Developments in Mathematics

Dorina Mitrea
Irina Mitrea
Marius Mitrea

Geometric
Harmonic
Analysis IV
Boundary Layer Potentials in Uniformly
Rectifiable Domains, and Applications
to Complex Analysis
Developments in Mathematics

Volume 75

Series Editors
Krishnaswami Alladi, Department of Mathematics, University of Florida,
Gainesville, FL, USA
Pham Huu Tiep, Department of Mathematics, Rutgers University, Piscataway, NJ,
USA
Loring W. Tu, Department of Mathematics, Tufts University, Medford, MA, USA

Aims and Scope

The Developments in Mathematics (DEVM) book series is devoted to publishing
well-written monographs within the broad spectrum of pure and applied mathe-
matics. Ideally, each book should be self-contained and fairly comprehensive in
treating a particular subject. Topics in the forefront of mathematical research that
present new results and/or a unique and engaging approach with a potential rela-
tionship to other fields are most welcome. High-quality edited volumes conveying
current state-of-the-art research will occasionally also be considered for publication.
The DEVM series appeals to a variety of audiences including researchers, postdocs,
and advanced graduate students.
Dorina Mitrea · Irina Mitrea · Marius Mitrea

Geometric Harmonic
Analysis IV
Boundary Layer Potentials in Uniformly
Rectifiable Domains, and Applications
to Complex Analysis
Dorina Mitrea Irina Mitrea
Department of Mathematics Department of Mathematics
Baylor University Temple University
Waco, TX, USA Philadelphia, PA, USA

Marius Mitrea
Department of Mathematics
Baylor University
Waco, TX, USA

ISSN 1389-2177 ISSN 2197-795X (electronic)

Developments in Mathematics
ISBN 978-3-031-29178-4 ISBN 978-3-031-29179-1 (eBook)
https://doi.org/10.1007/978-3-031-29179-1

Mathematics Subject Classification: 32A, 26B20, 31B, 35J, 42B

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Dedicated with love to our parents
Prefacing the Full Series

The current work is part of a series, comprised of five volumes, [68], [69], [70],
[71], [72]. In broad terms, the principal aim is to develop tools in Real and Harmonic
Analysis, of geometric measure theoretic flavor, capable of treating a broad spectrum
of boundary value problems formulated in rather general geometric and analytic
settings.
In Volume I ([68]) we establish a sharp version of Divergence Theorem (aka
Fundamental Theorem of Calculus) which allows for an inclusive class of vector
fields whose boundary trace is only assumed to exist in a nontangential pointwise
sense.
Volume II ([69]) is concerned with function spaces measuring size and/or smooth-
ness, such as Hardy spaces, Besov spaces, Triebel-Lizorkin spaces, Sobolev spaces,
Morrey spaces, Morrey-Campanato spaces, spaces of functions of Bounded Mean
Oscillations, etc., in general geometric settings. Work here also highlights the
close interplay between differentiability properties of functions and singular integral
operators.
The topic of singular integral operators is properly considered in Volume III
([70]), where we develop a versatile Calderón-Zygmund theory for singular integral
operators of convolution type (and with variable coefficient kernels) on uniformly
rectifiable sets in the Euclidean ambient, and the setting of Riemannian manifolds.
Applications to scattering by rough obstacles are also discussed in this volume.
In Volume IV ([71]) we focus on singular integral operators of boundary layer
type which enjoy more specialized properties (compared with generic, garden variety
singular integral operators treated earlier in Volume III). Applications to Complex
Analysis in several variables are subsequently presented, starting from the real-
izations that many natural integral operators in this setting, such as the Bochner-
Martinelli operator, are actual particular cases of double layer potential operators
associated with the complex Laplacian.
In Volume V ([72]), where everything comes together, finer estimates for a certain
class of singular integral operators (of chord-dot-normal type) are produced in a
manner which indicates how their size is affected by the (infinitesimal and global)
flatness of the “surfaces” on which they are defined. Among the library of double

vii
viii Prefacing the Full Series

layer potential operators associated with a given second-order system, we then iden-
tify those double layers which fall under this category of singular integral operators.
It is precisely for this subclass of double layer potentials that Fredholm theory may
then be implemented assuming the underlying domain has a compact boundary,
which is sufficiently flat at infinitesimal scales. For domains with unbounded bound-
aries, this very category of double layer potentials may be outright inverted, using
a Neumann series argument, assuming the “surface” in question is sufficiently flat
globally. In turn, this opens the door for solving a large variety of boundary value
problems for second-order systems (involving boundary data from Muckenhoupt
weighted Lebesgue spaces, Lorentz spaces, Hardy spaces, Sobolev spaces, BMO,
VMO, Morrey spaces, Hölder spaces, etc.) in a large class of domains which, for
example, are allowed to have spiral singularities (hence more general than domains
locally described as upper-graphs of functions). In the opposite direction, we show
that the boundary value problems formulated for systems lacking such special layer
potentials may fail to be Fredholm solvable even for really tame domains, like the
upper half-space, or the unit disk. Save for the announcement [67], all principal
results appear here in print for the first time.
We close with a short epilogue, attempting to place the work undertaken in this
series into a broader picture. The main goal is to develop machinery of geometric
harmonic analysis flavor capable of ultimately dealing with boundary value problems
of a very general nature. One of the principal tools (indeed, the piecè de résistance)
in this regard is a new and powerful version of the Divergence Theorem, devised in
Volume I, whose very formulation has been motivated and shaped from the outset
by its eventual applications to Harmonic Analysis, Partial Differential Equations,
Potential Theory, and Complex Analysis. The fact that its footprints may be clearly
recognized in the makeup of such a diverse body of results, as presented in Volumes
II-V, serves as testament to the versatility and potency of our brand of Divergence
Theorem. Alas, our enterprise is multifaceted, so its success is crucially dependent
on many other factors. For one thing, it is necessary to develop a robust Calderón-
Zygmund theory for singular integrals of boundary layer type (as we do in Volumes
III-IV), associated with generic weakly elliptic systems, capable of accommodating
a large variety of function spaces of interest considered in rather inclusive geometric
settings (of the sort discussed in Volume II). This renders these (boundary-to-domain)
layer potentials useful mechanisms for generating lots of null-solutions for the given
system of partial differential operators, whose format is compatible with the demands
in the very formulation of the boundary value problem we seek to solve. Next, in
order to be able to solve the boundary integral equation to which matters are reduced
in this fashion, the success of employing Fredholm theory hinges on the ability to
suitably estimate the essential norms of the (boundary-to-boundary) layer potentials.
In this vein, we succeed in relating the distance from such layer potentials to the
space of compact operators to the flatness of the boundary of the domain in question
(measured in terms of infinitesimal mean oscillations of the unit normal) in a desirable
manner which shows that, in a precise quantitative fashion, the flatter the domain the
smaller the proximity to compact operators. This subtle and powerful result, bridging
Prefacing the Full Series ix

between analysis and geometry, may be regarded as a far-reaching extension of the

pioneering work of Radon and Carleman in the early 1900’s.
Ultimately, ourwork aligns itself with the program stemming from A.P. Calderón’s
1978 ICM plenary address in which he advocates the use of layer potentials “for
much more general elliptic systems [than the Laplacian]” – see [9, p. 90], and may
be regarded as an optimal extension of the pioneering work of E.B. Fabes, M. Jodeit,
and N.M. Rivière in [24] (where layer potential methods have been first used to
solve boundary value problems for the Laplacian in bounded 𝒞1 domains). In this
endeavor, we have been also motivated by the problem1 posed by A.P. Calderón on
[9, p. 95], asking to identify the function spaces on which singular integral operators
(of boundary layer type) are well defined and continuous. This is relevant since, as
Calderón mentions, “A clarification of this question would be very important in the
study of boundary value problems for elliptic equations [in rough domains]. The
methods employed so far seem to be insufficient for the treatment of these problems.”
We also wish to mention that our work is also in line with the issue raised as an
open problem by C. Kenig in [44, Problem 3.2.2, pp. 116–117], where he asked
whether operators of layer potential type may be inverted on appropriate Lebesgue
and Sobolev spaces in suitable subclasses on NTA domains with compact Ahlfors
regular boundaries.
The task of making geometry and analysis work in unison is fraught with diffi-
culties, and only seldom can a two-way street be built on which to move between
these two worlds without loss of information. Given this, it is actually surprising
that in many instances we come very close to having optimal hypotheses, almost an
accurate embodiment of the slogan if it makes sense to write it, then it’s true.

Acknowledgments: The authors gratefully acknowledge partial support from the

Simons Foundation (through grants # 426669, # 958374, # 318658, # 616050,
# 637481), as well as NSF (grant # 1900938). Portions of this work have been
completed at Baylor University in Waco, Temple University in Philadelphia, the
Institute for Advanced Study in Princeton, MSRI in Berkeley, and the American
Institute of Mathematics in San Jose. We wish to thank these institutions for their
generous hospitality. Last, but not least, we are grateful to Michael E. Taylor for
gently yet persistently encouraging us over the years to complete this project.

Waco, USA Dorina Mitrea

Philadelphia, USA Irina Mitrea
Waco, USA Marius Mitrea

1 In the last section of [9], simply titled “Problems,” Calderón singles two directions for further

study. The first one is the famous question whether the smallness condition on a   L ∞ (the Lipschitz
constant of the curve {(x, a(x)) : x ∈ R} on which he proved the L 2 -boundedness of the Cauchy
operator) may be removed (as is well known, this has been solved in the affirmative by Coifman,
McIntosh, and Meyer in [15]). We are referring here to the second (and final) problem formulated
by Calderón on [9, p. 95].
Description of Volume IV

The bulk of the results in this volume amounts to a versatile Calderón-Zygmund

theory for singular integral operators of layer potential type in open sets with
uniformly rectifiable boundaries. As such, we find it appropriate to attempt to place
the portion of our work pertaining to this brand of Calderón-Zygmund theory into a
broader picture.
Traditionally, the label “Calderón-Zygmund theory” has been applied to a distin-
guished body of works primarily pertaining to the mapping properties of singular inte-
gral operators on Lebesgue spaces, considered in various ambients (ranging from the
standard Euclidean setting to arbitrary spaces of homogeneous type). For example,
thanks to the fundamental work of G. David and S. Semmes (cf. [19], [20]), it is now
understood that, within the Euclidean setting, the most general environment where
L p boundedness results, with 1 < p < ∞, are available for large classes of singular
integral operators is that of uniformly rectifiable subsets2 of Rn . This being said,
demands arising in related areas (e.g., boundary value problems in rough domains)
have stimulated interest in expanding the scope of the classical Calderón- Zygmund
theory as to include other category of function spaces, such as: Hardy-like spaces,
Sobolev-like spaces, Morrey spaces, Morrey-Campanato spaces, Lorentz spaces,
BMO spaces, VMO spaces, Hölder spaces, Besov spaces, Triebel-Lizorkin spaces,
among many others, while retaining the general nature of the geometric environ-
ment. Through the efforts of many people, steady progress has been registered (for
example, basic singular integral operators on Lipschitz surfaces are now fairly well-
understood) but significant obstacles have limited the success of such an endeavor.
Compared with Lebesgue spaces, some of the categorically distinct issues which arise
in dealing with singular integral operators acting on many of the scales of spaces
listed above are those of cancelation and/or smoothness. Such aspects are front and
center within the realm of Hardy spaces, Sobolev spaces, Morrey- Campanato spaces,
BMO spaces, Besov spaces, Triebel-Lizorkin spaces, etc. This calls for technology
capable of dealing with the issues of cancelation and smoothness in very general

2however, the quality of being uniformly rectifiable loses its central significance if in place of
Lebesgue spaces other scales of spaces (e.g., Hölder) are considered.

xi
xii Description of Volume IV

rough settings, such as sets which are uniformly rectifiable, or even merely Ahlfors
regular. The brand of Divergence Theorem developed in Volume I ([68]) goes a
long way in addressing this need, by yielding powerful and versatile integration
by parts formulas and trace results in very general settings. Succinctly put, various
basic aspects of the theory of singular integral operators require subtle cancelations
properties which our version of the Divergence Theorem can, for the first time,
accommodate.
To offer a concrete example, start in the context of the entire Euclidean space Rn
and consider a linear and continuous mapping T : 𝒮(Rn ) → 𝒮 (Rn ) which extends
to a bounded operator on L 2 (Rn , Ln ) and has the property that its Schwartz kernel
K (·, ·) satisfies
 
K ∈ L 1loc Rn × Rn \diag, Ln ⊗ Ln (0.0.1)

and there exist some constant C ∈ (0, ∞) together with some exponent γ ∈ (0, 1]
such that

for every x, y ∈ Rn with x = y, and each z ∈ Rn with |y − z|< 21 |x − y|,

C C|y − z|γ
one has |K (x, y)| ≤ n and |K (x, y) − K (x, z)| ≤ . (0.0.2)
|x − y| |x − y|n+γ

A classical result in harmonic analysis (see, e.g., the proof of [65, Théorème 3,
pp. 237–238]) is that3

T extends to a linear and continuous mapping from the Hardy

space H 1 (Rn ) into itself if and only if T T (1) = 0, in the sense

that T a has mean zero, i.e., T a dLn = 0 for all H 1 -atoms a. (0.0.3)
Rn

In the entire Euclidean space, there are certain natural ways of checking the cance-
lation condition required in (0.0.3) is satisfied, such as using the Fourier transform
in the case when T is a Fourier multiplier.
It has long know that this type of characterization of boundedness on Hardy spaces
is valid in the more general setting of spaces of homogeneous type, in the sense of
R. Coifman and G. Weiss. See, e.g., [16, p. 599], [22, Proposition 4.17, p. 104],
[22, Theorem 4.27, p. 112]. This being said, when the ambient Rn is replaced by a
uniformly rectifiable set  ⊆ Rn , an environment in which large classes of singular
integral operators are bounded on L 2 with respect to σ := Hn−1 , there are
 
3 the same is true with H 1 (Rn ) replaced by H p (Rn ) for each p ∈ n−1
n−1+γ ,1 .
Description of Volume IV xiii

basically no known examples of singular integral operators bounded on H 1 (, σ )

since the actual verification of the crucial cancelation condition

T a dσ = 0 whenever a is an atom on  (0.0.4)


becomes rather delicate even in the basic case when T is the (transpose) harmonic
double layer operator on . As we shall presently see, the brand of the Divergence
Theorem developed in Volume I ([68]) can efficiently deal with such an issue.
Ultimately, the picture that emerges is that Calderón-Zygmund theory is a multi-
faceted body of results aimed at describing how singular integral operators behave in
a multitude of geometric and analytic settings. The final goal becomes understanding
the intimate correlation between geometry and analysis from this perspective and,
eventually, building a “two-way street” allowing to pass information back and forth
between them in an optimal fashion. The study of singular integral operators on
Lebesgue spaces becomes a chapter, albeit a fundamental one, in this theory and our
own work contributes to this on-going program by vigorously promoting this more
general and inclusive point of view. The good news is that, as formidable as this
already is, such a version of Calderón-Zygmund theory is yet to reach the height of
its splendor. This is a vision worth sharing!
This portion of our work has also been motivated by the problem posed by A.P.
Calderón on [9, p. 95], asking to identify the function spaces on which singular inte-
gral operators (of boundary layer type) are well defined and continuous. Calderón
goes on to mention that: “A clarification of this question would be very important
in the study of boundary value problems for elliptic equations [in rough domains].
The methods employed so far seem to be insufficient for the treatment of these prob-
lems.” We shall employ the body of results established in this volume in the study of
boundary value problems in rather inclusive geometric settings and with boundary
data in a multitude of function spaces in Volume V ([72]).
Let us now describe the contents of the present volume in greater detail.
Chapter 1 deals with singular integral operators of boundary layer type on Lebesgue
and Sobolev spaces. One of the main points is that generic Calderón-Zygmund
convolution-type SIO’s (of the sort considered in [70, Chapter 2]) are not expected
to induce well-defined mappings on Sobolev spaces on UR sets. Indeed, for this to
happen, the integral kernel must possess a special algebraic structure. For example,
this is present in the conormal derivative, or a tangential derivative, of the funda-
mental solution of a weakly elliptic second-order system, and these are the types of
singular integral operators we focus on in this chapter. In addition to the discussion
on the history and physical interpretations of the classical harmonic layer poten-
tials from §1.1, topics treated in this chapter include “tangential” singular integral
operators, whose kernels exhibit a special algebraic structure, strongly reminiscent
of tangential derivatives (in §1.2), volume and integral operators of boundary layer
type associated with a given open set of locally finite perimeter and a given weakly
elliptic system (in §1.3), a multitude of relevant examples and alternative points of
view (in §1.4), a rich function theory of Calderón-Zygmund type for boundary layer
xiv Description of Volume IV

potentials associated with a given weakly elliptic system and an open set with a UR
boundary (in §1.5), the interpretation of the Cauchy and Cauchy-Clifford operators
as double layer potential operators, and the host of consequences naturally derived
from such a perspective (in §1.6), the description of kernels and images of singular
integral operators of boundary layer type (in §1.7), and how to modify boundary
layer potential operators as to increase the class of functions to which they may be
applied (in §1.8).
Chapter 2 is largely concerned with layer potential operators on Hardy, BMO,
VMO, and Hölder spaces defined on boundaries of UR domains. Once again, it takes
a special algebraic structure of the integral kernel for a singular integral operator to
map either of these spaces into itself, and the Divergence Theorem devised in Volume
I ([68]) plays a crucial role in ensuring this is indeed the case; see the discussion
surrounding (0.0.4). In fact, the same type of philosophy prevails in relation to the
action of double layer potential operators on Calderón, Morrey-Campanato, and
Morrey spaces, studied in Chapter 3, and also for the action of double layer potential
operators on Besov and Triebel-Lizorkin spaces, a subject discussed at length in
Chapter 4.
The above considerations bring into focus the following fundamental question:
describe the most general classes of singular integral operators on the boundary of
an arbitrary given UR domain  ⊆ Rn which map Hardy, BMO, VMO, Hölder,
Besov, and Triebel-Lizorkin spaces defined on ∂ boundedly into themselves. We
provide an answer to this basic question in Chapter 5 through the consideration
of what we call “generalized double layers.” The main attribute of these singular
integral operators is the fact that their integral kernels involve the inner product of the
outward unit normal (to the “surface” on which this integral operator is defined) with a
divergence-free vector-valued function. Such an algebraic structure confers excellent
cancelation properties (brought to bear by the Divergence Theorem) which, in turn,
allow us to establish boundedness results for these generalized double layers on a
multitude of basic scales of function spaces which, in addition to standard Lebesgue
spaces (and its Muckenhoupt weighted version), now also includes boundary Sobolev
spaces, Hardy spaces, Hölder spaces, the John-Nirenberg space BMO, the Sarason
space VMO, Besov spaces, and Triebel-Lizorkin spaces, among others. In the last
section of Chapter 5 we take another look at Riesz transforms from the perspecive
of generalized double layers.
In Chapter 6 we develop a theory of boundary layer potentials associated with
the Stokes system of linear hydrostatics, and related topics. Among other things, we
establish Green-type formulas, derive mapping properties for the aforementioned
boundary layer potential operators, and prove Fatou-type results, in settings which are
sharp from a geometric/analytic point of view. Once again, the brand of Divergence
Theorem discussed in Volume I ([68]) plays a prominent role in carrying out this
program.
Chapter 7 contains applications of the tools and results developed so far to anal-
ysis in several complex variables. It has long been known that Complex Analysis,
Geometric Measure Theory, and Harmonic Analysis tightly interface in the complex
plane (see, e.g., J. Garnett’s book [28] and the earlier references cited there). This
Description of Volume IV xv

is an area of mathematics which continues to undergo major transformations thanks

to spectacular contemporary advances, such as G. David’s characterization of the L 2
boundedness of the Cauchy operator in terms of Ahlfors regularity, and X. Tolsa’s
results on analytic capacity, to name just some. However, this rich interplay between
these branches of mathematics seems to have been considerably less explored in
the higher-dimensional setting, involving several complex variables. The main goal
in §7.1-§7.7 is to further our present understanding of this latter aspect. Themes
covered in this chapter include: CR-functions and differential forms on boundaries
of sets of locally finite perimeter (in §7.1), integration by parts formulas involving
¯
the ∂-operator on sets of locally finite perimeter (in §7.2), the Bochner-Martinelli
integral operator (in §7.3), a sharp version of the Bochner-Martinelli-Koppelman
formula (in §7.4), the Extension Problem for Hölder CR-functions on boundaries of
Ahlfors regular domains (in §7.5), the Extension Problem for Lebesgue and bounded
(or vanishing) mean oscillation functions on boundaries of uniformly rectifiable
¯
domains (in §7.6), as well as the ∂-operator and the Dolbeault complex on uniformly
rectifiable sets (in §7.7).
Lastly, in Chapter 8 we study Hardy spaces (in which the size is measured via the
nontangential maximal operator) for certain second-order weakly elliptic operators
in the complex plane. We start with Bitsadze’s operator ∂z̄2 in the unit disk of the
complex plane in §8.1, and expand the scope of this analysis in §8.2 by considering
the more inclusive family of second-order differential operators L λ := ∂z̄2 − λ2 ∂z2 ,
indexed by the parameter λ ∈ C. In all cases, the goal is to characterize the space
of null-solutions (satisfying approriate size conditions, decsribed in terms of the
nontangential maximal operator) and to identify precisely the corresponding spaces
of boundary traces.
Contents

Prefacing the Full Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Description of Volume IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1 Layer Potential Operators on Lebesgue and Sobolev Spaces . . . . . . . . 1
1.1 Comments on History and Physical Interpretations
of Harmonic Layer Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 “Tangential” Singular Integral Operators . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 A First Look at Layer Potential Operators . . . . . . . . . . . . . . . . . . . . . . 24
1.4 Examples and Alternative Points of View . . . . . . . . . . . . . . . . . . . . . . 43
1.5 Calderón-Zygmund Function Theory for Boundary Layer
Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue
and Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
1.7 Kernels and Images of Boundary Layer Potentials . . . . . . . . . . . . . . . 153
1.8 Modified Boundary Layer Potential Operators . . . . . . . . . . . . . . . . . . 173
2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder
Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
2.1 Double Layer Potential Operators on Hardy, BMO, VMO,
and Hölder Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
2.2 Single Layer Operators Acting from Hardy Spaces . . . . . . . . . . . . . . 271
2.3 Integral Operators of Layer Potential Type on Hardy-Based
Sobolev Spaces and BMO−1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
3 Layer Potential Operators on Calderón, Morrey-Campanato,
and Morrey Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
3.1 Boundary Layer Potentials on Calderón Spaces . . . . . . . . . . . . . . . . . 376
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces
and Their Pre-Duals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
3.3 Boundary Layer Potential Operators on Morrey Spaces
and Their Pre-Duals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394

xvii
xviii Contents

4 Layer Potential Operators Acting from Boundary Besov

and Triebel-Lizorkin Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
4.1 Boundary-to-Boundary Layer Potentials from Besov
and Triebel-Lizorkin Spaces into Themselves . . . . . . . . . . . . . . . . . . . 431
4.2 Boundary-to-Domain Layer Potentials from Besov Spaces
into Weighted Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
4.3 Boundary-to-Domain Layer Potentials from Besov Spaces
into Besov and Triebel-Lizorkin Spaces . . . . . . . . . . . . . . . . . . . . . . . . 481
4.4 Integral Representation Formulas of Layer Potential Type,
and Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
5 Generalized Double Layers in Uniformly Rectifiable Domains . . . . . . 531
5.1 Theory of Generalized Double Layers . . . . . . . . . . . . . . . . . . . . . . . . . 532
5.2 Generalized Double Layers with Matrix-Valued Kernels,
and Chord-Dot-Normal SIO’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
5.3 Another Look at Standard and Modified Riesz Transforms . . . . . . . 657
6 Green Formulas and Layer Potential Operators for the Stokes
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673
6.1 Green-Type Formulas for the Stokes System . . . . . . . . . . . . . . . . . . . . 674
6.2 Boundary Layer Potential Operators for the Stokes System:
Lebesgue, Sobolev, and Hardy Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 684
6.3 Other Integral Representations and Fatou-Type Results
for the Stokes System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 740
6.4 Layer Potentials for the Stokes System on Besov,
Triebel-Lizorkin, and Weighted Sobolev Spaces . . . . . . . . . . . . . . . . 762
7 Applications to Analysis in Several Complex Variables . . . . . . . . . . . . . 783
7.1 CR-Functions and Differential Forms on Boundaries
of Locally Finite Perimeter Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784
7.2 Integration by Parts Formulas Involving the ∂ Operator
on Sets of Locally Finite Perimeter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 797
7.3 The Bochner-Martinelli Integral Operator . . . . . . . . . . . . . . . . . . . . . . 812
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman
Formula and Related Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827
7.5 The Extension Problem for Hölder CR-Functions
on Boundaries of Ahlfors Regular Domains . . . . . . . . . . . . . . . . . . . . 857
7.6 The Extension Problem for L p /BMO/VMO/Morrey
Functions on Boundaries of Uniformly Rectifiable Domains . . . . . . 871
7.7 The ∂ Operator and the Dolbeault Complex on Uniformly
Rectifiable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 878
Contents xix

8 Hardy Spaces for Second-Order Weakly Elliptic Operators

in the Complex Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator
∂z2 in the Unit Disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884
8.2 Null-Solutions and Boundary Traces for the Operator
∂z2 − λ2 ∂z2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 909

A Terms and notation used in Volume IV . . . . . . . . . . . . . . . . . . . . . . . . . . . 923

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 973
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 979
Symbol Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983
Chapter 1
Layer Potential Operators on Lebesgue and
Sobolev Spaces

While certain features of singular integral operators are visible from geometric
properties in general spaces of homogeneous type (such as the Fractional Integration
Theorem, or the abstract boundedness criteria from [16]) other, more delicate prop-
erties (typically cancelation sensitive) require fully employing the resourcefulness
of the algebraic/geometric ambient and, crucially, involve differential calculus. The
boundedness of singular integral operators of boundary layer type on Sobolev spaces
falls under the latter category. Indeed, the boundary Sobolev spaces developed in
[69, Chapter 11] offer a functional context in which a variety of singular integral
operators (SIO’s for short) of boundary layer type act in a natural fashion. The main
goal of this chapter is to elaborate on this idea. In the process, we shall see that the
topics of boundary Sobolev spaces and SIO’s are closely intertwined. Specifically,
we are going to employ SIO’s as a tool to further our understanding of the brand of
boundary Sobolev spaces introduced in [69, Chapter 11], ultimately establishing a
two-way bridge between these two areas.

1.1 Comments on History and Physical Interpretations of

Harmonic Layer Potentials

To put matters in a broader perspective, we begin by making some remarks pertaining

to the history and origins of the classical boundary layer potentials associated with
the Laplace operator, i.e., when L = Δ. Here we shall also elaborate on ties with
mathematical physics, by highlighting connections between these boundary layer
potentials and charge distributions on surfaces. Throughout, we shall let Ω ⊆ Rn be
an Ahlfors regular domain, with geometric measure theoretic outward unit normal
ν, and use σ to abbreviate H n−1 ∂Ω.
In many ways, solving a PDE amounts to “undoing” (i.e., inverting) a differential
operator, and this continues to be the case when dealing with boundary value prob-
lems. For example, when working with the Laplacian Δ = ∂12 + · · · + ∂n2 in Rn , the
fundamental solution
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_1
2 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

⎧
⎪ 1 1
⎪
⎪ if n ≥ 3,
⎪
⎨ ωn−1 (2 − n) |x| n−2
⎪
EΔ (x) := (1.1.1)
⎪
⎪
⎪
⎪ 1
⎪ ln |x| if n = 2,
⎩ 2π
(where ωn−1 is the area of the unit sphere in Rn ) plays a basic role, since this may
be used to generate lots of harmonic functions in a given domain Ω ⊆ Rn , such as

Ω  x −→ EΔ (x − y) where y ∈ ∂Ω is an arbitrary fixed point. (1.1.2)

In fact, one may consider linear combinations of such mappings, i.e.,


N
Ω  x −→ EΔ (x − y j ) · λ j with {y j }1≤ j ≤ N ⊆ ∂Ω and {λ j }1≤ j ≤ N ⊆ C. (1.1.3)
j=1

More generally, one may consider a “density” f (y) ∈ C at each point y ∈ ∂Ω and
use it to create a weighted infinite “sum” of functions as in (1.1.2). This gives rise
to the boundary-to-boundary integral operator
∫
Ω  x −→ EΔ (x − y) f (y) dσ(y). (1.1.4)
∂Ω

We are therefore led to consider the classical

 harmonic single layer operator as the
mapping assigning to each function f ∈ L ∂Ω, 1+σ(x)
1
|x | n−2
if n ≥ 3 and each function

f ∈ L 1 ∂Ω, ln(2 + |x|)σ(x) if n = 2 the harmonic function defined at each point
x ∈ Rn \ ∂Ω by
∫
𝒮Δ f (x) := EΔ (x − y) f (y) dσ(y)
∂Ω
∫
⎧
⎪ 1 f (y)
⎪
⎪ dσ(y) if n ≥ 3,
⎪
⎨ (2 − n)ωn−1 ∂Ω |x − y| n−2
⎪
= ∫ (1.1.5)
⎪
⎪ 1
⎪
⎪ (ln |x − y|) f (y) dσ(y) if n = 2.
⎪ 2π
⎩ ∂Ω

Of course, one may re-fashion the process which, starting with (1.1.2), has led
to (1.1.4), but now taking directional derivatives of EΔ . For the purpose of dealing
with boundary value problems it is most natural to consider the normal derivative of
EΔ , a choice which gives rise to the boundary-to-boundary integral operator
∫
Ω  x −→ ∂ν(y) EΔ (x − y) f (y) dσ(y), (1.1.6)
∂Ω

where ν is the outward unit normal to Ω. This is how the boundary-to-boundary

harmonic double layer potential operator comes into play, as the mapping assigning
1.1 Comments on History and Physical Interpretations of Harmonic Layer Potentials 3

to each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
the harmonic function
∫
DΔ f (x) := ∂ν(y) EΔ (y − x) f (y) dσ(y)
∂Ω
∫
1 ν(y), y − x
= f (y) dσ(y) for each x ∈ Rn \ ∂Ω. (1.1.7)
ωn−1 ∂Ω |x − y| n
The single layer potential operator has already appeared in the work of Carl
Friedrich Gauss as early as 1838. More specifically, in [29] and [30], Gauss introduces
boundary integral equations of the first kind and used them to perform numerical
computations1. Among other things, Gauss has established jump-formulas for the
normal derivatives of the harmonic single layer, and re-discovered what is now
commonly referred to as Green-Gauss formula.
As with the single layer, the double layer is a mechanism for creating
 a multitude
on null-solutions for the Laplacian in Ω, since for each f ∈ L ∂Ω, 1+σ(x)
1
|x | n−1
the
function u := DΔ f satisfies

u ∈ 𝒞∞ (Ω) and Δu = 0 in Ω. (1.1.8)

Moreover, if Ω is a UR domain then the nontangential boundary trace of u may be

explicitly computed as
n.t.
u ∂Ω = ( 12 I + KΔ ) f at σ-a.e. point on ∂Ω, (1.1.9)

where I is the identity and KΔ is the principal-value singular integral operator

∫
(KΔ f )(x) := lim+ ∂ν(y) EΔ (x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (1.1.10)
ε→0
y ∈∂Ω
|x−y |>ε

As yet another motivation for the format of 𝒮Δ , recall Isaac Newton’s law of uni-
versal gravitation according to which two particles, located at x0, x ∈ R3 and having
masses m0, m ∈ (0, ∞), attract one another with a force of magnitude cm0 m/|x0 − x| 2 ,
where c is a universal constant. Hence, in vector notation which also incorporates
the direction, the force with which the body located at x is attracted to the body
located at x0 is given by
cm0 m x0 − x x0 − x
Fx0,x := = cm0 m . (1.1.11)
|x0 − x| 2 |x0 − x| |x0 − x| 3
Keeping x0 fixed and regarding this as a function of x yields a conservative vector
field in R3 \ {x0 }, since

1 Gauss actually published extensive tables and graphs of numerical results obtained, at least in
part, using such boundary integral equations; see [17] for an informative account on this topic
4 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
x0 − x  cm m
0
cm0 m = ∇ x for each x ∈ R3 \ {x0 }. (1.1.12)
|x0 − x| 3 |x0 − x|

This brings into focus the potential function possessed by the gravitational field, i.e.,
cm0 m
Ux0 (x) := for each x ∈ R3 \ {x0 }, (1.1.13)
|x0 − x|
which is a scalar-valued function satisfying Laplace’s equation

ΔUx0 = 0 in R3 \ {x0 }. (1.1.14)

Consider now the case when Ω ⊆ R3 is an Ahlfors regular domain, x0 is a point in

R3 \ ∂Ω, and an attractive mass is distributed over the “surface” (or “layer”) ∂Ω.
Denote by σ the “surface” measure H 2 ∂Ω, and denote by m(y) the mass density
at each point y ∈ ∂Ω. The superposition principle, according to which for all linear
systems the net response caused by a family of stimuli is the sum of the responses that
would have been caused by each stimulus individually, then shows that the overall
potential associated with this system is
∫
m(y)
cm0 dσ(y). (1.1.15)
∂Ω |x0 − y|

Up to normalization and readjusting notation, this agrees with the harmonic single
layer potential operator (1.1.5) (acting on the density m, and evaluated at x0 ) in the
three-dimensional setting.
The classical harmonic double layer potential operator, recalled in (1.1.7), is
typically associated with the names of Carl Neumann and Henri Poincaré. For this
reason, (1.1.7) or, rather, its principal-value version (cf. (A.0.101)), is sometimes
referred to as the Neumann-Poincaré operator2. Following Gauss’ pioneering study
of the first kind integral equation associated with the single layer potential operator,
the next major progress was registered with Carl Neumann’s work on the double
layer potential operator, a topic on which he has published extensively (including his
1877 monograph [82]). Neumann’s crowning achievement was his solution of the
second kind integral equation associated with the double layer potential operator in
convex domains, via what we presently call a Neumann series. Neumann’s proof of
the convergence of the method of iterations in this setting uses rather sophisticated
geometric and measure theoretic arguments.
In essence, this constitutes an early example of “hard analysis” in potential theory,
which set the tone for a great deal of work that followed. Indeed, many subsequent
generalizations of Neumann’s techniques have also been confined to hard harmonic
analysis. For example, in a genuine tour de force, Poincaré has subsequently succeed-
ed (in his long and technical paper [84] published in 1897) to replace the convexity
hypothesis by suitable smoothness assumptions. More specifically, Poincaré accom-
plished the task of finding an alternative proof the convergence of the Neumann
series for nonconvex domains, via an approach which required the underlying set

2 a name first used by T. Carleman in his Ph. D. thesis [12]

1.1 Comments on History and Physical Interpretations of Harmonic Layer Potentials 5

to have a 𝒞2 boundary and the functions involved to be fairly regular. Along the
way, Poincaré develops in [84] an astounding array of novel techniques. While he
himself scarcely employed them further, other mathematicians picked up the mantle
and continued to develop C. Neumann and H. Poincaré’s groundbreaking work in a
multitude of different directions, including Arthur Korn ([46], [47], [48], [49], [50])
Vladimir Steklov ([97], [98], [99], [100], [101]), Stanislaw Zaremba ([106], [107],
[108]), Johann Karl August Radon ([86], [87]), and Torsten Carleman ([12]), just to
name a few. The classical monographs [31] of N.M. Günter, [43] of O.D. Kellogg’s,
[91] of F. Riesz and B. Sz.-Nagy, and [95] of S.L. Sobolev are still quite readable,
informative accounts on this and related topics.
To elaborate on the physical significance of the harmonic double layer potential
operator (1.1.7) work in the three-dimensional setting. Specifically, consider an
Ahlfors regular domain Ω ⊆ R3 and suppose attractive mass is distributed over the
“surface” (or “layer”) ∂Ω. As before, we shall let σ denote the “surface” measure
H 2 ∂Ω, and denote by m(y) the mass density at each point y ∈ ∂Ω. Having fixed
some (unit) point mass x ∈ Ω, the gravitational force attracting each point mass y to
x is (cf. (1.1.11))
x−y
Fx (y) := c m(y) . (1.1.16)
|x − y| 3
This gives rise to a vector field Fx : ∂Ω → R3 , whose flux on the “surface” ∂Ω is
∫ ∫
 ν(y), x − y
ν(y), Fx (y) dσ(y) = c m(y) dσ(y) (1.1.17)
∂Ω ∂Ω |x − y| 3
which, up to normalization and readjusting notation, agrees with the harmonic double
layer potential operator (1.1.7) (acting on the density m, and evaluated at x) in the
three-dimensional setting.
Here another point of view which provides an alternative physical interpretation of
the harmonic double layer (1.1.7) in the three-dimensional setting. The starting point
is Coulomb’s law describing the field of electrostatic forces in a completely similar
manner to Newton’s law (1.1.11) (with m0 and m electric charges in this context).
Next, and suppose Ω ⊆ R3 is a open set with a sufficiently regular boundary,
assumed to be an insulator. Fix ε > 0 and define two3 “parallel layers,” namely
Sε± := {y ± εν(y) : y ∈ ∂Ω}, where ν is the outward unit normal vector to Ω.
Assume Sε± are conductors charged with distribution4 ±ε −1 ρ(y) for each y ∈ ∂Ω.
According to the superposition principle recalled earlier, the electric field generated
by these charges is given by ∇Uε where Uε is the associated potential function, i.e.,
∫  
ε −1 ρ(y) −ε −1 ρ(y)
Uε (x) := + dσ(y) for each x ∈ R3 \ ∂Ω.
∂Ω |x − (y + εν(y))| |x − (y − εν(y))|
(1.1.18)
Upon letting ε → 0+ , we see that for each x ∈ R3 \ ∂Ω and each y ∈ ∂Ω we have

3 hence the name, double layer

4 while attempting to use ±ε α ρ(y) one soon discovers that α = −1 is the only exponent for which
the limit in (1.1.19) is nontrivial
6 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 ε −1 −ε −1  ν(y), y − x
+ −→ (−2) (1.1.19)
|x − (y + εν(y))| |x − (y − εν(y))| |x − y| 3
which shows that the harmonic double layer potential can, up to normalization, be
thought of as the potential induced by a double layer of charges of opposite signs on
∂Ω.
In parallel to the integral methods used in potential theory, centered around
the Laplacian, a similar body of results has been constructed in elasticity theory,
corresponding to the Lamé system. Originally, these theories have been largely
independent and developed from within, though it has been noted early on that they
exhibit many formal analogies. One of the benefits of dealing with boundary layer
potentials associated with general weakly elliptic systems, as we do in this volume, is
that this builds a robust, inclusive theory which, in particular, subsumes the classical
cases of the Laplacian and the Lamé system at once.

1.2 “Tangential” Singular Integral Operators

We begin by considering integral operators whose kernels have the special alge-
braic structure described in the proposition below, strongly resembling a tangential
derivative.

Proposition 1.2.1 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with the

property that ∂Ω is a UR set. Abbreviate σ∗ := H n−1 ∂∗ Ω and σ := H n−1 ∂Ω. Also,
denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to
Ω. Next, consider a complex-valued function b ∈ Lloc 1 (Rn, L n ) with the property that

b Rn \{0} ∈ 𝒞 (R \ {0}) where N = N(n) ∈ N is a sufficiently large number, and

N n

such that ∇b is odd and positive homogeneous of degree 1 − n in Rn \ {0}. Finally,

 j, k ∈ {1, . . . , n} introduce the integral operators acting on
for each pair of indices
each function f ∈ L 1 ∂∗ Ω, 1+σ|y∗ (y)
| n−1
according to
∫
 
Tjk f (x) := ν j (y)(∂k b)(x − y) − νk (y)(∂j b)(x − y) f (y) dσ∗ (y), ∀x ∈ Ω,
∂∗ Ω
(1.2.1)

and, for σ-a.e. x ∈ ∂Ω,

∫
 
Tjk f (x) := lim+ ν j (y)(∂k b)(x − y) − νk (y)(∂j b)(x − y) f (y) dσ∗ (y),
ε→0
y ∈∂∗ Ω
|x−y |>ε
(1.2.2)
1.2 “Tangential” Singular Integral Operators 7

(which is meaningful thanks to [70, (2.3.15)]). Also, for each f ∈ L 1 ∂Ω, 1+σ(y) |y | n−1
define at σ∗ -a.e. x ∈ ∂∗ Ω (cf. [70, (2.3.15)])
∫
 
#
Tjk f (x) := lim+ ν j (x)(∂k b)(x − y) − νk (x)(∂j b)(x − y) f (y) dσ(y).
ε→0
y ∈∂Ω
|x−y |>ε
(1.2.3)

Then the following statements are true.

(i) For each aperture parameter κ ∈ (0, ∞) one has

κ−n.t.

Tjk f = Tjk f at σ∗ -a.e. point on ∂∗ Ω,

∂Ω  (1.2.4)
for every function f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
.
 
Also, for each f in the weighted boundary Sobolev space L11 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
(cf.
(A.0.131)), each index ∈ {1, . . . , n}, and each aperture parameter κ ∈ (0, ∞),
the pointwise nontangential boundary trace
  κ−n.t.
∂ Tjk f ∂Ω
exists at σ∗ -a.e. point on ∂∗ Ω. (1.2.5)
p,q
As a consequence of [69, (11.7.22)], this is true whenever f ∈ L1 (∂∗ Ω, σ∗ )
with p, q ∈ [1, ∞).
In addition, for each p, q ∈ [1, ∞) and κ > 0 there exists some finite constant
C > 0, depending only on ∂Ω, b, n, p, q, and κ, such that for each function
p,q
f ∈ L1 (∂∗ Ω, σ∗ ) one has
   
Nκ (Tjk f ) p + Nκ (∇Tjk f ) L q (∂Ω,σ) ≤ C f  L p, q (∂∗ Ω,σ∗ ) (1.2.6)
L (∂Ω,σ) 1

if p, q > 1, plus similar estimates in the case when p = 1 or q = 1, in which

scenario the corresponding L 1 -norms in the left-hand side are now replaced by
the quasi-norm L 1,∞ (∂Ω, σ).
(ii) Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operator
#
Tjk : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ∗ ) (1.2.7)

# in (1.2.7) is
is well defined, linear, and bounded. Moreover, the transpose of Tjk
 
−Tjk : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ). (1.2.8)

In fact, similar results are valid for the more general scale of Muckenhoupt
weighted Lebesgue spaces on ∂Ω, as well as for Lorentz spaces on ∂Ω.
(iii) Make the additional assumption that
8 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 
H n−1 ∂nta Ω \ ∂∗ Ω = 0 (1.2.9)

(a condition automatically satisfied if, e.g., Ω is a UR domain to begin with).

Then the operator
p,q p,q
Tjk : L1 (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.2.10)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
p p
Tjk : L1 (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.2.11)

is well defined, linear, and bounded for each p ∈ (1, ∞).

Moreover, given p, q ∈ (1, ∞), for each function
 σ∗ (x)
p
f ∈ Lloc (∂∗ Ω, σ∗ ) ∩ L 1 ∂∗ Ω, satisfying
1 + |x| n−1
 (1.2.12)
q σ∗ (x)
∂τi f ∈ Lloc (∂∗ Ω, σ∗ ) ∩ L ∂∗ Ω,
1
for all 1 ≤ i, ≤ n,
1 + |x| n−1

and each pair of indices r, s ∈ {1, . . . , n} one has

∂τr s (Tjk f ) = Tr#s (∂τ j k f ) at σ∗ -a.e. point on ∂∗ Ω, (1.2.13)

 
where ∂τ j k f is regarded as a function in L 1 ∂Ω, 1+σ(x)
|x | n−1
(where Tr#s is defined)
by extending it by zero outside ∂∗ Ω to the entire topological boundary ∂Ω. In
p,q
particular, formula (1.2.13) holds for every f ∈ L1 (∂∗ Ω, σ∗ ) with integrability
exponents p, q ∈ (1, ∞).
(iv) Retain the additional assumption made in (1.2.9). Then for each integrability
exponent p ∈ (1, ∞) it follows that Tjk # , originally acting on functions from

L (∂∗ Ω, σ∗ ) (regarding them as being extended by zero to the entire topological

p

boundary ∂Ω, and then applying Tjk # in the sense of (1.2.7)), further extends

uniquely to a linear, bounded operator, from the negative boundary Sobolev

p
space L−1 (∂∗ Ω, σ∗ ) into itself. Furthermore, if one adopts the same notation Tjk
#

for this extension, then the transpose of (1.2.11) is

p p
−Tjk
#
: L−1 (∂∗ Ω, σ∗ ) −→ L−1 (∂∗ Ω, σ∗ ) (1.2.14)

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

More generally, under the additional assumption made in (1.2.9), for each
p, q ∈ (1, ∞) it follows that Tjk# acting on functions from L p (∂ Ω, σ ) (in the
∗ ∗
manner described above) further extends uniquely to a linear and bounded
p,q
operator from the negative off-diagonal boundary Sobolev space L−1 (∂∗ Ω, σ∗ )
# for this extension, the transpose
into itself and, retaining the same notation Tjk
of (1.2.10) is
1.2 “Tangential” Singular Integral Operators 9

p,q p,q
−Tjk
#
: L−1 (∂∗ Ω, σ∗ ) −→ L−1 (∂∗ Ω, σ∗ ) (1.2.15)

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1.

Proof To justify (1.2.4), first observe that for each x ∈ Rn \ B(0, 2) we may use the
Fundamental Theorem of Calculus to write
∫ 1
     
b(x) − b |xx | = (∇b) t x + (1 − t) |xx | · x − |xx | dt. (1.2.16)
0

Bearing in mind that b is bounded on S n−1 , on account of the homogeneity property

satisfied by ∇b this readily implies

b(x) = O(|x|) as |x| → ∞. (1.2.17)

Since we also have b ∈ Lloc

1 (Rn, L n ), it follows that b induces a tempered distribution

in R via integration against Schwartz functions. Invoking [70, (2.5.19)] for this
n

tempered distribution then yields

ξk (∂ 
j b)(ξ) = ξ j (∂k b)(ξ) for each ξ = (ξ1, . . . , ξn ) ∈ R \ {0}.
n
(1.2.18)

Having established (1.2.18), for each given f ∈ L 1 ∂∗ Ω, 1+σ|y∗ (y) | n−1
we may now rely
on [70, Theorem 2.5.1] to compute
  κ−n.t.
Tjk f ∂Ω (x)

1   1  
= ∂k b ν(x) ν j (x) f (x) − ∂ j b ν(x) νk (x) f (x)
2i 2i
∫  
+ lim+ ν j (y)(∂k b)(x − y) − νk (y)(∂j b)(x − y) f (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

= (Tjk f )(x) at σ∗ -a.e. x ∈ ∂∗ Ω, (1.2.19)

finishing the proof of (1.2.4).

 
Next, pick f ∈ L11 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
. Then, for each point x ∈ Ω and ∈ {1, . . . , n},
we may write
∫
   
∂ Tjk f (x) = − ν j (y)∂yk − νk (y)∂y j (∂ b)(x − y) f (y) dσ∗ (y)
∂∗ Ω
∫
=− ∂τ j k (y) [(∂ b)(x − y)] f (y) dσ∗ (y)
∂∗ Ω
∫
= (∂ b)(x − y)(∂τ j k f )(y) dσ∗ (y). (1.2.20)
∂∗ Ω
10 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Above, the first equality is differentiation under the integral sign, while the second
equality uses the definition of ∂τ j k (y) . In the third equality in (1.2.20) we have
integrated by parts on the boundary, as permitted by [69, Lemma 11.1.7], bearing
in mind that the function ϕ := (∂ b)(x − ·) satisfies, thanks to our assumptions on
b, all properties demanded in [69, Lemma 11.1.7, (11.1.48)]. This finishes the proof
of (1.2.20). Once (1.2.20) has been established, we may invoke [70, Theorem 1.4.2]
  κ−n.t.
and [70, Theorem 2.5.1] to conclude that ∂ Tjk f ∂Ω exists at σ∗ -a.e. point on
∂∗ Ω.
p,q
Moreover, in the case when actually f ∈ L1 (∂∗ Ω, σ∗ ) with p, q ∈ [1, ∞) we may
rely on (1.2.20), [70, Theorem 1.4.2], and [70, Theorem 2.4.1] to estimate
   
Nκ (∂ Tjk f ) q ≤ C ∂τ j k f  L q (∂∗ Ω,σ∗ ) if 1 < q < ∞, (1.2.21)
L (∂Ω,σ)
   
Nκ (∂ Tjk f ) ≤ C ∂τ j k f  L 1 (∂∗ Ω,σ∗ ) if q = 1. (1.2.22)
L 1,∞ (∂Ω,σ)

From (1.2.1) and [70, Theorem 2.4.1] we also know that for each p ∈ [1, ∞) and κ > 0
there exists a finite constant C > 0 with the property that for every f ∈ L p (∂∗ Ω, σ∗ )
we have
 
Nκ (Tjk f ) p ≤ C f  L p (∂∗ Ω,σ∗ ) if 1 < p < ∞, (1.2.23)
L (∂Ω,σ)
 
Nκ (Tjk f ) 1,∞ ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (1.2.24)
L (∂Ω,σ)

Collectively, (1.2.21)-(1.2.24) prove that the claims in (1.2.6) and the subsequent
comment hold. This takes care of item (i).
The claims made in item (ii) are consequences of (1.2.2), (1.2.3), [70, Theo-
rem 2.3.2], and [70, (2.3.25)]. On to item (iii). Granted (1.2.4), (1.2.5), (1.2.23),
p,q
(1.2.6), it follows that for each given function f ∈ L1 (∂∗ Ω, σ∗ ) with p, q ∈ (1, ∞)
we may, thanks to (1.2.9), employ [69, Proposition 11.3.2] for the choice u := Tjk f
p,q
in Ω, to conclude that Tjk f ∈ L1 (∂∗ Ω, σ∗ ) and, taking into account (1.2.21), write
   
Tjk f  L p, q (∂∗ Ω,σ∗ ) ≤ C Nκ u L p (∂∗ Ω,σ∗ ) + Nκ (∇u) L q (∂∗ Ω,σ∗ )
1
 
≤ C f  L p (∂∗ Ω,σ∗ ) + C ∂τ j k f  L q (∂∗ Ω,σ∗ )

≤ C f  L p, q (∂∗ Ω,σ∗ ), (1.2.25)

1

for some constant C ∈ (0, ∞) independent of f . From this, the claims about the
operator in (1.2.10)-(1.2.11) are clear.
To justify (1.2.13), fix r, s ∈ {1, . . . , n} along with f as in (1.2.12), for some
 
p, q ∈ (1, ∞). In particular, f ∈ L11 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
, hence (1.2.20) holds at points
x ∈ Ω for each ∈ {1, . . . , n}. If we now define u := Tjk f ∈ 𝒞 N −1 (Ω), then
from (1.2.1), (1.2.20), (1.2.12), and [70, (2.4.8)] we see that the memberships in
[69, Proposition 11.3.2, (11.3.24)] are presently satisfied. Also, (1.2.4) and (1.2.5)
guarantee that the condition imposed on the nontangential traces in [69, Propo-
1.2 “Tangential” Singular Integral Operators 11

sition 11.3.2, (11.3.25)] holds. As such, [69, Proposition 11.3.2] is applicable to

the current function u. Together with (1.2.4), (1.2.20), [70, Theorem 2.5.1], [70,
(2.5.19)], and (1.2.3), formula [69, (11.3.26)] then implies that at σ∗ -a.e. point
x ∈ ∂∗ Ω we have

∂τr s (Tjk f )(x)

 κ−n.t.  κ−n.t.  κ−n.t.
= ∂τr s u ∂Ω (x) = νr (x) (∂s u) ∂Ω (x) − νs (x) (∂r u) ∂Ω (x)

1  
= νr (x)∂
s b ν(x) (∂τ j k f )(x)
2i
∫
+ lim+ νr (x)(∂s b)(x − y)(∂τ j k f )(y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

1  
− νs (x)∂r b ν(x) (∂τ j k f )(x)
2i
∫
− lim+ νs (x)(∂r b)(x − y)(∂τ j k f )(y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
∫
 
= lim+ νr (x)(∂s b)(x − y) − νs (x)(∂r b)(x − y) (∂τ j k f )(y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

= Tr#s (∂τ j k f )(x), (1.2.26)

proving (1.2.13). This takes care of item (iii).

Consider now the claims made in the first half of item (iv). To set the stage, fix
two exponents p, p ∈ (1, ∞) with 1/p + 1/p = 1, and consider the operator
 
T 
 
# : L p (∂ Ω, σ ) −→ L p (∂ Ω, σ ) defined by T # f , g := − f , T g
jk −1 ∗ ∗ −1 ∗ ∗ jk jk
p  p ∗ p
for all f ∈ L−1 (∂∗ Ω, σ∗ ) = L1 (∂∗ Ω, σ∗ ) and g ∈ L1 (∂∗ Ω, σ∗ )
(1.2.27)

where the angled brackets ·, · stand for the duality pairing between the space
p  p ∗ p
L−1 (∂∗ Ω, σ∗ ) = L1 (∂∗ Ω, σ∗ ) and L1 (∂∗ Ω, σ∗ ). Thanks to (A.0.136) and (1.2.11)
it follows that T
# is a well-defined, linear and bounded operator. To proceed, let us
jk
denote by E the mapping extending functions originally defined on ∂∗ Ω by zero to
the entire topological boundary ∂Ω. Then, since the transpose of Tjk
# in (1.2.7) is the

operator −Tjk in (1.2.8), we conclude from (1.2.27), and [69, (11.8.4), (11.8.5)] that
T# is the unique extension of the composition T # ◦ E (where T # is as in (1.2.7)) to
jk jk jk
12 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

a continuous operator on L−1 (∂∗ Ω, σ∗ ). If we now re-denote T

p # simply as T # , the
jk jk
claims in the first half of item (iv) follow. Finally, the claims in the second half of
item (iv) may be justified in a similar fashion. 

We continue by considering the issue of integration by parts on the boundary,

inside a principal value singular integral on a UR set.

Proposition 1.2.2 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set whose

boundary is a UR set; in particular, Ω is a set of locally finite perimeter. Denote
by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω and
abbreviate σ∗ := H n−1 ∂∗ Ω.
Consider a complex-valued function b ∈ 𝒞 N (Rn \ {0}), where N = N(n) ∈ N is
a sufficiently large number, which is even and positive homogeneous of degree 2 − n.
Also, fix some arbitrary indices j, k ∈ {1, . . . , n} and introduce the singular integral
operator
∫
Bg(x) := b(x − y)g(y) dσ∗ (y) at σ∗ -a.e. x ∈ ∂∗ Ω,
∂∗ Ω (1.2.28)
 
for every function g ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−2 .

Then
   
for each f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
with ∂τ j k f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−2
(1.2.29)
one has B(∂τ j k f ) = Tjk f at σ∗ -a.e. point on ∂∗ Ω,

and

r (∂ Ω, σ ) ∩ L 1 ∂ Ω, σ∗ (x)

for each g ∈ Lloc ∗ ∗ ∗ 1+ |x | n−2
with r ∈ (1, ∞)
(1.2.30)
one has ∂τ j k (Bg) = Tjk
# g at σ -a.e. point on ∂ Ω.
∗ ∗

 
Proof For each g ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−2
define
∫
ℬg(x) := b(x − y)g(y) dσ∗ (y), ∀x ∈ Ω. (1.2.31)
∂∗ Ω

Note that the integrability condition on g ensures that the integral in the right-hand
side of (1.2.31) is absolutely convergent for each x ∈ Ω. Thus, ℬg is a well-defined
function in Ω. To proceed, recall the operator Tjk from (1.2.1) and fix a function f
as in (1.2.29) along with an arbitrary point x ∈ Ω. Since ϕ := b(x − ·) satisfies all
demands in [69, Lemma 11.1.7, (11.1.48)] (relative to the present f ), we may use
[69, (11.1.49) in Lemma 11.1.7] to compute
1.2 “Tangential” Singular Integral Operators 13
∫
Tjk f (x) = ∂τk j (y) b(x − y) f (y) dσ∗ (y)
∂∗ Ω
∫
= b(x − y)(∂τ j k f )(y) dσ∗ (y)
∂∗ Ω

= ℬ(∂τ j k f )(x). (1.2.32)

Having fixed some κ > 0, we may now rely on (1.2.28), (1.2.32), the last part of
[70, Proposition 2.5.39] (also bearing in mind [68, Proposition 8.8.4]), and (1.2.4),
to write
∫
  κ−n.t.
B(∂τ j k f )(x) = b(x − y)(∂τ j k f )(y) dσ∗ (y) = ℬ(∂τ j k f ) ∂Ω (x)
∂∗ Ω
  κ−n.t.
= Tjk f ∂Ω
(x) = Tjk f (x) for σ∗ -a.e. x ∈ ∂∗ Ω. (1.2.33)

This establishes (1.2.29).

Moving on, fix a function g as in (1.2.30) and observe that

Bg ∈ Lloc
1
(∂∗ Ω, σ∗ ) and Tjk
#
g ∈ Lloc
r
(∂∗ Ω, σ∗ ), (1.2.34)

thanks to [68, (7.8.5)] and [70, (2.3.17)]. Also, for each ϕ ∈ 𝒞1c (Rn ) we have
∫ ∫ ∫
Bg(x)(∂τk j ϕ)(x) dσ∗ (x) = b(x − y)g(y) dσ∗ (y) (∂τk j ϕ)(x) dσ∗ (x)
∂∗ Ω ∂∗ Ω ∂∗ Ω
∫ ∫
= b(y − x)(∂τk j ϕ)(x) dσ∗ (x) g(y) dσ∗ (y)
∂∗ Ω ∂∗ Ω
∫
 
= B(∂τk j ϕ) (y)g(y) dσ∗ (y)
∂∗ Ω
∫
 
=− Tjk ϕ ∂∗ Ω (y)g(y) dσ∗ (y)
∂∗ Ω
∫
= ϕ(x)(Tjk
#
g)(x) dσ∗ (x). (1.2.35)
∂∗ Ω

Above, the first and third equalities come from (1.2.28), the second equality is a
consequence of Fubini’s Theorem and the fact that the function b is even, the fourth
equality is seen from (1.2.29) applied with f := ϕ ∂∗ Ω and the roles of j, k reversed
(cf. [69, (11.1.5)] which accounts for the minus sign), and the fifth equality is implied
by [70, (2.3.26)], (1.2.2), (1.2.3).
At this stage, from (1.2.34), (1.2.35), and (A.0.121) we conclude that the claim
in (1.2.30) is true. The proof of Proposition 1.2.2 is now complete. 
14 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

For future reference, it is useful to specialize Proposition 1.2.2 to the scale of

ordinary Lebesgue spaces in the manner made precise in the corollary below.

Corollary 1.2.3 Retain the assumptions on the set Ω and the function b made in
Proposition 1.2.2. Fix some arbitrary indices j, k ∈ {1, . . . , n} along with two
integrability exponents, p ∈ (1, ∞) and q ∈ (1, n − 1), then define
1 
1 −1
q∗ := q − n−1 . (1.2.36)

Finally, introduce the singular integral operator

∫
Bg(x) := b(x − y)g(y) dσ∗ (y) at σ∗ -a.e. x ∈ ∂∗ Ω,
∂∗ Ω (1.2.37)
for every function g ∈ L q (∂∗ Ω, σ∗ ).

Then
q ∗,q
B : L q (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.2.38)

is a well-defined, linear and bounded operator, with the property that

p,q
for each function f ∈ L1 (∂∗ Ω, σ∗ ) one has
(1.2.39)
B(∂τ j k f ) = Tjk f at σ∗ -a.e. point on ∂∗ Ω.

and such that

for every function g ∈ L q (∂∗ Ω, σ∗ ) one has
(1.2.40)
∂τ j k (Bg) = Tjk
# g at σ -a.e. point on ∂ Ω.
∗ ∗

Moreover, with prime indicating the Hölder conjugate exponent and with the
superscript star defined as in (1.2.36), it follows that5
∗
the transpose of B : L q (∂∗ Ω, σ∗ ) → L q (∂∗ Ω, σ∗ ) is
∗   (1.2.41)
the operator B : L (q ) (∂∗ Ω, σ∗ ) → L q (∂∗ Ω, σ∗ ).
Finally, in the case when ∂Ω is bounded, the same results are also true for every
q ∈ [n − 1, ∞), this time taking q∗ ∈ (1, ∞) arbitrary (and unrelated to q).

Proof All claims, except for (1.2.41), follow from Proposition 1.2.2, bearing in mind
[70, (2.5.549)] and the Fractional Integration Theorem (cf. [68, (7.8.7)]). Finally, the
claim in (1.2.41) is a direct consequences of the Fractional Integration Theorem and
Fubini’s Theorem. 
Retain the assumptions on the set Ω and the function b made in Proposition 1.2.2,
and recall the operator ℬ from (1.2.31). We aim to extend the action of this integral
operator to negative off-diagonal Sobolev spaces of the following sort. Fix some
 ∗
5 note that (q ∗ ) belongs to (1, n − 1) and satisfies (q ∗ ) = q
1.2 “Tangential” Singular Integral Operators 15

integrability
 exponent p ∈ (1, n − 1) along with some q ∈ (1, ∞), and denote by
p ∈ n−1 , ∞ and q  ∈ (1, ∞) their Hölder conjugate exponents. For each functional
n−2
p,q  p,q ∗
f ∈ L−1 (∂∗ Ω, σ∗ ) = L1 (∂∗ Ω, σ∗ ) we define
 
ℬ f (x) := p , q 
L1 (∂∗ Ω,σ∗ )
(b(x − ·) ∂∗ Ω, f p, q
L−1 (∂∗ Ω,σ∗ )
∫
= b(x − y) f0 (y) dσ∗ (y)
∂∗ Ω
∫
+ ∂τ j k (y) [b(x − y)] f jk (y) dσ∗ (y), ∀x ∈ Ω, (1.2.42)
∂∗ Ω

where the functions f0 ∈ L p (∂∗ Ω, σ∗ ), and f jk ∈ L q (∂∗ Ω, σ∗ ) with 1 ≤ j < k ≤ n,

are associated with the given functional f as in [69, Proposition 11.8.6].
In this regard, there are three basic remarks we wish to make. First, the version
of [69, Lemma 11.1.5] for off-diagonal Sobolev spaces implies (also taking into
account [68, Lemma 7.2.1]) that
p,q
b(x − ·) ∂∗ Ω ∈ L1 (∂∗ Ω, σ∗ ), ∀x ∈ Ω. (1.2.43)

Second, the final expression in (1.2.42) is unequivocally

 defined
 even though the
functional f does not determine the functions f0, ( f jk )1≤ j,k ≤n uniquely. Third, as
is apparent from (1.2.42),
p,q
ℬ f ∈ 𝒞0 (Ω) for each f ∈ L−1 (∂∗ Ω, σ∗ ). (1.2.44)

Proposition 1.2.4 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set whose

boundary is a UR set, and abbreviate σ∗ := H n−1 ∂∗ Ω, σ := H n−1 ∂Ω. Consider
a complex-valued function b ∈ 𝒞 N (Rn \ {0}), where N = N(n) ∈ N is a sufficiently
large number, which is even and positive homogeneous of degree 2 − n. Finally, fix
  −1
q ∈ (1, n − 1) and define q∗ := 1/q − 1/(n − 1) so, in particular, if (q∗ ) denotes
∗ ∗ 
the Hölder conjugate exponent of q then (q ) ∈ (1, n − 1).
In this setting, introduce two integral operators, acting on each f ∈ L q (∂∗ Ω, σ∗ )
according to
∫
ℬ f (x) := b(x − y) f (y) dσ∗ (y) for each x ∈ Ω, and (1.2.45)
∂∗ Ω
∫
B f (x) := b(x − y) f (y) dσ∗ (y) for σ∗ -a.e. x ∈ ∂∗ Ω. (1.2.46)
∂∗ Ω

Then, in relation to these, the following statements are true.

(i) The operator B induces well-defined, linear, and bounded mappings

16 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

q ∗,q
B : L q (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ), (1.2.47)
q,q ∗ q∗
B : L−1 (∂∗ Ω, σ∗ ) −→ L (∂∗ Ω, σ∗ ), (1.2.48)

which act in a compatible fashion with one another.

(ii) The (real) transpose of the operator B in (1.2.47) is the operator B in (1.2.48)
with q replaced by (q∗ ).
(iii) The operator B from (1.2.48) further extends to a linear and bounded mapping
q, p ∗
B : L−1 (∂∗ Ω, σ∗ ) −→ L q (∂∗ Ω, σ∗ ) + L p (∂∗ Ω, σ∗ ) (1.2.49)

for each p ∈ (1, ∞).

(iv) Fix κ > 0. Then for each p ∈ (1, ∞) one has, in a quantitative fashion,
∗ q, p
Nκ (ℬ f ) ∈ L q (∂Ω, σ) + L p (∂Ω, σ), ∀ f ∈ L−1 (∂∗ Ω, σ∗ ). (1.2.50)

In particular, there exists a finite constant C = C(Ω, b, q, κ) > 0 such that

 
Nκ (ℬ f ) q ∗ ≤ C f  L q, q ∗ (∂ Ω,σ )
L (∂Ω,σ) −1 ∗ ∗
(1.2.51)
q,q ∗
for all f ∈ L−1 (∂∗ Ω, σ∗ ).
q, p
(v) For each κ > 0, each p ∈ (1, ∞), and each f ∈ L−1 (∂∗ Ω, σ∗ ), the nontangential
κ−n.t.
pointwise trace ℬ f ∂Ω exists at σ∗ -a.e. point on ∂∗ Ω and, in fact, with B f
considered in the sense of (1.2.49) one has
κ−n.t.

ℬf = B f at σ∗ -a.e. point on ∂∗ Ω. (1.2.52)

∂Ω

(vi) For each κ > 0 there exists a finite constant C = C(Ω, b, q, κ) > 0 such that for
each function f ∈ L q (∂∗ Ω, σ∗ ) one has
 
Nκ (∇ℬ f ) q ≤ C f  L q (∂∗ Ω,σ∗ ) (1.2.53)
L (∂Ω,σ)

and κ−n.t.

(∇ℬ f ) exists σ∗ -a.e. on ∂∗ Ω. (1.2.54)

∂Ω
(vii) If ∂Ω is bounded then the same results in items (i)-(vi) are also true for every
q ∈ [n − 1, ∞), this time taking q∗ ∈ (1, ∞) arbitrary (and unrelated to q).

Proof The claim about (1.2.47) has been already noted in (1.2.38). Also, from the
Fractional Integration Theorem (cf. [68, (7.8.7)]) we know that
∗
B : L q (∂∗ Ω, σ∗ ) −→ L q (∂∗ Ω, σ∗ ) (1.2.55)

is well defined, linear and bounded. If q  denotes the Hölder conjugate exponent
of q, then Fubini’s Theorem shows (bearing in mind that b is even) that the (real)
transpose of (1.2.55) is the operator
1.2 “Tangential” Singular Integral Operators 17
∗  
B : L (q ) (∂∗ Ω, σ∗ ) −→ L q (∂∗ Ω, σ∗ ) (1.2.56)

whereas the (real) transpose of (1.2.47) is the operator

(q ∗ ),q 
B : L−1 (∂∗ Ω, σ∗ ) −→ L q (∂∗ Ω, σ∗ ). (1.2.57)
(q ∗ ),q ∗ 
Recall from [69, (11.8.30)] that L−1 (∂∗ Ω, σ∗ ) contains L (q ) (∂∗ Ω, σ∗ ), and the
latter space is dense in the former. Since the operators (1.2.47), (1.2.55) are com-
patible with each other, it follows that the operators (1.2.56), (1.2.57) also act in a
coherent fashion. The bottom line is that the operator B in (1.2.56) extends uniquely,
(q ∗ ),q
in a liner and bounded manner, to the larger space L−1 (∂∗ Ω, σ∗ ), where it agrees
with B . Let us also remark that if q is replaced by (q∗ ) then q  becomes the Hölder
conjugate exponent of (q∗ ), i.e., q∗ . These considerations simultaneously prove that
the operator (1.2.48) is well defined, linear, bounded, acting in a compatible fashion
with (1.2.47), and also that the (real) transpose of B in (1.2.47) is B in (1.2.48) with q
replaced by (q∗ ). The argument so far takes care of the claims made in items (i)-(ii).
Consider next the claim in item (iii). First, from (1.2.55), (A.0.123), and (1.2.40)
observe that

q ,p
Bφ ∈ L1 o o (∂∗ Ω, σ∗ ), ∀φ ∈ Lipc (∂Ω). (1.2.58)
n−2 <qo <∞
n−1

1<po <∞

q, p  q, p ∗
In particular, if p ∈ (1, ∞) and f ∈ L−1 (∂∗ Ω, σ∗ ) = L1 (∂∗ Ω, σ∗ ) , then for each
given function φ ∈ Lipc (∂∗ Ω) it follows that

the pairing L q, p (∂∗ Ω,σ∗ ) f , B φ L q, p (∂ Ω,σ ) is well defined. (1.2.59)
−1 1 ∗ ∗

Assuming f0 ∈ L q (∂∗ Ω, σ∗ ), and f jk ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ j < k ≤ n, are

associated with the given functional f as in [69, Proposition 11.8.6], further work
on said pairing yields the following sequence of equalities:
18 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

q, p
L−1 (∂∗ Ω,σ∗ ) f ,B φ L1
q , p 
(∂∗ Ω,σ∗ )
∫  ∫
= f0 Bφ dσ∗ + f jk ∂τ j k (Bφ) dσ∗
∂∗ Ω 1≤ j<k ≤n ∂∗ Ω

∫  ∫
= (B f0 )φ dσ∗ + #
f jk Tjk φ dσ∗
∂∗ Ω 1≤ j<k ≤n ∂∗ Ω

∫  ∫
 
= (B f0 )φ dσ∗ − Tjk f jk φ dσ∗
∂∗ Ω 1≤ j<k ≤n ∂∗ Ω

∫   
= B f0 − Tjk f jk φ dσ∗ . (1.2.60)
∂∗ Ω 1≤ j<k ≤n

Above, the first equality uses (1.2.59) and [69, Proposition 11.8.6], the second equal-
ity is based on (1.2.41) and (1.2.40), the third equality is a consequence of item (ii) in
Proposition 1.2.1, while the final equality is plain algebra. This establishes (1.2.60)
which, in concert with (1.2.56) and the first claim in item (ii) of Proposition 1.2.1,
goes to show that
q, p
the operator whose action on each f ∈ L−1 (∂∗ Ω, σ∗ ) is defined via

 f := B f0 −
B Tjk f jk ,
1≤ j<k ≤n
(1.2.61)
assuming the functions f0 ∈ L q (∂∗ Ω, σ∗ ) and f jk ∈ L p (∂∗ Ω, σ∗ ) with
1 ≤ j < k ≤ n are associated with the given functional f as in [69,
Proposition 11.8.6], is actually a well-defined, linear, and bounded
q, p ∗
mapping from L−1 (∂∗ Ω, σ∗ ) into L q (∂∗ Ω, σ∗ ) + L p (∂∗ Ω, σ∗ ).

As far as the claim in item (iii) is concerned, there remains to observe that the
operator B just defined agrees with B from (1.2.48). However, if f ∈ L q, p (∂∗ Ω, σ∗ )
−1
and φ ∈ Lipc (∂∗ Ω) are arbitrary, based on (1.2.60)-(1.2.61) and the manner in which
B has been defined in (1.2.48) we may write
∫

 f )φ dσ∗ = q, q ∗
(B f , B φ L q,(q ∗ ) (∂ Ω,σ )
L (∂ Ω,σ )
−1 ∗ ∗ ∗ ∗
∂∗ Ω 1

∫

= ∗ 
L (q ) (∂∗ Ω,σ∗ ) B f, φ ∗
L q (∂∗ Ω,σ∗ ) = (B f )φ dσ∗, (1.2.62)
∂∗ Ω

so the desired conclusion now follows with the help of [68, Corollary 3.7.3].
Next, the claim in item (iv) is seen from (1.2.42), [69, Proposition 11.8.6], [70,
Theorem 2.4.1], and [70, Proposition 2.5.39]. Also, the boundary trace formula
claimed in item (v) is a consequence of (1.2.42), [70, Theorem 2.4.1], [70, Proposi-
tion 2.5.39], [68, Proposition 8.8.4], and (1.2.4), bearing in mind the coincidence of
1.2 “Tangential” Singular Integral Operators 19

B with the operator B from (1.2.61). The claims in item (vi) are direct consequences
of [70, Theorem 2.4.1] and [70, Theorem 2.5.1]. Finally, when ∂Ω is bounded, the
addenda [68, (7.8.14)-(7.8.15)] to the Fractional Integration Theorem are in effect,
and this ultimately accounts for the claim in made in item (vii) of the statement. 

The following technical result is the basic ingredient in the proof of Proposi-
tion 1.2.6, stated a little later.

Lemma 1.2.5 Let Σ ⊆ Rn (where n ∈ N with n ≥ 2) be a closed Ahlfors regular set

and denote σ := H n−1 Σ. Fix ε ∈ (0, 1) and suppose b ∈ 𝒞1 (Rn \ {0}) is a function
with the property that there exists C0 ∈ (0, ∞) such that

|x| 2−n+ε if |x| ≥ 1,
|b(x)| ≤ C0 for all x ∈ Rn \ {0}, (1.2.63)
|x| 2−n−ε if |x| ≤ 1,

and
|(∇b)(x)| ≤ C0 |x| 1−n for all x ∈ Rn \ {0}. (1.2.64)
Then for each there exists a constant C ∈ (0, ∞), which depends only on Σ, n, C0 ,
and ε, such that
∫
|b(x − y) − b∗ (−y)| C
dσ(x) ≤ for all y ∈ Rn, (1.2.65)
Σ 1 + |x| n 1 + |y| n−1−ε

where b∗ := b · 1Rn \B(0,1) in Rn \ {0}.

Proof We split our discussion in a number of cases.

Case 1: |y| ≤ 1. Pick y ∈ Rn with |y| ≤ 1. Then b∗ (−y) = 0 and the term in the
left-hand side of (1.2.65) may be estimated as follows:
∫ ∫ ∫
|b(x − y)| |b(x − y)| |b(x − y)|
dσ(x) = dσ(x) + dσ(x)
Σ 1 + |x| 1 + |x| 1 + |x| n
n n
x ∈Σ x ∈Σ
|x−y | ≤1 |x−y |>1
∫ ∫
dσ(x) dσ(x)
≤ C0 + C0 ≤C
|x − y| n−2+ε 1 + |x| n−ε
x ∈Σ x ∈Σ
|x−y | ≤1 |x−y |>1

C
≤ . (1.2.66)
1 + |y| n−1−ε
Above, we have used (1.2.63), the two formulas in [68, (7.2.5)] (with μ := σ and ρ
the Euclidean distance in Rn ), and the fact that |y| ≤ 1. In particular, the last constant
C in (1.2.66) is positive, finite, and depends only on C0 , n, and the Ahlfors regularity
constants of Σ. This proves (1.2.65) in the current case.
Case 2: |y| > 1. Pick y ∈ Rn with |y| > 1. First suppose x ∈ Σ is such that 2|x| ≤ |y|.
Applying the Mean Value Theorem and (1.2.64) we obtain
20 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

2n−1
|b(x − y) − b∗ (−y)| ≤ C0 |x| sup |ξ − y| 1−n ≤ C0 |x| (1.2.67)
ξ ∈[0,x] |y| n−1

since, for every ξ ∈ [0, x] we have

|y| ≤ |ξ − y| + |ξ | ≤ |ξ − y| + |x| ≤ |ξ − y| + |y|/2, (1.2.68)

hence |y|/2 ≤ |ξ − y|. Consequently,

∫
|b(x − y) − b∗ (−y)|
dσ(x)
1 + |x| n
x ∈Σ
2 |x | ≤ |y |
∫ ∫
|x| |x|
≤C dσ(x) + C dσ(x)
|y| n−1 (1 + |x| n ) |y| n−1 (1 + |x| n )
x ∈Σ x ∈Σ
|y | ≥2 |x | ≥1 |x | ≤1/2

=: I + II. (1.2.69)

Then
∫
dσ(x) C C
II ≤ C ≤ ≤ (1.2.70)
|y| n−1 |y| n−1 1 + |y| n−1−ε
x ∈Σ
|x | ≤1/2

with C ∈ (0, ∞) dependent only on C1 , n, and the Ahlfors regularity constants of Σ.

To estimate I we first observe that there exists C = C(n, ε) ∈ (0, ∞) such that

1 C|y| ε
≤ if x ∈ Σ and |y| ≥ 2|x| ≥ 1. (1.2.71)
1 + |x| n−1 1 + |x| n−1+ε
Then (1.2.71) and the second formula in [68, (7.2.5)] imply
∫
C dσ(x)
I≤
|y| n−1 1 + |x| n−1
x ∈Σ
|y | ≥2 |x | ≥1
∫
C dσ(x) C
≤ ≤ (1.2.72)
|y| n−1−ε |x| n−1+ε 1 + |y| n−1−ε
x ∈Σ
|x | ≥1/2

for some C ∈ (0, ∞) depending only on C0 , n, the Ahlfors regularity constants of Σ,

and ε.
Second, suppose x ∈ Σ is such that 2|x| > |y| while maintaining the assumption
|y| > 1. Then (1.2.63), the second formula in [68, (7.2.5)], and the fact that |y| > 1
imply
1.2 “Tangential” Singular Integral Operators 21
∫ ∫
|b∗ (−y)| C0 dσ(x)
dσ(x) ≤
1 + |x| n |y| n−2−ε |x| n
x ∈Σ x ∈Σ
2 |x |> |y | 2 |x |> |y |

1 1 C
≤C = , (1.2.73)
|y| n−2−ε |y| |y| n−1−ε

where C ∈ (0, ∞) depends only on C0 , n, and the Ahlfors regularity constants of Σ.

Furthermore, invoking again (1.2.63), we may write
∫ ∫
|b(x − y)| |b(x − y)|
dσ(x) = dσ(x)
1 + |x| n 1 + |x| n
x ∈Σ x ∈Σ
2 |x |> |y | 2 |x |> |y |> |x |/2
∫
|b(x − y)|
+ dσ(x)
1 + |x| n
x ∈Σ
|y | ≤ |x |/2

:= III + IV. (1.2.74)

Decompose
∫ ∫
|b(x − y)| |b(x − y)|
III = dσ(x) + dσ(x)
1 + |x| n 1 + |x| n
x ∈Σ x ∈Σ
2 |x |> |y |> |x |/2 2 |x |> |y |> |x |/2
|x−y |<1 |x−y | ≥1
∫
dσ(x)
≤ C0
|x − y| n−2+ε |(1 + |x| n )
x ∈Σ
2 |x |> |y |> |x |/2
|x−y |<1
∫
dσ(x)
+ C0
|x − y| n−2−ε |(1 + |x| n )
x ∈Σ
2 |x |> |y |> |x |/2
|x−y | ≥1

:= IIIa + IIIb . (1.2.75)

Observe that whenever x ∈ Σ and 2|x| > |y| > |x|/2 we have |x−y| ≤ |x|+|y| ≤ 3|y|
which further gives
∫ ∫
dσ(x) C dσ(x)
IIIa ≤ C0 ≤
|x − y| n−2+ε |x| n |y| n |x − y| n−2+ε
x ∈Σ x ∈Σ
2 |x |> |y |> |x |/2 |x−y |<3 |y |

C C C
≤ · |y| 1−ε = ≤ , (1.2.76)
|y| n |y| n−1+ε |y| n−1−ε
22 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

for some C ∈ (0, ∞) depending only on C0 , n, ε, and the Ahlfors regularity constants
of Σ. For the third inequality in (1.2.76) we have used the first formula in [68, (7.2.5)],
while the last inequality in (1.2.76) is due to the fact that we are currently assuming
|y| > 1. Likewise,
∫ ∫
dσ(x) C dσ(x)
IIIb ≤ C0 ≤
|x − y| n−2−ε |x| n |y| n |x − y| n−2−ε
x ∈Σ x ∈Σ
2 |x |> |y |> |x |/2 |x−y |<3 |y |

C C
≤ · |y| 1+ε = . (1.2.77)
|y| n |y| n−1−ε
There remain to estimate IV. To do so, we further decompose
∫ ∫
|b(x − y)| |b(x − y)|
IV = dσ(x) + dσ(x)
1 + |x| n 1 + |x| n
x ∈Σ x ∈Σ
|y | ≤ |x |/2 |y | ≤ |x |/2
|x−y |<1 |x−y | ≥1

:= IVa + IVb . (1.2.78)

Note that if x ∈ Σ and |y| ≤ |x|/2 we have |x| ≤ |x − y| + |y| ≤ |x − y| + |x|/2,

hence |x|/2 ≤ |x − y|. Together with the second formula in [68, (7.2.5)], and the fact
that |y| > 1 this allows us to write
∫ ∫
dσ(x) dσ(x)
IVa ≤ C0 ≤ C
|x − y| n−2+ε |x| n |x| 2n−2+ε
x ∈Σ x ∈Σ
|y | ≤ |x |/2 |y | ≤ |x |/2

C C
≤ ≤ , (1.2.79)
|y| n−1+ε |y| n−1−ε

for some C ∈ (0, ∞) depending only on C0 , n, ε, and the Ahlfors regularity constants
of Σ. Similarly,
∫ ∫
dσ(x) dσ(x)
IVb ≤ C0 ≤C
|x − y| n−2−ε |x| n |x| 2n−2−ε
x ∈Σ x ∈Σ
|y | ≤ |x |/2 |y | ≤ |x |/2

C
≤ . (1.2.80)
|y| n−1−ε

Now estimate (1.2.65) in the case when |y| > 1 follows from (1.2.69)-(1.2.70)
combined with (1.2.72)-(1.2.80). 

The proposition below abstractly models the behavior of the modified single layer
potential operator, something we will consider later down the road.
1.2 “Tangential” Singular Integral Operators 23

Proposition 1.2.6 Let Σ ⊆ Rn (where n ∈ N with n ≥ 2) be a closed Ahlfors regular

set and denote σ := H n−1 Σ. Suppose b ∈ 𝒞1 (Rn \ {0}) is a function for which
there exist ε ∈ (0, 1) along with C0 ∈ (0, ∞) such that

|x| 2−n+ε if |x| ≥ 1,
|b(x)| ≤ C0 for all x ∈ Rn \ {0}, (1.2.81)
|x| 2−n−ε if |x| ≤ 1,

and
|(∇b)(x)| ≤ C0 |x| 1−n for all x ∈ Rn \ {0}. (1.2.82)
Then the operator
 σ(x)  σ(x)
Bmod : L 1 Σ, −→ L 1
Σ, (1.2.83)
1 + |x| n−1−ε 1 + |x| n
 
defined for each function f ∈ L 1 Σ, 1+ |xσ(x)
| n−1−ε
as
∫
 
Bmod f (x) := b(x − y) − b∗ (−y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (1.2.84)
Σ

where b∗ := b · 1Rn \B(0,1) in Rn \ {0}, is well defined, linear, and bounded.

 
Proof Let f ∈ L 1 Σ, 1+ |xσ(x)
| n−1−ε
be arbitrary. Making use of Fubini’s Theorem and
Lemma 1.2.5 we may write
∫ ∫
|b(x − y) − b∗ (−y)|| f (y)|
dσ(y) dσ(x)
Σ Σ 1 + |x| n
∫ ∫
|b(x − y) − b∗ (−y)|
= dσ(x) | f (y)| dσ(y)
Σ Σ 1 + |x| n
∫
| f (y)|
≤C dσ(y) < +∞, (1.2.85)
Σ 1 + |y| n−1−ε

for some C ∈ (0, ∞) independent of f . This goes to show that the integral defining
Bmod f (x) in (1.2.84) is absolutely convergent for σ-a.e. point x ∈ Σ. Hence, Bmod is
well defined in the context of (1.2.83), and the estimate derived in (1.2.85) gives
∫ ∫
|(Bmod f )(x)| | f (y)|
dσ(x) ≤ C dσ(y). (1.2.86)
Σ 1 + |x| n Σ 1 + |y|
n−1−ε

This proves that Bmod is indeed a bounded operator. Finally, its linearity is clear from
what we have proved so far and (1.2.84). 
24 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

1.3 A First Look at Layer Potential Operators

Here we shall introduce volume and boundary integral operators of layer type as-
sociated with a given open set and a weakly elliptic system. Specifically, suppose
n ∈ N with n ≥ 2 has been fixed. For some M ∈ N, consider a coefficient tensor

αβ
A = ar s 1≤r,s ≤n (1.3.1)
1≤α,β ≤M

with complex entries, with the property that the M × M second-order system (as
always, the summation convention over repeated indices is in effect)

αβ
L = L A := ar s ∂r ∂s 1≤α ≤M (1.3.2)
1≤β ≤M

is weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Also, denote by
A the transpose of A, i.e., the coefficient tensor

βα
A = asr 1≤r,s ≤n . (1.3.3)
1≤α,β ≤M

Let E = (Eγβ )1≤γ,β ≤M be the matrix-valued fundamental solution associated with

L as in [70, Theorem 1.4.2].
Given an arbitrary open set Ω ⊆ Rn , define the Newtonian (or volume)
potential operator associated with the system L in Ω as the mapping assigning
to each Lebesgue measurable function w : Ω → C M the expression
∫
(ΠΩ w)(x) := E(x − y)w(y) dy (1.3.4)
Ω

at points x ∈ Ω for which the above integral is absolutely convergent. Also, having
set σ := H n−1 ∂Ω, define the action of the boundary-to-domain single layer
potential operator 𝒮 on function
  σ(x) M
f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, (1.3.5)
1 + |x| n−2
according to
1.3 A First Look at Layer Potential Operators 25
∫ ∫ !
𝒮 f (x) := E(x − y) f (y) dσ(y) = Eγα (x − y) fα (y) dσ(y)
∂Ω ∂Ω 1≤γ ≤M
∫   !
= Eγ .(x − y) , f (y) dσ(y)
∂Ω 1≤γ ≤M
∫

= E L  (x − y) f (y) dσ(y)
∂Ω
∫   !

= E L  (x − y) .γ , f (y) dσ(y) for each x ∈ Ω. (1.3.6)
∂Ω 1≤γ ≤M

Above, Eγ . denotes the γ-th row of the matrix E. Also, E L  is the fundamental
solution 
  associated as in [70, Theorem 1.4.2] with the transpose system L , and
E L  . γ denotes the γ-th column of the matrix E L  . In the two-dimensional case,
the weight (1 + |x| n−2 )−1 appearing in (1.3.5) should be replaced by ln(2 + |x|), i.e.,
when n = 2 in place of (1.3.5) we shall (sometimes tacitly, but always) consider
  M
f ∈ L 1 ∂Ω, ln(2 + |x|)σ(x) . (1.3.7)

Since it is clear from (1.3.6) and [70, (1.4.20)] that

𝒮 f ∈ 𝒞∞ (Ω)
M
and L(𝒮 f ) = 0 in Ω, for each f as in (1.3.5), (1.3.8)

the operator 𝒮 may be regarded as a tool for generating a multitude of null-solutions

for the given system L in Ω. We will study this operator in detail, starting with
Theorem 1.5.1. For now, one thing that we can clarify is the decay at infinity in the
case of an exterior domain.
Lemma 1.3.1 Assume that Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is an exterior
domain and that σ := H n−1 ∂Ω is a finite measure. Consider a weakly elliptic,
second-order, homogeneous, constant (complex) coefficient M × M system in Rn .
Associate the single layer potential operator 𝒮 with Ω and L as above. Finally, define
M
u := 𝒮 f in Ω for some f ∈ L 1 (∂Ω, σ) . (1.3.9)

Then the following statements are true.

(i) Given any multi-index α ∈ N0n , one has (∂ α u)(x) = O(|x| 2−n− |α | ) as |x| → ∞
∫ −1
provided either |α| > 0, or n ≥ 3, or n = 2 and S 1 L(ξ) dH 1 (ξ) = 0.
∫
(ii) If ∂Ω f dσ = 0 then (∂ α u)(x) = O(|x| 1−n− |α | ) as |x| → ∞ for any multi-index
α ∈ N0n .
(iii) If n = 2 then
∫
lim u(x) exists (in C M ) if and only if ∂Ω f dσ = 0 if and only if
x→∞
(1.3.10)
lim u(x) = 0 if and only if u(x) = O(|x| −1 ) as |x| → ∞.
x→∞
26 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Also, in the two-dimensional setting,

∫
u(x) = E(x) f dσ + O(|x| −1 )
∂Ω
∫ ∫
ln |x|  −1
= O(1) − L(ξ) dH 1 (ξ) f dσ
4π 2 S1 ∂Ω

= O(ln |x|) as |x| → ∞, (1.3.11)

and, as a consequence,

if n = 2 then u(x) = o(ln |x|) as |x| → ∞ is equivalent to

∫ ∫ (1.3.12)
−1 1 (ξ)
S1
L(ξ) dH ∂Ω
f dσ = 0.

In particular, if n = 2 and
∫
−1
the M × M matrix L(ξ) dH 1 (ξ) is invertible (1.3.13)
S1

(which is always the case if the system L is assumed to actually satisfy the
Legendre-Hadamard strong ellipticity condition, and if M = 1 this is true if and
only if L = ∇ · A∇ for some A ∈ M20 ; cf. [70, Lemma 1.4.19] and [70,
∫ (1.4.186)])
then having u(x) = o(ln |x|) as |x| → ∞ is equivalent to having ∂Ω f dσ = 0.

Proof For each multi-index α ∈ N0n we have

∫
∂ α (𝒮 f )(x) = (∂ α E)(x − y) f (y) dσ(y) for all x ∈ Ω, (1.3.14)
∂Ω
∫
and under the additional condition that ∂Ω
f dσ = 0 we may further express
∫  
α
∂ (𝒮 f )(x) = (∂ α E)(x − y) − (∂ α E)(x) f (y) dσ(y) (1.3.15)
∂Ω

for all x ∈ Ω. The claims in items (i)-(ii) are then direct consequences of these
observations, [70, (1.4.24)], and the Mean Value Theorem. To treat item (iii), work
in the two-dimensional setting. For starters, write
1.3 A First Look at Layer Potential Operators 27
∫ ∫
 
u(x) = E(x) f (y) dσ(y) + E(x − y) − E(x) f (y) dσ(y)
∂Ω ∂Ω
 ∫ ∫
ln |x| −1
= Φ(x) − L(ξ) dH (ξ)1
f dσ + O(|x| −1 )
4π 2 S 1 ∂Ω
∫ ∫
ln |x|  −1
= O(1) − L(ξ) dH (ξ)1
f dσ + O(|x| −1 )
4π 2 S1 ∂Ω

= O(ln |x|) as |x| → ∞, (1.3.16)

where we have used [70, (1.4.22), (1.4.24)], the Mean Value Theorem, and the
fact that the function Φ : R2 \ {0} → C M×M is bounded (since it is continuous
and positive homogeneous of degree 0 in R2 \ {0}; cf. [70, (1.4.23)]). This proves
(1.3.11) which, in turn, implies (1.3.12).
The equivalences in (1.3.10) are seen from the first line in (1.3.11) and [70,
(1.4.47)]. The very last equivalence in the statement then follows from (1.3.12)
keeping in mind (1.3.13). 

Going further, strengthen the background hypotheses by assuming that the open
set Ω is of locally finite perimeter, and denote by ν = (ν1, . . . , νn ) its geometric
measure theoretic outward unit normal. In this context, we introduce the boundary-
to-domain double layer potential operator as the integral operator acting on
each function
  σ(x) M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, (1.3.17)
1 + |x| n−1
according to (recall that the summation convention over repeated indices is presently
in effect)
# ∫ $
βα
D f (x) := − νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y) (1.3.18)
∂∗ Ω
1≤γ ≤M

for every x ∈ Ω. In relation to this, we make two conventions. First,

whenever we wish to emphasize the dependence of the double
layer operator (1.3.18) on the coefficient tensor A from (1.3.1) (1.3.19)
we agree to write D A in place of D.
In contrast to the single layer potential operator (1.3.6) which is intrinsically associ-
ated with the given system L, to each coefficient tensor A belonging to the “library”
of (admissible) coefficient tensors of the system L, defined as
   αβ  
αβ
A L := A = ar s 1≤r,s ≤n : L = L A := ar s ∂r ∂s 1≤α,β ≤M , (1.3.20)
1≤α,β ≤M
28 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

there corresponds a double layer D A. In particular, the terminology a double layer,

as opposed to the double layer, is appropriate whenever referring to D associated
with a generic coefficient tensor A ∈ A L when no mention (explicit or implicit) of
the coefficient tensor A is being made.
Second, typically without any special mention,
we agree to retain the same formula for the double layer potential
(1.3.21)
operator, given in (1.3.18), at points x belonging to Rn \ Ω as well.
In terms of the conormal derivative operator (A.0.184), the double layer operator
(1.3.18) may be expressed more concisely as
∫   !
A    
D f (x) = ∂ν(y) E L  (x − y) . γ , f (y) dσ(y)
∂∗ Ω 1≤γ ≤M
∫   !
   
= ∂ν(y)
A
E L (x − y) γ . , f (y) dσ(y)
∂∗ Ω 1≤γ ≤M
∫  

= ∂ν(y)
A
E L  (x − y) f (y) dσ(y), ∀ x ∈ Ω. (1.3.22)
∂∗ Ω

Above, E L  is the fundamental

 solution
 associated as in [70, Theorem 1.4.2] with the
transpose system L  , and E L  . γ denotes the γ-th column of the matrix E L  . Also,
 
E L γ . denotes the γ-th row of the matrix E L , so the second equality in (1.3.22)
becomes a consequence of the first property in [70, (1.4.32)]. Finally, the last equality
in (1.3.22) follows from the second, keeping in mind that the conormal derivative
acts (in the variable y) as the M × M first-order system

A βα
∂ν(y) := νr (y)asr ∂ys (1.3.23)
1≤α,β ≤M

on the M × M matrix E L  (x − y) (in a natural manner which, on the algebraic side,

takes into account the ordinary multiplication of M × M matrices), and the transposed
of the resulting M × M matrix is then applied to the vector f (y) ∈ C M .
The double layer operator D may be regarded as a mechanism for generating lots
of null-solutions for the given system L in Ω since, as is apparent from (1.3.18) and
item (2) in [70, Theorem 1.4.2],

D f ∈ 𝒞∞ (Ω)
M
and L(D f ) = 0 in Ω, for each f as in (1.3.17). (1.3.24)

Indeed, given any function f as in (1.3.17), for each x ∈ Ω we may write

     
μγ
L D f (x) = a jk ∂j ∂k D f γ (x) (1.3.25)
1≤μ ≤M
∫ !
βα μγ
= νs (y)ar s ∂xr a jk (∂j ∂k Eγβ )(x − y) fα (y) dσ(y) = 0,
∂∗ Ω 1≤μ ≤M
1.3 A First Look at Layer Potential Operators 29

since the expression inside the brackets vanishes for each y ∈ ∂Ω, thanks to [70,
(1.4.21)].
It is also clear from (1.3.18) and [70, Theorem 1.4.2] that, in the case when Ω is
an exterior domain, for each function f as in (1.3.17) and each integer k ∈ N0 the
following decay condition holds:

O(|x| 1−n−k ),
∇ D f (x) =
k ∫ as |x| → ∞. (1.3.26)
O(|x| −n−k ) if ∂ Ω f dσ = 0,
∗

As we shall see in what follows, in addition to qualities displayed in (1.3.24),

(1.3.26) there are other, more specialized, properties that are characteristic to the
double layer potential operator defined in (1.3.18).
A fundamental property of the double layer potential operator D just introduced
is the ability of absorbing an arbitrary spatial derivative and eventually relocate it,
via integration by parts on the boundary, all the way to the function on which this
was applied to begin with. This is made precise in the following basic lemma.

Lemma 1.3.2 Let Ω ⊆ Rn be an open set of locally finite perimeter. Abbreviate

σ := H n−1 ∂Ω, and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω. Also, for some integer M ∈ N, consider a coefficient
αβ 
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Finally, associate with
A and Ω the double layer potential operator D as in (1.3.18), and consider a function
  M
f = ( fα )1≤α ≤M belonging to the weighted Sobolev space L11 ∂∗ Ω, 1+σ(x) |x | n−1
(cf.
(A.0.131)), i.e.,
  σ(x) M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω,
1 + |x| n−1
 σ(x) (1.3.27)
with the property that ∂τ j k fα ∈ L 1 ∂∗ Ω,
1 + |x| n−1
for all j, k ∈ {1, . . . , n} and α ∈ {1, . . . , M }.

Then, for each index ∈ {1, . . . , n} and at every x ∈ Ω, one has

#∫ $
  βα
∂ D f (x) = ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ(y) . (1.3.28)
∂∗ Ω
1≤γ ≤M

For example, if M := 1 and L := Δ, the Laplace operator in Rn , then (1.3.27)-

(1.3.28) amount to saying that for any scalar-valued function
30 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 σ(x)
f ∈ L 1 ∂∗ Ω, with the property that
1 + |x| n−1
 (1.3.29)
σ(x)
∂τ j k f ∈ L ∂∗ Ω,
1
for all j, k ∈ {1, . . . , n}
1 + |x| n−1

and any index ∈ {1, . . . , n} one has

n ∫

∂ (D f )(x) = (∂j EΔ )(x − y)(∂τ j f )(y) dσ(y) for each x ∈ Ω, (1.3.30)
j=1 ∂∗ Ω

where D is the standard boundary-to-domain harmonic double layer operator asso-

ciated with the set Ω, and EΔ is the standard fundamental solution for the Laplacian
(cf. (A.0.65)).
Proof of Lemma 1.3.2 Since the given function f is as in (1.3.17), we know from
M
(1.3.24) that D f is well defined and belongs to 𝒞∞ (Ω) . Having fixed an index
γ ∈ {1, . . . , M } and x ∈ Ω we may then compute
∫
  βα
∂ D f γ (x) = − νs (y)ar s (∂ ∂r Eγβ )(x − y) fα (y) dσ(y)
∂∗ Ω
∫
βα
=− ar s ∂τ s (y) [(∂r Eγβ )(x − y)] fα (y) dσ(y)
∂∗ Ω
∫
βα
= ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ(y). (1.3.31)
∂∗ Ω

Above, the first equality comes from (1.3.18) and differentiation under the integral
sign, while the second equality uses the definition of ∂τ s (y) and [70, (1.4.33)]. In the
third equality in (1.3.31) we have integrated by parts on the boundary, as permitted
by [69, Lemma 11.1.7], given that the function ϕ := (∂r Eγβ )(x − ·) is of class 𝒞1 in
a collar neighborhood of ∂Ω and, thanks to [70, (1.4.24)] and (1.3.27), satisfies
∫ ∫
| f ||∇ϕ| dσ < +∞, ∂τ s f |ϕ| dσ < +∞,
∂∗ Ω ∫ ∂∗ Ω
(1.3.32)
as well as | f (x)||ϕ(x)|(1 + |x|)−1 dσ(x) < +∞.
∂∗ Ω

This establishes (1.3.31), from which formula (1.3.28) now follows in view of the
arbitrariness of γ ∈ {1, . . . , M } and x ∈ Ω. 

In terms of the integral operators introduced above, we may recast the integral
representation formula from [70, Theorem 1.5.1] succinctly, as follows.

Theorem 1.3.3 Let Ω ⊆ Rn , where6 n ≥ 3, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure

6 see the last part in the statement for the case n = 2

1.3 A First Look at Layer Potential Operators 31

on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Let L be a homogeneous,
weakly elliptic, second-order M ×M system in Rn , with complex constant coefficients.
Suppose u ∈ Lloc 1 (Ω, L n ) M is a vector-valued function satisfying, for some aperture

parameter κ > 0,
  dy M
M×n
Lu ∈ L 1 Ω, and A∇u ∈ Lloc
1
(Ω, L n ) ,
1 + |y| n−2
(1.3.33)
κ−n.t. κ−n.t.
u ∂Ω and (A∇u) ∂Ω exist at σ-a.e. point on ∂nta Ω,

(with all derivatives taken in the sense of distributions, and using the piece of notation
introduced in (A.0.12)). In addition, assume the following integrability conditions
hold:
∫ ∫  
(Nκ u)(y) Nκ (A∇u) (y)
dσ(y) < ∞ and dσ(y) < ∞. (1.3.34)
∂Ω 1 + |y| 1 + |y| n−2
n−1
∂Ω

Next, recall the Newtonian potential operator ΠΩ from (1.3.4) and the single layer
potential operator 𝒮 from (1.3.6). Finally, pick a coefficient tensor A such that
L = L A and consider the potential operator D and the conormal derivative operator
∂νA associate with A and Ω as in (1.3.18) and (A.0.184), respectively.
Then at L n -a.e. point x ∈ Rn \ ∂Ω one has7

 κ−n.t.  u(x) if x ∈ Ω,
D u ∂Ω (x) − 𝒮(∂ν u)(x) + ΠΩ (Lu)(x) =
A
(1.3.35)
0 if x ∈ Rn \ Ω,

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds under the additional assumption that there exists
λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (1.3.36)
B(0,λ R)\B(0,R)

Finally, similar results are valid in the case when n = 2 provided either
∫
−1
L(ξ) dH 1 (ξ) = 0 ∈ C M×M , (1.3.37)
S1

or otherwise one assumes that ∂Ω is compact, replaces the first membership in

(1.3.33) by

7 We agree to retain the formulas for the double layer, single layer, and Newtonian potential operators
given, respectively, in (1.3.18), (1.3.6), and (1.3.4), at points x belonging to R n \ Ω as well. Here
we also allow the operator 𝒮 to act on functions originally defined only on ∂∗ Ω by extending them
by zero to the entire topological boundary ∂Ω; we shall occasionally make use of the latter option
in a tacit fashion in the future.
32 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

∫
|(Lu)(y)| ln(2 + |y|) dy < +∞, (1.3.38)
Ω
and, in the case when Ω is an exterior domain, replaces (1.3.36) by
⨏
 
|u| dL 2 = o ln1R as R → ∞. (1.3.39)
B(0,λ R)\B(0,R)

Proof This is obtained by merely recasting formulas [70, (1.5.4), (1.5.6)], in light
of (1.3.18), (1.3.6), (A.0.184), and (1.3.4). 

We wish to augment the above integral representation formula by considering

exterior domains in the two-dimensional setting. Corresponding to null-solutions of
the given system, we have the following result:

Theorem 1.3.4 Let Ω ⊆ R2 be an unbounded open set with a compact lower Ahlfors
regular boundary, and with the property that σ := H 1 ∂Ω is a doubling measure
on ∂Ω. Denote by ν the geometric measure theoretic outward unit normal to Ω. Let L
be a homogeneous, weakly elliptic, second-order M × M system in R2 , with complex
constant coefficients, and pick a coefficient tensor A such that L = L A. Suppose u is
a vector-valued function satisfying

u ∈ 𝒞∞ (Ω)
M
and Lu = 0 in Ω,
(1.3.40)
κ−n.t. κ−n.t.
u ∂Ω and (A∇u) ∂Ω exist at σ-a.e. point on ∂nta Ω.

In addition, assume that for some aperture parameter κ > 0 and some truncation
parameter ε > 0 the following integrability conditions hold:
∫ ∫
Nκε u dσ < ∞ and Nκε (A∇u) dσ < ∞, (1.3.41)
∂Ω ∂Ω

and assume that there exists λ ∈ (1, ∞) such that

⨏
|u| dL 2 = o(1) as R → ∞. (1.3.42)
B(0,λ R)\B(0,R)

Finally, consider the double layer potential operator D and the conormal derivative
operator ∂νA associated with A and Ω as in (1.3.18) and (A.0.184), respectively.
Then for each point x ∈ R2 \ ∂Ω one has

 κ−n.t.  u(x) if x ∈ Ω,
D u ∂Ω (x) − 𝒮(∂ν u)(x) =
A
(1.3.43)
0 if x ∈ R2 \ Ω.

Proof This is a direct consequence of [70, Theorem 1.5.7] used with N = 0 (in which
scenario the polynomial P vanishes identically, since its degree is ≤ 0 = −1) and
definitions. 
1.3 A First Look at Layer Potential Operators 33

For further use, let us also record here the following remarkable identity involving
the double and single layers.

Corollary 1.3.5 Let Ω ⊆ Rn , where n ≥ 2, be an Ahlfors regular domain, which is

assumed to have compact boundary if n = 2. Denote by ν the geometric measure
theoretic outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, consider a
homogeneous, weakly elliptic, second-order M × M system in Rn , with complex con-
stant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental
solution associated with L as in [70, Theorem 1.4.2].
Next, pick a coefficient tensor A such that L = L A and recall the double layer
potential operator D and the conormal derivative operator ∂νA associated with A
and Ω as in (1.3.18) and (A.0.184), respectively. Finally, recall the single layer
potential operator 𝒮 on associated with Ω and L as in (1.3.6).
Then for each point xo ∈ Ω and each index β ∈ {1, . . . , M } one has
   
D E. β (· − xo ) ∂Ω = 𝒮 ∂νA[E. β (· − xo )] in Ω. (1.3.44)

Proof This is a direct consequence of [70, Proposition 1.5.5], [70, Theorem 1.4.2],
and definitions. 

It turns out that the boundary-to-domain double layer maps constant functions in
a very specific fashion, made precise below.

Proposition 1.3.6 Let Ω ⊆ Rn , where n ≥ 2, be an open set of locally finite perimeter

with a compact boundary. Also, consider a homogeneous, weakly elliptic, second-
order M × M system L in Rn , with complex constant coefficients, and denote by D
a double layer potential operator associated with L and Ω as in (1.3.18). Then for
each constant λ ∈ C M one has

λ in Ω, if Ω is bounded,
Dλ = (1.3.45)
0 in Ω, if Ω is unbounded.

Furthermore, under the additional assumption that ∂Ω is a lower Ahlfors regular

set and that σ := H n−1 ∂Ω is a doubling measure on ∂Ω, it follows that for each
aperture parameter κ > 0 and each locally constant C M -valued function η in Ω
κ−n.t.
which is bounded and whose nontangential boundary trace η ∂Ω exists at σ-a.e.
on ∂Ω8 one has (with the convention made in (1.3.21))
⎧
⎪ η in Ω, if Ω is bounded,
⎪
⎪
⎪
⎪
 κ−n.t.  ⎪
⎨ 0 in Rn \ Ω,
⎪ if Ω is bounded,
D η ∂Ω = (1.3.46)
⎪
⎪ η − η∞ in Ω, if Ω is unbounded,
⎪
⎪
⎪
⎪
⎪ −η in Rn \ Ω, if Ω is unbounded,
⎩ ∞

8 this happens automatically if, e.g., Ω has finitely many connected components which are separated
34 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

where η∞ ∈ C M is the value of η in the (unique) unbounded connected component

of Ω.

Proof Fix λ = (λα )1≤α ≤M ∈ C M along with γ ∈ {1, . . . , M }. In a first stage,

assume that Ω is bounded and fix an arbitrary point x ∈ Ω. The idea is to apply [68,
n
Corollary 2.8.8] to a vector field F = (Fs )1≤s ≤n ∈ ℰ(Rn ) whose 𝒞∞ -singular
support is a compact subset of Ω and with the property that for each s ∈ {1, . . . , n}
its scalar component Fs coincides near the compact set Ω with
βα
−ar s (∂r Eγβ )(x − y)λα for L n -a.e. y ∈ Ω. (1.3.47)
 
Then [70, (1.4.33)] gives that divF Ω = λγ δx , where δx is the Dirac distribution
with mass at x in Ω. In view of this and (1.3.18), the Divergence Formula [68,
(2.8.57)] reduces precisely to (Dλ)γ (x) = λγ , so the version of (1.3.45) when Ω
is bounded follows on account of the arbitrariness of x ∈ Ω and γ ∈ {1, . . . , M }.
Parenthetically we wish to note that we could have also arrived to this conclusion as
a consequence of (1.3.35) used with u ≡ λ, but under stronger assumptions on the
underlying domain.
In the case when Ω is unbounded and ∂Ω is bounded, [68, Lemma 5.10.10] and
[68, (5.6.16)] imply that Ωc := Rn \ Ω is a compact set of locally finite perimeter,
satisfying ∂ ∗ (Rn \ Ω) = ∂ ∗ Ω, and whose geometric measure theoretic outward unit
normal is −ν. Having fixed an arbitrary point x ∈ Ω, apply the De Giorgi-Federer
version of the Divergence Theorem (cf. [68, Theorem 1.1.1]) to the set Ωc and a
vector field F ∈ 𝒞∞ n n which coincides in an open neighborhood O ⊆ Rn \ {x}
c (R )
of Rn \ Ω with
 
βα
O  y −→ − ar s (∂r Eγβ )(x − y)λα ∈ Cn . (1.3.48)
1≤s ≤n
 
Since (1.3.48) and [70, (1.4.33)] imply divF Ωc = 0, the Divergence Formula [68,
(1.1.8)] now simply becomes (Dλ)(x) = 0, thanks to (1.3.18). The proof of (1.3.45)
is therefore complete.
To justify (1.3.46), make the additional assumption that ∂Ω is a lower Ahlfors
regular set and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. Also, fix an arbitrary
aperture parameter κ > 0 and pick a locally constant C M -valued function η in Ω
which is bounded, and has a nontangential boundary trace. If Ω is bounded, then
the first two cases in (1.3.46) are seen directly from (1.3.35) (used with u := η). In
the case when Ω is unbounded, apply (1.3.35) to the function u := η − η∞ (which
vanishes at infinity) to obtain

 κ−n.t.  η − η∞ in Ω,
D η ∂Ω − η∞ = (1.3.49)
0 in Rn \ Ω.

From this, the version of (1.3.46) for unbounded sets follows upon invoking formula
(1.3.45), written with λ := η∞ and both for Ω as well as for Rn \ Ω (bearing in mind
that the geometric measure theoretic outward unit normal to the latter set is −ν). 
1.3 A First Look at Layer Potential Operators 35

There is also a boundary version of (1.3.45) involving a brand of principal-value

integral in which the radius of the ball excised goes to zero from within a suitably
chosen subset of (0, ∞), as described in our next proposition.

Proposition 1.3.7 Let Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) be a set of locally

finite perimeter with compact boundary. Denote by ν its geometric measure theoretic
outward unit normal and abbreviate σ := H n−1 ∂Ω. Also, for some M ∈ N, consider
αβ 
a coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the property that
1≤α,β ≤M
the M × M homogeneous second-order system L = L A associated with A in Rn as
in (1.3.2) is weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Finally,
let E L  be the matrix-valued fundamental solution associated with L  as in [70,
Theorem 1.4.2].
Then for σ-a.e. point x ∈ ∂∗ Ω one has
∫  1
+ 2 I M×M if Ω is bounded,
A
lim ∂ν(y) E L  (x − y) dσ(y) = (1.3.50)
ε→0+ − 12 I M×M if Ω is unbounded,
y ∈∂∗ Ω
|x−y |>ε

where the conormal derivative is applied to the columns of E L  , and I M×M is the
M × M identity matrix.
Moreover, under the stronger hypothesis that Ω is a Lebesgue measurable set
such that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors regular, there exists some number
N ∈ (0, ∞), depending only on n, L, and the upper Ahlfors regularity constant of
∂∗ Ω, with the property that
∫

sup ∂ν(y)
A
E L  (x − y) dσ(y) ≤ N for each x ∈ Rn . (1.3.51)
ε>0
y ∈∂∗ Ω
|x−y |>ε

Proof Denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental solution asso-

ciated with L as in [70, Theorem 1.4.2]. For each pair of indices α, γ ∈ {1, . . . , M }
define the vector field (recall that the summation convention over repeated indices is
presently in effect)
 βα 
k αγ := − ar s ∂r Eγβ 1≤s ≤n ∈ 𝒞∞ (Rn \ {0})
n
(1.3.52)

which is odd and positive homogeneous of degree 1 − n (cf. [70, Theorem 1.4.2]),
and which satisfies
βα
div k αγ = −ar s ∂r ∂s Eγβ = 0 in Rn \ {0}, (1.3.53)

thanks to [70, (1.4.33)]. Also, by virtue of the second equality in [70, (1.4.25)],
36 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫

ϑαγ := ω, k αγ (ω) dH n−1 (ω)
S n−1
∫
βα
=− ωs ar s (∂r Eγβ )(ω) dH n−1 (ω) = −δαγ . (1.3.54)
S n−1

To proceed, observe that for each y ∈ ∂ ∗ Ω and x ∈ Rn \ {y} we have


A βα
∂ν(y) E L  (x − y) = − νr (y)asr (∂s Eγβ )(x − y)
1≤α,γ ≤M
 
= ν(y), k αγ (x − y) . (1.3.55)
1≤α,γ ≤M

Granted this, all desired conclusions follow from [70, Lemma 2.5.9] and [70,
(2.5.151)]. 

Moving forward, we retain the assumptions on the M × M system L from (1.3.2)

and consider the case when the open set Ω ⊆ Rn has an Ahlfors regular boundary.
In such a scenario, [70, (2.5.549)] gives
  σ(x) M
M
L p (∂Ω, σ) → L 1 ∂Ω, whenever p ∈ [1, n − 1). (1.3.56)
1 + |x| n−2
In particular, the single layer potential operator (1.3.6) is meaningfully defined on
M
the space L p (∂Ω, σ) with p ∈ [1, n − 1). Considered as such, it follows from [70,
M
Proposition 2.5.39] (presently used with α = 1) that for each f ∈ L p (∂Ω, σ)
with p ∈ [1, n − 1) and each aperture parameter κ > 0 there exists a constant
C ∈ (0, ∞), which depends on p and κ but is independent of f , such that
 
Nκ (𝒮 f ) p ∗ ≤ C f [L p (∂Ω,σ)] M
L (∂Ω,σ)
(1.3.57)
if p ∈ (1, n − 1) and p1∗ = p1 − n−1 1
,

and  
Nκ (𝒮 f ) ≤ C f [L 1 (∂Ω,σ)] M if p = 1. (1.3.58)
L (n−1)/(n−2),∞ (∂Ω,σ)

Moreover, if ∂Ω is bounded then estimates in the same spirit hold in the range
p ∈ [n − 1, ∞). Specifically, [70, (2.5.556)-(2.5.557)] imply that, having fixed κ > 0,
for each p ∈ [n − 1, ∞) there exist finite constants, which now also depend on
diam(∂Ω), such that
M
if either n ≥ 3, or n = 2 and Ω is bounded, then for all f ∈ L p (∂Ω, σ)
 
we have Nκ (𝒮 f ) L ∞ (∂Ω,σ) ≤ Cp  f [L p (∂Ω,σ)] M if p ∈ (n − 1, ∞) and
 
Nκ (𝒮 f ) q ≤ Cq  f [L n−1 (∂Ω,σ)] M if p = n − 1 and q ∈ (1, ∞).
L (∂Ω,σ)
(1.3.59)
Also, corresponding to two-dimensional exterior domains,
1.3 A First Look at Layer Potential Operators 37
M
if n = 2 and Ω is unbounded then for all f ∈ L p (∂Ω, σ) and R > 0
 Ω∩B(0,R) 
we have Nκ (𝒮 f ) L ∞ (∂Ω,σ) ≤ CR, p  f [L p (∂Ω,σ)] M if p ∈ (1, ∞) and
 Ω∩B(0,R) 
Nκ (𝒮 f ) L q (∂Ω,σ) ≤ CR,q  f [L 1 (∂Ω,σ)] M if p = 1 and q ∈ (1, ∞).
(1.3.60)
Let us also remark here that [70, (2.5.552)] implies that for each p ∈ [1, n − 1)
there exists C ∈ (0, ∞) with the property that
 n−1 −1 
 p 
δ∂Ω 𝒮 f  ≤ C f [L p (∂Ω,σ)] M (1.3.61)
L ∞ (Ω, L n )

M
for all f ∈ L p (∂Ω, σ) .
Consider next the boundary-to-boundary single layer operator
∫ ∫ !
S f (x) := E(x − y) f (y) dσ(y) = Eγα (x − y) fα (y) dσ(y)
∂Ω ∂Ω 1≤γ ≤M
∫   !
= Eγ .(x − y) , f (y) dσ(y)
∂Ω 1≤γ ≤M
∫

= E L  (x − y) f (y) dσ(y)
∂Ω
∫   !

= E L  (x − y) .γ , f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
∂Ω 1≤γ ≤M
  σ(x) M
for each function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω , , (1.3.62)
1 + |x| n−2
with the same conventions9 as in (1.3.6). Then [70, (2.5.554)] ensures that this
operator is indeed well defined. By design, S is linear and, thanks to [70, (2.5.549)],
M
acts in a meaningful way on functions belonging to the space L p (∂Ω, σ) with
p ∈ [1, n − 1). From the Fractional Integration Theorem (cf. [68, (7.8.7), (7.8.14)-
(7.8.15)]), and the estimates for E from [70, Theorem 1.4.2] it follows that this gives
rise to a well-defined linear and continuous mapping
M ∗ M
S : L p (∂Ω, σ) −→ L p (∂Ω, σ)
(1.3.63)
provided p ∈ (1, n − 1) and 1
p∗ = 1
p − n−1 ,
1

and, corresponding to p = 1,

−→ L (n−1)/(n−2),∞ (∂Ω, σ)
M M
S : L 1 (∂Ω, σ) . (1.3.64)

9 In particular, when n = 2 the weight (1 + |x | n−2 )−1 should be replaced by ln(2 + |x |). Also,
we shall occasionally allow the operator S to act on functions originally defined only on ∂∗ Ω by
extending them by zero to the entire topological boundary ∂Ω.
38 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Note that if S A denotes the boundary-to-boundary single layer associated with

transpose coefficient tensor A from (1.3.3) in the same manner S has been associated
with A in (1.3.62), then
for each p ∈ (1, n − 1), the real transpose of (1.3.63) is the
∗  M  M (1.3.65)
operator S A : L (p ) (∂Ω, σ) −→ L p (∂Ω, σ) ,
 
1 −1
where (p∗ ) ∈ (1, n − 1) and p = (p1∗ ) − n−1 are the Hölder conjugate exponents
∗
1 1 −1

of the given p and of p := p − n−1 , respectively. Indeed, the claim in (1.3.65)
readily follows by virtue of (1.3.63), (1.3.6), [70, (1.4.32)], and Fubini’s Theorem
(whose applicability is ensured by the Fractional Integration Theorem; cf. [68,
(7.8.12)]).
We also wish to point out that, in the particular case when ∂Ω is bounded and
p ∈ [n − 1, ∞) (a scenario in which L p (∂Ω, σ) embeds into any L q (∂Ω, σ) with
M
q ∈ (1, n − 1)), the operator S from (1.3.62) maps L p (∂Ω, σ) boundedly into
M
L ∞ (∂Ω, σ) if p > n − 1 and, corresponding to p = n − 1, the operator S from
M M
(1.3.62) maps L n−1 (∂Ω, σ) boundedly into L q (∂Ω, σ) for any q ∈ (0, ∞).
For future reference, it is useful to note that (1.3.63) implies
M M
S : L ∞ (∂Ω, σ) → L q (∂Ω, σ) boundedly (1.3.66)
provided ∂Ω is a compact set and q ∈ (1, ∞).
We also wish to remark that, given any aperture parameter κ > 0, [70, Proposi-
tion 2.5.39] implies (assuming n ≥ 3 when ∂Ω is unbounded) that
  M
for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−2
the boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on Aκ (∂Ω) and, in fact, (1.3.67)
 κ−n.t.
𝒮 f ∂Ω (x) = (S f )(x) at σ-a.e. point x ∈ Aκ (∂Ω).

In particular, from [70, (2.5.549)] it follows that the above nontangential boundary
M
trace formula holds for each function f ∈ L p (∂Ω, σ) with p ∈ [1, n − 1).
Moving on, assume Ω ⊆ R (where n ∈ N with n ≥ 2) is a Lebesgue measurable
n

set, of locally finite perimeter. As in the past, abbreviate σ := H n−1 ∂Ω, and denote
by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω. In
such a setting, we define the boundary-to-boundary double layer potential
operator associated with L and Ω as the principal-value singular integral operators
acting on each function f as in (1.3.17) according to
# ∫ $
βα
K f (x) := − lim+ νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y) (1.3.68)
ε→0
y ∈∂∗ Ω 1≤γ ≤M
|x−y |>ε
1.3 A First Look at Layer Potential Operators 39

for σ-a.e. point x ∈ ∂∗ Ω. From [68, Proposition 5.6.7] we know that this limit
exists and K f is a σ-measurable function on ∂∗ Ω. Moreover, the last claim in [68,
Proposition 5.6.7] implies that if Ω is a Lebesgue measurable set whose topological
boundary ∂Ω is countably rectifiable (of dimension n − 1) and has locally finite
H n−1 measure (hence, in particular, if ∂Ω is a UR set), then for each function f as
in (1.3.17) the limit in (1.3.68) actually exists for σ-a.e. x ∈ ∂Ω and gives rise to a
σ-measurable C M -valued function on ∂Ω.
Throughout, make the convention that
whenever desirable to emphasize the dependence of the
double layer operator (1.3.68) on the coefficient tensor (1.3.69)
(1.3.1) we shall write K A in place of K.
With the same agreements as in the case of D (cf. (1.3.22)), for each function f
as in (1.3.17) we may refashion (1.3.68) as
∫    !
  
K f (x) = lim+ ∂ν(y)
A
E L  (x − y) . γ , f (y) dσ(y)
ε→0 1≤γ ≤M
y ∈∂∗ Ω
|x−y |>ε
∫   !
A   
= lim ∂ν(y) E L (x − y) γ . , f (y) dσ(y)
ε→0+ 1≤γ ≤M
y ∈∂∗ Ω
|x−y |>ε
∫  

= lim+ ∂ν(y)
A
E L  (x − y) f (y) dσ(y) (1.3.70)
ε→0
y ∈∂∗ Ω
|x−y |>ε

at σ-a.e. x ∈ ∂∗ Ω if Ω ⊆ Rn is a set of locally finite perimeter, or at σ-a.e. x ∈ ∂Ω if

Ω ⊆ Rn is a Lebesgue measurable set whose topological boundary ∂Ω is countably
rectifiable (of dimension n − 1) and has locally finite H n−1 measure.
Another singular integral operator of major interest, closely related to the one
introduced in (1.3.68), is the so-called “transpose” double layer operator K # .
To define this, assume Ω is a Lebesgue measurable set whose topological boundary
∂Ω is countably rectifiable (of dimension n − 1) and has locally finite H n−1 measure
(this is the case if, in particular, ∂Ω is a UR set). Then we define the action of K # on
each function
  σ(x) M
f = ( fγ )1≤γ ≤M ∈ L 1 ∂Ω, (1.3.71)
1 + |x| n−1
by setting
40 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
# ∫ $
βα
K f (x) :=
#
lim νs (x)ar s (∂r Eγβ )(x − y) fγ (y) dσ(y) (1.3.72)
ε→0+
y ∈∂Ω 1≤α ≤M
|x−y |>ε

at σ-a.e. x ∈ ∂∗ Ω. Based on [68, Corollary 5.3.6] we see that this definition is indeed
meaningful (here, [68, (5.6.23)] is also helpful). Once again, a similar convention
to (1.3.69) is in place for this operator, i.e., whenever desirable to emphasize the
dependence of the transpose double layer operator (1.3.72) on the coefficient tensor
(1.3.1) we agree to write K A# in place of K # . Also, for each function f as in (1.3.71)
and σ-a.e. point x ∈ ∂∗ Ω we may express
# ∫ $
   
K f (x) = lim+
#
∂ν(x) E L  (x − y) . γ
A
fγ (y) dσ(y)
ε→0 α
y ∈∂Ω 1≤α ≤M
|x−y |>ε
# ∫ $
  
A
= lim ∂ν(x) E L (x − y) γ . fγ (y) dσ(y) , (1.3.73)
ε→0+ α
y ∈∂Ω 1≤α ≤M
|x−y |>ε

where the second equality in (1.3.73) is a consequence of the first property in [70,
(1.4.32)]. Alternatively, we may recast (1.3.72) simply as
∫
A
K # f (x) = lim+ ∂ν(x) E L  (x − y) f (y) dσ(y) (1.3.74)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. x ∈ ∂∗ Ω.
As hinted in (1.3.19) and (1.3.69), the format of the double layer potential oper-
ators D, K, K # is strongly affected by the choice of a coefficient tensor A used to
represent the given system L as L A (cf. (A.0.139)). More on this later. For now we
wish to note that the difference between any two double layer potential operators
associated with any two coefficient tensors A0, A1 that can be used to write the system
L (cf. (1.3.20)) is a linear combination of “tangential” singular integral operators of
the sort introduced in (1.2.2). This is made precise in the proposition below.

Proposition 1.3.8 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a set of locally finite

perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω. Let L be a homogeneous, second-
order, weakly elliptic, M × M system with constant (complex) coefficients in Rn ,
and recall the matrix-valued fundamental solution E = (Eγβ )1≤γ,β ≤M associated
with L as in [70, Theorem 1.4.2]. For each β, γ ∈ {1, . . . , M } and r, s ∈ {1, . . . , n}
consider the “tangential” singular integral operator acting on each given function
φ ∈ L ∂∗ Ω, 1+σ(x)
1
|x | n−1
according to (compare with (1.2.2))
1.3 A First Look at Layer Potential Operators 41
 γβ 
Tr s φ (x) (1.3.75)
∫
 
:= lim+ νr (y)(∂s Eγβ )(x − y) − νs (y)(∂r Eγβ )(x − y) φ(y) dσ(y),
ε→0
y ∈∂∗ Ω
|x−y |>ε

which is meaningfully defined at σ-a.e. x ∈ ∂∗ Ω thanks to the first conclusion in [68,

Proposition 5.6.7].
Then for each function
  M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1 (1.3.76)

and any two coefficient tensors A0, A1 ∈ A L (cf. (1.3.20)), the difference between
K A0 f and K A1 f may be expressed as
   γβ βα 
K A0 − K A1 f = 12 Tr s br s fα 1≤γ ≤M at σ-a.e. point in ∂∗ Ω, (1.3.77)

where 
αβ
br s 1≤α,β ≤M := A0 − A1 . (1.3.78)
1≤r,s ≤n

Furthermore, if ∂Ω is countably rectifiable (of dimension n − 1) then the equality

in (1.3.77) is valid at σ-a.e. point in ∂Ω.

Proof For each function f as in (1.3.76), we see from (1.3.68) and (1.3.78) that
# ∫ $
  βα
K A0 − K A1 f (x) = − lim+ νs (y)br s (∂r Eγβ )(x − y) fα (y) dσ(y)
ε→0
y ∈∂∗ Ω 1≤γ ≤M
|x−y |>ε
(1.3.79)

at σ-a.e. point x ∈ ∂∗ Ω. In view of the fact that A0, A1 ∈ A L , we conclude from

(1.3.78) that
αβ αβ
br s = −bsr for all α, β ∈ {1, . . . , M } and all r, s ∈ {1, . . . , n}. (1.3.80)

Consequently, for each α, γ ∈ {1, . . . , M } we may write

βα
νs (y)br s (∂r Eγβ )(x − y)
βα βα
= 12 νs (y)br s (∂r Eγβ )(x − y) − 12 νs (y)bsr (∂r Eγβ )(x − y)
βα βα
= 12 νs (y)br s (∂r Eγβ )(x − y) − 12 νr (y)br s (∂s Eγβ )(x − y)

βα
= 12 νs (y)(∂r Eγβ )(x − y) − νr (y)(∂s Eγβ )(x − y) br s (1.3.81)
42 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

for each x ∈ ∂Ω and σ-a.e. point x ∈ ∂∗ Ω. Then (1.3.77) now follows by combining
(1.3.79) with (1.3.81) and (1.3.75). The final claim in the statement is a consequence
of what we have proved so far and the last part in [68, Proposition 5.6.7]. 

We conclude by recording the following variant of Proposition 1.3.8, for the dif-
ference of the transpose double layer corresponding to two choices of the coefficient
tensor.

Proposition 1.3.9 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a set of locally finite

perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω. Let L be a homogeneous, second-
order, weakly elliptic, M × M system with constant (complex) coefficients in Rn ,
and recall the matrix-valued fundamental solution E = (Eγβ )1≤γ,β ≤M associated
with L as in [70, Theorem 1.4.2]. For each β, γ ∈ {1, . . . , M } and r, s ∈ {1, . . . , n}
consider the “tangential” singular integral operator acting on each given function
ψ ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
according to (compare with (1.2.3))
 γβ 
r s ψ (x)
T (1.3.82)
∫
 
:= lim+ νr (x)(∂s Eγβ )(x − y) − νs (x)(∂r Eγβ )(x − y) ψ(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε

which is meaningfully defined at σ-a.e. x ∈ ∂∗ Ω thanks to the first conclusion in [68,

Proposition 5.6.7].
Then for each function
  M
f = ( fγ )1≤γ ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
(1.3.83)

and any two coefficient tensors A0, A1 ∈ A L (cf. (1.3.20)), the difference between
K A# 0 f and K A# 1 f may be expressed as
   γβ βα 
r s br s fγ
K A# 0 − K A# 1 f = − 12 T 1≤α ≤M at σ-a.e. point in ∂∗ Ω, (1.3.84)

where 
αβ
br s 1≤α,β ≤M := A0 − A1 . (1.3.85)
1≤r,s ≤n

Moreover, if ∂Ω is countably rectifiable (of dimension n − 1) then the equality in

(1.3.84) is valid at σ-a.e. point in ∂Ω.

Proof The same type of argument as in the proof of Proposition 1.3.8, now making
use of (1.3.72), yields all desired conclusions. 
1.4 Examples and Alternative Points of View 43

Remark 1.3.10 If the system L given in (1.3.2) is actually expressed as

 αβ 
L = Ar s ∂r ∂s with Ar s := ar s 1≤α,β ≤M ∈ C M×M for 1 ≤ r, s ≤ n, (1.3.86)

then we may recast the action of the boundary-to-domain double layer potential
operator on each vector-valued function f as in (1.3.17) simply as
∫
D f (x) = − νs (y)(∂r E)(x − y)Ar s f (y) dσ(y), ∀ x ∈ Ω. (1.3.87)
∂∗ Ω

In particular, this novel algebraic formalism makes it straightforward to see (bearing

in mind that LE = δ · I M×M in the sense of distributions in Rn ; cf. part (2) in [70,
Theorem 1.4.2]) that L(D f ) = 0 in Ω for each f as in (1.3.17) (compare with
(1.3.24)-(1.3.25)). Likewise, in terms of the writing (1.3.86) we may express the
action of the boundary-to-boundary double layer potential operator K (originally
associated with L and Ω as in (1.3.68)) on vector-valued functions f as in (1.3.17)
simply as
∫
K f (x) = − lim+ νs (y)(∂r E)(x − y)Ar s f (y) dσ(y) (1.3.88)
ε→0
y ∈∂∗ Ω
|x−y |>ε

for σ-a.e. point x ∈ ∂∗ Ω. Finally, the action of the “transpose” double layer operator
from (1.3.72) on each vector-valued function f as in (1.3.71) may be recast simply
as ∫
K # f (x) = lim+ νs (x)(Ar s ) (∂r E) (x − y) f (y) dσ(y) (1.3.89)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. x ∈ ∂∗ Ω.

1.4 Examples and Alternative Points of View

We begin by looking at a number of basic concrete manifestations of the recipes

discussed in §1.3 for associating boundary layer potentials with given sets of locally
finite perimeter and weakly elliptic operators.

Example 1.4.1 Let us first consider the case of the two-dimensional Laplacian L = Δ
in a nonempty, Lebesgue measurable, proper subset Ω of R2 , having locally finite
perimeter. Denote by ν the geometric measure theoretic outward unit normal to Ω,
and abbreviate σ := H 1 ∂Ω. In this setting, if
44 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
! !
10  1 i
A := and A := , (1.4.1)
01 −i 1
then
L A = L A = ∂12 + ∂22 = Δ. (1.4.2)
This being said, corresponding to the first choice of a matrix in (1.4.1), the recipes
(1.3.18), (1.3.68) presently yield the integral operators acting on each function

σ(x)
f ∈ L 1 ∂∗ Ω, 1+ |x | (1.4.3)

according to
∫
1 ν(y), y − x
D A f (x) = f (y) dσ(y) for all x ∈ Ω̊, (1.4.4)
2π ∂∗ Ω |x − y| 2
∫
1 ν(y), y − x
K A f (x) = lim+ f (y) dσ(y) (1.4.5)
ε→0 2π ∂∗ Ω\B(x,ε) |x − y| 2

for σ-a.e. x ∈ ∂∗ Ω, which are, respectively, the (two-dimensional) harmonic

boundary-to-domain double layer potential operator and the harmonic boundary-
to-boundary double layer potential operator, whereas under the natural identification
R2 ≡ C the operators D A, K A associated as in (1.3.18), (1.3.68) with Ω and the
 from (1.4.1) act on each function
matrix A

f ∈ L 1 ∂∗ Ω, H
1 (ζ )

1+ |ζ | , (1.4.6)

according to
∫
1 f (ζ)
D A f (z) = dζ for all z ∈ Ω̊, (1.4.7)
2πi ∂∗ Ω ζ −z
∫
1 f (ζ)
K A f (z) = lim+ dζ for H 1 -a.e. z ∈ ∂∗ Ω, (1.4.8)
ε→0 2πi ∂∗ Ω\B(z,ε) ζ − z

hence these are now the boundary-to-domain Cauchy integral operator and the
boundary-to-boundary Cauchy integral operator, respectively. In this vein, we wish
to recall that [68, Proposition 5.6.7] ensures that the limits in (1.4.5) and (1.4.8) do
exist as indicated. If, however, Ω is a Lebesgue measurable set whose topological
boundary ∂Ω is countably rectifiable (of dimension 1) and has locally finite H 1
measure (hence, in particular, if ∂Ω is a UR set), then the limits in (1.4.5) and (1.4.8)
actually exist for H 1 -a.e. point in ∂Ω.

Example 1.4.2 Consider next the higher-dimensional setting i.e., when n ∈ N with
n ≥ 2 is arbitrary. We are interested in the case of the Laplacian L = Δ in a nonempty,
1.4 Examples and Alternative Points of View 45

Lebesgue measurable, proper subset Ω of Rn , having locally finite perimeter. As in

the past, abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω. Writing the Laplacian Δ = ∂12 + · · · + ∂n2
as

n
 
Δ= a jk ∂j ∂k where A = a jk 1≤ j,k ≤n
j,k=1
(1.4.9)
has entries a jk := δ jk for each j, k ∈ {1, . . . , n}
the recipes (1.3.18), (1.3.68) yield the standard boundary-to-domain and boundary-
to-boundary harmonic double layer potential operators, acting on each function

f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
(1.4.10)

according to
∫
1 ν(y), y − x
D A f (x) = f (y) dσ(y) for all x ∈ Ω̊, (1.4.11)
ωn−1 ∂∗ Ω |x − y| n
∫
1 ν(y), y − x
K A f (x) = lim+ f (y) dσ(y) (1.4.12)
ε→0 ωn−1 ∂∗ Ω\B(x,ε) |x − y| n

for σ-a.e. x ∈ ∂∗ Ω. On the other hand, adopting the Clifford algebra formalism
recalled earlier in [68, §6.4] and expressing the Laplacian Δ = ∂12 + · · · + ∂n2 as


n
 
Δ=  = 
a jk ∂j ∂k where A a jk 1≤ j,k ≤n
j,k=1
(1.4.13)
has entries 
a jk := −e j  ek for each j, k ∈ {1, . . . , n}

now yields (in view of (1.3.22), (1.3.70), and the fact that ∂νA u = −(Du)  ν for each
Clifford algebra-valued function u) the integral operators acting on each function
Clifford algebra-valued function

f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
⊗Cn (1.4.14)

according to
∫
1 x−y
D A f (x) =  ν(y)  f (y) dσ(y) for all x ∈ Ω̊, (1.4.15)
ωn−1∂∗ Ω |x − y| n
∫
1 x−y
K A f (x) = lim+  ν(y)  f (y) dσ(y) (1.4.16)
ε→0 ωn−1 ∂∗ Ω\B(x,ε) |x − y| n

for σ-a.e. x ∈ ∂∗ Ω, i.e., the boundary-to-domain and boundary-to-boundary Cauchy-

Clifford integral operators.
46 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Once again, [68, Proposition 5.6.7] guarantees that the limits in (1.4.12) and
(1.4.16) do exist as specified. This being said, if Ω is a Lebesgue measurable set
whose topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and
has locally finite H n−1 measure (hence, in particular, if ∂Ω is a UR set), then the
limits in (1.4.12) and (1.4.16) actually exist for σ-a.e. point in ∂Ω.

Example 1.4.3 In the last part of Example 1.4.2, we could circumvent the direct use
of the Clifford algebras by employing instead matrix formalism. To elaborate on this,
fix n ∈ N with n ≥ 2, and consider a family of real matrices {E j }1≤ j ≤n satisfying
 2
Ej = −I2n ×2n for each j ∈ {1, . . . , n} and
(1.4.17)
E j Ek = −Ek E j for all j, k ∈ {1, . . . , n} with j  k.
The existence of such a family has been established in [68, (6.4.14)] (with m := n). To
proceed, define M := 2n and denote by I M×M the M × M identity matrix. Consider
the M × M second-order system in Rn defined as

L := Δ · I M×M (1.4.18)

where Δ = ∂12 + · · · + ∂n2 is the Laplacian in Rn . For each j, k ∈ {1, . . . , n} denote by

αβ
(a jk )1≤α,β ≤M the entries of the M × M matrix −E j Ek , i.e.,

αβ
a jk = −(E j Ek )αβ for each j, k ∈ {1, . . . , n}
(1.4.19)
and each α, β ∈ {1, . . . , M }.

Then, with the summation convention over repeated indices in effect, (1.4.17) implies
 αβ 
a jk ∂j ∂k 1≤α,β ≤M = −E j Ek ∂j ∂k = −(E j )2 ∂j2 = Δ · I M×M , (1.4.20)

hence,  αβ 
L = a jk ∂j ∂k 1≤α,β ≤M . (1.4.21)
Also, the fundamental solution E L associated with the weakly elliptic system L as
in [70, Theorem 1.4.2] is given by

E L := EΔ · I M×M (1.4.22)

where EΔ is the standard fundamental solution for the Laplacian in Rn , defined in

(A.0.65). In particular, for each x ∈ Rn \ {0} we have
δγβ x j
∂j (E L )γβ (x) = for 1 ≤ j ≤ n and 1 ≤ γ, β ≤ M. (1.4.23)
ωn−1 |x| n
1.4 Examples and Alternative Points of View 47

Pressing on, suppose Ω ⊆ Rn (where n ∈ N has n ≥ 2) is a set of locally finite

perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω. Finally, pick an arbitrary function
  σ(x) M
f ∈ L 1 ∂∗ Ω, . (1.4.24)
1 + |x| n−1
Then from (1.3.18), (1.4.19), and (1.4.23) we see that the boundary-to-domain double
layer potential operator associated as in (1.3.18) with the set Ω and the system L
written as in (1.4.21) acts on f according to
∫
1 
n
xj − yj  
n
D f (x) = Ej νk (y)Ek f (y) dσ(y) (1.4.25)
ωn−1 ∂∗ Ω j=1 |x − y| n k=1

for every x ∈ Ω̊. In terms of the boundary-to-domain Riesz transforms, R j with

1 ≤ j ≤ n, defined as in (A.0.188) (with Σ := ∂Ω), formula (1.4.25) becomes

1
n   n 
Df = Ej R j νk Ek f in Ω. (1.4.26)
2 j=1 k=1

Likewise, the boundary-to-boundary double layer potential operator associated as in

(1.3.68) with the set Ω and the system L written as in (1.4.21) acts on any function
f as in (1.4.24) according to

1
n   n 
Kf = E j Rj νk Ek f (1.4.27)
2 j=1 k=1

where R j for 1 ≤ j ≤ n are the boundary-to-boundary Riesz transforms defined as

in (A.0.187) with Σ := ∂∗ Ω.

Example 1.4.4 Let us take yet another point of view on the manner in which bound-
ary layer potentials may be associated with a given second-order weakly elliptic
system. Suppose L is a given second-order, homogeneous, constant (complex) co-
efficient, weakly elliptic M × M system in Rn , and denote by E = (Eγβ )1≤γ,β ≤M
the matrix-valued fundamental solution associated with L as in [70, Theorem 1.4.2].
Next, assume that
 
=  αγ γβ
B br 1≤r ≤n and B = bs 1≤s ≤n (1.4.28)
1≤α ≤M 1≤γ ≤ N
1≤γ ≤ N 1≤β ≤M

are two coefficient tensors (of type (1 × n, M × N) and (n × 1, N × M), respectively),

with complex entries, such that the associated first-order operators
48 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

 
 :=  αγ γβ
D br ∂r 1≤α ≤M and D := bs ∂s 1≤γ ≤ N (1.4.29)
1≤γ ≤ N 1≤β ≤M

may be used to factor L as follows:


L = DD. (1.4.30)

In particular, this permits us to express


αγ γβ
L=  br bs ∂r ∂s 1≤α ≤M (1.4.31)
1≤β ≤M

hence, 
αβ
L = L A where A := AD,D
 := ar s 1≤r,s ≤n
1≤α,β ≤M (1.4.32)
αβ αγ γβ
with each ar s := br bs .
Let us now consider a Lebesgue measurable nonempty proper set Ω ⊂ Rn of locally
finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the
geometric measure theoretic outward unit normal to Ω. Recall from [70, Conven-
tion 1.7.2] that whenever the function u = (uβ )1≤β ≤M is of class 𝒞1 in a neighbor-
hood of ∂∗ Ω, its conormal derivative associated with Ω and the coefficient tensor A
from (1.4.32) is defined as
 
αβ αγ γβ
∂νAu = νr ar s (∂s uβ ) ∂∗ Ω = (−i) iνr 
br bs (∂s uβ ) ∂∗ Ω
1≤α ≤M 1≤α ≤M

 ν)(Du)
= (−i)Sym( D; at σ-a.e. point on ∂∗ Ω (1.4.33)
∂∗ Ω

(see also [70, (1.7.42)] in this regard). Likewise, since

 
βγ  = −  γα
D = − bs ∂s 1≤γ ≤ N and D br ∂r 1≤α ≤M (1.4.34)
1≤β ≤M 1≤γ ≤ N

it follows that

 
βα βγ γα
∂νA u = νr asr (∂s uβ ) ∂∗ Ω = (−i) iνr 
bs br (∂s uβ ) ∂∗ Ω
1≤α ≤M 1≤α ≤M

 u)
= (−i)Sym(D ; ν)( D at σ-a.e. point on ∂∗ Ω. (1.4.35)
∂∗ Ω

In light of (1.3.22), (1.4.35), and [68, (1.7.17)] (as well as simple matrix formalism)
we may express the action of the boundary-to-domain double layer operator (1.3.18)
on any function f as in (1.3.17) at each point x ∈ Ω̊ as
1.4 Examples and Alternative Points of View 49
∫  !
      
D f (x) = −i y E L  (x − y) . , f (y) dσ(y)
Sym D ; ν(y) D γ
∂∗ Ω 1≤γ ≤M
∫   !
    
=i y E L  (x − y) . , Sym D; ν(y) f (y) dσ(y)
D γ
∂∗ Ω 1≤γ ≤M
∫
   
= (−i)  E L 
D (x − y) Sym D; ν(y) f (y) dσ(y), (1.4.36)
∂∗ Ω

where E L  is the fundamental  solution

 associated with the transpose system L 
as in [70, Theorem 1.4.2], E L  . γ denotes the γ-th column of the matrix E L  ,
and subscript y for D  indicates that said operator acts in the variable y. Also, in
 E L  is interpreted naturally, according to the ordinary
the last line of (1.4.36) D
multiplication of matrices. Bearing this in mind it is actually possible to eliminate
the transposition(s) altogether in the last line of (1.4.36) and simply write, for each
function f as in (1.3.17) and each point x ∈ Ω̊,
∫
   
D f (x) = (−i)  (x − y) Sym D; ν(y) f (y) dσ(y),
EL D (1.4.37)
∂∗ Ω

with the convention that E L D means ordinary multiplication of matrices (treating

the product between a function Eγ β and a partial derivative ∂r as being ∂r Eγ β ).
Another, more direct, way of proving formula (1.4.36) is to start with (1.3.18)
and then use (1.4.32), [70, (1.4.32)], (A.0.192), (1.4.34), to write
# ∫ $
βα  
D f (x) = − νs (y)ar s ∂r (E L )γβ (x − y) fα (y) dσ(y)
∂∗ Ω
1≤γ ≤M
# ∫ $
βδ  
= − νs (y)
br bδα
s ∂r (E L  )βγ (x − y) fα (y) dσ(y)
∂∗ Ω
1≤γ ≤M
#∫ $
 βδ   
= −
br ∂r (E L  )βγ (x − y) νs (y)bδα
s fα (y) dσ(y)
∂∗ Ω
1≤γ ≤M
#∫ $
    
= (−i) 
D EL δγ (x − y) Sym D; ν(y) f (y)
δ
dσ(y)
∂∗ Ω
1≤γ ≤M
∫
   
= (−i)  E L 
D (x − y) Sym D; ν(y) f (y) dσ(y), (1.4.38)
∂∗ Ω

for each function f as in (1.3.17) and each point x ∈ Ω̊.

In a similar fashion, we see from (1.3.70) and (1.4.35) that the action of the
principal-value double layer potential operator (1.3.68) on any function f as in
(1.3.17) may be recast as
50 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

K f (x)
∫   !
     
= −i lim+  
Sym D ; ν(y) Dy E L (x − y) . γ , f (y) dσ(y)

ε→0 1≤γ ≤M
y ∈∂∗ Ω
|x−y |>ε
∫   !
      
= i lim+ 
Dy E L  (x − y) . γ , Sym D; ν(y) f (y) dσ(y)
ε→0 1≤γ ≤M
y ∈∂∗ Ω
|x−y |>ε
∫
   
= (−i) lim+  E L 
D (x − y) Sym D; ν(y) f (y) dσ(y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
∫
   
= (−i) lim+  (x − y) Sym D; ν(y) f (y) dσ(y)
EL D (1.4.39)
ε→0
y ∈∂∗ Ω
|x−y |>ε

at σ-a.e. x ∈ ∂∗ Ω. From [68, Proposition 5.6.7] we know that the limit in (1.4.39)
exists as indicated, and if Ω is actually a Lebesgue measurable set whose topological
boundary ∂Ω is countably rectifiable (of dimension n−1) and has locally finite H n−1
measure (hence, in particular, if ∂Ω is a UR set) then the limit in (1.4.39) exists for
σ-a.e. point in ∂Ω.
Let us adopt the latter geometric setting, i.e., assume now that Ω is a Lebesgue
measurable set whose topological boundary ∂Ω is countably rectifiable (of dimension
n − 1) and has locally finite H n−1 measure (this is the case if, in particular, ∂Ω is a
UR set). In such a setting, in view of (1.3.73) we may refashion the singular integral
operator K # , originally defined in (1.3.72), as

K # f (x)
∫  !
   
= −i lim+ 
Sym D ; ν(x) D
x E L  (x − y) . γ fγ (y) dσ(y)
ε→0 α 1≤α ≤M
y ∈∂Ω
|x−y |>ε
∫
   
= (−i) lim+ Sym D ; ν(x) D E L  (x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
# ∫ $
   
= (−i)Sym D ; ν(x) 
lim 
D E L  (x − y) f (y) dσ(y) , (1.4.40)
ε→0+
y ∈∂Ω
|x−y |>ε
1.4 Examples and Alternative Points of View 51

for each function f as in (1.3.71), at σ-a.e. point x ∈ ∂∗ Ω (which is meaningful,

according to [68, Proposition 5.6.7]).
Moving on, it is of interest to record a version of the integral representation
formula from Theorem 1.3.3 emphasizing the role of factoring a given second-order
system as a product of two first-order systems.

Corollary 1.4.5 Let Ω ⊆ Rn , where10 n ≥ 3, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Let L be a homogeneous,
weakly elliptic, second-order M ×M system in Rn , with complex constant coefficients.
Suppose this may be factored as

L = DD (1.4.41)
where D and D are two homogeneous, constant (complex) coefficient, first-order
systems in Rn .
1 (Ω, L n ) M is a vector-valued function satisfying, for some
Next, assume u ∈ Lloc
aperture parameter κ > 0 and with all derivatives taken in the sense of distributions,
  dy M
Lu belongs to L 1 Ω, ,
1 + |y| n−2
Du has locally integrable components in Ω, (1.4.42)
κ−n.t. κ−n.t.
u ∂Ω and (Du) ∂Ω exist σ-a.e. on ∂nta Ω.

In addition, assume the following integrability conditions hold:

∫ ∫  
(Nκ u)(y) Nκ (Du) (y)
dσ(y) < ∞ and dσ(y) < ∞. (1.4.43)
∂Ω 1 + |y| ∂Ω 1 + |y|
n−1 n−2

Finally, recall the Newtonian potential operator ΠΩ from (1.3.4), the single layer
potential operator 𝒮 from (1.3.6), the boundary-to-domain double layer potential
operator D A D,
D
associated as in (1.3.18) with the set Ω and the coefficient tensor
 D are as in (1.4.29)), and the conormal derivative
A  (defined as in (1.4.32) if D,
D,D

operator ∂νD,D associate with the factorization (1.4.41) as in (A.0.185).
Then at L n -a.e. point x ∈ Ω one has
 κ−n.t.    
u(x) = D A D,
D
u ∂Ω (x) − 𝒮 ∂νD,D u (x) + ΠΩ (Lu)(x), (1.4.44)

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds under the additional assumption that there exists
λ ∈ (1, ∞) such that

10 see the last part in the statement for the case n = 2

52 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

⨏
|u| dL n = o(1) as R → ∞. (1.4.45)
B(0,λ R)\B(0,R)

As a corollary, if in place of the first line in (1.4.42) one now imposes the stronger
M
condition that Lu = 0 in Ω, then u ∈ 𝒞∞ (Ω) and at each point x ∈ Ω one has
(with the same caveat when Ω is an exterior domain)
 κ−n.t.    
u(x) = D A D,
D
u ∂Ω (x) − 𝒮 ∂νD,D u (x)
∫
      κ−n.t. 
=  E L   (x − y) (−i)Sym D; ν(y) u
D (y) dσ(y)
∂Ω
∂∗ Ω
∫
  κ−n.t.
−  ν(y) (Du)
E L (x − y)(−i)Sym D; (y) dσ(y), (1.4.46)
∂Ω
∂∗ Ω

where E L and E L  are, respectively, the fundamental solutions associated with L

and L  as in [70, Theorem 1.4.2] (and where D  acts on the columns of E L  ).
Lastly, similar results are valid in the case when n = 2 provided either
∫
−1
L(ξ) dH 1 (ξ) = 0 ∈ C M×M , (1.4.47)
S1

or otherwise one assumes that ∂Ω is compact, replaces the first membership in

(1.4.42) by ∫
|(Lu)(y)| ln(2 + |y|) dy < +∞, (1.4.48)
Ω
and, in the case when Ω is an exterior domain, replaces (1.4.45) by
⨏
 
|u| dL 2 = o ln1R as R → ∞. (1.4.49)
B(0,λ R)\B(0,R)

Proof All claims in the statement are clear from Theorem 1.3.3, definitions, (1.4.38),
and elliptic regularity (cf. [68, (6.5.40) in Theorem 6.5.7]). 

Remark 1.4.6 Consider the boundary layer representation formula (1.4.46) (in the
context of Corollary 1.4.5), in the case when the weakly elliptic M × M system
 with D, D
is factored as L = DD,  as in (1.4.29) (hence, the coefficient tensor
A := AD,D
 used to represent L as L A is as in (1.4.32)), and under the assumption
that
Du = 0 in Ω. (1.4.50)
A 
In such a scenario, as seen from (A.0.185), the conormal derivative ∂ν D, D u = ∂νD,D u

vanishes at σ-a.e. point on ∂∗ Ω. Also, Lu = D(Du) = 0 in Ω. As such, the boundary
layer representation formula (1.4.46) ultimately reduces to
1.4 Examples and Alternative Points of View 53
 n.t. 
u(x) = D A D,
D
u ∂Ω (x) (1.4.51)
∫
      n.t. 
=  E L   (x − y) (−i)Sym D; ν(y) u
D (y) dσ(y)
∂Ω
∂∗ Ω

for each point x ∈ Ω, where E L  is the fundamental solution associated with the
transpose system L  as in [70, Theorem 1.4.2].
The most basic manifestation of this phenomenon is Cauchy’s Reproducing For-
mula in complex analysis, allowing one to recover a holomorphic function from its
boundary trace, via the boundary-to-domain Cauchy integral operator. See (1.4.71)-
(1.4.73) in Example 2 presented a little later below. In a nutshell, the conormal
derivative associated with the factorization Δ = ∂∂ of the Laplacian in R2 ≡ C of
any function u which happens to be holomorphic in the open set Ω ⊆ R2 always
n.t.
vanishes (since (−i)Sym(∂; ν)(∂u) ∂Ω = 0 given that ∂u = 0 in Ω) and, as a result,
the corresponding Green’s Representation Formula for u regarded as a harmonic
function in Ω becomes precisely Cauchy’s Reproducing Formula for u viewed as a
holomorphic function in Ω.
A more general formulation of the result described in Remark 1.4.6 reads as
follows:
Theorem 1.4.7 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors
regular boundary, and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In
particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
Next, consider a homogeneous, first-order N × M system D with constant complex
coefficients in Rn (where N, M ∈ N) which is injectively elliptic (cf. [70, (1.3.18)]),
and suppose D  is a homogeneous first-order M × N system with constant complex
coefficients in Rn which complements D (i.e., [70, (1.3.21)] holds). In particular,
 is a weakly elliptic second-order M × M system in Rn . Let A  be the
L := DD D,D
coefficient tensor induced by the factorization DD  of the system L, defined as in
(1.4.32). Also, let D be the boundary-to-domain double layer potential operator
associated as in (1.3.18) with the set Ω and the coefficient tensor A := AD,D .
Finally, fix an aperture parameter κ ∈ (0, ∞) and suppose u : Ω → C M is a
vector-valued function with Lebesgue measurable components satisfying11

Nκ u ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
,
 κ−n.t.
u ∂Ω (x) exists for σ-a.e. x ∈ ∂nta Ω, (1.4.52)
  dy  N
and Du ∈ L 1 Ω, .
1 + |y| n−1

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
11 with Du considered in the sense of distributions in Ω
54 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
⨏
|u| dL n = o(1) as R → ∞. (1.4.53)
B(0,λ R)\B(0,R)

κ  −n.t.
Then for any κ  > 0 the nontangential trace u ∂Ω also exists at σ-a.e. point
on ∂nta Ω and is actually independent of κ . Moreover, with the dependence on the
aperture parameter dropped, for L n -a.e. x ∈ Rn \ ∂Ω one has
∫ 
 n.t.     u(x) if x ∈ Ω,
D u ∂Ω (x) −  E L  (x − y)(Du)(y) dy =
D (1.4.54)
Ω 0 if x ∈ Rn \ Ω,

where E L  is the matrix-valued fundamental solution associated with L  as in [70,

Theorem 1.4.2], and where D  acting on the columns of E L  .
As a corollary, if the last property in (1.4.52) is strengthened to

Du = 0 in Ω, (1.4.55)
M
then u ∈ 𝒞∞ (Ω) and (1.4.54) reduces to

 n.t.  u in Ω,
D u ∂Ω = (1.4.56)
0 in Rn \ Ω.

For example,
 αβ  starting with a given second-order, weakly elliptic M × M system
L = ar s ∂r ∂s 1≤α,β ≤M in Rn , with constant (complex) coefficients, the factorization
 with D, D
L = DD  as in (1.4.67) makes formula (1.4.54) take the form

 n.t.  ∫
βγ
D u ∂Ω (x) + ar s (∂r Eαβ )(x − y)(∂s uα )(y) dy
Ω 1≤γ ≤M

u(x) if x ∈ Ω,
= (1.4.57)
0 if x ∈ Rn \ Ω,

where E = (Eαβ )1≤α,,β ≤M is the matrix-valued fundamental solution associated

with L as in [70, Theorem 1.4.2], and (uα )1≤α ≤M are the scalar components of the
vector-valued function u.
Let also emphasize the remarkable corollary that, in the context of Theorem 1.4.7,
any function u satisfying

u ∈ 𝒞∞ (Ω) , Du = 0 in Ω, Nκ u ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
,
κ−n.t.
the nontangential trace u ∂Ω exists at σ-a.e. point on ∂nta Ω, (1.4.58)

and if Ω is an exterior domain then (1.4.53) also holds,

is reproduced by D in the sense that (cf. (1.4.56))

1.4 Examples and Alternative Points of View 55
 κ−n.t. 
u = D u ∂Ω in Ω. (1.4.59)

Simply put, any double layer reproduces something (specifically, the null-solutions
of the second factor in the factorization of the original system that has produced said
double layer in the first place).
The proof of Theorem 1.4.7 is presented next.
Proof of Theorem 1.4.7 The first two properties in (1.4.52) together with [68,
κ  −n.t.
Corollary 8.9.9] imply that for any κ  > 0 the nontangential trace u ∂Ω exist-
s at σ-a.e. point on ∂nta Ω and is actually independent of the parameter κ . Also,
(1.4.54) is obtained by the equation in [70, (1.6.3)] and (1.4.36) (while bearing in
mind [70, (1.6.1)]). Finally, if (1.4.55) holds then elliptic regularity implies that
M
u ∈ 𝒞∞ (Ω) , while (1.4.54) becomes (1.4.56).
Another way of justifying (1.4.56) is to note that if the equation (1.4.55) holds
then Lu = DDu = 0 in Ω, so the formula in question is a particular case of
Theorem 1.3.3 (or Corollary 1.4.5) when either n ≥ 3, or n = 2 and Ω is bounded,
and of Theorem 1.3.4, when n = 2 and Ω is an exterior domain. 
For a given second-order system L, formulas (1.4.36)-(1.4.40) allow us to asso-
ciate double layer potential operators with any factorization of L as in (1.4.30). This
procedure also prefigures how one should associate double layer potential operators
for a given second-order elliptic system in the manifold setting which is the com-
position, as in (1.4.30), of two first-order operators acting between vector bundles
over said differential manifold; see [74], [75], and [76] in this regard. Indeed, a
key advantage of working with a second-order system described through a (global)
factorization as in (1.4.30) in place of a (local) description via a coefficient tensor
A as in (1.4.32) is that the former (coordinate-free) formalism carries over to the
setting of manifolds.
Of course, whenever the second-order system L is expressed as


N
L= j D j
D (1.4.60)
j=1

 j , D j with 1 ≤ j ≤ N, we
for some families of first-order differential operators D
may refashion (1.4.60) as the factorization

D
% 1(
 where D
 := ( D
1, . . . , D &
N ) and D := & .. )) .
L = DD (1.4.61)
& . )
'D N *
Implementing (1.4.36)-(1.4.37) for the factorization (1.4.61) yields (retaining the
earlier setting, and with the same algebraic conventions as before)
56 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

N ∫
    
D f (x) = j E L  (x − y)(−i) Sym D j ; ν(y) f (y) dσ(y)
D
j=1 ∂∗ Ω

N ∫
    
=  j (x − y)(−i) Sym D j ; ν(y) f (y) dσ(y)
EL D (1.4.62)
j=1 ∂∗ Ω

for all x ∈ Ω̊, while implementing (1.4.39) and (1.4.40) for the factorization (1.4.61)
gives, at σ-a.e. point x ∈ ∂∗ Ω,


N ∫
   
K f (x) = lim+  j E L  (x − y)(−i) Sym D j ; ν(y) f (y) dσ(y)
D
ε→0
j=1 y∈∂∗ Ω
| x−y |>ε


N ∫
   
= lim+  j (x − y)(−i) Sym D j ; ν(y) f (y) dσ(y)
EL D (1.4.63)
ε→0
j=1 y∈∂∗ Ω
| x−y |>ε

and, respectively,
# ∫ $

N
   
K f (x) =
#
(−i)Sym Dj ; ν(x) lim 
D j E L  (x − y) f (y) dσ(y) .
ε→0+
j=1 y ∈∂Ω
|x−y |>ε
(1.4.64)

It is instructive to provide concrete examples to the approach for defining boundary

layer potentials in relation of a given second-order system proposed above.

Example 1.4.8 Suppose a second-order, weakly elliptic M × M system in Rn , with

constant (complex) coefficients,
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M , (1.4.65)
has been given. Hence, with div understood as the differential operator taking any
w = (wsβ ) 1≤s ≤n into divw := (∂s wsβ )1≤β ≤M , we have
1≤β ≤M

αβ
L = L A = divA∇ where A = ar s 1≤r,s ≤n . (1.4.66)
1≤α,β ≤M

Choosing
 := divA, i.e.,
D := ∇ and D

Du := ∂s uβ 1≤s≤n if u = (uβ )1≤β ≤ M and
1≤β ≤ M (1.4.67)
 αβ 
 :=
Dw ∂r ar s ws β 1≤α≤ M if w = (ws β ) 1≤s≤n
1≤β ≤ M
1.4 Examples and Alternative Points of View 57

yields the factorization


L = DD. (1.4.68)
Also,  
D w = − ∂s wsβ 1≤β ≤M if w = (wsβ ) 1≤s ≤n ,
1≤β ≤M
 
 u = − ∂r arαβ
D s uα 1≤s ≤n if u = (uα )1≤α ≤M ,
1≤β ≤M
  (1.4.69)
Sym(D ; ξ)w = − iξs wsβ 1≤β ≤M

if w = (wsβ ) 1≤s ≤n and ξ = (ξs )1≤s ≤n .

1≤β ≤M

Next, fix a Lebesgue measurable nonempty proper subset Ω of Rn , having locally

finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the
geometric measure theoretic outward unit normal to Ω. In this context, whenever
the function u = (uα )1≤α ≤M is of class 𝒞1 in a neighborhood of ∂∗ Ω, [70, Conven-
tion 1.7.2] tells us that at σ-a.e. point on ∂∗ Ω we have
    
 u αβ
(−i)Sym(D ; ν) D ∂∗ Ω
= νs ar s (∂r uα ) ∂∗ Ω = ∂νA u. (1.4.70)
1≤β ≤M

Bearing this in mind it follows that the boundary layer potential operators associated
as in formulas (1.4.36)-(1.4.40) with the factorization (1.4.68) of the second-order
system L, with D and D  as in (1.4.67), are precisely those considered earlier in
(1.3.22), (1.3.68), (1.3.72). Incidentally, these identifications remain valid if instead
of D, D as in (1.4.67) we take D := A∇ and D  := div.
The bottom line is that the “old” point of view, of associating boundary layer
potentials with a second-order system L as in (1.3.22), (1.3.68), (1.3.72), starting
from the representation of L as L A for some coefficient tensor A, may be subsumed
into the “new” point of view, of associating boundary layer potentials with a second-
order system L as in (1.4.36)-(1.4.40), starting from the factorization L = DD  for

some first-order systems D, D.
Example 1.4.9 Work in R2 ≡ C and consider the factorization of the two-
dimensional Laplacian Δ = ∂x2 + ∂y2 given by
 where D
Δ = DD  := ∂x − i∂y and D := ∂x + i∂y . (1.4.71)
Also, fix a Lebesgue measurable nonempty proper subset Ω of R2 ,
having locally
finite perimeter. Abbreviate σ := H 1 ∂Ω and denote by ν = (ν1, ν2 ) the geometric
measure theoretic outward unit normal to Ω, canonically identified with the complex-
valued function ν = ν1 +iν2 defined σ-a.e. on ∂∗ Ω. Then a simple direct computation
reveals that the boundary layer potentials associated with Δ as in (1.4.36),
 (1.4.39)
σ(ζ )
starting from the factorization given in (1.4.71) act on functions f ∈ L ∂∗ Ω, 1+
1
|ζ |
according to ∫
1 f (ζ)
D f (z) = dζ for all z ∈ Ω̊, (1.4.72)
2πi ∂∗ Ω ζ − z
and
58 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
1 f (ζ)
K f (z) = lim+ dζ for σ-a.e. z ∈ ∂∗ Ω. (1.4.73)
ε→0 2πi ζ−z
y ∈∂∗ Ω
|x−y |>ε

These are, of course, the boundary-to-domain and boundary-to-boundary versions

of the Cauchy integral operator (cf. the discussion in [70, §1.1]).
On the other hand, if in place of (1.4.71) for the same two-dimensional Laplacian
Δ = ∂x21 + ∂x22 we now consider the factorization
# $
∂x1
 where D
Δ = DD  := (∂x1 , ∂x2 ) and D := , (1.4.74)
∂x2

then the double layer potential operators associated with the two-dimensional Lapla-
cian as indicated in (1.4.36),
 (1.4.39) starting from the factorization (1.4.74) now act
σ(x)
on functions f ∈ L 1 ∂∗ Ω, 1+ |x | according to
∫
1 ν(y), y − x
D f (x) = f (y) dσ(y) for all x ∈ Ω̊, (1.4.75)
2π ∂∗ Ω |x − y| 2

and
∫
1 ν(y), y − x
K f (x) = lim+ f (y) dσ(y) for σ-a.e. x ∈ ∂∗ Ω. (1.4.76)
ε→0 2π |x − y| 2
y ∈∂∗ Ω
|x−y |>ε

These are the boundary-to-domain and boundary-to-boundary versions of the two-

dimensional harmonic double layer operator (compare with [70, (2.5.202), (2.5.203)]
in the n-dimensional setting). From [68, Proposition 5.6.7] we know that the limits
in (1.4.73) and (1.4.76) exist as indicated, but if Ω is actually a Lebesgue measurable
set whose topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and
has locally finite H n−1 measure (for example, if ∂Ω is a UR set) then said limits
exist for σ-a.e. point in ∂Ω.
In particular, he above discussion shows that in the two-dimensional setting
both the Cauchy integral operator and the harmonic double layer arise in the same
fashion, i.e., as boundary layer potential operators associated with two particular
factorizations of the two-dimensional Laplacian.
From (1.4.58)-(1.4.59) we know that the boundary-to-domain double layer asso-
 of a given weakly elliptic second-order
ciated as in (1.4.36) with a factorization DD
system L reproduces null-solutions of the first-order system D (with proper nontan-
gential boundary behavior). In the case of (1.4.71) we have chosen D to be (up to
normalization) the Cauchy-Riemann operator, so its null-solutions are holomorphic
functions in Ω. In turn, this yields the Cauchy reproducing formula for holomor-
phic functions with proper nontangential boundary behavior (see [70, (1.1.10)] in
this regard). On the other hand, in the case of (1.4.74) we have taken D to be the
gradient operator, so its null-solutions are locally constant functions in Ω. The cor-
1.4 Examples and Alternative Points of View 59

responding reproducing formula for the double layer in this scenario is discussed in
Proposition 1.3.6.

Example 1.4.10 Continue to work in R2 ≡ C and consider Bitsadze’s operator

 
LB := ∂z̄2 where ∂z̄ := 12 ∂x + i∂y (1.4.77)
(cf. [3], [4]). This is a weakly elliptic scalar operator and, up to an additive con-
stant, the fundamental solution associated with LB as in [70, Theorem 1.4.2] is (cf.
[70, Proposition 1.4.34] and [70, (1.4.281)]) the distribution induced by the locally
integrable function
z
E(z) = , z ∈ C. (1.4.78)
πz
Also, by design, we have a factorization
 
 where D
LB = DD  = D = 1 ∂x + i∂y . (1.4.79)
2

For further use, note that

 
 = D = − 1 ∂x + i∂y .
(LB ) = LB and D (1.4.80)
2

Next, fix a Lebesgue measurable nonempty proper subset Ω of R2 , having locally

finite perimeter. Abbreviate σ := H 1 ∂Ω and denote by ν = (ν1, ν2 ) the geometric
measure theoretic outward unit normal to Ω, canonically identified with the complex-
valued function ν = ν1 + iν2 defined at σ-a.e. point on ∂∗ Ω. According to [70,
Convention 1.7.2], for each function u of class 𝒞1 in a neighborhood of ∂∗ Ω, at
σ-a.e. point on ∂∗ Ω we have
  
 u
(−i)Sym(D ; ν) D = 12 ν(∂z̄ u) ∂∗ Ω . (1.4.81)
∂∗ Ω

In particular, since
1
(∂z̄ E)(z) = for each z ∈ C \ {0}, (1.4.82)
πz
it follows that for each z ∈ Ω and σ-a.e. ζ ∈ ∂∗ Ω we have

 ν(ζ)
(−i)Sym(D ; ν(ζ)) D ζ [E(z − ζ)] = − . (1.4.83)
2π(z − ζ)
Finally, recall from (A.0.62) that
1
ν(ζ) dσ(ζ) = dζ on ∂∗ Ω. (1.4.84)
i
Bearing these in mind, we then conclude that the boundary layer potential operators
associated as in formulas (1.4.36), (1.4.39) with the factorization (1.4.79) of Bitsadze
σ(ζ )
operator LB from (1.4.77) act on functions f ∈ L 1 ∂∗ Ω, 1+ |ζ | according to
60 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
1 f (ζ)
D f (z) = dζ for all z ∈ Ω̊, (1.4.85)
2πi ∂∗ Ω ζ−z

and ∫
1 f (ζ)
K f (z) = lim+ dζ for σ-a.e. z ∈ ∂∗ Ω. (1.4.86)
ε→0 2πi ζ−z
y ∈∂∗ Ω
|x−y |>ε

Surprisingly, these are the boundary-to-domain and boundary-to-boundary versions

of the Cauchy integral operator associated with Ω (compare with the discussion
in Example 1.4.9). Let us also remark that [68, Proposition 5.6.7] guarantees the
existence of the limit in (1.4.86), and if Ω is in fact a Lebesgue measurable set whose
topological boundary ∂Ω is countably rectifiable (of dimension 1) and has locally
finite H 1 measure (in particular, if ∂Ω is a UR set) then said limit exists for σ-a.e.
point in ∂Ω.
+
Example 1.4.11 Work in the three-dimensional setting and denote by Δ = 3j=1 ∂j2
the Laplacian in R3 . Use this to define the second-order homogeneous constant
coefficient system
Δ0 0
% (
L := && 0 Δ 0 )) = ΔI3×3 . (1.4.87)
' 0 0 Δ*
In other words, L is simply the three-dimensional “vector Laplacian,” which in
particular means that L  = L. As is well known, the three-dimensional vector
Laplacian may be written as

L = ∇div − curl curl, (1.4.88)

hence

2
L=  j D j where
D
j=1
(1.4.89)
1 := ∇,
D D1 := div, 2 := −curl,
D D2 := curl.
As noted earlier in (1.4.60)-(1.4.61), we may recast (1.4.89) simply as
!
div
 where D
L = DD  := (∇, −curl) and D := , (1.4.90)
curl

which we shall call the Maxwell factorization of the vector Laplacian. If EΔ is as in

(A.0.65) with n = 3, then
E L = EΔ I3×3 . (1.4.91)
Also, ! !
∇ −div
 =
D = . (1.4.92)
−curl −curl
Consequently, with the differential operators acting on columns,
1.4 Examples and Alternative Points of View 61

−∂ E −∂ E −∂ E
! EΔ 0 0 % 1 Δ 2 Δ 3 Δ(
−div & % ( & 0 ∂3 EΔ −∂2 EΔ ))
 E L 
D = & 0 EΔ 0 )) = && (1.4.93)
−curl &−∂3 EΔ 0 ∂1 EΔ ))
' 0 0 EΔ *
' ∂2 EΔ −∂1 EΔ 0 *
hence, further,

−∂ E 0 −∂3 EΔ ∂2 EΔ
  % 1 Δ (
 E L 
D &
= &−∂2 EΔ ∂3 EΔ 0 −∂1 EΔ )) . (1.4.94)
'−∂3 EΔ −∂2 EΔ ∂1 EΔ 0 *

To proceed, fix a Lebesgue measurable nonempty proper subset Ω of R3 , having

locally finite perimeter. Abbreviate σ := H 2 ∂Ω, and denote by ν = (ν1, ν2, ν3 )
the geometric measure theoretic outward unit normal to Ω. Having picked a vector-
  3
σ(y)
valued function f ∈ L 1 ∂∗ Ω, 1+ |y | 2
we therefore have
# $
ν· f
(−i)Sym(D; ν) = . (1.4.95)
ν× f
  
 E L   Sym(D; ν) f is identified, thanks to (1.4.93) and (1.4.95), as the
Then (−i) D
vector

−(∂ E )(ν · f ) − (∂3 EΔ )(ν × f )2 + (∂2 EΔ )(ν × f )3

% 1 Δ (
&−(∂2 EΔ )(ν · f ) + (∂3 EΔ )(ν × f )1 − (∂1 EΔ )(ν × f )3 ) . (1.4.96)
& )
−(∂
' 3 Δ E )(ν · f ) − (∂ E
2 Δ )(ν × f ) 1 + (∂ E
1 Δ )(ν × f ) 2*

Ultimately, this analysis leads to the conclusion that the double layer potential op-
erator associated with the three-dimensional vector-Laplacian (1.4.87) starting from
the Maxwell factorization (1.4.90) takes the form
∫  
   
D f (x) = (∇EΔ )(x − y) × ν × f (y) − ν · f (y)(∇EΔ )(x − y) dσ(y) (1.4.97)
∂∗ Ω

at each point x ∈ Ω̊. Likewise, the principal-value (or boundary-to-boundary) dou-

ble layer potential operator associated with the three-dimensional vector-Laplacian
(1.4.87) starting from the Maxwell factorization (1.4.90) is given by
∫  
   
K f (x) = lim+ (∇EΔ )(x − y) × ν × f (y) − ν · f (y)(∇EΔ )(x − y) dσ(y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
(1.4.98)
at σ-a.e. point x ∈ ∂∗ Ω. Integral operators of this sort are relevant in the treatment
of boundary value problems arising in the theory of electromagnetism; cf., e.g.,
[77]. The result recorded in [68, Proposition 5.6.7] guarantees the existence of the
62 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

limit specified in (1.4.98), and tells us that if Ω is in fact a Lebesgue measurable set
whose topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and
has locally finite H n−1 measure (in particular, if ∂Ω is a UR set) then the limit in
(1.4.98) actually exists for σ-a.e. point in ∂Ω.
Lastly, we wish to remark that the boundary-to-domain double layer (1.4.97),
which has been associated as in (1.4.36) with the Maxwell factorization of the vector
Laplacian given in (1.4.90), reproduces the null-solutions of the first-order system D,
i.e., divergence-free, curl-free, vector fields exhibiting proper nontangential boundary
behavior (in the sense of (1.4.58)-(1.4.59)).

Example 1.4.12 Work+in the Clifford algebra context and consider the factorization
of the Laplacian Δ = nj=1 ∂j2 in Rn given by


n
 where D
Δ = DD  := D := i e j ∂j . (1.4.99)
j=1

To implement the recipes given in (1.4.36)-(1.4.39) for this factorization, we find it

convenient to re-frame the discussion in the matrix formalism associated with the
Clifford algebra C n , as described in [68, (6.4.10)-(6.4.11)]. In such a scenario,
n ×2 n
identify e j := E jn ∈ R2 for each j ∈ {1, . . . , n}
(1.4.100)
and take  to be the ordinary multiplication of matrices.

Hence, if we abbreviate M := 2n , then

C n is canonically identified with the sub-algebra

(1.4.101)
of C M×M generated by the matrices {E jn }1≤ j ≤n .

Next, in the current language, the Dirac-type operator D from (1.4.99) becomes the
first-order, homogeneous, constant (complex) coefficient M × M system

n
D=i E jn ∂j . (1.4.102)
j=1

Furthermore, the second-order, homogeneous, constant (real) coefficient weakly

elliptic M × M system L := ΔI M×M , where I M×M denotes the M × M identity
 with D as in (1.4.102) and D
matrix, factors as DD  := D. Then L  = L, so
E L  = EΔ I M×M where EΔ is the standard fundamental solution for the Laplacian
(recalled in (A.0.65)). In addition, from [68, (6.4.16)] (cf. also [68, Lemma 6.4.1])
we see that
 = D = D.
D (1.4.103)
As a consequence of the aforementioned properties and [68, (6.4.16)], we may then
compute
1.4 Examples and Alternative Points of View 63
   
 E L 
D = DEΔ I M×M
 
n  
n
= i (∂j EΔ )E jn = (−i) (∂j EΔ )E jn . (1.4.104)
j=1 j=1

To proceed, fix a Lebesgue measurable nonempty proper subset Ω of Rn , having

locally finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn )
the geometric measure theoretic outward unit normal to Ω. Then

  
n
(−i)Sym D; ν = i ν j E jn (1.4.105)
j=1

and (1.4.104) gives

(−i)  x j − y j n
n
 
 E L 
D (x − y) = E
ωn−1 j=1 |x − y| n j (1.4.106)
for all x, y ∈ Rn with x  y.

In view of the identities recorded in (1.4.105)-(1.4.106), and after eventually reverting

to the standard Clifford algebra formalism, we then see that the boundary layer
potentials associated with the Laplacian as in (1.4.36)-(1.4.39)  starting from the
factorization given in (1.4.99) act on arbitrary functions f ∈ L ∂∗ Ω, 1+σ(y)
1
|y | n−1
⊗C n
according to
∫
1 x−y
D f (x) =  ν(y)  f (y) dσ(y) for all x ∈ Ω̊, (1.4.107)
ωn−1 ∂∗ Ω |x − y| n

and
∫
1 x−y
K f (x) = lim+  ν(y)  f (y) dσ(y) (1.4.108)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

for σ-a.e. x ∈ ∂∗ Ω. These are the boundary-to-domain and boundary-to-boundary

versions of the Cauchy-Clifford integral operator (cf. (A.0.53) and (A.0.54)). Once
again, [68, Proposition 5.6.7] ensures the existence of the limit specified in (1.4.108),
and tells us that if Ω is a Lebesgue measurable set whose topological boundary ∂Ω
happens to be countably rectifiable (of dimension n − 1) and has locally finite H n−1
measure (in particular, if ∂Ω is a UR set) then the limit in (1.4.108) exists for σ-a.e.
point in ∂Ω.
Let us momentarily adopt the latter geometric setting, i.e., assume for the time
being that Ω is a Lebesgue measurable set whose topological boundary ∂Ω is
countably rectifiable (of dimension n − 1) and has locally finite H n−1 measure (this
is the case if, in particular, ∂Ω is a UR set). In such a context, the principal-value
64 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

integral operator K # from (1.4.40) acts on each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
⊗C n
according to
∫
−1 x−y
K # f (x) = lim+ ν(x)   f (y) dσ(y) (1.4.109)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂∗ Ω, which is the “transpose” Cauchy-Clifford

+ operator (cf. (1.6.1)).
On the other hand, if for the same Laplacian Δ = nj=1 ∂j2 we now consider the
factorization
∂1
% .. ()
Δ = DD  := (∂1, . . . , ∂n ) and D := &
 where D .) (1.4.110)
&
'∂n *
in place of (1.4.99), then the boundary layer potential operators associated as in
(1.4.36), (1.4.39) with some Lebesgue measurable nonempty proper subset Ω of
Rn of locally finite perimeter and the n-dimensional Laplacian, starting from the
factorization (1.4.110), now become
∫
1 ν(y), y − x
D f (x) = f (y) dσ(y), ∀x ∈ Ω̊, (1.4.111)
ωn−1 ∂∗ Ω |x − y| n

and, respectively,
∫
1 ν(y), y − x
K f (x) = lim+ f (y) dσ(y) (1.4.112)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

for σ-a.e. x ∈ ∂∗ Ω, i.e., the boundary-to-domain and boundary-to-boundary versions

of the n-dimensional harmonic double layer operator (cf. [70, (2.5.202), (2.5.203)]).
Also, if Ω is a Lebesgue measurable set whose topological boundary ∂Ω is
countably rectifiable (of dimension n − 1) and has locally finite H n−1 measure, then
the boundary layer potential operator associated with the n-dimensional Laplacian
as in (1.4.40) starting from the factorization (1.4.110) now becomes
∫
1 ν(x), x − y
K f (x) = lim+
#
f (y) dσ(y) for σ-a.e. x ∈ ∂∗ Ω,
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε
(1.4.113)
which is the “transpose” principal-value harmonic double layer potential operator
(cf. (A.0.102)).
In particular, this highlights the fact that both the Cauchy-Clifford integral operator
and the harmonic double layer arise in the same fashion, i.e., as boundary layer
1.4 Examples and Alternative Points of View 65

potential operators associated with two particular factorizations of the Laplacian in

Rn .
Given that the factorization (1.4.99) of the Laplacian involves two copies of the
standard Dirac operator in Rn , from (1.4.58)-(1.4.59) we conclude that boundary-
to-domain double layer (1.4.107) (associated as in (1.4.36) with said factorization)
reproduces null-solutions of D, i.e., monogenic functions in Ω, with proper nontan-
gential boundary behavior. See [70, (1.2.30)] in this regard. By way of contrast, in
(1.4.110) we have taken D to be the gradient operator in Rn , so its null-solutions
are precisely locally constant functions in Ω. A suitable reproducing formula for the
corresponding double layer (cf. (1.4.111)) in this scenario is provided in Proposi-
tion 1.3.6.

Example 1.4.13 There are a number of natural variants of Example 1.4.12. First,
bring in the Clifford algebra C n , but work in Rn+1 . Label the variables in Rn+1
as x0, x1, . . . , xn , and factor the Laplacian in Rn+1 , regarded as the M × M system
L := ΔI2n ×2n , in the following fashion:

n 
n
 where D
Δ = DD  := ∂x0 − e j ∂x j and D := ∂x0 + e j ∂x j . (1.4.114)
j=1 j=1

With this factorization, we can then associate double layer potentials as in (1.4.36)
and (1.4.39). To be more specific, observe that L  = L, hence E L  = EΔ I M×M
where EΔ is the standard fundamental solution for the Laplacian in Rn+1 (as in
(A.0.65) with n replaced by n + 1). Once again relying on [68, (6.4.16)] (or using
[68, Lemma 6.4.1]), in place of (1.4.103) we now obtain

n
 = −∂x0 −
D e j ∂x j = −D. (1.4.115)
j=1

As a consequence of these properties and the fact that the transpose of e j (identified
with a matrix, as in (1.4.100)) is −e j for each j ∈ {1, . . . , n}, we therefore arrive at

    
n
 
 E L 
D = − DEΔ I M×M = −∂x0 EΔ + ∂x j EΔ e j . (1.4.116)
j=1

Thus, for all x, y ∈ Rn with x  y we have

1  xj − yj
n
  1 x0 − y0
 E L 
D (x − y) = − + ej . (1.4.117)
ωn |x − y| n+1 ωn j=1 |x − y| n+1

Next, fix a Lebesgue measurable nonempty proper subset Ω of Rn+1 , of locally finite

perimeter. If we abbreviate σ := H n ∂Ω and denote by ν = (ν0, ν1, . . . , νn ) the
geometric measure theoretic outward unit normal to Ω, then
66 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

  
n
(−i)Sym D; ν = ν0 + ν j e j at σ-a.e. point on ∂∗ Ω. (1.4.118)
j=1

The discussion above shows that the boundary layer potentials associated with the
(n + 1)-dimensional Laplacian as in (1.4.36)-(1.4.39) starting from the factorization
given in (1.4.114) act on arbitrary functions f ∈ L ∂∗ Ω, 1+σ(y)
1
|y | n ⊗ C n according to

∫ −(x − y ) + +n (x − y )e
1 0 0 j=1 j j j
D f (x) =
ωn |x − y| n+1
∂∗ Ω
 
n
 ν0 (y) + ν j (y)e j  f (y) dσ(y) (1.4.119)
j=1

at each point x ∈ Ω̊, and, respectively,

∫ +n
1 −(x0 − y0 ) + j=1 (x j − y j )e j
K f (x) = lim+
ε→0 ωn |x − y| n+1
y ∈∂∗ Ω
|x−y |>ε
 
n
 ν0 (y) + ν j (y)e j  f (y) dσ(y) (1.4.120)
j=1

at σ-a.e. point x ∈ ∂∗ Ω (see [68, Proposition 5.6.7] for the existence of the limit in
(1.4.120)). These should be compared with the Cauchy-Clifford integral operators
from (A.0.53) and (A.0.54). It may be verified without difficulty that, corresponding
√
to n = 1 (when there is only one imaginary unit e1 , which we identify with i = −1),
the operators (1.4.119)-(1.4.120) reduce precisely to the boundary-to-domain and
the boundary-to-boundary Cauchy integral operators in the plane, from (1.4.85)-
(1.4.86).
Second, denote by H the skew field of quaternions. Work in R4 , where we label
variables by x0, x1, x2, x3 . Then, if i, j, k are the standard anticommuting imaginary
units in H, we may factor the Laplacian in R4 as

 where D
Δ = DD  := ∂x0 − i∂x1 − j∂x2 − k∂x3
(1.4.121)
and D := ∂x0 + i∂x1 + j∂x2 + k∂x3 .

Once again, with this factorization we may associate double layer potentials as in
(1.4.36) and (1.4.39). To implement this, much as in the past it is convenient to
identify the multiplicative unit 1 in H with I4×4 , and regard i, j, k as antisymmetric
4 × 4 matrices, namely
1.4 Examples and Alternative Points of View 67

0 −1 0 0 0 0 −1 0 0 0 0 −1
% ( % ( % (
&1 0 0 0 ) &0 0 0 1)) &0 0 −1 0 ))
i ≡ && ),
) j ≡ && , k ≡ && . (1.4.122)
&0 0 0 −1) &1 0 0 0)) &0 1 0 0 ))
'0 0 1 0 * '0 −1 0 0* '1 0 0 0*

Then
 = −∂x0 − i∂x1 − j∂x2 − k∂x3 = −D.
D (1.4.123)
Next, denote by EΔ the standard fundamental solution for the Laplacian Δ in R4
(defined as in (A.0.65) with n := 4). Then L := ΔI4×4 factors as in (1.4.121), and
L  = L, so E L  = EΔ I4×4 . Consequently,
   
 E L 
D = − ∂x0 EΔ − (∂x1 EΔ )i − (∂x2 EΔ )j − (∂x3 EΔ )k

= −∂x0 EΔ + (∂x1 EΔ )i + (∂x2 EΔ )j + (∂x3 EΔ )k (1.4.124)

As such, for any pair of distinct points x, y ∈ R4 we have

  
 E L   (x − y)
D

1 x0 − y0 1 x1 − y1 1 x2 − y2 1 x3 − y3
=− + i+ j+ k. (1.4.125)
ω3 |x − y| 4 ω3 |x − y| 4 ω3 |x − y| 4 ω3 |x − y| 4

Going further, fix a Lebesgue measurable nonempty proper subset Ω of R4 , of locally

finite perimeter. If we abbreviate σ := H 3 ∂Ω and denote by ν = (ν0, ν1, ν2, ν3 ) the
geometric measure theoretic outward unit normal to Ω, then
 
(−i)Sym D; ν = ν0 + ν1 i + ν2 j + ν3 k at σ-a.e. point on ∂∗ Ω. (1.4.126)

These computations allow us to conclude that the boundary layer potentials asso-
ciated with the 4-dimensional Laplacian as in (1.4.36)-(1.4.39)
 starting from the
σ(y)
factorization given in (1.4.121) act on functions f ∈ L ∂∗ Ω, 1+ |y |3 ⊗ H according
1

to12
∫
1 −(x0 − y0 ) + (x1 − y1 )i + (x2 − y2 )j + (x3 − y3 )k
D f (x) = ·
ω3 |x − y| 4
∂∗ Ω

· ν0 (y) + ν1 (y)i + ν2 (y)j + ν3 (y)k · f (y) dσ(y) (1.4.127)

at each point x ∈ Ω̊, and, respectively,

12 with “·” denoting the multiplication in the field of quaternions H

68 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
1 −(x0 − y0 ) + (x1 − y1 )i + (x2 − y2 )j + (x3 − y3 )k
K f (x) = lim+ ·
ε→0 ω3 |x − y| 4
y ∈∂∗ Ω
|x−y |>ε

· ν0 (y) + ν1 (y)i + ν2 (y)j + ν3 (y)k · f (y) dσ(y) (1.4.128)

at σ-a.e. point x ∈ ∂∗ Ω (see [68, Proposition 5.6.7] for the existence of the limit
in (1.4.128)). One can make the case that these should be labeled as the Cauchy-
Hamilton integral operators associated with the set Ω.
Let us describe a more inclusive point of view, which also allows the consideration
of Cayley’s algebra of octonions. We briefly discuss a general algebraic construction.
Assume A = (A, +, ·) is a real, unital, associative algebra endowed with a linear
involution a → a c . Assume the latter to be a conjugation, meaning that (ab)c = bc a c
for every a, b ∈ A. In particular, if 1 A is the multiplicative unit in A, then 1cA = 1 A .
Define the trace of an element a ∈ A as Tr A (a) := 12 (a + a c ). Next, consider
the functor A → 𝒦(A) = (A × A, +, ·) where the addition in 𝒦(A) is done
componentwise while the multiplication and conjugation are, respectively, given by

(a, b) · (α, β) := (aα − β c b, βa + bα c ), (a, b)c := (a c, −b). (1.4.129)

Assuming the algebra A is as above, the following claims are straightforward:

(1) 𝒦(A) is a real, unital algebra, equipped with a conjugation.

(2) (1 A , 0) is the multiplicative unit in 𝒦(A). Also, if (e j ) j ∈J are imaginary units
which anticommute in A then (0, 1 A ) together with (e j , 0), (0, e j ), for j ∈ J, are
imaginary units which anticommute in 𝒦(A).
(3) If A is commutative then 𝒦(A) is associative.
(4) If A is associative and if any element has a scalar trace (i.e., if Tr A (a) ∈ R for
any a ∈ A), then 𝒦(A) is an alternative algebra, i.e.,

[x, y, z] := (xy)z − x(yz) (1.4.130)

is trilinear alternate, and any element in 𝒦(A) has a scalar trace.

(5) If Tr A (ab) = Tr A (ba) for any a, b ∈ A, then Tr𝒦(A) (xy) = Tr𝒦(A) (yx) for any
x, y ∈ 𝒦(A).
(6) If A is associative and Tr A (ab) = Tr A (ba) for any a, b ∈ A, then for any
x, y, z ∈ 𝒦(A) one has Tr𝒦(A) ((xy)z) = Tr𝒦(A) (y(zx)) or, equivalently,
Tr𝒦(A) [x, y, z] = 0.
(7) If A is a normed algebra and aa c = |a| 2 for each a ∈ A, then 𝒦(A) becomes
a normed algebra with |x| 2 = x x c = x c x = |a| 2 + |b| 2 for every x = (a, b) in
𝒦(A). In particular, the multiplicative inverse of any element x ∈ 𝒦(A) \ {0}
is x −1 = x c /|x| 2 .
(8) Assume that A is a normed algebra such that aa c = |a| 2 , and so that any element
in A has a scalar trace. Then 𝒦(A) turns into a real Hilbert space with respect
to the pairing
x, y := Tr𝒦(A) (xy c ) = 12 (xy c + yx c ). (1.4.131)
1.4 Examples and Alternative Points of View 69

(9) 𝒦(R) = C, the complex numbers, 𝒦(C) = H, the quaternions, and 𝒦(H) = O,
the Cayley algebra of octonions.

Specializing these general considerations to the case of the Cayley algebra of

octonions shows that the following properties are true:

(i) If i, j, k are the standard anticommuting imaginary units in H, then ξ1 := (i, 0),
ξ2 := ( j, 0), ξ3 := (k, 0), ξ4 := (0, 1), ξ5 := (0, i), ξ6 := (0, j), ξ7 := (0, k) are
anticommuting imaginary units which, along with ξ0 := (1, 0), the multiplicative
unit in O, form a basis for O. Accordingly, one can embed + R → O by identifying
8

each point x = (xα )α ∈ R with the Cayley number xα ξα ∈ O.

8

(ii) Equipped with the pairing (1.4.131), the Cayley algebra O becomes a real,
eight-dimensional Hilbert space, with {ξα }α an orthonormal basis.
+
(iii) If x = xα ξα ∈ O, with xα ∈ R, then x c = x0 − x1 ξ1 − · · · − x7 ξ7 . In particular,
TrO (x) = x0 .
(iv) For each x, y ∈ O one has the identities x(xy) = x 2 y, (xy)y = xy 2 , and also
(xy)y −1 = x = y −1 (yx) provided y  0.

In the setting of the Cayley algebra O, we may then introduce a first-order differ-
ential operator of Dirac type, namely

∂ 7
∂
D := + ξα , (1.4.132)
∂ x0 α=1 ∂ xα

which acts on O-valued functions (defined in R8 ) in a natural fashion. If we also set

∂  7
∂
D c := − ξα , (1.4.133)
∂ x0 α=1 ∂ xα

then the transpose of the operator (1.4.132) with respect to the pairing (1.4.131) is
D = −D c . Moreover, the Laplacian in R8 factors as
 where D is as in (1.4.132), and where D
Δ = DD  := D c . (1.4.134)

The key observation in this regard is that

x · y, z = y, x c · z , ∀x, y, z ∈ O. (1.4.135)

Indeed, by (1.4.131), this is equivalent to TrO ((xy)z c ) = TrO (y(z c x)) which is true,
by virtue of item (6) above. With (1.4.135) in hand, it is then straightforward to
justify (1.4.134).
Having established the factorization in (1.4.134), we may then associate dou-
ble layer potentials as in (1.4.36) and (1.4.39). Fix example, suppose a Lebesgue
measurable nonempty proper subset Ω of R8 , having locally finite perimeter has
been given. Denote by ν = (ν1, . . . , ν8 ) the geometric measure theoretic outward
70 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

unit normal to Ω and abbreviate σ := H 7 ∂Ω. Then the same type of argument as
in (1.4.116)-(1.4.118) shows that the boundary layer potentials associated with the
Laplacian as in (1.4.36)-(1.4.39)
 based on the factorization given in (1.4.134) act on
σ(y)
arbitrary functions f ∈ L ∂∗ Ω, 1+
1
|y | 7
⊗ O according to
∫
1 −(x − y)c  
D f (x) = · ν(y) · f (y) dσ(y) for all x ∈ Ω̊, (1.4.136)
ω7 ∂∗ Ω |x − y| 8

and, with the existence of the principal-value limit guaranteed by [68, Proposi-
tion 5.6.7],
∫
1 −(x − y)c  
K f (x) = lim+ · ν(y) · f (y) dσ(y) (1.4.137)
ε→0 ω7 |x − y| 8
y ∈∂∗ Ω
|x−y |>ε

for σ-a.e. x ∈ ∂∗ Ω, where the geometric measure theoretic outward unit+normal to

Ω is identified with an octonion-valued function ν : ∂∗ Ω → O, via ν = 8α=0 ξα να .
These may be regarded as the boundary-to-domain and the boundary-to-boundary
Cauchy-Cayley integral operators associated with Ω.
In all these cases, there are natural reproducing formulas for the appropriate
double layers, involving null-solutions of corresponding first-order operator D, as
indicated in (1.4.58)-(1.4.59).

Example 1.4.14 Recall the “deformation tensor” acting on any given vector-valued
distribution u = (u j )1≤ j ≤n defined in an open subset of Rn according to
  
Def u := 12 ∂k u j + ∂j uk . (1.4.138)
1≤ j,k ≤n

Its (real) transpose is acting on matrix-valued distribution w = (w jk )1≤ j,k ≤n accord-

ing to   
Def  w = − 12 ∂k w jk + ∂k wk j (1.4.139)
1≤ j ≤n

(cf. [70, (1.3.58)]). Let us also bring in the Jacobian operator. Specifically, for each
vector-valued distribution u = (u j )1≤ j ≤n defined in an open subset of Rn we set
 
∇u := ∂k u j 1≤ j,k ≤n . (1.4.140)

Its (real) transpose sends a matrix-valued distribution w = (w jk )1≤ j,k ≤n into

 
∇ w = − ∂k w jk 1≤ j ≤n (1.4.141)

(cf. [70, (1.3.57)]). In relation to these operators, it has been noted in [70, (1.7.44)]
that, having fixed some Lamé moduli μ, λ ∈ C then for each ζ ∈ C we may express
the complex Lamé system

Lλ,μ := μ Δ + (λ + μ)∇div (1.4.142)

1.4 Examples and Alternative Points of View 71

as
ζ D
Lλ,μ = D (1.4.143)
where

 Def
% (
ζ := − 2ζ Def , (λ + μ − ζ)∇, −(μ − ζ)∇ and D := & div )
D (1.4.144)
& )
' ∇ *
are homogeneous, constant coefficient, first-order systems in Rn . For future use, let
ζ is
us observe that the (real) transpose of D

−2ζ Def
% (
 &
Dζ = & −(λ + μ − ζ)div)) . (1.4.145)
' −(μ − ζ)∇ *
As in [70, (1.7.46)], we agree to abbreviate

Aζ := AD ζ ,D for each ζ ∈ C, (1.4.146)

where AD ζ ,D is the coefficient tensor associated with the systems D ζ , D as in [70,
(1.7.39)].
To proceed, fix a Lebesgue measurable nonempty proper subset Ω of Rn , having
locally finite perimeter. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. For each ζ ∈ C recall
the conormal derivative operator from [70, (1.7.48)]. According to [70, Conven-
tion 1.7.2], whenever u is a vector-valued function of class 𝒞1 in a neighborhood of
∂∗ Ω, at σ-a.e. point on ∂∗ Ω we have
A
ζ ; ν)(Du)
∂ν ζ u = (−i)Sym( D

= 2ζ (Def u)ν + (λ + μ − ζ)(divu)ν + (μ − ζ) (∇u)ν on ∂∗ Ω, (1.4.147)

A
for each ζ ∈ C. As noted earlier, ∂ν ζ is a one-parameter family of conormal derivative
operators, containing the traction conormal derivative operator [70, (1.7.49)] (to
which (1.4.147) reduces when ζ = μ), and also the pseudo-stress conormal derivative
operator defined when 3μ + λ  0 as in [70, (1.7.52)], corresponding to the choice
of ζ made in [70, (1.7.53)].
Going further, the goal is to explicitly identify the format of the double layer
potential operator produced by the recipe in (1.4.36), when the latter is applied for
the factorization of the Lamé system given in (1.4.143). To this end, in view of
the last
 formula
 in (1.4.144)
 it follows that for each given vector-valued function
n
f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
we have
72 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 
1
ν⊗ f + f ⊗ν
%2 (
(−i)Sym(D; ν) f = && ν· f ).
) (1.4.148)
' f ⊗ν *
For economy of space, the right-hand side in (1.4.148) has been written as a three-
component “vector” whose first and last components are n × n ordinary matrices. In
fact, this is a genuine column-vector
 with
 2n2 + 1 components, with the convention
that the n × n matrix 2 ν ⊗ f + f ⊗ ν is actually displayed as a column-vector with
1

n2 components, plus a similar convention for the n × n matrix f ⊗ ν.

Pressing on, to ensure the weak ellipticity of Lλ,μ assume

μ  0 and λ + 2μ  0 (1.4.149)

(cf. [70, (1.3.9)]) and bring in the matrix-valued fundamental solution E = E Lλ, μ of
Lμ,λ from [70, Proposition 1.4.4]. Our job is to identify the integrand in the last line
of (1.4.36), i.e.,
    
(−i) Dζ E L   (x − y)Sym D; ν(y) f (y). (1.4.150)
λ, μ

In this regard, first note that E Lλ, μ = E, since Lλ,μ is symmetric, so from (1.4.145)
we conclude that
−2ζ Def E. 1 −2ζ Def E. 2 ··· −2ζ Def E. n
% (
  &
Dζ E Lλ, μ = &−(λ + μ − ζ)divE. 1 −(λ + μ − ζ)divE. 2 · · · −(λ + μ − ζ)divE. n ))
' −(μ − ζ)∇E. 1 −(μ − ζ)∇E. 2 · · · −(μ − ζ)∇E. n *
(1.4.151)
where E. 1, . . . , E. n are the columns of E. To be concise, the right-hand side in
(1.4.151) has been written as a 3 × n “matrix” whose first and third rows are n × n
ordinary matrices. In reality, this is a genuine (2n2 +1)×n matrix with the convention
that each n × n matrix Def E. j is actually displayed as a column-vector with n2
components, plus a similar convention for each n × n matrix ∇E. j .
At this stage, we may use (1.4.148) and (1.4.151) (keeping in mind their respective
interpretations) to identify (1.4.150) as the vector
 
−ζ(Def E) ν ⊗ f + f ⊗ ν − (λ + μ − ζ)(ν · f )(divE) − (μ − ζ) (∇E)( f ⊗ ν), (1.4.152)

with the variables x, y suppressed. If for any two

+ matrices U = (U jk )1≤ j,k ≤n and
V = (Vjk )1≤ j,k ≤n we define the scalar U s V := nj,k=1 U jk Vjk then, above,
 
(Def E) ν ⊗ f + f ⊗ ν is the vector whose
  (1.4.153)
j-th component is (Def E. j ) s ν ⊗ f + f ⊗ ν ,

while  
divE := divE. j 1≤ j ≤n, (1.4.154)
and
1.4 Examples and Alternative Points of View 73

(∇E)( f ⊗ ν) is the vector whose

(1.4.155)
j-th component is (∇E. j ) s ( f ⊗ ν).
Ultimately, this analysis leads to the conclusion that the double layer potential op-
erator associated as in (1.4.36) with the Lamé system (1.4.142) starting from the
factorization (1.4.143)-(1.4.144) takes the form
∫ 
 
D A ζ f (x) = − ζ(Def E)(x − y) ν ⊗ f + f ⊗ ν (y)
∂∗ Ω

− (λ + μ − ζ)(ν · f )(y)(divE)(x − y)

− (μ − ζ) (∇E)(x − y)( f ⊗ ν)(y) dσ(y) (1.4.156)

at each point x ∈ Ω̊, with the conventions in (1.4.153)-(1.4.155) in effect.

Likewise, the principal-value (or boundary-to-boundary) double layer potential
operator associated as in (1.4.39) with the Lamé system (1.4.142) starting from the
factorization (1.4.143)-(1.4.144) is given by
∫   
K A ζ f (x) = lim+ − ζ(Def E)(x − y) ν ⊗ f + f ⊗ ν (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
− (λ + μ − ζ)(ν · f )(y)(divE)(x − y)

− (μ − ζ) (∇E)(x − y)( f ⊗ ν)(y) dσ(y) (1.4.157)

at σ-a.e. point x ∈ ∂∗ Ω, again with the conventions made in (1.4.153)-(1.4.155).

The existence of the principal-value limit is guaranteed by [68, Proposition 5.6.7],
which also ensures that if Ω is actually a Lebesgue measurable set whose topological
boundary ∂Ω is countably rectifiable (of dimension n−1) and has locally finite H n−1
measure (hence, in particular, if ∂Ω is a UR set) then the limit in (1.4.157) exists for
σ-a.e. point in ∂Ω.
There are other natural versions of the above construction, and we briefly elaborate
on one such variant. Again, having fixed some Lamé moduli μ, λ ∈ C, for each ζ ∈ C
we may factor the complex Lamé system Lλ,μ recalled in (1.4.142) as

 ζ
Lλ,μ = DD (1.4.158)

for the homogeneous, constant coefficient, first-order systems in Rn given by13

 −2ζ Def
% (
 := Def , ∇, ∇ and Dζ := &(λ + μ − ζ) div) .
D (1.4.159)
& )
' −(μ − ζ)∇ *

13 compared with (1.4.144), these differ only in the placement of the scalar coefficients −2ζ,
λ + μ − ζ, −(μ − ζ), which are now all attached with the second operator
74 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

It is then apparent from definitions that this factorization produces the same
conormal derivative as the one associated with the factorization in (1.4.143),

D,D  ,D
D
i.e., ∂ν ζ = ∂ν ζ and, in fact, even at the level of coefficient tensors we
have AD,D
 ζ = A  ζ ,D = Aζ . Consequently, the layer potentials associated as in
D
(1.4.36), (1.4.39) with the Lamé system (1.4.142) now starting from the factoriza-
tion (1.4.158)-(1.4.159) continue to be given by (1.4.156)-(1.4.157).
For each ζ ∈ C, the boundary-to-domain double layer D A ζ from (1.4.156) has
been associated as in (1.4.36) the factorization DD ζ of the Lamé system Lλ,μ given
in (1.4.158). As such, D A ζ reproduces null-solutions of the first-order system Dζ
exhibiting proper nontangential boundary behavior, in the sense of (1.4.58)-(1.4.59).
We wish to elaborate on the nature of such null-solutions. As seen from (1.4.159),
a vector-valued function ψ = (ψ j )1≤ j ≤n is a null-solutions of system Dζ in an open
set Ω ⊆ Rn if

ζ Def ψ = 0, (λ + μ − ζ) divψ = 0, (μ − ζ)∇ψ = 0, in Ω. (1.4.160)

If ζ  0 then the first condition above implies

∂i ψ j = −∂j ψi in Ω for each i, j ∈ {1, . . . , n}. (1.4.161)

In turn, this forces divψ = 0 (so the second condition in (1.4.160) is automatically
satisfied in this case), and for each i, j, k ∈ {1, . . . , n} we may write

∂i ∂j ψk = ∂i (∂j ψk ) = −∂i (∂k ψ j ) = −∂k (∂i ψ j )

= ∂k (∂j ψi ) = ∂j (∂k ψi ) = −∂j (∂i ψk ) = −∂i ∂j ψk , (1.4.162)

thus ∂i ∂j ψk = 0 in Ω. Hence, locally there exist constants a jk with ak j = −a jk

such that ∂j ψk = a jk . Ultimately this shows that locally (and also globally if Ω
is connected), there exist an antisymmetric matrix A ∈ Cn×n along with a vector
b ∈ Cn such that ψ(x) = Ax + b. Finally, the final condition in (1.4.160) is discarded
if ζ = μ, while if ζ  μ then said condition forces ψ to be a locally constant vector
field in Ω.

Example 1.4.15 Consider the complex Lamé system Lλ,μ in Rn defined in (1.4.142)
 αβLamé moduli λ, μ ∈ C. This system may be written in infinitely many ways as
for
ar s ∂r ∂s 1≤α,β ≤n . For example, we may express Lλ,μ as L A ζ , the system associated
as in (A.0.139) with the coefficient tensor Aζ := AD ζ ,D defined in (1.4.146) for each
ζ ∈ C. Recall that AD ζ ,D is the coefficient tensor associated as in [70, (1.7.39)] with
the systems D ζ , D defined in (1.4.144).

In a first stage, we wish to identify the latter coefficient tensor explicitly. To

accomplish this, define Υ to be the set consisting of elements γ of the form (γ , γ )
with γ  ∈ {1, 2, 3} and the convention that γ  ∈ {1, . . . , n}2 if either γ  = 1 or γ  = 3
while γ  is simply suppressed if γ  = 2. This permits us to identify
1.4 Examples and Alternative Points of View 75

Def
% (  n
 γβ 
&
D = & div )) = bs γ ∈Υ · ∂s (1.4.163)
1≤β ≤n
' ∇ *
s=1

where for each γ = (γ , γ ) ∈ Υ and β ∈ {1, . . . , n} we have set

 
⎧
⎪ 1
δγ (β,s) + δγ (s,β) if γ  = 1,
⎪
⎨
⎪ 2
γβ
bs := δsβ if γ  = 2, (1.4.164)
⎪
⎪
⎪ δ  if γ  = 3,
⎩ γ (β,s)
(with the Kronecker delta’s involving either two pairs or two singletons). Similarly,
we may express
 
n
 αγ 
ζ = − 2ζ Def , (λ + μ − ζ)∇, −(μ − ζ)∇ =
D 
br 1≤α ≤n · ∂r (1.4.165)
γ ∈Υ
r=1

where for each α ∈ {1, . . . , n} and γ = (γ , γ ) ∈ Υ

 
⎧
⎪ ζ δγ (α,r) + δγ (r,α) if γ  = 1,
⎪
⎨
⎪
 αγ
br := (λ + μ − ζ)δαr if γ  = 2, (1.4.166)
⎪
⎪
⎪ (μ − ζ)δ  if γ  = 3.
⎩ γ (α,r)

Then, according to [70, (1.7.39)], we have

 αβ 
Aζ = ar s (ζ) 1≤α,β ≤n (1.4.167)
1≤r,s ≤n

where each entry in Aζ is given by


αβ  αγ γβ
ar s (ζ) := br bs
γ ∈Υ
    
= ζ δγ (α,r) + δγ (r,α) 12 δγ (β,s) + δγ (s,β)
γ ∈ {1,...,n}2

+ (λ + μ − ζ)δαr δsβ

+ (μ − ζ)δγ (α,r) δγ (β,s)
γ ∈ {1,...,n}2
 
= ζ δαβ δr s + δαs δβr + (λ + μ − ζ)δαr δsβ + (μ − ζ)δαβ δr s

= μδαβ δr s + (λ + μ − ζ)δαr δsβ + ζ δαs δβr . (1.4.168)

In summary, the family of coefficients associated with the family of factorizations

of the Lamé system (1.4.142) as in (1.4.143) is explicitly described as
76 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 αβ 
Aζ = a jk (ζ) 1≤ j,k ≤n defined for each ζ ∈ C according to
1≤α,β ≤n
αβ
a jk (ζ) := μδ jk δαβ + (μ + λ − ζ)δ jα δkβ + ζ δ jβ δkα, (1.4.169)
for all j, k, α, β ∈ {1, . . . , n}.

In particular,

αβ
Lλ,μ = a jk (ζ)∂j ∂k = L A ζ for each ζ ∈ C. (1.4.170)
1≤α,β ≤n

In relation to this family of writings, it is of interest to directly identity the format

of the double layer potential operators associated as in (1.3.68), (1.3.18) for the
complex Lamé system Lμ,λ with Lamé moduli μ, λ ∈ C satisfying (1.4.149). For
this system, the fundamental solution E of Lμ,λ from [70, Theorem 1.4.2] has the
explicit form E = (E jk )1≤ j,k ≤n , a matrix whose ( j, k) entry is defined at each point
in Rn \ {0} according to [70, (1.4.72)].
Let us now fix a Lebesgue measurable nonempty proper subset Ω of Rn , having
locally finite perimeter. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. In such a setting, with
each choice of ζ ∈ C, associate a double layer potential operator K A ζ as in (1.3.68).
A direct computation based on [70, (1.4.72)], (1.4.169), and (1.3.68) then shows that
the integral kernel Θζ (x, y) of the principal-value double layer potential operator
K A ζ is an n × n matrix whose ( j, k) entry, 1 ≤ j, k ≤ n, is explicitly given by

ζ δ jk x − y, ν(y)
Θ jk (x, y) = −C1 (ζ)
ωn−1 |x − y| n
n x − y, ν(y) (x j − y j )(xk − yk )
− (1 − C1 (ζ))
ωn−1 |x − y| n+2
1 (x j − y j )νk (y) − (xk − yk )ν j (y)
− C2 (ζ) , (1.4.171)
ωn−1 |x − y| n

for σ-a.e. x ∈ ∂∗ Ω and y ∈ ∂∗ Ω, where the constants C1 (ζ), C2 (ζ) ∈ C are defined
as
μ(3μ + λ) − ζ(μ + λ) μ(μ + λ) − ζ(3μ + λ)
C1 (ζ) := , C2 (ζ) := . (1.4.172)
2μ(2μ + λ) 2μ(2μ + λ)
Thus, with notation introduced in (A.0.10), for each ζ ∈ C the integral kernel
Θζ (x, y) of K A ζ may be recast as
1.4 Examples and Alternative Points of View 77

1 x − y, ν(y)
Θζ (x, y) = −C1 (ζ) In×n
ωn−1 |x − y| n
n x − y, ν(y) (x − y) ⊗ (x − y)
− (1 − C1 (ζ))
ωn−1 |x − y| n+2
1 (x − y) ⊗ ν(y) − ν(y) ⊗ (x − y)
− C2 (ζ) , (1.4.173)
ωn−1 |x − y| n
for σ-a.e. x ∈ ∂Ω and y ∈ ∂∗ Ω, where In×n is the n×n identity matrix. Consequently,
  n
for each given vector-valued function f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n−1
and for σ-a.e. point
x ∈ ∂∗ Ω we have
∫
C1 (ζ) x − y, ν(y)
K A ζ f (x) = − lim f (y) dσ(y) (1.4.174)
ωn−1 ε→0+ |x − y| n
y ∈∂∗ Ω
|x−y |>ε
∫
n(1 − C1 (ζ)) x − y, ν(y) x − y, f (y)
− lim (x − y) dσ(y)
ωn−1 ε→0+ |x − y| n+2
y ∈∂∗ Ω
|x−y |>ε
∫
C2 (ζ) ν(y), f (y) (x − y) − x − y, f (y) ν(y)
− lim dσ(y)
ωn−1 ε→0+ |x − y| n
y ∈∂∗ Ω
|x−y |>ε

where the constants C1 (ζ), C2 (ζ) ∈ C are associated with each ζ ∈ C as in (1.4.172).
This agrees with the principal-value double layer potential operator for the Lamé
system defined in (1.4.157).
Likewise, the boundary-to-domain double layer potential operator D A ζ acts on
  n
each vector-valued function f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n−1
according to
∫
C1 (ζ) x − y, ν(y)
D Aζ f (x) = − f (y) dσ(y) (1.4.175)
ωn−1 |x − y| n
∂∗ Ω
∫
n(1 − C1 (ζ)) x − y, ν(y) x − y, f (y)
− (x − y) dσ(y)
ωn−1 |x − y| n+2
∂∗ Ω
∫
C2 (ζ) ν(y), f (y) (x − y) − x − y, f (y) ν(y)
− dσ(y)
ωn−1 |x − y| n
∂∗ Ω

for each x ∈ Ω̊. Reassuringly, this agrees with the (boundary-to-domain) double
layer potential operator associated the Lamé system in (1.4.156).
Example 1.4.16 Consider the factorization of the Laplacian Δ in R2n ≡ Cn given by
78 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

2∂
% z̄1 (
 where D &
 := (2∂z1 , . . . , 2∂zn ) and D := & .. )) .
Δ = DD (1.4.176)
& . )
'2∂z̄n *
Above, for each j ∈ {1, . . . , n} we have set
   
∂z j := 12 ∂x j −i∂y j and ∂z̄ j := 12 ∂x j +i∂y j , assuming the j-th complex
(1.4.177)
variable z j ∈ C is expressed as x j + iy j with x j , y j ∈ Rn .

Fix a Lebesgue measurable nonempty proper subset Ω of R2n ≡ Cn , having locally

finite perimeter. Abbreviate σ := H 2n−1 ∂Ω, and identify the geometric measure
theoretic outward unit normal

ν = (ν1, ν2, . . . , ν2n−1, ν2n ) ∈ R2n (1.4.178)

to Ω with the complex vector

νC := (ν1 + iν2, . . . , ν2n−1 + iν2n ) ∈ Cn . (1.4.179)

A straightforward interpretation of definitions then shows that the boundary lay-

er potentials associated with the Laplacian as in (1.4.36)-(1.4.39) starting from
the factorization
 given in (1.4.176) act on arbitrary complex-valued functions
σ(ζ )
f ∈ L ∂∗ Ω, 1+ |ζ |2n−1 according to
1

∫ 
1 νC (ζ), ζ − z
C
D f (z) = f (ζ) dσ(ζ) for all z ∈ Ω̊, (1.4.180)
ω2n−1 ∂∗ Ω |z − ζ | 2n
+
where u, w C := nj=1 u j w j for each u = (u j ) j ∈ Cn and w = (w j ) j ∈ Cn , is the
Hermitian complex-pairing, and
∫ 
1 νC (ζ), ζ − z
C
K f (z) = lim+ f (ζ) dσ(ζ) (1.4.181)
ε→0 ω2n−1 |z − ζ | 2n
y ∈∂∗ Ω
|x−y |>ε

for σ-a.e. z ∈ ∂∗ Ω. The result in [68, Proposition 5.6.7] guarantees the existence
of the above limit and also ensures that if Ω is in fact a Lebesgue measurable set
whose topological boundary ∂Ω is countably rectifiable (of dimension 2n − 1) and
has locally finite H 2n−1 measure (in particular, if ∂Ω is a UR set) then the limit in
(1.4.181) actually exists for σ-a.e. point in ∂Ω.
As we shall see later on, (1.4.180) and (1.4.181) are precisely the boundary-
to-domain and boundary-to-boundary versions of the Bochner-Martinelli integral
operator (cf. (7.5.2) and (7.5.26)).
Example 1.4.17 Fix two integers n, m ∈ N with n ≥ 2, and let (Rn )m be embedded
in A := ⊕C n , the sum of m copies of C n . That is, if x ∈ (Rn )m = Rnm we write
+n
x = (x1, . . . , xm ) with each x j = α=1 x αj eα ∈ C n for 1 ≤ j ≤ m. In this setting, the
1.4 Examples and Alternative Points of View 79

Dirac operator corresponding to the j-th copy of C n is


n
∂
Dj = eα , (1.4.182)
α=1
∂ x αj

while the global Dirac operator reads D := (D j )1≤ j ≤m . It follows that D is real (i.e.
D = D) and we have the factorization

m
Δ= Δ j = −D∗ D = −D D. (1.4.183)
j=1

Here Δ is the Laplacian in the whole space Rnm and Δ j is the Laplacian in the j-th
factor of the Cartesian product (Rn )m , for each 1 ≤ j ≤ m. Call a smooth C n -valued
function u defined in an open subset of Rnm separately monogenic if D j u = 0
for each j ∈ {1, . . . , m}. Note that this amounts to the requirement that Du = 0 (in
particular, a separately monogenic function is harmonic).
We may recast (1.4.183) as the factorization of L := Δ, the Laplacian in the whole
space Rnm , given by

D
% 1(
&
 := (−D1, . . . , −Dm ) and D := & .. )) .
 where D
Δ = DD (1.4.184)
& . )
' Dm *
Recall the standard fundamental solution for the Laplacian in (Rn )m , i.e., the function
defined for each x ∈ (Rn )m \ {0} as
⎧
⎪ 1 1
⎪
⎪ if nm ≥ 3,
⎪
⎨ ωnm−1 (2 − nm) |x|
⎪ nm−2
E(x) := (1.4.185)
⎪
⎪
⎪
⎪ 1
⎪ ln |x| if nm = 2,
⎩ 2π
where ωnm−1 stands for the area of the unit sphere in Rnm . Observe that

−D1
% (
&
 = & .. )) .
L  = L = Δ and D (1.4.186)
& . )
'−Dm *
From (1.4.185), (1.4.186), and [68, (6.4.16)] we may then compute

−D1 E
        %& . ()  
 EL = D
D  E = & . ) = D1 E, . . . , Dn E . (1.4.187)
& . )
'−Dm E*
80 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

To proceed, let us fix a Lebesgue measurable nonempty proper subset Ω of Rnm , hav-

ing locally finite perimeter. Abbreviate σ := H nm−1 ∂Ω and denote the geometric
measure theoretic outward unit normal to Ω by
 
ν = ν1, ν2, . . . , νnm = (ν j )1≤ j ≤m : ∂ ∗ Ω → (Rn )m → A
+n (1.4.188)
with ν j := α=1 ν αj eα for each j ∈ {1, . . . , m}.

Then
ν
% 1(
 
& .. )
(−i)Sym D; ν = & . ) (1.4.189)
& )
'νm *
so the expression defined for each pair of distinct points x ∈ Rn and y ∈ ∂ ∗ Ω as
    
 E L   (x − y)Sym D; ν(y)
Γ(x, y) := (−i) D

m
1 m
xj − yj
= (D j E)(x − y)  ν j (y) =  ν j (y)
j=1
ωnm−1 j=1
|x − y| nm

 n xα − yα
m 
1 j j
= eα  ν j (y)
ωnm−1 j=1 α=1
|x − y| nm


m  n xα − yα
n 
1 j j β
= ν j (y)eα  eβ (1.4.190)
ωnm−1 j=1 α=1 β=1
|x − y| nm

may be though of as the natural Cauchy kernel in this setting. In particular, the
boundary layer potential operators associated with the nm-dimensional Laplacian as
in (1.4.36)-(1.4.40)
 starting from the factorization (1.4.184) now act on each function
σ(y)
f ∈ L ∂∗ Ω, 1+ |y | nm−1 ⊗ C n according to
1

∫
D f (x) := Γ(x, y)  f (y) dσ(y) for each x ∈ Ω̊, (1.4.191)
∂∗ Ω

and, respectively,
∫
K f (x) := lim+ Γ(x, y)  f (y) dσ(y) for σ-a.e. x ∈ ∂∗ Ω. (1.4.192)
ε→0
y ∈∂∗ Ω
|x−y |>ε

These are the boundary-to-domain and boundary-to-boundary versions of the

Cauchy-Clifford integral operator in Clifford analysis for functions of several vari-
ables. The result in [68, Proposition 5.6.7] ensures the existence of the limit in
(1.4.192) and also guarantees that if Ω actually is a Lebesgue measurable set whose
topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and has lo-
1.4 Examples and Alternative Points of View 81

cally finite H n−1 measure (hence, in particular, if ∂Ω is a UR set) then the limit in
(1.4.192) exists for σ-a.e. point in ∂Ω.
Let us point out that while Γ(x, y) is harmonic in x, it is not separately monogenic
in the variable x when m > 1. Thus, the boundary-to-domain Cauchy-Clifford
operator (1.4.191) is separately monogenic if and only if m = 1. The fact that D
reproduces separately monogenic functions (in this regard see Remark 1.4.6 and
Theorem 1.4.7, as well as [60]) may then be viewed as a version of the classical
Bochner-Martinelli formula in Cn (cf. the discussion in §7.3).

Example 1.4.18 We employ notation and terminology from §7.3 to explain how the
higher-degree Bochner-Martinelli integral operator Bα,β with α, β ∈ {0, 1, . . . , n},
introduced in Definition 7.3.1, fits into the blueprint for producing integral operators
of double layer type presented in (1.4.36). Specifically, from (7.3.15)-(7.3.16) we
know that the complex Laplacian  := − 12 Δ in Cn ≡ R2n may be factored as

 with D
 = DD  := D := (∂¯ + ϑ). (1.4.193)
and
 from (7.3.20)-(7.3.21)
 we know that the double form Γα,β (ζ, z) of type
(α, β), (β, α) defined in (7.3.19) is a (suitably normalized) fundamental solution
for . For further reference, observe that

D ζ Γα,β (ζ, z) = ∂ζ Γα,β (ζ, z) + ϑζ Γα,β (ζ, z).
¯ (1.4.194)

To proceed, fix a Lebesgue measurable nonempty proper subset Ω of R2n ≡ Cn ,

having locally finite perimeter. Denote by ν its geometric measure theoretic outward
unit normal, and abbreviate σ := H 2n−1 ∂Ω. In this context, pick a differential form
 ) 
f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β . (1.4.195)

Then from (7.2.3) and (7.2.16) we know that, at σ-a.e. ζ ∈ ∂∗ Ω,

i i
Sym(D; ν(ζ)) f (ζ) = ν(ζ)0,1 ∧ f (ζ) − ν(ζ)1,0 ∨ f (ζ). (1.4.196)
2 2
On account of (1.4.194), (1.4.196), and simple degree considerations, the general
recipe in (1.4.36) presently yields the double layer operator acting on f from (1.4.195)
according to
∫
1 
D f (z) = − ν 0,1 (ζ) ∧ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫
1 
+ ν 1,0 (ζ) ∨ f (ζ), ϑζ Γα,β (ζ, z) dσ(ζ), ∀z ∈ Ω̊, (1.4.197)
2 ∂∗ Ω C

where the Hermitian inner product ·, · C is taken in the sense of (7.3.6).

The higher-degree Bochner-Martinelli integral operator Bα,β corresponds pre-
cisely to the first line of (1.4.197) (cf. Definition 7.3.1). In this vein, observe that
under the additional assumption that
82 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

f is complex tangential (cf. Definition 7.1.7),

(1.4.198)
i.e., ν(ζ)1,0 ∨ f (ζ) = 0 at σ-a.e. ζ ∈ ∂∗ Ω,

the integral in the second line of (1.4.197) drops out. The bottom line is that
the higher-degree Bochner-Martinelli integral operator Bα,β from
Definition 7.3.1 and the double layer operator D from (1.4.197),
constructed according to the general recipe described in (1.4.36)
(1.4.199)
for the factorization of L :=  given in (1.4.193), agree whenever
acting of differential forms f as in (1.4.195) which also happen to
be complex tangential (cf. (1.4.198)).
Similar considerations apply to the principal-value (boundary-to-boundary) Bochner-
Martinelli integral operator Bα,β from (7.3.68), vis-a-vis to the the principal-value
double layer operator K constructed according to the general blueprint described in
(1.4.39) for the factorization of the complex Laplacian L :=  given in (1.4.193).
Example 1.4.19 There are other factorizations of the Laplacian which yield bound-
ary layer potential operators of interest. For example, in the context of differential
forms, we have (recall that d, δ stand, respectively, for the exterior derivative operator
and its formal transpose)
# $
δ
 
Δ = DD with D := i(d, δ) and D := i , (1.4.200)
d
as well as
 with D
Δ = DD  := D := i(d + δ). (1.4.201)
These factorizations of the Hodge-Laplacian Δ = −(dδ + δd) lead to boundary layer
potentials of the sort discussed in [76] in the context of Riemannian manifolds,
generalizing those considered earlier in Example 1.4.11.
Example 1.4.20 Given a second-order M × M system L in Rn , there are many
choices of a coefficient tensor A which allows us to represent L as L A (the system
associated with A as in (1.3.2)), and all these choices correspond to typically different
double layer potential operators. In fact, this is the case even for scalar operators
(i.e., when M = 1). To illustrate this phenomenon, take the basic case when L = Δ,
the Laplacian in Rn . For any given antisymmetric matrix B = (b jk )1≤ j,k ≤n ∈ Cn×n
we may then write (with the summation convention over repeated indices in effect
throughout)
Δ = (δ jk + b jk )∂j ∂k , i.e., Δ = L A with (M = 1 and)
(1.4.202)
A := (a jk )1≤ j,k ≤n ∈ Cn×n where a jk := δ jk + b jk for 1 ≤ j, k ≤ n.
To proceed, fix a Lebesgue measurable nonempty proper subset Ω of Rn , having
locally finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn )
the geometric measure theoretic outward unit normal to Ω. Recall from [70, Con-
vention 1.7.2] that whenever the function u is of class 𝒞1 in a neighborhood of ∂∗ Ω,
1.4 Examples and Alternative Points of View 83

then at σ-a.e. point on ∂∗ Ω we may write


∂νA u = ν j ak j (∂k u) ∂∗ Ω = ν j (δ jk − b jk )(∂k u) ∂∗ Ω

= ν, (∇u) ∂∗ Ω − ν j b jk (∂k u) ∂∗ Ω
  
= ν, (∇u) ∂∗ Ω − 12 b jk ν j (∂k u) ∂∗ Ω − νk (∂j u) ∂∗ Ω . (1.4.203)

In view of this, if EΔ is the standard fundamental solution for the Laplacian in Rn

(cf. (A.0.65)), then the double layer potential operators D, K associated
 with A and
Ω as in (1.3.18) and (1.3.68), respectively, act on functions f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n−1
according to
∫
A
D f (x) = ∂ν(y) EΔ (x − y) f (y) dσ(y)
∂∗ Ω
∫

= ν(y), ∇y [EΔ (x − y)] f (y) dσ(y)
∂∗ Ω
∫
b jk
− ∂τ j k (y) EΔ (x − y) f (y) dσ(y)
2 ∂∗ Ω
∫
1 ν(y), y − x
= f (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
∫
b jk ν j (y)(yk − xk ) − νk (y)(y j − x j )
− f (y) dσ(y) (1.4.204)
2ωn−1 ∂∗ Ω |x − y| n

at every x ∈ Ω̊, and

∫
1 ν(y), y − x
(K f )(x) = lim+ f (y) dσ(y)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε
∫
b jk ν j (y)(yk − xk ) − νk (y)(y j − x j )
− lim f (y) dσ(y),
2ωn−1 ε→0+ |x − y| n
y ∈∂∗ Ω
|x−y |>ε
(1.4.205)

at σ-a.e. point x ∈ ∂∗ Ω. The existence of the limit in (1.4.205) is guaranteed by

[68, Proposition 5.6.7]. Also, if Ω is a Lebesgue measurable set whose topological
boundary ∂Ω is countably rectifiable (of dimension n − 1) and has locally finite
 the operator K associated with A and Ω as in (1.3.72) acts on
H n−1 measure, then #

functions f ∈ L 1 ∂Ω, 1+σ(y)

|y | n−1
at σ-a.e. point x ∈ ∂∗ Ω according to
84 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
1 ν(x), x − y
(K # f )(x) = lim+ f (y) dσ(y)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε
∫
b jk ν j (x)(xk − yk ) − νk (x)(x j − y j )
− lim f (y) dσ(y),
2ωn−1 ε→0+ |x − y| n
y ∈∂Ω
|x−y |>ε
(1.4.206)

with the existence of the limit once again ensured by [68, Proposition 5.6.7]. Ulti-
mately, the conclusion is that
as is visible from (1.4.203)-(1.4.206), the actual choice of the coef-
ficient matrix A in the writing of the Laplacian in (1.4.202) directly
(1.4.207)
affects the format of the conormal derivative as well as the boundary
layer potentials D, K, and K # associated with A.
It is also of significance to note that, as the above discussion indicates, even in
the case of the (scalar) Laplacian there exist infinitely many conormal derivatives,
hence infinitely many Neumann Problems. For example, if Ω is an Ahlfors regular
domain, κ > 0 is a fixed background aperture parameter, and p ∈ (1, ∞) is a given
integrability exponent, then each boundary value problem of the form
⎧ ∞
⎨ u ∈ 𝒞 (Ω),
⎪
⎪ Δu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t.   κ−n.t. κ−n.t. 
⎪
⎪ ν, (∇u) ∂Ω + 12 b jk ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω = f ∈ L p (∂Ω, σ),
⎩
(1.4.208)

is a legitimate Neumann Problem for the Laplacian in Ω, corresponding to the

conormal derivative ∂νA for the coefficient matrix A as in (1.4.202).
Of course, the same type of phenomenon occurs for a general  αβsecond-order

homogeneous weakly elliptic system L in Rn . Specifically, if A = ar s 1≤r,s ≤n is
1≤α,β ≤M
a coefficient tensor allowing us to represent L = L A as in (1.3.2), then for any other
coefficient tensor
 αβ 
B = br s 1≤r,s ≤n antisymmetric in the lower indices, i.e., satisfying
1≤α,β ≤M (1.4.209)
αβ αβ
br s = −bsr for each r, s ∈ {1, . . . , n} and α, β ∈ {1, . . . , M }

we have L = L A+B and yet the conormal derivative ∂νA as well as the boundary
layer potentials D A, K A, K A# associated with A are typically different from their
counterparts ∂νA+B , D A+B , K A+B , K A+B
# associated with A + B. In view of this, it
is remarkable that certain combinations involving the aforementioned operators are
unaffected when the underlying coefficient tensor changes. This is made precise in
the proposition below.
1.4 Examples and Alternative Points of View 85

Proposition 1.4.21 Let Ω ⊆ Rn , where n ≥ 3, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. Denote by ν the geometric measure theoretic outward unit normal to Ω. Let L
be a homogeneous, weakly elliptic, second-order M × M system in Rn , with complex
constant coefficients, and pick a coefficient tensor A with the property that L = L A.
Recall the single layer potential operator 𝒮 from (1.3.6). Also, let D A and ∂νA be,
respectively, the double layer potential operator and conormal derivative associated
with A and Ω as in (1.3.18) and, respectively, (A.0.184). Finally, fix an aperture
1,1 M
parameter κ > 0 and suppose u ∈ Wloc (Ω) is a vector-valued function satisfying

the nontangential pointwise limits

κ−n.t. κ−n.t.
u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂nta Ω, and
(1.4.210)
∫ ∫  
(Nκ u)(y) Nκ (∇u) (y)
dσ(y) < ∞ and dσ(y) < ∞.
∂Ω 1 + |y| n−1 ∂Ω 1 + |y|
n−2

 αβ 
Then for any other coefficient tensor B = br s 1≤r,s ≤n which is antisymmetric
1≤α,β ≤M
in the lower indices (cf. (1.4.209)) one has
 κ−n.t.   κ−n.t. 
D A u ∂Ω − 𝒮(∂νAu) = D A+B u ∂Ω − 𝒮(∂νA+B u) in Ω. (1.4.211)

Furthermore, a similar result is valid when n = 2 provided ∂Ω is compact.

Proof Denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental solution associ-

ated with L as in [70, Theorem 1.4.2] and write (uβ )1≤β ≤M for the scalar components
of the vector-valued function u. Also, fix an arbitrary index γ ∈ {1, . . . , M }, and pick
an arbitrary point x ∈ Ω. Assume first that either Ω is bounded,
  or ∂Ω is unbound-
ed. In such a scenario, consider the vector field Fx = Fs 1≤s ≤n with components
(throughout, the summation convention over repeated indices is in effect)
βα αβ
Fs := −br s (∂r Eγβ )(x − ·)uα − Eγα (x − ·)bsr ∂r uβ (1.4.212)

defined at L n -a.e. point in Ω. Reasoning as in the proof of [70, Theorem 1.5.1]

(while keeping in mind the antisymmetry property (1.4.209)) we see that
n
Fx ∈ Lloc
1 (Ω, L n ) , divFx = 0 in D (Ω),
 
NκΩ\K Fx ∈ L 1 (∂Ω, σ) if K := B x, 12 dist(x, ∂Ω) , (1.4.213)
κ−n.t.

and the trace Fx exists at σ-a.e. point on ∂nta Ω.

∂Ω

In fact, on account of (1.4.212), (1.3.18), (1.3.6), and (A.0.184), we have

86 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

  κ−n.t.    κ−n.t. 
D A+B u ∂Ω (x) − D A u ∂Ω (x) (1.4.214)
γ γ
∫
     κ−n.t. 
− 𝒮(∂νA+B u) γ (x) + 𝒮(∂νAu) γ (x) = ν · Fx ∂Ω
dσ.
∂∗ Ω

Granted (1.4.213)-(1.4.214), we may invoke [68, Theorem 1.4.1] and conclude

(based on Divergence Formula [68, (1.4.6)], as well as the arbitrariness of the index
γ ∈ {1, . . . , M } and the point x ∈ Ω) that identity (1.4.211) holds when either Ω is
bounded, or ∂Ω is unbounded. In the case when Ω is an exterior domain (a scenario
in which ∂Ω is bounded) we run the same argument as before but, this time, replace
u by ψu where ψ ∈ 𝒞∞ c (R ) is a function satisfying ψ ≡ 1 near ∂Ω. Finally, the
n

argument in the two-dimensional case is similar. 

1.5 Calderón-Zygmund Function Theory for Boundary Layer

Potentials

While certain features of boundary layer potentials are visible straight from invoking
generic results valid in the general setting of spaces of homogeneous type, other,
more delicate (typically cancelation and/or differential calculus sensitive) properties
require fully employing the resourcefulness of the algebraic/geometric ambient. For
example, while properties such as the fact that the boundary-to-boundary single layer
∗  
1 −1
S maps L p into L p if 1 < p < n − 1 and p∗ := p1 − n−1 is directly implied by the
Fractional Integration Theorem (which is valid in general spaces of homogeneous
type). This being said, showing that S has a regularizing/smoothing effect of order
one requires tools from differential calculus. Similar issues arise in connection to
double layer potential operators on Lebesgue and Sobolev spaces.
In this section we shall focus precisely on such aspects. Our first theorem elabo-
rates on the rich Calderón-Zygmund theory which may be developed in relation to
the boundary layer potential operators introduced earlier, for a given weakly elliptic
second-order system and an open set with a UR boundary. For related results for
boundary layer potential operators associated with higher-order weakly elliptic sys-
tems the reader is referred to the work in [78] in the class of Lipschitz domains, and
[40] in the class of uniformly rectifiable domains.
Theorem 1.5.1 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with the
property that ∂Ω is a UR set. Abbreviate σ∗ := H n−1 ∂∗ Ω and σ := H n−1 ∂Ω,
and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward
 αβ  unit normal
to Ω. Also, for some M ∈ N, consider a coefficient tensor A = ar s 1≤r,s ≤n with
1≤α,β ≤M
complex entries, with the property that the M × M homogeneous second-order system
L = L A associated with A in Rn as in (1.3.2) is weakly elliptic (in the sense of [70,
(1.3.3) in Definition 1.3.1]). Finally, consider the boundary layer potentials 𝒮, S,
D, K, K # associated with A and Ω as in (1.3.6), (1.3.62), (1.3.18), (1.3.68), and
(1.3.72), respectively.
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 87

Then, in relation to these operators, the following properties hold.

(i) For each p ∈ [1, ∞) and κ > 0 there exists a finite constant C > 0 with the
M
property that for every f ∈ L p (∂∗ Ω, σ∗ ) one has
 
Nκ (D f ) p ≤ C f [L p (∂∗ Ω,σ∗ )] M if 1 < p < ∞, (1.5.1)
L (∂Ω,σ)
 
Nκ (D f ) 1,∞ ≤ C f [L 1 (∂∗ Ω,σ∗ )] M if p = 1. (1.5.2)
L (∂Ω,σ)

Also, for each exponent p ∈ (1, ∞), each Muckenhoupt weight w ∈ Ap (∂Ω, σ),
and each aperture parameter κ > 0, there exists a constant C ∈ (0, ∞) with the
M
property that for every function f ∈ L p (∂∗ Ω, wσ∗ ) one has
 
Nκ (D f ) p ≤ C f [L p (∂∗ Ω,wσ∗ )] M . (1.5.3)
L (∂Ω,wσ)

Furthermore, given any p ∈ (1, ∞), q ∈ (0, ∞], and κ > 0, there exists a finite
M
constant C > 0 with the property that for every f ∈ L p,q (∂∗ Ω, σ∗ ) one has
 
Nκ (D f ) p, q ≤ C f [L p, q (∂∗ Ω,σ∗ )] M . (1.5.4)
L (∂Ω,σ)

Also, in view of work in [68, §8.5], similar estimates to (1.5.1)-(1.5.4) are true
with the nontangential maximal operator replaced by the tangential maximal
operator (associated as in [68, Definition 8.5.1] with a sufficiently large power;
cf. (A.0.146)).
In addition, whenever 1 < p < ∞ and p ≤ q ≤ ∞, there exists a constant
M
C ∈ (0, ∞) such that for each f ∈ L p (∂∗ Ω, σ∗ ) one has
 1+ n−1 − n 
 
δ∂Ω
p q
∇(D f )  ≤ C f [L p (∂∗ Ω,σ∗ )] M . (1.5.5)
L q (Ω, L n )
 
Finally, if one also imposes the condition that H n−1 ∂Ω\∂∗ Ω = 0 then, as noted
earlier in [69, (10.2.253)-(10.2.254)], there exists some constant C ∈ (0, ∞) with
. M
the property that for each function f ∈ L p (∂Ω, σ)∩ 𝒞η (∂Ω) with p ∈ (1, ∞)
and η ∈ (0, 1) one has14

sup (D f )(x) ≤ C f [H 1,∞ (∂Ω,σ)∗ ] M . (1.5.6)

x ∈Ω

(ii) For each function f belonging to the weighted boundary Sobolev space
  M
L11 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
(cf. (A.0.131)), each aperture parameter κ ∈ (0, ∞), and
each index ∈ {1, . . . , n}, the pointwise nontangential boundary trace
  κ−n.t.
∂ D f ∂Ω
exists (in C M ) at σ∗ -a.e. point on ∂∗ Ω. (1.5.7)

14 informally, this may be thought as the limiting case p = ∞ of (1.5.1)

88 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

p,q M
As a corollary of [69, (11.7.22)], this is true whenever f ∈ L1 (∂∗ Ω, σ∗ )
with p, q ∈ [1, ∞).
Moreover, for each p, q ∈ [1, ∞) and κ > 0 there exists some finite constant
C > 0, depending only on ∂Ω, A, n, κ, p, q, such that for each function
p,q M
f ∈ L1 (∂∗ Ω, σ∗ ) one has
   
Nκ (D f ) p + Nκ (∇D f ) L q (∂Ω,σ) ≤ C f [L p, q (∂∗ Ω,σ∗ )] M
L (∂Ω,σ) 1
(1.5.8)
if p ∈ (1, ∞) and q ∈ (1, ∞),

plus similar estimates in the case when either p = 1 or q = 1, this time with the
corresponding L 1 -norm in the left side replaced by the weak-L 1 (quasi-)norm.
In addition,
if p ∈ (1, ∞) and the weight w belongs to Ap (∂Ω, σ), then there exists
some constant C ∈ (0, ∞) with the property that for each f in the weighted
p M
Sobolev space
 L1 (∂∗ Ω, wσ∗ ) the
 following estimate holds
Nκ (D f ) p + Nκ (∇D f ) L p (∂Ω,wσ) ≤ C f [L p (∂∗ Ω,wσ∗ )] M .

L (∂Ω,wσ) 1
(1.5.9)

In addition, in view of work in [68, §8.5], similar estimates to (1.5.8)-(1.5.9)

are valid with the nontangential maximal operator replaced by the tangential
maximal operator (associated as in [68, Definition 8.5.1] with a sufficiently
large power; cf. (A.0.146)).
Finally, if Ω is also a uniform domain, then for each vector-valued function
p,q M
f = ( fα )1≤α ≤M ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ (1, ∞) and q ∈ (n − 1, ∞) one has
.
D f ∈ 𝒞η (Ω)
M
where η := 1 − (n − 1)/q ∈ (0, 1), (1.5.10)

and there exists a constant C = C(Ω, A, q) ∈ (0, ∞) with the property that

  
M 
n
 
D f  . η ≤C ∂τ fα  q
[𝒞 (Ω)] M jk L (∂∗ Ω,σ∗ )
α=1 j,k=1

≤ C f [L p, q (∂∗ Ω,σ∗ )] M . (1.5.11)

1

(iii) Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operator

M M
K # : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ∗ ) (1.5.12)

is well defined, linear, and bounded. Moreover, corresponding to p = 1, (1.3.72)

induces a well-defined, linear, and bounded operator
M M
K # : L 1 (∂Ω, σ) −→ L 1,∞ (∂∗ Ω, σ∗ ) , (1.5.13)

and the transpose of K # in (1.5.12) is

1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 89
 M  M
K : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ) . (1.5.14)

Corresponding to p = 1, (1.5.14) induces a well-defined, linear, and bounded

mapping
M M
K : L 1 (∂∗ Ω, σ∗ ) −→ L 1,∞ (∂Ω, σ) . (1.5.15)

In fact, similar results are valid for the operators K, K # acting on Lorentz
spaces and Muckenhoupt weighted Lebesgue spaces. Specifically, for each given
p ∈ (1, ∞), q ∈ (0, ∞], and w ∈ Ap (∂Ω, σ), the operators
M M
K : L p,q (∂∗ Ω, σ∗ ) −→ L p,q (∂Ω, σ) , (1.5.16)
M M
K # : L p,q (∂Ω, σ) −→ L p,q (∂∗ Ω, σ∗ ) , (1.5.17)
M M
K : L p (∂∗ Ω, wσ∗ ) −→ L p (∂Ω, wσ) , (1.5.18)
M M
K # : L p (∂Ω, wσ) −→ L p (∂∗ Ω, wσ∗ ) , (1.5.19)

are well-defined, linear, and bounded. Finally, if q > 1 and p, q  ∈ (1, ∞) are
such that 1/p + 1/p = 1 and 1/q + 1/q  = 1, then the (real) transpose of the
operator K in (1.5.16) is the operator K # in (1.5.17) with p, q replaced by p, q .
Also, the (real) transpose of the operator K in (1.5.18) is the operator K # in

(1.5.19) with p replaced by p and w replaced by w 1−p .
(iv) Having fixed some arbitrary aperture parameter κ ∈ (0, ∞), for each given
  M
vector-valued function f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
the following nontangential
boundary trace formula holds:
κ−n.t.
1 
Df = 2I + K f at σ-a.e. point on ∂∗ Ω, (1.5.20)
∂Ω

where I is the identity operator. In particular, the jump-formula (1.5.20) is valid

for each function f in the following function spaces:
M
• f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) (cf. [68, (7.7.106)]);
M
• f ∈ L p,q (∂∗ Ω, σ∗ ) with p ∈ (1, ∞) and q ∈ (0, ∞] (cf. [68, (7.7.107)]);
M
•f ∈ L p (∂∗ Ω, wσ∗ ) with p ∈ (1, ∞) and the weight w ∈ Ap (∂Ω, σ) (cf. [68,
(7.7.104)]);
M
• f ∈ M p,λ (∂Ω, σ) with p ∈ (1, ∞) and λ ∈ (0, n − 1) (cf. [69, (6.2.25)]).

Moreover, as a consequence of Proposition 1.3.6 and (1.5.20) (or, alternatively,

(1.3.50) and [70, (2.3.15)]), it follows that
90 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

if Ω has a compact boundary, then for each λ ∈ C M one has

 1
+ 2 λ at σ-a.e. point on ∂∗ Ω, if Ω is bounded, (1.5.21)
Kλ =
− 12 λ at σ-a.e. point on ∂∗ Ω, if Ω is unbounded.

Finally, from (1.3.44), (1.5.20), (1.3.67), and [68, Proposition 8.8.6] one sees
that for each point xo ∈ Ω and for each index β ∈ {1, . . . , M } one has
1  
2 I + K E . β (· − x o ) ∂Ω
= S ∂νA E. β (· − xo ) at σ-a.e. point on ∂∗ Ω.
(1.5.22)
(v) Under the additional assumption that
 
H n−1 ∂nta Ω \ ∂∗ Ω = 0, (1.5.23)

(which is automatically satisfied if, e.g., Ω is a UR domain to begin with) it

follows that the operator
p M p M
K : L1 (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.5.24)

is well defined, linear, and bounded for each p ∈ (1, ∞). More generally, if
(1.5.23) is satisfied then the operator
p,q M p,q M
K : L1 (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.5.25)

is well defined, linear, and bounded for each p, q ∈ (1, ∞), and
p M p M
K : L1 (∂∗ Ω, wσ∗ ) −→ L1 (∂∗ Ω, wσ∗ ) (1.5.26)

is well defined, linear, and bounded for each p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ).
(vi) Continue to retain the additional assumption made in (1.5.23). Then for
each p, q ∈ (1, ∞) it follows that K # , originally acting on functions from
M
L p (∂∗ Ω, σ) (regarding them as being extended by zero to the entire topo-
logical boundary ∂Ω, and then applying K # in the sense of (1.5.12)), further
extends uniquely to a linear and bounded operator from the negative bound-
p M
ary Sobolev space L−1 (∂∗ Ω, σ∗ ) into itself and, more generally, from the
p,q M
off-diagonal negative Sobolev space L−1 (∂∗ Ω, σ∗ ) into itself. Furthermore,
#
if one adopts the same notation K for said extensions, then the transpose of
(1.5.24) is
p M p M
K # : L−1 (∂∗ Ω, σ∗ ) −→ L−1 (∂∗ Ω, σ∗ ) , (1.5.27)

while the transpose of (1.5.25) is

p,q M p q M
K # : L−1 (∂∗ Ω, σ∗ ) −→ L−1 (∂∗ Ω, σ∗ ) , (1.5.28)
 1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 91

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1.
(vii) Temporarily strengthen the original hypotheses by assuming that Ω is actually
a UR domain. Then given any f belonging to the weighted boundary Sobolev
  M
space L11 ∂Ω, 1+σ(x)
|x | n−1
, at σ-a.e. point x ∈ ∂Ω one has
 
∂νA(D f ) (x) (1.5.29)
# ∫ $
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0
y ∈∂Ω 1≤μ ≤M
|x−y |>ε

where the conormal derivative ∂νA(D f ) is considered as in (A.0.184). As a

p,q M
corollary of [69, (11.7.22)], this is true whenever f ∈ L1 (∂Ω, σ) with
p, q ∈ [1, ∞).
Hence, in such a setting, the conormal derivative of the double layer induces an
operator
 
f → ∂νA D f := ∂νA(D f ) (1.5.30)

which for any given exponents p, q ∈ (1, ∞) is well defined, linear, and bounded
in the context
p,q M M
∂νA D : L1 (∂Ω, σ) −→ L q (∂Ω, σ) . (1.5.31)

Finally, it is apparent from (1.5.29) that (∂νA D) f does not jump across the
boundary (in the sense that it has the same nontangential boundary trace when
considered from Ω+ := Ω and Ω− := Rn \ Ω).
(viii) Continue to assume that Ω is actually a UR domain. Also, pick exponents
p, p, q, q  ∈ (1, ∞) satisfying 1/p + 1/p = 1 and 1/q + 1/q  = 1. Then for any
p,q M q, p M
f ∈ L1 (∂Ω, σ) and g ∈ L1 (∂Ω, σ) one has
∫ ∫
 A      
∂ν D f , g dσ = f , ∂νA D A g dσ (1.5.32)
∂Ω ∂Ω
 
where ∂νA along with D A and, ultimately ∂νA D A , are defined as before with
A now replaced by A . As a consequence, whenever Ω is a UR domain, the
operator (1.5.31) has a unique extension to a well-defined, linear, and bounded
mapping
M q, p M
∂νA D : L p (∂Ω, σ) −→ L−1 (∂Ω, σ) , (1.5.33)

namely the (real) transpose of the operator

 q, p M  M
∂νA D A : L1 (∂Ω, σ) −→ L p (∂Ω, σ) . (1.5.34)
92 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
  −1
(ix) Suppose p ∈ (1, n − 1) and p∗ := 1/p − 1/(n − 1) . Then the boundary-
to-boundary single layer potential operator S induces well-defined, linear, and
bounded mappings
M p ∗, p M
S : L p (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) , (1.5.35)
p, p ∗ M ∗ M
S : L−1 (∂∗ Ω, σ∗ ) −→ L p (∂∗ Ω, σ∗ ) , (1.5.36)

which act in a compatible fashion with one another. Also, if prime is used to
indicate the Hölder conjugate exponent, the (real) transpose of (1.5.35) is the
operator
(p ∗ ), p M  M
S A : L−1 (∂∗ Ω, σ∗ ) −→ L p (∂∗ Ω, σ∗ ) , (1.5.37)

where S A is associated with the coefficient tensor A (the real transpose of A)

in the same manner S has been associated with the original A.
In addition, the operator S from (1.5.36) further extends to a linear and bounded
mapping
p,q M ∗ M M
S : L−1 (∂∗ Ω, σ∗ ) −→ L p (∂∗ Ω, σ∗ ) + L q (∂∗ Ω, σ∗ )
(1.5.38)
for each given integrability exponent q ∈ (1, ∞).

Moreover, corresponding to n ≥ 3 and p = n−1, the following operator is linear

and bounded:

−→ L ∞ (∂Ω, σ)
M M
S : L n−1,1 (∂Ω, σ) . (1.5.39)

Finally, if ∂Ω is actually bounded then the same results are also true for the
exponent p ∈ [n − 1, ∞), this time regarding p∗ as an arbitrary index in (1, ∞)
(unrelated to p).
(x) Given p ∈ (1, n − 1) along with q ∈ (1, ∞), extend the action of the boundary-to-
domain single layer 𝒮, originally defined in (1.3.6), to the off-diagonal negative
p,q M
Sobolev space L−1 (∂∗ Ω, σ∗ ) by setting (again, with prime used to indicate
Hölder conjugation)
 
𝒮 f (x) := [L p, q (∂ Ω,σ )] M E(x − ·) ∂∗ Ω, f [L p, q (∂∗ Ω,σ∗ )] M ∀x ∈ Ω, (1.5.40)
1 ∗ ∗ −1

p,q M
for each f ∈ L−1 (∂∗ Ω, σ∗ ) . Then this is meaningfully defined and agrees
M
with 𝒮 from (1.3.6) when acting on the smaller space L p (∂∗ Ω, σ∗ ) . More-
over,
−→ 𝒞∞ (Ω)
p,q M M
𝒮 : L−1 (∂∗ Ω, σ∗ ) (1.5.41)
is continuous (when the space on the right is equipped with the Frechét topology
of uniform convergence of partial derivatives of any order on compact sets), and
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 93
   
∂γ 𝒮 f (x) = [L p, q (∂ Ω,σ )] M (∂γ E)(x − ·) ∂∗ Ω, f p, q
[L−1 (∂∗ Ω,σ∗ )] M (1.5.42)
1 ∗ ∗

p,q M
for each functional f ∈ L−1 (∂∗ Ω, σ∗ ) , each multi-index γ ∈ N0n , and each
point x ∈ Ω. Moreover,
p,q M
L(𝒮 f ) = 0 in Ω, for each functional f ∈ L−1 (∂∗ Ω, σ∗ ) . (1.5.43)

p,q M
Also, given any κ > 0, for each f ∈ L−1 (∂∗ Ω, σ∗ ) (hence, in particular, for
M
each function f ∈ L p (∂∗ Ω, σ∗ ) ; cf. [69, (11.8.30)]) the nontangential point-
κ−n.t.
wise trace 𝒮 f ∂Ω exists at σ∗ -a.e. point on ∂∗ Ω; in fact, with S f considered in
the sense of (1.5.38) one has
κ−n.t.

𝒮f = S f at σ∗ -a.e. point on ∂∗ Ω. (1.5.44)

∂Ω
  −1
Furthermore, if p ∈ (1, n − 1) and p∗ := 1/p − 1/(n − 1) , then for each
q ∈ (1, ∞) and κ > 0 one has, in a quantitative fashion,
∗
Nκ (𝒮 f ) ∈ L p (∂Ω, σ) + L q (∂Ω, σ)
M
(1.5.45)
p,q
for each f ∈ L−1 (∂∗ Ω, σ∗ ) .

In particular, there exists a finite constant C = C(Ω, A, p, κ) > 0 such that

 
Nκ (𝒮 f ) p ∗ ≤ C f [L p, p ∗ (∂ Ω,σ )] M
L (∂Ω,σ) −1 ∗ ∗
p, p ∗ M
(1.5.46)
for each f ∈ L−1 (∂∗ Ω, σ∗ ) ,

and, as a consequence of this and [69, (11.8.30)],

 
Nκ (𝒮 f ) p ∗ ≤ C f [L p (∂∗ Ω,σ∗ )] M
L (∂Ω,σ)
M
(1.5.47)
for each f ∈ L p (∂∗ Ω, σ∗ ) .

In addition, for each p ∈ [1, n − 1) and each κ > 0 there exists some constant
M
C = C(Ω, p, κ) ∈ (0, ∞) such that, for each function f ∈ L p (∂Ω, σ) ,
κ−n.t.
the nontangential trace (∇𝒮 f ) ∂Ω exists σ∗ -a.e. on ∂∗ Ω,
 
Nκ (∇𝒮 f ) p ≤ C f [L p (∂Ω,σ)] M if p ∈ (1, n − 1), (1.5.48)
L (∂Ω,σ)
 
Nκ (∇𝒮 f ) 1,∞ ≤ C f [L 1 (∂Ω,σ)] M provided p = 1.
L (∂Ω,σ)

Moreover, corresponding to p = n − 1,
if n ≥ 3 then 𝒮 f is bounded in Ω, for each function f
M (1.5.49)
belonging to the Lorentz space L n−1,1 (∂Ω, σ) .
94 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Once again, in view of work in [68, §8.5], similar estimates to those in (1.5.46)-
(1.5.48) are valid with the nontangential maximal operator replaced by the
tangential maximal operator (associated as in [68, Definition 8.5.1] with a
sufficiently large power; (A.0.146)).
Finally, all results in this item are also true for p ∈ [n − 1, ∞), now regarding p∗
as an arbitrary index in (1, ∞) (unrelated to p), in either of the following cases:
(1) n ≥ 3 and ∂Ω is bounded, (2) n = 2 and Ω is bounded, (3) n = 2, Ω is an
exterior domain, and the ordinary nontangential maximal operator is truncated.
(xi) Define the following modified version of the boundary-to-domain single layer
operator
∫
 
𝒮mod f (x) := E(x − y) − E∗ (−y) f (y) dσ(y) for each x ∈ Ω,
∂Ω (1.5.50)
  M
for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, where E∗ := E · 1R \B(0,1)
n .
   M
Then for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 the function 𝒮mod f is well defined,
M
belongs to the space 𝒞∞ (Ω) , and for each multi-index α ∈ N0n with |α| ≥ 1
one has
∫
α
∂ (𝒮mod f )(x) = (∂ α E)(x − y) f (y) dσ(y) for each x ∈ Ω. (1.5.51)
∂Ω

In particular, (1.5.51) implies that

     M
L 𝒮mod f = 0 in Ω for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 , (1.5.52)

while identity (1.5.51) and [70, Theorem 2.5.1] guarantee that, for each index
j ∈ {1, . . . , n}, each aperture parameter κ ∈ (0, ∞), and each function f in the
   M
space L 1 ∂Ω, 1+σ(x) |x | n−1
, the nontangential boundary trace

κ−n.t.

∂j (𝒮mod f ) exists at σ-a.e. point on ∂∗ Ω. (1.5.53)

∂Ω

In addition, as a consequence of (1.5.51), [70, (2.4.8)], and [70, Theorem 1.4.2],

for each aperture parameter κ ∈ (0, ∞) one has
  p
Nκ ∂j (𝒮mod f ) ∈ Lloc (∂Ω, σ), j ∈ {1, . . . , n},
   M (1.5.54)
for all f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n−1
∩ Lloc (∂Ω, σ) with p ∈ (1, ∞).
   M
Also, given any f = ( fβ )1≤β ≤M ∈ L 1 ∂Ω, 1+σ(x) |x | n−1 , with the piece of
notation introduced in [70, (2.9.53)] it follows that for each r ∈ {1, . . . , n} one
has
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 95

∂r (𝒮mod f )(x) = (∂r 𝒮αβ ) fβ (x) for each x ∈ Ω. (1.5.55)
1≤α ≤M
   M
In fact, if n ≥ 3 and f ∈ L 1 ∂Ω, 1+σ(x)|x | n−2 , or if n = 2 and the function
   M
M
f ∈ L 1 ∂Ω, ln(2 + |x|)σ(x) , (hence, in particular, if f ∈ L p (∂Ω, σ)
with p ∈ [1, n − 1)), then there exists a constant C f ∈ C M with the property that

𝒮mod f = 𝒮 f + C f at each point in Ω. (1.5.56)

M
If actually the function f belongs to the space L p (∂Ω, σ) for an arbitrary
integrability exponent p ∈ (1, ∞), then for each aperture parameter κ ∈ (0, ∞)
there exists some constant C = C(Ω, A, κ, p) ∈ (0, ∞) with the property that
 
n 
 
dist(·, ∂Ω)(n−1)/p · ∂j (𝒮mod f ) 
L ∞ (Ω, L n )
j=1

n
 
≤C Nκ (∂j (𝒮 f )) p ≤ C f [L p (∂Ω,σ)] M . (1.5.57)
mod L (∂Ω,σ)
j=1

   M
Furthermore, for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular,
M
if f ∈ L p (∂Ω, σ) with p ∈ [1, ∞)) the conormal derivative ∂νA𝒮mod f may be
meaningfully considered in the sense of (A.0.184), and one has the jump-formula
 
∂νA𝒮mod f = − 12 I + K A#  f at σ-a.e. point on ∂∗ Ω, (1.5.58)

where I is the identity operator, and K A#  is the operator associated as in (1.3.72)

   M
with the coefficient tensor A . Also, if n ≥ 3 and f ∈ L 1 ∂Ω, 1+σ(x) |x | n−2
M
(hence, in particular, if f ∈ L p (∂Ω, σ) with p ∈ [1, n − 1)), or if
   M
n = 2 and f ∈ L ∂Ω, ln(2 + |x|)σ(x)
1 , the conormal derivative ∂νA𝒮 f is
meaningfully defined in the sense of (A.0.184) and satisfies the jump-formula
 
∂νA𝒮 f = − 12 I + K A#  f at σ-a.e. point on ∂∗ Ω. (1.5.59)

In addition, for each integrability exponent p ∈ (1, ∞) there exists some constant
C = C(∂Ω, A, p) ∈ (0, ∞) such that
M
if n ≥ 3 and f ∈ L n−1 (∂Ω, σ) then |∇𝒮mod f | p dist(·, ∂Ω) p−1 dL n
p
is a Carleson measure in Ω with constant ≤ C f [L n−1 (∂Ω,σ)] M ;
(1.5.60)
96 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

in particular, corresponding to p = 2, it follows that15

M
if n ≥ 3 and f ∈ L n−1 (∂Ω, σ) then |∇𝒮mod f | 2 dist(·, ∂Ω) dL n is a
Carleson measure in Ω with constant ≤ C f [L
2
n−1 (∂Ω,σ)] M .

(1.5.61)
Moreover,
under the additional assumption that Ω is a uniform domain, for each
M
function f ∈ L p (∂Ω, σ) with p ∈ (n − 1, ∞) it follows that 𝒮mod f
η
. M
belongs
  to 𝒞 (Ω) where η := 1 − (n − 1)/p ∈ (0, 1) and one has
𝒮 f  . η M ≤ C f [L p (∂Ω,σ)] M for some finite positive constant
mod [𝒞 (Ω)]
C = C(Ω, A, p).
(1.5.62)

If ∂Ω is also bounded, then for each integrability exponent p ∈ (1, ∞) there exists
M
C = C(∂Ω, A, p) ∈ (0, ∞) with the property that for every f ∈ L p (∂Ω, σ)
one has

n
 
Nκ (∂j 𝒮 f ) ≤ C f [L p (∂Ω,σ)] M (1.5.63)
L p (∂Ω,σ)
j=1

plus a similar estimate with the nontangential maximal operator replaced by

the tangential maximal operator (associated as in [68, Definition 8.5.1] with a
sufficiently large power; (A.0.146)), and

|∇𝒮 f | p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω with

p M
constant ≤ C f [L n−1 (∂Ω,σ)] M if n ≥ 3 and f ∈ L n−1 (∂Ω, σ) .
(1.5.64)
In particular, corresponding to p = 2,
M
if n ≥ 3 and f ∈ L n−1 (∂Ω, σ) then |∇𝒮 f | 2 dist(·, ∂Ω) dL n is a
vanishing Carleson measure in Ω with constant ≤ C f [L 2
n−1 (∂Ω,σ)] M .

(1.5.65)

Furthermore, if ∂Ω is also bounded, then with K A#  denoting the operator (1.3.72)

M
associated with the coefficient tensor A , for each function f ∈ L p (∂Ω, σ)
with p ∈ [1, ∞) one has
 
∂νA𝒮 f = − 12 I + K A#  f at σ-a.e. point on ∂∗ Ω. (1.5.66)

When Ω is also a uniform domain with a compact boundary, then (1.5.62) may
be rephrased in terms of 𝒮 simply as the statement that

2
15 it is natural to refer to ∇𝒮mod f dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated
with f via the operator 𝒮mod
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 97
M . M
𝒮 : L p (∂Ω, σ) −→ 𝒞η (Ω) boundedly
(1.5.67)
if p ∈ (n − 1, ∞) and η := 1 − (n − 1)/p ∈ (0, 1).

If n ≥ 3 then for each truncation parameter ε > 0 one has

n−2 ,∞
n−1  
Nκε (𝒮mod f ) ∈ Lloc (∂Ω, σ) for each f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
(1.5.68)

and, as a consequence of this and [68, (6.2.36)],

  
Nκε (𝒮mod f ) ∈ Lloc (∂Ω, σ) for each f ∈ L 1 ∂Ω, 1+σ(x)
p M
|x | n−1
,
0<p< n−1
n−2
(1.5.69)

while
,
if n = 2 then Nκε (𝒮mod f ) ∈
p
Lloc (∂Ω, σ)
0<p<∞
 σ(x)  M
(1.5.70)
for each function f ∈ L 1 ∂Ω, 1+ |x | .

Finally, given any truncation parameter ε > 0 and any integrability exponent
p ∈ (1, ∞), it follows that
 
for each f ∈ L 1 ∂Ω, 1+σ(x)
p M
|x | n−1
∩ Lloc (∂Ω, σ)
(1.5.71)
on has Nκε (𝒮mod f ) ∈ Lloc (∂Ω, σ),
q

where q is any number in (1, ∞) if n = 2, while if n ≥ 3 then q is given by

⎧ 1  −1
⎪
⎪ − 1 if 1 < p < n − 1,
⎨ p n−1
⎪
q := any number in (1, ∞) if p = n − 1, (1.5.72)
⎪
⎪
⎪∞ if n − 1 < p < ∞.
⎩
(xii) In analogy with (1.5.50), define the following modified version of the boundary-
to-boundary single layer operator
∫
 
Smod f (x) := E(x − y) − E∗ (−y) f (y) dσ(y)
∂Ω
at σ-a.e. x ∈ ∂Ω, for each function (1.5.73)
 
f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
, where E∗ := E · 1Rn \B(0,1) .

Then this operator is meaningfully defined, via an absolutely convergent integral,

at σ-a.e. point in ∂Ω, and
   M
Smod : L 1 ∂Ω, 1+σ(x)
M
|x | n−1
−→ Lloc
1
(∂Ω, σ) (1.5.74)
98 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

as well as
   M
Smod : L 1 ∂Ω, 1+σ(x)
p p M
|x | n−1 ∩ Lloc
(∂Ω, wσ) −→ Lloc (∂Ω, wσ)
(1.5.75)
for each weight w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞),

and
   M    M
Smod : L 1 ∂Ω, 1+ |xσ(x)
| n−1−ε −→ L 1 ∂Ω, σ(x)
1+ |x | n
(1.5.76)
for each ε > 0,

are well-defined, linear, and continuous mappings. In addition, the following

mapping is well defined, linear, and bounded
  M
−→ L 1 ∂Ω, 1+σ(x)
M
Smod : L p (∂Ω, wσ) |x | n
(1.5.77)
for each weight w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞),

and
given any weight w ∈ Ap (∂Ω, σ) with exponent p ∈ (1, ∞), it
M
follows that for each sequence { f j } j ∈N ⊆ L p (∂Ω, wσ) which
M (1.5.78)
is weak-∗ convergent to some function f ∈ L p (∂Ω, wσ) one
M
has lim Smod f j = Smod f in (Lipc (∂Ω)) .
j→∞

Also, with the modified boundary-to-domain single layer operator 𝒮mod as in

  M
(1.5.50), for each aperture parameter κ > 0 and each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has
  κ−n.t.
the nontangential boundary limit 𝒮mod f ∂Ω (1.5.79)
exists (in C M ) at σ-a.e. point on Aκ (∂Ω),
and, in fact,
  κ−n.t.

𝒮mod f (x) = (Smod f )(x) at σ-a.e. point x ∈ Aκ (∂Ω),

∂Ω (1.5.80)
in particular, (cf. [68, Proposition 8.8.4]) at σ-a.e. point x ∈ ∂∗ Ω.
   M
Furthermore, if n ≥ 3 and the function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−2
(thus, in
M
particular, if f belongs to the space L p (∂Ω, σ) with p ∈ [1, n − 1)), or if
   M
n = 2 and f ∈ L ∂Ω, ln(2 + |x|)σ(x)
1 , then there exists some constant
C f ∈ C M with the property that

Smod f = S f + C f at σ-a.e. point x ∈ ∂∗ Ω. (1.5.81)

 1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 99

Finally, if for each pair of indices j, k ∈ {1, . . . , n} one considers the vector-
valued version of the integral operator (1.2.3) corresponding to choosing b to be
  M
the fundamental solution E, i.e., if for each function f ∈ L 1 ∂Ω, 1+σ(y) |y | n−1

one defines
∫
 
Tjk f (x) := lim+
#
ν j (x)(∂k E)(x − y) − νk (x)(∂j E)(x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
(1.5.82)

at σ-a.e. point x ∈ ∂∗ Ω, then

   M
for each f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, 1+σ(x)
p
|x | n−1
with p ∈ (1, ∞)
  (1.5.83)
one has ∂τ j k Smod f = Tjk
# f at σ-a.e. point on ∂ Ω.
∗

In particular, [68, Lemma 7.7.13] guarantees that the formula in the second
M
line of (1.5.83) holds whenever f ∈ L p (∂Ω, wσ) with p ∈ (1, ∞) and
w ∈ Ap (∂Ω, σ). Also, (1.5.83), (1.5.56), (1.5.80), and [68, (8.8.45)] ensure that
M
for each f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1)
  (1.5.84)
one has ∂τ j k S f = Tjk
# f at σ-a.e. point on ∂ Ω.
∗

(xiii) Make the assumption that Ω is a UR domain. Then the following operator
identities hold:
1   1   A 
2 I + K ◦ − 2 I + K = S ◦ ∂ν D
p,q M
on L1 (∂Ω, σ) with p ∈ (1, ∞) and q ∈ (1, n − 1), (1.5.85)
M  
as well as on L p (∂Ω, σ) with p ∈ n−1
n−2 , ∞ ,

1     
2I + K A#  ◦ − 12 I + K A#  = ∂νA D ◦ S
M
on L p (∂Ω, σ) with p ∈ (1, n − 1), as well as (1.5.86)
p, p ∗ M
on L−1 (∂Ω, σ) with p ∈ (1, n − 1) and 1
p∗ = 1
p − n−1 ,
1

S ◦ K A#  = K ◦ S
M
on L p (∂Ω, σ) with p ∈ (1, n − 1), as well as (1.5.87)
p, p ∗ M
on L−1 (∂Ω, σ) with p ∈ (1, n − 1) and 1
p∗ = 1
p − n−1 ,
1
 100 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
   
K A#  ◦ ∂νA D = ∂νA D ◦ K
p,q M
on L1 (∂Ω, σ) with p ∈ (1, ∞) and q ∈ (1, n − 1), (1.5.88)
M
as well as on L p (∂Ω, σ) with p ∈ (1, ∞).
M
Also, for each function f ∈ L p (∂Ω, wσ) with p ∈ (1, ∞) and each Mucken-
houpt weight w ∈ Ap (∂Ω, σ), at σ-a.e. point x ∈ ∂Ω one has (with the operators
# , 1 ≤ j, k ≤ n, defined as in (1.5.82))
Tjk

1   
2I + K A#  − 12 I + K A#  f (x) (1.5.89)
# ∫ $
μγ βα  # 
= lim νi (x)ai j ar s (∂r Eγβ )(x − y) Tjs f α (y) dσ(y)
ε→0+
y ∈∂Ω 1≤μ ≤M
|x−y |>ε
# ∫ $
μγ βα  
= lim νi (x)ai j ar s (∂r Eγβ )(x − y)∂τ j s Smod f α (y) dσ(y) .
ε→0+
y ∈∂Ω 1≤μ ≤M
|x−y |>ε

Moreover, if ∂Ω is bounded then one may allow any p, q ∈ (1, ∞) in (1.5.85)-

(1.5.88), this time taking p∗ ∈ (1, ∞) arbitrary (and unrelated to p).
(xiv) Revert back to assuming that Ω ⊆ Rn is an open set with the property that
∂Ω is a UR set. Denote by AWE (n, M) the collection of all coefficient tensors
 αβ 
A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L A associated with A in Rn as in (1.3.2)
is weakly elliptic. Fix j ∈ {1, . . . , n} and denote by ∂j S A the principal-value
singular integral operator on ∂Ω whose kernel is (∂j E A)(x − y), where E A is the
fundamental solution canonically associated with L A as in [70, Theorem 1.4.2].
  M
More specifically, for every f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1
set
∫
(∂j S A) f (x) := lim+ (∂j E A)(x − y) f (y) dσ(y) (1.5.90)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. Finally, recall the pieces of notation introduced in

(A.0.23), (A.0.24), and (A.0.171).
Then the following operator-valued assignments are continuous:
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 101

M
AWE (n, M)  A −→ ∂j S A ∈ Bd L p (∂Ω, σ) if 1 < p < ∞, (1.5.91)

M M
AWE (n, M)  A −→ ∂j S A ∈ Bd L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ) ,
(1.5.92)

M
AWE (n, M)  A → ∂j S A ∈ Bd L p (∂Ω, wσ) ,
(1.5.93)
if 1 < p < ∞ and w ∈ Ap (∂Ω, σ),

M
AWE (n, M)  A → ∂j S A ∈ Bd L p,q (∂Ω, σ)
(1.5.94)
if 1 < p < ∞ and 0 < q ≤ ∞,

and

M M
AWE (n, M)  A → ∂j S A ∈ Bd H p,q (∂Ω, σ) → L p,q (∂Ω, σ)
 
if p ∈ n−1 n , ∞ and q ∈ (0, ∞].
(1.5.95)

(xv) Make the assumption that Ω is a UR domain. Then for each p ∈ (1, ∞) the
following operator-valued assignments are continuous:

M
AWE (n, M)  A → K A ∈ Bd L p (∂Ω, σ) , (1.5.96)

p M
AWE (n, M)  A → K A ∈ Bd L1 (∂Ω, σ) , (1.5.97)

M
AWE (n, M)  A → K A# ∈ Bd L p (∂Ω, σ) , (1.5.98)

p M
AWE (n, M)  A → K A# ∈ Bd L−1 (∂Ω, σ) , (1.5.99)

p M M
AWE (n, M)  A → ∂νA D A ∈ Bd L1 (∂Ω, σ) → L p (∂Ω, σ) ,
(1.5.100)

M p M
AWE (n, M)  A → ∂νA D A ∈ Bd L p (∂Ω, σ) → L−1 (∂Ω, σ) .
(1.5.101)

If ∂Ω is compact, then for each p ∈ (1, ∞) the following operator-valued assign-

ments are also continuous:

M p M
AWE (n, M)  A → S A ∈ Bd L p (∂Ω, σ) → L1 (∂Ω, σ) , (1.5.102)

p M M
AWE (n, M)  A → S A ∈ Bd L−1 (∂Ω, σ) → L p (∂Ω, σ) . (1.5.103)
102 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Finally, similar results are valid on Lorentz spaces, off-diagonal Sobolev spaces,
as well as Muckenhoupt weighted Lebesgue and Sobolev spaces.

Proof The reader is reminded that, throughout the proof, the summation convention
over repeated indices is in effect.
Proof of claims in item (i): The nontangential maximal function estimates in (1.5.1)-
(1.5.2) are consequences of (1.3.68), [70, Theorem 1.4.2], and [70, Theorem 2.4.1],
while the nontangential maximal function estimate from (1.5.3) is implied by
(1.3.68), [70, Theorem 1.4.2], and [70, (2.4.18)]. Next, (1.5.4) is seen from (1.3.68),
[70, Theorem 1.4.2], and [70, (2.4.23)]. Finally, the estimate in (1.5.5) is a conse-
quence of [68, (6.5.31)] used with u := ∇D f , p := q, q := p, s := 1− p1 (also bearing
in mind (1.3.24) and [68, (6.5.40) in Theorem 6.5.7]), as well as [70, (2.4.34)] used
with Σ := ∂Ω.
  M
Proof of claims in item (ii): Having fixed f = ( fα )1≤α ≤M ∈ L11 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
,
Lemma 1.3.2 gives that for each index ∈ {1, . . . , n} we have, at every x ∈ Ω,
#∫ $
  βα
∂ D f (x) = ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ∗ (y) . (1.5.104)
∂∗ Ω
1≤γ ≤M

 
Since each ∂τ s fα belongs to L ∂∗ Ω, 1+σ|x∗ (x)
1
| n−1
, the claim in (1.5.7) follows from
(1.5.104), [70, Theorem 1.4.2], and [70, Theorem 2.5.1]. Also, the claims in (1.5.8)
and the subsequent comment are seen from (1.5.104), [70, Theorem 1.4.2], and [70,
Theorem 2.4.1]. The claim made in (1.5.9) may be justified, bearing [69, (11.7.21)] in
mind, based on [70, Theorem 1.4.2] and [70, (2.4.18)]. Finally, assume that Ω is also
p,q M
a uniform domain, and that f = ( fα )1≤α ≤M ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ (1, ∞)
and q ∈ (n − 1, ∞). In such a scenario, the Hölder regularity result in (1.5.10) and
the estimate in (1.5.11) are implied by (1.5.8) and [68, Corollary 8.6.8]. This takes
care of item (ii).
Proof of claims in item (iii): These are all consequences of (1.3.72), [70, Theo-
rem 1.4.2], [70, Theorem 2.3.2], (1.3.68), and [70, (2.3.25), (2.3.44), (2.3.60),
(2.3.61)].
Proof of claims in item (iv): Fix an index γ ∈ {1, . . . , M } along with some aperture
  M
parameter κ > 0, and observe that if f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
then
according to (1.3.18), [70, Theorem 1.4.2], [70, Theorem 2.5.1], and (1.3.68), at
σ∗ -a.e. x ∈ ∂∗ Ω we may write
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 103
  κ−n.t. 1 -  βα
Df γ ∂Ω (x) = − ∂r Eγβ ν(x) ar s νs (x) fα (x)
2i
∫
βα
− lim+ νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

1   −1 βα
=− L ν(x) ar s νr (x)νs (x) fα (x) + (K f )γ (x)
2 γβ

1   −1   
= L ν(x) L ν(x) fα (x) + (K f )γ (x)
2 γβ βα

= 12 δγα fα (x) + (K f )γ (x) = 2 fγ (x)

1
+ (K f )γ (x). (1.5.105)

On account of the arbitrariness of γ, formula (1.5.105) proves (1.5.20).

Proof of claims in item (v): Make the additional assumption stipulated in (1.5.23).
Then the claim that for every p ∈ (1, ∞) the operator K is well defined and bounded
in the context of (1.5.24) becomes a consequence of (1.5.20), (1.5.8), and [69,
Proposition 11.3.2] (whose applicability in the present case requires (1.5.23)). The
linearity of K is clear from (1.3.68). The more general cases pertaining to the action
of K on off-diagonal Sobolev spaces as in (1.5.25), and the action of K on weighted
Sobolev spaces as in (1.5.26) are dealt with in a similar fashion.
Proof of claims in item (vi): Having fixed p, p ∈ (1, ∞) such that 1/p + 1/p = 1,
we start by considering the operator

# : L p (∂∗ Ω, σ∗ )
K
Mp
−→ L−1 (∂∗ Ω, σ∗ )
M
defined by K# f , g  := f , Kg 
−1

p M p M ∗ p M
for all f ∈ L−1 (∂∗ Ω, σ∗ ) = L1 (∂∗ Ω, σ∗ ) and g ∈ L1 (∂∗ Ω, σ∗ )
(1.5.106)

where, in the present  context, the angled brackets ·, · denote the duality pairing
p M ∗ p M
between the spaces L1 (∂∗ Ω, σ∗ ) and L1 (∂∗ Ω, σ∗ ) . Thanks to (A.0.136)
# is a well-defined, linear and bounded operator. Bring
and (1.5.24) it follows that K
in the mapping E, extending functions originally defined on ∂∗ Ω by zero to the
entire topological boundary ∂Ω. Then, since the transpose of K # in (1.5.12) is K in
# is the unique
(1.5.14), we conclude from (1.5.106), and [69, (11.8.4), (11.8.5)] that K
extension of the composition K # ◦ E (where K # is as in (1.5.27)) to a continuous
p M # simply as K # then yields
operator from L−1 (∂∗ Ω, σ∗ ) into itself. Re-denoting K
the claims in item (vi) pertaining to the diagonal scale of boundary Sobolev spaces.
Finally, analogous claims for the off-diagonal scale of boundary Sobolev spaces may
be justified in a very similar fashion.
  M
Proof of claims in item (vii): Fix some f = ( fα )1≤α ≤M ∈ L11 ∂∗ Ω, 1+σ|x∗ (x) | n−1
.
Then the current item (ii) ensures that the conormal derivative ∂νA(D f ) is meaning-
104 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

fully defined as indicated in (A.0.184). More specifically, for each μ ∈ {1, . . . , M }

and ∈ {1, . . . , n}, at each point σ-a.e. x ∈ ∂Ω we may write

  μγ   κ−n.t.

(∂νA D) f μ (x) = νi (x)ai j ∂j D f γ (x)

∂Ω

1 -  βα μγ
= ∂r Eγβ ν(x) ar s ai j νi (x)(∂τ j s fα )(x)
2i
∫
μγ βα
+ lim+ νi (x)ai j ar s (∂r Eγβ )(x − y)(∂τ j s fα )(y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

1   −1 βα μγ
= L ν(x) ar s ai j νi (x)νr (x)(∂τ j s fα )(x)
2 γβ
∫
μγ βα
+ lim+ νi (x)ai j ar s (∂r Eγβ )(x − y)(∂τ j s fα )(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(1.5.107)

thanks to (A.0.184), (1.5.104), [70, Theorem 2.5.1], and [70, (1.4.30)] (with γ := er
and ξ := ν(x)). As regards the jump-term in (1.5.107), based on [69, (11.4.8)] we
may compute
  βα μγ
L(ν)−1 γβ ar s ai j νi νr ∂τ j s fα
  βα μγ  
= L(ν)−1 γβ ar s ai j νi νr ν j ∇tan fα s
  βα μγ  
− L(ν)−1 γβ ar s ai j νi νr νs ∇tan fα j
    βα  
= − L(ν)−1 γβ L(ν) μγ ar s νr ∇tan fα s
    μγ  
+ L(ν)−1 γβ L(ν) βα ai j νi ∇tan fα j
μα   μα  
= −ar s νr ∇tan fα s + ai j νi ∇tan fα j = 0. (1.5.108)

Collectively, (1.5.107) and (1.5.108) prove (1.5.29). The boundedness of the conor-
mal derivative operator in (1.5.31) when p, q ∈ (1, ∞) is then clear from (1.5.29) and
[70, Theorem 2.3.2].
Proof of claims in item (viii): Continue to assume that Ω is a UR domain in Rn .
Also, assume Ω  Rn since otherwise there is nothing to prove. Define Ω+ := Ω
and Ω− := Rn \ Ω. Then item (7) in [68, Lemma 5.10.9] ensures that Ω− is also a
UR domain, whose topological and geometric measure theoretic boundaries agree
with those of Ω, and whose geometric measure theoretic outward unit normal is −ν
at σ-a.e. point on ∂Ω.
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 105

p,q M q, p M
To proceed, select f ∈ L1 (∂Ω, σ) and g ∈ L1 (∂Ω, σ) where
p, p, q, q  ∈ (1, ∞) satisfy 1/p + 1/p = 1 and 1/q + 1/q  = 1. In order to stress the
dependence on the coefficient tensor, write D A for the boundary-to-domain double
layer associated with A as in (1.3.18). Also, let D A be the boundary-to-domain
double layer associated with the transpose coefficient tensor A (much as D is
associated with A in (1.3.18)). Finally, define u± := D A f and w ± := D A g in Ω± .
Then our earlier results in items (i), (ii), (iv), show that the pairs (u+, w + ) and
(u , w − ) satisfy [70, (1.7.76)] and the version of [70, (1.7.77)] with the nontangential
−

maximal function memberships replaced as in [70, (1.7.83)], as well as [70, (1.7.80)],

relative to Ω+ and to Ω− . In addition, in the case when Ω± is an exterior domain, it
is clear from (1.3.18), (1.5.104), and [70, (1.4.24)] that the functions u±, w ± satisfy

u± (x) = O(|x| 1−n ) and w ± (x) = O(|x| 1−n ) as |x| → ∞,

(1.5.109)
(∇u± )(x) = O(|x| −n ) and (∇w ± )(x) = O(|x| −n ) as |x| → ∞,

which further implies that condition [70, (1.7.79)] formulated for the exterior domain
Ω± holds in such a scenario. As such, we may invoke Green’s formula [70, (1.7.81)]
which, in light of (1.5.20) and (1.5.29), permits us to write
∫ 
 A    
∂ν D A f , ± 12 I + K A g dσ
∂Ω
∫      
= ± 12 I + K A f , ∂νA D A g dσ, (1.5.110)
∂Ω

where K A is the boundary-to-boundary double layer associated with the coefficient

tensor A as in (1.3.68), and K A is defined similarly with A replaced by A . Sub-
tracting the two versions of this equality then produces (1.5.32).
Proof of claims in items (ix)-(x): These are direct consequences of (1.3.6), (1.3.62),
and Proposition 1.2.4, whose applicability in the present setting is ensured by item
(3) in [70, Theorem 1.4.2]. We also wish to note that in the proof of the claim made
in relation to (1.5.37) the first formula in [70, (1.4.32)] is also used. Next, the fact
that if n ≥ 3 then S in (1.5.39) is a linear and bounded operator is a consequence of
[68, (7.8.11)] and [70, Theorem 1.4.2]. Finally, the claim in (1.5.49) may be justified
much as in [68, (7.8.10)-(7.8.11)].
Proof of claims in item (xi): The claims in the first part of the statement are easily
justified using the Mean Value Theorem, the estimates for the fundamental solution
from [70, Theorem 1.4.2], and [68, Lemma 7.2.1]. In the case when ∂Ω is actually
a UR set and f ∈ [L p (∂Ω, σ)] M with p ∈ (1, ∞), the second estimate in (1.5.57) is
implied by (1.5.51) and [70, Theorem 2.4.1], while the first estimate in (1.5.57) is a
consequence of [68, Lemma 8.6.6].
Turning now to the task of justifying the jump-formula recorded in (1.5.58), fix an
   M
arbitrary function f = ( fγ )1≤γ ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
. Then based on (A.0.184),
(1.5.55), and the jump-formula [70, (2.9.54)], for each α ∈ {1, . . . , M } we may write,
106 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

at σ-a.e. point x ∈ ∂∗ Ω,

αβ   κ−n.t.

(∂νA𝒮mod f )α (x) = νr (x)ar s ∂s (𝒮mod f )β (x)

∂Ω
 κ−n.t.
αβ
= νr (x)ar s (∂s 𝒮βγ ) fγ (x)
∂Ω

1 αβ
= a νr (x)νs (x)bβγ (x) fγ (x)
2 rs
∫
αβ
+ lim+ νr (x)ar s (∂s Eβγ )(x − y) fγ (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
1 αβ
= − fα (x) + lim+ νr (x)ar s (∂s Eβγ )(x − y) fγ (y) dσ(y),
2 ε→0
y ∈∂Ω
|x−y |>ε
(1.5.111)

where
  the last equality makes use of [70, (2.9.52)]. From (1.3.72) and the fact that
E L A = E L A (cf. [70, (1.4.32)]) we see that the above principal-value integral is
 
precisely K A#  f α (x). Altogether, we conclude that for each index α ∈ {1, . . . , M }
 
we have (∂νA𝒮mod f )α (x) = − 12 fα (x) + K A#  f α (x) at σ-a.e. x ∈ ∂∗ Ω. In view of the
arbitrariness of α, this establishes (1.5.58). Then the jump-formula (1.5.59) follows
on account of this and (1.5.56).
Next, the claim in (1.5.60) is readily seen from (1.5.51) and [70, (2.4.48)] used
with k := ∇E, p := n−1, and θ := p−1. That the Hölder estimate claimed in (1.5.62)
holds when Ω is a uniform domain whose boundary is a UR set is a consequence
of (1.5.57) and [68, (5.11.78)]. In the case when ∂Ω is bounded, 𝒮mod f defined
in (1.5.50) differs from 𝒮 f defined in (1.3.6) by a constant (which depends on
M
the function f ∈ L 1 (∂Ω, σ) ). Hence, in such a case, ∇𝒮mod f = ∇𝒮 f in Ω.
In particular, if ∂Ω is bounded then the nontangential maximal operator estimate
(1.5.57) and the jump-formula (1.5.58) become, respectively, (1.5.63) and (1.5.66).
As regards the vanishing Carleson measure property in (1.5.64), we first observe
that (1.5.60) presently gives that

|∇𝒮 f | p dist(·, ∂Ω) p−1 dL n is a Carleson measure in Ω with constant

p M (1.5.112)
≤ C f [L n−1 (∂Ω,σ)] M whenever f ∈ L n−1 (∂Ω, σ) .

There remains to show that the aforementioned Carleson measure is actually vanish-
ing. To this end, pick an exponent q ∈ (n−1, ∞) and define η := 1−(n−1)/q ∈ (0, 1).
M M
Also, select two arbitrary functions, f ∈ L n−1 (∂Ω, σ) and g ∈ L q (∂Ω, σ) .
M
Then the function g belongs to L n−1 (∂Ω, σ) (since ∂Ω is bounded), and for each
r ∈ 0, 2 diam(∂Ω) and x ∈ ∂Ω we may estimate
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 107
 ∫
1  p 1
p
  ∇ 𝒮( f − g) dist(·, ∂Ω) p−1 dL n
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

≤ C f − g[L n−1 (∂Ω,σ)] M , (1.5.113)

thanks to (1.5.112) written for f − g in place of f . In addition, we have

 ∫
1  p 1
p
  ∇ 𝒮g dist(·, ∂Ω) p−1 dL n
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
 ∫ 1
1 p
≤C   dist(·, ∂Ω) pη−1 dL n g[L q (∂Ω,σ)] M
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤ Cr η g[L q (∂Ω,σ)] M . (1.5.114)

The first inequality above uses the version of (1.5.57) written for 𝒮 in place of 𝒮mod
(and for g in place of f ), while the second inequality is based on [68, (8.6.101)] used
with λ := 1 − pη, α := 1, β := n − 1, and E := B(x, r) ∩ Ω. Together, (1.5.113) and
(1.5.114) imply
 ∫
1  p 1
p
lim+ sup   ∇ 𝒮 f dist(·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤ C f − g[L n−1 (∂Ω,σ)] M , (1.5.115)

for some constant C ∈ (0, ∞) independent of f and g. Having established this,

M
(1.5.64) now follows, on account of the arbitrariness of g ∈ L q (∂Ω, σ) and the
density of L (∂Ω, σ) in L (∂Ω, σ).
q n−1
  M
To justify the claims made in (1.5.68) and (1.5.70), pick f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
along with a truncation parameter ε > 0. Also, fix an arbitrary point x0 ∈ ∂Ω, and
choose some number r > 1 + 4|x0 | + 2ε(1 + κ). Split

f = f1 + f2 on ∂Ω, where
(1.5.116)
f1 := 1∂Ω∩B(x0,2r) · f and f2 := 1∂Ω\B(x0,2r) · f .

Hence,
𝒮mod f = 𝒮mod f1 + 𝒮mod f2 in Ω. (1.5.117)
Observe that
 
|x − y| ≤ |x − z| + |z − y| < (1 + κ) dist z, ∂Ω + |z − y| ≤ (2 + κ)|z − y|,
for each points x, y ∈ ∂Ω and each point z ∈ Γκ (x).
(1.5.118)

Based on (1.5.50), [70, (1.4.24)], and (1.5.118) we then conclude that there exists
some constant C = C(L, κ, r, x0 ) ∈ (0, ∞) with the property that if x ∈ ∂Ω ∩ B(x0, r)
108 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

and z ∈ Γκ (x) then in the case in which n ≥ 3 we have

∫
 
𝒮mod f1 (z) ≤ |E(z − y)|| f (y)| dσ(y)
∂Ω∩B(x0,2r)
∫
+ |E∗ (−y)|| f (y)| dσ(y)
∂Ω∩B(x0,2r)
∫ ∫
| f (y)|
≤C dσ(y) + C | f (y)| dσ(y)
∂Ω∩B(x0,2r) |z − y| n−2 ∂Ω∩B(x0,2r)
∫
| f (y)|
≤C dσ(y) + C f [L 1 (∂Ω, σn−1 )] M . (1.5.119)
∂Ω∩B(x0,2r) |x − y| n−2 1+|·|

Taking the supremum over z ∈ Γκ (x) this further implies

    
Nκ 𝒮mod f1 (x) ≤ C · I∂Ω∩B(x0,2r),1 | f | (x) + C f [L 1 (∂Ω, σ
)] M (1.5.120)
1+|·| n−1

for each x ∈ ∂Ω ∩ B(x0, r), where I∂Ω∩B(x0,2r),1 is the fractional integral operator
defined as in (A.0.100) with E := ∂Ω ∩ B(x0, 2r), d := n − 1, α := 1, μ := σ, and
ρ the ordinary Euclidean distance. From [68, (7.8.9)] we see (keeping in mind that
f is locally integrable) that, as a function of x ∈ ∂Ω ∩ B(x0, r), the right side of
 
(1.5.120) belongs to L n−2 ,∞ ∂Ω ∩ B(x0, r), σ . In concert with [68, (6.2.16)], [68,
n−1

(8.2.28)], and (1.5.120) this implies

   
Nκ 𝒮mod f1 ∈ L n−2 ,∞ ∂Ω ∩ B(x0, r), σ if n ≥ 3.
n−1
(1.5.121)

When n = 2, the nature of the singularity in the fundamental solution E changes

(see [70, (1.4.24)]).
 To handle this case, assume that x ∈ ∂Ω ∩ B(x0, r) and z ∈ Γκ (x)
with dist z, ∂Ω < ε. Since for each y ∈ ∂Ω ∩ B(x0, 2r) we have

|z − y| ≤ |z − x| + |x − x0 | + |x0 − y|
 
< (1 + κ) dist z, ∂Ω + r + 2r

< (1 + κ)ε + 3r, (1.5.122)

it follows from [70, (1.4.24)] and (1.5.118) that for each α ∈ (0, 1) there exists a
constant Cα ∈ (0, ∞), which depends on L, κ, ε, α, r, and x0 , with the property that
if n = 2 then
  Cα Cα
E(z − y) ≤ C 1 + ln |z − y| ≤ ≤ . (1.5.123)
|z − y| 1−α |x − y| 1−α

Granted this, when n = 2 in place of (1.5.119) we now see that, for each α ∈ (0, 1),
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 109
∫
  | f (y)|
𝒮mod f1 (z) ≤ C dσ(y) + C f [L 1 (∂Ω, 1+|·|
σ
)] M (1.5.124)
∂Ω∩B(x0,2r) |x − y| 1−α
 
for each z ∈ Γκ (x) with dist z, ∂Ω < ε, where C = C(L,  κ, α, ε, r, x0 ) ∈ (0, ∞).
After taking the supremum over all z ∈ Γκ (x) satisfying dist z, ∂Ω < ε, when n = 2
we arrive at
    
Nκε 𝒮mod f1 (x) ≤ C · I∂Ω∩B(x0,2r),α | f | (x) + C f [L 1 (∂Ω, 1+|·|
σ
)] M (1.5.125)

for each x ∈ ∂Ω ∩ B(x0, r), where I∂Ω∩B(x0,2r),α is the fractional integral operator of
order α ∈ (0, 1), defined as in (A.0.100) with E := ∂Ω ∩ B(x0, 2r), d := 1, μ := σ,
and ρ the ordinary Euclidean distance. Since f is a locally integrable function, from
[68, (7.8.9)] we then conclude that, as a function of the variable x ∈ ∂Ω ∩ B(x0, r),
 
the right side of (1.5.125) belongs to L α ,∞ ∂Ω ∩ B(x0, r), σ . Together with [68,
1

(6.2.16)], [68, (8.2.28)], and (1.5.125) this implies that

   
Nκε 𝒮mod f1 ∈ L α ,∞ ∂Ω ∩ B(x0, r), σ for each α ∈ (0, 1), if n = 2.
1
(1.5.126)

In concert with the embedding in [68, (6.2.38)], this ultimately proves the following
companion to (1.5.121):
   
Nκε 𝒮mod f1 ∈ L p ∂Ω ∩ B(x0, r), σ for each p ∈ (0, ∞), if n = 2. (1.5.127)

There remains to estimate the contribution from f2 , which we shall do working

under the assumption that n ≥ 2. To set the stage, continue to assume that the point
x ∈ ∂Ω ∩ B(x0, r) and pick some z ∈ Γκ (x) with dist(z, ∂Ω) < ε. Then

|z| ≤ |z − x| + |x − x0 | + |x0 |
 
< (1 + κ) dist z, ∂Ω + r + |x0 | ≤ (1 + κ)ε + r + |x0 |. (1.5.128)

In turn, (1.5.128) implies that for each ξ ∈ [0, z] and each y ∈ ∂Ω \ B(x0, 2r) we
have
3
2r + 14 |x0 − y| ≤ |x0 − y| ≤ |y − ξ | + |ξ | + |x0 |

≤ |y − ξ | + |z| + |x0 | ≤ |y − ξ | + ε(1 + κ) + 2|x0 | + r. (1.5.129)

This entails

1
2r − ε(1 + κ) − 2|x0 | + 14 |x0 − y| ≤ |y − ξ | (1.5.130)

which, bearing in mind the original choice of r, ultimately leads to the conclusion
that

4 |x0 − y| ≤ |y − ξ | for each ξ ∈ [0, z].

1
(1.5.131)
110 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Let us also observe that, since 2r > 1 + |x0 |, for each point y ∈ ∂Ω \ B(x0, 2r) we
have |y| > 1. Keeping this in mind, we now use (1.5.50), the Mean Value Theorem,
(1.5.128), (1.5.131), and [70, (1.4.24)] in [70, Theorem 1.4.2] to estimate
∫
 
𝒮mod f2 (z) ≤ |E(z − y) − E(−y)|| f (y)| dσ(y)
∂Ω\B(x0,2r)
∫ ∫
| f (y)| | f (y)|
≤C dσ(y) ≤ C dσ(y)
∂Ω\B(x0,2r) |x0 − y| n−1 ∂Ω\B(x0,2r) 1 + |y| n−1

≤ C f [L 1 (∂Ω, σ
)] M (1.5.132)
1+|·| n−1

for some C = C(L, κ, r, x0, ε) ∈ (0, ∞). After taking the supremum in (1.5.132) over
all z ∈ Γκ (x) with dist(z, ∂Ω) < ε we therefore arrive at
  
Nκε 𝒮mod f2 (x) ≤ C f [L 1 (∂Ω, σn−1 )] M for all x ∈ ∂Ω ∩ B(x0, r). (1.5.133)
1+|·|

In view of [68, (8.2.28)], this goes to show that

   
Nκε 𝒮mod f2 ∈ L ∞ ∂Ω ∩ B(x0, r), σ . (1.5.134)

At this stage, from (1.5.117), (1.5.121), (1.5.134), and [68, (6.2.16), (8.2.9), (8.2.28)]
we conclude that (1.5.68) holds, while (1.5.70) is implied by (1.5.117), (1.5.127),
and (1.5.134).
Finally, consider the claim made in (1.5.71)-(1.5.72). When n = 2 this is seen
directly from (1.5.70), so we shall focus on the case when n ≥ 3. In this scenario,
we re-run the argument which, starting with the decomposition in (1.5.116), has led
to (1.5.121). Since, thanks to (1.5.71), we now have f1 ∈ L p (∂Ω, σ), the mapping
properties of the fractional integration operator of order one (of the sort recorded in
[68, (7.8.7) and (7.8.14)-(7.8.15)]) now give that
   
Nκ 𝒮mod f1 ∈ L q ∂Ω ∩ B(x0, r), σ with q as in (1.5.72). (1.5.135)

With this in hand, the same argument as before then completes the proof of (1.5.71).
Proof of claims in item (xii): Fix x0 ∈ ∂Ω along with r ∈ (0, ∞). Given a function
  M
f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, for each x ∈ ∂Ω ∩ B(x0, r) split
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 111
∫
E(x − y) − E∗ (−y) | f (y)| dσ(y)
∂Ω
∫
= E(x − y) − E∗ (−y) | f (y)| dσ(y)
∂Ω\B(x,1)
∫
+ E(x − y) − E∗ (−y) | f (y)| dσ(y)
∂Ω∩B(x,1)

=: I(x) + I I(x). (1.5.136)

The same argument used to prove [70, (2.3.117)] presently gives that, for some
constant Cx0,r ∈ (0, ∞), we have

Cx0,r
E(x − y) − E∗ (−y) ≤ ,
1 + |y| n−1 (1.5.137)
for all x ∈ B(x0, r) and all y ∈ Rn \ B(x, 1).

Consequently, the integral defining I(x) is absolutely convergent and, moreover, the
assignment x → I(x) is measurable and bounded for x ∈ B(x0, r). In addition,
the version of the Fractional Integration Theorem recorded in [68, (7.8.9)] ensures
that the integral defining I I(x) in (1.5.136) is absolutely convergent for σ-a.e. point
x ∈ ∂Ω ∩ B(x0, r) and that, as a function of the variable x ∈ B(x0, r), the term I I(x)
is also absolutely integrable. This analysis ultimately shows that the definition of
(Smod f )(x) in (1.5.73) involves an absolutely convergent integral for σ-a.e. x ∈ ∂Ω,
and also that Smod in (1.5.74) is a well-defined, linear, and continuous operator.
To show that Smod is also a well-defined, linear, and continuous operator
in the context of (1.5.75), we reason much as above, this time starting with
 
f ∈ L 1 ∂Ω, 1+σ(x)
p M
|x | n−1
∩ Lloc (∂Ω, wσ) for some w ∈ Ap (∂Ω, σ), 1 < p < ∞.
M
Then g := f · 1∂Ω∩B(x0,r+1) ∈ L p (∂Ω, wσ) and for each x ∈ ∂Ω ∩ B(x0, r) we
may estimate
∫
I I(x) = E(x − y) − E∗ (−y) | f (y)| dσ(y)
∂Ω∩B(x,1)
∫
|g(y)|
≤C dσ(y)
∂Ω∩B(x,1) |x − y| n−2
∫
+ E∗ (−y) | f (y)| dσ(y). (1.5.138)
∂Ω∩B(x0,r+1)

Together with [68, Lemma 7.7.16] this implies that the assignment x → I I(x)
p
belongs to Lloc (∂Ω, wσ) from which we ultimately conclude that Smod in (1.5.75)
is indeed a well-defined, linear, and continuous operator. Finally, the claims about
(1.5.76) follow from (1.2.83) and [70, Theorem 1.4.2].
112 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Let us now deal with the claim in (1.5.78). To this end, fix p ∈ (1, ∞) and consider
M
a Muckenhoupt weight w ∈ Ap (∂Ω, σ). Also, assume { f j } j ∈N ⊆ L p (∂Ω, wσ)
M
is weak-∗ convergent to some function f ∈ L p (∂Ω, wσ) , i.e.,
∫ ∫

f , F dσ for each F ∈ L p (∂Ω, w  σ)
M
lim f j , F dσ = , (1.5.139)
j→∞ ∂Ω ∂Ω

where p ∈ (1, ∞) denotes the Hölder conjugate exponent of p and w  denotes the

dual weight of w, thus w  := w 1−p ∈ Ap (∂Ω, σ). To proceed, pick an arbitrary
M
C M -valued test function φ ∈ Lipc (∂Ω) . Choose some reference point x0 ∈ ∂Ω
and select R ∈ (0, ∞) large enough so that supp φ ⊆ ∂Ω ∩ B(x0, R). Then for each
j ∈ N we may decompose

Smod f j , φ = I j + II j + III j (1.5.140)

where the pairing in the left-hand side is taken in the sense of distributions on ∂Ω,
and where
∫  ∫ 
 
I j := f j (y), 1∂Ω\B(x0,2R) (y) · E(x − y) − E∗ (−y) φ(x) dσ(x) dσ(y),
∂Ω ∂Ω
(1.5.141)

∫ ∫ 
II j := − E∗ (−y)1∂Ω∩B(x0,2R) (y) f j (y) dσ(y), φ(x) dσ(x) , (1.5.142)
∂Ω ∂Ω

and
∫  ∫ 
III j := f j (y), 1∂Ω∩B(x0,2R) (y) · E(x − y) φ(x) dσ(x) dσ(y).
∂Ω ∂Ω∩B(x0,R)
(1.5.143)

If for each y ∈ ∂Ω we define

∫
 
F1 (y) := 1∂Ω\B(x0,2R) (y) · E(x − y) − E∗ (−y) φ(x) dσ(x), (1.5.144)
∂Ω

it follows that there exists a constant Cx0,R,φ ∈ (0, ∞) with the property that

Cx0,R,φ
|F1 (y)| ≤ for each y ∈ ∂Ω. (1.5.145)
1 + |y| n−1

Since F1 is a C M -valued σ-measurable function, we may then estimate

∫ ∫
 w (y)
|F1 (y)| p w (y) dσ(y) ≤ Cx0,R,φ n−1 ) p
dσ(y) < +∞, (1.5.146)
∂Ω ∂Ω (1 + |y|

 M
by (1.5.145) and [68, (7.7.101)]. Hence, F1 ∈ L p (∂Ω, w  σ) , so (1.5.139) gives
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 113

lim I j (1.5.147)
j→∞
∫ ∫
= lim f j (y), F1 (y) dσ(y) = f (y), F1 (y) dσ(y)
j→∞ ∂Ω ∂Ω
∫  ∫ 
 
= f (y), 1∂Ω\B(x0,2R) (y) · E(x − y) − E∗ (−y) φ(x) dσ(x) dσ(y).
∂Ω ∂Ω

To compute the limit of II j as j → ∞, define

F2 (y) := −E∗ (−y)1∂Ω∩B(x0,2R) (y) for each point y ∈ ∂Ω. (1.5.148)

∞ (∂Ω, σ) M×M  M×M

We then have F2 ∈ Lcomp ⊆ L p (∂Ω, w  σ) , so based on
(1.5.139) we may compute
∫ ∫ 
lim II j = lim F2 (y) f j (y) dσ(y), φ(x) dσ(x)
j→∞ j→∞ ∂Ω ∂Ω
∫ ∫ 
= F2 (y) f (y) dσ(y), φ(x) dσ(x)
∂Ω ∂Ω
∫ ∫ 
=− E∗ (−y)1∂Ω∩B(x0,2R) (y) f (y) dσ(y), φ(x) dσ(x) . (1.5.149)
∂Ω ∂Ω

Finally, to handle the limit of III j as j → ∞, for each y ∈ ∂Ω we define

∫
F3 (y) := 1∂Ω∩B(x0,2R) (y) · E(x − y) φ(x) dσ(x), (1.5.150)
∂Ω∩B(x0,R)

then since for each x ∈ supp φ ⊆ ∂Ω ∩ B(x0, R) and each y ∈ ∂Ω ∩ B(x0, 2R) we
may estimate |y − x| ≤ |y − x0 | + |x0 − x| < 2R + R = 3R, it follows that
∫
|φ(x)|
|F3 (y)| ≤ C dσ(x) for each y ∈ ∂Ω. (1.5.151)
|y − x| n−2
x ∈∂Ω
|y−x |<3R

Since F3 is a C M -valued σ-measurable function, from [68, Lemma 7.7.16] and

 M
(1.5.151) we then conclude that F3 ∈ L p (∂Ω, w  σ) which, in concert with
(1.5.139), permits us to write
∫ ∫
lim III j = lim f j (y), F3 (y) dσ(y) = f (y), F3 (y) dσ(y) (1.5.152)
j→∞ j→∞ ∂Ω ∂Ω
∫  ∫ 
= f (y), 1∂Ω∩B(x0,2R) (y) · E(x − y) φ(x) dσ(x) dσ(y).
∂Ω ∂Ω∩B(x0,R)

At this stage, from (1.5.140), (1.5.147), (1.5.149), and (1.5.152) we see that
114 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 
lim Smod f j , φ = Smod f , φ . (1.5.153)
j→∞

M
In view of the arbitrariness of φ ∈ Lipc (∂Ω) we therefore conclude that
M
lim Smod f j = Smod f in (Lipc (∂Ω)) , finishing the justification of (1.5.78).
j→∞
The fact that the operator (1.5.77) is well defined, linear, and bounded is a
consequence of (1.5.73), Proposition 1.2.6, [70, Theorem 1.4.2], and the embedding
in [68, (7.7.102)].
  M
To prove (1.5.79)-(1.5.80), fix an arbitrary function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
.
Having fixed a point x0 ∈ ∂Ω along with an arbitrary number r ∈ (0, ∞), decompose
𝒮mod f = 𝒮mod f1 + 𝒮mod f2 with f1, f2 as in (1.5.116). Note that since 𝒮mod f2 has
a continuous extension to B(x0, r) we trivially have that the nontangential trace
κ−n.t.

𝒮mod f2 exists at every point in Aκ (∂Ω) ∩ B(x0, r). More precisely,

∂Ω
∫
  κ−n.t.
 
𝒮mod f2 (x) = E(x − y) − E∗ (−y) f2 (y) dσ(y)
∂Ω ∂Ω
 
= Smod f2 (x) for each x ∈ Aκ (∂Ω) ∩ B(x0, r). (1.5.154)
M
Next, since f1 ∈ L 1 (∂Ω, σ) , at σ-a.e. point x ∈ Aκ (∂Ω) we may write
∫
  κ−n.t.
  κ−n.t.

𝒮mod f1 (x) = 𝒮 f1 (x) − E∗ (−y) f1 (y) dσ(y)

∂Ω ∂Ω ∂Ω
∫ ∫
= E(x − y) f1 (y) dσ(y) − E∗ (−y) f1 (y) dσ(y)
∂Ω ∂Ω
 
= Smod f1 (x), (1.5.155)

making use of [70, Proposition 2.5.39]. In turn, from (1.5.117) and (1.5.154)-
(1.5.155) we conclude that
  κ−n.t.

𝒮mod f (x) = (Smod f )(x) at σ-a.e. x ∈ B(x0, r) ∩ Aκ (∂Ω). (1.5.156)

∂Ω

Upon recalling that r > 0 has been arbitrarily chosen, this ultimately proves (1.5.80).
The claim made in relation to (1.5.81) is a consequence of (1.5.80) (and [68,
Proposition 8.8.4]), (1.5.56), and (1.3.67).
Let us justify the claim made in (1.5.83). To this end, fix j, k ∈ {1, . . . , n} and
   M
some function f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, 1+σ(x)
p
|x | n−1 with p ∈ (1, ∞). Also, choose
some aperture parameter κ ∈ (0, ∞). Then we may write
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 115
 
∂τ j k Smod f (x)
# $
  κ−n.t.

= ∂τ j k 𝒮mod f (x)
∂Ω

# $ # $
   κ−n.t.    κ−n.t.

= ν j (x) ∂k 𝒮mod f (x) − νk (x) ∂j 𝒮mod f (x)

∂Ω ∂Ω

 ∫ .
1  
= ν j (x) ∂k E ν(x) f (x) + lim+ (∂k E)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε
 ∫ .
1  
− νk (x) ∂j E ν(x) f (x) + lim+ (∂j E)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε
 # 
= Tjk f (x) at σ-a.e. x ∈ ∂∗ Ω. (1.5.157)

The first equality in (1.5.157) uses indentity (1.5.80) and [68, Proposition 8.8.4]. The
second equality in (1.5.157) is a consequence of [69, Proposition 11.3.2] applied
M
here with u := 𝒮mod f ∈ 𝒞∞ (Ω) , whose applicability in the current setting is
ensured by (1.5.69), (1.5.54), (1.5.79), [68, Proposition 8.8.4], (1.5.51), and [70,
Theorem 2.5.1]. The third equality in (1.5.157) is implied by [70, (2.5.4)] (keeping
in mind (1.5.51)). Finally, the last equality in (1.5.157) is seen from (1.5.82) upon
observing that at σ-a.e. point x ∈ ∂∗ Ω we have
   
ν j (x)∂ 
k E ν(x) − νk (x)∂ j E ν(x)

  −1   −1
= iν j (x)νk (x) L ν(x) − iνk (x)ν j (x) L ν(x) = 0, (1.5.158)

thanks to [70, Theorem 1.4.2]. This establishes (1.5.83).

Proof of claims in item (xiii): Throughout, work under the assumption that Ω is a
UR domain. To deal with the claims in (1.5.85) assume first that either Ω is bounded,
p,q M
or ∂Ω is unbounded. Pick a function f ∈ L1 (∂Ω, σ) with p ∈ (1, ∞) and
q ∈ (1, n − 1), then introduce u := D f in Ω. From (1.5.8), (1.5.7), (1.5.20), (1.3.24),
and [68, Lemma 7.2.1], it follows that u satisfies the hypotheses of Theorem 1.3.3. In
view of (1.5.20) and (1.3.24), the integral representation formula (1.3.35) presently
becomes    
D f = D 12 I + K f − 𝒮 ∂νA D f in Ω. (1.5.159)

On account of (1.5.20) and (1.5.44), taking nontangential traces produces

1  1 1   A 
2 I + K f = 2 I + K 2 I + K f − S ∂ν D f at σ-a.e. point on ∂Ω. (1.5.160)
116 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
    
After some simple algebra (1.5.160) implies 12 I + K − 12 I + K f = S ∂νA D f at
σ-a.e. point on ∂Ω. When Ω is an exterior domain, the same argument as above with
Ω replaced by Ω− := Rn \ Ω leads to the same conclusion. Here, item (7) in [68,
Lemma 5.10.9] is relevant. Hence, in all cases,
1      p,q M
2I + K ◦ − 12 I + K = S ◦ ∂νA D on L1 (∂Ω, σ)
(1.5.161)
with p ∈ (1, ∞) and q ∈ (1, n − 1).

This takes care of the first claim in (1.5.85). The second claim in (1.5.85) corresponds
M  
to functions belonging to L p (∂Ω, σ) with p ∈ n−1n−2 , ∞ . Having fixed such a p,
select q ∈ (1, n − 1) with the property that q∗ = p (i.e., 1/p = 1/q − 1/(n
 − 1)).
Then (1.5.161) ensures that we have 12 I + K ◦ − 12 I + K = S ◦ ∂νA D on
q ∗,q M
L1 (∂Ω, σ) , while from [69, (11.1.66)] we know that the latter space embeds
M
densely into L p (∂Ω, σ) . Based on this and the continuity properties of the
operators involved (established earlier in (1.5.33) with p := q∗ , and (1.5.28) with
p := q, as well as in the current item (iii)), the second claim in (1.5.85) follows. This
finishes the proof of (1.5.85).
Turning attention to (1.5.86), we first make the assumption that either Ω is bound-
M
ed, or ∂Ω is unbounded. Select f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1) and define
u := 𝒮 f in Ω. From (1.5.48), (1.3.67), (1.3.57), (1.3.8), and [68, Lemma 7.2.1], it
follows that u satisfies the hypotheses of Theorem 1.3.3. Thanks to (1.5.44), (1.5.59),
and (1.3.8), the integral representation formula (1.3.35) currently reads
 
𝒮 f = D(S f ) − 𝒮 − 12 I + K A#  f in Ω. (1.5.162)

In view of (1.5.59), taking the conormal derivative ∂νA of both sides then shows that,
at σ-a.e. point on ∂Ω, we have
 1      
− 2 I + K A#  f = ∂νA D (S f ) − − 12 I + K A#  − 12 I + K A#  f . (1.5.163)
    
This yields, after some simple algebra, 12 I + K A#  − 12 I + K A#  f = ∂νA D (S f ) at
σ-a.e. point on ∂Ω. If Ω is an exterior domain, we run the same argument as above
with Ω replaced by Ω− := Rn \ Ω and, in view of item (7) in [68, Lemma 5.10.9],
reach the same conclusion. This establishes that, in all cases, we have
1   1   A 
2 I + K A ◦ − 2 I + K A = ∂ν D ◦ S
# #

M
(1.5.164)
on L p (∂Ω, σ) with p ∈ (1, n − 1).

With this in hand, the second version in (1.5.86) may then be justified by a density
argument based on [69, (11.8.30)] and continuity properties of the operators involved
established earlier (cf. (1.5.36), (1.5.33), and (1.5.28)). This completes the proof of
(1.5.86).
To justify the claims in (1.5.87), assume for now that either Ω is bounded, or ∂Ω
M
is unbounded. Fix f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1) and recall the integral
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 117

identity established in (1.5.162). Taking nontangential traces then yields, on account

of (1.5.20) and (1.5.44),
   
S f = 12 I + K (S f ) − S − 12 I + K A#  f at σ-a.e. point on ∂Ω, (1.5.165)

from which we conclude, after some simple algebra, that K(S f ) = S(K A#  f ) at σ-a.e.
point on ∂Ω. In the case when Ω is an exterior domain, we run the same argument as
above with Ω replaced by Ω− := Rn \Ω and, thanks to item (7) in [68, Lemma 5.10.9],
reach the same conclusion. This proves that, in all cases, we have
M
S ◦ K A#  = K ◦ S on L p (∂Ω, σ) with p ∈ (1, n − 1). (1.5.166)

Having established this, the second version in (1.5.87) then follows by a density
argument based on [69, (11.8.30)] and continuity properties of the operators involved
established earlier (cf. (1.5.36), (1.5.28), and the current item (iii)). This finishes the
proof of (1.5.87).
Going further, consider the claims in (1.5.88). In a first stage, assume that either
p,q M
Ω is bounded, or ∂Ω is unbounded. Also, pick f ∈ L1 (∂Ω, σ) with p ∈ (1, ∞)
and q ∈ (1, n − 1). Then, as in the past, (1.5.159) holds. By applying the conormal
derivative operator ∂νA to both sides of (1.5.159) we arrive at the conclusion that
     
∂νA D f = ∂νA D 12 I + K f − − 12 I + K A#  ∂νA D f (1.5.167)

at σ-a.e. point on ∂Ω. Hence,

    p,q M
K A#  ◦ ∂νA D = ∂νA D ◦ K on L1 (∂Ω, σ)
(1.5.168)
with p ∈ (1, ∞) and q ∈ (1, n − 1).

When Ω is an exterior domain, the above argument with Ω replaced by Ω− := Rn \ Ω

produces the same conclusion (once again, item (7) in [68, Lemma 5.10.9] is relevant
here). Having proved (1.5.168), the version of the identity in (1.5.88) for functions
M
in L p (∂Ω, σ) with p ∈ (1, ∞) follows from this and a density argument (making
use of the continuity properties of the operators involved; cf. the current items (iii),
(vi), and (viii)). This finishes the proof of (1.5.88).
M
Moving on, the first equality in (1.5.89) corresponding to f ∈ L p (∂Ω, σ)
with p ∈ (1, n − 1) is a consequence of (1.5.86), (1.5.29), and (1.5.84). Extending
this formula to the case when p ∈ (1, ∞) is done via density, given that all operators
M
involved are bounded on L p (∂Ω, wσ) (cf. [70, (2.3.56)]). Having established
this, the second equality in (1.5.89) follows from the first, keeping in mind (1.5.83)
and [68, (7.7.104)].
An alternative proof of the first equality in (1.5.89) goes as follows. Pick an
M
arbitrary function f ∈ L p (∂Ω, σ) with p ∈ (1, ∞), then consider u := 𝒮mod f in
Ω. Also, fix an arbitrary aperture parameter κ ∈ (0, ∞). Results in the current item
(xi) then imply
118 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t. κ−n.t.
the boundary traces u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂Ω, (1.5.169)
κ−n.t.  
u ∂Ω = Smod f and ∂νAu = − 12 I + K A#  f .

In the case when Ω is an exterior domain we also have

⨏
|∇u| dL n = o(1) as R → ∞. (1.5.170)
B(0,2R)\B(0,R)

In concert with the last two equalities in (1.5.157) the first trace formula in the last
line of (1.5.169) shows that for each j, k ∈ {1, . . . , n} and α ∈ {1, . . . , M } we have
κ−n.t. κ−n.t.
ν j (∂k uα ) ∂Ω − νk (∂j uα ) ∂Ω = (Tjk
# f)
α
(1.5.171)
at σ-a.e. point on ∂Ω,

where the singular integral operator Tjk # is defined as in (1.5.82). As before, employ

the notation E = (Eγβ )1≤γ,β ≤M for the matrix-valued fundamental solution associat-
ed with L as in [70, Theorem 1.4.2]. For each ∈ {1, . . . , n} and each γ ∈ {1, . . . , M }
we may then rely on (1.5.169)-(1.5.171), [70, (1.5.230)], and (A.0.184) to write
∫
βα
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y)(Ts
#
f )α (y) dσ(y)
∂Ω
∫  
− (∂ Eγα )(x − y) − 12 I + K A#  f (y) dσ(y) (1.5.172)
∂Ω α

at each point x ∈ Ω. To proceed, fix an arbitrary index μ ∈ {1, . . . , M }. Multiply

μγ
both sides of (1.5.172) by a j , then go nontangentially to the boundary, subsequently
multiply by ν j , and finally sum up over j ∈ {1, . . . , n}. On account of the second
trace formula in the last line of (1.5.169) and [70, (1.5.230)], we therefore arrive at
 
− 12 I + K A#  f (x) (1.5.173)
μ
  μγ κ−n.t.
= ∂νAu μ (x) = ν j (x)a j (∂ uγ ) ∂Ω (x)
/∫ 0 κ−n.t.

μγ βα
= ν j (x) a j ar s (∂r Eγβ )(· − y)(Ts
#
f )α (y) dσ(y) (x)
∂Ω
∂Ω
/∫ 0 κ−n.t.
 
μγ
− ν j (x) a j (∂ Eγα )(· − y) − 1
2I + K A#  f (y) dσ(y) (x)
∂Ω α
∂Ω

at σ-a.e. point x ∈ ∂Ω. In view of [70, (2.5.4)], the term involving the first nontan-
gential trace above may be explicitly identified as
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 119
/∫ 0 κ−n.t.

μγ βα
ν j (x) a j ar s (∂r Eγβ )(· − y)(Ts
#
f )α (y) dσ(y) (x) (1.5.174)
∂Ω
∂Ω

μγ βα 1 -  #
= ν j (x)a j ar s ∂r Eγβ ν(x) (Ts f )α (x)
2i
∫
μγ βα
+ lim+ ν j (x)a j ar s (∂r Eγβ )(x − y)(Ts
#
f )α (y) dσ(y).
ε→0
y ∈∂Ω
|x−y |>ε

As regards the first term in the right-hand side of the equality in (1.5.174), at σ-a.e.
point x ∈ ∂Ω we have

μγ βα 1 -  #
ν j (x)a j ar s ∂r Eγβ ν(x) (Ts f )α (x) (1.5.175)
2i
1 μγ βα   −1 #
= ν j (x)a j ar s νr (x) L ν(x) γβ (Ts f )α (x)
2
∫
1 μγ βα   −1
= lim+ ν j (x)a j ar s νr (x) L ν(x) γβ ×
ε→0 2
y ∈∂Ω
|x−y |>ε
 
× ν (x)(∂s Eαδ )(x − y) − νs (x)(∂ Eαδ )(x − y) fδ (y) dσ(y),

thanks to [70, (1.4.30)] and (1.5.82). In view of (A.0.141), for each s ∈ {1, . . . , n}
and α ∈ {1, . . . , M } we have
μγ βα   −1 βα     −1
ν j (x)a j ar s νr (x) L ν(x) ν (x)
γβ 
= −ar s νr (x) L ν(x) μγ
L ν(x) γβ

βα μα
= −ar s νr (x)δμβ = −ar s νr (x), (1.5.176)

while for each ∈ {1, . . . , n} and α ∈ {1, . . . , M } we have

μγ βα   −1 μγ   −1  
ν j (x)a j ar s νr (x) L ν(x) ν (x)
γβ s
= −ν j (x)a j L ν(x) γβ
L ν(x) βα

μγ μα
= −ν j (x)a j δγα = −a j ν j (x). (1.5.177)

Combining (1.5.175)-(1.5.177) leads to the conclusion that

120 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

μγ βα 1 -  #
ν j (x)a j ar s ∂r Eγβ ν(x) (Ts f )α (x)
2i
∫ 
−1 μα
= lim+ ar s νr (x)(∂s Eαδ )(x − y)
ε→0 2
y ∈∂Ω
|x−y |>ε

μα
− a j ν j (x)(∂ Eαδ )(x − y) fδ (y) dσ(y)

= 0 at σ-a.e. point x ∈ ∂Ω. (1.5.178)

Going further, the term involving the second nontangential trace in (1.5.173) is equal
to (cf. [70, (1.5.230)], (1.5.51), and (1.5.58)),
/∫ 0 κ−n.t.
 
μγ
ν j (x) a j (∂ Eγα )(· − y) − 12 I + K A#  f (y) dσ(y) (x) (1.5.179)
∂Ω α
∂Ω
     
= ∂νA𝒮mod − 12 I + K A#  f (x) = − 12 I + K A#  − 12 I + K A#  f (x)
μ μ

at σ-a.e. point x ∈ ∂Ω. At this stage, from (1.5.173), (1.5.174), (1.5.178), and
(1.5.179) we readily conclude that the first equality in (1.5.89) holds for each function
M
f ∈ L p (∂Ω, σ) with p ∈ (1, ∞). The more general case when the function
M
f ∈ L p (∂Ω, wσ) for some w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞) then follows from
this and density, keeping in mind the continuity of all singular integral operators
involved on Muckenhoupt weighted Lebesgue spaces.
Finally, that in (1.5.85)-(1.5.88) we may allow p, q, p∗ ∈ (1, ∞) unrelated and
unrestricted when ∂Ω is bounded, follows from the fact that the Lebesgue scale is
nested, the addenda to the Fractional Integration Theorem in [68, (7.8.14)-(7.8.15)],
and keeping in mind that the weight functions in [70, (1.5.3)] now behave like
constants.
Proof of claims in item (xiv): Let us establish the continuity of the operator-valued
assignment in (1.5.91) for some fixed integrability exponent p ∈ (1, ∞). Fix a weakly
elliptic coefficient tensor A0 and pick another coefficient tensor A1 with | A0 − A1 |
small enough. Define

At := A0 + t(A1 − A0 ) for each t ∈ [0, 1]. (1.5.180)

Having fixed j ∈ {1, . . . , n}, let ∂j S At be the principal-value singular integral oper-
ator on ∂Ω whose kernel is (∂j E At )(x − y). Then ∂j S A0 − ∂j S A1 is a principal-value
singular integral operator on ∂Ω whose kernel is k(x−y) where, for each z ∈ Rn \{0},
∫ 1
k(z) := − k t (z) dt (1.5.181)
0

with
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 121

d
(∂j E At )(z) for every t ∈ [0, 1].
k t (z) := (1.5.182)
dt
In relation to this, we claim that

k t ∈ 𝒞∞ (Rn \ {0})
M×M
for every t ∈ [0, 1], (1.5.183)
k t (−z) = −k t (z) for every z ∈ Rn \ {0} and t ∈ [0, 1], (1.5.184)

k t (λz) = λ1−n k t (z) for every z ∈ Rn \ {0}, λ > 0, and t ∈ [0, 1], (1.5.185)
for every N ∈ N there exists C = C(N, A0 ) ∈ (0, ∞) such that
 
sup  k t S n−1 𝒞 N (S n−1 ) ≤ C| A0 − A1 |. (1.5.186)
t ∈[0,1]

Let us assume for the time being that (1.5.183)-(1.5.186) hold and indicate how
these may be used to finish the proof of the claim in item (15). Concretely, (1.5.183)-
(1.5.186) ensure that k from (1.5.181) satisfies the properties listed in [70, (2.3.3)].
As such, [70, Theorem 2.3.2] is applicable to the operator ∂j S A0 −∂j S A1 . In particular,
[70, (2.3.20)] permits us to estimate
   
∂j S A − ∂j S A  p 
M ≤ C k n−1
 N n−1
0 1 [L (∂Ω,σ)]
M p →[L (∂Ω,σ)] S 𝒞 (S )

≤ C| A0 − A1 |, (1.5.187)

where C ∈ (0, ∞) depends only on n, p, A0 , as well as the Ahlfors regularity and

UR constants of ∂Ω. Of course, once this has been established, the continuity of the
operator-valued assignment in (1.5.91) follows.
There remains to prove (1.5.183)-(1.5.186), a task to which we now turn. The key
ingredient in this regard is having a more transparent formula for k t (z). To set the
stage, write  αβ 
br s 1≤α,β ≤M := A1 − A0 (1.5.188)
1≤r,s ≤n

so that
 αβ   αβ 
ar s (t) 1≤α,β ≤M := At = A0 + t · br s 1≤α,β ≤M for each t ∈ [0, 1], (1.5.189)
1≤r,s ≤n 1≤r,s ≤n

then for each t ∈ [0, 1] introduce

d   αβ  −1
Θt (ξ) := ar s (t)ξr ξs )1≤α,β ≤M , ∀ξ ∈ Rn \ {0}. (1.5.190)
dt
From (1.5.182) and (A.0.66) we then conclude that, for each t ∈ [0, 1] and each point
z ∈ Rn \ {0},
 ∫ .
∂ 1 (n−1)/2
k t (z) = Δ | z, ξ |Θt (ξ) dH n−1 (ξ) (1.5.191)
∂z j 4(2πi)n−1 z S n−1
122 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

if n is odd, and
 ∫ .
∂ −1 (n−2)/2  
k t (z) = Δ ln | z, ξ | Θt (ξ) dH (ξ)
n−1
(1.5.192)
∂z j (2πi)n z S n−1

if n is even. To analyze (1.5.191)-(1.5.192), let us take a closer look at Θt (ξ) in

(1.5.190). Pick an arbitrary number to ∈ (0, 1) and, for ξ ∈ Rn \ {0} fixed, compute

Θto (ξ) (1.5.193)

 αβ  −1  αβ  −1
ar s (to + t)ξr ξs 1≤α,β ≤M − ar s (to )ξr ξs 1≤α,β ≤M
= lim
t→0 t
 αβ  −1
= − ar s (to )ξr ξs 1≤α,β ≤M ×
#  αβ   αβ  $
ar s (to + t)ξr ξs 1≤α,β ≤M − ar s (to )ξr ξs 1≤α,β ≤M
× lim ×
t→0 t
 αβ  −1
× ar s (to )ξr ξs 1≤α,β ≤M
   −1      −1
αβ αβ αβ
= − ar s (to )ξr ξs 1≤α,β ≤M br s ξr ξs 1≤α,β ≤M ar s (to )ξr ξs 1≤α,β ≤M .

In turn, from (1.5.193) we deduce that

Θt ∈ 𝒞∞ (Rn \ {0})
M×M
for every t ∈ [0, 1], (1.5.194)
Θt (ξ) is even in ξ ∈ Rn \ {0} for every t ∈ [0, 1], (1.5.195)

Θt (λξ) = λ−2 Θt (ξ) for every ξ ∈ Rn \ {0}, λ ∈ C \ {0}, and t ∈ [0, 1],
(1.5.196)

sup |Θt (ξ)| ≤ C|ξ | −2 | A0 − A1 | for every ξ ∈ Rn \ {0}, (1.5.197)

0≤t ≤1

where C ∈ (0, ∞) depends only on n, M, and A0 (recall that A1 is taken sufficiently

close to A0 to begin with). Hence Θt (ξ) behaves, in a quantitative fashion, uniformly
in t, like the inverse [L(ξ)]−1 appearing in (A.0.66). Bearing this in mind, we run the
same type of argument as in the proof of [66, Theorem 11.1, pp. 393-395] which,
starting with (A.0.66), establishes items (1), (3), (4) in [70, Theorem 1.4.2] (dealing,
respectively, with smoothness, parity, homogeneity, and estimates for derivatives
of the fundamental solution E), but for the function k t from (1.5.191)-(1.5.192) in
place of E. In the current case, this produces (1.5.183)-(1.5.186) and this finishes
the proof of the claim made in relation to (1.5.91). The remaining claims in item
(xiv) of the current theorem are dealt with analogously, making use of [70, (2.3.21),
(2.3.53), (2.3.57)].
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 123

Proof of claims in item (xv): All desired conclusions follow from the continuity re-
sults established in item (xiv), (1.3.68), (1.3.72), (1.5.29), (1.3.62), and the current
item (viii) (cf. also Proposition 1.5.6 a little further below).
The proof of Theorem 1.5.1 is therefore complete. 
Theorem 1.5.1 is a powerful tool, which may be used with great effect for a
variety of purposes. Below, we discuss a number of such applications. First, in
Proposition 1.5.2 we shall employ Theorem 1.5.1 to shed further light on the structure
of our brand of boundary Sobolev spaces on uniformly rectifiable sets.
Proposition 1.5.2 Pick an integer n ∈ N with n ≥ 2 and assume Ω  Rn is a UR
domain. Abbreviate σ := H n−1 ∂Ω, then introduce Ω+ := Ω and Ω− := Rn \ Ω.
Also, fix an integrability exponent p ∈ (1, ∞) along with an aperture parameter
κ ∈ (0, ∞). In this setting, consider the class of functions u± : Ω± → C satisfying

u± ∈ 𝒞∞ (Ω± ),
Nκ u± ∈ L p (∂Ω, σ), Nκ (∇u± ) ∈ L p (∂Ω, σ), (1.5.198)
κ−n.t.   κ−n.t.
u± ∂Ω and ∇u± ∂Ω exist σ-a.e. on ∂Ω.

Then the L p -based Sobolev space of order one on ∂Ω may be described as

 κ−n.t. κ−n.t. 
L1 (∂Ω, σ) = u+ ∂Ω − u− ∂Ω : u± are as in (1.5.198) .
p
(1.5.199)

Proof From item (7) in [68, Lemma 5.10.9] we know that Ω− is also a UR domain,
whose topological boundary agrees with that of Ω+ = Ω. Granted this, the right-to-
left inclusion in (1.5.199) becomes a consequence of [69, Proposition 11.3.2].
p
To prove the left-to-right inclusion in (1.5.199), pick f ∈ L1 (∂Ω, σ) arbitrary
and define
∫
1 ν(y), y − x
u± (x) := f (y) dσ(y) for x ∈ Ω±, (1.5.200)
ωn−1 ∂Ω |x − y| n
where ν is the geometric measure theoretic outward unit normal to Ω. Also, recall
the principal-value harmonic double layer K associated with Ω as in [70, (2.5.203)].
From [70, Theorem 2.4.1] we deduce that
 
Nκ u±  p ≤ C f  L p (∂Ω,σ), (1.5.201)
L (∂Ω,σ)

while from [70, (2.5.84)] we conclude that

κ−n.t.
 
u± = ± 12 I + K f at σ-a.e. point on ∂Ω. (1.5.202)
∂Ω

In particular,
κ−n.t. κ−n.t.

u+ − u− = f at σ-a.e. point on ∂Ω. (1.5.203)

∂Ω ∂Ω
Since from Theorem 1.5.1 we also know that
124 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
κ−n.t.

there exist ∇u± at σ-a.e. point on ∂Ω,

∂Ω (1.5.204)
 
and Nκ (∇u± ) L p (∂Ω,σ) ≤ C f  L p (∂Ω,σ),
1

we see that the functions u± are as in (1.5.198), so the desired conclusion follows.
Next, we rely on Theorem 1.5.1 to show that the point-spectra of K and K A# 
(acting on Lebesgue and Sobolev spaces on the boundary of a given UR domain)
are actually closely related.
Proposition 1.5.3 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain with
compact boundary and abbreviate σ := H n−1 ∂Ω. For some M ∈ N, consider
a coefficient tensor A with complex entries, with the property that the M × M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Recall the principal-
value double layer K from (1.3.68), and let K A#  be associated with A and Ω as in
(1.3.72). Finally, fix p ∈ (1, ∞) along with λ ∈ C \ {± 12 }.
p M
Then λ is an eigenvalue of K acting on L1 (∂Ω, σ) (respectively, on
M M
L p (∂Ω, σ) ) if and only if λ is an eigenvalue of K A#  acting on L p (∂Ω, σ)
p M
(respectively, on L−1 (∂Ω, σ) ).
p M
Proof Assume first that λ ∈ C\{± 12 } is an eigenvalue of K acting on L1 (∂Ω, σ) .
p M
Thus, there exists f ∈ L1 (∂Ω, σ) , f  0, such that K f = λ f . Then, if ∂νA D is as
M
in item (vii) of Theorem 1.5.1, the function g := ∂νA D f belongs to L p (∂Ω, σ)
(cf. (1.5.31)) and item (xiii) in Theorem 1.5.1 (cf. (1.5.88)) permits us to write
   
K A#  g = K A#  ∂νA D f = ∂νA D (K f )
   
= ∂νA D (λ f ) = λ ∂νA D f = λg. (1.5.205)

Also, if S is the boundary-to-boundary single layer potential operator associated with

A and Ω as in (1.3.62), based on item (xiii) in Theorem 1.5.1 (cf. (1.5.85)) we may
write
      
Sg = S ∂νA D f = 12 I + K − 12 I + K f = λ2 − 14 f  0, (1.5.206)

since λ2  14 and f  0. In turn, this forces g  0 which ultimately proves that

M
λ is an eigenvalue for K A#  on L p (∂Ω, σ) . The case when λ ∈ C \ {± 12 } is an
M
eigenvalue of K acting on L p (∂Ω, σ) is treated in a completely similar fashion,
bearing in mind (1.5.33) and the fact that the operator identities in (1.5.88) and
M
(1.5.85) continue to be valid on L p (∂Ω, σ) .
Going further, assume now that λ ∈ C \ {± 12 } is an eigenvalue of K A#  acting on
M M
L p (∂Ω, σ) . Thus, there exists some f ∈ L p (∂Ω, σ) , f  0, with the property
p M
that K A#  f = λ f . Then g := S f belongs to the Sobolev space L1 (∂Ω, σ) (cf.
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 125

item (ix) in Theorem 1.5.1) and item (xiii) in Theorem 1.5.1 (cf. (1.5.87)) permits
us to write
 
Kg = K(S f ) = S K A#  f = S(λ f ) = λS f = λg. (1.5.207)

In addition, thanks to item (xiii) in Theorem 1.5.1 (cf. (1.5.86)) we have

 A        
∂ν D g = ∂νA D (S f ) = 12 I + K A#  − 12 I + K A#  f = λ2 − 14 f  0, (1.5.208)

since λ2  14 and f  0. Consequently, g  0 which ultimately proves that λ is

p M
an eigenvalue for K on L1 (∂Ω, σ) . Finally, the case when λ ∈ C \ {± 12 } is an
p M
eigenvalue of K A#  acting on L−1 (∂Ω, σ) is dealt with in a completely analogous
fashion, once again based on item (ix) in Theorem 1.5.1 and the operator identities
in (1.5.87), (1.5.86). 

When specialized to the case L = Δ, the Laplacian in Rn , formula (1.5.89) from

Theorem 1.5.1 acquires a particularly pleasant format, as described in the proposition
below.

Proposition 1.5.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain. Abbreviate

σ := H n−1 ∂Ω, and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω. In this setting, bring in the “standard”
 principal-value
σ(x)
harmonic double layer defined for each function f ∈ L ∂Ω, 1+ |x | n−1 and σ-a.e.
1

point x ∈ ∂Ω according to
∫
1 ν(y), y − x
KΔ f (x) := lim+ f (y) dσ(y), (1.5.209)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

together with its formal transpose, i.e., the singular integral operator acting on each
given function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
as
∫
1 ν(x), x − y
KΔ# f (x) := lim+ f (y) dσ(y), (1.5.210)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

at σ-a.e. x ∈ ∂Ω. Also, consider the modified version of the boundary-to-boundary

harmonic single layer operator (compare with (1.5.73))
∫
 
SΔ,mod f (x) := EΔ (x − y) − EΔ (−y) · 1Rn \B(0,1) (−y) f (y) dσ(y)
∂Ω
 (1.5.211)
for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
and σ-a.e. point x ∈ ∂Ω.
126 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Additionally, for each pair of indices

 j, k ∈ {1, . . . , n} consider the operators16
σ(x)
defined for each function f ∈ L ∂Ω, 1+ |x | n−1 and σ-a.e. point x ∈ ∂Ω as
1

∫
 
Tjk f (x) := lim+ ν j (y)(∂k EΔ )(x − y) − νk (y)(∂j EΔ )(x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
(1.5.212)

and (compare with (1.5.82))

∫
 
Tjk f (x) := lim+
#
ν j (x)(∂k EΔ )(x − y) − νk (x)(∂j EΔ )(x − y) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(1.5.213)

where EΔ is the standard fundamental solution for the Laplacian in Rn recalled in

(1.1.1). Finally, fix some integrability exponent p ∈ (1, ∞) and pick some Mucken-
houpt weight w ∈ Ap (∂Ω, σ). Then

1   1  n

2 I + KΔ − 12 I + KΔ =
2 j,k=1
(Tjk )2 on L p (∂Ω, wσ), (1.5.214)

and
1   1  n

2 I + KΔ − 12 I + KΔ# = (T # )2
#
2 j,k=1 jk

1  
n
2
= ∂τ j k SΔ,mod on L p (∂Ω, wσ). (1.5.215)
2 j,k=1

Proof Select an arbitrary function f ∈ L p (∂Ω, wσ). Then formula (1.5.89) permits
αβ
us to write (with ar s := δr s , with M := 1, and with the Greek letter suppressed)

16 these are the versions of the singular integral operators from (1.2.2)-(1.2.3) corresponding to
choosing b to be the standard fundamental solution EΔ for the Laplacian in R n
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 127
1   
2I + KΔ# − 12 I + KΔ# f (x)
∫
 # 
= lim+ νi (x)δi j δr s (∂r EΔ )(x − y) Tjs f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
 # 
= lim+ ν j (x)(∂s EΔ )(x − y) Tjs f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫ 
1
= lim+ ν j (x)(∂s EΔ )(x − y)
2 ε→0
y ∈∂Ω
|x−y |>ε

# 
− νs (x)(∂j EΔ )(x − y) Tjs f (y) dσ(y)

1 #  # 
= T T f (x) at σ-a.e. point x ∈ ∂Ω, (1.5.216)
2 js js
with the summation convention over repeated indices in effect. In the third equality
# is antisymmetric in j and s, while the final
above we have used the fact that Tjs
equality is seen from (1.5.213).
From (1.5.216) we then conclude that the first equality in (1.5.215) holds. The
second equality in (1.5.215) is a consequence of (1.5.83) (plus the subsequent
comment) and definitions. Finally, formula (1.5.214) is implied by the first equality
in (1.5.215) and duality. 

Moving on, our next theorem elaborates on the properties of acoustic bound-
ary layer potentials from [70, §6.4], now considered in open sets with uniformly
rectifiable boundaries.

Theorem 1.5.5 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with the

property that ∂Ω is a compact UR set. Abbreviate σ := H n−1 ∂Ω, and denote by
ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal
 to Ω. Also,
for some M ∈ N, consider a complex coefficient tensor A := arJsI 1≤J,I ≤M which
1≤r,s ≤n
allows writing the vector Laplacian as Δ = divA∇, i.e.,
 
n
ΔI M×M = arJsI ∂r ∂s (1.5.217)
1≤J,I ≤M
r,s=1

where I M×M is the M × M identity matrix. Finally, fix a wave number k ∈ (0, ∞)

and recall from (A.0.75) the radiating fundamental solution Φk of the Helmholtz
operator Δ + k 2 . In this setting, for each vector-valued function f = ( fJ )1≤J ≤M
M
in [L 1 (∂∗ Ω, σ) define the boundary-to-domain acoustic double layer potential
operator at each x ∈ Ω as
128 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫ !
D f (x) := − νs (y)arI sJ (∂r Φk )(x − y) fJ (y) dσ(y) , (1.5.218)
∂∗ Ω 1≤I ≤M

and its principal-value version, the boundary-to-boundary acoustic double layer

potential operator defined at σ-a.e. x ∈ ∂Ω as
# ∫ $
K f (x) := − lim+ νs (y)arI sJ (∂r Φk )(x − y) fJ (y) dσ(y) . (1.5.219)
ε→0
y ∈∂∗ Ω 1≤I ≤M
|x−y |>ε

M
Next, for each f = ( fI )1≤I ≤M ∈ [L 1 (∂Ω, σ) define the “transpose” boundary-
to-boundary acoustic double layer potential operator at σ-a.e. x ∈ ∂∗ Ω as
# ∫ $
K # f (x) := lim νs (x)arI sJ (∂r Φk )(x − y) fI (y) dσ(y) , (1.5.220)
ε→0+
y ∈∂∗ Ω 1≤J ≤M
|x−y |>ε

as well as the boundary-to-domain acoustic single layer potential operator

∫ !
𝒮 f (x) := Φk (x − y) fI (y) dσ(y) for all x ∈ Ω, (1.5.221)
∂Ω 1≤I ≤M

and its boundary-to-boundary version

∫ !
S f (x) := Φk (x − y) fI (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (1.5.222)
∂Ω 1≤I ≤M

Then for each aperture parameter κ ∈ (0, ∞) and for each truncation parameter
ρ ∈ (0, ∞) the following properties hold.
M
(i) For every function f ∈ L 1 (∂∗ Ω, σ) one has

D f ∈ 𝒞∞ (Ω)
M
and (Δ + k 2 )D f = 0 in Ω. (1.5.223)

Moreover, if Ω is an exterior domain then D f radiates at infinity (cf. [70,

Definition 6.1.1]).
Also, for each p ∈ (1, ∞) there exists some constant C ∈ (0, ∞) such that for
M
every f ∈ L p (∂∗ Ω, σ) one has
 ρ 
Nκ (D f ) p ≤ C f [L p (∂∗ Ω,σ)] M . (1.5.224)
L (∂Ω,σ)

M
(ii) For each function f belonging to the boundary Sobolev space L11 (∂∗ Ω, σ)
and each index ∈ {1, . . . , n} the pointwise nontangential boundary trace
  κ−n.t.
∂ D f ∂Ω
exists (in C M ) at σ-a.e. point on ∂∗ Ω. (1.5.225)
 1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 129

Moreover, for each p ∈ (1, ∞) there exists some finite constant C > 0 such that
p M
for each function f ∈ L1 (∂∗ Ω, σ) one has
 ρ   ρ 
Nκ (D f ) p + Nκ (∇D f ) L p (∂Ω,σ) ≤ C f [L p (∂∗ Ω,σ)] M . (1.5.226)
L (∂Ω,σ) 1

(iii) Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operators

M M
K : L p (∂∗ Ω, σ) −→ L p (∂Ω, σ) , (1.5.227)
 M  M
K # : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ) , (1.5.228)

are well-defined, linear, bounded, and transpose to one another.

M
(iv) If I denotes the identity operator, then for each function f ∈ L 1 (∂∗ Ω, σ) the
following nontangential boundary trace formula holds:
κ−n.t.
1 
Df = 2I + K f at σ-a.e. point on ∂∗ Ω. (1.5.229)
∂Ω

(v) Strengthen the original hypotheses on the underlying set by assuming

 thataddi-
tional assumption that Ω is a UR domain; in particular, H n−1 ∂Ω \ ∂∗ Ω = 0.
Also, fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operators
p M p M
K : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) , (1.5.230)
p M p M
K # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) , (1.5.231)

are well-defined, linear, bounded, and transpose to one another.

(vii) Continue to assume that Ω is a UR domain and recall the conormal derivative
operator associated with the coefficient tensor A and the domain Ω as in [70,
(6.4.4)]. Also, fix some p ∈ (1, ∞). Then the conormal derivative of the double
layer induces a well-defined, linear, and bounded operator in the context
M   M
L p (∂Ω, σ)  f −→ ∂νA D f := ∂νA(D f ) ∈ L p (∂Ω, σ) . (1.5.232)

(viii) Continue to assume that Ω is actually a UR domain. Also, pick two integrability
exponents p, p ∈ (1, ∞) satisfying 1/p + 1/p = 1. Then for any two given
p M p M
vector-valued functions f ∈ L1 (∂Ω, σ) and g ∈ L1 (∂Ω, σ) one has
∫ ∫
 A      
∂ν D f , g dσ = f , ∂νA D A g dσ (1.5.233)
∂Ω ∂Ω
 
where ∂νA along with D A and, ultimately ∂νA D A , are defined as before with
A now replaced by A . As a consequence, whenever Ω is a UR domain, the
operator (1.5.31) has a unique extension to a well-defined, linear, and bounded
mapping
130 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
M p M
∂νA D : L p (∂Ω, σ) −→ L−1 (∂Ω, σ) , (1.5.234)

namely the (real) transpose of the operator

 p M  M
∂νA D A : L1 (∂Ω, σ) −→ L p (∂Ω, σ) . (1.5.235)

(ix) Once again assume Ω is a UR domain. Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1.
Then the boundary-to-boundary single layer potential operator S induces well-
defined, linear, bounded, and compatible fashion with one another mappings
M p M
S : L p (∂Ω, σ) −→ L1 (∂Ω, σ) , (1.5.236)
p M M
S : L−1 (∂Ω, σ) −→ L p (∂Ω, σ) , (1.5.237)

which are also transpose to one another.

(x) Assume Ω is a UR domain and fix p, p ∈ (1, ∞) satisfying 1/p+1/p = 1. Extend
the action of the boundary-to-boundary single layer 𝒮, originally defined in
(1.5.221), to each f = ( fI )1≤I ≤M belonging to the negative Sobolev space
p M
L−1 (∂Ω, σ) by setting
   
𝒮 f I (x) := [L p (∂Ω,σ)] M Φk (x − ·) ∂Ω, fI [L p (∂Ω,σ)] M
−1
1 (1.5.238)
for every x ∈ Ω and every I ∈ {1, . . . , M }.

Then this operator is meaningfully defined and agrees with 𝒮 from (1.5.221)
M
when acting on the smaller space L p (∂Ω, σ) . Moreover,

−→ 𝒞∞ (Ω)
p M M
𝒮 : L−1 (∂Ω, σ) (1.5.239)

is continuous (when the space on the right is equipped with the Frechét topology
of uniform convergence of partial derivatives of any order on compact sets), and
   
∂γ 𝒮 f I (x) = [L p (∂Ω,σ)] M (∂γ Φk )(x − ·) ∂Ω, fI [L p (∂Ω,σ)] M (1.5.240)
1 −1

p M
for each f = ( fI )1≤I ≤M ∈ L−1 (∂Ω, σ) , each index I ∈ {1, . . . , M }, each
multi-index γ ∈ N0n , and each point x ∈ Ω. Moreover,

p M
(Δ + k 2 )(𝒮 f ) = 0 in Ω, for each functional f ∈ L−1 (∂Ω, σ) , (1.5.241)

and if Ω is an exterior domain then 𝒮 f also radiates at infinity (cf. [70,

p M
Definition 6.1.1]). In addition, for each f ∈ L−1 (∂Ω, σ) the nontangential
κ−n.t.
pointwise trace 𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω; in fact, with S f considered
in the sense of (1.5.237) one has
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 131
κ−n.t.

𝒮f = S f at σ-a.e. point on ∂Ω. (1.5.242)

∂Ω

Furthermore, there exists a finite constant C > 0 such that

κ−n.t.
the nontangential trace (∇𝒮 f ) ∂Ω exists σ-a.e. on ∂Ω and
 ρ 
Nκ (𝒮 f ) p p
≤ C f [L p (∂Ω,σ)] M for all f ∈ L−1 (∂Ω, σ) ,
M
L (∂Ω,σ) −1
 ρ 
Nκ (∇𝒮 f ) p ≤ C f [L p (∂Ω,σ)] M for all f ∈ L p (∂Ω, σ)
M
L (∂Ω,σ)
(1.5.243)
M
Finally, for each f ∈ L 1 (∂Ω, σ) one has the jump-formula
 
∂νA𝒮 f = − 12 I + K A#  f at σ-a.e. point on ∂Ω, (1.5.244)

where I is the identity operator, and K A#  is the operator associated with the
coefficient tensor A and the set Ω as in (1.5.220).
(xi) Suppose that in fact Ω is a UR domain. For each pair of indices j, ∈ {1, . . . , n}
M
and each function f = ( fI )1≤I ≤M ∈ L 1 (∂Ω, σ) define
# ∫

#
Tj f (x) := lim+ ν j (x)(∂ Φk )(x − y)
ε→0
y ∈∂Ω
|x−y |>ε $

− ν (x)(∂j Φk )(x − y) fI (y) dσ(y)
1≤I ≤M
(1.5.245)

at σ-a.e. point x ∈ ∂Ω. Then

M
for each f ∈ L 1 (∂Ω, σ) and j, ∈ {1, . . . , n}
(1.5.246)
one has ∂τ j  (S f ) = Tj
# f at σ-a.e. point on ∂Ω.

(xii) Make the assumption that Ω is a UR domain. Then for each p ∈ (1, ∞) the
following operator identities hold:
1      M
2I + K ◦ − 12 I + K = S ◦ ∂νA D on L(∂Ω, σ) , (1.5.247)
1      p M
2I + K A#  ◦ − 12 I + K A#  = ∂νA D ◦ S on L−1 (∂Ω, σ) , (1.5.248)
p M
S ◦ K A#  = K ◦ S on L−1 (∂Ω, σ) , (1.5.249)
    M
K A#  ◦ ∂νA D = ∂νA D ◦ K on L p (∂Ω, σ) . (1.5.250)
132 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Proof This is a consequence of [70, Lemma 6.3.1] and Theorem 1.5.5 (together with
its proof) used for the weakly elliptic system L := ΔI M×M in Rn ; other results from
[70, Chapter
  6] also play a role. The key observation is that under the identification
E ≡ Φk δI J 1≤I,J ≤M the operators (1.3.18), (1.3.68), (1.3.72), (1.3.6), and (1.3.62)
correspond precisely to (1.5.218), (1.5.219), (1.5.220), (1.5.221), and (1.5.222),
respectively. 

Our next result elaborates on the specific manner in which tangential derivatives
commute with the action of the boundary-to-boundary double layer potential operator
on functions belonging to boundary Sobolev spaces.

Proposition 1.5.6 Suppose Ω ⊆ Rn is a UR domain. Abbreviate σ := H n−1 ∂Ω and

denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω.
Let L be a homogeneous, second-order, constant complex coefficient, weakly elliptic
M × M system in Rn and let E = (Eαβ )1≤α,β ≤M be the matrix-valued fundamental
solution
 associated with L as in [70, Theorem 1.4.2]. Also, pick a coefficient tensor
αβ
A = ar s 1≤α,β ≤M for which L A = L and bring in K, the boundary-to-boundary
1≤r,s ≤n
double layer potential operator associated with Ω and A as in (1.3.68). In addi-
tion, for each j, k ∈ {1, . . . , n} define the singular integral operator
 U jk acting on
each given function g = (gαs )1≤α ≤M with entries belonging to L 1 ∂Ω, 1+σ(x)
|x | n−1
as
1≤s ≤n
U jk g = (U jk g)1≤γ ≤M where, for every γ ∈ {1, . . . , M } and for σ-a.e. point x ∈ ∂Ω,

(U jk g)γ (x) (1.5.251)

∫
βα
:= − lim+ [νk (x) − νk (y)]ν j (y)ar s (∂r Eγβ )(x − y)gαs (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
+ lim+ [ν j (x) − ν j (y)]νk (y)ar s (∂r Eγβ )(x − y)gαs (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
+ lim+ [νk (y) − νk (x)]νs (y)ar s (∂r Eγβ )(x − y)gα j (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
− lim+ [ν j (y) − ν j (x)]νs (y)ar s (∂r Eγβ )(x − y)gαk (y) dσ(y).
ε→0
y ∈∂Ω
|x−y |>ε

Finally, fix some integrability exponent p ∈ (1, ∞) and consider a function

1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 133
  M
f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n−1
∩ Lloc (∂Ω, σ) such that

∂τ j k fα belongs to L 1 ∂Ω, 1+σ(x) (1.5.252)
p
|x | n−1
∩ Lloc (∂Ω, σ)
for all j, k ∈ {1, . . . , n} and α ∈ {1, . . . , M }.

Then for each j, k ∈ {1, . . . , n} one has

∂τ j k (K f ) = K(∂τ j k f ) + U jk (∇tan f ), (1.5.253)

where ∇tan f is regarded here as the M × n matrix-valued function whose (α, s) entry
is the s-th component of the tangential gradient ∇tan fα (cf. (A.0.78)).

As an example, consider the case when M := 1 and L := Δ, the Laplace operator

in Rn . Denote by KΔ the standard boundary-to-boundary harmonic double layer
operator associated with Ω, and recall the boundary-to-boundary Riesz transforms
Ri , 1 ≤ i ≤ n, on ∂Ω. Also, for each j ∈ {1, . . . , n}, denote by Mν j the operator
of pointwise multiplication by ν j , the j-th scalar component of ν. In this particular
setting, Proposition 1.5.6 amounts to saying that for any scalar function with the
property that there exists some p ∈ (1, ∞) such that
 σ(x) p
f ∈ L 1 ∂∗ Ω, ∩ Lloc (∂Ω, σ) and
1 + |x| n−1
 (1.5.254)
σ(x) p
∂τ j k f ∈ L ∂∗ Ω,
1
∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n}
1 + |x| n−1

it follows that for each pair of indices j, k ∈ {1, . . . , n} one has

∂τ j k (K f ) = K(∂τ j k f )
   
+ 1
2 Ri, Mνk ν j (∇tan f )i − 12 Ri, Mν j νk (∇tan f )i
   
− 12 Ri, Mνk νi (∇tan f ) j + 12 Ri, Mν j νi (∇tan f )k
   
= K(∂τ j k f ) + 12 Ri, Mν j ∂τi k f − 12 Ri, Mνk ∂τi j f , (1.5.255)

where the last equality comes from [69, (11.4.8)]. In summary, tangential partial
derivatives commute with the principal-value harmonic double layer, modulo com-
mutators between Riesz transforms and operators of pointwise multiplication by the
scalar components of the geometric measure theoretic outward unit normal.
Proof of Proposition 1.5.6 Fix j, k ∈ {1, . . . , n} along with γ ∈ {1, . . . , M } and
recall the definition of the boundary-to-domain double layer potential operator from
(1.3.18), as well as the definition of the operator introduced in [70, (2.9.53)]. Making
use of (1.3.28) and [69, (11.4.8)], for each x ∈ Ω we may then write
134 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
βα
∂j (D f )γ (x) = ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
∂Ω
∫
βα
− νs (y)ar s (∂r Eγβ )(x − y)(∇tan fα ) j (y) dσ(y)
∂Ω
βα    
= ar s (∂r 𝒮γβ ) ν j (∇tan fα )s (x) + D((∇tan f ) j ) γ (x), (1.5.256)

where (∇tan f ) j is the C M -valued function whose α-th component is (∇tan fα ) j . Fix
an aperture parameter κ > 0. From Lemma 1.3.2, [70, (2.4.8)], and [69, Proposi-
tion 11.3.2], we see that formula [69, (11.3.26)] holds for u := D f . Based on this,
(1.5.256), and the jump formulas in (1.5.20) and in [70, Corollary 2.9.5], at σ-a.e.
point x ∈ ∂Ω we may then write (bearing in mind [70, (2.9.52)])
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 135

∂τ j k (K f )γ (x) (1.5.257)

= ∂τ j k ( 12 f + K f )γ (x) − 12 ∂τ j k fγ (x)
κ−n.t. κ−n.t.
= ν j (∂k D f )γ ∂Ω
(x) − νk (∂j D f )γ ∂Ω
(x) − 12 ∂τ j k fγ (x)
βα
= − 12 ν j (x)νk (x)νr (x)ar s bγβ (x)(∇tan fα )s (x)
∫
βα
+ lim+ ν j (x)νk (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
  k
+ 12 ν j (x)(∇tan fγ )k (x) + ν j (x) K ∇tan f (x)
γ
βα
+ 12 ν j (x)νk (x)νr (x)ar s bγβ (x)(∇tan fα )s (x)
∫
βα
− lim+ νk (x)ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
  j
− 12 νk (x)(∇tan fγ ) j (x) − νk (x) K ∇tan f (x)
γ

− 12 ∂τ j k fγ (x)
∫
βα
= − lim+ νk (x)ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
+ lim+ ν j (x)νk (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
  j   k
+ K, Mνk ∇tan f (x) − K, Mν j ∇tan f (x)
γ γ
  
+ K ν j (∇tan f )k − νk (∇tan f ) j (x), (1.5.258)
γ

where for each ∈ {1, . . . , n} we have denoted by Mν the operator of pointwise
multiplication by ν , and K, Mν := K Mν − Mν K is the commutator of K with
Mν . Thanks to [69, (11.4.8)], the last line of (1.5.257) may be recast as
    
K ν j (∇tan f )k − νk (∇tan f ) j (x) = K(∂τ j k f ) γ (x). (1.5.259)
γ

Also, simple algebra gives

136 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫
βα
− lim+ νk (x)ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
+ lim+ ν j (x)νk (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
= − lim+ [νk (x) − νk (y)]ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
∫
βα
+ lim+ [ν j (x) − ν j (y)]νk (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y).
ε→0
y ∈∂Ω
|x−y |>ε
(1.5.260)

Having made these observations, there remains to note that if ∇tan f is regarded as
a matrix-valued function whose (α, s) entry is the s-th component of ∇tan fα , then
(1.5.257) may be rewritten as claimed in (1.5.253). 

Our final result in this section describes the manner in which the double layer
potential operators associated with a given weakly elliptic second-order M × M
system L in Rn and a set of locally finite perimeter Ω ⊆ Rn transform if one
performs a linear change of variables in the Euclidean space via W : Rn → Rn for
some non-singular matrix W ∈ Rn×n , or if L is multiplied (to the left, or to the right)
by a non-singular matrix C ∈ C M×M .

Proposition 1.5.7 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set of locally

finite perimeter. Abbreviate σ := H n−1 ∂Ω, and denote by ν the geometric measure
 αβ  unit normal to Ω. Also, for some M ∈ N, consider a coefficient
theoretic outward
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense that det [L(ξ)]  0 for each ξ ∈ Rn \ {0}).
Given a non-singular matrix W ∈ Rn×n , consider the system L ◦ W, formally
defined as 
αβ
L ◦ W := a jk (W∇) j (W∇)k . (1.5.261)
1≤α,β ≤M

From [70, Proposition 1.4.3] it is known that, with notation introduced in [70,
(1.4.56)-(1.4.57)],
 := W  ◦ A ◦ W ∈ A L◦W and L ◦ W is weakly elliptic.
A (1.5.262)
 := W  Ω has the same (geometric mea-
Also, from [39] it is know that the set Ω
sure theoretic) nature as Ω. Denote by ν its geometric measure theoretic outward
unit normal, and set   In particular, it makes sense to consider
σ := H n−1 ∂ Ω.
1.5 Calderón-Zygmund Function Theory for Boundary Layer Potentials 137

the (boundary-to-domain and boundary-to-boundary) double layer potential oper-

 as in (1.3.18) and (1.3.68),
ators17 associated with the system L ◦ W and the set Ω
respectively.
Then, for these operators the following formulas hold:
  
D A,
Ω
 −1
 g = D A,Ω (g ◦ W ) ◦ (W )

in Ω, (1.5.263)

and
  
K A,
Ω
 −1
 g = K A,Ω (g ◦ W ) ◦ (W ) at  
σ -a.e. point on ∂ Ω, (1.5.264)
  (x)  M
for each function g ∈ L 1 ∂∗ Ω, σ .
1+ |x | n−1
Furthermore, given any non-singular matrix C = (cβγ )1≤β,γ ≤M ∈ C M×M , the
systems LC and CL (interpreted in the sense of multiplication of M × M matrices)
are weakly elliptic, and if one defines
 αβ   αβ 
A  C := a jk cβγ 1≤α,γ ≤M , C  A := cγα a jk 1≤γ,β ≤M , (1.5.265)
1≤ j,k ≤n 1≤ j,k ≤n

then
A  C ∈ A LC , C  A ∈ AC L, (1.5.266)
and one has

D AC,Ω = C −1 ◦ D A,Ω ◦ C, DCA,Ω = D A,Ω, (1.5.267)

as well as

K AC,Ω = C −1 ◦ K A,Ω ◦ C, KCA,Ω = K A,Ω . (1.5.268)

Proof Formula (1.5.263) is seen from (1.3.18), (1.5.262), [70, (1.4.56)-(1.4.57)],

the “surface-to-surface” change of variable formula [69, (11.6.4)] (used for the
diffeomorphism F : Rn → Rn given by F(x) := W  x for each x ∈ Rn ), formula [69,
(11.6.1)] describing the geometric measure theoretic outward unit normal  
ν to Ω,
item (iii) in [70, Proposition 1.4.3] (describing the “canonical” fundamental solution
of L ◦ W), and the Chain Rule. Formula (1.5.264) is proved in a similar fashion, now
starting from (1.3.68) in place of (1.3.18). Finally, formulas (1.5.267)-(1.5.268) may
be justified directly from (1.5.265), (1.3.18), and (1.3.68). 

17 For the sake of clarity, we shall use both the coefficient tensor and the underlying set as subscripts
in the notation for the aforementioned operators
138 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and

Sobolev Spaces

From the discussion in (1.4.13)-(1.4.16) we know that the boundary-to-domain

and boundary-to-boundary Cauchy-Clifford integral operators may be regarded as
double layer operators associated with the Laplacian. In view of this, and granted
the availability of the rich Calderón-Zygmund theory developed in relation to layer
potential operators associated with arbitrary weakly elliptic second-order systems,
a host of results may instantaneously be deduced for the aforementioned Cauchy-
Clifford integral operators on Lebesgue and Sobolev spaces defined on UR sets.
Alternatively, the results which are now available from Theorem 1.5.1 and Propo-
sition 1.2.1 also allow for a quick and painless treatment of the Cauchy-Clifford
integral operator on boundary Sobolev spaces. Since this latter approach reveals
more about the algebraic structure of the Cauchy-Clifford integral operators, we
present it below, in the proof of the following proposition:

Proposition 1.6.1 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with the

property that ∂Ω is a UR set. Abbreviate σ∗ := H n−1 ∂∗ Ω and σ := H n−1 ∂Ω.
Also, denote by ν the geometric measure theoretic outward unit normal to Ω. Recall
the Cauchy-Clifford integral operators C, C associated with Ω as in (A.0.53) and
(A.0.54), respectively. In addition, define the action of the “transpose” Cauchy-
Clifford singular integral operator on any given Clifford algebra-valued function
f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
⊗Cn
∫
−1 x−y
C # f (x) := lim+ ν(x)   f (y) dσ(y) (1.6.1)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂∗ Ω. Then the following statements are true.

p,q
(i) For each function f ∈ L1 (∂∗ Ω, σ∗ ) ⊗ C n with exponents p, q ∈ [1, ∞), index
∈ {1, . . . , n}, and aperture parameter κ ∈ (0, ∞), the pointwise nontangential
boundary trace
  κ−n.t.
∂ C f ∂Ω
exists (in C n ) at σ∗ -a.e. point on ∂∗ Ω, (1.6.2)

and there exists some finite constant C > 0 depending only on ∂Ω, n, p, q, and
κ such that
   
Nκ (C f ) p + Nκ (∇C f ) L q (∂Ω,σ) ≤ C f  L p, q (∂∗ Ω,σ∗ )⊗ Cn
L (∂Ω,σ) 1
(1.6.3)
if p ∈ (1, ∞) and q ∈ (1, ∞),

plus similar estimates in the case when either p = 1 or q = 1, this time with the
corresponding L 1 -norm in the left side replaced by the weak-L 1 (quasi-)norm.
(ii) Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operator
1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 139

C # : L p (∂Ω, σ) ⊗ C n −→ L p (∂∗ Ω, σ∗ ) ⊗ C n (1.6.4)

is well defined, linear, and bounded. Also, corresponding to p = 1,

C # : L 1 (∂Ω, σ) ⊗ C n −→ L 1,∞ (∂∗ Ω, σ∗ ) ⊗ C n (1.6.5)

is well defined, linear, and bounded. Finally, the transpose of C # in (1.6.4) is

 
C : L p (∂∗ Ω, σ∗ ) ⊗ C n −→ L p (∂Ω, σ) ⊗ C n . (1.6.6)

(iii) Make the additional assumption that

 
H n−1 ∂nta Ω \ ∂∗ Ω = 0. (1.6.7)

Then for each p ∈ (1, ∞) one has

 2
C# = 14 I on L p (∂∗ Ω, σ∗ ) ⊗ C n . (1.6.8)

Also, the operator

p,q p,q
C : L1 (∂∗ Ω, σ∗ ) ⊗ C n −→ L1 (∂∗ Ω, σ∗ ) ⊗ C n (1.6.9)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
p p
C : L1 (∂∗ Ω, σ∗ ) ⊗ C n −→ L1 (∂∗ Ω, σ∗ ) ⊗ C n (1.6.10)

is well defined, linear, and bounded for each p ∈ (1, ∞).

(iv) Retain the additional assumption made in (1.6.7). Then for each p, q ∈ (1, ∞)
it follows that the mapping C # from (1.6.4) extends uniquely to a linear and
bounded operator from the negative off-diagonal boundary Sobolev space
p,q
L−1 (∂∗ Ω, σ∗ ) ⊗ C n into itself. Furthermore, if one retains the same notation C #
for this extension, the transpose of (1.6.9) is
p,q p,q
C # : L−1 (∂∗ Ω, σ∗ ) ⊗ C n −→ L−1 (∂∗ Ω, σ∗ ) ⊗ C n, (1.6.11)

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1. In
particular, the transpose of (1.6.10) is
p p
C # : L−1 (∂∗ Ω, σ∗ ) ⊗ C n −→ L−1 (∂∗ Ω, σ∗ ) ⊗ C n . (1.6.12)

As a corollary of [70, (2.5.324)], item (ii) in Proposition 1.6.1, density, and duality,
we also see that if Ω ⊆ Rn , where n ≥ 2, is a UR domain and σ := H n−1 ∂Ω then
140 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

C # is bounded on L p (∂Ω, w) ⊗ C n and

 2
we have C # = 14 I on L p (∂Ω, w) ⊗ C n (1.6.13)
whenever p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ).

This may also be seen from [70, Corollary 2.5.33] and duality.
Proof of Proposition 1.6.1 Upon recalling from [70, (2.5.311)] that

1 
n
Cf = Df + · e j  ek  R jk f in Ω
2 j,k=1 (1.6.14)

for each function f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n−1
⊗ C n,

where D is the harmonic double layer (cf. [70, Definition 2.5.17]) and the operators
R jk are as in (A.0.189), the claims in item (i) are direct consequences of part (ii) in
Theorem 1.5.1 and part (i) in Proposition 1.2.1. Next, the fact that the operator C #
is well defined, linear, and bounded both in the context of (1.6.4) and in the case of
(1.6.5) is guaranteed by [70, Theorem 2.3.2]. To proceed, fix p ∈ (1, ∞) is such that
1/p + 1/p = 1 and recall from [70, (2.5.312)] that

1 
n
C=K+ · e j  ek  R jk on L p (∂∗ Ω, σ∗ ) ⊗ C n, (1.6.15)
2 j,k=1

where K is the principal-value harmonic double layer (cf. [70, (2.5.203)]) and the
operators R jk have been defined in (A.0.190). In relation to the latter family of
operators, for each j, k ∈ {1, . . . , n} and f ∈ L p (∂Ω, σ) ⊗ C n let us also define
(with ν1, . . . , νn denoting the components of ν)
∫  
R#jk f (x) := lim+ ν j (x)(∂k EΔ )(y − x) − νk (x)(∂j EΔ )(y − x) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
∫
−1 ν j (x)(xk − yk ) − νk (x)(x j − y j )
= lim+ f (y) dσ(y)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε
(1.6.16)

for σ∗ -a.e. x ∈ ∂∗ Ω. Then, on the one hand, from (1.6.15), (1.6.16), part (iii) in
Theorem 1.5.1 (used with L := Δ), part (ii) in Proposition 1.2.1 (used with b := EΔ
which is even and positive homogeneous of degree 2 − n if n ≥ 3; the case n = 2 is
a minor variation), and [68, Lemma 6.4.1] it follows that the transpose of (1.6.15) is
the operator
1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 141

1 
n
K# − · ek  e j  R#jk
2 j,k=1

1 
n

= K# + · e j  ek  R#jk on L p (∂Ω, σ) ⊗ C n, (1.6.17)
2 j,k=1

where K # is defined as in (1.3.72) corresponding to L := Δ, i.e.,

∫
1 ν(x), x − y
K # f (x) = lim+ f (y) dσ(y) (1.6.18)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ∗ -a.e. x ∈ ∂∗ Ω. On the other hand, starting with (1.6.1) and reasoning much as
in [70, (2.5.310)] (based on [68, (6.4.2)-(6.4.3)]) we see that

1 
n
C# = K # + · e j  ek  R#jk . (1.6.19)
2 j,k=1

From these, the last claim in item (ii) then readily follows. Finally, in view of (1.6.15),
(1.6.19), and [70, (2.5.324)], the claims in items (iii)-(iv) are implied by parts (v)-(vi)
in Theorem 1.5.1, parts (iii)-(iv) in Proposition 1.2.1, and [70, Proposition 2.5.32]
(bearing in mind the current item (ii)). 

The next corollary contains a couple of useful identities involving Cauchy-Clifford

singular integral operators and Riesz transforms on the boundary of a given UR
domain.

Corollary 1.6.2 Let Ω ⊆ Rn , where n ≥ 2, be a UR domain. Set σ := H n−1 ∂Ω

and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal
to Ω. Also, let Mν be the operator of pointwise Clifford algebra multiplication from
the left by ν ≡ ν1 e1 + · · · + νn en . Then
 σ(x)
C # = Mν CMν on L 1 ∂Ω, ⊗Cn (1.6.20)
1 + |x| n−1

and for each w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞) one has

1 2 1 
n
CC # = − R + e j  ek [R j , Rk ] on L p (∂Ω, w) ⊗ C n (1.6.21)
4 j=1 j 4 1≤ j<k ≤n

where (R j )1≤ j ≤n are the Riesz transforms associated with Σ := ∂Ω as in (A.0.187).

Proof Formula (1.6.20) is a consequence of (A.0.54), (1.6.1), and the fact that at
σ-a.e. point on ∂Ω we have ν  ν = −1 (cf. [68, (5.6.21), (6.4.1)], (A.0.178), and the
fact that H n−1 (∂Ω \ ∂∗ Ω) = 0). Next observe from (A.0.53) and [68, (6.4.1)] that
142 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

1
n  σ(x)
CMν = − e j R j on L 1 ∂Ω, ⊗ C n. (1.6.22)
2 j=1 1 + |x| n−1

Fix w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞). From (1.6.20), (1.6.22), and [68, (7.7.106)] we
then deduce that, as operators on L p (∂Ω, w) ⊗ C n ,

      1 n  1 n
CC # = C Mν CMν = CMν CMν = − e j Rj − ek Rk
2 j=1 2 k=1

1  1 2 1 
n n
= e j  ek R j Rk = − R + e j  ek R j Rk
4 j,k=1 4 j=1 j 4 1≤ jk ≤n

1 2 1 
n
=− R + e j  ek (R j Rk − Rk R j ). (1.6.23)
4 j=1 j 4 1≤ j<k ≤n

Based on (1.6.23) and (1.6.20), we may therefore conclude that (1.6.21) holds. 

Special geometries lead to specialized formulas, as described in the next lemma.

Lemma 1.6.3 If Ω is either a ball or a half-space in Rn , then its outward unit normal
ν satisfies

−ν(x)  (x − y) = (x − y)  ν(y) for each x, y ∈ ∂Ω. (1.6.24)

Proof Suppose Ω is a ball in Rn . Via a translation and a dilation, there is no loss of

generality in assuming that Ω = B(0, 1). Then ν(x) = x for each x ∈ ∂Ω, hence

ν(x)  (x − y) + (x − y)  ν(x) = x  (x − y) + (x − y)  y

= −1 − x  y + x  y − (−1)

= 0 for each x, y ∈ ∂Ω, (1.6.25)

which establishes the formula claimed in (1.6.24) in this case. Finally, if for some
ξ ∈ S n−1 and x0 ∈ Rn we have Ω = x ∈ Rn : x − x0, ξ < 0 , then ν = ξ on ∂Ω.
Keeping this in mind, [68, (6.4.6)] allows to write

ν(x)  (x − y) + (x − y)  ν(x) = ξ  (x − y) + (x − y)  ξ

= −2 ξ, x − y

= 0 for each x, y ∈ ∂Ω, (1.6.26)

proving (1.6.24) in this case. 

As a result of Lemma 1.6.3, the Cauchy-Clifford singular integral agrees with its
(formal) transpose on spheres and hyperplanes.
1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 143

Lemma 1.6.4 If Ω is either

 a ball or a half-space in Rn , then C = C # as operators
H n−1 (x)
acting on the space L ∂Ω, 1+
1
|x | n−1
⊗ C n.

Proof This is a direct consequence of Lemma 1.6.3 and definitions (cf. (A.0.54),
(1.6.1)). 

In terms of Riesz transforms, the result in Lemma 1.6.4 yields the identities
described in the next corollary.

Corollary 1.6.5 Suppose Ω is either a ball or a half-space in Rn , where n ≥ 2, and

abbreviate σ := H n−1 ∂Ω. Let (R j )1≤ j ≤n are the Riesz transforms associated with
Σ := ∂Ω as in (A.0.187). Also, fix some w ∈ Ap (∂Ω, σ) with p ∈ (1, ∞). Then, as
operators on L p (∂Ω, w),

n
R2j = −I and R j Rk = Rk R j for each j, k ∈ {1, . . . , n}. (1.6.27)
j=1

Proof From (1.6.21), Lemma 1.6.4, and [70, (2.5.324)] we obtain

1 2 1 
n
1
I=− R + e j  ek [R j , Rk ] on L p (∂Ω, w) ⊗ C n . (1.6.28)
4 4 j=1 j 4 1≤ j<k ≤n

Then (1.6.27) follows by restricting ourselves to scalar-valued functions in L p (∂Ω, w)

and considering the vector and bi-vectors components in this identity separately. 

Here is another way of tying up Riesz transforms with single and double layers
for the (vector) Laplacian; this is going to be of relevance to us later on.

Proposition 1.6.6 Let Ω ⊆ Rn , where n ≥ 2, be a set of locally finite perimeter.

Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric measure
theoretic outward unit normal to Ω. Bring in the modified version 𝒮mod of the single
layer operator associated with the Laplacian Δ and the set Ω as in (1.5.50). Also,
recall boundary-to-domain Riesz transforms, R j indexed by j ∈ {1, . . . , n}, defined
as in (A.0.188) (with Σ := ∂Ω). Next, consider a family of 2n × 2n real matrices
{E j }1≤ j ≤n satisfying18
 2
Ej = −I2n ×2n for each j ∈ {1, . . . , n} and
(1.6.29)
E j Ek = −Ek E j for all j, k ∈ {1, . . . , n} with j  k,

and define first-order 2n × 2n system


n
D := E j ∂j . (1.6.30)
j=1

18 The existence of such a family has been established in [68, (6.4.14)] (with m := n)
144 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
   2n
Then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has19


n
2D𝒮mod f = E j R j f in Ω̊. (1.6.31)
j=1

Also, if D is the boundary-to-domain double layer potential operator associated as

in (1.3.18) with the set Ω and the vector Laplacian written as in (1.4.21), i.e., if for
each function
  σ(x)  2n
f ∈ L 1 ∂∗ Ω, (1.6.32)
1 + |x| n−1
and every x ∈ Ω̊ one defines
∫
1 
n
xj − yj  
n
D f (x) = Ej νk (y)Ek f (y) dσ(y), (1.6.33)
ωn−1 ∂∗ Ω j=1 |x − y| n k=1

then for each f as in (1.6.32) one has

 
n 
D f = D𝒮mod νj Ej f in Ω̊. (1.6.34)
j=1

Proof The identity claimed in (1.6.31) is seen from (1.5.50), (1.5.51), (A.0.65),
(A.0.188), and (1.6.30). In turn, formula (1.6.34) is implied by (1.6.33), (A.0.188),
and (1.6.31). 
It is natural to augment the picture emerging from Proposition 1.6.1 by discussing
separately the classical Cauchy integral operator on UR sets in the complex plane,
as the two-dimensional context accounts for a number of specialized properties
(compared with the higher-dimensional setting considered earlier).
Proposition 1.6.7 Suppose Ω ⊆ C ≡ R2 is an open set with the property that ∂Ω
is a UR set. Abbreviate σ∗ := H 1 ∂∗ Ω and σ := H 1 ∂Ω, then denote by ν the
geometric measure theoretic outward unit normal to Ω. In this setting, consider the
following Cauchy-type integral operators. 
σ∗ (ζ )
First, for each complex-valued function f ∈ L 1 ∂∗ Ω, 1+ |ζ | define (with dζ as in
(A.0.62)) ∫
1 f (ζ)
𝒞 f (z) := dζ for each z ∈ Ω, (1.6.35)
2πi ∂∗ Ω ζ − z
and its principal-value version
∫
1 f (ζ)
C f (z) := lim+ dζ at σ-a.e. z ∈ ∂Ω. (1.6.36)
ε→0 2πi ζ−z
ζ ∈∂∗ Ω
|z−ζ |>ε

19 with 𝒮mod acting on f componentwise

1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 145

σ(ζ )
Second, for each complex-valued function f ∈ L 1 ∂Ω, 1+ |ζ | set
# ∫ $
1 f (ζ)
C # f (z) := −ν(z) lim+ dσ(ζ) at σ∗ -a.e. z ∈ ∂∗ Ω, (1.6.37)
ε→0 2π ζ−z
ζ ∈∂Ω
|z−ζ |>ε

and make the convention that C # also acts in the same fashion on functions from
σ∗ (ζ )
L 1 ∂∗ Ω, 1+ |ζ | , regarded them as defined on the entire topological boundary ∂Ω
after being extended by zero outside ∂∗ Ω. In particular, this convention (1.6.36),
(1.6.36), and (A.0.62) imply (with ν denoting the complex conjugate of ν) that

C # f = −ν C(ν f ) at σ∗ -a.e. point in ∂∗ Ω, for each

σ∗ (ζ ) (1.6.38)
complex-valued function f from L 1 ∂∗ Ω, 1+ |ζ | .

Recall the Cauchy-Riemann operator together with its conjugate,

   
∂z̄ := 12 ∂z1 − 1i ∂z2 and ∂z := 12 ∂z1 + 1i ∂z2
(1.6.39)
where z = z1 + iz2 ∈ R + iR = C,

and consider the tangential derivative operator

∂τ := ∂τ12 (1.6.40)

where the latter is defined as in (A.0.183) with j := 1 and k = 2, i.e., abbreviate

∂τ := ν1 ∂ζ2 − ν2 ∂ζ1 where ν = ν1 + iν2 and ζ = ζ1 + iζ2 . (1.6.41)

Finally, fix an aperture parameter κ ∈ (0, ∞).

Then the following statements are true.

σ∗ (ζ )
(i) For each f ∈ L 1 ∂∗ Ω, 1+ |ζ | the function 𝒞 f is smooth in Ω and satisfies

∂z̄ (𝒞 f ) = 0 at each point belonging to the set Ω, and

κ−n.t. (1.6.42)
𝒞 f ∂Ω = ( 12 I + C) f at σ∗ -a.e. point in ∂∗ Ω.

Also, for each p ∈ [1, ∞) there exists C ∈ (0, ∞) depending only on ∂Ω, p, and
κ such that for each function f ∈ L p (∂∗ Ω, σ∗ ) one has
 
Nκ (𝒞 f ) p ≤ C f  L p (∂∗ Ω,σ∗ ) if p > 1, (1.6.43)
L (∂Ω,σ)
 
Nκ (𝒞 f ) ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (1.6.44)
L 1,∞ (∂Ω,σ)

σ∗ (ζ )
Finally, for each f ∈ L 1 ∂∗ Ω, 1+ |ζ | one has
146 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

2∂z 𝒮mod f = −𝒞(ν f ) in Ω, (1.6.45)

where 𝒮mod is the modified harmonic single layer potential operator associated
with Ω as in (1.5.50).
p
(ii) For each function f ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) one has

∂z (𝒞 f ) = (−i)𝒞(ν ∂τ f ) at each point in Ω. (1.6.46)

p
As a consequence, for each function f ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) the
κ−n.t.

pointwise nontangential boundary trace ∂z (𝒞 f ) exists (in C) at σ∗ -a.e.

∂Ω
point on ∂∗ Ω, and there exists some finite constant C > 0 depending only on
∂Ω, p, and κ such that
 
Nκ (∇𝒞 f ) p ≤ C∂τ f  L p (∂∗ Ω,σ∗ ) if p > 1, (1.6.47)
L (∂Ω,σ)
 
Nκ (∇𝒞 f ) ≤ C∂τ f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (1.6.48)
L 1,∞ (∂Ω,σ)

(iii) Suppose p, p ∈ (1, ∞) are such that 1/p + 1/p = 1. Then the operators

C : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ∗ ), (1.6.49)

C : L 1 (∂Ω, σ) −→ L 1,∞ (∂∗ Ω, σ∗ ), (1.6.50)

are well-defined, linear, and bounded. In addition,

the transpose of C in (1.6.49) is C  = C # , (1.6.51)

where C # is the linear and bounded mapping

 
C # : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ). (1.6.52)

Finally, corresponding to p = 1, the operator

C # : L 1 (∂∗ Ω, σ∗ ) −→ L 1,∞ (∂Ω, σ) (1.6.53)

is also well-defined, linear, and bounded.

(iv) Make the additional assumption that
 
H 1 ∂nta Ω \ ∂∗ Ω = 0 (1.6.54)

(which is automatically satisfied if Ω is a UR domain to begin with). Then the

operator
p p
C : L1 (∂∗ Ω, σ∗ ) −→ L1 (∂∗ Ω, σ∗ ) (1.6.55)

is well defined, linear, and bounded for each p ∈ (1, ∞).

1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 147

(v) Retain the additional assumption made in (1.6.54). Then for each p ∈ (1, ∞) it
follows that C # in (1.6.52) extends uniquely to a linear and bounded operator
p
from the negative boundary Sobolev space L−1 (∂∗ Ω, σ∗ ) into itself. Furthermore,
if one retains the same notation C # for this extension, the transpose of (1.6.55)
is
p p
C # : L−1 (∂∗ Ω, σ∗ ) −→ L−1 (∂∗ Ω, σ∗ ) (1.6.56)

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

(vi) Once again, retain the additional assumption made in (1.6.54). Then for each
p ∈ (1, ∞) one has
p
∂τ ◦ C + C # ◦ ∂τ = 0 on L1 (∂∗ Ω, σ∗ ). (1.6.57)

In particular, as a consequence of this and (1.6.38),

∂τ (C f ) = ν C(ν ∂τ f ) at σ∗ -a.e. point in ∂∗ Ω, for

p
each function f belonging to L1 (∂∗ Ω, σ∗ ). (1.6.58)


σ∗ (ζ )
Proof For each f ∈ L 1 ∂∗ Ω, 1+ |ζ | , it follows from (1.6.35) that the function 𝒞 f
is holomorphic in Ω. The jump-formula in (1.6.42) is a consequence of item (iv)
in Theorem 1.5.1 (in view of (1.4.72)-(1.4.73) in Example 1.4.9). The estimates in
(1.6.43)-(1.6.44) are direct consequence of item (i) in Theorem 1.5.1 (again, bearing
in mind the identification in (1.4.72)). 
σ∗ (ζ )
As far as (1.6.45) is concerned, for any f ∈ L 1 ∂∗ Ω, 1+ |ζ | (regarded as a function
defined on the entire topological boundary, extending it by zero on ∂Ω \ ∂∗ Ω we may
use (1.5.51) to write
∫
1 z−ζ
2∂z (𝒮mod f )(z) = f (ζ) dσ(ζ)
2π ∂∗ Ω |z − ζ | 2
∫
1 1
= f (ζ) dσ(ζ) = −𝒞(ν f ) in Ω. (1.6.59)
2π ∂∗ Ω z−ζ

This takes care of all claims in the current item (i).

To justify the first claim in item (ii), fix an arbitrary point z ∈ Ω. Since for each
ζ ∈ C \ {z} we have ∂ζ¯ ζ −z
1
= 0, at σ∗ -a.e. ζ ∈ ∂∗ Ω we may then write
 1   1   1 
ν(ζ)∂ζ = ν1 (ζ)∂ζ + iν2 (ζ)∂ζ
ζ−z ζ−z ζ−z
 
 1    1 
= ν1 (ζ) ∂ζ − ∂ζ¯ + iν2 (ζ) ∂ζ + ∂ζ¯
ζ−z ζ−z
 
 1 
= (−i) ν1 (ζ)∂ζ2 − ν2 (ζ)∂ζ1 . (1.6.60)
ζ−z
148 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
p
Consequently, given any f ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ [1, ∞), based on (1.6.35),
(1.6.60), (1.6.41), and (A.0.62) we may compute
∫  1 
1
∂z (𝒞 f )(z) = f (ζ)ν(ζ)∂z dσ∗ (ζ)
2π ∂∗ Ω ζ−z
∫  1 
1
=− f (ζ)ν(ζ)∂ζ dσ∗ (ζ)
2π ∂∗ Ω ζ−z
∫  1 
i
= f (ζ)∂τ(ζ ) dσ∗ (ζ)
2π ∂∗ Ω ζ−z
∫
1 1
= (∂τ f )(ζ) dσ∗ (ζ)
2πi ∂∗ Ω ζ−z

= (−i)𝒞(ν ∂τ f )(z) for each z ∈ Ω, (1.6.61)

where the fourth equality uses the integration by parts formula on the boundary from
[69, Lemma 11.1.7]. This finishes the proof of (1.6.46). With this in hand, all other
claims in item (ii) follow with the help of what we have proved already in item (i).
Moving on to item (iii), the fact that the operators (1.6.49), (1.6.50), (1.6.52),
(1.6.53) are all well-defined, linear, and bounded is a consequence of [70, Theo-
rem 2.3.2] and (A.0.62). To prove the claim made in (1.6.51), for each integrability
exponent p ∈ (1, ∞), consider the extension by zero operator

E : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ) given by


f on ∂∗ Ω, (1.6.62)
E f := for each f ∈ L p (∂∗ Ω, σ∗ ),
0 on ∂Ω \ ∂∗ Ω,

along with the operator of pointwise multiplication by the normal

M : L p (∂Ω, σ) → L p (∂∗ Ω, σ∗ ),
(1.6.63)
M f := ν f for each f ∈ L p (∂Ω, σ).

Clearly, these are both linear and bounded operators. Also, a moment’s reflection
shows that
the (real) transpose of the operator M in (1.6.63) is the com-
 
position E ◦ M ◦ E : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ), where (1.6.64)
p ∈ (1, ∞) is the Hölder conjugate exponent of p.
Next, bring in the principal-value singular integral operator

T : L p (∂Ω, σ) → L p (∂Ω, σ) (1.6.65)

acting on each f ∈ L p (∂Ω, σ) according to

1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 149
∫
1 f (ζ)
T f (z) := lim+ dσ(ζ) at σ-a.e. z ∈ ∂Ω. (1.6.66)
ε→0 2π ζ−z
ζ ∈∂Ω
|z−ζ |>ε

From [70, Theorem 2.3.2] (cf. [70, (2.3.15), (2.3.18)]) we know that T is well defined,
linear, and bounded. Moreover, [70, (2.3.25)] implies that
the (real) transpose of the singular integral operator (1.6.65) is
 
the operator −T : L p (∂Ω, σ) −→ L p (∂Ω, σ) where, once again, (1.6.67)

p ∈ (1, ∞) denotes the Hölder conjugate exponent of p.
Upon noting that the Cauchy singular integral operator defined in (1.6.36) in the
context C : L p (∂∗ Ω, σ∗ ) → L p (∂Ω, σ) may be expressed as C = T ◦ E ◦ M ◦ E, we
may invoke (1.6.64) and (1.6.67) to conclude that its (real) transpose is the operator
 
C  : L p (∂∗ Ω, σ∗ ) → L p (∂Ω, σ), given by C  = −M ◦ T = C # . This establishes
(1.6.51) and finishes the treatment of item (iii).
Moving on, the claims in the current items (iv)-(v) are direct consequences of
items (v)-(vi) in [70, Theorem 2.3.2] (in view of the identification made in (1.4.73)
in Example 1.4.9). At this stage, there remains to deal with item (vi). To this end, fix
p
some f ∈ L1 (∂∗ Ω, σ∗ ) with p ∈ [1, ∞). From [69, Proposition 11.3.2] applied to the
κ−n.t.
function u := 𝒞 f in Ω we see that the nontangential boundary trace u ∂Ω belongs
p
to the boundary Sobolev space L1 (∂∗ Ω, σ∗ ) and
 κ−n.t.  κ−n.t.  κ−n.t.
∂τ u ∂Ω = ν1 (∂z2 u) ∂Ω − ν2 (∂z1 u) ∂Ω at σ∗ -a.e. point on ∂∗ Ω. (1.6.68)

On the other hand,

 making  use of the fact that the function
 u is holomorphic in Ω,
we have ∂z2 u = i ∂z − ∂z̄ u = i∂z u and ∂z1 u = ∂z + ∂z̄ u = ∂z u. When used back in
(1.6.68) this permits us to conclude that
 κ−n.t.  κ−n.t.
∂τ u ∂Ω = iν (∂z u) ∂Ω at σ∗ -a.e. point on ∂∗ Ω. (1.6.69)

In turn, by combining the jump-formula in (1.6.42) with (1.6.69), (1.6.46), and

(1.6.38), we arrive at
 κ−n.t.  κ−n.t.
1
∂
2 τ f + ∂ τ (C f ) = ∂τ u ∂Ω
= iν (∂z u) ∂Ω
  κ−n.t.
= ν 𝒞(ν ∂τ f ) ∂Ω
= 12 ∂τ f + ν C(ν ∂τ f )

= 12 ∂τ f − C # (∂τ f ) at σ∗ -a.e. point on ∂∗ Ω. (1.6.70)

From this, (1.6.57) readily follows, completing the proof of Proposition 1.6.7. 
A remarkable intertwining identity in the two-dimensional setting, whose proof
involves Proposition 1.6.7, is discussed below (see also Propositions 2.3.13-2.3.14
in this regard).
150 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Corollary 1.6.8 Let Ω ⊆ R2 be a UR domain. Set σ := H 1 ∂Ω and denote by ν

the geometric measure theoretic outward unit normal to Ω. In this setting, recall the
two-dimensional the principal-value harmonic double layer acting on each function
 σ(x) 
f ∈ L 1 ∂Ω, 1+ |x | according to (cf. [70, (2.5.203)])
∫
1 ν(y), y − x
K f (x) = lim+ f (y) dσ(y) for σ-a.e. x ∈ ∂Ω, (1.6.71)
ε→0 2π |x − y| 2
y ∈∂Ω
|x−y |>ε

along with the two-dimensional “transpose” principal-value harmonic double layer

(cf. (1.6.18))
∫
1 ν(x), x − y
K f (x) = lim+
#
f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (1.6.72)
ε→0 2π |x − y| 2
y ∈∂Ω
|x−y |>ε

Finally, fix an integrability exponent p ∈ (1, ∞).

Then, with the tangential derivative operator

∂τ := ∂τ12 (1.6.73)

defined as in (A.0.183) with j := 1 and k := 2, one has

p
∂τ ◦ K = −K # ◦ ∂τ on L1 (∂Ω, σ). (1.6.74)

Also, with ∂ν D denoting the normal derivative of the harmonic double layer (cf.
(1.5.29)) and with the principal-value singular integral operator R defined for each
 σ(x) 
function f ∈ L 1 ∂Ω , 1+ |x | and σ-a.e. point x ∈ ∂Ω as
∫  
R f (x) := lim+ ν1 (y)(∂2 EΔ )(y − x) − ν2 (y)(∂1 EΔ )(y − x) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(1.6.75)

where EΔ is the standard fundamental solution for the two-dimensional Laplacian,

one has
p
∂ν D = −∂τ ◦ R on L1 (∂Ω, σ). (1.6.76)

An inspection of the proof given below shows that

formulas (1.6.74), (1.6.76) are actually true for the more inclusive
p,q
off-diagonal Sobolev space L1 (∂Ω, σ) with p, q ∈ (1, ∞) arbitrary. (1.6.77)

Under stronger assumptions on the set Ω, formulas (1.6.74), (1.6.76) remain valid on
yet even larger spaces. Specifically, via density (cf. [69, (11.8.4)]) and [69, Propo-
1.6 Cauchy and Cauchy-Clifford Operators on Lebesgue and Sobolev Spaces 151

sition 11.8.5], if Ω ⊆ R2 is an open set satisfying a two-sided local John condition

and whose boundary is compact and Ahlfors regular then the operator identities in
(1.6.74)-(1.6.76) are actually valid in the (larger) Lebesgue space L p (∂Ω, σ).
Proof of Corollary 1.6.8 Recall the singular integral operator R introduced in
 σ(x) 
(1.6.75). For each function f ∈ L 1 ∂Ω, 1+ |x | let us also define
∫  
R f (x) := lim+
#
ν1 (x)(∂2 EΔ )(y − x) − ν2 (x)(∂1 EΔ )(y − x) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(1.6.78)

at σ-a.e. point x ∈ ∂Ω, where (ν1, ν2 ) are the scalar components of the normal vector
ν and EΔ is the standard fundamental solution for the (two-dimensional) Laplacian
(cf. (1.1.1)). Then from (1.6.36), (1.6.71), and (1.6.75), on the one hand, and from
(1.6.37), (1.6.72), and (1.6.78), on the other hand, we see that

C = K − iR and C # = K # + iR# on L p (∂Ω, σ). (1.6.79)

p
Moreover, from (1.2.13) we conclude that for each f ∈ L1 (∂Ω, σ) we have

∂τ (R f ) = R# (∂τ f ) at σ-a.e. point on ∂Ω. (1.6.80)

Then (1.6.74) follows by combining (1.6.57), (1.6.79), and (1.6.80).

Let us now turn our attention to (1.6.76). Given any f ∈ L p (∂Ω, σ), at each x ∈ Ω
define
∫  
R f (x) := ν1 (y)(∂2 EΔ )(y − x) − ν2 (y)(∂1 EΔ )(y − x) f (y) dσ(y). (1.6.81)
∂Ω

Fix an aperture parameter κ > 0. From Proposition 1.2.1 it follows that

κ−n.t.
Nκ (R f ) ∈ L p (∂Ω, σ) and R f ∂Ω
= R f for each f ∈ L p (∂Ω, σ),
(1.6.82)
p
Nκ (∇R f ) ∈ L p (∂Ω, σ) for each f ∈ L1 (∂Ω, σ).

Also, since for each f ∈ L p (∂Ω, σ) the function 𝒞 f is holomorphic in Ω and

𝒞 f = D f − iR f in Ω, (1.6.83)

it follows that D f and −R f are conjugate harmonic functions in Ω. That is, for each
f ∈ L p (∂Ω, σ) we have

∂1 D f + ∂2 R f = 0 and ∂2 D f − ∂1 R f = 0 in Ω. (1.6.84)
p
As a consequence of (1.6.84) and [69, Proposition 11.3.2], for each f ∈ L1 (∂Ω, σ)
we may then write
152 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 κ−n.t.   κ−n.t.   κ−n.t. 
∂ν D f = ν · ∇(D f ) ∂Ω = ν1 ∂1 (D f ) ∂Ω + ν2 ∂2 (D f ) ∂Ω
 κ−n.t.   κ−n.t. 
= −ν1 ∂2 (R f ) ∂Ω + ν2 ∂1 (R f ) ∂Ω
 κ−n.t. 
= −∂τ (R f ) ∂Ω = −∂τ (R f ), (1.6.85)

proving (1.6.76). 
It is also of interest to augment the spectral results from Proposition 1.5.3 with
the following symmetry property (with respect to the origin) for the point-spectra of
the principal-value harmonic double layer acting on Lebesgue and Sobolev spaces
on the boundary of two-dimensional UR domains.
Corollary 1.6.9 Let Ω ⊆ R2 be an arbitrary UR domain and set σ := H 1 ∂Ω.
In this setting, recall the principal-value harmonic double layer K and its (real)
transpose K # from (1.6.71)-(1.6.72). Then for each exponent p ∈ (1, ∞) and each
complex number λ ∈ C \ {± 12 } the following statements are true.
p
(1) If λ is an eigenvalue of K acting on L p (∂Ω, σ) (respectively, on L1 (∂Ω, σ)) then
p
−λ is an eigenvalue of K acting on L p (∂Ω, σ) (respectively, on L1 (∂Ω, σ)).
(2) If λ is an eigenvalue of K # acting on the space L p (∂Ω, σ) (respectively, on
p
L−1 (∂Ω, σ)) then −λ is an eigenvalue of K # acting on the space L p (∂Ω, σ)
p
(respectively, on L−1 (∂Ω, σ)).
Proof Suppose λ ∈ C \ {± 12 } is an eigenvalue of K acting on L p (∂Ω, σ). Hence,
there exists f ∈ L p (∂Ω, σ), which is not zero σ-a.e. on ∂Ω, with the property that
K f = λ f . Then the function g := R f (where R is the singular integral operator
introduced in (1.6.75)) belongs to the Lebesgue space L p (∂Ω, σ) and, thanks to the
first equality in [70, (2.5.334)], satisfies

Kg = K(R f ) = −R(K f ) = −R(λ f ) = −λR f = −λg. (1.6.86)

As such, we may conclude that −λ is an eigenvalue of K acting on L p (∂Ω, σ) as

soon as we check that the function g is not zero σ-a.e. on ∂Ω. To see that this is the
case, we make use of the second equality in [70, (2.5.334)] to write
    
Rg = R2 f = 12 I + K − 12 I + K f = λ2 − 14 f  0, (1.6.87)

given that λ2  14 and f  0. This goes to show that, indeed, g  0. The above
p
argument also works for K acting on the Sobolev space L1 (∂Ω, σ) since the operator
p
identities in [70, (2.5.334)] are valid L1 (∂Ω, σ) as well (as may be seen from (1.5.24)
and (1.2.11)). The treatment of item (1) is therefore complete.
Finally, the claims in item (2) may be handled in an analogous fashion, this time
making use of the operator identities (with R# originally defined as in (1.6.78))
1    2
K # R# + R# K # = 0 and 2I + K# − 12 I + K # = R#
p
(1.6.88)
both on L p (∂Ω, σ) and on L−1 (∂Ω, σ),
1.7 Kernels and Images of Boundary Layer Potentials 153

which follow from [70, (2.5.334)] and transposition (bearing in mind item (vi) in
Theorem 1.5.1, and item (iv) in Proposition 1.2.1). 

For future endeavors, we find it useful to derive a version of the last formula from
(1.6.88) featuring the composition between the tangential derivative operator and the
modified harmonic single layer operator. This is made precise in the lemma below.

Lemma 1.6.10 Let Ω ⊆ R2 be an arbitrary UR domain and set σ := H 1 ∂Ω. In

this setting, recall the transpose double layer K # from (1.6.72) and bring in the
two-dimensional modified boundary-to-boundary single layer operator (cf. (1.5.73))
∫
 
Smod f (x) := E(x − y) − E∗ (−y) f (y) dσ(y) at σ-a.e. x ∈ ∂Ω,
∂Ω (1.6.89)
 σ(x)  −1 ln ·1
for each f ∈ L 1 ∂Ω, 1+ |x | , where E∗ := (2π) R \B(0,1)
2 .

Finally, consider the tangential derivative operator

∂τ := ∂τ12 (1.6.90)

defined as in (A.0.183) with j := 1 and k := 2, and fix an integrability exponent

p ∈ (1, ∞). Then, with I denoting the identity,
1    2
2I + K# − 12 I + K # = ∂τ Smod
(1.6.91)
as operators on L p (∂Ω, σ).

Proof From (1.6.78) and (1.5.82) we see that R# coincides with T12 # , the singular

integral operator defined as in (1.5.82) corresponding to j = 1, k = 2, M = 1, n = 2,

L = Δ, and E := (2π)−1 ln. Bearing this in mind, we then see from (1.5.83) that
 
∂τ Smod f = R# f for each f ∈ L p (∂Ω, σ). (1.6.92)

With this in hand, the operator identity claimed in (1.6.91) follows from (1.6.88). 

1.7 Kernels and Images of Boundary Layer Potentials

With an eye toward the goal of eventually inverting boundary layer potentials (some-
thing that is relevant in the context of solving boundary value problems, a topic
treated later on, in Volume IV; cf. [71]), one of the main goals in this section is to
study the kernels and images of singular integral operators of boundary layer type.
To set the stage, the reader is reminded
 αβ  (cf. [70, Definition 1.3.2]) that a coefficient
tensor with complex entries A = ar s 1≤r,s ≤n is said to be Legendre-Hadamard
1≤α,β ≤M
elliptic provided there exists a real number κ > 0 such that the following condition
is satisfied:
154 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
αβ
Re ar s ξr ξs ηα ηβ ≥ κ|ξ | 2 |η| 2 for all
(1.7.1)
ξ = (ξr )1≤r ≤n ∈ Rn and η = (ηα )1≤α ≤M ∈ C M .
A basic feature of a Legendre-Hadamard elliptic coefficient tensor A ∈ A(n, M),
which may be easily checked based on repeated applications of Plancherel’s theorem
and (1.7.1), is the fact that there exists a constant c = c(A, n) ∈ (0, ∞) with the
property that
∫ ∫
 M
Re A∇u, ∇u dL n ≥ c |∇u| 2 dL n, ∀ u ∈ W 1,2 (Rn ) . (1.7.2)
Rn Rn

Next, we make the following definition.

 αβ 
Definition 1.7.1 Fix n, M ∈ N, and consider a coefficient tensor A = ar s 1≤r,s ≤n
1≤α,β ≤M
with constant complex entries. Call A compliant provided A is positive semi-
definite, in the sense that
 
αβ β
Re Aζ, ζ = Re ar s ζs ζrα ≥ 0, ∀ζ = (ζrα ) 1≤r ≤n ∈ Cn×M , (1.7.3)
1≤α ≤M

and A has the property that for each ζ = (ζrα ) 1≤r ≤n ∈ Cn×M one has
1≤α ≤M

Re Aζ, ζ = 0 ⇐⇒ Aζ = 0. (1.7.4)

In relation to Definition 1.7.1, it is worth noting that

any weakly elliptic compliant coefficient tensor

(1.7.5)
is Legendre-Hadamard elliptic.
 αβ 
To justify (1.7.5), suppose A = ar s 1≤r,s ≤n is a weakly elliptic compliant coeffi-
1≤α,β ≤M
cient tensor, and fix an arbitrary vector ξ = (ξr )1≤r ≤n ∈ Rn \ {0}. The fact that A is
assumed to be weakly elliptic implies that
 αβ 
the M × M matrix ar s ξr ξs 1≤α,β ≤M is invertible. (1.7.6)

Suppose now η = (ηα )1≤α ≤M ∈ C M is such that

αβ
Re ar s ξr ξs ηα ηβ = 0. (1.7.7)

If we define
ζ := (ξr ηα ) 1≤r ≤n ∈ Cn×M (1.7.8)
1≤α ≤M

then formula (1.7.7) may be recast simply as Re Aζ, ζ = 0. In view of (1.7.4), this
 αβ   αβ 
forces 0 = Aζ = ar s ξs ηβ 1≤r ≤n , hence ar s ξr ξs ηβ 1≤α ≤M = 0 ∈ C M . Bearing
1≤α ≤M
in mind (1.7.6), this allows us to conclude that η = 0 ∈ C M . Thus, (1.7.7) can only
1.7 Kernels and Images of Boundary Layer Potentials 155

happen when η = 0 ∈ C M . The fact that A is positive semi-definite also implies

(writing (1.7.3) for ζ as in (1.7.8)) that
αβ
Re ar s ξr ξs ηα ηβ ≥ 0 for each η = (ηα )1≤α ≤M ∈ C M . (1.7.9)

In concert with the homogeneity of the bilinear form associated with A, these
properties ultimately show that there exists some c ∈ (0, ∞) with the property that
(1.7.1) holds. This finishes the proof of (1.7.5).
In the theorem below we establish injectivity properties for the boundary-to-
boundary versions of the single and double layer potential operators.

Theorem 1.7.2 Fix n ∈ N with n ≥ 2. Suppose Ω ⊆ Rn is an arbitrary UR domain

and abbreviate σ := H n−1 ∂Ω. Let A be a weakly elliptic coefficient tensor with
complex entries, and consider the boundary layer potentials S, K, K # associated
with A and Ω as in (1.3.62), (1.3.68), (1.3.72). Then the following properties hold.
(1) Suppose A is actually Legendre-Hadamard elliptic (cf. (1.7.1)). If n ≥ 3 the
boundary-to-boundary single layer operator S is injective in the context
M M
S : L 2(n−1)/n (∂Ω, σ) −→ L 2(n−1)/(n−2) (∂Ω, σ) (1.7.10)

(cf. (1.3.63)), while if n = 2 and ∂Ω is compact then

∫
M
if f ∈ L p (∂Ω, σ) with p ∈ (1, ∞) and f dσ = 0 then
∂Ω (1.7.11)
S f constant on ∂Ω forces f to vanish on ∂Ω.

(2) Suppose A is positive semi-definite and Legendre-Hadamard elliptic and fix a

complex number z ∈ C \ [− 12 , 12 ]. In addition, assume that either A is complex
symmetric20, or z is real. Then
M M
zI + K # : L 2(n−1)/n (∂Ω, σ) −→ L 2(n−1)/n (∂Ω, σ)
(1.7.12)
is an injective operator if n ≥ 3,

and the operator

M M
zI + K # : L p (∂Ω, σ) −→ L p (∂Ω, σ) is injective
(1.7.13)
for every p ∈ (1, ∞) if ∂Ω is compact and n = 2.

(3) Given p, p ∈ (1, ∞) with 1/p + 1/p = 1 along with z ∈ C, consider the space
 ∫
p M
L A,z (∂Ω, σ) := f ∈ L p (∂Ω, σ) : f , g dσ = 0 for each
∂Ω

p M
g ∈ L (∂Ω, σ) with (zI + K A)g = 0 . (1.7.14)

20 i.e., Hermitian self-adjoint

156 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Then for each p ∈ (1, ∞) and z ∈ C one has


M p
Im zI + K # ; L p (∂Ω, σ) = L A,z (∂Ω, σ), (1.7.15)

M
where the “bar” in the left-hand side indicates closure taken in L p (∂Ω, σ) .
Moreover,
M
± 12 I + K # : L 2(n−1)/n
A∗,±1/2
(∂Ω, σ) −→ L 2(n−1)/n (∂Ω, σ) injectively
(1.7.16)
if n ≥ 3 and A is assumed to be compliant (cf. Definition 1.7.1),

which, in concert with (1.7.15), shows that

± 12 I + K # : L 2(n−1)/n
A,±1/2
(∂Ω, σ) −→ L 2(n−1)/n
A,±1/2
(∂Ω, σ) injectively if n ≥ 3,
A is also assumed to be complex symmetric, and A is compliant.
(1.7.17)

(4) Assume A is positive definite, Ω is bounded, and Rn \ Ω is connected. Then the

operator
M M
1
2I + K # : L p (∂Ω, σ) −→ L p (∂Ω, σ) is injective
(1.7.18)
if either n ≥ 3 and 2(n − 1)/n ≤ p < ∞, or n = 2 and 1 < p < ∞.

(5) Suppose the coefficient tensor A is complex symmetric and positive definite.
Also, strengthen the hypotheses on the underlying domain by assuming that Ω
is a bounded UR domain satisfying a two-sided local John condition and such
that Rn \ Ω is connected. Then for each z ∈ C \ [− 12 , 12 ] it follows that
M M
zI + K : L12(n−1)/n (∂Ω, σ) −→ L12(n−1)/n (∂Ω, σ)
(1.7.19)
is an injective operator if n ≥ 3,

and the operator

p M p M
zI + K : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) is injective
(1.7.20)
for every p ∈ (1, ∞) if n = 2.

(6) Assume the coefficient tensor A is positive definite. Also, suppose Ω is a bounded
UR domain satisfying a two-sided local John condition and such that Rn \ Ω is
connected. Then
M M
1
2I + K : L12(n−1)/n (∂Ω, σ) −→ L12(n−1)/n (∂Ω, σ)
(1.7.21)
is an injective operator if n ≥ 3,

and the operator

1.7 Kernels and Images of Boundary Layer Potentials 157

p M p M
1
2I + K : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) is injective
(1.7.22)
for every p ∈ (1, ∞) if n = 2.

Proof Throughout, we denote by ν = (ν1, . . . , νn ) the geometric measure theoretic

outward unit normal to Ω, and define Ω+ := Ω as well as Ω− := Rn \ Ω. [68, Lem-
ma 5.10.9] then ensures that Ω− is also a UR domain, whose topological boundary
coincides with that of Ω, whose geometric measure theoretic boundary agrees with
that of Ω, and whose geometric measure theoretic
 αβ  outward unit normal is the vector
−ν at σ-a.e. point on ∂Ω. Also, we shall let ar s 1≤r,s ≤n be the entries of the coef-
1≤α,β ≤M
ficient tensor A, and denote by E = (Eγβ )1≤γ,β ≤M be the matrix-valued fundamental
solution associated with the system
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.7.23)

(i.e., L := L A = divA∇) as in [70, Theorem 1.4.2]. Finally, pick an aperture parameter

κ ∈ (0, ∞).
M
Proof of (1): First, assume that n ≥ 3 and pick a function f ∈ L 2(n−1)/n (∂Ω, σ)
satisfying S f = 0 on ∂Ω. If we define
∫
u(x) := E(x − y) f (y) dσ(y) for each x ∈ Rn \ ∂Ω, (1.7.24)
∂Ω

then [70, Proposition 2.5.40] implies that, when regarded as being defined L n -a.e.
in Rn , the function u satisfies
M nM
u ∈ L 2n/(n−1) (Rn, L n ) , ∇u ∈ L 2 (Rn, L n ) ,
(1.7.25)
1,2 n M
and u ∈ Wloc (R ) .

To proceed, introduce
u± := u Ω± . (1.7.26)
From item (x) in Theorem 1.5.1 we know that

u± ∈ 𝒞∞ (Ω± ) L Au± = 0 in Ω±,

M
, (1.7.27)

while (1.7.24), (1.5.47), (1.5.47), (1.3.67), (1.5.59), and (1.5.48) yield

Nκ u± ∈ L 2(n−1)/(n−2) (∂Ω, σ), Nκ (∇u± ) ∈ L 2(n−1)/n (∂Ω, σ),

κ−n.t.  
u± ∂Ω = S f , ∂νAu± = ∓ 12 I + K A#  f , (1.7.28)
κ−n.t.
and ∇u± ∂Ω exist at σ-a.e. point on ∂Ω.

In addition, thanks to (1.7.24) and [70, Theorem 1.4.2], it follows that for each k ∈ N0
we have
158 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

(∇k u)(x) = O(|x| 2−n−k ) as |x| → ∞ if ∂Ω compact. (1.7.29)

Granted the properties listed in (1.7.27)-(1.7.29), we may invoke [70, Corol-

lary 1.7.14] (presently used with p := 2(n − 1)/n, q := 2(n − 1)/(n − 2), a := n − 1,
b := n − 2) and, bearing in mind that the outward unit normal for Ω− is −ν, conclude
that
∫ ∫
  1  
A∇u±, ∇u± dL n = − 2 I ± K A#  f , S f dσ = 0 (1.7.30)
Ω± ∂Ω

since, by assumption, S f = 0 on ∂Ω. Collectively, from (1.7.25)-(1.7.30) we see that

∫

A∇u, ∇u ∈ L (R , L ) and
1 n n
A∇u, ∇u dL n = 0. (1.7.31)
Rn

To proceed, pick a real-valued function ϕ ∈ 𝒞∞

c (R ) which is identically one in
n

a neighborhood of the origin. If for each j ∈ N we set ϕ j (x) := ϕ(x/ j) for every
x ∈ Rn , then
∫

lim A∇(ϕ j u),∇(ϕ j u) dL n (1.7.32)
j→∞ Rn
∫

= lim ϕ2j A∇u, ∇u dL n
j→∞ Rn
∫
 
+ lim O |ϕ j ||∇ϕ j ||u||∇u| + |∇ϕ j | 2 |u| 2 dL n .
j→∞ Rn

Note that since |x| ≈ j on the support of ∇ϕ j , uniformly in j ∈ N, based on (1.7.25)

and (the trilinear version of) Hölder’s inequality we may estimate
∫ ∫
|ϕ j ||∇ϕ j ||u||∇u| dL n ≤ C j −1 |u||∇u| dL n
Rn
x ∈R n
|x |≈j

≤ C j −1/2 u[L 2n/(n−1) (Rn, L n )] M ∇u[L 2 (Rn, L n )]n M

= O( j −1/2 ) as j → ∞, (1.7.33)

and
∫ ∫
−2
|∇ϕ j | |u| dL ≤ C j
2 2 n
|u| 2 dL n
Rn
x ∈R n
|x |≈j

≤ C j −1 u[L
2
2n/(n−1) (R n , L n )] M

= O( j −1 ) as j → ∞. (1.7.34)
1.7 Kernels and Images of Boundary Layer Potentials 159

Thanks to (1.7.31), (1.7.33), (1.7.34), and Lebesgue’s Dominated Convergence The-

orem we may therefore conclude from (1.7.32) that
∫

lim A∇(ϕ j u), ∇(ϕ j u) dL n = 0. (1.7.35)
j→∞ Rn

M
Since the last membership in (1.7.25) ensures that ϕ j u ∈ W 1,2 (Rn ) for each
j ∈ N, the coercivity estimate (1.7.2) then gives, for some c ∈ (0, ∞),
1 ∫ 2

0 = lim Re A∇(ϕ j u), ∇(ϕ j u) dL n
j→∞ Rn
∫
≥ c lim |∇(ϕ j u)| 2 dL n
j→∞ Rn
∫
= c lim ϕ2j |∇u| 2 dL n
j→∞ Rn
∫
 
+ lim O |ϕ j ||∇ϕ j ||u||∇u| + |∇ϕ j | 2 |u| 2 dL n
j→∞ Rn
∫
=c |∇u| 2 dL n, (1.7.36)
Rn

where the last equality makes use of (1.7.33), (1.7.34), (1.7.31), and Lebesgue’s
Dominated Convergence Theorem. This ultimately proves that

∇u = 0 in Rn . (1.7.37)

In concert, the jump-formula (1.5.59), (A.0.184), and (1.7.37) establish

 1 
∓ 2 I + K A#  f = ∂νAu± = 0 at σ-a.e. point on ∂Ω. (1.7.38)

From (1.7.37) we finally conclude that

f = ∂νAu− − ∂νAu+ = 0, (1.7.39)

which goes to show that the boundary-to-boundary single layer operator S is injective
in the context of (1.7.10) if n ≥ 3.
Let us now consider the case when n = 2 and ∂Ω is compact. To fix ideas, assume
Ω+ is bounded (and Ω− is an exterior domain). Fix p ∈ (1, ∞). The
∫ goal is to prove
M
(1.7.11). To this end, pick a function f ∈ L p (∂Ω, σ) with ∂Ω f dσ = 0 and
satisfying S f = c, a constant on ∂Ω. Define u : Rn \ ∂Ω → C M as in (1.7.24). In the
present case, the effect of the extra cancelation assumption on f is the (improved)
decay property

(∇k u)(x) = O(|x| −1−k ) as |x| → ∞, for each k ∈ N0 . (1.7.40)

160 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

See Lemma 1.3.1. Also, [70, Proposition 2.5.41] implies that, when regarded as
being defined L 2 -a.e. in R2 , the function u satisfies
 M
 2M
u∈ L q (R2, L 2 ) and ∇u ∈ L q (R2, L 2 ) . (1.7.41)
2<q<∞ 1<q ≤2p

As in the past, introduce u± := u Ω± so

u± ∈ 𝒞∞ (Ω± ) and L Au± = 0 in Ω± .

M
(1.7.42)

Also, from (1.7.24), (1.7.40), the fact that ∂Ω is compact, item (x) in Theorem 1.5.1,
(1.3.67), (1.5.59), (1.5.48), and assumptions yield
,
Nκ u± ∈ 0<q<∞ L q (∂Ω, σ), Nκ (∇u± ) ∈ L p (∂Ω, σ),
κ−n.t.  
u± ∂Ω = c, ∂νAu± = ∓ 12 I + K A#  f , (1.7.43)
κ−n.t.
and ∇u± ∂Ω exist at σ-a.e. point on ∂Ω.

Thanks to these properties, [70, Corollary 1.7.14] applies (for the current p, and with
a := 2, b := 1) and gives that, on the one hand,
∫ ∫
±
  1  
±
A∇u , ∇u dL = 2
− 2 I ± K A#  f , c dσ. (1.7.44)
Ω± ∂Ω

On the other hand, since Ω+ is bounded,

∫ ∫
 1     
− 2 I + K A#  f , c dσ = f , − 12 I + K A c dσ = 0 (1.7.45)
∂Ω ∂Ω

thanks to the duality result from item (iii) of Theorem 1.5.1 and (1.5.21) (written
with A in place of A). In a similar fashion,
∫ ∫ ∫
 1 # 
  1  
− 2 I − K A f , c dσ = f , − 2 I − K A c dσ =
 f , −c dσ
∂Ω ∂Ω ∂Ω
∫ 
=− f dσ, c = 0, (1.7.46)
∂Ω

with the last equality provided by the vanishing moment condition for f . Collectively,
(1.7.44)-(1.7.46) imply that
∫

A∇u±, ∇u± dL 2 = 0. (1.7.47)
Ω±

Together with (1.7.41), this permits us to re-run the same argument which, starting
with (1.7.31), has produced (1.7.39) to once again conclude that f = 0.
1.7 Kernels and Images of Boundary Layer Potentials 161

Proof of (2): First, suppose n ≥ 3. Work under the assumption that A is positive
semi-definite, complex symmetric, as well as Legendre-Hadamard elliptic, and that
M
z ∈ C \ [− 12 , 12 ] is arbitrary. Consider a function f ∈ L 2(n−1)/n (∂Ω, σ) such that
(zI + K # ) f = 0. If we define
∫
u(x) := E  (x − y) f (y) dσ(y) for each x ∈ Rn \ ∂Ω, (1.7.48)
∂Ω

then [70, Proposition 2.5.40] implies

M nM
u ∈ L 2n/(n−1) (Rn, L n ) , ∇u ∈ L 2 (Rn, L n ) ,
(1.7.49)
1,2 n M
and u ∈ Wloc (R ) .

Next, we invoke [70, Corollary 1.7.14] which, granted the current assumptions,
applies to u Ω± (and the system L A ). On account of (1.3.67) and (1.5.59) (presently
used with A replaced by A ) we may therefore write
∫

0= (zI + K # ) f , S A f dσ
∂Ω
∫  
= (−z + 12 )(− 12 I + K # ) f + (z + 12 )( 12 I + K # ) f , S A f dσ
∂Ω
∫ ∫
  
   
= −z+ 1
2 A ∇u, ∇u dL + − z −n 1
2 A ∇u, ∇u dL n . (1.7.50)
Ω+ Ω−

At this point, bring in the elementary fact that (easily justified by inspecting the real
and imaginary parts)
.
a± ∈ [0, ∞), z ∈ C \ − 12 , 12
    =⇒ a+ = a− = 0. (1.7.51)
− z + 12 a+ + − z − 12 a− = 0

Recall that we are now assuming that A is positive semi-definite and complex
symmetric. Thus, A is also complex symmetric, and A∗ is positive semi-definite.
Thanks to these properties and the second membership in (1.7.49) we have
∫

a± := A ∇u, ∇u dL n ∈ [0, ∞). (1.7.52)
Ω±

Collectively, (1.7.50), (1.7.51), and (1.7.52) then permit us to conclude that

∫ ∫

 
A ∇u, ∇u dL = 0 and
n
A ∇u, ∇u dL n = 0, (1.7.53)
Ω+ Ω−

hence, ultimately, ∫

A ∇u, ∇u dL n = 0. (1.7.54)
Rn
162 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Granted (1.7.54), the same reasoning based on the Legendre-Hadamard ellipticity

of A (which also entails that A is Legendre-Hadamard elliptic) which, starting with
(1.7.31) has produced (1.7.37), continues to work in the present setting (thanks
to (1.7.48)). Much as before, from this we deduce that ∇u = 0 in Rn . Denoting
u± := u Ω± and availing ourselves of the jump-formulas
  
∂νA u± = ∓ 12 I + K # f at σ-a.e. point on ∂Ω (1.7.55)

(themselves seen from (1.5.59) in Ω± and with A in place of A), we eventually arrive
 
at the conclusion that f = ∂νA u− − ∂νA u+ = 0. The argument so far proves that if
n ≥ 3 then the operator zI + K is injective in the context of (1.7.12) provided A is
#

positive semi-definite, complex symmetric, as well as Legendre-Hadamard elliptic,

and z ∈ C \ − 12 , 12 is arbitrary.
Continue to assume that n ≥ 3, but suppose now that z ∈ R \ − 12 , 12 while A
is only positive semi-definite and Legendre-Hadamard elliptic. This time, in lieu of
(1.7.51) we use
.
a± ∈ C, Re a± ∈ [0, ∞), z ∈ R \ − 12 , 12
    =⇒ Re a+ = Re a− = 0, (1.7.56)
and − z + 12 a+ + − z − 12 a− = 0

which is itself a consequence of (1.7.51). When applied to

∫

a± := A ∇u, ∇u dL n ∈ C, (1.7.57)
Ω±

in place of (1.7.54) we now arrive at

∫

Re A ∇u, ∇u dL n = 0. (1.7.58)
Rn

This is nonetheless just as effective in the context of (1.7.36), so we may complete

the proof of the injectivity of the operator zI + K # in the context of (1.7.12) much
as before.
As an intermediate step, we next propose to show that

if n ≥ 2, ∂Ω is compact, z ∈ C such that z  − 12 if Ω is bounded and

M
z  12 if Ω is unbounded, and f ∈ L p (∂Ω, ∫ σ) with p ∈ (1, ∞) is (1.7.59)
such that (zI + K ) f = 0, then necessarily ∂Ω f dσ = 0.
#

To see that this is the case, observe first that [68, Lemma 5.10.10] implies that either
Ω+ is bounded, or Ω− is bounded. To fix ideas, assume Ω+ is bounded. In such a
scenario, applying Green’s formula [70, (1.7.81)] (with A playing the role of A)
to the functions u := 𝒮 A f and w := λ in Ω, for some arbitrary constant λ ∈ C M ,
yields
1.7 Kernels and Images of Boundary Layer Potentials 163
3∫ 4 ∫
   κ−n.t. 
− 12 I + K # f dσ, λ = ∂νA u, w ∂Ω
dσ
∂Ω ∂Ω
∫
κ−n.t. 
= u ∂Ω , ∂νA w dσ = 0, (1.7.60)
∂Ω

since ∂νA w = 0∫ at σ-a.e. point on ∂Ω. From this and the arbitrariness of λ ∈ C M we
conclude that ∂Ω (− 12 I + K # ) f dσ = 0. Upon noting that (− 12 I + K # ) f = (− 12 − z) f
∫
and recalling that z  − 12 , we conclude that ∂Ω f dσ = 0, as wanted. The case when
Ω− is bounded is dealt with similarly (applying Green’s formula [70, (1.7.81)] in
Ω− , and using the fact that z  12 ), and this finishes the proof of (1.7.59).
Returning to the principal subject of discussion, we now observe that the very
same argument used to establish the injectivity of (1.7.12) in the case n ≥ 3 may
be adapted to show that if ∂Ω is compact and n = 2 then the operator zI + K #
M
is injective on the space L p (∂Ω, σ) for every p ∈ (1, ∞). Indeed, in such a
M
case
∫ if f ∈ L p (∂Ω, σ) satisfies (zI + K # ) f = 0 then (1.7.59) guarantees that
∂Ω
f dσ = 0. In turn, this cancelation property on f translates into better-than-
expected decay properties for the function u defined in (1.7.48), namely

(∇k u)(x) = O(|x| −1−k ) as |x| → ∞, for each k ∈ N0 . (1.7.61)

Granted this, the previous argument (used to deal with (1.7.12) in the case n ≥ 3)
goes through, and shows that (1.7.13) is injective.
Proof of (3): That (1.7.15) holds for each p ∈ (1, ∞) and z ∈ C follows from item
(iii) in Theorem 1.5.1 and [69, (2.1.49)]. Assume next that n ≥ 3 and that A is
M
compliant. First, fix a function f ∈ L 2(n−1)/n
A∗,1/2
(∂Ω, σ) ⊆ L 2(n−1)/n (∂Ω, σ) with
1 
the property that 2 I + K f = 0. Set
#

u± := 𝒮 A f in Ω± . (1.7.62)

Then (1.5.66) (written for A in place of A) gives

 1 
∂νA u− = 2I + K # f = 0 on ∂Ω. (1.7.63)

Next consider the function

κ−n.t.
g := u− ∂Ω = S A f ∈ L 2(n−1)/(n−2) (∂Ω, σ)
M
, (1.7.64)

with the membership implied by (1.3.63) (used with p := 2(n − 1)/n). As a conse-
quence of (1.7.62), (1.7.63), (1.7.64), and (1.7.27)-(1.7.28) (written for A in place
of A), the integral representation formula from Theorem 1.3.3 (again, written for A
in place of A) presently gives

u− = D −A g in Ω− (1.7.65)
164 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

where D −A is the boundary-to-domain double layer operator associated with A

and Ω− as in (1.3.18). In view of (1.5.20) and the fact that the geometric measure
theoretic outward
 unitnormal for Ω− is −ν, going nontangentially to the boundary
then yields 12 I + K A g = 0. After taking complex conjugates this proves
1 
2I + K A∗ g = 0. (1.7.66)
1 
By also keeping in mind that f ∈ L 2(n−1)/n
A∗,1/2
(∂Ω, σ), g = S A f , and f = 2I − K# f
we may write
∫ ∫
1 #

0= f , g dσ = 2 I − K f , S A f dσ

∂Ω ∂Ω
∫

=− A ∇u+, ∇u+ dL n, (1.7.67)
Ω+

where the last equality comes from [70, Corollary 1.7.14] (using A in place of A,
and bearing in mind (1.7.62)). Consequently,
∫

Re A ∇u+, ∇u+ dL n = 0 (1.7.68)
Ω+

and, given that A is positive semi-definite (as 

 part of A being compliant; cf.
 +
Definition 1.7.1), this forces Re A ∇u , ∇u + = 0 in Ω+ . From this and the fact
that A is compliant, we conclude that

A ∇u+ = 0 in Ω+ (1.7.69)

which, after going nontangentially to the boundary and taking the dot product with
ν, further implies

∂νA u+ = 0 on ∂Ω. (1.7.70)
Collectively, (1.7.63), (1.7.70), (1.7.62), and (1.5.66) prove that
 
f = ∂νA u− − ∂νA u+ = 0 on ∂Ω. (1.7.71)

Hence, f = 0 which ultimately shows that the operator 12 I + K # is indeed injective

when acting from the space L 2(n−1)/n
A∗,1/2
(∂Ω, σ). Finally, the fact that − 12 I + K # is also
injective when acting from the space L 2(n−1)/n
A∗,−1/2
(∂Ω, σ) is established in a similar
fashion.
Proof of (4): Work under the assumption that the coefficient tensor A is positive
definite, that the set Ω is bounded, and that the set Ω− := Rn \ Ω is connected. Also,
  M
suppose 12 I + K # f = 0 for some function f such that f ∈ L 2(n−1)/n (∂Ω, σ)
M
if n ≥ 3 and f ∈∫ L p (∂Ω, σ) with p ∈ (1, ∞) arbitrary if n = 2. In particular,
(1.7.59) implies ∂Ω f dσ = 0. Define u± := 𝒮 A f in Ω± . The aforementioned
1.7 Kernels and Images of Boundary Layer Potentials 165

cancelation property of f ensures (cf. Lemma 1.3.1) that, in all dimensions n ≥ 2,

(∇k u− )(x) = O(|x| −n+1−k ) as |x| → ∞, for each k ∈ N0 . (1.7.72)

[70, Corollary 1.7.14] (with A in place of A) then gives

∫ ∫
  1  
A ∇u−, ∇u− dL n = − 2 I − K # f , S A f dσ = 0, (1.7.73)
Ω− ∂Ω
 
given that, as part of the present assumptions, we have − 12 I − K # f = 0 on ∂Ω. In
particular,
∫

Re A ∇u−, ∇u− dL n = 0 (1.7.74)
Ω−

and, upon noting that A is also positive definite (since A is so), this ultimately
implies
∇u− = 0 in Ω− . (1.7.75)
Given that Ω− is connected, this implies that u− is constant in Ω− . Since Ω− is an
exterior domain and u− decays at infinity (as seen from (1.7.72)), and we conclude
that
u− = 0 in Ω− . (1.7.76)
As a consequence,
κ−n.t.
S A f = u− ∂Ω = 0 on ∂Ω. (1.7.77)

As such, we may write

∫ ∫
 1 #
 
0= − 2 I + K f , S A f dσ =
 A ∇u+, ∇u+ dL n, (1.7.78)
∂Ω Ω+

where the last equality comes from [70, Corollary 1.7.14]. Consequently,
∫

Re A ∇u+, ∇u+ dL n = 0. (1.7.79)
Ω+

Given that A is positive definite (since A is so), this forces

∇u+ = 0 in Ω+ (1.7.80)

which further implies


∂νA u+ = 0 on ∂Ω. (1.7.81)
Together with (1.7.76) this finally yields
 
f = ∂νA u− − ∂νA u+ = 0 on ∂Ω, (1.7.82)
166 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

ultimately proving that the operator 12 I + K # is indeed injective when acting on

M
L 2(n−1)/n (∂Ω, σ) . From this, the desired conclusion follows, keeping in mind
that ∂Ω is compact.
Proof of (5): In this portion of the proof we shall work under the assumption that
the given coefficient tensor A is complex symmetric and positive definite. Also,
we shall assume that Ω is a bounded UR domain satisfying a two-sided local John
condition and such that Ω− is connected. First, consider the case when n ≥ 3. Pick
an arbitrary complex number z ∈ C \ [− 12 , 12 ] and suppose the vector-valued function
M
f = ( fα )1≤α ≤M ∈ L12(n−1)/n (∂Ω, σ) satisfies (zI + K) f = 0. Given the current
assumptions, we may invoke [69, (11.5.201)] to conclude that
M M
L12(n−1)/n (∂Ω, σ) → L 2(n−1)/(n−2) (∂Ω, σ) . (1.7.83)

For each x ∈ Rn \ ∂Ω let us now define

# ∫ $
βα
u(x) := − νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y) , (1.7.84)
∂Ω
1≤γ ≤M

and also introduce

u± := u Ω± ∈ 𝒞∞ (Ω± )
M
. (1.7.85)

Thanks to assumptions, (1.7.83), and Theorem 1.5.1 we have

Nκ u± ∈ L 2(n−1)/(n−2) (∂Ω, σ), Nκ (∇u± ) ∈ L 2(n−1)/n (∂Ω, σ),

κ−n.t.
(1.7.86)
u± ∂Ω = (± 12 I + K) f , ∂νAu± = ∂νA D f , and L Au± = 0 in Ω± .

In addition, from (1.7.84) and [70, (1.4.24)] we see that

(∇k u− )(x) = O(|x| 1−n−k ) as |x| → ∞, for each k ∈ N0 . (1.7.87)

Together, (1.7.86), (1.7.87), and [68, (8.6.51) in Proposition 8.6.3] further imply
M nM
u± ∈ L 2n/(n−2) (Ω±, L n ) and ∇u± ∈ L 2 (Ω±, L n ) . (1.7.88)

Also, [70, Corollary 1.7.14] applies to u± in Ω± and, on account of (1.7.86), yields

∫

0= (zI + K) f , ∂νA D f dσ
∂Ω
∫  
= (z + 12 )( 12 I + K) f + (−z + 12 )(− 12 I + K) f , ∂νA D f dσ
∂Ω
∫ ∫
  ∗
   
= z+ 1
2 A ∇u+, ∇u+ dL + z − n 1
2 A∗ ∇u−, ∇u− dL n . (1.7.89)
Ω+ Ω−
1.7 Kernels and Images of Boundary Layer Potentials 167

Given that A is complex symmetric and positive definite, (1.7.88) ensures that
∫

a± := A∗ ∇u±, ∇u± dL n ∈ [0, ∞). (1.7.90)
Ω±

From (1.7.89), (1.7.90), and (1.7.51) (used with −z in place of z) we conclude that
∫ ∫
∗
 
A ∇u+, ∇u+ dL = 0 and
n
A∗ ∇u−, ∇u− dL n = 0. (1.7.91)
Ω+ Ω−

Since A∗ = A is positive definite, this eventually forces

∇u± = 0 in Ω± . (1.7.92)

Since Ω− is connected, this implies that u− ≡ 0 in Ω− (since u− vanishes at infinity;

cf. (1.7.87)). In concert with (1.5.20), this further gives that
κ−n.t. κ−n.t. κ−n.t.
f = u+ ∂Ω − u− ∂Ω = u+ ∂Ω on ∂Ω. (1.7.93)

Recall that we are presently assuming Ω+ to be bounded. Writing (1.3.35) for u+ in

Ω+ and taking into account (1.7.92) also yields
 n.t. 
u+ = D u+ ∂Ω in Ω+ . (1.7.94)

From (1.7.93), (1.7.94), and (1.5.20) we then conclude that, on the one hand,

f = ( 12 I + K) f , hence K f = 1
2 f. (1.7.95)

On the other hand, K f = −z f with z  − 12 , so we ultimately conclude that f = 0.

This proves that for every given spectral parameter z ∈ C \ − 12 , 12 the operator
zI + K is injective in the context of (1.7.19) if n ≥ 3. Finally, the case of (1.7.20) in
the two-dimensional setting is dealt with in an absolutely analogous fashion.
Proof of (6): Assume A is positive definite, Ω+ is a bounded UR domain satisfying a
two-sided local John condition, and Ω− is connected. Let us first treat the case when
M
n ≥ 3. Suppose the function f ∈ L12(n−1)/n (∂Ω, σ) is such that ( 12 I + K) f = 0
and define u± as in (1.7.85). Then (1.7.86)-(1.7.88) continue to hold. Also, writing
(1.7.89) with z := 1/2 presently gives
∫ ∫
 
0= ( 2 I + K) f , ∂ν D f dσ =
1 A A∗ ∇u+, ∇u+ dL n . (1.7.96)
∂Ω Ω+

Since A∗ is positive definite, this forces ∇u+ = 0 in Ω+ . In view of item (vii) of

Theorem 1.5.1, this allows us to write

∂νAu− = ∂νAu+ = 0 on ∂Ω. (1.7.97)

Bearing this in mind and writing (1.7.89) for z := −1/2 then yields
168 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
∫ ∫
 
0= (− 12 I + K) f , ∂νA D f dσ = − A∗ ∇u−, ∇u− dL n . (1.7.98)
∂Ω Ω−

Given that A∗ is positive definite, Ω− is connected, and u− vanishes at infinity, this

implies that u− ≡ 0 in Ω− . Consequently,
κ−n.t.
(− 12 I + K) f = u− ∂Ω = 0 on ∂Ω, (1.7.99)

hence
f = ( 12 I + K) f − (− 12 I + K) f = 0 on ∂Ω. (1.7.100)
Lastly, the case of (1.7.22) is dealt with in an absolutely analogous fashion. 
Via transposition, Theorem 1.7.2 yields the following results pertaining to the
density of ranges of single and double layer potential operators.
Corollary 1.7.3 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary UR
domain and abbreviate σ := H n−1 ∂Ω. Also, let A be a coefficient tensor with
complex entries which is Legendre-Hadamard elliptic (cf. (1.7.1)), and consider the
boundary layer potentials S, K, K # associated with A and Ω as in (1.3.62), (1.3.68),
(1.3.72). Then the following claims are valid.
(1) The boundary-to-boundary single layer operator S has dense range in the context
 2(n−1) , 2(n−1) M  2(n−1) M
S : L−1n n−2 (∂Ω, σ) −→ L n−2 (∂Ω, σ) (1.7.101)

(cf. (1.5.36)) if n ≥ 3, and in the context

p,q M M5
S : L−1 (∂Ω, σ) −→ L q (∂Ω, σ) CM (1.7.102)

for every p, q ∈ (1, ∞) and n = 2.

(2) Suppose A is also positive semi-definite, and fix some z ∈ C \ [− 12 , 12 ]. If either
A is complex symmetric, or z is real, then the operator
M M
zI + K : L 2(n−1)/(n−2) (∂Ω, σ) −→ L 2(n−1)/(n−2) (∂Ω, σ) (1.7.103)

has dense range if n ≥ 3, and

M M
zI + K : L p (∂Ω, σ) −→ L p (∂Ω, σ) (1.7.104)

has dense range for every p ∈ (1, ∞) if ∂Ω is compact and n = 2.

(3) Assume A is positive definite, Ω is bounded, and Rn \ Ω is connected. Then the
operator
M M
1
2I + K : L p (∂Ω, σ) −→ L p (∂Ω, σ) has dense range
if either n ≥ 3 and 1 < p ≤ 2(n − 1)/(n − 2), (1.7.105)
or n = 2 and 1 < p < ∞.
1.7 Kernels and Images of Boundary Layer Potentials 169

(4) Suppose the coefficient tensor A is complex symmetric and positive definite.
Also, strengthen the hypotheses on the underlying domain by assuming that Ω
is a bounded UR domain satisfying a two-sided local John condition and such
that Rn \ Ω is connected. Then for any z ∈ C \ [− 12 , 12 ] the operator

2(n−1)/(n−2) M 2(n−1)/(n−2) M
zI + K # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (1.7.106)

(cf. (1.5.27)) has dense range if n ≥ 3, and

p M p M
zI + K # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (1.7.107)

has dense range for every p ∈ (1, ∞) if n = 2.

(5) Assume the coefficient tensor A is positive definite. Also, suppose Ω is a bounded
UR domain satisfying a two-sided local John condition and such that Rn \ Ω is
connected. Then the operator
2(n−1)/(n−2) M 2(n−1)/(n−2) M
1
2I + K # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ)
(1.7.108)
has dense range if n ≥ 3,

and the operator

p M p M
1
2I + K # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) has dense range
(1.7.109)
for every p ∈ (1, ∞) if n = 2.

Proof All desired conclusions follow from Theorem 1.7.2 via transposition, with
the help of items (iii), (vi), and (ix) in Theorem 1.5.1, as well as (A.0.136) and
(A.0.137). 

The boundary layer potential operators studied in Theorem 1.5.1 turn out to
be intimately connected with boundary value problems, even in the very general
geometric setting considered
 in our
 next proposition. To facilitate its statement, we
agree to denote by Im T : X → Y the image of a linear operator T : X → Y .

Proposition 1.7.4 Suppose Ω ⊆ Rn , where n ≥ 3, is a UR domain with compact

boundary, and abbreviate σ := H n−1 ∂Ω. With the summation convention over
repeated indices in effect, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.7.110)

be a homogeneous, weakly elliptic, second-order M × M system in Rn (for some

M ∈ N), with complex constant coefficients, and recall the principal-value double
layer K and the boundary single layer S associated, as in (1.3.68) and (1.3.62),
respectively, with the writing of the system L given in (1.7.110). Having fixed an
integrability exponent p ∈ (1, ∞) along with an aperture parameter κ > 0, define
p
the L1 Regularity Problem for the system L in Ω as
170 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

⎧
⎪ M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎪
⎨
⎪ M
Nκ u, Nκ (∇u) ∈ L p (∂Ω, σ), p
where f ∈ L1 (∂Ω, σ) , (1.7.111)
⎪
⎪
⎪
⎪u
κ−n.t.

⎩ ∂Ω = f at σ-a.e. point on ∂Ω,

with the understanding that, in the case when Ω is an exterior domain, one also
imposes the decay condition

u(x) = O(|x| 2−n ) as |x| → ∞. (1.7.112)

Then the following statements are equivalent:

p
(a) The L1 Regularity Problem for the system L in Ω is solvable (i.e., for each
p M
boundary datum f ∈ L1 (∂Ω, σ) one can find a function u as in (1.7.111)).
(b) The following operator is surjective:
p M M p M
Q : L1 (∂Ω, σ) ⊕ L p (∂Ω, σ) −→ L1 (∂Ω, σ)
1 
given by Q(g, h) := 2 I + K g + Sh for every (1.7.113)
p M M
g ∈ L1 (∂Ω, σ) and h ∈ L p (∂Ω, σ) .

(c) One has


p M p M
Im 12 I − K : L1 (∂Ω, σ) → L1 (∂Ω, σ)

M p M
⊆ Im S : L p (∂Ω, σ) → L1 (∂Ω, σ) . (1.7.114)
 αβ 
Proof Let A = ar s 1≤r,s ≤n be the coefficient tensor used in the representation
1≤α,β ≤M
of the system L in (1.7.110) and recall the conormal derivative operator ∂νA, with
p
respect to A and Ω, from (A.0.184). Assume first that the L1 Regularity Problem for
p M
the system L in Ω is solvable. Pick some function f ∈ L1 (∂Ω, σ) arbitrary and
let u solve (1.7.111) for this boundary datum. From the Fatou-type result established
in [70, Theorem 3.3.4] we know that u has the additional property that
κ−n.t.
(∇u) ∂Ω exists at σ-a.e. point on ∂Ω. (1.7.115)

Granted this, [70, Lemma 1.7.3] and [70, (1.7.15)] guarantee that the function
M
h := −∂νAu is well defined and belongs to L p (∂Ω, σ) . Also, since (1.7.112)
implies [70, (1.5.5)] given that n ≥ 3, the integral representation formula (1.3.35)
from Theorem 1.3.3 presently yields
 κ−n.t.   
u = D u ∂Ω − 𝒮 ∂νAu = D f + 𝒮h in Ω. (1.7.116)

In turn, from (1.7.116), (1.5.20), (1.5.44), and the last condition in (1.7.111), we
deduce that
1.7 Kernels and Images of Boundary Layer Potentials 171
1 
f = 2I + K f + Sh at σ-a.e. point on ∂Ω, (1.7.117)
 
or, equivalently, 12 I − K f = Sh. This ultimately goes to show that the inclusion in
(1.7.114) holds, hence (a) ⇒ (c).
p M
To prove the implication (c) ⇒ (b), observe that if an arbitrary f ∈ L1 (∂Ω, σ)
M
has been given, then (1.7.114) guarantees the existence of some h ∈ L p (∂Ω, σ)
 
with the property that 12 I − K f = Sh. In view of (1.7.113), this may be further
recast simply as Q( f , h) = f which proves that the operator Q is indeed surjective.
There remains to show that (b) ⇒ (a). To this end, start with some arbitrary
p M
f ∈ L1 (∂Ω, σ) then use the surjectivity of the operator Q from (1.7.113) to
p M M
conclude that there exist two functions, g ∈ L1 (∂Ω, σ) and h ∈ L p (∂Ω, σ) ,
1 
such that 2 I + K g + Sh = f . Then, thanks to (1.3.8), (1.3.24), Theorem 1.5.1 (cf.
p
(1.5.1), (1.5.20)), and [69, (11.8.4)], the function u := Dg + 𝒮h in Ω solves the L1
Regularity Problem (1.7.111). The proof of Proposition 1.7.4 is now complete. 

We next elaborate on the fact that, given a weakly elliptic second-order system
L, the solvability of the Regularity Problems for L on either side of a compact UR
surface is actually equivalent to the surjectivity of the single layer operator associated
with L on said surface.

Corollary 1.7.5 Let Ω ⊆ Rn , where n ≥ 3, be a bounded UR domain and set

σ := H n−1 ∂Ω. Define Ω+ := Ω and Ω− := Rn \ Ω. Also, fix an integrability
exponent p ∈ (1, ∞) along with an aperture parameter κ ∈ (0, ∞). Finally, for some
M ∈ N, consider a homogeneous, weakly elliptic, second-order M × M system L in
Rn with constant (complex) coefficients, and recall the boundary-to-boundary single
layer operator S associated with L and Ω as in (1.3.62).
Then the following statements are equivalent:
p
(i) The L1 Regularity Problems for the system L in Ω± are solvable.
M p M
(ii) The operator S : L p (∂Ω, σ) → L1 (∂Ω, σ) is surjective.

Proof This is a consequence of Proposition 1.7.4 (applied both to Ω+ and Ω− ; here

item (7) in [68, Lemma 5.10.9] is also relevant), upon observing that

p M p M
Im 12 I ∓ K : L1 (∂Ω, σ) → L1 (∂Ω, σ)

M p M
⊆ Im S : L p (∂Ω, σ) → L1 (∂Ω, σ) (1.7.118)

M p M
if and only if S : L p (∂Ω, σ) → L1 (∂Ω, σ) is surjective. 

We conclude this section by proving a uniqueness result for the mixed boundary
value problem formulated in a very general context.

Corollary 1.7.6 Suppose Ω ⊆ Rn , where n ≥ 2, is an open set with a lower Ahlfors

regular boundary and with the property that σ := H n−1 ∂Ω is a doubling measure.
172 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Hence, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. For some M ∈ N, assume A is a
complex coefficient tensor of type (n × n, M × M) which is positive definite, in the
sense that there exists some real number c > 0 such that
 
αβ β
Re Aζ, ζ = Re ar s ζs ζrα ≥ c|ζ | 2, ∀ζ = (ζrα ) 1≤r ≤n ∈ Cn×M . (1.7.119)
1≤α ≤M

Denote by L A the homogeneous constant (complex) coefficient second-order M × M

system in Rn associated with A as in (A.0.139). Finally, fix some aperture κ > 0 and
exponents p, q ∈ [1, ∞].
In this context, suppose u is a complex vector-valued function satisfying

⎧
⎪ u ∈ Wloc 1,1 M
(Ω) , Lu = 0 in Ω,
⎪
⎪
⎪
⎪
⎨ Nκ (∇u) ∈ L p (∂Ω, σ), Nκ u ∈ L q (∂Ω, σ),
⎪
(1.7.120)
⎪
⎪
⎪
⎪ κ−n.t. κ−n.t.
⎪
⎪ u ∂Ω and (∇u) ∂Ω exist σ-a.e. on ∂nta Ω,
⎩
and (with the dependence on κ dropped)

at σ-a.e. point x ∈ ∂∗ Ω one has that

 n.t.    (1.7.121)
either u ∂Ω (x) = 0, or ∂νAu (x) = 0.

Then, in any of the following scenarios:

(i) ∂Ω is unbounded, p = 2(n − 1)/n, and q = 2(n − 1)/(n − 2);
(ii) Ω is bounded, p ≥ 2(n − 1)/n, and q ≥ p/(p − 1);
(iii) Ω is an exterior domain, p ≥ 2(n − 1)/n, q ≥ p/(p − 1), and there exist two real
numbers, a > n/2 and b > n − 1 − a, such that

(∇u)(x) = O(|x| −a ) and u(x) = O(|x| −b ) as |x| → ∞, (1.7.122)

it follows that
the function u is locally constant in Ω. (1.7.123)

Proof From [70, Corollary 1.7.14] we know that formula [70, (1.7.162)] holds in the
present setting. Bearing in mind that L Au = 0 and that the boundary term vanishes,
thanks of (1.7.121), this implies that A∇u, ∇u = 0 at L n -a.e. point in Ω. In view
of positive definiteness property (1.7.119), this forces ∇u to vanish at L n -a.e. point
in Ω. Hence, ultimately, u is locally constant in Ω. 
1.8 Modified Boundary Layer Potential Operators 173

1.8 Modified Boundary Layer Potential Operators

As far as the boundary-to-domain version of the double layer potential operator

associated with a given M × M weakly elliptic system and an open set Ω ⊆ Rn with
an Ahlfors regular boundary is concerned, we can go one step further and adapt
the original definition from (1.3.17)-(1.3.18) as to allow functions from the space
  M
L 1 ∂Ω, 1+σ(x)
|x | n (where, as usual, σ := H n−1 ∂Ω), which is strictly bigger than
  M
the space L 1 ∂Ω, 1+σ(x) |x | n−1
considered in (1.3.17). One appealing virtue of such
an extension is that this larger weighted Lebesgue space contains a multitude of other
useful spaces, such as all Hölder functions as well as functions of bounded mean
oscillations, on ∂Ω.
Before stating our first main result in this regard, the reader is reminded that if
Ω ⊆ Rn is an open set with an Ahlfors regular boundary, a Carleson measure μ in Ω
is said to be vanishing provided
   .
μ B(x, r) ∩ Ω
lim sup   = 0. (1.8.1)
R→0+ x ∈∂Ω and H n−1 B(x, r) ∩ ∂Ω
r ∈(0,R)

Definition 1.8.1 Let Ω ⊆ Rn be an open set with an unbounded Ahlfors regu-

lar boundary. A Carleson measure μ in Ω (cf. [70, (2.4.1)]) is said to be super
vanishing provided the following three conditions are satisfied:
   .
μ B(x, r) ∩ Ω
lim sup   = 0, (1.8.2)
R→∞ x ∈∂Ω, r >0 H n−1 B(x, r) ∩ ∂Ω
B(x,r)⊆R n \B(0,R)
   .
μ B(x, r) ∩ Ω
lim sup   = 0, (1.8.3)
R→0+ x ∈∂Ω and H n−1 B(x, r) ∩ ∂Ω
r ∈(0,R)
   .
μ B(x, r) ∩ Ω
lim sup   = 0. (1.8.4)
R→∞ x ∈∂Ω and H n−1 B(x, r) ∩ ∂Ω
r ∈(R,∞)

We are now ready to state the theorem alluded to above.

Theorem 1.8.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the property

that ∂Ω is an Ahlfors regular set; in particular, Ω is a set of locally finite perimeter.
Denote by ν = (ν1, . . . , νn ) the geometric measure  outward unit normal to Ω
 αβtheoretic
and abbreviate σ := H n−1 ∂Ω. Also, let L = ar s ∂r ∂s 1≤α,β ≤M be a homogeneous,
weakly elliptic, constant (complex) coefficient, second-order M × M system in Rn
(for some M ∈ N).
In this setting, consider the following modified version of the double layer operator
in (1.3.18) acting on each function
174 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
  σ(x)  M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, (1.8.5)
1 + |x| n
according to (recall that the summation convention over repeated indices is presently
in effect):
 
Dmod f (x) (1.8.6)
# ∫ $
βα  (rγβ) 
:= − νs (y)ar s (∂r Eγβ )(x − y) − k1 (−y) fα (y) dσ(y)
∂∗ Ω
1≤γ ≤M

(rγβ)
at each point x ∈ Ω, where k1 := (∂r Eγβ ) · 1Rn \B(0,1) for every r ∈ {1, . . . , n},
γ, β ∈ {1, . . . , M }, and E = (Eγβ )1≤γ,β ≤M is the matrix-valued fundamental solution
associated with L as in [70, Theorem 1.4.2]. Then the following properties hold.

(1) The operator Dmod is meaningfully defined, and satisfies

M
Dmod f ∈ 𝒞∞ (Ω) and L(Dmod f ) = 0 in Ω,

  M (1.8.7)
for each f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n .

In addition, the operator Dmod is compatible with D from (1.3.18), in the sense
  M
that for each function f belonging to the smaller space L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
M
(hence, in particular, for each function f ∈ L p (∂∗ Ω, σ) with p ∈ [1, ∞)) the
difference

C f := Dmod f − D f is a constant (belonging to C M ) in Ω. (1.8.8)

As a consequence,
  σ(x) M
∇Dmod f = ∇D f in Ω for each f ∈ L 1 ∂∗ Ω, . (1.8.9)
1 + |x| n−1
Moreover,

Dmod maps constant (C M -valued) functions on ∂∗ Ω

(1.8.10)
into constant (C M -valued) functions in Ω.
In addition, at each point x ∈ Ω one may express
# ∫ $
μ  βα μ
∂ Dmod f (x) = − νs (y)ar s (∂ ∂r Eγβ )(x − y) fα (y) dσ(y)
∂∗ Ω
1≤γ ≤M
1 σ(x)  M
for each μ ∈ N0n with | μ| > 0 and f = ( fα )1≤α ≤M ∈ L ∂∗ Ω, 1+ |x | n .
(1.8.11)
1.8 Modified Boundary Layer Potential Operators 175

Finally, given any function

  σ(x)  M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω,
1 + |x| n
 σ(x) (1.8.12)
with the property that ∂τ j k fα ∈ L 1 ∂∗ Ω,
1 + |x| n−1
for all j, k ∈ {1, . . . , n} and α ∈ {1, . . . , M },

it follows that for each index ∈ {1, . . . , n} and each point x ∈ Ω one has
#∫ $
  βα
∂ Dmod f (x) = ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ(y) .
∂∗ Ω
1≤γ ≤M
(1.8.13)

(2) For each η ∈ (0, 1) there exists a constant C ∈ (0, ∞) with the property that
   
sup dist(x, ∂Ω)1−η ∇ Dmod f (x) ≤ C f [𝒞. η (∂Ω)] M (1.8.14)
x ∈Ω
. M
for every function f ∈ 𝒞η (∂Ω) . Moreover,

if Ω ⊆ Rn is a uniform domain with the property that ∂Ω is an Ahlfors

. M .   M
regular set then Dmod : 𝒞η (∂Ω) → 𝒞η Ω is a well-defined,
linear, and bounded operator for each given exponent η ∈ (0, 1),
(1.8.15)
whereas
if Ω ⊆ Rn is an NTA domain with an upper Ahlfors regular boundary
. M .   M
then Dmod : 𝒞ηvan (∂Ω) → 𝒞ηvan Ω is a well-defined, linear, and
bounded operator for each given exponent η ∈ (0, 1),
(1.8.16)
where the homogeneous vanishing Hölder spaces intervening above are defined
as in (A.0.48) (with Σ := ∂Ω and Σ := Ω, respectively). Also, for each η ∈ (0, 1)
and each p ∈ (1, ∞) there exists some C ∈ (0, ∞) with the property that for each
. M
function f ∈ 𝒞η (∂Ω) one has
# ∫ $ 1/p
1 p
sup ∇Dmod f dist(·, ∂Ω) p−1
dL n
x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,∞)

≤ C f [𝒞. η (∂Ω)] M (1.8.17)

and
176 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 # ∫ $ 1/p .
1 p
lim sup ∇Dmod f dist(·, ∂Ω) p−1
dL n
R→0+ x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,R)
 . M
≤ C dist f , 𝒞ηvan (∂Ω) , (1.8.18)
 .
where the distance is measured in the space 𝒞η (∂Ω) ,  · [𝒞. η (∂Ω)] M . As
M

a corollary, if the function f actually belongs to the homogeneous vanishing

. M
Hölder space 𝒞ηvan (∂Ω) for some η ∈ (0, 1), then for each p ∈ (1, ∞) one has
 # ∫ $ 1/p .
1 p
lim sup ∇Dmod f dist(·, ∂Ω) p−1
dL n
= 0.
R→0+ x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,R)
(1.8.19)

(3) Strengthen the original geometric hypotheses by assuming that ∂Ω is actually a

UR set. Also, fix an aperture parameter κ ∈ (0, ∞). Then, as a consequence of
(1.8.13) and [70, Theorem 2.5.1], the nontangential boundary trace
  κ−n.t.
∂ Dmod f ∂Ω
exists (in C M ) at σ-a.e. point on ∂∗ Ω,
(1.8.20)
for each function f as in (1.8.12) and each index ∈ {1, . . . , n}.

Another corollary of (1.8.13) and [70, (2.4.8)] (also keeping in mind [70, The-
orem 1.4.2]) is the fact that for each ε > 0 and each p ∈ (1, ∞)
 
Nκε ∇(Dmod f ) ∈ Lloc (∂Ω, σ) for each function
p

  σ(x)  M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, such that
1 + |x| n
(1.8.21)
 σ(x) p
∂τ j k fα ∈ L 1 ∂∗ Ω, ∩ Lloc (∂∗ Ω, σ)
1 + |x| n−1
for all j, k ∈ {1, . . . , n} and all α ∈ {1, . . . , M }.

In addition, as seen from (1.8.6), [70, (2.5.32)], and [70, Theorem 1.4.2], for
each truncation parameter ε ∈ (0, ∞) one has

Nκε (Dmod f ) ∈ Lloc (∂Ω, σ) for each function

p

   M (1.8.22)
f ∈ L 1 ∂∗ Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂∗ Ω, σ) with p ∈ (1, ∞).

Furthermore, for each function f as in (1.8.12) one has

1.8 Modified Boundary Layer Potential Operators 177
 
∂νA(Dmod f ) (x) (1.8.23)
# ∫ $
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0
y ∈∂∗ Ω 1≤μ ≤M
|x−y |>ε

where the conormal derivative is considered as in (A.0.184).

Next, let Kmod be the modified boundary-to-boundary double layer potential
   M
operator acting on each function f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n at
σ-a.e. x ∈ ∂Ω according to

Kmod f (x) (1.8.24)

# ∫ $
βα  (rγβ) (rγβ) 
:= − lim+ νs (y)ar s k ε (x − y) − k1 (−y) fα (y) dσ(y)
ε→0
∂∗ Ω 1≤γ ≤M

where
(rγβ)
kε := (∂r Eγβ ) · 1Rn \B(0,ε) for each ε > 0. (1.8.25)
Then, as seen from definitions and [68, Proposition 5.6.7], the operator Kmod
   M
is compatible with K (acting on functions from L 1 ∂∗ Ω, 1+σ(x)
|x | n−1 as in
(1.3.68)) in the sense that
 σ(x)  M
for each function f ∈ L 1 ∂∗ Ω , 1+ |x | n−1
(hence, in particular,
M
for each function f ∈ L p (∂∗ Ω, σ)
with p ∈ [1, ∞)) the difference
c f := Kmod f − K f is a constant (belonging to C M ) on ∂Ω and satisfies
|c f | ≤ C∂Ω, L ·  f [L 1 (∂ Ω, σ(x) )] M for some finite C∂Ω, L > 0.
∗
1+| x | n−1
(1.8.26)

Furthermore, the following jump-formula holds:

  κ−n.t.

1
Dmod f + Kmod f at σ-a.e. point on ∂∗ Ω,
= 2I
∂Ω
  σ(x)  M
(1.8.27)
for each given function f ∈ L 1 ∂∗ Ω, ,
1 + |x| n
where, as usual, I is the identity operator. As a consequence of (1.8.27) and
(1.8.10),

if ∂Ω is a UR set then the operator Kmod maps constant (C M -valued)

functions on ∂∗ Ω into constant (C M -valued) functions on ∂∗ Ω.
(1.8.28)
(4) Continue to work under the additional assumption that ∂Ω is a UR set. Then
for each integrability exponent p ∈ (1, ∞) there exists a constant C ∈ (0, ∞)
178 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
M
with the property that for each function f ∈ BMO(∂Ω, σ) the measure
 p
∇ Dmod f dist(·, ∂Ω) p−1 dL n is Carleson in Ω, in the quantitative sense that
∫
1  p
sup   ∇ Dmod f dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤ C f  .
p
(1.8.29)
[BMO(∂Ω,σ)] M

(with the piece of notation introduced in (A.0.19)). In particular, corresponding

to p = 2, it follows that21
 2
∇ Dmod f dist(·, ∂Ω) dL n is a Carleson measure in Ω,
M
(1.8.30)
for each function f ∈ BMO(∂Ω, σ) .

Moreover, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
M
that for each function f ∈ BMO(∂Ω, σ) one has
 ∫
1  p 1
p
lim+ sup   ∇ Dmod f dist(·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , [VMO(∂Ω, σ)] M (1.8.31)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)] M . In

particular,
 p
∇ Dmod f dist(·, ∂Ω) p−1 dL n
is a vanishing Carleson measure in Ω (1.8.32)
M
for each f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞)

and, corresponding to p = 2,
 2
∇ Dmod f dist(·, ∂Ω) dL n is a vanishing Carleson measure in Ω,
M
(1.8.33)
for each function f ∈ VMO(∂Ω, σ) .

Furthermore, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which
depends only on n, p, L, and the UR constants of ∂Ω, with the property that for
M
each function f ∈ BMO(∂Ω, σ) one has

2
21 it is natural to refer to ∇Dmod f dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated
with f via the operator Dmod
1.8 Modified Boundary Layer Potential Operators 179


max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫
 p 1/p
× ∇ Dmod f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim+ sup  1 ×
R→0 x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω

∫
 p 1/p
× ∇ Dmod f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ .
  p 1/p
× ∇ Dmod f dist (·, ∂Ω) p−1
dL n
B(x,r)∩Ω


M
≤ C dist f , CMO(∂Ω, σ) , (1.8.34)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)] M

(recall that CMO(∂Ω, σ) has been introduced in (A.0.52)). As a consequence
of (1.8.34) and Definition 1.8.1,
 p
∇ Dmod f dist (·, ∂Ω) p−1 dL n is a super vanishing Carleson measure in Ω,
M
for each function f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).
(1.8.35)

(5) If ∂Ω is bounded, then all properties listed in items (1)-(4) are valid for the
ordinary double layer operator D, as originally defined in (1.3.18), in place of
its modified version Dmod . In particular, if ∂Ω is a compact UR set then for each
p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property that for each
M
function f ∈ BMO(∂Ω, σ) one has
 ∫
1  p 1
p
lim+ sup   ∇ D f dist(·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , [VMO(∂Ω, σ)] M (1.8.36)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)] M . In

particular, if ∂Ω is a compact UR set then
180 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
  p
∇ Df dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
M
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞)
(1.8.37)
and, corresponding to p = 2, it follows that22
 2
∇ D f dist(·, ∂Ω) dL n is a vanishing Carleson measure in Ω,
M
(1.8.38)
for each function f ∈ VMO(∂Ω, σ)

(6) Strengthen the original geometric hypotheses by assuming now that Ω is a UR

domain. Then for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which
depends only on n, p, A, and the UR constants of ∂Ω, with the property that for
each function
  σ(x) M
p
f ∈ L 1 ∂Ω , ∩ Lloc (∂Ω, σ) (1.8.39)
1 + |x| n

each point xo ∈ ∂Ω, and each radius r ∈ (0, ∞) one has

⨏ ⨏ p 1/p
Kmod f − Kmod f dσ dσ (1.8.40)
Δ(x o ,r) Δ(x o ,r)
∫ ∞ ⨏ ⨏ p 1/p dλ
≤C f− f dσ dσ ,
1 Δ(x o ,λr) Δ(x o ,λr) λ2

where Δ(xo, R) := B(xo, R) ∩ ∂Ω for each R ∈ (0, ∞).

As a consequence of (1.8.40) and the definition of the Fefferman-Stein sharp
maximal operator (cf. (A.0.195)), it follows that for each p ∈ (1, ∞) there exists
some constant C ∈ (0, ∞) such that the following pointwise inequality holds:
 #
≤ C · fp# on ∂Ω, for every
Kmod f p
  M (1.8.41)
function f ∈ L 1 ∂Ω , 1+σ(x)
p
|x | n ∩ Lloc
(∂Ω, σ) .

Also, if p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 then given any functions
  M
f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, 1+σ(x)
p
|x | n together with
∫ (1.8.42)
p M
g ∈ Lcomp (∂Ω, σ) satisfying g dσ = 0 ∈ C M ,
∂Ω

if follows that

2
22 it is natural to refer to ∇D f dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated with
f via the operator D
1.8 Modified Boundary Layer Potential Operators 181
∫ ∫
|Kmod f ||g| dσ < +∞, | f ||K # g| dσ < +∞,
∂Ω ∂Ω
∫ ∫ (1.8.43)
and Kmod f , g dσ = f , K # g dσ.
∂Ω ∂Ω

Finally, for each p ∈ (1, ∞) there exists C ∈ (0, ∞), which depends only on
n, p, A, and the UR constants of ∂Ω, with the property that for each function
   M
f ∈ L 1 ∂Ω , 1+σ(x)
|x | n , each point xo ∈ ∂Ω, and each radius r ∈ (0, ∞) one
has
# ∫ $ 1/p
1   p
  ∇ Dmod f (x) dist (x, ∂Ω) p−1
dL (x)
n
(1.8.44)
σ Δ(xo, r) B(xo,r)∩Ω
∫ ∞ ⨏ ⨏ p 1/p dλ
≤C f− f dσ dσ
1 Δ(x o ,λr) Δ(x o ,λr) λ2

where Δ(xo, R) := B(xo, R) ∩ ∂Ω for each R ∈ (0, ∞).

As a corollary of (1.8.44) and the definition of the Fefferman-Stein sharp max-
imal operator (cf. (A.0.195)), it follows that for each p ∈ (1, ∞) there exists a
constant C ∈ (0, ∞) such that the pointwise inequality
# ∫ $ 1/p
1   p
sup   ∇ Dmod f dist (·, ∂Ω) p−1 dL n ≤ C · fp# (x)
r >0 σ Δ(x, r) B(x,r)∩Ω
(1.8.45)
  M
σ(x) 
holds for each point x ∈ ∂Ω and each function f ∈ L 1 ∂Ω , 1+ |x | n .

Before presenting the proof of this theorem we make a number of comments.

Comment 1: Similar results to those presented in Theorem 1.8.2 are valid for modi-
mod
fied boundary-to-domain “tangential” integral operators Tjk with j, k ∈ {1, . . . , n},
of the sort described in Proposition 1.8.16. Then, given any p ∈ (1, ∞), there exists
a constant C ∈ (0, ∞) such that for each function f ∈ BMO(∂Ω, σ) we have
 ∫
1  mod  p 1
p
lim+ sup   ∇ Tjk f dist(·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , VMO(∂Ω, σ) , (1.8.46)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ).
As a consequence,
 mod  p
∇ Tjk f dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
(1.8.47)
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).
182 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

In addition, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
that for each function f ∈ BMO(∂Ω, σ) we have


max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫
 mod  p 1/p
× ∇ Tjk f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim+ sup  1 ×
R→0 x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω

∫
 mod  p 1/p
× ∇ Tjk f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ .
 mod  p 1/p
× ∇ Tjk f dist (·, ∂Ω) p−1
dL n
B(x,r)∩Ω

 
≤ C dist f , CMO(∂Ω, σ) , (1.8.48)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ)
(recall that CMO(∂Ω, σ) has been introduced in (A.0.52)). As a corollary of (1.8.48)
and Definition 1.8.1,
 mod  p
∇ Tjk f dist (·, ∂Ω) p−1 dL n is a super vanishing Carleson measure in Ω,
for each function f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).
(1.8.49)

Comment 2: Analogous properties to those discussed in Theorem 1.8.2 are also

valid for the modified versions of the boundary-to-domain Cauchy-Clifford integral
operator Cmod defined in (1.8.238) in Corollary 1.8.24, hence further augmenting the
results in [70, Proposition 2.5.29] and Proposition 1.6.1. This may be seen from [70,
(2.5.311)], Theorem 1.8.2, and Comment 1 above.
Comment 3: Let DΔ,mod be the modified boundary-to-domain harmonic double layer
operator associated with Ω and L := Δ as in (1.8.6), i.e.,
∫  
1 ν(y), y − x ν(y), y
DΔ,mod f (x) := − · 1 R \B(0,1)
(y) f (y) dσ(y)
ωn−1 |x − y| n |y| n
n

∂∗ Ω
(1.8.50)
1.8 Modified Boundary Layer Potential Operators 183
 
for each f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n and each x ∈ Ω. If EΔ denotes the standard fundamental
solution for the Laplacian in Rn (cf. (A.0.65)), then formula (1.8.13) presently
becomes (with the summation convention in effect)
∫
 
∂ DΔ,mod f (x) = (∂j EΔ )(x − y)(∂τ j f )(y) dσ(y) (1.8.51)
∂∗ Ω

for each index ∈ {1, . . . , n}, each point x ∈ Ω, and each function
 σ(x)
f ∈ L 1 ∂∗ Ω, with the property that
1 + |x| n
 (1.8.52)
σ(x)
∂τ j k f ∈ L 1 ∂∗ Ω, for all j, k ∈ {1, . . . , n}.
1 + |x| n−1
This may be further streamlined if we bring in 𝒮Δ,mod , the modified boundary-to-
domain harmonic single layer operator23 associated with Ω and L := Δ as in (1.5.50),
i.e.,
∫
 
𝒮Δ,mod f (x) := EΔ (x − y) − EΔ (−y) · 1Rn \B(0,1) (−y) f (y) dσ(y)
∂Ω
 (1.8.53)
σ(x)
for each f ∈ L ∂Ω, 1+ |x | n−1 and each point x ∈ Ω.
1

Then, thanks to (1.5.51), we may recast (1.8.51) simply as (again with the summation
convention in effect)

∂ DΔ,mod f = ∂j 𝒮Δ,mod (∂τ j f ) in Ω, (1.8.54)

for each index ∈ {1, . . . , n} and each function f as in (1.8.52).

Comment 4: From [70, Proposition 2.3.3] we see that there are a number of al-
ternative characterizations of the boundary-to-boundary double layer potential op-
erator Kmod , originally introduced in (1.8.24). For one thing, given any function
  M
f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
along with any surface ball Δ ⊆ ∂Ω, there
exists a constant C f ,Δ ∈ R with the property that for σ-a.e. point x ∈ Δ we have
 
(Kmod f )(x) = K f 1Δ (x) + C f ,Δ (1.8.55)
#∫ $
βα  
+ νs (y)ar s (∂r Eγβ )(x − y) − (∂r Eγβ )(xΔ − y) fα (y) dσ(y)
∂∗ Ω\Δ
1≤γ ≤M

where xΔ ∈ ∂Ω is the center of the surface ball Δ.

For another thing, having fixed an arbitrary reference point x0 ∈ ∂Ω then for any
  M
given function f = ( fα )1≤α ≤M ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
it follows that the limit

23 recall the convention that when the single layer operator is applied to functions originally defined
only on ∂∗ Ω, these functions are regarded as being extended by zero outside ∂∗ Ω, to the entire ∂Ω
184 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

 
mod f )(x) := lim
(K K f 1B(x0, j)∩∂Ω (x)
j→∞
∫ .

βα
− νs (y)ar s (∂r Eγβ )(x0 − y) fα (y) dσ(y) (1.8.56)
1≤γ ≤M
y ∈∂∗ Ω
1≤ |x0 −y |< j

exists at σ-a.e. point x ∈ ∂Ω, and there exists a constant C f ,x0 ∈ R with the property
that
(K f ,x0 at σ-a.e. x ∈ ∂Ω.
mod f )(x) = (Kmod f )(x) + C (1.8.57)

We now turn to the proof of Theorem 1.8.2.

Proof of Theorem 1.8.2 All claims in item (1) with the exception of (1.8.10) and
(1.8.12)-(1.8.13) are direct consequences of definitions, [70, Corollary 2.5.3], and
[70, Theorem 1.4.2]. As far as the claim made in (1.8.10) is concerned, having fixed
 points x0, x1 ∈ Ω along with some constant λ ∈ C , we need to show
two arbitrary M

that (Dmod λ (x0 ) − (Dmod λ (x1 ) = 0. A moment’s reflection shows that this is further
equivalent to proving that for each fixed α, γ ∈ {1, . . . , M } we have
∫
βα  
νs (y)ar s (∂r Eγβ )(x0 − y) − (∂r Eγβ )(x1 − y) dσ(y) = 0. (1.8.58)
∂∗ Ω

To this end, consider the vector field F := (Fs )1≤s ≤n whose components are defined
at L n -a.e. point y ∈ Ω by
βα  
Fs (y) := ar s (∂r Eγβ )(x0 − y) − (∂r Eγβ )(x1 − y) , 1 ≤ s ≤ n. (1.8.59)

Hence,
n
F ∈ Lloc
1
(Ω, L n ) . (1.8.60)
Also, a direct computation which also uses [70, (1.4.33)] gives that, with the diver-
gence taken in the sense of distributions in Ω,
βα βα
divF = ∂s Fs = −ar s ∂yr ∂ys Eγβ (x0 − ·) + ar s ∂yr ∂ys Eγβ (x1 − ·)

= δαγ δx0 − δαγ δx1 ∈ ℰ(Ω) (1.8.61)

where, generally speaking, δx is the Dirac distribution in Ω with mass at x ∈ Ω.

Fix a compact set K ⊂ Ω whose interior contains both x0 and x1 . Then the same
argument which has established [70, (2.3.117)] presently gives that there exists a
constant CK ∈ (0, ∞) such that
CK
| F(y)| ≤ , ∀y ∈ Ω \ K. (1.8.62)
1 + |y| n

Granted this, [68, Lemma 8.3.7] shows that there exists C = C(K, κ, n) ∈ (0, ∞) such
that
1.8 Modified Boundary Layer Potential Operators 185
 C
NκΩ\K F (y) ≤ for each y ∈ ∂Ω. (1.8.63)
1 + |y| n
In turn, from (1.8.63), [68, (8.2.26)], and [68, Lemma 7.2.1], we conclude that

NκΩ\K F ∈ L 1 (∂Ω, σ). (1.8.64)

In addition, since F is continuous in a neighborhood of ∂Ω it follows that the

pointwise nontangential boundary trace
κ−n.t.
F ∂Ω
exists at every point in ∂nta Ω, (1.8.65)

and at σ-a.e. y ∈ ∂∗ Ω we have

 κ−n.t.
βα  
ν(y) · F ∂Ω (y) = νs (y)ar s (∂r Eγβ )(x0 − y) − (∂r Eγβ )(x1 − y) . (1.8.66)

Finally, we note that in the case when Ω is unbounded we have | F(y)| = O(|y| −n )
for y ∈ Ω with |y| → ∞. Hence, in such a scenario,
∫
|y · F(y)| dL n (y) = O(R) as R → ∞. (1.8.67)
[B(0,2R)\B(0,R)]∩Ω

At this stage, we may write

∫
βα  
νs (y)ar s (∂r Eγβ )(x0 − y) − (∂r Eγβ )(x1 − y) dσ(y)
∂∗ Ω
∫  κ−n.t.
 
= ν(y) · F (y) dσ(y) = (𝒞∞b (Ω))∗ divF, 1 𝒞∞b (Ω)
∂∗ Ω ∂Ω

= δαγ − δαγ = 0. (1.8.68)

Above, the first equality comes from (1.8.66), the second equality is formula [68,
(1.4.6)] (keeping in mind that the hypotheses of [68, Theorem 1.4.1] are satis-
fied, thanks to (1.8.60), (1.8.61), (1.8.65), (1.8.67)), the third equality is seen from
(1.8.61), and the final equality is obvious. In turn, (1.8.68) establishes (1.8.58). This
finishes the proof of (1.8.10).
Finally, the identity claimed in (1.8.13) for any function f as in (1.8.12) is estab-
lished staring from (1.8.11), then reasoning as in (1.3.31) based on the integration
by parts formula on the boundary from [69, Lemma 11.1.7]. The justification of the
claims in item (1) is now complete.
Moving on, the first claim in item (2), pertaining to (1.8.14), follows from (a
vector-valued version of) [70, Lemma 2.1.2], used with 𝒬 := ∇Dmod , bearing in
mind that the constant C2 defined in [70, (2.1.18)] presently vanishes, thanks to
(1.8.10) (upon also noting that dist(x, ∂Ω) ≤ dist(x, ∂∗ Ω) for each x ∈ Ω). Then
(1.8.15) follows by combining (1.8.14) with [68, (5.11.78)].
186 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

As regards (1.8.16), suppose Ω is as stated there, and fix an arbitrary function

. M
f ∈ 𝒞η (∂Ω) with η ∈ (0, 1). From [68, Proposition 5.9.16] we see that Ω is an
Ahlfors regular domain. In view of the present goal, there is no loss of generality in
assuming that Ω is also a uniform domain. More specifically, if ∂Ω is unbounded
then [68, Lemma 5.11.3] tells us that Ω is connected which, in concert with item (4)
in [68, Proposition 5.11.14], implies that Ω is a uniform domain. If ∂Ω is bounded,
then [68, Lemma 5.11.3] tells us that Ω has finitely many connected components
separated by a strictly positive distance from one another, and item (4) in [68,
Proposition 5.11.14] ensures that each such component is a uniform domain. Then
the conclusion we seek may be obtained by writing Dmod as a sum of modified double
layer potential operator corresponding to each connected component and by piecing
together information about each individual piece.
In summary, there is no loss of generality in assuming that Ω is both an Ahlfors
regular domain and a uniform domain. In particular, both (1.8.14) and (1.8.15) are
presently valid. In particular, the latter implies that for some constant C ∈ (0, ∞)
independent of f we have
.  
Dmod f ∈ 𝒞η (Ω) and Dmod f [𝒞. η (Ω)] M ≤ C f [𝒞. η (∂Ω)] M .
M
(1.8.69)
. M
Let us now show that the operator Dmod maps an arbitrary f ∈ 𝒞ηvan (∂Ω) into
the space (cf. (A.0.48))
 .
.η   M . η  M 
𝒞van Ω = u∈ 𝒞 Ω : lim+ sup u𝒞. η (B(x,r)∩Ω) = 0 . (1.8.70)
r→0
x ∈Ω

To this end, fix some arbitrary threshold ε > 0. Also, fix some β ∈ (η, 1), and invoke
[69, Theorem 3.2.2] to find a function
. .
g ∈ 𝒞η (∂Ω) ∩ 𝒞β (∂Ω)
M
with  f − g[𝒞. η (∂Ω)] M < ε. (1.8.71)
.   M
From (1.8.69) we know that u := Dmod f belongs to 𝒞η Ω . Making use of nota-
tion introduced in the formulation of the result recorded in [68, Proposition 5.11.15],
for each x ∈ ∂Ω and r ∈ (0, R) we may estimate
1.8 Modified Boundary Layer Potential Operators 187
 
D .
mod g η
[𝒞 (B(x,r)∩Ω)] M
   
= Dmod g [𝒞. η (B(x,r)∩Ω)] M ≤ C Dmod g [𝒞. η (Ω
x, r )]
M

 
≤ C sup ∇(Dmod g) · dist(·, ∂Ωx,r )1−η
Ω x, r
 
= C sup ∇(Dmod g) · dist(·, ∂Ωx,r )1−β · dist(·, ∂Ωx,r )β−η
Ω x, r
 
≤ Cr β−η · sup ∇(Dmod g) · dist(·, ∂Ωx,r )1−β
Ω x, r
 
≤ Cr β−η · sup ∇(Dmod g) · dist(·, ∂Ω)1−β
Ω x, r
 
≤ Cr β−η · sup ∇(Dmod g) · dist(·, ∂Ω)1−β
Ω

≤ Cr β−η · g[𝒞. β (∂Ω)] M , (1.8.72)

for some constant C ∈ (0, ∞) independent of x, r, and g. The first inequality in

(1.8.72) is simply a consequence of the monotonicity of Hölder semi-norm with re-
spect to the domain, and [68, Proposition 5.11.15]. The second inequality in (1.8.72)
is implied by [68, (5.11.78)] written with Ω replaced by Ωx,r . That the latter set is
actually a uniform domain (as required in [68, (5.11.78)]) is seen from [68, Proposi-
tion 5.11.15] and [68, (5.11.65)]. The third inequality in (1.8.72) is a consequence of
the fact that β − η > 0 and diam Ωx,r ≤ Cr, which imply dist(·, ∂Ωx,r )β−η ≤ Cr β−η
at each point in Ωx,r . The fourth inequality in (1.8.72) follows from the monotonicity
of the distance-to-the-boundary function (cf. [68, Lemma 8.1.1] and [68, Proposi-
tion 5.11.15]), bearing in mind that 1 − β > 0. The fifth inequality in (1.8.72) merely
uses the fact that Ωx,r is a subset of Ω, while the final inequality in (1.8.72) comes
from (1.8.14) written for g in place of f and β in place of η.
Having established (1.8.72), we may now conclude that there exists a constant
C ∈ (0, ∞) with the property that
 
sup Dmod g [𝒞. η (B(x,r)∩Ω)] M ≤ Cr β−η · g[𝒞. β (∂Ω)] M for each r ∈ (0, R). (1.8.73)
x ∈∂Ω

For each fixed r ∈ (0, R) we may now estimate

188 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

sup u[𝒞. η (B(x,r)∩Ω)] M

x ∈∂Ω
   
≤ sup Dmod g [𝒞. η (B(x,r)∩Ω)] M + sup Dmod ( f − g)[𝒞. η (B(x,r)∩Ω)] M
x ∈∂Ω x ∈∂Ω
   
≤ sup Dmod g [𝒞. η (B(x,r)∩Ω)] M + Dmod ( f − g)[𝒞. η (Ω)] M
x ∈∂Ω

≤ Cr β−η g[𝒞. β (∂Ω)] M + C f − g[𝒞. η (∂Ω)] M

≤ Cr β−η g[𝒞. β (∂Ω)] M + Cε. (1.8.74)

Above, we have used the triangle inequality in the first step, the monotonicity of
Hölder semi-norm with respect to the domain in the second step, (1.8.73) and
(1.8.14) (written for f − g in place of f ) in the third step, and (1.8.71) in the final
step. In view of the arbitrariness of ε > 0, from (1.8.74) we then readily conclude
that 
lim+ sup u[𝒞. η (B(x,r)∩Ω)] M = 0. (1.8.75)
r→0 x ∈∂Ω

Next, for each fixed r ∈ (0, R/5), let us estimate the Hölder semi-norm
u[𝒞. η (B(x,r)∩Ω)] M in the situation when x ∈ Ω \ ∂Ω = Ω. Consider first the case
when dist(x, ∂Ω) ≥ 4r. Then B(x, r) ⊆ Ω and for each y, z ∈ B(x, r) we may use the
Mean Value Theorem to write
|u(y) − u(z)|
≤ |y − z| 1−η · sup |(∇u)(ξ)|
|y − z| η ξ ∈[y,z]

≤ (2r)1−η · sup |∇(Dmod g)(ξ)|

ξ ∈[y,z]

+ (2r)1−η · sup |∇(Dmod ( f − g))(ξ)|

ξ ∈[y,z]

≤ Cr 1−η g[𝒞. β (∂Ω)] M · sup dist(ξ, ∂Ω)β−1

ξ ∈[y,z]

+ Cr 1−η  f − g[𝒞. η (∂Ω)] M · sup dist(ξ, ∂Ω)η−1

ξ ∈[y,z]

≤ Cr β−η g[𝒞. β (∂Ω)] M + Cε, (1.8.76)

where we have also relied on (1.8.14) (twice, first with g in place of f and β in place
of η, and, second, with f − g in place of f ), as well as on (1.8.71), plus the fact that
the distance from any point ξ ∈ B(x, r) to ∂Ω is at least 3r, while β − 1 < 0 and
η − 1 < 0. Bearing in mind that ε > 0 is arbitrary and that β − η > 0, from (1.8.76)
we then deduce that
1.8 Modified Boundary Layer Potential Operators 189

lim sup u[𝒞. η (B(x,r)∩Ω)] M = 0. (1.8.77)
r→0+ x ∈Ω with
dist(x,∂Ω)≥4r

At this stage, there remains to treat the case when the point x ∈ Ω satisfies
dist(x, ∂Ω) < 4r. Then there exists some point x∗ ∈ ∂Ω with the property that
|x − x∗ | < 4r, which further implies that we have B(x, r) ∩ Ω ⊆ B(x∗, 5r) ∩ Ω.
Consequently,

u[𝒞. η (B(x,r)∩Ω)] M ≤ u[𝒞. η (B(x ,5r)∩Ω)] M

∗

≤ sup u[𝒞. η (B(z,5r)∩Ω)] M . (1.8.78)

z ∈∂Ω

As such,

lim+ sup u[𝒞. η (B(x,r)∩Ω)] M
r→0 x ∈Ω with
dist(x,∂Ω)<4r

≤ lim+ sup u[𝒞. η (B(z,5r)∩Ω)] M = 0, (1.8.79)
r→0 z ∈∂Ω

with the last equality coming from (1.8.73). Together, (1.8.73), (1.8.77), and (1.8.79)
. M
prove that for each f ∈ 𝒞ηvan (∂Ω) the function u = Dmod f belongs to the space
described in (1.8.70). The justification of (1.8.16) is therefore complete.
Going further, (1.8.17) may be proved by relying on (1.8.14) and the fact that
there exists C ∈ (0, ∞) such that
∫
sup dist(·, ∂Ω)η p−1 dL n ≤ Cr n−1+η p for each r > 0, (1.8.80)
x ∈∂Ω B(x,r)∩Ω

itself a consequence of item (i) in [68, Proposition 8.7.1] (see [68, (8.7.3)], presently
used with y := x, N := 0, and α := 1 − ηp < 1).
. M
Let us now show that (1.8.19) holds when f ∈ 𝒞ηvan (∂Ω) . To this end, pick an
arbitrary ε > 0. Also, choose some β ∈ (η, 1) and select g as in (1.8.71). Then, on
the one hand,
# ∫ $ 1/p
1 p
sup ∇Dmod ( f − g) dist(·, ∂Ω) p−1
dL n
≤ Cε, (1.8.81)
x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,∞)

by (1.8.17) and (1.8.71). On the other hand,

190 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
# ∫ $ 1/p
1 p
sup ∇Dmod g dist(·, ∂Ω) p−1
dL n
x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,∞)

# ∫ $ 1/p
1 βp−1
≤ sup dist(·, ∂Ω) dL n
· g[𝒞. η (∂Ω)] M
x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,∞)

≤ Cr β−η g[𝒞. η (∂Ω)] M , (1.8.82)

by (1.8.17) used with g in place of f and with β in place of η, and by (1.8.80)

used with β in place of η. Collectively, (1.8.81) and (1.8.82) then imply (1.8.19), in
view of the arbitrariness of ε > 0. Finally, (1.8.18) is a consequence of (1.8.17) and
(1.8.19). This takes care of item (2).
To proceed, strengthen the hypotheses on Ω by assuming that ∂Ω is a UR set.
Then the jump-formula (1.8.27) follows from definitions, [70, Corollary 2.5.3], and
[70, Theorem 1.4.2]. With the exception of (1.8.23), all other claims in item (3) are
justified as explained in the very statement of the theorem. To justify (1.8.23), select
a function f = ( fα )1≤α ≤M as in (1.8.12). From what we have proved so far in item
(3) it follows that the conormal derivative ∂νA(Dmod f ) may be meaningfully defined,
as in (A.0.184). Specifically, at σ-a.e. point x ∈ ∂∗ Ω we have
 A   κ−n.t.
μγ
∂ν (Dmod f ) (x) = νi (x)ai j ∂j (Dmod f )γ ∂Ω (x) (1.8.83)
1≤μ ≤M
# ∫ $
μγ βα  
= lim νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0+
y ∈∂∗ Ω 1≤μ ≤M
|x−y |>ε

where the first equality comes from (A.0.184) and the second equality is a conse-
quence of (1.8.13) and the jump-formula from [70, Theorem 2.5.1] (reasoning as in
(1.5.107)-(1.5.108)). This establishes (1.8.23).
As regards item (4), the first order of business is to prove the estimate stated in
(1.8.29). However, having established (1.8.10), the argument proceeds very much
as in the case of the proof of [70, Corollary 2.4.2]. Specifically, pick an arbitrary
M
function
 f ∈ BMO(∂Ω,
 σ) . Having also fixed a point xo ∈ ∂Ω and a scale
r ∈ 0, 2 diam(∂Ω) , the same argument which has produced [70, (2.4.136)] (based
on the estimate in [70, (2.4.34)] and the decay of the integral kernel of ∇Dmod ) now
gives
∫
  p
∇ Dmod f (x) dist(x, ∂Ω) p−1 dx (1.8.84)
B(x o ,r)∩Ω
   
≤ Cσ B(xo, r) ∩ ∂Ω fp# (xo ) p + Cσ B(xo, r) ∩ ∂Ω f1# (xo ) p,
1.8 Modified Boundary Layer Potential Operators 191

where the L q -based Fefferman-Stein maximal function fq# , with q ∈ [1, ∞), has
been defined in (A.0.195). Granted this  estimate, the version of (1.8.29) with the
supremum taken in the regime r ∈ 0, 2 diam(∂Ω) follows on account of [68,
(7.4.111)]. Finally, the case when Ω is an exterior domain and r ≥ 2 diam(∂Ω) is
handled much as in [70, (2.4.142), (2.4.143)]. Going further, consider the claim
made in (1.8.31). To this end, pick p ∈ (1, ∞) and select an arbitrary function
M
f ∈ BMO(∂Ω, σ) . Also, fix some exponent η ∈ (0, 1) and choose an arbitrary
function .
g ∈ 𝒞η (∂Ω)
M M
∩ BMO(∂Ω, σ) . (1.8.85)
 
Then for each r ∈ 0, 2 diam(∂Ω) and x ∈ ∂Ω we may estimate
 ∫
1  p 1
p
  ∇ Dmod ( f − g) dist(·, ∂Ω) p−1 dL n
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤ C f − g[BMO(∂Ω,σ)] M , (1.8.86)

thanks to (1.8.29) written f − g in place of f . In addition, we have

 ∫
1  p 1
p
  ∇ Dmod g dist(·, ∂Ω) p−1 dL n
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
 ∫ 1
1 p
≤C   dist(·, ∂Ω) pη−1 dL n g[𝒞. η (∂Ω)] M
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤ Cr η g[𝒞. η (∂Ω)] M . (1.8.87)

Indeed, the first inequality above uses (1.8.14) (written for g in place of f ), while
the second inequality is based on [68, (8.6.101)] used with λ := 1 − pη, α := 1,
β := n − 1, E := B(x, r) ∩ Ω, and [68, (8.1.17)]. Collectively, (1.8.86) and (1.8.87)
imply that
 ∫
1  p p
1

lim+ sup   ∇ Dmod f dist(·, ∂Ω) p−1 dL n

R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤ C f − g[BMO(∂Ω,σ)] M , (1.8.88)

for some constant C ∈ (0, ∞) independent of f and g. With this in hand, (1.8.31)
follows on account of [69, Theorem 3.1.3] and the ability to choose g arbitrary as
in (1.8.85). In turn, (1.8.31) readily implies (1.8.32)-(1.8.33). The estimate claimed
in (1.8.34) is a particular case of a more general result established later in Theo-
rem 5.1.22. This takes care of item (4).
Next, the first claim in item (5) is clear from (1.8.8), while the remaining properties
follow from what we have proved in item (4), [69, (4.6.16)], and (1.8.8). As far as item
(6) is concerned, the estimates claimed in (1.8.40) and (1.8.44) are special cases of
more general results proved later in Theorem 5.1.15 and Theorem 5.1.8, respectively,
192 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

while the claim formulated in (1.8.43) is a consequence of [70, (2.3.36)], (1.8.24),

and (1.3.72). This completes the proof of Theorem 1.8.2. 

We continue by making three comments pertaining to Theorem 1.8.2. First, it is

   M
clear from (1.8.6) that for each function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x) |x | n , each
index j ∈ {1, . . . , n}, and each point x ∈ Ω, we have
# ∫ $
  βα
∂j Dmod f (x) = − νs (y)ar s (∂j ∂r Eγβ )(x − y) fα (y) dσ(y) . (1.8.89)
∂∗ Ω
1≤γ ≤M

In particular, this formula is valid when the function f belongs to the space
. M M
𝒞η (∂Ω) for some η ∈ (0, 1), or to the space BMO(∂Ω, σ) .
Our second comment further sheds light on the claims in item (2) of Theorem 1.8.2
. M
in the case when Ω is an Ahlfors regular domain. Specifically, fix f ∈ 𝒞η (∂Ω)
with η ∈ (0, 1), along with some j ∈ {1, . . . , n}. Also, fix some point x ∈ Ω. It
follows from (1.8.89) that we may express ∂j Dmod f (x) as

  ∫
αγ
∂j Dmod f (x) = m j, x (y) fα (y) dσ(y) (1.8.90)
∂Ω 1≤γ ≤M

where, for each α, γ ∈ {1, . . . , M }, we have set

αγ βα
m j,x (y) := −νs (y)ar s (∂j ∂r Eγβ )(x − y) for σ-a.e. y ∈ ∂Ω. (1.8.91)
αγ
In particular, m j,x is σ-measurable on ∂Ω and [70, Theorem 1.4.2] yields the estimate
αγ
m j,x (y) ≤ C|x − y| −n for σ-a.e. y ∈ ∂Ω. (1.8.92)

Pick now  n−1 

p := n−1
n−1+η ∈ n ,1 (1.8.93)
 
and note that this entails η = (n − 1) p1 − 1 . Also, choose some q ∈ (1, ∞) then
select
   q 
d ∈ (n − 1) qp − 1 , nq − (n − 1) and set θ := n−1
d p − 1 ∈ (0, 1). (1.8.94)

A direct computation based on (1.8.92) and [68, Lemma 7.2.1] then shows that
 αγ 
m  q (n−1−nq)/q
j,x L (∂Ω,σ) ≤ C dist(x, ∂Ω) (1.8.95)

and, if xo ∈ ∂Ω is such that dist(x, ∂Ω) = |x − xo |,

 αγ 
m  q (d+n−1−nq)/q
j,x L (∂Ω, | ·−x o | d σ) ≤ C dist(x, ∂Ω) . (1.8.96)

Collectively, (1.8.95)-(1.8.96) and (1.8.93)-(1.8.94) then permit us to estimate

1.8 Modified Boundary Layer Potential Operators 193
 αγ 1−θ  αγ θ
m  q · m  q
j,x L (∂Ω,σ) j,x L (∂Ω, | ·−x o | d σ)

≤ C dist(x, ∂Ω)(1−θ)(n−1−nq)/q+θ(d+n−1−nq)/q

= C dist(x, ∂Ω)η−1 . (1.8.97)

αγ  
In particular, m j,x ∈ L q ∂Ω, (1 + | · −xo | d )σ . Moreover, if eα denotes the constant
C M -valued function defined on ∂Ω whose α-th component is 1 and all others are
zero, then (1.8.10) implies
∫
αγ  
m j,x (y) dσ(y) = ∂j Dmod eα γ (x) = 0. (1.8.98)
∂Ω

Granted these properties, [69, Corollary 4.5.3] applies and gives that
αγ
m j,x is a multiple of a H p -molecule on ∂Ω (cf. [69, Definition 4.5.1]),
 αγ 
and we have m j,x  H p (∂Ω,σ) ≤ C · dist(x, ∂Ω)η−1 for some finite con- (1.8.99)
stant C > 0, independent of the given point x.
In turn, from (1.8.90), (1.8.99), and [69, Proposition 4.8.7] (which is currently
applicable since any H p -molecule on ∂Ω is also a multiple of a H 1 -molecule on
∂Ω), we deduce that, with ·, · denoting the duality bracket between functions
satisfying a homogeneous Hölder condition of order η on ∂Ω, modulo constants,
and the Hardy space H p on ∂Ω (cf. [69, Theorem 4.6.1]), we have

⎧
⎪ αγ 
  ⎪
⎨
⎪ [ fα ], m j,x 1≤γ ≤M if ∂Ω is unbounded,
∂j Dmod f (x) =  (1.8.100)
⎪
⎪ αγ 
⎪ fα , m if ∂Ω is bounded.
⎩ j,x 1≤γ ≤M

In concert with [69, (4.6.9)], (1.8.99), and [69, (4.8.46)] (which we invoke in the
case when ∂Ω is bounded, bearing (1.8.98) in mind), this ultimately allows us to
estimate

  
n 
M
 αγ 
∇ Dmod f (x) ≤  fα 𝒞. η (∂Ω) m j,x  H p (∂Ω,σ)
j=1 γ=1

≤ C · dist(x, ∂Ω)η−1  f [𝒞. η (∂Ω)] M . (1.8.101)

Since x ∈ Ω is arbitrary and C ∈ (0, ∞) is independent of x and f , this gives an

alternative proof of (1.8.14) (in the case when Ω is an Ahlfors regular domain).
Our third comment in relation to Theorem 1.8.2 is that all results there have
natural counterparts for the Cauchy-Clifford integral operator C, C associated with
Ω as in (A.0.53) and (A.0.54), respectively.
The fact that the normal derivative of the classical harmonic double layer potential
has no jump across the boundary is, in a smooth setting, occasionally referred to as the
194 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Lyapunov-Tauber theorem (cf., e.g., [31]). In the case of the Laplacian, the formula
for the normal derivative of the double layer takes a particularly aesthetic form. This
is indicated in the next proposition, which also serves as a natural justification of
the aforementioned Lyapunov-Tauber phenomenon in a very general geometric and
analytic setting.

Proposition 1.8.3 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain. Abbreviate

σ := H n−1 ∂Ω, and denote by ν the geometric measure theoretic outward unit
normal to Ω. Recall that DΔ,mod is the modified boundary-to-domain harmonic double
layer operator associated with Ω as in (1.8.50), and that SΔ,mod is the modified
boundary-to-boundary harmonic single layer operator from (1.5.211). Then for
each function
 σ(x)
f ∈ L 1 ∂Ω, such that, for some p ∈ (1, ∞),
1 + |x| n
 (1.8.102)
σ(x) p
∂τ j k f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n},
1 + |x| n−1
the following formula holds:
  
∂ν DΔ,mod f = ∂τ j k SΔ,mod ∂τ j k f in Ω. (1.8.103)
1≤ j<k ≤n

As a corollary, for each

 σ(x)
f ∈ L 1 ∂Ω, such that, for some p ∈ (1, ∞),
1 + |x| n−1
 (1.8.104)
σ(x) p
∂τ j k f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n},
1 + |x| n−1
one has   
∂ν DΔ f = ∂τ j k SΔ,mod ∂τ j k f in Ω, (1.8.105)
1≤ j<k ≤n

where DΔ is the ordinary boundary-to-domain harmonic double layer potential

operator associated with the set Ω.

Proof Denote by ν1, . . . , νn the scalar components of ν and select an aperture pa-
rameter κ > 0. Having fixed an arbitrary function f as in (1.8.102) we may write
1.8 Modified Boundary Layer Potential Operators 195


n
  κ−n.t. 
n
  κ−n.t.

∂ν DΔ,mod f = ν j ∂j DΔ,mod f = ν j ∂k 𝒮Δ,mod (∂τ j k f )

∂Ω ∂Ω
j=1 j,k=1

1   1  
n n
 κ−n.t.
 κ−n.t.

= ν j ∂k 𝒮Δ,mod (∂τ j k f ) − νk ∂j 𝒮Δ,mod (∂τ j k f )

2 j,k=1 ∂Ω 2 j,k=1 ∂Ω

1 
n    κ−n.t.  1 
n   
= ∂τ j k 𝒮Δ,mod ∂τ j k f = ∂τ j k SΔ,mod ∂τ j k f
2 j,k=1 ∂Ω 2 j,k=1


n   
= ∂τ j k SΔ,mod ∂τ j k f . (1.8.106)
1≤ j<k ≤n

Above, the first equality is simply the definition of the normal derivative (cf. (A.0.184)
for the Laplacian in its standard writing). The second equality comes from (1.8.54).
The third equality is a consequence of the fact that ∂τ j k is antisymmetric in j
and k. The fourth
 equality is implied by [69, (11.3.26)] written for the function
u := 𝒮Δ,mod ∂τ j k f , which satisfies the hypotheses of [69, Proposition 11.3.2].
Specifically, the first condition in [69, Proposition 11.3.2, (11.3.24)] is ensured by
(1.5.69)-(1.5.70), while the second condition in [69, Proposition 11.3.2, (11.3.24)] is
guaranteed by (1.5.54). Also, the existence of the nontangential traces in [69, Propo-
sition 11.3.2, (11.3.25)] is seen from (1.5.80) and (1.5.53). Next, the fifth equality
in (1.8.106) comes from (1.5.80), while the final equality in (1.8.106) is once again
a consequence of the antisymmetry of ∂τ j k in j and k. This finishes the proof of
(1.8.102).
Finally, (1.8.105) is a consequence of (1.8.102) and (1.8.8). 
Moving on, it is both useful and informative to provide concrete examples of
modified double layer potential operators for which the results in Theorem 1.8.2
apply.

Example 1.8.4 Work in R2 ≡ C and consider the factorization of the two-

dimensional Laplacian Δ = ∂x2 + ∂y2 given by

 where D
Δ = DD  := ∂x − i∂y and D := ∂x + i∂y . (1.8.107)
Let AD,D
 be the coefficient tensor induced by this factorization of the Laplacian,
 D as in (1.8.107). Also, let Ω ⊆ C ≡ R2 be an open set
defined as in (1.4.32) with D,
with an Ahlfors regular boundary. Then a direct computation reveals that the modified
boundary-to-domain double layer operator Dmod associated as in (1.8.6) with the set
σ(ζ )
Ω and the coefficient tensor A := AD,D
 acts on functions f ∈ L 1 ∂∗ Ω, 1+ |ζ | 2
according to
∫  .
1 1 1
Dmod f (z) = − 1 (ζ) f (ζ) dζ, ∀z ∈ Ω. (1.8.108)
2πi ζ − z ζ C\B(0,1)
∂∗ Ω
196 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Also, the modified boundary-to-boundary double layer operator Kmod associated as

in (1.8.24)-(1.8.25) with the set Ω and the coefficient tensor A := AD,D as above,

σ(ζ )
turns out to act on each function f ∈ L 1 ∂∗ Ω, 1+ |ζ |2 according to

∫  .
1 1 1
Kmod f (z) = lim+ 1 (ζ) − 1C\B(0,1) (ζ) f (ζ) dζ (1.8.109)
ε→0 2πi ζ − z C\B(z,ε) ζ
∂∗ Ω

at σ-a.e. z ∈ ∂Ω. Naturally, these may be thought of as the modified versions of the
standard boundary-to-domain and boundary-to-boundary Cauchy integral operators
(cf. the discussion in [70, §1.1]). For these operators, the results in Theorem 1.8.2
apply.

Example 1.8.5 Work +in the Clifford algebra context and consider the factorization
of the Laplacian Δ = nj=1 ∂j2 in Rn given by


n
 where D
Δ = DD  := D := i e j ∂j . (1.8.110)
j=1

Let AD,D
 be the coefficient tensor induced by this factorization of the Laplacian,
 D as in (1.8.110). In addition, let Ω ⊆ Rn be an open
defined as in (1.4.32) with D,
set with an Ahlfors regular boundary. Denote by ν the geometric measure theoretic
outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Then a direct computation
shows that the modified boundary-to-domain double layer operator Dmod associated
as in (1.8.6) with the set Ω and the coefficient tensor A := AD,D acts on functions

σ(y)
f ∈ L ∂∗ Ω, 1+ |y | n ⊗ C n according to
1

∫  .
1 x−y −y
Dmod f (x) = − 1 n (y)  ν(y)  f (y) dσ(y)
ωn−1 |x − y| n | − y| n R \B(0,1)
∂∗ Ω
(1.8.111)
at each point x ∈ Ω. Moreover, the modified boundary-to-domain double layer
operator Kmod associated as in (1.8.24)-(1.8.25) with the set Ωand the coefficient
tensor A := AD,D
 as above, acts on each given function f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n ⊗ C n
according to
∫ 
1 x−y
Kmod f (x) = lim+ 1 n (y) (1.8.112)
ε→0 ωn−1 |x − y| n R \B(x,ε)
∂∗ Ω .
−y
− 1 n (y)  ν(y)  f (y) dσ(y)
| − y| n R \B(0,1)

at σ-a.e. point x ∈ ∂Ω. These may be naturally regarded as the modified versions of
the standard boundary-to-domain and boundary-to-boundary Cauchy-Clifford inte-
1.8 Modified Boundary Layer Potential Operators 197

gral operators (cf. (A.0.53) and (A.0.54)). For these operators, the results described
in Theorem 1.8.2 are valid.

Example 1.8.6 Consider the factorization of the Laplacian Δ in R2n ≡ Cn given by

2∂
% z̄1 (
 where D &
 := (2∂z1 , . . . , 2∂zn ) and D := & .. ))
Δ = DD (1.8.113)
& . )
'2∂z̄n *
   
where, as in (1.4.177), we set ∂z j := 12 ∂x j − i∂y j and ∂z̄ j := 12 ∂x j + i∂y j for
each index j ∈ {1, . . . , n}. To proceed, let AD,D  be the coefficient tensor in-
duced by this factorization of the Laplacian, defined as in (1.4.32) with D,  D as
in (1.8.110). Also, fix an open set Ω ⊆ R with an Ahlfors regular boundary.
2n

Abbreviate σ := H 2n−1 ∂Ω and identify the geometric measure theoretic out-
ward unit normal ν = (ν1, ν2, . . . , ν2n−1, ν2n ) ∈ R2n to Ω with the complex vector
νC := (ν1 + iν2, . . . , ν2n−1 + iν2n ) ∈ Cn . A straightforward interpretation of defini-
tions then shows that the modified boundary-to-domain double layer operator Dmod
associated as in (1.8.6) with the set Ω and the coefficient tensor A := AD,D  acts on

σ(ζ )
arbitrary complex-valued functions f ∈ L ∂∗ Ω, 1+ |ζ |2n according to
1

∫  ν (ζ), ζ − z  νC (ζ), ζ
 .
1 C C C
Dmod f (z) = − 1Cn \B(0,1) (ζ) f (ζ) dσ(ζ)
ω2n−1 |z − ζ | 2n |ζ | 2n
∂∗ Ω

+ (1.8.114)
at each point z ∈ Ω, where u, w C := nj=1 u j w j for each u = (u j ) j ∈ Cn and
w = (w j ) j ∈ Cn , is the Hermitian complex-pairing. Furthermore, the modified
boundary-to-domain double layer operator Kmod associated as in (1.8.24)-(1.8.25)
with the set Ω and the coefficient tensor A := AD,D  as above, acts on each given

σ(ζ )
complex-valued function f ∈ L 1 ∂∗ Ω, 1+ |ζ |2n according to

∫  ν (ζ), ζ − z 
1 C C
Kmod f (z) = lim+ 1Cn \B(z,ε) (ζ) (1.8.115)
ε→0 ω2n−1 |z − ζ | 2n
∂∗ Ω  .
νC (ζ), ζ
C
− 1Cn \B(0,1) (ζ) f (ζ) dσ(ζ)
|ζ | 2n

at σ-a.e. point z ∈ ∂Ω. In a natural fashion, these may be thought of as the modified
versions of the standard boundary-to-domain and boundary-to-boundary Bochner-
Martinelli integral operators (cf. (7.5.2) and (7.5.26)). For these operators, the results
in Theorem 1.8.2 therefore apply.

Example 1.8.7 There are other factorizations of the Laplacian which yield boundary
layer potential operators of interest. For example, from (7.3.15)-(7.3.16) we know
that the complex Laplacian  := − 12 Δ in Cn ≡ R2n may be factored as
198 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

 with D
 = DD  := D := (∂¯ + ϑ). (1.8.116)
Also, in the context of differential forms, the Hodge-Laplacian Δ = −(dδ + δd) may
be factored as (recall that d, δ stand, respectively, for the exterior derivative operator
and its formal transpose)
# $
δ
Δ = DD with D  := i(d, δ) and D := i , (1.8.117)
d

as well as
 with D
Δ = DD  := D := i(d + δ). (1.8.118)
All these factorizations naturally lead to modified boundary layer potentials for which
the results in Theorem 1.8.2 are valid.
On to a new topic, in view of (1.8.26) and [69, (11.4.5)], the following result is a
generalization of Proposition 1.5.6.

Proposition 1.8.8 Assume Ω ⊆ Rn is a UR domain and abbreviate σ := H n−1 ∂Ω.

Consider a homogeneous, second-order, constant complex coefficient,
 weakly elliptic
αβ
M × M system L in R , and pick a coefficient tensor A = ar s 1≤α,β ≤M for
n
1≤r,s ≤n
which L A = L. Let K be the boundary-to-boundary double layer potential operator
associated with Ω and A as in (1.3.68), and bring in its modified version Kmod
from (1.8.24). Finally, recall the family of singular integral operators U jk with
j, k ∈ {1, . . . , n} defined in (1.5.251) and fix some integrability exponent p ∈ (1, ∞).
Then for each function
   M
f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) such that
 
∂τ j k fα belongs to L 1 ∂Ω, 1+σ(x) (1.8.119)
p
|x | n−1
∩ Lloc (∂Ω, σ)
for all j, k ∈ {1, . . . , n} and α ∈ {1, . . . , M },

and each pair of indices j, k ∈ {1, . . . , n} one has

 
∂τ j k Kmod f = K(∂τ j k f ) + U jk (∇tan f ) (1.8.120)

where, as in the case of (1.5.253), ∇tan f is regarded as the M × n matrix-valued

function whose (α, s) entry is the s-th component of the tangential gradient ∇tan fα
(cf. (A.0.78)).

In particular, Proposition 1.8.8 is applicable to all modified boundary-to-boundary

double layer potential operator Kmod given in Examples 1.8.4-1.8.7.
Proof of Proposition 1.8.8 Denote by ν = (ν1, . . . , νn ) the geometric measure the-
oretic outward unit normal to Ω. Also, fix an aperture parameter κ > 0, and pick
j, k ∈ {1, . . . , n} along with γ ∈ {1, . . . , M }. Given a function f = ( fα )1≤α ≤M as in
(1.8.119), we may then write
1.8 Modified Boundary Layer Potential Operators 199
   
∂τ j k Kmod f γ = ∂τ j k 12 f + Kmod f γ − 12 ∂τ j k fγ (1.8.121)
  κ−n.t.   κ−n.t.
= ν j ∂k (Dmod f )γ ∂Ω
− νk ∂j (Dmod f )γ ∂Ω
− 12 ∂τ j k fγ

where the second equality is justified by invoking [69, Proposition 11.3.2] with
u := Dmod f and p := 1 (its present applicability is ensured by (1.8.20), (1.8.27), and
(1.8.22)).
Next, recall the definition of the boundary-to-domain double layer potential op-
erator from (1.3.18), as well as the definition of the operator introduced in [70,
(2.9.53)]. Then making use of (1.8.13) and [69, (11.4.8)], for each x ∈ Ω we may
express
∫
βα
∂j (Dmod f )γ (x) = ν j (y)ar s (∂r Eγβ )(x − y)(∇tan fα )s (y) dσ(y)
∂Ω
∫
βα
− νs (y)ar s (∂r Eγβ )(x − y)(∇tan fα ) j (y) dσ(y)
∂Ω
βα    
= ar s (∂r 𝒮γβ ) ν j (∇tan fα )s (x) + D((∇tan f ) j ) γ (x), (1.8.122)

where (∇tan f ) j is the C M -valued function whose α-th component is (∇tan fα ) j . Grant-
ed (1.8.121) and (1.8.122), the same type of argument as in (1.5.257)-(1.5.260) proves
(1.8.120). 

Going further, the reader is reminded that homogeneous Sobolev space of order
one have been introduced in [69, Definition 11.5.3] on suitable Ahlfors regular sets
(cf. (A.0.127)-(A.0.128)).

Theorem 1.8.9 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the property

that ∂Ω is a UR set, and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous, weak-
ly elliptic, second-order M × M system L in Rn , with complex constant coefficients,
and recall the modified boundary-to-boundary single layer operator Smod associated
αβ
with L and Ω as in (1.5.73). Also, pick a coefficient tensor A = ar s 1≤α,β ≤M for
1≤r,s ≤n
which L A = L, and bring in the operator K A#  associated as in (1.3.72) with the
coefficient tensor A . Finally, fix an integrability exponent p ∈ (1, ∞). Then the
operator
M .p M
Smod : L p (∂Ω, σ) −→ L1 (∂Ω, σ) (1.8.123)

is well defined, linear, and bounded, when the target space is endowed with the
induced by semi-norm (A.0.128). In addition,
M .p 5 M
Smod : L p (∂Ω, σ) −→ L1 (∂Ω, σ) ∼ defined as
.p 5 M M
(1.8.124)
Smod f := Smod f ∈ L1 (∂Ω, σ) ∼ , ∀ f ∈ L p (∂Ω, σ)
200 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

is also a well-defined, linear, and bounded operator, when the quotient space is
endowed with the natural semi-norm24 introduced in [69, (11.5.138)].
Furthermore, with 𝒮mod denoting the modified version of the single layer operator
  M
acting on functions from L 1 ∂Ω, 1+σ(x)
|x | n−1
as in (1.5.50), for each aperture
parameter κ > 0 there exists some constant C = C(Ω, L, n, p, κ) ∈ (0, ∞) with the
M
property that for each function f ∈ L p (∂Ω, σ) and each truncation parameter
ε ∈ (0, ∞) one has:
M  
𝒮mod f ∈ 𝒞∞ (Ω) , L 𝒮mod f = 0 in Ω,
    
Nκ ∇𝒮mod f ∈ L p (∂Ω, σ), Nκ ∇𝒮mod f  L p (∂Ω,σ) ≤ C f [L p (∂Ω,σ)] M ,
 
Nκε (𝒮mod f ) ∈ Lloc (∂Ω, σ) for each q ∈ 0, n−1
q
n−2 ,
κ−n.t.
∇(𝒮mod f ) ∂Ω exists at σ-a.e. point on ∂∗ Ω,
 
∂νA𝒮mod f = − 12 I + K A#  f at σ-a.e. point on ∂∗ Ω,
  κ−n.t.

and 𝒮mod f (x) = (Smod f )(x) at σ-a.e. point x ∈ Aκ (∂Ω),

∂Ω
in particular (cf. [68, Proposition 8.8.4]), at σ-a.e. point x ∈ ∂∗ Ω.
(1.8.125)

Proof The claims pertaining to the operator (1.8.123) are justified based on Proposi-
tion 1.2.6, [70, Theorem 1.4.2], (1.5.83), [68, (7.7.106)], and the fact that the singular
integral operators defined in (1.5.82) induce bounded mappings (cf. [70, (2.3.20)]
and [70, Theorem 1.4.2])
M M
#
Tjk : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ) , ∀ j, k ∈ {1, . . . , n}. (1.8.126)

Having dealt with (1.8.123), the claims regarding (1.8.124) readily follow. The
properties in the first two lines of (1.8.125) are seen from item (xi) of Theorem 1.5.1
and [68, (7.7.106)]. The membership in the third line of (1.8.125) is a consequence
of [68, (7.7.106)] and (1.5.69)-(1.5.70). The existence of the nontangential boundary
trace in the fourth line of (1.8.125) comes from (1.5.53). Next, the jump-formula
in the fifth line of (1.8.125) comes from (1.5.58), bearing [68, (7.7.106)] in mind.
Finally, the boundary trace formula claimed in the last two lines of (1.8.125) is a
consequence of (1.5.80) and [68, (7.7.106)]. 

On boundaries of planar chord-arc domains, the invertibility properties of bound-

ary layer potentials for the Laplacian and related to one another in the manner
described in the proposition below (see also Corollary 2.3.15 in this regard).

24 recall from [69, Proposition 11.5.14] that said semi-norm is actually a genuine norm if Ω ⊆ R n
is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
1.8 Modified Boundary Layer Potential Operators 201

Proposition 1.8.10 Let Ω ⊆ R2 be a chord-arc domain with unbounded boundary

and abbreviate σ := H 1 ∂Ω. Recall the principal-value harmonic double layer K
and its (real) transpose K # from (1.6.71)-(1.6.72). Also, bring in the two-dimensional
modified boundary-to-boundary single layer operator Smod from (1.6.89). Finally, fix
p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then

the operators ± 12 I + K are isomorphisms on L p (∂Ω, σ), if and only if


the operators ± 12 I + K # are isomorphisms on L p (∂Ω, σ), if and only
 . p  5 (1.8.127)
if Smod : L p (∂Ω, σ) → L1 (∂Ω, σ) ∼ (defined as in (1.8.124)) is
an isomorphism.
Proof The first equivalence in (1.8.127) follows by duality (cf. item (iii) in Theo-
rem 1.5.1), while the second equivalence in (1.8.127) is a consequence of (1.8.124),
(1.6.91), and [69, Proposition 11.5.15]. 
Next we take a look at injectivity properties of the modified boundary-to-boundary
single layer operator in the context of (1.5.77), when the underlying domain is a half-
space.

Proposition 1.8.11 Consider a homogeneous, complex constant coefficient, second-

order M × M system L in Rn (where M, n ∈ N with n ≥ 2), satisfying the Legendre-
Hadamard (strong) ellipticity condition [70, (1.3.4) in Definition 1.3.2]. Given a
half-space Ω ⊆ Rn , abbreviate σ := H n−1 ∂Ω and consider the modified boundary-
to-boundary single layer operator Smod associated with L and Ω as in (1.5.73).
Finally, fix an integrability exponent p ∈ (1, ∞) along with a Muckenhoupt weight
M
w ∈ Ap (∂Ω, σ). Then for each function f ∈ L p (∂Ω, wσ) one has

Smod f is a constant on ∂Ω if and only if f = 0. (1.8.128)

Proof Making a translation and a rotation, there is no loss of generality in assuming

M
that Ω = R+n . Assume this is the case and consider a function f ∈ L p (∂Ω, wσ)
with the property that Smod f is constant on ∂Ω. Also, fix an aperture parameter
κ ∈ (0, ∞) and some arbitrary truncation parameter ε > 0. If we define Ω± := R±n
and set u± := 𝒮mod f in Ω± then the current assumptions, item (xi) in Theorem 1.5.1,
and [68, Proposition 8.4.9] imply that
⎧
⎪ M
u± ∈ 𝒞∞ (Ω± ) , Lu± = 0 in Ω±,
⎪
⎪
⎪
⎪  
⎪
⎪ Nκ ∇u± ∈ L p (∂Ω, wσ), Nκε u± ∈ Lloc (∂Ω, wσ),
p
⎪
⎪
⎨
⎪
κ−n.t.
  κ−n.t. (1.8.129)
⎪
⎪ u , ∇u exist σ-a.e. on ∂Ω,
⎪
⎪
±
∂Ω
±
∂Ω
⎪
⎪
⎪
⎪
κ−n.t.
⎪ and u±
⎪ = c at σ-a.e. point on ∂Ω,
⎩ ∂Ω

for some constant c ∈ C M . Introduce w ±j := ∂j u± in Ω± with 1 ≤ j ≤ n − 1. Then

for each j ∈ {1, . . . , n − 1} we have
202 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

⎧
⎪ M
⎪
⎪ w ±j ∈ 𝒞∞ (Ω± ) , Lw ±j = 0 in Ω±,
⎪
⎪
⎨
⎪
Nκ w ±j ∈ L p (∂Ω, wσ), and (1.8.130)
⎪
⎪
⎪
⎪
κ−n.t.
⎪
⎪ w ±j = 0 at σ-a.e. point on ∂Ω.
⎩ ∂Ω

To justify the property in the last line of (1.8.130) write (with ν = (0, . . . , 0, −1) ∈ Rn
denoting the outward unit normal to Ω)
κ−n.t. κ−n.t. κ−n.t.

w ±j = (∂j u± ) = −νn (∂j u± )

∂Ω ∂Ω ∂Ω
κ−n.t. κ−n.t.

= ν j (∂n u± ) − νn (∂j u± )
∂Ω ∂Ω
 κ−n.t.
= ∂τ j n u± ∂Ω = 0 at σ-a.e. point on ∂Ω, (1.8.131)

with the final equality a consequence of (1.8.129) and [69, Proposition 11.3.2]. Since
the system L satisfies the Legendre-Hadamard (strong) ellipticity condition, from
(1.8.130) and the uniqueness result for the weighted Dirichlet Problem for L in the
half-space proved in [61] we conclude that w ±j = 0 in Ω± for each j ∈ {1, . . . , n − 1}.
Hence,

∂j u± = 0 in Ω± for each j ∈ {1, . . . , n − 1}. (1.8.132)

Pick now  αβ 
A = a jk 1≤ j,k ≤n ∈ AL (1.8.133)
1≤α,β ≤M

and define
 αβ 
A jk = a jk 1≤α,β ≤M ∈ C M×M for each j, k ∈ {1, . . . , n}. (1.8.134)

Thus, L = A jk ∂j ∂k and from the first line in (1.8.129) and (1.8.132) we conclude
that, on the one hand,

0 = Lu± = A jk ∂j ∂k u± = Ann ∂n ∂n u± in Ω± . (1.8.135)

On the other hand, the fact that L is weakly elliptic (itself a consequence of the
Legendre-Hadamard strong ellipticity condition [70, (1.3.4) in Definition 1.3.2])
implies that its M × M symbol (or characteristic) matrix

L(ξ) = −ξ j ξk A jk , ξ = (ξ j )1≤ j ≤n ∈ Rn, (1.8.136)

satisfies
det L(ξ)  0, ∀ξ ∈ Rn \ {0}. (1.8.137)
In particular, choosing ξ := en = (0, . . . , 0, 1) ∈ Rn
shows that Ann = −L(en ) is an
invertible M × M matrix. Keeping this in mind we then conclude from (1.8.135) that
1.8 Modified Boundary Layer Potential Operators 203

∂n ∂n u± in Ω± . (1.8.138)

In concert, (1.8.132) and (1.8.138) prove that ∇(∂n u± ) = 0 in Ω± , hence ∂n u± is a

constant in Ω± , say
∂n u± = c± ∈ C M in Ω± . (1.8.139)
Consequently, (1.8.132), (1.8.139), and the fact that ν = (0, . . . , 0, −1) ∈ Rn entail
κ−n.t. κ−n.t.

∂νAu± = ν j A jk (∂k u± ) := ±νn Ann (∂n u± )

∂Ω ∂Ω

= ±Ann c± at σ-a.e. point on ∂Ω. (1.8.140)

Since, as seen from (1.5.58) (keeping [68, (7.7.104)] in mind), f = −∂νAu+ − ∂νAu−
at σ-a.e. point on ∂Ω, we deduce from (1.8.140) that f is a constant function on
M
∂Ω. In view of the fact that f ∈ L p (∂Ω, wσ) this ultimately forces f = 0, as
desired. This establishes the right-pointing implication in (1.8.128) and the opposite
implication is trivial. 

Moving on, we take on the task of establishing basic functional analytic proper-
ties of the modified boundary-to-domain double layer potential operator acting on
homogeneous Sobolev spaces defined on the boundary of a UR domain.

Theorem 1.8.12 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Denote

by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω and
αβ 
abbreviate σ := H n−1 ∂Ω. In addition, for some M ∈ N, let A = ar s 1≤r,s ≤n be
1≤α,β ≤M
a complex coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.8.141)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Also, let E = (Eγβ )1≤γ,β ≤M be

the matrix-valued fundamental solution associated with L as in [70, Theorem 1.4.2].
In this setting, recall the modified version of the double layer operator Dmod acting
  M
on functions from L 1 ∂Ω, 1+σ(x) |x | n as in (1.8.6). Finally, fix some integrability
exponent p ∈ (1, ∞) along with an aperture parameter κ ∈ (0, ∞).
Then there exists some constant C = C(Ω, A, n, p, κ) ∈ (0, ∞) with the property
.p M
that for each function f ∈ L1 (∂Ω, σ) it follows that
M  
Dmod f ∈ 𝒞∞ (Ω) , L Dmod f = 0 in Ω,
  κ−n.t.   κ−n.t.
Dmod f ∂Ω and ∇Dmod f ∂Ω exist at σ-a.e. point on ∂Ω,
    
Nκ ∇Dmod f ∈ L p (∂Ω, σ) and Nκ ∇Dmod f  L p (∂Ω,σ) ≤ C f [ L. p (∂Ω,σ)] M .
1
(1.8.142)
.p M
In fact, for each function f ∈ L1 (∂Ω, σ) one has
204 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

κ−n.t. 1 
(Dmod f ) ∂Ω = 2I + Kmod f at σ-a.e. point on ∂Ω, (1.8.143)
.p M
where I is the identity operator on L1 (∂Ω, σ) , and Kmod is the modified boundary-
to-boundary double layer potential operator from (1.8.24)-(1.8.25). Also,
.p M .   M
if p > n − 1 then the operator Dmod : L1 (∂Ω, σ) → 𝒞η Ω is
well defined, linear, and bounded, with η := 1 − n−1
p ∈ (0, 1), provided
either Ω ⊆ Rn is an open set satisfying a two-sided local John condi- (1.8.144)
tion and whose boundary is Ahlfors regular, or Ω is simultaneously a
uniform domain and a UR domain in Rn .

Moreover, given any function f = ( fα )1≤α ≤M belonging to the homogeneous

.p M
boundary Sobolev space L1 (∂Ω, σ) , at σ-a.e. point x ∈ ∂Ω one has
 
∂νA(Dmod f ) (x) (1.8.145)
# ∫ $
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0
y ∈∂Ω 1≤μ ≤M
|x−y |>ε

where the conormal derivative is considered as in (A.0.184).

Furthermore, the operator
.p M M
∂νA Dmod : L1 (∂Ω, σ) −→ L p (∂Ω, σ) defined as
 A . p M
(1.8.146)
∂ν Dmod ) f := ∂νA(Dmod f ) for each f ∈ L1 (∂Ω, σ)

is well defined, linear, and bounded, when the domain space is equipped with the
semi-norm induced by (A.0.128). In addition, for each q ∈ [1, ∞), the operator
(1.8.146) is an extension of the assignment (cf. [69, (11.5.58)] and (1.5.30)-(1.5.31))
q, p M   M
L1 (∂Ω, σ)  f −→ ∂νA D f := ∂νA(D f ) ∈ L p (∂Ω, σ) . (1.8.147)

Finally,
.p 5 M M
∂νA Dmod : L1 (∂Ω, σ) ∼ −→ L p (∂Ω, σ) defined as
.p M
(1.8.148)
∂νA Dmod [ f ] := ∂νA(Dmod f ) for each f ∈ L1 (∂Ω, σ)

is a well-defined, linear, and bounded operator, when the quotient space is equipped
with the natural semi-norm25 introduced in [69, (11.5.138)].

25 [69, Proposition 11.5.14] tells us that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
1.8 Modified Boundary Layer Potential Operators 205
.p M
Proof For each f ∈ L1 (∂Ω, σ) the jump-formula (1.8.143) is seen from (1.8.27)
(keeping in mind (A.0.127)). The claims in (1.8.142) are consequences of (A.0.127),
(1.8.7), (1.8.27), (1.8.20), (1.8.13), [70, (2.4.9)], and [70, Theorem 1.4.2]. Also, the
claim in (1.8.144) is a consequence of (1.8.142) and [68, Corollary 8.6.8] in the case
when Ω is simultaneously a uniform domain and a UR domain, and is seen from [69,
(11.5.60), (11.5.168), (11.5.201)] together with (1.8.15) in the case when Ω ⊆ Rn
is an open set satisfying a two-sided local John condition and whose boundary
is Ahlfors regular (in which scenario, Ω is known to be a UR domain; see [68,
(5.11.27)]).
Going further, formula (1.8.145) is a direct consequence of (1.8.83) and [69,
Definition 11.5.3] (cf. (A.0.127)-(A.0.128)). In turn, having established (1.8.145),
the claims made in relation to (1.8.146) follow with the help of [70, Theorem 2.3.2]
and [70, Theorem 1.4.2]. Also, the fact that the operator (1.8.146) is an extension of
the assignment (1.8.147) is clear from [69, (11.5.58)], (1.5.30)-(1.5.31), and (1.5.29).
Finally, the claims pertaining to (1.8.148) are consequences of what we have
proved so far and (1.8.10), which implies that ∂νA Dmod annihilates constants. 

Remark 1.8.13 The results in Theorem 1.8.12 are applicable to all modified
boundary-to-boundary double layer potential operators Dmod , Kmod described in
Examples 1.8.4-1.8.7.
Moreover, the operator ∂νA Dmod from (1.8.146) vanishes identically when Dmod is
as in (1.8.108) (see the last part in Remark 1.4.6). The same the peculiarity (i.e.,
that the operator ∂νA Dmod from vanishes identically) is present when Dmod is as in
(1.8.111) (see (1.4.33)).

Our next result deals with the modified boundary-to-boundary double layer po-
tential operator Kmod associated with a second-order weakly elliptic M × M system
L and a domain Ω ⊆ Rn satisfying suitable assumptions. The aim here is to iden-
.p M
tify geometric settings in which the homogeneous Sobolev space L1 (∂Ω, σ) is
invariant under the action of Kmod .

Theorem 1.8.14 Let Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) be a UR domain, and

αβ
abbreviate σ := H n−1 ∂Ω. Also, let L = ar s ∂r ∂s 1≤α,β ≤M be a homogeneous,
weakly elliptic, constant (complex) coefficient, second-order M × M system in Rn (for
some M ∈ N). In this context, consider the modified boundary-to-boundary double
layer potential operator Kmod from (1.8.24)-(1.8.25). Finally, select an integrability
exponent p ∈ (1, ∞). Then the following statements are true:

(1) If Ω is also assumed to satisfy a local John condition, then the operator
.p p M .p p M
Kmod : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) −→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
(1.8.149)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). As a consequence of this and (1.8.28),
206 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 . 6 M
p p
Kmod : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼
 . 6 M
p p
−→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼ (1.8.150)

defined as
 . 6 M
p p
Kmod [ f ] := Kmod f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼
.p (1.8.151)
p M
for each function f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)

is also a well-defined linear and bounded operator, when the quotient spaces are
equipped with the natural semi-norm introduced in [69, (11.5.138)]. In addition,
with U jk for j, k ∈ {1, . . . , n} denoting the family of singular integral operators
defined in (1.5.251), one has
 
∂τ j k Kmod f = K(∂τ j k f ) + U jk (∇tan f ) at σ-a.e. point on ∂Ω
.p p M
for each f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)and each j, k ∈ {1, . . . , n}.
(1.8.152)
(2) Impose the stronger assumption that Ω is an NTA domain with an Ahlfors
regular boundary26. Then the operator
.p M .p M
Kmod : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (1.8.153)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). As a corollary of (1.8.153) and (1.8.28), the following
is a well-defined linear and bounded27 operator:
.p 5 M .p 5 M
Kmod : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
.p 5 M . p M
Kmod [ f ] := Kmod f ∈ L1 (∂Ω, σ) ∼ , ∀ f ∈ L1 (∂Ω, σ) .
(1.8.154)
(3) Strengthen the original hypotheses by now assuming that Ω is an open set in
Rn satisfying a two-sided local John condition and whose boundary is Ahlfors
regular28. Then, if U jk with j, k ∈ {1, . . . , n} is the family of singular integral
operators defined in (1.5.251), one has
 
∂τ j k Kmod f = K(∂τ j k f ) + U jk (∇tan f ) at σ-a.e. point on ∂Ω
.p M
(1.8.155)
for each f ∈ L1 (∂Ω, σ) and each j, k ∈ {1, . . . , n}.

26 a scenario in which Ω is known to be a UR domain; cf. [68, (5.11.5)]

27 if ∂Ω is unbounded then the semi-norm introduced in [69, (11.5.138)] endowing the quotient
spaces in (1.8.154) is actually a genuine norm and the homogeneous Sobolev spaces of order one,
modulo constants, becomes Banach spaces; see [69, Proposition 11.5.14]
28 in which scenario, Ω is known to be a UR domain; see [68, (5.11.27)]
1.8 Modified Boundary Layer Potential Operators 207

It has been shown in [103] that

if Ω ⊆ Rn is an open set satisfying a two-sided corkscrew condition
and whose boundary is Ahlfors regular, then Ω satisfies the local John (1.8.156)
condition if and only if Ω satisfies the local Harnack chain condition.
As a consequence of (1.8.156) and the fact that the local John condition implies the
corkscrew condition we then obtain the following result:
if Ω ⊆ Rn is an open set whose boundary is Ahlfors regular, then
Ω satisfies a two-sided local John condition if and only if Ω is a (1.8.157)
two-sided NTA domain.
In view of this, we conclude that if Ω ⊆ Rn is an open set satisfying a two-sided local
John condition and whose boundary is Ahlfors regular then the operator Kmod is still
well defined, linear, bounded in the context of (1.8.153), and (1.8.154) continues to
be a well-defined linear and bounded operator.
Let us now present the proof of Theorem 1.8.14.
Proof of Theorem 1.8.14 Pick an aperture parameter κ ∈ (0, ∞). First, work under
the assumption that Ω ⊆ Rn is a UR domain satisfying a local John condition.
Consider an arbitrary function
.p p M
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) (1.8.158)
M
and define u := Dmod f in Ω. Then u ∈ 𝒞∞ (Ω) (cf. (1.8.7)), and the jump-formula
(1.8.143) gives
κ−n.t. 1 
u ∂Ω = 2I + Kmod f at σ-a.e. point on ∂Ω. (1.8.159)

From (1.8.142) we also know that

 
Nκ (∇u) ∈ L p (∂Ω, σ) and Nκ (∇u) L p (∂Ω,σ) ≤ C f [ L. p (∂Ω,σ)] M (1.8.160)
1

for some constant C ∈ (0, ∞) independent of f . Additionally, from (1.8.22) (whose

applicability in the present setting is guaranteed by (1.8.158) and the current geo-
metric assumptions on Ω) we know that for each truncation parameter ε ∈ (0, ∞) we
have
Nκε u ∈ Lloc (∂Ω, σ).
p
(1.8.161)
In view of these properties, we may reason as in the second half of the proof of [69,
Proposition 11.5.12] and obtain that
κ−n.t. .p p M
u ∂Ω belongs to the space L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
 κ−n.t.  (1.8.162)
and u ∂Ω [ L. p (∂Ω,σ)] M ≤ C f [ L. p (∂Ω,σ)] M
1 1
208 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

for some constant C ∈ (0, ∞) independent of f . In concert with (1.8.159), this

ultimately shows that
.p p M
Kmod f belongs to the space L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
  (1.8.163)
and Kmod f [ L. p (∂Ω,σ)] M ≤ C f [ L. p (∂Ω,σ)] M ,
1 1

from which the claims pertaining to (1.8.149) follow. In turn, the claims regarding
the operator (1.8.154) are readily seen from what we have just proved, (1.8.28), and
definitions. Finally, formula (1.8.155) is implied by Proposition 1.8.8 (bearing in
mind (1.8.158) and [69, Definition 11.5.3]; cf. (A.0.127)-(A.0.128)). This takes care
of item (1).
To deal with the claims in item (2), in place of the original geometric assumptions
let us now assume that Ω ⊆ Rn is an NTA domain with an upper Ahlfors regular
boundary. These hypotheses then guarantee (cf. [68, (5.10.24)]) that Ω is a UR
.p M
domain. Pick an arbitrary function f ∈ L1 (∂Ω, σ) , and define u := Dmod f in
Ω. Then u continues to enjoy the same properties as before, up to (and including)
(1.8.160). Granted these properties, we may invoke [69, Proposition 11.5.12] to
conclude that .p M
Kmod f belongs to the space L1 (∂Ω, σ)
  (1.8.164)
and Kmod f [ L. p (∂Ω,σ)] M ≤ C f [ L. p (∂Ω,σ)] M ,
1 1

for some constant C ∈ (0, ∞) independent of f . From this point on, the proof
proceeds as before.
As regards item (3), work under the stronger assumptions that Ω is an open set
in Rn satisfying a two-sided local John condition and whose boundary is Ahlfors
regular (as noted in [68, (5.11.27)], these hypotheses guarantee that Ω is a UR
domain). Then the desired result is a direct consequence of what we have proved in
item (1) and [69, Lemma 11.5.4]. 

Remark 1.8.15 The results in Theorem 1.8.14 are applicable to all modified
boundary-to-boundary double layer potential operators Kmod described in Exam-
ples 1.8.4-1.8.7.

The following result augments Proposition 1.2.1, through the consideration of

homogeneous Sobolev spaces. More general results of this type are contained in
item (9) of Theorem 5.1.1.

Proposition 1.8.16 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain. Abbreviate

σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω. Also, consider a complex-valued function b ∈ Lloc 1 (Rn, L n )

such that b Rn \{0} ∈ 𝒞 (R \ {0}) where N = N(n) ∈ N is a sufficiently large

N n

number, and with the property that ∇b is odd and positive homogeneous of degree
1 − n in Rn \ {0}. For each pair of indices j, k ∈ {1, . . . , n} introduce the modified
1.8 Modified Boundary Layer Potential Operators 209

boundary-to-domain “tangential”
 integral operator (compare with (1.2.1)) acting
σ(y)
on each function f ∈ L ∂Ω, 1+ |y | n−1 according to
1

∫ 
mod
Tjk f (x) := ν j (y) (∂k b)(x − y) − νk (y)(∂j b)(x − y) (1.8.165)
∂Ω
.
− (∂k b)(−y) − νk (y)(∂j b)(−y) · 1Rn \B(0,1) (y) f (y) dσ(y)

at each x ∈ Ω. Finally, fix an aperture parameter κ > 0 and pick an integrability

exponent p ∈ (1, ∞).
Then there exists some .constant C = C(Ω, b, n, p, κ) ∈ (0, ∞) with the property
p
that for each function f ∈ L1 (∂Ω, σ) it follows that
  mod  
Tjk f ∈ 𝒞 N −1 (Ω), Nκ ∇Tjk f  L p (∂Ω,σ) ≤ C f  L. p (∂Ω,σ),
mod

1
 mod  κ−n.t.  mod  κ−n.t.
(1.8.166)
Tjk f ∂Ω and ∇Tjk f ∂Ω exist at σ-a.e. point on ∂Ω.
.p
In fact, for each function f ∈ L1 (∂Ω, σ) one has
mod κ−n.t. mod
(Tjk f ) ∂Ω = Tjk f at σ-a.e. point on ∂Ω, (1.8.167)
mod
where Tjk is the modified version of the operator Tjk from (1.2.2). Specifically, for
each function
 σ(x)
f ∈ L 1 ∂Ω, (1.8.168)
1 + |x| n
one defines
∫ 
mod  
Tjk f (x) := lim+ ν j (y) Bk,ε (x − y) − Bk,1 (−y) (1.8.169)
ε→0
∂Ω
.
 
− νk (y) B j,ε (x − y) − B j,1 (−y) f (y) dσ(y)

at σ-a.e. point x ∈ ∂Ω, where

B,ε := (∂ b) · 1Rn \B(0,ε) for each ∈ {1, . . . , n} and ε > 0. (1.8.170)

Under the additional assumption that Ω satisfies a local John condition, the
operator
mod .p p .p p
Tjk : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) −→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) (1.8.171)

is well defined, linear, and bounded, when the above spaces are equipped with the
semi-norm (A.0.128).
210 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Finally, if the given set Ω is actually an NTA domain with an Ahlfors regular
boundary29 then the operator
mod .p .p
Tjk : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (1.8.172)

is well defined, linear, and bounded, when the spaces involved are endowed with the
semi-norm (A.0.128).
Proof The properties listed in (1.8.166) are justified by arguing as in the proof of
(1.8.142). With (1.8.166) in hand, the fact that (1.8.171)-(1.8.172) are well-defined,
linear, and bounded operators is then established much as in the proof of (1.8.149)
and (1.8.153), respectively. 
The boundary-to-domain version of the modified double layer potential operator,
along with the boundary-to-domain version of the modified single layer potential
operator, play a basic role in the formulation of the following fundamental integral
representation result.
Theorem 1.8.17 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the
property that ∂Ω is an Ahlfors regular set; in particular, Ω is a set of locally finite
perimeter. Denote by ν the geometric measure theoretic outward unit normal to Ω
αβ
and abbreviate σ := H n−1 ∂Ω. Also, for some M ∈ N, let A = ar s 1≤r,s ≤n be a
1≤α,β ≤M
complex coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.8.173)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . In this setting, recall

the modified version of the double layer operator Dmod acting on functions from
  M
L 1 ∂∗ Ω, 1+σ(x)
|x | n as in (1.8.6), and the modified version of the single layer oper-
  M
ator 𝒮mod acting on functions from L 1 ∂Ω, 1+σ(x) |x | n−1
as in (1.5.50). Finally, fix
an aperture parameter κ ∈ (0, ∞), a truncation parameter ε ∈ (0, ∞), and consider
a function u : Ω → C M satisfying
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκε u ∈ Lloc
1 (∂Ω, σ),

κ−n.t. κ−n.t.    M
u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n , (1.8.174)
κ−n.t.  σ(x)
(∇u) ∂Ω exists σ-a.e. on ∂nta Ω and Nκ (∇u) ∈ L 1 ∂Ω, .
1 + |x| n−1
In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.175)
B(0,λ R)\B(0,R)

29 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
1.8 Modified Boundary Layer Potential Operators 211

Then the conormal derivative ∂νAu, extended to the entire topological boundary
  M
by setting it to be zero outside ∂∗ Ω, belongs to L 1 ∂Ω , 1+σ(x)
|x | n−1
and there exists
some C -valued locally constant function cu in Ω with the property that
M

 κ−n.t.   
u = Dmod u ∂Ω − 𝒮mod ∂νAu + cu in Ω. (1.8.176)

In addition, if Dmod , 𝒮mod are now regarded as operators mapping into functions
defined in Rn \ Ω, then
 κ−n.t.   
Dmod u ∂Ω − 𝒮mod ∂νAu is a locally constant function in Rn \ Ω. (1.8.177)

Moreover, if in place of the second line in (1.8.174) one assumes that

κ−n.t. κ−n.t.    M
u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1 , (1.8.178)

then in place of (1.8.176) one may conclude that (again, for some C M -valued locally
constant function cu in Ω, and with the same caveat in the case when Ω is an exterior
domain)
 κ−n.t.   
u = D u ∂Ω − 𝒮mod ∂νAu + cu in Ω, (1.8.179)
where D is the “ordinary” double layer operator, associated with Ω and A as in
(1.3.18).
Also, if the last line in (1.8.174) is strengthened to
κ−n.t.  σ(x)
(∇u) ∂Ω exists σ-a.e. on ∂nta Ω and Nκ (∇u) ∈ L 1 ∂Ω, (1.8.180)
1 + |x| n−2

(with the weight (1 + |x| n−2 )−1 replaced by ln(2 + |x|) if n = 2) then in place of
(1.8.176) one may now conclude (once more, for some C M -valued locally constant
function cu in Ω, and with the imposition of the decay condition (1.8.175) when Ω
is an exterior domain) that
 κ−n.t.   
u = Dmod u ∂Ω − 𝒮 ∂νAu + cu in Ω, (1.8.181)

where 𝒮 is the “ordinary” single layer operator, associated with Ω and L as in

(1.3.6).
Finally, if in place of the second line in (1.8.174) one now assumes (1.8.178), and
in place of the last line in (1.8.174) one now assumes (1.8.180), then one concludes
(with the same notation and conventions as before) that
 κ−n.t.   
u = D u ∂Ω − 𝒮 ∂νAu + cu in Ω. (1.8.182)

It is of significance to observe that the locally function cu appearing in the

statement of Theorem 1.8.17 is actually a genuine constant (vector in C M ) when ∂Ω
is compact. Indeed, when Ω is bounded this follows from Theorem 1.3.3, (6.2.175),
212 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

and (1.5.56), while if Ω is an exterior domain this is implied by [70, Theorem 1.5.7]
(since (1.8.176) shows that u is bounded at infinity).
In addition, the locally constant function cu intervening in the statement of Theo-
rem 1.8.17 is in fact a genuine constant (vector in C M ) whenever ∂Ω is unbounded and
 
Nκ u ∈ L 1 ∂Ω , 1+σ(x)
|x | n . To justify this, fix two arbitrary distinct points x0, x1 ∈ Ω
along with an arbitrary index γ ∈ {1, . . . , n}. The idea is to employ the Diver-
gence Theorem in the version recorded in [68, Theorem 1.4.1] for the vector field
F = (Fs )1≤s ≤n with components given for each s ∈ {1, . . . , n} by
 
βα
Fs (y) := ar s (∂r Eγβ (x0 − y) − (∂r Eγβ (x1 − y) uα (y)
 
αβ
+ Eγα (x0 − y) − Eγα (x1 − y) asr (∂r uβ )(y) (1.8.183)

at L n -a.e. y ∈ Ω. This gives that the γ-th component of the vector cu (x0 ) − cu (x1 )
is zero, from which the desired conclusion readily follows.
Let us turn now to the proof of Theorem 1.8.17.
Proof of Theorem 1.8.17 Work under the assumption that u = (uβ )1≤β ≤M is as in
(1.8.174). Since the current hypotheses imply Nκε u, Nκε (∇u) ∈ Lloc
1 (∂Ω, σ), from
κ−n.t.
(1.8.174) and [69, Proposition 11.3.2] we then conclude that the function f := u ∂Ω
M
considered on ∂∗ Ω (cf. [68, (8.8.52)]) belongs to L1,loc
1 (∂ Ω, σ)
∗ and satisfies
 κ−n.t.  κ−n.t.
∂τ j k f = ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω
(1.8.184)
at σ-a.e. point on ∂∗ Ω, for each j, k ∈ {1, . . . , n},

where (ν1, . . . , νn ) are the scalar components of the geometric measure theoretic
outward unit normal ν to Ω. In concert with (1.8.174) this also entails
  σ(x)  M   σ(x) M
f ∈ L 1 ∂∗ Ω, and ∂τ f ∈ L 1
∂∗ Ω,
1 + |x| n jk
1 + |x| n−1 (1.8.185)
for each pair of indices j, k ∈ {1, . . . , n}.

To proceed, abbreviate (cf. (A.0.184))

   κ−n.t.
αβ
g := ∂νAu = νr ar s ∂s uβ ∂Ω at σ-a.e. point on ∂∗ Ω, (1.8.186)
1≤α ≤M

and note that (1.8.174) together with [68, (8.8.52), (8.9.8), (8.9.44)] ensure that
  σ(x) M
g ∈ L 1 ∂∗ Ω, . (1.8.187)
1 + |x| n−1
Going further, define

w := Dmod f − 𝒮mod g ∈ 𝒞∞ (Ω)

M
, (1.8.188)
1.8 Modified Boundary Layer Potential Operators 213

and denote by (wγ )1≤γ ≤M the scalar components of w. Also, write ( fα )1≤α ≤M and
(gα )1≤α ≤M for the scalar components of f and g, respectively. Finally, denote by
E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental solution associated with L as in
[70, Theorem 1.4.2]. Then for each index ∈ {1, . . . , n}, each index γ ∈ {1, . . . , M },
and each point x ∈ Ω we may compute
   
(∂ wγ )(x) = ∂ Dmod f γ (x) − ∂ 𝒮mod g γ (x)
∫
βα
= ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ(y)
∂∗ Ω
∫
− (∂ Eγα )(x − y)gα (y) dσ(y)
∂∗ Ω
∫
βα
= ar s (∂r Eγβ )(x − y)×
∂∗ Ω

  κ−n.t.   κ−n.t.  
× ν (y) (∂s uα ) ∂Ω (y) − νs (y) (∂ uα ) ∂Ω (y) dσ(y)
∫
αβ  κ−n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ ) ∂Ω (y) dσ(y)
∂∗ Ω

= (∂ uγ )(x), (1.8.189)

where the first equality above comes from (1.8.188), the second equality uses
(1.8.12)-(1.8.13), (1.8.185), (1.5.51), (1.8.187), the third equality utilizes (1.8.184),
and the final equality is provided by [70, (1.5.230)] (bearing in mind the properties
in (1.8.174)). From (1.8.189) we then conclude that ∇w = ∇u in Ω, which goes to
show that the difference cu := u − w is a C M -valued locally constant function in Ω.
The proof of (1.8.176) is therefore complete. The claim in (1.8.177) is established
similarly, now making use of [70, (1.5.232)].
Next, that (1.8.179) holds under the assumptions made in (1.8.178) in place of
the second line in (1.8.174) is implied by (1.8.176) and (1.8.8). Finally, (1.8.181)
and (1.8.182) are dealt with similarly, now also taking into account (1.5.56). 
Here is a version of Theorem 1.8.17 in which no size conditions are explicitly
imposed on the nontangential boundary trace of the function in question.

Corollary 1.8.18 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an Ahlfors regular domain

satisfying a local John condition. Abbreviate σ := H n−1 ∂Ω and denote by ν
the geometric
 αβ  measure theoretic outward unit normal to Ω. For some M ∈ N, let
A = ar s 1≤r,s ≤n be a complex coefficient tensor with the property that
1≤α,β ≤M

 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.8.190)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . In this setting, recall

the modified version of the double layer operator Dmod acting on functions from
214 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
 
L 1 ∂Ω, 1+σ(x)
M
|x | n as in (1.8.6), and the modified version of the single layer operator
  M
𝒮mod acting on functions from L 1 ∂Ω, 1+σ(x)|x | n−1
as in (1.5.50). Finally, fix an
aperture parameter κ ∈ (0, ∞), a truncation parameter ε ∈ (0, ∞), and consider a
function u : Ω → C M satisfying:
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,

Nκε u ∈ Lloc (∂Ω, σ) for some q ∈ [1, ∞],

q

(1.8.191)
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ (1, ∞),
κ−n.t. κ−n.t.
u ∂Ω and (∇u) ∂Ω exist σ-a.e. on ∂Ω.

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.192)
B(0,λ R)\B(0,R)

Then
κ−n.t. .p q, p M
u ∂Ω belongs to L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ,
(1.8.193)
M
∂νAu belongs to L p (∂Ω, σ) ,

and there exists some C M -valued locally constant function cu in Ω with the property
that  κ−n.t.   
u = Dmod u ∂Ω − 𝒮mod ∂νAu + cu in Ω. (1.8.194)
In addition, if Dmod , 𝒮mod are now regarded as operators mapping into functions
defined in Rn \ Ω, then
 κ−n.t.   
Dmod u ∂Ω − 𝒮mod ∂νAu is a locally constant function in Rn \ Ω. (1.8.195)

Proof Granted the current geometric assumptions, [69, Proposition 11.5.13] applies
and guarantees that the first property claimed in (1.8.193) holds. That the second
property claimed in (1.8.193) is a consequence of (1.8.191) and [70, Lemma 1.7.3].
Granted these, we then see that all hypotheses of Theorem 1.8.17 are currently
satisfied, so (1.8.194) is implied by (1.8.176), and (1.8.195) is implied by (1.8.177).

We continue by presenting a version of the integral representation formula from

Theorem 1.8.17 in an NTA domain with an Ahlfors regular boundary, for a null-
solution of a weakly elliptic system such that the nontangential maximal operator
of its gradient belongs to a Lebesgue space (a feature, ensuring that the nontan-
gential boundary trace of said function belongs a homogeneous Sobolev space). A
remarkable aspect of this result is the lack of explicit demands on the nontangential
maximal operator of the function itself.
1.8 Modified Boundary Layer Potential Operators 215

Theorem 1.8.19 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain such

that ∂Ω is an Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and denote by ν
 αβ unit normal to Ω. Also, fix an aperture
the geometric measure theoretic outward
parameter κ ∈ (0, ∞). Next, let A = ar s 1≤r,s ≤n (where M ∈ N) be a complex
1≤α,β ≤M
coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.8.196)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Recall the modified version

  M
of the double layer operator Dmod acting on functions from L 1 ∂Ω, 1+σ(x) |x | n as in
(1.8.6), and the modified version of the single layer operator 𝒮mod acting on functions
  M
from L 1 ∂Ω, 1+σ(x) |x | n−1
as in (1.5.50). Finally, consider a function u : Ω → C M
satisfying
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ).
(1.8.197)
for some integrability exponent p ∈ (1, ∞),

and, in the case when Ω is an exterior domain, make the additional assumption that
there exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.198)
B(0,λ R)\B(0,R)

Then
κ−n.t. .p p M
u ∂Ω exists σ-a.e. on ∂Ω, belongs to L1 (∂Ω, σ) ∩ L1,loc (∂Ω, σ) ,
κ−n.t. M
(∇u) ∂Ω exists σ-a.e. on ∂Ω, and ∂νAu belongs to L p (∂Ω, σ),
(1.8.199)
and there exists some C M -valued locally constant function cu in Ω with the property
that  κ−n.t.   
u = Dmod u ∂Ω − 𝒮mod ∂νAu + cu in Ω. (1.8.200)
In addition, if Dmod , 𝒮mod are now regarded as operators mapping into functions
defined in Rn \ Ω, then
 κ−n.t.   
Dmod u ∂Ω − 𝒮mod ∂νAu is a locally constant function in Rn \ Ω. (1.8.201)

Moreover, if p ∈ (1, n − 1) (which implicitly forces n ≥ 3) then there exists a

constant c ∈ C M such that
κ−n.t. ∗ M ∗
u ∂Ω − c ∈ L p (∂Ω, σ) and Nκ (u − c) ∈ L p (∂Ω, σ)
(1.8.202)
1 
1 −1
where p∗ := p − n−1 ∈ (1, ∞).
216 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

As a consequence, if p ∈ (1, n − 1) (again, this implicitly forces n ≥ 3) and p∗ is as

above then / 0M
κ−n.t.  σ(x)
p∗
u ∂Ω ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, (1.8.203)
1 + |x| n
and 
p∗ σ(x)
Nκ u ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, . (1.8.204)
1 + |x| n
κ−n.t.
Proof From the current assumptions and [69, Proposition 11.5.12] we see that u ∂Ω
.p p M
exists at σ-a.e. point on ∂Ω and belongs to L1 (∂Ω, σ) ∩ L1,loc (∂Ω, σ) . From
[68, Proposition 8.4.9] we also know that there exists some small ε > 0 such that

Nκε u ∈ Lloc (∂Ω, σ).

p
(1.8.205)

Next, the present hypotheses on Ω ensure (cf. [68, (5.10.24)]) that ∂Ω is a

UR set and Ω satisfies a two-sided corkscrew condition. In particular, Ω is a UR
domain. Keeping this in mind, the Fatou-type result from [70, Theorem 3.3.4]
κ−n.t.
guarantees the nontangential boundary trace (∇u) ∂Ω exists (in Cn·M ) at σ-a.e.
M
point on ∂Ω. In particular, ∂νAu is well defined and belongs to L p (∂Ω, σ) (cf.
[70, Lemma 1.7.3]). Hence, all conditions in (1.8.174) are satisfied, and this permits
us to invoke Theorem 1.8.17 to conclude that (1.8.200) and (1.8.201) hold.
For the remainder of the proof assume that p ∈ (1, n − 1) (hence necessarily
 
1 −1
n ≥ 3), and set p∗ := p1 − n−1 . Collectively, (1.8.205), [68, (8.9.8)], and the last
condition in (1.8.197) imply that the function g defined as in [69, (11.5.71)] belongs
to L p (∂Ω, σ). Having established this, from [68, (5.11.28)] and [69, Lemmas 11.5.8,
11.5.9] we infer the existence of a constant c ∈ C M such that the function
κ−n.t. ∗ M
u ∂Ω − c belongs to L p (∂Ω, σ) . (1.8.206)

In particular,
/ 0M
κ−n.t.  σ(x)
p∗ M p∗
u ∂Ω ∈ L (∂Ω, σ) + C M → Lloc (∂Ω, σ) ∩ L 1 ∂Ω, . (1.8.207)
1 + |x| n

Also, with c ∈ C M as above we may rely on (1.8.206), (1.8.8), and (1.8.10) to write
 κ−n.t.   κ−n.t. 
Dmod u ∂Ω = D u ∂Ω − c + c(1) in Ω. (1.8.208)

for some constant c(1) ∈ C M . To proceed, recall from the second line in (1.8.199)
M
that ∂νAu belongs to the space L p (∂Ω, σ) with p ∈ (1, n − 1). Based on this and
(1.5.56) we then conclude that there exists some constant c(2) ∈ C M such that

𝒮mod (∂νAu) = 𝒮(∂νAu) + c(2) in Ω. (1.8.209)

1.8 Modified Boundary Layer Potential Operators 217

Collectively, (1.8.208), (1.8.209), (1.8.200) imply that there exists a C M -valued

locally constant function 
cu in Ω (namely cu := c(1) − c(2) ) with the property that
 κ−n.t.   
u = D u ∂Ω − c − 𝒮 ∂νAu + 
cu in Ω. (1.8.210)

We next distinguish two cases. Assume first that ∂Ω is unbounded. Then [68,
Lemma 5.11.3] gives that Ω is connected so  cu in (1.8.210) is actually a genuine
constant, say cu ≡ co ∈ C M . Keeping this in mind, (1.8.210) implies
  κ−n.t.    
Nκ (u − co ) ≤ Nκ D u ∂Ω − c + Nκ 𝒮 ∂νAu (1.8.211)

at each point on ∂Ω. From (1.5.1), (2.2.39), and [68, (3.6.27)] we know that
  κ−n.t.  ∗
   ∗
Nκ D u ∂Ω − c ∈ L p (∂Ω, σ) and Nκ 𝒮 ∂νAu ∈ L p (∂Ω, σ). (1.8.212)

By combining (1.8.211), (1.8.212), and [68, (8.2.28)] we arrive at the conclusion

that
∗
Nκ (u − co ) ∈ L p (∂Ω, σ). (1.8.213)
κ−n.t. ∗ M
In view of [68, (8.9.8)], this entails u ∂Ω − co ∈ L p (∂Ω, σ) which, in concert
with (1.8.206) and the fact that we are presently assuming ∂Ω to be unbounded, forces
co = c. Having established this, (1.8.202) follows from (1.8.206) and (1.8.213). In
turn, (1.8.202) readily implies (1.8.203) and (1.8.204). Finally, consider the case
when ∂Ω is bounded. Much as before, from (1.8.210) and (1.8.212) we see that
∗
Nκ (u − 
cu ) ∈ L p (∂Ω, σ). (1.8.214)

Since in the current scenario [68, Lemma 5.11.3] implies that Ω has finitely many
connected components, the range of the locally constant C M -valued function  cu is
finite. Keeping in mind that now ∂Ω has finite measure, (1.8.214) and (1.8.206) imply
∗ κ−n.t. ∗ M
that Nκ u ∈ L p (∂Ω, σ) and u ∂Ω ∈ L p (∂Ω, σ) . Thus, (1.8.202) presently holds
with c := 0, a scenario in which (1.8.203) and (1.8.204) readily follow from this. 

A result of the same flavor as the integral representation formula from The-
orem 1.8.17, but for null-solutions of an injectively elliptic first-order system, is
presented in the corollary below. This is akin a higher-dimensional Cauchy integral
representation formula, in a very broad geometric setting and under very general
analytic assumptions on the function involved.

Corollary 1.8.20 Let Ω ⊆ Rn , where n ≥ 2, be an open set with an Ahlfors regular

boundary and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous first-order
N × M system D with constant complex coefficients in Rn (where N, M ∈ N) which
is injectively elliptic (cf. [70, (1.3.18)]), and suppose D is a homogeneous first-order
M × N system with constant complex coefficients in Rn which complements D (i.e.,
[70, (1.3.21)] holds). In particular, L := DD  is a weakly elliptic second-order M × M
system in Rn . Let AD,D be the coefficient  of
tensor induced by the factorization DD
218 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

the system L, defined as in (1.4.32), and bring in the modified version of the double
  M
layer operator Dmod acting on functions from L 1 ∂∗ Ω, 1+σ(x)
|x | n as in (1.8.6) for
 . Finally, fix an aperture parameter κ ∈ (0, ∞) and
the coefficient tensor A := AD,D
suppose u : Ω → C M is a vector-valued function satisfying
M
u ∈ 𝒞∞ (Ω) , Nκε u ∈ Lloc
Du = 0 in Ω, 1 (∂Ω, σ),

κ−n.t. κ−n.t.   σ(x)  M

u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, , (1.8.215)
1 + |x| n
κ−n.t.  σ(x)
(∇u) ∂Ω exists σ-a.e. on ∂nta Ω and Nκ (∇u) ∈ L 1 ∂Ω, .
1 + |x| n−1
In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.216)
B(0,λ R)\B(0,R)

Then there exists some C M -valued locally constant function cu in Ω with the
property that
 κ−n.t. 
u = Dmod u ∂Ω + cu in Ω. (1.8.217)
In addition, if Dmod is now regarded as an operator mapping into functions defined
in Rn \ Ω, then
 κ−n.t. 
Dmod u ∂Ω is a locally constant function in Rn \ Ω. (1.8.218)

Furthermore, if in place of the second line in (1.8.215) one now assumes that
κ−n.t. κ−n.t.    M
u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
, (1.8.219)

then in place of (1.8.217) one may conclude that (again, for some C M -valued locally
constant function cu in Ω, and with the same caveat in the case when Ω is an exterior
domain)
 κ−n.t. 
u = D u ∂Ω + cu in Ω, (1.8.220)
where D is the “ordinary” double layer operator, associated with the set Ω and the
coefficient tensor A := AD,D
 as in (1.3.18).

Proof Upon observing that Lu = DDu  = 0 in Ω, we may invoke Theorem 1.8.17

for the coefficient tensor A := AD,D  . Since, as seen from (1.4.33), the conormal
derivative ∂νAu vanishes at σ-a.e. point on ∂∗ Ω (given that we are presently assuming
that Du = 0 in Ω), it follows that (1.8.176) reduces precisely to (1.8.217). Likewise,
(1.8.177) turns into (1.8.218), while (1.8.179) now becomes (1.8.220). 
1.8 Modified Boundary Layer Potential Operators 219

Here is a version of Corollary 1.8.20 for null-solutions of an injectively elliptic

first-order system having the nontangential maximal functions of their gradients in
a Lebesgue space.

Corollary 1.8.21 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain such

that ∂Ω is an Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and fix an aperture
parameter κ ∈ (0, ∞) along with an integrability exponent p ∈ (1, ∞). Next, consider
a homogeneous first-order N × M system D with constant complex coefficients in
Rn (where N, M ∈ N) which is injectively elliptic (cf. [70, (1.3.18)]), and suppose
 is a homogeneous first-order M × N system with constant complex coefficients
D
in Rn which complements D (i.e., [70, (1.3.21)] holds). In particular, L := DD  is
a weakly elliptic second-order M × M system in R . Let AD,D
n
 be the coefficient
tensor induced by the factorization DD  of the system L, defined as in (1.4.32), and
bring in the modified version of the double layer operator Dmod acting on functions
  M
from L 1 ∂Ω, 1+σ(x)
|x | n as in (1.8.6) for the coefficient tensor A := AD,D
 . Finally,
consider a function u : Ω → C M satisfying

u ∈ 𝒞∞ (Ω)
M
, Du = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ). (1.8.221)

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.222)
B(0,λ R)\B(0,R)

Then
κ−n.t. .p p M
u ∂Ω exists σ-a.e. on ∂Ω, belongs to L1 (∂Ω, σ) ∩ L1,loc (∂Ω, σ) ,
κ−n.t. M ·n
(∇u) ∂Ω exists σ-a.e. on ∂Ω and belongs to L p (∂Ω, σ) ,
(1.8.223)
and there exists some C M -valued locally constant function cu in Ω with the property
that  κ−n.t. 
u = Dmod u ∂Ω + cu in Ω. (1.8.224)
In addition, if Dmod is now regarded as an operator mapping into functions defined
in Rn \ Ω, then
 κ−n.t. 
Dmod u ∂Ω is a locally constant function in Rn \ Ω. (1.8.225)

Proof This follows from Corollary 1.8.20, much as the first main claim in Theo-
rem 1.8.19 was proved by relying on Theorem 1.8.17. 

It is of interest to specialize the integral representation formulas deduced above to

the two-dimensional setting. First, Corollary 1.8.20 used with the Cauchy-Riemann
operator in the complex plane yields the following result:
220 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Corollary 1.8.22 Let Ω ⊆ C ≡ R2 be an open set with the property that ∂Ω is an

Ahlfors regular set. Abbreviate σ := H 1 ∂Ω and consider the following modified
version of the boundary-to-domain Cauchy integral operator acting on each function
 σ(ζ)
f ∈ L 1 ∂∗ Ω, (1.8.226)
1 + |ζ | 2
according to
∫  1 
  1 1
𝒞mod f (z) := − 1C\B(0,1) (ζ) f (ζ) dζ at each z ∈ Ω. (1.8.227)
2πi ∂∗ Ω ζ − z ζ

Also, consider a holomorphic function u : Ω → C satisfying

Nκε u ∈ Lloc
1 (∂Ω, σ) for some κ ∈ (0, ∞) and ε ∈ (0, ∞),

κ−n.t. κ−n.t.  σ(ζ)

u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, ,
1 + |ζ | 2 (1.8.228)
κ−n.t.  σ(ζ)
(∇u) ∂Ω exists σ-a.e. on ∂nta Ω, and Nκ (∇u) ∈ L 1 ∂Ω, .
1 + |ζ |
In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL 2 = o(1) as R → ∞. (1.8.229)
B(0,λ R)\B(0,R)

Then there exists some complex-valued locally constant function cu in Ω with the
property that
 κ−n.t. 
u = 𝒞mod u ∂Ω + cu in Ω. (1.8.230)
In addition, if 𝒞mod is now regarded as an operator mapping into functions defined
in C \ Ω, then
 κ−n.t. 
𝒞mod u ∂Ω is a locally constant function in C \ Ω. (1.8.231)

Moreover, if in place of the second line in (1.8.228) one now assumes that
κ−n.t. κ−n.t.  σ(ζ ) 
u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, 1+ |ζ | , (1.8.232)

then in place of (1.8.230) one may conclude that (again, for some C M -valued locally
constant function cu in Ω, and with the same caveat in the case when Ω is an exterior
domain)
 κ−n.t. 
u = 𝒞 u ∂Ω + cu in Ω, (1.8.233)
where 𝒞 is the “ordinary” boundary-to-domain Cauchy integral operator (associ-
ated with Ω as in (1.6.35)).
1.8 Modified Boundary Layer Potential Operators 221

Proof In view of the identification made in the first part of Example 1.4.9, this
becomes a direct consequence of Corollary 1.8.20 used with D := ∂x + i∂y which,
up to normalization, is the Cauchy-Riemann operator in the complex plane (cf.
(1.6.39)). 

Second, Theorem 1.8.19 specialized to the Cauchy-Riemann operator in the com-

plex plane yields the following result:

Corollary 1.8.23 Let Ω ⊆ C ≡ R2 be an NTA domain such that ∂Ω is an Ahlfors reg-

ular set. Abbreviate σ := H 1 ∂Ω and recall the modified version of the boundary-to-
domain Cauchy integral operator 𝒞mod acting on functions (1.8.226) as in (1.8.227).
Fix an aperture parameter κ ∈ (0, ∞) along with an integrability exponent p ∈ (1, ∞)
and consider a holomorphic function u : Ω → C satisfying30 Nκ (u ) ∈ L p (∂Ω, σ).
In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|u  | dL 2 = o(1) as R → ∞. (1.8.234)
B(0,λ R)\B(0,R)

Then
κ−n.t. .p p
u ∂Ω exists σ-a.e. on ∂Ω and belongs to L1 (∂Ω, σ) ∩ L1,loc (∂Ω, σ),
(1.8.235)
κ−n.t. 2
u  ∂Ω exists σ-a.e. on ∂Ω and belongs to L p (∂Ω, σ) ,

and there exists some complex-valued locally constant function cu in Ω with the
property that
 κ−n.t. 
u = 𝒞mod u ∂Ω + cu in Ω. (1.8.236)
In addition, if 𝒞mod is now regarded as an operator mapping into functions defined
in C \ Ω, then
 κ−n.t. 
𝒞mod u ∂Ω is a locally constant function in C \ Ω. (1.8.237)

Proof This is justified by reasoning much as in the proof of Theorem 1.8.19 (used
with D the Cauchy-Riemann operator in the complex plane), now making use of
Corollary 1.8.22 in place of Theorem 1.8.17 (and keeping in mind that |∇u| = |u  |).

In the same vein, it is also of interest to specialize our general integral representa-
tion formulas established earlier for injectively elliptic first-order systems to the case
when of the classical Dirac operator. First, Corollary 1.8.20 implies the following
result:

Corollary 1.8.24 Let Ω ⊆ Rn , where n ≥ 2, be an open set with an Ahlfors regular

boundary. Denote by ν the geometric measure theoretic outward unit normal to

30 with ‘prime’ denoting the ordinary complex derivative

222 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Ω and abbreviate σ := H n−1 ∂Ω. Define the modified version of the boundary-to-
 
domain Cauchy-Clifford operator Cmod acting on functions f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n ⊗ C n
according to
∫  .
1 x−y y
Cmod f (x) := + 1 n (y)  ν(y)  f (y) dσ(y),
ωn−1 |x − y| n |y| n R \B(0,1)
∂∗ Ω
(1.8.238)

at each point x ∈ Ω. Finally, fix an aperture parameter κ ∈ (0, ∞) and recall the

classical Dirac operator D from (A.0.55). In this setting, suppose u : Ω → C n is a
Clifford algebra-valued function satisfying

u ∈ 𝒞∞ (Ω) ⊗ C n, Nκε u ∈ Lloc

Du = 0 in Ω, 1 (∂Ω, σ),

κ−n.t. κ−n.t.  σ(x)

u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, ⊗ C n,
1 + |x| n (1.8.239)
κ−n.t.  σ(x)
(∇u) ∂Ω exists σ-a.e. on ∂nta Ω and Nκ (∇u) ∈ L 1 ∂Ω, .
1 + |x| n−1
In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.240)
B(0,λ R)\B(0,R)

Then there exists some Clifford algebra-valued locally constant function cu in Ω

with the property that
 κ−n.t. 
u = Cmod u ∂Ω + cu in Ω. (1.8.241)
In addition, if Cmod is now regarded as an operator mapping into functions defined
in Rn \ Ω, then
 κ−n.t. 
Cmod u ∂Ω is a locally constant function in Rn \ Ω. (1.8.242)

Furthermore, if in place of the second line in (1.8.239) one now assumes that
κ−n.t. κ−n.t.  
u ∂Ω exists σ-a.e. on ∂nta Ω and u ∂Ω ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
⊗ C n, (1.8.243)

then in place of (1.8.241) one may conclude that (again, for some Clifford algebra-
valued locally constant function cu in Ω, and with the same caveat in the case when
Ω is an exterior domain)
 κ−n.t. 
u = C u ∂Ω + cu in Ω, (1.8.244)

where C is the “ordinary” Cauchy-Clifford operator, associated with the set Ω as in

(A.0.53).
1.8 Modified Boundary Layer Potential Operators 223

 and D are
Proof This is a direct consequence of Corollary 1.8.20 used when both D
the classical Dirac operator (A.0.55). 
Lastly, Corollary 1.8.21 written for the Dirac operator produces the following
result:
Corollary 1.8.25 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain such
that ∂Ω is an Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and fix an aperture
parameter κ ∈ (0, ∞) along with an integrability exponent p ∈ (1, ∞). Bring in the
modified version of the boundary-to-domain Cauchy-Clifford operator Cmod acting
 
on functions f ∈ L 1 ∂Ω, 1+σ(x)
|x | n ⊗ C n as in (1.8.238), and recall the classical
Dirac operator D from (A.0.55). Finally. consider a Clifford algebra-valued function
u : Ω → C n satisfying

u ∈ 𝒞∞ (Ω) ⊗ C n, Du = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ). (1.8.245)

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.8.246)
B(0,λ R)\B(0,R)

Then
κ−n.t.
the nontangential trace u ∂Ω exists σ-a.e. on ∂Ω,
κ−n.t. .p p 
u ∂Ω belongs to L1 (∂Ω, σ) ∩ L1,loc (∂Ω, σ) ⊗ C n, (1.8.247)
κ−n.t. n
(∇u) ∂Ω exists σ-a.e. on ∂Ω and belongs to L p (∂Ω, σ) ⊗ C n ,

and there exists some Clifford algebra-valued locally constant function cu in Ω with
the property that
 κ−n.t. 
u = Cmod u ∂Ω + cu in Ω. (1.8.248)
In addition, if Cmod is now regarded as an operator mapping into functions defined
in Rn \ Ω, then
 κ−n.t. 
Cmod u ∂Ω is a locally constant function in Rn \ Ω. (1.8.249)

 and
Proof This follows from Corollary 1.8.21 specialized to the case when both D
D are the classical Dirac operator (A.0.55). 
Singular integrals on rough surfaces do not constitute an algebra of operators. This
being said, there are some remarkable composition identities involving the modified
boundary-to-boundary double layer, its transpose version, the modified boundary-
to-boundary single layer, and the conormal derivative of the modified double layer.
These are made precise in Theorem 1.8.26 stated below (which should be compared
with item (xiii) of Theorem 1.5.1).
224 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

Theorem 1.8.26 Fix n ∈ N with n ≥ 2 and assume Ω ⊆ Rn is a UR domain. Denote

by ν the geometric measure theoretic outward
 αβ unit normal to Ω and abbreviate
σ := H n−1 ∂Ω. For some M ∈ N, let A = ar s 1≤r,s ≤n be a complex coefficient
1≤α,β ≤M
tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (1.8.250)

(i.e., L := L A) is a weakly elliptic M×M system in Rn . Having fixed some integrability

exponent p ∈ (1, ∞), recall the operators Smod from (1.8.123), together with ∂νA Dmod
from (1.8.146), and Kmod from (1.8.24). Finally, let K A#  be the operator associated
with the coefficient tensor A and the set Ω as in (1.3.72). Then the following
statements are true.
M
(1) For each function f ∈ L p (∂Ω, σ) , at σ-a.e. point on ∂Ω one has
1      
2I + K A#  − 12 I + K A#  f = ∂νA Dmod Smod f (1.8.251)

and there exists c f , which is the nontangential trace on ∂Ω of some C M -valued

locally constant function in Ω, with the property that
   
Smod K A#  f = Kmod Smod f + c f at σ-a.e. point on ∂Ω. (1.8.252)

(2) Impose the additional hypothesis that Ω satisfies a local John condition, and
.p p M
recall Kmod from (1.8.149). Then for each f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) one
has
    
K A#  ∂νA Dmod f = ∂νA Dmod Kmod f at σ-a.e. point on ∂Ω, (1.8.253)

and there exists c f , which is the nontangential trace on ∂Ω of some C M -valued

locally constant function in Ω, with the property that at σ-a.e. point on ∂Ω one
has
1   1   
2 I + Kmod − 2 I + Kmod f = Smod ∂νA Dmod f + c f . (1.8.254)

(3) If the original assumptions on the underlying domain are strengthened by now
asking that Ω is actually an NTA domain with an Ahlfors regular boundary31
then, with the operator Kmod as in (1.8.153), both formula (1.8.253) as well as
.p M
formula (1.8.254) actually hold for each function f ∈ L1 (∂Ω, σ) .

Proof The identity claimed in (1.8.251) is a consequence of (1.5.89), the fact that
the operator (1.8.123) is well defined, linear, and bounded, plus (1.8.145)-(1.8.146).
M
To prove (1.8.252), start with f ∈ L p (∂Ω, σ) and define u := 𝒮mod f in Ω.
Having fixed an aperture parameter κ ∈ (0, ∞) along with a truncation parameter

31 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
1.8 Modified Boundary Layer Potential Operators 225

ε ∈ (0, ∞), we conclude from (1.8.125) that u satisfies:

M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
Nκ (∇u) ∈ L p (∂Ω, σ),
 
Nκε u ∈ Lloc (∂Ω, σ) for each q ∈ 0, n−1
q
n−2 ,
κ−n.t.
(∇u) ∂Ω exists at σ-a.e. point on ∂Ω, (1.8.255)
 
∂νAu = − 12 I + K A#  f at σ-a.e. point on ∂Ω,
κ−n.t.
and u ∂Ω = Smod f at σ-a.e. point on ∂Ω.

In addition, if Ω is an exterior domain then from (1.5.51) and [70, (1.4.24)] we see
that ⨏
|∇u| dL n = o(1) as R → ∞. (1.8.256)
B(0,2R)\B(0,R)

Granted these properties (and also keeping in mind (1.8.123), (A.0.127), and [68,
(7.7.106)]), we may invoke Theorem 1.8.17 and conclude from (1.8.176) that there
exists some C M -valued locally constant function cu in Ω with the property that
   
u = Dmod Smod f − 𝒮mod − 12 I + K A#  f + cu in Ω. (1.8.257)
κ−n.t.
Taking nontangential boundary traces in (1.8.257) and denoting c f := cu ∂Ω then
yields
    
Smod f = 12 I + Kmod Smod f − Smod − 12 I + K A#  + c f , (1.8.258)

on account of (1.8.255), (1.8.123), (1.8.143), (1.5.12), and the last property in

(1.8.125). Having established (1.8.258), simple algebra then yields (1.8.252). This
takes care of (1).
Consider next the first claim made in item (2). To set the stage, bring in the
matrix-valued fundamental solution E = (Eγβ )1≤γ,β ≤M associated with L as in [70,
Theorem 1.4.2]. Also, fix an arbitrary function
.p p M
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) (1.8.259)

and select an arbitrary index μ ∈ {1, . . . , M }. In a first stage, our goal is to show that
  
K A#  ∂νA Dmod f (x) (1.8.260)
μ
∫ 
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s Kmod f α (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. To this end, define u := Dmod f in Ω, and pick an aperture
parameter κ ∈ (0, ∞) along with a truncation parameter ε ∈ (0, ∞). Then (1.8.142),
226 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

(1.8.22), and (1.8.143) ensure that the following properties are satisfied:
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
Nκε u ∈ Lloc (∂Ω, σ),
p
Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t. κ−n.t. (1.8.261)
the boundary traces u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂Ω,
κ−n.t.    
u ∂Ω = 12 I + Kmod f and ∂νAu = ∂νA Dmod f .

In addition, [70, Theorem 1.4.2] implies that if Ω is an exterior domain we have

⨏
|∇u| dL n = o(1) as R → ∞. (1.8.262)
B(0,2R)\B(0,R)

Let also note here that, as seen from (1.8.121) and (1.8.261), for each , s ∈ {1, . . . , n}
and each α ∈ {1, . . . , M } we have
 κ−n.t.   κ−n.t.   
ν (∂s uα ) ∂Ω − νs (∂ uα ) ∂Ω = ∂τ s ( 12 I + Kmod ) f α . (1.8.263)

For each ∈ {1, . . . , n} and each γ ∈ {1, . . . , M } we may then rely on (1.8.261)-
(1.8.263), [70, (1.5.230)], and (A.0.184) to write
∫
βα  
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y)∂τ s ( 12 I + Kmod ) f α (y) dσ(y)
∂Ω
∫  
− (∂ Eγα )(x − y) ∂νA Dmod f (y) dσ(y) (1.8.264)
∂Ω α

at each point x ∈ Ω. Next, fix an arbitrary index μ ∈ {1, . . . , M }. Multiply both sides

μγ
of (1.8.264) by a j , then go nontangentially to the boundary, subsequently multiply
by ν j , and finally sum up over j ∈ {1, . . . , n}. On account of [70, (1.5.230)], we
therefore arrive at
 
∂νA Dmod f (x) (1.8.265)
μ
μγ
= ν j (x)a j (∂ uγ )(x)
/∫ 0 κ−n.t.

μγ βα  
= ν j (x) a j ar s (∂r Eγβ )(· − y)∂τ s ( 12 I + Kmod ) f α (y) dσ(y) (x)
∂Ω
∂Ω
/∫ 0 κ−n.t.
 
μγ
− ν j (x) a j (∂ Eγα )(· − y) ∂νA Dmod f (y) dσ(y) (x)
∂Ω α
∂Ω

at σ-a.e. point x ∈ ∂Ω. The term involving the second nontangential trace in above
is equal to (cf. [70, (1.5.230)], (1.5.51), and (1.5.58)),
1.8 Modified Boundary Layer Potential Operators 227
/∫ 0 κ−n.t.
 
μγ
ν j (x) a j (∂ Eγα )(· − y) ∂νA Dmod f (y) dσ(y) (x) (1.8.266)
∂Ω α
∂Ω
     
= ∂νA𝒮mod ∂νA Dmod f (x) = − 12 I + K A#  ∂νA Dmod f (x)
μ μ

at σ-a.e. point x ∈ ∂Ω.

In view of [70, (2.5.4)], the term involving the first nontangential trace in (1.8.265)
may be explicitly identified as
/∫ 0 κ−n.t.

μγ βα  
ν j (x) a j ar s (∂r Eγβ )(· − y)∂τ s ( 12 I + Kmod ) f α (y) dσ(y) (x)
∂Ω
∂Ω

μγ βα 1 -   
= ν j (x)a j ar s ∂r Eγβ ν(x) ∂τ s ( 12 I + Kmod ) f α (x) (1.8.267)
2i
∫
μγ βα  
+ lim+ a j ar s (∂r Eγβ )(x − y)∂τ s ( 12 I + Kmod ) f α (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. Bearing in mind [70, (1.4.30)] and (1.8.263), at σ-a.e. point
x ∈ ∂Ω we may recast the first term in the right-hand side of (1.8.267) as

μγ βα 1 -   
ν j (x)a j ar s ∂r Eγβ ν(x) ∂τ s ( 12 I + Kmod ) f α (x) (1.8.268)
2i
1 μγ βα   −1
= ν j (x)a j ar s νr (x) L ν(x) γβ ×
2
  κ−n.t.   κ−n.t.  
× ν (x) (∂s uα ) ∂Ω (x) − νs (x) (∂ uα ) ∂Ω (x) .

Recall from (1.5.176)-(1.5.177) that for each s ∈ {1, . . . , n} and α ∈ {1, . . . , M } we

have
μγ βα   −1 μα
ν j (x)a j ar s νr (x) L ν(x) γβ ν (x) = −ar s νr (x), (1.8.269)

while for each ∈ {1, . . . , n} and α ∈ {1, . . . , M } we have

μγ βα   −1 μα
ν j (x)a j ar s νr (x) L ν(x) ν (x)
γβ s
= −a j ν j (x). (1.8.270)

From (1.8.268)-(1.8.270) we then see that

228 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

μγ βα 1 -   
ν j (x)a j ar s ∂r Eγβ ν(x) ∂τ s ( 12 I + Kmod ) f α (x)
2i
1  μα  κ−n.t. 
μα  κ−n.t.  
=− ar s νr (x) (∂s uα ) ∂Ω (x) − a j ν j (x) (∂ uα ) ∂Ω (x) .
2
= 0 at σ-a.e. point x ∈ ∂Ω. (1.8.271)

Taking into account (1.8.271), (1.8.145), and (1.8.153), it follows that (1.8.267)
becomes
/∫ 0 κ−n.t.
μγ βα  1 
ν j (x) a j ar s (∂r Eγβ )(· − y)∂τ s ( 2 I + Kmod ) f α (y) dσ(y) (x)
∂Ω
∂Ω

1  
= ∂νA Dmod f (x)
2 μ
∫
μγ βα  
+ lim+ a j ar s (∂r Eγβ )(x − y)∂τ s Kmod f α (y) dσ(y) (1.8.272)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. Together, (1.8.265), (1.8.266), and (1.8.272) readily prove
(1.8.260) (after some natural re-labeling). This completes the proof of (1.8.260)
under the current assumptions. Finally, (1.8.253) is a consequence of (1.8.260),
.p p M
(1.8.145)-(1.8.146), and the fact that the space L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) is p-
resently invariant under the action of Kmod (see Theorem 1.8.14).
There is yet another, more direct, proof of (1.8.253) in item (2) we would like
to discuss. Specifically, granted the properties noted in (1.8.261)-(1.8.262), we may
invoke Corollary 1.8.18 and conclude from (1.8.194) that there exists a C M -valued
locally constant function cu in Ω with the property that
   
u = Dmod 12 I + Kmod f − 𝒮mod ∂νA Dmod f + cu in Ω. (1.8.273)

Applying ∂νA to both sides then yields, on account of (1.8.261), (1.8.146), the jump-
formula in the fifth line of (1.8.125), and the fact that ∂νA cu = 0 (as may be seen
from (A.0.184)),
 A       
∂ν Dmod f = ∂νA Dmod 12 I + Kmod f − − 12 I + K A#  ∂νA Dmod f , (1.8.274)

from which (1.8.253) now readily follows after canceling like-terms.

To justify (1.8.254), start with f as in (1.8.259) and once again define u := Dmod f
κ−n.t.
in Ω. Then (1.8.273) continues to hold. If we introduce c f := cu ∂Ω , then going
nontangentially to the boundary in (1.8.273) yields
1  1 1   A 
2 I + Kmod f = 2 I + Kmod 2 I + Kmod f − Smod ∂ν Dmod f + c f , (1.8.275)
1.8 Modified Boundary Layer Potential Operators 229

thanks to the first formula in the fourth line of (1.8.261), the jump-formula (1.8.27)
(bearing in mind (A.0.127) and the fact that the operator Kmod leaves the space
.p p M
L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) invariant; see Theorem 1.8.14), and the last property
in (1.8.125) (also keeping in mind (1.8.146)). With (1.8.275) in hand, the identity
claimed in (1.8.254) follows after some simple algebra. This completes the treatment
of item (2).
To deal with item (3), work under the assumption that Ω is an NTA domain
with an Ahlfors regular boundary. The proof of (1.8.253) for an arbitrary function
.p M
f ∈ L1 (∂Ω, σ) proceeds along similar lines to the argument in the either of the
two proofs of this identity provided in item (2). As far as the first proof is concerned,
.p M
given any f ∈ L1 (∂Ω, σ) , we once again define u := Dmod f in Ω. This time,
(1.8.142) and (1.8.143) imply (compare with (1.8.261))
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t. κ−n.t.
the boundary traces u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂Ω, (1.8.276)
κ−n.t.    
u ∂Ω = 12 I + Kmod f and ∂νAu = ∂νA Dmod f .

In a first stage, the goal is to establish (1.8.260). For this we argue as before, the main
difference is the justification of (1.8.263). Granted the current assumptions, we may
invoke [68, Proposition 8.4.9] and conclude from [68, (8.4.109)] and (1.8.261) that
there exists a truncation parameter ε ∈ (0, ∞) such that

Nκε u ∈ Lloc (∂Ω, σ).

p
(1.8.277)

In turn, (1.8.277) and (1.8.261) permit us to call upon [69, Proposition 11.3.2] to
conclude from [69, (11.3.26)] and (1.8.276) that
 κ−n.t.   κ−n.t. 
ν (∂s uα ) ∂Ω − νs (∂ uα ) ∂Ω
 κ−n.t.  
= ∂τ s u ∂Ω = ∂τ s ( 12 I + Kmod ) f α, (1.8.278)

at σ-a.e. point on ∂Ω. Thus, (1.8.263) continues to hold under the current geometric
assumptions, and from this point on the proof of (1.8.260) is completed as before. At
this stage, (1.8.253) becomes a consequence of (1.8.260), (1.8.145)-(1.8.146), and
.p M
the fact that the homogeneous Sobolev space L1 (∂Ω, σ) is presently invariant
under the action of Kmod (see Theorem 1.8.14).
Let us also note that formula (1.8.253) may also be justified for each function
.p M
f ∈ L1 (∂Ω, σ) much as in the second proof given in item (2), now relying on
the integral representation formula (1.8.200) from Theorem 1.8.19 for the function
u := Dmod f , then taking conormal derivatives.
Finally, we may also establish the veracity of formula (1.8.254) for each
.p M
f ∈ L1 (∂Ω, σ) by once again employing the integral representation formula
230 1 Layer Potential Operators on Lebesgue and Sobolev Spaces

(1.8.200) from Theorem 1.8.19 for the function u := Dmod f , then going nontangen-
tially to the boundary. 
Remark 1.8.27 It is of interest to consider the operator identities described in The-
orem 1.8.12 in the case when they involve modified boundary-to-boundary double
layer potential operators of the sort described in Examples 1.8.4-1.8.5. Indeed, in
such scenarios, the operator ∂νA Dmod vanishes identically (as noted in the last part in
Remark 1.4.6 and in (1.4.33)). As such, the identity recorded in (1.8.254) simplifies in
each of these settings. For example, re-branding the modified boundary-to-boundary
Cauchy-Clifford integral operator (1.8.112) as Cmod , we conclude from (1.8.254) that
.p
for each f ∈ L1 (∂Ω, σ) ⊗ C n with p ∈ (1, ∞) there exists c f , which is the non-
tangential trace on ∂Ω of some C n -valued locally constant function in Ω, with the
property that
1   1 
2 I + C mod − 2 I + Cmod f = c f at σ-a.e. point on ∂Ω, (1.8.279)

or, equivalently,  
Cmod Cmod f = 1
4 f + c f on ∂Ω. (1.8.280)
In particular, if ∂Ω is connected and we let Cmod be the operator [ f ] → Cmod f
(with brackets denoting equivalence classes modulo constants), we therefore obtain
2 .p
Cmod = 14 I on L1 (∂Ω, σ)/∼ ⊗ C n with p ∈ (1, ∞). (1.8.281)

See also (2.1.186)-(2.1.189) for related identities.

In the same setting as above, .(1.8.252) yields (reasoning in a similar fashion)
p
that for each given function f ∈ L1 (∂Ω, σ) ⊗ C n with p ∈ (1, ∞) there exists some
constant c f , which is the nontangential trace on ∂Ω of some C n -valued locally
constant function in Ω, such that
   
Smod C # f = Cmod Smod f + c f . (1.8.282)

Of course, similar results are valid for the Cauchy operator in the complex plane
(a scenario in which n = 2, and we identify R2 ≡ C). For example, re-branding the
modified boundary-to-boundary Cauchy integral operator (1.8.109) as Cmod , much
as above we conclude from (1.8.254) that if ∂Ω is connected we have
2 .p
Cmod = 14 I on L1 (∂Ω, σ)/∼ with p ∈ (1, ∞), (1.8.283)

where Cmod denotes the operator [ f ] → Cmod f (again, with brackets denoting
equivalence classes modulo constants).
Here is an application to the theory of Hardy spaces in which some of the above
considerations play a role.
Proposition 1.8.28 Let Ω ⊆ C ≡ R2 be a two-sided NTA domain such that ∂Ω is
an unbounded Ahlfors regular set. Abbreviate σ := H 1 ∂Ω and fix an integrability
exponent p ∈ (1, ∞) along with some aperture parameter κ > 0. Next, set
1.8 Modified Boundary Layer Potential Operators 231

Ω+ := Ω, Ω− := C \ Ω, (1.8.284)

and consider the “positive/negative” homogeneous Hardy spaces with regularity on

∂Ω, defined as
.p  κ−n.t.
H1,± (∂Ω, σ) := u ∂Ω : u holomorphic in Ω± and

Nκ (∇u) ∈ L p (∂Ω, σ) . (1.8.285)

Then, with ∼ standing for equivalence modulo constants,

.p .p
H1,± (∂Ω, σ)/∼ are closed subspaces of L1 (∂Ω, σ)/∼ and
.p  .  . (1.8.286)
p p
L1 (∂Ω, σ)/∼ = H1,+ (∂Ω, σ)/∼ ⊕ H1,− (∂Ω, σ)/∼ , direct sum.

In fact, if Cmod is the modified boundary-to-boundary

 Cauchy singular integral
σ(ζ )
operator acting on each function f ∈ L 1 ∂Ω, 1+ |ζ | 2
according to

∫  .
1 1 1
Cmod f (z) := lim+ 1 (ζ) − 1C\B(0,1) (ζ) f (ζ) dζ (1.8.287)
ε→0 2πi ζ − z C\B(z,ε) ζ
∂Ω

at σ-a.e. z ∈ ∂Ω (see (1.8.109)), and if Cmod denotes the operator [ f ] → Cmod f

(with brackets denoting equivalence classes modulo constants), then
.p  .p
H1,± (∂Ω, σ)/∼ = Ker ∓ 12 I + Cmod ; L1 (∂Ω, σ)/∼
 .p
= Im ± 12 I + Cmod ; L1 (∂Ω, σ)/∼ . (1.8.288)

Proof For starters, observe that the present assumptions imply (see the last part in
[68, Lemma 5.11.3]) that
Ω± are connected sets. (1.8.289)
.p
In addition, from (1.8.285) and [69, Proposition 11.5.12] we see that H1,± (∂Ω, σ)
.p
are well-defined subspaces of L1 (∂Ω, σ). To proceed, from Example 1.8.4 and
.p
Remark 1.8.13 we conclude that for each f ∈ L1 (∂Ω, σ) the functions

u± := 𝒞mod f in Ω± (1.8.290)

are holomorphic and satisfy Nκ (∇u± ) ∈ L p (∂Ω, σ) as well as

κ−n.t. 1 
u± ∂Ω = 2I ± Cmod f at σ-a.e. point on ∂Ω. (1.8.291)

In view of (1.8.285), we therefore have

232 1 Layer Potential Operators on Lebesgue and Sobolev Spaces
1  .p
2I ± Cmod f ∈ H1,± (∂Ω, σ), (1.8.292)
1   
and since f = 2I + Cmod f + 12 I − Cmod f , we obtain
.p .p .p
L1 (∂Ω, σ) = H1,+ (∂Ω, σ) + H1,− (∂Ω, σ), (1.8.293)

hence also
.p  .  .
p p
L1 (∂Ω, σ)/∼ = H1,+ (∂Ω, σ)/∼ + H1,− (∂Ω, σ)/∼ . (1.8.294)
.p .p
Suppose now f ∈ H1,+ (∂Ω, σ) ∩ H1,− (∂Ω, σ). By definition, this implies the ex-
istence of two holomorphic functions w± in Ω± with Nκ (∇w± ) ∈ L p (∂Ω, σ) and
κ−n.t.
satisfying w± ∂Ω = f . Then (1.8.236) implies (bearing in mind (1.8.289)) that
there exists a constant c ∈ C such that
 κ−n.t.  κ−n.t.
w− = 𝒞mod w− ∂Ω + c = 𝒞mod w+ ∂Ω + c in Ω− (1.8.295)

which, according to (1.8.237), means that w− is a constant function in Ω− (again,

here (1.8.289) is relevant). In a very similar fashion, we also obtain that w+ is a
constant
 function  in Ω + . Thus, 𝒞
 mod f are constant functions in Ω± which then forces
f = 12 I + Cmod f + 12 I − Cmod f to be itself a constant function on ∂Ω. In concert
with (1.8.294), this ultimately yields the direct sum decomposition in the second line
of (1.8.286).
There is more one can extract from the above considerations. Specifically, recall
.p
(from the definition in (1.8.285)) that for any f ∈ H1,+ (∂Ω, σ) there exists a holo-
κ−n.t.
morphic function u in Ω+ with Nκ (∇u) ∈ L p (∂Ω, σ) and such that f = u ∂Ω . Then
(1.8.236) implies (keeping in mind (1.8.289)) that there exists a constant c ∈ C such
that  κ−n.t.
u = 𝒞mod u ∂Ω + c = 𝒞mod f + c in Ω+ . (1.8.296)

After going nontangentially to the boundary we arrive at

κ−n.t. 1 
f = u ∂Ω = 2I + Cmod f + c. (1.8.297)

Hence .p
Cmod [ f ] = 12 [ f ] for each [ f ] ∈ H1,+ (∂Ω, σ)/∼ (1.8.298)
and, analogously,
.p
Cmod [ f ] = − 12 [ f ] for each [ f ] ∈ H1,− (∂Ω, σ)/∼ . (1.8.299)

These prove the first equality in (1.8.288) from which we also deduce that
.p .p
H1,± (∂Ω, σ)/∼ are closed subspaces of L1 (∂Ω, σ)/∼. Finally, the second equali-
ty in (1.8.288) is a consequence of (1.8.283). 
1.8 Modified Boundary Layer Potential Operators 233

In closing we note that similar results are valid for homogeneous Hardy spaces
with regularity in the higher-dimensional setting, now employing Clifford algebra-
valued null-solutions of the Dirac operator in Rn in place of holomorphic functions
in the complex plane.
Chapter 2
Layer Potential Operators on Hardy, BMO,
VMO, and Hölder Spaces

The reader is reminded that Hölder, BMO, VMO, and Hardy spaces in the general
setting of spaces of homogeneous type and on Ahlfors regular sets have been dis-
cussed at length in [68, §§ 7.3-7.4] and [69, Chapters 3-4]. This chapter is primarily
focused on layer potential operators acting on Hardy, BMO, VMO, and Hölder
spaces defined on boundaries of UR domains. A key aspect in this analysis is that a
special algebraic structure is required of the integral kernel for a singular integral
operator to map either of these spaces into itself, and the brand of Divergence Theo-
rem produced in [68] plays a crucial role in ensuring this is indeed the case. In order
to be more specific, let us pick a UR set Σ ⊆ Rn and abbreviate σ := H n−1 Σ. Then
all garden-variety Calderón-Zygmund convolution-type singular integral operators
T (of the sort considered in [70, Chapter 2]) are well defined and bounded in the
context T : H p (Σ, σ) → L p (Σ, σ) for n−1
n < p ≤ 1. This being said, they hopelessly
fail to map the Hardy space H (Σ, σ) into itself for n−1
p
n < p ≤ 1. Well, a long time
ago (in the late 1970’s to be more precise; cf. [16]) R. Coifman and G. Weiss have
taught us that when regarding Σ as a space of homogeneous type, i.e., equipped with
the Euclidean metric and the doubling measure σ, for a linear and bounded inte-
gral operator T : L 2 (Σ, σ) → L 2 (Σ, σ) associated with an integral kernel satisfying
“standard” size and regularity properties to actually map the Hardy space H 1 (Σ, σ)
boundedly into itself it is necessary and sufficient that

m := T a is a molecule on Σ, for each atom a on Σ (2.0.1)

(in a quantitative fashion). The delicate aspect is that, by∫ design, a molecule m on
Σ is supposed to satisfy the vanishing moment condition Σ m dσ = 0, so we would
need ∫
T a dσ = 0 for each atom a on Σ. (2.0.2)
Σ
Alas, plain convolution-type operators (i.e., integral operators with integral kernel
k(x − y), where k is a smooth, odd, and positive homogeneous of degree 1 − n
function in Rn \ {0}) are really dull, as in general they lack any type of cancelation
properties. This deficiency is predicated by the inability of the integral kernel k
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 235
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_2
236 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

to “see the underlying surface” Σ, to which this is completely unrelated. This is

why operators like the transpose harmonic double layer KΔ# (from (A.0.102)) are
decisively different. For example, corresponding to the case when Σ := ∂Ω for some
UR domain Ω ⊆ Rn , we may verify (2.0.2) for KΔ# playing the role of T by writing,
for each atom a on ∂Ω,
∫ ∫
 1 
KΔ# a dσ = − 2 I + KΔ# a dσ
∂Ω ∂Ω
∫  n.t.  ∫
= ν · F∂Ω dσ = divF dL n = 0, (2.0.3)
∂Ω ∂Ω

where ν is the geometric measure theoretic outward unit normal to Ω, and the vector
field F is given by ∇𝒮a (with 𝒮 denoting the boundary-to-domain single layer
potential operator for the Laplacian in Ω). The first equality uses the vanishing
moment condition of the atom itself, the second equality is a particular case of
the jump-formula (1.5.58), the third equality comes from an application of the
Divergence Formula from [68, Theorem 1.2.1], and the last equality is a consequence
of the fact that F is actually divergence-free in Ω (since div∇𝒮a = Δ𝒮a = 0).
This is the line of attack we adopt for proving mapping properties for transpose
double layers associated with weakly elliptic second-order systems in UR domains.
Once the action of the these double layers on the Hardy scale has been established,
one can deal with similar issues on BMO, VMO, and Hölder spaces via duality.
Much of the work in this chapter elaborates on this program.

2.1 Double Layer Potential Operators on Hardy, BMO, VMO,

and Hölder Spaces

Consider the principal-value (p.v.) singular integral operator K # from (1.3.72) in the
case when the set Ω ⊆ Rn is a UR domain and σ := H n−1 ∂Ω. In light of item (6)
of [70, Theorem 2.3.2] (while also bearing in mind the qualities of the fundamental
solution highlighted in [70, Theorem 1.4.2]) it follows that K # , originally acting on
Lebesgue spaces on ∂Ω (as in (1.5.12)), extends uniquely to a linear and bounded
operator
M M  n−1 
K # : H p (∂Ω, σ) −→ L p (∂Ω, σ) , p∈ n ,∞ . (2.1.1)
M
Specifically, given any f = ( fγ )1≤γ ≤M ∈ H p (∂Ω, σ) , the operator K # acts on f

n , 1 , the action of K on f
as in (1.3.72) in the case when p ∈ (1, ∞) and, if p ∈ n−1 #

is defined at σ-a.e. x ∈ ∂Ω according to

2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 237

K # f (x) (2.1.2)
m ∫
βα
= νs (x)ar s · lim λγ, j lim+ (∂r Eγβ )(x − y)aγ, j (y) dσ(y)
m→∞ ε→0
j=1 y ∈∂Ω 1≤α ≤M
|x−y |>ε

where, for each γ ∈ {1, . . . , n}, we have relied on [69, Theorem

  4.4.1] pto expand

fγ = ∞ j=1 λ γ, j aγ, j (with convergence in H p (∂Ω, σ)), with λ
γ, j j ∈N
∈  and each
aγ, j a (p, q)-atom on ∂Ω for some fixed background exponent q ∈ [1, ∞] with q > p.
In fact, we may use real interpolation (as in [69, Theorem 4.3.1] and [68, (6.2.48)])
to conclude from (2.1.1) that K # also induces a linear and bounded operator
M M
K # : H p,q (∂Ω, σ) −→ L p,q (∂Ω, σ) ,
  (2.1.3)
for p ∈ n−1n , ∞ and q ∈ (0, ∞].

The main point of our first theorem in this section is that, due to its specific algebraic
nature, the operator K # actually maps Hardy spaces into Hardy spaces in a linear
and bounded fashion.

Theorem 2.1.1 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). In this context,
consider the boundary layer potential operator K # associated with the system L and
the set Ω as in (1.3.72).
Then the operator K # , originally acting on Lebesgue spaces on ∂Ω (as in (1.5.12)),
extends uniquely to a linear and bounded mapping
M M  n−1
K # : H p (∂Ω, σ) −→ H p (∂Ω, σ) , p∈ n ,1 . (2.1.4)

Moreover, various choices of the exponent p yield operators which are compatible
with one another.
Ultimately, as a consequence of (2.1.4) and (1.5.12), one has a (unique) family
of operators
M M  n−1 
K # : H p (∂Ω, σ) −→ H p (∂Ω, σ) , p∈ n ,∞ , (2.1.5)

which are well-defined, linear, continuous, compatible with one another, and which
agree with (1.3.72) when p ∈ (1, ∞). In addition,
 
if p ∈ n−1
n , ∞ , the composition to the left of K from (2.1.5) with
#
M M
the L p -filtering operator H : H p (∂Ω, σ) → L p (∂Ω, σ) (2.1.6)
M M
(cf. [69, (4.9.2)]) is K # : H p (∂Ω, σ) → L p (∂Ω, σ) de-
fined as in (2.1.1)-(2.1.2).
238 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Finally, the operator K # in the context of (2.1.5) depends in a continuous fashion

on the underlying coefficient tensor (in the sense of item (xv) in Theorem 1.5.1).

Proof Fix an arbitrary exponent p ∈ n−1 n , 1 . The gist of the strategy is to show
that K # maps vector-valued atoms into a fixed multiple of vector-valued molecules
for the vector Hardy space in question. Concretely, pick q ∈ (1, ∞) and consider an
arbitrary C M -valued (p, q)-atom on ∂Ω. Recall from [69, (4.4.167)-(4.4.168)] that
this means that a : ∂Ω → C M is some σ-measurable  function with the property that
there exist a point xo ∈ ∂Ω and some number r ∈ 0, 2 diam(∂Ω) such that

supp a ⊆ B(xo, r) ∩ ∂Ω,

  1/q−1/p
a [L q (∂Ω,σ)] M ≤ σ B(xo, r) ∩ ∂Ω , (2.1.7)
∫
a dσ = 0 ∈ C M .
∂Ω

In addition, in the case when ∂Ω is compact, we agree that each vector in C M of

norm ≤ σ(∂Ω)−1/p is also considered, by definition, to be a C M -valued (p, q)-atom
on ∂Ω. The claim that we make is that
the function m := K # a is a fixed multiple of a C M -valued (p, q, ε)-
molecule on ∂Ω (i.e., a C M -valued function whose scalar com- (2.1.8)
ponents are as in [69, Definition 4.5.1]), with ε := n−1 1
.

To prove this, first note that according to Theorem 1.5.1 the function m is mean-
M
ingfully defined and belongs to the space L q (∂Ω, σ) . In fact, thanks to item (iii)
in Theorem 1.5.1 and (2.1.7), we have

m [L q (∂Ω,σ)] M = K#a [L q (∂Ω,σ)] M ≤C a [L q (∂Ω,σ)] M

  1/q−1/p
≤ Cσ B(xo, r) ∩ ∂Ω , (2.1.9)

for some finite constant C = C(∂Ω, L, q) > 0 independent of the atom.

To study the decay of m, observe that based on (2.1.7), the Mean Value Theorem,
and [70, Theorem 1.4.2], we may estimate
∫
|m(x)| ≤ C |(∇E)(x − y) − (∇E)(x − xo )||a(y)| dσ(y)
B(x o ,r)∩∂Ω
r   1−1/p
≤C σ B(xo, r) ∩ ∂Ω (2.1.10)
|x − xo | n

at every x ∈ ∂Ω \ B(xo, 2r). To proceed, for each k ∈ N define the boundary annulus
k (xo, r) := B(xo, 2k+1 r) \ B(xo, 2k r) ∩ ∂Ω.
A (2.1.11)

We may then rely on (2.1.10) and the Ahlfors regularity of ∂Ω to obtain to that, for
each k ∈ N,
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 239
∫  1/q
|m| q dσ
k (x o ,r)
A
r   1−1/p   1/q
≤C σ B(xo, r) ∩ ∂Ω σ B(xo, 2k+1 r) ∩ ∂Ω
(2k r)n
  1/q−1/p
≤ C2k(n−1)[1/q−1−1/(n−1)] σ B(xo, r) ∩ ∂Ω (2.1.12)

for some finite constant C = C(∂Ω, L) > 0 independent of the atom.

At this stage, the claim in (2.1.8) follows as soon as we check that
∫
m dσ = 0. (2.1.13)
∂Ω

To this end, recall the matrix-valued fundamental solution E = (Eαβ )1≤α,β ≤M

associated with L as in [70, Theorem 1.4.2]. For any given indices α, β ∈ {1, . . . , M }
and j ∈ {1, . . . , n} then introduce two integral operators acting on each scalar-valued
function f ∈ L po (∂Ω, σ) with po ∈ (1, ∞) according to
∫
j
Qαβ f (x) := (∂j Eαβ )(x − y) f (y) dσ(y) for each x ∈ Ω, (2.1.14)
∂Ω

and, respectively,
∫
j
Q αβ f (x) := lim+ (∂j Eαβ )(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.1.15)
ε→0
y ∈∂Ω
|x−y |>ε

Fix a background parameter κ > 0 and pick f ∈ L po (∂Ω, σ) with po ∈ (1, ∞).
Also, select α, β ∈ {1, . . . , M } and j ∈ {1, . . . , n}. Then [70, Theorem 2.3.2] im-
j
plies that the limit defining Q αβ f in (2.1.15) exists for σ-a.e. x ∈ ∂Ω. Also, [70,
Theorem 2.5.1] and [70, (1.4.30)] ensure that at σ-a.e. point x ∈ ∂Ω we have the
jump-formula

  κ−n.t. 1  
(x) = ∂
j j
Qαβ f  j Eαβ ν(x) f (x) + (Q αβ f )(x)
∂Ω 2i
1 j
= bαβ (x)ν j (x) f (x) + (Q αβ f )(x), (2.1.16)
2
where ν = (ν1, . . . , νn ) is the geometric measure theoretic outward unit normal to Ω,
and
bαβ (x) is the (α, β)-entry in the M × M matrix
   −1
γδ (2.1.17)
L(ν(x))−1 = − ar s νr (x)νs (x)
1≤γ,δ ≤M

(with the summation convention over repeated indices in effect). Pushing forth, fix
an arbitrary index α ∈ {1, . . . , M } and bring in the vector field
240 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
βα r
F = (Fs )1≤s ≤n with Fs := ar s Qγβ aγ in Ω, for each s ∈ {1, . . . , n}, (2.1.18)

where aγ is the γ-th component of the C M -valued function a. Then, by design,

F ∈ 𝒞∞ (Ω)
n
(2.1.19)

and, thanks to (2.1.18), (2.1.14), [70, (1.4.33)], we have

 βα
(divF)(x) = ar s (∂s Qγβr
aγ )(x)
∫
βα
= ar s (∂s ∂r Eγβ )(x − y)aγ (y) dσ(y)
∂Ω

= 0 for each x ∈ Ω. (2.1.20)

Since for each s ∈ {1, . . . , n} the jump-formula (2.1.16) implies

κ−n.t. κ−n.t.
Fs ∂Ω (x) = ar s (Qγβ aγ )∂Ω (x)
βα r

1 βα βα
= bγβ (x)ar s νr (x)aγ (x) + ar s (Qγβ
r
aγ )(x) (2.1.21)
2
at σ-a.e. point x ∈ ∂Ω, it follows that
κ−n.t.
F∂Ω exists at σ-a.e. point on ∂Ω. (2.1.22)

Formula (2.1.21) also permits us to compute at σ-a.e. point x ∈ ∂Ω

 κ−n.t.   κ−n.t. 
ν(x) · F∂Ω (x) = νs (x) Fs ∂Ω (x)

1 βα βα
= bγβ (x)ar s νr (x)νs (x)aγ (x) + νs (x)ar s (Qγβ
r
aγ )(x)
2
∫
1 βα
= − δγα aγ (x) + νs (x)ar s (∂r Eγβ )(x − y)aγ (y) dσ(y)
2 ∂Ω

1 1
= − aα (x) + (K # a)α (x) = − aα (x) + mα (x), (2.1.23)
2 2
upon recalling (2.1.17), (2.1.15), (1.3.72), and the definition of m in (2.1.8). Since, as
is apparent from (2.1.7), the components of the function a : ∂Ω → C M are multiples
of (1, q)-atoms on ∂Ω, we may invoke [70, (2.4.14)] with p = 1 in order to conclude
that

Nκ F ∈ L 1 (∂Ω, σ). (2.1.24)

Finally, the vanishing moment property of the atom (cf. the last line in (2.1.7))
together with (2.1.18), (2.1.14), and [70, (1.4.24)], imply that
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 241

in the case when Ω is an exterior domain we have


F(x) = O(|x| −n ) as x ∈ Ω satisfies |x| → ∞; hence, (2.1.25)
the pointwise decay property [68, (1.2.9)] is satisfied.
Collectively, (2.1.19), (2.1.20), (2.1.22), (2.1.24), (2.1.25) guarantee the validity
of the Divergence Formula [68, (1.2.2)] which, in light of (2.1.20) and (2.1.23),
presently gives
∫ ∫
 κ−n.t. 
0= divF dL n = ν · F ∂Ω dσ
Ω ∂Ω
∫ ∫ ∫
1
=− aα dσ + mα dσ = mα dσ, (2.1.26)
2 ∂Ω ∂Ω ∂Ω

bearing in mind the vanishing

∫ moment property in (2.1.7). Having established this,
we then conclude that ∂Ω mα dσ = 0 for each index α ∈ {1, . . . , M }, finishing the
proof of (2.1.13).
Let us record our progress: from (2.1.9), (2.1.12), and (2.1.13) we deduce that,
whenever a is as in (2.1.7), the components of the function m := K # a : ∂Ω → C M
are fixed multiples (i.e., the multiplicative constant is independent of the atom a)
of molecules for the Hardy space H p (∂Ω, σ), in the sense of [69, Definition 4.5.1].
Granted this, it follows from [69, (4.5.6)] that
M
m ∈ H p (∂Ω, σ) and m [H p (∂Ω,σ)] M ≤ C(∂Ω, L, p) ∈ (0, ∞). (2.1.27)

Also, in the case when ∂Ω is compact and the (p, q)-atom a is C M -valued con-
stant function on ∂Ω, of absolute value ≤ σ(∂Ω)−1/p , it follows from item (ii-
M
i) of Theorem 1.5.1 that the fucntion m := K # a belongs to L 2 (∂Ω, σ) and
satisfies m [L 2 (∂Ω,σ)] M ≤ C(∂Ω, L, p) ∈ (0, ∞). In view of this and the fact
M M
that L 2 (∂Ω, σ) presently embeds continuously into H p (∂Ω, σ) (cf. [69,
(4.2.13)]), we see that the conclusions in (2.1.27) hold in this case as well.
Having established (2.1.27) in all circumstances, we may now invoke [69, The-
orem 4.4.7] (whose applicability in the present setting makes use of (1.5.12)) to
conclude that, indeed, the mapping K # , originally considered as in (1.5.12), extends
M
uniquely to a linear and bounded operator from the Hardy space H p (∂Ω, σ) into
 n−1
itself. Finally, that various choices of p ∈ n , 1 in (2.1.4) yield operators which are
compatible with one another may now be seen with the help of [69, Theorem 4.4.3].
Let
 us now  deal with the claim made in (2.1.6). In this regard, #fix some exponent
p ∈ n−1 n , ∞ and observe that both the composition to the left of K from (2.1.5) with
M M
the L p -filtering operator H : H p (∂Ω, σ) → L p (∂Ω, σ) and the operator
M
K# from (2.1.1) are well-defined continuous mappings from H p (∂Ω, σ) into
M
L p (∂Ω, σ) . Fix an exponent q ∈ (1, ∞) and recall from [69, (4.4.114)] that the
M M
space H p (∂Ω, σ) ∩ L q (∂Ω, σ) is dense in H p (∂Ω, σ) . Given that, thanks to
M
(1.5.12) and [69, (4.9.3)], said operators agree on H p (∂Ω, σ) ∩ L q (∂Ω, σ) , the
desired conclusion follows.
242 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

To conclude the proof of Theorem 2.1.1 there remains to justify the very last
claim in the statement. First, from (2.1.27) and the fact that the intervening constant
C(∂Ω, L, p) ∈ (0, ∞) depends in a bounded fashion on the coefficient tensor A (used
to write L), we conclude that the operator K # in the context of (2.1.5) depends
in a bounded manner on the underlying coefficient tensor. Granted this, the desired
conclusion follows from (1.5.96) and an interpolation inequality (of the sort discussed
a little later in (2.1.37)). 

Recall the Lorentz-based Hardy spaces from [69, Definition 4.2.3] (cf. (A.0.81)).
 
Corollary 2.1.2 Retain the setting of Theorem 2.1.1. Then for each p ∈ n−1 n ,∞
and q ∈ (0, ∞], the operator K # , originally acting on Lebesgue spaces on ∂Ω (as in
(1.5.12)), induces a linear and bounded mapping
M M
K # : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) (2.1.28)

whose action continues to be compatible with that of K # in (2.1.5). Also,

 
for each p ∈ n−1 n , ∞ and q ∈ (0, ∞], the composition to
the left of K # from (2.1.28) with the operator H mapping
M M
H p,q (∂Ω, σ) into L p,q (∂Ω, σ) (cf. [69, (4.9.5) in The- (2.1.29)
orem 4.9.1]) is the transpose double layer potential operator
M M
K # : H p,q (∂Ω, σ) → L p,q (∂Ω, σ) from (2.1.3).

Moreover,
M M
K # : L 1 (∂Ω, σ) −→ H 1,∞ (∂Ω, σ) (2.1.30)

is a well-defined, linear and bounded mapping, and

the composition to the left of K # from (2.1.30) with the filtering

M M
operator H : H 1,∞ (∂Ω, σ) → L 1,∞ (∂Ω, σ) (cf. [69,
(2.1.31)
(4.9.5) in Theorem 4.9.1]) is the principal value operator
M M
K # : L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ) from (1.5.13).

Proof The claims about (2.1.28) are seen from Theorem 2.1.1, [69, (4.3.3)], and
real interpolation. In particular, (2.1.28) with p = 1 and q = ∞ gives that K # maps
M
H 1,∞ (∂Ω, σ) boundedly into itself. Granted this, the fact that K # in (2.1.30) is a
well-defined, linear and bounded mapping follows on account of [69, (4.2.28)] (used
with Σ := ∂Ω).    n−1 
To prove (2.1.29), fix p ∈ n−1 n , ∞ and q ∈ (0, ∞], then select p0, p1 ∈ n ,∞
such that p0 < p < p1 . Since the operators (2.1.5) act in a compatible fashion with
one another, we may extend
M M
K # : H p0 (∂Ω, σ) + H p1 (∂Ω, σ) −→ H p0 (∂Ω, σ) + H p1 (∂Ω, σ) (2.1.32)
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 243
M
by setting K # ( f0 + f1 ) := K # f0 + K # f1 for each f j ∈ H p j (∂Ω, σ) with j ∈ {0, 1}.
Likewise, starting from the fact that the operators (2.1.1) act coherently, we may
once again naturally extend
M M
K # : H p0 (∂Ω, σ) + H p1 (∂Ω, σ) −→ L p0 (∂Ω, σ) + L p1 (∂Ω, σ) (2.1.33)

Given that [69, Theorem 4.9.1] also allows us to canonically extend

M M
H : H p0 (∂Ω, σ) + H p1 (∂Ω, σ) −→ L p0 (∂Ω, σ) + L p1 (∂Ω, σ) , (2.1.34)

we may invoke (2.1.6) to conclude that the composition to the left of K # from
(2.1.32) with H from (2.1.34) is the operator K # from (2.1.33). With this in hand,
and recalling from [69, (1.3.41), (4.3.3)] that

H p,q (∂Ω, σ) ⊆ H p0 (∂Ω, σ) + H p1 (∂Ω, σ), (2.1.35)

the conclusion in (2.1.29) follows.

At this stage in the proof there remains to deal with the claim in (2.1.31). To
this end, observe first that both the composition to the left of the operator K # from
M M
(2.1.30) with the filtering operator H : H 1,∞ (∂Ω, σ) → L 1,∞ (∂Ω, σ) and the
M M
principal-value singular integral operator K # : L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ)
M
from (1.5.13) are well-defined continuous mappings from the space L 1 (∂Ω, σ)
M M
into L 1,∞ (∂Ω, σ) . Fix q ∈ (1, ∞). Since L 1 (∂Ω, σ) ∩ L q (∂Ω, σ) is dense in
M M
L 1 (∂Ω, σ) and said operators agree on L 1 (∂Ω, σ)
∩ L q (∂Ω, σ) (thanks to
(1.5.12) and [69, (4.9.3)]), the desired conclusion follows. 
To sate our next result, recall the piece of notation introduced in (A.0.171).

Proposition 2.1.3 Retain assumptions made in Theorem 2.1.1 on the set Ω ⊆ Rn

and the M × M system L. In addition, suppose
n−1
n < po < p ≤ 1 < qo < ∞, 0 < q < ∞, (2.1.36)

and consider θ ∈ (0, 1) such that 1/p = (1 − θ)/po + θ/qo . Then there exists a finite
constant C > 0, depending only on ∂Ω, n, po , p, qo , q, with the property that the
norm of the operator K # in (2.1.28) may be estimated as
θ
K# Bd([H p, q (∂Ω,σ)] M ) ≤ C K# 1−θ
Bd([H p o (∂Ω,σ)] M )
· K# Bd([L q o (∂Ω,σ)] M )
. (2.1.37)

Proof This is a consequence of Theorem 2.1.1, [69, (4.3.3)], the interpolation es-
timate from [69, Proposition 1.3.7, (1.3.64)], and (A.0.171). Attention should be
paid to the fact that [69, (1.3.64)] works when the target spaces are quasi-normed
lattices of functions (which is not the case for Hardy spaces H p with p ≤ 1). One
remedy is to apply [69, (1.3.64)] to the sub-linear operator T f := (K # f )γ (i.e., the
Fefferman-Stein grand maximal function of K # f ; cf. [69, (4.1.6)]) which now takes
values in Lebesgue spaces. Alternatively, we may take advantage of the fact that K #
244 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

is a linear operator and use the version of [69, Proposition 1.3.7] discussed in [68,
Remark 1.3.8].
In the special case when q = p (a scenario in which H p,q (∂Ω, σ) simply be-
comes H p (∂Ω, σ); cf. [69, (4.2.25)]), a direct proof of the estimate recorded in
(2.1.37) is as follows. Pick some arbitrary C M -valued (p, qo )-atom a on ∂Ω. Consid-
qo M
∫ case when said atom has vanishing moment, i.e., a ∈ Lcomp (∂Ω, σ)
er first the
satisfies ∂Ω a dσ = 0, and there exist a point xo ∈ ∂Ω along with some radius
 
r ∈ 0, 2 diam(∂Ω) such that supp a ⊆ B(xo, r) ∩ ∂Ω and
  1/qo −1/p
a [L q o (∂Ω,σ)] M ≤ σ B(xo, r) ∩ ∂Ω . (2.1.38)
  1/p−1/po
In particular, σ B(xo, r) ∩ ∂Ω · a is a C M -valued (po, qo )-atom on ∂Ω.
Writing that its quasi-norm in [H (∂Ω, σ)] M is bounded by a constant which
p o

depends only on the environment then yields

  1/po −1/p
a [H p o (∂Ω,σ)] M ≤ C · σ B(xo, r) ∩ ∂Ω . (2.1.39)

Bearing in mind the choice of θ, from (2.1.38)-(2.1.39) we conclude that

θ
a 1−θ
[H p o (∂Ω,σ)] M
· a [L q o (∂Ω,σ)] M
≤ C, (2.1.40)

for some finite constant C > 0 independent of the atom. Since the above discussion
guarantees that a ∈ [H po (∂Ω, σ)] M ∩[L qo (∂Ω, σ)] M , it follows from Theorem 2.1.1
and (1.5.12) that K # a also belongs to [H po (∂Ω, σ)] M ∩ [L qo (∂Ω, σ)] M . Based on
this, [69, Proposition 4.2.2], (A.0.171), and (2.1.40), we may then estimate
θ
K#a [H p (∂Ω,σ)] M ≤ C K#a 1−θ
[H p o (∂Ω,σ)] M
K#a [L q o (∂Ω,σ)] M

θ
≤ C K# 1−θ
Bd([H p o (∂Ω,σ)] M )
· K# Bd([L q o (∂Ω,σ)] M )
×

θ
× a 1−θ
[H p o (∂Ω,σ)] M
· a [L q o (∂Ω,σ)] M

θ
≤ C K# 1−θ
Bd([H p o (∂Ω,σ)] M )
· K# Bd([L q o (∂Ω,σ)] M )
. (2.1.41)

The same type of estimate is valid when ∂Ω is bounded and the atom a is constant.
With this in hand, [69, Theorem 4.4.7] applies and yields (2.1.37) in the case when
q = p. 
# defined in (1.2.3)
Pressing on, we discuss the action of the integral operators Tjk
on the scale of Lorentz-based Hardy spaces on ∂Ω.
Theorem 2.1.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate
σ := H n−1  ∂Ω. Consider a complex-valued function b ∈ Lloc1 (Rn, L n ) with the

property b Rn \{0} ∈ 𝒞 (R \ {0}) for some sufficiently large number N = N(n) ∈ N,
N n

and such that ∇b is odd and positive homogeneous of degree 1−n in Rn \{0}. Finally,
fix j, k ∈ {1, . . . , n} and recall the integral operator Tjk
# defined as in (1.2.3).
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 245

Then for each p ∈ n−1
n , 1 the operator T jk from (1.2.3), originally acting on
#

Lebesgue spaces on ∂Ω (as in item (ii) of Proposition 1.2.1), extends uniquely to a

linear and bounded mapping
#
Tjk : H p (∂Ω, σ) −→ H p (∂Ω, σ), (2.1.42)

and there exists a constant C = C(Ω, p) ∈ (0, ∞) with the property that
 #  
T  p ≤C· sup ∂ α (∇b). (2.1.43)
jk H (∂Ω,σ)→H (∂Ω,σ)
p
n−1
|α | ≤ N −1 S

As a consequence of (2.1.42), item (ii) of Proposition 1.2.1, [69, (4.3.3)], and

real interpolation, for each p ∈ n−1
n , ∞ and q ∈ (0, ∞] the operator

#
Tjk : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) (2.1.44)

is well defined, linear, bounded, and satisfies a similar norm estimate as above.

Proof The strategy, similar to that used in the proof of Theorem 2.1.1, is to show
that if a : ∂Ω → C is an arbitrary (p, q)-atom, where q ∈ (1, ∞) is a fixed exponent,
then m := Tjk# a is a fixed multiple of a molecule for the Hardy space H p (∂Ω, σ) (cf.
 
[69, Definition 4.5.1]). Specifically, assuming that xo ∈ ∂Ω and r ∈ 0, 2 diam(∂Ω)
are such that
  1/q−1/p
supp a ⊆ B(xo, r) ∩ ∂Ω, a L q (∂Ω,σ) ≤ σ B(xo, r) ∩ ∂Ω ,
∫ (2.1.45)
and a dσ = 0,
∂Ω

the goal is to show that there exists some finite constant C = C(∂Ω, n, b, p, q) > 0
independent of the atom in question with the property that
  1/q−1/p
m L q (∂Ω,σ) ≤ Cσ B(xo, r) ∩ ∂Ω and, for each k ∈ N,
∫  1/q   1/q−1/p
|m| q dσ ≤ C2k(n−1)[1/q−1−1/(n−1)] σ B(xo, r) ∩ ∂Ω ,
k (x o ,r)
A
(2.1.46)

where
k (xo, r) := B(xo, 2k+1 r) \ B(xo, 2k r) ∩ ∂Ω,
A ∀k ∈ N, (2.1.47)

as well as
∫
m dσ = 0. (2.1.48)
∂Ω
246 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

The estimates in (2.1.46) may be established by closely mimicking the arguments

used in the proof (2.1.9) and (2.1.12) (here, item (ii) in Proposition 1.2.1 plays a
role).
To justify the vanishing moment condition claimed in (2.1.48), for each index
 ∈ {1, . . . , n} we introduce two integral operators acting on scalar-valued functions
f ∈ L po (∂Ω, σ) with po ∈ (1, ∞) according to
∫
B f (x) := (∂ b)(x − y) f (y) dσ(y) for each x ∈ Ω, (2.1.49)
∂Ω

and, respectively,
∫
B f (x) := lim+ (∂ b)(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.1.50)
ε→0
y ∈∂Ω
|x−y |>ε

To proceed, select some background parameter κ > 0 and fix an arbitrary function

f ∈ L po (∂Ω, σ) with po ∈ (1, ∞). In addition, pick an integer  ∈ {1, . . . , n}. [70,
Theorem 2.3.2] then ensures that the limit defining B f in (2.1.50) exists for σ-a.e.
point x ∈ ∂Ω. Also, [70, Theorem 2.5.1] implies that at σ-a.e. point x ∈ ∂Ω we
have the jump-formula

  κ−n.t. 1  
B f  (x) = ∂b ν(x) f (x) + (B f )(x), (2.1.51)
∂Ω 2i
where ν = (ν1, . . . , νn ) is the geometric measure theoretic outward unit normal to Ω.
Let us now define the vector field

F := (Bk a)e j − (B j a)ek ∈ 𝒞∞ (Ω) .

n
(2.1.52)

Thanks to (2.1.52) and (2.1.49) we then have


(divF)(x) = ∂j (Bk a)(x) − ∂k (B j a)(x)
∫ ∫
= (∂j ∂k b)(x − y)a(y) dσ(y) − (∂k ∂j b)(x − y)a(y) dσ(y)
∂Ω ∂Ω

= 0 for each x ∈ Ω. (2.1.53)

Furthermore, the jump-formula (2.1.51) implies

κ−n.t. κ−n.t. κ−n.t.
F∂Ω (x) = (Bk a)∂Ω (x)e j − (B j a)∂Ω (x)ek

1  
= ∂k b ν(x) a(x)e j + (Bk a)(x)e j
2i
1  
− ∂j b ν(x) a(x)ek − (B j a)(x)ek (2.1.54)
2i
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 247

at σ-a.e. point x ∈ ∂Ω. Consequently, at σ-a.e. point x ∈ ∂Ω we have

 κ−n.t.  1  
ν(x) · F∂Ω (x) =  ∂k b ν(x) ν j (x)a(x) + ν j (x)(Bk a)(x)
2i
1  
− ∂j b ν(x) νk (x)a(x) − νk (x)(B j a)(x)
2i
∫
 
= ν j (x)(∂k b)(x − y) − νk (x)(∂j b)(x − y) a(y) dσ(y)
∂Ω

= (Tjk
#
a)(x) = m(x), (2.1.55)

by also relying on [70, (2.5.19)], (2.1.50), and (1.2.3). Upon noting from (2.1.45)
that the function a : ∂Ω → C is a multiple of an (1, q)-atom on ∂Ω, we may use [70,
(2.4.14)] with p = 1 to conclude that

Nκ F ∈ L 1 (∂Ω, σ). (2.1.56)

Finally, the vanishing moment property of the atom (cf. the last property in (2.1.45))
together with (2.1.52) and (2.1.49) imply that
in the case when Ω is an exterior domain we have

F(x) = O(|x| −n ) as x ∈ Ω satisfies |x| → ∞; thus, the (2.1.57)
decay condition [68, (1.2.9)] is presently satisfied.
Together, (2.1.52), (2.1.53), (2.1.54), (2.1.56), and (2.1.57) ensure the validity of
the Divergence Formula [68, (1.2.2)]. In light of (2.1.53) and (2.1.55) this currently
permits us to write
∫ ∫ ∫
 κ−n.t. 
0= divF dL n = ν · F ∂Ω dσ = m dσ, (2.1.58)
Ω ∂Ω ∂Ω

finishing the proof of (2.1.48). In summary, whenever a is as in (2.1.45), the function

m := Tjk# a is a fixed multiple of a molecule for the Hardy space H p (∂Ω, σ), in the

sense of [69, Definition 4.5.1]. Having established this, it follows from [69, (4.5.6)]
that there exists some finite constant C > 0 independent of the atom a such that

m ∈ H p (∂Ω, σ) and m H p (∂Ω,σ) ≤ C. (2.1.59)

At this stage, there remains to consider the case when ∂Ω is compact and the
atom a is of the form σ(∂Ω)−1/p . In this scenario, item (ii) of Proposition 1.2.1
implies that the function m := Tjk # a belongs to the space L 2 (∂Ω, σ) and satisfies

m L 2 (∂Ω,σ) ≤ C(∂Ω, b, p) ∈ (0, ∞). Bearing in mind that, in the current setting,
L 2 (∂Ω, σ) embeds continuously into H p (∂Ω, σ) (cf. [69, (4.2.13)]), we see that
the conclusions in (2.1.59) are valid in this case as well. Thus, (2.1.59) is true in
all circumstances. Granted this, we may once again rely on [69, Theorem 4.4.7]
(keeping (1.2.7) in mind) to conclude that, as claimed, the mapping Tjk # , originally
248 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

considered as in (1.2.3), extends uniquely to a linear and bounded operator from the
Hardy space H p (∂Ω, σ) into itself. 
A corollary of Theorems 2.1.1-2.1.4 worth mentioning at this stage pertains to
the transpose Cauchy-Clifford singular integral operator C # from (1.6.1). To put
matters in perspective, recall that the principal-value singular integral operator C #
from (1.6.1), when Ω ⊆ Rn is a UR domain and σ := H n−1 ∂Ω, may be expressed
at σ-a.e. x ∈ ∂Ω as
∫
1 x−y
C # f (x) = −ν(x) lim+ f (y) dσ(y) . (2.1.60)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

As such, item (6) of [70, Theorem 2.3.2] guarantees that C # , originally acting on
Lebesgue spaces on ∂Ω (as in (1.6.4)), extends uniquely to a linear and bounded
operator
 
C # : H p (∂Ω, σ) ⊗ Cn −→ L p (∂Ω, σ) ⊗ Cn, p ∈ n−1 n ,∞ . (2.1.61)

We may further interpolate (based on [69, Theorem 4.3.1] and [68, (6.2.48)]) to
obtain a linear and bounded operator

C # : H p,q (∂Ω, σ) ⊗ Cn −→ L p,q (∂Ω, σ) ⊗ Cn,

  (2.1.62)
for p ∈ n−1n , ∞ and q ∈ (0, ∞].

This being said, due to its special algebraic nature, the operator C # turns out to map
Hardy spaces (respectively, Lorentz-based Hardy spaces) into themselves in a linear
and bounded fashion. This, along with other related properties, are discussed in the
theorem below.
Theorem 2.1.5 Let Ω ⊆ Rn be a UR domain and set σ := H n−1 ∂Ω. Then C # ,
originally considered on Lebesgue spaces as in item (ii) of Proposition 1.6.1, extends
to a linear and bounded operator in the context
 
C # : H p (∂Ω, σ) ⊗ Cn −→ H p (∂Ω, σ) ⊗ Cn, ∀p ∈ n−1
n ,∞ (2.1.63)
 
and the operators C # corresponding to various values of p ∈ n−1 n , ∞ are compatible
with one another. Also,
 
for each p ∈ n−1n , ∞ , the composition to the left of C from (2.1.63) with
#

the L -filtering operator H : H (∂Ω, σ) ⊗ Cn −→ L p (∂Ω, σ) ⊗ Cn

p p
(2.1.64)
(cf. [69, (4.9.2)]) is C # : H p (∂Ω, σ) ⊗ Cn → L p (∂Ω, σ) ⊗ Cn from
(2.1.61).
In addition,
 2  n−1 
C# = 14 I on H p (∂Ω, σ) ⊗ Cn with p ∈ n ,∞ . (2.1.65)
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 249
 n−1 
More generally, for each p ∈ n , ∞ and q ∈ (0, ∞] the operator

C # : H p,q (∂Ω, σ) ⊗ Cn −→ H p,q (∂Ω, σ) ⊗ Cn (2.1.66)

is well defined, linear bounded, satisfies

 2
C# = 14 I on H p,q (∂Ω, σ) ⊗ Cn, (2.1.67)

and
 
for each given exponents p ∈ n−1 n , ∞ and q ∈ (0, ∞], the composition to
the left of the operator C # defined in (2.1.66) with the filtering operator H :
H p,q (∂Ω, σ)⊗Cn −→ L p,q (∂Ω, σ)⊗Cn (cf. [69, (4.9.5) in Theorem 4.9.1])
is the operator C # : H p,q (∂Ω, σ) ⊗ Cn −→ L p,q (∂Ω, σ) ⊗ Cn considered
in (2.1.62).
(2.1.68)

Finally,

C # : L 1 (∂Ω, σ) ⊗ Cn −→ H 1,∞ (∂Ω, σ) ⊗ Cn (2.1.69)

is a well-defined, linear and bounded mapping, and

the composition to the left of the operator C # from (2.1.69) with the filtering
operator H : H 1,∞ (∂Ω, σ) ⊗ Cn −→ L 1,∞ (∂Ω, σ) ⊗ Cn (cf. [69, (4.9.5) in
Theorem 4.9.1]) is the operator C # : L 1 (∂Ω, σ) ⊗ Cn −→ L 1,∞ (∂Ω, σ) ⊗ Cn
considered in the principal-value sense (cf. (1.6.5)).
(2.1.70)

Before presenting the proof of this theorem we wish to note that since (2.1.67)
implies
 2
C# = 14 I on H 1,∞ (∂Ω, σ) ⊗ Cn, (2.1.71)

and since [69, (4.2.28)] gives

L 1 (∂Ω, σ) ⊗ Cn → H 1,∞ (∂Ω, σ) ⊗ Cn, (2.1.72)

we conclude that
 
C# C# f = 1
4 f for each f ∈ L 1 (∂Ω, σ) ⊗ Cn, (2.1.73)

where C # f is considered in the sense of (2.1.69), and the subsequent action of C #

on this distribution is taken in the sense of (2.1.66) with p = 1 and q = ∞.
Proof of Theorem 2.1.5 The claims made in regard to (2.1.63) follow from for-
mula (1.6.19), elaborating on the nature of the components of C # , and Theorem-
s 2.1.1-2.1.4. That the operators C # in (2.1.5) corresponding to various values of
250 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 
p ∈ n−1 n , ∞ are compatible with one another is a consequence of [69, Theo-
rem 4.4.3] and (1.6.1). Going further, that (2.1.66) is a well-defined linear and
bounded operator is implied by (2.1.63), item (ii) of Proposition 1.6.1, [69, (4.3.3)],
and real interpolation. Next, recall from [70, Proposition 2.5.32] that C 2 = 14 I on
L q (∂Ω, σ) ⊗ Cn with q ∈ (1, ∞). In concert with item (ii) in Proposition 1.6.1 this
gives
 2
C# = 14 I on L q (∂Ω, σ) ⊗ Cn with q ∈ (1, ∞). (2.1.74)

Fix now p ∈ n−1 n , 1 along with q ∈ (1, ∞). Then (2.1.74) implies that we have
 #2 1 p,q
C = 4 I on the space Hfin (∂Ω, σ) ⊗ Cn . Granted this, from [69, (4.4.114)] and
(2.1.63) we then conclude that
 2  n−1
C# = 14 I on H p (∂Ω, σ) ⊗ Cn with p ∈ n ,1 . (2.1.75)

Combining (2.1.74) and (2.1.75) we see that (2.1.65) holds.

Next, (2.1.64) is justified by reasoning as in the proof of (2.1.6), while (2.1.68)
is established much as (2.1.29). Taking p = 1 and q = ∞ in (2.1.66) gives that
C # maps H 1,∞ (∂Ω, σ) ⊗ Cn boundedly into itself. Having established this, we may
now conclude that C # in (2.1.69) is a well-defined, linear and bounded mapping on
account of [69, (4.2.28)] (used here with Σ := ∂Ω). Also, the claim in (2.1.70) may
be proved along the lines of (2.1.31).
Finally, there remains to prove (2.1.67). Suppose n−1n < p0 < p1 < ∞. From the
 #2
first part of the proof we know that C acts like 4 I both on H p0 (∂Ω, σ) ⊗ Cn
1

and on H (∂Ω, σ) ⊗ Cn . Since C on H 0 (∂Ω, σ) ⊗ Cn is compatible with C #

p 1 # p
 2
on H p1 (∂Ω, σ) ⊗ Cn , we conclude that, on the one hand, C # acts like 14 on
H p0 (∂Ω, σ) ⊗ Cn + H p1 (∂Ω, σ) ⊗ Cn . On the other hand, from [69, (1.3.41),
(4.3.3)] we see that

H p,q (∂Ω, σ) ⊗ Cn ⊆ H p0 (∂Ω, σ) ⊗ Cn + H p1 (∂Ω, σ) ⊗ Cn (2.1.76)

 2
whenever p ∈ (p0, p1 ) and q ∈ (0, ∞]. Thus, ultimately, we obtain C # = 14 I on
H p,q (∂Ω, σ) ⊗ Cn whenever
 p0 < p < p1 and 0 < q ≤ ∞. Given that p0, p1 have
n , ∞ , this finishes the proof of (2.1.67).
been arbitrarily chosen in n−1 

The Clifford algebra formalism allows us to consider the Riesz transforms bundled
together, into a single entity, we call the boundary-to-domain Clifford-Riesz
transform. Specifically, given any Ahlfors regular domain Ω ⊆ Rn , abbreviate
σ := H n−1 ∂Ω then define the action of the Clifford-Riesz transform on functions
f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
⊗ Cn as
∫
2 x−y
R C f (x) := f (y) dσ(y), ∀x ∈ Ω. (2.1.77)
ωn−1 ∂Ω |x − y| n
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 251

At least formally, the scalar components of the Clifford-Riesz transform agree, up

to normalization, with the first-order partial derivatives of the single layer operator.
The study of the Clifford-Riesz transform in arbitrary UR domains is taken up next.

Theorem 2.1.6 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. In relation to the Clifford-Riesz transform (2.1.77) the following
properties are true.

(a) With D denoting the Dirac operator (A.0.55), one has

   
D R C f = 0 in Ω, ∀ f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1 ⊗ Cn . (2.1.78)

(b) For each integrability exponent p ∈ (1, ∞) and each aperture parameter κ > 0
one has1 p
R C : L p (∂Ω, σ) ⊗ Cn −→ Nκ (Ω; σ) ⊗ Cn
(2.1.79)
linearly and boundedly.
Hence, for each given exponent p ∈ (1, ∞) and κ ∈ (0, ∞) there exists a constant
C = C(Ω, p, κ) ∈ (0, ∞) such that for each f ∈ L p (∂Ω, σ) ⊗ Cn one has
  
Nκ R C f  p ≤ C f L p (∂Ω,σ)⊗ C n (2.1.80)
L (∂Ω,σ)

Moreover, corresponding to p = 1, there exists a constant C = C(Ω, κ) ∈ (0, ∞)

such that   
Nκ R C f  1,∞ ≤ C f L 1 (∂Ω,σ)⊗ C n (2.1.81)
L (∂Ω,σ)

for each f ∈ L 1 (∂Ω, σ) ⊗ Cn . Finally, for each f ∈ L p (∂Ω, σ) ⊗ Cn with
p ∈ [1, ∞) and κ > 0 one has

  κ−n.t.  
ν RC f  = I − 2C # f at σ-a.e. point on ∂Ω, (2.1.82)
∂Ω

where C # is the principal-value singular integral operator from (1.6.1).


(c) Given any integrability exponent p ∈ n−1 n , 1 and aperture parameter κ > 0,
the Clifford-Riesz transform (2.1.79) extends uniquely to a linear and bounded
operator
p
R C : H p (∂Ω, σ) ⊗ Cn −→ Nκ (Ω; σ) ⊗ Cn . (2.1.83)
Specifically, if for each j ∈ {1, . . . , n} one introduces
1 xj
Φ j (x) := , ∀x = (x1, . . . , xn ) ∈ Rn \ {0}, (2.1.84)
ωn−1 |x| n
then said extension of (2.1.79) acts according to2

p
1 with Nκ (Ω; σ) defined as in [68, Proposition 8.3.5]
2 with duality brackets ·, · in the sense of [69, Theorem 4.6.1]
252 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

  n 
  
R C f (x) = 2 Φ j (x − ·)∂Ω , fJ e j eJ
(2.1.85)
 J j=1
for each f = J fJ eJ ∈ H p (∂Ω, σ) ⊗ Cn and x ∈ Ω,

in the case when ∂Ω is unbounded, and a similar formula in the case when
∂Ω is bounded
 (omitting taking the equivalence class, modulo constants, of
Φ j (x − ·)∂Ω this time). Hence, given any p ∈ n−1 n , ∞ and κ ∈ (0, ∞), the
Clifford-Riesz transform R C (defined as in (2.1.77), (2.1.85)) satisfies
  
Nκ R C f  p ≤ C f H p (∂Ω,σ)⊗ C n (2.1.86)
L (∂Ω,σ)

for each f ∈ H p (∂Ω, σ) ⊗ Cn , for some constant C = C(Ω, p, κ) ∈ (0, ∞)

independent of f . In addition,
 
D R C f = 0 in Ω, ∀ f ∈ H p (∂Ω, σ) ⊗ Cn . (2.1.87)
 
(d) For each f ∈ H p (∂Ω, σ) ⊗ Cn with p ∈ n−1 n , ∞ , the Clifford-Riesz transform
R C (defined as in (2.1.77), (2.1.85)) satisfies
   
ν • R C f = I − 2C # f in H p (∂Ω, σ) ⊗ Cn, (2.1.88)

where the Clifford bullet product is defined as in (A.0.167), and where C # is the
“transpose” Cauchy-Clifford operator from (2.1.63).

Proof The claim in item (a) is a straightforward consequence of definitions, bearing

in mind that the function Rn \ {0}  x → x/|x| n ∈ Cn is a null-solution of D. As
regards item (b), the claims in (2.1.79)-(2.1.81) are clear from [70, Theorem 2.4.1],
while the claim in (2.1.82) is implied by (2.1.77), [70, Theorem 2.5.1], and (1.6.1).
Alternatively, we may rely on (A.0.53), [70, (2.5.309)], and the observation that,
with ν denoting the geometric measure theoretic outward unit normal to Ω, we have
 
R C f = −2C(ν f ) in Ω, for each f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1
⊗ Cn . (2.1.89)

Moving on to item (c), the claims pertaining to (2.1.83)-(2.1.86) are consequences

of definitions and [70, Theorem 2.4.1], while (2.1.87) follows with the help of
Lemma 2.2.1. 
n , 1 . Thanks to item
To deal with the claim in item (d), assume first that p ∈ n−1
(c) and the discussion in [69, Example 10.2.14], the mapping
 
H p (∂Ω, σ) ⊗ Cn  f −→ ν • R C f ∈ H p (∂Ω, σ) ⊗ Cn (2.1.90)

is well defined, linear and bounded. From (2.1.63) we know that C # is also a well-
defined, linear and bounded operator from H p (∂Ω, σ) ⊗ Cn into itself. As such,
the claim in (2.1.88) follows as soon as we show that these two operators agree on
a dense subspace of H p (∂Ω, σ) ⊗ Cn . To this end, pick q ∈ (1, ∞) and consider
p,q
f ∈ Hfin (∂Ω, σ) ⊗ Cn . Then from [69, (10.2.104)] and (2.1.82) we see that for each
fixed aperture parameter κ > 0 we have
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 253

    κ−n.t.  
ν • RC f = ν RC f  = I − 2C # f . (2.1.91)
∂Ω

On account of [69, (4.4.114)] and the fact that the principal-value singular integral
operator C # from (1.6.1) acts in a compatible fashion with C # in (2.1.63), the desired
conclusion now follows. Finally, that the same conclusion holds when p ∈ (1, ∞) is
seen from (2.1.91) which remains valid when f ∈ L p (∂Ω, σ) ⊗ Cn . 

There are many well-documented instances when the space L ∞ is inadequate and
BMO turns out to be the correct substitute. For example, the Riesz transforms in the
Euclidean space are not bounded on L ∞ but (their modified versions) are bounded on
BMO, and the latter is the smallest space containing L ∞ with this property. We wish
to show that the principal-value double layer operator associated with a given weakly
elliptic system in a UR domain may be suitably extended to the space of functions of
bounded mean oscillations in a manner that renders that map continuous from said
space into itself. This is made precise in our next theorem.

Theorem 2.1.7 Let Ω ⊆ Rn (where  αβ n  ∈ N, n ≥ 2) be a UR domain and define

σ := H n−1 ∂Ω. Also, let L = ar s ∂r ∂s 1≤α,β ≤M be a homogeneous, weakly elliptic,
constant (complex) coefficient, second-order M × M system in Rn (for some M ∈ N).
In this context, consider the boundary layer potential operators K and K # , associated
with the system L and the set Ω as in (1.3.68) and (1.3.72), respectively. Finally,
recall the modified boundary-to-boundary double layer potential operator Kmod from
(1.8.24)-(1.8.25).
Then the operators
M M
Kmod : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) (2.1.92)

and
 M  M
Kmod : BMO(∂Ω, σ) ∼ −→ BMO(∂Ω, σ) ∼ defined as
(2.1.93)
M
Kmod [ f ] := Kmod f for each function f ∈ BMO(∂Ω, σ)

are well-defined, linear, and bounded. Moreover, in the case when ∂Ω is bounded, a
scenario in which one has
M
 M
BMO(∂Ω, σ) ⊆ L p (∂Ω, σ) , (2.1.94)
0<p<∞

M
the operator K acting on the Lebesgue scale L p (∂Ω, σ) with p ∈ (1, ∞) has
M
BMO(∂Ω, σ) as an invariant subspace, and its restriction
M M
K : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) (2.1.95)

is a well-defined, linear and bounded operator. Finally,

254 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

the (real) transpose of the operator K # from (2.1.4) with p = 1 is,

respectively, Kmod from (2.1.93) if ∂Ω is unbounded, and K from (2.1.96)
(2.1.95) if ∂Ω is bounded,
M M
that is, for each f ∈ BMO(∂Ω, σ) and g ∈ H 1 (∂Ω, σ) one has (with duality
brackets as in [69, Theorem 4.6.1])
   
Kmod f , g = [ f ], K # g if ∂Ω is unbounded, (2.1.97)
   
K f , g = f , K # g if ∂Ω is bounded. (2.1.98)

Proof Recall from [68, (7.4.118)] that

  σ(x)  M
M
BMO(∂Ω, σ) ⊆ L 1 ∂Ω, . (2.1.99)
1 + |x| n
M
If we now pick an arbitrary function f = ( fα )1≤α ≤M ∈ BMO(∂Ω, σ) , from
(2.1.99), [70, (2.3.35)], (1.8.24), and [68, (7.4.105)] we see that
 M
p
Kmod f ∈ Lloc (∂Ω, σ) . (2.1.100)
1≤p<∞

Let us now consider an arbitrary C M -valued (1, ∞)-atom a = (aγ )1≤γ ≤M on ∂Ω with
vanishing moment3. Concretely, a is C M -valued, σ-measurable
 function
 defined on
∂Ω, with the property that there exist xo ∈ ∂Ω and r ∈ 0, 2 diam(∂Ω) such that
  −1
supp a ⊆ B(xo, r) ∩ ∂Ω, a [L ∞ (∂Ω,σ)] M ≤ σ B(xo, r) ∩ ∂Ω ,
∫ (2.1.101)
and a dσ = 0 ∈ C M .
∂Ω

From (2.1.8) and [69, (4.5.33)] we know that

∫
| f ||K # a| dσ < +∞, (2.1.102)
∂Ω
∫
and we make the claim that we also have ∂Ω
|Kmod f ||a| dσ < +∞ as well as
∫ ∫
Kmod f , a dσ =  f , K # a dσ. (2.1.103)
∂Ω ∂Ω

Indeed, this is a direct consequence of (1.8.43). A direct proof of (2.1.103) goes as

follows. First, bearing (2.1.102) in mind, we write (using the summation convention
over repeated indices, as in the past)

3 this is always the case if ∂Ω is unbounded

2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 255
∫ ∫ ∫
fα (K # a)α dσ = fα (K # a)α dσ + fα (K # a)α dσ
∂Ω ∂Ω∩B(x o ,2r) ∂Ω\B(x o ,2r)

=: I + II, (2.1.104)

with the last equality defining I and II. In relation to these, note that on account of
item (iii) in Theorem 1.5.1 and [68, (7.4.105)] we may re-write I as
∫ ∫ 
   
I= f · 1∂Ω∩B(xo,2r) α (K a)α dσ =
#
K f · 1∂Ω∩B(xo,2r) aγ dσ
γ
∂Ω ∂Ω
∫ 
 
= Kmod f · 1∂Ω∩B(xo,2r) aγ dσ (2.1.105)
γ
∂Ω
   
since K f · 1∂Ω∩B(xo,2r) differs from Kmod f · 1∂Ω∩B(xo,2r) by a constant (from C M )
on ∂Ω, thanks to [70, (2.3.34)] and the fact that the atom has integral zero (cf. the
last property in (2.1.101)). Also,
256 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
∫ ∫
βα
II = fα (x) νs (x)ar s (∂r Eγβ )(x − y)aγ (y) dσ(y) dσ(x)
∂Ω\B(x o ,2r) ∂Ω

∫ ∫
βα (rγβ)
= fα (x) νs (x)ar s k ε (x − y)aγ (y) dσ(y) dσ(x)
∂Ω\B(x o ,2r) ∂Ω

∫ ∫
βα  (rγβ)
= fα (x) νs (x)ar s k ε (x − y)
∂Ω\B(x o ,2r) ∂Ω

(rγβ) 
+ k1 (−x) aγ (y) dσ(y) dσ(x)

∫ ∫
βα  (rγβ)
= fα (x) νs (x)ar s k ε (x − y)
∂Ω\B(x o ,2r) ∂Ω∩B(x o ,r)

(rγβ) 
+ k1 (−x) aγ (y) dσ(y) dσ(x)

∫ ∫
βα  (rγβ)
= νs (x)ar s k ε (x − y)
∂Ω∩B(x o ,r) ∂Ω\B(x o ,2r)

(rγβ) 
+ k1 (−x) fα (x) dσ(x) aγ (y) dσ(y)

∫ ∫
βα  (rγβ)
= lim+ νs (x)ar s k ε (x − y)
ε→0 ∂Ω\B(x o ,2r)
∂Ω

(rγβ) 
+ k1 (−x) fα (x) dσ(x) aγ (y) dσ(y)

∫ 
 
= Kmod f · 1∂Ω\B(xo,2r) (y) aγ (y) dσ(y). (2.1.106)
γ
∂Ω

The first equality in (2.1.106) is implied by (1.3.72) bearing in mind that, thanks
to the first property in (2.1.101), the variables x, y are uniformly separated. The
second equality in (2.1.106) uses (1.8.25) and is valid for each choice ε ∈ (0, r).
The third equality in (2.1.106) is a consequence of the cancelation property of the
atom (cf. the last property in (2.1.101)), while the fourth equality in (2.1.106) is
seen from the first property in (2.1.101). The fifth equality in (2.1.106) follows from
Fubini’s Theorem whose applicability is presently ensured by the fact that the double
integral is absolutely convergent, thanks to the properties listed in the first line of
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 257

(2.1.101), the estimate in [70, (2.3.117)] (with a constant which stays bounded for
x in a compact subset of ∂Ω), and (2.1.99). The sixth equality in (2.1.106) uses the
fact that the inner integral is actually independent of ε ∈ (0, r), and also the support
condition for the atom. Finally, the last equality in (2.1.106) is seen from (1.8.24).
At this stage, from (2.1.104)-(2.1.106) we conclude that
∫ ∫
 
fα (K # a)α dσ = Kmod f γ aγ dσ, (2.1.107)
∂Ω ∂Ω

from which (2.1.103) follows.

Let us now suppose that ∂Ω is unbounded. Then, on the one hand, based on
(5.1.232) and [69, Proposition 4.8.6] (whose applicability with g := K # a is ensured
by (2.1.4) and (2.1.102)), for each C M -valued (1, ∞)-atom a = (aγ )1≤γ ≤M on ∂Ω
we may write
∫
   
Kmod f γ aγ dσ = [ f ], K # a (2.1.108)
∂Ω

where ·, · denotes the duality bracket between the John-Nirenberg space of func-
tions of bounded mean oscillations on ∂Ω, modulo constants, and the Hardy space
H 1 on Σ (cf. [69, Theorem 4.6.1]). In concert with [69, (4.6.9)], this permits us to
estimate
∫     
 
 K f aγ dσ  =  [ f ], K # a 
mod γ
∂Ω

≤C f [BMO(∂Ω,σ)] M · K#a [H 1 (∂Ω,σ)] M

≤C f [BMO(∂Ω,σ)] M , (2.1.109)

where the last inequality is based on (2.1.4) and [69, (4.5.5)-(4.5.6)]. On the other
hand, from (2.1.100), (A.0.20), [68, Proposition 7.4.12], and (2.1.101) we see that
 
K f  (2.1.110)
mod [BMO(∂Ω,σ)] M
 ∫
   
 
≤ C · sup  Kmod f γ aγ dσ  : a = (aγ )1≤γ ≤M
∂Ω

is a C -valued (1, ∞)-atom on ∂Ω .
M

Together, (2.1.109) and (2.1.110) give

 
K f  ≤C f [BMO(∂Ω,σ)] M , (2.1.111)
mod [BMO(∂Ω,σ)] M

for some C ∈ (0, ∞) independent of f . Hence, the operator Kmod is a well-defined,

M
linear, bounded mapping from BMO(∂Ω, σ) into itself when ∂Ω is unbounded.
258 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces


In view of the format of the norm on BMO(∂Ω, σ) (cf. [68, (7.4.95)]) and (1.8.28), we
then also conclude that the mapping in (2.1.93) is well defined, linear, and bounded,
when ∂Ω is unbounded. With these properties in hand, proving (2.1.97) comes down
(thanks to the last property in [69, (4.4.114)] and the continuity of K # on the Hardy
scale) to showing that    
Kmod f , a = [ f ], K # a (2.1.112)
M
for each f ∈ BMO(∂Ω, σ) and each C M -valued (1, ∞)-atom a on ∂Ω. This,
however, is clear from (2.1.108) and the duality result from [69, Theorem 4.6.1,
M
(4.6.8)] (bearing in mind that we already know that Kmod f ∈ BMO(∂Ω, σ) ).
At this stage, all claims pertaining to (2.1.92)-(2.1.93) have been justified when
∂Ω is an unbounded set. As such, there remains to treat the case when ∂Ω is a bounded
set. In such a scenario, (2.1.94) holds thanks to [68, (7.4.105)]. As a consequence of
M
this and [70, (2.3.34)], for each function f = ( fα )1≤α ≤M ∈ BMO(∂Ω, σ) and
each a = (aγ )1≤γ ≤M as in (2.1.101) we have
∫ ∫
fα (K # a)α dσ = (K f )γ aγ dσ. (2.1.113)
∂Ω ∂Ω

Then the same argument which, starting with (5.1.232), has produced (5.1.239)
presently gives
M
Kf . ≤C f for each f ∈ BMO(∂Ω, σ) .
[BMO(∂Ω,σ)] M
[BMO(∂Ω,σ)] M
(2.1.114)
From this, (1.5.12), and [69, (4.6.18)] we then conclude that K is a well-defined,
linear, and bounded operator in the context of (2.1.95).
Next, (2.1.94) and [70, (2.3.34)] currently give that Kmod f = [K f ] for each
M
f ∈ BMO(∂Ω, σ) . With this in hand, it follows from (2.1.95) that the mapping in
(2.1.93) is well defined, linear, and bounded. Granted this, for each f ∈ BMO(∂Ω, σ)
we may estimate, bearing in mind (A.0.20), the fact that C f := Kmod f − K f is a
constant as in [70, (2.3.34)], and [69, (4.6.18)]:
   
K f  ≤ Kmod f − K f [BMO(∂Ω,σ)] M + K f BMO(∂Ω,σ)
mod [BMO(∂Ω,σ)] M

≤ σ(∂Ω)|C f | + C f [BMO(∂Ω,σ)] M

≤C f [L 1 (∂Ω,σ)] M +C f [BMO(∂Ω,σ)] M

≤C f [BMO(∂Ω,σ)] M . (2.1.115)

This completes the proof of the fact that the mapping (2.1.92) is well defined and
M
bounded. Next, for each f = ( fα )1≤α ≤M ∈ BMO(∂Ω, σ) and each C M -valued
(1, ∞)-atom a = (aγ )1≤γ ≤M on ∂Ω we may write
∫ ∫
   
K f, a = (K f )γ aγ dσ = fα (K # a)α dσ = f , K # a (2.1.116)
∂Ω ∂Ω
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 259

by [69, Proposition 4.8.6] (keeping in mind (2.1.94) and (1.5.12)), (1.2.7), and
(2.1.4). This proves (2.1.98) and completes the proof of Theorem 2.1.7. 

Remark 2.1.8 The results in Theorem 2.1.7 are applicable to all boundary-to-
boundary double layer potential operators K, K # from Examples 1.4.9-1.4.20 and
their modified versions Kmod from Examples 1.8.4-1.8.7.

In particular, Theorem 2.1.7 permits us to verify the integral identities claimed

below.

Corollary 2.1.9 Suppose Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is a UR domain,

and abbreviate σ := H n−1 ∂Ω. Also, suppose L is a homogeneous, weakly elliptic,
constant (complex) coefficient, second-order M × M system in Rn (for some M ∈ N),
and consider the boundary layer potential operator K # associated with the system L
and the set Ω as in (1.3.72).
M
Then for each function f ∈ H 1 (∂Ω, σ) one has
M M
f ∈ L 1 (∂Ω, σ) , K # f ∈ L 1 (∂Ω, σ) , (2.1.117)

and, with all integrals involved absolutely convergent,

∫
⎧
⎪ 1
⎪
⎪ + f dσ if Ω is bounded,
∫ ⎪
⎪ 2 ∂Ω
⎪
⎨
⎪ ∫
K # f dσ = 1 (2.1.118)
∂Ω ⎪
⎪ − f dσ if Ω is unbounded and ∂Ω is bounded,
⎪
⎪ 2 ∂Ω
⎪
⎪
⎪ if ∂Ω is unbounded.
⎩ 0
M
Proof Pick f ∈ H 1 (∂Ω, σ) arbitrary. Then the claims in (2.1.117) are immedi-
ate consequences of [69, (4.2.10)] and (2.1.4) (used with p := 1). Next, the branch
of (2.1.118) corresponding to the case when ∂Ω is unbounded is seen from Theo-
rem 2.1.1 and [69, (4.4.9)]. Suppose now that ∂Ω is bounded. Then for each λ ∈ C M
we may write
#∫ $ ∫  
K f dσ , λ =
#
K # f , λ dσ = [H 1 (∂Ω,σ)] M K # f , λ [BMO(∂Ω,σ)] M
∂Ω ∂Ω
 
= [H 1 (∂Ω,σ)] M f , Kλ [BMO(∂Ω,σ)] M
 
= [H 1 (∂Ω,σ)] M f , ± 12 λ [BMO(∂Ω,σ)] M
∫
1# $
=± f dσ , λ (2.1.119)
2 ∂Ω
where the ± signs correspond to Ω being bounded and unbounded, respectively.
Indeed, the first equality in (2.1.119) is trivial, the second one is a consequence
of Theorem 2.1.1 and [69, Proposition 4.8.6], the third one comes from (2.1.98),
260 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

the fourth one is guaranteed by (1.5.21), and the fifth one is once again implied by
[69, Proposition 4.8.6]. Having proved (2.1.119), on account of the arbitrariness of
λ ∈ C M we then conclude that the branches of (2.1.118) corresponding to the case
when ∂Ω is bounded are valid as well. 

A companion result to Theorem 2.1.7 is Theorem 2.1.10 below, to the effect that
the action of the principal-value double layer operator associated with a given weakly
elliptic system in a UR domain may also be extended, in a linear and bounded fashion,
to the entire scale of Hölder spaces on the boundary. See also Proposition 2.1.12 for
a result of similar flavor.

Theorem 2.1.10 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). In this context,
consider the boundary layer potential operators K and K # , associated with the
system L and the set Ω as in (1.3.68) and (1.3.72), respectively. In addition, bring
back the operator Kmod defined in (1.8.24). Lastly, consider
 
η ∈ (0, 1) and define p := n−1+η
n−1
∈ n−1n ,1 . (2.1.120)

or, equivalently, start with

  1 
p ∈ n−1 n , 1 and define η := (n − 1) p − 1 ∈ (0, 1). (2.1.121)

Then the operators

. .
Kmod : 𝒞η (∂Ω) −→ 𝒞η (∂Ω)
M M
(2.1.122)

and
.  M .  M
Kmod : 𝒞η (∂Ω) ∼ −→ 𝒞η (∂Ω) ∼ defined as
.η M
(2.1.123)
Kmod [ f ] := Kmod f for each function f ∈ 𝒞 (∂Ω)

are well defined, linear, and bounded. Moreover, in the case when ∂Ω is bounded, a
scenario in which one has
. 
𝒞η (∂Ω) = 𝒞η (∂Ω)
M M M
⊆ L q (∂Ω, σ) , (2.1.124)
0<q ≤∞

M
the operator K acting on the Lebesgue scale L q (∂Ω, σ) with q ∈ (1, ∞) has
M
𝒞η (∂Ω) as an invariant subspace, and its restriction

K : 𝒞η (∂Ω) −→ 𝒞η (∂Ω)
M M
(2.1.125)

is a well-defined, linear and bounded operator. Finally,

2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 261
 n−1 
the (real) transpose of the operator K # from (2.1.4) with p ∈ n , 1
is, respectively, Kmod from (2.1.123) if ∂Ω is unbounded, and K from (2.1.126)
(2.1.125) if ∂Ω is bounded,
. M M
that is, for each f ∈ 𝒞η (∂Ω) and g ∈ H p (∂Ω, σ) one has (with duality
brackets as in [69, Theorem 4.6.1])
   
Kmod f , g = [ f ], K # g if ∂Ω is unbounded, (2.1.127)
   
K f , g = f , K # g if ∂Ω is bounded. (2.1.128)

Proof All results are largely dealt with as their counterparts treated in Theorem 2.1.7.
More specifically, from [68, (7.4.119)] we know that
.  σ(x) 
𝒞η (∂Ω) ⊆ L 1 ∂Ω, . (2.1.129)
1 + |x| n
. M
Then for any function f = ( fα )1≤α ≤M ∈ 𝒞η (∂Ω) , from (2.1.129), [70, (2.3.35)],
and (1.8.24) we see that
 M
q
Kmod f ∈ Lloc (∂Ω, σ) . (2.1.130)
1≤q<∞

Also, from (2.1.8) and [69, Lemma 4.5.5] we know that

∫
| f ||K # a| dσ < +∞. (2.1.131)
∂Ω

With this in hand, the same argument that has produced (2.1.107) currently gives
∫ ∫
 
fα (K a)α dσ =
#
Kmod f γ aγ dσ (2.1.132)
∂Ω ∂Ω

for each C M -valued (p, ∞)-atom a = (aγ )1≤γ ≤M on ∂Ω with vanishing moment.
Granted this, we may reason as in (2.1.108)-(2.1.111), now relying on [69, Proposi-
tion 4.8.7] in place of [69, Proposition 4.8.6], and [68, Proposition 7.4.8] in place of
[68, Proposition 7.4.12], to conclude that there exists C ∈ (0, ∞) such that
  .
K f [𝒞. η (∂Ω)] M ≤ C f . for each f ∈ 𝒞η (∂Ω)
M
. (2.1.133)
mod [𝒞η (∂Ω)] M

From this and (1.8.28), the claims concerning the operators (2.1.122)-(2.1.123)
follow. Once these have been established, proving (2.1.127) reduces (in view of the
last property in [69, (4.4.114)]) to checking that
   
Kmod f , a = [ f ], K # a for each
. M (2.1.134)
function f ∈ 𝒞η (∂Ω) and each (p, ∞)-atom a on ∂Ω.
262 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
. M
Since we already know that Kmod f ∈ 𝒞η (∂Ω) , from the duality result in [69,
Theorem 4.6.1, (4.6.8)] we conclude that
∫
 
Kmod f , a = (Kmod f )γ aγ dσ. (2.1.135)
∂Ω

Observe that since a is a (p, ∞)-atom, hence also a multiple of some (1, ∞)-atom,
from (2.1.4) we have
M M
K # a ∈ H 1 (∂Ω, σ) ∩ H p (∂Ω, σ) . (2.1.136)

Thanks to (2.1.131) and (2.1.136), [69, Proposition 4.8.7] applies and gives
∫
 
[ f ], K a =
#
fα (K # a)α dσ. (2.1.137)
∂Ω

Now (2.1.134) follows from (2.1.135), (2.1.137), and (2.1.132).

In the case when ∂Ω is bounded, starting with the fact that (2.1.113) holds for
M
each f ∈ 𝒞η (∂Ω) and each C M -valued (p, ∞)-atom a = (aγ )1≤γ ≤M on ∂Ω, the
same argument which has proved (2.1.133), outlined above, now gives

for each f ∈ 𝒞η (∂Ω)

M
Kf . ≤C f [𝒞η (∂Ω)] M . (2.1.138)
[𝒞η (∂Ω)] M

M
In addition, for each function f ∈ 𝒞η (∂Ω) and σ-a.e. point x ∈ ∂Ω we may use
(1.5.21) and [68, (7.2.5)] (with X := ∂Ω, r := 2 diam(∂Ω), d := n − 1, and δ := η)
to estimate
  
|(K f )(x)| ≤  K( f − f (x)) (x) + 12 | f (x)|
∫
dσ(y)
≤ C f [𝒞. η (∂Ω)] M + 12 | f (x)|
∂Ω |x − y|
n−1−η

η
≤ C diam(∂Ω) f . + 12 sup | f |, (2.1.139)
[𝒞η (∂Ω)] M
∂Ω

for some C ∈ (0, ∞) which depends only on L, n, and the lower ADR constant of
Ω. In view of this and the fact that, as seen from (2.1.138), the operator K maps
M
𝒞η (∂Ω) into continuous functions on ∂Ω, we conclude that there exists some
constant C ∈ (0, ∞) with the property that

for each f ∈ 𝒞η (∂Ω)

M
sup |(K f )(x)| ≤ C f [𝒞η (∂Ω)] M . (2.1.140)
x ∈∂Ω

Ultimately, (2.1.140) and (2.1.138) imply

for each f ∈ 𝒞η (∂Ω)

M
Kf [𝒞η (∂Ω)] M ≤C f [𝒞η (∂Ω)] M . (2.1.141)
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 263

This proves that the operator K is well defined and bounded in the context of
(2.1.125). Finally, (2.1.128) is justified as before, based on density (cf. [69, (4.4.114)])
M
and the fact that (2.1.116) continues to hold for each function f ∈ 𝒞η (∂Ω) and
each C M -valued (p, ∞)-atom a on ∂Ω (cf. (2.1.125), the bounded set version of the
duality result from [69, Theorem 4.6.1, (4.6.8)], item (iii) in Theorem 1.5.1, and
[69, Corollary 4.8.11], keeping in mind (2.1.8)). The proof of Theorem 2.1.10 is
therefore complete. 

Similar results as in the first part of Theorem 2.1.10 are also valid on our scale of
vanishing Hölder spaces, introduced in [69, §3.2]. A formal statement is contained
in our next corollary.

Corollary 2.1.11 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). In this context,
consider the boundary-to-boundary layer potential operator K and its modified
version Kmod associated with the system L and the set Ω in as in (1.3.68) and
(1.8.24), respectively. Finally, fix an arbitrary exponent

η ∈ (0, 1) (2.1.142)
.
and recall the (homogeneous and inhomogeneous) vanishing Hölder spaces 𝒞ηvan (∂Ω)
and 𝒞ηvan (∂Ω) defined as in (A.0.48) and (A.0.49), respectively, with Σ := ∂Ω.
Then the operators
. .
Kmod : 𝒞ηvan (∂Ω) −→ 𝒞ηvan (∂Ω)
M M
(2.1.143)

and
.  M .  M
Kmod : 𝒞ηvan (∂Ω) ∼ −→ 𝒞ηvan (∂Ω) ∼ defined as
.η M
(2.1.144)
Kmod [ f ] := Kmod f for each function f ∈ 𝒞van (∂Ω)

are well defined, linear, and bounded. Moreover, in the case when ∂Ω is bounded, a
scenario in which one has
.
𝒞ηvan (∂Ω) = 𝒞ηvan (∂Ω) ⊆ 𝒞η (∂Ω)
M M M
, (2.1.145)
M
the operator K acting on the ordinary Hölder space 𝒞η (∂Ω) (as discussed in
M
Theorem 2.1.10) has 𝒞ηvan (∂Ω) as an invariant subspace, and its restriction

K : 𝒞ηvan (∂Ω) −→ 𝒞ηvan (∂Ω)

M M
(2.1.146)

is a well-defined, linear and bounded operator.

Proof All results are consequences of Theorem 2.1.10, the definitions made in
(A.0.48), and (A.0.49), as well as the density result from [69, Theorem 3.2.2]. 
264 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

In Theorem 2.1.10 we have established that the principal-value double layer

potential operators is a well-defined linear and bounded operator on the scale of
Hölder spaces considered on boundaries of UR domains. Remarkably, such a result
continues to be true when the uniform rectifiability assumption for the boundary is
replaced by a mere upper Ahlfors regularity demand. This is made precise in the
proposition below.

Proposition 2.1.12 Let Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) be a Lebesgue

measurable set with the property that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors
regular (in particular, [68, (5.6.35)] guarantees that Ω is a set of locally finite
perimeter). Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit
 αβ  σ := H ∂Ω. Also, for some M ∈ N, consider a
normal to Ω and abbreviate n−1

coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the property that
1≤α,β ≤M
the M × M homogeneous second-order system L = L A associated with A in Rn as
in (1.3.2) is weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Finally,
let E = (Eγβ )1≤γ,β ≤M be the matrix-valued fundamental solution associated with L
as in [70, Theorem 1.4.2]. Then the following statements are true.
M
(i) For each function f = ( fα )1≤α ≤M ∈ 𝒞η (∂∗ Ω) with η ∈ (0, 1) the limit
∫
βα
K f (x) := − lim+ νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y)
ε→0
y ∈∂∗ Ω 1≤γ ≤M
|x−y |>ε
(2.1.147)

exists at σ-a.e. point x ∈ ∂∗ Ω. In fact, at σ-a.e. point x ∈ ∂∗ Ω the principal-value

double layer potential defined in (2.1.147) may be expressed as
∫
βα  
K f (x) = − νs (y)ar s (∂r Eγβ )(x − y) fα (y) − fα (x) dσ(y)
∂∗ Ω 1≤γ ≤M

+ 12 f (x) if Ω is bounded,
+
− 12 f (x) if Ω is unbounded.
(2.1.148)
M
(ii) If for each f = ( fα )1≤α ≤M ∈ 𝒞η (∂∗ Ω) with η ∈ (0, 1) one redefines K f on
a σ-nullset contained in ∂∗ Ω as being the function in (2.1.148) at each point
x ∈ ∂∗ Ω, the assignment f → K f thus defined induces a well-defined, linear,
and bounded operator

K : 𝒞η (∂∗ Ω) −→ 𝒞η (∂∗ Ω)
M M
(2.1.149)

for each η ∈ (0, 1), with the property that for each λ ∈ C M one has
2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 265

+ 12 λ on ∂∗ Ω, if Ω is bounded,
Kλ ≡ (2.1.150)
− 12 λ on ∂∗ Ω, if Ω is unbounded.

(iii) Let D be the boundary-to-domain double layer operators associated with the
set Ω and the coefficient tensor A as in (1.3.18). Also, fix an arbitrary aperture
parameter κ > 0. Then, under the additional assumption that the set Ω is open,
the following jump-formula (where I denotes the identity operator) is valid:
M
for each f ∈ 𝒞η (∂∗ Ω) with η ∈ (0, 1) one has
 
 κ−n.t.
  (2.1.151)
Df  = 12 I + K f at σ-a.e. point on ∂∗ Ω.
∂Ω

Proof For each α, γ ∈ {1, . . . , M } define (with the summation convention over
 βα 
repeated indices in effect) kαγ := − ar s ∂r Eγβ 1≤s ≤n ∈ 𝒞∞ (Rn \ {0}) which is
n

an odd, positive homogeneous of degree 1 − n, divergence-free vector field (cf. [70,

Theorem 1.4.2]). Thanks to [70, (1.4.25)], we also have
∫
 
ϑαγ := ω, kαγ (ω) dH n−1 (ω)
S n−1
∫
βα
=− ωs ar s (∂r Eγβ )(ω) dH n−1 (ω) = −δαγ . (2.1.152)
S n−1

If Zαγ is the integral operator associated as in [70, (2.5.157)] with kαγ , then it
M
becomes apparent from (1.3.18) that for each f = ( fα )1≤α ≤M ∈ L 1 (∂∗ Ω, σ) we
have
 
D f = Zαγ fα 1≤γ ≤M in Ω. (2.1.153)

Also, if Zαγ is the principal-value singular integral operator associated as in [70,

(2.5.166)] with kαγ , then from (2.1.147) it follows that for each vector-valued function
M
f = ( fα )1≤α ≤M ∈ L 1 (∂∗ Ω, σ) we have
 
K f = Zαγ fα 1≤γ ≤M at σ-a.e. point on ∂∗ Ω. (2.1.154)

Granted (2.1.152)-(2.1.154), all desired claims follow from [70, Proposition 2.5.16].
Moving on, Theorem 2.1.7, Theorem 2.1.10, and [69, Theorem 3.1.3] make
it possible to handle the (modified version of the) principal-value double layer
potential operator on the Sarason space, of functions of vanishing mean oscillations,
considered on the boundary of an arbitrary UR domain.
Corollary 2.1.13 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain and
set σ := H n−1 ∂Ω. Let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N), and recall the
boundary layer potential operator K, and its modified version Kmod , associated with
the system L and the set Ω as in (1.3.68) and (1.8.24), respectively.
266 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
M
Then the operator Kmod acting on BMO(∂Ω, σ) (cf. Theorem 2.1.7) has
M
VMO(∂Ω, σ) as an invariant subspace, hence its restriction
M M
Kmod : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (2.1.155)

is a well-defined, linear and bounded operator. In addition, the following operator

is also well defined, linear and bounded:
 M  M
Kmod : VMO(∂Ω, σ) ∼ −→ VMO(∂Ω, σ) ∼ defined as
(2.1.156)
M
Kmod [ f ] := Kmod f for each function f ∈ VMO(∂Ω, σ) .
M
Finally, if ∂Ω is compact then the operator K acting on BMO(∂Ω, σ) (cf.
M
(2.1.95)) has VMO(∂Ω, σ) as an invariant subspace, hence its restriction
M M
K : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (2.1.157)

is a well-defined, linear and bounded operator, and

the (real) transpose of the operator (2.1.157) is the operator K # :

M M (2.1.158)
H 1 (∂Ω, σ) −→ H 1 (∂Ω, σ) (cf. Theorem 2.1.1).

Proof This is a consequence of Theorem 2.1.7, Theorem 2.1.10, [69, Theorem 3.1.3],
[69, (3.1.50)], Theorem 2.1.1, (1.8.28), and [69, (4.6.22)-(4.6.23)]. 

We wish to note that the same template used in the proofs of Theorem 2.1.7,
Theorem 2.1.10, Corollary 2.1.11, and Corollary 2.1.13 can handle other important
classes of operators. To provide further concrete example, suppose Ω ⊆ Rn is an
UR domain and abbreviate σ := H n−1 ∂Ω. Also, fix a complex-valued function
1 n n 
b ∈ Lloc (R , L ) with b Rn \{0} ∈ 𝒞 N (Rn \ {0}) for some sufficiently large number
N = N(n) ∈ N, and such that ∇b is odd and positive homogeneous of degree 1 − n in
mod
Rn \ {0}. Assume first that ∂Ω is unbounded and bring in the operators Tjk defined
mod
as in (1.8.168)-(1.8.170). Then for each j, k ∈ {1, . . . , n} the operators Tjk map
constant functions on ∂Ω into constant functions on ∂Ω and, with brackets denoting
equivalence classes modulo constants, the naturally induced maps
mod
Tjk 
: BMO(∂Ω, 
σ) −→ BMO(∂Ω, σ),
mod mod
defined as Tjk [ f ] := Tjk f ] for each (2.1.159)
 
f ∈ BMO(∂Ω, σ) ⊂ L 1 ∂Ω, 1+σ(x)
|x | n ,

and, for every η ∈ (0, 1),

2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 267
mod .η  .η 
Tjk : 𝒞 (∂Ω) ∼ −→ 𝒞 (∂Ω) ∼,
mod mod
defined as Tjk [ f ] := Tjk f ] for each (2.1.160)
.  
f ∈ 𝒞η (∂Ω) ⊂ L 1 ∂Ω, 1+σ(x)
|x | ,
n

are all well defined, linear, and bounded. Boundedness results similar to (2.1.160)
are also valid on homogeneous vanishing Hölder spaces (see Corollary 2.1.11). In
addition, with the duality brackets as in [69, Theorem 4.6.1], for each pair of integers
j, k ∈ {1, . . . , n} we have

mod   
Tjk f , g = − [ f ], Tjk# g for each
  (2.1.161)
f ∈ BMO(∂Ω, σ) ⊂ L 1 ∂Ω, 1+σ(x)
|x | n and g ∈ H 1 (∂Ω, σ),

as well as
 mod   
Tjk f , g = − [ f ], Tjk# g for each

.  
f ∈ 𝒞η (∂Ω) ⊂ L 1 ∂Ω, 1+σ(x)
|x | n and g ∈ H (∂Ω, σ)
p (2.1.162)
  1 
with p ∈ n−1n , 1 and η := (n − 1) p − 1 ∈ (0, 1),

# is as in (2.1.42). In fact, each T mod

where the operator Tjk jk is also well defined,
linear, and bounded on plain BMO(∂Ω, σ) (i.e., without modding out constants).
Considered as such, said operator has VMO(∂Ω, σ) as an invariant subspace, and
both
mod
Tjk : VMO(∂Ω, σ) −→ VMO(∂Ω, σ), (2.1.163)
together with
mod
Tjk 
: VMO(∂Ω, 
σ) −→ VMO(∂Ω, σ),
mod mod
defined as Tjk [ f ] := Tjk f ] for each (2.1.164)
 
f ∈ VMO(∂Ω, σ) ⊂ L 1 ∂Ω, 1+σ(x)
|x | n ,

are well-defined, linear, and bounded operators.

Moreover, in the case when ∂Ω is bounded, the principal-value singular integral
operator Tjk from (1.2.2) has BMO(∂Ω, σ), 𝒞η (∂Ω) and 𝒞ηvan (∂Ω) with η ∈ (0, 1),
as well as VMO(∂Ω, σ) as invariant subspaces, and its restrictions

Tjk : BMO(∂Ω, σ) −→ BMO(∂Ω, σ), (2.1.165)

Tjk : 𝒞η (∂Ω) −→ 𝒞η (∂Ω) for each η ∈ (0, 1), (2.1.166)

Tjk : 𝒞η (∂Ω)van −→ 𝒞ηvan (∂Ω) for each η ∈ (0, 1), (2.1.167)

Tjk : VMO(∂Ω, σ) −→ VMO(∂Ω, σ), (2.1.168)

268 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

are well-defined, linear, and bounded operators which satisfy

   
Tjk f , g = − f , Tjk
#
g , ∀ f ∈ BMO(∂Ω, σ), ∀g ∈ H 1 (∂Ω, σ), (2.1.169)
   
Tjk f , g = − f , Tjk
#
g , ∀ f ∈ 𝒞η (∂Ω), ∀g ∈ H p (∂Ω, σ), (2.1.170)
 n−1   
if p ∈ n , 1 and η := (n − 1) p1 − 1 ∈ (0, 1).
The following result, which complements Corollary 2.1.13, pertains to the action
of the modified boundary-to-boundary double layer potential operator on the space
CMO.

Corollary 2.1.14 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain with an

unbounded boundary, and define σ := H n−1 ∂Ω. Also, let L be a homogeneous,
weakly elliptic, constant (complex) coefficient, second-order M × M system in Rn
(for some M ∈ N). In this context, consider the modified boundary-to-boundary
double layer potential operator Kmod as in (1.8.24)-(1.8.25), and the corresponding
transpose double layer potential operator K # associated with the system L and the
set Ω as in (1.3.72). Then the operators
M M
Kmod : CMO(∂Ω, σ) −→ CMO(∂Ω, σ) (2.1.171)

and
 M  M
Kmod : CMO(∂Ω, σ) ∼ −→ CMO(∂Ω, σ) ∼ defined as
(2.1.172)
M
Kmod [ f ] := Kmod f for each function f ∈ CMO(∂Ω, σ)

are well-defined, linear, and bounded. Moreover,

under the identification [69, (4.6.17)] the (real) transpose of the oper-
ator Kmod from (2.1.171) is the operator K # from (2.1.4) with p = 1,
while under the identification [69, (4.6.21)] the (real) transpose of the (2.1.173)
operator Kmod from (2.1.172) is the operator K # from (2.1.4) with
p = 1.
M . M
Proof Given any ϕ ∈ Lipc (∂Ω) we have ϕ ∈ 𝒞η (∂Ω) for each η ∈ (0, 1).
As such, Theorem 2.1.10 ensures that
.
Kmod ϕ ∈ 𝒞η (∂Ω)
M
for each η ∈ (0, 1). (2.1.174)

From (1.8.26), (1.3.68), and [70, Theorem 1.4.2] we also see that there exists a
constant Cϕ ∈ C M with the property that
 
Kmod ϕ (x) = Cϕ + O(|x| 1−n ) as ∂Ω  x → ∞. (2.1.175)

Together, (2.1.174) and (2.1.175) prove that

2.1 Double Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces 269
M
for each ϕ ∈ Lipc (∂Ω) there exists some Cϕ ∈ C M such that (2.1.176)
Kmod ϕ−Cϕ is a continuous functions on ∂Ω which vanish at infinity.

Based on this, (A.0.52), [69, (4.6.14)], and the fact that the operator Kmod is well
defined, linear, and bounded in the context of (2.1.92), we then conclude that Kmod
M
maps the space CMO(∂Ω, σ) linearly and boundedly into itself. This takes care
of (2.1.171), and the claims about (2.1.172) follow on account of the first line in [69,
(4.6.14)].
Finally, the claims in (2.1.173) are consequences of [69, (4.6.17), (4.6.21)] and
(2.1.97). 

Remark 2.1.15 The results in Theorem 2.1.10, Corollary 2.1.11, and Corollar-
ies 2.1.13-2.1.14 are applicable to all boundary-to-boundary double layer potential
operators K, K # from Examples 1.4.9-1.4.20 and their modified versions Kmod from
Examples 1.8.4-1.8.7.

Let us further elaborate on Remark 2.1.8 and Remark 2.1.15 by describing the
analogues of Theorem 2.1.7, Theorem 2.1.10, Corollary 2.1.11, and Corollary 2.1.14
for the Cauchy-Clifford singular integral operator C defined in (A.0.54). These
results further complement the picture emerging from [70, Proposition 2.5.29] and
Proposition 1.6.1 in relation to this operator. Specifically, having fixed a UR domain
Ω ⊆ Rn , abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. Assume first that ∂Ω is unbounded. For each function
 σ(x) 
f ∈ L 1 ∂Ω, ⊗ Cn (2.1.177)
1 + |x| n
we define the modified boundary-to-boundary Cauchy-Clifford integral
∫ 
1 x−y
Cmod f (x) := lim+ 1 n (y) (2.1.178)
ε→0 ωn−1 |x − y| n R \B(x,ε)
∂Ω 
y
+ n 1Rn \B(0,1) (y) ν(y) f (y) dσ(y)
|y|

at σ-a.e. x ∈ ∂Ω. Then Cmod maps constant (Cn -valued) functions on ∂Ω into
constant (Cn -valued) functions on ∂Ω and induces well-defined, linear, and bounded
mappings in the following contexts:
   
Cmod : BMO(∂Ω, σ) ⊗ Cn ∼ −→ BMO(∂Ω, σ) ⊗ Cn ∼
(2.1.179)
Cmod [ f ] := Cmod f ] for each f ∈ BMO(∂Ω, σ) ⊗ Cn,

   
Cmod : VMO(∂Ω, σ) ⊗ Cn ∼ −→ VMO(∂Ω, σ) ⊗ Cn ∼
(2.1.180)
Cmod [ f ] := Cmod f ] for each f ∈ VMO(∂Ω, σ) ⊗ Cn,
270 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
   
Cmod : CMO(∂Ω, σ) ⊗ Cn ∼ −→ CMO(∂Ω, σ) ⊗ Cn ∼
(2.1.181)
Cmod [ f ] := Cmod f ] for each f ∈ CMO(∂Ω, σ) ⊗ Cn,
and, for every η ∈ (0, 1),
.  . 
Cmod : 𝒞η (∂Ω) ⊗ Cn ∼ −→ 𝒞η (∂Ω) ⊗ Cn ∼
. (2.1.182)
Cmod [ f ] := Cmod f ] for each f ∈ 𝒞η (∂Ω) ⊗ Cn,

.  . 
Cmod : 𝒞ηvan (∂Ω) ⊗ Cn ∼ −→ 𝒞ηvan (∂Ω) ⊗ Cn ∼
. (2.1.183)
Cmod [ f ] := Cmod f ] for each f ∈ 𝒞ηvan (∂Ω) ⊗ Cn,
with brackets denoting equivalence classes modulo constants. Moreover, these op-
erators are compatible (in a suitable sense) with the action of the principal-value
Cauchy-Clifford singular integral operator C on Lebesgue spaces. In addition, with
the duality brackets as in [69, Theorem 4.6.1], from (2.1.97) we presently deduce
that    
Cmod f , g = [ f ], C # g for each
 
f ∈ BMO(∂Ω, σ) ⊗ Cn ⊂ L 1 ∂Ω, 1+σ(x) |x | n ⊗ Cn
(2.1.184)
and g ∈ H 1 (∂Ω, σ) ⊗ Cn,
while (2.1.127) currently implies
   
Cmod f , g = [ f ], C # g for each
.  
f ∈ 𝒞η (∂Ω) ⊗ Cn ⊂ L 1 ∂Ω, 1+σ(x)|x | n ⊗ Cn
(2.1.185)
and g ∈ H p (∂Ω, σ) ⊗ Cn with
 n−1   
p ∈ n , 1 and η := (n − 1) p1 − 1 ∈ (0, 1).

In concert with (2.1.65) and the duality result from [69, Theorem 4.6.1], these further
entail
2  
Cmod = 14 I on BMO(∂Ω, σ) ⊗ Cn ∼, (2.1.186)
2  
Cmod = 14 I on VMO(∂Ω, σ) ⊗ Cn ∼, (2.1.187)
2  
Cmod = 14 I on CMO(∂Ω, σ) ⊗ Cn ∼, (2.1.188)
. 
= 14 I on 𝒞η (∂Ω) ⊗ Cn ∼ with η ∈ (0, 1),
2
Cmod (2.1.189)
. 
= 14 I on 𝒞ηvan (∂Ω) ⊗ Cn ∼ with η ∈ (0, 1).
2
Cmod (2.1.190)

Similar results are also valid in the case when Ω has a bounded boundary. Specif-
ically, if ∂Ω is a bounded set, then BMO(∂Ω, σ) ⊗ Cn , VMO(∂Ω, σ) ⊗ Cn , as well
2.2 Single Layer Operators Acting from Hardy Spaces 271

as 𝒞η (∂Ω) ⊗ Cn and 𝒞ηvan (∂Ω) ⊗ Cn with η ∈ (0, 1) are all invariant subspaces of
the standard Cauchy-Clifford singular integral operator C acting on Lebesgue spaces
(as in item (ii) of Proposition 1.6.1), the induced mappings

C : BMO(∂Ω, σ) ⊗ Cn −→ BMO(∂Ω, σ) ⊗ Cn, (2.1.191)

C : VMO(∂Ω, σ) ⊗ Cn −→ VMO(∂Ω, σ) ⊗ Cn, (2.1.192)

C : 𝒞η (∂Ω) ⊗ Cn −→ 𝒞η (∂Ω) ⊗ Cn with η ∈ (0, 1), (2.1.193)

C : 𝒞η (∂Ω) ⊗ Cn −→ 𝒞ηvan (∂Ω) ⊗ Cn with η ∈ (0, 1), (2.1.194)

are well defined, linear, bounded,

   
C f , g = f , C# g (2.1.195)

(with the duality brackets as in [69, Theorem 4.6.1]) when either

f ∈ BMO(∂Ω, σ) ⊗ Cn and g ∈ H 1 (∂Ω, σ) ⊗ Cn, (2.1.196)

or
f ∈ 𝒞η (∂Ω) ⊗ Cn and g ∈ H p (∂Ω, σ) ⊗ Cn
  1  (2.1.197)
with p ∈ n−1
n , 1 and η := (n − 1) p − 1 ∈ (0, 1),

and, finally,

C 2 = 14 I on BMO(∂Ω, σ) ⊗ Cn, (2.1.198)

C 2 = 14 I on VMO(∂Ω, σ) ⊗ Cn, (2.1.199)

C 2 = 14 I on 𝒞η (∂Ω) ⊗ Cn with η ∈ (0, 1), (2.1.200)

C 2 = 14 I on 𝒞ηvan (∂Ω) ⊗ Cn with η ∈ (0, 1). (2.1.201)

2.2 Single Layer Operators Acting from Hardy Spaces

As a prelude to the treatment of certain classes of integral operators (of single layer
type) on boundary Hardy spaces later on in this section, in the next lemma we look
at the issue of “differentiation under the duality pairing” which ultimately decides
the regularity of the integral operators in question.

Lemma 2.2.1 Consider a closed Ahlfors

 regular set Σ ⊆ Rn and let σ := H n−1 Σ.
Fix a function b ∈ 𝒞 N +1 R \ {0} for some N ∈ N0 , with the property that for each
n

multi-index α ∈ N0n of length |α| ≤ N and each ε > 0 one has

sup |(∂ α b)(x)| < +∞. (2.2.1)

x ∈R n \B(0,ε)
272 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Then for each α ∈ N0n with |α| ≤ N and each x ∈ Rn \ Σ one has
 
(∂ α b)(x − ·)Σ ∈ 𝒞η (Σ) ⊂ L ∞ (Σ, σ) ⊂ BMO(Σ, σ). (2.2.2)
0<η<1


Also, if for each α ∈ N0n with |α| ≤ N and each f ∈ H p (Σ, σ) with p ∈ n−1
n , 1 one
considers the function
 α 
(∂ b)(x − ·), f if Σ is bounded,
(Tα f )(x) :=   ∀x ∈ Rn \ Σ, (2.2.3)
α
[(∂ b)(x − ·)], f if Σ is unbounded,

where ·, · stands for the duality bracket on Σ described in [69, Theorem 4.6.1],
then Tα f is well defined, and satisfies

Tα f ∈ 𝒞 N (Rn \ Σ) and ∂ β (Tα f ) = Tα+β f

(2.2.4)
for each β ∈ N0n with | β| ≤ N − |α|.

Proof Pick α ∈ N0 with

n
 |α| ≤ N and an arbitrary pointαx ∈ R \ Σ. Using (2.2.1)
n

with ε ∈ 0, dist(x, Σ) then proves that the function (∂ b)(x − ·) is both bounded
and Lipschitz in Rn \ B(0, ε). Since our choice of ε ensures that Σ is contained in this
set, and since both boundedness
 and Lipschitzianity are hereditary properties, we
conclude that (∂ α b)(x − ·)Σ is both bounded and Lipschitz on Σ. With this in hand,
the membership in (2.2.2) follows with the help of [68, (7.4.105)] and [68, (7.3.25)].
Having proved (2.2.2), from the duality result recorded in [69, Theorem 4.6.1] we
may now conclude that Tα f is well defined in Rn \ Σ for each given f ∈ H p (Σ, σ)
and α ∈ N0n with |α| ≤ N. Next, fix an arbitrary f ∈ H p (Σ, σ) and suppose until
mentioned otherwise that Σ is unbounded. According to [69, Theorem 4.4.1], there
exist a numerical sequence {λ j } j ∈N ∈  p (N) along with a sequence {a j } j ∈N of
(p, 2)-atoms on Σ such that

f = lim fm in H p (Σ, σ) where, for each m ∈ N,

m→∞
m
(2.2.5)
fm := λ j a j ∈ Lcomp
2
(Σ, σ).
j=1

If for each m ∈ N we now set


um (x) := [b(x − ·)], fm , ∀x ∈ Rn \ Σ, (2.2.6)

then the duality result from [69, Theorem 4.6.1, (4.6.8)] permits us to express
2.2 Single Layer Operators Acting from Hardy Spaces 273
m ∫
um (x) = λj b(x − y)a j (y) dσ(y)
j=1 Σ

∫
= b(x − y) fm (y) dσ(y), ∀x ∈ Rn \ Σ. (2.2.7)
Σ

It is then clear from this representation that for each m ∈ N we have

um ∈ 𝒞 N (Rn \ Σ) and, for each α ∈ N0n with |α| ≤ N and x ∈ Rn \ Σ,

∫
  (2.2.8)
(∂ α um )(x) = (∂ α b)(x − y) fm (y) dσ(y) = [(∂ α b)(x − ·)], fm ,
Σ

where the last equality relies on (2.2.2) and the duality result from [69, Theorem 4.6.1,
(4.6.8)]. If for each given α ∈ N0n with |α| ≤ N and each given x ∈ Rn \Σ we introduce
  n−1 
(∂ α b)(x − ·) 𝒞. (n−1)(1/p−1) (Σ) if p ∈ ,1 ,
Cbα (x)
n
:= (2.2.9)
(∂ α b)(x − ·) BMO(Σ,σ) if p = 1,

then based on (2.2.3), (2.2.8), [69, (4.6.9)], and (2.2.9) we may estimate
     
(Tα f )(x) − (∂ α um )(x) =  [(∂ α b)(x − ·)], f − [(∂ α b)(x − ·)], fm 
  

=  [(∂ α b)(x − ·)], f − fm 

≤ Cbα (x) f − fm H p (Σ,σ), (2.2.10)

for each α ∈ N0n with |α| ≤ N, m ∈ N, and x ∈ Rn \ Σ. An inspection of the argument

used to justify (2.2.2) reveals that, for each xo ∈ Rn \ Σ and α ∈ N0n with |α| ≤ N
fixed, the quantity (2.2.9) satisfies

sup Cbα (x) < +∞ if 0 < r < dist(xo, Σ). (2.2.11)

x ∈B(x o ,r)

In concert with (2.2.5) and (2.2.10), this proves that

 
for each α ∈ N0n with |α| ≤ N, the sequence ∂ α um m converges
(2.2.12)
to the function Tα f uniformly on compact subsets of Rn \ Σ.
From this and the differentiation theorem for sequences of functions we then readily
conclude that (2.2.4) holds in the case when Σ is unbounded. Finally, the case when
Σ is bounded is treated, mutatis mutandis, identically. 

We continue by considering integral operators of single layer type and studying

the manner in which their weak tangential derivatives act on inhomogeneous Hardy-
based Sobolev spaces.
274 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Proposition 2.2.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. For some sufficiently large N = N(n) ∈ N, let b ∈ 𝒞 N (Rn \ {0})
be a complex-valued function which is even and positive homogeneous of degree
 
2 − n. Consider the integral operator acting on each f ∈ L 1 ∂Ω, 1+σ(x) |x | n−2
(hence,
in particular, for each f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1)) according to
∫
ℬ f (x) := b(x − y) f (y) dσ(y) for each x ∈ Ω. (2.2.13)
∂Ω
 n−1 
Then, for each p ∈ n , n − 1 and q ∈ (0, ∞], the operator (2.2.13) extends to a
continuous linear mapping

ℬ : H p,q (∂Ω, σ) −→ 𝒞 N (Ω) (2.2.14)

which, for each j, k ∈ {1, . . . , n} and each f ∈ H p,q (∂Ω, σ), satisfies
.  
∂τ j k ℬ f = Tjk #
f in H p,q (∂Ω, σ), (2.2.15)

where the operator Tjk# is as in (2.1.44).

Let us also consider ℬmod , the modified version of (2.2.13) acting on each function
 
f belonging to the larger space L 1 ∂Ω, 1+σ(x)|x | n−1
(hence, in particular, for each
f ∈ L p (∂Ω, σ) with p ∈ (1, ∞)) according to
∫ % &
ℬmod f (x) := b(x − y) − b(−y)1Rn \B(0,1) (y) f (y) dσ(y) (2.2.16)
∂Ω

at every x ∈ Ω. Then for each function f ∈ L p,q (∂Ω, σ), with exponents p ∈ (1, ∞),
q ∈ (0, ∞], and each j, k ∈ {1, . . . , n} one has
.  
∂τ j k ℬmod f = Tjk #
f at σ-a.e. point on ∂Ω, (2.2.17)

# is now interpreted as in item (ii) of Proposition 1.2.1.

where the operator Tjk

Before presenting the proof of Proposition 2.2.2 we wish to comment on some

of its immediate ramifications. To set the stage, let Ω ⊆ Rn be a UR domain and set
σ := H n−1 ∂Ω. Suppose b ∈ 𝒞 N (Rn \ {0}) (for some sufficiently large N ∈ N) is a
complex-valued function which is even and positive homogeneous of degree 2 − n.
Also, pick two arbitrary indices j, k ∈ {1, . . . , n} and recall the integral operator
Tjk associated with Ω and b as in (1.2.1). Finally, consider ℬ as in (2.2.14). As
consequence of [69, Lemma 11.10.3] and [69, Corollary 11.1.8], we then see that
for each distribution f belonging to the Hardy-based inhomogeneous  Sobolev space
q, p
H1 (∂Ω, σ) (cf. [69, Definition 11.10.6]) with p ∈ n−1 n , n − 1 and q ∈ (1, ∞) we
have
 
Tjk f = ℬ ∂τ j k f in Ω. (2.2.18)
2.2 Single Layer Operators Acting from Hardy Spaces 275

In concert, (2.2.18) and (2.2.15) further imply that for every j, k, r, s ∈ {1, . . . , n}
we have (compare with (1.2.13))
.    
∂τr s Tjk f = Tr#s ∂τ j k f in H p (∂Ω, σ), for each
q, p   (2.2.19)
f ∈ H1 (∂Ω, σ) with p ∈ n−1 n , n − 1 and q ∈ (1, ∞).

Here is the proof of Proposition 2.2.2.

Proof of Proposition 2.2.2 The extension of the operator ℬ, originally considered
as in (2.2.13), to a continuous linear mapping in the context of (2.2.14) may be
justified based on Lemma 2.2.1 and real-interpolation (cf. [69, Theorem 4.3.1]); see
also [70, (2.4.24)]
  regard. As such, given any aperture parameter κ > 0 along
in this
with p ∈ n−1 n , n − 1 and q ∈ (0, ∞], we conclude from [70, (2.4.27)] that there
exists a constant C ∈ (0, ∞) with the property that
 
Nκ (∇ℬ f ) p, q ≤ C f H p, q (∂Ω,σ) for each f ∈ H p,q (∂Ω, σ). (2.2.20)
L (∂Ω,σ)

To proceed, fix two indices j, k ∈ {1, . . . , n}. Granted (2.2.20), the result described
in [69, Example 10.2.2] applies and [69, (10.2.13)] gives that
 
for each f ∈ H p,q (∂Ω, σ) with p ∈ n−1 , n − 1 and q ∈ (0, ∞],
. n
the distribution ∂τ j k (ℬ f ) belongs to the Lorentz-based Hardy space
(2.2.21)
H p,q (∂Ω, σ) and there exists
 . a finite constant C > 0, depending only
on Ω, n, b, p, q, such that ∂τ j k (ℬ f ) H p, q (∂Ω,σ) ≤ C f H p, q (∂Ω,σ) .

# is also well-defined, linear, and bounded on

In view of the fact that the operator Tjk
H (∂Ω, σ), in order to conclude (2.2.15) it suffices to show that
p,q

.  
∂τ j k ℬ f = Tjk
#
f for each f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1). (2.2.22)

Indeed,
 n−1via density
 this identity continues to hold for any f ∈ H p (∂Ω, σ) with
p ∈ n , n − 1 from which (2.2.15) is then deduced using real-interpolation (cf.
[69, Theorem 4.3.1]).
To justify (2.2.22), fix an arbitrary function f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1).
Based on [69, Example 10.2.10] (see the equality in [69, (10.2.92)], in particular)
.  
and [70, (2.4.9)] we conclude that ∂τ j k ℬ f belongs to Lloc
1 (∂Ω, σ) and, for some

background aperture parameter κ > 0,

.   κ−n.t. κ−n.t.
∂τ j k ℬ f = ν j (∂k ℬ f )∂Ω − νk (∂j ℬ f )∂Ω at σ-a.e. point on ∂Ω, (2.2.23)

where ν = (ν1, . . . , νn ) stands for the geometric measure theoretic outward unit
normal to Ω. In concert with the jump-formula described in (2.1.49)-(2.1.51) and
the identity proved in (1.2.18), this permits us to compute
276 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
.  
∂τ j k ℬ f (x)
∫
1  
= ν j (x)
∂k b ν(x) f (x) + lim+ ν j (x)(∂k b)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε
∫
1  
− νk (x)∂
j b ν(x) f (x) − lim+ νk (x)(∂j b)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε
∫
 
= lim+ ν j (x)(∂k b)(x − y) − νk (x)(∂j b)(x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

= Tjk
#
f (x) at σ-a.e. point x ∈ ∂Ω, (2.2.24)

with the last equality coming from (1.2.3). Hence, (2.2.22) is established and, with
it, the proof of (2.2.15) is complete.
 
Turning our attention to (2.2.17), first observe that for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
and each index  ∈ {1, . . . , n} we have
∫
 
∂ ℬmod f (x) = (∂ b)(x − y) f (y) dσ(y) for all x ∈ Ω. (2.2.25)
∂Ω

In concert with [70, Theorem 2.4.1], this proves that for each p ∈ (1, ∞) and
q ∈ (0, ∞] there exists a constant C ∈ (0, ∞) with the property that
  
Nκ ∇(ℬ f )  p, q ≤ C f L p, q (∂Ω,σ) (2.2.26)
mod L (∂Ω,σ)

for all f ∈ L p,q (∂Ω, σ). Going further, pick two indices j, k ∈ {1, . . . , n}. Having
shown (2.2.26), we may once again invoke [69, Example 10.2.2] to conclude that

for each function f ∈ L p,q (∂Ω, σ) with integrability exponents

.  
p ∈ (1, ∞) and q ∈ (0, ∞], the distribution ∂τ j k ℬmod f belongs
to the Lorentz space L p,q (∂Ω, σ) and there exists some finite con- (2.2.27)
 . C > 0, depending
stant  only on Ω, n, b, p, q, with the property that
∂τ ℬ f ) p, q ≤ C f L p, q (∂Ω,σ) .
jk mod L (∂Ω,σ)

Given that the operator Tjk # is also well-defined, linear, and bounded on L p,q (∂Ω, σ)

(cf. item (ii) in Proposition 1.2.1), the formula claimed in (2.2.17) follows via density
and embeddings (cf. [68, (6.2.51)-(6.2.52)]) as soon as we show that
.  
∂τ j k ℬmod f = Tjk
#
f for each f ∈ L p (∂Ω, σ) with p ∈ (1, n − 1). (2.2.28)

However, (2.2.28) is implied by the fact that, as seen from [69, Example 10.2.2], the
.
operator ∂τ j k annihilates constants and the observation that, as seen from definitions,
2.2 Single Layer Operators Acting from Hardy Spaces 277
 
for each function f belonging to the space L 1 ∂Ω, 1+σ(x)
|x | n−1
(in partic-
ular, for each function f ∈ L (∂Ω, σ) with exponent p ∈ [1, ∞)) the
p (2.2.29)
difference C f := ℬmod f − ℬ f is a constant in Ω.

This completes the proof of (2.2.17). 

One of the main points of the theorem below is that the action of the boundary-
to-domain version of the single layer, originally defined as in (1.3.6) on Lebesgue
space L p with p ∈ [1, n − 1), may be extended  in a natural fashion to the scale of
Hardy spaces H p for p in the range n−1n , n − 1 if n ≥ 3. Moreover, there is a natural
version of said single layer in the two-dimensional setting, acting on the scale of
Hardy spaces H p with p ∈ 12 , 1 .

Theorem 2.2.3 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with an Ahlfors

regular boundary ∂Ω. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn )
the geometric measure theoretic
 outward unit to Ω. For some M ∈ N, consider a
αβ
coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the property that
1≤α,β ≤M
the M × M second-order system (as always, the summation convention over repeated
indices is in effect)  
αβ
L A := ar s ∂r ∂s (2.2.30)
1≤α,β ≤M

is weakly elliptic. Also, denote by A the transpose of the coefficient tensor A (as
defined in (1.3.3)). Finally, let E = (Eαβ )1≤α,β ≤M be the fundamental solution
associated with the system L A as in [70, Theorem 1.4.2].
n , 1 define the boundary-to-domain single layer
In this setting, for each p ∈ n−1
M
potential operator 𝒮 acting on each f = ( fβ )1≤β ≤M ∈ H p (∂Ω, σ) according to
  
⎧
⎪ Eαβ (x − ·)∂Ω, fβ if ∂Ω is bounded,
⎪
⎨
⎪ 1≤α ≤M
𝒮 f (x) := 

 (2.2.31)
⎪
⎪ 
⎪ [Eαβ (x − ·)∂Ω ], fβ if ∂Ω is unbounded,
⎩ 1≤α ≤M

at every x ∈ Ω, where ·, · stands for the duality bracket on ∂Ω, described in [69,
Theorem 4.6.1] (with Σ := ∂Ω). Then the following properties hold.

(1) For each p ∈ n−1 n , 1 , the single layer induces a well-defined, linear operator
in the context
−→ 𝒞∞ (Ω)
M M
𝒮 : H p (∂Ω, σ) (2.2.32)
M
which is also continuous when 𝒞∞ (Ω) is equipped with the Frechét topology
of uniform convergence of partial derivatives on compact sets. In additions,
the operators in (2.2.32) corresponding to various values of p ∈ n−1n , 1 are
compatible with one another.
 n−1 for each f = ( fβ )1≤β ≤M ∈ [H (∂Ω, σ)] with integrability exponent
Moreover, p M

p ∈ n , 1 one has
278 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

 
L A 𝒮 f = 0 in Ω, (2.2.33)
and, for each multi-index γ ∈ N0n and each point x ∈ Ω,
  
⎧
⎪ (∂γ Eαβ )(x − ·)∂Ω, fβ if ∂Ω is bounded,
  ⎪
⎨
⎪ 1≤α ≤M
∂γ 𝒮 f (x) = 
⎪
⎪   
⎪ [(∂γ Eαβ )(x − ·)∂Ω ], fβ if ∂Ω is unbounded.
⎩ 1≤α ≤M
(2.2.34)

In addition, for each f ∈ [H p (∂Ω, σ)] M with p ∈ n−1 n , 1 and each γ ∈ N0 one
n

has
 
∂γ 𝒮 f (x) = O(|x| 2−n− |γ | ) as |x| → ∞,
(2.2.35)
if Ω is an exterior domain, and either n ≥ 3 or |γ| > 0.

Finally,
M M
if n = 2 then for each f ∈ H 1 (∂Ω, σ) the function 𝒮 f ∈ 𝒞∞ (Ω) ,
originally defined for each x ∈ Ω as in (2.2.31), extends (via the same
formula) to a continuous function in R2 , and if Ω is not an exterior domain
it is bounded by C f [H 1 (∂Ω,σ)] M , for some C = C(Ω, A) ∈ (0, ∞).
(2.2.36)
 n−1
(2) For each p ∈ n , 1 , the operator 𝒮 from (2.2.31)-(2.2.32) satisfies
∫
𝒮 f (x) = E(x − y) f (y) dσ(y) for each x ∈ Ω, whenever
∂Ω (2.2.37)
M M
f ∈ H p (∂Ω, σ) ∩ L q (∂Ω, σ) with q ∈ (1, n − 1).

Hence, the operator 𝒮 from (2.2.31)-(2.2.32) is compatible with the single

layer operator introduced earlier in (1.3.6). In particular, the latter operator
may be regarded as a natural extension of the former. Together, they give rise to
a well-defined, linear and bounded mapping
  n−1 
M ∞ M n , n − 1 if n ≥ 3,
𝒮 : H (∂Ω, σ)
p
−→ 𝒞 (Ω) , p ∈  (2.2.38)
2, 1 if n = 2,
1


defined as in (2.2.31)-(2.2.32)
 whenever p ∈ n−1n , 1 , and defined as in (1.3.6)
whenever n ≥ 3 and p ∈ 1, n − 1). Theoperator 𝒮 in (2.2.38) has the property
n , n−1 and each aperture parameter
that for each integrability exponent p ∈ n−1
κ ∈ (0, ∞) there exists some constant C = C(Ω, A, κ, p) ∈ (0, ∞) such that for
M
each f ∈ H p (∂Ω, σ) one has
2.2 Single Layer Operators Acting from Hardy Spaces 279

if either n ≥ 3, or n = 2 and Ω is not an exterior domain, then

    −1 (2.2.39)
Nκ (𝒮 f ) p ∗ ≤ C f [H p (∂Ω,σ)] M with p∗ := 1 − 1 ,
L (∂Ω,σ) p n−1

and
if n = 2 and Ω is an exterior domain then for each R > 0
 Ω∩B(0,R)    −1
Nκ (𝒮 f ) L p ∗ (∂Ω,σ) ≤ CR f [H p (∂Ω,σ)] M with p∗ := p1 − 1 .
(2.2.40)

Moreover, in the two-dimensional case one may allow p = 1 in (2.2.39)-(2.2.40)

(in which scenario one has p∗ = ∞), that is,

if n = 2 and Ω is not an exterior domain, then

  (2.2.41)
Nκ (𝒮 f ) ≤C f [H 1 (∂Ω,σ)] M for all f ∈ H 1 (∂Ω, σ)
M
,
L ∞ (∂Ω,σ)

and
if n = 2 and Ω is an exterior domain, then for each R > 0
 Ω∩B(0,R) 
Nκ (𝒮 f ) ∞ ≤ CR f [H 1 (∂Ω,σ)] M for all f ∈ H 1 (∂Ω, σ)
M
.
L (∂Ω,σ)
(2.2.42)

Finally, when either n ≥ 3 or ∂Ω is bounded, the single layer operator from

(1.3.6) corresponding to p = 1, i.e.,
M M M
𝒮 : L 1 (∂Ω, σ) → 𝒞∞ (Ω) mapping each f ∈ L 1 (∂Ω, σ)
∫
into 𝒮 f (x) := E(x − y) f (y) dσ(y) for every x ∈ Ω,
∂Ω
(2.2.43)

extends the operator 𝒮 from (2.2.38) corresponding to p = 1. Moreover, for each

given p ∈ n−1 n , 1 , the operator 𝒮 in (2.2.43) agrees with the operator 𝒮 from
M M M
(2.2.38) on the subspace H p (∂Ω, σ) ∩ H 1 (∂Ω, σ) of L 1 (∂Ω, σ) .
(3) Assume ∂Ω is a UR set, and pick an aperture
 parameter κ > 0. Then for each
integrability exponent p ∈ n−1
n , n − 1 if n ≥ 3, or p ∈ 12 , 1 if n = 2, there
exists a constant C = C(Ω, A, κ, p) ∈ (0, ∞) with the property that for every
M
f ∈ H p (∂Ω, σ) one has
n
 
Nκ (∂j 𝒮 f ) ≤C f [H p (∂Ω,σ)] M . (2.2.44)
L p (∂Ω,σ)
j=1
280 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

(4) Suppose ∂Ω is a UR set. Then, if f ∈ [H p (∂Ω, σ)] M with p ∈ n−1 n , 1 , the
.A
weak conormal derivative ∂ν 𝒮 f , taken in the sense of [69, Definition 10.2.18],
is meaningfully defined in the Hardy space [H p (∂Ω, σ)] M and satisfies, for a
constant C = C(Ω, A, p) ∈ (0, ∞) independent of f ,
 .A 
∂ 𝒮 f  p ≤ C f [H p (∂Ω,σ)] M . (2.2.45)
ν [H (∂Ω,σ)] M
.
In addition, the weak conormal derivative ∂νA𝒮 f , considered in the sense of [69,
Definition 10.2.18], is also meaningfully defined for every f ∈ [L 1 (∂Ω, σ)] M
and satisfies, for some C = C(Ω, A) ∈ (0, ∞) independent of f ,
 .A 
∂ 𝒮 f  1, ∞ ≤ C f [L 1 (∂Ω,σ)] M . (2.2.46)
ν [H (∂Ω,σ)] M
 n−1
If in fact Ω is a UR domain, then for every f ∈ [H p (∂Ω, σ)] M with p ∈ n ,1
one actually has the jump-formula
.   M
∂νA𝒮 f = − 12 I + K A#  f in H p (∂Ω, σ) , (2.2.47)

where I is the identity operator, and K A#  is the operator associated as in (2.1.4)

with the transpose coefficient tensor A (in place of A).
 
(5) Fix κ > 0 along with q ∈ (0, ∞] and p ∈ n−1 n , n − 1 . Then the single layer
operator (2.2.38) further extends to a linear and continuous mapping

−→ 𝒞∞ (Ω)
M M
𝒮 : H p,q (∂Ω, σ) (2.2.48)

with the property that there exists some constant C = C(Ω, A, κ, p, q) ∈ (0, ∞)
M
such that for each f ∈ H p,q (∂Ω, σ) one has
n
 
Nκ (∂j 𝒮 f ) ≤C f [H p, q (∂Ω,σ)] M , (2.2.49)
L p, q (∂Ω,σ)
j=1

as well as
if n = 2 and Ω is not an exterior domain, then
   
1 −1
Nκ (𝒮 f ) ∗ ≤ C f [H p, q (∂Ω,σ)] M with p∗ := p1 − ,
L p , q (∂Ω,σ) n−1
(2.2.50)

and
if n = 2 and Ω is an exterior domain, then for each R > 0
 Ω∩B(0,R) 
Nκ (𝒮 f ) L p ∗ , q (∂Ω,σ) ≤ CR f [H p, q (∂Ω,σ)] M (2.2.51)
 
1 −1
with p∗ := p1 − n−1 .
2.2 Single Layer Operators Acting from Hardy Spaces 281

In particular, (2.2.50)-(2.2.51) with q := p give the following sharper versions

of (2.2.39)-(2.2.40):

if n = 2 and Ω is not an exterior domain, then

   
1 −1
Nκ (𝒮 f ) p ∗ , p ≤ C f [H p (∂Ω,σ)] M with p∗ := p1 − ,
L (∂Ω,σ) n−1
(2.2.52)

and, respectively,

if n = 2 and Ω is an exterior domain, then for each R > 0

 Ω∩B(0,R)   
1 −1
Nκ (𝒮 f ) p ∗ , p ≤ CR f [H p (∂Ω,σ)] M with p∗ := 1 − ,
L (∂Ω,σ) p n−1
(2.2.53)
M
Also, whenever f = ( fβ )1≤β ≤M ∈ H p,q (∂Ω, σ) has compact support (as a
distribution, which is automatically the case if ∂Ω is compact) and the function
ψ ∈ Lipc (∂Ω) is identically one near supp f one has
  #  $
𝒮 f α (x) = Lipc (∂Ω) ψEαβ (x − ·)∂Ω, fβ (Lipc (∂Ω))
(2.2.54)
for each x ∈ Ω and each α ∈ {1, . . . , M }.

Moreover, if in fact Ω is a UR domain, then for every f ∈ [H p,q (∂Ω, σ)] M one
has the jump-formulas
.   M
∂νA𝒮 f = − 12 I + K A#  f in H p,q (∂Ω, σ) if p ≤ 1,
  (2.2.55)
M
∂νA𝒮 f = − 12 I + K A#  f in L p,q (∂Ω, σ) if p > 1,

as well as
' κ−n.t. (
αβ 
νr ar s ∂s (𝒮 f )β  = − 12 H f + K A#  f at σ-a.e. point on ∂Ω,
∂Ω 1≤α ≤M
with H as in [69, Theorem 4.9.1, (4.9.5)] and K A#  now regarded
M M
as a mapping from H p,q (∂Ω, σ) into L p,q (∂Ω, σ) (cf. (2.1.3)).
(2.2.56)

(6) Assume Ω is a UR domain, and consider a factorization of the original system

of the form

L = DD (2.2.57)
where D is a homogeneous, constant (complex) coefficient, first-order M × N
system in Rn , and D is a homogeneous, constant (complex) coefficient, first-order
N × M system in Rn , say
282 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

 n   n 
= αγ γβ
D b j ∂j 1≤α ≤M and D = bk ∂k 1≤γ ≤ N . (2.2.58)
j=1 1≤γ ≤ N k=1 1≤β ≤M

Also, define the coefficient tensor (with the summation convention over repeated
indices in effect)
 αβ  αβ αγ γβ
AD,D
 := a jk 1≤ j,k ≤n where each a jk := 
b j bk . (2.2.59)
1≤α,β ≤M
 
Then for every κ > 0 and every f ∈ [H p,q (∂Ω, σ)] M with p ∈ n−1 n , ∞ and
q ∈ (0, ∞] one has the jump-formulas
   
(−i)Sym D;  ν • (D𝒮 f ) = − 1 I + K #  f if p ≤ 1, (2.2.60)
2 A )D
D,

  κ−n.t.  
 ν (D𝒮mod f )
(−i)Sym D; = − 12 I + K A#  f if p > 1, (2.2.61)
∂Ω )D
D,

where the distribution in the left-hand side of (2.2.60) is defined as in [69,

Proposition 10.2.11], and where K A#  is the transpose double layer associated
)D
D,
 
with the domain Ω and the transpose coefficient tensor AD,D  .
(7) Continue to assume that Ω is a UR domain,  and fix two arbitrary indices
j, k ∈ {1, . . . , n}. Then for each p ∈ n−1
n , ∞ and q ∈ (0, ∞] the operator Tjk
#

from (1.5.82) extends to a linear and bounded mapping

M M
#
Tjk : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) (2.2.62)

and, with this interpretation,

M  
for each f ∈ H p,q (∂Ω, σ) n , n − 1 and q ∈ (0, ∞]
with n−1
.   # f in H p,q (∂Ω, σ) M .
(2.2.63)
one has ∂τ j k 𝒮 f = Tjk

Also,
M
if n = 2 then for each f ∈ H 1 (∂Ω, σ) one has
.   # f in H 1 (∂Ω, σ) M .
(2.2.64)
∂τ j k 𝒮 f = Tjk

M
Finally, for each function f ∈ L p,q (∂Ω, σ) with p ∈ (1, ∞) and q ∈ (0, ∞]
one has
.  
∂τ j k 𝒮mod f = Tjk
#
f at σ-a.e. point on ∂Ω, (2.2.65)

# is now interpreted as a bounded operator on L p,q (∂Ω, σ) M

where Tjk .
2.2 Single Layer Operators Acting from Hardy Spaces 283

It is also worth noting that, as an inspection of the proof of Theorem 2.2.3 reveals,
all results in items (1)-(2), with the exception of (2.2.33) and (2.2.36), are of purely
real variable nature, as they utilize only generic size and regularity properties of the
fundamental solution E. As such, analogous results are valid for the more general
class of operators in which E is replaced by a kernel function b ∈ 𝒞∞ (Rn \ {0})
satisfying (2.2.1) for each multi-index α ∈ N0n , as well as

|(∂ α b)(x)| ≤ C|x| −(n−2+ |α |) for each x ∈ Rn \ {0}

(2.2.66)
and each multi-index α ∈ N0n of length |α| ≤ 1.

We now turn to the task of giving the proof of Theorem 2.2.3.

Proof of Theorem 2.2.3 That the operator in (2.2.32) is well defined, linear, con-
tinuous, and that (2.2.34) holds, follow from Lemma 2.2.1 and [70, Theorem 1.4.2]
when n ≥ 3, and from [70, (1.4.22)-(1.4.23)], [69, Lemma 4.8.2], and [70, (1.4.44)]
when n = 2. In turn, (2.2.34) readily implies (2.2.33) (cf. [70, (1.4.21)]).
 Next,
that the operators in (2.2.32) corresponding to various values of p in n−1 n , 1 are
compatible with one another is a consequence of (2.2.31) and [69, Theorem 4.4.3].
To justify the claim made in (2.2.35), work under the assumption that Ω is an
exterior domain and fix f = ( fβ )1≤β ≤M ∈ [H p (∂Ω, σ)] M where p ∈ n−1 n , 1 along
with γ ∈ N0n . Then from (2.2.34) we know that for each α ∈ {1, . . . , M } we have
    
∂γ 𝒮 f α (x) = (H p (∂Ω,σ))∗ (∂γ Eαβ )(x − ·)∂Ω, fβ H p (∂Ω,σ) (2.2.67)
 
at each point x ∈ Ω. Suppose first that p < 1 and set η := (n − 1) p1 − 1 ∈ (0, 1). In
concert with the duality result from [69, Theorem 4.6.1] this implies that, for each
point x ∈ Ω,
 γ    
∂ 𝒮 f (x) ≤ C (∂γ E)(x − ·) η f [H p (∂Ω,σ)] M . (2.2.68)
[𝒞 (∂Ω)] M ×M

To estimate the Hölder norm above, fix some R ∈ (0, ∞) large enough so that
∂Ω ⊆ B(0, R) and make use of [68, (7.3.25)] and [70, Theorem 1.4.2] to write
 γ 
(∂ E)(x − ·) η M ×M
[𝒞 (∂Ω)]
     
≤ C sup (∂γ E)(x − y) + (∂γ E)(x − ·)[Lip(∂Ω)] M ×M
y ∈∂Ω
    
≤ C sup (∂γ E)(x − y) + sup (∇∂γ E)(x − y)
|y | ≤R |y | ≤R

= O(|x| 2−n− |γ | ) as |x| → ∞, (2.2.69)

assuming that either n ≥ 3, or |γ| > 0. The case when p = 1 is dealt with similarly,
now estimating
284 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 γ    
∂ 𝒮 f (x) ≤ C (∂γ E)(x − ·) f [H 1 (∂Ω,σ)] M
[BMO(∂Ω,σ)] M ×M
 
≤ C sup (∂γ E)(x − y) f [H 1 (∂Ω,σ)] M
y ∈∂Ω

= O(|x| 2−n− |γ | ) as |x| → ∞, (2.2.70)

again, assuming that either n ≥ 3, or |γ| > 0. The proof of (2.2.35) is therefore
complete.
Finally, the claim made in (2.2.36) is seen from [70, (1.4.22)-(1.4.23)] and [69,
Lemma 4.8.2] (since a similar result as in [69, Lemma 4.8.2] is valid with the
logarithm replaced by the function Φ from [70, (1.4.23)]). This takes care of the
claims in item (1).
Turning attention to item (2). In concert with [69, Proposition 4.8.6] if p = 1, and

with [69, Proposition 4.8.7] (together with [69, Proposition 4.2.2]) if p ∈ n−1 n ,1 ,
the estimate in [70, (2.5.558)] shows that the operator 𝒮 from (2.2.31)-(2.2.32)
satisfies the compatibility condition in (2.2.37). In particular, this justifies retaining
the same notation for the operator in (2.2.38).
Consider next the task of proving the estimate claimed in (2.2.39), working first
under the assumption that n ≥ 3. In view of [68, (3.6.27)], the range p ∈ (1, n − 1) is
covered by (1.3.57). In fact, using notation introduced in (A.0.168) we may rephrase
(1.3.57) as the statement that
M q∗
𝒮 : L q (∂Ω, σ) −→ Nκ (Ω; σ) is a well-defined, bounded
  −1 (2.2.71)
operator if q ∈ (1, n − 1) and q∗ := 1/q − 1/(n − 1) .

To proceed, consider the case when p ∈ n−1 n , 1 . Fix some q ∈ (1, n−1) and consider
an arbitrary
 C M -valued (p, q)-atom a on ∂Ω. Hence, there exist x ∈ ∂Ω along with
 o
r ∈ 0, 2 diam(∂Ω) such that

supp a ⊆ B(xo, r) ∩ ∂Ω,

  1/q−1/p
a [L q (∂Ω,σ)] M ≤ σ B(xo, r) ∩ ∂Ω ,
(2.2.72)
∫
and a dσ = 0.
∂Ω

If ∂Ω is bounded then, by definition, constant functions of absolute value less than or

equal to σ(∂Ω)−1/p are also considered to be (p, q)-atoms. Then, based on Hölder’s
inequality, the Ahlfors regularity of ∂Ω, (2.2.39), and (2.2.72), we may estimate
2.2 Single Layer Operators Acting from Hardy Spaces 285
∫
  ∗
Nκ (𝒮a) p dσ
B(x o ,2r)∩∂Ω
∫   ∗  p∗ /q∗   1−p∗ /q∗
≤ Nκ (𝒮a)q dσ · σ B(xo, 2r) ∩ ∂Ω
B(x o ,2r)∩∂Ω
  p∗
· Nκ (𝒮a) L q ∗ (∂Ω,σ)
∗ /q ∗ )
≤ C · r (n−1)(1−p
∗ /q ∗ ) p∗
≤ C · r (n−1)(1−p · a [L q (∂Ω,σ)] M
∗ /q ∗ ) ∗ (n−1)(1/q−1/p)
≤ C · r (n−1)(1−p · rp = C, (2.2.73)

for some constant C ∈ (0, ∞) independent of the atom in question. Next, pick
an arbitrary point x ∈ ∂Ω \ B(xo, 2r), along with some arbitrary z ∈ Γκ (x) and
y ∈ ∂Ω ∩ B(xo, r). Then the Mean Value Theorem, the estimates for E available
from [70, Theorem 1.4.2], and [68, (8.1.8)] imply that

|E(z − y) − E(z − xo )| ≤ Cr |z − xo | −(n−1) ≤ Cr |x − xo | −(n−1) (2.2.74)

for some constant C ∈ (0, ∞) independent of x, xo, z, r. Since a belongs to

M
L q (∂Ω, σ) , has bounded support and integrates to zero, (2.2.37) permits us
to write
∫
 
𝒮a(z) = E(z − y) − E(z − xo ) a(y) dσ(y). (2.2.75)
∂Ω

Collectively, (2.2.75), (2.2.74), and (2.2.72) allow us to estimate (bearing in mind

that ∂Ω is Ahlfors regular)
∫
 
𝒮a(z) ≤ Cr |x − xo | −(n−1) |a(y)| dσ(y)
∂Ω∩B(x o ,r)
  1−1/q
≤ Cr |x − xo | −(n−1) · a [L q (∂Ω,σ)] M · σ B(xo, r) ∩ ∂Ω

r 1+(n−1)(1−1/p)
≤C . (2.2.76)
|x − xo | n−1

By taking the supremum of both sides of (2.2.76) in z ∈ Γκ (x) we therefore arrive at

  r 1+(n−1)(1−1/p)
Nκ (𝒮a) (x) ≤ C for all x ∈ ∂Ω \ B(xo, 2r). (2.2.77)
|x − xo | n−1
On account of (2.2.77) and [68, (7.2.5)] we then obtain (bearing in mind that
p∗ (n − 1) > n − 1)
286 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
∫ ∫ ∗ ∗
  ∗ r p +p (n−1)(1−1/p)
Nκ (𝒮a) p dσ ≤ C dσ
∂Ω\B(x o ,2r) ∂Ω\B(x o ,2r) |x − xo | (n−1)p∗
∗ +p ∗ (n−1)(1−1/p) ∗ (n−1)
≤ C · rp · r (n−1)−p

= C, (2.2.78)

for some constant C ∈ (0, ∞) independent of the atom a. At this stage, combining
(2.2.73) with (2.2.78) shows that for each (p, q)-atom a with zero integral on ∂Ω we
have
 
Nκ (𝒮a) p ∗ ≤C (2.2.79)
L (∂Ω,σ)

for some constant C ∈ (0, ∞) independent of a. Recall that when ∂Ω is bounded,

constant functions of absolute value ≤ σ(∂Ω)−1/p are also considered, by definition,
to be (p, q)-atoms. In such a scenario, (2.2.79) continues to hold, by virtue of (1.3.59),
since we are currently assuming n ≥ 3. Thus, (2.2.79) is true for all (p, q)-atoms a
on ∂Ω in such a case.
Granted (2.2.71) and (2.2.79), we may now invoke [69, Theorem 4.4.7] with (X, τ)
q∗ p∗
and (Y, · ) taken to be, respectively, Nκ (Ω; σ) and Nκ (Ω; σ) (both of which are
Banach spaces; cf. [68, Proposition 8.3.5]). In light of the remark made in [69,
(4.4.146)] (presently used with θ := 1 and s := 1) this guarantees that the operator
𝒮 from (2.2.71) extends uniquely, in the manner described in [69, (4.4.144)], to a
linear and bounded operator

 : H p (∂Ω, σ) M p∗
𝒮 −→ Nκ (Ω; σ). (2.2.80)

To complete the proof of (2.2.39) in the case when n ≥ 3 there remains to show that
 n−1
 f = 𝒮 f for every f ∈ H p (∂Ω, σ)
𝒮
M
with p ∈ ,1 . (2.2.81)
n

M
With this finality in mind, pick an arbitrary distribution f ∈ H p (∂Ω, σ) with

integrability exponent p ∈ n−1n , 1 and, for some fixed exponent q ∈ (1, n − 1),
consider a numerical sequence {λi }i ∈N ∈  p (N) along with a sequence {ai }i ∈N of
j
C M -valued (p, q)-atoms on ∂Ω with the property that f j := i=1 λi ai converges to
f in H p (∂Ω, σ)
M
as j → ∞. Then [69, (4.4.144)] implies that 𝒮  f = lim 𝒮 f j
j→∞
p∗
in Nκ (Ω; σ). From this and [68, (8.3.33)] we further conclude that 𝒮 f j converges
 f pointwise in Ω as j → ∞. As such, for each x ∈ Ω we may write, assuming
to 𝒮
that ∂Ω is unbounded,
 
 f (x) = lim 𝒮 f j (x) = lim [E(x − ·)], f j
𝒮
j→∞ j→∞
 
= [E(x − ·)], f = 𝒮 f (x), (2.2.82)
2.2 Single Layer Operators Acting from Hardy Spaces 287

where we have also used (2.2.31) in the second and fourth equalities. Hence, for
each x ∈ Ω we have 𝒮  f (x) = 𝒮 f (x), and the same conclusion holds when ∂Ω is
bounded (via a similar argument). The proof of (2.2.39) is therefore complete, in the
case when n ≥ 3.
To deal with the two-dimensional setting, first observe that (2.2.41)-(2.2.42)
  are
implied by (2.2.36) and [68, (8.2.28)]. Next, assume n = 2, fix p ∈ 12 , 1 , and set
  −1
p∗ := p1 − 1 . We largely reason as before. The main difference is that if a is as
in (2.2.72) (this time, with q ∈ [1, ∞] arbitrary), then in place of (2.2.73) we now
estimate (again, for some constant C ∈ (0, ∞) independent of the atom in question)
∫
  ∗   ∗  
Nκ (𝒮a) p dσ ≤ Nκ (𝒮a) p ∞ · σ B(xo, 2r) ∩ ∂Ω
L (∂Ω,σ)
B(x o ,2r)∩∂Ω

p∗
≤ C ·r · a [H 1 (∂Ω,σ)] M
    p∗ (1−1/p)
≤ C · r · σ B(xo, r) ∩ ∂Ω ≤ C, (2.2.83)

thanks to (2.2.41), the fact that [69, (4.4.6)] and [69, Theorem 4.4.1] imply
    1−1/p
a [H 1 (∂Ω,σ)] M ≤ C σ B(xo, r) ∩ ∂Ω , (2.2.84)

the Ahlfors regularity of ∂Ω, and the definition of p∗ . The argument in (2.2.74)-
(2.2.78) goes through and, as before, we conclude that (2.2.79) holds if either Ω is
bounded, or ∂Ω is unbounded. With this in hand, the same type of reasoning as in
the end-game of the case n ≥ 3 establishes (2.2.39) when n = 2 and either Ω is
bounded, or ∂Ω is unbounded.
Finally, when n = 2 and Ω is an exterior domain, (2.2.79) is true provided a
truncated nontangential maximal operator is used (see (1.3.60)). This establishes
(2.2.40).
Going further, the compatibility between (2.2.43) and 𝒮 from (2.2.38) with p = 1
n , 1 the
is a consequence of [69, Lemma 4.6.5]. Finally, the fact that for each p ∈ n−1
operator 𝒮 in (2.2.43) agrees with the operator 𝒮 from (2.2.38) when considered on
M
H p (∂Ω, σ) ∩ H 1 (∂Ω, σ) is a consequence of [69, Lemma 4.6.6]. This concludes
the treatment of item (2).
As regards item (3), if ∂Ω is a UR set then the nontangential estimate in (2.2.44)
is implied by (2.2.34), [70, (2.4.14)], [70, (2.4.9)] (keeping in mind [68, (3.6.27)]),
and [70, Theorem 1.4.2].
Next, let us deal with the claims in item (4). For now, assume that ∂Ω is a UR
M 
set. First, having fixed an arbitrary f ∈ H p (∂Ω, σ) with p ∈ n−1n , 1 along with
some background aperture parameter κ > 0, the proof so far guarantees that the
function u := 𝒮 f satisfies

u ∈ 𝒞∞ (Ω)
M
, L Au = 0 in Ω, and Nκ (∇u) ∈ L p (∂Ω, σ). (2.2.85)
288 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

In fact, there exists a constant C = C(Ω, A, κ, p) ∈ (0, ∞) independent of f such that

Nκ (∇u) L p (∂Ω,σ) ≤C f [H p (∂Ω,σ)] M . (2.2.86)

. M
Granted these, [69, Theorem 10.2.24] applies and gives ∂νA𝒮 f ∈ H p (∂Ω, σ)
and
 .A   
∂ 𝒮 f  p ≤ C Nκ (∇𝒮 f ) L p (∂Ω,σ) ≤ C f [H p (∂Ω,σ)] M , (2.2.87)
ν [H (∂Ω,σ)] M

proving (2.2.45). In particular, we have a well-defined, linear, and bounded assign-

ment .
M M
H p (∂Ω, σ)  f −→ ∂νA𝒮 f ∈ H p (∂Ω, σ) . (2.2.88)
Estimate in (2.2.46) is established in a similar manner, making use of [69, Theo-
rem 10.2.24] and the estimate in the last line in (1.5.48).
Going further, strengthen the hypotheses by assuming that Ω is a UR domain
in Rn . In this setting, fix p ∈ n−1 n , 1 along with some q ∈ (1, n − 1). Suppose
first that the function f is an arbitrary C M -valued (p, q)-atom on ∂Ω. Then f is a
M
multiple of a C M -valued (1, q)-atom, so f also belongs to the space H 1 (∂Ω, σ) .
If we define u := 𝒮 f then, in light of (2.2.86), the latter membership ensures that
Nκ (∇u) belongs to L 1 (∂Ω, σ). In addition, from (2.2.34) and [69, Proposition 4.8.7]
it follows that for each j ∈ {1, . . . , n} and each α ∈ {1, . . . , M } we have
∫
(∂j uα )(x) = (∂j Eαβ )(x − y) fβ (y) dσ(y), ∀x ∈ Ω. (2.2.89)
∂Ω

Then [70, Theorem 2.5.1] applies and gives that, for every j ∈ {1, . . . , n} and
α ∈ {1, . . . , M }, the nontangential trace
κ−n.t.
(∂j uα )∂Ω exists σ-a.e. on ∂Ω. (2.2.90)

At this stage, item (3) in [69, Theorem 10.2.24] applies and the pointwise formula
. M M
for the weak conormal derivative ∂νAu ∈ H 1 (∂Ω, σ) ⊂ L 1 (∂Ω, σ) given in
[69, (10.2.186)] then permits us to express
. 
αβ   κ−n.t. 
∂νAu = νr ar s ∂s (𝒮 f )β 
∂Ω 1≤α ≤M
 
= − 12 I + K A#  f at σ-a.e. point on ∂Ω, (2.2.91)

where the last equality is a consequence of (2.2.37), (1.5.59), and the fact that,
M
as a C M -valued (p, q)-atom, the function f belongs to L q (∂Ω, σ) (recall that
q ∈ (1, n − 1)). Let us record our progress. From (2.2.45) and (2.2.91) it follows that
the assignment
M . M
H p (∂Ω, σ)  f −→ ∂νA𝒮 f ∈ H p (∂Ω, σ) (2.2.92)
2.2 Single Layer Operators Acting from Hardy Spaces 289

is well defined, linear, continuous, and agrees with the operator − 12 I + K A#  when act-
ing on arbitrary C M -valued (p, q)-atoms on ∂Ω. Since from Theorem 2.1.1 we know
M
that − 12 I +K A#  is also a well-defined linear and bounded operator on H p (∂Ω, σ) ,
formula (2.2.47) now follows via a standard density argument (relying on [69,
(4.4.114)]).
Next, with the exception of (2.2.54) and (2.2.56), the claims in item (5) are
consequences of the properties established so far, along with [69, Theorem 4.3.1],
(1.5.59), and [69, Proposition 1.3.7] (with the manner in which 𝒮 acts in the context
of (2.2.48) determined by [69, (1.3.41), (4.3.3)], and (2.2.31)). To justify (2.2.54),
we reason much as in [70, (2.4.93), (2.4.94)]. Specifically, fix a compactly supported
M
distribution f = ( fβ )1≤β ≤M ∈ H p,q (∂Ω, σ) along with ψ ∈ Lipc (∂Ω) which is
 n−1 
identically one near supp f . Also, pick p0, p1 ∈ n , n − 1 such that p0 < p < p1
and decompose f as f (0) + f (1) where f (i) = ( fβ(i) )1≤β ≤M ∈ H pi (∂Ω, σ)
M
for
i ∈ {0, 1}. Next, choose a function θ ∈ 𝒞∞ c (R ) satisfying θ ≡ 1 near the origin in
n

Rn , then for each R > 0 set θ R (x) := θ(x/R) for every x ∈ Rn . Finally, fix a point
x ∈ Ω along with an index α ∈ {1, . . . , M }. We may then rely on [68, (7.3.17)] (with
α := 1) to conclude that
  .
lim θ R Eα .(x − ·)∂Ω = Eα .(x − ·)∂Ω in 𝒞η (∂Ω)
M
for each η ∈ (0, 1). (2.2.93)
R→∞

Together with (2.2.31) and [69, Lemma 4.6.4], this permits us to write
 (i)  #  $
𝒮 f α (x) = lim Lipc (∂Ω) θ R Eαβ (x − ·)∂Ω, fβ(i) (Lipc (∂Ω)) for i ∈ {0, 1}.
R→∞
(2.2.94)
Summing up the two formulas in (2.2.94) then yields
  #  $
𝒮 f α (x) = lim Lipc (∂Ω) θ R Eαβ (x − ·)∂Ω, fβ (Lipc (∂Ω))
R→∞
#  $
= lim Lip c (∂Ω) θ R ψEαβ (x − ·)∂Ω, fβ (Lipc (∂Ω))
R→∞
#  $
= Lipc (∂Ω) ψEαβ (x − ·)∂Ω, fβ (Lipc (∂Ω)) (2.2.95)

which proves (2.2.54). The jump-formula in (2.2.56) may be justified based on

[70, (2.5.65)], upon noting that using the notation introduced in (A.0.141) for each
α, γ ∈ {1, . . . , M } we may compute, based on [70, (1.4.30)] (used with γ := es and
ξ := ν(x)),
290 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
   
1 αβ *   1 αβ −1
νr (x)ar s (∂s Eβγ ) ν(x) = νr (x)ar s iνs (x) L(ν(x))
2i 2i βγ

1  
−1

= − L(ν(x)) L(ν(x))
2 αβ βγ

1
= − δγα for σ-a.e. x ∈ ∂∗ Ω. (2.2.96)
2
Going further, the claims in item (6) are consequences of the current item (5), [69,
Proposition 10.2.21], and (1.5.55). As regards item (7), the fact that Tjk # , original-

ly defined as in (1.5.82), extends to a linear and bounded mapping in the context

of (2.2.62) is a direct consequence of Theorem 2.1.4 (keeping in mind [70, Theo-
rem 1.4.2]; cf. (2.1.44)), whereas the formula claimed in (2.2.63) is seen directly
from Proposition 2.2.2 when n ≥ 3 (cf. (2.2.15)), and from a slight variant of it
when n = 2, in which (2.2.22) is now tested for f ∈ H 1 (∂Ω, σ) (as a byproduct,
the latter also justifies (2.2.64)). Finally, (2.2.65) is a direct consequence of the last
part in Proposition 2.2.2 (bearing in mind [70, Theorem 1.4.2]; cf. (2.2.17)) when
n ≥ 3, and from a slight variant of it when n = 2 (in which (2.2.28) is now tested for
f ∈ H 1 (∂Ω, σ)). This completes the proof of Theorem 2.2.3. 

We now present the version of [69, Corollary 10.2.28] regarding the (uniform)
membership to the weak Hardy space H 1,∞ of conormal and tangential derivatives of
the fundamental solution of a weakly elliptic homogeneous constant complex coef-
ficient elliptic second-order system when the singularity is located on the boundary
of the domain. Such a version, which is going to be useful in the proof of Proposi-
tion 2.2.5 stated a little later, involves the notion of principal-value distribution on a
given UR set (cf. [69, Proposition 11.9.1]).

Corollary 2.2.4 Suppose Ω ⊆ Rn , where n ∈ N satisfies n ≥ 2, is a UR domain.

Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
  measure
αβ
theoretic outward unit normal to Ω. Also, for some M ∈ N, let A = ar s 1≤r,s ≤n
1≤α,β ≤M
be a constant (complex) coefficient tensor with the property that the associated
αβ
homogeneous second-order M × M system L A := ar s ∂r ∂s 1≤α,β ≤M (with the
summation convention over repeated indices in effect) is weakly elliptic in Rn . Finally,
denote by E = (Eαβ )1≤α,β ≤M the matrix-valued fundamental solution associated
with L A as in [70, Theorem 1.4.2]. Then
  αγ  
the distribution P.V. νr ar s (∂s Eγβ )(xo − ·)∂Ω belongs to
1≤α,β ≤M
(2.2.97)
M×M
the weak Hardy space H 1,∞ (∂Ω, σ) for σ-a.e. point xo ∈ ∂Ω,

in a uniform fashion, i.e., there exists C ∈ (0, ∞) such that

2.2 Single Layer Operators Acting from Hardy Spaces 291

M   αγ   

P.V. νr ar s (∂s Eγβ )(xo − ·)∂Ω  ≤ C for σ-a.e. xo ∈ ∂Ω.
H 1, ∞ (∂Ω,σ)
α,β=1
(2.2.98)
Moreover, for any α, β ∈ {1, . . . , M } and j, k ∈ {1, . . . , n}, one has
     
P.V. ν j (∂k Eαβ )(xo − ·)∂Ω − P.V. νk (∂j Eαβ )(xo − ·)∂Ω ∈ H 1,∞ (∂Ω, σ)

for σ-a.e. point xo ∈ ∂Ω, in a uniform fashion.

(2.2.99)
In addition, for every index β ∈ {1, . . . , M } and σ-a.e. point xo ∈ ∂Ω one has
.   αγ  
∂νA E. β (· − xo ) = − 12 δxo eβ − P.V. νr ar s (∂s Eγβ )(xo − ·)∂Ω
1≤α ≤M
M
as distributions in H 1,∞ (∂Ω, σ) .
(2.2.100)

Finally, for each fixed β ∈ {1, . . . , M } the difference

. .
∂νA E. β (· − x0 ) − ∂νA E. β (· − x1 )
+ (2.2.101)
M
belongs to H p (∂Ω, σ) for σ-a.e. x0, x1 ∈ ∂Ω,
n <p<∞
n−1

and for each fixed α, β ∈ {1, . . . , M } the difference

     
P.V. ν j (∂k Eαβ )(x0 − ·)∂Ω − P.V. νk (∂j Eαβ )(x0 − ·)∂Ω
     
− P.V. ν j (∂k Eαβ )(x1 − ·)∂Ω + P.V. νk (∂j Eαβ )(x1 − ·)∂Ω

belongs to H p (∂Ω, σ) for σ-a.e. x0, x1 ∈ ∂Ω, (2.2.102)
n <p<∞
n−1

though the uniformity of the membership is now lost.

Proof All claims except (2.2.100) are seen from [69, (4.2.17)], [69, (4.2.34)], and
[69, Proposition 11.9.2] by specializing k as in [69, (10.2.241), (10.2.242)]. Finally,
(2.2.100) follows from (A.0.182) and [69, (11.9.30)] (used for F as in [69, (11.9.18)]
and with k as in [69, (10.2.241)]), bearing in mind [70, (1.4.25)] and the fact that
∇E is odd. 
We are now ready to discuss the action of single and double layer potential
operators in the context of Theorem 2.2.3 (cf. (2.2.54)) and Corollary 2.1.2 (cf.
(2.1.28)) on certain concrete distributions belonging to the scale of Lorentz-based
Hardy spaces.
Proposition 2.2.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain. Abbreviate
σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric measure theoret-
ic outward unit normal to Ω. Let L be a homogeneous, weakly elliptic, constant
292 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

(complex) coefficient, second-order M × M system in Rn (for some M ∈ N). De-

note by E = (Eαβ )1≤α,β ≤M the matrix-valued fundamental solution associated with
L as in [70,
 Theorem 1.4.2]. Finally, pick a constant (complex) coefficient tensor
αβ
A = ar s 1≤r,s ≤n with the property that L = L A. Then the following claims are
1≤α,β ≤M
true.
(1) If 𝒮 is the boundary-to-domain single layer potential operator (associated with
the system L and the set Ω) acting on the scale of Lorentz-based Hardy spaces
as in (2.2.48), then for every index β ∈ {1, . . . , M } and σ-a.e. point xo ∈ ∂Ω
one has
𝒮(δxo eβ ) = E. β (· − xo ) in Ω. (2.2.103)
(2) If K # is the boundary layer potential operator (associated with the coefficient
tensor A and the set Ω) acting on the scale of Lorentz-based Hardy spaces as in
Corollary 2.1.2, then for every index β ∈ {1, . . . , M } and σ-a.e. point xo ∈ ∂Ω
one has
  γα  
K # (δxo eβ ) = − P.V. νr asr (∂s Eβγ )(xo − ·)∂Ω (2.2.104)
1≤α ≤M
. 
= ∂νA (E L  ) . β (· − xo ) + 12 δxo eβ ∈ H 1,∞ (∂Ω, σ)
M
.

In particular, for every index β ∈ {1, . . . , M } and σ-a.e. point xo ∈ ∂Ω one has
  .
− 12 I + K A#  (δxo eβ ) = ∂νA E. β (· − xo ) in H 1,∞ (∂Ω, σ)
M
. (2.2.105)
M M
(3) If K # is now regarded as a mapping from H 1,∞ (∂Ω, σ) into L 1,∞ (∂Ω, σ)
(cf. (2.1.3)), then for every index β ∈ {1, . . . , M } and σ-a.e. point xo ∈ ∂Ω one
has
  
K # (δxo eβ ) = − νr asr (∂s Eβγ )(xo − ·)∂Ω
γα
at σ-a.e. point on ∂Ω.
1≤α ≤M
(2.2.106)

(4) If H is the filtering operator, considered as in [69, Theorem 4.9.1, (4.9.5)] with
p := 1 and q := ∞, then for every indices α, β ∈ {1, . . . , M } and σ-a.e. point
xo ∈ ∂Ω one has
  γα   
H P.V. νr asr (∂s Eβγ )(xo − ·)∂Ω = νr asr (∂s Eβγ )(xo − ·)∂Ω
γα
(2.2.107)
at σ-a.e. point on ∂Ω.

Proof Fix β ∈ {1, . . . , M }. From [69, Example 4.2.4] we know that for each point
M
xo ∈ ∂Ω the distribution δxo eβ belongs to H 1,∞ (∂Ω, σ) . Since δxo eβ has also
compact support, (2.2.54) implies (2.2.103). Next, the first jump-formula in (2.2.55)
presently gives
.   M
∂νA𝒮(δxo eβ ) = − 12 I + K A#  (δxo eβ ) in H 1,∞ (∂Ω, σ) . (2.2.108)
2.2 Single Layer Operators Acting from Hardy Spaces 293

On the one hand, from (2.2.100) and (2.2.103) we see that for σ-a.e. xo ∈ ∂Ω we
have
.   αγ  
∂νA𝒮(δxo eβ ) = − 12 δxo eβ − P.V. νr ar s (∂s Eγβ )(xo − ·)∂Ω
1≤α ≤M (2.2.109)
  M
as distributions in Lipc (∂Ω) .

Comparing now (2.2.108) with (2.2.109) yields the first equality in (2.2.104) after
slight adjustments in notation (taking into account [70, (1.7.2)] and the first formula
in [70, (1.4.32)]). The second equality in (2.2.104) is seen from (2.2.100) written
with A in place of A.
Going further, the claim made in item (3) is justified by writing the jump-formula
M
(2.2.56) with A in place of A and for the choice f := δxo eβ ∈ H 1,∞ (∂Ω, σ) (cf.
[69, (4.2.34)]), then invoking [69, (4.9.8)] and (2.2.54). Finally, the claim in item (4)
is seen by applying H to (2.2.104), then taking into account (2.1.29) and (2.2.106).

We shall temporarily digress for the purpose of further elaborating on the scope
of Proposition 2.2.5. To set stage, recall from Theorem 2.1.5 that the boundary-to-
boundary transpose Cauchy-Clifford operator C # is well defined, linear, and bounded
on the scale of Lorentz-based Hardy spaces considered on the boundary of a UR
domain Ω ⊆ Rn (cf. (2.1.66)). In this regard, it is remarkable that for σ-a.e. point
xo ∈ ∂Ω we have
  
C # δxo = P.V. ν Φ(xo − ·)∂Ω ∈ H 1,∞ (∂Ω, σ) ⊗ Cn (2.2.110)

where ν is the geometric measure theoretic outward unit normal to Ω, the measure
σ := H n−1 ∂Ω, and
1 x
Φ(x) := , ∀x ∈ Rn \ {0}. (2.2.111)
ωn−1 |x| n
For example, in the particular case when we take Ω := R+n , the upper half-space
in Rn , formula (2.2.110) implies (as may be seen after unraveling notation) that
x
R j δ0 = ωn−1
2
P.V. |x |jn , where R j is j-th Riesz transform in Rn−1 and δ0 denotes the
Dirac distribution with mass at the origin in Rn−1 . Hence, it is natural to think of
(2.2.110) as a generalization of this classical fact in Harmonic Analysis.  
Formula (2.2.110) is proved by computing the Clifford bullet product ν• R C δxo
in two ways. Specifically, first we employ (2.1.88) to conclude that for each xo ∈ ∂Ω
we have
   
ν • R C δxo = I − 2C # δxo in H 1,∞ (∂Ω, σ) ⊗ Cn . (2.2.112)

Second, starting from the observation that for each xo ∈ ∂Ω and each x ∈ Ω we have
 
R C δxo (x) = 2Φ(x − xo ), (2.2.113)
294 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

we may use the very definition of the Clifford bullet product given in (A.0.167)
to compute (along the lines of the argument which has eventually produced [69,
(11.9.30)])
  
ν • Φ(· − xo ) = 12 δxo − P.V. ν Φ(xo − ·)∂Ω (2.2.114)

as distributions on ∂Ω, for σ-a.e. point xo ∈ ∂Ω. Comparing (2.2.112) with (2.2.114)
(while keeping in mind (2.2.113)) yields (2.2.110). To close, we remark that if H is
the filtering operator considered as in [69, Theorem 4.9.1, (4.9.5)] (with p := 1 and
q := ∞) then for σ-a.e. point xo ∈ ∂Ω we have (compare with (2.2.107))
    
H P.V. ν Φ(xo − ·)∂Ω = ν Φ(xo − ·)∂Ω at σ-a.e. point on ∂Ω. (2.2.115)

Returning now to the task of studying the action of the single layer potential
operator on Hardy spaces, the philosophy emerging from our theorem below is that
the boundary-to-boundary version of the single layer operator acts naturally from the
n < p < ∞, if n ≥ 3. When
entire scale of boundary Hardy spaces, i.e., H p with n−1
n = 2, the boundary-to-boundary single layer acts naturally from Hardy spaces H p
with 12 < p < 1.

Theorem 2.2.6 Let Ω ⊆ Rn , where n ∈ N satisfies4 n ≥ 3, be an open set with the

property that ∂Ω is an Ahlfors regular set, and denote σ := H n−1 ∂Ω. For some
M ∈ N, consider a coefficient tensor A of type (n × n, M × M) as in (1.3.1) with the
property that the M × M second-order system L A defined as in (2.2.30) is weakly
elliptic and denote by E the fundamental solution associated with the system L A as
in [70, Theorem 1.4.2].
In this setting, recall the definition of the boundary-to-boundary single layer
operator S from (1.3.62). In relation to this operator, the following statements are
true.

(1) For each p ∈ n−1 n , 1 , the operator S from (1.3.63) further extends, in a unique
fashion, to a linear and bounded operator5
M ∗ M
S : H p (∂Ω, σ) −→ L p (∂Ω, σ)
  −1  (2.2.116)
where p∗ := p1 − n−1
1
n−2 .
∈ 1, n−1

Furthermore, various choices of p yield operators which are compatible with

one another.

4 The two-dimensional case is dealt with separately, in item (8)

5 When Ω ⊆ R n is an open set satisfying a two-sided local John condition and whose boundary is
a compact Ahlfors regular set, (2.2.116) may be seen from [69, Proposition 11.11.5] and item (ix)
in Theorem 1.5.1.
2.2 Single Layer Operators Acting from Hardy Spaces 295
 n−1
Finally, for each given p ∈ n , 1 , the operator S in (2.2.116) agrees with the
M M
operator S from (1.3.64) on the subspace H p (∂Ω, σ) ∩ H 1 (∂Ω, σ) of
M
L 1 (∂Ω, σ) .
(2) Suppose r ∈ (n − 1, ∞) and define η := 1 − (n − 1)/r ∈ (0, 1). Then the operator
S from (1.3.63) further extends, in a unique fashion, to a linear and bounded
mapping
⎧
⎪ M
M ⎨ S f ∈ 𝒞η (∂Ω)
⎪ if ∂Ω is bounded,
L (∂Ω, σ)
r
 f −→ .η
⎪
⎪ [S f ] ∈ 𝒞 (∂Ω)/∼ M
if ∂Ω is unbounded.
⎩
(2.2.117)
Moreover, if
 n−1   
p∈ n ,1 , η := (n − 1) p1 − 1 ∈ (0, 1),
  −1   (2.2.118)
p∗ := p1 − n−1
1
n−2 ,
∈ 1, n−1

and  
1 −1
(p∗ ) := 1 − p∗ ∈ (n − 1, ∞) (2.2.119)
denotes the Hölder conjugate exponent of p∗ ,
then for each distribution f in the
M ∗  M
space H (∂Ω, σ)
p and each function g ∈ L (p ) (∂Ω, σ) one has
∫   
[H p (∂Ω,σ)] M f , [SL  g] [𝒞. η (∂Ω)/∼] M if ∂Ω is unbounded,
S f , g dσ =  
[H p (∂Ω,σ)] M f , SL  g [𝒞η (∂Ω)] M if ∂Ω is bounded,
∂Ω
(2.2.120)
where SL  is the single layer potential operator associated with L  , the (real)
transpose of L, and the brackets in the right side indicate duality in the sense of
[69, Theorem 4.6.1].
(3) The operator S from (1.3.63) also extends, in a unique fashion, to a linear and
bounded mapping
⎧
⎪ M
M ⎨ S f ∈ VMO(∂Ω, σ)
⎪ if ∂Ω is bounded,
L n−1
(∂Ω, σ)  f →
⎪ 
⎪ [S f ] ∈ VMO(∂Ω, σ)
M
if ∂Ω is unbounded.
⎩
(2.2.121)

In particular,
⎧
⎪ M
M ⎨ S f ∈ BMO(∂Ω, σ)
⎪ if ∂Ω is bounded,
L n−1
(∂Ω, σ)  f →
⎪ 
⎪ [S f ] ∈ BMO(∂Ω, σ)
M
if ∂Ω is unbounded,
⎩
(2.2.122)
296 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

is also a well-defined, linear, and continuous operator. Furthermore, for each

M M
given pair of functions, f ∈ H 1 (∂Ω, σ) and g ∈ L n−1 (∂Ω, σ) , one has
∫
S f , g dσ (2.2.123)
∂Ω
  
[H 1 (∂Ω,σ)] M f , [SL  g] [BMO(∂Ω,σ)/∼] M if ∂Ω is unbounded,
=  
[H 1 (∂Ω,σ)] M f , SL  g [BMO(∂Ω,σ)] M if ∂Ω is bounded,

where SL  is the single layer potential operator associated with L  , the (real)
transpose of L, and the brackets in the right side indicate duality in the sense of
[69, Theorem 4.6.1].
(4) The operators from items (1)-(3) act in a coherent fashion with one another, as
well as with the operator S from (1.3.63). As such, they may be glued together
M
to create an operator acting from the global Hardy scale H p (∂Ω, σ) with
 n−1 
p ∈ n , ∞ as follows:
 n−1 
⎧ ∗
L p (∂Ω, σ)
M
if p ∈ ,n−1 , = −
⎪ n−1 ,
1 1 1
⎪
⎪ n p∗ p
⎪
⎪
⎪
⎪ VMO(∂Ω, σ)
M
if p = n − 1 and ∂Ω is bounded,
⎪
⎪
⎪
⎪
⎪
⎪  M
⎪
⎪ BMO(∂Ω, σ) if p = n − 1 and ∂Ω is unbounded,
⎪
⎨
⎪
M
S : H p (∂Ω, σ) → 𝒞η (∂Ω)
M
if p ∈ (n − 1, ∞), η = 1 − p ,
n−1
⎪
⎪
⎪
⎪
⎪
⎪ and ∂Ω is bounded,
⎪
⎪
⎪
⎪ .
⎪
⎪ 𝒞η (∂Ω)/∼
M
if p ∈ (n − 1, ∞), η = 1 − p ,
n−1
⎪
⎪
⎪
⎪
⎪ and ∂Ω is unbounded.
⎩
(2.2.124)
See also (2.3.12) and (4.3.43) in this regard.
(5) In addition to the boundary-to-boundary single layer S from (2.2.124), recall its
M
modified version Smod from (1.5.73). Then for each function f ∈ L p (∂Ω, σ)
with 1 < p < ∞ one has

Smod f + C f if p ∈ (1, n − 1) or ∂Ω is bounded,
Sf = (2.2.125)
Smod f if p ∈ [n − 1, ∞) and ∂Ω is unbounded,

for some constant C f ∈ C M which depends on f .

 
(6) Given an exponent p ∈ n−1 n , n − 1 , recall the boundary-to-domain single
layer operator 𝒮 from (2.2.38). Also, fix some arbitrary aperture parameter
M
κ ∈ (0, ∞). Then for each f ∈ H p (∂Ω, σ) it follows that
2.2 Single Layer Operators Acting from Hardy Spaces 297
κ−n.t.
the nontangential boundary limit (𝒮 f )∂Ω exists (in C M ) at σ-a.e.
point on Aκ (∂Ω) (hence also at σ-a.e. point on ∂∗ Ω; cf. [68, (8.8.45)]),
(2.2.126)

and, with S denoting the boundary-to-boundary single layer operator from

(2.2.124), one has
 κ−n.t. 
(𝒮 f )∂Ω (x) = (S f )(x) at σ-a.e. point x ∈ Aκ (∂Ω)
(2.2.127)
(hence also at σ-a.e. point x ∈ ∂∗ Ω; cf. [68, (8.8.45)]).
 
(7) For each p ∈ n−1n , n − 1 and q ∈ (0, ∞], the operator S from (2.2.124) induces
a well-defined, linear and bounded mapping
M ∗ M
S : H p,q (∂Ω, σ) −→ L p ,q (∂Ω, σ)
  −1 (2.2.128)
where p∗ := p1 − n−1
1
∈ (1, ∞).

In particular,
M ∗ M
S : H p (∂Ω, σ) −→ L p , p (∂Ω, σ) boundedly
    −1 (2.2.129)
if p ∈ n−1 , n − 1 and p∗ := 1 − 1
.
n p n−1

Moreover, having fixed an arbitrary aperture parameter κ > 0, for each

M  
f ∈ H p,q (∂Ω, σ) with p ∈ n−1 n , n − 1 and q ∈ (0, ∞] it follows that
κ−n.t.
the nontangential boundary limit (𝒮 f )∂Ω exists (in C M ) at σ-a.e. point on
Aκ (∂Ω), and, with S denoting the boundary-to-boundary single layer operator
from (2.2.128), the boundary trace formula (2.2.127) continues to hold.
(8) Assume n = 2. Consider the operator

−→ L ∞ (∂Ω, σ) ∩ 𝒞0 (∂Ω)
M M
S : H 1 (∂Ω, σ) (2.2.130)
M
acting on each f = ( fα )1≤α ≤M ∈ H 1 (∂Ω, σ) according to
  
⎧
⎪ Eαβ (x − ·)∂Ω, fβ if ∂Ω is bounded,
⎪
⎨
⎪ 1≤α ≤M
S f (x) := 

 (2.2.131)
⎪
⎪ 
⎪ [Eαβ (x − ·)∂Ω ], fβ if ∂Ω is unbounded,
⎩ 1≤α ≤M

at every point x ∈ ∂Ω (where ·, · stands for the duality bracket on ∂Ω, described
in [69, Theorem 4.6.1], used with Σ := ∂Ω). Then this is a well-defined, linear,
M
continuous mapping. Moreover, for each distribution f ∈ H 1 (∂Ω, σ) one

has (𝒮 f ) = S f , where 𝒮 f ∈ 𝒞0 (Ω)
M
∂Ω
is considered as in (2.2.36). In
298 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
M
particular, for each aperture parameter κ > 0 and each f ∈ H 1 (∂Ω, σ) one
has
κ−n.t.
(𝒮 f )∂Ω = S f at σ-a.e. point on Aκ (∂Ω)
(2.2.132)
(hence also at σ-a.e. point on ∂∗ Ω; cf. [68, (8.8.45)]).
 
Also, for each p ∈ 12 , 1 the operator (2.2.130)-(2.2.131) extends uniquely to a
linear and bounded mapping
∗ 1  −1
where p∗ :=
M M
S : H p (∂Ω, σ) −→ L p (∂Ω, σ) p −1 , (2.2.133)

M
which continues to satisfy (2.2.132) for each κ > 0 and each f ∈ H p (∂Ω, σ) .
In addition, via real interpolation (cf. [69, (4.3.3)] and [68, (6.2.48)]), from
(2.2.133) one further obtains that
M ∗ M
S : H p,q (∂Ω, σ) −→ L p ,q (∂Ω, σ) boundedly
1    −1 (2.2.134)
if p ∈ 2 , 1 , q ∈ (0, ∞], and p∗ := p1 − 1 ,

hence, in particular,
M ∗ M
S : H p (∂Ω, σ) −→ L p , p (∂Ω, σ) boundedly
1    −1 (2.2.135)
whenever p ∈ 2 , 1 and p∗ := p1 − 1 ,

M 1 
and (2.2.132) remains true for each f ∈ H p,q (∂Ω, σ) with p ∈ 2, 1 and
q ∈ (0, ∞].
Finally, if

the set ∂Ω is bounded,

1 
p ∈ 2 , 1 , η := p1 − 1 ∈ (0, 1), (2.2.136)
  −1   −1
p∗ := p1 − 1 ∈ (1, ∞), (p∗ ) := 1 − p1∗ ∈ (1, ∞)

then the operator S from (1.3.63) further extends, in a unique fashion, to a linear
and bounded mapping
∗ 
S : L (p ) (∂Ω, σ) −→ 𝒞η (∂Ω)
M M
(2.2.137)
M
and for each distribution f ∈ H p (∂Ω, σ) and each function g in the space
∗  M
L (p ) (∂Ω, σ) one has
∫
 
S f , g dσ = [H p (∂Ω,σ)] M f , SL  g [𝒞η (∂Ω)] M (2.2.138)
∂Ω
2.2 Single Layer Operators Acting from Hardy Spaces 299

where SL  is the single layer potential operator associated with L  , the (real)
transpose of L, and the brackets in the right side indicate duality in the sense of
[69, Theorem 4.6.1].

An inspection of the proof of Theorem 2.2.6 reveals that all results in items (1)-(7)
are of a purely real variable nature, making use of only generic size and regularity
properties of the fundamental solution E. Consequently, analogous results are valid
for more general classes of operators in which E is replaced by a kernel function
enjoying similar size and regularity properties.
We wish to remark that formula (2.2.127) is optimal, in the sense that Aκ (∂Ω)
is the largest subset of ∂Ω where it is meaningful to consider the nontangential
κ−n.t.
trace 𝒮 f ∂Ω . Let us also note that, if E = (Eαβ )1≤α,β ≤M denotes the matrix-
valued fundamental solution associated with L as in [70, Theorem 1.4.2], then since
M
δxo eβ ∈ H 1,∞ (∂Ω, σ) for each β ∈ {1, . . . , M } (cf. [69, (4.2.34)]) the last claim
in item (7) and (2.2.103) imply that for each β ∈ {1, . . . , M } and each xo ∈ ∂Ω we
have
   
S δxo eβ (x) = Eαβ (x − xo ) 1≤α ≤M for σ-a.e. x ∈ ∂nta Ω. (2.2.139)

We now turn to the proof of Theorem 2.2.6.


Proof of Theorem 2.2.6 To prove the claims in item (1) fix an exponent p ∈ n−1 n ,1
∗
1 
1 −1
along with some q ∈ (1, n − 1) and set q := q − n−1 . Also, consider an arbitrary
C M -valued
 (p, q)-atom
 a on ∂Ω. Thus, there exist a point xo ∈ ∂Ω and some radius
r ∈ 0, diam(∂Ω) such that the properties in (2.2.72) hold. In addition, in the
case when ∂Ω is bounded, constant functions of absolute value ≤ σ(∂Ω)−1/p are
also considered to be (p, q)-atoms. Making use of Hölder’s inequality, the doubling
property of σ, (1.3.63), and (2.2.72), we may then write
∫
∗
|Sa| p dσ
B(x o ,2r)∩∂Ω
∫ ∗
 p∗ /q∗   1−p∗ /q∗
≤ |Sa| q dσ · σ B(xo, 2r) ∩ ∂Ω
B(x o ,2r)∩∂Ω

p∗   1−p∗ /q∗
≤ Sa ∗
[L q (∂Ω,σ)] M
· σ B(xo, 2r) ∩ ∂Ω

p∗   1−p∗ /q∗
≤C· a [L q (∂Ω,σ)] M
· σ B(xo, r) ∩ ∂Ω = C, (2.2.140)

for some constant C ∈ (0, ∞) independent of the given atom. Going further, pick an
arbitrary point x ∈ ∂Ω \ B(xo, 2r), along with some arbitrary y ∈ ∂Ω ∩ B(xo, r).
Then the Mean Value Theorem and the estimates for E from [70, Theorem 1.4.2]
imply
|E(x − y) − E(x − xo )| ≤ Cr |x − xo | −(n−1) (2.2.141)
300 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

for some constant C ∈ (0, ∞) independent of x, xo, r. Bearing in mind that a belongs
M
to L q (∂Ω, σ) , has bounded support and integrates to zero, (1.3.62) allows us to
express
∫
 
(Sa)(x) = E(x − y) − E(x − xo ) a(y) dσ(y). (2.2.142)
∂Ω

In turn, (2.2.75), (2.2.74), and (2.2.72) permit us to estimate (upon recalling that ∂Ω
is Ahlfors regular)
∫
−(n−1)
|(Sa)(x)| ≤ Cr |x − xo | |a(y)| dσ(y)
∂Ω∩B(x o ,r)
  1−1/q
≤ Cr |x − xo | −(n−1) · a [L q (∂Ω,σ)] M · σ B(xo, r) ∩ ∂Ω

r 1+(n−1)(1−1/p)
≤C . (2.2.143)
|x − xo | n−1

Given that p∗ (n − 1) > n − 1, from (2.2.143) and [68, (7.2.5)] we then obtain
∫ ∫ ∗ ∗
p∗ r p +p (n−1)(1−1/p)
|Sa| dσ ≤ C dσ(x)
∂Ω\B(x o ,2r) ∂Ω\B(x o ,2r) |x − xo | (n−1)p∗
∗ +p ∗ (n−1)(1−1/p) ∗ (n−1)
≤ C · rp · r (n−1)−p = C, (2.2.144)

for some constant C ∈ (0, ∞) independent of the atom a. Finally, combining (2.2.73)
with (2.2.78) proves that for each C M -valued (p, q)-atom a with zero integral on ∂Ω
we have

Sa ∗
[L p (∂Ω,σ)] M ≤C (2.2.145)

for some constant C ∈ (0, ∞) independent of a. Recall that when ∂Ω is bounded,

constant functions of absolute value less than or equal to σ(∂Ω)−1/p are also consid-
ered to be (p, q)-atoms. In such a case, (2.2.145) continues to hold thanks to (1.3.66).
Hence, (2.2.145) is true for all C M -valued (p, q)-atoms a on ∂Ω.
Now that (1.3.63) and (2.2.145) have been established, the plan is now to invoke
∗ ∗
[69, Theorem 4.4.7] with (X, τ) := L q (∂Ω, σ) and (Y, · ) := L p (∂Ω, σ). In
view of the remark made in [69, (4.4.146)] (presently used with θ := 1 and s := 1)
this abstract extension result guarantees that S from (1.3.63) extends uniquely, in the
manner described in [69, (4.4.144)], to a linear
 and bounded operator as in (2.2.116).
That various choices of p in the interval n−1 n , 1 yield operators in (2.2.116) which
are compatible with one another may be justified with the help of the simultaneous
convergence result proved in [69, Theorem 4.4.3].
To justify the final claim in item (1), fix some q ∈ (1, n − 1) and pick an arbitrary
M M
function f ∈ H p (∂Ω, σ) ∩ H 1 (∂Ω, σ) . Then [69, Theorem 4.4.3] ensures
the existence of a sequence
2.2 Single Layer Operators Acting from Hardy Spaces 301

q M M M
{ f N } N ∈N ⊆ Lcomp (∂Ω, σ) ∩ H p (∂Ω, σ) ∩ H 1 (∂Ω, σ)
M M
so that lim f N = f both in H p (∂Ω, σ) and H 1 (∂Ω, σ) , (2.2.146)
N →∞
M
hence also in the space L 1 (∂Ω, σ) .

To clarify notation, let us temporarily write SH p for the operator S in (2.2.116), and
write SL 1 for the operator S in (1.3.64). From what we have proved already it follows
that, for each N ∈ N, we have
∫
SH p f N (x) = E(x − y) f N (y) dσ(y) = SL 1 f N (x) for σ-a.e. x ∈ ∂Ω. (2.2.147)
∂Ω

Based on (2.2.146), the boundedness of (2.2.116), the boundedness of (1.3.64), and

[68, Lemma 6.2.6], we may then write
M
lim SL 1 f N = SL 1 f in L (n−1)/(n−2),∞ (∂Ω, σ) ,
N →∞ (2.2.148)
1 (∂Ω, σ) M
hence also in Lloc ,

and
∗ M
lim SH p f N = SH p f in L p (∂Ω, σ) ,
N →∞ (2.2.149)
1 (∂Ω, σ) M
hence also in Lloc .
Finally, from (2.2.147), (2.2.148), (2.2.149) we conclude that SL 1 f = SH p f at σ-a.e.
point on ∂Ω, hence the desired conclusion follows. This finishes the treatment of the
claims made in item (1).
On to item (2), given r ∈ (n − 1, ∞) define
 n 1  −1  n − 1  n−1
p := − ∈ , 1 and η := 1 − ∈ (0, 1), (2.2.150)
n−1 r n r
then consider p∗ ∈ (1, ∞) such that 1/p∗ = 1/p − 1/(n − 1). These choices entail
1 
(n − 1) − 1 = η and (p∗ ) = r. (2.2.151)
p
Let S A be the boundary-to-boundary single layer associated with transpose coef-
ficient tensor A from (1.3.3) in the same manner S has been associated with A in
(1.3.62). From the current item (1) we know that
M ∗ M
S A : H p (∂Ω, σ) −→ L p (∂Ω, σ) (2.2.152)

is a well-defined linear and bounded operator. Thanks to (2.2.151) and the duality
result recoded in [69, Theorem 4.6.1], its transpose is
⎧
⎪ M
  M ⎨ 𝒞η (∂Ω)
⎪ if ∂Ω is bounded,
S A : L r (∂Ω, σ) −→ . (2.2.153)
⎪
⎪ 𝒞η (∂Ω)/∼ M
if ∂Ω is unbounded.
⎩
302 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

We claim that this is compatible the operator S from (1.3.63). To see that this is
the case, suppose first that ∂Ω is unbounded. In such a scenario, pick an arbitrary
M M
f ∈ L r (∂Ω, σ) ∩ L q (∂Ω, σ) with q ∈ (1, n − 1), and select a representative
  .  
g for S A f , i.e., select a function g ∈ 𝒞η (∂Ω) such that S A f = [g]. Finally,
let q∗ ∈ (1, ∞) satisfy 1/q∗ = 1/q − 1/(n − 1), and denote by ∗ 
 (q ) ∈ (1, n − 1) the
Hölder conjugate exponent of q∗ . Then for each C M -valued p, (q∗ ) -atom a on ∂Ω
we may write
∫
    
g, a dσ = [g], a = S A f , a
∂Ω
∫ ∫
=  f , S A a dσ = S f , a dσ, (2.2.154)
∂Ω ∂Ω

where S A a is considered in the sense of (1.3.62), i.e.,

∫
 
S A a (x) = E A (x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.2.155)
∂Ω

The last equality in (2.2.154) is a consequence of Fubini’s Theorem, whose applica-

bility is ensured by the fact that if I1 is the fractional integral operator of order 1 on
∂Ω (cf. (A.0.100)) then [70, (1.4.24)], Hölder’s inequality, and [68, (7.8.7)] imply
∫ ∫
|E A (x − y)|| f (x)||a(y)| dσ(x) dσ(y)
∂Ω ∂Ω
∫
≤C (I1 | f |)(y)||a(y)| dσ(y)
∂Ω
 
≤ C  I1 | f |  L q ∗ (∂Ω,σ) a [L (q
∗ )
(∂Ω,σ)] M

≤C f [L q (∂Ω,σ)] M a [L (q
∗ )
(∂Ω,σ)] M < +∞. (2.2.156)

This finishes the proof of (2.2.154). In turn, from (2.2.154) and [69, Lemma 4.6.9]
we conclude that [S f ] = [g] = S A f , where S f is computed in the sense of
(1.3.62), Thus, the operator S from (1.3.63) further extends to a linear and bounded
operator as in (2.2.117). Thanks to [68, (3.1.14)], such an extension is necessarily
unique. Dealing with the case when ∂Ω is bounded is very similar, and this completes
the treatment of the first claim in item (2).
Finally, the claim in (2.2.120) may be justified by first considering the case when
∗ 
f is a (p, 2)-atom and when g ∈ L (p ) (∂Ω, σ) ∩ L 2 (∂Ω, σ). In such a scenario, the
desired conclusion is seen from [69, Lemma 4.6.6], (2.2.117) used for the exponent
r := (p∗ ) ∈ (n−1, ∞), and (1.3.62). Having established this, we may then invoke [69,
∗  ∗ 
(4.4.172)], the density of L (p ) (∂Ω, σ) ∩ L 2 (∂Ω, σ) in L (p ) (∂Ω, σ), and continuity
of both sides of (2.2.120) with respect to f and g (cf. (2.2.116), (2.2.117), and the
duality result from [69, Theorem 4.6.1]) to conclude that (2.2.120) in full generality.
2.2 Single Layer Operators Acting from Hardy Spaces 303

Turning to item (3), we run the same type of argument used in the treatment
of item (2) in which we now take r := n − 1. Such a choice leads to considering
p = 1 in (2.2.150) and the role of the Hölder spaces is now played by BMO.
We therefore arrive at the conclusion that S from (1.3.63) extends, in a unique
fashion, to a linear and bounded operator as in (2.2.122). In fact, this conclusion
may be further refined. Specifically, pick an integrability exponent r ∈ (n − 1, ∞)
M M
and, given any f ∈ L n−1 (∂Ω, σ) , select a sequence { f j } j ∈N ⊆ L r (∂Ω, σ)
M
with the property that f j → f in L r (∂Ω, σ) as j → ∞. Then, if we define
η := 1 − (n − 1)/r ∈ (0, 1), from item (2) and (2.2.122) it follows that the sequence
M M
{S f j } j ∈N ⊆ 𝒞η (∂Ω) converges to S f in BMO(∂Ω, σ) if ∂Ω is bounded,
  .η
and the sequence [S f j ] ⊆ 𝒞 (∂Ω)
M 
converges to [S f ] in BMO(∂Ω, σ)
M
j ∈N
if ∂Ω is unbounded. On account of [69, (3.1.50)] and [69, Theorem 3.1.3] we then
M
conclude that actually S f ∈ BMO(∂Ω, σ) , finishing the proof of the first claim
made in item (3).
To deal with the claim in (2.2.123), first consider the case when f is a (1, 2)-atom
and when g ∈ L n−1 (∂Ω, σ) ∩ L 2 (∂Ω, σ). In such a setting, the desired conclusion
is seen from [69, Lemma 4.6.5], (2.2.122), and (1.3.62). Having treated this special
case, we may then rely on [69, (4.4.172)], the density of L n−1 (∂Ω, σ) ∩ L 2 (∂Ω, σ)
in L n−1 (∂Ω, σ), and continuity of both sides of (2.2.123) with respect to f and g (cf.
(2.2.116), (2.2.122), and the duality result from [69, Theorem 4.6.1]) to conclude
that (2.2.123) holds as stated.
As regards item (4), since we already know that the operators from items (1)-(3)
are compatible with S from (1.3.63), we only have to show that said operators act in
a coherent fashion with one another. The latter property is, however, a consequence
of the former and the simultaneous convergence results from [69, Theorem 4.4.3]
and [68, (3.1.14)].
Moving on to item (5), assume first that either p ∈ (1, n − 1) or ∂Ω is bounded,
M
and fix some function f ∈ L p (∂Ω, σ) . Then (1.5.73) and (1.3.62) imply (that all
integrals involved are absolutely convergent is ensured by the Fractional Integration
Theorem, [70, (1.4.24)], and [68, Lemma 7.2.1])
∫ ∫
(Smod f )(x) = E(x − y) f (y) dσ(y) − E∗ (−y) f (y) dσ(y)
∂Ω ∂Ω

= (S f )(x) − C f for σ-a.e. x ∈ ∂Ω, (2.2.157)

∫
with C f := ∂Ω E∗ (−y) f (y) dσ(y) ∈ C M . Consider next the case when ∂Ω is un-
bounded and p ∈ [n − 1, ∞). Fix some integrability exponent q ∈ (1, n − 1) and,
M
having picked an arbitrary function f ∈ L p (∂Ω, σ) , choose a sequence of func-
M M
tions { f j } j ∈N ⊆ L p (∂Ω, σ) ∩ L q (∂Ω, σ) which converges to f in the space
M
L p (∂Ω, σ) . Write SL p and SL q for the versions of the single layer operator from
M M
(2.2.124) acting from the spaces L p (∂Ω, σ) and L q (∂Ω, σ) , respectively.

1 (∂Ω, σ) ∼ M
Then, with convergence in Lloc , we have
304 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Smod f = lim Smod f j = lim SL q f j = lim SL p f j , (2.2.158)

j→∞ j→∞ j→∞

M
where the first equality is implied by (1.5.74), bearing in mind that L p (∂Ω, σ)
  M
embeds continuously into L 1 ∂Ω, 1+σ(x) |x | n−1
, the second equality is a consequence
of (2.2.157) written for f j in place of f , and the third equality is guaranteed by the
first claim in the current item (4). In turn, from (2.2.158) we conclude that, after
eventually passing to a subsequence, SL p f jk converges to Smod f pointwise σ-a.e.
on ∂Ω as k → ∞. On the other hand, the current item (4) gives that lim SL p f jk = S f
k→∞
 M . M
in BMO(∂Ω, σ) or 𝒞η (∂Ω)/∼ , depending on whether p = n−1 or p > n−1
(in which case η = 1 − (n − 1)/p) which goes to show that SL p f jk has a subsequence
which converges to S f pointwise σ-a.e. on ∂Ω. Altogether, this argument shows that
S f = Smod f at σ-a.e. point in ∂Ω, completing the treatment of item (5).
Consider next the claims in item (6). The case when p ∈ (1,n − 1) is covered
by (1.3.67), so there remains to deal with situation when p ∈ n−1 n , 1 . Fix such
an exponent p along with an aperture parameter κ > 0. The idea is to use [68,
Proposition 6.2.11] for the following choices. First, take 𝒳 := Ω ∪ Aκ (∂Ω) presently
endowed with the topology inherited from the ambient Euclidean space, and X :=
Aκ (∂Ω). Hence, 𝒳 \ X = Ω. Second, we take μ := H n−1  Aκ (∂Ω) which is a locally
finite complete Borel-regular measure on X by virtue of [68, Lemma 3.6.4] (the
fact that the hypotheses in [68, (3.6.25)] are presently satisfied is seen from [68,
(8.8.5)] and the upper Ahlfors regularity of ∂Ω). If for each x ∈ X = Aκ (∂Ω)
we take Γ(x) := Γκ (x), then condition [68, (6.2.71)] is satisfied thanks to (A.0.2).
M
Next, we take Y := H p (∂Ω, σ) which is a quasi-Banach space, and consider the
operator T mapping vectors f ∈ Y into C M -valued continuous functions defined in
𝒳 \ X = Ω according to T f := 𝒮 f (cf. (2.2.32)). Since 𝒮 is linear, [68, (6.2.72)]
p,q M
holds. Going further, fix some q ∈ (1, n − 1) and take V := Hfin (∂Ω, σ) . From
[69, (4.4.114)] we know that V is a dense linear subspace of Y . Finally, take the
integrability exponent p appearing in the statement of [68, Proposition 6.2.11] to
presently be p∗ . These choices imply that the associated maximal operator (cf. [68,
(6.2.73)]) is
M
(T f )(x) = Nκ (𝒮 f )(x), ∀ f ∈ H p (∂Ω, σ) , ∀x ∈ Aκ (∂Ω). (2.2.159)

Since, according to item (3) in Theorem 2.2.3, this maximal operator maps
M ∗  
H p (∂Ω, σ) boundedly into the Lorentz space L p ,∞ Aκ (∂Ω), H n−1  Aκ (∂Ω) ,
it follows that hypothesis [68, (6.2.74)] is satisfied. Finally, that for every f ∈ V the
limit in [68, (6.2.75)] exists becomes a consequence of (1.3.67) (bearing in mind
[69, (4.4.114)] and our choice of q).
At this stage, the application of [68, Proposition 6.2.11] in the manner just de-
scribed guarantees that
2.2 Single Layer Operators Acting from Hardy Spaces 305
M 
for each f ∈ H p (∂Ω, σ) with p ∈ n−1
n , 1 the nontangential
κ−n.t.

limit (𝒮 f) exists at H -a.e. point in Aκ (∂Ω), and we have
n−1
(2.2.160)
 κ−n.t.  ∂Ω

(𝒮 f ) ∂Ω  ≤ Nκ (𝒮 f ) at H n−1 -a.e. point in Aκ (∂Ω).

Having proved this, [68, Corollary 8.9.7] (used with μ := H n−1 ∂Ω) additionally
gives that
M 
for each f ∈ H p (∂Ω, σ) with p ∈ n−1 , 1 the function defined
 κ−n.t.n (2.2.161)

H -a.e. as Aκ (∂Ω)  x −→ (𝒮 f ) ∂Ω (x) is H n−1 -measurable.
n−1

From (2.2.160),
 (2.2.161), and item (3) in Theorem 2.2.3 we then conclude that for
each p ∈ n−1
n , 1 there exists C ∈ (0, ∞) with the property that
 κ−n.t. 

(𝒮 f )∂Ω  p ∗ ≤ C f [H p (∂Ω,σ)] M , (2.2.162)
[L (A κ (∂Ω), H
n−1  A κ (∂Ω))]
M

M
for every f ∈ H p (∂Ω, σ) . Hence,
κ−n.t. ∗ 
 f −→ (𝒮 f )∂Ω ∈ L p Aκ (∂Ω), H n−1  Aκ (∂Ω)
M M
H p (∂Ω, σ)
is a well-defined, linear and bounded operator.
(2.2.163)

On the other hand, item (1) of the current theorem implies that S also induces a
well-defined, linear and bounded operator in the same functional analytic context.
Moreover, as seen from item (2) in Theorem 2.2.3, item (2) of the current theorem,
p,q M
(1.3.6), (1.3.62), and (1.3.67), when acting on Hfin (∂Ω, σ) , these two operators
yield functions which agree H n−1 -a.e. on Aκ (∂Ω). Since the operators in question-
p,q M M
s are continuous and Hfin (∂Ω, σ) is dense in H p (∂Ω, σ) , we ultimately
M
conclude that (2.2.127) holds for each f ∈ H p (∂Ω, σ) .
As regards the claims in item (7), the boundedness of the boundary-to-boundary
single layer operator in the context of (2.2.128) is a consequence of the current item
(4), [69, Theorem 4.3.1], [68, (6.2.48)], and real interpolation. Finally, that for each
κ−n.t.
the nontangential boundary limit (𝒮 f )∂Ω exists σ-a.e. on
M
f ∈ H p,q (∂Ω, σ)
Aκ (∂Ω), and that the boundary trace formula (2.2.127) continues to hold in this case,
may be justified by reasoning as in the proof of (2.2.126)-(2.2.127).
Let us now deal with item (8), working in the two-dimensional setting. For starters,
all claims up to, and including (2.2.132), made in relation to the operator (2.2.130)-
 
(2.2.131) are direct consequence of (2.2.36). To proceed, fix an exponent p ∈ 12 , 1
  −1
and set p∗ := p1 − 1 . Also, pick some C M -valued (p, 2)-atom a on ∂Ω. Retaining
earlier notation, in place of (2.2.140) we now have
306 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
∫
  p∗  
|Sa| p dσ ≤ Sa[L ∞ (∂Ω,σ)] M · σ B(xo, 2r) ∩ ∂Ω
∗

B(x o ,2r)∩∂Ω
p∗
≤ Cr · a [H 1 (∂Ω,σ)] M
≤ C, (2.2.164)

thanks to the boundedness of the operator (2.2.130), the Ahlfors regularity of ∂Ω,
the estimate recorded in (2.2.84), and the definition of p∗ . Above, C ∈ (0, ∞) is
a constant independent of the atom in question. Since the argument in (2.2.141)-
(2.2.144) continues to work when n = 2, we ultimately conclude that we presently
still have (2.2.145) for all C M -valued (p, 2)-atoms a on ∂Ω. With this in hand, we
may now invoke [69, Theorem 4.4.7] with
∗
q := 1, (X, τ) := L ∞ (∂Ω, σ), and (Y, · ) := L p (∂Ω, σ). (2.2.165)

Bearing in mind [69, (4.4.146)] (presently used with θ := 1 and s := 1) this

guarantees that S from (2.2.130)-(2.2.131) extends uniquely, in the manner described
in [69, (4.4.144)], to a linear and bounded operator in the context of (2.2.133).
Next, the fact that the single layer in the context of (2.2.133) continues to satisfy
M
(2.2.132) for each κ > 0 and each f ∈ H p (∂Ω, σ) may be justified arguing
much as in the proof of the claims made in the current item (6), now making use of
(2.2.133). The very last claim in item (8), concerning the validity of (2.2.132) for
M  
each f ∈ H p,q (∂Ω, σ) with p ∈ 12 , 1 and q ∈ (0, ∞], is seen from what we
have just proved and [69, (1.3.41), (4.3.3)].
Finally, under the assumptions in (2.2.136) the claims in (2.2.136)-(2.2.138) may
be justified by reasoning much as in the proofs of (2.2.117) and (2.2.120), making
use of (2.2.133). This finishes the proof of Theorem 2.2.6. 

In the last part of this section, we wish to augment the result established in
Theorem 1.8.19 by now working in a setting which places the (weak) conormal
derivative of the function in question in a Hardy space H p with p ≤ 1. Specifically,
we have the basic integral representation formula stated in following theorem6:

Theorem 2.2.7 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain with the

property that ∂Ω is an Ahlfors regular set. Denote by ν the geometric measure
theoretic outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, for some
M ∈ N, let L be a weakly elliptic homogeneous second-order M × M system in Rn ,
with constant complex coefficients. Next, fix an aperture parameter κ ∈ (0, ∞) and
consider a function u : Ω → C M satisfying
M
u ∈ 𝒞∞ (Ω) ,
Lu = 0 in Ω, and
 (2.2.166)
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1
n ,1 .

Then the following statements are true.

6 for a related result, see Theorem 2.3.16

2.2 Single Layer Operators Acting from Hardy Spaces 307
κ−n.t.
(a) The nontangential boundary trace u∂Ω exists at σ-a.e. point on ∂Ω and, as a
  M
function, this trace belongs to the weighted Lebesgue space L 1 ∂Ω, 1+σ(x) |x | n .
1 
(b) If either n ≥ 3, or n = 2 and p ∈ 2 , 1 , it follows that there exists some constant
κ−n.t.
c ∈ C M with the property that u∂Ω − c belongs to the space L p (∂Ω, σ) ,
∗ M

 
1 −1
κ−n.t. p∗
∈ (1, ∞). In particular, u∂Ω ∈ Lloc (∂Ω, σ)
M
where p∗ := p1 − n−1
in this scenario. Finally, in the remaining case, i.e., when n = 2 and p = 1,
κ−n.t.
the nontangential boundary trace u∂Ω belongs to the John-Nirenberg space
M
BMO(∂Ω, σ) .
(c) In the case when Ω is an exterior domain make the additional assumption that
there exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (2.2.167)
B(0,λ R)\B(0,R)
 αβ 
Then for each complex coefficient tensor A = ar s 1≤r,s ≤n with the property
.A 1≤α,β ≤M
that L = L A the weak conormal derivative ∂ν u belongs to the Hardy space
M
H p (∂Ω, σ) and there exists some C M -valued locally constant function cu
in Ω with the property that
 κ−n.t.  . 
u = Dmod u∂Ω − 𝒮 ∂νAu + cu in Ω, (2.2.168)

where Dmod is the modified double layer potential operator associated with the
  M
coefficient tensor A (acting on functions from L 1 ∂Ω, 1+σ(x)
|x | n as in (1.8.6)),
and where the single layer potential operator acts on distributions in the Hardy
M
space H p (∂Ω, σ) as in (2.2.32).
 
(d) Assume that either n ≥ 3, or n = 2 and p ∈ 12 , 1 and Ω not an exterior domain.
Then there exists a constant c ∈ C M such that
∗  
1 −1
Nκ (u − c) ∈ L p (∂Ω, σ) where p∗ := p1 − n−1 ∈ (1, ∞), (2.2.169)

and, as a consequence, in this case one has

 σ(x) 
p∗
Nκ u ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, . (2.2.170)
1 + |x| n
 
Also, in the case when n = 2, p ∈ 12 , 1 , and Ω is an exterior domain, the same
results in (2.2.169)-(2.2.170) remain true provided a truncated nontangential
maximal operator, say NκΩ∩B(0,R) with R ∈ (0, ∞), is employed in place of
Nκ = NκΩ .
Finally, in the case when n = 2 and p = 1 one has
308 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Nκε u ∈ Lloc (∂Ω, σ) for each ε > 0 and each q ∈ (0, ∞).
q
(2.2.171)

Through the consideration of the sub-unital integrability exponents, Theo-

rem 2.2.7 complements the result in Theorem 1.8.19 (which deals with the super-
unital range). An integral representation similar in spirit to (2.2.168) is also es-
tablished in Theorem 2.3.16 under weaker geometric hypotheses on the underlying
domain Ω but for stronger analytical assumptions on the function u. See also Corol-
lary 2.3.17 for a version of (2.2.168) in which we are allowed to take cu = 0.
Proof of Theorem 2.2.7 For starters, [68, Proposition 8.9.22] and [68, (5.2.4)] en-
sure that the nontangential trace
κ−n.t.
u∂Ω exists at σ-a.e. point on ∂Ω, (2.2.172)

and the function

⎧
⎨ Nκ (∇u) if ∂Ω is unbounded,
⎪
⎪
g :=  κ−n.t.  (2.2.173)
⎪
⎪ u∂Ω  + Nκ (∇u) if ∂Ω is bounded,
⎩
belongs to L p (∂Ω, σ). Granted this, from [68, (5.11.28)] and [69, Lemma 11.5.8]
we see that there exists a constant C ∈ (0, ∞) independent of u with the property that
  κ−n.t.   κ−n.t.  
 
 u ∂Ω (x) − u∂Ω (y) ≤ C|x − y| · g(x) + g(y)
(2.2.174)
for σ-a.e. points x, y ∈ ∂Ω.
κ−n.t.
Then [69, Lemma 11.5.9] applies with f := u∂Ω and Σ := ∂Ω. If either n ≥ 3 or
 
n = 2 and p ∈ 12 , 1 , this guarantees the existence of some constant c ∈ C M such
κ−n.t.
that the function u∂Ω − c belongs to L p (∂Ω, σ) . On account of this, we may
∗ M

also conclude that

κ−n.t.   σ(x)   M
u∂Ω ∈ L p (∂Ω, σ)
∗ M
+ C M → L 1 ∂Ω, , (2.2.175)
1 + |x| n
in this scenario. In the remaining case, i.e., when n = 2 and p = 1, from item (2) in
[69, Lemma 11.5.9] and [68, (7.4.118)] (used here with p = 1, d = 1, and ε = 1) we
see that
κ−n.t.   σ(x)   M
u∂Ω ∈ BMO(∂Ω, σ)
M
→ L 1 ∂Ω, . (2.2.176)
1 + |x| 2
Altogether, the argument so far takes care of the claims made in items (a)-(b).
Going further, from the second line in (2.2.166) and [68, (8.6.51)] we see (also
bearing in mind (A.0.105)) that
np/(n−1) M ·n
∇u ∈ Lbdd (Ω, L n ) . (2.2.177)
2.2 Single Layer Operators Acting from Hardy Spaces 309

In addition, from [68, Proposition 8.4.9] we deduce that

ρ p
Nκ u ∈ Lloc (∂Ω, σ) for each ρ > 0. (2.2.178)

In concert with [68, (8.1.18), (8.6.51)], this also implies

np/(n−1) M
u ∈ Lbdd (Ω, L n ) . (2.2.179)

Ultimately, from (2.2.177)-(2.2.179) and (A.0.208) we conclude that

1, np/(n−1) M
u ∈ Wbdd (Ω) . (2.2.180)

For later use, let us also note here that (2.2.172), (2.2.178), [68, Corollary 8.9.9],
[68, (8.8.52)], and [68, (5.2.4)] guarantee that
κ −n.t.
u∂Ω exists σ-a.e. on ∂Ω for each  κ ∈ (0, ∞), and this (2.2.181)
trace is actually independent of the aperture parameter.
Henceforth, in the case when Ω is an exterior domain, make the additional
assumption that (2.2.167)
 αβholds
 for some λ ∈ (1, ∞). To proceed, pick a complex
coefficient tensor A = ar s 1≤r,s ≤n with the property that L = L A. Then, granted
1≤α,β ≤M
the present hypotheses, [70, Theorem 3.3.1] yields a number of conclusions. First,
for any , s ∈ {1, . . . , n} we have
. M . M
∂τ s u ∈ H p (∂Ω, σ) and ∂νAu ∈ H p (∂Ω, σ) . (2.2.182)

Second, if E = (Eγβ )1≤γ,β ≤M is the matrix-valued fundamental solution associated

with L as in [70, Theorem 1.4.2] and if (uγ )1≤γ ≤M are the scalar components
of the function u, then with the duality pairings understood in the sense of [69,
Theorem 4.6.1] with Σ := ∂Ω and the summation convention over repeated indices
in effect it follows that for each  ∈ {1, . . . , n} and each γ ∈ {1, . . . , M } we have
#  . $
(∂ uγ )(x) = ar s (∂r Eγβ )(x − ·)∂Ω , ∂τ s uα
βα

#  .  $
− (∂ Eγβ )(x − ·)∂Ω , ∂νAu β , for all x ∈ Ω, (2.2.183)

assuming ∂Ω is unbounded. A similar formula holds when ∂Ω is bounded, this time

omitting taking equivalence classes of functions modulo constants in the duality
pairings in (2.2.183). The idea is to further transform the first duality pairing in
(2.2.183). We shall do so while henceforth assuming that ∂Ω is unbounded, since a
very similar (and slightly simpler) argument works in the case when ∂Ω is bounded
as well.
To get started, fix , s, r ∈ {1, . . . , n} along with α,
 β, γ ∈ {1, . . . , M }. Also,
pick x ∈ Ω then choose some ε ∈ 0, 12 dist(x, ∂Ω) . Consider a scalar-valued
function η ∈ 𝒞∞ (Rn ) satisfying η = 0 on B(0, 1), η = 1 on Rn \ B(0, 2), and define
310 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 
ηε (y) := η (y − x)/ε for every y ∈ Rn . This ensures that

ηε ∈ 𝒞∞ (Rn ), ηε ≡ 1 on Rn \ B(x, 2ε), ηε ≡ 0 on B(x, ε). (2.2.184)

As a consequence, we have that the function Eγβ (x − ·)ηε belongs to 𝒞∞ (Rn ) and
coincides with Eγβ (x − ·) near ∂Ω. Next, choose a function θ ∈ 𝒞∞
c (R ) satisfying
n

0 ≤ θ ≤ 1, θ ≡ 1 on B(0, 1), θ ≡ 0 on Rn \ B(0, 2) and, for each R > 0, define

θ R (y) := θ(y/R) for every y ∈ Rn . The general weak-∗ convergence results from
[69, Lemma 4.8.4] and [69, Lemma 4.8.1] (also bearing in mind the trivial bounded
embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)) imply that we have
  
(∂r Eγβ )(x − ·)∂Ω = (∂r Eγβ )(x − ·)ηε ∂Ω
    ∗
= lim θ R (∂r Eγβ )(x − ·)ηε ∂Ω weak-∗ in H p (∂Ω, σ)
R→∞

⎧ . 
⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼ if p < 1,
⎪
= (2.2.185)
⎪ 
⎪ BMO(∂Ω,
⎩ σ) if p = 1.

Consequently,
, -
#  . $    .
(∂r Eγβ )(x − ·)∂Ω , ∂τ s uα = lim θ R (∂ Eγβ )(x − ·)ηε ∂Ω , ∂τ s uα .
R→∞

(2.2.186)

Let us momentarily abbreviate

Ψ := θ R (∂r Eγβ )(x − ·)ηε . (2.2.187)

Then Ψ ∈ 𝒞∞ c (R ) and for each R > 0 we have

n

, -
   .
θ R (∂r Eγβ )(x − ·)ηε  , ∂τ uα
∂Ω s

, -
   .
= Lipc (∂Ω) θ R (∂r Eγβ )(x − ·)ηε ∂Ω , ∂τ s uα (Lip c (∂Ω))

∫
 
= (∂s uα )(∂ Ψ) − (∂ uα )(∂s Ψ) dL n, (2.2.188)
Ω

thanks to [69, Lemma 4.6.4] and [68, Example 4.2.4] (cf. (A.0.175)-(A.0.176) in the
Glossary). At this stage, we may invoke [68, Proposition 2.8.17] for the vector field

F := uα (∂ Ψ)es − uα (∂s Ψ)e (2.2.189)

2.2 Single Layer Operators Acting from Hardy Spaces 311

and with p replaced by np/(n − 1) ∈ (1, ∞), its applicability in the present context
being ensured by (2.2.180) and (2.2.181). From this and the independence of the
nontangential boundary trace on the aperture parameter (cf. (2.2.181)) we may then
conclude that
, -
   .
θ R (∂r Eγβ )(x − ·)ηε  , ∂τ uα∂Ω s

∫
 κ−n.t.  %   &
= uα ∂Ω νs (∂ Ψ)∂Ω − ν (∂s Ψ)∂Ω dσ. (2.2.190)
∂Ω

Upon recalling the definition of Ψ from (2.2.187) and bearing in mind that ηε ≡ 1
near ∂Ω while ∇ηε is supported away from ∂Ω (as may be seen from (2.2.184)), we
may then conclude from (2.2.186) and (2.2.190) that
#  . $
ar s (∂r Eγβ )(x − ·)∂Ω , ∂τ s uα = lim IR − lim IIR
βα
(2.2.191)
R→∞ R→∞

where, for each R > 0, we have set

∫ % &
βα  κ−n.t. 
IR := ar s uα ∂Ω νs (∂ θ R )(∂r Eγβ )(x − ·) − ν (∂s θ R )(∂r Eγβ )(x − ·) dσ,
∂Ω
(2.2.192)

and
∫ % &
βα  κ−n.t. 
IIR := ar s uα ∂Ω νs θ R (∂ ∂r Eγβ )(x − ·) − ν θ R (∂s ∂r Eγβ )(x − ·) dσ.
∂Ω
(2.2.193)

Since for each fixed x ∈ Ω, each y ∈ ∂Ω, and each R > 0 we have
Cx
|(∇θ R )(y)| ≤ CR−1 1 |y |≈R and |(∇E)(x − y)| ≤ , (2.2.194)
1 + |y| n−1
it follows that
∫
    κ−n.t.    −1  1
IR  ≤ C(x, A, ∂Ω)  u (y) R 1 |y |≈R dσ(y)
∂Ω 1 + |y| n−1
∂Ω

∫   κ−n.t. 
 u (y)
∂Ω
≤ C(x, A, ∂Ω) dσ(y). (2.2.195)
1 + |y| n
y ∈∂Ω
|y |≈R

κ−n.t.  
In view of the fact that u∂Ω ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n (cf. (2.2.175)-(2.2.176)),
Lebesgue’s Dominated Convergence Theorem gives
312 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

∫   κ−n.t. 
 u (y)
∂Ω
lim dσ(y) = 0. (2.2.196)
R→∞ 1 + |y| n
y ∈∂Ω
|y |≈R

From (2.2.195) and (2.2.196) we then conclude that

lim IR = 0. (2.2.197)
R→∞

Next, for each R > 0 decompose

IIR = II(a) (b)

R + II R (2.2.198)

where
∫
βα  κ−n.t. 
II(a)
R := ar s uα ∂Ω νs θ R (∂ ∂r Eγβ )(x − ·) dσ, (2.2.199)
∂Ω

and
∫
βα  κ−n.t. 
II(b)
R := − ar s uα ∂Ω ν θ R (∂s ∂r Eγβ )(x − ·) dσ. (2.2.200)
∂Ω

βα
Upon noting that ar s (∂s ∂r Eγβ )(x − ·) = 0 on ∂Ω (as seen from [70, (1.4.33)]), we
deduce that
II(b)
R = 0 for each R > 0. (2.2.201)
Also, since for each y ∈ ∂Ω we have
Cx
lim θ R (y) = 1, |θ R (y)| ≤ C, and |(∇2 E)(x − y)| ≤ , (2.2.202)
R→∞ 1 + |y| n
Lebesgue’s Dominated Convergence Theorem presently applies, keeping in mind
κ−n.t.   M
that, as noted in (2.2.175)-(2.2.176), we have u∂Ω ∈ L 1 ∂Ω, 1+σ(x) |x | n . This
permits us to compute
∫
 κ−n.t. 
νs (y)ar s θ R (y)(∂ ∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
lim II(a)
R = lim
R→∞ R→∞ ∂Ω
∫
 κ−n.t. 
νs (y)ar s (∂ ∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
=
∂Ω
  κ−n.t.  
= −∂ Dmod u∂Ω (x), (2.2.203)
γ

with the last equality coming from (1.8.11) (again, keeping in mind (2.2.175)-
(2.2.176)).
Collectively, (2.2.191), (2.2.197), (2.2.198), (2.2.201), and (2.2.203) then imply
that the first term in the right-hand side of (2.2.183) may be recast as
2.2 Single Layer Operators Acting from Hardy Spaces 313
#  . $   κ−n.t.  
(∂r Eγβ )(x − ·)∂Ω , ∂τ s uα = ∂ Dmod u∂Ω
βα
ar s (x). (2.2.204)
γ

From (2.2.34) we also see that the second term in the right-hand side of (2.2.183)
may be expressed as
#  .  $ .
− (∂ Eγβ )(x − ·)∂Ω , ∂νAu β = −∂ 𝒮(∂νAu) γ (x). (2.2.205)

Altogether, from (2.2.183), (2.2.204), and (2.2.205) we conclude that

  κ−n.t.   .
(∂ uγ )(x) = ∂ Dmod u∂Ω (x) − ∂ 𝒮(∂νAu) γ (x), ∀x ∈ Ω. (2.2.206)
γ

In view of the arbitrariness of  ∈ {1, . . . , n} and each γ ∈ {1, . . . , M }, this further

implies that there exists some C M -valued locally constant function cu in Ω such that
(2.2.168) holds. This concludes the treatment of item (c).
As regards item (d), work under the assumption that either n ≥ 3, or n = 2 and
  κ−n.t.
p ∈ 12 , 1 . Then, as we have seen in item (b), there exists c ∈ C M such that u∂Ω − c
∗ M
belongs to L p (∂Ω, σ) . Use (2.2.168) to write
 κ−n.t.  . 
u = Dmod u∂Ω − c − 𝒮 ∂νAu + cu(1) in Ω, (2.2.207)

where cu(1) := cu + Dmod c is a locally constant C M -valued function in Ω (as may

κ−n.t.
be seen from (1.8.10)). Since u∂Ω − c ∈ L p (∂Ω, σ)
∗ M
with p∗ ∈ (1, ∞) in the
current case, we may invoke (1.8.8) to conclude that the difference
 κ−n.t.   κ−n.t. 
cu(2) := Dmod u∂Ω − c − D u∂Ω − c
(2.2.208)
is a constant (belonging to C M ) in Ω.

cu := cu(1) + cu(2) is a locally constant C M -valued functions in Ω and

In particular, 
(2.2.207)-(2.2.208) entail
 κ−n.t.  . 
u = D u∂Ω − c − 𝒮 ∂νAu + 
cu in Ω. (2.2.209)

In turn, this implies that for each Lebesgue measurable set E ⊆ Ω we have
  κ−n.t.    . 
cu ) ≤ NκE D u∂Ω − c + NκE 𝒮 ∂νAu
NκE (u −  (2.2.210)

at each point on ∂Ω. Recall from (1.5.1) and (2.2.39), (2.2.40) that we have
  κ−n.t. 
Nκ D u∂Ω − c ∈ L p (∂Ω, σ),
∗
(2.2.211)

as well as
314 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
  .  ∗
Nκ 𝒮 ∂νAu ∈ L p (∂Ω, σ)
(2.2.212)
if either n ≥ 3, or n = 2 and Ω is not an exterior domain,

and
  . 
NκΩ∩B(0,R) 𝒮 ∂νAu ∈ L p (∂Ω, σ)
∗

(2.2.213)
for each R ∈ (0, ∞) if n = 2 and Ω is an exterior domain.

Then from (2.2.210), (2.2.211)-(2.2.213), and [68, (8.2.28)] we conclude that there
exists a locally constant C M -valued function 
cu in Ω such that
∗
Nκ (u − 
cu ) ∈ L p (∂Ω, σ)
(2.2.214)
if either n ≥ 3, or n = 2 and Ω is not an exterior domain,

while

NκΩ∩B(0,R) (u − 
∗
cu ) ∈ L p (∂Ω, σ)
(2.2.215)
for each R ∈ (0, ∞) if n = 2 and Ω is an exterior domain.

If ∂Ω is unbounded then [68, Lemma 5.11.3] gives that Ω is connected so  cu is

actually a genuine constant in C M , which establishes (2.2.169) in this case. If ∂Ω
is bounded then [68, Lemma 5.11.3] tells us that Ω has finitely many connected
components, so the range of the locally constant C M -valued function  cu is finite.
∗
Since in this case ∂Ω has finite measure, (2.2.214) implies that Nκ u ∈ L p (∂Ω, σ) if
either n ≥ 3, or n = 2 and Ω is not an exterior domain, so (2.2.169) presently holds
with c := 0. In turn, (2.2.170) is a consequence of (2.2.169). Also, (2.2.215) implies
that if n = 2 and Ω is an exterior domain then NκΩ∩B(0,R) u ∈ L p (∂Ω, σ) for each
∗

R ∈ (0, ∞), so the desired conclusions in this case follow as well.

Finally, that (2.2.171) holds in the case when n = 2 and p = 1 is seen from
κ−n.t.
(2.2.168), the fact that we now have u∂Ω ∈ BMO(∂Ω, σ) , the last embedding
M

in [68, (7.4.105)], [68, (7.4.118)] used with d := 1 and ε := 1, (1.8.22) used with
n = 2 and q in place of p, (2.2.41)-(2.2.42), and the fact that (as already observed
above) any locally constant function in Ω has finite range. 

We conclude with a result in the spirit of [69, Theorem 11.10.10] in which

no explicit assumptions are made on the nontangential boundary trace and the
nontangential maximal operator of the function itself.

Theorem 2.2.8 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain whose

boundary is an Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω, and fix an aperture
parameter κ ∈ (0, ∞). Also, suppose u : Ω → C is a function satisfying

u belongs to 𝒞1 (Ω) and

 n−1  (2.2.216)
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n ,∞ .
2.2 Single Layer Operators Acting from Hardy Spaces 315
κ−n.t.
Then the nontangential boundary trace u∂Ω exists at σ-a.e. point on ∂Ω and
κ−n.t.
is independent of the aperture parameter κ. As a function, u∂Ω belongs to the
.p
homogeneous Hardy-based Sobolev space H1 (∂Ω, σ), and satisfies
 κ−n.t.  .
∂τ j k u∂Ω = ∂τ j k u ∈ H p (∂Ω, σ) for all j, k ∈ {1, . . . , n}, (2.2.217)

as well as  κ−n.t.   
u  .p ≤ C Nκ (∇u) L p (∂Ω,σ) (2.2.218)
∂Ω 1
H (∂Ω,σ)

for some constant C = C(Ω, n, κ, p) ∈ (0, ∞) independent of u.

It is noteworthy that Theorem 2.2.8 may be used in concert with [70, Theo-
rem 3.3.1] and item (i) in Theorem 2.3.1 to give an alternative proof of Theo-
rem 2.2.7.
Proof of Theorem 2.2.8 The case when p ∈ (1, ∞) is a consequence of [70, Lem-
ma 2.5.9], [69, (4.2.9), (11.10.30)], [69, Proposition 10.2.11], and [68, Proposi-
tion 8.4.9], so we shall restrict attention for the rest of the proof to the case when
 n−1
p∈ n ,1 . (2.2.219)
First, [68, Proposition 8.9.22] and [68, (5.2.4)] guarantee that the nontangential
trace κ−n.t.
u∂Ω exists at σ-a.e. point on ∂Ω, (2.2.220)
and that
⎧
⎨ Nκ (∇u) if ∂Ω is unbounded,
⎪
⎪
g :=  κ−n.t.  (2.2.221)
⎪  
⎪ u∂Ω  + Nκ (∇u) if ∂Ω is bounded,
⎩
is a well-defined function belonging to L p (∂Ω, σ). As such, [68, (5.11.28)] and [69,
Lemma 11.5.8] show that there exists a constant C ∈ (0, ∞) independent of u with
the property that
  κ−n.t.   κ−n.t.  
 
 u ∂Ω (x) − u∂Ω (y) ≤ C|x − y| · g(x) + g(y)
(2.2.222)
for σ-a.e. points x, y ∈ ∂Ω.
κ−n.t.
Granted this, [69, Lemma 11.5.9] applies with f := u∂Ω and Σ := ∂Ω. When
1 
either n ≥ 3, or n = 2 and p ∈ 2 , 1 , this implies the existence of some constant
κ−n.t.
c ∈ C such that u∂Ω − c is a function belonging to the space L p (∂Ω, σ), where
∗

 1 −1

p∗ := p1 − n−1 ∈ (1, ∞). With this in hand, we then conclude that in this scenario
we have 
κ−n.t. σ(x) 
u∂Ω ∈ L p (∂Ω, σ) + C → L 1 ∂Ω,
∗
. (2.2.223)
1 + |x| n
316 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

In the remaining case, i.e., when n = 2 and p = 1, from item (2) in [69, Lemma 11.5.9]
and [68, (7.4.118)] (presently used with p := 1, d := 1, and ε := 1) we see that
κ−n.t.  σ(x) 
u∂Ω ∈ BMO(∂Ω, σ) → L 1 ∂Ω, . (2.2.224)
1 + |x| 2
The argument so far shows that, in all cases,
κ−n.t.  σ(x) 
u∂Ω ∈ L 1 ∂Ω, ⊆ Lloc
1
(∂Ω, σ). (2.2.225)
1 + |x| n
Next, from the second line in (2.2.216) and [68, (8.6.51) in Proposition 8.6.3] we
see (also keeping in mind (A.0.105)) that
np/(n−1) n
∇u ∈ Lbdd (Ω, L n ) . (2.2.226)

In addition, from [68, Proposition 8.4.9] we deduce that

ρ p
Nκ u ∈ Lloc (∂Ω, σ) for each ρ > 0. (2.2.227)

In concert with [68, (8.6.51) in Proposition 8.6.3] and [68, (8.1.18)], this also implies
np/(n−1)
u ∈ Lbdd (Ω, L n ). (2.2.228)

Together, (2.2.226), (2.2.228), and (A.0.208) permit us to conclude that

1,np/(n−1)
u ∈ Wbdd (Ω). (2.2.229)

Let us also record here that, as seen from (2.2.220), (2.2.227), [68, Corollary 8.9.9],
and [68, (5.2.4), (8.8.52)], the nontangential trace
κ −n.t.
u∂Ω exists σ-a.e. on ∂Ω for each  κ ∈ (0, ∞), and this (2.2.230)
trace is actually independent of the aperture parameter.

Fix now an arbitrary test function ψ ∈ Lipc (∂Ω), and consider a function Ψ
satisfying 
Ψ ∈ Lip( Ω ), Ψ∂Ω = ψ, and Ψ ≡ 0
(2.2.231)
outside of some compact subset of Ω.
Let us also fix a pair of indices j, k ∈ {1, . . . , n}. We claim that

there exists some sequence {ϕ } ∈N ⊆ 𝒞∞

c (R ) such that
n

supp ϕ ⊆ K for all  ∈ N and some compact set K ⊆ Rn,

ϕ −→ Ψ uniformly on compact sets in Rn as  → ∞, (2.2.232)
   
∇ϕ Ω −→ ∇Ψ Ω weak-∗ in L ∞ (Ω, L n ) as  → ∞,
n

∂τ j k ϕ converges to ∂τ j k ψ weak-∗ in L ∞ (∂Ω, σ) as  → ∞.

2.2 Single Layer Operators Acting from Hardy Spaces 317

Indeed, such a sequence is constructed as in [69, Remark 11.1.13] (with ∂∗ Ω replaced

by ∂Ω, f replaced by ψ, and F replaced by Ψ) by mollifying Ψ. The property in the
third line of (2.2.232) comes from [69, Lemma 11.1.12], and the property in the last
line of (2.2.232) is proved as in [69, (11.3.13)].
To proceed, for each  ∈ N define the vector field

F := u(∂k ϕ )e j − u(∂j ϕ )ek ∈ 𝒞1 (Ω) .

n
(2.2.233)

From (2.2.233) and (2.2.231) we then see that

F belongs to the space W 1,np/(n−1) (Ω) ,

n
(2.2.234)
and vanishes outside of K ∩ Ω,

while (2.2.230) further implies that

κ −n.t.
F ∂Ω exists σ-a.e. on ∂Ω for each κ ∈ (0, ∞), and this trace (2.2.235)
is actually independent of the aperture parameter.

Let us also observe that, thanks to (2.2.233) and (2.2.220), at σ-a.e. point on ∂Ω we
have
κ−n.t.  κ−n.t.    κ−n.t.  
F ∂Ω = u∂Ω (∂k ϕ )∂Ω e j − u∂Ω (∂j ϕ )∂Ω ek ,
(2.2.236)
and div F = (∂j u)(∂k ϕ ) − (∂k u)(∂j ϕ ).

Observe from (2.2.225) that

κ−n.t.
u∂Ω belongs to L 1loc (∂Ω, σ), hence
 κ−n.t.    (2.2.237)
∂τ j k u∂Ω ∈ Lipc (∂Ω) by [69, Definition 11.2.1].

Also, from (2.2.216) and [69, Example 10.2.2] we see that

.
∂τ j k u is a well-defined distribution in H p (∂Ω, σ)
.   (2.2.238)
satisfying ∂τ j k u H p (∂Ω,σ) ≤ C Nκ (∇u) L p (∂Ω,σ),

for some constant C ∈ (0, ∞) independent of u.

Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal
to Ω. Bearing in mind the memberships from (2.2.237) and (2.2.238), we are now
prepared to compute (with all pointy brackets denoting distributional pairings on
∂Ω):
318 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
# ∫
 κ−n.t.  $  κ−n.t. 
∂τ j k u∂Ω , ψ = − u∂Ω ∂τ j k ψ dσ
∂Ω
∫
 κ−n.t. 
= − lim u∂Ω ∂τ j k ϕ dσ
→∞ ∂Ω
∫
 κ−n.t.   
= − lim u∂Ω ν j (∂k ϕ ) − νk (∂j ϕ ) dσ
→∞ ∂Ω
∫
 κ−n.t. 
= − lim ν · F ∂Ω dσ
→∞ ∂Ω
∫
= − lim div F dL n
→∞ Ω
∫
 
= − lim (∂j u)(∂k ϕ ) − (∂k u)(∂j ϕ ) dL n
→∞ Ω
∫
 
=− (∂j u)(∂k Ψ ) − (∂k u)(∂j Ψ ) dL n
Ω
. 
= ∂τ j k u , ψ . (2.2.239)

Above, the first equality comes from [69, Definition 11.2.1], while the second equality
is based on (2.2.232) and (2.2.225). The third equality in (2.2.239) is implied by
(A.0.183), and the fourth equality in (2.2.239) is a consequence of the property
recorded in the first line of (2.2.236). The fifth equality in (2.2.239) is a consequence
of [68, Proposition 2.8.17], whose applicability in the current setting is guaranteed
by the assumptions on Ω, together with (2.2.234) and the fact that (np)/(n − 1) > 1
(see (2.2.219)), as well as (2.2.235). The sixth equality in (2.2.239) is seen from
(2.2.236), while the seventh equality in (2.2.239) is deduced from (2.2.232) and
(2.2.229) (again, bearing in mind that (np)/(n − 1) > 1 thanks to (2.2.219)). The
final equality in (2.2.239) follows from (2.2.231) and [69, Example 10.2.2].
Having established (2.2.239), in view of the arbitrariness of the test function
ψ ∈ Lipc (∂Ω) we conclude that
 κ−n.t.  .  
∂τ j k u∂Ω = ∂τ j k u in Lipc (∂Ω) . (2.2.240)

From this and (2.2.238) we may now conclude that (2.2.217) holds. Finally, the fact
κ−n.t. .p
that u∂Ω belongs to the homogeneous Hardy-based Sobolev space H1 (∂Ω, σ) and
satisfies (2.2.218) then becomes a consequence of (2.2.217), (2.2.238), (2.2.237),
and [69, Definition 11.10.5] (cf. (A.0.89)-(A.0.90)). 
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 319

2.3 Integral Operators of Layer Potential Type on Hardy-Based

Sobolev Spaces and BMO−1

In the theorem below we shall augment Theorem 1.5.1 and Theorem 2.2.6, by
including results involving the scale of Hardy-based Sobolev spaces and BMO−1 .
Theorem 2.3.1 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Abbreviate
σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
αβ 
normal to Ω. For M ∈ N given, consider a coefficient tensor A = ar s 1≤r,s ≤n
1≤α,β ≤M
with complex entries, with the property that the M × M homogeneous second-order
system L = L A associated with A in Rn as in (1.3.2) is weakly elliptic (in the sense
of [70, (1.3.3) in Definition 1.3.1]). Finally, fix an aperture parameter κ ∈ (0, ∞)
along with an exponent 
n ,1 .
p ∈ n−1 (2.3.1)
Then the following properties hold.
(i) Recall the modified boundary-to-domain double layer potential operator Dmod
  M
acting on functions from the space L 1 ∂Ω, 1+σ(x)
|x | n as in (1.8.6). Then for
.p M
each f = ( fα )1≤α ≤M ∈ H1 (∂Ω, σ) , γ ∈ {1, . . . , M },  ∈ {1, . . . , n}, and
x ∈ Ω one has (using the summation convention)
#  $
⎧
⎪ −a
βα
(∂ )(x − ·)  , ∂τ fα if ∂Ω bounded,
  ⎪
⎨
⎪ r s r E γβ ∂Ω  s
∂ Dmod f γ (x) = # $
⎪
⎪ 
⎪ −arβα (∂ Eγβ )(x − ·)  , ∂τ fα if ∂Ω unbounded,
⎩ s r ∂Ω  s

(2.3.2)

where the pairings in (2.3.2) are understood in the sense of [69, Theorem 4.6.1]
(keeping in mind that each tangential derivative ∂τ s fα belongs to the Hardy
space H p (∂Ω, σ)). As a consequence of (2.3.2), [70, Theorem 1.4.2], and [70,
Corollary 2.5.4],
.p M
for each f ∈ H1 (∂Ω, σ) , the nontangential boundary trace
κ−n.t. (2.3.3)
(∇Dmod f )∂Ω exists (in C M ·n ) at σ-a.e. point on ∂Ω.

Also, there exists some finite constant C > 0, depending only on ∂Ω, A, n, κ,
.p M
and p, such that for each f ∈ H1 (∂Ω, σ) one has
 
Nκ (∇D f ) p
mod L (∂Ω,σ)
≤ C f [ H. p (∂Ω,σ)] M . (2.3.4)
1

(ii) The following operator is well defined, linear, and bounded:

. .p M M
∂νA Dmod : H1 (∂Ω, σ) −→ H p (∂Ω, σ) defined as
.
 A . . p M
(2.3.5)
∂ν Dmod ) f := ∂νA(Dmod f ) for each f ∈ H1 (∂Ω, σ) ,
320 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

where the weak conormal derivative is considered in the sense of [69, Defini-
tion 10.2.18]. In turn, the operator (2.3.5) induces a bounded linear mapping
. q, p M M
∂νA D : H1 (∂Ω, σ) −→ H p (∂Ω, σ) for each q ∈ [1, ∞). (2.3.6)
.  n−1
The operators ∂νA D thus considered for various values of p ∈ n , 1 and
q ∈ [1, ∞) are compatible with another. In addition,

having fixed integrability exponents p0, q0 ∈ (1, ∞) and q ∈ [1, ∞),

q, p M p ,q M
it follows that for each f ∈ H1 (∂Ω, σ) ∩ L1 0 0 (∂Ω, σ) the
M
function ∂νA D f ∈ L q0 (∂Ω, σ) with ∂νA D as in (1.5.31) induces (2.3.7)
.A M .
the same distribution on ∂Ω as ∂ν D f ∈ H p (∂Ω, σ) with ∂νA D
as in (2.3.6).
Moreover,
if Ω ⊆ Rn is an open set satisfying a two-sided local John condition whose
boundary is compact and Ahlfors regular and q ∈ (1, ∞), then the mapping
. M M M
→ Lip (∂Ω)
q, p
∂νA D : H1 (∂Ω, σ) → H p (∂Ω, σ) induced by
the operator (2.3.6) composed with the inclusion [69, (4.2.8)] is compatible
M M M
→ Lip (∂Ω) , the
q
with ∂νA D : L q (∂Ω, σ) → L−1 (∂Ω, σ)
mapping induced by the operator (1.5.33) composed with the inclusion
[69, (11.8.15)].
(2.3.8)
(iii) The boundary-to-boundary double layer potential operator K from (1.5.24)
extends to a linear and bounded mapping
q, p M q, p M
K : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) for each q ∈ (1, ∞). (2.3.9)

(iv) The boundary-to-boundary single layer potential operator from Theorem 2.2.6
induces a linear and bounded mapping
M .p M
S : H p (∂Ω, σ) −→ H1 (∂Ω, σ) . (2.3.10)
# is the operator in (2.1.42) corresponding to b := E (the matrix-
Also, if Tjk
valued fundamental solution associated with L as in [70, Theorem 1.4.2]) then
for each j, k ∈ {1, . . . , n} one has
M
∂τ j k (S f ) = Tjk
#
f for each f ∈ H p (∂Ω, σ) . (2.3.11)

Finally, the operator S from (2.2.116) induces a linear and bounded mapping
M p ∗, p M
S : H p (∂Ω, σ) −→ H1 (∂Ω, σ)
 
1 −1
 (2.3.12)
where p∗ := p1 − n−1 n−2 .
∈ 1, n−1
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 321

(v) Assume n ≥ 3. Then the (real) transpose of the operator (2.3.12) corresponding
to p := 1, i.e.,
M n−1
,1 M
S : H 1 (∂Ω, σ) −→ H1n−2 (∂Ω, σ) (2.3.13)

is the linear and bounded mapping

⎧
⎪ M
 M ⎨ BMO(∂Ω, σ)
⎪ if ∂Ω is bounded,
S : BMO−1 (∂Ω, σ) −→
⎪ BMO(∂Ω, σ)  ∼
⎪ M
if ∂Ω is unbounded.
⎩
(2.3.14)

Also, the (real) transpose of the operator (2.3.9) corresponding to p := 1 and

q := n−1
n−2 , i.e.,

n−1
,1 M n−1
,1 M
K : H1n−2 (∂Ω, σ) −→ H1n−2 (∂Ω, σ) , (2.3.15)

is the linear and bounded mapping

K  : BMO−1 (∂Ω, σ)
M M
−→ BMO−1 (∂Ω, σ) . (2.3.16)

This mapping is compatible with K # , in the sense that

M
K  f = K # f as functionals in BMO−1 (∂Ω, σ)
(2.3.17)
M M
for each f ∈ L n−1 (∂Ω, σ) → BMO−1 (∂Ω, σ) .
M
In addition, VMO−1 (∂Ω, σ) is an invariant subspace of K  in (2.3.16),
hence

K  : VMO−1 (∂Ω, σ)
M M
−→ VMO−1 (∂Ω, σ) (2.3.18)

is a well-defined, linear, and bounded mapping.

(vi) Continue to assume n ≥ 3 and, this time, also make the additional assumption
that ∂Ω is compact. Then the boundary-to-domain single layer potential operator
associated with L and Ω induces a well-defined linear mapping in the context

−→ 𝒞∞ (Ω)
M M
𝒮 : BMO−1 (∂Ω, σ) (2.3.19)
M
if for each f = ( fβ )1≤β ≤M ∈ BMO−1 (∂Ω, σ) one sets
  
𝒮 f (x) := Eαβ (x − ·)∂Ω, fβ for all x ∈ Ω. (2.3.20)
1≤α ≤M

In formula (2.3.20) above, E = (Eαβ )1≤α,β ≤M is the matrix-valued funda-

mental solution associated with the system L as in [70, Theorem 1.4.2], and
322 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
n−1
,1
·, · stands for the duality bracket between the space H1n−2 (∂Ω, σ) and the
 n−1 ,1 ∗
space BMO−1 (∂Ω, σ) = H1n−2 (∂Ω, σ) (cf. [69, Definition 11.10.9]; see also
M
(A.0.32)). In addition, for each f ∈ BMO−1 (∂Ω, σ) one has
 
L A 𝒮 f = 0 in Ω, (2.3.21)

and for each p ∈ (1, ∞) there exists some constant C = C(Ω, L, p) ∈ (0, ∞) such
that
|∇𝒮 f | p dist(·, ∂Ω) p−1 dL n is a Carleson measure in
p
the set Ω with constant ≤ C f [BMO (∂Ω,σ)] M . (2.3.22)
−1

In particular, corresponding to p = 2, it follows that7

|∇𝒮 f | 2 dist(·, ∂Ω) dL n is a Carleson measure in Ω with constant

M (2.3.23)
≤ C f [BMO2
(∂Ω,σ)] M
, for each f ∈ BMO−1 (∂Ω, σ) .
−1

M
Also, for each given f ∈ BMO−1 (∂Ω, σ) , the boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
 κ−n.t.    (2.3.24)
𝒮 f ∂Ω (x) = (SL  ) f (x) at σ-a.e. point x ∈ ∂Ω,

where SL  is the boundary-to-boundary single layer potential operator associ-

M
ated with L  , considered (cf. (2.3.13)) as a mapping from H 1 (∂Ω, σ) into
n−1
,1 M
H1n−2 (∂Ω, σ) , and (SL  ) stands for its (real) transpose (cf. (2.3.14)).
M
Finally, for each f ∈ BMO−1 (∂Ω, σ) one has
 ∫ 1
1    p
lim+ sup   ∇ 𝒮 f  dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , [VMO−1 (∂Ω, σ)] M (2.3.25)

where the distance in the right-hand side is considered in [BMO−1 (∂Ω, σ)] M .
As a consequence,

|∇𝒮 f | p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure

M (2.3.26)
in the set Ω for each f ∈ VMO−1 (∂Ω, σ) and p ∈ (1, ∞),

hence, corresponding to p = 2,

|∇𝒮 f | 2 dist(·, ∂Ω) dL n is a vanishing Carleson

M (2.3.27)
measure in Ω for each f ∈ VMO−1 (∂Ω, σ) .
 2
7 it is natural to think of  ∇𝒮 f  dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated with
f via the operator 𝒮
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 323

(vii) Strengthen the original hypotheses by assuming that n ≥ 3 and Ω is an open set
satisfying a two-sided local John condition and whose boundary is compact and
Ahlfors regular. Then S  from (2.3.14) is compatible with SL  , the single layer
potential operator associated with L  and Ω as in (1.3.62), in the sense that

S  f = SL  f at σ-a.e. point on ∂Ω, for each

M M
(2.3.28)
f ∈ L n−1 (∂Ω, σ) → BMO−1 (∂Ω, σ) .
M
Moreover, the restriction of the operator (2.3.14) to VMO−1 (∂Ω, σ) induces
a well-defined, linear, and bounded mapping

S  : VMO−1 (∂Ω, σ)
M M
−→ VMO(∂Ω, σ) . (2.3.29)

Parenthetically, we wish to note that (2.2.139), [69, (4.2.17)], and (2.3.10) offer an
alternative proof of the result recorded in [69, (11.10.66)], corresponding to (special,
matrix-valued case) b := E and n ≥ 3.
.p M
Proof of Theorem 2.3.1 Fix some f = ( fα )1≤α ≤M ∈ H1 (∂Ω, σ) . Also, pick
some γ ∈ {1, . . . , M },  ∈ {1, . . . , n}, and select an arbitrary point x ∈ Ω. Since
  M
f ∈ L 1 ∂Ω, 1+σ(y)
|y | n , it follows that Dmod f is meaningfully defined, and we may
use (1.8.11) together with the first two lines in (1.3.31) to write (using the summation
convention over repeated indices)
∫
  βα
∂ Dmod f γ (x) = − ar s ∂τ s (y) [(∂r Eγβ )(x − y)] fα (y) dσ(y). (2.3.30)
∂Ω
   
To proceed, choose ε ∈ 0, dist(x, ∂Ω) , set Uε := y ∈ Rn : dist(y, ∂Ω) < ε
and fix an index β ∈ {1, . . . , M } along with r ∈ {1, . . . , n}. Then the function
ϕ(y) := (∂r Eγβ )(x − y) for each y ∈ Uε , satisfies

ϕ ∈ 𝒞1 (Uε ) and there exists C ∈ (0, ∞) such that

C C (2.3.31)
|ϕ(y)| ≤ and |(∇ϕ)(y)| ≤ for all y ∈ Uε .
1 + |y| n−1 1 + |y| n

Then (2.3.31) ensures the applicability of [69, Lemma 11.10.4] for the current choice
of ϕ which, for each s ∈ {1, . . . , n}, allows us to write
∫
∂τ s (y) [(∂r Eγβ )(x − y)] fα (y) dσ(y)
∂Ω
#  $
⎧
⎪ − (∂ )(x − ·)  , ∂ if ∂Ω bounded,
⎪
⎨
⎪ E
r γβ f
∂Ω τ s α
= # $ (2.3.32)
⎪
⎪ 
⎪ − (∂r Eγβ )(x − ·) , ∂τ s fα if ∂Ω unbounded,
⎩ ∂Ω
324 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

where the pairings in (2.3.32) are understood in the sense of [69, Theorem 4.6.1]
(keeping in mind that ∂τ s fα ∈ H p (∂Ω, σ)). Collectively, (2.3.30) and (2.3.32)
prove (2.3.2). Since the pairings in the right-hand side of (2.3.2) are of the form [70,
(2.4.16)] (corresponding to the kernel k := ∂r Eγβ ) and since ∂τ s fα ∈ H p (∂Ω, σ),
we may apply [70, (2.4.14)] in [70, Theorem 2.4.1] to conclude that

  M n
Nκ (∂ D f ) p ≤C ∂τ s fα H p (∂Ω,σ) (2.3.33)
mod L (∂Ω,σ)
α=1 s=1

for some finite C > 0 independent of f . This establishes (2.3.4) and finishes the
treatment of the claims made in item (i).
Moving on to item (ii), the claims about the operator (2.3.5) are clear from
(A.0.89), (1.3.24), (2.3.4), and [69, Theorem 10.2.24]. Collectively, these imply that
.p M M
for each f ∈ H1 (∂Ω, σ) , the function u := D f belongs to 𝒞∞ (Ω) , and
satisfies L A.u = 0 in Ω as well as Nκ (∇u) ∈ L (∂Ω, σ). As such, the weak conormal
p

derivative ∂νAu, considered in the sense of [69, Definition 10.2.18], satisfies

. .   
and ∂νAu[H p (∂Ω,σ)] M ≤ C Nκ (∇u) L p (∂Ω,σ)
M
∂νAu ∈ H p (∂Ω, σ) (2.3.34)

for some constant C = C(Ω, A, κ, p) ∈ (0, ∞). The estimate in (2.3.34) in combination
with (2.3.4) then yields
 .A 
∂ (D f ) p ≤ C f [ H. p (∂Ω,σ)] M , (2.3.35)
ν mod [H (∂Ω,σ)] M 1

as desired.
Next, the fact that the operator (2.3.6) is well defined, linear, and bounded is seen
from what we have proved so far and (1.8.8). Also, from (A.0.181)-(A.0.182) and
q, p M
(A.0.165)-(A.0.166) we see that for each f ∈ H1 (∂Ω, σ) and Φ ∈ Lipc (Rn )
we have
#. ∫
 $  
A 
[Lip c (∂Ω) ] M ∂ν D f , Φ ∂Ω [Lip c (∂Ω)] M = A∇(D f ), ∇Φ dL n . (2.3.36)
Ω
.
From this it is clear that the mappings induced by ∂νA D in the context of (2.3.6)
corresponding to various values of p ∈ n−1 n , 1 and q ∈ [1, ∞) are compatible with
another.
Let us now check the compatibility claim made in (2.3.7). With this goal in
q, p M p ,q M
mind, fix an arbitrary function f ∈ H1 (∂Ω, σ) ∩ L1 0 0 (∂Ω, σ) for some
p0, q0 ∈ (1, ∞). Then from [69, Definitions 11.1.2 and 11.10.6] it follows that ∂τ j k f
M M
belongs to H p (∂Ω, σ) ∩ L q0 (∂Ω, σ) for each indices j, k ∈ {1, . . . , n}.
Since L 0 (∂Ω, σ) = H 0 (∂Ω, σ) (recall that q0 > 1), [69, Proposition 4.2.2] implies
q q
M q,1 M
∂τ j k f ∈ H 1 (∂Ω, σ) for each j, k ∈ {1, . . . , n}. Hence f ∈ H1 (∂Ω, σ) ,
.A
Since the mappings induced by ∂ν D in the context of (2.3.6) are compatible with
.
one another, henceforth we shall consider ∂νA D f regarding f as a function in
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 325

q,1 M
H1 (∂Ω, σ) . Then item (i) in the current theorem (applied with p = 1) together
with (1.8.8) give
Nκ (∇D f ) ∈ L 1 (∂Ω, σ). (2.3.37)
M
Moreover, [69, (4.2.10)] guarantees that ∂τ j k f belongs to L 1 (∂Ω, σ) for each
q,1 M
j, k ∈ {1, . . . , n}, thus f ∈ L1 (∂Ω, σ) .
As such, we may apply item (ii) in
Theorem 1.5.1 (with p := q, q := 1, and keeping in mind that H n−1 (∂Ω \ ∂∗ Ω) = 0
in our setting; cf. the definition of a UR domain from [68, Definition 5.10.6]) to
obtain that   κ−n.t.
∇D f ∂Ω exists at σ-a.e. point on ∂Ω. (2.3.38)
Having established (2.3.37) and (2.3.38), we may invoke item (3) in [69, Theo-
.
rem 10.2.24] to conclude that ∂νA(D f ), considered in the sense of [69, Defini-
n·M
tion 10.2.18], actually belongs to L 1 (∂Ω, σ) and has the pointwise formula
  κ−n.t. 
. αβ 
∂νA(D f ) = νr ar s ∂s uβ ∂Ω at σ-a.e. point on ∂Ω. (2.3.39)
1≤α ≤M
The desired conclusion now follows upon observing that the right-hand side of
(2.3.39) is actually (∂νA D) f , with the operator ∂νA D as in (1.5.31) (see item (vii) in
Theorem 1.5.1).
Let us now justify the compatibility claim made in (2.3.8). To this end, work
under the assumption that Ω ⊆ Rn is an open set satisfying a two-sided local John
condition and whose boundary is compact and Ahlfors regular. Also, fix q ∈ (1, ∞)
q, p M M
and pick f ∈ H1 (∂Ω, σ) → L q (∂Ω, σ) arbitrary. The goal is to show that
.
∂νA D f = ∂νA D f in Lip (∂Ω)
M
, (2.3.40)
. M M
where we have considered ∂νA D f ∈ H p (∂Ω, σ) → Lip (∂Ω) via (2.3.6)
M M
Lip (∂Ω)
q
and [69, (4.2.8)], while ∂νA D f ∈ L−1 (∂Ω, σ) → via (1.5.33) and
M
[69, (11.8.15)]. Pick an arbitrary vector-valued test function φ ∈ Lip (∂Ω) , with
the aim of proving that
 .A   
[Lip (∂Ω) ] M ∂ν D f , φ [Lip (∂Ω)] M = [Lip (∂Ω) ] M ∂νA D f , φ [Lip (∂Ω)] M . (2.3.41)
 n−1 
To this end, we distinguish two cases. First, suppose p ∈ n , 1 and define
1 
η := (n − 1) p − 1 ∈ (0, 1). (2.3.42)

M
Bring in the sequence {ϕ } ∈N ⊆ 𝒞∞
c (R )
n associated with the scalar components
M
of the function f := φ ∈ L1∞ (∂Ω, σ) as in [69, Remark 11.1.13]. Then [69,
(11.1.92)-(11.1.94)] guarantee that

there exists a compact set K ⊂ Rn such that supp ϕ ⊆ K for all  ∈ N, (2.3.43)

and the functions ψ := ϕ ∂Ω , for each  ∈ N, satisfy the following properties:
326 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
%  &
lim sup ψ (x) − φ(x) = 0, (2.3.44)
→∞ x ∈∂Ω

sup ψ [𝒞α (∂Ω)] M < +∞ for each α ∈ (0, 1), (2.3.45)

∈N

and
M
∂τ j k ψ −→ ∂τ j k φ weak-∗ in L ∞ (∂Ω, σ) as  → ∞,
(2.3.46)
for each pair of indices j, k ∈ {1, . . . , n}.
As a consequence of (2.3.44), (2.3.45), and [69, (4.8.29) in Lemma 4.8.4] we also
have
lim ψ = φ weak-∗ in 𝒞η (∂Ω) ,
M
(2.3.47)
→∞

whereas (2.3.44) implies

lim ψ = ϕ in L ∞ (∂Ω, σ)
M M
, hence also in BMO(∂Ω, σ) . (2.3.48)
→∞

Finally, as a consequence of (2.3.43), (2.3.44), (2.3.46), and [69, Lemma 11.8.3] we

have
q M
lim ψ = φ weak-∗ in L1 (∂Ω, σ) . (2.3.49)
→∞

To get going, write

 .A  . 
[Lip (∂Ω) ] M ∂ν D f , φ [Lip (∂Ω)] M = [H p (∂Ω,σ)] M ∂νA D f , φ [𝒞η (∂Ω)] M
 .A 
= lim [H p (∂Ω,σ)] M ∂ν D f , ψ [𝒞η (∂Ω)] M
→∞
 .A 
= lim [Lip (∂Ω) ] M ∂ν D f , ψ [Lip (∂Ω)] M , (2.3.50)
→∞

with the first and last equalities coming from [69, (4.6.31)] (bearing in mind (2.3.6)),
and the second equality implied by (2.3.47).
Let us now fix  ∈ N. From (A.0.181)-(A.0.182) and (A.0.165)-(A.0.166) we see
that
∫
 .A   
[Lip(∂Ω) ] M ∂ν D f , ψ [Lip c (∂Ω)] M = A∇(D f ), ∇ϕ dL n . (2.3.51)
Ω

Also,
∫ ∫
   
A∇(D f ), ∇ϕ dL =
n
A ∇ϕ , ∇(D f ) dL n (2.3.52)
Ω Ω
∫ ∫
  1    
= ∂νA ϕ , 2 I + K f dσ − L  ϕ , D f dL n
∂Ω Ω
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 327

by the Green type formula [70, (1.7.121)], used with A in place of A, with u := ϕ Ω
q, p M
and w := D f . The fact that f ∈ H1 (∂Ω, σ) implies, in view of (1.5.1), (2.3.4),
and (1.8.8),

Nκ (D f ) ∈ L q (∂Ω, σ) and Nκ (∇D f ) ∈ L p (∂Ω, σ). (2.3.53)

In concert with [68, (8.6.51)] and the fact that p > n−1
n , the latter membership further
entails (with K as in (2.3.43))
nM nM
∇D f ∈ L np/(n−1) (Ω ∩ K, L n ) → L 1 (Ω ∩ K, L n ) . (2.3.54)
κ−n.t.  
Also, from (1.5.20) we know that for our choice of w we have w ∂Ω = 12 I + K f at
σ-a.e. point on ∂Ω, for any aperture parameter κ ∈ (0, ∞). These properties ensure
that the hypotheses of [70, Theorem 1.7.12] are presently satisfied, so (2.3.52) is
justified. Combining (2.3.51)-(2.3.52) leads to the conclusion that
∫
 .A   A   

[Lip(∂Ω) ] M ∂ ν D f , ψ [Lip c (∂Ω)] M = ∂ν ϕ , 12 I + K f dσ
∂Ω
∫
 
− L  ϕ , D f dL n . (2.3.55)
Ω

Going further, continue to keep  ∈ N arbitrary and fixed. We make the claim that
M
for each function g ∈ L q (∂Ω, σ) we have
∫ ∫
 A      
∂ν ϕ , 12 I + K g dσ − L ϕ , Dg dL n .
∂Ω Ω
∫
  
= g, ∂νA D A ψ dσ. (2.3.56)
∂Ω

The justification of this claim rests on two observations. The first observation we
make is that both sides of (2.3.56) depend linearly and continuously on the function
M
g ∈ L q (∂Ω, σ) . Indeed, this is apparent from definitions, the continuity of K on
M
the space L q (∂Ω, σ) (cf. item (iii) in Theorem 1.5.1), the fact that (1.5.1) and
[68, (8.6.51)] imply the continuity of
M M
D : L q (∂Ω, σ) −→ L 1 (Ω ∩ K, L n ) , (2.3.57)
  M   −1
and the fact that ∂νA D A ψ ∈ L q (∂Ω, σ) where q  := 1 − q1 ∈ (1, ∞) (cf.
[69, (11.1.81)] and (1.5.31)).
The second observation we wish to make in relation to (2.3.56) is that said
q M
formula is true whenever g ∈ L1 (∂Ω, σ) . To see this, note that since we have the
q M q, p M
inclusion L1 (∂Ω, σ) ⊆ H1 (∂Ω, σ) (as is apparent from [68, (3.6.27)], [69,
(4.2.21)], and definitions), the same argument that led to formula (2.3.55) presently
328 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

q M
shows that, for each g ∈ L1 (∂Ω, σ) , we have
∫ ∫
 A 1     
∂ν ϕ , 2 I + K g dσ − L ϕ , Dg dL n
∂Ω Ω
. 
= [Lip(∂Ω) ] M ∂νA Dg, ψ [Lipc (∂Ω)] M . (2.3.58)

q M .
Knowing that g ∈ L1 (∂Ω, σ) allows us to replace ∂νA Dg by ∂νA Dg (see the
compatibility result in (2.3.7), here used with p0 = q0 = q). Hence,
∫ ∫
 A      
∂ν ϕ , 12 I + K g dσ − L ϕ , Dg dL n
∂Ω Ω
∫
   
= [Lip(∂Ω) ] M ∂νA Dg, ψ [Lipc (∂Ω)] M = ∂νA Dg, ψ dσ
∂Ω
∫
  
= g, ∂νA D A ψ dσ, (2.3.59)
∂Ω

where the second equality above is a consequence of [68, (4.1.47)], and the fi-
nal equality in (2.3.59) is provided by (1.5.32) (bearing in mind that the function
q M
ψ ∈ L1 (∂Ω, σ) ; cf. [69, (11.1.81)]). Formula (2.3.59) justifies (2.3.56) in the
q M
case when g belongs to the space L1 (∂Ω, σ) . Recall from [69, (11.1.66)] that
q M M
L1 (∂Ω, σ) → L q (∂Ω, σ)densely. In view of this and the fact that, as al-
ready noted, both sides of (2.3.56) depend linearly and continuously on the function
M
g ∈ L q (∂Ω, σ) , we may ultimately conclude that (2.3.56) holds for each function
M
g ∈ L q (∂Ω, σ) .
M
Having established (2.3.56) for arbitrary functions g ∈ L q (∂Ω, σ) , write
(2.3.56) for g := f then combine the resulting formula with (2.3.55) to deduce that,
for each fixed  ∈ N, we have
∫
 .A    

[Lip(∂Ω) ] M ∂ ν D f , ψ [Lip c (∂Ω)] M = f , ∂νA D A ψ dσ. (2.3.60)
∂Ω

If we now pass to limit, as  → ∞, and also bring into the mix (2.3.50), we arrive at
the conclusion that, on the one hand,
∫
 .A    

[Lip (∂Ω) ] M ∂ ν D f , φ [Lip (∂Ω)] M = lim f , ∂νA D A ψ dσ. (2.3.61)
→∞ ∂Ω

On the other hand, for each fixed  ∈ N, we may rely on (1.5.33)-(1.5.34) to

M
write (viewing f simply as a function in L q (∂Ω, σ) , and ψ as a function in
q M
L1 (∂Ω, σ) )
∫
    
f , ∂νA D A ψ dσ = [L q (∂Ω,σ)] M ∂νA D f , ψ [L q (∂Ω,σ)] M (2.3.62)
−1
∂Ω 1
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 329

where ∂νA D acts as in (1.5.33) (with p := q). Together, (2.3.61), (2.3.62), (2.3.49),
and [69, (11.8.16)] then permit us to write
 .A   
[Lip (∂Ω) ] M ∂ν D f , φ [Lip (∂Ω)] M = lim q
[L−1 (∂Ω,σ)] M ∂νA D f , ψ q
[L1 (∂Ω,σ)] M
→∞
 
= [L q (∂Ω,σ)] M ∂νA D f , φ [L q (∂Ω,σ)] M
−1 1
 
= [Lip (∂Ω) ] M ∂νA D f, φ [Lip (∂Ω)] M . (2.3.63)

This finishes the proof of (2.3.41), hence (2.3.40) is established in the case when
p < 1. Finally, the case when p = 1 follows along very similar lines, now using
the H 1 -BMO duality in place of the H p -𝒞η duality (cf. [69, Lemma 4.6.4]), and
(2.3.48). Hence, (2.3.8) is justified for any p as in (2.3.1). This completes the proof
of item (ii).
 q, p M
Next, assume p ∈ n−1 n , 1 and q ∈ (1, ∞) and take f ∈ H1 (∂Ω, σ) . By the
current item (i) and (1.8.9) we have

Nκ ∇D f ) ∈ L p (∂Ω, σ) and
  (2.3.64)
Nκ (∇D f ) p ≤ C f [ H. p (∂Ω,σ)] M .
L (∂Ω,σ) 1

Moreover,

Nκ (D f ) ∈ L q (∂Ω, σ) → Lloc
1
(∂Ω, σ) and (2.3.65)
κ−n.t.

(D f ) = ( 12 I + K) f at σ-a.e. point on ∂Ω, (2.3.66)
∂Ω

thanks to item (i) in Theorem 1.5.1 (with p := q), and item (iv) in Theorem 1.5.1
(bearing in mind that H n−1 (∂Ω \ ∂∗ Ω) = 0 in the current setting). Given that
D f ∈ [𝒞∞ (Ω)] M , the properties listed in (2.3.64)-(2.3.66) ensure the applicability
of [69, Theorem 11.10.10] which presently yields

∂τ j k (K f ) ∈ H p (∂Ω, σ) for each j, k ∈ {1, . . . , n}, (2.3.67)

and
n
 
∂τ (K f ) p ≤C f .p . (2.3.68)
jk H (∂Ω,σ) [ H1 (∂Ω,σ)] M
j,k=1
M
Since from [70, (2.3.18)] we also know that K maps L q (∂Ω, σ) boundedly into
itself, based on (2.3.67)-(2.3.68), (2.3.4), and [69, Definition 11.10.6] we ultimately
conclude that
q, p M
K f ∈ H1 (∂Ω, σ) and
(2.3.69)
Kf [H1
q, p
(∂Ω,σ)] M ≤C f [H1
q, p
(∂Ω,σ)] M .
330 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Hence, the claim made in item (iii) follows.

 
1 −1
Turning to the claim made in item (iv), define p∗ := p1 − n−1 ∈ (1, ∞] and fix
j, k ∈ {1, . . . , n} arbitrary. From Theorem 2.2.6 we know that
M ∗ M
S : H p (∂Ω, σ) −→ L p (∂Ω, σ) (2.3.70)

is well defined, linear, and bounded. In concert with [69, (11.2.6)] this implies that
M   M
∂τ j k S : H p (∂Ω, σ) −→ Lipc (∂Ω) continuously. (2.3.71)

Since from Theorem 2.1.4 (cf. (2.1.42)) and [69, (4.2.8)] we also have
M   M
#
Tjk : H p (∂Ω, σ) −→ Lipc (∂Ω) continuously, (2.3.72)

in order to conclude that (2.3.11) holds it suffices to show (cf. [69, Theorem 4.4.1])
M
that for each (p, ∞)-atom a for H p (∂Ω, σ) we have
  M
∂τ j k (Sa) = Tjk
#
a in Lipc (∂Ω) . (2.3.73)

M
To this end, consider an arbitrary test function φ ∈ Lipc (∂Ω) . With S A being
the single layer potential associated with the coefficient tensor A , the real transpose
of A, in the same manner S has been associated with the original A, we may then
write
 
[(Lip c (∂Ω)) ] M ∂τ j k (Sa), φ [Lip (∂Ω)] M
c
 
= −[(Lipc (∂Ω)) ] M Sa, ∂τ j k φ [Lip
c (∂Ω)]
M

∫ ∫
=− Sa, ∂τ j k φ dσ = − a, S A (∂τ j k φ) dσ
∂Ω ∂Ω
∫ ∫
=− a, Tjk φ dσ = Tjk
#
a, φ dσ
∂Ω ∂Ω
 # 
= [(Lipc (∂Ω)) ] M Tjk a, φ [Lip (2.3.74)
c (∂Ω)]
M

thanks to (2.3.70), [69, Definition 11.2.1], Fubini’s Theorem, a slight variant of

(1.2.29), item (ii) in Proposition 1.2.1, and [68, Proposition 4.1.4]. This establishes
(2.3.73), finishing the proof of (2.3.11).
In turn, from (2.3.70), the fact that
.  σ(x) 
L q (∂Ω, σ) ⊆ L 1 ∂Ω, , (2.3.75)
1≤q ≤∞
1 + |x| n

(2.3.11), (2.1.42), and [69, Definition 11.10.5] (cf. (A.0.89)-(A.0.90)) we conclude

that the boundary-to-boundary single layer induces a well-defined, linear, and bound-
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 331

ed mapping in the context of (2.3.10). Finally, the claim about (2.3.12) follows from
what we have just proved, (2.3.70), and [69, Definition 11.10.6]. This finishes the
proof of item (iv).
That S  , the (real) transpose of (2.3.12) corresponding to p = 1, is a well-defined,
linear, and bounded mapping in the context of (2.3.14) follows from [69, (1.2.67)], the
duality result recorded in [69, Theorem 4.6.1], and (A.0.32). Likewise, that (2.3.16)
is a well-defined linear and bounded operator follows from (2.3.9) (used with p := 1
and q := n−1 
n−2 ) and (A.0.32). To show that the operator K , considered in the context
M
of (2.3.16), is compatible with K # acting on the space L n−1 (∂Ω, σ) , pick an
M M
arbitrary function f ∈ L n−1 (∂Ω, σ) . Hence also f ∈ BMO−1 (∂Ω, σ) (cf.
n−2 ,1
n−1
M
[69, (11.10.52)]) and for each g ∈ H1 (∂Ω, σ) we may write
 
[BMO−1 (∂Ω,σ)] M K f , g n−1 ,1
[H1n−2 (∂Ω,σ)] M
 
= [BMO−1 (∂Ω,σ)] M f , Kg n−1 ,1
[H1n−2 (∂Ω,σ)] M
 
= [L n−1 (∂Ω,σ)] M f , Kg n−1
[L n−2 (∂Ω,σ)] M
 
= [L n−1 (∂Ω,σ)] M K # f , g n−1
[L n−2 (∂Ω,σ)] M
 
= [BMO−1 (∂Ω,σ)] M K # f , g n−1 ,1 . (2.3.76)
[H1n−2 (∂Ω,σ)] M

The first equality above is implied by the fact that K  is the transpose of (2.3.15).
The second equality in (2.3.76) comes from the compatibility property recorded
in [69, (11.10.55)]. The third equality in (2.3.76) is a consequence of item (iii) in
Theorem 1.5.1, while the fourth equality in (2.3.76) is once again seen from [69,
(11.10.55)]. With (2.3.76) in hand, (2.3.17) follows.
M
Next, that K  from (2.3.16) has VMO−1 (∂Ω, σ) as an invariant sub-
space is seen from the boundedness of (2.3.16), (2.3.17), the fact that K # maps
M
L n−1 (∂Ω, σ) into itself (cf. item (iii) in Theorem 1.5.1), (A.0.207), and [69,
Lemma 1.2.20].
Consider next the task of addressing the claims made in item (vi). Throughout,
work under the assumption that n ≥ 3 and that ∂Ω is bounded. To get started,
M
fix some arbitrary functional f = ( fβ )1≤β ≤M ∈ BMO−1 (∂Ω, σ) . From [69,
Proposition 11.10.7] and (A.0.32) we know that for each β ∈ {1, . . . , M } there exist
(β) (β)
f0 ∈ L n−1 (∂Ω, σ) and f jk ∈ BMO(∂Ω, σ) with 1 ≤ j < k ≤ n satisfying (for
some constant C ∈ (0, ∞) independent of f )
(β) (β)
f0 L n−1 (∂Ω,σ) + f jk BMO(∂Ω,σ) ≤ C fβ BMO−1 (∂Ω,σ) (2.3.77)
1≤ j<k ≤n

n−1
,1
and, for every function g ∈ H1n−2 (∂Ω, σ),
332 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 
BMO−1 (∂Ω,σ) fβ, g n−1 ,1 (2.3.78)
H1n−2 (∂Ω,σ)
∫
(β)  (β) 
= f0 g dσ + BMO(∂Ω,σ) f jk , ∂τ j k g .
∂Ω H 1 (∂Ω,σ)
1≤ j<k ≤n

Having picked x ∈ Ω and α ∈ {1, . . . , M }, we may then use (2.3.78) to write

  #  $
𝒮 f (x) = BMO (∂Ω,σ) fβ, Eαβ (x − ·)
α −1 n−1 ,1
∂Ω
H1n−2 (∂Ω,σ)
∫
(β)
= Eαβ (x − y) f0 (y) dσ(y)
∂Ω
# $
(β)
+ BMO(∂Ω,σ) f jk , ∂τ j k Eαβ (x − ·)
H 1 (∂Ω,σ)
1≤ j<k ≤n
∫
(β)
= Eαβ (x − y) f0 (y) dσ(y)
∂Ω
∫
(β)
+ ∂τ j k (y) Eαβ (x − y) f jk (y) dσ(y), (2.3.79)
1≤ j<k ≤n ∂Ω

where the last equality is based on [69, Lemma 4.6.5] (whose applicability in the
present setting is guaranteed by [68, (7.4.106)] and [70, Theorem 1.4.2]). Going
further, for any two pairs of indices, j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M },
introduce the integral operators acting on each function φ ∈ L 1 (∂Ω σ) according to
∫
αβ  
Tjk φ(x) := ν j (y)(∂k Eαβ )(x − y) − νk (y)(∂j Eαβ )(x − y) φ(y) dσ(y)
∂Ω
(2.3.80)

at each x ∈ Ω, where ν = (ν1, . . . , νn ) is the geometric measure theoretic outward

(β)
unit normal to Ω. In this notation, and with f0 := ( f0 )1≤β ≤M , the formula recorded
in (2.3.79) may be recast (bearing in mind [68, (7.4.106)]) simply as
 αβ (β) 
𝒮 f = 𝒮 f0 − Tjk f jk 1≤α ≤M in Ω. (2.3.81)
1≤ j<k ≤n

It is then clear from (2.3.81) that (2.3.19) is a well-defined linear mapping, and that
(2.3.21) holds. In addition, from (2.3.81), (1.5.60), (1.5.56), [70, Corollary 2.4.2],
and (2.3.77) we see that (2.3.22) is true. From (2.3.81), (1.5.44), (1.2.4), and [68,
κ−n.t.
(7.4.106)] we also conclude that 𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact, at
σ-a.e. point on ∂Ω we have
κ−n.t.  αβ (β) 
𝒮 f ∂Ω = S f0 − Tjk f jk 1≤α ≤M (2.3.82)
1≤ j<k ≤n
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 333

where, for each j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M },

αβ
Tjk φ(x) (2.3.83)
∫
 
:= lim+ ν j (y)(∂k Eαβ )(x − y) − νk (y)(∂j Eαβ )(x − y) φ(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε

for every function φ ∈ L 1 (∂Ω, σ) and σ-a.e. point x ∈ ∂Ω. Having established this,
M
for each test function ψ = (ψα )1≤α ≤M ∈ Lip(∂Ω) we may then write, thanks to
(1.3.62) and [70, (2.3.25)],
∫ #
κ−n.t. $
𝒮 f ∂Ω , ψ dσ
∂Ω
∫ ∫
   αβ (β) 
= S f0, ψ dσ − Tjk f jk ψα dσ
∂Ω 1≤ j<k ≤n ∂Ω

∫ ∫
  (β)   αβ  # 
= f0, SL  ψ dσ + f jk Tjk ψα dσ (2.3.84)
∂Ω 1≤ j<k ≤n ∂Ω

αβ
where for each j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M } we have denoted by Tjk the
integral operator acting on each function ϕ ∈ L 1 (∂Ω, σ) according to
 αβ  #
Tjk ϕ(x) (2.3.85)
∫
 
:= lim+ ν j (x)(∂k Eαβ )(x − y) − νk (x)(∂j Eαβ )(x − y) ϕ(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. In relation to this operator we wish to observe that, as seen
from (1.5.83) and the first formula in [70, (1.4.32)], for each j, k ∈ {1, . . . , n} and
each β ∈ {1, . . . , M } we have
   αβ  #
∂τ j k SL  ψ β = Tjk ψα at σ-a.e. point on ∂Ω. (2.3.86)

Together, (2.3.84) and (2.3.86) prove that

334 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
∫ # ∫
κ−n.t. $  
𝒮 f ∂Ω , ψ dσ = f0, SL  ψ dσ
∂Ω ∂Ω
∫
(β)  
+ f jk ∂τ j k SL  ψ β dσ
1≤ j<k ≤n ∂Ω

∫
(β)  
= f0 SL  ψ β dσ
∂Ω
#   $
(β)
+ BMO(∂Ω,σ) f jk , ∂τ j k SL  ψ β
H 1 (∂Ω,σ)
1≤ j<k ≤n
   
= BMO−1 (∂Ω,σ) fβ, SL  ψ β n−1 ,1
H1n−2 (∂Ω,σ)
 
= [BMO−1 (∂Ω,σ)] M f , SL  ψ n−1 ,1
[H1n−2 (∂Ω,σ)] M
 
= [BMO(∂Ω,σ)] M (SL  ) f , ψ [H 1 (∂Ω,σ)] M
∫ # $
= (SL  ) f , ψ dσ, (2.3.87)
∂Ω

where the second equality uses [69, Lemma 4.6.5] (whose applicability in the present
setting is ensured by [68, (7.4.106)] and [70, Theorem 1.4.2]). In view of [68,
κ−n.t.
(3.7.23)], we may now conclude from (2.3.87) that 𝒮 f ∂Ω = (SL  ) f at σ-a.e.
point on ∂Ω. This finishes the proof of (2.3.24).
To complete the treatment of the claims in item (vi) there remains to justify
M
(2.3.25). To this end, pick p ∈ (1, ∞) and select an arbitrary f ∈ BMO−1 (∂Ω, σ) .
M
Also, choose an arbitrary function g ∈ L n−1 (∂Ω, σ)
. In particular, we have
M  
g ∈ BMO−1 (∂Ω, σ) by [69, (11.10.52)]. Then for each r ∈ 0, 2 diam(∂Ω) and
x ∈ ∂Ω we may estimate
 ∫ 1
1   
  ∇ 𝒮( f − g)  p dist(·, ∂Ω) p−1 dL n p
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤C f −g [BMO−1 (∂Ω,σ)] M , (2.3.88)

thanks to (2.3.22) (written for f − g in place of f ). In addition, from (1.5.64) we

know that
 ∫ 1
1    p
lim+ sup   ∇ 𝒮g  dist(·, ∂Ω) p−1 dL n p = 0.
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
(2.3.89)

Collectively, (2.3.88) and (2.3.89) imply

2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 335
 ∫ 1
1    p
lim+ sup   ∇ 𝒮 f  dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤C f −g [BMO−1 (∂Ω,σ)] M , (2.3.90)

for some constant C ∈ (0, ∞) independent of f and g. With this in hand, (2.3.25)
M
follows on account of [69, (11.10.53)] and the arbitrariness of g ∈ L n−1 (∂Ω, σ) .
To deal with the claims in item (vii), assume Ω ⊆ Rn , where n ∈ N with n ≥ 3,
is an open set satisfying a two-sided local John condition and whose boundary is
compact and Ahlfors regular. To justify the identity in (2.3.28), pick an arbitrary
M M
function f ∈ L n−1 (∂Ω, σ) → BMO−1 (∂Ω, σ) (cf. [69, (11.10.52)]). Fix
 
q, q ∈ (1, ∞) with 1/q + 1/q = 1 and recall that a fractional integration operator of
M  M
order 1 maps L n−1 (∂Ω, σ) boundedly into L q (∂Ω, σ) . For each C M -valued
(q, 1)-atom g on ∂Ω we may then write
∫
 
S  f , g dσ = [BMO(∂Ω,σ)] M S  f , g [H 1 (∂Ω,σ)] M
∂Ω
 
= [BMO−1 (∂Ω,σ)] M f , Sg n−1 ,1
[H1n−2 (∂Ω,σ)] M
 
= [L n−1 (∂Ω,σ)] M f , Sg n−1
[L n−2 (∂Ω,σ)] M
∫ ∫
=  f , Sg dσ = SL  f , g dσ. (2.3.91)
∂Ω ∂Ω

The first equality above is implied [69, Lemma 4.6.5] (bearing in mind [68,
(7.4.106)]). The second equality in (2.3.91) uses the fact that S  is the transpose of
S in (2.3.13). The third equality in (2.3.91) comes from the compatibility property
recorded in [69, (11.10.55)]. For the fourth equality in (2.3.91), we have used the
n−1
fact that the duality between L n−1 (∂Ω, σ) and L n−2 (∂Ω, σ) is given by the integral
pairing. Finally, the fifth equality in (2.3.91) is a consequence of (1.3.62), Fubini’s
Theorem, and the first formula in [70, (1.4.32)]. Once (2.3.91) has been established,
(2.3.28) follows with the help of [69, Lemma 4.6.9].
M
Finally, that the restriction of the operator (2.3.14) to VMO−1 (∂Ω, σ) in-
duces a well-defined, linear, and bounded mapping in the context described in
M
(2.3.29), is seen from (2.3.14), (2.3.28), the fact that SL  maps L n−1 (∂Ω, σ)
M
into L1n−1 (∂Ω, σ) (as seen from item (ix) in Theorem 1.5.1), [69, (11.5.203)],
and [69, Lemma 1.2.20]. This completes the treatment of item (vii), and finishes the
proof of Theorem 2.3.1. 
For further use, we note here that Theorem 2.3.1 implies the following result.
Corollary 2.3.2 Suppose Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is a UR domain.
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
 theoretic outward
αβ 
unit normal to Ω. For M ∈ N, consider a coefficient tensor A = ar s 1≤r,s ≤n with
1≤α,β ≤M
336 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

complex entries, with the property that the M × M homogeneous second-order system
L = L A associated with A in Rn as in (1.3.2) is weakly elliptic (in the sense of [70,
 1.3.1]). Finally, fix an aperture parameter κ ∈ (0, ∞) along with
(1.3.3) in Definition
an exponent p ∈ n−1 n , 1 . Then the following properties hold.

(i) Recall the modified boundary-to-domain double layer potential operator Dmod
  M
acting on functions from L 1 ∂Ω, 1+σ(x)
|x | n as in (1.8.6). Then the operator
. .p  M M
∂νA Dmod : H1 (∂Ω, σ) ∼ −→ H p (∂Ω, σ) defined as
.A .A .p M
(2.3.92)
∂ν Dmod [ f ] := ∂ν (Dmod f ) for each f ∈ H1 (∂Ω, σ) ,

is well defined, linear, and bounded, when the quotient space is equipped with
the semi-quasinorm8 introduced in (A.0.92).
(ii) Recall the boundary-to-boundary single layer potential operator S from Theo-
rem 2.2.6. Then the operator
M .p  M
[S] : H p (∂Ω, σ) −→ H1 (∂Ω, σ) ∼ defined as
.p  M M
(2.3.93)
[S] f := [S f ] ∈ H1 (∂Ω, σ) ∼ , ∀ f ∈ H p (∂Ω, σ) .

is well defined, linear, and bounded, when the quotient space is equipped with
the semi-quasinorm introduced in (A.0.92).

Proof All claims are direct consequences of Theorem 2.3.1 and definitions. 
Up to additive constants, functions in homogeneous Hardy-based Sobolev spaces
actually belong to a Lebesgue space corresponding to a sharp embedding exponent.
This is made precise in the theorem below.

Theorem 2.3.3 Let Ω ⊆ Rn (where n ∈ N with n ≥ 2) be an open set satisfying a

two-sided local John condition with an Ahlfors regular boundary, and abbreviate
σ := H n−1 ∂Ω. Also,
 n−1   
1 −1
fix p ∈ n , n − 1 and define p∗ := p1 − n−1 ∈ (1, ∞). (2.3.94)
.p
Then for each function f ∈ H1 (∂Ω, σ) one may find some number c = c( f ) ∈ C
such that
f − c belongs to the space L p∗ (∂Ω, σ), (2.3.95)
and, in addition,

8 if in fact Ω ⊆ R n is an open set satisfying a two-sided local John condition and whose boundary
is an unbounded Ahlfors regular set, then Proposition 2.3.8 guarantees that said semi-quasinorm
.p 
becomes a genuine quasinorm, making H1 (∂Ω, σ) ∼ a quasi-Banach space
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 337

if ∂Ω is unbounded, there exists C = C(Ω, n, p) ∈ (0, ∞) so that

n
  (2.3.96)
f −c L p∗ (∂Ω,σ) ≤C ∂τ f  p
jk H (∂Ω,σ)
= C f H. p (∂Ω,σ) .
1
j,k=1

In fact, with C canonically identified with the space of constant functions on ∂Ω, one
has .p p ,p
H1 (∂Ω, σ) = C + H1 ∗ (∂Ω, σ). (2.3.97)
As a consequence, .p p
H1 (∂Ω, σ) ⊆ Lloc∗ (∂Ω, σ), (2.3.98)
and
.p
if ∂Ω is unbounded then for each f ∈ H1 (∂Ω, σ) one has
(2.3.99)
f H. p (∂Ω,σ) = 0 ⇐⇒ f is a constant function on ∂Ω.
1

Lastly, one has

1, p .p p ,p
H1 (∂Ω, σ) = H1 (∂Ω, σ) = H1 ∗ (∂Ω, σ)
(2.3.100)
as sets, provided ∂Ω is bounded,

as well as
1, p
H1 (∂Ω, σ) → L p∗ (∂Ω, σ)
(2.3.101)
continuously and densely, if ∂Ω is bounded,
and
if ∂Ω is bounded, there exists a constant C ∈ (1, ∞) such that
C −1 f p∗ , p
H1 (∂Ω,σ) ≤ f H1
1, p
(∂Ω,σ)
≤C f H1
p∗ , p
(∂Ω,σ) (2.3.102)
1, p .p p ,p
for all f ∈ H1 (∂Ω, σ) = H1 (∂Ω, σ) = H1 ∗ (∂Ω, σ).
  1 
1 −1
 
Note that if p ∈ n−1n , 1 then p∗ := p − n−1 ∈ 1, n−1
n−2 . That the latter range
is optimal as far as the conclusion in (2.3.95) is concerned, may be seen from [69,
(11.10.66)].
Proof of Theorem 2.3.3 Consider first the case when p ∈ (1, n − 1). Then from [69,
(11.10.30)], item (1) in [69, Theorem 11.5.16], [69, (11.5.102)], [69, (11.1.27)], and
[68, Corollary 3.7.3] we see that all claims made in the statement of the theorem are
true. We are therefore left with consider the case when either

n ≥ 3 and p ∈ n−1 n ,1 , (2.3.103)

or 1 
n = 2 and p ∈ 2, 1 . (2.3.104)
338 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

We shall simultaneously consider both scenarios described in (2.3.103)-(2.3.104).

The key claim we make is that. there exists some constant C = C(Ω, n, p) ∈ (0, ∞)
p
such that for each given f ∈ H1 (∂Ω, σ) we may find some non-negative function
g ∈ L p (∂Ω, σ) with the property that

| f (x) − f (y)| ≤ |x − y| · g(x) + g(y) for σ-a.e. x, y ∈ ∂Ω, (2.3.105)

and for which

g L p (∂Ω,σ) ≤ C f H. p (∂Ω,σ) if ∂Ω is unbounded. (2.3.106)

1

To justify this claim, we argue largely as in the proof of [69, Lemma 11.5.10].
For starters, define Ω+ := Ω and Ω− := Rn \ Ω. From the current hypotheses and
[68, (5.11.27)] together with item (7) in [68, Lemma 5.10.9] we then conclude
that both Ω+ and Ω− are UR domains satisfying a local John condition, and with
∂Ω+ = ∂Ω = ∂Ω− . As in the proof of [69, Lemma 11.5.10], denote by Kmod the
modified principal-value harmonic double layer operator
.p associated with Ω and
L := Δ as in (1.8.24). Fix an arbitrary function f ∈ H1 (∂Ω, σ), so
 
f ∈ L 1 ∂Ω, 1+σ(x)
|x | n ⊆ Lloc (∂Ω, σ) satisfying
1
(2.3.107)
∂τ j k f ∈ H p (∂Ω, σ) for all j, k ∈ {1, . . . , n}

(see (A.0.89)), and decompose

 
f = f+ − f− where f± := ± 12 I + Kmod f , (2.3.108)

To proceed, define u± := ±Dmod ± f in Ω , where D ± are the modified boundary-

± mod
to-domain harmonic double layer operators associated with Ω± and L := Δ as in
(1.8.6). Then from Theorem 1.8.2, and Theorem 2.3.1 we see that u± : Ω± → C
are well-defined (harmonic) functions belonging to 𝒞∞ (Ω± ) which, for any given
aperture parameter κ > 0, satisfy
κ−n.t.  
u± ∂Ω = ± 12 I + Kmod f = f± at σ-a.e. point on ∂Ω,
Nκ (∇u± ) belongs to the space L p (∂Ω, σ), and
(2.3.109)
  n
 
Nκ (∇u± ) p ≤C ∂τ f  p = C f H. p (∂Ω,σ),
L (∂Ω,σ) jk H (∂Ω,σ) 1
j,k=1

for some constant C = C(Ω, n, κ, p) ∈ (0, ∞). At this point, the discussion branches
out, depending on whether ∂Ω is unbounded, or bounded. For now, assume that ∂Ω
is unbounded, a scenario in which [69, Lemma 11.5.8] implies

| f± (x) − f± (y)| ≤ |x − y| · g± (x) + g± (y) for σ-a.e. x, y ∈ ∂Ω,

(2.3.110)
where g± := C · Nκ (∇u± ).
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 339

If we now define g := g+ + g− then the claims in (2.3.105)-(2.3.106) in the case

when ∂Ω is unbounded are seen from (2.3.109), and (2.3.110).
Moving on, consider the scenario in which (2.3.103) holds and ∂Ω is bounded.
Define f± and u± as before. These are functions which continue to enjoy the properties
described in (2.3.108)-(2.3.109). What changes is that now [69, Lemma 11.5.8] gives

| f± (x) − f± (y)| ≤ |x − y| · g± (x) + g± (y) for σ-a.e. x, y ∈ ∂Ω,

 κ−n.t.   (2.3.111)
where g± := C u± ∂Ω  + Nκ (∇u± ) .

If we define g := g+ + g− then (2.3.105) holds, but proving the membership of g to

L p (∂Ω, σ) requires more work than before (due to the presence of the nontangential
boundary traces in the formula for g± in (2.3.111)). Specifically, we first note that
κ−n.t. 
u± ∂Ω = f± ∈ L 1,∞ (∂Ω, σ) ⊆ L q (∂Ω, σ), (2.3.112)
0<q<1

thanks to the formula in the first line of (2.3.109), the mapping property recorded in
[70, (2.3.33)], the membership in the first line of [69, (11.5.105)], the fact that ∂Ω is
bounded, and the embedding in [68, (6.2.38)]. In turn, from (2.3.112), the definitions
of g± from (2.3.111), and the membership in the second line of (2.3.109) we then
see that 
g± ∈ L p (∂Ω, σ) + L q (∂Ω, σ), (2.3.113)
0<q<1

hence 
g = g+ + g− ∈ L p (∂Ω, σ) + L q (∂Ω, σ). (2.3.114)
0<q<1

If p < 1, this puts g in L p (∂Ω, σ), and we are done. If p = 1, we arrive at the
conclusion that 
g∈ L q (∂Ω, σ). (2.3.115)
0<q<1

As noted earlier, the function g satisfies (2.3.105). This permits us to invoke [69,
Lemma 11.5.9] and deduce that
  1 
1 −1
whenever q ∈ n−1
n , 1 and q∗ := q − n−1 then
 n−1  (2.3.116)
q∗ ∈ 1, n−2 and f belongs to the space L q∗ (∂Ω, σ).

Finally, from (2.3.116) and [70, (2.3.35)] we infer that

κ−n.t.  
u± ∂Ω = ± 12 I + Kmod f ∈ L q∗ (∂Ω, σ), (2.3.117)

which, in light of [69, (11.5.118)] and the definitions in [69, (11.5.113)], implies

g ∈ L q∗ (∂Ω, σ) + L 1 (∂Ω, σ) = L 1 (∂Ω, σ), (2.3.118)

340 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

since ∂Ω is presently assumed to be bounded. This establishes (2.3.105) for some

non-negative function g ∈ L p (∂Ω, σ) in all cases described in (2.3.103)-(2.3.104).
Having proved (2.3.105)-(2.3.106),
.p we may rely on item (1) in [69, Lemma 11.5.9]
and conclude that for each f ∈ H1 (∂Ω, σ) there exists c = c( f ) ∈ C such that (2.3.95)
and (2.3.96) are true. In turn, from (2.3.95) and (A.0.89) we see that (2.3.97) holds.
If ∂Ω is bounded, the first equality in (2.3.100) is implied by [69, (11.10.41)],
while the second equality in (2.3.100) is a consequence of (2.3.97) (bearing in mind
p ,p
that C ⊆ H1 ∗ (∂Ω, σ) now).
Next, consider the task of proving (2.3.101). Henceforth, assume that ∂Ω is
bounded. Then [69, (11.10.41)] and (2.3.98) imply that
1, p
H1 (∂Ω, σ) ⊆ L p∗ (∂Ω, σ) as sets. (2.3.119)

Consequently, the identity operator

1, p
H1 (∂Ω, σ)  f −→ f ∈ L p∗ (∂Ω, σ) (2.3.120)

is a well-defined linear mapping. We claim that its graph is closed in the product
1, p
topology on H1 (∂Ω, σ) × L p∗ (∂Ω, σ). To justify this claim, suppose
1, p 1, p
{ f j } j ∈N ⊆ H1 (∂Ω, σ) and (g, h) ∈ H1 (∂Ω, σ) × L p∗ (∂Ω, σ)
1, p
are such that lim f j = g in H1 (∂Ω, σ) and lim f j = h in L p∗ (∂Ω, σ).
j→∞ j→∞
(2.3.121)
We then conclude from (A.0.94) that lim f j = g in L 1 (∂Ω, σ). Given that we
j→∞
also have lim f j = h in L 1 (∂Ω, σ) (since ∂Ω is bounded and p∗ > 1), it follows
j→∞
that g = h. Thus, (g, h) belongs to the graph (2.3.120), proving that this graph is
1, p
closed. Since H1 (∂Ω, σ) and L p∗ (∂Ω, σ) are quasi-Banach spaces, the version of
the Closed Graph Theorem for quasi-Banach spaces proved in [73, Corollary 6.78,
p. 442] implies that (2.3.120) is a bounded mapping. This establishes the continuity
of the embedding in (2.3.101). The fact that this embedding also has dense range is
a consequence of [69, (11.10.39)] and [68, (3.7.22)]. Finally, the first inequality in
(2.3.102) is implied by (2.3.101), while the second inequality in (2.3.102) follows
from definitions and the fact that L p∗ (∂Ω, σ) → L 1 (∂Ω, σ) continuously. 
A version of Theorem 2.3.3 corresponding to the forbidden values n = 2 and
p = 1 is presented below.

Proposition 2.3.4 Let Ω ⊆ R2 be an open set satisfying a two-sided local John

condition whose boundary is an Ahlfors regular set, and abbreviate σ := H 1 ∂Ω.
Then .
H11 (∂Ω, σ) ⊆ VMO(∂Ω, σ) ⊆ BMO(∂Ω, σ) (2.3.122)
.
and the inclusion H11 (∂Ω, σ) → BMO(∂Ω, σ) is actually continuous if ∂Ω is un-
bounded. More specifically,
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 341

if ∂Ω is unbounded, there exists C = C(Ω) ∈ (0, ∞) such that

. (2.3.123)
f BMO(∂Ω,σ) ≤ C f H. 1 (∂Ω,σ) for all f ∈ H11 (∂Ω, σ).
1

As a consequence of (2.3.123) and [68, (7.4.93)],

.
if ∂Ω is unbounded then for each f ∈ H11 (∂Ω, σ) one has
(2.3.124)
f H. 1 (∂Ω,σ) = 0 ⇐⇒ f is a constant function on ∂Ω.
1

Finally,
H11,1 (∂Ω, σ) → VMO(∂Ω, σ)
(2.3.125)
continuously and densely, if ∂Ω is bounded,
hence, in particular,

if ∂Ω is bounded, there exists some constant C ∈ (0, ∞) such that

 
f BMO(∂Ω,σ) ≤ C f L 1 (∂Ω,σ) + f H. 1 (∂Ω,σ) = C f H 1,1 (∂Ω,σ) (2.3.126)
.1 1 1

for all functions f ∈ H1 (∂Ω, σ) = H11,1 (∂Ω, σ) (cf. [69, (11.10.41)]).

Proof Reasoning as in the proof of Theorem 2.3.3 with n = 2 and p =. 1, we see that
there exists some constant C = C(Ω) ∈ (0, ∞) such that for each f ∈ H11 (∂Ω, σ) it is
possible to find some non-negative function g ∈ L 1 (∂Ω, σ) with the property that

| f (x) − f (y)| ≤ |x − y| · g(x) + g(y) for σ-a.e. x, y ∈ ∂Ω, (2.3.127)

and for which

g L 1 (∂Ω,σ) ≤ C f H. 1 (∂Ω,σ) if ∂Ω is unbounded. (2.3.128)

1

Granted these, we may invoke item (2) in [69, Lemma 11.5.9] (with n = 2) to
conclude that f belongs to VMO(∂Ω, σ) and that (2.3.123) holds.
For the remainder of the proof assume that ∂Ω is bounded. Then [69, (11.10.41)]
and (2.3.122) imply that

H11,1 (∂Ω, σ) ⊆ VMO(∂Ω, σ) as sets. (2.3.129)

In particular, the identity operator

H11,1 (∂Ω, σ)  f −→ f ∈ VMO(∂Ω, σ) (2.3.130)

is a well-defined linear mapping. Also, we claim that its graph is closed in the product
topology on H11,1 (∂Ω, σ) × VMO(∂Ω, σ). Indeed, if
342 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

{ f j } j ∈N ⊆ H11,1 (∂Ω, σ) and (g, h) ∈ H11,1 (∂Ω, σ) × VMO(∂Ω, σ)

are such that lim f j = g in H11,1 (∂Ω, σ) and lim f j = h in BMO(∂Ω, σ)
j→∞ j→∞
(2.3.131)
we conclude from (A.0.94) and [68, (7.4.105)] that lim f j = g in L 1 (∂Ω, σ) and
j→∞
lim f j = h in L 1 (∂Ω, σ). Hence, g = h. Consequently, (g, h) belongs to the
j→∞
graph (2.3.130), proving that said graph is closed. Given that both H11,1 (∂Ω, σ) and
VMO(∂Ω, σ) are Banach spaces, the Closed Graph Theorem implies that (2.3.130)
is a bounded mapping. This establishes the continuity of the embedding in (2.3.125),
That this embedding also has dense range is implied by [69, (3.1.51), (11.10.39)].
Finally, (2.3.126) is seen from (2.3.125). 
Remark 2.3.5 Assume Ω ⊆ R2 is an open set satisfying a two-sided local John
condition, with a compact Ahlfors regular boundary, and abbreviate σ := H 1 ∂Ω.
Then Proposition 2.3.4, [69, Definition 11.10.6], and [68, (7.4.105)] imply that
q,1
H11,1 (∂Ω, σ) = H1 (∂Ω, σ) for each q ∈ [1, ∞). (2.3.132)

In such a setting, we then proceed to define (compare with [69, Definition 11.10.9];
see also (A.0.32))
 ∗  ∗
q,1
BMO−1 (∂Ω, σ) := H11,1 (∂Ω, σ) = H1 (∂Ω, σ) (2.3.133)

for each q ∈ [1, ∞).

Also,

for an arbitrary integrability exponent q ∈ (1, ∞) we define

(2.3.134)
VMO−1 (∂Ω, σ) := the closure of L q (∂Ω, σ) in BMO−1 (∂Ω, σ).

In relation to this, first we note that (2.3.133) and [69, (11.10.47)] imply we have a
well-defined continuous embedding

L q (∂Ω, σ) → BMO−1 (∂Ω, σ) for each q ∈ (1, ∞) (2.3.135)

which, in turn, makes the definition of VMO−1 (∂Ω, σ) given above meaningful.
Also, the fact that L q1 (∂Ω, σ) → L q0 (∂Ω, σ) continuously and densely whenever
1 < q0 < q1 < ∞ makes the definition of VMO−1 (∂Ω, σ) unambiguous.
Retain the setting of Remark 2.3.5. Then, as is apparent from definitions,

VMO−1 (∂Ω, σ) is a closed subspace of BMO−1 (∂Ω, σ), and the space
L q (∂Ω, σ) embeds continuously and densely into VMO−1 (∂Ω, σ) for (2.3.136)
each q ∈ (1, ∞).
 
In fact, since φ|∂Ω : φ ∈ 𝒞∞ (R2 ) is dense in any L q (∂Ω, σ) with q ∈ (1, ∞) (cf.
[68, (3.7.22)]), we conclude from (2.3.136) that
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 343
 
the space φ|∂Ω : φ ∈ 𝒞∞ (R2 ) is dense in VMO−1 (∂Ω, σ); as a
consequence, VMO−1 (∂Ω, σ) is the closure of Lip(∂Ω) in the space (2.3.137)
BMO−1 (∂Ω, σ).
Finally, we wish to observe that, [69, (11.10.49)], (2.3.133), and [69, (1.2.25)] imply
that the following compatibility result is valid for each q, q  ∈ (1, ∞) satisfying
1/q + 1/q  = 1:
∫
   
f g dσ = L q (∂Ω,σ) f , g L q (∂Ω,σ) = BMO−1 (∂Ω,σ) f , g H q,1 (∂Ω,σ)
∂Ω 1

for all f ∈

L q (∂Ω, σ) → BMO−1 (∂Ω, σ) (2.3.138)
q,1
and all g ∈ H1 (∂Ω, σ) → L q (∂Ω, σ).

We continue by presenting the versions of [69, Proposition 11.11.5] and [69,

Proposition 11.11.6] corresponding to the forbidden values n = 2 and p = 1.
Proposition 2.3.6 Let Ω ⊆ R2 be an open set satisfying a two-sided local John
condition, with a compact Ahlfors regular boundary, and abbreviate σ := H 1 ∂Ω.
Then
 ∗
H 1 (∂Ω, σ) → H11,1 (∂Ω, σ) = BMO−1 (∂Ω, σ) continuously, (2.3.139)

and
   
H 1 (∂Ω,σ) f, g BMO(∂Ω,σ)
= BMO−1 (∂Ω,σ) f, g H11,1 (∂Ω,σ)
for every f ∈ H (∂Ω, σ) → BMO
1
−1 (∂Ω, σ) (2.3.140)
and g ∈ H11,1 (∂Ω, σ) → VMO(∂Ω, σ) → BMO(∂Ω, σ),

where the above embeddings are provided by (2.3.139), (2.3.125), and [69, (3.1.2)].
Proof That we have the continuous embedding claimed in (2.3.139) follows from
(2.3.125), [69, (4.6.22)], and [69, Lemma 1.2.1]. A direct proof, which also gives the
compatibility of the pairings claimed in (2.3.140) goes as follows. First,
 observe that
∗
there exists a constant C ∈ (0, ∞) with the property that for each Λ ∈ VMO(∂Ω, σ)
we have
% &  
sup |Λ f | : f ∈ H11,1 (∂Ω, σ), f H 1,1 (∂Ω,σ) ≤ 1 ≤ C Λ(VMO(∂Ω,σ))∗ . (2.3.141)
1

Indeed, this is a direct consequence of the continuity of the embedding in (2.3.125).

We next consider
 ∗  ∗
Φ : VMO(∂Ω, σ) −→ H11,1 (∂Ω, σ) defined as
 ∗   ∗ (2.3.142)
VMO(∂Ω, σ)  Λ → Φ(Λ) := Λ H 1,1 (∂Ω,σ) ∈ H11,1 (∂Ω, σ) .
1

The fact that this is a well defined linear operator is then seen from (2.3.141) which
∗
actually implies that for each Λ ∈ VMO(∂Ω, σ) we have
344 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
  ∗
Λ belongs to H11,1 (∂Ω, σ) and
H11,1 (∂Ω,σ)
    
   Λ
(2.3.143)
Λ H 1,1 (∂Ω,σ)  1,1 ∗
≤ C (VMO(∂Ω,σ))∗
.
1 (H1 (∂Ω,σ))

In fact, (2.3.143) also implies that Φ is continuous in the context of (2.3.142). Going
further,
 we claim that  ∗ Φ from (2.3.142) is injective.
 To this end,
 ∗ assume the functional
Λ ∈ VMO(∂Ω, σ) is such that Φ(Λ) = 0 in H11,1 (∂Ω, σ) . This forces Λ f = 0 for
each function f ∈ H11,1 (∂Ω, σ). In particular, Λ f = 0 for each f ∈ Lip(∂Ω), thanks
to [69, (11.10.39)]. In concert with [69, (3.1.51)], this ultimately forces  ∗ Λ f = 0 for
each f ∈ VMO(∂Ω, σ), hence Λ = 0 as a functional in VMO(∂Ω, σ) . This proves
that Φ is indeed injective in the context of (2.3.142).
 ∗ function f ∈ H (∂Ω, σ), let us associate the functional
Next, 1
 with any given
Λ f ∈ VMO(∂Ω, σ) by setting
 
Λ f (φ) := H 1 (∂Ω,σ) f , φ BMO(∂Ω,σ) for each φ ∈ VMO(∂Ω, σ). (2.3.144)

According to [69, (4.6.22)-(4.6.23)],

 ∗
the assignment H 1 (∂Ω, σ)  f −→ Λ f ∈ VMO(∂Ω, σ) is
the concrete embodiment
 of the implicit
 ∗ identification in the (2.3.145)
formula H 1 (∂Ω, σ) = VMO(∂Ω, σ) .

Then
the inclusion in (2.3.139) is given by
 ∗ (2.3.146)
H 1 (∂Ω, σ)  f → Φ(Λ f ) ∈ H11,1 (∂Ω, σ) .
Granted (2.3.145) and the properties of Φ established so far, it follows that the map
in (2.3.146) is well defined, linear, continuous, and injective. Thus, (2.3.139) holds.
As far as (2.3.140) is concerned, pick two arbitrary functions, f ∈ H 1 (∂Ω, σ) and
g ∈ H11,1 (∂Ω, σ). Then, since g ∈ H11,1 (∂Ω, σ) → VMO(∂Ω, σ) (cf. (2.3.125)), we
may use (2.3.146), (2.3.142), and (2.3.144) to write
   
(H 1,1 (∂Ω,σ))∗ f , g H 1,1 (∂Ω,σ) = (H 1,1 (∂Ω,σ))∗ Φ(Λ f ), g H 1,1 (∂Ω,σ)
1 1 1 1
#  
= (H 1,1 (∂Ω,σ))∗ Λf  H11,1 (∂Ω,σ)
,g H11,1 (∂Ω,σ)
1
 
= Λ f (g) = H 1 (∂Ω,σ) f, g BMO(∂Ω,σ)
(2.3.147)

as wanted. 
We next study the action of the boundary-to-domain single layer potential operator
on the negative Sobolev space BMO−1 in the two-dimensional setting.
Theorem 2.3.7 Assume Ω ⊆ R2 is an open set satisfying a two-sided local John con-
dition whose boundary is a compact Ahlfors regular set. Abbreviate σ := H 1 ∂Ω.
Let L be a homogeneous, weakly elliptic, constant (complex) coefficient, second-
order M × M system in R2 (for some M ∈ N), which is weakly elliptic (in the sense of
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 345

[70, (1.3.3) in Definition 1.3.1]) Finally, denote by E = (Eαβ )1≤α,β ≤M the matrix-
valued fundamental solution associated with the system L as in [70, Theorem 1.4.2].
Then the boundary-to-domain single layer potential operator associated with L
and Ω as in (1.3.6) induces a well-defined linear mapping in the context

−→ 𝒞∞ (Ω)
M M
𝒮 : BMO−1 (∂Ω, σ) (2.3.148)
M
if for each f = ( fβ )1≤β ≤M ∈ BMO−1 (∂Ω, σ) one sets
  
𝒮 f (x) := Eαβ (x − ·)∂Ω, fβ for all x ∈ Ω, (2.3.149)
1≤α ≤M

where the pairing ·, · above stands for the duality

 bracket between the space
∗
1,1 1,1
H1 (∂Ω, σ) and the space BMO−1 (∂Ω, σ) = H1 (∂Ω, σ) (cf. (2.3.133)). In
M
addition, for each f ∈ BMO−1 (∂Ω, σ) one has
 
L 𝒮 f = 0 in Ω, (2.3.150)

and for each p ∈ (1, ∞) there exists some constant C = C(Ω, L, p) ∈ (0, ∞) such that

|∇𝒮 f | p dist(·, ∂Ω) p−1 dL 2 is a Carleson measure in

p
the set Ω with constant ≤ C f [BMO (∂Ω,σ)] M . (2.3.151)
−1

In particular, corresponding to p = 2, it follows that9

|∇𝒮 f | 2 dist(·, ∂Ω) dL 2 is a Carleson measure in Ω with constant

M (2.3.152)
≤ C f [BMO2
(∂Ω,σ)] M
, for each functional f ∈ BMO−1 (∂Ω, σ) .
−1

M
Also, for each given f ∈ BMO−1 (∂Ω, σ) , the boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
 κ−n.t.    (2.3.153)
𝒮f ∂Ω
(x) = (SL  ) f (x) at σ-a.e. point x ∈ ∂Ω,

where SL  is the boundary-to-boundary single layer potential operator associated

M M
with L  , considered as a mapping from H 1 (∂Ω, σ)
q,1
into H1 (∂Ω, σ) with
q ∈ (1, ∞) (cf. (2.3.13)), and (SL  ) , mapping
 
q,1 M ∗ M
H1 (∂Ω, σ) = BMO−1 (∂Ω, σ) (2.3.154)

into  
M ∗ M
H 1 (∂Ω, σ) = BMO(∂Ω, σ) , (2.3.155)

 2
9 it is natural to regard  ∇𝒮 f  dist(·, ∂Ω) dL 2 as the Littlewood-Paley measure associated with f
via the operator 𝒮
346 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

stands for its (real) transpose (cf. the duality result from [69, Theorem 4.6.1] and
Remark 2.3.5).
M
Finally, for each f ∈ BMO−1 (∂Ω, σ) one has
 ∫ 1
1    p
lim+ sup   ∇ 𝒮 f  dist(·, ∂Ω) p−1 dL 2 p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , [VMO−1 (∂Ω, σ)] M (2.3.156)

where the distance in the right-hand side is considered in [BMO−1 (∂Ω, σ)] M . As a
consequence,

|∇𝒮 f | p dist(·, ∂Ω) p−1 dL 2 is a vanishing Carleson measure

M (2.3.157)
in Ω for each f ∈ VMO−1 (∂Ω, σ) and p ∈ (1, ∞),

hence, corresponding to p = 2,

|∇𝒮 f | 2 dist(·, ∂Ω) dL 2 is a vanishing Carleson mea-

M (2.3.158)
sure in Ω for each f ∈ VMO−1 (∂Ω, σ) .

Proof Throughout, fix q, q  ∈ (1, ∞) with 1/q + 1/q  = 1. Then [69, Proposi-

tion 11.10.7] and Remark 2.3.5 guarantee the existence of a constant C ∈ (0, ∞) with
the property that for each functional
 
M q,1 M ∗
f = ( fβ )1≤β ≤M ∈ BMO−1 (∂Ω, σ) = H1 (∂Ω, σ) (2.3.159)

(β)  (β)
we can find two families of functions, f0 ∈ L q (∂Ω, σ) and f12 ∈ BMO(∂Ω, σ)
with β ∈ {1, . . . , M }, satisfying
M %   (β)  &
 f (β)   +  f12 BMO(∂Ω,σ) ≤ C f [BMO−1 (∂Ω,σ)] M (2.3.160)
0 L q (∂Ω,σ)
β=1

as well as
 
[BMO−1 (∂Ω,σ)] M f, g q,1
[H1 (∂Ω,σ)] M
∫
(β)  (β) 
= f0 gβ dσ + BMO(∂Ω,σ) f12 , ∂τ12 gβ H 1 (∂Ω,σ) (2.3.161)
∂Ω
q,1 M
for every function g = (gβ )1≤β ≤M ∈ H1 (∂Ω, σ) .

(β)  M
Let us abbreviate f0 := ( f0 )1≤β ≤M ∈ L q (∂Ω, σ) . Also, for each pair of indeces
αβ
α, β ∈ {1, . . . , M } denote by T12 the integral operator defined as in (2.3.80) with
j := 1 and k := 2. Then the same argument which has produced (2.3.81) currently
gives
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 347
 αβ (β) 
𝒮 f = 𝒮 f0 − T12 f12 1≤α ≤M in Ω. (2.3.162)

From (2.3.162) it is then clear that (2.3.148) is a well-defined linear mapping and
that (2.3.150) holds. In addition, the same type of argument as in (2.3.82)-(2.3.87)
leads to the conclusion that the boundary trace formula (2.3.153) is true.
Next, the claims in (2.3.151) follow from (2.3.162), [70, Corollary 2.4.2], and
(2.3.160), as soon as we show that:
M
if p, q ∈ (1, ∞) and h ∈ L q (∂Ω, σ) then
|∇𝒮h| p dist(·, ∂Ω) p−1 dL 2 is a Carleson measure (2.3.163)
p
in Ω with constant ≤ C h [L q (∂Ω,σ)] M .

M
To justify (2.3.163), fix h ∈ L q (∂Ω, σ) and p ∈ (1, ∞). Set u := 𝒮h in Ω. Pick
 
R ∈ (0, ∞) large enough so that x ∈ R2 : dist(x, ∂Ω) ≤ 2 diam ∂Ω is contained in
B(0, R). We claim that

|∇u| p dist(·, ∂Ω) p−1 ∈ L 2 (ΩR, L 2 ) where ΩR := Ω ∩ B(0, R) and

  (2.3.164)
  p
|∇u| p dist(·, ∂Ω) p−1  2 ≤ C h [L q (∂Ω,σ)] M .
L (Ω R , L )
2

 
Assuming this claim for the time being, for each x ∈ ∂Ω and r ∈ 0, 2 diam ∂Ω we
may estimate
∫
|∇𝒮h| p dist(·, ∂Ω) p−1 dL 2
Ω∩B(x,r)
∫  1/2     1/2
≤ |∇u| 2p dist(·, ∂Ω)2(p−1) dL 2 L 2 Ω ∩ B(x, r)
Ω∩B(x,r)
      1/2
 
≤ |∇u| p dist(·, ∂Ω) p−1  · L 2 B(x, r)
L 2 (Ω R , L 2 )

p
≤ Cr h [L q (∂Ω,σ)] M
, (2.3.165)

from which (2.3.163) follows. There remains to prove the claim made in (2.3.164).
To this end, observe that (1.5.56) and (1.5.57) presently give (for some fixed aperture
parameter κ > 0)
    

dist(·, ∂Ω)1/q · ∇u ∞ ≤ C Nκ (∇u) L q (∂Ω,σ)
L (Ω, L )
2

≤C h [L q (∂Ω,σ)] M . (2.3.166)

In concert with [68, (8.6.51)] this permits us to conclude that

  

dist(·, ∂Ω)1/q · ∇u ∞ ≤ C h [L q (∂Ω,σ)] M , (2.3.167)
L (Ω R , L )
2
348 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

and
 
 
|∇u|  ≤C h [L q (∂Ω,σ)] M . (2.3.168)
L 2q (Ω R , L 2 )

If q ≤ p, use (2.3.167)-(2.3.168) to write

 2 ∫
 
|∇u| p dist(·, ∂Ω) p−1  2 = |∇u| 2p dist(·, ∂Ω)2(p−1) dL 2
L (Ω R , L )
2
ΩR
∫   2(p−q)
= |∇u| 2q |∇u||dist(·, ∂Ω)1/q dist(·, ∂Ω)2p(1−1/q) dL 2
ΩR
  2(p−q)  2q
  
≤ (2R)2p(1−1/q) dist(·, ∂Ω)1/q · ∇u ∞ |∇u|  2q
L (Ω R , L 2 ) L (Ω R , L 2 )

2p
≤ C(2R)2p(1−1/q) h [L q (∂Ω,σ)] M
, (2.3.169)

from which (2.3.164) follows. If p < q, use Hölder’s inequality and (2.3.168) to
write
 2 ∫
 
|∇u| p dist(·, ∂Ω) p−1  2 = |∇u| 2p dist(·, ∂Ω)2(p−1) dL 2
L (Ω R , L )
2
ΩR
∫  p/q   1−p/q
≤ |∇u| 2q dist(·, ∂Ω)2q(p−1)/p dL 2 · L 2 (ΩR )
ΩR
 2p
 
≤ C(2R)2(p−1) R2(1−p/q) |∇u|  2q
L (Ω R , L 2 )

2p
≤ CR2p(1−1/q) h [L q (∂Ω,σ)] M
, (2.3.170)

from which (2.3.164) once again follows. This completes the proof of (2.3.151).
M
Finally, let us justify (2.3.156). To this end, pick some f ∈ BMO−1 (∂Ω, σ)
M
and select p ∈ (1, ∞). Also, choose an arbitrary function g ∈ L p (∂Ω, σ) . In
M  
particular, g ∈ BMO−1 (∂Ω, σ) by (2.3.135). Then for each r ∈ 0, 2 diam(∂Ω)
and x ∈ ∂Ω we may estimate
 ∫ 1
1   
  ∇ 𝒮( f − g)  p dist(·, ∂Ω) p−1 dL 2 p
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
≤C f −g [BMO−1 (∂Ω,σ)] M , (2.3.171)

thanks to (2.3.151) (written for f − g in place of the function f ). In addition, with

η := 1 − 1/p ∈ (0, 1) we have
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 349
 ∫ 1
1    p
  ∇ 𝒮g  dist(·, ∂Ω) p−1 dL 2 p
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
 ∫  p1
1
≤C   dist(·, ∂Ω) pη−1 dL 2 g [L p (∂Ω,σ)] M
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

≤ Cr η g [L p (∂Ω,σ)] M . (2.3.172)

The first inequality above uses the version of (1.5.57) written for 𝒮 in place of 𝒮mod
(and for g in place of f ), while the second inequality is based on [68, (8.6.101)]
used with n := 2, λ := 1 − pη, α := 1, β := 1, E := B(x, r) ∩ Ω, and [68, (8.1.17)].
Together, (2.3.171) and (2.3.172) imply
 ∫ 1
1    p
lim+ sup   ∇ 𝒮 f  dist(·, ∂Ω) p−1 dL 2 p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤C f −g [BMO−1 (∂Ω,σ)] M , (2.3.173)

for some constant C ∈ (0, ∞) independent of f and g. With this in hand, (2.3.156)
M
follows on account of the arbitrariness of the function g in L p (∂Ω, σ) and the
fact that the latter space is dense in [VMO−1 (∂Ω, σ)] M (see (2.3.136)). 
.p
In a favorable geometric setting, the homogeneous Hardy-based Sobolev space H1
(modulo constants) becomes a quasi-Banach space when equipped with the quasi-
norm assigning to each function the sum of the H p quasi-norms of its tangential
derivatives. This is made precise in our next proposition.

Proposition 2.3.8 Assume Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an open set

satisfying a two-sided local John condition and whose boundary is an unbounded

Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and pick some exponent p ∈ n−1
n ,1 .
.p 
Let H1 (∂Ω, σ) ∼ denote the quotient space of classes [·] of equivalence modulo
.p
constants of functions in H1 (∂Ω, σ), equipped with the semi-quasinorm (A.0.92).
.p  .p 
Then (A.0.92) is a genuine quasinorm on H1 (∂Ω, σ) ∼, and H1 (∂Ω, σ) ∼ is a
quasi-Banach space when equipped with the quasinorm (A.0.92).

Proof
. p That the semi-quasinorm (A.0.92) is actually a quasinorm on the space
H1 (∂Ω, σ) ∼ is readily seen from (2.3.99) if n ≥ 3 and (2.3.124) if n = 2. To
.p 
show that H1 (∂Ω, σ) ∼ is complete when equipped with the quasinorm (A.0.92),
.p  
let { fα }α∈N ⊆ H1 (∂Ω, σ) be such that [ fα ] α∈N is a Cauchy sequence in the
.p 
quotient space H1 (∂Ω, σ) ∼. Then for each fixed j, k ∈ {1, . . . , n} it follows that
 
∂τ j k fα α∈N is a Cauchy sequence in H p (∂Ω, σ). As the latter space is complete, it
follows that there exists g jk ∈ H p (∂Ω, σ) such that

∂τ j k fα −→ g jk in H p (∂Ω, σ) as α → ∞. (2.3.174)

To proceed, we shall consider two cases.

350 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 
Case I: Assume that either n ≥ 3, or n = 2 and we have p ∈ 12 , 1 . Define the
1 
1 −1
exponent p∗ := p − n−1 and note that, thanks to the working assumptions,
p∗ ∈ (1, ∞). The current assumptions also allow us to invoke Theorem 2.3.3 which
. p constant C = C(Ω, n, p) ∈ (0, ∞) with the property
guarantees the existence of some
that for each function f ∈ H1 (∂Ω, σ) we may find c( f ) ∈ C such that f − c( f )
belongs to the space L p∗ (∂Ω, σ) and
n
 
f − c( f ) L p∗ (∂Ω,σ) ≤C ∂τ f  p = C f H. p (∂Ω,σ) . (2.3.175)
jk H (∂Ω,σ) 1
j,k=1

Since ∂Ω is unbounded,
.p it follows that c( f ) is uniquely determined by f . In particular,
the assignment H1 (∂Ω, σ)  f → c( f ) ∈ C is linear. Hence, if we set


fα := fα − c( fα ) for each α ∈ N, (2.3.176)

it follows that
  

fα − 
fβ L p∗ (∂Ω,σ) =  fα − c( fα ) − fβ − c( fβ )  L p∗ (∂Ω,σ)
 
= ( fα − fβ ) − c( fα − fβ ) L p∗ (∂Ω,σ)
 
≤ C fα − fβ H. p (∂Ω,σ) = C [ fα − fβ ] H. p (∂Ω,σ)/∼
1 1
 
= C [ fα ] − [ fβ ] H. p (∂Ω,σ)/∼ for each α, β ∈ N. (2.3.177)
1

  .p 
In view of the fact that [ fα ] α∈N is a Cauchy sequence in H1 (∂Ω, σ) ∼, this
implies that { 
fα }α∈N is a Cauchy sequence in L p∗ (∂Ω, σ). In particular, there exists
f ∈ L (∂Ω, σ) such that
p∗


fα −→ f in L p∗ (∂Ω, σ) as α → ∞. (2.3.178)
 
In addition, since ∂τ j k 
fα = ∂τ j k fα in Lipc (∂Ω) for each j, k ∈ {1, . . . , n} and
α ∈ N, from (2.3.174) we conclude that for each j, k ∈ {1, . . . , n} we have

∂τ j k 
fα −→ g jk in H p (∂Ω, σ) as α → ∞. (2.3.179)

Consequently, for each j, k ∈ {1, . . . , n} and each test function ϕ ∈ Lipc (∂Ω) we
may write
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 351
∫
 
(Lip c (∂Ω)) ∂τ j k f , ϕ Lip c (∂Ω)
=− f (∂τ j k ϕ) dσ
∂Ω
∫
= − lim 
fα (∂τ j k ϕ) dσ
α→∞ ∂Ω
 
= lim (Lip c (∂Ω)) ∂τ j k 
fα, ϕ Lip c (∂Ω)
α→∞
 
= lim H p (∂Ω) ∂τ j k 
fα, [ϕ] (H p (∂Ω))∗
α→∞
 
= H p (∂Ω)g jk , [ϕ] (H p (∂Ω))∗
 
= (Lipc (∂Ω)) g jk , ϕ Lip (∂Ω), (2.3.180)
c

thanks to [69, Definition 11.2.1], (2.3.178), the duality result recorded in [69, The-
orem 4.6.1], [69, Lemma 4.6.4], and (2.3.179). In turn, this implies that

∂τ j k f = g jk ∈ H p (∂Ω, σ) for each j, k ∈ {1, . . . , n}. (2.3.181)

 
Upon recalling that f ∈ L p∗ (∂Ω, σ) ⊆ L 1 ∂Ω, 1+σ(x)
|x | , we then conclude that
n
.p
f ∈ H1 (∂Ω, σ).
 Finally,
 from (2.3.174), (2.3.181), and (A.0.92) we deduce that. the sequence
p 
[ fα ] α∈N converges to [ f ], the class of f , in the quotient space H1 (∂Ω, σ) ∼.
.p 
This proves that, in Case I, the space H1 (∂Ω, σ) ∼ is indeed complete when
equipped with the quasinorm (A.0.92).
Case II: Assume n = 2 and p = 1. From Proposition 2.3.4 and the fact that
BMO(∂Ω, σ) is complete, it follows that there exists some f ∈ BMO(∂Ω, σ) such
that
fα −→ f in BMO(∂Ω, σ) as α → ∞. (2.3.182)
 
σ(x)
In view of [68, (7.4.118)], we then have f ∈ L 1 ∂Ω, 1+ |x | 2
and fα −→ f in
1 (∂Ω, σ) as α → ∞. Granted these, then the same type of argument as before
Lloc
shows

∂τ j k f = g jk ∈ H 1 (∂Ω, σ) for each j, k ∈ {1, . . . , 2}, (2.3.183)

where the functions g jk ∈ H 1 (∂Ω, σ) are as in (2.3.174) with p = 1. Hence, f

.  
belongs to H11 (∂Ω, σ) and the sequence [ fα ] α∈N converges to [ f ], the class of f ,
. 
in the quotient space H11 (∂Ω, σ) ∼. The final conclusion is that, in the current case,
.1 
the space H1 (∂Ω, σ) ∼ is complete when equipped with the quasinorm (A.0.92).

Next we study the mapping properties of the modified boundary-to-boundary

double layer potential
.p operator acting on the scale of homogeneous Hardy-based
Sobolev space H1 (modulo constants).
352 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Theorem 2.3.9 Fix n, M ∈ N with n ≥ 2. Assume Ω is an NTA domain with

an Ahlfors regular boundary10
 αβ  and abbreviate σ := H n−1 ∂Ω. Also, consider a
coefficient tensor A = ar s 1≤r,s ≤n having complex entries, with the property that
1≤α,β ≤M
the M × M homogeneous second-order system L = L A associated with A in Rn as
in (1.3.2) is weakly elliptic (in the sense of [70, (1.3.3) in Definition
 1.3.1]). Lastly,
fix an aperture parameter κ ∈ (0, ∞) along with an exponent p ∈ n−1 n ,1 .
Then the modified boundary-to-boundary double layer potential operator Kmod ,
originally defined as in (1.8.24), induces a linear and bounded mapping
.p M .p M
Kmod : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) . (2.3.184)

As a consequence of this and (1.8.28), the operator

.p  M .p  M
Kmod : H1 (∂Ω, σ) ∼ −→ H1 (∂Ω, σ) ∼ defined as
.p  M . p M
(2.3.185)
Kmod [ f ] := Kmod f ∈ H1 (∂Ω, σ) ∼ , ∀ f ∈ H1 (∂Ω, σ) ,

is well defined, linear, and bounded, when all quotient spaces are equipped with the
semi-quasinorm11 introduced in (A.0.92).
.p M
Proof Pick f ∈ H1 (∂Ω, σ) and set u := Dmod f in Ω. From (A.0.89), (1.8.7),
and item (i) in Theorem 2.3.1 we know that u : Ω → C M is a well-defined function
which, for any aperture parameter κ > 0, satisfies
M
u belongs to 𝒞∞ (Ω) and Lu = 0 in Ω,
κ−n.t.  1 
u∂Ω = 2 I + Kmod f at σ-a.e. point on ∂Ω,
(2.3.186)
Nκ (∇u) belongs to the space L p (∂Ω, σ),
 
and Nκ (∇u) L p (∂Ω,σ) ≤ C f [ H. p (∂Ω,σ)] M ,
1

for some constant C = C(Ω, n, κ, p, L) ∈ (0, ∞). In particular, Theorem 2.2.8 applies
κ−n.t. .p
and gives that the nontangential boundary trace u∂Ω belongs to H1 (∂Ω, σ) ,
M
.p M
and (2.2.218) holds. In turn, this implies that Kmod f belongs to H1 (∂Ω, σ) and,
for some constant C = C(Ω, n, κ, p, L) ∈ (0, ∞) independent of f , we may write

10 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
11 Proposition 2.3.8 tells us that if Ω ⊆ R n is an open set satisfying a two-sided local John condition
and whose boundary is an unbounded Ahlfors regular set, then said semi-quasinorm is actually a
.p 
genuine quasinorm, and H1 (∂Ω, σ) ∼ becomes a quasi-Banach space
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 353
  
Kmod f .p
[ H1 (∂Ω,σ)] M
≤ C 1 I + K
2
 .
mod f [ H p (∂Ω,σ)] M + C f
.p
[ H1 (∂Ω,σ)] M
1

 κ−n.t. 
≤ C u∂Ω [ H. p (∂Ω,σ)] M + C f .p
[ H1 (∂Ω,σ)] M
1
 
≤ C Nκ (∇u) L p (∂Ω,σ) + C f .p
[ H1 (∂Ω,σ)] M

≤C f .p . (2.3.187)
[ H1 (∂Ω,σ)] M

Thus, Kmod induces a well-defined, linear, and bounded mapping in the context of
(2.3.184). The claims pertaining to (2.3.185) readily follow from this. 

Remark 2.3.10 The results in Theorem 2.3.9 apply to all modified boundary-to-
boundary double layer potential operators Kmod described in Examples 1.8.4-1.8.7.

We are now in a position to augment Theorem 1.8.26 with composition identities

involving the modified boundary-to-boundary double layer, its transpose version,
the modified boundary-to-boundary single layer, and the conormal derivative of
the modified double layer, now considered on homogeneous Hardy-based Sobolev
spaces (for related results, see also Theorem 2.3.18).

Theorem 2.3.11 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an NTA domain whose

boundary is an Ahlfors regular set. Denote by ν the geometric measure theoretic
outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Next, for some M ∈ N,
αβ
let A = ar s 1≤r,s ≤n be a complex coefficient tensor with the property that
1≤α,β ≤M

 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (2.3.188)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Having fixed an exponent


p ∈ n−1
n ,1 , (2.3.189)
.
recall the operators S from (2.3.10), ∂νA Dmod from (2.3.5), and Kmod from (2.3.184).
Finally, let K A#  be the operator associated with the coefficient tensor A and the set
Ω as in (2.1.4). Then the following statements are true.
.p M
(1) For each f ∈ H1 (∂Ω, σ) there exists c f , which is the nontangential trace
on ∂Ω of some C M -valued locally constant function in Ω, with the property that
at σ-a.e. point on ∂Ω one has
1   1    .  
2 I + Kmod − 2 I + Kmod f = S ∂νA Dmod f + c f . (2.3.190)

.p M
(2) For each function f ∈ H1 (∂Ω, σ) one has
 .A   . 
∂ν Dmod Kmod f = K A#  ∂νA Dmod f at σ-a.e. point on ∂Ω. (2.3.191)
354 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
M
(3) For each f ∈ H p (∂Ω, σ) there exists c f , which is the nontangential trace
on ∂Ω of some C M -valued locally constant function in Ω, with the property that
   
S K A#  f = Kmod S f + c f at σ-a.e. point on ∂Ω. (2.3.192)
M
(4) For each f ∈ H p (∂Ω, σ) , at σ-a.e. point on ∂Ω one has
1     .  
2 I + K A − 12 I + K A#  f = ∂νA Dmod S f .
#
(2.3.193)

Proof The argument largely proceeds as in the proof of Theorem 1.8.26, now substi-
tuting the integral representation formulas (1.8.176) from Theorem 1.8.17, (1.8.194)
from Corollary 1.8.18, and (1.8.200) from Theorem 1.8.19, with the integral repre-
sentation formula (2.2.168) from Theorem 2.2.7, and keeping in mind the mapping
properties for layer potentials from Theorem 2.1.1, Theorem 2.3.1, Theorem 2.3.9,
Theorem 2.2.3, and Theorem 2.2.6.
Turning to specifics, observe first that the present hypotheses imply (cf. [68,
(5.10.33)]) that Ω is a UR domain. Fix an aperture parameter κ ∈ (0, ∞). To justify
.p M
the claim made in item (1), pick some arbitrary function f ∈ H1 (∂Ω, σ) and
define u := Dmod f in Ω. In view of (A.0.89), we see that
 σ(x)   M
f ∈ L 1 ∂Ω, (2.3.194)
1 + |x| n
which goes to show that u is a well-defined function (cf. (1.8.5)-(1.8.6)). In addition,
(1.8.7), (1.8.27), (2.3.3), (2.3.4), and (2.3.5) guarantee that
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t. κ−n.t.
the boundary traces u∂Ω , (∇u)∂Ω exist σ-a.e. on ∂Ω,
κ−n.t.   .p (2.3.195)
u∂Ω = 12 I + Kmod f ∈ H1 (∂Ω, σ) , and
M

. .  M
∂νAu = ∂νA Dmod f ∈ H p (∂Ω, σ) .

Moreover, [70, Theorem 1.4.2] implies that if Ω is an exterior domain we have

⨏
|∇u| dL n = o(1) as R → ∞. (2.3.196)
B(0,2R)\B(0,R)

Granted these properties, Theorem 2.2.7 applies, and according to the integral
representation formula (2.2.168) we may write
 κ−n.t.  . 
u = Dmod u∂Ω − 𝒮 ∂νAu + cu
    .  
= Dmod 12 I + Kmod f − 𝒮 ∂νA Dmod f + cu in Ω, (2.3.197)
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 355

for some C M -valued locally constant function cu in Ω. The idea is to now take non-
κ−n.t.
tangential boundary traces in (2.3.197). With c f := cu ∂Ω , on account of (1.8.27),
(2.2.127) (if n ≥ 3), and (2.2.132) (plus the comment right after (2.2.133), if n = 2),
we obtain
1  1   1    .  
2 I + Kmod f = 2 I + Kmod 2 I + Kmod f − S ∂ν Dmod f + c f
A
(2.3.198)

at σ-a.e. point on ∂Ω. With this in hand, (2.3.190) readily follows, after simple
algebra. This takes care of the claim made in item (1).
Taking weak conormal derivatives in (2.3.197) leads to
 .A  .     .  .   .
∂ν Dmod f = ∂νA Dmod 12 I + Kmod f − ∂νA𝒮 ∂νA Dmod f + ∂νA cu

.        .  
= ∂νA Dmod 12 I + Kmod f − − 12 I + K A#  ∂νA Dmod f , (2.3.199)
.
thanks to (2.3.5), (2.2.47), and the fact that ∂νA cu = 0 (as may be seen from [69,
Definition 10.2.18]). Having established (2.3.199), the claim made in (2.3.191)
readily follows after some simple algebra. The treatment of item (2) is therefore
complete.
Consider next the claims made in items (3)-(4). To set the stage, pick an arbitrary
M
distribution f ∈ H p (∂Ω, σ) and define u := 𝒮 f in Ω. Then Theorem 2.2.3
ensures that u is a well-defined function in Ω satisfying:
M
u ∈ 𝒞∞ (Ω) ,
Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ),
κ−n.t. κ−n.t.
the boundary traces u ∂Ω
, (∇u) ∂Ω
exist σ-a.e. on ∂Ω,
κ−n.t. .p
u∂Ω = S f ∈ H1 (∂Ω, σ) ,
M (2.3.200)
.   M
∂νAu = − 12 I + K A#  f ∈ H p (∂Ω, σ) ,
and (∇u)(x) = O(|x| 1−n ) as |x| → ∞.

See (2.2.32), (2.2.33), (2.2.35), (2.2.44), (2.2.47), (2.2.127) (if n ≥ 3), and (2.2.132)
(plus the comment right after (2.2.133), if n = 2). We should also point out that
the memberships in lines 3-4 above come from (2.3.10) and (2.1.4), respectively.
In particular, in the case when Ω is an exterior domain, the last property listed in
(2.3.200) entails
⨏
|∇u| dL n = o(1) as R → ∞. (2.3.201)
B(0,2R)\B(0,R)

In turn, these properties permit us to invoke Theorem 2.2.7, and the integral repre-
sentation formula (2.2.168) presently gives that there exists some C M -valued locally
constant function cu in Ω such that
356 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 κ−n.t.  . 
u = Dmod u∂Ω − 𝒮 ∂νAu + cu
  
= Dmod (S f ) − 𝒮 − 12 I + K A#  f in Ω. (2.3.202)

κ−n.t.
To proceed, define c f := cu ∂Ω . After taking nontangential boundary traces in
(2.3.202) we arrive at
    
S f = 12 I + Kmod (S f ) − S − 12 I + K A#  f + c f (2.3.203)

at σ-a.e. point on ∂Ω, thanks to (2.3.200), (2.2.127) (if n ≥ 3), and (2.2.132) (plus
the comment right after (2.2.133), if n = 2), also bearing in mind the memberships
 1  M .p M
− 2 I + K A#  f ∈ H p (∂Ω, σ) (see Theorem 2.1.1) and S f ∈ H1 (∂Ω, σ)
(see (2.3.10)). Having proved (2.3.203), the claim in (1.8.252) follows after simple
algebra. This takes care of item (3).
To justify (2.3.193), observe that taking weak conormal derivatives in (2.3.202)
leads to
 1  . . .    .
− 2 I + K A#  f = ∂νAu = ∂νA Dmod (S f ) − ∂νA𝒮 − 12 I + K A#  f + ∂νA cu
.    1  
= ∂νA Dmod (S f ) − − 12 I + K A#  − 2 I + K A#  f , (2.3.204)
.
by virtue of (2.3.200), (2.3.5), (2.2.47), and the fact that ∂νA cu = 0 (as may be seen
from [69, Definition 10.2.18]). At this stage, (2.3.193) follows from (2.3.204) after
canceling like-terms. 

Remark 2.3.12 From Remark 1.8.27, Theorem 2.3.11 (cf. (2.3.190)), and Theo-
rem 2.3.9 we conclude that the modified boundary-to-boundary Cauchy-Clifford
integral operator Cmod (cf. (1.8.112)) is well defined, linear and bounded in the
context
.p .p  
Cmod : H1 (∂Ω, σ) ⊗ Cn −→ H1 (∂Ω, σ) ⊗ Cn for p ∈ n−1 n ,∞ , (2.3.205)

and, if ∂Ω is connected and we let Cmod be the operator [ f ] → Cmod f (with

brackets denoting equivalence classes modulo constants), we have
2 .p  n−1 
Cmod = 14 I on H1 (∂Ω, σ)/∼ ⊗ Cn with p ∈ n ,∞ . (2.3.206)

Below we extend the scope of the intertwining identity proved in Corollary 1.6.8.

Proposition 2.3.13 Let Ω ⊆ R2 be an open set satisfying a two-sided local John

condition and having an Ahlfors regular boundary. Abbreviate σ := H 1 ∂Ω and
denote by ν the geometric measure theoretic outward unit normal to Ω. Let KΔ,mod
be the modified boundary-to-boundary harmonic double layer operator associated
with Ω and L := Δ as in (1.8.24), i.e.,
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 357
∫ 
1 ν(y), y − x
KΔ,mod f (x) := lim+ · 1R2 \B(x,ε) (y)
ε→0 2π |x − y| 2
y ∈∂Ω
|x−y |>ε

ν(y), y
− · 1R2 \B(0,1) (y) f (y) dσ(y) (2.3.207)
|y| 2
 σ(y) 
for each function f ∈ L 1 ∂Ω, 1+ |y | 2
and for σ-a.e. point x ∈ ∂Ω. In addition,
recall from Theorem 2.1.1 that the (two-dimensional) “transpose” principal-value
 1 layer K from (1.6.72) induces a bounded mapping on H (∂Ω, σ)
harmonic double # p

for each p ∈ 2 , 1 .
Then, with the tangential derivative operator

∂τ := ∂τ12 (2.3.208)

defined as in (A.0.183) with j := 1 and k := 2, one has

.p 1
∂τ ◦ KΔ,mod = −K # ◦ ∂τ on H1 (∂Ω, σ) with p ∈ 2, 1 . (2.3.209)

Proof For starters, we note that the hypotheses on the underlying set imply that Ω is
a UR domain (see [68, (5.10.33)]). Let DΔ,mod be the modified boundary-to-domain
harmonic double layer operator associated, in the plane, with Ω and L := Δ as in
(1.8.6), i.e.,
∫  
1 ν(y), y − x ν(y), y
DΔ,mod f (x) := − · 1R2 \B(0,1) (y) f (y) dσ(y)
2π |x − y| 2 |y| 2
∂Ω
 (2.3.210)
σ(y)  −1 ln |x| for
for each f ∈ L 1 ∂Ω, 1+ |y | 2 and each x ∈ Ω. With EΔ (x) := (2π)
x ∈ R2 \ {0} denoting the standard fundamental solution for the Laplacian in the
two-dimensional setting, let us also define the following modified version of the
boundary-to-domain operator (1.6.81):

R mod f (x) (2.3.211)

∫  
:= ν1 (y)(∂2 EΔ )(y − x) − ν2 (y)(∂1 EΔ )(y − x)
∂Ω

 
− ν1 (y)(∂2 EΔ )(y) − ν2 (y)(∂1 EΔ )(y) · 1R2 \B(0,1) (y) f (y) dσ(y)

 σ(y) 
for each f ∈ L 1 ∂Ω, 1+ |y | 2
and each x ∈ Ω. Finally, recall the modified version of
the boundary-to-domain Cauchy integral operator 𝒞mod acting on functions (1.8.226)
 σ(x) 
as in (1.8.227). Since for each real-valued function f ∈ L 1 ∂Ω, 1+ |x | 2
we have
358 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

𝒞mod f is a holomorphic function in Ω,

(2.3.212)
and 𝒞mod f = DΔ,mod f − iR mod f in Ω,

it follows that DΔ,mod f and −R mod f are conjugate harmonic functions in Ω. In turn,
this implies that
 σ(x) 
for each (complex-valued) function f ∈ L 1 ∂Ω, 1+ |x | 2
we have
DΔ,mod f , R mod f ∈ 𝒞∞ (Ω), ΔDΔ,mod f = ΔR mod f = 0 in Ω, (2.3.213)
∂1 DΔ,mod f + ∂2 R mod f = 0 and ∂2 DΔ,mod f − ∂1 R mod f = 0 in Ω.

Also, from Theorem 1.8.12 and Proposition 1.8.16 we know that

   
Nκ ∇DΔ,mod f ∈ L p (∂Ω, σ) and Nκ ∇R mod f ∈ L p (∂Ω, σ)
.p (2.3.214)
for each function f ∈ H1 (∂Ω, σ).

To proceed, recall that the tangential derivative operator ∂τ := ∂τ12 is defined as

in (A.0.183) with j := 1 and k := 2. Consider the boundary-to-domain single layer
potential operator 𝒮 associated with Ω and L := Δ as in (2.2.31).. Finally, recall
p
the duality brackets from [69, Theorem 4.6.1]. Then for each f ∈ H1 (∂Ω, σ) with
1
p ∈ 2 , 1 and each  ∈ {1, 2} we may write
∫ % &
−∂ R mod f (x) = ν1 (y)(∂ ∂2 EΔ )(y − x) − ν2 (y)(∂ ∂1 EΔ )(y − x) f (y) dσ(y)
∂Ω
∫
 
= ∂τ(y) (∂ EΔ )(y − x) f (y) dσ(y)
∂Ω
  
⎧ 
⎨ (∂ EΔ )(x − ·) ∂Ω, ∂τ f
⎪
⎪ if ∂Ω is bounded,
=   
⎪
⎪ (∂ EΔ )(x − ·) , ∂τ f if ∂Ω is unbounded,
⎩ ∂Ω

= ∂ 𝒮(∂τ f )(x) for each x ∈ Ω, (2.3.215)

thanks to (2.3.211), (A.0.183), [69, Lemma 11.10.4], and (2.2.34). Thus,

∇R mod f = −∇𝒮(∂τ f ) in Ω,
.p 1 (2.3.216)
for each f ∈ H1 (∂Ω, σ) with p ∈ 2, 1 .
.p 
Consider now an arbitrary f ∈ H1 (∂Ω, σ) with p ∈ 12 , 1 . Then f belongs to
 σ(x) 
L 1 ∂Ω, 1+ |x | 2
and has ∂τ f ∈ H p (∂Ω, σ). Granted the current geometric assump-
tions on Ω, (2.3.98) (when 12 < p < 1), as well as (2.3.122) and [68, (7.4.105)]
(when p = 1), allow us to conclude that
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 359
.
q
f ∈ Lloc (∂Ω, σ). (2.3.217)
1<q<∞

In concert with [70, (2.5.32)], this further implies that each truncation parameter
ε ∈ (0, ∞) we have
.
Nκε (DΔ,mod f ) ∈
q
Lloc (∂Ω, σ) ⊆ Lloc
1
(∂Ω, σ). (2.3.218)
1<q<∞

Finally, we note from (2.3.214) and [68, (8.6.51)] that

2
∇(DΔ,mod f ) ∈ Lbdd
1
(Ω, L 2 ) , (2.3.219)

and recall from (1.8.27) that we have the jump-formula

  κ−n.t.  1 
DΔ,mod f  = 2 I + KΔ,mod f at σ-a.e. point on ∂Ω. (2.3.220)
∂Ω

Keeping these properties in mind, we may write

  .
− 12 I + K # (∂τ f ) = ∂ν 𝒮(∂τ f ) = ν • ∇𝒮(∂τ f )
 
= −ν • ∇R mod f = −ν • ∂2 DΔ,mod f , −∂1 DΔ,mod f
.   κ−n.t. 
= ∂τ DΔ,mod f = −∂τ DΔ,mod f ∂Ω
1 
= −∂τ 2I + KΔ,mod f in H p (∂Ω, σ). (2.3.221)

Above, the first equality is a consequence of (2.2.47), the second equality comes
from the definition of the weak normal derivative (cf. (A.0.181)-(A.0.182)), the third
equality has been proved in (2.3.216), the fourth equality is contained in (2.3.213),
the fifth equality is seen from [69, Example 10.2.2], the sixth equality is provided
by [69, Proposition 11.3.1] (whose applicability in the present setting is ensured by
(2.3.218), (2.3.219), and (2.3.220)), and the final equality in (2.3.221) uses (2.3.220).
At this stage, the conclusion claimed in (2.3.209) is readily seen from (2.3.221).
Our next proposition contains intertwining identities in the two-dimensional set-
ting, in the spirit of Corollary 1.6.8 and Proposition 2.3.13.

Proposition 2.3.14 Let Ω ⊆ R2 be an open set satisfying a two-sided local John

condition and whose boundary is an Ahlfors regular set. Introduce σ := H 1 ∂Ω
and abbreviate
∂τ := ∂τ12 (2.3.222)
with the latter tangential derivative operator on ∂Ω defined as in (A.0.183) with
j := 1 and k := 2. Lastly, bring back the modified boundary-to-boundary harmonic
double layer operator KΔ,mod from (2.3.207), and recall the “transpose” principal-
value harmonic double layer K # from (1.6.72). Then one has
360 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
.p
∂τ ◦ KΔ,mod = −K # ◦ ∂τ on L1 (∂Ω, σ) with p ∈ (1, ∞). (2.3.223)

In particular, for each p ∈ (1, ∞) one has

∂τ ◦ KΔ,mod = −K # ◦ ∂τ
.p  (2.3.224)
as operators from L1 (∂Ω, σ) ∼ into L p (∂Ω, σ),

where .p 
∂τ : L1 (∂Ω, σ) ∼ −→ L p (∂Ω, σ) is defined as
.p (2.3.225)
∂τ [ f ] := ∂τ f for each f ∈ L1 (∂Ω, σ),
and (compare with (1.8.154))
.p  .p 
KΔ,mod : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ is defined as
.p  .p (2.3.226)
KΔ,mod [ f ] := KΔ,mod f ∈ L1 (∂Ω, σ) ∼ for each f ∈ L1 (∂Ω, σ).

Finally, as a consequence of (2.3.223) and (2.3.209), one also has

.p  
∂τ ◦ KΔ,mod = −K # ◦ ∂τ on H1 (∂Ω, σ) with p ∈ 12 , ∞ . (2.3.227)

Proof. Fix some aperture parameter κ ∈ (0, ∞), and pick an arbitrary function
p
f ∈ L1 (∂Ω, σ) with 1 < p < ∞. Then from Theorem 1.8.12 we know that
 
DΔ,mod f ∈ 𝒞∞ (Ω), Nκ ∇DΔ,mod f ∈ L p (∂Ω, σ) and
κ−n.t.   (2.3.228)
(DΔ,mod f )∂Ω = 12 I + KΔ,mod f at σ-a.e. point on ∂Ω,

and also   κ−n.t.

∇DΔ,mod f ∂Ω exist at σ-a.e. point on ∂Ω. (2.3.229)
In addition, from [69, (11.5.59)] and (1.8.22) we see that

Nκε (DΔ,mod f ) ∈ Lloc (∂Ω, σ) for each ε > 0.

p
(2.3.230)

Then, if EΔ is the standard fundamental solution for the Laplacian in Rn (cf. (A.0.65)),
at σ-a.e. point x ∈ ∂Ω we may compute
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 361
1   κ−n.t. 
∂τ 2I + KΔ,mod f (x) = ∂τ DΔ,mod f ∂Ω (x)
  κ−n.t.    κ−n.t. 
= ν1 (x) ∂2 (DΔ,mod f ) ∂Ω (x) − ν2 (x) ∂1 (DΔ,mod f ) ∂Ω (x)
∫
1
= ν1 (x)2 (∂τ f )(x) − lim+ ν1 (x)(∂2 EΔ )(x − y)(∂τ f )(y) dσ(y)
2 ε→0
y ∈∂Ω
|x−y |>ε
∫
1
+ ν2 (x)2 (∂τ f )(x) − lim+ ν2 (x)(∂1 EΔ )(x − y)(∂τ f )(y) dσ(y)
2 ε→0
y ∈∂Ω
|x−y |>ε

= 12 (∂τ f )(x) − K # (∂τ f )(x). (2.3.231)

The first equality above is a consequence of the second line in (2.3.228). The second
equality above is implied by [69, Proposition 11.3.2], whose present applicability
is guaranteed by (2.3.228)-(2.3.230). The third equality above may be justified with
the help of (1.8.51) and the jump-formula [70, (2.5.341) in Proposition 2.5.35].
The final equality in (2.3.231) is implied by the definition of K # (cf. (1.6.72))
and the fact that ν12 + ν22 = 1 at σ-a.e. point on ∂Ω. At this stage, (2.3.223) follows
readily from (2.3.231), and everything else is a direct consequence of this identity
and definitions.
In closing, we wish to note that an alternative proof of (2.3.223) is obtain by
combining the version of the formula (1.6.74) recorded in (1.6.77) together with [69,
(11.5.161)], and keeping in mind (1.8.28) plus item (5) in [69, Lemma 11.1.3]. 

Availing ourselves of the intertwining formulas established in Proposition 2.3.14

we may further augment Proposition 1.8.10 as follows:

Corollary 2.3.15 Let Ω ⊆ R2 be a chord-arc domain with unbounded boundary

and abbreviate σ := H 1 ∂Ω. Recall the transpose harmonic double layer K # from
(1.6.72), and bring in the modified boundary-to-boundary harmonic double layer
operator KΔ,mod acting as in (2.3.226). Then for each p ∈ (1, ∞) and each z ∈ C it
follows that

the operator zI + K # is an isomorphism on L p (∂Ω,

. pσ) if and only if (2.3.232)
the operator −zI + KΔ,mod is an isomorphism on L1 (∂Ω, σ) ∼.

Proof This is a direct consequence of the intertwining formula (2.3.224) and the
isomorphism proved in [69, Proposition 11.5.15]. 

We next take up the task of establishing the analogue of the integral representation
formula (1.3.35) from Theorem 1.3.3 for functions
 u satisfying, among other things,
Nκ (∇u) ∈ L p (∂Ω, σ) with p ∈ n−1 n , n − 1 . The chief novel difficulty is that now
Nκ (∇u) may not even be locally integrable, so the hypotheses of Theorem 1.3.3 are
clearly violated. Related results are contained in Theorem 2.2.7 and Theorem 1.8.19.
362 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

Before stating our new result, the reader is reminded that the maximal operator P
has been defined in (A.0.173).

Theorem 2.3.16 Suppose Ω ⊆ Rn , where n ∈ N satisfies12 n ≥ 3, is an Ahlfors

regular domain. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
 αβ  unit normal to Ω. Also, for some M ∈ N, consider a coefficient
theoretic outward
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). In this setting, consid-
er the potential operators D, 𝒮, ΠΩ , associated with A and Ω as in (1.3.18), (1.3.6),
1,1 M
(1.3.4). Suppose u ∈ Wloc (Ω) is a vector-valued function with the property that,
for some aperture parameter κ ∈ (0, ∞), satisfies
κ−n.t.
the trace u∂Ω exists at σ-a.e. point on ∂Ω,
∫
(Nκ u)(y) (2.3.233)
dσ(y) < +∞ for some δ > 0,
∂Ω 1 + |y|
n−1−δ
 
and Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1
n ,n−1 .

In addition, with Lu considered in the sense of distributions in Ω, assume that

M
Lu ∈ Lloc
1
(Ω, L n ) and P(Lu) ∈ L p (∂Ω, σ). (2.3.234)

Finally, fix a point x ∈ Rn \ ∂Ω with the property that

∫
|(Lu)(y)|
dy < +∞. (2.3.235)
Ω |x − y| n−2
.
Then the weak conormal derivative ∂νAu, defined as in [69, Theorem 10.2.24,
M
(10.2.148)], belongs to the Hardy space H p (∂Ω, σ) and one has (with the single
layer operator acting on distributions in the Hardy space as in (2.2.31))

 κ−n.t. 

.A u(x) if x ∈ Ω,
D u ∂Ω (x) − 𝒮(∂ν u)(x) + ΠΩ (Lu)(x) = (2.3.236)
0 if x ∈ Rn \ Ω.

if either ∂Ω is unbounded, or Ω is bounded. In the case when Ω is an exterior

domain, the same conclusion is valid under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (2.3.237)
B(0,λ R)\B(0,R)

In fact, a similar result is valid when n = 2 provided either

12 see the last claim in the statement for n = 2

2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 363
∫
−1
L(ξ) dH 1 (ξ) = 0 ∈ C M×M , (2.3.238)
S1

or otherwise one assumes that ∂Ω is compact, replaces (2.3.235) by

∫
  
|(Lu)(y)| 1 +  ln |x − y|  dy < +∞, (2.3.239)
Ω

and, in the case when Ω is an exterior domain, replaces (2.3.237) by

⨏
 
|u| dL 2 = o ln1R as R → ∞. (2.3.240)
B(0,λ R)\B(0,R)

We note that when n = 2 and Ω is an exterior domain, (2.3.240) is implied by

the weaker decay condition (2.3.237), provided (2.3.234)-(2.3.235) are replaced by
Lu = 0 in Ω. Indeed, this is seen from [70, (1.5.205), (1.5.213)]. We also wish to
note that [68, Lemma 7.2.1] and Hölder’s inequality imply that
the integrability condition from the second line of (2.3.233) is
(2.3.241)
automatically satisfied if Nκ u ∈ L q (∂Ω, σ) for some q ∈ [1, ∞).

Proof of Theorem 2.3.16 Assume first that n ≥ 3. For starters, (2.3.233) and [68,
Lemma 8.3.1] imply that

the function u is locally Lipschitz in Ω. (2.3.242)

In addition, from the first two lines in (2.3.233), [68, (8.9.8)], [68, Proposition 8.8.4],
[68, Corollary 8.9.6], and the fact that H n−1 (∂Ω \ ∂∗ Ω) = 0 we conclude that
κ−n.t.   σ(y)   M
u∂Ω ∈ L 1 ∂Ω, . (2.3.243)
1 + |y| n−1
Moreover, from [69, (10.2.182)] and (2.3.233)-(2.3.234) we know that
. M
∂νAu ∈ H p (∂Ω, σ) . (2.3.244)

We find it convenient to divide the remaining portion of the proof into two cases,
depending on the size of p.

Case I: Suppose p ∈ n−1 n , 1 . Assume for now that ∂Ω is unbounded, and fix a
point x ∈ Ω for which (2.3.235) holds. Also, select an arbitrary index α ∈ {1, . . . , M }.
Consider a scalar-valued function η ∈ 𝒞∞ (Rn ) with
 the property that η = 0 on B(0, 1)
and η = 1 on Rn \ B(0, 2). For each number ε ∈ 0, 12 dist(x, ∂Ω) define ηε : Rn → R
by setting
y − x
ηε (y) := η for every y ∈ Rn . (2.3.245)
ε
Then ηε ∈ 𝒞∞ (Rn ) is a bounded function satisfying
364 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

lim ηε (y) = 1 for every y ∈ Rn \ {x}, (2.3.246)

ε→0+
 
and there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we
have

1 − ηε ∈ 𝒞∞
c (Ω), ηε ≡ 0 on B(x, ε), supp (∇ηε ) ⊆ B(x, 2ε) \ B(x, ε),
and |(∇ j η ε )(y)| ≤ Cε −j for every j ∈ N0 and every y ∈ Rn .
(2.3.247)

In particular, for each β ∈ {1, . . . , M } the function Eαβ (x − ·)ηε belongs to 𝒞∞ (Rn )
and coincides with Eαβ (x − ·) near ∂Ω.
To proceed, fix ε ∈ 0, 12 dist(x, ∂Ω) and observe that according to (2.2.31) we
have (with the summation convention over repeated indices in effect)
 .A  #  . $
𝒮∂ν u α (x) = Eαβ (x − ·)∂Ω , (∂νAu)β
#   . $
= Eαβ (x − ·)ηε ∂Ω , (∂νAu)β (2.3.248)

where ·, · stands for the duality bracket on ∂Ω, described in [69, Theorem 4.6.1]
(with Σ := ∂Ω). Let us also fix some number λ ∈ (1, ∞) and select a function
θ ∈ 𝒞∞c (R ) satisfying 0 ≤ θ ≤ 1, θ ≡ 1 on B(0, 1), θ ≡ 0 on R \ B(0, λ) and, for
n n

each R > 0, define θ R (x) := θ(x/R) for every x ∈ R . For each β ∈ {1, . . . , M } we
n

then have
    
lim θ R Eαβ (x − ·)ηε ∂Ω = Eαβ (x − ·)ηε ∂Ω
R→∞

⎧ . 
⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼ if p < 1,
⎪ (2.3.249)
weak-∗ in
⎪ 
⎪ BMO(∂Ω,
⎩ σ) if p = 1.

This is implied by the general weak-∗ convergence results established in [69, Lem-
ma 4.8.4] and, respectively, [69, Lemma 4.8.1] (in the latter case also bearing in
mind the trivial bounded embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)). At this stage,
from (2.3.248), (2.3.249), (2.3.244), the duality result from [69, Theorem 4.6.1], and
[69, Lemma 4.6.4] we conclude that
 .A  # .   $
𝒮∂ν u α (x) = lim (Lipc (∂Ω)) (∂νAu)β, θ R Eαβ (x − ·)ηε ∂Ω Lipc (∂Ω) . (2.3.250)
R→∞

Pressing on, recall from (A.0.181)-(A.0.182) that the weak conormal derivative of
u associated with the coefficient tensor A is defined as the distribution
.      M
∂νAu := ν • Fγ 1≤γ ≤M in Lipc (∂Ω) where
  (2.3.251)
γβ
Fγ := (A∇u)γ = ar s ∂s uβ 1≤r ≤n for each γ ∈ {1, . . . , M }.
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 365

Based on (2.3.250), (2.3.251), and [68, Proposition 4.2.3] we obtain

∫
 .A 
𝒮∂ν u α (x) = lim (divFγ )θ R Eαγ (x − ·)ηε dL n
R→∞ Ω
∫
  
+ lim Fγ, ∇ θ R Eαγ (x − ·)ηε dL n
R→∞ Ω
∫
 
= lim Lu , θ R Eα .(x − ·)ηε dL n
R→∞ Ω
∫
γβ  
+ lim ar s (∂s uβ )∂r θ R Eαγ (x − ·)ηε dL n . (2.3.252)
R→∞ Ω

Thanks to (2.3.235), [70, (1.4.24)], and Lebesgue’s Dominated Convergence Theo-

rem,
∫
   
lim+ lim Lu, θ R Eα .(x − ·)ηε dL n = ΠΩ (Lu) α (x). (2.3.253)
ε→0 R→∞ Ω

Together, (2.3.252) and (2.3.253) permit us to conclude that

 .   
𝒮∂νAu α (x) − ΠΩ (Lu) α (x) (2.3.254)
∫
γβ  
= lim+ lim (∂s uβ )ar s ∂r θ R Eαγ (x − ·)ηε dL n .
ε→0 R→∞ Ω
 n−1
Recall that θ R E(x − ·)ηε ∈ 𝒞∞
c (R ) and that Nκ (∇u) ∈ L (∂Ω, σ) with p ∈
n p
n ,1
implies, via [68, (8.6.51)], that ∇u ∈ 1 (Ω, L n ) M ·n .
Lbdd
Also, the integrability
condition in the second line of (2.3.233) implies that Nκ u ∈ Lloc
1 (∂Ω, σ). Bearing

these in mind and relying on the integration by parts formula established in [68,
Theorem 1.7.1], we may further transform this formula into
 .   
𝒮∂νAu α (x) − ΠΩ (Lu) α (x)
∫
γβ  
= − lim+ lim uβ ar s ∂r ∂s θ R Eαγ (x − ·)ηε dL n
ε→0 R→∞ Ω
∫
 κ−n.t.  γβ  
+ lim νs uβ ∂Ω ar s ∂r θ R Eαγ (x − ·) dσ
R→∞ ∂Ω

=: lim+ Iε + II. (2.3.255)

ε→0

Using the product rule for the derivative ∂r yields

366 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
∫
 κ−n.t.  γβ
II = lim νs uβ ∂Ω ar s (∂r θ R )Eαγ (x − ·) dσ
R→∞ ∂Ω
∫
 κ−n.t.  γβ
− lim νs uβ ∂Ω ar s θ R (∂r Eαγ )(x − ·) dσ
R→∞ ∂Ω

(1)
=: II − II(2) . (2.3.256)

Based on [70, (1.4.24)] plus the fact that for every j ∈ N we have

sup |∇ j θ R | ≤ CR−j and supp(∇ j θ R ) ⊆ B(0, λ R) \ B(0, R) for each R > 0,

Rn
(2.3.257)

we may estimate
∫  κ−n.t.  γβ 
 
 νs uβ ∂Ω ar s (∂r θ R )Eαγ (x − ·) dσ 
∂Ω
∫   κ−n.t.  
 u (y)
∂Ω
≤C dσ(y)
∂Ω∩[B(0,λ R)\B(0,R)] 1 + |y| n−1
= o(1) as R → ∞, (2.3.258)

where the last equality is a consequence of Lebesgue’s Dominated Convergence

Theorem, whose applicability is ensured by (2.3.243). Thus,

II(1) = 0. (2.3.259)

In view of (1.3.18), from (2.3.243) and Lebesgue’s Dominated Convergence Theo-

rem we also conclude that
∫   κ−n.t.  
 κ−n.t.  γβ
II(2) = νs uβ ∂Ω ar s (∂r Eαγ )(x − ·) dσ = − D u∂Ω (x). (2.3.260)
∂Ω α

γβ
Going further, recall that ar s (∂r ∂s Eαγ )(x − ·) = 0 in Rn \ {x} (cf. [70, (1.4.33)])
for each index β ∈ {1, . . . , M }. We split Iε from (2.3.255) as

Iε = I(1) + I(2) + I(3) + I(4) (5) (6)

ε + Iε + Iε (2.3.261)

where
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 367
∫
γβ
I(1) := lim uβ ar s (∂r θ R )(∂s Eαγ )(x − ·) dL n,
R→∞ Ω
∫
γβ
I(2) := lim uβ ar s (∂s θ R )(∂r Eαγ )(x − ·) dL n,
R→∞ Ω
∫
(3) γβ
I := − lim uβ ar s (∂r ∂s θ R )Eαγ (x − ·) dL n,
R→∞ Ω
∫
γβ
I(4)
ε := lim uβ ar s θ R (∂r Eαγ )(x − ·)∂s ηε dL n,
R→∞ Ω
∫
γβ
I(5)
ε := lim uβ ar s θ R (∂s Eαγ )(x − ·)∂r ηε dL n,
R→∞ Ω
∫
γβ
I(6)
ε := − lim uβ ar s θ R Eαγ (x − ·)∂r ∂s ηε dL n . (2.3.262)
R→∞ Ω

Note that since we are presently assuming ∂Ω to be unbounded, estimate [68,

(8.6.51)] is applicable with ER := Ω ∩ B(0, λ R) \ B(0, R) and p := (n − 1)/n, and
this gives

u L 1 (E R , L n ) ≤ C Nκ u L (n−1)/n (πκ (E R ),σ), ∀R > 0. (2.3.263)

On the other hand, based on Hölder’s inequality, [68, (8.1.17)], and the upper-Ahlfors
regularity of ∂Ω we may estimate

Nκ u L (n−1)/n (πκ (E R ),σ)

∫ n−1
 n−1
n

= (Nκ u)(y) n dσ(y)

πκ (E R )
/∫ 0 n−1
n
 (N u)(y)  n−1   n−1
κ n
= · 1 + |y| n−1−δ n dσ(y)
πκ (E R ) 1 + |y|
n−1−δ

∫  ∫   n−1
(Nκ u)(y) n−1−δ  n−1
1

≤ dσ(y) · 1 + |y| dσ(y)

πκ (E R ) 1 + |y|
n−1−δ
πκ (E R )
 ∫ 
(Nκ u)(y)
≤ CRn−δ dσ(y) . (2.3.264)
∂Ω 1 + |y|
n−1−δ

Collectively, (2.3.262), (2.3.257), [70, (1.4.24)], (2.3.263), (2.3.264), and (2.3.233)

imply
368 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
% ∫ &
 (1) 
I  ≤ C lim sup R−n |u| dL n
R→∞ Ω∩[B(0,λ R)\B(0,R)]
% &
= C lim sup R−n u L 1 (E R , L n )
R→∞
∫ (Nκ u)(y) 
≤C dσ(y) · lim sup R−δ = 0 (2.3.265)
∂Ω 1 + |y| n−1−δ R→∞

hence

I(1) = 0. (2.3.266)

Moreover, in a completely analogous manner we also have

I(2) = 0 and I(3) = 0. (2.3.267)

Next, we decompose
∫
γβ
I(4)
ε = uβ (y)ar s (∂r Eαγ )(x − y)(∂s ηε )(y) dy = I(4a)
ε + I(4b)
ε (2.3.268)
Ω

where
∫
  γβ
I(4a)
ε := uβ (y) − uβ (x) ar s (∂r Eαγ )(x − y)(∂s ηε )(y) dy,
Ω ∫ (2.3.269)
γβ
I(4b)
ε := uβ (x) ar s (∂r Eαγ )(x − y)(∂s ηε )(y) dy.
Ω

In relation to I(4a)
ε observe that (2.3.242) permits us to estimate
⨏
   
lim sup I(4a)
ε
 ≤ C · lim sup u(y) − u(x) dy = 0. (2.3.270)
ε→0+ ε→0+ B(x,2ε)

Also, [70, (1.4.33)] eventually allows us to conclude that for each β ∈ {1, . . . , M }
we have
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 369
∫
γβ
lim ar s (∂r Eαγ )(x − y)(∂s ηε )(y) dy
ε→0+ Ω
∫
γβ
= lim+ ar s (∂r Eαγ )(x − y)∂s (ηε − 1)(y) dy
ε→0 Ω
# $
γβ
= − lim+ D  (Ω) ar s ∂r Eαγ (x − ·) , ∂s (ηε − 1) D(Ω)
ε→0
# $
γβ
= lim+ D  (Ω) ar s ∂s ∂r Eαγ (x − ·) , ηε − 1 D(Ω)
ε→0

= lim+ δαβ (ηε − 1)(x) = −δαβ . (2.3.271)

ε→0

By combining (2.3.268)-(2.3.271) we then arrive at the conclusion that

lim I(4)
ε = −uα (x). (2.3.272)
ε→0+

In a completely similar fashion, we also have

lim I(5) (6)

ε = −uα (x) and lim+ Iε = uα (x). (2.3.273)
ε→0+ ε→0

Bringing together (2.3.255), (2.3.256), (2.3.259), (2.3.260), (2.3.261), (2.3.266),

(2.3.267), (2.3.272), and (2.3.273) now yields the formula
 .A      κ−n.t.  
𝒮∂ν u α (x) − ΠΩ (Lu) α (x) = D u∂Ω (x) − uα (x). (2.3.274)
α

Given that α ∈ {1, . . . , M } is arbitrary, this establishes (2.3.236) in the case when
∂Ω is unbounded and x ∈ Ω.
In the case when ∂Ω is unbounded and x ∈ Rn \ Ω, we reason in a completely
similar manner. Keeping in mind that this time we actually have ηε ≡ 1 in Ω, it is
clear from (2.3.262) that we now have I(4) (5) (6)
ε = Iε = Iε = 0. Ultimately, this explains
why the right-hand side of (2.3.236) is zero in this case. Next, the case when Ω is
bounded (and x ∈ Rn \ ∂Ω) is treated, mutatis mutandis, identically, since (2.3.263)
remains valid in such a scenario.
To complete the treatment in Case I, there remains to consider the case when
Ω is an exterior domain, when estimate (2.3.263) may fail to hold with C ∈ (0, ∞)
independent of R (see the nature of the constant involved in [68, (8.6.51) in Proposi-
tion 8.6.3]). However, in such a situation the decay condition (2.3.237) turns out to
be a good substitute, as this once again implies (as in the first line of (2.3.265)) that
(2.3.266)-(2.3.267) hold.
Case II: Suppose p ∈ (1, n − 1). The argument proceeds largely as before, with
some natural alterations. First, (2.3.244) becomes
. M
∂νAu ∈ L p (∂Ω, σ) , (2.3.275)
370 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

and in place of (2.3.248) we now have

∫
 .A  .
𝒮∂ν u α (x) = Eαβ (x − ·)ηε (∂νAu)β dσ. (2.3.276)
∂Ω

The integral in (2.3.276) is absolutely convergent since if p∫ is the Hölder conjugate

exponent of p, then the membership p ∈ (1, n − 1) implies ∂Ω |E(x − ·)| p dσ < +∞
thanks to [70, (1.4.24)] and [68, (7.2.5)]. This observation also permits us to write
∫
 .A  .
𝒮∂ν u α (x) = lim θ R Eαβ (x − ·)ηε (∂νAu)β dσ, (2.3.277)
R→∞ ∂Ω

which shows that (2.3.250) is valid in the present case as well. From this point on,
the proof proceeds as before, and the desired conclusion follows.
In fact, the same argument as above works when n = 2 provided either (2.3.238)
holds, or otherwise we assume that ∂Ω is compact, replace (2.3.235) by (2.3.239)
and, in the case when Ω is an exterior domain, replace [70, (1.5.5)] by (2.3.240) (see
[70, (1.4.24)]). This completes the proof of Theorem 2.3.16. 

A consequence of Theorem 2.3.16 worth stating as a stand-alone result is the

integral representation formula in the corollary below.

Corollary 2.3.17 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain with the

property that ∂Ω is a compact Ahlfors regular set. Denote by ν the geometric measure
theoretic outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, for some
M ∈ N, let L be a weakly elliptic homogeneous second-order M × M system in Rn ,
with constant complex coefficients. Next, fix an aperture parameter κ ∈ (0, ∞) and
consider a function u : Ω → C M satisfying
M
u ∈ 𝒞∞ (Ω) ,
Lu = 0 in Ω, and
  (2.3.278)
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1
n ,n−1 .

Finally, in the case when Ω is an exterior domain make the additional assumption
that there exists λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (2.3.279)
B(0,λ R)\B(0,R)
 αβ 
Then for each complex coefficient tensor A = ar s 1≤r,s ≤n with the proper-
.A 1≤α,β ≤M
ty that L = L A the weak conormal derivative ∂ν u belongs to the Hardy space
κ−n.t.
H p (∂Ω, σ) , the nontangential boundary trace u∂Ω exists at σ-a.e. point on
M

∗ M  
1 −1
∂Ω and, as a function, belongs to L p (∂Ω, σ) where p∗ := p1 − n−1 ∈ (1, ∞).
With these interpretations, the following integral representation formula holds:
 κ−n.t.  . 
u = D u∂Ω − 𝒮 ∂νAu in Ω, (2.3.280)
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 371

where D is the double layer potential operator associated with the coefficient tensor
A and the domain Ω as in (1.3.18), and where the single layer potential operator
M
acts on distributions in the Hardy space H p (∂Ω, σ) as in (2.2.31), (1.3.6).

Proof Recall that any NTA domain with an Ahlfors regular boundary is an Ahlfors
regular domain (cf. (A.0.1) and [68, (5.2.4)]). Since ∂Ω is bounded, Theorem 2.2.7
κ−n.t.
and Theorem 1.8.19 enure that the nontangential boundary trace u∂Ω exists at
∗
σ-a.e. point on ∂Ω, and Nκ u ∈ L p (∂Ω, σ). In particular, all conditions in (2.3.233)
are satisfied. Granted this, (2.3.280) follows from Theorem 2.3.16 (see also the first
comment following its statement). 

Having established in Theorem 2.3.1 mapping properties for boundary  layer

q, p
potentials involving the Hardy-based Sobolev spaces H1 (∂Ω, σ) with p ∈ n−1 n ,1
and q ∈ [1, ∞] from [69, Definition 11.10.6], the goal of our next theorem in this
section is to extend the operator identities from item (xiii) in Theorem 1.5.1 to the
latter scale (for related results, see also Theorem 2.3.11).

Theorem 2.3.18 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. In the

case when n = 2 make the additional assumption that ∂Ω is compact. Abbreviate
σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward
 αβ  unit normal
to Ω. Also, for some M ∈ N, consider a coefficient tensor A = ar s 1≤r,s ≤n with
1≤α,β ≤M
complex entries, with the property that the M × M homogeneous second-order system
L = L A associated with A in Rn as in (1.3.2) is weakly elliptic (in the sense of [70,
.
(1.3.3) in Definition 1.3.1]). In this setting, consider the boundary layer potential
operators S, ∂νA D, K, associated with A and Ω as in Theorem 2.3.1, and let K A# 
be the operator associated with A and Ω as in Theorem 2.1.1. Then for any two
exponents, q ∈ (1, ∞) and
  n−1
 n−1  n−1  n , 1 if n ≥ 3,
p ∈ n ,1 ∩ n ,n−1 =   (2.3.281)
1
2 , 1 if n = 2,

the following operator identities hold:

1    .  q, p M
2I + K ◦ − 12 I + K = S ◦ ∂νA D on H1 (∂Ω, σ) , (2.3.282)
1    .  M
2I + K A#  ◦ − 12 I + K A#  = ∂νA D ◦ S on H p (∂Ω, σ) , (2.3.283)
M
S ◦ K A#  = K ◦ S on H p (∂Ω, σ) , (2.3.284)
.  .  q, p M
K A#  ◦ ∂νA D = ∂νA D ◦ K on H1 (∂Ω, σ) . (2.3.285)

Henceforth, strengthen the original geometric assumptions, now demanding that

Ω is an open set in Rn satisfying a two-sided local John condition with an Ahlfors
regular boundary. In the two-dimensional setting, also assume that ∂Ω is compact.
.p M
Then for each f ∈ H1 (∂Ω, σ) there exists c f , which is the nontangential trace
372 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces

on ∂Ω of some C M -valued locally constant function in Ω, with the property that at

σ-a.e. point on ∂Ω one has
1   1    .  
2 I + Kmod − 2 I + Kmod f = S ∂νA Dmod f + c f , (2.3.286)
.
with S as in (2.3.10), ∂νA Dmod as in (2.3.5), and Kmod as in (2.3.184). In addition, for
.p M
each function f ∈ H1 (∂Ω, σ) one has
 .A   . 
∂ν Dmod Kmod f = K A#  ∂νA Dmod f at σ-a.e. point on ∂Ω. (2.3.287)

q, p M
Proof To prove (2.3.282), pick a function f ∈ H1 (∂Ω, σ) and define u := D f
in Ω. From (2.3.4), (1.5.1), (1.5.20), (1.3.24), and (2.3.241), it follows that u satisfies
the hypotheses of Theorem 2.3.16 (including (2.3.237) in the case when Ω is an
exterior domain). In view of (1.5.20) and (1.3.24), the integral representation formula
(2.3.236) presently becomes
    . 
D f = D 12 I + K f − 𝒮 (∂νA D) f in Ω. (2.3.288)

Upon recalling (1.5.20), (2.3.6), (2.2.127) (also keeping in mind (A.0.1)), and taking
nontangential traces then yields
1      . 
2I + K f = 12 I + K 12 I + K f − S (∂νA D) f at σ-a.e. point on ∂Ω. (2.3.289)

After some simple algebra this proves (2.3.282).

M
Moving on, select f ∈ H p (∂Ω, σ) and define u := 𝒮 f in Ω. From (2.2.39)-
(2.2.41), (2.2.44), (2.2.127), (2.2.132), (1.3.8), and (2.3.241), it follows that u satisfies
the hypotheses of Theorem 2.3.16 in the case when either Ω is bounded or ∂Ω is
unbounded. Thanks to (2.2.127) (also keeping in mind (A.0.1)), (2.2.33), (2.2.47),
and (2.1.4), the integral representation formula (2.3.236) currently reads
  
𝒮 f = D(S f ) − 𝒮 − 12 I + K A#  f in Ω. (2.3.290)

Upon recalling (2.2.47), (2.3.12), and (2.3.6), taking the weak conormal derivative
.
∂νA of both sides of (2.3.290) then shows that
  .    
− 12 I + K A#  f = ∂νA D (S f ) − − 12 I + K A#  − 12 I + K A#  f (2.3.291)
M
in H p (∂Ω, σ) . After some simple algebra, this establishes (2.3.283) in the case
when either Ω is bounded or ∂Ω is unbounded. Also, if staring with (2.3.290) we now
take nontangential traces then, on account of (1.5.20) and (2.2.127) (also bearing in
mind (A.0.1)), we arrive at
    
S f = 12 I + K (S f ) − S − 12 I + K A#  f at σ-a.e. point on ∂Ω, (2.3.292)
2.3 Integral Operators of Layer Potential Type on Hardy-Based . . . 373

from which (2.3.284) follows in the case when either Ω is bounded or ∂Ω is un-
bounded, after some simple algebra. Finally, when Ω is an exterior domain, we may
prove (2.3.283) and (2.3.284) in a similar fashion with Ω replaced by Ω− := Rn \ Ω.
q, p M
Consider (2.3.285). Given any f ∈ H1 (∂Ω, σ) , write (2.3.288) then apply
.A
the weak conormal derivative operator ∂ν to both sides to obtain
 .A  .     . 
∂ν D f = ∂νA D 12 I + K f − − 12 I + K A#  (∂νA D) f , (2.3.293)

on account of (2.3.6) and (2.2.47). From this, (2.3.285) readily follows. This takes
care of all claims in the first half of the statement of the theorem.
To deal with the claims in the second half of the statement, assume now that Ω
is an open set in Rn satisfying a two-sided local John condition whose boundary
is Ahlfors regular. In the two-dimensional setting make the additional assumption
that ∂Ω is compact. Finally, suppose p is an in (2.3.281). The latter condition
 
1 −1
enures that p∗ := p1 − n−1 is a well-defined number, belonging to (1, ∞). Fix
.p M
an arbitrary function f ∈ H1 (∂Ω, σ) . From Theorem 2.3.3 we know that there
exists c0 ∈ C M with the property that
∗

f := f − c0 ∈ L p (∂Ω, σ)
M
. (2.3.294)

Define
u := D 
f in Ω, (2.3.295)
and note that, by virtue of (2.3.294) and (1.8.10), we may recast this as

u = Dmod f + c1 for some constant c1 ∈ C M . (2.3.296)

As a consequence of (2.3.294)-(2.3.296), (1.3.24), items (i) and (iv) in Theorem 1.5.1,

as well as items (i) and (iv) in Theorem 2.3.1, for any aperture parameter κ > 0 we
have
M
u belongs to 𝒞∞ (Ω) and Lu = 0 in Ω,
Nκ (∇u) belongs to the space L p (∂Ω, σ),
∗
.  σ(x) 
Nκ u ∈ L p (∂Ω, σ) ⊆ L 1 ∂Ω, , (2.3.297)
δ>0
1 + |x| n−1−δ
1  κ−n.t.  1 
 
2 I + K f = u ∂Ω = 2 I + Kmod f + c1,
. .
and ∂νAu = (∂νA Dmod ) f .
In the case when Ω is an exterior domain, we also have
⨏
|u| dL n = o(1) as R → ∞. (2.3.298)
B(0,2R)\B(0,R)

Granted these properties, Theorem 2.3.16 applies, and from (2.3.236) and (2.3.297)
we see that
374 2 Layer Potential Operators on Hardy, BMO, VMO, and Hölder Spaces
 κ−n.t.  .
u = D u∂Ω − 𝒮(∂νAu) in Ω. (2.3.299)

Note that for some constants c2, c3, c4 ∈ C M we may write

 κ−n.t.       
D u∂Ω = D 12 I + K  f = Dmod 12 I + K  f + c2 (2.3.300)
     
= Dmod 1
2I + Kmod 
f + c3 = Dmod 12 I + Kmod f + c4 in Ω,

thanks to (1.8.10), (1.8.28), and (2.3.297). After combining (2.3.299) with (2.3.300)
and the last property in (2.3.297) we arrive at
   .
u = Dmod 12 I + Kmod f − 𝒮(∂νA Dmod ) f + c4 in Ω. (2.3.301)

Going nontangentially to the boundary in (2.3.301) and making use of (2.3.297),

(2.3.184), (A.0.89), (1.8.27), (2.3.5), and (2.2.126), further yields
1  1   1   .
2 I + Kmod f + c1 = 2 I + Kmod 2 I + Kmod f − S(∂νA Dmod ) f + c4, (2.3.302)

from which (2.3.286) follows with c f := c1 − c4 ∈ C M . Applying the weak conormal

.
derivative ∂νA to all terms in (2.3.301) and relying on (2.3.297), (2.3.184), (2.3.5),
and (2.2.47) produces
. .      .
(∂νA Dmod ) f = (∂νA Dmod ) 12 I + Kmod f − − 12 I + K A#  (∂νA Dmod ) f , (2.3.303)

which readily justifies (2.3.287). 

Chapter 3
Layer Potential Operators on Calderón,
Morrey-Campanato, and Morrey Spaces

The reader is reminded that Morrey-Campanato spaces, Morrey spaces, and their pre-
duals on Ahlfors regular sets in Rn have been discussed at length in [69, Chapter 6].
The main aim in this chapter is to study singular integral operators of boundary
layer type on Calderón spaces (cf. §3.1), Morrey-Campanato spaces (cf. §3.2), and
Morrey spaces (cf. §3.3), on domains with uniformly rectifiable boundaries.
In §3.1 we introduce Calderón spaces (aka Cpα -spaces) on Ahlfors regular sets, by
requiring the membership of a “fractional” Fefferman-Stein sharp maximal function
(obtained by modifying the standard version, used in connection with the John-
Nirenberg space BMO, as indicated in (3.1.1)) to a Lebesgue space. See (3.1.10)
and (3.1.14) in this regard. In the entire Euclidean space, such Calderón spaces have
been studied at length in [23], [14], [6], [80], [94] and the references included there.
The main novel aspect of our analysis is the consideration of the action of singular
integral operators of boundary layer type on Calderón spaces defined on uniformly
rectifiable sets. Our main result in this regard is Theorem 3.1.1, whose proof makes
use of cancelation properties specific to (modified) double layer operators which, in
turn, are a consequence of the Divergence Theorem.
Boundary-to-boundary double layer potential operators are studied on Morrey-
Campanato spaces in §3.2. The starting point is establishing the boundedness result
described in Theorem 3.2.1 for the transpose double layer K # associated with a given
weakly elliptic system on the boundary of an arbitrary UR domain, acting on the
pre-dual of the Morrey-Campanato space introduced in [69, Chapter 6]. The proof
of Theorem 3.2.1 makes essential use of the atomic/molecular theory developed in
[69, Chapter 6] in relation to these pre-dual spaces. In the process of proving that K #
maps atoms to molecules, the full force of our Divergence Theorem from [68, §1.2]
is required. With this in hand, a duality argument then gives the boundedness of
double layer potential operators K on Morrey-Campanato spaces; see Theorem 3.2.2
for details.
Finally, in §3.3 we take up the task of studying the action of singular integral
operators of layer potential type on Morrey spaces and their pre-duals, considered
on boundaries of uniformly rectifiable domains. At its core, the approach we adopt in
this endeavor rests on two basic aspects. First, in earlier work special care has been
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 375
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_3
376 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

taken to establish integral identities and operator identities in a sufficiently general

functional analytic setting (i.e., involving spaces of functions in which Morrey spaces
embed naturally). See [70, Theorem 1.7.10], Proposition 1.2.1, Proposition 1.2.2,
Lemma 1.3.2, along with the jump-formulas (1.5.20) and (1.5.58) in Theorem 1.5.1,
in this regard. Second, in [70, Theorem 2.6.1], [70, Proposition 2.6.2], and [70,
Proposition 2.6.3] we have proved norm estimates in Morrey spaces and their pre-
duals for generic convolution-type singular integral operators. In turn, these garden-
variety singular integral operators serve as building blocks for the algebraically more
sophisticated boundary layer potential operators we presently consider. In concert,
these two aspects work with great efficiency and yield a satisfactory theory for
boundary layer potential operators acting on Morrey spaces and their pre-duals,
considered on uniformly rectifiable sets.
More specifically, in Theorem 3.3.1 we prove boundedness properties, jump-
relations, and nontangential maximal function estimates for boundary layer potential
operators on Morrey and block spaces in arbitrary UR domains. In Theorem 3.3.2
we further extend the scope of this theorem by now allowing our layer potential
operators to act on Morrey-based Sobolev spaces. Any boundary-to-domain version
of the double layer potential operator associated with a given weakly elliptic system
and UR domain with compact boundary is shown to satisfy certain fractional Car-
leson measure estimates when acting on Morrey spaces, of the sort established in
Theorem 3.3.3. Next, Theorem 3.3.5 deals with the modified single layer potential
operator mapping Morrey and block spaces into Morrey-based and, respectively,
block-based Sobolev spaces. Mapping properties for modified boundary-to-domain
double layer potential operators involving Morrey-based and block-based Sobolev
spaces are subsequently treated in Theorem 3.3.6. In Theorem 3.3.8 we then estab-
lish mapping properties for modified boundary-to-boundary double layer potential
operators acting on the scales of Morrey-based and block-based Sobolev spaces.
Moving on, in Theorem 3.3.10 we prove a very useful integral representation
formula for null-solutions of weakly elliptic second-order systems in NTA domains
with Ahlfors regular boundaries with the property that the nontangential maximal
operator of their gradients belong to Morrey and block spaces. Finally, this is then
used in Theorem 3.3.12 to derive operator identities akin to those established in The-
orem 1.8.26, now involving Morrey-based and block-based homogeneous Sobolev
spaces.

3.1 Boundary Layer Potentials on Calderón Spaces

We begin by introducing two maximal operators, then associate with them what
we shall call Calderón spaces (aka Cpα -spaces) on uniformly rectifiable sets, with
the ultimate goal of study singular integral operators in such a setting. Specifically,
let Σ ⊆ Rn be a closed Ahlfors regular set and abbreviate σ := H n−1 Σ. Having
fixed an exponent q ∈ [1, ∞) along with a power η ∈ R, we consider the following
“fractional” version of the Fefferman-Stein sharp maximal operator (A.0.195), acting
3.1 Boundary Layer Potentials on Calderón Spaces 377

on each f ∈ Lloc
1 (Σ, σ) as

 
⨏  q  1/q
#
fq,η (x) := sup R −η  f (y) − fΔ(x,R)  dσ(y) , ∀x ∈ Σ, (3.1.1)
R>0 Σ∩B(x,R)

⨏
where fΔ(x,R) := Σ∩B(x,R) f dσ for each R > 0 and x ∈ Σ. This turns out to be a
σ-measurable function1 and, obviously,

( f − c)#q,η = fq,η
#
for each c ∈ C. (3.1.2)

For each f ∈ Lloc

1 (Σ, σ) let us also set

 
⨏  1/q

fq,η (x) := sup R−η
# | f (y) − f (x)| dσ(y)
q
, ∀x ∈ Σ. (3.1.3)
R>0 Σ∩B(x,R)

Since for any two points x, y ∈ Σ and any R > 0 we have

 ⨏ 
 
 f (y) − fΔ(x,R)  ≤ | f (y) − f (x)| +  f (x) − 
f dσ 
Σ∩B(x,R)
⨏
≤ | f (y) − f (x)| + | f (x) − f (z)| dσ(z), (3.1.4)
Σ∩B(x,R)

it follows that
⨏ ⨏
 
 f (y) − fΔ(x,R) q dσ(y) ≤ C | f (y) − f (x)| q dσ(y)
Σ∩B(x,R) Σ∩B(x,R)
⨏
+ | f (x) − f (z)| q dσ(z)
Σ∩B(x,R)
⨏
=C | f (y) − f (x)| q dσ(y) (3.1.5)
Σ∩B(x,R)

hence, ultimately,

#
fq,η (x) ≤ C · # (x) at each x ∈ Σ.
fq,η (3.1.6)

In the opposite direction, we claim that if η > 0 then there exists C ∈ (0, ∞) with the
property that for each function f ∈ Lloc
1 (Σ, σ) we have

# (x) ≤ C · f # (x) at σ-a.e. x ∈ Σ.

fq,η (3.1.7)
q,η

Indeed, we shall show that (3.1.7) holds at each Lebesgue point x ∈ Σ for f . To
see that this is the case, fix such a point and, given the nature of the conclusion we

1 the same argument used in the proofs of [68, Theorems 6.3.3 and 7.6.1] applies
378 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

# (x) < +∞. In

seek, observe that there is no loss of generality in assuming that fq,η
q
turn, this ensures that actually f ∈ Lloc (Σ, σ). To proceed, pick a scale R > 0. The
estimate in [68, (7.4.121)] written with 2−j−1 R in place of r then gives that for each
j ∈ N we have
⨏    1/q
 f (y) − fΔ(x,2− j−1 R) q dσ(y)
Σ∩B(x,R)

j ⨏
  1/q
 
≤C  f (y) − fΔ(x,2−k R) q dσ(y)
k=0 Σ∩B(x,2−k R)


∞
≤C (2−k R)η fq,η
#
(x) = CRη fq,η
#
(x), (3.1.8)
k=0

for some finite C > 0 (since η > 0). After passing to limit as j → ∞ we therefore
arrive at
⨏  1/q
| f (y) − f (x)| q dσ(y) ≤ CRη fq,η
#
(x), (3.1.9)
Σ∩B(x,R)

on account of Lebesgue’s Differentiation Theorem (cf. [68, Proposition 7.4.4]

whose applicability in the present setting is guaranteed by [68, Lemma 3.6.4])
and Lebesgue’s Dominated Convergence Theorem (which uses the membership of
f to Lloc (Σ, σ)). At this stage, (3.1.7) follows from (3.1.9) after dividing by Rη and
q

taking the supremum over R > 0.

With Σ, σ as above, for each p ∈ [1, ∞], q ∈ [1, ∞), and η ∈ R we define
.p
Cq,η (Σ, σ) := f ∈ Lloc
1
(Σ, σ) : fq,η
#
∈ L p (Σ, σ) (3.1.10)

equipped with the semi-norm

.p
f C. p #
:= fq,η L p (Σ,σ)
, ∀ f ∈ Cq,η (Σ, σ). (3.1.11)
q, η (Σ,σ)

Note that if we set

[ f ] C. p := f C. p for each f ∈ Lloc

1
(Σ, σ), (3.1.12)
q, η (Σ,σ)/∼ q, η (Σ,σ)

then [·] C. p becomes a genuine norm on the quotient space

q, η (Σ,σ)/∼

.p .p
Cq,η (Σ, σ)/∼ := [ f ] : f ∈ Cq,η (Σ, σ) . (3.1.13)

Let us also introduce

3.1 Boundary Layer Potentials on Calderón Spaces 379
p .p
Cq,η (Σ, σ) : = L p (Σ, σ) ∩ Cq,η (Σ, σ)

= f ∈ L p (Σ, σ) : fq,η
#
∈ L p (Σ, σ) (3.1.14)

and equip this space with the norm

p
f p
Cq, η (Σ,σ) := f L p (Σ,σ) + fq,η
#
,
L p (Σ,σ)
∀ f ∈ Cq,η (Σ, σ). (3.1.15)

Since the maximal functions from (3.1.3) and (3.1.1) have been first introduced in the
Euclidean setting in [8] and [10], respectively, we shall refer to (3.1.10) and (3.1.14)
as (homogeneous and inhomogeneous) Calderón spaces. A wealth of information
concerning the Euclidean version of these Calderón spaces may be found in [23],
[14], [6], [80], [94] and the references therein. The novelty here is the consideration
of the action of singular integral operators on Calderón spaces defined on uniformly
rectifiable sets (cf. Theorem 3.1.1 below in this regard).
In relation to these scales of spaces we wish to make two comments. First, since
[68, (7.4.115)] (used with X := Σ, μ := σ, d := n − 1, ε := 1, and p := 1) gives that
for each function f ∈ Lloc 1 (Σ, σ), each point x ∈ Σ, and each scale r ∈ (0, ∞), we
0
have
∫  
f (x) − fΔ(x0,r) 
  n dσ(x)
r + |x − x0 |
Σ
∫ ⨏  1/q dλ
C ∞  
≤  f (x) − fΔ(x ,λr) q dσ(x)
r 1 Σ∩B(x0,λr)
0
λ2
∫
C  ∞
dλ  #
≤ f (x0 ), (3.1.16)
r 1−η 1 λ2−η q,η
it follows that
.p  σ(x) 
p
Cq,η (Σ, σ) ⊂ Cq,η (Σ, σ) ⊂ L 1 Σ, whenever η < 1. (3.1.17)
1 + |x| n
Our second comment is that, as is visible from [68, Proposition 7.4.9] (whose present
applicability is ensured by [68, Lemma 3.6.4] used with s := n − 1), (A.0.195), and
[68, (7.4.113)], corresponding to the case when p = ∞ we have
 .
.∞ 𝒞η (Σ) if η > 0,
Cq,η (Σ, σ) = (3.1.18)
BMO(Σ, σ) if η = 0.

The action of the boundary-to-boundary version of double layer potential op-

erators associated with a given homogeneous, weakly elliptic, constant coefficient,
second-order system on uniformly rectifiable sets has been studied on Hölder spaces
in Theorem 2.1.10 and on the John-Nirenberg space of functions of bounded mean
oscillations in Theorem 2.1.7. We wish to augment this body of results by consider-
380 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

ing below mapping properties for said double layer potential operators on Calderón
spaces with p < ∞.

Theorem 3.1.1 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. In addition, let L be a homogeneous, weakly elliptic, constant
(complex) coefficient, second-order M × M system in Rn (for some M ∈ N). In this
setting, let K be a principal-value double layer operator associated with Ω and L
as in (1.3.68), and recall its modified version Kmod introduced in (1.8.24)-(1.8.25).
Also, fix q ∈ (1, ∞) and η ∈ (0, 1).
Then there exists a constant C ∈ (0, ∞) with the property that, with the maximal
operator defined as in (3.1.1) (with Σ := ∂Ω), one has
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Kmod f (3.1.19)
   M
for every function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n . As a consequence, given any p ∈ [1, ∞),
it follows that the operator
 .p  M  .p  M
[Kmod ] : Cq,η (∂Ω, σ) ∼ −→ Cq,η (∂Ω, σ) ∼
   .p M (3.1.20)
defined by [Kmod ][ f ] := Kmod f for each f ∈ Cq,η (∂Ω, σ)
 .p  M
is well defined, linear, and bounded (considering the space Cq,η (∂Ω, σ) ∼
equipped with the norm defined as in (3.1.12)). Moreover,
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Kf (3.1.21)
 M
for every function f ∈ L p (∂Ω, σ) with p ∈ [1, ∞). Finally, whenever one has
p ∈ (1, ∞) it follows that the operator
 p M  p M
K : Cq,η (∂Ω, σ) −→ Cq,η (∂Ω, σ) (3.1.22)
 p M
is well defined, linear, and bounded (considering the space Cq,η (∂Ω, σ) equipped
with the norm defined as in (3.1.15)).

As is apparent from the proof given below, similar results hold for double layer
potential operators acting on Lorentz-based Calderón spaces, defined in a similar
fashion to (3.1.10), (3.1.14), this time demanding the membership of the maximal
function to Lorentz spaces in lieu of Lebesgue spaces.
   M
Proof of Theorem 3.1.1 Select a function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x) |x | n

and pick x ∈ ∂Ω at which (3.1.7) holds (with Σ := ∂Ω). If for each R > 0 we define
 
gR (y) := f (y) − f (x) 1∂Ω∩B(x,2R) (y), ∀y ∈ ∂Ω, (3.1.23)

then Hölder’s inequality and (3.1.7) (keeping in mind (3.1.3)) permit us to estimate
3.1 Boundary Layer Potentials on Calderón Spaces 381
⨏  1/q
gR [L 1 (∂Ω,σ)] M ≤ CRn−1 | f (y) − f (x)| q dσ(y)
Σ∩B(x,2R)

≤ CRn−1+η fq,η
#
(x), ∀R > 0. (3.1.24)

As a consequence, for each fixed scale R > 0 we have

∫
| f (y) − f (x)|
dσ(y)
∂Ω\B(x,2R) |y − x| n
∞ ∫
 | f (y) − f (x)|
= dσ(y)
j=1 ∂Ω∩[B(x,2 j+1 R)\B(x,2 j R)] |y − x| n


∞
  −n
≤ 2j R g2 j R [L 1 (∂Ω,σ)] M
j=1


∞
  −n   n−1+η
≤ 2j R 2j R #
fq,η (x)
j=1

= CRη−1 fq,η
#
(x), (3.1.25)

for some constant C > 0, independent of f and x, which is finite since η < 1. In a
similar fashion, for each R > 0 we may estimate
∫
| f (y) − f (x)|
dσ(y)
∂Ω∩B(x,2R) |y − x|
n−1

∞ ∫
 | f (y) − f (x)|
= dσ(y)
j=0 ∂Ω∩[B(x,2− j+1 R)\B(x,2− j R)] |y − x| n−1


∞
  −(n−1)
≤ 2−j R g2− j R [L 1 (∂Ω,σ)] M
j=0


∞
  −(n−1)   n−1+η
≤ 2−j R 2−j R #
fq,η (x)
j=0

= CRη fq,η
#
(x), (3.1.26)

for some constant C > 0, independent of f and x, which is finite since η > 0.
Going further, from (1.8.28) we see that for each z ∈ ∂Ω we have
         
Kmod f (z) − Kmod f (x) = Kmod f − f (x) (z) − Kmod f − f (x) (x). (3.1.27)
382 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 αβ 
Write L explicitly as ar s ∂r ∂s 1≤α,β ≤M . If ν = (ν1, . . . , νn ) denotes the geometric
measure theoretic outward unit normal to Ω, then for each fixed γ ∈ {1, . . . , M } and
(rγβ)
z ∈ B(x, R) ∩ ∂Ω we may then employ (3.1.27) and (1.8.24) to express (with k ε
as in (1.8.25))
   
Kmod f γ (z) − Kmod f γ (x)
∫
βα (rγβ) (rγβ)  
= lim+ νs (y)ar s k ε (x − y) − k ε (z − y) fα (y) − fα (x) dσ(y)
ε→0
∂Ω
∫
βα  
= νs (y)ar s (∂r Eγβ )(x − y) − (∂r Eγβ )(z − y) fα (y) − fα (x) dσ(y)
∂Ω\B(x,2R)
∫
βα (rγβ)    
+ lim+ νs (y)ar s k ε (x − y) fα (y) − fα (x) dσ(y) + KgR (z).
ε→0
∂Ω∩B(x,2R)
(3.1.28)

On account of (3.1.28), the Mean Value Theorem, [70, (1.4.24)], (3.1.25), and
(3.1.25), for each point z ∈ B(x, R) ∩ ∂Ω we may then estimate
    
 K f (z) − K f (x)
mod mod
∫
   | f (y) − f (x)|
≤  KgR (z) + C|z − x| · dσ(y)
|x − y| n
∂Ω\B(x,2R)
∫
| f (y) − f (x)|
+C dσ(y)
|x − y| n−1
∂Ω∩B(x,2R)
  
≤  KgR (z) + CRη fq,η
#
(x). (3.1.29)

In turn, after raising the most extreme sides to the q-th power and integrating over
∂Ω ∩ B(x, R) with respect to σ in the variable z, this permits us to write
⨏       1/q
 K f (z) − K f (x)q dσ(z)
mod mod
∂Ω∩B(x,R)

≤ CR−(n−1)/q KgR [L q (∂Ω,σ)] M

+ CRη fq,η
#
(x)

≤ CR−(n−1)/q gR [L q (∂Ω,σ)] M
+ CRη fq,η
#
(x)

≤ CRη fq,η
#
(x), (3.1.30)
 M
thanks to boundedness of K on L q (∂Ω, σ) (cf. Theorem 1.5.1), (3.1.23), and
(3.1.7).
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 383

At this stage, (3.1.19) follows from (3.1.30) and (3.1.6), bearing in mind that σ-a.e.
x ∈ ∂Ω is a Lebesgue point for f (cf. [68, Proposition 7.4.4] and [68, Lemma 3.6.4]).
As far. as the claim pertaining to the operator in (3.1.20) is concerned, for each given
p
f ∈ Cq,η (∂Ω, σ) we may write

[Kmod ][ f ] [C. p = [Kmod f ] [C. p

q, η (∂Ω,σ)/∼] q, η (∂Ω,σ)/∼]
M M

= Kmod f .p = (Kmod f )#q,η L p (∂Ω,σ)

[Cq, η (∂Ω,σ)] M

≤ C fq,η
#
L p (∂Ω,σ) =C f .p
[Cq, η (∂Ω,σ)] M

= C [ f ] [C. p (3.1.31)
q, η (∂Ω,σ)/∼]
M

thanks to (3.1.20), (3.1.11), (3.1.19), and (3.1.12). The desired conclusions follow
from this. Next, the claim in (3.1.21) becomes a consequence of (3.1.19), (1.8.24),
(1.3.68), and (3.1.2). Finally, that the operator K in (3.1.22) is well defined, linear,
and bounded whenever p ∈ (1, ∞) is seen from (3.1.21), (3.1.14)-(3.1.15), and the
 M
boundedness of K on L p (∂Ω, σ) (cf. Theorem 1.5.1). 

3.2 Boundary Layer Potentials on Morrey-Campanato Spaces

and Their Pre-Duals

In this section we turn our attention to singular integral operators on Morrey-

Campanato spaces, as well as their pre-duals, on uniformly rectifiable sets. We
begin by establishing the boundedness result described in the theorem below for
the transpose (principal-value) double layers, associated with a given weakly elliptic
system on the boundary of an arbitrary UR domain, acting on the pre-dual space of
the Morrey-Campanato space introduced in [69, Chapter 6].

Theorem 3.2.1 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). In this context,
consider the boundary layer potential operator K # associated with the system L and
the set Ω as in (1.3.72). Finally, select q ∈ (1, ∞) and λ ∈ (0, n − 1).
Then the operator
 M  M q(n − 1)
K # : L r (∂Ω, σ) −→ L r (∂Ω, σ) with r := (3.2.1)
n − 1 + λ(q − 1)
 M
(cf. (1.5.12)), has ℋq,λ (∂Ω, σ) as an invariant subspace (cf. [69, (6.1.22)]),
and
384 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 M 
K # : ℋq,λ (∂Ω, σ) , · [ℋ q, λ (∂Ω,σ)] M

 M 
−→ ℋq,λ (∂Ω, σ) , · [ℋ q, λ (∂Ω,σ)] M (3.2.2)

is a linear and bounded mapping.

 M
Proof Assume first that ∂Ω is unbounded and let a ∈ L q (∂Ω, σ) be an arbitrary
C M -valued
 ℋ q,λ -atom on ∂Ω. Then (cf. [69, (6.1.15)]) there exist x ∈ ∂Ω and
 o
R ∈ 0, 2 diam(∂Ω) such that
1 ∫
λ( q −1)
supp a ⊆ B(xo, R) ∩ ∂Ω, a [L q (∂Ω,σ)] M ≤R , a dσ = 0. (3.2.3)
∂Ω

Contrasting the above properties to [69, (4.4.2)] shows that R(λ−n+1)(1−1/q) a is an

ordinary C M -valued (1, q)-atom (i.e., an L q -normalized atom for the Hardy space
H 1 on ∂Ω). Consequently, (2.1.8) gives that

the function m := R(λ−n+1)(1−1/q) K # a is a fixed multiple of a

C M -valued (1, q, ε)-molecule on ∂Ω, in the sense considered in (3.2.4)
[69, Definition 4.5.1] for the choice ε := 1/(n − 1).

Using the boundedness of K # on Lebesgue spaces (cf. (1.5.12)) and the normalization
of the atom in (3.2.3), we estimate
∫ ∫
|K # a| q dσ ≤ C |a| q dσ ≤ CRλ(1−q), (3.2.5)
∂Ω ∂Ω

for some constant C ∈ (0, ∞) independent of the ℋq,λ -atom a. Also, if we choose
(for ε := 1/(n − 1), as above)
 
θ ∈ (n − 1)(q − 1) , (n − 1)[(1 + ε)q − 1] , (3.2.6)

then by relying on (3.2.4) and [69, (4.5.1)] we may write (using the piece of notation
introduced in [69, (4.5.2)] with Σ := ∂Ω)
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 385
∫
|(K # a)(x)| q |x − xo | θ dσ(x)
∂Ω
∫
= R(n−1−λ)(q−1) |m(x)| q |x − xo | θ dσ(x)
∂Ω


∞ ∫
≤ CR(n−1−λ)(q−1) (2k R)θ |m| q dσ
k=0 A k (x o ,R)


∞
  1−q
≤ CR(n−1−λ)(q−1)+θ 2k(n−1)[1−(1+ε)q]+kθ σ B(xo, R) ∩ ∂Ω
k=0


∞ 
≤ CR(n−1−λ)(q−1)+θ 2k {θ−(n−1)[(1+ε)q−1]} R−(n−1)(q−1)
k=0

= CRλ(1−q)+θ , (3.2.7)

for some C ∈ (0, ∞) independent of the ℋq,λ -atom a. In addition, from (3.2.4) and
(2.1.13) we see that
∫
K # a dσ = 0. (3.2.8)
∂Ω

Collectively, (3.2.5), (3.2.7), (3.2.8), and [69, (6.1.37)] prove that

the function K # a is a fixed multiple of a C M -valued

(3.2.9)
ℋq,λ,θ -molecule on ∂Ω, for θ chosen as in (3.2.6).
In turn, from (3.2.9) and [69, (6.1.38)] we conclude that there exists a constant
C ∈ (0, ∞) independent of the ℋq,λ -atom a with the property that
 M
K # a ∈ ℋq,λ (∂Ω, σ) and K # a [ℋ q, λ (∂Ω,σ)] M ≤ C. (3.2.10)
 M
Suppose now that some function f ∈ ℋq,λ (∂Ω, σ) has been given. Then
there exist some numerical sequence {λ j } j ∈N ∈ 1 (N) along with a sequence {a j } j ∈N
of C M -valued ℋq,λ -atoms on ∂Ω such that f = ∞ λ a in the sense of distri-
j=1 j j 
butions on ∂Ω. Thanks to (A.0.85)-[69, (6.1.22)], the series ∞ j=1 λ j a j actually
 r M q(n−1)
converges to f in L (∂Ω, σ) with r := n−1+λ(q−1) . Also, (3.2.10) readily im-
 N 
plies that the sequence K # λ j a j N ∈N is Cauchy, hence convergent, in the
 j=1
M 
Banach space ℋq,λ (∂Ω, σ) , · [ℋq, λ (∂Ω,σ)] M . In light of [69, (6.1.22)], this
 M
latter convergence takes place in L r (∂Ω, σ) as well. Since K # is continuous on
 r M   
L (∂Ω, σ) , it follows that K # j=1 λ j a j N ∈N converges to K # f in the space
N
 q,λ M
ℋ (∂Ω, σ) . Bearing (3.2.10) in mind, this argument ultimately proves that
386 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 M
K # f ∈ ℋq,λ (∂Ω, σ) and
(3.2.11)
K# f [ℋ q, λ (∂Ω,σ)] M ≤C f [ℋ q, λ (∂Ω,σ)] M ,

from which all desired conclusions follow in the case when ∂Ω is unbounded.
Finally, in the case when ∂Ω is bounded we reason similarly, except that now
we also have to consider the action of K # on an atom with constant components.
Since on account of (1.5.12) and [69, (6.1.24)] the resulting function belongs to
 q,λ M
ℋ (∂Ω, σ) , all desired conclusions follow in this case as well. 

In turn, Theorem 3.2.1 is a key ingredient in the proof of the fact that the principal-
value double layer potential operators associated with a given weakly elliptic system
on the boundary of an arbitrary UR domain act in a natural fashion on Morrey-
Campanato spaces (both homogeneous and inhomogeneous), as indicated in our
next theorem.

Theorem 3.2.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N), and consider a
principal-value double layer potential operator K associated with the system L and
the set Ω as in (1.3.68). Finally, select p ∈ (1, ∞) and λ ∈ (0, n − 1).
Then the operator
 M  M
K : L p (∂Ω, σ) −→ L p (∂Ω, σ) (3.2.12)

(which is well defined, linear and bounded; cf. Theorem 1.5.1), has the inhomo-
 M
geneous Morrey-Campanato space L p,λ (∂Ω, σ) (cf. (A.0.119)) as an invariant
subspace, and
 M 
K : L p,λ (∂Ω, σ) , · [L p, λ (∂Ω,σ)] M
 M 
−→ L p,λ (∂Ω, σ) , · [L p, λ (∂Ω,σ)] M (3.2.13)

is a linear and bounded mapping.

Moreover, if Kmod is the modified version of the principal-value double layer
operator introduced in (1.8.24)-(1.8.25), the assignment
.  M .  M
[Kmod ] : L p,λ (∂Ω, σ) ∼ −→ L p,λ (∂Ω, σ) ∼
. M (3.2.14)
[Kmod ][ f ] := [Kmod f ] for each f ∈ L p,λ (∂Ω, σ)

is well defined, linear and bounded, assuming each quotient space above equipped
with the norm [·] [L. p, λ (∂Ω,σ)/∼] M (cf. [69, (6.1.9)]). In particular,
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 387
. M . M
Kmod : L p,λ (∂Ω, σ) → L p,λ (∂Ω, σ) is well defined,
linear, and there exists some constant C ∈ (0, ∞) with the
property that Kmod f [L. p, λ (∂Ω,σ)] M ≤ C f [L. p, λ (∂Ω,σ)] M for (3.2.15)
. M
each function f ∈ L p,λ (∂Ω, σ) .

Finally, assuming that q ∈ (1, ∞) satisfies 1/p + 1/q = 1,

the (real) transpose #

  of the operator K from (3.2.2) is,
respectively, Kmod from (3.2.14) if ∂Ω is unbounded, (3.2.16)
and K from (3.2.13) if ∂Ω is bounded,
. M  M
that is, for each f ∈ L p,λ (∂Ω, σ) and g ∈ ℋq,λ (∂Ω, σ) one has
    
Kmod f , g = [ f ], K # g if ∂Ω is unbounded, (3.2.17)
   
K f , g = f , K # g if ∂Ω is bounded, (3.2.18)

where ·, · denotes the duality bracket between the Morrey-Campanato space and
its pre-dual (cf. [69, (6.1.25)]).
Proof To fix ideas, assume ∂Ω is unbounded (the case when ∂Ω is bounded is very
similar). Recall the operator K # associated with the given system L and the set Ω as
in (1.3.72). Let q ∈ (1, ∞) be the exponent satisfying 1/p + 1/q = 1 and consider
consider the operator
.   .  
 : L p,λ (∂Ω, σ) ∼ M −→ L p,λ (∂Ω, σ) ∼ M
K
   
defined by K[  f ], g := [ f ], K # g for every (3.2.19)
. M  M
f ∈ L p,λ (∂Ω, σ) and g ∈ ℋq,λ (∂Ω, σ)

where the angled brackets ·, · stand for the duality pairing between the homogeneous
Morrey-Campanato space, modulo constants, and its pre-dual (cf. [69, (6.1.25)]).
Thanks to Theorem 3.2.1 and [69, (6.1.25)] it follows that K  is a well-defined, linear
and bounded operator. This is going to be of significance shortly.
For now, fix an arbitrary C M -valued ℋq,λ -atom a = (aα )1≤α ≤M on ∂Ω. Also,
. M
pick an arbitrary function f = ( fα )1≤α ≤M ∈ L p,λ (∂Ω, σ) . Since from [69,
(6.1.3), (6.1.14)] (presently used with Σ := ∂Ω) we know that
 p M  1 σ(x)   M
f ∈ Lloc (∂Ω, σ) ∩ L ∂Ω, , (3.2.20)
1 + |x| n
we may invoke [70, (2.3.35)] (bearing in mind (1.8.24)) to conclude that
 p M
Kmod f ∈ Lloc (∂Ω, σ) . (3.2.21)
∫
We also have ∂Ω | f ||K # a| dσ < +∞, thanks to (3.2.9) and [69, Lemma 6.1.3] (used
with Σ := ∂Ω). Granted this and (3.2.20), we may reason as in (2.1.104)-(2.1.106)
388 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

to conclude that (cf. (2.1.107))

∫ ∫
 
fα (K # a)α dσ = Kmod f α aα dσ, (3.2.22)
∂Ω ∂Ω

using the summation convention over repeated indexes. As such, we may write
∫ ∫
   
(Kmod f )α aα dσ =  f ], a ,
fα (K # a)α dσ = [ f ], K # a = K[ (3.2.23)
∂Ω ∂Ω

where the second equality above is a consequence of (3.2.9) and [69, Lemma 6.1.3]
(again, used with Σ := ∂Ω), while the third equality in (3.2.23) comes from (3.2.19).
. M
If we now pick a function h = (hα )1≤α ≤M ∈ L p,λ (∂Ω, σ)  f ],
such that [h] = K[
then (3.2.23), [69, Lemma 6.1.3], and [69, (6.1.88)] permit us to write
∫ ∫
 
(Kmod f )α aα dσ = [h], a = hα aα dσ. (3.2.24)
∂Ω ∂Ω

On account of this, the arbitrariness of the ℋq,λ -atom a, the fact that any such
atom is a multiple of an ordinary (1, q)-atom (for the scale of Hardy spaces on ∂Ω),
[69, Lemma 4.6.9], and the fact that for each α ∈ {1, . . . , M } both (Kmod f )α and hα
p
belong to Lloc (∂Ω, σ), we finally conclude that

 f ].
[Kmod f ] = [h] = K[ (3.2.25)

In turn, based on (3.2.25), the boundedness of the operator (3.2.19), and [69, (6.1.9)],
we may estimate
 f ] . p, λ
[Kmod f ] [L. p, λ (∂Ω,σ)/∼] M = K[ [L (∂Ω,σ)/∼] M

≤ C [ f ] [L. p, λ (∂Ω,σ)/∼] M , (3.2.26)

. M
for some constant C ∈ (0, ∞) independent of the function f ∈ L p,λ (∂Ω, σ) .
This proves that the assignment (3.2.14) is indeed well defined, linear and bounded.
The version of the claims in (3.2.16)-(3.2.18) corresponding to the case when ∂Ω is
unbounded is also implicit in what we have proved so far.
 M  p . M
In the case when f ∈ L p,λ (∂Ω, σ) = L (∂Ω, σ) ∩ L p,λ (∂Ω, σ) , we know
from [70, (2.3.34)] that the difference Kmod f − K f is a constant on ∂Ω. On account
of this, [69, (6.1.8), (6.1.9)], and (3.2.26) we may then compute

K f [L. p, λ (∂Ω,σ)] M = Kmod f [L. p, λ (∂Ω,σ)] M = [Kmod f ] [L. p, λ (∂Ω,σ)/∼] M

≤ C [ f ] [L. p, λ (∂Ω,σ)/∼] M = C f [L. p, λ (∂Ω,σ)] M . (3.2.27)

Since the operator (3.2.12) is also bounded, from (3.2.27) and (A.0.120) we ulti-
mately obtain
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 389

Kf [L p, λ (∂Ω,σ)] M ≤C f [L p, λ (∂Ω,σ)] M , (3.2.28)

 M
for some C ∈ (0, ∞) independent of f ∈ L p,λ (∂Ω, σ) . With this in hand, all
desired conclusions in the statement of the theorem about the operator K (including
the version of the claims in (3.2.16)-(3.2.18) when ∂Ω is bounded) now readily
follow. 

It is also of interest to study the action of the boundary-to-domain double layer

potential operators associated with a given weakly elliptic system in an arbitrary
UR domain on homogeneous and inhomogeneous Morrey-Campanato spaces. This
topic is addressed in the theorem below.

Theorem 3.2.3 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. Let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N) and recall the
boundary-to-domain double layer operator D associated with L and Ω as in (1.3.18),
as well as its modified version Dmod defined in (1.8.6). Also, fix an integrability
exponent p ∈ (1, ∞) along with a parameter λ ∈ (0, n − 1).
Then there exists a constant C ∈ (0, ∞) with the property that
   
∇ D f (x) ≤ C f [L. p, λ (∂Ω,σ)] M
n−1−λ
sup dist(x, ∂Ω)1+ p
mod (3.2.29)
x ∈Ω
. M
for each function f ∈ L p,λ (∂Ω, σ) . In particular, there exists some constant
C ∈ (0, ∞) such that
   
∇ D f (x) ≤ C f . p, λ
n−1−λ
sup dist(x, ∂Ω)1+ p
[L (∂Ω,σ)] M
x ∈Ω

≤C f [L p, λ (∂Ω,σ)] M (3.2.30)
 M
for each function f ∈ L p,λ (∂Ω, σ) (see also [70, (2.6.5)]).

Proof Having fixed an arbitrary point x ∈ Ω, recall from (1.8.90)-(1.8.92) that for
   M
each given function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n we may express

  ∫ 
αγ
∂j Dmod f (x) = m j,x (y) fα (y) dσ(y) , 1 ≤ j ≤ n, (3.2.31)
∂Ω 1≤γ ≤M

αγ
for some family of σ-measurable functions m j,x defined on ∂Ω which satisfy, for
some constant C ∈ (0, ∞) independent of x,
 αγ 
m (y) ≤ C|x − y| −n at σ-a.e. y ∈ ∂Ω, (3.2.32)
j,x

as well as (cf. (1.8.98)) ∫

αγ
m j, x (y) dσ(y) = 0. (3.2.33)
∂Ω
390 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

To proceed, let q ∈ (1, ∞) be such that 1/p+1/q = 1, define η := 1+(n−1−λ)/p,

and pick a number θ such that (n − 1)(q − 1) < θ < nq − n + 1 (which is always
possible). Also, let xo ∈ ∂Ω be such that |x − xo | = dist(x, ∂Ω) and abbreviate
R := dist(x, ∂Ω). Finally, fix j ∈ {1, . . . , n} along with α, γ ∈ {1, . . . , M }. Then
αγ
the function Rη · m j,x is σ-measurable and there exists some constant C ∈ (0, ∞),
independent of x, with the property that
∫
 η 
 R · mαγ (y)q dσ(y)
j,x
∂Ω
∫
dσ(y)
≤ CRηq
∂Ω |x − y| qn
∫ ∫
dσ(y) dσ(y)
= CRηq + CRηq
B(x o ,2R)∩∂Ω |x − y| qn ∂Ω\B(x o ,2R) |x − y| qn
∫
  dσ(y)
≤ CRηq R−nq σ B(xo, 2R) ∩ ∂Ω + CRηq
∂Ω\B(x o ,2R) |y − xo | qn

≤ CRηq Rn−1−nq = CRλ(1−q) . (3.2.34)

Indeed, the first inequality above is a consequence of (3.2.32), the last inequality
is implied by the upper Ahlfors regularity of ∂Ω together with [68, Lemma 7.2.1]
(bearing in mind that nq > n − 1; cf. [68, (7.2.5)]), and the final equality is implied
by the choice of η and q. In a similar fashion, since qn − θ > n − 1 we may estimate
∫
 η 
 R · mαγ (y)q |y − xo | θ dσ(y)
j,x
∂Ω
∫
ηq |y − xo | θ
≤ CR dσ(y)
∂Ω |x − y| qn
∫
|y − xo | θ
= CRηq dσ(y)
B(x o ,2R)∩∂Ω |x − y| qn
∫
|y − xo | θ
+ CRηq dσ(y)
∂Ω\B(x o ,2R) |x − y| qn
∫
ηq θ−nq   ηq dσ(y)
≤ CR R σ B(xo, 2R) ∩ ∂Ω + CR
∂Ω\B(x o ,2R) |y − xo | qn−θ

≤ CRηq Rn−1−nq+θ = CRλ(1−q)+θ , (3.2.35)

for some constant C ∈ (0, ∞) independent of x. Collectively, (3.2.33), (3.2.34),

(3.2.35) prove that
αγ
each Rη · m j,x is a fixed multiple of some ℋq,λ,θ -molecule
(3.2.36)
on the set ∂Ω (in the sense described in [69, (6.1.37)]).
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 391

In turn, from (3.2.36) and [69, (6.1.88)] we conclude that there exists a constant
C ∈ (0, ∞) independent of x with the property that
αγ
Rη · m j,x ℋ q, λ (∂Ω,σ) ≤C (3.2.37)
. M
Given any function f = ( fα )1≤α ≤M ∈ L p,λ (∂Ω, σ) , for each j ∈ {1, . . . , n}
and γ ∈ {1, . . . , M } we may now estimate (bearing in mind [69, (6.1.14)])
       
∂j D  = Rη ∂j D f (x)
n−1−λ
dist(x, ∂Ω)1+ p
mod f γ (x) mod γ
∫ 
 αγ 
= Rη · m j,x (y) fα (y) dσ(y)
∂Ω

⎧  αγ  
⎨  [ fα ], R · m j,x  if ∂Ω is unbounded
⎪ η
⎪
=  
⎪
⎪  fα, Rη · mαγ  if ∂Ω is bounded
⎩ j,x

αγ
≤ C f [L. p, λ (∂Ω,σ)] M Rη · m j,x ℋ q, λ (∂Ω,σ)

≤ C f [L. p, λ (∂Ω,σ)] M . (3.2.38)

In (3.2.38), the first equality is implied the formulas for R and η, the second equality
uses (3.2.31), the third equality is a consequence of (3.2.36) and [69, Lemma 6.1.3],
the first inequality is seen from [69, (6.1.9), (6.1.29)] in the case when ∂Ω is
unbounded as well as their counterparts in the case when ∂Ω is bounded, and the
final inequality comes from (3.2.37).
In view of the arbitrariness of x ∈ Ω, j ∈ {1, . . . , n}, and γ ∈ {1, . . . , M }, the
estimate claimed in (3.2.29) now readily follows from (3.2.38). Finally, (3.2.30) is a
consequence of (3.2.29) and (1.8.9) (bearing in mind [68, (7.7.106)]). 

The (modified version of the) double layer potential operator associated with a
weakly elliptic system in a UR domain acting on the Morrey-Campanato spaces also
satisfies a fractional Carleson measure estimate of the sort described in our next
theorem (see also Example 5.1.13 for a more general result of this flavor).

Theorem 3.2.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. For some M ∈ N, suppose L is an M × M homogeneous, weakly
elliptic, constant (complex) coefficient, second-order system in Rn , and recall the
modified boundary-to-domain double layer operator Dmod associated with L and Ω
as in (1.8.6). Finally, fix p ∈ (1, ∞) and λ ∈ (0, n − 1).
Then there exists a constant C ∈ (0, ∞) with the property that the fractional
Carleson measure estimate
392 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 ∫  p1
   p
−λ ∇ D 
mod f (y) dist(y, ∂Ω) dL n (y)
p−1
sup r
x ∈∂Ω and r >0 B(x,r)∩Ω

≤ C f [L. p, λ (∂Ω,σ)] M (3.2.39)

holds for each function f belonging to the homogeneous Morrey-Campanato space

 . p,λ M
L (∂Ω, σ) .

As is apparent from an inspection of the proof of Theorem 3.2.4, the same type
of fractional Carleson measure estimate holds for other types of singular integral
operators which exhibit similar size and cancellation properties as the family of
double layers associated with weakly elliptic systems. For example, this is the case
for the operators Qi j defined in [70, (2.4.115)] hence, in particular, for the operators
∇R jk considered in [70, (2.5.266)]. Similar fractional Carleson measure estimates
for the gradient of the Cauchy-Clifford operator ∇C from [70, (2.5.298)] may also
be obtained, thanks to the discussion in Example 1.4.12, as a particular case of
Theorem 3.2.4.
. M
Proof of Theorem 3.2.4 Fix an arbitrary function f ∈ L p,λ (∂Ω, σ) . Also, fix a
point xo ∈ ∂Ω and for each R > 0 abbreviate
⨏
ΔR := B(xo, R) ∩ ∂Ω, T(ΔR ) := B(xo, R) ∩ Ω, and fΔ R := f dσ. (3.2.40)
ΔR
 
Then (A.0.118) ensures that for each R ∈ 0, 2 diam(∂Ω) we have
⨏
 
 f − fΔ  p dσ ≤ CR−(n−1−λ) f p. . (3.2.41)
R p, λ
ΔR
M [L (∂Ω,σ)]
 
In addition, pick a scale r ∈ 0, 2 diam(∂Ω) and consider a cutoff function η in Rn
satisfying
η ∈ 𝒞∞c (R ),
n 0 ≤ η ≤ 1, η ≡ 1 on B(xo, 2r),

η ≡ 0 outside B(xo, 4r), and (3.2.42)

|∂ α η(x)| ≤ Cα r − |α |, ∀x ∈ Rn and ∀α ∈ N0n .

We then proceed to split

f = η( f − fΔ4r ) + (1 − η)( f − fΔ4r ) + fΔ4r . (3.2.43)

From (3.2.43) and the property of Dmod recorded in (1.8.10) we see that
       
∇ Dmod f = ∇ Dmod η( f − fΔ4r ) + ∇ Dmod (1 − η)( f − fΔ4r ) . (3.2.44)

Thus,
3.2 Boundary Layer Potentials on Morrey-Campanato Spaces and Their Pre-Duals 393
∫  
   p
∇ Dmod f (x) dist(x, ∂Ω) p−1 dx
T (Δr )
∫  
     p
≤C ∇ Dmod η( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
T (Δr )
∫  
     p
+C ∇ Dmod (1 − η)( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
T (Δr )

=: I + II. (3.2.45)

We shall show that

· r λ and II ≤ C f . p, λ · rλ.
p p
I ≤ C f . p, λ (3.2.46)
[L (∂Ω,σ)] M [L (∂Ω,σ)] M

To justify the first inequality in (3.2.46), with D denoting a boundary-to-domain

double layer operator associated with L and Ω as in (1.3.18), we write
∫  
     p
I≤C ∇ Dmod η( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
Ω
∫  
     p
=C ∇ D η( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
Ω
∫ ∫
   
≤C η( f − fΔ ) p dσ ≤ C  f − fΔ  p dσ
4r 4r
∂Ω Δ4r
⨏
 
= Cr n−1  f − fΔ  p dσ ≤ Cr λ f p. . (3.2.47)
4r p, λ
Δ4r [L (∂Ω,σ)] M

Above, the first inequality follows from the definition of I in (3.2.45), the subsequent
equality is justified by (1.8.9), [69, (6.1.3)], and [68, (7.7.106)], the second inequality
is implied by [70, (2.4.34)], (1.3.18), and [70, Theorem 1.4.2], the third inequality
is clear from the support properties of the function η introduced in (3.2.42), the
subsequent equality is obvious, and the fourth inequality comes from (3.2.41).
To justify the second inequality in (3.2.46), we first observe that for each x ∈ T(Δr )
we have
  ∫
     | f (y) − fΔ4r |
∇ Dmod (1 − η)( f − fΔ4r ) (x) ≤ C dσ(y), (3.2.48)
∂Ω\Δ2r |y − xo | n

based on (1.8.11), the properties of the fundamental solution E from [70, The-
orem 1.4.2] and of the function η from [70, (2.4.132)], and the fact that since
x ∈ T(Δr ) we have |x − y| ≈ |xo − y| uniformly for y ∈ ∂Ω \ Δ2r . In turn, from the
first inequality in [68, (7.4.115)] (used with X = ∂Ω, μ := σ, p := 1, d := n − 1,
ε := 1, q := p, and with the letter t used in place of λ) we obtain
394 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
∫
| f (y) − fΔ4r |
dσ(y)
∂Ω\Δ2r |y − xo | n
∫
| f (y) − fΔ4r |
≤C   n dσ(y)
∂Ω r + |y − xo |
∫ ⨏
C ∞  
 f (x) − fΔ(x , tr)  p dσ(x)
 1/p dt
≤ o
r 1 Δ(x o , tr) t2
∫ ∞ dt 
≤ C f [L. p, λ (∂Ω,σ)] M · r −1 (tr)−(n−1−λ)/p 2
1 t
= C f [L. p, λ (∂Ω,σ)] M · r −1−(n−1−λ)/p . (3.2.49)

In the third inequality above

∫ ∞ we dthave used (3.2.41), and the final inequality takes
into account the fact that 1 t 2+(n−1−λ)/p < +∞. By combining (3.2.48) and (3.2.49)
we obtain
∫
p −p−(n−1−λ)
II ≤ C f . p, λ ·r dist(x, ∂Ω) p−1 dx
[L (∂Ω,σ)]
M
T (Δr )

· r −p−(n−1−λ) · r p−1 · r n
p
≤ C f . p, λ
[L (∂Ω,σ)] M

· rλ.
p
= C f . p, λ (3.2.50)
[L (∂Ω,σ)] M

This establishes the second inequality in (3.2.46). At this stage, the version of (3.2.39)
with the supremum taken in the regime 0 < r < 2 diam(∂Ω) follows from (3.2.45)
and (3.2.46). Finally, the version of (3.2.39) with the supremum taken in the regime
r ≥ 2 diam(∂Ω) is trivially implied by what we have just proved (since in this case
we have B(x, r) ∩ Ω = Ω for each x ∈ ∂Ω). 

3.3 Boundary Layer Potential Operators on Morrey Spaces and

Their Pre-Duals

The results in this section deal with singular integral operators of boundary layer
type, acting on Morrey spaces and their pre-duals, considered on boundaries of uni-
formly rectifiable domains. In a nutshell, the approach we adopt in the treatment of
singular integral operators of boundary layer type rests on two basic aspects. First, in
earlier work special care has been taken to establish integral formulas and operator
identities (such as those in [70, Theorem 1.7.10], Proposition 1.2.1, Proposition 1.2.2,
Lemma 1.3.2, along with the jump-formulas (1.5.20) and (1.5.58) in Theorem 1.5.1,
etc.) in an inclusive enough functional analytic setting which, in particular, allows the
consideration of Morrey spaces and their pre-duals. Second, in [70, Theorem 2.6.1],
[70, Proposition 2.6.2], and [70, Proposition 2.6.3] we have proved norm estimates
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 395

in Morrey spaces and their pre-duals for generic, garden-variety singular integral
operators, which in turn serve as building blocks for the algebraically more sophis-
ticated boundary layer potential operators we wish to consider. Together, these two
aspects work with great efficiency to yield very satisfactory theory for boundary
layer potential operators acting on Morrey spaces and their pre-duals, considered
on uniformly rectifiable sets. Here is the theorem which backs up these heuristic
considerations.
Theorem 3.3.1 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Define
σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
normal to Ω. Also, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). Write L = L A
for a choice of a coefficient tensor A and consider the layer potential operators K,
D, 𝒮, S, 𝒮mod , and K # associated with the coefficient tensor A and the set Ω as
in (1.3.68), (1.3.18), (1.3.6), (1.3.62), (1.5.50), and (1.3.72), respectively. Finally,
select p, q ∈ (1, ∞) such that 1/p + 1/q = 1, along with λ ∈ (0, n − 1), and some
aperture parameter κ > 0.
Then the following statements are true.
(i) The operators
 M  M
K, K # : M p,λ (∂Ω, σ) −→ M p,λ (∂Ω, σ) (3.3.1)

are well-defined, linear, and bounded. In addition, for each multi-index α ∈ N0n
there exists C ∈ (0, ∞), depending only on the Ahlfors regularity constants of
∂Ω, L, n, p, λ, and α, with the property that
 α  
dist (x, ∂Ω) |α |+ ∂ D f (x) ≤ C f [M p, λ (∂Ω,σ)] M
n−1−λ
sup p
x ∈Ω (3.3.2)
 M
for each function f ∈ M p,λ (∂Ω, σ) ,

and, with C ∈ (0, ∞) now depending on the UR character of ∂Ω, L, n, κ, p, and

λ,
   
Nκ D f M p, λ (∂Ω,σ) + Nκ ∇(𝒮mod f ) M p, λ (∂Ω,σ) ≤ C f [M p, λ (∂Ω,σ)] M
 M
for each function f ∈ M p,λ (∂Ω, σ) .
(3.3.3)

In light of work in [68, §8.5], a similar estimate to (3.3.3) is true with the non-
tangential maximal operator replaced by the tangential maximal operator (as-
sociated as in [68, Definition 8.5.1] with a sufficiently large power; (A.0.146)).
 M
Furthermore, for each given function f in the Morrey space M p,λ (∂Ω, σ)
the following nontangential boundary trace formulas hold (with I denoting the
identity operator)
κ−n.t.  

D f = 12 I + K f at σ-a.e. point on ∂Ω, (3.3.4)
∂Ω
396 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

and
κ−n.t.
∇(𝒮mod f )∂Ω exists at σ-a.e. point on ∂Ω and
  (3.3.5)
∂νA𝒮mod f = − 12 I + K A#  f at σ-a.e. point on ∂Ω,

where K A#  is the operator associated as in (1.3.72) with the coefficient tensor

A .
In addition, the operators K, K # in the context of (3.3.1) depend in a contin-
uous fashion on the underlying coefficient tensor (in the sense of item (xv) in
Theorem 1.5.1).
Finally, similar results are valid with the Morrey space M p,λ (∂Ω, σ) replaced
throughout by its version M̊ p,λ (∂Ω, σ) (defined as in (A.0.149) with Σ := ∂Ω).
(ii) Assume n ≥ 3 and suppose
1 1  −1
λ ∈ (0, n − 2), 1 < p < n − 1 − λ, p∗ := − . (3.3.6)
p n−1−λ
Then the boundary-to-boundary single layer potential operator induces well-
defined, linear, and bounded mappings in the following settings:
 M  M
S : M p,λ (∂Ω, σ) −→ M p∗,λ (∂Ω, σ) , (3.3.7)
 M  p , p,λ M
S : M p,λ (∂Ω, σ) −→ M1 ∗ (∂Ω, σ) . (3.3.8)

Also, there exists a constant C ∈ (0, ∞) such that

 
Nκ 𝒮 f M p∗, λ (∂Ω,σ) ≤ C f [M p, λ (∂Ω,σ)] M
 M (3.3.9)
for each function f ∈ M p,λ (∂Ω, σ) ,

plus a similar estimate in which the nontangential maximal operator is replaced

by the tangential maximal operator (associated as in [68, Definition 8.5.1] with
a sufficiently large power; (A.0.146)), and
 M
for each f ∈ M p,λ (∂Ω, σ) the boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact, (3.3.10)
 κ−n.t. 
𝒮f ∂Ω
(x) = (S f )(x) at σ-a.e. point x ∈ ∂Ω.

Moreover, similar results are valid for the pre-duals of Morrey spaces, as
well as their own pre-duals. Specifically, the jump-formula in (3.3.10) remains
 M
valid when the function f belongs the space M̊ p,λ (∂Ω, σ) or to the space
 p,λ M
B (∂Ω, σ) , and there exists some constant C ∈ (0, ∞) with the property
that
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 397
 
max Sf [M p∗ , λ (∂Ω,σ)] M , Nκ 𝒮 f M p∗ , λ (∂Ω,σ)
, Sf [M1
p∗ , p, λ
(∂Ω,σ)] M
 M
≤C f [M p, λ (∂Ω,σ)] M , for each f ∈ M̊ p,λ (∂Ω, σ) ,
(3.3.11)

and
 
max Sf [B p∗ , λ (∂Ω,σ)] M , Nκ 𝒮 f B p∗ , λ (∂Ω,σ)
, Sf p , p, λ
[ℋ1 ∗ (∂Ω,σ)] M
 M
≤C f [B p, λ (∂Ω,σ)] M for each f ∈ B p,λ (∂Ω, σ) .
(3.3.12)

(iii) The operators

 M  M
K, K # : B q,λ (∂Ω, σ) −→ B q,λ (∂Ω, σ) (3.3.13)

are well-defined, linear, bounded, and depend in a continuous fashion on the

underlying coefficient tensor (in the sense of item (xv) in Theorem 1.5.1). In fact,
the real transpose of K in (3.3.13) (respectively, K # in (3.3.13)) is the operator
K # in (3.3.1) (respectively, K in (3.3.1)). Also, the real transpose of K acting on
 p,λ M  M
M̊ (∂Ω, σ) (respectively, K # acting on M̊ p,λ (∂Ω, σ) ) is the operator
K # in (3.3.13) (respectively, K in (3.3.13)).
Moreover, there exists a constant C ∈ (0, ∞) such that
   
Nκ D f B q, λ (∂Ω,σ) + Nκ ∇(𝒮mod f ) B q, λ (∂Ω,σ) ≤ C f [B q, λ (∂Ω,σ)] M
 M
for each function f ∈ B q,λ (∂Ω, σ) .
(3.3.14)

Thanks to work in [68, §8.5], an estimate analogous to (3.3.14) holds if the

nontangential maximal operator is replaced by the tangential maximal oper-
ator (associated as in [68, Definition 8.5.1] with a sufficiently large power;
(A.0.146)).
Finally, the jump-formulas (3.3.4)-(3.3.5) remain valid for arbitrary vector-
 M
valued functions f ∈ B q,λ (∂Ω, σ) .
(iv) The following property holds:
 M
for each given sequence { f j } j ∈N ⊆ M p,λ (∂Ω, σ) which is
 p,λ M
weak-∗ convergent to some function f ∈ M (∂Ω, σ) one (3.3.15)
 
M
has lim Smod f j = Smod f in (Lipc (∂Ω)) .
j→∞

(v) Make the additional assumption that ∂Ω is compact. Then the boundary-
to-boundary single layer potential operator induces well-defined, linear, and
bounded mappings in the following settings:
398 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 M  p,λ M
S : M p,λ (∂Ω, σ) −→ M1 (∂Ω, σ) , (3.3.16)
 M  p,λ M
S : M̊ p,λ (∂Ω, σ) −→ M̊1 (∂Ω, σ) , (3.3.17)
 M  q,λ M
S : B q,λ (∂Ω, σ) −→ B1 (∂Ω, σ) . (3.3.18)

Also, for each large R ∈ (0, ∞) there exists a constant C ∈ (0, ∞) such that
   
NκΩ∩B(0,R) 𝒮 f + Nκ ∇𝒮 f M p, λ (∂Ω,σ) ≤ C f
M p, λ (∂Ω,σ) [M p, λ (∂Ω,σ)] M
 M
for each function f ∈ M p,λ (∂Ω, σ) ,
(3.3.19)

and
   
NκΩ∩B(0,R) 𝒮 f + Nκ ∇𝒮 f B q, λ (∂Ω,σ) ≤ C f
B q, λ (∂Ω,σ) [B q, λ (∂Ω,σ)] M
 M
for each function f ∈ B q,λ (∂Ω, σ) ,
(3.3.20)

with the convention that NκΩ∩B(0,R) may be replaced by Nκ if n ≥ 3. Finally, the

boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
 κ−n.t.  (3.3.21)
𝒮 f ∂Ω (x) = (S f )(x) at σ-a.e. point x ∈ ∂Ω,
 M  M
for each function f belonging to either M p,λ (∂Ω, σ) , or B p,λ (∂Ω, σ) .
(vi) Once again make the additional assumption that ∂Ω is compact. Then the
boundary-to-domain single layer potential operator induces a well-defined lin-
ear mapping in the context
 p,λ M  M
𝒮 : M−1 (∂Ω, σ) −→ 𝒞∞ (Ω) (3.3.22)
 p,λ M
if for each f = ( fβ )1≤β ≤M ∈ M−1 (∂Ω, σ) one sets
  
𝒮 f (x) := Eαβ (x − ·)∂Ω, fβ for all x ∈ Ω, (3.3.23)
1≤α ≤M

where E = (Eαβ )1≤α,β ≤M is the matrix-valued fundamental solution associated

with L as in [70, Theorem 1.4.2], and ·, · stands for the duality bracket between
q,λ p,λ
the spaces B1 (∂Ω, σ) and M−1 (∂Ω, σ) (cf. (A.0.156)). In addition, for each
 p,λ M
given f ∈ M−1 (∂Ω, σ) one has
 
L A 𝒮 f = 0 in Ω, (3.3.24)
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 399

for each large R > 0 there exists some constant C = C(Ω, L, κ, p, λ, R) ∈ (0, ∞)
such that

NκΩ∩B(0,R) (𝒮 f ) M p, λ (∂Ω,σ)
≤C f p, λ
[M−1 (∂Ω,σ)] M
(3.3.25)

(with the convention that NκΩ∩B(0,R) may be replaced by Nκ if n ≥ 3), and the
boundary trace
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
 κ−n.t.    (3.3.26)
𝒮f ∂Ω
(x) = (SL  ) f (x) at σ-a.e. point x ∈ ∂Ω,

where SL  is the boundary-to-boundary single layer potential operator associ-

 M  q,λ M
ated with L  , acting from B q,λ (∂Ω, σ) into B1 (∂Ω, σ) (cf. (3.3.17)),
and (SL  ) stands for its (real) transpose, i.e.,
 p,λ M  M
(SL  ) : M−1 (∂Ω, σ) −→ M p,λ (∂Ω, σ) (3.3.27)

(see [69, Definition 11.8.9] or (A.0.156), and [69, Proposition 6.2.8]).

(vii) Once more make the additional assumption that ∂Ω is compact. Then the
boundary-to-domain single layer potential operator induces a well-defined lin-
ear mapping in the context
 q,λ M  M
𝒮 : B−1 (∂Ω, σ) −→ 𝒞∞ (Ω) (3.3.28)
 q,λ M
with the convention that for each f = ( fβ )1≤β ≤M ∈ B−1 (∂Ω, σ) one defines
  
𝒮 f (x) := Eαβ (x − ·)∂Ω, fβ for all x ∈ Ω, (3.3.29)
1≤α ≤M

where E = (Eαβ )1≤α,β ≤M is the matrix-valued fundamental solution associated

with L as in [70, Theorem 1.4.2], and ·, · stands for the duality bracket be-
p,λ q,λ
tween M̊1 (∂Ω, σ) and B−1 (∂Ω, σ) (cf. (A.0.37)). Moreover, for each given
 q,λ M
distribution f ∈ B−1 (∂Ω, σ) one has
 
L A 𝒮 f = 0 in Ω, (3.3.30)

for each large R > 0 there exists some constant C = C(Ω, L, κ, q, λ, R) ∈ (0, ∞)
such that

NκΩ∩B(0,R) (𝒮 f ) B q, λ (∂Ω,σ)
≤C f q, λ
[B−1 (∂Ω,σ)] M
(3.3.31)

(with the convention that NκΩ∩B(0,R) may be replaced by Nκ if n ≥ 3), and the
boundary trace
 400 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
κ−n.t.
𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
 κ−n.t.    (3.3.32)
𝒮 f ∂Ω (x) = (SL  ) f (x) at σ-a.e. point x ∈ ∂Ω,

where SL  is the boundary-to-boundary single layer potential operator associat-

 M  q,λ M
ed with L  , considered as a mapping from M̊ p,λ (∂Ω, σ) into M̊1 (∂Ω, σ)
(cf. (3.3.18)), and (SL  ) stands for its (real) transpose, i.e.,
 q,λ M  M
(SL  ) : B−1 (∂Ω, σ) −→ B q,λ (∂Ω, σ) (3.3.33)

(see [69, Definition 11.8.9] or (A.0.37), and [69, Proposition 6.2.16]).

(viii) Strengthen the hypotheses on Ω by now assuming that this is an open set sat-
isfying a two-sided local John condition and whose boundary is compact and
Ahlfors regular. Then the operator (3.3.27) induces a well-defined, linear, bound-
ed mapping
 p,λ M  M
(SL  ) : M̊−1 (∂Ω, σ) −→ M̊ p,λ (∂Ω, σ) , (3.3.34)

and for each large R > 0

 p,λ M
NκΩ∩B(0,R) (𝒮 f ) ∈ M̊ p,λ (∂Ω, σ) for each f ∈ M̊−1 (∂Ω, σ) , (3.3.35)

(with the convention that NκΩ∩B(0,R) may be replaced by Nκ if n ≥ 3), where

p,λ
M̊−1 (∂Ω, σ) is the space introduced in [69, Definition 11.8.12]; cf. (A.0.149).

Proof The claims about the operators (3.3.1) are direct consequences of definitions,
[70, (2.6.1)], and [69, (6.2.5)]. The fact that the operators K, K # in the context of
(3.3.1) depend in a continuous fashion on the underlying coefficient tensor is a
consequence of [69, Proposition 6.2.12] and item (xv) in Theorem 1.5.1. Next, the
estimate in (3.3.2) follows from [70, (2.6.5)] and [69, (6.2.5)] (in this regard, see
also (3.2.29) and [69, (6.2.34)], bearing in mind (1.8.8) and [69, (6.2.25)]), while the
estimate in (3.3.3) is implied by [70, (2.6.4)], [69, (6.2.5)], and (1.5.51). Also, the
jump-formulas (3.3.4), (3.3.5) are direct consequences of (1.5.20), (1.5.53), (1.5.58),
and [69, (6.2.25)]. The very last claim in item (i) is a direct consequence of the last
property recorded in part (1) of [70, Theorem 2.6.1]. Next, all claims in item (ii)
with the exception of those pertaining to S mapping into Sobolev-like spaces (such
as (3.3.8), etc.) are direct consequences of [70, Propositions 2.6.2-2.6.3] (used with
Σ := ∂Ω and α := 1) and [70, Theorem 1.4.2] (the fact that K, K # in (3.3.13) depend
in a continuous fashion on the underlying coefficient tensor may be seen from the
 M
corresponding continuity result on M̊ p,λ (∂Ω, σ) from item (i), and the duality
result from [69, Proposition 6.2.16]). To deal with (3.3.8), we begin by noting that
from [68, (6.2.23)] (with τ := 1 and a := n − 2), (3.3.6), and [69, (6.2.7)] we have
 M   σ(x)  M
p
M p,λ (∂Ω, σ) → L 1 ∂Ω, ∩ L (∂Ω, σ) . (3.3.36)
1 + |x| n−2 loc
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 401

Granted this, we may invoke (1.2.30) which, together with [70, (2.6.1)], [70,
 M
Theorem 1.4.2], and [69, (6.2.3)], gives that for each f ∈ M p,λ (∂Ω, σ) and
j, k ∈ {1, . . . , n} we have
 M
∂τ j k (S f ) ∈ M p,λ (∂Ω, σ) and
(3.3.37)
∂τ j k (S f ) [M p, λ (∂Ω,σ)] M
≤C f [M p, λ (∂Ω,σ)] M

for some constant C ∈ (0, ∞) independent of f . Together with (3.3.7) and [69,
(11.7.14)], this proves that the operator S in (3.3.8) is well defined, linear and
bounded. All other claims pertaining to S mapping into Sobolev-like spaces of the
sort considered in the last portion of item (ii) are dealt with in a similar fashion.
Next, the claims in item (iii) are consequences of part (4) in [70, Theorem 2.6.1]
(keeping in mind [69, (6.2.74)]).
The claim in (3.3.15) is justified by reasoning much as in the proof of (1.5.78).
 M
Specifically, suppose the sequence { f j } j ∈N ⊆ M p,λ (∂Ω, σ) is weak-∗ convergent
 M
to some f ∈ M p,λ (∂Ω, σ) . In view of [69, Proposition 6.2.8] this amounts to
∫ ∫
 M
lim f j , F dσ = f , F dσ for each F ∈ B q,λ (∂Ω, σ) . (3.3.38)
j→∞ ∂Ω ∂Ω
 M
Pick an arbitrary C M -valued test function φ ∈ Lipc (∂Ω) , then choose a reference
point x0 ∈ ∂Ω along with some radius R ∈ (0, ∞) large  enough  to ensure that
supp φ ⊆ ∂Ω ∩ B(x0, R). Then for each j ∈ N decompose Smod f j , φ as in (1.5.140).
Consider the function F1 defined on ∂Ω as in (1.5.144). Since (1.5.145) holds, we
 M
conclude from [69, (6.2.245)] that F1 ∈ B q,λ (∂Ω, σ) given that N := n − 1
λ(q−1)+n−1
satisfies n − 1 > q (recall that λ < n − 1). As such, we conclude (much as
in (1.5.147), keeping in mind (3.3.38)) that
∫

lim I j = f (y), F1 (y) dσ(y). (3.3.39)
j→∞ ∂Ω

Next, thanks to [69, (6.2.69)] the matrix-valued function F2 defined on ∂Ω as in

 ∞  M×M  q,λ  M×M
(1.5.148) satisfies F2 ∈ Lcomp (∂Ω, σ) ⊆ B (∂Ω, σ) . Consequently,
the same type of argument as in (1.5.149) (now taking into account (3.3.38)) yields
∫ ∫
lim II j = F2 (y) f (y) dσ(y), φ(x) dσ(x) . (3.3.40)
j→∞ ∂Ω ∂Ω

There remains to handle the limit of III j as j → ∞. To this end, define F3 as in

(1.5.150) then use (1.5.151) and [68, (7.7.120)] to estimate

|F3 | ≤ CM ∂Ω φ pointwise on ∂Ω, (3.3.41)

where, as usual, M ∂Ω is the Hardy-Littlewood maximal operator on the set ∂Ω.

 ∞  M  q,λ M
Since [69, (6.2.69)] also ensures that φ ∈ Lcomp (∂Ω, σ) ⊆ B (∂Ω, σ) , [69,
402 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

Corollary 6.2.11] guarantees that M ∂Ω φ belongs to B q,λ (∂Ω, σ). In view of this,
 M
we may then conclude from (3.3.41) and [69, (6.2.75)] that F3 ∈ B q,λ (∂Ω, σ) .
With this in hand, we deduce from (3.3.38) that
∫
lim III j = f (y), F3 (y) dσ(y). (3.3.42)
j→∞ ∂Ω

In turn, from (1.5.140), (3.3.39), (3.3.40), and (3.3.42) we see that

   
lim Smod f j , φ = Smod f , φ , (3.3.43)
j→∞

 M
which ultimately proves that lim Smod f j = Smod f in (Lipc (∂Ω)) , as wanted.
j→∞
Moving on, all the claims made in item (v) of the theorem are consequences
of what we have proved so far, [68, Lemma 7.7.16], [70, (2.5.548)], and [69,
Corollaries 6.2.11, 6.2.13]. Consider next the claims made in item (vi). Work
under the assumption that ∂Ω is compact. To get started, fix some arbitrary
 p,λ M
f = ( fβ )1≤β ≤M ∈ M−1 (∂Ω, σ) . From [69, Proposition 11.8.10] we know that
(β) (β)
for each β ∈ {1, . . . , M } there exist f0 , f jk ∈ M p,λ (∂Ω, σ), with 1 ≤ j < k ≤ n,
satisfying

(β) (β)
f0 M p, λ (∂Ω,σ) + f jk M p, λ (∂Ω,σ) ≤ C fβ M p, λ (∂Ω,σ) (3.3.44)
−1
1≤ j<k ≤n

and
∫   
  (β) (β)
p, λ
M−1 (∂Ω,σ)
fβ, g q, λ
B1 (∂Ω,σ)
= f0 g + f jk ∂τ j k g dσ
∂Ω 1≤ j<k ≤n (3.3.45)
q,λ
for every function g ∈ B1 (∂Ω, σ).

For each x ∈ Ω and α ∈ {1, . . . , M } we may then write

    
𝒮 f α (x) = M p, λ (∂Ω,σ) fβ, Eαβ (x − ·)∂Ω B q, λ (∂Ω,σ)
−1 1
∫
(β)
= Eαβ (x − y) f0 (y) dσ(y)
∂Ω
 ∫
  (β)
+ ∂τ j k (y) Eαβ (x − y) f jk (y) dσ(y). (3.3.46)
1≤ j<k ≤n ∂Ω

To proceed, for any two pairs of indices, j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M },

introduce the integral operators acting on each function φ ∈ L 1 (∂Ω, σ) according to
∫
αβ
Tjk φ(x) := ν j (y)(∂k Eαβ )(x − y) − νk (y)(∂j Eαβ )(x − y) φ(y) dσ(y)
∂Ω
(3.3.47)
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 403

at each x ∈ Ω, where ν = (ν1, . . . , νn ) is the geometric measure theoretic outward

(β)
unit normal to Ω. In terms of this piece of notation, and with f0 := ( f0 )1≤β ≤M , the
formula recorded in (3.3.46) may be recast (bearing in mind [69, (6.2.25)]) simply
as
 
αβ (β) 
𝒮 f = 𝒮 f0 − Tjk f jk 1≤α ≤M in Ω. (3.3.48)
1≤ j<k ≤n

It is then apparent from (3.3.48) that (3.3.22) is a well-defined linear mapping and
that (3.3.24) holds. In addition, from (3.3.48), (3.3.19), [70, (2.6.4)], [69, (6.2.5)],
and (3.3.44) we see that (3.3.25) is true. From (3.3.48), (3.3.21), (1.2.4), and [69,
κ−n.t.
(6.2.25)] we also conclude that 𝒮 f ∂Ω exists at σ-a.e. point on ∂Ω and, in fact, at
σ-a.e. point on ∂Ω we have
κ−n.t.  
αβ (β) 
𝒮 f ∂Ω = S f0 − Tjk f jk 1≤α ≤M (3.3.49)
1≤ j<k ≤n

where, for each j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M },

αβ
Tjk φ(x) (3.3.50)
∫
:= lim+ ν j (y)(∂k Eαβ )(x − y) − νk (y)(∂j Eαβ )(x − y) φ(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε

for every function φ ∈ L 1 (∂Ω, σ) and σ-a.e. point x ∈ ∂Ω. Having established this,
 M
for each function ψ = (ψα )1≤α ≤M ∈ Lip(∂Ω) we may then write, thanks to
(1.3.62) and [70, (2.6.15)],
 κ−n.t.

[M p, λ (∂Ω,σ)] M 𝒮 f ∂Ω , ψ (3.3.51)
q, λ[B (∂Ω,σ)]
M

 
= [M p, λ (∂Ω,σ)] M S f0, ψ [B q, λ (∂Ω,σ)] M
  αβ (β) 
− M p, λ (∂Ω,σ) Tjk f jk , ψα
B q, λ (∂Ω,σ)
1≤ j<k ≤n
 
= [M p, λ (∂Ω,σ)] M f0, SL  ψ [B q, λ (∂Ω,σ)] M
  
(β)  αβ  #
+ M p, λ (∂Ω,σ) f jk , Tjk ψα
B q, λ (∂Ω,σ)
1≤ j<k ≤n

where for each j, k ∈ {1, . . . , n} and α, β ∈ {1, . . . , M } we have set

404 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
 αβ  #
Tjk ϕ(x) (3.3.52)
∫
:= lim+ ν j (x)(∂k Eαβ )(x − y) − νk (x)(∂j Eαβ )(x − y) ϕ(y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε

for every function ϕ ∈ L 1 (∂Ω, σ) and σ-a.e. point x ∈ ∂Ω. In relation to this operator
we wish to observe that, as seen from (1.5.83) and the first formula in [70, (1.4.32)],
for each j, k ∈ {1, . . . , n} and each β ∈ {1, . . . , M } we have
   αβ  #
∂τ j k SL  ψ β = Tjk ψα at σ-a.e. point on ∂Ω. (3.3.53)

Together, (3.3.51) and (3.3.53) prove (bearing in mind the nature of the duality
pairing involved; cf. [69, Proposition 6.2.8]) that
 κ−n.t.

[M p, λ (∂Ω,σ)] M 𝒮 f ∂Ω , ψ q, λ M [B (∂Ω,σ)]
 (β) 
= M p, λ (∂Ω,σ) f0 , (SL  ψ)β B q, λ (∂Ω,σ)
    
(β)
+ M p, λ (∂Ω,σ) f jk , ∂τ j k SL  ψ β
B q, λ (∂Ω,σ)
1≤ j<k ≤n
∫ 
(β)   
(β)   
= f0 SL  ψ β + f jk ∂τ j k SL  ψ β dσ
∂Ω 1≤ j<k ≤n
   
= p, λ
M−1 (∂Ω,σ)
fβ, SL  ψ β q, λ
B1 (∂Ω,σ)
 
= [M p, λ (∂Ω,σ)] M f , SL  ψ q, λ
[B1 (∂Ω,σ)] M
−1
 
= [M p, λ (∂Ω,σ)] M (SL  ) f , ψ [B q, λ (∂Ω,σ)] M
. (3.3.54)

In view of [69, (6.2.70)] and [69, Proposition 6.2.8], we conclude from formula
κ−n.t.  M
(3.3.54) that 𝒮 f ∂Ω = (SL  ) f as functions in M p,λ (∂Ω, σ) . This finishes the
proof of (3.3.26). The treatment of the claims in item (vi) is therefore complete. In
fact, all claims made in item (vii) are dealt with similarly, now making use of [69,
Proposition 6.2.16] and [69, (6.2.20)].
Finally, there remains to deal with the claims made in item (viii). Work under
the assumption that Ω is an open set satisfying a two-sided local John condition and
whose boundary is compact and Ahlfors regular. Also, define s := p(n−1)n−1−λ ∈ (p, ∞).
We claim that (SL  ) from (3.3.27) is compatible with S, the single layer potential
operator (associated with L and Ω) considered as in item (ix) of Theorem 1.5.1, in
the sense that
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 405

(SL  ) f = S f at σ-a.e. point on ∂Ω, for each

 s M  p,λ M (3.3.55)
f ∈ L−1 (∂Ω, σ) → M−1 (∂Ω, σ) (cf. [69, (11.8.43)]).
 s M  p,λ M
To justify (3.3.55), fix an arbitrary f ∈ L−1 (∂Ω, σ) → M−1 (∂Ω, σ) . Also,
q(n−1)
denote by r := n−1+λ(q−1) ∈ (1, q) the Hölder conjugate exponent of s. In par-
q,λ
ticular, from [69, (11.7.27)] we know that B1 (∂Ω, σ) → L1r (∂Ω, σ). Then for
 M
each function g ∈ B q,λ (∂Ω, σ) the embedding in [69, (6.2.71)] guarantees that
 r M
g ∈ L (∂Ω, σ) , and we may write
 
[M p, λ (∂Ω,σ)] M (SL  ) f , g [B q, λ (∂Ω,σ)] M
 
= [M p, λ (∂Ω,σ)] M f , SL  g [B q, λ (∂Ω,σ)] M
−1 1
 
s (∂Ω,σ)] M f , SL  g
= [L−1 [L1r (∂Ω,σ)] M
 
= [L s (∂Ω,σ)] M S f , g [L r (∂Ω,σ)] M
∫
= S f , g dσ
∂Ω
 
= [M p, λ (∂Ω,σ)] M S f , g [B q, λ (∂Ω,σ)] M . (3.3.56)

The first equality above is implied by the fact that (SL  ) is the transpose of SL  acting
 M  q,λ M
from B q,λ (∂Ω, σ) into B1 (∂Ω, σ) (cf. (3.3.18)). The second equality in
(3.3.56) comes from the compatibility property recorded in [69, (11.8.44)]. The
third equality in (3.3.56) is a consequence of item (ix) in Theorem 1.5.1, and [69,
(11.5.202)]. For the fourth equality in (3.3.56) we have used the fact that duality
between L s (∂Ω, σ) and L r (∂Ω, σ) is given by the integral pairing. Finally, the fifth
equality in (3.3.56) is seen from [69, Proposition 6.2.8]. Once (3.3.56) has been
established, (3.3.55) follows by invoking [69, (6.2.80)].
 s M  M
Since S maps L−1 (∂Ω, σ) into L s (∂Ω, σ) (cf. item (ix) of Theorem 1.5.1),
we conclude from [69, Definition 11.8.12] (see (A.0.149)), (3.3.55), (A.0.149),
(3.3.27), and [69, Lemma 1.2.20] that actually (SL  ) is a well-defined, linear, and
bounded operator in the setting described in (3.3.34).
To justify (3.3.35), first assume that n ≥ 3. We claim that the mapping
 p,λ M
M−1 (∂Ω, σ)  f −→ Nκ (𝒮 f ) ∈ M p,λ (∂Ω, σ) (3.3.57)
 p,λ M
is continuous. To justify this claim, pick f , g ∈ M−1 (∂Ω, σ) arbitrary. Then
(3.3.31) gives Nκ (𝒮 f ), Nκ (𝒮g) ∈ M p,λ (∂Ω, σ). In particular,
   
Nκ (𝒮 f ) (x) < +∞ and Nκ (𝒮g) (x) < +∞ for σ-a.e. point x ∈ ∂Ω. (3.3.58)
406 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

Granted this, we may invoke [68, (8.2.11)] to write

      

 Nκ (𝒮 f ) (x) − Nκ (𝒮g) (x) ≤ Nκ (𝒮( f − g)) (x) (3.3.59)

at σ-a.e. point x ∈ ∂Ω. In concert with [69, (6.2.3)] and (3.3.25), this gives

Nκ (𝒮 f ) − Nκ (𝒮g) M p, λ (∂Ω,σ)
≤ Nκ (𝒮( f − g)) M p, λ (∂Ω,σ)

≤C f −g p, λ
[M−1 (∂Ω,σ)] M
(3.3.60)

for some C ∈ (0, ∞) independent of f , g. Finally, (3.3.60) readily implies that the
mapping (3.3.57) is indeed continuous.
 p,λ M
To proceed, fix an arbitrary f ∈ M̊−1 (∂Ω, σ) . From [69, Definition 11.8.12]
 s M
(cf. (A.0.149)) we know that there exists a sequence { f j } j ∈N ⊆ L−1 (∂Ω, σ)
 p,λ M
which converges to f in M−1 (∂Ω, σ) . As a result of this, (1.5.45), and the
continuity of the mapping (3.3.57) we conclude that the sequence Nκ (𝒮 f j ) j ∈N
is contained in L s (∂Ω, σ) and converges to Nκ (𝒮 f ) in M p,λ (∂Ω, σ). In view of
(A.0.149), this places Nκ (𝒮 f ) in M̊ p,λ (∂Ω, σ), thus establishing (3.3.35). Finally,
the case when n = 2 and a truncated nontangential maximal operator is employed is
dealt with very similarly. 

There are also natural results (of the sort discussed in Theorem 1.5.1 for the
brand of boundary Sobolev spaces introduced in [69, Chapter 11]) involving layer
potential operators acting on Morrey-based Sobolev spaces, as defined in (A.0.150)-
(A.0.151), or even more generally, on off-diagonal Morrey-based Sobolev spaces, as
defined in (A.0.152)-(A.0.153), and the related brands from (A.0.154), as well as on
block-based Sobolev spaces, as defined in (A.0.33)-(A.0.34), and their off-diagonal
versions from (A.0.35)-(A.0.36).

Theorem 3.3.2 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Abbreviate

σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
normal to Ω. Net, let L be a homogeneous, weakly elliptic, constant (complex)
coefficient, second-order M × M system in Rn (for some M ∈ N). Write L = L A
for a choice of a coefficient tensor A and consider the layer potential operators K,
D, 𝒮, S, 𝒮mod , and K # associated with the coefficient tensor A and the set Ω as in
(1.3.68), (1.3.18), (1.3.6), (1.3.62), (1.5.50), and (1.3.72), respectively. Finally, pick
two arbitrary integrability exponents p, q ∈ (1, ∞), along with a number λ ∈ (0, n−1),
and an aperture parameter κ ∈ (0, ∞). Then the following statements are true.
 p,q,λ M
(i) For each f in the off-diagonal Morrey-based Sobolev space M1 (∂Ω, σ)
it follows that
  κ−n.t.
the nontangential boundary trace ∂ D f ∂Ω exists (3.3.61)
(in C M ) at σ-a.e. point on ∂Ω, for each ∈ {1, . . . , n},
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 407

and there exists some finite constant C > 0, depending only on the UR character
of ∂Ω, L, n, κ, p, q, λ, such that

Nκ (D f ) M p, λ (∂Ω,σ)
+ Nκ (∇D f ) M q, λ (∂Ω,σ)
≤C f p, q, λ
[M1 (∂Ω,σ)] M
.
(3.3.62)

In addition, the operator

 p,q,λ M  p,q,λ M
K : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) (3.3.63)

is well defined, linear, bounded, and depends in a continuous fashion on the

underlying coefficient tensor (in the sense of item (xv) in Theorem 1.5.1).
(ii) Denote by ν = (ν1, . . . , νn ) the geometric measure
 αβ  theoretic outward unit normal
to Ω, and choose a coefficient tensor A = ar s 1≤r,s ≤n such that L = L A. Then,
1≤α,β ≤M
given any function f = ( fα )1≤α ≤M belonging to the off-diagonal Morrey-based
 p,q,λ M
Sobolev space M1 (∂Ω, σ) , at σ-a.e. point x ∈ ∂Ω one has
 
∂νA(D f ) (x) (3.3.64)
! ∫ "
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0
y ∈∂Ω 1≤μ ≤M
|x−y |>ε

where the conormal derivative ∂νA(D f ) is originally defined as in (A.0.184).

Moreover, the conormal derivative of the double layer
 
f → ∂νA D f := ∂νA(D f ) (3.3.65)

induces a well-defined, linear, and bounded operator

 p,q,λ M  M
∂νA D : M1 (∂Ω, σ) −→ M q,λ (∂Ω, σ) . (3.3.66)

Also, if the exponents p, q  ∈ (1, ∞) are such 1/p+1/p = 1 and 1/q +1/q  = 1,
 p,q,λ M
then for any two given vector-valued functions f ∈ M1 (∂Ω, σ) and
 q, p,λ M
g ∈ B1 (∂Ω, σ) one has
∫ ∫
 A     A  
∂ν D f , g dσ = f , ∂ν D A g dσ (3.3.67)
∂Ω ∂Ω

A 
where ∂ν along with D A and, ultimately ∂νA D A , are defined as before with
A now replaced by A .
(iii) Similar results to those in items (i)-(ii) are valid for functions in the space
 p,q,λ M  p,q,λ M
M̊1 (∂Ω, σ) defined in (A.0.154), as well as the space B1 (∂Ω, σ)
408 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

defined in (A.0.35)-(A.0.36). Finally, the operator identities in item (xiii) of

Theorem 1.5.1 also have natural counterparts on these scales of spaces.
 p,q,λ M
Proof Given an arbitrary function f = ( fα )1≤α ≤M ∈ M1 (∂Ω, σ) , from
Lemma 1.3.2 and [69, (11.7.24)] we conclude that for each index ∈ {1, . . . , n} we
have
!∫ "
  βα
∂ D f (x) = ar s (∂r Eγβ )(x − y)(∂τ s fα )(y) dσ(y) , (3.3.68)
∂Ω
1≤γ ≤M

for all x ∈ Ω. Granted this, [70, Theorem 2.5.1] then justifies the claim made in
(3.3.61) (bearing in mind [69, (6.2.25), (11.7.14)]). Also, the estimate in (3.3.62) is
implied by (3.3.68), [69, (11.7.14)], [70, (2.6.4)], and [70, Theorem 1.4.2]. Next, the
fact that the operator K in (3.3.63) is well defined, linear, and bounded, follows from
 p,q,λ M
(3.3.1), [69, (11.3.26)] (used for u := D f with f ∈ M1 (∂Ω, σ) ; the fact
that [69, Proposition 11.3.2] applies to this function is ensured by (3.3.62), (3.3.3),
(3.3.4), (3.3.61), and [69, (6.2.25)]), and [69, (6.2.3)]. The fact that the operator
K in the context of (3.3.63) depends continuously on the underlying coefficient
tensor may be seen using the corresponding continuity result for K in (3.3.1) and
Proposition 1.5.6. This takes care of item (i).
As regards the claims in item (ii), the validity of the formula (3.3.64) for each given
 p,q,λ M
function f = ( fα )1≤α ≤M ∈ M1 (∂Ω, σ) is ensured by (1.5.29), bearing in
mind the second embedding in [69, (11.7.24)]. In turn, from (3.3.64), [69, (11.7.14)],
[70, (2.6.1)], [70, Theorem 1.4.2], and [69, (6.2.5)] we conclude that, indeed, the
operator in (3.3.66) is well defined, linear, and bounded. Next, the integral identity
claimed in (3.3.67) may be justified in the same manner as (1.5.32), bearing in
mind that the main ingredients in the proof of the (1.5.32) (namely Green’s formula
[70, (1.7.81)], as well as the jump-formula (1.5.20), and (1.5.29)) are all valid for
Morrey spaces and their pre-duals (cf. [70, (1.7.75)], (3.3.4), and (3.3.64)). Finally,
the claims in item (iii) are dealt with in a very similar fashion. 

It turns out that the boundary-to-domain double layer potential operators asso-
ciated with a given weakly elliptic system and UR domain with compact boundary
satisfy certain fractional Carleson measure estimates when acting on Morrey spaces.
This is made precise in the theorem below (see also Example 5.1.10 for a more
general result of this flavor).

Theorem 3.3.3 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. For some M ∈ N, suppose L is a homogeneous, weakly elliptic,
constant (complex) coefficient, second-order, M × M system in Rn , and consider
a boundary-to-domain double layer operator D associated with L and Ω as in
(1.3.18). Also, fix an integrability exponent p ∈ (1, ∞) along with some parameter
λ ∈ (0, n − 1).
Then there exists a constant C ∈ (0, ∞) with the property that the fractional
Carleson measure estimate
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 409
 ∫  p1
   
sup r −λ ∇ D f (y) p dist(y, ∂Ω) p−1 dL n (y)
x ∈∂Ω and r >0 B(x,r)∩Ω

≤C f [M p, λ (∂Ω,σ)] M (3.3.69)
 M
holds for each function f in the Morrey space M p,λ (∂Ω, σ) . In addition, there
 M
exists C ∈ (0, ∞) with the property that for each function f ∈ M p,λ (∂Ω, σ) one
has
 ∫ 1
  
lim+ sup r −λ ∇ D f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and B(x,r)∩Ω
r ∈(0,R)
 
≤ C dist f , [ M̊ p,λ (∂Ω, σ)] M , (3.3.70)
 M
where the distance is measured in M p,λ (∂Ω, σ) . In particular, for each function
 M
f belonging to the space M̊ p,λ (∂Ω, σ) one has
 ∫ 1
  
lim+ sup r −λ ∇ D f  p dist(·, ∂Ω) p−1 dL n p = 0. (3.3.71)
R→0 x ∈∂Ω and B(x,r)∩Ω
r ∈(0,R)

Proof The fractional Carleson measure estimate claimed in (3.3.69) is a consequence

of Theorem 3.2.4, [69, (6.2.44)], (1.8.9), and [68, (7.7.106)] (see also [69, (6.2.49)]
in the case when ∂Ω is bounded). Consider next the claim made in (3.3.70). To this
 M
end, fix some η ∈ (0, 1) and select an arbitrary function f ∈ M p,λ (∂Ω, σ) . Also,
 M
 some g ∈  Lipc (∂Ω) . Then for each location x ∈ ∂Ω and each scale
choose
r ∈ 0, 2 diam(∂Ω) we may rely on (3.3.69) to write
 ∫ 1
  
r −λ ∇ D( f − g)  p dist(·, ∂Ω) p−1 dL n p
B(x,r)∩Ω

≤C f −g [M p, λ (∂Ω,σ)] M . (3.3.72)

Also, we have
 ∫ 1
  
r −λ ∇ Dg  p dist(·, ∂Ω) p−1 dL n p
B(x,r)∩Ω
 ∫ 1
1   
∇ Dg  p dist(·, ∂Ω) p−1 dL n p
n−1−λ
≤ Cr p  
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

η+ n−1−λ
≤ C g [𝒞. η (∂Ω)] M · r p , (3.3.73)
410 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

where the last inequality comes from (1.8.87). Given that η + (n − 1 − λ)/p > 0,
from (3.3.72) and (3.3.73) we then conclude that
 ∫ 1
  
lim+ sup r −λ ∇ D f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and B(x,r)∩Ω
r ∈(0,R)

≤C f −g [M p, λ (∂Ω,σ)] M , (3.3.74)

for some constant C ∈ (0, ∞) independent of f and g. Having established this,

(3.3.70) follows on account of [69, (6.2.20)] and the arbitrariness of the function
 M
g ∈ Lipc (∂Ω) . Finally, (3.3.70) readily implies (3.3.71). 

The results presented in this section about the double layer operators associated
with arbitrary weakly elliptic systems are in effect for particular embodiments of
this class of singular integral operators, such as those discussed in Examples 1-7,
just prior to the statement of Proposition 1.4.21 in the build-up to Theorem 1.5.1. In
particular, they apply to the Cauchy-Clifford integral operator. In relation to this, we
wish to single out the following version of [70, Proposition 2.5.32] which opens the
door for developing a rich theory of Hardy spaces in the context of Morrey spaces
and their pre-duals.

Proposition 3.3.4 Suppose Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is a UR domain

and abbreviate σ := H n−1 ∂Ω. Also, pick p, q ∈ (1, ∞) such that 1/p + 1/q = 1 and
select λ ∈ (0, n − 1) arbitrary.
Then the principal-value Cauchy-Clifford integral operator C associated with Ω
as in (A.0.54) induces well-defined linear and bounded mappings

C : M p,λ (∂Ω, σ) ⊗ C n −→ M p,λ (∂Ω, σ) ⊗ C n (3.3.75)

C : M̊ p,λ (∂Ω, σ) ⊗ C n −→ M̊ p,λ (∂Ω, σ) ⊗ C n (3.3.76)

C : B q,λ (∂Ω, σ) ⊗ C n −→ B q,λ (∂Ω, σ) ⊗ C n (3.3.77)

and has the property that

C 2 = 14 I on either of the spaces

(3.3.78)
M p,λ (∂Ω, σ) ⊗ C n, M̊ p,λ (∂Ω, σ) ⊗ C n, or B q,λ (∂Ω, σ) ⊗ C n .

Likewise, the transpose Cauchy-Clifford integral operator C # associated with Ω

as in (1.6.1) induces well-defined linear and bounded mappings

C # : M p,λ (∂Ω, σ) ⊗ C n −→ M p,λ (∂Ω, σ) ⊗ C n (3.3.79)

C # : M̊ p,λ (∂Ω, σ) ⊗ C n −→ M̊ p,λ (∂Ω, σ) ⊗ C n (3.3.80)

C # : B q,λ (∂Ω, σ) ⊗ C n −→ B q,λ (∂Ω, σ) ⊗ C n (3.3.81)

3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 411

and has the property that

 2
C# = 14 I on either of the spaces
(3.3.82)
M p,λ (∂Ω, σ) ⊗ C n, M̊ p,λ (∂Ω, σ) ⊗ C n, or B q,λ (∂Ω, σ) ⊗ C n .

Finally,

the (real) transpose of C from (3.3.76) is C # from (3.3.81), (3.3.83)

the (real) transpose of C from (3.3.77) is C # from (3.3.79), (3.3.84)

the (real) transpose of C # from (3.3.80) is C from (3.3.77), (3.3.85)

the (real) transpose of C # from (3.3.81) is C from (3.3.75). (3.3.86)

Proof Combining [70, Theorem 2.6.1] together with (A.0.54) and (1.6.1) yields the
claims pertaining to the operators (3.3.75)-(3.3.77) and (3.3.79)-(3.3.81). Having
established these boundedness properties, the claims in (3.3.83)-(3.3.86) then follow
on account of [70, (2.6.15), (2.6.16)], (A.0.54), (1.6.1), and [68, Lemma 6.4.1].
To prove that C 2 = 14 I on M p,λ (∂Ω, σ), let f ∈ M p,λ (∂Ω, σ) ⊗ C n be arbitrary
and define u := C f in Ω where the boundary-to-domain Cauchy-Clifford operator C
is as in (A.0.53). Also, pick an arbitrary aperture parameter κ > 0. Based on (3.3.3)
and (3.3.4) we conclude that

u ∈ 𝒞∞ (Ω) ⊗ C n, Du = 0 in Ω,
κ−n.t.
u∂Ω = ( 12 I + C) f at σ-a.e. point on ∂Ω,
(3.3.87)
Nκ u belongs to the Morrey space M p,λ (∂Ω, σ),
u(x) = O(|x| 1−n ) if Ω is an exterior domain.

In concert with [69, (6.2.25)], the next-to-last property above implies that
∫
(Nκ u)(x)
dσ(x) < +∞, (3.3.88)
∂Ω + |x|
1 n−1

while the last property in (3.3.87) implies that, in the case when Ω is an exterior
domain, ∫
|u(x)| dL n = o(Rn ) as R → ∞. (3.3.89)
[B(0,2R)\B(0,R)]∩Ω

With (3.3.87)-(3.3.89) in hand, we may rely on [70, Theorem 1.2.2] to conclude that
∫  κ−n.t. 
1 x−y
u(x) =  ν(y)  u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
 
= C( 12 I + C) f (x) at each x ∈ Ω. (3.3.90)
412 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

Taking the nontangential boundary traces of the most extreme sides in (3.3.90) then
yields (on account of (3.3.4))

( 12 I + C) f = ( 12 I + C) f ( 12 I + C) f at σ-a.e. point on ∂Ω. (3.3.91)

This goes to show that, indeed, C 2 = 14 I on M p,λ (∂Ω, σ) ⊗ C n , hence also on

M̊ p,λ (∂Ω, σ) ⊗ C n by [69, (6.2.15)] and (3.3.76). That we also have C 2 = 14 I on
B q,λ (∂Ω, σ) ⊗ C n is established in a similar fashion, this time taking into account
item (iii) of Theorem 3.3.1 and the embedding in [69, (6.2.230)]. This completes the
proof of (3.3.78).
 2
Finally, from (3.3.78) and (3.3.83)-(3.3.84) we see that C # = 14 I both on
 2
M p,λ (∂Ω, σ) ⊗ C n and on B q,λ (∂Ω, σ) ⊗ C n . The former also implies C # = 14 I
on M̊ p,λ (∂Ω, σ) ⊗ C n , thanks to (3.3.80) and [69, (6.2.15)]. 
We continue by discussing mapping properties of the modified single layer po-
tential operator involving Morrey-based and block-based Sobolev spaces in arbitrary
UR domains.

Theorem 3.3.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain, and abbreviate

σ := H n−1 ∂Ω. Also, consider a homogeneous, weakly elliptic, second-order M ×M
system L in Rn , with complex constant coefficients, and recall the modified boundary-
to-boundary single layer operator Smod associated with L and Ω as in (1.5.73).
Finally, fix some integrability exponents p, q ∈ (1, ∞) along with some parameter
λ ∈ (0, n − 1). Then the following properties are true.
(1) The modified boundary-to-boundary single layer operator induces a mapping
 M  . p,λ M
Smod : M p,λ (∂Ω, σ) −→ M1 (∂Ω, σ) (3.3.92)

which is well defined, linear, and bounded, when the target space is endowed
with the semi-norm induced by (A.0.158).
(2) The following operator,
   M  . p,λ  M
Smod : M p,λ (∂Ω, σ) −→ M1 (∂Ω, σ) ∼ defined as
     . p,λ  M  p,λ M
Smod f := Smod f ∈ M1 (∂Ω, σ) ∼ , ∀ f ∈ M (∂Ω, σ) .
(3.3.93)

is well defined, linear, and bounded, when the quotient space is equipped with
the semi-norm2 introduced in [69, (11.13.51)].
(3) With 𝒮mod denoting the modified version of the single layer operator acting
   M
on functions from L 1 ∂Ω, 1+σ(x)
|x | n−1
as in (1.5.50), for each given aperture

2 Recall from [69, Proposition 11.13.10] that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 413

parameter κ > 0 there exists some constant C = C(Ω, L, n, p, λ, κ) ∈ (0, ∞) with

 M
the property that for each given function f ∈ M p,λ (∂Ω, σ) one has
 M  
𝒮mod f ∈ 𝒞∞ (Ω) , L 𝒮mod f = 0 in Ω,
 
Nκ ∇𝒮mod f ∈ M p,λ (∂Ω, σ),
 
Nκ ∇𝒮mod f M p, λ (∂Ω,σ) ≤ C f [M p, λ (∂Ω,σ)] M ,
  κ−n.t. 
and 𝒮mod f  (x) = (Smod f )(x) at σ-a.e. point x ∈ Aκ (∂Ω)
∂Ω
hence, (cf. [68, Proposition 8.8.4] and (A.0.204)), at σ-a.e. point x ∈ ∂Ω.
(3.3.94)

(4) Similar properties to those described in items (1)-(3) are valid for block spaces
(and block-based homogeneous Sobolev spaces) in place of Morrey spaces (and
Morrey-based homogeneous Sobolev spaces). More specifically, the operator
 M  . q,λ M
Smod : B q,λ (∂Ω, σ) −→ B1 (∂Ω, σ) (3.3.95)

is well defined, linear, and bounded, when the target space is endowed with the
semi-norm induced by (A.0.40). Also,
   M  . q,λ  M
Smod : B q,λ (∂Ω, σ) −→ B1 (∂Ω, σ) ∼ defined as
     . q,λ  M  q,λ M
Smod f := Smod f ∈ B1 (∂Ω, σ) ∼ , ∀ f ∈ B (∂Ω, σ)
(3.3.96)

is a well-defined linear and bounded operator, in the scenario in which the

quotient space is endowed with the semi-norm3 introduced in [69, (11.13.70)].
Finally, for each κ > 0 there exists a constant C = C(Ω, L, n, q, λ, κ) ∈ (0, ∞)
 M
with the property that for each function f ∈ B q,λ (∂Ω, σ) one has
 M  
𝒮mod f ∈ 𝒞∞ (Ω) , L 𝒮mod f = 0 in Ω,
 
Nκ ∇𝒮mod f ∈ B q,λ (∂Ω, σ),
  (3.3.97)
Nκ ∇𝒮mod f B q, λ (∂Ω,σ) ≤ C f [B q, λ (∂Ω,σ)] M ,
  κ−n.t. 
and 𝒮mod f  (x) = (Smod f )(x) at σ-a.e. point x ∈ ∂Ω.
∂Ω

(5) Analogous properties to those presented in items (1)-(3) above are also valid for
vanishing Morrey spaces M̊ p,λ (∂Ω, σ)
. p,λ(cf. (A.0.149)) and vanishing Morrey-
based homogeneous Sobolev spaces M1 (∂Ω, σ) (cf. [69, Definition 11.13.15],

3 according to [69, Proposition 11.13.12], said semi-norm is actually a genuine norm if Ω ⊆ R n

is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
414 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

or (A.0.159)-(A.0.160)) in place of Morrey spaces and Morrey-based homoge-

neous Sobolev spaces, respectively.

Proof The claims regarding the operator (3.3.92) are seen from (A.0.157), (1.5.75)
(with w ≡ 1), (1.5.76) (with 0 < ε < (n − 1 − λ)/p), [68, (6.2.23)] (with τ = 1 and
a = n − 1 − ε), [69, (6.2.25)], (1.5.83), and the fact that the singular integral operators
defined in (1.5.82) induce bounded mappings (cf. [70, (2.6.1)], [69, (6.2.5)], and [70,
Theorem 1.4.2])
 M  M
#
Tjk : M p,λ (∂Ω, σ) −→ M p,λ (∂Ω, σ) , ∀ j, k ∈ {1, . . . , n}. (3.3.98)

Once we have dealt with (3.3.92), the claims regarding (3.3.93) readily follow.
The properties in the first line of (3.3.94) are seen from (1.5.51), (1.5.52), and
[69, (6.2.25)]. The properties in the second line of (3.3.94) are consequences of
[69, (6.2.25)], (1.5.51), [70, Theorem 1.4.2], [70, (2.6.4)], and [69, (6.2.5)]. The
boundary trace formula in (3.3.94) is implied by (1.5.80) and [69, (6.2.25)]. Next,
the claims in item (4) are justified in an analogous fashion, now making use of the
fact that
 M  M
#
Tjk : B q,λ (∂Ω, σ) −→ B q,λ (∂Ω, σ) , ∀ j, k ∈ {1, . . . , n} (3.3.99)

are well-defined, linear, and bounded operators (as may be seen from [70, (2.6.14)],
[69, (6.2.74)], and [70, Theorem 1.4.2]). Finally, the claim in item (5) is seen by
reasoning in a similar fashion, bearing in mind (A.0.159) and making use of the fact
that
 M  M
Tjk#
: M̊ p,λ (∂Ω, σ) −→ M̊ p,λ (∂Ω, σ) , ∀ j, k ∈ {1, . . . , n} (3.3.100)

are well-defined, linear, and bounded operators (cf. [70, Theorem 2.6.1] and [69,
(6.2.17)]). 

Going further, we consider mapping properties of the modified boundary-to-

domain double layer potential operators involving Morrey-based and block-based
Sobolev spaces in arbitrary UR domains.

Theorem 3.3.6 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Denote

by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω and
αβ 
abbreviate σ := H n−1 ∂Ω. In addition, for some M ∈ N, let A = ar s 1≤r,s ≤n be
1≤α,β ≤M
a complex coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (3.3.101)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Also, let E = (Eγβ )1≤γ,β ≤M be

the matrix-valued fundamental solution associated with L as in [70, Theorem 1.4.2].
In this setting, recall the modified version of the double layer operator Dmod acting
   M
on functions from L 1 ∂Ω, 1+σ(x) |x | n as in (1.8.6). Finally, fix some integrability
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 415

exponents p, q ∈ (1, ∞) along with a number λ ∈ (0, n−1), and an aperture parameter
κ ∈ (0, ∞). Then the following statements are true.

(1) There exists some constant C = C(Ω, A, n, p, λ, κ) ∈ (0, ∞) with the property that
 . p,λ M
for each function f ∈ M1 (∂Ω, σ) it follows that
 M  
Dmod f ∈ 𝒞∞ (Ω) , L Dmod f = 0 in Ω,
  κ−n.t.   κ−n.t.
Dmod f ∂Ω and ∇Dmod f ∂Ω exist at σ-a.e. point on ∂Ω,
  (3.3.102)
Nκ ∇Dmod f belongs to M p,λ (∂Ω, σ) and
 
Nκ ∇Dmod f M p, λ (∂Ω,σ) ≤ C f [ M. p, λ (∂Ω,σ)] M .
1

 . p,λ M
In fact, for each function f ∈ M1 (∂Ω, σ) one has
κ−n.t.  
(Dmod f )∂Ω = 12 I + Kmod f at σ-a.e. point on ∂Ω, (3.3.103)
 . p,λ M
where I is the identity operator on M1 (∂Ω, σ) , and Kmod is the modified
boundary-to-boundary double layer potential operator from (1.8.24)-(1.8.25).
(2) Given any function f = ( fα )1≤α ≤M belonging to the Morrey-based homoge-
 . p,λ M
neous boundary Sobolev space M1 (∂Ω, σ) , at σ-a.e. point x ∈ ∂Ω one
has
 A 
∂ν (Dmod f ) (x) (3.3.104)
! ∫ "
μγ βα  
= lim+ νi (x)ai j ar s (∂r Eγβ )(x − y) ∂τ j s fα (y) dσ(y)
ε→0
y ∈∂Ω 1≤μ ≤M
|x−y |>ε

where the conormal derivative is considered as in (A.0.184).

(3) The operator
 . p,λ M  M
∂νA Dmod : M1 (∂Ω, σ) −→ M p,λ (∂Ω, σ) defined as
 A  . p,λ M (3.3.105)
∂ν Dmod ) f := ∂νA(Dmod f ) for each f ∈ M1 (∂Ω, σ)

is well defined, linear, and bounded, when the domain space is equipped with
the semi-norm induced by (A.0.158). In addition,
   . p,λ  M  M
∂νA Dmod : M1 (∂Ω, σ) ∼ −→ M p,λ (∂Ω, σ) defined as
 A   . p,λ M
∂ν Dmod [ f ] := ∂νA(Dmod f ) for each f ∈ M1 (∂Ω, σ) ,
(3.3.106)
416 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

is a well-defined linear and bounded operator, when the quotient space is

equipped with the semi-norm4 introduced in [69, (11.13.51)].
(4) Similar properties to those described in items (1)-(3) are valid for block spaces
(and block-based homogeneous Sobolev spaces) in place of Morrey spaces
(and Morrey-based homogeneous Sobolev spaces). Concretely, there exists a
constant C = C(Ω, A, n, q, λ, κ) ∈ (0, ∞) with the property that for each function
 . q,λ M
f ∈ B1 (∂Ω, σ) one has
 M  
Dmod f ∈ 𝒞∞ (Ω) , L Dmod f = 0 in Ω,
  κ−n.t.  
Dmod f ∂Ω = 12 I + Kmod f at σ-a.e. point on ∂Ω,
  κ−n.t.
∇Dmod f ∂Ω exist at σ-a.e. point on ∂Ω, (3.3.107)
 
Nκ ∇Dmod f belongs to B q,λ (∂Ω, σ) and
 
Nκ ∇Dmod f B q, λ (∂Ω,σ) ≤ C f [ B. q, λ (∂Ω,σ)] M .
1

 . q,λ M
Also, formula (3.3.104) remains true for f = ( fα )1≤α ≤M ∈ B1 (∂Ω, σ)
and the operator
 . q,λ M  M
∂νA Dmod : B1 (∂Ω, σ) −→ B q,λ (∂Ω, σ) defined as
 A  . q,λ M (3.3.108)
∂ν Dmod ) f := ∂νA(Dmod f ) for each f ∈ B1 (∂Ω, σ)

is well defined, linear, and bounded, when the domain space is equipped with
the semi-norm induced by [69, (11.13.66)]. Finally,
   . q,λ  M  M
∂νA Dmod : B1 (∂Ω, σ) ∼ −→ B q,λ (∂Ω, σ) defined as
 A   . q,λ M
∂ν Dmod [ f ] := ∂νA(Dmod f ) for each f ∈ B1 (∂Ω, σ)
(3.3.109)
is a well-defined linear and bounded operator, assuming the quotient space is
endowed with the semi-norm5 introduced in [69, (11.13.70)].

. p,λ are also valid

(5) Analogous properties to those presented in items (1)-(3) above
for vanishing Morrey-based homogeneous Sobolev spaces M1 (∂Ω, σ) (cf. [69,
Definition 11.13.15], or (A.0.159)-(A.0.160)) in place of Morrey-based homo-
geneous Sobolev spaces.

4 recall from [69, Proposition 11.13.10] that said semi-norm is in fact a genuine norm if Ω ⊆ R n
is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
5 from [69, Proposition 11.13.10] we know that this semi-norm is actually a genuine norm if Ω ⊆ R n
is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 417
 . p,λ M
Proof That the jump-formula (3.3.103) is true for each f ∈ M1 (∂Ω, σ) follows
from (1.8.27) (bearing in mind (A.0.157)). The claims in (3.3.102) are consequences
of (A.0.157), (1.8.7), (1.8.27), (1.8.20), (1.8.13), [70, Theorem 2.6.1], and [70,
Theorem 1.4.2].
Also, (3.3.104) is a consequence of (1.8.83) and [69, Definition 11.13.1] (cf.
(A.0.157)-(A.0.158)). Having established (3.3.104), the claims made in relation to
(3.3.105) follow with the help of [70, Theorem 2.6.1] and [70, Theorem 1.4.2].
The claim pertaining to (3.3.106) is a consequence of what we have proved so far
and (1.8.10), which implies that ∂νA Dmod annihilates constants. Finally, the claims in
items (4) and (5) are justified in a similar fashion. 

Remark 3.3.7 The results in Theorem 3.3.6 are applicable to all modified boundary-
to-boundary double layer potential operators Dmod , Kmod described in Exam-
ples 1.8.4-1.8.7. In addition, the operator ∂νA Dmod vanishes identically in the context
of (3.3.105) and (3.3.108) when Dmod is as in (1.8.108) (see the last part in Re-
mark 1.4.6). The same the peculiarity (i.e., that ∂νA Dmod from vanishes identically)
is present when Dmod is as in (1.8.111) (see (1.4.33)).

The next item on the current agenda is to establish mapping properties for the
modified boundary-to-boundary double layer potential operator acting on the scales
of Morrey-based and block-based Sobolev spaces in the geometric setting described
in the theorem below.

Theorem 3.3.8 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain  satisfying

 a local
αβ
John condition, and abbreviate σ := H n−1 ∂Ω. Also, let L = ar s ∂r ∂s 1≤α,β ≤M
be a homogeneous, weakly elliptic, constant (complex) coefficient, second-order
M × M system in Rn (for some integer M ∈ N). In this context, consider the modified
boundary-to-boundary double layer potential operator Kmod from (1.8.24)-(1.8.25).
Finally, select some integrability exponents p, q ∈ (1, ∞) along with a parameter
λ ∈ (0, n − 1). Then the following statements are valid.
(1) The modified boundary-to-boundary double layer potential operator induces a
mapping
 . p,λ M  . p,λ M
Kmod : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) (3.3.110)

which is well defined, linear, and bounded, when the spaces involved are endowed
with the semi-norm (A.0.158). As a corollary of (3.3.110) and (1.8.28), the
operator
   . p,λ  M  . p,λ  M
Kmod : M1 (∂Ω, σ) ∼ −→ M1 (∂Ω, σ) ∼ defined as
     . p,λ  M  . p,λ M
Kmod [ f ] := Kmod f ∈ M1 (∂Ω, σ) ∼ , ∀ f ∈ M1 (∂Ω, σ)
(3.3.111)
418 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

is well defined, linear, and bounded, assuming all quotient spaces are endowed
with the semi-norm6 introduced in [69, (11.13.51)].
(2) If U jk with j, k ∈ {1, . . . , n} is the family of singular integral operators defined
in (1.5.251), then
 
∂τ j k Kmod f = K(∂τ j k f ) + U jk (∇tan f ) at σ-a.e. point on ∂Ω
 . p,λ M (3.3.112)
for each f ∈ M1 (∂Ω, σ) and each j, k ∈ {1, . . . , n}.

(3) Similar properties to those described in items (1)-(2) are valid for block-based
homogeneous Sobolev spaces in place Morrey-based homogeneous Sobolev
spaces. Specifically,
 . q,λ M  . q,λ M
Kmod : B1 (∂Ω, σ) −→ B1 (∂Ω, σ) (3.3.113)

is a well-defined, linear, and bounded operator when the spaces involved are
endowed with the semi-norm [69, (11.13.66)]. Also,
   . q,λ  M  . q,λ  M
Kmod : B1 (∂Ω, σ) ∼ −→ B1 (∂Ω, σ) ∼ defined as
     . q,λ  M  . q,λ M
Kmod [ f ] := Kmod f ∈ B1 (∂Ω, σ) ∼ , ∀ f ∈ B1 (∂Ω, σ)
(3.3.114)
is a well-defined linear and bounded mapping, when all quotient spaces are
endowed with the semi-norm7 introduced in [69, (11.13.70)]. Finally,
 
∂τ j k Kmod f = K(∂τ j k f ) + U jk (∇tan f ) at σ-a.e. point on ∂Ω
 . q,λ M (3.3.115)
for each f ∈ B1 (∂Ω, σ) and each j, k ∈ {1, . . . , n}.

. p,λ are also valid

(4) Analogous properties to those presented in items (1)-(2) above
for vanishing Morrey-based homogeneous Sobolev spaces M1 (∂Ω, σ) (cf. [69,
Definition 11.13.15], or (A.0.159)-(A.0.160)) in place of Morrey-based homo-
geneous Sobolev spaces.
Proof Fix an aperture parameter κ ∈ (0, ∞) along with a truncation parameter
ε ∈ (0, ∞). Next, consider an arbitrary function
 . p,λ M  1 σ(x)  p
M
f ∈ M1 (∂Ω, σ) ⊆ L ∂Ω, ∩ L (∂Ω, σ) (3.3.116)
1 + |x| n loc

(cf. (A.0.157)), and define u := Dmod f in Ω. Then from (3.3.116), (3.3.102), (1.8.22),
and the jump-formula (3.3.103) we see that
6 recall from [69, Proposition 11.13.10] that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
7 according to [69, Proposition 11.13.12] said semi-norm is actually a true norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 419
 M κ−n.t.  
u ∈ 𝒞∞ (Ω) , u∂Ω = 12 I + Kmod f at σ-a.e. point on ∂Ω,
Nκε u belongs to Lloc (∂Ω, σ),
p
Nκ (∇u) belongs to M p,λ (∂Ω, σ), (3.3.117)
and Nκ (∇u) M p, λ (∂Ω,σ)
≤C f . p, λ ,
[ M1 (∂Ω,σ)] M

for some constant C = C(Ω, L, κ, p, λ) ∈ (0, ∞) independent of f . Granted these

properties and the current geometric assumptions, we may invoke [69, Proposi-
tion 11.13.8] to conclude that
κ−n.t.  . p,λ M
u∂Ω belongs to the space M1 (∂Ω, σ)
κ−n.t. (3.3.118)
and u ∂Ω
. p, λ
[ M1 M ≤ C f
(∂Ω,σ)]
. p, λ M.
[ M1 (∂Ω,σ)]

Then from (3.3.117) and (3.3.118) we deduce that

 . p,λ M
Kmod f belongs to the space M1 (∂Ω, σ)
(3.3.119)
and Kmod f . p, λ ≤C f . p, λ ,
[ M1 (∂Ω,σ)] M [ M1 (∂Ω,σ)] M

from which the claims regarding (3.3.110) follow. Next, the claims concerning
the operator (3.3.111) are readily seen from what we have just proved, (1.8.28),
and definitions. Also, formula (3.3.112) is a consequence of Proposition 1.8.8 and
[69, Definition 11.13.1] (cf. (A.0.157)-(A.0.158)). Finally, the claims in items (3)-
(4) are established in a similar fashion, now making use of the trace results from
[69, Proposition 11.13.14] (as well as the embedding in [69, (6.2.71)]), and [69,
Proposition 11.13.17] (together with (5.1.53)), respectively. 

Remark 3.3.9 The results in Theorem 3.3.8 are applicable to all modified boundary-
to-boundary double layer potential operators Kmod described in Examples 1.8.4-
1.8.7.

Changing topics, we now turn to integral representation formulas involving null-

solutions of weakly elliptic second-order systems in NTA domains with Ahlfors
regular boundaries. The goal is to produce a version of Theorem 1.8.19 with Morrey
and block spaces playing the role of the Lebesgue spaces in the formulation of said
result.

Theorem 3.3.10 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain such that ∂Ω

is an Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric
αβ 
measure theoretic outward unit normal to Ω. Let A = ar s 1≤r,s ≤n (where M ∈ N)
1≤α,β ≤M
be a complex coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (3.3.120)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Once again, recall

the modified version of the double layer operator Dmod acting on functions from
420 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
   M
L 1 ∂Ω, 1+σ(x)
|x | n as in (1.8.6), and the modified version of the single layer opera-
   M
tor 𝒮mod acting on functions from L 1 ∂Ω, 1+σ(x) |x | n−1
as in (1.5.50). Finally, fix an
aperture parameter κ ∈ (0, ∞) along with an integrability exponent p ∈ (1, ∞) and
a number λ ∈ (0, n − 1). In this setting, consider a function u : Ω → C M satisfying
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ M p,λ (∂Ω, σ). (3.3.121)

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.122)
B(0,λ R)\B(0,R)

Then
κ−n.t.  . p,λ M
u∂Ω exists σ-a.e. on ∂Ω and belongs to M1 (∂Ω, σ) ,
(3.3.123)
κ−n.t.  M
(∇u)∂Ω exists σ-a.e. on ∂Ω and ∂νAu belongs to M p,λ (∂Ω, σ) ,

and there exists some C M -valued locally constant function cu in Ω with the property
that  κ−n.t.   
u = Dmod u∂Ω − 𝒮mod ∂νAu + cu in Ω. (3.3.124)
In particular, if in place of the last condition in (3.3.121) one assumes
that Nκ (∇u) belongs to M̊ p,λ (∂Ω, σ) (which is a subspace of M p,λ (∂Ω, σ); cf.
κ−n.t.  . p,λ M
(A.0.149)) then u∂Ω ∈ M1 (∂Ω, σ) , the conormal derivative ∂νAu belongs
 M
to M̊ p,λ (∂Ω, σ) , and the integral representation formula (3.3.124) continues to
hold.
Finally, if the last condition in (3.3.121) is changed to Nκ (∇u) ∈ B q,λ (∂Ω, σ)
κ−n.t.
for some exponent q ∈ (1, ∞) then u∂Ω exists at σ-a.e. point on ∂Ω and belongs
 . q,λ M κ−n.t.
to B1 (∂Ω, σ) , the trace (∇u)∂Ω exists at σ-a.e. point on ∂Ω, the conormal
 M
derivative ∂νAu belongs to B q,λ (∂Ω, σ) , and the integral representation formula
(3.3.124) once again continues to hold.
κ−n.t.
Proof The current assumptions and [69, Proposition 11.13.7] imply that u∂Ω
 . p,λ M
exists at σ-a.e. point on ∂Ω and belongs to M1 (∂Ω, σ) . From [69, (6.2.7)]
and [68, Proposition 8.4.9] we also see that there exists some ε > 0 such that
Nκε u ∈ Lloc (∂Ω, σ). The present hypotheses on Ω ensure (cf. [68, (5.10.24)]) that Ω
p

is a UR domain. Bearing this in mind, the Fatou-type result from [70, Theorem 3.3.4]
κ−n.t.
ensures that the nontangential boundary trace (∇u)∂Ω exists (in Cn·M ) at σ-a.e.
 M
point on ∂Ω. As such, ∂νAu is well defined and belongs to M p,λ (∂Ω, σ) (cf.
[70, (1.7.16)], [69, (6.2.3)], and [69, (6.2.5)]). Thus, all conditions in (1.8.174) are
satisfied, and this permits us to invoke Theorem 1.8.17 to conclude that (3.3.124)
holds. Next, the claims made in the scenario in which the last condition in (3.3.121)
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 421

is strengthened to Nκ (∇u) ∈ M̊ p,λ (∂Ω, σ) are seen from what we have proved so
far, (A.0.149), [69, Proposition 11.13.16], and [69, (6.2.17), (6.2.18)]. Finally, the
claims in the very last portion of the statement of the theorem are justified in a similar
fashion, bearing in mind [69, (6.2.71)]. 
We continue by presenting a variant of Corollary 1.8.18 in which the Lebesgue
spaces are replaced by appropriate Morrey and block spaces. In contrast to Theo-
rem 3.3.10, this is formulated in an Ahlfors regular domain satisfying a local John
condition.

Theorem 3.3.11 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an Ahlfors regular domain

satisfying a local John condition. Abbreviate σ := H n−1 ∂Ω and denote
 αβ  by ν the
geometric measure theoretic outward unit normal to Ω. Let A = ar s 1≤r,s ≤n
1≤α,β ≤M
(where M ∈ N) be a complex coefficient tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (3.3.125)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . In this setting, bring

in the modified version of the double layer operator Dmod acting on functions from
 1  M
L ∂Ω, 1+σ(x)
|x | n as in (1.8.6), and the modified version of the single layer operator
   M
𝒮mod acting on functions from L 1 ∂Ω, 1+σ(x) |x | n−1
as in (1.5.50). Finally, fix an
aperture parameter κ ∈ (0, ∞) along with a truncation parameter ε ∈ (0, ∞), an
integrability exponent p ∈ (1, ∞), a number λ ∈ (0, n − 1), and consider a function
u : Ω → C M satisfying:
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
κ−n.t. κ−n.t.
u∂Ω and (∇u)∂Ω exist σ-a.e. on ∂Ω,
(3.3.126)
Nκε u
p
belongs to the space Lloc (∂Ω, σ),

Nκ (∇u) belongs to the space M p,λ (∂Ω, σ).

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.127)
B(0,λ R)\B(0,R)

Then κ−n.t.  . p,λ M

u∂Ω belongs to M1 (∂Ω, σ) ,
(3.3.128)
 M
∂νAu belongs to M p,λ (∂Ω, σ) ,
and there exists some C M -valued locally constant function cu in Ω with the property
that  κ−n.t.   
u = Dmod u∂Ω − 𝒮mod ∂νAu + cu in Ω. (3.3.129)
422 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

In particular, if in place of the last condition in (3.3.126) one assumes that Nκ (∇u)
belongs to M̊ p,λ (∂Ω, σ) (which is a subspace of M p,λ (∂Ω, σ); cf. (A.0.149)) then
κ−n.t.  . p,λ M  M
u∂Ω ∈ M1 (∂Ω, σ) the conormal derivative ∂νAu belongs to M̊ p,λ (∂Ω, σ) ,
and the integral representation formula (3.3.129) continues to hold.
Finally, if for some exponent q ∈ (1, ∞) the last two lines in (3.3.126) are changed
to
Nκε u ∈ Llocλ (∂Ω, σ) with qλ := n−1+λ(q−1)
q q(n−1)
∈ (1, q),
(3.3.130)
and Nκ (∇u) belongs to the space B q,λ (∂Ω, σ),
κ−n.t.  . q,λ M
then u∂Ω belongs to B1 (∂Ω, σ) , the conormal derivative ∂νAu belongs to
 q,λ M
B (∂Ω, σ) , and the integral representation formula (3.3.129) once again con-
tinues to hold.
Proof In view of the current geometric assumptions, from (3.3.126), [69, Proposi-
tion 11.13.8], and (A.0.157) we conclude that
κ−n.t.  . p,λ M   σ(x)   M
u∂Ω ∈ M1 (∂Ω, σ) → L 1 ∂Ω, . (3.3.131)
1 + |x| n
In particular, the first membership in (3.3.128) holds. Also, (A.0.184), (3.3.126),
[68, (8.9.8)], [68, Corollary 8.9.6], [69, (6.2.3)], and [69, (6.2.5)] we see that the
second membership in (3.3.128) is true as well.
Going further, we note that [69, (6.2.25)] implies
 σ(x) 
M p,λ (∂Ω, σ) → L 1 ∂Ω, . (3.3.132)
1 + |x| n−1
From (3.3.126), (3.3.127), (3.3.131), and (3.3.132) we then see that all hypotheses
of Theorem 1.8.17 are presently satisfied. As such, (1.8.176) implies (3.3.129).
Let us now replace the last condition in (3.3.126) with the assumption that Nκ (∇u)
belongs to M̊ p,λ (∂Ω, σ). Then [69, Proposition 11.13.17] guarantees that the function
κ−n.t.  . p,λ M
u∂Ω belongs to the space M1 (∂Ω, σ) . Also, (A.0.184), [68, (8.9.8)], [68,
Corollary 8.9.6], and [69, (6.2.17), (6.2.18)] we see that the conormal derivative
 M
∂νAu now belongs to M̊ p,λ (∂Ω, σ) . Finally, since M̊ p,λ (∂Ω, σ) is a subspace of
M p,λ (∂Ω, σ) (cf. (A.0.149)), from what we have proved earlier we conclude that the
integral representation formula (3.3.129) continues to hold in this case.
As regards the claims in the last paragraph of the statement, fix some exponen-
t q ∈ (1, ∞) and replace the last two lines in (3.3.126) by (3.3.130). Then [69,
Proposition 11.13.13] and (A.0.38) guarantee that
κ−n.t.  . q,λ M   σ(x)   M
u∂Ω ∈ B1 (∂Ω, σ) → L 1 ∂Ω, . (3.3.133)
1 + |x| n
In addition, (A.0.184), the first two lines in (3.3.126), (3.3.130), [68, (8.9.8)], [68,
Corollary 8.9.6], [69, (6.2.75)], and [69, (6.2.74)] ensure that the conormal derivative
 M
∂νAu now belongs to the space B q,λ (∂Ω, σ) . Next, we note that [69, (6.2.71)]
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 423

and [68, (7.7.106)] imply

 σ(x) 
B q,λ (∂Ω, σ) → L 1 ∂Ω, . (3.3.134)
1 + |x| n−1
Collectively, the first two lines in (3.3.126), (3.3.130), (3.3.127), (3.3.133), and
(3.3.134) imply that all hypotheses of Theorem 1.8.17 are presently satisfied. As a
result, (1.8.176) follows from (3.3.129). 
Together with Theorem 1.8.17, Theorem 3.3.11 is one of the main ingredients in
the proof of certain operator identities akin to those established in Theorem 1.8.26,
now involving Morrey-based and block-based homogeneous Sobolev spaces.
Theorem 3.3.12 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain. Denote
by ν the geometric measure theoretic outward
 αβ unit normal to Ω and abbreviate
σ := H n−1 ∂Ω. For some M ∈ N, let A = ar s 1≤r,s ≤n be a complex coefficient
1≤α,β ≤M
tensor with the property that
 αβ 
L := ar s ∂r ∂s 1≤α,β ≤M (3.3.135)

(i.e., L := L A) is a weakly elliptic M × M system in Rn . Fix some integrability

exponents p, q ∈ (1, ∞) along with a parameter λ ∈ (0, n − 1), then recall the
operators Smod from Theorem 3.3.5, ∂νA Dmod from Theorem 3.3.6, and associate the
modified boundary-to-boundary double layer potential operator Kmod with Ω and A
as in (1.8.24)-(1.8.25). Finally, let K A#  be the operator associated with the coefficient
tensor A and the set Ω as in (1.3.72). Then the following statements are true.
 M
(1) For each function f belonging to either M p,λ (∂Ω, σ) (in particular, to
 p,λ M  q,λ M
M̊ (∂Ω, σ) ; cf. [69, (6.2.15)]), or B (∂Ω, σ) , there exists c f , which
is the nontangential trace on ∂Ω of some C M -valued locally constant function
in Ω, with the property that
   
Smod K A#  f = Kmod Smod f + c f at σ-a.e. point on ∂Ω. (3.3.136)

In particular, if ∂Ω is connected then

    
Smod K A#  = Kmod Smod acting from either
 p,λ M  p,λ M  M (3.3.137)
M (∂Ω, σ) , M̊ (∂Ω, σ) , or B q,λ (∂Ω, σ) .
 M
(2) For each function f belonging to either M p,λ (∂Ω, σ) (in particular, to
 p,λ M  q,λ M
M̊ (∂Ω, σ) ; cf. [69, (6.2.15)]), or B (∂Ω, σ) , at σ-a.e. point on ∂Ω
one has
1    
#    
2 I + K A − 12 I + K A#  f = ∂νA Dmod Smod f . (3.3.138)

In particular,
424 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces
1     
2I − 12 I + K A#  = ∂νA Dmod Smod acting from either
+ K A# 
 p,λ M  p,λ M  M (3.3.139)
M (∂Ω, σ) , M̊ (∂Ω, σ) , or B q,λ (∂Ω, σ)/∼ .

(3) Strengthen the original hypotheses by now also assuming that Ω satisfies a local
 . p,λ M
John condition. Then for each function f belonging to either M1 (∂Ω, σ)
 . p,λ M
(hence, in particular, to the space M1 (∂Ω, σ) ; cf. [69, (11.13.87)]), or
 . q,λ M
B1 (∂Ω, σ) , there exists c f , which is the nontangential trace on ∂Ω of
some C M -valued locally constant function in Ω, with the property that at σ-a.e.
point on ∂Ω one has
1   1     
2 I + Kmod − 2 I + Kmod f = Smod ∂νA Dmod f + c f . (3.3.140)

In particular, if ∂Ω is connected then

1    1     
2 I + Kmod − 2 I + Kmod = Smod ∂νA Dmod acting from either
 . p,λ M  . p,λ M  . q,λ M
M1 (∂Ω, σ)/∼ , M1 (∂Ω, σ)/∼ , or B1 (∂Ω, σ)/∼ .
(3.3.141)

(4) Continue to impose the additional assumption that Ω satisfies a local John
 . p,λ M
condition. Then for each function f belonging to either M1 (∂Ω, σ) (in
 . p,λ M  . q,λ M
particular, to M1 (∂Ω, σ) ; cf. [69, (11.13.87)]), or B1 (∂Ω, σ) , one
has
 A    
∂ν Dmod Kmod f = K A#  ∂νA Dmod f at σ-a.e. point on ∂Ω. (3.3.142)

In particular,
    
∂νA Dmod Kmod = K A#  ∂νA Dmod acting from either
 . p,λ M  . p,λ M  . q,λ M
M1 (∂Ω, σ)/∼ , M1 (∂Ω, σ)/∼ , or B1 (∂Ω, σ)/∼ .
(3.3.143)

Proof To set the stage, fix an aperture parameter κ ∈ (0, ∞) along with a truncation
parameter ε ∈ (0, ∞). Consider the claims made in item (1). Having pick an arbitrary
function
 M   σ(x)  M
p
f ∈ M p,λ (∂Ω, σ) → L 1 ∂Ω, ∩ L (∂Ω, σ) , (3.3.144)
1 + |x| n−1 loc

(cf. [69, (6.2.25)]), define u := 𝒮mod f in Ω and note that

3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 425
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,

Nκ (∇u) ∈ M p,λ (∂Ω, σ), Nκε u ∈ Lloc 1 (∂Ω, σ),

κ−n.t.
(∇u) ∂Ω
exists at σ-a.e. point on ∂Ω, (3.3.145)
 
∂νAu = − 12 I + K A#  f at σ-a.e. point on ∂Ω,
κ−n.t.  p,λ M    M
u∂Ω = Smod f ∈ M1 (∂Ω, σ) → L 1 ∂Ω, 1+σ(x)
|x | n ,

thanks to (3.3.144), (3.3.94), (1.5.71), (3.3.5), (3.3.92), and (A.0.157). Moreover, if

Ω is an exterior domain then from (1.5.51) and [70, (1.4.24)] we see that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.146)
B(0,2R)\B(0,R)

Granted these properties, we may invoke Theorem 1.8.17 and conclude from
(1.8.176) that there exists some C M -valued locally constant function cu in Ω with
the property that
   
u = Dmod Smod f − 𝒮mod − 12 I + K A#  f + cu in Ω. (3.3.147)
κ−n.t.
Taking nontangential boundary traces in (3.3.147) and denoting c f := cu ∂Ω then
yields
    
Smod f = 12 I + Kmod Smod f − Smod − 12 I + K A#  + c f , (3.3.148)

on account of the last line in (3.3.145), (1.8.27), (3.3.1), and the last property in
(3.3.94). With (3.3.148) in hand, simple algebra then establishes (3.3.136) for arbi-
 M
trary functions f belonging to M p,λ (∂Ω, σ) (hence, in particular, for arbitrary
 p,λ M
functions in M̊ (∂Ω, σ) ; cf. [69, (6.2.15)]).
 M
The proof of (3.3.136) for arbitrary functions f ∈ B q,λ (∂Ω, σ) is carried out
along similar lines, making use of item (4) in Theorem 3.3.5 (and also keeping in
mind [69, Definition 11.13.11], or (A.0.38)-(A.0.40), together with the embeddings
in [69, (6.2.71)] and [68, (7.7.106)]). Finally, the claims in (3.3.137) is a direct
consequence of (3.3.136). This takes care of (1).
Moving on, justifying (3.3.138) proceeds as above, up to (3.3.147), at which
point we apply conormal derivatives to all terms involved. Thanks to (1.5.58) and
Theorem 3.3.6, this ultimately proves (3.3.138) (bearing in mind that ∂νA cu = 0; cf.
(A.0.184)). In turn, this readily implies (3.3.139). All claims in item (2) are therefore
established.
To deal with items (3)-(4), work under the stronger assumptions that Ω is a UR
domain satisfying a local John condition. To get going, pick an arbitrary function
 . p,λ M   σ(x)  M
p
f ∈ M1 (∂Ω, σ) → L 1 ∂Ω, ∩ L (∂Ω, σ) , (3.3.149)
1 + |x| n loc
426 3 Layer Potential Operators on Calderón, Morrey-Campanato, and Morrey Spaces

(see (A.0.157)), and define u := Dmod f in Ω. Then u is well defined (cf. item (1) in
Theorem 1.8.2) and satisfies:
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,

Nκ (∇u) ∈ M p,λ (∂Ω, σ), Nκε u ∈ Lloc (∂Ω, σ),

p

κ−n.t.
(∇u)∂Ω exists at σ-a.e. point on ∂Ω, (3.3.150)
 
∂νAu = ∂νA Dmod f at σ-a.e. point on ∂Ω,
κ−n.t.    . p,λ M    M
u∂Ω = 12 I + Kmod f ∈ M1 (∂Ω, σ) → L 1 ∂Ω, 1+σ(x)
|x | n ,

by virtue of (3.3.149), item (1) in Theorem 3.3.6, item (1) in Theorem 3.3.8, (1.8.22),
and (1.8.20). In addition, when Ω is an exterior domain we see from (3.3.149),
(1.8.11), and [70, (1.4.24)] that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.151)
B(0,2R)\B(0,R)

Once the aforementioned properties have been validated, Theorem 3.3.11 applies
(see (3.3.126)) and (3.3.129) presently implies that there exists some C M -valued
locally constant function cu in Ω with the property that
   
u = Dmod 12 I + Kmod f − 𝒮mod ∂νA Dmod f + cu in Ω. (3.3.152)
κ−n.t.
If we now set c f := cu ∂Ω and go nontangentially to the boundary in (3.3.152) we
arrive at
1  1 1   A 
2 I + Kmod f = 2 I + Kmod 2 I + Kmod f − Smod ∂ν Dmod + c f , (3.3.153)

after taking into account (3.3.150), (3.3.110), (3.3.103), (3.3.105), and the last prop-
erty in (3.3.94). In turn, (3.3.153) readily establishes (3.3.140) for arbitrary func-
 . p,λ M
tions f belonging to M1 (∂Ω, σ) (hence, in particular, for arbitrary functions
 . p,λ M
in M1 (∂Ω, σ) ; cf. [69, (11.13.87)]).
 . q,λ M
The version of (3.3.140) for arbitrary functions f ∈ B1 (∂Ω, σ) is dealt with
analogously, now relying on item (4) in Theorem 3.3.5, item (4) in Theorem 3.3.6,
item (3) in Theorem 3.3.8, and keeping in mind [69, Definition 11.13.11] (cf.
(A.0.38)-(A.0.40)). Upon noting that (3.3.141) is a direct consequence of (3.3.140),
the treatment of item (3) is complete.
Finally, the proof of (3.3.142) proceeds as above, up to the integral representation
formula (3.3.152), at which stage we now take the conormal derivatives of all
terms involved. In view of Theorem 3.3.5, Theorem 3.3.6, Theorem 3.3.8, and
Theorem 3.3.1, this leads to (3.3.138) (once again, keeping in mind that ∂νA cu = 0;
cf. (A.0.184)). This takes care of the current item (4), and finishes the proof of the
theorem. 
3.3 Boundary Layer Potential Operators on Morrey Spaces and Their Pre-Duals 427

Remark 3.3.13 It is of particular interest to consider the operator identities present-

ed in Theorem 3.3.12 when they involve the brand of modified boundary-to-boundary
double layer potential operators described in Examples 1.8.4-1.8.5. This is because the
operator ∂νA Dmod vanishes identically in these scenarios (as noted in the last
part in Remark 1.4.6 and in (1.4.33)). As a result, the identity recorded in (3.3.140)
simplifies in each of these instances.
For example, for the modified boundary-to-boundary Cauchy-Clifford integral
Cmod defined in (2.1.178) we obtain from (3.3.140) that for each function f belonging
. p,λ . q,λ
to either M1 (∂Ω, σ) ⊗ C n or B1 (∂Ω, σ) ⊗ C n there exists c f , which is the
nontangential trace on ∂Ω of some C n -valued locally constant function in Ω, with
the property that
1   1  
2 I + C mod − 2 I + C mod f = c f at σ-a.e. point on ∂Ω. (3.3.154)
 
Hence, Cmod Cmod f = 14 f + c f on ∂Ω. In particular, if ∂Ω is connected and we let
   
Cmod be the operator [ f ] → Cmod f (with brackets denoting equivalence classes
modulo constants), we therefore obtain
 2
Cmod = 14 I (3.3.155)
 . p,λ   . q,λ 
on M1 (∂Ω, σ)/∼ ⊗ C n and B1 (∂Ω, σ)/∼ ⊗ C n . See also (1.8.279) and
(2.1.186)-(2.1.189) for related identities.
Chapter 4
Layer Potential Operators Acting from
Boundary Besov and Triebel-Lizorkin Spaces

Besov and Triebel-Lizorkin spaces on Ahlfors regular subsets of Rn have been

discussed at length in [69, Chapter 18]. The primary goal in this chapter is to
explore the mapping properties of singular integral operators of layer potential
type acting from (boundary) Besov and Triebel-Lizorkin spaces considered in very
broad geometric settings. Given the nature of the spaces involved, the operators in
question are expected to have certain inherent cancelation properties (see, e.g., the
boundedness criterion for an operator T from [32, Theorem A, p. 129] requiring that
T(1) = 0 in a suitable sense; cf. also [22, Theorem 4.26, p. 123] in this regard), which
generic Calderón-Zygmund convolution type operators typically fail to satisfy. This
explains the focus on singular integral operators of layer potential type adopted here.
From this perspective, we find it instructive to briefly elaborate on the crucial
cancelation property “K(1) = 0” alluded to above, for a boundary-to-boundary
double layer operator K associated with a given weakly elliptic second-order M × M
system and a UR domain Ω ⊆ Rn with compact boundary. Using the brand of
“wavelets” introduced in [22, Definition 2.2, p. 28], this ultimately boils down to
checking that
∫
K # ψ dσ = 0 for each C M -valued wavelet ψ on ∂Ω, (4.0.1)
∂Ω

where K # stands for the transpose double layer and σ := H n−1 ∂Ω is the “surface
measure” on the topological boundary of Ω. It turns out that the size and cancelation
properties of such a vector-valued wavelet ψ on ∂Ω permit us to conclude that actually
 M
ψ belongs to the Hardy space H 1 (∂Ω, σ) . This is the manner in which the present
enterprise ties up with analysis from §2.1, since Corollary 2.1.9 then guarantees that
(4.0.1) holds. For the Besov scale, this ultimately proves that in the setting described
above any boundary-to-boundary double layer induces a well-defined linear and
bounded mapping

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 429
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_4
430 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 p,q M  p,q M
K : Bs (∂Ω, σ) −→ Bs (∂Ω, σ) ,
    (4.0.2)
p ∈ n−1
n ,∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1.

The next step is to use duality in order to obtain boundedness results for the transpose
 p,q M
double layer K # on Besov spaces Bs (∂Ω, σ) in the range 1 < p, q < ∞ and
0 < s < 1 (where duality works). We desire to further augment this range by
allowing sub-unital exponents p, q (and negative smoothness). The idea is to start
M
from the fact that K # is known to be bounded on the Hardy scale [H p (∂Ω, σ)
 
with p ∈ n−1 n , 1 and then employ [69, Theorem 7.8.21] where the envelopes of
Hardy spaces (considered in the sense discussed at length in the first part of [69,
§7.8]) have been identified as being Besov spaces of the nature we currently desire.
This typifies the flavor of §4.1, where the Triebel-Lizorkin scale and other singular
integral operators of interest are also considered.
Changing gears, in §4.2 we consider mapping properties of boundary-to-domain
layer potential operators acting from Besov spaces into some suitably weighted
Sobolev spaces, in the context of open sets with compact Ahlfors regular boundaries.
In this regard, we first establish mapping properties for certain boundary-to-domain
integral operators acting from Besov spaces with a positive amount of smoothness
into Lebesgue spaces suitably weighted in terms of the distance to the boundary. See
Theorem 4.2.1 and Theorem 4.2.6, where we identify the key qualities of the integral
operators in question which produce desirable results of this flavor. The mapping
properties thus obtained are subsequently specialized to double layer and single
layer operators, and some of the main results derived in this fashion are contained
in Theorem 4.2.3 and Theorem 4.2.10. See also Theorem 4.2.5 where the boundary
behavior of the double layer potential operator acting from boundary Besov spaces
into weighted Sobolev spaces is discussed. A program of similar aims is carried
out in §4.3, this time in relation to the action of boundary-to-domain layer potential
operators acting from (boundary) Besov spaces into Besov and Triebel-Lizorkin
spaces (defined in open sets).
In §4.4, the final section in this chapter, we derive a number of useful inte-
gral representation formulas of layer potential type in the context of Besov and
Triebel-Lizorkin spaces. For example, in Theorem 4.4.1 we prove a basic integral
representation formula in such a setting, to the effect that a function which is a
null-solution of a weakly elliptic system may be expressed as the action of the
(boundary-to-domain) double layer operator on the boundary trace of said function
and the action of the (boundary-to-domain) single layer operator on the conormal
derivative of the original function. Other results of similar flavor are subsequently
derived (including some in relation to first-order systems). In turn, such integral
representation formulas are used to conclude a number of remarkable properties for
functions in Besov and Triebel-Lizorkin spaces (in open subsets of Rn ) which are
null-solutions of a weakly elliptic system.
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 431

4.1 Boundary-to-Boundary Layer Potentials from Besov and

Triebel-Lizorkin Spaces into Themselves

Recall the principal-value (p.v.) singular integral operator K associated with a given
a weakly elliptic system L and a UR domain Ω ⊆ Rn as in (1.3.68). It turns out that
thanks to its specific algebraic nature, the operator K acts naturally on the scales of
Besov and Triebel-Lizorkin spaces on ∂Ω. Specifically, we have the following result.

Theorem 4.1.1 Suppose Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is a UR domain with

compact boundary and abbreviate σ := H n−1 ∂Ω. Also, let L be a homogeneous,
weakly elliptic, constant (complex) coefficient, second-order M × M system in Rn
(for some M ∈ N). In this setting, consider the boundary layer potential operator K
associated with the system L and the set Ω as in (1.3.68).
Then the operator K, originally acting on Lebesgue spaces on ∂Ω (cf. (1.5.14)),
extends uniquely to linear and bounded mappings
 p,q M  p,q M
K : Bs (∂Ω, σ) −→ Bs (∂Ω, σ) ,
    (4.1.1)
n ,∞ ,
p ∈ n−1 q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
 p,q M  p,q M
K : Fs (∂Ω, σ) −→ Fs (∂Ω, σ) ,
     1  (4.1.2)
n ,∞ ,
p ∈ n−1 q ∈ n−1n ,∞ , (n − 1) min{p,q } − 1 + < s < 1.

Moreover, various choices of the exponents yield operators which are compatible
with one another.

Proof Since ∂Ω is compact, from [69, Proposition 7.7.2], and [69, (7.1.55), (7.7.5)]
we see that
 p,q M  M
Bs (∂Ω, σ) → L 1 (∂Ω, σ) whenever
    (4.1.3)
p ∈ n−1n ,∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and

p,q M  M
Fs (∂Ω, σ) → L 1 (∂Ω, σ) whenever
 n−1   n−1   1  (4.1.4)
p ∈ n , ∞ , q ∈ n , ∞ , (n − 1) min{p,q } − 1 + < s < 1.

As a consequence of these embeddings and [70, Theorem 2.3.2], it follows that K,

originally defined as a principal-value singular integral operator on the Lebesgue
 M
space L 1 (∂Ω, σ) (cf. [70, (2.3.19)]), acts on a meaningful manner on functions
belonging to the family of Besov and Triebel-Lizorkin spaces described in (4.1.1)-
(4.1.2).
432 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

To see that actually K induces well-defined, linear, and bounded mappings in

such settings, the idea is to check that the one-sided T(1) theorem from [32, The-
orem A, p. 129] (cf. also [33], [22, Theorem 4.26, p. 123], [56], and [21]), which
guarantees the boundedness of singular integral operators on Besov and Triebel-
Lizorkin spaces considered in the abstract setting of spaces of homogeneous type, is
presently applicable. For starters, from (1.3.68) and Theorem 1.5.1 we see that K is a
Calderón-Zygmund operator in the sense of [22, Definition 1.9, p. 14], whose kernel
satisfies conditions (1.3)-(1.5) on p. 126 of [32] (and also conditions (1.10)-(1.11)
in [22, Definition 1.8, p. 14]). In addition, as noted in the comment following [22,
Definition 1.15, p. 19], the fact that K is bounded on the scale of Lebesgue spaces (cf.
Theorem 1.5.1) guarantees that K satisfies the weak boundedness property (cf. [22,
Definition 1.8, p. 14]). As such, matters are reduced to checking that K(1) = 0. The
latter cancelation condition should be suitably understood. Indeed, according to the
comment following [22, Theorem 4.27, p. 123] (cf. also item (iii) of Theorem 1.5.1),
what this ultimately boils down to is having
∫
K # ψ dσ = 0 for each C M -valued wavelet ψ on ∂Ω. (4.1.5)
∂Ω

The notion of wavelet referred to above is introduced in [22, Definition 2.2, p. 28].
An inspection of this definition (as well as [22, Definition 2.1, p. 27]) reveals that
wavelets are test functions in the sense of [69, Definition 7.1.1] satisfying a vanishing
 ψ : ∂Ω → C has the property that there exist two
moment condition. That is, M

numbers γ ∈ (0, ∞), β ∈ 0, 1], along with a point x0 ∈ ∂Ω and a radius r ∈ (0, ∞)
such that (for some constant C ∈ (0, ∞))
rγ
|ψ(x)| ≤ C   n−1+γ for each x ∈ ∂Ω, (4.1.6)
r + |x − x0 |

r γ |x − y| β
|ψ(x) − ψ(y)| ≤ C   n−1+γ+β
r + |x − x0 | (4.1.7)
whenever x, y ∈ ∂Ω are such that |x − y| < (r + |x − x0 |)/4,
and
∫
ψ dσ = 0. (4.1.8)
∂Ω

The claim that we make in relation to such a function ψ is that, irrespective of

whether ∂Ω is bounded,
 M
ψ ∈ H 1 (∂Ω, σ) . (4.1.9)

In turn, this may be justified by invoking [69, Corollary 4.5.3] with p := 1. Specifi-
cally, choosing
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 433
 
q ∈ (1, ∞) arbitrary and d ∈ (n − 1)(q − 1), (n − 1)(q − 1) + qγ (4.1.10)
   M
ensures (in light of (4.1.6) and [68, (7.2.5)]) that ψ ∈ L q ∂Ω, (1 + | · −x0 | d )σ .
Granted this membership and the cancelation condition (4.1.8), [69, Corollary 4.5.3]
applies and gives (4.1.9). Having proved (4.1.9),∫ we may now rely on Corollary 2.1.9
together with (4.1.8) to conclude that we have ∂Ω K # ψ dσ = 0.
Having established (4.1.5), the door is open for employing the one-sided T(1)
theorem from [32, Theorem A, p. 129]. This gives that K is well defined, linear, and
bounded both in the context of (4.1.1) and (4.1.2). Finally, that various versions of
K in (4.1.1)-(4.1.2) are compatible with one another (as well as with other earlier
versions of K considered so far) is clear from (4.1.3)-(4.1.4). 
Remark 4.1.2 If in place of [32, Theorem A, p. 129] we use the one-sided T(1)
theorem from [22, Theorem 4.26, p. 123] then the same type of argument as in the
proof of Theorem 4.1.1 gives that whenever Ω ⊆ Rn is a UR domain with unbounded
boundary then K also acts naturally on homogeneous Besov and Triebel-Lizorkin
spaces on the boundary, i.e.,
 . p,q M  . p,q M
K : Bs (∂Ω, σ) −→ Bs (∂Ω, σ) ,
    (4.1.11)
p ∈ n−1
n ,∞ , q ∈ (0, ∞), (n − 1) p1 − 1 + < s < 1,

and
 . p,2 M  . p,2 M
K : Fs (∂Ω, σ) −→ Fs (∂Ω, σ) ,
   1  (4.1.12)
p ∈ n−1
n ,∞ , (n − 1) min{p,2} − 1 + < s < 1.

are well-defined, linear, bounded operators, which act in a compatible fashion with
one another.
As commented at length on earlier occasions, there is a large variety of double
layer operators associated with a given weakly elliptic system, and Theorem 4.1.1 is
applicable to all such operators. A concrete case involves the boundary-to-boundary
Cauchy-Clifford integral operator C defined in (A.0.54). Indeed, as noted in Exam-
ple 1.4.12 this is a special case of a double layer (associated with the Laplacian).
In light of this, Theorem 4.1.1 implies that whenever Ω ⊆ Rn is a UR domain with
compact boundary and σ := H n−1 ∂Ω then
p,q p,q
C : Bs (∂Ω, σ) ⊗ Cn −→ Bs (∂Ω, σ) ⊗ Cn,
    (4.1.13)
n ,∞ ,
p ∈ n−1 q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
p,q p,q
C : Fs (∂Ω, σ) ⊗ Cn −→ Fs (∂Ω, σ) ⊗ Cn,
 n−1     1  (4.1.14)
p ∈ n , ∞ , q ∈ n−1 n ,∞ , (n − 1) min{p,q } − 1 + < s < 1,
434 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

are well-defined, linear, and bounded operators. In addition, from [70, (2.5.332)],
(4.1.13)-(4.1.14), and [69, Lemma 7.1.10] it follows that
p,q
C 2 = 14 I on Bs (∂Ω, σ) ⊗ Cn when
    (4.1.15)
p ∈ n−1
n ,∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
p,q
C 2 = 14 I on Fs (∂Ω, σ) ⊗ Cn when
     1  (4.1.16)
n ,∞ ,
p ∈ n−1 q ∈ n−1 n ,∞ , (n − 1) min{p,q } − 1 + < s < 1,

first when q < ∞, then the end-point q = ∞ is included as a result of this and loose
embeddings (cf. items (iii)-(iv) in [69, Proposition 7.7.1]).
Mapping properties analogous to (4.1.13)-(4.1.14) also hold for the boundary-to-
boundary version of the ordinary Cauchy operator in the complex plane, and for the
boundary-to-boundary Bochner-Martinelli integral operator in the context of several
complex variables (cf. Example 1.4.9 and Example 1.4.16). Finally, there are also
results in the spirit of Remark 4.1.2 for all these operators.
Another type of singular integral operator which acts naturally on boundary Besov
and Triebel-Lizorkin spaces with a positive amount of smoothness is discussed in
the theorem below.

Theorem 4.1.3 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain with compact

boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν j )1≤ j ≤n the geometric
measure theoretic outward unit normal to Ω. Next, let b ∈ Lloc1 (Rn, L n ) be a complex-

valued function with the property that b Rn \{0} ∈ 𝒞 (R \ {0}) where N = N(n) ∈ N
N n

is a sufficiently large number, and such that ∇b is odd and positive homogeneous of
degree 1 − n in Rn \ {0}. Finally, for each pair of indices j, k ∈ {1, . . . , n} introduce
the integral operator acting on each function f ∈ L 1 (∂Ω, σ) according to
∫
Tjk f (x) := lim+ ν j (y)(∂k b)(x − y) − νk (y)(∂j b)(x − y) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(4.1.17)

for σ-a.e. x ∈ ∂Ω. Then the operator Tjk extends uniquely to linear and bounded
mappings
p,q p,q
Tjk : Bs (∂Ω, σ) −→ Bs (∂Ω, σ),
 n−1    (4.1.18)
p ∈ n , ∞ , q ∈ (0, ∞), (n − 1) p1 − 1 + < s < 1,

and
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 435
p,q p,q
Tjk : Fs (∂Ω, σ) −→ Fs (∂Ω, σ),
 n−1   1  (4.1.19)
p, q ∈ n , ∞ , (n − 1) min{p,q } − 1 + < s < 1.

Furthermore, different choices of the exponents yield operators which are compatible
with one another.
Proof Given any  ∈ {1, . . . , n} define
∫
B f (x) := (∂ b)(x − y) f (y) dσ(y) at each x ∈ Ω,
∂Ω (4.1.20)
for every function f ∈ L 1 (∂Ω, σ).

Also, fix κ ∈ (0, ∞) along with  ∈ {1, . . . , n}. Then for each f ∈ L 1 (∂Ω, σ) we have
κ−n.t. 1  
B f ∂Ω (x) = ∂ b ν(x) f (x)
2i
∫
+ lim+ (∂ b)(x − y) f (y) dσ(y) (4.1.21)
ε→0
y ∈∂Ω
|x−y |>ε

at σ-a.e. point x ∈ ∂Ω. For each j, k ∈ {1, . . . , n} we may then rely on (4.1.21),
(1.2.18), and (1.2.3) to compute
κ−n.t. κ−n.t.
ν j (x) Bk f ∂Ω (x) − νk (x) B j f ∂Ω (x)
∫
1  
= ν j (x)
∂k b ν(x) f (x) + lim+ ν j (x)(∂k b)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε
∫
1  
− νk (x)∂j b ν(x) f (x) − lim+ νk (x)(∂j b)(x − y) f (y) dσ(y)
2i ε→0
y ∈∂Ω
|x−y |>ε

= (Tjk
#
f )(x) at σ-a.e. point x ∈ ∂Ω. (4.1.22)

To proceed, fix some ψ ∈ H 1 (∂Ω, σ) along with j, k ∈ {1, . . . , n} and define the

vector field
   
F := Bk ψ e j − B j ψ ek in Ω. (4.1.23)

Then from (4.1.23), [70, (2.4.14)] (with p := 1), and (4.1.22) we see that
 n
F ∈ 𝒞1 (Ω) , divF = 0 in Ω, Nκ F ∈ L 1 (∂Ω, σ),
κ−n.t. (4.1.24)
and F ∂Ω = Tjk # ψ at σ-a.e. point x ∈ ∂Ω.
436 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

In the case when Ω is not an exterior domain [68, Theorem 1.2.1] applies and the
Divergence Formula recorded in [68, (1.2.2)] presently gives
∫ ∫ ∫
 κ−n.t. 
#
Tjk ψ dσ = ν · F ∂Ω dσ = divF dL n = 0, (4.1.25)
∂Ω ∂Ω Ω

hence, in this case, ∫

#
Tjk ψ dσ = 0. (4.1.26)
∂Ω

When Ω is an exterior domain, then working with the bounded UR domain Rn \ Ω

in place of Ω (cf. item (7) in [68, Lemma 5.10.9]) and once again arrive at the same
conclusion as in (4.1.26).
Together with item (ii) in Proposition 1.2.1, the cancelation property established
in (4.1.26) then permits us to invoke the one-sided T(1) theorem [32, Theorem A,
p. 129] (cf. also [22, Theorem 4.26, p. 123]) to conclude, much as in the proof
of Theorem 4.1.1, that Tjk induces well-defined, linear, and continuous mappings
(which act in a coherent fashion) in the context of both (4.1.18) and (4.1.19). 
We now proceed to study the action of the “transpose” double layer K # (originally
defined on Lebesgue spaces in (1.3.72)) on Besov and Triebel-Lizorkin scales defined
on boundaries of UR domains. The staring point the following partial result obtained
from Theorem 4.1.1 and duality. Given the format of [69, Proposition 7.6.1], for
now we restricting ourselves to portions of the aforementioned scales consisting of
Banach spaces.

Corollary 4.1.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain with compact

boundary and set σ := H n−1 ∂Ω. Also, let L be a homogeneous, weakly elliptic,
constant (complex) coefficient, second-order M × M system in Rn (for some M ∈ N).
In this context, consider the boundary layer potential operator K # associated with
the system L and the set Ω as in (1.3.72).
Then the operator K # , originally acting on Lebesgue spaces on ∂Ω (cf. (1.5.12)),
extends uniquely to linear and bounded mappings
 p,q M  p,q M
K # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ) ,
(4.1.27)
p ∈ (1, ∞], q ∈ (0, ∞], s ∈ (0, 1),

and
 p,q M  p,q M
K # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ) ,
(4.1.28)
p ∈ (1, ∞), q ∈ (1, ∞], s ∈ (0, 1).

In addition, various choices of the parameters p, q, s yield operators which are

compatible with one another. Finally, if the exponents p, q, p , q ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q and s ∈ (0, 1), then
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 437
   
p, q
[B−s (∂Ω,σ)] M K# f , g = [B−s
p, q
p ,q
[B s (∂Ω,σ)] M f , Kg [B sp , q (∂Ω,σ)] M
(∂Ω,σ)] M
 p,q M  p ,q M
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ) ,
(4.1.29)

and
   
p, q
[F−s (∂Ω,σ)] M K# f , g = [F−s
[Fs
p, q
p ,q (∂Ω,σ)] M f , Kg [Fsp , q (∂Ω,σ)] M
(∂Ω,σ)] M
 p,q M  p ,q M
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ) .
(4.1.30)

Proof First, in the case when q ∈ (1, ∞], the claimed mapping properties are con-
sequence of Theorem 4.1.1 and [69, Proposition 7.6.1] (also bearing in mind [69,
Lemma 7.1.10] and item (iii) in Theorem 1.5.1). Second, the extension to q ∈ (0, ∞]
in (4.1.27) is achieved by invoking the real interpolation result from [69, (7.4.2)].
Finally, (4.1.29)-(4.1.30) are seen from what we have proved so far, item (iii) in
Theorem 1.5.1, [69, (7.1.62)], and Theorem 4.1.1. 

In our next theorem we further augment the results from Corollary 4.1.4 by now
allowing sub-unital integrability exponents.

Theorem 4.1.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain with compact

boundary and set σ := H n−1 ∂Ω. Consider a homogeneous, weakly elliptic, con-
stant (complex) coefficient, second-order M × M system L in Rn (for some M ∈ N),
and bring in the boundary layer potential operator K # associated with the system L
and the set Ω as in (1.3.72).
Then the operator K # , originally considered as in Corollary 4.1.4, further extends,
in a unique fashion, to linear and bounded mappings
 p,q M  p,q M
K # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (4.1.31)
provided s ∈ (0, 1), p ∈ n−1
n−s , ∞], q ∈ (0, ∞],

and
 p,q M  p,q M
K # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ) ,
  (4.1.32)
provided s ∈ (0, 1), p ∈ n−1
n−s , ∞), q ∈ n−1
n−s , ∞].

Moreover, various choices of the parameters p, q, s yield operators which are com-
patible with one another.

Proof Let us first observe

 that
 whenever s, p, q are as in (4.1.31) the lower bound on
p implies s + (n − 1) p1 − 1 < 1, hence we may choose some exponent p ∈ (1, ∞)
with the property that
1 1
s := s + (n − 1) −
 < 1. (4.1.33)
p p
438 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

As such, we may invoke [69, Theorem 7.7.4] to conclude that, with 

q ∈ (1, ∞)
arbitrary, we have a continuous embedding
 p,q M  p, q M
B−s (∂Ω, σ) → B−s (∂Ω, σ) . (4.1.34)

Since from (4.1.27) in Corollary 4.1.4 we already know that K # is well defined on
 p, q M
the larger space B−s (∂Ω, σ) it follows that it is meaningful to consider K # f
 p,q M
whenever f ∈ B−s (∂Ω, σ) with s, p, q as in (4.1.31). In particular, K # stays
compatible with its earlier version considered in Corollary 4.1.4.  n−1 
Suppose now that two arbitrary parameters s∗ ∈ (0, 1) and p∗ ∈ n−s ∗ , 1 have
been given. Define
 
s ∗ −1
p := 1
p∗ + n−1 (4.1.35)

and note that this entails

1 
n−1
n < p < p∗ ≤ 1 and s∗ = (n − 1) p − 1
p∗ . (4.1.36)

Granted this, [69, (7.8.153)] applies and yields

 M  p∗, p∗ M
E p∗ H p (∂Ω, σ) = B−s∗ (∂Ω, σ) . (4.1.37)

In turn, from [69, (7.1.60)], (4.1.37), (2.1.4) in Theorem 2.1.1, and [69, (7.8.56) in
Proposition 7.8.9] we conclude that
 n−1 
whenever s∗ ∈ (0, 1) and p∗ ∈ n−s ∗ , 1 , it follows that the operator
 p,2 M  p,2 M  
K : F0 (∂Ω, σ)
# → F0 (∂Ω, σ) , with p ∈ n−1 n , 1 defined (4.1.38)
as in (4.1.35), extends (in a unique fashion) to a linear and bounded
 p∗, p∗ M  p∗, p∗ M
mapping K # : B−s∗ (∂Ω, σ) → B−s∗ (∂Ω, σ) .

From embeddings (cf. (4.1.34)) we also see that the various extensions K # described
in (4.1.38) act in a coherent fashion with one another, and they are also compatible
with K # from Corollary 4.1.4. For this reason, we agree to drop the “hat”, and simply
refer to K # as K # .  
Consider now s ∈ (0, 1), p ∈ n−1 ∗
n−s , 1 , and q ∈ (0, ∞]. Choose s0 ∈ (0, s) and
∗
s1 ∈ (s, 1) close enough to s so that

n−s0∗ , n−s1∗ < p,

n−1 n−1
max (4.1.39)

then pick θ ∈ (0, 1) such that s = (1 − θ)s0∗ + θs1∗ . Finally, define p∗ := p. Then,
according to the real interpolation result from [69, (7.4.2)] (presently used with
p := p∗ , q0 := p∗ , q1 := p∗ , s0 := −s0∗ , s1 := −s1∗ ) we have
 p ∗, p ∗  M  p∗, p∗ M  p,q M
B−s∗ (∂Ω, σ) , B−s∗ (∂Ω, σ) = B−s (∂Ω, σ) . (4.1.40)
0 1 θ,q
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 439

In concert, (4.1.38), [69, Proposition 1.3.7], and (4.1.40) then prove that the operator
K # is bounded in the context of
 p,q M  p,q M
K # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ) ,
  (4.1.41)
s ∈ (0, 1), p ∈ n−1n−s , 1 , q ∈ (0, ∞].

Another proof of the fact that the operator K # is bounded in the context of
 p, p M
(4.1.41) is as follows. Start with an arbitrary f ∈ B−s (∂Ω, σ) , with s ∈ (0, 1)
 n−1 
and p ∈ n−s , 1 , then invoke [69, Corollary 7.2.9] to decompose it as
  p, p M
f =g+ λ j a j in B−s (∂Ω, σ) , (4.1.42)
j ∈N

for some function

 p, p M   ∗ M
g ∈ B−s (∂Ω, σ) ∩ L p (∂Ω, σ) (4.1.43)
1≤p ∗ ≤∞

satisfying

g[B−s
p, p
(∂Ω,σ)] M + sup g[L p ∗ (∂Ω,σ)] M ≤ C f  p, p
M (4.1.44)
1≤p ∗ ≤∞ [B−s (∂Ω,σ)

and for some family {a j } j ∈N of η-smooth atoms of type (p, −s) on ∂Ω along with
some numerical sequence {λ j } j ∈N ∈  p satisfying
 1/p
|λ j | p ≤ C f [B−s
p, p
(∂Ω,σ)] M . (4.1.45)
j ∈N

From [69, (7.2.18)] and (2.1.8) we then conclude that, for each j ∈ N,

the function m j := K # a j is a fixed multiple of a C M -valued

(po, qo, ε)-molecule on ∂Ω (cf. [69, Definition 4.5.1]), if we (4.1.46)
 
s −1
 
take po := p1 + n−1 ∈ n−1 n , 1 , qo ∈ [1, ∞], and ε := n−1 .
1

In turn, (4.1.46) and [69, (7.2.16)] imply that

for each j ∈ N, the function m j := K # a j is a fixed multiple

of some C M -valued (p, qo, ε, 1 − s)-rough molecule (in the (4.1.47)
sense of [69, Definition 7.2.3]).
Combining (4.1.47) with [69, Theorem 7.2.4] then proves (on account of (4.1.34)
and (4.1.27) in Corollary 4.1.4) that
440 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
   
 #    
K λj aj  p, p
= λj K#aj  p, p
[B−s (∂Ω,σ)] M [B−s (∂Ω,σ)] M
j ∈N j ∈N

 1/p
≤ |λ j | p
j ∈N

≤ C f [B−s
p, p
(∂Ω,σ)] M , (4.1.48)

where the last inequality is provided by (4.1.45). From this, (4.1.42)-(4.1.44),

 p, p M
(1.5.12), and [69, (7.7.8)] we then deduce that K # maps B−s (∂Ω, σ) boundedly
 n−1 
into itself whenever s ∈ (0, 1) and p ∈ n−s , 1 . Via real interpolation (cf. [69, The-
orem 7.4.1]) we then conclude that the operator in (4.1.41) is indeed well defined,
linear, and bounded.
At this stage, the claims pertaining to (4.1.31) now become consequences of
(4.1.27) and (4.1.41).
Alternatively, we could have established the boundedness of K # in the context
of (4.1.31) by relying on the one-sided T(1) theorem from [32, Theorem B, p. 129].
More specifically, the latter result requires several hypotheses which guarantee the
desired boundedness result. First, we need K # to satisfy T ∗ (1) = 0. This is a condition
amounting to K(1) = 0, which has been already checked in the course of the proof
of Theorem 4.1.1. Second, we need K # to satisfy the weak boundedness property
(aka WBP). This follows from the fact that the operator in question is bounded
 M
on L 2 (∂Ω, σ) (cf. item (iii) in Theorem 1.5.1). Finally, K # is required to have
an integral kernel satisfying the estimates demanded in [32, (1.3), (1.4), and (1.6),
p. 126], and this is clear from (1.3.72) and [70, Theorem 1.4.2] (bearing in mind that
we are currently assuming ∂Ω to be compact).
In fact, the one-sided T(1) theorem from [32, Theorem B, p. 129] also gives
boundedness results on the scale of Triebel-Lizorkin spaces in the setting of spaces
of homogeneous type, for a class of operators to which K # belongs (as just noted
above). When specialized to the present setting, this theorem then establishes the
boundedness of K # in the context of (4.1.32). 

Remark 4.1.6 Much as Theorem 4.1.1 has given (4.1.13)-(4.1.14), Theorem 4.1.5
applies to the “transpose” principal-value Cauchy-Clifford operator C # (defined in
(1.6.1)) and shows that, whenever Ω ⊆ Rn is a UR domain with compact boundary
and σ := H n−1 ∂Ω, then
p,q p,q
C # : B−s (∂Ω, σ) ⊗ Cn −→ B−s (∂Ω, σ) ⊗ Cn,
  (4.1.49)
s ∈ (0, 1), p ∈ n−1
n−s , ∞ , q ∈ (0, ∞],

and
p,q p,q
C # : F−s (∂Ω, σ) ⊗ Cn −→ F−s (∂Ω, σ) ⊗ Cn,
  (4.1.50)
s ∈ (0, 1), p ∈ n−1
n−s , ∞), q ∈ n−1
n−s , ∞],
4.1 Boundary-to-Boundary Layer Potentials from Besov and Triebel-Lizorkin Spaces 441

are well-defined, linear, and bounded operators. Moreover, whenever the exponents
p, q, p , q ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q and s ∈ (0, 1), then (compare
with item (ii) in Proposition 1.6.1)
 #   
B−s (∂Ω,σ)⊗ Cn C f , g B p , q (∂Ω,σ)⊗ C = B−s (∂Ω,σ)⊗ Cn f , Cg B p , q (∂Ω,σ)⊗ C
p, q p, q
s n s n

p,q p ,q
for each f ∈ B−s (∂Ω, σ) ⊗ Cn and g ∈ Bs (∂Ω, σ) ⊗ Cn,
(4.1.51)

and
   
p, q
F−s (∂Ω,σ)⊗ Cn C# f , g p ,q
Fs (∂Ω,σ)⊗ Cn
= F−s
p, q
(∂Ω,σ)⊗ Cn f , Cg F p , q (∂Ω,σ)⊗ Cn
s

p,q p ,q
for each f ∈ F−s (∂Ω, σ) ⊗ Cn and g ∈ Fs (∂Ω, σ) ⊗ Cn .
(4.1.52)

Finally, from (1.6.8), (4.1.49)-(4.1.50), and [69, Lemma 7.1.10] it follows that
p,q
(C # )2 = 14 I on B−s (∂Ω, σ) ⊗ Cn when
  (4.1.53)
s ∈ (0, 1), p ∈ n−1 n−s , ∞ , q ∈ (0, ∞],

and
p,q
(C # )2 = 14 I on F−s (∂Ω, σ) ⊗ Cn when
  (4.1.54)
s ∈ (0, 1), p ∈ n−1 n−s , ∞), q ∈ n−1
n−s , ∞],

first when q < ∞, then the end-point q = ∞ is included as a result of this and loose
embeddings (cf. items (iii)-(iv) in [69, Proposition 7.7.1]).
Here is another category of singular integral operators which behave naturally on
boundary Besov and Triebel-Lizorkin spaces with a negative amount of smoothness.
Theorem 4.1.7 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain with compact
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν j )1≤ j ≤n the geometric
1 (Rn, L n ) be a complex-
measure theoretic outward unit normal to Ω. Next, let b ∈ Lloc

valued function with the property that b Rn \{0} ∈ 𝒞 (R \ {0}) where N = N(n) ∈ N
N n

is a sufficiently large number, and such that ∇b is odd and positive homogeneous of
degree 1 − n in Rn \ {0}. Finally, for each given pair of indices j, k ∈ {1, . . . , n}
introduce the integral operator acting on each function f ∈ L 1 (∂Ω, σ) according to
∫
Tjk f (x) := lim+
#
ν j (x)(∂k b)(x − y) − νk (x)(∂j b)(x − y) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(4.1.55)

at σ-a.e. x ∈ ∂Ω. Then the operator Tjk

# extends uniquely to linear and bounded

mappings
442 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p,q p,q
−s (∂Ω, σ) −→ B−s (∂Ω, σ),
# : B
Tjk
  (4.1.56)
s ∈ (0, 1), p ∈ n−1 n−s , ∞ , q ∈ (0, ∞],

and
p,q p,q
−s (∂Ω, σ) −→ F−s (∂Ω, σ),
# : F
Tjk
  (4.1.57)
s ∈ (0, 1), p ∈ n−1n−s , ∞), q ∈ n−1
n−s , ∞].

Also, various choices of the exponents yield operators which are compatible with
one another. Finally, if p, q, p , q ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q and
s ∈ (0, 1), then
 #   
B−s (∂Ω,σ) T jk f , g B sp , q (∂Ω,σ) = − B−s (∂Ω,σ) f , T jk g B sp , q (∂Ω,σ)
p, q p, q

(4.1.58)
p,q p ,q
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ),

and
 #   
p, q
F−s (∂Ω,σ) Tjk f , g F p , q (∂Ω,σ)
= − F−s
p, q
(∂Ω,σ) f , T jk g F p , q (∂Ω,σ)
s s
(4.1.59)
p,q p ,q
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ).

Proof The claims about (4.1.56)-(4.1.57) may be justified with the help of Theo-
rem 4.1.3, item (ii) in Proposition 1.2.1, and Theorem 2.1.4 much in the same way
Theorem 4.1.5 has been proved making use of Theorem 4.1.1, item (iii) in Theo-
rem 1.5.1, and Theorem 2.1.1. Finally, (4.1.58)-(4.1.59) are seen from what we have
proved so far, Theorem 4.1.3, item (ii) in Proposition 1.2.1, and [69, (7.1.62)]. 

4.2 Boundary-to-Domain Layer Potentials from Besov Spaces

into Weighted Sobolev Spaces

In this section we take up the task of studying mapping properties of boundary-to-

domain layer potential operators acting from Besov spaces into weighted Sobolev
spaces.
For starters, we shall prove mapping properties for certain boundary-to-domain
integral operators acting from Besov spaces with a positive amount of smoothness
into Lebesgue spaces suitably weighted in terms of the distance to the boundary.
Below, we shall establish a general result of this nature which identifies the key
qualities of said integral operator which produce a desirable result of this flavor.
Subsequently, we shall specialize this to a number of concrete integral operators
of practical interest. For now, the reader is reminded that, having fixed background
parameter θ ∈ (0, 1), the solid maximal function u ,θ is associated with any given
function u as in [68, §6.6]. Also, recall that for each number a ∈ R we use the
abbreviation (a)+ := max{a, 0}.
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 443

Theorem 4.2.1 Let Ω be an open set in Rn , n ≥ 2, whose boundary is an Ahlfors

regular set, and abbreviate σ := H n−1 ∂Ω. Also, consider a measurable function
q : Ω × ∂Ω → C with the property that there exist ε ∈ [0, 1), k ∈ R, and some finite
constant co > 0 such that
δ∂Ω (x) −ε 1
|q(x, y)| ≤ co δ∂Ω (x)−k
|x − y| |x − y| n (4.2.1)
for every point x ∈ Ω and σ-a.e. point y ∈ ∂Ω.
 
Associated with it, consider the integral operator acting on each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−ε
according to ∫
Q f (x) := q(x, y) f (y) dσ(y) for x ∈ Ω, (4.2.2)
∂Ω
and assume that
Q annihilates constants. (4.2.3)
Then, if 1 
n−1
n−ε < p ≤ ∞ and (n − 1) p −1 + < s < 1 − ε, (4.2.4)
it follows that for each θ ∈ (0, 1) there exists some C = C(Ω, ε, k, co, p, s, θ) > 0 such
that
 k+1− 1 −s   
 Q f   p, p
δ∂Ω
p
,θ
≤ C f Bsp, p (∂Ω,σ), ∀ f ∈ Bs (∂Ω, σ), (4.2.5)
L (Ω, L )
p n

 
with the understanding that when p > 1 the solid maximal function Q f  is
  ,θ
replaced by Q f .

A few remarks are in order. First, from (4.2.1) it follows that

co δ∂Ω (x)−k−ε
|q(x, y)| ≤ for all x ∈ Ω and y ∈ ∂Ω, (4.2.6)
|x − y| n−ε
which, in concert with [68, Lemma 7.2.1], implies that

the operator Q acts on functions from the weighted Lebesgue space

  (4.2.7)
L 1 ∂Ω, 1+σ(x)
|x | n−ε meaningfully (via an absolutely convergent integral).

In particular, it follows that the action of Q on constants functions is meaningful,

hence it makes sense to impose the demand in (4.2.3).
Second, from [69, Proposition 7.9.4], [69, (7.1.59)], and [68, (7.4.119)], we see
that  
Bs (∂Ω, σ) → L 1 ∂Ω, 1+σ(x)
p, p
|x | n−ε
(4.2.8)
if ε ∈ [0, 1) and p, s are as in (4.2.4).
Together with (4.2.7) this shows that Q acts in a meaningful fashion on the Besov
spaces considered in Theorem 4.2.1.
444 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 
Third, [68, (6.6.6) in Lemma 6.6.1] ensures that Q f  ,θ is a Lebesgue measurable

function in Ω, which bounds |Q f  at L n -a.e. point in Ω. As a consequence of this
and (4.2.5) we therefore have (with p, s as in (4.2.4))
 k+1− 1 −s 
  p, p
δ∂Ω
p
· Q f ≤ C f Bsp, p (∂Ω,σ), ∀ f ∈ Bs (∂Ω, σ). (4.2.9)
L p (Ω, L n )

Fourth, an inspection of the proof reveals that the same type of result is true in
the case when the kernel q is matrix-valued, and the functions f are vector-valued
(with the understanding that the definition of Q in (4.2.2) now takes into account the
natural action of the matrix q(x, y) on the vector f (y)).
We now turn to the task of presenting the proof of Theorem 4.2.1.
Proof of Theorem 4.2.1 Throughout, fix θ ∈ (0, 1). We shall show that whenever
p, s are as in (4.2.4) there exists a constant C ∈ (0, ∞) such that for each function
p, p
f ∈ Bs (∂Ω, σ) we have
 k+1− 1 −s   
 Q f  
δ∂Ω
p
,θ 
≤ C f Bsp, p (∂Ω,σ), (4.2.10)
L p (Ω, L n )

with the convention that when p > 1 the solid maximal function |Q f  ,θ is simply

replaced by |Q f . Consider first the case p = 1, in which scenario we shall prove
that there exists some finite C > 0 such that if 0 < s < 1 − ε then
 
 k−s   
δ∂Ω Q f ,θ  1 ≤ C f B1,1 (∂Ω,σ), ∀ f ∈ Bs1,1 (∂Ω, σ). (4.2.11)
L (Ω, L )
n s

To justify this, fix an arbitrary function f ∈ Bs1,1 (∂Ω, σ). Thanks to (4.2.3) we may
write
∫
(Q f )(x) = q(x, y)( f (y) − f (z)) dσ(y) for all x ∈ Ω, z ∈ ∂Ω. (4.2.12)
∂Ω

Combining (4.2.12) with (4.2.1) further yields

∫
 
(Q f )(x) ≤ co δ∂Ω (x)−k−ε |x − y| −n+ε | f (y) − f (z)| dσ(y), (4.2.13)
∂Ω

for all x ∈ Ω and z ∈ ∂Ω. Next, fix c > 1 and for each x ∈ Ω define the set

Ex,c := z ∈ ∂Ω : |x − z| < cδ∂Ω (x) . (4.2.14)

Also, consider x ∗ ∈ ∂Ω with the property that |x − x ∗ | = δ∂Ω (x). Then for each
z ∈ Ex,c we have |z − x ∗ | ≤ |x − z| + |x − x ∗ | < (c + 1)δ∂Ω (x). Moreover, if
0 < λ < c − 1, then for every z ∈ ∂Ω satisfying |z − x ∗ | < λδ∂Ω (x) we may estimate
|x − z| ≤ |x − x ∗ | + |x ∗ − z| < cδ∂Ω (x). As such,
   
B x ∗, λδ∂Ω (x) ∩ ∂Ω ⊆ Ex,c ⊆ B x ∗, (c + 1)δ∂Ω (x) ∩ ∂Ω. (4.2.15)
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 445

It is also apparent from (4.2.14) that Ex,c is a relatively open subset of ∂Ω, hence
σ-measurable. On account of this, (4.2.15), and the Ahlfors regularity of ∂Ω, we
conclude that
σ(Ex,c ) ≈ δ∂Ω (x)n−1 . (4.2.16)
Having established this, we proceed to take the integral average with respect to σ
over the set Ex,c of both sides of (4.2.13). In doing so, and relying on (4.2.16), we
arrive at the conclusion that, for each x ∈ Ω,
∫ ∫
  | f (y) − f (z)|
(Q f )(x) ≤ Cδ∂Ω (x)−n+1−k−ε dσ(y) dσ(z). (4.2.17)
E x, c ∂Ω |x − y|
n−ε

Pressing on, pick an arbitrary point x 0 ∈ Ω. Elementary considerations show that

⎧
⎪ (1 − θ)δ∂Ω (x 0 ) ≤ δ∂Ω (x) ≤ (1 + θ)δ∂Ω (x 0 ),
  ⎪
⎨
⎪
x ∈ B x 0, θ · δ∂Ω (x 0 ) =⇒ and (4.2.18)
⎪
⎪
⎪ E ⊆ E 0 where c := c(1 + θ) + θ.
⎩ x,c x ,c0 0

Furthermore,
 
x ∈ B x 0, θ · δ∂Ω (x 0 ) =⇒ |x − y| ≈ |x 0 − y| uniformly for y ∈ ∂Ω. (4.2.19)

Consequently,
 from (4.2.18)-(4.2.19)
 and (4.2.17) we deduce that for each point
x ∈ B x 0, θ · δ∂Ω (x 0 ) we have
∫ ∫
  | f (y) − f (z)|
(Q f )(x) ≤ Cδ∂Ω (x 0 )−n+1−k−ε dσ(y) dσ(z). (4.2.20)
∂Ω |x − y|
0 n−ε
Ex 0, c
0

 
Taking the supremum over x ∈ B x 0, θ · δ∂Ω (x 0 ) then gives (keeping in mind
(A.0.194))
∫ ∫
  | f (y) − f (z)|
(Q f ) (x 0 ) ≤ Cδ∂Ω (x 0 )−n+1−k−ε dσ(y) dσ(z),
,θ
∂Ω |x − y|
0 n−ε
Ex 0, c
0
(4.2.21)
for every x 0 ∈ Ω. Hence, if we re-denote x 0 by x, then multiply both sides of (4.2.21)
by δ∂Ω (x)k−s and, finally, integrate the resulting expressions over Ω with respect to
x, we obtain (after an application of Fubini’s Theorem)
446 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

∫
 
δ∂Ω (x)k−s Q f  ,θ
(x) dx (4.2.22)
Ω
∫ ∫ ∫
| f (y) − f (z)|
≤C δ∂Ω (x) 1−n−s−ε
dσ(y) dσ(z) dx
Ω E x, c0 ∂Ω |x − y| n−ε
∫ ∫ ∫
δ∂Ω (x)1−n−s−ε
=C | f (y) − f (z)| dx dσ(y) dσ(z),
∂Ω ∂Ω Γκ (z) |x − y| n−ε

where κ := c0 − 1 > 0 and

Γκ (z) := x ∈ Ω : |x − z| < (1 + κ)δ∂Ω (x) . (4.2.23)

At this point we invoke [68, Proposition 8.7.8] with Σ := ∂Ω, β := −1 + n + s + ε,

and M := n − ε (note that, since p = 1 and 0 < s < 1 − ε, these choices entail β < n
and M > n − β, as required by the hypotheses of said proposition). As a consequence
of this result, the integral over Γκ (z) from (4.2.22) is bounded by C|z − y| −(n−1+s) .
The latter used back in (4.2.22) then yields (4.2.11), upon recalling [69, (7.9.11)].
Now we turn our attention to the case when p = ∞ and 0 < s < 1 − ε. The goal
is to show that there exists a constant C ∈ (0, ∞) with the property that for each
f ∈ Bs∞,∞ (∂Ω, σ) we have
 
 k+1−s  
δ∂Ω Q f  ∞ ≤ C f B∞,∞ (∂Ω,σ) . (4.2.24)
n L (Ω, L ) s

To this end, fix f ∈ Bs∞,∞ (∂Ω, σ). Also, let x ∈ Ω be arbitrary and, again, denote
by x ∗ ∈ ∂Ω a point with the property that |x − x ∗ | = δ∂Ω (x). Then, by once more
appealing to (4.2.3), we may estimate
∫

k+1−s 

δ∂Ω (x) 
Q f (x) ≤ δ∂Ω (x) k+1−s
|q(x, y)|| f (y) − f (x ∗ )| dσ(y). (4.2.25)
∂Ω

Since f ∈ Bs∞,∞ (∂Ω, σ) = 𝒞s (∂Ω) (recall [69, (7.1.59)]), we have

| f (y) − f (x ∗ )| ≤  f B∞,∞
s
∗ s
(∂Ω,σ) |y − x | , ∀y ∈ ∂Ω. (4.2.26)

To proceed, set r := |x − x ∗ | = δ∂Ω (x) and bound the right-hand side of (4.2.25) by
I1 + I2 , where I1, I2 correspond to taking y ∈ B(x ∗, 2r) ∩ ∂Ω and y ∈ ∂Ω \ B(x ∗, 2r),
respectively, in the integral in (4.2.25). Making use of (4.2.26), (4.2.1), and the fact
that if y ∈ B(x ∗, 2r) ∩ ∂Ω then |x − y| ≥ r hence also |y − x ∗ | ≤ 2r ≤ 2|x − y|, we
obtain
∫
|y − x ∗ | s
I1 ≤ C f B∞,∞
s (∂Ω,σ) r k+1−s
k+ε |x − y| n−ε
dσ(y)
B(x ∗,2r)∩∂Ω r
−n+1  
≤ C f B∞,∞
s (∂Ω,σ) r σ B(x ∗, 2r) ∩ ∂Ω ≤ C f B∞,∞ s (∂Ω,σ), (4.2.27)
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 447

where in the last inequality in (4.2.27) we have used the Ahlfors regularity of ∂Ω.
Turning our attention to I2 , observe that if the point y ∈ ∂Ω \ B(x ∗, 2r) then we have
|y − x ∗ | ≤ |x − y| + r ≤ |x − y| + 12 |y − x ∗ | which further implies |y − x ∗ | ≤ 2|x − y|.
Thus, starting again with (4.2.26) and (4.2.1), we may now write
∫
|y − x ∗ | s
I2 ≤ C f B∞,∞ (∂Ω,σ) r k+1−s
dσ(y)
s
∂Ω\B(x ∗,2r) r
k+ε |x − y| n−ε
∫
dσ(y)
≤ C f B∞,∞ (∂Ω,σ) r 1−s−ε
∗ n−s−ε
∂Ω\B(x ∗,2r) |y − x |
s

≤ C f B∞,∞
s (∂Ω,σ), (4.2.28)

thanks to the first estimate in [68, (7.2.5)] (whose applicability is ensured by the
fact that we presently have n − s − ε > n − 1). Now (4.2.24) follows by combining
(4.2.27) and (4.2.28). This completes the treatment of the case when p = ∞ and
0 < s < 1 − ε.
To treat the case when 1 < p < ∞, the idea is to use what we have proved so
far and interpolation. More precisely, assume 0 < s0 < s1 < 1 − ε and consider the
family of linear operators
k+z−[(1−z)s0 +zs1 ]
Lz := δ∂Ω Q for z ∈ C with 0 ≤ Re z ≤ 1. (4.2.29)

Observe that Lz depends analytically on z and

k−s0
Re z = 0 =⇒ |Lz f | = δ∂Ω |Q f |, (4.2.30)
k+1−s1
Re z = 1 =⇒ |Lz f | = δ∂Ω |Q f |. (4.2.31)

Based on this observation, our results for p = 1 and p = ∞ (the former in concert
with [68, (6.6.6) in Lemma 6.6.1]) lead to the conclusion that the operators

Lz : Bs1,1
0
(∂Ω, σ) −→ L 1 (Ω, L n ) for Re z = 0, (4.2.32)

Lz : Bs∞,∞
1
(∂Ω, σ) −→ L ∞ (Ω, L n ) for Re z = 1, (4.2.33)

are well-defined, linear, and bounded. Granted these, the complex interpolation
results from [69, Theorem 7.5.2] may be used in concert with Stein’s interpolation
theorem for analytic families of operators (see, e.g., [11, Theorem 3.4, pp. 151-152]
for a versatile variant). This allows us to conclude that the operator
k+1− p1 −s p, p
δ∂Ω Q : Bs (∂Ω, σ) −→ L p (Ω, L n ) (4.2.34)

is well defined, linear, and bounded for each s ∈ (0, 1 − ε) and p ∈ [1, ∞].
Alternatively, we may reach the same conclusions regarding the operator (4.2.34)
using the real method of interpolation. Specifically, given any s ∈ (0, 1 − ε), from
(4.2.11) and [68, Lemma 6.6.1] we know that the operator
448 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 −1 n 
δ∂Ω
k+1−s
Q : Bs1,1 (∂Ω, σ) −→ L 1 Ω, δ∂Ω L (4.2.35)

is well defined, linear, and bounded, whereas (4.2.24) implies that we also have a
well-defined, linear, and bounded operator
 −1 n 
δ∂Ω
k+1−s
Q : Bs∞,∞ (∂Ω, σ) −→ L ∞ (Ω, L n ) = L ∞ Ω, δ∂Ω L (4.2.36)

(with the above equality a consequence of the fact that the measures L n and δ∂Ω −1 L n

are mutually absolutely continuous in Ω). Then from (4.2.35)-(4.2.36), the real
interpolation result from [69, (7.4.18)] (used with Σ := ∂Ω, p0 := 1, p1 := ∞),
and well-known real interpolation results for generic Lebesgue spaces (cf., e.g., [2,
Theorem 5.2.1, p. 109]; see also [26] for useful results regarding the real interpolation
of generic weighted Lebesgue spaces) we conclude that the operator
p, p  −1 n 
δ∂Ω
k+1−s
Q : Bs (∂Ω, σ) −→ L p Ω, δ∂Ω L (4.2.37)

is well defined, linear, and bounded whenever 1 ≤ p ≤ ∞ and 0 < s < 1 − ε.

A moment’s refection shows that this is equivalent to having the operator (4.2.34)
well-defined, linear, and bounded for each s ∈ (0, 1 − ε) and p ∈ [1, ∞]. In turn, this
amounts to having the estimate claimed (4.2.10) valid in the range 1 ≤ p ≤ ∞ and
0 < s < 1 − ε (with the agreement that when p > 1 the subscripts , θ are omitted).
This finishes the proof of estimate (4.2.10) in the range 1 ≤ p ≤ ∞ and 0 < s < 1 − ε
(again, with the understanding that when p > 1 the subscripts , θ areomitted).
n−1
We are left analyzing the case when n−ε < p < 1 and (n − 1) p1 − 1 < s < 1 − ε.
In this scenario, we shall first prove that for each fixed η ∈ (s, 1) there exists some
finite constant C > 0 such that
 k+1− 1 −s   
 Qa 
δ∂Ω
p
,θ
≤ C for each smooth η-block a of type (p, s)
p L (Ω, L )
n

(4.2.38)
(cf. [69, Definition 7.2.1]). To this end, fix k and η as above, and assume that
n−1
a L ∞ (∂Ω,σ) ≤ r s− p for some r > 0, (4.2.39)
supp a ⊆ B(xa, r) ∩ ∂Ω for some xa ∈ ∂Ω, (4.2.40)

p −η
n−1
|a(x) − a(y)| ≤ r s− |x − y| η for all x, y ∈ ∂Ω. (4.2.41)

For τ ∈ R to be specified later, consider the re-normalization

a(x) := r τ a(x) for each x ∈ ∂Ω.

 (4.2.42)
 
For now, assume that τ + s − (n − 1) p1 − 1 ∈ (0, 1), i.e.,
1   
(n − 1) p − 1 − s < τ < 1 + (n − 1) p1 − 1 − s. (4.2.43)
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 449

Then, using the properties of a it is easy tocheckthat 

a satisfies conditions (4.2.39)-
(4.2.41) with s replaced by τ + s − (n − 1) p1 − 1 and with p replaced by 1. Thus,  a
 1 
is a smooth η-block of type 1, τ + s − (n − 1) p − 1 if we also assume
1 
η > τ + s − (n − 1) p −1 . (4.2.44)

Granted this, [69, Theorem 7.2.8] ensures that

a ∈ B1,1
 (∂Ω, σ) and 
a B1,1 (∂Ω,σ) ≤ C, (4.2.45)
τ+s−(n−1)( p1 −1) 1 −1)
τ+s−(n−1)( p

for some finite constant C > 0 independent of a. Under the additional assumption
that  
τ < 1 − ε + (n − 1) p1 − 1 − s, (4.2.46)
this allows us to use the bounds already proved for the operator Q corresponding to
p = 1 in order to obtain that
 k−τ−s+(n−1)( 1 −1)   
 Q 
a
p
δ∂Ω ,θ 
≤ C
a B1,1 (∂Ω,σ) ≤ C. (4.2.47)
L 1 (Ω, L n ) 1 −1)
τ+s−(n−1)( p

Recall that θ ∈ (0, 1) has been fixed at the beginning of the proof. Applying Hölder’s
inequality and using (4.2.47) we may then estimate
∫
  p
δ∂Ω (x)k+1− p −s Qa ,θ (x) dx
1

B(x a ,2(1−θ)−1 r)∩Ω

∫
 p
= r −τ p δ∂Ω (x)k p+p−1−sp Q
a  ,θ (x) dx
B(x a ,2(1−θ)−1 r)∩Ω
∫ 1   p
≤r −τ p
δ∂Ω (x) k−τ−s+(n−1) p −1 Q 
a (x) dx ×
B(x a ,2(1−θ)−1 r)∩Ω
∫  1−p
τp
× δ∂Ω (x) 1−p −n dx
B(x a ,2(1−θ)−1 r)∩Ω
∫  1−p
τp
≤ Cr −τ p
δ∂Ω (x) 1−p −n dx . (4.2.48)
B(x a ,2(1−θ)−1 r)∩Ω

If we further impose the condition

1 
(n − 1) p − 1 < τ, (4.2.49)
τp
we may then apply [68, (8.7.3)] with α := n − 1−p and N := 0 (also, with Σ := ∂Ω,
−1
y := xa , and r replaced by 2(1 − θ) r) to conclude that
450 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
∫
τp τp
δ∂Ω (x) 1−p −n dx ≤ Cr 1−p . (4.2.50)
B(x a ,2(1−θ)−1 r)∩Ω

The key observation now is that conditions (4.2.43), (4.2.46), (4.2.44), and (4.2.49)
may be simultaneously satisfied for a suitable choice of τ, since 0 < s < 1 − ε and
s < η < 1. Having chosen such a number τ, a combination of (4.2.48), (4.2.50), and
(4.2.47) yields
∫
  p
δ∂Ω (x)k+1− p −s Qa ,θ (x) dx ≤ C,
1
(4.2.51)
B(x a ,2(1−θ)−1 r)∩Ω

for some finite constant C > 0 which is independent of the atom in question.
Next, we turn our attention to the contribution away from the support of the atom.
Using (4.2.39), (4.2.40), and (4.2.1) we see that
∫
  |a(y)|
(Qa)(x) ≤ C dσ(y)
δ
B(x a ,r)∩∂Ω ∂Ω (x) k+ε |x − y| n−ε

C r s+(n−1)(1− p )
1

≤ if x ∈ Ω \ B(xa, 2r). (4.2.52)

δ∂Ω (x)k+ε |x − xa | n−ε
To proceed, observe that
   
x 0 ∈ Ω\ B xa, 2(1− θ)−1 r and x ∈ B x 0, θ · δ∂Ω (x 0 ) =⇒ |x − xa | ≥ 2r. (4.2.53)

Indeed, the second membership forces δ∂Ω (x) > (1 − θ)δ∂Ω (x 0 ). Hence, in the case
in which we have δ∂Ω (x 0 ) ≥ 2(1 − θ)−1 r we may write

|x − xa | ≥ δ∂Ω (x) > (1 − θ)δ∂Ω (x 0 ) ≥ 2r, (4.2.54)

as wanted. On the other hand, if δ∂Ω (x 0 ) < 2(1 − θ)−1 r then

|x − xa | ≥ |x 0 − xa | − |x − x 0 | ≥ 2(1 − θ)−1 r − θ · δ∂Ω (x 0 )

> 2
1−θ r − 2θ
1−θ r = 2r, (4.2.55)

finishing the justification of (4.2.53). In turn, from (4.2.53), (4.2.52), (A.0.194), and
(4.2.18)-(4.2.19) we obtain

C r s+(n−1)(1− p )
1
   
Qa (x
0
) ≤ if x 0 ∈ Ω \ B xa, 2(1 − θ)−1 r . (4.2.56)
,θ δ∂Ω (x ) |x − xa |
0 k+ε 0 n−ε

At this point we find it convenient to revert to denoting x 0 by x, and use (4.2.56) and
[68, (8.7.5) in Proposition 8.7.1] with α := 1 + p(s − 1 + ε) and N := p(n − ε) (note
that α < 1 since s < 1 − ε, while α > n − N since s > (n − 1)(1/p − 1)) in order to
conclude that
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 451
∫
 p
δ∂Ω (x)(k+1− p −s)p Qa ,θ (x) dx ≤ C.
1
(4.2.57)
Ω\B(x a ,2(1−θ)−1 r)

Now estimate (4.2.38) follows from (4.2.51) and (4.2.57).

 
Going further, fix n−εn−1
< p < 1 along with (n−1) p1 −1 < s < 1− ε and consider
p, p
some arbitrary f ∈ Bs (∂Ω, σ). Then [69, Theorem 7.2.7] ensures the existence of
a sequence {a j } j ∈N of smooth η-blocks of type (p, s) on ∂Ω along with a numerical
sequence {λ j } j ∈N belonging to  p (a space with which bp, p (∂Ω) identifies; cf. [69,
(7.2.24) in Definition 7.2.5]) with the property that

∞ 1/p
|λ j | p ≤ C f Bsp, p (∂Ω,σ) < +∞, (4.2.58)
j=1

for some constant C ∈ (0, ∞) independent of f , and such that if


m
fm := λ j a j for each m ∈ N (4.2.59)
j=1

then
p, p
fm −→ f in Bs (∂Ω, σ) as m → ∞. (4.2.60)
Let us also observe that if we set
1 
s∗ := s − (n − 1) p − 1 ∈ (0, s) ⊆ (0, 1 − ε), (4.2.61)

s∗
then 0 < s∗ < s < 1 and p1 − n−1
s
= 1 − n−1 . As such, [69, Theorem 7.7.4] (presently
employed with Σ := ∂Ω) gives
p, p
Bs (∂Ω, σ) → Bs1,1
∗
(∂Ω, σ) continuously. (4.2.62)

In particular, from (4.2.60) and (4.2.62) we conclude that

fm −→ f in Bs1,1
∗
(∂Ω, σ) as m → ∞. (4.2.63)

To proceed, abbreviate
k+1− p1 −s
F := δ∂Ω (Q f ) ,θ and (4.2.64)

k+1− p1 −s
Fm := δ∂Ω (Q fm ) ,θ for each m ∈ N. (4.2.65)

In relation to these functions we make several remarks. First, based on (4.2.65),

(4.2.59), [68, (6.6.4), (6.6.9)], (4.2.38), and (4.2.58), for each m ∈ N we may
estimate (bearing in mind that we are presently assuming 0 < p < 1)
452 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces


m  k+1− 1 −s   p
 
|λ j | p δ∂Ω p Qa j 
p
Fm  L p (Ω, L n ) ≤ ,θ  p
L (Ω, L n )
j=1


m
p
≤C |λ j | p ≤ C f B p, p (∂Ω,σ), (4.2.66)
s
j=1

for some constant C ∈ (0, ∞) which is independent of f and m. Likewise, given any
m , m ∈ N with m ≤ m , we may use (4.2.65), [68, (6.6.4), (6.6.10)], (4.2.59), and
(4.2.38) to estimate (again, mindful of the fact that 0 < p < 1)
   k+1− 1 −s       p
Fm − Fm  p p   
L (Ω, L )
n = δ∂Ω
p
 Q fm ,θ − Q fm ,θ  p
L (Ω, L n )
 k+1− 1 −s   p
 
≤ δ∂Ω p Q( fm − fm ) ,θ  p
L (Ω, L n )


m  k+1− 1 −s   p
 
≤ |λ j | p δ∂Ω p Qa j  ,θ  p
L (Ω, L n )
j=m


m
≤C |λ j | p, (4.2.67)
j=m

for some constant C ∈ (0, ∞) which is independent of f , m, m . Finally, in a similar

fashion, for each m ∈ N we may also employ (4.2.11) in order to write
 −1+ 1    
 −1+ p1   k−s      
δ∂Ω · F − δ∂Ω · Fm  = δ∂Ω  Qf − Q fm ,θ 
p

L 1 (Ω, L n ) ,θ L 1 (Ω, L n )
 
 k−s   
≤ δ∂Ω Q( f − fm ) ,θ  L 1 (Ω, L n )

≤ C f − fm B1,1 (∂Ω,σ), (4.2.68)

s

for some constant C ∈ (0, ∞) which is independent of f , m.

Let us summarize our progress. From (4.2.66), (4.2.67), (4.2.68), (4.2.58),
(4.2.60), and (4.2.63) we conclude that

Fm ∈ L p (Ω, L n ) for all m ∈ N and sup Fm  L p (Ω, L n ) ≤ C f Bsp, p (∂Ω,σ),

m∈N
(4.2.69)

{Fm }m∈N is a Cauchy sequence in L p (Ω, L n ), and (4.2.70)

−1+ p1 −1+ p1
δ∂Ω · Fm −→ δ∂Ω · F in L 1 (Ω, L n ) as m → ∞. (4.2.71)
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 453

In turn, (4.2.70) implies that there exists some F  ∈ L p (Ω, L n ) such that Fm → F  in
L p (Ω, L n ) as m → ∞. Consequently, there exists a subsequence {Fmi }i ∈N with the
property that
Fmi → F  at L n -a.e. point in Ω as i → ∞. (4.2.72)
Also, from (4.2.69) we see that
 L p (Ω, L n ) ≤ C f B p, p (∂Ω,σ)
F (4.2.73)
s

for some constant C ∈ (0, ∞) which is independent of the function f . Finally, (4.2.71)
−1+ p1 −1+ p1
guarantees that δ∂Ω · Fmi → δ∂Ω · F in L 1 (Ω, L n ) as i → ∞, hence there exists
−1+ 1 −1+ 1
a sub-subsequence {Fmi j } j ∈N with the property that δ∂Ω p · Fmi j → δ∂Ω p · F at
L n -a.e. point in Ω as j → ∞. Since the latter further implies that Fmi j → F at
L n -a.e. point in Ω as j → ∞, we then conclude from this and (4.2.72) that F = F 
at L -a.e. point in Ω. In concert with (4.2.73)
n
 and (4.2.64), this ultimately yields
(4.2.10) when n−εn−1
< p < 1 and (n − 1) p1 − 1 < s < 1 − ε. At this point, the estimate
claimed in (4.2.10) has been established for the full range of indices p, s indicated in
(4.2.4). The proof of Theorem 4.2.1 is therefore complete. 
Theorem 4.2.1 has many remarkable consequences and in a series of corollaries
we single out several applications to double layer like integral operators.
Corollary 4.2.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set whose boundary is
Ahlfors regular. Denote by ν = (ν1, . . . , νn ) the geometric measure
 αβtheoretic
 outward
unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, let L = ar s ∂r ∂s 1≤α,β ≤M be
an M×M second-order, homogeneous, constant (complex) coefficient, weakly elliptic
system in Rn , and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental
solution associated with L as in [70, Theorem 1.4.2]. Finally, fix θ ∈ (0, 1) and
assume that 1 
n < p ≤ ∞ and (n − 1) p − 1 + < s < 1.
n−1
(4.2.74)
Then the following conclusions are valid.
(a) If D is the double layer potential operator associated with L and Ω as in (1.3.18)
then given any multi-index γ ∈ N0n with |γ| > 0 there exists some finite constant
 p, p M
C = C(Ω, L, p, s, θ, γ) > 0 such that for each f ∈ Bs (∂Ω, σ) one has
 |γ |− 1 −s   
 ∂γ (D f ) 
δ∂Ω
p
,θ 
≤ C f [Bsp, p (∂Ω,σ)] M
L p (Ω, L n ) (4.2.75)
provided either p < ∞, or ∂Ω is a bounded set.

(b) Make the additional assumption that ∂Ω is bounded. For each pair of in-
dices j, k ∈ {1, . . . , n} consider the integral operator U jk acting on any
 M
f ∈ L 1 (∂∗ Ω, σ) according to
∫
U jk f (x) := ν j (y)(∂k E)(x − y) − νk (y)(∂j E)(x − y) f (y) dσ(y) (4.2.76)
∂∗ Ω
454 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

for all x ∈ Ω. Then for each γ ∈ N0n with |γ| > 0 there exists some finite
constant C = C(Ω, L, p, s, θ, γ) > 0 with the property that for each function
 p, p M
f ∈ Bs (∂Ω, σ) one has
n 
 
 |γ |− p1 −s  γ  
δ∂Ω ∂ (U jk f ) ,θ  L p (Ω, L n )
≤ C f [Bsp, p (∂Ω,σ)] M . (4.2.77)
j,k=1

(c) Continue to work under the additional assumption that ∂Ω is bounded. Then
the following versions of (4.2.75) and (4.2.77) corresponding to the case when
|γ| = 0 also hold: for each cutoff function ψ ∈ 𝒞∞ c (R ) there exists a constant
n

C = C(Ω, L, p, s, θ, ψ) ∈ (0, ∞) with the property that for each given function
 p, p M
f ∈ Bs (∂Ω, σ) one has
 1− 1 −s  

δ∂Ω ψ D f  ≤ C f [Bsp, p (∂Ω,σ)] M and
p

L p (Ω, L n )
 1− 1 −s    (4.2.78)
 
δ∂Ω D f  ,θ 
≤ C f  if Ω bounded,
p
p, p
[B s (∂Ω,σ)] M
L p (Ω, L n )

as well as
n 
  1− p1 −s  
δ∂Ω ψ U jk f  ≤ C f [Bsp, p (∂Ω,σ)] M and
L p (Ω, L n )
j,k=1
n 
 
 1− p1 −s   
δ∂Ω U jk f  ,θ 
≤ C f [Bsp, p (∂Ω,σ)] M if Ω bounded.
L p (Ω, L n )
j,k=1
(4.2.79)

Proof From [69, Theorem 7.7.4] and [69, (7.9.10)-(7.9.11)] we conclude that
1 
n < p < ∞ and (n − 1) p − 1 + < s < 1
if n−1
there exist p∗ ∈ (1, ∞) and s∗ ∈ (0, 1) such that (4.2.80)
p, p p ,p
Bs (∂Ω, σ) → Bs∗∗ ∗ (∂Ω, σ) → L p∗ (∂Ω, σ).

Also, corresponding to p = ∞, the identification in [69, (7.1.59)] implies that

if ∂Ω is a compact set and 0 < s < 1 then

(4.2.81)
Bs∞,∞ (∂Ω, σ) = 𝒞s (∂Ω) → L p∗ (∂Ω, σ) for each p∗ ∈ (0, ∞].

From (4.2.80)-(4.2.81) and (1.3.24) we see that whenever p, s are as in (4.2.74) the
 p, p M
double layer potential operator D is well defined on the Besov space Bs (∂Ω, σ)
and, in fact,
 M
D f ∈ 𝒞∞ (Ω) and L(D f ) = 0 in Ω,
 p, p M (4.2.82)
for each f ∈ Bs (∂Ω, σ) .
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 455

To proceed, pick an arbitrary multi-index γ ∈ N0n with |γ| > 0 and define the
operator Q as in (4.2.2) for the matrix-valued integral kernel
βα
q(x, y) := − νs (y)ar s (∂γ ∂r Eμβ )(x − y)1∂∗ Ω (y)
1≤μ,α ≤M (4.2.83)
for every x ∈ Ω and σ-a.e. y ∈ ∂Ω.

From (4.2.83) and [70, Theorem 1.4.2] we then see that there exists C ∈ (0, ∞) with
the property that

C δ∂Ω (x)1− |γ |
|q(x, y)| ≤ C|∂γ (∇E)(x − y)| ≤ ≤ C
|x − y| n−1+ |γ | |x − y| n (4.2.84)
for every x ∈ Ω and σ-a.e. y ∈ ∂Ω.

As such, estimate (4.2.1) is satisfied with ε := 0 and k := |γ| − 1. In addition, if Dmod

is as in (1.8.6), then (1.8.11), (1.8.10), and the fact that |γ| > 0 permit us to write
 
Q e = ∂γ Dmod e = 0 in Ω for each  ∈ {1, . . . , n}. (4.2.85)

Hence, all hypotheses of Theorem 4.2.1 are satisfied by the current choice of Q.
Upon noting that, as seen from (4.2.80)-(4.2.81) and (1.3.18),
 p, p M
Q f = ∂γ (D f ) for each f ∈ Bs (∂Ω, σ)
(4.2.86)
provided either p < ∞, or ∂Ω is a bounded set,

we conclude from (4.2.5), (4.2.82), [68, (6.5.40) in Theorem 6.5.7], and [68, (6.6.91)]
that (4.2.75) holds. This takes care of item (a).
As regards the operators U jk from (4.2.76), observe that if ∂Ω is bounded then
 M
given any j, k ∈ {1, . . . , n} for each function f ∈ L11 (∂∗ Ω, σ) we may write
∫
U jk f (x) = E(x − y)(∂τ j k f )(y) dσ(y), ∀x ∈ Ω, (4.2.87)
∂∗ Ω

thanks to (A.0.183) and the boundary integration by parts formula [69, (11.1.62)]. In
particular, the cancelation condition (4.2.3) is satisfied by ∂γ U jk for each γ ∈ N0n .
Since when |γ| > 0 the integral kernel of ∂γ U jk also satisfies (4.2.1) with ε := 0
and k := |γ| − 1, Theorem 4.2.1 applies and yields (4.2.77) (bearing in mind [68,
(6.5.40) in Theorem 6.5.7], [68, (6.6.91)], and the fact that L(U jk f ) = 0 in Ω for
 p, p M
each function f ∈ Bs (∂Ω, σ) ). This takes care of item (b).
There remains to deal with the claims in item (c). Again, work under the assump-
tion that ∂Ω is bounded. Fix an arbitrary ψ ∈ 𝒞∞ c (R ) and consider the following
n

version of the double layer:

456 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
⨏
⎧
⎪
⎨ ψD f − ψ
⎪ f dσ if Ω is bounded,
 f :=
D ∂Ω (4.2.88)
⎪
⎪ ψD f
⎩ if Ω is unbounded.

Since under the present assumptions we have the continuous embedding (cf. (4.2.80)-
(4.2.81))
p, p
Bs (∂Ω, σ) → L 1 (∂Ω, σ), (4.2.89)
it follows that the above definition is meaningful. Note that, thanks to Proposi-
 annihilates constants. Also, since ∂Ω is bounded it follows
tion 1.3.6, the operator D
 satisfies (4.2.1) with k := 0 and ε := 0. Granted these
that the integral kernel of D
properties we may invoke Theorem 4.2.1, and (4.2.5) (with k := 0) presently ensures
(bearing in mind [68, (6.6.6)]) that there exists C ∈ (0, ∞) with the property that for
 p, p M
each f ∈ Bs (∂Ω, σ) we have
 1− 1 −s  
  f 
δ∂Ω D ≤ C f [Bsp, p (∂Ω,σ)] M .
p
(4.2.90)
L p (Ω, L n )

In turn, from (4.2.90), (4.2.88), (4.2.89), and item (i) in [68, Proposition
 8.7.1],

applied here with Σ := ∂Ω, r := 2 diam(Ω ∩ supp ψ) + 2 dist ∂Ω, Ω ∩ supp ψ ,
α := 1 − p(1 − s), and N := 0, we then conclude
 1− 1 −s  

δ∂Ω ψ D f  ≤ C f [Bsp, p (∂Ω,σ)] M
p
(4.2.91)
L p (Ω, L n )

 p, p M
for every f ∈ Bs (∂Ω, σ) . This is the first inequality in (4.2.78). When Ω
is bounded, the second estimate claimed in (4.2.78) is a consequence of the first
(now taking the cutoff function ψ to be identically one near Ω), and also relying on
(1.3.24), [68, (6.5.40) in Theorem 6.5.7], and [68, (6.6.91)].
Next, the first estimate claimed in (4.2.79) follows from Theorem 4.2.1, whose
applicability is ensured by (4.2.87) and the observation that the integral kernel of
ψU jk satisfies (4.2.1) with ε := 0 and k := 0. Finally, in the case when Ω is bounded,
the second estimate in (4.2.79) is a consequence of the first estimate in (4.2.79), also
keeping in mind [68, (6.5.40) in Theorem 6.5.7], [68, (6.6.91)], and the fact that we
 p, p M
have L(U jk f ) = 0 in Ω for each f ∈ Bs (∂Ω, σ) . 

Recall the agreement made in [69, Convention 8.3.7] (cf. also (A.0.217)), and
the scale of weighted maximal Sobolev spaces from [69, Definition 8.6.1] (see also
(A.0.215)-(A.0.216)).

Theorem 4.2.3 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with a compact

Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Let L be a second-order,
homogeneous, constant (complex) coefficient, weakly elliptic M × M system in Rn ,
and let D be the boundary-to-domain double layer potential operator associated
with L and Ω as in (1.3.18). Finally, fix
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 457
1 
n−1
n < p < ∞, (n − 1) p −1 + < s < 1,
(4.2.92)
and set a := 1 − s − p1 .

Then, with [69, Convention 8.3.7] (cf. also (A.0.217)) assumed throughout, the
following operators are well defined, linear, and continuous:
 p, p M  1, p M
D : Bs (∂Ω, σ) −→ Wa (Ω)bdd and
 p, p M  1, p M (4.2.93)
D : Bs (∂Ω, σ) −→ Wa, (Ω) ∩ Ker L if Ω is bounded.

Strengthen the original geometric hypotheses by assuming now that ∂Ω is a UR

set, and now restrict to p ∈ (1, ∞). Then, corresponding to the limiting case s = 0 of
(4.2.93), the following operators are well defined, linear, and continuous
 M  1, p M
D : L p (∂∗ Ω, σ) −→ W 1 (Ω)bdd and
1− p
 p M  1, p M (4.2.94)
D : L (∂∗ Ω, σ) −→ W 1 (Ω) ∩ Ker L if Ω is bounded,
1− p ,

while corresponding to the limiting case s = 1 the following operators are well
defined, linear, and continuous:
 p M  2, p M
D : L1 (∂∗ Ω, σ) −→ W 1 (Ω)bdd and
1− p
 p M  2, p M (4.2.95)
D : L1 (∂∗ Ω, σ) −→ W 1 (Ω) ∩ Ker L if Ω is bounded.
1− p ,

Finally, similar results are valid for the family of integral operators U jk defined
as in (4.2.76) for each j, k ∈ {1, . . . , n}.

Proof Combining (4.2.75) (with |γ| = 1) and the first estimate in (4.2.78). takes
care of the first operator in (4.2.93). With this in hand, the claims about the second
operator in (4.2.93) follow (assuming Ω is bounded) with help from (1.3.24), [68,
(6.5.40) in Theorem 6.5.7], and [68, (6.6.91)].
Let us now work under the stronger assumption that ∂Ω is a UR set. Fix some
p ∈ (1, ∞). Then (1.5.5) with q := p yields
 1− 1  

δ∂Ω ∇(D f ) ≤ C f [L p (∂∗ Ω,σ)] M .
p
(4.2.96)
L p (Ω, L n )

We need a similar estimate without the gradient. To this end, pick a cutoff function
ψ ∈ 𝒞∞c (R ). Also, recall from (1.3.18) that the integral kernel of the operator D is
n

the matrix-valued function defined at each x ∈ Ω and σ-a.e. y ∈ ∂Ω by

βα
k(x, y) := − 1∂∗ Ω (y)ν j (y)ar j (∂r Eγβ )(x − y) (4.2.97)
1≤γ,α ≤M

 αβ 
where L = ar j ∂r ∂j 1≤α,β ≤M is the writing of the given system with respect to
which the double layer operator D has been set up, ν = (ν j )1≤ j ≤n is the geometric
458 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

measure theoretic outward unit normal to Ω, and E = (Eγβ )1≤γ,β ≤M is the matrix-
valued fundamental solution associated with L as in [70, Theorem 1.4.2]. From
(4.2.97) and the estimates in [70, Theorem 1.4.2] we see that for each ε∗ < 1
there exists a constant C ∈ (0, ∞) such that |ψ(x)k(x, y)| ≤ C|x − y| −(n−ε∗ ) for
all x ∈ Ω and σ-a.e. y ∈ ∂Ω. As such, choosing r > 0 large enough so that
Ω ∩ supp ψ ⊆ x ∈ Ω : δ∂Ω (x) < r along with ε∗ ∈ (0, 1), permits us to invoke
[68, Proposition 8.7.10] (for the operator T := ψD) with β := 1 − p1 to conclude that
 M
there exists a constant C ∈ (0, ∞) such that for every f ∈ L p (∂Ω, σ) we have
 1− 1  

δ∂Ω ψD f  ≤ C f [L p (∂Ω,σ)] M .
p
(4.2.98)
L p (Ω, L n )

In concert, (4.2.96) and (4.2.98) readily imply the claims about the first operator in
(4.2.94). When Ω is bounded, the claims about the second operator in (4.2.94) are
consequences of what we have just proved and [69, Lemma 8.6.2]. Likewise, the
claims about the operators in (4.2.95) are consequences of (4.2.94), Lemma 1.3.2,
item (4) in [70, Theorem 2.4.1] (cf. [70, (2.4.34)]), [70, Theorem 1.4.2], [68,
(6.6.91)], (1.3.24), and [68, (6.5.40) in Theorem 6.5.7].
Finally, since the family of operators U jk , defined for j, k ∈ {1, . . . , } as in
(4.2.76), satisfy the same key analytical and algebraic properties that allowed us to
treat the double layer D (cf. (4.2.77), (4.2.87), and the fact that L(U jk f ) = 0 in Ω
 M
for each f ∈ L 1 (∂∗ Ω, σ) in particular), similar mapping properties continue to
hold for said family of operators. 

We have already commented that there is a large variety of double layer operators
associated with a given weakly elliptic system, and Corollary 4.2.2 together with
Theorem 4.2.3 apply to all such operators. A case in point is as follows. As noted
in Example 1.4.12, the boundary-to-domain Cauchy-Clifford integral operator C
(defined in (A.0.53)) is a particular example of a double layer (associated with the
Laplacian). As such, Corollary 4.2.2 implies that whenever Ω ⊆ Rn (where n ∈ N,
n ≥ 2) is an open set with an Ahlfors regular boundary, σ := H n−1 ∂Ω, and p, s
are as in (4.2.74), then for each θ ∈ (0, 1) and each γ ∈ N0n with |γ| > 0 there exists
some finite constant C = C(Ω, p, s, θ, γ) > 0 such that
 |γ |− 1 −s   
 ∂γ (C f ) 
δ∂Ω
p
,θ  L p (Ω, L n )
≤ C f Bsp, p (∂Ω,σ)⊗ Cn (4.2.99)

p, p
for each f ∈ Bs (∂Ω, σ) ⊗ Cn . In the same geometric setting, Theorem 4.2.3
further implies that
p, p 1, p
C : Bs (∂Ω, σ) ⊗ Cn −→ Wa (Ω)bdd ⊗ Cn

n < p < ∞,
is a continuous operator if n−1 (4.2.100)
1 
(n − 1) p − 1 + < s < 1 and a := 1 − s − p1 ,

and
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 459
p, p 1, p
C : Bs (∂Ω, σ) ⊗ Cn −→ Wa, (Ω) ⊗ Cn
is a bounded operator whenever Ω is bounded, (4.2.101)
 
n−1
n < p < ∞, (n − 1) p1 − 1 + < s < 1, and a := 1 − s − p1 .

Similar considerations also apply to the boundary-to-domain version of the ordinary

Cauchy operator in the complex plane, and to the boundary-to-domain Bochner-
Martinelli integral operator in the context of several complex variables (cf. Exam-
ple 1.4.9 and Example 1.4.16).
Theorem 4.2.1 is also directly applicable to the following brand of boundary-to-
domain integral operators (with kernels of a purely real variable nature, lacking any
direct link to partial differential equations).

Corollary 4.2.4 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set with the

property that ∂Ω is an Ahlfors regular set; in particular, Ω is a set of locally finite
perimeter. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit
normal to Ω, and abbreviate σ := H n−1 ∂Ω. Next, suppose k ∈ 𝒞2 (Rn \ {0}) is
function which is positive homogeneous of degree −N for some N ≥ n − 1, and
for each i, j ∈ {1, . . . , n} consider the integral operator Qi j acting on functions
 
f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n according to
∫     
Qi j f (x) := νi (y)∂y j k(x − y) − ν j (y)∂yi k(x − y) f (y) dσ(y) (4.2.102)
∂∗ Ω

for each x ∈ Ω. Finally, fix θ ∈ (0, 1) and assume that

1 
n < p ≤ ∞ and (n − 1) p − 1 + < s < 1.
n−1
(4.2.103)

Then there exists some finite constant C = C(Ω, k, p, s, θ, N, n) > 0 with the
p, p
property that for each f ∈ Bs (∂Ω, σ) one has
 N −n+2− 1 −s   
 
δ∂Ω
p
Qi j f ,θ  ≤ C f Bsp, p (∂Ω,σ) (4.2.104)
L p (Ω, L n )
 
with the convention that when p > 1 the solid maximal function Qi j f ,θ is replaced
by Qi j f . In particular,

if N := n − 1 and a := 1 − s − p1 , then the operator

p, p  ap 
Qi j : Bs (∂Ω, σ) → L p Ω, δ∂Ω L n is well defined, (4.2.105)
linear, and bounded.
Proof Upon recalling from [70, (2.4.120), (2.4.130)] that

Qi j 1 ≡ 0 in Ω, (4.2.106)

the fact that the estimate in (4.2.104) holds follows from Theorem 4.2.1 presently
used with ε := 0, k := N − (n − 1), and
460 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
    
q(x, y) := νi (y)∂y j k(x − y) − ν j (y)∂yi k(x − y) 1∂∗ Ω (y) (4.2.107)

for each x ∈ Ω and y ∈ ∂Ω. In turn, (4.2.104) implies readily implies the claim in
(4.2.105). 
Next, we discuss the boundary behavior of the double layer potential operator,
acting from boundary Besov spaces into weighted Sobolev spaces in open sets.
Theorem 4.2.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an (ε, δ)-domain whose
boundary is a compact UR set. Abbreviate σ := H n−1  ∂Ω and make the additional
assumption that ∂∗ Ω has full measure in ∂Ω, i.e., σ ∂Ω \ ∂∗ Ω = 0. Next, let L be a
second-order, homogeneous, constant (complex) coefficient, weakly elliptic M × M
system in Rn , and let D be the double layer potential operator associated with
the system L and the set Ω as in (1.3.18). Finally, pick an integrability exponent
p ∈ (1, ∞), a smoothness index s ∈ (0, 1), and introduce
 
a := 1 − s − p1 ∈ − p1 , 1 − p1 . (4.2.108)

Then the double layer operator acting in the context of

 p, p M  1, p M
D : Bs (∂Ω, σ) −→ Wa (Ω)bdd (4.2.109)

(cf. (4.2.93)), satisfies the jump-formula

 p, p M
TrΩ→∂Ω ◦ D = 12 I + K on Bs (∂Ω, σ) , (4.2.110)
 1, p M  p, p M
where TrΩ→∂Ω : Wa (Ω)bdd → Bs (∂Ω, σ) is the boundary trace oper-
ator considered in [69, Theorem 8.3.6] (further extended as in Remark 2 follow-
 p, p M
ing its statement), I denotes the identity operator on Bs (∂Ω, σ) , and K is
 p, p M
the boundary-to-boundary double layer operator acting on Bs (∂Ω, σ) as in
(4.1.1) of Theorem 4.1.1.
Proof Since the set ∂Ω is compact, [68, (5.11.35)] ensures that rad (Ω) > 0. Hence,
all operators involved are well defined, linear, and bounded, thanks to Theorem 4.2.3,
[69, Theorem 8.3.6], and Theorem 4.1.1. Fix a sufficiently large aperture parameter
κ > 0 so that the conclusion in [69, Corollary 8.3.9] holds. Also, pick an arbitrary
cutoff function ψ ∈ 𝒞∞ (Rn ) satisfying ψ ≡ 1 near ∂Ω. Then for each given function
  M c p, p M
f ∈ Lip (∂Ω) ⊆ Bs (∂Ω, σ) we may write

      κ−n.t.
TrΩ→∂Ω ◦ D f = TrΩ→∂Ω ψD f = ψD f 
∂Ω

= ( 12 I + K) f at σ-a.e. point on ∂Ω, (4.2.111)

by virtue of [69, Corollary 8.3.9], item (iv) of Theorem 1.5.1, and the fact that
ψ ≡ 1 near ∂Ω. As a consequence of this, the operators TrΩ→∂Ω ◦ D and 12 I + K
 M  p, p M
agree on Lip (∂Ω) . Since said operators are continuous on Bs (∂Ω, σ) and
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 461
 M  p, p M
since Lip (∂Ω) is a dense subset of Bs (∂Ω, σ) (cf. [69, Lemma 7.1.10]),
we ultimately conclude that the jump-formula claimed in (4.2.110) holds for each
 p, p M
function belonging to Bs (∂Ω, σ) . 
Theorem 4.2.5 is applicable to the entire family of double layer operators asso-
ciated with a given weakly elliptic system. This includes the boundary-to-domain
Cauchy-Clifford integral operator C (which fits into such a setting as indicated in
Example 1.4.12). In such a case, Theorem 4.2.5 gives that whenever Ω ⊆ Rn (where
n ∈ N, n ≥ 2) is an (ε, δ)-domain whose boundary is a compact UR set with the prop-
erty that σ ∂Ω \ ∂∗ Ω = 0, σ := H n−1 ∂Ω, p ∈ (1, ∞), s ∈ (0, 1), and a := 1 − s − p1 ,
then
p, p
TrΩ→∂Ω ◦ C = 12 I + C on Bs (∂Ω, σ) ⊗ Cn (4.2.112)
1, p p, p
where TrΩ→∂Ω : Wa (Ω)bdd ⊗ Cn → Bs (∂Ω, σ) ⊗ Cn is (the Clifford algebra
version of) the boundary trace operator from [69, Theorem 8.3.6] (further extended
as in Remark 2 following its statement), C is the boundary-to-domain version of the
Cauchy-Clifford integral operator considered in the context of (4.2.100), I denotes
p, p
the identity operator on Bs (∂Ω, σ) ⊗ Cn , and C is the boundary-to-boundary
p, p
Cauchy-Clifford integral operator acting on Bs (∂Ω, σ) ⊗ Cn as in (4.1.13) (with
q = p). Analogous considerations also apply to the boundary-to-domain version of
the ordinary Cauchy operator in the complex plane, and to the boundary-to-domain
Bochner-Martinelli integral operator in the context of several complex variables (cf.
Example 1.4.9 and Example 1.4.16).
Moving on, we propose to study mapping properties of boundary-to-domain
integral operators acting from Besov spaces with a negative amount of smoothness
into suitably weighted Lebesgue spaces. Once again, the strategy is to establish a
general result of this nature which identifies those basic features of said integral
operator which are responsible for results of this flavor.
Theorem 4.2.6 Let Ω be an open set in Rn (where n ∈ N, n ≥ 2) with a compact
Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Consider a function
r : Ω × ∂Ω → C with the property that there exist ε ∈ [0, 1), k ∈ R, and a finite
constant C0 > 0 such that
δ∂Ω (x) −ε 1
|r(x, y)| ≤ C0 δ∂Ω (x)−k
|x − y| |x − y| n−1 (4.2.113)
for every point x ∈ Ω and every point y ∈ ∂Ω.

In addition, suppose there exists a constant C1 ∈ (1, ∞) such that

δ∂Ω (x) −ε |y − z|
|r(x, y) − r(x, z)| ≤ C0 δ∂Ω (x)−k .
|x − y| |x − y| n (4.2.114)
for every x ∈ Ω and every y, z ∈ ∂Ω with |x − y| ≥ C1 |y − z|.

In this context, consider the integral operator acting on each function f ∈ L 1 (∂Ω, σ)
according to
462 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
∫
R f (x) := r(x, y) f (y) dσ(y), x ∈ Ω. (4.2.115)
∂Ω
Finally, select a function ψ ∈ 𝒞∞c (R ),
n
and assume
 
n−1
n−ε < p ≤ ∞ and (n − 1) p1 − 1 + < s < 1 − ε. (4.2.116)
p, p
Then the action of the operator R may be extended to Bs−1 (∂Ω, σ) and there
exists a finite constant C = C(Ω, ψ, ε, C0, C1, p, s) > 0 with the property that for
p, p
every f ∈ Bs−1 (∂Ω, σ) one has
 k+1− 1 −s 
 
δ∂Ω
p
· ψR f  ≤ C f B p, p (∂Ω,σ) . (4.2.117)
L p (Ω, L n ) s−1

Before presenting the proof of this theorem we wish to make two comments.
The first comment pertains to the nature of the estimate demanded in (4.2.114).
Specifically, assume Ω satisfies the following connectivity property:

there exists some constant c ∈ (0, ∞) such that for each pair of points
y, z ∈ ∂Ω there exists a rectifiable path t → γy,z (t) joining y and z in
Ω, parametrized with respect to its arc-length (hence |γy,z (t)| = 1 for (4.2.118)
H 1 -a.e. t), and with the property that length (γy,z ) ≤ c|y − z|.

For example, any open set with compact boundary and satisfying a local John
condition has the aforementioned connectivity property. Also, assume that for each
fixed x ∈ Ω, we may extend r(x, ·) to a function of class 𝒞1 in Ω \ {x} with the
property that there exists C ∈ (0, ∞) such that

δ∂Ω (x) −ε 1
|∇y r(x, y)| ≤ Cδ∂Ω (x)−k for each y ∈ Ω \ {x}. (4.2.119)
|x − y| |x − y| n
We then claim that (4.2.114) is presently satisfied. To see that this the case, fix x ∈ Ω
and pick a constant C1 > 0 much larger than c. Also, select a pair of arbitrary points
y, z ∈ ∂Ω and set L := length (γy,z ). Then, on account of (4.2.113) and (4.2.118),
for each x ∈ Ω with |x − y| ≥ C1 |y − z| we may estimate
∫ L
|r(x, y) − r(x, z)| ≤ |(∇y r)(x, γy,z (t))| dt
0

δ∂Ω (x)−k−ε
≤ C · L · sup
0<t<L |x − γy,z (t)| n−ε
δ∂Ω (x)−k−ε
≤ C|y − z| , (4.2.120)
|x − y| n−ε

where we have used the fact that |x − γy,z (t)| ≥ C|x − y| for every t ∈ [0, L] if
|x − y| ≥ C1 |y − z|, and that L ≤ c|y − z| (cf. (4.2.118)).
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 463

The second comment is that the same type of result continues to hold in the case
when the kernel r is matrix-valued, and the functions f are vector-valued (with the
understanding that the definition of R in (4.2.115) now takes into account the natural
action of the matrix r(x, y) on the vector f (y)).
After this digression, we turn to the proof of Theorem 4.2.6.
Proof of Theorem 4.2.6 The proof of (4.2.117) consists of treating several special
cases, then using interpolation to cover the whole range of indices. We begin by
considering the situation when p = ∞. In this scenario, fix a bounded subset Ωo of
Ω. The crucial step is establishing that there exists C ∈ (0, ∞) with the property that
 
1,1
r(x, ·) ∈ B1−s (∂Ω, σ) and r(x, ·)B1,1 (∂Ω,σ) ≤ Cδ∂Ω (x)s−k−1
1−s (4.2.121)
for every point x ∈ Ωo .

To this end, fix x ∈ Ωo and observe that, thanks to [69, (7.9.10)], the claims in
(4.2.121) follow as soon as we prove that
∫ ∫
|r(x, y) − r(x, z)|
dσ(y) dσ(z) ≤ Cδ∂Ω (x)s−k−1 (4.2.122)
|y − z| n−s
∂Ω ∂Ω

and ∫
|r(x, y)| dσ(y) ≤ Cδ∂Ω (x)s−k−1 (4.2.123)
∂Ω

for some C ∈ (0, ∞) independent of x. As regards (4.2.122), with the constant C1

as in (4.2.114), for z ∈ ∂Ω arbitrary split the inner integral according to whether
∫|x − y| < C1 |y − z|, or |x − y| ≥ C1 |y − z|. The focus then becomes bounding
I (z) dσ(z) by Cδ∂Ω (x)s−k−1 for  ∈ {1, 2, 3}, where
∂Ω 
∫
|r(x, y)|
I1 (z) := dσ(y), (4.2.124)
y ∈∂Ω, |x−y |<C1 |y−z | |y − z| n−s
∫
|r(x, z)|
I2 (z) := dσ(y), (4.2.125)
y ∈∂Ω, |x−y |<C1 |y−z | |y − z| n−s
∫
|r(x, y) − r(x, z)|
I3 (z) := dσ(y). (4.2.126)
y ∈∂Ω, |x−y | ≥C1 |y−z | |y − z| n−s

We start by making use of (4.2.113) to estimate I1 (z) as follows

∫
dσ(y)
I1 (z) ≤ Cδ∂Ω (x)−k−ε , (4.2.127)
|x − y| n−1−ε |y − z| n−s
y ∈∂Ω
|x−y |<C1 |y−z |

which further implies that

464 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
∫
I1 (z) dσ(z)
∂Ω
∫ ∫
1 dσ(z)
≤ Cδ∂Ω (x)−k−ε dσ(y)
∂Ω |x − y| n−1−ε ∂Ω\B(y, |x−y |/C1 ) |y − z| n−s
∫
1
≤ Cδ∂Ω (x)−k−ε |x − y| s−1 dσ(y)
∂Ω |x − y| n−1−ε

≤ Cδ∂Ω (x)−k−ε δ∂Ω (x)s−1+ε = Cδ∂Ω (x)s−k−1 . (4.2.128)

The first inequality in (4.2.128) is due to Fubini’s Theorem, while for the second and
third inequalities we have used [68, (7.2.5)] and [68, (8.7.92)].
∫
In fact, the same argument works to bound the integral ∂Ω I2 (z) dσ(z). Indeed,
|x − y| < C1 |y − z| implies |x − z| < (1 + C1 )|y − z| which allows us to enlarge
the domain of integration so that, after using Fubini’s Theorem, we then return to a
similar expression to the one just handled∫ above.
Consider next the task of estimating ∂Ω I3 (z) dσ(z). A combination of (4.2.114)
with (4.2.126) produces
∫
−k−ε dσ(y)
I3 (z) ≤ Cδ∂Ω (x) (4.2.129)
|x − y| n−ε |y − z| n−1−s
y ∈∂Ω
|x−y | ≥C1 |y−z |

which, in concert with [68, (7.2.5)] and [68, (8.7.92)], allows us to estimate
∫
I3 (z) dσ(z)
∂Ω
∫ ∫
−k−ε 1 dσ(z)
≤ Cδ∂Ω (x) dσ(y)
|x − y| n−ε |y − z| n−1−s
∂Ω ∂Ω∩B(y, |x−y |/C1 )
∫
dσ(y)
≤ Cδ∂Ω (x)−k−ε ≤ Cδ∂Ω (x)−k−ε δ∂Ω (x)−1+s+ε
|x − y| n−ε−s
∂Ω

= Cδ∂Ω (x)s−k−1 . (4.2.130)

∫
Collectively, the estimates for ∂Ω I (z) dσ(z) with  = {1, 2, 3} yield the bound
(4.2.122). This concludes the proof of (4.2.122).
Turning our attention to (4.2.123), we use (4.2.113) and [68, (8.7.92)] to estimate
∫ ∫
1
|r(x, y)| dσ(y) ≤ C0 δ∂Ω (x)−k−ε dσ(y)
|x − y| n−1−ε
∂Ω ∂Ω

≤ Cδ∂Ω (x)−k−ε ≤ Cδ∂Ω (x)s−k−1, (4.2.131)

4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 465

where the last inequality is based on the observation that our hypotheses imply
−ε − s + 1 > 0 and that we have δ∂Ω (x) stays bounded on the bounded set Ωo . This
establishes (4.2.123) and finishes the proof of (4.2.121).
Going further, recall from [69, Proposition 7.6.1] that
∞,∞  1,1 ∗
Bs−1 (∂Ω, σ) = B1−s (∂Ω, σ) . (4.2.132)

On account of (4.2.132) and (4.2.121), we may then define the action of the operator
∞,∞ 1,1 ∗
R on each f ∈ Bs−1 (∂Ω, σ), regarded as a functional in B1−s (∂Ω, σ) , according
to
 
(R f )(x) := B1,1 (∂Ω,σ) r(x, ·), f (B1,1 (∂Ω,σ))∗ for every x ∈ Ω. (4.2.133)
1−s 1−s

This is compatible with the original definition of R and, in concert with (4.2.121),
shows that there exists a constant C ∈ (0, ∞) with the property that
 
 k+1−s  ∞,∞
δ∂Ω · ψR f  ∞ ≤ C f B∞,∞ (∂Ω,σ) for all f ∈ Bs−1 (∂Ω, σ). (4.2.134)
L (Ω, L )
n s−1

Consider next the case when p = 1 and s ∈ (0, 1 − ε). The first order of business is
1,1
to extend the action of the operator R to the Besov space Bs−1 (∂Ω, σ). With this aim
in mind, observe that (4.2.114) implies that for each fixed point x ∈ Ω the function
r(x, ·) is Lipschitz on ∂Ω. Since the latter set is compact, it follows that, on the one
hand,
r(x, ·) ∈ 𝒞1−s (∂Ω) for each x ∈ Ω. (4.2.135)
On the other hand, [69, Proposition 7.6.1] and [69, (7.1.59)] ensure that
1,1  1,1  ∗∗  ∗
Bs−1 (∂Ω, σ) → Bs−1 (∂Ω, σ) = 𝒞1−s (∂Ω) . (4.2.136)

Based on (4.2.135)-(4.2.136) we may then naturally define the action of the operator
1,1
R on each given f ∈ Bs−1 (∂Ω, σ) by setting
 
(R f )(x) := 𝒞1−s (∂Ω) r(x, ·), f (𝒞1−s (∂Ω))∗ for every x ∈ Ω. (4.2.137)

Our next priority is to show that there exists some finite constant C > 0 with the
property that
   1−s ∗
 k−s 
δ∂Ω · ψR f  1 ≤ C f  (𝒞1−s (∂Ω))∗ for each f ∈ 𝒞 (∂Ω) . (4.2.138)
L (Ω, L )
n

In light of (4.2.136), this then proves (4.2.117) when p = 1 and s ∈ (0, 1 − ε).
As a preamble to the proof of (4.2.138) we first propose to show that for each
fixed function g ∈ L ∞ (Ω, L n ) with g L ∞ (Ω, L n ) ≤ 1 the estimate
∫ 
 
 δ∂Ω (x)k−s ψ(x)r(x, ·)g(x) dx  1−s ≤C (4.2.139)
Ω 𝒞 (∂Ω)
466 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

holds for some finite constant C > 0 independent of g. With this goal in mind,
observe that (4.2.139) is a consequence of the following two inequalities. First, we
claim that for every y, z ∈ ∂Ω the inequality
∫   

 g(x)δ∂Ω (x)k−s ψ(x) r(x, y) − r(x, z) dx  ≤ C |y − z| 1−s (4.2.140)
Ω

holds for some constant C ∈ (0, ∞) independent of y, z. Second, we claim that there
exists a finite constant C > 0 such that we have the inequality
∫ 
 
 g(x)δ∂Ω (x)k−s ψ(x)r(x, y) dx  ≤ C, ∀y ∈ ∂Ω. (4.2.141)
Ω

To prove (4.2.140), fix two arbitrary points y, z ∈ ∂Ω and recall (4.2.118). We can

then bound the integral in the left-hand side of (4.2.140) by
∫
C δ∂Ω (x)k−s |r(x, y)| dx
Ω∩B(y,C1 |y−z |)
∫
+C δ∂Ω (x)k−s |r(x, z)| dx
Ω∩B(y,C1 |y−z |)
∫
+C δ∂Ω (x)k−s |r(x, y) − r(x, z)| dx
Ω\B(y,C1 |y−z |)

=: I + II + III. (4.2.142)

As far as I is concerned, note that

∫
δ∂Ω (x)−s−ε
I≤C dx ≤ C|y − z| 1−s, (4.2.143)
Ω∩B(y,C1 |y−z |) |x − y| n−1−ε

by (4.2.113) and by [68, (8.7.3)] applied with r :=  C1 |y − z|, α := s + ε, and

N := n − 1 − ε. Since Ω ∩ B(y, C1 |y − z|) ⊆ Ω ∩ B z, (1 + C1 )|y − z| , a similar
argument applies to II. Thus, we are left with estimating III. For this, we use (4.2.114)
together with [68, (8.7.5)] applied with r := C1 |y − z|, α := s + ε, and N := n − ε
(choices for which we have n − N < α < 1, as required for the applicability of [68,
(8.7.5)]) and obtain
∫
δ∂Ω (x)−s−ε
III ≤ C|y − z| dx
Ω\B(y,C1 |y−z |) |x − y|
n−ε

≤ C|y − z| · |y − z| −s = C|y − z| 1−s, (4.2.144)

as desired. This finishes the proof of (4.2.140). As far as (4.2.141) is concerned,

choose a finite number R > sup{|a − b| : a ∈ ∂Ω, b ∈ Ω ∩ supp ψ}. Also, pick an
arbitrary point y ∈ ∂Ω. Then [68, (8.7.3)] used with r := R, α := s + ε < 1, and
N := n − 1 − ε < n − α gives
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 467
∫  ∫
δ∂Ω (x)−s−ε
 
 g(x)δ∂Ω (x)k−s ψ(x)r(x, y) dx  ≤ C dx ≤ C R1−s .
Ω |x − y| n−1−ε
B(y,R)\∂Ω
(4.2.145)
This establishes the estimate claimed in (4.2.141), and completes the proof of
(4.2.139).  ∗
Having justified (4.2.139), fix some f ∈ 𝒞1−s (∂Ω) and for each function
∞ (Ω, L n ) with g ∞
g ∈ Lcomp L (Ω, L n ) ≤ 1 use (4.2.137) together with (4.2.139) to
write
∫ 
 
 g(x)δ∂Ω (x)k−s ψ(x)(R f )(x) dx 
Ω
 ∫  
 
= 𝒞1−s (∂Ω) δ∂Ω (x)k−s ψ(x)r(x, ·)g(x) dx, f (𝒞1−s (∂Ω))∗ 
Ω
∫ 
 
≤ δ∂Ω (x)k−s ψ(x)r(x, ·)g(x) dx   f (𝒞1−s (∂Ω))∗
Ω 𝒞1−s (∂Ω)

≤ C f (𝒞1−s (∂Ω))∗ . (4.2.146)

On account of the arbitrariness of g, this ultimately establishes (4.2.138). Hence, the

proof of (4.2.117) in the case when p = 1 and s ∈ (0, 1 − ε) is finished.
Moving on, we treat (4.2.117) in the case when 1 < p < ∞ and 0 < s < 1 − ε.
The idea is to rely on what we proved so far and invoke Stein’s interpolation theorem
for analytic families of operators. Specifically, suppose 0 < s0 < s1 < 1 − ε and
consider the family of linear operators
k+z−[(1−z)s0 +zs1 ]
Lz f := δ∂Ω ψR f for z ∈ C with 0 ≤ Re z ≤ 1. (4.2.147)

This family exhibits an analytical dependence on the parameter z, and satisfies

k−s0
Re z = 0 =⇒ |Lz f | = δ∂Ω |ψR f |, (4.2.148)
k+1−s1
Re z = 1 =⇒ |Lz f | = δ∂Ω |ψR f |. (4.2.149)

This observation and the results already proved for p = 1 and p = ∞ then ensure
that the operators

Lz : Bs1,1
0 −1
(∂Ω, σ) −→ L 1 (Ω, L n ) for Re z = 0, (4.2.150)

Lz : Bs∞,∞
1 −1
(∂Ω, σ) −→ L ∞ (Ω, L n ) for Re z = 1, (4.2.151)

are well-defined, linear, and bounded. Granted these, Stein’s interpolation theorem
for analytic families of operators applies and, when used in concert with our complex
interpolation results from [69, Theorem 7.5.2], allows us to conclude that the operator
k+1− p1 −s p, p
δ∂Ω · ψR : Bs−1 (∂Ω, σ) −→ L p (Ω, L n ) (4.2.152)
468 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

is well defined, linear, and bounded, for every s ∈ (0, 1 − ε) and p ∈ [1, ∞]. This
finishes the proof of estimate (4.2.117) for the range of indices s ∈ (0, 1 − ε) and
p ∈ [1, ∞].
Alternatively, we may arrive at the same conclusions regarding the operator
(4.2.152) using the real method of interpolation. Concretely, given any s ∈ (0, 1 − ε),
from (4.2.136)-(4.2.138) we may conclude that the operator
1,1  −1 n 
δ∂Ω
k+1−s
· ψR : Bs−1 (∂Ω, σ) −→ L 1 Ω, δ∂Ω L (4.2.153)

is well defined, linear, and bounded, while from (4.2.132)-(4.2.134) we know that
we have a well-defined, linear, and bounded operator
∞,∞  −1 n 
δ∂Ω
k+1−s
· ψR : Bs−1 (∂Ω, σ) −→ L ∞ (Ω, L n ) = L ∞ Ω, δ∂Ω L (4.2.154)
−1 L n
(with the above equality a consequence of the fact that the measures L n and δ∂Ω
are mutually absolutely continuous in Ω). Then from (4.2.153)-(4.2.154), the real
interpolation result from [69, (7.4.18)] (used with Σ := ∂Ω, p0 := 1, p1 := ∞),
and well-known real interpolation results for generic Lebesgue spaces (cf., e.g., [2,
Theorem 5.2.1, p. 109]) we conclude that the operator
p, p  −1 n 
δ∂Ω
k+1−s
· ψR : Bs−1 (∂Ω, σ) −→ L p Ω, δ∂Ω L (4.2.155)

is well defined, linear, and bounded whenever 1 ≤ p ≤ ∞ and 0 < s < 1 − ε. In

turn, this is equivalent to having estimate (4.2.117) valid in the range 1 ≤ p ≤ ∞
and 0 < s < 1 − ε.
At this stage, there remains to analyze the case when
1 
n−ε < p < 1 and (n − 1) p − 1 < s < 1 − ε.
n−1
(4.2.156)

In this scenario, the plan is to eventually use the decomposition from [69, Theo-
rem 7.2.7] of f into a linear combination of smooth blocks and smooth atoms. To
be specific, fix η ∈ (1 − s, 1) and recall from [69, Definition 7.2.1] that an η-smooth
atom of type (p, s − 1) is a function a ∈ L ∞ (∂Ω, σ) with the property that there exist
r > 0 and xa ∈ ∂Ω such that

(1) supp a ⊆ B(xa, r) ∩ ∂Ω, (4.2.157)

n−1
(2) a L ∞ (∂Ω,σ) ≤ r s−1− p , (4.2.158)

(3) |a(x) − a(y)| ≤ r p −η

s−1− n−1
· |x − y| η for all x, y ∈ ∂Ω, (4.2.159)
∫
(4) a dσ = 0. (4.2.160)
∂Ω

An η-smooth block enjoys similar properties, except that (4.2.160) is not necessarily
satisfied. This being said, a key provision in [69, Theorem 7.2.7] is that all blocks in-
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 469
p, p
volved in the decomposition of a given distribution f ∈ Bs−1 (∂Ω, σ) have uniformly
large supports (say, surface balls of a common radius ρ).
In view of these considerations, the immediate goal is to show that for each fixed
η ∈ (1 − s, 1) there exists some constant C = C(Ω, ψ, C0, C1, p, s) ∈ (0, ∞) such that
 k+1− 1 −s 
 
δ∂Ω
p
· ψRa ≤C (4.2.161)
L p (Ω, L n )

whenever a is either a smooth η-atom on ∂Ω, or a smooth η-block of type (p, s − 1)

with r = ρ, a fixed positive number which depends only on ∂Ω. To justify this
claim, first assume the function a ∈ L ∞ (∂Ω, σ) is as in (4.2.157)-(4.2.160). Fix
x ∈ Ω \ B(xa, C1 r) and, for each point y ∈ ∂Ω ∩ B(xa, r), use the vanishing moment
condition of the atom together with estimate (4.2.114) and the Ahlfors regularity of
∂Ω to write
  ∫   
   
|ψ(x)| (Ra)(x) = |ψ(x)|  r(x, xa ) − r(x, y) a(y) dσ(y)
∂Ω∩B(x a ,r)

δ∂Ω (x)−k−ε
≤ C r n a L ∞ (∂Ω,σ)
|x − xa | n−ε
δ∂Ω (x)−k−ε
≤ C r s+(n−1)(1− p )
1
. (4.2.162)
|x − xa | n−ε
Consequently,
∫
 p
δ∂Ω (x)k p+p−1−ps |ψ(x)| p (Ra)(x) dx
Ω\B(x a ,C1 r)
∫
δ∂Ω (x) p−pε−1−ps
≤ Cr ps+(n−1)(p−1) dx. (4.2.163)
|x − xa | p(n−ε)
Ω\B(x a ,C1 r)

With α := 1 − p(1 − s − ε) and N := p(n − ε), the conditions 1 > α > n − N

are satisfied, given that we assume (4.2.156). Thus, [68, Proposition 8.7.1] applies
and allows us to further bound the right-hand side in (4.2.163) by Cr ps+(n−1)(p−1) ·
r n−(1−p(1−s−ε))−p(n−ε) = C. This proves that
∫
 p
δ∂Ω (x)k p+p−1−ps |ψ(x)| p (Ra)(x) dx ≤ C, (4.2.164)
Ω\B(x a ,C1 r)

for some finite constant C > 0, depending only on Ω, ψ, p, s, Co , C1 .

The treatment of the situation when (4.2.160) is dropped and, instead, it is assumed
that r = ρ, where ρ ∈ (0, ∞) is a fixed number depending only on ∂Ω, is similar.
In this case, the fact that x belongs to a bounded subset of Ω (namely Ω ∩ supp ψ)
compensates the fact that we no longer assume the vanishing moment condition
(4.2.160). Specifically, the same type of argument that has produced (4.2.162) now
470 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 
gives (using (4.2.113), the fact that |x − y| ≥ 1 − C1−1 |x − xa | for each point
x ∈ Ω \ B(xa, C1 r) and each y ∈ B(xa, r) ∩ ∂Ω, and the Ahlfors regularity of ∂Ω),
 
|ψ(x)| (Ra)(x) (4.2.165)

δ∂Ω (x) −ε 1 |x − xa |
≤ C r s+(n−1)(1− p ) δ∂Ω (x)−k
1
|ψ(x)|,
|x − xa | |x − xa | n r

for each x ∈ Ω \ B(xa, C1 r). Upon noticing that

   
|x − xa |/r |ψ(x)| ≤ ρ−1 sup |ψ| sup |x − y| : x ∈ supp ψ, y ∈ ∂Ω
Rn

= C(∂Ω, ψ) < ∞, (4.2.166)

we deduce that ψ(x)(Ra)(x) satisfies a bound of the same nature as in (4.2.162).

Thus, (4.2.164) holds in this case as well.
Let us now consider the contribution coming from integrating near the support
of the function a ∈ L ∞ (∂Ω, σ). The strategy is to rely on a rescaling argument in
order to be able to use what we have already proved in the case p = 1. To do so, pick
τ ∈ R and introduce
a(x) := r τ a(x) for x ∈ ∂Ω.
 (4.2.167)
For now, assume that
   
(n − 1) p1 − 1 − s < τ < (n − 1) p1 − 1 + 1 − s. (4.2.168)

Based on the properties of a (i.e., (4.2.157)-(4.2.160), or (4.2.157)-(4.2.159)) it is

not difficult to check that 
a is an η-smooth atom of type 1, τ + s − (n − 1)( p1 − 1) − 1 ,
 
or an η-smooth block of type 1, τ + s − (n − 1)( p1 − 1) − 1 , depending on the original
nature of a ∈ L ∞ (∂Ω, σ), provided
 
η > 1 − τ − s + (n − 1) p1 − 1 . (4.2.169)

Assuming this is the case, [69, Theorem 7.2.8] guarantees the existence of a constant
C ∈ (0, ∞), independent of a, such that

a ∈ B1,1
 (∂Ω, σ) and 
a B1,1 (∂Ω,σ) ≤C (4.2.170)
τ+s−(n−1)( p1 −1)−1 1 −1)−1
τ+s−(n−1)( p

If we further require that

1 
τ < (n − 1) p − 1 + 1 − ε − s, (4.2.171)

we may use the bounds already proved for the operator R corresponding to p = 1 in
order to obtain
 k−τ−s+(n−1)( 1 −1) 
 
δ∂Ω
p
· ψR
a 1 ≤ C a B1,1 (∂Ω,σ) ≤ C. (4.2.172)
n L (Ω, L ) 1 −1)−1
τ+s−(n−1)( p
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 471

Hence, by first applying Hölder’s inequality, then (4.2.172), and then [68, (8.7.3)]
τp
with α := n − 1−p and N := 0, we may further estimate
∫
  p
δ∂Ω (x)k+1− p −s |ψ(x)| Ra(x)
1
dx
B(x a ,C1 r)∩Ω
p
 ∫
  "
p −1) a(x) dx ##
|ψ(x)| R
1
−τ p
≤ Cr δ∂Ω (x)k−τ−s+(n−1)(
!B(x a,C1 r)∩Ω $
1−p
 ∫ "
τp
× δ∂Ω (x) 1−p −n dx ##
!B(x a ,C1 r)∩Ω $
1−p
 ∫ "
τp
≤ Cr −τ p
δ∂Ω (x) 1−p −n dx ## ≤ C, (4.2.173)
!B(x a,C1 r)∩Ω $
provided 1 
(n − 1) p − 1 < τ. (4.2.174)
Note that (4.2.174) is needed in order to ensure that the necessary condition α < 1,
under which [68, (8.7.3)] with N = 0 holds, is satisfied. Given the assumptions in
(4.2.116) and bearing in mind that 1 − s < η < 1, it follows that it is possible to select
τ simultaneously satisfying (4.2.168), (4.2.169), (4.2.171), and (4.2.174). For such
a τ fixed, estimate (4.2.173) holds. Now (4.2.161) follows by combining (4.2.164)
and (4.2.173), completing the proof of (4.2.161) in the case when s and p satisfy
(4.2.156).
Moving on, we now consider the case when f is an arbitrary
  distribution in the
p, p
Besov space Bs−1 (∂Ω, σ) with n−ε
n−1
< p < 1 and (n − 1) p1 − 1 < s < 1 − ε. [69,
Theorem 7.2.7] guarantees the existence of a sequence {a j } j ∈N consisting of either
η-smooth atoms of type (p, s − 1), or η-smooth blocks of type (p, s − 1) supported
in surface balls of a common radius ρ, along with a numerical sequence {λ j } j ∈N
belonging to  p (a space with which bp, p (∂Ω) naturally identifies; cf. [69, (7.2.24)
in Definition 7.2.5]) satisfying

∞ 1/p
|λ j | p ≤ C f B p, p (∂Ω,σ) < +∞, (4.2.175)
s−1
j=1

for some constant C ∈ (0, ∞) independent of f , and such that if


m
fm := λ j a j for each m ∈ N (4.2.176)
j=1

then
472 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p, p
fm −→ f in Bs−1 (∂Ω, σ) as m → ∞. (4.2.177)
Observe that if we set
1 
s∗ := s − (n − 1) p − 1 ∈ (0, s) ⊆ (0, 1 − ε), (4.2.178)

s∗ −1
then 0 < s∗ < s < 1 and p1 − s−1
n−1 = 1− n−1 . Granted these, [69, Theorem 7.7.4]
(used with Σ := ∂Ω) gives
p, p
Bs−1 (∂Ω, σ) → Bs1,1
∗ −1
(∂Ω, σ) continuously. (4.2.179)

As a consequence of (4.2.177) and (4.2.179) we conclude that

fm −→ f in Bs1,1
∗ −1
(∂Ω, σ) as m → ∞. (4.2.180)

Let us now introduce

k+1− p1 −s k+1− p1 −s
G := δ∂Ω · ψR f and G m := δ∂Ω · ψR fm for each m ∈ N. (4.2.181)

These functions enjoy several remarkable properties. First, as consequence of

(4.2.181), (4.2.176), (4.2.161), and (4.2.175), for each m ∈ N we may estimate
(mindful that we are presently assuming 0 < p < 1)

m  k+1− 1 −s p
p  
G m  L p (Ω, L n ) ≤ |λ j | p δ∂Ω p · ψRa j  p
L (Ω, L n )
j=1


m
p
≤C |λ j | p ≤ C f B p, p (∂Ω,σ), (4.2.182)
s−1
j=1

for some C ∈ (0, ∞) independent of f and m. Similarly, given any m , m ∈ N with

m ≤ m , based on (4.2.181), (4.2.176), and (4.2.161) to estimate (again, keeping in
mind that 0 < p < 1)
   k+1− 1 −s p
G m − G m  p p  
L (Ω, L n )
≤ δ∂Ω p · ψR( fm − fm ) p
L (Ω, L n )


m  k+1− 1 −s p
 
≤ |λ j | p δ∂Ω p · ψRa j  p
L (Ω, L n )
j=m


m
≤C |λ j | p, (4.2.183)
j=m

for some constant C ∈ (0, ∞) independent of f , m, m . Lastly, for each m ∈ N we

may employ (4.2.138) to write
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 473
 −1+ 1   
 −1+ p1   k−s 
δ∂Ω · G − δ∂Ω · G m  = δ∂Ω · ψR( f − fm )
p

L 1 (Ω, L n ) L 1 (Ω, L n )

≤ C f − fm B1,1 (∂Ω,σ), (4.2.184)

s−1

for a constant C ∈ (0, ∞) independent of f , m. Collectively, (4.2.182), (4.2.183),

(4.2.184), (4.2.175), (4.2.177), and (4.2.180) permit us to conclude that

G m ∈ L p (Ω, L n ) for all m ∈ N and sup G m  L p (Ω, L n ) ≤ C f B p, p (∂Ω,σ),

s−1
m∈N
(4.2.185)

{G m }m∈N is a Cauchy sequence in L p (Ω, L n ), and (4.2.186)

−1+ p1 −1+ p1
δ∂Ω · G m −→ δ∂Ω · G in L 1 (Ω, L n ) as m → ∞. (4.2.187)

 ∈ L p (Ω, L n ) such that

Next, from (4.2.186) we see that there exists some function G

G m → G in L (Ω, L ) as m → ∞. As such, there exists a subsequence {G mi }i ∈N
p n

with the property that

 at L n -a.e. point in Ω as i → ∞.
G mi → G (4.2.188)

Also, (4.2.185) implies that

 L p (Ω, L n ) ≤ C f B p, p (∂Ω,σ)
 G (4.2.189)
s−1

for some constant C ∈ (0, ∞) which is independent of the function f . Finally,

−1+ p1 −1+ p1
(4.2.187) ensures that δ∂Ω · G mi → δ∂Ω · G in L 1 (Ω, L n ) as i → ∞, hence there
−1+ 1 −1+ 1
exists a sub-subsequence {G mi j } j ∈N with the property that δ∂Ω p ·G mi j → δ∂Ω p ·G
at L n -a.e. point in Ω as j → ∞. Since the latter further implies that G mi j → G at
L n -a.e. point in Ω as j → ∞, we then deduce from this and (4.2.188) that G = G  at
L n -a.e. point in Ω. Together with (4.2.73) and (4.2.181),
 this
 ultimately establishes
n−1
(4.2.117) in the case when n−ε < p < 1 and (n − 1) p1 − 1 < s < 1 − ε. Hence, the
estimate claimed in (4.2.117) has been now proved for the full range of indices p, s
described in (4.2.116). This finishes the proof of Theorem 4.2.6. 

Theorem 4.2.6 has a host of remarkable applications, and in the next three corol-
laries we explore some of its concrete manifestations.

Corollary 4.2.7 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with a

compact Ahlfors regular boundary and abbreviate σ := H n−1 ∂Ω. Also, assume
b ∈ 𝒞1 (Rn \ {0}) is a function with the property that there exist No, N ∈ R and some
C ∈ (0, ∞) such that

|b(z)| ≤ C|z| −No and |(∇b)(z)| ≤ C|z| −N −1

(4.2.190)
for each z ∈ B(0, 1) \ {0}.
474 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Define the integral operator acting on each function f ∈ L 1 (∂Ω, σ) according to

∫
(B f )(x) := b(x − y) f (y) dσ(y), x ∈ Ω. (4.2.191)
∂Ω

Finally, select a function ψ ∈ 𝒞∞

c (R ) and assume p, s, k satisfy
n

 
n < p ≤ ∞, (n − 1) p1 − 1 + < s < 1,
n−1
(4.2.192)
k ≥ max{N, No } − n + 1 and k > s − 1.
p, p
Then B may be extended to the Besov space Bs−1 (∂Ω, σ) and there exists a finite
constant C > 0, which depends only on Ω, b, ψ, N, No, p, s, k, with the property that
p, p
for each f ∈ Bs−1 (∂Ω, σ) one has
 k+1− 1 −s 
 
δ∂Ω
p
· ψB f  ≤ C f B p, p (∂Ω,σ) . (4.2.193)
L p (Ω, L n ) s−1

Proof Define

r(x, y) := b(x − y), for each x ∈ Ω and y ∈ ∂Ω, (4.2.194)

and note that, thanks to (4.2.190) we have

|r(x, y)| ≤ C|x − y| −No for all x ∈ Ω, y ∈ ∂Ω, (4.2.195)

and also using the Mean Value Theorem,

|r(x, y) − r(x, z)| ≤ C|y − z||x − y| −N −1 for all x ∈ Ω, y, z ∈ ∂Ω

(4.2.196)
with the property that |x − y| ≥ 2|y − z|.

Let us next remark that, since Ω is assumed to be bounded,

for each α, β ∈ R satisfying α ≥ (β)+ there exists a constant
C ∈ (0, ∞) with the property that δ∂Ω (x)α ≤ C|x − y| β for (4.2.197)
each x ∈ Ω and each y ∈ ∂Ω.

Set N := max{N, No }. Based on (4.2.192) and the observation that 0 < 1 − s < 1,
we may choose a number ε ∈ [0, 1) (which is smaller than, but very close to, 1 − s)
satisfying
0 ≤ ε < 1 − s and k ≥ max{ N  − n + 1, −ε}. (4.2.198)
Together with the first line in (4.2.192), the upper bound for ε (in the double inequality
above) ensures that
 
(n − 1) p1 − 1 + < s < 1 − ε and n−ε n−1
< p ≤ ∞. (4.2.199)

Also, since the last inequality in (4.2.198) implies

4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 475

 − n + 1 + ε, 0} = ( N
k + ε ≥ max{ N  − n + 1 + ε)+ (4.2.200)
 − n + 1 + ε) to
we may invoke the estimate in (4.2.197) (with α := k + ε and β := N
conclude that there exists some finite constant C > 0 with the property that

δ∂Ω (x)k+ε ≤ C|x − y| N −n+1+ε for all x ∈ Ω and y ∈ ∂Ω. (4.2.201)

In turn, from (4.2.201) and (4.2.195)-(4.2.196) it readily follows that the func-
tion (4.2.194) satisfies (4.2.113)-(4.2.114) for the present ε and k, with C1 := 2.
Since (4.2.192) is also satisfied (cf. (4.2.198)) by the current choice of ε, it follows
that Theorem 4.2.6 applies and gives that B may be extended to the Besov space
p, p
Bs−1 (∂Ω, σ) in such a way that (4.2.193) is satisfied. 

Next, we turn our attention to the boundary-to-domain single layer acting on

Besov spaces. First, we introduce this operator property.

Proposition 4.2.8 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with a compact

Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, let L be a second-
order, homogeneous, constant (complex) coefficient, weakly elliptic M × M system
in Rn , and recall the boundary-to-domain single layer potential operator 𝒮 defined
in relation to L and Ω as in (1.3.6), where E = (Eαβ )1≤α,β ≤M is the fundamental
solution associated with the system L as in [70, Theorem 1.4.2]. Finally, define
  
𝒮 f (x) := Eαβ (x − ·)∂Ω, fβ (Lip (∂Ω))
Lip (∂Ω)
1≤α ≤M (4.2.202)
  M
for each f = ( fβ )1≤β ≤M ∈ Lip (∂Ω) and each x ∈ Ω.

Then, if
1 
n−1
n < p ≤ ∞, 0 < q ≤ ∞, (n − 1) p −1 + < s < 1, (4.2.203)

the action of the single layer potential operator 𝒮 may be naturally adapted to
 p,q M
the Besov space Bs−1 (∂Ω, σ) in the following fashion. Since (4.2.203) entails
 
p − n−1 ∈ − n−1 , 1 , if p < ∞ it is possible to choose p∗ ∈ (1, ∞) and s∗ ∈ (0, s)
1 s s
s∗
such that p1 − n−1s
= p1∗ − n−1 . According to [69, Theorem 7.7.4], this guarantees
that in this case for any q∗ ∈ (1, ∞) one has
p,q p ,q
Bs−1 (∂Ω, σ) → Bs∗∗−1∗ (∂Ω, σ). (4.2.204)

When p = ∞, then [69, (7.7.6)] may be invoked to ensure that (4.2.204) holds for any
 p, p M
p∗, q∗ ∈ (1, ∞) and s∗ ∈ (0, s). Now, given any f = ( fβ )1≤β ≤M ∈ Bs−1 (∂Ω, σ)
define
  
𝒮 f (x) := p ,q
(B s∗∗−1 ∗ (∂Ω,σ))∗ Eαβ (x − ·)∂Ω, fβ B p∗, q∗ (∂Ω,σ) (4.2.205)
s∗ −1 1≤α ≤M
476 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p ,q
for each and each x ∈ Ω, presently viewing each fα in Bs∗∗−1∗ (∂Ω, σ) (via (4.2.204))
  p ,q ∗
and considering each Eαβ (x − ·)∂Ω in Bs∗∗−1∗ (∂Ω, σ) via
 p∗,q∗  p ,q ∗
Eαβ (x − ·)∂Ω ∈ Lip (∂Ω) ⊆ B1−s∗
(∂Ω, σ) = Bs∗∗−1∗ (∂Ω, σ) (4.2.206)

where p∗, q∗ are the Hölder conjugate exponent of p∗ , q∗ (cf. [69, Proposition 7.6.1]).
With this interpretation it follows that
 p,q M  M
𝒮 : Bs−1 (∂Ω, σ) −→ 𝒞∞ (Ω) (4.2.207)
 M
is a well-defined linear operator which is also continuous when 𝒞∞ (Ω) is
equipped with the Frechét topology of uniform convergence of partial derivatives on
compact sets. Moreover, one may take derivatives of 𝒮 f by differentiating under the
duality pairing (4.2.205) in a natural fashion, i.e.,
  
∂γ (𝒮 f )(x) = p ,q
(B s∗∗−1 ∗ (∂Ω,σ))∗ (∂γ Eαβ )(x − ·)∂Ω, fβ B p∗, q∗ (∂Ω,σ)
s∗ −1 1≤α ≤M
for each multi-index γ ∈ N0n and each point x ∈ Ω,
(4.2.208)
and one has
 p,q M
L(𝒮 f ) = 0 in Ω for each f ∈ Bs−1 (∂Ω, σ) . (4.2.209)

Finally,
if Ω also satisfies a two-sided local John condition then the single
layer operator in the context of (4.2.207) is compatible with the (4.2.210)
single layer operator in the context of (1.5.41).
Proof From [69, Proposition 7.6.2] we know that the definition in (4.2.205) is
unambiguous. The fact that (4.2.207) is a well-defined, linear, continuous operators,
and that (4.2.208) holds may be seen by reasoning as in the proof of Lemma 2.2.1
(in which (2.2.5) is now replaced by f = lim PN f , with family of operators
N →∞
{PN } N ∈N defined as in [69, (7.3.37)]). Also, (4.2.209) is implied by (4.2.208) and
[70, Theorem 1.4.2].
To justify the compatibility claim in (4.2.210), work under the additional assump-
tion that Ω also satisfies a two-sided local John condition. Then (4.2.204) and [69,
(11.8.15), (11.11.12)] imply that we have the following continuous embeddings:
p,q p ,q p  
Bs−1 (∂Ω, σ) → Bs∗∗−1∗ (∂Ω, σ) → L−1∗ (∂Ω, σ) → Lip (∂Ω) . (4.2.211)
 p,q M
Fix f = ( fβ )1≤β ≤M ∈ Bs−1 (∂Ω, σ) . Then for each index α ∈ {1, . . . , M } and
fixed point x ∈ Ω we may write
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 477
    
𝒮f α (x) = (B p∗, q∗ (∂Ω,σ))∗ Eαβ (x − ·)∂Ω, fβ B p∗, q∗ (∂Ω,σ)
s∗ −1 s∗ −1

  
= Lip(∂Ω) Eαβ (x − ·)∂Ω, fβ (Lip(∂Ω))
  
= (L p∗ (∂Ω,σ))∗ Eαβ (x − ·)∂Ω, fβ L p∗ (∂Ω,σ) (4.2.212)
−1 −1

thanks to (4.2.205), (4.2.206), [69, (7.6.9), (11.8.16)], and (A.0.136). This proves
(4.2.210). 

We shall now establish the following result.

Corollary 4.2.9 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with a compact

Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, let L be a second-
order, homogeneous, constant (complex) coefficient, weakly elliptic M × M system
in Rn , and recall the boundary-to-domain single layer potential operator 𝒮 acting
in the context of Besov spaces as in Proposition 4.2.8. Finally, assume that
1 
n < p ≤ ∞ and (n − 1) p − 1 + < s < 1.
n−1
(4.2.213)

Then for each multi-index γ ∈ N0n and each number k ∈ R satisfying

k ≥ |γ| − 1 and k > s − 1 (4.2.214)

and each parameter θ ∈ (0, 1) there exists a finite constant C > 0, which depends
 p, p M
only on Ω, L, p, s, γ, k, θ, with the property that for each f ∈ Bs−1 (∂Ω, σ) one
has  k+1− 1 −s 
  
δ∂Ω
p ∂γ 𝒮 f   ≤ C f [B p, p (∂Ω,σ)] M
,θ L p (Ω, L n ) s−1 (4.2.215)
provided the set Ω is bounded.
Moreover, for each cutoff function ψ ∈ 𝒞∞ c (R ) there exists a finite constant
n

C > 0, which depends only on Ω, L, p, s, γ, k, ψ, with the property that for each
 p, p M
f ∈ Bs−1 (∂Ω, σ) one has
 k+1− 1 −s  

δ∂Ω
p
ψ ∂γ 𝒮 f  ≤ C f [B p, p (∂Ω,σ)] M . (4.2.216)
L p (Ω, L n ) s−1

Proof Let E be the fundamental solution associated with L as in [70, Theorem 1.4.2].
The idea is to apply Corollary 4.2.7 with b := ∂γ E. If either n ≥ 3 or |γ| > 0, then the
hypotheses in (4.2.190) and (4.2.192) are satisfied with No := N := n − 2 + |γ| and
the current p, s, k. In the remaining case, when n = 2 and |γ| = 0, then the hypotheses
in (4.2.190) and (4.2.192) are satisfied with any No ∈ (0, s), N := 0, and the current
p, s, k. Granted these, (4.2.215) follows from the version of (4.2.193) for a bounded
domain (in which case we may take the cutoff function ψ ∈ 𝒞∞ c (R ) to be identically
n

one near Ω), also bearing in mind (4.2.209), [68, (6.5.40) in Theorem 6.5.7], and
[68, (6.6.91)]. Finally, (4.2.216) is implied by (4.2.193) as is. 
478 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

In turn, Corollary 4.2.9 is the main ingredient in the proof of the fact that the
boundary-to-domain single layer operator maps Besov spaces into weighted maximal
Sobolev spaces.

Theorem 4.2.10 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with a compact

Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, suppose L is a
second-order, homogeneous, constant (complex) coefficient, weakly elliptic M × M
system in Rn , and recall the boundary-to-domain single layer potential operator 𝒮,
associated with L and Ω, acting in the context of Besov spaces as in Proposition 4.2.8.
Finally, let  
n < p < ∞, (n − 1) p1 − 1 + < s < 1,
n−1
(4.2.217)
and set a := 1 − s − p1 .
Then, with [69, Convention 8.3.7] (cf. also (A.0.217)) assumed throughout, the
following operators are well defined, linear, and continuous:
 p, p M  1, p M
𝒮 : Bs−1 (∂Ω, σ) −→ Wa (Ω)bdd and
 p, p M  1, p M (4.2.218)
𝒮 : Bs−1 (∂Ω, σ) −→ Wa, (Ω) ∩ Ker L if Ω is bounded.

Furthermore, corresponding to the end-point case p = ∞,

if Ω is also assumed to be a locally uniform domain,

 ∞,∞ M  s M
then the operator 𝒮 : Bs−1 (∂Ω, σ) −→ 𝒞bdd (Ω) (4.2.219)
is well defined, linear, and continuous for each s ∈ (0, 1).

Next, strengthen the original hypotheses on Ω by assuming that ∂Ω is actually a

UR set. Then, corresponding to the limiting case s = 0 of (4.2.218), the operators
 p M  1, p M
𝒮 : L−1 (∂∗ Ω, σ) −→ W 1 (Ω)bdd and
1− p
 p M  1, p M (4.2.220)
𝒮 : L−1 (∂∗ Ω, σ) −→ W 1 (Ω) ∩ Ker L if Ω is bounded,
1− p ,

are well defined, linear, and continuous for each p ∈ (1, ∞), while corresponding to
the limiting case s = 1 of (4.2.218) the operators
 M  2, p M
𝒮 : L p (∂∗ Ω, σ) −→ W 1 (Ω)bdd and
1− p
 p M  2, p M (4.2.221)
𝒮 : L (∂Ω, σ) −→ W 1 (Ω) ∩ Ker L if Ω is bounded
1− p ,

are also well defined, linear, and continuous for each p ∈ (1, ∞).

Proof Throughout, fix an arbitrary cutoff function ψ ∈ 𝒞∞ c (R ). Specializing

n

(4.2.216) twice, first to the case when |γ| = 0 and k = 0, then to the case when
|γ| = 1 and k = 0, implies that there exists some constant C ∈ (0, ∞) with the
property that
4.2 Boundary-to-Domain Layer Potentials from Besov into Weighted Sobolev Spaces 479
   
 a    a 
δ∂Ω ψ 𝒮 f  + δ∂Ω ψ ∇𝒮 f  ≤ C f [B p, p (∂Ω,σ)] M (4.2.222)
L p (Ω, L n ) L p (Ω, L n ) s−1

 p, p M
for each f ∈ Bs−1 (∂Ω, σ) . From this and (A.0.211) we conclude that the first
operator in (4.2.218) is well defined, linear, and continuous. When Ω is bounded, fix
some θ ∈ (0, 1) and specialize (4.2.215) twice, first taking |γ| = 0 and k = 0, then
taking |γ| = 1 and k = 0. Collectively, these imply that there exists C ∈ (0, ∞) such
that
    
 a     a   
δ∂Ω 𝒮 f ,θ  p + δ ∂Ω ∇𝒮 f ,θ
≤ C f [B p, p (∂Ω,σ)] M (4.2.223)
L (Ω, L )
n p n L (Ω, L ) s−1

 p, p M
for each f ∈ Bs−1 (∂Ω, σ) . Hence, the claims regarding the second operator in
(4.2.218) follow from this in light of (A.0.211) and (4.2.209).
For the next segment in the proof, work under the additional assumption that Ω is
a locally uniform domain. Define Ωr := x ∈ Ω : dist(x, ∂Ω) ≥ r for a sufficiently
small threshold r > 0, and fix s ∈ (0, 1). We shall use the version of (4.2.216)
corresponding to p = ∞ when |γ| = 1 and k = 0. In concert with [68, (5.11.75)],
this guarantees (bearing in mind that ψ is compactly supported) that there exists a
constant C = C(Ω, L, s) ∈ (0, ∞) such that
    

ψ𝒮 f [𝒞s (Ω)] M ≤ C δ1−s ∇(ψ𝒮 f )
∂Ω + C · sup ψ𝒮 f 
L ∞ (Ω, L n ) Ωr

≤ C f [B∞,∞ (∂Ω,σ)] M (4.2.224)

s−1

 ∞,∞ M
for each function f ∈ Bs−1 (∂Ω, σ) . Thus, the claim made in (4.2.219) is
justified.
For the remainder of the proof strengthen make the assumption that ∂Ω is actually
a UR set, and fix some p ∈ (1, ∞). To set the stage, we make the claim that there
 M
exists some constant C ∈ (0, ∞) with the property that for each h ∈ L p (∂Ω, σ)
we have
 1− 1  

δ∂Ω ψ ∂γ 𝒮h p ≤ Ch[L p (∂Ω,σ)] M whenever |γ| ≤ 2.
p
(4.2.225)
L (Ω, L )
n

Indeed, when |γ| ≤ 1, use [69, (7.7.8)] to embed

p, p
L p (∂Ω, σ) → Bs−1 (∂Ω, σ) for any s ∈ (0, 1). (4.2.226)

1− 1 1−s− 1
Fix s ∈ (0, 1). Upon observing that δ∂Ω p ≤ Cδ∂Ω p on Ω ∩ supp ψ, we may invoke
(4.2.216) to estimate
 1− 1    1−s− 1  
  ∂γ 𝒮h
δ∂Ω ψ ∂γ 𝒮h p ≤ δ ψ
p p
C ∂Ω
L (Ω, L )
n p n L (Ω, L )

≤ Ch[B p, p (∂Ω,σ)] M ≤ Ch[L p (∂Ω,σ)] M , (4.2.227)

s−1
480 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

as wanted. Finally, if |γ| = 2, say γ = ei + e j for i, j ∈ {1, . . . , n}, and if we set

∫
Tj h(x) := (∂j E)(x − y)h(y) dσ(y), ∀x ∈ Ω, (4.2.228)
∂Ω

then, thanks to [70, (2.4.34)], we have

 1− 1    1− 1  
 
δ∂Ω ψ ∂γ 𝒮h p
p γ 
p
≤ C δ ∂Ω
∂ 𝒮h
L (Ω, L )
n L p (Ω, L n )
∫ 1/p
≤C |(∂i Tj h)(x)| p dist(x, ∂Ω) p−1 dx
Ω

≤ Ch[L p (∂Ω,σ)] M , (4.2.229)

which finishes the proof of (4.2.225).

 p M
Turning to the claims in (4.2.220) in earnest, pick some f ∈ L−1 (∂∗ Ω, σ)
and recall from (1.5.40) the manner in which 𝒮 acts on f . Specifically, from
 M
[69, Proposition 11.8.2] we know that there exist f0, f jk ∈ L p (∂∗ Ω, σ) , for all
1 ≤ j < k ≤ n, satisfying

 f0 [L p (∂∗ Ω,σ)] M +  f jk [L p (∂∗ Ω,σ)] M ≤ C f [L p (∂∗ Ω,σ)] M (4.2.230)
−1
1≤ j<k ≤n

for some finite constant C > 0 independent of f and, with 1 < p < ∞ such that
1/p + 1/p = 1,
∫ 
 
p
[L (∂∗ Ω,σ)] M f , g p
[L (∂ Ω,σ)] M
=  f0, g +  f jk , ∂τ j k g dσ
−1 ∗
1 ∂∗ Ω 1≤ j<k ≤n
 p M
for every function g ∈ L1 (∂∗ Ω, σ) .
(4.2.231)
Then if E = (Eαβ )1≤α,β ≤M is the matrix-valued fundamental solution associated
with L in Rn as in [70, Theorem 1.4.2], then from (1.5.40) and (4.2.231) we conclude
that for each x ∈ Ω we have
∫
(𝒮 f )(x) = E(x − y) f0 (y) dσ(y)
∂∗ Ω
 ∫
+ ∂τ j k (y) [E(x − y)] f jk (y) dσ(y)
1≤ j<k ≤n ∂∗ Ω

  
= (𝒮 
f0 )(x) + ∂j 𝒮(νk 
f jk )(x) − ∂k 𝒮(ν j 
f jk )(x) , (4.2.232)
1≤ j<k ≤n

 M
where f0 , 
f jk are the extensions of the functions f0, f jk ∈ L p (∂∗ Ω, σ) by zero
outside ∂∗ Ω, to the entire ∂Ω. Thanks to this representation formula, the estimate
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 481

established in (4.2.225) (also used with ψ replaced by various derivatives of ψ, as

needed), and (4.2.230), we then conclude that
 1− 1    1− 1  
 
δ∂Ω ψ𝒮 f  p
p 
p
+ δ ∂Ω
∇(ψ𝒮 f ) ≤ C f [L p (∂∗ Ω,σ)] M , (4.2.233)
L (Ω, L )
n L (Ω, L )
p n −1

for some constant C ∈ (0, ∞) independent of f . This proves that the first operator in
(4.2.220) is well defined, linear, and bounded. With this in hand, the claims about
the second operator in (4.2.220) are seen to be true by virtue of [69, Lemma 8.6.2].
Finally, the fact that the first operator in (4.2.221) is well defined, linear, and
bounded is seen directly from (4.2.225) and definitions (cf. (A.0.211)), while the
claims about the second operator in (4.2.221) follow from this and [69, Lem-
ma 8.6.2]. 

4.3 Boundary-to-Domain Layer Potentials from Besov Spaces

into Besov and Triebel-Lizorkin Spaces

We are now in a position to elucidate the smoothing effect of the boundary-to-domain

double layer potential operator, measured on Besov and Triebel-Lizorkin scales.

Theorem 4.3.1 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an (ε, δ)-domain with a

compact Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, suppose
L is a second-order, homogeneous, constant (complex) coefficient, weakly elliptic
M × M system in Rn , and recall the boundary-to-domain double layer potential
operator D associated with L and Ω as in (1.3.18). Then, with [69, Convention 8.3.7]
(cf. also (A.0.217)) assumed throughout, the following assertions are true.

(1) The double layer operator D induces a well-defined, linear, and continuous
mapping
 p,q M % &M
p,q
D : Bs (∂Ω, σ) −→ B 1 (Ω)bdd whenever
s+ p (4.3.1)
 
n < p < ∞, (n − 1) p1 − 1 + < s < 1, 0 < q ≤ ∞.
n−1

(2) The double layer operator D induces a well-defined, linear, and continuous
mapping
 p, p M % &M
p,q
D : Bs (∂Ω, σ) −→ F 1 (Ω)bdd whenever
s+ p (4.3.2)
1 
n−1
n < p < ∞, (n − 1) p − 1 + < s < 1, n+s+1/p
n
< q ≤ ∞.

(3) Corresponding to taking s := 2 − p1 and q := 2 in (4.3.2), the double layer

operator D induces a well-defined, linear, and continuous mapping
482 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
% &M  p M
p, p
D : B 1 (∂Ω, σ) −→ h2 (Ω)bdd with n
n+1 < p < 1, (4.3.3)
2− p

p
where h2 (Ω) is the local Hardy-based Sobolev space of order 2 in Ω (cf. [69,
(9.2.43)]).
Also, corresponding to having s := 1 − p1 and q := 2 in (4.3.2), the double layer
operator D induces a well-defined, linear, and continuous mapping
% &M  1, p M
p, p
D : B 1 (∂Ω, σ) −→ Wbdd (Ω) for each p ∈ (1, ∞). (4.3.4)
1− p

(which is a particular case of Theorem 4.2.3).

(4) Strengthen the original hypotheses on Ω by assuming that ∂Ω is actually a UR
set. Then, as a limiting case of (4.3.1)-(4.3.2), formally corresponding to making
s := 0, the operators
 M % &M
p,q
D : L p (∂∗ Ω, σ) −→ B 1 (Ω)bdd with
p (4.3.5)
1 < p < ∞ and p ≤ q ≤ ∞,

as well as % &M
 M p,q
D : L p (∂∗ Ω, σ) −→ F 1 (Ω)bdd with
p (4.3.6)
1 < p < ∞ and n
n+1/p < q ≤ ∞,
are well defined, linear, and continuous. In particular, corresponding to the case
when p = q = 2, it follows that (recall (A.0.86) and [69, (9.2.22)])
 M  1/2  M
D : L 2 (∂∗ Ω, σ) −→ Hbdd (Ω) (4.3.7)

is a well-defined, linear, and continuous operator. Furthermore, as a limiting

case of (4.3.1)-(4.3.2), formally corresponding to making s := 1, the operators
 p M % &M
p,q
D : L1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd with
1+ p (4.3.8)
1 < p < ∞ and p ≤ q ≤ ∞,

as well as % &M
 p M p,q
D : L1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd with
1+ p (4.3.9)
1 < p < ∞ and n+1+1/p <
n
q ≤ ∞,
are well-defined, linear, and continuous. In particular, corresponding to the case
when p = q = 2, it follows that (recall (A.0.86) and [69, (9.2.22)])
 M  3/2  M
D : L12 (∂∗ Ω, σ) −→ Hbdd (Ω) (4.3.10)

is a well-defined, linear, and continuous operator.

4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 483

(5) Similar results to those recorded in items (1)-(4) above for the double layer
D also hold for the integral operators U jk , defined for j, k ∈ {1, . . . , } as in
(4.2.76).

Before presenting the proof of this result we wish to remark that whenever the set
p,q p,q p,q p p
Ω is actually bounded, we have B 1 (Ω)bdd = B 1 (Ω), F 1 (Ω), h2 (Ω)bdd = h2 (Ω),
s+ p s+ p s+ p
1, p
Wbdd (Ω) = W 1, p (Ω), etc., i.e., the subscript bdd may be omitted in all cases.
Here is the proof of Theorem 4.3.1.
Proof of Theorem 4.3.1 Assume first that
 
n < p < ∞, (n − 1) p1 − 1 + < s < 1, < q ≤ ∞.
n−1 n
n+s+1/p (4.3.11)

The fact that 1

p − s
n−1 < 1, itself a consequence of the bounds for s in (4.3.11),
implies that
s∗
there exists s∗ ∈ (0, s) such that 1
p − s
n−1 + n−1 < 1. (4.3.12)

Consequently,

s∗
−1  
p∗ := 1
p − s
n−1 + n−1 =⇒ p∗ ∈ max{1, p}, ∞ . (4.3.13)

In particular, from (4.3.12)-(4.3.13), [69, Theorem 7.7.4], and [69, (7.9.10)-(7.9.11)]

we conclude that we have the continuous embeddings
 p, p M  p ,p M  M
Bs (∂Ω, σ) → Bs∗∗ ∗ (∂Ω, σ) → L p∗ (∂Ω, σ) . (4.3.14)
 p, p M
To proceed, fix an arbitrary f ∈ Bs (∂Ω, σ) . Then the function u := D f
is a smooth null-solution of the system L in Ω (cf. (1.3.24)). In addition, from
(4.3.13)-(4.3.14), (1.5.1), and [68, (8.6.51)] we see that
 p M
u ∈ Lbdd (Ω, L n ) . (4.3.15)

In concert with Theorem 4.2.3, the proof so far shows that if a := 1 − s − 1

p then
 p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 1, p M (4.3.16)
ψu ∈ Wa (Ω) for each ψ ∈ 𝒞∞
c (R ),
n

with quantitative control. Granted this, we may then conclude from [69, Corol-
lary 9.2.32] (used with k := 0) that the double layer potential operator induces a
well-defined, linear, and continuous mapping in the context of (4.3.2).
In turn, from (4.3.2) and real interpolation, based on [69, (7.4.4) in Theorem 7.4.1]
and [69, Theorem 9.2.22], we conclude that the double layer potential operator also
induces a well-defined, linear, and continuous mapping in the context of (4.3.1).
484 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Next, the claim pertaining to (4.3.3) is seen by specializing the result in (4.3.2)
to the case when n+1 n
< p < 1, s := 2 − p1 , and q := 2, bearing in mind the
identification result from [69, Proposition 9.2.7] (currently used with k := 2). Also,
in the case when Ω is bounded, the assertions made in relation to (4.3.5)-(4.3.7) are
clear from item (11) of [70, Theorem 2.4.1] (whose present applicability is ensured
by (1.3.18) and [70, Theorem 1.4.2]). When Ω is an exterior domain, apply what we
have just proved to the bounded set ΩR := Ω ∩ B(0, R) for some sufficiently large
R > 0, identifying function from L p (∂∗ Ω, σ) with function in L p (∂∗ ΩR, σR ) (where
σR := H n−1 ∂ΩR ) by extending them by zero from ∂∗ Ω to ∂∗ ΩR .
 M
Alternatively, given any f ∈ L p (∂∗ Ω, σ) , the function u := D f is known
from (4.2.94), (1.5.1), and [68, (8.6.51)] to satisfy
 p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 1, p M (4.3.17)
ψu ∈ W 1 (Ω) for each ψ ∈ 𝒞∞c (R ),
n
1− p

with quantitative control. Having established this, we may then conclude from [69,
Corollary 9.2.32] (with k := 0, s := 0, and a := 1 − p1 ) that the double layer potential
operator induces a well-defined, linear, and continuous mapping both in the context
of (4.3.6) and in the context of (4.3.5).
Going further, consider the claims concerning the double layer potential operator
 p M
in (4.3.8)-(4.3.10). Given any f ∈ L1 (∂∗ Ω, σ) with 1 < p < ∞, the function
u := D f is a smooth null-solution of the system L in Ω (cf. (1.3.24)). Also, [68,
 p M
(8.6.51)] together with (1.5.1) ensure that u ∈ Lbdd (Ω, L n ) . In concert with
(4.2.95) from Theorem 4.2.3 the argument so far gives that
 p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 2, p M (4.3.18)
ψu ∈ W 1 (Ω) for each ψ ∈ 𝒞∞c (R ),
n
1− p

with quantitative control. With this in hand, [69, Corollary 9.2.32] permits us to
conclude that the double layer potential operator induces a well-defined, linear,
and continuous mapping in the context of (4.3.8)-(4.3.9). In turn, (4.3.10) is a
consequence of this and [69, (9.2.22)].
Finally, since the operators U jk , defined for j, k ∈ {1, . . . , } as in (4.2.76), satisfy
the same key analytical and algebraic properties that allowed us to deduce the
mapping properties for the double layer D recorded in items (1)-(4) (cf. (4.2.77),
 M
(4.2.87), and the fact that L(U jk f ) = 0 in Ω for each f ∈ L 1 (∂∗ Ω, σ) in
particular), said mapping properties continue to hold for this family of operators. 

Theorem 4.3.1 specialized to the case of the boundary-to-domain Cauchy-Clifford

integral operator C (regarded as a special case of a double layer associated, as
explained in Example 1.4.12, with the Laplacian) yields the conclusion that whenever
Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an (ε, δ)-domain with a compact Ahlfors regular
boundary, and σ := H n−1 ∂Ω, then
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 485
p,q p,q
C : Bs (∂Ω, σ) ⊗ Cn −→ B 1 (Ω)bdd ⊗ Cn with
s+ p
1  (4.3.19)
n < p < ∞, (n − 1) p − 1 + < s < 1, 0 < q ≤ ∞,
n−1

p, p p,q
C : Bs (∂Ω, σ) ⊗ Cn −→ F 1 (Ω)bdd ⊗ Cn with
s+ p
1  (4.3.20)
n−1
n < p < ∞, (n − 1) p − 1 + < s < 1, n+s+1/p
n
< q ≤ ∞,
and
p, p p
C:B (∂Ω, σ) ⊗ Cn −→ h2 (Ω)bdd ⊗ Cn with n
n+1 < p < 1, (4.3.21)
2− p1

are all well-defined, linear, and continuous operators. Moreover, the analogues of the
claims in item (4) of Theorem 4.3.1 (regarding the end-point cases s = 0 and s = 1
of (4.3.19)-(4.3.20)) hold for the the boundary-to-domain Cauchy-Clifford integral
operator C as well. Finally, similar considerations also apply to the boundary-to-
domain version of the ordinary Cauchy operator in the complex plane, and to the
boundary-to-domain Bochner-Martinelli integral operator in the context of several
complex variables (cf. Example 1.4.9 and Example 1.4.16).
We continue by discussing the boundary behavior of the (boundary-to-domain
version of the) double layer potential operator acting from boundary Besov spaces
into Besov and Triebel-Lizorkin spaces defined on domains.

Theorem 4.3.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an (ε, δ)-domain whose

 Abbreviate σ := H ∂Ω and make the additional
boundary is a compact UR set. n−1

assumption that σ ∂Ω \ ∂∗ Ω = 0. Finally, suppose L is a second-order, homoge-
neous, constant (complex) coefficient, weakly elliptic M × M system in Rn , and recall
the boundary-to-domain double layer potential operator D associated with L and
Ω as in (1.3.18). Then the following claims are true.

(1) One has the jump-formula

 p,q M
TrΩ→∂Ω ◦ D = 12 I + K on Bs (∂Ω, σ) , whenever
  (4.3.22)
n < p < ∞, (n − 1) p1 − 1 + < s < 1, 0 < q ≤ ∞,
n−1

 p,q M  p,q M
where D : Bs (∂Ω, σ) → Bs+1/p (Ω)bdd is the double layer potential
 p,q M  p,q M
operator considered in (4.3.1), TrΩ→∂Ω : Bs+1/p (Ω)bdd → Bs (∂Ω, σ)
is the boundary trace operator from [69, (9.4.91) in item (ii) of Theorem 9.4.5]
(further extended as in Remark 3 following the statement of [69, Theorem 9.4.5]),
 p,q M
I denotes the identity operator on Bs (∂Ω, σ) , and K is the boundary-to-
 p,q M
boundary double layer operator acting on Bs (∂Ω, σ) as in (4.1.1) of
Theorem 4.1.1.
(2) One has the jump-formula
486 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 p, p M
TrΩ→∂Ω ◦ D = 12 I + K on Bs (∂Ω, σ) , whenever
  (4.3.23)
n < p < ∞, (n − 1) p1 − 1 + < s < 1, n+s+1/p < q ≤ ∞,
n−1 n

 p,q M  p, p M
where, this time, TrΩ→∂Ω : Fs+1/p (Ω)bdd → Bs (∂Ω, σ) is the bound-
ary trace operator from [69, (9.4.93) in item (ii) of Theorem 9.4.5] (further
extended as in Remark 3 following the statement of [69, Theorem 9.4.5]),
 p, p M  p,q M
D : Bs (∂Ω, σ) → Fs+1/p (Ω)bdd is the double layer potential op-
 p, p M
erator considered in (4.3.2), I denotes the identity operator on Bs (∂Ω, σ)
and, finally, K is the boundary-to-boundary double layer operator acting on
 p, p M
Bs (∂Ω, σ) as in (4.1.1) of Theorem 4.1.1.

We wish to note that, as a particular case of Theorem 4.3.2, similar results are
valid for the Cauchy-Clifford integral operator, the ordinary Cauchy operator in the
complex plane, and the Bochner-Martinelli integral operator in the context of several
complex variables.
Proof of Theorem 4.3.2 Consider the claim made in item (1). For starters, observe
that all operators involved are well-defined, linear, and continuous, in the contexts
specified there thanks to item (1) of Theorem 4.3.1, item (ii) of [69, Theorem 9.4.5],
and Theorem 4.1.1. Fix a sufficiently large aperture parameter κ > 0 as in [69, Corol-
  M  p,q M
lary 8.3.9]. Then for each function f belonging to Lip (∂Ω) ⊆ Bs (∂Ω, σ)
we may rely on [69, Corollary 8.3.9] and item (iv) of Theorem 1.5.1 to write

      κ−n.t.
TrΩ→∂Ω ◦ D f = TrΩ→∂Ω D f = D f 
∂Ω

= ( 12 I + K) f at σ-a.e. point on ∂Ω. (4.3.24)

 M
Hence, the operators TrΩ→∂Ω ◦ D and 12 I + K agree on Lip (∂Ω) . Given that
 p,q M  M
said operators are continuous on Bs (∂Ω, σ) and given that Lip (∂Ω) is a
 p,q M
dense subset of Bs (∂Ω, σ) (cf. [69, Lemma 7.1.10]), we ultimately conclude
that the jump-formula claimed in (4.3.22) holds for each function belonging to
 p,q M
Bs (∂Ω, σ) . Finally, the proof of the jump-formula in the context described in
item (2) is similar, relying on the same main ingredients as before, and this finishes
the proof of the theorem. 

Next, we turn our attention to the smoothing properties of the boundary-to-domain

single layer potential operator, measured on Besov and Triebel-Lizorkin scales.

Theorem 4.3.3 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an (ε, δ)-domain with a

compact Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, suppose
L is a second-order, homogeneous, constant (complex) coefficient, weakly elliptic
M × M system in Rn , and recall the boundary-to-domain single layer potential
operator 𝒮, associated with L and Ω, acting in the context of Besov spaces as in
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 487

Proposition 4.2.8. Then, with [69, Convention 8.3.7] (cf. also (A.0.217)) assumed
throughout, the following assertions are true.

(1) The single layer operator 𝒮 induces a well-defined, linear, and continuous
mapping
 p,q M % &M
p,q
𝒮 : Bs−1 (∂Ω, σ) −→ B 1 (Ω)bdd whenever
s+ p (4.3.25)
 
n < p < ∞, (n − 1) p1 − 1 + < s < 1, 0 < q ≤ ∞.
n−1

(2) The single layer operator 𝒮 induces a well-defined, linear, and continuous
mapping
 p, p M % &M
p,q
𝒮 : Bs−1 (∂Ω, σ) −→ F 1 (Ω)bdd whenever
s+ p (4.3.26)
 
n−1
n < p < ∞, (n − 1) p1 − 1 + < s < 1, n+s+1/p
n
< q ≤ ∞.

(3) Corresponding to s := 2 − p1 and q := 2 in (4.3.25)-(4.3.26), the single layer

operator induces a well-defined, linear, and continuous mapping
% &M  p M
p, p
𝒮 : B 1 (∂Ω, σ) −→ h2 (Ω)bdd with n
n+1 < p < 1, (4.3.27)
1− p

p
where h2 (Ω) is the local Hardy-based Sobolev space of order 2 in Ω (cf. [69,
(9.2.43)]).
(4) Strengthen the original hypotheses on Ω by assuming that ∂Ω is actually a UR
set. Then, as a limiting case of (4.3.25)-(4.3.26), formally corresponding to
making s := 0, the operators
 p M % &M
p,q
𝒮 : L−1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd with
p (4.3.28)
1 < p < ∞ and p ≤ q ≤ ∞,

as well as % &M
 p M p,q
𝒮 : L−1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd with
p (4.3.29)
1 < p < ∞ and n
n+1/p < q ≤ ∞,
are well-defined, linear, and continuous. In particular, corresponding to the case
when p = q = 2, it follows that (recall (A.0.86) and [69, (9.2.22)])
 2 M  1/2  M
𝒮 : L−1 (∂∗ Ω, σ) −→ Hbdd (Ω) (4.3.30)

is a well-defined, linear, and continuous operator. Furthermore, as a limiting

case of (4.3.25)-(4.3.26), formally corresponding to making s := 1, the operators
488 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

 M % &M
p,q
𝒮 : L p (∂Ω, σ) −→ B 1 (Ω)bdd with
1+ p (4.3.31)
1 < p < ∞ and p ≤ q ≤ ∞,

as well as % &M
 M p,q
𝒮 : L p (∂Ω, σ) −→ F 1 (Ω)bdd with
1+ p (4.3.32)
1 < p < ∞ and n+1+1/p <
n
q ≤ ∞,
are well defined, linear, and continuous. In particular, corresponding to the case
when p = q = 2, it follows that (recall (A.0.86) and [69, (9.2.22)])
 M  3/2  M
𝒮 : L 2 (∂Ω, σ) −→ Hbdd (Ω) (4.3.33)

is a well-defined, linear, and continuous operator.

Proof Consider first the case when the indices p, q, s are as in (4.3.26). Select some
 p, p M
arbitrary function f ∈ Bs−1 (∂Ω, σ) and consider u := 𝒮 f in Ω. From (4.2.216)
written for |γ| = 0 and k := s − 1 + p1 we see that, in a quantitative fashion,
 p M
u ∈ Lbdd (Ω, L n ) . (4.3.34)

If we now set a := 1 − s − p1 , from (4.3.34), (4.2.207), (4.2.209), and Theorem 4.2.10

we deduce that  p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 1, p M (4.3.35)
ψu ∈ Wa (Ω) for each ψ ∈ 𝒞∞ c (R ),
n

with quantitative control. Having established this, we may then conclude from [69,
Corollary 9.2.32] that the single layer potential operator induces a well-defined,
linear, and continuous mapping in the context of (4.3.26). With this in hand, from
(4.3.26) and real interpolation (using [69, (7.4.4) in Theorem 7.4.1] and [69, The-
orem 9.2.22]) we conclude that the single layer potential operator also induces a
well-defined, linear, and bounded mapping in the context of (4.3.25). Next, the claim
regarding (4.3.27) is seen by specializing the result in (4.3.26) to the case when
n+1 < p < 1, s := 2 − p , and q := 2, keeping in mind the identification result from
n 1

[69, Proposition 9.2.7] (used here with k := 2).

 p M
Turning our attention to item (4), fix an arbitrary f ∈ L−1 (∂∗ Ω, σ) and define
u := 𝒮 f in Ω. From [68, (8.6.51)] and (1.5.45) (used with q := p) we see that
 p M
u ∈ Lbdd (Ω, L n ) in a quantitative fashion. Combining this with (4.2.220) we
then conclude that
 p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 1, p M (4.3.36)
ψu ∈ W 1 (Ω) for each ψ ∈ 𝒞∞ c (R ),
n
1− p
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 489

with quantitative control. With this in hand, [69, Corollary 9.2.32] then gives that
the single layer potential operator induces a well-defined, linear, and continuous
mapping both in the context of (4.3.28) and in the context of (4.3.29).
Let us now analyze the single layer operator in the context of (4.3.32). Given
 M
f ∈ L p (∂Ω, σ) with 1 < p < ∞, consider the function u := 𝒮 f in Ω. Then from
(4.3.34) and (4.2.221) we conclude that
 p M
u ∈ Lbdd (Ω, L n ) ∩ Ker L such that
 2, p M (4.3.37)
ψu ∈ W 1 (Ω) for each ψ ∈ 𝒞∞c (R ),
n
1− p

with quantitative control. Once this has been established, we may invoke [69, Corol-
lary 9.2.32] to conclude that the single layer potential operator induces a well-defined,
linear, and continuous mapping both in the context of (4.3.31) and in the context of
(4.3.32). 

Let us now consider the action of the boundary-to-boundary single layer potential
operator acting on Besov scales.

Theorem 4.3.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with a compact

Ahlfors regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, suppose L is a
second-order, homogeneous, constant (complex) coefficient, weakly elliptic M × M
system in Rn , and recall the boundary-to-boundary single layer potential operator
S, initially associated with L and Ω as in (1.3.62). Then the following statements are
valid.

(1) The single layer operator S from (2.2.116) further extends, in a unique fashion,
to a linear and bounded mapping
 p,q M  ∗ M
S : Bs−1 (∂Ω, σ) −→ L p ,q (∂Ω, σ)
whenever n−1
n < p < 1, 0 < q ≤ ∞, (4.3.38)
1  −1
(n − 1) p − 1 < s < 1, and p∗ := 1
p − s
n−1 .

Moreover, various choices of p, q, s, p∗ as above yield operators which are com-

patible with one another. In particular, the single layer operator S from (2.2.116)
induces well-defined linear and bounded mappings, which are compatible with
each other, in the context
 p, p M  ∗ M
S : Bs−1 (∂Ω, σ) −→ L p (∂Ω, σ) whenever
1  −1 (4.3.39)
n−1
< p < 1, (n − 1) − 1 < s < 1, p ∗ := 1 − s .
n p p n−1

(2) If, in addition to the original hypotheses, Ω is also assumed to satisfy a two-
sided local John condition, then the single layer operator S from (1.3.62) further
extends, in a unique fashion, to a linear and bounded mapping
490 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
 p,q M  p,q M
S : Bs−1 (∂Ω, σ) −→ Bs (∂Ω, σ)
(4.3.40)
whenever 1 < p < ∞, 0 < q ≤ ∞, 0 < s < 1.

Once again, various choices of p, q, s as above yield operators which are com-
patible with one another. Moreover, if the exponents p, q, p , q ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q and s ∈ (0, 1), then
   
[B s (∂Ω,σ)] M S f, g [B−s = [B p, q (∂Ω,σ)] M f, SL  g [B p , q (∂Ω,σ)] M
p, q p ,q
(∂Ω,σ)] M s−1 1−s
 p,q M  p ,q M
for each f ∈ Bs−1 (∂Ω, σ) and g ∈ B−s (∂Ω, σ) ,
(4.3.41)

where SL  is associated with the system L  (the real transpose of L) in the same
manner S has been associated with the original system L.
(3) In addition to the original hypotheses, make also the assumption that Ω is an
(ε, δ)-domain. Then the single layer operator S from (4.3.39) further extends, in
a unique fashion, to a linear and bounded mapping
 p,q M  p,q M
S : Bs−1 (∂Ω, σ) −→ Bs (∂Ω, σ) whenever
  (4.3.42)
n < p < ∞, 0 < q ≤ ∞, (n − 1) p1 − 1 + < s < 1.
n−1

Various choices of p, q, s as above yield operators which are compatible with one
another. Also, as a consequence of (4.3.42) and [69, (7.7.54)], it follows that the
single layer potential operator induces a well-defined, linear, and continuous
mapping
p ∗,q ∗
S : H p (∂Ω, σ) −→ B (∂Ω, σ) whenever
1−(n−1)( p1 − p1∗ )
(4.3.43)
n−1
n < p < p∗ ≤ 1 and 0 < q∗ ≤ ∞.

Furthermore, one has the trace formula

 p,q M
TrΩ→∂Ω ◦ 𝒮 = S on Bs−1 (∂Ω, σ) (4.3.44)

when
 p,q M  p,q M
TrΩ→∂Ω : Bs+1/p (Ω)bdd → Bs (∂Ω, σ) is the boundary
trace operator from [69, (9.4.91) in item (ii) of Theorem 9.4.5]
(further extended as indicated in Remark 3 following its state-
 p,q M  p,q M (4.3.45)
ment), 𝒮 : Bs−1 (∂Ω, σ) → Bs+1/p (Ω)bdd is the boundary-
to-domain single layer from (4.3.25), and S is the boundary-to-
boundary single layer potential operator from (4.3.42),
and also when
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 491
 p,q M  p, p M
TrΩ→∂Ω : Fs+1/p (Ω)bdd → Bs (∂Ω, σ) is the boundary
trace operator from [69, (9.4.93) in item (ii) of Theorem 9.4.5]
(and further extended as indicated in Remark 3 following its state-
 p, p M  p,q M
ment), 𝒮 : Bs−1 (∂Ω, σ) → Fs+1/p (Ω)bdd is the boundary- (4.3.46)
to-domain single layer from (4.3.26), and S is the boundary-to-
 p, p M
boundary single layer potential operator mapping Bs−1 (∂Ω, σ)
 p, p M
the space into Bs (∂Ω, σ) as in (4.3.42) with q := p.

Finally, if p ∈ (1, ∞), q := p, s ∈ (0, 1), and a := 1 − s − p1 , the trace formula

(4.3.44) also holds when assuming
 1, p M  p, p M
TrΩ→∂Ω : Wa (Ω)bdd → Bs (∂Ω, σ) is the boundary
trace operator discussed in [69, Theorem 8.3.6] (and further ex-
tended as indicated in Remark 2 following the statement of the
 p, p M  1, p M
theorem), 𝒮 : Bs−1 (∂Ω, σ) → Wa (Ω)bdd is the boundary- (4.3.47)
to-domain single layer from (4.2.218), and S is the boundary-to-
 p, p M
boundary single layer potential operator mapping Bs−1 (∂Ω, σ)
 p, p M
into Bs (∂Ω, σ) as in (4.3.42) with q := p.

(4) If, in addition to the original hypotheses, Ω is also assumed to be a locally

uniform domain, then the single layer operator S from (4.3.39) induces a linear
and bounded mapping
 ∞,∞ M  M
S : Bs−1 (∂Ω, σ) −→ Bs∞,∞ (∂Ω, σ) for each s ∈ (0, 1). (4.3.48)

Furthermore, this extension satisfies

  ∞,∞ M
S f = (𝒮 f )∂Ω for each f ∈ Bs−1 (∂Ω, σ) where
 s M (4.3.49)
𝒮 f is regarded as a function locally in 𝒞 (Ω) (cf. (4.2.219)).

(5) If, in addition to the original hypotheses, Ω is also assumed to an (ε, δ)-domain
satisfying an exterior local John condition, then
the duality formula (4.3.41) remains valid if either 1 < p < ∞,
0 < q < ∞, 0 < s < 1, or p = 1, 0 < q ≤ 1, 0 < s < 1 (with (4.3.50)
the convention made in [69, (7.6.1)]).
Proof Let p, s, p∗ be as in (4.3.39) and define
−1  n−1 
q := 1−s
n−1 + 1
p ∈ n ,p . (4.3.51)

In particular,
1 
n−1
n < q < p ≤ 1 and s − 1 = −(n − 1) q − 1
p . (4.3.52)
492 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Granted this, [69, (7.8.153)] applies and yields

 M  p, p M
Ep H q (∂Ω, σ) = Bs−1 (∂Ω, σ) . (4.3.53)

Also, [69, (7.8.35)] guarantees that

 ∗ M  ∗ M
Ep L p (∂Ω, σ) = L p (∂Ω, σ) . (4.3.54)

From item (1) of Theorem 2.2.6 we know that the single layer operator S induces a
linear and bounded mapping in the context of
 M  ∗ M
S : H q (∂Ω, σ) −→ L p (∂Ω, σ) . (4.3.55)

In turn, from (4.3.53), (4.3.54), (4.3.55), and [69, (7.8.56) in Proposition 7.8.9] we
conclude that the single layer operator S from (4.3.55) extends (in a unique fashion)
to a linear and bounded mapping
 p, p M  ∗ M
S' : Bs−1 (∂Ω, σ) → L p (∂Ω, σ) . (4.3.56)

Via embeddings it may be seen that the operators S' corresponding to various values
of p, s, p∗ as in (4.3.39) act in a coherent fashion with one another, and they are also
compatible with S from Theorem 2.2.6. As such, we may drop the “hat”, and simply
refer to S' from (4.3.56) simply as S. This establishes the claims made in relation to
(4.3.39). With this in hand, the claims regarding (4.3.38) are then justified using the
real interpolation results from [69, (7.4.2)] and [68, (6.2.48)].
Moving on, make the additional assumption that Ω satisfies a two-sided local
John condition (cf. [68, Definition 5.11.7]; see also (A.0.104)). In particular, from
[68, (5.10.24), (5.11.26)] we conclude that Ω is actually a UR domain. Thanks to
this, the results in item (ix) of Theorem 1.5.1 imply that for each p ∈ (1, ∞) the single
layer potential operator S induces well-defined, linear, and bounded mappings
 M  p M
S : L p (∂Ω, σ) −→ L1 (∂Ω, σ) ,
 p M  M (4.3.57)
S : L−1 (∂Ω, σ) −→ L p (∂Ω, σ) ,

which act in a compatible fashion with one another. Granted this, the claims per-
taining to (4.3.40) in the current part (2) follow from our real interpolation results
obtained in [69, Theorem 11.12.2] and [69, (11.12.60) in Corollary 11.12.3]. Finally,
(4.3.41) is seen from what we have proved so far, item (ix) in Theorem 1.5.1, and
[69, (7.1.62)].
To deal with the claims made in part (3) of the theorem, make the additional
assumption that Ω is an (ε, δ)-domain and suppose
 
n < p < ∞, 0 < q ≤ ∞, (n − 1) p1 − 1 + < s < 1.
n−1
(4.3.58)
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 493

Then the result in item (1) of Theorem 4.3.3 gives that the boundary-to-domain single
layer potential operator 𝒮 induces a well-defined, linear, and continuous mapping
 p,q M % &M
p,q
𝒮 : Bs−1 (∂Ω, σ) −→ B 1 (Ω)bdd . (4.3.59)
s+ p

Also, from item (ii) of [69, Theorem 9.4.5] we know that the (vector-valued) bound-
ary trace operator
% &M  p,q M
p,q
TrΩ→∂Ω : B 1 (Ω)bdd −→ Bs (∂Ω, σ) (4.3.60)
s+ p

is well defined, linear, and continuous. Consequently, the composition of the opera-
tors in (4.3.59)-(4.3.60), i.e.,
 p,q M  p,q M
S := TrΩ→∂Ω ◦ 𝒮 : Bs−1 (∂Ω, σ) −→ Bs (∂Ω, σ) (4.3.61)

is also a well-defined, linear, and continuous operator. The goal now is to show that S
from (4.3.61) agrees with S when acting on “nice” functions, in the common domain.
To this end, fix a sufficiently large aperture parameter κ > 0 as in [69, Corol-
 M
lary 8.3.9]. Then for each function f ∈ Lip (∂Ω) we may rely on [69, Corol-
lary 8.3.9], (1.3.67), [68, (8.8.69)] (also bearing in mind [68, Lemma 5.11.9)] and
[68, (5.11.35)]) to write
   
S f = TrΩ→∂Ω ◦ 𝒮 f = TrΩ→∂Ω 𝒮 f

  κ−n.t.
= 𝒮f  = S f at σ-a.e. point on ∂Ω. (4.3.62)
∂Ω
 M
Hence, the operators S and S agree on Lip (∂Ω) . As a consequence of this and
  M  p,q M
the boundedness of (4.3.61), for each f ∈ Lip (∂Ω) ⊆ Bs−1 (∂Ω, σ) we have
 p,q M
S f = S f ∈ Bs (∂Ω, σ) and
 
S f [Bsp, q (∂Ω,σ)] M = 
S f [B p, q (∂Ω,σ)] M ≤ C f [B p, q (∂Ω,σ)] M , (4.3.63)
s s−1

 M
for some constant C ∈ (0, ∞) independent of the function f . Given that Lip (∂Ω)
 p,q M
is a dense subset of Bs (∂Ω, σ) (cf. [69, Lemma 7.1.10]), we ultimately con-
clude that the single layer operator S from (4.3.39) extends, in a unique fashion, to
a linear and bounded mapping in the context of (4.3.42).
Going further, the fact that the trace formulas (4.3.44)-(4.3.47) are valid is implicit
in the manner in which the single layer potential operator S has been defined in the
context of (4.3.42) (see (4.3.62) and recall the boundedness properties of 𝒮 from
(4.3.25), (4.3.26), (4.2.218), together with the boundedness properties TrΩ→∂Ω from
[69, (9.4.91) in item (ii) of Theorem 9.4.5] as well as [69, (9.4.93) in item (ii) of
Theorem 9.4.5], and [69, (8.3.38)]).
494 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Under the assumption that Ω is also a locally uniform domain, the claim made in
item (4) is proved in a similar manner, making use of (4.2.219) and [69, (7.1.59)],
now taking S to be the mapping
 M   M
∞,∞
Bs−1 (∂Ω, σ)  f −→ (𝒮 f )∂Ω ∈ Bs∞,∞ (∂Ω, σ) . (4.3.64)

Finally, consider the claim made in item (5), now assuming that Ω is also
(ε, δ)-domain satisfying an exterior local John condition. Together, [68, Proposi-
tion 5.11.14] and [68, (5.11.28)] then ensure that Ω is also a locally uniform domain
satisfying an interior local John condition. Granted these properties, the claim made
in (4.3.50) may be justified based on the continuity/compatibility result just estab-
lished, the duality formula (4.3.41), the density result from [69, Lemma 7.1.10], and
the compatibility of pairings from [69, Proposition 7.6.2]. Specifically, assume that
either 1 < p < ∞, 0 < q < ∞, 0 < s < 1, or p = 1, 0 < q ≤ 1, 0 < s < 1. Also, pick
s∗ ∈ (s, 1) and p∗ ∈ (1, p ). Granted these choices, [69, Proposition 7.7.2] gives that
p ,q p ,p
B−s (∂Ω, σ) ⊆ B−s∗ ∗ ∗ (∂Ω, σ). (4.3.65)
p,q
Also, having fixed an arbitrary f ∈ Bs−1 (∂Ω, σ), [69, Lemma 7.1.10] tells us that
there exists a sequence
 M  p,q M
{ f j } j ∈N ⊆ Lip(∂Ω) such that lim f j = f in Bs−1 (∂Ω, σ) . (4.3.66)
j→∞

In particular, (4.3.66) and the continuity of the operator (4.3.42) gives

p,q
lim S f j = S f in Bs (∂Ω, σ). (4.3.67)
j→∞

 p ,q M
From (4.3.42) and (4.3.48), it follows that SL  g ∈ B1−s (∂Ω, σ) for each g ∈
p ,q
B−s (∂Ω, σ) it follows . Bearing this in mind we may then write
 
[B s (∂Ω,σ)] M S f , g [B p , q (∂Ω,σ)] M
p, q
−s
 
= lim p, q
[B s (∂Ω,σ)] M S fj, g p ,q
[B−s (∂Ω,σ)] M
j→∞
 
= lim (p∗ ) ,(p∗ ) S fj, g p , q∗
[B−s∗∗ (∂Ω,σ)] M
j→∞ [B s∗ (∂Ω,σ)] M

 
= lim (p∗ ) ,(p∗ ) f j , SL  g p ,q
∗ ∗
[B1−s (∂Ω,σ)] M
j→∞ [B s∗ −1 (∂Ω,σ)] M ∗

 
= lim p, q
[B s−1 (∂Ω,σ)] M f j , SL  g p ,q
[B1−s (∂Ω,σ)] M
j→∞
 
= [B p, q (∂Ω,σ)] M f , SL  g [B p , q (∂Ω,σ)] M
. (4.3.68)
s−1 1−s
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 495

Above, the first equality is implied by (4.3.67) and the continuity of the duality
pairing (cf. item (i) in [69, Proposition 7.6.1]). The second equality in (4.3.68)
 (p∗ ) ,(p∗ ) M
uses (4.3.65), the fact that the membership of f j to Bs∗ −1 (∂Ω, σ) implies
 (p∗ ) ,(p∗ ) M
S f j ∈ Bs∗ (∂Ω, σ) (cf. (4.3.40)), and the compatibility of pairings from
[69, Proposition 7.6.2]. The third equality in (4.3.68) comes from (4.3.41). The
fourth equality in (4.3.68) is once again implied by the compatibility of pairings
from [69, Proposition 7.6.2]. The final equality in (4.3.68) is a result of (4.3.66).
This establishes (4.3.50), so the proof of Theorem 4.3.4 is complete. 

Our next theorem is concerned with the action of the conormal derivative of the
double layer potential operator on boundary Besov spaces.

Theorem 4.3.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set satisfying

a two-sided local John condition and whose boundary is compact and Ahlfors
regular. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. Next, for some fixed M ∈ N, consider a coefficient
αβ 
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Finally, fix

1 < p < ∞, 0 < q ≤ ∞, 0 < s < 1. (4.3.69)

Then the operator ∂νA D from (1.5.31) extends, in a unique fashion, to a bounded
linear mapping
 p,q M  p,q M
∂νA D : Bs (∂Ω, σ) −→ Bs−1 (∂Ω, σ) (4.3.70)

and various choices of p, q, s as in (4.3.69) yield operators which are compatible

with one another. Moreover, if s ∈ (0, 1) and the exponents p, q, p , q ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q then
 A 
[B s−1 (∂Ω,σ)] M ∂ν D f , g [B p , q (∂Ω,σ)] M
p, q
1−s
  
= [Bsp, q (∂Ω,σ)] M f , ∂νA D A g [B p , q (∂Ω,σ)] M (4.3.71)
−s
 p,q M  p ,q M
for each f ∈ Bs (∂Ω, σ) and g ∈ B1−s (∂Ω, σ) ,

where ∂νA D A is associated with A (the real transpose of A) in the same manner
∂νA D has been associated with the original coefficient tensor A.
Furthermore, under the additional assumption that Ω is an (ε, δ)-domain, one
has the following compatibility result
     
∂νA D f , 0 = ∂νA ψD f , L(ψD f ) = ∂νA D f
 p,q M
for all f ∈ Bs (∂Ω, σ) and each cutoff function (4.3.72)
ψ ∈ 𝒞∞
c (R ) with ψ ≡ 1 near ∂Ω,
n
496 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

in each of the following scenarios:

 p, p M
(a) Assume q := p and, for each f ∈ Bs (∂Ω, σ) , the conormal derivatives
in (4.3.72) are considered as in [69, Remark 8.5.4] and [69, Proposition 8.5.3]
 1, p M
(by viewing D f as a function in Wa (Ω)bdd with a := 1 − s − p1 ; cf.
 A 
Theorem 4.2.3), while ∂ν D f in the right-hand side of (4.3.72) is taken in the
sense of (4.3.70) with q := p.
 p,q M
(b) Assume q ∈ (1, ∞) and, for each f ∈ Bs (∂Ω, σ) , the conormal derivatives
in (4.3.72) are considered as in [69, Remark 9.5.3] and [69, Proposition 9.5.2]
 p,q M
with A := B (by viewing D f as a function in Bs+1/p (Ω)bdd ; cf. item (1) in
 
Theorem 4.3.1), while ∂νA D f in the right-hand side of (4.3.72) is taken in the
sense of (4.3.70).
 p, p M
(c) Assume q := p and, for each f ∈ Bs (∂Ω, σ) , the conormal derivatives in
(4.3.72) are considered as in [69, Remark 9.5.3] and [69, Proposition 9.5.2] with
 p,qo M
A := F (by viewing D f as a function in Fs+1/p (Ω)bdd with 1 < qo < ∞; cf.
 A 
item (2) in Theorem 4.3.1), while ∂ν D f in the right-hand side of (4.3.72) is
taken in the sense of (4.3.70) with q := p.

Finally, retaining the additional assumption that Ω is an (ε, δ)-domain, it follows

that, given any p ∈ (1, ∞) and s ∈ (0, 1), for each s∗ ∈ (s, 1) the operator
 p, p M  p, p M
∂νA D : Bs (∂Ω, σ) −→ Bs−1 (∂Ω, σ) (4.3.73)
 M  M
maps Bs∞,∞
∗
(∂Ω, σ) into Bs∞,∞
∗ −1
(∂Ω, σ) and the map induced in this context,
i.e.,  M  M
∂νA D : Bs∞,∞
∗
(∂Ω, σ) −→ Bs∞,∞
∗ −1
(∂Ω, σ) , (4.3.74)
is well defined, linear, and bounded.

Proof From [68, (5.2.4), (5.10.24), (5.11.26)] and [68, ] we see that Ω is actually a
UR domain. Having established this, the results in items (vii)-(viii) of Theorem 1.5.1
imply that for each p ∈ (1, ∞) the operator ∂νA D induces well-defined, linear, and
bounded mappings
 p M  M
∂νA D : L1 (∂Ω, σ) −→ L p (∂Ω, σ) ,
 M  p M (4.3.75)
∂νA D : L p (∂Ω, σ) −→ L−1 (∂Ω, σ) ,

which are compatible with each other. Granted this, all claims pertaining to (4.3.70)
follow from the real interpolation results from [69, Theorem 11.12.2] and [69,
(11.12.60) in Corollary 11.12.3]. Furthermore, (4.3.71) is seen from what we have
proved so far, item (viii) in Theorem 1.5.1, and [69, (7.1.62)].
To deal with the claims in the second part of the statement of the theorem, make
the additional assumption that Ω is an (ε, δ)-domain. Consider first the situation
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 497
 p M
when f ∈ L1 (∂Ω, σ) , and fix an aperture parameter κ > 0. From items (i)-(ii)
in Theorem 1.5.1 we know that
κ−n.t.   κ−n.t.
D f ∂Ω , ∇D f ∂Ω exist at σ-a.e. point on ∂Ω, and
(4.3.76)
Nκ (D f ), Nκ (∇D f ) belong to L p (∂Ω, σ).

Hence, if we now introduce u = (uβ )1≤β ≤M := D f in Ω, it is then meaningful to

consider ∂νAu in the sense of (A.0.184). In fact, according to (A.0.184) and item (vii)
in Theorem 1.5.1 (cf. (1.5.29) and (1.5.30)) we have
 
∂νAu = ∂νA D f . (4.3.77)
 M
To proceed, let us now pick an arbitrary function ϕ ∈ Lip (∂Ω) . It is a classical
 M
fact that this may be extended to some Φ = (Φα )1≤α ≤M ∈ Lipc (Rn ) . Also, pick
a cutoff function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near ∂Ω. Observe that
n

  αβ
A∇(ψu), ∇Φ = a jk ∂k (ψuβ )∂j Φα belongs to L 1 (Ω, L n ), (4.3.78)

thanks to (4.3.76) and [68, (8.6.51)]. Then, with p ∈ (1, ∞) denoting the Hölder
conjugate exponent of p, we may write
∫
 A  
∂ν D f , ϕ dσ
∂Ω
∫ ∫
 κ−n.t.   κ−n.t. 
= ∂νAu, Φ∂Ω dσ = ∂νA(ψu), Φ∂Ω dσ
∂Ω ∂Ω
∫ ∫
αβ  
= a jk ∂k (ψuβ )∂j Φα dL n + L(ψu), Φ dL n
Ω Ω
     
= ([B p , p (∂Ω,σ)] M )∗
∂νA ψu, L(ψu) , TrΩ→∂Ω ΦΩ [B p , p (∂Ω,σ)] M
1−s 1−s
   
= ([B p , p (∂Ω,σ)] M )∗
∂νA ψD f , L(ψD f ) , ϕ [B p , p (∂Ω,σ)] M
. (4.3.79)
1−s 1−s

κ−n.t. 
The first equality above comes from (4.3.77) and the fact that Φ∂Ω = Φ∂Ω = ϕ
at every point on Aκ (∂Ω) (cf. [68, (8.9.10)]), hence at σ-a.e. point on ∂Ω (cf. [68,
Corollary 8.8.9]). The second equality in (4.3.79) is implied by the Green type
formula [70, (1.7.121)], whose present applicability is ensured by (4.3.78) and the
 M
fact that L A(ψu) ∈ 𝒞∞ c (Ω) (cf. (1.3.24)). The third equality uses the generalized
Green’s formula from [69, (8.5.18)]. To see that the latter is applicable in the present
p
context requires two observations. First, since the space L1 (∂Ω, σ) is contained in
p, p  1, p M
Bs (∂Ω, σ) (cf. [69, (11.11.12)]), Theorem 4.2.3 guarantees that ψu ∈ Wa (Ω)
  1, p M
with a := 1 − s − p1 . Second, the function ΦΩ does belong to W−a (Ω) , as may
498 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

be seen from [69, (8.3.36)]. The final equality in (4.3.79) is a consequence of the
definition of u and [69, (8.3.39)].
In turn, from (4.3.79) and [69, Lemma 7.1.10] we ultimately conclude that if
 p M    M
f ∈ L1 (∂Ω, σ) then ∂νA D f , originally viewed as a function in L p (∂Ω, σ)
 p ,p M ∗
(cf. (1.5.31)), induces a linear and continuous functional in B1−s (∂Ω, σ)
 
which actually coincides  with  the conormal derivative
 ∂νA ψD  f , L(ψD f ) . Since
both assignments f → ∂νA D f and f → ∂νA ψD f , L(ψD f ) are continuous from
 p, p M  p, p M
Bs (∂Ω, σ) into Bs−1 (∂Ω, σ) (as evident from (4.3.70), and [69, Proposi-
tion 8.5.3] in combination with Theorem 4.2.3), we finally conclude, on account of
 p M  p, p M
the density of L1 (∂Ω, σ) into Bs (∂Ω, σ) (cf. [69, Proposition 11.11.3]),
that the identification claimed in (4.3.72) holds in the scenario described in item (a)
of the theorem.
That (4.3.72) is also valid in the scenarios described in items (b)-(c) of the theorem
is then proved in a similar fashion, making use of [69, Proposition 9.5.2] in place of
[69, Proposition 8.5.3], and of Theorem 4.3.1 in place of Theorem 4.2.3.
To deal with the final claim in the statement of the theorem, retain the additional
assumption that Ω is an (ε, δ)-domain. In view of the fact that the conormal derivative
of the double layer ∂νA D does not jump across the boundary (cf. as seen from (1.5.29)
and the fact that ∂νA D from (4.3.70) is the unique extension of (1.5.31)), there is
no loss of generality in assuming that Ω is also bounded (cf. [68, Lemma 5.10.10]).
To proceed, fix p ∈ (1, ∞), s ∈ (0, 1), and s∗ ∈ (s, 1). From (4.3.70) we know that
the operator (4.3.73) is well defined, linear, and bounded. Also, item (i) in [69,
 M  p, p M
Proposition 7.7.2] implies that Bs∞,∞ (∂Ω, σ) is a subspace of Bs (∂Ω, σ) .
 M ∗

Given any f ∈ Bs∞,∞ ∗

(∂Ω, σ) set u := D f and use (4.3.72) and [69, (8.5.32)] to
write
 M
(∂νA D) f = ∂νA(D f , 0) = ∂νA(u, 0) = ∂νAu ∈ Bs∞,∞ ∗ −1
(∂Ω, σ) , (4.3.80)

where the conormal derivative ∂νAu is defined as in [69, (8.5.29)-(8.5.30)]. This

 M  M
proves that the operator (4.3.73) maps Bs∞,∞ ∗
(∂Ω, σ) into Bs∞,∞∗ −1
(∂Ω, σ) .
Finally, from (4.3.80), [69, (8.5.31)], (1.8.14), (1.8.9), and [69, (7.1.59)] we deduce
that
 A     
(∂ D) f  ∞,∞  1−s∗  
ν [B (∂Ω,σ)] M
= ∂νAu[B∞,∞ (∂Ω,σ)] M ≤ C δ∂Ω ∇u  ∞
s∗ −1 s∗ −1 n L (Ω, L )
 
 1−s∗ 
= C δ∂Ω ∇(D f ) ≤ C f [𝒞. s∗ (∂Ω)] M
L ∞ (Ω, L n )

≤ C f [B∞,∞
s∗ (∂Ω)]
M. (4.3.81)

This establishes the fact that the operator (4.3.73) induces a well defined, linear, and
bounded mapping in the context of (4.3.74). 
4.3 Boundary-to-Boundary Layer Potentials from Besov into Besov/Triebel-Lizorkin Spaces 499

The conormal derivative of the boundary-to-domain single layer potential operator

acting on Besov spaces satisfies jump-formulas of the sort described in the theorem
below.
Theorem 4.3.6 Let Ω ⊆ Rn be an NTA domain with a compact Ahlfors regular
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
 αβ  unit normal to Ω. Next, for some M ∈ N, consider a coefficient
theoretic outward
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]). Recall the boundary-
to-domain single layer potential operator 𝒮 (associated with L and Ω) acting in the
context of Besov spaces as in Proposition 4.2.8, and the boundary-to-boundary
“transpose” double layer potential operator K A#  (associated with A and Ω) acting
in the context of Besov spaces as in Theorem 4.1.5. Finally, fix

1 < p < ∞, 1 < q < ∞, 0 < s < 1. (4.3.82)

Then the jump-formula

     
∂νA 𝒮 f , 0 = ∂νA ψ𝒮 f , φ = − 12 I + K A#  f
 p,q M
for all f ∈ Bs−1 (∂Ω, σ) and all ψ ∈ 𝒞∞ c (R )
n (4.3.83)
 M
with ψ ≡ 1 near ∂Ω, where φ := L(ψ𝒮 f ) ∈ 𝒞∞ c (Ω) ,

(with I denoting the identity operator) holds in each of the following scenarios:
 p, p M
(a) Assume q := p, for each f ∈ Bs−1 (∂Ω, σ) consider the conormal derivatives
in (4.3.83) as in [69, Remark 8.5.4] and [69, Proposition 8.5.3] (by regarding
 1, p M
𝒮 f as a function in Wa (Ω)bdd with a := 1 − s − p1 ; cf. Theorem 4.2.10),
and interpret K A#  as in (4.1.27) (with A in place of A, with q := p, and with s
replaced by 1 − s).
 p,q M
(b) For each f ∈ Bs−1 (∂Ω, σ) , the conormal derivatives in (4.3.83) are con-
sidered as in [69, Remark 9.5.3] and [69, Proposition 9.5.2] with A := B (by
 p,q M
viewing 𝒮 f in Bs+1/p (Ω)bdd ; cf. item (1) in Theorem 4.3.3), while K A#  in
(4.3.83) is understood as in (4.1.27) (with A in place of A, and with s replaced
by 1 − s).
 p, p M
(c) Assume q := p and, for each f ∈ Bs−1 (∂Ω, σ) , the conormal derivatives
in (4.3.83) are considered in the sense of [69, Remark 9.5.3] and [69, Proposi-
 p,qo M
tion 9.5.2] with A := F (by viewing 𝒮 f as a function in Fs+1/p (Ω)bdd with
1 < qo < ∞; cf. item (2) in Theorem 4.3.3), while K A#  in (4.3.83) is understood
as in (4.1.27) (with A in place of A, with q := p, and with s replaced by 1 − s).
Proof Let us establish the jump-formula (4.3.83) in the scenario described in item
(a) in the statement of the theorem. For starters, observe that [68, (5.10.24)] implies
500 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

that Ω is actually a UR domain. Granted this, Corollary 4.1.4 applies and gives that
 p, p M  p, p M
K A#  : Bs−1 (∂Ω, σ) −→ Bs−1 (∂Ω, σ) boundedly. (4.3.84)

In addition, having fixed an arbitrary cutoff function ψ ∈ 𝒞∞

c (R ) with ψ ≡ 1 near
n

∂Ω, from Theorem 4.2.10 and [69, Proposition 8.5.3] we also see that
 p, p M    p, p M
Bs−1 (∂Ω, σ)  f −→ ∂νA ψ𝒮 f , φ f ∈ Bs−1 (∂Ω, σ)
 M (4.3.85)
where φ f := L(ψ𝒮 f ) ∈ 𝒞∞ c (Ω) ,

is a well-defined, linear, and bounded operator. On account of (4.3.84)-(4.3.85),

the jump-formula claimed in (4.3.83) follows as soon as we show that the op-
 p, p M
erators in (4.3.84)-(4.3.85) agree on a dense subset of Bs−1 (∂Ω, σ) . By [69,
Lemma 7.1.10], it therefore suffices to prove that
     M
∂νA ψ𝒮 f , φ f = − 12 I + K A#  f for each f ∈ Lip (∂Ω) . (4.3.86)
 M
With this goal in mind, fix an arbitrary f ∈ Lip (∂Ω) and set
 M
u = (uβ )1≤β ≤M := ψ𝒮 f ∈ 𝒞∞ (Ω) . (4.3.87)

Also, fix an aperture parameter κ > 0. From (4.3.87) and item (x) in Theorem 1.5.1
we see that
κ−n.t. κ−n.t.
u∂Ω , (∇u)∂Ω exist at σ-a.e. point on ∂Ω,
 M (4.3.88)
Lu = φ f ∈ 𝒞∞ c (Ω) and Nκ (∇u) ∈ L p (∂Ω, σ).

It is then meaningful to consider ∂νAu in the pointwise sense of (A.0.184). In fact,

according to item (xi) in Theorem 1.5.1 we have
 
∂νAu = − 12 I + K A#  f . (4.3.89)
 M
Going further, let us now pick an arbitrary function ϕ ∈ Lip (∂Ω) . As is well
 M 
known, there exists some Φ = (Φα )1≤α ≤M ∈ Lipc (Rn ) such that Φ∂Ω = ϕ.
Thanks to (4.3.88) and [68, (8.6.51)] we have
  αβ
A∇u, ∇Φ = a jk (∂k uβ )(∂j Φα ) belongs to L 1 (Ω, L n ), (4.3.90)

Let p ∈ (1, ∞) denote the Hölder conjugate exponent of p. Then, with justifications
to be provided momentarily, we may compute
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 501
  
p ,p
([B1−s (∂Ω,σ)] M )∗
− 12 I + K A#  f , ϕ p ,p
[B1−s (∂Ω,σ)] M
.
∫ ∫
    κ−n.t. 
= − 1
2I + K A#  f , ϕ dσ = ∂νAu, Φ∂Ω dσ
∂Ω ∂Ω
∫ ∫
  αβ
= φ f , Φ dL n + a jk ∂k uβ ∂j Φα dL n
Ω Ω
   
= ([B p , p (∂Ω,σ)] M )∗
∂νA(u, φ f ), TrΩ→∂Ω ΦΩ [B p , p (∂Ω,σ)] M
1−s 1−s
 
= ([B p , p (∂Ω,σ)] M )∗
∂νA(ψ𝒮 f , φ f ), ϕ p ,p
[B1−s (∂Ω,σ)] M
. (4.3.91)
1−s

κ−n.t. 
Above, the first equality comes from (4.3.89) and the fact that Φ∂Ω = Φ∂Ω = ϕ
at every point on Aκ (∂Ω) (cf. [68, (8.9.10)]), hence at σ-a.e. point on ∂Ω (cf.
[68, Corollary 8.8.9]). In view of the fact that Lu = φ f in Ω (cf. (4.3.88)), the
second equality in (4.3.91) is implied by the Green type formula [70, (1.7.121)],
whose present applicability is ensured by (4.3.90). The third equality in (4.3.91)
uses the generalized Green’s formula from [69, (8.5.18)], which is applicable in
p, p
the present context. Indeed, since Lip (∂Ω) is contained in Bs (∂Ω, σ) (cf. [69,
 1, p M
(7.1.62)]), Theorem 4.2.10 ensures that u ∈ Wa (Ω) with a := 1 − s − p1 . Also,
  1, p M
[69, (8.3.36)] implies that ΦΩ ∈ W−a (Ω) . The last equality in (4.3.91) is a
consequence of the definition of u and [69, (8.3.39)]. Thus, (4.3.91) is fully justified.
At this stage, from the resulting identity in (4.3.91) and [69, Lemma 7.1.10] we
   M
may conclude that − 12 I + K A#  f , originally viewed as a function in L p (∂Ω, σ)
 p ,p M ∗
(cf. (1.5.12)), induces a linear and continuous functional in B1−s (∂Ω, σ)
which actually coincides with the conormal derivative ∂νA(ψ𝒮 f , φ f ). This establishes
(4.3.86) which, together with [69, (8.5.23)], proves the jump-formula (4.3.83) in the
scenario described in item (a).
Finally, that (4.3.83) is also valid in the scenarios described in items (b)-(c) of the
theorem may be justified in a similar manner, now relying on [69, Remark 9.5.3] in
place of [69, Remark 8.5.4], and of Theorem 4.3.3 in place of Theorem 4.2.10. 

4.4 Integral Representation Formulas of Layer Potential Type,

and Consequences

The first main result in this section is a basic integral representation formula of a null-
solution of a weakly elliptic system, which allows us to recover such a function from
the action of the (boundary-to-domain) double layer operator on the boundary trace
of said function and the action of the (boundary-to-domain) single layer operator on
the conormal derivative of the original function.
502 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Theorem 4.4.1 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an (ε, δ)-domain with

a compact Ahlfors regular boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν
 αβ unit normal to Ω. Next, for some M ∈ N,
the geometric measure theoretic outward
consider a coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the
1≤α,β ≤M
property that the M × M homogeneous second-order system L = L A associated with
A in Rn as in (1.3.2) is weakly elliptic (in the sense of [70, (1.3.3) in Definition 1.3.1]).
 M
Finally, let u ∈ 𝒞∞ (Ω) be a null-solution of the system L in Ω which, in the case
when Ω is an exterior domain, satisfies
⨏
|u| dL 2 = o(1) as R → ∞, (4.4.1)
B(0,λ R)\B(0,R)

for some λ ∈ (1, ∞).

Then the integral representation formula
   
u = D TrΩ→∂Ω u − 𝒮 ∂νA(u, 0) in Ω (4.4.2)

holds in any of the following scenarios:

(1) One assumes that
 p,q M
u belongs to the Besov space B 1 (Ω)bdd with
s+ p
(4.4.3)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,

 p,q M  p,q M
TrΩ→∂Ω : B 1 (Ω)bdd → Bs (∂Ω, σ) is the boundary
s+ p
trace operator from [69, (9.4.105)] (thus, TrΩ→∂Ω u lies in the (4.4.4)
 p,q M
space Bs (∂Ω, σ) ),

 p,q M  p,q M
D : Bs (∂Ω, σ) → B 1 (Ω)bdd is the boundary-to-domain
ps+
double layer potential operator associated with A and Ω, which is well (4.4.5)
defined in such a context by item (1) in Theorem 4.3.1,

∂νA is the conormal derivative operator associated with A and Ω as in

 p,q M (4.4.6)
[69, Remark 9.5.3] with A := B (so ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ) ),

and
 p,q M  p,q M
𝒮 : Bs−1 (∂Ω, σ) → B 1 (Ω)bdd is the boundary-to-domain
s+ p (4.4.7)
single layer operator (associated with A and Ω) as in (4.3.25).
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 503

(2) One assumes that

 p,q M
u belongs to the Triebel-Lizorkin space F 1 (Ω)bdd with
s+ p
(4.4.8)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,

 p,q M  p, p M
TrΩ→∂Ω : F 1 (Ω)bdd → Bs (∂Ω, σ) is the boundary trace
s+ p
 p, p M (4.4.9)
operator from [69, (9.4.106)] (hence TrΩ→∂Ω u ∈ Bs (∂Ω, σ) ),

 p, p M  p,q M
D : Bs (∂Ω, σ) → F 1 (Ω)bdd is the boundary-to-domain
s+ p
double layer potential operator associated with A and Ω, which is well
defined in such a context by item (2) in Theorem 4.3.1,
(4.4.10)

∂νA is the conormal derivative operator associated with A and Ω as in

 p, p M
[69, Remark 9.5.3] with A := F (so ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ) ),
(4.4.11)
and
 p, p M  p,q M
𝒮 : Bs−1 (∂Ω, σ) → F 1 (Ω)bdd is the boundary-to-domain
s+ p
single layer operator (associated with A and Ω) as in (4.3.26).
(4.4.12)
(3) One also assumes that Rn \ Ω is n-thick, and that
 1, p M
u belongs to the weighted Sobolev space Wa (Ω)bdd , with
(4.4.13)
1 < p < ∞, 0 < s < 1, a := 1 − s − p1 ,

 1, p M  p, p M
TrΩ→∂Ω : Wa (Ω)bdd → Bs (∂Ω, σ) is the trace from
 p, p M
[69, Theorem 8.3.6] (so that TrΩ→∂Ω u ∈ Bs (∂Ω, σ) by [69, (4.4.14)
(8.3.38)]),

 p, p M  1, p M
D : Bs (∂Ω, σ) → Wa (Ω)bdd is the boundary-to-domain
double layer potential operator associated with A and Ω, which is well
defined in the present context thanks to Theorem 4.2.3,
(4.4.15)
504 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

∂νA is the conormal derivative operator associated with A and Ω as in

 p, p M
[69, Remark 8.5.4] (so, ∂νA(u, 0) ends up in Bs−1 (∂Ω, σ) ),
(4.4.16)
and
 p, p M  1, p M
𝒮 : Bs−1 (∂Ω, σ) → Wa (Ω)bdd is the boundary-to-domain
single layer operator (associated with A and Ω) as in (4.2.218).
(4.4.17)

Furthermore, if Ω is an exterior domain and, in place of (4.4.1), one now only

assumes ⨏
|u| dL 2 = o(R) as R → ∞, (4.4.18)
B(0,λ R)\B(0,R)

for some λ ∈ (1, ∞), then in place of the integral representation formula (4.4.2) one
now concludes that there exists c ∈ C M such that
   
u = D TrΩ→∂Ω u − 𝒮 ∂νA(u, 0) + c in Ω, (4.4.19)

in any of the scenarios described in items (1)-(3) above.

Proof Consider the scenario specified in item (1). Fix a point x ∈ Ω and choose a
scalar-valued function θ ∈ 𝒞∞ (Rn ) with the property that θ = 0 on B(0, 1) and θ = 1
on Rn \ B(0, 2). For each

ε ∈ 0, min 1, 12 dist(x, ∂Ω) (4.4.20)

define θ ε : Rn → R by setting
y−x
θ ε (y) := θ for every y ∈ Rn . (4.4.21)
ε
Then θ ε ∈ 𝒞∞ (Rn ) is a bounded function satisfying

lim θ ε (y) = 1 for every y ∈ Rn \ {x}, (4.4.22)

ε→0+

and there exists a constant C ∈ (0, ∞) such that for each ε as in (4.4.20) we have

1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε),
supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε), and (4.4.23)
|(∇ j θ ε )(y)| ≤ Cε −j for every j ∈ N0 and every y ∈ Rn .

Also, pick a scalar-valued function ψ ∈ 𝒞∞c (R ) such that ψ = 1 on B(0, 1), and for
n

each R > 0 define ψR := ψ(·/R). Henceforth, fix R ∈ (0, ∞) sufficiently large, as to

have
∂Ω ⊆ B(0, R) as well as B(x, 1) ⊆ B(0, R). (4.4.24)
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 505

To proceed, bring in the matrix-valued fundamental solution E = (Eγβ )1≤γ,β ≤M

associated with the system L as in [70, Theorem 1.4.2]. Fix γ ∈ {1, . . . , M } and
observe that Eγβ (x − ·)θ ε ∈ 𝒞∞ (Ω) for each β ∈ {1, . . . , M }. Hence,
   M
w := Eγβ (x − ·)θ ε ψR 1≤β ≤M ∈ 𝒞∞ (Ω)
(4.4.25)
and w vanishes outside of a bounded subset of Ω.
% &M
p ,q
As a consequence, w ∈ B (Ω) where p , q are the Hölder conjugate
1−s+ p1
exponents of p, q. Since the assumptions on p, q, s from (4.4.3) ensure that
p,q
Bs+1/p (Ω) → L p (Ω, L n ) → L 1 (Ω, L n ), (4.4.26)
% ∗& M
p,q
(recall that Ω has finite measure), we my also define g ∈ Bs+1/p (Ω) by setting
∫
 
p, q
[(B s+1/p (Ω))∗ ] M g, G p, q
[B s+1/p (Ω)] M := (L  w)α Gα dL n
Ω
% &M (4.4.27)
p,q
for each G = (Gα )1≤α ≤M ∈ B 1 (Ω) .
s+ p

 M
In particular, the function L  w ∈ 𝒞∞ (Ω) satisfies
 M
L  w = gΩ in D (Ω) . (4.4.28)

We now invoke the version of the generalized “full” Green’s formula [69, (9.5.20)] for
Besov spaces, applied with u the function from the statement of the theorem (which
% ∗& M
p ,q
is currently assumed to satisfy (4.4.3)), with f := 0 ∈ B1−s+1/p (Ω) , and with
w and g as above. Specifically, having selected some cutoff function ξ ∈ 𝒞∞ c (R )
n

with ξ ≡ 1 near both ∂Ω and the support of w (playing the role of ψ in [69, (9.5.20)]),
we obtain
 A 
([B
p ,q ∂ (u, 0), TrΩ→∂Ω w [B p , q (∂Ω,σ)] M
(∂Ω,σ)] M )∗ ν
1−s 1−s

  
− ([Bsp, q (∂Ω,σ)] M )∗ ∂νA (w, g), TrΩ→∂Ω u p, q
[B s (∂Ω,σ)] M
 
= −[(B p, q (Ω))∗ ] M g, ξu [B p, q (Ω)] M
s+1/p s+1/p

∫ ∫

=− (L w)α ξuα dL = − n
(L  w)α uα dL n, (4.4.29)
Ω Ω

where the last equality comes from (4.4.27). Note that thanks to (4.4.25) and Propo-
sition 4.2.8 we have
506 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
   
p ,q
([B1−s (∂Ω,σ)] M )∗
∂νA(u, 0), TrΩ→∂Ω w p ,q
[B1−s (∂Ω,σ)] M
= 𝒮 ∂νA(u, 0) (x).
γ
(4.4.30)

Also, if (ν1, . . . , νn ) are the components of ν then

 A 
([B s (∂Ω,σ)] M )∗ ∂ν (w, g), TrΩ→∂Ω u [B s (∂Ω,σ)] M
p, q p, q

∫
  
ν j ak j (∂k wβ )∂Ω TrΩ→∂Ω u α dσ
βα
=
∂∗ Ω
∫
βα  
=− ν j ak j (∂k Eγβ )(x − ·) TrΩ→∂Ω u α dσ
∂∗ Ω

 
= D TrΩ→∂Ω u (x), (4.4.31)
γ

by [69, Proposition 9.5.4], (4.4.25), and (1.3.18). Let us take a second look at the
last integral in (4.4.29). For starters, use the product rule and support considerations
(cf. (4.4.24) and (4.4.20)) to expand
βα
−(L  w)α uα = −ak j (∂k ∂j wβ )uα
βα βα
= ak j (∂j Eγβ )(x − ·)(∂k θ ε )uα + ak j (∂j Eγβ )(x − ·)(∂k ψR )uα
βα βα
+ ak j (∂k Eγβ )(x − ·)(∂j θ ε )uα + ak j (∂k Eγβ )(x − ·)(∂j ψR )uα
βα βα
− ak j Eγβ (x − ·)(∂j ∂k θ ε )uα − ak j Eγβ (x − ·)(∂j ∂k ψR )uα
βα
− ak j (∂j ∂k Eγβ )(x − ·)θ ε ψR uα, (4.4.32)

and observe that the last term above vanishes in Ω\{x} since for each α ∈ {1, . . . , M }
we have
βα
ak j (∂j ∂k Eγβ )(x − ·) = 0 in Rn \ {x}, (4.4.33)

by [70, (1.4.33)]. Thus,

∫
− (L  w)α uα dL n = Iε + Iε + IIε + IIε + IIIε + IIIε
Ω
+ IVR + VR + VIR, (4.4.34)

where
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 507
∫
βα  
Iε := ak j (∂j Eγβ )(x − y)(∂k θ ε )(y) uα (y) − uα (x) dy,
Ω
∫
βα  
IIε := ak j (∂k Eγβ )(x − y)(∂j θ ε )(y) uα (y) − uα (x) dy, (4.4.35)
Ω
∫
βα  
IIIε := − ak j Eγβ (x − y)(∂j ∂k θ ε )(y) uα (y) − uα (x) dy,
Ω

then
∫ 
βα
Iε := ak j (∂j Eγβ )(x − y)(∂k θ ε )(y) dy uα (x),
Ω
∫ 
βα
IIε := ak j (∂k Eγβ )(x − y)(∂j θ ε )(y) dy uα (x), (4.4.36)
Ω
∫ 
βα
IIIε := − ak j Eγβ (x − y)(∂j ∂k θ ε )(y) dy uα (x),
Ω

and, finally,
∫
βα
IVR := ak j (∂j Eγβ )(x − y)(∂k ψR )(y)uα (y) dy,
Ω
∫
βα
VR := ak j (∂k Eγβ )(x − y)(∂j ψR )(y)uα (y) dy, (4.4.37)
Ω
∫
βα
VIR := − ak j Eγβ (x − y)(∂j ∂k ψR )(y)uα (y) dy.
Ω

In relation to Iε we note that since u is continuous at x we may estimate (using [70,

(1.4.24)] and [70, (1.5.25)])
⨏
   
lim sup Iε  ≤ C lim sup uα (y) − uα (x) dy = 0. (4.4.38)
ε→0+ ε→0+ B(x,2ε)

As regards the expression appearing in the curly brackets in Iε , by relying on (4.4.23)

and [70, (1.4.33)] we may conclude that for each fixed index α ∈ {1, . . . , M } we
have
508 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
∫
βα
lim+ ak j (∂j Eγβ )(x − y)(∂k θ ε )(y) dy
ε→0 Ω
∫
βα
= lim+ ak j (∂j Eγβ )(x − y)∂k (θ ε − 1)(y) dy
ε→0 Ω
 
βα  
= − lim+ D (Ω) ak j ∂j Eγβ (x − ·) , ∂k (θ ε − 1) D(Ω)
ε→0
   
βα
= lim+ D (Ω) ak j ∂k ∂j Eγβ (x − ·) , θ ε − 1 D(Ω)
ε→0
 
= δαγ lim+ D (Ω) δx, θ ε − 1 D(Ω)
ε→0

= δαγ lim+ (θ ε − 1)(x) = −δαγ . (4.4.39)

ε→0

Consequently,

lim Iε = −uγ (x). (4.4.40)

ε→0+

Likewise,
 
lim sup IIε  = 0 and lim+ IIε = −uγ (x). (4.4.41)
ε→0+ ε→0

In a very similar fashion,

lim IIIε = uγ (x) (4.4.42)

ε→0+

and, if n ≥ 3,
⨏
   
lim sup IIIε  ≤ C lim sup uα (y) − uα (x) dy = 0. (4.4.43)
ε→0+ ε→0+ B(x,2ε)
  
When n = 2 we only have |E(x − y)| ≤ C0 1 +  ln |x − y|  for y ∈ Rn \ {x} (cf.
[70, (1.4.24)]), but u is of class 𝒞∞ in Ω (being a null-solution of the weakly elliptic
system L; cf. [68, Theorem 6.5.7]), hence locally Lipschitz, so we may employ the
Mean Value Theorem to estimate
⨏
      
  
lim sup IIIε ≤ C lim sup 1 + ln ε  uα (y) − uα (x) dy
ε→0+ ε→0+ B(x,2ε)
  
≤ C lim sup 1 +  ln ε  ε = 0. (4.4.44)
ε→0+

Finally, from definitions, (4.4.1), [70, (1.5.205), (1.5.213)], and [70, Theorem 1.4.2]
we see that

lim IVR = 0, lim VR = 0, lim VIR = 0. (4.4.45)

R→∞ R→∞ R→∞
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 509

Collectively, (4.4.32)-(4.4.45) prove that

∫
− (L  w)α uα dL n = −uγ (x). (4.4.46)
Ω

At this stage, from (4.4.29), (4.4.30), (4.4.31), and (4.4.46) we conclude that
   
uγ (x) = D TrΩ→∂Ω u (x) − 𝒮 ∂νA(u, 0) (x). (4.4.47)
γ γ

In view of the arbitrariness of γ ∈ {1, . . . , M } and x ∈ Ω, this establishes the integral

representation formula (4.4.2) in the scenario described in item (1) of the theorem.
Going further, that the integral representation formula (4.4.2) also holds in the
scenarios described in items (2)-(3) in the statement of the theorem is established in
a completely analogous fashion (for the situation in item (3), the generalized “full”
Green formula from [69, (8.5.26)] is now used, and [69, Proposition 8.5.6] is now
employed in place of [69, Proposition 9.5.4]).
There remains to justify formula (4.4.19), working now under the assumption that
Ω is an exterior domains and, in place of (4.4.1), we only have (4.4.18). To this end,
fix a point x0 ∈ Rn \ Ω and, in lieu of (4.4.25), now consider
   M
w := Eγβ (x − ·) − Eγβ (x0 − ·) θ ε ψR ∈ 𝒞∞ (Ω) . (4.4.48)
1≤β ≤M

The idea is to run the same argument as above for this choice of w. The manner
in which the above alteration manifests itself in this process is via the following
estimates (themselves consequences of [70, Theorem 1.4.2] and the Mean Value
Theorem)  
|∇ψR | (∇E)(x − ·) − (∇E)(x0 − ·) ≤ CR−n−1 and
  (4.4.49)
|∇2 ψR | E(x − ·) − E(x0 − ·) ≤ CR−n−1,
which are valid for some C = C(L, x, x0 ) ∈ (0, ∞) independent of R. Together with
(4.4.18), they permit us to justify (the corresponding version of) (4.4.45) in the
present setting. Much as before, we arrive at the conclusion that (4.4.19) holds in
any of the scenarios described in items (1)-(3) with c := (cγ )1≤γ ≤M ∈ C M given by
∫
βα  
cγ := νs (y)ar s (∂r Eγβ )(x0 − y) TrΩ→∂Ω u α (y) dσ(y)
∂Ω
  
+ ([B p , q (∂Ω,σ)] M )∗
∂νA(u, 0), Eγ .(x0 − ·)∂Ω [B p , q (∂Ω,σ)] M
(4.4.50)
1−s 1−s

for each γ ∈ {1, . . . , M }. The proof of Theorem 4.4.1 is therefore complete. 

The next goal is to establish operator identities involving the double and single
layer potentials associated with a weakly elliptic system, similar in format to those
from item (xiii) in Theorem 1.5.1, when considering said boundary layer potentials
on Besov spaces.
510 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Theorem 4.4.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set satisfying a two-

sided local John condition and whose boundary is a compact Ahlfors regular set.
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric
 αβmeasure
 theoretic outward
unit normal to Ω. Next, for some M ∈ N, let A = ar s 1≤r,s ≤n be a coefficient
1≤α,β ≤M
tensor with complex entries, with the property that the M × M homogeneous second-
order system L = L A associated with A in Rn as in (1.3.2) is weakly elliptic (in the
sense of [70, (1.3.3) in Definition 1.3.1]). Consider the boundary layer potentials
K, K #, S, ∂νA D canonically associated with A and Ω. Finally, let K A#  be associated
with A (the real transpose of A) in the same manner K # has been associated with
the original coefficient tensor A. Then the following operator identities hold:
1       p,q M
2I + K ◦ − 12 I + K = S ◦ ∂νA D on Bs (∂Ω, σ)
(4.4.51)
with p ∈ (1, ∞), q ∈ (0, ∞], and s ∈ (0, 1),

1       p,q M
2I + K A#  ◦ − 12 I + K A#  = ∂νA D ◦ S on Bs−1 (∂Ω, σ)
(4.4.52)
with p ∈ (1, ∞), q ∈ (0, ∞], and s ∈ (0, 1),

 p,q M
S ◦ K A#  = K ◦ S on Bs−1 (∂Ω, σ)
(4.4.53)
with p ∈ (1, ∞), q ∈ (0, ∞], and s ∈ (0, 1),

     p,q M
K A#  ◦ ∂νA D = ∂νA D ◦ K on Bs (∂Ω, σ)
(4.4.54)
with p ∈ (1, ∞), q ∈ (0, ∞], and s ∈ (0, 1).

Moreover,
under the additional assumption that Ω is a uniform domain, the oper-
(4.4.55)
ator identities in (4.4.51)-(4.4.54) also hold p = q = ∞ and s ∈ (0, 1).

Proof Without loss of generality (otherwise working with Rn \ Ω in place of Ω; cf.

[68, Lemma 5.10.9] and [68, Lemma 5.10.10]) we may assume that Ω is bounded. The
action of the boundary-to-boundary single layer potential operator and the conormal
derivative of the double layer on Besov scales have been established in Theorem 4.3.4
and Theorem 4.3.5 for the type of open sets Ω considered here. Granted the current
hypotheses, from [68, (5.2.4), (5.10.24), (5.11.26)] we also see that Ω is a UR domain.
As such, Theorem 4.1.1 and Theorem 4.1.5 (dealing with mapping properties for the
boundary-to-boundary double layer potential operator and its formal transpose on the
scale of Besov spaces) are also valid in this context. The aforementioned theorems,
together with the operator identities from item (xiii) of Theorem 1.5.1 and the density
results from [69, Lemma 7.1.10] then yield (4.4.51)-(4.4.54), at least if q < ∞. The
case when q = ∞ may be covered a posteriori, based on what we have just proved
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 511

and loose embeddings (cf. items (iii)-(iv) in [69, Proposition 7.7.1]). Alternatively,
in the case when Ω is also assumed to be an (ε, δ)-domain, the operator identities
(4.4.51)-(4.4.54) may be justified starting from Green’s representation formula from
Theorem 4.4.1 on account of the boundary behavior of intervening layer potentials
(cf. Theorem 4.3.2, (4.3.72), (4.3.44), (4.3.83)). Finally, that the operator identities
in (4.4.51)-(4.4.54) continue to be valid when p = q = ∞ and s ∈ (0, 1) if Ω is
also assumed to be a uniform domain follows from what we have proved so far,
loose embeddings, and the mapping properties for S, ∂νA D, K, K A#  on Besov scales
with p = q = ∞ established in Theorem 4.3.4, Theorem 4.3.5, Theorem 4.1.1, and
Theorem 4.1.5. 

The relationship between Besov and Triebel-Lizorkin spaces exhibiting the same
first integrability exponent and the same amount of smoothness has been described
in [69, (9.2.11)]. For any homogeneous constant (complex) coefficient second-
order weakly elliptic M × M system L in Rn , our next corollary shows that a
new phenomenon
 occurs when considering the intersections of said spaces with
 ∞ M
Ker L := u ∈ 𝒞 (Ω) : Lu = 0 in Ω . Also, the first equality in (4.4.59)
below may be interpreted as saying that
informally speaking, pointwise multiplication by δ∂Ω
a “adds” a units of

smoothness (on Besov and Triebel-Lizorkin scales), i.e., multiplication

(4.4.56)
a acts like “fractional integration” of order a if a ≥ 0, and like
by δ∂Ω
“fractional differentiation” of order −a if a ≤ 0.
Corollary 4.4.3 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular
boundary and abbreviate σ := H n−1 ∂Ω. Also, fix an exponent p ∈ (1, ∞) along
with a number s ∈ (0, 1). Finally, consider a homogeneous constant (complex)
coefficient second-order
 weakly elliptic M × M system
 L in Rn , and recall the
 ∞ M
notation Ker L := u ∈ 𝒞 (Ω) : Lu = 0 in Ω . Then
 p,q M  
F (Ω) ∩ Ker L is independent of q ∈ n+s+1/p , ∞
n
. (4.4.57)
s+ p1

In particular, corresponding to q := p,
 p,q M  p, p M  n 
F 1 (Ω) ∩Ker L = B 1 (Ω) ∩Ker L for each q ∈ n+s+1/p , ∞ . (4.4.58)
s+ p s+ p

Moreover, under the additional assumption that the set Rn \ Ω is n-thick one has
(compare with [69, Corollary 9.2.31] and [69, Corollary 9.2.38])
 1, p M  p, p M  p,q M
Wa (Ω) ∩ Ker L = B 1 (Ω) ∩ Ker L = F 1 (Ω) ∩ Ker L
s+ p s+ p
 1  (4.4.59)
with a := 1 − s − p ∈ − p , 1 − p and n+s+1/p < q < ∞.
1 1 n

   p,q M
Proof Suppose q ∈ n+s+1/p , ∞
n
and consider a function u ∈ F 1 (Ω) ∩ Ker L.
s+ p
Pick an exponent qo ∈ (1, ∞) such that qo ≥ q and recall from [69, Corol-
512 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p,q p,q
lary 9.2.1] and [69, Theorem 9.1.1] that F 1 (Ω) → F 1o (Ω). In particular,
s+ p s+ p
 p,qo M
u ∈ F 1 (Ω) ∩ Ker L. Then item (2) in Theorem 4.4.1 (used with q := qo )
s+ p
ensures that
   
u = D TrΩ→∂Ω u − 𝒮 ∂νA(u, 0) in Ω, with
 p, p M  p, p M (4.4.60)
TrΩ→∂Ω u ∈ Bs (∂Ω, σ) and ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ) .

Granted this, from item (1) of Theorem 4.3.1 and item (1) of Theorem 4.3.3 we see
 p, p M
that u belongs to B 1 (Ω) ∩ Ker L. This proves the left-to-right inclusion in
s+ p
(4.4.58).
 p, p M
Conversely, given an arbitrary function u in the space B 1 (Ω) ∩ Ker L, we
s+ p
may invoke item (1) of Theorem 4.4.1 to conclude that the integral representation
formula (4.4.60) holds. Based on this, item (2) of Theorem 4.3.1 and item (2) of
 p,q M  n 
Theorem 4.3.3 it follows that u ∈ F 1 (Ω) ∩ Ker L for each q ∈ n+s+1/p ,∞ ,
s+ p
proving the right-to-left inclusion in (4.4.58). Hence (4.4.58) has been established,
and the claim in (4.4.57) is a direct consequence of it.
Finally, that (4.4.59) holds under the additional assumption that the set Rn \ Ω is
n-thick may be seen by reasoning as above, plus the help of item (3) in Theorem 4.4.1,
Theorem 4.2.3, and Theorem 4.2.10. 
As a consequence of (4.4.57), in the context of Corollary 4.4.3 the conormal
derivative operator from [69, Proposition 9.5.2] induces a well-defined, linear, and
continuous mapping
 p,q M  p, p M
F (Ω)∩ Ker L  u −→ ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ)
s+ p1
 n  (4.4.61)
whenever p ∈ (1, ∞), s ∈ (0, 1), and q ∈ n+s+1/p ,∞ .

Moving on, recall the class of injectively elliptic first-order systems in Rn in-
troduced in [70, Definition 1.3.4]. For further reference, let us also make here the
following convention. Given a homogeneous constant (complex) first-order injec-
tively elliptic N × M system in Rn along with some arbitrary open set Ω ⊆ Rn , we
agree to abbreviate
  M 
Ker D := u ∈ 𝒞∞ (Ω) : Du = 0 in Ω . (4.4.62)

We now discuss a basic integral representation formula for null-solutions of a first-

order injectively elliptic system D, which belong to suitable Besov, Triebel-Lizorkin,
or weighted Sobolev spaces in a bounded (ε, δ)-domain with an Ahlfors regular
boundary.
Theorem 4.4.4 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Next, let L be a homogeneous constant (complex)
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 513

second-order weakly elliptic M × M system in Rn . Consider a factorization of L of

the form

L = DD (4.4.63)
where

n 
n
= αγ γβ
D b j ∂j 1≤α ≤M and D = bk ∂k 1≤γ ≤ N (4.4.64)
j=1 1≤γ ≤ N j=k 1≤β ≤M

are homogeneous, constant (complex) coefficient, first-order systems in Rn , and

define the coefficient tensor (with the summation convention over repeated indices
in effect)
 αβ  αβ αγ γβ
AD,D
 := a jk 1≤ j,k ≤n where each a jk := 
b j bk (4.4.65)
1≤α,β ≤M

(so that, in particular, L = L A D,

(D
). Let D be the boundary-to-domain double layer
potential operator associated with the coefficient tensor AD,D and the set Ω as in
(1.3.18). Also, let K be the boundary-to-boundary double layer potential operator
associated with the coefficient tensor AD,D  and the set Ω, acting on the scale of
boundary Besov spaces as in (4.1.1). Finally, consider a vector-valued function
 M
u ∈ 𝒞∞ (Ω) satisfying Du = 0 in Ω.
Then u admits the integral representation formula
 
u = D TrΩ→∂Ω u in Ω (4.4.66)

and one has   

− 12 I + K TrΩ→∂Ω u = 0 on ∂Ω, (4.4.67)
in any of the scenarios described below:
 p,q M
(1) assume the function u belongs to B 1 (Ω) with n−1
n < p < ∞, 0 < q ≤ ∞,
s+ p
1 
and (n − 1) p − 1 + < s < 1;
 p,q M
(2) assume the function u belongs to the Triebel-Lizorkin space F 1 (Ω) with
s+ p
1 
n < p < ∞, 0 < q ≤ ∞, and (n − 1) p − 1 + < s < 1;
n−1

(3) make the additional assumption that Rn \ Ω is n-thick and suppose the function
 1, p M
u belongs to the weighted Sobolev space Wa (Ω) with p ∈ (1, ∞) and
 1 
a ∈ − p , 1 − p1 .
Proof Consider first the scenario described in item (1). We may then employ [69,
Corollary 9.2.1] to conclude that there exist po, qo ∈ (1, ∞) and so ∈ (0, 1) such that
p ,q  p ,q M
p,q
B 1 (Ω) ⊆ B o o1 (Ω). Since Lu = D(Du) = 0, it follows that u ∈ B o o1 (Ω)
s+ p so + p o so + p o
is a null-solution of the system L in Ω. Moreover, from [69, (9.5.36)] and the fact
that Du = 0 we see that
A D,
∂ν
(D
 ν)(Du, 0) = 0.
(u, 0) = (−i)Sym( D; (4.4.68)
514 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

Granted these properties of u, from item (1) of Theorem 4.4.1 we see that Green’s
integral representation formula (4.4.2) is presently valid and, in light of (4.4.68),
reduces precisely to (4.4.66). In turn, (4.4.67) readily follows from (4.4.66) and the
jump-formula from item (1) of Theorem 4.3.2. The situations described in items (2),
(3) are handled similarly (making use of [69, (8.5.49)] in the latter scenario). 
Our next result is the version of Corollary 4.4.3 for first-order injectively elliptic
systems (cf. also [69, (9.2.11)] and (4.4.56)).

Corollary 4.4.5 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular

boundary and abbreviate σ := H n−1 ∂Ω. Also, consider a homogeneous constant
(complex) coefficient first-order N × M system D in Rn which is injectively
 elliptic,
n < p < ∞ and (n − 1) p − 1 + < s < 1.
and recall (4.4.62). Finally, suppose n−1 1

Then
 p,q M  n 
F 1 (Ω) ∩ Ker D is independent of q ∈ n+s+1/p ,∞ (4.4.69)
s+ p

hence, corresponding to q := p,
 p,q M  p, p M
F (Ω) ∩ Ker D = B 1 (Ω) ∩ Ker D
s+ p1 s+ p
 n  (4.4.70)
for each q ∈ n+s+1/p , ∞ .

Finally, under the additional assumption that the set Rn \ Ω is n-thick and p ∈ (1, ∞),
one has
 1, p M  p, p M  p,q M
Wa (Ω) ∩ Ker D = B 1 (Ω) ∩ Ker D = F 1 (Ω) ∩ Ker D
s+ p s+ p
 1  (4.4.71)
with a := 1 − s − p ∈ − p , 1 − p and n+s+1/p < q < ∞.
1 1 n

Proof If D∗ denotes the Hermitian adjoint of D, then L := D∗ D is a homogeneous,

constant (complex) coefficient, weakly elliptic, second-order M ×M system in Rn (cf.
[70, (1.3.28)]). Intersecting (4.4.57)-(4.4.59) (written for the second-order weakly
elliptic system L) with Ker D then yields all desired conclusions, since it is immediate
that Ker L ∩ Ker D = Ker D.
Another, more direct, proof of the claims made in (4.4.69)-(4.4.71) goes as
follows. Fix some exponent q ∈ n+s+1/p n
, ∞ and consider an arbitrary function
 p,q M
u ∈ F 1 (Ω) ∩ Ker D. Then item (2) in Theorem 4.4.4 (applied for L := D∗ D)
s+ p
guarantees that we have the integral representation formula
 
u = D TrΩ→∂Ω u in Ω. (4.4.72)

Based on this, [69, (9.4.93) in item (ii) of Theorem 9.4.5], and item (1) in The-
orem 4.3.1 (with q := p) we then conclude that u belongs to the Besov space
 p, p M
B 1 (Ω) . This proves the left-to-right inclusion in (4.4.70). The opposite inclu-
s+ p
sion in (4.4.70) is justified in a similar manner, using item (1) in Theorem 4.4.4,
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 515

[69, (9.4.91) in item (ii) of Theorem 9.4.5] (with q := p), and item (2) in Theo-
rem 4.3.1. With (4.4.70) in hand, the claim in (4.4.69) also follows. Finally, under
the additional assumption that Rn \ Ω is n-thick and p ∈ (1, ∞), the claim in (4.4.71)
is justified in a similar manner, relying on item (3) in Theorem 4.4.4, [69, (8.3.38)],
and Theorem 4.2.3. 
We also have the following versions of the Maximum Principle adapted to null-
solutions of a first-order injectively elliptic system, belonging to Besov, Triebel-
Lizorkin, or weighted Sobolev spaces in a bounded (ε, δ)-domain with an Ahlfors
regular boundary.
Corollary 4.4.6 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Finally, let D be a homogeneous constant
(complex) first-order injectively elliptic N × M system in Rn , and recall (4.4.62).
1 
n < p < ∞, 0 < q ≤ ∞, and (n − 1) p − 1 + < s < 1. Then one has
(1) Assume n−1
the Besov-themed Maximum Principle
 
u[B p, q (Ω)] M ≈ TrΩ→∂Ω u[B p, q (∂Ω,σ)] M ,
1 s
 p,q M
s+ p
(4.4.73)
uniformly for u ∈ B 1 (Ω) ∩ Ker D.
s+ p

As a consequence,
 p,q M  p,q M
TrΩ→∂Ω : B 1 (Ω) ∩ Ker D −→ Bs (∂Ω, σ)
s+ p (4.4.74)
is an injective operator with closed range.
1 
(2) Assume n−1n < p < ∞, (n − 1) p − 1 + < s < 1, and n+s+1/p < q < ∞. Then
n

one has the Triebel-Lizorkin-themed Maximum Principle

 
u[F p, q (Ω)] M ≈ TrΩ→∂Ω u[B p, p (∂Ω,σ)] M ,
1
s+ p s
 p,q M (4.4.75)
uniformly for u ∈ F 1 (Ω) ∩ Ker D.
s+ p

In particular,
 p,q M  p, p M
TrΩ→∂Ω : F 1 (Ω) ∩ Ker D −→ Bs (∂Ω, σ)
s+ p
(4.4.76)
is an injective operator with closed range.

 that the set R \ Ω is1 n-thick and suppose

(3) Make the additional n
 assumption
p ∈ (1, ∞) and a ∈ − p , 1 − p . Then, with s := 1 − a − p one has the weighted
1 1

Sobolev-themed Maximum Principle

 
u[W 1, p (Ω)] M ≈ TrΩ→∂Ω u[B p, p (∂Ω,σ)] M ,
a s
 1, p M (4.4.77)
uniformly for u ∈ Wa (Ω) ∩ Ker D,
516 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

and, as such,
 1, p M  p, p M
TrΩ→∂Ω : Wa (Ω) ∩ Ker D −→ Bs (∂Ω, σ)
(4.4.78)
is an injective operator with closed range.

Proof To set the stage, recall from [70, (1.3.28)] that if D∗ is the Hermitian adjoint
of D then L := D∗ D is a homogeneous, constant (complex) coefficient, second-order
M × M system in Rn , which is weakly elliptic (in the sense of [70, (1.3.3) in Defini-
 p,q M
tion 1.3.1]). Suppose first that u ∈ B 1 (Ω) ∩ Ker D with p, q, s as described in
s+ p
item (1). Then the right-pointing inequality (4.4.73) is a direct consequence of [69,
(9.4.91) in item (ii) of Theorem 9.4.5], while the left-pointing inequality (4.4.73)
is a consequence of the integral representation formula (4.4.66) (considered in the
setting of item (1) in Theorem 4.4.4, applied with D  := D∗ ), and the mapping prop-
erties of the double layer operator from (4.3.1). This establishes the Besov-themed
Maximum Principle (4.4.73), and (4.4.74) is a direct consequence of it.
Next, the Triebel-Lizorkin-themed Maximum Principle (4.4.75) is justified in an
analogous fashion, based on [69, (9.4.93) in item (ii) of Theorem 9.4.5], the integral
representation formula (4.4.66) (in the setting of item (2) in Theorem 4.4.4, once
again applied with D  := D∗ ), and the mapping properties of the double layer operator
from (4.3.2). Finally, the claim in item (3) is dealt with similarly, now relying on
[69, (8.3.38)], the integral representation formula (4.4.66) (in the setting of item (3)
in Theorem 4.4.4, with D  := D∗ ), and the mapping properties of the double layer
operator from Theorem 4.2.3. 

In the next theorem we shall establish a basic integral representation formula

for null-solutions of injectively elliptic first-order systems allowed to have a low
amount of regularity (measured on Besov, Triebel-Lizorkin, and weighted Lebesgue
scales) to the point that the ordinary boundary trace may not exist. Instead, the
integral representation formula (4.4.80) (which should be compared with (4.4.66)
and the last line in (1.4.36)) employes the principal symbol map defined in [69,
Proposition 9.5.5] and [69, Proposition 8.5.7].

Theorem 4.4.7 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular

boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Next, let L be a homogeneous constant (complex)
second-order weakly elliptic M × M system in Rn . Consider a factorization of L of
the form
L = DD (4.4.79)
where D is a homogeneous, constant (complex) coefficient, first-order M × N system
in R , and D is a homogeneous, constant (complex) coefficient, first-order N × M
n

system in Rn . Bring in the matrix-valued fundamental solution E L  associated with

the transpose system L  as in [70, Theorem 1.4.2]. Finally, consider a vector-valued
 M
function u ∈ 𝒞∞ (Ω) satisfying Du = 0 in Ω.
Then u admits the integral representation formula
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 517
   
u(x) =  E L 
D (x − ·)∂Ω, (−i)Sym(D; ν)(u, 0)
 
  
= 
Dx E L  (x − ·) . γ  , (−i)Sym(D; ν)(u, 0) (4.4.80)
∂Ω 1≤γ ≤M

for all x ∈ Ω, in any of the scenarios described below.

 p,q M
(1) Assume u belongs to the Besov space B 1 (Ω) with p, q ∈ (1, ∞)
s+ p −1
 p,q N
and s ∈ (0, 1). Also, (−i)Sym(D; ν)(u, 0) ∈ Bs−1 (∂Ω, σ) is defined as
in [69, Proposition 9.5.5] (with A := B) and the brackets in the right-
hand side of (4.4.80) stand for the canonical duality pairing between the
 p,q N ∗  p ,q N   −1
space Bs−1 (∂Ω, σ) = B1−s (∂Ω, σ) (with p := 1 − p1 and
  −1  p,q  N
q := 1 − q1 ) and the space Bs−1 (∂Ω, σ) .
 p,q M
(2) Assume the function u belongs to the Triebel-Lizorkin space F 1 (Ω)
s+ p −1
 p, p N
with p, q ∈ (1, ∞) and s ∈ (0, 1). Also, (−i)Sym(D; ν)(u, 0) ∈ Bs−1 (∂Ω, σ)
is defined as in [69, Proposition 9.5.5] (with A := F) and the brackets in the
right-hand side of (4.4.80) stand for the canonical duality pairing between the
 p, p  N ∗  p ,p N   −1
space Bs−1 (∂Ω, σ) = B1−s (∂Ω, σ) (with p := 1 − p1 ) and the
 p, p N
space Bs−1 (∂Ω, σ) .
(3) Make the additional assumption that Rn \ Ω is n-thick and suppose the function
  ap  M
u belongs to the weighted Lebesgue space L p Ω, δ∂Ω L n with p ∈ (1, ∞)
 1   p, p N
and a ∈ − p , 1 − p . Finally, (−i)Sym(D; ν)(u, 0) ∈ Bs−1 (∂Ω, σ)
1
(with
s := 1 − a − p ∈ (0, 1)) is defined as in [69, Proposition 8.5.7] and the brack-
1

ets in the right-hand side of (4.4.80) stand for the canonical duality pairing
 p, p N ∗  p ,p N   −1
between Bs−1 (∂Ω, σ) = B1−s (∂Ω, σ) (with p := 1 − p1 ) and
 p, p N
Bs−1 (∂Ω, σ) .

Proof Work in the scenario described in item (1). To fix notation, assume


n 
n
= αγ γβ
D b j ∂j 1≤α ≤M and D = bk ∂k 1≤γ ≤ N . (4.4.81)
j=1 1≤γ ≤ N k=1 1≤β ≤M

Also, define the coefficient tensor (with the summation convention over repeated
indices in effect)
 αβ  αβ αγ γβ
AD,D
 := a jk 1≤ j,k ≤n where each a jk := 
b j bk , (4.4.82)
1≤α,β ≤M

so that  αβ 
L = L A D,
(D
= a jk ∂j ∂k 1≤α,β ≤M . (4.4.83)
518 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

To proceed, pick a point x ∈ Ω and select a scalar-valued function θ ∈ 𝒞∞ (Rn )

 θ = 0 on B(0, 1), and θ = 1 on R \ B(0, 2). Next, for each
with the property that n

ε ∈ 0, 12 dist(x, ∂Ω) define θ ε : Rn → R by setting

y−x
θ ε (y) := θ for every y ∈ Rn . (4.4.84)
ε
Hence, θ ε ∈ 𝒞∞ (Rn ) is a bounded function satisfying

lim θ ε (y) = 1 for every y ∈ Rn \ {x}, (4.4.85)

ε→0+
 
and there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we
have

1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε), supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε),
and |(∇ j θ ε )(y)| ≤ Cε −j for every j ∈ N0 and every y ∈ Rn .
(4.4.86)

Going further, fix μ ∈ {1, . . . , M } and define

    N
 E L  (x − ·)
w := θ ε D ∈ 𝒞∞ (Ω) . (4.4.87)
γμ 1≤γ ≤ N

Then

n
γβ
D w = − bk ∂k wγ
1≤β ≤M
k=1


n
  
γβ  E L  (x − ·)
=− bk (∂k θ ε ) D γμ , (4.4.88)
1≤β ≤M
k=1

since

n
  
γβ  E L  (x − ·)
θ ε bk ∂k D γμ 1≤β ≤M
k=1

 
 E L  (x − ·)
= θ ε D D βμ 1≤β ≤M

 
= θ ε L  E L  βμ (x − ·) = 0. (4.4.89)
1≤β ≤M

In particular,
 M
D w ∈ 𝒞∞c (Ω) . (4.4.90)
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 519

Writing the generalized integration by parts formula [69, (9.5.32)] for the present u,
 N
with f := 0 (so that Du = f Ω in D (Ω) ), and with w as in (4.4.87) then gives
   
the μ-th component of  E L 
D (x − ·)∂Ω, (−i)Sym(D; ν)(u, 0)
 
= ([B p , q (∂Ω,σ)] N )∗
(−i)Sym(D; ν)(u, 0), TrΩ→∂Ω w p ,q
[B1−s (∂Ω,σ)] N
1−s
∫
  γβ   
= − u, D w =  E L  (x − ·)uβ dL n
bk (∂k θ ε ) D γμ
Ω

=: Iε + IIε, (4.4.91)

where
∫
γβ     
Iε :=  E L  (x − y) uβ (y) − uβ (x) dy
bk (∂k θ ε )(y) D (4.4.92)
γμ
Ω

and
∫
γβ   
IIε := uβ (x)  E L  (x − y) dy.
bk (∂k θ ε )(y) D (4.4.93)
γμ
Ω

Next, based on [70, Theorem 1.4.2] and (4.4.86) we may estimate

⨏
   
 
lim sup Iε ≤ C lim sup u(y) − u(x) dy = 0. (4.4.94)
ε→0+ ε→0+ B(x,2ε)

Also, since 1 − θ ε ∈ 𝒞∞c (Ω) (cf. (4.4.86)), we may write

∫  
γβ    
 E L  (x − y) dy = D
bk (∂k θ ε )(y) D  E L  (x − ·), bγβ ∂k [θ ε − 1]
γμ γμ k
Ω
     
γβ  E L  (x − ·) , θ ε − 1
= − bk ∂k D γμ
  
 E L  (x − ·), θ ε − 1
= − D D βμ
  
= − L  E L  βμ (x − ·), θ ε − 1
 
= − δx δβμ, θ ε − 1 = −(θ ε (x) − 1)δβμ

= δβμ . (4.4.95)

In concert with (4.4.93) this proves that

 
IIε = uμ (x) for each ε ∈ 0, 12 dist(x, ∂Ω) . (4.4.96)
520 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

At this stage, (4.4.80) follows from (4.4.91), (4.4.92), (4.4.93), (4.4.94), and (4.4.96).
Finally, the validity of the integral representation formula (4.4.80) in the scenarios
described in items (2)-(3) is established in a similar fashion (we only wish to note
that for item (3) the generalized integration by parts formula [69, (8.5.45)] is used).

The results in the corollary below should be compared with Corollary 4.4.3 and
Corollary 4.4.5 (cf. also [69, (9.2.11)] and (4.4.56)).

Corollary 4.4.8 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular

boundary and abbreviate σ := H n−1 ∂Ω. Also, consider a homogeneous constant
(complex) coefficient first-order N × M system D in Rn , which is injectively elliptic.
Finally, fix p ∈ (1, ∞) along with s ∈ (0, 1) and recall (4.4.62). Then
 p,q M  
F (Ω) ∩ Ker D is independent of q ∈ n+s+1/p , ∞
n
. (4.4.97)
s+ p −1
1

In particular, corresponding to q := p,
 p,q M  p, p M
F (Ω) ∩ Ker D = B 1 (Ω) ∩ Ker D
s+ p1 −1 s+ p −1
 n  (4.4.98)
for each q ∈ n+s+1/p , ∞ .

Furthermore, under the additional assumption that the set Rn \ Ω is n-thick one
has
 p ap  M  p, p M  p,q M
L Ω, δ∂Ω L n ∩ Ker D = B 1 (Ω) ∩ Ker D = F 1 (Ω) ∩ Ker D
s+ p −1 s+ p −1
 
with a := 1 − s − p1 ∈ − p1 , 1 − p1 and n+s+1/p
n
< q < ∞.
(4.4.99)

Proof For starters, recall from [70, (1.3.28)] that if D∗ is the Hermitian adjoint
of D then L := D∗ D is a homogeneous, constant (complex) coefficient, second-
order M × M system in Rn , which is weakly elliptic (in the sense of [70, (1.3.3) in
Definition 1.3.1]). To proceed, suppose q ∈ n+s+1/p
n
, ∞ and consider an arbitrary
 p,q M
function u ∈ F 1 (Ω) ∩Ker D. Select an exponent qo ∈ (1, ∞) such that qo ≥ q
s+ p −1
p,q p,q
and recall from [69, Corollary 9.2.1] that F 1 (Ω) → F 1o (Ω). Consequently,
s+ p −1 s+ p −1
 p,qo M
u ∈ F 1 (Ω) ∩ Ker D. Placing ourselves in the scenario described in item (2)
s+ p −1
in Theorem 4.4.7 (used with q := qo ) permits us to express
   
u(x) = DE L  (x − ·)∂Ω, (−i)Sym(D; ν)(u, 0) , ∀x ∈ Ω,
 p, p N (4.4.100)
with (−i)Sym(D; ν)(u, 0) ∈ Bs−1 (∂Ω, σ) .

With this in hand, from item (1) of Theorem 4.3.3 and [69, (9.2.8)] we see that the
 p, p M
function u belongs to the space B 1 (Ω) ∩ Ker D. This proves the left-to-right
s+ p −1
inclusion in (4.4.98).
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 521
 p, p M
In the converse direction, whenever u ∈ B 1 (Ω) ∩ Ker D we are in the
s+ p −1
the context described in item (1) of Theorem 4.4.7 (with q := p) and this permits
us to conclude that the integral representation formula (4.4.60) holds. Together
with item (2) of Theorem 4.3.3 and [69, (9.2.8)], this implies that u belongs to
 p,q M  n 
F 1 (Ω) ∩Ker D for each q ∈ n+s+1/p , ∞ , proving the right-to-left inclusion
s+ p −1
in (4.4.98). This finishes the proof of (4.4.98), and the claim in (4.4.97) is a direct
consequence of (4.4.98). Finally, the fact that (4.4.99) holds under the additional
assumption that Rn \ Ω is n-thick may be seen by reasoning as above, now invoking
item (3) in Theorem 4.4.7 and Theorem 4.2.10. 

From Theorem 4.4.7 and [70, (1.3.40), (1.3.41)] we conclude that the gradient of
any null-solution of a homogeneous constant coefficient weakly elliptic second-order
system in a given bounded (ε, δ)-domain with an Ahlfors regular boundary satisfies
an integral representation formula. The aforementioned formula is identified in the
theorem below (via a conceptually different proof).

Theorem 4.4.9 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular

boundary with the property that
 
H n−1 ∂Ω \ ∂∗ Ω = 0. (4.4.101)

Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic out-
ward unit  normal to Ω. Next, for some M ∈ N, consider a coefficient tensor
αβ
A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M ho-
1≤α,β ≤M
mogeneous second-order system L = L A associated with A in Rn as in (1.3.2) is
weakly elliptic (in the sense of [70, (1.3.3)  in Definition 1.3.1]), and bring in the
matrix-valued fundamental solution E = Eαβ 1≤α,β ≤M associated with L as in [70,
 M
Theorem 1.4.2]. Finally, consider a null-solution u = (uα )1≤α ≤M ∈ 𝒞∞ (Ω) of
the system L in Ω.
Then, for each  ∈ {1, . . . , n} and γ ∈ {1, . . . , M }, the integral representation
formula
   
(∂ uγ )(x) = ar s (∂r Eγβ )(x − ·)∂Ω, ∂τ s TrΩ→∂Ω uα
βα

  
− (∂ Eγ .)(x − ·)∂Ω, ∂νA(u, 0) , ∀x ∈ Ω, (4.4.102)

(where the summation convention over repeated indices is in effect, and where Eγ .
denotes the γ-th row of the matrix-valued function E) holds in any of the following
scenarios:

(1) One assumes that

 p,q M
u belongs to the Besov space B 1 (Ω) with
s+ p
(4.4.103)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,
522 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

p,q p,q
TrΩ→∂Ω : B (Ω) → Bs (∂Ω, σ) is the boundary trace
s+ p1
operator from [69, (9.4.91) in item (ii) of Theorem 9.4.5] (4.4.104)
p,q
(thus TrΩ→∂Ω uα ∈ Bs (∂Ω, σ) for each index α),

∂τ s is the tangential derivative operator considered as in [69,

p,q (4.4.105)
(11.12.76)] (thus ∂τ s TrΩ→∂Ω uα ∈ Bs−1 (∂Ω, σ) for all α, , s),

∂νA is the conormal derivative associated with A and Ω as in [69,

 p,q M (4.4.106)
Proposition 9.5.2] with A := B (so ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ) ),

and
the brackets in the right-hand side of (4.4.102) are the duality pair-
p ,q  p,q ∗
ings between B1−s (∂Ω, σ) = Bs−1 (∂Ω, σ) (with p , q denoting (4.4.107)
p,q
the Hölder conjugate exponents of p, q) and Bs−1 (∂Ω, σ).

(2) One assumes that

 p,q M
u belongs to the Triebel-Lizorkin space F 1 (Ω) with
s+ p
(4.4.108)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,

p,q p, p
TrΩ→∂Ω : F (Ω) → Bs (∂Ω, σ) is the boundary trace
s+ p1
operator from [69, (9.4.93) in item (ii) of Theorem 9.4.5] (4.4.109)
 p, p M
(hence TrΩ→∂Ω uα ∈ Bs (∂Ω, σ) for each α),

∂τ s is the tangential derivative operator

 as in [69, (11.12.76)] with
p, p (4.4.110)
q := p (so that ∂τ s TrΩ→∂Ω uα ∈ Bs−1 (∂Ω, σ) for all α, , s),

∂νA is the conormal derivative associated with A and Ω as in [69,

 p, p M (4.4.111)
Proposition 9.5.2] with A := F (so ∂νA(u, 0) ∈ Bs−1 (∂Ω, σ) ),

and
the brackets appearing in the right-hand side of (4.4.102) are the
p ,p  p, p ∗
duality pairings between B1−s (∂Ω, σ) = Bs−1 (∂Ω, σ) (with (4.4.112)
p, p
p denoting the conjugate exponent of p) and Bs−1 (∂Ω, σ).

(3) Strengthen (4.4.101) by assuming that Rn \ Ω is n-thick, and suppose that

4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 523
 1, p M
u belongs to the weighted Sobolev space Wa (Ω) , with
(4.4.113)
1 < p < ∞, 0 < s < 1, a := 1 − s − p1 ,

1, p p, p
TrΩ→∂Ω : Wa (Ω) → Bs (∂Ω, σ) is the boundary trace
p, p
from [69, Theorem 8.3.6] (thus TrΩ→∂Ω uα ∈ Bs (∂Ω, σ) (4.4.114)
for each index α),

∂τ s is the tangential derivative operator

 as in [69, (11.12.76)] with
p, p (4.4.115)
q := p (so that ∂τ s TrΩ→∂Ω uα ∈ Bs−1 (∂Ω, σ) for all α, , s),

∂νA is the conormal derivative operator associated with A and Ω as

 p, p M
in [69, Proposition 8.5.3] (so, ∂νA(u, 0) ends up in Bs−1 (∂Ω, σ) ),
(4.4.116)
and
the brackets appearing in the right-hand side of (4.4.102) are the
p ,p  p, p ∗
duality pairings between B1−s (∂Ω, σ) = Bs−1 (∂Ω, σ) (with p (4.4.117)
p, p
denoting the conjugate exponent of p) and Bs−1 (∂Ω, σ).

Proof Let us establish the integral representation formula (4.4.102) in the scenario
described in the current item (1). Starting with the Green-type formula (4.4.2) in
the context specified in item (1) of Theorem 4.4.1 and applying ∂ , for some fixed
 ∈ {1, . . . , n}, to both sides yields
   
(∂ u)(x) = ∂ D TrΩ→∂Ω u (x) − ∂ 𝒮 ∂νA(u, 0) (x) for all x ∈ Ω. (4.4.118)

To proceed, denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit

normal to Ω. Having also fixed γ ∈ {1, . . . , M } and x ∈ Ω we may then compute
 
∂ D(TrΩ→∂Ω u) γ (x) (4.4.119)
∫
βα  
=− νs (y)ar s (∂ ∂r Eγβ )(x − y) TrΩ→∂Ω uα (y) dσ(y)
∂Ω
∫
βα  
=− ar s ∂τ s (y) [(∂r Eγβ )(x − y)] TrΩ→∂Ω uα (y) dσ(y)
∂Ω
   
ar s (∂r Eγβ )(x − ·)∂Ω, ∂τ s TrΩ→∂Ω uα B p, q (∂Ω,σ) .
βα
= p ,q
B1−s (∂Ω,σ) 1−s

Above, the first equality comes from (1.3.18), (4.4.101), and differentiation under
the integral sign. The second equality in (4.4.119) uses the definition of ∂τ s (y) and
[70, (1.4.33)]. The final equality in (4.4.119) is implied by [69, (7.6.9)] and [69,
Proposition 11.12.5)] (bearing in mind (4.4.104)-(4.4.105)). Moreover, on account
of (4.4.106), Proposition 4.2.8 presently gives
524 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
    
∂ 𝒮 ∂νA(u, 0) (x) = (∂ Eγ .)(x − ·)∂Ω, ∂νA(u, 0) . (4.4.120)
γ

At this stage, (4.4.102) follows by combining (4.4.118) with (4.4.119)-(4.4.120).

Finally, that the integral representation formula (4.4.102) also holds in the scenar-
ios described in items (2)-(3) in the statement of the theorem is justified in a similar
fashion. 

Our next result explores the nature of spaces of monogenic functions (i.e., null-
solutions of the Dirac operator in Rn ) which exhibit a certain prescribed amount
of smoothness measured on the scales of Besov, Triebel-Lizorkin, and weighted
Sobolev spaces in a bounded NTA domain with an Ahlfors regular boundary.

Theorem 4.4.10 Let Ω ⊆ Rn be a bounded NTA domain with an Ahlfors regular

boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Recall the formalism associated
) with the Clifford
algebra Cn from [68, §6.4], and the Dirac operator D := nj=1 e j  ∂j from (A.0.55).
In this context, define the Besov-Hardy spaces
 
p,q p,q
ℬα (Ω; D) := u ∈ Bα (Ω) ⊗ Cn : Du = 0 in Ω
(4.4.121)
0 < p ≤ ∞, 0 < q ≤ ∞, α ∈ R,
p,q
equipped in each instance with the quasi-norm inherited from Bα (Ω) ⊗ Cn . Then,
in relation to this scale of spaces, the following statements are true.

(1) Whenever
1 
n−1
n < p < ∞, 0 < q ≤ ∞, (n − 1) p −1 + < s < 1, (4.4.122)

the boundary-to-domain Cauchy-Clifford integral operator (cf. (4.3.19))

p,q p,q
C : Bs (∂Ω, σ) ⊗ Cn −→ ℬ (Ω; D) (4.4.123)
s+ p1

is well defined, linear, bounded, and surjective. In addition, one has the following
Plemelj-type jump-formula
p,q
TrΩ→∂Ω ◦ C = 12 I + C on Bs (∂Ω, σ) ⊗ Cn (4.4.124)

where C is the boundary-to-boundary Cauchy-Clifford integral operator from

(4.1.13).
(2) For each p, q, s as in (4.4.122) define the boundary Besov-Hardy space
 
p,q p,q
ℬs (∂Ω; D) := TrΩ→∂Ω u : u ∈ ℬ 1 (Ω; D) (4.4.125)
s+ p

p,q
and equip it with the quasi-norm inherited from Bs (∂Ω, σ) ⊗ Cn . Then
p,q p,q
ℬs (∂Ω; D) is a closed linear subspace of Bs (∂Ω, σ) ⊗ Cn and the operator
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 525
p,q p,q
1
2I + C is a projection1 of Bs (∂Ω, σ) ⊗ Cn onto ℬs (∂Ω; D). Furthermore,
% &
p,q p,q p,q
ℬs (∂Ω; D) = Ker 1
2I − C : Bs (∂Ω, σ) ⊗ Cn → Bs (∂Ω, σ) ⊗ Cn
% &
p,q p,q
= Im 1
2I + C : Bs (∂Ω, σ) ⊗ Cn → Bs (∂Ω, σ) ⊗ Cn
(4.4.126)

and
p,q p,q
C : ℬs (∂Ω; D) −→ ℬ (Ω; D) isomorphically,
s+ p1
p,q p,q (4.4.127)
with inverse TrΩ→∂Ω : ℬ 1 (Ω; D) −→ ℬs (∂Ω; D).
s+ p

p,q
(3) If p, q, s are as in (4.4.122), then for each function u ∈ ℬ (Ω; D) one has the
s+ p1
Cauchy Reproducing Formula
 
u = C TrΩ→∂Ω u in Ω, (4.4.128)

and the Besov-themed Maximum Principle

 
u p, q B ≈ TrΩ→∂Ω u
(Ω)⊗ Cn Bs
p, q
(∂Ω,σ)⊗ Cn
(4.4.129)
1
s+ p

(where the intervening proportionality constants are independent of u).

(4) Similar properties2 to those described in items (1)-(3) are valid for the Triebel-
Lizorkin-Hardy spaces
 
p,q p,q
ℱα (Ω; D) := u ∈ Fα (Ω) ⊗ Cn : Du = 0 in Ω
(4.4.130)
0 < p < ∞, 0 < q ≤ ∞, α ∈ R,

and for the weighted Sobolev-Hardy spaces

 
1, p 1, p
𝒲a (Ω; D) := u ∈ Wa (Ω) ⊗ Cn : Du = 0 in Ω
  (4.4.131)
1 < p < ∞, a ∈ − p1 , 1 − p1 .

1 
(5) Whenever n−1
n < p < ∞ and (n − 1) p −1 + < s < 1 it follows that
p,q  
ℱ (Ω; D) is independent of q ∈ n+s+1/p , ∞
n
, (4.4.132)
s+ p1

in particular, corresponding to q := p,

1 i.e., a linear, bounded, and idempotent operator

2 now employing diagonal Besov spaces on the boundary
526 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p,q p, p  
ℱ (Ω; D) =ℬ (Ω; D) for each q ∈ n+s+1/p , ∞
n
. (4.4.133)
s+ p1 s+ p1

Moreover,
1, p p,q p, p
𝒲a (Ω; D) = ℱ (Ω; D) =ℬ (Ω; D)
s+ p1 s+ p1
whenever 1 < p < ∞, 0 < s < 1, (4.4.134)
a := 1 − s − p1 , and n
n+s+1/p < q < ∞.

Proof For starters, observe that the present geometric assumptions imply that Ω
is both a bounded (ε, δ)-domain whose complement is n-thick, as well as a UR
domain (cf. [68, (5.11.66)], [68, (5.10.24)], [68, (5.2.4)], and [68, (5.1.6)]). That the
operator (4.4.123) is well defined, linear, bounded is clear from (4.3.19), (4.4.121),
p,q
and the fact that D(C f ) = 0 in Ω for each f ∈ Bs (∂Ω, σ) ⊗ Cn (cf. (2.1.78)). The
jump-formula (4.4.124) is implied by item (1) of Theorem 4.3.2. Next, the Cauchy
Reproducing Formula (4.4.128) is a particular case of the integral representation
formula (4.4.66) in the context described in item (1) of Theorem 4.4.4 (taking D
to be the Dirac operator (A.0.55)). In concert with [69, (9.4.91) in item (ii) of
Theorem 9.4.5], the Cauchy Reproducing Formula (4.4.128) then proves that the
Cauchy-Clifford integral operator (4.4.123) is surjective. Going further, (4.4.74)
p,q
(used with D the Dirac operator (A.0.55)) presently implies that ℬs (∂Ω; D) is a
p,q
closed linear subspace of Bs (∂Ω, σ) ⊗ Cn . Also, knowing that the Cauchy-Clifford
integral operator (4.4.123) is surjective and having established the Plemelj-type
jump-formula (4.4.124) permits us to recast (4.4.125) as
  
p,q p,q
ℬs (∂Ω; D) = 12 I + C f : f ∈ Bs (∂Ω, σ) ⊗ Cn . (4.4.135)

 2 p,q
Since from (4.1.13), (4.1.15) we see that 12 I + C = 12 I + C on Bs (∂Ω, σ) ⊗ Cn ,
we ultimately conclude (bearing (4.4.135) in mind) that 12 I + C is a projection of
p,q p,q
Bs (∂Ω, σ) ⊗ Cn onto ℬs (∂Ω; D). Furthermore, (4.4.126) follows from (4.4.67)
and (4.4.135), while (4.4.129) comes from (4.4.73). Next, that the Cauchy-Clifford
operator C is an isomorphism in the context of (4.4.127), with the trace operator
serving as inverse, is clear from (4.4.128), (4.4.125), (4.4.124), and (4.4.126). The
argument so far yields all claims made in items (1)-(3), and the claim made in
item (4) is dealt with in a very similar fashion. Finally, the claims in item (5) are
consequences of Corollary 4.4.5 (applied with D the Dirac operator (A.0.55)). 

To state our final result in this section, recall the Clifford-Riesz transform R C ,
originally defined as in (2.1.77) for a given Ahlfors regular domain Ω ⊆ Rn with
compact boundary. Its action may then be extended to distributions by setting
  )) n   
R C f (x) := 2 
Lip (∂Ω) Φ j (x − ·) ∂Ω, fJ (Lip (∂Ω)) e j  eJ
J j=1
  (4.4.136)
)
for each f = J fJ eJ ∈ Lip (∂Ω) ⊗ Cn and x ∈ Ω,
4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 527
xj
where Φ j (x) := 1
ω n−1 |x | n for all x = (x1, . . . , xn ) ∈ Rn \ {0}, with 1 ≤ j ≤ n.

Theorem 4.4.11 Suppose Ω ⊆ Rn is a bounded NTA domain with an Ahlfors regular

boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Recall the Dirac operator D from (A.0.55), the
scale of Besov-Hardy spaces from (4.4.121), and the Clifford-Riesz transform R C
associated with Ω as in (4.4.136). Then the following claims are true.

(1) The operator

p,q p,q
R C : Bs−1 (∂Ω, σ) ⊗ Cn −→ ℬ (Ω; D)
s+ p1 −1
1  (4.4.137)
n−1
n < p < ∞, 0 < q ≤ ∞, (n − 1) p −1 + < s < 1,

is well defined, linear, and bounded.

p,q
(2) If 1 < p, q < ∞ and 0 < s < 1, then for each f ∈ Bs−1 (∂Ω, σ) ⊗ Cn one has
the jump-formula
   
(−i)Sym(D; ν) R C f , 0 = I − 2C # f . (4.4.138)

where the principal symbol map (associated with the Dirac operator D) is
defined as in [69, Proposition 9.5.5], and where C # is the boundary-to-boundary
“transpose” Cauchy-Clifford operator (cf. (4.1.49)).
(3) Assume 1 < p, q < ∞ and 0 < s < 1. Then the principal symbol map
p,q p,q
ℬ (Ω; D)  u −→ (−i)Sym(D; ν)(u, 0) ∈ Bs−1 (∂Ω, σ) ⊗ Cn (4.4.139)
s+ p1 −1

(associated with the Dirac operator D as in [69, Proposition 9.5.5]) is well

defined, linear, and bounded. Moreover, if one defines
 
p,q p,q
ℬs−1 (∂Ω; D) := (−i)Sym(D; ν)(u, 0) : u ∈ ℬ 1 (Ω; D) (4.4.140)
s+ p −1

p,q
and equip it with the quasi-norm inherited from Bs−1 (∂Ω, σ) ⊗ Cn , then
p,q p,q
ℬs−1 (∂Ω; D) is a closed linear subspace of Bs−1 (∂Ω, σ) ⊗ Cn and the op-
p,q p,q
erator 12 I − C # is a projection3 of Bs−1 (∂Ω, σ) ⊗ Cn onto ℬs−1 (∂Ω; D). In
addition,
% &
p,q p,q p,q
ℬs−1 (∂Ω; D) = Ker 12 I + C # : Bs−1 (∂Ω, σ) ⊗ Cn → Bs−1 (∂Ω, σ) ⊗ Cn
% &
p,q p,q
= Im 1
2I − C # : Bs−1 (∂Ω, σ) ⊗ Cn → Bs−1 (∂Ω, σ) ⊗ Cn
(4.4.141)

and

3 i.e., a linear, bounded, and idempotent operator

528 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces
p,q p,q
the operator R C : ℬs−1 (∂Ω; D) −→ ℬ (Ω; D)
s+ p1 −1
is an isomorphism whose inverse is the principal symbol map (4.4.142)
p,q p,q
ℬ 1 (Ω; D)  u → (−i/2)Sym(D; ν)(u, 0) ∈ ℬs−1 (∂Ω; D).
s+ p −1

p,q
(4) If 1 < p, q < ∞ and 0 < s < 1, then each u ∈ ℬ (Ω; D) has the integral
s+ p1 −1
representation formula
 
u = 12 R C (−i)Sym(D; ν)(u, 0) in Ω, (4.4.143)

and the following Besov-themed Maximum Principle holds

 
uB p, q (Ω)⊗ Cn ≈ (−i)Sym(D; ν)(u, 0)B p, q (∂Ω,σ)⊗ Cn (4.4.144)
1 −1
s+ p s−1

p,q
uniformly for u ∈ ℬ (Ω; D).
s+ p1 −1

(5) Similar properties4 to those described in items (1)-(4) are valid for the scale of
Triebel-Lizorkin-Hardy spaces (cf. (4.4.130)). In fact, whenever 1 < p < ∞ and
0 < s < 1 it follows that
p,q  n 
ℱ 1 (Ω; D) is independent of q ∈ n+s+1/p ,∞ , (4.4.145)
s+ p −1

in particular, corresponding to q := p,
p,q p, p  
ℱ (Ω; D) =ℬ (Ω; D) for each q ∈ n+s+1/p , ∞
n
. (4.4.146)
s+ p1 −1 s+ p1 −1

Moreover,
 p ap   p,q p, p
L Ω, δ∂Ω L n ⊗ Cn ∩ Ker D = ℱ 1 (Ω; D) = ℬ (Ω; D)
s+ p −1 s+ p1 −1

if 1 < p < ∞, 0 < s < 1, a := 1 − s − p1 , and n

n+s+1/p < q < ∞.
(4.4.147)

Proof In concert with [68, (5.11.66)], [68, (5.10.24)], [68, (5.2.4)], and [68, (5.1.6)],
the present geometric assumptions imply that Ω is both a bounded (ε, δ)-domain
whose complement is n-thick, as well as a UR domain. To proceed, bring in the
boundary-to-domain single layer potential operator 𝒮Δ associated with the Laplacian
in Ω, and observe that
 
R C = 2D𝒮Δ on Lip (∂Ω) ⊗ Cn . (4.4.148)

In particular,
 
DR C = 2D2 𝒮Δ = −2Δ𝒮Δ = 0 on Lip (∂Ω) ⊗ Cn . (4.4.149)

4 now employing diagonal Besov spaces on the boundary

4.4 Integral Representation Formulas of Layer Potential Type, and Consequences 529

From (4.4.148), (4.4.149), item (1) of Theorem 4.3.3, and [69, (9.2.8)] we may
then conclude that the Clifford-Riesz transform induces a well-defined, linear, and
p,q
bounded operator in the context of (4.4.137). Next, if f ∈ Bs−1 (∂Ω, σ) ⊗ Cn with
1 < p, q < ∞ and 0 < s < 1, then we may compute
   
(−i)Sym(D; ν) R C f , 0 = (−2i)Sym(D; ν) D𝒮Δ f , 0
A−D, D    
= −2∂ν 𝒮Δ f , 0 = I − 2C # f , (4.4.150)

thanks to (4.4.148), [69, (9.5.36)], and the jump-formula (4.3.83) in the context of
item (b) in Theorem 4.3.6 (also bearing in mind the discussion in Example 1.4.12; cf.
(1.4.109) and the comment following it). Also, the fact that the principal symbol map
(4.4.139) is well defined, linear, and bounded is a direct consequence of (4.4.121)
 2
and [69, Proposition 9.5.5]. Since from (4.1.53) we also see that 12 I +C # = 12 I +C #
p,q p,q
on Bs−1 (∂Ω, σ) ⊗ Cn , we deduce that 12 I + C # is a projection of Bs−1 (∂Ω, σ) ⊗ Cn .
In particular, the second equality in (4.4.141) follows.
p,q
Going further, for each u ∈ ℬ 1 (Ω; D) with p, q ∈ (1, ∞) and s ∈ (0, 1) the
s+ p −1
integral representation formula (4.4.80) presently becomes

u = D𝒮Δ (−i)Sym(D; ν)(u, 0) in Ω. (4.4.151)

On account of this and (4.4.148) wee then conclude that the integral representation
formula (4.4.143) holds. In addition, from (4.4.151) and (4.4.138) we see that
  
(−i)Sym(D; ν)(u, 0) = 12 I − C # (−i)Sym(D; ν)(u, 0) , (4.4.152)

hence
1  
2I + C # (−i)Sym(D; ν)(u, 0) = 0. (4.4.153)

In concert with (4.4.140), this proves the left-to-right inclusion in the first equality
in (4.4.141). Since (4.4.140), (4.4.137), and (4.4.138) ensure that the space in the
p,q
last line of (4.4.141) is contained in ℬs−1 (∂Ω; D), the circular chain of inclusions
shown so far simultaneously proves all equalities in (4.4.141). This has several
consequences. First, as a byproduct of (4.4.141) and the continuity of (4.1.49)
p,q p,q
we conclude that ℬs−1 (∂Ω; D) is a closed linear subspace of Bs−1 (∂Ω, σ) ⊗ Cn .
p,q
Second, we conclude from (4.4.141) that 12 I − C # projects Bs−1 (∂Ω, σ) ⊗ Cn onto
p,q
ℬs−1 (∂Ω; D).
Pressing on, the integral representation formula recorded in (4.4.143) readily im-
plies (in view of (4.4.137), (4.4.139), and (4.4.140)) that the Clifford-Riesz transform
is an isomorphism in the context of (4.4.142). Next, the right-pointing inequality in
(4.4.144) is a direct consequence of the boundedness of the principal symbol map
in (4.4.139), while the left-pointing inequality in (4.4.144) is implied by (4.4.143),
530 4 Layer Potential Operators Acting from Boundary Besov and Triebel-Lizorkin Spaces

(4.4.139), and the boundedness of the Clifford-Riesz transform in (4.4.137). The

proof so far addresses all claims made in items (1)-(4), and the very first claim made
in item (5) is dealt with in a similar fashion. Finally, the claims in (4.4.145)-(4.4.147)
are consequences of Corollary 4.4.8. 
Chapter 5
Generalized Double Layers in Uniformly
Rectifiable Domains

Matching classes of singular integrals with function spaces on which they behave in
a natural fashion is a topic at the core of the classical Calderón-Zygmund theory in
the entire Euclidean space. See, for instance, [65] where the focus is on Calderón-
Zygmund operators considered on Lebesgue (plain and Muckenhoupt weighted),
Hölder, and Sobolev spaces in Rn . The main tools employed are wavelets and
the T(1) Theorem of David and Journé. In this chapter we are interested in the
situation when the Euclidean space is replaced by a more general “surface” and the
philosophy that emerges is that not all singular integral operators are created equal.
For example, not every singular integral operator bounded on Lebesgue spaces L p
p
(with 1 < p < ∞) is bounded on Sobolev spaces L1 , as a certain algebraic structure
is needed, linking said operator to the underlying surface. This is the case for double
layer potential operators we have studied earlier. In fact, these enjoy a host of rather
specialized properties, which are not generally shared by “ordinary”, garden variety
SIO’s. Here the goal is to further nuance such distinctions.
The trade-mark characteristic of what we shall call a generalized double layer
operator is the fact that its integral kernel is the inner product of the outward unit
normal (to the “surface” on which this integral operator is defined) with a divergence-
free vector-valued kernel; see (5.1.1)-(5.1.2) and (5.1.4)-(5.1.6) below. The algebraic
structure just described confers excellent cancelation properties (brought to fruition
by the Divergence Theorem) which, in turn, permit us to establish boundedness
results for these generalized double layers for a multitude of basic scales of function
spaces which, in addition to standard L p spaces with p ∈ (1, ∞), now also includes
boundary Sobolev spaces, Hardy spaces, Hölder spaces, the John-Nirenberg space
BMO, the Sarason space VMO, Besov spaces, and Triebel-Lizorkin spaces, among
others.
In fact, our generalized double layer operators make up the largest class of singular
integral operators enjoying the aforementioned properties.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 531
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_5
532 5 Generalized Double Layers in Uniformly Rectifiable Domains

5.1 Theory of Generalized Double Layers

A theory of generalized double layers, of the sort advertised earlier, is presented in

Theorem 5.1.1 below (see also Theorem 5.1.8 and Theorem 5.1.15 further in this
section). Through its proof, this establishes a firm link between our versions of the
Divergence Theorem from [68, Chapter 1] and singular integral operators on UR sets
acting on scales of function spaces which, directly or implicitly, involve cancelations.
Theorem 5.1.1 (Theory of Generalized Double Layers) Fix n ∈ N with n ≥ 2,
and suppose Ω ⊆ Rn is an arbitrary UR domain1. Abbreviate σ := H n−1 ∂Ω and
denote by ν the geometric measure theoretic outward unit normal to Ω. Having fixed
a sufficiently large integer N = N(n) ∈ N, consider a vector-valued function
 n
k = (k j )1≤ j ≤n ∈ 𝒞 N (Rn \ {0})
(5.1.1)
odd, positive homogeneous of degree 1 − n,

and satisfying (with the summation convention over repeated indices in effect)

div k = ∂j k j = 0 in Rn \ {0}. (5.1.2)

Finally, set
∫
ϑ := 
ω, k(ω) dH n−1 (ω) ∈ C. (5.1.3)
S n−1
 
In this setting, introduce integral operators2 acting on f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
according to
∫
T f (x) :=  − y) f (y) dσ(y) for all x ∈ Ω,
ν(y), k(x (5.1.4)
∂Ω
 
and for each f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
consider
∫
T f (x) := lim+  − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ν(y), k(x (5.1.5)
ε→0
y ∈∂Ω
|x−y |>ε

as well as
∫
T # f (x) := lim+  − x) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω.
ν(x), k(y (5.1.6)
ε→0
y ∈∂Ω
|x−y |>ε

1 If Ω ⊆ R n is merely an open set with a UR boundary, a large number of conclusions continue to

hold, with at most minor alterations (namely, eventually replacing ∂Ω with ∂∗ Ω).
2 which we think of as “generalized double layer potential operators”
5.1 Theory of Generalized Double Layers 533

Then the following statements are true.

 
(1) For each aperture parameter κ > 0, each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, and
σ-a.e. point x ∈ ∂Ω one has
 κ−n.t.  ϑ

T f (x) = − f (x) + (T f )(x). (5.1.7)
∂Ω 2
 
Also, if for each f ∈ L 1 ∂Ω, 1+σ(x)|x | n−1
one defines
∫
W f (x) :=  − y) dσ(y) for all x ∈ Ω,
f (y) k(x (5.1.8)
∂Ω
 
then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has
 n 
W f ∈ 𝒞 N (Ω) and div W f = 0 in Ω. (5.1.9)
 
Moreover, for each κ > 0 and f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, the nontangential boundary
trace κ−n.t.
(W f )∂Ω exists at σ-a.e. point on ∂Ω, (5.1.10)
and at σ-a.e. point x ∈ ∂Ω one has
 κ−n.t.  ϑ

ν(x) · W f  (x) = − f (x) − (T # f )(x). (5.1.11)
∂Ω 2
Finally, if ∂Ω is compact then both T and T map constant functions to constant
functions and, in fact, at each point in Ω we have

−ϑ if Ω is bounded,
T1 = (5.1.12)
0 if Ω is an exterior domain,

(recall that an exterior domain is the complement of a compact set in Rn ), while

at σ-a.e. point on ∂Ω we have

− ϑ2 if Ω is bounded,
T1 = (5.1.13)
+ ϑ2 if Ω is an exterior domain.

(2) For each p ∈ [1, ∞) and κ > 0 there exists some finite constant C > 0, depending
 n, p, and κ, such that for each function f ∈ L p (∂Ω, σ) one has
only on ∂Ω, k,

max Nκ (T f ) , Nκ (W f )
L p (∂Ω,σ) L p (∂Ω,σ)
≤C f L p (∂Ω,σ) if p > 1,
(5.1.14)
534 5 Generalized Double Layers in Uniformly Rectifiable Domains

plus similar estimates in the case when p = 1 in which scenario the corre-
sponding L 1 -norms in the left-hand side are now replaced by the quasi-norm in
L 1,∞ (∂Ω, σ).
Moreover, the action of the operator W, originally considered as in (5.1.8),
may be further extended in a unique and coherent fashion (cf. [70, (2.4.15),
(2.4.16), (2.4.24)]) to the scale of Lorentz-based Hardy spaces H p,q (∂Ω, σ)
with p ∈ n−1 n , ∞ and q ∈ (0, ∞] and said extension satisfies (for some constant
C = C(∂Ω, k,  n, p, q) ∈ (0, ∞))

Nκ (W f ) L p, q (∂Ω,σ)
≤ C f H p, q (∂Ω,σ)
(5.1.15)
for all f ∈ H p,q (∂Ω, σ).

Also, whenever f ∈ H p,q (∂Ω, σ) has compact support (as a distribution, which
is automatically the case if ∂Ω is compact) and ψ ∈ Lipc (∂Ω) is identically one
near supp f one has
   
 − ·) , f (Lip (∂Ω))
W f (x) = Lipc (∂Ω) ψ k(x ∂Ω c
(5.1.16)
for each x ∈ Ω.

(3) For each p ∈ (1, ∞), the operators

T : L p (∂Ω, σ) → L p (∂Ω, σ) and T # : L p (∂Ω, σ) → L p (∂Ω, σ) (5.1.17)

are well defined, linear, and bounded. Moreover, given any p, p ∈ (1, ∞) with
1/p + 1/p = 1 it follows that

the transpose of T acting on L p (∂Ω, σ)

 (5.1.18)
is the operator T # acting on L p (∂Ω, σ).

Corresponding to the end-point p = 1, both operators T and T # map L 1 (∂Ω, σ)

  into L (∂Ω, σ). Next, if for each vector-valued function
linearly 1,∞
 and
 boundedly n
g ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
one considers
∫
V g(x) := lim+  − y), g(y) dσ(y) for σ-a.e. x ∈ ∂Ω,
k(x (5.1.19)
ε→0
y ∈∂Ω
|x−y |>ε

then
 
V( f ν) = T f for each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1 (5.1.20)

and for each p ∈ (1, ∞) the following operator is well defined, linear, and
bounded:
5.1 Theory of Generalized Double Layers 535
 n
V : L p (∂Ω, σ) −→ L p (∂Ω, σ). (5.1.21)

Finally, similar results are valid for Muckenhoupt weighted Lebesgue spaces,
Lorentz spaces, and Morrey spaces (as well as their duals and their preduals)
on ∂Ω.
 
(4) Fix p ∈ n−1n , 1 . Then the operator T , originally acting on Lebesgue spaces as
#

in (the first part of) item (3), extends to a linear and bounded mapping from the
Hardy space H p (∂Ω, σ) into itself,

T # : H p (∂Ω, σ) −→ H p (∂Ω, σ) (5.1.22)

 
and the operators corresponding to various choices of the index p ∈ n−1n , 1 are
compatible with one another. As far as the dependence of the operator norm for
T # in (5.1.22) on the kernel k is concerned, homogeneity considerations dictate
that   
T # H p (∂Ω,σ)→H p (∂Ω,σ) ≤ C sup |∂ α k | , (5.1.23)
n−1
|α | ≤ N S

where the constant C ∈ (0, ∞) depends only on n, p, and the UR character of

∂Ω. In addition,

if p ∈ n−1
n , ∞ , the composition to the left of T from (5.1.22) with
#

the L -filtering operator H : H (∂Ω, σ) → L p (∂Ω, σ) (cf. [69,

p p
(5.1.24)
(4.9.2)]) is the operator T # : H p (∂Ω, σ) → L p (∂Ω, σ) considered
as in [70, (2.3.27)].

In fact, T # maps the scale of Lorentz-based Hardy spaces on ∂Ω continuously

into itself, specifically

T # : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) is well defined,

 (5.1.25)
linear, and bounded for p ∈ n−1 n , ∞ and q ∈ (0, ∞].

For example, for σ-a.e. xo ∈ ∂Ω one has (with δxo ∈ H 1,∞ (∂Ω, σ) and T #
acting as in (5.1.25) with p = 1 and q = ∞)
  
 o − ·)
T # δxo = −P.V. ν, k(x ∂Ω
 
(5.1.26)
as distributions in Lipc (∂Ω) ,

and for σ-a.e. x0, x1 ∈ ∂Ω one has (with δx0 − δx1 ∈ n−1 <p<1 H p (∂Ω, σ) and
 n
T # acting as in (5.1.25) with p ∈ n−1n , 1 and q = p)
     
 0 − ·)
T # (δx0 − δx1 ) = −P.V. ν, k(x  1 − ·)
+ P.V. ν, k(x
∂Ω ∂Ω
 
(5.1.27)
as distributions in Lipc (∂Ω) .
536 5 Generalized Double Layers in Uniformly Rectifiable Domains

In addition, for each f ∈ H p,q (∂Ω, σ) with p ∈ n−1 n , ∞ and q ∈ (0, ∞] one
has
ϑ
ν • W f = − f − T# f, (5.1.28)
2
where the “bullet product” is defined as in [68, Proposition 4.2.3].
Finally, for each function f ∈ H 1 (∂Ω, σ) one has

f ∈ L 1 (∂Ω, σ), T # f ∈ L 1 (∂Ω, σ), (5.1.29)

and, with all integrals involved absolutely convergent,

∫
⎧
⎪ ϑ
⎪
⎪ − f dσ if Ω is bounded,
∫ ⎪
⎪ 2 ∂Ω
⎪
⎨
⎪ ∫
T # f dσ = ϑ
∂Ω ⎪
⎪ + f dσ if Ω is unbounded and ∂Ω is bounded,
⎪
⎪ 2 ∂Ω
⎪
⎪
⎪ if ∂Ω is unbounded.
⎩ 0
(5.1.30)

(5) Select q ∈ (1, ∞) along with λ ∈ (0, n − 1), and recall ℋq,λ (∂Ω, σ), the pre-dual
to the Morrey-Campanato space, defined as in (A.0.84) (with Σ := ∂Ω). Then
the operator

q(n − 1)
T # : L r (∂Ω, σ) −→ L r (∂Ω, σ) with r := (5.1.31)
n − 1 + λ(q − 1)

has ℋq,λ (∂Ω, σ) as an invariant subspace (cf. (5.1.17) and [69, (6.1.22)]), and

T # : ℋq,λ (∂Ω, σ) −→ ℋq,λ (∂Ω, σ) (5.1.32)

is a well defined, linear, and bounded operator.

(6) For each given function f belonging to the weighted boundary Sobolev space

L11 ∂Ω, 1+σ(x)
|x | n−1
(cf. (A.0.131)), each index ∈ {1, . . . , n}, and each aperture
parameter κ ∈ (0, ∞), the pointwise nontangential boundary trace
 κ−n.t.
∂ T f ∂Ω exists at σ-a.e. point on ∂Ω. (5.1.33)
p,q
As a consequence of [69, (11.7.22)], this is true whenever f ∈ L1 (∂Ω, σ) with
p, q ∈ [1, ∞).
Furthermore, for each p, q ∈ [1, ∞) and κ > 0 there exists some finite constant
C > 0, depending only on ∂Ω, k,  n, p, q, and κ, such that for each function
p,q
f ∈ L1 (∂Ω, σ) one has

Nκ (T f ) L p (∂Ω,σ)
+ Nκ (∇T f ) L q (∂Ω,σ)
≤C f p, q
L1 (∂Ω,σ) if p, q > 1,
(5.1.34)
5.1 Theory of Generalized Double Layers 537

plus similar estimates in the case when p = 1 or q = 1, in which scenario the

corresponding L 1 -norms in the left-hand side are now replaced by the quasi-
norm L 1,∞ (∂Ω, σ).
(7) The operator
p,q p,q
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.1.35)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
p p
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.1.36)

is well defined, linear, and bounded for each p ∈ (1, ∞). In addition,
p p
T : L1 (∂Ω, w) −→ L1 (∂Ω, w) is well defined, linear, and bounded
(5.1.37)
for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ).

Moreover, for each function

 σ(x)  p
f ∈ L11 ∂Ω, ∩ L1,loc (∂Ω, σ) for some p ∈ (1, ∞) (5.1.38)
1 + |x| n−1
it follows that
p
T f ∈ L1,loc (∂Ω, σ), (5.1.39)
and for each pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.1.19))
    
∂τr s (T f ) = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f ) (5.1.40)

at σ-a.e. point on ∂Ω. In particular, formula (5.1.40) holds for every function
p,q p
f ∈ L1 (∂Ω, σ) with p, q ∈ (1, ∞), as well as for every f ∈ L1 (∂Ω, w) with
p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ). Formula (5.1.40) also holds for each function
p,λ q,λ
in M1 (∂Ω, σ) or in B1 (∂Ω, σ) with p, q ∈ (1, ∞) and λ ∈ (0, n − 1) (cf.
(A.0.150), (A.0.33)). In fact,
p,λ
the operator T maps each of the Sobolev spaces M1 (∂Ω, σ),
p,λ q,λ
M̊1 (∂Ω, σ), B1 (∂Ω, σ) with p, q ∈ (1, ∞) and λ ∈ (0, n − 1) (cf. (5.1.41)
(A.0.150), (A.0.155), (A.0.33)) boundedly into themselves.

(8) For each p ∈ (1, ∞) it follows that T # , originally acting on functions from
L p (∂Ω, σ), further extends uniquely to a linear, bounded operator, from the
p
negative boundary Sobolev space L−1 (∂Ω, σ) into itself. Furthermore, if one
retains the same notation T # for said extension, then the transpose of (5.1.36) is
538 5 Generalized Double Layers in Uniformly Rectifiable Domains

p p
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.1.42)

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

More generally, for each p, q ∈ (1, ∞) it follows that T # acting on functions
from L p (∂Ω, σ) further extends uniquely to a linear and bounded operator from
p,q
the negative off-diagonal boundary Sobolev space L−1 (∂Ω, σ) into itself and,
#
adopting the same notation T for this extension, the transpose of (5.1.35) is
p,q p,q
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.1.43)

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1.
In addition, for each exponent p ∈ (1, ∞) and each Muckenhoupt weight w in
Ap (∂Ω, σ), it follows that T # , originally acting on L p (∂Ω, w), further extends
uniquely to a linear, bounded operator, from the negative boundary Sobolev
p
space L−1 (∂Ω, w) into itself which, in fact, is the transpose of T acting on

L1 (∂Ω, w ) where p := (1 − 1/p)−1 ∈ (1, ∞) is the conjugate exponent of p
p

and w  := w 1−p ∈ Ap (∂Ω, σ) is the conjugate weight of w (cf. [68, item (2) in
Lemma 7.7.1]).
(9) Consider the following modified version of the generalized double layer operator

in (5.1.4) acting on each function3 f ∈ L 1 ∂Ω, 1+σ(x)
|x | n according to
∫  

Tmod f (x) :=  − y) − k(−y)
ν(y), k(x  · 1Rn \B(0,1) (y) f (y) dσ(y) (5.1.44)
∂Ω

for all x ∈ Ω. Then the operator Tmod is meaningfully defined, and is compatible
with T from (5.1.4) in the sense that for each function f belonging to the smaller

space L 1 ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular, for each function f ∈ L p (∂Ω, σ)
with p ∈ [1, ∞)) the difference

C f := Tmod f − T f is a constant in Ω. (5.1.45)

As a consequence,
 σ(x) 
∇Tmod f = ∇T f in Ω for each f ∈ L 1 ∂Ω, . (5.1.46)
1 + |x| n−1
In addition, at each point x ∈ Ω one may express
∫
  
∂ Tmod f (x) =  − y) f (y) dσ(y)
ν(y), (∂ k)(x
∂Ω
(5.1.47)
1 σ(y)
for each ∈ {1, . . . , n} and f ∈ L ∂Ω, 1+ |y | n .

3 The reader is alerted to the change in power (from n − 1 to n) for the weight intervening in
 σ(x)  σ(x)
L 1 ∂Ω, 1+| x | n , compared with L ∂Ω, 1+| x | n−1 . In particular, the former space is more inclusive
1

than the latter one.

5.1 Theory of Generalized Double Layers 539

Moreover,

Tmod maps constant functions on ∂Ω into constant functions in the set Ω.

(5.1.48)
In fact, if in place of (5.1.44) one now considers the modified operator acting

on each given f ∈ L 1 ∂Ω, 1+σ(x)
|x | n according to
∫  

mod f (x) :=
T  − y) − k(−y)(1
ν(y), k(x  − ψ(y)) f (y) dσ(y) (5.1.49)
∂Ω

for all x ∈ Ω, where ψ ∈ 𝒞∞ c (R ) is a fixed function satisfying ψ ≡ 1 near

n

0 ∈ R , then
n

∫


Tmod 1 (x) = −ϑ − ∇ψ · k dL n for each x ∈ Ω. (5.1.50)
Ω

Next, given any function

 σ(x) 
f ∈ L 1 ∂Ω, with the property that
1 + |x| n
 (5.1.51)
σ(x) 
∂τ j  f ∈ L 1 ∂Ω, for each j, ∈ {1, . . . , n},
1 + |x| n−1
.p
(hence, in particular, for each f ∈ L1 (∂Ω, σ) with 1 < p < ∞) it follows that for
each index ∈ {1, . . . , n} and each point x ∈ Ω one has (using the summation
convention)
∫

∂ Tmod f (x) = k j (x − y)(∂τ j f )(y) dσ(y). (5.1.52)
∂Ω

As a consequence of (5.1.52) and [70, Theorem 2.5.1], given any aperture

parameter κ ∈ (0, ∞) it follows that the nontangential boundary trace
 κ−n.t.
∂ Tmod f ∂Ω exists at σ-a.e. point on ∂Ω,
(5.1.53)
for each function f as in (5.1.51) and each index ∈ {1, . . . , n}.

Another corollary of (5.1.52) and [70, (2.4.8)] is the fact that for each aperture
parameter κ > 0, each truncation parameter ε ∈ (0, ∞), and each exponent
p ∈ (1, ∞),

Nκε ∇(Tmod f ) ∈ Lloc (∂Ω, σ) for each function
p

 σ(x) 
f ∈ L 1 ∂Ω, with the property that (5.1.54)
1 + |x| n
 σ(x)  p
∂τ j k f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n}.
1 + |x| n−1
540 5 Generalized Double Layers in Uniformly Rectifiable Domains

In addition, as seen from (5.1.44) and [70, (2.5.32)], for each integrability
exponent p ∈ (1, ∞), each aperture parameter κ ∈ (0, ∞), and each truncation
parameter ε ∈ (0, ∞) one has

Nκε (Tmod f ) ∈ Lloc (∂Ω, σ) for each function

p

 σ(x) 
(5.1.55)
p
f ∈ L 1 ∂Ω, ∩ L (∂Ω, σ).
1 + |x| n loc

.p  
Finally, for each f ∈ H1 (∂Ω, σ) with p ∈ n−1 n , 1 , each ∈ {1, . . . , n}, and
each x ∈ Ω one has (using the summation convention)
  
⎧
⎪ (x − ·)  , ∂τ f if ∂Ω bounded,
 ⎪
⎨
⎪ k j ∂Ω  j
∂ Tmod f (x) =   (5.1.56)
⎪
⎪   
⎪ k j (x − ·) , ∂τ j f if ∂Ω unbounded,
⎩ ∂Ω

where the pairings in (5.1.56) are understood in the sense of [69, Theorem 4.6.1]
(keeping in mind that each tangential derivative ∂τ j f belongs to the Hardy space
H p (∂Ω, σ)). As
 a consequence
 of this and [70, (2.4.14)] in [70, Theorem 2.4.1],
for each p ∈ n−1n , 1 and each aperture parameter κ ∈ (0, ∞) there exists some
finite constant C >.p 0, depending only on ∂Ω, k, n, κ, and p, such that for each
distribution f ∈ H1 (∂Ω, σ) one has

Nκ (∇Tmod f ) L p (∂Ω,σ)
≤ C f H. p (∂Ω,σ) . (5.1.57)
1

Also, as a consequence of (5.1.56) and [70, Corollary 2.5.4],

.p  
for each f ∈ H1 (∂Ω, σ) with p ∈ n−1 n , 1 , the nontangential
κ−n.t. (5.1.58)
trace (∇T f )
mod ∂Ω
exists (in Cn ) at σ-a.e. point on ∂Ω.

(10) For each α ∈ (0, 1) there exists a constant C ∈ (0, ∞) with the property that
  
sup dist(x, ∂Ω)1−α ∇ Tmod f (x) ≤ C f 𝒞. α (∂Ω) (5.1.59)
x ∈Ω
.
for every function f ∈ 𝒞α (∂Ω). Moreover,
if Ω ⊆ Rn is a uniform domain
. . property
with the  that ∂Ω is an Ahlfors
regular set then Tmod : 𝒞α (∂Ω) → 𝒞α Ω is a well-defined, linear,
and bounded operator for each given exponent α ∈ (0, 1),
(5.1.60)
whereas
5.1 Theory of Generalized Double Layers 541

if Ω ⊆ Rn is. an NTA domain

.  with an upper Ahlfors regular boundary
α (∂Ω) → 𝒞α Ω is a well-defined, linear, and bounded
then Tmod : 𝒞van van
operator for each given exponent α ∈ (0, 1),
(5.1.61)
where the homogeneous vanishing Hölder spaces are defined as in (A.0.48)
(with Σ := ∂Ω and Σ := Ω, respectively). Also, for each α ∈ (0, 1) and each
p ∈ (1,
. ∞) there exists some C ∈ (0, ∞) with the property that for each function
f ∈ 𝒞α (∂Ω) one has
 ∫  1/p
1  p
sup ∇T
mod f  dist(·, ∂Ω) p−1 dL n ≤ C f 𝒞. α (∂Ω)
x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,∞)
(5.1.62)
and
 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
 .α
≤ C dist f , 𝒞van (∂Ω) , (5.1.63)
. 
where the distance is measured in the space 𝒞α (∂Ω), · 𝒞. α (∂Ω) . As a corol-
. α if the function f actually belongs to the homogeneous vanishing Hölder space
lary,
𝒞van (∂Ω) for some α ∈ (0, 1), then for each p ∈ (1, ∞) one has
 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n = 0.
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
(5.1.64)
(11) Let Tmod be the modified version of the singular integral operator (5.1.5), acting

on each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n according to4
∫
 
Tmod f (x) := lim+ ν(y), kε (x − y) − k1 (−y) f (y) dσ(y) (5.1.65)
ε→0
∂Ω

at σ-a.e. x ∈ ∂Ω, where

kε := k · 1Rn \B(0,ε) for each ε > 0. (5.1.66)

Then Tmod is compatible with T from (5.1.5), in the sense that

4 alternative descriptions of Tmod are seen from [70, Proposition 2.3.3]

542 5 Generalized Double Layers in Uniformly Rectifiable Domains

for each function f belonging to the space L 1 ∂Ω, 1+σ(x)|x | n−1
(hence,
in particular, for each f ∈ L (∂Ω, σ) with p ∈ [1, ∞)) the d-
p
(5.1.67)
ifference C f := Tmod f − T f is a constant function on ∂Ω with
|C f | ≤ C∂Ω, k · f L 1 (∂Ω, σ(x) ) for some finite C∂Ω, k > 0.
1+| x | n−1


Also, for each aperture parameter κ > 0 and each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n
the following jump-formula holds at σ-a.e. point on ∂Ω:
κ−n.t. ϑ

Tmod f  = − f + Tmod f . (5.1.68)
∂Ω 2
.p
In particular, (5.1.68) holds for each f ∈ L1 (∂Ω, σ) with 1 < p < ∞. As a
consequence of (5.1.68) and (5.1.48),

Tmod maps constant functions on ∂Ω

(5.1.69)
into constant functions on ∂Ω.

Moreover,

if p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 then given any functions
  ∫
p σ(x) p
f ∈ Lloc (∂Ω, σ) ∩ L ∂Ω, 1+ |x | n and g ∈ Lcomp (∂Ω, σ) with
1
g dσ = 0,
∂Ω
∫ ∫
it follows that |Tmod f ||g| dσ < +∞, | f ||T # g| dσ < +∞,
∂Ω ∂Ω
∫ ∫
and (Tmod f )g dσ = f (T # g) dσ.
∂Ω ∂Ω
(5.1.70)

Finally, if p ∈ (1, ∞) then for each function


f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) such that
 (5.1.71)
f ∈ L 1 ∂Ω, 1+σ(x)
p
∂τ j k |x | n−1
∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n}
.p
(hence for each f ∈ L1 (∂Ω, σ) if Ω satisfies a two-sided local John condition)
it follows that
p
Tmod f ∈ L1,loc (∂Ω, σ), (5.1.72)
and for each pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.1.19))
     
∂τr s Tmod f = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f ) (5.1.73)

at σ-a.e. point on ∂Ω.

5.1 Theory of Generalized Double Layers 543

(12) For each p ∈ (1, ∞) there exists a constant C ∈ (0,

 ∞) with
 p the property that for
each function f ∈ BMO(∂Ω, σ) the measure5 ∇ Tmod f  dist(·, ∂Ω) p−1 dL n is
Carleson in Ω in the quantitative sense that
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω

≤C f
p
. . (5.1.74)
BMO(∂Ω,σ)

In particular, corresponding to p = 2, it follows that6

  2
∇ T f  dist(·, ∂Ω) dL n is a Carleson measure in Ω,
mod
(5.1.75)
for each function f ∈ BMO(∂Ω, σ).

Moreover, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
that for each function f ∈ BMO(∂Ω, σ) one has
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω
r ∈(0,R)

≤ C dist f , VMO(∂Ω, σ) (5.1.76)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ).
As a corollary,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
mod

for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞)

(5.1.77)
and, corresponding to p = 2,
  2
∇ T f  dist(·, ∂Ω) dL n is a vanishing Carleson measure in Ω,
mod
(5.1.78)
for each function f ∈ VMO(∂Ω, σ).

(13) Make the additional assumption that ∂Ω is bounded. Then all properties listed in
items (9)-(12) above are valid for the operator T , as originally defined in (5.1.4),
in place of its modified version Tmod . In particular, for each p ∈ (1, ∞) there exists
a constant C ∈ (0, ∞) with the property that for each function f ∈ BMO(∂Ω, σ)
one has

5 i.e., the Littlewood-Paley measure associated with f via the modified generalized double layer
potential operator Tmod
 2
6 it is natural to refer to  ∇Tmod f  dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated with
f via the operator Tmod
544 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
p
≤C f dtBMO(∂Ω,σ)
, (5.1.79)

and
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤ C dist f , VMO(∂Ω, σ) (5.1.80)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ).
In particular,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).
(5.1.81)
(14) The operators

Tmod : BMO(∂Ω, σ) −→ BMO(∂Ω, σ), (5.1.82)

Tmod : VMO(∂Ω, σ) −→ VMO(∂Ω, σ), (5.1.83)

are well defined, linear, and bounded. In particular, (5.1.82)-(5.1.83) together

with (5.1.69) imply that both
   
Tmod : BMO(∂Ω, σ) ∼ −→ BMO(∂Ω, σ) ∼ defined as
    (5.1.84)
Tmod [ f ] := Tmod f for each function f ∈ BMO(∂Ω, σ),

and
   
Tmod : VMO(∂Ω, σ) ∼ −→ VMO(∂Ω, σ) ∼ defined as
    (5.1.85)
Tmod [ f ] := Tmod f for each function f ∈ VMO(∂Ω, σ),

are well-defined, linear, and bounded operators. Also, if ∂Ω is unbounded one

has
    
Tmod f , g = [ f ], T # g , ∀ f ∈ BMO(∂Ω, σ), ∀g ∈ H 1 (∂Ω, σ), (5.1.86)

where ·, · denotes the duality bracket between the John-Nirenberg space of

functions of bounded mean oscillations on ∂Ω, modulo constants, and the Hardy
space H 1 on ∂Ω (cf. [69, Theorem 4.6.1]), and where T # is presently considered
as in (5.1.22) with p := 1. In addition, if ∂Ω is unbounded, the operators

Tmod : CMO(∂Ω, σ) −→ CMO(∂Ω, σ), (5.1.87)

5.1 Theory of Generalized Double Layers 545

and
   
Tmod : CMO(∂Ω, σ) ∼−→ CMO(∂Ω, σ) ∼ defined as
    (5.1.88)
Tmod [ f ] := Tmod f for each function f ∈ CMO(∂Ω, σ),

are also well defined, linear, and bounded.

In the case when ∂Ω is bounded, a scenario in which one has

BMO(∂Ω, σ) ⊆ L p (∂Ω, σ), (5.1.89)
0<p<∞

the operator T acting on the Lebesgue scale L p (∂Ω, σ) with p ∈ (1, ∞) (cf.
(5.1.17)) has BMO(∂Ω, σ) as an invariant subspace, and its restriction

T : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) (5.1.90)

is a well-defined, linear, and bounded operator, which satisfies

   
T f , g = f , T # g , ∀ f ∈ BMO(∂Ω, σ), ∀g ∈ H 1 (∂Ω, σ) (5.1.91)

where ·, · denotes the duality bracket between the John-Nirenberg space of

functions of bounded mean oscillations on ∂Ω and the Hardy space H 1 on ∂Ω
(cf. [69, Theorem 4.6.1]) and where T # is presently considered as in (5.1.22)
with p := 1.
Moreover, when ∂Ω is bounded the operator T acting on BMO(∂Ω, σ) (cf.
(5.1.90)) has VMO(∂Ω, σ) as an invariant subspace, hence its restriction

T : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (5.1.92)

is a well-defined, linear, and bounded operator.

 1
(15) Fix p ∈ n−1n , 1 and set α := (n − 1) p − 1 ∈ (0, 1). Then the operators
. .
Tmod : 𝒞α (∂Ω) −→ 𝒞α (∂Ω), (5.1.93)
.
α α
.
Tmod : 𝒞van (∂Ω) −→ 𝒞van (∂Ω), (5.1.94)
.    . 
𝒞α (∂Ω) ∼  [ f ] → Tmod f ∈ 𝒞α (∂Ω) ∼, (5.1.95)
.α    .α 
𝒞van (∂Ω) ∼  [ f ] → Tmod f ∈ 𝒞van (∂Ω) ∼, (5.1.96)

are well defined, linear, bounded. As a consequence of (5.1.93)-(5.1.96), the

operators
  .  . 
Tmod : 𝒞α (∂Ω) ∼ −→ 𝒞α (∂Ω) ∼ defined as
    . (5.1.97)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞α (∂Ω),
546 5 Generalized Double Layers in Uniformly Rectifiable Domains

  .α  .α 
Tmod : 𝒞van (∂Ω) ∼ −→ 𝒞van (∂Ω) ∼ defined as
    .α (5.1.98)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞van (∂Ω),

are well defined, linear, and bounded. In addition, if ∂Ω is unbounded one has
     .
Tmod f , g = [ f ], T # g , ∀ f ∈ 𝒞α (∂Ω), ∀g ∈ H p (∂Ω, σ) (5.1.99)

with ·, · denoting the duality bracket between a homogeneous Hölder space

on ∂Ω, modulo constants, and the corresponding Hardy space on ∂Ω (cf. [69,
Theorem 4.6.1]), and where T # is presently considered as in (5.1.22). Finally,
in the case when ∂Ω is bounded, a scenario in which one has
.α . 
α
𝒞van (∂Ω) = 𝒞van (∂Ω) ⊆ 𝒞α (∂Ω) = 𝒞α (∂Ω) ⊆ L q (∂Ω, σ), (5.1.100)
0<q ≤∞

α (∂Ω)
the operator T acting on the Lebesgue scale as in (5.1.17) has both 𝒞van
α
and 𝒞 (∂Ω) as invariant subspaces, its restrictions
α α
T : 𝒞van (∂Ω) −→ 𝒞van (∂Ω) (5.1.101)

and
T : 𝒞α (∂Ω) −→ 𝒞α (∂Ω) (5.1.102)
are well-defined, linear, bounded operators, and the latter operator satisfies
   
T f , g = f , T # g , ∀ f ∈ 𝒞α (∂Ω), ∀g ∈ H p (∂Ω, σ) (5.1.103)

with ·, · denoting the duality bracket between Hölder and Hardy spaces on ∂Ω
(cf. [69, Theorem 4.6.1]) and with T # considered as in (5.1.22).
(16) Select p, q ∈ (1, ∞) with 1/p + 1/q = 1 along with λ ∈ (0, n − 1). Then the
operator T from (5.1.17) has the inhomogeneous Morrey-Campanato space
L p,λ (∂Ω, σ) (defined as in (A.0.119) with Σ := ∂Ω) as an invariant subspace,
and
   
T : L p,λ (∂Ω, σ), · L p, λ (∂Ω,σ) −→ L p,λ (∂Ω, σ), · L p, λ (∂Ω,σ) (5.1.104)

is a linear and bounded mapping. Moreover, if Tmod is the modified version of the
singular integral operator T defined in (5.1.65), the assignment
.  . 
[Tmod ] : L p,λ (∂Ω, σ) ∼ −→ L p,λ (∂Ω, σ) ∼
. (5.1.105)
[Tmod ][ f ] := [Tmod f ] for each f ∈ L p,λ (∂Ω, σ)

is well defined, linear and bounded. In particular,

5.1 Theory of Generalized Double Layers 547
. p,λ . p,λ
Tmod : L (∂Ω, σ) −→ L (∂Ω, σ) is well defined, lin-
ear, and there exists some constant C ∈ (0, ∞) with the
property that Tmod f L. p, λ (∂Ω,σ) ≤ C f L. p, λ (∂Ω,σ) for each (5.1.106)
. p,λ
f ∈L (∂Ω, σ).
Finally,

the (real) transpose #

  of the operator T from (5.1.32) is,
respectively, Tmod from (5.1.105) if ∂Ω is unbounded, (5.1.107)
and T from (5.1.104) if ∂Ω is bounded,
.
that is, for each f ∈ L p,λ (∂Ω, σ) and g ∈ ℋq,λ (∂Ω, σ) one has
    
Tmod f , g = [ f ], T # g if ∂Ω is unbounded, (5.1.108)
   
T f , g = f , T # g if ∂Ω is bounded, (5.1.109)

where ·, · denotes the duality bracket between the Morrey-Campanato space

and its pre-dual (cf. [69, (6.1.25)]).
(17) Fix q ∈ (1, ∞) and η ∈ (0, 1) and recall the Calderón spaces defined in (3.1.10)-
(3.1.15). Then there exists a constant C ∈ (0, ∞) with the property that, with the
maximal operator defined as in (3.1.1) (with Σ := ∂Ω), one has
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tmod f (5.1.110)
 
for every function f ∈ L 1 ∂Ω, 1+σ(x)|x | n . As a consequence, given any p ∈ [1, ∞),
it follows that
.p  .p 
[Tmod ] : Cq,η (∂Ω, σ) ∼ −→ Cq,η (∂Ω, σ) ∼
  .p (5.1.111)
defined by [Tmod ][ f ] := Tmod f for each f ∈ Cq,η (∂Ω, σ)

is a well-defined, linear, and bounded operator. Moreover,

 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tf (5.1.112)

for every function f ∈ L p (∂Ω, σ) with p ∈ [1, ∞). Finally, whenever p ∈ (1, ∞)
it follows that
p p
T : Cq,η (∂Ω, σ) −→ Cq,η (∂Ω, σ) (5.1.113)
is a well-defined, linear, and bounded operator.
(18) If
1
∈ {1, . . . , N }, n−1
n < p ≤ ∞ and (n − 1) p −1 + < s < 1, (5.1.114)
548 5 Generalized Double Layers in Uniformly Rectifiable Domains

 , p, s, θ) > 0
then for each θ ∈ (0, 1) there exists a finite constant C = C(Ω, k,
such that
− 1 −s  
δ∂Ω p ∇ T f  ≤C f p, p (5.1.115)
mod ,θ L p (Ω, L n ) Bs (∂Ω,σ)

p, p
for all f ∈ Bs (∂Ω,
 σ), with
 the understanding
 that when p > 1 the solid
maximal function ∇ Tmod f  ,θ is replaced by ∇ Tmod f .
(19) Make the additional assumption that ∂Ω is compact. Then the operator T,
originally acting on Lebesgue spaces on ∂Ω (cf. (5.1.17)), extends uniquely to
linear and bounded mappings
p,q p,q
T : Bs (∂Ω, σ) −→ Bs (∂Ω, σ),
 n−1   (5.1.116)
p ∈ n , ∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
p,q p,q
T : Fs (∂Ω, σ) −→ Fs (∂Ω, σ),
 n−1    1 (5.1.117)
p ∈ n , ∞ , q ∈ n−1 n ,∞ , (n − 1) min{p,q } −1 + < s < 1.

Moreover, various choices of the exponents yield operators which are compatible
with one another. In addition, the operator T # , originally considered acting on
Lebesgue spaces on ∂Ω (cf. (5.1.17)) further extends, in a unique fashion, to
linear and bounded mappings
p,q p,q
T # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (5.1.118)
with s ∈ (0, 1), p ∈ n−1n−s , ∞], q ∈ (0, ∞],

and
p,q p,q
T # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ),
  (5.1.119)
with s ∈ (0, 1), p ∈ n−1n−s , ∞), q ∈ n−1
n−s , ∞].

Again, various choices of the parameters p, q, s yield operators which are com-
patible with one another. In all cases,

the operator norms of T from (5.1.116)-(5.1.117)

 as well as T # from
α
(5.1.118)-(5.1.119) are dominated by C |α | ≤ N supS n−1 |∂ k | with
C ∈ (0, ∞) depending only on n, p, q, s, and the UR character of ∂Ω.
(5.1.120)
Finally, if p, q, p, q  ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q  and s ∈ (0, 1),
then
5.1 Theory of Generalized Double Layers 549
 #   
B−s (∂Ω,σ) T f , g B p , q  (∂Ω,σ) = f, Tg p , q 
p, q p, q
s
B−s (∂Ω,σ) Bs (∂Ω,σ)
(5.1.121)
p,q p,q
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ),

and
 #   
p, q
F−s (∂Ω,σ) T f , g F p, q (∂Ω,σ) = F−s
p, q
(∂Ω,σ) f , T g F p , q  (∂Ω,σ)
s s
(5.1.122)
p,q p,q
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ).

(20) Strengthen the original hypotheses on the underlying domain by assuming that
Ω is a UR domain satisfying a local John condition. Also, fix some p ∈ (1, ∞).
Then the operator
.p p .p p
Tmod : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) −→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) (5.1.123)

is well defined, linear, and bounded, when the spaces involved are endowed with
.p
the semi-norm (A.0.128). Moreover, formula (5.1.73) holds for each function
p
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ).
Finally, if it is assumed that Ω ⊆ Rn is an NTA domain with an upper Ahlfors
regular boundary7 then for each integrability exponent p ∈ (1, ∞) the operator
.p .p
Tmod : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.1.124)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). In this case, it follows from (5.1.124) and (5.1.69) that
for each p ∈ (1, ∞) the operator
  .p  .p 
Tmod : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
    .p  .p (5.1.125)
Tmod [ f ] := Tmod f ∈ L1 (∂Ω, σ) ∼ for all f ∈ L1 (∂Ω, σ),

is well defined, linear, and bounded, when all quotient spaces are endowed with
the natural semi-norm8 introduced in [69, (11.5.138)].
(21) Assume Ω is a UR domain satisfying a local John condition9, and fix an inte-
grability exponent p ∈ (1, ∞) along with a parameter λ ∈ (0, n − 1). Then the
operator
. p,λ . p,λ
Tmod : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) (5.1.126)

7 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
8 [69, Proposition 11.5.14] tells us that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
9 for example, this is the case if Ω ⊆ R n is an NTA domain with an upper Ahlfors regular boundary
550 5 Generalized Double Layers in Uniformly Rectifiable Domains

is well defined, linear, and bounded, when the spaces involved are en-
dowed. with the semi-norm (A.0.158). Also, (5.1.73) holds for each function
p,λ
f ∈ M1 (∂Ω, σ). As a consequence of (5.1.126) and (5.1.69), the operator
  . p,λ  . p,λ 
Tmod : M1 (∂Ω, σ) ∼ −→ M1 (∂Ω, σ) ∼ defined as
    . p,λ  . p,λ
Tmod [ f ] := Tmod f ∈ M1 (∂Ω, σ) ∼ for all f ∈ M1 (∂Ω, σ),
(5.1.127)
is well defined, linear, and bounded, when all quotient spaces are endowed with
the semi-norm10 introduced in [69, (11.13.51)].
. p,λ are valid for vanishing Morrey-based homoge-
In fact, analogous properties
neous Sobolev spaces M1 (∂Ω, σ) (cf. [69, Definition 11.13.15], or (A.0.159)-
(A.0.160)) in place of Morrey-based homogeneous Sobolev spaces. Final-
. q,λsimilar properties are valid for block-based homogeneous Sobolev spaces
ly,
B1 (∂Ω, σ) with q ∈ (1, ∞) in place of Morrey-based homogeneous Sobolev
spaces.
(22) Strengthen the hypotheses on the underlying domain by assuming that Ω ⊆ Rn
is an NTA domain with an upper Ahlfors regular boundary11. Then the modified
boundary-to-boundary operator Tmod , originally defined as in (5.1.65), induces
a linear and bounded mapping
.p .p  
Tmod : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) for each p ∈ n−1n ,1 . (5.1.128)

Before presenting the proof of Theorem 5.1.1 we shall discuss some relevant
examples.

Example 5.1.2 (The Cauchy Operator in the Plane) Work in the two-dimensional
setting, and identify R2 ≡ C. Let Ω ⊆ C be a UR domain, set σ := H 1 ∂Ω, and
denote by ν = (ν1, ν2 ) ≡ ν1 + iν2 the geometric measure theoretic outward unit
normal to Ω. Also, consider
 
 := − 1 1 , i for each z ∈ C \ {0}.
k(z) (5.1.129)
2π z z
This is a smooth vector-valued function which is odd, positive homogeneous of degree
−1, and satisfies
 1 1 1 i 1 1
div k (z) = −∂x − ∂y = − (∂x + i∂y )
2π z 2π z 2π z
1 1 
=− ∂ = 0 for each z ∈ C \ {0}, (5.1.130)
π z

10 From [69, Proposition 11.13.10] it is known that this semi-norm is actually a genuine norm
if Ω ⊆ R n is an open set satisfying a two-sided local John condition and whose boundary is an
unbounded Ahlfors regular set
11 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
5.1 Theory of Generalized Double Layers 551

where ∂ := 12 (∂x + i∂y ) is the Cauchy-Riemann operator in the plane. Hence all
hypotheses of Theorem 5.1.1 are satisfied in this case, and (5.1.3) presently becomes
∫ ∫ 1 i
 1
ϑ= ω, k(ω) dH 1 (ω) = − ω1 + ω2 dH 1 (ω)
S1 2π S 1 ω ω
∫ 1 ∫
1 1 1
=− (ω1 + iω2 ) dH 1 (ω) = − ω dH 1 (ω)
2π S 1 ω 2π S 1 ω
∫
1
=− dH 1 (ω) = −1. (5.1.131)
2π S 1
Note since for each z ∈ Ω we have
     i 
 − ζ) = − 1 ν1 (ζ) 1
ν(ζ), k(z −
1
ν2 (ζ)
2π z−ζ 2π z−ζ
1   1 
=− ν1 (ζ) + iν2 (ζ)
2π z−ζ
1 ν(ζ)
= for σ-a.e. ζ ∈ ∂Ω, (5.1.132)
2π ζ − z

and since iν(ζ) dσ(ζ) = dζ on ∂Ω (cf. (A.0.62)), the operator T constructed as

in (5.1.4) for k as in (5.1.129) becomes precisely the boundary-to-domain Cauchy
 σ(ζ )
integral operator, acting on each function f ∈ L 1 ∂Ω, 1+ |ζ | according to (cf.
(1.6.35)) ∫
1 f (ζ)
𝒞 f (z) = dζ for all z ∈ Ω. (5.1.133)
2πi ∂Ω ζ − z
Also, the principal-value singular integral operator T from (5.1.5) presently becomes
∫
1 f (ζ)
C f (z) := lim+ dζ for σ-a.e. z ∈ ∂Ω, (5.1.134)
ε→0 2πi ζ−z
ζ ∈∂Ω
|z−ζ |>ε

i.e., the principal-value Cauchy integral operator (1.6.36), while T # from (5.1.6)
currently acquires the format
 ∫ 
1 f (ζ)
C f (z) := −ν(z) lim+
#
dσ(ζ) for σ-a.e. z ∈ ∂Ω, (5.1.135)
ε→0 2π ζ−z
ζ ∈∂Ω
|z−ζ |>ε

i.e., T # is the transpose Cauchy singular integral operator from (1.6.37). In partic-
ular, Theorem 5.1.1 specialized to this setting becomes compatible with the results
established earlier in Proposition 1.6.7.
552 5 Generalized Double Layers in Uniformly Rectifiable Domains

Example 5.1.3 (The Cauchy-Clifford Operator) Let Ω ⊆ Rn be an arbitrary UR

domain (where n ∈ N with n ≥ 2). Abbreviate σ := H n−1 ∂Ω and denote by ν the
geometric measure theoretic outward unit normal to Ω. Recall the Clifford algebra
with n imaginary units C n from [68, §6.4] and consider the vector field with Clifford
algebra-valued components
 n
k = (k j )1≤ j ≤n ∈ 𝒞∞ (Rn \ {0}) ⊗ C n (5.1.136)

given by
1 x
k j (x) :=  e j for 1 ≤ j ≤ n and x ∈ Rn \ {0}. (5.1.137)
ωn−1 |x| n

In particular, k is odd and positive homogeneous of degree 1 − n in Rn \ {0}, and

satisfies (with the summation convention over repeated indices in effect)
 1  x 
div k (x) = (∂j k j )(x) = ∂j  ej
ωn−1 |x| n
1  x 
= DR = 0 in Rn \ {0}, (5.1.138)
ωn−1 |x| n
where DR is the version of the Dirac operator acting from the right (cf. (A.0.59)).
Also, (5.1.3) presently becomes
∫ ∫  ω 
 1
ϑ= ω, k(ω) dH (ω) =
n−1
ωj  e j dH n−1 (ω)
S n−1 ωn−1 S n−1 |ω| n
∫ ∫
1 ω 1
=  ω dH (ω) =
n−1
(−1) dH n−1 = −1, (5.1.139)
ωn−1 S n−1 |ω| n ωn−1 S n−1
on account of [68, (6.4.1)]. Given that for each x ∈ Ω we have
   x−y 
 − y) = ν j (y) 1
ν(y), k(x  e j
ωn−1 |x − y| n
1 x−y
=  ν(y) at σ-a.e. y ∈ ∂Ω, (5.1.140)
ωn−1 |x − y| n

the operator T from (5.1.4) for k as in (5.1.129) becomes precisely the boundary-to-

domain Cauchy-Clifford integral operator, acting on each f ∈ L 1 ∂Ω, 1+σ(y)|y | n−1
⊗C n
according to (cf. (A.0.53))
∫
1 x−y
C f (x) =  ν(y)  f (y) dσ(y) for all x ∈ Ω. (5.1.141)
ωn−1 |x − y| n
∂Ω

Moreover, the principal-value singular integral operator T from (5.1.5) presently

becomes
5.1 Theory of Generalized Double Layers 553
∫
1 x−y
C f (x) = lim+  ν(y)  f (y) dσ(y) (5.1.142)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, i.e., the principal-value Cauchy-Clifford integral operator

(A.0.54), while T # from (5.1.6) currently becomes
∫
1 x−y
C f (x) = − lim+
#
ν(x)   f (y) dσ(y) (5.1.143)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, i.e., T # is the transpose Cauchy-Clifford singular integral oper-
ator from (1.6.1). In particular, Theorem 5.1.1 specialized to this setting becomes
compatible with the results established earlier in Proposition 1.6.1, Theorem 2.1.5,
(2.1.179)-(2.1.192), (4.1.13)-(4.1.14), and (4.1.49)-(4.1.50).

Example 5.1.4 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate

σ := H n−1 ∂Ω. Consider a complex-valued function b ∈ 𝒞∞ (Rn \ {0}) such that
∇b is odd and positive homogeneous of degree 1 − n (this is the case if, for example,
b is even and positive homogeneous of degree 2 − n). For each given pair of indices
i, j ∈ {1, . . . , n} define the vector field
 n
ki j := (∂i b)e j − (∂j b)ei ∈ 𝒞∞ (Rn \ {0}) (5.1.144)

which is odd, positive homogeneous of degree 1 − n, and divergence-free in Rn \ {0}.

Note that (5.1.3) presently becomes
∫
ϑ= ω, ki j (ω) dH n−1 (ω)
S n−1
∫
! "
= ω j (∂i b)(ω) − ωi (∂j b)(ω) dH n−1 (ω)
S n−1
∫

= ∂τ j i b (ω) dH n−1 (ω) = 0, (5.1.145)
S n−1

thanks to (A.0.183) and [69, (11.1.7)]. The operators T , T, T # , associated as in

(5.1.4), (5.1.5), (5.1.6) with the vector field from (5.1.144) have been considered ear-
lier in (1.2.1), (1.2.2), (1.2.3). The results in Theorem 5.1.1 are then compatible with
those in Proposition 1.2.1, (2.1.159)-(2.1.168), [70, Corollary 2.4.2], Theorem 4.1.3,
and Theorem 4.1.7. In closing, we wish to note that
when Ω := R+n the (non-zero) principal-value generalized dou-
ble layer operators corresponding to the kernels (5.1.144) for
the choice b := EΔ , the standard fundamental solution for the (5.1.146)
Laplacian in Rn , coincide (up to normalization) with the family
of ordinary Riesz transforms on ∂Ω ≡ Rn−1 .
554 5 Generalized Double Layers in Uniformly Rectifiable Domains

Example 5.1.5 (The Harmonic Double Layer) Let Ω ⊆ Rn be an arbitrary UR

domain (where n ∈ N with n ≥ 2). Abbreviate σ := H n−1 ∂Ω and denote by ν the
geometric measure theoretic outward unit normal to Ω. If we consider the vector
field
 n
k = (k j )1≤ j ≤n ∈ 𝒞∞ (Rn \ {0}) (5.1.147)

given by

 1 x
k(x) := − for all x ∈ Rn \ {0}, (5.1.148)
ωn−1 |x| n

then k is odd, positive homogeneous of degree 1 − n, and divergence-free in Rn \ {0}.

Also, (5.1.3) presently becomes
∫ ∫
 1 ω, ω
ϑ= ω, k(ω) dH n−1 (ω) = − dH n−1 (ω) = −1. (5.1.149)
S n−1 ω n−1 S n−1 |ω| n

Given that for each x ∈ Ω we have

  1 ν(y), y − x
 − y) =
ν(y), k(x at σ-a.e. y ∈ ∂Ω, (5.1.150)
ωn−1 |x − y| n

the operator T from (5.1.4) for k as in (5.1.147)-(5.1.148) becomes precisely the

classical boundary-to-domain harmonic double layer operator, acting on each func-

tion f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
according to (cf. [70, (2.5.202)])
∫
1 ν(y), y − x
D f (x) = f (y) dσ(y) for all x ∈ Ω. (5.1.151)
ωn−1 ∂Ω |x − y| n
Moreover, the principal-value singular integral operator T from (5.1.5) presently
becomes
∫
1 ν(y), y − x
K f (x) = lim+ f (y) dσ(y) (5.1.152)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, i.e., the harmonic double layer operator [70, (2.5.203)], while T #
from (5.1.6) currently becomes
∫
1 ν(x), x − y
K f (x) = lim+
#
f (y) dσ(y) (5.1.153)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, i.e., T # is the transpose harmonic double layer operator from
(A.0.102). In particular, Theorem 5.1.1 is applicable to these integral operators.
5.1 Theory of Generalized Double Layers 555

Example 5.1.6 (Genuine Double Layers Associated with Second-Order System-

s) Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and abbreviate σ := H n−1 ∂Ω.
αβ
For some M ∈ N, consider a coefficient tensor A = ar s 1≤r,s ≤n with complex
1≤α,β ≤M
entries, with the property that the M × M homogeneous second-order system L = L A
associated with A in Rn as in (1.3.2) is weakly elliptic (in the sense of [70, (1.3.3) in
Definition 1.3.1]). Let E = (Eγβ )1≤γ,β ≤M be the matrix-valued fundamental solution
associated with L as in [70, Theorem 1.4.2].
For each fixed α, γ ∈ {1, . . . , M } define the vector field (recall that the summation
convention over repeated indices is presently in effect)
 βα  n
kαγ := − ar s ∂r Eγβ 1≤s ≤n ∈ 𝒞∞ (Rn \ {0}) (5.1.154)

which is odd and positive homogeneous of degree 1 − n (cf. [70, Theorem 1.4.2]),
and which satisfies
βα
div kαγ = −ar s ∂r ∂s Eγβ = 0 in Rn \ {0}, (5.1.155)

thanks to [70, (1.4.33)]. Also, (5.1.3) presently becomes

∫
ϑ= ω, kαγ (ω) dH n−1 (ω)
S n−1
∫
βα
=− ωs ar s (∂r Eγβ )(ω) dH n−1 (ω) = −δαγ (5.1.156)
S n−1

by virtue of the second equality in [70, (1.4.25)].

If Tαγ denotes the integral operator associated as in (5.1.4) with kαγ defined
in (5.1.154), then it becomes apparent from (1.3.18) that for each given function
   M
f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
we have

D f = Tαγ fα 1≤γ ≤M in Ω. (5.1.157)

Also, if Tαγ denotes the principal-value singular integral operator associated as in

(5.1.5) with kαγ defined in (5.1.154), then from (1.3.68) it follows that for each given
   M
function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 we have

K f = Tαγ fα 1≤γ ≤M at σ-a.e. point on ∂Ω, (5.1.158)

hence

K = Tαγ 1≤γ,α ≤M . (5.1.159)

αγ from (5.1.154) then, as seen

# is defined as in (5.1.6) in relation to k
Finally, if Tαγ
   M
from (1.3.72), for each f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
we have
556 5 Generalized Double Layers in Uniformly Rectifiable Domains
 #
K # f = Tαγ fγ 1≤α ≤M at σ-a.e. point on ∂Ω, (5.1.160)

thus,
 #
K # = Tαγ 1≤α,γ ≤M . (5.1.161)

As such, Theorem 5.1.1 is compatible with results established earlier for D, K,

K # (and their modified versions) in Theorem 1.5.1, Theorem 1.8.2, Theorem 2.1.1,
Corollary 2.1.2, Theorem 2.1.7, Corollary 2.1.9, Theorem 2.1.10, Corollary 2.1.13,
Theorem 3.1.1, Theorem 3.2.2, Theorem 3.3.1, Theorem 4.1.1, Corollary 4.1.4, and
Theorem 4.1.5.

Example 5.1.7 (Chord-Dot-Normal Singular Integral Operators) Let Ω ⊆ Rn be

an arbitrary UR domain (where n ∈ N with n ≥ 2). Abbreviate σ := H n−1 ∂Ω and
denote by ν the geometric measure theoretic outward unit normal to Ω. Consider a
function θ ∈ 𝒞∞ (Rn \ {0}) which is even and positive homogeneous of degree −n,
and define the vector field

 := z θ(z) for each z ∈ Rn \ {0}.

k(z) (5.1.162)
 n
Then k ∈ 𝒞∞ (Rn \ {0}) is odd and positive homogeneous of degree 1 − n. The
homogeneity property of θ entails Euler’s identity

z, (∇θ)(z) = −n θ(z) for each z ∈ Rn \ {0}. (5.1.163)

As such,

 = ∂j z j θ(z) = n θ(z) + z j (∂j θ)(z)
div k(z)

= n θ(z) + z, (∇θ)(z) = 0 for each z ∈ Rn \ {0}, (5.1.164)

which shows that k from (5.1.162) satisfies (5.1.2). In particular, Theorem 5.1.1
applies to the integral operators associated with kernels of the form (5.1.162). We
shall consider this class of integral operators in detail a little later, in Theorem 5.2.2.
For now, we wish to offer a couple of examples of this nature.
First, consider the case when, for some fixed pair of indices j, k ∈ {1, . . . , n} we
take θ := θ jk , where θ(z) = z j zk /|z| n+2 for each z ∈ Rn \ {0}. For this choice, the
operator T from (5.1.4) becomes12
∫
(x j − y j )(xk − yk )
Θ jk f (x) = ν(y), x − y f (y) dσ(y)
∂Ω |x − y| n+2 (5.1.165)
 
for each f ∈ L 1 ∂Ω, 1+σ(y)|y | n−1
and each x ∈ Ω.

12 this is relevant in relation to double layer potential operators for the Lamé and Stokes systems
5.1 Theory of Generalized Double Layers 557

Second, work when n = 2 under the identification R2 ≡ C. For each integers

j, k ∈ {1, . . . , n} consider
 the principal-value
 singular integral operator13 acting on
σ(ζ )
each function f ∈ L ∂Ω, 1+ |ζ | according to
1

∫
ν(z), z − ζ ν(ζ), z − ζ
U jk f (z) := lim+ f (ζ) dσ(ζ) (5.1.166)
ε→0 (z − ζ)3
y ∈∂Ω
|x−y |>ε

for σ-a.e. z ∈ ∂Ω. Then U jk may be expressed as a finite linear combination

(with coefficients scalar components of ν) of operators associated as in (5.1.5) with
vector-valued kernels of the form

C \ {0}  ξ −→ ξ (ξ1, ξ2 )/ξ 3 ∈ C2, ∈ {1, 2}. (5.1.167)

We are now ready to present the proof of Theorem 5.1.1.

Proof of Theorem 5.1.1 Before commencing the proof in earnest, wemake several
observations. First, in view of the decay of k at infinity, L 1 ∂Ω, 1+σ(x)
|x | n−1
is the largest
space guaranteeing  that the integral
 in (5.1.4) is absolutely convergent. Second, that
σ(x)
for each f ∈ L ∂Ω, 1+ |x | n−1 the limits in (5.1.5), (5.1.6) exist for σ-a.e. point
1

on ∂Ω is guaranteed by [70, (2.3.15)], (the last condition in) the definition of a

UR domain from [68, Definition 5.10.6], and [68, (5.6.32)]. Third, as noted in [68,
(4.5.46)] we have

div k = ϑδ in D (Rn ), (5.1.168)

where δ is Dirac’s distribution with mass at the origin in Rn . Fourth, each k j induces
(via integration against Schwartz functions) a tempered distribution, and recall from
[70, (2.5.7)] that each k#j is given by a continuous function in Rn \ {0}. As a
consequence of (5.1.168) we further obtain
1 1 $ ϑ
k#j (ξ)ξ j = ∂$
j k j (ξ) = div k(ξ) = for each ξ ∈ Rn \ {0}. (5.1.169)
i i i
After this preamble, we now turn to the actual proof in earnest.
Proof of claims in item (1): Denote by ν1, . . . , νn the scalar components of the geo-
metric measure theoretic outward unit normal to Ω, and fix an aperture parameter
κ > 0 along with a function f ∈ L ∂Ω, 1+σ(x)
1
|x | n−1
. At σ-a.e. x ∈ ∂Ω we may then
compute

13 which is related to the radiosity singular integral operator

558 5 Generalized Double Layers in Uniformly Rectifiable Domains
 κ−n.t.  1 

T f (x) = k#j ν(x) ν j (x) f (x) + (T f )(x)
∂Ω 2i
ϑ
=− f (x) + (T f )(x), (5.1.170)
2
by [70, (2.5.4)], (5.1.4), (5.1.5), and (5.1.169). This establishes (5.1.7). Likewise, at
κ−n.t.
σ-a.e. point x ∈ ∂Ω the nontangential boundary trace (W f )∂Ω exists and we have
(also keeping in mind that k is odd)
 κ−n.t.  1 

ν(x) · W f  (x) = k#j ν(x) ν j (x) f (x)
∂Ω 2i
∫
+ lim+  − y) f (y) dσ(y)
ν(x), k(x
ε→0
y ∈∂Ω
|x−y |>ε

ϑ
=− f (x) − (T # f )(x). (5.1.171)
2
This prove (5.1.10)-(5.1.11). Also, the claims in (5.1.9) are direct consequences of
(5.1.8), (5.1.1), (5.1.2).
Assume next that Ω is bounded. Fix an aperture parameter κ > 0 and pick an
arbitrary point x ∈ Ω. Then [68, Theorem 1.4.1] applies to the vector field defined

at L n -a.e. y ∈ Ω as F(y)  − y) and, on account of (5.1.168), the Divergence
:= k(x
Formula [68, (1.4.6)] gives
∫
 κ−n.t. 
(T 1)(x) = ν · F ∂Ω dσ = (𝒞∞ (Ω))∗ divF, 1 𝒞∞ (Ω)
b
∂Ω b


= − (𝒞∞ (Ω))∗ ϑδx, 1 𝒞∞
b
(Ω) = −ϑ. (5.1.172)
b

Hence, T 1 ≡ −ϑ in Ω. With this in hand, (5.1.170) then gives that T1 = − ϑ2 at σ-a.e.

point on ∂Ω. In the case when Ω is an exterior domain, the same type of reasoning
applies except that now we also need to take into account the contribution of F at
infinity. According to [68, Proposition 4.7.1], the latter quantity may be expressed
as the limit
∫ y − x

[F]∞ = lim  − y) dH n−1 (y)
· k(x
R→∞ |x−y |=R R
∫
= 
ω, k(ω) dH n−1 (ω) = −ϑ, (5.1.173)
S n−1

where we have changed variables ω := (y − x)/R, used the homogeneity and parity
 and also recalled (5.1.3). Granted this, the Divergence Formula [68, (1.4.5)]
of k,
permits us to compute
5.1 Theory of Generalized Double Layers 559
∫
 κ−n.t. 
(T 1)(x) = ν · F ∂Ω dσ = (𝒞∞ (Ω))∗ divF,
1 𝒞∞
b
(Ω)
∞
− [F]
∂Ω b


= − (𝒞∞ (Ω))∗ ϑδx, 1 𝒞∞
b
(Ω) + ϑ = 0. (5.1.174)
b

We therefore have T 1 ≡ 0 in Ω this time which, in concert with (5.1.170), shows that
now T1 = + ϑ2 at σ-a.e. point on ∂Ω. This completes the proof of (5.1.12)-(5.1.13).
Proof of claims in item (2): The claim in (5.1.16) is dealt with much as in the proof
of (2.2.54), while all other results are consequences of items (3) and (5) in [70,
Theorem 2.4.1].
Proof of claims in item (3): All desired results are consequences of items (3) and (5)
in [70, Theorem 2.3.2].
 
Proof of claims in item (4): Fix two exponents, p ∈ n−1 n , 1 and q ∈ (1, ∞), then
pick an arbitrary function f ∈ H p (∂Ω, σ) ∩ L q (∂Ω, σ). Also, select an aperture
parameter κ > 0, and denote by I the identity operator. Then, on the one hand, in the
sense of distributions on ∂Ω we may write
 κ−n.t.  

ν •W f = ν · W f = − (ϑ/2)I + T # f , (5.1.175)
∂Ω

thanks to [69, Proposition 10.2.9] (whose applicability is ensured by (5.1.10) and

the fact that we currently have Nκ (W f ) ∈ L q (∂Ω, σ) ⊂ Lloc
1 (∂Ω, σ); cf. (5.1.14)),

and the jump-formula proved in (5.1.11) (bearing in mind (5.1.3)). On the other
hand, [69, Theorem 10.2.1] (whose applicability in the present with F := W f is
guaranteed by (5.1.9) and (5.1.15)) gives that

ν • W f ∈ H p (∂Ω, σ) and ν • W f H p (∂Ω,σ) ≤C f H p (∂Ω,σ) (5.1.176)

 p) ∈ (0, ∞) independent of f . From (5.1.175) and

for some constant C = C(∂Ω, k,
(5.1.176) we conclude that

T# f H p (∂Ω,σ) ≤C f H p (∂Ω,σ)
(5.1.177)
for each f ∈ H p (∂Ω, σ) ∩ L q (∂Ω, σ).

In view of the density of H p (∂Ω, σ) ∩ L q (∂Ω, σ) in H p (∂Ω, σ) (cf. the last claim
in [69, (4.4.114)]), we conclude that T # , originally acting on L q (∂Ω, σ), extends to
a linear and bounded operator from H p (∂Ω, σ) into itself. The fact that T # further
extends to the scale of Lorentz-based Hardy spaces H p,q (∂Ω, σ) as indicated in
(5.1.25) is then a consequence of what we have proved so far and the interpolation
results from [69, Theorem 4.3.1]. 
To justify the identity claimed in (5.1.28) fix p ∈ n−1 n , ∞ along with q ∈ (0, ∞]
and observe that the assignment
560 5 Generalized Double Layers in Uniformly Rectifiable Domains

H p,q (∂Ω, σ)  f −→ ν • W f ∈ H p,q (∂Ω, σ)

(5.1.178)
is well defined, linear, and bounded,

as may be seen from [69, Theorem 10.2.1], (5.1.9), and (5.1.15). From [69, Propo-
sition 10.2.9] and (5.1.11) we also know that (5.1.175) continues to hold for each
f ∈ H p,q (∂Ω, σ) ∩ L 2 (∂Ω, σ). Based on this, (5.1.178), (5.1.25), and [69, Lem-
ma 4.3.3] we may then conclude that (5.1.28) holds whenever f ∈ H p,q (∂Ω, σ) with
q < 1. Finally, that the end-point q = ∞ may also be allowed follows from what we
have just proved and the fact that H p,∞ (∂Ω, σ) embeds into H p0 (∂Ω, σ) + H p1 (∂Ω, σ)
whenever n−1n < p0 < p < p1 < ∞ (cf. [69, (4.3.146)]).
To prove the formula claimed in (5.1.26), at σ-a.e. point xo ∈ ∂Ω we write, in

the sense of distributions (i.e., in Lipc (∂Ω) ),

ϑ    
 o − ·)
− δxo + P.V. ν, k(x  − xo ) = ν • Wδxo
= ν • k(·
2 ∂Ω

ϑ
= − δxo − T # δxo , (5.1.179)
2

thanks to [69, (11.9.30)] and the fact that k is odd, (5.1.16) with f := δxo , as well as
(5.1.28) written for f := δxo ∈ H 1,∞ (∂Ω, σ) (cf. [69, Example 4.2.4]). Now, (5.1.26)
readily follows from (5.1.179). Also, (5.1.27) is a direct consequence of (5.1.26) and
[69, (4.2.17)].
It is instructive to give a direct proof
 ofthe claims made in relation to the operator
in (5.1.22). To this end, fix p ∈ n−1 n , 1 . As in the proof of Theorem 2.1.1, the
strategy is to show that T # maps atoms into a fixed multiple of molecules for the
Hardy space in question. Concretely, pick q ∈ (1, ∞) and consider a (p, q)-atom a on
∂Ω. Recall from [69, (4.4.167)-(4.4.168)] that this means that a : ∂Ω → C is some
σ-measurable  function with the property that there exist a point xo ∈ ∂Ω and some
number r ∈ 0, 2 diam(∂Ω) such that

supp a ⊆ B(xo, r) ∩ ∂Ω,

 1/q−1/p
a L q (∂Ω,σ) ≤ σ B(xo, r) ∩ ∂Ω , (5.1.180)
∫
a dσ = 0.
∂Ω

In addition, in the case when ∂Ω is compact, we agree that a constant of absolute

value ≤ σ(∂Ω)−1/p is also considered, by definition, to be a (p, q)-atom on ∂Ω. The
claim that we make is that
m := T # a is a fixed multiple of a (p, q, ε)-molecule on ∂Ω
(5.1.181)
(i.e., a function as in [69, Definition 4.5.1]), with ε := n−1
1
.

To prove this, first note that according to the current item (3) the function m
is meaningfully defined and belongs to the space L q (∂Ω, σ). In fact, thanks to the
current item (3) and (5.1.180), we have
5.1 Theory of Generalized Double Layers 561

m L q (∂Ω,σ) = T #a L q (∂Ω,σ) ≤C a L q (∂Ω,σ)

 1/q−1/p
≤ Cσ B(xo, r) ∩ ∂Ω , (5.1.182)

 q) > 0 independent of the atom.

for some finite constant C = C(∂Ω, k,
To study the decay of m, observe that based on (5.1.180) and the Mean Value
Theorem, for each x ∈ ∂Ω \ B(xo, 2r), we may estimate
∫
|m(x)| ≤ C  − y) − k(x
| k(x  − xo )||a(y)| dσ(y)
B(x o ,r)∩∂Ω
r  1−1/p
≤C σ B(xo, r) ∩ ∂Ω . (5.1.183)
|x − xo | n
To proceed, for each ∈ N define the boundary annulus
 
A (xo, r) := B(xo, 2 +1 r) \ B(xo, 2 r) ∩ ∂Ω. (5.1.184)

We may then rely on (5.1.183) and the Ahlfors regularity of ∂Ω to obtain to that, for
each ∈ N,
∫  1/q
|m| q dσ
A (x o ,r)
r  1−1/p 
σ B(xo, 2 +1 r) ∩ ∂Ω
1/q
≤C σ B(xo, r) ∩ ∂Ω
(2 r) n


≤ C2 (n−1)[1/q−1−1/(n−1)] σ B(xo, r) ∩ ∂Ω
1/q−1/p
(5.1.185)

 > 0 independent of the atom. In particular,

for some finite constant C = C(∂Ω, k)
from (5.1.182) and (5.1.185) we conclude that
∫
|m| dσ < +∞. (5.1.186)
∂Ω

In view of (5.1.182), (5.1.185), and [69, Definition 4.5.1] the claim in (5.1.181)
follows as soon as we check that
∫
m dσ = 0. (5.1.187)
∂Ω

To this end, fix some background parameter κ > 0 and bring in the vector field

F := Wa in Ω. (5.1.188)

Thanks to (5.1.9) we have

 n
F ∈ 𝒞 N (Ω) and divF = 0 in Ω. (5.1.189)
562 5 Generalized Double Layers in Uniformly Rectifiable Domains

Also, (5.1.11) gives

κ−n.t.
F∂Ω exists at σ-a.e. point on ∂Ω (5.1.190)

and
 κ−n.t. ϑ ϑ
ν · F∂Ω = − a − T # a = − a − m at σ-a.e. point on ∂Ω. (5.1.191)
2 2
Since, as is apparent from (5.1.180), the function a : ∂Ω → C is a multiple of an
(1, q)-atom on ∂Ω, we may invoke (5.1.15) (with p = q = 1) to conclude that

Nκ F ∈ L 1 (∂Ω, σ). (5.1.192)

Finally, the vanishing moment property of the atom (cf. the last line in (5.1.180))
together with (5.1.188) imply that
in the case when Ω is an exterior domain we have

F(x) = O(|x| −n ) as x ∈ Ω satisfies |x| → ∞; hence, (5.1.193)
the pointwise decay property [68, (1.2.9)] is satisfied.
Collectively, (5.1.189), (5.1.190), (5.1.192), (5.1.193) guarantee the validity of
the Divergence Formula [68, (1.2.2)] which, in light of (5.1.189) and (5.1.191),
presently gives
∫ ∫
 κ−n.t.
0= divF dL n = ν · F ∂Ω dσ
Ω ∂Ω
∫ ∫ ∫
ϑ
=− a dσ − m dσ = − m dσ, (5.1.194)
2 ∂Ω ∂Ω ∂Ω

bearing in mind the vanishing

∫ moment property in (5.1.180). Having established
this, we then conclude that ∂Ω m dσ = 0, finishing the proof of (5.1.187).
Let us record our progress: from (5.1.182), (5.1.185), and (5.1.187) we deduce
that, whenever a is as in (5.1.180), the function m := T # a is a fixed multiple of
a molecule for the Hardy space H p (∂Ω, σ), in the sense of [69, Definition 4.5.1].
Granted this, it follows from [69, (4.5.6)] that

m ∈ H p (∂Ω, σ) and m H p (∂Ω,σ)

 p) ∈ (0, ∞).
≤ C(∂Ω, k, (5.1.195)

Moreover, in the case when ∂Ω is compact and the (p, q)-atom a is some con-
stant function on ∂Ω, of absolute value ≤ σ(∂Ω)−1/p , it follows from the cur-
rent item (3) that the function m := T # a belongs to L 2 (∂Ω, σ) and satisfies
m L 2 (∂Ω,σ) ≤ C(∂Ω, k, p) ∈ (0, ∞). Keeping in mind that, in the present set-
ting, L (∂Ω, σ) embeds continuously into H p (∂Ω, σ) (cf. [69, (4.2.13)]), we deduce
2

that the conclusions in (5.1.195) are valid in this case as well. Having established
(5.1.195) in all circumstances, we may now invoke [69, Theorem 4.4.7] (whose ap-
plicability in the present setting makes use of the current item (3)) to conclude that,
5.1 Theory of Generalized Double Layers 563

indeed, the mapping T # , originally considered as in the present item (3), extends u-
 from the Hardy space H (∂Ω, σ) into itself.
niquely to a linear and bounded operator p

Finally, that various choices of p ∈ n−1 n , 1 in (5.1.22) yield operators which are
compatible with one another may now be seen with the help of [69, Theorem 4.4.3].
Moving on, for each f ∈ H 1 (∂Ω, σ) the memberships in (5.1.29) are immediate
consequences of [69, (4.2.10)] and (5.1.22) (used with p := 1). To justify (5.1.30), by
continuity (cf. [69, (4.2.10)], (5.1.22)) and [69, Theorem 4.4.1] it suffices to consider
the case when f is an atom for the Hardy space H 1 (∂Ω, σ), say f = a where a is
as in [69, (4.4.2)-(4.4.3)] (with Σ := ∂Ω and p := 1). When ∂Ω is bounded we may
rely on (5.1.18) and (5.1.13) to write
∫
⎧
⎪ ϑ
∫ ∫ ⎪
⎪ − a dσ if Ω is bounded,
⎨ 2 ∂Ω
⎪
T a dσ =
#
aT1 dσ = ∫
∂Ω ∂Ω ⎪
⎪ ϑ
⎪
⎪ + a dσ if Ω is an exterior domain,
⎩ 2 ∂Ω
(5.1.196)
which is in agreement with (5.1.30). When ∂Ω is unbounded, we may invoke the
vanishing moment condition in [69, (4.4.2)] and (5.1.11) to write (for some arbitrary
aperture parameter κ > 0)
∫ ∫ ∫  κ−n.t. 
ϑ 
T # a dσ = 2 a + T #
a dσ = − ν · Wa  dσ = 0, (5.1.197)
∂Ω ∂Ω ∂Ω ∂Ω

thanks to (5.1.9) and the Divergence Formula [68, (1.2.2)] (whose applicability with
F := Wa is guaranteed by (5.1.9), (5.1.10), and (5.1.15) with p = q = 1). This
finishes the proof of (5.1.30).
Proof of claims in item (5): The claim made in item (5) may be justified by reasoning
much as in the proof of Theorem 3.2.1, now relying on what we have proved already
in the current item (4).
Proof of claims in item (6): Suppose f is an arbitrary function belonging to the

weighted boundary Sobolev space L11 ∂Ω, 1+σ(x)
|x | n−1
and fix ∈ {1, . . . , n} arbitrary.
Then for each x ∈ Ω we may write
564 5 Generalized Double Layers in Uniformly Rectifiable Domains

∂ (T f )(x)
∫
= ν j (y)∂x [k j (x − y)] f (y) dσ(y)
∂Ω
∫
=− ν j (y)∂y [k j (x − y)] f (y) dσ(y)
∂Ω
∫
= ∂τ j (y) [k j (x − y)] f (y) dσ(y)
∂Ω
∫

= k j (x − y) ∂τ j  f (y) dσ(y). (5.1.198)
∂Ω

Above, the first equality is obtained by differentiating under the integral sign in
(5.1.4), the second inequality is a simple consequence of chain rule, the third equality
takes into account (A.0.183) as well as the fact that ∂y j [k j (x − y)] = 0 by (5.1.2),
and the last equality is justified by [69, Lemma 11.1.7]. Then (5.1.33) follows from
(5.1.198) and [70, Theorem 2.5.1]. Collectively, (5.1.198), (5.1.4), and item (3) in
[70, Theorem 2.4.1] also prove the estimates claimed in (5.1.34) and the subsequent
comment.
Proof of claims in item (7): Start by choosing an arbitrary function
 
f ∈ L11 ∂Ω, 1+σ(x)
p
|x | n−1
∩ L1,loc (∂Ω, σ) for some p ∈ (1, ∞). (5.1.199)

First, from (5.1.4), (5.1.198), and item (2) in [70, Theorem 2.4.1] (cf. [70, (2.4.8)])
we see that
p
if f is as in (5.1.199) then Nκ (T f ), Nκ (∇T f ) ∈ Lloc (∂Ω, σ). (5.1.200)

Second, based on (5.1.198), [69, (11.4.8)], and (5.1.4), for each ∈ {1, . . . , n} we
may compute

∂ (T f )(x) = T (∇tan f ) (x)
∫
−  − y), (∇tan f )(y) dσ(y) for all x ∈ Ω. (5.1.201)
ν (y) k(x
∂Ω

As a consequence of (5.1.201) and (5.1.170), at σ-a.e. x ∈ ∂Ω we then have

5.1 Theory of Generalized Double Layers 565
  κ−n.t.  ϑ 

∂ Tf  (x) = − (∇tan f ) + T (∇tan f ) (x)
∂Ω 2
1 #
− k j ν(x) ν (x)(∇tan f ) j (x)
2i
∫
− lim+ k j (x − y)ν (y)(∇tan f ) j (y) dσ(y). (5.1.202)
ε→0
y ∈∂Ω
|x−y |>ε

From (5.1.7), (5.1.202), (5.1.200), and [69, Proposition 11.3.2] (used with u := T f )
p
we conclude that T f ∈ L1,loc (∂Ω, σ), hence (5.1.39) holds. Moreover, for each
r, s ∈ {1, . . . , n} we may then compute, at σ-a.e. x ∈ ∂Ω,
566 5 Generalized Double Layers in Uniformly Rectifiable Domains

∂τr s (T f )(x) (5.1.203)

 ϑ  ϑ
= ∂τr s − f + T f (x) + ∂τr s f (x)
2 2
 κ−n.t.  ϑ

= ∂τr s T f  (x) + ∂τr s f (x)
∂Ω 2
  κ−n.t.    κ−n.t.  ϑ
 
= νr (x) ∂s T f  (x) − νs (x) ∂r T f  (x) + ∂τr s f (x)
∂Ω ∂Ω 2
ϑ 
= − νr (x)(∇tan f )s (x) + νr (x)T (∇tan f )s (x)
2
1 #
− k j ν(x) νr (x)νs (x)(∇tan f ) j (x)
2i
∫
− lim+ k j (x − y)νr (x)νs (y)(∇tan f ) j (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

ϑ 
+ νs (x)(∇tan f )r (x) − νs (x)T (∇tan f )r (x)
2
1 #
+ k j ν(x) νs (x)νr (x)(∇tan f ) j (x)
2i
∫
+ lim+ k j (x − y)νs (x)νr (y)(∇tan f ) j (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

ϑ
+ ∂τ f (x)
2 rs
    
= T ∂τr s f (x) + Mνr , T (∇tan f )s (x) − Mνs , T (∇tan f )r (x)
∫

− lim+ k j (x − y) νr (x)νs (y) − νs (x)νr (y) (∇tan f ) j (y) dσ(y).
ε→0
y ∈∂Ω
|x−y |>ε

The first equality in (5.1.203) is just algebra, the second equality uses the jump-
formula (5.1.7), the third equality is provided by [69, Proposition 11.3.2] used with
u := T f (whose present applicability is ensured by (5.1.7), (5.1.202), and (5.1.200)),
the fourth equality is based on (5.1.202), while the final equality is a consequence
of [69, (11.4.8)]. Ultimately, (5.1.203) shows that at σ-a.e. x ∈ ∂Ω we have
5.1 Theory of Generalized Double Layers 567

∂τr s (T f )(x) (5.1.204)

    
= T ∂τr s f (x) + Mνr , T (∇tan f )s (x) − Mνs , T (∇tan f )r (x)
∫
 
− lim+  − y), (∇tan f )(y) νr (x)νs (y) − νs (x)νr (y) dσ(y),
k(x
ε→0
y ∈∂Ω
|x−y |>ε

and (5.1.40) readily follows from this and (5.1.19). The claims pertaining to (5.1.35)-
(5.1.37) are then seen from this and the current item (3).
p,λ
That formula (5.1.40) also holds for each function f belonging to M1 (∂Ω, σ)
q,λ
or B1 (∂Ω, σ) with p, q ∈ (1, ∞) and λ ∈ (0, n − 1) is clear from (5.1.38), (A.0.150),
(A.0.33), [69, (6.2.7)], and [69, (6.2.71)]. In turn, this also ensures that the claims in
(5.1.41) are valid.
Proof of claims in item (8): All results in item (8) are implied by the current item
(7), item (3), (A.0.136), (A.0.137), and duality.
Proof of claims in item (9): All claims in item (9) with the exception of (5.1.48),
(5.1.49)-(5.1.50), and (5.1.51)-(5.1.52) are direct consequences of definitions and
[70, Corollary 2.5.3]. As far as the claim made in (5.1.48) is concerned, having fixed
two arbitrary points x0, x1 ∈ Ω, we need to show that (Tmod 1 (x0 ) = (Tmod 1 (x1 ),
which is further equivalent to proving that
∫
 
ν(y), k(x  1 − y) dσ(y) = 0.
 0 − y) − k(x (5.1.205)
∂Ω

To this end, consider the vector field defined by

 0 − y) − k(x
 := k(x
F(y)  1 − y) at L n -a.e. point y ∈ Ω. (5.1.206)

Hence  1 n
F ∈ Lloc (Ω, L n ) (5.1.207)
and (5.1.168) gives that, with the divergence taken in the sense of distributions in Ω,

divF = −ϑδx0 + ϑδx1 ∈ ℰ(Ω) (5.1.208)

where, generally speaking, δx is the Dirac distribution in Ω with mass at x ∈ Ω.

Fix a compact set K ⊂ Ω whose interior contains both x0 and x1 . Then the same
argument which has established [70, (2.3.117)] presently gives that there exists a
constant CK ∈ (0, ∞) such that

 CK
| F(y)| ≤ , ∀y ∈ Ω \ K. (5.1.209)
1 + |y| n

 κ, n) ∈ (0, ∞)
Granted this, [68, Lemma 8.3.7] shows that there exists C = C(K, k,
such that
568 5 Generalized Double Layers in Uniformly Rectifiable Domains
  C
NκΩ\K F (y) ≤ for each y ∈ ∂Ω. (5.1.210)
1 + |y| n
In turn, from (5.1.210), [68, (8.2.26)], and [68, Lemma 7.2.1], we conclude that

NκΩ\K F ∈ L 1 (∂Ω, σ). (5.1.211)

In addition, since F is continuous in a neighborhood of ∂Ω it follows that the

pointwise nontangential boundary trace
κ−n.t.
F∂Ω exists at every point in ∂nta Ω, (5.1.212)

and at σ-a.e. y ∈ ∂Ω we have

 κ−n.t.   
ν(y) · F∂Ω (y) = ν(y), k(x
 0 − y) − k(x
 1 − y) . (5.1.213)


Finally, we note that in the case when Ω is unbounded we have | F(y)| = O(|y| −n )
for y ∈ Ω with |y| → ∞. Hence, in such a scenario,
∫

|y · F(y)| dL n (y) = O(R) as R → ∞. (5.1.214)
[B(0,2R)\B(0,R)]∩Ω

At this stage, we may write

∫
 
 0 − y) − k(x
ν(y), k(x  1 − y) dσ(y)
∂Ω
∫  κ−n.t. 
 
= ν(y) · F 1
(y) dσ(y) = (𝒞∞b (Ω))∗ divF, 𝒞∞ (Ω)
∂Ω ∂Ω b

= −ϑ + ϑ = 0. (5.1.215)

Above, the first equality comes from (5.1.213), the second equality is formula [68,
(1.4.6)] (keeping in mind that the hypotheses of [68, Theorem 1.4.1] are satisfied,
thanks to (5.1.207), (5.1.208), (5.1.212), (5.1.214)), the third equality is seen from
(5.1.208), and the final equality is obvious. In turn, (5.1.215) establishes (5.1.205).
This finishes the proof of (5.1.48).
Next, the identity claimed in (5.1.52) for any function f as in (5.1.51) is established
staring from (5.1.47), then reasoning as in (5.1.198) based on the integration by parts
formula on the boundary from [69, Lemma 11.1.7].
Consider now the task of justifying (5.1.50) for the operator (5.1.49). To this end,
fix a function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near the origin, and pick an arbitrary point
n

x ∈ Ω. Then the vector field

 − y) − k(−y)(1
Fx (y) := k(x  − ψ(y)) for L n -a.e. y ∈ Ω (5.1.216)
5.1 Theory of Generalized Double Layers 569
 1  n κ−n.t.
belongs to Lloc (Ω, L n ) ∩ 𝒞∞ Ω \ {x} . In particular, Fx ∂Ω exists at every point
in ∂nta Ω. Also, since ∇ψ · k ∈ 𝒞∞c (R ), from (5.1.168) and (5.1.2) we see that the
n


divergence of Fx , taken in the sense of distributions in Ω, is given by


divFx = −ϑδx + div[(1 − ψ) k]

= −ϑδx − ∇ψ · k ∈ ℰ(Ω) + L 1 (Ω, L n ). (5.1.217)

Finally, much as in (5.1.211), (5.1.214), we presently have

NκΩ\K Fx ∈ L 1 (∂Ω, σ) for any compact neighborhood K ⊆ Ω of x, (5.1.218)

and
∫
|y · Fx (y)| dL n (y) = O(R) as R → ∞. (5.1.219)
[B(0,2R)\B(0,R)]∩Ω

Granted these properties, we may then write

∫  κ−n.t. 
  
mod 1 (x) =
T ν(y) · Fx  (y) dσ(y) = (𝒞∞b (Ω))∗ divFx, 1 𝒞∞ (Ω)
∂Ω ∂Ω b

∫
= −ϑ − ∇ψ · k dL n, (5.1.220)
Ω

as seen from (5.1.49), (5.1.216), [68, (1.4.6)], and (5.1.217). This establishes (5.1.50),
so the justification of the claims in item (9) is now complete.
Proof of claims in item (10): The claim pertaining to (5.1.59) follows from [70, Lem-
ma 2.1.2], used with 𝒬 := ∇Tmod , bearing in mind that the constant C2 defined in [70,
(2.1.18)] presently vanishes, thanks to (5.1.48). The claim concerning (5.1.60) then
follows by combining (5.1.59) with [68, (5.11.78)]. Finally, the claims in (5.1.61)
and (5.1.62)-(5.1.64) are justified in a manner very similar to the proof of (1.8.16)
and (1.8.17)-(1.8.19), respectively.
Proof of claims in item (11): The claim in (5.1.67) is implied by [70, (2.3.34)]. The
jump-formula (5.1.68) follows from definitions and [70, Corollary 2.5.3]. The claim
made in (5.1.70) is a consequence of [70, (2.3.36)], bearing in mind (5.1.65) and
(5.1.6). To prove (5.1.73), pick a function f as in (5.1.71). Then for each pair of
indices r, s ∈ {1, . . . , n} we may write
 
∂τr s Tmod f = ∂τr s − ϑ2 f + Tmod f + ϑ2 ∂τr s f
 κ−n.t.  κ−n.t.
= νr ∂s (Tmod f ) ∂Ω − νs ∂r (Tmod f ) ∂Ω + ϑ2 ∂τr s f (5.1.221)

where the second equality is justified by invoking [69, Proposition 11.3.2] with
u := Tmod f and p := 1 (its present applicability is ensured by (5.1.68), (5.1.53),
(5.1.54), and (5.1.55)). Next, observe from (5.1.52) and [69, (11.4.8)] that for each
570 5 Generalized Double Layers in Uniformly Rectifiable Domains

index ∈ {1, . . . , n} we have

∫

∂ Tmod f (x) = k j (x − y)(∂τ j f )(y) dσ(y) (5.1.222)
∂Ω

= T (∇tan f ) (x)
∫
−  − y), (∇tan f )(y) dσ(y) for all x ∈ Ω.
ν (y) k(x
∂Ω

With (5.1.221) and (5.1.222) in hand, the same type of argument as in (5.1.203)-
(5.1.204) then establishes (5.1.73). Finally, (5.1.72) follows from [70, (2.3.35)],
(5.1.73), and [70, (2.3.17)].
Proof of claims in item (12): The main step is to prove the estimate stated in (5.1.74),
from which all other claims in this item then follow. However, having estab-
lished (5.1.48), the argument proceeds very much as in the case of the proof
 Corollary 2.4.2]. Specifically, having fixed a point xo ∈ ∂Ω and a scale
of [70,
r ∈ 0, 2 diam(∂Ω) , the same argument which have produced [70, (2.4.136)] (based
on the estimate in [70, (2.4.34)] and the decay of the integral kernel of ∇Tmod ) now
gives
∫   p
 
∇ Tmod f (x) dist(x, ∂Ω) p−1 dx (5.1.223)
B(x o ,r)∩Ω
 
≤ Cσ B(xo, r) ∩ ∂Ω fp# (xo ) p + Cσ B(xo, r) ∩ ∂Ω f1# (xo ) p,

where the L q -based Fefferman-Stein maximal function fq# has been defined in
(A.0.195). Granted
 this, the version of (5.1.74) with the supremum taken in the
regime r ∈ 0, 2 diam(∂Ω) follows on account of [68, (7.4.111)]. Finally, the case
when Ω is an exterior domain and r ≥ 2 diam(∂Ω) is handled much as in [70,
(2.4.142), (2.4.143)].
Next, consider the claim made in (5.1.76). To this end, pick p ∈ (1, ∞) and select
an arbitrary function f ∈ BMO(∂Ω, σ). Also, fix some α ∈ (0, 1) choose some
arbitrary function
.
g ∈ 𝒞α (∂Ω) ∩ BMO(∂Ω, σ). (5.1.224)

Then for each r ∈ 0, 2 diam(∂Ω) and x ∈ ∂Ω we may estimate
 ∫ 1
1   
 ∇ T ( f − g)  p dist(·, ∂Ω) p−1 dL n p
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
mod

≤C f −g BMO(∂Ω,σ), (5.1.225)

thanks to (5.1.74) written with f − g in place of f . In addition, having fixed some

arbitrary α ∈ (0, 1), we may write
5.1 Theory of Generalized Double Layers 571
 ∫ 1
1   
 ∇ T g  p dist(·, ∂Ω) p−1 dL n p
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
mod

 ∫  p1
1
≤ C g 𝒞. α (∂Ω)  dist(·, ∂Ω) pα−1 dL n
σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

≤ C g 𝒞. α (∂Ω) r α . (5.1.226)

Indeed, the first inequality above uses (5.1.59) (written for g in place of f ), while
the second inequality is based on [68, (8.6.101)] used with λ := 1 − pα, α := 1,
β := n − 1, and E := B(x, r) ∩ Ω. Collectively, (5.1.225) and (5.1.226) imply that
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω
r ∈(0,R)

≤C f −g BMO(∂Ω,σ), (5.1.227)

for some constant C ∈ (0, ∞) independent of f and g. With this in hand, (5.1.76)
follows on account of [69, Theorem 3.1.3] and the ability to choose g arbitrary as in
(5.1.224). In turn, (5.1.76) readily implies (5.1.77)-(5.1.78).
Proof of claims in item (13): All claims clear from what we have proved in the current
items (9)-(12) and (5.1.45).
Proof of claims in item (14): Recall from [68, (7.4.118)] that
 σ(x) 
BMO(∂Ω, σ) ⊆ L 1 ∂Ω, . (5.1.228)
1 + |x| n

If we now pick an arbitrary function f ∈ BMO(∂Ω, σ), from (5.1.228), [70, (2.3.35)],
(5.1.65), and [68, (7.4.105)] we see that

p
Tmod f ∈ Lloc (∂Ω, σ). (5.1.229)
1≤p<∞

Let us also consider an arbitrary (1, ∞)-atom a : ∂Ω → C with vanishing moment14.

This is some σ-measurable function with the property that there exist xo ∈ ∂Ω and
r ∈ 0, 2 diam(∂Ω) such that
 −1
supp a ⊆ B(xo, r) ∩ ∂Ω, a L ∞ (∂Ω,σ) ≤ σ B(xo, r) ∩ ∂Ω ,
∫ (5.1.230)
and a dσ = 0.
∂Ω

From (5.1.181) and [69, (4.5.33)] we know that

14 this is always the case if ∂Ω is unbounded

572 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫
| f ||T # a| dσ < +∞, (5.1.231)
∂Ω
∫  
and we make the additional claim that T f  |a| dσ < +∞ and
∂Ω mod

∫ ∫

f (T a) dσ =
#
Tmod f a dσ. (5.1.232)
∂Ω ∂Ω

Indeed, these are all consequences of (5.1.70). A direct proof of (5.1.232) goes as
follows. First, bearing (5.1.231) in mind, we write
∫ ∫ ∫
f T # a dσ = f T # a dσ + f T # a dσ
∂Ω ∂Ω∩B(x o ,2r) ∂Ω\B(x o ,2r)

=: I + II, (5.1.233)

with the last equality defining I and II. In relation to these, note that on account of
the current item (3) and [68, (7.4.105)] we may re-write I as
∫ ∫  
 
I= f · 1∂Ω∩B(xo,2r) T # a dσ = T f · 1∂Ω∩B(xo,2r) a dσ
∂Ω ∂Ω
∫  

= Tmod f · 1∂Ω∩B(xo,2r) a dσ (5.1.234)
∂Ω
 
since T f · 1∂Ω∩B(xo,2r) differs from Tmod f · 1∂Ω∩B(xo,2r) by a constant on ∂Ω,
thanks to [70, (2.3.34)] and the fact that the atom has integral zero (cf. the last
property in (5.1.230)). Also,
5.1 Theory of Generalized Double Layers 573
∫ ∫ 
II = f (x)  − x) a(y) dσ(y) dσ(x)
ν(x), k(y
∂Ω\B(x o ,2r) ∂Ω

∫ ∫ 
= f (x) ν(x), kε (y − x) a(y) dσ(y) dσ(x)
∂Ω\B(x o ,2r) ∂Ω

∫ ∫ 
 
= f (x) ν(x), kε (y − x) − k1 (−x) a(y) dσ(y) dσ(x)
∂Ω\B(x o ,2r) ∂Ω

∫ ∫ 

= f (x) ν(x), kε (y − x) − k1 (−x) a(y) dσ(y) dσ(x)
∂Ω\B(x o ,2r) ∂Ω∩B(x o ,r)

∫ ∫ 
 
= ν(x), kε (y − x) − k1 (−x) f (x) dσ(x) a(y) dσ(y)
∂Ω∩B(x o ,r) ∂Ω\B(x o ,2r)

∫ ∫ 
 
= lim ν(x), kε (y − x) − k1 (−x) f (x) dσ(x) a(y) dσ(y)
ε→0+ ∂Ω\B(x o ,2r)
∂Ω
∫  

= Tmod f · 1∂Ω\B(xo,2r) (y) a(y) dσ(y). (5.1.235)
∂Ω

The first equality in (5.1.235) is implied by (5.1.6) bearing in mind that, thanks
to the first property in (5.1.230), the variables x, y are uniformly separated. The
second equality in (5.1.235) uses (5.1.66) and is valid for each choice ε ∈ (0, r).
The third equality in (5.1.235) is a consequence of the cancelation property of the
atom (cf. the last property in (5.1.230)), while the fourth equality in (5.1.235) is
seen from the first property in (5.1.230). The fifth equality in (5.1.235) follows from
Fubini’s Theorem whose applicability is presently ensured by the fact that the double
integral is absolutely convergent, thanks to the properties listed in the first line of
(5.1.230), the estimate in [70, (2.3.117)] (with a constant which stays bounded for x
in a compact subset of ∂Ω), and (5.1.228). The sixth equality in (5.1.235) uses the
fact that the inner integral is actually independent of ε ∈ (0, r), and also the support
condition for the atom. Finally, the last equality in (5.1.235) is seen from (5.1.65).
At this stage, from (5.1.233)-(5.1.235) we conclude that (5.1.232) holds for each a
as in (5.1.230).
Let us now suppose that ∂Ω is unbounded. Then, on the one hand, based on
(5.1.232) and [69, Proposition 4.8.6] (whose applicability with g := T # a is ensured
by (5.1.22) and (5.1.231)), for each (1, ∞)-atom a on ∂Ω we may write
∫
  
Tmod f a dσ = [ f ], T # a (5.1.236)
∂Ω

where ·, · denotes the duality bracket between the John-Nirenberg space of func-
tions of bounded mean oscillations on ∂Ω, modulo constants, and the Hardy space
574 5 Generalized Double Layers in Uniformly Rectifiable Domains

H 1 on Σ (cf. [69, Theorem 4.6.1]). In concert with [69, (4.6.9)], this permits us to
estimate
∫    
 
 Tmod f a dσ  =  [ f ], T # a 
∂Ω

≤C f BMO(∂Ω,σ) · T #a H 1 (∂Ω,σ)

≤C f BMO(∂Ω,σ), (5.1.237)

where the last inequality is based on (5.1.22) and [69, (4.5.5)-(4.5.6)]. On the other
hand, from (5.1.229), (A.0.20), [68, Proposition 7.4.12], and (5.1.230) we see that

∫  
 
Tmod f BMO(∂Ω,σ) ≤ C · sup  Tmod f a dσ  : a (1, ∞)-atom on ∂Ω .
∂Ω
(5.1.238)
Together, (5.1.237) and (5.1.238) give

Tmod f BMO(∂Ω,σ)
≤C f BMO(∂Ω,σ), (5.1.239)

for some C ∈ (0, ∞) independent of f . Hence, the operator Tmod is a well-defined,

linear, bounded mapping from BMO(∂Ω, σ) into itself when ∂Ω is unbounded. In
&
view of the format of the norm on BMO(∂Ω, σ) (cf. [68, (7.4.95)]) and (5.1.69),
we then also conclude that the mapping in (5.1.84) is well defined, linear, and
bounded, when ∂Ω is unbounded. With these properties in hand, in the case when
∂Ω is unbounded, proving (5.1.86) comes down (thanks to the last property in [69,
(4.4.114)] and the continuity of T # on the Hardy scale) to showing that
    
Tmod f , a = [ f ], T # a (5.1.240)

for each f ∈ BMO(∂Ω, σ) and each (1, ∞)-atom a on ∂Ω. This, however, is clear
from (5.1.236) and the duality result in [69, Theorem 4.6.1, (4.6.8)] (bearing in mind
that we already know that Tmod f ∈ BMO(∂Ω, σ)).
At this stage, all claims pertaining to (5.1.82), (5.1.84), and (5.1.86) have been
justified when ∂Ω is unbounded. From (5.1.82), (5.1.93) (whose proof is independent
of the present considerations), and [69, Theorem 3.1.3] we then conclude that the
operator (5.1.83) is also well defined, linear, and bounded. In concert with (5.1.69),
this also takes care of (5.1.85) in the case when ∂Ω is unbounded.
Let us now treat the case when ∂Ω is bounded. In such a scenario, (5.1.89) holds
thanks to [68, (7.4.105)]. As a consequence of this and (5.1.67), for each a as in
(5.1.230) we have
∫ ∫
f T # a dσ = (T f ) a dσ. (5.1.241)
∂Ω ∂Ω

Then the same argument which, starting with (5.1.232), has produced (5.1.239)
presently gives
5.1 Theory of Generalized Double Layers 575

Tf . ≤C f BMO(∂Ω,σ) for each f ∈ BMO(∂Ω, σ). (5.1.242)

BMO(∂Ω,σ)

From this, (5.1.17), and [69, (4.6.18)] we then conclude that T is a well-defined,
linear, and bounded operator in the context of (5.1.90).
 
Next, (5.1.89) and (5.1.67) currently give that Tmod f = [T f ] for each function
f ∈ BMO(∂Ω, σ). With this in hand, it follows from (5.1.90) that the mapping
in (5.1.84) is bounded. Granted this, for each f ∈ BMO(∂Ω, σ) we may estimate,
bearing in mind (A.0.20), the fact that C f := Tmod f − T f is a constant as in (5.1.67),
and [69, (4.6.18)]:

Tmod f BMO(∂Ω,σ)
≤ Tmod f − T f BMO(∂Ω,σ)
+ Tf BMO(∂Ω,σ)

≤ σ(∂Ω)|C f | + C f BMO(∂Ω,σ)

≤C f L 1 (∂Ω,σ) +C f BMO(∂Ω,σ)

≤C f BMO(∂Ω,σ) . (5.1.243)

This proves that the mapping (5.1.82) is well defined and bounded. In turn, from
(5.1.82), [69, (3.1.50)], and (5.1.102) (whose proof is independent of the present
considerations), we also see that (5.1.83) is well-defined and bounded operator.
Pressing on, for each f ∈ BMO(∂Ω, σ) and each (1, ∞)-atom a on ∂Ω we may
write
∫ ∫
   
T f, a = (T f )a dσ = f T # a dσ = f , T # a (5.1.244)
∂Ω ∂Ω

by [69, Proposition 4.8.6] (keeping in mind (5.1.89) and (5.1.17)), (5.1.18), and
(5.1.22). In concert with the last property in [69, (4.4.114)] and the continuity of
T # on the Hardy scale, this proves (5.1.91). The claims pertaining to (5.1.92) are
consequences of (5.1.90), (5.1.102), and [69, (3.1.50)].
Finally, the claims made in relation to (5.1.87) and (5.1.88) may be justified based
on (5.1.82), (5.1.84), and (5.1.93), as in the proof of Corollary 2.1.14.
Proof of claims in item (15): All results in item (15) dealing with the ordinary Hölder
scale are established much as their counterparts have been dealt with in the current
item (14). More specifically, from [68, (7.4.119)] we know that
.  σ(x) 
𝒞α (∂Ω) ⊆ L 1 ∂Ω, . (5.1.245)
1 + |x| n
.
Then for any function f ∈ 𝒞α (∂Ω), from (5.1.245), [70, (2.3.35)], and (5.1.65) we
see that 
q
Tmod f ∈ Lloc (∂Ω, σ). (5.1.246)
1≤q<∞

Also, from (5.1.181) and [69, Lemma 4.5.5] we know that

576 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫
| f ||T # a| dσ < +∞. (5.1.247)
∂Ω

With this in hand, the same argument that has produced (5.1.232) currently gives
∫ ∫

f T a dσ =
#
Tmod f a dσ (5.1.248)
∂Ω ∂Ω

for each (p, ∞)-atom a on ∂Ω with vanishing moment. Granted this, we may reason
as in (5.1.236)-(5.1.239), now relying on [69, Proposition 4.8.7] in place of [69,
Proposition 4.8.6], and [68, Proposition 7.4.8] in place of [68, Proposition 7.4.12],
to conclude that there exists C ∈ (0, ∞) such that
.
Tmod f 𝒞. α (∂Ω) ≤ C f 𝒞. α (∂Ω) for each f ∈ 𝒞α (∂Ω). (5.1.249)

From this and (5.1.69), the claims concerning the operators (5.1.93)-(5.1.95) follow.
Once these have been established, in the case when ∂Ω is unbounded, proving
(5.1.99) reduces (in view of the last property in [69, (4.4.114)]) to checking that
    
Tmod f , a = [ f ], T # a for each
. (5.1.250)
f ∈ 𝒞α (∂Ω) and each (p, ∞)-atom a on ∂Ω.
.
Since we already know that Tmod f ∈ 𝒞α (∂Ω), from the duality result in [69, Theo-
rem 4.6.1, (4.6.8)] we conclude that
∫
  
Tmod f , a = (Tmod f )a dσ. (5.1.251)
∂Ω

Observe that since a is a (p, ∞)-atom, hence also a multiple of a (1, ∞)-atom, from
(5.1.22) we have
T # a ∈ H 1 (∂Ω, σ) ∩ H p (∂Ω, σ). (5.1.252)
Thanks to (5.1.247) and (5.1.252), [69, Proposition 4.8.7] applies and gives
∫
 
[ f ], T # a = f T # a dσ. (5.1.253)
∂Ω

Now (5.1.250) follows from (5.1.251), (5.1.253), and (5.1.248).

In the case when ∂Ω is bounded, starting with the fact that (5.1.241) holds for
each f ∈ 𝒞α (∂Ω) and each (p, ∞)-atom a on ∂Ω, the same argument which has
proved (5.1.249), outlined above, now gives

T f 𝒞. α (∂Ω) ≤ C f 𝒞α (∂Ω) for each f ∈ 𝒞α (∂Ω). (5.1.254)

Moreover, for each function f ∈ 𝒞α (∂Ω) and σ-a.e. point x ∈ ∂Ω we may use
(5.1.13) and [68, (7.2.5)] (with X := ∂Ω, r := 2 diam(∂Ω), d := n − 1, and δ := α)
to estimate
5.1 Theory of Generalized Double Layers 577
 
|(T f )(x)| ≤  T( f − f (x)) (x) + |ϑ2 | | f (x)|
∫
dσ(y)
≤ C f 𝒞. α (∂Ω) + |ϑ2 | | f (x)|
∂Ω |x − y| n−1−α

 α
≤ C diam(∂Ω) f 𝒞. α (∂Ω) + |ϑ2 | sup | f |, (5.1.255)
∂Ω

for some C ∈ (0, ∞) which depends only on k,  n, and the lower ADR constant of
Ω. Keeping in mind that, as seen from (5.1.254), the operator T maps 𝒞α (∂Ω) into
continuous functions on ∂Ω, we then conclude from (5.1.255) that there exists some
constant C ∈ (0, ∞) with the property that

sup |(T f )(x)| ≤ C f 𝒞α (∂Ω) for each f ∈ 𝒞α (∂Ω). (5.1.256)

x ∈∂Ω

Ultimately, (5.1.255) and (5.1.256) imply that there exists C ∈ (0, ∞) such that

Tf 𝒞α (∂Ω) ≤C f 𝒞α (∂Ω) for each f ∈ 𝒞α (∂Ω). (5.1.257)

This proves that T is well defined and bounded in the context of (5.1.102). Lastly,
(5.1.103) is justified as before, based on density (cf. [69, (4.4.114)]) and the fact that
(5.1.244) continues to hold for each f ∈ 𝒞α (∂Ω) and each (p, ∞)-atom a on ∂Ω (cf.
[69, Corollary 4.8.11], keeping in mind (5.1.181), and the bounded set version of
the duality result from [69, Theorem 4.6.1, (4.6.8)]).
Finally, all results in item (15) dealing with the scale of vanishing Hölder s-
paces follow from what we have proved so far and the density result from [69,
Theorem 3.2.2].
Proof of claims in item (16): All desired conclusions may be justified by reasoning
as in the proof of Theorem 3.2.2, making use of the results established so far.
Proof of claims in item (17): The claims in item (17) are established by closely fol-
lowing the argument presented in the proof of Theorem 3.1.1, and relying the results
obtained in earlier items.
Proof of claims in item (18): Fix ∈ {1, . . . , N } and pick a multi-index α ∈ N0 with
|α| = . From (5.1.47) we know that
∫
  
∂ α Tmod f (x) =  − y) f (y) dσ(y)
ν(y), (∂ α k)(x
∂Ω
(5.1.258)
1 σ(y)
for each f ∈ L ∂Ω, 1+ |y | n and x ∈ Ω.

In addition, (5.1.48) implies

∂ α (Tmod 1) = 0 in Ω. (5.1.259)
578 5 Generalized Double Layers in Uniformly Rectifiable Domains

Granted (5.1.258)-(5.1.259), the claims made in item (18) then become consequences
of Theorem 4.2.1 (used with k := − 1 and ε := 0).
Proof of claims in item (19): All results may be justified by reasoning as in the proofs
of Theorem 4.1.1, Corollary 4.1.4, and Theorem 4.1.5. The key cancelation property,
formulated as in (4.1.5) with T # in place of K # , follows from (4.1.9) and (5.1.30).
Proof of claims in item (20): The justification of the claims in item (20) proceed
along the lines of the proof of Theorem 1.8.14, making use of what we have
established earlier.
Proof of claims in item (21): All claims in item (21) may be justified by reasoning
much as in the proof of Theorem 3.3.8, employing results that are available to us
from earlier work.
Proof of claims in item (22): Reason as in the proof of Theorem 2.3.9, this time
making use of (5.1.14), (5.1.45), (5.1.57), (5.1.68), and (5.1.69). 

We next discuss a basic estimate which, in view of its specific format, can be
thought of as a “space-less” Carleson-type estimate for modified generalized double
layer operators in uniformly rectifiable domains.

Theorem 5.1.8 Pick n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an arbitrary UR

domain. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. For a sufficiently large integer N = N(n) ∈ N, consider
a vector-valued function
 n
k ∈ 𝒞 N (Rn \ {0}) odd, positive homogeneous of degree 1 − n,
(5.1.260)
and satisfying div k = 0 in Rn \ {0}.

With the set Ω and the kernel k, associate as in (5.1.44) the modified generalized

double layer operator Tmod , i.e., the mapping sending each f ∈ L 1 ∂Ω, 1+σ(x) |x | n into
the function defined at each point x ∈ Ω according to
∫  

Tmod f (x) :=  − y) − k(−y)
ν(y), k(x  · 1Rn \B(0,1) (y) f (y) dσ(y). (5.1.261)
∂Ω

Then for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which depends only
 and the UR constants of ∂Ω, with the property that for each function
on n, p, k,

f ∈ L ∂Ω, 1+σ(x)
1
|x | n , each point xo ∈ ∂Ω, and each radius r ∈ (0, ∞) one has

 ∫  1/p
1   p
 ∇ T 
mod f (x) dist(x, ∂Ω) dL n (x)
p−1
(5.1.262)
σ Δ(xo, r) B(x o ,r)∩Ω
∫ ∞ ⨏  ⨏  p  1/p dλ
 
≤C f − f dσ  dσ
1 Δ(x o ,λr) Δ(x o ,λr) λ2
5.1 Theory of Generalized Double Layers 579

where Δ(xo, R) := B(xo, R) ∩ ∂Ω for each R ∈ (0, ∞).

As a corollary of (5.1.262) and the definition of the Fefferman-Stein sharp max-
imal operator (cf. (A.0.195)), it follows that for each p ∈ (1, ∞) there exists some
constant C ∈ (0, ∞) such that the pointwise inequality
 ∫  1/p
1   p
sup  ∇ T  dist (·, ∂Ω) p−1 dL n ≤ C · fp# (x)
mod f
r >0 σ Δ(x, r) B(x,r)∩Ω

 (5.1.263)
σ(x)
holds for each point x ∈ ∂Ω and each function f ∈ L 1 ∂Ω , 1+ |x | n .

The membership of f to L 1 ∂Ω, 1+σ(x) |x | n is required simply to ensure that the
action of Tmod on f (as defined in (5.1.261)) is meaningful (so the formulation of
(5.1.262) makes sense). The absence of a recognizable norm in the formulation of
the estimate recorded in (5.1.262), together with the specific format of its left-hand
side, validate the point of view (espoused earlier), to the effect that (5.1.262) can be
thought of as a “space-less” Carleson-type estimate for modified generalized double
layer operators in uniformly rectifiable domains.
In terms of the maximal operator P mapping any given L n -measurable function
u : Ω → C into the function defined at every point x ∈ ∂Ω as
∫ 
1
(Pu)(x) := sup  |u| dL ∈ [0, ∞],
n
(5.1.264)
r >0 σ ∂Ω ∩ B(x, r) Ω∩B(x,r)

the estimate recorded in (5.1.263) simply reads

  p  
P ∇ Tmod f  dist (·, ∂Ω) p−1 ≤ C · fp#
p
on ∂Ω, (5.1.265)
 σ(x)
for any given exponent p ∈ (1, ∞) and each function f ∈ L 1 ∂Ω , 1+ |x | n .
Let us now present the proof of Theorem 5.1.8.

Proof of Theorem 5.1.8 Pick an arbitrary function f ∈ L 1 ∂Ω, 1+σ(x) |x | n , and fix
some integrability exponent p ∈ (1, ∞), some point xo ∈ ∂Ω, and some radius r > 0.
Since Tmod maps constant functions on ∂Ω into constant functions in Ω (cf. (5.1.48)),
we see that
∇Tmod 1 ≡ 0 in Ω. (5.1.266)
Going forward, for each R > 0 abbreviate
⨏
ΔR := B(xo, R) ∩ ∂Ω, T(ΔR ) := B(xo, R) ∩ Ω, and fΔ R := f dσ. (5.1.267)
ΔR

In addition, pick a scale r ∈ 0, 2 diam(∂Ω) and consider a cutoff function η in Rn
satisfying
580 5 Generalized Double Layers in Uniformly Rectifiable Domains

η ∈ 𝒞∞
c (R ),
n 0 ≤ η ≤ 1, η ≡ 1 on B(xo, 2r),

η ≡ 0 outside B(xo, 4r), |∂ α η(x)| ≤ Cα r − |α |, (5.1.268)

for all x ∈ Rn and for all α ∈ N0n .

We then proceed to split

f = η( f − fΔ4r ) + (1 − η)( f − fΔ4r ) + fΔ4r . (5.1.269)

From (5.1.269) and the cancelation property (5.1.266) we see that

 
∇Tmod f = ∇Tmod η( f − fΔ4r ) + ∇Tmod (1 − η)( f − fS4r ) . (5.1.270)

As such,
∫  p
 
∇Tmod f (x) dist(x, ∂Ω) p−1 dx
T (Δr )
∫  p
  
≤C ∇Tmod η( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
T (Δr )
∫  p
  
+C ∇Tmod (1 − η)( f − fΔ4r ))(x) dist(x, ∂Ω) p−1 dx
T (Δr )

=: I + II. (5.1.271)

Next, write
∫  p
  
I≤C ∇Tmod η( f − fΔ4r ) (x) dist(x, ∂Ω) p−1 dx
Ω
∫ ∫
   
≤C η( f − fΔ ) p dσ ≤ C  f − fΔ  p dσ
4r 4r
∂Ω Δ4r
⨏
 
≤ Cσ(Δr )  f − fΔ  p dσ. (5.1.272)
4r
Δ4r

Above, the first inequality follows from the definition of I in (5.1.271), the second
inequality follows from [70, (2.4.34)], the third inequality is clear from the support
properties of the function η introduced in (5.1.268), and the last inequality is due to
the fact that σ is doubling (itself, a consequence of the Ahlfors regularity of ∂Ω).
As regards II, we first observe that for each x ∈ T(Δr ) we have
  ∫
   | f (y) − fΔ4r |
∇Tmod (1 − η)( f − fΔ4r ))(x)  ≤ C dσ(y) (5.1.273)
∂Ω\Δ2r r + |xo − y| n

thanks to the definition of ∇Tmod (1 − η)( f − fΔ4r ) , the properties of the function η
from (5.1.268), and the fact that
5.1 Theory of Generalized Double Layers 581

|x − y| ≈ |xo − y| + r uniformly for x ∈ T(Δr ) and y ∈ ∂Ω \ Δ2r . (5.1.274)

In turn, from (5.1.271) and (5.1.273) we obtain

∫ | f (y) − fΔ4r | p ∫
II ≤ C dσ(y) dist(x, ∂Ω) p−1 dx
∂Ω\Δ2r r + |xo − y|
n
T (Δr )
 ∫ | f (y) − f | p
Δ4r
≤ Cr p−1+n
dσ(y)
∂Ω r + |xo − y|
n

 ∫ p
| f (y) − fΔ4r |
≤ Cσ(Δr ) r dσ(y) , (5.1.275)
∂Ω r + |xo − y|
n

where the last equality is a consequence of the lower Ahlfors regularity of ∂Ω and
the fact that we are presently assuming r ∈ 0, 2 diam(∂Ω) .
We continue by noting that the first inequality in [68, (7.4.115)], used with
X := ∂Ω, μ := σ, ρ := | · − · |, d := n − 1, p := 1, q := 1, and ε := 1, yields
∫ ∫ ∞⨏
| f (y) − fΔ4r |   
r dσ(y) ≤ C  f − fΔ  dσ dλ
∂Ω r + |xo − y|
λr
n
1 Δλr λ2
∫ ∞⨏  1/p dλ
 
≤C  f − fΔ  p dσ , (5.1.276)
λr
1 Δλr λ2

where the last step comes from Hölder’s inequality. Combining (5.1.275) with
(5.1.276) then gives
∫ p
∞  ⨏   p  1/p dλ
II ≤ Cσ(Δr )  f − fΔ  dσ . (5.1.277)
λr
1 Δλr λ2

Also, from [68, (7.4.125)] used with X := ∂Ω, μ := σ, ρ := | · − · |, j := 2, p := 1,

q := p, and ε := 1 we see that
⨏    1/p ∫ 8 ⨏    1/p dλ
 f − fΔ  p dσ ≤C  f − fΔ  p dσ
λr
Δ4r
4r
4 Δλr λ2
∫ ∞ ⨏  1/p dλ
 
≤C  f − fΔ  p dσ . (5.1.278)
λr
1 Δλr λ2

From (5.1.272) and (5.1.278) we then conclude that

∫ p
∞ ⨏   p  1/p dλ
I ≤ Cσ(Δr )  
f − fΔλr dσ . (5.1.279)
1 Δλr λ2

Collectively, (5.1.271), (5.1.277) and (5.1.279) prove that there exists C ∈ (0, ∞)
such that
582 5 Generalized Double Layers in Uniformly Rectifiable Domains
 ∫  1/p
1  p
∇T f  dist(·, ∂Ω) p−1 dL n
σ(Δr ) B(x o ,r)∩Ω
mod

∫ ⨏  1/p dλ
∞  
≤C  f − fΔ  p dσ (5.1.280)
λr
1 Δλr λ2

whenever 0 < r < 2 diam(∂Ω). From this, the desired conclusion immediately
follows in the case when ∂Ω is unbounded, or Ω is bounded. To complete the proof
 when Ω is an
of the theorem, there remains to establish a similar estimate in the case
exterior domain and when the supremum is taken in the regime r ∈ 2 diam(∂Ω), ∞ .
In such a scenario, the inequality we seek becomes
 ∫  1/p
1  p
∇T f  dist(·, ∂Ω) p−1 dL n
σ(∂Ω) B(x o ,r)∩Ω
mod

⨏    1/p
≤C  f − f∂Ω  p dσ , (5.1.281)
∂Ω
⨏
where f∂Ω := ∂Ω
f dσ. Equivalently, we may recast this as
∫  1/p
 p
∇T f  dist(·, ∂Ω) p−1 dL n ≤ C f − f∂Ω . (5.1.282)
mod L p (∂Ω,σ)
B(x o ,r)∩Ω

Observe that if f does not belong to L p (∂Ω, σ), then there is nothing to prove (since
the right-hand side of (5.1.282) is infinite). In the case when f ∈ L p (∂Ω, σ) we rely
on [70, (2.4.34)] (with Σ := ∂Ω and f − f∂Ω in place of f ) together with (5.1.46)
and (5.1.266) to write
∫  1/p
|(∇Tmod f )(x)| p dist(x, ∂Ω) p−1 dx ≤ C f − f∂Ω L p (∂Ω,σ) . (5.1.283)
R n \∂Ω

In turn, (5.1.283) readily implies (5.1.282), so the proof of Theorem 5.1.8 is com-
plete. 

Our next corollary amounts to a user-friendly version of Theorem 5.1.8, to the

effect that control of the mean oscillations of a given function at a certain scale
implies control of the Carleson-like semi-norm of the “Littlewood-Paley measure”
associated with said function (in relation to some generalized double layer potential
operator).

Corollary 5.1.9 Suppose Ω is an arbitrary UR domain in Rn (where n ∈ N with

n ≥ 2). Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. For
 a sufficiently nlarge integer N = N(n) ∈ N, consider a
vector-valued function k ∈ 𝒞 N (Rn \ {0}) which is odd, positive homogeneous of
degree 1 − n, and divergence-free in Rn \ {0}. Associate with the set Ω and the kernel
5.1 Theory of Generalized Double Layers 583

k the modified generalized double layer operator Tmod acting on functions from the

space L 1 ∂Ω, 1+σ(x)
|x | n as in (5.1.261). Next, assume

φ : (0, ∞) → [0, ∞) is a function satisfying

∫ ∞
1  dt (5.1.284)
φ ∈ Lloc (0, ∞), L 1 and φ(t) 2 < +∞,
1 t

then define φ : (0, ∞) → [0, ∞) by setting

∫ ∞
dt
φ(r) := r φ(t) 2 for each r ∈ (0, ∞). (5.1.285)
r t

Finally, fix some p ∈ (1, ∞) and assume f ∈ Lloc

1 (∂Ω, σ) is such that

⨏  ⨏  p  1/p
 
sup f − f dσ  dσ ≤ φ(r) for each r ∈ (0, ∞), (5.1.286)
x ∈∂Ω Δ(x,r) Δ(x,r)

where Δ(x, r) := B(x, r) ∩ ∂Ω. 

Then f belongs to the space L 1 ∂Ω, 1+σ(x)
|x | n and there exists a constant C ∈ (0, ∞),

which depends only on n, p, k, and the UR constants of ∂Ω, with the property that
for each r ∈ (0, ∞) one has
 ∫  1/p
1  p
sup  ∇T f  dist(·, ∂Ω) p−1 dL n ≤ C φ(r). (5.1.287)
x ∈∂Ω σ Δ(x, r) B(x,r)∩Ω
mod

Proof From (5.1.284) we see that

∫ ∞
dt
φ(t) 2 < +∞ for each r ∈ (0, ∞). (5.1.288)
r t
In concert with [68, (7.4.115)], used with X := ∂Ω, μ := σ, ρ := | · − · |, d := n − 1,
p := 1, q := 1, and ε := 1, for each xo ∈ ∂Ω and r ∈ (0, ∞) this allows us to estimate
∫ ∫
| f (x) − fΔ(xo,r) | C ∞ dλ
dσ(x) ≤ φ(λr) 2
∂Ω r + |x − x o | n r 1 λ
∫ ∞
dt
=C φ(t) 2 < +∞. (5.1.289)
r t

In turn, this shows that f belongs to the space L 1 ∂Ω, 1+σ(x)
|x | n . Having established
this, Theorem 5.1.8 applies and (5.1.262) readily implies (5.1.287), on account of
(5.1.285). 

Corollary 5.1.9 brings into focus the quantity φ(r), which bounds the L p -based
mean oscillations of a given function at scale r. There are many function spaces
in which such a bound occurs naturally, something we elaborate on in a series of
584 5 Generalized Double Layers in Uniformly Rectifiable Domains

examples, below. In all these cases we adopt the background context employed in
Corollary 5.1.9.
The first such paradigm, has to do with controlling the mean oscillations of a
given function in terms of its BMO semi-norm.

Example 5.1.10 Given an arbitrary function f ∈ BMO(∂Ω, σ), define

φ : (0, ∞) → [0, ∞), φ(t) := f . for each t > 0. (5.1.290)

BMO (∂Ω,σ)
∫∞
Hence, φ satisfies the conditions in (5.1.284), (5.1.286), and since r r
dt
t2
= 1 it
follows that φ defined as in (5.1.285) is presently given by

φ(r) := f . for each r > 0. (5.1.291)

BMO (∂Ω,σ)

Then (5.1.287) implies, after also taking the supremum over all r ∈ (0, ∞), it follows
that for each p ∈ (1, ∞) we have
 ∫  1/p
1  p
sup  ∇T f  dist(·, ∂Ω) p−1 dL n
x ∈∂Ω σ Δ(x, r) B(x,r)∩Ω
mod

r ∈(0,∞)

≤C f . . (5.1.292)
BMO (∂Ω,σ)

This corresponds to the Carlson estimate recorded in (5.1.74).

Next we consider the situation in which the mean oscillations of a given function
are controlled in terms of its Lebesgue norm.

Example 5.1.11 Pick an integrability exponent p ∈ (1, ∞). Having fixed an arbitrary
function f ∈ L p (∂Ω, σ), introduce

φ : (0, ∞) → [0, ∞), φ(t) := C f L p (∂Ω,σ) · t (1−n)/p for each t > 0, (5.1.293)

where C ∈ (0, ∞) is a fixed constant which depends only on p and the upper Ahlfors
regularity constant of ∂Ω. If C is sufficiently large, it follows that the conditions
stipulated in (5.1.284), (5.1.286) are satisfied by our present choice of φ. In addition,
since ∫ ∞
dt  p 
r t (1−n)/p 2 = r (1−n)/p (5.1.294)
r t n−1+p
we see that φ defined as in (5.1.285) currently takes the form
 p 
φ(r) := C f L p (∂Ω,σ) · r (1−n)/p for each r > 0. (5.1.295)
n−1+p
Then (5.1.287) implies, in view of (5.1.46), [70, (2.3.32)], the lower Ahlfors regularity
of ∂Ω, and after also taking the supremum over r > 0, that the boundary-to-domain
generalized double layer potential operator T from (5.1.4) satisfies
5.1 Theory of Generalized Double Layers 585
∫  1/p
 
∇T f  p dist(·, ∂Ω) p−1 dL n ≤C f L p (∂Ω,σ) . (5.1.296)
Ω

This corresponds to the L p -based area-function estimate from [70, (2.4.34)] (written
for Σ := ∂Ω).
We continue by considering the scenario in which one controls the mean oscilla-
tions of a given function in terms of its Morrey norm.
Example 5.1.12 Having fixed an integrability exponent p ∈ (1, ∞) and some number
λ ∈ (0, n − 1), pick an arbitrary function f ∈ M p,λ (∂Ω, σ) and define

φ : (0, ∞) → [0, ∞), φ(t) := 2 f · t (λ−n+1)/p for each t > 0.

M p, λ (∂Ω,σ)
(5.1.297)
Thus, φ satisfies the conditions in (5.1.284), (5.1.286), and since
∫ ∞
dt  p 
r t (λ−n+1)/p 2 = r (λ−n+1)/p (5.1.298)
r t n−1−λ+p

it follows that φ defined as in (5.1.285) is presently given by

 2p 
φ(r) := f M p, λ (∂Ω,σ) · r (λ−n+1)/p for each r > 0. (5.1.299)
n−1−λ+p
Then (5.1.287) implies, in view of (5.1.46), [69, (6.2.25)], and after also taking
the supremum over r > 0, that the boundary-to-domain generalized double layer
potential operator T from (5.1.4) satisfies
 ∫  1/p
1  
sup ∇T f  p dist(·, ∂Ω) p−1 dL n ≤C f M p, λ (∂Ω,σ) .
x ∈∂Ω rλ B(x,r)∩Ω
r ∈(0,∞)
(5.1.300)

This is a generalization of the fractional Carleson measure estimates on Morrey

spaces from Theorem 3.3.3.
It is clear from the above discussion that this machinery extends to other (Morrey-
like) spaces, such as generalized Morrey spaces, and generalized Orlicz-Morrey
spaces. One can also control the mean oscillations of a function in terms of its
Morrey-Campanato semi-norm. This is analyzed below:
Example 5.1.13 Again, start by fixing p ∈ . (1, ∞) along with some λ ∈ (0, n − 1).
This time, pick an arbitrary function f ∈ L p,λ (∂Ω, σ) and define

φ : (0, ∞) → [0, ∞), φ(t) := 2 f L. p, λ (∂Ω,σ) · t (λ−n+1)/p for each t > 0.

(5.1.301)
Consequently, φ satisfies the conditions in (5.1.284), (5.1.286), and in view of
(5.1.298) it follows that φ defined as in (5.1.285) is presently given by
586 5 Generalized Double Layers in Uniformly Rectifiable Domains
 2p 
φ(r) := f L. p, λ (∂Ω,σ) · r (λ−n+1)/p for each r > 0. (5.1.302)
n−1−λ+p

Then (5.1.287) implies, after first dividing by r (λ−n+1)/p and then taking the supremum
over r > 0, that
 ∫  1/p
1  p
sup ∇T f  dist(·, ∂Ω) p−1 dL n ≤ C f L. p, λ (∂Ω,σ) .
x ∈∂Ω rλ B(x,r)∩Ω
mod

r ∈(0,∞)
(5.1.303)
This is a generalization of the fractional Carleson measure estimate on Morrey-
Campanato spaces from Theorem 3.2.4.
Furthermore, a similar result is valid for generalized Morrey-Campanato spaces.
We conclude with an example in which the mean oscillations are controlled in terms
of the Hölder semi-norm.
Example 5.1.14 To set .the stage, fix some exponent α ∈ (0, 1) along with some
arbitrary function f ∈ 𝒞α (∂Ω), and introduce

φ : (0, ∞) → [0, ∞), φ(t) := f 𝒞. α (∂Ω) · (2t)α for each t > 0. (5.1.304)

Note that φ satisfies the conditions in (5.1.284), (5.1.286), and since

∫ ∞
dt rα
r tα 2 = for each r > 0, (5.1.305)
r t 1−α

we conclude that φ defined as in (5.1.285) is presently given by

 2α 
φ(r) := f 𝒞. α (∂Ω) · r α for each r > 0. (5.1.306)
1−α
Then (5.1.287) implies, after first dividing by r α and then taking the supremum over
r > 0, that
 ∫  1/p
1  p
sup ∇T
mod f  dist(·, ∂Ω) p−1 dL n ≤ C f 𝒞. α (∂Ω) .
x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,∞)
(5.1.307)
This corresponds precisely to the fractional Carleson measure estimate on Hölder
spaces from item (10) in Theorem 5.1.1 (see (5.1.62)).
We wish to augment the “space-less” Carleson-type estimate for boundary-to-
domain modified generalized double layer operators established in Theorem 5.1.8
with a companion “space-less” estimate15 for boundary-to-boundary modified gen-
eralized double layer operators, of the sort described in the theorem below.
15 in the sense that said estimate does not involve any recognizable norm, employed in standard
function spaces
5.1 Theory of Generalized Double Layers 587

Theorem 5.1.15 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an arbitrary UR

domain. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. For a sufficiently large integer N = N(n) ∈ N, consider
a vector-valued function
 n
k ∈ 𝒞 N (Rn \ {0}) odd, positive homogeneous of degree 1 − n,
(5.1.308)
and satisfying div k = 0 in Rn \ {0}.

With the set Ω and the kernel k, associate as in (5.1.65) the boundary-to-boundary
modified generalized double layer operator Tmod , i.e., the mapping sending each

f ∈ L 1 ∂Ω, 1+σ(x)
|x | n into the function defined at σ-a.e. point x ∈ ∂Ω according to
∫
 
Tmod f (x) := lim+ ν(y), kε (x − y) − k1 (−y) f (y) dσ(y) (5.1.309)
ε→0
∂Ω

where
kε := k · 1Rn \B(0,ε) for each ε > 0. (5.1.310)
Then for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which depends only
 and the UR constants of ∂Ω, with the property that for each function
on n, p, k,
 σ(x)  p
f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ), (5.1.311)
1 + |x| n

each point xo ∈ ∂Ω, and each radius r ∈ (0, ∞) one has

⨏  ⨏  p  1/p
 
Tmod f − Tmod f dσ  dσ (5.1.312)
Δ(x o ,r) Δ(x o ,r)
∫ ∞ ⨏  ⨏  p  1/p dλ
 
≤C f − f dσ  dσ ,
1 Δ(x o ,λr) Δ(x o ,λr) λ2

where Δ(xo, R) := B(xo, R) ∩ ∂Ω for each R ∈ (0, ∞).

As a consequence of (5.1.312) and the definition of the Fefferman-Stein sharp
maximal operator (cf. (A.0.195)), it follows that for each p ∈ (1, ∞) there exists some
constant C ∈ (0, ∞) such that the following pointwise inequality holds:
 #
Tmod f ≤ C · fp# on ∂Ω, for every
p
  (5.1.313)
function f ∈ L 1 ∂Ω , 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ).

From [70, (2.3.35)] we know that for each f as in (5.1.311) the function Tmod f is
p
well-defined and belongs to the space Lloc (∂Ω, σ). As such, the estimate in (5.1.312)
is meaningfully formulated.
588 5 Generalized Double Layers in Uniformly Rectifiable Domains

Proof of Theorem 5.1.15 Fix a function f as in (5.1.311), along with some point
xo ∈ ∂Ω and a radius r ∈ (0, ∞). Decompose
 
Tmod f = Tmod ( f − fΔ(xo,2r) ) · 1∂Ω\Δ(xo,2r)
 
+ Tmod ( f − fΔ(xo,2r) ) · 1Δ(xo,2r) + Tmod fΔ(xo,2r), (5.1.314)
⨏
where fΔ(xo,2r) := Δ(x ,2r) f dσ. For each fixed x ∈ Δ(xo, r) we may then further
o
decompose
 
Tmod ( f − fΔ(xo,2r) ) · 1∂Ω\Δ(xo,2r) (x) = g(x) + C (1)
f , (5.1.315)

where
∫
 
g(x) :=  − y) − k(x
ν(y), k(x  o − y) f (y) − fΔ(xo,2r) dσ(y) (5.1.316)
∂Ω\Δ(x o ,2r)

and C (1)
f is a constant (depending on f , xo , r), defined as
∫
 
C (1)
f :=
 o − y) − k1 (−y)
ν(y), k(x f (y) − fΔ(xo,2r) dσ(y). (5.1.317)
∂Ω\Δ(x o ,2r)

Then
⨏  p1
|g(x)| p dσ(x)
Δ(x o ,r)

∫
1  
=  − y) − k(x
ν(y), k(x  o − y) ×
 1
σ Δ(xo, r) p ∂Ω\Δ(x o ,2r)


× f (y) − fΔ(xo,2r) dσ(y)
L p (Δ(x o ,r),σ)
∫ ∫ 1
1  
≤ k(x − y) − k(x
 o − y) p dσ(x) p ×
 1
σ Δ(xo, r) p ∂Ω\Δ(x o ,2r) Δ(x o ,r)

 
×  f (y) − fΔ(xo,2r)  dσ(y)
∫  
 f (y) − fΔ(x ,2r) 
≤ Cr
o
dσ(y)
∂Ω\Δ(x o ,2r) |y − xo | n
∫ ∞ ⨏  ⨏  p  1/p dλ
 
≤C f − f dσ  dσ . (5.1.318)
1 Δ(x o ,λr) Δ(x o ,λr) λ2
5.1 Theory of Generalized Double Layers 589

Above, the first step is obvious, the second one is Minkowski’s inequality, the third
one uses the Mean Value Theorem and the properties of k,  and the fourth is comes
from the first inequality in [68, (7.4.115)] (used with X := ∂Ω, ρ := | · − · |, μ := σ,
p := 1, q := p, ε := 1, d := n − 1, and r replaced by 2r).
Next, since the function ( f − fΔ(xo,2r) ) · 1Δ(xo,2r) belongs to L p (∂Ω, σ), from
(5.1.67) we see that there exists a constant C (2)
f , depending on f , xo , r, such that

   
Tmod ( f − fΔ(xo,2r) ) · 1Δ(xo,2r) = T ( f − fΔ(xo,2r) ) · 1Δ(xo,2r) + C (2)
f . (5.1.319)

In addition,
⨏     p  p1

T ( f − fΔ(xo,2r) ) · 1Δ(xo,2r)  dσ
Δ(x o ,r)

1  
≤  1
T ( f − fΔ(xo,2r) ) · 1Δ(xo,2r)
L p (∂Ω,σ)
σ Δ(xo, r) p

C 
≤  f − fΔ(xo,2r) · 1Δ(xo,2r)
1
L p (∂Ω,σ)
σ Δ(xo, r) p

⨏   1
≤C  f − fΔ(x ,2r)  p dσ p
o
Δ(x o ,2r)
∫ ∞ ⨏  ⨏  p  1/p dλ
 
≤C  f − f dσ  dσ , (5.1.320)
1 Δ(x o ,λr) Δ(x o ,λr) λ2

where the first inequality is obvious, the second one is a consequence of the bound-
edness of T on L p (∂Ω, σ) (see (5.1.17)), the third one uses the fact that σ is a
doubling measure on ∂Ω, and the last one is implied by [68, (7.4.125)]. Finally, we
recall from (5.1.69) that

C (3)
f := Tmod fΔ(x o ,2r) (5.1.321)

is a constant, depending on f , xo , r. If we now introduce

C f := C (1) (2) (3)

f + Cf + Cf , (5.1.322)

it follows that C f is a constant, depending on f , xo , r, and from (5.1.314)-(5.1.321)

we see that
590 5 Generalized Double Layers in Uniformly Rectifiable Domains
⨏  ⨏  p  1/p
 
Tmod f − Tmod f dσ  dσ (5.1.323)
Δ(x o ,r) Δ(x o ,r)
⨏    1/p
≤2 T f − C f  p dσ
mod
Δ(x o ,r)
∫ ∞ ⨏  ⨏  p  1/p dλ
 
≤C f − f dσ  dσ ,
1 Δ(x o ,λr) Δ(x o ,λr) λ2

as wanted. 

It is of interest to record a user-friendly version of Theorem 5.1.15, to the effect

that control of the mean oscillations of a given function at a certain scale implies
(a suitably correlated) control of the mean oscillations for any modified generalized
double layer acting on said function. This is done in our next corollary.

Corollary 5.1.16 Suppose Ω is an arbitrary UR domain in Rn (where n ∈ N with

n ≥ 2). Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. For  a sufficiently large integer N = N(n) ∈ N, consider
a vector-valued function k ∈ 𝒞 N (Rn \ {0}) which is odd, positive homogeneous
n

of degree 1 − n, and divergence-free in Rn \ {0}. With the set Ω and the kernel
 associate the boundary-to-boundary modified generalized double layer operator
k,

Tmod acting on functions from the space L 1 ∂Ω, 1+σ(x)
|x | n as in (5.1.65)-(5.1.66). Next,
assume
φ : (0, ∞) → [0, ∞) is a function satisfying
∫ ∞
1  dt (5.1.324)
φ ∈ Lloc (0, ∞), L 1 and φ(t) 2 < +∞,
1 t
then define φ : (0, ∞) → [0, ∞) by setting
∫ ∞
dt
φ(r) := r φ(t) 2 for each r ∈ (0, ∞). (5.1.325)
r t

Finally, fix some p ∈ (1, ∞) and assume f ∈ Lloc

1 (∂Ω, σ) is such that

⨏  ⨏  p  1/p
 
sup f − f dσ  dσ ≤ φ(r) for each r ∈ (0, ∞), (5.1.326)
x ∈∂Ω Δ(x,r) Δ(x,r)

where Δ(x, r) := B(x, r) ∩ ∂Ω. 

Then f belongs to the space L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) and there exists a
constant C ∈ (0, ∞), which depends only on n, p, k,  and the UR constants of ∂Ω,
with the property that for each r ∈ (0, ∞) one has
⨏  ⨏  p  1/p
 
sup Tmod f − Tmod f dσ  dσ ≤ C φ(r). (5.1.327)
x ∈∂Ω Δ(x,r) Δ(x,r)
5.1 Theory of Generalized Double Layers 591

Proof That f belongs to the space L 1 ∂Ω, 1+σ(x)|x | n is seen from (5.1.289), while
p
the membership of f to Lloc (∂Ω, σ) is clear from (5.1.324) and (5.1.326). Final-
ly, (5.1.327) follows from (5.1.312), (5.1.326), a natural change of variables, and
(5.1.325). 

Retaining the background context employed in Corollary 5.1.16, we with to

elaborate on the estimate (5.1.327) in a series of examples, included below. In the
first such example, we start by controlling the mean oscillations of a given function
in terms of its BMO semi-norm.

Example 5.1.17 Fix an arbitrary function f ∈ BMO(∂Ω, σ) and define

φ : (0, ∞) → [0, ∞), φ(t) := f . for each t > 0. (5.1.328)

BMO (∂Ω,σ)
∫∞
Hence, φ satisfies the conditions in (5.1.324), (5.1.326), and since r r
dt
t2
= 1 it
follows that φ defined as in (5.1.325) is currently given by

φ(r) := f . for each r > 0. (5.1.329)

BMO (∂Ω,σ)

Then (5.1.327) implies, after also taking the supremum over all r ∈ (0, ∞), that for
each p ∈ (1, ∞) we have
⨏  ⨏  p  1/p
  .
sup Tmod f − Tmod f dσ  dσ ≤ C f BMO (∂Ω,σ)
.
x ∈∂Ω, r >0 Δ(x,r) Δ(x,r)
(5.1.330)

Ultimately, this provides a new proof of the fact that Tmod is a well-defined, linear,
and bounded operator on BMO(∂Ω, σ) (cf. (5.1.82)).

We next consider the scenario in which one controls the mean oscillations of a
given function in terms of its Morrey-Campanato semi-norm (a similar result is valid
for generalized Morrey-Campanato spaces).

. some p ∈ (1, ∞) along with some λ ∈ (0, n − 1). Pick

Example 5.1.18 Start by fixing
an arbitrary function f ∈ L p,λ (∂Ω, σ) and define

φ : (0, ∞) → [0, ∞), φ(t) := 2 f L. p, λ (∂Ω,σ) · t (λ−n+1)/p for each t > 0.

(5.1.331)
Note that φ satisfies the conditions in (5.1.324), (5.1.326), and in view of (5.1.298)
it follows that φ defined as in (5.1.325) is presently given by
 2p 
φ(r) := f L. p, λ (∂Ω,σ) · r (λ−n+1)/p for each r > 0. (5.1.332)
n−1−λ+p

Then the estimate recorded in (5.1.327) (λ−n+1)/p and

 implies, after first dividing by r
then taking the supremum over r ∈ 0, 2 diam ∂Ω , that
592 5 Generalized Double Layers in Uniformly Rectifiable Domains

⨏  ⨏ p  1
 
≤ C f L. p, λ (∂Ω,σ) .
n−1−λ p
sup r p Tmod f − Tmod f dσ  dσ
x ∈∂Ω and Δ(x,r) Δ(x,r)
0<r <2 diam ∂Ω
(5.1.333)
In view of (A.0.118), this gives a new proof of the fact that Tmod is a well defined,
.
linear, and bounded operator on L p,λ (∂Ω, σ) (cf. (5.1.106)).
In our final example, the mean oscillations are controlled in terms of the Hölder
semi-norm.
Example 5.1.19 To set .the stage, fix some exponent α ∈ (0, 1) along with some
arbitrary function f ∈ 𝒞α (∂Ω), and introduce

φ : (0, ∞) → [0, ∞), φ(t) := f 𝒞. α (∂Ω) · (2t)α for each t > 0. (5.1.334)

Observe that φ satisfies the conditions in (5.1.324), (5.1.326), and in view of (5.1.305)
we conclude that φ defined as in (5.1.325) is currently given by
 2α 
φ(r) := f 𝒞. α (∂Ω) · r α for each r > 0. (5.1.335)
1−α
Then (5.1.327) implies, after first dividing by r α and then taking the supremum over
r ∈ (0, ∞), that
⨏ ⨏ 
1    p  1/p
 
sup α Tmod f − Tmod f dσ  dσ ≤ C f 𝒞. α (∂Ω) .
x ∈∂Ω and r Δ(x,r) Δ(x,r)
0<r <∞
(5.1.336)
Bearing in mind the characterization of the class of Hölder functions from [68,
Proposition 7.4.9] (whose present applicability is guaranteed by [68, Lemma 3.6.4]
used with s := n − 1), this ultimately provides a new proof of the fact that Tmod is a
.
well defined, linear, and bounded operator on 𝒞α (∂Ω) (see (5.1.93)).
All the above examples are subsumed by the following corollary, generalizing
work in the entire Euclidean setting from [83].
Corollary 5.1.20 Suppose Ω is an arbitrary UR domain in Rn (where n ∈ N with
n ≥ 2). Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. For  aNsufficiently large integer N = N(n) ∈ N, consider
 n
a vector-valued function k ∈ 𝒞 (R \ {0}) which is odd, positive homogeneous
n

of degree 1 − n, and divergence-free in Rn \ {0}. With the set Ω and the kernel
 associate the boundary-to-boundary modified generalized double layer operator
k,

Tmod acting on functions from the space L 1 ∂Ω, 1+σ(x)
|x | n as in (5.1.65)-(5.1.66). Next,
assume
φ : (0, ∞) → (0, ∞) is L 1 -measurable and there exists C ∈ (0, ∞)
∫ ∞ (5.1.337)
dt
such that r φ(t) 2 ≤ Cφ(r) for each r ∈ (0, ∞).
r t
5.1 Theory of Generalized Double Layers 593

Finally, fix some p ∈ (1, ∞) and introduce the generalized BMO space16

BMOφ, p (∂Ω, σ) := f ∈ Lloc

1
(∂Ω, σ) : f BMO φ, p (∂Ω,σ) < +∞ , (5.1.338)

where for each f ∈ Lloc

1 (∂Ω, σ) one defines

⨏ ⨏ 
1    p  1/p
 
f BMO φ, p (∂Ω,σ) := sup f − f dσ  dσ .
x ∈∂Ω and φ(r) Δ(x,r) Δ(x,r)
r ∈(0,∞)
(5.1.339)

Then one has

 σ(x)  p
BMOφ, p (∂Ω, σ) ⊆ L 1 ∂Ω, ∩ Lloc (∂Ω, σ), (5.1.340)
1 + |x| n
and the operator

Tmod : BMOφ, p (∂Ω, σ) −→ BMOφ, p (∂Ω, σ) (5.1.341)

is well defined, linear, and bounded.

Parenthetically, we wish to note that functional inequalities in the spirit of the

inequality in (5.1.337) are discussed in [102, Lemma 2.3, p. 515] and [81, Lemma 2,
p. 98].
Let us also remark that, whenever φ is as in (5.1.337), we may employ notation
introduced in (5.1.338)-(5.1.339) to recast (5.1.287) as the following generalized
Carleson measure estimate17 valid for each function f ∈ BMOφ, p (∂Ω, σ):
 ∫  1/p 
1 1  p
sup  ∇T f  dist(·, ∂Ω) p−1 dL n
x ∈∂Ω and φ(r) σ Δ(x, r) B(x,r)∩Ω mod
r ∈(0,∞)

≤C f BMO φ, p (∂Ω,σ) . (5.1.342)

It is relevant to remark that the brand of BMO space introduced in (5.1.338)-

(5.1.339) may be considered on an arbitrary closed Ahlfors regular set Σ ⊆ Rn
(rather than the topological boundary of a UR domain), and for an arbitrary function
φ : (0, ∞) → (0, ∞). Specifically, with σ := H n−1 Σ, for each given p ∈ [1, ∞) set

BMOφ, p (Σ, σ) := f ∈ Lloc

1
(Σ, σ) : f BMO φ, p (Σ,σ) < +∞ (5.1.343)

where, for each f ∈ Lloc

1 (Σ, σ), we define

16 alternatively, (φ, p)-BMO space, or simply φ-BMO space

17 alternatively, (φ, p)-Carleson measure estimate, or simply φ-Carleson measure estimate
594 5 Generalized Double Layers in Uniformly Rectifiable Domains
⨏ ⨏ 
1    p  1/p
 
f BMO φ, p (Σ,σ) := sup f − f dσ  dσ . (5.1.344)
x ∈Σ and φ(r) Δ(x,r) Δ(x,r)
r ∈(0,∞)

In relation to this we wish to note that if the function φ is quasi-increasing and

doubling (see (5.1.349)) then the space BMOφ, p (Σ, σ) is actually independent of
p ∈ [1, ∞). More specifically, with φ : (0, ∞) → (0, ∞) arbitrary define

BMOφ (Σ, σ) := f ∈ Lloc

1
(Σ, σ) : f BMO φ (Σ,σ) < +∞ , (5.1.345)

where, for each f ∈ Lloc

1 (Σ, σ), we have set

⨏ ⨏ 
1    
 
f BMO φ (Σ,σ) := sup f − f dσ  dσ . (5.1.346)
x ∈Σ and φ(r) Δ(x,r) Δ(x,r)
r ∈(0,∞)

Let us now assume that

there exist two constants C1, C2 ∈ (1, ∞) such that
(5.1.347)
φ(t) ≤ C1 φ(r) for each r ∈ (0, ∞) and t ∈ (0, C2 r).

Via iterations, this is equivalent to asking that

 ∈ (0, ∞) such that

for each given C ∈ (0, ∞) there exists C
(5.1.348)

φ(t) ≤ Cφ(r) whenever r ∈ (0, ∞) and t ∈ (0, Cr).

It is also straightforward to check that (5.1.347) (or (5.1.348)) is equivalent to

demanding that

φ is quasi-increasing (i.e., there exists C ∈ (0, ∞) such that

φ(t1 ) ≤ Cφ(t2 ) whenever 0 < t1 ≤ t2 < ∞), and φ is dou-
(5.1.349)
bling (i.e., there exist C1, C2 ∈ (1, ∞) such that φ(C1 t) ≤ C2 φ(t)
for each t ∈ (0, ∞)).
The claim we make is that if we assume (5.1.347) (or, equivalently, (5.1.348), or
(5.1.349)) then

BMOφ, p (Σ, σ) = BMOφ (Σ, σ) for each p ∈ (1, ∞)

(5.1.350)
(equal vector spaces with equivalent semi-norms).

To justify (5.1.350), fix some p ∈ (1, ∞) along with some function f ∈ Lloc 1 (Σ, σ).

Also, pick an arbitrary x ∈ Σ together with an

 arbitrary r ∈ (0, ∞). Consider a surface
ball Δ ⊆ Δ(x, 5r) and having a radius t ∈ 0, 2 diam Σ . Then the Ahlfors regularity
of Σ implies that t < C3 r for some C3 ∈ (0, ∞) which depends only on n and the
Ahlfors regularity constants of Σ. Bearing this in mind, we may then estimate
5.1 Theory of Generalized Double Layers 595
⨏  ⨏  p  1/p
1  
f− f dσ  dσ
φ(r) Δ(x,r) Δ(x,r)
 ⨏  ⨏  
C  
≤ sup f − f dσ  dσ
φ(r) Δ ⊆Δ(x,5r) Δ Δ

C  
≤ sup φ(t) · f BMO φ (Σ,σ)
φ(r) t ∈(0,C3 r)

≤C f BMOφ (Σ,σ), (5.1.351)

thanks to [68, Lemma 7.4.10], the observation made in [68, (7.4.61)] (according
to which the supremum in the first line of (5.1.351) may be taken over all surface
balls Δ on Σ contained in Δ(x, 5r) and having radii < 2 diam Σ), and the additional
assumption made in (5.1.348). After taking the supremum over all x ∈ Σ and
r ∈ (0, ∞) we arrive at

f BMO φ, p (Σ,σ) ≤C f BMO φ (Σ,σ) . (5.1.352)

The opposite inequality in (5.1.352) is also true (with C = 1), as a consequence of

Hölder’s inequality. Altogether, this establishes the claim made in (5.1.350).
For example,

φ : (0, ∞) → (0, ∞) given by φ(t) := 1 for each t ∈ (0, ∞)

has the properties described in (5.1.337), (5.1.347), and (5.1.353)
 
BMOφ (∂Ω, σ) = BMO(∂Ω, σ), · BMO . ,
(∂Ω,σ)

while for each α ∈ (0, 1),

φ : (0, ∞) → (0, ∞) with φ(t) := t α for all t > 0

is as in (5.1.337), (5.1.347), and we have (5.1.354)
.α
BMOφ (∂Ω, σ) = 𝒞 (∂Ω).

On the other hand, for each p ∈ (1, ∞) and λ ∈ (0, n − 1),

φ : (0, ∞) → (0, ∞) with φ(t) := t (λ−n+1)/p for each t > 0

. (5.1.355)
is as in (5.1.337) and BMOφ, p (∂Ω, σ) = L p,λ (∂Ω, σ).

This being said, the function φ from (5.1.355) fails to satisfy (5.1.347) (specifically,
.
φ fails to be quasi-increasing), and the space BMOφ, p (∂Ω, σ) = L p,λ (∂Ω, σ) is
strongly dependent on the choice of the parameter p ∈ (1, ∞).
We now turn to the task of presenting the proof of Corollary 5.1.20.
Proof of Corollary 5.1.20 The inclusion (5.1.340) is a consequence of the first
conclusion in Corollary 5.1.16. As regards (5.1.341), pick an arbitrary function
596 5 Generalized Double Layers in Uniformly Rectifiable Domains

f ∈ BMOφ, p (∂Ω, σ) and note that this entails


⨏  ⨏  p  1/p
 
sup f − f dσ  dσ ≤ f BMO φ, p (∂Ω,σ) · φ(r) (5.1.356)
x ∈∂Ω Δ(x,r) Δ(x,r)

for each r ∈ (0, ∞). Granted this, we may then invoke (5.1.327), (5.1.325), and the
inequality in (5.1.337) to write

Tmod f BMO φ, p (∂Ω,σ)

⨏ ⨏ 
1    p  1/p
 
= sup Tmod f − Tmod f dσ  dσ
x ∈∂Ω and φ(r) Δ(x,r) Δ(x,r)
r ∈(0,∞)

≤C f BMO φ, p (∂Ω,σ) . (5.1.357)

This shows that the operator (5.1.341) is indeed well defined and bounded. 

We close this section with a result essentially stating that modified boundary-to-
domain generalized double layers map CMO functions into (certain densities of)
super vanishing Carleson measures (a notion introduced in Definition 1.8.1). We
shall actually show more in Theorem 5.1.22, stated a little further below. To set the
stage, we first prove the following lemma.

Lemma 5.1.21 Pick n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is a UR domain for

which ∂Ω is unbounded. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric
measure theoretic outward unit normal to Ω. For a sufficiently large N = N(n) ∈ N,
consider a vector-valued function
 n
k ∈ 𝒞 N (Rn \ {0}) odd, positive homogeneous of degree 1 − n,
(5.1.358)
and satisfying div k = 0 in Rn \ {0}.

With the set Ω and the kernel k, associate as in (5.1.44) the modified generalized

double layer operator Tmod , i.e., the mapping sending each f ∈ L 1 ∂Ω , 1+σ(x)|x | n into
the function defined at each point x ∈ Ω according to
∫  

Tmod f (x) :=  − y) − k(−y)
ν(y) , k(x  · 1Rn \B(0,1) (y) f (y) dσ(y). (5.1.359)
∂Ω
∞ (∂Ω, σ) one has
Then for each p ∈ (1, ∞) and each function f ∈ Lcomp
 ∫  1/p
  p
lim sup  1 ∇ T  dist (·, ∂Ω) p−1 dL n =0
mod f
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω B(x,r)∩Ω
(5.1.360)
and
5.1 Theory of Generalized Double Layers 597
 ∫  1/p
  p
lim sup  1 ∇ T  dist (·, ∂Ω) p−1 dL n
mod f
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω B(x,r)∩Ω
B(x,r)⊆R n \B(0,R)

= 0. (5.1.361)

Proof Start by fixing an integrability exponent p ∈ (1, ∞) and pick an arbitrary

∞ (∂Ω, σ). To simplify notation, let Δ denote an arbitrary surface
function f ∈ Lcomp
ball on ∂Ω, i.e., a set of the form Δ = B(x,
⨏ r) ∩ ∂Ω for some x ∈ ∂Ω and r > 0. For
any such surface ball abbreviate fΔ := Δ f dσ. We claim that

there exists some constant C f ∈ (0, ∞) such that

⨏    1/p Cf (5.1.362)
 f − fΔ  p dσ ≤ for all Δ.
Δ σ(Δ) 1/p

Indeed, since f is compactly supported and bounded there exists some surface ball
Δ∗ ⊆ ∂Ω (depending on f ) such that

|f| ≤ f L ∞ (∂Ω,σ) · 1Δ∗ pointwise on ∂Ω. (5.1.363)

Hence, for every Δ we may estimate

⨏  1/p  σ(Δ )  1/p
∗
| f | p dσ ≤ f L ∞ (∂Ω,σ) . (5.1.364)
Δ σ(Δ)
Consequently, invoking the triangle inequality and Hölder’s inequality it follows that,
for every Δ,
⨏    1/p ⨏  1/p
 f − fΔ  p dσ ≤2 | f | p dσ
Δ Δ
 σ(Δ )  1/p
∗
≤2 f L ∞ (∂Ω,σ) . (5.1.365)
σ(Δ)

From this the claim in (5.1.362) follows by choosing C f := 2σ(Δ∗ )1/p · f L ∞ (∂Ω,σ) .
Next, given any ball B centered on ∂Ω, say B := B(x, r) for some x ∈ ∂Ω and
r > 0, we agree to abbreviate

λ B := B(x, λr) for each λ > 0,

ΔB := Δ(x, r) := B(x, r) ∩ ∂Ω, (5.1.366)
and CB := B(x, r) ∩ Ω.

Fix now a ball B centered on ∂Ω and define the function

⨏    1/p
FB (λ) :=  f − fΔ  p dσ , ∀ λ ∈ (0, ∞). (5.1.367)
λB
ΔλB
598 5 Generalized Double Layers in Uniformly Rectifiable Domains

Then, by (5.1.362) and the Ahlfors regularity of ∂Ω, there exists a finite constant
C f > 0 such that

Cf Cf
FB (λ) ≤ ≤ for each λ ∈ (0, ∞). (5.1.368)
σ(ΔλB )1/p λ
n−1
p σ(ΔB )1/p
By also invoking (5.1.262), it follows that
 ∫  1/p
1   p
∇ T  dist (·, ∂Ω) p−1 dL n
mod f
σ(ΔB ) CB
∫ ∞ ∫ ∞
dλ Cf 1 dλ
≤C FB (λ) ≤ ·
1 λ2 σ(ΔB )1/p 1 λ
n−1
p λ2

Cf
= . (5.1.369)
σ(ΔB )1/p
In particular, (5.1.369) implies (5.1.360), bearing in mind that ∂Ω is an Ahlfors
regular set.
Our next goal is to show that
 
for each fixed λ > 0, one has lim sup FB (λ) = 0. (5.1.370)
R→∞ B ⊆R n \B(0,R)
B ball centered on ∂Ω

To see why this is true, fix λ > 0 and pick ε ∈ (0, ∞) arbitrary. Let R > 0 and
(C ) p
consider a ball B in Rn \ B(0, R) centered on ∂Ω. If σ(ΔB ) ≥ ε p ·λf n−1 , then by
(C ) p
(5.1.368) we have FB (λ) ≤ ε. On the other hand, if σ(ΔB ) ≤ ε p ·λf n−1 , then the radius
of the ball B is bounded by a constant and since B ⊆ Rn \ B(0, R) we must have that
λB is disjoint from the support of f provided R > R0 for some sufficiently large R0
(relative to λ, f , ε, n, p, and ∂Ω). This forces FB (λ) = 0 when B ⊆ Rn \ B(0, R)
with R ≥ R0 . Consequently, supB ⊆Rn \B(0,R) FB (λ) ≤ ε for R ≥ R0 . Since ε ∈ (0, ∞)
is arbitrary, (5.1.370) follows.
Having proved (5.1.370), let us fix R > 0 and use (5.1.262) to write
 ∫  1/p
1   p
sup ∇ T  dist (·, ∂Ω) p−1 dL n
mod f
B ⊆R n \B(0,R) σ(ΔB ) CB
B ball centered on ∂Ω

∫ ∞
dλ
≤C· sup FB (λ)
B ⊆R n \B(0,R) 1 λ2
B ball centered on ∂Ω

∫  
∞
dλ
≤C sup FB (λ) . (5.1.371)
1 B ⊆R n \B(0,R) λ2
B ball centered on ∂Ω
5.1 Theory of Generalized Double Layers 599

It is clear from (5.1.367) and (5.1.365) that for each λ > 0,

 
sup sup FB (λ) ≤ 2 f L ∞ (∂Ω,σ), (5.1.372)
R>0 B ⊆R n \B(0,R)
B ball centered on ∂Ω

and we note that the constant 2 f L ∞ (∂Ω,σ) is absolutely integrable on (1, ∞) with
respect to the measure dλ
λ2
. The latter, (5.1.372), (5.1.370), and Lebesgue’s Dominated
Convergence Theorem imply
∫ ∞ 
dλ
lim sup FB (λ) 2 = 0. (5.1.373)
R→∞ 1 B ⊆R \B(0,R)
n λ
B ball centered on ∂Ω

In turn, (5.1.373) may be used in connection with (5.1.371) to conclude that (5.1.361)
holds. 

Before stating our final result in this section, the reader is advised to recall the
notion of super vanishing Carleson measure we have introduced in Definition 1.8.1.

Theorem 5.1.22 Pick n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is a UR domain with

∂Ω an unbounded set. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric
measure theoretic outward unit normal to Ω. For a sufficiently large N = N(n) ∈ N,
consider a vector-valued function
 n
k ∈ 𝒞 N (Rn \ {0}) odd, positive homogeneous of degree 1 − n,
(5.1.374)
and satisfying div k = 0 in Rn \ {0}.

With the set Ω and the kernel k, associate as in (5.1.44) the modified generalized

double layer operator Tmod , i.e., the mapping sending each f ∈ L 1 ∂Ω , 1+σ(x)|x | n into
the function defined at each point x ∈ Ω according to
∫  

Tmod f (x) :=  − y) − k(−y)
ν(y) , k(x  · 1Rn \B(0,1) (y) f (y) dσ(y). (5.1.375)
∂Ω

Then for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which depends only
 and the UR constants of ∂Ω, with the property that for each function
on n, p, k,
f ∈ BMO(∂Ω, σ) one has
600 5 Generalized Double Layers in Uniformly Rectifiable Domains


max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→0+ x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω

∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ 
  p  1/p
× ∇ T  dist (·, ∂Ω) p−1 dL n
mod f
B(x,r)∩Ω


≤ C dist f , CMO(∂Ω, σ) , (5.1.376)

where the distance in the right-hand side is considered in BMO(∂Ω, σ). As a conse-
quence of (5.1.376) and Definition 1.8.1,
  
∇ T f  p dist (·, ∂Ω) p−1 dL n is a super vanishing Carleson measure in Ω,
mod

for each function f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).

(5.1.377)

Proof Fix an exponent p ∈ (1, ∞) and a function f ∈ BMO(∂Ω, σ). Then, for each
g ∈ Lipc (∂Ω) we have f − g ∈ BMO(∂Ω, σ) and, as already proved in (5.1.227),
 ∫  p1
  p
lim sup 
1 ∇ T f  dist (·, ∂Ω) p−1 dL n
mod
R→0+ x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω B(x,r)∩Ω

≤C f −g BMO(∂Ω,σ), (5.1.378)

for some constant C ∈ (0, ∞) independent of f and g. On the other hand, the
triangle inequality and estimate (5.1.74) show that there exists a constant C ∈ (0, ∞),
independent of f and g, with the property that for each x ∈ ∂Ω and r > 0 we have
 ∫  1/p
  
 1 ∇ T f  p dist (·, ∂Ω) p−1 dL n
mod
σ B(x,r)∩∂Ω B(x,r)∩Ω
 ∫  1/p
  
≤  1 ∇ T g  p dist (·, ∂Ω) p−1 dL n
mod
σ B(x,r)∩∂Ω B(x,r)∩Ω

+C f −g BMO(∂Ω,σ) . (5.1.379)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 601

Taking now the supremum over x ∈ ∂Ω and r > 0 such that B(x, r) ⊆ Rn \ B(0, R) of
both sides of (5.1.379) and then passing to limit as R → ∞ in the resulting inequality
yields, on account of (5.1.361),
 ∫  1/p
  
lim sup  1 ∇ T f  p dist (·, ∂Ω) p−1 dL n
mod
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω B(x,r)∩Ω
B(x,r)⊆R n \B(0,R)

≤C f −g BMO(∂Ω,σ) . (5.1.380)

Likewise, taking the supremum over x ∈ ∂Ω and r > R of both sides of (5.1.379),
and passing to limit as R → ∞ in the resulting inequality we obtain, this time relying
on (5.1.360),
 ∫  1/p
  
lim sup  1 ∇ T f  p dist (·, ∂Ω) p−1 dL n
mod
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω B(x,r)∩Ω

≤C f −g BMO(∂Ω,σ) . (5.1.381)

Now (5.1.376) follows by combining (5.1.378), (5.1.380), (5.1.381) and the fact
that the space CMO(∂Ω, σ) is the closure of Lipc (∂Ω) in BMO(∂Ω, σ) (cf. [69,
(4.6.13)]). Finally, (5.1.377) is clear from (5.1.376) and Definition 1.8.1, so the
proof of Theorem 5.1.22 is complete. 

5.2 Generalized Double Layers with Matrix-Valued Kernels, and

Chord-Dot-Normal SIO’s

Our first result in this section may be regarded as a version of Theorem 5.1.1 for
matrix-valued kernels.
Theorem 5.2.1 Fix a dimension n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an
arbitrary UR domain18. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric
measure theoretic outward unit normal to Ω. Let D be a homogeneous first-order
M × M  system (where M, M  ∈ N are arbitrary) with constant complex coefficients

n 
αβ
D= a j ∂j 1≤α ≤M (5.2.1)
j=1 1≤β ≤M 

(with the summation convention over repeated indices in effect) and recall that its
(principal) symbol is the M × M  matrix-valued function
 
αβ
Sym(D; ξ) := i a j ξ j 1≤α ≤M for each ξ = (ξ1, . . . , ξn ) ∈ Rn . (5.2.2)
1≤β ≤M 

18 When Ω ⊆ R n is just an open set with a UR boundary, a large number of conclusions are still
valid, with at most minor alterations (namely, eventually replacing ∂Ω with ∂∗ Ω).
602 5 Generalized Double Layers in Uniformly Rectifiable Domains

Having fixed some M  ∈ N together with a sufficiently large integer N = N(n) ∈ N,

consider a matrix-valued function
   M  ×M 
Θ = Θβγ 1≤β ≤M  ∈ 𝒞 N (Rn \ {0}) (5.2.3)
1≤γ ≤M 

which is odd, positive homogeneous of degree 1 − n in Rn \ {0}, and satisfies (with

D acting on the columns of Θ)

D Θ = 0 · I M×M  in Rn \ {0}. (5.2.4)

Finally, set
∫

ϑ := (−i)Sym(D; ω)Θ(ω) dH n−1 (ω) ∈ C M×M . (5.2.5)
S n−1
   M
In this setting, for each vector-valued function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1 define
∫  
T f (x) := (−i)Sym(D; ν(y))Θ(x − y) f (y) dσ(y) for all x ∈ Ω, (5.2.6)
∂Ω

and
∫  
T f (x) := lim+ (−i)Sym(D; ν(y))Θ(x − y) f (y) dσ(y) (5.2.7)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, where the superscript  indicates (real) transposition. In addition,
    M 
for each given vector-valued function g ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
consider
∫
T # f (x) := lim+ (−i)Sym(D; ν(x))Θ(y − x)g(y) dσ(y) (5.2.8)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω.

Then, for each fixed aperture parameter κ > 0, the following statements are true.
   M
(1) For each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 one has

 κ−n.t. 
 1
T f (x) = − ϑ f (x) + (T f )(x) at σ-a.e. x ∈ ∂Ω. (5.2.9)
∂Ω 2
    M 
Also, if for each f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1 one defines
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 603
∫
W f (x) := Θ(x − y) f (y) dσ(y) for all x ∈ Ω, (5.2.10)
∂Ω
    M 
then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 one has

  M 
W f ∈ 𝒞 N (Ω) and D W f = 0 ∈ C M in Ω. (5.2.11)
   M 
Moreover, for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 the nontangential boundary trace

κ−n.t.
(W f )∂Ω exists (in C M ) at σ-a.e. point on ∂Ω,

(5.2.12)

and at σ-a.e. point x ∈ ∂Ω one has

  κ−n.t. 
 1
(−i)Sym D; ν(x) W f  (x) = − ϑ f (x) − (T # f )(x). (5.2.13)
∂Ω 2
Finally, if ∂Ω is compact then for each λ ∈ C M both T λ and T λ are constant
functions and, in fact, at each point in Ω we have

−ϑ λ if Ω is bounded,
Tλ =  (5.2.14)
0∈ CM if Ω is an exterior domain19,
while at σ-a.e. point on ∂Ω we have

− 12 ϑ λ if Ω is bounded,
Tλ = (5.2.15)
+ 12 ϑ λ if Ω is an exterior domain.

(2) For each p ∈ [1, ∞) there exists some finite constant C > 0, depending only on
  M 
∂Ω, Θ, D, n, p, and κ, such that for each function f ∈ L p (∂Ω, σ) one has

max Nκ (T f ) L p (∂Ω,σ)
, Nκ (W f ) L p (∂Ω,σ)
≤C f [L p (∂Ω,σ)] M
 (5.2.16)

if p > 1, plus a similar estimate when p = 1 in which case the corresponding

L 1 -norms in the left-hand side are now replaced by the quasi-norm in the space
 1,∞  M 
L (∂Ω, σ) .
Moreover, the action of the operator W, originally considered as in (5.2.10), may
be further extended in a unique and coherent fashion (cf. [70, (2.4.15), (2.4.16),
  M 
(2.4.24)]) to the scale of Lorentz-based Hardy spaces H p,q (∂Ω, σ) with
 n−1
p ∈ n , ∞ and q ∈ (0, ∞] and for each κ > 0 said extension satisfies (for
some constant C = C(∂Ω, Θ, D, n, p, q, κ) ∈ (0, ∞))

Nκ (W f ) L p, q (∂Ω,σ)
≤C f [H p, q (∂Ω,σ)] M


  M  (5.2.17)
for all f ∈ H p,q (∂Ω, σ) .
604 5 Generalized Double Layers in Uniformly Rectifiable Domains

(3) For each p ∈ (1, ∞), the operators

 M   M 
T : L p (∂Ω, σ) → L p (∂Ω, σ) , (5.2.18)
  M   M
T # : L p (∂Ω, σ) → L p (∂Ω, σ) , (5.2.19)

are well defined, linear, and bounded. Also, given any p, p ∈ (1, ∞) with 1/p +
1/p = 1 it follows that
 M   M 
the transpose of T : L p (∂Ω, σ) → L p (∂Ω, σ)
   M    M (5.2.20)
is the operator T # : L p (∂Ω, σ) → L p (∂Ω, σ) .

Corresponding to the end-point p = 1, both operators T and T # map vector-

valued functions with components in L 1 (∂Ω, σ) linearly and continuously into
  in L (∂Ω, σ). Next, if for each vector-
vector-valued functions 1,∞
 with
 components n
valued function g ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
one considers
∫
V g(x) := lim+ (−i)Sym(D; g(y))Θ(x − y) dσ(y) (5.2.21)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, then for each p ∈ (1, ∞) the following operator is well
defined, linear, and bounded:
 n   M×M 
V : L p (∂Ω, σ) −→ L p (∂Ω, σ) . (5.2.22)

Finally, similar results are valid for Muckenhoupt weighted Lebesgue spaces,
Lorentz spaces, and Morrey spaces (as well as their duals and their preduals)
on ∂Ω.
 
(4) Fix p ∈ n−1 n , 1 . Then the operator T , originally considered as in (5.2.19),
#

extends to a linear and bounded mapping on Hardy spaces, i.e.,

  M   M
T # : H p (∂Ω, σ) −→ H p (∂Ω, σ) (5.2.23)
 
and the operators corresponding to various choices of the index p ∈ n−1 n ,1
are compatible with one another. In fact, T # maps the scale of Lorentz-based
Hardy spaces on ∂Ω continuously into itself, specifically
  M   M
T # : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) is well defined,
 n−1 (5.2.24)
linear, and bounded for p ∈ n , ∞ and q ∈ (0, ∞].
  M  
In addition, for each f ∈ H p,q (∂Ω, σ) with p ∈ n−1
n , ∞ and q ∈ (0, ∞]
one has
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 605

1
(−i)Sym(D; ν) • W f = − ϑ f − T # f , (5.2.25)
2
where the “bullet product” is defined as in [69, (10.2.94) in Proposition 10.2.11].
Finally, for each function f ∈ H 1 (∂Ω, σ) one has
  M   M
f ∈ L 1 (∂Ω, σ) , T # f ∈ L 1 (∂Ω, σ) , (5.2.26)

and, with all integrals involved absolutely convergent,

∫
⎧
⎪ 1
⎪−
⎪ ϑ f dσ if Ω is bounded,
∫ ⎪
⎪ 2 ∂Ω
⎪
⎨
⎪ ∫
T # f dσ = + 1 ϑ f dσ if Ω is unbounded and ∂Ω is bounded,
∂Ω ⎪
⎪
⎪ 2 ∂Ω
⎪
⎪
⎪
⎪ 0 ∈ C M if ∂Ω is unbounded.
⎩
(5.2.27)

(5) Pick q ∈ (1, ∞) and λ ∈ (0, n − 1) and recall ℋq,λ (∂Ω, σ), the pre-dual to
the Morrey-Campanato space, defined as in (A.0.84) (with Σ := ∂Ω). Then the
operator
  M   M q(n − 1)
T # : L r (∂Ω, σ) −→ L r (∂Ω, σ) with r := (5.2.28)
n − 1 + λ(q − 1)
induces a well-defined, linear, and bounded mapping in the context
  M   M
T # : ℋq,λ (∂Ω, σ) −→ ℋq,λ (∂Ω, σ) . (5.2.29)
  M
(6) For each given function f in the space L11 ∂Ω, 1+σ(x) |x | n−1
, the vector ver-
sion of the weighted boundary Sobolev space defined in (A.0.131), each index
∈ {1, . . . , n}, and each aperture parameter κ > 0, the pointwise nontangential
boundary trace
 κ−n.t.
∂ T f ∂Ω exists (in C M ) at σ-a.e. point on ∂Ω.

(5.2.30)
 p,q  M 
As a consequence of [69, (11.7.22)], this is true whenever f ∈ L1 (∂Ω, σ)
with p, q ∈ [1, ∞).
Furthermore, for each p, q ∈ [1, ∞) and κ > 0 there exists some finite constant
C > 0, depending only on ∂Ω, Θ, D, n, p, q, and κ, such that for each function
 p,q  M 
f ∈ L1 (∂Ω, σ) one has

Nκ (T f ) L p (∂Ω,σ)
+ Nκ (∇T f ) L q (∂Ω,σ)
≤C f p, q
[L1 (∂Ω,σ)] M
 (5.2.31)
606 5 Generalized Double Layers in Uniformly Rectifiable Domains

if p, q > 1, plus similar estimates in the case when p = 1 or q = 1, in which

scenario the corresponding L 1 -norms in the left-hand side are now replaced by
the quasi-norm L 1,∞ (∂Ω, σ).
(7) The operator
 p,q M  p,q  M 
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.32)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
 p M  p  M 
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.33)

is well defined, linear, and bounded for each p ∈ (1, ∞). Actually,
 p M  p  M 
T : L1 (∂Ω, w) −→ L1 (∂Ω, w) is well defined, linear, bounded
for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ).
(5.2.34)

Moreover, for each function

  σ(x)  M
p
f ∈ L11 ∂Ω, ∩ L (∂Ω, σ) for some p ∈ (1, ∞) (5.2.35)
1 + |x| n−1 1,loc

it follows that
 p  M 
T f ∈ L1,loc (∂Ω, σ) (5.2.36)
and for each pair of indices r, s ∈ {1, . . . , n} at σ-a.e. point on ∂Ω one has (with
V as in (5.2.21))
    
∂τr s (T f ) = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
+ Mνr , V (νs ∇tan f ) − Mνs , V (νr ∇tan f ). (5.2.37)
 p,q M
In particular, formula (5.2.37) holds for every function f ∈ L1 (∂Ω, σ)
 p M
with p, q ∈ (1, ∞), as well as for every function f ∈ L1 (∂Ω, w) with p in
(1, ∞) and w ∈ Ap (∂Ω, σ). Formula (5.2.37) also holds for each function in
 p,λ M  q,λ M
M1 (∂Ω, σ) or B1 (∂Ω, σ) with p, q ∈ (1, ∞) and λ ∈ (0, n − 1) (cf.
(A.0.150), (A.0.33)). In fact,
 p,λ M  p,λ  M 
T : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) , (5.2.38)
 p,λ M  p,λ  M 
T : M̊1 (∂Ω, σ) −→ M̊1 (∂Ω, σ) , (5.2.39)
 q,λ M  q,λ  M 
T : B1 (∂Ω, σ) −→ B1 (∂Ω, σ) , (5.2.40)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 607

are all well-defined, linear, and continuous operators whenever p, q ∈ (1, ∞)

and λ ∈ (0, n − 1) (cf. (A.0.150), (A.0.155), (A.0.33)).
(8) For each p ∈ (1, ∞) it follows that T # , originally acting as in (5.2.19), further
extends uniquely to a linear, bounded operator on negative boundary Sobolev
spaces:
 p  M   p M
T # : L−1 (∂Ω, σ) → L−1 (∂Ω, σ) . (5.2.41)

Furthermore, if one retains the same notation T # for said extension, then the
transpose of (5.2.33) is
 p  M   p M
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.2.42)

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

More generally, for each p, q ∈ (1, ∞) it follows that T # acting as in (5.2.19)
further extends uniquely to a linear and bounded operator from the negative
 p,q  M   p,q M
off-diagonal boundary Sobolev space L−1 (∂Ω, σ) into L−1 (∂Ω, σ)
and, adopting the same notation T # for this extension, the transpose of (5.2.32)
is
 p,q  M   p,q M
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.2.43)

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1.
Also, for each integrability exponent p ∈ (1, ∞) and each Muckenhoupt weight
  M 
w ∈ Ap (∂Ω, σ), it follows that T # , originally acting from L p (∂Ω, w) into
 p M
L (∂Ω, w) , further extends uniquely to a linear, bounded operator, from the
 p  M   p M
negative boundary Sobolev space L−1 (∂Ω, w) into L−1 (∂Ω, w) which,
 p 
 M   p M
in fact, is the transpose of T acting from L1 (∂Ω, w ) into L1 (∂Ω, w )
where p := (1 − 1/p)−1 ∈ (1, ∞) is the Hölder conjugate exponent of p and

w  := w 1−p ∈ Ap (∂Ω, σ) is the conjugate weight of w (cf. [68, item (2) in
Lemma 7.7.1]).
(9) Consider the following modified version of the generalized double layer potential
  M
operator in (5.2.6), acting on each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n according
to

Tmod f (x) (5.2.44)
∫ 
! "
:= (−i)Sym(D; ν(y)) Θ(x − y) − Θ(−y) · 1Rn \B(0,1) (y) f (y) dσ(y)
∂Ω

for all x ∈ Ω. Then the operator Tmod is meaningfully defined, and is compatible
with T from (5.2.6) in the sense that for each f belonging to the smaller space
 1 M  M
L ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular, for each function f ∈ L p (∂Ω, σ)
608 5 Generalized Double Layers in Uniformly Rectifiable Domains

with p ∈ [1, ∞)) the difference


C f := Tmod f − T f is a constant (belonging to C M ) in Ω. (5.2.45)

As a consequence,
  σ(x)   M
∇Tmod f = ∇T f in Ω for each f ∈ L 1 ∂Ω, . (5.2.46)
1 + |x| n−1
Moreover,

Tmod maps (C M -valued) constant functions on ∂Ω

 (5.2.47)
into (C M -valued) constant functions in Ω.

In addition, at each point x ∈ Ω one may express

∫  

∂ Tmod f (x) = (−i)Sym(D; ν(y))(∂ Θ)(x − y) f (y) dσ(y)
∂Ω
(5.2.48)
 1 σ(y) M
for each ∈ {1, . . . , n} and f ∈ L ∂Ω, 1+ |y | n .

Also,
 κ−n.t.
∂ Tmod f ∂Ω exists (in C M ) at σ-a.e. point on ∂Ω,


  σ(x)   M
for each ∈ {1, . . . , n} and each f ∈ L 1 ∂Ω, satisfying
1 + |x| n
  σ(x)  M
∂τ j k f ∈ L 1 ∂Ω, for all j, k ∈ {1, . . . , n}.
1 + |x| n−1
(5.2.49)
Furthermore, for each truncation parameter ε ∈ (0, ∞) and each integrability
exponent p ∈ (1, ∞) it follows that

Nκε ∇(Tmod f ) ∈ Lloc (∂Ω, σ) for each function
p

  σ(x)   M
f ∈ L 1 ∂Ω, with the property that
1 + |x| n
  σ(x)  M
p
∂τ j k f ∈ L 1 ∂Ω, ∩ L (∂Ω, σ) for all j, k ∈ {1, . . . , n},
1 + |x| n−1 loc
(5.2.50)
and
Nκε (Tmod f ) ∈ Lloc (∂Ω, σ) for each function
p

  σ(x)  M (5.2.51)
p
f ∈ L 1 ∂Ω, ∩ L (∂Ω, σ) .
1 + |x| n loc
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 609
 
Finally, for each p ∈ n−1 n , 1 and each aperture parameter κ ∈ (0, ∞) there
exists some finite constant C > 0, depending only on ∂Ω, n, Θ, κ, and p, with
 .p M
the property that for each distribution f ∈ H1 (∂Ω, σ) one has

Nκ (∇Tmod f ) L p (∂Ω,σ)
≤C f .p
[ H1 (∂Ω,σ)] M
. (5.2.52)

(10) For each α ∈ (0, 1) there exists a constant C ∈ (0, ∞) with the property that
  
sup dist(x, ∂Ω)1−α ∇ Tmod f (x) ≤ C f . (5.2.53)
[𝒞α (∂Ω)] M
x ∈Ω
. M
for every function f ∈ 𝒞α (∂Ω) . Moreover,

if Ω ⊆ Rn is a uniform domain with the property that ∂Ω is

. M  .   M 
an Ahlfors regular set then Tmod : 𝒞α (∂Ω) → 𝒞α Ω
(5.2.54)
is a well-defined, linear, and bounded operator for each given
exponent α ∈ (0, 1),
whereas
if Ω ⊆ Rn is an NTA domain with an upper Ahlfors regular
.α M  . α   M 
boundary then Tmod : 𝒞van (∂Ω) → 𝒞van Ω is a well-
(5.2.55)
defined, linear, and bounded operator for each given exponent
α ∈ (0, 1),
with the vanishing Hölder spaces defined as in (A.0.48) (with Σ := ∂Ω and
Σ := Ω, respectively). Also, for each α ∈ (0, 1) and each p ∈ (1, ∞) there exists
. M
some C ∈ (0, ∞) with the property that for each function f ∈ 𝒞α (∂Ω) one
has
 ∫  1/p
1  p
sup ∇T f  dist(·, ∂Ω) p−1
dL n
n−1+αp mod
x ∈∂Ω r B(x,r)∩Ω
r ∈(0,∞)

≤C f . (5.2.56)
[𝒞α (∂Ω)] M

and
 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
 .α M
≤ C dist f , 𝒞van (∂Ω) , (5.2.57)
 . M 
where the distance is measured in the space 𝒞α (∂Ω) , · [𝒞. α (∂Ω)] M . As
a corollary, if the function f actually belongs to the homogeneous vanishing
.α M
Hölder space 𝒞van (∂Ω) for some α∈(0, 1), then for each p∈(1, ∞) one has
610 5 Generalized Double Layers in Uniformly Rectifiable Domains
 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n = 0.
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
(5.2.58)
(11) Let Tmod be the modified version of the singular integral operator (5.2.7), acting
  M
on each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n according to20
∫ 
! " 
Tmod f (x) := lim+ (−i)Sym(D; ν(y)) Θε (x − y) − Θ1 (−y) f (y) dσ(y)
ε→0
∂Ω
(5.2.59)

at σ-a.e. x ∈ ∂Ω, where

Θε := Θ · 1Rn \B(0,ε) for each ε > 0. (5.2.60)

  M
Then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n one has the jump-formula
κ−n.t.
 1
Tmod f  = − ϑ f + Tmod f at σ-a.e. point on ∂Ω. (5.2.61)
∂Ω 2
.p M
In particular, (5.2.61) holds for each f ∈ L1 (∂Ω, σ) with 1 < p < ∞. As a
consequence of (5.2.61) and (5.2.47),

Tmod maps (C M -valued) constant functions on ∂Ω

 (5.2.62)
into (C M -valued) constant functions on ∂Ω.

Moreover,

if p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 then given

   M
any functions f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, 1+σ(x)
p
|x | n and
∫
 p  M  
g ∈ Lcomp (∂Ω, σ) satisfying g dσ = 0 ∈ C M , (5.2.63)
∫ ∫∂Ω
one has |Tmod f ||g| dσ < +∞, | f ||T # g| dσ < +∞,
∂Ω ∂Ω
∫ ∫
as well as Tmod f , g dσ = f , T # g dσ.
∂Ω ∂Ω

Lastly, given any p ∈ (1, ∞), for each function

20 alternative descriptions of Tmod are seen from [70, Proposition 2.3.3]

5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 611
  M
f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) such that
  M
∂τ j k f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n−1
∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n}
(5.2.64)
.p M
(hence for each f ∈ L1 (∂Ω, σ) if Ω also satisfies a two-sided local John
condition) it follows that
 p  M 
Tmod f ∈ L1,loc (∂Ω, σ) (5.2.65)

and for each pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.2.21))
     
∂τr s Tmod f = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
+ Mνr , V (νs ∇tan f ) − Mνs , V (νr ∇tan f ) (5.2.66)

at σ-a.e. point on ∂Ω. In particular, formula (5.2.66) holds for every function
 p,q M
f ∈ L1 (∂Ω, σ) with exponents p, q ∈ (1, ∞), as well as for every function
 p M
f ∈ L1 (∂Ω, w) with p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ).
(12) For each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property that
 M   p
for each f ∈ BMO(∂Ω, σ) the measure ∇ Tmod f  dist(·, ∂Ω) p−1 dL n is
Carleson in Ω in the quantitative sense that
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω

≤C f
p
. . (5.2.67)
[BMO(∂Ω,σ)] M

In particular, corresponding to p = 2, it follows that21

  2
∇ T f  dist(·, ∂Ω) dL n is a Carleson measure in Ω,
mod
 M (5.2.68)
for each function f ∈ BMO(∂Ω, σ) .

Moreover, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
 M
that for each function f ∈ BMO(∂Ω, σ) one has
 ∫  p1
1   p
lim+ sup  ∇ T f  dist(·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
mod

r ∈(0,R)

≤ C dist f , [VMO(∂Ω, σ)] M (5.2.69)

 2
21 it is natural to regard  ∇Tmod f  dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated with
f via the operator Tmod
612 5 Generalized Double Layers in Uniformly Rectifiable Domains
 M
where the distance in the right-hand side is considered in BMO(∂Ω, σ) . As
a corollary,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
mod
 M
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞)
(5.2.70)
and, corresponding to p = 2,
  2
∇ T f  dist(·, ∂Ω) dL n is a vanishing Carleson measure in Ω,
mod
 M (5.2.71)
for each function f ∈ VMO(∂Ω, σ) .

Finally, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
 M
that for each function f ∈ BMO(∂Ω, σ) one has


max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→0+ x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω

∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ 
  p  1/p
× ∇ T  dist (·, ∂Ω) p−1 dL n
mod f
B(x,r)∩Ω


≤ C dist f , [CMO(∂Ω, σ)] M , (5.2.72)
 M
where the distance in the right-hand side is considered in BMO(∂Ω, σ) . As
a consequence of (5.1.376) and Definition 1.8.1,
  
∇ T f  p dist (·, ∂Ω) p−1 dL n is a super vanishing Carleson measure in Ω,
mod
 M
for each function f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).
(5.2.73)
(13) Make the additional assumption that ∂Ω is bounded. Then all properties listed
in items (9)-(12) above are valid for the operator T , as originally defined in
(5.2.6), in place of its modified version Tmod . In particular, for each p ∈ (1, ∞)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 613

there exists a constant C ∈ (0, ∞) with the property that for each function
 M
f ∈ BMO(∂Ω, σ) one has
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω

≤C f
p
. . (5.2.74)
[BMO(∂Ω,σ)] M

and
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤ C dist f , [VMO(∂Ω, σ)] M (5.2.75)
 M
where the distance in the right-hand side is considered in BMO(∂Ω, σ) . In
particular,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
 M
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).
(5.2.76)
(14) The operators
 M   M 
Tmod : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) and (5.2.77)
 M   M 
Tmod : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (5.2.78)

are well defined, linear, and bounded. As a corollary, (5.2.77)-(5.2.78) and

(5.2.62) imply that both
    M    M 
Tmod : BMO(∂Ω, σ) ∼ −→ BMO(∂Ω, σ) ∼ defined as
     M
Tmod [ f ] := Tmod f for each function f ∈ BMO(∂Ω, σ)
(5.2.79)

and
    M    M 
Tmod : VMO(∂Ω, σ) ∼ −→ VMO(∂Ω, σ) ∼ defined as
     M
Tmod [ f ] := Tmod f for each function f ∈ VMO(∂Ω, σ)
(5.2.80)

are well-defined, linear, and bounded operators. Moreover, if ∂Ω is unbounded

then
614 5 Generalized Double Layers in Uniformly Rectifiable Domains
    
Tmod f , g = [ f ], T # g ,
 M   M  (5.2.81)
∀ f ∈ BMO(∂Ω, σ) , ∀g ∈ H 1 (∂Ω, σ)
where ·, · denotes the duality bracket between the John-Nirenberg space of
functions of bounded mean oscillations on ∂Ω, modulo constants, and the Hardy
space H 1 on ∂Ω (cf. [69, Theorem 4.6.1]), the operator Tmod is as in (5.2.77),
and T # is presently considered as in (5.2.23) with p = 1. In addition, if ∂Ω is
unbounded, the operators
 M   M 
Tmod : CMO(∂Ω, σ) −→ CMO(∂Ω, σ) , (5.2.82)

and
    M    M 
Tmod : CMO(∂Ω, σ) ∼ −→ CMO(∂Ω, σ) ∼ defined as
     M
Tmod [ f ] := Tmod f for each function f ∈ CMO(∂Ω, σ) ,
(5.2.83)

are also well defined, linear, and bounded.

Furthermore, in the case when the set ∂Ω is bounded (a scenario in which the
inclusion BMO(∂Ω, σ) ⊆ L p (∂Ω, σ) holds for each p ∈ (1, ∞)), the operator T
acting as in (5.2.18) induces a well-defined, linear, and bounded mapping
 M   M 
T : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) (5.2.84)

which satisfies
   
T f , g = f , T #g ,
 M   M  (5.2.85)
∀ f ∈ BMO(∂Ω, σ) , ∀g ∈ H 1 (∂Ω, σ)

where ·, · denotes the duality bracket between the John-Nirenberg space of

functions of bounded mean oscillations on ∂Ω and the Hardy space H 1 on ∂Ω
(cf. [69, Theorem 4.6.1]), and T # is presently considered as in (5.2.23) with
p = 1.
Finally, when ∂Ω is bounded the operator T acting as in (5.2.84) induces a
linear and bounded mapping
 M   M 
T : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) . (5.2.86)
 n−1 1
(15) Fix p ∈ n , 1 and set α := (n − 1) p − 1 ∈ (0, 1). Then the operators
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 615
. M .  M 
Tmod : 𝒞α (∂Ω) −→ 𝒞α (∂Ω) , (5.2.87)
.α M .α  M 
Tmod : 𝒞van (∂Ω) −→ 𝒞van (∂Ω) , (5.2.88)
.α  M   .   M 
𝒞 (∂Ω) ∼  [ f ] → Tmod f ∈ 𝒞α (∂Ω) ∼ (5.2.89)

are well defined, linear, bounded. In particular, (5.2.87)-(5.2.89) imply that

  .  M .   M 
Tmod : 𝒞α (∂Ω) ∼ −→ 𝒞α (∂Ω) ∼ defined as
    .α M (5.2.90)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞 (∂Ω)

and
  .α  M .α   M 
Tmod : 𝒞van (∂Ω) ∼ −→ 𝒞van (∂Ω) ∼ defined as
    .α M (5.2.91)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞van (∂Ω)

are well-defined, linear, and bounded operators. Also, if ∂Ω is unbounded one

has     
Tmod f , g := [ f ], T # g ,
. M   M  (5.2.92)
∀ f ∈ 𝒞α (∂Ω) , ∀g ∈ H p (∂Ω, σ) ,
where ·, · denotes the duality bracket between a homogeneous Hölder space
on ∂Ω, modulo constants, and the corresponding Hardy space on ∂Ω (cf. [69,
Theorem 4.6.1]), the operator Tmod is as in (5.2.87), and where #
 n−1the operator T
is presently considered as in (5.2.23) with exponent p ∈ n , 1 . Finally, in
. α situation when
the the set. ∂Ω is bounded (a scenario in which one actually has
𝒞van α (∂Ω) ⊆ 𝒞α (∂Ω) = 𝒞α (∂Ω) ⊆ L p (∂Ω, σ) for each p ∈ (1, ∞)),
(∂Ω) = 𝒞van
the operator T acting as in (5.2.18) induces well-defined, linear and bounded
mappings
 α M  α  M 
T : 𝒞van (∂Ω) −→ 𝒞van (∂Ω) (5.2.93)
and  M   M 
T : 𝒞α (∂Ω) −→ 𝒞α (∂Ω) (5.2.94)
the latter of which satisfies
     M   M 
T f , g = f , T #g , ∀ f ∈ 𝒞α (∂Ω) , ∀g ∈ H p (∂Ω, σ) (5.2.95)

with ·, · denoting the duality bracket between Hölder and Hardy spaces on ∂Ω
(cf. [69, Theorem 4.6.1]),the operator T is as in (5.2.94), and the operator T #
is as in (5.2.23) with p ∈ n−1
n ,1 .
(16) Select p, q ∈ (1, ∞) with 1/p + 1/q = 1 along with λ ∈ (0, n − 1), and recall the
Morrey-Campanato space L p,λ (∂Ω, σ) (defined as in (A.0.119) with Σ := ∂Ω).
Then the operator T from (5.2.18) induces a linear and bounded mapping in the
context
616 5 Generalized Double Layers in Uniformly Rectifiable Domains
 M 
T : L p,λ (∂Ω, σ) , · [L p, λ (∂Ω,σ)] M

  M  
→ L p,λ (∂Ω, σ) , · [L p, λ (∂Ω,σ)] M
 (5.2.96)

Moreover, if Tmod is the modified version of the singular integral operator T

defined in (5.2.59), the assignment
.  M .   M 
[Tmod ] : L p,λ (∂Ω, σ) ∼ −→ L p,λ (∂Ω, σ) ∼
. M (5.2.97)
[Tmod ][ f ] := [Tmod f ] for each f ∈ L p,λ (∂Ω, σ)

is well defined, linear and bounded. In particular,

. M .  M 
Tmod : L p,λ (∂Ω, σ) → L p,λ (∂Ω, σ) is a well-defined, lin-
ear, operator and there exists some constant C ∈ (0, ∞) with the
property that Tmod f [L. p, λ (∂Ω,σ)] M  ≤ C f [L. p, λ (∂Ω,σ)] M for all (5.2.98)
. M
f ∈ L p,λ (∂Ω, σ) .

Finally,

the (real)
 transpose
 of the operator T # from (5.2.29) is, respec-
tively, Tmod from (5.2.97) if ∂Ω is unbounded, and T from (5.2.99)
(5.2.96) if ∂Ω is bounded,
. M   M 
that is, for each f ∈ L p,λ (∂Ω, σ) and g ∈ ℋq,λ (∂Ω, σ) one has
    
Tmod f , g = [ f ], T # g if ∂Ω is unbounded, (5.2.100)
   
T f , g = f , T # g if ∂Ω is bounded, (5.2.101)

where ·, · denotes the duality bracket between vector Morrey-Campanato

spaces and their vector pre-duals (cf. [69, (6.1.25)]).
(17) Fix q ∈ (1, ∞) and η ∈ (0, 1) and recall the Calderón spaces defined in (3.1.10)-
(3.1.15). Then there exists a constant C ∈ (0, ∞) with the property that, with the
maximal operator defined as in (3.1.1) (with Σ := ∂Ω), one has
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tmod f (5.2.102)
   M
for every function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n . As a consequence, given any expo-
nent p ∈ [1, ∞), it follows that
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 617
 .p  M  .p   M 
[Tmod ] : Cq,η (∂Ω, σ) ∼ −→ Cq,η (∂Ω, σ) ∼
   .p M (5.2.103)
defined by [Tmod ][ f ] := Tmod f for each f ∈ Cq,η (∂Ω, σ)

is a well-defined, linear, and bounded operator. Moreover,

 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tf (5.2.104)
 M
for every function f ∈ L p (∂Ω, σ) with p ∈ [1, ∞). Finally, whenever the
exponent p ∈ (1, ∞) it follows that
 p M  p  M 
T : Cq,η (∂Ω, σ) −→ Cq,η (∂Ω, σ) (5.2.105)

is a well-defined, linear, and bounded operator.

(18) Assume
1
∈ {1, . . . , N }, n−1
n < p ≤ ∞ and (n − 1) p −1 + < s < 1. (5.2.106)

Then for each θ ∈ (0, 1) there exists a finite C = C(Ω, Θ, D, , p, s, θ) > 0 such
that
− 1 −s  
δ∂Ω p ∇ T f  ≤C f p, p (5.2.107)
mod ,θ L p (Ω, L n ) [B s (∂Ω,σ)] M

 p, p M
for all f ∈ Bs (∂Ω, σ)  , with the understanding
  when p > 1 the solid
that
maximal function ∇ Tmod f  ,θ is replaced by ∇ Tmod f .
(19) Make the additional assumption that ∂Ω is compact. Then the operator T,
originally acting on Lebesgue spaces on ∂Ω (cf. (5.2.18)), extends uniquely to
linear and bounded mappings
 p,q M  p,q  M 
T : Bs (∂Ω, σ) −→ Bs (∂Ω, σ) ,
 n−1  1 (5.2.108)
p ∈ n , ∞ , q ∈ (0, ∞], (n − 1) p − 1 + < s < 1,

and
 p,q M  p,q  M 
T : Fs (∂Ω, σ) −→ Fs (∂Ω, σ) ,
 n−1  n−1   1 (5.2.109)
p ∈ n , ∞ , q ∈ n , ∞ , (n − 1) min{p,q } − 1 + < s < 1.

Moreover, various choices of the exponents yield operators which are compatible
with one another. In addition, the operator T # , originally considered acting on
Lebesgue spaces on ∂Ω (cf. (5.2.19)) further extends, in a unique fashion, to
linear and bounded mappings
618 5 Generalized Double Layers in Uniformly Rectifiable Domains
 p,q  M   p,q M
T # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (5.2.110)
with s ∈ (0, 1), p ∈ n−1 n−s , ∞], q ∈ (0, ∞],

and
 p,q  M   p,q M
T # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ) ,
  (5.2.111)
with s ∈ (0, 1), p ∈ n−1 n−s , ∞), q ∈ n−1
n−s , ∞].

Again, various choices of the parameters p, q, s yield operators which are com-
patible with one another. Finally, if the exponents p, q, p, q  ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q  and s ∈ (0, 1), then
 # 
[B−s (∂Ω,σ)] M T f , g [B sp , q  (∂Ω,σ)] M
p, q

 
= [B−s
p, q
(∂Ω,σ)] M
 f , T g p , q 
[B s (∂Ω,σ)] M
 (5.2.112)
 p,q  M   p,q M
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ) ,

and
 # 
p, q
[F−s (∂Ω,σ)] MT f , g [F p, q (∂Ω,σ)] M
s
 
= [F−s
p, q
(∂Ω,σ)] M  f , T g p , q 
[Fs (∂Ω,σ)] M
 (5.2.113)
 p,q  M   p,q M
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ) .

(20) Strengthen the original hypotheses on the underlying domain by assuming that
Ω is a UR domain satisfying a local John condition. Then the operator
.p p M .p p  M 
Tmod : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) → L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
(5.2.114)

is well defined, linear, and bounded, when the spaces involved are endowed
with the semi-norm (A.0.128). In addition, (5.2.66) holds for each function
.p p M
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) .
Finally, under the assumption that Ω ⊆ Rn is an NTA domain with an upper
Ahlfors regular boundary22, for each integrability exponent p ∈ (1, ∞) the
operator
.p M .p  M 
Tmod : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.115)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). In this case, it follows from (5.2.115) and (5.2.62) that
for each p ∈ (1, ∞) the operator

22 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 619
  .p  M .p   M 
Tmod : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
    .p  .p M
Tmod [ f ] := Tmod f ∈ L1 (∂Ω, σ) ∼ for all f ∈ L1 (∂Ω, σ) ,
(5.2.116)
is well defined, linear, and bounded, when all quotient spaces are endowed with
the natural semi-norm23 introduced in [69, (11.5.138)].
(21) Assume Ω is a UR domain satisfying a local John condition24, and fix an
integrability exponent p ∈ (1, ∞) along with a parameter λ ∈ (0, n − 1). Then
the operator
 . p,λ M  . p,λ  M 
Tmod : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) (5.2.117)

is well defined, linear, and bounded, when the spaces involved are endowed
with the semi-norm (A.0.158). As a consequence of (5.2.117) and (5.2.62), the
operator
   . p,λ  M .p   M 
Tmod : M1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
     . p,λ   M   . p,λ M
Tmod [ f ] := Tmod f ∈ M1 (∂Ω, σ) ∼ for all f ∈ M1 (∂Ω, σ)
(5.2.118)
is well defined, linear, and bounded when all quotient spaces are equipped with
the semi-norm25 introduced in [69, (11.13.51)].
. p,λ are valid for vanishing Morrey-based homoge-
Finally, analogous properties
neous Sobolev spaces M1 (∂Ω, σ) (cf. [69, Definition 11.13.15], or (A.0.159)-
. q,λ
(A.0.160)) with p ∈ (1, ∞) and for block-based homogeneous Sobolev spaces
B1 (∂Ω, σ) with q ∈ (1, ∞) in place of Morrey-based homogeneous Sobolev
spaces.
(22) Strengthen the hypotheses on the underlying domain by assuming that Ω ⊆ Rn
is an NTA domain with an upper Ahlfors regular boundary26. Then the modified
boundary-to-boundary operator Tmod (originally defined as in (5.2.59)-(5.2.60))
induces a linear and bounded mapping
 .p M  .p  M   
Tmod : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) for all p ∈ n−1
n ,1 . (5.2.119)

(23) The vector-valued versions of Theorem 5.1.8 and Corollary 5.1.9 are valid for
the boundary-to-domain modified generalized double layer potential operator

23 [69, Proposition 11.5.14] tells us that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
24 for example, this is the case if Ω ⊆ R n is an NTA domain with an upper Ahlfors regular boundary
25 Recall from [69, Proposition 11.13.10] that this semi-norm is actually a genuine norm if Ω ⊆ R n
is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
26 again, this is the case if Ω is an open set satisfying a two-sided local John condition and whose
boundary is Ahlfors regular; cf. (1.8.157)
620 5 Generalized Double Layers in Uniformly Rectifiable Domains

Tmod from (5.2.44), while the vector-valued versions of Theorem 5.1.15, Corol-
lary 5.1.16, and Corollary 5.1.20 are valid for the boundary-to-boundary modi-
fied generalized double layer potential operator Tmod defined in (5.2.59)-(5.2.60).
To offer an example, work under the assumption n = 2 and identify R2 ≡ C.
Consider D := ∂ = 12 (∂x + i∂y ), the Cauchy-Riemann operator in the plane, and take
Θ(z) := −1/(2πz) for each z ∈ C \ {0} (hence, M  = M  = 1, and N = ∞). Then
Θ is odd, positive homogeneous of degree −1 in C \ {0}, and satisfies D Θ = 0 in
C \ {0} (hence (5.2.4) holds). In  this scenario,
 the action of the integral operator
σ(ζ )
(5.2.6) on each function f ∈ L ∂Ω, 1+ |ζ | then becomes
1

∫ ∫
1 f (ζ) 1 f (ζ)
T f (z) = ν(ζ) dσ(ζ) = dζ for all z ∈ Ω. (5.2.120)
2π ∂Ω ζ−z 2πi ∂Ω ζ−z
Thus, T becomes precisely the boundary-to-domain Cauchy integral operator 𝒞
associated with Ω (as in (1.6.35)).
Here is the proof of Theorem 5.2.1.
Proof of Theorem 5.2.1 For each α ∈ {1, . . . , M } and γ ∈ {1, . . . , M  } define the
vector field
 αβ  n
kαγ := a j Θβγ 1≤ j ≤n ∈ 𝒞 N (Rn \ {0}) (5.2.121)

which is odd, positive homogeneous of degree 1 − n, and satisfies

αβ 
div kαγ = a j ∂j Θβγ = D Θ . γ α = 0 in Rn \ {0}, (5.2.122)

thanks to (5.2.4). Associate the integral operators Tαγ , Tαγ , Tαγ# with the kernel kαγ
as in (5.1.4), (5.1.5), and (5.1.6), respectively. Henceforth we also agree to abbreviate
∫
 
ϑαγ := ω, kαγ (ω) dH n−1 (ω) ∈ C. (5.2.123)
S n−1

In particular, from (5.2.5), (5.2.2), and (5.2.121) we see that

∫ 
αβ
ϑ= a j ω j Θβγ (ω) dH n−1 (ω)
S n−1 1≤α ≤M
1≤γ ≤M 
∫ 
 
= ω, kαγ (ω) dH n−1
(ω)
S n−1 1≤α ≤M
1≤γ ≤M 

= ϑαγ 1≤α ≤M . (5.2.124)
1≤γ ≤M 

To proceed, let (ν j )1≤ j ≤n be the scalar components of the geometric measure

theoretic outward unit normal ν to Ω. Then, keeping (5.2.2) in mind, we see that for
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 621
   M
each given vector-valued function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(y) |y | n−1
we may
recast (5.2.6) in terms of the vector fields introduced in (5.2.121) as
∫ 
αβ
T f (x) = a j ν j (y)Θβγ fα (y) dσ(y)
∂Ω
1≤γ ≤M 
∫ 
 
= ν(y), kαγ (x − y) fα (y) dσ(y)
∂Ω
1≤γ ≤M 

= Tαγ fα (x) 1≤γ ≤M  for all x ∈ Ω, (5.2.125)

and we may reformulate (5.2.7) as

 ∫ 
 
T f (x) = lim+ ν(y), kαγ (x − y) fα (y) dσ(y)
ε→0
y ∈∂Ω 1≤γ ≤M 
|x−y |>ε

= Tαγ fα (x) 1≤γ ≤M  for σ-a.e. x ∈ ∂Ω. (5.2.126)
    M 
Similarly, for each vector-valued function g = (gγ )1≤γ ≤M  ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1

we may refashion (5.2.8) as

 ∫ 
 
T g(x) = lim
# 
ν(x), k αγ (y − x) gγ (y) dσ(y)
ε→0+
y ∈∂Ω 1≤α ≤M
|x−y |>ε
 #
= Tαγ gγ (x) 1≤α ≤M for σ-a.e. x ∈ ∂Ω. (5.2.127)

The above identifications make it possible to rely on Theorem 5.1.1 to deal with
the current claims. First, a combination of (5.2.125), (5.2.126), (5.1.7) and (5.2.124)
   M
gives that for each vector-valued function f ∈ L 1 ∂Ω, 1+σ(x)|x | n−1
we have

 κ−n.t.   κ−n.t. 
 
T f (x) = Tαγ fα  (x)
∂Ω ∂Ω 1≤γ ≤M 
 1  
= − ϑαγ fα (x) + Tαγ fα (x)
2 1≤γ ≤M 

1
= − ϑ f (x) + (T f )(x) at σ-a.e. x ∈ ∂Ω, (5.2.128)
2
which establishes (5.2.9). Next, the claims in (5.2.11) are implied by (5.2.10), (5.2.4),
and (5.2.3). Also, (5.2.12) is a direct consequence of assumptions and [70, Theo-
622 5 Generalized Double Layers in Uniformly Rectifiable Domains

rem 2.5.1]. To justify (5.2.13), first note that at σ-a.e. point x ∈ ∂Ω we

have
 1     
(−i)Sym D; ν(x) # ν(x) = (−i)i aαβ ν j (x) 1 Θ
Θ # βγ ν(x)
2i j 2i 1≤α ≤M
1≤γ ≤M 

1 $

= ν(x) · kαγ ν(x) 1≤α ≤M
2i 1≤γ ≤M 

1'  
=− div kαγ ν(x) 1≤α ≤M
2 1≤γ ≤M 
1  1
=− ϑαγ 1≤α ≤M = − ϑ, (5.2.129)
2 1≤γ ≤M  2

thanks to (5.2.2), (5.2.121), (5.2.123), (5.1.168) (keeping in mind that # δ = 1), and
    M 
σ(x)
(5.2.124). Pick now an arbitrary function f ∈ L ∂Ω, 1+ |x | n−1
1 and use [70,
Theorem 2.5.1] to write, at σ-a.e. point x ∈ ∂Ω,

  κ−n.t. 

(−i)Sym D;ν(x) W f  (x)
∂Ω

 1  
= (−i)Sym D; ν(x) # ν(x) f (x)
Θ
2i
 ∫ 
+ (−i)Sym(D; ν(x)) lim+ Θ(x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε

1
= − ϑ f (x) − (T # f )(x), (5.2.130)
2
where the last equality uses (5.2.129) and (5.2.8) (plus the fact that Θ is odd). This
finishes the justification of (5.2.13).
For each α ∈ {1, . . . , M } and γ ∈ {1, . . . , M } consider the operator acting on
n
each vector-valued function g ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
according to
∫
 
Vαγ g(x) := lim+ kαγ (x − y), g(y) dσ(y) for σ-a.e. x ∈ ∂Ω, (5.2.131)
ε→0
y ∈∂Ω
|x−y |>ε

where kαγ is as in (5.2.121). Then, based on (5.2.21),

 (5.2.131), and   n(5.2.121) for
σ(y)
each given vector-valued function g = (g j )1≤ j ≤n ∈ L ∂Ω, 1+ |y | n−1
1 we have
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 623
∫
V g(x) = lim+ (−i)Sym(D; g(y))Θ(x − y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
 ∫ 
αβ
= lim a j Θβγ (x − y)g j (y) dσ(y)
ε→0+ 1≤α ≤M
y ∈∂Ω 1≤γ ≤M 
|x−y |>ε
 ∫ 
 
= lim kαγ (x − y), g(y) dσ(y)
ε→0+ 1≤α ≤M
y ∈∂Ω 1≤γ ≤M 
|x−y |>ε

= Vαγ g(x) 1≤α ≤M for σ-a.e. x ∈ ∂Ω. (5.2.132)
1≤γ ≤M 

In turn, given any r, s ∈ {1, . . . , n} and any function f = ( fα )1≤α ≤M as in (5.2.35),

based on (5.2.126), (5.1.40), and (5.2.132) we may compute

∂τr s (T f ) = ∂τr s (Tαγ fα )γ = ∂τr s Tαγ fα γ
        
= Tαγ ∂τr s fα + Mνr , Tαγ (∇tan fα )s − Mνs , Tαγ (∇tan fα )r
γ γ γ
     
− Mνr , Vαγ (νs ∇tan fα ) + Mνs , Vαγ (νr ∇tan fα )
γ γ
    
= T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
+ Mνr , V (νs ∇tan f ) − Mνs , V (νr ∇tan f ) (5.2.133)

at σ-a.e. point on ∂Ω. This establishes (5.2.37).

All other claims made in the statement of Theorem 5.2.1 are justified in a similar
fashion, based on (5.2.121)-(5.2.127) and the corresponding properties in Theo-
rem 5.1.1. 
Our next theorem amounts to a comprehensive theory (including nontangential
maximal function estimates, jump-formulas, boundedness results on a large variety
of basic function spaces, duality results, Carleson measure estimates, among many
other things), for “chord-dot-normal” singular integral operators in arbitrary UR
domains, of the sort discussed in (5.2.135)-(5.2.137) below.
Theorem 5.2.2 (Chord-Dot-Normal Singular Integral Operators) Fix an integer
n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an arbitrary UR domain27. Abbreviate
σ := H n−1 ∂Ω, denote by ν the geometric measure theoretic outward unit normal
to Ω, and pick an aperture parameter κ > 0. Also, having fixed a sufficiently large
27 If Ω ⊆ R n is merely an open set with a UR boundary, a large number of conclusions continue to
hold, with at most minor alterations (namely, eventually replacing ∂Ω with ∂∗ Ω).
624 5 Generalized Double Layers in Uniformly Rectifiable Domains

integer N = N(n) ∈ N, consider a function k ∈ 𝒞 N (Rn \ {0}) which is even and

positive homogeneous of degree −n, and define
∫
ϑ := k(ω) dH n−1 (ω) ∈ C. (5.2.134)
S n−1

In this setting, introduce chord-dot-normal singular integral operators of the

following sort. For each function f ∈ L 1 ∂Ω, 1+σ(y) |y | n−1
define the boundary-to-
domain integral operator
∫
T f (x) := ν(y), x − y k(x − y) f (y) dσ(y) for all x ∈ Ω, (5.2.135)
∂Ω
 
and for each f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
consider the boundary-to-boundary singular
integral operators
∫
T f (x) := lim+ ν(y), x − y k(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.136)

as well as
∫
T f (x) := lim+
#
ν(x), y − x k(y − x) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω.
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.137)

Then, in relation to these chord-dot-normal singular integral operators, the fol-

lowing statements are true.
 
(1) For each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has

 κ−n.t.  ϑ

T f (x) = − f (x) + (T f )(x) at σ-a.e. x ∈ ∂Ω. (5.2.138)
∂Ω 2
 
In addition, if for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 one defines
∫
W f (x) := f (y)k(x − y)(x − y) dσ(y) for all x ∈ Ω, (5.2.139)
∂Ω
 
then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has
 n 
W f ∈ 𝒞 N (Ω) and div W f = 0 in Ω. (5.2.140)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 625
 
Furthermore, for each κ > 0 and f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, the nontangential
boundary trace
κ−n.t.
(W f )∂Ω exists at σ-a.e. point on ∂Ω, (5.2.141)

and satisfies
 κ−n.t.  ϑ

ν(x) · W f  (x) = − f (x) − (T # f )(x) at σ-a.e. x ∈ ∂Ω. (5.2.142)
∂Ω 2
Finally, if ∂Ω is bounded then both T 1 and T1 are constant functions and, in
fact, at each point in Ω we have

−ϑ if Ω is bounded,
T1 = (5.2.143)
0 if Ω is an exterior domain28,

while at σ-a.e. point on ∂Ω we have

− ϑ2 if Ω is bounded,
T1 = (5.2.144)
+ ϑ2 if Ω is an exterior domain.

(2) For each p ∈ [1, ∞) there exists some finite constant C > 0, depending only on
∂Ω, k, n, p, and κ, such that for each function f ∈ L p (∂Ω, σ) one has

max Nκ (T f ) , Nκ (W f )
L p (∂Ω,σ) L p (∂Ω,σ)
≤C f L p (∂Ω,σ) (5.2.145)

if p > 1, plus similar estimates in the case when p = 1 in which scenario the
corresponding L 1 -norms in the left-hand side are now replaced by the quasi-
norm L 1,∞ (∂Ω, σ).
Moreover, the action of the operator W, originally considered as in (5.2.139),
may be further extended in a unique and coherent fashion (cf. [70, (2.4.15),
(2.4.16), (2.4.24)])
 n−1 to the scale of Lorentz-based Hardy spaces H p,q (∂Ω, σ)
with p ∈ n , ∞ and q ∈ (0, ∞] and for each κ > 0 said extension satisfies
(for some constant C = C(∂Ω, k, n, p, q, κ) ∈ (0, ∞))

Nκ (W f ) L p, q (∂Ω,σ)
≤C f H p, q (∂Ω,σ) (5.2.146)

for all f ∈ H p,q (∂Ω, σ).

(3) For each p ∈ (1, ∞), the operators

T : L p (∂Ω, σ) → L p (∂Ω, σ) and T # : L p (∂Ω, σ) → L p (∂Ω, σ) (5.2.147)

are both well defined, linear, and bounded. Also, given any p, p ∈ (1, ∞) satis-
fying 1/p + 1/p = 1 it follows that
626 5 Generalized Double Layers in Uniformly Rectifiable Domains

the transpose of T acting on L p (∂Ω, σ)

 (5.2.148)
is the operator T # acting on L p (∂Ω, σ).

Corresponding to the end-point p = 1, both operators T and T # map L 1 (∂Ω, σ)

  into L (∂Ω, σ). Next, if for each vector-valued function
linearly 1,∞
 and
 boundedly n
g ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
one considers
∫
V g(x) := lim+ k(x − y) x − y, g(y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.149)

then
 
V( f ν) = T f for each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1 , (5.2.150)

and for each p ∈ (1, ∞) the following operator is well defined, linear, and
bounded:
 n
V : L p (∂Ω, σ) −→ L p (∂Ω, σ) (5.2.151)

Finally, similar results are valid for Muckenhoupt weighted Lebesgue spaces,
Lorentz spaces, and Morrey spaces (as well as their duals and their preduals)
on ∂Ω.
 
(4) Fix p ∈ n−1n , 1 . Then the operator T , originally acting on Lebesgue spaces as
#

in (the first part of) item (3), extends to a linear and bounded mapping from the
Hardy space H p (∂Ω, σ) into itself,

T # : H p (∂Ω, σ) −→ H p (∂Ω, σ) (5.2.152)

 
and the operators corresponding to various choices of the index p ∈ n−1 n ,1
are compatible with one another. In fact, T # maps the scale of Lorentz-based
Hardy spaces on ∂Ω continuously into itself, specifically

T # : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) is well defined,

 (5.2.153)
linear, and bounded for p ∈ n−1 n , ∞ and q ∈ (0, ∞].

As far as the dependence of the operator norm for T # in (5.2.153) on the kernel
k is concerned, homogeneity considerations dictate that
  
T # H p, q (∂Ω,σ)→H p, q (∂Ω,σ) ≤ C sup |∂ α k | , (5.2.154)
n−1
|α | ≤ N S

where the constant C ∈ (0, ∞) depends only on n, p, q, and the UR character
of ∂Ω. In addition, for each f ∈ H p,q (∂Ω, σ) with p ∈ n−1
n , ∞ and q ∈ (0, ∞]
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 627

one has
ϑ
ν•Wf =− f − T# f . (5.2.155)
2
Finally, for each function f ∈ H 1 (∂Ω, σ) one has

f ∈ L 1 (∂Ω, σ), T # f ∈ L 1 (∂Ω, σ), (5.2.156)

and, with all integrals involved absolutely convergent,

∫
⎧
⎪ ϑ
⎪
⎪ − f dσ if Ω is bounded,
∫ ⎪
⎪ 2 ∂Ω
⎪
⎨
⎪ ∫
T # f dσ = ϑ
∂Ω ⎪
⎪ + f dσ if Ω is unbounded and ∂Ω is bounded,
⎪
⎪ 2 ∂Ω
⎪
⎪
⎪ if ∂Ω is unbounded.
⎩ 0
(5.2.157)

(5) Pick q ∈ (1, ∞) and λ ∈ (0, n − 1) and recall ℋq,λ (∂Ω, σ), the pre-dual to
the Morrey-Campanato space, defined as in (A.0.84) (with Σ := ∂Ω). Then the
operator

q(n − 1)
T # : L r (∂Ω, σ) −→ L r (∂Ω, σ) with r := (5.2.158)
n − 1 + λ(q − 1)

has ℋq,λ (∂Ω, σ) as an invariant subspace (cf. (5.2.147) and [69, (6.1.22)]),
and

T # : ℋq,λ (∂Ω, σ) −→ ℋq,λ (∂Ω, σ) (5.2.159)

is a well defined, linear, and bounded operator.

(6) For each given function f belonging to the weighted boundary Sobolev space

L11 ∂Ω, 1+σ(x)
|x | n−1
(cf. (A.0.131)), and each index ∈ {1, . . . , n}, the pointwise
nontangential boundary trace
 κ−n.t.
∂ T f ∂Ω exists at σ-a.e. point on ∂Ω. (5.2.160)
p,q
As a consequence of [69, (11.7.22)], this is true whenever f ∈ L1 (∂Ω, σ) with
p, q ∈ [1, ∞).
Furthermore, for each p, q ∈ [1, ∞) there exists C ∈ (0, ∞), depending only on
p,q
∂Ω, k, n, p, q, and κ, such that for each function f ∈ L1 (∂Ω, σ) one has

Nκ (T f ) L p (∂Ω,σ)
+ Nκ (∇T f ) L q (∂Ω,σ)
≤C f L1
p, q
(∂Ω,σ) (5.2.161)

if p, q > 1, plus similar estimates in the case when p = 1 or q = 1, in which

scenario the corresponding L 1 -norms in the left-hand side are now replaced by
the quasi-norm in L 1,∞ (∂Ω, σ).
628 5 Generalized Double Layers in Uniformly Rectifiable Domains

(7) The operator

p,q p,q
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.162)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
p p
T : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.163)

is well defined, linear, and bounded for each p ∈ (1, ∞). In fact,
p p
T : L1 (∂Ω, w) −→ L1 (∂Ω, w) is well defined, linear, and bounded
for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ).
(5.2.164)

Moreover, for each function

 σ(x)  p
f ∈ L11 ∂Ω, ∩ L1,loc (∂Ω, σ) for some p ∈ (1, ∞) (5.2.165)
1 + |x| n−1
it follows that
p
T f ∈ L1,loc (∂Ω, σ), (5.2.166)
and for each pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.2.149))
    
∂τr s (T f ) = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r (5.2.167)
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f ) at σ-a.e. point on ∂Ω.
p,q
In particular, formula (5.2.167) holds for every function f ∈ L1 (∂Ω, σ) with
p
exponents p, q ∈ (1, ∞), as well as for every function f ∈ L1 (∂Ω, w) with
exponent p ∈ (1, ∞) and weight w ∈ Ap (∂Ω, σ). Formula (5.2.167) also holds
p,λ q,λ
for each function in M1 (∂Ω, σ) or in B1 (∂Ω, σ) with p, q ∈ (1, ∞) and
λ ∈ (0, n − 1) (cf. (A.0.150), (A.0.33)). In fact,
p,λ
the operator T maps each of the Sobolev spaces M1 (∂Ω, σ),
p,λ q,λ
M̊1 (∂Ω, σ), B1 (∂Ω, σ) with p, q ∈ (1, ∞) and λ ∈ (0, n − 1) (5.2.168)
(cf. (A.0.150), (A.0.155), (A.0.33)) boundedly into themselves.

(8) For each p ∈ (1, ∞) it follows that T # , originally acting on functions from
L p (∂Ω, σ), further extends uniquely to a linear, bounded operator, from the
p
negative boundary Sobolev space L−1 (∂Ω, σ) into itself. Furthermore, if one
#
retains the same notation T for said extension, then the transpose of (5.2.163)
is
p p
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.2.169)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 629

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

More generally, for each p, q ∈ (1, ∞) it follows that T # acting on functions
from L p (∂Ω, σ) further extends uniquely to a linear and bounded operator from
p,q
the negative off-diagonal boundary Sobolev space L−1 (∂Ω, σ) into itself and,
adopting the same notation T # for this extension, the transpose of (5.2.162) is
p,q p,q
T # : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) (5.2.170)

where p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q  = 1.
Also, for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ), it follows
that the operator T # , originally acting on L p (∂Ω, w), further extends unique-
ly to a linear, bounded operator, from the negative boundary Sobolev space
p
L−1 (∂Ω, w) into itself which, in fact, is the transpose of T acting on the bound-
p
ary Sobolev space L1 (∂Ω, w ) where p := (1 − 1/p)−1 ∈ (1, ∞) is the conjugate

exponent of p and w  := w 1−p ∈ Ap (∂Ω, σ) is the conjugate weight of w (cf.
[68, item (2) in Lemma 7.7.1]).
(9) Consider the following modified version of the double layer operator in (5.2.135)

acting on each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n according to
∫  

Tmod f (x) := ν(y), (x − y)k(x − y) − (−y)k(−y) · 1Rn \B(0,1) (y) f (y) dσ(y)
∂Ω
(5.2.171)
for all x ∈ Ω. Then the operator Tmod is meaningfully defined, and is compatible
with T from (5.1.4) in the sense that for each function f belonging to the smaller

space L 1 ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular, for each function f ∈ L p (∂Ω, σ)
with p ∈ [1, ∞)) the difference

C f := Tmod f − T f is a constant in Ω. (5.2.172)

As a consequence,
 σ(x) 
∇Tmod f = ∇T f in Ω for each f ∈ L 1 ∂Ω, . (5.2.173)
1 + |x| n−1
Moreover,
Tmod maps constant functions on ∂Ω
(5.2.174)
into constant functions in Ω.
In addition, at each point x ∈ Ω one may express
630 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫
  
∂ Tmod f (x) = ν(y), ∂x [(x − y)k(x − y)] f (y) dσ(y)
∂Ω
∫
= ν (y)k(x − y) + ν(x), x − y (∂ k)(x − y) f (y) dσ(y) (5.2.175)
∂Ω

for each ∈ {1, . . . , n} and each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n .

Also, given any function

 σ(x) 
f ∈ L 1 ∂Ω, with the property that
1 + |x| n
 (5.2.176)
σ(x) 
∂τ j  f ∈ L 1 ∂Ω, for each j, ∈ {1, . . . , n},
1 + |x| n−1
.p
(in particular, for each function f ∈ L1 (∂Ω, σ) with 1 < p < ∞) it follows that
for each index ∈ {1, . . . , n} and each point x ∈ Ω one has
∫

∂ Tmod f (x) = (x j − y j )k(x − y)(∂τ j f )(y) dσ(y). (5.2.177)
∂Ω

As a consequence of (5.2.177) and [70, Theorem 2.5.1],

 κ−n.t.
the boundary trace ∂ Tmod f ∂Ω exists at σ-a.e. point on ∂Ω,
(5.2.178)
for each function f as in (5.2.176) and each index ∈ {1, . . . , n}.

Another corollary of (5.2.177) and [70, (2.4.8)] is the fact that for each trunca-
tion parameter ε ∈ (0, ∞), and each exponent p ∈ (1, ∞),

Nκε ∇(Tmod f ) belongs to the space Lloc (∂Ω, σ)
p

 σ(x) 
for each function f ∈ L 1 ∂Ω, with the property that
1 + |x| n
 σ(x)  p
∂τ j k f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) for all j, k ∈ {1, . . . , n}.
1 + |x| n−1
(5.2.179)
Also, as seen from (5.2.171) and [70, (2.5.32)], for each integrability exponent
p ∈ (1, ∞) and each truncation parameter ε ∈ (0, ∞) one has

Nκε (Tmod f ) ∈ Lloc (∂Ω, σ) for each function

p

 σ(x) 
(5.2.180)
p
f ∈ L 1 ∂Ω, ∩ L (∂Ω, σ).
1 + |x| n loc

 
Finally, for each exponent p ∈ n−1n , 1 and each aperture parameter κ ∈ (0, ∞),
there exists some finite constant C > 0, depending only on ∂Ω, k, n, κ, and p,
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 631
.p
with the property that for each distribution f ∈ H1 (∂Ω, σ) one has

Nκ (∇Tmod f ) L p (∂Ω,σ)
≤ C f H. p (∂Ω,σ) . (5.2.181)
1

(10) Given any exponent α ∈ (0, 1) there exists a constant C ∈ (0, ∞) with the
property that
  
sup dist(x, ∂Ω)1−α ∇ Tmod f (x) ≤ C f 𝒞. α (∂Ω) (5.2.182)
x ∈Ω
.
for each function f ∈ 𝒞α (∂Ω). Moreover,
if Ω ⊆ Rn is a uniform domain with. the property . that
 ∂Ω
is an Ahlfors regular set then Tmod : 𝒞α (∂Ω) → 𝒞α Ω is a (5.2.183)
well-defined, linear, and bounded operator for each α ∈ (0, 1),
whereas
if Ω ⊆ Rn is an NTA .domain with .an upper
 Ahlfors regular
α (∂Ω) → 𝒞α Ω is a well-defined,
boundary then Tmod : 𝒞van van
(5.2.184)
linear, and bounded operator for each exponent α ∈ (0, 1),
with the homogeneous vanishing Hölder spaces defined as in (A.0.48) (with
Σ := ∂Ω and Σ := Ω, respectively). Also, for each α ∈ (0, 1) and each exponent
p ∈ (1,
. ∞) there exists some C ∈ (0, ∞) with the property that for each function
f ∈ 𝒞α (∂Ω) one has
 ∫  1/p
1  p
sup ∇T
mod f  dist(·, ∂Ω) p−1 dL n ≤ C f 𝒞. α (∂Ω)
x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,∞)
(5.2.185)
and
 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
 .α
≤ C dist f , 𝒞van (∂Ω) , (5.2.186)
.
where the distance is measured in the space 𝒞α (∂Ω), · 𝒞. α (∂Ω) . As a corollary,
if. the function f actually belongs to the homogeneous vanishing Hölder space
𝒞vanα (∂Ω) for some α ∈ (0, 1), then for each p ∈ (1, ∞) one has

 ∫  1/p 
1  p
lim sup ∇T f  dist(·, ∂Ω) p−1 dL n = 0.
mod
R→0+ x ∈∂Ω r n−1+αp B(x,r)∩Ω
r ∈(0,R)
(5.2.187)
632 5 Generalized Double Layers in Uniformly Rectifiable Domains

(11) Let Tmod be the modified version of the singular integral operator (5.2.136),

acting on each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n according to29
∫
 
Tmod f (x) := lim+ ν(y), kε (x − y) − k1 (−y) f (y) dσ(y) (5.2.188)
ε→0
∂Ω

at σ-a.e. x ∈ ∂Ω, where

kε (z) := z k(z) · 1Rn \B(0,ε) (z) for each ε > 0 and z ∈ Rn . (5.2.189)

Then for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n one has the jump-formula

κ−n.t. ϑ

Tmod f  = − f + Tmod f at σ-a.e. point on ∂Ω. (5.2.190)
∂Ω 2
.p
In particular, (5.2.190) holds for each f ∈ L1 (∂Ω, σ) with 1 < p < ∞. As a
consequence of (5.2.190) and (5.2.174),

Tmod maps constant functions on ∂Ω

(5.2.191)
into constant functions on ∂Ω.

Furthermore,

if p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,

 
then given any f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω, 1+σ(x)
p
|x | n

p
∫
and g ∈ Lcomp (∂Ω, σ) with ∂Ω g dσ = 0,
(5.2.192)
∫ ∫
it follows that |Tmod f ||g| dσ < +∞, | f ||T # g| dσ < +∞,
∂Ω ∂Ω
∫ ∫
and (Tmod f )g dσ = f (T # g) dσ.
∂Ω ∂Ω

Finally, given any integrability exponent p ∈ (1, ∞), for each function
 
f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) with the property that

 σ(x) 
∂τ j k f ∈ L 1 ∂Ω,
p
∩ Lloc (∂Ω, σ) (5.2.193)
1 + |x| n−1
for all j, k ∈ {1, . . . , n}

29 alternative descriptions of Tmod may be obtained from [70, Proposition 2.3.3]

5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 633
.p
(hence for each f ∈ L1 (∂Ω, σ) if Ω also satisfies a two-sided local John
condition) it follows that
p
Tmod f ∈ L1,loc (∂Ω, σ), (5.2.194)

and for each pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.2.149))
     
∂τr s Tmod f = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f ) (5.2.195)

at σ-a.e. point on ∂Ω.

(12) For each p ∈ (1, ∞) there exists a constant C ∈ (0,
 ∞) with
 p the property that for
each function f ∈ BMO(∂Ω, σ) the measure ∇ Tmod f  dist(·, ∂Ω) p−1 dL n is
Carleson in Ω in the quantitative sense that
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω

≤C f
p
. . (5.2.196)
BMO(∂Ω,σ)

In particular, corresponding to p = 2, it follows that30

  2
∇ T f  dist(·, ∂Ω) dL n is a Carleson measure in Ω,
mod
(5.2.197)
for each function f ∈ BMO(∂Ω, σ).

In addition, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the
property that for each function f ∈ BMO(∂Ω, σ) one has
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω
mod
B(x,r)∩Ω
r ∈(0,R)

≤ C dist f , VMO(∂Ω, σ) (5.2.198)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ).
As a corollary,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n
mod

is a vanishing Carleson measure in Ω, (5.2.199)

for each f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞)

and, corresponding to p = 2,

 2
30 it is natural to refer to  ∇Tmod f  dist(·, ∂Ω) dL n as the Littlewood-Paley measure associated
with f via the operator Tmod
634 5 Generalized Double Layers in Uniformly Rectifiable Domains
  2
∇ T f  dist(·, ∂Ω) dL n
mod

is a vanishing Carleson measure in Ω, (5.2.200)

for each f ∈ VMO(∂Ω, σ).

Finally, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property
that for each function f ∈ BMO(∂Ω, σ) one has

max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→0+ x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω

∫  1/p
  p
× ∇ T f  dist (·, ∂Ω) p−1 dL n ,
mod
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ 
  p  1/p
× ∇ T  dist (·, ∂Ω) p−1 dL n
mod f
B(x,r)∩Ω


≤ C dist f , CMO(∂Ω, σ) , (5.2.201)

where the distance in the right-hand side is considered in the space BMO(∂Ω, σ).
As a consequence of (5.2.201) and Definition 1.8.1,
  
∇ T f  p dist (·, ∂Ω) p−1 dL n
mod

is a super vanishing Carleson measure in Ω, (5.2.202)

for each f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).

(13) Make the additional assumption that ∂Ω is bounded. Then all properties listed
in items (9)-(12) above are valid for the operator T , as originally defined in
(5.2.135), in place of its modified version Tmod . In particular, for each p ∈ (1, ∞)
there exists a constant C ∈ (0, ∞) with the property that for each function
f ∈ BMO(∂Ω, σ) one has
∫
1   
sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

≤C f
p
. . (5.2.203)
BMO(∂Ω,σ)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 635

and
 ∫ 1
1   
lim+ sup  ∇ T f  p dist(·, ∂Ω) p−1 dL n p
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
r ∈(0,R)

≤ C dist f , VMO(∂Ω, σ) (5.2.204)

where the distance in the right-hand side is considered in BMO(∂Ω, σ). As a

consequence,
  
∇ T f  p dist(·, ∂Ω) p−1 dL n
is a vanishing Carleson measure in Ω, (5.2.205)
for each f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).

(14) The operators

Tmod : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) and (5.2.206)

Tmod : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (5.2.207)

are well defined, linear, and bounded. As a consequence of (5.2.206)-(5.2.207),

the operators
   
Tmod : BMO(∂Ω, σ) ∼ −→ BMO(∂Ω, σ) ∼ defined as
    (5.2.208)
Tmod [ f ] := Tmod f for each function f ∈ BMO(∂Ω, σ)

and
   
Tmod : VMO(∂Ω, σ) ∼ −→ VMO(∂Ω, σ) ∼ defined as
    (5.2.209)
Tmod [ f ] := Tmod f for each function f ∈ VMO(∂Ω, σ)

are well defined, linear, and bounded. Also, if ∂Ω is unbounded one has
    
Tmod f , g = [ f ], T # g ,
(5.2.210)
∀ f ∈ BMO(∂Ω, σ), ∀g ∈ H 1 (∂Ω, σ),

where ·, · denotes the duality bracket between the John-Nirenberg space of

functions of bounded mean oscillations on ∂Ω, modulo constants, and the Hardy
space H 1 on ∂Ω (cf. [69, Theorem 4.6.1]), and where T # is considered here as
in (5.2.152) with p := 1. In addition, if ∂Ω is unbounded, the operators

Tmod : CMO(∂Ω, σ) −→ CMO(∂Ω, σ), (5.2.211)

and
636 5 Generalized Double Layers in Uniformly Rectifiable Domains
   
Tmod : CMO(∂Ω, σ) ∼−→ CMO(∂Ω, σ) ∼ defined as
    (5.2.212)
Tmod [ f ] := Tmod f for each function f ∈ CMO(∂Ω, σ),

are also well defined, linear, and bounded.

Corresponding to the case when ∂Ω is bounded (a scenario in which one has
BMO(∂Ω, σ) ⊆ L p (∂Ω, σ) for each p ∈ (1, ∞)), the operator T acting on the
Lebesgue scale L p (∂Ω, σ) with p ∈ (1, ∞) has BMO(∂Ω, σ) as an invariant
subspace, and its restriction

T : BMO(∂Ω, σ) −→ BMO(∂Ω, σ) (5.2.213)

is a well-defined, linear, and bounded operator, which satisfies

   
T f , g = f , T # g , ∀ f ∈ BMO(∂Ω, σ), ∀g ∈ H 1 (∂Ω, σ) (5.2.214)

where ·, · denotes the duality bracket between the John-Nirenberg space of

functions of bounded mean oscillations on ∂Ω and the Hardy space H 1 on ∂Ω
(cf. [69, Theorem 4.6.1]) and where T # is presently considered as in (5.2.152)
with p := 1.
Moreover, when ∂Ω is bounded the operator T acting on BMO(∂Ω, σ) (cf.
(5.2.213)) has VMO(∂Ω, σ) as an invariant subspace, hence its restriction

T : VMO(∂Ω, σ) −→ VMO(∂Ω, σ) (5.2.215)

is a well-defined, linear, and bounded operator.

 1
(15) Fix p ∈ n−1n , 1 and set α := (n − 1) p − 1 ∈ (0, 1). Then the operators
. .
Tmod : 𝒞α (∂Ω) −→ 𝒞α (∂Ω), (5.2.216)
.α .α
Tmod : 𝒞van (∂Ω) −→ 𝒞van (∂Ω), (5.2.217)
.    . 
𝒞α (∂Ω) ∼  [ f ] → Tmod f ∈ 𝒞α (∂Ω) ∼ (5.2.218)

are well defined, linear, and bounded. As a consequence of (5.2.216)-(5.2.218),

the operators
  .  . 
Tmod : 𝒞α (∂Ω) ∼−→ 𝒞α (∂Ω) ∼ defined as
    . (5.2.219)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞α (∂Ω),

and
  .α  .α 
Tmod : 𝒞van (∂Ω) ∼−→ 𝒞van (∂Ω) ∼ defined as
    .α (5.2.220)
Tmod [ f ] := Tmod f for each function f ∈ 𝒞van (∂Ω),

are well defined, linear, and bounded. Also, if ∂Ω is unbounded one has
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 637
     .α
Tmod f , g := [ f ], T # g , ∀ f ∈ 𝒞 (∂Ω), ∀g ∈ H p (∂Ω, σ), (5.2.221)

where ·, · denotes the duality bracket between a homogeneous Hölder space

on ∂Ω, modulo constants, and the corresponding Hardy space on ∂Ω (cf. [69,
Theorem #
 n−1 4.6.1]), and where T is presently considered as in (5.2.152) (with
p ∈ n , 1 ). Finally, in the case when the set ∂Ω is bounded (a scenario in
.α .
α (∂Ω) ⊆ 𝒞α (∂Ω) = 𝒞α (∂Ω) ⊆ L p (∂Ω, σ) for
which one has 𝒞van (∂Ω) = 𝒞van
each p ∈ (1, ∞)), the operator T acting on the Lebesgue scale as in item (3) has
α (∂Ω) and 𝒞α (∂Ω) as invariant subspaces, and its restrictions
both 𝒞van
α α
T : 𝒞van (∂Ω) −→ 𝒞van (∂Ω) (5.2.222)

and
T : 𝒞α (∂Ω) −→ 𝒞α (∂Ω) (5.2.223)
are well-defined, linear and bounded operators, the latter of which satisfies
   
T f , g = f , T # g , ∀ f ∈ 𝒞α (∂Ω), ∀g ∈ H p (∂Ω, σ) (5.2.224)

where ·, · is the duality bracket between Hölder and Hardy spaces on ∂Ω (cf.
[69, Theorem 4.6.1]), and where T # is currently considered as in (5.2.152) (with
p ∈ n−1n , 1 ).
(16) Select p, q ∈ (1, ∞) with 1/p + 1/q = 1 along with λ ∈ (0, n − 1). Then the
operator T from (5.2.147) has the inhomogeneous Morrey-Campanato space
L p,λ (∂Ω, σ) (defined as in (A.0.119) with Σ := ∂Ω) as an invariant subspace,
and
   
T : L p,λ (∂Ω, σ), · L p, λ (∂Ω,σ) −→ L p,λ (∂Ω, σ), · L p, λ (∂Ω,σ) (5.2.225)

is a linear and bounded mapping. Moreover, if Tmod is the modified version of the
singular integral operator T defined in (5.2.188), the assignment
.  . 
[Tmod ] : L p,λ (∂Ω, σ) ∼ −→ L p,λ (∂Ω, σ) ∼
. (5.2.226)
[Tmod ][ f ] := [Tmod f ] for each f ∈ L p,λ (∂Ω, σ)

is well defined, linear and bounded. In particular,

. .
the operator Tmod : L p,λ (∂Ω, σ) −→ L p,λ (∂Ω, σ) is well de-
fined, linear, and there exists some constant C ∈ (0, ∞) with
the property that Tmod f L. p, λ (∂Ω,σ) ≤ C f L. p, λ (∂Ω,σ) for each (5.2.227)
. p,λ
f ∈L (∂Ω, σ).
Finally,
638 5 Generalized Double Layers in Uniformly Rectifiable Domains

the (real)
 transpose
 of the operator T # from (5.2.159) is, respec-
tively, Tmod from (5.2.226) if ∂Ω is unbounded, and T from (5.2.228)
(5.2.225) if ∂Ω is bounded,
.
that is, for each f ∈ L p,λ (∂Ω, σ) and g ∈ ℋq,λ (∂Ω, σ) one has
    
Tmod f , g = [ f ], T # g if ∂Ω is unbounded, (5.2.229)
   
T f , g = f , T # g if ∂Ω is bounded, (5.2.230)

where ·, · denotes the duality bracket between the Morrey-Campanato space

and its pre-dual (cf. [69, (6.1.25)]).
(17) Select q ∈ (1, ∞) and η ∈ (0, 1) and recall the Calderón spaces defined in
(3.1.10)-(3.1.15). Then there exists a constant C ∈ (0, ∞) with the property that,
with the maximal operator defined as in (3.1.1) (with Σ := ∂Ω), one has
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tmod f (5.2.231)
 
for each function f ∈ L 1 ∂Ω, 1+σ(x)|x | n . Consequently, given any p ∈ [1, ∞), it
follows that
.p  .p 
[Tmod ] : Cq,η (∂Ω, σ) ∼ −→ Cq,η (∂Ω, σ) ∼
  .p (5.2.232)
defined by [Tmod ][ f ] := Tmod f for each f ∈ Cq,η (∂Ω, σ)

is a well-defined, linear, and bounded operator. Furthermore, for each exponent

p ∈ [1, ∞) there exists some constant C ∈ (0, ∞) such that
 #
q,η (x) ≤ C fq,η (x) at σ-a.e. x ∈ ∂Ω,
#
Tf (5.2.233)

for each function f ∈ L p (∂Ω, σ). Finally, whenever p ∈ (1, ∞) it follows that
p p
T : Cq,η (∂Ω, σ) −→ Cq,η (∂Ω, σ) (5.2.234)

is a well-defined, linear, and bounded operator.

(18) If
1
∈ {1, . . . , N }, n−1
n < p ≤ ∞ and (n − 1) p −1 + < s < 1, (5.2.235)

 , p, s, θ) ∈ (0, ∞) such
it follows that for each θ ∈ (0, 1) there exists C = C(Ω, k,
that
− 1 −s  
δ∂Ω p ∇ Tmod f  ,θ p ≤ C f Bsp, p (∂Ω,σ) (5.2.236)
nL (Ω, L )
p, p
for all f ∈ Bs (∂Ω,
 σ), with
 the understanding
 that when p > 1 the solid
maximal function ∇ Tmod f  ,θ is replaced by ∇ Tmod f .
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 639

(19) Make the additional assumption that ∂Ω is compact. Then the operator T,
originally acting on Lebesgue spaces on ∂Ω (cf. (5.2.147)), extends uniquely to
linear and bounded mappings
p,q p,q
T : Bs (∂Ω, σ) −→ Bs (∂Ω, σ),
 n−1   (5.2.237)
p∈ n , ∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
p,q p,q
T : Fs (∂Ω, σ) −→ Fs (∂Ω, σ),
 n−1    1 (5.2.238)
p ∈ n , ∞ , q ∈ n−1 n ,∞ , (n − 1) min{p,q } −1 + < s < 1.

Moreover, various choices of the exponents yield operators which are compatible
with one another. In addition, the operator T # , originally considered acting on
Lebesgue spaces on ∂Ω (cf. (5.2.147)) further extends, in a unique fashion, to
linear and bounded mappings
p,q p,q
T # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (5.2.239)
with s ∈ (0, 1), p ∈ n−1n−s , ∞], q ∈ (0, ∞],

and
p,q p,q
T # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ),
  (5.2.240)
with s ∈ (0, 1), p ∈ n−1n−s , ∞), q ∈ n−1
n−s , ∞].

Again, various choices of the parameters p, q, s yield operators which are com-
patible with one another. In all cases,

the operator norms of T in (5.2.237)-(5.2.238) #

 as well as of T in
α
(5.2.239)-(5.2.240) are all dominated by C |α | ≤ N supS n−1 |∂ k |
where the constant C ∈ (0, ∞) depends only on n, p, q, s, and the UR
character of ∂Ω.
(5.2.241)
Finally, if p, q, p, q  ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q  and s ∈ (0, 1),
then
 #   
B−s (∂Ω,σ) T f , g B sp , q  (∂Ω,σ) = B−s (∂Ω,σ) f , T g B sp , q  (∂Ω,σ)
p, q p, q

(5.2.242)
p,q p,q
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ),

and
 #   
p, q
F−s (∂Ω,σ) T f , g F p, q (∂Ω,σ) = F−s
p, q
(∂Ω,σ) f , T g F p , q  (∂Ω,σ)
s s
(5.2.243)
p,q p,q
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ).
640 5 Generalized Double Layers in Uniformly Rectifiable Domains

(20) Impose stronger hypotheses on the underlying domain by now assuming that Ω
is a UR domain satisfying a local John condition. Also, fix some p ∈ (1, ∞).
Then the operator
.p p .p p
Tmod : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) −→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) (5.2.244)

is well defined, linear, and bounded, when the spaces involved are endowed with
.p
the semi-norm (A.0.128). Moreover, identity (5.2.195) holds for each function
p
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ).
Finally, if one now assumes that Ω ⊆ Rn is an NTA domain with an upper
Ahlfors regular boundary31, then for each integrability exponent p ∈ (1, ∞) the
operator
.p .p
Tmod : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (5.2.245)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). In this case, it follows from (5.2.245) and (5.2.191)
that for each p ∈ (1, ∞) the operator
  .p  .p 
Tmod : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
    .p  .p (5.2.246)
Tmod [ f ] := Tmod f ∈ L1 (∂Ω, σ) ∼ for all f ∈ L1 (∂Ω, σ),

is well defined, linear, and bounded, when all quotient spaces are endowed with
the natural semi-norm32 introduced in [69, (11.5.138)].
(21) Suppose Ω is a UR domain satisfying a local John condition33, and fix an
integrability exponent p ∈ (1, ∞) along with a parameter λ ∈ (0, n − 1). Then
the operator
. p,λ . p,λ
Tmod : M1 (∂Ω, σ) −→ M1 (∂Ω, σ) (5.2.247)

is well defined, linear, and bounded, when the spaces involved are endowed
with the semi-norm (A.0.158). In turn, (5.2.247) and (5.2.191) imply that the
operator
  . p,λ  . p,λ 
Tmod : M1 (∂Ω, σ) ∼ −→ M1 (∂Ω, σ) ∼ defined as
    . p,λ  . p,λ (5.2.248)
Tmod [ f ] := Tmod f ∈ M1 (∂Ω, σ) ∼ ∀ f ∈ M1 (∂Ω, σ)

31 recall that this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
32 [69, Proposition 11.5.14] tells us that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
33 for example, this is the case if Ω ⊆ R n is an NTA domain with an upper Ahlfors regular boundary
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 641

is well defined, linear, and bounded, when the quotient spaces are endowed with
the semi-norm34 introduced [69, (11.13.51)].
. p,λ
Furthermore, similar properties are valid for vanishing Morrey-based homoge-
neous Sobolev spaces M1 (∂Ω, σ) (cf. [69, Definition 11.13.15], or (A.0.159)-
(A.0.160)) with p. ∈ (1, ∞), λ ∈ (0, n − 1), and for block-based homogeneous
q,λ
Sobolev spaces B1 (∂Ω, σ) with q ∈ (1, ∞), λ ∈ (0, n − 1), in place of Morrey-
based homogeneous Sobolev spaces.
(22) Strengthen the hypotheses on the underlying domain by assuming that Ω ⊆ Rn
is an NTA domain with an upper Ahlfors regular boundary35. Then the mod-
ified boundary-to-boundary operator Tmod (originally defined as in (5.2.188)-
(5.2.189)) induces a linear and bounded mapping
.p .p  
Tmod : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) for each p ∈ n−1
n ,1 . (5.2.249)

(23) Theorem 5.1.8 and Corollary 5.1.9 are valid for the boundary-to-domain modi-
fied chord-dot-normal integral operator Tmod from (5.2.171). In addition, Theo-
rem 5.1.15, Corollary 5.1.16, and Corollary 5.1.20 are valid for the boundary-
to-boundary modified chord-dot-normal singular integral operator Tmod from
(5.2.188)-(5.2.189).

Proof If we define the vector-valued function

 := z k(z) for each z ∈ Rn \ {0},
k(z) (5.2.250)
 N n n
then k ∈ 𝒞 (R \ {0}) is odd and positive homogeneous of degree 1 − n, since
the current scalar-valued function k is even and positive homogeneous of degree −n
in Rn \ {0}. The latter condition implies that for each z ∈ Rn \ {0} and t ∈ (0, ∞) we
have k(tz) = t −n k(z). Differentiating in t then yields z, (∇k)(tz) = −n t −n−1 k(z)
and, after specializing this by taking t = 1, we arrive at Euler’s identity

z, (∇k)(z) = −n k(z) for each z ∈ Rn \ {0}. (5.2.251)

Consequently,

 = ∂j z j k(z) = n k(z) + z j (∂j k)(z)
div k(z)

= n k(z) + z, (∇k)(z) = 0 for each z ∈ Rn \ {0}, (5.2.252)

which ultimately shows that k from (5.2.250) satisfies (5.1.2). Note that

34 We know from [69, Proposition 11.13.10] that this semi-norm is actually a genuine norm if
Ω ⊆ R n is an open set satisfying a two-sided local John condition and whose boundary is an
unbounded Ahlfors regular set
35 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
642 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫ ∫

ω, k(ω) dH n−1 (ω) = k(ω) dH n−1 (ω), (5.2.253)
S n−1 S n−1

so ϑ from (5.1.3) agrees with ϑ from (5.2.134). Also, the integral operators (5.1.4)-
(5.1.6) for k as in (5.2.250) agree with (5.2.135)-(5.2.137). Granted these, all results
become direct consequences of Theorem 5.1.1. 
Of course, there is a natural version of Theorem 5.2.2 for chord-dot-normal
singular integral operators with matrix-valued kernels, i.e., as in (5.2.135)-(5.2.137)
where now k is matrix-valued (and the respective SIO’s now acting on vector-valued
functions). Double layer potential operators associated with distinguished coefficient
tensors fall under this category (we shall elaborate on this later, in the next volume).
It is also of interest to work out a version of Theorem 5.2.2 corresponding to
variable coefficient kernels, of the sort presented below.

Theorem 5.2.3 For each n ∈ N with n ≥ 2 there exists a positive integer M = M(n)
with the following significance. Let b(x, z) be a function which is even and positive
homogeneous of degree −n in the variable z ∈ Rn \ {0}, and such that ∂zα b(x, z)
is continuous and bounded on Rn × S n−1 for each multi-index α ∈ N0n satisfying
|α| ≤ M. Also, let Ω ⊆ Rn be a nonempty open set with the property that ∂Ω is a UR
set; in particular, Ω is a set of locally finite perimeter. Abbreviate σ := H n−1 ∂Ω
and denote by ν the geometric measure theoretic outward unit normal to Ω. Define
∫
ϑ(x) := b(x, ω) dH n−1 (ω) ∈ C for each x ∈ Rn . (5.2.254)
S n−1

In this setting, introduce variable coefficient chord-dot-normal

 singular  integral
operators of the following sort. For each function f ∈ L 1 ∂Ω, 1+σ(y)|y | n−1
define the
boundary-to-domain integral operators
∫
T f (x) := ν(y), x − y b(x, x − y) f (y) dσ(y) for all x ∈ Ω, (5.2.255)
∂Ω

and
∫
 f (x) :=
T ν(y), x − y b(y, x − y) f (y) dσ(y) for all x ∈ Ω. (5.2.256)
∂Ω
 
Also, for each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1 consider the boundary-to-boundary
singular integral operators
∫
T f (x) := lim+ ν(y), x − y b(x, x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.257)
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 643
∫
T # f (x) := lim+ ν(x), y − x b(y, y − x) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.258)
as well as
∫
 f (x) := lim
T ν(y), x − y b(y, x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω,
+
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.259)
and
∫
# f (x) := lim
T ν(x), y − x b(x, y − x) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω.
+
ε→0
y ∈∂Ω
|x−y |>ε
(5.2.260)
Then, in relation to these chord-dot-normal singular integral operators, the fol-
lowing statements are true.
 
(1) For each function f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
and κ > 0 one has

 κ−n.t.  ϑ(x)

T f (x) = − f (x) + (T f )(x) at σ-a.e. x ∈ ∂Ω, (5.2.261)
∂Ω 2
and
 κ−n.t.  ϑ(x)
 f 
T (x) = −  f )(x) at σ-a.e. x ∈ ∂Ω.
f (x) + (T (5.2.262)
∂Ω 2
 
In addition, if for each function f ∈ L 1 ∂Ω, 1+σ(y) |y | n−1
one defines the vector-
valued functions
∫
W f (x) := f (y)b(y, x − y)(x − y) dσ(y) for all x ∈ Ω, (5.2.263)
∂Ω

and
∫
( f (x) :=
W f (y)b(x, x − y)(x − y) dσ(y) for all x ∈ Ω, (5.2.264)
∂Ω

then for each κ > 0 the nontangential boundary traces

κ−n.t. κ−n.t.
( f )
(W f )∂Ω and (W ∂Ω
exists at σ-a.e. point on ∂Ω, (5.2.265)

and satisfy
 κ−n.t.  ϑ(x)

ν(x) · W f  (x) = − f (x) − (T # f )(x) at σ-a.e. x ∈ ∂Ω, (5.2.266)
∂Ω 2
644 5 Generalized Double Layers in Uniformly Rectifiable Domains

 κ−n.t.  ϑ(x)
ν(x) · W( f  (x) = − # f )(x) at σ-a.e. x ∈ ∂Ω. (5.2.267)
f (x) − (T
∂Ω 2
(2) For each p ∈ [1, ∞) and κ > 0 there exists some finite constant C > 0, depending
only on ∂Ω, b, n, p, and κ, such that for each function f ∈ L p (∂Ω, σ) one has

max Nκ (T f ) , Nκ (W f )
L p (∂Ω,σ) L p (∂Ω,σ)
≤C f L p (∂Ω,σ) (5.2.268)

and

max  f)
Nκ (T (f)
, Nκ (W ≤C f L p (∂Ω,σ) (5.2.269)
L p (∂Ω,σ) L p (∂Ω,σ)

if p > 1, plus similar estimates in the case when p = 1 in which scenario the
corresponding L 1 -norms in the left-hand side are now replaced by the quasi-
norm L 1,∞ (∂Ω, σ).
Moreover, the action of the operator W,( originally considered as in (5.2.264),
may be further extended in a unique and coherent fashion (cf. [70, (2.4.15),
(2.4.16), (2.4.24)])
 n−1 to the scale of Lorentz-based Hardy spaces H p,q (∂Ω, σ)
with p ∈ n , ∞ and q ∈ (0, ∞] and for each κ > 0 said extension satisfies
(for some constant C = C(∂Ω, b, n, p, q, κ) ∈ (0, ∞))
(f )
Nκ (W ≤C f H p, q (∂Ω,σ) (5.2.270)
L p, q (∂Ω,σ)

for all f ∈ H p,q (∂Ω, σ).

Finally, a similar extension result and estimate holds for the operator W in a
slightly more restrictive setting than originally assumed, namely when
  1
∂Ω is compact, p ∈ n−1 n , 1 , r > (n − 1) p − 1 , and whenever
α ∈ N0n has |α| ≤ M the function ∂zα b(·, z) belongs to 𝒞r (Rn ), (5.2.271)
uniformly for z ∈ S n−1 .

(3) For each p ∈ (1, ∞), the operators

T
T, T #, T, # : L p (∂Ω, σ) −→ L p (∂Ω, σ) (5.2.272)

are all well defined, linear, and bounded. Also, given any p, p ∈ (1, ∞) satisfying
1/p + 1/p = 1 it follows that

the transpose of T acting on L p (∂Ω, σ)

 (5.2.273)
is the operator T # acting on L p (∂Ω, σ),

while
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 645

 acting on L p (∂Ω, σ)
the transpose of T
(5.2.274)
# acting on L p (∂Ω, σ).
is the operator T

Corresponding to the end-point p = 1, the operators T, T # , T, T # map L 1 (∂Ω, σ)

linearly and boundedly into L 1,∞ (∂Ω, σ). Next, if for each vector-valued function
  n
g ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one considers
∫
V g(x) := lim+ b(x, x − y) x − y, g(y) dσ(y) (5.2.275)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, and

∫
g(x) := lim
V b(y, x − y) x − y, g(y) dσ(y) (5.2.276)
ε→0+
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω, then the operators

 
 : L p (∂Ω, σ) n −→ L p (∂Ω, σ), p ∈ (1, ∞),
V, V
  (5.2.277)
 : L 1 (∂Ω, σ) n −→ L 1,∞ (∂Ω, σ),
V, V

are well defined, linear, bounded, and for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
one has

 f ν) = T
V( f ν) = T f and V( f . (5.2.278)

Finally, similar results are valid for Muckenhoupt weighted Lebesgue spaces,
Lorentz spaces, and Morrey spaces (as well as their duals and their preduals)
on ∂Ω.
 
(4) Fix p ∈ n−1 #
n , 1 . Then the operators T and T , originally acting on Lebesgue
#

spaces as in (the first part of) item (3), extend to linear and bounded mappings
# : H p (∂Ω, σ) −→ L p (∂Ω, σ)
T #, T (5.2.279)
 n−1 
and the operators corresponding to various choices of the index p ∈ n ,1
are compatible with one another.
(5) Work under the stronger assumption that whenever α, β ∈ N0n satisfy |α| ≤ M
β
and | β| ≤ 1 the function ∂x ∂zα b(x, z) is continuous and bounded on Rn × S n−1 .
p
Then for each function f belonging to the boundary Sobolev space L1 (∂Ω, σ)
with 1 < p < ∞, each index j ∈ {1, . . . , n}, and each aperture parameter κ > 0,
the pointwise nontangential boundary traces
 κ−n.t.  κ−n.t.
f 
∂j T f ∂Ω and ∂j T ∂Ω
exist at σ-a.e. point on ∂Ω. (5.2.280)
646 5 Generalized Double Layers in Uniformly Rectifiable Domains

Furthermore, for each p ∈ (1, ∞) and κ > 0 there exists some finite constant
C > 0, depending only on ∂Ω, b, n, p, and κ, such that for each function
p
f ∈ L1 (∂Ω, σ) one has

Nκ (T f ) L p (∂Ω,σ)
+ Nκ (∇T f ) L p (∂Ω,σ)
≤C f p
L1 (∂Ω,σ) (5.2.281)

and
 f)
Nκ (T  f)
+ Nκ (∇T ≤C f L1 (∂Ω,σ) .
p (5.2.282)
L p (∂Ω,σ) L p (∂Ω,σ)

(6) Continue to assume that whenever α, β ∈ N0n satisfy |α| ≤ M and | β| ≤ 1

β
the function ∂x ∂zα b(x, z) is continuous and bounded on Rn × S n−1 . Then the
operators
 : L (∂Ω, σ) −→ L (∂Ω, σ)
T, T
p p
(5.2.283)
1 1

are well defined, linear, and bounded for each p ∈ (1, ∞). In fact,

 : L (∂Ω, w) → L (∂Ω, w) are well defined, linear, bounded

T, T
p p
1 1
(5.2.284)
for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ).
p
Moreover, for each function f ∈ L1 (∂Ω, σ) with 1 ≤ p < ∞ and for any given
pair of indices r, s ∈ {1, . . . , n} one has (with V as in (5.2.149))
    
∂τr s (T f ) = T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f )

+ Br s f at σ-a.e. point on ∂Ω, (5.2.285)

where, at σ-a.e. point x ∈ ∂Ω,

∫
Br s f (x) := lim+ ν(y), x − y (∂τr s (x) b)(x, x − y) f (y) dσ(y), (5.2.286)
ε→0
y ∈∂Ω
|x−y |>ε

(here (∂τr s (x) b)(x, x − y) indicates that the tangential derivative ∂τr s is applied
to the function b(x, z) in the variable x and subsequently z taken to be x − y)
induces linear and bounded mappings

Br s : L p (∂Ω, σ) −→ L p (∂Ω, σ) if p ∈ (1, ∞),

(5.2.287)
Br s : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ) if p = 1.

 Finally,
A similar result is valid for T.
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 647

the operators T and T  map the Sobolev spaces M p,λ (∂Ω, σ),
1
p,λ q,λ
M̊1 (∂Ω, σ), and B1 (∂Ω, σ) with exponents p, q ∈ (1, ∞) (5.2.288)
and λ ∈ (0, n−1) (cf. (A.0.150), (A.0.155), (A.0.33)) boundedly
into themselves.
(7) Once again make the assumption that whenever α, β ∈ N0n satisfy |α| ≤ M and
β
| β| ≤ 1 the function ∂x ∂zα b(x, z) is continuous and bounded on Rn × S n−1 . Then
for each p ∈ (1, ∞) it follows that T # and T # , originally acting on functions
from L p (∂Ω, σ), further extend uniquely to linear, bounded operators, from the
p
negative boundary Sobolev space L−1 (∂Ω, σ) into itself. Furthermore, if one
retains the same notation for said extensions, then the transpose operators of
(5.2.283) are
 
# : L p (∂Ω, σ) −→ L p (∂Ω, σ)
T #, T (5.2.289)
−1 −1

where p ∈ (1, ∞) is such that 1/p + 1/p = 1.

Also, for each exponent p ∈ (1, ∞) and each weight w ∈ Ap (∂Ω, σ), it follows
# , originally acting on L p (∂Ω, w), further extend uniquely to linear,
that T # and T
p
bounded operators, from the negative boundary Sobolev space L−1 (∂Ω, w) into
itself which, in fact, are the transpose of T and T  acting on the Sobolev space
p
L1 (∂Ω, w ) where p := (1 − 1/p)−1 ∈ (1, ∞) is the conjugate exponent of p

and w  := w 1−p ∈ Ap (∂Ω, σ) is the conjugate weight of w (cf. [68, item (2) in
Lemma 7.7.1]).
(8) Adopt the stronger assumption that whenever α, β ∈ N0n satisfy |α| ≤ M and
β
| β| ≤ 1 the function ∂x ∂zα b(x, z) is continuous and bounded on Rn × S n−1 .
Then the operators T and T  originally acting on Lebesgue spaces on ∂Ω (cf.
(5.2.272)), extend uniquely to linear and bounded mappings

 : Bs (∂Ω, σ) −→ Bs (∂Ω, σ),

T, T
p,q p,q

 n−1   (5.2.290)
p ∈ n , ∞ , q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
 : Fs (∂Ω, σ) −→ Fs (∂Ω, σ),
T, T
p,q p,q

 n−1    1 (5.2.291)
p ∈ n , ∞ , q ∈ n−1 n ,∞ , (n − 1) min{p,q } −1 + < s < 1.

Moreover, various choices of the exponents yield operators which are compatible
with one another. In addition, the operators T # and T# , originally considered
acting on Lebesgue spaces on ∂Ω (cf. (5.2.272)) further extend, in a unique
fashion, to linear and bounded mappings
p,q p,q
T # : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (5.2.292)
with s ∈ (0, 1), p ∈ n−1n−s , ∞], q ∈ (0, ∞],
648 5 Generalized Double Layers in Uniformly Rectifiable Domains

and
p,q p,q
T # : F−s (∂Ω, σ) −→ F−s (∂Ω, σ),
  (5.2.293)
with s ∈ (0, 1), p ∈ n−1n−s , ∞), q ∈ n−1
n−s , ∞].

Again, various choices of the parameters p, q, s yield operators which are com-
patible with one another. Finally, if the exponents p, q, p, q  ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q  and s ∈ (0, 1), then
 #   
B−s (∂Ω,σ) T f , g B sp , q  (∂Ω,σ) = B−s (∂Ω,σ) f , T g B sp , q  (∂Ω,σ)
p, q p, q

(5.2.294)
p,q p,q
for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ),

and
 #   
p, q
F−s (∂Ω,σ) T f , g F p, q (∂Ω,σ) = F−s
p, q
(∂Ω,σ) f , T g F p , q  (∂Ω,σ)
s s
(5.2.295)
p,q p,q
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ),

 and T
plus two other similar duality formulas now involving the operators T # .

Proof The idea is to reduce matters to the case of “constant coefficient” kernel
already treated in Theorem 5.2.2 via a spherical harmonics expansion, much as in
the proof of [70, Theorem 2.5.38]. We shall freely use notation and results from the
proof on [70, Theorem 2.5.38]. To get started, suppose
! "
Ψi ∈N0, 1≤i ≤H is an orthonormal basis for L 2 (S n−1, H n−1 ),
(5.2.296)
consisting of spherical harmonics as in [70, (2.5.483)].

In particular,

Ψi L 2 (S n−1, H n−1 ) = 1 for each ∈ N0 and 1 ≤ i ≤ H , (5.2.297)

and
Ψi : S n−1 → R is an odd function whenever is odd. (5.2.298)
Also, recall from [70, (2.5.481), (2.5.482)] that

H0 = 1 and H ≤ Cn n−1
for ≥ 1. (5.2.299)

If for each ∈ N0 and 1 ≤ i ≤ H we define

∫
ai (x) := b(x, ω)Ψi (ω) dH n−1 (ω), for each x ∈ Rn, (5.2.300)
S n−1

it follows from (5.2.298) and the current assumptions on b(x, z) that

the function ai is continuous in Rn and
(5.2.301)
is identically zero whenever is odd.
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 649

As in [70, (2.5.497)], we may then expand (bearing in mind (5.2.301), and the fact
that now b(x, z) is positive homogeneous of degree −n in the variable z)

 
H

b(x, z) = ai (x)ki (z), ∀ (x, z) ∈ Rn × Rn \ {0} , (5.2.302)
∈2N0 i=1

where for each ∈ 2N0 and 1 ≤ i ≤ H we have set

 z 
ki (z) := Ψi |z| −n for all z ∈ Rn \ {0}, (5.2.303)
|z|
and where the coefficients ai are rapidly decreasing, in a uniform manner, in the
sense that for each number m ∈ N with 2m ≤ M there exists a constant Cn,m ∈ (0, ∞)
such that
 
sup ai (x) ≤ Cn,m · Cb · max{1, }−2m for ∈ 2N0 and 1 ≤ i ≤ H , (5.2.304)
x ∈R n

where  α 
Cb := Cn · sup  ∂ b (x, z) ∈ (0, ∞). (5.2.305)
z
(x,z)∈R n ×S n−1
|α | ≤M

Also, much as in [70, (2.5.498)], if the number d is as in [70, (2.5.487)] then there
exists some constant Cn, N ∈ (0, ∞) with the property that

ki S n−1 𝒞 N (S n−1 ) ≤ Cn, N · max{1, } d if ∈ N0 and 1 ≤ i ≤ H . (5.2.306)

Consequently, if for each ∈ 2N0 and 1 ≤ i ≤ H we introduce

∫ ∫
ϑi := ki (ω) dH (ω) =
n−1
Ψi (ω) dH n−1 (ω) ∈ C, (5.2.307)
S n−1 S n−1

where the second equality comes from (5.2.303), then the normalization condition
recorded in (5.2.297) implies that there exists a constant Cn ∈ (0, ∞) such that

|ϑi | ≤ Cn for all ∈ N0 and 1 ≤ i ≤ H . (5.2.308)

Based on (5.2.302) and (5.2.254), for each x ∈ Rn we may compute

    ∫
ai (x)ϑi = ai (x)ki (ω) dH n−1 (ω)
∈2N0 1≤i ≤H ∈2N0 1≤i ≤H S n−1

∫
= b(x, ω) dH n−1 (ω) = ϑ(x), (5.2.309)
S n−1

where all series are absolutely convergent and may be interchanged with integration
thanks to (5.2.299), (5.2.304), and (5.2.308).
650 5 Generalized Double Layers in Uniformly Rectifiable Domains

Moving on, for each ∈ 2N0 and 1 ≤ i ≤ H , define  the boundary-to-domain


integral operator T acting on each function f ∈ L ∂Ω, 1+σ(x)
i 1
|x | n−1
according to
∫
T i
f (x) := ν(y), x − y ki (x − y) f (y) dσ(y) for all x ∈ Ω, (5.2.310)
∂Ω

and consider the boundary-to-boundary

 singular
 integral operator T i on ∂Ω acting
σ(x)
on any given function f ∈ L 1 ∂Ω, 1+ |x | n−1 according to
∫
T i f (x) := lim+ ν(y), x − y ki (x − y) f (y) dσ(y) (5.2.311)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω.

Then, with justifications for the convergence of the series involved as in the proof
of [70, Theorem 2.5.38], we have
 
T= ai T i (5.2.312)
∈2N0 1≤i ≤H
 
with convergence in B L p (∂Ω, σ) → L p (∂Ω, σ) if 1 < p < ∞ and, corresponding
 
to p = 1, in B L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ) . Also, for each f ∈ L p (∂Ω, σ) with
1 ≤ p < ∞ and each κ > 0, at σ-a.e. x ∈ ∂Ω we may compute

 κ−n.t.   
H  κ−n.t. 
 
T f (x) = ai (x) T i f  (x)
∂Ω ∂Ω
∈2N0 i=1

1    
H H
=− ai (x)ϑi (x) f (x) + ai (x)(T i f )(x)
2 ∈2N i=1 ∈2N i=1
0 0

ϑ(x)
=− f (x) + (T f )(x), (5.2.313)
2
where the first equality is justified much as in [70, (2.5.520)-(2.5.525)], and the
subsequent steps follow from (5.2.138), (5.2.312), and (5.2.309).
We shall now use the above result to prove the jump-formula stated in (5.2.261).

To get started, select a function f ∈ L 1 ∂Ω , 1+σ(x)
|x | n−1
, fix a point x0 ∈ ∂Ω and pick
an arbitrary number r ∈ (0, ∞). Decompose

f = f1 + f2 on ∂Ω, where
(5.2.314)
f1 := 1∂Ω∩B(x0,2r) · f and f2 := 1∂Ω\B(x0,2r) · f ,
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 651

then split
T f = T f1 + T f2 in Ω. (5.2.315)
Choose an aperture parameter κ > 0. In view of the fact that T f2 has a continuous
κ−n.t.

extension to B(x0, r), it trivially follows that the nontangential trace T f2  exists
∂Ω
at every point in ∂nta Ω ∩ B(x0, r). Specifically,
κ−n.t. ∫
 
T f2  (x) = ν(y), x − y b(x, x − y) f2 (y) dσ(y)
∂Ω ∂Ω

= T f2 (x) for each x ∈ ∂nta Ω ∩ B(x0, r). (5.2.316)

From (5.2.316) and item (iii) in [68, Proposition 8.8.6] we then see that

 κ−n.t. 

T f2  (x) = T f2 (x) at σ-a.e. x ∈ ∂∗ Ω ∩ B(x0, r). (5.2.317)
∂Ω

Next, since f1 ∈ L 1 (∂Ω, σ), we may rely on (5.2.313) to conclude that, at σ-a.e.
point x ∈ ∂∗ Ω ∩ B(x0, r), we have

 κ−n.t. ϑ(x)

T f1  (x) = − f1 (x) + (T f1 )(x). (5.2.318)
∂Ω 2
Finally, from (5.2.317), (5.2.318), and (5.2.314) we conclude (bearing in mind the
arbitrariness of r > 0) that (5.2.261) holds.
Next, the goal is to prove (5.2.281). As a preamble, for each ∈ 2N0 and
1 ≤ i ≤ H we recall from (5.1.201) (written for k(z)  := ki (z)z) that for any
j ∈ {1, . . . , n} and at each x ∈ Ω, we have

∂j (T i f )(x) = T i (∇tan f ) j (x)
∫
− ν j (y) x − y, (∇tan f )(y) ki (x − y) dσ(y) (5.2.319)
∂Ω
p
for any function f ∈ L1 (∂Ω, σ)
with p ∈ (1, ∞). Based on this, (5.2.310), and [70,
(2.4.9)] we then conclude that for each p ∈ (1, ∞) and κ > 0 there exists a constant
p
C∂Ω, p,n,κ ∈ (0, ∞) with the property that for any f ∈ L1 (∂Ω, σ) we have

Nκ (T i f ) L p (∂Ω,σ)
+ Nκ (∇T i f ) L p (∂Ω,σ)

≤ C∂Ω, p,n,κ · ki S n−1 𝒞 N (S n−1 )
f L1 (∂Ω,σ) .
p (5.2.320)

From (5.2.310), (5.2.311), (5.2.138), (5.2.161) (wit q := p), and [69, Proposi-
p
tion 11.3.4] we conclude that for any function f ∈ L1 (∂Ω, σ) with p ∈ (1, ∞) and
any κ > 0 we have
652 5 Generalized Double Layers in Uniformly Rectifiable Domains

 κ−n.t. ϑ
 i p
Ti f = T i f  + f ∈ L1 (∂Ω, σ) (5.2.321)
∂Ω 2
and there exists some C = C(∂Ω, p, n, κ) ∈ (0, ∞) such that

Ti f p
L1 (∂Ω,σ) ≤ C Nκ (T i f ) L p (∂Ω,σ)
+ C Nκ (∇T i f ) L p (∂Ω,σ)

+ C|ϑi | f L1 (∂Ω,σ) .
p (5.2.322)

Combining (5.2.322) with (5.2.320), (5.2.308), and (5.2.306) we arrive at the con-
clusion that for each p ∈ (1, ∞) there exists a constant C∂Ω, p,n ∈ (0, ∞) with the
property that

Ti f p
L1 (∂Ω,σ) ≤ C∂Ω, p,n · max{1, } d · f p
L1 (∂Ω,σ)
p
(5.2.323)
for each f ∈ L1 (∂Ω, σ), if ∈ 2N0 and 1 ≤ i ≤ H .

Let us now work under the stronger assumptions adopted in item (5). Much as in
[70, (2.5.542)], these imply that for each m ∈ N with 2m ≤ M there exists a constant
Cn,m ∈ (0, ∞) such that
    β 
sup (∇ai )(x) ≤ Cn,m · sup sup  ∂x ∂zα b (x, z) · max{1, }−2m
x ∈R n |α | ≤M x ∈R n
|β | ≤1 z ∈S n−1
whenever ∈ N0 and 1 ≤ i ≤ H .
(5.2.324)
From [69, Corollary 11.1.19] we see that

given any scalar function a ∈ 𝒞1 (Rn ) with the property that

supx ∈Rn |a(x)| < ∞ as well as supx ∈Rn |(∇a)(x)| < ∞, for each
p
f ∈ L1 (∂Ω, σ) with p ∈ [1, ∞] it follows that a · f belongs to (5.2.325)
p
L1 (∂Ω, σ) and for each   j, k ∈ {1,
 pair of indices  . . . , n} one has

∂τ j k (a · f ) = ν j (∂k a)∂Ω − νk (∂j a)∂Ω · f + a∂Ω · ∂τ j k f .

Then a combination of (5.2.312), (5.2.324), (5.2.304), and (5.2.325) proves that, for
each integrability exponent p ∈ (1, ∞), the series
   
p p
ai T i converges to T in B L1 (∂Ω, σ) → L1 (∂Ω, σ) . (5.2.326)
∈2N0 1≤i ≤H

p
As a consequence, T defines a linear and bounded operator from L1 (∂Ω, σ) into
itself, for each p ∈ (1, ∞). A similar argument is valid for T,  and this establishes
(5.2.283). The same circle of ideas may be employed to justify all claims made in
item (5).
Let us now turn our attention to (5.2.285). We shall work under the hypotheses
adopted in item (6) so, in particular, (5.2.324) continues to hold. To set the stage, for
each ∈ N0 and 1 ≤ i ≤ H consider the  n singular integral operator acting on each
vector-valued function g ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ according to
5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 653
∫
V i g(x) := lim+ ki (x − y) x − y, g(y) dσ(y) (5.2.327)
ε→0
y ∈∂Ω
|x−y |>ε

for σ-a.e. x ∈ ∂Ω. Thanks to (5.2.302) and (5.2.275), much as with (5.2.312) we
have  
V= ai V i (5.2.328)
∈2N0 1≤i ≤H
 
with convergence in B L p (∂Ω, σ) → L p (∂Ω, σ) if 1 < p < ∞ and, corresponding
 
to p = 1, in B L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ) .
To proceed, observe that (5.2.324) permits us to differentiate the series in (5.2.302)
term by term and obtain that, for each j ∈ {1, . . . , n},

 
H

(∂x j b)(x, z) = (∂x j ai )(x)ki (z), ∀ (x, z) ∈ Rn × Rn \ {0} . (5.2.329)
∈2N0 i=1

Next, given any r, s ∈ {1, . . . , n}, for each f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ we may
use (5.2.286) to write
∫ 
Br s f (x) = lim+ ν(y), x − y νr (x)(∂xs b)(x, x − y)
ε→0
y ∈∂Ω
|x−y |>ε

− νs (x)(∂xr b)(x, x − y) f (y) dσ(y)
 ∫ 
= νr (x) lim+ ν(y), x − y (∂xs b)(x, x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
 ∫ 
− νs (x) lim+ ν(y), x − y (∂xr b)(x, x − y) f (y) dσ(y)
ε→0
y ∈∂Ω
|x−y |>ε
 
= νr (x)(∂xs ai )(x)(T i f )(x)
∈2N0 1≤i ≤H
 
− νs (x)(∂xr ai )(x)(T i f )(x)
∈2N0 1≤i ≤H
   
= νr (x)(∂xs ai )(x) − νs (x)(∂xr ai )(x) · (T i f )(x)
∈2N0 1≤i ≤H
(5.2.330)
654 5 Generalized Double Layers in Uniformly Rectifiable Domains
p
at σ-a.e. point x ∈ ∂Ω. Finally, for each function f ∈ L1 (∂Ω, σ) with 1 < p < ∞
we may now compute
     
∂τr s (T f ) = νr (∂s ai )∂Ω − νs (∂r ai )∂Ω · T i f
∈2N0 1≤i ≤H
 
+ ai ∂τr s (T i f )
∈2N0 1≤i ≤H
  
= Br s f + ai T i ∂τr s f
∈2N0 1≤i ≤H
   
+ ai Mνr , T i (∇tan f )s
∈2N0 1≤i ≤H
   
− ai Mνs , T i (∇tan f )r
∈2N0 1≤i ≤H
   
− ai Mνr , V i (νs ∇tan f )
∈2N0 1≤i ≤H
   
+ ai Mνs , V i (νr ∇tan f )
∈2N0 1≤i ≤H
    
= Br s f + T ∂τr s f + Mνr , T (∇tan f )s − Mνs , T (∇tan f )r
   
− Mνr , V (νs ∇tan f ) + Mνs , V (νr ∇tan f ). (5.2.331)

Above, the first equality uses (5.2.326), item (iv) in [69, Proposition 11.1.9] (used
with q := p), (5.2.163) (used with T := T i ), and (5.2.325). The second equality in
(5.2.331) is based on (5.2.330) and (5.2.167) (used with T := T i and V := V i ). The
last equality in (5.2.331) is a consequence of (5.2.312) and (5.2.328). The proof of
(5.2.285) is therefore complete.
All other claims in the statement of Theorem 5.2.3 are proved in a similar fashion,
using the spherical harmonic decomposition technique employed in the proof of [70,
Theorem 2.5.38] together with the constant coefficient kernel case already dealt with
in Theorem 5.2.2. 

To offer examples of chord-dot-normal singular integral operators of a different

nature, work in the two-dimensional setting, and employ the notation

x ∗ := (x2, −x1 ) for each x = (x1, x2 ) ∈ R2 . (5.2.332)

Note that, with ‘dot’ denoting the ordinary scalar product in R2 , we have

x · x ∗ = 0 and x · y = x ∗ · y ∗, ∀x, y ∈ R2, (5.2.333)

which further entails the following useful orthogonal decomposition:

5.2 Generalized Double Layers with Matrix-Valued Kernels . . . 655

y · x∗ ∗ y · x
y= x + x, ∀x ∈ R2 \ {0}, ∀y ∈ R2 . (5.2.334)
|x| 2 |x| 2

Lemma 5.2.4 Let Ω be an Ahlfors regular domain in R2 . Denote by ν geometric

measure theoretic outward unit normal to Ω, and let τ = iν be the geometric
measure theoretic unit tangent along ∂Ω (cf. [68, (5.6.29)-(5.6.31)]). Also, assume
that F : R2 \ {0} → R is a function of class 𝒞 N , for some N ∈ N, which is positive
homogeneous of degree zero.
Then the function

k : R2 \ {0} −→ R given by
(∇F)(x) · x ∗ (5.2.335)
k(x) := − for each x ∈ R2 \ {0}
|x| 2

is of class 𝒞 N −1 , positive homogeneous of degree −2, and one has36

∂τ(y) [F(x − y)] = x − y, ν(y) k(x − y)

(5.2.336)
for H 1 -a.e. point y ∈ ∂Ω and each x ∈ R2 \ {y},

as well as
∂τ(x) [F(x − y)] = − x − y, ν(x) k(x − y)
(5.2.337)
for H 1 -a.e. point x ∈ ∂Ω and each y ∈ R2 \ {x}.

Finally, if F is also even, then so is k.

Proof Since F is positive homogeneous of degree zero we have

F(t x) = F(x) ∀t > 0, ∀x ∈ R2 \ {0}. (5.2.338)

Differentiating with respect to t in the above equality implies (∇F)(t x) · x = 0 for all
t ∈ (0, ∞) and x ∈ R2 \ {0}. In particular, corresponding to t = 1, we have

(∇F)(x) · x = 0, ∀x ∈ R2 \ {0}. (5.2.339)

Recall from [68, (5.6.30)] that if (ν1, ν2 ) are the scalar components of the vector ν
then τ = (−ν2, ν1 ) at H 1 -a.e. point on ∂Ω. In the notation introduced in (5.2.332),
this translates into
τ = ν ∗ at H 1 -a.e. point on ∂Ω. (5.2.340)
For H 1 -a.e. point y ∈ ∂Ω and each x ∈ R2 \ {y} we may then write

36 where ∂τ denotes the directional derivative operator along the unit vector τ
656 5 Generalized Double Layers in Uniformly Rectifiable Domains

∂τ(y) [F(x − y)] = τ(y) · ∇y [F(x − y)] = −τ(y) · (∇F)(x − y)

 (∇F)(x − y) · (x − y)∗
= −ν ∗ (y) · (x − y)∗
|x − y| 2
(∇F)(x − y) · (x − y) 
+ (x − y)
|x − y| 2
(∇F)(x − y) · (x − y)∗
= − [ν ∗ (y) · (x − y)∗ ]
|x − y| 2
(∇F)(x − y) · (x − y)∗
= − x − y, ν(y)
|x − y| 2

= x − y, ν(y) k(x − y). (5.2.341)

Above, the first equality is simply the definition of ∂τ , the directional derivative
operator along the unit vector τ. The second equality in (5.2.341) is just Chain Rule,
and the third equality comes from (5.2.334). The fourth equality in (5.2.341) is
then implied by (5.2.339), while the fifth equality follows from the second identity
in (5.2.333) (used with ν(y) in place of x and with x − y in place of y). The last
equality in (5.2.341) is seen from (5.2.335). This establishes (5.2.336), and (5.2.337)
is justified in a very similar fashion.
Next, differentiating (5.2.338) with respect to x yields t(∇F)(t x) = (∇F)(x) for
each t ∈ (0, ∞) and each x ∈ R2 \ {0}, ergo ∇F is a positive homogeneous function
of degree −1 in R2 \ {0}. Thus, the fact that the assignment

(∇F)(x) · x ∗
R2 \ {0}  x → is positive homogeneous of degree −2 (5.2.342)
|x| 2

follows easily from the observation above and the fact that R2 \ {0}  x → x ∗ is pos-
itive homogeneous of degree 1, while R2 \ {0}  x → |x| 2 is positive homogeneous
of degree 2. In view of (5.2.335), this analysis proves that k is positive homogeneous
of degree −2 in R2 \ {0}. It is also clear from definitions that k is a function of class
𝒞 N −1 .
Finally, it is clear from (5.2.335) that k is an even function whenever F is so. 

From Lemma 5.2.4 we see that:

singular integral operators whose kernels are the directional derivative,
along the unit tangent vector to a two-dimensional UR domain, of
(5.2.343)
smooth even functions which are positive homogeneous of degree zero
in R2 \ {0} are singular integral operators of “chord-dot-normal” type.
To all such singular integral operators the results from Theorem 5.2.2 then apply.
Examples of functions F which are smooth, even, and positive homogeneous of
degree zero in R2 \ {0} include
5.3 Another Look at Standard and Modified Riesz Transforms 657

Pm (x)
F(x) := for each x ∈ R2 \ {0}, (5.2.344)
|x| m

where m ∈ N is even, and Pm is a homogeneous polynomial of degree m in R2 . In

particular, we may take
x1 x2
F(x) := for each x = (x1, x2 ) ∈ R2 \ {0}, (5.2.345)
|x| 2
or, using complex notation,

zk
F(z) := for each z ∈ C \ {0} and k ∈ N. (5.2.346)
zk

5.3 Another Look at Standard and Modified Riesz Transforms

We first consider standard and modified Riesz transforms in the entire Euclidean
space. Throughout, fix n ∈ N satisfying n ≥ 2. For each j ∈ {1, . . . , n − 1} define the
(ordinary) j-th Riesz transform in Rn−1 as the singular integral operator R j acting
 
on any given function f ∈ L 1 Rn−1, 1+ |xdx | n−1 according to
∫
2 xj − yj
R j f (x ) := lim+ f (y ) dy  (5.3.1)
ε→0 ωn−1 |x  − y  | n
y  ∈R n−1
|x  −y  |>ε

for L n−1 -a.e. x  ∈ Rn−1 . These operators fall withing the scope of [70, Theo-
rem 2.3.2], [70, Theorem 2.6.1], and Theorem 2.1.4 stated for the closed UR set

Σ := Rn−1 × {0} ⊆ Rn, (5.3.2)

canonically identified with the Euclidean space Rn−1 , and the kernels

k j ∈ 𝒞∞ (Rn \ {0}), k j (x) := x j /|x| n for each x ∈ Rn \ {0}, (5.3.3)

for j ∈ {1, . . . , n − 1}, which are smooth, odd, and positive homogeneous of degree
1 − n in Rn \ {0}. The aforementioned Riesz transforms are also special instances
of the singular integral operators discussed in Proposition 1.2.1, corresponding to
the particular case when Ω := R+n , a scenario in which we have the identifications
∂Ω = Rn−1 × {0} ≡ Rn−1 , σ ≡ L n−1 , and ν = −en . If we also take the function b
from the statement of Proposition 1.2.1 to be

b ∈ 𝒞∞ (Rn \ {0}), b(x) := |x| n−2 for each x ∈ Rn \ {0} (5.3.4)

then, up to normalization, the “tangential” singular integral operators Tn j corre-

sponding to 1 ≤ j ≤ n − 1, defined as in (1.2.2) but for the present setting, agree with
658 5 Generalized Double Layers in Uniformly Rectifiable Domains

the Riesz transforms R j with 1 ≤ j ≤ n − 1, defined as in (5.3.1). In addition, the

# with 1 ≤ j ≤ n − 1, defined as in (1.2.3) for the choice
singular integral operators Tnj
of Ω and b as above, also agree, up to normalization, with the Riesz transforms R j
with 1 ≤ j ≤ n − 1, defined in (5.3.1).
Significantly, for each j ∈ {1, . . . , n − 1} the j-th Riesz transform R j also falls
under the umbrella of Theorem 5.1.1 specialized to the case when Ω := R+n , with
∂Ω = Rn−1 × {0} canonically identified with Rn−1 . Specifically, with b as in (5.3.4),
when the vector-valued function (5.1.1) is presently taken to be
 n
k := (∂j b)en − (∂n b)e j ∈ 𝒞∞ (Rn \ {0}) (5.3.5)

(which is odd, positive homogeneous of degree 1−n, and divergence-free in Rn \{0}),

then the generalized double layers T and T # , defined as in (5.1.5)-(5.1.6) for these
choices, agree, up to normalization, with R j . See also Example 5.1.4 in this regard.
From [70, Theorem 2.3.2], Proposition 1.2.1, Theorem 2.1.4, and [70, Theo-
rem 2.6.1] we then see that each Riesz transform acts naturally on a multitude of
function spaces in Rn−1 . For example, for each index j ∈ {1, . . . , n−1}, the following
are well-defined, linear, and bounded operators:
5.3 Another Look at Standard and Modified Riesz Transforms 659

R j : L p (Rn−1, L n−1 ) → L p (Rn−1, L n−1 ), 1 < p < ∞, (5.3.6)

R j : L p (Rn−1, w) → L p (Rn−1, w), 1 < p < ∞, w ∈ Ap (Rn−1, L n−1 ), (5.3.7)

R j : L 1 (Rn−1, L n−1 ) → L 1,∞ (Rn−1, L n−1 ), (5.3.8)

R j : L p,q (Rn−1, L n−1 ) → L p,q (Rn−1, L n−1 ), 1 < p < ∞, 0 < q ≤ ∞, (5.3.9)

R j : H p (Rn−1, L n−1 ) → H p (Rn−1, L n−1 ), n−1

n < p < ∞, (5.3.10)

R j : H p,q (Rn−1, L n−1 ) → H p,q (Rn−1, L n−1 ), n−1

n < p < ∞, 0 < q ≤ ∞, (5.3.11)
p p
R j : L1 (Rn−1, L n−1 ) → L1 (Rn−1, L n−1 ), 1 < p < ∞, (5.3.12)
p p
R j : L−1 (Rn−1, L n−1 ) → L−1 (Rn−1, L n−1 ), 1 < p < ∞, (5.3.13)
p p
R j : L1 (Rn−1, w) → L1 (Rn−1, w), 1 < p < ∞, w ∈ Ap (Rn−1, L n−1 ), (5.3.14)
p p
R j : L−1 (Rn−1, w) → L−1 (Rn−1, w), 1 < p < ∞, w ∈ Ap (Rn−1, L n−1 ), (5.3.15)

R j : M p,λ (Rn−1, L n−1 ) → M p,λ (Rn−1, L n−1 ), 1 < p < ∞, 0 < λ < n − 1,
(5.3.16)
p,λ p,λ
R j : M1 (Rn−1, L n−1 ) → M1 (Rn−1, L n−1 ), 1 < p < ∞, 0 < λ < n − 1,
(5.3.17)

R j : M̊ p,λ (Rn−1, L n−1 ) → M̊ p,λ (Rn−1, L n−1 ), 1 < p < ∞, 0 < λ < n − 1,
(5.3.18)
p,λ p,λ
R j : M̊1 (Rn−1, L n−1 ) → M̊1 (Rn−1, L n−1 ), 1 < p < ∞, 0 < λ < n − 1,
(5.3.19)

R j : B q,λ (Rn−1, L n−1 ) → B q,λ (Rn−1, L n−1 ), 1 < q < ∞, 0 < λ < n − 1,
(5.3.20)
q,λ q,λ
R j : B1 (Rn−1, L n−1 ) → B1 (Rn−1, L n−1 ), 1 < q < ∞, 0 < λ < n − 1,
(5.3.21)

R j : L p,λ (Rn−1, L n−1 ) → L p,λ (Rn−1, L n−1 ), 1 < p < ∞, 0 < λ < n − 1, (5.3.22)
p p
R j : Cq,η (Rn−1, L n−1 ) → Cq,η (Rn−1, L n−1 ), 1 < p, q < ∞, 0 < η < 1. (5.3.23)

Higher order versions of the smoothness spaces above are also allowed.
Slightly digressing, here is an invertibility result involving linear combinations of
Riesz transforms in the entire Euclidean setting which is going to be relevant later
on.
660 5 Generalized Double Layers in Uniformly Rectifiable Domains

Proposition 5.3.1 Fix n ∈ N with n ≥ 2. Also, pick a family of n numbers

λ0, λ1 . . . , λn−1 ∈ C, not all zero, and define


n−1
T := λ0 I + λ j Rj, (5.3.24)
j=1

where I is the identity operator and R j is the j-th Riesz transform in Rn−1 (cf. (5.3.1)).
Then
if n ≥ 3, then T is a linear, bounded, injective operator
(5.3.25)
with dense range from L 2 (Rn−1, L n−1 ) into itself.
In addition, the following properties are equivalent:
(a) The operator T is invertible on L p (Rn−1, L n−1 ) for each p ∈ (1, ∞).
(b) The operator T is invertible on L 2 (Rn−1, L n−1 ).
(c) One has


n−1
λ j ξ j  (−i)λ0 |ξ  | for each ξ  = (ξ1, . . . , ξn−1 ) ∈ Rn−1 \ {0}. (5.3.26)
j=1

Proof For starters, from (5.3.24) and (5.3.6) we see that T is a well-defined linear
and bounded operator on each L p (Rn−1, L n−1 ) with p ∈ (1, ∞). Bring in the Fourier
transform F  in Rn−1 . Since, as is well known (see, e.g., [66, (4.9.15), p. 183]), for
each f ∈ L 2 (Rn−1, L n−1 ) and each j ∈ {1, . . . , n − 1} we have
 ξj 
F  R j f (ξ ) = (−i)  F  f (ξ ) for all ξ  ∈ Rn−1 \ {0}, (5.3.27)
|ξ |
it follows that

F  T f = mF  f for each f ∈ L 2 (Rn−1, L n−1 ), (5.3.28)

where

n−1
λj ξj
m(ξ ) := λ0 + (−i) for all ξ  = (ξ1, . . . , ξn−1 ) ∈ Rn−1 \ {0}. (5.3.29)
j=1
|ξ  |

In particular,

m belongs to 𝒞∞ Rn−1 \ {0} ∩ L ∞ (Rn−1, L n−1 )
(5.3.30)
and is positive homogeneous of degree zero.

We also claim that

if n ≥ 3 then m(ξ )  0 for L n−1 -a.e. ξ  ∈ Rn−1 . (5.3.31)

5.3 Another Look at Standard and Modified Riesz Transforms 661

When λ0 = 0, this simply follows by observing that the zeros ξ  ∈ Rn−1 \ {0} of m
are contained in the intersection of the following two hyperplanes in Rn−1 ,


n−1 
n−1
(Re λ j )ξ j = 0 and (Im λ j )ξ j = 0, (5.3.32)
j=1 j=1

at least one of which is non-degenerate. There remains to justify (5.3.31) in the case
when λ0  0. Note that if ξ  ∈ Rn−1 \ {0} is such that m(ξ ) = 0 then formula
(5.3.29) implies that we necessarily have (−i)λ0 |ξ  | = n−1
j=1 λ j ξ j . Hence, the zeros
of m are among the solutions of the equation


n−1 2
−λ02 |ξ  | 2 = λj ξj (5.3.33)
j=1

or, equivalently, the polynomial equation

 
n−1
λ02 ξ12 + · · · + ξn−1
2
+ λ j λk ξ j ξk = 0. (5.3.34)
j,k=1
If n ≥ 3, the polynomial in the left-side of (5.3.34) is not identically zero, since the
latter eventuality amounts to having

λ02 + λ2 = 0 and λ j λk = 0 for all , j, k ∈ {1, . . . , n − 1} with j  k, (5.3.35)

itself an impossibility for n ≥ 3 (given that we are presently assuming λ0  0). As

such, the claim made in (5.3.31) follows as soon as we show that, for each n ∈ N
with n ≥ 2, 
L n−1 φ−1 ({0}) = 0 for any polynomial function
(5.3.36)
φ : Rn−1 → C not identically zero.
We shall prove (5.3.36) by induction on n. If n = 2, this is readily implied by the
Fundamental Theorem of Algebra, which ensures that φ−1 ({0}) has finite cardinality
in this case. Assume (5.3.36) holds for a given n ∈ N with n ≥ 2, and consider a
polynomial function φ : Rn−1 → C which is not identically zero. Write φ as


N
φ(x , xn ) = φ j (x )xn for each (x , xn ) ∈ Rn = Rn−1 × R,
j
(5.3.37)
j=0

for some integer N ∈ N0 and some polynomial functions φ j : Rn−1 → C with index

j ∈ {0, 1, . . . , N }, at least one of which is not identically zero. Denote by j∗ the index
for which the latter property holds. Also, for each x  ∈ Rn−1 define the polynomial
of one variable
662 5 Generalized Double Layers in Uniformly Rectifiable Domains


N
ψx (t) := φ j (x )t j for each t ∈ R. (5.3.38)
j=0

Next, decompose the closed set φ−1 ({0}) as

φ−1 ({0}) = Z1 ∪ Z2 (5.3.39)

where

Z1 := (x , xn ) ∈ Rn : φ j (x ) = 0 for each j ∈ {0, 1, . . . , N } , (5.3.40)

and

Z2 := (x , xn ) ∈ Rn : there exists some jo ∈ {0, 1, . . . , N }

with φ jo (x )  0 and xn ∈ ψx−1 ({0}) . (5.3.41)


The inductive hypothesis implies L n−1 φ−1 −1
j∗ ({0}) = 0 and since Z1 ⊆ φ j∗ ({0}) × R
we conclude (based on Fubini’s Theorem) that

L n (Z1 ) = 0. (5.3.42)

Also, for each (x , xn ) ∈ Z2 the polynomial ψx is not identically zero, so the
Fundamental Theorem of Algebra guarantees that ψx−1 ({0}) has finite cardinality. In
particular, if we set
!
Z2 := x  ∈ Rn−1 : there exists xn ∈ R such that (x , xn ) ∈ Z2 , (5.3.43)

then L 1 ψx−1 ({0}) = 0 for each x  ∈ Z2 . Based on this and Fubini’s Theorem we
may now compute
∫ ∫ ∫ 
L (Z2 ) =
n
1 dL =
n
1 dL 1 dL n−1 (z ) = 0. (5.3.44)
Z2 Z2 ψz−1 ({0})


From (5.3.39), (5.3.42), and (5.3.44) we ultimately conclude that L n φ({0}) = 0.
This finishes the proof of (5.3.36), so (5.3.31) is now fully justified.
In turn, from (5.3.28), (5.3.31), and the fact that F  is an isomorphism of
L (Rn−1, L n−1 ) we deduce that
2

T : L 2 (Rn−1, L n−1 ) → L 2 (Rn−1, L n−1 )

(5.3.45)
is an injective operator if n ≥ 3.

With λ0, λ1, . . . , λn−1 replaced by λ0, −λ1, . . . , −λn−1 , this also proves that
5.3 Another Look at Standard and Modified Riesz Transforms 663

T ∗ = λ0 I − n−1
j=1 λ j R j : L 2 (Rn−1, L n−1 ) → L 2 (Rn−1, L n−1 )
(5.3.46)
is injective if n ≥ 3.

Having established (5.3.46) we then conclude from [69, (2.1.45)] that the operator

T : L 2 (Rn−1, L n−1 ) → L 2 (Rn−1, L n−1 ) has dense range if n ≥ 3. (5.3.47)

Together, (5.3.45) and (5.3.47) finish the proof of (5.3.25).

Let us now turn to the equivalences claimed in the last part of the statement. First,
it is clear that (a) ⇒ (b). To prove the implication (b) ⇒ (c), assume T is invertible
on L 2 (Rn−1, L n−1 ). Since F  is an isomorphism of L 2 (Rn−1, L n−1 ), from (5.3.28)
we conclude that the mapping

L 2 (Rn−1, L n−1 )  f −→ m f ∈ L 2 (Rn−1, L n−1 ) (5.3.48)

is an isomorphism. This is equivalent to having

m(ξ )  0 for L n−1 -a.e. ξ  ∈ Rn−1 and m−1 ∈ L ∞ (Rn−1, L n−1 ). (5.3.49)

Keeping (5.3.29) in mind, (5.3.26) is readily implied by (5.3.49).

There remains to justify (c) ⇒ (a). To this end, fix an arbitrary p ∈ (1, ∞) and
assume (5.3.26) holds. The latter property ensures that

m−1 belongs to 𝒞∞ Rn−1 \ {0} ∩ L ∞ (Rn−1, L n−1 )
(5.3.50)
and is positive homogeneous of degree zero.

Consequently, for each multi-index α ∈ N0n−1 the function ∂ α m−1 is continuous in

Rn−1 \ {0} and positive homogeneous of degree −|α| hence, in particular, there exists
Cα ∈ (0, ∞) such that
 α −1  
∂ m (ξ ) ≤ Cα |ξ  | − |α | for each ξ  ∈ Rn−1 \ {0}. (5.3.51)

Granted this, Mikhlin’s Multiplier Theorem (cf., e.g., [27, Theorem 6.3, p. 210])
applies and gives that there exists a linear operator

Q bounded both on L p (Rn−1, L n−1 ) and on L p (Rn−1, L n−1 ),

(5.3.52)
with F (Q f ) = m−1 F  f if f ∈ L p (Rn−1, L n−1 ) ∩ L 2 (Rn−1, L n−1 ).

From this, (5.3.28), the fact that F  is an isomorphism of L 2 (Rn−1, L n−1 ), and that
T is bounded both on L 2 (Rn−1, L n−1 ) and on L p (Rn−1, L n−1 ), we conclude that
Q(T f ) = f and T(Q f ) = f for each f ∈ L p (Rn−1, L n−1 ) ∩ L 2 (Rn−1, L n−1 ). By
density, it follows that QT = I = TQ on the entire space L p (Rn−1, L n−1 ), hence T is
invertible on L p (Rn−1, L n−1 ). 
mod
Moving on, for each j ∈ {1, . . . , n − 1} we shall denote by R j the j-th modified
Riesz transform in Rn−1 , i.e., the singular integral operator acting on any given
664 5 Generalized Double Layers in Uniformly Rectifiable Domains
 
n−1 -a.e. point x  ∈ Rn−1 according to
function f ∈ L 1 Rn−1, 1+dx
|x  | n at L


R j f (x )
mod
(5.3.53)
∫
2 xj − yj
:= lim+ 1 n−1  (y )
ε→0 ωn−1 |x  − y  | n R \Bn−1 (x ,ε)
R n−1

−y j
− 1 n−1  (y ) f (y ) dy 

| − y  | n R \Bn−1 (0 ,1)

where, generally speaking, Bn−1 (z , r) is the (n − 1)-dimensional ball centered at

z  ∈ Rn−1 and of radius r.
For each j ∈ {1, . . . , n − 1}, this agrees, up to normalization, with the operator
mod
Tn j defined as in (1.8.168)-(1.8.170) in the case when Ω := R+n and for b as in
(5.3.4).
mod
The format of the j-th modified Riesz transform R j also agrees with the general
recipe from (A.0.202), presently implemented for Σ as in (5.3.2), which is a closed
UR subset of Rn , and the integral kernel k j as in (5.3.3). It also agrees, up to
normalization, with the modified generalized double layer defined as in (5.1.65) in
the case when Ω := R+n and for the vector-valued kernel k as in (5.3.5). In particular,
(5.1.69) implies that
mod
Rj maps constant functions on Rn−1
(5.3.54)
into constant functions on Rn−1 , for each j ∈ {1, . . . , n − 1}.

This makes it possible to consider the action of the modified Riesz transforms on
quotient spaces (of equivalence classes modulo constants), a scenario in which we
define  mod   mod 
R j [ f ] := R j f . (5.3.55)
With this convention in mind, from items (14), (15), (16), (17), (20), (21), (22) in
Theorem 5.1.1 we see that for each j ∈ {1, . . . , n − 1} the following are well-defined,
linear, and bounded operators:
5.3 Another Look at Standard and Modified Riesz Transforms 665
mod .p .p
Rj : L1 (Rn−1, L n−1 ) → L1 (Rn−1, L n−1 ) with 1 < p < ∞, (5.3.56)
 mod  .p  .p 
Rj : L1 (Rn−1, L n−1 ) ∼ → L1 (Rn−1, L n−1 ) ∼ with 1 < p < ∞, (5.3.57)
mod . p,λ . p,λ
Rj : M1 (Rn−1, L n−1 ) → M1 (Rn−1, L n−1 )
with 1 < p < ∞, 0 < λ < n − 1, (5.3.58)
 mod  . p,λ  . p,λ 
Rj : M1 (Rn−1, L n−1 ) ∼ → M1 (Rn−1, L n−1 ) ∼
with 1 < p < ∞, 0 < λ < n − 1, (5.3.59)
mod . p,λ . p,λ
Rj : M1 (Rn−1, L n−1 ) → M1 (Rn−1, L n−1 )
with 1 < p < ∞, 0 < λ < n − 1, (5.3.60)
 mod  . p,λ  . p,λ 
Rj : M1 (Rn−1, L n−1 ) ∼ → M1 (Rn−1, L n−1 ) ∼
with 1 < p < ∞, 0 < λ < n − 1, (5.3.61)
mod . q,λ . q,λ
Rj : B1 (Rn−1, L n−1 ) → B1 (Rn−1, L n−1 )
with 1 < p < ∞, 0 < λ < n − 1, (5.3.62)
 mod  . q,λ  . q,λ 
Rj : B1 (Rn−1, L n−1 ) ∼ → B1 (Rn−1, L n−1 ) ∼
with 1 < p < ∞, 0 < λ < n − 1, (5.3.63)
mod .p .p
Rj : H1 (Rn−1, L n−1 ) → H1 (Rn−1, L n−1 ) with n−1
n < p < ∞, (5.3.64)
 mod  .p  .p 
Rj : H1 (Rn−1, L n−1 ) ∼ → H1 (Rn−1, L n−1 ) ∼
with n−1
n < p < ∞, (5.3.65)
. .
: 𝒞α (Rn−1 ) −→ 𝒞α (Rn−1 ) with 0 < α < 1,
mod
Rj (5.3.66)
  .  . 
: 𝒞α (Rn−1 ) ∼ → 𝒞α (Rn−1 ) ∼ with 0 < α < 1,
mod
Rj (5.3.67)
mod α
. α
.
Rj : 𝒞van (Rn−1 ) −→ 𝒞van (Rn−1 ) with 0 < α < 1, (5.3.68)
 mod  . α n−1  . α n−1 
Rj : 𝒞van (R ) ∼ → 𝒞van (R ) ∼ with , 0 < α < 1, (5.3.69)

as well as
666 5 Generalized Double Layers in Uniformly Rectifiable Domains
mod
Rj : BMO(Rn−1, L n−1 ) −→ BMO(Rn−1, L n−1 ), (5.3.70)
 
mod
Rj & n−1, L n−1 ) → BMO(R
: BMO(R & n−1, L n−1 ), (5.3.71)
mod
Rj : VMO(Rn−1, L n−1 ) −→ VMO(Rn−1, L n−1 ), (5.3.72)
 
mod
Rj & n−1, L n−1 ) → VMO(R
: VMO(R & n−1, L n−1 ), (5.3.73)
mod
Rj : CMO(Rn−1, L n−1 ) −→ CMO(Rn−1, L n−1 ), (5.3.74)
 
mod
Rj & n−1, L n−1 ) → CMO(R
: CMO(R & n−1, L n−1 ), (5.3.75)
mod . .
Rj : L p,λ (Rn−1, L n−1 ) → L p,λ (Rn−1, L n−1 )
with 1 < p < ∞, 0 < λ < n − 1, (5.3.76)
 mod  .  . 
Rj : L p,λ (Rn−1, L n−1 ) ∼ → L p,λ (Rn−1, L n−1 ) ∼
with 1 < p < ∞, 0 < λ < n − 1, (5.3.77)
mod .p .p
Rj : Cq,η (Rn−1, L n−1 ) → Cq,η (Rn−1, L n−1 ), 1 ≤ p < ∞
with 1 < q < ∞, 0 < η < 1, (5.3.78)
 mod  .p  .p 
Rj : Cq,η (Rn−1, L n−1 ) ∼ → Cq,η (Rn−1, L n−1 ) ∼
with 1 < p, q < ∞, 0 < η < 1. (5.3.79)

In addition, there are natural mapping properties of the Riesz transforms acting on
Besov and Triebel-Lizorkin spaces in the entire Euclidean ambient, like the ones in
item (19) of Theorem 5.1.1.
Recall that for each j ∈ {1, . . . , n − 1} we have identified (up to a common
normalization constant) the “ordinary” j-th Riesz transform R j with the operator
mod mod
Tn#j , and the j-th modified Riesz transform R j with the operator Tn j . In view of
these identifications and (2.1.161)-(2.1.162) we then conclude that for each index
j ∈ {1, . . . , n − 1} we have (with the duality brackets as in [69, Theorem 4.6.1])
mod 
   
R j f , g = − [ f ], R j g for any two functions
 
 (5.3.80)
f ∈ BMO(Rn−1, L n−1 ) ⊂ L 1 Rn−1, 1+dx 
|x | n and g ∈ H 1 (Rn−1, L n−1 ),

as well as
5.3 Another Look at Standard and Modified Riesz Transforms 667
 mod    
Rjf , g = − [ f ], R j g for each
.  

f ∈ 𝒞α (Rn−1 ) ⊂ L 1 Rn−1, 1+dx 
|x | n and g ∈ H p (Rn−1, L n−1 ) (5.3.81)
 n−1 
with p ∈ n , 1 and α := (n − 1) p1 − 1 ∈ (0, 1).

From (5.1.146) and (5.1.71)-(5.1.73) we also see that if p ∈ (1, ∞) then for each
function
  p
f ∈ L 1 Rn−1, 1+dx
|x  | n ∩ L loc (R
n−1, L n−1 ) such that

  p
∂k f ∈ L 1 Rn−1, 1+ |xdx | n−1 ∩ Lloc (Rn−1, L n−1 ) (5.3.82)

for all k ∈ {1, . . . , n − 1}

it follows that for each j ∈ {1, . . . , n − 1} we have

mod p
Rj f ∈ L1,loc (Rn−1, L n−1 ) (5.3.83)

and  mod 
∂k R j f = R j ∂k f for each k ∈ {1, . . . , n − 1}. (5.3.84)
In particular, for each j, k ∈ {1, . . . , n − 1} and each p ∈ (1, ∞) it follows that
 mod  .p
∂k R j f = R j ∂k f for each f ∈ L1 (Rn−1, L n−1 ),
p
(5.3.85)
and ∂k (R j f ) = R j (∂k f ) for each f ∈ L1 (Rn−1, L n−1 ).

Shifting perspectives, bring in the modified boundary-to-boundary Cauchy-

Clifford integral operator Cmod , originally introduced in (1.8.112) and re-branded
as such in Remark 1.8.27. When specialized to the case when Ω := R+n , with
∂Ω = Rn−1 × {0} canonically identified with Rn−1 and with ν = −en , this be-
comes the
 principal-value singular integral operator acting on any given function

f ∈ L 1 Rn−1, 1+dx 
|x | n ⊗ C n at L n−1 -a.e. point x  ∈ Rn−1 according to

Cmod f (x ) (5.3.86)
∫
−1 x − y
= lim+ 1 n  (y )
ε→0 ωn−1 |x  − y  | n R \Bn−1 (x ,ε)
R n−1

−y 
− 1 n  (y )  en  f (y ) dy  .

| − y  | n R \Bn−1 (0 ,1)

Hence, in terms of the modified Riesz transforms introduced earlier, we may express
this as

1
n−1
mod
 dx  
Cmod = − e j  en  R j on L 1 Rn−1, ⊗ C n. (5.3.87)
2 j=1 1 + |x  | n
668 5 Generalized Double Layers in Uniformly Rectifiable Domains

Then the fact that the formula noted in (1.8.281) presently becomes (with I denoting
the identity operator)
 2 .p 
Cmod = 14 I on L1 (Rn−1, L n−1 ) ⊗ C n ∼ with p ∈ (1, ∞), (5.3.88)

implies that, when acting

. pon equivalence classes (modulo constants) of scalar-valued
functions in the space L1 (Rn−1, L n−1 ) with p ∈ (1, ∞), we have

 2  
n−1
 mod    
n−1
 mod  
I = 4 Cmod = e j  en R j ek  en Rk
j=1 k=1


n−1
 mod   mod  
n−1
 mod   mod 
= e j  e n  ek  e n R j Rk = e j  ek R j Rk
j,k=1 j,k=1


n−1
 mod 2   mod   mod  
=− Rj + e j  ek Rj , Rk , (5.3.89)
j=1 1≤ j<k ≤n−1

where the last set of brackets stand for the commutator [A, . p B] := AB − BA. In
turn, (5.3.89) amounts to saying that, on the quotient space L1 (Rn−1, L n−1 )/∼ with
p ∈ (1, ∞), we have

n−1
 mod  2
Rj = −I, (5.3.90)
j=1

and
 mod  mod   mod   mod 
Rj Rk = Rk Rj for each j, k ∈ {1, . . . , n − 1}. (5.3.91)

In view of (5.3.55) and (5.3.54), we.may equivalently recast these operator identities
p
as the statement that, on the space L1 (Rn−1, L n−1 ) with p ∈ (1, ∞),


n−1
 mod 2
Rj = −I modulo constants, (5.3.92)
j=1

and
mod mod mod mod
R j Rk = Rk R j modulo constants, for each j, k ∈ {1, . . . , n − 1}. (5.3.93)

Moreover, we observe that

5.3 Another Look at Standard and Modified Riesz Transforms 669

the very same identities (5.3.92)-(5.3.93) are also valid on a variety of

other function spaces of interest, in place of the homogeneous Sobolev
space employed. above, namely on the homogeneous Morrey-based
p,λ
Sobolev space M1 (Rn−1, L n−1 ) with 1 < p < ∞ and 0 < λ < n −
. q,λ
1, on the homogeneous block-based Sobolev space B1 (Rn−1, L n−1 ) (5.3.94)
with 1 < q < ∞ and 0. < λ < n − 1, on the homogeneous Hardy-
p
based Sobolev space H1 (Rn−1, L n−1 ) with n−1 n < p < ∞, on the
. α n−1
homogeneous Hölder space 𝒞 (R ) with 0 < α < 1, and on the
John-Nirenberg space BMO(Rn−1, L n−1 ).
Indeed, the same proof which, starting with (5.3.88), has produced (5.3.92)-(5.3.93)
continues to work in any of those settings, thanks to Remark 3.3.13, Remark 2.3.12,
(2.1.189), and (2.1.186).
In particular, from (5.3.92) written on the space BMO(Rn−1, L n−1 ) and (5.3.72)
we conclude that
for any function f ∈ BMO(Rn−1, L n−1 ) one has
mod
R j f ∈ VMO(Rn−1, L n−1 ) (5.3.95)
f ∈ VMO(Rn−1, L n−1 ) ⇔
for all j ∈ {1, . . . , n − 1}.

This should be compared with the similar result, in the unit, sphere established in
Proposition 5.3.2 (cf. (5.3.107)).
. p,λAlso, from (5.3.92) written on the homogeneous
Morrey-based Sobolev space M1 (Rn−1, L n−1 ) and (5.3.60) we conclude that
. p,λ
for any f ∈ M1 (Rn−1, L n−1 ) with 1 < p < ∞ and 0 < λ < n − 1 one has
. p,λ mod . p,λ
f ∈ M1 (Rn−1, L n−1 ) ⇔ R j f ∈ M1 (Rn−1, L n−1 ) for all j ∈ {1, . . . , n − 1}.
(5.3.96)
From (5.3.92) and (5.3.56)-(5.3.71) we may also obtain “mixed” regularity results
of the following sort:

for any given f ∈ BMO(Rn−1, L n−1 ) and α ∈ (0, 1) one has

. . (5.3.97)
f ∈ 𝒞α (Rn−1 ) ⇔ R j f ∈ 𝒞α (Rn−1 ) for all j ∈ {1, . . . , n − 1}.
mod

All the considerations so far in this section apply to the Hilbert transform on
the real line, i.e., the singular integral operator H acting on any given function
f ∈ L 1 R, 1+dx|x | according to
∫
1 f (y)
H f (x) := lim+ dy for L 1 -a.e. x ∈ R, (5.3.98)
ε→0 π x−y
y ∈R
|x−y |>ε

and its modified version,

 Hmod , which 1is the singular integral operator acting on any
given function f ∈ L 1 R, 1+dx
|x | 2
at L -a.e. point x ∈ R according to
670 5 Generalized Double Layers in Uniformly Rectifiable Domains
∫ 
 1 1 1
Hmod f (x) := lim+ 1R\[x−ε,x+ε] (y) + 1R\[−1,1] (y) f (y) dy.
ε→0 π x−y y
R
(5.3.99)

In particular,
 2
Hmod = −I modulo constants, (5.3.100)
on any of the spaces mentioned in relation to (5.3.92) above (cf. (5.3.94)).
In particular, from (5.3.100) written on the John-Nirenberg space BMO(R, L 1 )
and the fact that Hmod maps VMO(R, L 1 ) into itself we conclude that

for any given function f ∈ BMO(R, L 1 ) one has

(5.3.101)
f ∈ VMO(R, L 1 ) ⇐⇒ Hmod f ∈ VMO(R, L 1 ).

Our last result in this section elaborates on the mapping properties of the Riesz
transforms on smooth surfaces, and contains a remarkable characterization of the
Sarason space VMO in this setting (cf. (5.3.107)).
Proposition 5.3.2 Fix n ∈ N with n ≥ 2, and consider the Riesz transforms (R j )1≤ j ≤n
associated as in (A.0.187) with Σ := S n−1 . Then for each j ∈ {1, . . . , n}, the operator
R j induces mappings

R j : BMO(S n−1, H n−1 ) −→ BMO(S n−1, H n−1 ), (5.3.102)

R j : VMO(S n−1, H n−1 ) −→ VMO(S n−1, H n−1 ), (5.3.103)

p p
R j : L1 (S n−1, H n−1 ) −→ L1 (S n−1, H n−1 ) with p ∈ (1, ∞), (5.3.104)

R j : 𝒞α (S n−1 ) −→ 𝒞α (S n−1 ) with α ∈ (0, 1), (5.3.105)

α α
R j : 𝒞van (S n−1 ) −→ 𝒞van (S n−1 ) with α ∈ (0, 1), (5.3.106)

that are well defined, linear, and bounded. Moreover,

for any given function f ∈ BMO(S n−1, H n−1 ) one has

R j f ∈ VMO(S n−1, H n−1 ) (5.3.107)
f ∈ VMO(S n−1, H n−1 ) ⇔
for all j ∈ {1, . . . , n},

for each given function f ∈ 𝒞α (S n−1 ) with α ∈ (0, 1) one has

(5.3.108)
α (S n−1 ) ⇐⇒ R f ∈ 𝒞α (S n−1 ) for all j ∈ {1, . . . , n},
f ∈ 𝒞van j van

for each given function f ∈ L p (S n−1, H n−1 ) with p ∈ (1, ∞) one has
p p
(5.3.109)
f ∈ L1 (S n−1, H n−1 ) ⇔ R j f ∈ L1 (S n−1, H n−1 ) for all j ∈ {1, . . . , n}.
5.3 Another Look at Standard and Modified Riesz Transforms 671

Finally, all results are valid with the unit sphere S n−1 = ∂Ω replaced by the
boundary ∂Ω of any bounded domain Ω ⊆ Rn of class 𝒞1+ε , ε ∈ (0, 1), with the
understanding that one now takes α ∈ (0, ε) in (5.3.105)-(5.3.106) and (5.3.108).

Proof Work in the Clifford algebra context. Let Mν be the operator of pointwise
left-multiplication in C n by ν, where ν(x) ≡ x1 e1 + · · · + xn en for each x ∈ S n−1 .
Then (1.6.22) tells us that

1
n

CMν = − e j R j on L 1 S n−1, H n−1 ⊗ C n . (5.3.110)
2 j=1

From [69, Proposition 4.4.8] we know that H 1 (S n−1, H n−1 ) is a module over the
ring of smooth functions on S n−1 . Via duality (cf. [69, Theorem 4.6.1] and [69,
Lemma 4.6.9]) we then see that

BMO(S n−1, H n−1 ) is a module over the ring

(5.3.111)
of smooth functions on S n−1 (in a quantitative fashion).

Obviously,

𝒞α (S n−1 ) and 𝒞van

α (S n−1 ) with α ∈ (0, 1) are modules over the ring
(5.3.112)
of smooth functions on S n−1 (in a quantitative fashion).

From (5.3.111)-(5.3.112) and [69, (3.1.50)] we then conclude that

VMO(S n−1, H n−1 ) is a module over the ring

(5.3.113)
of smooth functions on S n−1 (in a quantitative fashion).

Finally, item (v) in [69, Proposition 11.1.9] implies that

p
L1 (S n−1, H n−1 ) with p ∈ (1, ∞) is a module over the ring
(5.3.114)
of smooth functions on S n−1 (in a quantitative fashion).

Collectively, (2.1.191), (5.3.110), and (5.3.111) then prove that each Riesz trans-
form induces a well-defined, linear, and bounded mapping in the context of (5.3.102).
The claim pertaining to (5.3.103)-(5.3.106) are dealt with similarly, making use of
(5.3.110), (5.3.112)-(5.3.114), (2.1.192), (2.1.193), (2.1.194), and (1.6.10).
Consider next the claim made in (5.3.107). The left-to-right implication is a con-
sequence of (5.3.103). To prove the opposite implication, pick an arbitrary function
f in the space BMO(S n−1, H n−1 ) with the property that R j f ∈ VMO(S n−1, H n−1 )
for all j ∈ {1, . . . , n}. Then f ∈ L 2 (S n−1, H n−1 ) (cf. [68, (7.4.106)]) and, as such,
we may write
n
f =− R j (R j f ) ∈ VMO(S n−1, H n−1 ) (5.3.115)
j=1
672 5 Generalized Double Layers in Uniformly Rectifiable Domains

where the equality comes from (1.6.27) and the membership is a consequence
of assumptions and (5.3.103). Properties (5.3.108) and (5.3.109) are dealt with
similarly.
Finally, the very last claim in the statement
  nis proved in a completely similar
fashion, bearing in mind that now ν ∈ 𝒞ε (∂Ω) . 
Chapter 6
Green Formulas and Layer Potential Operators
for the Stokes System

To say that a pair, consisting of vector-valued function u, playing the role of velocity,
together with a scalar-valued function π, playing the role of pressure, is a null-
solution of the Stokes system of linear hydrostatics in an open set Ω ⊆ Rn , where
n ∈ N with n ≥ 2, amounts to having
 
u ∈ 𝒞∞ (Ω) , π ∈ 𝒞∞ (Ω), with
(6.0.1)
Δu − ∇π = 0 in Ω and div u = 0 in Ω,

where the Laplacian Δ = nj=1 ∂j2 acts on u componentwise. Due to its special alge-
braic format (specifically, the divergence-free condition imposed on u and the fact
that the pressure function π plays a different role than the scalar components of u),
the Stokes system does not fit directly into the general framework of homogeneous
constant coefficient second-order systems, treated earlier in §1.5, §1.7, and Chap-
ter 3 in Volume III ([70]), as well as Chapter 1, Chapter 3, and Chapter 4 in the
current volume. As such, the partial differential equations (6.0.1) warrant separate
consideration. In this chapter we shall examine aspects of the theory built around the
Stokes system in which our brand of Divergence Theorem developed in Volume I
([68]) plays a prominent role, such as Green-type formulas, boundary layer potential
operators, and Fatou-type theorems, with the goal of producing results which are
sharp from a geometric/analytic point of view.
More specifically, in §6.1 we derive a number of basic Green-type formulas for
the Stokes system in open subsets of Rn with a lower Ahlfors regular boundary and
a doubling “surface” measure. In §6.2 we treat boundary layer potential operators
for the Stokes system in open sets with uniformly rectifiable boundaries, acting from
Lebesgue, Sobolev, and Hardy spaces. In addition to other integral representation
formulas of interest, in §6.3 we establish quantitative Fatou-type theorems for the
Stokes system in UR domains. Lastly, in §6.4 we deal with boundary layer potentials
for the Stokes system on Besov, Triebel-Lizorkin, and weighted Sobolev spaces.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 673
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_6
674 6 Green Formulas and Layer Potential Operators for the Stokes System

6.1 Green-Type Formulas for the Stokes System

We begin by elaborating on the algebraic formalism associated with the Stokes

system. For a parameter λ ∈ C fixed, let
αβ
a jk (λ) := δ jk δαβ + λδ jβ δkα, 1 ≤ j, k, α, β ≤ n, (6.1.1)

and, adopting the summation convention over repeated indices, consider the second-
order, homogeneous, constant (complex) coefficient, n × n system Lλ given by
 
αβ
Lλ := a jk (λ)∂j ∂k . (6.1.2)
1≤α,β ≤n

Then for any vector u = (u1, . . . , un ) with components distributions in an open subset
of Rn we have
   
αβ
Lλ u = a jk (λ)∂j ∂k uβ = Δuα + λ ∂α (∂β uβ )
1≤α ≤n 1≤α ≤n

= Δ
u + λ ∇div
u, (6.1.3)

where the Laplacian is applied to u componentwise. Note that when acting on

divergence-free vectors, Lλ simply becomes the (vector) Laplacian.
One aspect directly affected by special algebraic format of the Stokes system is
the manner in which one associates a notion of conormal derivative with this system.
Specifically, Ω ⊆ Rn is an open set with the property that ∂Ω is lower Ahlfors
regular and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In particular, Ω is a set
of locally finite perimeter, and its geometric measure theoretic outward unit normal
ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂∗ Ω. Fix λ ∈ C and κ ∈ (0, ∞). Assume two
functions,
 1 n
u = (uβ )1≤β ≤n ∈ Lloc (Ω, L n ) with
 αβ   1  n×n
a jk (λ)∂k uβ α, j ∈ Lloc (Ω, L n ) and such that (6.1.4)
 αβ  κ−n.t.

a jk (λ)∂k uβ exists σ-a.e. on ∂∗ Ω for 1 ≤ α, j ≤ n,

∂Ω

together with
κ−n.t.
π ∈ Lloc
1
(Ω, L n ) such that π ∂Ω exists σ-a.e. on ∂∗ Ω, (6.1.5)

have been given. Then various considerations dictate that the conormal derivative
for the Stokes system, in relation to the manner in which the system in (6.1.3) has
been written, be defined as1

1 the choice λ := 1 is ubiquitous in the literature; for example, this is used in Ladyzhenskaya’s book
[54]
6.1 Green-Type Formulas for the Stokes System 675
   κ−n.t. κ−n.t. 
αβ
∂νλ (
u, π) := ν j a jk (λ)∂k uβ ∂Ω
− να π ∂Ω
1≤α ≤n
κ−n.t.
 κ−n.t. 
= ∇
u + λ(∇
u) ν − π ∂Ω ν, at σ-a.e. point on ∂∗ Ω. (6.1.6)
∂Ω

αβ
Above, the coefficients a jk (λ) are as in (6.1.1), ∇u := (∂k u j )1≤ j,k ≤n denotes the
Jacobian matrix of the vector-valued function u, and the superscript indicates
transposition of matrices.
In the next two theorems we present sharp versions of Green-type formulas for
the Stokes system. To state our first result on this topic, for each given λ ∈ C we
introduce the bilinear form
αβ β
Aλ (ξ, ζ) := a jk (λ)ξ jα ζk for all
β (6.1.7)
ξ = (ξ jα ) j,α ∈ Cn×n, ζ = (ζk )k,β ∈ Cn×n,

αβ
where the coefficients a jk (λ) are as in (6.1.1).

Theorem 6.1.1 Suppose Ω ⊆ Rn (where n ≥ 2) is an open set with a lower Ahlfors

regular boundary and with the property that σ := H n−1 ∂Ω is a doubling measure.
Hence, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Next, fix λ ∈ C and recall the
homogeneous constant (complex) coefficient second-order n × n system Lλ in Rn
defined as in (6.1.3). Also, fix p ∈ [1, ∞], denote by p its Hölder conjugate exponent,
pick some aperture parameters κ, κ > 0, and consider three functions
 1,1  n 1,1
u, w ∈ Wloc (Ω) and π ∈ Wloc (Ω) (6.1.8)

satisfying (with all derivatives taken in the sense of distributions in Ω and the bilinear
form Aλ (·, ·) as in (6.1.7)):

u), Nκ w < ∞ at σ-a.e. point on ∂Ω,

Nκ π, Nκ (∇
 
u) · Nκ w belongs to L 1 (∂Ω, σ),
Nκ π + Nκ (∇
κ−n.t. κ−n.t. κ −n.t.
u) ∂Ω , π ∂Ω , and w ∂Ω
(∇ exist σ-a.e. on ∂nta Ω, (6.1.9)

Aλ (∇  − π(divw)
u, ∇ w)  belongs to L 1 (Ω, L n ), and
 1 n
Lλ u ∈ Lloc (Ω, L n ) , Lλ u − ∇π, w ∈ L 1 (Ω, L n ).

Then the following Green-type formula (involving absolutely convergent integrals,

and the conormal derivative defined as in (6.1.6)) holds
676 6 Green Formulas and Layer Potential Operators for the Stokes System
∫ ∫
κ −n.t.
Lλ u − ∇π, w
 dL n = ∂νλ (
u, π), w ∂Ω dσ
Ω ∂∗ Ω
∫
 
− Aλ (∇  − π(divw)
u, ∇ w)  dL n, (6.1.10)
Ω

in the case when either Ω is bounded, or ∂Ω is unbounded. Moreover, formula

(6.1.10) also holds if Ω is unbounded and ∂Ω is bounded (i.e., when Ω is an exterior
domain) provided there exists μ ∈ (1, ∞) such that
∫
 
|∇  dL n = o(R) as R → ∞.
u | + |π| | w| (6.1.11)
B(0, μR)\B(0,R)

For examples, all conditions in the first two lines of (6.1.9) are satisfied if

u) ∈ L p (∂Ω, σ), Nκ w ∈ L p (∂Ω, σ),

Nκ π, Nκ (∇
(6.1.12)
for some p, p ∈ [1, ∞] with 1/p + 1/p = 1.

In such a scenario, [68, Proposition 8.9.8] implies that for any other given aperture
parameter κ > 0 we have

Nκ π, Nκ (∇
u) ∈ L p (∂Ω, σ), Nκ w ∈ L p (∂Ω, σ), (6.1.13)

and the nontangential boundary traces

κ −n.t. κ −n.t. κ −n.t.
π ∂Ω , (∇
u) ∂Ω , w ∂Ω
exist at σ-a.e. point on ∂nta Ω,
(6.1.14)
and agree with the nontangential boundary traces in (6.1.9).

We now present the proof of Theorem 6.1.1.

Proof of Theorem 6.1.1 The first line in (6.1.9) together with [68, Lemma 8.3.1]
guarantee that
∞
 ∞  n×n  ∞ n
π ∈ Lloc (Ω, L n ), ∇
u ∈ Lloc (Ω, L n ) , w ∈ Lloc (Ω, L n ) . (6.1.15)

To proceed, consider the vector field F = (Fj )1≤ j ≤n with scalar components given
by
αβ
Fj := a jk (λ)(∂k uβ )wα − πw j , 1 ≤ j ≤ n. (6.1.16)

In particular, from (6.1.16) and (6.1.15) we see that

 ∞ n  1 n
F ∈ Lloc (Ω, L n ) ⊂ Lloc (Ω, L n ) . (6.1.17)

To compute the divergence of F, in the sense of distributions in Ω, fix an arbitrary

scalar-valued function ϕ ∈ 𝒞∞
c (Ω) and write
6.1 Green-Type Formulas for the Stokes System 677
∫
D (Ω)
 ϕ
divF, D(Ω) =− Fj ∂j ϕ dL n (6.1.18)
Ω
∫ ∫
αβ
=− a jk (λ)(∂k uβ )wα ∂j ϕ dL n + πw j ∂j ϕ dL n .
Ω Ω

As in the past,  parameter ε > 0 consider the region defined

 for each sufficiently small
by Ωε := x ∈ Ω : dist(x, ∂Ω) > ε and use a Friedrichs mollifier to construct a
 n
sequence w ε = (wαε )1≤α ≤n ∈ 𝒞∞ (Ωε ) such that

for each compact set K ⊂ Ω we have

 n
w ε −−−−→
+
w in L 1 (K, L n ) and
ε→0
w ε −−−−→ w at L n -a.e. point in Ω,
ε→0+
  n×n (6.1.19)
∇ w ε −−−−→
+
∇ w in L 1 (K, L n ) ,
ε→0
and there exists some number εK > 0
so that sup0<ε<εK  w ε [L ∞ (K, L n )]n < ∞.
 1 n
Based on (6.1.15), (6.1.19), the fact that Lλ u ∈ Lloc (Ω, L n ) (cf. the last line in
(6.1.9)), and Lebesgue’s Dominated Convergence Theorem, we may then compute
678 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
αβ
− a jk (λ)(∂k uβ )wα ∂j ϕ dL n
Ω
∫
αβ
= − lim+ a jk (λ)(∂k uβ )wαε ∂j ϕ dL n
ε→0 Ω
∫
αβ
= − lim+ a jk (λ)(∂k uβ )∂j (wαε ϕ) dL n
ε→0 Ω
∫
αβ
+ lim+ a jk (λ)(∂k uβ )(∂j wαε )ϕ dL n
ε→0 Ω
 
αβ
= − lim+ D (Ω) a jk (λ)∂k uβ, ∂j (wαε ϕ) D(Ω)
ε→0
∫
αβ
+ a jk (λ)(∂k uβ )(∂j wα )ϕ dL n
Ω
  ∫
αβ
= lim+ D (Ω) a jk (λ)∂j ∂k uβ, wαε ϕ D(Ω) + Aλ (∇  dL n
u, ∇ w)ϕ
ε→0 Ω
  ∫
= lim+ D (Ω) (Lλ u)α, wαε ϕ D(Ω) + Aλ (∇  dL n
u, ∇ w)ϕ
ε→0 Ω
∫ ∫
= lim+ (Lλ u)α wαε ϕ dL n + Aλ (∇  dL n
u, ∇ w)ϕ
ε→0 Ω Ω
∫ ∫
= (Lλ u)α wα ϕ dL n + Aλ (∇  dL n
u, ∇ w)ϕ
Ω Ω
∫ ∫
= Lλ u, w ϕ dL n + Aλ (∇  dL n .
u, ∇ w)ϕ (6.1.20)
Ω Ω

Also, (6.1.15), (6.1.19), and Lebesgue’s Dominated Convergence Theorem permit

us to compute
∫ ∫
πw j ∂j ϕ dL = lim+
n
πw εj ∂j ϕ dL n
Ω ε→0 Ω
∫ ∫
= − lim+ (∂j π)w εj ϕ dL n − lim+ π(∂j w εj )ϕ dL n
ε→0 Ω ε→0 Ω
∫ ∫
=− (∂j π)w j ϕ dL − n
π(∂j w j )ϕ dL n
Ω Ω
∫ ∫
=−  dL −
∇π, wϕ n
 dL n .
π(divw)ϕ (6.1.21)
Ω Ω

Collectively, (6.1.18), (6.1.20), and (6.1.21) prove that

6.1 Green-Type Formulas for the Stokes System 679

divF = Lλ u − ∇π, w + Aλ (∇  − π(divw)

u, ∇ w)  in D (Ω). (6.1.22)

In concert with the last two lines in (6.1.9), this implies

divF ∈ L 1 (Ω, L n ). (6.1.23)

Moving on, define κ := min{κ, κ } > 0 and observe from (6.1.16) that there
exists a constant Cλ ∈ (0, ∞) such that
 
0 ≤ Nκ F ≤ Cλ Nκ (∇u) + Nκ π Nκ w
 
u) + Nκ π Nκ w on ∂Ω.
≤ Cλ Nκ (∇ (6.1.24)

From this, (6.1.9), and [68, (8.2.26)] we then conclude that

Nκ F ∈ L 1 (∂Ω, σ). (6.1.25)

Furthermore, (6.1.9) and (6.1.16) ensure that the nontangential boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω (6.1.26)

∂Ω

and, in fact, at σ-a.e. point on ∂nta Ω we have

κ −n.t.   κ−n.t.   κ −n.t.   κ−n.t.   κ −n.t. 
αβ
F = a jk (λ) ∂k uβ ) ∂Ω wα ∂Ω
− π ∂Ω w j ∂Ω
. (6.1.27)
∂Ω 1≤ j ≤n

From this, (A.0.184), (6.1.6), and [68, Proposition 8.8.6] we conclude that
 κ −n.t.  κ −n.t.
ν · F ∂Ω
= ∂νλ (
u, π), w ∂Ω
at σ-a.e. point on ∂∗ Ω. (6.1.28)

Finally, we remark that when Ω is an exterior domain, (6.1.16) and (6.1.11) guarantee
the validity of the integral growth condition [68, (1.2.3)] for the current vector field

F.
Granted the aforementioned properties of F,  [68, Theorem 1.2.1] applies and
the Divergence Formula [68, (1.2.2)] presently yields [70, (1.7.121)] on account of
(6.1.22) and (6.1.28). 

Here is our second Green-type formula for the Stokes system.

Theorem 6.1.2 Assume Ω ⊆ Rn (where n ≥ 2) is an open set with a lower Ahlfors

regular boundary and with the property that σ := H n−1 ∂Ω is a doubling measure.
Thus, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Fix a parameter λ ∈ C and recall
the homogeneous constant (complex) coefficient second-order n × n system Lλ in
Rn from (6.1.3). In addition, pick some aperture parameters κ, κ > 0, and consider
four functions
680 6 Green Formulas and Layer Potential Operators for the Stokes System
 1,1  n 1,1
u, w ∈ Wloc (Ω) and π, ρ ∈ Wloc (Ω) (6.1.29)
satisfying (with all derivatives taken in the sense of distributions in Ω and the bilinear
form Aλ (·, ·) as in (6.1.7)):

Nκ π, Nκ (∇  Nκ w < ∞ at σ-a.e. point on ∂Ω,

u), Nκ u, Nκ ρ, Nκ (∇ w),
   
Nκ π + Nκ (∇  ∈ L 1 (∂Ω, σ),
u) · Nκ w ∈ L 1 (∂Ω, σ) and Nκ u · Nκ ρ + Nκ (∇ w)
κ−n.t. κ−n.t. κ−n.t. κ −n.t. κ −n.t. κ −n.t.
u ∂Ω , (∇
u) ∂Ω , π ∂Ω , w  ∂Ω , ρ ∂Ω exist σ-a.e. on ∂nta Ω,
, (∇ w)∂Ω
 1 n
both Lλ u and Lλ w belong to the space Lloc (Ω, L n ) , and, finally,
Lλ u − ∇π, w − π(divw)
 − u, Lλ w − ∇ρ + ρ(div
u) ∈ L 1 (Ω, L n ).
(6.1.30)
Then the following Green-type formula (involving absolutely convergent integrals,
and with the conormal derivatives defined as in (6.1.6)) holds
∫  
Lλ u − ∇π, w − π(divw)
 − u, Lλ w − ∇ρ + ρ(div u) dL n
Ω
∫  
κ −n.t. κ−n.t.
= ∂νλ (
u, π), w ∂Ω
− u ∂Ω , ∂νλ (w,
 ρ) dσ, (6.1.31)
∂∗ Ω

in the case when either Ω is bounded, or ∂Ω is unbounded. Furthermore, formula

(6.1.31) continues to be valid in the case when Ω is an exterior domain provided
there exists some μ ∈ (1, ∞) such that
∫     
|∇  + |∇ w|
u | + |π| | w|  + | ρ| | u| dL n = o(R) as R → ∞. (6.1.32)
B(0, μR)\B(0,R)

For example, the assumptions in the first two lines of (6.1.30) are satisfied if

Nκ π, Nκ (∇
u) ∈ L p (∂Ω, σ), Nκ u ∈ L q (∂Ω, σ),
 ∈ L q (∂Ω, σ),
Nκ ρ, Nκ (∇ w) Nκ w ∈ L p (∂Ω, σ), (6.1.33)
with p, q, p , q ∈ [1, ∞] such that 1/p + 1/p = 1/q + 1/q = 1.

In such a scenario, [68, Proposition 8.9.8] ensure that for any given aperture param-
eter κ > 0 we have
Nκ π, Nκ (∇
u) ∈ L p (∂Ω, σ), Nκ u ∈ L q (∂Ω, σ),
(6.1.34)
 ∈ L q (∂Ω, σ),
Nκ ρ, Nκ (∇ w) Nκ w ∈ L p (∂Ω, σ),

and the nontangential boundary traces

6.1 Green-Type Formulas for the Stokes System 681
κ −n.t. κ −n.t. κ −n.t. κ −n.t. κ −n.t. κ −n.t.
π ∂Ω , u ∂Ω , (∇
u) ∂Ω , ρ ∂Ω , w ∂Ω
 ∂Ω
, (∇ w)
exist σ-a.e. on ∂nta Ω and agree with the respective (6.1.35)

pointwise boundary traces from (6.1.30).

Here is the proof of Theorem 6.1.2.

Proof of Theorem 6.1.2 For starters, observe that (6.1.29), the first line in (6.1.30),
and [68, Lemma 8.3.1] imply
 1,∞  n  n ∞
u, w ∈ Wloc (Ω) = Liploc (Ω) and π, ρ ∈ Lloc (Ω, L n ). (6.1.36)

To proceed, consider the vector field F = (Fj )1≤ j ≤n : Ω → Cn with components

given by (with the summation convention over repeated indices enforced throughout
the proof)
αβ βα
Fj := a jk (λ)(∂k uβ )wα − πw j − a jk (λ)(∂k wα )uβ + ρu j . (6.1.37)

In particular, from (6.1.37) and (6.1.36) we see that

 ∞ n  1 n
F ∈ Lloc (Ω, L n ) ⊂ Lloc (Ω, L n ) . (6.1.38)

The next order of business is to compute divF in the sense of distributions in Ω.

To this end, fix an arbitrary scalar-valued function ϕ ∈ 𝒞∞c (Ω) and write
∫

D (Ω) div F, ϕ D(Ω) = − Fj ∂j ϕ dL n
Ω
∫ ∫
αβ
=− a jk (λ)(∂k uβ )wα ∂j ϕ dL n + πw j ∂j ϕ dL n
Ω Ω
∫ ∫
βα
+ a jk (λ)(∂k wα )uβ ∂j ϕ dL n − ρu j ∂j ϕ dL n
Ω Ω

=: I + I I + I I I + IV . (6.1.39)
 
For each sufficiently small ε > 0 consider Ωε := x ∈ Ω : dist(x, ∂Ω) > ε . Using
 n
a Friedrichs mollifier, we may construct a sequence uε = (uβε )1≤β ≤n ∈ 𝒞∞ (Ωε )
such that
682 6 Green Formulas and Layer Potential Operators for the Stokes System

uε −−−−→
+
u uniformly on compact subsets of Ω,
ε→0
uε −−−−→
and ∇ +
∇
u at L n -a.e. point in Ω;
ε→0
moreover, for each fixed compact set K ⊂ Ω
 n (6.1.40)
we have Lλ uε −−−−→
+
Lλ u in L 1 (K, L n )
ε→0
and there exists some small εK > 0 such that
  ε 
u [L ∞ (K, L n )]n×n < ∞,
sup  uε [L ∞ (K, L n )]n + ∇
0<ε<ε K
 n
along with a sequence w ε = (wαε )1≤α ≤n ∈ 𝒞∞ (Ωε ) such that

w ε −−−−→
+
w uniformly on compact subsets of Ω,
ε→0
and ∇ w ε −−−−→
+
∇ w at L n -a.e. point in Ω;
ε→0
moreover, for each fixed compact set K ⊂ Ω
 1 
n) n (6.1.41)
we have Lλ w ε −−−−→
+
Lλ 
w in L (K, L
ε→0
and there exists some small εK > 0 such that
   
sup  w ε [L ∞ (K, L n )]n + ∇ w ε [L ∞ (K, L n )]n×n < ∞.
0<ε<ε K

By relying on [70, (1.7.97)], (6.1.40), (6.1.41), (6.1.3), the penultimate line in

(6.1.30), and (6.1.7) we may then compute
∫
αβ
I = − lim+ a jk (λ)(∂k uβε )wαε ∂j ϕ dL n
ε→0 Ω
∫ ∫
αβ αβ
= − lim+ a jk (λ)(∂k uβε )∂j (wαε ϕ) dL n + lim+ a jk (λ)(∂k uβε )(∂j wαε )ϕ dL n
ε→0 Ω ε→0 Ω
∫ ∫
= lim+ (Lλ uε )α wαε ϕ dL n + lim+ uε, ∇ w ε )ϕ dL n
Aλ (∇
ε→0 Ω ε→0 Ω
∫ ∫
= Lλ u, w ϕ dL n + lim+ uε, ∇ w ε )ϕ dL n .
Aλ (∇ (6.1.42)
Ω ε→0 Ω

In a similar fashion,
∫ ∫
βα
 β ϕ dL n − lim+
I I I = − uβ (Lλ w) a jk (λ)(∂j uβε )(∂k wαε )ϕ dL n
Ω ε→0 Ω
∫ ∫
=− u, Lλ w ϕ dL n − lim+ uε, ∇ w ε )ϕ dL n,
Aλ (∇ (6.1.43)
Ω ε→0 Ω

so that
6.1 Green-Type Formulas for the Stokes System 683
∫ ∫
I + III = Lλ u, w ϕ dL n − u, Lλ w ϕ dL n . (6.1.44)
Ω Ω

Also, (6.1.36), (6.1.40)-(6.1.41), and Lebesgue’s Dominated Convergence Theorem

permit us to compute
∫
I I = lim+ πw εj ∂j ϕ dL n
ε→0 Ω
∫ ∫
= − lim+ (∂j π)w εj ϕ dL n − lim+ π(∂j w εj )ϕ dL n
ε→0 Ω ε→0 Ω
∫ ∫
=−  dL −
∇π, wϕ n
 dL n,
π(divw)ϕ (6.1.45)
Ω Ω

and, likewise,
∫ ∫
IV = ∇ρ, uϕ dL n + ρ(div
u)ϕ dL n . (6.1.46)
Ω Ω

By combining (6.1.39), and (6.1.44)-(6.1.46) we conclude that

divF = Lλ u − ∇π, w − π(divw)

 − u, Lλ w − ∇ρ + ρ(div
u) in D (Ω). (6.1.47)

Granted this, the last line in (6.1.30) ensures that

divF ∈ L 1 (Ω, L n ). (6.1.48)

Moreover, if κ := min{κ, κ } > 0 then from (6.1.37) we see that at each point on
∂Ω we have
    
Nκ F ≤ C Nκ (∇ u) + Nκ π Nκ w + Nκ (∇ w)  + Nκ ρ Nκ u
    
u) + Nκ π Nκ w + Nκ (∇ w)
≤ C Nκ (∇  + Nκ ρ Nκ u , (6.1.49)

for some constant C = C(λ) ∈ (0, ∞) which depends only on λ. In turn, from (6.1.49),
(6.1.34), and [68, (8.2.26)] we conclude that

Nκ F ∈ L 1 (∂Ω, σ). (6.1.50)

Let us also observe that (6.1.37) and (6.1.30) imply that the nontangential boundary
κ −n.t.
trace F ∂Ω exists at σ-a.e. point on ∂nta Ω. Concretely, at σ-a.e. point on ∂nta Ω we
have
684 6 Green Formulas and Layer Potential Operators for the Stokes System

κ −n.t.
αβ  κ−n.t.   κ −n.t.   κ−n.t.   κ −n.t. 
F = a jk (λ) (∂k uβ ) ∂Ω wα ∂Ω
− π ∂Ω w j ∂Ω
∂Ω


 κ−n.t.  βα  κ −n.t.   κ −n.t.  κ−n.t. 
− uβ ∂Ω
a jk (λ) (∂k wα ) ∂Ω + ρ ∂Ω uj ∂Ω
. (6.1.51)
1≤ j ≤n

In turn, from (6.1.51), (6.1.6), and [68, Proposition 8.8.6], we conclude that
 κ −n.t.  κ −n.t. κ−n.t.
ν · F ∂Ω
= ∂νλ (
u, π), w ∂Ω − u ∂Ω , ∂νλ (w,
 ρ) (6.1.52)

at σ-a.e. point on ∂∗ Ω. Finally, in the case when Ω is an exterior domain, (6.1.32)

guarantees that the vector field F satisfies the growth condition [68, (1.2.3)]. Granted
 [68, Theorem 1.2.1] applies and the Divergence Formula [68,
these properties of F,
(1.2.2)] currently yields (6.1.31) on account of (6.1.47) and (6.1.52). 

6.2 Boundary Layer Potential Operators for the Stokes System:

Lebesgue, Sobolev, and Hardy Spaces

The goal here is to introduce and study boundary layer potential operators for the
Stokes system considered in a general class of subsets of Rn , with  n≥  2. To set the
stage, recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n of the
Stokes system in Rn , whose ( j, k)-entry is defined at each x = (x j )1≤ j ≤n ∈ Rn \ {0}
by  
⎧
⎪ 1 1 δ jk x j xk
⎪
⎪ − + if n ≥ 3,
⎪
⎨ 2ωn−1 n − 2 |x| n−2
⎪ |x| n
E jk (x) :=   (6.2.1)
⎪
⎪ x j xk
⎪
⎪ 1
⎪− −δ jk ln |x| + if n = 2,
⎩ 4π |x| 2
and the accompanying pressure vector q given by
  1 x
 = q j (x) 1≤ j ≤n := −
q(x) , ∀x ∈ Rn \ {0}. (6.2.2)
ωn−1 |x| n
See, e.g., [66, Theorem 10.29, p. 382] when n ≥ 3 and [41, (2.3.3), p. 63] when
n = 2. Regarding them as functions defined L n -a.e. in Rn , these are locally integrable
functions of (at most) slow growth at infinity. As such, the distributions they induce
in Rn are tempered, i.e.,
    n×n    n
E = E jk 1≤ j,k ≤n ∈ 𝒮 (Rn ) and q = q j 1≤ j ≤n ∈ 𝒮 (Rn ) . (6.2.3)

Interpreted as such, they satisfy in the sense of distributions in Rn ,

6.2 Boundary Layer Potential Operators for the Stokes System . . . 685

ΔE jk − ∂j qk = δ jk δ for each j, k ∈ {1, . . . , n}, (6.2.4)

∂j E jk = 0 for each k ∈ {1, . . . , n} and ∂k E jk = 0 for each j ∈ {1, . . . , n},

(6.2.5)

E jk = Ek j and ∂j qk = ∂k q j for each j, k ∈ {1, . . . , n}, (6.2.6)

q = −∇EΔ and divq = ∂j q j = −δ and Δq = −∇δ, (6.2.7)

where δ is the Dirac delta-function and EΔ is the standard fundamental solution for
the Laplacian in Rn (cf. (A.0.65)). Moreover, for each j, k,  ∈ {1, . . . , n} we have
ξj ξ ξ δ jk 
qj (ξ) = i and 
∂ E jk (ξ) = iξ
j k
−
|ξ | 2 |ξ | 4 |ξ | 2 (6.2.8)
for each ξ = (ξr )1≤r ≤n ∈ Rn \ {0}.

Indeed, the first formula above is a particular case of [66, Corollary 4.65, p. 147].
As regards the second formula in (6.2.8), the case n ≥ 3 is clear from [66, (10.6.15),
p. 382], while the case n = 2 may be treated using [66, Proposition 4.73, pp. 153-
154].
In particular, (6.2.4)-(6.2.7) imply that at each x ∈ Rn \ {0} we have

∂k E jk (x) = 0 for 1 ≤ j ≤ n, ∂j E jk (x) = 0 for 1 ≤ k ≤ n,

ΔE jk (x) = ΔEk j (x) = ∂k q j (x) = ∂j qk (x) for 1 ≤ j, k ≤ n, (6.2.9)
 = 0, and ∂j q j (x) = 0.
Δq(x)

Also,

Δ2 E jk (x) = 0 for 1 ≤ j, k ≤ n, and x ∈ Rn \ {0}. (6.2.10)

Consider next an open set Ω ⊆ Rn and abbreviate σ := H n−1 ∂Ω. In this setting,
define the action of the boundary-to-domain single layer potential operator
for the Stokes system 𝒮 on each function2
 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂Ω, (6.2.11)
1 + |x| n−2
according to
∫
𝒮 f(x) := E(x − y) f(y) dσ(y)
∂Ω
∫ 
= E jk (x − y) fk (y) dσ(y) for each x ∈ Ω. (6.2.12)
∂Ω 1≤ j ≤n

2 when n = 2, the weight (1 + |x | n−2 )−1 should be replaced by ln(2 + |x |)

686 6 Green Formulas and Layer Potential Operators for the Stokes System

The corresponding (single layer) pressure potential acts on functions

 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂Ω, (6.2.13)
1 + |x| n−1
according to
∫
Q f(x) :=  − y), f(y) dσ(y)
q(x
∂Ω
∫
= q j (x − y) f j (y) dσ(y) for each x ∈ Ω. (6.2.14)
∂Ω

Then, on account of (6.2.9), it follows that for each function f as in (6.2.11) we have

Δ𝒮 f − ∇Q f = 0 and div𝒮 f = 0 in Ω. (6.2.15)

This may be interpreted as saying that, for each function f as in (6.2.11), the pair
 
𝒮 f, Q f is a null-solution for the Stokes system in Ω.
Let us now consider an open set Ω ⊆ Rn of locally finite perimeter and abbre-
viate σ := H n−1 ∂Ω. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic
outward unit normal to Ω, which is well defined at σ-a.e. point on ∂∗ Ω. In such a
setting, the action of the boundary-to-domain double layer potential operator
for the Stokes system Dλ on an arbitrary function
 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω, (6.2.16)
1 + |x| n−1
is given at each x ∈ Ω by3

3 with E. γ denoting the γ-th column of the matrix-valued function E

6.2 Boundary Layer Potential Operators for the Stokes System . . . 687
 ∫ 
 
λ  
Dλ f(x) := ∂ν(y) E. γ (y − x), qγ (y − x) , f(y) dσ(y)
∂∗ Ω 1≤γ ≤n
 ∫ 
 
λ  
= ∂ν(y) E. γ (x − y), −qγ (x − y) , f(y) dσ(y)
∂∗ Ω 1≤γ ≤n
 ∫

αβ
= − ν j (y)a jk (λ)(∂k Eβγ )(x − y)
∂∗ Ω


+ να (y)qγ (x − y) fα (y)(y) dσ(y)
1≤γ ≤n
 ∫

= − νk (y)(∂k E jγ )(x − y) − λνk (y)(∂j Ekγ )(x − y)
∂∗ Ω


+ ν j (y)qγ (x − y) f j (y) dσ(y) . (6.2.17)
1≤γ ≤n

In a more explicit fashion, for each vector-valued function f as in (6.2.16) we have

Dλ f(x) (6.2.18)
 ∫
δ jγ x − y, ν(y)
= (λ − 1)
2ωn−1 |x − y| n
∂∗ Ω

n x − y, ν(y)(x j − y j )(xγ − yγ )
− (λ + 1)
2ωn−1 |x − y| n+2
! 
1 ν j (y)(xγ − yγ ) − νγ (y)(x j − y j )
+ (λ − 1) f j (y) dσ(y) .
2ωn−1 |x − y| n
1≤γ ≤n

We wish to note here that, as a consequence of (6.2.17), (6.2.4), (6.2.5), and the
Divergence Formula in [68, Corollary 1.5.2], much as in the case of (1.3.45), for
each constant c ∈ Cn we have
c in Ω, if Ω is bounded,
if ∂Ω is compact then Dλ c = (6.2.19)
0 in Ω, if Ω is unbounded.

Let us also define the action of the corresponding (double layer) pressure
potential on functions
688 6 Green Formulas and Layer Potential Operators for the Stokes System
 σ(x)  n
f ∈ L 1 ∂∗ Ω, (6.2.20)
1 + |x| n
by setting, for each x ∈ Ω,
∫
Pλ f(x) := −(1 + λ)  − y), f(y) dσ(y)
ν j (y) (∂j q)(x
∂∗ Ω
∫
= −(1 + λ) ν j (y)(∂j qk )(x − y) fk (y) dσ(y). (6.2.21)
∂∗ Ω

Then (6.2.9) implies that for each function f as in (6.2.16) we have

ΔDλ f − ∇Pλ f = 0 and divDλ f = 0 in Ω. (6.2.22)

 
Hence, for each function f as in (6.2.16), the pair Dλ f, Pλ f is a null-solution for
the Stokes system in Ω.
In relation to these operators we have the following boundary layer integral
representation formulas for null-solutions of the Stokes system.

Theorem 6.2.1 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on  n ∂∗ Ω. Also, fix an aperture
parameter κ > 0. Suppose the functions u ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) satisfy

Δ
u − ∇π = 0 in Ω, u = 0 in Ω, and
div
κ−n.t. κ−n.t. κ−n.t. (6.2.23)
u ∂Ω , (∇
u) ∂Ω , π ∂Ω exist σ-a.e. on ∂nta Ω.

Then the following conclusions are true.

(a) If n ≥ 3 and one has

∫ ∫  
(Nκ u)(y) Nκ (∇
u) (y) + (Nκ π)(y)
dσ(y) < ∞ and dσ(y) < ∞,
1 + |y| n−1 1 + |y| n−2
∂Ω ∂Ω
(6.2.24)
κ −n.t. κ −n.t. κ −n.t.
then for any κ > 0 the nontangential traces u ∂Ω (∇ u) ∂Ω , π ∂Ω also exist
at σ-a.e. point on ∂nta Ω and are actually independent of the aperture parameter
κ . Moreover, with the dependence on the parameter κ dropped, for each λ ∈ C
one has4
 n.t.   
u(x) = Dλ u ∂Ω (x) − 𝒮 ∂νλ (u, π) (x), for all x ∈ Ω, (6.2.25)

4 here we allow the operator 𝒮 to act on functions originally defined only on ∂∗ Ω by extending
them by zero to the entire topological boundary ∂Ω
6.2 Boundary Layer Potential Operators for the Stokes System . . . 689

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, formula (6.2.25) continues to be true under the additional assumption
that there exists μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL n = o(1) as R → ∞. (6.2.26)
B(0, μR)\B(0,R)

(b) If n ≥ 3 and one has

∫ ∫  
(Nκ u)(y) Nκ (∇
u) (y) + (Nκ π)(y)
dσ(y) < ∞ and dσ(y) < ∞,
1 + |y| n 1 + |y| n−1
∂Ω ∂Ω
(6.2.27)
κ −n.t. κ −n.t. κ −n.t.
then for any κ > 0 the nontangential traces u ∂Ω (∇ u) ∂Ω , π ∂Ω also exist
at σ-a.e. point on ∂nta Ω and are actually independent of κ . Furthermore, with
the dependence on the parameter κ dropped, for each λ ∈ C one has5
 n.t.   
π(x) = Pλ u ∂Ω (x) − Q ∂νλ (u, π) (x), for all x ∈ Ω, (6.2.28)

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that
there exists μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL n = o(R) as R → ∞. (6.2.29)
B(0, μR)\B(0,R)

(c) Assume n ≥ 2 and Ω is an exterior domain. Also, suppose there exists a trunca-
tion parameter ε > 0 for which

Nκε u, Nκε (∇

u), Nκε π ∈ L 1 (∂Ω, σ). (6.2.30)

Finally, make the assumption that there exists some μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL n = o(R) as R → ∞. (6.2.31)
B(0,μR)\B(0,R)

κ −n.t. κ −n.t. κ −n.t.

Then for any κ > 0 the nontangential traces u ∂Ω (∇ u) ∂Ω , π ∂Ω also exist
at σ-a.e. point on ∂nta Ω and are actually independent of κ (so the dependence
on the aperture parameter may be dropped).
In addition, there exists a constant c ∈ Cn with the property that for each λ ∈ C
one has6

5 again, allowing the operator Q to act on functions originally defined only on ∂∗ Ω by extending
them by zero to the entire topological boundary ∂Ω
6 once more allowing the operators 𝒮, Q to act on functions originally defined only on ∂∗ Ω by
extending them by zero to the entire topological boundary ∂Ω
690 6 Green Formulas and Layer Potential Operators for the Stokes System
n.t.   
u(x) = Dλ u ∂Ω (x) − 𝒮 ∂νλ (
u, π) (x) + c, for all x ∈ Ω, (6.2.32)

and
 n.t.   
π(x) = Pλ u ∂Ω (x) − Q ∂νλ (
u, π) (x), for all x ∈ Ω, (6.2.33)

(d) Make the assumption that n = 2, the set ∂Ω is compact, and there exists a
truncation parameter ε > 0 for which (6.2.30) holds.
κ −n.t. κ −n.t. κ −n.t.
Then for any κ > 0 the nontangential traces u ∂Ω (∇ u) ∂Ω , π ∂Ω also exist
at σ-a.e. point on ∂nta Ω and are actually independent of κ .
Moreover, if Ω is bounded then the integral representation formulas in (6.2.25)
and (6.2.28) are valid. Finally, if Ω is unbounded then (6.2.25) holds provided
there exists μ ∈ (1, ∞) such that
⨏
   
| u| + R|π| dL 2 = o ln1R as R → ∞, (6.2.34)
B(0,μR)\B(0,R)

while (6.2.28) holds provided there exists some μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL 2 = o(R) as R → ∞. (6.2.35)
B(0,μR)\B(0,R)

Proof Let us first deal with item (a), under the additional assumption made in
(6.2.24). First, from (6.2.24) and [68, Corollary 8.9.9] we deduce that for any κ > 0
the nontangential traces
κ −n.t. κ −n.t. κ −n.t.
u ∂Ω , (∇
u) ∂Ω , π ∂Ω exist at σ-a.e. point on ∂nta Ω,
(6.2.36)
and are actually independent of the aperture parameter κ .

In addition, with the dependence on κ dropped, from (6.2.24), [68, (8.9.8)], [68,
(8.9.44)], and [68, (8.8.52)] we conclude that
∫  n.t. 
∫  n.t.   n.t. 
u ∂Ω (y) u) ∂Ω (y) + π ∂Ω (y)
(∇
dσ(y) < ∞ and dσ(y) < ∞.
1 + |y| n−1 1 + |y| n−2
∂∗ Ω ∂∗ Ω
(6.2.37)
In particular, having selected some λ ∈ C, from (6.2.37) and (6.2.1)-(6.2.2) it
 n.t. 
follows that, for each fixed point x ∈ Ω, the integrals defining Dλ u ∂Ω (x) and
 λ 
𝒮 ∂ν (
u, π) (x) are absolutely convergent.
To proceed, fix an arbitrary index γ ∈ {1, . . . , n}, pick an arbitrary point x ∈ Ω,
and consider the vector field Fx = Fj 1≤ j ≤n with components given at L n -a.e. point
in Ω by (recall that, throughout, the summation convention over repeated indices is
in effect)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 691
αβ
Fj := −a jk (λ)(∂k Eβγ )(x − ·)uα + qγ (x − ·)u j
αβ
− Eγα (x − ·)a jk (λ)∂k uβ + Eγ j (x − ·)π. (6.2.38)

Then from (6.2.38) and (6.2.1)-(6.2.2) we see that

 1 n
Fx ∈ Lloc (Ω, L n ) . (6.2.39)

To compute divFx in the sense of distributions in the open set Ω, first observe from
 − ·) = −q(·
definitions that we have E(x − ·) = E(· − x) and q(x  − x). Also, we find it
convenient to express (∂k Eβγ )(x − ·) as −∂k [Eβγ (· − x)] then, using (6.2.4)-(6.2.7),
the first line in (6.2.23), and (6.1.3), write
 
αβ αβ
divFx = ∂j Fj = a jk (λ)∂j ∂k [Eβγ (· − x)] uα + ∂k [Eβγ (· − x)]a jk (λ)(∂j uα )

− ∂j [qγ (· − x)]u j − qγ (· − x)(∂j u j )

αβ αβ
− ∂j [Eγα (· − x)]a jk (λ)∂k uβ − Eγα (· − x)a jk (λ)∂j ∂k uβ

+ ∂j [Eγ j (· − x)]π + Eγ j (· − x)∂j π

 
= Lλ [E. γ (· − x)] α uα − ∂j [qγ (· − x)]u j − Eγα (· − x)(Lλ u)α

+ Eγ j (· − x)∂j π
 
= Lλ [E. γ (· − x)] − ∇[qγ (· − x)] uα − Eγα (· − x)(Lλ u − ∇π)α
α

= δαγ uα δx = uγ (x)δx (6.2.40)

where δx is Dirac’s delta distribution with mass at x. Hence,

divFx = uγ (x)δx ∈ ℰ (Ω). (6.2.41)

 
Next, consider K := B x, 12 dist(x, ∂Ω) which is a compact subset of Ω, and
based on (6.2.38), (6.2.1)-(6.2.2), and [68, Lemma 8.3.7] estimate, at each y ∈ ∂Ω,
     
NκΩ\K Fx (y) ≤ C Nκ u (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
      
+ C Nκ (∇
u) (y) + Nκ π (y) · sup |x − z| 2−n
z ∈Γκ (y)\K
     !
Nκ u (y) Nκ (∇
u) (y) + Nκ π (y)
≤C + , (6.2.42)
|x − y| n−1 |x − y| n−2
692 6 Green Formulas and Layer Potential Operators for the Stokes System

for some constant C = C(Ω, n, λ, κ) ∈ (0, ∞). From (6.2.42), (6.2.24), and [68,
(8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (6.2.43)

Moreover, from (6.2.38), [68, (8.9.10)-(8.9.11)], and (6.2.36) we conclude that

n.t.

Fx exists at σ-a.e. point on ∂nta Ω (6.2.44)

∂Ω

and, in fact, for each j ∈ {1, . . . , n} at σ-a.e. point y ∈ ∂nta Ω we have

 n.t.   n.t.   n.t. 
αβ
Fj (y) = −a jk (λ)(∂k Eβγ )(x − y) uα ∂Ω (y) + qγ (x − y) u j ∂Ω
(y)
∂Ω

αβ  n.t. 
− Eγα (x − y)a jk (λ) (∂k uβ ) ∂Ω (y)
 n.t. 
+ Eγ j (x − y) π ∂Ω (y). (6.2.45)

From (6.2.45), (6.1.6), and [68, (8.8.52)] we then conclude that at σ-a.e. point
y ∈ ∂∗ Ω we have
 n.t.   n.t. 
ν(y) · Fx (y) = ν j (y) Fj (y)
∂Ω ∂Ω
   n.t. 
αβ
= − ν j (y)a jk (λ)(∂k Eβγ )(x − y) + να (y)qγ (x − y) uα ∂Ω (y)
  n.t.   n.t.  
αβ
− Eγα (x − y) ν j (y)a jk (λ) (∂k uβ ) ∂Ω (y) − να (y) π ∂Ω (y)
   n.t.  
λ 
= ∂ν(y) E. γ (x − y), −qγ (x − y) , u ∂Ω (y)
 
− Eγ .(x − y), ∂νλ (
u, π)(y) . (6.2.46)

In the case when Ω is an exterior domain, condition (6.2.26) self-improves thanks to

[70, Lemma 1.5.6] (bearing in mind that Δ2 u = 0) to
⨏   
| u| + R |∇
u | + |π| dL n = o(1) as R → ∞. (6.2.47)
B(0, μR)\B(0,R)

In turn, (6.2.47) together with (6.2.1)-(6.2.2) guarantee that

the vector field Fx satisfies the decay condition [68, (1.4.8)]. (6.2.48)

Collectively, (6.2.39), (6.2.41), (6.2.44), and (6.2.48) ensure that, for each point
x ∈ Ω such that [70, (1.5.12)] holds, the vector field Fx satisfies the hypotheses
of [68, Theorem 1.4.1]. On account of [68, (4.6.21)], (6.2.41), and (6.2.46), the
6.2 Boundary Layer Potential Operators for the Stokes System . . . 693

Divergence Formula [68, (1.4.6)] presently yields

 
uγ (x) = ℰ (Ω) uγ (x)δx, 1 ℰ(Ω) = (𝒞∞b (Ω))∗ uγ (x)δx, 1 𝒞∞b (Ω)
∫
   n.t. 
= (𝒞∞b (Ω))∗ divFx, 1 𝒞∞b (Ω) = ν · Fx ∂Ω
dσ
∂∗ Ω
∫ 
λ    n.t.  
= ∂ν(y) E. γ (x − y), −qγ (x − y) , u ∂Ω (y) dσ(y)
∂∗ Ω
∫  
− Eγ .(x − y), ∂νλ (
u, π)(y) dσ(y),
∂∗ Ω
 n.t. 
   
= Dλ u ∂Ω (x) − 𝒮 ∂νλ (
u, π) (x) (6.2.49)
γ

for each point x ∈ Ω and each γ ∈ {1, . . . , n}. This establishes (6.2.25).
Moving on, the first claims in item (b) are dealt with in a similar manner, so
we focus on establishing the integral representation formula (6.2.28), under the
  To this end, fix an arbitrary point x ∈ Ω and consider
assumption made in (6.2.27).
the vector field G x = G j 1≤ j ≤n with components given at L n -a.e. point in Ω by

αβ
G j := −(1 + λ)(∂j qk )(x − ·)uk − qα (x − ·)a jk (λ)∂k uβ + q j (x − ·)π. (6.2.50)

Note that (6.2.50) and (6.2.1)-(6.2.2) guarantee that

 1 n
G x ∈ Lloc (Ω, L n ) . (6.2.51)

The next order of business is to compute divG x in the sense of distributions in

 − ·) = −q(·
Ω. In this vein, first recall that q(x  − x). Also, express (∂j qk )(x − ·) as
∂j [qk (· − x)] then rely on (6.1.1), (6.1.3), the first line in (6.2.23), and (6.2.6)-(6.2.7)
to write
694 6 Green Formulas and Layer Potential Operators for the Stokes System
   
divG x = ∂j G j = −(1 + λ) ∂j ∂j [qk (· − x)] uk − (1 + λ) ∂j [qk (· − x)] (∂j uk )
αβ αβ
+ ∂j [qα (· − x)]a jk (λ)∂k uβ + qα (· − x)a jk (λ)∂j ∂k uβ

− ∂j [q j (· − x)]π − q j (· − x)∂j π
   
= −(1 + λ) Δ[qk (· − x)] uk − (1 + λ) ∂j [qk (· − x)] (∂j uk )
   
+ ∂j [qα (· − x)] δαβ δ jk + λδ jβ δkα ∂k uβ + qα (· − x) Lλ u α

+ πδx − q j (· − x)∂j π
 
= (1 + λ)(∂k δx )uk − (1 + λ) ∂j [qk (· − x)] (∂j uk )

+ ∂j [qα (· − x)]∂j uα + λ∂β [qα (· − x)]∂α uβ + qα (· − x)∂α π

+ πδx − q j (· − x)∂j π

= (1 + λ)uk (∂k δx ) + π(x)δx . (6.2.52)

This shows that

divG x = (1 + λ)uk (∂k δx ) + π(x)δx ∈ ℰ (Ω). (6.2.53)

For further reference, let us observe that

 
(𝒞∞

(Ω))∗ divG x , 1 𝒞∞ (Ω) = ℰ (Ω) (1 + λ)uk (∂k δx ) + π(x)δx , 1 ℰ(Ω)
b b

= (1 + λ)ℰ (Ω) uk (∂k δx ), 1 ℰ(Ω) + π(x)ℰ (Ω) δx, 1 ℰ(Ω)

= (1 + λ)ℰ (Ω) ∂k δx, uk ℰ(Ω) + π(x)

= −(1 + λ)ℰ (Ω) δx, div

u ℰ(Ω) + π(x)

= π(x), (6.2.54)

thanks to [68, (4.6.21)] and the fact that u is divergence-free (cf. (6.2.23)).
 
Next, with the compact K := B x, 12 dist(x, ∂Ω) as before, based on (6.2.50),
(6.2.2), and [68, Lemma 8.3.7], at each y ∈ ∂Ω we may estimate
     
NκΩ\K G x (y) ≤ Nκ u (y) · sup |x − z| −n
z ∈Γκ (y)\K
      
+ Nκ (∇
u) (y) + Nκ π (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
      !
Nκ u (y) Nκ (∇
u) (y) + Nκ π (y)
≤C + , (6.2.55)
|x − y| n |x − y| n−1
6.2 Boundary Layer Potential Operators for the Stokes System . . . 695

for some constant C = C(Ω, λ, κ) ∈ (0, ∞). From (6.2.55), (6.2.27), and [68, (8.2.26)]
it follows that

NκΩ\K G x ∈ L 1 (∂Ω, σ). (6.2.56)

In addition, from (6.2.50), the second line in (6.2.23), [68, (8.9.10)-(8.9.11)], and
[70, (1.5.18)] we conclude that
n.t.

G x exists at σ-a.e. point on ∂nta Ω (6.2.57)

∂Ω

and, in fact, for each j ∈ {1, . . . , n} at σ-a.e. point y ∈ ∂nta Ω we have

 n.t.   n.t. 
Gj (y) = −(1 + λ)(∂j qk )(x − y) uk ∂Ω
(y)
∂Ω

αβ  n.t.   n.t. 
− qα (x − y)a jk (λ) (∂k uβ ) ∂Ω (y) + q j (x − ·) π ∂Ω (y). (6.2.58)

Using (6.2.58), (6.1.6), and [68, (8.8.52)] we then see that at σ-a.e. point y ∈ ∂∗ Ω
we have
 n.t.   n.t. 
ν(y) · G x (y) = −(1 + λ)ν j (y)(∂j qk )(x − y) uk ∂Ω (y)
∂Ω

αβ  n.t. 
− qα (x − y)ν j (y)a jk (λ) (∂k uβ ) ∂Ω (y)
 n.t. 
+ q j (x − y)ν j (y) π ∂Ω (y)
 n.t. 
 − y), u ∂Ω (y)
= −(1 + λ)ν j (y) (∂j q)(x
 
− qα (x − y) ∂νλ (
u, π) α (y). (6.2.59)

Finally, when Ω is an exterior domain condition (6.2.29) self-improves (much as

(6.2.26) implied (6.2.47) via interior estimates) to
⨏   
| u| + R |∇
u | + |π| dL n = o(R) as R → ∞. (6.2.60)
B(0, μR)\B(0,R)

Together with (6.2.2), this guarantees that

the vector field G x satisfies the decay condition [68, (1.4.8)]. (6.2.61)

Granted (6.2.51), (6.2.53), (6.2.56), (6.2.57), (6.2.61), it follows that the vector
field G x satisfies the hypotheses of [68, Theorem 1.4.1]. As such we may rely on
(6.2.54), and the Divergence Formula [68, (1.4.6)] to write, on account of (6.2.59),
696 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
   n.t. 
π(x) = (𝒞∞b (Ω))∗ divG x, 1 𝒞∞ (Ω) = ν · G x ∂Ω
dσ
b ∂∗ Ω
∫
 n.t. 
= −(1 + λ)  − y), u ∂Ω (y) dσ(y)
ν j (y) (∂j q)(x
∂∗ Ω
∫
 
− q j (x − y) ∂νλ (
u, π) j (y) dσ(y)
∂∗ Ω
n.t.   
= Pλ u ∂Ω (x) − Q ∂νλ (
u, π) (x) (6.2.62)

for each point x ∈ Ω. This establishes (6.2.28) and completes the treatment of item
(b).
In fact, a very similar treatment applies to item (d), since the assumptions made
on that occasion guarantee that all hypotheses of [68, Corollary 1.5.2] are satisfied,
and that [68, (1.5.22)] holds (with n = 2).
There remains to deal with item (c). Hence, we shall work under the assumption
that Ω is an exterior domain (in particular, ∂Ω is compact). The opening claims in
item (c) are justified as before, so we focus on the integral representation formulas
claimed in (6.2.32) and (6.2.33). In the case of (6.2.33), the very same argument as
in the proof of item (b) works in the current setting, if we now employ the Divergence
Theorem recorded in [68, Corollary 1.5.2], bearing in mind that (6.2.31) is identical
to (6.2.29).
To justify the integral representation formula claimed in (6.2.32), select an ar-
bitrary index γ ∈ {1, . . . , n} and fix two arbitrary points x0, x1 ∈ Ω. In relation to
these, define the vector field

Fx0,x1 := Fx0 − Fx1 (6.2.63)

where the vector fields Fx0 and Fx1 are associated with the points x0 and x1 , respec-
 
tively, like the vector field Fx = Fj 1≤ j ≤n has been associated with the point x ∈ Ω
in (6.2.38). By (6.2.63), (6.2.39), and (6.2.40) we have
 1 n
Fx0,x1 ∈ Lloc (Ω, L n ) and div Fx0,x1 = uγ (x0 )δx0 − uγ (x1 )δx1 ∈ ℰ (Ω). (6.2.64)

From definitions, (6.2.1)-(6.2.2), and (6.2.30) we also see that

Nκε Fx0,x1 ∈ L 1 (∂Ω, σ). (6.2.65)

Together, (6.2.63), (6.2.44), (6.2.46) imply

n.t.

Fx0,x1 exists at σ-a.e. point on ∂nta Ω, (6.2.66)

∂Ω

and at σ-a.e. point y ∈ ∂nta Ω we have

6.2 Boundary Layer Potential Operators for the Stokes System . . . 697
     n.t.  
λ 
n.t.

ν(y) · Fx0,x1 (y) = ∂ν(y) E. γ (x0 − y) , −qγ (x0 − y) , u ∂Ω (y)

∂Ω
 
− Eγ . (x0 − y) , ∂νλ (
u, π)(y)
   n.t.  
λ 
− ∂ν(y) E. γ (x1 − y) , −qγ (x1 − y) , u ∂Ω (y)
 
+ Eγ . (x1 − y) , ∂νλ (
u, π)(y) . (6.2.67)

Finally, from (6.2.63), (6.2.38), (6.2.1)-(6.2.2), the Mean Value Theorem, and
(6.2.31) we deduce that
⨏
Fx0,x1 dL n = o(R1−n ) as R → ∞. (6.2.68)
B(0,μR)\B(0,R)

Collectively, (6.2.64), (6.2.65), (6.2.66), and (6.2.68) ensure that all hypotheses of
[68, Corollary 1.5.2] are satisfied, and that [68, (1.5.22)] holds for the vector field
Fx0,x1 . Consequently, for this vector field we may use the Divergence Formula in the
version recorded in [68, (1.5.20)] which, thanks to (6.2.64), (6.2.67), (6.2.17), and
(6.2.12), presently gives
  n.t.     
uγ (x0 ) − uγ (x1 ) = Dλ u ∂Ω (x0 ) − 𝒮 ∂νλ (
u, π) (x0 )
γ γ
 n.t.     
− Dλ u ∂Ω γ (x1 ) + 𝒮 ∂νλ (
u, π) (x1 ). (6.2.69)
γ

In view of the arbitrariness of x0, x1 ∈ Ω, this may be interpreted as saying that the
function
  n.t.     
Ω  x → uγ (x) − Dλ u ∂Ω (x) + 𝒮 ∂νλ ( u, π) (x) ∈ C (6.2.70)
γ γ

is constant in Ω. Denoting its value by cγ ∈ C and then defining c := (cγ )1≤γ ≤n ∈ Cn

ultimately establishes (6.2.32). This finishes the proof of Theorem 6.2.1. 

There is yet another set of integral representation formulas for null-solutions of

the Stokes system, of the sort described in the theorem below.

Theorem 6.2.2 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit
normal to Ω, and fix an aperture
n parameter κ ∈ (0, ∞). Suppose the functions
u = (u1, . . . , un ) ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) satisfy
698 6 Green Formulas and Layer Potential Operators for the Stokes System

Δ
u − ∇π = 0 in Ω, u = 0 in Ω,
div
κ−n.t. κ−n.t.
(∇
u) ∂Ω and π ∂Ω exist σ-a.e. on ∂nta Ω,
(6.2.71)
∫  
Nκ (∇
u) (y) + (Nκ π)(y)
dσ(y) < +∞.
∂Ω 1 + |y| n−1
In the case when Ω is an exterior domain, make the additional assumption that there
exists some μ ∈ (1, ∞) such that
⨏
 
|∇u| + |π| dL n = o(1) as R → ∞. (6.2.72)
B(0, μ R)\B(0,R)

Then, given any r, γ ∈ {1, . . . , n} and λ ∈ C, one has

∫   κ−n.t. 
(∂r uγ )(x) = (∂k E jγ )(x − y) νr (y) (∂k u j ) ∂Ω (y)
∂∗ Ω
 κ−n.t.  
− νk (y) (∂r u j ) ∂Ω (y) dσ(y)
∫   κ−n.t. 
+λ (∂j Ekγ )(x − y) νr (y) (∂k u j ) ∂Ω (y)
∂∗ Ω
 κ−n.t.  
− νk (y) (∂r u j ) ∂Ω (y) dσ(y)
∫
 κ−n.t. 
+ qγ (x − y)ν j (y) (∂r u j ) ∂Ω (y) dσ(y)
∂∗ Ω
∫
 
− (∂r Eγα )(x − y) ∂νλ (
u, π) α (y) dσ(y), ∀x ∈ Ω, (6.2.73)
∂∗ Ω

and
∫
 κ−n.t. 
π(x) = (1 + λ) q j (x − y)νk (y) ∂j uk ∂Ω
(y) dσ(y)
∂∗ Ω
∫
 
− qα (x − y) ∂νλ (
u, π) α (y) dσ(y), ∀x ∈ Ω. (6.2.74)
∂∗ Ω

Proof Fix an arbitrary point x ∈ Ω along with two indexes r, γ ∈ {1, . . . , n}.
Consider the vector field defined at L n -a.e. point in Ω as
  
Fx := (∂k E jγ )(x − ·) + λ(∂j Ekγ )(x − ·) (∂k u j )er − (∂r u j )ek
 αβ 
+ qγ (x − ·)(∂r u j )e j − (∂r Eγα )(x − ·) a jk (λ)(∂k uβ )e j − πeα . (6.2.75)

From (6.2.75) and (6.2.1)-(6.2.2) we see that

 1 n
Fx ∈ Lloc (Ω, L n ) ∩ 𝒞∞ (Ω \ {x}) . (6.2.76)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 699

while from (6.2.75), the second line in (6.2.71) we conclude that

n.t.

Fx exists at σ-a.e. point on ∂nta Ω. (6.2.77)

∂Ω

Moreover, a direct computation based on (6.2.75), (6.1.1), (6.1.3), (6.1.6), (6.2.4)-

(6.2.7), (6.2.9), and the first line in (6.2.71) gives that

divFx = (∂r uγ )(x)δx in D (Ω), (6.2.78)

hence, in particular,

divFx ∈ ℰ (Ω). (6.2.79)

 
Abbreviate K := B x, 12 dist(x, ∂Ω) and note that K is a compact subset of Ω.
Based on (6.2.75), (6.2.1)-(6.2.2), and [68, Lemma 8.3.7] we may estimate
 Ω\K      
Nκ Fx (y) ≤ C Nκ (∇ u) (y) + (Nκ π)(y) · sup |x − z| 1−n
z ∈Γκ (y)\K
 
Nκ (∇
u) (y) + (Nκ π)(y)
≤C , ∀y ∈ ∂Ω, (6.2.80)
|x − y| n−1

for some constant C = C(Ω, n, λ, κ) ∈ (0, ∞). From (6.2.80), the last line in (6.2.71),
and [68, (8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (6.2.81)

Finally, in the case when Ω is an exterior domain, it follows from (6.2.75), (6.2.72),
and (6.2.1)-(6.2.2) that
∫

|x · F(x)| dL n (x) = o(R2 ) as R → ∞,
A μ, R ∩Ω (6.2.82)
where Aμ,R := B(0, μ R) \ B(0, R).

Together, the above properties ensure that the vector field Fx satisfies the hypotheses
of [68, Theorem 1.4.1]. As such, on account of (6.2.78), the Divergence Formula
[68, (1.4.6)] currently gives (6.2.73).
The justification of (6.2.74) uses the same circle of ideas, this time starting with
the vector field defined at L n -a.e. point in Ω by
 αβ 
G x := (1 + λ)q j (x − ·)(∂j uk )ek − qα (x − ·) a jk (λ)(∂k uβ )e j − πeα , (6.2.83)

where x ∈ Ω is an arbitrary fixed point. 

Our next major result is Theorem 6.2.4, stated a little later, which amounts to
a brand of Calderón-Zygmund theory customized to fit the specific format of the
boundary layer operators naturally associated with Stokes system. As a preamble, in
700 6 Green Formulas and Layer Potential Operators for the Stokes System

the lemma below we study how derivatives applied to the Stokes double layer (both
for the velocity and the pressure), acting on boundary Sobolev functions, may be
absorbed under the integral sign as weak tangential derivatives.

Lemma 6.2.3 Let Ω ⊆ Rn be an open set with an upper Ahlfors regular boundary
and abbreviate σ∗ := H n−1 ∂∗ Ω. Also, fix λ ∈ C. Then for each f = ( f j )1≤ j ≤n be-
  n
longing to the weighted boundary Sobolev space L11 ∂∗ Ω, 1+σ|x∗ (x)| n−1
(cf. (A.0.131))
and each index r ∈ {1, . . . , n} one has
∫
    
∂r Dλ f(x) = (∂k E jγ )(x − y) ∂τ f j (y) + λ(∂j Ekγ )(x − y) ∂τ f j (y)
rk rk
∂∗ Ω

  
+ qγ (x − y) ∂τ j r f j (y) dσ∗ (y) (6.2.84)
1≤γ ≤n

at every point x ∈ Ω. In addition, for each function

  n
f = ( f )1≤ ≤n ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n such that
 σ(x) 
(6.2.85)
∂τ j k f ∈ L 1 ∂∗ Ω , for all j, k,  ∈ {1, . . . , n}
1 + |x| n−1
one has
∫
 
Pλ f(x) = (1 + λ) q j (x − y) ∂τk j fk (y) dσ∗ (y) (6.2.86)
∂∗ Ω

at every point x ∈ Ω.

Proof Fix r, γ ∈ {1, . . . , n} along with x ∈ Ω arbitrary. Given f as in the statement,

from (6.2.17) we see that
∫
  
∂r Dλ f γ (x) = − νk (y)(∂r ∂k E jγ )(x − y) − λνk (y)(∂r ∂j Ekγ )(x − y)
∂∗ Ω

+ ν j (y)(∂r qγ )(x − y) f j (y) dσ∗ (y). (6.2.87)

Write
 
−νk (y)(∂r ∂k E jγ )(x − y) = νk (y)∂yr (∂k E jγ )(x − y) (6.2.88)
   
= ∂τk r (y) (∂k E jγ )(x − y) + νr (y)∂yk (∂k E jγ )(x − y)
 
= ∂τk r (y) (∂k E jγ )(x − y) − νr (y)(ΔE jγ )(x − y)
 
= ∂τk r (y) (∂k E jγ )(x − y) − νr (y)(∂j qγ )(x − y),

and
6.2 Boundary Layer Potential Operators for the Stokes System . . . 701

−λνk (y)(∂r ∂j Ekγ )(x − y)

 
= λνk (y)∂yr (∂j Ekγ )(x − y)
   
= λ∂τk r (y) (∂j Ekγ )(x − y) + λνr (y)∂yk (∂j Ekγ )(x − y)
   
= λ∂τk r (y) (∂j Ekγ )(x − y) + λνr (y)∂y j (∂k Ekγ )(x − y)
 
= λ∂τk r (y) (∂j Ekγ )(x − y) . (6.2.89)

Also, write
 
ν j (y)(∂r qγ )(x − y) = −ν j (y)∂yr qγ (x − y)
 
= ∂τr j (y) [qγ (x − y) − νr (y)∂y j [qγ (x − y)

= ∂τr j (y) [qγ (x − y) + νr (y)(∂j qγ )(x − y) (6.2.90)

and observe that the last term above cancels the last term in (6.2.88). With this in
mind, we conclude from (6.2.87)-(6.2.90) that
∫ 
   

∂r Dλ f γ (x) = ∂τk r (y) (∂k E jγ )(x − y) f j (y)
∂∗ Ω
 
+ λ∂τk r (y) (∂j Ekγ )(x − y) f j (y)
  
+ ∂τr j (y) qγ (x − y) f j (y) dσ∗ (y). (6.2.91)

On account of this and [69, Lemma 11.1.7] (whose applicability, with ϕ taken to be
one of the components in (∇E)(x − ·) or q(x  − ·), is ensured by (6.2.1)-(6.2.2)) we
therefore obtain
∫
      
∂r Dλ f γ (x) = (∂k E jγ )(x − y) ∂τr k f j (y) + λ(∂j Ekγ )(x − y) ∂τr k f j (y)
∂∗ Ω
  
+ qγ (x − y) ∂τ j r f j (y) dσ∗ (y). (6.2.92)

This proves (6.2.84). As regards (6.2.86), fix an arbitrary vector-valued function f

as in (6.2.85). Based on (6.2.21) we may write
702 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
Pλ f(x) = −(1 + λ) ν j (y)(∂j qk )(x − y) fk (y) dσ∗ (y)
∂∗ Ω
∫
= −(1 + λ) ν j (y)(∂k q j )(x − y) fk (y) dσ∗ (y)
∂∗ Ω
∫
 
= (1 + λ) ν j (y)∂yk q j (x − y) fk (y) dσ∗ (y)
∂∗ Ω
∫
 
= (1 + λ) ∂τ j k (y) q j (x − y) fk (y) dσ∗ (y), (6.2.93)
∂∗ Ω

from which (6.2.86) follows by once again appealing to [69, Lemma 11.1.7] (with ϕ
 − ·)).
now taken to be one of the components of q(x 

We shall also need to consider the principal-value version of the Stokes double
layer operator. To be specific, suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a set
of locally finite perimeter. Much as before, abbreviate σ := H n−1 ∂Ω, and denote
by ν = (ν j )1≤ j ≤n the geometric measure theoretic outward unit normal to Ω. Also,
fix some λ ∈ C. In this setting, define the action of the boundary-to-boundary
(or principal-value) double layer potential operator for the Stokes
system on each function
 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω, (6.2.94)
1 + |x| n−1
at σ-a.e. x ∈ ∂∗ Ω according to
 ∫

Kλ f(x) := lim − νk (y)(∂k E jγ )(x − y) − λνk (y)(∂j Ekγ )(x − y)
ε→0+
y ∈∂∗ Ω
|x−y |>ε 

+ ν j (y)qγ (x − y) f j (y) dσ(y) . (6.2.95)
1≤γ ≤n

Then [68, Proposition 5.6.7] ensures that this limit exists and Kλ f is a σ-measurable
function on ∂∗ Ω. Furthermore, the last result in [68, Proposition 5.6.7] ensures that
if Ω is a Lebesgue measurable set whose topological boundary ∂Ω is countably
rectifiable (of dimension n − 1) and has locally finite H n−1 measure (hence, in
particular, if ∂Ω is a UR set), then for each function f as in (6.2.94) the limit in
(6.2.95) actually exists for σ-a.e. x ∈ ∂Ω and gives rise to a σ-measurable Cn -valued
function on ∂Ω. Let us also note here that, for each vector-valued function f as in
(6.2.94), at σ-a.e. x ∈ ∂∗ Ω we may write
6.2 Boundary Layer Potential Operators for the Stokes System . . . 703

Kλ f(x) (6.2.96)
 ∫
δ jγ x − y, ν(y)
= lim (λ − 1)
ε→0+ 2ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

n x − y, ν(y)(x j − y j )(xγ − yγ )
− (λ + 1)
2ωn−1 |x − y| n+2
! 
1 ν j (y)(xγ − yγ ) − νγ (y)(x j − y j )
+ (λ − 1) f j (y) dσ(y) .
2ωn−1 |x − y| n
1≤γ ≤n

Assume next that Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a Lebesgue measurable

set whose topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and
has locally finite H n−1 measure (this is the case if, e.g., ∂Ω is a UR set). Once again
abbreviate σ := H n−1 ∂Ω and denote by ν = (ν j )1≤ j ≤n the geometric measure
theoretic outward unit normal to Ω. In this context, let us define the action of the
transpose double layer potential operator for the Stokes system on each
vector-valued function
 σ(x)  n
f = ( fγ )1≤γ ≤n ∈ L 1 ∂Ω, (6.2.97)
1 + |x| n−1
according to
 ∫

Kλ# f(x) := lim+ νk (x)(∂k E jγ )(x − y) + λνk (x)(∂j Ekγ )(x − y)
ε→0
y ∈∂Ω
|x−y |>ε 

− ν j (x)qγ (x − y) fγ (y) dσ(y) (6.2.98)
1≤ j ≤n

at σ-a.e. point x ∈ ∂∗ Ω. From [68, (5.6.23)] and [68, Corollary 5.3.6] we know
that this definition is indeed meaningful in the present geometric context. In a more
explicit fashion, for each vector-valued function f as in (6.2.97), at σ-a.e. x ∈ ∂∗ Ω
we may write
704 6 Green Formulas and Layer Potential Operators for the Stokes System

Kλ# f(x) (6.2.99)

 ∫
δ jγ x − y, ν(x)
= lim − (λ − 1)
ε→0+ 2ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

n x − y, ν(x)(x j − y j )(xγ − yγ )
+ (λ + 1)
2ωn−1 |x − y| n+2
! 
1 ν j (x)(xγ − yγ ) − νγ (x)(x j − y j )
− (λ − 1) fγ (y) dσ(y) .
2ωn−1 |x − y| n
1≤ j ≤n

Here is the version of Calderón-Zygmund theory designed to accommodate the

specific format of the boundary layer operators naturally associated with Stokes
system, advertised earlier.

Theorem 6.2.4 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set with the

property that ∂Ω is a UR set. Abbreviate σ∗ := H n−1 ∂∗ Ω and σ := H n−1 ∂Ω,
and denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal
to Ω. Also, fix some λ ∈ C and consider the boundary layer potential operators 𝒮,
Q, Dλ , Pλ , Kλ , Kλ# associated with Ω as in (6.2.12), (6.2.14), (6.2.17), (6.2.21),
(6.2.95), (6.2.98). Then the following properties hold.

(i) For each p ∈ [1, ∞) and κ > 0 there exists  n a finite constant C > 0 with the
property that for every f ∈ L p (∂∗ Ω, σ∗ ) one has
 
Nκ (Dλ f) ≤ C f[L p (∂∗ Ω,σ∗ )]n if 1 < p < ∞, (6.2.100)
L p (∂Ω,σ)
 
Nκ (Dλ f) ≤ C f[L 1 (∂∗ Ω,σ∗ )]n if p = 1, (6.2.101)
L 1,∞ (∂Ω,σ)
 n
and for every f ∈ L p (∂Ω, σ) one has
 
Nκ (Q f) ≤ C f[L p (∂Ω,σ)]n if 1 < p < ∞, (6.2.102)
L p (∂Ω,σ)
 
Nκ (Q f) ≤ C f[L 1 (∂Ω,σ)]n if p = 1. (6.2.103)
L 1,∞ (∂Ω,σ)

(ii) For each function f belonging to the weighted boundary Sobolev space
 1  n
L1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
(cf. (A.0.131)), aperture parameter κ ∈ (0, ∞), and in-
dex r ∈ {1, . . . , n}, the pointwise nontangential boundary traces
κ−n.t.   κ−n.t.
Pλ f ∂Ω , ∂r Dλ f ∂Ω
exist (in Cn ) at σ∗ -a.e. point on ∂∗ Ω. (6.2.104)

Also, for each aperture parameter κ ∈ (0, ∞) and exponents p, q ∈ [1, ∞) there
exists some constant C ∈ (0, ∞), depending
n only on ∂Ω, λ, n, κ, p, q, such that
every function f ∈ L1 (∂∗ Ω, σ∗ ) one has
p,q
6.2 Boundary Layer Potential Operators for the Stokes System . . . 705
     
Nκ (Dλ f) p +Nκ (∇Dλ f) L q (∂Ω,σ) + Nκ (Pλ f) L q (∂Ω,σ)
L (∂Ω,σ)

≤ C f [L p, q (∂∗ Ω,σ∗ )] M if p, q > 1, (6.2.105)

1

plus similar estimates in the case when either p = 1 or q = 1, this time with the
corresponding L 1 -norm in the left side replaced by the weak-L 1 (quasi-)norm.
(iii) Fix p, p ∈ (1, ∞) with 1/p + 1/p = 1. Then the operators
 n  n
Kλ : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ) ,
 n  n (6.2.106)
Kλ# : L p (∂Ω, σ) −→ L p (∂∗ Ω, σ∗ ) ,

are well-defined, linear, and bounded. Moreover, corresponding to p = 1, the

operators in (6.2.95)-(6.2.98) induce well-defined, linear, and bounded map-
pings
 n  n
Kλ : L 1 (∂∗ Ω, σ∗ ) −→ L 1,∞ (∂Ω, σ) ,
 n  n (6.2.107)
Kλ# : L 1 (∂Ω, σ) −→ L 1,∞ (∂∗ Ω, σ∗ ) .

In addition, the transpose of K # in (6.2.106) is

 n  n
Kλ : L p (∂∗ Ω, σ∗ ) −→ L p (∂∗ Ω, σ∗ ) . (6.2.108)
  n
(iv) Having fixed some κ ∈ (0, ∞), for each f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
the following
nontangential boundary trace formula (aka jump-formula) holds:
 κ−n.t.    
Dλ f (x) = 1
2I + Kλ f (x) at σ∗ -a.e. point x ∈ ∂∗ Ω. (6.2.109)
∂Ω
 n
In particular, (6.2.109) holds for each f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞).
Moreover, as a consequence of (6.2.19) and (6.2.109), it follows that

if Ω has a compact boundary, then for each c ∈ Cn one has

+ 12 c at σ-a.e. point on ∂∗ Ω, if Ω is bounded, (6.2.110)
Kλ c =
− 12 c at σ-a.e. point on ∂∗ Ω, if Ω is unbounded.
  n
(v) For each aperture parameter κ ∈ (0, ∞) and each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, at
σ-a.e. point x ∈ ∂∗ Ω one has the jump-formula
 κ−n.t.  ∫
1
Q f (x) = ν(x), f(x) + lim+  − y), f(y) dσ(y). (6.2.111)
q(x
∂Ω 2 ε→0
y ∈∂Ω
|x−y |>ε

 n
In particular, (6.2.111) holds for each f ∈ L p (∂Ω, σ) with p ∈ [1, ∞).
 706 6 Green Formulas and Layer Potential Operators for the Stokes System

(vi) The single layer potential operator 𝒮 for the Stokes system, defined in (6.2.12),
along with its boundary-to-boundary version
∫
S f(x) := E(x − y) f(y) dσ(y) for x ∈ ∂Ω, (6.2.112)
∂Ω
 
(where E = E jk 1≤ j,k ≤n is the Kelvin matrix-valued fundamental solution
for the Stokes system in Rn recalled in (6.2.1)) satisfy similar properties as
those of the single layers for generic weakly elliptic, homogeneous, constant
(complex) coefficient, second-order M × M systems described in items (ix)-(xii)
of Theorem 1.5.1, as well as Theorem 2.2.3 and Theorem 2.2.6. This time, the
jump-formula (1.5.59) should be interpreted as
   
∂νλ 𝒮 f, Q f = − 12 I + Kλ# f at σ-a.e. point on ∂∗ Ω, (6.2.113)
  n
if n ≥ 3 and f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−2
(hence, in particular, whenever we have
 p n
f ∈ L (∂Ω, σ) with p ∈ [1, n − 1)).
(vii) Strengthen the original hypotheses by assuming that Ω is actually a UR domain.
Then the operator
 p,q n  p,q n
Kλ : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (6.2.114)

is well defined, linear, and bounded for each p, q ∈ (1, ∞). In particular, the
operator
 p n  p n
Kλ : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (6.2.115)

is well defined, linear, and bounded for each p ∈ (1, ∞).

(viii) Retain the assumption that Ω is a UR domain. Then for each p, q ∈ (1, ∞) it
follows that Kλ# in (6.2.106) extends uniquely to a linear and bounded operator
 p n
from the negative boundary Sobolev space L−1 (∂Ω, σ) into itself and, more
 p,q n
generally, from the off-diagonal negative Sobolev space L−1 (∂Ω, σ) into
itself. Furthermore, if one adopts the same notation Kλ# for said extensions, and
if p , q ∈ (1, ∞) are such that 1/p + 1/p = 1 and 1/q + 1/q = 1, then the
transpose of (6.2.114) is
 p ,q n  pq n
Kλ# : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) , (6.2.116)

while the transpose of (6.2.115) is

 p n  p n
Kλ# : L−1 (∂Ω, σ) −→ L−1 (∂Ω, σ) . (6.2.117)

(ix) Continue to assume that Ω is actually a UR domain. Then given any vector-
  n
valued function f = ( f j )1≤ j ≤n in the Sobolev space L11 ∂Ω, 1+σ(x)
|x | n−1
(cf.
6.2 Boundary Layer Potential Operators for the Stokes System . . . 707

(A.0.131)), for each α ∈ {1, . . . , n} one has

  
∂νλ Dλ f, Pλ f (x) (6.2.118)
α
∫   
= lim+ ν (x)(∂k E jα )(x − y) ∂τ k f j (y)
ε→0
y ∈∂Ω
|x−y |>ε
 
+ λν (x)(∂j Ekα )(x − y) ∂τ k f j (y)
 
+ ν (x)qα (x − y) ∂τ j  f j (y)
 
+ λν (x)(∂k E j )(x − y) ∂ταk f j (y)
 
+ λ2 ν (x)(∂j Ek )(x − y) ∂ταk f j (y)
 
+ λν (x)q (x − y) ∂τ j α f j (y)
  
− (1 + λ)να (x)q j (x − y) ∂τk j fk (y) dσ(y)
 
at σ-a.e. point x ∈ ∂Ω, where the conormal derivative ∂νλ Dλ f, Pλ f is defined
as in (6.1.6). As a corollary of [69, (11.7.22)], this is true whenever we have
f ∈ L1 (∂Ω, σ) with p, q ∈ [1, ∞).
p,q n

 exponents p, q ∈ (1, ∞) the conormal

Hence, in such a setting, for any given
derivative (6.1.6) of the pair Dλ, Pλ induces a well-defined, linear, and bound-
ed operator
   p,q n  n
∂νλ Dλ, Pλ : L1 (∂Ω, σ) −→ L q (∂Ω, σ) . (6.2.119)
 
Finally, (6.2.118) implies that ∂νλ Dλ f, Pλ f does not jump across the boundary
(in the sense that it has the same nontangential boundary trace when considered
from Ω+ := Ω and Ω− := Rn \ Ω).
(x) Continue to assume that Ω is actually a UR domain. Also, pick exponents
p, p , q, q ∈ (1, ∞) satisfying 1/p + 1/p = 1 and 1/q + 1/q = 1. Then for all
 p,q n  q ,p n
functions f ∈ L1 (∂Ω, σ) and g ∈ L1 (∂Ω, σ) one has
∫ ∫
   
∂νλ Dλ f, Pλ f , g dσ = f, Dλ g, Pλ g dσ. (6.2.120)
∂Ω ∂Ω

As a consequence, whenever Ω is a UR domain, the operator (6.2.119) has a

unique extension to a well-defined, linear, and bounded mapping
   n  q, p n
∂νλ Dλ, Pλ : L p (∂Ω, σ) −→ L−1 (∂Ω, σ) , (6.2.121)

namely the (real) transpose of the operator

708 6 Green Formulas and Layer Potential Operators for the Stokes System
   q ,p n  n
∂νλ Dλ, Pλ : L1 (∂Ω, σ) −→ L p (∂Ω, σ) . (6.2.122)

(xi) Make the assumption that Ω is a UR domain. Then the following operator
identities hold:
1   1  λ
 
2 I + Kλ ◦ − 2 I + Kλ = S ◦ ∂ν Dλ, Pλ
 p,q n
on L1 (∂Ω, σ) with p ∈ (1, ∞) and q ∈ (1, n − 1), (6.2.123)
 p n  n−1 
as well as on L (∂Ω, σ) with p ∈ n−2 , ∞ ,

1     
2I+ Kλ# ◦ − 12 I + Kλ# = ∂νλ Dλ, Pλ ◦ S
 n
on L p (∂Ω, σ) with p ∈ (1, n − 1), as well as (6.2.124)
 p, p∗ n
on L−1 (∂Ω, σ) with p ∈ (1, n − 1) and p1∗ = p1 − n−1 ,
1

S ◦ Kλ# = Kλ ◦ S
  n
on L p (∂Ω, σ) with p ∈ (1, n − 1), as well as (6.2.125)
 p, p∗ n
on L−1 (∂Ω, σ) with p ∈ (1, n − 1) and p1∗ = p1 − n−1 ,
1

   
Kλ# ◦ ∂νλ Dλ, Pλ = ∂νλ Dλ, Pλ ◦ Kλ
 p,q n
on L1 (∂Ω, σ) with p ∈ (1, ∞) and q ∈ (1, n − 1), (6.2.126)
 n
as well as on L p (∂Ω, σ) with p ∈ (1, ∞).

Moreover, if ∂Ω is bounded then one may allow any p, q ∈ (1, ∞), this time
taking p∗ ∈ (1, ∞) arbitrary (and unrelated to p).
(xii) Under the assumption that Ω is a UR domain, similar results to those presented
in Theorem 3.3.1, Theorem 3.3.2, and Theorem 3.3.3 are valid for the boundary
layer operators associated with Stokes system considered earlier in this section
acting on Morrey spaces and their pre-duals.

Proof The nontangential maximal function estimates in item (i) are consequences
of (6.2.17), (6.2.14), (6.2.1)-(6.2.2), and [70, Theorem 2.4.1]. Likewise, all claims
in item (ii) may be justified based on Lemma 6.2.3, (6.2.1)-(6.2.2), and [70, The-
orem 2.4.1]. Next, the claims in item (iii) are readily implied by (6.2.95)-(6.2.98),
(6.2.1)-(6.2.2), [70, Theorem 2.3.2], and [70, (2.3.25)].
To prove the jump-formula from item (iv), we first introduce some notation.
Specifically, for each given , j, k ∈ {1, . . . , n} and each complex-valued function
 σ(x) 
g ∈ L 1 ∂Ω, (6.2.127)
1 + |x| n−1
6.2 Boundary Layer Potential Operators for the Stokes System . . . 709

define
∫
(T , j,k g)(x) := (∂ E jk )(x − y)g(y) dσ(y) for each x ∈ Ω, (6.2.128)
∂Ω
∫
(Q j g)(x) := q j (x − y)g(y) dσ(y) for each x ∈ Ω, (6.2.129)
∂Ω

along with their principal-value versions

∫
(T , j,k g)(x) := lim+ (∂ E jk )(x − y)g(y) dσ(y) for σ-a.e. x ∈ ∂Ω,
ε→0
y ∈∂Ω
|x−y |>ε
(6.2.130)
∫
(Q j g)(x) := lim+ q j (x − y)g(y) dσ(y) for σ-a.e. x ∈ ∂Ω. (6.2.131)
ε→0
y ∈∂Ω
|x−y |>ε

Also, fix an aperture parameter κ > 0. Then from the jump-formula recorded in [70,
(2.5.4)] and (6.2.8) we conclude that for σ-a.e. point x ∈ ∂∗ Ω we have

  κ−n.t.
1  
T , j,k g (x) = ν (x) ν j (x)νk (x) − δ jk g(x) + (T , j,k g)(x), (6.2.132)
∂Ω 2
  κ−n.t.
1
Qjg (x) = ν j (x)g(x) + (Q j g)(x). (6.2.133)
∂Ω 2
In turn, from (6.2.17) and (6.2.130)-(6.2.133) we deduce that for each given vector-
  n
valued function f = ( f j )1≤ j ≤n belonging to the space L 1 ∂∗ Ω, 1+σ(x)
|x | n−1 and for
each index γ ∈ {1, . . . , n} we have

  κ−n.t.
1    
Dλ f γ (x) = − νk (x)νk (x) ν j (x)νγ (x) − δ jγ f j (x) − Tk, j,γ (νk f j ) (x)
∂Ω 2
λ    
− νk (x)ν j (x) νk (x)νγ (x) − δkγ f j (x) − λ Tk, j,γ (νk f j ) (x)
2
1  
+ ν j (x)νγ (x) f j (x) + Qγ (ν j f j ) (x)
2
1  
= fγ (x) + Kλ f γ (x) at σ-a.e. point x ∈ ∂∗ Ω. (6.2.134)
2
On account of the arbitrariness of γ, this proves the jump-formula (6.2.109). The
jump-formula claimed in item (v) is seen directly from (6.2.133) and (6.2.14).
Going further, the first claim in item (vi) is a consequence of the fact that the
integral kernel of the single layer potential operators for the Stokes system have
the same analytical properties as in the case of the single layers for generic weakly
710 6 Green Formulas and Layer Potential Operators for the Stokes System

elliptic, homogeneous, constant (complex) coefficient, second-order M × M systems

treated earlier in Theorem 1.5.1.
  n
As regards the jump-formula (6.2.113), if n ≥ 3 and f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−2
we
may write
  
∂νλ 𝒮 f, Q f (x)
α
  κ−n.t.   κ−n.t. 
αβ
= ν j (x)a jk (λ) ∂k 𝒮 f β ∂Ω
(x) − να (x) Q f ∂Ω (x)
  κ−n.t.    κ−n.t. 
= ν j (x) ∂j 𝒮 f α ∂Ω
(x) + λν j (x) ∂α 𝒮 f j ∂Ω
(x)
 κ−n.t. 
− να (x) Q f ∂Ω (x)
1  
= ν j (x)ν j (x) να (x)νk (x) − δαk fk (x) + ν j (x)(Tj,α,k fk )(x)
2
λ  
+ ν j (x)να (x) ν j (x)νk (x) − δ jk fk (x) + λν j (x)(Tα, j,k fk )(x)
2
1
− να (x)ν j (x) f j (x) − να (x)(Q j f j )(x)
2
1  
= − fα (x) + Kλ# f α (x) at σ-a.e. point x ∈ ∂∗ Ω, (6.2.135)
2
thanks to (6.1.6), (6.1.1), (6.2.132)-(6.2.133), and (6.2.98).
Henceforth strengthen the hypotheses on the underlying set by assuming that Ω
is actually a UR domain. The claim that for each p, q ∈ (1, ∞) the operator Kλ is
well defined and bounded in the context of (6.2.114) is a consequence of (6.2.109),
(6.2.105), and [69, Proposition 11.3.2]. This takes care of item (vii). Next, the claims
in the current item (viii) may be dealt with based on what we have proved so far,
reasoning as in the treatment of item (vi) in Theorem 1.5.1.
Consider next the claims in item (ix). To get started, fix a vector-valued function
   n
f = ( f j )1≤ j ≤n ∈ L11 ∂Ω, 1+σ(x)
|x | n−1
, along with some α ∈ {1, . . . , n}. For starters,
from (6.1.6), (6.1.1), we see that at σ-a.e. point on ∂Ω we have
     κ−n.t.   κ−n.t. 
αβ
∂νλ Dλ f, Pλ f = ν j a jk (λ) ∂k Dλ f β ∂Ω − να Pλ f ∂Ω
α
  κ−n.t.    κ−n.t. 
=ν ∂ Dλ f α ∂Ω
+ λν ∂α Dλ f ∂Ω

 κ−n.t. 
− να Pλ f ∂Ω . (6.2.136)

Let us consider the terms above containing nontangential traces separately. First,
based on (6.2.84) and (6.2.132)-(6.2.133), at σ-a.e. point on ∂Ω we may write
6.2 Boundary Layer Potential Operators for the Stokes System . . . 711
  κ−n.t.  1     
ν ∂ Dλ f α ∂Ω
= ν νk ν j να − δ jα ∂τ k f j + ν Tk, j,α ∂τ k f j
2
λ     
+ ν ν j νk να − δkα ∂τ k f j + λν Tj,k,α ∂τ k f j
2
1    
+ ν να ∂τ j  f j + ν Q α ∂τ j  f j , (6.2.137)
2
and
  κ−n.t.  λ     
λν ∂α Dλ f ∂Ω
= ν νk ν j ν − δ j ∂ταk f j + λν Tk, j, ∂ταk f j
2
λ2     
+ ν ν j νk ν − δk ∂ταk f j + λ2 ν Tj,k, ∂ταk f j
2
λ    
+ ν ν ∂τ j α f j + λν Q ∂τ j α f j . (6.2.138)
2
In addition, (6.2.86) and (6.2.133) imply that at σ-a.e. point on ∂Ω we have
 κ−n.t.  1+λ    
Pλ f ∂Ω = ν j ∂τk j fk + (1 + λ)Q j ∂τk j fk , (6.2.139)
2
hence
 κ−n.t.  1+λ    
−να Pλ f ∂Ω = − να ν j ∂τk j fk − (1 + λ)να Q j ∂τk j fk . (6.2.140)
2
By making repeated use of [69, (11.4.3), (11.4.8)] we may express the jump-terms
appearing in (6.2.137) and (6.2.138) as
1   
ν νk ν j να − δ jα ν (∇tan f j )k − νk (∇tan f j )
2
λ   
+ ν ν j νk να − δkα ν (∇tan f j )k − νk (∇tan f j )
2
1  
+ ν να ν j (∇tan f j ) − ν (∇tan f j ) j
2
λ 1
= − ν j (∇tan f j )α − να (∇tan f j ) j (6.2.141)
2 2
and, respectively,
712 6 Green Formulas and Layer Potential Operators for the Stokes System

λ   
ν νk ν j ν − δ j να (∇tan f j )k − νk (∇tan f j )α
2
λ2   
+ ν ν j νk ν − δk να (∇tan f j )k − νk (∇tan f j )α
2
λ  
+ ν ν ν j (∇tan f j )α − να (∇tan f j ) j
2
λ 
= ν j (∇tan f j )α − να (∇tan f j ) j . (6.2.142)
2
Combining (6.2.141) with (6.2.142) yields
λ 1 λ 
− ν j (∇tan f j )α − να (∇tan f j ) j + ν j (∇tan f j )α − να (∇tan f j ) j
2 2 2
1+λ 1+λ
=− να (∇tan f j ) j = − να νk ∂τk j f j
2 2
1+λ
= να ν j ∂τk j fk , (6.2.143)
2
where we have also made use of (A.0.78) and [69, (11.1.24)]. Observe that the last
expression in (6.2.143) cancels the jump-term in (6.2.140). Keeping this in mind,
we may now conclude from (6.2.136)-(6.2.140) that at σ-a.e. point x ∈ ∂Ω we have
  
∂νλ Dλ f, Pλ f (x)
α
       
= ν (x) Tk, j,α ∂τ k f j (x) + λTj,k,α ∂τ k f j (x) + Q α ∂τ j  f j (x)
       
+ λν (x) Tk, j, ∂ταk f j (x) + λTj,k, ∂ταk f j (x) + Q ∂τ j α f j (x)
 
− (1 + λ)να (x)Q j ∂τk j fk (x). (6.2.144)

On account of (6.2.130)-(6.2.131), this proves (6.2.118). Granted this, the fact that
(6.2.119) is a well-defined, linear, and bounded operator becomes a consequence of
(6.2.1)-(6.2.2) and [70, Theorem 2.3.2].
Let us now turn our attention to the claims made in item (x). Assume Ω  Rn ,
else there is nothing to prove. Define Ω+ := Ω and Ω− := Rn \ Ω. Item (7) in [68,
Lemma 5.10.9] then guarantees that Ω± are two UR domains, whose topological and
geometric measure theoretic boundaries agree with those of Ω, and whose geometric
measure theoretic outward unit normals are ±ν at σ-a.e. point on ∂Ω. Next, pick
 p,q n  q ,p n
f ∈ L1 (∂Ω, σ) and g ∈ L1 (∂Ω, σ) with p, p , q, q ∈ (1, ∞) satisfying
1/p + 1/p = 1 and 1/q + 1/q = 1, then define
6.2 Boundary Layer Potential Operators for the Stokes System . . . 713

u± := Dλ f and π± := Pλ f in Ω± ,
(6.2.145)
w± := Dλ g and ρ± := Pλ g in Ω± .

Then our earlier results in the current

 items
 (ii) and (iv), together
 with (6.2.22), imply
that the pairs ( u+, π+ ), (w+, ρ+ ) and ( u−, π− ), (w−, ρ− ) satisfy (6.1.29)-(6.1.30)
(with the roles of p, q interchanged), relative to Ω+ and to Ω− . In addition, in the case
when Ω± is an exterior domain, it is clear from (6.2.17), (6.2.21), and (6.2.1)-(6.2.2)
that the functions u±, π±, w±, ρ± satisfy, for each multi-index α ∈ N0n ,
   
∂ α u± (x) = O(|x| 1−n− |α | ) and ∂ α w± (x) = O(|x| 1−n− |α | ) as |x| → ∞,
π± (x) = O(|x| −n ) and ρ± (x) = O(|x| −n ) as |x| → ∞.
(6.2.146)
In turn, these properties imply that condition (6.1.32) formulated for the exterior
domain Ω± holds in such a scenario. As such, we may invoke Green’s formula
(6.1.31) which, in light of (6.2.109), (6.2.118), and (6.2.22), permits us to write
∫ 
    
∂νλ Dλ f, Pλ f , ± 12 I + Kλ g dσ
∂Ω
∫    
= ± 12 I + Kλ f, ∂νλ Dλ g, Pλ g dσ. (6.2.147)
∂Ω

Subtracting the two versions of this equality yields (6.2.120).

On to the claims in item (xi). As a preamble, it is useful to observe that, in
concert
 pwith (6.2.113),
n formula (6.2.28) written for u := 𝒮 f and π := Q f, where

f ∈ L (∂Ω, σ) with p ∈ [1, n − 1), implies
  
Q 1
2I + Kλ# f = Pλ (S f) in Ω. (6.2.148)

Likewise, writing formula (6.2.28) for the pair u := Dλ f and π := Pλ f, where
 n
f ∈ L p (∂Ω, σ) with p ∈ [1, ∞), and then making use of (6.2.109) yields
     
Q ∂νλ Dλ f, Pλ f = Pλ − 12 I + Kλ f in Ω. (6.2.149)

The operator identities claimed in current item (xi) may now be justified in an
analogous fashion to those in item (xiii) of Theorem 1.5.1, making use of the
integral representation formulas from Theorem 6.2.1, as well as the jump-formulas
for the Stokes layer potential operators established earlier in this proof, and (6.2.148)-
(6.2.149).
Finally, the claim in item (xii) may be justified by reasoning as in the proofs of
Theorems 3.3.1-3.3.3, making use of [70, Theorem 2.6.1], [70, Proposition 2.6.2],
and our earlier results for the boundary layer operators associated with Stokes system
in this section. 
714 6 Green Formulas and Layer Potential Operators for the Stokes System

The machinery developed so far permits us to show that tangential derivatives

commute with any of the principal-value double layer potential operators for the S-
tokes system introduced before at the expense of commutators of Calderón-Zygmund
singular integrals with operators of pointwise multiplication with scalar components
of the geometric measure theoretic outward unit normal. This claim is made precise
in the theorem below.
Theorem 6.2.5 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the property
that ∂Ω is a UR set and  
H n−1 ∂nta Ω \ ∂∗ Ω = 0 (6.2.150)
(this condition is automatically satisfied if the set Ω is a UR domain to begin with).
Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric measure
theoretic outward unit normal to Ω. Next, pick some λ ∈ C and consider the principal-
value double layer potential operator Kλ associated with Ω as in (6.2.95). Finally,
fix an integrability exponent p ∈ (1, ∞). Then for each vector-valued function
  n
f = ( fμ )1≤μ ≤n ∈ L 1 ∂∗ Ω, 1+σ(x)
p
|x | n−1
∩ L loc (∂∗ Ω, σ) such that
 σ(x)  p
∂τ j k fμ ∈ L 1 ∂∗ Ω , ∩ Lloc (∂∗ Ω, σ) for all j, k, μ ∈ {1, . . . , n}
1 + |x| n−1
(6.2.151)
it follows that  p n
Kλ f ∈ L1,loc (∂∗ Ω, σ) (6.2.152)
and for each pair of indices j, k ∈ {1, . . . , n} one has
      
∂τ j k (Kλ f )γ = Kλ ∂τ j k f + Mν j , T ,μ,γ (∂τk  fμ ) − Mνk , T ,μ,γ (∂τ j  fμ )
γ
   
+ λ Mν j , Tμ, ,γ (∂τk  fμ ) − λ Mνk , Tμ, ,γ (∂τ j  fμ )
   
+ Mν j , Qγ (∂τμ, k fμ ) − Mνk , Qγ (∂τμ, j fμ ), (6.2.153)

where Mν is the operator of pointwise multiplication with ν for each  ∈ {1, . . . , n},
and the family of operators T , j,k , Q has been defined in (6.2.130)-(6.2.131).
Proof Pick a vector-valued function f as in (6.2.151) and select an aperture pa-
rameter κ > 0. Recall the boundary-to-domain double layer potential operator Dλ
associated with Ω as in (6.2.17). From Lemma 6.2.3, [70, (2.4.8)] in [70, Theo-
rem 2.4.1], and [70, Theorem 2.5.1] we then see that
   
Nκ Dλ f ∈ Lloc (∂Ω, σ), Nκ ∇Dλ f ∈ Lloc (∂Ω, σ),
p p
(6.2.154)

and (bearing in mind (6.2.150)) the nontangential traces

κ−n.t.
  κ−n.t.

Dλ f and ∂j Dλ f for j ∈ {1, . . . , n}

∂Ω ∂Ω (6.2.155)
exist at σ-a.e. point on ∂nta Ω.
6.2 Boundary Layer Potential Operators for the Stokes System . . . 715

Granted these properties, [69, Proposition 11.3.2] applies and, in light of item (iv)
in Theorem 6.2.4, its first conclusion guarantees that (6.2.152) holds.
To proceed, fix j, k, γ ∈ {1, . . . , n}. Then, thanks to (6.2.109) and [69, Proposi-
tion 11.3.2, (11.3.26)] (whose applicability in the current setting has been justified
above), we may compute
1
∂τ j k (Kλ f )γ = ∂τ j k 2 f + Kλ f )γ − 12 ∂τ j k fγ

  κ−n.t.
  κ−n.t.

= ν j ∂k Dλ f γ − νk ∂j Dλ f γ − 12 ∂τ j k fγ . (6.2.156)
∂Ω ∂Ω

Using notation and results from Lemma 6.2.3 and (6.2.128)-(6.2.133) we may then
compute

  κ−n.t.

ν j ∂k Dλ f γ
∂Ω
!
κ−n.t. κ−n.t.

= νj T ,μ,γ (∂τk  fμ ) + λTμ, ,γ (∂τk  fμ ) + Q γ (∂τμk fμ )

∂Ω ∂Ω

   
= νj 2ν
1
νμ νγ − δμγ ∂τk  fμ + T ,μ,γ (∂τk  fμ )
   
+ λν j 2 νμ
1
ν νγ − δ γ ∂τk  fμ + Tμ, ,γ (∂τk  fμ )

 
+ νj 2 νγ ∂τμk fμ
1
+ Qγ (∂τμ, k fμ ) (6.2.157)

and, similarly,

  κ−n.t.    
νk ∂j Dλ f γ = νk 2ν
1
νμ νγ − δμγ ∂τ j  fμ + T ,μ,γ (∂τ j  fμ )
∂Ω
   
+ λνk 2 νμ
1
ν νγ − δ γ ∂τ j  fμ + Tμ, ,γ (∂τ j  fμ )

 
+ νk 2 νγ ∂τμ j fμ
1
+ Qγ (∂τμ, j fμ ) . (6.2.158)

By also making use of [69, Proposition 11.4.2], we see that the jump-terms in
  κ−n.t.
ν j ∂k Dλ f γ are
∂Ω
716 6 Green Formulas and Layer Potential Operators for the Stokes System

2 νj ν νμ νγ νk (∇tan fμ ) − 12 ν j νμ ν νγ ν (∇tan fγ )k − 12 ν j ν νk (∇tan fγ )

1

+ 12 ν j ν ν (∇tan fγ )k + λ2 ν j νμ ν νγ νk (∇tan fμ ) − λ2 ν j νμ ν νγ ν (∇tan fμ )k

− λ2 ν j νμ νk (∇tan fμ )γ + λ2 ν j νμ νγ (∇tan fμ )k + 12 ν j νγ νμ (∇tan fμ )k

− 12 ν j νγ νk (∇tan fμ )μ . (6.2.159)

Re-write (6.2.159) with j and k interchanged, then subtract it from (6.2.159). After
canceling like terms (in which j, k play symmetric roles) and making repeated use
of [69, Proposition 11.4.2], we arrive at the conclusion that the jump-terms in the
  κ−n.t.   κ−n.t.
expression ν j ∂k Dλ f γ − νk ∂j Dλ f γ simply amount to 12 ∂τ j k fγ . In the
∂Ω ∂Ω
ultimate analysis, this is going to cancel the very last term in (6.2.156). Returning to
(6.2.156) and putting it altogether we therefore obtain

∂τ j k (Kλ f )γ = ν j T ,μ,γ (∂τk  fμ ) + λν j Tμ, ,γ (∂τk  fμ ) + ν j Qγ (∂τμ, k fμ )

− νk T ,μ,γ (∂τ j  fμ ) − λνk Tμ, ,γ (∂τ j  fμ ) − νk Qγ (∂τμ, j fμ ). (6.2.160)

Note that, by once again employing [69, Proposition 11.4.2], we have

 
ν j T ,μ,γ (∂τk  fμ ) = Mν j , T ,μ,γ (∂τk  fμ )
 
+ T ,μ,γ ν j νk (∇tan fμ ) − ν j ν (∇tan fμ )k ,
 
ν j Tμ, ,γ (∂τk  fμ ) = Mν j , Tμ, ,γ (∂τk  fμ )
 
+ Tμ, ,γ ν j νk (∇tan fμ ) − ν j ν (∇tan fμ )k ,
 
νk T ,μ,γ (∂τ j  fμ ) = Mνk , T ,μ,γ (∂τ j  fμ )
 
+ T ,μ,γ νk ν j (∇tan fμ ) − νk ν (∇tan fμ ) j ,
 
νk Tμ, ,γ (∂τ j  fμ ) = Mνk , Tμ, ,γ (∂τ j  fμ )
 
+ Tμ, ,γ νk ν j (∇tan fμ ) − νk ν (∇tan fμ ) j , (6.2.161)

and
   
ν j Qγ (∂τμ, k fμ ) = Mν j , Qγ (∂τμ, k fμ ) + Qγ ν j νμ (∇tan fμ )k − ν j νk (∇tan fμ )μ ,

   
νk Qγ (∂τμ, j fμ ) = Mνk , Qγ (∂τμ, j fμ ) + Qγ νk νμ (∇tan fμ ) j − νk ν j (∇tan fμ )μ .
(6.2.162)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 717

Then (6.2.160)-(6.2.162) and [69, Proposition 11.4.2] yield

∂τ j k (Kλ f )γ (6.2.163)
     
= Mν j , T ,μ,γ (∂τk  fμ ) − Mνk , T ,μ,γ (∂τ j  fμ ) − T ,μ,γ ν ∂τ j k fμ
     
+ λ Mν j , Tμ, ,γ (∂τk  fμ ) − λ Mνk , Tμ, ,γ (∂τ j  fμ ) − λTμ, ,γ ν ∂τ j k fμ
     
+ Mν j , Qγ (∂τμ, k fμ ) − Mνk , Qγ (∂τμ, j fμ ) + Qγ νμ ∂τ j k fμ .

Combining (6.2.163) with (6.2.95) and (6.2.130)-(6.2.131) leads to the conclusion

that (6.2.153) holds. 
 
Recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n for the
Stokes system in Rn from (6.2.1), and the pressure vector q = (q j )1≤ j ≤n from (6.2.2).
Suppose Ω ⊆ Rn is an open set and abbreviate σ := H n−1 ∂Ω. In this setting, define
the modified boundary-to-domain single layer potential operator for the
Stokes system
∫
 
𝒮mod f(x) := E(x − y) − E∗ (−y) f(y) dσ(y) for each x ∈ Ω,
∂Ω (6.2.164)
   n
for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, where E∗ := E · 1R \B(0,1)
n .

  n
This definition implies that for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
the function 𝒮mod f is
 ∞ n
well defined, belongs to the space 𝒞 (Ω) , and for each multi-index α ∈ N0n with
|α| ≥ 1 one has
∫
∂ α (𝒮mod f )(x) = (∂ α E)(x − y) f(y) dσ(y) for each x ∈ Ω. (6.2.165)
∂Ω

  n
Moreover, for each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
we have
 n
𝒮mod f ∈ 𝒞∞ (Ω) , Q f ∈ 𝒞∞ (Ω),
  (6.2.166)
Δ 𝒮mod f − ∇Q f = 0 and div 𝒮mod f = 0 in Ω.

In analogy with (6.2.164), we also consider the following modified version of the
boundary-to-boundary single layer operator for the Stokes system
∫
 

Smod f (x) := E(x − y) − E∗ (−y) f (y) dσ(y) at σ-a.e. x ∈ ∂Ω,
∂Ω (6.2.167)
 1 σ(x)  n

for each f ∈ L ∂Ω, 1+ |x | n−1 , where E∗ := E · 1Rn \B(0,1),

which is meaningfully defined, via an absolutely convergent integral, at σ-a.e. point

in ∂Ω.
718 6 Green Formulas and Layer Potential Operators for the Stokes System

To proceed, we agree to abbreviate, for each ε > 0,

( , j,k)
Θε := (∂ E jk ) · 1Rn \B(0,ε) for every , j, k ∈ {1, . . . , n}, (6.2.168)

as well as
(j)
θ ε := q j · 1Rn \B(0,ε) for every j ∈ {1, . . . , n}. (6.2.169)

Let us now consider an open set Ω ⊆ Rn of locally finite perimeter and define
σ := H n−1 ∂Ω. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward
unit normal to Ω, which is well defined at σ-a.e. point on ∂∗ Ω. Also, pick an arbitrary
λ ∈ C. Use this parameter and the pieces of notation from (6.2.168)-(6.2.169) to
define the action of the modified boundary-to-domain double layer potential
operator for the Stokes system (compare with (6.2.16)-(6.2.17)) on each function
 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω , (6.2.170)
1 + |x| n
by setting at each x ∈ Ω

Dλ f(x)
mod
(6.2.171)
 ∫
  
(k, j,γ)
:= − νk (y) (∂k E jγ )(x − y) − Θ1 (−y)
∂∗ Ω
 (j,k,γ) 
− λνk (y) (∂j Ekγ )(x − y) − Θ1 (−y)

  
(γ)
+ ν j (y) qγ (x − y) − θ 1 (−y) f j (y) dσ(y) .
1≤γ ≤n

We also define the modified boundary-to-boundary double layer potential

operator for the Stokes system (compare with (6.2.94)-(6.2.95)), acting on each
function f = ( f j )1≤ j ≤n as in (6.2.170) according to

Kλ f(x)
mod
(6.2.172)
 ∫ 
 (k, j,γ) (k, j,γ) 
:= lim+ − νk (y) Θε (x − y) − Θ1 (−y)
ε→0
∂∗ Ω
 (j,k,γ) (j,k,γ) 
− λνk (y) Θε (x − y) − Θ1 (−y)

  
(γ) (γ)
+ ν j (y) θ ε (x − y) − θ 1 (−y) f j (y) dσ(y)
1≤γ ≤n

at σ-a.e. x ∈ ∂∗ Ω. It is then apparent from definitions and [68, Proposition 5.6.7]

  n
that Kλ is compatible with Kλ (acting on functions from L 1 ∂∗ Ω, 1+σ(x)
mod
|x | n−1
as
6.2 Boundary Layer Potential Operators for the Stokes System . . . 719

in (6.2.95)) in the sense that

   n
for each function f ∈ L 1 ∂∗ Ω , 1+σ(x)
|x | n−1
the difference
(6.2.173)
c  := K f − Kλ f is a constant (belonging to Cn ) on ∂∗ Ω.
mod
f λ

The immediate goal is to develop a versatile Calderón-Zygmund theory for the

modified boundary layer operators associated with Stokes system as above, in the
spirit of Theorem 6.2.4.

Theorem 6.2.6 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the property

that ∂Ω is an Ahlfors regular set; in particular, Ω is a set of locally finite perimeter.
Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to
Ω and abbreviate σ := H n−1 ∂Ω. Also, fix an aperture parameter κ and a number
λ ∈ C. Then the following properties hold.
mod
(1) The operator Dλ is meaningfully defined, and
 σ(x) 
n  n
for each f ∈ L 1 ∂∗ Ω , one has Dλ f ∈ 𝒞∞ (Ω)
mod
1+ |x | n
(6.2.174)
as well as Δ(Dλ f) − ∇Pλ f = 0 and div Dλ f = 0 in Ω.
mod mod

mod
In addition, the operator Dλ is compatible with Dλ from (6.2.17), in the sense
  n
that for each function f belonging to the smaller space L 1 ∂∗ Ω , 1+σ(x)
|x | n−1
the difference

C f := Dλ f − Dλ f is a constant (belonging to Cn ) in Ω.
mod
(6.2.175)

Consequently,
 σ(x)  n
∇Dλ f = ∇Dλ f in Ω for each f ∈ L 1 ∂∗ Ω ,
mod
. (6.2.176)
1 + |x| n−1
Moreover,
mod
Dλ maps constant (Cn -valued) functions on ∂∗ Ω (6.2.177)
into constant (Cn -valued) functions in the set Ω.
In addition,

for each multi-index α ∈ N0n with |α| > 0 and each

 σ(x)  n (6.2.178)
function f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω , ,
1 + |x| n
at each point x ∈ Ω one may express
720 6 Green Formulas and Layer Potential Operators for the Stokes System
 mod 
∂ α Dλ f (x) (6.2.179)
 ∫

= − νk (y)(∂ α ∂k E jγ )(x − y) − λνk (y)(∂ α ∂j Ekγ )(x − y)
∂∗ Ω


+ ν j (y)(∂ α qγ )(x − y) f j (y) dσ(y) .
1≤γ ≤n

Finally, given any function

 σ(x)  n
f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω , with the property that
1 + |x| n
 (6.2.180)
σ(x) 
∂τr s f j ∈ L 1 ∂∗ Ω , for all r, s, j ∈ {1, . . . , n},
1 + |x| n−1

it follows that for each index r ∈ {1, . . . , n} and each point x ∈ Ω one has
∫
 mod    

∂r D f (x) = (∂k E jγ )(x − y) ∂τ f j (y)
λ rk
∂∗ Ω
 
+ λ(∂j Ekγ )(x − y) ∂τr k f j (y)

  
+ qγ (x − y) ∂τ j r f j (y) dσ(y) (6.2.181)
1≤γ ≤n

as well as
∫
 
Pλ f(x) = (1 + λ) q j (x − y) ∂τk j fk (y) dσ(y). (6.2.182)
∂∗ Ω

(2) For each η ∈ (0, 1) there exists a constant C ∈ (0, ∞) with the property that
  mod  
sup dist (x, ∂Ω)1−η ∇ Dλ f (x) ≤ C f[𝒞. η (∂Ω)]n (6.2.183)
x ∈Ω
. n
for every function f ∈ 𝒞η (∂Ω) . Moreover,

if Ω ⊆ Rn is a uniform domain with the property that ∂Ω is an Ahlfors

. n  .  n
regular set then Dλ : 𝒞η (∂Ω) → 𝒞η Ω
mod
is a well-defined,
linear, bounded operator for any η ∈ (0, 1),
(6.2.184)
whereas
if Ω ⊆ Rn is an NTA domain with an upper Ahlfors regular boundary
. n  .   n
then Dλ : 𝒞ηvan (∂Ω) → 𝒞ηvan Ω is a well-defined, linear, and
mod

bounded operator for each η ∈ (0, 1),

(6.2.185)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 721

where the homogeneous vanishing Hölder spaces intervening above are defined
as in (A.0.48) (with Σ := ∂Ω and Σ := Ω, respectively). Also, for each η ∈ (0, 1)
and each p ∈ (1,
. η∞) there
 n exists some C ∈ (0, ∞) with the property that for each

function f ∈ 𝒞 (∂Ω) one has
 ∫  1/p
1 mod
p
sup ∇Dλ f dist (·, ∂Ω) p−1
dL n
x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,∞)

≤ C f[𝒞. η (∂Ω)]n (6.2.186)

and
 ∫  1/p !
1
f dist(·, ∂Ω) p−1 dL n
mod p
lim sup ∇Dλ
R→0+ x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,R)
 .  n
≤ C dist f, 𝒞ηvan (∂Ω) , (6.2.187)
 . n 
where the distance is measured in the space 𝒞η (∂Ω) ,  · [𝒞. η (∂Ω)]n . As

. η function
a corollary, ifthe  n f actually belongs to the homogeneous vanishing
Hölder space 𝒞van (∂Ω) for some η ∈ (0, 1), then for each p ∈ (1, ∞) one has
 ∫  1/p !
1 mod
p
lim sup ∇Dλ f dist (·, ∂Ω) p−1
dL n
= 0.
R→0+ x ∈∂Ω r n−1+η p B(x,r)∩Ω
r ∈(0,R)
(6.2.188)
(3) Strengthen the original geometric hypotheses by assuming that ∂Ω is actually a
UR set. Also, fix an aperture parameter κ ∈ (0, ∞). Then, as a consequence of
(6.2.181) and [70, Theorem 2.5.1], the nontangential boundary trace
 mod  κ−n.t.
∂ Dλ f ∂Ω
exists (in Cn ) at σ-a.e. point on ∂∗ Ω,
(6.2.189)
for all functions f as in (6.2.180) and all  ∈ {1, . . . , n}.

Another corollary of (6.2.181) and [70, (2.4.8)] is the fact that for each ε > 0
and each p ∈ (1, ∞)
 
Nκε ∇(Dλ f) ∈ Lloc (∂Ω, σ) for each function
mod p

 σ(x)  n
f = ( fα )1≤α ≤n ∈ L 1 ∂∗ Ω, such that
1 + |x| n (6.2.190)
 σ(x)  p
∂τ j k fα ∈ L 1 ∂∗ Ω , ∩ Lloc (∂∗ Ω, σ)
1 + |x| n−1
for all j, k ∈ {1, . . . , n} and all α ∈ {1, . . . , n}.
722 6 Green Formulas and Layer Potential Operators for the Stokes System

In addition, as seen from (6.2.171) and [70, (2.5.32)], for each truncation pa-
rameter ε ∈ (0, ∞) one has

Nκε (Dλ f) ∈ Lloc (∂Ω, σ) for each function

mod p

  n (6.2.191)
f ∈ L 1 ∂∗ Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂∗ Ω, σ) with p ∈ (1, ∞).

Next, given any function f = ( f j )1≤ j ≤n as in (6.2.180) and given any index
α ∈ {1, . . . , n}, at σ-a.e. point x ∈ ∂∗ Ω one has (where the conormal derivative
is considered as in (6.1.6))
  mod 
∂νλ Dλ f, Pλ f (x) (6.2.192)
α
∫   
= lim+ ν (x)(∂k E jα )(x − y) ∂τ k f j (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
 
+ λν (x)(∂j Ekα )(x − y) ∂τ k f j (y)
 
+ ν (x)qα (x − y) ∂τ j  f j (y)
 
+ λν (x)(∂k E j )(x − y) ∂ταk f j (y)
 
+ λ2 ν (x)(∂j Ek )(x − y) ∂ταk f j (y)
 
+ λν (x)q (x − y) ∂τ j α f j (y)
  
− (1 + λ)να (x)q j (x − y) ∂τk j fk (y) dσ(y).

(4) Once more strengthen the original geometric hypotheses by assuming that ∂Ω
is actually a UR set. Then the following jump-formula holds:

 mod 
κ−n.t.
1 
Dλ f f at σ-a.e. point on ∂∗ Ω,
mod
= 2I + Kλ
∂Ω
 σ(x)  n
(6.2.193)
for each given function f ∈ L 1 ∂∗ Ω , ,
1 + |x| n
where, as usual, I is the identity operator. As a consequence of (6.2.193) and
(6.2.177),
mod
if ∂Ω is a UR set, the operator Kλ maps constant (Cn -valued)
(6.2.194)
functions on ∂∗ Ω into constant (Cn -valued) functions on ∂∗ Ω.

Moreover, if Ω is actually a UR domain and if p, p ∈ (1, ∞) are such that

1/p + 1/p = 1, then given any functions
6.2 Boundary Layer Potential Operators for the Stokes System . . . 723
  n
f ∈ Lloc (∂Ω, σ) ∩ L 1 ∂Ω , σ(x)
p
1+ |x | n together with
∫ (6.2.195)
 p n
g ∈ L comp (∂Ω, σ) satisfying g dσ = 0 ∈ Cn,
∂Ω

it follows that
∫ ∫
Kλ f | g | dσ < +∞, | f||Kλ# g | dσ < +∞,
mod

∂Ω ∂Ω
∫ ∫ (6.2.196)
and
mod

Kλ f , g dσ = f, Kλ# g dσ.
∂Ω ∂Ω

(5) Work under the stronger assumption that ∂Ω is a UR set. For each vector-valued
  n
function f belonging to the space L 1 ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular, for
 p n
each vector-valued function f ∈ L (∂Ω, σ) with p ∈ [1, ∞)) the following
jump-formula holds:
   
∂νλ 𝒮mod f, Q f = − 12 I + Kλ# f at σ-a.e. point on ∂∗ Ω, (6.2.197)

where the conormal derivative is defined as in (6.1.6).

(6) Continue to work under the stronger assumption that ∂Ω is a UR set. Then for
each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property that for each
 n  mod  p
f ∈ BMO(∂Ω, σ) the measure ∇ Dλ f dist (·, ∂Ω) p−1 dL n is Carleson
in Ω, in the quantitative sense that
∫
1  mod  p
sup   ∇ Dλ f dist (·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω

≤ C f .
p
(6.2.198)
[BMO(∂Ω,σ)] n

(with the piece of notation introduced in (A.0.19)). Moreover, for each p ∈ (1, ∞)
 n C ∈ (0, ∞) with the property that for each function
there exists a constant
f ∈ BMO(∂Ω, σ) one has
 ∫  p1
1  mod  p
lim+ sup   ∇ Dλ f dist (·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω
r ∈(0,R) B(x,r)∩Ω
 
≤ C dist f, [VMO(∂Ω, σ)]n (6.2.199)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)]n . In

particular,
724 6 Green Formulas and Layer Potential Operators for the Stokes System
 mod 
∇ Dλ f
p
dist (·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
 n
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).
(6.2.200)
Furthermore, for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞), which
depends only on n, p, and the UR  n constants of ∂Ω, with the property that for
each function f ∈ BMO(∂Ω, σ) one has


max lim sup  1 ×
R→∞ x ∈∂Ω, r >0 σ B(x,r)∩∂Ω
B(x,r)⊆R n \B(0,R)
∫  1/p
 mod  p
× ∇ Dλ f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim sup  1 ×
R→0+ x ∈∂Ω, r ∈(0,R) σ B(x,r)∩∂Ω
∫  1/p
 mod  p
× ∇ Dλ f dist (·, ∂Ω) p−1 dL n ,
B(x,r)∩Ω

lim sup  1 ×
R→∞ x ∈∂Ω, r >R σ B(x,r)∩∂Ω

∫ !
 p  1/p
f dist (·, ∂Ω) p−1 dL n
mod
× ∇ Dλ
B(x,r)∩Ω

   n
≤ C dist f, CMO(∂Ω, σ) , (6.2.201)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)]n

(recall that CMO(∂Ω, σ) has been introduced in (A.0.52)). As a consequence
of (6.2.201) and Definition 1.8.1,
 mod 
∇ Dλ f
p
dist (·, ∂Ω) p−1 dL n is a super vanishing Carleson measure in Ω,
 n
for each function f ∈ CMO(∂Ω, σ) and each p ∈ (1, ∞).
(6.2.202)
(7) If ∂Ω is bounded, then all properties listed in items (1)-(4) are valid for the
ordinary double layer operator Dλ , as originally defined in (6.2.17), in place
mod
of its modified version Dλ . In particular, if ∂Ω is a compact UR set then for
each p ∈ (1, ∞)
 there exists a nconstant C ∈ (0, ∞) with the property that for each
function f ∈ BMO(∂Ω, σ) one has
6.2 Boundary Layer Potential Operators for the Stokes System . . . 725
 ∫  p1
1  p
lim+ sup   ∇ Dλ f dist (·, ∂Ω) p−1 dL n
R→0 x ∈∂Ω and σ B(x, r) ∩ ∂Ω
r ∈(0,R) B(x,r)∩Ω
 
≤ C dist f, [VMO(∂Ω, σ)]n (6.2.203)

where the distance in the right-hand side is considered in [BMO(∂Ω, σ)]n . In

particular, if ∂Ω is a compact UR set then
 
∇ Dλ f
p
dist (·, ∂Ω) p−1 dL n is a vanishing Carleson measure in Ω,
 n
for each function f ∈ VMO(∂Ω, σ) and each p ∈ (1, ∞).
(6.2.204)
(8) Strengthen the original assumptions by demanding now that Ω is a UR domain.
Then, having fixed an integrability exponent p ∈ (1, ∞), for each vector-valued
function
  n
f = ( fμ )1≤μ ≤n ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) such that
 σ(x)  p
∂τ j k fμ ∈ L 1 ∂Ω , ∩ Lloc (∂Ω, σ) for all j, k, μ ∈ {1, . . . , n}
1 + |x| n−1
(6.2.205)
it follows that
 p n
Kλ f ∈ L1,loc (∂Ω, σ)
mod
(6.2.206)
and for each pair of indices j, k ∈ {1, . . . , n} one has
 mod 
∂τ j k Kλ f γ (6.2.207)
       
= Kλ ∂τ j k f + Mν j , T ,μ,γ (∂τk  fμ ) − Mνk , T ,μ,γ (∂τ j  fμ )
γ
   
+ λ Mν j , Tμ, ,γ (∂τk  fμ ) − λ Mνk , Tμ, ,γ (∂τ j  fμ )
   
+ Mν j , Qγ (∂τμ, k fμ ) − Mνk , Qγ (∂τμ, j fμ ),

where, for each  ∈ {1, . . . , n}, Mν is the operator of pointwise multiplication
with ν and the family of operators T , j,k , Q has been defined in (6.2.130)-
(6.2.131).
(9) Make the stronger assumption that ∂Ω is a UR set. Fix an integrability exponent
p ∈ (1, ∞). Then the operator
 n .p n
Smod : L p (∂Ω, σ) −→ L1 (∂Ω, σ) (6.2.208)

is well defined, linear, and bounded, when the target space is endowed with the
semi-norm induced by (A.0.128). In addition,
726 6 Green Formulas and Layer Potential Operators for the Stokes System
   n .p " n
Smod : L p (∂Ω, σ) −→ L1 (∂Ω, σ) ∼ defined as
    .p " n  p n (6.2.209)
Smod f := Smod f ∈ L1 (∂Ω, σ) ∼ , ∀ f ∈ L (∂Ω, σ)

is also a well-defined, linear, and bounded operator, when the quotient space is
endowed with the natural semi-norm7 introduced in [69, (11.5.138)]. Further-
more, with 𝒮mod denoting the modified version of the single layer operator for
  n
the Stokes system acting on functions from L 1 ∂Ω , 1+σ(x) |x | n−1
as in (6.2.164),
there exists some constant
 p C =
nC(Ω, n, p, κ) ∈ (0, ∞) with the property that for

each function f ∈ L (∂Ω, σ) and each truncation parameter ε ∈ (0, ∞) one
has:
 n  
𝒮mod f ∈ 𝒞∞ (Ω) , Δ 𝒮mod f − ∇Q f = 0 and div 𝒮mod f = 0 in Ω,
    
Nκ ∇𝒮mod f ∈ L p (∂Ω, σ), Nκ ∇𝒮mod f  L p (∂Ω,σ) ≤ C f[L p (∂Ω,σ)]n ,
    
N κ Q f ∈ L p (∂Ω, σ), N κ Q f  L p (∂Ω,σ) ≤ C f[L p (∂Ω,σ)]n ,
 
Nκε (𝒮mod f) ∈ Lloc (∂Ω, σ) for each q ∈ 0, n−1
q
n−2 ,
κ−n.t.
∇(𝒮mod f) ∂Ω exists at σ-a.e. point on ∂∗ Ω,
   
∂νλ 𝒮mod f, Q f = − 12 I + Kλ# f at σ-a.e. point on ∂∗ Ω,
  κ−n.t. 
and 𝒮mod f (x) = (Smod f )(x) at σ-a.e. point x ∈ Aκ (∂Ω),
∂Ω
in particular (cf. [68, Proposition 8.8.4]), at σ-a.e. point x ∈ ∂∗ Ω.
(6.2.210)

(10) Assume the set Ω is actually a UR domain. Fix some integrability exponent
p ∈ (1, ∞). Then there exists some constant
.p C = C(Ω, n, p, κ) ∈ (0, ∞) with the
property that for each function f ∈ L1 (∂Ω, σ) one has
n

 n
Dλ f ∈ 𝒞∞ (Ω) , Δ(Dλ f) − ∇Pλ f = 0 and div Dλ f = 0 in Ω,
mod mod mod

 mod  κ−n.t.  mod  κ−n.t.

Dλ f ∂Ω and ∇Dλ f ∂Ω exist at σ-a.e. point on ∂Ω,
 mod    mod  
Nκ ∇Dλ f ∈ L p (∂Ω, σ), Nκ ∇Dλ f  L p (∂Ω,σ) ≤ C f[ L. p (∂Ω,σ)]n ,
    
1

N κ Pλ f ∈ L p (∂Ω, σ), N κ Pλ f  L p (∂Ω,σ) ≤ C f[ L. p (∂Ω,σ)]n .

1

.p n (6.2.211)
In fact, for each function f ∈ L1 (∂Ω, σ) one has
κ−n.t. 1 
(Dλ f) ∂Ω = f at σ-a.e. point on ∂Ω,
mod mod
2I + Kλ (6.2.212)

7 recall from [69, Proposition 11.5.14] that said semi-norm is actually a genuine norm if Ω ⊆ R n
is an open set satisfying a two-sided local John condition and whose boundary is an unbounded
Ahlfors regular set
6.2 Boundary Layer Potential Operators for the Stokes System . . . 727
.p M mod
where I is the identity operator on L1 (∂Ω, σ) , and Kλ is the modified
boundary-to-boundary double layer potential operator for the Stokes system
from (6.2.172). Also,
.p n  .   n
if p > n − 1 then Dλ : L1 (∂Ω, σ) → 𝒞η Ω
mod
is a well-
defined, linear, and bounded operator, with η := 1 − p ∈ (0, 1),
n−1

provided either Ω ⊆ Rn is an open set satisfying a two-sided local (6.2.213)

John condition and whose boundary is Ahlfors regular, or Ω is
simultaneously a uniform domain and a UR domain in Rn .

Moreover, formula (6.2.192) holds for any function f = ( f j )1≤ j ≤n belonging

.p n
to the homogeneous boundary Sobolev space L1 (∂Ω, σ) . Furthermore, the
operator
 mod  .p n  n
∂νλ Dλ , Pλ : L1 (∂Ω, σ) −→ L p (∂Ω, σ) defined as
 mod   mod  .p n (6.2.214)
∂νλ Dλ , Pλ f := ∂νλ Dλ f, Pλ f for each f ∈ L1 (∂Ω, σ)

is well defined, linear, and bounded, when the domain space is equipped with
the semi-norm induced by (A.0.128). In addition,
  mod   . p " n  n
∂νλ Dλ , Pλ : L1 (∂Ω, σ) ∼ −→ L p (∂Ω, σ) defined as
 λ  mod   mod  .p n
∂ν Dλ , Pλ [ f] := ∂νλ Dλ f, Pλ f for each f ∈ L1 (∂Ω, σ)
(6.2.215)
is a well-defined, linear, and bounded operator, when the quotient space is
equipped with the natural semi-norm8 introduced in [69, (11.5.138)].
Finally, similar properties to those described in Theorem 2.3.1 and Theo-
rem 2.3.7 hold for the layer potentials
 associated
 n with the Stokes system
 n in
a UR domain Ω ⊆ Rn acting on BMO−1 (∂Ω, σ) and VMO−1 (∂Ω, σ) .
(11) If Ω is a UR domain satisfying a local John condition, then the operator
mod .p p n .p p n
Kλ : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) −→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
(6.2.216)

is well defined, linear, and bounded, when the spaces involved are endowed with
the semi-norm (A.0.128). As a consequence of this and (6.2.194),
 mod   . p p
# n
Kλ : L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼
. # n
p p
−→ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼ (6.2.217)

8 [69, Proposition 11.5.14] tells us that this semi-norm is fact a genuine norm if Ω ⊆ R n is an
open set satisfying a two-sided local John condition and whose boundary is an unbounded Ahlfors
regular set
728 6 Green Formulas and Layer Potential Operators for the Stokes System

defined as
   mod  . # n
[ f] := Kλ f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) ∼
mod p p
Kλ
.p n (6.2.218)
for each function f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ)
p

is also a well-defined linear and bounded operator, when the quotient spaces are
equipped with the natural semi-norm introduced in [69, (11.5.138)]. In addition,

the identity recorded in (6.2.207) is true for each

.p n (6.2.219)
function f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) .
p

(12) Impose the stronger assumption that Ω is an open set in Rn satisfying a two-sided
local John condition and whose boundary is Ahlfors regular9. Then the operator
mod .p n .p n
Kλ : L1 (∂Ω, σ) −→ L1 (∂Ω, σ) (6.2.220)

is well defined, linear, and bounded, when the spaces involved are endowed
with the semi-norm (A.0.128). As a corollary of (6.2.220) and (6.2.194), the
following is a well-defined linear and bounded10 operator:
mod  .p " n .p " n
Kλ : L1 (∂Ω, σ) ∼ −→ L1 (∂Ω, σ) ∼ defined as
 mod   mod   . p " n .p n
Kλ [ f ] := Kλ f ∈ L1 (∂Ω, σ) ∼ , ∀ f ∈ L1 (∂Ω, σ) .
(6.2.221)
Furthermore,

the identity recorded in (6.2.207) holds for each

.p n (6.2.222)
vector-valued function f ∈ L1 (∂Ω, σ) .

(13) Strengthen the original hypotheses by now assuming that Ω is an NTA domain
mod
with an Ahlfors regular boundary11. Then the operator Kλ is still well defined,
linear, bounded in the context of (6.2.220), and (6.2.221) continues to be a
well-defined linear and bounded operator.
(14) Similar properties to those discussed in items (9)-(13) above hold for the modified
layer potential operators for the Stokes system acting on Morrey-based and
block-based Sobolev spaces.

Proof All claims may be justified employing the same ideas as in the proofs of The-
orem 1.8.2, Proposition 1.8.8, Theorem 1.8.9, Theorem 1.8.12, and Theorem 1.8.14,
now making use of the definitions given in (6.2.164), (6.2.167), (6.2.171), (6.2.172)
9 in which scenario, Ω is known to be a UR domain; see [68, (5.11.27)]
10 if ∂Ω is unbounded then the natural semi-norm introduced in [69, (11.5.138)] endowing the
quotient spaces in (6.2.221) is actually a genuine norm and the homogeneous Sobolev spaces of
order one, modulo constants, becomes Banach spaces; see [69, Proposition 11.5.14]
11 a scenario in which Ω is known to be a UR domain; cf. [68, (5.11.5)]
6.2 Boundary Layer Potential Operators for the Stokes System . . . 729

and relying on the formalism associated with the Stokes system developed so far in
this chapter. 

We shall now use the modified layer potential operators for the Stokes system to
prove a basic integral representation result.

Theorem 6.2.7 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the property

that ∂Ω is an Ahlfors regular set; in particular, Ω is a set of locally finite perimeter.
Denote by ν the geometric measure theoretic outward unit normal to Ω and abbreviate
σ := H n−1 ∂Ω. Also, fix an arbitrary number λ ∈ C.
mod
In this setting, recall the modified version of the double layer operator Dλ for
 1 
 n
the Stokes system acting on vector-valued functions from L ∂∗ Ω , 1+σ(x) |x | n as in
(6.2.171), and the modified version of the single layer operator 𝒮mod for the Stokes
  n
system acting on vector-valued functions from L 1 ∂Ω , 1+σ(x) |x | n−1
as in (6.2.164).
Finally, fix an aperture parameter κ ∈ (0, ∞), a truncation parameter ε ∈ (0, ∞),
and consider a pair of functions, u : Ω → Cn and π : Ω → C, satisfying
 n
u ∈ 𝒞∞ (Ω) , π ∈ 𝒞∞ (Ω),
Δu − ∇π = 0 and div u = 0 in Ω,
 σ(x) 
Nκε u ∈ Lloc
1
u), Nκ π ∈ L 1 ∂Ω ,
(∂Ω, σ), Nκ (∇ 1+ |x | n−1
, (6.2.223)
κ−n.t. κ−n.t. κ−n.t.
the traces u ∂Ω , (∇
u) ∂Ω , π ∂Ω
exist σ-a.e. on ∂nta Ω,
κ−n.t.   n
and the function u ∂Ω belongs to the space L 1 ∂∗ Ω , 1+σ(x)
|x | n .

In the case when Ω is an exterior domain make the additional assumption that there
exists μ ∈ (1, ∞) such that
⨏
 
|∇
u | + |π| dL n = o(1) as R → ∞. (6.2.224)
B(0,μR)\B(0,R)

Then the conormal derivative ∂νλ (

u, π), extended to the entire topological bound-
  n
ary by setting it to be zero outside ∂∗ Ω, belongs to L 1 ∂Ω , 1+σ(x)|x | n−1
, and the
κ−n.t.
nontangential trace u ∂Ω considered as a function on ∂∗ Ω (cf. [68, (8.8.52)]) be-
longs to the space
    n    
f ∈ L 1 ∂∗ Ω , 1+σ(x)
|x | n : ∂τ jk
 ∈ L 1 ∂∗ Ω , σ(x)n−1 n, 1 ≤ j, k ≤ n .
f 1+ |x |
(6.2.225)
Furthermore, there exists some Cn -valued locally constant function cu,π  in Ω with
the property that
 κ−n.t.   
− 𝒮mod ∂νλ (
mod
u = Dλ u ∂Ω u, π) + cu,π
 in Ω, (6.2.226)

and
730 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
 κ−n.t. 
π(x) = (1 + λ) q j (x − y)∂τk j uk ∂Ω
(y) dσ(y)
∂∗ Ω
 
− Q ∂νλ (
u, π) (x) at each point x ∈ Ω. (6.2.227)

Proof Work under the assumption that the functions u = (u j )1≤ j ≤n and π are
as in (6.2.223). The current hypotheses imply that Nκε u, Nκε (∇
u) ∈ Lloc
1 (∂Ω, σ),

so we may rely on (6.2.223) and [69, Proposition 11.3.2] to conclude that the
κ−n.t.
function f := u ∂Ω considered on ∂∗ Ω (cf. [68, (8.8.52)]) belongs to the space
 1 n
L1,loc (∂∗ Ω, σ) and satisfies
 κ−n.t.   κ−n.t. 
∂τ j k f = ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω
(6.2.228)
at σ-a.e. point on ∂∗ Ω, for each j, k ∈ {1, . . . , n},

where (ν1, . . . , νn ) are the scalar components of the geometric measure theoretic
outward unit normal ν to Ω. In concert with (6.2.223) this also entails
 σ(x)  n  σ(x)  n
f ∈ L 1 ∂∗ Ω , and ∂τ j k f ∈ L 1 ∂∗ Ω ,
1 + |x| n 1 + |x| n−1 (6.2.229)
for each pair of indices j, k ∈ {1, . . . , n}.

To proceed, abbreviate (cf. (6.1.6))

  κ−n.t.  κ−n.t. 
g := ∂νλ (
u, π) = ∇u + λ(∇
u) ∂Ω
ν − π ∂Ω ν (6.2.230)

at σ-a.e. point on ∂∗ Ω, and note that (6.2.223) together with [68, (8.8.52), (8.9.8),
(8.9.44)] ensure that
 σ(x)  n
g ∈ L 1 ∂∗ Ω , . (6.2.231)
1 + |x| n−1
Going further, extend g to the entire topological boundary by setting it to be zero
outside ∂∗ Ω, and define
 n
w := Dλ f − 𝒮mod g ∈ 𝒞∞ (Ω) .
mod
(6.2.232)

Also, denote by (wγ )1≤γ ≤n the scalar components of w.   Finally,

 recall from (6.2.1)
the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n for the Stokes
system in Rn , and from (6.2.2) the pressure vector q = (q j )1≤ j ≤n . Then for each
index r ∈ {1, . . . , n}, each index γ ∈ {1, . . . , n}, and each point x ∈ Ω we may
compute
6.2 Boundary Layer Potential Operators for the Stokes System . . . 731

(∂r wγ )(x)
 mod   
= ∂r Dλ f γ (x) − ∂r 𝒮mod g γ (x)
∫  
 κ−n.t.   κ−n.t. 
= (∂k E jγ )(x − y) νr (y) (∂k u j ) ∂Ω (y) − νk (y) (∂r u j ) ∂Ω (y) dσ(y)
∂∗ Ω
∫  
 κ−n.t.   κ−n.t. 
+λ (∂j Ekγ )(x − y) νr (y) (∂k u j ) ∂Ω (y) − νk (y) (∂r u j ) ∂Ω (y) dσ(y)
∂∗ Ω
∫  
 κ−n.t.   κ−n.t. 
+ qγ (x − y) ν j (y) (∂r u j ) ∂Ω (y) − νr (y) (∂j u j ) ∂Ω (y) dσ(y)
∂∗ Ω
∫
 
− (∂r Eγα )(x − y) ∂νλ (
u, π) α (y) dσ(y)
∂∗ Ω

= (∂r uγ )(x). (6.2.233)

The first equality above comes from (6.2.232), the second equality uses (6.2.180)-
(6.2.181), (6.2.229), (6.2.165), (6.2.231), the third equality utilizes (6.2.228), and
κ−n.t.
the last equality is provided by (6.2.73) (bearing in mind that (∂j u j ) ∂Ω = 0 since
u is divergence-free). From (6.2.233) we then conclude that ∇ w = ∇ u in Ω, which
shows that the difference cu,π  := 
u − w is a C n -valued locally constant function

in Ω. This concludes the proof of (6.2.226). Finally, (6.2.227) is a consequence of

(6.2.74), the fact that u is divergence-free plus (6.2.228), and the definition made in
(6.2.14). 

We are now prepared to augment the results in item (xi) of Theorem 6.2.4 by prov-
ing some remarkable composition identities involving the modified layer potentials
for the Stokes system.

Theorem 6.2.8 Pick n ∈ N with n ≥ 2 and assume Ω ⊆ Rn is a UR domain. Denote

by ν the geometric measure theoretic outward unit normal to Ω and abbreviate
σ := H n−1 ∂Ω. Select a number λ ∈ C along with an integrability exponent
p ∈ (1, ∞), and recall the modified boundary-to-boundary layer potential operators
 mod 
for the Stokes system: Smod from (6.2.208), ∂νλ Dλ , Pλ from (6.2.214), and Kλ
mod

from item (11) of Theorem 6.2.6 (cf. also (6.2.172)). Finally, let Kλ# be the operator
associated with the parameter λ and the set Ω as in (6.2.98). Then the following
statements are valid.
 n
(1) Given any vector-valued function f ∈ L p (∂Ω, σ) , at σ-a.e. point on ∂Ω one
has
1     mod  
#  #  
2 I + Kλ − 1
2 I + Kλ f = ∂ν
λ
D λ , Pλ Smod f (6.2.234)
732 6 Green Formulas and Layer Potential Operators for the Stokes System

and there exists c f, which is the nontangential trace on ∂Ω of some Cn -valued
locally constant function in Ω, such that
  mod  
Smod Kλ# f = Kλ Smod f + c f at σ-a.e. point on ∂Ω. (6.2.235)

(2) Make the additional hypothesis that Ω satisfies a local John condition, and
mod
recall the operator Kλ from (6.2.216). Then, for each vector-valued function
.p n
f ∈ L1 (∂Ω, σ) ∩ Lloc (∂Ω, σ) , at σ-a.e. point on ∂Ω one has
p

  mod     mod    mod 

Kλ# ∂νλ Dλ , Pλ f = ∂νλ Dλ , Pλ Kλ f (6.2.236)

and there exists c f, which is the nontangential trace on ∂Ω of some Cn -valued
locally constant function in Ω, with the property that at σ-a.e. point on ∂Ω one
has
1 
mod   mod 
   mod  
I + K − 1
I + K  =S
f ∂ν
λ
D , Pλ f + c f. (6.2.237)
2 λ 2 λ mod λ

(3) Strengthen the original geometric assumptions by now asking that Ω ⊆ Rn is an

NTA domain with an upper Ahlfors regular boundary12. Then, with the operator
mod
Kλ now interpreted as in (6.2.220), both formula (6.2.236) as well as formula
.p n
(6.2.237) actually hold for each vector-valued function f ∈ L1 (∂Ω, σ) .
Proof All claims are established by reasoning as in the proof of Theorem 1.8.26,
now making use of the integral representation formulas from Theorem 6.2.7 and the
functional analytic properties of the modified boundary layer potential operators for
the Stokes system from Theorem 6.2.6. 
Shifting perspectives, we shall now define the Stokes weak conormal derivative
a distribution given by a suitable bullet product. For each λ ∈ C, recall the system
Lλ from (6.1.1)-(6.1.2).
Definition 6.2.9 Fix λ ∈ C and let Ω ⊆ Rn be an arbitrary open set. Consider
a vector-valued function u = (uα )1≤α ≤n along with a complex-valued function π
satisfying (with all derivatives taken in the sense of distributions in Ω)
 1 n  1  n×n
u ∈ Lloc (Ω, L n ) , ∇ u ∈ Lbdd (Ω, L n ) ,
 1 n (6.2.238)
π ∈ Lbdd (Ω, L ) and Lλ u − ∇π ∈ Lbdd (Ω, L ) .
1 n n

Introduce the family of vector fields

Fα := ∇uα + λ ∂α u − π eα for each α ∈ {1, . . . , n}. (6.2.239)

In this setting, define the Stokes weak conormal derivative of the pair u, π
as the distribution
12 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 733
.     n
u, π) := ν • Fα 1≤α ≤n ∈ Lipc (∂Ω)
∂νλ ( . (6.2.240)

Since (6.2.238) implies that for each α ∈ {1, . . . , n} we have

 1 n  
Fα ∈ Lbdd (Ω, L n ) and divFα = Lλ u − ∇π α ∈ Lbdd 1
(Ω, L n ), (6.2.241)
    n
from [68, Proposition 4.2.3] we see that ν • Fα 1≤α ≤n ∈ Lipc (∂Ω) . Thus,
Definition 6.2.9 is meaningful.
Our next theorem concerns finer integrability and compatibility properties of the
Stokes weak conormal derivative introduced in Definition 6.2.9.

Theorem 6.2.10 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an open set with an

Ahlfors regular boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν j )1≤ j ≤n
the geometric measure theoretic outward unit normal to Ω. Also, fix some λ ∈ C
and define the system Lλ as in (6.1.1)-(6.1.2). Finally, consider a Cn -valued function
u = (uα )1≤α ≤n along with a complex-valued function π which
 n−1  is L -measurable in
n

Ω satisfying, for some integrability exponents p ∈ n , ∞ , q ∈ (0, ∞], and some

aperture parameter κ > 0,
 1,1  n
u belongs to the local Sobolev space Wloc (Ω) ,
 1 n  
Lλ u − ∇π ∈ Lloc (Ω, L n ) , P Lλ u − ∇π ∈ L p,q (∂Ω, σ), (6.2.242)

Nκ (∇
u) ∈ L p,q (∂Ω, σ) and Nκ π ∈ L p,q (∂Ω, σ).

Then the following statements are true.

(1) There exists a constant C(Ω, λ, κ, p, q) ∈ (0, ∞) with the property that the Stokes
weak conormal derivative of the pair ( u, π), taken in the sense of Definition 6.2.9,
satisfies
.  n
∂νλ (
u, π) belongs to the Lorentz-based Hardy space H p,q (∂Ω, σ) and
 .λ     
∂ (u, π) p, q 
n ≤ C Nκ (∇ u) p, q + C Nκ π  p, q
ν [H (∂Ω,σ)] L (∂Ω,σ) L (∂Ω,σ)
  
+C P Lλ u − ∇π  L p, q (∂Ω,σ) .
(6.2.243)
(2) Whenever
Ω is actually an Ahlfors regular domain and, in addition to
(6.2.242), one assumes that Nκ π, Nκ (∇
u) ∈ Lloc
1 (∂Ω, σ) and
κ−n.t. κ−n.t. (6.2.244)
the nontangential pointwise traces π ∂Ω and (∇
u) ∂Ω exist at
σ-a.e. point on ∂Ω,

it follows that the Stokes weak conormal derivative of the pair ( u, π) agrees with
the distribution associated (as in [68, Proposition 4.1.4]) with the conormal
derivative of (u, π) taken in a pointwise sense as in (6.1.6), i.e.,
734 6 Green Formulas and Layer Potential Operators for the Stokes System
.  κ−n.t. κ−n.t. 
αβ
∂νλ (
u, π) = ν j a jk (λ)(∂k uβ ) ∂Ω − να π ∂Ω
1≤α ≤n
  κ−n.t.  κ−n.t. κ−n.t. 
= ν j (∂j uα ∂Ω
+ λ ν j (∂α u j ∂Ω
− να π ∂Ω
1≤α ≤n
κ−n.t.
 κ−n.t.    n
= ∇
u + λ(∇
u) ν − π ∂Ω ν in Lipc (∂Ω) . (6.2.245)
∂Ω

(3) If, in fact,

Ω is actually an Ahlfors regular domain, the conditions in
(6.2.242) are assumed to hold with p ∈ [1, ∞) and q ∈ (0, p],
κ−n.t.
and one also asks that the nontangential traces π ∂Ω and (6.2.246)
κ−n.t.
u) ∂Ω exist σ-a.e. on ∂Ω,
(∇
.  n
then the Stokes weak conormal derivative ∂νλ ( u, π) ∈ H p,q (∂Ω, σ) from
 p n
(6.2.243) actually belongs to the Lebesgue space L (∂Ω, σ) and one has the
pointwise formula
.   κ−n.t.  κ−n.t. κ−n.t. 
∂νλ (
u, π) = ν j (∂j uα ∂Ω + λ ν j (∂α u j ∂Ω − να π ∂Ω
1≤α ≤n
κ−n.t.
 κ−n.t. 
= ∇
u + λ(∇
u) ν − π ∂Ω ν (6.2.247)
∂Ω

at σ-a.e. point on ∂Ω.

 n−1 
Proof Given u, π as in (6.2.242) for some p ∈ n , ∞ , q ∈ (0, ∞], we have
 1  n×n
∇
u ∈ Lbdd (Ω, L n ) , π ∈ Lbdd
1 (Ω, L n ),

 1 n (6.2.248)
and Lλ u − ∇π ∈ Lbdd (Ω, L n ) ,

thanks to [68, (8.6.51) in Proposition 8.6.3] and [69,

 1 (10.1.4)]. nAlso, each of the
vector fields Fα defined as in (6.2.239) belongs to Lbdd (Ω, L n ) and satisfies
 
Nκ Fα ∈ L p,q (∂Ω, σ), divFα = Lλ u − ∇π α ∈ Lloc
1 (Ω, L n ),

     (6.2.249)
and P divFα = P Lλ u − ∇π α ∈ L p,q (∂Ω, σ).

As such, [69, Theorem 10.2.1] applies and from [69, (10.2.4)-(10.2.5)] we conclude
that (6.2.243) holds. Next, the claim in item (2) is implied by (6.2.240) and the
compatibility property established in [69, Proposition 10.2.9]. In turn, the claim in
item (3) is a consequence of (2) in view of the fact that the present conditions on p, q
entail L p,q (∂Ω, σ) ⊆ L p (∂Ω, σ) ⊆ Lloc
1 (∂Ω, σ); cf. [68, (6.2.25)-(6.2.26)]. 

We augment Theorem 6.2.4 with results pertaining to the action of boundary layer
potential operators associated with the Stokes system on Hardy spaces.
6.2 Boundary Layer Potential Operators for the Stokes System . . . 735

Theorem 6.2.11 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a UR domain and set

σ := H n−1 ∂Ω. Also, fix some λ ∈ C and consider the boundary layer potential
operators 𝒮, Q, Dλ , Pλ , Kλ , Kλ# associated with Ω as in (6.2.12), (6.2.14), (6.2.17),
mod
(6.2.21), (6.2.95), (6.2.98). Also, recall the modified versions Dλ from (6.2.171),
mod
and Kλ from (6.2.172). Then the following properties hold.

(1) The operator Kλ# , originally acting on Lebesgue spaces on ∂Ω (as in (6.2.106)),
extends uniquely to a linear and bounded operator
 n  n  
Kλ# : H p (∂Ω, σ) −→ H p (∂Ω, σ) , p ∈ n−1
n ,1 . (6.2.250)

Moreover, various choices of p yield operators which are compatible with one
another.
 
(2) More generally, for each p ∈ n−1 n , ∞ and q ∈ (0, ∞], the operator Kλ , origi-
#

nally acting on Lebesgue spaces on ∂Ω (as in (6.2.106)), induces a linear and

bounded mapping
 n  n
Kλ# : H p,q (∂Ω, σ) −→ H p,q (∂Ω, σ) (6.2.251)

whose action continues to be compatible with that of Kλ# in (6.2.106).

 
(3) Given p ∈ n−1 n , 1 , there exists C ∈ (0, ∞) with the property that for every
f ∈ [H p (∂Ω, σ)]n one has
 n
𝒮 f ∈ 𝒞 ∞ (Ω) , Q f ∈ 𝒞 ∞ (Ω),
  (6.2.252)
Δ 𝒮 f − ∇Q f = 0 and div 𝒮 f = 0 in Ω,

as well as
   
Nκ (∇𝒮 f) p + Nκ (Q f) L p (∂Ω,σ) ≤ C f[H p (∂Ω,σ)]n . (6.2.253)
L (∂Ω,σ)

 
Also, for every f ∈ [H p (∂Ω, σ)]n with p ∈ n−1 n , 1 the following jump-formula
holds:
.      n
∂νλ 𝒮 f, Q f = − 12 I + Kλ# f in H p (∂Ω, σ) . (6.2.254)
 
(4) For each p ∈ n−1 n , 1 and q ∈ [1, ∞) there exists some finite constant C > 0,
 q, p n
depending only on ∂Ω, λ, n, κ, and p, such that for each f ∈ H1 (∂Ω, σ)
one has
   
Nκ (Pλ f) p
L (∂Ω,σ)
+ Nκ (∇Dλ f) L p (∂Ω,σ) ≤ C f[ H. p (∂Ω,σ)]n . (6.2.255)
1

 n−1   λ

(5) For each p ∈ n , 1 and q ∈ [1, ∞), the operator ∂ν Dλ, Pλ from (6.2.119)
extends to a bounded linear mapping
736 6 Green Formulas and Layer Potential Operators for the Stokes System
.λ    q, p n  n
∂ν Dλ, Pλ : H1 (∂Ω, σ) −→ H p (∂Ω, σ) . (6.2.256)
 
Furthermore, for each p ∈ n−1 n , 1 the following operator is well defined,
linear, and bounded:
.  mod   . p n  n
∂νλ Dλ , Pλ : H1 (∂Ω, σ) −→ H p (∂Ω, σ) defined as
.  mod  .  mod   .p  n (6.2.257)
∂νλ Dλ , Pλ f := ∂νλ Dλ f, Pλ f for each f ∈ H1 (∂Ω, σ) ,

where the weak conormal derivative is considered in the sense of Definition 6.2.9.
As a consequence, the operator
.  mod   . p " n  n
∂νλ Dλ , Pλ : H1 (∂Ω, σ) ∼ → H p (∂Ω, σ) given by
.  mod  .  mod   .p n
∂νλ Dλ , Pλ [ f ] := ∂νλ Dλ f, Pλ f for each f ∈ H1 (∂Ω, σ) ,
(6.2.258)
is well defined, linear, and bounded, when the quotient space is equipped with
the semi-quasinorm13 introduced in (A.0.92).
 
(6) If p ∈ n−1 n , 1 and q ∈ (1, ∞) then the operator Kλ , originally acting on
boundary Sobolev spaces as in (6.2.115), extends to a linear and bounded
mapping
 q, p n  q, p n
Kλ : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) . (6.2.259)

In addition, if Ω ⊆ Rn is an NTA domain  with an upper Ahlfors regular

n , 1 the modified boundary-to-
boundary14, it follows that for each p ∈ n−1
mod
boundary double layer potential operator Kλ induces a linear and bounded
mapping
mod  .p n  .p n
Kλ : H1 (∂Ω, σ) −→ H1 (∂Ω, σ) . (6.2.260)

As a consequence of this and (6.2.194), the operator

 mod   .p " n  .p " n
Kλ : H1 (∂Ω, σ) ∼ −→ H1 (∂Ω, σ) ∼ defined as
 mod   mod   . p " n  .p n
Kλ [ f ] := Kλ f ∈ H1 (∂Ω, σ) ∼ , ∀ f ∈ H1 (∂Ω, σ) ,
(6.2.261)

13 if in fact Ω ⊆ R n is an open set satisfying a two-sided local John condition and whose boundary
is an unbounded Ahlfors regular set, then Proposition 2.3.8 guarantees that said semi-quasinorm
.p "
becomes a genuine quasinorm, making H1 (∂Ω, σ) ∼ a quasi-Banach space
14 in particular, this is the case if Ω is an open set satisfying a two-sided local John condition and
whose boundary is Ahlfors regular; cf. (1.8.157)
6.2 Boundary Layer Potential Operators for the Stokes System . . . 737

is well defined, linear, and bounded, when all quotient spaces are equipped with
the semi-quasinorm15 introduced in (A.0.92).
(7) The boundary-to-boundary version of the single layer potential operator S for
the Stokes system, defined in (6.2.112), satisfies properties analogous to those
for single layers associated with generic weakly elliptic second-order
 systems,
described in Theorem 2.2.6. In addition, if p ∈ n−1 , n − 1 and p∗ ∈ (1, ∞) is
n
such that 1/p∗ = 1/p − 1/(n − 1), then the operator S from (6.2.112) induces a
linear and bounded mapping in the context
 n  p∗, p n
S : H p (∂Ω, σ) −→ H1 (∂Ω, σ) , (6.2.262)

and, corresponding to the case when p = 1 in the two-dimensional setting, the

following operator is well defined, linear, and bounded:
 2  2
S : H 1 (∂Ω, σ) −→ L ∞ (∂Ω, σ) ∩ 𝒞0 (∂Ω) if n = 2. (6.2.263)
 
Also, if p ∈ n−1 n , 1 then the boundary-to-boundary single layer potential
operator S for the Stokes system induces a linear and bounded mapping
 n  .p n
S : H p (∂Ω, σ) −→ H1 (∂Ω, σ) . (6.2.264)
 
As a consequence, for each p ∈ n−1 n , 1 the following operator is well-defined,
linear, and bounded:
 n  .p " n
[S] : H p (∂Ω, σ) −→ H1 (∂Ω, σ) ∼ defined as
 .p " n  p n (6.2.265)
[S] f := [S f ] ∈ H1 (∂Ω, σ) ∼ , ∀ f ∈ H (∂Ω, σ) ,

when the quotient space is equipped with the semi-quasinorm introduced in

(A.0.92).
 
(8) Recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n of the
Stokes system in Rn , and the accompanying pressure vector q = (q j )1≤ j ≤n , from
   .p n
(6.2.1)-(6.2.2). Given p ∈ n−1 
n , 1 , for each f = ( f j )1≤ j ≤n ∈ H1 (∂Ω, σ) ,
each γ, r ∈ {1, . . . , n}, and each x ∈ Ω one has (using the summation convention)
 mod    
∂r Dλ f γ (x) = (∂k E jγ )(x − ·) ∂Ω , ∂τr k f j
  
+ λ (∂j Ekγ )(x − ·) ∂Ω , ∂τr k f j
  
+ qγ (x − ·) ∂Ω , ∂τ j r f j (6.2.266)

15 Proposition 2.3.8 tells us that if Ω ⊆ R n is an open set satisfying a two-sided local John condition
and whose boundary is an unbounded Ahlfors regular set, then said semi-quasinorm is actually a
.p "
genuine quasinorm, and H1 (∂Ω, σ) ∼ becomes a quasi-Banach space
738 6 Green Formulas and Layer Potential Operators for the Stokes System

and   
Pλ f(x) = (1 + λ) q j (x − ·) ∂Ω , ∂τk j fk (6.2.267)

if ∂Ω is unbounded, plus similar formulas without equivalence classes of func-

tions modulo constants if ∂Ω is bounded, with the pairings in (6.2.266)-(6.2.267)
understood in the sense of [69, Theorem 4.6.1] (keeping in mind that all tangen-
tial derivatives belong to the Hardy space H p (∂Ω, σ)). If some aperture param-
eter κ > 0 has also been fixed then, as a consequence of (6.2.266)-(6.2.267),
(6.2.1)-(6.2.2), and [70, Corollary 2.5.4],
 .p n
for each given f ∈ H1 (∂Ω, σ) , the nontangential
κ−n.t. κ−n.t.
boundary traces (∇Dλ f) ∂Ω and (Pλ f) ∂Ω exist (in (6.2.268)
mod

Cn·n and Cn , respectively) σ-a.e. on the set ∂Ω.

From (6.2.266)-(6.2.267), (6.2.1)-(6.2.2), and item (3) in [70, Theorem 2.4.1],
 . p constant nC > 0, depending only on ∂Ω, n,
one also sees that there exists a finite
κ, and p, such that for each f ∈ H1 (∂Ω, σ) one has
   
Nκ (∇D mod f)
λ L p (∂Ω,σ)
+ Nκ (Pλ f) L p (∂Ω,σ) ≤ C f[ H. p (∂Ω,σ)]n . (6.2.269)
1

 n−1 
Proof To deal with item (1), fix p ∈ n , 1 and q ∈ (1, ∞). As in the case of
Theorem 2.1.1, the gist of the proof is to show that if a : ∂Ω → Cn is an arbitrary
(p, q)-atom with vanishing moment,
∫
a dσ = 0, (6.2.270)
∂Ω
 n
then m := Kλ# a is a fixed multiple of a molecule for the Hardy space H p (∂Ω, σ)
(cf. [69, Definition 4.5.1]). The estimates in [69, (4.5.1)-(4.5.2)] may be established
by closely mimicking the arguments used in the proof (2.1.9) and (2.1.12). To justify
the vanishing moment condition
∫
 dσ = 0,
m (6.2.271)
∂Ω

observe that

 = Kλ# a = ∂νλ (𝒮a, Q a) + 12 a.

m (6.2.272)

Thanks to this and (6.2.270), we see that (6.2.271) follows as soon as we establish
that ∫
∂νλ (𝒮a, Q a) dσ = 0. (6.2.273)
∂Ω
To this end, recall that
6.2 Boundary Layer Potential Operators for the Stokes System . . . 739

∂νλ (𝒮a,Q a) (6.2.274)

  κ−n.t.  κ−n.t. κ−n.t. 
= ν j (∂j (𝒮a)α ∂Ω
+ λ ν j (∂α (𝒮a) j ∂Ω
− να (Q a) ∂Ω .
1≤α ≤n

To proceed, fix α ∈ {1, . . . , n} and consider the vector field

 n
F := ∂j (𝒮a)α e j + λ ∂α (𝒮a) j e j − (Q a)eα ∈ 𝒞∞ (Ω) . (6.2.275)

Then (6.2.275) and (6.2.15) permit us to compute

divF = ∂j ∂j (𝒮a)α + λ ∂α ∂j (𝒮a) j − ∂α (Q a)

 
= Δ𝒮a + λ div𝒮a − ∇Q a = 0 in Ω. (6.2.276)
α

Also, having fixed a background aperture parameter κ > 0, [70, Theorem 2.5.1]
implies that
κ−n.t.
F ∂Ω exists at σ-a.e. point on ∂Ω (6.2.277)

and, in fact, the design of F is such that

 κ−n.t.   
ν · F ∂Ω = ∂νλ (𝒮a, Q a) at σ-a.e. point on ∂Ω. (6.2.278)
α

Since the components of the function a : ∂Ω → Cn are multiples of (1, q)-atoms on

∂Ω, we may invoke [70, (2.4.14)] with p = 1 to also conclude that

Nκ F ∈ L 1 (∂Ω, σ). (6.2.279)

Finally, the vanishing moment property of the atom together with (6.2.275) and
(6.2.1)-(6.2.2) imply that
in the case when Ω is an exterior domain we have

F(x) = O(|x| −n ) as x ∈ Ω satisfies |x| → ∞; as such, (6.2.280)
condition [68, (1.2.9)] is currently satisfied.
Collectively, (6.2.275), (6.2.276), (6.2.277), (6.2.279), (6.2.280), guarantee the va-
lidity of the Divergence Formula [68, (1.2.2)] which, in light of (6.2.278), (2.1.20),
and the arbitrariness of α ∈ {1, . . . , n}, presently gives (6.2.273). This establishes
(6.2.271) which, in concert with [69, Theorem 4.4.7], ultimately implies that Kλ#
extends uniquely a linear and bounded operator in the context  of (6.2.250). Finally,
[69, Theorem 4.4.3] ensures that various choices of p ∈ n−1 ,
n 1 in (6.2.250) yield
operators which are compatible with one another. This concludes the treatment of
item (1). Next, the claims in item (2) are consequences of what we have just proved
and real interpolation (cf. [69, Theorem 4.3.1]).
Turning attention to item (3), the properties claimed in (6.2.252) follows from
Lemma 2.2.1 and definitions. Next, the estimate in (6.2.253) is seen directly from
740 6 Green Formulas and Layer Potential Operators for the Stokes System

(6.2.1)-(6.2.2) and [70, (2.4.14)]. As regards the jump-formula (6.2.254), we first

observe that, thanks to (6.2.243), (6.2.15), and (6.2.253),
 p n .  n
H (∂Ω, σ)  f −→ ∂νλ (𝒮 f, Q f ) ∈ H p (∂Ω, σ) (6.2.281)

is a well-defined, linear, and bounded assignment. Moreover, from item (3) in Theo-
rem 6.2.10, (6.2.113), and (6.1.6) we know that this assignment agrees with − 12 I +Kλ#
when acting on arbitrary Cn -valued (p, q)-atoms on ∂Ω (with q ∈ (1, ∞)). Since from
the current item (1) we also know that − 12 I + Kλ# is a well-defined linear and bounded
 n
operator on H p (∂Ω, σ) , the jump-formula (6.2.254) now follows via a standard
density argument (based on [69, (4.4.114)]). The claims in items (4)-(7) are justified
by reasoning much as in the proof of Theorem 2.3.1, making use of Theorem 6.2.4
and Theorem 6.2.10 (also reasoning as in the proof of Theorem 2.2.6 to justify
(6.2.262) and (6.2.263)).
As regards item (8), formula (6.2.266) follows from identity (6.2.91) and [69,
Lemma 11.10.4] (currently applied with ϕ one of the entries in (∇E)(x − ·)), while
formula (6.2.267) is implied by (6.2.93) and [69, Lemma 11.10.4] (now used with ϕ
 − ·)).
one of the entries in (∇q)(x 

6.3 Other Integral Representations and Fatou-Type Results for

the Stokes System

We debut by presenting integral representation formulas for derivatives of the com-

ponents of the velocity function as well as for the pressure function, working solely
with integrability properties of the nontangential maximal operator, without making
any assumptions on the existence of nontangential pointwise boundary traces for
said functions.

Theorem 6.3.1 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an Ahlfors regular domain.

Abbreviate σ := H n−1 ∂Ω andfix some aperture parameter κ > 0. Assume the pair
n
consisting of u = (u j )1≤ j ≤n ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) is a null-solution of the
Stokes system in Ω, i.e.,

Δ
u − ∇π = 0 in Ω, u = 0 in Ω,
div (6.3.1)

which also satisfies

 n−1

Nκ (∇
u) ∈ L p (∂Ω, σ) and Nκ π ∈ L p (∂Ω, σ) for some p ∈ , ∞ . (6.3.2)
n
 
Finally, recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n of
the Stokes system in Rn , and the accompanying pressure vector q = (q j )1≤ j ≤n , from
(6.2.1)-(6.2.2).
Then for each λ ∈ C and each , s, j ∈ {1, . . . , n} one has
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 741
. .λ  n
∂τ s u j ∈ H p (∂Ω, σ) and ∂ν (
u, π) ∈ H p (∂Ω, σ) . (6.3.3)

Moreover, given any r, γ ∈ {1, . . . , n} and λ ∈ C, one has (with the duality pairings
understood in the sense of [69, Theorem 4.6.1] with Σ := ∂Ω, and the summation
convention over repeated indices in effect)
 . 
(∂r uγ )(x) = (∂k E jγ )(x − ·) ∂Ω , ∂τr k u j
 . 
+ λ (∂j Ekγ )(x − ·) ∂Ω , ∂τr k u j
  . 
+ qγ (x − ·) ∂Ω , ∂τ j r u j
 .  
− (∂r Eγα )(x − ·) ∂Ω , ∂νλ (
u, π) α , for all x ∈ Ω, (6.3.4)

and
  . 
π(x) = (1 + λ) q j (x − ·) ∂Ω , ∂τk j uk
  .  
− qα (x − ·) ∂Ω , ∂νλ (
u, π) α , for all x ∈ Ω, (6.3.5)
 
assuming that p ∈ n−1 , 1 and ∂Ω is unbounded. Similar formulas hold when
  n
p ∈ n−1n , 1 and Ω is bounded, this time omitting taking equivalence classes of
functions modulo
  constants in the duality pairings in (6.3.4)-(6.3.5). In the case
when p ∈ n−1 n , 1 and Ω is an exterior domain, these formula continue to be valid
under the additional assumption that there exists some μ ∈ (1, ∞) such that
⨏
 
|∇u| + |π| dL n = o(1) as R → ∞. (6.3.6)
B(0, μ R)\B(0,R)

Finally, similar integral representation formulas hold when p ∈ (1, ∞), this time
interpreting all duality pairings as integration on ∂Ω with respect to σ, i.e.,
∫
. 
(∂r uγ )(x) = (∂k E jγ )(x − y) ∂τr k u j (y) dσ(y)
∂Ω
∫
. 
+λ (∂j Ekγ )(x − y) ∂τr k u j (y) dσ(y)
∂Ω
∫
. 
+ qγ (x − y) ∂τ j r u j (y) dσ(y)
∂Ω
∫
. 
− (∂r Eγα )(x − y) ∂νλ (
u, π) α (y) dσ(y), for all x ∈ Ω, (6.3.7)
∂Ω

and
742 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
. 
π(x) = (1 + λ) q j (x − y) ∂τk j uk (y) dσ(y)
∂Ω
∫
. 
− qα (x − y) ∂νλ (
u, π) α (y) dσ(y), for all x ∈ Ω, (6.3.8)
∂Ω

with the same caveat as before when Ω is an exterior domain.

Proof The claims in (6.3.3) are consequences of assumptions, [69, (10.2.14)], and
Theorem 6.2.10. To prove the integral representation formula (6.3.4), fix some arbi-
trary r, γ ∈ {1, . . . , n}, λ ∈ C, and x ∈ Ω. The strategy is to use the version of the
Divergence Formula established in [68, Theorem 1.9.4] to the vector field define at
L n -a.e. point in Ω as
  
F := (∂k E jγ )(x − ·) + λ(∂j Ekγ )(x − ·) (∂k u j )er − (∂r u j )ek

+ qγ (x − ·)(∂r u j )e j
 
− (∂r Eγα )(x − ·) (∂j uα )e j + λ (∂α u j )e j − πeα . (6.3.9)
 1 n
Upon recalling (6.2.1)-(6.2.2), we clearly have F ∈ Lbdd (Ω, L n ) . With the diver-
gence considered in the sense of distributions in Ω, we also have
 
divF = − ∂r ∂k [E jγ (x − ·)] + λ∂r ∂j [Ekγ (x − ·)] (∂k u j )
 
+ (∂k E jγ )(x − ·) + λ(∂j Ekγ )(x − ·) (∂r ∂k u j )
 
+ ∂k ∂k [E jγ (x − ·)] + λ∂k ∂j [Ekγ (x − ·)] (∂r u j )
 
− (∂k E jγ )(x − ·) + λ(∂j Ekγ )(x − ·) (∂k ∂r u j )

− (∂j qγ )(x − ·)(∂r u j ) + qγ (x − ·)(∂j ∂r u j )

 
+ ∂j ∂r [Eγα (x − ·)] (∂j uα ) + λ (∂α u j )
 
− (∂r Eγα )(x − ·) (∂j ∂j uα ) + λ (∂j ∂α u j )

− ∂α ∂r [Eγα (x − ·)]π + (∂r Eγα )(x − ·)∂α π. (6.3.10)

Using (6.2.6), the first line and the sixth line cancel. The second line and fourth line
also cancel. In the third line, write ∂k ∂k [E jγ (x − ·)] = Δ[E jγ (x − ·)] and use the fact
that λ∂k ∂j [Ekγ (x − ·)] = 0 (cf. (6.2.5)). In the fifth line, we have ∂j ∂r u j = 0. In
the seventh line, write ∂j ∂j uα = Δuα and note that λ (∂j ∂α u j ) = λ ∂α (div u) = 0.
Finally, in the eighth line ∂α ∂r [Eγα (x − ·)] = 0 (cf. (6.2.5)). Implementing these
observations leads to
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 743
 
divF = (ΔE jγ )(x − ·)] − (∂j qγ )(x − ·) (∂r u j )
 
− (∂r Eγα )(x − ·) Δuα − ∂α π

= δ jγ δx (∂r u j ) = (∂r uγ )(x)δx, (6.3.11)

thanks to (6.2.4) and (6.3.1). In particular, divF ∈ ℰ (Ω), so condition [68, (1.9.29)]
is presently satisfied.
Next, with μ as in (6.3.6)
 if Ω is an exterior domain and μ := 2 otherwise, pick a
function φ ∈ 𝒞∞ c B(0, μ) with the property that φ ≡ 1 on B(0, 1). Then the family

ℱ := {φ R }R>0, where φ R := φ(·/R) for each R ∈ (0, ∞), (6.3.12)

becomes a system of auxiliary functions (in the sense of [68, (1.3.3)]). From [68,
(1.9.30)] we know that
∫

[F]ℱ = − lim R −1  dL n (y)
(∇φ)(y/R) · F(y) (6.3.13)
R→∞ Ω

Let us assume for the moment that either Ω is bounded, or ∂Ω is unbounded. We may
then rely on (6.2.1)-(6.2.2), [68, Proposition 8.6.3], (6.3.2), and Hölder’s inequality
to estimate, for R large,
∫
  
R−1  − y)| |(∇
|(∇φ)(y/R)| |(∇E)(x − y)| + | q(x u)(y)| + |π(y)| dL n (y)
Ω
∫
 
≤ CR−n |∇
u | + |π| dL n
Ω∩[B(0, μ R)\B(0,R)]

∫   n−1
np  n−1
np   1− n−1
≤ CR−n |∇
u | + |π| dL n Rn np

Ω
 
≤ CR−
n−1
p Nκ (∇
u) L p (∂Ω,σ) + Nκ π L p (∂Ω,σ)

= o(1) as R → ∞. (6.3.14)

From (6.3.9), (6.3.13), and (6.3.14) we then conclude that

 ℱ = 0.
if either Ω is bounded, or ∂Ω is unbounded, then [F] (6.3.15)

In the case when Ω is an exterior domain, thanks to (6.3.6) we may directly estimate
744 6 Green Formulas and Layer Potential Operators for the Stokes System
∫
  
R−1 |(∇φ)(y/R)| |(∇E)(x − y)| + |q(x − y)| |(∇
u)(y)| + |π(y)| dL n (y)
Ω
⨏
 
≤C |∇
u | + |π| dL n
B(0, μ R)\B(0,R)

= o(1) as R → ∞. (6.3.16)

Thus,
 ℱ = 0.
if Ω is an exterior domain then [F] (6.3.17)
Collectively, (6.3.15) and (6.3.17) show that in all cases the limit in [68, (1.9.30)]
written for F exists, and is actually zero. 
To proceed, assume first that p ∈ n−1 n , 1 and ∂Ω is unbounded. Choose a
function η ∈ 𝒞∞ c (Ω) satisfying η ≡ 1 near x. Granted [70, (3.3.118), (3.3.119)], [68,
Definition 4.2.6] permits us to define ν • F as
   
ν • F = ν • (1 − η)F in Lipc (∂Ω) (6.3.18)
 
where ν • (1−η)F is now interpreted in the sense of [68, Proposition 4.2.3] (bearing
in mind that both the components of the vector field (1 − η)F and its divergence are
absolutely integrable on arbitrary bounded measurable subsets of Ω). Hence, with
the piece of notation introduced in [68, (1.9.31)], we have
     
 1 = ν • (1 − η)F , 1
ν • F, ℱ ℱ
   
= lim (Lip c (∂Ω)) ν • (1 − η)F , φ R ∂Ω Lip c (∂Ω)
R→∞

= I + II + III + IV, (6.3.19)

where
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 745
   
I := lim (Lip c (∂Ω)) ν • (1 − η)(∂k E jγ )(x − ·) (∂k u j )er − (∂r u j )ek ,
R→∞

φR ∂Ω Lip c (∂Ω)
   
II := lim (Lip c (∂Ω)) ν • λ(1 − η)(∂j Ekγ )(x − ·) (∂k u j )er − (∂r u j )ek ,
R→∞

φR ∂Ω Lip c (∂Ω)
   
III := lim (Lip c (∂Ω)) ν • (1 − η)(∂r Eγα )(x − ·) (∂j uα + λ ∂α u j )e j − πeα ,
R→∞

φR ∂Ω Lip c (∂Ω)
   
IV := lim (Lip c (∂Ω)) ν • (1 − η)qγ (x − ·)(∂r u j )e j , φ R ∂Ω Lip c (∂Ω)
. (6.3.20)
R→∞

As regards the first term above, we may write

   
I = lim (Lipc (∂Ω)) (1 − η)(∂k E jγ )(x − ·) ν • (∂k u j )er − (∂r u j )ek ,
R→∞ ∂Ω

φR ∂Ω Lip c (∂Ω)
  . 
= lim (Lip c (∂Ω)) (∂k E jγ )(x − ·) ∂ u ,φ
∂Ω τr k j R ∂Ω Lip c (∂Ω)
R→∞
.   
= lim (Lip c (∂Ω)) ∂τr k u j , φ R (∂k E jγ )(x − ·) ∂Ω Lip c (∂Ω)
R→∞
.   
= lim H p (∂Ω,σ) ∂τr k u j , φ R (∂k E jγ )(x − ·) ∂Ω (H p (∂Ω,σ))∗
R→∞
. 
= H p (∂Ω,σ) ∂τr k u j , (∂k E jγ )(x − ·) ∂Ω (H p (∂Ω,σ))∗ . (6.3.21)

In (6.3.21), the first equality uses [68, (4.2.14)]. The second equality relies on two
things, namely the observation that 1 − η = 1 on ∂Ω, and the definition of the
weak tangential derivative from [68, Example 4.2.4] (cf. (A.0.175)-(A.0.176) in
the Glossary). The third equality is implied by [68, (4.1.43)]. The fourth equality
follows on account of (6.3.3) and [69, Lemma 4.6.4] ((2.2.2) is also helpful in this
regard). Finally, the last equality in (6.3.21) is a consequence of the fact that for each
j, k ∈ {1, . . . , n} we have
746 6 Green Formulas and Layer Potential Operators for the Stokes System
   
lim (∂k E jγ )(x − ·)φ R ∂Ω
= (∂k E jγ )(x − ·) ∂Ω weak-∗ in
R→∞

⎧ . "
 ∗ ⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼ if p < 1,
⎪ (6.3.22)
H p (∂Ω, σ) =
⎪ $
⎪ BMO(∂Ω,
⎩ σ) if p = 1.

In turn, this is implied by the general weak-∗ convergence results established in [69,
Lemma 4.8.4] (here (2.2.2) helps) and, respectively, [69, Lemma 4.8.1] (also bearing
in mind the trivial bounded embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ) in the latter
case). This finishes the justification of (6.3.21).
Going further, in a similar fashion we also obtain
. 
II = H p (∂Ω,σ) ∂τr k u j , λ(∂j Ekγ )(x − ·) ∂Ω (H p (∂Ω,σ))∗ (6.3.23)

and
 .  
III = H p (∂Ω,σ) ∂νλ (
u, π) α, (∂r Eγα )(x − ·) ∂Ω (H p (∂Ω,σ))∗ . (6.3.24)

In addition, upon noting that (∂r u j )e j = (∂r u j )e j − (∂j u j )er , the same type of
argument gives
.  
IV = H p (∂Ω,σ) ∂τ j r u j , qγ (x − ·) ∂Ω (H p (∂Ω,σ))∗ . (6.3.25)

In fact, similar arguments may be employed

 to prove (6.3.21), (6.3.23), (6.3.24),
(6.3.25) in the case when p ∈ n−1 n , 1 and ∂Ω is bounded, a scenario in which
the dual of H p and the corresponding duality bracket no  longer  involve classes of
functions modulo constants. Thus, in all cases, if p ∈ n−1 n , 1 then from (6.3.19),
(6.3.21), and (6.3.23)-(6.3.25) we obtain
   . 
 1 = (∂k E jγ )(x − ·)
ν • F, , ∂ u
τr k j
ℱ ∂Ω
 . 
+ λ(∂j Ekγ )(x − ·) ∂Ω , ∂τr k u j
  . 
+ qγ (x − ·) ∂Ω , ∂τ j r u j
 .  
+ (∂r Eγα )(x − ·) ∂Ω , ∂νλ (
u, π) α (6.3.26)

where all brackets in the right-hand side refer to the duality pairing from [69,
Theorem 4.6.1].  
At this stage, if p ∈ n−1n , 1 then [68, Theorem 1.9.4] applies and, on account
of (6.3.11), (6.3.15), (6.3.17), and (6.3.26), the Divergence Formula [68, (1.9.32)]
presently yields (6.3.4). Finally, the case when p ∈ (1, ∞) is handled analogously,
keeping in mind the compatibility of the duality pairing with the ordinary integral
pairing on ∂Ω, and using Lebesgue’s Dominated convergence Theorem in place of
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 747

(6.3.22). In such a scenario we arrive at the integral representation formula claimed

in (6.3.7).
To finish the proof of Theorem 6.3.1, there remains to deal with the integral
representations for the pressure, claimed in (6.3.5) and (6.3.8). This time, the idea
is to apply the version of the Divergence Formula from [68, Theorem 1.9.4] to the
vector field
 
G := (1 + λ)q j (x − ·)(∂j uk )ek − qα (x − ·) (∂j uα )e j + λ (∂α u j )e j − πeα , (6.3.27)

where x ∈ Ω is an arbitrary, fixed point. In this regard, retaining the system of

auxiliary functions ℱ from (6.3.12), by reasoning much as in (6.3.14), (6.3.16) we
arrive at the conclusion that
∫
 ℱ = − lim R−1 (∇φ)(y/R) · G(y)
[G]  dL n (y) = 0. (6.3.28)
R→∞ Ω

Also,

divG = (1 + λ)∂k [q j (x − ·)](∂j uk ) + (1 + λ)q j (x − ·)(∂k ∂j uk )

   
− ∂j [qα (x − ·)] (∂j uα ) + λ (∂α u j ) − qα (x − ·) (∂j ∂j uα ) + λ (∂j ∂α u j )

+ ∂α [qα (x − ·)]π + qα (x − ·)∂α π

= π(x)δx, (6.3.29)
 
n , 1 and
where the last equality makes use of (6.2.6)-(6.2.7). Moreover, if p ∈ n−1
∂Ω is unbounded then the same type of argument as in (6.3.19)-(6.3.26) now gives
    . 
 1 = (1 + λ) q j (x − ·)
ν • G, , ∂τ u k
ℱ ∂Ω kj

  .  
− qα (x − ·) ∂Ω , ∂νλ (
u, π) α (6.3.30)

where all brackets in the right-hand side refer to the duality pairing from [69,
Theorem 4.6.1]. In fact, similar formulas valid are valid when ∂Ω is bounded, or
p ∈ (1, ∞). Thanks to them, (6.3.28),
 and(6.3.29), the Divergence Formula [68,
n , 1 and ∂Ω is unbounded (plus a similar
(1.9.32)] now yields (6.3.5) if p ∈ n−1
version when ∂Ω is bounded), as well as (6.3.8) when p ∈ (1, ∞). 
In turn, Theorem 6.3.1 is one of the key ingredients in the proof of our first
Fatou-type result for null-solutions of the Stokes system in arbitrary UR domains,
presented below.

Theorem 6.3.2 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be a UR domain. Abbreviate

σ := H n−1 ∂Ω and denote by ν the geometric  n measure theoretic outward unit
normal to Ω. Assume the functions u ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) satisfy, for some
aperture parameter κ > 0,
748 6 Green Formulas and Layer Potential Operators for the Stokes System

Δ
u − ∇π = 0 in Ω, u = 0 in Ω, and
div
  (6.3.31)
Nκ (∇
u), Nκ π ∈ L p (∂Ω, σ) for some p ∈ n−1
n ,∞ .

Then
κ−n.t. κ−n.t.
the nontangential pointwise traces ∇u ∂Ω and π ∂Ω exist, (6.3.32)
in Cn×n and C, respectively, at σ-a.e. point on ∂Ω.

In addition, these traces are independent of the aperture parameter κ ∈ (0, ∞), the
κ−n.t.   n×n κ−n.t.
function (∇u) ∂Ω belongs to L p (∂Ω, σ) , the function π ∂Ω belongs to the
space L p (∂Ω, σ), and
 κ−n.t.   
(∇
u)  ≤ Nκ (∇
u) L p (∂Ω,σ), (6.3.33)
∂Ω [L p (∂Ω,σ)] n×n
 κ−n.t.   
π  ≤ Nκ π  L p (∂Ω,σ) . (6.3.34)
∂Ω L p (∂Ω,σ)

Furthermore, having fixed some arbitrary λ ∈ C,

  .λ  n
if p ∈ n−1 n , 1 then ∂ν (
u, π) ∈ H p (∂Ω, σ) and
.  n (6.3.35)
∂τ j k u ∈ H p (∂Ω, σ) for each j, k ∈ {1, . . . , n},

and the following H p /L p -boundary maximum principle holds:

   
Nκ (∇
u ) p + Nκ π  p (6.3.36)
L (∂Ω,σ) L (∂Ω,σ)

.  %
n
.   
≈ ∂νλ (
u, π)[H p (∂Ω,σ)]n + ∂τ u p
jk [H (∂Ω,σ)] n
if p ∈ n−1
n ,1
j,k=1

and
   
Nκ (∇
u) L p (∂Ω,σ) + Nκ π  L p (∂Ω,σ) (6.3.37)
 κ−n.t.   κ−n.t. 
≈ (∇
u) ∂Ω [L p (∂Ω,σ)]n×n + π ∂Ω  L p (∂Ω,σ) if p ∈ (1, ∞)

provided, in the case when Ω is an exterior domain, it is also assumed that there
exists some number μ ∈ (1, ∞) such that
⨏
 
|∇u| + |π| dL n = o(1) as R → ∞. (6.3.38)
B(0, μ R)\B(0,R)

Proof From (6.3.4)-(6.3.5) (plus their counterparts when Ω is bounded), as well as

(6.3.7)-(6.3.8), on account of (6.3.3), [70, Theorem 2.5.1], and item (6) in Theo-
rem 2.2.6 we conclude that
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 749
κ−n.t. κ−n.t.
the nontangential pointwise traces (∇
u) ∂Ω and π ∂Ω exist at σ-a.e.
point on ∂Ω if either Ω is bounded, or ∂Ω is unbounded, or Ω is an (6.3.39)
exterior domain and the decay condition (6.3.6) holds.
This Fatou-type property further self-improves. Specifically, in the case when Ω is an
exterior domain, working with ΩR := B(0, R) ∩ Ω in place of Ω (where R ∈ (0, ∞) is
a sufficiently large number), much as in the proof of [70, Theorem 3.1.6], eliminates
the need of asking that the decay condition (6.3.6) holds. Simply put, (6.3.32) holds
as stated. The integrability properties claimed next in the statement, along with
the independence of the nontangential traces on the aperture parameter and the
estimates in (6.3.33)-(6.3.34), follow from assumptions, [68, Proposition 8.9.8], [68,
Corollary 8.9.6], [68, (8.9.8)], [68, (8.9.8)], and [68, Proposition 8.8.6].
Going further, (6.3.35) is implied by (6.3.3). The left-pointing inequality in
(6.3.36) is a consequence of (6.3.4), (6.3.5), (6.3.3), [70, (2.4.14)], and (6.2.1)-
(6.2.2). The right-pointing inequality in (6.3.36) comes from (6.2.243) and [69,
(10.2.14)]. Next, the left-pointing inequality in (6.3.37) follows from (6.3.7), (6.3.8),
and [70, (2.4.9)]. Finally, granted what we have proved already, the right-pointing
inequality in (6.3.37) is seen from [68, (8.9.8)]. 

It is possible to further expand upon the approach taken to prove Theorem 6.3.1
as to allow membership of the nontangential maximal functions of null-solutions
of the Stokes system to Generalized Banach Function Spaces in place of Lebesgue
spaces. Here is a formal statement:

Theorem 6.3.3 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an Ahlfors regular domain.

Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward
unit normal to Ω. Suppose X is a Generalized Banach Function Space on (∂Ω, σ)
with the property that

M ∂Ω : X → X and M ∂Ω : X → X
(6.3.40)
are well-defined bounded mappings,

where M ∂Ω is the Hardy-Littlewood maximal operator on ∂Ω, and X is the associ-

ated space of X (cf. [69, Definitions 5.1.4, 5.1.11]).  n
Assume the pair consisting of u = (u j )1≤ j ≤n ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) is a
null-solution of the Stokes system in Ω, i.e.,

Δ
u − ∇π = 0 in Ω, div u = 0 in Ω, (6.3.41)

which also satisfies, for some aperture parameter κ > 0,

Nκ (∇
u) ∈ X and Nκ π ∈ X. (6.3.42)

In the case when Ω is an exterior domain make the additional assumption that there
exists some μ ∈ (1, ∞) such that
750 6 Green Formulas and Layer Potential Operators for the Stokes System
⨏
 
|∇u| + |π| dL n = o(1) as R → ∞. (6.3.43)
B(0,μR)\B(0,R)
 
Finally, recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n of
the Stokes system in Rn , and the accompanying pressure vector q = (q j )1≤ j ≤n , from
(6.2.1)-(6.2.2).
Then for each λ ∈ C and each , s, j ∈ {1, . . . , n} one has
. .  n
∂τ s u j ∈ X and ∂νλ (
u, π) ∈ X . (6.3.44)

Moreover, given any r, γ ∈ {1, . . . , n} and λ ∈ C, one has (with absolutely convergent
integrals)
∫
. 
(∂r uγ )(x) = (∂k E jγ )(x − y) ∂τr k u j (y) dσ(y)
∂Ω
∫
. 
+λ (∂j Ekγ )(x − y) ∂τr k u j (y) dσ(y)
∂Ω
∫
. 
+ qγ (x − y) ∂τ j r u j (y) dσ(y)
∂Ω
∫
. 
− (∂r Eγα )(x − y) ∂νλ (
u, π) α (y) dσ(y) (6.3.45)
∂Ω

for all x ∈ Ω, and

∫
. 
π(x) = (1 + λ) q j (x − y) ∂τk j uk (y) dσ(y)
∂Ω
∫
. 
− qα (x − y) ∂νλ (
u, π) α (y) dσ(y), for all x ∈ Ω. (6.3.46)
∂Ω

Proof Thanks to (6.3.40), we may invoke [69, Proposition 5.2.7] which guarantees
that
there exist q ∈ (1, ∞) and ε ∈ (0, 1) such that
 σ(x)  (6.3.47)
X → L q ∂Ω , continuously.
1 + |x| n−1−ε

From [69, Proposition 10.2.6], Definition 6.2.9, [69, Example 10.2.2], (6.3.40), and
(6.3.47) we then conclude that the memberships claimed in (6.3.44) are indeed true.
Having established this, we now run the same argument as in the proof of The-
orem 6.3.1, with two key differences singled out below. First, in the case when
either Ω is bounded, or ∂Ω is unbounded, in place of (6.3.14) we now make use of
(6.2.1)-(6.2.2), Hölder’s inequality, [68, Proposition 8.6.3], [68, (8.1.17)], (6.3.42),
and (6.3.47) to estimate (with q and ε as in (6.3.47)):
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 751
∫
  
R−1  − y)| |(∇
|(∇φ)(y/R)| |(∇E)(x − y)| + | q(x u)(y)| + |π(y)| dL n (y)
Ω
∫
 
≤ CR−n |∇
u | + |π| dL n
Ω∩[B(0,μR)\B(0,R)]

∫   n−1
nq  n−1
nq   1− n−1
≤ CR−n |∇
u | + |π| dL n Rn nq

Ω∩[B(0,μR)\B(0,R)]

∫  1/q
− n−1
≤ CR q |Nκ (∇
u)| q dσ
∂Ω∩B(0,μ(2+κ)R)
!
∫  1/q
+ |Nκ π| dσq
∂Ω∩B(0,μ(2+κ)R)

∫ |Nκ (∇u)(z)| q  1/q

≤ CR−
n−1
q Rn−1−ε dσ(z)
∂Ω∩B(0,μ(2+κ)R) 1 + |z| n−1−ε
∫ |(Nκ π)(z)| q  1/q
+ CR−
n−1
q Rn−1−ε dσ(z)
∂Ω∩B(0,μ(2+κ)R) 1 + |z| n−1−ε
!
− qε
   
≤ CR u )
Nκ (∇  σ(z)
 + Nκ π   σ(z)

L q ∂Ω, L q ∂Ω,
1+|z | n−1−ε 1+|z | n−1−ε

ε
 
≤ CR− q Nκ (∇
u)X + Nκ πX

= o(1) as R → ∞. (6.3.48)

As in the case of (6.3.14), this suits our purposes. The second significant adjustment
we have to make in the proof Theorem 6.3.1 regards (6.3.22). In lieu of this we shall
presently use the fact that for each j, k ∈ {1, . . . , n}, each fixed x ∈ Ω, and any f ∈ X
(ultimately playing the role of functions appearing in (6.3.44)) we have
∫ ∫
lim (∂k E jγ )(x − ·)φ R f dσ = (∂k E jγ )(x − ·) f dσ. (6.3.49)
R→∞ ∂Ω ∂Ω
 σ(y) 
In turn, (6.3.49) is a consequence of the fact that f ∈ L 1 ∂Ω , 1+ |y | n−1
as seen from
(6.3.47), that |(∇E)(x − y)| ≤ Cx (1 + |y| n−1 )−1
for all y ∈ ∂Ω as seen from (6.2.1),
and Lebesgue’s Dominated Convergence Theorem.
With these alterations implemented, the rest of the argument in the proof of
Theorem 6.3.1 goes through and gives (6.3.45). The justification of (6.3.46) is
similarly based on the proof of Theorem 6.3.1. 
In turn, the integral representation formulas established in Theorem 6.3.3 allow us
to prove the following versatile Fatou-type theorem for the Stokes system in arbitrary
752 6 Green Formulas and Layer Potential Operators for the Stokes System

UR domains, in which control of the nontangential maximal function is in terms of

a Generalized Banach Function Space.

Theorem 6.3.4 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be a UR domain. Abbreviate

σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
normal to Ω. Assume X is a Generalized Banach Function Space on (∂Ω, σ) with
the property that
M ∂Ω : X → X and M ∂Ω : X → X
(6.3.50)
are well-defined bounded mappings,
where M ∂Ω is the Hardy-Littlewood maximal operator on ∂Ω, and X is the as-
sociated space of X (cf. [69, Definitions 5.1.4, 5.1.11]). Finally, having fixed some
aperture parameter κ > 0, assume
 n
u ∈ 𝒞∞ (Ω) , π ∈ 𝒞∞ (Ω),
Δ
u − ∇π = 0 in Ω, div u = 0 in Ω, (6.3.51)
Nκ (∇
u) ∈ X, Nκ π ∈ X.

Then
κ−n.t. κ−n.t.
the nontangential pointwise traces ∇ u ∂Ω and π ∂Ω exist, (6.3.52)
in Cn×n and C, respectively, at σ-a.e. point on ∂Ω.

In addition, these traces are actually independent of the parameter κ ∈ (0, ∞), the
κ−n.t.   n×n κ−n.t.
function (∇
u) ∂Ω belongs to X , the function π ∂Ω belongs to X, and
 κ−n.t.   
(∇
u) ∂Ω [X]n×n ≤ Nκ (∇
u)X, (6.3.53)
 κ−n.t.   
π  ≤ Nκ π  . (6.3.54)
∂Ω X X

Proof All claims follow from Theorem 6.3.4, [69, Proposition 5.2.7], [70, Theo-
rem 2.5.1], (6.2.1)-(6.2.2), [68, Corollary 8.9.9], [69, (5.1.12)], [68, (8.9.8)], and
[68, Corollary 8.9.6]. 

We continue by presenting another brand of Fatou-type result for null-solutions of

the Stokes system in arbitrary UR domains, which augments that in Theorem 6.3.2.

Theorem 6.3.5 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be a UR domain. Abbreviate

σ := H n−1  ∂Ω and fix some aperture parameter κ > 0. Assume the functions
n
u ∈ 𝒞∞ (Ω) and π ∈ 𝒞∞ (Ω) satisfy

Δ
u − ∇π = 0 in Ω, u = 0 in Ω,
div
 
Nκ (∇
u), Nκ π ∈ L p (∂Ω, σ) for some p ∈ n−1n ,n−1 , (6.3.55)
∗  
1 −1
and Nκ u ∈ L p (∂Ω, σ) where p∗ := p1 − n−1 .
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 753

Then
the nontangential pointwise boundary traces
κ−n.t. κ−n.t. κ−n.t. (6.3.56)
u ∂Ω , ∇u ∂Ω , and π ∂Ω exist σ-a.e. on ∂Ω.
Moreover, these traces are actually independent of κ ∈ (0, ∞), and
κ−n.t.  p∗, p n κ−n.t.   n×n κ−n.t.
u ∂Ω ∈ H1 (∂Ω, σ) , (∇ u) ∂Ω ∈ L p (∂Ω, σ) , π ∂Ω ∈ L p (∂Ω, σ),
(6.3.57)
in a quantitative sense, i.e., there exists some C = C(Ω, κ, p) ∈ (0, ∞) such that
 κ−n.t.     
u  p∗, p u) L p (∂Ω,σ) + C Nκ u L p ∗ (∂Ω,σ),
≤ C Nκ (∇ (6.3.58)
∂Ω [H 1
(∂Ω,σ)] n

 κ−n.t.   
(∇
u)  ≤ Nκ (∇
u) L p (∂Ω,σ), (6.3.59)
∂Ω [L p (∂Ω,σ)] n×n
 κ−n.t.   
π  ≤ C Nκ π  L p (∂Ω,σ) . (6.3.60)
∂Ω L p (∂Ω,σ)

Finally, for each given λ ∈ C, the Stokes weak conormal derivative of the pair
u, π satisfies
.  n
∂νλ (
u, π) belongs to the Hardy space H p (∂Ω, σ) and
 .λ      (6.3.61)
∂ (   u) L p (∂Ω,σ) + C Nκ π  L p (∂Ω,σ)
ν u, π) [H p (∂Ω,σ)] n ≤ C Nκ (∇

for some constant C ∈ (0, ∞) which depends only on Ω, n, λ, κ, p and, with the layer
potential operators Dλ , Pλ , 𝒮, Q associated with Ω and λ as in (6.2.17), (6.2.21),
(6.2.12), (6.2.14), respectively, one has the integral representation formulas
 κ−n.t.  . 
u = Dλ u ∂Ω − 𝒮 ∂νλ (
u, π) in Ω, (6.3.62)
 κ−n.t.  . 
π = Pλ u ∂Ω − Q ∂νλ (
u, π) in Ω, (6.3.63)

with the understanding that if Ω is an exterior domain one also assumes that there
exists μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL n = o(1) as R → ∞, (6.3.64)
B(0, μ R)\B(0,R)

in the case of (6.3.62), and

⨏
 
| u| + R|π| dL n = o(R) as R → ∞, (6.3.65)
B(0, μ R)\B(0,R)

in the case of (6.3.63).

Proof For starters, the claims in (6.3.61) are consequences of (6.2.243). Also, rea-
soning as in the proof of [70, (3.3.111)], the current assumptions on u = (u j )1≤ j ≤n
imply that
754 6 Green Formulas and Layer Potential Operators for the Stokes System
∗
for all j, k ∈ {1, . . . , n} the distribution ν • (u j ek ) belongs to L p (∂Ω, σ)
 
and ν • (u j ek ) L p ∗ (∂Ω,σ) ≤ CNκ u L p ∗ (∂Ω,σ) + CNκ (∇ u) L p (∂Ω,σ)
(6.3.66)

for some constant C ∈ (0, ∞) depending only on Ω, n, κ, p.

To proceed, fix γ ∈ {1, . . . , n} along with x ∈ Ω and consider the vector field
Fx = (Fj )1≤ j ≤n whose scalar components are defined as in (6.2.38). Specifically,
with the summation convention over repeated indices assumed throughout, at L n -a.e.
point in Ω define the j-th component of Fx as
αβ
Fj := −a jk (λ)(∂k Eβγ )(x − ·)uα + qγ (x − ·)u j
αβ
− Eγα (x − ·)a jk (λ)∂k uβ + Eγ j (x − ·)π. (6.3.67)

The strategy is to apply [68, Theorem 1.9.4] to the vector field Fx just introduced.
As a prelude, from (6.3.67) and (6.2.1)-(6.2.2) we note that
 1 n  1 n
Fx ∈ Lbdd (Ω, L n ) ⊆ Lbdd (Ω, L n ) + ℰ (Ω) , (6.3.68)

and recall from (6.2.40) that the divergence of Fx , taken in the sense of distributions
in Ω, is

divFx = uγ (x)δx ∈ ℰ (Ω) ⊆ L 1 (Ω, L n ) + ℰ (Ω). (6.3.69)

Together, (6.3.68) and (6.3.69) prove that the vector field Fx satisfies [68, (1.9.29)].
Next, with μ as in (6.3.64)-(6.3.65)
  if Ω is an exterior domain and μ := 2 otherwise,
pick a function φ ∈ 𝒞∞ c B(0, μ) with the property that φ ≡ 1 on B(0, 1). Then the
family

ℱ := {φ R }R>0, where φ R := φ(·/R) for each R ∈ (0, ∞), (6.3.70)

becomes a system of auxiliary functions (in the sense of [68, (1.3.3)]). Estimates
similar in nature to [70, (3.3.122), (3.3.123)] prove that

if either Ω is bounded, or ∂Ω is unbounded, then [Fx ]ℱ = 0. (6.3.71)

In the case when Ω is an exterior domain, condition (6.3.64) self-improves (thanks

to interior estimates, bearing in mind that Δ2 u = 0) to
⨏   
| u| + R |∇
u | + |π| dL n = o(1) as R → ∞. (6.3.72)
B(0, μR)\B(0,R)

In turn, (6.3.72) together with (6.2.1)-(6.2.2) guarantee that

if Ω is an exterior domain then [Fx ]ℱ = 0. (6.3.73)

6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 755

From (6.3.71) and (6.3.73) we then conclude that, in all cases, the limit in [68,
(1.9.30)] written for Fx exists, and is actually zero.
Next, reasoning as in [70, (3.3.129)-(3.3.133)] (while availing ourselves of
(6.3.61) and (6.3.66)), in place of [70, (3.3.134)] we presently obtain
 
ν • Fx, 1 ℱ
∫ 
   
= − ν • (u j ek ) (y)(∂k E jγ )(x − y) − λ ν • (u j ek ) (y)(∂j Ekγ )(x − y)
∂Ω
  
+ ν • (u j e j ) (y)qγ (x − y) dσ(y)
 . 
− 𝒮(∂νλ (
u, π))(x) . (6.3.74)
γ

Granted (6.3.68)-(6.3.69), (6.3.71), (6.3.73), it follows that [68, Theorem 1.9.4]

may be applied to Fx . On account of (6.3.74), (6.3.69), (6.3.71), and (6.3.73), the
Divergence Formula [68, (1.9.32)] written for Fx then gives
∫ 
   
uγ (x) = − ν • (u j ek ) (y)(∂k E jγ )(x − y) − λ ν • (u j ek ) (y)(∂j Ekγ )(x − y)
∂Ω
  
+ ν • (u j e j ) (y)qγ (x − y) dσ(y)
 . 
− 𝒮(∂νλ (
u, π))(x) . (6.3.75)
γ

In view of the arbitrariness of γ ∈ {1, . . . , n} and x ∈ Ω, from (6.3.75), (6.3.66),

[70, Theorem 2.5.1], item (6) in Theorem 2.2.6, and (6.3.61), we conclude that
κ−n.t.
the nontangential trace u ∂Ω exists at σ-a.e. point on ∂Ω if either
Ω is bounded, or ∂Ω is unbounded, or Ω is an exterior domain (6.3.76)
and the decay condition (6.3.64) holds.
In fact, this Fatou-type property further self-improves. Specifically, in the case when
Ω is an exterior domain, working with ΩR := B(0, R) ∩ Ω in place of Ω (where the
radius R ∈ (0, ∞) is sufficiently large), much as in the proof of [70, Theorem 3.1.6],
eliminates the need of asking that the decay condition (6.3.64) holds. Simply put, at
this stage we have shown that, in all cases,
κ−n.t.
the nontangential trace u ∂Ω exists σ-a.e. on ∂Ω. (6.3.77)

With this in hand, we may now apply [69, Proposition 10.2.9] to each vector field
u j ek with j, k ∈ {1, . . . , n} to conclude that if ν = (ν1, . . . , νn ) is the geometric
measure theoretic outward unit normal to Ω then
   κ−n.t. 
ν • u j ek = νk u j ∂Ω for each j, k ∈ {1, . . . , n}. (6.3.78)
756 6 Green Formulas and Layer Potential Operators for the Stokes System

Returning with (6.3.78) in (6.3.75) ultimately produces (6.3.62), in view of (6.2.17).

Having proved (6.3.77), we may now rely on [69, Definition 11.10.6] and [69,
Corollary 11.10.11] to conclude that the first claim in (6.3.57) along with the estimate
in (6.3.58) hold as well.
κ−n.t.
From Theorem 6.3.2 we also know that the nontangential pointwise traces ∇ u ∂Ω
κ−n.t.
and π ∂Ω exist at σ-a.e. point on ∂Ω, they are independent of the aperture param-
eter as well as p-th power integrable on ∂Ω with respect to the measure σ, and
the estimates (6.3.33)-(6.3.34) hold. In particular, this takes care of the last two
memberships in (6.3.57) and (6.3.59)-(6.3.60).
We are therefore left with proving the integral representation formula claimed in
 in mind, fix an arbitrary point x ∈ Ω and bring back the
(6.3.63). With this goal
vector field G x = G j 1≤ j ≤n with components defined as in (6.2.50), i.e.,

αβ
G j := −(1 + λ)(∂j qk )(x − ·)uk − qα (x − ·)a jk (λ)∂k uβ + q j (x − ·)π (6.3.79)

at L n -a.e. point in Ω. Note that (6.3.79) and (6.2.1)-(6.2.2) guarantee that

 1 n
G x ∈ Lbdd (Ω, L n ) . (6.3.80)

Also, from (6.2.52) we know that the divergence of G x computed in the sense of
distributions in Ω is given by

divG x = (1 + λ)uk (∂k δx ) + π(x)δx ∈ ℰ (Ω) ⊆ L 1 (Ω, L n ) + ℰ (Ω). (6.3.81)

In particular, much as in (6.2.54), we have

 
(𝒞∞

(Ω))∗ divG x , 1 𝒞∞ (Ω) = π(x). (6.3.82)
b b

Based on estimates similar to [70, (3.3.122), (3.3.123)], and also relying on (6.3.65)
in the case when Ω is an exterior domain, we obtain

[G x ]ℱ = 0. (6.3.83)

In an analogous fashion to (6.3.74) we have

∫
   
ν • G x, 1 ℱ = −(1 + λ) ν • (uk e j ) (y)(∂j qk )(x − y) dσ(y)
∂Ω
.λ
− Q(∂ν (
u, π))(x). (6.3.84)

Hence, thanks to (6.3.84), (6.3.78), and (6.2.21),

6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 757
∫
   κ−n.t. 
ν • G x, 1 ℱ = −(1 + λ) ν j (y) uk ∂Ω
(y)(∂j qk )(x − y) dσ(y)
∂Ω
.λ
− Q(∂ν (
u, π))(x)
 κ−n.t.  .
= Pλ u ∂Ω (x) − Q(∂νλ (
u, π))(x). (6.3.85)

At this stage, the Divergence Formula [68, (1.9.32)] written for G x ultimately yields
(6.3.63), on account of (6.3.83), (6.3.82), and (6.3.85). 
Lastly, here is a powerful result which uses our earlier Fatou-type results for
null-solutions of the Stokes system to produce integral representation formulas in an
inclusive setting.
Theorem 6.3.6 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an NTA domain with the
property that ∂Ω is an Ahlfors regular set. Denote by ν the geometric measure
theoretic outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, fix an
arbitrary number λ ∈ C along with an aperture parameter κ ∈ (0, ∞).
mod
In this setting, recall the modified version of the double layer operator Dλ for
  n
the Stokes system acting on vector-valued functions from L 1 ∂Ω , 1+σ(x) |x | n as in
(6.2.171), and the modified version of the single layer operator 𝒮mod for the Stokes
  n
system acting on vector-valued functions from L 1 ∂Ω , 1+σ(x)|x | n−1
as in (6.2.164).
The reader is also reminded that actions of the boundary–to-domain single layer
operator 𝒮 for the Stokes system and the corresponding pressure potential Q have
been extended to Hardy spaces in item (3) of Theorem 6.2.11. Finally, consider a
pair of functions, u = (u j )1≤ j ≤n : Ω → Cn and π : Ω → C, satisfying
 n
u ∈ 𝒞∞ (Ω) , π ∈ 𝒞∞ (Ω),
u − ∇π = 0 and div u = 0 in Ω,
Δ (6.3.86)
 
Nκ (∇
u), Nκ π ∈ L p (∂Ω, σ) for some p ∈ n−1
n , ∞ .

In the case when Ω is an exterior domain make the additional assumption that there
exists μ ∈ (1, ∞) such that
⨏
 
|∇
u | + |π| dL n = o(1) as R → ∞. (6.3.87)
B(0,μR)\B(0,R)

Then the following properties hold: First,

κ−n.t. κ−n.t.
the nontangential pointwise traces ∇ u ∂Ω and π ∂Ω ex- (6.3.88)
ist, in Cn×n and C, respectively, at σ-a.e. point on ∂Ω,

and these traces are actually independent of the aperture parameter κ ∈ (0, ∞). Sec-
κ−n.t.   n×n κ−n.t.
ond, the function (∇
u) ∂Ω belongs to L p (∂Ω, σ) , the function π ∂Ω belongs
to L p (∂Ω, σ), and
758 6 Green Formulas and Layer Potential Operators for the Stokes System
 κ−n.t.   
(∇
u)  ≤ Nκ (∇
u) L p (∂Ω,σ), (6.3.89)
∂Ω [L p (∂Ω,σ)] n×n
 κ−n.t.   
π  ≤ Nκ π  L p (∂Ω,σ) . (6.3.90)
∂Ω L p (∂Ω,σ)
 n
In particular, the pointwise conormal derivative ∂νλ (
u, π) belongs to L p (∂Ω, σ) .
κ−n.t.
Third, the nontangential boundary trace u ∂Ω exists (in Cn ) at σ-a.e. point on ∂Ω
and is independent of the aperture parameter κ. As a function,
κ−n.t.  .p n
u ∂Ω belongs to H1 (∂Ω, σ) and satisfies
 κ−n.t.  .  n (6.3.91)
∂τ j k u ∂Ω = ∂τ j k u ∈ H p (∂Ω, σ) for all j, k ∈ {1, . . . , n},

as well as  κ−n.t.   
u  .p ≤ C Nκ (∇
u) L p (∂Ω,σ) (6.3.92)
∂Ω [ H (∂Ω,σ)] n 1

. some constant C = C(Ω, n, κ, p) ∈ (0,

for  ∞). Fourth,
 n the weak conormal derivative
∂νλ (
u, π) belongs to the Hardy space H p (∂Ω, σ) , and there exists a constant
C = C(Ω, n, κ, p, λ) ∈ (0, ∞) such that
 .λ   
∂ (   u) L p (∂Ω,σ) .
ν u, π) [H p (∂Ω,σ)] n ≤ C Nκ (∇ (6.3.93)

Fifth, there exists some Cn-valued locally constant function cu,π

 in Ω with the
property that whenever p ∈ n−1n , 1 one has
 κ−n.t.  . 
− 𝒮 ∂νλ (
mod
u = Dλ u ∂Ω u, π) + cu,π
 in Ω, (6.3.94)

and
  κ−n.t.   . 
π = (1 + λ)Q ∂τk j uk ∂Ω
− Q ∂νλ (
u, π)
1≤ j ≤n

  . κ−n.t.
= Pλ u ∂Ω − Q ∂νλ (
u, π) in Ω, (6.3.95)

while for p ∈ (1, ∞) one has

 κ−n.t.   
− 𝒮mod ∂νλ (
mod
u = Dλ u ∂Ω u, π) + cu,π
 in Ω, (6.3.96)

and
∫
 κ−n.t.   
π(x) = (1 + λ) q j (x − y)∂τk j uk ∂Ω
(y) dσ(y) − Q ∂νλ (
u, π) (x)
∂Ω
 κ−n.t.   
= Pλ u ∂Ω (x) − Q ∂νλ (
u, π) (x) at each point x ∈ Ω. (6.3.97)

Sixth, for each truncation parameter ε > 0 one has

6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 759

p∗
Nκε u belongs to the space Lloc (∂Ω, σ)
  ∗
1 
1 −1
(6.3.98)
if p ∈ n−1
n , n − 1 and p := p − n−1 ∈ (1, ∞),

and
Nκε u belongs to the space Lloc (∂Ω, σ)
q
(6.3.99)
if n = 2, p = 1, and q ∈ (1, ∞) is arbitrary.

Proof The properties claimed in (6.3.88)-(6.3.90), and (6.3.93) are consequences

of the Fatou-type result for the Stokes system from Theorem 6.3.2 (bearing in mind
that any NTA domain with an Ahlfors regular boundary is a UR domain), while the
properties
 claimed in (6.3.91)-(6.3.92) are seen from Theorem 2.2.8. Next, assume
that p ∈ n−1
n , 1 , with the goal of proving (6.3.94). To this end, define
 κ−n.t.  .   n
− 𝒮 ∂νλ (
u, π) ∈ 𝒞∞ (Ω) ,
mod
w = (wγ )1≤γ ≤n := Dλ u ∂Ω (6.3.100)
 
and recall the Kelvin matrix-valued fundamental solution E = E jk 1≤ j,k ≤n for the
Stokes system in Rn from (6.2.1), and the pressure vector q = (q j )1≤ j ≤n from (6.2.2).
For the clarity of the exposition, assume first that ∂Ω is unbounded. Then for each
index r ∈ {1, . . . , n} and each index γ ∈ {1, . . . , n} we may use (6.3.100), (6.2.266),
and Lemma 2.2.1 to compute (with all pointy brackets indicating duality in the sense
of [69, Theorem 4.6.1])
 mod  κ−n.t.    . 
(∂r wγ )(x) = ∂r Dλ u ∂Ω (x) − ∂r 𝒮 ∂νλ (
u, π) (x)
γ γ
   κ−n.t.  
= (∂k E jγ )(x − ·) ∂Ω , ∂τr k u ∂Ω j
   κ−n.t.  
+ λ (∂j Ekγ )(x − ·) ∂Ω , ∂τr k u ∂Ω j
   κ−n.t.  
+ qγ (x − ·) ∂Ω , ∂τ j r u ∂Ω j
  .  
− (∂r Eγ j )(x − ·) ∂Ω , ∂νλ (
u, π) j (6.3.101)

at each point x ∈ Ω. On account of the last line in (6.3.91) this further shows that, at
each point x ∈ Ω,
760 6 Green Formulas and Layer Potential Operators for the Stokes System
  . 
(∂r wγ )(x) = (∂k E jγ )(x − ·) ∂Ω , ∂τr k u j
  . 
+ λ (∂j Ekγ )(x − ·) ∂Ω , ∂τr k u j
  . 
+ qγ (x − ·) ∂Ω , ∂τ j r u j
  .  
− (∂r Eγ j )(x − ·) ∂Ω , ∂νλ (
u, π) j

= (∂r uγ )(x), (6.3.102)

with the last equality provided by (6.3.4). From (6.3.102) we then conclude that
∇ w = ∇u in Ω, which shows that the difference cu,π  := u  − w is a Cn -valued locally
constant function in Ω. This concludes the proof of (6.3.94) in the case when ∂Ω is
unbounded. Note that, in the case when ∂Ω is unbounded, the first equality in formula
(6.3.95) is directly implied by (6.3.5) and the last line in (6.3.91). The second equality
in (6.3.95) is then a consequence of the first, (6.3.91), and [69, Lemma 11.10.4].
The proofs of (6.3.94)-(6.3.95) in the case when ∂Ω is bounded is carried out in a
similar fashion.
In addition, from Theorem 6.2.7 we see that (6.3.96)-(6.3.97) are  true when
p ∈ (1, ∞). Finally, the validity of (6.3.98)-(6.3.99) in the range p ∈ n−1 n , 1 may
be justified based on (6.3.94) reasoning as in the proof of item (d) in Theorem 2.2.7,
while the validity of (6.3.98)-(6.3.99) in the range p ∈ (1, ∞) may be justified based
on (6.3.96) by reasoning as in the proof of the last claim in Theorem 1.8.19. 
We conclude this section by complementing the operator identities from item
(xi) in Theorem 6.2.4 and Theorem 6.2.8 by now considering similar formulas on
Hardy spaces and Hardy-based Sobolev spaces. We do this in two installments, in
Theorems 6.3.7-6.3.8 below.
Theorem 6.3.7 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a UR domain. Abbreviate
σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
normal to .Ω. Finally, fix λ ∈ C and consider the Stokes boundary layer potential
operators ∂νλ (Dλ, Pλ ), Kλ , Kλ# , S, associated with λ and Ω as in (6.2.256), (6.2.259),
(6.2.250), and (6.2.262). Then for any two exponents
 
p ∈ n−1 n , 1 and q ∈ (1, ∞) (6.3.103)

the following operator identities hold:

1   1   .λ   q, p n
2 I + Kλ ◦ − 2 I + Kλ = S ◦ ∂ν (Dλ, Pλ ) on H1 (∂Ω, σ) , (6.3.104)
1    .   n
2I + Kλ# ◦ − 12 I + Kλ# = ∂νλ (Dλ, Pλ ) ◦ S on H p (∂Ω, σ) , (6.3.105)
 n
S ◦ Kλ# = Kλ ◦ S on H p (∂Ω, σ) , (6.3.106)
 .λ   .λ q, p  n
Kλ# ◦ ∂ν Dλ = ∂ν (Dλ, Pλ ) ◦ Kλ on H1 (∂Ω, σ) . (6.3.107)
6.3 Other Integral Representations and Fatou-Type Results for the Stokes System 761

Proof All claims may be justified in a manner analogous to the proof of The-
orem 2.3.18 (cf. also the proof of item (xi) in Theorem 6.2.4), making use of
Theorem 6.2.11, Theorem 6.2.10, and the integral representation formulas from
Theorem 6.3.5. 

Here is the final installment of composition identities, now also involving modified
layer potentials for the Stokes system on Hardy spaces.

Theorem 6.3.8 Pick n ∈ N with n ≥ 2 and assume Ω ⊆ Rn is an NTA domain with

an Ahlfors regular boundary. Denote by ν the geometric measure theoretic outward
unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Select a number λ ∈ C and fix an
exponent  
p ∈ n−1n , 1 . (6.3.108)
Bring back the following operators associated with the Stokes system in the set Ω,
.  mod 
for the given parameter λ: S from (6.2.264), ∂νλ Dλ , Pλ from (6.2.257), Kλ from
mod

#
(6.2.260), and Kλ from (6.2.250). Then, in relation to these, the following results are
valid.
 n
(1) Given any f ∈ H p (∂Ω, σ) , at σ-a.e. point on ∂Ω one has
1      .  mod   
2 I + Kλ
#
− 12 I + Kλ# f = ∂νλ Dλ , Pλ S f (6.3.109)

and there exists c f, which is the nontangential trace on ∂Ω of some Cn -valued
locally constant function in Ω, such that
  mod  
S Kλ# f = Kλ S f + c f at σ-a.e. point on ∂Ω. (6.3.110)
mod
(2) Recall the operator Kλ from (6.2.216) and (6.2.220). Then, given any distri-
 .p n
bution f ∈ H1 (∂Ω, σ) , at σ-a.e. point on ∂Ω one has
 .  mod    .  mod    mod 
Kλ# ∂νλ Dλ , Pλ f = ∂νλ Dλ , Pλ Kλ f (6.3.111)

and there exists c f, which is the nontangential trace on ∂Ω of some Cn -valued
locally constant function in Ω, with the property that at σ-a.e. point on ∂Ω one
has
1 
mod   mod 
 .   
I + K − 1
I + K  = S ∂νλ D mod , Pλ f + c .
f (6.3.112)
2 λ 2 λ λ f

Proof The desired conclusions may be justified by reasoning as in the proof of

Theorem 2.3.11, now making use of the integral representation formulas from The-
orem 6.3.6 and availing ourselves of the functional analytic properties enjoyed by
the boundary layer potential operators associated with the Stokes system on Hardy
spaces from Theorem 6.2.11. 
762 6 Green Formulas and Layer Potential Operators for the Stokes System

6.4 Layer Potentials for the Stokes System on Besov,

Triebel-Lizorkin, and Weighted Sobolev Spaces

The vast majority of the results from §4 pertaining to the action of boundary layer
potentials associated with generic constant coefficient homogeneous second-order
systems on Besov, Triebel-Lizorkin, and weighted Sobolev spaces have natural coun-
terparts for the Stokes system16 as well. Here, our goal is to briefly elaborate on this
topic. Our first result of this nature reads as follows.

Theorem 6.4.1 Suppose Ω ⊆ Rn (where n ∈ N satisfies n ≥ 2) is a UR domain with

compact boundary and abbreviate σ := H n−1 ∂Ω. Also, fix λ ∈ C arbitrary and
recall the boundary layer potential operators Kλ and Kλ# from (6.2.95), (6.2.98).
Then the operator Kλ , originally acting on Lebesgue spaces on ∂Ω (cf. item (iii)
in Theorem 6.2.4), extends uniquely to linear and bounded mappings
 p,q n  p,q n
Kλ : Bs (∂Ω, σ) −→ Bs (∂Ω, σ) ,
    (6.4.1)
n ,∞ ,
p ∈ n−1 q ∈ (0, ∞], (n − 1) p1 − 1 + < s < 1,

and
 p,q n  p,q n
Kλ : Fs (∂Ω, σ) −→ Fs (∂Ω, σ) ,
     1  (6.4.2)
p ∈ n−1
n ,∞ , q ∈ n−1
n ,∞ , (n − 1) min{p,q } − 1 + < s < 1.

Also, the operator Kλ# , originally acting on Lebesgue spaces on ∂Ω (cf. item (iii)
in Theorem 6.2.4), extends uniquely to linear and bounded mappings
 p,q n  p,q n
Kλ# : B−s (∂Ω, σ) −→ B−s (∂Ω, σ)
 (6.4.3)
provided s ∈ (0, 1), p ∈ n−1 n−s , ∞], q ∈ (0, ∞],

and
 p,q n  p,q n
Kλ# : F−s (∂Ω, σ) −→ F−s (∂Ω, σ) ,
  (6.4.4)
n−s , ∞),
provided s ∈ (0, 1), p ∈ n−1 q ∈ n−1
n−s , ∞].

In all cases, various choices of the exponents yield operators which are compatible
with one another. Finally, if p, q, p , q ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q
and s ∈ (0, 1), then

p, q
[B−s (∂Ω,σ)] n Kλ# f, g = [B−s
p ,q
[B s
p, q
(∂Ω,σ)] n
  [B p , q (∂Ω,σ)]n
(∂Ω,σ)] n f , Kλ g s
 p,q n  p ,q n (6.4.5)

for each f ∈ B−s (∂Ω, σ) and g ∈ Bs (∂Ω, σ) ,

16 due to its special algebraic format, the Stokes system does not fit directly into the category of
generic constant coefficient homogeneous second-order systems
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 763

and
p, q
[F−s (∂Ω,σ)] n Kλ# f, g p ,q
[Fs
= [F−s
p, q
(∂Ω,σ)] n
  [F p , q (∂Ω,σ)]n
(∂Ω,σ)] n f , Kλ g s
 p,q n  p ,q n (6.4.6)
for each f ∈ F−s (∂Ω, σ) and g ∈ Fs (∂Ω, σ) .

Proof All claims are justified by reasoning as in the proofs of Theorem 4.1.1 and
Theorem 4.1.5, bearing in mind (6.2.271), item (1) in Theorem 6.2.11, and item (iii)
in Theorem 6.2.4. 

We next describe the smoothing effect of the boundary-to-domain double layer

potential operator associated with the Stokes system, measured on weighted Sobolev,
Besov, and Triebel-Lizorkin scales.

Theorem 6.4.2 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set whose boundary

is compact and Ahlfors regular. Abbreviate σ := H n−1 ∂Ω and pick λ ∈ C. In this
setting, recall the boundary-to-domain double layer Dλ for the Stokes system from
(6.2.17), and the corresponding potential for the pressure Pλ from (6.2.21). Finally,
fix  
p ∈ (1, ∞), s ∈ (0, 1), and a := 1 − s − p1 ∈ − p1 , 1 − p1 . (6.4.7)
Then the following operators are well-defined, linear, and bounded:
 p, p n  1, p n
Dλ : Bs (∂Ω, σ) −→ Wa (Ω)bdd ∩ Ker Δ2, (6.4.8)
 p, p n   ap n
Pλ : Bs (∂Ω, σ) −→ L p Ω, δ∂Ω L n ∩ Ker Δ, (6.4.9)
 p, p n  1, p n
𝒮 : Bs−1 (∂Ω, σ) −→ Wa (Ω)bdd ∩ Ker Δ2, (6.4.10)
 p, p n  p  ap n
Q : Bs−1 (∂Ω, σ) −→ Lbdd Ω, δ∂Ω L n ∩ Ker Δ. (6.4.11)

In addition, assuming
1 
n−1
n < p < ∞, (n − 1) p −1 + < s < 1, and with a := 1 − s − p1 , (6.4.12)

the operators
 p, p n  1, p n
Dλ : Bs (∂Ω, σ) −→ Wa, (Ω) , (6.4.13)
 p, p n  p ap n
Pλ : Bs (∂Ω, σ) −→ L Ω, δ∂Ω L n , (6.4.14)
 p, p n  1, p n
𝒮 : Bs−1 (∂Ω, σ) −→ Wa, (Ω) , (6.4.15)
 p, p n  p ap n
Q : Bs−1 (∂Ω, σ) −→ L Ω, δ∂Ω L n , (6.4.16)

are well-defined, linear, and bounded, provided Ω is bounded.

Finally, strengthen the original hypotheses by assuming this time that Ω is an
(ε, δ)-domain whose boundary is a compact UR set with the additional property
764 6 Green Formulas and Layer Potential Operators for the Stokes System
 
that H n−1 ∂Ω \ ∂∗ Ω = 0. Then, whenever s, p, s are as in (6.4.7), the double layer
operator Dλ , acting in the context of (6.4.8), satisfies the jump-formula
 p, p n
TrΩ→∂Ω ◦ Dλ = 12 I + Kλ on Bs (∂Ω, σ) , (6.4.17)
 1, p n  p, p n
where TrΩ→∂Ω : Wa (Ω)bdd → Bs (∂Ω, σ) is the boundary trace operator
 p, p n
from [69, Theorem 8.3.6], I denotes the identity operator on Bs (∂Ω, σ) , and
 p, p n
Kλ is the boundary-to-boundary double layer operator acting on Bs (∂Ω, σ) as
in (6.4.1) of Theorem 6.4.1.

Proof The claim pertaining to (6.4.8) is seen from Theorem 4.2.1, much as in the
proof of Theorem 4.2.3 (recalling from (6.2.84) that ∇Dλ annihilates all constant
vectors). As regards the operator Pλ in (6.4.9), upon noting that thanks to (6.2.6)-
 n
(6.2.7) for each f ∈ L 1 (∂∗ Ω, σ) and each x ∈ Ω we may express
∫
Pλ f(x) = −(1 + λ) ν j (y)(∂j qk )(x − y) fk (y) dσ(y) (6.4.18)
∂∗ Ω
∫
= −(1 + λ) ν j (y)(∂k q j )(x − y) fk (y) dσ(y)
∂∗ Ω
∫     
= (1 + λ) ν j (y)∂yk q j (x − y) − νk (y)∂y j q j (x − y) fk (y) dσ(y),
∂∗ Ω

the desired conclusion follows from Corollary 4.2.4. The claim about the operator
(6.4.10) is justified by reasoning much as in the proof of Theorem 4.2.10 (bearing
in mind (6.2.10)). Next, if 𝒮Δ denotes the boundary-to-domain single layer potential
operator associated with L := Δ, the Laplacian in Rn , and the set Ω, then we extend
the action of Q, originally considered as in (6.2.14), to any vector distribution on
∂Ω according to

Q f = −div𝒮Δ f = −∂j 𝒮Δ f j in Ω,
  n (6.4.19)
for each f = ( f j )1≤ j ≤n ∈ Lip (∂Ω) .

From this and Theorem 4.2.10 it follows that the operator Q is well defined, linear,
and bounded in the context of (6.4.11).
When p, s, a are as in (6.4.12), the fact that the operators (6.4.13)-(6.4.16) are well-
defined, linear and bounded may be justified much as in the proof of Theorem 4.2.3
and Theorem 4.2.10.
Finally, the claim regarding the validity of the jump-formula (6.4.17) may be
established by reasoning as in the proof of Theorem 4.2.5, based on what we have
proved so far and item (iv) in Theorem 6.2.4. 

We now turn to the issue of boundedness of hydrostatic layer potentials on Besov

and Triebel-Lizorkin scales.
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 765

Theorem 6.4.3 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an (ε, δ)-domain whose

boundary is a compact Ahlfors regular set, and abbreviate σ := H n−1 ∂Ω. Also,
fix λ ∈ C and recall the boundary-to-domain double layer Dλ for the Stokes system
from (6.2.17), and the corresponding potential for the pressure Pλ from (6.2.21).
Then the following assertions are true.
(1) The following operators are well-defined, linear, and bounded:
 p,q n p,q
n
Dλ : Bs (∂Ω, σ) −→ B 1 (Ω)bdd ,
s+ p
 p,q n p,q
n
Pλ : Bs (∂Ω, σ) −→ B 1 (Ω)bdd ,
s+ p −1
 p,q n p,q
n
𝒮 : Bs−1 (∂Ω, σ) −→ B 1 (Ω)bdd , (6.4.20)
s+ p
 p,q n p,q
n
Q : Bs−1 (∂Ω, σ) −→ B 1 (Ω)bdd ,
s+ p −1
 
if n−1
n < p < ∞, (n − 1) p1 − 1 + < s < 1, 0 < q ≤ ∞.

(2) The following operators are well-defined, linear, and bounded:

 p, p n p,q
n
Dλ : Bs (∂Ω, σ) −→ F 1 (Ω)bdd ,
s+ p
 p, p n p,q
n
Pλ : Bs (∂Ω, σ) −→ F 1 (Ω)bdd ,
s+ p −1
 p, p n p,q
n
𝒮 : Bs−1 (∂Ω, σ) −→ F 1 (Ω)bdd , (6.4.21)
s+ p
 p, p n p,q
n
Q : Bs−1 (∂Ω, σ) −→ F 1 (Ω)bdd ,
s+ p −1
 
if n−1
n < p < ∞, (n − 1) p1 − 1 + < s < 1, n
n+s+1/p < q ≤ ∞.

(3) Corresponding to taking s := 2 − p1 and q := 2 in (4.3.2), one has the following

well-defined, linear, and bounded mappings

p, p
n  p n
Dλ : B (∂Ω, σ) −→ h2 (Ω)bdd ,
2− p1
p, p
n  p n
Pλ : B (∂Ω, σ) −→ h1 (Ω)bdd ,
2− p1
p, p
n  p n (6.4.22)
𝒮: B (∂Ω, σ) −→ h2 (Ω)bdd ,
1− p1
p, p
n  p n
Q: B (∂Ω, σ) −→ h1 (Ω)bdd ,
1− p1

provided n
n+1 < p < 1,
p
where hk (Ω), k ∈ N, is the scale of local Hardy-based Sobolev spaces in Ω (cf.
[69, (9.2.43)]).
766 6 Green Formulas and Layer Potential Operators for the Stokes System

(4) Strengthen the original hypotheses on Ω by assuming that ∂Ω is actually a UR

set. Then, as a limiting case of (6.4.20)-(6.4.21), formally corresponding to
making s := 0, the operators
 n p,q
n
Dλ : L p (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
p
 n p,q
n
Pλ : L p (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
p −1
 p n p,q
n
(6.4.23)
𝒮 : L−1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
p
 p n p,q
n
Q : L−1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
p −1

provided 1 < p < ∞ and p ≤ q ≤ ∞,

as well as the operators

 n p,q
n
Dλ : L p (∂∗ Ω, σ) −→ F 1 (Ω)bdd ,
p
 n p,q
n
Pλ : L p (∂∗ Ω, σ) −→ F 1 (Ω)bdd ,
p −1
 p n p,q
n
𝒮 : L−1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd , (6.4.24)
p
 p n p,q
n
Q : L−1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd ,
p −1

provided 1 < p < ∞ and n

n+1/p < q ≤ ∞,

are all well-defined, linear, and bounded. In particular, corresponding to the

case p = q = 2, it follows that (recall (A.0.86) and [69, (9.2.22)])
 n  1/2  n
Dλ : L 2 (∂∗ Ω, σ) −→ Hbdd (Ω) ,
 2 n  −1/2  n
Pλ : L (∂∗ Ω, σ) −→ Hbdd (Ω) ,
 2 n  1/2  n (6.4.25)
𝒮 : L−1 (∂∗ Ω, σ) −→ Hbdd (Ω) ,
 2 n  −1/2  n
Q : L−1 (∂∗ Ω, σ) −→ Hbdd (Ω) ,

are well-defined, linear, and bounded operators. Furthermore, as a limiting case

of (6.4.20)-(6.4.21), formally corresponding to making s := 1, the operators
 p n p,q
n
Dλ : L1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
1+ p
 p n p,q
n
Pλ : L1 (∂∗ Ω, σ) −→ B 1 (Ω)bdd ,
p (6.4.26)
 n p,q
n
𝒮 : L p (∂Ω, σ) −→ B 1 (Ω)bdd ,
1+ p

provided 1 < p < ∞ and p ≤ q ≤ ∞,

6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 767

as well as
 p n p,q
n
Dλ : L1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd ,
1+ p
 p n p,q
n
Pλ : L1 (∂∗ Ω, σ) −→ F 1 (Ω)bdd ,
p
 n p,q
n
𝒮 : L p (∂Ω, σ) −→ F 1 (Ω)bdd , (6.4.27)
1+ p
 n p,q
n
Q : L p (∂Ω, σ) −→ F 1 (Ω)bdd ,
p

provided 1 < p < ∞ and n

n+1+1/p < q ≤ ∞,
are well-defined, linear, and bounded. In particular, corresponding to p = q = 2,
it follows that (recall (A.0.86) and [69, (9.2.22)])
 n  3/2  n
Dλ : L12 (∂∗ Ω, σ) −→ Hbdd (Ω) ,
 2 n  1/2  n
Pλ : L1 (∂∗ Ω, σ) −→ Hbdd (Ω) ,
 n  3/2  n (6.4.28)
𝒮 : L 2 (∂Ω, σ) −→ Hbdd (Ω) ,
 2 n  1/2  n
Q : L (∂Ω, σ) −→ Hbdd (Ω) ,

are well-defined, linear, and bounded operators.

(5) Continue to enforce the additional
 assumption that ∂Ω is a UR set and also
suppose that H n−1 ∂Ω \ ∂∗ Ω = 0 (e.g., any NTA domain with a compact
Ahlfors regular boundary will do). Then the jump-formula
 p,q n
TrΩ→∂Ω ◦ Dλ = 12 I + Kλ on Bs (∂Ω, σ) , whenever
  (6.4.29)
n < p < ∞, (n − 1) p1 − 1 + < s < 1, 0 < q ≤ ∞,
n−1

 p,q n  p,q n
holds if Dλ : Bs (∂Ω, σ) → Bs+1/p (Ω)bdd is the double layer potential
 p,q n  p,q n
operator considered in (6.4.20), TrΩ→∂Ω : Bs+1/p (Ω)bdd → Bs (∂Ω, σ)
is the boundary trace operator from [69,
 p,q(9.4.91) in
 nitem (ii) of Theorem 9.4.5],
I denotes the identity operator on Bs (∂Ω, σ) , and Kλ is the boundary-
 p,q n
to-boundary double layer operator acting on Bs (∂Ω, σ) as in (6.4.1) of
Theorem 6.4.1. Furthermore, the jump-formula
 p, p n
TrΩ→∂Ω ◦ Dλ = 12 I + Kλ on Bs (∂Ω, σ) , whenever
  (6.4.30)
n < p < ∞, (n − 1) p1 − 1 + < s < 1, n+s+1/p < q ≤ ∞,
n−1 n

 p,q n  p, p n
holds if, this time, TrΩ→∂Ω : Fs+1/p (Ω)bdd → Bs (∂Ω, σ) is the
boundary
 p, ptrace operator
n from [69, (9.4.93)
n in item (ii) of Theorem 9.4.5],
p,q
Dλ : Bs (∂Ω, σ) → Fs+1/p (Ω)bdd is the double layer potential opera-
 p, p n
tor considered in (6.4.21), I denotes the identity operator on Bs (∂Ω, σ)
768 6 Green Formulas and Layer Potential Operators for the Stokes System

and,
 p, pfinally, Kλn is the boundary-to-boundary double layer operator acting on
Bs (∂Ω, σ) as in (6.4.1) of Theorem 6.4.1.

Proof The mapping properties for the double layer potential operator Dλ may be
dealt with as in the treatment of items (1)-(4) in the proof of Theorem 4.3.1. As
regards the operator Pλ , from (6.4.18) and (6.2.2) we see that for each function
 n
f = ( fk )1≤k ≤n ∈ L 1 (∂∗ Ω, σ) we have
∫ 
   
Pλ f(x) = (1 + λ) ν j (y)∂yk q j (x − y) − νk (y)∂y j q j (x − y) fk (y) dσ(y)
∂∗ Ω
 
= −(1 + λ)∂j U jk fk (x), for all x ∈ Ω, (6.4.31)

where, for each j, k ∈ {1, . . . , n}, the integral operator U jk is associated as in

(4.2.76) to the Laplace operator L := Δ and the set Ω. From this, item (5) in
Theorem 4.3.1, and [69, (9.2.8)] then all desired mapping properties for Pλ follow.
Next, the mapping properties of the operator 𝒮 are justified by the same argument
used in the proof of Theorem 4.3.3, while the mapping properties of the operator
Q are seen from (6.4.19) and Theorem 4.3.3. Finally, the jump-formulas claimed in
(6.4.29)-(6.4.30) may be established based on the mapping properties for Dλ in the
context of (6.4.20)-(6.4.21), [69, Theorem 9.4.5], Theorem 6.4.1, and (6.2.109). 

Remark 6.4.4 The boundary-to-boundary version of the single layer potential op-
erator S associated with the Stokes system (cf. (6.2.112)) satisfies the mapping prop-
erties described in Theorem 4.3.4 (with M := n). Indeed, this may be justified in the
manner as in the proof of Theorem 4.3.4, now relying on item (7) of Theorem 6.2.11,
item (vi) of Theorem 6.2.4, and the mapping properties of the boundary-to-domain
version of the single layer potential operator 𝒮 from Theorem 6.4.3.

We next introduce and study the conormal derivative operators ∂νλ , with λ ∈ C,
associated with the Stokes system in the context of Besov and Triebel-Lizorkin
spaces.

Proposition 6.4.5 Let Ω ⊆ Rn be a bounded (ε, δ)-domain with an Ahlfors regular

boundary, and abbreviate σ := H n−1 ∂Ω. Also, pick p, q ∈ (1, ∞) along with
  −1   −1
s ∈ (0, 1) then set p := 1 − p1 and q := 1 − q1 . Finally, fix λ ∈ C and recall
the system Lλ from (6.1.3). In this context, define the conormal derivative operator
∂νλ associated with the Stokes system as the mapping from the space

 p,q n  p ,q ∗ n
u, π; f ) ∈ Bs+1/p (Ω) ⊕ Bs+1/p−1 (Ω) ⊕ B1−s+1/p (Ω)
p,q
( :
!
Lλ u − ∇π = f Ω in Ω (6.4.32)

(where
 p,q the convention
n introduced in [69, (9.5.1)] has been used), with values in
u, π; f ) from (6.4.32) the functional
Bs−1 (∂Ω, σ) , associating to each triplet (
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 769
  n∗  p,q n
p ,q
u, π; f ) ∈
∂νλ ( B1−s (∂Ω, σ) = Bs−1 (∂Ω, σ) (6.4.33)

αβ
acting according to (recall that the coefficients a jk (λ) are as in (6.1.1), and that the
summation convention over repeated indices is presently in effect)

p ,q
([B1−s (∂Ω,σ)] n )∗
u, π; f ), ϕ
∂νλ ( p ,q
[B1−s (∂Ω,σ)] n

αβ  + f, Φ

:= a jk (λ)∂k uβ, ∂j Φα − π, divΦ
(6.4.34)
 p ,q n  p ,q n
for all ϕ ∈ B1−s (∂Ω, σ)  = (Φα )α ∈
and Φ B1−s+1/p (Ω)
 = ϕ,
satisfying TrΩ→∂Ω Φ 

where the first two sets of brackets in the right side of the first line above are
p,q p ,q
understood as the duality pairing between Bs+1/p−1 (Ω) and B−s+1/p (Ω) (cf. [69,
(9.2.141)]) and the final set of brackets in the right side of the first line above is the
 p ,q ∗ n  p ,q n
canonical duality pairing between B1−s+1/p (Ω) and B1−s+1/p (Ω) .
Then the conormal derivative operator ∂νA considered in (6.4.32)-(6.4.34) is well
defined, linear, and bounded in the sense that there exists a constant C ∈ (0, ∞) with
the property that
 λ 
∂ (  
ν u, π; f ) [B p, q (∂Ω,σ)] n (6.4.35)
s−1
 
≤ C  u[B p, q (Ω)] n + πB p, q (Ω) +  f[(B p , q (Ω))∗ ] n
s+1/p s+1/p−1 1−s+1/p

u, π; f ) belonging to the domain of ∂νλ (cf. (6.4.32)).

for each triplet (
 p ,q n
Moreover, for each triplet ( u, π; f ) as in (6.4.32) and each w ∈ B1−s+1/p (Ω)
one has the following generalized “half” Green’s formula:

p ,q
([B1−s (∂Ω,σ)] n )∗
u, π; f ), TrΩ→∂Ω w
∂νλ ( p ,q
[B1−s (∂Ω,σ)] n

αβ
= p, q
B s+1/p−1 (Ω) a jk (λ)∂k uβ, ∂j wα p ,q
B−s+1/p (Ω)

− B p, q (Ω) π, divw p ,q
B−s+1/p (Ω)
s+1/p−1

+ [(B p , q (Ω))∗ ] n
f, w p ,q
[B1−s+1/p (Ω)] n
(6.4.36)
1−s+1/p

where (uβ )1≤β ≤n are the scalar components of u and (wα )1≤α ≤n are the scalar
components of w.
u, π; f ) belonging to (6.4.32), and (w,
Furthermore, for any two triplets, (  ρ; g )
belonging to the analogue of (6.4.32) with p, q, s replaced by p , q , 1 − s, one has
the following generalized “full” (or “symmetric”) Green’s formula:
770 6 Green Formulas and Layer Potential Operators for the Stokes System

p ,q
([B1−s (∂Ω,σ)] n )∗
u, π; f ), TrΩ→∂Ω w
∂νλ ( p ,q
[B1−s (∂Ω,σ)] n

− ([Bsp, q (∂Ω,σ)]n )∗ ∂νλ (w,

 ρ; g ), TrΩ→∂Ω u p, q
[B s (∂Ω,σ)] n

= p, q
B s+1/p−1 (Ω) u, ρ
div p ,q
B−s+1/p (Ω)
− B p, q (Ω) π, divw p ,q
B−s+1/p (Ω)
s+1/p−1

+ [(B p , q (Ω))∗ ] n
f, w p ,q
[B1−s+1/p (Ω)] n
1−s+1/p

− [(B p, q (Ω))∗ ] n g, u [B s+1/p (Ω)] n .

p, q (6.4.37)
s+1/p

Finally, similar results are valid on the scale of Triebel-Lizorkin spaces (in the
spirit of [69, Proposition 9.5.2] with A := F) and the scale of weighted Sobolev
spaces (in the spirit of [69, Proposition 8.5.3]).

Proof All claims are justified by reasoning as in the proofs of [69, Proposition 8.5.3]
and [69, Proposition 9.5.2]. 

Much as in [69, Remark 9.5.3] (see also the subsequent comments), it is possible
to define the conormal derivative in the more general setting described below.

Remark 6.4.6 Retain the assumptions on Ω, λ, p, q, s from Proposition 6.4.5, along

with the conventions made there. For A ∈ {B, F}, set q∗ := q if A = B and q∗ := p
if A = F, and denote by q∗ the Hölder conjugate exponent of q∗ . In addition,
fix an arbitrary cutoff function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near ∂Ω. Recall [69,
n

Convention 8.3.7] (cf. also (A.0.217)). Assume

 p,q n p,q
u = (uβ )1≤β ≤n ∈ As+1/p (Ω)bdd , π ∈ As+1/p−1 (Ω)bdd ,
 p ,q ∗ n (6.4.38)
f ∈ A1−s+1/p (Ω)

with the property that

 n
Lλ u − ∇π = f Ω in D (Ω) (6.4.39)

and define
 p,q M
&
u, π; f ) := ∂νλ (
∂νλ ( uψ, πψ ; fψ ) ∈ Bs−1∗ (∂Ω, σ) (6.4.40)
with
 p,q n p,q
uψ := ψ u ∈ As+1/p (Ω) , πψ := ψπ ∈ As+1/p−1 (Ω)
 p ,q ∗ n
and fψ ∈ A1−s+1/p (Ω) given by
 
αβ αβ αβ
fψ := a jk (λ)(∂j ∂k ψ)uβ + a jk (λ)(∂j ψ)(∂k uβ ) + a jk (λ)(∂k ψ)(∂j uβ )
1≤α ≤n
+ψ f + π∇ψ.
(6.4.41)
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 771

Then the above definition is meaningful, and does not depend on the particular
 p ,q n
cutoff function ψ. Furthermore, for each w = (wα )1≤α ≤n ∈ A1−s+1/p (Ω) which
vanishes outside a bounded subset of Ω the generalized “half” Green’s formula

p , q∗ u, π; f ), TrΩ→∂Ω w
∂νλ ( p , q∗
([B1−s (∂Ω,σ)] n )∗ [B1−s (∂Ω,σ)] n

αβ
= A p, q (Ω) a jk (λ)ψ∂k uβ, ∂j wα p ,q
A−s+1/p (Ω)
s+1/p−1

− A p, q (Ω) ψπ, divw p ,q

A−s+1/p (Ω)
s+1/p−1

+ [(A p , q (Ω))∗ ] n
f, w p ,q
[A1−s+1/p (Ω)] n
(6.4.42)
1−s+1/p

holds for each cutoff function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near both ∂Ω and the support
n


of w.
 ρ; g ) satisfying analogous properties to (6.4.38)-
In addition, for any triplet (w,
(6.4.39) with p, q, s replaced by p , q , 1 − s, and such that both w and ρ vanish
outside a bounded subset of Ω, the generalized “full” Green’s formula

p , q∗ u, π; f ), TrΩ→∂Ω w
∂νλ ( p , q∗
([B1−s (∂Ω,σ)] n )∗ [B1−s (∂Ω,σ)] n

− ([Bsp, q∗ (∂Ω,σ)]n )∗ ∂νλ (w,

 ρ; g ), TrΩ→∂Ω u p, q∗
[B s (∂Ω,σ)] n

= A p, q (Ω) u, ρ
div p ,q
A−s+1/p (Ω)
− A p, q (Ω) π, divw p ,q
A−s+1/p (Ω)
s+1/p−1 s+1/p−1

+ [(A p , q (Ω))∗ ] n
f, w p ,q
[A1−s+1/p (Ω)] n
1−s+1/p

− [(A p, q (Ω))∗ ] n g, u p, q

[A s+1/p (Ω)] n (6.4.43)
s+1/p

holds for each cutoff function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near both ∂Ω as well as the
n

supports of w and ρ.
Finally, similar results are valid on the scale of weighted Sobolev spaces.

Our next result deals with the conormal derivative of the pair consisting of the
hydrostatic double layer and the associated pressure potential.

Theorem 6.4.7 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set satisfying a

two-sided local John condition and whose boundary is compact and Ahlfors regular.
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward
unit normal to Ω. Also, fix λ ∈ C along with

1 < p < ∞, 0 < q ≤ ∞, 0 < s < 1. (6.4.44)

 
Then the operator ∂νλ Dλ, Pλ from item (ix) of Theorem 6.2.4 extends in a unique
fashion to a bounded linear mapping
772 6 Green Formulas and Layer Potential Operators for the Stokes System
 n    p,q n
 f −→ ∂νλ Dλ f, Pλ f ∈ Bs−1 (∂Ω, σ)
p,q
Bs (∂Ω, σ) (6.4.45)

and various choices of p, q, s as in (6.4.44) yield operators which are compatible

with one another. Moreover, if s ∈ (0, 1) and the exponents p, q, p , q ∈ (1, ∞) satisfy
1/p + 1/p = 1 = 1/q + 1/q then
   
∂ λ D f, P f , g [B p , q (∂Ω,σ)]n
p, q
[B s−1 (∂Ω,σ)] n ν λ λ
 
1−s

 λ
= [Bsp, q (∂Ω,σ)]n f , ∂ν Dλ g, Pλ g [B p , q (∂Ω,σ)]n (6.4.46)
−s
 p,q n  p ,q n
for each f ∈ Bs (∂Ω, σ) and g ∈ B1−s (∂Ω, σ) .

Finally, under the additional assumption that Ω is an (ε, δ)-domain, one has the
following compatibility result
     p,q n
∂νλ Dλ f, Pλ f = ∂νλ Dλ f, Pλ f; 0 for each f ∈ Bs (∂Ω, σ) , (6.4.47)

where the expression in the left-hand side is considered as in (6.4.45), and the
conormal derivative in the right-hand side is taken in either of the scenarios described
in Proposition 6.4.5.
Proof The same type of argument as in the proof of Theorem 4.3.5, now making use
of item (ix) of Theorem 6.2.4 and Proposition 6.4.5, yields all desired conclusions.
The jump-formula for the conormal derivative of the pair (𝒮, Q) is discussed in
our next theorem.
Theorem 6.4.8 Let Ω ⊆ Rn be an NTA domain with a compact Ahlfors regular
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Also, select λ ∈ C and pick

1 < p < ∞, 1 < q < ∞, 0 < s < 1. (6.4.48)

Then, with I denoting the identity operator, one has the jump-formula
     p,q n
∂νλ 𝒮 f, Q f; 0 = − 12 I + Kλ# f for each f ∈ Bs−1 (∂Ω, σ) , (6.4.49)

where the conormal derivative in the left-hand side of (6.4.49) is considered in either
of the scenarios described in Proposition 6.4.5.
Proof This parallels the proof of Theorem 4.3.6, now making use of item (vi) in
Theorem 6.2.4, Proposition 6.4.5, and Theorem 6.4.1. 
The basic Green-type integral representation formulas involving hydrostatic layer
potentials for functions belonging to Besov, Triebel-Lizorkin, and weighted Sobolev
spaces make the object of the theorem below.
Theorem 6.4.9 Let Ω ⊆ Rn be an (ε, δ)-domain with a compact Ahlfors regular
boundary. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 773
 n
theoretic outward unit normal to Ω. Also, fix some λ ∈ C and let u ∈ 𝒞∞ (Ω) and
π ∈ 𝒞∞ (Ω) solve the Stokes system in Ω (cf. (6.0.1)). Then the integral representation
formulas    
u = Dλ TrΩ→∂Ω u − 𝒮 ∂νλ ( u, π; 0) in Ω, (6.4.50)
and    
π = Pλ TrΩ→∂Ω u − Q ∂νλ (
u, π; 0) in Ω, (6.4.51)
hold with the understanding that if Ω is an exterior domain one also assumes that
there exists μ ∈ (1, ∞) such that
⨏
 
| u| + R|π| dL n = o(1) as R → ∞, (6.4.52)
B(0, μ R)\B(0,R)

in the case of (6.4.50), and

⨏
 
| u| + R|π| dL n = o(R) as R → ∞, (6.4.53)
B(0, μ R)\B(0,R)

in the case of (6.4.51), in any for the following scenarios:

(1) One assumes that

 p,q n p,q
u ∈ B 1 (Ω)bdd and π ∈ B 1 (Ω)bdd with
s+ p s+ p −1
(6.4.54)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,

and
 p,q n  p,q n
TrΩ→∂Ω : B 1 (Ω)bdd → Bs (∂Ω, σ) is the boundary trace
s+ p
operator from [69, (9.4.91) in item (ii) of Theorem 9.4.5] (hence
p,q n
TrΩ→∂Ω u ∈ Bs (∂Ω, σ) ), the layer potential operators Dλ , (6.4.55)
 p,q n
Pλ , 𝒮, Q are as in (6.4.20), and ∂νλ (u, π; 0) ∈ Bs−1 (∂Ω, σ) is
the conormal derivative defined as in Proposition 6.4.5 and further
extended in Remark 6.4.6.

(2) One assumes that

 p,q n p,q
u ∈ F 1 (Ω)bdd and π ∈ F 1 (Ω)bdd with
s+ p s+ p −1
(6.4.56)
1 < p < ∞, 0 < s < 1, 1 < q < ∞,

and
774 6 Green Formulas and Layer Potential Operators for the Stokes System
 p,q n  p, p n
TrΩ→∂Ω : F 1 (Ω)bdd → Bs (∂Ω, σ) is the boundary trace
s+ p
operator from [69, (9.4.93) in item (ii) of Theorem 9.4.5] (hence
p, p n
TrΩ→∂Ω u ∈ Bs (∂Ω, σ) ), the layer potential operators Dλ , (6.4.57)
 p, p n
Pλ , 𝒮, Q are as in (6.4.21), and ∂νλ (u, π; 0) ∈ Bs−1 (∂Ω, σ) is
the conormal derivative defined as in Proposition 6.4.5 and further
extended in Remark 6.4.6.
(3) One also assumes that Rn \ Ω is n-thick, that
 1, p n p  ap 
u ∈ Wa (Ω)bdd and π ∈ Lbdd Ω, δ∂Ω L n with
(6.4.58)
1 < p < ∞, 0 < s < 1, a := 1 − s − p1 ,

and
 1, p n  p, p n
TrΩ→∂Ω : Wa (Ω)bdd → Bs (∂Ω, σ) is the boundary trace
operator
 p, p from [69, Theorem 8.3.6] (hence TrΩ→∂Ω u belongs to
n
Bs (∂Ω, σ) ), the layer potential operators Dλ , Pλ , 𝒮, Q are
 p, p n (6.4.59)
as in (6.4.8)-(6.4.11) and, finally, ∂νλ (
u, π; 0) ∈ Bs−1 (∂Ω, σ) is
the conormal derivative defined as in the very last part of Proposi-
tion 6.4.5 and further extended in Remark 6.4.6.
Finally, if Ω is an exterior domain and in place of (6.4.52) one imposes the
weaker assumption
⨏
 
| u| + R|π| dL n = o(R) as R → ∞, (6.4.60)
B(0, μ R)\B(0,R)

then in place of (6.4.50) one now concludes that there exists a constant c ∈ Cn with
the property that
   
u = Dλ Tr Ω→∂Ω u − 𝒮 ∂νλ (
u, π; 0) + c in Ω, (6.4.61)

in any of the scenarios described in items (1)-(3) above.

Proof We shall carry out the proof in the case when Ω is assumed to be bounded,
since the case when Ω is an exterior domain largely proceeds along similar lines.
Consider the scenario specified in item (1). Fix a point x ∈ Ω and choose a scalar-
valued function θ ∈ 𝒞∞ (Rn ) with the property that θ = 0 identically
 on B(0, 1) and
θ = 1 identically on Rn \ B(0, 2). For each ε ∈ 0, 12 dist(x, ∂Ω) define θ ε : Rn → R
by setting
y − x
θ ε (y) := θ for every y ∈ Rn, (6.4.62)
ε
so that
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 775

θ ε ∈ 𝒞∞ (Rn ), 1 − θ ε ∈ 𝒞∞
c (Ω),
(6.4.63)
θ ε ≡ 0 on B(x, ε) and θ ε ≡ 1 on Rn \ B(x, 2ε).

Next, bring in the fundamental solution for the Stokes system E = (E jk )1≤ j,k ≤n with
entries as in (6.2.1), and recall the accompanying pressure vector q = (q j )1≤ j ≤n
from (6.2.2). Fix j ∈ {1, . . . , n} and introduce
   n
w := E jk (x − ·)θ ε 1≤k ≤n ∈ 𝒞∞ (Ω) and ρ := −q j (x − ·)θ ε ∈ 𝒞∞ (Ω).
(6.4.64)

Then a direct computation (based on (6.2.4)-(6.2.5) and (6.4.63)) gives that

divw = E jk (x − ·)(∂k θ ε ) ∈ 𝒞∞
c (Ω), (6.4.65)

and

Lλ w − ∇ρ = − 2(∂ E jk )(x − ·)(∂ θ ε ) + E jk (x − ·)(Δθ ε ) − q j (x − ·)(∂k θ ε )

− λ(∂k E j )(x − ·)(∂ θ ε ) + λE j (x − ·)(∂k ∂ θ ε ) (6.4.66)
1≤k ≤n
 p ,q n
is a vector with components in 𝒞∞  ∈
c (Ω). Furthermore, w B (Ω) and
1−s+ p1
p ,q
ρ∈B (Ω) where p , q are the Hölder conjugate exponents of p, q. Define next
−s+ p1

 n
g := Lλ w − ∇ρ ∈ 𝒞∞c (Ω) . (6.4.67)
 p,q ∗ n
When naturally regarded as a functional g ∈ Bs+1/p (Ω) via integral pairing
over Ω, this trivially satisfies
 n
Lλ w − ∇ρ = g Ω in D (Ω) . (6.4.68)

In addition, if (ν1, . . . , νn ) are the components of ν, then from the memberships in

(6.4.64), Proposition 6.4.5, and (6.1.10) we see that
   
αβ
∂νλ (w,
 ρ; g) = νr ar s (λ) ∂s wβ ∂Ω − να ρ ∂Ω . (6.4.69)
1≤α ≤n

Granted these properties, the generalized “full” Green’s formula (6.4.37) for Besov
spaces, applied with u, π as the statement of the theorem (which are currently assumed
 p ,q ∗ n
to be as in (6.4.54)), with f := 0 ∈ B (Ω)
1−s+1/p
 ρ, and g as
, and with w,
above, yields
776 6 Green Formulas and Layer Potential Operators for the Stokes System

p ,q
([B1−s (∂Ω,σ)] n )∗
∂νλ (
u, π; 0), TrΩ→∂Ω w p ,q
[B1−s (∂Ω,σ)] n

− ([Bsp, q (∂Ω,σ)]n )∗ ∂νλ (w,

 ρ; g), TrΩ→∂Ω u [B s
p, q
(∂Ω,σ)] n

= −B p, q (Ω) π, divw p ,q
B−s+1/p (Ω)
− [(B p, q (Ω))∗ ] n g, u p, q
[B s+1/p (Ω)] n
s+1/p−1 s+1/p

∫
= − D (Ω) π, divw D(Ω) − Lλ w − ∇ρ, u dL n . (6.4.70)
Ω

Note that, thanks to (6.4.64) and Proposition 4.2.8,

p ,q
([B1−s (∂Ω,σ)] n )∗
∂νλ (
u, π; 0),TrΩ→∂Ω w p ,q
[B1−s (∂Ω,σ)] n
  
= 𝒮 ∂νλ (
u, π; 0) (x). (6.4.71)
j

By virtue of (6.4.69), (6.4.64), and (6.2.17) we also have

p, q
([B s (∂Ω,σ)] n )∗ ∂νλ (w,
 ρ; g), TrΩ→∂Ω u [Bsp, q (∂Ω,σ)]n
∫  
αβ   
= νr ar s (λ) ∂s wβ ∂Ω − να ρ ∂Ω TrΩ→∂Ω u α dσ
∂∗ Ω
∫  
αβ
= − νr (y)ar s (λ)(∂s E jβ )(x − y) + να (y)q j (x − y) ×
∂∗ Ω
 
× TrΩ→∂Ω u α (y) dσ(y)
  
= Dλ TrΩ→∂Ω u (x). (6.4.72)
j

There remains to consider the last line of (6.4.70). In this regard, based on
(6.4.65)-(6.4.66) we may express
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 777
∫
− D (Ω) π , divw D(Ω) − Lλ w − ∇ρ, u dL n
Ω

= − D (Ω) E jk (x − ·), (∂k θ ε )π D(Ω)

+ D (Ω) 2(∂ E jk )(x − ·), (∂ θ ε )uk D(Ω)

− D (Ω) E jk (x − ·), (Δθ ε )uk D(Ω)

+ D (Ω) q j (x − ·), (∂k θ ε )uk D(Ω)

− D (Ω) λ(∂k E j )(x − ·), (∂ θ ε )uk D(Ω)

+ D (Ω) λE j (x − ·), (∂k ∂ θ ε )uk D(Ω)

=: I + II + III + IV + V + VI. (6.4.73)

 n
Observe that since ∇θ ε ∈ 𝒞∞c (Ω) and θ ε − 1 ∈ 𝒞∞
c (Ω), we may use (6.2.5) to
write

I = − D (Ω) E jk (x − ·), (∂k θ ε )π D(Ω)

 
= − D (Ω) E jk (x − ·), ∂k (θ ε − 1)π D(Ω)

+ D (Ω) E jk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= D (Ω) E jk (x − ·), (θ ε − 1)(∂k π) D(Ω) . (6.4.74)

In a similar fashion,

II + III = D (Ω) E jk (x − ·), (Δθ ε )uk + 2(∂ θ ε )(∂ uk ) D(Ω)

 
= D (Ω) E jk (x − ·), Δ (θ ε − 1)uk D(Ω)

− D (Ω) E jk (x − ·), (θ ε − 1)(Δuk ) D(Ω)

= D (Ω) Δ[E jk (x − ·)], (θ ε − 1)uk D(Ω)

− D (Ω) E jk (x − ·), (θ ε − 1)(Δuk ) D(Ω) . (6.4.75)

Next, given that u is divergence-free, we have

IV = D (Ω) q j (x − ·), (∂k θ ε )uk D(Ω)

 
= D (Ω) q j (x − ·), ∂k (θ ε − 1)uk D(Ω)

= − D (Ω) ∂k [q j (x − ·)], (θ ε − 1)uk D(Ω) . (6.4.76)

Finally, by once again using the fact that u is divergence-free we obtain

778 6 Green Formulas and Layer Potential Operators for the Stokes System
 
V + VI = − D (Ω) λE j (x − ·), ∂k (∂ θ ε )uk D(Ω)

+ D (Ω) λE j (x − ·), (∂k ∂ θ ε )uk D(Ω) = 0. (6.4.77)

Bearing in mind (6.2.4) and the fact that Δ

u = ∇π in Ω, from (6.4.74)-(6.4.77) we
conclude that

I + II + III + IV + V + VI

= D (Ω) δ jk δx, (θ ε − 1)uk D(Ω) = −u j (x). (6.4.78)

At this stage, the integral representation formula (6.4.50) follows from (6.4.70),
(6.4.71), (6.4.72), (6.4.87), and (6.4.78), on account of the arbitrariness of the index
j ∈ {1, . . . , n} and point x ∈ Ω.
Let us now turn our attention to the integral representation formula claimed in
(6.4.51). To get started, introduce
    n  p ,q n
 − ·)θ ε = qk (x − ·)θ ε 1≤k ≤n ∈ 𝒞∞ (Ω) ⊆ B
ω := q(x (Ω) , (6.4.79)
1−s+ p1

and observe that, thanks to (6.2.7) and (6.4.63),

divω = qk (x − ·)(∂k θ ε ) ∈ 𝒞∞
c (Ω), (6.4.80)

and

h := Lλ ω = 2∂ [qk (x − ·)](∂ θ ε ) + qk (x − ·)(Δθ ε ) + λ∂k [q j (x − ·)](∂j θ ε )
  n
+ λq j (x − ·)(∂k ∂j θ ε ) ∈ 𝒞∞c (Ω) . (6.4.81)
1≤k ≤n

 ∗ n
When naturally regarded as a functional h ∈
p,q
Bs+1/p (Ω) via integral pairing
over Ω, this trivially satisfies
 n
Lλ ω = h Ω in D (Ω) . (6.4.82)

In addition, from the memberships in (6.4.64), Proposition 6.4.5, (6.1.10), and (6.1.1)
we see that
   
∂νλ (ω,  = νr arαβ
 0; h) s (λ) ∂s ωβ ∂Ω
1≤α ≤n
   
= νs ∂s [qα (x − ·)] ∂Ω + λνβ ∂α [qβ (x − ·)] ∂Ω
1≤α ≤n 1≤α ≤n
 
= (1 + λ) νs ∂s [qα (x − ·)] ∂Ω , (6.4.83)
1≤α ≤n

where the last equality uses (6.2.6). The idea is now to invoke the version of the
generalized “full” Green’s formula (6.4.37) for Besov spaces, applied with u, π as
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 779

the statement of the theorem (which are currently assumed to be as in (6.4.54)),

 p ,q ∗ n
with f := 0 ∈ B (Ω)
1−s+1/p
 ρ := 0, and g := h as above.
, and with w := ω,
Specifically, we obtain

p ,q
([B1−s (∂Ω,σ)] n )∗
∂νλ (
u, π; 0), TrΩ→∂Ω ω p ,q
[B1−s (∂Ω,σ)] n

− ([Bsp, q (∂Ω,σ)]n )∗ ∂νλ (ω,  TrΩ→∂Ω u

 0; h), [B s
p, q
(∂Ω,σ)] n

= −B p, q (Ω) π, divω p ,q − [(B p, q (Ω))∗ ] n

 u
h, p, q
[B s+1/p (Ω)] n
s+1/p−1 B−s+1/p (Ω) s+1/p

∫
= − D (Ω) π, divω D(Ω) −  u dL n .
h, (6.4.84)
Ω

Note that thanks to (6.4.79) and the manner the action of the integral operator Q
from (6.2.14) is extended to distributions on ∂Ω (cf. (6.4.19)) we have
 
([B
p ,q ∂ λ (
(∂Ω,σ)] n )∗ ν
u, π; 0), TrΩ→∂Ω ω [B p , q (∂Ω,σ)]n = Q ∂νλ (
u, π; 0) (x).
1−s 1−s
(6.4.85)

On account of (6.4.83) and (6.2.21) we also have

p, q
([B s (∂Ω,σ)] n )∗ ∂νλ (ω,  TrΩ→∂Ω u [B p, q (∂Ω,σ)]n
 0; h), s
∫
 
= (1 + λ) νs (y)∂ys [qα (x − y)] TrΩ→∂Ω u α (y) dσ(y)
∂∗ Ω
 
= Pλ TrΩ→∂Ω u (x). (6.4.86)

Let us take a closer look at the last line of (6.4.70). For starters, use (6.4.65)-(6.4.66)
to expand
∫
− D (Ω) π , divω D(Ω) −  u dL n
h,
Ω

= − D (Ω) qk (x − ·), (∂k θ ε )π D(Ω)

− D (Ω) 2∂ [qk (x − ·)], (∂ θ ε )uk D(Ω)

− D (Ω) qk (x − ·), (Δθ ε )uk D(Ω)

− D (Ω) λ∂k [q j (x − ·)], (∂j θ ε )uk D(Ω)

− D (Ω) λq j (x − ·), (∂k ∂j θ ε )uk D(Ω)

=: I + II + III + IV + V . (6.4.87)
780 6 Green Formulas and Layer Potential Operators for the Stokes System
 n
Since from (6.4.63) we know that ∇θ ε ∈ 𝒞∞ c (Ω) and θ ε − 1 ∈ 𝒞∞
c (Ω), while
from (6.2.7) we see that ∂k [qk (x − ·)] = −δx , we may compute

I = − D (Ω) qk (x − ·), (∂k θ ε )π D(Ω)

 
= − D (Ω) qk (x − ·), ∂k (θ ε − 1)π D(Ω) + D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= D (Ω) ∂k [qk (x − ·)], (θ ε − 1)π D(Ω) + D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= D (Ω) δx, (θ ε − 1)π D(Ω) + D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= −π(x) + D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω) . (6.4.88)

Making use of (6.2.7), the fact that u is divergence-free, that Δ

u = ∇π, and that
∇θ ε ≡ 0 on B(x, ε) (cf. (6.4.63)), we also obtain
 
II + III = D (Ω) qk (x − ·), 2∂ (∂ θ ε )uk D(Ω)

− D (Ω) qk (x − ·), (Δθ ε )uk D(Ω)

= D (Ω) qk (x − ·), (Δθ ε )uk + 2(∂ θ ε )(∂ uk ) D(Ω)

 
= D (Ω) qk (x − ·), Δ (θ ε − 1)uk − (θ ε − 1)(Δuk ) D(Ω)

= D (Ω) Δ[qk (x − ·)], (θ ε − 1)uk D(Ω)

− D (Ω) qk (x − ·), (θ ε − 1)(Δuk ) D(Ω)

= − D (Ω) ∂k δx, (θ ε − 1)uk D(Ω) − D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= D (Ω) δx, (∂k θ ε )uk D(Ω) − D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω)

= − D (Ω) qk (x − ·), (θ ε − 1)(∂k π) D(Ω) . (6.4.89)

Lastly, given that u is divergence-free, we have

IV + V = − D (Ω) λ∂k [q j (x − ·)], (∂j θ ε )uk D(Ω)

 
− D (Ω) λq j (x − ·), ∂k (∂j θ ε )uk D(Ω) = 0. (6.4.90)

Collectively, (6.4.88)-(6.4.90) prove that

I + II + III + IV + V = −π(x). (6.4.91)

In concert with (6.4.84), (6.4.85), (6.4.86), and (6.4.87), this ultimately establishes
the integral representation formula in (6.4.51). This takes care of the claims made
in the context of item (1) in the statement of the theorem. Finally, the validity of
(6.4.50)-(6.4.51) in items (2)-(3) in the statement is proved in a similar fashion.
6.4 Stokes Layer Potentials on Besov, Triebel-Lizorkin, and Weighted Sobolev Spaces 781

At this stage, there remains to justify formula (6.4.61), working now under the
assumption that Ω is an exterior domains and, in place of (6.4.52), imposing the
weaker assumption formulated in (6.4.60). With this goal in mind, fix x0 ∈ Rn \ Ω
and, in place of (6.4.64), now consider
    n
w := E jk (x − ·) − E jk (x0 − ·) θ ε ∈ 𝒞∞ (Ω) ,
1≤k ≤n (6.4.92)
 
and ρ := − q j (x − ·) − q j (x0 − ·) θ ε ∈ 𝒞∞ (Ω).

Thanks to (6.2.1)-(6.2.2) and the Mean Value Theorem, these altered functions have
one extra unit of decay at infinity, compared to their original counterparts in (6.4.64).
In turn, this permits us to run the same argument as above for this choice of (w,  ρ),
even though we are now only assuming (6.4.60). Collecting the contributions made
by the introduction of the terms containing x0 into a constant c ∈ Cn then proves
(6.4.61). The proof of Theorem 6.4.9 is now complete. 

Lastly, we deduce some basic functional analytic identities involving hydrostatic

boundary layer potential operators acting on Besov spaces, akin to those discussed
in Theorem 4.4.2.

Theorem 6.4.10 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set satisfying a

two-sided local John condition and whose boundary is a compact Ahlfors regular
set. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. Also, fix λ ∈ C and consider p, q ∈ (1, ∞) and s ∈ (0, 1).
Then the following operator identities hold:
1   1  λ   p,q n
2 I + Kλ ◦ − 2 I + Kλ = S ◦ ∂ν Dλ, Pλ on Bs (∂Ω, σ) , (6.4.93)
1    λ    p,q  n
2 I + Kλ ◦ − 2 I + Kλ = ∂ν Dλ, Pλ ◦ S on B−s (∂Ω, σ) ,
# 1 #
(6.4.94)
     p,q n
Kλ# ◦ ∂νλ Dλ, Pλ = ∂νλ Dλ, Pλ ◦ Kλ on Bs (∂Ω, σ) , (6.4.95)
 p,q n
S ◦ Kλ# = Kλ ◦ S on B−s (∂Ω, σ) . (6.4.96)

Proof One way to see this is to rely on the operator identities from item (xi) of
Theorem 6.2.4, the density result from [69, Lemma 7.1.10], as well as the mapping
properties of the boundary layer potential operators from Theorem 6.4.3 and The-
orem 6.4.7. Another way of justifying the operator identities from (6.4.93)-(6.4.96)
is to carry out the same type of argument as in the proof of the operator identities
from item (xi) of Theorem 6.2.4, now starting with the Green-type representation
formula from Theorem 6.4.9 and then using the jump-relations from item (5) of
Theorem 6.4.3, Theorem 6.4.7, and Theorem 6.4.8. 
Chapter 7
Applications to Analysis in Several Complex
Variables

In this chapter we shall present a multitude of applications of the body of results

developed so far in the area of Geometric Harmonic Analysis to the field of Complex
Analysis of Several Variables. As is well known, Complex Analysis, Geometric
Measure Theory, and Harmonic Analysis interface tightly in the complex plane
(see, e.g., J. Garnett’s book [28] and the references therein). The complexity of
these connections is quite fascinating, and there are many aspects yet to be fully
understood. At the same time, progress continues to be registered, as this area of
mathematics continues to undergo seismic transformations thanks to spectacular
advances made in the last decades. These includes G. David’s characterization of the
L 2 boundedness of the Cauchy operator in terms of Ahlfors regularity, and X. Tolsa’s
results on analytic capacity, to name just a couple. Nonetheless, this fruitful interplay
between these branches of mathematics appears to have been much less explored in
the higher-dimensional setting, involving several complex variables.
Here we wish to venture into such territory, employing our brand of Divergence
Theorem from Volume I ([68]), the type of function space theory developed in
Volume II ([69]), and the sort of Calderón-Zygmund theory for singular integral
operators of boundary layer type devised in the current volume. Concrete topics
addressed in this chapter include: CR-functions and differential forms on boundaries
of sets of locally finite perimeter (in §7.1), integration by parts formulas involving
the ∂-operator
¯ on sets of locally finite perimeter (in §7.2), the Bochner-Martinelli
integral operator (in §7.3), a sharp version of the Bochner-Martinelli-Koppelman
formula (in §7.4), the Extension Problem for Hölder CR-functions on boundaries of
Ahlfors regular domains (in §7.5), the Extension Problem for Lebesgue and bounded
(or vanishing) mean oscillation functions on boundaries of uniformly rectifiable
domains (in §7.6), as well as the ∂-operator
¯ and the Dolbeault complex on uniformly
rectifiable sets (in §7.7).

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 783
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_7
784 7 Applications to Analysis in Several Complex Variables

7.1 CR-Functions and Differential Forms on Boundaries of

Locally Finite Perimeter Sets

Having fixed n ∈ N, we have a natural identification

Cn = C × · · · × C ≡ (R × R) × · · · × (R × R) ≡ R2n . (7.1.1)

Specifically, if for each j ∈ {1, . . . , n} we let z j denote the complex variable in the
j-th factor of the Cartesian product Cn , then set x j := Re z j and y j := Im z j , we shall
identify
Cn  (z1, . . . , zn ) ≡ (x1, y1, x2, y2, . . . , xn, yn ) ∈ R2n . (7.1.2)
In the converse direction, given a real vector ξ = (ξ1, ξ2, . . . , ξ2n−1, ξ2n ) ∈ R2n , we
agree to identify it with its complex version

ξC := (ξ1 + iξ2, . . . , ξ2n−1 + iξ2n ) ∈ Cn . (7.1.3)

For each j ∈ {1, . . . , n}, the relationship between the individual complex variables
z j ∈ C and their real components (x j , y j ) ∈ R × R implies

z j = x j + iy j , dz j = dx j + idy j , d z̄ j = dx j − idy j ,
dx j = 2−1 (dz j + d z̄ j ), dy j = (−i)2−1 (dz j − d z̄ j ),
∂z j = 2−1 (∂x j − i∂y j ) = 2−1 (∂2j−1 − i∂2j ), (7.1.4)
∂z̄ j = 2−1 (∂x j + i∂y j ) = 2−1 (∂2j−1 + i∂2j ),
∂x j = ∂z j + ∂z̄ j , ∂y j = i(∂z j − ∂z̄ j ).

Going further, assume that Ω is a set of locally finite perimeter in R2n ≡ Cn .

Abbreviate σ := H 2n−1 ∂Ω, denoted by ν its geometric measure theoretic outward
unit normal, and recall that νC stands for its complexified version (defined as in
(7.1.3)). In such a setting, for each j, k ∈ {1, . . . , n} we introduce the complex
tangential derivative operator ∂τCj k by setting
     
∂τCj k ψ := (νC ) j ∂z̄k ψ  − (νC )k ∂z̄ j ψ 
∂∗ Ω ∂∗ Ω
for every complex-valued function ψ of class (7.1.5)
𝒞1 in some open neighborhood of ∂∗ Ω in Cn .

We aim to extend the action of the complex tangential derivative operators to other
classes of functions than those just described.

Definition 7.1.1 Let Ω be a set of locally finite perimeter in R2n ≡ Cn . Denote by ν its
geometric measure theoretic outward unit normal, and abbreviate σ := H 2n−1 ∂Ω.
Also, fix two arbitrary indices j, k ∈ {1, . . . , n}. Say that the complex tangential
derivative operator ∂τCj k maps a given complex-valued function f ∈ Lloc 1 (∂ Ω, σ)
∗
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 785

1 (∂ Ω, σ) provided that there exists some complex-valued function h in

into Lloc ∗
Lloc (∂∗ Ω, σ) such that
1

∫ ∫
 C 
f ∂τ j k ψ dσ = − hψ dσ for each ψ ∈ 𝒞1c (Cn ), (7.1.6)
∂∗ Ω ∂∗ Ω

with ∂τCj k ψ in the left-hand side understood in the sense of (7.1.5).

Finally, given any p ∈ [1, ∞] define

p,1 p p
LC,loc (∂∗ Ω, σ) := f ∈ Lloc (∂∗ Ω, σ) : ∂τCj k f belongs to Lloc (∂∗ Ω, σ)

for each j, k ∈ {1, . . . , n} . (7.1.7)

From [68, Proposition 3.7.2] (whose applicability in the present setting is ensured
by [68, Lemma 3.6.4] and [68, (5.6.33)]) it follows that there could be at most one
function h ∈ Lloc1 (∂ Ω, σ) doing the job in (7.1.6). As such, there is no ambiguity if
∗
for a function f ∈ Lloc1 (∂ Ω, σ) which is mapped by the complex tangential derivative
∗
C 1 (∂ Ω, σ) we define
operator ∂τ j k into Lloc ∗

∂τCj k f := h at σ-a.e. point on ∂∗ Ω. (7.1.8)

In particular, in the context of the above definition, formula (7.1.6) becomes

∫ ∫
 C   C 
f ∂τ j k ψ dσ = − ∂τ j k f ψ dσ for each ψ ∈ 𝒞1c (Cn ). (7.1.9)
∂∗ Ω ∂∗ Ω

We wish to stress that using the symbol ∂τCj k f for the function h doing the job
in (7.1.6) creates no ambiguity with the definition given in (7.1.5) in the case when
f ∈ 𝒞1c (Cn ). Indeed, in such a scenario, (7.1.9) holds by virtue of the De Giorgi-
Federer version of the Divergence Theorem recorded in [68, Theorem 1.1.1] (see
Lemma 7.1.6 for details in similar circumstances).
We are going to be particularly interested in the case when, for a given integrability
exponent p ∈ (1, ∞), the complex tangential derivative operator ∂τCj k maps a certain
function f ∈ L p (∂∗ Ω, σ) into the space L p (∂∗ Ω, σ). Thanks to (7.1.9), Riesz’s Du-
ality Theorem, and [68, Proposition 3.7.1] (whose applicability is presently ensured
by [68, Lemma 3.6.4] and [68, (5.6.33)]), if p ∈ (1, ∞) is such that 1/p + 1/p = 1,
this amounts to having
∫ 
  C  
sup  f ∂τ j k ψ dσ  < +∞. (7.1.10)

ψ ∈𝒞1c (C n ) ∂∗ Ω
ψ | ∂∗ Ω L p  (∂ Ω, σ) ≤1
∗

Moreover, in (7.1.10) we may restrict ourselves to functions ψ ∈ 𝒞cN (U), where

U ⊆ Cn is an open neighborhood of ∂Ω, and N ∈ N ∪ {∞}.
786 7 Applications to Analysis in Several Complex Variables

Here is a formal definition of the partial Sobolev spaces involving the action of the
complex tangential differential operators considered in the sense of Definition 7.1.1
(cf. also (7.1.8)).

Definition 7.1.2 Assume Ω is a set of locally finite perimeter in R2n ≡ Cn . Abbreviate

σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit
normal. In this context, for each p ∈ [1, ∞] define the complex boundary Sobolev
space

LCp,1 (∂∗ Ω, σ) := f ∈ L p (∂∗ Ω, σ) : ∂τCj k f belongs to L p (∂∗ Ω, σ)

for each j, k ∈ {1, . . . , n} (7.1.11)

and equip it with the norm

n
f p,1
LC (∂∗ Ω,σ)
:= f L p (∂∗ Ω,σ) + ∂τCj k f L p (∂∗ Ω,σ)
j,k=1 (7.1.12)
for each f ∈ LCp,1 (∂∗ Ω, σ).

In particular, in the above setting, for each j, k ∈ {1, . . . , n} the operator

∂τCj k : LCp,1 (∂∗ Ω, σ) −→ L p (∂∗ Ω, σ) (7.1.13)

is well defined, linear, and bounded. Moreover, we have

∂τCj k = −∂τCk j , 1 ≤ j, k ≤ n, hence ∂τCj j = 0, ∀ j ∈ {1, . . . , n}. (7.1.14)

The operators introduced in (7.1.13) should be contrasted with the family

∂τ m 1≤,m≤2n of real tangential derivative operators on ∂∗ Ω. On smooth
functions, the latter operators act according to
     
∂τ m ψ := ν ∂m ψ  − νm ∂ ψ  for every ψ ∈ 𝒞1 (Cn ), (7.1.15)
∂∗ Ω ∂∗ Ω

(for each , m ∈ {1, . . . , 2n}) and their action is further extended to functions in the
p
boundary Sobolev space L1 (∂∗ Ω, σ). See [69, Chapter 11] for details. It is then clear
from definitions that
p
L1 (∂∗ Ω, σ) ⊆ LCp,1 (∂∗ Ω, σ) (7.1.16)
and that
1    
∂τCj k = ∂τ2k−1, 2 j−1 − ∂τ2k, 2 j + i ∂τ2k, 2 j−1 + ∂τ2k−1, 2 j (7.1.17)
2
for all j, k ∈ {1, . . . , n}.

Proposition 7.1.3 Suppose Ω is a set of locally finite perimeter in R2n ≡ Cn and, as

usual, abbreviate σ := H 2n−1 ∂Ω. Also, select an open neighborhood U ⊆ Cn of
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 787

1,1
∂Ω and consider two function, f ∈ LC,loc (∂∗ Ω, σ) along with ψ ∈ 𝒞1 (U), satisfying
∫ ∫
n
  n
 C 
|f| ∂z̄ ψ  dσ < +∞, ∂τ f  |ϕ| dσ < +∞,
j jk

∂∗ Ω j=1 ∂∗ Ω j,k=1
∫ (7.1.18)
−1
as well as | f (z)||ψ(z)|(1 + |z|) dσ(z) < +∞.
∂∗ Ω

Then for each j, k ∈ {1, . . . , n} one has

∫ ∫
   
f ∂τCj k ψ dσ = − ∂τCj k f ψ dσ. (7.1.19)
∂∗ Ω ∂∗ Ω

Consequently, the boundary integration by parts formula (7.1.19) is true for all
j, k ∈ {1, . . . , n} if f ∈ LC (∂∗ Ω, σ) with p ∈ [1, ∞] and, with p ∈ [1, ∞] such that
p,1

1/p + 1/p = 1,

ψ ∈ 𝒞1 (U) satisfies ψ ∂∗ Ω ∈ L p (∂∗ Ω, σ) and


 (7.1.20)
(∂z̄ j ψ)∂∗ Ω ∈ L p (∂∗ Ω, σ) for each j ∈ {1, . . . , n}.


p,1
In particular, (7.1.19) holds for every f ∈ LC (∂∗ Ω, σ) with p ∈ [1, ∞] and
ψ ∈ 𝒞1 (U) in the case when ∂∗ Ω is bounded.

Proof The same argument as in the proof of [69, Lemma 11.1.7] works virtually
verbatim in the present setting. 

In contrast with the real tangential derivative operators ∂τ j k 1≤ j,k ≤n (from [69,
Definition 11.1.2]) which, at least when considered on Sobolev spaces defined on
the boundary of an open set Ω ⊆ Rn with an Ahlfors regular boundary and which
satisfies a two-sided local John condition, annihilate precisely the locally constant
functions on ∂Ω (cf. [69, Proposition 11.4.3]), the common null-space for the family
of complex tangential derivative operators ∂τCj k 1≤ j,k ≤n is much richer, consisting
of CR-functions on ∂Ω as defined next.

Definition 7.1.4 Let Ω be a set of locally finite perimeter in R2n ≡ Cn and, as usual,
abbreviate σ := H 2n−1 ∂Ω. In this setting, call f ∈ Lloc
1 (∂ Ω, σ) a CR-function
∗
provided

∂τCj k f = 0 at σ-a.e. point on ∂∗ Ω for all j, k ∈ {1, . . . , n}. (7.1.21)

In the context of the above definition, it follows from Definition 7.1.2 that
any CR-function f ∈ L p (∂∗ Ω, σ) with p ∈ [1, ∞]
p,1 (7.1.22)
actually belongs to the Sobolev space LC (∂∗ Ω, σ).

Trivially, the last formula in (7.1.14) implies that

788 7 Applications to Analysis in Several Complex Variables

when n = 1 it follows that any f ∈ L p (∂∗ Ω, σ)

(7.1.23)
with p ∈ [1, ∞] is actually a CR-function.
Also, from (7.1.21) and (7.1.5) we deduce that, in the context of Definition 7.1.4,

if U is an open neighborhood of ∂∗ Ω in Cn and ψ ∈ 𝒞1 (U) is a

(7.1.24)
holomorphic function in U, then ψ ∂∗ Ω is a CR-function.

An incisive generalization of this observation is contained in the proposition be-

low. Before stating it, the reader is reminded that, given an open set Ω ⊆ Cn , a
function F : Ω → C is said to be holomorphic provided F is separately holomor-
phic (or separately analytic) in the sense of one complex variable analysis. That is,
for each j ∈ {1, . . . , n} and each z = (z1, . . . , zn ) ∈ Ω, the function
 
Oz, j  ζ −→ F z1, . . . , z j−1, ζ, z j+1, . . . , zn ∈ C
  (7.1.25)
where Oz, j := ζ ∈ C : z1, . . . , z j−1, ζ, z j+1, . . . , zn ∈ Ω

is holomorphic, in the classical one-variable sense. In this vein, it is worth recalling

that Hartogs’ classical theorem on separate analyticity asserts that any such “sepa-
rately analytic” function F turns out to be continuous and, ultimately, 𝒞∞ smooth in
Ω (see, e.g., [37], [51, §2.4, pp. 107–110]).

Proposition 7.1.5 Let Ω be an open nonempty proper subset of R2n ≡ Cn with a

lower Ahlfors regular boundary and with the property that σ := H 2n−1 ∂Ω is a
doubling measure on ∂Ω. Fix an aperture parameter κ ∈ (0, ∞) and a truncation
1 (Ω, L 2n ) is a complex-valued function satisfying
parameter ε > 0. Suppose F ∈ Lloc
κ−n.t.
Nκε F ∈ Lloc
1
(∂Ω, σ), the trace F ∂Ω exists σ-a.e. on ∂nta Ω (7.1.26)

and such that, for some fixed j, k ∈ {1, . . . , n}, (with derivatives taken in the sense
of distributions)

∂z̄ j F, ∂z̄k F ∈ Lloc

1 (Ω, L 2n ), Nκε (∂z̄ j F), Nκε (∂z̄k F) ∈ Lloc
1 (∂Ω, σ), and

κ−n.t. κ−n.t. (7.1.27)

the traces (∂z̄ j F)∂Ω , (∂z̄k F)∂Ω exist at σ-a.e. point on ∂nta Ω.
κ  −n.t. κ  −n.t.
Then for any other κ  > 0 the nontangential traces F ∂Ω , (∂z̄ j F)∂Ω ,
κ  −n.t.
(∂z̄k F)∂Ω exist at σ-a.e. point on ∂nta Ω, and are actually independent of κ . Also,
when the dependence on κ  is dropped, and when considered on ∂∗ Ω, the function
n.t.
F ∂Ω belongs to Lloc
1 (∂ Ω, σ) and is mapped by ∂ C into L 1 (∂ Ω, σ), the functions
∗ τj k loc ∗
n.t. n.t.

(∂z̄ F) , (∂z̄ F)  belong to L (∂∗ Ω, σ), and
1
j ∂Ω k ∂Ω loc
 n.t.   n.t.   n.t. 
∂τCj k F ∂Ω = (νC ) j (∂z̄k F)∂Ω − (νC )k (∂z̄ j F)∂Ω (7.1.28)

at σ-a.e. point on ∂∗ Ω.
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 789

As a corollary, if F : Ω → C is a holomorphic function satisfying (7.1.26) then

κ−n.t.
F  considered on ∂∗ Ω is a CR-function.
∂Ω

Proof The present assumptions imply that Ω is a set of locally finite perimeter. In
particular, it is meaningful to consider its geometric measure theoretic outward unit
normal ν. Consider U := z ∈ Cn : dist(z, ∂Ω) < ε , an open neighborhood of ∂Ω
in Cn , and pick an arbitrary function ψ ∈ 𝒞1c (U). Also, select j, k ∈ {1, . . . , n} and,
with e 1≤ ≤2n denoting the standard orthonormal basis in R2n , define the vector
field

H := (ψ ∂z̄k F + F ∂z̄k ψ) (e2j−1 + ie2j )

− (ψ ∂z̄ j F + F ∂z̄ j ψ) (e2k−1 + ie2k ) in Ω. (7.1.29)

 1  2n
Then H ∈ Lloc (Ω, L 2n ) and, with the divergence considered as an operator in
R2n ,    
divH = 2 ∂z̄ j ∂z̄k F ψ − 2 ∂z̄k ∂z̄ j F ψ = 0 in D (Ω). (7.1.30)
Going further, if we abbreviate
κ−n.t. κ−n.t. κ−n.t.
f := F ∂Ω , f j := (∂z̄ j F)∂Ω , fk := (∂z̄k F)∂Ω , (7.1.31)

then f , f j , fk ∈ Lloc
1 (∂ Ω, σ), thanks to (7.1.26), (7.1.27), [68, (8.9.8)], [68, (8.9.44)],
∗
and [68, (8.8.52)]. Also
κ−n.t.
H ∂Ω exists at σ-a.e. point on ∂nta Ω (7.1.32)

and, by design,
 κ−n.t.   
ν · H ∂Ω = (νC ) j fk − (νC )k f j ψ + f ∂τCj k ψ at σ-a.e. on ∂∗ Ω. (7.1.33)

Note that since

√   √  
Nκ H ≤ 2 Nκε F Nκ (∇ψ) + 2 Nκε (∂z̄ j F) + Nκε (∂z̄k F) Nκ ψ (7.1.34)

pointwise on ∂Ω, and since Nκ ψ, Nκ (∇ψ) ∈ Lcomp ∞ (∂Ω, σ), the first condition in

(7.1.26) together with the second condition in (7.1.27) and [68, (8.2.26)] imply that
Nκ H ∈ L 1 (∂Ω, σ). Lastly, if Ω is an exterior domain then the compact support
condition on ψ ensures that H vanishes in a neighborhood of infinity. Granted these
properties, [68, Theorem 1.2.1] applies and the Divergence Formula [68, (1.2.2)]
written for H yields (on account of (7.1.30) and (7.1.33))
∫ ∫
 
f ∂τCj k ψ dσ = (νC ) j fk − (νC )k f j ψ dσ. (7.1.35)
∂∗ Ω ∂∗ Ω
790 7 Applications to Analysis in Several Complex Variables

In light of Definition 7.1.1 (cf. also (7.1.8)) from (7.1.35) and the arbitrariness of ψ
in 𝒞1c (U) we ultimately conclude that ∂τCj k f = (νC ) j fk − (νC )k f j on ∂∗ Ω. From this,
(7.1.28) follows on account of (7.1.31).
Finally, when F is actually a holomorphic function in Ω, the right-hand side
of (7.1.28) vanishes for each indices j, k ∈ {1, . . . , n}. As noted in (7.1.21), this
κ−n.t.
amounts to saying that F ∂Ω is indeed a CR-function. 

We next review notation pertaining to differential forms used frequently in the

remainder of this volume. For more background on this and related issues, the reader
is referred to the monographs [13], [59], and [88]. Throughout, the wedge symbol
∧ will denote the exterior product of differential forms. Given α, β ∈ {0, 1, . . . , n}
denote by Λα,β Cn the space of differential forms of (type) degree (α, β) with complex
coefficients, i.e., objects represented as

u= u I,J dz I ∧ d z̄ J . (7.1.36)
|I |=α, |J |=β

The sum in (7.1.36) is performed over strictly increasing arrays. That is, we have
I = (i1, . . . , iα ) ∈ {1, . . . , n}α satisfying i1 < · · · < iα of length |I | = α, and
J = ( j1, . . . , jβ ) ∈ {1, . . . , n} β satisfying j1 < · · · < jβ of length |J | = β, for which
we have set dz I := dzi1 ∧ · · · ∧ dziα and d z̄ J := d z̄ j1 ∧ · · · ∧ d z̄ jβ , and u I,J ∈ C. The
complex conjugate of a form u ∈ Λα,β Cn written as in (7.1.36) is defined to be

ū := u I,J d z̄ I ∧ dz J = (−1)αβ u I,J dz J ∧ d z̄ I . (7.1.37)

|I |=α |J |=β |J |=β |I |=α

Hence,
ū ∈ Λβ,α Cn and ū¯ = u for each u ∈ Λα,β Cn . (7.1.38)
 
Also, for each u ∈ Λα,β Cn and w ∈ Λα ,β Cn we have

u ∧ w = ū ∧ w̄. (7.1.39)

Pressing on, recall from (A.0.67) that for any two arrays J, K (not necessarily
ordered), the generalized Kronecker symbol εK J
is given by
  
det (δ jk ) j ∈J, k ∈K if J, K agree as sets,
εK
J
:= (7.1.40)
0 otherwise,

where δ jk := 1 if j = k, and zero if j  k. Put differently,

⎧
⎪ 0 if J, K do not agree as sets,
⎪
⎨
⎪
εK
J
= +1 if J differs from K by an even permutation, (7.1.41)
⎪
⎪
⎪ −1 if J differs from K by an odd permutation.
⎩
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 791

We shall employ the Hermitian inner product ·, ·C on differential forms uniquely
defined by the requirement that
 I 
dz ∧ d z̄ J , dz A ∧ d z̄ B = 2 |I |+ |J | ε IA εBJ for all arrays I, J, A, B. (7.1.42)
C

The power of 2 is an artifact of dz j = dx j + idy j having length 21/2 (rather than being
of unit length). Thus, in particular, if α, β ∈ {0, 1, . . . , n} then

 f , gC = 2α+β fI,J gI,J if f , g ∈ Λα,β Cn are given by

|I |=α, |J |=β
(7.1.43)
f = fI,J dz I ∧ d z̄ J and g = gI,J dz I ∧ d z̄ J .
|I |=α, |J |=β |I |=α, |J |=β

Note that if f , g ∈ Λα,β Cn then

 f , gC = g, f  C = ḡ, f¯ C (7.1.44)

and that  
 f , gC = 0 if f ∈ Λα,β Cn and g ∈ Λα ,β Cn
(7.1.45)
with either α  α  or β  β  .
As customary, write
   1/2
| f |C :=  f , f C = 2α+β | fI,J | 2
|I |=α |J |=β
(7.1.46)
for each f = fI,J dz I ∧ d z̄ J ∈ Λα,β Cn .
|I |=α, |J |=β

The volume element in Cn ≡ R2n is given by

dV = dx1 ∧ dy1 ∧ · · · ∧ dxn ∧ dyn

= (−2i)−n dz1 ∧ d z̄1 ∧ · · · ∧ dzn ∧ d z̄n

= (2i)−n (−1)n(n−1)/2 d z̄1 ∧ · · · ∧ d z̄n ∧ dz1 ∧ · · · ∧ dzn . (7.1.47)

In the context of integration over open subsets of Cn ≡ R2n , we shall tacitly identify
dV with the Lebesgue measure L 2n . Next, consider the Hodge star operator ∗ in
Cn ≡ R2n . For each given α, β ∈ {0, 1, . . . , n} this may be characterized as the
unique linear isomorphism

∗ : Λα,β Cn −→ Λn−β,n−α Cn (7.1.48)

with the property that

u ∧ (∗ū) = |u|C2 dV for every u ∈ Λα,β Cn . (7.1.49)

792 7 Applications to Analysis in Several Complex Variables

In particular, ∗1 = dV. In fact, it can be checked that

if I, J, K are increasing, mutually disjoint subsets of {1, . . . , n} then

  K 
∗ dz I ∧ d z̄ J ∧ dz ∧ d z̄ = in 2 M−n (−1) M(M−1)/2+ |I | × (7.1.50)
  {1,...,n}\(I∪J∪K)
× dz I ∧ d z̄ J ∧ dz ∧ d z̄

where M := |I | + |J | + 2|K | and, if K = (k1, . . . , k ), we have abbreviated

 K
dz ∧ d z̄ := (dzk1 ∧ d z̄k1 ) ∧ · · · ∧ (dzk ∧ d z̄k ). (7.1.51)

Using the Hodge-star operator, we define the interior product between a 1-form θ
and an -form u by setting
θ ∨ u := ∗(θ ∧ ∗u). (7.1.52)
For further reference as well as for the convenience of the reader, some basic,
elementary properties of these objects are summarized in the following lemma.

Lemma 7.1.6 For arbitrary one-forms θ, η, and any -form u, -form ω, ( + 1)-form
w, and (2n − )-form λ in R2n ≡ Cn , the following are true:
(1) ∗ ∗ u = (−1) u, u, ∗λC = (−1) ∗u, λC , and ∗u, ∗λC = u, λC ;
(2) θ ∧ (θ ∧ u) = 0 and θ ∨ (θ ∨ u) = 0;
(3) θ ∧ (η ∨ u) + η ∨ (θ ∧ u) = θ, η̄C u and θ ∧ u, wC = u, θ̄ ∨ wC ;
(4) ∗(θ ∧ u) = (−1) θ ∨ (∗u) and ∗(θ ∨ u) = (−1)−1 θ ∧ (∗u);
(5) ∗ū = ∗u and u ∧ ∗ω̄ = u, ωC dV.
Moreover, if θ is normalized such that θ, θC = 1, then also:
(6) u = θ ∧ (θ̄ ∨ u) + θ̄ ∨ (θ ∧ u) and |u|C2 = |θ ∧ u|C2 + | θ̄ ∨ u|C2 ;
(7) | θ̄ ∧ (θ ∨ u)|C = |θ ∨ u|C and |θ ∨ (θ̄ ∧ u)|C = | θ̄ ∧ u|C .
Lastly, if θ ∈ Λ1,0 Cn and η ∈ Λ0,1 Cn then

θ∧ : Λα,β Cn −→ Λα+1,β Cn, θ∨ : Λα,β Cn −→ Λα,β−1 Cn,

(7.1.53)
η∧ : Λα,β Cn −→ Λα,β+1 Cn, η∨ : Λα,β Cn −→ Λα−1,β Cn .

For practical purposes it is relevant to point out that if α, β ∈ {0, 1, . . . , n} then

n
η ∧ u = (−1)α εJ η j u M,I dz M ∧ d z̄ J ∈ Λα,β+1 Cn if
jI

|M |=α |J |=β+1 |I |=β j=1

n
u= u M,I dz M ∧ d z̄ I ∈ Λα,β Cn and η = η j d z̄ j ∈ Λ0,1 Cn,
|M |=α |I |=β j=1
(7.1.54)

and
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 793
n
εJ θ j w M,I dz J ∧ d z̄ I ∈ Λα+1,β Cn if
jM
θ∧w =
|M |=α |J |=α+1 |I |=β j=1
n
w= w M,I dz M ∧ d z̄ I ∈ Λα,β Cn and θ = θ j dz j ∈ Λ1,0 Cn .
|M |=α |I |=β j=1
(7.1.55)

These identities can be seen directly from definitions. In addition, given any two
indices α, β ∈ {0, 1, . . . , n}, we may conclude from (7.1.50) and (7.1.52) that
n
εK η j wK,J dz I ∧ d z̄ J ∈ Λα−1,β Cn if
jI
η∨w =2
|K |=α |J |=β |I |=α−1 j=1
n (7.1.56)
w= wK,J dz K ∧ d z̄ J ∈ Λα,β Cn and η = η j d z̄ j ∈ Λ0,1 Cn,
|K |=α |J |=β j=1

and
n
θ ∨ u = 2(−1)α εJ θ j uK,J dz K ∧ d z̄ I ∈ Λα,β−1 Cn if
jI

|K |=α |J |=β |I |=β−1 j=1

n
u= uK,J dz K ∧ d z̄ J ∈ Λα,β Cn and θ = θ j dz j ∈ Λ1,0 Cn .
|K |=α |J |=β j=1
(7.1.57)

Before moving on, we wish to remark that the operators ∧, ∨, ∗ extend to differential
forms with variable coefficients by considering their action in a natural pointwise
fashion.
Given any real vector ξ = (ξ1, ξ2, . . . , ξ2n−1, ξ2n ) ∈ R2n , recall its complex version
ξC ∈ Cn from (7.1.3), and define
n n
ξ 1,0 := (ξC ) j dz j ∈ Λ1,0 Cn, ξ 0,1 := (ξC ) j dz j ∈ Λ0,1 Cn . (7.1.58)
j=1 j=1

From definitions it follows that for each ξ ∈ R2n we have

2ξ = ξ 1,0 + ξ 0,1, ξ 0,1 = ξ 1,0, ξ 1,0 = ξ 0,1,

(7.1.59)
ξ 1,0, ξ 0,1 C = 0, |ξ 1,0 |C = |ξ 0,1 |C = 21/2 |ξ |.

In particular, all the above considerations apply to the geometric measure theoretic
unit normal ν = (ν1, ν2, . . . , ν2n−1, ν2n ) ∈ R2n of a set of locally finite perimeter
Ω ⊆ R2n ≡ Cn . In such a setting, recall from (7.1.3) that the complex outward unit
normal to Ω is given by
   
νC = ν1 + iν2, . . . , ν2n−1 + iν2n = ν2j−1 + iν2j 1≤ j ≤n ∈ Cn . (7.1.60)
794 7 Applications to Analysis in Several Complex Variables

Also, in keeping with (7.1.58), let us define

n n
ν 1,0 := (νC ) j dz j ∈ Λ1,0 Cn, ν 0,1 := (νC ) j dz j ∈ Λ0,1 Cn . (7.1.61)
j=1 j=1

If the vector ν = (ν1, ν2, . . . , ν2n−1, ν2n ) is further identified with the 1-form

ν = ν1 dx1 + ν2 dy1 + · · · + ν2n−1 dxn + ν2n dyn, (7.1.62)

then from (7.1.59) we see that

2ν = ν 1,0 + ν 0,1, ν 0,1 = ν 1,0, ν 1,0 = ν 0,1,

(7.1.63)
ν 1,0, ν 0,1 C = 0, |ν 1,0 |C = |ν 0,1 |C = 21/2 .

Definition 7.1.7 Assume Ω is a set of locally finite perimeter in R2n ≡ Cn . Abbreviate

σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit
normal. Also, fix some α, β ∈ {0, 1, . . . , n} and consider a Λα,β Cn -valued function
f defined σ-a.e. on ∂∗ Ω. Then call
de f
f complex tangential ⇐⇒ ν 1,0 ∨ f = 0 at σ-a.e. point on ∂∗ Ω, (7.1.64)

and
de f
f complex normal ⇐⇒ ν 0,1 ∧ f = 0 at σ-a.e. point on ∂∗ Ω. (7.1.65)

In the context of the above definition, based on item (6) of Lemma 7.1.6 and
(7.1.63) we may write
 
2 f = ν 0,1, ν 1,0 f = ν 0,1 ∧ (ν 1,0 ∨ f ) + ν 1,0 ∨ (ν 0,1 ∧ f ), (7.1.66)
C

hence
f = ftan,C + fnor,C (7.1.67)
where

ftan,C := 1
2 ν 1,0 ∨ (ν 0,1 ∧ f ) and fnor,C := 1
2 ν 0,1 ∧ (ν 1,0 ∨ f ). (7.1.68)

Also, a direct consequence of the definitions in (7.1.68) and (7.1.66) is that

ν 0,1 ∧ f = ν 0,1 ∧ ftan,C and ν 1,0 ∨ f = ν 1,0 ∨ fnor,C . (7.1.69)

Moving on, recall that the d-bar operator acts on a continuously differentiable
complex-valued function f defined in an open subset of Cn according to
n
∂¯ f := (∂z̄ j f ) d z̄ j . (7.1.70)
j=1
7.1 CR-Functions and Differential Forms on Boundaries of Locally Finite Perimeter Sets 795

Continue to assume that Ω is a set of locally finite perimeter in R2n ≡ Cn . As

in the past, denote by ν its geometric measure theoretic outward unit normal, and
abbreviate σ := H 2n−1 ∂Ω. In this context, given any complex-valued function f
of class 𝒞1 in a neighborhood of ∂∗ Ω, we have

ν 0,1 ∧ ∂¯ f = (νC ) j ∂z̄k f − (νC )k ∂z̄ j f d z̄ j ∧ d z̄k

1≤ j<k ≤n

= ∂τCj k f d z̄ j ∧ d z̄k at σ-a.e. point on ∂∗ Ω. (7.1.71)

1≤ j<k ≤n

From this, Definition 7.1.7, Definition 7.1.4, and (7.1.70) it follows that

if f is a scalar function of class 𝒞1 in a neighborhood of ∂∗ Ω

in Cn then its restriction to ∂∗ Ω is a CR-function if and only if (7.1.72)
the restriction of ∂¯ f to ∂∗ Ω is a complex normal form.
A significantly more refined version of this observation is described in our next
proposition.

Proposition 7.1.8 Let Ω be an open nonempty proper subset of R2n ≡ Cn with a

lower Ahlfors regular boundary and with the property that σ := H 2n−1 ∂Ω is a
doubling measure on ∂Ω. Fix an integrability exponent p ∈ [1, ∞], an aperture
parameter κ ∈ (0, ∞), and a truncation parameter ε > 0. In this setting, assume
1 (Ω, L 2n ) is a complex-valued function satisfying
F ∈ Lloc
κ−n.t.
Nκε F ∈ Lloc (∂Ω, σ), the trace F ∂Ω exists σ-a.e. on ∂nta Ω
p
(7.1.73)

and such that (with the operator ∂¯ applied in the sense of distributions)
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λ0,1, Nκε (∂F)
∂F ¯ ∈ L p (∂Ω, σ), and
loc loc
κ−n.t. (7.1.74)
¯ 
the trace (∂F) exists at σ-a.e. point on ∂nta Ω.
∂Ω

κ  −n.t. κ  −n.t.
Then for any other κ  > 0 the nontangential traces F ∂Ω , (∂F) ¯ 
∂Ω
exist σ-a.e.
on ∂nta Ω, and are actually independent of κ . Also, when the dependence on κ  is
n.t.
dropped, and when considered on ∂∗ Ω, the function F ∂Ω belongs to LC,loc (∂∗ Ω, σ),
p,1
n.t.
¯  belong to L p (∂∗ Ω, σ) ⊗ Λ0,1 and
the form (∂F) ∂Ω loc
 n.t.   n.t. 
¯ 
ν 0,1 ∧ (∂F) = ∂τCj k F ∂Ω d z̄ j ∧d z̄k at σ-a.e. point on ∂∗ Ω. (7.1.75)
∂Ω
1≤ j<k ≤n

n.t. n.t.
¯  is complex normal if and only if F  is a CR-function.
As a corollary, (∂F) ∂Ω ∂Ω

Proof The fact that the nontangential traces do not depend on the aperture and
belong to the specified spaces is seen as in Proposition 7.1.5. Together with (7.1.61),
796 7 Applications to Analysis in Several Complex Variables

the latter result (cf. (7.1.28)) also permits us to compute

 n.t.    n.t.   n.t. 
¯ 
ν 0,1 ∧ (∂F) = (νC ) j (∂z̄ F)  − (νC ) k (∂z̄ F)  d z̄ j ∧ d z̄k
∂Ω k ∂Ω j ∂Ω
1≤ j<k ≤n
 n.t. 
= ∂τCj k F ∂Ω d z̄ j ∧ d z̄k (7.1.76)
1≤ j<k ≤n

at σ-a.e. point on ∂∗ Ω. This establishes (7.1.75), and the very last claim in the
statement is a consequence of it. 

Going forward, in keeping with an earlier notational convention, if (X, μ) is a

given measure space then for each α, β ∈ {0, 1, . . . , n} and each p ∈ (0, ∞] we shall
abbreviate
L p (X, μ) ⊗ Λα,β := L p (X, μ) ⊗ Λα,β Cn . (7.1.77)
In other words, the space L p (X, μ) ⊗ Λα,β is the is the space of differential forms
of type (α, β) with coefficients from L p (X, μ). This becomes a quasi-Banach space
when equipped with the quasi-norm

f L p (X,μ)⊗Λα, β := fI,J L p (X,μ)

|I |=α, |J |=β
(7.1.78)
if f = fI,J dz I ∧ d z̄ J ∈ L p (X, μ) ⊗ Λα,β .
|I |=α, |J |=β

Definition 7.1.9 Suppose Ω is a set of locally finite perimeter in R2n ≡ Cn . Abbre-

viate σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit
normal. For each pair of degrees α, β ∈ {0, 1, . . . , n} and each exponent p ∈ (0, ∞]
consider

Ltan,C (∂∗ Ω, σ) ⊗ Λα,β := { f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β : f is complex tangential},

p

(7.1.79)

and

Lnor,C (∂∗ Ω, σ) ⊗ Λα,β := { f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β : f is complex normal},

p

(7.1.80)

where complex tangentiality and complex normality are as introduced in Defini-

tion 7.1.7.

It follows from definitions that Ltan,C (∂∗ Ω, σ) ⊗ Λα,β and Lnor,C (∂∗ Ω, σ) ⊗ Λα,β
p p

are closed subspaces of L p (∂∗ Ω, σ) ⊗ Λα,β , satisfying

 p   p 
Ltan,C (∂∗ Ω, σ) ⊗ Λα,β ∩ Lnor,C (∂∗ Ω, σ) ⊗ Λα,β = {0}. (7.1.81)
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 797

Consequently, the following decomposition holds both algebraically and topologi-

cally:
 p   p 
L p (∂∗ Ω, σ) ⊗ Λα,β = Ltan,C (∂∗ Ω, σ) ⊗ Λα,β ⊕ Lnor,C (∂∗ Ω, σ) ⊗ Λα,β .
(7.1.82)

When p = 2 the above direct sum is in fact orthogonal. For further use, let us also
note here that, as is apparent from (7.1.53), (7.1.67)-(7.1.68), and simple degree
considerations,

Ltan,C (∂∗ Ω, σ) ⊗ Λα,0 = L p (∂∗ Ω, σ) ⊗ Λα,0,

p

(7.1.83)
Lnor,C (∂∗ Ω, σ) ⊗ Λα,n = L p (∂∗ Ω, σ) ⊗ Λα,n .
p

Moving on, in the same context as in Definition 7.1.2, for each α, β ∈ {0, 1, . . . , n}
let us now consider the space

LCp,1 (∂∗ Ω, σ) ⊗ Λα,β (7.1.84)


:= f = fK,I dz K ∧ d z̄ I : fK,I ∈ LCp,1 (∂∗ Ω, σ)
|K |=α |I |=β

for each K ∈ {1, . . . , n}α and I ∈ {1, . . . , n} β

equipped with the norm

n
f p,1
LC (∂∗ Ω,σ)⊗Λα, β
:= f L p (∂∗ Ω,σ)⊗Λα, β + ∂τCj k fK,I L p (∂∗ Ω,σ)
|K |=α |I |=β j,k=1

whenever f = fK,I dz K ∧ d z̄ I ∈ LCp,1 (∂∗ Ω, σ) ⊗ Λα,β .

|K |=α |I |=β
(7.1.85)
The space LC (∂∗ Ω, σ) ⊗ Λα,β can be thought of as a partial Sobolev space of
p,1

(α, β)-differential forms on the geometric measure theoretic boundary of Ω which is

well-adapted to the family of operators ∂τCj k 1≤ j,k ≤n .

7.2 Integration by Parts Formulas Involving the ∂ Operator on

Sets of Locally Finite Perimeter

We begin by briefly discussing the algebraic formalism associated with the ∂-operator
¯
(for more on this topic see, e.g., [13], [59], [88]). To facilitate the subsequent dis-
cussion we find it useful to introduce the following piece of notation. Given an
arbitrary open Ω subset of Cn and an arbitrary subspace𝒳 of D (Ω), we agree to
denote by 𝒳 ⊗ Λα,β the space of differential forms u = |I |=α, |J |=β u I,J dz I ∧ d z̄ J
798 7 Applications to Analysis in Several Complex Variables

with each u I,J belonging to the space 𝒳. Whenever dealing with differential forms
whose coefficients are actually functions, earlier operations with forms with complex
coefficients are now understood in a pointwise sense.
To start in earnest, recall that the exterior derivative operator d in R2n may be
decomposed as
n n
d= ∂x j dx j ∧ · + ∂y j dy j ∧ ·
j=1 j=1

n n
= ∂z j dz j ∧ · + ∂z̄ j d z̄ j ∧ · = ∂ + ∂,
¯ (7.2.1)
j=1 j=1

where, as customary, we have set

n n
∂¯ := ∂z̄ j d z̄ j ∧ · and ∂ := ∂z j dz j ∧ · (7.2.2)
j=1 j=1

for the standard d-bar operator and its complex conjugate, respectively. Their prin-
cipal symbols are given by
i 0,1 i
Sym(∂;
¯ ξ) := ξ ∧ · and Sym(∂; ξ) := ξ 1,0 ∧ ·
2 2 (7.2.3)
for each ξ ∈ R2n .

For each α, β ∈ {0, 1 . . . , n} and each open subset Ω of Cn , the operators

∂¯ : D (Ω) ⊗ Λα,β −→ D (Ω) ⊗ Λα,β+1,

(7.2.4)
∂ : D (Ω) ⊗ Λα,β −→ D (Ω) ⊗ Λα+1,β,

are well-defined, linear, and continuous. Explicitly, if

u= u M,I dz M ∧ d z̄ I ∈ D (Ω) ⊗ Λα,β, (7.2.5)

|M |=α |I |=β

it follows that
n
¯ = (−1)α
∂u
jI
εJ ∂z̄ j u M,I dz M ∧ d z̄ J , (7.2.6)
|J |=β+1 |M |=α |I |=β j=1

and
n
jM
∂u = εK ∂z j u M,I dz K ∧ d z̄ I . (7.2.7)
|K |=α+1 |M |=α |I |=β j=1

Note that
¯ = ∂ ū for each u ∈ D (Ω) ⊗ Λα,β .
∂u (7.2.8)
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 799

Also, the fact that d ◦ d = 0 forces (via simple degree considerations)

∂ ◦ ∂ = 0, ∂¯ ◦ ∂¯ = 0, ∂ ◦ ∂¯ + ∂¯ ◦ ∂ = 0. (7.2.9)

Next, if we set
ϑ := − ∗ ∂ ∗ and ϑ̄ := − ∗ ∂∗
¯ (7.2.10)
then, for each α, β ∈ {0, 1 . . . , n} and each open subset Ω of Cn , the operators

ϑ : D (Ω) ⊗ Λα,β −→ D (Ω) ⊗ Λα,β−1,

(7.2.11)
ϑ̄ : D (Ω) ⊗ Λα,β −→ D (Ω) ⊗ Λα−1,β,

are well-defined, linear, and continuous. More transparently, if α, β ∈ {0, 1, . . . , n}

then for each open set Ω ⊆ Cn and for each form

ψ= ψ M,J dz M ∧ d z̄ J ∈ D (Ω) ⊗ Λα,β (7.2.12)

|M |=α, |J |=β

we have
n
ϑψ = 2(−1)α+1
jI
εJ ∂z j ψ M,J dz M ∧ d z̄ I . (7.2.13)
|M |=α |I |=β−1 |J |=β j=1

Also,

∂ = ∗ϑ∗, ∂¯ = ∗ϑ̄∗, ϑ ◦ ϑ = 0, ϑ̄ ◦ ϑ̄ = 0, ϑ ◦ ϑ̄ + ϑ̄ ◦ ϑ = 0. (7.2.14)

Consider next an arbitrary open set Ω ⊆ Cn and fix some α, β ∈ {0, 1, . . . , n}.

Given two arbitrary forms u ∈ 𝒞1 (Ω) ⊗ Λα,β−1 and w ∈ 𝒞1 (Ω) ⊗ Λα,β , we may then
compute

d(u ∧ ∗w̄) = (du) ∧ ∗w̄ + (−1)α+β+1 u ∧ d(∗w̄)

= (∂u + ∂u)
¯ ∧ ∗w̄ + u ∧ ∗ ∗ (∂ + ∂)(∗
¯ w̄)

= (∂u) ∧ ∗w̄ + (∂u)

¯ ∧ ∗w̄ − u ∧ ∗ϑ w̄ − u ∧ ∗ϑ̄ w̄
       
= ∂u, w dV + ∂u, ¯ w dV − u, ϑ̄w dV − u, ϑw dV
C C C C
   
= ∂u, w dV − u, ϑw dV .
¯ (7.2.15)
C C

Above, the first equality is the Leibniz product formula for the exterior derivative
operator, the second equality is implied by the decomposition in (7.2.1) and item (1)
in Lemma 7.1.6, the third equality uses (7.2.10), the third equality is a consequence
of item (5) in Lemma 7.1.6, and the final equality is seen from simple degree
considerations (taking into account (7.2.4), (7.2.11), and (7.1.45)). In turn, from
(7.2.15) and a most elementary version of Stokes’ theorem we conclude that
800 7 Applications to Analysis in Several Complex Variables
∫ ∫ ∫
   
∂u,
¯ w dV − u, ϑw dV = d(u ∧ ∗w̄) = 0
C C
Ω Ω Ω
whenever u ∈ 𝒞1c (Ω) ⊗ Λα,β−1 and w ∈ 𝒞1 (Ω) ⊗ Λα,β .

This shows that

¯  = ϑ̄ and ∂  = ϑ
(∂) (7.2.16)
in the sense that the real transpose of ∂¯ (respectively, ∂) acting on (α, β − 1)-forms
is the operator ϑ̄ (respectively, ϑ) acting on (α, β)-forms. Let us also note that, as is
apparent from (7.2.16), [68, (1.7.17)-(1.7.18)], (7.2.3), and (7.1.58),
i i
Sym(ϑ; ξ) = − ξ 1,0 ∨ · and Sym(ϑ̄; ξ) = − ξ 0,1 ∨ ·
2 2 (7.2.17)
for each given real vector ξ ∈ R2n .

Below, we discuss a basic integration by parts formula for the ∂¯ operator in a very
general setting.

Theorem 7.2.1 Let Ω be an open nonempty proper subset of Cn ≡ R2n with a lower
Ahlfors regular boundary, and such that σ := H 2n−1 ∂Ω is a doubling measure on
∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Fix an aperture parameter
κ ∈ (0, ∞) and assume that

u ∈ Lloc
1
(Ω, L 2n ) ⊗ Λα,β and w ∈ Lloc
1
(Ω, L 2n ) ⊗ Λα,β+1, (7.2.18)

where α, β ∈ {0, 1, . . . , n}, satisfy

Nκ u < ∞ and Nκ w < ∞ at σ-a.e. point on ∂Ω,

Nκ u · Nκ w belongs to the space L 1 (∂Ω, σ),
κ−n.t. κ−n.t.
the nontangential traces u∂Ω , w ∂Ω exist σ-a.e. on ∂nta Ω, (7.2.19)
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λα,β+1, ϑw ∈ L 1 (Ω, L 2n ) ⊗ Λα,β,
∂u loc loc
   
and ∂u, w − u, ϑw belongs to L 1 (Ω, L 2n ),
¯
C C

with all partial differential operators considered in the sense of distributions in Ω.

In the case when Ω is an exterior domain make the additional assumption that there
exists λ ∈ (1, ∞) such that
∫
|u|C |w|C dL 2n = o(R) as R → ∞. (7.2.20)
B(0,λ R)\B(0,R)

Then
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 801
∫  ∫ 
1 κ−n.t. κ−n.t. 
∂u,
¯ w − u, ϑw dV =
C C ν 0,1 ∧ u∂Ω , w ∂Ω dσ
Ω 2 ∂∗ Ω C

∫  κ−n.t.
1 κ−n.t. 
= u∂Ω , ν 1,0 ∨ w ∂Ω dσ. (7.2.21)
2 ∂∗ Ω C

As a corollary, if the integrability condition in the last line of (7.2.19) is strength-

ened to
   
¯ w ∈ L 1 (Ω, L 2n ) and u, ϑw ∈ L 1 (Ω, L 2n )
∂u, (7.2.22)
C C

then (7.2.21) may be recast as the integration by parts formula

∫ ∫ ∫ 
κ−n.t. κ−n.t. 
¯ w dV = u, ϑw dV + 1
∂u, ν 0,1
∧ u  , w ∂Ω dσ
C C
2 ∂∗ Ω ∂Ω
Ω Ω C
∫ ∫ 
1 κ−n.t. κ−n.t. 
= u, ϑwC dV + u∂Ω , ν 1,0 ∨ w ∂Ω dσ. (7.2.23)
Ω 2 ∂∗ Ω C

For example, all hypotheses demanded in the first two lines of (7.2.19) are satisfied

if we assume Nκ u ∈ L p (∂Ω, σ) and Nκ w ∈ L p (∂Ω, σ) for some integrability
exponents p, p ∈ [1, ∞] with 1/p + 1/p = 1. In such a scenario, it follows that
κ  −n.t. κ  −n.t.
for any other κ  > 0 the nontangential traces u∂Ω , w ∂Ω exist at σ-a.e. point on
∂nta Ω and are actually independent of κ .
By taking conjugates, a number of related versions of (7.2.21) may be established
from it. For example, (7.2.21) and (7.1.44) it follows that if the differential forms
u, w are as in (7.2.18)-(7.2.20) then
∫  ∫ 
1 κ−n.t. κ−n.t. 
ϑw, uC − w, ∂u
¯
C dV = − ν 1,0 ∨ w ∂Ω , u∂Ω dσ
Ω 2 ∂∗ Ω C

∫ 
1 κ−n.t. κ−n.t. 
=− w ∂Ω , ν 0,1 ∧ u∂Ω dσ. (7.2.24)
2 ∂∗ Ω C

Proof of Theorem 7.2.1 The idea is to invoke [68, Theorem 1.7.2] for the choice
D := ∂¯ (and with w replaced by w̄). As may seen from (7.2.16), such a choice
implies that D = ϑ̄. Since in the present setting (−i)Sym(D; ν) may be identified
with (−i)Sym(∂; ¯ ν) = 1 ν 0,1 ∧ (cf. (7.2.3)), formula (7.2.21) then follows directly
2
from [68, (1.7.24)]. 
Our next goal is to introduce a partial Sobolev space of differential forms on the
geometric measure theoretic boundary of a set of locally finite perimeter in Cn which
is well-adapted to the ∂-formalism.
¯ In order to be able to do so, we first make the
definition below.
Definition 7.2.2 Let Ω be a set of locally finite perimeter in R2n ≡ Cn . Denote by ν its
geometric measure theoretic outward unit normal, and abbreviate σ := H 2n−1 ∂Ω.
Also, fix a pair of arbitrary degrees α, β ∈ {0, 1, . . . , n}. In this context, say that
802 7 Applications to Analysis in Several Complex Variables

1 (∂ Ω, σ) ⊗ Λα,β has the property that ∂¯ f ∈ L 1 (∂ Ω, σ) ⊗ Λα,β+2 provided

f ∈ Lloc ∗ τ loc ∗
1 (∂ Ω, σ) ⊗ Λα,β+2 such that
there exists some g ∈ Lloc ∗
∫ ∫
 0,1 
ν ∧ f , ϑψ dσ = g, ψC dσ for all ψ ∈ 𝒞∞ c (C ) ⊗ Λ
n α,β+2
. (7.2.25)
C
∂∗ Ω ∂∗ Ω

In the setting of Definition 7.2.2, from [68, (5.2.6)] and [68, Lemma 3.6.4]
we see that σ∂∗ Ω is a complete, locally finite, Borel-regular measure on ∂∗ Ω
(where the latter set is endowed with the topology inherited from the Euclidean
ambient). Granted this, [68, Proposition 3.7.2] applies and gives that the form g is
unambiguously defined by the condition imposed in (7.2.25). In order to stress the
dependence of such g on the given form f we shall henceforth employ the notation

∂¯τ f := g. (7.2.26)

In the same setting as in Definition 7.2.2, given any p ∈ [1, ∞] we then define
p, ∂¯
Lα,β τ (∂∗ Ω, σ) := f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β : ∂¯τ f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 .
(7.2.27)

In this notation, we may now recast (7.2.25) as the integration by parts formula on
the boundary
∫ ∫
 0,1   
ν ∧ f , ϑψ dσ = ∂¯τ f , ψ dσ
C C
∂∗ Ω ∂∗ Ω
(7.2.28)
p, ∂¯
for all f ∈ Lα,β τ (∂∗ Ω, σ) and ψ ∈ 𝒞∞
c (C ) ⊗ Λ
n α,β+2 .

p, ∂¯
We also agree to equip the space Lα,β τ (∂∗ Ω, σ) with the natural norm

f p, ∂¯
Lα, β τ (∂∗ Ω,σ)
:= f L p (∂∗ Ω,σ)⊗Λα, β + ∂¯τ f L p (∂∗ Ω,σ)⊗Λα, β+2 (7.2.29)

p, ∂¯
for each f ∈ Lα,β τ (∂∗ Ω, σ).

Proposition 7.2.3 Consider a set of locally finite perimeter Ω ⊆ R2n ≡ Cn . Abbre-

viate σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit
normal. Also, pick some function f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β with exponent p ∈ [1, ∞]
and α, β ∈ {0, 1, . . . , n}. Then
p, ∂¯ p, ∂¯
f ∈ Lα,β τ (∂∗ Ω, σ) if and only if ftan,C ∈ Lα,β τ (∂∗ Ω, σ), (7.2.30)

and if either of the above memberships holds then

∂¯τ f = ∂¯τ ( ftan,C ) at σ-a.e. point on ∂∗ Ω. (7.2.31)

Proof This is a direct consequence of (7.2.25) and (7.1.69). 

7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 803

p, ∂¯
As is apparent from our next result, the partial Sobolev space Lα,β τ (∂∗ Ω, σ)
introduced in (7.2.27) turns out to be rather rather rich.

Proposition 7.2.4 Suppose Ω is a set of locally finite perimeter in R2n ≡ Cn . As usual,

abbreviate σ := H 2n−1 ∂Ω, and pick p ∈ [1, ∞] along with α, β ∈ {0, 1, . . . , n}.
Then one has the inclusion
p, ∂¯
LCp,1 (∂∗ Ω, σ) ⊗ Λα,β ⊆ Lα,β τ (∂∗ Ω, σ) (7.2.32)

and for each differential form

f = fK,I dz K ∧ d z̄ I ∈ LCp,1 (∂∗ Ω, σ) ⊗ Λα,β (7.2.33)

|K |=α |I |=β

there holds

∂¯τ f = 2−1 (−1)α+1 (∂τCj k fK,I )dz K ∧ d z̄ j ∧ d z̄k ∧ d z̄ I . (7.2.34)

|K |=α |I |=β 1≤ jk ≤n

In particular, corresponding to α = β = 0, one has

∂¯τ f = − ∂τCj k f d z̄ j ∧ d z̄k for each f ∈ LCp,1 (∂∗ Ω, σ). (7.2.35)

1≤ j<k ≤n


Proof Pick ψ = |M |=α, |J |=β+2 ψ M,J dz M ∧d z̄ J ∈ 𝒞∞c (C ) ⊗ Λ
n α,β+2 arbitrary along

with some increasing multi-indices K ∈ {1, . . . , n} and I ∈ {1, . . . , n} β . Then, if

α

ν denotes the geometric measure theoretic outward unit normal to Ω, the (K, I)-th
component of ν 0,1 ∨ ϑψ on ∂∗ Ω is given at σ-a.e. point by
n
  jI
ν 0,1 ∨ ϑψ K,I =2 ε M (ν 0,1 ) j (ϑψ)K, M
|M |=β+1 j=1

n n
= 4(−1)α+1
jI
ε M εJk M (ν 0,1 ) j ∂zk ψK,J
|M |=β+1 j=1 |J |=β+2 k=1

= 4(−1)α+1
k jI
εJ (ν 0,1 ) j ∂zk ψK,J
|J |=β+2 1≤ jk ≤n
 
= 2(−1)α+1
k jI
εJ (ν 0,1 ) j ∂zk − (ν 0,1 )k ∂z j ψK,J
|J |=β+2 1≤ jk ≤n
 
= 2(−1)α+1
k jI
εJ (νC ) j ∂z̄k − (νC )k ∂z̄ j ψ K,J
|J |=β+2 1≤ jk ≤n

= 2(−1)α+1
k jI
εJ (∂τCj k ψ K,J ). (7.2.36)
|J |=β+2 1≤ jk ≤n
804 7 Applications to Analysis in Several Complex Variables

Above, the first equality is a consequence of (7.1.57) and the second equality follows
from (7.2.13). Furthermore, the third identity in (7.2.36) is a consequence of the fact
that
jI k jI k jI
εJk M ε M = εJk M εk M = εJ . (7.2.37)
|M |=β+1 |M |=β+1
k jI jk I
The fourth identity in (7.2.36) follows from the fact that εJ = −εJ while the
fifth one follows from (7.1.61). Finally, the sixth equality follows easily from the
properties of complex conjugation while the last equality uses the definition (7.1.5)
of the complex tangential derivative operator ∂τCj k .
Hence, for any given differential form f as in (7.2.33) we may write
∫
 0,1 
ν ∧ f , ϑψ dσ
C
∂∗ Ω
∫
 
= f , ν 0,1 ∨ ϑψ dσ
C
∂∗ Ω
∫
α+β
=2 fK,I (ν 0,1 ∨ ϑψ)K,I dσ
|K |=α |I |=β ∂∗ Ω
∫
α+β+1 α+1 k jI
=2 (−1) εJ fK,I (∂τCj k ψ K,J ) dσ,
|K |=α |I |=β |J |=β+2 1≤ jk ≤n ∂∗ Ω

(7.2.38)

where the first identity above is a consequence of item (3) in Lemma 7.1.6, the
second equality follows from (7.1.43), and the last equality uses (7.2.36).
To proceed, let us define

g := (∂τCj k fK,I )dz K ∧ d z̄ j ∧ d z̄k ∧ d z̄ I ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 .

|K |=α |I |=β 1≤ jk ≤n
(7.2.39)
Making use of (7.2.38), (7.1.9), (7.1.14), and (7.2.39) we may then compute
∫
 0,1 
ν ∧ f , ϑψ dσ
C
∂∗ Ω
∫
= 2α+β+1 (−1)α
k jI
εJ (∂τCj k fK,I )ψ K,J dσ
|K |=α |I |=β |J |=β+2 1≤ jk ≤n ∂∗ Ω
∫
= 2α+β+1 (−1)α+1 gK,J ψ K,J dσ
|K |=α |J |=β+2 ∂∗ Ω
∫
= 2−1 (−1)α+1 g, ψC dσ. (7.2.40)
∂∗ Ω
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 805

p, ∂¯
In light of (7.2.27), this proves that f ∈ Lα,β τ (∂∗ Ω, σ) (ultimately establishing the
inclusion in (7.2.32)) and that ∂¯τ f = 2−1 (−1)α+1 g ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 , (hence
(7.2.34) holds). 
We augment the result established in Proposition 7.2.4 by showing that the inclu-
sion in (7.2.32) actually becomes an equality when β = 0.

Proposition 7.2.5 Assume Ω is a set of locally finite perimeter in R2n ≡ Cn and,

as usual, abbreviate σ := H 2n−1 ∂Ω. Then for each degree α ∈ {0, 1, . . . , n} and
integrability exponent p ∈ [1, ∞] one has
p, ∂¯
Lα,0 τ (∂∗ Ω, σ) = LCp,1 (∂∗ Ω, σ) ⊗ Λα,0 . (7.2.41)

Proof The right-to-left inclusion in (7.2.41) is a special case of Proposition 7.2.4. To

 p, ∂¯
prove the opposite inclusion in (7.2.41), pick f = |K |=α fK dz K ∈ Lα,0 τ (∂∗ Ω, σ).
Also, select a test function ϕ ∈ 𝒞∞c (C ) along with two indices jo, k o ∈ {1, . . . , n}
n

satisfying jo < k o and some ordered array Ko ∈ {1, . . . , n}α . Consider next the
differential form

ψ := ϕ̄ dz Ko ∧ d z̄ jo ∧ d z̄ko ∈ 𝒞∞
c (C ) ⊗ Λ
n α,2
. (7.2.42)

Then, if ν denotes the geometric measure theoretic outward unit normal to Ω, by

specializing formula (7.2.38) to the present choice of f and ψ we obtain
∫ ∫
 0,1 
ν ∧ f , ϑψ dσ = 2α+2 (−1)α+1 fKo ∂τCj o k o ϕ dσ. (7.2.43)
C
∂∗ Ω ∂∗ Ω

p, ∂¯
Next, in light of (7.2.27), the membership of f to the space Lα,0 τ (∂∗ Ω, σ) ensures the
 
existence of some g = |K |=α 1≤ j<k ≤n gK, jk dz K ∧ d z̄ j ∧ d z̄k ∈ L p (∂∗ Ω, σ) ⊗ Λα,2
with the property that
∫ ∫ ∫
 0,1  α+2
ν ∧ f , ϑψ dσ = g, ψC dσ = 2 gKo, jo ko ϕ dσ, (7.2.44)
C
∂∗ Ω ∂∗ Ω ∂∗ Ω

where the last equality uses (7.1.42) and the specific formats of g and ψ. In concert,
(7.2.43) and (7.2.44) ultimately prove that
∫ ∫
(−1)α+1 fKo ∂τCj o k o ϕ dσ = g jo ko ϕ dσ, ∀ϕ ∈ 𝒞∞ c (C ).
n
(7.2.45)
∂∗ Ω ∂∗ Ω

Having established (7.2.45), from Definition 7.1.1 (cf. also (7.1.8)) we may then con-
clude that ∂τCj o k o fKo = (−1)α gKo, jo ko ∈ L p (∂∗ Ω, σ). Given that fKo ∈ L p (∂∗ Ω, σ)
and the indeces jo, k o are arbitrary, Definition 7.1.2 shows that we necessarily have
p,1
fKo ∈ LC (∂∗ Ω, σ). Ultimately, since Ko has been arbitrarily chosen this finishes
the proof of (7.2.41). 
806 7 Applications to Analysis in Several Complex Variables

In applications, it is useful to know that a version of the integration by parts

formula on the boundary recorded in (7.2.28) also holds if the compact support
condition for the intervening test functions is replaced by suitable integrability
conditions. The specific format of this result is recorded below.

Proposition 7.2.6 Let Ω be a set of locally finite perimeter in R2n ≡ Cn . Abbreviate

σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit nor-
mal. Also, fix p ∈ [1, ∞] and α, β ∈ {0, 1, . . . , n}, along with an open neighborhood
p, ∂¯
U ⊆ Cn of ∂Ω. Consider a differential form f ∈ Lα,β τ (∂∗ Ω, σ) and suppose

ψ ∈ 𝒞1 (U) ⊗ Λα,β+2 is such that ψ ∂∗ Ω ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2


 (7.2.46)
and (ϑψ)∂∗ Ω ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+1,


where p ∈ [1, ∞] is the Hölder conjugate exponent of p. Then

∫ ∫
 0,1   
ν ∧ f , ϑψ dσ = ∂¯τ f , ψ dσ. (7.2.47)
C C
∂∗ Ω ∂∗ Ω

In particular, (7.2.47) holds for every ψ ∈ 𝒞1 (U) ⊗ Λα,β+2 in the case when ∂∗ Ω
is bounded.

Proof Pick a scalar-valued function θ ∈ 𝒞∞ c (C ) with the property that θ ≡ 1 on

n

B(0, 1) and, for each R > 0, define θ R (z) := θ(z/R) for every z ∈ Cn . The idea is to
write
∫ ∫
 0,1   0,1 
ν ∧ f , ϑψ dσ = lim ν ∧ f , θ R ϑψ dσ
C R→∞ C
∂∗ Ω ∂∗ Ω
∫
 
= lim ν 0,1 ∧ f , ϑ(θ R ψ) dσ
R→∞ C
∂∗ Ω
∫
 
− lim ν 0,1 ∧ f , (∂θ R ) ∨ ψ dσ
R→∞ C
∂∗ Ω
∫
 
= lim ∂¯τ f , θ R ψ dσ
R→∞ C
∂∗ Ω
∫
 
= lim ∂¯τ f , ψ dσ. (7.2.48)
R→∞ C
∂∗ Ω

Indeed, the first equality above is a consequence of Lebesgue’s Dominated Conver-

gence Theorem (bearing in mind that the last condition in (7.2.46) and the fact that
f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β ). The second equality uses a Leibniz product rule to the
effect that

ϑ(θ R ψ) = θ R ϑψ − (∂θ R ) ∨ ψ (7.2.49)

7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 807

which, in turn, is seen from (7.2.10) and items (1) and (4) in in Lemma 7.1.6. The
third equality is based on two facts. First, since θ R ψ ∈ 𝒞1c (U) ⊗ Λα,β+2 for each
R > 0, we may use a standard mollifier argument which ultimately allows us to
invoke (7.2.28) in order to conclude that
∫ ∫
 0,1   
ν ∧ f , ϑ(θ R ψ) dσ = ∂¯τ f , θ R ψ dσ for each R > 0. (7.2.50)
C C
∂∗ Ω ∂∗ Ω
 
Second, since ∂θ R  ≤ CR−1 at σ-a.e. point in ∂∗ Ω, we may estimate
C

∫   
 
 ν 0,1 ∧ f , (∂θ R ) ∨ ψ dσ  (7.2.51)
C
∂∗ Ω

≤ CR−1 f L p (∂∗ Ω,σ)⊗Λ0, β ψ ∂∗ Ω 
L p (∂∗ Ω,σ)⊗Λ0, β+2
→ 0 as R → ∞.

As such, the second limit in the second line of (7.2.48) vanishes. There remains
to observe that the last equality in (7.2.48) is implied by Lebesgue’s Dominated
Convergence
 Theorem (bearing in mind that ∂¯τ f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 and
ψ ∂∗ Ω ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 .



In our next proposition we study how the operator ∂¯τ interacts with the pointwise
nontangential boundary trace operator.

Proposition 7.2.7 Consider an open nonempty proper subset Ω of Rn with a lower

Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on
∂Ω; in particular, Ω is a set of locally finite perimeter. Denote by ν the geometric
measure theoretic outward unit normal to Ω. Also, fix α, β ∈ {0, 1, . . . , n}, an aper-
ture parameter κ ∈ (0, ∞), an integrability exponent p ∈ [1, ∞], and a truncation
parameter ε > 0. In this context, pick a differential form u satisfying

u ∈ Lloc
1
(Ω, L 2n ) ⊗ Λα,β, ∂u
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λ0,β+1,
loc (7.2.52)

Nκε u ∈ L p (∂Ω, σ), Nκε (∂u)

¯ ∈ L p (∂Ω, σ), and (7.2.53)
κ−n.t. κ−n.t.
the traces u∂Ω and (∂u)
¯ 
∂Ω
exist σ-a.e. on ∂nta Ω. (7.2.54)

Then κ−n.t.
u∂Ω belongs to the space Lα,β τ (∂∗ Ω, σ) and
p, ∂¯

 κ−n.t.   κ−n.t.  (7.2.55)

∂¯τ u∂Ω = −ν 0,1 ∧ (∂u)
¯ 
∂Ω
on ∂∗ Ω.

In particular, at σ-a.e. point on ∂∗ Ω one has

 κ−n.t.    κ−n.t.    κ−n.t. 
ν 1,0 ∨ ∂¯τ u∂Ω = −ν 1,0 ∨ ν 0,1 ∧ (∂u)
¯ 
∂Ω
¯ 
= −2 (∂u) ∂Ω
. (7.2.56)
tan,C

Moreover, there exists a finite constant C > 0, independent of u and ε > 0, such that
808 7 Applications to Analysis in Several Complex Variables
κ−n.t.  
u∂Ω p, ∂¯
≤ C Nκε u L p (∂Ω,σ)
+ Nκε (∂u)
¯
L p (∂Ω,σ)
. (7.2.57)
Lα, β τ (∂Ω,σ)

Proof With the function Φε as in [68, Lemma 6.1.2], define Ψε := 1 − Φε . Then

[68, (6.1.5)-(6.1.6)] imply

Ψε ∈ 𝒞∞ (Ω), 0 ≤ Ψε (x) ≤ 1 for all x ∈ Ω, (7.2.58)

 
sup (∂ α Ψε )(x) ≤ Cα ε − |α | for each α ∈ N0n, (7.2.59)
x ∈Ω

Ψε ≡ 0 on Ω \ Oε and Ψε ≡ 1 on Oε/N . (7.2.60)

Consequently, Ψε ∈ W 1,∞ (Ω) and supp (∇Ψε ) ⊆ Oε \ Oε/N . If we now introduce

uε := Ψε u in Ω, (7.2.61)

then
uε ∈ Lloc
1
(Ω, L 2n ) ⊗ Λα,β and ∂u
¯ ε = ∂Ψ
¯ ε ∧ u + Ψε ∂u.
¯ (7.2.62)
Note that we have Nκ uε ≤ Nκε u
pointwise on ∂Ω which, thanks to (7.2.53) and [68,
(8.2.26)], gives that Nκ uε ∈ L p (∂Ω, σ).
 Similarly, since the indentity in (7.2.62)
implies Nκ (∂u¯ ε ) ≤ C Nκε u + Nκε (∂u)
¯ pointwise on ∂Ω, we may also infer that
Nκ (∂u
¯ ε ) ∈ L p (∂Ω, σ). Given that uε = u and ∂u ¯ ε = ∂u¯ in Oε/N , we conclude
that at σ-a.e. point on ∂nta Ω we also have
κ−n.t. κ−n.t. κ−n.t. κ−n.t.
uε ∂Ω = u∂Ω and (∂u
¯ ε )
∂Ω
= ( ¯ 
∂u) ∂Ω
. (7.2.63)

To proceed, consider U := z ∈ Cn : dist(z, ∂Ω) < ε/N , which is an open

neighborhood of ∂Ω in Cn , and pick an arbitrary form ψ ∈ 𝒞1c (U) ⊗ Λα,β+2 . Then
∞ (∂Ω, σ). Moreover,
ϑψ has compact support, ϑ(ϑψ) = 0 in U, and Nκ (ϑψ) ∈ Lcomp
estimate [68, (8.6.51) in Proposition 8.6.3] ensures that
 
∂u
¯ ε, ϑψ ∈ L 1 (Ω, L 2n ). (7.2.64)
C

From these we conclude that conditions (7.2.19) for the forms uε (playing the role of
u) and w := ϑψ are satisfied, hence Theorem 7.2.1 permits us to write (on account
of (7.2.63))
∫ ∫
 0,1  κ−n.t.    0,1  κ−n.t.  
ν ∧ u∂Ω , ϑψ dσ = ν ∧ uε ∂Ω , ϑψ dσ
C C
∂∗ Ω ∂∗ Ω
∫
 
=2 ∂u
¯ ε, ϑψ dV . (7.2.65)
C
Ω

Integrating by parts once more, now moving the partial derivatives from ψ to uε
(i.e., employing (7.2.21) with ∂u
¯ ε playing the role of u and with ψ playing the role
of w, since once again the conditions (7.2.19) can be easily verified), we obtain
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 809
∫ ∫
    κ−n.t.  
2 ∂u
¯ ε, ϑψ dV = − ¯ ε
ν 0,1 ∧ ∂u , ψ dσ
C ∂Ω C
Ω ∂∗ Ω
∫
  κ−n.t.  
=− ¯ 
ν 0,1 ∧ ∂u , ψ dσ, (7.2.66)
∂Ω C
∂∗ Ω

κ−n.t.
where the last equality uses (7.2.63). All in all, introducing f := u∂Ω , the above
reasoning shows that f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β and
∫ ∫
 0,1  κ−n.t.  
ν ∧ f , ϑψC dσ = −
0,1
ν ∧ ∂u ¯  , ψ dσ (7.2.67)
∂Ω C
∂∗ Ω ∂∗ Ω

for all ψ ∈ 𝒞1c (U) ⊗ Λα,β+2 . In addition, thanks to the hypotheses made in (7.2.53)-
 κ−n.t. 
(7.2.54), the membership ν 0,1 ∧ ∂u ¯  ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 holds. On account
∂Ω
of 7.2.27, this establishes (7.2.55). With (7.2.55) in hand, all remaining conclusions
in the statement of the proposition now easily follow. 

To proceed, in the definition below we introduce a very general notion of CR-form

(aka, Cauchy-Riemann differential form).

Definition 7.2.8 Suppose Ω is an open set of locally finite perimeter in R2n ≡ Cn .

Abbreviate σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward
unit normal. Also, pick an arbitrary pair of degrees α, β ∈ {0, 1, . . . , n}. In this
1 (∂ Ω, σ) ⊗ Λα,β a CR-form provided
context, call f ∈ Lloc ∗
∫
 0,1 
ν ∧ f , ϑψ dσ = 0 for each ψ ∈ 𝒞∞ c (C ) ⊗ Λ
n α,β+2
. (7.2.68)
C
∂∗ Ω

A standard mollifier argument shows that an equivalent definition is obtained by

considering test forms ψ ∈ 𝒞1c (U) ⊗ Λα,β+2 , for some U ⊆ Cn open neighborhood
of ∂Ω. Moreover, by (7.2.28) and [68, Proposition 3.7.2] (whose applicability in the
present setting is ensures by [68, (5.2.6)] and [68, Lemma 3.6.4]) we see that for
each α, β ∈ {0, 1, . . . , n} the following holds:

if Ω ⊆ R2n ≡ Cn is an open set of locally finite perimeter, and

f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β with p ∈ [1, ∞] and σ := H 2n−1 ∂Ω, then (7.2.69)
p, ∂¯
f is a CR-form ⇐⇒ f ∈ Lα,β τ (∂Ω, σ) and ∂¯τ f = 0 on ∂∗ Ω.

In particular, corresponding to the case when α = β = 0, from (7.2.69), Propo-

sition 7.2.4, and Proposition 7.2.5 we see that a complex-valued scalar function
p,1
f ∈ L p (∂∗ Ω, σ) with p ∈ [1, ∞] is a CR-form provided f ∈ LC (∂∗ Ω, σ) and
∂τ j k f = 0 for each j, k ∈ {1, . . . , n}. In view of Definition 7.1.4 and (7.1.22) this
C

ultimately shows that

in the scalar case, the concept of CR-form naturally
(7.2.70)
reduces to the notion of CR-function.
810 7 Applications to Analysis in Several Complex Variables

Let us momentarily digress for the purpose of considering in greater detail the
particular case when Ω is a bounded 𝒞1 domain in R2n ≡ Cn . In such a scenario, if ν
and σ retain their earlier significance, then the latter may be naturally identified with
the “volume" element dσ on the 𝒞1 manifold ∂Ω. Lastly, denote by ι : ∂Ω → Cn
the canonical inclusion mapping. In particular, ι∗ denotes the pull-back from Cn to
∂Ω. Next, fix α, β ∈ {0, 1, . . . , n} and assume a differential form f ∈ 𝒞0 (∂Ω) ⊗ Λα,β
has been given. Below, we shall tacitly identify f with an arbitrary extension of
itself to a form with continuous coefficients in an open neighborhood of ∂Ω. Pick an
arbitrary test form ψ ∈ 𝒞∞ c (C ) ⊗ Λ
n α,β+2 and introduce

ϕ := 2(−1)β ∗ ψ̄ ∈ 𝒞∞
c (C ) ⊗ Λ
n n−α,n−2−β
. (7.2.71)

The stage is set for us to compute

 0,1     
ν ∧ f , ϑψ dσ = 2 ν ∧ f , ϑψ dσ = 2ι∗ f ∧ ∗ϑψ
C C

   
= 2(−1)β+2 ι∗ f ∧ ∗ϑ ∗ ∗ψ = 2(−1)β ι∗ f ∧ ∂(∗
¯ ψ̄)
 
= ι∗ f ∧ ∂ϕ
¯ on ∂Ω, (7.2.72)

where the first equality is based on (7.1.63) and simple degree considerations, the
second equality uses the fact that, generally speaking, for any continuous -form u
and ( + 1)-form ω in a neighborhood of ∂Ω we have (cf., e.g., [79, Lemma 3.1])
     
ι∗ (u ∧ ∗ω̄) = ν ∧ u∂Ω , ω∂Ω dσ, (7.2.73)
C

the third equality in (7.2.72) is implied by item (1) in Lemma 7.1.6, the fourth
equality relies on (7.2.14), and the final equality simply takes (7.2.71) into account.
In concert, (7.2.68) and (7.2.71)-(7.2.72) prove the following:

if Ω is a bounded 𝒞1 domain in R2n ≡ Cn and f ∈ 𝒞0 (∂Ω) ⊗ Λα,β ,

having f be a CR-form in∫ the sense
 of Definition
 7.2.8 becomes equiv-
¯ = 0 for each differential form (7.2.74)
alent to the demand that ∂Ω ι∗ f ∧ ∂ϕ
ϕ ∈ 𝒞∞ c (C ) ⊗ Λ
n n−α,n−2−β .

In particular, corresponding to the scalar case α = β = 0, if Ω is a bounded 𝒞1

domain in R2n ≡ Cn , then having some given f ∈ 𝒞0 (∂Ω) be a CR-function in the
sense of Definition 7.1.4 becomes equivalent to asking that
∫
 
ι∗ f ∂ϕ
¯ = 0 for each ϕ ∈ 𝒞∞ c (C ) ⊗ Λ
n n,n−2
. (7.2.75)
∂Ω

Retaining the assumption that Ω is a bounded 𝒞1 domain in R2n ≡ Cn , we may

further rephrase the condition that a given f ∈ 𝒞1 (∂Ω) is a CR-function as follows.
Identifying f with an extension of itself to a function of class 𝒞1 in an open
neighborhood of ∂Ω, for each ϕ ∈ 𝒞∞ c (C ) ⊗ Λ
n n,n−2 we may write
7.2 Integrating ∂ by Parts on Sets of Locally Finite Perimeter 811
∫ ∫ ∫ ∫
       
ι∗ f ∂ϕ
¯ = ι∗ f dϕ = ι∗ d( f ϕ) − ι∗ df ∧ ϕ
∂Ω ∂Ω ∂Ω ∂Ω
∫ ∫ ∫
     
= d∂Ω ι∗ ( f ϕ) − ι∗ df ∧ ϕ = − ι∗ df ∧ ϕ , (7.2.76)
∂Ω ∂Ω ∂Ω

using (7.2.1) (and keeping in mind that ϕ is of degree (n, n − 2)), the product
rule for d, the fact that pull-back commutes with the exterior derivative operator
(d∂Ω denoting its version on the manifold ∂Ω), and Stokes’ Theorem. Expressing
ϕ ∈ 𝒞∞ c (C ) ⊗ Λ
n n,n−2 as dz ∧ · · · ∧ dz ∧ ψ with ψ ∈ 𝒞∞ (Cn ) ⊗ Λ0,n−2 arbitrary,
1 n c
from (7.2.76) and (7.2.70) we ultimately conclude that

if Ω is a bounded 𝒞1 domain in R2n ≡ Cn and f ∈ 𝒞1 (∂Ω), then

having f be a CR-function in the sense of Definition 7.1.4 becomes
(7.2.77)
equivalent to the demand that the pull-back to the manifold ∂Ω of the
differential form d f ∧ dz1 ∧ · · · ∧ dzn vanishes identically on ∂Ω.
The latter vanishing property is sometimes referred to as Wirtinger’s condition (cf.
[90]).
The point of view adopted in (7.2.75) is very popular. For example, this is how
the concept of CR-function is introduced within the class of continuous functions
defined on boundaries of 𝒞1 domains in Cn in [59, Definition 8.17, p. 44], [52, (6.9),
p. 62], and [53, Definition 3.1.2, p. 76]. We wish to emphasize that this commonly
held point of view (based on Stokes formula and pull-back to ∂Ω, assumed to have
a manifold structure of at least class 𝒞1 ) is no longer practical if ∂Ω is lacking
regularity. In this regard, the progress that we register here is dispensing with any
manifold structure assumption on the boundary of the given set Ω.
We close this section by proving that, in a suitable context, nontangential boundary
traces of null-solutions of the ∂-operator
¯ are CR-forms.

Proposition 7.2.9 Let Ω ⊆ Cn ≡ R2n be a UR domain. Abbreviate σ := H 2n−1 ∂Ω.

Also, fix a pair of degrees α, β ∈ {0, 1, . . . , n}, an exponent p ∈ [1, ∞), an aperture
κ > 0, and a truncation parameter ε > 0. In this context, if u ∈ 𝒞1 (Ω) ⊗ Λα,β is a
differential form satisfying

Nκε u ∈ L p (∂Ω, σ), ∂u

¯ = 0 in Ω,
κ−n.t. (7.2.78)
and u∂Ω exists σ-a.e. on ∂Ω,

then κ−n.t.  κ−n.t. 

u∂Ω ∈ Lα,β τ (∂Ω, σ) and ∂¯τ u∂Ω = 0.
p, ∂¯
(7.2.79)
In particular,
κ−n.t.
u∂Ω is a CR-form. (7.2.80)

Proof The claims in (7.2.79) are immediate from (7.2.55). The claim in (7.2.80)
then follows from (7.2.79) and (7.2.69). 
812 7 Applications to Analysis in Several Complex Variables

7.3 The Bochner-Martinelli Integral Operator

Having fixed some n ∈ N, we begin by considering double forms in thevariables ζ, z

in Cn . Given integers α, β, γ, δ ∈ {0, 1, . . . , n}, a double form of type (α, β), (γ, δ)
is an object which may be expressed as
IJ    
Θ= θK L dζ ∧ d ζ
I ¯J ⊗ dz K ∧ d z̄ L , (7.3.1)
|I |=α |J |=β |K |=γ |L |=δ

where each θ KIJ

L is a function in the variables ζ, z. These objects can then be made
subject to natural operations with ordinary forms, simply by viewing them as differ-
ential forms in the variable ζ whose coefficients are themselves differential forms in
the variable z, or the other way around. For example, the complex conjugate of Θ in
(7.3.1) is
 I J  
Θ := θK
IJ
L d ζ ∧ dζ
¯ ⊗ d z̄ K ∧ dz L
|I |=α |J |=β |K |=γ |L |=δ

 J   
= (−1)αβ+γδ θK
IJ ¯ I ⊗ dz L ∧ d z̄ K ,
L dζ ∧ d ζ (7.3.2)
|J |=β |I |=α |L |=δ |K |=γ
 
which is a double form of type (β, α), (δ, γ) . We may also apply differential opera-
tors, such as ∂,
¯ ∂, ϑ, and ϑ̄, in either the variable ζ, or the variable z (each time we
agree to indicate this by appending the variable to the operator as a subscript). For
example, if Θ is as in (7.3.1) then
n
 IJ   ¯   
∂¯ζ Θ := ∂ζ¯j θ K L d ζ j ∧ dζ ∧ d ζ
I ¯J ⊗ dz K ∧ d z̄ L
j=1 |I |=α |J |=β |K |=γ |L |=δ

 n 
jJ  IJ     
= (−1)α ε M ∂ζ¯j θ K L dζ I ∧ dζ¯ M ⊗ dz K ∧ d z̄ L ,
|I |=α |K |=γ j=1 |J |=β
|M |=β+1 |L |=δ
(7.3.3)
 
which is a double form of type (α, β +1), (γ, δ) . Similar conventions are in effect for
the action of the Hodge star operator ∗, the exterior product, and the interior product
with one forms (considered either in the variable ζ, or the variable z) on double
forms. One natural venue through which double forms arise is taking the tensor

product of two ordinary forms. Concretely, suppose u = |I |=α, |J |=β u I,J dζ I ∧ dζ¯J

is an ordinary (α, β)-form and w =   wK, L dz ∧ d z̄ is an ordinary
|K |=γ, |L |=δ
K L

(γ, δ)-form, we define u ⊗ w as the (α, β), (γ, δ) double form given by
   
u ⊗ w := u I,J wK, L dζ I ∧ dζ¯J ⊗ dz K ∧ d z̄ L . (7.3.4)
|I |=α |J |=β |K |=γ |L |=δ
7.3 The Bochner-Martinelli Integral Operator 813

Going further, we wish to extend the Hermitian inner product (7.1.43) to the
case when one of the forms participating in the inner product is now a double form.

Specifically, if Θ is as in (7.3.1) and u = |I |=α, |J |=β u I,J dz I ∧ d z̄ J is an ordinary
(α , β )-form we define
 
u, Θ := 0 if (α , β )  (α, β) (7.3.5)
C

and, corresponding to the case when (α , β ) = (α, β), we set

   
u, Θ := 2α+β u I,J θ K
IJ
L d z̄ ∧ dz
K L
C
|K |=γ |L |=δ |I |=α |J |=β
 
= 2α+β (−1)γδ u I,J θ K
IJ
L dz ∧ d z̄ .
L K
(7.3.6)
|L |=δ |K |=γ |I |=α |J |=β

In particular,

the inner product

 of an ordinary
 (α, β)-form with a double
(7.3.7)
form of type (α, β), (γ, δ) is an ordinary (δ, γ)-form.

In fact, there is a natural Hermitian inner product for double forms of the following
sort. If the double form Θ is as in (7.3.1) and

=        
Θ θ IJ dζ I ∧ dζ¯J ⊗ dz K ∧ d z̄ L , (7.3.8)
KL
|
I |=α | J|=β | K
 |=γ | L
 |=δ

we set
 
 Θ := 2α+β+γ+δ
Θ, 
θK L θK L .
IJ IJ
(7.3.9)
C
|I |=α |J |=β |K |=γ |L |=δ
 
In this vein, it is useful to note that for any double form Θ of type (α, β), (γ, δ) , and
any two ordinary forms, u of type (α, β) and w of type (γ, δ), we have
     
u ⊗ w, Θ = w, ū, Θ . (7.3.10)
C
C C

Finally, the Hermitian inner product of double forms defined in (7.3.9) may be further
extended to the case when, for some given open subset Ω ⊆ Cn , the coefficients of
 in (7.3.8) belong to D (Ω) while the coefficients of Θ in (7.3.1) belong to D(Ω)
Θ
by setting
   IJ IJ 

D  (Ω) Θ, Θ := 2α+β+γ+δ 
D  (Ω) θ K L , θ K L D(Ω) . (7.3.11)
D(Ω)
|I |=α |J |=β |K |=γ |L |=δ

 in (7.3.8) belong to ℰ(Ω) while the coefficients

Similarly, when the coefficients of Θ
of Θ in (7.3.1) belong to ℰ(Ω), we define
814 7 Applications to Analysis in Several Complex Variables

]quad
   

ℰ (Ω) Θ, Θ ℰ(Ω) := 2
α+β+γ+δ 
θK L , θ K L ℰ(Ω) . (7.3.12)
IJ IJ
 ℰ (Ω)
|I |=α |J |=β |K |=γ |L |=δ


Based
  definitions, it may be checked that if Θ is a double form of type
on these
(α, β), (γ, δ) with coefficients in the space D (Ω) and Θ is a double form of type
(α, β + 1), (γ, δ) with coefficients in D(Ω) then
   
¯  
D  (Ω) ∂ζ Θ, Θ D(Ω) = D  (Ω) Θ, ϑζ Θ D(Ω), (7.3.13)

plus a similar formula when D (Ω), D(Ω) are replaced by ℰ(Ω) and ℰ(Ω), respec-
tively.
Changing topics, consider the complex Laplacian in Cn defined as

 := ∂ϑ
¯ + ϑ ∂.
¯ (7.3.14)

In particular, from (7.3.14), (7.2.9), and (7.2.14) we see that

 2
 = ∂¯ + ϑ . (7.3.15)

A direct computation also shows that

n
 = −2 ∂zk ∂z̄k = − 12 Δ (7.3.16)
k=1

where Δ denotes the real Laplacian in R2n , i.e.,

n
Δ := (∂x2j + ∂y2j ) in R2n . (7.3.17)
j=1

We are interested in manufacturing a fundamental solution for . To this end, for

each z, ξ ∈ Cn with z  ξ consider
⎧
⎪ 1
⎪
⎪ − ln |ζ − z| 2 for n = 1,
⎨ 2π
⎪
En (ζ, z) := (7.3.18)
⎪
⎪ (n − 2)!
⎪
⎪ |ζ − z| 2−2n for n ≥ 2.
⎩ 2π n

Since the surface area of the unit ball in R2n is given by ω2n−1 = 2π n /(n − 1)! it
follows that En is −2 times the standard fundamental solution for the real Laplacian
Δ in R2n . Given the goal we have in mind, this shows that the choice of En (ζ, z)
 Next, for each α, β ∈ {0, 1, . . . , n} consider the double
in (7.3.18) is indeed natural.
form of type (α, β), (β, α) given by
7.3 The Bochner-Martinelli Integral Operator 815

Γα,β (ζ, z) := 2−α−β En (ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (d z̄ J ∧ dz I )

|J |=α |I |=β

= 2−α−β (−1)αβ En (ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J ). (7.3.19)

|J |=α |I |=β

Then, by design,

ζ Γα,β (ζ, z) = 2−α−β (−1)αβ δz (ζ) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J ), (7.3.20)

|J |=α |I |=β

z Γα,β (ζ, z) = 2−α−β (−1)αβ δζ (z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J ), (7.3.21)

|J |=α |I |=β

where δz (ζ) the Dirac distribution in the variable ζ ∈ Cn with mass at z, and where
δζ (z) the Dirac distribution in the variable z ∈ Cn with mass at ζ. Moreover, it is
clear from (7.3.19) that
Γα,β (z, ζ) = Γβ,α (ζ, z). (7.3.22)
We claim that we also have

ϑζ Γα,β (ζ, z) = ∂z Γα,β−1 (ζ, z), (7.3.23)

∂ζ Γα,β (ζ, z) = ϑz Γα,β+1 (ζ, z). (7.3.24)

Indeed, by (7.2.12)-(7.2.13) and (7.3.19) we have

ϑζ Γα,β (ζ, z) = 21−α−β (−1)αβ+α+1 × (7.3.25)

n
jM  
× εI ∂ζ j En (ζ, z) (dζ J ∧ dζ¯ M ) ⊗ (dz I ∧ d z̄ J ),
|M |=β−1 j=1 |J |=α |I |=β

whereas (7.2.7)-(7.2.5) imply that

∂z Γα,β−1 (ζ, z) = 21−α−β (−1)α(β−1) × (7.3.26)

n
jI  
× εK ∂z j En (ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz K ∧ d z̄ J ).
|K |=β j=1 |J |=α |I |=β−1
   
Upon interchanging M with I, and noting that ∂ζ j En (ζ, z) = −∂z j En (ζ, z) for
each j, it follows that the two right-hand sides of (7.3.25) and (7.3.26) match. This
establishes (7.3.23), and (7.3.24) follows from it by interchanging z, ζ (bearing in
mind (7.3.22)), then readjusting notation.
The Bochner-Martinelli kernel for (0, β)-forms in Cn with β ∈ {0, 1, . . . , n}
is then defined as the double differential form

Knβ (ζ, z) := − ∗ ∂ζ Γ0,β (ζ, z), (7.3.27)

816 7 Applications to Analysis in Several Complex Variables

where the Hodge star isomorphism is applied in the variable ζ. If Ω is a bounded

𝒞1 domain in R2n ≡ Cn , then ∂Ω is a 𝒞1 -smooth submanifold of R2n ≡ Cn . In
such a context, the Bochner-Martinelli integral operator is defined on a continuous
(0, β)-form f on ∂Ω as
∫  
B0,β f (z) := ι∗ζ f (ζ) ∧ Knβ (ζ, z) , ∀z ∈ Cn \ ∂Ω, (7.3.28)
∂Ω

where ι : ∂Ω → Cn is the canonical inclusion and ι∗ζ indicates pull-back in the

variable ζ. In view of (7.2.73), an equivalent way of defining the Bochner-Martinelli
integral operator on a continuous (0, β)-form f on the boundary of a bounded 𝒞1
domain Ω ⊆ Cn is
∫
 
B0,β f (z) := − ν(ζ) ∧ f (ζ), ∂¯ζ Γ0,β (ζ, z) dσ(ζ), (7.3.29)
C
∂Ω

at each z ∈ Cn \ ∂Ω. As explained before, it is this expression which we find most

suitable for extending the Bochner-Martinelli integral operator to situations when Ω
is lacking smoothness in a traditional sense.
While this is done in Definition 7.3.1 below, for now we make one other ob-
servation regarding the structure of (7.3.29). Specifically, in the case when f is a
Λ0,β -valued function, using the second identity in (7.1.63) we may write
1,β 0,β+1
ν(ζ) ∧ f (ζ) = 12 ν 1,0 (ζ) ∧ f (ζ) + 12 ν 0,1 (ζ) ∧ f (ζ) ∈ Λζ + Λζ . (7.3.30)
 
From (7.3.19) we see that the double form Γ0,β(ζ, z) is of type  (0, β), (β, 0) and,
hence, (7.3.3) gives that ∂¯ζ Γ0,β (ζ, z) is of type (0, β + 1), (β, 0) . Based on degree
considerations (see (7.3.7)) we may therefore conclude that, whenever Ω is a bounded
𝒞1 domain in R2n ≡ Cn with outward unit normal ν and surface measure σ, we may
further express the action of the Bochner-Martinelli integral operator (7.3.29) on a
continuous Λ0,β -valued function f defined on ∂Ω as
∫
1  0,1 
B0,β f (z) = − ν (ζ) ∧ f (ζ), ∂¯ζ Γ0,β (ζ, z) dσ(ζ), (7.3.31)
2 ∂Ω C

at each z ∈ Cn \ ∂Ω, where the Hermitian inner product ·, ·C is taken in the sense
of (7.3.6).
In the most general geometric and algebraic setting, we shall introduce the
(higher-degree) Bochner-Martinelli integral operator as follows.

Definition 7.3.1 Given a set Ω ⊆ R2n ≡ Cn (where n ∈ N) of locally finite perimeter,

denote by ν its geometric measure theoretic outward unit normal, and abbreviate
σ := H 2n−1 ∂Ω. Also, fix α, β ∈ {0, 1, . . . , n}. In this setting, define the action of
the (higher-degree) Bochner-Martinelli integral operator Bα,β associated with Ω on
an arbitrary σ-measurable function f : ∂∗ Ω → Λα,β Cn satisfying
7.3 The Bochner-Martinelli Integral Operator 817
∫
| f (ζ)|C
dσ(ζ) < +∞ (7.3.32)
1 + |ζ | 2n−1
∂∗ Ω

according to
∫
1  
Bα,β f (z) := − ν 0,1 (ζ) ∧ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ), (7.3.33)
2 C
∂∗ Ω

at each z ∈ Cn \ ∂Ω, where the Hermitian inner product ·, ·C is taken in the sense
of (7.3.6).

As regards Definition 7.3.1, note that the conditions imposed on f ensure that
the integral in the right-hand side of (7.3.33) is absolutely convergent for each
z ∈ Cn \ ∂Ω. In fact,

Bα,β f ∈ 𝒞∞ (Ω) ⊗ Λα,β and Δ(Bα,β f ) = 0 in Ω, (7.3.34)

where Δ := ∂12 + · · · + ∂2n

2 is the Laplacian in R2n . Moreover, in the case when ∂Ω

is an upper Ahlfors regular set, [68, Lemma 7.2.1] implies that any differential form
f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β with p ∈ [1, ∞) satisfies (7.3.32).
The reader is also reminded that (1.4.199) elaborates on the relationship between
the higher-degree Bochner-Martinelli integral operator Bα,β from Definition 7.3.1
and the class of double layer potential operators, constructed according to the general
recipe described in (1.4.36).
The main point of our next proposition is that, when acting on differential forms
belonging to an appropriate space on the boundary, the Bochner-Martinelli integral
operator interacts well with the ∂-operator.
¯

Proposition 7.3.2 Let Ω ⊆ R2n ≡ Cn be an open set with the property that ∂Ω is
upper Ahlfors regular; in particular, Ω is a set of locally finite perimeter. Abbreviate
σ := H 2n−1 ∂Ω and denote by ν its geometric measure theoretic outward unit nor-
p, ∂¯
mal. Also, fix α, β ∈ {0, 1, . . . , n}. Then for each differential form f ∈ Lα,β τ (∂∗ Ω, σ)
with p ∈ [1, ∞) one has
∫
1  
∂(Bα,β f )(z) =
¯ (∂¯τ f )(ζ), ∂¯ζ Γα,β+1 (ζ, z) dσ(ζ) (7.3.35)
2 ∂∗ Ω C

at every z ∈ Cn \ ∂Ω.

Proof By taking ∂¯ in both sides in the formula (7.3.33) it follows that for every
z ∈ Cn \ ∂Ω we may write
818 7 Applications to Analysis in Several Complex Variables
∫
 1  
∂¯ Bα,β f )(z) = − ν 0,1 (ζ) ∧ f (ζ), ∂¯ζ ∂z Γα,β (ζ, z) dσ(ζ)
2 C
∂∗ Ω
∫
1  
=− ν 0,1 (ζ) ∧ f (ζ), ∂¯ζ ϑζ Γα,β+1 (ζ, z) dσ(ζ)
2 C
∂∗ Ω
∫
1  
=− ν 0,1 (ζ) ∧ f (ζ), (ζ − ϑζ ∂¯ζ )Γα,β+1 (ζ, z) dσ(ζ)
2 C
∂∗ Ω
∫
1  
= ν 0,1 (ζ) ∧ f (ζ), ϑζ ∂¯ζ Γα,β+1 (ζ, z) dσ(ζ)
2 C
∂∗ Ω
∫
1  
= (∂¯τ f )(ζ), ∂¯ζ Γα,β+1 (ζ, z) dσ(ζ), (7.3.36)
2 C
∂∗ Ω

where the second equality follows from (7.3.23), the third equality is a consequence
of (7.3.14), the fourth equality uses (7.3.20) (bearing in mind that z ∈ Cn \ ∂Ω and
ζ ∈ ∂∗ Ω ⊆ ∂Ω ensure that we have z  ζ), and the last equality may be justified
making use of Proposition 7.2.6 (whose present applicability is ensured by (7.3.19)
and [68, Lemma 7.2.1]). This completes the proof of the proposition. 

A significant consequence of Proposition 7.3.2 is that, when applied to CR-forms,

the Bochner-Martinelli integral operator yields null-solution of the ∂-operator.
¯

Proposition 7.3.3 Consider an open set Ω ⊆ R2n ≡ Cn with the property that ∂Ω is
upper Ahlfors regular; in particular, Ω is a set of locally finite perimeter. Abbreviate
σ := H 2n−1 ∂Ω and fix α, β ∈ {0, 1, . . . , n} along with p ∈ [1, ∞). In this context, if
f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β is a CR-form then

Bα,β f is ∂-closed
¯ in Ω, (7.3.37)

i.e., Bα,β f is a null-solution of the ∂-operator

¯ in Ω.

Proof This is a direct consequence of (7.2.69) and Proposition 7.3.2. 

Pressing on, in the proposition below we introduce an integral operator which is

closely connected to the Bochner-Martinelli integral operator, through the action of
∂¯ and ∂¯τ (cf. (7.3.41)), and study its mapping properties and boundary behavior.

Proposition 7.3.4 Let n ∈ N and suppose Ω ⊆ R2n ≡ Cn is a nonempty open set

such that ∂Ω is a UR set; in particular, Ω has locally finite perimeter. Denote by ν its
geometric measure theoretic outward unit normal and abbreviate σ := H 2n−1 ∂Ω.
Also, fix a pair of arbitrary degrees α, β ∈ {0, 1, . . . , n}, along with an aperture
 ) 
parameter κ > 0. Lastly, for each differential form f ∈ L 1 ∂Ω, 1+σ(ζ |ζ | 2n−1
⊗ Λα,β+1 ,
define
7.3 The Bochner-Martinelli Integral Operator 819
∫
 
Aα,β f (z) := − f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ), ∀z ∈ Ω, (7.3.38)
C
∂Ω
∫
 
Aα,β f (z) := − lim+ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ for σ-a.e. z ∈ ∂Ω. (7.3.39)
ε→0 C
ζ ∈∂Ω
|z−ζ |>ε

Then the following properties hold.

 ) 
(i) For each differential form f ∈ L 1 ∂Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β+1 (hence, in particular,
for each differential form f ∈ L p (∂Ω, σ) ⊗ Λα,β+1 with p ∈ [1, ∞)) the limit in
(7.3.39) exists for σ-a.e. z ∈ ∂Ω. Moreover,

Aα,β f ∈ 𝒞∞ (Ω) ⊗ Λα,β and Δ(Aα,β f ) = 0 in Ω, (7.3.40)

where Δ := ∂12 + · · · + ∂2n

2 is the Laplacian in R2n .
p, ∂¯
(ii) Pick an arbitrary differential form f ∈ Lα,β τ (∂∗ Ω, σ) with integrability exponent
p ∈ [1, ∞), and canonically regard ∂¯τ f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 as being a form
in L p (∂Ω, σ) ⊗ Λα,β+2 by extending it by zero from ∂∗ Ω to the entire topological
boundary ∂Ω. Then, with this convention, one has

∂(B
¯ α,β f ) = − 1 Aα,β+1 (∂¯τ f ) in Ω.
2 (7.3.41)

Also, given any f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β with p ∈ [1, ∞), upon agreeing to regard
ν 0,1 ∧ f as a form in L p (∂Ω, σ) ⊗ Λα,β+1 by extending it by zero from ∂∗ Ω to
∂Ω, one has
Bα,β f = 12 Aα,β (ν 0,1 ∧ f ) in Ω. (7.3.42)
(iii) The operators

Aα,β : L p (∂Ω, σ) ⊗ Λα,β+1 −→ L p (∂Ω, σ) ⊗ Λα,β, p ∈ (1, ∞), (7.3.43)

Aα,β : L 1 (∂Ω, σ) ⊗ Λα,β+1 −→ L 1,∞ (∂Ω, σ) ⊗ Λα,β, (7.3.44)

are well-defined, linear, and bounded. In addition, corresponding to the end-

point p = ∞, the operator Aα,β induces a linear and bounded mapping

Aα,β : L ∞ (∂Ω, σ) ⊗ Λα,β+1 −→ BMO(∂Ω, σ) ⊗ Λα,β . (7.3.45)

The operator (7.3.39) further extends to a linear and bounded mapping

Aα,β : H p (∂Ω, σ) ⊗ Λα,β+1 −→ L p (∂Ω, σ) ⊗ Λα,β, n−1

n < p ≤ 1. (7.3.46)
 n−1 
(iv) For each p ∈ n , ∞ there exists a constant C = C(Ω, p, κ, n) ∈ (0, ∞) such
that
 820 7 Applications to Analysis in Several Complex Variables

Nκ (Aα,β f ) L p (∂Ω,σ)⊗Λα, β
≤C f L p (∂Ω,σ)⊗Λα, β+1 if 1 < p < ∞, (7.3.47)

Nκ (Aα,β f ) L 1,∞ (∂Ω,σ)⊗Λα, β

≤C f L 1 (∂Ω,σ)⊗Λα, β+1 , (7.3.48)

Nκ (Aα,β f ) L p (∂Ω,σ)⊗Λα, β
≤C f H p (∂Ω,σ)⊗Λα, β+1 if n−1
n < p ≤ 1.
(7.3.49)

(v) For each p ∈ (1, ∞) there exists a constant C = C(Ω, p, n) ∈ (0, ∞) with the
property that for each differential form f ∈ L p (∂Ω, σ) ⊗ Λα,β+1 one has
∫  1/p
|(∇Aα,β f )|Cp dist(·, ∂Ω) p−1 dL 2n ≤ C f L p (∂Ω,σ)⊗Λα, β+1 , (7.3.50)
Ω

where ∇Aα,β f stands for the collection of all first-order partial derivatives of
all components of the differential form Aα,β f . In particular, corresponding to
p = 2, one has the following L 2 -square function estimate
∫ ∫
|(∇Aα,β f )|C2 dist(·, ∂Ω) dL 2n ≤ C | f |C2 dσ. (7.3.51)
Ω ∂Ω

(vi) There exists a constant C = C(Ω, n) ∈ (0, ∞) with the property that for each
differential form f ∈ L ∞ (∂Ω, σ) ⊗ Λα,β+1 one has the following Carleson
measure estimate:
 ∫  1/2
1  
sup   (∇Aα,β ) f 2 dist(·, ∂Ω) dL 2n
x ∈∂Ω, r >0 σ ∂Ω ∩ B(x, r)
C
B(x,r)∩Ω

≤C f L ∞ (∂Ω,σ)⊗Λα, β+1 . (7.3.52)

(vii) For each form f ∈ L p (∂Ω, σ) ⊗ Λα,β+1 with 1 ≤ p < ∞ the following jump-
formula holds:
κ−n.t. 1

Aα,β f  = ν 1,0 ∨ f + Aα,β f for σ-a.e. point in ∂∗ Ω. (7.3.53)
∂Ω 2
(viii) If Ω is actually a bounded NTA domain in R2n ≡ Cn with an Ahlfors regular
boundary and p ∈ (1, ∞), then for each q ∈ n+1/p
n
, ∞ the operators

Aα,β : L p (∂Ω, σ) ⊗ Λα,β+1 −→ F1/p (Ω) ⊗ Λα,β,

p,q

Aα,β : L p (∂Ω, σ) ⊗ Λα,β+1 −→ B1/p (Ω) ⊗ Λα,β,

p, p
(7.3.54)
Aα,β : L 2 (∂Ω, σ) ⊗ Λα,β+1 −→ H 1/2 (Ω) ⊗ Λα,β,

are well-defined, linear, and bounded.

In preparation for the proof of the above proposition, we introduce some notation.
Given an open set of locally finite perimeter Ω ⊆ R2n ≡ Cn , let σ := H 2n−1 ∂Ω. In
7.3 The Bochner-Martinelli Integral Operator 821

this setting, for each j ∈ {1, . . . , n} we define the complex Riesz transform R C, j
as the integral operator acting on σ-measurable functions f : ∂Ω → C satisfying
∫
| f (ζ)|
dσ(ζ) < +∞ (7.3.55)
1 + |ζ | 2n−1
∂Ω

according to
∫
2 z j − ζj
R C, j f (z) := f (ζ) dσ(ζ), for all z ∈ Ω. (7.3.56)
ω2n−1 |z − ζ | 2n
∂Ω

Its complex conjugate is the operator R cj given by R C,

c f := R
j
¯
C, j f . Explicitly,

∫
2 z̄ j − ζ¯j
R C,
c
j f (z) := f (ζ) dσ(ζ), for all z ∈ Ω. (7.3.57)
ω2n−1 |z − ζ | 2n
∂Ω

When ∂Ω is a UR set it also makes sense to define their boundary-to-boundary

versions, i.e.,
∫
2 z j − ζj
RC, j f (x) := lim+ f (ζ) dσ(ζ) (7.3.58)
ε→0 ω2n−1 |z − ζ | 2n
|z−ζ |>ε
ζ ∈∂Ω

 ) 
for σ-a.e. z ∈ ∂Ω, where f ∈ L 1 ∂Ω, 1+σ(ζ
|ζ | 2n−1
. Finally, consider the complex
C, j f for 1 ≤ j ≤ n. Specifically,
c f := R
conjugate of the operators RC, j , that is, RC, ¯
j
for each j ∈ {1, . . . , n} we have
∫
2 z̄ j − ζ¯j
c
RC, f (x) = lim+ f (ζ) dσ(ζ) (7.3.59)
j ε→0 ω2n−1 |z − ζ | 2n
|z−ζ |>ε
ζ ∈∂Ω

 ) 
for σ-a.e. z ∈ ∂Ω, where f ∈ L 1 ∂Ω, 1+σ(ζ
|ζ | 2n−1
.
Proof of Proposition 7.3.4 To set the stage, we note that
822 7 Applications to Analysis in Several Complex Variables

∂¯ζ Γα,β (ζ, z) (7.3.60)

 n 
= ∂ζ¯j dζ¯j ∧ · 2−α−β (−1)αβ En (ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J )
j=1 |J |=α |I |=β

n
= 2−α−β (−1)αβ ∂ζ¯j [En (ζ, z)](dζ¯j ∧ dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J )
j=1 |J |=α |I |=β

= 2−α−β (−1)αβ+α ×
n
jI
× εK ∂ζ¯j [En (ζ, z)](dζ J ∧ dζ¯K ) ⊗ (dz I ∧ d z̄ J ).
j=1 |J |=α |I |=β |K |=β+1

 
In particular, ∂¯ζ Γβ (ζ, z) is a double form of type (α, β + 1), (β, α) . As such, if the
differential form f ∈ L p (∂Ω, σ) ⊗ Λα,β+1 , with p ∈ [1, ∞), is explicitly written as

f (ζ) = fJ,K (ζ) dζ J ∧ dζ¯K for σ-a.e. ζ ∈ ∂Ω, (7.3.61)

|J |=α |K |=β+1

then
 
− f (ζ), ∂¯ζ Γα,β (ζ, z) (7.3.62)
C

n
= 2(−1)αβ+α+1
jI
εK ∂ζ¯j [En (ζ, z)] fJ,K (ζ) d z̄ I ∧ dz J
j=1 |J |=α |I |=β |K |=β+1

n
= 2(−1)α+1
jI
εK ∂ζ j [En (ζ, z)] fJ,K (ζ) dz J ∧ d z̄ I
j=1 |J |=α |I |=β |K |=β+1

2(−1)α
n
jI ζ¯j − z̄ j
= εK fJ,K (ζ) dz J ∧ d z̄ I .
ω2n−1 j=1 |J |=α |I |=β |K |=β+1
|ζ − z| 2n

Consequently, in terms of the complex Riesz transforms discussed in (7.3.56)-

(7.3.59), for f as in (7.3.61) we have
n
Aα,β f = (−1)α+1
jI
εK R C,
c
j fJ,K dz ∧ d z̄ in Ω,
J I
(7.3.63)
j=1 |J |=α |I |=β |K |=β+1

and
n
Aα,β f = (−1)α+1
jI
εK RC,
c
j fJ,K dz ∧ d z̄ on ∂Ω.
J I
(7.3.64)
j=1 |J |=α |I |=β |K |=β+1
7.3 The Bochner-Martinelli Integral Operator 823

The first claim in item (i) in the statement of the proposition is then clear from
(7.3.64) and [70, Theorem 2.3.2]. The second claim in item (i) is seen straight
from definitions (with [68, Lemma 7.2.1] ensuring the absolute convergence of the
integral defining A β f in (7.3.38)). Next, the first claim in item (ii) follows from
Proposition 7.3.2, in view of (7.3.38), while the second claim in item (ii) is implied
by Definition 7.3.1 and (7.3.38).
Going further, all claims in item (iii) are direct consequences of (7.3.64) and [70,
Theorem 2.3.2] (bearing in mind that the complex Riesz transforms do fall under
the scope of the latter theorem). Likewise, all claims in items (iv), (v), and (vi) are
immediate consequences of (7.3.63) and the corresponding results for the complex
Riesz transforms implied by [70, Theorem 2.4.1].
As regards the jump-formula in item (vii), we first observe that in the present
context [70, Theorem 2.5.1] implies that for each j ∈ {1, . . . , n} and each scalar
 ) 
function g ∈ L 1 ∂Ω, 1+σ(ζ
|ζ | 2n−1
, the j-th complex Riesz transform satisfies

  κ−n.t.
R C, j g  = −(νC ) j g + RC, j g at σ-a.e. point on ∂∗ Ω. (7.3.65)
∂Ω

Taking complex conjugates this yields (after re-denoting ḡ by g)

  κ−n.t.
R C,
c
jg  = −(νC ) j g + RC,
c
j g at σ-a.e. point on ∂∗ Ω. (7.3.66)
∂Ω
 ) 
To proceed, fix an arbitrary differential form f ∈ L 1 ∂Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λ0,β+1 . Based
on (7.3.63)-(7.3.64), (7.1.57), (7.3.66), and (7.1.61), we may then compute

  κ−n.t. n
jI  c  κ−n.t. J
Aα,β f  = (−1)α+1 εK R C, j J,K 
f dz ∧ d z̄ I
∂Ω ∂Ω
j=1 |J |=α |I |=β |K |=β+1

n
= (−1)α
jI
εK (νC ) j fJ,K dz J ∧ d z̄ I
j=1 |J |=α |I |=β |K |=β+1

n
+ (−1)α+1
jI
εK RC,
c
j fJ,K dz ∧ d z̄
J I

j=1 |J |=α |I |=β |K |=β+1

= 1
2 ν 1,0 ∨ f + Aα,β f at σ-a.e. point on ∂∗ Ω. (7.3.67)

This proves (7.3.53). Lastly, the claim in item (viii) is a consequence of item (4) of
[70, Theorem 2.4.1] (bearing in mind (7.3.40)). 
Our main result pertaining to the nature of the Bochner-Martinelli operator Bα,β
defined in (7.3.33) is contained in the theorem below.

Theorem 7.3.5 Let n ∈ N and suppose Ω ⊆ R2n ≡ Cn is a nonempty open set

with the property that ∂Ω is a UR set; in particular, Ω has locally finite perimeter.
824 7 Applications to Analysis in Several Complex Variables

Abbreviate σ := H 2n−1 ∂Ω and denote by ν the geometric measure theoretic out-
ward unit normal to Ω. Also, fix a pair of arbitrary degrees α, β ∈ {0, 1, . . . , n},
along with an aperture parameter κ > 0. Finally, for each differential form
 ) 
f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β , define the principal-value (boundary-to-boundary)
Bochner-Martinelli integral operator
∫
1  0,1 
Bα,β f (z) := − lim+ ν (ζ) ∧ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ) (7.3.68)
2 ε→0 C
ζ ∈∂∗ Ω
|z−ζ |>ε

for σ-a.e. point z ∈ ∂Ω. Then the following properties hold.

 ) 
(i) For each differential form f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β (hence, in particular,
for each differential form f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β with p ∈ [1, ∞)) the limit in
(7.3.39) exists for σ-a.e. z ∈ ∂Ω. Moreover, the operators

Bα,β : L p (∂∗ Ω, σ) ⊗ Λα,β −→ L p (∂Ω, σ) ⊗ Λα,β, p ∈ (1, ∞), (7.3.69)

Bα,β : L 1 (∂∗ Ω, σ) ⊗ Λα,β −→ L 1,∞ (∂Ω, σ) ⊗ Λα,β, (7.3.70)

are well-defined, linear, and bounded. In addition, if ∂Ω is compact, then corre-

sponding to the end-point p = ∞ the operator Bα,β induces a linear and bounded
mapping

Bα,β : L ∞ (∂∗ Ω, σ) ⊗ Λα,β −→ BMO(∂Ω, σ) ⊗ Λα,β . (7.3.71)

(ii) For each p ∈ [1, ∞) there exists a constant C = C(Ω, p, κ, n) ∈ (0, ∞) such that
for each differential form f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β one has

Nκ (Bα,β f ) L p (∂Ω,σ)⊗Λα, β
≤C f L p (∂∗ Ω,σ)⊗Λα, β if 1 < p < ∞, (7.3.72)

and, corresponding to p = 1,

Nκ (Bα,β f ) L 1,∞ (∂Ω,σ)⊗Λα, β

≤C f L 1 (∂∗ Ω,σ)⊗Λα, β . (7.3.73)

(iii) For each p ∈ [1, ∞) there exists some constant C = C(Ω, p, κ, n) ∈ (0, ∞) with
the property that if the differential form f belongs to the partial Sobolev space
p, ∂¯
Lα,β τ (∂∗ Ω, σ) then

Nκ (∂B
¯ α,β f )
L p (∂Ω,σ)⊗Λα, β
≤ C ∂¯τ f L p (∂∗ Ω,σ)⊗Λα, β+2 if 1 < p < ∞,
(7.3.74)

and, corresponding to p = 1,

Nκ (∂B
¯ α,β f )
L 1,∞ (∂Ω,σ)⊗Λα, β
≤ C ∂¯τ f L 1 (∂∗ Ω,σ)⊗Λα, β+2 . (7.3.75)

Also, the nontangential boundary limit

7.3 The Bochner-Martinelli Integral Operator 825
κ−n.t.
¯ α,β f 
∂B exists at σ-a.e. point on ∂∗ Ω. (7.3.76)
∂Ω

(iv) For each p ∈ (1, ∞) there exists a constant C = C(Ω, p, n) ∈ (0, ∞) with the
property that for each form f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β one has
∫  1/p
|(∇Bα,β f )|Cp dist(·, ∂Ω) p−1 dL 2n ≤ C f L p (∂∗ Ω,σ)⊗Λα, β , (7.3.77)
Ω

where ∇Bα,β f stands for the collection of all first-order partial derivatives of
all components of the differential form Bα,β f . In particular, corresponding to
p = 2, one has the following L 2 -square function estimate:
∫ ∫
|(∇Bα,β f )|C dist(·, ∂Ω) dL ≤ C
2 2n
| f |C2 dσ. (7.3.78)
Ω ∂∗ Ω

(v) There exists a constant C = C(Ω, n) ∈ (0, ∞) with the property that for each
differential form f ∈ L ∞ (∂∗ Ω, σ)⊗Λα,β one has the following Carleson measure
estimate:
 ∫  1/2
1  
sup   (∇Bα,β ) f 2 dist(·, ∂Ω) dL 2n
x ∈∂Ω, r >0 σ ∂Ω ∩ B(x, r)
C
B(x,r)∩Ω

≤C f L ∞ (∂∗ Ω,σ)⊗Λα, β . (7.3.79)

(vi) If Ω is actually a bounded NTA domain in R2n ≡ Cn with an Ahlfors regular

boundary, then whenever p ∈ (1, ∞) and q ∈ n+1/p
n
, ∞ the operators

Bα,β : L p (∂Ω, σ) ⊗ Λα,β −→ F1/p (Ω) ⊗ Λα,β,

p,q

Bα,β : L p (∂Ω, σ) ⊗ Λα,β −→ B1/p (Ω) ⊗ Λα,β,

p, p
(7.3.80)

Bα,β : L 2 (∂Ω, σ) ⊗ Λα,β −→ H 1/2 (Ω) ⊗ Λα,β,

are well-defined, linear, and bounded.

 ) 
(vii) For each form f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β the following jump-formula holds:
κ−n.t. 1

Bα,β f  = ν 1,0 ∨ (ν 0,1 ∧ f ) + Bα,β f
∂Ω 4
1
= ftan,C + Bα,β f at σ-a.e. point on ∂∗ Ω. (7.3.81)
2
In particular, (with I denoting the identity operator),

if f ∈ Ltan,C (∂∗ Ω, σ) ⊗ Λα,β with 1 ≤ p < ∞ then

p

κ−n.t.   (7.3.82)
Bα,β f ∂Ω = 12 I + Bα,β f at σ-a.e. point on ∂∗ Ω.
 826 7 Applications to Analysis in Several Complex Variables

(viii) Suppose the set Ω  R2n ≡ Cn is actually a UR domain, and fix an arbitrary

 ) 
differential form f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β . Then, if one defines

± 
Ω+ := Ω, Ω− := Rn \ Ω, and Bα,β f := Bα,β f  , (7.3.83)
Ω±

it follows that Ω− is also a UR domain whose topological and geometric measure

theoretic boundaries agree with those of Ω, and whose geometric measure
theoretic outward unit normal is −ν at σ-a.e. point on ∂Ω. In addition, one has
the jump-formulas:
κ−n.t.
± 
Bα,β f = ± 12 ftan,C + Bα,β f at σ-a.e. point on ∂Ω. (7.3.84)
∂Ω

As a consequence,

if Ω is a UR domain and f ∈ Ltan,C (∂∗ Ω, σ) ⊗ Λα,β with 1 ≤ p < ∞

p

κ−n.t.  1
± f

then Bα,β ∂Ω
= ± 2 I + Bα,β f at σ-a.e. point on ∂Ω.
(7.3.85)

Thanks to [70, Theorem 2.6.1], under the assumption that Ω is a UR domain

we also have mapping properties on Morrey spaces which are similar to those for
the double layer operator presented in Theorem 3.3.1. Also, thanks to (1.4.199) and
Theorem 3.3.3, we see that if Ω is a UR domain, p ∈ (1, ∞), and λ ∈ (0, 2n − 1),
then there exists a constant C ∈ (0, ∞) with the property that the fractional Carleson
measure estimate
 ∫  p1
  
sup r −λ ∇ Bα,β f  p dist(·, ∂Ω) p−1 dL 2n
z ∈∂Ω and B(z,r)∩Ω
0<r <2 diam(∂Ω)

≤C f M p, λ (∂Ω,σ)⊗Λα, β (7.3.86)

holds for each differential form f ∈ M p,λ (∂Ω, σ)⊗Λα,β which is complex tangential.
Proof of Theorem 7.3.5 The first claim in item (i) is a consequence of [70, Theo-
 ) 
rem 2.3.2]. Moreover, given any differential form f ∈ L 1 ∂∗ Ω, 1+σ(ζ |ζ | 2n−1
⊗ Λα,β , if
 
we agree to regard ν 0,1 ∧ f as a form in the space L 1 ∂Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β+1 by
extending it by zero from ∂∗ Ω to ∂Ω then (7.3.68) and (7.3.39) imply

Bα,β f = 12 Aα,β (ν 0,1 ∧ f ) on ∂Ω. (7.3.87)

Bearing this in mind, all other claims in the current item (i) are implied by part (iii)
of Proposition 7.3.4. In fact, granted (7.3.87) and given the relationship between the
family of operators Bα,β and Aα,β identified in (7.3.41)-(7.3.42) (and also keeping
in mind that the exterior product with ν 0,1 is a bounded operator on Lebesgue spaces
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 827

on ∂∗ Ω with respect to the measure σ), all claims in the current items (ii)-(vi) are
direct consequences of Proposition 7.3.4.
 ) 
As regards the first claim in item (vii), having fixed f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
⊗ Λα,β
we may invoke (7.3.42), (7.3.53), (7.3.87), and (7.1.68) in order to write
κ−n.t. 1 κ−n.t.
 
Bα,β f  = Aα,β (ν 0,1 ∧ f )
∂Ω 2 ∂Ω

1 1 1,0
= ν ∨ (ν 0,1 ∧ f ) + Aα,β (ν 0,1 ∧ f )
2 2
1
= ftan,C + Bα,β f at σ-a.e. point on ∂∗ Ω. (7.3.88)
2
This proves the jump-formula (7.3.81), and (7.3.82) follows from it. Lastly, all
the claims in item (viii) are consequences of (7.3.81)-(7.3.82) and part (7) of [68,
Lemma 5.10.9]. 

7.4 A Sharp Version of the Bochner-Martinelli-Koppelman

Formula and Related Topics

In its classical form, the Bochner-Martinelli-Koppelman formula involves smooth

domains and differential forms, and here the main goal is to establish a geometri-
cally and analytically sharp version of this basic result. See Theorem 7.4.3 and the
comments preceding its statement. For now, we discuss the following generalization
of the Cauchy-Pompeiu integral representation formula from [70, Theorem 1.1.1] to
functions of several complex variables.

Theorem 7.4.1 Let Ω ⊆ R2n ≡ Cn be an open set with a lower Ahlfors regular
boundary, and with the property that σ := H 2n−1 ∂Ω is a doubling measure on
∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Also, fix a pair of degrees
α, β ∈ {0, 1, . . . , n} along with some aperture parameter κ > 0.
In this context, suppose u ∈ Lloc 1 (Ω, L 2n ) ⊗ Λα,β is a differential form satisfying

(with all partial differential operators considered in the sense of distributions in Ω)

 L 2n (ζ)   L 2n (ζ) 
α,β+1
¯ ∈ L 1 Ω,
∂u ⊗ Λ , ϑu ∈ L 1
Ω, ⊗ Λα,β−1,
1 + |ζ | 2n−1 1 + |ζ | 2n−1 (7.4.1)
κ−n.t.

the nontangential trace u ∂Ω exists at σ-a.e. point on ∂nta Ω,

and ∫
(Nκ u)(ζ)
dσ(ζ) < +∞. (7.4.2)
∂Ω 1 + |ζ | 2n−1
828 7 Applications to Analysis in Several Complex Variables
κ  −n.t.
Then for any κ  > 0 the nontangential trace u∂Ω also exists σ-a.e. on ∂nta Ω
and is actually independent of κ . Moreover, with the dependence on the parameter
κ  dropped, for L 2n -a.e. point z ∈ Ω one has (with absolutely convergent integrals)
∫  
1  n.t. 
u(z) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1  n.t. 
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
+ (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ)
C
Ω
∫  
+ (ϑu)(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ) (7.4.3)
C
Ω

if either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω is

an exterior domain, the same conclusion holds true under the additional assumption
that there exists λ ∈ (1, ∞) such that
⨏
|u|C dL 2n = o(1) as R → ∞. (7.4.4)
B(0,λ R)\B(0,R)

Furthermore, for each z ∈ Cn \ Ω one has

∫  
1  n.t. 
0=− ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1  n.t. 
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
+ (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ)
C
Ω
∫  
+ (ϑu)(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ), (7.4.5)
C
Ω

with the same caveat as in (7.4.4) when Ω is an exterior domain.

Prior to presenting the proof of this theorem we make several comments. First,
if ∂Ω is upper Ahlfors regular then the integrability condition in (7.4.2) is satisfied
whenever Nκ u ∈ L p (∂Ω, σ) for some p ∈ [1, ∞). Moreover, if ∂Ω is actually
compact, then the integrability conditions in (7.4.2) simply reduce to the membership

Nκ u ∈ L 1 (∂Ω, σ). (7.4.6)

7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 829

Second, recall from [68, Lemma 3.5.7] that the integrability conditions in the first
line of (7.4.1) are equivalent with having for L 2n -a.e. point z ∈ Cn the finiteness
condition
∫
|(∂u)(ζ)|
¯
C + |(ϑu)(ζ)|C
dL 2n (ζ) < +∞. (7.4.7)
Ω |ζ − z| 2n−1
Third, if [68, Theorem 1.5.1] is employed in lieu of [68, Theorem 1.4.1] in the proof
of Theorem 7.4.1, it is possible to relax the doubling assumption on σ := H 2n−1 ∂Ω
to merely demanding that this is a locally finite measure. In such a scenario, we need
to impose the condition that the aperture parameter κ is sufficiently large (depending
on Ω), and the flexibility of changing κ when considering nontangential boundary
traces may be lost. Nonetheless, modulo these nuances, the format of the main results
(i.e., formulas (7.4.3) and (7.4.5)) remains the same.
Fourth, in the case when n = 1 and α = β = 0, Theorem 7.4.1 becomes [70,
Theorem 1.1.1]. In particular, (7.4.3) reduces in this case precisely to the Cauchy-
Pompeiu formula [70, (1.1.8)].
After this preamble, we now turn to the proof of Theorem 7.4.1.
Proof of Theorem 7.4.1 We debut by observing that, in concert, (7.4.2) and [68,
Lemma 8.3.1] imply that
∞
u ∈ Lloc (Ω, L 2n ) ⊗ Λα,β . (7.4.8)

Moreover, (7.4.1), (7.4.2), and [68, Corollary 8.9.9] ensure that for any κ  > 0 the
κ  −n.t.
nontangential trace u∂Ω exists at σ-a.e. point on ∂nta Ω and is actually independent
of the parameter κ  > 0. In addition, with the dependence on κ  dropped, from [68,
(8.9.8)], [68, (8.9.44)], and (7.4.2) we conclude that

∫   n.t.  
 u (ζ)
∂Ω C
dσ(ζ) < +∞. (7.4.9)
∂∗ Ω 1 + |ζ | 2n−1

In particular, from (7.4.9) and (7.3.18)-(7.3.19) it follows that, for each point z ∈ Ω,
the boundary integrals in the first two lines of (7.4.3) are absolutely convergent.
The strategy for actually proving formula (7.4.3) is to apply [68, Theorem 1.2.1] to
a suitably constructed vector field. Specifically, fix a Lebesgue point z ∈ Ωfor (all the
n
coefficients of) u with the property that (7.4.7) holds, and define F : Ω → Λα,β Cn
by requiring that

 1  0,1  1 
ξ · F(ζ) =− ξ ∧ u(ζ), ∂¯ζ Γα,β (ζ, z) + ξ 1,0 ∨ u(ζ), ϑζ Γα,β (ζ, z)
2 C 2 C

for all real vectors ξ ∈ R2n, at L 2n -a.e. point ζ ∈ Ω.

(7.4.10)

Above, as in (7.1.58), for each real vector ξ = (ξ1, ξ2, . . . , ξ2n−1, ξ2n ) ∈ R2n we have
denoted
830 7 Applications to Analysis in Several Complex Variables
n n
ξ 1,0 := (ξC ) j dz j ∈ Λ1,0 Cn and ξ 0,1 := (ξC ) j dz j ∈ Λ0,1 Cn, (7.4.11)
j=1 j=1

where ξC := (ξ1 + iξ2, . . . , ξ2n−1 + iξ2n ) ∈ Cn is the complex version of ξ ∈ R2n .

With this piece of notation, it is clear that the right-hand side of (7.4.10) depends
linearly in the variable ξ ∈ R2n , hence the demand in (7.4.10) determines F uniquely
(as a function defined L 2n -a.e. in Ω) and unambiguously. Moreover, from (7.4.8),
(7.4.10), and (7.3.18)-(7.3.19) we have
 1 n
F ∈ Lloc (Ω, L 2n ) ⊗ Λα,β . (7.4.12)

Granted this, it is meaningful to compute the distributional divergence of F in Ω.

To identify this explicitly, fix an arbitrary real-valued test function ϕ ∈ 𝒞∞
c (Ω) and
observe that
 
(∇ϕ)C = ∂x1 ϕ + i∂y1 ϕ, . . . , ∂xn ϕ + i∂yn ϕ
 
= 2 ∂z̄1 ϕ, . . . , ∂z̄n ϕ . (7.4.13)

As a consequence of (7.4.13), (7.4.11), and (7.2.2) we therefore have

(∇ϕ)1,0 = 2 ∂ϕ and (∇ϕ)0,1 = 2 ∂ϕ.

¯ (7.4.14)

We may then write

 

D  (Ω)⊗Λα, β div F, ϕ D(Ω)
∫
 
 ∇ϕ D(Ω) = −
= − D  (Ω)⊗Λα, β F, ∇ϕ · F dL 2n
Ω
∫  
1   0,1
= ∇ϕ(ζ) ∧ u(ζ), ∂¯ζ Γα,β (ζ, z)
2 C
Ω
  1,0 
− ∇ϕ(ζ) ∨ u(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ)
C
∫  
= (∂ϕ)(ζ)
¯ ∧ u(ζ), ∂¯ζ Γα,β (ζ, z)
C
Ω
 
− (∂ϕ)(ζ) ∨ u(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ), (7.4.15)
C

thanks to (7.4.10) (presently used with ξ := ∇ϕ viewed as a real vector in R2n ),

(7.4.11), (7.1.4), and (7.4.14). Since the Leibniz product rule for ∂,
¯ ϑ currently gives
that, in the sense of distributions in Ω,

(∂ϕ)
¯ ∧ u = ∂(ϕu)
¯ − ϕ ∂u,
¯ (∂ϕ) ∨ u = −ϑ(ϕu) + ϕϑu, (7.4.16)
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 831

∞ (Ω) ⊗ Λα,β
it follows from (7.4.8), (7.4.16), and the first line in (7.4.1) that ϕu ∈ Lcomp
satisfies ∂(ϕu) ∈ Lcomp (Ω, L ) ⊗ Λ
¯ 1 2n α,β+1 and ϑ(ϕu) ∈ Lcomp (Ω, L 2n ) ⊗ Λα,β−1 .
1

Furthermore, in light of (7.4.7) we also have

∫
| ∂(ϕu)(ζ)|
¯
C + |ϑ(ϕu)(ζ)|C
dL 2n (ζ) < +∞. (7.4.17)
Ω |ζ − z| 2n−1
As such, we may further express
 
D  (Ω)⊗Λα, β
 ϕ
divF, = I + I I, (7.4.18)
D(Ω)

where
∫    
I := − ϕ(ζ) (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) + (ϑu)(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ),
C C
Ω
(7.4.19)

and
∫    
I I := ∂(ϕu)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) + ϑ(ϕu)(ζ), ϑζ Γα,β (ζ, z) dL 2n (ζ).
C C
Ω
(7.4.20)

To handle term I I, bring in a scalar-valued function θ ∈ 𝒞∞ (R2n ) with

 the property
that θ = 0 on B(0, 1) and θ = 1 on R2n \ B(0, 2). For each ε ∈ 0, 12 dist(z, ∂Ω)
define θ ε : R2n ≡ Cn → R by setting
ζ − z
θ ε (ζ) := θ for every ζ ∈ Cn ≡ R2n . (7.4.21)
ε
Then
θ ε ∈ 𝒞∞ (Cn ) is a bounded function uniformly in ε,
(7.4.22)
satisfying lim+ θ ε (ζ) = 1 for every fixed ζ ∈ Cn \ {z},
ε→0
 
and there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(z, ∂Ω) we
have
θ ε − 1 ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(z, ε),

supp (∇θ ε ) ⊆ B(z, 2ε) \ B(z, ε), and (7.4.23)

|(∇θ ε )(ζ)| ≤ Cε −1 for every ζ ∈ Cn .

In particular, both θ ε (ζ)∂¯ζ Γα,β (ζ, z) and θ ε (ζ)ϑζ Γα,β (ζ, z) are 𝒞∞ in the variable
ζ ∈ Ω. Based on Lebesgue’s Dominated Convergence Theorem (whose applicability
is presently ensures by (7.4.22) and (7.4.17)) may therefore write
832 7 Applications to Analysis in Several Complex Variables
∫  
I I = lim+ ∂(ϕu)(ζ),
¯ θ ε (ζ)∂¯ζ Γα,β (ζ, z)
ε→0 C
Ω
 
+ ϑ(ϕu)(ζ), θ ε (ζ)ϑζ Γα,β (ζ, z) dL 2n (ζ)
C
∫ 
 
= lim+ (ϕu)(ζ), ϑζ θ ε (ζ)∂¯ζ Γα,β (ζ, z)
ε→0 C
Ω
  
+ (ϕu)(ζ), ∂¯ζ θ ε (ζ)ϑζ Γα,β (ζ, z) dL 2n (ζ). (7.4.24)
C

Note that, on account of Leibniz’ product rule for ∂, ¯ ϑ, (7.3.14), and (7.3.20), we
may compute in the sense of distributions
   
ϑζ θ ε (ζ)∂¯ζ Γα,β (ζ, z) + ∂¯ζ θ ε (ζ)ϑζ Γα,β (ζ, z)
 
= θ ε (ζ) ϑζ ∂¯ζ + ∂¯ζ ϑζ Γα,β (ζ, z)
¯ ε )(ζ) ∧ ϑζ Γα,β (ζ, z) − (∂θ ε )(ζ) ∨ ϑζ Γα,β (ζ, z)
+ (∂θ

= (∂θ
¯ ε )(ζ) ∧ ϑζ Γα,β (ζ, z) − (∂θ ε )(ζ) ∨ ϑζ Γα,β (ζ, z). (7.4.25)

In turn, this suggests that we split

I I = I I I + IV (7.4.26)

where
∫ 
I I I := lim+ (ϕu)(ζ) − (ϕu)(z), (∂θ
¯ ε )(ζ) ∧ ϑζ Γα,β (ζ, z)
ε→0
Ω

− (∂θ ε )(ζ) ∨ ϑζ Γα,β (ζ, z) dL 2n (ζ), (7.4.27)
C

and
∫ 
IV := lim+ (ϕu)(z), (∂θ
¯ ε )(ζ) ∧ ϑζ Γα,β (ζ, z)
ε→0
Ω

− (∂θ ε )(ζ) ∨ ϑζ Γα,β (ζ, z) dL 2n (ζ). (7.4.28)
C

Since z is a Lebesgue point for u, we may estimate (based on (7.4.23) and (7.3.18)-
(7.3.19))
⨏
 
|I I I | ≤ C lim sup (ϕu)(ζ) − (ϕu)(z) dL 2n (ζ) = 0, (7.4.29)
C
ε→0+ B(z,2ε)
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 833

thus
I I I = 0. (7.4.30)
Also, since ηε := θ ε − 1 belongs to 𝒞∞c (Ω) and since (ϕu)(z) does not depend on ζ,
we may reverse-engineer the first equality in (7.4.25) in order to write
∫ 
IV = lim+ (ϕu)(z), (∂η
¯ ε )(ζ) ∧ ϑζ Γα,β (ζ, z)
ε→0 Ω

− (∂ηε )(ζ) ∨ ϑζ Γα,β (ζ, z) dL 2n (ζ)
C
 
= lim+ ℰ(Ω)⊗Λα, β (ϕu)(z), (∂η
¯ ε ) ∧ ϑ · Γα,β (·, z) − (∂ηε )∨ ϑ · Γα,β (·, z) ℰ (Ω)⊗Λα, β
ε→0
   
= − lim+ ℰ(Ω)⊗Λα, β (ϕu)(z), ηε ϑ · ∂¯· + ∂¯· ϑ · Γα,β (·, z) ℰ (Ω)⊗Λα, β
ε→0

= − lim+ ηε (z)(ϕu)(z) = (ϕu)(z), (7.4.31)

ε→0

(with all distributional pairings in the “dot" variable), where in the penultimate
equality we have also made use of (7.3.20), and the last equality relies on the fact
that ηε (z) = −1 for each ε ∈ (0, εo ) (cf. (7.4.23)).
Collectively, (7.4.18)-(7.4.20), (7.4.26)-(7.4.28), (7.4.30)-(7.4.31) establish that
   
divF = − ∂u,
¯ ∂¯· Γα,β (·, z) − ϑu, ϑ · Γα,β (·, z)
C C

+ u(z)δz in D (Ω) ⊗ Λα,β, (7.4.32)

where δz is the Dirac distribution in Ω with mass at z. In particular, from (7.4.32)

and (7.4.7) we conclude that

divF ∈ ℰ(Ω) ⊗ Λα,β + L 1 (Ω, L 2n ) ⊗ Λα,β . (7.4.33)

 
Next, K := B z, 12 dist(z, ∂Ω) is a compact subset of Ω, and based on (7.4.10),
(7.3.18)-(7.3.19), and [68, Lemma 8.3.7] we may estimate
     
NκΩ\K F (ζ) ≤ C Nκ u (ζ) · sup |z − ξ | 1−2n
ξ ∈Γκ (ζ )\K
 
Nκ u (ζ)
≤C , for all ζ ∈ ∂Ω, (7.4.34)
|z − ζ | 2n−1

for some constant C = C(Ω, n, κ) ∈ (0, ∞). In turn, from (7.4.34), (7.4.2), and [68,
(8.2.26)] it follows that

NκΩ\K F ∈ L 1 (∂Ω, σ). (7.4.35)

Moreover, from (7.4.10), the second line in (7.4.1), and [68, (8.9.10)-(8.9.11)] we
conclude that
834 7 Applications to Analysis in Several Complex Variables
κ−n.t.

F exists at σ-a.e. point on ∂nta Ω (7.4.36)
∂Ω

and at σ-a.e. point ζ ∈ ∂∗ Ω we have

 κ−n.t.  1  κ−n.t.  

ν(ζ) · F (ζ) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z)
∂Ω 2 C

1    κ−n.t.  
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) . (7.4.37)
2 C

Let us also remark that condition (7.4.4) together with (7.3.18)-(7.3.19) guarantee
that
if the open set Ω ⊆ R2n is an exterior domain, then the
(7.4.38)
vector field F satisfies the growth condition [68, (1.4.8)].

Together, (7.4.12), (7.4.33), and (7.4.36) guarantee that the vector field F satisfies
the hypotheses of [68, Theorem 1.4.1]. On account of [68, (4.6.19)], (7.4.32), and
(7.4.37), the Divergence Formula recorded in [68, (1.4.6)] presently yields (7.4.3).
Finally, formula (7.4.5) is established in a similar (and simpler) fashion, keeping
in mind that if z ∈ Cn \ Ω then Γα,β (·, z) has coefficients in 𝒞∞ (Ω). As such, there
is no need to bring in the cutoff function θ ε and, this time, the Dirac distribution in
(7.4.32) is no longer present. 

Theorem 7.4.1 contains as a particular case the following geometrically sharp

version of the Bochner-Martinelli integral representation formula.

Corollary 7.4.2 Suppose Ω ⊆ R2n ≡ Cn is an open set with a lower Ahlfors regular
boundary, and with the property that σ := H 2n−1 ∂Ω is a doubling measure on ∂Ω.
In particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Fix some aperture parameter κ > 0.
In this context, suppose F ∈ Lloc 1 (Ω, L 2n ) is a complex–valued function satisfying

(with all partial derivatives considered in the sense of distributions in Ω)

 L 2n (ζ) 
∂z̄ j F ∈ L 1 Ω, for each j ∈ {1, . . . , n},
1 + |ζ | 2n−1 (7.4.39)
κ−n.t.
the trace F ∂Ω exists at σ-a.e. point on ∂nta Ω,

and ∫
(Nκ F)(ζ)
dσ(ζ) < +∞. (7.4.40)
∂Ω + |ζ |
1 2n−1

κ  −n.t.
Then for any κ  > 0 the nontangential trace F ∂Ω also exists σ-a.e. on ∂nta Ω
and is in fact independent of κ . Furthermore, with the dependence on the parameter
κ  dropped, for L 2n -a.e. point z ∈ Ω one has (with absolutely convergent integrals)
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 835
∫
 n.t.  2
n
ζ¯j − z̄ j
F(z) = B0,0 F ∂Ω (z) − (∂z̄ F)(ζ) dL 2n (ζ) (7.4.41)
ω2n−1 j=1
|ζ − z| 2n j
Ω

if either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω is

an exterior domain, the same conclusion holds true under the additional assumption
that there exists λ ∈ (1, ∞) such that
⨏
|F | dL 2n = o(1) as R → ∞. (7.4.42)
B(0,λ R)\B(0,R)

Furthermore, for each z ∈ Cn \ Ω holds one has

∫
 n.t.  2
n
ζ¯j − z̄ j
0 = B0,0 F ∂Ω (z) − (∂z̄ F)(ζ) dL 2n (ζ) (7.4.43)
ω2n−1 j=1
|ζ − z| 2n j
Ω

with the same provision as in (7.4.42) when Ω is an exterior domain.

As a consequence, if F is actually a holomorphic function in Ω with the property
that κ−n.t.
F ∂Ω exists at σ-a.e. point on ∂nta Ω,
∫ (7.4.44)
(Nκ F)(ζ)
and dσ(ζ) < +∞,
∂Ω 1 + |ζ |
2n−1

κ  −n.t.
then for any other κ  > 0 the nontangential trace F ∂Ω also exists σ-a.e. on ∂nta Ω,
is in fact independent of κ  and, with the dependence on the parameter κ  dropped,
one has

 n.t.  F(z) if z ∈ Ω,

B0,0 F ∂Ω (z) = for each z ∈ Cn \ ∂Ω, (7.4.45)
0 if z ∈ Cn \ Ω,

provided one also assumes the decay condition (7.4.42) in the case when Ω is an
exterior domain.
Proof All claims follow by specializing Theorem 7.4.1 to the case when α = β = 0.
Theorem 7.4.3 stated below is a sharp rendition of the Bochner-Martinelli-
Koppelman formula. Our version generalizes [88, Theorem 1.10, p. 154] and [59,
Theorem 4.11, p. 23] which assume that underlying set Ω is a bounded 𝒞1 domain
and the differential form u is of class 𝒞1 (Ω), and [36, Theorem 1.11.1, p. 57] where
it is assumed that Ω is a bounded domain with piecewise 𝒞2 boundary, and that both
u and ∂u ¯ are continuous on Ω.
Theorem 7.4.3 Suppose Ω ⊆ R2n ≡ Cn is an open set with a lower Ahlfors regular
boundary, and with the property that σ := H 2n−1 ∂Ω is a doubling measure on
∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Also, fix a pair of degrees
α, β ∈ {0, 1, . . . , n} along with some aperture parameter κ > 0.
836 7 Applications to Analysis in Several Complex Variables

In this context, assume

 L 2n (ζ) 
u ∈ L 1 Ω, ⊗ Λα,β (7.4.46)
1 + |ζ | 2n−1
is a differential form satisfying (with all partial differential operators considered in
the sense of distributions in Ω)
 L 2n (ζ) 
∂u
¯ ∈ L 1 Ω, ⊗ Λα,β+1, (7.4.47)
1 + |ζ | 2n−1
κ−n.t.
u∂Ω exists at σ-a.e. point on ∂nta Ω, (7.4.48)
∫
(Nκ u)(ζ)
dσ(ζ) < +∞. (7.4.49)
∂Ω 1 + |ζ |
2n−1

Then the differential form

∫  
U(z) := u(ζ), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) for L 2n -a.e. z ∈ Ω,
Ω C (7.4.50)
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λα,β .
is locally integrable in Ω and satisfies ∂U loc

κ  −n.t.
Also, for any κ  > 0 the nontangential trace u∂Ω exists σ-a.e. on ∂nta Ω and is
actually independent of κ . Moreover, with the dependence on the parameter κ 
dropped, for L 2n -a.e. point z ∈ Ω one has
∫  
1  n.t. 
u(z) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
+ (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ)
Ω C

∫  
+ ∂¯z u(ζ), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) , (7.4.51)
Ω C

where the action of ∂¯z is considered in the sense of distributions in Ω (cf. (7.4.50)).

Proof We debut by observing that, in concert, (7.4.49) and [68, Lemma 8.3.1] imply
that
∞
u ∈ Lloc (Ω, L 2n ) ⊗ Λα,β . (7.4.52)

Moreover, (7.4.48), (7.4.49), and [68, Corollary 8.9.9] ensure that for any κ  > 0 the
κ  −n.t.
nontangential trace u∂Ω exists at σ-a.e. point on ∂nta Ω and is actually independent
of the parameter κ  > 0. In addition, with the dependence on κ  dropped, from [68,
(8.9.8), (8.9.44)] and (7.4.49) we conclude that
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 837

∫   n.t.  
 u (ζ)
∂Ω C
dσ(ζ) < +∞. (7.4.53)
∂∗ Ω 1 + |ζ | 2n−1

In particular, from (7.4.53) and (7.3.18)-(7.3.19) it follows that, for each fixed point
z ∈ Ω, the boundary integral in the first line of (7.4.51) is absolutely convergent.
Finally, we observe from the proof of [68, Lemma 3.5.7] (cf. [68, (3.5.32)]) that,
collectively, the integrability conditions in (7.4.46) and the first line of (7.4.47) imply
that ∫
|u(ζ)|C + |(∂u)(ζ)|
¯
C
the expression dL 2n (ζ)
Ω |ζ − z| 2n−1
(7.4.54)
1 (Cn, L 2n ) as a function of z.
belongs to Lloc
The strategy for actually proving formula (7.4.51) is to apply [68, Theorem 1.2.1]
to a suitable domain and vector field. Specifically, define

D := Ω × Cn (7.4.55)

which is an open set in R4n ≡ C2n . Next,

fix w ∈ 𝒞∞
c (Ω) ⊗ Λ
β,α
and define Fw : D −→ C2n (7.4.56)

by requiring (using notation introduced in (7.3.4) and (7.3.9)) that

1   0,1  
(ξ, η) · Fw (ζ, z) = − ξ ∧ u(ζ) ⊗ w(z), ∂¯ζ Γα,β (ζ, z)
2 C

1    
+ u(ζ) ⊗ η0,1 ∨ w(z) , ∂¯ζ Γα,β−1 (ζ, z) (7.4.57)
2 C

for all vectors (ξ, η) ∈ R2n × R2n, at L 2n ⊗ L 2n -a.e. point (ζ, z) ∈ D.

Indeed, since the rems in the right-hand side of (7.4.57) depend linearly in the
variables ξ, η ∈ R2n , the demand in (7.4.57) determines Fw unambiguously as an
L 2n ⊗ L 2n -a.e. defined function in D which is C2n -valued. This also shows that F is
L 2n ⊗ L 2n -measurable. Observe next that (7.4.57) and (7.3.18)-(7.3.19) imply the
existence of a purely dimensional constant C ∈ (0, ∞) such that
 
Fw (ζ, z) ≤ C |u(ζ)|C |w(z)|C for all (ζ, z) ∈ D with ζ  z. (7.4.58)
|ζ − z| 2n−1
In concert with (7.4.54) and (7.4.55)-(7.4.56) this allows us to conclude that
  4n
Fw ∈ L 1 (D, L 2n ⊗ L 2n ) . (7.4.59)

Granted this, it is meaningful to compute the distributional divergence of Fw in D.

To identify this explicitly, fix two arbitrary real-valued test functions ϕ ∈ 𝒞∞
c (Ω) and
ψ ∈ 𝒞∞c (C ). Upon recalling from (7.4.14) that (∇ϕ)
n 0,1 = 2 ∂ϕ
¯ and (∇ψ)0,1 = 2 ∂ψ,
¯
838 7 Applications to Analysis in Several Complex Variables

we may then write

 

D  (D) div Fw , ϕ ⊗ ψ D(D)
  
= − D  (D) Fw, (∇ϕ) ⊗ ψ, ϕ ⊗ (∇ψ) D(D)
∫
 
=− (∇ϕ) ⊗ ψ, ϕ ⊗ (∇ψ) · Fw dL 2n ⊗ dL 2n
D
∫  
1   0,1 
= ∇ϕ(ζ) ∧ u(ζ) ⊗ (ψw)(z), ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
2 D C
∫  
1   0,1 
− (ϕu)(ζ) ⊗ ∇ψ(ζ) ∨ w(z) , ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z)
2 D C
∫  
 
= (∂ϕ)(ζ)
¯ ∧ u(ζ) ⊗ (ψw)(z), ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
D C
∫  
 
− (ϕu)(ζ) ⊗ (∂ψ)(z)
¯ ∨ w(z) , ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z)
D C

=: I + II, (7.4.60)

thanks to (7.4.57) presently used with ξ := (∇ϕ) ⊗ ψ and η := ϕ ⊗ (∇ψ) viewed as

reals vector in R2n . Making use of Leibniz’ product rule for ∂¯ (cf. (7.4.16)) we may
further split

I = I a + Ib, (7.4.61)

where
∫  
 
Ia := − ϕ(ζ)(∂u)(ζ)
¯ ⊗ (ψw)(z), ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
D C
∫  
=− (∂u)(ζ)
¯ ⊗ w(z), ∂¯ζ Γα,β (ζ, z) (ϕ ⊗ ψ)(ζ, z) dL 2n (ζ) dL 2n (z)
D C
(7.4.62)

and
∫  
 
Ib := ∂(ϕu)(ζ)
¯ ⊗ (ψw)(z), ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z). (7.4.63)
D C

To handle term Ib , bring in a scalar-valued function θ ∈ 𝒞∞ (R2n ) with the property

that θ = 0 on B(0, 1) and θ = 1 on R2n \ B(0, 2). For each ε > 0 define

θ ε : R2n × R2n ≡ Cn × Cn −→ R (7.4.64)

by setting
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 839
ζ − z
θ ε (ζ, z) := θ for every ζ, z ∈ Cn ≡ R2n . (7.4.65)
ε
Then
θ ε ∈ 𝒞∞ (Cn × Cn ) is a bounded function uniformly in ε,
(7.4.66)
satisfying lim+ θ ε (ζ, z) = 1 for each fixed ζ, z ∈ Cn with ζ  z,
ε→0

and there exists a constant C ∈ (0, ∞) such that for each ε > 0 we have

θ ε (ζ, z) = 0 if |ζ − z| < ε, θ ε (ζ, z) = 1 if |ζ − z| ≥ 2ε,

supp (∇θ ε ) ⊆ (ζ, z) ∈ Cn × Cn : ε ≤ |ζ − z| ≤ 2ε , (7.4.67)
and |(∇θ ε )(ζ, z)| ≤ Cε −1 for all (ζ, z) ∈ Cn × Cn .

Based on Lebesgue’s Dominated Convergence Theorem (whose applicability in the

present setting is ensured by (7.4.66) and (7.4.54)) and (7.3.13) we may then write
∫  
 
Ib = lim+ ∂(ϕu)(ζ)
¯ ⊗ (ψw)(z), θ ε (ζ, z)∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
ε→0 C
D
   
= lim+ ℰ (D) ∂(ϕu)(ζ)
¯ ⊗ (ψw)(z), θ ε (ζ, z)∂¯ζ Γα,β (ζ, z) ℰ(D)
ε→0
   
= lim+ ℰ (D) (ϕu)(ζ) ⊗ (ψw)(z), ϑζ θ ε (ζ, z)∂¯ζ Γα,β (ζ, z) ℰ(D) .
ε→0
∫ 
  
= lim+ (ϕu)(ζ) ⊗ (ψw)(z), ϑζ θ ε (ζ, z)∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z).
ε→0 C
D
(7.4.68)

On the other hand, on account of Leibniz’ product rule for ∂,

¯ ϑ, (7.3.14), and (7.3.20),
we may compute
840 7 Applications to Analysis in Several Complex Variables
 
ϑζ θ ε (ζ, z)∂¯ζ Γα,β (ζ, z) (7.4.69)

= −∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) + θ ε (ζ, z)ϑζ ∂¯ζ Γα,β (ζ, z)
 
= −∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) + θ ε (ζ, z) ϑζ ∂¯ζ + ∂¯ζ ϑζ Γα,β (ζ, z)

− θ ε (ζ, z)∂¯ζ ϑζ Γα,β (ζ, z)

= −∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) + θ ε (ζ, z)ζ Γα,β (ζ, z)

− θ ε (ζ, z)∂¯ζ ∂z Γα,β−1 (ζ, z)

= −∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z)

 
− ∂z θ ε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) + ∂z θ ε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z).

In turn, this suggests that we split

Ib = III − IV + V. (7.4.70)

Above, III := − lim+ IIIε where, for each ε > 0,

ε→0
∫  
IIIε := (ϕu)(ζ) ⊗ (ψw)(z), ∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z).
C
D
(7.4.71)

Also, IV is given by
∫ 
  
lim+ (ϕu)(ζ) ⊗ (ψw)(z), ∂z θ ε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ)dL 2n (z)
ε→0 C
D
∫  
= lim+ (ϕu)(ζ) ⊗ ϑ̄(ψw)(z), θ ε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ)dL 2n (z)
ε→0 C
D
∫  
= (ϕu)(ζ) ⊗ ϑ̄(ψw)(z), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ)dL 2n (z), (7.4.72)
C
D

with the last step justified by Lebesgue’s Dominated Convergence Theorem (which,
in turn, relies on (7.4.66) and (7.4.54)). Finally, V := lim+ Vε where, for each ε > 0,
ε→0
∫  
Vε := (ϕu)(ζ) ⊗ (ψw)(z), ∂z θ ε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ)dL 2n (z).
C
D
(7.4.73)

Observe that since (∂ψ)

¯ ∨ w + ϑ̄(ψw) = ψ ϑ̄w we may combine
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 841
∫  
II − IV = − (ϕu)(ζ) ⊗ (ψ ϑ̄w)(z), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z)
C
D
∫  
=− u(ζ) ⊗ (ϑ̄w)(z), ∂¯ζ Γα,β−1 (ζ, z) (ϕ ⊗ ψ)(ζ, z)dL 2n (ζ) dL 2n (z).
C
D
(7.4.74)

Pressing on, break up

V = V a + Vb (7.4.75)
where Va is the version of V in which we freeze the coefficients of the differential
form ϕu at the point z (while retaining the dζ’s and dζ’s ¯ originally present in the
writing of u unaffected), i.e.,
∫ 
Va := lim+ (ϕu)(z) ⊗ (ψw)(z) ,
ε→0 D

∂z θ ε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z), (7.4.76)
C

and Vb = V − Va is the remainder given by

∫ 
 
Vb := lim+ (ϕu)(ζ) − (ϕu)(z) ⊗ (ψw)(z) ,
ε→0 D

∂z θ ε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z). (7.4.77)
C

 
Note that if we set εw := 12 dist supp w, ∂Ω > 0 and for each point z ∈ supp w and
each ε ∈ (0, εw ) we define
⨏
 
fε (z) := (ϕu)(ζ) − (ϕu)(z) dL 2n (ζ) (7.4.78)
C
B(z,2ε)

then by invoking Lebesgue’s Dominated Convergence Theorem (whose applicability

is ensured by Lebesgue’s Differentiation Theorem and (7.4.52)) we conclude from
(7.4.77), (7.4.67), and (7.3.18)-(7.3.19) that
∫
|Vb | ≤ C lim+ fε (z) dL 2n (z) = 0, hence Vb = 0. (7.4.79)
ε→0 supp w

Similarly, break up
III = IIIa + IIIb (7.4.80)
where IIIa is the version of III in which we freeze the coefficients of the differential
form ϕu at the point z, i.e.,
842 7 Applications to Analysis in Several Complex Variables
∫ 
IIIa := − lim+ (ϕu)(z) ⊗ (ψw)(z) ,
ε→0 D

∂ζ θ ε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z), (7.4.81)
C

then estimate the disagreement term IIIb := III − IIIa as in (7.4.79) to conclude that
∫
|IIIb | ≤ C lim+ fε (z) dL 2n (z) = 0, hence IIIb = 0. (7.4.82)
ε→0 supp w

At this stage, abbreviating

ηε := θ ε − 1 and Θ := (ϕu)(z) ⊗ (ψw)(z), (7.4.83)

from (7.4.75)-(7.4.77), (7.4.79), and (7.4.80)-(7.4.82) we see that we may express

III + V as

III + V = IIIa + Va (7.4.84)

∫ 
= lim+ Θ, ∂z ηε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z)
ε→0 D

− ∂ζ ηε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z).
C

To proceed, the idea is to reverse-engineer (7.4.69) (with the role of θ ε now played
by ηε ) and compute, in the sense of distributions:

∂z ηε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z) − ∂ζ ηε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z)

 
= ∂z ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) − ηε (ζ, z)∂¯ζ ∂z Γα,β−1 (ζ, z)
 
+ ϑζ ηε (ζ, z)∂¯ζ Γα,β (ζ, z) − ηε (ζ, z)ϑζ ∂¯ζ Γα,β (ζ, z)
   
= ∂z ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) + ϑζ ηε (ζ, z)∂¯ζ Γα,β (ζ, z)

− ηε (ζ, z)(∂¯ζ ϑζ + ϑζ ∂¯ζ )Γα,β (ζ, z). (7.4.85)

Let us also observe that, in the sense of distributions in D,

(∂¯ζ ϑζ + ϑζ ∂¯ζ )Γα,β (ζ, z) = ζ Γα,β (ζ, z) (7.4.86)

= 2−α−β (−1)αβ δ(ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J )

|J |=α |I |=β

where δ(ζ, z) ∈ D (D) is the diagonal Dirac distribution defined as

∫
 
D (D) δ(ζ, z), Φ(ζ, z) D(D) :=
 Φ(z, z) dL 2n (z) (7.4.87)
Ω
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 843

for each test function Φ ∈ 𝒞∞

c (D). In particular, since ηε (z, z) = −1 for each z ∈ Ω,
it follows that

−ηε (ζ, z)(∂¯ζ ϑζ + ϑζ ∂¯ζ )Γα,β (ζ, z) (7.4.88)

= 2−α−β (−1)αβ δ(ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J ).

|J |=α |I |=β

Since, by assumption,

u ∈ Lloc
1
(Ω, L 2n ) ⊗ Λα,β and ∂u
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λα,β+1,
loc (7.4.89)

using a standard mollifier argument yields a sequence

{u j } j ∈N ⊂ 𝒞∞ (Ω, L 2n ) ⊗ Λα,β with u j → u in Lloc

1 (Ω, L 2n ) ⊗ Λα,β,

∂u ¯ in L 1 (Ω, L 2n ) ⊗ Λα,β+1 as j → ∞, and

¯ j → ∂u (7.4.90)
loc
u j (z) → u(z) as j → ∞ at each Lebesgue point z of u in Ω.

Moreover, thanks to (7.4.52), we may also assume that

for each compact K ⊂ Ω there exists jK ∈ N

(7.4.91)
with the property that sup supK |u j |C < +∞.
j ≥ jK

Associated with each u j then define the double form

Θ j := (ϕu j )(z) ⊗ (ψw)(z), for all z ∈ Ω. (7.4.92)

Returning to (7.4.84) then write

∫ 
III + V = lim+ lim Θ j , ∂z ηε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z)
ε→0 j→∞ D

− ∂ζ ηε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
C

= lim+ lim D(D) Θ j , ∂z ηε (ζ, z) ∧ ∂¯ζ Γα,β−1 (ζ, z)
ε→0 j→∞

− ∂ζ ηε (ζ, z) ∨ ∂¯ζ Γα,β (ζ, z) D  (D)

= lim+ lim VIε, j + VIIε, j + VIII j , (7.4.93)

ε→0 j→∞

where, for each ε > 0 and j ∈ N, we have set

   
VIε, j := D(D) Θ j , ∂z ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) D  (D) (7.4.94)
844 7 Applications to Analysis in Several Complex Variables
   
VIIε, j := D(D) Θ j , ϑζ ηε (ζ, z)∂¯ζ Γα,β (ζ, z) D  (D) (7.4.95)

and
 
VIII j := D(D) Θ j , 2−α−β (−1)αβ δ(ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (dz I ∧ d z̄ J ) D  (D)
|J |=α
|I |=β
∫
 
= ϕu j , ψ w̄ dL 2n . (7.4.96)
C
Ω

Let us now consider the linear functional

∫
Λ : 𝒞c (D) → C,
0
ΛΦ := Φ(z, z) dL 2n (z), ∀Φ ∈ 𝒞0c (D). (7.4.97)
Ω

Since this obviously has the property that for each compact set K ⊆ D

sup |ΛΦ| : Φ ∈ 𝒞0c (D), |Φ| ≤ 1 on D, and supp Φ ⊆ K < +∞, (7.4.98)

from Riesz’ Representation Theorem (cf. [68, Proposition 3.9.1]) we conclude that
there exists Borel measure on D, call it μdiag , which is locally finite and Borel-regular,
with the property that
∫
ΛΦ = Φ(ζ, z) dμdiag (ζ, z) for every Φ ∈ 𝒞0c (D). (7.4.99)
D

In particular, if we change u on a L 2n -nullset to make it Borel measurable (which,

as is well known, is always possible – see, e.g., [92, Exercise 5, p. 56] – and which
does not affect any of the original properties of u) we may then write
∫
 
lim VIII j = lim ϕu j , ψw dL 2n
j→∞ j→∞ C
Ω
∫
 
= lim u j (z), w̄(z) (ϕ ⊗ ψ)(z, z) dL 2n (z)
j→∞ C
Ω
∫
 
= lim u j (ζ), w̄(z) (ϕ ⊗ ψ)(ζ, z) dμdiag (ζ, z)
j→∞ C
D
∫
 
= u(ζ), w̄(z) (ϕ ⊗ ψ)(ζ, z) dμdiag (ζ, z), (7.4.100)
C
D

by relying on (7.4.97), (7.4.99), (7.4.90)-(7.4.91), and Lebesgue’s Dominated Con-

vergence Theorem. Note that for each ε > 0 and j ∈ N we have
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 845
 
VIε, j = D(D) ϑ̄z Θ j , ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) D  (D)

   
= D(D) (ϕu j )(z) ⊗ ϑ̄(ψw) (z), ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) D  (D)
∫  
 
= (ϕu j )(z) ⊗ ϑ̄(ψw) (z), ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z)
C
D
(7.4.101)

hence, as a simple application of Lebesgue’s Dominated Convergence Theorem

shows,
∫ 
 
lim VIε, j = (ϕu)(z) ⊗ ϑ̄(ψw) (z) ,
j→∞ D 
ηε (ζ, z)∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z) (7.4.102)
C

and, ultimately,

lim lim VIε, j = 0, (7.4.103)

ε→0+ j→∞

once again by Lebesgue’s Dominated Convergence Theorem. Also, since the coeffi-
cients of the double form Θ j are independent of ζ, we have
 
VIIε, j = D(D) ∂¯ζ Θ j , ηε (ζ, z)∂¯ζ Γα,β (ζ, z) D  (D) = 0 (7.4.104)

for all ε > 0 and all j ∈ N. Together, (7.4.60), (7.4.61), (7.4.70), (7.4.74), (7.4.75),
(7.4.79), (7.4.80), (7.4.82), (7.4.93), (7.4.100), (7.4.103), and (7.4.104) prove that
 

D  (D) div Fw , ϕ ⊗ ψ D(D)
∫  
=− (∂u)(ζ)
¯ ⊗ w(z), ∂¯ζ Γα,β (ζ, z) (ϕ ⊗ ψ)(ζ, z) dL 2n (ζ) dL 2n (z)
C
D
∫  
− u(ζ) ⊗ (ϑ̄w)(z), ∂¯ζ Γα,β−1 (ζ, z) (ϕ ⊗ ψ)(ζ, z) dL 2n (ζ) dL 2n (z)
C
D
∫
 
+ u(ζ), w̄(z) (ϕ ⊗ ψ)(ζ, z) dμdiag (ζ, z). (7.4.105)
C
D

In view of the arbitrariness of ϕ ∈ 𝒞∞ ∞ n

c (Ω) and ψ ∈ 𝒞c (C ), if we now define
 
Ψw (ζ, z) := u(ζ), w̄(z) for (ζ, z) ∈ D, (7.4.106)
C

this further implies that

846 7 Applications to Analysis in Several Complex Variables
 
divFw = − (∂u)
¯ ⊗ w, ∂¯ζ Γα,β
C
 
− u ⊗ (ϑ̄w), ∂¯ζ Γα,β−1 + Ψw μdiag in D (D). (7.4.107)
C

In particular, since w is smooth and compactly supported, u is locally essentially

bounded, and μdiag is a locally finite Borel measure, it follows that

Ψw μdiag ∈ CBM(D), (7.4.108)

from (7.4.107), (7.4.54), and (7.4.108) we conclude that

divFw ∈ L 1 (D, L 2n ⊗ L 2n ) + CBM(D). (7.4.109)

Going forward, straight from definitions we see that

D ⊆ C2n is an open set, with ∂D = ∂Ω × Cn, and ∂∗ D = ∂∗ Ω × Cn . (7.4.110)

In particular, D has locally finite perimeter, and if we consider σD := H 4n−1 ∂D

and let νD denote the geometric measure theoretic outward unit normal to D, then
 
σD = σ ⊗ L 2n and νD (ζ, z) = ν(ζ), 0
(7.4.111)
at σD -a.e. point (ζ, z) ∈ ∂∗ D = ∂∗ Ω × Cn .

In addition, for each κ > 0 we have

ΓΩ,κ (ζo ) × {zo } ⊆ ΓD,κ (ζo, zo ) ⊆ ΓΩ,κ (ζo ) × Cn

(7.4.112)
for all (ζo, zo ) ∈ ∂D = ∂Ω × Cn,

which, in particular, proves that

∂nta D = ∂nta Ω × Cn . (7.4.113)

Next, if we now fix some compact neighborhood K of supp w in the open set

Ω, based on (7.4.10), (7.4.112), and [68, Lemma 8.3.7] we may find some constant
C = C(Ω, K, w, κ, n) ∈ (0, ∞) such that for all (ζo, zo ) ∈ ∂D = ∂Ω × Cn we have
!
 D\(K×K)     
Nκ Fw (ζo, zo ) ≤ C Nκ u (ζo ) · sup sup |ζ − z| 1−2n
z ∈supp w ζ ∈Γκ (ζo )\K
 
Nκ u (ζo )
≤C . (7.4.114)
1 + |ζo | 2n−1
In turn, from (7.4.114), (7.4.49), and [68, (8.1.18), (8.2.26)] it follows that

NκD\(K×K) Fw ∈ Lloc

1
(∂D, σD ). (7.4.115)
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 847

Moreover, from (7.4.57), (7.4.47), and [68, (8.9.10)-(8.9.11)] we conclude that

κ−n.t.

Fw  exists at σD -a.e. point on ∂nta D, (7.4.116)
∂D

and, by also taking (7.4.111), (7.4.110), and [68, (8.8.52)] into account, that
 κ−n.t.     κ−n.t. 

νD (ζ, z) · Fw  (ζ, z) = ν(ζ), 0 · Fw  (ζ, z)
∂D ∂D

1    κ−n.t.   
=− ν 0,1 (ζ) ∧ u∂Ω (ζ) ⊗ w(z), ∂¯ζ Γα,β (ζ, z)
2 C

at σD -a.e. point (ζ, z) ∈ ∂∗ D = ∂∗ Ω × Cn . (7.4.117)

As a consequence of (7.4.117), there exists a constant C ∈ (0, ∞) with the property

that for σD -a.e. point (ζ, z) ∈ ∂∗ D we have
  κ−n.t.  
  κ−n.t.    u (ζ) |w(z)|C
   ∂Ω
νD (ζ, z) · Fw 
C
(ζ, z) ≤ C . (7.4.118)
∂D |ζ − z| 2n−1
Together with (7.4.53) and [68, (8.9.44)], this goes to show (also keeping in mind
that w is a smooth compactly supported function in Ω) that in fact
 κ−n.t. 

νD · Fw  belongs to the space L 1 (∂∗ D, σD ). (7.4.119)
∂D

Together, (7.4.59), (7.4.109), (7.4.115), (7.4.116), and (7.4.119) guarantee that the
vector field Fw satisfies the hypotheses of [68, Theorem 1.4.1]. Moreover, for each
R > 0 the membership in (7.4.59) allows us to estimate
∫
|(ζ, z) · Fw (ζ, z)| dL 2n (ζ) dL 2n (z) ≤ 2R Fw L 1 (D, L 2n ⊗ L 2n ) (7.4.120)
(ζ,z)∈D
R ≤ |(ζ,z) |<2R

which, in turn, ensures that [68, (1.4.8)] holds. On account of (7.4.107) and (7.4.117),
the Divergence Formula recorded in [68, (1.4.6)] then currently yields
∫
   n.t. 
(𝒞∞ (D)) ∗ div 
Fw , 1 ∞
𝒞 b (D) = νD · Fw ∂D dσD . (7.4.121)
b ∂∗ D

On the other hand, (7.4.107) and [68, (4.6.19)] imply

848 7 Applications to Analysis in Several Complex Variables
∫  
 

(𝒞∞b (D)) divFw, 1 =− (∂u)
∗ 𝒞∞ (D)
¯ ⊗ w, ∂¯ζ Γα,β dL 2n dL 2n
b
D C
∫  
− u ⊗ (ϑ̄w), ∂¯ζ Γα,β−1 dL 2n dL 2n
D C
∫
+ Ψw dμdiag =: A1 + A2 + A3 . (7.4.122)
D

Use (7.3.10) and Fubini’s Theorem to expand (keeping in mind that w is compactly
supported in Ω)
∫    
A1 = − w(z), (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z)
C
Ω×C n C
∫  ∫ 
 
=− w(z), (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) dL 2n (z), (7.4.123)
C
Ω Ω C

as well as
∫    
A2 = − (ϑ̄w)(z), u(ζ), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z)
C
Ω×C n C
∫  ∫ 
 
=− (ϑ̄w)(z), u(ζ), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ) dL 2n (z). (7.4.124)
C
Ω Ω C

Also,
∫ ∫ ∫
   
A3 = Ψw dμdiag = u, w̄ dL 2n = w(z), ū(z) dL 2n (z). (7.4.125)
C C
D Ω Ω

In a similar fashion, (7.4.117) implies

∫
 n.t. 
νD · Fw ∂D dσD (7.4.126)
∂∗ D
∫ 
1   κ−n.t.   
=− w(z), ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)dL 2n (z)
2 C
C
∂∗ D
∫  ∫ 
1   κ−n.t.  
=− w(z), ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ) dL 2n (z).
2 C
∂∗ Ω C
Ω

Observe that (7.4.54) ensures that the differential form U, originally defined in
1 (Ω, L 2n ) ⊗ Λα,β−1 . Keeping this in mind,
(7.4.50), actually belongs to the space Lloc
from (7.4.121)-(7.4.126), (7.2.16), and the fact that w ∈ 𝒞∞c (Ω) ⊗ Λ
β,α is arbitrary

we may now conclude that

7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 849
∫  
∂U
¯ = ∂¯z u(ζ), ∂¯ζ Γα,β−1 (ζ, z) dL 2n (ζ)
C
Ω
∫
 
=− (∂u)(ζ),
¯ ∂¯ζ Γα,β (ζ, z) dL 2n (ζ) + u(z)
C
Ω
∫
1   κ−n.t.  
+ ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ) (7.4.127)
2 C
∂∗ Ω

in the sense of distributions in Ω. Since, once again thanks to (7.4.54), the right-hand
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λα,β .
side of (7.4.127) is locally integrable in Ω, we deduce that ∂U loc
This proves (7.4.50) and ultimately formula (7.4.51) is a reinterpretation of (7.4.127)
with this observation in mind. This finishes the proof of Theorem 7.4.3. 

Here is a companion to Theorem 7.4.1. The integral formulas in this result are
going to be useful in the proof of Theorem 7.4.6, stated a little later.

Theorem 7.4.4 Fix n ∈ N with n ≥ 2 and let Ω ⊆ R2n ≡ Cn be an open set with
a lower Ahlfors regular boundary, with the property that σ := H 2n−1 ∂Ω is a
doubling measure on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its
geometric measure theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Also,
fix a pair of degrees α, β ∈ {0, 1, . . . , n} along with some aperture parameter κ > 0.
1 (Ω, L 2n ) ⊗ Λα,β is a differential form satisfying (with all
Lastly, suppose u ∈ Lloc
partial differential operators considered in the sense of distributions in Ω):
¯ ∈ L 1 (Ω, L 2n ) ⊗ Λα,β+1, ϑu ∈ L 1 (Ω, L 2n ) ⊗ Λα,β−1,
∂u loc loc
 L 2n (ζ) 
Δu belongs to the space L 1 Ω, ⊗ Λα,β, (7.4.128)
1 + |ζ | 2n−2
κ−n.t. κ−n.t. κ−n.t.
u∂Ω
¯ 
, (∂u) ∂Ω
, (ϑu) ∂Ω
exist at σ-a.e. point on ∂ Ω, nta

and
∫ ∫
(Nκ u)(ζ) Nκ (∂u)(ζ)
¯ + Nκ (ϑu)(ζ)
dσ(ζ) < ∞, dσ(ζ) < ∞. (7.4.129)
1 + |ζ | 2n−1 1 + |ζ | 2n−2
∂Ω ∂Ω

κ  −n.t. κ  −n.t. κ  −n.t.

Then for any κ  > 0 the nontangential traces u∂Ω , (∂u) ¯ 
∂Ω
, (ϑu)∂Ω also
exist σ-a.e. on ∂nta Ω and are actually independent of κ . Moreover, with the depen-
dence on the parameter κ  dropped, for L 2n -a.e. point z ∈ Ω one has (with absolutely
convergent integrals)
850 7 Applications to Analysis in Several Complex Variables
∫  
1  n.t. 
u(z) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1  n.t. 
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1 n.t. 
+ ¯  (ζ), ν 0,1 (ζ) ∧ Γα,β (ζ, z)
(∂u) dσ(ζ)
2 ∂Ω
∂∗ Ω C

∫  
1 n.t. 
− (ϑu)∂Ω (ζ), ν 1,0 (ζ) ∨ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1
− (Δu)(ζ), Γα,β (ζ, z) dL 2n (ζ) (7.4.130)
2 C
Ω

if either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω is

an exterior domain, the same conclusion holds true under the additional assumption
that there exists λ ∈ (1, ∞) such that
⨏
|u|C dL 2n = o(1) as R → ∞, and
B(0,λ R)\B(0,R)

⨏ (7.4.131)
¯ + |ϑu| dL 2n = o(R) as R → ∞.
| ∂u|C C
B(0,λ R)\B(0,R)

Furthermore, for each z ∈ Cn \ Ω one has

∫  
1  n.t. 
0=− ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1  n.t. 
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1  n.t. 
+ ¯  (ζ), ν 0,1 (ζ) ∧ Γα,β (ζ, z) dσ(ζ)
(∂u)
2 ∂∗ Ω ∂Ω
C

∫  
1  n.t. 
− (ϑu)∂Ω (ζ), ν 1,0 (ζ) ∨ Γα,β (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫  
1
− (Δu)(ζ), Γα,β (ζ, z) dL 2n (ζ), (7.4.132)
2 C
Ω

with the same caveat as in (7.4.131) when Ω is an exterior domain.

Finally, suitable versions of these results hold when n = 1, taking into account the
fact that now the fundamental solution for the Laplacian in C ≡ R2 has a logarithmic
behavior.
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 851

Proof For starters, (7.4.129) and [68, Lemma 8.3.1] imply

∞ (Ω, L 2n ) ⊗ Λα,β,
u ∈ Lloc ¯ ∈ L ∞ (Ω, L 2n ) ⊗ Λα,β+1,
∂u loc
(7.4.133)
∞ (Ω, L 2n ) ⊗ Λα,β−1 .
and ϑu ∈ Lloc

In addition, from (7.4.128), (7.4.129), and [68, Corollary 8.9.9] it follows that for
κ  −n.t. κ  −n.t. κ  −n.t.
any κ  > 0 the nontangential traces u∂Ω , (∂u) ¯ 
∂Ω
, (ϑu) 
∂Ω
also exist σ-a.e. on

∂nta Ω and are actually independent of the parameter κ . Also, with the dependence
on κ  dropped, from [68, (8.9.8)], [68, (8.9.44)], and the last condition in (7.4.128)
we see that
∫  un.t.  (ζ)
∂Ω C
dσ(ζ) < +∞ and
1 + |ζ | 2n−1
∂∗ Ω
(7.4.134)
∫  (∂u) n.t.    n.t.  
¯  (ζ) +  (ϑu) (ζ)
∂Ω C ∂Ω C
dσ(ζ) < +∞.
1 + |ζ | 2n−2
∂∗ Ω

In turn, this ensure that, for each fixed point z ∈ Ω, the boundary integrals in
the first four lines of (7.4.130) are absolutely convergent. Also, recall from [68,
Lemma 3.5.7] that the integrability condition in the second line of (7.4.128) is
equivalent with having for L 2n -a.e. point z ∈ Cn the finiteness condition
∫
|(Δu)(ζ)|C
dL 2n (ζ) < +∞. (7.4.135)
Ω |ζ − z|
2n−2

As in the proof of Theorem 7.4.1, the strategy for actually proving formula
(7.4.130) is to apply [68, Theorem 1.2.1] to a suitable vector field. Concretely, fix a
Lebesgue point z ∈ Ω for (allthe coefficients
n of) u with the property that (7.4.135)
holds, and define F : Ω → Λα,β Cn by requiring (with convention (7.4.11) in
place) that

 1  0,1  1 
ξ · F(ζ) =− ξ ∧ u(ζ), ∂¯ζ Γα,β (ζ, z) + ξ 1,0 ∨ u(ζ), ϑζ Γα,β (ζ, z)
2 C 2 C


1 ¯  1  
+ (∂u)(ζ), ξ 0,1 ∧ Γα,β (ζ, z) − (ϑu)(ζ), ξ 1,0 ∨ Γα,β (ζ, z)
2 C 2 C

for all real vectors ξ ∈ R2n, at L 2n -a.e. point ζ ∈ Ω. (7.4.136)

Note that the right-hand side of (7.4.136) depends linearly in ξ ∈ R2n . As such, the
demand in (7.4.136) determines F uniquely and unambiguously. In addition, from
(7.4.133), (7.4.136), and (7.3.18)-(7.3.19) we have
 1 n
F ∈ Lloc (Ω, L 2n ) ⊗ Λα,β . (7.4.137)
852 7 Applications to Analysis in Several Complex Variables

A computation in the spirit of (7.4.32) (cf. also [70, (1.5.37)] for details in similar
circumstances) then gives
 
divF = − u, Γα,β (·, z) + u(z)δz
C
 
= 1
2 Δu, Γα,β (·, z) + u(z)δz in D (Ω) ⊗ Λα,β, (7.4.138)
C

where δz is the Dirac distribution in Ω with mass at z. In particular, from (7.4.138)

and (7.4.135) we conclude that

divF ∈ ℰ(Ω) ⊗ Λα,β + L 1 (Ω, L 2n ) ⊗ Λα,β . (7.4.139)

 
Introducing the compact subset of Ω given by K := B z, 12 dist(z, ∂Ω) , we may
conclude from [68, Lemma 8.3.7], (7.4.136), (7.3.18)-(7.3.19), (7.4.129), and [68,
(8.2.26)] that

NκΩ\K F ∈ L 1 (∂Ω, σ). (7.4.140)

Furthermore, from (7.4.136), the second line in (7.4.128), and [68, (8.9.10)-(8.9.11)]
we conclude that
κ−n.t.

F exists at σ-a.e. point on ∂nta Ω (7.4.141)
∂Ω

and at σ-a.e. point ζ ∈ ∂nta Ω we have

 κ−n.t.  1  κ−n.t.  

ν(ζ) · F (ζ) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z)
∂Ω 2 C

1    κ−n.t.  
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z)
2 C

1   ¯ n.t.  
+ (∂u) ∂Ω (ζ), ν 0,1 (ζ) ∧ Γα,β (ζ, z)
2 C


1  n.t.  
− (ϑu)∂Ω (ζ), ν 1,0 (ζ) ∨ Γα,β (ζ, z) . (7.4.142)
2 C

Next, condition (7.4.131) together with (7.3.18)-(7.3.19) ensures that

if the open set Ω ⊆ R2n is an exterior domain, then the

(7.4.143)
vector field F satisfies the growth condition [68, (1.4.8)].

Collectively, (7.4.137), (7.4.139), and (7.4.141) imply that the vector field F satisfies
the hypotheses of [68, Theorem 1.4.1]. On account of [68, (4.6.19)], (7.4.138),
and (7.4.142), the Divergence Formula recorded in [68, (1.4.6)] currently yields
(7.4.130). Lastly, formula (7.4.132) is established similarly (the proof is actually
simpler since for z ∈ Cn \ Ω the form Γα,β (·, z) has coefficients in 𝒞∞ (Ω)). 
7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 853

Before going any further, we wish to comment on the nature of the integral
representation formula (7.4.130). Specifically, this involves several integral operators
which are worth singling out. First, we have what may be considered a complex
 ) 
double layer operator, whose action on a form f ∈ L 1 ∂∗ Ω, 1+σ(ζ |ζ | 2n−1
⊗ Λα,β is
defined at each z ∈ Ω as
∫  
1
Dα,β f (z) := f (ζ), ν 0,1 (ζ) ∧ ϑζ Γα,β (ζ, z) − ν 1,0 (ζ) ∨ ∂¯ζ Γα,β (ζ, z) dσ(ζ).
2 C
∂∗ Ω
(7.4.144)
In particular, with Δ := ∂12 + · · · + ∂2n
2 denoting the Laplacian in R2n , we have

(compare with [70, (2.5.205)])

 
Dα,β f ∈ 𝒞∞ (Ω) ⊗ Λα,β and Δ Dα,β f = 0 in Ω. (7.4.145)

Also, corresponding to α = β = 0,

D0,0 = B0,0 . (7.4.146)

Second, for suitable differential forms f : ∂∗ Ω → Λα,β Cn let us define the boundary-
to-domain single layer potential operator
∫
 
𝒮α,β f (z) := f (ζ), Γα,β (ζ, z) dσ(ζ), ∀z ∈ Ω, (7.4.147)
C
∂∗ Ω

Its boundary-to-boundary version, i.e.,

∫
 
Sα,β f (z) := f (ζ), Γα,β (ζ, z) dσ(ζ), z ∈ ∂Ω. (7.4.148)
C
∂∗ Ω

is going to be important later on (in the proof of Theorem 7.4.6). Finally, in the same
context as above, define the complex volume (Newtonian) potential operator acting
on suitable differential forms U : Ω → Λα,β Cn according to
∫
1  
Πα,β U(z) := − U(ζ), Γα,β (ζ, z) dL 2n (ζ), z ∈ Ω. (7.4.149)
2 Ω C

In this notation, the integral representation formula (7.4.130) may be succinctly

recast as
 n.t. 
u(z) = Dα,β u (z) ∂Ω
  n.t.   n.t.  
− 𝒮α,β 2ν
1 0,1
∧ (ϑu)∂Ω − 12 ν 1,0 ∨ (∂u)
¯ 
∂Ω
(z)

+ Πα,β (Δu)(z). (7.4.150)

The density-form on which the single layer is acting in the second line above, i.e.,
854 7 Applications to Analysis in Several Complex Variables
 n.t.   n.t. 
2ν
1 0,1
∧ (ϑu)∂Ω − 12 ν 1,0 ∨ (∂u)
¯ 
∂Ω
 n.t.   n.t. 
¯ ν) (ϑu)
= (−i) Sym(∂; ¯ 
+ (−i) Sym(ϑ; ν) (∂u) (7.4.151)
∂Ω ∂Ω

is precisely the co-normal derivative associated with the factorization of the complex
Laplacian  as ∂ϑ ¯ + ϑ ∂¯ (cf. (7.3.14)) acting on u. In this vein, it is worth noting that
the integral kernel of the complex double layer (7.4.144) is precisely this conormal
derivative acting on Γα,β (ζ, z) in the variable ζ. Ultimately, formula (7.4.150) is the
natural analogue of Green’s third identity discussed in [70, Theorem 1.5.1] (cf. [70,
(1.5.4)]) for the complex Laplacian.
In our next proposition we prove energy identities for the complex Laplacian 
in a very general geometric setting.

Proposition 7.4.5 Suppose Ω ⊆ R2n ≡ Cn is a bounded open set with a lower

Ahlfors regular boundary, and with the property that σ := H 2n−1 ∂Ω is a doubling
measure on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric
measure theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Pick a pair of
arbitrary degrees α, β ∈ {0, 1, . . . , n} and an aperture parameter κ > 0. In this
context, suppose the differential form u ∈ 𝒞∞ (Ω) ⊗ Λα,β satisfies:

Δu = 0 in Ω, Nκ u ∈ L (2n−1)/(n−1) (∂Ω, σ),

κ−n.t. κ−n.t. κ−n.t.
u∂Ω , (∂u)
¯ 
∂Ω
, (ϑu)∂Ω exist σ-a.e. on ∂nta Ω, (7.4.152)

¯ Nκ (ϑu) ∈ L (2n−1)/n (∂Ω, σ).

and Nκ (∂u),

Then for any other specified aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ  −n.t. κ  −n.t.
boundary traces u∂Ω , (∂u) ¯ 
∂Ω
, and (ϑu)∂Ω exist at σ-a.e. point on ∂nta Ω and
are actually independent of κ . Moreover, with the dependence on the parameter
κ  dropped, the following energy identity, involving absolutely convergent integrals,
holds:
∫  ∫ 
  2  2 n.t. n.t. 
¯  + ϑu dL 2n = − 1
∂u ν 1,0 ∨ (∂u) ¯  , u dσ
C C 2 ∂∗ Ω ∂Ω ∂Ω
Ω C

∫ 
1 n.t. n.t. 
+ ν 0,1 ∧ (ϑu)∂Ω, u∂Ω dσ. (7.4.153)
2 ∂∗ Ω C

Proof To get started, observe that (2n−1)/(n−1) and (2n−1)/n are Hölder conjugate
exponents and that, thanks to [68, (8.6.51)], the last line in (7.4.152) implies
¯ ∈ L 2 (Ω, L 2n ) ⊗ Λα,β+1,
∂u ϑu ∈ L 2 (Ω, L 2n ) ⊗ Λα,β−1 . (7.4.154)

To proceed, introduce the first-order differential operators

7.4 A Sharp Version of the Bochner-Martinelli-Koppelman Formula and Related Topics 855
!
∂¯  
D :=  := ϑ, ∂¯ .
and D (7.4.155)
ϑ

Then (7.3.14)-(7.3.16) imply that

 = ϑ ∂¯ + ∂ϑ
L := DD ¯ =  = − 1 Δ. (7.4.156)
2

Moreover, choosing w := ū it follows from (7.2.3) and (7.2.17) that

  n.t.  n.t. 
 ν) (Du)
(−i)Sym( D; , w ∂Ω
∂Ω

" n.t. )
  (∂u) 
# ¯ ∂Ω & n.t.
= − 2ν ∨ ·,
1 1,0
2ν
∧· $
1 0,1
n.t. ' , u ∂Ω
(ϑu) 
% ∂Ω ( C
 n.t.  n.t.  n.t. 
¯  + 1 ν 0,1 ∧ (ϑu) , u
= − 12 ν 1,0 ∨ (∂u) . (7.4.157)
∂Ω 2 ∂Ω ∂Ω
C

In addition, from (7.4.155) and (7.2.16) we see that

! !
ϑ 
∂
D = = , (7.4.158)
∂¯ ϑ̄

hence with the same choice of w as above we have

" ! !)
  ∂u
¯ ∂u
¯  2  2
 w =
Du, D 
, = ∂u
¯  + ϑu . (7.4.159)
ϑu ϑu C C
C

With these identifications in hand, all desired conclusions now follow directly from
[70, Theorem 1.7.18]. This finishes the proof of the proposition. 
Pressing on, recall the complex double layer and single layer operators, Dα,β
and 𝒮α,β , Sα,β , introduced in (7.4.144) and (7.4.147)-(7.4.148), respectively. The
point of our next theorem is that even though the kernel of the complex double layer
Dα,β is merely harmonic in the variable z, having a form u reproduced by Dα,β
acting on its nontangential boundary trace turns out to be equivalent to having u
simultaneously ∂-closed
¯ and ϑ-closed.
Theorem 7.4.6 Fix n ∈ N with n ≥ 2 and suppose Ω is a bounded UR domain in
R2n ≡ Cn . Abbreviate σ := H 2n−1 ∂Ω, select α, β ∈ {0, 1, . . . , n}, and pick some
κ > 0. In this context, assume u ∈ 𝒞1 (Ω) ⊗ Λα,β is a differential form such that
κ−n.t.
Nκ u ∈ L (2n−1)/(n−1) (∂Ω, σ) and u∂Ω exists σ-a.e. on ∂Ω. Then

Nκ (∂u),
¯ Nκ (ϑu) ∈ L (2n−1)/n (∂Ω, σ) ⎫
⎪
⎬
⎪
 κ−n.t.  ⇐⇒ ∂u
¯ = 0 and ϑu = 0 in Ω.
u = Dα,β u∂Ω at each point in Ω ⎪ ⎪
⎭
(7.4.160)
856 7 Applications to Analysis in Several Complex Variables

As a corollary (corresponding to the scalar case, when α = β = 0), if F ∈ 𝒞1 (Ω)

is a complex-valued function with the property that Nκ F ∈ L (2n−1)/(n−1) (∂Ω, σ) and
κ−n.t.
F ∂Ω exists σ-a.e. on ∂Ω, then

¯ ∈ L (2n−1)/n (∂Ω, σ) ⎫
Nκ (∂F) ⎪
⎬
⎪
   ⇐⇒ F is holomorphic in Ω. (7.4.161)
κ−n.t.
and F = B0,0 F ∂Ω in Ω ⎪ ⎪
⎭
Proof Consider the right-pointing implication in (7.4.160). The fact that, by as-
sumption,
 κ−n.t. 
u = Dα,β u∂Ω in Ω (7.4.162)
implies (cf. (7.4.145)) that u is harmonic in Ω. Granted this and the hypotheses made
in (7.4.160), we may invoke [70, (3.1.148)] to conclude that
κ−n.t. κ−n.t.
¯ 
(∂u) , (ϑu)  exist σ-a.e. on ∂Ω. (7.4.163)
∂Ω ∂Ω

As such, all hypotheses of Theorem 7.4.4 are satisfied and, with the dependence on
the aperture parameter κ dropped, at each point z ∈ Ω the integral representation
formula (7.4.130) currently becomes
∫  
1  n.t. 
u(z) = − ν 0,1 (ζ) ∧ u∂Ω (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ∂Ω C
∫  
1  n.t. 
+ ν 1,0 (ζ) ∨ u∂Ω (ζ), ϑζ Γα,β (ζ, z) dσ(ζ)
2 ∂Ω C
∫  
1  n.t. 
+ ¯  (ζ), ν 0,1 (ζ) ∧ Γα,β (ζ, z) dσ(ζ)
(∂u)
2 ∂Ω ∂Ω
C
∫  
1  n.t. 
− (ϑu)∂Ω (ζ), ν 1,0 (ζ) ∨ Γα,β (ζ, z) dσ(ζ). (7.4.164)
2 ∂Ω C

Contrasting this with (7.4.162) then leads to the conclusion that

∫  
1  n.t. 
¯  (ζ), ν 0,1 (ζ) ∧ Γα,β (ζ, z) dσ(ζ)
(∂u)
2 ∂Ω ∂Ω
C
∫  
1  n.t. 
− (ϑu)∂Ω (ζ), ν 1,0 (ζ) ∨ Γα,β (ζ, z) dσ(ζ) = 0
2 ∂Ω C
(7.4.165)

at each point z ∈ Ω. In terms of (7.4.147) this translates into saying that

𝒮α,β f = 0 in Ω, (7.4.166)

where we have abbreviated

7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 857

1 1,0  ¯ n.t.  1 0,1  n.t. 

f := ν ∨ (∂u) ∂Ω − ν ∨ (ϑu)∂Ω ∈ L (2n−1)/n (∂Ω, σ) ⊗ Λα,β .
2 2
(7.4.167)

Going nontangentially to the boundary in (7.4.166) then proves (cf. [70, Proposi-
tion 2.5.39]) that, on the one hand,

Sα,β f = 0 on ∂Ω. (7.4.168)

On the other hand, by arguing as in the proof of [76, Proposition 3.5, p. 126] (while
keeping part (7) of [68, Lemma 5.10.9] in mind) we conclude that the operator

Sα,β : L (2n−1)/n (∂Ω, σ) ⊗ Λα,β −→ L (2n−1)/(n−1) (∂Ω, σ) ⊗ Λα,β

(7.4.169)
is well defined, linear, bounded, injective, with dense range.

Collectively, (7.4.167)-(7.4.169) imply that f = 0, i.e.,

1 1,0  ¯ n.t.  1 0,1  n.t. 
ν ∨ (∂u) ∂Ω − ν ∨ (ϑu)∂Ω = 0 at σ-a.e. point on ∂Ω. (7.4.170)
2 2
The working assumptions on u guarantee that the hypotheses of Proposition 7.4.5 are
satisfied. As such, we may write the energy identity (7.4.153) which, given (7.4.170),
simply reduces to
∫ 
 2   2
∂u
¯  + ϑu dL 2n = 0. (7.4.171)
C C
Ω

In turn, this forces ∂u

¯ = 0 and ϑu = 0 in Ω, finishing the proof of the right-pointing
implication in (7.4.160).
As regards the left-pointing implication in (7.4.160), since the differential u now
satisfies the hypotheses of Theorem 7.4.1, we may write the integral representation
formula (7.4.3). Given that we are presently assuming ∂u ¯ = 0 and ϑu = 0 in Ω,
the latter reduces precisely to (7.4.162). Having established this, all other desired
conclusions are immediate. 

7.5 The Extension Problem for Hölder CR-Functions on

Boundaries of Ahlfors Regular Domains

As a prelude to discussing the Extension Problem for Hölder CR-functions on

boundaries of Ahlfors regular domains, we shall first study the action of the Bochner-
Martinelli integral operator in such a context. To set the stage, we make the following
observation. Let Ω ⊆ R2n ≡ Cn be a set of locally finite perimeter. Abbreviate
σ := H 2n−1 ∂Ω and denote by ν the geometric measure theoretic outward unit
normal to Ω. In this setting, for each σ-measurable, complex-valued, function defined
on ∂∗ Ω, satisfying
858 7 Applications to Analysis in Several Complex Variables
∫
| f (ζ)|
dσ(ζ) < +∞, (7.5.1)
1 + |ζ | 2n−1
∂∗ Ω

the scar case of formula (7.3.33) (i.e., when α = β = 0) corresponds to

∫
1  0,1 
B0,0 f (z) = − ν (ζ), ∂¯ζ Γ0,0 (ζ, z) f (ζ) dσ(ζ)
2 ∂∗ Ω C

∫ n
=− (νC ) j (ζ)∂ζ j [En (ζ, z)] f (ζ) dσ(ζ)
∂∗ Ω j=1

∫ n
1 ζ¯j − z̄ j
= (ν ) j (ζ) f (ζ) dσ(ζ)
ω2n−1 ∂∗ Ω j=1 |ζ − z| 2n C

∫  
1 νC (ζ), ζ − z
C
= f (ζ) dσ(ζ), ∀z ∈ Cn \ ∂Ω. (7.5.2)
ω2n−1 ∂∗ Ω |z − ζ | 2n

Theorem 7.3.5 specialized to the case α = β = 0 yields a wealth of information

about the nature of the operator (7.5.2). We wish to augment this body of results by
establishing the properties in the proposition below.
Proposition 7.5.1 Suppose Ω ⊆ R2n ≡ Cn is an open set of locally finite perimeter,
and abbreviate σ := H 2n−1 ∂Ω. Then the following properties hold.
(i) For every σ-measurable function f : ∂∗ Ω → C satisfying the integrability
condition stated in (7.5.1) one has (with Δ denoting the Laplacian in R2n )

B0,0 f ∈ 𝒞∞ (Cn \ ∂Ω) and Δ(B0,0 f ) = 0 in Cn \ ∂Ω. (7.5.3)

(ii) Under the additional assumption that ∂Ω is bounded one has


1 in Ω, if Ω is bounded,
B0,0 1 ≡ (7.5.4)
0 in Ω, if Ω is unbounded.

(iii) Under the additional assumption that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors
regular, it follows that for each α ∈ (0, 1) there exists some C ∈ (0, ∞), depending
only on n, α, diam(∂∗ Ω), and the upper Ahlfors regularity constant of ∂∗ Ω, such
that for every function f ∈ 𝒞α (∂∗ Ω) one has
    
sup (B0,0 f )(z) + sup dist(z, ∂∗ Ω)1−α ∇(B0,0 f )(z) ≤ C f 𝒞α (∂∗ Ω) . (7.5.5)
z ∈Ω z ∈Ω

In particular, if Ω ⊆ R2n ≡ Cn is a uniform domain whose boundary is compact

and upper Ahlfors regular then B0,0 induces a well-defined, linear, and bounded
operator in the context
 
B0,0 : 𝒞α (∂∗ Ω) −→ 𝒞α Ω , ∀α ∈ (0, 1). (7.5.6)
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 859

As a corollary, if Ω ⊆ R2n ≡ Cn is simultaneously a uniform domain, and an

Ahlfors regular domain with compact boundary, then for each α ∈ (0, 1) the
operator B0,0 is well defined, linear, and bounded in the context
 
B0,0 : 𝒞α (∂Ω) −→ 𝒞α Ω . (7.5.7)

(iv) Strengthen the local finite perimeter assumption on Ω by demanding that ∂Ω is

p
a UR set. Then for each function f belonging to L1 (∂∗ Ω, σ) with p ∈ [1, ∞) (the
L p -based Sobolev space of order one on ∂∗ Ω, relative to the ambient R2n ; cf.
[69, Chapter 11]) one has
 
Nκ ∇B0,0 f L p (∂Ω,σ) ≤ C f L p (∂∗ Ω,σ) if 1 < p < ∞, (7.5.8)
1
 
Nκ ∇B0,0 f L 1,∞ (∂Ω,σ)
≤C f L11 (∂∗ Ω,σ) if p = 1, (7.5.9)

where ∇ denotes the gradient operator in the ambient R2n and C ∈ (0, ∞) is a
constant independent of f .
(v) Under the additional assumption that ∂∗ Ω is an upper Ahlfors regular set, given
p,1
any function f ∈ LC (∂∗ Ω, σ) with p ∈ [1, ∞) and any index j ∈ {1, . . . , n}, for
every z ∈ C \ ∂Ω one has
n

∫ n
  1 ζ¯k − z̄k  C 
∂z̄ j B0,0 f (z) = ∂τ f (ζ) dσ(ζ). (7.5.10)
ω2n−1 |ζ − z| 2n j k
∂∗ Ω k=1

As a corollary of (7.5.10), item (3) in [70, Theorem 2.4.1], and [70, Theo-
rem 2.5.1], if Ω ⊆ R2n ≡ Cn is open and ∂Ω is a UR set then for each
p,1
f ∈ LC (∂∗ Ω, σ) with p ∈ [1, ∞) and each κ > 0 the nontangential boundary
trace
  κ−n.t.
∂z̄ j B0,0 f ∂Ω exists at σ-a.e. point on ∂∗ Ω
(7.5.11)
for each index j ∈ {1, . . . , n},

and there exists a constant C ∈ (0, ∞), independent of f , with the property that
n
 
Nκ ∂z̄ j B0,0 f L p (∂Ω,σ)
≤C f p,1
LC (∂∗ Ω,σ)
if 1 < p < ∞, (7.5.12)
j=1

n
 
Nκ ∂z̄ j B0,0 f L 1 (∂Ω,σ)
≤C f LC1,1 (∂∗ Ω,σ) if p = 1. (7.5.13)
j=1

(vi) Make the additional assumption that ∂∗ Ω is an upper Ahlfors regular set, and
fix a function
860 7 Applications to Analysis in Several Complex Variables

f ∈ L p (∂∗ Ω, σ) with p ∈ [1, ∞) such that

(7.5.14)
∂τCj k f = 0 on ∂∗ Ω for each j, k ∈ {1, . . . , n}.

Then

B0,0 f is holomorphic in Cn \ ∂Ω (7.5.15)

and
if one also assumes that ∂Ω is bounded and n > 1, then
the function B0,0 f vanishes identically in the unbounded (7.5.16)
connected component of the set Cn \ ∂Ω.
(vii) Under the additional assumption that ∂Ω is a compact UR set, for each exponent
p ∈ (1, ∞) there exists some constant C ∈ (0,  ∞) with pthe property that for each
function f ∈ BMO(∂Ω, σ) it follows that ∇(B0,0 f ) dist(·, ∂Ω) p−1 dL 2n is a
Carleson measure in Ω in the quantitative sense that
∫
1  
sup   ∇(B0,0 f ) p dist(·, ∂Ω) p−1 dL 2n
z ∈∂Ω, r >0 σ B(z, r) ∩ ∂Ω B(z,r)∩Ω
p
≤C f BMO(∂Ω,σ)
. (7.5.17)

Moreover, for each p ∈ (1, ∞) and f ∈ VMO(∂Ω, σ) it follows that

 
∇(B0,0 f ) p dist(·, ∂Ω) p−1 dL 2n
(7.5.18)
is a vanishing Carleson measure in Ω.

Proof Starting with (7.5.2) and unraveling definitions we see that

n
B0,0 = D + i R (2j−1)(2j) (7.5.19)
j=1

where D is the (boundary-to-domain) harmonic double layer potential operator

associated with the present domain Ω as in [70, Definition 2.5.17], and each R (2j−1)(2j)
is the operator associated with the given domain Ω as in (A.0.189). Granted this, all
claims in items (i)-(iii) become direct consequences of [70, Lemmas 2.5.18, 2.5.19],
and [70, Proposition 2.5.26]. Moreover, the claim in item (vii) also follows from
(7.5.19), in light of [70, Propositions 2.5.24, 2.5.26].
p
Consider next the claim in item (iv). Having fixed a function f ∈ L1 (∂∗ Ω, σ) with
p ∈ [1, ∞), it follows from (7.5.19) that for each  ∈ {1, . . . , 2n} the (real) spatial
partial derivative ∂ B0,0 f may be expressed as
n
∂ B0,0 f = ∂ D f + i ∂ R (2j−1)(2j) f in Ω. (7.5.20)
j=1
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 861

On the other hand, from (A.0.189) and [69, Lemma 11.1.7] we see that for each
j, k,  ∈ {1, . . . , 2n} we have
∫
∂ R jk f (z) = − (∂ EΔ )(z − ζ)(∂τ j k f )(ζ) dσ(ζ), ∀z ∈ Ω, (7.5.21)
∂∗ Ω

where EΔ denotes the standard fundamental solution for the Laplacian in R2n (defined
as in (A.0.65) with n replaced by 2n), and ∂τ j k is the real tangential partial differential
operator (defined as in [69, Chapter 11] relative to the ambient R2n ). Then the
estimates in (7.5.8)-(7.5.9) are consequences of (7.5.20)-(7.5.21), estimate (1.5.8)
and the subsequent comment in Theorem 1.5.1 (applied to the harmonic double layer
introduced in [70, Definition 2.5.17]), and item (3) in [70, Theorem 2.4.1] (which
provides nontangential maximal function estimates for the integral operators in the
right-hand side of (7.5.21)).
Next we deal with the claim made in item (v). To this end, having fixed an arbitrary
p,1
function f ∈ LC (∂∗ Ω, σ) with p ∈ [1, ∞), along with some j ∈ {1, . . . , n}, for each
z ∈ C \ ∂Ω we may compute
n

∫ n
 
∂z̄ j B0,0 f (z) = − (νC )k (ζ)∂z̄ j ∂ζk [En (ζ, z)] f (ζ) dσ(ζ)
∂∗ Ω k=1

∫ n
= (νC )k (ζ)∂ζ¯j ∂ζk [En (ζ, z)] f (ζ) dσ(ζ)
∂∗ Ω k=1

∫ n  
= ∂τC ∂ζk [En (ζ, z)] f (ζ) dσ(ζ)
k j (ζ )
∂∗ Ω k=1

∫ n
 
= ∂ζk [En (ζ, z)] ∂τCj k f (ζ) dσ(ζ). (7.5.22)
∂∗ Ω k=1

Above, the first equality is obtained by differentiating under the integral sign in
(7.5.2), the second equality is seen from (7.3.18), the third equality uses (7.1.5) and
the fact that
n
∂ζ¯k ∂ζk [En (ζ, z)] = Δζ [En (ζ, z)] = 0, (7.5.23)
k=1

and the final equality is a consequence of Proposition 7.1.3 (whose applicability is

presently ensured by the assumption that ∂∗ Ω is upper Ahlfors regular). This justifies
(7.5.22), and (7.5.10) readily follows from it on account of (7.3.18).
Let us now deal with the claims in item (vi), under the additional assumption
that ∂∗ Ω is upper Ahlfors regular. First, that the function B0,0 f is holomorphic in
Cn \ ∂Ω if f is as in (7.5.14) is clear from item (v). Second, to prove the claim in
862 7 Applications to Analysis in Several Complex Variables

(7.5.16), let us also assume that ∂Ω is bounded and that n ≥ 2. With f as in (7.5.14),
abbreviate F := B0,0 f . Then F is holomorphic in Cn \ ∂Ω and, as is apparent from
(7.5.2), decays at infinity. Consider next a number R ∈ (0, ∞) which is sufficiently
large so that if

Q := B(0, R)n := B(0, R) × · · · × B(0, R) ⊆ Cn (7.5.24)

(where each factor in the Cartesian product is the origin-centered ball of radius R in
C), then ∂Ω ⊆ Q. Pick an index j ∈ {1, . . . , n} and fix a point z ∗j ∈ C \ B(0, R), then
define the function G j : Cn−1 → C by setting

G j (z1, . . . , zn−1 ) := F(z1, . . . , z j−1, z ∗j , z j+1, . . . , zn−1 )

(7.5.25)
for each (z1, . . . , zn−1 ) ∈ Cn−1 .

Note that (z1, . . . , z j−1, z ∗j , z j+1, . . . , zn−1 )  Q for each (z1, . . . , zn−1 ) ∈ Cn−1 which,
in turn, implies (z1, . . . , z j−1, z ∗j , z j+1, . . . , zn−1 ) ∈ Cn \ ∂Ω. This ensures that G j is
well defined. Also, by design, G j is holomorphic in Cn−1 and decays at infinity.
Granted these (and bearing in mind that n ≥ 2), Liouville’s Theorem applies and
yields that G j ≡ 0 in Cn−1 . Given the arbitrariness of j ∈ {1, . . . , n} and
z ∗j ∈ C \ B(0, R), this ultimately translates into saying that F ≡ 0 in Cn \ Q. Denote
by U the unbounded connected component of the set Cn \ ∂Ω. Since U overlaps
with Cn \ Q on a nonempty open set, unique continuation implies that F necessarily
vanishes in U. This concludes the proof of (7.5.16).
Lastly, the claims in the current item (viii) follows from items (4)-(5) in The-
orem 1.8.2 bearing in mind that, as discussed in Example 1.4.16, the Bochner-
Martinelli integral operator is a particular example of a double layer potential oper-
ator associated with a certain factorization of the Laplacian in R2n . 

Let Ω ⊆ R2n ≡ Cn be a set of locally finite perimeter. Abbreviate σ := H 2n−1 ∂Ω

and denote by ν the geometric measure theoretic outward unit normal to Ω. In this
(or a possibly more restrictive) geometric context, we would like to make sense
of the principal-value
  Bochner-Martinelli integral operator acting on any function
f ∈ L 1 ∂∗ Ω, 1+ | ·σ|2n−1 according to:
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 863
∫
1  
B0,0 f (z) = − lim+ ν 0,1 (ζ), ∂¯ζ Γ0,0 (ζ, z) f (ζ) dσ(ζ)
ε→0 2 C
ζ ∈∂∗ Ω
|z−ζ |>ε

∫ n
1 ζ¯j − z̄ j
= lim+ (ν ) j (ζ) f (ζ) dσ(ζ) (7.5.26)
ε→0 ω2n−1 j=1
|ζ − z| 2n C
ζ ∈∂∗ Ω
|z−ζ |>ε

∫  
1 νC (ζ), ζ − z
C
= lim+ f (ζ) dσ(ζ) for σ-a.e. z ∈ ∂∗ Ω.
ε→0 ω2n−1 |z − ζ | 2n
ζ ∈∂∗ Ω
|z−ζ |>ε

According to [68, Proposition 5.6.7], this limit exists and B0,0 f is a σ-measurable
function on ∂∗ Ω. Furthermore, the last claim in [68, Proposition 5.6.7] guarantees
that if Ω ⊆ Cn is a Lebesgue measurable set whose topological boundary ∂Ω is
countably rectifiable (of dimension 2n − 1) and has locally finite H 2n−1 measure
(hence, in particular, if ∂Ω is a UR set), then for each function f ∈ L 1 ∂∗ Ω, 1+ | ·σ|2n−1
the limit in (7.5.26) actually exists for σ-a.e. x ∈ ∂Ω and gives rise to a σ-measurable
complex-valued function on ∂Ω.
Theorem 7.3.5 (with α = β = 0) already provides a great deal of information
about this singular integral operator, and our next result further elaborates on the
properties enjoyed by the principal-value Bochner-Martinelli integral operator just
considered. Before stating it we make one more definition. Specifically, in the same
geometric context as above, for each p, q ∈ [1, ∞] consider the off-diagonal (partial)
Sobolev spaces
1, p,q
LC (∂∗ Ω, σ) := f ∈ L p (∂∗ Ω, σ) : ∂τCj k f ∈ L q (∂∗ Ω, σ) for 1 ≤ j, k ≤ n
(7.5.27)
equipped with the natural norm

f 1, p, q
LC (∂∗ Ω,σ)
:= f L p (∂∗ Ω,σ) + ∂τCj k f L q (∂∗ Ω,σ)
(7.5.28)
1≤ j,k ≤n

1, p,q
for each f ∈ LC (∂∗ Ω, σ). Note that, by design,
1, p, p
LC (∂∗ Ω, σ∗ ) = LC1, p (∂∗ Ω, σ∗ ) for each p ∈ [1, ∞]. (7.5.29)

Proposition 7.5.2 Let Ω  R2n ≡ Cn be a Lebesgue measurable set with a compact

upper Ahlfors regular boundary and abbreviate σ := H 2n−1 ∂Ω. Then for each
α ∈ (0, 1) the following statements are true.

(i) For each given complex-valued function f ∈ 𝒞α (∂∗ Ω) the limit defining the
principal-value Bochner-Martinelli integral operator B0,0 f (z) in (7.5.26) exists
for σ-a.e. point z ∈ ∂∗ Ω.
864 7 Applications to Analysis in Several Complex Variables

(ii) After possibly redefining B0,0 f on a σ-nullset contained in ∂∗ Ω, the assignment

f → B0,0 f thus defined induces a well-defined, linear, and bounded operator

B0,0 : 𝒞α (∂∗ Ω) −→ 𝒞α (∂∗ Ω) (7.5.30)

with the property that


+ 12 on ∂∗ Ω, if Ω is bounded,
B0,0 1 ≡ (7.5.31)
− 12 on ∂∗ Ω, if Ω is unbounded.

(iii) Under the additional assumption that the set Ω is open and ∂Ω is also lower
Ahlfors regular, the jump-formula (where I denotes the identity operator and
κ > 0 is an arbitrary fixed number) is valid:

for every complex-valued function f ∈ 𝒞α (∂∗ Ω) one has

  κ−n.t. (7.5.32)
B0,0 f  = ( 12 I + B0,0 ) f at σ-a.e. point on ∂∗ Ω.
∂Ω

(iv) In the case when it is also assumed that H 2n−1 (∂Ω \ ∂∗ Ω) = 0, one may replace
∂∗ Ω by ∂Ω in the formulation of (7.5.30), (7.5.31), and (7.5.32).
(v) Suppose the set Ω is actually an Ahlfors regular domain with compact boundary
and define Ω+ := Ω, Ω− := R2n \ Ω. In this setting, let B0,0 ± be the Bochner-

Martinelli integral operators associated with Ω± . That is, for each complex-
valued function f ∈ L 1 (∂∗ Ω, σ) define

±
  
B0,0 f := B0,0 f  . (7.5.33)
Ω±

Then the sets Ω± are also Ahlfors regular domains with compact boundaries,
∂(Ω± ) = ∂Ω, and ∂∗ (Ω+ ) = ∂∗ Ω = ∂∗ (Ω− ). In addition, the geometric measure
theoretic outward unit normals to Ω± are ±ν at σ-a.e. point on ∂Ω, where ν is
the geometric measure theoretic outward unit normal to Ω. Moreover, given any
κ > 0,

for every complex-valued function f ∈ 𝒞α (∂Ω) one has

 ±  κ−n.t. (7.5.34)
B0,0 f  = (± 12 I + B0,0 ) f at σ-a.e. point on ∂Ω.
∂Ω

(vi) Suppose Ω ⊆ R2n ≡ Cn is an open set with the property that

 
∂Ω is a UR set and H 2n−1 ∂nta Ω \ ∂∗ Ω = 0. (7.5.35)
 ) 
Then for each f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
the principal-value Bochner-Martinelli
integral operator (7.5.26) is well defined. Moreover, for each p ∈ (1, ∞) this
induces a linear and bounded mapping in the following settings:
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 865
p p
B0,0 : L1 (∂∗ Ω, σ) −→ L1 (∂∗ Ω, σ), (7.5.36)

B0,0 : LCp,1 (∂∗ Ω, σ) → LCp,1 (∂∗ Ω, σ). (7.5.37)

More generally, if (7.5.35) is satisfied then

p,q p,q
B0,0 : L1 (∂∗ Ω, σ) −→ L1 (∂∗ Ω, σ), (7.5.38)
1, p,q 1, p,q
B0,0 : LC (∂∗ Ω, σ) −→ LC (∂∗ Ω, σ), (7.5.39)

are well-defined, linear, and bounded operators whenever p, q ∈ (1, ∞).

Proof From (7.5.26) and definitions it follows that

n
B0,0 = K + i R(2j−1)(2j) (7.5.40)
j=1

where K is the boundary-to-boundary (i.e., principal-value) harmonic double layer

potential operator associated with the present domain Ω as in [70, (2.5.203)], and
each R(2j−1)(2j) is the operator associated with the given domain Ω as in (A.0.190).
Having noticed this, all claims in items (i)-(iv) become direct consequences of
Proposition 7.5.2 and [70, Proposition 2.5.27]. Going further, the claims in item (v)
readily follow from part (6) of [68, Lemma 5.10.9] and the current item (iv).
To deal with the claims in item (vi), assume Ω is an open set satisfying the
properties listed in (7.5.35). Then [70, Theorem 2.3.2] implies that for each function
 ) 
f ∈ L 1 ∂∗ Ω, 1+σ(ζ
|ζ | 2n−1
the limit defining B0,0 f (z) in (7.5.26) exists at σ-a.e. z ∈ ∂Ω.
Next, the fact that B0,0 in (7.5.36) is well defined, linear, and bounded is a direct
consequence of (7.5.40) and Theorem 1.5.1. As regards the corresponding claims
for the operator B0,0 in (7.5.37), observe first that (7.3.69) specialized to the case
when α = β = 0 gives that

B0,0 : L p (∂∗ Ω, σ) −→ L p (∂Ω, σ) is well defined, linear, and bounded. (7.5.41)

p,1
Next, fix an arbitrary function f ∈ LC (∂∗ Ω, σ) along some j ∈ {1, . . . , n} and some
aperture parameter κ > 0. Then at σ-a.e. point on ∂∗ Ω we may write
 
∂τCj k B0,0 f = ∂τCj k ( 12 I + B0,0 ) f − 12 ∂τCj k f (7.5.42)
    κ−n.t.     κ−n.t. 1 C
= (νC ) j ∂z̄k B0,0 f  − (νC )k ∂z̄ j B0,0 f  − 2 ∂τ j k f ,
∂Ω ∂Ω

where the second equality is a consequence of (7.3.82) (specialized to α = β = 0

while bearing (7.1.83) in mind), and (7.1.28) in Proposition 7.1.5 used with
F := B0,0 f (its present applicability being ensured by (7.5.35), item (v) in Proposi-
tion 7.5.1, together with items (ii) and (vii) in Theorem 7.3.5 used with α = β = 0).
Collectively, (7.5.41), (7.5.42), and (7.1.11)-(7.1.12) then prove that B0,0 in (7.5.37)
866 7 Applications to Analysis in Several Complex Variables

is indeed a well-defined, linear, and bounded operator. Finally, the claims about the
operators (7.5.38)-(7.5.39) are implicit in what we have proved already. 

We momentarily digress to include the characterization of balls in terms of

(Hermitian) self-adjointness of the associated Bochner-Martinelli singular integral
operator in the class of bounded UR domains, given in the proposition below. This
is a significant improvement over the main result in [5] where the more restrictive
class of bounded 𝒞1 domains has been considered.

Proposition 7.5.3 Let Ω ⊆ Cn , where n ≥ 2, be a bounded UR domain and set

σ := H 2n−1 ∂Ω. Then the principal-value Bochner-Martinelli singular integral
operator B0,0 (cf. (7.5.26)) is Hermitian self-adjoint on L 2 (∂Ω, σ) if and only if Ω is
a ball.

Proof In one direction, if the operator B0,0 : L 2 (∂Ω, σ) → L 2 (∂Ω, σ) is Hermitian

self-adjoint, then from (7.5.40) we conclude (upon noting that both K and the
operators R(2j−1)(2j) with 1 ≤ j ≤ n have real-valued integral kernels) that the
principal-value harmonic double layer potential operator K associated with the set
Ω ⊆ Cn ≡ R2n as in [70, (2.5.203)] is symmetric on L 2 (∂Ω, σ). Granted this, we
may then invoke [38, Theorem 4.23] (bearing in mind [68, (5.10.52)]) to conclude
that Ω is a ball. As far as the converse direction is concerned, if Ω is a ball we then
   
trivially have νC (ζ), ζ − z = νC (z), z − ζ for each z, ζ ∈ ∂Ω (where ν denotes
C C
the outward unit normal to Ω). On account of (7.5.26), this goes to show that the
operator B0,0 : L 2 (∂Ω, σ) → L 2 (∂Ω, σ) is Hermitian self-adjoint. 

Returning to the main topic of conversation, we shall prove that the square of
the principal-value version of the Bochner-Martinelli integral operator is, up to
normalization, the identity when acting on CR-functions in the context described in
the next proposition.

Proposition 7.5.4 Suppose Ω ⊆ R2n ≡ Cn is an open set with the property that ∂Ω
is a UR set and such that
 
H 2n−1 ∂nta Ω \ ∂∗ Ω = 0. (7.5.43)

Abbreviate σ := H 2n−1 ∂Ω and fix an exponent p ∈ (1, ∞). Then for each CR-
function f ∈ L p (∂∗ Ω, σ) one has
2
B0,0 f = 1
4 f at σ-a.e. point on ∂∗ Ω. (7.5.44)

Proof Fix some κ > 0 and, having picked a CR-function f ∈ L p (∂∗ Ω, σ), introduce
F := B0,0 f in Ω. Then from items (ii), (vii) in Theorem 7.3.5, and (7.5.15), we
conclude that
κ−n.t.
F ∂Ω = ( 12 I + B0,0 ) f at σ-a.e. point on ∂∗ Ω,
(7.5.45)
Nκ F ∈ L p (∂Ω, σ), and F is holomorphic in Ω.
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 867

Moreover,

if Ω is an exterior domain then F(z) = O(|z| 1−2n ) as |z| → ∞. (7.5.46)

In turn, (7.5.45)-(7.5.46) and (7.5.43) ensure that the integral representation formula
(7.4.45) holds for the current function F, i.e.,
 κ−n.t. 
F = B0,0 F ∂Ω in Ω. (7.5.47)

Going nontangentially to the boundary in (7.5.47) then yield

 
( 12 I + B0,0 ) f = ( 12 I + B0,0 ) ( 12 I + B0,0 ) f at σ-a.e. point on ∂∗ Ω, (7.5.48)

on account of (7.5.45) and (7.3.82). With this in hand, formula (7.5.44) follows after
some simple algebra. 
We are now ready to discuss the main result in this section. To place matters in
a broader perspective we first make a series of remarks pertaining to the history of
extension phenomena in several complex variables. Virtually all results of this nature
originate in the pioneering work of Friedrich Hartogs who, in 1906, by relying on
Cauchy’s integral formula for functions of several complex variables proved1 the
following:
[Hartogs’ Kugelsatz; [34]] If n ≥ 2 and Ω is an open bounded subset
of Cn with connected boundary, then any holomorphic function f in a (7.5.49)
neighborhood U of ∂Ω extends to a holomorphic function in the set Ω.

In the limiting case when the neighborhood U shrinks to ∂Ω, the standard holomor-
phicity assumption for f should be replaced by the demand that f is a CR-function
on ∂Ω. The latter condition is meaningful under suitable assumptions on f and ∂Ω,
a scenario in which we shall refer to this question as the Extension Problem2. The
first solution to the Extension Problem was given in 1931 by Francesco Severi in
the case when both ∂Ω and f are real-analytic. Severi proved such a theorem in
[93] for the case n = 2 for tangential analytic functions in the sense of Wirtinger
(cf. the discussion in (7.2.77)) via a technique involving passing from real variables
to complex variables at the level of power series which naturally extends to higher
dimensions. He first established a local version which then was globalized using the
classical Hartogs extension theorem (cf. (7.5.49)). Seemingly unaware of Severi’s
earlier work, in 1936 Helmuth Kneser solved the Extension Problem in [45] for
bounded strictly pseudoconvex domains of class 𝒞2 in C2 . In this paper Kneser also
observed that Wirtinger’s tangential analyticity condition d f ∧ dz1 ∧ · · · ∧ dzn = 0
on ∂Ω (which is meaningful only in the category of continuously differentiable
functions) may be recast as a Morera type condition on ∂Ω, which permitted him to
1 It was actually not until 2007 that Hartogs’ original analytic disk method was shown to work in
full generality by J. Merker and E. Porten in [64]
2 A positive resolution of the Extension Problem in any reasonable setting yields (7.5.49) as a
corollary
868 7 Applications to Analysis in Several Complex Variables

holomorphically
∫ extend functions f which are merely continuous on ∂Ω and satisfy
bD
f dz 1 ∧ dz 2 = 0 for every 𝒞1 subdomain D of the manifold ∂Ω (with bD denot-
ing the topological boundary of D relative to this ambient). Apparently unaware of
Kneser’s work, in 1956 Hand Lewy reproved the extension theorem for continuously
differentiable CR-functions in the setting of bounded strictly pseudoconvex domains
of class 𝒞2 in C2 in [57]3.
Using potential theoretic methods (specifically, the solvability of the Dirichlet
Problem), in 1957 Gaetano Fichera succeeded in extending Severi’s 1931 glob-
al extension theorem without demanding real analyticity for the data and without
assuming pseudoconvexity for the underlying domain. Concretely, in [25] Fichera
solved the Extension Problem for bounded domains of class 𝒞1+ε , with ε > 0 arbi-
trary. Inspired by this work, in 1961 Enzo Martinelli re-visited his 1942 proof of the
classical Hartogs theorem (based on the Bochner-Martinelli integral formula) and, in
[62], adapted it to produce a conceptually simple proof of the global extension theo-
rem of Severi-Fichera in the class of bounded 𝒞1 domains Ω and for CR-functions
f ∈ 𝒞1 (∂Ω). In the late 1960’s B. Weinstock dealt with the Extension Problem for
continuous CR-functions defined on boundaries of bounded 𝒞∞ domains in [104],
[105].
For an informative account on the early history of this subject, see [89], [90].
As regards more recent work, in [59, Theorem 8.20, p. 45] a version of the Exten-
sion Problem is presented4 which involves bounded 𝒞1 domains and continuous
CR-functions. Also, the 1976 survey [35] of G. Henkin and E. Chirka contains a
discussion of the Extension Problem for integrable functions on compact Lyapunov
surfaces (which essentially are topological boundaries of bounded domains of class
𝒞1+ε for some ε > 0). This continues to be an active area of research and the
interested reader is referred to [55] for a review of related work up to the early
1990’s.
In this volume we contribute to this line of research, aimed at identifying the most
general geometric/analytic setting in which the Extension Problem may be solved, by
considering domains so rough that their topological boundaries are lacking any type
of manifold structure, and also consider CR-functions which are not differentiable in
a traditional sense. Such a setting gives rise to a number of significant challenges. For
example, one has to find a suitable notion of tangential Cauchy-Riemann equations
on a rough “surface" and in Definition 7.1.4 we have done just that, assuming that
the surface in question is the topological boundary of a set which is merely of locally
finite perimeter. As regards our main results on this topic, in Theorem 7.5.5 we first
discuss the solution to the Extension Problem for Hölder CR-functions in Ahlfors
regular domains. Subsequently, in Theorem 7.6.1 we solve the Extension Problem
in the class of uniformly rectifiable (UR) domains for CR-functions in Lebesgue,

3 Motivated by this topic (specifically, in an effort to show that the Dolbeault complex for the
boundary d-bar operator is almost never exact), one year later Lewy published his celebrated
example of a smooth first-order linear partial differential equation with no solution in [58] (which
shows that the Cauchy-Kovalevskaya theorem does not have a natural analog in the smooth category)
4 this is referred to by I. Lieb and J. Michel as “a deep generalization of the Kugelsatz"
7.5 Extending Hölder CR-Functions from Boundaries of Ahlfors Regular Domains 869

John-Nirenberg, and Sarason spaces (L p /BMO/VMO). In both cases, our earlier

results on the nature of the Bochner-Martinelli integral operator play a crucial role.
Theorem 7.5.5 Having fixed n ∈ N with n ≥ 2, assume Ω ⊆ R2n ≡ Cn is a bounded
Ahlfors regular domain with the property that R2n \ Ω is a connected set, and
abbreviate σ := H 2n−1 ∂Ω.
(1) Existence/Uniqueness: Given any α ∈ (0, 1), there exists a constant C ∈ (0, ∞)
with the property that for each CR-function f ∈ 𝒞α (∂Ω) one can find a unique
holomorphic function F in Ω which extends f nontangentially, in the sense that
for each κ > 0 one has
κ−n.t.
F ∂Ω = f at σ-a.e. point on ∂Ω, (7.5.50)

and such that

  
sup |F(z)| + sup dist(z, ∂Ω)1−α (∇F)(z) ≤ C f 𝒞α (∂Ω) . (7.5.51)
z ∈Ω z ∈Ω

(2) Strong Uniqueness: Given any f ∈ L 1 (∂Ω, σ) and any κ > 0, there could be at
most one holomorphic function F in Ω with Nκ F ∈ L 1 (∂Ω, σ) and such that
κ−n.t.
F ∂Ω = f at σ-a.e. point on ∂Ω. In particular, any CR-function f ∈ 𝒞α (∂Ω)
with α ∈ (0, 1) has a unique bounded holomorphic nontangential extension to
Ω (in the sense of (7.5.50)).
(3) Further Regularity: If Ω is also a uniform domain, then the holomorphic non-
tangential extension F of the
 CR-function
 f ∈ 𝒞α (∂Ω) with α ∈ (0, 1) from part
(1) actually belongs to 𝒞α Ω and satisfies

F 𝒞α (Ω) ≤C f 𝒞α (∂Ω) . (7.5.52)


In such a scenario one has f = F ∂Ω , the ordinary restriction of F to ∂Ω.

In the context of the above theorem, recall from (7.1.21) that f ∈ L 1 (∂Ω, σ)
is a CR-function provided ∂τCj k f = 0 on ∂Ω (in the sense of Definition 7.1.1) for
all j, k ∈ {1, . . . , n}. From Proposition 7.1.5 we know that being a CR-function is
a necessary condition for the existence of a holomorphic nontangential extension
(with control of the nontangential maximal function).
Proof of Theorem 7.5.5 Define Ω+ := Ω and Ω− := R2n \ Ω. Then, having fixed
some arbitrary CR-function f ∈ 𝒞α (∂Ω) with α ∈ (0, 1), introduce
  
F ± := B0,0 f  . (7.5.53)
Ω±

Note that Ω± ⊆ Cn \ ∂Ω, so (7.5.14)-(7.5.15) imply that F ± are holomorphic in

Ω± . Moreover, since n > 1 and Ω− is assumed to be connected, from (7.5.16) we
conclude that actually F − vanishes identically in Ω− . In concert with (7.5.34), for
each κ > 0 this permits us to write
870 7 Applications to Analysis in Several Complex Variables

f = ( 12 I + B0,0 ) f − (− 12 I + B0,0 ) f

 +  κ−n.t.  −  κ−n.t.
= B0,0 f  − B0,0 f 
∂Ω ∂Ω
κ−n.t.

= F+ at σ-a.e. point on ∂Ω. (7.5.54)
∂Ω

Hence, F := F + = B0,0 + f is a holomorphic nontangential extension of the given f

which also satisfies (7.5.51), thanks to (7.5.5) (also bearing in mind the inequality
dist(·, ∂Ω) ≤ dist(·, ∂∗ Ω), which holds since ∂∗ Ω ⊆ ∂Ω).
That for any given function in L 1 (∂Ω, σ), and any κ > 0, there could be at most
κ−n.t.
one holomorphic function F in Ω with Nκ F ∈ L 1 (∂Ω, σ) and such that F ∂Ω
matches the given function at σ-a.e. point on ∂Ω, is then clear from the reproducing
formula the last part of Corollary 7.4.2 (cf. (7.4.45)). In particular, this establishes
uniqueness for the extension F + of f constructed above.
Lastly, if Ω is also a uniform domain, then from (7.5.6) we conclude that F +
belongs to 𝒞α Ω and F + 𝒞α (Ω) ≤ C f 𝒞α (∂Ω) for some constant C ∈ (0, ∞)
independent of f . 
As a byproduct of our solution to the Extension Problem from Theorem 7.5.5 we
have the following result, to the effect that the class of Hölder CR-functions is stable
under pointwise multiplication.

Corollary 7.5.6 Assume Ω ⊆ R2n ≡ Cn , where n ∈ N with n ≥ 2, is a bounded

Ahlfors regular domain with the property that R2n \ Ω is a connected set. Then

𝒜α := f ∈ 𝒞α (∂Ω) : f is a CR-function
(7.5.55)
is a sub-algebra of 𝒞α (∂Ω) for each α ∈ (0, 1).

Proof Fix α ∈ (0, 1). Clearly, 𝒜α is a linear subspace of 𝒞α (∂Ω). To show that this is
actually an algebra, we need to check that it is stable under pointwise multiplication.
With this goal in mind, pick two arbitrary CR-functions f , g ∈ 𝒞α (∂Ω) and use
Theorem 7.5.5 to extend them holomorphically to Ω. Specifically, there exist two
bounded holomorphic functions F, G in Ω with the property that for each κ > 0 we
have κ−n.t. κ−n.t.
 
F = f and G = g at σ-a.e. point on ∂Ω, (7.5.56)
∂Ω ∂Ω

where σ := H 2n−1 ∂Ω. Then their product, H := FG, is a bounded holomorphic

function in Ω which, for each κ > 0 satisfies
κ−n.t.

H = f g at σ-a.e. point on ∂Ω. (7.5.57)
∂Ω

Granted this, the very last claim in the statement of Proposition 7.1.5 guarantees
that f g is a CR-function. Since f g also belongs to 𝒞α (∂Ω), the desired conclusion
follows. 
7.6 Extending L p /BMO/VMO/Morrey Functions in Uniformly Rectifiable Domains 871

7.6 The Extension Problem for L p /BMO/VMO/Morrey

Functions on Boundaries of Uniformly Rectifiable Domains

In this section, the principal goal is to present the solution to the Extension Problem
formulated for L p /BMO/VMO/Morrey functions in UR domains.

Theorem 7.6.1 Fix n ∈ N with n ≥ 2 and assume that Ω ⊆ R2n ≡ Cn is a bounded

UR domain with the property that the set R2n \ Ω is connected. Also, abbreviate
σ := H 2n−1 ∂Ω.
(1) Existence: Given any p ∈ [1, ∞), there exists a constant C ∈ (0, ∞) with the
property that for each CR-function f ∈ L p (∂Ω, σ) one can find a holomorphic
function F in Ω which extends f nontangentially, in the sense that for each κ > 0
one has κ−n.t.
F ∂Ω = f at σ-a.e. point on ∂Ω, (7.6.1)
and such that

Nκ F L p (∂Ω,σ)
≤C f L p (∂Ω,σ) if 1 < p < ∞, (7.6.2)

Nκ F L 1,∞ (∂Ω,σ)
≤C f L 1 (∂Ω,σ) if p = 1. (7.6.3)

Also, corresponding to p = ∞, for each CR-function f ∈ BMO(∂Ω, σ) there

exists a holomorphic function F in Ω which extends f nontangentially (in
the sense of (7.6.1)) and which, for each q ∈ (1, ∞) and with some constant
C ∈ (0, ∞) independent of f , satisfies the following Carleson measure estimate:
∫
 q
sup  1  ∇F  dist(·, ∂Ω)q−1 dL 2n ≤ C f q .
BMO(∂Ω,σ)
z ∈∂Ω, r >0 σ ∂Ω∩B(z,r) B(z,r)∩Ω
(7.6.4)

(2) Strong Uniqueness: For each holomorphic function F in Ω with the property that
κ−n.t.
Nκ F ∈ L p (∂Ω, σ), for some κ > 0, the nontangential boundary trace F ∂Ω
exists at σ-a.e. point on ∂Ω and satisfies
κ−n.t.
F ∂Ω L 1 (∂Ω,σ)
≈ Nκ F L 1 (∂Ω,σ)
(7.6.5)

where the (implicit) proportionality constants are independent of F. As a con-

sequence, given any f ∈ L 1 (∂Ω, σ) and any κ > 0, there could be at most one
κ−n.t.
holomorphic function F in Ω with Nκ F ∈ L 1 (∂Ω, σ) and such that F ∂Ω = f
at σ-a.e. point on ∂Ω. In particular, given any CR-function f ∈ L p (∂Ω, σ)
with p ∈ (1, ∞) the holomorphic nontangential extension F of f to Ω con-
structed in part (1) is unique (in the class of holomorphic functions in Ω whose
nontangential maximal functions are in L p (∂Ω, σ)).
(3) Further Regularity: For each p, q ∈ [1, ∞) there exists a constant C ∈ (0, ∞) with
the property that if f is a CR-function belonging to the off-diagonal Sobolev
872 7 Applications to Analysis in Several Complex Variables
p,q
spaces L1 (∂Ω, σ) then, in addition to (7.6.2)-(7.6.3), its holomorphic nontan-
gential extension F to Ω constructed in part (1) also satisfies

Nκ (∇F) L q (∂Ω,σ)
≤ C ∇tan f L q (∂Ω,σ) if 1 < q < ∞, (7.6.6)

Nκ (∇F) L 1,∞ (∂Ω,σ)

≤ C ∇tan f L 1 (∂Ω,σ) if q = 1. (7.6.7)

Also, for each p ∈ (1, ∞) there exists a constant C = C(Ω, p, n) ∈ (0, ∞) such that
for any CR-function f ∈ L p (∂Ω, σ), the holomorphic nontangential extension
F of f to Ω constructed in part (1) has the additional property that
∫  1/p
|∇F | p dist(·, ∂Ω) p−1 dL 2n ≤ C f L p (∂Ω,σ) . (7.6.8)
Ω

In addition, for any CR-function f ∈ VMO(∂Ω, σ) the holomorphic nontan-

gential extension F of f to Ω constructed in part (1) has the property that for
each p ∈ (1, ∞) the measure |∇F | p dist(·, ∂Ω) p−1 dL 2n is a vanishing Carleson
measure in Ω, in the sense that
∫
1
lim+ sup   |∇F | p dist(·, ∂Ω) p−1 dL 2n = 0.
R→0 z ∈∂Ω, r ∈(0,R) σ B(z, r) ∩ ∂Ω B(z,r)∩Ω
(7.6.9)

Moreover, given any integrability exponent p ∈ (1, ∞) along with a parameter

λ ∈ (0, 2n − 1), there exists a constant C = C(Ω, n, p, λ) ∈ (0, ∞) with the
property that for any CR-function f belonging to the Morrey space M p,λ (∂Ω, σ)
the holomorphic nontangential extension F of f to Ω devised in part (1) satisfies
the following fractional Carleson measure estimate:
 ∫  p1
sup r −λ |∇F | p dist(·, ∂Ω) p−1 dL 2n (7.6.10)
z ∈∂Ω and B(z,r)∩Ω
0<r <2 diam(∂Ω)
≤C f M p, λ (∂Ω,σ) .

Finally, when Ω is actually a bounded NTA domain

 n with  Ahlfors regular bound-
ary, then for each p ∈ (1, ∞) and q ∈ n+1/p , ∞ there exists a constant
C ∈ (0, ∞) with the property that for any CR-function f ∈ L p (∂Ω, σ) the holo-
morphic nontangential extension F of f to Ω constructed in part (1) belongs to
p,q p, p
F1/p (Ω) ∩ B1/p (Ω) and satisfies

max F F1/p (Ω), F ≤C f L p (∂Ω,σ) . (7.6.11)
p, q p, p
B1/p (Ω)

In particular, corresponding to p = q = 2,

F H 1/2 (Ω) ≤C f L 2 (∂Ω,σ) . (7.6.12)

7.6 Extending L p /BMO/VMO/Morrey Functions in Uniformly Rectifiable Domains 873

Recall from (7.1.21) that, in the context of the above theorem, f ∈ L 1 (∂Ω, σ)
is a CR-function whenever ∂τCj k f = 0 on ∂Ω (in the sense of Definition 7.1.1) for
all j, k ∈ {1, . . . , n}. In this vein, it is worth pointing out that, as is apparent from
Proposition 7.1.5, being a CR-function is a necessary condition for the existence of
a holomorphic nontangential extension with an integrable nontangential maximal
function.
We also wish to note that, as far as extending arbitrary CR-functions in a nontan-
gential fashion to holomorphic functions inside the given domain, assuming n > 1
is a necessary condition. Indeed, bearing (7.1.23) in mind, it is easy to see that
such an extension may not exist even in the case of the unit disk in the plane. Also,
simple counterexamples in an annulus (taking f to be two different constants on the
two spheres making up the boundary) show that the connectivity hypothesis on the
complement is a necessary condition as well.
Proof of Theorem 7.6.1 Fix some arbitrary CR-function f ∈ L p (∂Ω, σ) for some
exponent p ∈ [1, ∞). Having defined Ω+ := Ω and Ω− := R2n \ Ω, consider
  
F ± := B0,0 f  . (7.6.13)
Ω±

Since Ω± ⊆ Cn \∂Ω, it follows from (7.5.14)-(7.5.15) imply that F ± are holomorphic

in Ω± . Given that n > 1 and Ω− is connected, (7.5.16) actually forces the function F −
to vanish identically in the set Ω− . Bearing this in mind, for each κ > 0 we may then
invoke (7.3.85) (with α = β = 0, in which case we have the natural identification
p
Ltan,C (∂Ω, σ) ⊗ Λ0,0 ≡ L p (∂Ω, σ); cf. (7.1.64) and (7.1.79) in this regard) to write
1   
f = 2I + B0,0 f − − 12 I + B0,0 f

 +  κ−n.t.  −  κ−n.t.
= B0,0 f  − B0,0 f 
∂Ω ∂Ω
κ−n.t.

= F+ at σ-a.e. point on ∂Ω. (7.6.14)
∂Ω

Consequently, F := F + = B0,0 + f is a holomorphic nontangential extension of the f

which, by virtue of (7.3.72)-(7.3.73) (presently used with α = β = 0), also satisfies

(7.6.2)-(7.6.3). The argument in the case when f ∈ BMO(∂Ω, σ) is similar, relying
on (7.5.17) in place of nontangential maximal function estimates.
Next, the claims in item (2) up to, and including, (7.6.5) are consequences of [70,
Theorem 3.1.6] (cf. [70, (3.1.103)], [70, (3.1.112)]), and the discussion in Comment 6
(following the statement of said theorem). Parenthetically, we wish to note that we
can also use the reproducing formula (7.4.45) to conclude that there could be at most
one nontangential extension of a given f ∈ L 1 (∂Ω, σ) to a holomorphic function F
in Ω satisfying Nκ F ∈ L 1 (∂Ω, σ) for some κ > 0.
With the exception of (7.6.10), the additional regularity properties of the extension
specified in the last part of the statement are direct consequences of the mapping
properties of the operator B0,0 discussed in (7.5.8)-(7.5.9), (7.3.77), (7.3.80), and
874 7 Applications to Analysis in Several Complex Variables

item (vii) in Proposition 7.5.1 (also keeping in mind the estimates established in [69,
Proposition 11.4.2]).
Finally, that the fractional Carleson measure estimate recorded in (7.6.10) holds
whenever the function f belonging to the Morrey space M p,λ (∂Ω, σ) is a conse-
quence of (3.3.69), bearing in mind that F = B0,0 + f and the fact that the operator
+ is a particular case of a double layer (as discussed in Example 1.4.16).
B0,0 

In turn, the solution to the version of the Extension Problem from Theorem 7.6.1
allows us to establish the following result pertaining to the stability of the class of
CR-functions in Lebesgue spaces under pointwise multiplication.

Corollary 7.6.2 Assume Ω ⊆ R2n ≡ Cn , where n ∈ N with n ≥ 2, is a bounded UR

domain with the property that R2n \Ω is a connected set. Abbreviate σ := H 2n−1 ∂Ω
and fix p, q ∈ (1, ∞) with the property that 1/p + 1/q = 1/r for some r ≥ 1. In this
setting, consider two CR-functions, f ∈ L p (∂Ω, σ) and g ∈ L q (∂Ω, σ). Then f g is
a CR-function in L r (∂Ω, σ).

Proof Clearly, f g ∈ L r (∂Ω, σ) (in particular f g ∈ L 1 (∂Ω, σ) since ∂Ω has finite

measure) so there remains to show that f g is a CR-function. Fix κ > 0. According
to Theorem 7.6.1 there exist two holomorphic functions F, G in Ω satisfying

Nκ F ∈ L p (∂Ω, σ), Nκ G ∈ L q (∂Ω, σ),

κ−n.t. κ−n.t. (7.6.15)
 
F = f and G = g at σ-a.e. point on ∂Ω.
∂Ω ∂Ω

Then the product

H := FG is a holomorphic function in Ω with the property that

κ−n.t. (7.6.16)

Nκ H ∈ L r (∂Ω, σ) and H  = f g at σ-a.e. point on ∂Ω.
∂Ω

Having established (7.6.16), the very last claim in Proposition 7.1.5 ensures that f g
is indeed a CR-function. 

Together, the Fatou-type result from [70, Theorem 3.1.6] (cf. [70, (3.1.144)] in
particular) and the solution to the version of the Extension Problem from Theo-
rem 7.6.1 directly impact the theory of Hardy spaces for holomorphic functions of
several complex variables. A concrete example is offered by the following theorem.

Theorem 7.6.3 Let Ω ⊆ R2n ≡ Cn be a UR domain and abbreviate σ := H 2n−1

 ∂Ω.
Having fixed a background aperture parameter κ > 0, for each p ∈ 2n−1
2n , ∞ define
the holomorphic Hardy space in Ω as

H p (Ω) := F : Ω → C : F holomorphic, Nκ F ∈ L p (∂Ω, σ),

and lim F(z) = 0 if Ω is an exterior domain , (7.6.17)

|z |→∞
7.6 Extending L p /BMO/VMO/Morrey Functions in Uniformly Rectifiable Domains 875

and equip it with the quasi-norm H p (Ω)  F → F H p (Ω) := Nκ F L p (∂Ω,σ) .

Then the following statements are true.
(i) The space H p (Ω) is quasi-Banach, and the nontangential trace mapping
κ−n.t.
H p (Ω)  F −→ F ∂Ω ∈ L p (∂Ω, σ) (7.6.18)

is well defined, linear, and bounded. Moreover, if p ∈ (1, ∞) then the image of
(7.6.18) is contained in f ∈ L p (∂Ω, σ) : f is a CR-function .
(ii) Whenever p ∈ (1, ∞), the Bochner-Martinelli integral operator

B0,0 : f ∈ L p (∂Ω, σ) : f is a CR-function −→ H p (Ω) (7.6.19)

is well defined, linear, bounded, and surjective, with the nontangential trace map
(7.6.18) serving as a right-inverse.
(iii) If p ∈ (1, ∞), the set Ω is bounded, n > 1, and Cn \ Ω is connected, then the
operator B0,0 is actually an isomorphism in the context of (7.6.19).
p
Proof By design, H p (Ω) is a subspace of Nκ (Ω; σ), defined as in (A.0.168). Keep-
ing this in mind, we may then rely on [68, Proposition 8.3.5] to conclude that H p (Ω)
is indeed a quasi-Banach space. Also, from [70, (3.1.144)] we know that for each
κ−n.t.
F ∈ H p (Ω) the nontangential trace F ∂Ω exists at σ-a.e. point on ∂Ω. Finally, the
very last claim in item (i) is a consequence of what we have proved so far and the
last conclusion in the statement of Proposition 7.1.5.
Moving on, that the operator (7.6.19) is well defined, linear, bounded when
p ∈ (1, ∞) is seen from item (vi) of Proposition 7.5.1 together with items (ii)
and (vii) of Theorem 7.3.5. In addition, the very last claim in the current item (i)
together with the very last claim in Corollary 7.4.2 prove that the nontangential trace
κ−n.t.
H p (Ω)  F → F ∂Ω is indeed as a right-inverse for the operator (7.6.19).
Finally, assume p ∈ (1, ∞), the set Ω is bounded, n > 1, and Cn \ Ω is connected.
In this scenario, if f ∈ L p (∂Ω, σ) is a CR-function with the property that B0,0 f = 0
in Ω, then (7.6.14) simply reduces to f = 0. This proves that the operator (7.6.19) is
injective, hence ultimately an isomorphism, in this case. 
Our next three theorems are concerned with characterizing the quality of be-
ing a CR-functions in terms of the action of the principal-value version of the
Bochner-Martinelli integral operator. The emerging philosophy is that, under favor-
able geometric and analytic assumptions, the space of CR-functions is precisely the
null-space of the operator − 12 I + B0,0 .

Theorem 7.6.4 Fix n ∈ N with n ≥ 2 and suppose Ω ⊆ R2n ≡ Cn is a bounded UR

domain. Abbreviate σ := H 2n−1 ∂Ω, select α ∈ {0, 1, . . . , n}, and pick a differential
form
f ∈ L (2n−1)/(n−1) (∂Ω, σ) ⊗ Λα,0 with the property that
(7.6.20)
∂¯τ f ∈ L (2n−1)/n (∂Ω, σ) ⊗ Λα,2
876 7 Applications to Analysis in Several Complex Variables

(in particular, the conditions in (7.6.20) are satisfied if f is an arbitrary differential

p, ∂¯
form belonging to the space Lα,0 τ (∂Ω, σ) with (2n − 1)/(n − 1) ≤ p ≤ ∞). Then

Bα,0 f = 1
2 f at σ-a.e. point on ∂Ω =⇒ f is a CR-form. (7.6.21)

Proof Given a form f as in (7.6.20) which also satisfies Bα,0 f = 1

2 f at σ-a.e. point
on ∂Ω, introduce
u := Bα,0 f ∈ 𝒞∞ (Ω) ⊗ Λα,0 . (7.6.22)
Thanks to (7.6.20), items (ii)-(iii) in Theorem 7.3.5, and simple degree considerations
(based on (7.2.11)) we then have

Nκ u ∈ L (2n−1)/(n−1) (∂Ω, σ), ¯ ∈ L (2n−1)/n (∂Ω, σ),

Nκ (∂u)
(7.6.23)
and ϑu = 0 in Ω.

Moreover, from (7.3.82), (7.1.83), and the fact that we are assuming that Bα,0 f = 1
2 f
at σ-a.e. point on ∂Ω we obtain
κ−n.t.  
u∂Ω = 12 I + Bα,0 f = f at σ-a.e. point on ∂Ω. (7.6.24)

In particular, (7.6.24) and (7.6.22) entail

 κ−n.t. 
u = Bα,0 u∂Ω in Ω. (7.6.25)

Collectively, (7.6.23)-(7.6.25) and Theorem 7.4.6 then imply that u is ∂-closed

¯ in Ω.
Having established this, from Proposition 7.2.9 and (7.6.24) we may conclude that
κ−n.t.
f = u∂Ω is indeed a CR-form. 

To state our next result, recall the off-diagonal (partial) Sobolev spaces defined
in (7.5.27).

Theorem 7.6.5 Fix n ∈ N with n ≥ 2 and assume that Ω ⊆ R2n ≡ Cn is a bounded

UR domain such that R2n \Ω is connected. Abbreviate σ := H 2n−1 ∂Ω and consider
a complex-valued function
1, p,q
f ∈ LC (∂Ω, σ) with 2n−1
n−1 ≤ p ≤ ∞ and 2n−1
n ≤q≤∞ (7.6.26)

(in particular, this is the case if f is an arbitrary function belonging to the space
p,1
LC (∂Ω, σ) with (2n − 1)/(n − 1) ≤ p ≤ ∞). Then

f is a CR-function ⇐⇒ B0,0 f = 1
2 f at σ-a.e. point on ∂Ω. (7.6.27)

It is relevant to note that, under the geometric assumptions on the set Ω made
in first part of Theorem 7.6.5, we may rephrase (7.6.26)-(7.6.27)
  simply
 2n−1by saying

that, for any given integrability exponents p ∈ 2n−1 n−1 , ∞ and q ∈ n , ∞ , the
null-space of the operator − 12 I + B0,0 acting from the off-diagonal (partial) Sobolev
7.6 Extending L p /BMO/VMO/Morrey Functions in Uniformly Rectifiable Domains 877
1, p,q
space LC (∂Ω, σ) into itself consists precisely of all CR-functions belonging to
L p (∂Ω, σ).
Proof of Theorem 7.6.5 As regards the right-pointing implication, if f is a CR-
function then from Theorem 7.6.1 and its proof we know that f extends nontan-
gentially to a holomorphic function in Ω which is actually given by F := B0,0 f .
In concert with (7.3.82) (used with α = β = 0), this further implies that for each
κ−n.t.  
κ > 0 we have f = F ∂Ω = 12 I + B0,0 f at σ-a.e. point on ∂Ω, which ultimate-
ly goes to show that B0,0 f = 12 f on ∂Ω. Lastly, the left-pointing implication in
(7.6.29) is a direct consequence of Theorem 7.6.4 (presently invoked with α = 0),
Proposition 7.2.4, and Proposition 7.2.5. 
We conclude this section by discussing the following characterization of CR-
functions in terms of the principal-value Bochner-Martinelli singular integral oper-
ator.
Theorem 7.6.6 Fix n ∈ N with n ≥ 2 and assume that Ω ⊆ R2n ≡ Cn is a bounded
NTA domain with an Ahlfors regular boundary and with the property that the set
R2n \ Ω is connected. Abbreviate σ := H 2n−1 ∂Ω and recall the critical exponent
pΩ associated with Ω as in [70, (5.7.46)]. Finally, pick a complex-valued function
1, p,q  
f ∈ LC (∂Ω, σ) with p ∈ 2n−1 n−1 , ∞
  (7.6.28)
satisfying p > pΩ, and q ∈ 2n−1 n ,∞

in particular, this is the case if f is an arbitrary function belonging to the space
p,1 
LC (∂Ω, σ) with max{(2n − 1)/(n − 1), pΩ } < p ≤ ∞ . Then

B0,0 f = 1
2 f on ∂Ω ⇐⇒ f is a CR-function ⇐⇒ B0,0
2
f = 1
4 f on ∂Ω. (7.6.29)

In the case when Ω is a bounded Lipschitz domain in R2n ≡ Cn with a connected

complement, then all hypotheses of geometric nature in Theorem 7.6.6 are satisfied
and, in addition, pΩ ∈ [1, 2) (cf. [18]). In particular, the demand in (7.6.28) that
p > pΩ becomes superfluous in this situation.
Proof of Theorem 7.6.6 For starters, we wish to note that the geometric assumptions
imply that Ω is a UR domain (cf. [68, Proposition 5.10.4] and [68, (5.2.4)]). Granted
this, the first equivalence in (7.6.29) is provided by Theorem 7.6.5. Also, the right-
pointing implication in the second equivalence in (7.6.29) is seen directly from
Proposition 7.5.4 (keeping in mind that condition (7.5.43) is presently satisfied
thanks to [68, (5.2.4)] and [68, (8.8.52)]).
As regards the left-pointing implication in the second equivalence in (7.6.29), we
start by defining the complex-valued function

F := B0,0 f in Ω. (7.6.30)

Observe that, thanks to (7.3.34), (7.3.82), item (v) in Proposition 7.5.1, and (7.3.72),
for each fixed κ > 0 we have
878 7 Applications to Analysis in Several Complex Variables

F ∈ 𝒞∞ (Ω), Nκ F ∈ L p (∂Ω, σ), Nκ (∂F) ¯ ∈ L (2n−1)/n (∂Ω, σ),

κ−n.t. κ−n.t. (7.6.31)
both F  ∂Ω
¯ 
and ∂F ∂Ω
exist σ-a.e. on ∂Ω, and ΔF = 0 in Ω,

where Δ := ∂12 + · · · + ∂2n

2 is the Laplacian in R2n . Moreover, (7.3.82), (7.1.83), and

(7.3.69) imply that actually

κ−n.t.  
F ∂Ω = 12 I + B0,0 f ∈ L p (∂Ω, σ). (7.6.32)

2 f = 1 f on ∂Ω, another
In particular, given that we are currently assuming that B0,0 4
application of (7.3.82) gives (once again bearing in mind (7.1.83))
  κ−n.t.   κ−n.t.  1  
B0,0 F ∂Ω  = 2 I + B0,0 12 I + B0,0 f
∂Ω
1 2   
= 4I + B0,0 + B0,0 f = 12 I + B0,0 f . (7.6.33)

At this stage, we make the claim that

 κ−n.t. 
F = B0,0 F ∂Ω in Ω. (7.6.34)

To prove this, observe that by (7.6.31)-(7.6.32) on the one hand, and by (7.3.34),
(7.3.72), (7.6.33) on the other hand, both sides of (7.6.34)
 solve the L p Dirichlet
Problem for the Laplacian in Ω with boundary datum 12 I + B0,0 f ∈ L p (∂Ω, σ),
i.e.,
⎧ ∞
⎪ u ∈ 𝒞 (Ω),
⎪
⎪
⎪
⎪
⎨ Δu = 0 in Ω,
⎪
(7.6.35)
⎪
⎪ Nκ u ∈ L p (∂Ω, σ),
⎪ κ−n.t.
⎪
⎪
⎪ u 1 
⎩ ∂Ω = 2 I + B0,0 f σ-a.e. on ∂Ω.
Granted this, from the assumption on p and the uniqueness portion of [70, Theo-
rem 5.7.7] we conclude that (7.6.34) holds, as claimed.
Together, (7.6.31), (7.6.34), and the very last claim in Theorem 7.4.6 then imply
that F is holomorphic in Ω. Having established this, from Proposition 7.1.5 and
κ−n.t.
(7.6.32) we may conclude that f = F ∂Ω is indeed a CR-function. 

7.7 The ∂ Operator and the Dolbeault Complex on Uniformly

Rectifiable Sets

In this section we introduce a suitable version of the ∂¯ operator on the boundaries of

UR domains and explain how this may be used to set up the Dolbeault cohomology
complex in such a general geometric context. We begin establishing the following
result, regarding the action of ∂¯τ .
7.7 The ∂ Operator and the Dolbeault Complex on Uniformly Rectifiable Sets 879

Proposition 7.7.1 Consider a UR domain Ω ⊆ Cn ≡ R2n and set σ := H 2n−1 ∂Ω.

Also, fix α, β ∈ {0, 1, . . . , n} along with p ∈ [1, ∞). Then the operator
p, ∂¯
∂¯τ : Lα,β τ (∂Ω, σ) −→ Lnor,C (∂Ω, σ) ⊗ Λα,β+2
p
(7.7.1)

is well defined, linear, and bounded.

p, ∂¯
Proof Pick an arbitrary form f ∈ Lα,β τ (∂Ω, σ). Since ∂¯τ f ∈ L p (∂Ω, σ) ⊗ Λα,β+2
by design, we only need to show that ∂¯τ f is complex normal (cf. Definition 7.1.7
and the subsequent discussion).
With this goal in mind, denote by ν the geometric measure theoretic outward
unit normal to Ω, and introduce Ω+ := Ω and Ω− := Rn \ Ω. Next, for each
± as in (7.3.38) with the domains
α, β ∈ {0, 1, . . . , n} define the integral operators Aα,β
Ω± playing the role of Ω. Also, recall the operators Bα,β ± introduced in (7.3.83).

Having fixed some aperture parameter κ > 0, make use of Proposition 7.3.2 and
Proposition 7.3.4 (both written for the UR domains Ω± , mindful of item (7) in [68,
Lemma 5.10.9]) in order to write
  κ−n.t. 1 ±  κ−n.t.
¯ ± f 
∂B = − Aα,β+1 (∂¯τ f ) 

α,β
∂Ω 2 ∂Ω

1  1
= − ± ν 1,0 ∨ ∂¯τ f + Aα,β (∂¯τ f )
2 2
= ∓ 14 ν 1,0 ∨ ∂¯τ f − 12 Aα,β+1 (∂¯τ f ) at σ-a.e. point on ∂Ω. (7.7.2)

If we now abbreviate

u± := Bα,β
±
f ∈ 𝒞∞ (Ω± ) ⊗ Λα,β, (7.7.3)

then (7.7.2) allows us to express

 +  κ−n.t.  −  κ−n.t.
− 12 ν 1,0 ∨ ∂¯τ f = ∂B
¯
α,β f  − ∂B
¯
α,β f 
∂Ω ∂Ω
κ−n.t. κ−n.t.
¯ )
= (∂u + ¯ − )
− (∂u . (7.7.4)
∂Ω ∂Ω

Observe that thanks to items (ii), (iii), and (viii) in Theorem 7.3.5, we may invoke
the conclusions of Proposition 7.2.7 applied to the functions u± in Ω± . Concretely,
based on (7.7.4), Proposition 7.2.7, the linearity of the operator ∂¯τ , the jump-formulas
recorded in (7.3.84), and (7.2.31), we may write
880 7 Applications to Analysis in Several Complex Variables
 + κ−n.t.   − κ−n.t. 
¯ )
− 12 ν 0,1 ∧ (ν 1,0 ∨ ∂¯τ f ) = ν 0,1 ∧ (∂u − ν 0,1 ∧ (∂u ¯ )
∂Ω ∂Ω
  κ−n.t.    κ−n.t. 
= − ∂¯τ u+ ∂Ω − ∂¯τ u− ∂Ω
 κ−n.t. κ−n.t. 
= −∂¯τ u+ ∂Ω − u− ∂Ω = −∂¯τ ( ftan,C )

= −∂¯τ f at σ-a.e. point on ∂Ω. (7.7.5)

In light of (7.1.69), the above identity entails

∂¯τ f = 12 ν 0,1 ∧ (ν 1,0 ∨ ∂¯τ f ) = (∂¯τ f )nor,C at σ-a.e. point on ∂Ω. (7.7.6)

This amounts to saying that ∂¯τ f is complex normal, as wanted. 

To proceed, in the definition below we introduce a very general version of the

boundary ∂-bar
¯ operator.

Definition 7.7.2 Let Ω ⊆ Cn ≡ R2n be a UR domain. Denote by ν the geometric

measure theoretic outward unit normal to Ω and abbreviate σ := H 2n−1 ∂Ω. Also,
fix α, β ∈ {0, 1, . . . , n} along with p ∈ [1, ∞). In this context, define
p, ∂¯ p, ∂¯
Lα,β b (∂Ω, σ) := Lα,β τ (∂Ω, σ) ∩ Ltan,C (∂Ω, σ) ⊗ Λα,β,
p
(7.7.7)

and introduce the boundary d-bar operator

p, ∂¯
∂¯b : Lα,β b (∂Ω, σ) −→ Ltan,C (∂Ω, σ) ⊗ Λα,β+1
p
(7.7.8)

by setting
p, ∂¯
∂¯b f := ν 1,0 ∨ ∂¯τ f for each f ∈ Lα,β b (∂Ω, σ). (7.7.9)

From (7.2.27), Definition 7.1.9, and item (2) in Lemma 7.1.6 it follows that the
operator (7.7.8)-(7.7.9) is well defined, linear, and bounded. In addition, it satisfies
the properties presented in the proposition below.

Proposition 7.7.3 Let Ω ⊆ Cn ≡ R2n be a UR domain and set σ := H 2n−1 ∂Ω.

Also, fix α, β ∈ {0, 1, . . . , n} along with p ∈ [1, ∞). Then the operator ∂¯b in (7.7.8)
is well defined, linear, bounded, and satisfies
p, ∂¯
∂¯τ f = 12 ν 0,1 ∧ ∂¯b f , ∀ f ∈ Lα,β b (∂Ω, σ). (7.7.10)

Furthermore,
p, ∂¯
∂B
¯ α,β f = − 1 Bα,β+1 (∂¯b f ) in Cn \ ∂Ω,
2 ∀ f ∈ Lα,β b (∂Ω, σ). (7.7.11)

Proof The first claim in the statement, pertaining to the operator ∂¯b , follows from
Definition 7.7.2, Proposition 7.7.1, and Lemma 7.1.6. Turning our attention to the
p, ∂¯
second claim, fix an arbitrary form f ∈ Lα,β b (∂Ω, σ). Based on the definition of ∂¯b ,
7.7 The ∂ Operator and the Dolbeault Complex on Uniformly Rectifiable Sets 881

item (3) in Lemma 7.1.6, (7.1.63), the fact that the differential form ∂¯τ f is complex
normal (cf. Proposition 7.7.1), and (7.1.63), we may write

ν 0,1 ∧ ∂¯b f = ν 0,1 ∧ (ν 1,0 ∨ ∂¯τ f )

 
= ν 0,1, ν 1,0 ∂¯τ f − ν 1,0 ∨ (ν 0,1 ∧ ∂¯τ f )
C

 
= ν 0,1, ν 0,1
∂¯τ f = |ν 0,1 |C2 ∂¯τ f = 2 ∂¯τ f . (7.7.12)
C

This proves (7.7.10). Lastly, for each z ∈ Cn \ ∂Ω, Proposition 7.3.2 and (7.7.10)
allow us to compute
∫
  1  
∂¯ Bα,β f (z) = (∂¯τ f )(ζ), ∂¯ζ Γα,β+1 (ζ, z) dσ(ζ)
2 ∂∗ Ω C

∫
1  0,1 
= ν (ζ) ∧ (∂¯b f )(ζ), ∂¯ζ Γα,β+1 (ζ, z) dσ(ζ)
4 ∂Ω C

1
= − Bα,β+1 (∂¯b f )(z). (7.7.13)
2
This establishes (7.7.11), and finishes the proof of the proposition. 

In the class of uniformly rectifiable domains, (7.7.9) and (7.7.10) show that

whenever Ω ⊆ R2n ≡ Cn is a given UR domain and

p, ∂¯
f ∈ Lα,β b (∂Ω, σ) with p ∈ [1, ∞) and σ := H 2n−1 ∂Ω, (7.7.14)
f is a CR-form ⇐⇒ ∂¯b f = 0 at σ-a.e. point on ∂Ω.

Our final result in this section opens the doors for considering the Dolbeault
cohomology complex in the context of UR sets.

Proposition 7.7.4 Consider a UR domain Ω ⊆ Cn ≡ R2n , and set σ := H 2n−1 ∂Ω.

Also, fix α, β ∈ {0, 1, . . . , n} along with p ∈ [1, ∞). Then the operator ∂¯b from (7.7.8)
actually induces a well-defined, linear, and bounded mapping
p, ∂¯ p, ∂¯b
∂¯b : Lα,β b (∂Ω, σ) −→ Lα,β+1 (∂Ω, σ) (7.7.15)

which satisfies
∂¯b ◦ ∂¯b = 0. (7.7.16)
p, ∂¯
Proof Pick f ∈ Lα,β b (∂Ω, σ). If ν stands for the geometric measure theoretic
outward unit normal to Ω, then (7.7.9) gives

g := ∂¯b f = ν 1,0 ∨ ∂¯τ f ∈ Ltan,C (∂Ω, σ) ⊗ Λα,β+1 .

p
(7.7.17)

Having selected an arbitrary form ψ ∈ 𝒞∞

c (C ) ⊗ Λ
n α,β+3 , we may compute
882 7 Applications to Analysis in Several Complex Variables
∫ ∫
   
ν 0,1 ∧ g, ϑψ dσ = ν 0,1 ∧ (ν 1,0 ∨ ∂¯τ f ), ϑψ dσ
C C
∂∗ Ω ∂∗ Ω
∫ ∫
   
=2 (∂¯τ f )nor,C, ϑψ dσ = 2 ∂¯τ f , ϑψ dσ
C C
∂∗ Ω ∂∗ Ω
∫
 
=2 f , ϑ(ϑψ) dσ = 0, (7.7.18)
C
∂∗ Ω

thanks to (7.7.17), (7.1.68), (7.7.1), and (7.2.14). In concert with (7.7.17), (7.2.27)
p, ∂¯b
(and (7.2.26)), this proves that ∂¯b f ∈ Lα,β+1 (∂Ω, σ) and ∂¯b (∂¯b f ) = 0. All conclu-
sions in the proposition then readily follow from this. 
Chapter 8
Hardy Spaces for Second-Order Weakly Elliptic
Operators in the Complex Plane

In this chapter we shall work  in the complex plane C ≡ R2 . Recall the Cauchy-
Riemann operator ∂z̄ := 2 ∂x + i∂y ) and its complex conjugate ∂z := 12 ∂x − i∂y ).
1

From the discussion in [70, §1.4] (cf. [70, (1.4.186)]) we know that there are three
prototypes of scalar, constant (complex) coefficient, elliptic weakly elliptic operators
in the plane, namely ∂z ∂z̄ which, up to a multiplicative factor is the two-dimensional
Laplacian, ∂z̄2 also referred to as Bitsadze’s operator (cf. [3], [4]), and its complex
conjugate ∂z2 .
We have already considered in [70, §5.5] Hardy spaces of harmonic functions
(i.e., null-solutions of the Laplacian) in NTA domains. In this section the focus is
the study of Hardy spaces of null-solutions for the operator ∂z̄2 in the unit disk of the
complex plane, a scale of spaces which interfaces tightly with the Dirichlet Problem
for ∂z̄2 in the unit disk. In this regard, it has been noted by A.V. Bitsadze that
   
(1 − |z| 2 )W : W ∈ 𝒞0 D and W holomorphic in D (8.0.1)

is an infinite dimensional subspace of the space of null-solutions for the classical

Dirichlet Problem for ∂z̄2 in the unit disk, i.e., the space of complex-valued functions
u satisfying
 
u ∈ 𝒞∞ (D) ∩ 𝒞0 D , ∂z̄2 u = 0 in D,
 (8.0.2)
and u∂D = 0 at each point on ∂D.

It was this observation that really thrusted the operator ∂z̄2 into the spotlight. Here
our goal is to precisely describe the space of null-solutions, as well as the space of
admissible boundary data, for the Dirichlet Problem for the Bitsadze operator in the
unit disk. For example, our work in this chapter (see (8.1.32) and (8.1.38)) shows
that the space
    
u ∈ 𝒞∞ D : ∂z̄2 u = 0 in D and u∂D = 0 (8.0.3)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 883
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1_8
884 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

coincides with
   
(1 − |z| 2 )W : W ∈ 𝒞∞ D and W holomorphic in D . (8.0.4)

We will primarily concern ourselves with the L p Dirichlet Problem for the Bit-
sadze operator in the unit disk, in which the boundary condition is taken in the
nontangential sense, and for which the size of the solution is measured using the
nontangential maximal operator. Ultimately, our analysis in §8.1 paints a very precise
picture of the failure of Fredholm solvability of the Dirichlet and Regularity Prob-
lems for Bitsadze’s operator in the unit disk of the complex plane. In §8.2 we carry
out a program with similar aims for a more inclusive family of scalar second-order
operators in the complex plane, namely Lλ := ∂z̄2 − λ2 ∂z2 , which contains Bitsadze’s
operator as a specal case (corresponding to λ = 0).

8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator

∂2z in the Unit Disk

In this section we shall work with Bitsadze’s operator ∂z̄2 in the complex plane. The
goal is to precisely describe its space of null-solutions as well as the corresponding
spaces of boundary traces, assuming only nontangential maximal function control.
Throughout, we abbreviate
 
D+ := D := z ∈ C : |z| < 1 , D− := C \ D, σ := H 1 ∂D. (8.1.1)

Fix some integrability exponent p ∈ (1, ∞) along with some aperture parameter
κ ∈ (0, ∞), and recall the Hardy space of holomorphic functions in D+ possessing
p-th power integrable nontangential maximal functions
 
H p (D+ ) := H p (D) := U holomorphic in D : Nκ U ∈ L p (∂D, σ) (8.1.2)

and, in the case of D− , also vanishing at infinity

 
H p (D− ) = U holomorphic in D− : Nκ U ∈ L p (∂D, σ) and U(∞) = 0 . (8.1.3)

We shall equip these spaces with the norms

U  H p (D± ) := Nκ U  L p (∂D,σ) for each U ∈ H p (D± ). (8.1.4)

Next, denote by 𝒞 the boundary-to-domain Cauchy operator associated with the unit
disk, acting on each function f ∈ L 1 (∂D, σ) according to
∫
1 f (ζ)
𝒞 f (z) := dζ, ∀z ∈ C \ ∂D. (8.1.5)
2πi ∂D ζ − z
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 885

This induces well-defined, linear, bounded mappings

𝒞 : L p (∂D, σ) −→ H p (D± ). (8.1.6)

We shall also need the corresponding boundary Hardy spaces of nontangential traces
of functions in H p (D± ), i.e.,
 κ−n.t. 
H p (∂D±, σ) := U ∂D± : U ∈ H p (D± ) . (8.1.7)

According to a classical theorem of Fatou, if u is a bounded harmonic function

on the unit disc, D in the complex plane C, then u has nontangential boundary traces
at H 1 -a.e. point on the unit circle ∂D. In fact, the same conclusion remains valid if u
is merely assumed to be bounded from below. The latter result has a local analogue
due to I. Privalov [85] (see also [109]). Specifically,
if u is a harmonic function in the unit disc D with the property that
for each point eiθ of a measurable subset E of the unit circle ∂D
there exists some κθ ∈ (0, ∞) such that u is bounded from below on the
nontangential approach region Γκθ (eiθ ), then for any aperture parameter (8.1.8)
 κ−n.t. 
κ ∈ (0, ∞) the nontangential boundary trace u∂D (z) exists for H 1 -
a.e. point z ∈ E.
See also A.P. Calderón [7] and E.M. Stein [96] for extensions to higher dimensions
(with the unit disk replaced by the upper half-space), and D. Jerison and C.E. Kenig
[42] for a local Fatou theorem in NTA domains.
The classical Fatou theorem mentioned above implies that
κ−n.t.
U ∂D± exists σ-a.e. on ∂D for each U ∈ H p (D± ), (8.1.9)

so the boundary Hardy spaces H p (∂D±, σ) are indeed well defined. In fact, as is
well known,

H p (∂D±, σ) are closed subspaces of L p (∂D, σ) and

(8.1.10)
L p (∂D, σ) = H p (∂D+, σ) ⊕ H p (∂D−, σ), direct sum.

Also, we have the Cauchy reproducing formulas

 κ−n.t. 
U = 𝒞 U ∂D± in D± for each U ∈ H p (D± ), (8.1.11)

as well as the Cauchy vanishing formulas

𝒞 f = 0 in D∓ for each f ∈ H p (∂D±, σ). (8.1.12)

In fact,  
f ∈ L p (∂D, σ) : 𝒞 f = 0 in D± = H p (∂D∓, σ). (8.1.13)
886 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

We also claim that

 ∫ 
H (∂D+, σ) = f ∈ L (∂D, σ) :
p p
f (ζ)ζ k dσ(ζ) = 0 for all k ∈ N , (8.1.14)
∂D

and
 ∫ 
H p (∂D−, σ) = f ∈ L p (∂D, σ) : f (ζ)ζ −k dσ(ζ) = 0 for all k ∈ N0 .
∂D
(8.1.15)
To justify (8.1.14), note that
 
ζ 
  = 1 < 1 for all z ∈ C \ D and ζ ∈ ∂D, (8.1.16)
 z  |z|

hence for each f ∈ L p (∂D, σ) and z ∈ C \ D we may write

∫ ∫
1 f (ζ) 1 f (ζ)
(𝒞 f )(z) = dζ = dζ
2πi ∂D ζ − z 2πi ∂D (−z)(1 − ζ/z)
∞ ∫ ∞ ∫
1 f (ζ) ζ k 1 1
=− dζ = − f (ζ)ζ k dζ .
2πi k=0 ∂D z z 2πi k=0
z k+1 ∂D

∞ ∫
1 1
=− f (ζ)ζ k+1 dσ(ζ), (8.1.17)
2π k=0
z k+1 ∂D

where the last equality above makes use of the identity dζ = iζ dσ(ζ). In turn, the
power series expansion just established in (8.1.17) proves that for each given function
f ∈ L p (∂D, σ) we have
∫
𝒞 f = 0 in C \ D ⇐⇒ f (ζ)ζ k dσ(ζ) = 0 for all k ∈ N, (8.1.18)
∂D

so (8.1.14) follows from this and (8.1.13). The characterization in (8.1.15) may be
justified in a similar fashion. Among other things, (8.1.14) readily yields

z · H p (∂D+, σ) = H p (∂D+, σ) ⊕ z̄, (8.1.19)

where we define z̄ to be the (one-dimensional) linear span over the field C of the
function ∂D   z⨏→ z̄ ∈ C. Indeed,
⨏ the left-to-right inclusion is seen by decomposing
z̄ · f = z̄ · f − ∂D f dσ + z̄ · ∂D f dσ and using (8.1.14) for the first term, while
the right-to-left inclusion is a consequence of the fact that we have f = z̄(z f ) and
that multiplication by z keeps you in H p (∂D+, σ).
Going further, we also consider Hardy spaces with regularity in D± , i.e.,
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 887
p p
H1 (D) := H1 (D+ )
 
:= U holomorphic in D : Nκ U, Nκ (∇U) ∈ L p (∂D, σ) , (8.1.20)

and
p p
H1 (C \ D) := H1 (D− ) (8.1.21)
 
:= U holomorphic in D : Nκ U, Nκ (∇U) ∈ L p (∂D, σ), U(∞) = 0 .

We can then define boundary Hardy spaces with regularity, namely

 κ−n.t. 
H1 (∂D±, σ) := U ∂D± : U ∈ H1 (D± ) .
p p
(8.1.22)

In relation to these, it is useful to note that

p p
H p (∂D±, σ) ∩ L1 (∂D, σ) = H1 (∂D±, σ). (8.1.23)

Indeed, the left-to-right inclusion is seen from Cauchy’s reproducing formula,

(1.4.7), and (1.5.8). The right-to-left inclusion in (8.1.23) is implied by [69, Propo-
sition 11.3.4] and definitions. In concert with (8.1.14)-(8.1.15), this implies
 ∫ 
p p
H1 (∂D+, σ) = f ∈ L1 (∂D, σ) : f (ζ)ζ k dσ(ζ) = 0 for all k ∈ N , (8.1.24)
∂D

and
 ∫ 
f (ζ)ζ −k dσ(ζ) = 0 for all k ∈ N0 .
p p
H1 (∂D−, σ) = f ∈ L1 (∂D, σ) :
∂D
(8.1.25)
Also, as seen from definitions, (1.4.7), and (6.2.105),

the boundary-to-domain Cauchy operator 𝒞 induces

p p (8.1.26)
well-defined mappings from L1 (∂D, σ) into H1 (D± ).

Finally, (8.1.10) implies that we have the direct sum decomposition

p p p
L1 (∂D, σ) = H1 (∂D+, σ) ⊕ H1 (∂D−, σ). (8.1.27)

Our first result amounts to a “structure theorem” for null-solutions of the Bitsadze
operator ∂z̄2 in the unit disk of the complex plane. To facilitate its statement, we agree
to abbreviate
 
δ∂D (z) := dist z, ∂D = 1 − |z|, for each z ∈ D. (8.1.28)
 
Theorem 8.1.1 Let D := z ∈ C : |z| < 1 be the unit disk of the complex plane.
Then, for a given complex-valued function u defined in D, the conditions
888 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

u ∈ 𝒞∞ (D) and ∂z̄2 u = 0 in D, (8.1.29)

are equivalent to having the decomposition

u = (1 − |z| 2 )W + U + λ z̄ in D for some

(8.1.30)
holomorphic functions U, W in D and some λ ∈ C.

Furthermore, U, W, λ are uniquely determined by the function u namely,

λ = (∂z̄ u)(0), (8.1.31)

while, with ‘prime’ denoting the ordinary complex differentiation of holomorphic

functions, for each z ∈ D one has
⎧
⎪ (∂ u)(z) − (∂z̄ u)(0)
⎨ − z̄
⎪ if z ∈ D \ {0},
W(z) = z (8.1.32)
⎪
⎪ −(∂z̄ u)(0)
⎩ if z = 0,

and
⎧
⎪ (∂ u)(z) − (∂z̄ u)(0)
⎨ u(z) − z̄(∂z̄ u)(z) + z̄
⎪ if z ∈ D \ {0},
U(z) = z (8.1.33)
⎪
⎪ u(0) + (∂z̄ u)(0)
⎩ if z = 0.

In relation to the decomposition (8.1.30) of a function u as in (8.1.29), the

following additional properties are valid for any integrability exponent p ∈ (1, ∞)
and any aperture parameter κ ∈ (0, ∞):

(1) For any function u as in (8.1.29) one has (recall (8.1.1), (8.1.2), and (8.1.28))

Nκ u ∈ L p (∂D, σ) if and only if

  (8.1.34)
U ∈ H p (D) and Nκ δ∂D · W ∈ L p (∂D, σ).

(2) For any function u as in (8.1.29) one has

κ−n.t.
u∂D exists at σ-a.e. point on ∂D if and only if
κ−n.t.   κ−n.t. (8.1.35)
U ∂D exists and δ∂D · W ∂D = 0 at σ-a.e. point on ∂D.

(3) For any function u as in (8.1.29) satisfying two additional properties, namely
κ−n.t.
that Nκ u ∈ L p (∂D, σ) and the nontangential boundary trace u∂D exists at
σ-a.e. point on ∂D, it follows that
κ−n.t.
the function u∂D − λ z̄ belongs to
(8.1.36)
the boundary Hardy space H p (∂D+, σ),
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 889

and, with 𝒞 denoting the boundary-to-domain Cauchy operator associated with

the unit disk (cf. (8.1.5)), one has
 κ−n.t. 
U = 𝒞 u∂D in D. (8.1.37)

(4) For any function u as in (8.1.29) and with the additional property that Nκ u
belongs to L p (∂D, σ) one has
κ−n.t.
u∂D = 0 at σ-a.e. point on ∂D if and only if
  κ−n.t. (8.1.38)
U = 0, λ = 0, and δ∂D · W ∂D = 0 at σ-a.e. point on ∂D.

(5) For any function u as in (8.1.29), the following equivalences are true: first,

Nκ (∂z̄ u) ∈ L p (∂D, σ) if and only if W ∈ H p (D), (8.1.39)

second,
Nκ (∇u) ∈ L p (∂D, σ) if and only if
p (8.1.40)
W ∈ H p (D) and U ∈ H1 (D),
third,

Nκ (∇u) ∈ L p (∂D, σ) if and only if

κ−n.t. (8.1.41)
∂z̄ u ∈ H p (D), Nκ u ∈ L p (∂D, σ), u∂D ∈ L1 (∂D, σ),
p

and, finally,

Nκ (∇u) ∈ L p (∂D, σ) if and only if

κ−n.t. (8.1.42)
∂z̄ u ∈ H p (D), Nκ u ∈ L p (∂D, σ), u∂D ∈ H1 (∂D, σ) ⊕ z̄
p

where, as before, z̄ denotes the (one-dimensional) linear span over C of the
function ∂D  z → z̄ ∈ C.

We wish to remark that having u ∈ 𝒞∞ (D) with ∂z̄2 u = 0 in D implies that

the function ∂z̄ u is holomorphic in D. In particular, it makes sense to speak of the
complex derivative (∂z̄ u)(0) in (8.1.32). Let also note that, under the assumptions
κ−n.t.
in item (3), u∂D is a function in L p (∂D, σ), so the Cauchy operator acts in a
meaningful manner in (8.1.37). Finally, we observe that

there exists a bounded function u ∈ 𝒞 ∞ (D) with ∂z̄2 u = 0 in D and

such that for any aperture parameter κ > 0 the nontangential boundary (8.1.43)
κ−n.t.
trace u∂D fails to exist at every point on ∂D.

This is implicit in the discussion in [63, §2] where the function in question, defined
(for some fixed even α ∈ N sufficiently large) as
890 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
∞
α k z α for each z ∈ D,
k
u(z) := (1 − |z| 2 ) (8.1.44)
k=1

is actually shown to fail to possess a radial limit at each point on ∂D. Another
counterexample is described in [1, §4.3, p. 233]. Thus, in stark contrast with the
case of harmonic functions, null-solutions u of the operator ∂z̄2 in the unit disk
may fail to have a nontangential boundary trace at any point on ∂D even when
Nκ u ∈ L p (∂D, σ) for each κ > 0 and each p ∈ (0, ∞).
We next give the proof of Theorem 8.1.1.
Proof of Theorem 8.1.1 Assume first that u is an in (8.1.30). A moment’s reflection
then shows that both conditions in (8.1.29) are satisfied. Furthermore, applying the
Cauchy-Riemann operator to the decomposition formula in (8.1.30) yields

∂z̄ u = −zW + λ in D (8.1.45)

which, in turn, forces (∂z̄ u)(0) = λ (so we necessarily have (8.1.31)) and

(∂z̄ u)(z) − (∂z̄ u)(0)

W(z) = − for each z ∈ D \ {0}. (8.1.46)
z
In view of the fact that W is continuous (by virtue of being holomorphic) in D and
that, as noted earlier, ∂z̄ u is holomorphic in D, we conclude from (8.1.46) that

W(0) = −(∂z̄ u)(0). (8.1.47)

Collectively, (8.1.46) and (8.1.47) prove (8.1.32). Then a straightforward computa-

tion based on (8.1.31) and (8.1.32) shows that U = u − (1 − |z| 2 )W − λ z̄ is necessarily
as claimed in (8.1.33). In particular, U, W, λ are uniquely determined by the function
u.
In the opposite direction, suppose u is as in (8.1.29). If we define U, W, λ as in
(8.1.31)-(8.1.33) then basic algebra shows that the decomposition claimed in (8.1.30)
holds. Moreover, the fact that ∂z̄ u is holomorphic in D renders W holomorphic in D.
In concert with the observation that, as seen from (8.1.29), the function

D  x −→ u(z) − z̄(∂z̄ u)(z) is holomorphic in D, (8.1.48)

this also proves that U defined as in (8.1.33) is holomorphic in D. This establishes

the equivalence between (8.1.29) and (8.1.30).
To deal with the claims in items (1)-(5), fix some p ∈ (1, ∞) along with some
κ ∈ (0, ∞).
Proof of item (1): In one direction, assume Nκ u ∈ L p (∂D, σ). Based on this and the
last part in [68, Corollary 8.9.13] we deduce that
 
Nκ δ∂D · ∇u ∈ L p (∂D, σ). (8.1.49)

From (8.1.49) and (8.1.32) we then see that

8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 891
ρ 
Nκ δ∂D · W ∈ L p (∂D, σ) for each truncation parameter ρ ∈ (0, 1/2). (8.1.50)

Since W is bounded on compact subsets of D (by virtue of being continuous in D),

we ultimately conclude that
 
Nκ δ∂D · W ∈ L p (∂D, σ). (8.1.51)

Thus, the last property claimed in the second line of (8.1.34) holds. Next, observe
that
1 − |z| 2 = (1 + |z|)δ∂D (z) for each z ∈ D. (8.1.52)
In particular, we may recast the decomposition formula from (8.1.30) as
 
U = u − (1 + |z|) δ∂D · W − λ z̄ in D. (8.1.53)

From this, (8.1.51), and the current working assumption we then conclude that
Nκ U ∈ L p (∂D, σ). In view of (8.1.2), this shows that the first property claimed in
the second line of (8.1.34) holds as well. This proves the direct implication in the
equivalence stated in (8.1.34).
In the opposite direction, from the second line of (8.1.34) and the decomposition
formula from (8.1.30) (reformulated as in (8.1.53)) we readily conclude that Nκ u
lies in L p (∂D, σ).
Proof of item (2): In one direction, assume u is a function as in (8.1.29) with the
κ−n.t.
additional property that the nontangential boundary trace u∂D exists at σ-a.e. point
on ∂D. In concert with the very last part in [68, Proposition 8.9.11] this implies that

for each aperture parameter κ  ∈ (0, κ) we have

  κ  −n.t. (8.1.54)
δ∂D · ∇u ∂D = 0 at σ-a.e. point on ∂D.

In particular,
  κ  −n.t.
δ∂D · (∂z̄ u) ∂D = 0 at σ-a.e. point on ∂D, for each κ  ∈ (0, κ). (8.1.55)

From (8.1.55) and (8.1.32) we then deduce that

  κ  −n.t.
δ∂D · W ∂D = 0 at σ-a.e. point on ∂D, for each κ  ∈ (0, κ). (8.1.56)

Based on (8.1.56), the fact that we are presently assuming that the nontangential
κ−n.t.
boundary trace u∂D exists at σ-a.e. point on ∂D, and (8.1.53) we then conclude
that
for each κ  ∈ (0, κ) the nontangential boundary trace
κ  −n.t. (8.1.57)
U ∂D exists (in C) at σ-a.e. point on ∂D.
Having established this, we may invoke the local Fatou theorem recalled in (8.1.8)
(bearing in mind that U is a holomorphic function in D) and conclude that
892 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
κ−n.t.
U ∂D exists at σ-a.e. point on ∂D. (8.1.58)
κ−n.t.
From this, the fact that we are currently assuming that u∂D exists at σ-a.e. point
on ∂D, and (8.1.53) we see then that
  κ−n.t.
δ∂D · W ∂D exists σ-a.e. on ∂D. (8.1.59)

Finally, combining (8.1.54) and (8.1.59) we arrive at the conclusion that

  κ−n.t.
δ∂D · W ∂D = 0 at σ-a.e. point on ∂D. (8.1.60)

Together, (8.1.58) and (8.1.60) prove the claims made in the second line of (8.1.35).
This takes care of the direct implication in the equivalence (8.1.35). The opposite
implication in the equivalence claimed in (8.1.35) is clear from (8.1.53).
Proof of item (3): Consider a function u as in (8.1.29), with Nκ u ∈ L p (∂D, σ), and
κ−n.t.
such that u∂D exists at σ-a.e. point on ∂D. From item (2) and (8.1.53) we then see
that  κ−n.t.   κ−n.t. 
U ∂D (z) = u∂D (z) − λ z̄ for σ-a.e. point z ∈ ∂D. (8.1.61)
Let us also note that, given the current working assumptions, item (1) guarantees
that U ∈ H p (D). Based on this, (8.1.61), the Cauchy reproducing and vanishing
formulas recorded in (8.1.11)-(8.1.12), and also keeping in mind that the function

D  z → z −1 ∈ C belongs to H p (∂D−, σ), (8.1.62)

we may then write

 κ−n.t.   κ−n.t.   
U = 𝒞 U ∂D = 𝒞 u∂D + 𝒞 λ z̄
 κ−n.t.     κ−n.t. 
= 𝒞 u∂D + λ𝒞 z −1 = 𝒞 u∂D in D, (8.1.63)

establishing (8.1.37). Finally, since U ∈ H p (D), from (8.1.61), (8.1.7), and (8.1.31)
we conclude that the membership claimed in (8.1.36) holds as well.
Proof of item (4): Suppose u is as in (8.1.29) and has Nκ u ∈ L p (∂D, σ). In one
κ−n.t.
direction, if u∂D = 0 at σ-a.e. point on ∂D then (8.1.37) gives at once that U = 0
in D. Also, from the current working assumptions, (8.1.36), (8.1.62), the fact that the
sum in (8.1.10) is direct, and (8.1.31) we deduce that λ = 0. Since item (2) presently
  κ−n.t.
implies that δ∂D · W ∂D = 0 at σ-a.e. point on ∂D, the direct implication in
(8.1.38) is established. The converse implication in (8.1.38) is a simple consequence
of (8.1.53).
Proof of item (5): Throughout, fix a function u as in (8.1.29). From the definition of
W in (8.1.32) it is clear that
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 893

for each truncation parameter ρ ∈ (0, 1/2) one has

ρ ρ
(8.1.64)
Nκ (∂z̄ u) ∈ L p (∂D, σ) if and only if Nκ W ∈ L p (∂D, σ).

In view of the fact that both ∂z̄ u and W are bounded near the origin, this readily
implies that

Nκ (∂z̄ u) ∈ L p (∂D, σ) if and only if Nκ W ∈ L p (∂D, σ). (8.1.65)

Bearing in mind (8.1.2), this establishes the equivalence in (8.1.39).

Assume next that Nκ (∇u) ∈ L p (∂D, σ). Then (8.1.65) guarantees that W be-
longs to the Hardy space H p (D) which, in concert with the last part in [68, Corol-
lary 8.9.13], further implies that
 
Nκ δ∂D · ∇W ∈ L p (∂D, σ). (8.1.66)

Also, applying the gradient to both sides of the decomposition formula in (8.1.30)
gives

∇u = −2zW + (1 − |z| 2 )∇W + ∇U + λ, −iλ

 
= −2zW + (1 + |z|) δ∂D · ∇W + ∇U + λ, −iλ in D. (8.1.67)

From this, the fact that both Nκ (∇u) and Nκ W belong to L p (∂D, σ), and (8.1.66) we
then conclude that Nκ (∇U) belongs to L p (∂D, σ). Since from item (1) we know that
p
U ∈ H p (D), we ultimately conclude that U actually belongs to H1 (D) (defined in
(8.1.20)). Another, alternative way of showing that U ∈ H (D) is to rely on (8.1.37)
p
κ−n.t.
and the fact that we presently have u∂D ∈ L1 (∂D, σ). This finishes the proof of
p

the direct implication in (8.1.40). In the opposite direction, assume W ∈ H p (D)

p
and U ∈ H1 (D) Then the last part in [68, Corollary 8.9.13] implies that (8.1.66)
holds, so we conclude from (8.1.67) that Nκ (∇u) ∈ L p (∂D, σ). The equivalence in
(8.1.40) is therefore justified.
Moving on, assume Nκ (∇u) ∈ L p (∂D, σ). Then ∂z̄ u is a holomorphic function
in D with Nκ (∂z̄ u) ∈ L p (∂D, σ), hence ∂z̄ u ∈ H p (D) (cf. (8.1.2)). Also, [68,
Proposition 8.4.9], [68, Proposition 8.9.22], and [69, Proposition 11.3.4] enure that
 κ−n.t.
Nκ u ∈ L p (∂D, σ) and that the nontangential boundary trace u∂D exists and belongs
p
to L1 (∂D, σ). This proves the direct implication in (8.1.41).
In the opposite direction, assume u is a function as in (8.1.29) and has the following
κ−n.t.
additional properties: ∂z̄ u ∈ H p (D), the nontangential boundary trace u∂D exists
p
and belongs to L1 (∂D, σ), and Nκ u ∈ L p (∂D, σ). In concert with item (3) and
(8.1.26), these properties imply that
 κ−n.t. 
U = 𝒞 u∂D ∈ H1 (D).
p
(8.1.68)
894 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

In addition, the current assumptions and (8.1.39) guarantee that W ∈ H p (D). Based
on this, (8.1.68), and (8.1.40) we conclude that Nκ (∇u) ∈ L p (∂D, σ). This completes
the proof of the equivalence claimed in (8.1.41).
Finally, consider the equivalence claimed in (8.1.42). In one direction, assume u is
a function as in (8.1.29) with Nκ (∇u) ∈ L p (∂D, σ). From (8.1.41) we then conclude
κ−n.t.
that ∂z̄ u ∈ H p (D), Nκ u ∈ L p (∂D, σ), and the nontangential boundary trace u∂D
p
exists and belongs to L1 (∂D, σ). Granted these properties, item (3) guarantees that
κ−n.t.
the function u∂D − λ z̄ belongs to the boundary Hardy space H p (∂D+, σ). As such,
said function belongs to
p p
H p (∂D+, σ) ∩ L1 (∂D, σ) = H1 (∂D+, σ), (8.1.69)
κ−n.t.
with the equality provided by (8.1.23). Ultimately, this shows that u∂D belongs
p
to H1 (∂D, σ) ⊕ z̄, so the direct implication in (8.1.42) holds. The opposite
implication is readily seen from (8.1.41) and (8.1.23), so the proof of the
equivalence claimed in (8.1.42) is complete. 

In the theorem below we identify both the space of admissible boundary data and
the space of null-solutions for the L p Dirichlet Problem for the Bitsadze operator in
the unit disk of the complex plane. The main ingredient in the proof is the structural
result established earlier in Theorem 8.1.1, centered around the decomposition in
(8.1.30). In particular, the nature of the piece U decisively determines the nature
of the space of admissible boundary data, while the nature of the piece W is solely
responsible for the nature of the space of null-solutions.

Theorem 8.1.2 Denote by D be the unit disk in the complex plane, and abbreviate
σ := H 1 ∂D. Fix an integrability exponent p ∈ (1, ∞) and some aperture parameter
κ ∈ (0, ∞). Then the space of admissible boundary data for the L p Dirichlet Problem
for the Bitsadze operator in the unit disk has the following description:
 κ−n.t.
u∂D : u ∈ 𝒞∞ (D), ∂z̄2 u = 0 in D, Nκ u ∈ L p (∂D, σ), (8.1.70)
κ−n.t. 
and u∂D exists σ-a.e. on ∂D = H p (∂D+, σ) ⊕ z̄,

where z̄ is defined as the (one-dimensional) linear span over the field C of the
function ∂D  z → z̄ ∈ C. Alternatively,
 κ−n.t.
u∂D : u ∈ 𝒞∞ (D), ∂z̄2 u = 0 in D, Nκ u ∈ L p (∂D, σ), (8.1.71)
κ−n.t. 
and u∂D exists σ-a.e. on ∂D
 ∫ 
= f ∈ L p (∂D, σ) : f (z)z k dσ(z) = 0 for all k ∈ N with k ≥ 2 .
∂D

In addition, the space of null-solutions for the L p Dirichlet Problem for the
Bitsadze operator in the unit disk, i.e.,
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 895

u ∈ 𝒞∞ (D) : ∂z̄2 u = 0 in D, Nκ u ∈ L p (∂D, σ),
κ−n.t. 
and u∂D = 0 at σ-a.e. point on ∂D , (8.1.72)

coincides with the infinite dimensional space

  
(1 − |z| 2 )W : W holomorphic in D, Nκ δ∂D · W ∈ L p (∂D, σ),
  κ−n.t. 
and δ∂D · W ∂D = 0 at σ-a.e. point on ∂D . (8.1.73)

In light of the classical decomposition (8.1.10), we conclude from (8.1.70) that

the cokernel in the full space of boundary data L p (∂D, σ) of the space of admissible
boundary data for the L p Dirichlet Problem for the Bitsadze operator in the unit disk
(described in the left side of (8.1.70)) is isomorphic to

H p (∂D−, σ) z̄, (8.1.74)

which is an infinite dimensional space. Also, since

 
(1 − |z| 2 )W : W ∈ H p (D) (8.1.75)

is an infinite dimensional subspace of (8.1.73), it follows that

the space of null-solutions for the L p Dirichlet Problem for
the Bitsadze operator in the unit disk (described in (8.1.72)) (8.1.76)
is infinite dimensional.
The space (8.1.73) is far richer than the space identified by A.V. Bitsadze in (8.0.1).
Indeed, as opposed to the latter, the former contains functions whose nontangential
trace fails to be absolutely integrable. An example of this sort is offered by
1
W∗ : D → C, W∗ (z) := for each z ∈ D. (8.1.77)
1−z
 
Obviously, this is holomorphic in D and belongs to 𝒞∞ D \ {1} . In particular,
  κ−n.t.
δ∂D · W∗ ∂D = 0 at each point in ∂D \ {1}, hence at σ-a.e. point  on ∂D. Since we
also have δ ∂D (z) = 1 − |z| ≤ |1 − z| for each z ∈ D, it follows that δ∂D · W∗  ≤ 1 in D,
hence Nκ δ∂D · W∗ ∈ L ∞ (∂D, σ). Thus, W∗ belongs to (8.1.73) but not to (8.0.1).
As a byproduct, the function

1 − |z| 2
u∗ : D → C, u∗ (z) := for each z ∈ D (8.1.78)
1−z
 
belongs to the space (8.1.72) but fails to be in 𝒞0 D . More generally, for any two
finite families θ 1, . . . , θ N ∈ [0, 2π) and c1, . . . , c N ∈ C, the function
896 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

N
cj
Wo : D → C, Wo (z) := for each z ∈ D, (8.1.79)
j=1
eiθ j − z

belongs to (8.1.73) but not to (8.0.1).

Finally, we note that thanks to the result recorded in (8.1.43) it is actually necessary
κ−n.t.
to assume the existence of the nontangential boundary trace u∂D in (8.1.70) and
(8.1.71).
Here is the proof of Theorem 8.1.2.
Proof of Theorem 8.1.2 The left-to-right inclusion in (8.1.70) follows from item
(3) in Theorem 8.1.1 (cf. (8.1.36)). To justify the right-to-left inclusion in (8.1.70),
observe (cf. (8.1.2)) that any function f belonging to H p (∂D+, σ) ⊕ z̄ is of the
form κ−n.t.
f = U ∂D + λ z̄ (8.1.80)
for some holomorphic function U in D with Nκ U ∈ L p (∂D, σ) and some complex
number λ. Then the function u : D → C given by u(z) := U(z) + λ z̄ for each z ∈ D
satisfies
u ∈ 𝒞∞ (D), ∂z̄2 u = 0 in D, Nκ u ∈ L p (∂D, σ),
κ−n.t. κ−n.t. (8.1.81)
and u∂D = U ∂D + λ z̄ = f at σ-a.e. point on ∂D.

This places f in the space appearing in the left side of (8.1.70), so the proof of (8.1.70)
is complete. The alternative characterization offered in (8.1.71) is a consequence of
(8.1.70) and the fact that

H p (∂D+, σ) ⊕ z̄ (8.1.82)

 ∫ 
= f ∈ L (∂D, σ) :
p
f (z)z k dσ(z) = 0 for all k ∈ N with k ≥ 2 .
∂D

In turn, (8.1.82) is justified by observing that the left-to-right inclusion is directly

implied by (8.1.14), while the right-to-left inclusion follows from the observation
that if f belongs to the space in the right side of (8.1.82) then the function
⨏
∂D  z −→ f (z) − f (ζ)ζ dσ(ζ) z̄ ∈ C (8.1.83)
∂D

belongs to the space in the right side of (8.1.14), hence ultimately to H p (∂D+, σ).
This places f in H p (∂D+, σ) ⊕ z̄, finishing the proof of (8.1.82). The claim made
in (8.1.71) is therefore established.
Next, from the decomposition in (8.1.30) and items (1), (4) of Theorem 8.1.1 we
see that the space (8.1.72) is contained in (8.1.73). Conversely, given any holomor-
    κ−n.t.
phic function W in D with Nκ δ∂D ·W ∈ L p (∂D, σ) and such that δ∂D ·W ∂D = 0
at σ-a.e. point on ∂D, it follows that u := (1 − |z| 2 )W belongs to 𝒞∞ (D), solves
κ−n.t.
∂z̄2 u = 0 in D, has Nκ u ∈ L p (∂D, σ), and satisfies u∂D = 0 at σ-a.e. point on
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 897

∂D. This places u in the space (8.1.72). Hence, the space (8.1.73) is contained in
(8.1.72). 

We wish to augment Theorem 8.1.2 with a result of similar flavor, describing

p
the space of admissible boundary data and the space of null-solutions for the L1
Regularity Problem for the Bitsadze operator in the unit disk of the complex plane.

Theorem 8.1.3 Let D be the unit disk in the complex plane, and set σ := H 1 ∂D.
Also, pick an exponent p ∈ (1, ∞) and some aperture parameter κ ∈ (0, ∞). Then the
p
space of admissible boundary data for the L1 Regularity Problem for the Bitsadze
operator in the unit disk has the following description:
 κ−n.t. 
u∂D : u ∈ 𝒞∞ (D), ∂z̄2 u = 0 in D, Nκ (∇u) ∈ L p (∂D, σ)
p
= H1 (∂D+, σ) ⊕ z̄ (8.1.84)

where, as in the past, z̄ is the (one-dimensional) linear span over C of the function
∂D  z → z̄ ∈ C. Alternatively,
 κ−n.t.
u∂D : u ∈ 𝒞∞ (D), ∂z̄2 u = 0 in D, Nκ (∇u) ∈ L p (∂D, σ), (8.1.85)
κ−n.t. 
and u∂D exists σ-a.e. on ∂D
 ∫ 
p
= f ∈ L1 (∂D, σ) : f (z)z k dσ(z) = 0 for all k ∈ N with k ≥ 2 .
∂D
p
Moreover, the space of null-solutions for the L1 Regularity Problem for the
Bitsadze operator in the unit disk, i.e.,

u ∈ 𝒞∞ (D) : ∂z̄2 u = 0 in D, Nκ (∇u) ∈ L p (∂D, σ),
κ−n.t. 
and u∂D = 0 at σ-a.e. point on ∂D , (8.1.86)

coincides with
 
(1 − |z| 2 )W : W ∈ H p (D) . (8.1.87)

Before presenting the proof of this theorem we make a three comments. First,
given any function u ∈ 𝒞∞ (D) with Nκ (∇u) ∈ L p (∂D, σ), the result established in
κ−n.t.
[68, (8.9.236)] implies that the nontangential trace u∂D exists at σ-a.e. point on
∂D. This is relevant in the context of (8.1.84).
Second, from (8.1.27) and (8.1.84) we see that
p
the quotient space between the full space of boundary data L1 (∂D, σ)
p
and the space of admissible boundary data for the L1 Regularity Prob-
(8.1.88)
lem for the Bitsadze operator in D (described in the left side of(8.1.84))
p
is isomorphic to the infinite dimensional space H1 (∂D−, σ) z̄.
898 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

Third, as seen from the last part in the statement of the above theorem,
p
the space of null-solutions for the L1 Regularity Problem for
the Bitsadze operator in the unit disk (described in (8.1.86)) is (8.1.89)
infinite dimensional.
The proof of Theorem 8.1.3 is given next:
Proof of Theorem 8.1.3 The left-to-right inclusion in (8.1.84) is seen from (8.1.42).
As regards the right-to-left inclusion in (8.1.84), first note from (8.1.22) that any
κ−n.t.
function f belonging to H1 (∂D+, σ) ⊕ z̄ is of the form f = U ∂D + λ z̄ for some
p

holomorphic function U in D with Nκ (∇U) ∈ L p (∂D, σ) and some complex number

λ. Then the function u : D → C given for each z ∈ D by u(z) := U(z) + λ z̄ belongs
to the space in the left side of (8.1.84) and satisfies
κ−n.t. κ−n.t.
u∂D = U ∂D + λ z̄ = f at σ-a.e. point on ∂D. (8.1.90)

As a consequence, f belongs to the space appearing in the left side of (8.1.84).

This establishes (8.1.84). The alternative characterization given in (8.1.85) is then
implied by (8.1.85) and the fact that
p
H1 (∂D+, σ) ⊕ z̄ (8.1.91)
 ∫ 
p
= f ∈ L1 (∂D, σ) : f (z)z k dσ(z) = 0 for all k ∈ N with k ≥ 2 ,
∂D

itself a consequence of (8.1.91) and (8.1.23).

Next, let us justify the fact that the spaces in (8.1.86) and (8.1.87) coincide. First,
from the decomposition in (8.1.30) item (4) of Theorem 8.1.1 and (8.1.40) it follows
that the space (8.1.86) is contained in (8.1.87). For the opposite inclusion, given any
function W ∈ H p (D) it readily follows that u := (1 − |z| 2 )W belongs to 𝒞∞ (D),
κ−n.t.
solves ∂z̄2 u = 0 in D, and satisfies u∂D = 0 at σ-a.e. point on ∂D. Also, (8.1.40)
(and the uniqueness of the decomposition in (8.1.30)) gives that Nκ u ∈ L p (∂D, σ).
Hence, u belongs to the space (8.1.86), thus finishing the proof of the fact that the
space (8.1.87) is contained in (8.1.86). 

Moving on, we present a “structure theorem” for null-solutions of the Bitsadze

operator ∂z̄2 in the complement of the closed unit disk in the complex plane (compare
with Theorem 8.1.1). The reader is advised to recall (8.1.1), (8.1.3), (8.1.7), and
(8.1.28).
 
Theorem 8.1.4 Let D := z ∈ C : |z| < 1 be the unit disk of the complex plane
and abbreviate D− := C \ D. Then, for a given complex-valued function u defined in
D− , the conditions
u ∈ 𝒞∞ (D− ) and ∂z̄2 u = 0 in D−, (8.1.92)
are equivalent to having the decomposition
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 899

u = (1 − |z| 2 )W + U in D−
(8.1.93)
for some holomorphic functions U, W in D− .

Moreover, U, W are uniquely determined by the function u, namely

(∂z̄ u)(z)
W(z) = − for each z ∈ D−, (8.1.94)
z
and
(∂z̄ u)(z)
U(z) = u(z) − z̄(∂z̄ u)(z) + for each z ∈ D− . (8.1.95)
z
Also,
u(∞) = 0 if and only if
(8.1.96)
U(∞) = 0 and W(z) = o(1/|z| 2 ) as z → ∞,
and
u(z) = O(1) as z → ∞ if and only if there exists c ∈ C such that
(8.1.97)
U(z) = c + O(1/|z|) and W(z) = O(1/|z| 2 ) as z → ∞.

Regarding the decomposition (8.1.93) of a function u as in (8.1.92), the following

additional properties are valid for any integrability exponent p ∈ (1, ∞) and any
aperture parameter κ ∈ (0, ∞):
(1) Let u be a function as in (8.1.92). Then

Nκ u ∈ L p (∂D, σ) if and only if

  (8.1.98)
Nκ U ∈ L p (∂D, σ) and Nκ (1 − |z| 2 )W ∈ L p (∂D, σ),

Thus, as a consequence of (8.1.98), (8.1.96), and (8.1.3),

Nκ u ∈ L p (∂D, σ) and u(∞) = 0 if and only if

 
U ∈ H p (D− ), Nκ (1 − |z| 2 )W ∈ L p (∂D, σ), (8.1.99)
and W(z) = o(1/|z| 2 ) as z → ∞,

while as a consequence of (8.1.98), (8.1.97), and (8.1.3),

Nκ u ∈ L p (∂D, σ) and u(z) = O(1) as z → ∞ if and only if

 
U ∈ H p (D− ) ⊕ C, Nκ (1 − |z| 2 )W ∈ L p (∂D, σ), (8.1.100)
and W(z) = O(1/|z| 2 ) as z → ∞.

(2) For any function u as in (8.1.92) one has

κ−n.t.
u∂D exists at σ-a.e. point on ∂D if and only if
κ−n.t.   κ−n.t. (8.1.101)
U ∂D exists and (1 − |z| 2 )W ∂D = 0 at σ-a.e. point on ∂D.
900 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

(3) For any given function u as in (8.1.92) satisfying the additional properties that
κ−n.t.
Nκ u ∈ L p (∂D, σ), the nontangential boundary trace u∂D exists at σ-a.e. point
on ∂D, and u(∞) = 0 it follows that
κ−n.t.
u∂D ∈ H p (∂D−, σ), (8.1.102)

and, with 𝒞 denoting the boundary-to-domain Cauchy operator associated with

D− , one has
 κ−n.t. 
U = 𝒞 u∂D in D− . (8.1.103)
If in place of u vanishing at infinity one now assumes that u(z) = O(1) as z → ∞,
then in place of (8.1.102)-(8.1.103) one obtains
κ−n.t.
u∂D ∈ H p (∂D−, σ) ⊕ C (8.1.104)

and, respectively,
 κ−n.t. 
U = c + 𝒞 u∂D in D−, for some c ∈ C. (8.1.105)

(4) For any given function u as in (8.1.92) and with the additional properties that
Nκ u ∈ L p (∂D, σ) and u(z) = O(1) as z → ∞ (hence, in particular, if u(∞) = 0)
one has
κ−n.t.
u∂D = 0 at σ-a.e. point on ∂D if and only if
  κ−n.t. (8.1.106)
U = 0 in D− and (1 − |z| 2 )W ∂D = 0 at σ-a.e. point on ∂D.

(5) For any function u as in (8.1.92) plus the additional property that u(∞) = 0, the
following equivalences are true: first,

Nκ (∂z̄ u) ∈ L p (∂D, σ) if and only if W ∈ H p (D− ), (8.1.107)

second,
Nκ (∇u) ∈ L p (∂D, σ) if and only if
p (8.1.108)
W ∈ H p (D− ) and U ∈ H1 (D− ),
third,

Nκ (∇u) ∈ L p (∂D, σ) if and only if

κ−n.t. (8.1.109)
∂z̄ u ∈ H p (D− ), Nκ u ∈ L p (∂D, σ), u∂D ∈ L1 (∂D, σ),
p

and, finally,

Nκ (∇u) ∈ L p (∂D, σ) if and only if

κ−n.t. (8.1.110)
∂z̄ u ∈ H p (D− ), Nκ u ∈ L p (∂D, σ), u∂D ∈ H1 (∂D−, σ).
p
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 901

Finally, if in place of u vanishing at infinity one now assumes that u(z) = O(1) as
z → ∞, then (8.1.107) and (8.1.109) remain valid, while in place of (8.1.108)
and (8.1.110) one now has

Nκ (∇u) ∈ L p (∂D, σ) if and only if

p (8.1.111)
W ∈ H p (D− ) and U ∈ H1 (D− ) ⊕ C,

and, respectively,

Nκ (∇u) ∈ L p (∂D, σ) if and only if

κ−n.t. (8.1.112)
∂z̄ u ∈ H p (D− ), Nκ u ∈ L p (∂D, σ), u∂D ∈ H1 (∂D−, σ) ⊕ C.
p

Proof Clearly, any function as in (8.1.93) satisfies both conditions in (8.1.92). If we

now start with u as in (8.1.92) and define W, U as in (8.1.94)-(8.1.95), then since
both ∂z̄ u and u − z̄∂z̄ u are holomorphic in D− , it follows that W, U are holomorphic
functions in D− and the decomposition in (8.1.93) holds. Moreover, applying ∂z̄ to
said decomposition reveals that W, U are necessarily given by the formulas (8.1.94)-
(8.1.95). In particular, W, U are uniquely determined by u.
Next, the equivalence claimed in (8.1.96) is a consequence of (8.1.93), the fact
that W, U are holomorphic functions in D− , the function u is a null-solution of
the weakly elliptic operator ∂z̄2 in D− , and [70, Lemma 1.5.6]. The equivalence in
(8.1.97) is justified similarly, now also making use of [70, Corollary 1.6.8] (with
D := ∂z̄ ).
To deal with the equivalence claimed in (8.1.98), assume first that Nκ u belongs
to L p (∂D, σ). In concert with the last part in [68, Corollary 8.9.13] this implies
 
Nκ δ∂D · ∂z̄ u ∈ L p (∂D, σ). (8.1.113)

In view of (8.1.28) and the fact that the function D−  z → (1 + |z|)/z ∈ C is

bounded, from (8.1.113) we conclude that

Nκ D−  z → (1 − |z| 2 ) (∂z̄ u)(z)

z ∈ L p (∂D, σ). (8.1.114)
 
Glancing at (8.1.94) then shows that Nκ (1 − |z| 2 )W ∈ L p (∂D, σ). From (8.1.93)
and the working hypothesis we also see that Nκ U ∈ L p (∂D, σ). This proves the
direct implication in (8.1.98). The converse implication in (8.1.98) is clear from
(8.1.93).
The claim in item (2) is justified by arguing much as in the proof of (8.1.35) (while
keeping in mind (8.1.52)). The claims in the first part of item (3) are consequences
of what we have proved in items (1)-(2) and Cauchy’s reproducing formula (8.1.11).
The claims in the second part of item (3), corresponding to the case when we now
assume that u(z) = O(1) as z → ∞, are dealt with similarly. The claims in item (4)
are implied by items (1)-(3).
There remains to justify the claims made in item (5). Assume u is a function as
in (8.1.92). If u(z) = O(1) as z → ∞, then (8.1.97) and [70, Lemma 1.5.6] imply
902 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

(∇u)(z) = O(1/|z|), (∇U)(z) = O(1/|z| 2 ),

(8.1.115)
and W(z) = O(1/|z| 2 ) as z → ∞.

Consequently, for any fixed truncation parameter ρ ∈ (0, ∞) we have:

ρ
Nκ (∂z̄ u) ∈ L p (∂D, σ) if and only if Nκ (∂z̄ u) ∈ L p (∂D, σ),
ρ
Nκ (∇u) ∈ L p (∂D, σ) if and only if Nκ (∇u) ∈ L p (∂D, σ),
ρ
(8.1.116)
Nκ (∇U) ∈ L p (∂D, σ) if and only if Nκ (∇U) ∈ L p (∂D, σ),
ρ
Nκ W ∈ L p (∂D, σ) if and only if Nκ W ∈ L p (∂D, σ).

Keeping these in mind, then the same type of argument as in the proof of (8.1.39)-
(8.1.42), now making use of (8.1.94)-(8.1.95) and what we have proved already in
the current items (1)-(4), yields all desired conclusions. 
There are similar results to those established in Theorem 8.1.2, identifying both
the space of admissible boundary data and the space of null-solutions for the L p
Dirichlet Problem for the Bitsadze operator in the complement of the closed unit
disk of the complex plane.

Theorem 8.1.5 Let D be the unit disk in C, and set σ := H 1 ∂D. Also, pick an
exponent p ∈ (1, ∞) and some aperture parameter κ ∈ (0, ∞). Then the space of
admissible boundary data for the L p Dirichlet Problem for the Bitsadze operator in
the complement of the closed unit disk has the following description:
 κ−n.t.  
u∂D : u ∈𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, Nκ u ∈ L p (∂D, σ), (8.1.117)
κ−n.t. 
u(∞) = 0, and u∂D exists σ-a.e. on ∂D = H p (∂D−, σ).

Alternatively,
 κ−n.t.  
u∂D : u ∈ 𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, Nκ u ∈ L p (∂D, σ), (8.1.118)
κ−n.t. 
u(∞) = 0, and u∂D exists σ-a.e. on ∂D
 ∫ 
= f ∈ L p (∂D, σ) : f (z)z−k dσ(z) = 0 for all k ∈ N0 .
∂D

Also, if in place of the vanishing condition at infinity one now assumes mere
boundedness for the function u, then in place of (8.1.117)-(8.1.118) one obtains
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 903
 κ−n.t.  
u∂D : u ∈𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, Nκ u ∈ L p (∂D, σ),
κ−n.t. 
u(z) = O(1) as z → ∞, u∂D exists σ-a.e. on ∂D
= H p (∂D−, σ) ⊕ C
 ∫ 
= f ∈ L p (∂D, σ) : f (z)z −k dσ(z) = 0 for all k ∈ N .
∂D
(8.1.119)

Moreover, the space of null-solutions for the L p Dirichlet Problem for the Bitsadze
operator in the complement of the closed unit disk, i.e.,
  
u ∈ 𝒞∞ C \ D : ∂z̄2 u = 0 in C \ D, Nκ u ∈ L p (∂D, σ),
κ−n.t. 
u(z) = o(1) as z → ∞, u∂D = 0 at σ-a.e. point on ∂D ,
(8.1.120)

coincides with
  
(1 − |z| 2 )W : W holomorphic in C \ D, Nκ (1 − |z| 2 )W ∈ L p (∂D, σ),
  κ−n.t.
(1 − |z| 2 )W ∂D = 0 at σ-a.e. point on ∂D,

and W(z) = o(1/|z| 2 ) as z → ∞ . (8.1.121)

Finally, the same characterization remains valid provided little “o” is replaced by
big “O” both in (8.1.120) and in (8.1.121).
From (8.1.10) and (8.1.117) we see that the quotient space between the full space
of boundary data L p (∂D, σ) and the space of admissible boundary data for the L p
Dirichlet Problem for the Bitsadze operator in the complement of the closed unit
disk (described in the left side of (8.1.117)) is isomorphic to the boundary Hardy
space H p (∂D+, σ), which is infinite dimensional. Also, since
 1 − |z| 2 
: k∈N (8.1.122)
z k+2
is an infinite dimensional subspace of (8.1.121), the last part in the statement of the
above theorem implies that
the space of null-solutions for the L p Dirichlet Problem for the
Bitsadze operator in the complement of the closed unit disk is (8.1.123)
infinite dimensional.
Here is the proof of Theorem 8.1.5:
Proof of Theorem 8.1.5 This is justified by reasoning much as in the proof of The-
orem 8.1.2, now using Theorem 8.1.4 in place of Theorem 8.1.1, and also keeping
in mind (8.1.15). 
904 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

Remark 8.1.6 As a result of the jump-formula valid for any double layer across the
boundary of a UR domain Ω ⊂ Rn , the sum of the spaces of admissible boundary
data for the L p Dirichlet Problems in Ω and Rn \ Ω, for any given weakly elliptic,
second-order, homogeneous, constant (complex) coefficient, M × M system in Rn is
 M
always the full space L p (∂Ω, σ) (where, as usual, σ := H n−1 ∂Ω).
In the case L = ∂z̄ , the Bitsadze operator in the plane (hence n = 2 and M = 1),
2

and with Ω := D, the open unit disk of the complex plane, we see from (8.1.70),
(8.1.117), and (8.1.10) that the sum of the two spaces of admissible boundary data
referred to in the previous paragraph is

H p (∂D+, σ) ⊕ z + H p (∂D−, σ) = L p (∂D, σ), (8.1.124)

as anticipated.
Finally, we are also in a position to identify the space of null-solutions and the
p
space of admissible boundary data for the L1 Regularity Problem for the Bitsadze
operator in the complement of the closed unit disk in the complex plane.
Theorem 8.1.7 Let D be the unit disk in C, and set σ := H 1 ∂D. Also, pick an
exponent p ∈ (1, ∞) and some aperture parameter κ ∈ (0, ∞). Then the space of
p
admissible boundary data for the L1 Regularity Problem for the Bitsadze operator
in the complement of the closed unit disk, i.e.,
 κ−n.t.   
u∂D : u ∈ 𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, Nκ (∇u) ∈ L p (∂D, σ), u(∞) = 0
(8.1.125)

is precisely the boundary Hardy space with regularity

p
H1 (∂D−, σ). (8.1.126)

Alternatively,
 κ−n.t.   
u∂D : u ∈ 𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, Nκ (∇u) ∈ L p (∂D, σ), u(∞) = 0
 ∫ 
f (z)z−k dσ(z) = 0 for all k ∈ N0 . (8.1.127)
p
= f ∈ L1 (∂D, σ) :
∂D

In addition, if in place of the vanishing condition at infinity one now assumes mere
boundedness for the function u, then in place of (8.1.125)-(8.1.127) one has
 κ−n.t.  
u∂D : u ∈ 𝒞∞ C \ D , ∂z̄2 u = 0 in C \ D, (8.1.128)

Nκ (∇u) ∈ L p (∂D, σ), and u(z) = O(1) as z → ∞
p
= H1 (∂D−, σ) ⊕ C
 ∫ 
f (z)z−k dσ(z) = 0 for all k ∈ N .
p
= f ∈ L1 (∂D, σ) :
∂D
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 905
p
Furthermore, the space of null-solutions for the L1 Regularity Problem for the
Bitsadze operator in the complement of the closed unit disk, i.e.,
  
u ∈ 𝒞∞ C \ D : ∂z̄2 u = 0 in C \ D, Nκ (∇u) ∈ L p (∂D, σ), (8.1.129)
κ−n.t. 
u(z) = o(1) as z → ∞, u∂D = 0 at σ-a.e. point on ∂D ,

coincides with
 
(1 − |z| 2 )W : W ∈ H p (D− ) and W(z) = o(1/|z| 2 ) as z → ∞ . (8.1.130)

Finally, the same identification remains valid if little “o” is replaced by big “O”
both in (8.1.129) and in (8.1.130).
 
Given any function u ∈ 𝒞∞ C \ D with Nκ (∇u) ∈ L p (∂D, σ), the result
κ−n.t.
established in [68, (8.9.236)] guarantees that the nontangential trace u∂D exist-
s at σ-a.e. point on ∂D. This observation is relevant in the context of (8.1.125) and
(8.1.128).
Theorem 8.1.7 has two significant consequences. First, the quotient space between
p
the full space of boundary data L1 (∂D, σ) and the space of admissible boundary data
p
for the L1 Regularity Problem for the Bitsadze operator in the complement of the
closed unit disk (described in (8.1.125)) is isomorphic to the boundary Hardy space
H p (∂D+, σ), which is infinite dimensional. Second, the space of null-solutions for
p
the L1 Regularity Problem for the Bitsadze operator in the complement of the closed
unit disk is infinite dimensional.
Proof of Theorem 8.1.7 This is justified by reasoning like in the proof of Theo-
rem 8.1.3, now employing Theorem 8.1.4 in lieu of Theorem 8.1.1, and also bearing
in mind (8.1.25). 

As a quick inspection of the proofs reveals, analogous results hold for other spaces
of boundary data:

Remark 8.1.8 Natural versions of Theorem 8.1.2, Theorem 8.1.3, Theorem 8.1.5,
and Theorem 8.1.7 are valid for Muckenhoupt weighted Lebesgue spaces, Morrey
spaces, vanishing Morrey spaces, block spaces, as well as for their Sobolev space
counterparts.

In the last portion of this section we indicate how our earlier results for the scalar
Bitsadze’s operator ∂z̄2 may be relatively painlessly “lifted” to genuine systems. To
set the stage, we first describe an abstract algebraic tool. Given any complex vector
space X of complex-valued functions defined on a common fixed set, we agree to
denote
    
Pairs [X] := Re f , Im f + i Re g, Im g : f , g ∈ X . (8.1.131)

Then this is itself a complex vector space, and we have the linear isomorphism
906 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

Pairs [X]  X ⊕ X via

    (8.1.132)
Re f , Im f + i Re g, Im g → f ⊕ g.

More generally, for any linear subspace Y of X we have the linear isomorphism

Pairs [X] X X
 ⊕ . (8.1.133)
Pairs [Y ] Y Y

Indeed, if [·]X/Y denotes the equivalence class in X/Y , then the linear assignment
   
Re f , Im f + i Re g, Im g → [ f ]X/Y ⊕ [g]X/Y (8.1.134)

maps Pairs [X] onto (X/Y ) ⊕ (X/Y ) and its kernel is precisely Pairs [Y ], so (8.1.133)
is a consequence of the First Group Isomorphism Theorem in algebra. Finally, we
wish to observe that
if X has the property that Re f ∈ X for each f ∈ X, then
(8.1.135)
in place of (8.1.132) one actually has Pairs [X] = X ⊕ X.

Consider now the second-order, homogeneous, real constant coefficient, 2 × 2

system  
1 ∂x − ∂y −2∂x ∂y
2 2
LB := . (8.1.136)
4 2∂x ∂y ∂x2 − ∂y2
The characteristic matrix of the system L is given by
 
−1 ξ1 − ξ2 −2ξ1 ξ2
2 2
LB (ξ) = at each ξ = (ξ1, ξ2 ) ∈ R2 . (8.1.137)
4 2ξ1 ξ2 ξ12 − ξ22

Hence, at each ξ = (ξ1, ξ2 ) ∈ R2 \ {0} we have

 2 
det [LB (ξ)] = 16
1
(ξ1 − ξ22 )2 + (2ξ1 ξ2 )2 = 16
1
(ξ12 + ξ22 )2 = 16 |ξ |
1 4
 0, (8.1.138)

which goes to show that

the system LB from (8.1.136) is weakly elliptic. (8.1.139)

The system LB defined in (8.1.136) is closely related to Bitsadze’s operator LB := ∂z̄2 .

Specifically, if we denote by π1, π2 : C2 → C the canonical coordinate projections,
defined as

π1 (z1, z2 ) := z1 and π2 (z1, z2 ) = z2 for each (z1, z2 ) ∈ C2, (8.1.140)

then
8.1 Null-Solutions and Boundary Traces for Bitsadze’s Operator ∂z2 in the Unit Disk 907

∂z̄2 (u1 + iu2 ) = π1 LB (u1, u2 ) + iπ2 LB (u1, u2 )

for any open set Ω ⊆ R2 ≡ C and any two (8.1.141)
complex-valued functions u1, u2 ∈ 𝒞2 (Ω).
In particular,    
LB Re u, Im u = Re (∂z̄2 u), Im (∂z̄2 u)
for any open set Ω ⊆ R2 ≡ C and any (8.1.142)
complex-valued function u ∈ 𝒞2 (Ω).
To continue, fix an integrability exponent p ∈ (1, ∞) and an aperture parameter
κ ∈ (0, ∞). We may then translate Theorem 8.1.2 using the above formalism and
obtain the following characterization of the space of admissible boundary data for
the L p Dirichlet Problem for the system LB defined in (8.1.136) in the unit disk in
the plane:
 κ−n.t.  2
U ∂D : U ∈ 𝒞∞ (D) , LB U = 0 in D, Nκ U ∈ L p (∂D, σ), (8.1.143)
κ−n.t.   
and U ∂D exists σ-a.e. on ∂D = Pairs H p (∂D+, σ) ⊕ z̄ .

Also, from (8.1.72)-(8.1.73) we see that the space of null-solutions for the L p
Dirichlet Problem for the system LB in the unit disk, i.e.,
  2
U ∈ 𝒞∞ (D) : LB U = 0 in D, Nκ U ∈ L p (∂D, σ),
κ−n.t. 
and U ∂D = 0 at σ-a.e. point on ∂D ,
(8.1.144)

coincides with the infinite dimensional space (1 − |z|)2 · Pairs [X] where
  
X := W holomorphic in D : Nκ δ∂D · W ∈ L p (∂D, σ) and
  κ−n.t. 
δ∂D · W ∂D = 0 at σ-a.e. point on ∂D .
(8.1.145)

Likewise, from Theorem 8.1.3 we see that the space of admissible boundary data
p
for the L1 Regularity Problem for the system LB in the unit disk has the following
description:
 κ−n.t.  2 
U ∂D : U ∈ 𝒞∞ (D) , LB U = 0 in D, Nκ (∇U) ∈ L p (∂D, σ) (8.1.146)
 
p
= Pairs H1 (∂D+, σ) ⊕ z̄ ,

p
while the space of null-solutions for the L1 Regularity Problem for system LB in the
unit disk, i.e.,
908 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
  2
U ∈ 𝒞∞ (D) : LB U = 0 in D, Nκ (∇U) ∈ L p (∂D, σ),
κ−n.t. 
and U ∂D = 0 at σ-a.e. point on ∂D ,
(8.1.147)

coincides with
  
(1 − |z| 2 )W : W ∈ Pairs H p (D) . (8.1.148)

Of course, similar results may be obtained in relation to Theorem 8.1.5 and Theo-
rem 8.1.7. In addition, analogous results are valid for the transpose of the system
LB , namely for  
1 ∂x − ∂y 2∂x ∂y
2 2

LB = . (8.1.149)
4 −2∂x ∂y ∂x2 − ∂y2
Indeed, these are naturally derived from the corresponding results for LB recorded
above keeping in mind that

LB (u1, u2 ) = π1 LB (u1, −u2 ), −π2 LB (u1, −u2 )

for any open set Ω ⊆ R2 ≡ C and any two (8.1.150)
complex-valued functions u1, u2 ∈ 𝒞2 (Ω);

in particular,
LB (u1, u2 ) = 0 ⇐⇒ LB (u1, −u2 ) = 0
for any open set Ω ⊆ R2 ≡ C and any (8.1.151)
complex-valued functions u1, u2 ∈ 𝒞2 (Ω).
Our final comment has to do with yet another manner of transferring results from
the scalar Bitsadze operator to genuine systems. For example, we consider
 
∂z̄2 0
LB := (8.1.152)
0 ∂z̄2

then LB is a second-order, homogeneous, complex constant coefficient, 2 × 2 weakly

elliptic system. Keeping in mind the simple relationship this system bears to the
scalar Bitsadze operator, we see from Theorem 8.1.2 that the space of admissible
boundary data for the L p Dirichlet Problem for the system LB in the unit disk in the
plane may be characterized as follows:
 κ−n.t.  2
U ∂D : U ∈ 𝒞∞ (D) , LB U = 0 in D, Nκ U ∈ L p (∂D, σ), (8.1.153)
κ−n.t.    2
and U ∂D exists σ-a.e. on ∂D = H p (∂D+, σ) ⊕ z̄ ,
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 909

while the space of null-solutions for the L p Dirichlet Problem for the system LB in
the unit disk, i.e.,
  2
U ∈ 𝒞∞ (D) : LB U = 0 in D, Nκ U ∈ L p (∂D, σ),
κ−n.t. 
and U ∂D = 0 at σ-a.e. point on ∂D , (8.1.154)
 
coincides with the infinite dimensional space U = (u1, u2 ) : u1, u2 ∈ X where
  
X := (1 − |z|)2W : W holomorphic in D : Nκ δ∂D · W ∈ L p (∂D, σ)
  κ−n.t. 
and δ∂D · W ∂D = 0 at σ-a.e. point on ∂D .
(8.1.155)

Similarly, Theorem 8.1.3 readily implies that the space of admissible boundary
p
data for the L1 Regularity Problem for the system LB in the unit disk may be
described as
 κ−n.t.  2 
U ∂D : U ∈ 𝒞∞ (D) , LB U = 0 in D, Nκ (∇U) ∈ L p (∂D, σ) (8.1.156)
 2
p
= H1 (∂D+, σ) ⊕ z̄ ,

p
whereas the space of null-solutions for the L1 Regularity Problem for system LB in
the unit disk, i.e.,
  2
U ∈ 𝒞∞ (D) : LB U = 0 in D, Nκ (∇U) ∈ L p (∂D, σ),
κ−n.t. 
and U ∂D = 0 at σ-a.e. point on ∂D , (8.1.157)

coincides with
  2
(1 − |z| 2 )W : W ∈ H p (D) . (8.1.158)

As expected, analogous results for the system LB may be derived now starting from
Theorem 8.1.5 and Theorem 8.1.7.

8.2 Null-Solutions and Boundary Traces for the Operator

∂2z − λ 2 ∂2z

Work in the complex plane. Bring in the Cauchy-Riemann operator ∂z̄ := 12 (∂x +i∂y ),
and denote by ∂z := 12 (∂x − i∂y ) its complex conjugate. We are interested in studying
the family of scalar, homogeneous, constant (complex) coefficient, second-order
operators
910 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
  
Lλ := ∂z̄2 − λ2 ∂z2 = ∂z̄ + λ∂z ∂z̄ − λ∂z , ∀λ ∈ C. (8.2.1)

In real variables we therefore have

 
Lλ = 14 (1 − λ2 )∂x2 + 2i(1 + λ2 )∂x ∂y − (1 − λ2 )∂y2 , ∀λ ∈ C. (8.2.2)

Hence, for each λ ∈ C we may express Lλ as

 
1 1 − λ2 2i(1 + λ2 )
Lλ = divA∇ where A := . (8.2.3)
4 2i(1 + λ2 ) −(1 − λ2 )

From [70, Example 1.4.25] we then see (bearing in mind [70, (1.4.178)]) that, for
any given λ ∈ C,
Lλ does not satisfy the Legendre-Hadamard (strong) ellipticity condi-
(8.2.4)
tion, and Lλ is weakly elliptic if and only if λ  ±i and λλ2 +1
2
−1
∈ C \ iR.

Next, for each fixed λ ∈ C consider the transformations

Tλ± : C −→ C given by
(8.2.5)
Tλ± (z) := z ± λz for each z ∈ C.

Note that if ζ = z ± λz then −ζ ± λζ = (|λ| 2 − 1)z. From this we conclude that

Tλ± is invertible if and only if |λ|  1, a scenario in which

 ±  −1 (8.2.6)
Tλ (ζ) = 1−1|λ|2 ζ ∓ 1−λ|λ|2 ζ for each ζ ∈ C.

Henceforth we agree to denote by 𝒪(Ω) the collection of all holomorphic functions

in a given open set Ω ⊆ C.

Lemma 8.2.1 Suppose Ω ⊆ C is an arbitrary open set and consider some function
w ∈ 𝒞1 (Ω). Also, pick some λ ∈ C with |λ|  1. Then
 
∂z̄ ± λ∂z w = 0 in Ω (8.2.7)

if and only if
 
w = ϕ ◦ Tλ∓ for some function ϕ ∈ 𝒪 (Tλ∓ )−1 Ω . (8.2.8)

Proof Using the Chain Rule involving a holomorphic function (recalled  in [70,
(1.4.146)]) it may be checked without any difficulty that for each ϕ ∈ 𝒪 (Tλ∓ )−1 Ω
we have   
∂z̄ ± λ∂z ϕ(z ∓ λz) = 0 for each z ∈ Ω. (8.2.9)
This proves the left-pointing implication in the statement. In the opposite direction,
given any w ∈ 𝒞1 (Ω) satisfying (8.2.7), if for each ζ ∈ (Tλ∓ )−1 (Ω) we define
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 911

 1 λ
ϕ(ζ) := w Tλ∓ )−1 (ζ) = w ζ± ζ , (8.2.10)
1 − |λ| 2 1 − |λ| 2
then
 a patient
 application of the ordinary Chain Rule shows  that ϕ satisfies
∂ζ¯ ϕ(ζ) = 0 for each ζ ∈ (Tλ∓ )−1 (Ω). Hence ϕ ∈ 𝒪 (Tλ∓ )−1 Ω , as wanted. 

We are now prepared to prove the following “structure theorem” concerning the
null-solutions of the operator Lλ from (8.2.1).

Theorem 8.2.2 Let Ω ⊆ C be a simply connected1 set, and fix some λ ∈ C \ {0}

with |λ|  1. Also, consider a function u ∈ 𝒞2 (Ω). Then Lλ u = 0 in Ω if and only if
there exist two functions,
   
ϕ ∈ 𝒪 (Tλ− )−1 Ω and ψ ∈ 𝒪 (Tλ+ )−1 Ω , (8.2.11)

such that
u = ϕ ◦ Tλ− + ψ ◦ Tλ+ in Ω. (8.2.12)
In addition, the above holomorphic functions ϕ, ψ are uniquely determined by u
up to additive constants and, in fact,
  
ϕ  = − 2λ
1
∂z̄ − λ∂z u ◦ (Tλ− )−1,
   (8.2.13)
ψ  = 2λ
1
∂z̄ + λ∂z u ◦ (Tλ+ )−1 .

Proof In one direction, suppose u ∈ 𝒞2 (Ω) has Lλ u = 0 in Ω. Then

 
w := ∂z̄ − λ∂z u ∈ 𝒞1 (Ω) (8.2.14)

satisfies  
∂z̄ + λ∂z w = Lλ u = 0 in Ω. (8.2.15)
 
Lemma 8.2.1 then guarantees the existence of a function Φ ∈ 𝒪 (Tλ− )−1 Ω such that

w = Φ ◦ Tλ− in Ω. (8.2.16)

Given that we are currently assuming λ  0, and since (Tλ− )−1 (Ω) is a simply
connected set, we may invoke the equivalence between items (1) and (7) in [68,
Proposition 5.8.1] to conclude that there exists some
 
ϕ ∈ 𝒪 (Tλ− )−1 Ω satisfying ϕ  = (−2λ)−1 Φ in (Tλ− )−1 (Ω), (8.2.17)

where ‘prime’ stands for the ordinary complex derivative of holomorphic functions.
Granted this, an application of the Chain Rule involving a holomorphic function
(recalled in [70, (1.4.146)]) presently gives

1 hence, Ω is also open and connected

912 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
  
∂z̄ − λ∂z ϕ(z − λz) = (−2λ)ϕ (z − λz) = Φ(z − λz) = w(z)
  
= ∂z̄ − λ∂z u (z) for each z ∈ Ω. (8.2.18)

Hence,   
∂z̄ − λ∂z u(z) − ϕ(z − λz) = 0 for each z ∈ Ω, (8.2.19)
so yet another application
 of Lemma 8.2.1 proves the existence of some function
ψ ∈ 𝒪 (Tλ+ )−1 Ω such that

u(z) − ϕ(z − λz) = ψ(z + λz) for each z ∈ Ω. (8.2.20)

From this, (8.2.12) follows.

In the opposite direction, given any functions ϕ, ψ as in (8.2.11), Lemma 8.2.1
tells us that for each z ∈ Ω we have
     
∂z̄ + λ∂z ϕ(z − λz) = 0 and ∂z̄ − λ∂z ψ(z + λz) = 0. (8.2.21)

In view of this and the factorization in (8.2.1) we then conclude that

 
Lλ ϕ(z − λz) + ψ(z + λz) = 0 for each z ∈ Ω, (8.2.22)

as wanted. There remains to prove the very last claim in the statement. In this regard
we note that if (8.2.12) holds then
        
∂z̄ + λ∂z u (z) = ∂z̄ + λ∂z ϕ(z − λz) + ∂z̄ + λ∂z ψ(z + λz)

= 2λψ (z + λz) for each z ∈ Ω. (8.2.23)

Thus,
  
ψ (z + λz) = 1
2λ ∂z̄ + λ∂z u (z) for each z ∈ Ω (8.2.24)

and, likewise,
  
ϕ (z − λz) = −1
2λ ∂z̄ − λ∂z u (z) for each z ∈ Ω. (8.2.25)

These formulas prove (8.2.13) and also make it clear that the holomorphic functions
ϕ, ψ are uniquely determined by u up to additive constants. 
We are now in prepared to describe both the space of admissible boundary data
and the space of null-solutions for the Homogeneous Regularity Problem for the
operator Lλ in the upper half-plane.
Theorem 8.2.3 Fix an aperture parameter κ > 0 along with an integrability expo-
nent p ∈ (1, ∞) and define the homogeneous Hardy spaces with regularity (as in
(1.8.285) for the upper half-plane)
.p  κ−n.t. 
H1,± (R, L 1 ) := w ∂C± : w ∈ 𝒪(C± ) with Nκ (∇w) ∈ L p (R, L 1 ) . (8.2.26)
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 913

Also, pick a real number

λ ∈ (−1, 1) \ {0}. (8.2.27)
Then the space of admissible boundary data for the Homogeneous Regularity.p
Problem for the operator Lλ in the upper half-plane, i.e., the subspace of L1 (R, L 1 )
described as
.p  κ−n.t. 
𝒰1,λ := u∂C+ : u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ) , (8.2.28)
.p
coincides with H1,+ (R, L 1 ) (the “positive” homogeneous Hardy space with regular-
.p 
ity on the real line). Moreover, the cokernel of 𝒰1,λ ∼ (i.e., the space of equivalence
.p
classes of function in 𝒰1,λ modulo constants) into the full space of boundary data
.p 
modulo constants, L1 (R, L 1 ) ∼, is isomorphic to the “negative” homogeneous
.p 
Hardy space with regularity modulo constants, H1,− (R, L 1 ) ∼, i.e.,
.p 
L1 (R, L 1 ) ∼ .p 
.p   H1,− (R, L 1 ) ∼ . (8.2.29)
𝒰1,λ ∼

In particular,
the codimension of the space of admissible boundary data, modulo
constants, for the Homogeneous Regularity. Problem for the operator
p  (8.2.30)
Lλ in the upper half-plane (i.e., the space 𝒰1,λ ∼) into the full space
.p 
of boundary data modulo constants, L1 (R, L 1 ) ∼, is +∞.

Finally, the space on null-solutions for the Homogeneous Regularity Problem for
the operator Lλ in the upper half-plane, namely
 κ−n.t. 
u ∈ 𝒞∞ (C+ ) : Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ), u∂C+ = 0 , (8.2.31)

coincides with the infinite dimensional space

    
φ (1 + λ)(z − λz) − φ (1 − λ)(z + λz) : φ ∈ 𝒪(C+ )

with Nκ (∇φ) ∈ L p (R, L 1 ) . (8.2.32)

The fact that the nontangential boundary trace exists in the context of (8.2.28) and
.p .p
that 𝒰1,λ is indeed a subspace of L1 (R, L 1 ) is seen from [69, Proposition 11.5.12].
Also, the space (8.2.32) is, as claimed, infinite dimensional. For example, this may
be seen by considering the holomorphic functions

φk (z) := (i + z)−k for each z ∈ C+ with k ∈ N arbitrary. (8.2.33)

From [68, Lemma 8.3.7] we see that Nκ (∇φk ) ∈ L p (R, L 1 ) for each k ∈ N, and we
claim that
914 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
    
φk (1 + λ)(z − λz) − φk (1 − λ)(z + λz)
k ∈N (8.2.34)
is a family of linearly independent functions in C+ .

Indeed, if for some integer N ∈ N and some coefficients ck ∈ C, with 1 ≤ k ≤ N,

we have
N     
ck φk (1 + λ)(z − λz) − φk (1 − λ)(z + λz) = 0, ∀z ∈ C+, (8.2.35)
k=1

then applying ∂z̄ − λ∂z we obtain

N
2λ(1 + λ)k ck
  k+1 = 0, ∀z ∈ C+, (8.2.36)
k=1 i + (1 + λ)(z − λz)

hence also
N
  N −k
k ck i + (1 + λ)(z − λz) = 0, ∀z ∈ C+ . (8.2.37)
k=1

Inspecting the coefficients of z (from highest to lowest) in this polynomial equation

reveals that all ck ’s vanish. As such, (8.2.34) follows, so the space (8.2.32) is infinite
dimensional.
We now give the proof of Theorem 8.2.3.
.p
Proof of Theorem 8.2.3 In one direction, given any f ∈ H1,+ (R, L 1 ) there exists a
function
w ∈ 𝒪(C+ ) with Nκ (∇w) ∈ L p (R, L 1 ) and
κ−n.t. (8.2.38)
w ∂C+ (z) = f (z) at L 1 -a.e. point z ∈ ∂C+ ≡ R.

If we define
z − λz
u : C+ → C+, u(z) := w for each z ∈ C+, (8.2.39)
1−λ
∞
then u ∈ 𝒞 (C+ ) and the Chain Rule involving a holomorphic function (recalled in
[70, (1.4.146)]) gives  
∂z̄ ± λ∂z u = 0 in C+ . (8.2.40)
In particular, Lλ u = 0 in C+ (cf. (8.2.1)). Also, in view of the fact that

z − λz
C+  z → ∈ C+ is a bi-Lipschitz homeomorphism, (8.2.41)
1−λ
the properties listed in (8.2.38) together with [68, Lemma 8.1.7] and [68, Proposi-
tion 8.9.8] help us conclude that Nκ (∇u) ∈ L p (R, L 1 ) and
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 915

κ−n.t. κ−n.t. z − λz κ−n.t.

u∂C+ (z) = w ∂C+ = w ∂C+ (z)
1−λ
= f (z) at L 1 -a.e. point z ∈ ∂C+ ≡ R, (8.2.42)

since z = z on ∂C+ ≡ R. All these ultimately put f in the space of admissible

boundary data for the Homogeneous Regularity Problem for the operator Lλ in the
.p
upper half-plane, described in (8.2.28). The conclusion is that the space H1,+ (R, L 1 )
embeds into (8.2.28).
To prove the opposite inclusion, start with a function satisfying

u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ). (8.2.43)

Based on these properties, Theorem 8.2.2 (used with Ω := C+ and λ as in (8.2.27)),

the fact that the transformations2
Tλ± : C+ → C+ given by Tλ± (z) := z ± λz for all z ∈ C+
(8.2.44)
are bi-Lipschitz homeomorphisms,

[68, Lemma 8.1.7], and [68, Proposition 8.4.1], we then conclude that there exist
two functions

ϕ, ψ ∈ 𝒪(C+ ) with Nκ (∇ϕ), Nκ (∇ψ) ∈ L p (R, L 1 ), (8.2.45)

and such that

u(z) = ϕ(z − λz) + ψ(z + λz) for each z ∈ C+ . (8.2.46)
If we now define
Φ, Ψ : C+ −→ C+ given at each z ∈ C+
    (8.2.47)
by Φ(z) := ϕ (1 − λ)z and Ψ(z) := ψ (1 + λ)z ,

then
Φ, Ψ ∈ 𝒪(C+ ) and Nκ (∇Φ), Nκ (∇Ψ) ∈ L p (R, L 1 ). (8.2.48)
In particular (cf. (8.2.26)),
κ−n.t. κ−n.t. .p
Φ∂C+ and Ψ∂C+ belong to H1,+ (R, L 1 ). (8.2.49)

Also, once again bearing in mind that z = z on ∂C+ ≡ R, we see from (8.2.44)-
(8.2.48), [68, Lemma 8.1.7], and [68, Proposition 8.9.8] that

2 it is not difficult to see that the transformations Tλ± map the upper half-plane C+ bijectively onto
itself if and only if λ ∈ (−1, 1)
916 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
κ−n.t. κ−n.t.   κ−n.t.  
u∂C+ (z) = ϕ∂C+ (1 − λ)z + ψ ∂C+ (1 + λ)z
κ−n.t. κ−n.t.
= Φ∂C+ (z) + Ψ∂C+ (z) at L 1 -a.e. point z ∈ ∂C+ ≡ R (8.2.50)

which, in concert with (8.2.49), proves that

κ−n.t. κ−n.t. κ−n.t. .p
u∂C+ = Φ∂C+ + Ψ∂C+ ∈ H1,+ (R, L 1 ). (8.2.51)
.p
From this we then conclude that the space (8.2.28) is contained in H1,+ (R, L 1 ).
.p
The coincidence of the space (8.2.28) with H1,+ (R, L 1 ) has therefore been estab-
lished. Granted this, the isomorphism claimed in (8.2.29) becomes a consequence
of (1.8.286). In turn, (8.2.29) readily implies (8.2.30) since the homogeneous Hardy
.p
space with regularity H1,− (R, L 1 ) is infinite dimensional.
Next, given any φ ∈ 𝒪(C+ ) with Nκ (∇φ) ∈ L p (R, L 1 ) it follows from Theo-
rem 8.2.2 and [68, Lemma 8.1.7] that the function

u : C+ → C defined for each z ∈ C+ by

    (8.2.52)
u(z) := φ (1 + λ)(z − λz) − φ (1 − λ)(z + λz)

satisfies

u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ). (8.2.53)

In addition, [69, Proposition 11.5.12] and [68, Lemma 8.1.7] allow us to compute
 κ−n.t.   κ−n.t.     κ−n.t.   
u∂C+ (z) = φ∂C+ (1 + λ)(z − λz) − φ∂C+ (1 − λ)(z + λz)
 κ−n.t.     κ−n.t.   
= φ∂C+ (1 + λ)(1 − λ)z − φ∂C+ (1 − λ)(1 + λ)z

= 0 at L 1 -a.e. point z ∈ R ≡ ∂C+, (8.2.54)

since complex conjugation leaves points in R ≡ ∂C+ invariant. Thus, the space
described in (8.2.32) is contained in the space on null-solutions for the Homogeneous
Regularity Problem for the operator Lλ in the upper half-plane, explicitly recorded
in (8.2.31).
To establish the opposite inclusion, start with a function u satisfying
κ−n.t.
u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ), u∂C+ = 0. (8.2.55)

The structural result proved in Theorem 8.2.2 guarantees (also bearing in mind [68,
Lemma 8.1.7] and [68, Proposition 8.4.1]) that

u(z) = ϕ(z − λz) + ψ(z + λz) for each z ∈ C+ for some

(8.2.56)
ϕ, ψ ∈ 𝒪(C+ ) with Nκ (∇ϕ), Nκ (∇ψ) ∈ L p (R, L 1 ).
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 917

If we now introduce
Φ, Ψ : C+ −→ C+ given at each z ∈ C+
    (8.2.57)
by Φ(z) := ϕ (1 − λ)z and Ψ(z) := ψ (1 + λ)z ,

then
Φ, Ψ ∈ 𝒪(C+ ) satisfy Nκ (∇Φ), Nκ (∇Ψ) ∈ L p (R, L 1 ), (8.2.58)
and [68, Lemma 8.1.7] permit us to compute, at L 1 -a.e. point z ∈ R ≡ ∂C+ ,
 κ−n.t.   κ−n.t.   κ−n.t. 
0 = u∂C+ (z) = ϕ∂C+ (z − λz) + ψ ∂C+ (z + λz)
 κ−n.t.     κ−n.t.   
= ϕ∂C+ (1 − λ)z + ψ ∂C+ (1 + λ)z
 κ−n.t.   κ−n.t. 
= Φ∂C+ (z) + Ψ∂C+ (z), (8.2.59)

hence κ−n.t.
(Φ + Ψ)∂C+ = 0 at L 1 -a.e. point on R ≡ ∂C+ . (8.2.60)
From this and the integral representation formula (1.8.236) from Corollary 1.8.23
we first conclude that Φ + Ψ is a constant in C+ , then another appeal to (8.2.60) gives
that actually
Φ + Ψ = 0 in C+ . (8.2.61)
If we now define
z
φ(z) := Φ for each z ∈ C+, (8.2.62)
1 − λ2
it follows from (8.2.62), (8.2.56), (8.2.57), (8.2.58), (8.2.61), [68, Lemma 8.1.7],
and [68, Proposition 8.4.1] that φ ∈ 𝒪(C+ ) has Nκ (∇φ) ∈ L p (R, L 1 ) and
   
u(z) = φ (1 + λ)(z − λz) − φ (1 − λ)(z + λz) for each z ∈ C+ . (8.2.63)

This shows that (8.2.31) is indeed included in the space described in (8.2.32). 

We continue by presenting the following counterpart to Theorem 8.2.3, describing

the space of admissible boundary data as well as the space of null-solutions for the
Dirichlet Problem for the operator Lλ in the upper half-plane. The reader is reminded
that the Hilbert transform
 on the real line is the singular integral operator acting on
each function f ∈ L 1 R, 1+dx|x | according to
∫
1 f (y)
H f (x) := lim+ dy for L 1 -a.e. x ∈ R. (8.2.64)
ε→0 π x−y
y ∈R
|x−y |>ε

Theorem 8.2.4 Pick some aperture parameter κ > 0 together with some integrability
exponent p ∈ (1, ∞) and recall the ordinary Hardy spaces on the real line
918 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
 κ−n.t. 
H± (R, L 1 ) = w ∂C± : w ∈ 𝒪(C± ) with Nκ w ∈ L p (R, L 1 )
p

 
= f ∈ L p (R, L 1 ) : H f = ∓i f , (8.2.65)

where H is the Hilbert transform on the real line (cf. (8.2.64)). Finally, pick a real
number
λ ∈ (−1, 1) \ {0}. (8.2.66)
Then the space of admissible boundary data for the L p Dirichlet Problem for the
operator Lλ in the upper half-plane, i.e., the subspace of L p (R, L 1 ) described as
 κ−n.t.
𝒰λ := u∂C+ : u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+,
p

κ−n.t. 
Nκ u ∈ L p (R, L 1 ) and u∂C+ exists L 1 -a.e. on R , (8.2.67)

p
coincides with the “positive” Hardy space H+ (R, L 1 ). Moreover, the cokernel of
p
𝒰λ into the full space of boundary data L p (R, L 1 ) is isomorphic to the “negative”
Hardy space H−p (R, L 1 ), i.e.,

L p (R, L 1 )
p  H−p (R, L 1 ). (8.2.68)
𝒰λ

In particular,
the codimension of the space of admissible boundary data for the L p
Dirichlet Problem for the operator Lλ in the upper half-plane (i.e., the (8.2.69)
p
space 𝒰λ ) into the full space of boundary data L p (R, L 1 ) is +∞.

Finally, the space on null-solutions for the L p Dirichlet Problem for the operator
Lλ in the upper half-plane, namely
 κ−n.t. 
u ∈ 𝒞∞ (C+ ) : Lλ u = 0 in C+, Nκ (∇u) ∈ L p (R, L 1 ), u∂C+ = 0 , (8.2.70)

is infinite dimensional.

Proof Let u be a complex-valued function defined in the upper half-plane and

satisfying

u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ u ∈ L p (R, L 1 ),

κ−n.t. (8.2.71)
and u∂C+ exists at L 1 -a.e. point on R.

If we define
uε (z) := u(z + iε) for each z ∈ C+ (8.2.72)
then, based on (8.2.71)-(8.2.72) and interior estimates for null-solutions of the weakly
elliptic operator Lλ (also bearing in mind [68, Proposition 8.4.1]), for each ε > 0
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 919

the following properties are satisfied:

 
uε ∈ 𝒞∞ C+ , Lλ uε = 0 in C+, Nκ (∇uε ) ∈ L p (R, L 1 ),
   (8.2.73)
and uε   ≤ Nκ u at every point on R.
∂C+

From Lebesgue’s Dominated Convergence Theorem we also see that

 κ−n.t.
fε := uε ∂C+ converges in L p (R, L 1 ) to u∂C+ as ε → 0+ . (8.2.74)

Note that each fε belongs to the space of admissible boundary data for the Homoge-
neous Regularity Problem for the operator Lλ in the upper half-plane, i.e., the space
.p
𝒰1,λ described in (8.2.28). According to identification made in Theorem 8.2.3, each
fε therefore belongs to the “positive” homogeneous Hardy space with regularity
.p
H1,+ (R, L 1 ). In view of this and (1.8.288) we then conclude that for each ε > 0 there
exists a constant cε ∈ C such that

Cmod fε = 1
2 fε + cε . (8.2.75)

On the other hand, since fε ∈ L p (R, L 1 ), much as in (1.8.26) we have that Cmod fε
differs from (i/2)H fε by an additive constant. Ultimately, this implies (bearing in
mind that H maps L p (R, L 1 ) into itself and any constant function in L p (R, L 1 ) is
actually identically zero) that (8.2.75) simply reduces to

H fε = −i fε for each ε > 0. (8.2.76)

Passing to limit ε → 0+ and keeping in mind (8.2.74) plus the boundedness of H on

L p (R, L 1 ), we then conclude that
 κ−n.t.  κ−n.t. 
H u∂C+ ) = −i u∂C+ . (8.2.77)
κ−n.t.
In light of (8.2.65), this places the function u∂C+ in the “positive” Hardy spaces
p p p
H+ (R, L 1 ). In summary, 𝒰λ ⊆ H+ (R, L 1 ).
p
To justify the opposite inclusion, start with an arbitrary f ∈ H+ (R, L 1 ). This
means that there exists a function
w ∈ 𝒪(C+ ) with Nκ w ∈ L p (R, L 1 ) and
κ−n.t. (8.2.78)
w ∂C+ (z) = f (z) at L 1 -a.e. point z ∈ ∂C+ ≡ R.

Use this to define

z − λz
u : C+ → C+, u(z) := w for each z ∈ C+ . (8.2.79)
1−λ
Then u ∈ 𝒞∞ (C+ ) and the Chain Rule involving a holomorphic function (cf. [70,
(1.4.146)]) presently gives
920 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane
 
∂z̄ ± λ∂z u = 0 in C+ . (8.2.80)

Hence, Lλ u = 0 in C+ (cf. (8.2.1)). Moreover, in view of the fact that

z − λz
C+  z → ∈ C+ is a bi-Lipschitz homeomorphism, (8.2.81)
1−λ
the properties listed in (8.2.78) together with [68, Lemma 8.1.7] and [68, Proposi-
tion 8.9.8] permit us to conclude that Nκ u ∈ L p (R, L 1 ) and
κ−n.t. κ−n.t. z − λz κ−n.t.
u∂C+ (z) = w ∂C+ = w ∂C+ (z)
1−λ
= f (z) at L 1 -a.e. point z ∈ ∂C+ ≡ R, (8.2.82)
p
since z = z on ∂C+ ≡ R. As a result, f belongs to 𝒰λ , the space of admissible
boundary data for the L p Dirichlet Problem for the operator Lλ in the upper half-
p p
plane. The conclusion is that the space H+ (R, L 1 ) embeds into 𝒰λ .
The reasoning above proves that
p p
𝒰λ = H+ (R, L 1 ). (8.2.83)
p
Recall that the Hardy spaces H± (R, L 1 ) are closed subspaces of L p (R, L 1 ) and we
have the direct sum decomposition
p
L p (R, L 1 ) = H+ (R, L 1 ) ⊕ H−p (R, L 1 ). (8.2.84)

Then (8.2.68) is a consequence of (8.2.83) and (8.2.84). In particular, since

H−p (R, L 1 ) is infinite dimensional, the claim in (8.2.69) also follows.
Finally, that the space on null-solutions for the L p Dirichlet Problem for the oper-
ator Lλ in the upper half-plane, described in (8.2.70), is infinite dimensional is seen
from (8.2.34) (also bearing in mind [68, Lemma 8.1.7] and [68, Proposition 8.9.8]),
upon noting that each function φk defined as in (8.2.33) satisfies Nκ φk ∈ L p (R, L 1 )
(cf. [68, Lemma 8.3.7]). 

Lastly, here is a result complementing Theorems 8.2.3-8.2.4, describing both the

space of admissible boundary data and the space of null-solutions for the Inhomo-
geneous Regularity Problem for the operator Lλ in the upper half-plane.

Theorem 8.2.5 Select some aperture parameter κ > 0 together with some inte-
grability exponent p ∈ (1, ∞) and define the inhomogeneous Hardy spaces with
regularity
 κ−n.t. 
H1,± (R, L 1 ) := w ∂C± : w ∈ 𝒪(C± ) with Nκ w, Nκ (∇w) ∈ L p (R, L 1 )
p

 
p
= f ∈ L1 (R, L 1 ) : H f = ∓i f , (8.2.85)
8.2 Null-Solutions and Boundary Traces for the Operator ∂z2 − λ2 ∂z2 921

where H is the Hilbert transform on the real line (cf. (8.2.64)). Also, pick a real
number
λ ∈ (−1, 1) \ {0}. (8.2.86)
Then the space of admissible boundary data for the Inhomogeneous Regularity
p
Problem for the operator Lλ in the upper half-plane, i.e., the subspace of L1 (R, L 1 )
described as
 κ−n.t. 
𝒰1,λ := u∂C+ : u ∈ 𝒞∞ (C+ ), Lλ u = 0 in C+, Nκ u, Nκ (∇u) ∈ L p (R, L 1 ) ,
p

(8.2.87)
p
coincides with H1,+ (R, L 1 ) (the “positive” inhomogeneous Hardy space with reg-
p
ularity on the real line). Furthermore, the cokernel of 𝒰1,λ into the full space of
p
boundary data L1 (R, L 1 ) is isomorphic to the “negative” inhomogeneous Hardy
p
space with regularity H1,− (R, L 1 ), i.e.,
p
L1 (R, L 1 ) p
p  H1,− (R, L 1 ). (8.2.88)
𝒰1,λ

As a consequence,
the codimension of the space of admissible boundary data for the
Inhomogeneous Regularity Problem for the operator Lλ in the upper
p
half-plane (i.e., the space 𝒰1,λ ) into the full space of boundary data (8.2.89)
p
L1 (R, L ), is +∞.
1

Finally, the space on null-solutions for the Inhomogeneous Regularity Problem

for the operator Lλ in the upper half-plane, namely
 κ−n.t. 
u ∈ 𝒞∞ (C+ ) : Lλ u = 0 in C+, Nκ u, Nκ (∇u) ∈ L p (R, L 1 ), u∂C+ = 0 ,
(8.2.90)

coincides with the infinite dimensional space

    
φ (1 + λ)(z − λz) − φ (1 − λ)(z + λz) : φ ∈ 𝒪(C+ ) (8.2.91)

with Nκ φ, Nκ (∇φ) ∈ L p (R, L 1 ) .

Proof This follows from Theorems 8.2.3-8.2.4 and their proofs. 

We conclude with three remarks, the first of which ties up with the discussion in
Remark 8.1.6.

Remark 8.2.6 Similar results to those described in Theorems 8.2.3-8.2.5 are valid
with the upper half-plane C+ replaced by lower half-plane C− . For example, the
space of admissible boundary data for the L p Dirichlet Problem for the operator
922 8 Hardy Spaces for Second-Order Weakly Elliptic Operators in the Complex Plane

Lλ in the lower half-plane turns out to be the “negative” Hardy space H−p (R, L 1 ).
Together with the identification made in Theorem 8.2.4, this shows that the spaces
of admissible boundary data for the L p Dirichlet Problem for the operator Lλ in C±
p
are, respectively, H+ (R, L 1 ) and H−p (R, L 1 ). Remarkably, the latter Hardy spaces
sum up (directly) to the full space of boundary data L p (R, L 1 ) (cf. (8.2.84)). Related
considerations also hold in the case of the identifications made in Theorem 8.2.3
and Theorem 8.2.5.

Our second remark briefly elaborates

 on the nature of the operator Lλ considered
in the unit disk D := z ∈ C : |z| < 1 in the complex plane.

Remark 8.2.7 Pick an arbitrary complex number λ ∈ C. Then the space of null-
solutions for the classical Dirichlet Problem for the operator Lλ in the unit disk,
namely
    
u ∈ 𝒞∞ D : Lλ u = 0 in D and u∂D = 0 , (8.2.92)

is infinite dimensional. A key observation in this regard is that for any φ ∈ 𝒪(C) the
function u : D → C given by
   
u(z) := φ (z + λz)2 − 4λ − φ (z − λz)2 for each z ∈ D (8.2.93)

belongs to
 the
 space described in (8.2.92). Indeed, it is clear from this definition that
u ∈ 𝒞∞ D . Next, that Lλ u = 0 in D is seen from the factorization in (8.2.2) plus
the Chain
 Rule involving a holomorphic function recalled in [70, (1.4.146)]. Finally,
that u∂D = 0 follows from the observation that

if z ∈ ∂D then z = z −1 hence
(8.2.94)
(z + λz)2 − 4λ = (z + λ/z)2 − 4λ = (z − λ/z)2 = (z − λz)2 .

Choosing φ(ζ) := ζ k with k ∈ N arbitrary then yields an infinite family of functions

in the space (8.2.92), which are linearly independent if λ  0.

The final remark reads as follows:

Remark 8.2.8 Our earlier results pertaining to the failure of Fredholm solvability
for boundary value problems for the scalar operator Lλ may be “lifted” to genuine
systems by considering  
Lλ 0
Lλ := (8.2.95)
0 Lλ
which, for each or each λ ∈ (−1, 1) \ {0}, is a second-order, homogeneous, complex
constant coefficient, 2 × 2 weakly elliptic system.
Appendix A
Terms and notation used in Volume IV

A
Ahlfors regular domain (cf. [68, Definition 5.9.15]):

a nonempty open subset Ω of Rn such that ∂Ω is

(A.0.1)
an Ahlfors regular set and H n−1 (∂Ω \ ∂∗ Ω) = 0

Aκ (∂Ω), the κ-accessible (from within Ω) subset of ∂Ω (cf. [68, (8.8.2)]):

 
Aκ (∂Ω) := x ∈ ∂Ω : x ∈ Γκ (x) (A.0.2)

[w] A p , the characteristic of the weight w on a space of homogeneous type (X, ρ, μ)

(cf. [68, §7.7]):
⨏ ⨏  p−1
[w] A p := sup w dμ w −1/(p−1) dμ (A.0.3)
B ρ -ball B B

Ap (X, ρ, μ), the Muckenhoupt Ap -class on a space of homogeneous type (X, ρ, μ)

(cf. [68, §7.7]):
 
Ap (X, ρ, μ) := w weight function : [w] A p < ∞ (A.0.4)

A∞ (X, ρ, μ), the Muckenhoupt A∞ -class on a space of homogeneous type (X, ρ, μ)

(cf. [68, §7.7]): 
A∞ (X, ρ, μ) := Ap (X, μ) (A.0.5)
1≤p<∞

V ⊥ , the annihilator V ⊥ of a subspace V of a Banach space X:

 
V ⊥ := Λ ∈ X ∗ : Λ(x) = 0 for all x ∈ V (A.0.6)

⊥ W, the annihilator of a subspace W of X ∗ (where X is Banach):

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 923
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1
924 A Terms and notation used in Volume IV

⊥
 
W := x ∈ X : Λ(x) = 0 for all Λ ∈ W (A.0.7)

T ∗ , the adjoint of the operator T ∈ L(X → Y ):

T ∗ : Y ∗ −→ X ∗, T ∗ (Λ) := Λ ◦ T for each Λ ∈ Y ∗ (A.0.8)

A p , the absolutely p-convex hull of a subset A of a vector space X, with p ∈ (0, 1],
defined as (cf. [68, §7.8.3]):

M
A p := λ j v j : M ∈ N, {v j }1≤ j ≤M ⊆ A,
j=1

M
{λ j }1≤ j ≤M ⊆ C with |λ j | p ≤ 1 (A.0.9)
j=1

a ⊗ b, the tensor product of vectors a = (a j )1≤ j ≤ N ∈ C N and b = (bk )1≤k ≤M ∈ C M ,

defined as the N × M matrix

a ⊗ b := a j bk 1≤ j ≤ N (A.0.10)
1≤k ≤M

αβ
A , the transpose of a coefficient tensor A := ar s 1≤r,s ≤n (cf. [70, §2.7]):
1≤α,β ≤M

βα αβ
A sr := ar s for all α, β, r, s (A.0.11)

A∇u, the action of the coefficient tensor A on the Jacobian matrix ∇u, defined if
αβ M
A = ar s 1≤r,s ≤n u = (uβ )1≤β ≤ N ∈ D (Ω) as (cf. [70, §2.7]):
1≤α,β ≤M

αβ
A∇u := ar s ∂s uβ 1≤α ≤M (A.0.12)
1≤r ≤n

  αβ
A·, · , the bilinear form associated with the coefficient tensor A = ar s 1≤r,s ≤n
1≤α,β ≤M
(cf. [70, §2.7]):
  αβ β β
Aζ, η := ar s ζs ηrα for all ζ := (ζs )β,s ∈ C N ×n and η := (ηrα )α,r ∈ C M×n
(A.0.13)
1,1 n
A q,κ , the L q -based area-function in Rn \ Σ, acting on each u ∈ Wloc (R \ Σ) as (cf.
[70]):
∫
1/q
(A q,κ u)(x) := c
|(∇u)(y)| q |x − y| q−n dy , x∈Σ (A.0.14)
ΓΣκ (x)
A Terms and notation used in Volume IV 925

AWE (n, M), the class of weakly elliptic coefficient tensors, defined as the collection
αβ
of all coefficient tensors A = ar s 1≤α,β ≤M with complex entries, with the property
1≤r,s ≤n
that the M × M homogeneous second-order system L A associated with A in Rn is
weakly elliptic.
Aα,β , the boundary-to-domain integral operator acting on any given differential form
f ∈ L 1 ∂Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β+1 at each z ∈ Ω as in (7.3.38):
∫
 
Aα,β f (z) := − f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
C
∂Ω

Aα,β , the boundary-to-boundary integral operator acting on any given differential

form f ∈ L 1 ∂Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β+1 at σ-a.e. z ∈ ∂Ω as in (7.3.39):
∫
 
Aα,β f (z) := − lim+ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ
ε→0 C
ζ ∈∂Ω
|z−ζ |>ε

p,q
As (Ω), the Besov/Triebel-Lizorkin space in the open set Ω ⊆ Rn (with A := B
corresponding to Besov spaces, and with A := F corresponding to Triebel-Lizorkin
spaces), with 0 < p, q ≤ ∞ and s ∈ R, defined as (cf. [69, §9.2]):
  
As (Ω) := u ∈ D (Ω) : there exists U ∈ As (Rn ) such that U Ω = u (A.0.15)
p,q p,q

and equipped with the quasi-norm

 
(Rn ), U Ω = u
p,q
u Ap, q
s (Ω)
:= inf U  Ap, q n : U ∈ As
s (R )
(A.0.16)


As (Ω)bdd , the space of all distributions u in Ω such that ψ Ω u ∈ As (Ω) for each
p,q p,q
∞
cutoff function ψ ∈ 𝒞c (R ) (cf. [69, Convention 8.3.7] and (A.0.217)).
n

B
Bρ (x, r), the ρ-ball with center at x ∈ X and radius r > 0 in the quasi-metric space
(X, ρ) (cf. [68, §7.1]):

Bρ (x, r) := {y ∈ X : ρ(x, y) < r } (A.0.17)

 
Bn−1 (x , r) := y  ∈ Rn−1 : |y  − x  | < r , the (n − 1)-dimensional (open) ball in
Rn−1 centered at x  ∈ Rn−1 and of radius r ∈ (0, ∞)
BMO1 , the BMO-based Sobolev spaces of order one (locally integrable functions
with distributional first-order partial derivatives in BMO)
 f ∗ (Δ), the local BMO norm of the function f on the surface ball Δ (cf. [68, §7.4]):
926 A Terms and notation used in Volume IV
⨏
 f ∗ (Δ) := sup | f − fΔ | dμ (A.0.18)
Δ ⊆Δ Δ

.
 f BMO(X,μ) , the homogeneous BMO semi-norm of the function f in the context of
a space of homogeneous type (X, ρ, μ) (cf. [68, §7.4]):
⨏
 
.
 f BMO(X,μ) := sup  f − fB (x,r)  dμ (A.0.19)
ρ
x ∈X, r >0 B ρ (x,r)

 · BMO(X,μ) , the inhomogeneous BMO “norm” in the context of a space of homo-

geneous type (X, ρ, μ) (cf. [68, §7.4]):
⎧
⎪ .
 f BMO(X,μ) if X is unbounded
⎪
⎨
⎪
 f BMO(X,μ) :=  ∫  (A.0.20)
⎪  
⎪ .
⎪  f BMO(X,μ) + f dμ if X is bounded
⎩ X

BMO X, μ , the space of functions of bounded mean oscillations for a space of

homogeneous type (X, ρ, μ) (cf. [68, §7.4]):

BMO X, μ := f ∈ Lloc 1
(X, μ) :  f BMO(X,μ) < +∞ (A.0.21)


BMO(X, μ), the space BMO modulo constants for a space of homogeneous type
(X, ρ, μ) (cf. [68, (7.4.96)]):
  

BMO(X, μ) := BMO(X, μ) ∼ = [ f ] : f ∈ BMO(X, μ) (A.0.22)

∂E, the topological boundary of the set E

Borelτ (X), the Borelians of the topological space (X, τ)
[A; B] := [A, B] := AB − BA, the commutator of A and B
{ A; B} := AB + BA, the anti-commutator of A and B
Bd(X → Y ), the space of linear and bounded operators from the quasi-normed vector
space X,  · X into the quasi-normed vector space Y,  · Y (cf. [69, §1.2]):
 
Bd(X → Y ) := T : X → Y : T linear mapping with T X→Y < +∞ (A.0.23)

Bd(X), the space of linear and bounded operators from the quasi-normed vector
space X into itself:

Bd(X) := Bd(X → X) (A.0.24)

B(X → Y ), the space of linear and bounded operators from X to Y , where X, Y are
two linear topological spaces (cf. [69, §1.1]):
A Terms and notation used in Volume IV 927
 
B(X → Y ) := T : X −→ Y : T linear and bounded (A.0.25)

B q,λ (Σ, σ), the block space on the Ahlfors regular set Σ ⊆ Rn , defined for q ∈ (1, ∞)
and λ ∈ (0, n − 1) as (cf. [69, §6.2]):


B q,λ (Σ, σ) := f ∈ Lipc (Σ) : there exist a sequence {λ j } j ∈N ∈ 1 (N) and
a family {b j } j ∈N of B q,λ -blocks on Σ so that
∞

f = λ j b j with convergence in Lipc (Σ)
j=1
(A.0.26)

and equipped with the norm


∞ 
∞

 f  B q, λ (Σ,σ) := inf |λ j | : f = λ j b j in Lipc (Σ) with (A.0.27)
j=1 j=1

{λ j } j ∈N ∈ 1
(N) and each b j a B q,λ -block on Σ

. p,q
Bs (Σ, σ), the homogeneous Besov space on the Ahlfors regular set Σ ⊆ Rn , defined
(cf. [69, Definition 7.1.2]) for
 
s ∈ (−1, 1), max n−1 n , n+s < p ≤ ∞,
n−1
0 < q ≤ ∞,
  
max (s)+, −s + (n − 1) p1 − 1 < β < 1, (A.0.28)
+
    
max s − p , (n
n−1
− 1) 1
p − 1 , −s + (n − 1) p1 − 1 < γ < 1
+

β,γ ∗
as the collection of “distributions” f on Σ (specifically, functionals f ∈ G̊0 (Σ) )
for which
  1/q
q
 f B. p, q (Σ,σ) := 2ks Ek f  L p (Σ,σ) <∞ (A.0.29)
s
k ∈Z

with the natural alterations when p = ∞ or q = ∞

p,q
Bs (Σ, σ), the inhomogeneous Besov space on the Ahlfors regular set Σ ⊆ Rn ,
defined (cf. [69, §7.1]) for
 
s ∈ (−1, 1), n , n+s < p ≤ ∞,
max n−1 n−1
0 < q ≤ ∞,
    
max (s)+ , −s + (n − 1) p1 − 1 < β < 1, (n − 1) p1 − 1 < γ < 1,
+ +
(A.0.30)
928 A Terms and notation used in Volume IV

as the collection of all “distributions” f on Σ such that, if {Ek }k ∈Z, k ≥κΣ is the family
of conditional expectation operators on the Ahlfors regular set Σ, then
  N
(
κΣ,τ)   p  1/p
 f Bsp, q (Σ,σ) := σ(QτκΣ,ν ) mQ κΣ ,ν |EκΣ f |
τ
τ ∈IκΣ ν=1

  1/q
 q
+ 2 Ek f  L p (Σ,σ)
ks
<∞ (A.0.31)
k ∈Z
k ≥
κΣ +1

p,q
Bs (Rn ), the Besov space in Rn equipped with the quasi-norm  · Bsp, q (Rn ) for
0 < p, q ≤ ∞ and s ∈ R (cf. [69, §9.1])
p,q
Bs (Ω), the Besov space in the open set Ω ⊆ Rn with 0 < p, q ≤ ∞ and s ∈ R,
as the collection of u ∈ D (Ω) for which there exists U ∈ Bs (Rn ) such that
p,q
defined


U Ω = u, and equipped with the quasi-norm (cf. [69, (9.2.1)])
 
uBsp, q (Ω) := inf U Bsp, q (Rn ) : U ∈ Bs (Rn ), U Ω = u
p,q

BMO−1 (∂Ω, σ), the BMO-based negative Sobolev space of order minus one on ∂Ω
(cf. [69, Definition 11.10.9]):
 n−1 ,1 ∗
BMO−1 (∂Ω, σ) := H1n−2 (∂Ω, σ) (A.0.32)

q,λ
B1 (∂Ω, σ), the block-based Sobolev space of order one on ∂Ω (cf. [69, (11.7.20)]):


q,λ
B1 (∂Ω, σ) := f ∈ B q,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.33)

one has ∂τ j k f ∈ B q,λ (∂Ω, σ)

equipped with the norm


n
 
q,λ
B1 (∂Ω, σ)  f −→  f  B q, λ (∂Ω,σ) + ∂τ f  q, λ (A.0.34)
jk B (∂Ω,σ)
j,k=1

p,q,λ
B1 (∂Ω, σ), the off-diagonal block-based Sobolev space of order one on ∂Ω (cf.
[69, (11.7.18)-(11.7.19)]):

p,q,λ
B1 (∂Ω, σ) := f ∈ B p,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.35)

one has ∂τ j k f ∈ B q,λ (∂Ω, σ)

A Terms and notation used in Volume IV 929

equipped with the norm


n
 
p,q,λ
B1 (∂Ω, σ)  f −→  f  B p, λ (∂Ω,σ) + ∂τ f  q, λ (A.0.36)
jk B (∂Ω,σ)
j,k=1

q,λ
B−1 (∂Ω, σ), the block-based negative Sobolev space of order minus one on ∂Ω (cf.
[69, Definition 11.8.9]):
 ∗
q,λ p,λ
B−1 (∂Ω, σ) := M̊1 (∂Ω, σ) (A.0.37)

if p, q ∈ (1, ∞) are such that 1/p + 1/q = 1 and λ ∈ (0, n − 1)

. q,λ
B1 (∂Ω, σ), the block-based homogeneous Sobolev space of order one on ∂Ω, for
q ∈ (1, ∞) and λ ∈ (0, n − 1) (cf. [69, Definition 11.13.11]):

. q,λ  σ(x)  qλ
B1 (∂Ω, σ) := f ∈ L 1 ∂Ω, ∩Lloc (∂Ω, σ) : for 1 ≤ j, k ≤ n
1 + |x| n

one has that ∂τ j k f ∈ B q,λ (∂Ω, σ)

(A.0.38)

where
q(n − 1)
qλ := ∈ (1, q), (A.0.39)
n − 1 + λ(q − 1)
and equipped with the semi-norm

. q,λ 
n
B1 (∂Ω, σ)  f →  f  B. q, λ (∂Ω,σ) := ∂τ j k f  B q, λ (∂Ω,σ) (A.0.40)
1
j,k=1

B0,β , the classical boundary-to-domain Bochner-Martinelli integral operator on

(0, β)-forms associated with a bounded 𝒞1 domain Ω ⊆ R2n ≡ Cn , whose action on
a continuous (0, β)-form f on ∂Ω is defined as in (7.3.28):
∫ ∫
∗
 
B0,β f (z) := ιζ f (ζ) ∧ Knβ (ζ, z) = − ν(ζ) ∧ f (ζ), ∂¯ζ Γ0,β (ζ, z) dσ(ζ)
C
∂Ω ∂Ω

for each z ∈ Cn \ ∂Ω, where ι : ∂Ω → Cn is the canonical inclusion and ι∗ζ indicates
pull-back in the variable ζ
Bα,β , the higher degree boundary-to-domain Bochner-Martinelli integral operator
in a given set of locally finite perimeter Ω ⊆ Cn , whose action on an arbitrary
(α, β)-form f ∈ L 1 ∂∗ Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β at each z ∈ Cn \ ∂Ω is as in (7.3.33):
930 A Terms and notation used in Volume IV
∫
1  
Bα,β f (z) := − ν 0,1 (ζ) ∧ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 C
∂∗ Ω

Bα,β , the boundary-to-boundary Bochner-Martinelli integral operator associated

with in a given set of locally finite perimeter Ω ⊆ Cn , whose action on an arbi-
trary (α, β)-form f ∈ L 1 ∂∗ Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β at each σ-a.e. z ∈ ∂Ω is as in
(7.3.68):
∫
1  0,1 
Bα,β f (z) := − lim+ ν (ζ) ∧ f (ζ), ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 ε→0 C
ζ ∈∂∗ Ω
|z−ζ |>ε

p,q
ℬα (Ω; D), the (interior) Besov-Hardy space of null-solutions of the Dirac operator
n
D= e j  ∂j in an open set Ω ⊆ Rn , defined for any p ∈ (0, ∞], any q ∈ (0, ≤ ∞],
j=1
and any α ∈ R as in (4.4.121):
p,q  p,q 
ℬα (Ω; D) := u ∈ Bα (Ω) ⊗ C n : Du = 0 in Ω
p,q
and equipped with the quasi-norm inherited from Bα (Ω) ⊗ C n
p,q
ℬs (∂Ω; D), the boundary Besov-Hardy space associated with the Dirac operator
n
D= e j  ∂j in a bounded NTA domain Ω ⊆ Rn with an Ahlfors regular boundary,
j=1
defined for integrability exponents p ∈ n−1 n , ∞ and q ∈ (0, ∞], and smoothness
index (n − 1) p1 − 1 + < s < 1 as in (4.4.125):
p,q  p,q 
ℬs (∂Ω; D) := TrΩ→∂Ω u : u ∈ ℬ 1 (Ω; D)
s+ p

p,q
and equipped with the quasi-norm inherited from Bs (∂Ω, σ) ⊗ C n
C
U, the closure of the set U ⊆ Rn
𝒞k (Ω), the space of functions of class 𝒞k in an open neighborhood of Ω
𝒞kc (Ω), the space of functions of class 𝒞k with compact support in the open set Ω
𝒞kb (Ω), the space of bounded functions of class 𝒞k in Ω
 ∗
𝒞∞
b (Ω) , the algebraic dual of 𝒞∞
b (Ω)
CBM(Ω), the space of complex Borel measures in the open set Ω ⊆ Rn
C n , the Clifford algebra (C n, +, ) generated by n imaginary units defined as the
minimal enlargement of Rn to a unitary real algebra which is not generated (as
an algebra) by any proper subspace of Rn , and such that x  x = −|x| 2 for each
x ∈ Rn → C n (cf. [68, §6.4])
A Terms and notation used in Volume IV 931

 · 𝒞. α (U,ρ) , the homogeneous Hölder space semi-norm of order α > 0 in the set
U ⊆ X, in the context of a quasi-metric space (X, ρ), defined for each function
f : U → R as (cf. [68, (7.3.2)]):

| f (x) − f (y)|
 f 𝒞. α (U,ρ) := sup (A.0.41)
x,y ∈U ρ(x, y)α
xy

.
𝒞α (U, ρ), the homogeneous Hölder space of order α > 0 in the set U ⊆ X, defined
in the context of a quasi-metric space (X, ρ) as (cf. [68, (7.3.1)]):
. 
𝒞α (U, ρ) := f : U → R :  f 𝒞. α (U,ρ) < +∞ (A.0.42)

.
𝒞α (U, ρ)/∼, the homogeneous Hölder space of order α > 0 modulo constants, in the
set U ⊆ X, defined in the context of a quasi-metric space (X, ρ) as (cf. [68, (7.3.6)]):
.  . 
𝒞α (U, ρ)/∼ := [ f ] : f ∈ 𝒞α (U, ρ) (A.0.43)
. α (U, ρ), the local homogeneous Hölder space of order α > 0 in the set U ⊆ X,
𝒞loc
defined in the context of a quasi-metric space (X, ρ) as (cf. [68, (7.3.7)]):
.α   .
𝒞loc (U, ρ) := f : U → C : f Bρ (x,r)∩U ∈ 𝒞α Bρ (x, r) ∩ U, ρ

for each x ∈ U and r ∈ (0, ∞) (A.0.44)

 · 𝒞α (U,ρ) , the inhomogeneous Hölder space norm of order α > 0 in the set U ⊆ X,
in the context of a quasi-metric space (X, ρ), defined for each function f : U → R
as (cf. [68, (7.3.20)]):

 f 𝒞α (U,ρ) := sup | f | +  f 𝒞. α (U,ρ), ∀ f ∈ 𝒞α (U, ρ) (A.0.45)

U

𝒞α (U, ρ), the inhomogeneous Hölder space of order α > 0 in the set U ⊆ X, defined
in the context of a quasi-metric space (X, ρ) as (cf. [68, (7.3.19)]):
 . 
𝒞α (U, ρ) := f ∈ 𝒞α (U, ρ) : f is bounded in U (A.0.46)

𝒞αc (U, ρ), the space of Hölder functions of order α > 0 with ρ-bounded support
in the set U ⊆ X, defined in the context of a quasi-metric space (X, ρ) as (cf. [68,
(7.3.26), (7.3.27)]):
 . 
𝒞αc (U, ρ) := f ∈ 𝒞α (U, ρ) : f vanishes outside of a ρ-bounded subset of U
(A.0.47)
932 A Terms and notation used in Volume IV
.γ
𝒞van (Σ), the homogeneous “vanishing” Hölder space of order γ on the set Σ (cf. [69,
(3.2.5)]):
.  .  
𝒞γvan (Σ) := f ∈ 𝒞γ (Σ) : lim+ sup  f 𝒞. γ (B(x,r)∩Σ) = 0 (A.0.48)
r→0 x ∈Σ

𝒞γvan (Σ), the inhomogeneous “vanishing” Hölder space of order γ on the set Σ (cf.
[69, (3.2.8)]):
  
𝒞γvan (Σ) := f ∈ 𝒞γ (Σ) : lim+ sup  f 𝒞. γ (B(x,r)∩Σ) = 0 (A.0.49)
r→0 x ∈Σ

Cp(X → Y ), the space of compact linear operators from the topological vector space
X into the topological vector space Y :
 
Cp(X → Y ) := T : X → Y : T linear compact mapping (A.0.50)

Cp(X), the space of compact linear operators from the topological vector space X
into itself:

Cp(X) := Cp(X → X) (A.0.51)

CMO(Σ, σ), the Coifman-Weiss space (cf. [69, (4.6.11)]):

CMO(Σ, σ) is the closure in BMO(Σ, σ) of 𝒞00 (Σ), the space of

(A.0.52)
all continuous functions on Σ which vanish at infinity

C, the boundary-to-domain Cauchy-Clifford integral operator associated with a given

set Ω ⊆ Rn of locally finite perimeter, whose action on any given function f in
L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ C n at each x ∈ Ω̊ is
∫
1 x−y
C f (x) :=  ν(y)  f (y) dσ∗ (y) (A.0.53)
ωn−1 |x − y| n
∂∗ Ω

C, the boundary-to-boundary Cauchy-Clifford integral operator, acting on any func-

tion f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ C n at σ∗ -a.e. point x ∈ ∂∗ Ω as:
∫
1 x−y
C f (x) := lim+  ν(y)  f (y) dσ∗ (y) (A.0.54)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

C # , the transpose Cauchy-Clifford integral operator associated with a set Ω ⊆ Rn

of locally finite perimeter, acting on functions f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
⊗ C n at σ-a.e.
point x ∈ ∂∗ Ω as in (1.6.1):
A Terms and notation used in Volume IV 933
∫
−1 x−y
C # f (x) := lim+ ν(x)   f (y) dσ(y)
ε→0 ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

𝒞mod , the modified boundary-to-domain Cauchy integral operator in the plane, acting
σ(ζ )
on f ∈ L 1 ∂∗ Ω, 1+ |ζ | 2
at each z ∈ Ω as in (1.8.227):
∫  1
1 1
(𝒞mod )(z) := − 1C\B(0,1) (ζ) f (ζ) dζ
2πi ∂∗ Ω ζ − z ζ

Cmod , the modified boundary-to-domain Cauchy-Clifford integral operator, acting on

any function f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n ⊗ C n at each point x ∈ Ω as in (1.8.238):
∫ 
1 x−y y
Cmod f (x) := + 1 n (y)  ν(y)  f (y) dσ(y)
ωn−1 |x − y| n |y| n R \B(0,1)
∂∗ Ω

.p
Cq,η (Σ, σ), the homogeneous Calderón space on the closed Ahlfors regular set
Σ ⊆ Rn for p ∈ [1, ∞], q ∈ [1, ∞), and η ∈ R, defined as in (3.1.10), (3.1.11):
.p 
Cq,η (Σ, σ) := f ∈ Lloc
1
(Σ, σ) : fq,η
#
∈ L p (Σ, σ)

and equipped with the semi-norm

 #  .p
 f C. p (Σ,σ) :=  fq,η  p
L (Σ,σ)
, ∀ f ∈ Cq,η (Σ, σ)
q, η

 ⨏   1/q 
# (x) := sup R−η
where fq,η  f (y) − fΔ(x,R) q dσ(y) for all x ∈ Σ
Δ(x,R)
R>0
p
Cq,η (Σ, σ), the inhomogeneous Calderón space on the closed Ahlfors regular set
Σ ⊆ Rn , defined for p ∈ [1, ∞], q ∈ [1, ∞), and η ∈ R as in (3.1.14), (3.1.15)
.p 
p
Cq,η (Σ, σ) := L p (Σ, σ) ∩ Cq,η (Σ, σ) = f ∈ L p (Σ, σ) : fq,η
#
∈ L p (Σ, σ)

and equip this space with the norm

 # 
 f Cq,p η (Σ,σ) :=  f  L p (Σ,σ) +  fq,η  p p
L (Σ,σ)
, ∀ f ∈ Cq,η (Σ, σ)

D
u · w = u, w , the dot product of two vectors u, w ∈ Rn
 the divergence of the vector field F
divF,
D (Ω), the space of distributions in the open set Ω
D  (Ω) ·, · D(Ω) , the distributional pairing in the open set Ω
Δ := ∂12 + · · · + ∂n2 , the Laplace operator in Rn
934 A Terms and notation used in Volume IV

δx , the Dirac distribution with mass at x

D, the classical (homogeneous) Dirac operator in Rn defined as (cf. [68, (6.4.139)]):

n
D= e j  ∂j (A.0.55)
j=1
 
αβ
D= n
j=1 a j ∂j + bαβ 1≤α ≤ N , a (generic) N × N  first-order system
1≤β ≤ N 
D , the (real) transpose of the first-order system D:
 n 
αβ
D := − a j ∂j + bαβ 1≤β ≤ N  (A.0.56)
j=1 1≤α ≤ N

D, the complex conjugate of the first-order system D

D∗ , the Hermitian adjoint of the first-order system D
d, the exterior derivative operator acting on the differential form u = J uJ dx J
according to (cf. [68, (1.11.32)], and also [68, (6.4.140)-(6.4.141)] for the Clifford
algebra context):
n 
∂uJ
du = dx j ∧ dx J (A.0.57)
j=1 J
∂ x j

δ, the formal adjoint of the exterior derivative operator d on differential forms (see
also [68, (6.4.142)] for the Clifford algebra context)
δ jk , the Kronecker symbol, i.e., δ jk := 1 if j = k and δ jk := 0 if j  k
δ∂Ω (·), the distance function to the boundary of Ω
UV := (U \ V) ∪ (V \ U), the symmetric difference of the sets U and V
Δ(x, r) := B(x, r) ∩ ∂Ω, the surface ball on ∂Ω with center at x ∈ ∂Ω and radius
r>0
D := D L , the (homogeneous) Dirac operator in Rn acting from the left

n
D := D L := e j  ∂j (A.0.58)
j=1

DR , the (homogeneous) Dirac operator acting from the right on the Clifford algebra-
valued function u according to

n
DR u := (∂j u)  e j (A.0.59)
j=1
A Terms and notation used in Volume IV 935

Dk (X), the k-th generation of dyadic cubes in the geometrically doubling quasi-
metric space X, defined as in [68, Proposition 7.5.4]:

Dk (X) := {Q αk }α∈Ik (A.0.60)

D(X), the dyadic grid on the geometrically doubling quasi-metric space X, defined
as in [68, Proposition 7.5.4]:

D(X) := Dk (X) (A.0.61)
k ∈Z, k ≥κ X

dim X, the dimension of a vector space X

dζ, the complex arc-length, defined as:

dζ := iν(ζ) dσ(ζ) (A.0.62)

Def, the deformation tensor, acting on the vector-distribution u = (u1, . . . , un ) as:

Def u := 1
2 ∂j uk + ∂k u j 1≤ j,k ≤n (A.0.63)

Δ + k 2 , the Helmholtz operator with wave number k

dV, the volume element in Cn , defined as in (7.1.47):

dV := dx1 ∧ dy1 ∧ · · · ∧ dxn ∧ dyn = (−2i)−n dz1 ∧ d z̄1 ∧ · · · ∧ dzn ∧ d z̄n

D = D A, the boundary-to-domain double layer for a second-order, homogeneous,

αβ
constant coefficient, M × M weakly elliptic system L A := ar s ∂r ∂s 1≤α,β ≤M cor-
αβ
responding to a given coefficient tensor A := ar s 1≤α,β ≤M (with canonical fun-
1≤r,s ≤n
damental solution E = Eγβ 1≤γ,β ≤M ), whose action on any vector-valued function
σ(x) M
f = ( fα )1≤α ≤M ∈ L1∂Ω, 1+ |x | n−1
is defined at each point x ∈ Ω as in (1.3.18):
 ∫ 
βα
D A f (x) := − νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y)
1≤γ ≤M
∂∗ Ω

Dmod , the boundary-to-domain modified double layer potential operator for an M × M

αβ
weakly elliptic system L A := ar s ∂r ∂s 1≤α,β ≤M corresponding to the coefficient
αβ
tensor A := ar s 1≤α,β ≤M (with fundamental solution E = Eγβ 1≤γ,β ≤M ) whose
1≤r,s ≤n
action on f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n at each x ∈ Ω is defined as in
(1.8.6):
 ∫ 
βα  (rγβ) 
Dmod f (x) := − νs (y)ar s (∂r Eγβ )(x − y) − k1 (−y) fα (y) dσ(y)
∂∗ Ω 1≤γ ≤M
936 A Terms and notation used in Volume IV

(rγβ)
where k1 := (∂r Eγβ ) · 1Rn \B(0,1) for every r, γ, β
DΔ,mod , the boundary-to-domain modified harmonic double layer potential operator
whose action on each f ∈ L 1 ∂∗ Ω, 1+σ(y)
|y | n is defined at each x ∈ Ω as in (1.8.50):
∫
1  ν(y), y − x ν(y), y 
DΔ,mod f (x) := − · 1Rn \B(0,1) (y) f (y) dσ(y)
ωn−1 |x − y| n |y| n
∂∗ Ω

Dλ , the boundary-to-domain double layer for the Stokes system associated with a
given open set Ω ⊆ Rn of locally finite perimeter and with the coefficient tensor
Aλ := δ jk δαβ + λδ jβ δkα 1≤α,β ≤n corresponding to any λ ∈ C, whose action on
  1≤ j,k ≤n  n

each f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω, 1+σ(x) at each x ∈ Ω as in (6.2.17), (6.2.18):
|x | n−1

! ∫ 
δ jγ x − y, ν(y)
Dλ f(x) = (λ − 1)
2ωn−1 |x − y| n
∂∗ Ω

n(λ + 1) x − y, ν(y) (x j − y j )(xγ − yγ )

−
2ωn−1 |x − y| n+2
 "
λ − 1 ν j (y)(xγ − yγ ) − νγ (y)(x j − y j )
+ f j (y) dσ(y)
2ωn−1 |x − y| n
1≤γ ≤n

Dα,β , the boundary-to-domain complex double layer potential operator associated

with Ω ⊆ Cn , acting on any given (α, β)-form f ∈ L 1 ∂∗ Ω, 1+σ(ζ )
|ζ | 2n−1
⊗ Λα,β at each
z ∈ Ω as in (7.4.144):
∫ # $
1
Dα,β f (z) := f (ζ), ν 0,1 (ζ) ∧ ϑζ Γα,β (ζ, z) − ν 1,0 (ζ) ∨ ∂¯ζ Γα,β (ζ, z) dσ(ζ)
2 C
∂∗ Ω

D  (Ω) ·, · D(Ω) , the distributional Hermitian inner product for double forms (7.3.11)

E
(ε, δ)-domain: (cf. [68, Definition 5.11.8])
a nonempty, open, proper subset Ω of Rn with the property that
for any x, y ∈ Ω with |x − y| < δ there exists a rectifiable curve
γ : [0, 1] → Ω such that γ(0) = x, γ(1) = y, as well as length(γ) ≤ (A.0.64)
|z−x | |z−y |
ε |x − y| and ≤ ε1 dist(z, ∂Ω) for each z ∈ γ([0, 1])
1
|x−y |

e j , the unit vector in the j-th direction in Rn , defined for each j ∈ {1, . . . , n} as
e j := (δ jk )1≤k ≤n ∈ Rn where δ jk is the Kronecker symbol
A Terms and notation used in Volume IV 937

{e j }1≤ j ≤n , the standard orthonormal basis in Rn

EΔ , the standard fundamental solution for the Laplacian, given for each x ∈ Rn \ {0}
by
⎧
⎪ 1 1
⎪
⎨ ωn−1 (2 − n) |x| n−2 if n ≥ 3,
⎪
⎪
EΔ (x) := (A.0.65)
⎪ 1
⎪
⎪
⎪ ln |x| if n = 2
⎩ 2π
ℰ(Ω), the space of distributions compactly supported in the open set Ω ⊆ Rn
[x]X/Y := x + Y , the equivalence class of the vector x ∈ X in the quotient space X/Y

E p (X), the p-envelope of a quasi-normed space X whose dual separates

  points,
defined for p ∈ (0, 1] as the completion of X in the quasi-norm  ·  p (cf. [69,
Definition 7.8.4])
Ex∂Ω→Ω , the extension operator from ∂Ω to Ω (cf. [69, Theorem 8.4.1])
E = E L = Eαβ 1≤α,β ≤M , the canonical fundamental solution of the weakly elliptic
M × M system L defined as in [70, Theorem 1.4.2]:
 ∫ 
⎧
⎪ Δ (n−1)/2  
⎪
⎪ − x −1
 x, ξ  L(ξ) dH (ξ) if n is odd,
n−1
⎪
⎪
⎪
⎪ 4(2πi)n−1
⎪
⎪
⎨
⎪ S n−1
E(x) := (A.0.66)
⎪
⎪  ∫ 
⎪
⎪ Δx(n−2)/2  
⎪
⎪ −1
ln  x, ξ  L(ξ) dH n−1 (ξ) if n is even.
⎪
⎪
⎪
⎪ (2πi) n
⎩ S n−1

εBA, the generalized Kronecker symbol defined for any two arrays A, B as (cf. [68,
(6.4.116)]): 
det (δab )a ∈ A,b ∈B if | A| = |B|,
εB :=
A
(A.0.67)
0 otherwise

E  (Ω) ·, · E(Ω) , the (compact support) distributional Hermitian inner product for
double forms (7.3.12)
En (·, ·), the fundamental solution for the complex Laplacian  := ∂ϑ ¯ + ϑ ∂¯ in Cn
(7.3.18)
F
Φ(X → Y ), the collection of Fredholm operators from the linear topological space
X into the linear topological space Y (cf. [69, Definition 2.2.1])
Φ+ (X → Y ), the collection of finite-dim kernel semi-Fredholm operators from the
Banach space X into the Banach space Y (cf. [69, §2.1])
Φ− (X → Y ), the collection of finite-dim cokernel semi-Fredholm operators from
the Banach space X into the Banach space Y (cf. [69, §2.1])
938 A Terms and notation used in Volume IV

fγ , the Fefferman-Stein grand maximal function (with parameter γ ∈ (0, 1)), asso-

ciating to each “distribution” f ∈ Lipc (Σ) on the Ahlfors regular set Σ ⊆ Rn the
function defined at each point x ∈ Σ by (cf. [69, (4.1.6)]):
 
fγ (x) := sup  f , ψ  (A.0.68)
ψ ∈ Tγ (x)

.
where Tγ (x) ⊆ Lipc (Σ) is the collection of all normalized (in 𝒞γ (Σ)) bump functions
centered at the point x
ρFred (T; X), the Fredholm (or essential spectral) radius of T ∈ Bd (X) (cf. [69,
Definition 2.2.5]):
 
ρFred (T; X) := inf r > 0 : zI − T ∈ Φ(X → X) for each z ∈ C \ B(0, r)
(A.0.69)

 ∞ , the contribution of the vector field F at infinity defined (for a system of

[F]
auxiliary functions, i.e., a family {φ R }R>0 of smooth compactly supported functions
in Rn which are globally bounded and progressively become pointwise equal to 1 on
compact sets, in a uniform fashion) as in [68, Chapters 1 and 4]:
∫
 ∞ := − lim
[F] ∇φ R · F dL n (A.0.70)
R→∞ Ω

. p,q
Fs (Σ, σ), the homogeneous Triebel-Lizorkin space on the Ahlfors regular set
Σ ⊆ Rn , defined (cf. [69, Definition 7.1.2]) for
   
s ∈ (−1, 1), max n−1 n , n+s < p ≤ ∞,
n−1
n , n+s < q ≤ ∞,
max n−1 n−1

  
max (s)+, −s + (n − 1) p1 − 1 < β < 1, (A.0.71)
+
    
max s − p , (n
n−1
− 1) 1
p − 1 , −s + (n − 1) 1
p −1 < γ < 1,
+

as the collection of all “distributions” f on the closed set Σ (specifically, functionals

β,γ ∗
in G̊0 (Σ) ) with the property that
   1/q 
 
 f F. p, q (Σ,σ) :=  
q
2ks |Ek f |  <∞ (A.0.72)
s
k ∈Z L p (Σ,σ)

whenever p < ∞ (with natural alterations when q = ∞) and, corresponding to the

case when p = ∞,
%⨏ 
∞ & 1/q
q
 f F. ∞, q (Σ,σ) := sup sup 2ks |Ek f | dσ <∞ (A.0.73)
s
 ∈Z τ ∈I Qτ k=
A Terms and notation used in Volume IV 939

again with natural alterations when q = ∞

p,q
Fs (Σ, σ), the inhomogeneous Triebel-Lizorkin space on the Ahlfors regular set
Σ ⊆ Rn , defined as the collection of all “distributions” f on Σ with the property that
 f Fsp, q (Σ,σ) < +∞ (cf. [69, Definition 7.1.2])
p,q
Fs (Rn ), the (inhomogeneous) Triebel-Lizorkin space in Rn for 0 < p, q ≤ ∞ and
s ∈ R, equipped with the quasi-norm  · Fsp, q (Rn ) (cf. [69, (9.1.4)-(9.1.5)])
  
Fs (Ω) := u ∈ D (Ω) : there exists U ∈ Fs (Rn ) such that U Ω = u , the
p,q p,q

Triebel-Lizorkin space in the (arbitrary) open set Ω ⊆ Rn for 0 < p, q ≤ ∞ and

s ∈ R, equipped with the quasi-norm (cf. [69, (9.2.1)])
 
uFsp, q (Ω) := inf U Fsp, q (Rn ) : U ∈ Fs (Rn ), U Ω = u
p,q

⨏
fΔ := Δ
f dσ, the integral average of f on the “surface ball” Δ
φ', the Fourier transform of φ in Rn :
∫
φ'(ξ) := e−i x, ξ
φ(x) dx, ξ ∈ Rn (A.0.74)
Rn

Φk (·), the (unique) radiating fundamental solution of the Helmholtz operator Δ + k 2

in Rn , given at each x ∈ Rn \ {0} by the formula (cf. [70, (6.1.7)]):

1 H (1) (k |x|)
(n−2)/2 (n−2)/2
Φk (x) := k (A.0.75)
4i(2π)(n−2)/2 |x| (n−2)/2

(   1/2
| f |C := f, f C = 2α+β |J |=β | fI,J |
2 , the “complex” norm of the dif-
|I |=α
ferential form f = fI,J dz I ∧ d z̄ J (7.1.46)
|I |=α, |J |=β

2 ν ∨ (ν 0,1 ∧ f ), the tangential complex part of a form f (7.1.68)

1 1,0
ftan,C :=

2 ν ∧ (ν 1,0 ∨ f ), the normal complex part of a form f (7.1.68)

1 0,1
fnor,C :=
p,q
ℱα (Ω; D), the Triebel-Lizorkin-Hardy space of null-solutions of the Dirac oper-
n
ator D = e j  ∂j in an open set Ω ⊆ Rn defined for p ∈ (0, ∞), q ∈ (0, ∞], and
j=1
α ∈ R as in (4.4.130):
p,q  p,q 
ℱα (Ω; D) := u ∈ Fα (Ω) ⊗ C n : Du = 0 in Ω

G
g= g jk dx j ⊗ dxk , the Riemannian metric tensor
1≤ j,k ≤n
940 A Terms and notation used in Volume IV

∇u, the gradient (Jacobian matrix) of a C M -valued function u = (uα )1≤α ≤M defined
in an open subset of Rn , defined as:
⎡ ∂1 u1 · · · ∂n u1 ⎤⎥
⎢
⎢ .. ⎥
∇u := ∂j uα 1≤α ≤M = ⎢ ... ..
. . ⎥⎥ (A.0.76)
1≤ j ≤n ⎢
⎢ ∂1 u M · · · ∂n u M ⎥⎦
⎣

∇ , the gradient operator in Rn−1

Γκ (x) = ΓΩ,κ (x), the (κ-)nontangential approach region with vertex at x ∈ ∂Ω,
defined as (cf. [68, (8.1.2)]):
 
Γκ (x) = ΓΩ,κ (x) := y ∈ Ω : |x − y| < (1 + κ)δ∂Ω (y) , ∀x ∈ ∂Ω (A.0.77)

X, a Generalized Banach Function Space on the measure space (X, M, μ) equipped

with the norm  · X (cf. [69, Definition 5.1.4])
∞ (X, μ)  · X ∞
X̊ := Lcomp , the closure of Lcomp in the Generalized Banach Function Space
X,  · X (cf. [69, Definition 5.2.6])
X, a Generalized Banach Function Space on the measure space (X, M, μ) equipped
with the norm  · X (cf. [69, Definition 5.1.4])
∞ (X, μ)  · X ∞
X̊ := Lcomp , the closure of Lcomp in the Generalized Banach Function Space
X,  · X (cf. [69, Definition 5.2.6])
∇tan f , the tangential gradient of the function f ∈ L1,loc
1 (∂ Ω, σ ) (cf. [69, §11.4]):
∗ ∗


n 
∇tan f := νk ∂τk j f at σ∗ -a.e. point on ∂∗ Ω (A.0.78)
1≤ j ≤n
k=1

∇tan
A , the tangential gradient of u = (u ) associated with the coefficient tensor
β β
αβ
A := ar s 1≤α,β ≤M , defined (with ∇tan denoting the tangential gradient acting on
1≤r,s ≤n
scalar functions along ∂Ω) as
 κ−n.t. 
u := νr ar s ∇tan uβ ∂Ω
αβ
∇tan
A
s 1≤α ≤M

GΩ (·, ·), the Green function for the Laplacian, where Ω is a bounded open set in Rn
Γα,β , the double form of type (α, β), (β, α) , defined for α, β ∈ {0, 1, . . . , n} as in
(7.3.19):
 
Γα,β (ζ, z) := 2−α−β En (ζ, z) (dζ J ∧ dζ¯ I ) ⊗ (d z̄ J ∧ dz I )
|J |=α |I |=β
A Terms and notation used in Volume IV 941

H
H n−1 , the (n − 1)-dimensional Hausdorff measure in Rn
H s , the s-dimensional Hausdorff measure in Rn
H p (Σ, σ), the Lebesgue-based Hardy space on the Ahlfors regular set Σ ⊆ Rn ,
defined for p ∈ n−1n , ∞ and γ ∈ (n − 1) p − 1 +, 1 as (cf. [69, Definition 4.2.1])
1



H p (Σ, σ) := f ∈ Lipc (Σ) : fγ ∈ L p (Σ, σ) (A.0.79)

and equipped with the quasi-norm

 
 f  H p (Σ,σ) :=  fγ  L p (Σ,σ), ∀ f ∈ H p (Σ, σ) (A.0.80)

.
H p (Σ, σ), the homogeneous
. p Hardy space
 on the Ahlfors regular set Σ ⊆ Rn , defined
for each p ∈ n , ∞ as H (Σ, σ) := f ∈ H (Σ, σ) : f , 1 = 0 if Σ is bounded,
n−1 p
.
and simply as H p (Σ, σ) := H p (Σ, σ) if Σ is unbounded [69, (4.2.12)]
H p,q (Σ, σ), the Lorentz-based Hardy space on the Ahlfors regular set Σ ⊆ Rn ,
defined for p ∈ n−1 n , ∞ , p ∈ (0, ∞], and γ ∈ (n − 1) p − 1 +, 1 as (cf. [69,
1

Definition 4.2.3])


H p,q (Σ, σ) := f ∈ Lipc (Σ) : fγ ∈ L p,q (Σ, σ) (A.0.81)

and equipped with the quasi-norm

 
 f  H p, q (Σ,σ) :=  fγ  L p, q (Σ,σ), ∀ f ∈ H p,q (Σ, σ) (A.0.82)

p,q
Hfin (Σ, σ), the vector space of all finite linear combinations of (p, q)-atoms on
Σ equipped with the quasi-norm  f  H p, q (Σ,σ) defined as the infimum of all
 N  1/p N
fin

|λ j | p such that f = λ j a j for {λ j }1≤ j ≤ N ⊆ C and (p, q)-atoms {a j }1≤ j ≤ N

j=1 j=1
(cf. [69, (4.4.113)])
H, the L p -filtering operator, acting on each given distribution f ∈ H p (Σ, σ) with
n < p < ∞ at each x ∈ Σ according to (cf. [69, Theorem 4.9.1]):
n−1

 
(H f )(x) := lim+ (H p (Σ,σ))∗ St (x, ·), f H p (Σ,σ) (A.0.83)
t→0

where {St (·, ·)}t are the integral kernels of a suitable approximation to the identity
ℋq,λ (Σ, σ), the space defined for any given q ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69,
§6.1]):
942 A Terms and notation used in Volume IV


ℋq,λ (Σ, σ) := f ∈ Lipc (Σ) : there exist a sequence {λ j } j ∈N ∈ 1
(N) and
a family {a j } j ∈N of ℋq,λ -atoms on Σ so that
∞

f = λ j a j with convergence in Lipc (Σ) ,
j=1
(A.0.84)

and equipped with the norm


∞ 
∞

 f ℋq, λ (Σ,σ) := inf |λ j | : f = λ j a j in Lipc (Σ) with (A.0.85)
j=1 j=1

{λ j } j ∈N ∈ 1
(N) and each a j a ℋq,λ -atom on Σ

H s (Ω), the L 2 -based fractional Sobolev space of order s ∈ R in an open set Ω ⊆ Rn

(cf. [69, §9.2]):
   
H s (Ω) := U Ω : U ∈ H s (Rn ) = U Ω : U ∈ Fs2,2 (Rn ) , ∀s ∈ R, (A.0.86)

and equipped with the norm

 
u H s (Ω) := inf U  H s (Rn ) : U ∈ H s (Rn ) such that u = U Ω
 
≈ inf U F 2,2 (Rn ) : U ∈ Fs2,2 (Rn ) such that u = U Ω (A.0.87)
s

p
Hκ (Ω), the Hardy space of harmonic functions u in the open set Ω ⊆ Rn with a p-th
power integral nontantangential maximal function:
 
Hκ (Ω) := u ∈ 𝒞∞ (Ω) : Δu = 0 in Ω, and Nκ u ∈ L p (∂Ω, σ)
p
(A.0.88)

Hλ(1) (·), the Hankel function of the first kind with index λ ∈ R
.p
H1 (∂Ω, σ), the Hardy-based homogeneous Sobolev space of order one on ∂Ω,
defined for each p ∈ n−1
n , ∞ as (cf. [69, Definition 11.10.5]):

.p  σ(x) 
H1 (∂Ω, σ) := f ∈ L 1 ∂Ω , : ∂τ j k f ∈ H p (∂Ω, σ) (A.0.89)
1 + |x| n

for each j, k ∈ {1, . . . , n}

and equipped with the semi-norm

A Terms and notation used in Volume IV 943

.p 
n
H1 (∂Ω, σ)  f →  f  H. p (∂Ω,σ) := ∂τ j k f  H p (∂Ω,σ) (A.0.90)
1
j,k=1

.p 
H1 (∂Ω, σ) ∼, the quotient space of classes [·] of equivalence modulo constants of
.p
functions in H1 (∂Ω, σ) defined for p ∈ n−1
n , ∞ as (cf. [69, (11.10.33)-(11.10.34)]):
.p   .p 
H1 (∂Ω, σ) ∼ := [ f ] : f ∈ H1 (∂Ω, σ) (A.0.91)
equipped with the semi-quasinorm

.p    
n
 
H1 (∂Ω, σ) ∼  [ f ] −→ [ f ] H. p (∂Ω,σ)/∼ := ∂τ f  p
jk H (∂Ω,σ)
(A.0.92)
1
j,k=1

q, p
H1 (∂Ω, σ), the Hardy-based inhomogeneous Sobolev space of order one on ∂Ω,
defined for p ∈ n−1
n , ∞ and q ∈ [1, ∞] as (cf. [69, Definition 11.10.6]):

q, p .p
H1 (∂Ω, σ) := L q (∂Ω, σ) ∩ H1 (∂Ω, σ)

= f ∈ L q (∂Ω, σ) : ∂τ j k f ∈ H p (∂Ω, σ) for 1 ≤ j, k ≤ n
(A.0.93)

and equipped with the quasi-norm

q, p
 f  H q, p (∂Ω,σ) :=  f  L q (∂Ω,σ) +  f  H. p (∂Ω,σ), ∀ f ∈ H1 (∂Ω, σ) (A.0.94)
1 1

H p (Ω; D), the Hardy space in Ω associated with the first-order N × M system D,
defined as
M
H p (Ω; D) is the collection of all functions u ∈ 𝒞∞ (Ω) satisfying
Nκ u ∈ L p (∂Ω, σ) and Du = 0 in Ω, and which also vanish at infinity (A.0.95)
(in the sense described in [70, Definition 1.6.3]) when Ω is an exterior
domain
and equipped with the quasi-norm
 
u H p (Ω;D) := Nκ u L p (∂Ω,σ), ∀u ∈ H p (Ω; D) (A.0.96)

p
ℋ• (∂Ω; D), the “bullet” boundary Hardy space associated with the Dirac operator
D := nj=1 e j  ∂j in the open set Ω ⊆ Rn , defined as in (A.0.97):
p  
ℋ• (∂Ω; D) := ν 
• u : u ∈ H p (Ω; D) (A.0.97)

and equipped with the quasi-norm inherited from H p (∂Ω, σ) ⊗ C n

944 A Terms and notation used in Volume IV

∗, the Hodge star operator in Cn ≡ R2n whose action satisfies u ∧ (∗ū) = |u|C2 dV for
u ∈ Λα,β Cn and α, β ∈ {0, 1, . . . , n} (7.1.48)
.p
H1,± (∂Ω, σ), the “positive/negative” homogeneous Hardy spaces with regularity on
∂Ω, defined as in (1.8.285):
.p  κ−n.t.
H1,± (∂Ω, σ) := u∂Ω : u holomorphic in Ω± and

Nκ (∇u) ∈ L p (∂Ω, σ)

where Ω+ := Ω and Ω− := C \ Ω
Hmod , the modified Hilbert transform on the real line, whose action on each function
f ∈ L 1 R, 1+dx
|x | 2
at L 1 -a.e. point x ∈ R is defined as in (5.3.99):
∫ 
1 1 1
Hmod f (x) := lim+ 1R\[x−ε,x+ε] (y) + 1R\[−1,1] (y) f (y) dy
ε→0 π x−y y
R

H p (Ω), the (several variable) holomorphic Hardy space in the UR domain Ω in

R2n ≡ Cn defined for each p ∈ 2n−1 2n , ∞ as in (7.6.17):

H p (Ω) := F : Ω → C : F holomorphic, Nκ F ∈ L p (∂Ω, σ),

and lim F(z) = 0 if Ω is an exterior domain ,

|z |→∞

and equipped with the quasi-norm H p (Ω)  F → F  H p (Ω) := Nκ F  L p (∂Ω,σ)

·, · C , the Hermitian inner product on the space of differential forms uniquely defined
by the requirement that one has dz I ∧ d z̄ J , dz A ∧ d z̄ B C = 2 |I |+ |J | ε IA εBJ for all arrays
I, J, A, B (7.1.42)
·, · C , the pointwise Hermitian inner product for double forms
I
·, · , the (real) inner product in C M defined for any vectors u = (uk )1≤k ≤M ∈ C M
and w = (wk )1≤k ≤M ∈ C M as:


M
u, w := u k wk (A.0.98)
k=1

√
i := −1 ∈ C, the complex imaginary unit
ι∗ , the pull-back map induced by the canonical inclusion ι
Ů, the interior of the set U ⊆ Rn
⨏ ∫
E
1
f dμ := μ(E) E
f dμ, the integral average of the function f on the set E ⊆ X, in
a measure space (X, μ)
A Terms and notation used in Volume IV 945

fBρ (x,r) , the integral average of f over the ρ-ball Bρ (x, r), in the context of a space
of homogeneous type (X, ρ, μ), defined as (cf. [68, (7.4.9)]):
⨏ ∫
1
fBρ (x,r) := f dμ := f (y) dμ(y) (A.0.99)
B ρ (x,r) μ Bρ (x, r) Bρ (x,r)

IE,α , the fractional integral operator of order α on the set E contained in a metric
space (X, ρ) equipped with an upper d-dimensional Borel measure μ on (X, τρ ),
μ(x)
acting on functions f ∈ L 1 E, 1+ρ(x,x ) d−α
according to (cf. [68, (7.8.3)]):
0

∫
f (y)
IE,α f (x) := dμ(y) for μ-a.e. x ∈ E (A.0.100)
E ρ(x, y)d−α

Im (T : X → Y ) := {T x : x ∈ X }, the image (or range) of the operator T : X → Y

 
(X0, X1 )θ,q := x ∈ X0 + X1 :  x(X0,X1 )θ, q < +∞ , the intermediate space for the
real method of interpolation between the compatible pair of quasi-Banach spaces X0
and X1 , equipped with the real interpolation quasi-norm  · (X0,X1 )θ, q (cf. [69, §1.3])

ln hΦ (t) ln hΦ (t)
i(Φ) := sup ln t = lim+ ln t where hΦ (t) := sup Φ(st)
Φ(s) for t ∈ (0, ∞), the lower
0<t<1 t→0 s>0
dilation index of the Young function Φ (cf. [69, (5.3.14)])
ln hΦ (t) ln hΦ (t)
I(Φ) := inf ln t = lim ln t where hΦ (t) := sup Φ(st)
Φ(s) for t ∈ (0, ∞), the
1<t<∞ t→∞ s>0
upper dilation index of the Young function Φ (cf. [69, (5.3.15)])
K
KΔ , the boundary-to-boundary harmonic double layer potential, defined as (cf. [68,
(1.1.32)]):
∫
1 ν(y), y − x
KΔ f (x) := lim+ f (y) dH n−1 (y), x ∈ ∂Ω
ε→0 ωn−1 ∂Ω\B(x,ε) |x − y| n
(A.0.101)
KΔ# , the transpose harmonic double layer potential, defined as (cf. [68, (1.1.33)]):
∫
1 ν(x), x − y
KΔ f (x) := lim+
#
f (y) dH n−1 (y), x ∈ ∂Ω
ε→0 ωn−1 ∂Ω\B(x,ε) |x − y| n
(A.0.102)
Ker (T : X → Y ) := {x ∈ X : T x = 0}, the kernel (or null-space) of the operator T

M
Ker L := u ∈ 𝒞∞ (Ω) : Lu = 0 in Ω , the null-space of the M × M system L,
in an open set Ω
K = K A, the boundary-to-boundary double layer a given second-order M × M
αβ
elliptic system written as L A := ar s ∂r ∂s 1≤α,β ≤M corresponding to a choice of
αβ
the coefficient tensor A := ar s 1≤α,β ≤M (with canonical fundamental solution
1≤r,s ≤n
946 A Terms and notation used in Volume IV

E = Eγβ 1≤γ,β ≤M ) acting on any function f = ( fα )1≤α ≤M ∈ L 1 ∂Ω, 1+σ(x)

M
|x | n−1
at σ-a.e. point x ∈ ∂∗ Ω as in (1.3.68):
 ∫ 
βα
K f (x) := − lim+ νs (y)ar s (∂r Eγβ )(x − y) fα (y) dσ(y)
ε→0 1≤γ ≤M
y ∈∂∗ Ω
|x−y |>ε

Kmod , the modified boundary-to-boundary double layer potential operator for some
αβ
second-order M × M weakly elliptic system written as L A := ar s ∂r ∂s 1≤α,β ≤M
αβ
corresponding to a choice of the coefficient tensor A := ar s 1≤α,β ≤M (with fun-
1≤r,s ≤n
damental solution E = Eγβ 1≤γ,β ≤M ) acting on any f = ( fα )1≤α ≤M belonging to
 M
L 1 ∂∗ Ω, 1+σ(x)
|x | n at σ-a.e. point x ∈ ∂Ω as in (1.8.24):

 ∫ 
βα (rγβ) (rγβ)
Kmod f (x) := − lim+ νs (y)ar s {k ε (x − y) − k1 (−y)} fα (y) dσ(y)
ε→0 1≤γ ≤M
∂∗ Ω

(rγβ)
where k ε := (∂r Eγβ ) · 1Rn \B(0,ε) for each ε > 0
K# = K A# ,the transpose double layer for a given second-order M × M weakly
αβ
elliptic system written as L A := ar s ∂r ∂s 1≤α,β ≤M corresponding to a choice of
αβ
the coefficient tensor A := ar s 1≤α,β ≤M (with canonical fundamental solution
1≤r,s ≤n
E = Eγβ 1≤γ,β ≤M ) acting on any function f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
at σ-a.e. point
x ∈ ∂∗ Ω as in (1.3.72):
 ∫ 
βα
K f (x) := lim+
#
νs (x)ar s (∂r Eγβ )(x − y) fγ (y) dσ(y)
ε→0 1≤α ≤M
y ∈∂Ω
|x−y |>ε

Ker D, the null-space of a first-order N × M constant complex coefficient system

 n 
αβ
D= a j ∂j 1≤α ≤ N in Rn which is injectively elliptic (i.e., its symbol, given by
j=1 1≤β ≤M
n
αβ
the formula Sym(D; ξ) := i aj ξj 1≤α ≤ N is an injective mapping from C M into
j=1 1≤β ≤M
C N , for each vector ξ ∈ Rn \ {0}), defined as in (4.4.62):

Ker D := u ∈ 𝒞∞ (Ω)
M
: Du = 0 in Ω

Kλ , the boundary-to-boundary double layer for the Stokes system associated with
the coefficient tensor Aλ := δ jk δαβ + λδ jβ δkα 1≤α,β ≤n corresponding to any λ ∈ C,
1≤ j,k ≤n
A Terms and notation used in Volume IV 947
  n
acting on each f = ( f j )1≤ j ≤n ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at σ-a.e. point x ∈ ∂∗ Ω as in
(6.2.95), (6.2.96):
! ∫ 
δ jγ x − y, ν(y)

Kλ f (x) = lim+ (λ − 1)
ε→0 2ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

n(λ + 1) x − y, ν(y) (x j − y j )(xγ − yγ )

−
2ωn−1 |x − y| n+2
 "
λ − 1 ν j (y)(xγ − yγ ) − νγ (y)(x j − y j )
+ f j (y) dσ(y)
2ωn−1 |x − y| n
1≤γ ≤n

Kλ# , the transpose double layer for the Stokes system associated with the coefficient
tensor Aλ := δ jk δαβ + λδ jβ δkα 1≤α,β ≤n corresponding to any λ ∈ C, acting on
  1≤ j,k ≤nn
 σ(x)
each f = ( f j )1≤ j ≤n ∈ L ∂∗ Ω, 1+ |x | n−1
1 at σ-a.e. point x ∈ ∂∗ Ω as in (6.2.98),
(6.2.99):
! ∫ 
#  δ jγ x − y, ν(x)
Kλ f (x) = lim+ − (λ − 1)
ε→0 2ωn−1 |x − y| n
y ∈∂Ω
|x−y |>ε

n(λ + 1) x − y, ν(x) (x j − y j )(xγ − yγ )

+
2ωn−1 |x − y| n+2
 "
λ − 1 ν j (x)(xγ − yγ ) − νγ (x)(x j − y j )
− fγ (y) dσ(y)
2ωn−1 |x − y| n
1≤ j ≤n

Knβ (·, ·), the Bochner-Martinelli kernel for (0, β)-forms in Cn with β ∈ {0, 1, . . . , n},
defined as in (7.3.27):
Knβ (ζ, z) := − ∗ ∂ζ Γ0,β (ζ, z)
with the Hodge star isomorphism applied in ζ
k x0 , the Poisson kernel for the Laplacian in the domain Ω, defined as

dω x0
k x0 := , (A.0.103)
dσ
i.e., the Radon-Nikodym derivative of the harmonic measure ω x0 with pole at x0 ∈ Ω
with respect to the surface measure σ := H n−1 ∂Ω on the topological boundary ∂Ω
L
local John condition, satisfied by an open set Ω ⊆ Rn (cf. [68, Definition 5.11.7]):
948 A Terms and notation used in Volume IV

there exist θ ∈ (0, 1), Mo ∈ (1, ∞), and R ∈ 0, diam ∂Ω (the

latter required to be ∞ if ∂Ω is unbounded) such that for every
x ∈ ∂Ω and r ∈ (0, R∗ ) one can find a point xr ∈ B(x, r) ∩ Ω
with the property that B(xr , θr) ⊆ Ω and for each y ∈ Δ(x, r) it (A.0.104)
is possible to find a rectifiable path γy : [0, 1] → Ω whose length
is ≤ Mo · r, which satisfies γy (0) = y, γy (1) = xr , and such that
dist γy (t), ∂Ω > θ · |γy (t) − y| for every t ∈ (0, 1]

L n , the (n-dimensional) Lebesgue measure in Rn

∞ , the space of essentially bounded functions with compact support
Lcomp
L 0 (X, μ), the space of measurable functions which are pointwise finite μ-a.e. on X
p
Lbdd (Ω, L n ), the space of functions which are p-th power integrable on bounded
subsets of Ω (cf. [68, (4.2.4)]):
p
Lbdd (Ω, L n ) be the collection of all L n -measurable functions defined
in Ω which are p-th power absolutely integrable with respect to the
Lebesgue measure on each bounded L n -measurable subset of the set Ω.
(A.0.105)
 · Lip(X) , the natural semi-norm on Lip(X), defined in the context of a metric space
(𝒳, d) as (cf. [68, (3.7.1)]):

| f (x) − f (y)|
 f Lip(X) := sup (A.0.106)
x,y ∈X, xy d(x, y)

Lip(X), the space of Lipschitz functions on the (quasi-)metric space X, defined as

(cf. [68, (3.7.2)]):
 
Lip(X) := f : X → C :  f Lip(X) < +∞ (A.0.107)

Lipc (X), the space of Lipschitz functions with bounded support in the (quasi-)metric
space X

Lipc (Σ) , the space distributions on a given set Σ ⊆ Rn , defined as (cf. [68,
(4.1.34)]):
the topological dual of Lipc (Σ), τ𝒟 (A.0.108)

(Lip c (Σ)) ·, · Lip c (Σ) , or simply ·, · , the distributional pairing on the set Σ
· L p, q (X,μ), the Lorentz space quasi-norm, defined as (cf. [68, (6.2.14)]):

⎧ ∫ ∞ q dt  1/q
⎪
⎪ t 1/p fX∗ (t) if 0 < p, q < ∞,
⎪
⎪
⎨ 0
⎪ t
 f  L p, q (X,μ) := sup 1/p f ∗ (t) if 0 < p ≤ ∞, q = ∞, (A.0.109)
⎪
⎪ t>0 t
⎪
⎪
X
⎪f ∞ if p = ∞, 0 < q ≤ ∞
⎩ L (X,μ)
A Terms and notation used in Volume IV 949

L p,q (X, μ), the Lorentz space on X with respect to the measure μ defined as (cf. [68,
(6.2.13)]):

L p,q (X, μ) := f : X → R μ-measurable :  f  L p, q (X,μ) < +∞ (A.0.110)

p,q
L (Ω, μ), the maximal Lorentz space with respect to the Borel measure μ in the
open set Ω ⊆ Rn , defined as (cf. [68, (6.6.41)]):
p,q  
L (Ω, μ) := u : Ω → C : u is L n -measurable and u,θ ∈ L p,q (Ω, μ)
(A.0.111)
p
L (Ω, μ), the maximal Lebesgue space with respect to the Borel measure μ in the
open set Ω ⊆ Rn , defined as (cf. [68, (6.6.43)]):
p p, p
L (Ω, μ) := L (Ω, μ) (A.0.112)
 
= u : Ω → C : u is L n -measurable and u,θ ∈ L p (Ω, μ)

log+ , the positive part of ln, defined for each t ∈ [0, ∞) as (cf. [68, (7.6.68)]):

0 if t ∈ [0, 1],
log+ t := (A.0.113)
ln t if t ∈ [1, ∞)

L(X → Y ), the space of linear and continuous operators from X to Y

L Φ (X, μ), the Orlicz space on the sigma-finite measure space (X, M, μ) associated
with a Young function Φ (cf. [69, §5.3]):
 
L Φ (X, μ) := f ∈ ℳ(X, μ) :  f  L Φ (X,μ) < ∞ (A.0.114)

where the Luxemburg norm  ·  L Φ (X,μ) is defined as

 ∫
 f  L Φ (X,μ) := inf λ > 0 : Φ(| f (x)|/λ) dμ(x) ≤ 1 (A.0.115)
X
 ∫ 
∞ 
inf λ > 0 : μ x ∈ X : | f (x)| > λΦ−1 (t) dt ≤ 1 (A.0.116)
0

 ∫ α
L p (log L)α (X, μ) = f ∈ ℳ(X, μ) : X | f (x)| p ln(e + | f (x)|) dμ(x) < +∞ for
p ∈ (1, ∞) and α ∈ R, Zygmund’s space (cf. [69, §5.3])
L p (Ω, wL n ), the weighted L p Lebesgue space on the set Ω ⊆ Rn , equipped with
∫  1/p
the natural norm u L p (Ω, w L n ) := Ω |u| p w dL n (cf. [69, §8.3])
 
Ls (Ω) := U Ω : U ∈ Ls (Rn ) , the Bessel potential space in an open set Ω ⊆ Rn
p p

for p ∈ (1, ∞) and s ∈ R, equipped with the norm (cf. [69, §9.2]):
950 A Terms and notation used in Volume IV
 
u Lsp (Ω) := inf U  Lsp (Rn ) : U ∈ Ls (Rn ), u = U Ω
p

.
L p,λ (Σ, σ), the homogeneous Morrey-Campanato space on a given Ahlfors regular
set Σ ⊆ Rn , defined for each integrability exponent p ∈ (1, ∞) and each parameter
λ ∈ (0, n − 1) as (cf. [69, §6.1]):
. 
L p,λ (Σ, σ) := f ∈ Lloc1
(Σ, σ) :  f L. p, λ (Σ,σ) < +∞ (A.0.117)

where, for each f ∈ Lloc

1 (Σ, σ),

 
n−1−λ ⨏   1
 f  . p, λ := sup R p  f (y) − fΔ(x,R)  p dσ(y) p
L (Σ,σ)
x ∈Σ and Σ∩B(x,R)
0<R<2 diam(Σ)
(A.0.118)

L p,λ (Σ, σ), the inhomogeneous Morrey-Campanato space equippe on a given Ahlfors
regular set Σ ⊆ Rn , defined for each integrability exponent p ∈ (1, ∞) and each
parameter λ ∈ (0, n − 1) as (cf. [69, §6.1]):
. 
L p,λ (Σ, σ) := L p (Σ, σ) ∩ L p,λ (Σ, σ) = f ∈ L p (Σ, σ) :  f L. p, λ (Σ,σ) < +∞
(A.0.119)

where
 f L p, λ (Σ,σ) :=  f  L p (Σ,σ) +  f L. p, λ (Σ,σ) (A.0.120)
p
L1 (∂∗ Ω, σ∗ ), the L p -based Sobolev space of order one on ∂∗ Ω, defined for each
p ∈ [1, ∞] as (cf. [69, Definition 11.1.2]):

p
L1 (∂∗ Ω, σ∗ ) := f ∈ L p (∂∗ Ω, σ∗ ) : ∂τ j k f exists in L p (∂∗ Ω, σ∗ )

for each j, k ∈ {1, . . . , n} (A.0.121)

and equipped with the norm (cf. [69, Proposition 11.1.9]):


p
L1 (∂∗ Ω, σ∗ )  f →  f  L p (∂∗ Ω,σ∗ ) :=  f  L p (∂∗ Ω,σ∗ ) + ∂τ j k f  L p (∂∗ Ω,σ∗ )
1
1≤ j,k ≤n
(A.0.122)
p,q
L1 (∂∗ Ω, σ∗ ), the off-diagonal Sobolev space of order one on ∂∗ Ω, defined for each
p, q ∈ [1, ∞] as (cf. [69, Definition 11.1.2]):

p,q
L1 (∂∗ Ω, σ∗ ) := f ∈ L p (∂∗ Ω, σ∗ ) : ∂τ j k f exists in L q (∂∗ Ω, σ∗ )

for each j, k ∈ {1, . . . , n} (A.0.123)

A Terms and notation used in Volume IV 951
p,q
and equipped with the norm define for each f ∈ L1 (∂∗ Ω, σ∗ ) as (cf. [69, Proposi-
tion 11.1.9]):

 f  L p, q (∂∗ Ω,σ∗ ) :=  f  L p (∂∗ Ω,σ∗ ) + ∂τ j k f  L q (∂∗ Ω,σ∗ ) (A.0.124)
1
1≤ j,k ≤n

p,q
L1,loc (∂∗ Ω, σ∗ ), the local off-diagonal (boundary) Sobolev space defined for each
pair of exponents p, q ∈ [1, ∞] as (cf. [69, Definition 11.1.2]):

p,q p q
L1,loc (∂∗ Ω, σ∗ ) := f ∈ Lloc (∂∗ Ω, σ∗ ) : ∂τ j k f exists in Lloc (∂∗ Ω, σ∗ )

for each j, k ∈ {1, . . . , n} (A.0.125)

p
L1,loc (∂∗ Ω, σ∗ ), the (boundary) Sobolev space defined for the exponent p ∈ [1, ∞] as
(cf. [69, Definition 11.1.2]):
p p, p
L1,loc (∂∗ Ω, σ∗ ) := L1,loc (∂∗ Ω, σ∗ ) (A.0.126)

.p
L1 (∂Ω, σ), the L p -based homogeneous Sobolev space of order one on ∂Ω as (cf.
[69, Definition 11.5.3]):

.p  σ(x) 
L1 (∂Ω, σ) := f ∈ L 1 ∂Ω, : ∂τ j k f ∈ L p (∂∗ Ω, σ) (A.0.127)
1 + |x| n

for each j, k ∈ {1, . . . , n}

and equipped with the semi-norm

.p 
n
L1 (∂Ω, σ)  f →  f  L. p (∂Ω,σ) := ∂τ j k f  L p (∂∗ Ω,σ) (A.0.128)
1
j,k=1

p
L1 (∂∗ Ω, wσ∗ ), the weighted Sobolev space of order one on ∂∗ Ω, defined for an
exponent p ∈ [1, ∞] and a generic weight w on ∂∗ Ω as (cf. [69, §11.7]):
p 
L1 (∂∗ Ω, wσ∗ ) := f ∈ L p (∂∗ Ω, wσ∗ ) : ∂τ j k f ∈ L p (∂∗ Ω, wσ∗ )

for each j, k ∈ {1, . . . , n} (A.0.129)

equipped with the norm


n
p
L1 (∂∗ Ω, wσ∗ )  f −→  f  L p (∂∗ Ω,wσ∗ ) + ∂τ j k f  L p (∂∗ Ω,wσ∗ ) (A.0.130)
j,k=1
952 A Terms and notation used in Volume IV
p σ∗ (x)
L1 ∂∗ Ω, 1+ |x | a , the “power” weighted Sobolev space on ∂∗ Ω, defined for each
p ∈ [1, ∞] and a ∈ [0, ∞) as (cf. [69, (11.7.6)]):

p σ∗ (x) σ∗ (x) σ∗ (x)
L1 ∂∗ Ω, 1+ |x | a = f ∈ L ∂∗ Ω, 1+ |x | a : ∂τ j k f ∈ L ∂∗ Ω, 1+ |x | a
p p
(A.0.131)

for each j, k ∈ {1, . . . , n}

p
L1 (∂Ω, w), the Muckenhoupt weighted Sobolev space of order one on ∂Ω, defined
for p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ) as (cf. [69, §11.7]):
p  
L1 (∂Ω, w) := f ∈ L p (∂Ω, w) : ∂τ j k f ∈ L p (∂Ω, w), 1 ≤ j, k ≤ n (A.0.132)

and equipped with the norm


n
p
L1 (∂Ω, w)  f →  f  L p (∂Ω,w) :=  f  L p (∂Ω,w) + ∂τ j k f  L p (∂Ω,w) (A.0.133)
1
j,k=1

p,q
L1 (∂Ω, σ), the Lorentz-based Sobolev space of order one on ∂Ω, defined for
p ∈ (1, ∞) and q ∈ (0, ∞] as (cf. [69, §11.7]):
p,q  
L1 (∂Ω, σ) := f ∈ L p,q (∂Ω, σ) : ∂τ j k f ∈ L p,q (∂Ω, σ), 1 ≤ j, k ≤ n
(A.0.134)
and equipped with the quasi-norm

n
p,q
L1 (∂Ω, σ)  f →  f  L p, q (∂Ω,σ) :=  f  L p, q (∂Ω,σ) + ∂τ j k f  L p, q (∂Ω,σ)
1
j,k=1
(A.0.135)
p
L−1 (∂∗ Ω, σ∗ ),
the negative Sobolev space of order minus one on ∂∗ Ω
(L p -based)

defined for p, p ∈ (1, ∞) with p1 + p1 = 1 as (cf. [69, Definition 11.8.1]):
  ∗
p p
L−1 (∂∗ Ω, σ∗ ) := L1 (∂∗ Ω, σ∗ ) (A.0.136)

p,q
L−1 (∂∗ Ω, σ∗ ), the off-diagonal negative Sobolev space on ∂∗ Ω defined for each two
exponents p, q ∈ (1, ∞) as (cf. [69, (11.8.28)]):
   ∗
p,q p ,q
L−1 (∂∗ Ω, σ∗ ) := L1 (∂∗ Ω, σ∗ ) where p1 + p1 = 1 and q1 + q1 = 1 (A.0.137)

p
L−1 (∂Ω, w), the Muckenhoupt weighted negative Sobolev space on ∂Ω defined for
p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ) as (cf. [69, Definition 11.8.7]):
  ∗
L−1 (∂Ω, w) := L1 (∂Ω, w )
p p
(A.0.138)
A Terms and notation used in Volume IV 953

where p := (1 − 1/p)−1 ∈ (1, ∞) is the Hölder conjugate exponent of the given p


and where w  := w 1−p ∈ Ap (∂Ω, σ) is the conjugate weight of w
αβ
L = ar s ∂r ∂s 1≤α,β ≤M , a homogeneous, second-order, constant (complex) coeffi-
cient, M × M system in Rn
L A, the second-order homogeneous constant coefficient system in Rn associated with
αβ
a given coefficient tensor A := ar s 1≤r,s ≤n as
1≤α ≤M
1≤β ≤ N
 
αβ
L A := ar s ∂r ∂s 1≤α ≤M (A.0.139)
1≤β ≤ N

L(ξ), the characteristic matrix of the homogeneous (constant-coefficient) higher-

order system L = ∂ α Aαβ ∂ β , defined as (cf. [68, (6.5.39)]):
|α |= |β |=m

L(ξ) := (−1)m ξ α+β Aαβ, ∀ξ ∈ Rn (A.0.140)
|α |= |β |=m

L(ξ), the characteristic matrix of the homogeneous complex constant coefficient

αβ
second-order M × M system L = ar s ∂r ∂s 1≤α,β ≤M in Rn , defined for each vector
ξ = (ξr )1≤r ≤n ∈ Rn as
αβ
L(ξ) := − ar s ξr ξs 1≤α,β ≤M (A.0.141)

Lλ,μ , the complex Lamé system with Lamé moduli λ, μ ∈ C defined as:

Lλ,μ := μ Δ + (λ + μ)∇div (A.0.142)

Lλ,μ (ξ) := −μ|ξ | 2 In×n − (λ + μ)ξ ⊗ ξ for each ξ ∈ Rn , the characteristic matrix of
the complex Lamé system Lλ,μ := μ Δ + (λ + μ)∇div
L A,z (∂Ω, σ), the space defined for p, p ∈ (1, ∞) with 1/p + 1/p = 1 and z ∈ C as
p

in (1.7.14):
 ∫
p M
L A,z (∂Ω, σ) := f ∈ L (∂Ω, σ)
p
: f , g dσ = 0 for each
∂Ω
 M
g ∈ L p (∂Ω, σ) with (zI + K A)g = 0

Lλ , the Stokes system in the special writing (for λ ∈ C) as in (6.1.2):

Lλ := (δ jk δαβ + λδ jβ δkα )∂j ∂k 1≤α,β ≤n

LB , the Bitsadze operator, defined as in (1.4.77):

LB := ∂z̄2 = 14 (∂x + i∂y )2

954 A Terms and notation used in Volume IV
p,1
LC,loc (∂∗ Ω, σ), the space defined for p ∈ [1, ∞] as (7.1.7):

p,1 p p
LC,loc (∂∗ Ω, σ) := f ∈ Lloc (∂∗ Ω, σ) : ∂τCj k f belongs to Lloc (∂∗ Ω, σ)

for each j, k ∈ {1, . . . , n}

p,1
LC (∂∗ Ω, σ), the complex boundary Sobolev space, defined for p ∈ [1, ∞] as in
(7.1.11):

LCp,1 (∂∗ Ω, σ) := f ∈ L p (∂∗ Ω, σ) : ∂τCj k f belongs to L p (∂∗ Ω, σ)

for each j, k ∈ {1, . . . , n}

Λα,β Cn , the space of differential forms of (type) degree (α, β) with complex coef-
ficients, defined for α, β ∈ {0, 1, . . . , n} as in (7.1.36), namely the collection of all
u = u I,J dz I ∧ d z̄ J where the sum is performed over strictly increasing
|I |=α, |J |=β
arrays and u I,J ∈ C for all I, J
L p (X, μ) ⊗ Λα,β := L p (X, μ) ⊗ Λα,β Cn where (X, μ) is a given measure space,
α, β ∈ {0, 1, . . . , n} and p is an exponent in (0, ∞], the space of differential form-
s of type (α, β) with coefficients from L p (X, μ), equipped with the quasi-norm
 f  L p (X,μ)⊗Λα, β :=  fI,J  L p (X,μ) if f = fI,J dz I ∧d z̄ J (7.1.77),
|I |=α, |J |=β |I |=α, |J |=β
(7.1.78)
Ltan,C (∂∗ Ω, σ) ⊗ Λα,β , the complex tangential forms defined for α, β ∈ {0, 1, . . . , n}
p

and p ∈ (0, ∞] as the collection of all f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β satisfying ν 1,0 ∨ f = 0

at σ-a.e. point on ∂∗ Ω (7.1.79), (7.1.64)
Lnor,C (∂∗ Ω, σ) ⊗ Λα,β , the complex normal forms defined for α, β ∈ {0, 1, . . . , n}
p

and p ∈ (0, ∞] as the collection of all f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β satisfying ν 0,1 ∧ f = 0

at σ-a.e. point on ∂∗ Ω (7.1.80), (7.1.65)
LC (∂∗ Ω, σ) ⊗ Λα,β , the space of differential forms of type (α, β) with coefficients
p,1
p,1
from LC (∂∗ Ω, σ) and equipped with the norm (cf. (7.1.84), (7.1.85))
  
n
 C 
 f  L p,1 (∂∗ Ω,σ)⊗Λα, β :=  f  L p (∂∗ Ω,σ)⊗Λα, β + ∂τ fK,I  p
C
jk L (∂∗ Ω,σ)
|K |=α |I |=β j,k=1

for each f = |I |=β fK,I dz K ∧ d z̄ I

|K |=α
1, p,q
LC (∂∗ Ω, σ), the off-diagonal (partial) Sobolev space defined for p, q ∈ [1, ∞]
as the collection of all f ∈ L p (∂∗ Ω, σ) satisfying ∂τCj k f ∈ L q (∂∗ Ω, σ) for all
1 ≤ j, k ≤ n, equipped with the norm (cf. (7.5.27), (7.5.28))
A Terms and notation used in Volume IV 955
  
 f  L 1, p, q (∂∗ Ω,σ) :=  f  L p (∂∗ Ω,σ) + ∂τC f  q
C
jk L (∂∗ Ω,σ)
1≤ j,k ≤n

p, ∂¯ p, ∂¯
Lα,β b (∂Ω, σ) := Lα,β τ (∂Ω, σ) ∩ Ltan,C (∂Ω, σ) ⊗ Λα,β for α, β ∈ {0, 1, . . . , n} and
p

p ∈ [1, ∞) (7.7.7)
p, ∂¯
Lα,β τ (∂∗ Ω, σ), the space defined for p ∈ [1, ∞] as the collection of all forms f in
L p (∂∗ Ω, σ) ⊗ Λα,β with the property that ∂¯τ f ∈ L p (∂∗ Ω, σ) ⊗ Λα,β+2 (7.2.27)
M
M X,s,α , the L s -based fractional Hardy-Littlewood maximal operator of order α in
the space of homogeneous type (X, ρ, μ), defined for each μ-measurable function f
on X as (cf. [68, (7.6.1)]):
0 1
⨏  s1
α
M X,s,α f (x) := sup μ(Bρ (x, r)) | f | dμ
s
, ∀x ∈ X (A.0.143)
r >0 B ρ (x,r)

M X,s , the L s -based Hardy-Littlewood maximal operator in the space of homoge-

neous type (X, ρ, μ), defined for each μ-measurable function f on X as (cf. [68,
(7.6.7)]):
⨏  s1
M X,s f (x) := sup | f | s dμ , ∀ x ∈ X (A.0.144)
r >0 B ρ (x,r)

M X , the Hardy-Littlewood maximal operator on the space of homogeneous type

(X, ρ, μ), defined for each μ-measurable function f on X as (cf. [68, (7.6.16)]):
∫
1
M X f (x) := sup | f | dμ, ∀ x ∈ X (A.0.145)
r ∈(0,∞) μ Bρ (x, r) B ρ (x,r)

M , the tangential maximal function of u (with exponent M), defined at each x ∈ ∂Ω

umax
as (cf. [68, (8.5.2)]):

M : ∂Ω −→ [0, +∞] defined by

umax
  δ (y)  M  (A.0.146)
 ∂Ω 
umax (x) := u(y)  ∞ for each x ∈ ∂Ω
M
|x − y| L y (Ω, L n )

ℳ(X, μ), the collection of μ-measurable functions defined on an arbitrary measure

space (X, M, μ)
ℳ+ (X, μ), the collection of non-negative μ-measurable functions defined on an
arbitrary measure space (X, M, μ)
M p,λ (Σ, σ), the Morrey space on the Ahlfors regular set Σ ⊆ Rn , defined for each
p ∈ (0, ∞) and λ ∈ (0, n − 1) as (cf. [69, §6.2]):
956 A Terms and notation used in Volume IV

M p,λ (Σ, σ) := f : Σ → C : f is σ-measurable and  f  M p, λ (Σ,σ) < +∞
(A.0.147)
where, for each σ-measurable function f on Σ,
 
n−1−λ  ⨏  p1
 f  M p, λ (Σ,σ) := sup R p | f | dσ
p
(A.0.148)
x ∈Σ and Σ∩B(x,R)
0<R<2 diam(Σ)

M̊ p,λ (Σ, σ), the vanishing Morrey space on the Ahlfors regular set Σ ⊆ Rn with
p ∈ (0, ∞) and λ ∈ (0, n − 1) (cf. [69, §6.2]):
p(n−1)
M̊ p,λ (Σ, σ) := the closure of L s (Σ, σ) with s := n−1−λ in M p,λ (Σ, σ) (A.0.149)

p,λ
M1 (∂Ω, σ), the Morrey-based Sobolev space of order one on ∂Ω, defined for
p ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, (11.7.12)-(11.7.13)]):

p,λ
M1 (∂Ω, σ) := f ∈ M p,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.150)

one has ∂τ j k f ∈ M p,λ (∂Ω, σ)

equipped with the norm


n
 
p,λ
M1 (∂Ω, σ)  f −→  f  M p, λ (∂Ω,σ) + ∂τ f  p, λ (A.0.151)
jk M (∂Ω,σ)
j,k=1

p,q,λ
M1 (∂Ω, σ), the off-diagonal Morrey-based Sobolev space of order one on ∂Ω,
defined for p, q ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, (11.7.14)-(11.7.15)]):

p,q,λ
M1 (∂Ω, σ) := f ∈ M p,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.152)

one has ∂τ j k f ∈ M q,λ (∂Ω, σ) ,

and endowed with the norm


n
 
M1
p,q,λ
(∂Ω, σ)  f −→  f  M p, λ (∂Ω,σ) + ∂τ f  q, λ (A.0.153)
jk M (∂Ω,σ)
j,k=1

p,q,λ
M̊1 (∂Ω, σ), the vanishing off-diagonal Morrey-based Sobolev space of order one
on ∂Ω, defined for p, q ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, (11.7.16)]):

p,q,λ
M̊1 (∂Ω, σ) := f ∈ M̊ p,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.154)

one has ∂τ j k f ∈ M̊ q,λ (∂Ω, σ)

A Terms and notation used in Volume IV 957
p,λ
M̊1 (∂Ω, σ), the vanishing Morrey-based Sobolev space of order one on ∂Ω, defined
for p ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, (11.7.17)]):

p,λ
M̊1 (∂Ω, σ) := f ∈ M̊ p,λ (∂Ω, σ) : for each j, k ∈ {1, . . . , n} (A.0.155)

one has ∂τ j k f ∈ M̊ p,λ (∂Ω, σ)

p,λ
M−1 (∂Ω, σ), the Morrey-based negative Sobolev space on ∂Ω, defined for any
p, q ∈ (1, ∞) with 1/p + 1/q = 1 and λ ∈ (0, n − 1) as (cf. [69, Definition 11.8.9]):
 ∗
p,λ q,λ
M−1 (∂Ω, σ) := B1 (∂Ω, σ) (A.0.156)

. p,λ
M1 (∂Ω, σ), the homogeneous Morrey-based Sobolev space of order one on ∂Ω
defined for p ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, Definition 11.13.1]):

. p,λ  σ(x)  p
M1 (∂Ω, σ) := f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) : ∂τ j k f ∈ M p,λ (∂Ω, σ)
1 + |x| n

for 1 ≤ j, k ≤ n

(A.0.157)

and equipped with the semi-norm

. p,λ 
n
M1 (∂Ω, σ)  f →  f  M. p, λ (∂Ω,σ) := ∂τ j k f  M p, λ (∂Ω,σ) (A.0.158)
1
j,k=1

. p,λ
M1 (∂Ω, σ), the homogeneous vanishing Morrey-based Sobolev space of order one
on ∂Ω, defined for p ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [69, Definition 11.13.15]):

. p,λ  σ(x)  p
M1 (∂Ω, σ) := f ∈ L 1 ∂Ω, ∩ Lloc (∂Ω, σ) : ∂τ j k f ∈ M̊ p,λ (∂Ω, σ)
1 + |x| n

for 1 ≤ j, k ≤ n

(A.0.159)

and equipped with the semi-norm

. p,λ 
n
M1 (∂Ω, σ)  f →  f M. p, λ (∂Ω,σ) := ∂τ j k f  M p, λ (∂Ω,σ) (A.0.160)
1
j,k=1

Mb , the operator of pointwise multiplication by the function b ∈ L ∞ (Σ, σ)

958 A Terms and notation used in Volume IV

[Mb, T], the commutator of Mb (the pointwise multiplication by b) with the operator
T, defined by [Mb, T] f := b(T f ) − T(b f )
N
N0 := N ∪ {0} = {0, 1, 2, . . . }
NTA domain: a nonempty open subset Ω of Rn satisfying a two-sided corkscrew
condition as well as a Harnack chain condition (cf. [68, Definition 5.11.1])
two-sided NTA domain: a nonempty open subset Ω of Rn satisfying a two-sided
corkscrew condition as well as a two-sided Harnack chain condition (cf. [68, Defi-
nition 5.11.1])
one-sided NTA domain (or interior NTA domain): a nonempty open subset Ω of Rn
satisfying an interior corkscrew condition as well as a Harnack chain condition (cf.
[68, Definition 5.11.1])
Nκ u, the (κ-)nontangential maximal operator acting on the function measurable
u : Ω → Rn according to (cf. [68, (8.2.1)]):

Nκ u : ∂Ω −→ [0, +∞], (Nκ u)(x) := u L ∞ (Γκ (x), L n ) for all x ∈ ∂Ω (A.0.161)

Nκε u, the (κ-)nontangential maximal operator truncated at height ε > 0, acting on

the function u : Ω → Rn according to (cf. [68, (1.5.5)]):
 
Nκε u := Nκ (u · 1 Oε ) where Oε := x ∈ Ω : dist(x, ∂Ω) < ε . (A.0.162)

NκE u, the restricted nontangential maximal function of u : Ω → R, relative to the

set E (cf. [68, (8.2.4)]):

NκE u : ∂Ω −→ [0, +∞] defined as

(A.0.163)
(NκE u)(x) := u L ∞ (Γκ (x)∩E, L n ) for each x ∈ ∂Ω

ν, the geometric measure theoretic outward unit normal to a set Ω ⊆ Rn of locally

finite perimeter, defined via the requirement that (cf. [68, (5.6.2)-(5.6.3)]):
there exist a locally finite Borel-regular measure σ∗ in Rn and a
n
vector-valued function ν ∈ L ∞ (Rn, σ∗ ) satisfying |ν(x)| = 1 at (A.0.164)
n
σ∗ -a.e. x ∈ Rn and with such that ∇1Ω = −νσ∗ in D (Rn )

νg , the geometric measure theoretic outward unit normal induced by the metric
tensor g
ν E , the geometric measure theoretic outward unit normal induced by the standard
Euclidean metric
ν • F,  the “bullet” product involving a vector field F ∈ L 1 (Ω, L n ) n (where Ω
bdd
is an arbitrary open subset of Rn ) whose divergence, considered in the sense of
distributions in Ω, satisfies divF ∈ Lbdd
1 (Ω, L n ), defined as a functional acting on
A Terms and notation used in Volume IV 959

each ψ ∈ Lipc (∂Ω) according to (cf. [68, Proposition 4.2.3]):

∫ ∫
 
 ψ :=
ν • F, F · ∇Ψ dL n + (divF)Ψ  dL n, (A.0.165)
Ω Ω

where Ψ is any complex-valued function satisfying


Ψ ∈ Lip(Ω), Ψ∂Ω = ψ, and Ψ ≡ 0 outside
(A.0.166)
of some compact subset of Ω

ν
• u, the Clifford bullet product of ν with u (cf. [69, (10.2.100)]):

ν
• u := (−i)Sym(D; ν) • u (A.0.167)

p
Nκ (Ω; μ), the space of measurable functions in Ω with a p-th power integrable non-
tangential maximal function on ∂Ω with respect to the measure μ (cf. [68, (8.3.31)]):

p 
Nκ (Ω; μ) := u : Ω → C : u is L n -measurable, and

u Nκp (Ω;μ) := Nκ u L p (∂Ω, μ) < +∞ (A.0.168)

κ−n.t.
u|∂Ω (x), the nontangential trace of the function u : Ω → R at the point x ∈ ∂Ω
such that x ∈ Γκ (x), defined as (cf. [68, Definition 8.9.1]):
κ−n.t.
u|∂Ω (x) is the number a ∈ R with the property that for every
ε > 0 there exists some r > 0 such that |u(y) − a| < ε for L n -a.e. (A.0.169)
point y ∈ Γκ (x) ∩ B(x, r)

T X→Y , the operator “norm” of a positively homogeneous mapping T acting from

the quasi-normed vector space X,  · X with values in the quasi-normed vector
space Y,  · Y
 
T X→Y := sup TuY : u ∈ X, uX = 1 ∈ [0, +∞] (A.0.170)

T Bd(X→Y) := T X→Y
T Bd(X) , the “norm” of a positively homogeneous mapping T acting from the quasi-
normed vector space X,  · X into itself

T Bd(X) := T Bd(X→X) = sup T xX (A.0.171)

x ∈X, x  X =1

ess
T X→Y , the essential norm of the operator T ∈ Bd(X → Y ), where X, Y are
quasi-normed spaces (cf. [69, §1.2]):
960 A Terms and notation used in Volume IV
ess
T X→Y := dist T, Cp(X → Y )
 
= inf T − K X→Y : K ∈ Cp(X → Y ) (A.0.172)

 · (X0,X1 )θ, q , the real interpolation quasi-norm

νC , the complex outward unit normal of a set of locally finite perimeter Ω ⊆ R2n ≡ Cn
defined as in (7.1.60):

νC := ν1 + iν2, . . . , ν2n−1 + iν2n ∈ Cn

if ν = (ν1, ν2, . . . , ν2n−1, ν2n ) ∈ R2n is the geometric measure theoretic unit normal
to Ω
O
1E , the characteristic function of a given set E
ωn−1 := H n−1 (S n−1 ), the surface area of S n−1 (the (n − 1)-dimensional sphere in
Rn )
 
Oε := x ∈ Ω : δ∂Ω (x) < ε , the one-sided collar neighborhood of ∂Ω of “width”
ε>0
𝒪(Ω), the collection of all holomorphic functions in an open set Ω ⊆ C
P
P, the P-maximal operator acting on a Lebesgue measurable function u : Ω → Rn
at the point x ∈ ∂Ω according to (cf. [69, §10.1]):
 ∫ 
1
(Pu)(x) := sup |u| dL n
(A.0.173)
0<r <2 diam(∂Ω) σ ∂Ω ∩ B(x, r) Ω∩B(x,r)


ℰ (Ω) ·, · ℰ(Ω) , the pairing between a compactly supported distribution u in Ω and a
smooth function f ∈ 𝒞∞ (supp u), say f ∈ 𝒞∞ (O) with O ⊆ Ω open set containing
supp u, defined for each F ∈ 𝒞∞ (Ω) with the property that F = f near supp u as (cf.
[68, (2.2.33)]):    
ℰ (Ω) u, f ℰ(Ω) := ℰ (Ω) u, F ℰ(Ω) (A.0.174)

∧, the exterior product of differential forms

∨, the interior product of differential forms
.
∂τ j k u, the weak tangential derivative of a given function u ∈ Lloc
1 (Ω, L n ) with the
n
property that ∇u ∈ Lbdd (Ω, L ) , defined as (cf. [68, Example 4.2.4]):
1 n

.
∂τ j k u := ν • (∂k u)e j − (∂j u)ek , (A.0.175)

i.e., the “bullet” product ν • Fjk

u , where
A Terms and notation used in Volume IV 961

Fjk
u
:= (∂k u)e j − (∂j u)ek (A.0.176)

∂∗ E, the measure theoretic boundary of a Lebesgue measurable set E ⊆ Rn , defined

as (cf. [68, (5.2.1)]):
 L n (B(x, r) ∩ E)
∂∗ E := x ∈ Rn : lim sup > 0 and
r→0+ rn
L n (B(x, r) \ E)
lim sup >0 (A.0.177)
r→0+ rn

∂ ∗ E, the reduced boundary of a set E ⊆ Rn of locally finite perimeter, defined as

(cf. [68, (5.6.13)]):
∂ ∗ E consists of all points x ∈ ∂E satisfying the following three
properties:
⨏ 0 < σ∗ (B(x, r)) < +∞ for each r ∈ (0, ∞), formula (A.0.178)
lim+ B(x,r) ν dσ∗ = ν(x) is valid, and |ν(x)| = 1.
r→0

∂nta Ω, the nontangentially accessible boundary of an open proper subset Ω of Rn ,

defined as (cf. [68, Definition 8.8.5]):
2  
∂nta Ω := Aκ (∂Ω) = x ∈ ∂Ω : x ∈ Γκ (x) for each κ > 0 (A.0.179)
κ>0

(a)+ := max{a, 0}, the positive part of the number a ∈ R

πκ (E) = πΩ,κ (E), the “shadow” (or projection) of a given set E ⊆ Ω onto ∂Ω,
defined as (cf. [68, (8.1.15)]):
 
πκ (E) = πΩ,κ (E) := x ∈ ∂Ω : Γκ (x) ∩ E  (A.0.180)

pX , the lower Boyd index of a rearrangement invariant Banach function space X on

a non-atomic sigma-finite measure space (X, M, μ)
∂νA(·, ·), the conormal derivative operator with respect to the coefficient tensor
A acting from weighted Sobolev spaces (cf. [69, Proposition 8.5.3]), and from
Besov/Triebel-Lizorkin spaces (cf. [69, Proposition 9.5.2])
.
∂νAu, the weak conormal derivative of u with respect to the coefficient tensor A
defined (cf. [69, Definition 10.2.18]) by first introducing
αβ
Fα := (A∇u)α = ar s ∂s uβ 1≤r ≤n for each α ∈ {1, . . . , M } (A.0.181)

and then setting

.  
 M
∂νAu := ν • Fα 1≤α ≤M ∈ Lipc (∂Ω) (A.0.182)
962 A Terms and notation used in Volume IV

∂τ j k ϕ with ϕ ∈ 𝒞1 (O), the pointwise tangential derivative operator (cf. [69, §11.1]):
 
∂τ j k ϕ := ν j ∂k ϕ  O∩∂∗ Ω − νk ∂j ϕ  O∩∂∗ Ω at σ∗ -a.e. point on O ∩ ∂∗ Ω (A.0.183)

∂τ j k , the tangential derivative operator defined in a weak, or distributional, sense (cf.

[69, (11.1.17) and Definition 11.1.2], and [69, Definition 11.2.1])
∂ := ∂z̄ := 12 (∂x + i∂y ), the Cauchy-Riemann operator
∂ := ∂z := 12 (∂x − i∂y ), the (complex) conjugate of the Cauchy-Riemann operator
∂νA, the pointwise conormal derivative operator with respect to the coefficient tensor
αβ N
A := ar s 1≤α,β ≤M acting on some u = (uβ )1≤β ≤ N ∈ D (Ω) (which is of func-
1≤r,s ≤n
tion type near ∂Ω) at H n−1 -a.e. point on ∂∗ Ω as (cf. [70, (1.7.9) in Definition 1.7.1]):
 κ−n.t. 
∂νAu := νr ar s ∂s uβ ∂Ω
αβ
(A.0.184)
1≤α ≤M


∂νD,D , the conormal derivative operator associated with the factorization of the
 where D
second-order system L as DD,  and D are homogeneous, constant complex
N
coefficient, first-order systems in R , acting on a given u ∈ D (Ω) (which is of
n

function type near ∂Ω) as:

 κ−n.t.
 ν)(Du)
∂νD,D u := (−i)Sym( D; (A.0.185)
∂Ω


P.V. b k(x − ·)Σ , the principal-value distribution associated with the smooth odd
kernel k and the bounded function b on the countably rectifiable upper Ahlfors regular
set Σ ⊆ Rn , acting on each test function φ ∈ Lipc (Σ) as (cf. [69, Proposition 11.9.1]):
∫
  
P.V. b k(x − ·)Σ , φ := lim+ b(y)k(x − y)φ(y) dσ(y)
ε→0
y ∈Σ
|y−x |>ε

ΠΩ , the Newtonian (volume) potential operator associated with the system L in the
set Ω acting on a given function w as (cf. (1.3.4)):
∫
ΠΩ w := E(· − y)w(y) dy
Ω

where E is the canonical fundamental solution of the system L

∂τ , the tangential partial derivative in the two-dimensional setting, defined as in
(1.6.40):
∂τ := ∂τ12 = ν1 ∂2 − ν2 ∂1
A Terms and notation used in Volume IV 963

∂νλ (
u, π), the (pointwise) conormal derivative for the Stokes system associated with
the coefficient tensor Aλ := δ jk δαβ + λδ jβ δkα 1≤α,β ≤n for each λ ∈ C at σ-a.e.
1≤ j,k ≤n
point on ∂∗ Ω as (cf. (6.1.6)):
κ−n.t. κ−n.t.
∂νλ (
u, π) := ∇
u + λ(∇
u)  ν − π ∂Ω ν
∂Ω

.
∂νλ (
u, π), the weak conormal derivative for the Stokes system associated with λ ∈ C
for the coefficient tensor Aλ := δ jk δαβ + λδ jβ δkα 1≤α,β ≤n as in Definition 6.2.9,
1≤ j,k ≤n
by first introducing the family of vector fields (cf. (6.2.239))

Fα := ∇uα + λ ∂α u − π eα for each α ∈ {1, . . . , n}

and then defining (cf. (6.2.240)):

.  
 n
u, π) := ν • Fα
∂νλ ( 1≤α ≤n ∈ Lipc (∂Ω)

∂τCj k , the complex tangential derivative operator whose action on a complex-valued

function ψ of class 𝒞1 in some open neighborhood of ∂∗ Ω in Cn is defined as (cf.
(7.1.5)):  
∂τCj k ψ := (νC ) j ∂z̄k ψ ∂∗ Ω − (νC )k ∂z̄ j ψ ∂∗ Ω

∂,
¯ the d-bar operator acting on a continuously differentiable complex-valued function
f defined in an open subset of Cn as (cf. (7.1.70)):

n
∂¯ f := (∂z̄ j f ) d z̄ j
j=1

p, ∂¯
∂¯b , the boundary d-bar operator from the space Lα,β b (∂Ω, σ) into the space
p, ∂¯
Ltan,C (∂Ω, σ) ⊗ Λα,β+1 acting on each (α, β)-form f ∈ Lα,β b (∂Ω, σ) as in (7.7.8),
p

(7.7.9), Definition 7.7.2:

∂¯b f := ν 1,0 ∨ ∂¯τ f

∂,
¯ the standard d-bar operator acting on forms as (cf. (7.2.2)):


n
∂¯ := ∂z̄ j d z̄ j ∧ ·
j=1

∂, the complex conjugate of the standard d-bar operator acting on forms as (cf.
(7.2.2)):
n
∂ := ∂z j dz j ∧ ·
j=1
964 A Terms and notation used in Volume IV

∂¯τ , the “partial” tangential derivative introduced in Definition 7.2.2: an (α, β)-form
f ∈ Lloc 1 (∂ Ω, σ) ⊗ Λα,β is said to have ∂¯ f in the space L 1 (∂ Ω, σ) ⊗ Λα,β+2
∗ τ loc ∗
provided there exists some form g ∈ L 1 (∂ Ω, σ) ⊗ Λα,β+2 with the property that
∫ ∫ loc ∗

∂Ω
ν 0,1 ∧ f , ϑψ C dσ = ∂ Ω g, ψ C dσ for all ψ ∈ 𝒞∞ c (C ) ⊗ Λ
n α,β+2
∗ ∗
Pλ , the (double layer) pressure operator for the Stokes system acting on any function
f ∈ L 1 ∂∗ Ω, σ(x)n
n
1+ |x | at each point x ∈ Ω as (cf. (6.2.21)):
∫
Pλ f(x) := −(1 + λ)  − y), f(y) dσ(y)
ν j (y) (∂j q)(x
∂∗ Ω

where the pressure vector q is defined as in (6.2.2) by the formula q(x)

 := − ωn−1
1 x
|x | n
for all points x ∈ R \ {0}
n

Πα,β , the complex volume (Newtonian) potential operator acting on differential

forms U : Ω → Λα,β Cn at each point z ∈ Ω as (cf. (7.4.149)):
∫
 
Πα,β U(z) := − 2
1
U(ζ), Γα,β (ζ, z) dL 2n (ζ)
C
Ω

Q
X/Y , the quotient space of a vector space X and a linear subspace Y of X
qX , the upper Boyd index of a rearrangement invariant Banach function space X on
a non-atomic sigma-finite measure space (X, M, μ)
Q, the (single layer) pressure potential for the Stokes system acting on any function
f ∈ L 1 ∂Ω, 1+σ(x)
n
|x | n−1
at each point x ∈ Ω as in (6.2.14):
∫
Q f(x) :=  − y), f(y) dσ(y)
q(x
∂Ω

where the pressure vector q is defined as in (6.2.2) by the formula q(x)

 := − ωn−1
1 x
|x | n
for each x ∈ R \ {0}
n

R
R+n , the (open) upper half-space in Rn
R−n , the (open) lower half-space in Rn
rad(Ω), the number associated with any nonempty open set Ω ⊆ Rn as (cf. [68,
(5.11.31)]):

rad(Ω) := inf inf sup |x − y|, where

j x ∈Ω j y ∈Ω j
(A.0.186)
{Ω j } j are the connected components of Ω

R j , the j-th boundary-to-boundary Riesz transform, with j ∈ {1, . . . , n}, acting on

f ∈ L 1 Σ, 1+ |σ· | n−1 at σ-a.e. x ∈ Σ as:
A Terms and notation used in Volume IV 965
∫
2 xj − yj
R j f (x) := lim+ f (y) dσ(y) (A.0.187)
ε→0 ωn−1 |x − y| n
|x−y |>ε
y ∈Σ

R j , the j-th boundary-to-domain Riesz transform, with j ∈ {1, . . . , n}, acting on

each f ∈ L 1 Σ, 1+ |σ· | n−1 at every point x ∈ Rn \ Σ as:
∫
2 xj − yj
R j f (x) := f (y) dσ(y) (A.0.188)
ωn−1 |x − y| n
Σ

R jk , the boundary-to-domain integral operator, with j, k ∈ {1, . . . , n}, acting on each

f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at any point x ∈ Ω as:
∫
−1 ν j (y)(xk − yk ) − νk (y)(x j − y j )
R jk f (x) := f (y) dσ∗ (y) (A.0.189)
ωn−1 |x − y| n
∂∗ Ω

R jk , the principal-value singular integral operator, with j, k ∈ {1, . . . , n}, acting on

each f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at σ∗ -a.e. point x ∈ ∂∗ Ω as:
∫  
R jk f (x) := lim+ ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
(A.0.190)
(R jk )max , the maximal operator, with j, k ∈ {1, . . . , n}, acting on each function
f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at any point x ∈ ∂∗ Ω as:
 ∫

(R jk )max f (x) := sup  ν j (y)(∂k EΔ )(y − x)
ε>0
y ∈∂∗ Ω
|x−y |>ε


− νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
(A.0.191)

R#jk , the principal-value singular integral operator on ∂Ω, for j, k ∈ {1, . . . , n}, acting
on any f ∈ L p (∂Ω, σ) ⊗ C n at σ∗ -a.e. point x ∈ ∂∗ Ω as (cf. (1.6.16)):
∫
−1 ν j (x)(x k −yk )−νk (x)(x j −y j )
R#jk f (x) := lim+ |x−y | n f (y) dσ(y)
ε→0 ωn−1
y ∈∂Ω
|x−y |>ε
966 A Terms and notation used in Volume IV

R C , the boundary-to-domain Clifford-Riesz transform associated with a given set

Ω ⊂ Rn , acting on any f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
⊗ C n at each point x ∈ Ω as (cf.
(2.1.77)): ∫
2 x−y
R C f (x) :=  f (y) dσ(y)
ωn−1 ∂Ω |x − y| n
mod
R j , the j-th modified Riesz transform (with j ∈ {1, . . . , n − 1}) in the Euclidean

space Rn−1 , acting on any given function f ∈ L 1 Rn−1, 1+dx |x  | n at L
n−1 -a.e. point

x  ∈ Rn−1 as (cf. (5.3.53)):

∫ 
mod 2 x j −y j 
R j f (x) = lim+ |x  −y  | n 1R n−1 \B n−1 (x ,ε) (y )
ε→0 ωn−1
R n−1
−y j 
− |−y  | n 1R n−1 \B n−1 (0,1) (y ) f (y ) dy 

where, generally speaking, Bn−1 (z , r) is the (n − 1)-dimensional ball centered at

z  ∈ Rn−1 and of radius r
R C, j , the j-th boundary-to-domain complex Riesz transform (with j ∈ {1, . . . , n}),
acting on any function f ∈ L 1 ∂Ω, 1+σ(ζ )
|ζ | 2n−1
at each point z ∈ Ω as (cf. (7.3.56)):
∫
2 z j − ζj
R C, j f (z) := f (ζ) dσ(ζ)
ω2n−1 |z − ζ | 2n
∂Ω

RC, j , the j-th boundary-to-boundary complex Riesz transform (with j ∈ {1, . . . , n}),
acting on any function f ∈ L 1 ∂Ω, 1+σ(ζ )
|ζ | 2n−1
at σ-a.e. point z ∈ ∂Ω as (cf. (7.3.58)):
∫
z j −ζ j
RC, j f (x) := lim+ ω2n−1
2
|z−ζ | 2n
f (ζ) dσ(ζ)
ε→0
|z−ζ |>ε
ζ ∈∂Ω

S
σ := H n−1 ∂Ω, the surface measure on ∂Ω
σ∗ := H n−1 ∂∗ Ω, the surface measure on ∂∗ Ω
σ := H n−1 Σ, the surface measure on the closed Ahlfors regular set Σ ⊆ Rn
 n 
αβ
Sym(D; ·), the principal symbol of the first-order system D = a j ∂j 1≤α ≤ N
j=1 1≤β ≤M
defined at each ξ ∈ Rn as:

n 
αβ
Sym(D; ξ) := i aj ξj 1≤α ≤ N (A.0.192)
j=1 1≤β ≤M
A Terms and notation used in Volume IV 967

∗, the Hodge star operator

S n−1 := ∂B(0, 1), the unit sphere in Rn
S±n−1 := S n−1 ∩ R±n , the upper and lower (open) hemispheres
supp f , the support of the μ-measurable function f defined on a topological space
(X, τ) (where μ is a Borel measure measure on (X, τ)), defined as (cf. [68, (3.8.7) in
Definition 3.8.3]):
 ∫
supp f := x ∈ X : | f | dμ > 0 for each open set O ⊆ X with x ∈ O
O
(A.0.193)
𝒮(Rn ), the space of (smooth, rapidly decreasing) Schwartz functions in Rn
𝒮(Rn ), the space of tempered distributions in Rn
u,θ , the solid maximal function of u : Ω → C, defined at each point x ∈ Ω according
to (cf. [68, (6.6.2)]):

u,θ (x) := u L ∞ (B(x,θ δ∂Ω (x)), L n ) ∈ [0, ∞] (A.0.194)

fp# , the L p -based Fefferman-Stein sharp maximal function of f ∈ Lloc 1 (X, μ), defined

as (cf. [68, (7.4.110)]):

⨏    1/p
fp# (x) := sup  f (y) − fB (x,r)  p dμ(y) , ∀x ∈ X (A.0.195)
ρ
r >0 B ρ (x,r)

ρinv (T; X), the spectral radius of T ∈ Bd (X), with X a quasi-Banach space (cf. [69,
Definition 2.2.5]):

ρinv (T; X) := inf r > 0 : zI − T : X → X homeomorphism

for each point z ∈ C \ B(0, r)

(A.0.196)
𝒮, the boundary-to-domain single layer potential operator for an M × M second-
order weakly elliptic system L (with fundamental solution E) acting on any given
function f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−2
at each point x ∈ Ω as in (1.3.6):
∫
𝒮 f (x) := E(x − y) f (y) dσ(y)
∂Ω
𝒮mod , the boundary-to-domain modified single layer potential operator for an M × M
second-order weakly elliptic system L (with fundamental solution E) acting on any
f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−2
at each point x ∈ Ω as in (1.5.50):
∫
𝒮mod f (x) := {E(x − y) − E(−y) · 1Rn \B(0,1) (−y)} f (y) dσ(y)
∂Ω
968 A Terms and notation used in Volume IV

𝒮Δ,mod , the boundary-to-domain modified harmonic single layer potential operator

acting on any f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
at each point x ∈ Ω as in (1.8.53):
∫
 
𝒮Δ,mod f (x) := EΔ (x − y) − EΔ (−y) · 1Rn \B(0,1) (−y) f (y) dσ(y)
∂Ω

S, the boundary-to-boundary single layer for an M × M second-order weakly elliptic

system L (with canonical fundamental solution E) acting on each given function
f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−2
(when n ≥ 3) at σ-a.e. point x ∈ ∂Ω as in (1.3.62):
∫
S f (x) := E(x − y) f (y) dσ(y)
∂Ω

Smod , the boundary-to-boundary modified single layer potential operator for an M × M

second-order weakly elliptic system L (with fundamental solution E) acting on any
given function f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
at σ-a.e. point x ∈ ∂Ω as in (1.5.73):
∫
Smod f (x) := {E(x − y) − E(−y) · 1Rn \B(0,1) (−y)} f (y) dσ(y)
∂Ω

𝒮, the boundary-to-domain single layer for the Stokes system acting on any given
function f ∈ L 1 ∂Ω, 1+σ(x)
n
|x | n−2
at each point x ∈ Ω as in (6.2.12):
∫
𝒮 f(x) := E(x − y) f(y) dσ(y)
∂Ω

where E is the Kelvin matrix-valued fundamental solution of the Stokes system in

Rn (cf. (6.2.1))
𝒮α,β , the boundary-to-domain complex single layer potential operator acting on a
form f : ∂∗ Ω → Λα,β Cn at each z ∈ Ω as in (7.4.147):
∫
 
𝒮α,β f (z) := f (ζ), Γα,β (ζ, z) dσ(ζ)
C
∂∗ Ω

Sα,β , the boundary-to-boundary complex single layer operator acting on a form

f : ∂∗ Ω → Λα,β Cn at σ-a.e. point z ∈ ∂Ω as in (7.4.148):
∫
 
Sα,β f (z) := f (ζ), Γα,β (ζ, z) dσ(ζ)
C
∂∗ Ω

T
Tγ (x), the family of “bump” (i.e., localized, and normalized in the Hölder norm)
functions centered at x ∈ Σ (cf. [69, §4.1])
A Terms and notation used in Volume IV 969
ap
TrRn →Σ , the trace operator from Rn to Σ defined for each u ∈ W 1, p Rn, δΣ L n as
the limit (cf. [69, Theorem 8.1.1]):
⨏
(TrRn →Σ u)(x) := [u]Rn (x) := lim+ u(y) dy (A.0.197)
r→0 B(x,r)

TrΩ→∂Ω , the trace operator from the Euclidean space Rn into the set ∂Ω defined for
1, p ap
each u ∈ Wa (Ω) := W 1, p Ω, δ∂Ω L n as the limit (cf. [69, Theorem 8.3.6]):
⨏
(TrΩ→∂Ω u)(x) := [u]Ω (x) := lim+ u(y) dy (A.0.198)
r→0 B(x,r)∩Ω

TrΩ→∂Ω , the trace operator from the open set Ω ⊆ Rn to its boundary
 ∂Ω, acting on
each given u ∈ Aα (Ω) (and with w ∈ Aα (Rn ) such that w Ω = u) according to
p,q p,q

(cf. [69, Theorem 9.4.5]):

⨏
TrΩ→∂Ω u (x) := lim+ w dL n at σ-a.e. x ∈ ∂Ω (A.0.199)
r→0 B(x,r)

Tε , the truncated singular integral operator acting on each f ∈ L 1 ∂Ω, 1+σ(x)

|x | n at any
point x ∈ ∂Ω as:
∫
(Tε f )(x) := k(x − y) f (y) dσ(y) (A.0.200)
y ∈∂Ω, |x−y |>ε

Tmax , the maximal operator acting on each f ∈ L 1 ∂Ω, 1+σ(x)|x | n at any point x ∈ ∂Ω
as:  
(Tmax f )(x) := sup (Tε f )(x) (A.0.201)
ε>0

Tmod , the modified principal-value singular integral operator acting on each function
f ∈ L 1 Σ, 1+σ(x)
|x | n at σ-a.e. point in Σ as:
∫
 
Tmod f := lim+ k ε (· − y) − k1 (−y) f (y) dσ(y) (A.0.202)
ε→0
Σ

where k ε := k · 1Rn \B(0,ε) for each ε > 0

Tmod , the boundary-to-domain modified integral operator, acting on any function
f ∈ L 1 ∂Ω, 1+σ(x)
|x | n at each point x ∈ Ω as:
∫
 
(Tmod f )(x) := k(x − y) − k1 (−y) f (y) dσ(y) (A.0.203)
∂Ω

where k1 := k · 1Rn \B(1,0)

970 A Terms and notation used in Volume IV

ϑ := − ∗ ∂∗ and ϑ̄ := − ∗ ∂∗
¯ (cf. (7.2.10))
U
UR set (cf. [68, Definition 5.10.1]): a closed set Σ ⊂ Rn which is (upper) Ahlfors
regular and has Big Pieces of Lipschitz Images (in a uniform, quantitative, scale-
invariant fashion)
UR domain (cf. [68, Definition 5.10.6]): a nonempty open subset Ω of Rn such that
∂Ω is a UR set and
H n−1 (∂Ω \ ∂∗ Ω) = 0 (A.0.204)

UC(X, ρ), the space of uniformly continuous functions on the metric space (X, ρ)
[u]∞
A , the contribution at infinity of a null-solution u for the vector-Helmholtz oper-

ator, associated with the coefficient tensor A := arJsI 1≤J,I ≤M in the writing of the
1≤r,s ≤n
vector Laplacian as Δ = divA∇, defined at each x ∈ Rn as:
∫ 
[u]∞A
:= lim '
ys arI sJ (∂r Φk )(x − y)uJ (y)
R→∞ |y |=R
 
+ asr
IJ
Φk (x − y)(∂r uJ )(y) dH n−1 (x)
1≤I ≤M
(A.0.205)

where Φk is the radiating fundamental solution of the Helmholtz operator Δ + k 2

u ⊗ w, the tensor product of differential forms (cf. (7.3.4))
U " V, the union of two disjoint sets U, V
V
VMO(X, μ), the Sarason space of functions of vanishing mean oscillations on the
measure metric space (X, ρ, μ) defined (with UC(X, ρ) denoting the space of uni-
formly continuous functions on (X, ρ)) as the space (cf. [69, §3.1]):

VMO(X, μ) := the closure of UC(X, ρ) ∩ BMO(X, μ) in BMO(X, μ) (A.0.206)

VMO−1 (∂Ω, σ), the VMO-based negative Sobolev space on ∂Ω, defined for n ≥ 3
as [69, Definition 11.10.9]):

VMO−1 (∂Ω, σ) := the closure of L n−1 (∂Ω, σ) in BMO−1 (∂Ω, σ) (A.0.207)

VMO−1 (∂Ω, σ) in the two-dimensional setting defined as the closure of L q (∂Ω, σ)

in BMO−1 (∂Ω, σ), for any q ∈ (1, ∞) (2.3.134)
∨, the interior product of differential forms (7.1.52)
W
weakly elliptic system: a system L whose characteristic matrix has the property that
det [L(ξ)]  0 for each ξ ∈ Rn \ {0}
A Terms and notation used in Volume IV 971

weakly elliptic coefficient tensor: a coefficient tensor A with the property that the
canonically associated second-order system L A is weakly elliptic, i.e., such that
det [L A(ξ)]  0 for each ξ ∈ Rn \ {0}
W k, p (Ω), the L p -based Sobolev space of order k in Ω (intrinsically defined)
k, p
Wloc (Ω), the local L p -based Sobolev space of order k in Ω
k, p
Wbdd (Ω), the space of Sobolev functions on any bounded measurable subset of Ω
(cf. [68, (3.0.4)]):
k, p k, p
Wbdd (Ω) denote the space of functions u ∈ Wloc (Ω) with the property
that ∂ α u ∈ L p (O, L n ) for each α ∈ N0n with |α| ≤ k and each bounded (A.0.208)
Lebesgue measurable subset O of Ω.

W k, p (Rn, wL n ), the weighted Sobolev space in Rn , defined for p ∈ (0, ∞), k ∈ N0

and a generic weight w in Rn as (cf. [69, §8.1]):

k,1 n
W k, p (Rn, wL n ) := f ∈ Wloc (R ) :  f W k, p (Rn, w L n ) < +∞ (A.0.209)

where
 3∫ 4 1/p
β
 f W k, p (Rn, w L n ) := |∂ f | w dL
p n
(A.0.210)
|β | ≤k Rn

k, p
Wa (Ω), the weighted Sobolev space in Ω defined as in [69, (8.3.5)] for the weight
ap
w := δ∂Ω , with k ∈ N0 , p ∈ (0, ∞), a ∈ R, and equipped with the quasi-norm (cf.
[69, Definition 8.3.4]):
 ∫  1/p
uW k, p (Ω) := |(∂ α u)(x)| p δ∂Ω (x)ap dx (A.0.211)
a
|α | ≤k Ω

1, p
W̊a (Ω), the weighted Sobolev space in the open set Ω, defined for p ∈ (1, ∞) and
a ∈ (−1/p, 1 − 1/p) as (cf. [69, (8.3.65)]):

W̊a (Ω) := the closure of 𝒞∞

1, p 1, p
c (Ω) in W a (Ω) (A.0.212)

−1, p
Wa (Ω), the negative weighted Sobolev space in Ω, defined for p ∈ (1, ∞) and
a ∈ (−1/p, 1 − 1/p) as (cf. [69, (8.5.1)-(8.5.2)]):
 
n
−1, p
∂j f j ∈ D (Ω) : f j ∈ L p Ω, δ∂Ω L n , 0 ≤ j ≤ n
ap
Wa (Ω) := f = f0 +
j=1
(A.0.213)
equipped with the norm
972 A Terms and notation used in Volume IV


n+1 
n
−1, p
Wa (Ω)  f → inf  f j  L p (Ω,δ a p L n ) : f = f0 + ∂j f j , (A.0.214)
∂Ω
j=0 j=1

ap
f j ∈ L p Ω, δ∂Ω L n , 0 ≤ j ≤ n

k, p
Wa, (Ω), the weighted maximal Sobolev space in Ω, defined for p ∈ (0, ∞), k ∈ N0 ,
and a ∈ R as (cf. [69, Definition 8.6.1]):

(Ω) : ∂ α u ∈ L Ω, δ∂Ω L n for all α ∈ N0n with |α| ≤ k
k, p k,1 p ap
Wa, (Ω) := u ∈ Wloc
(A.0.215)
equipped with the quasi-norm

uW k, p (Ω) := ∂ α u Lp (Ω,δ a p L n )
a, ∂Ω
|α | ≤k

 ∫   1
≈ (∂ α u),θ  p δ ap dL n p for θ ∈ (0, 1) (A.0.216)
∂Ω
|α | ≤k Ω

1, p
𝒲a (Ω; D), the weighted Sobolev-Hardy space of null-solutions of the Dirac opera-
n
tor D = e j  ∂j in an open set Ω ⊆ Rn defined for p ∈ (0, ∞) and a ∈ − p1 , 1 − p1
j=1
as in (4.4.131):
1, p  1, p 
𝒲a (Ω; D) := u ∈ Wa (Ω) ⊗ C n : Du = 0 in Ω

X
∗
X ∗ ·, · X , the duality pairing between a vector space X and its algebraic dual X
X, the associated space (aka Köthe dual) of the Generalized Banach Function Space
X, equipped with the norm  · X (cf. [69, Definition 5.1.11])
      
 x  :=  ·  := inf λ > 0 : λ−1 x ∈ BX (0, 1) p for all x ∈ X, the Minkowski
p X, p
functional associated with the absolutely p-convex hull of the unit ball in X (cf. [69,
(7.8.6)])
𝒳bdd (Ω), or 𝒳(Ω)bdd , the space of distributions in the open set Ω ⊆ Rn defined as
(cf. [69, Convention 8.3.7]):
 
u ∈ D (Ω) : ψ Ω u ∈ 𝒳(Ω) for each ψ ∈ 𝒞∞ c (R )
n
(A.0.217)
References

1. M.B. Balk, Polyanalytic functions and their generalizations, pp. 195–261 in Encyclopaedia
of Mathematical Sciences, Vol. 85, Springer-Verlag, Berlin, Heidelberg, 1997.
2. J. Bergh and J. Löfström, Interpolation Spaces. An Introduction, Springer-Verlag, Berlin/New
york, 1976.
3. A.V. Bitsadze, Ob edinstvennosti resheniya zadachi Dirichlet dlya ellipticheskikh uravnenii s
chastnymi proizvodnymi, Uspekhi Matem. Nauk, 3 (1948) no. 6, 211–212.
4. A.V. Bitsadze, Boundary Value Problems for Second-Order Elliptic Equations, North-
Holland, Amsterdam, 1968.
5. H.P. Boas, A geometric characterization of the ball and the Bochner-Martinelli kernel, Math.
Ann., 248 (1980), no. 3, 275–278.
6. B. Bojarski, Sharp maximal operator of fractional order and Sobolev embedding inequalities,
Bull. Polish Acad. Sci. Math., 33 (1985), 7–16.
7. A.P. Calderón, On the behavior of harmonic functions near the boundary, Trans. Amer. Math.
Soc., 68 (1950), 47–54.
8. A.P. Calderón, Estimates for singular integral operators in terms of maximal functions, Studia
Math., 44 (1972), 167–186.
9. A.P. Calderón, Commutators, singular integrals on Lipschitz curves and applications, pp. 85–
96 in “Proceedings of the International Congress of Mathematicians” (Helsinki, 1978), Acad.
Sci. Fennica, Helsinki, 1980.
10. A.P. Calderón and R. Scott, Sobolev type inequalities for p > 0, Studia Math., 62 (1978),
75–92.
11. A.P. Calderón and A. Torchinsky, Parabolic maximal functions associated with a distribution.
II, Adv. in Math., 24 (1977), 101–171.
12. T. Carleman, Über das Neumann-Poincarésche Problem für ein Gebiet mit Ecken, Almqvist
& Wiksell, 1916.
13. S.-C. Chen and M.-C. Shaw, Partial Differential Equations in Several Complex Variables,
AMS/IP Studies in Advanced Mathematics, Vol. 19, American Mathematical Society, Provi-
dence, RI; International Press, Boston, MA, 2001.
14. M. Christ, The extension problem for certain function spaces involving fractional orders of
differentiability, Arkiv för Matematik, 22 (1984), no. 1-2, 63–81.
15. R. Coifman, A. McIntosh, and Y. Meyer, L’intégrale de Cauchy definit un opérateur borné
sur L 2 pour les courbes lipschitziennes, Ann. Math., 116 (1982), 361–388.
16. R.R. Coifman and G. Weiss, Extensions of Hardy spaces and their use in analysis, Bull. Amer.
Math. Soc., 83 (1977), no. 4, 569–645.

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 973
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1
974 References

17. M. Costabel, Some historical remarks on the positivity of boundary integral operators, pp. 1–
27 in “Boundary Element Analysis”, M. Schanz and O. Steinbach eds., Lecture Notes in
Applied and Computational Mechanics, Vol. 29, Springer, Berlin, Heidelberg, 2007.
18. B.E.J. Dahlberg, Estimates of harmonic measure, Arch. Rational Mech. Anal., 65 (1977),
no. 3, 275–288.
19. G. David and S. Semmes, Singular Integrals and Rectifiable Sets in R n : Beyond Lipschitz
Graphs, Astérisque, No. 193, 1991.
20. G. David and S. Semmes, Analysis of and on Uniformly Rectifiable Sets, Mathematical Surveys
and Monographs, AMS Series, 1993.
21. D. Deng and Y.S. Han, T 1 theorem for Besov and Triebel-Lizorkin spaces, Sci. China Ser. A,
48 (2005), 657–665.
22. D. Deng and Y. Han, Harmonic Analysis on Spaces of Homogeneous Type, Lecture Notes in
Mathematics, Vol. 1966, Springer, 2009.
23. R.A. DeVore and R.C. Sharpley, Maximal Functions Measuring Smoothness, Memoirs of the
Amer. Math. Soc., Vol. 47, No. 293, AMS, Providence, 1984.
24. E.B. Fabes, M. Jodeit Jr., and N.M. Rivière, Potential techniques for boundary value problems
on C 1 -domains, Acta Math., 141 (1978), no. 3–4, 165–186.
25. G. Fichera, Caratterizazione della traccia, sulla frontiera di un campo, di una funzione
analitica di più variabili complesse, Rend. Acc. Naz. Lincei VII, 23 (1957), 706–715.
26. D. Freitag Real interpolation of weighted L p -spaces, Mathematische Nachrichten, 86 (1978),
no. 1, 15–18.
27. J. García-Cuerva and J.L. Rubio de Francia, Weighted Norm Inequalities, North Holland,
Mathematics Studies Vol. 116, 1985.
28. J.B. Garnett, Bounded Analytic Functions, Revised first edition, Graduate Texts in Mathemat-
ics, 236, Springer, New York, 2007.
29. C.F. Gauss, Allgemeine Theorie des Erdmagnetismus, (1838). Werke 5 (1867), 127–193.
30. C.F. Gauss, Atlas des Erdmagnetismus, 1840. Werke 12 (1929), 326–408.
31. N.M. Günter, Potential Theory and its Applications to Basic Problems of Mathematical
Physics, F. Ungar, 1967.
32. Yanchang Han, T 1 Theorems for inhomogeneous Besov and Triebel-Lizorkin spaces over
space of homogeneous type, Vietnam Journal of Mathematics, 36 (2008), no. 2, 125–136.
33. Y. Han, S. Lu, and D. Yang, Inhomogeneous Besov and Triebel-Lizorkin spaces on spaces of
homogeneous type, Approx. Theory Appl. (N.S.), 15 (1999), no. 3, 37–65.
34. Fr. Hartogs, Einige Folgerungen aus der Cauchyschen Integralformel bei Funktionen mehrerer
Verä-nderlichen, Sitzungsberichte der K.B. Akademie der Wissenschaften zu München,
Mathematisch-Physikalische Klasse, 36 (1906), 223–242.
35. G.M. Henkin and E.M. Chirka, Boundary properties of holomorphic functions of several
complex variables, pp. 12-142 in “Current Problems in Mathematics”, Vol. 4, Akad. Nauk
SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Informatsii, Moscow, 1975 (English translation in J.
Soviet. Math., 5 (1976), no. 5).
36. G.M. Henkin and J. Leiterer, Theory of Functions on Complex Manifolds, Monographs in
Mathematics, Birhäuser, 1984.
37. M. Hervé, Analytic and Plurisubharmonic Functions in Finite and Infinite Dimensional S-
paces, Springer Lecture Notes, Vol. 198, Springer-Verlag, Berlin, 1971.
38. S. Hofmann, E. Marmolejo-Olea, M. Mitrea, S. Perez-Esteva, and M. Taylor, Hardy spaces,
singular integrals and the geometry of Euclidean domains of locally finite perimeter, Geo-
metric and Functional Analysis (GAFA), 19 (2009), no. 3, 842–882.
39. S. Hofmann, M. Mitrea, and M. Taylor, Geometric and transformational properties of Lips-
chitz domains, Semmes-Kenig-Toro domains, and other classes of finite perimeter domains,
J. Geom. Anal., 17 (2007), no. 4, 593–647.
40. G. Hoepfner, P. Liboni, D. Mitrea, I. Mitrea, and M. Mitrea, Multi-layer potentials for higher
order systems in rough domains, to appear in Analysis and PDE, (2021).
41. G.C. Hsiao and W.L. Wendland, Boundary Integral Equations, Applied Mathematical Sci-
ences, Vol. 164, Springer, 2008.
References 975

42. D.S. Jerison and C.E. Kenig, Boundary behavior of harmonic functions in nontangentially
accessible domains, Advances in Mathematics, 46 (1982), no. 1, 80–147.
43. O.D. Kellogg, Foundations of Potential Theory, Dover Publications, Inc., 1953 (republication
of the work printed in 1929 by J. Springer, Berlin).
44. C.E. Kenig, Harmonic Analysis Techniques for Second Order Elliptic Boundary Value Prob-
lems, CBMS Regional Conference Series in Mathematics, Vol. 83, AMS, Providence, RI,
1994.
45. H. Kneser, Die Randwerte einer analytischen Funktion zweier Veränderlichen, Monatsh. für
Math. u. Phys., 43 (1936), 364–380.
46. A. Korn, Lehrbuch der Potentialtheorie. Allgemeine Theorie des Potentials und der Poten-
tialfunctionen im Raume, Ferd. Dümmler, Berlin, 1899.
47. A. Korn, Sur la méthode de Neumann et le problème de Dirichlet, C.R., 130 (1900), 557.
48. A. Korn, Lehrbuch der Potentialtheorie. II. Allgemeine Theorie des logarithmischen Potentials
und der Potentialfunctionen in der Ebene, Ferd. Dümmler, Berlin 1901.
49. A. Korn, Sur les équations de l’élasticité, Ann. Sci. Éc. Norm. Sup., 24 (1907), no. 3, 9–75.
50. A. Korn, Sur certaines questions qui se rattachent au problème des efforts dans la théorie de
lâĂŹélasticité Ann. Fac. Sci.Toulouse, 2 (1910), no. 3, 7–18.
51. S.G. Krantz, Functions of Several Complex Variables, 2nd edition, AMS Chelsea Publishing,
American Mathematical Society, Providence, Rhode Island, 2001.
52. A.M. Kytmanov, The Bochner-Martinelli Integral and Its Applications, Birkhäuser Verlag,
Bassel, 1995.
53. A.M. Kytmanov and S.G. Myslivets, Multidimensional Integral Representations. Problems of
Analytic Continuation, Springer, 2015.
54. O.A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, Gordon and
Breach Science Publishers, New York-London, 1963.
55. C. Laurent-Thiébaut, Phénomène de Hartogs-Bochner dans les variétés CR, pp. 233–247 in
“Topics in Complex Analysis”, Banach Center Publications, Vol. 31, Warszawa, 1995.
56. P.G. Lemarié, Continuité sur les espaces de Besov des opérateurs définis par des intégrales
singulièrs, Ann. Inst. Fourier, 35 (1985), 175–187.
57. H. Lewy, On the local character of the solutions of an atypical linear differential equation in
three variables and a related theorem for regular functions of two complex variables, Ann.
Math., 64 (1956), 514–522.
58. H. Lewy, An example of a smooth linear partial differential equation without solutions, Ann.
Math., 66 (1957), 155–158.
59. I. Lieb and J. Michel, The Cauchy-Riemann Complex. Integral Formulae and Neumann Prob-
lem, Aspects of Mathematics, E34, Friedr. Vieweg & Sohn, Braunschweig, 2002.
60. E. Marmolejo-Olea and M. Mitrea, Harmonic analysis for general first order differential op-
erators in Lipschitz domains, pp. 91–114 in “Clifford Algebras: Application to Mathematics,
Physics, and Engineering”, Birkhäuser Progress in Mathematical Physics Series, 2003.
61. J.M. Martell, D. Mitrea, I. Mitrea, and M. Mitrea, The Dirichlet problem for elliptic systems
with data in Köthe function spaces, Revista Matemática Iberoamericana, 32 (2016), no. 3,
913–970.
62. E. Martinelli, Sulla determinazione di una funzione analitica di più variabili complesse in un
campo, assegnatane la traccia sulla frontiera, Ann. Mat. Pura e Appl., 55 (1961), 191–202.
63. M.Ya. Mazalov, On the existence of angular boundary values for polyharmonic functions in
the unit ball, Journal of Mathematical Sciences, 234 (2018), no. 3, 362–368.
64. J. Merker and E. Porten, A Morse theoretical proof of the Hartogs extension theorem, J. Geom.
Anal., 17 (2007), no. 3, 513–546.
65. Y. Meyer, Ondelettes et Opérateurs II. Opérateurs de Calderón-Zygmund, Hermann, Paris,
1990.
66. D. Mitrea, Distributions, Partial Differential Equations, and Harmonic Analysis, Second
edition, Springer Nature Switzerland, 2018.
67. D. Mitrea, I. Mitrea, and M. Mitrea, A sharp divergence theorem with nontangential traces,
Notices of the American Mathematical Society, 67 (2020), no. 9, 1295–1305.
976 References

68. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis I: A Sharp Divergence The-
orem with Nontangential Pointwise Traces, Developments in Mathematics Vol. 72, Springer
Nature, Switzerland, 2022.
69. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis II: Function Spaces Mea-
suring Size and Smoothness on Rough Sets, Developments in Mathematics Vol. 73, Springer
Nature, Switzerland, 2022.
70. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis III: Integral Representa-
tions, Calderón-Zygmund Theory, Fatou Theorems, and Applications to Scattering, Develop-
ments in Mathematics Vol. 74, Springer Nature, Switzerland, 2022.
71. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis IV: Boundary Layer Poten-
tials in Uniformly Rectifiable Domains, and Applications to Complex Analysis, Developments
in Mathematics Vol. 75, Springer Nature, Switzerland, 2022.
72. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis V: Fredholm Theory and
Finer Estimates for Integral Operators, with Applications to Boundary Problems, Develop-
ments in Mathematics Vol. 76, Springer Nature, Switzerland, 2022.
73. D. Mitrea, I. Mitrea, M. Mitrea, and S. Monniaux, Groupoid Metrization Theory with Appli-
cations to Analysis on Quasi-Metric Spaces and Functional Analysis, Birkhäuser, 2013.
74. D. Mitrea, I. Mitrea, M. Mitrea, and B. Schmutzler, Calderón–Zygmund theory for second-
order elliptic systems on Riemannian manifolds, pp. 413–426 in “Integral Methods in Science
and Engineering,” C. Constanda and A. Kirsch editors, Birkhäuser, 2015.
75. D. Mitrea, I. Mitrea, M. Mitrea, and B. Schmutzler, Layer Potential Techniques for Boundary
Value Problems on Rough Subdomains of Riemannian Manifolds monograph in preparation,
2017.
76. D. Mitrea, I. Mitrea, M. Mitrea, and M. Taylor, The Hodge-Laplacian: Boundary Value Prob-
lems on Riemannian Manifolds, Studies in Mathematics, Vol. 64, De Gruyter, 2016.
77. D. Mitrea, M. Mitrea, and J. Pipher, Vector potential theory on Lipschitz domains in R3 and
applications to electromagnetic scattering, Journal of Fourier Analysis and Applications, 3
(1997), no. 2, 131–192.
78. I. Mitrea and M. Mitrea, Multi-Layer Potentials and Boundary Problems for Higher-Order
Elliptic Systems in Lipschitz Domains, Lecture Notes in Mathematics, Vol. 2063, Springer,
Berlin, 2013.
79. M. Mitrea and M. Muether, An integration by parts formula in submanifolds of positive
codimension, Math. Methods in Applied Sciences, 27 (2004), no. 14, 1711–1723.
80. A. Miyachi, Atomic decompositions for Sobolev spaces and for the C pα spaces on general
domains, Tsukura J. Math., 21 (1997), no. 1, 59–96.
81. E. Nakai, Hardy-Littlewood maximal operator, singular integral operators, and Riesz poten-
tials on generalized Morrey spaces, Math. Nachr., 166 (1994), 95–103.
82. C. Neumann, Untersuchungen über das Logarithmische und Newtonsche Potential, B.G. Teub-
ner, Leipzig, 1877.
83. J. Peetre, On convolution operators leaving L p, λ spaces invariant, Annali di Matematica
Pura ed Applicata, 72 (1966), 295–304.
84. H. Poincaré, La méthode de Neumann et le problème de Dirichlet, Acta Mathematica, 20
(1897), 59–142.
85. I. Privalov, Randeigenschaften analytischer Funktionen, Deutscher Verlag der Wis-
senschaften, Berlin, 1956.
86. J. Radon, Über lineare Funktionaltransformationen und Funktionalgleichungen, Sitzungs-
berichte Akad. Wiss., Abt. 2a, Wien, 128 (1919), 1083–1121.
87. J. Radon, Über die Randwertaufgaben beim logarithmischen Potential, Sitzungsberichte
Akad. Wiss., Abt. 2a, Wien, 128 (1919), 1123–1167.
88. R.M. Range, Holomorphic Functions and Integral Representations in Several Complex Vari-
ables, Graduate Texts in Mathematics, Vol. 108, Springer-Verlag, New York, 1986.
89. R.M. Range, Extension phenomena in multidimensional complex analysis: correction of the
historical record, The Mathematical Intelligencer, 24 (2002), no. 2, 4–12.
90. R.M. Range, Some landmarks in the history of the tangential Cauchy Riemann equations,
Rend. Mat. Appl. Serie VII, 30 (2010), no. 3-4, 275–283.
References 977

91. F. Riesz and B. Sz.-Nagy, Functional Analysis, Ungar Publishing Co., New York, 1955, French
original: Akad. Kiado, Budapest, 1952.
92. W. Rudin, Real and Complex Analysis, 3-rd edition, McGraw-Hill, Boston, Massachusetts,
1987.
93. F. Severi, Risoluzione generale del problema di Dirichlet per le funzioni biarmoniche, Rend.
Reale Accad. Lincei, 23 (1931), 795–804.
94. R. Sharpley and Y.-S. Shim, Singular integrals on C pα , Studia Math., XCIL (1986), 285–293.
95. S.L. Sobolev, Partial Differential Equations of Mathematical Physics, Pergamon Press, Ox-
ford, 1964.
96. E.M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton Math-
ematical Series, No. 30, Princeton University Press, Princeton, NJ, 1970.
97. W. Stekloff, Les méthodes générales pour résoudre les problèmes fondamentaux de la physique
mathématique, Ann. Fac. Sci. Toulouse, 2 (1900), no. 2, 207–272.
98. W. Stekloff, Mémoire sur les fonctions harmoniques de M. H. Poincaré, Ann. Fac. Sci.
Toulouse, 2 (1900), no. 2, 273–303.
99. W. Stekloff, Sur la méthode de Neumann et le problème de Dirichlet, C. R., 130 (1900),
396–399.
100. W. Stekloff, Remarque à une note de M. A. Korn: “Sur la méthode de Neumann et le problème
de Dirichlet.”, C. R., 130 (1900), 826–827.
101. W. Stekloff, Sur les problèmes fondamentaux de la physique mathématique, Ann. de l’Éc.
Norm., 19 (1902), no. 3, 191–259.
102. J.-O. Strömberg, Bounded mean oscillations with Orlicz norms and duality of Hardy spaces,
Indiana Univ. Math. J., 28 (1979), no. 3, 511–544.
103. O. Tapiola and X. Tolsa, Connectivity conditions and boundary Poincaré inequalities,
preprint, (2022).
104. B.M. Weinstock, Continuous boundary values of analytic functions of several complex vari-
ables, Proc. Amer. Math. Soc., 21 (1969), no. 2, 463–466.
105. B.M. Weinstock, An approximation theorem for ∂-closed
¯ forms of type (n, n − 1), Proc. Amer.
Math. Soc., 26 (1970), no. 4, 625–628.
106. S. Zaremba, Sur le problème de Dirichlet, Ann. Sci. Éc. Norm. Sup., 14 (1897), no. 3,
251–258.
107. S. Zaremba, Sur l’équation aux dérivées partielles Δu + ξu + f = 0 et sur les fonctions
harmoniques, Ann. Sci. Éc. Norm. Sup., 16 (1899), no. 3, 427–464.
108. S. Zaremba, Contribution à la théorie des fonctions fondamentales, Ann. Sci. Éc. Norm. Sup.,
20 (1903), no. 3, 9–26.
109. A. Zygmund, Trigonometric Series, Cambridge University Press, 1968.
Subject Index

(ε, δ)-domain, 936 modified double layer for the

P-maximal function, 960 Stokes system, 718
CR-form, 809 single layer (for a generic system),
CR-function, 787 37
single layer for the Stokes system,
Ahlfors regular
706
domain, 923
Analytic function transpose double layer (for a
separately (several variables), 788 generic system), 39
transpose double layer for the
Besov space Stokes system, 703
inhomogeneous, 928 Boundary-to-domain
Bitsadze’s operator, 59 Bochner-Martinelli integral
Bochner-Martinelli operator on (α, β)-forms, 817
integral operator (on functions), 78 Clifford-Riesz transform, 250
integral operator on (0, β)-forms,
double layer (for a generic
816
system), 27
integral operator on (α, β)-forms,
double layer for the Stokes system,
817
686
Boundary
measure theoretic, 961 modified double layer for the
Boundary d-bar operator, 880 Stokes system, 718
Boundary-to-boundary modified single layer for the
Bochner-Martinelli integral Stokes system, 717
operator on (α, β)-forms, 824 single layer (for a generic system),
double layer (for a generic 24
system), 38 single layer for the Stokes system,
double layer for the Stokes system, 685
702 Bullet product, 959

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 979
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1
980 Subject Index

Calderón space, 379 Ahlfors regular, 923

Carleson measure Double form, 812
super vanishing, 173 Double layer potential operator
vanishing, 173 boundary-to-boundary, 38
Cauchy integral operator boundary-to-boundary, for the
boundary-to-boundary modified, Stokes system, 702
230 boundary-to-domain, 27, 686
Cauchy integral operator, boundary-to-domain, for the
boundary-to-domain modified, Stokes system, 686
220 modified boundary-to-domain,
Cauchy-Clifford integral operator 718
boundary-to-boundary modified,
230 Exterior derivative operator, 934
boundary-to-domain modified,
222 Fourier transform, 939
transpose, 138 Fredholm radius, 938
Cayley algebra, 69
Hardy space
Chord-dot-normal singular integral
Lebesgue-based, 941
operator, 623
Lorentz-based, 941
Clifford-Riesz transform, 250 Hardy-based
Coefficient tensor inhomogeneous Sobolev space,
compliant, 154 943
Legendre-Hadamard elliptic, 153 Hermitian inner product
positive semi-definite, 154 for double forms, 813
Complex double layer operator, 853 on differential forms, 791
Complex Laplacian (in Cn ), 814 Hilbert transform, 669
Complex normal (function), 794 modified, 670
Complex outward unit normal, 793 operator, 917
Complex Riesz transform, 821 Hodge star operator, 791, 967
Complex tangential (function), 794 Holomorphic
Complex tangential derivative function (of several variables), 788
operator, 784 Hardy space, 874, 944
Conormal derivative for the Stokes separately, 788
system in Holomorphic functions
Besov/Triebel-Lizorkin spaces, space, 910
768 Homogeneous
Triebel-Lizorkin space, 938
D-bar operator, 794 Homogeneous Hardy spaces with
Differential form regularity, 231
double, 812 Homogeneous Sobolev space
of degree (α, β), 790 Morrey-based, 957
Distribution Homogeneous, with regularity
on an arbitrary set, 948 Hardy spaces, 231
Domain
(ε, δ), 936 Inhomogeneous Besov space, 928
Subject Index 981

Inhomogeneous Sobolev space boundary-to-boundary, for the

Hardy-based, 943 Stokes system, 718
Integration by parts on the boundary Morrey-based
involving ∂¯τ , 802 homogeneous Sobolev space, 957
with complex tangential negative Sobolev space, 957
derivatives, 785 Morrey-Campanato space
homogeneous, 950
Lamé system, 953 inhomogeneous, 950
Layer potential operator Muckenhoupt weighted
boundary-to-boundary negative Sobolev space, 952
double (for a generic system), 38
double, for the Stokes system, Newtonian (volume) potential
702 operator, 24
modified double, 177 Nontangential
modified double, for the Stokes maximal operator, 958
system, 718 Nontangential maximal function
modified single, 97 truncated, 958
modified single (for the Stokes Norm
system), 717 Luxemburg, on an Orlicz space,
single (for a generic system), 37 949
boundary-to-domain
double (for a generic system), 27 Octonions, 69
double (for the Stokes system), Off-diagonal Sobolev space
686 block-based, 928
modified double, 174 Operator
modified harmonic double, 182 Bitsadze’s, 59
modified harmonic single, 183 Orlicz space, 949
modified single, 94
modified single (for the Stokes Pressure operator for the Stokes
system), 717 system
single (for a generic system), 24 double layer, 688
single (for the Stokes system), single layer, 686
685
modified boundary-to-domain Riesz transform
double (for the Stokes system), complex, 821
718 modified operator, 664
transpose double, 39
transpose double (for the Stokes Sarason space VMO, 970
system), 703 Separately monogenic, 79
Local (boundary) Sobolev space, 951 Single layer potential operator
Local John condition, 947 boundary-to-boundary, 37
Luxemburg norm on an Orlicz space, boundary-to-domain, 24, 685
949 modified boundary-to-domain,
717
Modified double layer potential Singular integral operator
operator chord-dot-normal, 623
982 Subject Index

Sobolev space Sarason, VMO, 970

L p -based homogeneous, 951 weighted maximal Sobolev, 972
block-based, 928 weighted Sobolev, 951
block-based homogeneous, 929 Space of holomorphic functions in
complex (boundary), 786 an open set, 910
homogeneous, 951 Standard fundamental solution for
vanishing Morrey-based, 957 the Laplacian, 937
local, 951 Super vanishing Carleson measure,
local off-diagonal (boundary), 951 173
Lorentz-based, 952 Support of a measurable function,
Morrey-based homogeneous, 957 967
negative
definition, 952 Tangential derivative
Morrey-based, 957 real, 786
Muckenhoupt weighted, 952 weak, 960
off-diagonal (boundary), 950 Tangential derivative operator
vanishing Morrey-based, 957 complex, 784
weighted, 951 in a pointwise sense, 962
Space Transpose double layer for the Stokes
L p -based Sobolev, 950 system, 703
L p -based homogeneous Sobolev, Transpose double layer potential
951 operator, 39
Hardy (Lebesgue-based), 941 Triebel-Lizorkin space
Hardy (Lorentz-based), 941 homogeneous, 938
homogeneous Morrey-Campanato,
950 Vanishing Carleson measure, 173
homogeneous Triebel-Lizorkin, Vanishing mean oscillations, 970
938
inhomogeneous Besov , 928 Weak conormal derivative
inhomogeneous for the Stokes system, 732
Morrey-Campanato, 950 Weak tangential derivative, 960
local off-diagonal (boundary) Weighted
Sobolev, 951 maximal Sobolev space, 972
negative Sobolev, 952 Sobolev space, 951
of vanishing mean oscillations, Sobolev space in Ω, for the weight
ap
970 w := δ∂Ω , 971
Orlicz, 949 Sobolev spaces in Rn , 971
Symbol Index

a ⊗ b tensor product of vectors a, b, ·, · Lipc (Σ) (or simply ·, · )

(Lip c (Σ))
924 distributional pairing, 948
u ⊗ w tensor product of differential ·, · C pointwise Hermitian inner
forms, 970 product for double forms, 944
∗ Hodge star operator in Cn , 944 D  (Ω) ·, · D(Ω) distributional
∗ Hodge star operator, 967 Hermitian inner product for
∧ exterior product of differential double forms, 936
forms, 960 ℰ (Ω) ·, · ℰ(Ω) (compact support)
∨ interior product, 970 distributional Hermitian inner
∨ interior product of differential product for double forms, 937
forms, 960 ·, · C Hermitian inner product on
Δ + k 2 Helmholtz operator, 935 differential forms, 944
u · w = u, w dot product of two
Δ Laplace operator, 933
vectors u, w ∈ Cn , 944
∇u gradient (Jacobian matrix) of u,
u · w = u, w dot product of two
940
vectors u, w ∈ Rn , 933
∇  gradient operator in Rn−1 , 940
[Mb, T] commutator of Mb with T,
∇tan tangential gradient, 940 958
∇tan
A , 940  
Aζ, η bilinear form associated with
A∇u, 924 the coefficient tensor A, 924
Δ = Δ(x, r) surface ball, 934  ∞ contribution of F at infinity,
[F]
UV symmetric difference of U and 938
V, 934 [u]∞ A contribution at infinity of

U " V the union of two disjoint sets null-solution for the

U, V, 970 vector-Helmholtz operator, 970
D (Ω) ·, · D(Ω) distributional pairing
  · Bd(X→Y) operator norm, 959
in Ω, 933  · Bd(X) operator norm, 959
X ∗ ·, · X , 972  · X→Y operator norm, 959

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 983
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis IV, Developments in Mathematics 75,
https://doi.org/10.1007/978-3-031-29179-1
984 Symbol Index
ess
 ·  X→Y essential norm, 959 Φ− (X → Y ) finite-dim cokernel
 ·  Minkowski functional of semi-Fredholm operators from
p 
BX (0, 1) p , 972 X into Y , 937
 · (X0,X1 )θ, q real interpolation Γκ (x) nontangential approach region,
quasi-norm, 945, 960 940
(a)+ := max{a, 0}, 961 Γα,β (ζ, z), 940
X/Y quotient space, 964 Λα,β Cn differential forms of (type)
[x]X/Y equivalence class of x ∈ X in degree (α, β), 954
X/Y , 937 νg GMT unit normal induced by the
1E characteristic function of E, 960 metric tensor g, 958
⨏fΔ integral average of f in Δ, 939 ν E GMT unit normal induced by the
E
f dμ integral average of f on E, standard Euclidean metric, 958
944 
ν • F the bullet product of ν with F,
Ů interior of the set U, 944 959
U closure of the set U, 930
ν • u the Clifford bullet product of ν
V ⊥ annihilator of a subspace V of a
with u, 959
Banach space X, 923
⊥ W annihilator of a subspace W of νC complex outward unit normal, 960
∗ ωn−1 surface area of S n−1 , 960
√ X , where X is Banach, 923
i = −1 ∈ C complex imaginary ρinv (T; X) spectral radius of
unit, 944 T ∈ Bd (X), 967
[A; B] := [A, B] := AB − BA the ρFred (T; X) Fredholm (or essential
commutator of A and B, 926 spectral) radius of T ∈ Bd (X),
{ A; B} := AB + BA the 938
anti-commutator of A and B, σ∗ = H n−1 ∂∗ Ω surface measure,
926 966
dζ complex arc-length, 935 σ = H n−1 ∂Ω surface measure on
d exterior derivative operator, 934 ∂Ω, 966
δ formal adjoint of the exterior σ := H n−1 Σ surface measure on
derivative operator d, 934 the closed Ahlfors regular set
δ jk Kronecker symbol, 934 Σ ⊆ Rn , 966
δx Dirac distribution with mass at x,
∂nta Ω nontangentially accessible
934
boundary of Ω, 961
δ∂Ω (·) distance function to the
boundary, 934 ∂∗ E measure theoretic boundary of
φ' = ℱφ Fourier transform of φ, 939 E, 961
∗
∂ E reduced boundary of E, 961
εBA generalized Kronecker symbol,
937 ∂νA conormal derivative operator
Φk (·) radiating fundamental solution with respect to the coefficient
of the Helmholtz operator, 939 tensor A acting from Besov and
Φ(X → Y ) Fredholm operators from Triebel-Lizorkin spaces, 961
X into Y , 937 ∂νA conormal derivative operator
Φ+ (X → Y ) finite-dim kernel with respect to the coefficient
semi-Fredholm operators from tensor A acting from weighted
X into Y , 937 Sobolev spaces, 961
Symbol Index 985
.A
∂ν weak conormal derivative Aκ (∂Ω) accessibility set, 923
operator with respect to the Ap (X, ρ, μ) Muckenhoupt class, 923
coefficient tensor A, 961 [w] A p characteristic of the
∂νA pointwise conormal derivative Muckenhoupt weight w, 923
operator with respect to the A∞ (X, ρ, μ) Muckenhoupt class, 923
coefficient tensor A, 962 A p absolutely p-convex hull of the

D,D
∂ν pointwise conormal derivative set A, 924
p,q
associated with the As (Ω) Besov/Triebel-Lizorkin
 962
factorization L = DD, space in Ω, 925
λ
∂ν ( u, π) conormal derivative for the  ·  Ap, q quasi-norm in
s (Ω)
Besov/Triebel-Lizorkin space
. Stokes system, 963
∂νλ (u, π) weak conormal derivative in Ω, 925
for the Stokes system, 963 A q,κ L q -based area-function, 924
∂τ tangential partial derivative in the AWE (n, M) weakly elliptic coefficient
two-dimensional setting, 962 tensors, 925
∂¯τ , 964 B(X → Y ) linear and (topologically)
∂τCj k complex tangential derivative bounded operators from X to
operator, 963 Y , 926
∂τ j k pointwise tangential derivative Bd(X) linear and bounded operators
operator, 962 on X, 926
∂τ j k tangential derivative operator, Bd X → Y linear and (norm)
bounded operators from X to
. 962
∂τ j k weak tangential derivative, 960 Y , 926
∂¯b boundary d-bar operator, 963 BMO−1 (∂Ω, σ), 928
∂¯ d-bar operator, 963 BMO(X, μ) space of functions of
∂, ∂z the conjugate of the bounded mean oscillations, 926
Cauchy-Riemann operator, 962 .
 · BMO(X,μ) , homogeneous BMO
∂, ∂z̄ the Cauchy-Riemann operator, semi-norm, 926
962  · BMO(X,μ) inhomogeneous BMO
∂,
¯ ∂, d-bar operator and its complex “norm”, 926
conjugate on forms, 963  f ∗ (Δ) local BMO norm of f on Δ,
ΠΩ Newtonian (volume) potential 925
operator, 962 
BMO(X, μ) the space BMO modulo
Πα,β complex volume (Newtonian) constants, 926
potential operator, 964 B0,β boundary-to-domain
πκ (E), πΩ,κ (E) “shadow” (or Bochner-Martinelli integral
projection) of E ⊆ Ω onto ∂Ω, operator on (0, β)-forms, 929
961 Bα,β boundary-to-domain
ϑ̄, 970 Bochner-Martinelli integral
ϑ, 970 operator on (α, β)-forms, 929
A (global) transpose of A, 924 Bα,β boundary-to-boundary
Aα,β boundary-to-domain integral Bochner-Martinelli integral
operator, 925 operator on (α, β)-forms, 930
Aα,β boundary-to-boundary integral Bn−1 (x , r) open ball with center x 
operator, 925 and radius r in Rn−1 , 925
986 Symbol Index

Bρ (x, r) ρ-ball with center at x and C boundary-to-domain

radius r, 925 Cauchy-Clifford integral
p,q
ℬα (Ω; D) Besov-Hardy space of operator, 932
null-solutions of D in Ω, 930 Cmod modified boundary-to-domain
p,q
ℬs (∂Ω; D) boundary Besov-Hardy Cauchy-Clifford integral
operator, 933
. p,q space, 930 C boundary-to-boundary
Bs (Σ, σ) homogeneous Besov
space on the set Σ, 927 Cauchy-Clifford integral
operator, 932
 · B. p, q (Σ,σ) homogeneous Besov
s 𝒞mod modified boundary-to-domain
quasi-seminorm, 927 Cauchy integral operator, 933
p,q
Bs (Σ, σ) (inhomogeneous) Besov C L left-handed
space on the set Σ, 928 boundary-to-boundary
 · Bs q (Σ,σ) (inhomogeneous)
p, Cauchy-Clifford integral
Besov quasi-norm, 928 operator, 932
p,q
Bs (Rn ) Besov space in Rn , 928 C # transpose Cauchy-Clifford
 · Bsp, q (Rn ) quasi-norm in the Besov integral operator, 932
space in Rn , 928 𝒞k (Ω) functions of class 𝒞k in an
q,λ
B1 (∂Ω, σ) block-based Sobolev open neighborhood of Ω, 930
space, 928 𝒞kc (Ω) functions of class 𝒞k with
p,q,λ compact support in the open
B1 (∂Ω, σ) off-diagonal
set Ω, 930
block-based Sobolev space,
𝒞kb (Ω) bounded functions of class
. q,λ 928 𝒞k in Ω, 930
B1 (∂Ω, σ) block-based ∞ ∗
homogeneous Sobolev space, 𝒞b (Ω) the algebraic dual of
929 𝒞∞b (Ω), 930
.
B q,λ (Σ, σ) block space, 927 𝒞α (U, ρ) homogeneous Hölder
space, 931
 ·  B q, λ (Σ,σ) norm on block space,
927  · 𝒞. α (U,ρ) homogeneous Hölder
q,λ
B−1 (∂Ω, σ) block-based negative . space semi-norm, 931
Sobolev space, 929 𝒞α (U, ρ)/∼ homogeneous Hölder
Borelτ (X) Borelians of the . α space modulo constants, 931
topological space (X, τ), 926 𝒞loc (U, ρ) local homogeneous
p,q Hölder space, 931
Bs (Ω) Besov space in Ω, 928
𝒞α (U, ρ) inhomogeneous Hölder
 · Bsp, q (Ω) quasi-norm in the Besov
space, 931
space in Ω, 928
 · 𝒞α (U,ρ) inhomogeneous Hölder
 ·  B q, λ (∂Ω,σ) norm in the space norm, 931
1
block-based Sobolev space of 𝒞αc (U, ρ) Hölder functions with
order one on ∂Ω, 928 ρ-bounded support, 931
.
 ·  B p, q, λ (∂Ω,σ) norm in the 𝒞γvan (Σ) homogeneous vanishing
1
off-diagonal block-based Hölder space, 932
Sobolev space of order one on 𝒞γvan (Σ) inhomogeneous vanishing
∂Ω, 929 Hölder space, 932
Symbol Index 987

CBM(Ω) complex Borel measures in D homogeneous Dirac operator in

Ω, 930 Rn , 934
CMO(Σ, σ), 932 divF the divergence of the vector
C n Clifford algebra generated by n field F,  933
imaginary units, 930 dim X dimension of X, 935
Cp(X → Y ) space of compact linear dV volume element in Cn , 935
operators from X into Y , 932 E = Eαβ 1≤α,β ≤M fundamental
Cp(X) space of compact linear solution of the system L, 937
operators from X into itself, EΔ standard fundamental solution for
932 the Laplacian, 937
.p ℰ(Ω) distributions compactly
Cq,η (Σ, σ) homogeneous Calderón
space, 933 supported in Ω, 937
p
Cq,η (Σ, σ) inhomogeneous Calderón En (ζ, z), 937
space, 933 E p (X) p-envelope of X, 937
Ex∂Ω→Ω , the extension operator
D(X) dyadic grid on X, 935
from ∂Ω to Ω, 937
DΔ,mod boundary-to-domain modified
e j standard j-th unit vector in Rn ,
harmonic double layer
936
potential operator, 936
{e j }1≤ j ≤n standard orthonormal
Dk (X), 935
basis in Rn , 937
D, D A boundary-to-domain double p,q
Fs (Rn ) Triebel-Lizorkin space in
layer (for a generic system), Rn , 939
935  · Fsp, q (Rn ) quasi-norm in the
Dλ boundary-to-domain double Triebel-Lizorkin space in Rn ,
layer for the Stokes system, 936
Dmod boundary-to-domain modified . p,q 939
Fs (Σ, σ) homogeneous
double layer potential, 935 Triebel-Lizorkin space on the
Dα,β complex double layer potential, set Σ, 938
936  · F. p, q (Σ,σ) homogeneous
D (Ω) space of distributions in Ω, s
Triebel-Lizorkin
933 quasi-seminorm, 938
D = nj=1 e j  ∂j Dirac operator in p,q
Fs (Σ, σ) (inhomogeneous)
Rn , 934 Triebel-Lizorkin space on the
D first-order system, 934 set Σ, 939
D (real) transpose of the first-order  · Fsp, q (Σ,σ) (inhomogeneous)
system D, 934 Triebel-Lizorkin quasi-norm,
D complex conjugate of the 939
p,q
first-order system D, 934 ℱα (Ω; D) Triebel-Lizorkin-Hardy
D∗ Hermitian adjoint of the space of null-solutions of D in
first-order system D, 934 Ω, 939
Def deformation tensor, 935 ℱφ = φ' Fourier transform of φ, 939
D L Dirac operator acting from the fBρ (x,r) integral average of f over
left, 934 Bρ (x, r), 945
DR Dirac operator acting from the fγ Fefferman-Stein grand maximal
right, 934 function of f , 938
988 Symbol Index

fp# L p -based Fefferman-Stein sharp H. p (Σ, σ) Hardy space, 941

maximal function, 967 H p (Σ, σ) homogeneous Hardy
p,q
Fs (Ω) Triebel-Lizorkin space in Ω, space, 941
939  ·  H p (Σ,σ) quasi-norm on Hardy
 · Fs q (Ω) quasi-norm in
p, space, 941
Triebel-Lizorkin space in Ω, H p,q (Σ, σ) Lorentz-based Hardy
939 space, 941
ftan,C , fnor,C , 939  ·  H p, q (Σ,σ) quasi-norm on
GΩ (·, ·) Green function for the Lorentz-based Hardy space,
Laplacian, 940 941
g = 1≤ j,k ≤n g jk dx j ⊗ dxk p,q
Hfin (Σ, σ) finite linear combinations
Riemannian metric tensor, 939 of atoms, 941
H n−1 the (n − 1)-dimensional  ·  H p, q (Σ,σ) quasi-norm on
Hausdorff measure in Rn , 941 fin
p,q
Hfin (Σ, σ), 941
H s-dimensional Hausdorff
s
p
Hκ (Ω) Hardy space of harmonic
measure in Rn , 941
functions, 942
Hmod modified Hilbert transform, 944
ℋq,λ (Σ, σ), 941
H L p -filtering operator, 941
 · ℋq, λ (Σ,σ) , 942
H s (Ω) L 2 -based fractional Sobolev
space in Ω, 942 𝒪(Ω) holomorphic functions in Ω,
 ·  H s (Ω) norm in the L 2 -based 960
fractional Sobolev space in Ω, IE,α fractional integral operator of
order α on E, 945
. p 942 Im T : X → Y image of T : X → Y ,
H1 (∂Ω, σ) Hardy-based
homogeneous Sobolev space, 945
i(Φ) lower dilation index of Φ, 945
. p 942  I(Φ) upper dilation index of Φ, 945
H1 (∂Ω, σ) ∼ classes of
equivalence, modulo ι∗ pull-back map induced by the
. p constants, canonical inclusion ι, 944
of functions in H1 (∂Ω, σ), 943
p,q
H1 (∂Ω, σ) Hardy-based K, K A boundary-to-boundary double
inhomogeneous Sobolev space, layer (for a generic system),
946
. p 943
H1,± (∂Ω, σ) homogeneous Hardy Kmod boundary-to-boundary modified
spaces with regularity, 944 double layer potential operator,
Hλ(1) (·) Hankel function of the first 946
kind, 942 K # , K A# transpose double layer (for a
p
ℋ• (∂Ω; D) “bullet” boundary Hardy generic system), 946
space, 943 KΔ boundary-to-boundary harmonic
H p (Ω; D) Hardy space in Ω, double layer potential, 945
associated with the first-order KΔ# transpose harmonic double layer
operator D, 943 potential, 945
 ·  H p (Ω;D) quasi-norm in the Hardy Kλ boundary-to-boundary double
space H p (Ω; D), 943 layer for the Stokes system, 947
H (Ω) holomorphic Hardy space in
p Kλ# transpose double layer for the
Ω, 944 Stokes system, 947
Symbol Index 989

Lnor,C (∂∗ Ω, σ) ⊗ Λα,β complex

p
Ker T : X → Y kernel of
T : X → Y , 945 normal forms, 954
p
Ker D null-space of the system D, L1 (∂∗ Ω, σ∗ ) L p -based (boundary)
946 Sobolev space, 950
Ker L null-space of the system L, 945  ·  L p (∂∗ Ω,σ∗ ) norm on Sobolev
1
Knβ (ζ, z) Bochner-Martinelli kernel space, 950, 951
for (0, β)-forms, 947 p
L1,loc (∂∗ Ω, σ∗ ) local (boundary)
k x0 Radon-Nikodym derivative of Sobolev space, 951
p
ω x0 with respect to σ, 947 L1 (∂Ω, w) Muckenhoupt weighted
L homogeneous second-order system (boundary) Sobolev space, 952
p
of differential operators, 953 L1 (∂∗ Ω, wσ∗ ) weighted Sobolev
L A second-order system associated . p space, 951
to a coefficient tensor A, 953 L1 (∂Ω, σ) homogeneous Sobolev
L(ξ) characteristic matrix of L, 953 space, 951
L(X → Y ) linear and continuous L p,q (X, μ) Lorentz space on X with
operators from X to Y , 949 respect to the measure μ, 949
L n Lebesgue measure in Rn , 948  ·  L p, q (X,μ) Lorentz space
LB = ∂z̄2 Bitsadze’s operator, 953 quasi-norm, 948
p,q
Lλ Stokes system (in a special L (Ω, μ) maximal Lorentz space,
writing), 953 949
p,q
Lμ,λ the Lamé system, 953 L1 (∂Ω, σ) Lorentz-based Sobolev
Lλ,μ (ξ) characteristic matrix of the space, 952
p
Lamé system, 953 Ls (Ω) Bessel potential space in Ω,
Lip(X) space of Lipschitz functions 950
on X, 948  ·  Lsp (Ω) norm in the Bessel
Lipc (X) space of Lipschitz functions potential space in Ω, 950
with bounded support in X, L Φ (X, μ) Orlicz space, 949
948  ·  L Φ (X,μ) Luxemburg norm on the

Lipc (Σ) distributions on Σ, 948 Orlicz space L Φ (X, μ), 949
L (X, μ) measurable functions which
0 L (log L)α Zygmund’s space, 949
p
p,q
are a.e. pointwise finite, 948 L1 (∂∗ Ω, σ∗ ) off-diagonal
L (X, μ) ⊗ Λα,β differential forms of
p (boundary) Sobolev space, 950
p,q
type (α, β), 954 L1,loc (∂∗ Ω, σ∗ ) local off-diagonal
∞
Lcomp essentially bounded functions (boundary) Sobolev space, 951
p
with compact support, 948 L−1 (∂∗ Ω, σ∗ ) negative Sobolev
p
Lbdd (Ω, L n ) p-th power integrable space, 952
p
functions over bounded subsets L−1 (∂Ω, w) Muckenhoupt weighted
of Ω, 948 negative Sobolev space, 952
p,q
L p (Ω, wL n ) weighted L p Lebesgue L−1 (∂∗ Ω, σ∗ ) off-diagonal negative
space over Ω, 949 Sobolev space, 952
p 1, p,q
L (Ω, μ) maximal Lebesgue space, LC (∂∗ Ω, σ) off-diagonal (partial)
949 Sobolev space, 954
Ltan,C (∂∗ Ω, σ) ⊗ Λα,β complex
p
log+ positive ln, 949
p
tangential forms, 954 L A,z (∂Ω, σ), 953
990 Symbol Index
p,1
LC (∂∗ Ω, σ) (boundary) complex M X Hardy-Littlewood maximal
Sobolev space, 954 operator on X, 955
p,1
LC,loc (∂∗ Ω, σ), 954 M X,s L s -based Hardy-Littlewood
maximal operator, 955
LC (∂∗ Ω, σ) ⊗ Λα,β , 954
p,1

p, ∂¯
M X,s,α fractional Hardy-Littlewood
Lα,β b (∂Ω, σ), 955 maximal operator, 955
p, ∂¯
Lα,β τ (∂∗ Ω, σ), 955 N0 = N ∪ {0}, 958
p
. Nκ (Ω; μ), 959
L p,λ (Σ, σ) homogeneous
Nκ nontangential maximal operator,
Morrey-Campanato space, 950
958
 · L. p, λ (Σ,σ) Morrey-Campanato
NκE the nontangential maximal
semi-norm, 950 operator restricted to E, 958
L p,λ (Σ, σ) inhomogeneous Nκε the nontangential maximal
Morrey-Campanato space, 950 function truncated at height ε,
 · L p, λ (Σ,σ) Morrey-Campanato 958
norm, 950 Oε one-sided collar neighborhood of
Mb operator of pointwise ∂Ω, 960
multiplication by the function pX lower Boyd index, 961
b, 957 qX upper Boyd index, 964
M p,λ (Σ, σ) Morrey space, 955 P maximal function of Carleson
 ·  M p, λ (Σ,σ) norm on Morrey space, type, 960
956 P.V. b k(x − ·)|Σ principal-value
M̊ (Σ, σ) vanishing Morrey space,
p,λ
distribution on the set Σ, 962
956 Pλ (double layer) pressure potential
p,λ
M1 (∂Ω, σ) Morrey-based Sobolev for the Stokes system, 964
space, 956 Q (single layer) pressure potential for
p,q,λ
M̊1 (∂Ω, σ) off-diagonal the Stokes system, 964
vanishing Morrey-based R+n upper half-space in Rn , 964
. p,λ Sobolev space, 956 R−n lower half-space in Rn , 964
M1 (∂Ω, σ) homogeneous R j boundary-to-boundary Riesz
Morrey-based Sobolev space, transform, 964
mod
957 R j modified Riesz transform, 966
p,λ
M̊1 (∂Ω, σ) vanishing RC, j boundary-to-boundary complex
Morrey-based Sobolev space, Riesz transform, 966
. p,λ 957 R jk , 965
M1 (∂Ω, σ) homogeneous (R jk )max , 965
vanishing Morrey-based R#jk , 965
Sobolev space, 957 R j boundary-to-domain Riesz
p,λ
M−1 (∂Ω, σ) Morrey-based negative transform, 965
Sobolev space, 957 R C, j boundary-to-domain complex
ℳ+ (X, μ) non-negative Riesz transform, 966
μ-measurable functions on X, R jk , 965
955 R C boundary-to-domain
ℳ(X, μ) μ-measurable functions on Clifford-Riesz transform, 966
X, 955 rad(Ω), 964
Symbol Index 991
κ−n.t.
S boundary-to-boundary single layer u|∂Ω (x) nontangential trace of u
(for a generic system), 968 at x ∈ ∂Ω, 959
Smod boundary-to-boundary modified u,θ solid maximal function of u, 967
single layer potential operator, UC(X, ρ) the space of uniformly
968 continuous functions on the
Sα,β boundary-to-boundary complex metric space (X, ρ), 970
single layer operator, 968 umax
M tangential maximal function of
S n−1 unit sphere in Rn , 967 u, 955
S±n−1 upper/lower hemispheres of VMO(X, μ) space of functions of
S n−1 , 967 vanishing mean oscillations,
Sym(D; ξ) principal symbol of the 970
first-order system D, 966 VMO−1 (∂Ω, σ), 970
𝒮 boundary-to-domain single layer VMO−1 (∂Ω, σ) in the
(for a generic system), 967 two-dimensional setting, 970
𝒮 boundary-to-domain single layer W k, p (Ω) L p -based Sobolev space of
for the Stokes system, 968 order k in Ω, 971
𝒮mod boundary-to-domain modified k, p
Wbdd (Ω), 971
single layer potential operator, k, p
Wloc (Ω) local L p -based Sobolev
967 space of order k in Ω, 971
𝒮Δ,mod boundary-to-domain modified W k, p (Rn, wL n ) weighted Sobolev
harmonic single layer potential spaces in Rn , 971
operator, 968
 · W k, p (Rn, w L n ) norm in the
𝒮α,β boundary-to-domain complex weighted Sobolev spaces in
single layer operator, 968 Rn , 971
𝒮(Rn ) Schwartz functions, 967 k, p
Wa (Ω) weighted Sobolev space in
𝒮(Rn ) tempered distributions, 967 ap
Ω, for the weight w := δ∂Ω , 971
supp f support of the measurable  · W k, p (Ω) quasi-norm in the
function f , 967 a
weighted Sobolev space
T ∗ adjoint of T, 924 k, p
Wa (Ω), 971
Tε truncated singular integral
W̊a (Ω) closure of 𝒞∞
1, p
c (Ω) in
operator, 969 1, p
Wa (Ω), 971
Tmax maximal operator, 969 k, p
Tmod modified principal-value Wa, (Ω) weighted maximal Sobolev
singular integral operator, 969 space, 972
TrΩ→∂Ω trace operator from Ω to  · W k, p (Ω) quasi-norm in weighted
a,
∂Ω, 969 maximal Sobolev space, 972
−1, p
TrRn →Σ trace operator from Rn to Σ, Wa (Ω) weighted Sobolev space of
969 order −1 in Ω, 971
TrΩ→∂Ω trace operator from Ω to  · W −1, p (Ω) norm on weighted
a
∂Ω, 969 Sobolev space of order −1 in
Tmod boundary-to-domain modified Ω, 971
1, p
integral operator, 969 𝒲a (Ω; D) weighted
Tγ (x) bump functions centered at x, Sobolev-Hardy space of
968 null-solutions of D in Ω, 972
992 Symbol Index

(X0, X1 )θ,q real interpolation  · X norm on the Generalized

intermediate space, 945 Banach Function Space X, 940
𝒳bdd (Ω), 𝒳(Ω)bdd , 925, 972 X associated space of X, 972
X Generalized Banach Function  · X norm on the associated space
Space on the measure space of X, 972
(X, M, μ), 940 ∞
X̊ closure of Lcomp in X, 940