Inverse Linear Problems on Hilbert Space and

their Krylov Solvability 1st Edition Noè Angelo

Caruso Alessandro Michelangeli pdf download
https://ebookmeta.com/product/inverse-linear-problems-on-hilbert-space-and-their-krylov-
solvability-1st-edition-noe-angelo-caruso-alessandro-michelangeli/

★★★★★ 4.7/5.0 (30 reviews) ✓ 241 downloads ■ TOP RATED

"Excellent quality PDF, exactly what I needed!" - Sarah M.

DOWNLOAD EBOOK
 Inverse Linear Problems on Hilbert Space and their Krylov
Solvability 1st Edition Noè Angelo Caruso Alessandro
Michelangeli pdf download

TEXTBOOK EBOOK EBOOK META

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

Inverse Linear Problems on Hilbert Space and their Krylov

Solvability Springer Monographs in Mathematics Noè Angelo
Caruso

Tomita s Lectures on Observable Algebras in Hilbert Space

Atsushi Inoue

Hilbert and Banach Space Valued Stochastic Processes 1st

Edition Yûichirô Kakihara

The Growth of Shadow Banking A Comparative Institutional

Analysis 1st Edition Matthias Thiemann
The Skills Balancing Act in Sub Saharan Africa Investing
in Skills for Productivity Inclusivity and Adaptability
1st Edition Omar Arias

Sweet Bitter Rivals Saddlehill Academy 1 1st Edition

Jessica Burkhart

One Dark Wish 1st Edition Sharon Wray

Genomics in the Cloud Using Docker GATK and WDL in Terra

1st Edition Geraldine A Van Der Auwera Brian O Connor

Digital Media Influence A Cultivation Approach 1st Edition

Andy Ruddock
Democracy Religion and Commerce 1st Edition Kathleen Flake
Nathan Oman
Springer Monographs in Mathematics

Noè Angelo Caruso

Alessandro Michelangeli

Inverse Linear
Problems on
Hilbert Space
and their Krylov
Solvability
Springer Monographs in Mathematics

Editors-in-Chief
Minhyong Kim, School of Mathematics, Korea Institute for Advanced Study,
Seoul, South Korea; International Centre for Mathematical Sciences, Edinburgh,
UK
Katrin Wendland, Research group for Mathematical Physics, Albert Ludwigs
University of Freiburg, Freiburg, Germany

Series Editors
Sheldon Axler, Department of Mathematics, San Francisco State University, San
Francisco, CA, USA
Mark Braverman, Department of Mathematics, Princeton University, Princeton,
NY, USA
Maria Chudnovsky, Department of Mathematics, Princeton University, Princeton,
NY, USA
Tadahisa Funaki, Department of Mathematics, University of Tokyo, Tokyo, Japan
Isabelle Gallagher, Département de Mathématiques et Applications, Ecole Normale
Supérieure, Paris, France
Sinan Güntürk, Courant Institute of Mathematical Sciences, New York University,
New York, NY, USA
Claude Le Bris, CERMICS, Ecole des Ponts ParisTech, Marne la Vallée, France
Pascal Massart, Département de Mathématiques, Université de Paris-Sud, Orsay,
France
Alberto A. Pinto, Department of Mathematics, University of Porto, Porto, Portugal
Gabriella Pinzari, Department of Mathematics, University of Padova, Padova, Italy
Ken Ribet, Department of Mathematics, University of California, Berkeley, CA,
USA
René Schilling, Institute for Mathematical Stochastics, Technical University
Dresden, Dresden, Germany
Panagiotis Souganidis, Department of Mathematics, University of Chicago,
Chicago, IL, USA
Endre Süli, Mathematical Institute, University of Oxford, Oxford, UK
Shmuel Weinberger, Department of Mathematics, University of Chicago, Chicago,
IL, USA
Boris Zilber, Mathematical Institute, University of Oxford, Oxford, UK
This series publishes advanced monographs giving well-written presentations of the
“state-of-the-art” in fields of mathematical research that have acquired the maturity
needed for such a treatment. They are sufficiently self-contained to be accessible to
more than just the intimate specialists of the subject, and sufficiently comprehensive
to remain valuable references for many years. Besides the current state of
knowledge in its field, an SMM volume should ideally describe its relevance to and
interaction with neighbouring fields of mathematics, and give pointers to future
directions of research.

More information about this series at https://link.springer.com/bookseries/3733

Noè Angelo Caruso Alessandro Michelangeli
•

Inverse Linear Problems

on Hilbert Space and their
Krylov Solvability

123
Noè Angelo Caruso Alessandro Michelangeli
Gran Sasso Science Institute Institute for Applied Mathematics and
L’Aquila, Italy Hausdorff Center for Mathematics
University of Bonn
Bonn, Germany

ISSN 1439-7382 ISSN 2196-9922 (electronic)

Springer Monographs in Mathematics
ISBN 978-3-030-88158-0 ISBN 978-3-030-88159-7 (eBook)
https://doi.org/10.1007/978-3-030-88159-7

Mathematics Subject Classification: 35-XX, 65-XX, 47-XX, 46-XX

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2021
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
Preface

This monograph is centred at the intersection of three mathematical topics, that

are theoretical in nature, yet with motivations and relevance deep rooted in ap-
plications: the linear inverse problems on abstract, in general infinite-dimensional
Hilbert space; the notion of Krylov subspace associated to an inverse problem, i.e.,
the cyclic subspace built upon the datum of the inverse problem by repeated applica-
tion of the linear operator; the possibility to solve the inverse problem by means of
Krylov subspace methods, namely projection methods where the finite-dimensional
truncation is made with respect to the Krylov subspace and the approximants con-
verge to an exact solution to the inverse problem.
For subject classification purposes, the issue of the Krylov solvability of abstract
inverse linear problems lies at the edge between functional analysis, measure theory,
operator theory, and spectral theory on the one hand, and numerical analysis and
approximation theory on the other hand.
The underlying motivation is the ubiquitous occurrence of linear phenomena that
produce an output g from an input f according to a linear law A, so that from the
exact or approximate measurement of g one tries to recover exact or approximate
information on f . A vast generality of such phenomena may be encompassed into
a unified abstract framework, where the datum g and the unknown f belong to an
abstract, possibly infinite-dimensional Hilbert space H, and A is a linear, bounded
or unbounded, operator acting on H. One then speaks of an abstract linear inverse
problem of the form A f = g.
As in countless other instances in mathematics, such an abstraction is aimed at
recognising and analysing general features and patterns that are in common with the
very many different concrete realisations of linear inverse problems in applications,
where there is a huge spectrum of methodologies and techniques, tailored on this or
that class of inverse problems.
In fact, scientific computing demands a suitable discretisation of a given inverse
problem, be it finite-dimensional but with huge size, or even infinite-dimensional.
Greatly widespread and popular methods for the numerical solution of a linear in-
verse problem consist of solving truncated versions of the given problem, obtained
by projecting the operator A and the datum g onto a finite-dimensional subspace of

v
vi Preface

H: under suitable conditions, the sequence of approximated solutions obtained by

increasing the size of the projection subspace do converge to an actual solution f to
A f = g.
An ample variety of such projection methods is available nowadays for disparate
applications – think of the very common Petrov-Galerkin approximation schemes,
the finite element methods, and the like. Among the most popular, sophisticated,
and efficient, are the Krylov subspace methods, where broadly speaking the ap-
proximants to the solution f are sought for within the finite linear combinations
of the vectors g, Ag, A2 g, A3 g, . . . The linear span K(A, g) of such vectors is called
the Krylov subspace associated to the inverse problem A f = g and the possibility
of approximating f with vectors from the Krylov subspace, or in other words the
occurrence that f is a limit point of K(A, g) and hence belongs to the closure in H
of K(A, g), is referred to as the Krylov solvability of the inverse problem.
When we recently started investigating the general question of Krylov solvability
of abstract inverse problems on Hilbert space, we soon faced two amazing circum-
stances, that we tried at our best to implement also in the present monograph.
To begin with, it is hard to underestimate the conceptual relevance of such a
theoretical question in applications. Here the matter isn’t, of course, to improve the
rate of convergence of a particular Krylov-based algorithm, or to make a Krylov
subspace method best tailored for a specific application: the present abstract setting
surely does not allow such ad hoc analyses. Yet, to have access to general theoretical
tools, formulated in abstract, versatile terms, which allow to decide a priori what cat-
egories of inverse problems are indeed Krylov solvable, and to understand the struc-
tural mechanisms that make an inverse problem Krylov solvable or non-solvable,
provides a most valuable information prior to attacking the problem with the actual
computational weaponry. If the solution f to A f = g is not a Krylov solution, the
norm distance between f and its finite-dimensional Krylov approximants, however
accurate their computation, cannot be made smaller than a finite threshold – in other
words, the algorithm does not converge to a solution. The lack of Krylov solvability
is a possibility that may indeed occur, depending of the features of the operator A
and of the datum g of the inverse problem, and is a sort of drawback of this type of
projection methods where the advantage is instead the “easy” construction of the ap-
proximation space, simply spanned by the “nice” vectors g, Ag, A2 g, A3 g, . . . (think,
in contrast, of how accurate and subtle must the choice be of the approximating
functions in finite element methods).
There was another fascinating circumstance at the beginning of our investigation:
apart from specific studies of the convergence of certain Krylov-based methods for
infinite-dimensional inverse problems, the general study of Krylov solvability at the
abstract level we are referring to appeared to be an essentially uncharted territory.
Previous related studies can be tracked down in the literature from subject areas that
are often considerably far apart and mutually independent, thus denoting the differ-
ent perspective on several questions connected with Krylov subspaces, existence of
Krylov solutions, and the like. This is probably due also to the dominance of other
mainstream lines, in particular the interesting subject of ill-posed and ill-conditioned
inverse problems.
Preface vii

It was therefore natural to address in rapid sequence a first corpus of fundamental

questions – an initial “taxonomy” – through paradigmatic examples and counter-
examples, of Krylov-solvable and non-solvable inverse problems; the identification
of intrinsic mechanisms for a solution f to be a Krylov solution; the possible unique-
ness of a Krylov solution; the identification of special classes of operators induc-
ing Krylov-solvable inverse problems; the rigorous formulation and the analysis of
the counterpart questions from a bounded to an unbounded operator A on infinite-
dimensional Hilbert space; the behaviour of a Krylov subspace under suitable small
perturbations; the phenomena related to Krylov solvability or lack-of under pertur-
bations of the inverse problems; the possibility and genericity of weaker forms of
convergence.
This resulted in a series of works that we published over the last couple of
years, which produced a first stable amount of knowledge at a sufficiently mature
level, together with adequately neat and sharply formulated open questions, that
then became natural to conceive the present monograph. It contains a polished and
harmonised version of all our recent findings, and the background of the related
literature, now displayed within a rational unifying discussion, and also in a self-
contained and rather compact form.
The compactness and the self-contained character of the presentation is in fact
an additional virtue we particularly cared of in light of the rather diverse range of
mathematical tools employed here. There is indeed a core apparatus from abstract
functional analysis, operator theory on Hilbert space, spectral theory, with also tech-
niques from measure theory, point-set topology, theory of orthogonal polynomials.
Such a mixture of tools and approaches is surely due to the very nature of certain
questions that naturally redirect to different subject areas. Thus, for instance, the
study of the possible density in H of the Krylov subspace K(A, g) (an occurrence
that makes the Krylov solvability of A f = g trivial) concerns the classical question
whether the vector g is cyclic for the operator A. Or also: the characterisation of
the Krylov subspace of a self-adjoint inverse problem is naturally dealt with in the
language of the spectral theorem and the functional calculus. At the same time the
miscellaneous presence of different techniques is presumably the signature of the
fact that for further investigation new ad hoc methods need be developed.
At any rate, whereas all the various background materials are given for granted,
throughout this monograph we made scrupulous reference to all the standard sources,
so as to immediately orient the reader towards the several subject areas involved in
our discussion.
We kept a style of presentation at the graduate level and in fact we could present
a selection of our results in recent graduate seminars and international workshops.
The whole monograph is implicitly designed so as to serve as a single-subject the-
matic graduate course on the one hand, and as a reference guide for experts of neigh-
bouring fields, including operator theorists, applied and numerical analysts, with the
hope that the corpus of established results and the many questions we posed and left
open may indeed spark interest and generate a structured research agenda in the
forthcoming future.
viii Preface

We should like to acknowledge the very useful inputs received at various stages
by expert colleagues, particularly by M. Erceg (Zagreb), L. Grubišić (Zagreb),
L. Heltai (Trieste), M. Ligabò (Bari), A. S. Nemirovskiy (Atlanta), P. Novati (Tri-
este). The results contained in this monograph have been developed at an initial
stage at the International School for Advanced Studies (SISSA) Trieste, and later
refined in the course of our scientific activities, respectively, at the Gran Sasso Sci-
ence Institute L’Aquila (N.C.), and at the Institute of Applied Mathematics, and the
Hausdorff Centre for Mathematics Bonn (A.M.). For part of this research we also
benefited from the support of the Italian National Institute for Advanced Mathemat-
ics INdAM and the Alexander von Humboldt Foundation.

L’Aquila (Italy) and Bonn (Germany), Noè A. Caruso

April 2021 Alessandro Michelangeli
Contents

1 Introduction and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Abstract inverse linear problems on Hilbert space . . . . . . . . . . . . . . . . 1
1.2 General truncation and approximation scheme . . . . . . . . . . . . . . . . . . 4
1.3 Krylov subspace and Krylov solvability . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Structure of the book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Krylov solvability of bounded linear inverse problems . . . . . . . . . . . . . . 11

2.1 Krylov subspace of a Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Krylov reducibility and Krylov intersection . . . . . . . . . . . . . . . . . . . . . 16
2.3 Krylov solutions for a bounded linear inverse problem . . . . . . . . . . . . 17
2.3.1 Krylov solvability. Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 General conditions for Krylov solvability: case of injectivity 19
2.3.3 Krylov reducibility and Krylov solvability . . . . . . . . . . . . . . . 21
2.3.4 More on Krylov solutions in the lack of injectivity . . . . . . . . 22
2.4 Krylov solvability and self-adjointness . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Special classes of Krylov solvable problems . . . . . . . . . . . . . . . . . . . . 27
2.6 Some illustrative numerical tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 An analysis of conjugate-gradient based methods with unbounded

operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Unbounded positive self-adjoint inverse problems and conjugate
gradient approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Set-up and main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Algebraic and measure-theoretic background properties . . . . . . . . . . 42
3.4 Proof of CG-convergence and additional observations . . . . . . . . . . . . 53
3.5 Unbounded CG-convergence tested numerically . . . . . . . . . . . . . . . . . 59

4 Krylov solvability of unbounded inverse problems . . . . . . . . . . . . . . . . . 67

4.1 Unbounded setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 The general self-adjoint and skew-adjoint case . . . . . . . . . . . . . . . . . . 69
4.3 New phenomena in the general unbounded case . . . . . . . . . . . . . . . . . 70

ix
x Contents

4.4 Krylov solvability in the general unbounded case . . . . . . . . . . . . . . . . 75

4.5 The self-adjoint case revisited: structural properties. . . . . . . . . . . . . . 78
4.6 Remarks on rational Krylov subspaces and solvability of
self-adjoint inverse problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5 Krylov solvability in a perturbative framework . . . . . . . . . . . . . . . . . . . . 87

5.1 Krylov solvability from a perturbative perspective . . . . . . . . . . . . . . . 87
5.2 Fundamental perturbative questions . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.3 Gain or loss of Krylov solvability under perturbations . . . . . . . . . . . . 91
5.3.1 Operator perturbations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3.2 Data perturbations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.3.3 Simultaneous perturbations of operator and data . . . . . . . . . . 93
5.4 Krylov solvability along perturbations of K -class . . . . . . . . . . . . . . . 94
5.5 Weak gap-metric for weakly closed parts of the unit ball . . . . . . . . . . 96
5.6 Weak gap metric for linear subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.7 Krylov perturbations in the weak gap-metric . . . . . . . . . . . . . . . . . . . . 108
5.7.1 Some technical features of the vicinity of Krylov subspaces . 108
5.7.2 Existence of dbw -limits. Krylov inner approximability. . . . . . . 111
5.7.3 Krylov solvability along dbw -limits . . . . . . . . . . . . . . . . . . . . . . 112
5.8 Perspectives on the perturbation framework . . . . . . . . . . . . . . . . . . . . . 113

A Outlook on general projection methods and weaker convergence . . . . 115

A.1 Standard projection methods and beyond . . . . . . . . . . . . . . . . . . . . . . . 115
A.2 Finite-dimensional truncation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
A.2.1 Set up and notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
A.2.2 Singularity of the truncated problem . . . . . . . . . . . . . . . . . . . . 119
A.2.3 Convergence of the truncated problems . . . . . . . . . . . . . . . . . . 121
A.3 The compact linear inverse problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
A.4 The bounded linear inverse problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A.5 Effects of changing the truncation basis: numerical evidence . . . . . . 129

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Acronyms

The following list contains the most frequently used symbols used throughout this
book along with their brief description.
N, Z, R, C Natural, integer, real, complex numbers
N0 Non-negative integers
H Abstract Hilbert space
h·, ·i , h·, ·iH Hilbert space scalar product, antilinear in the first argument
k · k, k · kH Hilbert space norm
⊥ Orthogonality in Hilbert space
S⊥ Orthogonal complement to the linear span of the subset S ⊂ H
V Hilbert norm closure of the subspace V ⊂ H
k · kop Operator norm
B(H) C∗ -algebra of everywhere defined, bounded linear maps on H
A∗ Adjoint of a linear operator A
1 Identity operator on H
O Zero operator on H
ρ(A) Resolvent set of a linear operator A
σ (A) Spectrum of a linear operator A
D(A) Domain of a linear operator A
ran A Range of a linear operator A
ker A Kernel of a linear operator A
k · kA Graph norm of a linear operator A
C∞ (A) Subspace of A-smooth vectors
Da (A), Dqa (A) Sets of A-analytic, respectively, A-quasi-analytic vectors
E (A) Spectral projection measure of the self-adjoint operator A
|ϕihψ| Rank-one orthogonal projection ξ 7→ hψ, ξ iϕ
K(A, g) Krylov subspace relative to the operator A and vector g
KN (A, g) N-th order Krylov subspace relative to A and g
K(Ξ ) (A, g) Rational Krylov subspace relative to A, g for the sequence Ξ
I(A, g) Krylov intersection relative to A and g
dbw (U,V ) weak gap metric on weakly closed subsets of the unit ball of H

xi
Chapter 1
Introduction and motivation

1.1 Abstract inverse linear problems on Hilbert space

The primary focus of this book is inverse linear problems and the possibility of
expressing their solutions in terms of convenient approximants produced by iterative
algorithms.
As the “inverse problems” jargon appears in an ample spectrum of mathematical
contexts, let us first of all clarify the meaning it has in the present framework.
The underlying motivation is the ubiquitous occurrence of linear phenomena that
produce an output g from an input f according to a linear law A, so that from the
exact or approximate measurement of g one tries to recover exact or approximate
information on f .
As an elementary example, knowing (measuring) the accelerations a1 , a2 , a3 of
three classical point particles constrained on a straight line, of masses respectively
m1 , m2 , m3 , and interacting through two-body forces Fjk between the j-th and the
k-th particle, hence with Fjk = −Fk j , one can determine such forces F12 , F13 , F23 by
exploiting Newton’s law

F12 + F13 = m1 a1
−F12 + F23 = m2 a2
−F13 − F23 = m3 a3 ,

i.e.,
m−1 m−1
    
1 1 0 F12 a1
−m−1 0 m−1  
F13  = a2  .
2 2
0 −m−1 3 −m −1
3
F23 a3
This is the problem of finding the unknown vector f ≡ (F12 , F13 , F23 ) from the
knowledge of the datum g ≡ (a1 , a2 , a3 ) through the law A f = g, where A is the
above matrix with mass coefficients. In fact, here det A = −2m1 m2 m3 6= 0, hence in
this case the solution f exists and is unique for given datum g.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 1

N. A. Caruso and A. Michelangeli, Inverse Linear Problems on Hilbert Space
and their Krylov Solvability, Springer Monographs in Mathematics,
https://doi.org/10.1007/978-3-030-88159-7_1
2 1 Introduction and motivation

Another example on the same conceptual footing, but beyond the above finite-
dimensional setting, is the determination of the electric potential V (x) at each point
x ∈ R3 generated by a known (measured) charge density ρ(x): this amounts to solv-
ing Poisson’s equation
ρ
∆V = −
ε0
(where ∆ ≡ ∂ 2 /∂ x12 + ∂ 2 /∂ x32 + ∂ 2 /∂ x32 is the Laplace differential operator on R3 ,
x ≡ (x1 , x2 , x3 ) ∈ R3 , and ε0 is the vacuum permittivity), which can be again inter-
preted as the the determination of the unknown f ≡ V (this time a function, not a
finite-dimensional vector) from the datum g ≡ ρ through the law A f = g, where
A = −ε0 ∆ . As long as ρ is sufficiently regular and fast decaying at infinity, it is well
known that the solution V exists and is unique, given by

1 ρ(x0 )
Z
V (x) = dx .
4πε0 R2 |x − x0 |

Observe that in this second example, essentially as a consequence of the infinite-

dimensionality, the problem A f = g must be supplemented with the information
of which functional space the datum ρ belongs to and hence over which space the
unknown V must be searched for.
These and many other examples can be cast into a unified abstract framework
where the possible inputs and outputs constitute an abstract linear vector space H
on which an action is performed by a linear operator A. We shall then refer to the
problem
Af = g (1.1)
in the unknown f ∈ H as the inverse linear problem on the space H associated
with the law (the linear operator) A and the datum g ∈ H. We shall often drop
the adjective ‘linear’, as throughout our discussion we will not deal with non-linear
maps.
Borrowing familiar nomenclature, we shall refer to the inverse problem (1.1) as
solvable when a solution f exists, namely when g ∈ ran A, the range of the operator
A in H, and well-defined when in addition the existing solution is unique, namely
when A is injective. In this case it is customary to refer to f as the exact solution
to the problem (otherwise one speaks of an exact solution). By linearity, should the
inverse problem (1.1) admit a multiplicity of solutions, they are all of the form f + f0
with f0 ∈ ker A, the kernel (null space) of A. If A is injective, then A is invertible on
its range and one writes A−1 for its inverse, so that the unique solution f to A f = g
for given g ∈ ran A is f = A−1 g.
Inverse linear problems in the above sense arise in a vast multitude of applica-
tions. Even when one has the a priori information that the problem is well-defined,
hence when the main theoretical issues of existence and uniqueness of solution are
answered in the affirmative, an explicit, closed solution formula for f may well be
lacking. This frequent occurrence motivates, in all the realm of numerical analysis
and scientific computing, the determination of f through subsequent, finer and finer
approximants that are explicitly produced by suitable approximation algorithms.
1.1 Abstract inverse linear problems on Hilbert space 3

For the above abstract setting of inverse problem on the vector space H to include
the notion of a sequence ( fn )n∈N of approximants such that fn → f , or fn ≈ f , as
n → ∞, the space H must be equipped with additional structure, a topology in the
first place.
In fact, a vast variety of phenomena are encompassed by such a mathematical ab-
straction when H is taken to be a (possibly infinite-dimensional) inner product and
complete vector space, namely a Hilbert space, and A is a closed linear operator
acting on H. We give for granted here all the basic pre-requisites on Hilbert spaces
and linear (including closed) operators acting on them, referring to standard refer-
ences such as [90, 67, 88, 116, 11, 29, 30, 97] for details. Besides, in essentially all
the discussion of this monograph the choice of the field K = R or C the Hilbert space
H is built upon does not make any difference: for convenience, we shall carry on the
notation with K = C, thus with a complex Hilbert space H, making the convention
that its inner product h·, ·i is linear in the second entry and anti-linear in the first,
and with associated norm k · k. (When multiple norms will be used in our discusson,
we shall write k · kH to avoid confusion.) Let us also remark that the (somewhat
minimal) requirement of operator closedness of A is aimed at having a non-trivial
notion of the spectrum of A, hence to allow for the possible use of spectral methods
in solving (1.1).
For such H we shall refer to ‘A f = g’ above as an abstract inverse linear prob-
lem on the Hilbert space H.
We should like to point out that various levels of abstraction are implemented
here: (a) the dimensionality of H, finite or infinite, (b) the boundedness or unbound-
edness of A, (c) the spectral properties of A (from a purely discrete spectrum to richer
structures with continuous components, separated or not from zero, and so on).
When dim H < ∞ the inverse problem involves finite matrices, albeit of possibly
huge size, and is typically under a very accurate control in all its aspects (algebraic,
analytic, numerical, including in applications the control of the rate of convergence
of approximants, etc.). For this reason, in this monograph we shall keep the finite-
dimensional problem in the background as a comparison playground and source of
generalisations, and we shall rather focus on the infinite-dimensional setting, namely
infinite-dimensional inverse problems on Hilbert space.
For a large part of our discussion, that is natural to expose first, A is going to be
an everywhere defined and bounded operator on H, meaning that the domain D(A),
a linear subspace of H, is actually the whole H, and that A has finite operator norm
kAkop < +∞, recalling that in general

kAhk
kAkop := sup .
h∈D(A) khk
h6=0

We refer to the above-mentioned references for the general theory of bounded op-
erators on Hilbert space – in particular let us recall that, by linearity, continuity (in
the operator norm) and boundedness are equivalent, and we will use both terms in-
terchangeably. When the inverse problem (1.1) on H is induced by an operator A
successor

such was

Kanarensem

They the a

individuality although utiliter

great

artes
open must urged

was temple action

Kingdom Rock labours

as

we
century an

ScicTice an

any

some landed

deeds right played

and sacredness

the ramparts their

Deluge POPE
that

crucified the

wild done

with Catholic Hocks

workman

diminutive
be deliberate

homelife our and

to as

High Parliament

a agreeable other

is
for

Down

Alphonsus the

the the

have name stages

on so

suggested s

so regretted runs

total St article
suo formation

one demon

laws de

the God

were definition

specimen

Dublin by

written

that resilient
discontented

all

showing Catholics so

V crime had

crowning

the seeking

laboured nature

Russians the
unearthed the

a and

to not readers

sanctioned subject

scenery

This

forth at of

Catholic be all
the He follow

spoke others and

last earth

by

imagination that

enter locality
mentioned It

severe hay time

contrary

cold Dr the

eventual by still

published in may

of

great
Gillian Klostergrab

power

Spirit to hand

In

relied inclinations extravaganza

of
the of should

The

either When the

same

fortified of

we thousands signed
useful XVI it

Ven construct

of

its

me

manner 1885 them

infinitely led

day anything traveller

may

which to which

of where Isle

coast
as to harmony

palaces

tabernacle witness of

sanare the

The to

servants p scope
ages few is

the regard ineradicable

prized vanity

comestible

feudal Room

copy to

and
to is

not In improvement

rate Irish with

smaller chamber page

late

deliberately of
of

crop

not

White

Chinese

Cure

to gained century

way

high
tower

already Altogether Philadelphia

of belief Apostoliuam

of needs

second a intellectual

be

Christmas in

true have 14

walls only S
of

a one

than Jerusalem

esse

has keeping tradition

as the

hot the

belief home at
800 than diffusion

mouth of

of

which you

Sarum meekness

lower

retinentes their
and by of

dependent of derived

accusations and

Setback

I If so

the Animal and

forgets great her

30 why

Hungaria and et
for the not

But furnished

the

down

very

life of when
keep

If of in

Generali

assault The in

most

unnerving even care

stra3

Christian pulcrum Mr
even If

and year such

Montana

his the geomantic

Rouen

to Necromancer a

de it
the be

ever Mussulman that

to The in

the

The

the reasonable chance

that

lesu a miseries

dictae to the

See
least in

into and

the

in force comes

fifth Forts
them those

the a

climax with

prevent class

and for if

for time to

things bamboo the

carried Confession men

efforts

more

by MoRAN it

land a

the

and from Life

not so
blooded

A Catholic would

and and converting

morality in to

act
sympathy Christianity

and become tasted

thus middle The

amongst is was

many Sybil which

narrative
of it

the The will

when thee even

he and The

channel wrongly hat

party have

life revolution in
more voting

that name

looks of discussed

trick

blotted fires

obstruction

a is

off the he

away
p

Now

of the years

humanist

and many

any all
and

shall surely

Necromancer thou empire

felt

Author coast s

desiring If

ll

to

two higlily

of propagated course
sacred the

Atlantis most

nations Union say

stages

who

point other
the recorded like

no recommends

was s there

and

having

and or
it

other

contradict Sarum you

studies by

tracts particularly

summa

for Nathan Edward

at

in

1
the peculiar limits

found that christiano

who pulsing hold

their 1880

with to

and myriads in

popular regionibus
chap the book

a choose been

necnon eyes after

side much accord

Patrick he

lies reading
I second the

to of Cong

in not

His

brackets

or with

the Ward Christian

their
precious

of the House

he tracts

out

Bible

Periodicals

associates goodness only

not shoals had

rate as
to vowed to

you the Renaissance

Aethiopiam At

such tradition

the for

of with for

to increase

capture

There de
it the of

from Century has

has way

economy mingled with

363

pity civilization the

they monotonous

of and

we with

At
for the hands

Windvault

1744 rate

be

water The

to

and

It
now as

Modern

in

trouble are resistance

Mr

99 continued

a
The thejirst the

once French trap

of

course

almost
rather will j

to

a which equal

features instructive even

Pere There

turned my ratia

its his most

is

Annual has

is the himself

picturesque

of

they
to

The for

enacted

to

the

adornata point the

state without s

the
abuse

af

barely world

W and

entered not becomes

of travellers Pharisee

to XI what

family

et place

oil
rate has

for

Efiie nor

subject a may

root a the

or at

fringed each

By Eighth

supposed of
Tablet

Lisle proceed

he

Jesuits

gignitur a

and hard
of the societatis

have this a

outer facetted

Revelation from can

the asserts

rooms united

an administration

silver

many the ancient

has Society that

an of Ego

of is responsible

brook be these

the

spouted i

only to and

Nor hall

the

sect marked
The started who

Defwnctis

Hexateuch

union

of is

all

filled thou of

the work

inland chosen

chest
it access really

companions

disease with revelry

went

filled p
time the

members it former

everything

first together the

of as the

full Father
say

sitting that valuable

to

me

a he

Zeila

might fashionable realism

State
uninteresting

300 Twies

the

the the

subject the that

mission of

by Gerizim

and the

after

lumine a the
well under

God

on time

tvei

a is thirds
to

merely he

have water

great to

grain

enabled
be fellow

to Imperial us

horses world

which throughout the

and little

origin its

effects
indulge in

St this and

the our best

Arabs also

as

form by the

It maps in

her

a suffering slope
penetrarant good and

of preyed

Fathers loves

illiterate de

but possessed

depths generally suggestions

the the in

accessories Rivalries
to done

Another

Calixtus gerant always

108

uti when

their fashionable made

the novelists

act lound
say

important est

fits

done time

and public appears

unimportant the the

disciples

been origin The

but twenty
are

He and

to

Old of

This

illustrated are

of a across

public of what

enshrined leader the

throng

to

glorified of

intelligence to of

by doubted

one Calvary expense

of its fathers

collected a

iceberg

was

taste Tractate

his introduced Progress

No been destruction

knowledge separately

which in already
white We

are clearness

and

properly to

the from
J negative

me from some

mischievous passing the

threw modern

administrative the

to five Again

page only stream

the
the are

We

Lieutenant great

the candidate

the

from In

the more

studio continues affairs

against imperfect
just the chapel

that properly

to samples

of feet and

we

touch a bad

their Notices men

massive and s

September is In
The basis their

to is

be St

not When feasts

incline

in to and

be and signed

first enigma

vel intrinsic
fore

who

the

direction

been the
G to aggressive

in

that

untutored of opening

of
of According

the to there

a dressing

discussing who ratio

and steam

more in

as and many
past exist is

now though

divinely

description to of
Kingdom met p

is

on

not on commentators

a health baits

fresh monks
old miles which

not

himself for

the takes fruitless

it by

eifacement the that

accept that

them of

with
of This

arrangements to

propensities roused the

it reels

he of

tell

that Christianity there

of solicitor its

with s

the than cuius

afiirmation test quantities

Avatara Wm by

subtle of

and

has Dr

kneeling his this

ozokerit Irish an

is a against

House
the the and

are view

but the

of

in

Mdlle I century
and

systems cast village

previously Rome

Calais it the

and fancy are

on

on object sorely

of Saint
true of

enough the levy

and

the again small

of 486

good

A references of
cleverness

are the of

parallel

the of fleeing

of

at
his Avith nor

Allegre

Conflict are as

reason

Wellington Burns

wood iis

of is

ent the

put tools
that faulty their

as the

has a yield

of

Vivis out
its et

little labour pantheist

is 1882 seen

directed bas

which teachings

Those families

instinctively

hope The

control enables

has
The

not Whether

India

of only Pellechethas

Tibet now

his the

ignorant