C A M B R I D G E T R AC T S I N M AT H E M AT I C S

General Editors

J . B E RTO I N , B . B O L L O B Á S , W. F U LTO N , B . K R A , I . M O E R D I J K ,
C . P R A E G E R , P. S A R NA K , B . S I M O N , B . TOTA RO

233 Category and Measure

 C A M B R I D G E T R AC T S I N M AT H E M AT I C S

G E N E R A L E D I TO R S

J. BERTOIN, B. BOLLOBÁS, W. FULTON, B. KRA, I. MOERDIJK,

C. PRAEGER, P. SARNAK, B. SIMON, B. TOTARO

A complete list of books in the series can be found at www.cambridge.org/mathematics.

Recent titles include the following:
200. Singularities of the Minimal Model Program. By J. Kollár
201. Coherence in Three-Dimensional Category Theory. By N. Gurski
202. Canonical Ramsey Theory on Polish Spaces. By V. Kanovei, M. Sabok, and J. Zapletal
203. A Primer on the Dirichlet Space. By O. El-Fallah, K. Kellay, J. Mashreghi, and
T. Ransford
204. Group Cohomology and Algebraic Cycles. By B. Totaro
205. Ridge Functions. By A. Pinkus
206. Probability on Real Lie Algebras. By U. Franz and N. Privault
207. Auxiliary Polynomials in Number Theory. By D. Masser
208. Representations of Elementary Abelian p-Groups and Vector Bundles. By D. J. Benson
209. Non-homogeneous Random Walks. By M. Menshikov, S. Popov, and A. Wade
210. Fourier Integrals in Classical Analysis (Second Edition). By C. D. Sogge
211. Eigenvalues, Multiplicities and Graphs. By C. R. Johnson and C. M. Saiago
212. Applications of Diophantine Approximation to Integral Points and Transcendence.
By P. Corvaja and U. Zannier
213. Variations on a Theme of Borel. By S. Weinberger
214. The Mathieu Groups. By A. A. Ivanov
215. Slenderness I: Foundations. By R. Dimitric
216. Justification Logic. By S. Artemov and M. Fitting
217. Defocusing Nonlinear Schrödinger Equations. By B. Dodson
218. The Random Matrix Theory of the Classical Compact Groups. By E. S. Meckes
219. Operator Analysis. By J. Agler, J. E. McCarthy, and N. J. Young
220. Lectures on Contact 3-Manifolds, Holomorphic Curves and Intersection Theory.
By C. Wendl
221. Matrix Positivity. By C. R. Johnson, R. L. Smith, and M. J. Tsatsomeros
222. Assouad Dimension and Fractal Geometry. By J. M. Fraser
223. Coarse Geometry of Topological Groups. By C. Rosendal
224. Attractors of Hamiltonian Nonlinear Partial Differential Equations. By A. Komech and
E. Kopylova
225. Noncommutative Function-Theoretic Operator Function and Applications. By J. A. Ball
and V. Bolotnikov
226. The Mordell Conjecture. By A. Moriwaki, H. Ikoma, and S. Kawaguchi
227. Transcendence and Linear Relations of 1-Periods. By A. Huber and G. Wüstholz
228. Point-Counting and the Zilber–Pink Conjecture. By J. Pila
229. Large Deviations for Markov Chains. By A. D. De Acosta
230. Fractional Sobolev Spaces and Inequalities. By D. E. Edmunds and W. D. Evans
231. Families of Varieties of General Type. By J. Kollár
232. The Art of Working with the Mathieu Group M24 . By R. T. Curtis
233. Category and Measure. By N. H. Bingham and A. J. Ostaszewski
234. The Theory of Countable Borel Equivalence Relations. By A. S. Kechris
 Category and Measure
Infinite Combinatorics, Topology and Groups

N. H. BINGHAM
Imperial College London

A DA M J . O S TA S Z E W S K I
London School of Economics and Political Science
 Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi - 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467

Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9780521196079
DOI: 10.1017/9781139048057
© N. H. Bingham and Adam J. Ostaszewski 2025
This publication is in copyright. Subject to statutory exception and to the provisions
of relevant collective licensing agreements, no reproduction of any part may take
place without the written permission of Cambridge University Press & Assessment.
When citing this work, please include a reference to the DOI 10.1017/9781139048057
First published 2025
A catalogue record for this publication is available from the British Library
A Cataloging-in-Publication data record for this book is available from the Library of
Congress
ISBN 978-0-521-19607-9 Hardback
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this
publication and does not guarantee that any content on such websites is, or will
remain, accurate or appropriate.
 To Cecilie

Nick

To Monika, and to Juliana and Konstancja

Adam
 Contents

Preface page xi
Prologue: Regular Variation 1
1 Preliminaries 24
1.1 Littlewood’s Three Principles 24
1.2 Topology: Preliminaries and Notation 25
1.3 Convergence Properties 28
1.4 Miscellaneous 30
2 Baire Category and Related Results 34
2.1 Baire’s Theorem 34
2.2 Banach’s Category Theorem 42
2.3 Countability Conditions – Topological 43
2.4 Countability Conditions – Games of Banach–Mazur Type 45
2.5 Notes 48
3 Borel Sets, Analytic Sets and Beyond: ∆21 51
3.1 Borel Sets and Analytic Sets 51
3.2 Beyond Analytic Sets: Projective Hierarchy 60
3.3 * Character Theorems 62
3.4 * Appendix: Baire Hull 68
4 Infinite Combinatorics in Rn : Shift-Compactness 71
4.1 Generic Dichotomy 71
4.2 Kestelman–Borwein–Ditor (KBD) Theorem: First Proof 73
5 Kingman Combinatorics and Shift-Compactness 78
5.1 Definitions and Notation 79
6 Groups and Norms: Birkhoff–Kakutani Theorem 86
6.1 Introduction 86
6.2 Analytic Shift Theorem 95

vii
viii Contents

6.3 The ‘Squared Pettis Theorem’ 98

6.4 Shifted-Covering Compactness 99
7 Density Topology 101
7.1 The Lebesgue Density Theorem 101
8 Other Fine Topologies 114
8.1 The Fine Topology of Potential Theory: Polar Sets 114
8.2 Analytically Heavy Topologies 120
8.3 Other Fine Topologies 123
8.4 * Topologies from Functions, Base Operators or Density
Operators 124
8.5 * Fine Topologies from Ideals I: Base Operators 127
8.6 * Fine Topologies from Ideals II: Modes of Convergence 131
8.7 * Coarse versus Fine Topologies 134
9 Category–Measure Duality 137
9.1 Introduction 137
9.2 Steinhaus Dichotomy 139
9.3 Subgroup Dichotomy 140
9.4 Darboux Dichotomy 144
9.5 The Fubini Null and the Kuratowski–Ulam Theorems 147
9.6 * Universal Kuratowski–Ulam Spaces 148
9.7 * Baire Products 158
10 Category Embedding Theorem and Infinite Combinatorics 161
10.1 Preliminaries 161
10.2 The Category Embedding Theorem, CET 162
10.3 Kestelman–Borwein–Ditor Theorem: CET Proof 165
10.4 Kestelman–Borwein–Ditor Theorem: Second Proof 167
10.5 Near-Closure and No Trumps 168
11 Effros’ Theorem and the Cornerstone Theorems of Functional
Analysis 172
11.1 Introduction 172
11.2 Action, Microtransitive Action, Shift-Compactness 175
11.3 Effros and Crimping Properties: E, EC and C 178
11.4 From Effros to the Open Mapping Theorem 181
12 Continuity and Coincidence Theorems 184
12.1 Continuity Theorems: Introduction 184
12.2 Banach–Neeb Theorem 188
12.3 Levi Coincidence Theorem 192
12.4 Semi-Polish Theorem 196
12.5 * Pol’s Continuity Theorem 199
 Contents ix

13 * Non-separable Variants 203

13.1 Non-separable Analyticity and the Baire Property 203
13.2 Nikodym Actions: Non-separable Case 211
13.3 Coincidence Theorems: Non-separable Case 212
14 Contrasts between Category and Measure 222
14.1 Classical Results 222
14.2 Modern Results: Forcing 229
14.3 Category Duality Revisited 231
15 Interior-Point Theorems: Steinhaus–Weil Theory 235
15.1 The Steinhaus–Piccard Theorem 235
15.2 The Common Basis Theorem 236
15.3 Measure Subcontinuity and Interior Points 239
15.4 The Simmons–Mospan Converse 251
15.5 The Steinhaus Property AA−1 versus AB−1 255
15.6 Borell’s Interior-Point Property 259
15.7 Haar-Meagre Sets in an Abelian Polish Group 261
15.8 Haar-Null Sets Revisited (the Abelian Case) 265
16 Axiomatics of Set Theory 268
16.1 The Three Elephants in the Room 268
16.2 Hypotheses and Axioms 269
16.3 Ordinals and Cardinals 272

Epilogue: Topological Regular Variation 278

References 291
Index 324
 Preface

The origins of this book stem largely from two earlier books, its ‘parents’, as
it were. The first is John Oxtoby’s (lovely, succinct) Measure and Category:
A Survey of the Analogies between Topological and Measure Spaces (1971,
1980) (the term category is used here in the sense of Baire category, not of
functors and categories in algebraic topology). Our own title Category and
Measure: Infinite Combinatorics, Topology and Groups both pays tribute to
Oxtoby’s book as motivation and inspiration and declares our main theme: for
us, it is category, rather than measure, that is paramount, while for Oxtoby it is
the other way round.
The second is the first author’s first book, Regular Variation (Bingham,
Goldie and Teugels, 1987, 1989) – ‘BGT’ hereafter, for brevity. Regular vari-
ation is described in the preface of BGT as ‘essentially a chapter in classical
real variable theory, together with its applications . . .’. BGT is now well estab-
lished, widely used and widely cited; it was intended to be comprehensive in
its treatment of the theory and in surveying the applications. It left three ‘open
ends’. The first two were the foundational question (BGT, p. 11: measurability
suffices, the Baire property suffices, neither includes the other; what is actu-
ally needed?), and the contextual question (what is its natural context?). The
third was the difficulty of the hardest theorem in the ‘theory’ part of BGT (Th.
1.4.3/Th. 3.2.5). The challenge there was to dissolve this difficulty. All three
matters have now been satisfactorily resolved.
While mathematics progresses by journal papers, it is consolidated by books.
Thus as the number of papers one writes in an area grows, one must confront
the fact that one’s work will be ‘for experts only’ unless one synthesizes the
corpus into a book, aimed at and addressed to the general mathematical public.
While a glance at our extensive References will show that our sources are wide-
ranging, our aim here is to bring the unifying essentials to the attention and put
them at the disposal of the ‘mathematician on the street’.

xi
xii Preface

Doing the above provided the motivation for the first phase of our joint work
(the first half, up to 2011) and its unifying connecting thread. Accordingly, we
addressed ourselves to writing it up in book form, under the title suggested by
the two ‘parent texts’, Topological Regular Variation. But under the momentum
generated by this first half of our corpus to date, the second half emerged. We
had the sense to realise that we should pause on writing the book, lest we ‘write
the wrong book’, and instead let the papers emerge, let the dust settle, and write
the book resulting from ‘both halves, rather than just the one’.
The result is the present book. Its title reflects its focus and viewpoint. Its
formal prerequisites are those of BGT: ‘a good background in analysis at, say,
first-year graduate level, including in particular a knowledge of measure theory’
(BGT, xviii), set theory – particularly descriptive set theory – and mathematical
logic. Any background here will only be helpful, but none is assumed; see for
instance our 2019 survey ‘Set Theory and the Analyst’.
The most important theorem in the ground we cover is the classic Baire Cate-
gory Theorem of 1899, while measure theory (Lebesgue, 1905) was mentioned
above. Each gives a family of small sets (meagre for category, null for measure).
The meagre sets and the null sets form σ-ideals, M and N , respectively. These,
along with their interplay, similarities and contrasts, pervade the book. So too
does the theory of analytic sets, much influenced by C. A. (Ambrose) Rogers,
the second author’s PhD supervisor.
The other main mathematical ingredient is infinite combinatorics, a field
which owes much to the work of Paul Erdős (1913–1996). Here we had to
fashion some of the tools we needed for ourselves, one reason for the fourteen-
year delay since 2011.
Regarding regular variation: this is in no sense a book about regular variation.
(Despite the large citation count of BGT, the first author has often been struck
by how many good, knowledgeable mathematicians have never heard of it –
though probabilists like himself use it all the time.) That said, the links between
this book and regular variation are too important to be ‘written in invisible ink’.
We have chosen to use regular variation as a ‘framing device’ for the book. The
book opens with a Prologue, containing a summary of ‘regular variation up to
BGT’. This can be read as part of the main text, skipped, or referred back to for
reference, depending on the taste and background of the reader. It closes with an
Epilogue, ‘Topological Regular Variation’, our summary of our own work here
since. (Much else has happened in regular variation since BGT, particularly in
probability and in higher dimensions, but this would take us too far afield.) This
framing device is unusual in a mathematical monograph but familiar enough in
other contexts. It is common to see several minutes of action before the credit
titles of a film (‘Prologue’). And we recall with pleasure Conrad’s use of the
 Preface xiii

narrator Marlow as a framing device in a number of his novels (‘Prologue and

Epilogue’).
We refer for background in several areas to standard works. Our main sources
are Engelking (Eng1989) for topology, Kechris (Kec1995) for descriptive set
theory and Bogachev (Bog2007a) for measure theory. In addition, we refer
repeatedly to Oxtoby’s classic above, and to the paper by C. A. Rogers and
J. E. Jayne (RogJ1980). This is contained in the proceedings of the London
Mathematical Society’s Instructional Conference on Analytic Sets, University
College London, 1978. This conference led eventually to our collaboration and
to this book. Analytic sets and their influence permeate the whole book.
While most of the text is addressed to the mathematical public generally,
some parts are more specialized. These are marked with an asterisk.
Long bibliographies when numbered burden the reader with remembering
numbers. We use instead ‘name plus date’, to guide the reader’s eye when
looking up references. For the name, we always use the first three letters of the
first author’s name (with more if needed to avoid ambiguity), plus the first letter
of the second and third authors’ names (thus our 2020 paper with Jabłońska and
Jabłoński is BinJJ2020). Dates are given in full. This is consistent with usage in
ordinary speech, and needed as there are three centuries involved (going back
to, e.g. Baire, 1899).
While chapters (Chapter n), sections (§n.p) and results (Theorem n.p.q) are
numbered in full, equations are numbered sparingly, and locally. Thus (*), (**)
are the first and second such equations in the current section or chapter.
Unless otherwise stated, e or eG will denote the group identity.
We cannot resist mentioning Tom Körner’s description of Baire’s category
theorem as ‘a profound triviality’, and Jean-Pierre Kahane quoting this and
describing the use he and Bob Kaufman had made of it as profound. Alas,
Rogers (see FalGO2015) and Kahane are no longer with us, but their influence
pervades the book, and we salute their memory with gratitude.
To close, we thank all those who have helped us over the years, most par-
ticularly our collaborators (some alas no longer with us, including the much-
missed Harry I. Miller (1939–2018)), in particular Eliza Jabłońska and Tomasz
Natkaniec for a careful reading respectively of §§15.7–15.8 and §§9.5–9.7. We
are most grateful to David Tranah and Roger Astley of Cambridge University
Press, for their kind forbearance during the book’s unconscionably long gesta-
tion period. Last and most, we thank our wives and families for their love and
support during the writing of this book.
 Prologue: Regular Variation

P.1 Introduction
Regular variation is a subject both of theoretical interest and of great use in
a variety of applications. These include analytic number theory (asymptotics
of arithmetic functions, Prime Number Theorem), complex analysis (entire
functions – Levin–Pfluger theory) and, particularly, probability theory (limit
theorems). The standard work here, covering theory and applications, is Bin-
GoT1987 (BGT below for brevity). As it happens, matters left open in BGT –
the foundational question, p. 11 (on measurability and the Baire property), and
the contextual question (Appendix 1, on contexts beyond the real analysis to
which the bulk of the text is devoted) – motivated our joint work. So this book
is motivated by two much earlier and by now well-established texts, Oxtoby
(Oxt1980) and BGT. But these serve only as background and motivation here;
this book is self-contained and may be read without reference to either.
To make the above more specific, here are some instances of ‘what regular
variation can do for the mathematician in the street’.

P.2 Probability Theory

The prototypical limit theorem in the subject is Kolmogorov’s 1 strong law of
large numbers: that if (X n ) is a sequence of independent copies of a random
variable X drawn from some distribution (or law) F, the averages Sn /n :=
1 X k /n converge as n increases to some limit c with probability 1 (‘almost
Pn

1 In 2023, the postal address of Moscow State University became 1 Kolmogorov Street; cf. 2
Stefan Banach Street, Warsaw, the postal address of the mathematics department of Warsaw
University.

1
2 Prologue: Regular Variation

surely’, a.s.) if and only if X has a mean (first moment, expectation) (meaning
E [ |X | ] < ∞), and then the limit is the mean µ = E [X]:
Sn /n → µ (n → ∞) a.s., where µ = E[X]. (LLN)
Second only to this is the central limit theorem: if also one has finite variance,
σ 2 say (finite second moment), and centres at means (subtracting E[Sn ] = nµ),
√
the right scaling is by nσ, and one then has a limit distribution, the standard
normal (or Gaussian) law Φ = N (0, 1):
Z x − 1 u2
√ e 2
P((Sn − nµ)/σ n ≤ x) → Φ(x) := √ du (n → ∞) for all x ∈ R.
−∞ 2π
(CLT)
Because these results are so important, and because one does not always have
a (finite) mean and variance, it was of great interest to find versions of them
which held under weaker moment conditions. It was realized by Sakovich in
1956 (Sak1956) that regular variation gave the right language here: what one
needs for the first is that the truncated mean is slowly varying,
Z x
udF (u) ∼ ` 1 (x) (x → ∞),
−x
and for the second that the truncated variance is slowly varying,
Z x
u2 dF (u) ∼ ` 2 (x) (x → ∞),
−x
where ` 1 , ` 2 are slowly varying (below).
Note. Oddly, despite their importance, these results were overlooked at the
time, and they were re-discovered and given prominence in Feller’s book
(Fel1966). The first author saw them there then (‘love at first sight’).
More curiously still, although regular variation dates back to 1930 (below),
the classic monograph by Gnedenko and Kolmogorov (GneK1954) (the Rus-
sian original is from 1949) made no use of it. So its treatment of these and
related matters is unnecessarily complicated, and in particular the analysis and
the probability are not properly separated. We note that Sakovich’s PhD was
supervised by Gnedenko.
For more on Gnedenko’s work, and his very interesting life, see Bin2014. 2
Then one has the third member of the trilogy, LLN–CLT–LIL: the law of
the iterated logarithm. Here the norming, which gives the result its name, is
intermediate between those in (LLN) and (CLT), and the conclusion is of a
different type:
lim sup (Sn − nµ)/σ 2n log log n = +1 a.s., lim inf · · · = −1 a.s.
p

2 The text of a talk given by the first author at the Gnedenko Centenary Memorial Meeting,
Moscow State University, 2012.
 Prologue: Regular Variation 3

Indeed,
p
(Sn − nµ)/σ 2n log log n →→ [−1, 1] a.s., (LIL)

meaning that all points in [−1, 1], and no others, are limits of subsequences,
a.s.
Stable laws (limit laws of centred and normed sums of independent copies)
provide another good example. See, e.g., two approaches to the ‘domain of
attraction’ problem here by Pitman and Pitman (PitP2016) and Ostaszewski
(Ost2016a).
For more on the early history of regular variation in probability theory, see
Bin2007.

Extremes The extreme values in a sample – sample maximum and minimum –

have always been of great practical importance (the strength of a chain is that of
its weakest link, etc.). The theory here dates back to Fisher and Tippett in 1928,
so to before regular variation, though the relevance of regular variation was soon
realized. The area is growing in importance nowadays, e.g. because of climate
change and global warming. There was pioneering early work by Gnedenko in
1943, but the systematic use of regular variation to study extremes stems from
de Haan in 1970 (Haa1970). For background and historical comments, see,
e.g., our recent survey BinO2021b and the references there.
While Hardy himself was not interested in probability, the Tauberian theory
to which he and Littlewood contributed so much has proved very useful in
probability theory; see, e.g., Bin2015b.

P.3 Complex Analysis

Recall (see, e.g., BGT, Ch. 4) that an Abelian theorem passes from a stronger
mode of convergence to a weaker one (such results are usually easy); a Taube-
rian one gives a converse, under an additional condition (a Tauberian condi-
tion); Mercerian theorems (see, e.g., BGT, Ch. 5) are hybrids, going from a
condition on both to a stronger conclusion, under no Tauberian condition. A
prototypical Abelian result will pass from a function

f (x) ∼ x ρ `(x) (` ∈ R0 ) (x → ∞)

to a Mellin convolution
Z ∞
( f ∗ k)(x) := k (t) f (x/t)dt/t,
0
4 Prologue: Regular Variation

where ρ lies in the vertical strip in the complex s-plane where the Mellin
transform Z ∞ Z ∞
k̂ (s) := t k (t)dt/t =
−s
us k (1/u)du/u
0 0
converges absolutely, giving
( f ∗ k)(x) ∼ k̂ ( ρ)x ρ `(x) (x → ∞);
the factor k̂ ( ρ) is to be expected, since if f (x) = x ρ , ( f ∗ k)(x) = k̂ ( ρ)x ρ .
The Tauberian converse reverses the implication for kernels satisfying Wiener’s
condition that k̂ (s) be non-vanishing for Re s = ρ ((NV) below), under suit-
able Tauberian conditions on f . The Mercerian (or ‘ratio Tauberian’, below)
statement assumes convergence of the quotient,
( f ∗ k)(x)/ f (x) → c (x → ∞)
for some constant c, and deduces regular variation of both as above, with
c = k̂ ( ρ).
For entire functions of finite order, one can look (in discs centre 0 and large
radius r) at growth rates of the function, its maximum modulus M (r) and the
zero-counting function n(r) (both in discs centre 0 and radius r). Matters split
between integer and non-integer order. Functions with real negative zeros are
simplest; write Eρ for the class of entire f with order ρ < ∞ and negative
zeros. For f ∈ Eρ , one has the Valiron–Titchmarsh theorem (BGT, §7.2, Th.
7.2.2), proved by Tauberian methods involving regular variation (BGT, Ch. 4),
based on the linear integral transform (Stieltjes transform)
Z ∞
zn(t)
log f (z) = dt/t (| arg z| < π).
0 t+z
For non-integral order, regular variation of either of n(r), log f (reiθ ) implies
regular variation of the other, and convergence of the quotient to a non-zero
limit. For more on the Valiron–Titchmarsh Theorem, see DrasS1970 and the
references cited there.
The question of whether convergence of this quotient implies regular vari-
ation of both functions has been called of ‘ratio Tauberian’ type; it is in fact
Mercerian (BGT, Ch. 5). The first such results are due to Edrei and Fuchs
(EdrF1966) and Shea (She1969), for f ∈ Eρ and the Stieltjes transform
above. Such results were extended by Drasin (Dras1968) and Drasin and Shea
(DrasS1976) to more general kernels, using Wiener Tauberian theory. Drasin
and Shea had non-negative kernels k, for which the relevant Mellin transforms
converge absolutely in their strip of convergence. Matters are more complicated
when the kernel can change sign (as with Fourier sine and cosine transforms,
 Prologue: Regular Variation 5

and Hankel transforms), as here there can be strips of conditional convergence

(or Abel summability) also; see Jor1974. The Fourier and Hankel cases were
considered in detail by Bingham and Inoue (BinI1997; BinI1999).
As may be seen from the Wiener Tauberian Theorem, P.7.1: if regular varia-
tion of index ρ (membership of Rρ ) is to appear in the hypothesis and conclu-
sion, the key condition on the kernel is the non-vanishing condition

k̂ (s) , 0 (Re s = ρ). (NV)

In the corresponding Mercerian results, the key condition on k is the no-repeat

condition
k̂ (s) = k̂ ( ρ) on Re s = ρ only for s = ρ (NR)

(She1969; Jor1974; cf. PalW1934, IV, (18.09)).

One can also consider the Nevanlinna characteristic
Z π
1
T (r) = T (r, f ) := log+ | f (reiθ )| dθ
2π −π
(see, e.g., Haym1964). For f ∈ Eρ , this is given by the non-linear integral
transform
(Z ∞ )
T (r) = sup P(r/t, θ)N (t)dt/t : θ ∈ (0, π) ,
0

where
r
sin θ
Z
1
N (r) := n(t)dt/t, P(t, θ) := .
0 π t + 2 cos θ + t −1
Baernstein (Bae1969) obtained a non-linear Tauberian theorem (the passage
from N to T is Abelian, and simple, EdrF1966): for f entire of genus 0, if
T (r) ∼ r λ `(r) as r → ∞ for ` slowly varying, then

(a) if λ ∈ [0, 21 ], then N (r) ∼ r λ `(r);

(b) if λ ∈ [ 12 , 1], and f has only negative zeros, then N (r) ∼ sin πλ r λ `(r).

The corresponding Mercerian (or ratio Tauberian) theorem was proved by Edrei
and Fuchs (EdrF1966), for λ ∈ [ 21 , 1]: if the ratio converges, then both N (r)
and T (r) are regularly varying.
Baernstein (Bae1969) conjectured that his results extend to M ρ , the class
of meromorphic functions of finite order ρ, negative zeros and positive poles,
but this is not the case. For counterexamples and discussion, see Dras2010.
However, they do extend to the subclass Jρ of M ρ whose zeros an and poles
bn are symmetrically related, an = −bn (Will1972). Edrei (Edr1969) removes
6 Prologue: Regular Variation

geometric restrictions on the zeros and poles, but at the cost of obtaining
only ‘locally Tauberian’ results, in which r → ∞ only through the union of a
well-chosen sequence of large intervals.
One can extend to real zeros. One can use the language of proximate orders,
due to Valiron in 1913, which can be shown to be equivalent to that of regular
variation (and thus that Valiron may be credited with initiating the subject).
The resulting Levin–Pfluger theory (A. Pfluger in 1938, B. Ya. Levin in 1964;
BGT, Ch. 7) may be regarded as weakening the severe geometric restriction
that all zeros lie on one ray, or one line, as far as possible.
The contrasts between key examples throw light on the theory, which they
inspired. To quote BGT (end of §7.6): ‘It is instructive to compare the two
examples sin πz and 1/Γ(z). Their rates of growth differ, as above; their zeros
differ not so much in their density as in their geometry. An extensive study
of the integer-order case has been given by Pfluger (1946), motivated by the
contrasts between these examples.’
As well as the maximum modulus, the minimum modulus of an entire func-
tion is of interest:

M (r) := sup{| f (z)| : |z| ≤ r }, m(r) := inf{| f (z)| : |z| ≤ r }.

One has the cos π ρ theorem (see BGT, §7.7; Bae1974; Ess1975 and the refer-
ences there for details): if f is entire of order ρ ∈ [0, 1),
log m(r)
lim sup ≥ cos π ρ.
log M (r)
Functions extremal here are particularly interesting; see DrasS1969. Here one
encounters exceptional sets, of logarithmic density 0, which cannot be avoided
(Haym1970).

Pólya Peaks. Pólya peaks (of the ‘first and second kinds’) are a device in
real analysis, introduced by Pólya (Poly1923) for the study of entire func-
tions. They were named by Edrei in the 1960s. Their use was extended
to meromorphic functions by Hayman (Haym1964, §4.4); for details, see
DrasS1972. It turns out that they are intimately linked to the Matuszewska
indices (BGT, §2.1) α( f ), β( f ) of regular variation: both kinds of peak ex-
ist in the interval [ β( f ), α( f )] and nowhere else (the Pólya Peak Theorem;
BGT1987, Th. 2.5.2).
In fact, the use of Pólya peaks in the results above (Edrei, Drasin and Shea,
Jordan) may be avoided; see Bingham and Inoue (BinI2000a). This may be
preferred on thematic grounds in complex analysis, as well as to simplify the
proofs.
 Prologue: Regular Variation 7

Recently, essential use of O-regular variation has been made in the the-
ory of ultraholomorphic functions; see JimSS2019 for background and
details.

P.4 Analytic Number Theory

For background on Abelian, Tauberian and Mercerian theorems as mentioned
above, see, e.g., BGT, §4.5; Kor2004.
Tauberian theorems such as the Hardy–Littlewood–Karamata theorem are
extensively used in analytic number theory (see, e.g., Ten2015; BGT, Ch. 4).
So too is e.g. Kohlbecker’s Tauberian Theorem on asymptotics of partitions
(BGT, Th. 4.12.1). Tauberian (and Mercerian) theorems can be used to give
short proofs of the Prime Number Theorem (BGT, §6.2).
The prime divisor functions,
ω(n) := # distinct prime divisors of n,
Ω(n) := # prime divisors of n (counted with multiplicity)
(Ten2015, I.2.2), illustrate our approach well. The classical estimates are (Hardy
and Ramanujan in 1917; Ten2015, I.3.6,7)
1X
ω(n) = log log x + c1 + O(1/ log x) (x → ∞),
x n≤x
with c1 a known constant, and similarly for Ω(n) with a different known constant
c2 > c1 . One also has the classical Erdős–Kac central limit theorem of 1939,
1
|{n ≤ x : ω(n) ≤ log log x+λ log log x}| → Φ(λ) (x → ∞ for all λ ∈ R),
p
x
the beginning of probabilistic number theory, and its refinementpof Berry–
Esseen type, due to Rényi and Turán in 1958, with error term O(1/ log log x)
uniform in λ (Ten2015, III.4.15). Our methods give (BinI2000b)
1 X 1X log λ
ω(n) − ω(n) ∼ (x → ∞ for all λ ∈ R).
λ x n≤λx x n≤x log x

This is a statement of regular-variation type, so it has a representation theorem,

namely
Z x
1X dt
ω(n) = C + (1 + o(1) + o(1/ log x)
x n≤x 2 t log t
(note the two error terms, one inside the integral, one outside). This is not com-
parable to the classical results. There, it is the size of the error terms that counts,
8 Prologue: Regular Variation

but there is no information on behaviour under differencing; here, matters are

reversed. Similarly for results of Mertens (Ten2015, I.1.4; HarW2008, Th. 425,
427) on sums over primes p (BinI2000b),
X log p X
, 1/p.
p ≤x
p p ≤x

P.5 Regular Variation: Preliminaries

The subject of regular variation originates with the Yugoslav mathematician
Jovan Karamata (1902–1967) in 1930 (Kar2009). It concerns limiting relations
of the form
f (λ x)/ f (x) → g(λ) (x → ∞) ∀ λ > 0, (K)
for positive functions f on R+ . Relevant here is the multiplicative group of
positive reals, (R+, ×), with Haar measure dx/x. While this formulation is the
one useful for applications, for theory it is more convenient to pass to the
additive group of reals, (R, +), Haar measure Lebesgue measure dx, where we
write this as
h(u + x) − h(x) → k (u) (x → ∞) ∀ u ∈ R. (K+ )
We can pass at will between these two formulations via the exp/log isomor-
phism. The core of the resulting theory is treated in full in Chapter 1 of BGT,
with further results (e.g. with lim replaced by lim sup – where one may lose
measurability, by ‘character degradation’) in Chapter 2 of BGT.
The limit function g in (K) satisfies the Cauchy functional equation
g(λ µ) = g(λ)g(µ) (λ, µ > 0). (CFE)
For background on functional equations, see AczD1989; the classic context is
Ban1920.

P.6 Topological Regular Variation

Solutions to (CFE), as is typical with functional equations, exhibit a sharp di-
chotomy: they are either very nice (continuous, here) or very nasty (pathological
– unbounded above and below on every interval, or even on any non-meagre
Baire set or non-null measurable set). Since (as we shall see below) such ‘bad’
solutions can be manufactured at will from a Hamel basis (of the reals, as a
vector space over the rationals), we will call this the Hamel pathology.
 Prologue: Regular Variation 9

Under mild regularity conditions (such as measurability, the Baire property,

No Trumps NT, etc.), this gives
g(λ) ≡ λ ρ
for some ρ ∈ R. Then f is called regularly varying with index ρ, f ∈ Rρ .
Functions in class R0 are called slowly varying, written ` (for lente, or langsam).
By taking logarithms, (K) may be written in terms of the limit of the difference
of a function at arguments λ x and x. It turns out that this may be fruitfully
generalized by introducing a denominator ` ∈ R0 :
[ f (λ x) − f (x)]/`(x) → k (λ) (x → ∞) ∀λ > 0 (BK/DH)
(using a denominator in Rρ for ρ , 0 gives nothing new; see, e.g., BGT, §3.2).
This study goes back to Bojanic and Karamata (BojK1963), and independently
to de Haan (Haa1970), whence the name (BK/DH); see BGT, Chapter 3 for a
full account.
There are three key theorems that underlie any form of regular variation (there
are two forms above; more will follow). These are (under mild conditions):
the Uniform Convergence Theorem, UCT: the convergence in (K), (BK/DH)
takes place uniformly on compact λ-sets in R+ ;
the Representation Theorem: giving that ` ∈ R0 if and only if it may be written
in the form
Z x
`(x) = c(x) exp{  (u) du/u} (x ≥ 1) (RT)
1
where
c(x) → c ∈ R+,  (x) → 0 (x → ∞)
(here  (. ) may be taken as smooth as desired, while c(. ) may be as
rough as the mild regularity conditions allow);
the Characterization Theorem: giving the form of g(λ) as λ ρ as above and
that of k in (BK/DH) as
Z λ
k (λ) = chρ (λ), c ∈ R+ ; hρ (λ) := uρ−1 du = (λ ρ −1)/ρ (λ > 0),
1
with the usual ‘l’Hospital convention’ that the right-hand side above
is taken as log λ when ρ = 0.
Even with the simplest functional equation that arises here (the Cauchy),
some mild regularity condition is required. There is a dichotomy: as above
solutions are either very nice (powers λ ρ or the hρ (λ) as above) or very nasty
– pathological (e.g. unbounded above and below on any non-negligible set).
10 Prologue: Regular Variation

Exceptional Sets. There are situations in which the passage to the limit in
slow and regular variation needs to avoid some exceptional set; see BGT, §2.9.
Examples arise in complex analysis: BGT, Th. 7.2.4 (a result of Titchmarsh in
1927), and in work of Drasin and Shea (DrasS1976) on functions extremal for
the cos π ρ theorem of Wiman and Valiron mentioned above. We will need such
exceptional sets below, in dealing with sequential aspects of regular variation.

Thinning (Quantifier Weakening). Another key question, going back to a

conjecture of Karamata, concerns weakening the quantifier, ∀, in (K), (BK/DH):
requiring the convergence to take place for some but not all λ > 0. Rather than
having a continuum of conditions to check, one may be able to reduce this to
finitely many, or even to just two (of course, one could not expect just one to
suffice!). Results of this kind – which one might call thinning results, as they
involve thinning of the λ-set on which convergence is required (cf. BinO2010a)
– go back to Heiberg (Hei1971) and Seneta (Sen1973). Interestingly, given
the side-condition of Heiberg–Seneta type, one no longer needs to impose the
regularity condition needed above to eliminate pathology.
Matters were taken further in Bingham and Goldie (BinGo1982a) and Bing-
ham and Ostaszewski (BinO2018a; BinO2020a): with

g ∗ (λ) := lim sup f (λ x)/ f (x),

x→∞
assume
lim sup g ∗ (λ) ≤ 1.
λ↓1

Then the following are equivalent (for positive f ):

(i) there exists ρ ∈ R such that

f (λ x)/ f (x) → λ ρ (x → ∞) ∀λ > 0;

(ii) g(λ) := limx→∞ f (λ x)/ f (x) exists, finite, for a λ-set of positive measure
[a non-meagre Baire set];
(iii) g(λ) exists, finite, in a λ-set dense in R+ ;
(iv) g(λ) exists, finite, for λ = λ 1, λ 2 with (log λ 1 )/ log λ 2 irrational.

The reader may recognize that Kronecker’s Theorem (HarW2008, Ch. XXIII)
lies behind (iv) here.
There are corresponding thinning results for (BK/DH); see BGT, Th. 3.2.5,
Th. 1.4.3. As remarked there, the result for (K) is no easier, despite its simpler
context. This is because the thinning takes place in the quantifier over λ, which
affects only the numerator in (BK/DH). The effect is that the general ` in the
 Prologue: Regular Variation 11

denominator is no harder to handle than ` ≡ 1, when (BK/DH) reduces to the

logarithmic form of (K).
Remark The proofs of these two thinning results from BGT are the hardest in
that book on the theory of regular variation as such (that of the Drasin–Shea–
Jordan theorem, BGT, §5.2, is harder, but belongs rather to Mercerian theory).
The search for simpler proofs was the motivation behind several of our papers,
on thinning (BinO2010a) and quantifier weakening (BinO2018a; BinO2020a).

Frullani Integrals. The Frullani integral (G. Frullani, 1828; see Ostr1976 for
the history) of a locally integrable function ψ on R+ is the improper integral
Z ∞− Z X
I = I (ψ; a, b) := {ψ(λt) − ψ(t)}dt/t = lim {ψ(λt) − ψ(t)}dt/t.
0+  ↓0,X ↑∞ 

Writing bt = u, we see that I (ψ; a, b) = I (ψ; a/b, 1) = I (ψ, a/b), say. So we

may restrict attention to
Z ∞− Z X
I = I (ψ; λ) := {ψ(λt) − ψ(t)}dt/t = lim {ψ(λt) − ψ(t)}dt/t.
0+  ↓0,X ↑∞ 

Now
Z X Z λX Z λ
{ψ(λt) − ψ(t)}dt/t = ψ(t)dt/t − ψ(t)dt/t.
 X 

So the two limits, concerning behaviour at ∞ and at 0, may be handled sepa-

rately. One obtains (BinGo1982b, §6; BGT Th. 1.6.6):

Theorem For ψ and its Frullani integral I (ψ; λ) as above, the following are
equivalent:

(i) I (ψ; λ) exists for all λ ∈ R+ ;

(ii) I (ψ; λ) exists for λ in a set of positive measure [a non-meagre Baire set];
(iii) I (ψ; λ) exists for λ in a dense set in R+ (or for λ = λ 1, λ 2 with
(log λ 1 )/ log λ 2 irrational), and
Z λx
lim inf lim inf ψ(t)dt/t ≥ 0,
λ↓1 x→∞ x
Z λx
lim inf lim inf ψ(1/t)dt/t ≥ 0.
λ↓1 x→∞ x

Each of (i)–(iii) holds if and only if both of the following finite limits exist for
some (all) σ > 0:
12 Prologue: Regular Variation
Z x
M = M (ψ) = lim σx −σ
uσ ψ(u)du/u,
x→∞ 1
Z x
m = m(ψ) = lim σx −σ
uσ ψ(1/u)du/u.
x→∞ 1

Then the Frullani integral is given by

I (ψ, λ) = (M − m) log λ (λ ∈ R+ ).

This result extends those of Hardy and Littlewood (HarL1924), where σ = 1.

Frullani integrals occur in probability theory, e.g. in the fluctuation theory
of Lévy processes; see, e.g., Bert1996, III.1, p. 73.

Convergence and Cesàro Convergence. The mathematics of the Frullani

integral above yields as a by-product the results below (BinGo1982b, §6),
showing exactly what is needed for a Cesàro convergent function or sequence
to converge. For functions: for φ locally integrable on [0, ∞),
Z x
1
φ(t)dt → c (x → ∞)
x 0

if and only if
Z ∞
φ(x) = a(x) + b(x), where a(x) → c, b(t)dt/t is convergent.
1

For sequences:
n
1X
sk → c (n → ∞)
n 1

if and only if
∞
X
s n = an + bn, where an → c, bn /n is convergent.
1

We include the proof (due to G. E. H. Reuter) as it is so short.

Proof If bn /n converges, (b1 + · · · + bn )/n → 0 by Kronecker’s Lemma.

P
So if s n = an + bn as above, s n → c (C1 ).
Conversely, if s n → c (C1 ), set an+1 := (s1 + · · · + s n )/n. Then s n = an +
n(an+1 −an ), and this is the required decomposition, since if bn := n(an+1 −an ),
bn /n converges.
P

 Prologue: Regular Variation 13

Beurling Slow Variation. For φ : R → R+ Baire or measurable, φ is called

Beurling slowly varying if

φ(x) = o(x) and φ(x + tφ(x))/φ(x) → 1 for all t ∈ R, (x → ∞). (BSV)

This originated in unpublished lecture notes of Beurling in 1957 on his Taube-

rian theorem (below); see Kor2004, IV.11.
If also the convergence in (BSV) is locally uniform in t, φ is called self-
neglecting, written φ ∈ SN. The term and the concept arose in probability
theory; see, e.g., BinO2021b and the references there.
It was shown by Bloom (Blo1976) that for φ continuous, the convergence in
(BSV) is indeed locally uniform. In fact Bloom’s proof needs only the Darboux
property, or intermediate-value property, that φ takes every value between any
two values it attains. For this and other results, see BinO2014; Ost2015a. Here
it is enough to require that φ takes a dense set of values between any two values
attained, but the question of whether a Darboux-like property can be dropped
altogether remains open.
The Representation Theorem gives the Beurling slowly varying φ as those
positive functions of the form
Z x
φ(x) = c(x)  (u)du (x ∈ R),
0

where  is with  (x) → 0 as x → ∞, and c is Baire/measurable with

C∞
c(x) → c ∈ (0, ∞) as x → ∞ (BGT Th. 2.11.3; BinO2014, §9; Ost2015a,
p. 731).

P.7 Tauberian Theorems

We first recall Wiener’s Tauberian Theorem (see, e.g., Har1949, XII; Kor2004,
II):

Theorem P.7.1 (Wiener’s Tauberian Theorem) Suppose K ∈ L 1 (R) with

Fourier transform K̂ non-vanishing on R, and H ∈ L ∞ (R). If
Z Z
K (x − y)H (y) dy → c K (y) dy (x → ∞),

then, for all G ∈ L 1 (R),

Z Z
G(x − y)H (y) dy → c G(y) dy (x → ∞).
14 Prologue: Regular Variation

Here integrals are

R ∞over R and we use additive notation; one may work multi-
plicatively with 0 K (x/y)H (y)dy/y, etc. The Tauberian condition here is of
O-type: H ∈ L ∞ (R), or H = O(1).
Beurling’s Tauberian Theorem generalizes Wiener’s Tauberian Theorem, to
which it reduces in the special case φ ≡ 1:
Theorem P.7.2 (Beurling’s Tauberian Theorem) Suppose K ∈ L 1 (R) with
Fourier transform K̂ non-vanishing on R, with φ Beurling slowly varying and
H ∈ L ∞ (R). If
Z ! Z
x−y
K H (y) dy/φ(x) → c K (y) dy (x → ∞),
φ(x)
then, for all G ∈ L 1 (R),
Z ! Z
x−y
G H (y) dy/φ(x) → c G(y) dy (x → ∞).
φ(x)
Note that the arguments x − y in Theorem P.7.1 involve the additive group
(R, +), and the x/y of its multiplicative version that of the multiplicative group
(R+, ×), while the (x − y)/φ(x) of Theorem P.7.2 involve the ring structure of
R. The two results are thus structurally distinct.
For a short and elegant reduction of Beurling’s Tauberian Theorem to
Wiener’s, see Kor2004, IV, Th. 11.1. For an early use of Beurling’s Taube-
rian Theorem in probability theory, see Bin1981.

The Borel–Tauber Theorem

The two most basic families of summability methods are the Cesàro Cα (α > 0)
and Abel methods A; see, e.g., Har1949, V–VII; Kor2004, I. Perhaps next in im-
portance, though harder, are the Euler E p (p ∈ (0, 1)) and Borel methods; see,
e.g., Har1949, VIII, IX; Kor2004, VI. The Euler and Borel methods (plus those
of Taylor and Meyer–König) belong to the family of circle methods (German:
Kreisverfahrung; see MeyK1949). The name derives from the circle of conver-
gence of a power series; such methods were used for analytic continuation by
power series.
The key Tauberian theorem for the Borel method – ‘Borel–Tauber Theorem’
– is:
Theorem P.7.3 (Borel–Tauber Theorem) For s n := 0n ak : if
P

∞
X
e−x s n x n /n! → s (x → ∞)
0
√
and an = O(1/ n), then
sn → c (n → ∞).
 Prologue: Regular Variation 15

The setting of Theorem P.7.3 is discrete, involving a sum, while that of

Theorems P.7.1 and P.7.2 is continuous, involving an integral. To pass between
them, one may either use ‘Wiener’s second theorem’: see, e.g., Har1949, 12.7,
which demands less of the integrand (so H (y)dy becomes a Stieltjes integral,
dU (y), say) but more of the kernel K; or use an auxiliary approximation
argument. There is much more to be said here, but we must refer for further
detail to, e.g., Kor2004, VI.
√
The Tauberian Condition. Here the Tauberian condition is an = O(1/ n),
and this is best-possible in that no weaker O-condition would suffice here.
But because the weights e−x x n /n! (of course those of the Poisson distribution
P(x) with parameter x) are non-negative, a one-sided Tauberian condition
√ √
suffices: an = O L (1/ n), meaning that nan is bounded below (or with O R
and bounded above). In fact such ‘pointwise’ conditions on the individual an
are not needed, but rather ‘averaged’ forms of them involving differences of the
s n . The classical one is of ‘slow-decrease’ type, due to R. Schmidt in 1925 (see
Kor2004, VI.12):
√ √
lim inf (s m − s n ) ≥ 0 (m, n → ∞, 0 ≤ m − n → 0).
Such one-sided Tauberian conditions are studied at length in Bingham and
Goldie (BinGo1983).

Valiron Methods. For β ∈ (0, 1), write Vβ for the Valiron summability method
(Bin1984b), given by writing
∞ ( )
1 X 1
√ s k exp − (x − k) /x2 2β
→ s (x → ∞)
x β 2π 0 2
as
s n → s. (Vβ )
Our principal concern is with the case β = 21 (see BinT1986). One sees that the
√ is a discrete form of the condition in Theorem P.7.2 with K (x) =
sum above
e−x /2 / 2π. This is the standard normal probability density Φ or N (0, 1), with
2

Fourier transform (characteristic function) exp{− 12 t 2 }, which is non-vanishing

as in Theorems P.7.1 and P.7.2. This K may thus serve as a Wiener kernel. In
the notation of Theorems P.7.1 and P.7.2, one can obtain boundedness of H (. )
from the other conditions; see Har1949, p. 220 for the pointwise Tauberian
√
condition an = O(1/ n) and Har1949, p. 225 for a reference to Vijayaraghan’s
method of monotone minorants for the slow-decrease Tauberian condition. This
allows an easy proof of Theorem P.7.3 from Theorem P.7.2.
16 Prologue: Regular Variation

That the methods V 1 and B are intimately linked has been known since Hardy
2
and Littlewood in 1916 (HarL1916). In probabilistic language, this link reflects
the central limit theorem: the Poisson law P(x) with large parameter x is an
n-fold convolution of P(x/n) with itself, and so approaches normality. Much
more is true: the rapid tail-decay of the Poisson laws allows the use of large-
deviation methods. See Kor2004, VI, Th. 6.1, where the range |n − x| < xγ
occurs, where 1/2 < γ < 2/3. As Korevaar remarks, this parameter range is
the ‘signature’ of large deviations.

Jakimovski and Karamata–Stirling Methods. There are other summability

methods whose weights exhibit central-limit behaviour. We consider indepen-
dent random variables X n , integer-valued (so that the weights will form a
matrix, below), with partial sums Sn = 1n Xk ; write
P

ank := P(Sn = k),

and write A = (ank ) for the summability matrix. The classical case is of
Jakimovski methods (Jak1959; ZelB1970, §70); here the X n are Bernoulli (0–1
valued), with
P(X n = 1) = pn, P(X n = 0) = qn := 1 − pn .
Writing pn = 1/(1 + d n ) (d n ≥ 0), this gives
n n
x + dj
Y ! X
= ank x k ,
j=1
1 + d j k=0

and the Jakimovski method [F, d n ]. The motivating examples are:

(i) the Euler methods, with d n = 1/λ, say written E(λ);
(ii) the Karamata–Stirling methods KS(λ), with d n = (n − 1)/λ. Here
ank = λ k Snk /(λ)n,
with (Snk ) the Stirling numbers of the second kind and
(λ)n := λ(λ + 1) · · · (λ + n − 1).
See Bin1988 for their Tauberian theory and BinS1990 for LLN and LIL
results.
Turning from the non-identically distributed Bernoulli case to the identi-
cally distributed general integer-valued case gives the random-walk methods
(Bin1984a).
All the summability methods considered here are closely enough linked to
be equivalent for bounded sequences (as are Euler and Borel, and indeed as are
Cesàro and Abel).
 Prologue: Regular Variation 17

Riesz Means and Moving Averages. With K as above, taking H (x) = Ha (x) :=
a−1 I[0,a] (x) gives conclusions of the form
1 X
√ sk → s (n → ∞),
a n √
n≤k<n+a n

passing back from integrals to sums as above. These are Riesz means (HarR1915,
IV; Har1949, §4.16, §5.16); ‘typical means’ there and in ChanM1952, or mov-
ing averages in the language of probability and statistics. For more on Riesz
means and Beurling moving averages, see Bin1981; Bin2019. For related mov-
ing averages, see BinG2015; BinO2016a; BinG2017.
The Fourier transform of Ha here is Ĥa (t) = (exp(iat) − 1)/(iat), which has
real zeros, so H cannot be used as a Wiener kernel. But two such Ha with a1 /a2
irrational may be used, as their Fourier transforms have no common zeros (see,
e.g., Wie1933, §10 Th. 6; BinI2000b).
In addition to Riesz means and moving averages, there is a third mode
of convergence relevant here, ‘perturbed Cesàro convergence with rate’. For
β ∈ (0, 1), one has (BinT1986, Th. 3) the equivalence as n → ∞ of
 
sn → σ R(exp n1−β , 1),
1 X
s k → s, for some (all) u > 0,
unβ β
n≤k<n+un
n
1 X  
(s k +  k ) = s + o 1/n1−β for some  n → 0.
n+1 0

The most important case, β = 21 , is in Bin1981, Th. 2 and BinGo1983, Th. 3.

It has distinguished antecedents. That the third statement is sufficient for Borel
(and so Euler) convergence without the  n terms (so is clearly sufficient with
them) is due to Hardy (Har1904, p. 55; Har1949, Th. 149: the first predates
Karamata’s work, the second does not). An approach via regular variation gives
sufficiency of the general result: the relevant Representation Theorem gives the
 n , which plays the role of the error term within the sum or integral there.

P.8 General Regular Variation

One can usefully combine and generalize all three forms of regular varia-
tion (Karamata, Bojanic–Karamata/de Haan, Beurling) encountered above. In
BinO2020a we study general regular variation, in which one has

[ f (x + tφ(x)) − f (x)]/h(x) → K (t) locally uniformly in t. (GRV)

18 Prologue: Regular Variation

Here f is the function under study, φ ∈ BSV and h are auxiliary, and the limit
K is called the kernel. By using the algebraic machinery of Popa groups, one
can substantially reduce the theory to those of the earlier three. In addition,
one encounters a number of functional equations: Cauchy, Gołab–Schinzel,
˛
Chudziak–Jabłońska, Beurling–Goldie, Goldie. See BinO2020a for further de-
tail (and the planned sequel to this book).

Sequential Results: Kendall’s Theorem. As above, regular variation is a

continuous-variable theory, while our preferred tool, the Baire Category The-
orem, is a discrete-variable theorem about sequences. But it has long been
recognized that sequential results are possible and useful; see, e.g., BGT, §1.9.
One finds there reference to early work by Croft (Cro1957), Kingman (Kin1964)
and Kendall (in particular Ken1968, Th. 16):
Theorem (Kendall’s Theorem) If
lim sup x n = ∞, lim sup x n+1 /x n = 1
x→∞ x→∞

and, for some continuous positive functions f and g, interval I = (a, b),
0 < a < b < ∞ and sequence (an ),
an f (λ x n ) → g(λ) ∈ R+ (n → ∞) for all λ ∈ (a, b),
then f varies regularly.
If then f (x) ∼ x ρ `(x), one has (BinO2020b)
−ρ
an ∼ cx n `(x n ).
Because of the importance of Kendall’s Theorem in applications, we should
thus generalize this result as far as possible, in the light of what is now known.
It turns out that one can generalize all three of f , g, I above, but at the cost of
introducing an exceptional set (BGT, §2.9; DrasS1976). For a function f , say
that f (x) has essential limit L = L( f ), finite, as x → ∞,
ess-lim f (. ) = L,
if for all  > 0 there exist X = X (, f ) ∈ R and meagre M = M (, f ) such that
| f (x) − L| <  for all x > X, x < M.
Then (BinO2020b, Th. 2.3) one can weaken continuity of f to being Baire,
continuity of g to being positive, and I an interval to being a non-meagre Baire
set. The weakened conclusion is that
K (s) := ess-limx→∞ f (sλ)/ f (λ)
 Prologue: Regular Variation 19

exists, finite and multiplicative. One calls such f weakly quasi-regularly vary-
ing. If further g is Baire, then (BinO2020b, Th. 2.5)
K (s) ≡ s κ for some κ ∈ R;
one calls such f strongly quasi-regularly varying.
There is also a character-degradation theorem (BinO2020b, Th. 8.1): if k =
k (. , . ) is Borel, then K, where
K (s) := ess-limx→∞ k (s, x)
is of ambiguous analytic class ∆12 (see Chapter 7).

Functional Equations: Hamel Bases. The definition

f (λ x)/ f (x) → g(λ) (x → ∞) for all λ ∈ (0, ∞)
leads immediately to
g(λ µ) = g(λ)g(µ) for all λ, µ ∈ (0, ∞)
(BGT, 1.4.1). This is the Cauchy functional equation, in multiplicative form.
While this is the form preferred for applications, for theory it is better to change
from this multiplicative setting in (R+, ×) to the corresponding additive setting
in (R, +) by writing h(x) := log f (e x ), k (x) := log g(e x ), giving
k (u + v) = k (u) + k (v) for all u, v ∈ R, (CFE)
the Cauchy functional equation on the line. Such functions k are called additive.
From (CFE), one obtains
k (mu) = mk (u), k (u/n) = k (u)/n for all u ∈ R, m ∈ N, n ∈ N \ 0,
so
k (qu) = q k (u) for all u ∈ R, q ∈ Q.
Thus, writing c := k (1),
k (x) = c x for all x ∈ R
if k is continuous, by approximation. So, continuous additive functions are
linear.
One can easily extend this result vastly beyond continuity (BGT1987, 1.1.3).
One obtains (Ostr1929, for measurable k; Meh1964, for the Baire case) that if
an additive function k is bounded above or below on a non-null measurable set
[a non-meagre Baire set], k is linear. Thus, an additive function k is linear or
(highly) pathological.
20 Prologue: Regular Variation

To proceed, we need to invoke the Axiom of Choice, AC, in some form

(e.g. Zorn’s Lemma); that is, to extend the axiom system we work with from
ZF (Zermelo–Fraenkel) to ZFC (i.e. ZF + AC). One can now prove easily that
every vector space has a basis (see, e.g., Jec1973, 2.2.2). Conversely, it was
shown by Blass in 1984 that existence of bases implies AC (Bla1984).
Regarding the real line R as a vector space over the rationals Q as ground
field, R(Q) say, if we work in ZFC this shows (G. Hamel in 1905, Ham1905)
that we have a basis, H say (‘H for Hamel’, below) for R over Q. Of course,
H is uncountable; indeed, it has the power c of the continuum (Kucz1985, Th.
IV.2.3, p. 82).
We may now define, at will, any function g : H → R. This may be extended
uniquely to a homomorphism f : R 7→ R: each x ∈ R may be written uniquely
as a finite linear combination
X
x= αi bi (ci ∈ Q, bi ∈ H).

Then
X
f (x) := αi g(bi ).

If x ∈ H, the above representation of x reduces to x = x, so

f | H = g.

Also, if y ∈ R has the representation

X
y= βi bi,
X
f (y) = βi g(bi ).

Then x + y has the representation i (αi + βi )bi (the range of summation here
P
being the union of those in the two finite sums for x and y), so
X X X
f (x + y) = (αi + βi )g(bi ) = αi g(bi ) + βi g(bi ) = f (x) + f (y) :

f is additive. Were f continuous, we could make it discontinuous by changing

its value at one point. But then by the Ostrowski and Mehdi results, f would
be unbounded above and below on every interval, say. As no such change can
be induced in a continuous function by changing its value at one point, we
conclude that f is already discontinuous. Thus a Hamel basis gives us a way
of manufacturing pathological (discontinuous) additive functions at will. We
call this behaviour the Hamel pathology. Such pathological functions – or,
identifying a function with its graph, functions with graph a Hamel basis in the
plane – are called Hamel functions.
 Prologue: Regular Variation 21

The argument above can be presented for additive functions k : Rd → R

(Kucz1985, V.2); we take d = 1 here for simplicity.
Additive functions thus have the property that even a little regularity forces
great regularity (the form c x). Ostaszewski (Ost2015a, p. 729) lists ways in
which this can happen: additive functions are continuous if they are:
• Baire (Ban1932, I §3 Th.4);
• measurable (Fre1913; Fre1914);
• bounded on a non-null measurable set (Ostr1929);
• bounded on a non-meagre Baire set (Meh1964).
See BinO2011a for details and references.

P.9 Hamel Bases

Despite the pathological behaviour of the Hamel functions above, Hamel bases
as sets may not themselves be pathological. In 1920 Sierpiński (Sie1920)
showed that:
• a Hamel basis H can be (Lebegue-)measurable;
• (Th. I) any measurable Hamel basis has measure 0;
• any Hamel basis has inner measure 0;
• a Hamel basis can be non-measurable.
Thus the classes H1 , H2 of measurable and non-measurable Hamel bases are
both non-empty. Sierpiński also showed (Th. II) that no Hamel base can be an
analytic set – indeed, it cannot even be a Borel set. He ends with a corollary of
his proofs: There exist two measurable sets X, Y ⊆ R such that the set of sums
{x + y : x ∈ X, y ∈ Y } is non-measurable.
Being a Hamel basis is a purely algebraic concept, while we can switch
between the measure and category cases by switching between the density
topology (Chapter 7) and the Euclidean topology. We conclude that the classes
of Hamel bases with and without the Baire property are both non-empty:
• a Hamel basis may or may not have the Baire property, both cases being
possible.
• Sierpiński (Sie1935) also showed this, assuming the Continuum Hypothesis,
CH, for part of it.
F. Burton Jones showed in 1942 that an additive function continuous on a set
T which is analytic and contains a Hamel basis is continuous (Jon1942b); see
also Jon1942b. Kominek proved in 1981 the analogous result with ‘continuous’
22 Prologue: Regular Variation

replaced by ‘bounded’ (Kom1981). Motivated by the analogy between these

two results, the present authors (BinO2010a) gave a result with both the Jones
and Kominek theorems as corollaries, using Choquet’s Capacitability Theorem.
They also deduced Jones’ theorem from Kominek’s and gave another proof of
the Uniform Convergence Theorem for slowly varying functions.
Płotka (Plo2003) showed that every function f : R → Q can be represented
as the pointwise sum of two Hamel functions.
Recall that a perfect set is a non-empty closed set with no isolated points.
A subset A of a Polish space X is called Marczewski measurable if for every
perfect set P ⊆ X either P ∩ A or P \ A contains a perfect set. If every perfect set
P contains a perfect subset which misses A, then A is called Marczewski null.
Marczewski (Mar1935) (writing as E. Szpilrajn) showed that the Marczewski
measurable sets form a σ-field, and the Marczewski null sets form a σ-ideal.
Miller and Popvassilev (MillP2000) show:
• (Th. 10) There exists a Hamel basis H for R which is Marczewski null.
• (Th. 8) There exists a Hamel basis H for R2 which is Marczewski null.
• (Th. 14) There exists a Hamel basis H for R which is Marczewski measurable
and perfectly dense.
Dorais, Filipów and Natkaniec (DorFN2013) show (Th. 4.2) ‘deep differ-
ences between Lebesgue or Baire measurability and Marczewski measurability
by constructing a discontinuous additive function that is Marczewski mea-
surable’. They also show (Ex. 4.1) that there exist additive (discontinuous)
functions that are not Marczewski measurable. For further background, see
Kha2004.

P.10 Scaling and Fechner’s Law

Fechner’s Law (Gustav Fechner (1801–1887) in 1860) may be viewed as stating
that, when two related physically meaningful functions f and g have no natural
scale in which to measure their units, and are reasonably smooth, then their
relationship is given by a power law:
f = cg α . (F)
For background, see, e.g., Bin2015a; Han2004, §5.6.
Fechner’s Law emerges naturally from regular variation, as follows (we re-
strict attention to the basic case, with f , g positive, increasing and unbounded).
They satisfy some unknown functional relationship, say,
f (x) = φ(g(x)) : f = φ ◦ g.
 Prologue: Regular Variation 23

As there is no natural scale, then at least asymptotically this relationship should

be scale-independent regarding x. So changing scale by λ,
f (λ x) ∼ ψ(λ) f (x) for all λ > 0 (RV)
for some function ψ(.) > 0. Under a minimal smoothness assumption (the Baire
property or measurability suffice), f is regularly varying with index α > 0, and
ψ is a power:
[
f ∈ Rα ⊆ R := Rα .
α>0

Similarly from g = φ← ◦ f , g ∈ R, and from ψ = f ◦ g ← with f , g ∈ R, φ ∈ R

also:
φ(λ) = λ α `(λ) ∈ Rα,
with ` ∈ R0 .
The classically important special case is the simplest one, ` constant, ` ≡ c:
φ(x) = cx α ; f (x) = cg(x) α : f = cg α,
giving Fechner’s Law.

Illustrative Example: Athletics Times. For aerobic running below ultra dis-
tances (800 m to the marathon, say), time t and distance d show Fechner
dependence:
t = cd α .
Here c (time per unit distance) reflects the quality of the athlete, while α is
approximately constant between athletes. This is illustrated on a real data set
(the first author’s half-marathon and marathon times) in Bingham and Fry
(BinF2010, §8.2.3).
The statistics needed (regression) extends to the study of ageing also. The
Rule of Thumb for ageing athletes (over 40, say) is: expect to lose a minute a
year on your marathon time through ageing alone. It is well borne out by this
data set (BinF2010, Ex. 1.3, Ex. 9.6).
 1

Preliminaries

1.1 Littlewood’s Three Principles

We shall be dealing extensively with measurable sets and functions, and begin
by recalling Littlewood’s three principles (J.E. Littlewood (1885–1977) from
1944; see Lit1944, §4), according to which a general situation is ‘nearly’ an
easy situation:

(i) any measurable set is nearly a finite union of intervals;

(ii) any measurable function is nearly continuous;
(iii) any convergent sequence of measurable functions is nearly uniformly con-
vergent.

These statements can be made precise, as follows.

Littlewood’s first principle is essentially the regularity of Lebesgue measure.
That is, with |. | denoting Lebesgue measure, for A (Lebesgue) measurable | A|
is the infimum of |U | over open sets U ⊇ A (open supersets of A) and the
supremum of |K | over compact subsets K of A. So one can approximate to
within any  > 0 from without by open sets and from within by compact sets;
taking  = 1/n (with n = 1, 2, . . .), one can find a Gδ set G ⊇ A and a Fσ set
F ⊆ A with |G \ A| = 0 and | A \ F | = 0. (For the place of Gδ and Fσ sets
in the Borel hierarchy, see pages 32 and 51.) As each open set on the line is a
countable disjoint union of open intervals, for each  > 0 one can find a finite
(disjoint) union U of open intervals whose symmetric difference with A has
measure |U∆A| < . See, e.g., Bog2007a, §1.5 or Roy1988, §3.3.
Littlewood’s second principle is essentially Lusin’s Theorem (N.N. Lusin, or
Luzin (1883–1950), in 1912; see, e.g., Bog2007a, Th. 2.2.10 or Hal1950, §55).

Theorem 1.1.1 (Lusin’s Continuity Theorem, or Almost-Continuity Theorem)

Given a regular measure and a finite measure space (e.g. Lebesgue measure on

24
 1.2 Topology: Preliminaries and Notation 25

a compact interval), for f measurable and a.e. finite, f is almost continuous:

for  > 0, there exists a closed set F on which f is continuous and whose
complement has measure |F c | < .

This property of almost continuity in fact characterizes measurability.

Littlewood’s third principle is essentially Egorov’s Theorem (D.F. Egorov
(1869–1931) in 1911; see, e.g., Bog2007a, Th. 2.2.1; Hal1950, Th. 21A).

Theorem 1.1.2 (Egorov’s Theorem) For A measurable of finite measure, and

f n a sequence of measurable functions convergent a.e. to f on A, f n converges
almost uniformly: for each  > 0 there exists B ⊆ A with | A \ B| <  and
f n → f uniformly on B.

Thus almost everywhere convergence (in particular, pointwise convergence)

implies almost uniform convergence.
For textbook accounts of Littlewood’s three principles, see SteS2005, §4.3;
Roy1988, §3.6.
Our standard reference texts for measure theory will be Bogachev (Bog2007a;
Bog2007b) and Fremlin (Fre2000b; Fre2001; Fre2002; Fre2003; Fre2008).

1.2 Topology: Preliminaries and Notation

We gather here a variety of results which we will need later. These are all
known to the experts; others may prefer to return to this for reference as may
be needed.
Our standard references for general topology, as already mentioned in the
Preface, will be Engelking (Eng1989) and also the Handbook of Set-Theoretic
Topology (KunV1984).
In the earlier parts of the book we will, for the most part, work in met-
ric spaces. But we may need to use alternative metrics, for instance to take
advantage of completeness. So it will be convenient to work from the start with
topological spaces (Hausdorff, by assumption), and these may often turn out
to be metrizable. Under such circumstances there may also often be a second
stronger, or as we shall say finer, topology in play (i.e. one with more open sets)
– which we call submetrizable. A helpful analogy is the interplay of weak and
strong topologies in function spaces.
In general we develop topological machinery when needed. But here we
briefly review basic concepts to establish conventions and notation – for details
and proofs of results mentioned below we refer to Eng1989. So below we are
concerned with:
26 Preliminaries

(i) separation properties (regular, completely regular, normal, etc.);

(ii) covering and refinement properties (compactness, countable compactness,
local finiteness and paracompactness, etc.);
(iii) base properties (second countability and σ-local finiteness, etc.);
(iv) neighbourhood base properties (first countability, regular bases).

Notation. Inclusion will be denoted by A ⊆ B, proper inclusion by A ⊂ B,

complement by Ac , symmetric difference by A∆B := ( A \ B) ∪ (B \ A). By Ā
we will denote the closure of A when there is only one topology in play. Oth-
erwise the closure will be written cl T ( A) or just cl T A to imply the relevant
topology T .
By ω := N ∪ {0} we denote the set of finite ordinals.
A subset A of natural numbers ω will often be identified with a real number
in (0,1). Indeed, A is identified uniquely by its indicator function (on the natural
numbers) with 1 A (n) = 1 or 0, according as n ∈ A or not. In turn, 1 A, as a
binary sequence, determines a real number in (0,1) with binary expansion of
that sequence.

Separation Properties. A topological space X is regular if for each neigh-

bourhood U of any point x there is a neighbourhood V with x ∈ V ⊆ V̄ ⊆ U.
Further X is completely regular (or Tychonoff) if for each neighbourhood U of
any point x there is a continuous function f with f (x) = 1 and f = 0 outside
U and it is normal (or Urysohn) if for any disjoint pair of closed sets A, B there
is a continuous function f with f = 1 on A and f = 0 on B.
A space is pseudonormal if matters are as in the last definition but one of the
two closed sets is countable (e.g. a convergent sequence). Thus a pseudonormal
space is completely regular.

Covering and Refinement Properties. A space X is compact if every open

covering, i.e. open family U covering X (family of open sets with union X),
contains a finite subcovering (finite subfamily U 0 covering X). The space is
countably compact if any countable open covering has a finite subcovering.
This is to be contrasted with sequential compactness, which demands that
every sequence hx n i must have a convergent subsequence; such a space is
countably compact (see Eng1989, Th. 3.10.30). A space X is pseudocompact
if every real-valued continuous function is bounded. A countably compact
completely regular space is pseudocompact. Every normal pseudocompact
space is countably compact. In a metrizable space these three concepts are
equivalent.
A space is Lindelöf if any open covering contains (has) a countable subcov-
ering.
 1.2 Topology: Preliminaries and Notation 27

A family F is locally finite if each point x has a neighbourhood U meeting

(intersecting) only finitely many members of F . If each x has a neighbourhood
U meeting at most one member of F , then F is discrete.
The family is σ-locally finite (or σ-discrete) if F = i ∈ω Fi and each Fi is
S
locally finite (resp. discrete). Any open σ-locally finite covering has a locally
finite refinement (by sets not necessarily open) – see, e.g., Eng1989, Lemma
5.1.10.
A family V refines the family U if each member of V is included in a
member of U.
A space is paracompact if every open covering U has a locally finite open
refinement. Examples are Lindelöf (and in particular compact) spaces and
metrizable spaces. By Stone’s Theorem (Eng1989, Th. 4.4.1) every open cover
of a metrizable space has an open refinement that is both locally finite and
σ-discrete.
Every paracompact space is normal. A regular space is paracompact if and
only if every open cover has an open σ-locally finite refinement (Eng1989, Th.
5.1.11).
By analogy with countable compactness, X is countably paracompact if
every countable open covering U has a locally finite open refinement.
Dowker’s Theorem (Eng1989, Th. 5.2.8) asserts that X is normal and count-
ably paracompact if and only if X × [0, 1] is normal. This links normal spaces
to Borsuk’s Homotopy Extension Theorem (see, e.g., Eng1989, §5.5.21 and
note in particular Starbird’s Theorem).

Base Properties. The simplest base property is its countability: a topology is

called second countable (i.e. satisfies the second axiom of countability) if it has
a countable base. The topology of a metric space is second countable if and
only if the space is separable (has a countable dense subset). Generalizations of
this simplest of all topological countability properties include σ-discrete bases
and σ-locally finite bases.
For instance, the Nagata–Smirnov Theorem (Eng1989, Th. 4.4.7) asserts
that a space is metrizable if and only if it is regular and has a σ-locally finite
base. In the same spirit is Bing’s Theorem (Eng1989, Th. 4.4.8) that a space is
metrizable if and only if it is regular and has a σ-discrete base.
A further generalization is provided through regular neighbourhood bases
below.

Neighbourhood Base Properties. A neighbourhood base is an assignment

to each point x in space of a family B(x) of neighbourhoods of x such that
for every neighbourhood U of any point x there is a smaller neighbourhood
28 Preliminaries

B in B(x). The simplest neighbourhood base property is again countability

(for each x): a topology is called first countable (i.e. satisfies the first axiom
of countability) if it has a neighbourhood base assigning to each point x a
countable family B(x) as above.
One obtains a neighbourhood base from a base B by setting B(x) := {B ∈
B : x ∈ B}. This leads to the notion of a base B that is point-regular by
requiring that every point x has a neighbourhood U such that all but finitely
many members of B(x) lie in U (i.e. all but finitely many of those members B
of B that contain x lie in U). This notion motivates a ‘localized’ version.
A base B is regular if for every neighbourhood U of any point x there is
a smaller neighbourhood V such that all but finitely many members B of B
meeting V lie in U. Arhangelskii’s Theorem (Eng1989, Th. 5.4.6) asserts in
particular that a Hausdorff space is metrizable if and only if it has a regular
base.

1.3 Convergence Properties

We shall need various modes of convergence, which we now discuss briefly.
Let hΩ, S, mi be a measure space. For Φ a property of subsets of Ω we write
m{Φ} as an abbreviation for m({ω ∈ Ω : Φ(ω)}). In particular, if m is finite
we can divide by m(Ω) to make m a probability, so without loss of generality
m is a probability if finite. For hX, T i a topological space, recall that in this
latter context a random variable with values in X is an S-measurable map
Y : Ω → X; we write L 0 (Ω, X ) for the set of random variables.
Modes of Convergence. For hX, di a metric space and hYn : n ∈ ωi, a sequence
of random variables with values in X the sequence converges to Y0 :
(i) m-a.e. or almost surely (a.s.) if Yn (ω) → Y0 (ω) almost everywhere;
(ii) in measure/in probability if, for every  > 0, m{d(Yn, Y0 ) >  } → 0 as
n → ∞.
Convergence a.e. implies convergence in measure/in probability (Bog2007a,
2.2.3), but not conversely. The standard example here is constructed from the
subintervals I j := [ j/2n−1 − 1, ( j + 1)/2n−1 − 1] for 2n−1 ≤ j < 2n of [0, 1]. As
these have lengths shrinking to zero, their indicator functions f j (ω) regarded
as random variables on Ω = [0, 1], equipped with Lebesgue measure, converge
to zero in probability. They do not converge almost surely: for any irrational
ω there is an infinite sequence n j (ω) where the n j th function is 1 and an
infinite sequence m j (ω) where the n j th function is 0. So we do not have a.e.
convergence, but do have a.e. convergence along a subsequence. This example
 1.3 Convergence Properties 29

is canonical, as a result below (the ‘subsequence theorem’, due to F. Riesz in

1909) shows.
Thus convergence a.e. (or a.s. in the probability case) is a strong mode of
convergence and convergence in measure (or in probability) a weaker one. This
is reflected in the terminology of the basic limit theorems of probability theory,
the strong law of large numbers (SLLN) and the weak law of large numbers
(WLLN) (see, e.g., Dud1989, Ch. 8). Other strong modes of convergence are
convergence in L p (or in pth mean; we shall only need p = 1 – convergence in
mean – and p = 2 – convergence in mean square). These are not comparable
to convergence a.e. At the other extreme is convergence in distribution, or in
law, as in the central limit theorem (CLT) of probability theory; this is implied
by convergence in measure/probability, but not conversely unless the limit is
constant. See, e.g., Dud1989, Ch. 9.
Convergence in pth mean is metric and generated by the norm of L p . Conver-
gence in measure/probability is metric and generated by the Ky Fan metric, α
say. Convergence in distribution is metric and generated by the Prohorov met-
ric, ρ say. That convergence in probability implies convergence in distribution
follows from ρ ≤ α; see, e.g., Dud1989, Th. 11.3.5. These results also show
that convergence a.e. is metrizable only when it coincides with convergence
in measure. This does not happen in general, as the example below shows –
indeed, in general convergence a.e. is not even topological (see, e.g., Dud1989,
Problem 9.2.2). But it does happen if the measure space is purely atomic, as
then there are no non-trivial null sets; examples such as the one above, which
take place on [0, 1] under the Lebesgue-measurable sets and Lebesgue measure,
do not then apply.

Say that the sequence hYn : n ∈ ωi has the sub-subsequence property relative
to the null sets if for every subsequence hYn(k) i there is a sub-subsequence
hYn(k ( j)) i converging a.e. to Y0 .

Theorem 1.3.1 (Subsequence Theorem; Dud1989, Th. 9.2.1; Bog2007a,

2.2.5(i)) For hYn : n ∈ ωi a sequence of random variables with values in
a separable metric space hX, di, the sequence hYn : n ∈ ωi converges to Y0
in probability if and only if hYn : n ∈ ωi has the sub-subsequence property
relative to the null sets.

Proof Suppose that hYn : n ∈ ωi does not converge to Y0 in probability. Then

for some ε > 0 there is a subsequence n(k) such that

P{d(Yn(k), Y0 ) > ε} > ε, for all k.

30 Preliminaries

Thus for any sub-subsequence hYn(k ( j)) i we have also

P{d(Yn(k ( j)), Y0 ) > ε} > ε, for all j,
and so Yn(k ( j)) (ω) does not converge to Y0 (ω) for a non-null set of ω, i.e.
hYn : n ∈ ωi does not have the sub-subsequence property.
Now suppose that hYn : n ∈ ωi converges in probability to Y0 . Then so does
hYn(k) : k ∈ ωi for any given subsequence n(k). Choose for each j an integer
k ( j) > j such that
P{d(Yn(k ( j)), Y0 ) > 1/ j} < 1/ j 2 .
Hence, by summability of 1/ j 2, the set
\ [
{ω : d(Yn(k ( j)) (ω), Y0 (ω)) > 1/ j}
k j>k

has P-measure 0, i.e.

[ \
{ω : d(Yn(k ( j)) (ω), Y0 (ω)) ≤ 1/ j}
k j>k

has P-measure 1. So for ω in this latter set and any ε > 0 there is some k = k (ω)
such that for j > max{k (ω), 1/ε}
d(Yn(k ( j)) (ω), Y0 (ω)) ≤ 1/ j < ε.
That is, hYn(k ( j)) i converges almost surely to Y0 . 
Definition The sequence hYn i is Cauchy or fundamental in measure if
P{sup d(Yn, Yk ) ≥ ε} → 0 as n → ∞.
k ≥n

Lemma 1.3.2 (Almost Sure Convergence Criterion; Dud1989, 9.2.4; cf.

Bog2007a, 2.2.5 (ii)) If hYn i is a Cauchy/fundamental sequence, then Yn
a.s. converges to some Y0 a.s. (and so also in measure).
Remark Wagner and Wilczyński (WagW2000) proved that the category ver-
sion also holds.

1.4 Miscellaneous
We will use F , G for the families of closed and open sets (‘f for fermé, g for
geöffnet’), H as our usual letter for a family of sets in general, K for the family
of compact sets (‘k for kompakt’). A family H is multiplicative if it is closed
under finite intersection, and so analogously a set-valued map F : H → H is
multiplicative if F ( A ∩ B) = F ( A) ∩ F (B). Dually, say that the family H is
 1.4 Miscellaneous 31

additive if it is closed under finite unions, and so also a set-valued map F is

additive if F ( A ∪ B) = F ( A) ∪ F (B). A family H is maximally additive if
whenever A ∪ B is in H , then at least one of A or B is in H ; this is motivated
by a context in which a family H with some property P may be extended to a
maximal such family, and the extension process permits the inclusion for each
A ∪ B one of A or B, without violating P. We write σ(H ) for the σ-algebra
generated by (the smallest σ-algebra containing) H . Given a σ-algebra, we will
often need to consider a sub-σ-algebra of ‘small sets’ (prototypical examples:
null sets or meagre sets). Such a sub-σ-algebra will be closed under intersection
with any set in the σ-algebra, and so will have the structure of an ideal (using
the viewpoint of Boolean algebra, with intersection as product). We will use
I to denote an ideal generically, qualifying this to distinguish one ideal from
another; we write M for the ideal of meagre sets (see Chapter 2), and N for the
ideal of null sets. Here the measure µ will be Lebesgue measure, n-dimensional
Lebesgue or Haar measure. Given a sequence σ := hσ1, σ2, . . .i, write σ | n
for the first n terms. For H a family of sets, write S(H ) for the class of sets of
the form
[ \∞
H (σ), where H (σ) := H (σ | n),
σ n=1

with each H (σ | n) ∈ H . Here S is the Souslin operation (M. Ya. Souslin, or

Suslin (1894–1919), in 1917), and the sets in S(H ) are the Souslin-H sets (the
term analytic and notation A are also used; see, e.g., Bog2007a, §1.10). The
Souslin operation is important in descriptive set theory, and we shall return to
it later in §3.1. Meanwhile we note the following:

(i) The Souslin operation is idempotent: S(S(H )) = S(H ).

(ii) The class of measurable sets is closed under the Souslin operation (Lusin
and Sierpiński in 1918, Nikodym in 1925, Marczewski (as Szpilrajn) in
1929 and 1933 – see, e.g., RogJ1980, Cor. 2.9.3). We will meet the sets
with the Baire property in Chapter 2; for them we have the dual result,
also due to Marczewski (see, e.g., RogJ1980, Cor. 2.9.4):
(iii) the class of sets with the Baire property is closed under the Souslin
operation.

Evidently S(H ) includes the family of all countable intersections of members

of H , i.e. S(H ) ⊇ Hδ ; with Hausdorff’s δ for Durchschnitt notation, writing
i | n = (i 1, . . . , i n ), and noting that
[ \ [∞ [ \∞ 
H (i 1, . . . , i n ) = H (i 2, . . . , i n ) ,
i∈N N n ii =1 i∈N N n=2
32 Preliminaries

we see also the inclusion of countable unions, i.e. S(H ) ⊇ Hσ, so that
Hσ ⊆ Hσδ ⊆ Hσδσ ⊆ · · · ⊆ S(H ).
Beyond the initial levels displayed, the Borelian-H hierarchy is thus obtained,
by transfinite induction through the countable ordinals, by alternating the σ
and δ operations at successor ordinals and at limit ordinals ‘coalescing’ (taking
unions of) the preceding families. The hierarchy is thus included in S(H ). We
shall refer to the case H = K where K = K (X ) denotes the family of compact
sets in X .
Taking H = G, the open sets of a space (recall ‘G for geöffnet’), and H = F ,
the closed sets (‘F for fermé’), we see that the Borelian-H and Borelian-F
hierarchies are also included in the corresponding S(H ):
Gσ = G ⊆ Gδσ ⊆ Gδσδ ⊆ · · · ⊆ S(G),
Fσ ⊆ Fσδ ⊆ Fσδσ ⊆ · · · ⊆ S(F ) .
Evidently, in a metric space G ⊆ F σ , and similarly F ⊆ G δ . Continuing in
this way, Fσ ⊆ G δσ, Gδ ⊆ F σδ and so on. So taking the union the two hi-
erarchies amalgamate, creating the Borel hierarchy, which is closed under
complementarity. We note, for future comparison, the capitalized Greek nota-
tion, a compromise between the two languages, emanating from mathematical
logic using bold-face Σ for sum and Π for product:
Π00 ⊆ Π01 ⊆ Π02 ⊆ · · ·
Σ00 ⊆ Σ01 ⊆ Σ02 ⊆ · · ·
with Π00 for F and Σ00 for G and later levels beyond these initial ones indexed by
ordinals. The notation stresses the implied reference to the existential quantifier
∃n use of the integers (0-level objects) in the countable union operation over a
sequence of sets and the complementary universal quantifier ∀n in the countable
intersection. The bold-face signals the implied coding of the sequences involved
which are in general unconstructively enumerated. When constructive (more
accurately ‘effectively enumerated’) one drops down to light-face notation.
The sets in S(F ) are the Souslin-F sets; in the context of the closed subsets
of a complete metric space, these are called absolutely analytic (subsets) since
such subsets when embedded in any ‘enveloping’ metric space are Souslin-F
in any enveloping metric space. We address their properties in later chapters.
We use ‘measure’ to mean ‘countably additive measure’. If a measure is
defined on the power set ℘(X ) of all subsets of a countable set X, and van-
ishes on singletons, it vanishes identically by countable additivity (note that a
finitely additive measure may very well vanish on singletons but not identically).
 1.4 Miscellaneous 33

The same statement holds for X of cardinality that of the first uncountable
ordinal (Ulam’s Theorem of 1930 – see Bog2007a, Th. 1.12.40).
A cardinal κ is called non-measurable if whenever a measure is defined on
all subsets of a set of cardinality κ, and vanishes on singletons, it vanishes
identically. Other cardinals are called measurable. Whereas one thinks of mea-
surable sets as being ‘nice’, here it is non-measurable cardinals that are ‘nice’.
For background, see, e.g., Bog2007a; Bog2007b; Fre2008. We will meet non-
measurable cardinality in connection with results of Pol, 2.1.6 and §12.5*, and
in §16.3.
We will study category–measure duality in Chapter 9, and a certain non-
metric topology, the density topology D, in Chapter 7. This is convenient
for our purposes because using it one can bring the category and measure
aspects together. We note here that this can only happen because the topology
is non-metric. For decompositions of metric measure spaces into two parts, one
meagre (small in category) and one null (small in measure), see MarS1949.
See also Oxt1980, Ch. 16.
 2

Baire Category and Related Results

2.1 Baire’s Theorem

In measure theory, as in Chapter 1, one has a natural notion of smallness of a
set A, namely A having measure zero (A null, | A| = 0), and an approximation
to it in having | A| <  for  > 0 small. We now turn to a topological analogue,
which will turn out to be even more important.
As we will work throughout with topological spaces, we abbreviate ‘topolog-
ical space’ to ‘space’ below. A neighbourhood of a point x is a set containing
an open set containing x – that is, a set containing x in its interior. In particular,
neighbourhoods are non-empty. We say that a set A in a space is dense if its
closure A is the whole space, nowhere dense if A has empty interior. In partic-
ular, if A is nowhere dense, then so is A. Following René Baire (1874–1932) in
1899 (Bai1899; Bai1909), we say that A is meagre (or of the first category) if
it is contained in a countable union of nowhere dense sets, non-meagre (or of
the second category) otherwise. Call a set co-meagre (or residual) if its com-
plement is meagre. Note that a countable union of meagre sets is meagre. This
is the topological analogue of the measure-theoretic statement that a countable
union of null sets is null.
We consider in tandem various notions of largeness defined by intersection
with target sets (having members in prescribed target sets). Thus a dense set
A has members in any neighbourhood (i.e. meets any non-empty open set). A
set of positive measure meets (has members in) any co-null set. Likewise a
non-meagre set A meets any co-meagre set.
A first test that a set is large (in whatever sense) is to give a construction
identifying at least one member in some target set. Two related classic construc-
tions are provided by compactness and completeness. When X is (countably)
compact and Fn+1 ⊆ Fn is a nested sequence of non-empty closed sets, their
intersection is non-empty. Likewise, when X is a complete metric space and

34
 2.1 Baire’s Theorem 35

Fn+1 ⊆ Fn is a nested sequence of non-empty closed sets with diameters tending

to 0, their intersection is non-empty, in fact a singleton. (The latter is Cantor’s
Nested Sets Theorem, see below; cf. Mun1975, Lemma 7.30.)
The two results are closely connected and motivate the notion of topological
completeness.
One way to state Cantor’s Theorem (omitting the uniqueness aspect), or
equivalently to assert completeness (see, e.g., Eng1989, 4.3.9), is to identify
a sequence of open families Gn (i.e. families of open sets), e.g. the family of
balls of radius 1/n under a complete metric, with the property that if Un ∈ Gn
and Un+1 ⊆ Un , then
\ \
∅, Un+1 ⊆ Gn .
n n
It is clear from here that a Gδ subspace of a complete metric space satisfies
Cantor’s Theorem. More is in fact true: a Gδ subspace Y of a complete metric
space X may be re-metrized so as to be a complete metric space in its own right.
This is Alexandroff(-Hausdorff) Remetrization Theorem (see, e.g., Oxt1980,
Ch. 12).
Suppose that Gn is a sequence of open families in a compact space X each
covering a subset Y , with
\ [
Y= G n, where G n := Gn .
n
T
Then for Un ∈ Gn and Un+1 ⊆ Un , one has by compactness ∅ , nUn+1 ⊆
n G n = Y . Motivated by this similarity, one says that a space is topologi-
T
cally complete (or Čech-complete) if it is a Gδ subset of some compact space
(e.g. in its Stone–Čech compactification). Thus a Gδ subset of a topologically
complete space is itself topologically complete. Likewise a closed subset of
a topologically complete space is topologically complete (cf. Eng1989, Th.
3.9.6).
The connection between topological completeness and complete metriz-
ability is that a metric space is completely metrizable (i.e. has an equivalent
complete metric) if and only if regarded as a metrizable space it is topologically
complete. (See Eng1989, 4.3.26; cf. Mazurkiewicz’s Theorem, Dug1966.)
Either of these two contexts gives rise directly to a proof of Baire’s Theorem
when stated in the form that the intersection of a sequence of dense open sets
is dense. This may be paraphrased as saying that an intersection of large open
sets is large. The proof in either context proceeds by exhibiting a member of
the intersection within a target set – in this case a neighbourhood, as follows.
It is typical that it is just as easy to find a point in the intersection as finding it
in any neighbourhood. The following pair of statements are together known as
the Baire Category Theorem, or more briefly as Baire’s Theorem, due to Baire
36 Baire Category

(in 1899, Bai1899). It is amazing to note how conceptually similar his proof is
to Cantor’s first proof (in 1874) that the reals are uncountable (see Kan2009).

Theorem 2.1.1 (Baire’s Theorem; Kec1995, 8B; Eng1989, 3.9.3; Kel1955,

6.34) In a topologically complete space, in particular in a locally compact
Hausdorff space or a completely metrizable space, the intersection of a count-
able sequence of dense open sets is dense.

Proof Let Vn be dense and open and U any neighbourhood. As V1 is dense

and open it meets U, so by regularity there is a neighbourhood U1 with U1 ⊆
U ∩ V 1 . Continuing inductively, choose non-empty open sets Un ⊆ U such that
Un+1 ⊆ Un ∩ V n (the latter is possible because, as before, Vn is open and has
T T
points in common with Un by density). Then ∅ , nUn+1 ⊆ U ∩ n Vn . 

An important corollary is its equivalent re-statement, dualized to the closed

sets.

Theorem 2.1.2 (Baire’s Theorem – Variant) If a topologically complete space,

in particular a locally compact Hausdorff space or a completely metrizable
space, is a union of closed sets Fn , then one of the sets Fn has non-empty
interior.

Proof Suppose otherwise. Then each Fn has as complement a dense open set
G n , and any point common to these open sets is not covered by the Fn . 

One may check that the following three properties of a topological space
X, each of which may be thought of as giving a sense in which X resembles
familiar spaces such as Euclidean space, are equivalent:

(i) each non-empty open set in X is non-meagre;

(ii) each co-meagre set in X is dense;
(iii) the intersection of countably many dense open sets in X is dense.

One calls such a topological space a Baire space. We note in passing that a Gδ
(in particular, an open) subset of a Baire space is Baire. So:

(a) A completely metrizable space is Baire.

(b) A locally compact Hausdorff space is Baire.

In (b) above, the point of the (Hausdorff) space being locally compact is that it
has a one-point (Alexandroff) compactification.

(a) Any topologically complete space is Baire.

 2.1 Baire’s Theorem 37

For background on Baire spaces, see, e.g., AarL1974; HawoM1977.

We think of meagre sets as small in the category (or topological) sense, just
as null sets are small in the measure sense. The two concepts of smallness are,
however, quite distinct. The real line, a prototypical big set, may be written as
the disjoint union of a null set and a meagre set, each small in one way (but big
in the other); see, e.g., Oxt1980, Th. 1.6.
Baire’s Theorem is a powerful tool in producing non-constructive existence
proofs. One shows that things exist not by exhibiting examples but by showing
that they exist generically. This is a topological counterpart to the probabilistic
method, for which see AloS2008; Spe1994.
A set A is said to have the property of Baire if it is the symmetric difference
of an open set with a meagre set: A = G∆M with G open and M meagre. We
shall abbreviate this by saying simply that A is Baire. It can easily be shown that
the Baire sets are also the sets A = F∆M with F closed and M meagre. The
class of Baire sets is closed under complements, and under disjoint unions. It is
a σ-algebra – the smallest σ-algebra containing the open sets and the meagre
sets (and so, is the σ-algebra generated by the Borel σ-algebra and the meagre
sets). The Baire sets are those of the form U ∪ M1 with U ∈ Fσ and M1 meagre,
and those of the form V \ M2 with V ∈ Gδ and M2 meagre; see, e.g., Oxt1980,
Ch. 4. In particular, open, closed, Fσ and Gδ sets are Baire. One sees that the
Baire sets are analogous to the measurable sets, in view of Littlewood’s first
principle (§1.1).
We recall from §1.4 that Souslin-F sets, and in particular Borel sets, are
Baire (by Nikodym’s Theorem; see §1.4). A set S in a space X may have the
Baire property in X, but its restriction to a subspace Y, namely S ∩ Y, does
not necessarily have that property in Y . Indeed, although open-ness is retained
under restriction, the same is not the case with ‘nowhere denseness’: if N
is nowhere dense in X, then, for any neighbourhood U meeting N, there is
a non-empty sub-neighbourhood V missing N, and, passing to the subspace
Y = N itself, we no longer have V ∩ Y non-empty. Nevertheless, some sets
S do preserve the Baire property under restriction, and in so doing exhibit a
stronger form of the Baire property. Such sets have come to be called restricted
Baire, an unfortunate term as the more appropriate ‘hereditarily Baire’ has
taken on the meaning of ‘F -hereditarily Baire’. Following Sto1963, one might
also speak of the completely hereditarily Baire sets. Fortunately, if S is a Borel
set in X, then S ∩ Y is a Borel set in Y ; but Borel sets are generated from
open sets (which are Baire) by operations which preserve the Baire property,
and so Borel sets preserve the Baire property under restriction. More gener-
ally, this is the case with the larger family of Souslin-F sets (for definition
see §1.4).
38 Baire Category

Theorem 2.1.3 (Kur1966, §11.VI and VII) Souslin-F sets, and in particular
Borel sets, preserve the Baire property under restriction, i.e. inherit the Baire
property on arbitrary subspaces.

A function f is said to have the property of Baire, or more briefly (as above)
to be Baire, if the inverse image under f of any open set is Baire. One sees
that the Baire functions are the category (or topological) analogues of the
measurable functions of §1.1.
The alternative usage ‘Baire measurable’ refers to the smallest class of func-
tions including the continuous functions and closed under pointwise limits. We
will study the relation between this concept and that considered here in Chapter
12 on Continuity and Coincidence Theorems.
One reason why Baire functions are of interest is the following Continuity
Theorem, a category analogue of Lusin’s Continuity Theorem. As usual, we
write f | C for the function f restricted to a subset C of its domain.

Theorem 2.1.4 (Baire’s Continuity Theorem; Kec1995, Th. 8.38) If f : X →

Y is Baire-measurable with Y second countable, then f is continuous off a
meagre set, i.e. for some co-meagre subset C the restriction f | C is continuous.

Proof Let BY be a countable base for the topology of Y . For B ∈ BY , as f −1 (B)

is Baire, we may choose U = UB open in X such that NB := UB 4 f −1 (B) is
meagre; so UB \NB ⊂ f −1 (B). Then C := X\ B NB is co-meagre and f | C
S
is continuous on C. Indeed, for c ∈ C and f (c) ∈ B ∈ BY we have c ∈ UB (for
otherwise c ∈ f −1 (B)\UB , and so c ∈ UB \NB ⊂ f −1 (B)). 

As an illustration of how countability conditions in topology yield general-

izations of classical theorems beyond their original separable metric context,
we now give a generalization of Baire’s Continuity Theorem. We need the fol-
lowing definition, which extends the usage of §1.2 (cf. Bing’s Theorem cited
there).

Definition A map f : X → Y is σ-discrete if there is a σ-discrete collection

B f ⊆ ℘(X ), called a σ-discrete base for f , such that, for each open V in Y ,
there is B f (V ) ⊆ B f with f −1 (V ) = B f (V ).
S

Remarks 1. For X or Y metrizable a continuous map f : X → Y is σ-discrete.

Indeed, for X metrizable it is enough to take B f to be any σ-discrete base for
the open sets of X. On the other hand, for Y metrizable, and BY any σ-discrete
base for the open sets of Y , the set B f := { f −1 (V ) : V ∈ BY } is also σ-discrete.
2. If Y is second countable with countable base BY , then B f := { f −1 (B) : B ∈
BY } is countable, so σ-discrete, and of course B f (V ) := { f −1 (B) : B ⊆ V }.
 2.1 Baire’s Theorem 39

To motivate Hansell’s Theorem below, we note that if f | ( A\M) is contin-

uous and V is open in Y, then f −1 (V ) is an open set modulo some subset of M.

Theorem 2.1.5 (Hanse1971, §3.7, Th. 9) For f : X → Y a σ-discrete map

on a metrizable space, f has the Baire property if and only if there is a meagre
set M in X such that f | (X\M) is continuous.

A function may have the stronger property that inverse images of open sets
have the (stronger) restricted Baire property. When this is so the function itself
is said to have the restricted Baire property. The following is a characterization
of such functions in a fairly broad category of spaces. See §12.5.

Theorem 2.1.6 (Pol’s Theorem; Pol1976) A function f from a compact space

X to a metric space Y of non-measurable cardinality has the restricted Baire
property if and only if for every subspace A of X there is a subset M ⊆ A
meagre in A such that f | ( A\M) is continuous.

This yields a far-reaching consequence for functions that are known to have
the restricted Baire property. Below, X, Y are as in Pol’s Theorem, 2.1.6.

Corollary 2.1.7 For X a Borel subspace in some compactification, Y a metric

space of non-measurable cardinality, and f a function from X to Y with a
Souslin-F (X ×Y ) graph, there is a meagre subset M in X such that f | (X\M)
is continuous.

Proof Let f have Souslin-F (X × Y ) graph G. As X is Souslin-F (K ), it is

analytic. Hence the image set f (X ), being the projection of G ∩ (X × Y ), is
analytic. Likewise f −1 (U) is the projection of X × (U ∩ f (X )) ∩ G and so
Souslin-F (X ). So f −1 (U) inherits the Baire property on arbitrary subspaces.
That is, f has the restricted Baire property. The conclusion now follows from
Pol’s Theorem, 2.1.6. 

Remark It was previously shown by Hansell (Hanse1971, §3.7, Cor. 8) that

in the case of X metrizable and Borel in some compactification (i.e. absolutely
Borel), this conclusion holds without any cardinality restriction on Y .
Recall that B is a pseudo-base if every non-empty open set contains a member
of B; such a collection B is locally countable if each member of B contains
only countably many members of B. The product of two Baire spaces, one
of which has a locally countable pseudo-base, is Baire. An arbitrary product
of Baire spaces, each of which has a locally countable pseudo-base, is Baire
(Oxt1960, Th. 3). Without the countability assumption, the product need not
be Baire (Oxt1960, §4). See also §9.7.
40 Baire Category

Remark We say that a topological space satisfies Blumberg’s Continuity Theo-

rem (or has Blumberg’s property, or is Blumberg) if for any real-valued function
on X there is a dense set D such that f | D is continuous. Note that Blumberg
implies Baire (see Bradford and Goffman 1961, BraG1960). The converse is
true in a metric space, or more generally a space with a σ-disjoint pseudo-base
(as defined above). See Whi1974.
We have noted that the property of being a Baire space is hereditary for the
open subspaces of a Baire space X; the following is a generalization to a less
restricted class of subspaces (as it applies to all dense subspaces).

Theorem 2.1.8 (ComN1965, L. 1.2) If X is Baire and A ⊆ X meets each non-

empty Gδ of X, then A is a Baire subspace, so in particular A is non-meagre.

Proof Let G be open in A. For a sequence of sets Un that are dense and open
in A, choose W and Vn open in X such that Un = Vn ∩ A and G = W ∩ A.
T
Then Vn is dense in A and so, as X is Baire, W ∩ n Vn is a non-empty Gδ . By
T
assumption, the latter set meets A and so G ∩ n Un is non-empty. 

Theorem 2.1.9 (ComN1965, L. 1.3) For an arbitrary index set I, if each Xi

is Baire and each Ai ⊆ Xi meets each non-empty Gδ of Xi , then the (Tychonoff)
Q Q
product space i Ai meets every non-empty Gδ subset of the product i Xi .
So in particular the product is Baire.

Proof For Vn open in i Xi, write

Q
Y \
Vn = Vi,n,m, and V := Vn,
i n

with each Vi,n,m open in Xi and with Vi,n,m = Xi for cofinitely many m. Put
\
Vi := Vi,n,m .
n,m

Then Vi is a Gδ in Xi, and so meets Ai, in ai, say. Then a = hai i ∈ Ai ∩V .

Q
i 

We close with an extension of Baire’s Theorem obtained by weakening its

assumption of completeness and so its reliance on open sets in a useful way. We
recall from §1.2 that a topological space X is regular if for each neighbourhood
U of any point there is a neighbourhood V of that point with V̄ ⊆ U; further,
X is completely regular if for each neighbourhood U of any point x there is a
continuous function f with f (x) = 1 and f = 0 outside U.
In the definition below, we have a sequence hBn i of pseudo-bases, which can
be thought of as playing the role of ‘shrinking diameters’.

Definition (Oxt1960, Section 5) A space is pseudocomplete if

 2.1 Baire’s Theorem 41

(i) there is a sequence of pseudo-bases Bn such that every nested sequence of

sets Bn ∈ Bn with B̄n+1 ⊆ Bn has non-empty intersection;
(ii) the space is quasi-regular, i.e. every neighbourhood contains the closure of
some neighbourhood.

Thus a pseudocomplete space is Baire. Furthermore, an arbitrary product of

pseudocomplete spaces is pseudocomplete.
In this connection recall from §1.2 that a completely regular space is pseudo-
compact if continuous functions are bounded. Significant here is the following
equivalent definition in terms of nested sequences of open sets.

Theorem 2.1.10 (GilJ1960, Lemma 9.13) A completely regular space is

T
pseudocompact if and only if every nested sequence of open sets Vn has n V̄n
non-empty.
T
Proof Suppose X is pseudocompact, but n V̄n is empty for some nested
sequence of open sets Vn . Pick distinct points x n and open sets Wn with x n ∈
Wn ⊆ W̄n ⊆ Vn ; then {x n : n ∈ ω} is discrete and also closed, since any
T
limit point would be in n V̄n . In fact more is true: every point x of X has a
neighbourhood that meets at most finitely many sets Wn ; otherwise, x would
be in n V̄n . Now let f n be continuous with f n (x n ) = n and f n = 0 on X\Wn .
T
Then f (x) = f n (x) is a series with at most one non-zero term and for any
P
x the function f coincides with some gn on a neighbourhood of x. So f is
continuous and unbounded, contradicting pseudocompactness.
For the converse, note that any unbounded continuous function f yields the
nested sequence of open level sets Vn := {x : | f (x)| > n} for which n V̄n is
T
empty. 

We note that completely regular countably compact spaces are pseudocom-

pact; conversely, normal pseudocompact spaces are countably compact (see,
e.g., Eng1989, 3.10).
One thus has

Theorem 2.1.11 (ComN1965, Cor. 2.8) Any product of pseudocompact spaces,

and so in particular of countably compact spaces, is Baire.

Proof A pseudocompact space is pseudocomplete – for Bn take the non-empty

open sets and refer to the preceding theorem. 

We obtain better stability behaviour in §2.3 by introducing somewhat stronger

properties which lie between Baire and complete.
42 Baire Category

2.2 Banach’s Category Theorem

Banach’s Category Theorem (Oxt1980, Ch. 16) or Banach’s Union Theo-
rem (Kur1966, §10III) is also known as Banach’s Localization Principle. See
Kel1955, Th. 6.35; RogJ1980. This is also a result we constantly rely on through-
out the book. The proof below illustrates how an elementary proof, applicable
in the second-countable case (outlined in the Remark below), may easily be
lifted to the more general metric setting by the use of σ-discrete open cover-
ings (for which see §1.2). That notion was not yet available when Banach based
his ingenious proof on an open disjoint almost covering (i.e. omitting only a
nowhere dense set). We need some definitions.

Definitions Say that T is locally meagre at a point s (not necessarily in T) if

there is a neighbourhood Us of s with T ∩ Us meagre, and write L M (T ) for the
set of points at which T is locally meagre.
Say that T is everywhere non-meagre (in T) if there are no points of T where
T is locally meagre; then for each neighbourhood Ut of t ∈ T the set Ut ∩ T is
non-meagre.
Recall that s is an essential limit point of T if for each neighbourhood Us of
s the set T ∩ Us is non-meagre.

Theorem 2.2.1 (Banach’s Category Theorem, or Localization Principle) If a

topological space S is locally meagre, then S is meagre. That is, a union of any
family of meagre open sets is meagre.

Proof For U any maximal disjoint family of open sets U with U ∩ S meagre,
put US = U. By maximality, S\ cl US is empty. (Otherwise, U ∩ S is meagre
S
for some U ⊆ X\ cl US .) It suffices to show S ∩ US is meagre, as (cl US )\US is
nowhere dense.
For U ∈ U choose Nn (U) nowhere dense so that
[
U∩S= Nn (U).
n∈ω

Put
[
Nn := {Nn (U) : U ∈ U },

which is nowhere dense, because U disjoint implies for U ∈ U that

[
Nn ∩ U = Nn (U), and so S ∩ US = Nn .
n∈ω

So, if an open V meets Nn, then V meets Nn (U) for some unique U ∈ U,
and so V ∩ U contains a non-empty subset W disjoint from Nn (U). Then W is
disjoint from Nn , as W meeting Nn gives that V meets Nn, and so meets Nn (U),
 2.3 Countability Conditions – Topological 43

implying that W meets Nn (U). This is a contradiction, since W is disjoint from

Nn (U). 

Theorem 2.2.2 (Banach’s Category Theorem – Variant) In a metrizable

space, for T a Baire set, the set L M (T ) of points at which T is locally meagre
is itself meagre. So quasi all points of T are essential limit points of T, and if T
is non-meagre, then T\L M (T ) is everywhere non-meagre.

Proof By Bing’s Theorem (§1.2), select a σ-discrete base for the open sets,
say B = i ∈ω Bi with each Bi discrete. For L = L M (T ) and each point s ∈ L
S
choose an open set Us ∈ B with T ∩Us meagre. Then L i := {T ∩Us : Us ∈ Bi } is
covered by the discrete family {Us ∈ Bi }. For fixed i, put S(i) := {s : Us ∈ Bi }
and, for s ∈ S(i), write T ∩ Us := m Nms,i with Nms,i nowhere dense.
S
i := S s,i
We now show that each set Nm s ∈S(i) Nm is nowhere dense, so that
L = i ∈ω m∈ω Nm is meagre.
i
S S
Fixing m, by discreteness, each x has a neighbourhood Wx meeting at most
one set Us with s ∈ S(i). If W meets no such set, then W avoids L i, so assume
p,i pi
now that it meets such a set, say Up . Since Nm ⊆ Up and Nm is nowhere
pi
dense, there is a neighbourhood W ⊆ W with W avoiding Nm . Then W 0
0 0
pi
avoids all the sets Nmsi for s , p in S(i), because W does so, and it avoids Nm
by construction. So W 0 avoids Nm i , so N i is nowhere dense.
m
For the rest: if now T is non-meagre, then S := T\L M (T ) is non-meagre
and each point of S is an essential limit point of T . So quasi all points of
T are essential limit points of T, as asserted. Furthermore, L M (S) = ∅: for
if t ∈ L M (S), then for some Ut the set Ut ∩ S is meagre, and so also is
Ut ∩ (T ∪ L(T )) ⊇ Ut ∩ T, i.e. t ∈ L M (T ), contradicting that S is disjoint from
L M (T ). 

Remark Were the space second countable with countable base B, one would
argue that, since for each s there is Us ∈ B with T ∩ Us meagre, the countable
family {T ∩ U : T ∩ U is meagre and U ∈ B} covers L, so has meagre union,
and so L is meagre. The proof above embellishes this ‘by localization’.
We next pass to other notions of countability that yield results of the Baire
Category Theorem type.

2.3 Countability Conditions – Topological

Countability is present rather obviously in measure theory via σ-additivity, and
this is the basis of the measure–category duality in Oxt1980. We extended the
scope of category methods in the last section, replacing metric completeness
44 Baire Category

with compactness (to obtain topological, or Čech, completeness). Here we im-

prove on the generalization by weakening the role of compactness. By contrast,
the next section brings to bear the mechanics of game theory. Our aim is to
extend the scope of category–measure duality.
The over-riding concern is to relax all countability aspects of metrizability
as far as possible. There are two modes of thought:

(i) distill local properties shared by all metric spaces into bases, or into types
of covering refinements, such as point-finiteness or local finiteness (see,
e.g., Eng1989, §4.4) – this we delay;
(ii) impose global structures less rigid than a metric – this we now do.

Recall from the last section that we were guided by the sequence of open
covers hGn i comprising balls of radius shrinking to zero with n, say 2−n . They
form the core of metrizability theory; see Gru1984 for a wide overview of
different structures motivated by covers hGn i as a vehicle of countability in a
topological space X.
We start with a particularly simple condition: the requirement that the diag-
onal set in X 2 be a Gδ -set, i.e. that {(x, x) : x ∈ X } = n Vn with Vn open in
T
X 2 . This is reminiscent of ‘uniformities qua neighbourhoods of the diagonal’
(cf. Eng1989), but the pertinent features which this yields are:

(i) the sequence of open covers hGn i defined by Gn := {G open : G×G ⊆ Vn },

known as the Gδ -sequence; and
(ii) their ‘stars’ defined by
[
st(x, Gn ) := {G : x ∈ G ∈ Gn }.

For an illustration of the power of this structure, we mention Chaber’s Theorem

(see Gru1984); a countably compact space with a Gδ -diagonal is metrizable.
See also Ost1980.
Our next idea introduces a very weak form of compactness: this is the notion
of an M-space (M for Morita; again see Gru1984). One demands the existence
of open covers hGn i with:

(i) clustering: if a sequence of points hx n i satisfies x n ∈ st(x, Gn ) for each n,

then the set {x n : n ∈ ω} has a cluster point; and
(ii) star-refinement: {st(x, Gn+1 ) : x ∈ X } refines Gn for each n.

As an illustration, we cite Morita’s result that a space is metrizable if and

only if it is an M-space and has a Gδ -diagonal. For an application of star-
refinement, we recall first that a topology τ on X is called submetrizable
if there is a coarser metrizable topology τ 0 on X (i.e. τ 0 ⊂ τ): in brief, τ
refines some metrizable topology. (Thus the density topology on R, to be
 2.4 Games 45

defined in Chapter 7, is submetrizable since it refines the usual Euclidean

metric topology.) A connection between the new ideas established so far is
given by the following theorem.
Theorem 2.3.1 (See, e.g., Gru1984, Th. 2.5) A space X is submetrizable if
and only if X has a Gδ -sequence such that {st(x, Gn+1 ) : x ∈ X } refines Gn for
each n.

2.4 Countability Conditions – Games of Banach–Mazur Type

Under what circumstances will a nested sequence have non-empty intersec-
tion? This is an existence question (on the existence of a common element).
So far we have two answers: for closed sets completeness, or compactness;
for open sets Baire’s Theorem identifies density. We now introduce general-
izations which yield non-empty intersections, but need to be formulated in the
language of game theory. These existence assertions concern the existence of
a winning strategy, or more accurately the absence of a winning strategy of the
counterparty (see the comments by Jean Saint-Raymond in Sai1983). The idea
of defining classes of topological spaces via games was initiated by Choquet
(see Cho1969, p. 116), but the approach goes back to Banach and Mazur, who
worked in a simpler context: the real line. They formulated their game as gener-
ating a decimal expansion of a real number. This idea proved to be potent in two
ways. As stated, it eventually revolutionized our understanding of ‘definable
sets of reals’ (see §16.1). Restated as generating a nested sequence of intervals,
it gave topologists novel ways to introduce useful countability conditions into
topology.
The games considered below are related and so have a similar structure. All
have stages labelled by the natural numbers and all have two players I and II
(also referred to, in opposite order, as α and β, and sometimes He and She,
a practice we occasionally follow below). Their actions/choices, taken at the
various stages, generate in particular a nested sequence of open sets. Each
game has a rule specifying the winning player as a function of the actions and
referring to the intersection of the nested sequence. A strategy for the player i is
a function which assigns at each stage n a choice for player i given the actions
taken so far. (So this is a game of perfect information, like chess.)
Interest focusses on characterizing when a particular player, say i, has a
winning strategy, i.e. a strategy under which i wins no matter what the choices
of the other party are.
For the Banach–Mazur game in a space X with target A Players I and II
alternately choose nested non-empty open sets Un and Vn respectively at stage
n = 0, 1, 2, . . . , with Player I choosing first. Thus
46 Baire Category

U0 ⊇ V1 ⊇ U1 ⊇ · · · Vn ⊇ Un+1 ⊇ Vn+1 ⊇ · · · .

Player II wins the game if

\
Vn ⊆ A.
n
T
Otherwise I wins (i.e. n Vn meets the complement of A). In particular, II
wins if the intersection is empty. The fundamental theorem in this context, due
to Banach and Mazur, is that a set A is co-meagre in X if and only if II has
a winning strategy (see Oxt1980, Ch. 6). Mazur proposed the game (this is
Problem 43 in the Scottish Book; Mau1981), and Banach responded in 1935
by demonstrating its determinacy for target sets A having the Baire property.
Further development is needed to characterize a winning strategy for I.
The special case with A = ∅ is known as the Choquet game J (X ) (‘J for
jeu’). So here the choices are exactly as in the Banach–Mazur game, and now
Player II wins if
\ \
Un = ∅, i.e. Vn = ∅,
n n

as in the special case above. Here I is called β.

Theorem 2.4.1 (Oxt1980) The space X is a Baire space if and only if X is

β-unfavourable (i.e. I has no winning strategy) in the Choquet game J (X ).

The theorem is mute about whether Player II has a winning strategy. The
space is said to be a Choquet space (or an α-favourable space) if in fact Player II
(Player α) has a winning strategy.

Theorem 2.4.2 (Kec1995, 8.11) The space X is Choquet if and only if X is

co-meagre in its completion.

The choices in the strong Choquet game are as in the Banach–Mazur game
with the additional stipulation that at each stage n, Player I chooses also a point
x n ∈ Un and Player II’s choice is constrained so that

x n ∈ Vn .

Here Player II wins if

\ \
Vn , ∅, i.e. Un , ∅.
n n

The space is called strongly Choquet if Player II has a winning strategy in

the strong Choquet game. (Note that if the game is played without recall of the
previous moves and Player II has a winning strategy, then instead one speaks
of strongly α-favourable spaces.)
 2.4 Games 47

Theorem 2.4.3 (Kec1995, 8.17(i)) The space X is strongly Choquet if and

only if X is a Gδ in its completion, i.e. X is completely metrizable.

Theorem 2.4.4 (Kec1995, 8.33(ii)) For the Banach–Mazur game on a Cho-

quet space which is metrizable, Player I has a winning strategy if and only if A
is meagre in the neighbourhood of some point.

The Christensen games, for which see Chr1974, used to great effect by J.
Saint-Raymond in Sai1983, generalize the Choquet game, bringing the latter
closer to the original Banach–Mazur game by allowing Player II (at least in the
limit) to select a target set. In the points version, Jpt (X ), Player II at stage n
selects a point x n . Now Player I wins if
\
Un ∩ {x n : n ∈ ω} , ∅.
n

Two further generalizations arise from replacing point selection by set selec-
tion. In one version, Jcpt , Player II creates an increasing sequence of compact
sets by selecting at stage n a compact set Kn with Kn ⊇ Kn−1 . Now Player I
wins if
\ [
Un ∩ {Kn : n ∈ ω} , ∅.
n n

In an alternative version, denoted Janalytic, an increasing sequence of K-analytic

sets An (see §3.1.1 for a definition) replaces the compact sets with II selecting
at stage n an analytic set An with An ⊇ An−1 . This time Player I wins if
\ [
Un ∩ { An : n ∈ ω} , ∅.
n n

One may compare various games on X. Following Deb1986, we say that J2

is a stronger game than J1, or is less favourable to α than J1, and write J1 ≺ J2,
provided that for each space X

(i) if J2 (X ) is α-favourable, then so too is J1 (X ); and

(ii) if J2 (X ) is β-unfavourable, then so too is J1 (X ).

Thus
J ≺ Janalytic ≺ Jcpt ≺ Jpt .

Note that for a given X the game Ji (X ) need not be determined; however,
if X is regular and K-analytic, then the games JP (X ) for P = pt, cpt, analytic
are determined. One area of application is concerned with the issue of when
separate continuity implies joint continuity (see Chapter 12). For the time being
we content ourselves with indicating that the game-theoretic approach properly
extends the category of ‘Baire-like’ spaces.
48 Baire Category

We need a notion of point separation as follows. When, as with Ba(X ), the

family of Baire sets of the space X, the family H is closed under complemen-
tation, then the separation is symmetric.

Definition For A, B ⊆ X, say that A is countably distinguished from B by a

family of sets H if there is a countable subfamily H 0 such that for every pair
of points a ∈ A, b ∈ B there is H ∈ H 0 with a ∈ H and b < H.

Theorem 2.4.5 (Deb1987, Th. 3) For a compact space X and subspaces

A ⊆ A0 with A dense in X and Baire, if A is countably distinguished from X\A0
by Ba(X ), then A0 is β-unfavourable for the game Jpt .

We refer to the preceding subsection for the notion of complete regularity.

Theorem 2.4.6 (Deb1987, Cor. 7) For X completely regular, if X is K-

analytic, then Jpt (X ) is α-favourable.

In fact a weaker hypothesis, that instead X be countably determined (i.e. is

the image of a compact-valued upper semi-continuous image of a metrizable
space), renders Jpt (X ) a β-unfavourable game, for which see again Deb1987.
We close this section by noting that products of Choquet spaces are Choquet
and open subspaces of Choquet spaces are Choquet. Strongly Choquet spaces
are Choquet. Products of strongly Choquet spaces are strongly Choquet, and
their non-empty Gδ subspaces are strongly Choquet. (For these, see
Kec1995, §8.)

2.5 Notes
1. Plumed Spaces.
Intermediate between the two notions of §2.3 in the use of compactness is
Arhangelskii’s notion of a p-space (‘plumed’ space), which is a simultaneous
generalization of both topological completeness and metrizability. We give
two related versions under the simplifying assumption that X is completely
regular (for which notion see §1.2), an assumption that ensures that X has a
compactification (namely the Stone–Čech compactification).
Say that X is a p-space if there is a compactification Y in which there are
covers hUn i of X by sets open in Y such that the star-refinement satisfies
\
st(x, Un ) ⊆ X, ∀x ∈ X .
n

Here, for x a point in the space X and U a family of its subsets, st(x, U ) :=
S
{U : x ∈ U ∈ U } (see p. 44).
 2.5 Notes 49

A p-space space is called a strict p-space if, in addition,

\ \
st(x, Un ) = st(x, Un ).
n n
All metrizable spaces and all topologically complete spaces are p-spaces (see,
e.g., Arh1995). We refer to Baire p-spaces in Bouziad’s Theorem in Theorem
2.5.1.
Internal characterization is given by reference to a type of finite-intersection
property, as follows (again see, e.g., Gru1984).
(a) X is a strict p-space if and only if there exist open covers hGn i such that
T
K x := n st(x, Gn ) is compact for each x, and for each open G with
K x ⊆ G there is n with st(x, Gn ) ⊆ G (Arh1963).
(b) X is a p-space if and only if there exist open covers hGn i such that if
T
x ∈ G n ∈ Gn , then K x := m Ḡ m is compact for each x, and for each x
T
and open G with K x ⊆ G there is n such m≤n Ḡ m ⊆ G (Arh1963).
As examples of their usefulness, observe that a countable product of p-spaces,
or of M-spaces, is again a p-space, or respectively an M-space.
See Arh1995, §7 for a wide-ranging discussion of the significance of p-
spaces.

2. An Application of Games to p-Spaces.

The following result, noticed by Bouziad, illustrates how a p-space structure
on a Baire space makes it almost as good as a topologically complete space.
Theorem 2.5.1 (Bouz1993, Prop. 3.6) A Baire p-space is β-unfavourable for
the Christensen game Jpt, i.e. Player I has no winning strategy.
Proof Let hGn i be a plumage for X with the properties given in the definition
above. Suppose that at stage n II has played V1, . . . , Vn . Pick z n ∈ G n ∩ Vn with
G n ∈ Gn and Un with z n ∈ Ūn ⊆ G n ∩ Vn . Recall from Theorem 2.4.1 that,
since X is Baire, the strong Choquet game is unfavourable for I. Let I play Un
and let Player II apply her winning strategy against Un . That determines her
play of say Vn . Then
\ \
Un = Vn , ∅.
n n
T
As Ūn+1 ⊆ Un and Ūn ⊆ G n , this implies that K := n Ḡ n is non-empty;
by the definition of p-space K is compact. Now some point of K is a cluster
point of hx n i; otherwise, each point k ∈ K has a neighbourhood Uk avoiding
Z := {z n : n ∈ ω} and so, by compactness, some finite union of these yields
an open set U with U ⊇ K and avoiding Z. But Ḡ m ⊆ U for some m and this
implies that Um+1 avoids Z, a contradiction since z m ∈ Ūm ⊆ Ḡ m ⊆ U which
50 Baire Category

avoids Z. So Z has a cluster point z ∈ K . But z n ∈ Ūn for all z, so z ∈ Ūn for
all n; so
\
z∈ Un ∩ {z n : n ∈ ω} , ∅,
n
guaranteeing a win in the Christensen game. 
For fixed-point theorems in the context of game theory, see Bor1989.
 3

Borel Sets, Analytic Sets and Beyond: ∆21

3.1 Borel Sets and Analytic Sets

The theory of analytic sets dates from work of Souslin in 1916, Luzin in
1917 (Lus1917) and Luzin and Sierpiński in 1918 (LusS1923). For mono-
graph treatments, see Lus1930; RogJ1980. The historical origins, in an error
of Lebesgue in 1905, are given there – in Lebesgue’s preface to Lus1930 and
in RogJ1980, §1.3: projections of Borel sets need not be Borel, whence the
degradation studied above. Modern interest in analytic sets dates from its use
in measure theory in work in the 1950s by, e.g., Roy Davies and Choquet – see
Del1980.
In the Euclidean, and also in a metrizable topological, context the open sets
and the closed sets both generate the same σ-algebra – the Borel sets. Indeed,
one passes from the open sets G – the topology – to the closed sets F via
complements; but one may avoid complementation altogether by viewing the
closed sets as a subfamily of the Gδ sets (or the open sets as the subfamily
of the Fσ ). At this point we signal that on occasion it is worth drawing a
distinction between σ-algebras generated from a family H by reference only
to the ‘positive’ operations of countable intersections and unions – that is
precluding the ‘negative’ operation corresponding to complementation. Making
this distinction, one obtains for H = G or F the Borelian-G and Borelian-F ,
which coincide in a metric setting, and more importantly, corresponding to the
compact sets K there are also the Borelian-K sets, which coincide with the
earlier σ-algebras in a locally compact (and so in the Euclidean) setting.
The Borelian hierarchy has a parallel development as sets of points satis-
fying some property naturally described by a formula written out in a logical
symbolism using (in addition to the single real number playing the role of the
argument) only the usual connectives (& and ‘or’) and universal and existential
quantifiers over the natural numbers – typically identified as a function σ ∈ ωω

51
52 Borel Sets, Analytic Sets and Beyond: ∆21

(e.g. via the continued-fraction expansion). The function σ acts as a coding

device in two natural contexts: coded reference to a countable ordinal α (needed
to construct a Borel set at the αth level in the hierarchy) and reference to a given
open set G through a listing of the indices n of basic open sets Bn (taken from
a fixed basis B) with Bn ⊆ G (so that G is the union of the coded basic open
sets). One singles out the use of an effectively computable σ from among all
possible σ.
Some Euclidean sets rely in their natural (rigorous, but informal) definition
on the use of a universal or existential quantifier ranging over the reals. The use
of an existential quantifier may be interpreted as projection; thus a set on the
line described as
E := {x ∈ R : (∃y)Φ(x, y)}

referring to the plane Borel set

F := {(x, y) ∈ R2 : Φ(x, y)}

described using logical symbolism may also be written

E = proj1 F.

The question arises of when a Borel set F has a Borel projection. The standard
answers are: when

(i) each vertical {x} × R meets F in at most one point: here projection is a
one-to-one continuous mapping; more generally
(ii) each vertical {x} ×R meets F in a σ-compact set (e.g. a countable set). This
is the (Rogers-)Arsenin–Kunugui Theorem (see, e.g., RogJ1980, Thms.
5.9.1, 5.9.2).

Beyond this, in general E is not Borel. In the Euclidean context sets of the
general form E above are Souslin-K sets (see §1.4), i.e. of the form
[ \
ω
K (σ), where K (σ) := K (σ | n),
σ ∈ω n∈ω

where this representation has diam K (σ|n) ≤ 2−n and K (σ | n + 1) ⊆ K (σ |

n). In view of this, sets of the form E are simply known as Souslin sets or
analytic sets. See RogW1966 for their projections.
Remark While in R the concepts of being analytic and Souslin coincide, this
is not so in general. They coincide here because of upper semi-continuity (see
§1.4) of the mapping K : ωω → R; indeed, for open G, if
\
K (σ) ⊆ G, i.e. G c ∩ K (σ | n) = ∅,
n∈ω
 3.1 Borel Sets and Analytic Sets 53

then, by (countable) compactness, G c ∩ K (σ | N ) = ∅ for some N, and so, for

τ with τ | N = σ | N,
K (τ) ⊆ G.
Regarding ω as a discrete space metrized so that distinct points are unit distance
apart, the countable product topology on ωω is metrized by discounted first
differences:
d(σ, τ) = 2−n, n = inf{m : σm , τm }.
where
Under this structure on ωω , the mapping σ 7→ K (σ) := n∈ω K (σ | n) is
T
continuous on the closed set {σ : K (σ) , ∅}.
Finally we mention a result of particular significance to the hierarchy of
projective sets presented in §3.2.
Theorem 3.1.1 (Souslin’s Theorem) In a Polish space, a set is Borel if and
only if it and its complement are both analytic.
The projection of a Baire set may be, but need not be, Baire: under Gödel’s
Axiom of Constructibility,
V = L,
asserting that all sets are constructible (see Dev1973 and Chapter 1), there are
non-Baire sets which are projections of co-analytic sets (which are themselves
Baire), as the example (BinO2010b, Th. 4) demonstrates. However, the weaker
concept of shift-compactness, studied in Chapters 4 and 6, is preserved under
projection.
Theorem 3.1.2 (BinO2010b, Th. 1) For a shift-compact set in R2 its projec-
tion is shift-compact in R.

3.1.1 Analyticity and K-Analyticity

As before, we denote by K = K (X ) the family of compact subsets of a
topological space X, by G = G(X ) the open sets, and by F = F (X ) the
closed sets. As above, a set A in X is obtained from a family H of subsets of
X by the Souslin operation (briefly, is Souslin-H ), if there is a determining
system H = hH (i | n)i assigning to each finite sequence of positive integers
i | n := (i 1, i 2, . . . , i n ) (taken to be empty when n = 0) a set H (i | n) ∈ H with
[ [ \ \
A= H (i) = H (i | n), where H (i) := H (i | n).
i ∈I i ∈I n∈ω n∈ω
Here, as in Chapter 1, we write
I := NN,
54 Borel Sets, Analytic Sets and Beyond: ∆21

which is endowed with the product topology (treating N as discrete), so that for
i ∈ I its restrictions may be denoted

i | n := (i 1, . . . , i n ).

The behaviour of the mapping i 7→ H (i) is of significance, but before we turn

to considering this we mention two important dual results.

Theorem 3.1.3 (Nikodym’s Theorem and Corollary; Nik1925) If the sets in

H have the Baire property, then each Souslin-H set has the Baire property.
Hence Souslin-F sets are Baire.

Theorem 3.1.4 (Marczewski’s Theorem and Corollary) If the sets in H are

measurable with respect to a σ-finite regular measure µ on X, then each
Souslin-H set is µ-measurable.
Hence Souslin-F sets are measurable.

These results will be established in a more general non-separable metriz-

able context in Chapter 13. They both rely on the existence of a Baire (resp.
measurable) hull (see §3.4) of an arbitrary set A, a Baire/measurable expansion
M ⊇ A, minimal in the sense that, for any other Baire/measurable M 0 with
M ⊇ M 0 ⊇ A, the difference M\M 0 is meagre/null.

Definition (K -Analytic Spaces) We recall from RogJ1980 that for X a Haus-

dorff space, a map K : I → ℘(X ) is compact-valued if K (i) is compact for each
i ∈ I, and singleton-valued if each K (i) is a singleton. (For useful background
on compact-valued maps, see Ber1963; Hil1974.) The map K is upper semi-
continuous if, for each i ∈ I and each open U in X with K (i) ⊆ U, there is n
such that K (i 0 ) ⊆ U for each i 0 with i 0 | n = i | n.

A subset in X is K -analytic if it may be represented in the form K (I) for

some compact-valued, upper semi-continuous map K : I → ℘(X ). We say that
X is a K -analytic space if X itself is a K -analytic set.

Notation We put
I (i | n) := {i 0 : i 0 | n = i | n}

and
I< (i | n) := {i 0 ∈ I : i m
0
≤ i m for m ≤ n}.

So
\
I< (i) := I< (i | n)
n
 3.1 Borel Sets and Analytic Sets 55

is compact (by Tychonoff’s Theorem when viewing {i 0 | n : i m 0 ≤ i for m ≤ n}

m
as a compact set). Correspondingly, we write
[
K (i | n) :=K (I (i | n)) = {K (i 0 ) : i 0 | n = i | n},
K < (i | n) :=K (I< (i | n)), and K < (i) = K (I< (i));
the latter is also compact if K is compact-valued and upper semi-continuous.
Definition If K : I → K (X ) is upper semi-continuous, say that K is H -
circumscribed if there is a determining system hH (i | n)i assigning to each
i | n a set H (i | n) ∈ H with K (i) = H (i) := n∈ω H (i | n) such that
T
H (i) ⊆ U for U open implies that H (i | n) ⊆ U for some n.
Remarks 1. If K is upper semi-continuous, as above, and X is Hausdorff, then
[ \
K (I) = cl K (i | n),
i ∈I n∈ω
so K is F -circumscribed. In particular a K -analytic set is Souslin-F . Indeed,
T
as K (i) ⊆ n∈ω cl K (i | n), the inclusion of the left in the right is clear; for
T
the other direction, if x ∈ ( n∈ω cl K (i | n))\K (i) for some i, then there is U
open with x <clU and K (i) ⊆ U, and so K (i | n) ⊆ U for some n, yielding the
contradiction that x < cl K (i | n) ⊆ cl U.
2. If further X is a metric space and G(i | n) := B1/n (K (i | n)) = {x : d(x, y) <
1/n for some y ∈ K (i | n)}, then K (I) = i ∈I n∈ω G(i | n) is a Souslin-G
S T
representation such that:
(i) n∈ω G(i | n) = K (i) is compact;
T
(ii) K (i) ⊆ U, for U open, implies that G(i | n) ⊆ U for some n.
Then K is G-circumscribed. Indeed, if K (i) ⊆ U, then by compactness there
is n such that B2/n (K (i)) ⊆ U. But for m > n large enough, K (i | m) ⊆
B1/n (K (i)), and so B1/m (K (i | m)) ⊆ B1/m (B1/n (K (i))) ⊆ B2/n (K (i)) ⊆ U.
We will return to this observation in Theorem 8.2.1.
3. In a non-separable metric space a more relevant generalization has I
replaced by
J := κ N,
with κ an arbitrary infinite cardinal (again viewed, like N above, as a discrete
space). See Chapter 13.
Definition (Analytic Spaces) Call a Hausdorff space X analytic if X is the
continuous image of a closed subset of I, or equivalently of a Polish space. So an
analytic space is K -analytic, since singletons are compact; a K -analytic subset
of an analytic set is analytic. A K -analytic space is more general, since it is
56 Borel Sets, Analytic Sets and Beyond: ∆21

a continuous image of a Lindelöf–Čech-complete space – by Jayne’s Theorem

(RogJ1980, Th. 2.8.1); cf. Hansell (Hanse1992, Th. 3.1).
Note that if X is metric and K -analytic, then without loss of generality we
may arrange to have diamX (K (i | n)) < 2−n . This implies that K (i) = {k (i)}
on a closed subset of I on which k is continuous. Since the closed subspace of
I is analytic, this yields a space which is a continuous image of I, so again an
analytic space. See RogJ1980, §2.8 for background on different definitions of
analyticity, and §2.11 therein for applications to Banach spaces under the weak
topology.
In a metric analytic space open sets, being Fσ, are analytic. Note that an
open subset of a Polish space is again a Polish space. This yields the following
generalizations.
Lemma 3.1.5 (i) In an analytic space, all open sets are analytic.
(ii) A regular K -analytic space (in particular a regular analytic space) is
Lindelöf, so open sets are K -analytic if and only if they are Fσ .
Proof (i) If X is the continuous image of I under α, then each open U in X is
the continuous image under α of α −1 (U), an open subset in I.
(ii) By regularity, for U open, U has an open covering by sets V with cl V ⊆ U.
If U is K -analytic, it is Lindelöf, and so U has a countable covering by open sets
Vn with cl Vn ⊆ U; so U = n cl Vn is an Fσ . For the converse, as Souslin-Fσ
S
subsets of a K -analytic space are K -analytic, open sets are K -analytic if they
are Fσ . 
Remarks 1. Of course for U open, (cl U)\U is closed nowhere dense, so
modulo a meagre set an open set is closed, so ‘almost K -analytic’.
2. The two parts of Lemma 3.1.5 are close since (Lev1983; HruA2005) every
analytic Baire space has a dense completely metrizable subspace; in fact every
regular analytic space has a finer topology in which it is metrizable (just declare
the countably many sets cl α(I (i | n)) to be open). Condition (i) of the lemma
is related to the existence of winning strategies in certain topological games
(see Ost2011, Proposition L3 in §3 and Whi1975, Th. 11).

3.1.2 Analytic Cantor Theorems

The following result is implicit in a number of situations and goes back to
Frolík’s characterization of Čech-complete spaces as Gδ in some compacti-
fication (Fro1960; see Eng1989, §3.9); it may be used to lift theorems about
Polish spaces to results on analytic metric spaces and to characterize ana-
lytic sets. For example, Frolík (Fro1970, Th. 2) characterizes analytic sets as
 3.1 Borel Sets and Analytic Sets 57

intersections of a Gδ and a set that is Souslin in its Stone–Čech compact-

ification; in similar spirit Fremlin (Fre1980) develops the theory of Čech-
analytic sets (cf. also HanseJR1983). In the opposite direction, Aarts et al.
(AarGM1970b; AarGM1970a) use similar machinery to characterize com-
pleteness via compactness. Recall that Cantor’s Theorem on the intersection of
a nested sequence of closed (or compact, as appropriate) sets has two formula-
tions: (i) referring to vanishing diameters (in a complete-space setting); and (ii)
to (countable) compactness. In the spirit of these, we now give two topological
versions. Refer to §3.1.1 for notation.
Theorem 3.1.6 (Analytic Cantor Theorem; Ost2011) Let X be a Hausdorff
space and A = K (I) be K -analytic in X, with K compact-valued and upper
semi-continuous.
If Fn is a decreasing sequence of (non-empty) closed sets in X such that
F ∩ K (I (i 1, . . . , i n )) , ∅, for some i = (i 1, . . .) ∈ I and each n, then K (i) ∩
Tn
n Fn , ∅.
Equivalently, if there are open sets Vn in I with cl Vn+1 ⊆ Vn and diamI Vn ↓ 0
such that Fn ∩ K (Vn ) , ∅, for each n, then:
T
(i) n cl Vn is a singleton, {i} say;
T
(ii) K (i) ∩ n Fn , ∅.
Proof If not, then n K (i) ∩Fn = ∅ and so, by compactness, K (i) ∩Fp = ∅ for
T
some p, i.e. K (i) ⊆ X\Fp . So by upper semi-continuity Fp ∩ K (I (i 1, . . . , i n )) =
∅ for some n ≥ p, yielding the contradiction Fn ∩ K (I (i 1, . . . , i n )) = ∅. 
Theorem 3.1.6 has the following filter base (finite intersection property, ‘fip’)
generalization.
Theorem 3.1.7 In the setting of Theorem 3.1.6, for H a filter base, if for
some i and each n ∈ ω, H0 ∈ H , there is m > n and H ∈ H with H ⊆ H0
meeting K (I (i 1, . . . , i m )), then
\
K (i) ∩ {cl H : H ∈ H } , ∅.
Proof If not, and ∅ = K (i)∩ {cl H : H ∈ H }, then, for some finite subfamily
T
H 0, we have ∅ = K (i) ∩ {cl H : H ∈ H 0 }, and so K (i) ⊆ {X\ cl H : H ∈
T S
H 0 }. By upper semi-continuity, K (I (i 1, . . . , i n )) ⊆ {X\ cl H : H ∈ H 0 }, for
S
some n. As H is a filter base, there exists H0 ∈ H with H0 ⊆ {H : H ∈ H 0 }.
T
Now for some m > n and some H1 ∈ H with H1 ⊆ H0, the set K (I (i 1, . . . , i m ))
meets H1, contradicting the fact that ∅ = K (i) ∩ {H : H ∈ H }.
T

The filter base version is usually rendered employing ‘inclusion’ as below –
suggesting the shrinking ‘diameters’ of Cauchy’s criterion, ultimately the
58 Borel Sets, Analytic Sets and Beyond: ∆21

inspiration of Frolík’s (Fro1960); cf. again Eng1989, §3.9; Hanse1992, §3,

p. 281 – rather than the ‘trace on K (i)’ property above (‘intersection with
K (i)’). This has a similar but simpler proof, given below for the sake of com-
pleteness. In fact the inclusion version implies the trace version (see the Remark
following the next theorem). In §2.3 we see their duals in the Banach–Mazur
‘inclusion’ games and the Choquet ‘trace’ games.
Theorem 3.1.8 In the setting of Theorem 3.1.6, for H a filter base, if, for
some i, each K (I (i 1, . . . , i n )) contains a member of H , then
\
∅, {cl H : H ∈ H } ⊆ K (i).
Proof The inclusion is clear. If ∅ = K (i) ∩ {cl H : H ∈ H }, then for
T
some finite subfamily H 0 we have ∅ = K (i) ∩ {cl H : H ∈ H 0 }, and so
T
K (i) ⊆ {X\ cl H : H ∈ H 0 }. By upper semi-continuity, K (I (i 1, . . . , i n )) ⊆
S
{X\ cl H : H ∈ H 0 }, for some n. But K (I (i 1, . . . , i n )) ⊇ H0 for some (non-
S
empty!) H0 ∈ H , and so ∅ = H0 ∩ H 0 for each H 0 ∈ H 0, contradicting
the fact that H is a filter sub-base, unless H 0 = ∅. But then K (i) = ∅ ⊇
K (I (i 1, . . . , i n )) ⊇ H0 , giving a final contradiction. 
Remark To see why Theorem 3.1.8 implies Theorem 3.1.7, consider a filter
base H with the intersection property of 3.1.7 relative to i ∈ I. One may pick for
each H0 ∈ H and n ∈ N an integer m = m(n) > n and a set H = H n (H0 ) ⊆ H0
in H such that Hin (H0 ) := H ∩ K (i 1, . . . , i m(n) ) , ∅. Then {Hin (H0 ) : n ∈ N,
H0 ∈ H } is a filter sub-base satisfying the hypothesis of 3.1.8, and so
\ \
∅ , K (i) ∩ {cl Hin (H0 ) : n ∈ N, H0 ∈ H } ⊆ K (i) ∩ {cl H0 : H0 ∈ H }.
A similar argument can be conducted with a weaker hypothesis by exploiting
the compactness of K < (i) (defined in §3.1.1).
Theorem 3.1.9 In the setting of Theorem 3.1.6, suppose now that the nested
sequence Fn satisfies Fn ∩ K (I< (i 1, . . . , i n )) , ∅, for some i = (i 1, . . .) ∈ I and
T
each n. Then K < (i) ∩ n Fn , ∅.
T
Equivalently, if there are open sets Vn in I with H := n cl Vn non-empty
T
compact sets such that Fn ∩ K (Vn ) , ∅ for each n, then K (H) ∩ n Fn , ∅.
Proof If not, then K < (i)∩ n Fn = ∅. Since K < (i) is compact, K < (i)∩Fp = ∅,
T
for some p. By upper semi-continuity, for each j ∈ I< (i) there is n( j) such that
K ( j | n( j)) ⊆ X\Fp . Since I< (i) is compact, there are j (1), . . . , j (t) in I< (i)
and integers ns = n( j (s)) such that {I ( j (s) | ns ) : s = 1, . . . , t} is a finite open
covering of I< (i). Put q = p + maxs ≤t ns .
For j ∈ I< (i | q), consider j 0 with j 0 | q = j | q and j 0 ∈ I< (i). (For
instance, take j 0 = j1, . . . , jq, i q+1, i q+2, . . .). Refer to the finite covering to find
 3.1 Borel Sets and Analytic Sets 59

s with j 0 | ns = j (s) | ns . Then K ( j | q) ⊆ K ( j (s) | ns ) ⊆ X\Fp . So K (I< (i |

q)) ⊆ X\Fp, and in particular K (I< (i | q)) ∩ Fq = ∅, a contradiction. 

The following generalization of Theorem 3.1.6 is at the heart of both the

proof of the Gandy–Harrington Theorem (see §8.2) and likewise of van Mill’s
proof (Mil2004, Prop. 2.2) of his Analytic Baire Theorem (concerning the
countable intersection of dense, heavy sets).

Theorem 3.1.10 Let X be a Hausdorff space and An = Kn (I) be K -analytic

in X, with Kn taking singleton or empty values and upper semi-continuous.
If Fn is a decreasing sequence of (non-empty) closed sets in X such that
\
Fn ∩ Km (I (i 1, . . . , i n )) , ∅,
m≤n

for some i = (i 1, . . .) ∈ I and each n, then

T T
n Fn ∩ n An (i) , ∅.

Proof If not, write Hn := Kn (i) and Kn (i) := {x n } (whenever Kn (i) is

non-empty). By compactness, since n (Fn ∩ Hn ) = ∅, there is p with Fp ∩
T

n≤p Hn = ∅. If x m < Fp or Hm = ∅ for some m ≤ p, then K m (i) ⊆ X\Fp, and

T
so Fp ∩ Km (In (i | n)) = ∅ for some n > p + m. Then Fn ∩ Km (In (i | n)) = ∅,
a contradiction. So x m ∈ Fp for all m ≤ p. Since Fp ∩ n≤p Hn = ∅, for some
T
m, m 0 we have x m , x m0 . As X is Hausdorff, for some disjoint U, V we have
x m ∈ U and x m0 ∈ V . So for some n > m + m 0 + p we have Km (i | n) ⊆ U and
Km0 (i | n) ⊆ V . So Fn ∩ Km (i | n) ∩ Km0 (i | n) = ∅, a contradiction. 

We generalize the last result beyond the singleton-valued to the compact-

valued case, which needs a separation lemma. (This generalizes the well-known
result that in a Hausdorff space disjoint compact sets may be separated by
disjoint open sets – cf. Kel1955, Th. 5.9.) The next result is shown for regular
Hausdorff spaces in DavJO1977 (in the course of a proof of Th. 1 there). We
work in subspaces that are K -analytic, so the following more intuitive proof
applies. Recall from Lemma 3.1.5 that a K -analytic space A is Lindelöf and
that a regular Lindelöf space is normal (cf. Kel1955, Lemma 4.1). So a regular
K -analytic space A is normal.

Lemma 3.1.11 (Separation Lemma) In a normal space, and so also in a

locally compact Hausdorff space X, for an ordered finite sequence of compact
sets hK1, . . . , Kn i with empty intersection and n ≥ 2, there is a corresponding
ordered finite sequence hU1, . . . , Un i of open sets with empty intersection such
that Ki ⊆ Ui .

Proof Suppose given the compact sequence hK1, . . . , Kn i. First assume that
X is normal. For each i the set Ki is disjoint from the set K−i := j,i K j .
T
60 Borel Sets, Analytic Sets and Beyond: ∆21

As in Urysohn’s Separation Lemma (Chapter 1), let f i : X → [0, 1] be a

continuous function with zero set Ki such that f i (K−1 ) = 1. Put f := i ≤n f i ;
P
then f (x) = 0 if and only if x ∈ i ≤n Ki = ∅, and so f > 0 on X . Now
T
Ui := {x : f i (x) < f (x)/n} is open and Ki ⊆ Ui , as Ki is the zero set of f i .
It follows that hU1, . . . , Un i has empty intersection; indeed, if x ∈ i ≤n Ui, for
T
some x, then summing the relations f i (x) < f (x)/n we obtain the contradiction
that f (x) < f (x).
Now assume that X is locally compact and Hausdorff; we may choose U
open containing i ≤n Ki with Y = cl U compact. As Y is normal, we may find
S
sets Vi open in Y separating the sets Ki as required, but in Y . Taking Ui = Vi ∩U
we obtain the desired separation in U and so in X . 

Remark For an alternative proof, using subnets, see Ost2011, §1.2.

We now give the promised generalization of Theorem 3.1.10 from singleton-
valued to compact-valued representations.

Theorem 3.1.12 (Multiple K -Analytic Targets) Let X be regular Hausdorff

and An = Kn (I) be K -analytic in X, with Kn compact-valued and upper semi-
continuous. If Fn is a decreasing sequence of (non-empty) closed sets in X
with
\
Fn ∩ Km (i 1 (m), . . . , i n (m)) , ∅,
m≤n

for some i(n) = (i 1 (n), . . .) ∈ I and each n, then n Fn ∩ n Kn (i(n)) , ∅.

T T

Proof If not, write Hn := Kn (i(n)). Put Y = n Hn . As Y contains the sets

S
Fn ∩ m≤n Km (I (i 1 (m), . . . , i n (m))), and regularity is subspace hereditary, we
T
may as well assume that X = Y . By compactness, since n (Fn ∩ Hn ) = ∅,
T
there is p with Fp ∩ n≤p Hn = ∅. If Fp ∩ Hm = ∅ for some m ≤ p, then
T
Km (i(m)) ⊆ X\Fp, and so Fp ∩ Km (I (i(m) | n)) = ∅ for some n > p + m. Then
Fn ∩ Km (I (i(m) | n)) = ∅, a contradiction. So the compact set Hm 0 := F ∩ H
p m
is non-empty for each m ≤ p. Since n≤p Hn = ∅, for some open-in-Fp
0
T

n≤p Un = ∅ (by the Separation Lemma 3.1.11 –

sets Um ⊇ Hm 0 we have T

as X is now assumed Lindelöf, and so normal). So for some n > p we have

Km (i(m) | n) ⊆ Um for each m ≤ p, and so Fn ∩ n≤p Km (i(m) | n) = ∅, a
T
contradiction. 

3.2 Beyond Analytic Sets: Projective Hierarchy

For the most part we work with the mindset of the practising analyst, that is in
‘naive’ set theory. As usual, we work in the standard mathematical framework
 3.2 Beyond Analytic Sets: Projective Hierarchy 61

of Zermelo–Fraenkel set theory (ZF), i.e. we do not make use of the Axiom
of Choice (AC) unless we say so explicitly. Our interest in the complexities
induced by the limsup operation points us in the direction of definability and
descriptive set theory because of the question of whether certain specific sets,
encountered in the course of analysis, have the Baire property. The answer
depends on what further axioms one admits. For us there are two alternatives
yielding the kind of decidability we seek. Firstly, there is Gödel’s Axiom of
Constructibility (V = L), as an appropriate strengthening of the Axiom of
Choice, which creates definable sets without the Baire property (or without
measurability). Secondly, at the opposite pole, there is the Axiom of Projective
Determinacy (PD) (see MycSw1964 or Kec1995, 5.38.C), which guarantees
the Baire property in the kind of definable sets we encounter. Thus to decide
whether sets of the kind we encounter below have the Baire property, or are
measurable, the answer is: it depends on the axioms of set theory that one
adopts. See §14.3 and Chapter 16.

3.2.1 Projective/Analytical Hierarchy

To formulate our results we need the language of descriptive set theory, for
which see, e.g., RogJ1980; Kec1995; Mos2009. Within such an approach we
will regard a function as a set, namely its graph. We need the beginning of the
projective hierarchy (or analytical hierarchy in Euclidean space (see Kec1995,
S. 37.A), in particular the following classes:
the analytic sets Σ11 ;
their complements, the co-analytic sets Π11 ;
the common part of the previous two classes, the ambiguous class ∆11 :=
Σ11 ∩Π11, that is, by Souslin’s Theorem (3.1.1; RogJ1980, p. 5, and MartK1980,
p. 407 or Kec1995, 14.C) the Borel sets;
the projections (continuous images) of Π11 sets, forming the class Σ12 ;
their complements, forming the class Π12 ;
the ambiguous class ∆21 := Σ12 ∩ Π12 ;
and then: Σ1n+1, the projections of Π1n ; their complements Π1n+1 ; and the
ambiguous class ∆n+1 1 := Σ1n+1 ∩ Π1n+1 .
The notation (bold-faced to refer to coding, cf. page 32) reflects the fact that
the canonical expression of the logical structure of their definitions, that is with
the quantifiers (ranging over the reals, hence the superscript 1, as reals are type 1
objects – integers are of type 0) all at the front, is determined by a string of
alternating quantifiers starting with an existential or universal quantifier (resp.
Σ or Π). Here the subscript accounts for the number of alternations.
62 Borel Sets, Analytic Sets and Beyond: ∆21

In §3.3 we will examine the projective character of sets naturally associated

with asymptotic behaviour of regularly varying functions, i.e. we will verify
their position in the hierarchy. The concern there will be with the cases n = 1, 2
or 3.

3.3 * Character Theorems

The Baire/measurable property discussed at various points in the Prologue is
usually satisfied in mathematical practice. Indeed, any analytic subset of R
possesses these properties (RogJD1980, Part 1 §2.9; Kec1995, 29.5), hence
so do all the sets in the σ-algebra that they generate (the C-sets, Kec1995,
§29.D, C for crible – see Burg1983a; Burg1983b, cf. BinO2011a). There is a
broader class still. Recall first that an analytic set may be viewed as a projection
of a planar Borel set P, so is definable as {x : Φ(x)} via the Σ11 formula
Φ(x) := (∃y ∈ R)[(x, y) ∈ P]; here the notation Σ11 indicates one quantifier
block (the subscripted value) of existential quantification, ranging over reals
(type 1 objects – the superscripted value). As in §1.4 use of the bold-face
version of the symbol indicates the need to refer to arbitrary coding (by reals
not necessarily in an effective manner, for which see Gao2009, §1.5) of the
various open sets needed to construct P. (An open set U is coded by the
sequence of rational intervals contained in U.) Effective variants are rendered
in light-face.
Consider a set A such that both A and R\A may be defined by a Σ12 formula,
say respectively as {x : Φ(x)} and {x : Ψ(x)}, where Φ(x) := (∃y ∈ R)(∀z ∈
R)(x, y, z) ∈ P} now, and similarly Ψ. This means that A is both Σ12 and
Π12 (with Π indicating a leading universal quantifier block), and so is in the
ambiguous class ∆21 . If in addition the equivalence
Φ(x) ⇐⇒ ¬Ψ(x),
where ¬, meaning negation (see BelS1969), is read ‘not’, is provable in Z F,
i.e. without reference to AC, then A is said to be provably ∆21 . It turns out that
such sets have the Baire/measurable property – see FenN1973, where these are
generalized to the universally (= absolutely) measurable sets (cf. BinO2017,
§2); the idea is ascribed to Solovay in Kan2003, Ch. 3, Ex. 14.4. How much
further this may go depends on what axioms of set theory are admitted, a matter
to which we presently turn.
Our interest in such matters derives from the Character Theorems of regular
variation, as noted in BinO2010b, §3 (revisited in BinO2016a, §11), which
identify the logical complexity of the function
 3.3 * Character Theorems 63

h∗ (x) := lim sup h(t + x) − h(t),

t→∞

which is ∆21 if the function h (more precisely, its graph) is Borel (and is Π12 if
h is analytic, and Π13 if h is co-analytic). We argued in BinO2010b, §5 that ∆21
is a natural setting in which to study regular variation.
Interest in the character of a function h is motivated by an interest within the
theory of regular variation in the character of the level sets

H k := {s : |h(s)| < k} = {s : (∃t)[(s, t) ∈ h & |t| < k]}

for k ∈ N (where as above h is identified with its graph). The set H k is thus
the projection of h ∩ (R × [0, k]) and hence is Σ1n if h is Σ1n, e.g. it is Σ11, i.e.
analytic, if h is analytic (in particular, Borel). Also

H k = {s : (∀t)[(s, t) ∈ h =⇒ |t| ≤ k]} = {s : (∀t)[(s, t) < h or |t| ≤ k]},

and so this is also Π1n if h is Σ1n . Thus if h is Σ1n , then H k is ∆n1 . So if ∆n1 sets
are Baire, for some k the set H k is Baire/non-null, and hence shift-compact
(Chapter 4), as
[
R= Hk .
k ∈ω

With this in mind, it suffices to consider upper limits; as before, we prefer to

work with the additive formulation. Consider the definition

h∗ (x) := lim sup[h(t + x) − h(t)].

t→∞

Thus in general h∗ takes values in the extended real line. The problem is
that the function h∗ is in general less well behaved than the function h – for
example, if we assume h measurable, h∗ need not be measurable, and similarly
if h has the Baire property, h∗ need not. The problem we address here is the
extent of this degradation – saying exactly how much less regular than h the
limsup h∗ may be. The nub is the set S on which h∗ is finite. This set S is an
additive semi-group (cf. §15.2) on which the function h∗ is subadditive (see
BinO2008) – or additive, if limits exist (see BinO2009b). Furthermore, if h has
Borel graph, then h∗ has ∆21 graph (see below). But in the presence of certain
axioms of set-theory (for which see below), the ∆21 sets have the Baire property
and are measurable; hence if S is large in either of these two senses, then in
fact S contains a half-line. The extent of the degradation in passing from h to
h∗ is addressed in the following result (taken from BinO2016a), which we call
the First Character Theorem, and then contrast it with two alternative character
theorems. Undefined terms are explained below in the course of the proof; as
64 Borel Sets, Analytic Sets and Beyond: ∆21

in BGT1987, p. 17, we reserve the name Characterization Theorem (CT) for a

result identifying the limit function arising there concerning limsupm, cf. §P.6.
Theorem 3.3.1 (First Character Theorem)
(i) If h is Borel (has Borel graph), then the graph of the function h∗ (x) is a
difference of two analytic sets, hence is measurable and ∆21 . If the graph
of h is Fσ , then the graph of h∗ (x) is Borel.
(ii) If h is analytic (has analytic graph), then the graph of the function h∗ (x)
is Π12 .
(iii) If h is co-analytic (has co-analytic graph), then the graph of the function
h∗ (x) is Π13 .
In our next theorem we assume much more:
Theorem 3.3.2 (Second Character Theorem) Suppose h ∈ ∆21 and the follow-
ing limit exists:
∂h(x) := lim [h(t + x) − h(t)].
t→∞

Then the graph of ∂h is ∆21 .

The point of the next theorem is that it may be applied under the assumption of
Gödel’s Axiom V = L (see Dev1973), as this axiom implies that ∆21 ultrafilters
on ω exist (see, for instance, Zap1999; Zap2001, where Ramsey ultrafilters are
considered). Recall (Chapter 1) that sets of natural numbers are identified with
real numbers (via their indicator functions) and so ultrafilters are regarded as
sets of reals. For information on various types of ultrafilter on ω, see ComN1974
or HinS1998 (see KomT2008 for an introduction). In particular this means that
we have a midway position between the results of the First and Second Character
Theorems.
Theorem 3.3.3 (Third Character Theorem) Suppose the following are of class
∆21 : the function h and an ultrafilter U on ω. Then the following is of class ∆21 :
∂U h(t) := U- lim[h(t + n) − h(n)].
n

Comment 1. In the circumstances of Theorem 3.3.3 ∂U h(t) is an additive

function, whereas in those of Theorem 3.3.1 one has only subadditivity. See
BGT1987, p. 62, equation (2.0.3).

Comment 2. One may also consider replacing h(t + n) − h(n) by h(t + x(n)) −
h(x(n)), as in the Equivalence Theorem of BinO2009c, so as to take limits along
a specified sequence x : ω → ωω, in which case to have an ‘effective’ version
of Theorem 3.3.3 one would need to specify the effective descriptive character
of x. (Here again ωω is identified with the reals via indicator functions.)
 3.3 * Character Theorems 65

3.3.1 Proofs
Proof of the First Character Theorem
(i) Let us suppose that h is Borel (i.e. h has a Borel graph). As a first step
consider the graph of the function of two variables h(t + x) − h(t), namely the
set
G = {(x, t, y) : y = h(t + x) − h(t)}.

One expects this to be a Borel set and indeed it is. For a proof, we must
refer back to the set h itself, and to do this we must re-write the defining clause
appropriately. This re-writing brings out explicitly an implicit use of quantifiers,
a common enough occurrence in analysis (see end of BinO2010b for another
important example). We have

y = h(t + x) − h(t) ⇔ (∃u, v, w ∈ R)r (x, t, y, u, v, w),

where

r (x, t, y, u, v, w) = [ y = u − v & w = t + x & (w, u) ∈ h& (t, v) ∈ h]. (*)

From a geometric viewpoint, the set of points

{(x, t, y, u, v, w) : r (x, t, y, u, v, w)}

is Borel in R6, hence the set

G = {(x, t, y) : (∃u, v, w ∈ R)r (x, t, y, u, v, w)},

being a projection of a Borel set, is an analytic set in R3, and in general not
Borel. However, in the particular present context the ‘sections’

{(u, v, w) : r (x, y, z, t)},

corresponding to fixed (x, t, y) ∈ G, are single points (since u, v and w are

defined uniquely by the values of x and t). In consequence, the projection here
is Borel. The reason for this is that any Borel set is a continuous injective
image of the irrationals (RogJ1980, §3.6, p. 69), and so a continuous injective
image, as here under projection, of a Borel set is Borel. (So here the hidden
quantifiers are ‘innocuous’, in that they do not degrade the character of G.) The
current result may also be seen as the simplest instance of a more general result,
the Rogers–Kunugui–Arsenin Theorem, which asserts that if the sections of a
Borel set are Fσ (i.e. countable unions of closed sets), then its projection is
Borel (RogJ1980, pp. 147/148).
By abuse of notation, let us put h(t, x) := h(t + x) − h(t) and think of t as
parameterizing a family of functions. By assumption, the family of functions
66 Borel Sets, Analytic Sets and Beyond: ∆21

h(t, x) is Borel; that is, the graph {(x, y, t) : y = h(t, x)} is a Borel set (we will
weaken this restriction appropriately below).
As a second step, we now consider the formal definition of h∗ (x), again writ-
ten out as in predicate calculus using a semi-formal apparatus. The definition
comes naturally as a conjunction of two clauses:

y = h∗ (x) ⇔ P(x, y) & Q(x, y),

where

P =(∀n)(∀q ∈ Q+ )(∃t ∈ R)(∃z ∈ R)[t > n & z = h(t, x) & |z − y| < q],
Q =(∀q ∈ Q+ )(∃m)(∀t ∈ R)(∀z ∈ R)[t > n & (t, x, z) ∈ h =⇒ z < y + q].

The first clause (predicate) asserts that y is a limit point of the set {h(t, x) :
t ∈ R} and this requires an existential quantifier; the second clause asserts that,
with finitely many exceptions, no member of the set exceeds y by more than q
and this requires a universal quantifier.
From a geometric viewpoint, for fixed q > 0 the set of points

G1 = {(x, y, z, t) : p(x, y, z)},

where
p(x, y, z, t) = [(t, x, z) ∈ h & |z − y| < q],

is Borel in R4, hence again the set {(x, y) : (∃z, t ∈ R)p(x, y, z)}, being a
projection of a Borel set, is an analytic set in R2 . Again, for fixed (x, y) we look
at the section of G1 . Evidently {z : |z − y| < q} is an open set, so Fσ . However,
only if we assume that h is Fσ can we deduce that

{(x, y) : (∃t ∈ R)(∃z ∈ R)[t > n & z = h(t, x) & |z − y| < q}

is Borel. Otherwise it is merely analytic.

Since the quantifiers in (∃z ∈ R)(∃t ∈ R)p(x, y, z, t) are at the front of the
defining formula, we see that the formula is Σ11 . See RogJ1980 for a modern side-
by-side exposition of the two viewpoints of mathematical logic and geometry.
Finally, the set {(x, y) : P(x, y)} is seen to be obtainable from analytic sets (or
Borel in the special case) by use of countable union and intersection operations.
It is thus an analytic set (or Borel as the case may be).
By contrast, for Q(x, y, z, t, q) := [[z = h(t, x)] =⇒ z < y + q], the set
{(x, y) : (∀z, t ∈ R)(∀q ∈ Q+ )Q(x, y, z, t, q)}, is co-analytic, since its comple-
ment is the analytic set

{(x, y) : (∃z, t ∈ R)(∃q ∈ Q+ ))[z = h(t, x) & z ≥ y + q]}.

 3.3 * Character Theorems 67

Again for given q and for arbitrary fixed (x, y) the section of {(x, y, z, t) : [z =
h(t, x) & z ≥ y + q]} will be Fσ if the graph of h is Fσ, but is otherwise
analytic. Thus {(x, y) : Q(x, y)} is seen to be obtainable from co-analytic sets
(or at best Borel sets) by use of countable union and intersection operations. It
is thus co-analytic, or Borel as the case may be.
Again, as with Q above, the formula (∀z ∈ R)q(x, y, z, t) is Π11 , since the
opening quantifier is universal of order 1.
The set {(x, y) : Q(x, y)} is seen to be obtainable from co-analytic sets by
use of countable union and intersection operations. It is thus co-analytic since
such operations preserve this character. Finally, note that the sets which are
differences of analytic sets are both in the classes Π11 and Σ12 , and so in their
intersection ∆21 . We have of course neglected the possibility that the lim sup is
infinite, but for this case we need only note that
h∗ (x) =∞ ⇔ (∀n)(∃t ∈ R)(∃z ∈ R)[t > n & z = h(t, x) & z > n],
h∗ (x) <∞ ⇔ (∃y ∈ R)(y = h∗ (x)),
so that this case is simultaneously Σ11 and Π11 .
We have thus proved part (i).
(ii) Now assume that h has an analytic graph. Then G, being the projection of
an analytic set, is now analytic. That is, we may write
y = h(t, x) ⇔ (∃w ∈ R)F (t, x, y, w),
where the set {(t, x, y, w) : F (t, x, y, w)} is Borel. Then
{(x, y) : (∃z ∈ R)(∃w ∈ R)[F (t, x, z, w) & |z − y| < q]}
is only analytic, since we have no information about special sections; however,
the set
{(x, y) : (∀z ∈ R)(∃w ∈ R)[t > n & F (t, x, z, w) =⇒ z < y + q]}
requires for its definition a quantifier alternation which begins with a universal
quantifier, so is Π12 . Since Σ11 sets are necessarily a subclass of Π12 sets, the
graph of lim supt f (t, x) in this case is Π12 , which proves (ii).
(iii) Now, suppose that the function h(x) has a co-analytic graph. Then by (*)
the set G is of class Σ12, i.e. the function h(t, x) has a Σ12 graph. That is, we now
have to write
y = h(t, x) ⇔ (∃u ∈ R)(∀w ∈ R)F (t, x, y, u, w),
where as before the set {(t, x, y, w) : F (t, x, y, u, w)} is Borel. Then
{(x, y) : (∃z, u ∈ R)(∀w ∈ R)[F (t, x, z, u, w) & |z − y| < q]}
68 Borel Sets, Analytic Sets and Beyond: ∆21

is now Σ12 . On the other hand, the set

{(x, y) : (∀z ∈ R)(∃u ∈ R)(∀w ∈ R)[F (t, x, z, u, w) =⇒ z < y + q]}
is Π13 . Since Σ11 sets are necessarily a subclass of Π13 sets, the graph of
lim supt h(t, x) in this case is Π13 . 
Proof of the Second Character Theorem Here we have
y = ∂h(x) ⇐⇒ (∀q ∈ Q+ )(∃n ∈ ω)(∀t > n)(∀zuvw)P,
where
P = [z = u − v & w = t + x & (t, v) ∈ h & (w, u) ∈ h] =⇒ |z − y| < q ,
 

and
y , ∂h(x) ⇐⇒ (∀q ∈ Q+ )(∃n ∈ ω)(∀t > n)(∀zuvw)Q,
where
Q = [z = u − v & w = t + x & (v, t) ∈ h & (u, w) ∈ h] =⇒ |z − y| ≥ q . 
 

Proof of the Third Character Theorem By (*) the function y = h(t, x) is of

class Σ12 . We show that y = ∂U h(t) is of class Σ12 . The result will follow since
the negation satisfies
y , ∂U h(t) ⇐⇒ ∃z z , y & [z = h∗ (t) orh∗ (t) = ±∞] ,
 

and so is of class Σ12 . Finally,

y = ∂U h(t) ⇐⇒ (∀ε ∈ Q+ )(∃U)(∀n ∈ ω)(∃t)P,
where
P = [U ∈ U & n ∈ U & (n, t) ∈ x & |t − y| < ε],
and
∂U h(t) = ∞ ⇐⇒ (∀M ∈ Q+ )(∃U ∈ U )(∀n ∈ U)(∃t)[(n, t) ∈ x & t > M].


3.4 * Appendix: Baire Hull

The proof below relies on Banach’s Category Theorem. Here M denotes the
meagre sets and Ba the sets with the Baire property. A point x is a heavy point
for an arbitrary set A if each neighbourhood of x meets A in a non-meagre set.
The heavy points form a closed set. One way to see this is to consider the set of
 3.4 * Appendix: Baire Hull 69

light points of A, namely those having a neighbourhood meeting A in a meagre

set. The light points evidently form an open set.
A set A that is not Baire is of course not a meagre set. Consequently, there is
a natural expansion of A less a meagre set that is Baire, namely the closed set
of points on which A is ‘heavy’. This set is minimal among Baire sets in that
any smaller expansion of A differs by a meagre set, as in the next result.

Theorem 3.4.1 (Baire Hull) For any A, there is B ∈ Ba with A ⊆ B such

that for any B 0 ∈ Ba, if A ⊆ B 0 ⊆ B, then B\B 0 ∈ M.

Proof For any A, we begin by decomposing X into a part on which A is light

and one on which it is heavy, as follows. Write

L( A) = L N ( A) := {U ∈ G(X ) : U ∩ A ∈ M},
[
L( A) := L( A), H ( A) := X\L( A).

Claim for A ∈ Ba, that H ( A)\A ∈ M.

Choose G open agreeing with A modulo M. Put F = cl G; then A\F ⊆ A\G.

So A ∩ (X\F) ∈ M, and (X\F) ∈ L( A). So X\F ⊆ L( A), giving H ( A) ⊆ F.
So, since (cl G)\G and (G\A) ∈ M,

H ( A)\A ⊆ F\A = (cl G)\A ⊆ (cl G)\G ∪ (G\A) ∈ M.

We next show that A\H ( A) = A ∩ L( A) ∈ M for arbitrary A, implying that

A is covered by the union of H ( A) and a meagre set (so that the symmetric
difference A∆H ( A) ∈ M).
For L 0 ( A) a maximal family of disjoint open subsets in L( A), let
[ / [
Q := L( A) L 0 ( A).

We have that Q is nowhere dense. For suppose that Q is dense in an open set V .
Then V ∩U is non-empty for some U ∈ L( A). Since A∩U is in M, so A∩U ∩V
is in M, i.e. U ∩ V is in L( A). Now Q is dense in the open subset V ∩ U. As
S 0 S
L ( A) is open, Q is closed as a subset of L( A). Hence Q contains V ∩ U,
and so U ∩ V is disjoint from L 0 ( A), contradicting the maximality of L 0 ( A).
So after all Q is nowhere dense.
We now show that A ∩ L 0 ( A) ∈ M. For each U in L 0 ( A), write
S
[
A∩U = Mn (U),
n∈N

with Mn (U) nowhere dense sets and put

[
Mn := {Mn (U) : U ∈ L 0 ( A)}.
70 Borel Sets, Analytic Sets and Beyond: ∆21

Then Mn is nowhere dense, by the Banach Category Theorem (as Mn (U) ⊆ U

and L 0 ( A) is disjoint). Finally,
[
A\H ( A) = A ∩ L( A) ⊆ Q ∪ Mn ∈ M.
n∈N
So
[
A ⊆ B := H ( A) ∪ H0, with H0 = Q̄ ∪ M̄n ∈ M,
n∈N
exhibiting an Fσ cover of A, as H ( A) is closed. Now consider B 0 ∈ Ba with
A ⊆ B 0 ⊆ B. Then
L(M) ⊆ L(B 0 ) ⊆ L( A), and H ( A) ⊆ H (B 0 ) ⊆ H (B).
Now
B\B 0 = H0 \B 0 ∪ H ( A)\B 0 ⊆ H0 \B 0 ∪ H (B 0 )\B 0 .
But both H0 and H (B 0 )\B 0 ∈ M, the latter by the claim at the outset. So B is
a Baire hull. 
 4

Infinite Combinatorics in Rn : Shift-Compactness

The results of this chapter develop a new aspect of measure–category duality.

This has powerful applications.

4.1 Generic Dichotomy

We will meet many examples of situations which divide into two sharply
contrasting cases – usually where behaviour is, in some appropriate sense, either
very good or very bad. Such dichotomy often has a generic aspect: if something
works at all, it works nearly everywhere. We address the source of this genericity
below: a property inheritable by supersets either holds generically or fails
outright.

Definition For X the reals R+ with the Euclidean or density topology (Chapter
7), denote by Ba(X ), or just Ba, the Baire sets of the space X, and recall these
form a σ-algebra. Say that a correspondence F : Ba → Ba is monotonic if
F (S) ⊆ F (T ) for S ⊆ T .

The nub is the following simple result, which we call the Generic Dichotomy
Principle.

Theorem 4.1.1 (Generic Dichotomy Principle; BinO2010f) For F : Ba →

Ba monotonic: either

(i) there is a non-meagre S ∈ Ba with S ∩ F (S) = ∅; or,

(ii) for every non-meagre T ∈ Ba, T ∩ F (T ) is quasi almost all of T .

Equivalently: the existence condition, that S ∩ F (S) , ∅ should hold for all
non-meagre S ∈ Ba, implies the genericity condition that, for each non-meagre
T ∈ Ba, T ∩ F (T ) is quasi almost all of T .

71
72 Infinite Combinatorics

Proof Suppose that (i) fails. Then S∩F (S) , ∅ for every non-meagre S ∈ Ba.
We show that (ii) holds. Suppose otherwise; then, for some T non-meagre in
Ba, the set T ∩ F (T ) is not almost all of T . Then the set U := T\F (T ) ⊆ T is
non-meagre (it is in Ba as T and F (T ) are), and so

∅ , U ∩ F (U) (S ∩ F (S) , ∅ for every non-meagre S)

⊆ U ∩ F (T ) (U ⊆ T and F monotonic).

But, as U := T\F (T ), we have U ∩ F (T ) = ∅, a contradiction.

The final assertion simply rephrases the dichotomy as an implication. 

The next result transfers the onus of verifying the existence condition above
to topological completeness: it suffices to test the F property on Gδ sets.

Theorem 4.1.2 (Generic Completeness Principle; BinO2010f) For F : Ba →

Ba monotonic, if W ∩ F (W ) , ∅ for all non-meagre W ∈ Gδ, then, for each
non-meagre T ∈ Ba, T ∩ F (T ) is quasi almost all of T .
That is, either

(i) there is a non-meagre S ∈ Gδ with S ∩ F (S) = ∅; or,

(ii) for every non-meagre T ∈ Ba, T ∩ F (T ) is quasi almost all of T .

Proof By Baire’s Theorem, for S non-meagre in Ba there is a non-meagre

W ⊆ S with W ∈ Gδ . So W ∩ F (W ) , ∅ and thus ∅ , W ∩ F (W ) ⊆ S ∩ F (S),
by monotonicity. By Theorem 4.1.1, for every non-meagre T ∈ Ba, T ∩ F (T )
is quasi almost all of T . 

The rest of this section is concerned with examples of monotonic corre-

spondences which play a part in the development of our subject. Each of the
correspondences F to be introduced below will give rise to a correspondence
Φ( A) := F ( A) ∩ A which is a lower or upper density (see Chapter 7, and
also LukMZ1986) and so gives rise to a fine topology on the real line (see
Chapter 8).

Example (R, D). Here the meagre sets are the null sets. Let B denote a
countable basis of Euclidean open neighbourhoods. For any set T and 0 < α < 1
put
Bα (T ) := {I ∈ B : |I ∩ T | > α|I |},

which is countable, and

\ [
F (T ) := {I : I ∈ Bα (T )}.
α∈Q∩(0,1)
 4.2 KBD – First Proof 73

Thus F is increasing in T, F (T ) is measurable (even if T is not) and x ∈ F (T )

if and only if x is a density point of T . The genericity assertion is the Lebesgue
Density Theorem of Chapter 7; see, e.g., Saks1964, IV.10; Bog2007a, Th.5.6.2.
Further related examples follow in §8.6.

4.2 Kestelman–Borwein–Ditor (KBD) Theorem: First Proof

The main result of this section is the following theorem establishing the funda-
mental infinite combinatorics which apply to both Baire and measurable sets.
These dual foundations, originating in single-variable theory, were unified in
two ways in BinO2009b and BinO2010h. In one they are unified structurally
by their identical combinatorics. In the other, both are derived from a single
source: Baire category. Both views translate immediately to Rd .
Theorem 4.2.1 (Theorem KBD – Kestelman–Borwein–Ditor Theorem) Let
{z n } → 0 be a null sequence of reals. If T is Baire/Lebesgue measurable, then,
for generically all t ∈ T, there is an infinite set Mt such that
{t + z m : m ∈ Mt } ⊆ T .
Variants of this theorem, depending on context, permeate this book. One in-
stance, Theorem 10.5.2, is concerned with bounded sequences, and there such
sequences will have subsequences contained in the target set T, up to trans-
lation as above. It seems appropriate to call this property shift-compactness,
borrowing an established probabilistic term, coined by Parthasarathy (Par1967),
describing a very similar feature. The property here does in fact imply a finite
sub-covering result: see §6.4.

Infinitely Often versus Co-finitely Often

There is some slight difference (which need not detain us here) between the
combinatorics of category and measure, indeed of no significance to conver-
gence properties, in view of the sub-subsequence theorem.
In the case of a non-meagre Baire target set T, the infinite combinatorics
of Theorem 4.2.1 can be strengthened by replacing the infinite set of indices
m ∈ M for which the translates t + z m are in T with a co-finite set of indices.
For a study of this aspect in a general group setting, see MilleMO2021. 1
1 Before naming them shift-compact, our earlier work used sub-universal for sets containing, up
to translation as in Theorem 4.2.1, a subsequence from every null sequence. This followed
Kestelman’s use in Kes1947a of universal to describe sets co-finitely containing all convergent
sequences up to translation. For an analysis of the subtle relationship between Parthasarathy’s
term and ours, see BinO2024.
74 Infinite Combinatorics

For examples where co-finite combinatorics fail, see BorwD1978; Mille1989

and, e.g., Komj1988.
The stronger combinatorics are of immediate use when checking the continu-
ity of an additive function on T, say when establishing f (t) = limn→∞ f (t + z n ),
when t ∈ T and t + z n are all but finitely many in T .
But a subsequence approach, albeit less elegant, is equally helpful. It suffices
to establish that f (t) = limm∈M0 f (t + z m ) for some infinite M0 ⊆ M of
an arbitrary sequence M ⊆ N. Indeed, if there is a neighbourhood U of f (t)
and an infinite subset M of N omitting { f (t + z m ) : m ∈ M}, then f (t) =
limn∈M0 f (t + z m ) cannot hold for any M0 ⊆ M.
In summary: when studying convergence properties, the infinitely often com-
binatorics are adequate and available for both measure and category contexts.
Hence we speak of identical combinatorics.
Our first proof of the KBD Theorem 4.2.1 follows from the two results
below, both important in their own right. The first and its corollary address
displacements of open sets in the density and the Euclidean topologies; it
is mentioned (in passing) in a note added in proof (p. 32) in Kemperman,
Kem1957, Th. 2.1, p. 30, for which we give an alternative proof. The second
parallels an elegant result for the measure case treated in BergHW1997.
Theorem 4.2.2 (Displacements Lemma – Kemperman’s Theorem; Kem1957
Th. 2.1 with Bi = E, ai = t) If E is non-null Borel, then f (x) := |E ∩ (E + x)|
is continuous at x = 0, and so, for some ε = ε(E) > 0,
E ∩ (E + x) is non-null for |x| < ε.
More generally, f (x 1, . . . , x p ) := |(E + x 1 ) ∩ · · · ∩ (E + x p )| is continuous at
x = (0, . . . , 0), and so, for some ε = ε p (E) > 0,
(E + x 1 ) ∩ · · · ∩ (E + x p ) is non-null for |x i | < ε, (i = 1, . . . , p).
First Proof (After BergHW1997; cf. e.g. BinO2010g, Th. 7.5) Let t be a
density point of E. Choose ε > 0 such that
3
|E ∩ Bε (t)| >
|Bε (0)|.
4
Now |Bε (t)\Bε (t + x)| ≤ (1/4)|Bε (t + x)| for x ∈ Bε/2 (0), so
1
|E ∩ Bε (t + x)| >
|Bε (0))|.
2
By invariance of Lebesgue measure we have
3
|(E + x) ∩ Bε (t + x)| > |Bε (0))|.
4
 4.2 KBD – First Proof 75

But, again by invariance, as Bε (t) + x = Bε (0) + t + x this set has measure

|Bε (0)|. Using | A1 ∪ A2 | = | A1 | + | A2 | − | A1 ∩ A2 | with A1 := E ∩ Bε (t + x)
and A2 := (E + x) ∩ Bε (t + x) now yields
5
|E ∩ (E + x)| ≥ |E ∩ (E + x) ∩ (Bε (t) + x)| > |Bε (0)| − |Bε (0)| > 0.
4
Hence, for x ∈ Bε/2 (0), we have |E ∩ (E + x)| > 0.
For the p-fold form we need some notation. Let t again denote a density point
of E and x = (x 1, . . . , x n ) a vector of variables. Set A j := Bε (t) ∩ E ∩ (E + x j )
for 1 ≤ j ≤ n. For each multi-index i = (i(1), . . . , i(d)) with 0 < d < n, put
f i (x) := | Ai(1) ∩ · · · ∩ Ai(d) |;
f n (x) := | A1 ∩ · · · ∩ An |, f 0 = |Bε (t) ∩ E|.
We have already shown that the functions f j (x) = |Bε (t) ∩ E ∩ (E + x j )| are
continuous at 0. Now argue inductively: suppose that, for i of length less than
n, the functions f i are continuous at (0, . . . , 0). Then for given ε > 0, there is
δ > 0 such that for k xk < δ and each such index i we have
−ε < f i (x) − f 0 < ε,
where f 0 = |Bε (t) ∩ E|. Noting that
[n
Ai ⊂ B(t) ∩ E,
i=1
and using upper or lower approximations, according to the signs in the inclusion–
exclusion identity
[n X X \
Ai = | Ai | − | Ai ∩ A j | + · · · + (−1) n−1 Ai ,
i=1 i i< j i

one may compute linear functions L(ε), R(ε) such that

L(ε) < f n (x) − f 0 < R(ε).
Indeed, taking x i = 0 in the identity, both sides collapse to the value f 0 .
Continuity follows. 
Second Proof Apply instead Theorem 61.A of Hal1950, Ch. XII, p. 266 to
establish the base case, and then proceed inductively as before. 
Corollary 4.2.3 Kemperman’s Theorem (4.2.2) holds for non-meagre Baire
sets E in place of Borel sets in the form: for each p in N there exists ε =
ε p (E) > 0 such that
(E + x 1 ) ∩ · · · ∩ (E + x p ) non-meagre, for |x i | < ε, (i = 1, . . . , p).
Proof A non-meagre Baire set differs from an open set by a meagre set. 
76 Infinite Combinatorics

We will now prove Theorem 4.2.1 using the Generic Completeness Principle
(Theorem 4.1.1); this amounts to proceeding in two steps. To motivate the proof
strategy, note that the embedding property is upward hereditary (i.e. monotonic
in the sense of §4.1): if T includes a subsequence of z n by a shift t in T, then so
does any superset of T . We first consider a non-meagre Gδ /non-null closed set
T, just as in BergHW1997, modified to admit the consecutive format so that,
for infinitely many m, z m and all of its p consecutive terms in the null sequence,
which we denote by

z̄ pm := {z m, z m+1, . . . , z m+p },

translate into the target set T. We next deduce the theorem by appeal to Gδ
inner-regularity of category/measure and Generic Dichotomy. (The subset E of
exceptional shifts can only be meagre/null.) The next result is a generalization
of a lemma due to Bergelson, Hindman and Weiss proving the existence of a
sequence embedding.

Theorem 4.2.4 (See BergHW1997, Lemma 2.2) For T Baire non-meagre/

measurable non-null and a null sequence z n → 0, there exist t ∈ T and an
infinite Mt such that
{t + z̄ pm : m ∈ Mt } ⊆ T .

Proof The conclusion of the theorem is inherited by supersets (is upward

hereditary), so without loss of generality we may assume that T is Baire non-
meagre/measurable non-null and completely metrizable, say under a metric
ρ = ρT . (For T measurable non-null we may pass down to a compact non-null
subset, and for T Baire non-meagre we simply take away a meagre set to leave a
Baire non-meagre Gδ subset.) Since this is an equivalent metric, for each a ∈ T
ρ
and ε > 0 there is δ = δ(ε) > 0 such that Bδ (a) ⊆ Bε (a). Thus, by taking
ε = 2−n−1 , the δ-ball Bδ (a) has ρ-diameter less than 2−n .
Working inductively in steps of length p, we define subsets of T (of possible
translators) Bpm+i of ρ-diameter less than 2−m for i = 1, . . . , p as follows.
With m = 0, we take B0 = T. Given n = pm and Bn open in T, choose
N such that |zk | < min{ 21 |x n |, ε p (Bn )}, for all k > N . For i = 1, . . . , p, let
x n−1+i = z N +i ; then by Kemperman’s Theorem 4.2.2 or its Corollary, 4.2.3,
Bn ∩ (Bn − x n ) ∩ · · · ∩ (Bn − x n+p ) is non-empty (and open). We may now
choose a non-empty subset Bn+i of T which is open in A with ρ-diameter less
than 2−m−1 such that clT Bn+i ⊂ Bn ∩ (Bn − x n ) ∩ · · · ∩ (Bn − x n+i ) ⊆ Bn+i−1 .
T
By completeness, the intersection n∈N Bn is non-empty. Let
\
t∈ Bn ⊂ T .
n∈N
 4.2 KBD – First Proof 77

Now t + x n ∈ Bn ⊂ T, as t ∈ Bn+1, for each n. Hence Mt := {m : z mp+1 =

x n for some n ∈ N} is infinite. Moreover, if z pm+1 = x n , then z pm+2 =
x n+1, . . . , z pm+p = x n+p−1 , and so
{t + z̄ pm : m ∈ Mt } ⊆ T . 
We now apply Theorem 4.1.1 (Generic Dichotomy) to extend what we now
know from an existence to a genericity statement, thus completing the proof of
Theorem 4.2.1.
Theorem 4.2.5 For T Baire/measurable and z n → 0, for generically all t ∈ T
there exists an infinite Mt such that
{t + z̄ pm : m ∈ Mt } ⊆ T .
Hence, if Z ⊆ X accumulates at 0 (has an accumulation point there), then for
some t ∈ T the set Z ∩ (T − t) accumulates at 0 (along Z). Such a t may be
found in any open set with which T has non-null intersection.
Proof Working as usual in X, the correspondence
\ [
F (T ) := [(T − z pm+1 ) ∩ · · · ∩ (T − z pm+p )]
n∈ω m>n
takes Baire sets to Baire sets and is monotonic. Here t ∈ F (T ) if and only if
there exists an infinite Mt such that {t + z̄ pm : m ∈ Mt } ⊆ T . By Theorem
4.2.2 F (T ) ∩ T , ∅, for T Baire non-meagre, so we may appeal to the Generic
Dichotomy Principle (Theorem 4.1.1) to deduce that F (T ) ∩ T is quasi all of T
(cf. the example of §4.1).
With the main assertion proved, let Z ⊆ X accumulate at 0 and suppose that
z n in Z converges to 0. Take p = 1. Then, for some t ∈ T, there is an infinite
Mt such that {t + z m : n ∈ Mt } ⊆ T . Thus {z m : n ∈ Mt } ⊆ Z ∩ (T − t) has 0
as a joint accumulation point. 
 5

Kingman Combinatorics and

Shift-Compactness

The KBD theorem of the previous chapter is about shift-embedding of sub-

sequences of a null sequence {z n } into a single set T with an assumption
of regularity (Baire/measurable). Our generalizations below of a theorem of
Kingman (Kin1963; Kin1964) have been motivated by the wish to establish
‘multiple embedding’ versions of KBD: we seek conditions on a sequence {z n }
and a family of sets {Tk }k ∈ω which together guarantee that one shift embeds
(different) subsequences of {z n } into all members of the family.
Evidently, if t + z n lies in several sets infinitely often, then the sets in question
have a common limit point, a sense in which they are contiguous at t. Thus
contiguity conditions are one goal, the other two being regularity conditions
on the family, and admissibility conditions on the null sequences.
We view the original Kingman Theorem as studying contiguity at infinity,
so that divergent sequences z n (i.e. with z n → +∞) there replace the null
sequences of KBD. The theorem uses openness as a regularity condition on the
family, cofinality at infinity (e.g. unboundedness on the right) as the simplest
contiguity condition at infinity, and

z n+1 /z n → 1 (multiplicative form),

(*)
z n+1 − z n → 0 (additive form)

as the admissibility condition on the divergent sequence z n ; (*) follows from

regular variation by Weissman’s Lemma (BGT1987, Lemma 1.9.6). Taken
together, these three guarantee multiple embedding (at infinity).
One can switch from ±∞ to 0 by an inversion x → 1/x, and thence to
any τ by a shift y → y + τ. Openness remains the regularity condition, a
property of density at zero becomes the analogous admissibility condition on
null sequences, and cofinality (or accumulation) at τ the contiguity condition.
The transformed theorem then asserts that for admissible null sequences ζ n

78
 5.1 Definitions and Notation 79

there exists a scalar σ such that the sequence σζ n + τ has subsequences in all
the open sets Tk provided these all accumulate at τ.
Bitopology: The Euclidean and Density Topologies The two main themes of
this book, category and measure, have many similarities and many contrasts
(as the Table of Contents will show). Each has a topology intimately linked
with it. For category, it is the ordinary (Euclidean) topology, E. For measure,
it is the density topology, D (Kec1995, 17.47). We defer a full treatment of the
density topology to Chapter 7.
Below, we will replace Kingman’s regularity condition of openness by the
Baire property, or alternatively measurability, to obtain two versions of King-
man’s Theorem – one for category and one for measure. As above, we develop
the regularity theme bitopologically, working with two topologies, so as to
deduce the measure case from the Baire case by switching from the Euclidean
E to the density topology D (see the end of §1.4 and Chapter 7).

5.1 Definitions and Notation

5.1.1 Essential Contiguity Conditions
We use the notation Br (x) := {y : |x − y| < r } and ω := {0, 1, 2, . . .}. Likewise
ρ
for a ∈ A ⊆ R and metric ρ = ρ A on A; Br (a) := {y ∈ A : ρ(a, y) < r } and
cl A denotes closure in A. For S given, put S >m = S\Bm (0). Denote by R+ the
(strictly) positive reals. When we regard R+ as a multiplicative group, we write
A · B := {ab : a ∈ A, b ∈ B}, A−1 := {a−1 : a ∈ A},
for A, B subsets of R+ .
Call a Baire set S essentially unbounded if for each m ∈ N the set S >m is
non-meagre. This may be interpreted in the sense of the metric (Euclidean)
topology, or as we see later in the measure sense by recourse to the density
topology. To distinguish the two, we will qualify the term by referring to the
category/metric or the measure sense.
Say that a set S ⊂ R+ accumulates essentially at 0 if S −1 is essentially
unbounded. (In BergHW1997 such sets are called measurably/Baire large at 0.)
Say that S ⊂ R+ accumulates essentially at t if (S − t) ∩ R+ accumulates
essentially at 0.
We begin by simplifying essential unboundedness modulo meagre/null sets.
Theorem 5.1.1 In R+ with the Euclidean or density topology, and with S
Baire/measurable and essentially unbounded, there exists an open/density-
open unbounded G and meagre/null M with G\M ⊂ S.
80 Kingman Combinatorics

Proof Choose integers mn inductively with m0 = 0 and mn+1 > mn the least
integer such that (mn, mn+1 ) ∩ S is non-meagre. For, given mn , the integer mn+1
is well defined, as otherwise, for each m > mn , we would have (mn, m) ∩ S
meagre, and so also
[
(mn, ∞) ∩ S = (mn, m) ∩ S meagre,
m>mn

contradicting the essential unboundedness of S. Now, as (mn, mn+1 ) ∩ S is

Baire/measurable, we may choose G n open/density-open and Mn, Mn0 meagre
subsets of (mn, mn+1 ) such that
((mn, mn+1 ) ∩ S) ∪ Mn = G n ∪ Mn0 .
Hence G n is non-empty. Put G := n G n and M := n Mn . Then M is
S S
meagre and G is open unbounded and, since M ∩ (mn, mn+1 ) = Mn and
G ∩ (mn, mn+1 ) = G n,
[ [ [
G\M = G n \M = G n \Mn ⊂ (mn, mn+1 ) ∩ S = S,
n n n
as asserted. 

Weakly Archimedean Property – An Admissibility Condition

Let I be N or Q, with the ordering induced from the reals. Our purpose will be
to take limits through subsets J of I which are unbounded on the right (more
briefly: unbounded). According as I is N or Q, we will write n → ∞ or q → ∞.
Denote by X the line with either the metric or density topology and say that
a family {hi : i ∈ I} of self-homeomorphisms of the topological space X is
weakly Archimedean if, for each non-empty open set V in X and any j ∈ I, the
open set
[
U j (V ) := hi (V )
i≥j

meets every essentially unbounded set in X .

Theorem 5.1.2 (Implicit in BGT1987, Th. 1.9.1 (i)) In the multiplicative
group of positive reals R∗+ with the Euclidean topology, if d n is divergent
and the multiplicative form of condition (*) above holds, then the functions
hn (x) = d n x for n = 1, 2, . . . , are homeomorphisms and {hn : n ∈ N } is
weakly Archimedean. For any interval J = (a, b) with 0 < a < b and any m,
[
Um (J) := dn J
n≥m
contains an infinite half-line, and so meets every unbounded open set. Similarly
this is the case in the additive group of reals R with hn (x) = d n + x and
Um (J) = n≥m (d n + J).
S
 5.1 Definitions and Notation 81

Proof For given ε > 0 and all large enough n, we have 1−ε < d n /d n+1 < 1+ε.
Write x := (a + b)/2 ∈ J. For ε small enough, a < x(1 − ε) < x(1 + ε) < b,
and then a < xd n /d n+1 < b, hence xd n ∈ d n+1 J, and so d n J meets d n+1 J.
Thus for large enough n consecutive d n J overlap; as d n → ∞, their union is
thus a half-line. 
Remark Some such condition as (*) is necessary, otherwise the set Um (J) risks
missing an unbounded sequence of open intervals. For an indirect example, see
the remark in BGT1987 after Th. 1.9.2 and G.E.H. Reuter’s elegant coun-
terexample to a ‘corollary’ of Kingman’s Theorem, a breakdown caused by
the absence of our condition. For a direct example, note that if d n = r n log n
with r > 1 and J = (0, 1), then d n + J and d n+1 + J miss the interval (1 +
r n log n, r n+1 log(1 + n)) and the omitted intervals have union an unbounded
open set; to see that the omitted intervals are non-degenerate note that their
lengths are unbounded:
r n+1 log(1 + n) − r n log n − 1 → ∞ (n → ∞).
Theorem 5.1.2 does not extend to the real line under the density topology;
the homeomorphisms hn (x) = nx are no longer weakly Archimedean, as we
demonstrate by an example in BinO2010f, Theorem 4.6. We are thus led to an
alternative approach driven by the following lemma, a multiplicative version of
an additive result in Sie1920.
Lemma 5.1.3 For a, b density points of their respective measurable sets A,
B in R+ and for n = 1, 2, . . ., there exist positive rationals qn and points an , bn
converging to a, b through A, B, respectively, such that bn = qn an .
Proof For n = 1, 2, . . . and the consecutive values ε = 1/n the sets Bε (a) ∩
A and Bε (b) ∩ B are measurable non-null, so by Steinhaus’ Theorem (in
multiplicative rather than additive form; see BGT1987, Th.1.1.1 or Chapter 15)
the set [B ∩ Bε (b)] · [A ∩ Bε (a)]−1 contains interior points, and so in particular
a rational point qn . Thus for some an ∈ Bε (a) ∩ A and bn ∈ Bε (b) ∩ B we have
qn = bn an−1, and as |a − an | < 1/n and |b − bn | < 1/n, an → a, bn → b. 
Remarks 1. For the purposes of the next theorem, we observe that qn may
be selected arbitrarily large, for fixed a, by taking b sufficiently large (since
qn → ba−1 ).
2. The lemma addresses D-open sets but also holds in the metric topology (the
proof is similar but simpler), and so may be restated bitopologically as follows.
Theorem 5.1.4 For R+ with either the Euclidean or the density topology, if
a, b are respectively in the open sets A, B, then, for n = 1, 2, . . ., there exist
82 Kingman Combinatorics

positive rationals qn and points an , bn converging metrically to a, b through

A, B respectively, such that
bn = qn an .

We now prove the density-topology analogue of Theorem 5.1.2.

Theorem 5.1.5 In the multiplicative group of reals R∗+ with the density topol-
ogy, the family of homeomorphisms {hq : q ∈ Q+ } defined by hq (x) := qx,
where Q+ has its natural order, is weakly Archimedean. For any density-open
set A and any j ∈ Q+ ,
[
U j ( A) := qA
q ≥ j,q ∈Q+

contains almost all of an infinite half-line, and so meets every unbounded

density-open set.

Proof Let B be Baire and essentially unbounded in the D-topology. Then B is

measurable and essentially unbounded in the sense of measure. From Theorem
5.1.1, we may assume that B is density-open. Let A be non-empty density-open.
Fix a ∈ A and j ∈ Q+ . Since B is unbounded, we may choose b ∈ B such that
b > ja. By Theorem 5.1.4 there is a q ∈ Q+ with j < q < ba−1 such that
qa 0 = b0, with a 0 ∈ A and b0 ∈ B. Thus

U j ( A) ∩ B ⊇ hq ( A) ∩ B = q A ∩ B , ∅,

as required.
If U j ( A) fails to contain almost all of any infinite half-line, then its comple-
ment B := R+ \U j ( A) is essentially unbounded in the sense of measure and so,
as above, must meet U j ( A), a contradiction. 

Remarks 1. For A the set of irrationals in (0, 1) the set U j ( A) is again a set
of irrationals which contains almost all, but not all, of an infinite half-line. Our
result is thus best possible.
2. Note that (*) is relevant to the distinction between integer and rational
skeletons; see the prime-divisor example on p. 53 of BGT1987. Theorem 5.1.5
holds with Q+ replaced by any countable dense subset of R∗+ , although later we
use the fact that Q+ is closed under multiplication. There is an affinity here with
the use of a dense ‘skeleton set’ in the Heiberg–Seneta Theorem, Th. 1.4.3 of
BGT1987, and its extension, Th. 3.2.5, therein.

Theorem 5.1.6 (Bitopological Kingman Theorem; Kin1963, Th. 1; Kin1964,

where I = N) If X is a Baire space, and
 5.1 Definitions and Notation 83

(i) {hi : i ∈ I} is a countable, linearly ordered, weakly Archimedean family

of self-homeomorphisms of X; and
(ii) {Sk : k = 1, 2, . . .} are essentially unbounded Baire sets, then, for quasi
all η ∈ X and all k ∈ N, there exists an unbounded subset Jηk of I with

{h j (η) : j ∈ Jηk } ⊂ Sk .

Equivalently, if (i) and

(ii0) { Ak : k = 1, 2, . . .} are Baire and all accumulate essentially at 0, then for

quasi all η and every k = 1, 2, . . ., there exists Jηk unbounded with

{h j (η) −1 : j ∈ Jηk } ⊂ Ak .

Proof We will apply Theorem 4.1.1 (Generic Dichotomy), so consider an ar-

bitrary non-meagre Baire set T . We may assume without loss of generality that
T = V \M with V non-empty open and M meagre. For each k = 1, 2, . . .,
choose G k open and Nk and Nk0 meagre such that Sk ∪ Nk0 = G k ∪ Nk .
Put N := M ∪ n,k h−1 n (Nk ); then N is meagre (as h n, and so h n , is a
S 0 −1

homeomorphism).
As Sk is essentially unbounded, G k is unbounded (otherwise, for some m,
G k ⊂ (−m, m), and so Sk ∩ (m, ∞) ⊂ Nk is meagre). Define the open sets
G jk := i ≥ j hi−1 (G k ). We first show that each G jk is dense. Suppose, for some
S
j, k, there is a non-empty open set V such that V ∩ G jk = ∅. Then for all i ≥ j,

V ∩ hi−1 (G k ) = ∅; G k ∩ hi (V ) = ∅.

So G k ∩ i ≥ j hi (V ) = ∅, i.e. for U j the open set U j := i ≥ j hi (V ), we have

S S
G k ∩ U j = ∅. But as G k is unbounded, this contradicts {hi } being a weakly
Archimedean family.
Thus the open set G jk is dense (meets every non-empty open set); so, as I is
countable, the Gδ set
\∞ \
H := G jk
k=1 j ∈I

is dense (as X is a Baire space). So as V is a non-empty open subset we may

choose η ∈ (H ∩ V )\N . (Otherwise N ∪ (X\H) and hence V is meagre.) Thus
η ∈ T and for all k = 1, 2, . . . ,
\ [
η∈V∩ hi−1 (G k ) and η < N . (**)
j ∈I i≥j

For all m, as hm (η) < hm (N ) we have for all m, k, hm (η) < Nk0 . Using (**), for
each k select an unbounded Jηk such that for j ∈ Jηk , η ∈ h−1
j (G k ); for such j we
have η ∈ h j (Sk ). That is, for some η ∈ T we have
−1
84 Kingman Combinatorics

{h j (η) : j ∈ Jηk } ⊂ Sk .

Now
\∞ \ [
F (T ) := T ∩ hi−1 (G k )
k=1 j ∈I i≥j

takes Baire sets to Baire sets and is monotonic. Moreover, η ∈ F (T ) if and only
if η ∈ T and for each k there is an unbounded Jηk with {h j (η) : j ∈ Jηk } ⊂ Sk .
We have just shown that T ∩ F (T ) , ∅ for T arbitrary non-meagre, so the
Generic Dichotomy Principle, Theorem 4.1.1 implies that X ∩ F (X ) is quasi
almost all of X, i.e. for quasi all η in X and each k there is an unbounded Jηk
with {h j (η) : j ∈ Jηk } ⊂ Sk . 

Working in the density and the Euclidean topology in turn, we obtain the
following conclusions.

Theorem 5.1.7 (Kingman Theorem for Category) If {Sk : k = 1, 2, . . .} are

Baire and essentially unbounded in the category sense, then for quasi all η and
each k ∈ N there exists an unbounded subset Jηk of N with

{nη : n ∈ Jηk } ⊂ Sk .

In particular this is so if the sets Sk are open.

Theorem 5.1.8 (Kingman Theorem for Measure) If {Sk : k = 1, 2, . . .} are

measurable and essentially unbounded in the measure sense, then for almost
all η and each k ∈ N there exists an unbounded subset Jηk of Q+ with

{qη : q ∈ Jηk } ⊂ Sk .

In the following corollary, Jtk refers to unbounded subsets of N or Q+ accord-

ing to the category/measure context. It specializes down to a KBD result for a
single set T when Tk ≡ T, but it falls short of KBD in view of the extra admis-
sibility assumption and the factor σ (the latter an artefact of the multiplicative
setting).

Corollary 5.1.9 For {Tk : k ∈ ω} Baire/measurable and z n → 0 admissible,

for generically all t ∈ R there exist σt and unbounded Jtk such that for k =
1, 2, . . .,

t ∈ Tk =⇒ {t + σt z m : m ∈ Jtk } ⊂ Tk .
 5.1 Definitions and Notation 85

Proof For T Baire/measurable, let N = N (T ) be the set of points t ∈ T

which are not points of essential accumulation of T; then t ∈ N if for some
n = n(t) the set T ∩ B1/n (t) is meagre/null. As R with the Euclidean topology
is (hereditarily) second countable, it is hereditarily Lindelöf (see Eng1989,
Th. 3.8.1 or Dug1966, Th. 8.6.3), so for some countable S ⊂ N
[
N⊂ T ∩ B1/n(t) (t),
t ∈S
and so N is meagre/null. Thus the set Nk of points t ∈ Tk such that Tk − t does
not accumulate essentially at 0 is meagre/null, as is N = k Nk . For t < N,
S
put Ωt := {k ∈ ω : Tk − t accumulates essentially at 0}. Applying Kingman’s
Theorem, 5.1.6, to the sets {Tk − t : k ∈ Ωt } and the sequence z n → 0, there
exist σt and unbounded Jtk such that for k ∈ Ωt ,
{σt z m : m ∈ Jtk } ⊂ Tk − t, i.e. {t + σt z m : m ∈ Jtk } ⊂ Tk .
Thus for t < N, so for generically all t, there exist σt and unbounded Jtk such
that, for k = 1, 2, . . .,
t ∈ Tk =⇒ {t + σt z m : m ∈ Jtk } ⊂ Tk . 
 6

Groups and Norms: Birkhoff–Kakutani Theorem

6.1 Introduction
Group-norms, which behave like the usual vector norms except that scaling
is restricted to the basic scalars of group theory (the units ±1 in an abelian
context and the exponents ±1 in the non-commutative context), have played a
part in the early development of topological group theory, and created an early
context for functional analysis as conducted by such American mathematicians
as Michal (Mic1940; Mic1947) and Bartle (Bar1955). They appear naturally in
the study of groups of homeomorphisms. See the long paper, BinO2010g, by
the present authors for a detailed development and also ArhT2008. For some
relatively recent work on group norms, see CabC2002, Sect. 2.1.1 which uses
embedding of quasi-normed groups into Banach spaces in considering Ulam’s
problem (see Ula1960) on the global approximation of nearly additive functions
by additive functions. See also BurBI2001.
The basic metrization theorem for groups, the Birkhoff–Kakutani Theorem
of 1936 (Bir1936; Kak1936; see Kel1955; Kle1952; Bou1966; ArhM1998,
compare also Eng1989), is usually stated as asserting that a first-countable
Hausdorff topological group has a right-invariant metric. It is, properly speak-
ing, a ‘normability’ theorem in the style of Kolmogorov’s Theorem (Kol1934 or
Rud1973; in this connection see also Jam1943, where strong forms of connect-
edness are used in an abelian setting to generate norms), as we shall see below
(see also JamMW1947). Indeed the metric construction in Kak1936 is reminis-
cent of the more familiar construction of a Minkowski functional (for which see
Rud1973), but is implicitly a supremum norm – as defined below; in Rudin’s
derivation of the metric (for a topological vector space setting, Rud1973) this
norm is explicit.
We say that the group X is normed if it has a group-norm as defined below
(cf. DezDD2006).

86
 6.1 Introduction 87

Definition We say that k. k : X → R+ is a group-norm if the following

properties hold:

(i) Subadditivity (triangle inequality): k x yk ≤ k xk + k yk;

(ii) Positivity: k xk > 0 unless x = e;
(iii) Inversion (symmetry): k x −1 k = k xk.
If (i) holds we speak of a group semi-norm; if (i) and (iii) and kek = 0 hold
we speak of a pseudo-norm (cf. Pet1950); if (i) and (ii) hold we speak of a
group pre-norm (see Low1997 for a full vocabulary).
We say that a group pre-norm, and so also a group-norm, is abelian, or more
precisely cyclically permutable, if

(iv) Abelian norm (cyclic permutation): k xyk = k yxk for all x, y.

Proposition 6.1.1 If k · k is a group-norm on X, then
d(x, y) = d R (x, y) = d RX (x, y) := k xy −1 k
is a right-invariant metric; equivalently,
˜ y)) = d L (x, y) = d X (x, y) := d(x −1, y −1 ) = k x −1 yk
d(x, L

is the conjugate left-invariant metric on the group.

Conversely, if d(x, y) is a right-invariant metric, then
k xk := d(x, e) = d(x,
˜ e)

is a group-norm. Thus the metric k · k is bi-invariant if and only if k x y −1 k =

k x −1 yk = k y −1 xk, i.e. if and only if the group-norm is abelian.
Furthermore, for (X, k · k) a normed group, the inversion mapping x 7→ x −1
from (X, d) to (X, d) ˜ is an isometry and hence a homeomorphism.

Proof Given a group-norm put d(x, y) = k xy −1 k. Then k xy −1 k = 0 if and

only if x y −1 = e, i.e. if and only if x = y. Symmetry follows from inversion
as d(x, y) = k(xy −1 ) −1 k = k yx −1 k = d(y, x). Finally, d obeys the triangle
inequality, since
k xy −1 k = k xz −1 zy −1 k ≤ k xz −1 k + kzy −1 k.
As for the converse, given a right-invariant metric d, put k xk := d(e, x). Now
k xk = d(e, x) = 0 if and only if x = e. Next, k x −1 k = d(e, x −1 ) = d(x, e) =
k xk, and so
d(xy, e) = d(x, y −1 ) ≤ d(x, e) + d(e, y −1 ) = k xk + k yk.
Also d(xa, ya) = k xaa−1 y −1 k = d(x, y).
88 Groups and Norms: Birkhoff–Kakutani Theorem

The metric d is bi-invariant if and only if d(e, yx −1 ) = d(x, y) = d(e, x −1 y)

if and only if k yx −1 k = k x −1 yk. Inverting the first term yields the abelian
property of the group-norm.
Finally, for (X, k · k) a normed group and with the notation d(x, y) = k x y −1 k
etc., the mapping x → x −1 from (X, d RX ) → (X, d LX ) is an isometry and so a
homeomorphism, as d L (x −1, y −1 ) = d R (x, y). 
The following result clarifies the relationship between the conjugate metrics
and the group structure. We define the ε-swelling of a set K in a metric space
X for a given (e.g. right-invariant) metric d X , to be
[
Bε (K ) := {z : d X (z, k) < ε for some k ∈ K } = Bε (k)
k ∈K
and for the conjugate (resp. left-invariant) case we can write similarly
B̃ε (K ) := {z : d˜X (z, k) < ε for some k ∈ K }.
We write Bε (x 0 ) for Bε ({x 0 }), so that

0 k < ε} = {wx 0 : w = zx 0 , kwk < ε} = Bε (e)x 0 .

Bε (x 0 ) := {z : kzx −1 −1

When x 0 = eX , the ball Bε (eX ) is the same under either of the conjugate
metrics, as
Bε (eX ) := {z : kzk < ε}.
Proposition 6.1.2 (i) In a locally compact group X, for K compact and for
ε > 0 small enough so that the closed ε-ball Bε (eX ) is compact, the swelling
Bε/2 (K ) is precompact.
(ii) Bε (K ) = {wk : k ∈ K, kwkX < ε} = Bε (eX )K, where the notation refers
to swellings for d X a right-invariant metric; similarly for d˜X , the conjugate
metric,
B̃ε (K ) = K Bε (eX ).
Proof (i) If x n ∈ Bε/2 (K ), then we may choose k n ∈ K with d(k n, x n ) < ε/2.
Without loss of generality k n converges to k. Thus there exists N such that, for
n > N, d(k n, k) < ε/2. For such n, we have d(x n, k) < ε. Thus the sequence
x n lies in the compact closed ε-ball centred at k and so has a convergent
subsequence.
(ii) Let d X (x, y) be a right-invariant metric, so that d X (x, y) = k x y −1 k. If
kwk < ε, then d X (wk, k) = d X (w, e) = kwk < ε, so wk ∈ Bε (K ). Conversely,
if ε > d X (z, k) = d X (zk −1, e), then, putting w = zk −1, we have z = wk ∈
Bε (K ). 
Theorem 6.1.3 (Invariance of Norm Theorem) (a) The group-norm is abelian
(and the metric is bi-invariant) if and only if
 6.1 Introduction 89

k x y(ab) −1 k ≤ k xa−1 k + k yb−1 k

for all x, y, a, b, or equivalently,
kuabvk ≤ kuvk + kabk
for all a, b, u, v.
(b) Hence a metric d on the group X is bi-invariant if and only if the Klee
property holds:
d(ab, x y) ≤ d(a, x) + d(b, y). (Klee)
In particular, this holds if the group X is itself abelian. (See Kle1952.)
(c) The group-norm is abelian if and only if the norm is preserved under
conjugacy (inner automorphisms).
Proof (a) If the group-norm is abelian, then by the triangle inequality,
k x yb−1 · a−1 k = ka−1 x yb−1 k ≤ ka−1 xk + k yb−1 k.
For the converse we demonstrate bi-invariance in the form kba−1 k = ka−1 bk.
In fact it suffices to show that k yx −1 k ≤ k x −1 yk; for then bi-invariance follows,
since taking x = a, y = b, we get kba−1 k ≤ ka−1 bk, whereas taking x = b−1 ,
y = a−1 we get the reverse ka−1 bk ≤ kba−1 k. As for the claim, we note that
k yx −1 k ≤ k yx −1 yy −1 k ≤ k yy −1 k + k x −1 yk = k x −1 yk.

(b) Klee’s result is deduced as follows. If d is a bi-invariant metric, then k · k

is abelian. Conversely, for d a metric, let k xk := d(e, x). Then k.k is a group-
norm, as
d(ee, x y) ≤ d(e, x) + d(e, y).
Hence d is right-invariant and d(u, v) = kuv −1 k. Now we conclude that the
group-norm is abelian since
k xy(ab) −1 k = d(x y, ab) ≤ d(x, a) + d(y, b) = k xa−1 k + k yb−1 k.
Hence d is also left-invariant.
(c) Suppose the norm is abelian. Then for any g, by the cyclic property
kg −1 bgk = kgg −1 bk = kbk. Conversely, if the norm is preserved under au-
tomorphism, then we have bi-invariance, since kba−1 k = ka−1 (ba−1 )ak =
ka−1 bk. 
Note that, taking b = v = e, we have the triangle inequality. Thus the result
(a) characterizes maps k.k with the positivity property as group pre-norms
which are abelian. In regard to conjugacy, see also the Uniformity Theorem
for Conjugation in BinO2010g, Th. 12.4. We now state the following classical
result.
90 Groups and Norms: Birkhoff–Kakutani Theorem

Theorem 6.1.4 (Normability Theorem for Groups – Birkhoff–Kakutani Theo-

rem) Let X be a first-countable topological group and let Vn be a symmetric
local base at eX with
4
Vn+1 ⊆ Vn .
Let r = ∞ −n be presented as a terminating representation of the
P
n=1 cn (r)2
dyadic number r, and put
Y
A(r) = Vn .
cn (r )=1

Then
p(x) := inf{r : x ∈ A(r)}
is a group-norm. If, further, X is locally compact and non-compact, then p may
be arranged such that p is unbounded on X, but bounded on compact sets.
For a detailed proof see that offered in Rud1973, for Th. 1.24 (pp. 18–
19), which derives a metrization of a topological vector space in the form
d(x, y) = p(x − y) and makes no use of commutativity or of the scalar field
(so note how symmetric neighbourhoods here replace the ‘balanced’ ones in
a topological vector space). That proof may be rewritten verbatim with x y −1
substituting for the additive notation x − y. We give a sketch account below in
association with a reformulation.
Theorem 6.1.5 (Birkhoff–Kakutani Normability Theorem) A first-countable
right topological group X is a normed group if and only if inversion and multi-
plication are continuous at the identity. In particular, a metrizable topological
group is normed.
Sketch Proof (Ost2013c) A normed group under its right norm topology is
a first-countable right topological group. It follows directly from the defining
three axioms above that inversion and multiplication are continuous at the iden-
tity e. For the converse, consider any open neighbourhood U of e. Continuity
of multiplication at e implies that there is an open neighbourhood V of e such
that V 4 ⊆ U. Next, using continuity of inversion at e, one may choose an
open neighbourhood N ⊆ V of e such that N −1 ⊆ V . Then, as right shifts are
homeomorphisms,
W := N N −1 = {Ng −1 : g ∈ N }
is an open neighbourhood of e with W −1 = W ⊆ V 2 . So, since W 2 ⊆ V 4 ⊆ U,
we conclude that for any open neighbourhood U of e one may choose an
open neighbourhood W of e with W −1 = W ⊆ W 2 ⊆ U. One may now
 6.1 Introduction 91

follow verbatim the argument in Kak1936. In broad outline: take a sequence of

neighbourhoods U1, U1/2, U1/4, . . . of e that is a basis at e with

n = U1/2 n ⊆ U1/2 n ⊆ U1/2 n−1 , . . .

−1 2
U1/2
for n ∈ N. Then construct a (Urysohn) function f : G → R+ such that f (x) ≤
2−n if and only if x ∈ U1/2n and define a metric by
d(g, h) := sup | f (gx) − f (hx)|.
x

This is a right-invariant metric on G compatible with the topology, as shown

by Kakutani (Kak1936). 
A group-norm defines two metrics: the right-invariant metric, which we
denote, as in Proposition 6.1.1, by d R (x, y) := k xy −1 k, and the conjugate left-
invariant metric, here to be denoted d L (x, y) := d R (x −1, y −1 ) = k x −1 yk. There
is correspondingly a right and a left metric topology, which we term the right or
left norm-topology. We write →R for convergence under d R , etc. Both metrics
give rise to the same norm, since d L (x, e) = d R (x −1, e) = d R (e, x) = k xk, and
hence define the same balls centred at the origin e:
BRd (e, r) := {x : d(e, x) < r } = BLd (e, r).
Denoting this commonly determined set by B(r), we have seen in Proposition
6.1.2 that
BR (a, r) = {x : x = ya and d R (a, x) = d R (e, y) < r } = B(r)a,
BL (a, r) = {x : x = ay and d L (a, x) = d L (e, y) < r } = aB(r).
Thus the open balls are right- or left-shifts of the norm balls at the origin. This
is best viewed in the current context as saying that under d R the right-shift
ρ a : x → xa is right uniformly continuous, since
d R (xa, ya) = d R (x, y),
and likewise that under d L the left-shift λ a : x → ax is left uniformly continu-
ous, since
d L (ax, ay) = d L (x, y).
In particular, under d R we have y →R b if and only if yb−1 →R e, as
d R (e, yb−1 ) = d R (y, b). Likewise, under d L , we have x →L a if and only
if a−1 x →L e, as d L (e, a−1 x) = d L (x, a).
Thus either topology is determined by the neighbourhoods of the identity
(origin) and according to choice makes the appropriately sided shift continuous;
said another way, the topology is determined by the neighbourhoods of the
92 Groups and Norms: Birkhoff–Kakutani Theorem

identity and the chosen shifts. We noted earlier that the triangle inequality
implies that multiplication is jointly continuous at the identity e, as a mapping
from (X, d R ) to (X, d R ). Likewise inversion is also continuous at the identity
by the symmetry axiom. (See Theorem 6.1.5.) To obtain similar results other
than at the identity one needs to have continuous conjugation, and this is linked
to the equivalence of the two norm topologies (see the Equivalence Theorem,
6.1.9). The conjugacy map under g ∈ G (inner automorphism) is defined by

γg (x) := gxg −1 .

Evidently the inverse of γg is given by conjugation under g −1 and that γg

is a homomorphism. Its continuity, as a mapping from (X, d R ) to (X, d R ), is
thus determined by behaviour at the identity, as we verify below. We work with
the right topology (under d R ), and may sometimes leave unsaid equivalent
assertions about the isometric case of (X, d L ) replacing (X, d R ).

Lemma 6.1.6 The homomorphism γg is right-to-right continuous at any point

if and only if it is right-to-right continuous at e.

Proof This is immediate since x →R a if and only if xa−1 →R e, and

γg (x) →R γg (a) if and only if γg (xa−1 ) →R γg (e) since

kgxg −1 (gag −1 ) −1 k = kgxa−1 g −1 k. 

Working under d R, we will relate inversion to left-shifts. We begin with the

following.

Lemma 6.1.7 If inversion is right-to-right continuous, then x →R a if and

only if a−1 x →R e.

Proof For x →R a, we have d R (e, a−1 x) = d R (x −1, a−1 ) → 0, assum-

ing continuity. Conversely, for a−1 x →R e we have d R (a−1 x, e) → 0, i.e.
d R (x −1, a−1 ) → 0. So since inversion is assumed to be right-continuous and
(x −1 ) −1 = x, etc., we have d R (x, a) → 0. 

We now expand this.

Theorem 6.1.8 The following are equivalent:

(i) inversion is right-to-right continuous;

(ii) left-open sets are right-open;
(iii) for each g, the conjugacy γg is right-to-right continuous at e; i.e. for every
ε > 0 there is δ > 0 such that

gB(δ)g −1 ⊂ B(ε);
 6.1 Introduction 93

(iv) left-shifts are right-continuous.

Proof We show that (i)⇐⇒(ii)⇐⇒(iii)⇐⇒(iv).

Assume (i). For any a and any ε > 0, by continuity of inversion at a, there
is δ > 0 such that, for x with d R (x, a) < δ, we have d R (x −1, a−1 ) < ε, i.e.
d L (x, a) < ε. Thus

B(δ)a = BR (a, δ) ⊂ BL (a, ε) = aB(ε), (incl)

i.e. left-open sets are right-open, giving (ii). For the converse, we just reverse
the last argument. Let ε > 0. As a ∈ BL (a, ε) and BL (a, ε) is left open, it is
right open and so there is δ > 0 such that

BR (a, δ) ⊂ BL (a, ε).

Thus for x with d R (x, a) < δ, we have d L (x, a) < ε, i.e. d R (x −1, a−1 ) < ε, i.e.
inversion is right-to-right continuous, giving (i).
To show that (ii)⇐⇒(iii) note that the inclusion (incl) is equivalent to

a−1 B(δ)a ⊂ B(ε),

i.e. to
γa−1 [B(δ)] ⊂ B(ε);

that is, to the assertion that γa−1 (x) is continuous at x = e (and so continuous,
by Lemma 6.1.6). The property (iv) is equivalent to (iii) since the right-shift is
right-continuous and γa (x)a = λ a (x) is equivalent to γa (x) = λ a (x)a−1 . 

We saw in the Birkhoff–Kakutani Theorem that metrizable topological

groups are normable (equivalently, have a right-invariant metric); we now
formulate a converse, showing when the right-invariant metric derived from a
group-norm equips its group with a topological group structure. As this is a
characterization of metric topological groups, we will henceforth refer to them
synonymously as normed topological groups. Below the condition (adm) refers
to admissible topologies.

Theorem 6.1.9 (Equivalence Theorem) A normed group is a topological

group under the right (resp. left) norm-topology if and only if each conjugacy

γg (x) := gxg −1

is right-to-right (resp. left-to-left) continuous at x = e (and so everywhere);

i.e. for z n →R e and any g
gz n g −1 →R e. (adm)
94 Groups and Norms: Birkhoff–Kakutani Theorem

Equivalently, it is a topological group if and only if left/right-shifts are contin-

uous for the right/left norm-topology, or if and only if the two norm topologies
are themselves equivalent.
In particular, if also the group structure is abelian, then the normed group
is a topological group.

Proof Only one direction needs proving. We work with the d R topology, the
right topology. By Theorem 6.1.8 we need only show that under it multiplication
is jointly right-continuous. First we note that multiplication is right-continuous
if and only if

d R (x y, ab) = k xyb−1 a−1 k, as (x, y) →R (a, b).

Here, we may write Y = yb−1 so that Y →R e if and only if y →R b, and we

obtain the equivalent condition

d R (xY b, ab) = d R (xY, a) = k xY a−1 k, as (x, Y ) →R (a, e).

Again by Theorem 6.1.8, as inversion is right-to-right continuous, the preceding

Lemma 6.1.6 justifies re-writing the second convergence condition with X =
a−1 x and X →R e, yielding the equivalent condition

d R (aXY b, ab) = d R (aXY, a) = kaXY a−1 k, as (X, Y ) →R (e, e).

But, by Lemma 6.1.6 again, this is equivalent to continuity of conjugacy. 

The final assertion is related to a result of Żelazko (Zel1960) (cf. Com1984).

We will later apply the Equivalence Theorem several times in conjunction with
the following result.

Lemma 6.1.10 (Weak Continuity Criterion) For fixed x, if, for all null se-
quences wn , we have γx (wn(k) ) → eX down some subsequence wn(k), then γx
is continuous.

Proof We are to show that for every ε > 0 there is δ > 0 and N such that, for
all n > N,
xB(δ)x −1 ⊂ B(ε).

Suppose not. Then there is ε > 0 such that, for each k = 1, 2, . . . and
each δ = 1/k, there is n = n(k) > k and wk with kwk k < 1/k and
k xwk x −1 k > ε. So wk → 0. By assumption, down some subsequence n(k) we
have k xwn(k) x −1 k → 0; but this contradicts k xwn(k) x −1 k > ε. 
 6.2 Analytic Shift Theorem 95

6.2 Analytic Shift Theorem

The main result of the section is a shift-compactness theorem which, thanks to
its restriction to analytic target sets, is applicable in normed groups that need
not be abelian. We will need some preliminary results. The first is a simple
and immediate corollary of the Analytic Cantor Theorem (Theorem 3.1.6) and
refers to the fact that Br (x) = Br (eX )x.
A Convergence Criterion In a normed group, for r n & 0 and α n = an · · · a1
with cl Brn+1 (an+1 ) ⊆ Brn (e)an , if X = K (I) is an analytic subset and
K (i 1, . . . , i n ) ∩ Brn (α n ) , ∅ for some i ∈ I and all n, then the sequence
{α n } is convergent.
Proof Indeed, α n → α, where {α} = K (i) ∩ n Fn for Fn = cl(Brn (α n )). 
T

Theorem 6.2.1 (Analytic Baire Theorem) In a normed group X under d RX ,

if X contains a non-meagre analytic set, then X is Baire and in fact, up to a
meagre set, X is analytic (and separable).
Proof Let A be non-meagre and analytic with upper semi-continuous repre-
sentation A = K (I). Let {Fn0 : n ∈ ω} be closed nowhere-dense subsets of
X and G an arbitrary non-empty open set. Say g ∈ G. As A is analytic, by
Nikodym’s Theorem, A has the Baire property, so A = U\N ∪ M for some
open U and meagre M, N. As A is non-meagre, U is non-meagre and so non-
empty. Say u ∈ U\N . Since each mapping ρt (x) = xt is a homeomorphism,
H := Gg −1 u is open and meets U in u ∈ A. As G is arbitrary, it suffices to
show that H meets X\ m∈ω Fm0 in a point of A. (Here we are using the fact
S
that a normed group which is locally Baire at some point is Baire – for a more
general result along these lines, see TopH1980, Prop. 2.2.3.)
We choose inductively integers i n, points x n , radii r n > 0 with r n+1 < r n /2
and nowhere dense closed sets {Fmn : m ∈ ω} such that K (i 1, . . . , i n ) ∩ B(x n, r n )
is non-meagre with
B(x n, r n ) ⊆ B(x n−1, r n−1 ) ⊆ H,
and
/[
K (i 1, . . . , i n ) ⊇ B(x n, r n ) Fn and
m∈ω m
[
B(x n, r n ) ∩ Fmk = ∅.
k,m<n

Begin by taking x 0 = eX and selecting r 0 arbitrarily so that B(x 0, r 0 ) ∩ K (I) is

non-meagre. To verify the inductive step, note that
[
K (i 1, . . . , i n ) ∩ B(x n, r n ) = K (i 1, . . . , i n, m) ∩ B(x n, r n ),
m∈ω
96 Groups and Norms: Birkhoff–Kakutani Theorem

so there is i n+1 such that K (i 1, . . . , i n, i n+1 ) ∩ B(x n, r n ) is non-meagre. So,

(again by Nikodym’s Theorem) for some non-meagre open V, closed nowhere
dense sets Fmn+1 and meagre Nn+1, we have
[
K (i 1, . . . , i n, i n+1 ) ∩ B(x n, r n )\Nn+1 = V \ Fmn+1 .
m∈ω
By analyticity, V is hereditarily separable, so we may choose x n+1 , r n+1 (with
r n+1 < r n /2 ) such that x n+1 ∈ cl(B(x n+1, r n+1 )) ⊆ V ∩ B(x n, r n )\ m<n+1 Fmk
S
and
K (i 1, . . . , i n+1 ) ∩ B(x n+1, r n+1 )
is non-meagre. With the induction verified, by the Convergence Criterion above
(or directly from the Analytic Cantor Theorem, 3.1.6), there is a such that
{a} = K (i) ∩ n∈ω B(x n, r n ) ⊆ H. So a ∈ H (and a = limn x n ). Furthermore,
T
for any n, we have B(x n+1, r n+1 ) ∩ Fn0 = ∅, so a < m∈ω Fm0 .
S
As for the second claim, recall that A = U\N ∪ M with U non-empty and
N, M meagre with N ⊆ U and U ∩ M = ∅. Suppose that B := Bε (a) ⊆ U;
then B ∩ A = B\N and so B\N is analytic. Now {Bx : x ∈} is an open
cover of X, so has a σ-discrete refinement, say V := n∈ω Vn with Vn :=
S
{Vtn : t ∈ Tn } discrete. Suppose that Vtn ⊆ Bx tn . Put Ntn := (N ∩ B)x tn,
which is meagre (since right-shifts are homeomorphisms); then Vtn \Ntn ⊆
(B\N )x tn ⊆ Ax tn . Without loss of generality Ntn is a meagre Fσ subset of
Vtn and so Vtn \Ntn = (Vtn ∩ ( Ax tn ))\Ntn is analytic and non-meagre. By
Banach’s Category Theorem, 2.2.1, N 0 := n∈ω {Ntn : t ∈ Tn } is meagre,
S S
since Vn is discrete. By a theorem of Montgomery (Mon1935), for each n the
S S
set {Vtn \Ntn : t ∈ Tn } is analytic, being locally analytic on {Vtn : t ∈ Tn },
and is non-meagre. As V covers X, we have
[ [
X\N 0 = {Vtn \Ntn : t ∈ Tn },
n∈ω
and so X\N 0 is analytic, being a countable union of analytic sets, and non-
meagre. 
Lemma 6.2.2 (Displacements Lemma – Baire Case) Working under d RX ,
for X a Baire space, A Baire non-meagre and for almost all a ∈ A, there is
ε = ε A (a) > 0 such that A∩ Aa−1 xa is non-meagre for all x with k xk < ε A (a).
Indeed, for any x with k xk < ε A (a), modulo meagre sets, Aa−1 ∩ Aa−1 x ⊇
B(x, s(x)) for some s(x) > 0, and so A ∩ Aa−1 xa ⊇ B(xa, s(x)).
Proof Omitting a meagre subset of A we may assume without loss of gen-
erality that e ∈ Aa−1 = U\N for some U open and N meagre. So, as
Aa−1 is non-meagre, there is ε > 0 such that B(e, ε) ⊆ Aa−1 modulo mea-
gre sets. For any x with k xk < ε and s := min{ε, 21 (ε − k xk)} > 0, we
 6.2 Analytic Shift Theorem 97

have B(e, ε) ⊇ B(x, s) = B(e, s)x. Indeed, if d R (y, x) < s, then d(y, e) <
d(y, x) + d(x, e) < s + ε = 21 (ε − k xk) + k xk < ε. Modulo meagre sets: as
Aa−1 x ⊇ B(e, s)x, we have

Aa−1 ⊇ Aa−1 ∩ Aa−1 x ⊇ B(e, ε) ∩ B(e, s)x ⊇ B(x, s),

which is non-meagre as the space X is Baire. See also Corollary 4.2.3 

Theorem 6.2.3 (Analytic Shift Theorem) In a normed group under the topol-
ogy d RX , for z n → eX null, A a K -analytic and non-meagre subset: for a
non-meagre set of a ∈ A with co-meagre Baire envelope, there is an infinite set
Ma and points an ∈ A converging to a such that
−1
{aam z m am : m ∈ Ma } ⊆ A.

In particular, if the normed group is topological, for quasi all a ∈ A, there is

an infinite set Ma such that {az m : m ∈ Ma } ⊆ A.
−1 z a converges to a; indeed,
Remark Notice that in the context above aam m m
−1
d RX (aam z m am, a) = kaam
−1
z m am a−1 k ≤ kaam
−1
k + kz m k + kam a−1 k.

Proof By Theorem 6.2.1 (Analytic Baire Theorem), we may assume that all
non-empty open sets are Baire and so non-meagre. Next, note that any non-
meagre Baire set is equal modulo meagre sets to a non-meagre Gδ . So if the
Baire envelope of the set of t ∈ A with the asserted property has non-meagre
complement, we may assume that this complement is analytic. It thus suffices
to prove that a non-meagre analytic set A contains at least one point t for which
there exists an infinite set Mt and points an ∈ A converging to t such that
{tam−1 z a : m ∈ M } ⊆ A.
m m t
We proceed to prove this. So write A = K (I) with K upper semi-continuous
and single-valued. For each n ∈ ω we find inductively integers i n, points x n ,
yn , an with an ∈ A, numbers r n > 0, s n > 0, analytic subsets An of A, and
closed nowhere dense sets {Fmn : m ∈ ω} such that
[
K (i 1, . . . , i n ) ⊇ An ∩ ( An an−1 x n an ) ⊇ B(yn, s n )\ Fmn
m∈ω

and
[
yn ∈ B(x n an, r n ) and B(yn, s n ) ∩ Fmk = ∅.
m,k<n

Assuming this done for n, since K (i 1, . . . , i n ) = k K (i 1, . . . , i n, k) is non-

S
meagre, there is i n+1 such that K (i 1, . . . , i n, i n+1 ) ∩ An ∩ ( An an−1 x n an ) is non-
meagre. Put

An+1 := K (i 1, . . . , i n, i n+1 ) ∩ An ∩ ( An an−1 x n an ) ⊆ K (i 1, . . . , i n+1 ).

98 Groups and Norms: Birkhoff–Kakutani Theorem

As An is non-meagre, we may pick an+1 ∈ An+1 as in Lemma 6.2.2 (Displace-

ments Lemma), and m(n) so large that kz m k < ε(an+1, An+1 ) for m ≥ m(n).
Pick x n+1 = z m(n)
−1 . Then, since A
n+1 ⊆ K (i 1, . . . , i n+1 ) and in view of Lemma
6.2.2, there is r n+1 and closed nowhere dense sets {Fmn+1 : m ∈ ω} such that
K (i 1, . . . , i n+1 ) ⊇ An+1 ∩ An+1 an+1−1
x n+1 an+1
[
⊇ B(x n+1 an+1, r n+1 ) F n+1 .
m∈ω m

Since the set m,k<n+1 Fmk is closed and nowhere dense, there is yn+1 ∈
S
B(x n+1 an+1, r n+1 ) and s n+1 > 0 so small that B(yn+1, s n+1 ) ⊆ B(x n+1 an+1, r n+1 )
and B(yn+1, s n+1 ) ∩ m,k<n+1 Fmk = ∅. Hence
S

[ [
B(x n+1 an+1, r n+1 ) Fmn+1 ⊇ B(yn+1, s n+1 ) Fmn+1 .
m∈ω m∈ω
By Theorem 3.1.6 (Analytic Cantor Theorem), there is t with
\ \
{t} = K (i) ∩ B(yn, s n ) ⊆ An ∩ ( An an−1 x n an ).
n n
So t ∈ A. Fix n. One has t < m∈ω Fmn (since B(ym+1, s m+1 ) ∩ k<m Fkn = ∅
S S
for each m), and so
[
t ∈ B(yn, s n ) Fmn ⊆ An ∩ ( An an−1 x n an ) ⊆ K (i 1, . . . , i n ) ⊆ A.
m∈ω

As t ∈ An an−1 x n an, n an = tan z m(n) an ∈ An ⊆ A. So

one has tan−1 x −1 −1

{tan z m(n) an : n ∈ ω} ⊆ A. Moreover, d R (x n an, t) = d R (x n, tan−1 ) → 0,

−1

so since x n → e, we have tan−1 → e, i.e. an →R t. 

6.3 The ‘Squared Pettis Theorem’

Theorem 6.3.1 In a normed group X, for T ⊆ X almost complete, U open
with T ∩ U non-meagre, and z n → eX , the set SU of t ∈ T ∩ U for which there
exist points t m ∈ T with t m →R t and an infinite Mt with
{tt −1
m z m t m : m ∈ Mt } ⊆ T,

is non-meagre.
Proof Suppose not; then there is an open set U such that SU is meagre. Letting
H be a meagre Fσ cover of SU , the set T 0 := (T\H) ∩ U is Baire and non-
meagre. But then by the Analytic Shift Theorem (Theorem 6.2.3) there exists
points t, t m ∈ T 0 and infinite set Mt such that
{tt −1 0
m z m t m : m ∈ Mt } ⊆ T ⊆ T ∩ U,

a contradiction. 
 6.4 Shifted-Covering Compactness 99

Our next theorem is named after a simpler Baire category result (without the
square), due to Pettis (Pet1950), true in simpler circumstances. See Chapter 15,
which is devoted to interior-point theorems.

Theorem 6.3.2 (Squared Pettis Theorem) Let X be a K -analytic (e.g. topo-

logically complete) normed group and A in X Baire non-meagre under the
right norm-topology. Then eX is an interior point of ( AA−1 ) 2 .

Proof Suppose not. We may assume that A is analytic. Otherwise, since A

is Baire, write A = (U\M) ∪ N with M an Fσ , and M, N both meagre. Then
U\M is analytic and non-meagre. Now we may select z n ∈ B1/n (e)\( AA−1 ) 2 .
As z n → e, we apply Theorem 6.2.3 (Analytic Shift Theorem) to A, to find
t ∈ A, Mt infinite and t m ∈ A for m ∈ Mt such that tt −1
m z m t m ∈ A for all
m ∈ Mt . So, for m ∈ Mt ,

z m ∈ AA−1 AA−1 = ( AA−1 ) 2,

a contradiction. 

6.4 Shifted-Covering Compactness

The following two theorems assert that a ‘covering property modulo shift’ is sat-
isfied by bounded (right) shift-compact sets (BinO2010g). It will be convenient
to make the following definitions.

Definitions 1. Say that D := {D1, . . . , Dh } shift-covers a subset X of G or is

a shifted-cover of X if, for some d 1, . . . , d h in G

X = D1 d 1 ∪ . . . ∪ Dh d h .

Say that X is compactly shift-covered if every open cover U of X contains

a finite subfamily D which shift-covers X.
2. For N a neighbourhood of eG say that D := {D1, . . . , Dh } N-strongly
shift-covers a subset A of G or is an N-strong shifted-cover of A if, for some
d 1, . . . , d h in N
A ⊆ (D1 − d 1 ) ∪ · · · ∪ (Dh − d h ).

Say that A is compactly strongly shift-covered, or compactly shift-covered

with arbitrarily small shifts, if every open cover U of A contains for each
neighbourhood N of eG a finite subfamily D which N-strongly shift-covers A
(BinO2011a).
100 Groups and Norms: Birkhoff–Kakutani Theorem

Theorem 6.4.1 Let A be a shift-compact subset of a separable normed topo-

logical group G. Then A is compactly shift-covered, i.e. for any norm-open
cover U of A, there is a finite subset V of U , and for each member of V a
translator, such that the corresponding translates of V cover A.
Proof Let U be an open cover of A. Since G is second-countable we may
assume that U is a countable family. Write U = {Ui : i ∈ ω}. Let Q = {q j :
j ∈ ω} enumerate a dense subset of G. Suppose, contrary to the assertion, that
there is no finite subset V of U such that elements of V, translated each by a
corresponding member of Q, cover A. For each n, choose an ∈ A not covered
by {Ui qi : i, j < n}. As A is precompact, so we may assume, by passing to a
subsequence (if necessary), that an converges to some point a0, and also that,
for some t, the sequence an t lies entirely in A. Let Ui in U cover a0 t. Without
loss of generality we may assume that an t ∈ Ui for all n. So an ∈ Ui t −1
for all n. Thus we may select V := Ui q j to be a translation of Ui such that
an ∈ V = Ui q j for all n. But this is a contradiction, since an is not covered by
{Ui0 q j 0 : i 0, j 0 < n} for n > max{i, j}. 
The above proof may be improved to strong shift-covering, with only a minor
modification (replacing Q with a set Q ε = {q εj : j ∈ ω} which enumerates, for
given ε > 0, a dense subset of the ε ball about e), yielding the following.
Theorem 6.4.2 Let A be a strongly shift-compact subset of a separable
normed topological group G. Then A is compactly strongly shift-covered, i.e.
for any norm-open cover U of A, and any neighbourhood of eG , there is a finite
subset V of U , and, for each member of V, a translator in N such that the
corresponding translates of V cover A.
 7

Density Topology

The first result below is the Lebesgue Density Theorem, 7.1.1. Closely related
to this is the density topology, D, the subject of this chapter. As we shall see,
D and the Euclidean topology E enable us to work bitopologically, with D
playing for the measure case the role played by E for the category case. This
enables us to align as closely as possible qualitative measure properties with
their Baire analogues. This is the theme of Chapter 9, on category–measure
duality.

7.1 The Lebesgue Density Theorem

The density topology has a number of properties that will be crucially useful for
us. However, it lacks some properties commonly regarded as desirable. From
our point of view, the density topology is a wonderful topology (we would even
say a beautiful one). We quote in this connection from page 1 of LukMZ1986:
‘It was proved that the “nice” system of all approximately continuous functions
is exactly the class of all continuous functions in the “bad” density topology.’
We, like the authors of LukMZ1986, wish to rehabilitate the density topology.
Let B denote a countable basis of open neighbourhoods. For any set T put

Bα (T ) := {I ∈ B : |I ∩ T | ∗ > α|I |}

(with |. | ∗ for outer Lebesgue measure), which is countable, and

\ [
F (T ) := {I : I ∈ Bα (T )}.
α∈Q∩(0,1)

Thus F is increasing in T, F (T ) is measurable (even if T is not), and x ∈ F (T )

if and only if x is a density point of T .
The following is one form of the Lebesgue Density Theorem.

101
102 Density Topology

Theorem 7.1.1 (Density Theorem) If K is compact and non-null, then K has

a density point. Hence almost all points of a measurable set are density points.

Before beginning the proof, we give a definition, a lemma and a corollary.

Working in R, for given K compact, put

α(x) = α K (x) := lim sup |I ∩ K |/|I | (x ∈ K ),

x ∈I ∈B, |I |→0

where K, if omitted, is to be understood from context.

Lemma 7.1.2 For I ∈ B, if α(. ) is bounded away from 1 on I ∩ K, then

|I ∩ K | = 0.

Proof Suppose not, and that α(x) ≤ A < 1 on I ∩ K . Take k > |I ∩ K |.

By the outer regularity of |. |, there is a sequence In ∈ B with In ⊂ I such
that I ∩ K ⊂ n In and n |In | < k. We may assume that each In ∩ K , ∅
S P
(discarding any others). Thus if x ∈ In ∩ K, we have α(x) ≤ A, and so

|In ∩ K | ≤ A|In |.

So
X X
|I ∩ K | ≤ |In ∩ K | ≤ A |In | ≤ Ak.
n n

Letting k & |I∩K | gives |I∩K | ≤ A|I∩K |. As A < 1, this gives |I∩K | = 0. 

Corollary 7.1.3 If K is compact and |K | > 0, then there is I0 ∈ B such that

α(. ) is not bounded away from 1 on K ∩ I0 and |K ∩ I0 | > 0.

Proof Suppose otherwise; then for each I ∈ B either |K ∩ I | = 0 or α(. ) is

bounded away from 1 on K ∩ I for each I ∈ B. For each k ∈ K choose Ik ∈ B
with k ∈ Ik (possible as B is a basis). By assumption, either |K ∩ Ik | = 0
or α(. ) is bounded away from 1 on K ∩ Ik . Either way |K ∩ Ik | = 0 (by the
lemma). By compactness, there is a finite set F ⊆ K such that {Ik : k ∈ F}
covers K . Then |K | ≤ k ∈F |K ∩ Ik | = 0, so |K | = 0, a contradiction.
P


Proof of Theorem 7.1.1 Suppose otherwise. Then for each x ∈ K there is

α = α(x) with 0 < α < 1 such that for I ∈ B, if x ∈ I, then

|I ∩ K | ≤ α|I |.

For K with |K | > 0, we construct a nested sequence hIn i of non-empty sets

in B whose intersection contains a density point of K . Pick I0 ∈ B such that
α(. ) is not bounded away from 1 on K ∩ I0 and |K ∩ I0 | > 0. By passing to a
 7.1 The Lebesgue Density Theorem 103

subinterval J with |K ∩ J | > 0 on which α(. ) is not bounded away from 1, we

claim that we may without loss of generality assume also that
1
|K ∩ I0 | ≥ |I0 |.
2
Otherwise, for each I 0 ⊂ I0 with I 0 ∈ B and |K ∩ I 0 | > 0, we would have
1 0
|K ∩ I 0 | <
|I |,
2
implying that α(x) < 1/2 on K ∩ I0, and so, by the lemma, that K ∩ I0 is null.
As K ∩ I0 is precompact, we may choose I1 ∈ B with I¯1 ⊂ I0 and diam(I1 ) <
diam(I0 )/2 such that α(. ) is not bounded away from 1 on K ∩ I1 and such that
3
|K ∩ I1 | ≥ |I1 |.
4
Continue inductively. Thus, if
\
{x} = K ∩ I¯n,
n∈ω

then x ∈ In for n = 1, 2, . . . and

|K ∩ In | ≥ (1 − 2−n )|In |.
Thus x ∈ F (K ), a contradiction. Thus after all K has a density point.
We now prove the final assertion. Let T be a measurable set of positive
measure. In the notation above, we are to prove that S := T\F (T ) ⊆ T has
measure zero. Suppose otherwise. Then we may choose K ⊆ S compact with
|K | > 0 (by inner regularity of Lebesgue measure). Note that S is disjoint from
F (T ). We have just shown that K has a density point k, i.e. for all 0 < α < 1,
there is I with k ∈ I ∈ B such that
|I ∩ K | > α|I |.
But k ∈ K ⊆ S ⊆ T, so for all 0 < α < 1, there is I with k ∈ I ∈ B such that
|I ∩ T | ≥ |I ∩ S| ≥ |I ∩ K | > α|I |,
i.e. k is a density point of T, equivalently k ∈ F (T ). This contradicts disjointness
of S from F (T ). 
The last step is an instance of the Generic Dichotomy Principle of Chapter 4.
Lebesgue proved his Density Theorem as the special case for indicator func-
tions f = I A of his version of the fundamental theorem of calculus for Lebesgue
integrals: that if one takes the indefinite integral of a Lebesgue-measurable
function and differentiates it, one recovers the original function almost every-
where. For this extra degree of generality, the Vitali covering theorem and
104 Density Topology

Hardy–Littlewood maximal function are relevant; for textbook expositions, see

Bog2007a, Ch. 5; Saks1964, IV.10; Ste1970, pp. 5, 12. See also BinO2018b;
Bru1971.

Corollary 7.1.4 For A measurable, write φ( A) for the set of all density points
of A. Then the symmetric difference of A and φ( A) is (Lebesgue-)null:

| A∆φ( A)| = 0.

Proof First, A \ φ( A) is (Lebesgue-)null, by the Lebesgue Density Theorem.

Also φ( A) \ A ⊆ Ac \ φ( Ac ), which as Ac is measurable is similarly null.
Combining, the symmetric difference is null. 

For A, B measurable, write A ∼ B if and only if | A∆B| = 0. Then ∼ is an

equivalence relation on the measurable sets. We note some properties (Oxt1980,
Th. 3.21).

(i) (Quasi-identity) φ( A) ∼ A. This is the previous corollary.

(ii) (Neglecting property) A ∼ B implies φ( A) ∼ φ(B); φ(∅) ∼ ∅, φ(R) ∼ R.
This follows from the definition of φ(. ).
(iii) (Multiplicative property) φ( A ∩ B) = φ( A) ∩ φ(B).
For, if I is an interval, I \ ( A∩ B) = (I \ A) ∪ (I \ B). So |I | − |I ∩ A∩ B| ≤
|I | − |I ∩ A| + |I | − |I ∩ B|, or |I ∩ A| + |I ∩ B| − |I | ≤ |I ∩ A ∩ B|. Take
I = [x − h, x + h], divide by h and let h ↓ 0: if x is a density point of both
A and B, the limit of the left is 1, whence x is a density point of A ∩ B.
(iv) (Monotonicity) If A ⊆ B, then φ( A) ⊆ φ(B).
This follows from (iii).

One can define a mapping φ with these properties in more general measure
spaces. Such a mapping is called a lower density; see Chapter 8. The existence
of lower densities was proved by von Neumann and Maharam (Mah1958) and
is connected with the Ionescu Tulcea theory of lifting (IonT1961; IonT1969);
see Oxt1980, Th. 22.4.
In what follows (X, A, m) is a measure space and m∗ is an outer measure,
that is a non-negative countably subadditive set function defined on all subsets
of X (see, e.g., Bog2007a, §1.11: the value +∞ for m∗ is allowed). Typically,
m∗ will arise from the measure m via
∞
X [∞ 
m∗ ( A) := inf m( An ) : An ∈ A, A ⊆ An .
n=1 n=1

One calls a set A m-measurable if, for all  > 0, there is a set A ∈ A
with the symmetric difference A∆A having m∗ ( A∆A ) < . Equivalently, the
 7.1 The Lebesgue Density Theorem 105

measurable sets A are those that can be approximated from within and without
by sets Ai , Ao ∈ A:
Ai ⊆ A ⊆ Ao, m∗ ( Ao \ Ai ) = 0.
Then Ai can be taken in A σδ and Ao in A δσ . We call Ao the measurable cover
(or measurable envelope) of A; Ai the measurable core (or measurable kernel)
of A.
We shall be dealing principally here with Lebesgue measure, which is infinite.
However, our main interest here is in local aspects, for which we may by
truncation restrict to some compact set, on which Lebesgue measure is finite.
On such a finite measure space, matters simplify to
m∗ ( A) = inf{m(M) : A ⊆ M ∈ A}.
Call B a density basis (cf. derivation basis, HayP1970, differentiation basis,
LukMZ1986) if, for each x, there is a sequence hIn i with x ∈ In ∈ B such that
m(In ) & 0. Write B(x) for the collection of such sequences hIn i in B. The
corresponding upper outer density is defined by
D̄∗ ( A, x) = sup{lim supn m∗ (E ∩ In )/m(In ) : hIn i ∈ B(x)}.
A lower outer density is defined similarly. If they are equal we speak of a density.
We note in passing that, for fixed x, the set function D̄∗ ( A, x) is monotone and
subadditive (as with the outer measure).
Call x an outer density point of A if the outer density is 1, and an outer
dispersion point of A if the outer density is zero. The word outer is omitted
when A is measurable.
Definition For B a density basis and D̄∗ the corresponding upper outer density
function, put
U := {U : (∀u ∈ U) D̄∗ (X\U, u) = 0}.
Theorem 7.1.5 (Marti1964, Th. 4.1) For B a density basis, U as above is a
topology on X .

Proof This follows from subadditivity of D̄∗ ( A, x) for fixed x. In particular if

U1 and U2 are in U and U = U1 ∩ U2, then for u ∈ U, since U ⊆ Ui one has
D̄∗ (X\U, u) = D̄∗ ((X\U1 ) ∪ (X\U2 ), u)
≤ D̄∗ ((X\U1 ), u) + D̄∗ ((X\U2 ), u) = 0.
So U ∈ U. If {Ui : i ∈ I} is any subfamily of U and u ∈ U = I Ui, then for
S
some i ∈ I we have u ∈ Ui and so
D̄∗ (X\U, u) ≤ D̄∗ (X\Ui, u) = 0,
so U ∈ U. Evidently X and ∅ are in U. 
106 Density Topology

The main result is the measurability of the sets in U. We need a lemma and
a corollary of it.

Lemma 7.1.6 (Cf. Marti1964, 3.6) If E is a measurable cover of A and J is

measurable, then E ∩ J is a measurable cover of A ∩ J.

Proof If not, then there is F with A ∩ J ⊆ F ⊆ E ∩ J with m(F) = m∗ ( A ∩ J)

and m(F) < m(E ∩ J). But F ∪ (E\J) is measurable, covers A and satisfies

m∗ ( A) ≤ m(F ∪ (E\J)) = m(F) + m(E\J)

< m(E ∩ J) + m(E\J) = m(E),

contradicting m∗ ( A) = m(E). 

Corollary 7.1.7 If E is a measurable cover of A, then every density point of

E is an outer density point of A.

Proof Suppose x is a density point of E. For any α < 1 there is I ∈ B(x)

such that m(E ∩ I) > αm(I). So m∗ ( A ∩ I) = m(E ∩ I) > αm(I). Hence x is
an outer density point of A. 

Say that a density theorem holds for the density basis B if almost all points of
any set A are outer density points of A (equivalently, by the preceding corollary,
almost all points of any measurable set A are density points of A).

Theorem 7.1.8 (Marti1964, 4.3) If a density theorem holds for B, then a set
A is measurable if and only if almost all points in X\A are outer dispersion
points of A.

Proof If A is measurable, then so is X\A, and so by the assumed density

theorem almost all points of X\A are density points of X\A and hence dispersion
points of A, as asserted.
For the converse, let A have the asserted property and suppose that A is not
measurable. Let E be a measurable cover of A. Recall that a set F ⊆ E is
measurable if and only if

m(E) = m∗ (F) + m∗ (E\F).

As m(E) = m∗ ( A) and A is non-measurable we have m∗ (E\A) > 0. We next

show that almost all points of E\A are points of outer density of A, contradicting
the assumption that almost all points of E\A are outer dispersion points of A.
To this end note, by the assumed density theorem, that almost all points of
E are density points of E. By Corollary 7.1.7 all points of E are outer density
 7.1 The Lebesgue Density Theorem 107

points of A. Hence almost all points of E, and a fortiori almost all points of
E\A, are outer density points of A. 

Theorem 7.1.9 (Marti1964, 4.4. and 4.5) If a density theorem holds for B,
then each set U ∈ U is measurable. So in particular all points of U are density
points of U.

Proof By definition, if U ∈ U then all points in U = X\(X\U) are outer

dispersion points of X\U. By the preceding theorem, X\U is measurable and
so also is U. 

The motivating example in the material above is the class of intervals (x −

hn, x + hn ) containing a point x, with hn ↓ 0 rational (e.g. hn = 1/n). This
class is a density basis; that a density theorem holds for it is the content of the
Lebesgue Density Theorem. Call a set density open if all its points are points
of density. Then by Theorem 7.1.9, the class of such density open sets forms
a topology, the density topology D. The density topology is due to Haupt and
Pauc (HauP1952; Gof1962) (see also GofNN1961; GofW1961). However, it is
implicit in the work of Denjoy in 1914 (Den1915).
We shall need a number of properties of D.
(1) The density topology D is finer than the Euclidean topology E.
For, every point of a Euclidean neighbourhood is a density point. Thus every
Euclidean neighbourhood is a density neighbourhood.
The density topology is thus an example of a fine topology – a topology on
Euclidean space finer than the Euclidean topology – see Chapter 8.
(2) The D-interior of a measurable set M is φ(M), the set of density points of
M contained in M.
For, by the Lebesgue Density Theorem M \ φ(M) is null, so φ(M) is mea-
surable as M is, and each of its points is a density point of M by definition of
φ. So φ(M) is D-open. But any D-open subset of M is contained in φ(M), by
definition of D. So:
(20) If x ∈ M and M is measurable, M is a D-neighbourhood of x if and only
if x is a density point of M, i.e. x ∈ φ(M).
(3) The D-nowhere dense sets are the Lebesgue-null sets.
For, if A is Lebesgue-null, all points of its complement Ac are density points
of Ac , so Ac is D-open, so A is D-closed. Being null, A = A has empty interior.
So A is nowhere dense. Conversely, if A is D-nowhere dense, its closure A has
empty interior, so contains no density point of A. So A is null by the Lebesgue
Density Theorem.
(4) The D-meagre sets are the Lebesgue-null sets.
108 Density Topology

For, meagre sets are countable unions of nowhere dense sets (i.e. of null sets
by above) and so are null.
(5) (R, D) is a Baire space.
For, a countable union of D-nowhere dense sets, that is of null sets by above,
is null and so D-meagre. More is true: (R, D) is hereditarily Baire: any subset
of the real line is Baire under the induced topology (the proof is as above:
subsets of null sets are null).
(6) The D-Borel sets are the (Lebesgue-)measurable sets. In fact these are all
density-Gδ .
For, by outer regularity of Lebesgue measure, any measurable set is a (Eu-
clidean, so density) Gδ set less a null set. The null sets are closed in the D
topology (by the density theorem) and so any Lebesgue-measurable set is Dδ
and so D-Borel. Conversely, any D set is measurable, and so the D-Borel sets
are measurable (since the measurable sets are a σ-algebra).
This result is the D-analogue of regularity of Lebesgue measure, and so may
be regarded as the density analogue of Littlewood’s First Principle (§1.2).

Definition For a family H of subsets of X, a topology T on X is said to

have the H -insertion property if, for T -open U and T -closed F ⊇ U, there is
H ∈ H with
U ⊆ H ⊆ F.

Thus D has the ‘density-Gδ -insertion property’, or ‘Gδ (D)-insertion prop-

erty’. Since a topology is the class of its open sets, we can abbreviate this
to:
(60) D has the Dδ -insertion property.
The close relationship between Euclidean Gδ -subsets and the density topol-
ogy clarifies the connection between the algebraic and the two topological
structures in play here. We remind the reader that a topological group is a
topological space endowed with a group structure such that the group opera-
tions (x, y) → x y and x → x −1 are continuous. Where there may be separate
but not joint continuity of the group operation (x, y) → x y, we speak of a
paratopological group.
(7) (R, D) under addition is a paratopological group but not a topological
group.
Separate continuity of addition follows from the commutativity of addition
and shift-invariance of the density topology. However, joint continuity fails.
Suppose otherwise – i.e. that f (x, y) = x + y is jointly continuous as a map
from (R, D) 2 → (R, D). For any G non-empty D-open, e.g. an interval, the
 7.1 The Lebesgue Density Theorem 109

set G\Q is non-empty D-open with no rational elements. Then f −1 (G\Q)

is non-empty D-open and so contains a non-empty set of the form U × V
with U, V D-open and of positive measure. By inner regularity of Lebesgue
measure, these two sets have measurable kernels that are Euclidean Gδ -subsets,
say HU and HV . By Steinhaus’s Theorem (see Chapter 15), HU + HV contains
an interval and so rational points. But f (HU × HV ) ⊆ G\Q, a contradiction.
(Compare Scheinberg, Sch1971, Arhangelskii and Reznichenko, ArhR2005, or
Heath and Poerio. 1 )
When, as here, our interest is on local aspects, one may localize onto a
compact set on which Lebesgue measure is finite. One says that a complete,
finite measure µ is a category measure if the µ-null sets are the meagre sets.
The term is due to Oxtoby (Oxt1980, Ch. 23).
(8) Lebesgue measure on, say, ([0, 1], D) is a category measure.
Call a function f defined on a Euclidean neighbourhood of x approximately
continuous if there is a measurable set M with x a density point of M and
f (y) → f (x) as y → x with y ∈ M (the term is due to A. Denjoy in 1915).
(9) The function f is approximately continuous at x if and only if f is
D-continuous at x.
For, approximate continuity at x implies D-continuity at x, by (3). The con-
verse, due to GofNN1961, uses the Lusin–Menchoff Theorem (see LukMZ1986,
Ch. 3 and §6A and also Theorem 7.1.11).
(90) The density topology is thus the coarsest topology on the line making
the approximately continuous functions (in Denjoy’s sense above) continuous.
In consequence, the density topology is completely regular (cf. LukMZ1986,
Th. 2.3).
Thus the density topology plays for the approximately continuous functions
the role played by the fine topology for the superharmonic functions (see
Chapter 8).
(10) The function f is D-continuous a.e. if and only if it has the Baire property
with respect to D if and only if it is measurable. This is the Denjoy–Stepanoff
theorem (Denjoy in 1915, W. Stepanoff in 1942; see, e.g., LukMZ1986, 6.20).
The last two properties correspond roughly to Littlewood’s Second Principle.
Recall (§1.2) that if T is submetrizable, then T is finer than (i.e. refines) a
metrizable topology, Tρ say, with ρ a metric generating this, so that T ⊇ T ρ .
This has two consequences:

1 Paper at Conference on Topology and Theoretical Computer Science in honour of Peter

Collins and Mike Reed, 2006, unpublished.
110 Density Topology

(i) any Tρ -closed set (being the complement of a Tρ -open set which is also
T -open) is T -closed;
(ii) cl T A ⊆ clρ A, as there are more open sets.

Definition A submetrizable topology T , say, refining a metrizable topology

Tρ from a metric ρ, is cometrizable relative to Tρ , if for each x and every
T -neighbourhood U there is a T -neighbourhood V such that

x ∈ V ⊆ clρ V ⊆ U.

That is, the Tρ -closures of T -neighbourhoods form a T -neighbourhood-base.

Since cl T V ⊆ clρ V this property implies regularity.

Definition The topology T is said to have the Lusin–Menchoff property

relative to a topology τ (cf. LukMZ1986, Th. 3.11, p. 85) if for each τ-closed
set H and every T -neighbourhood U of H there is a T -neighbourhood V of
H such that
H ⊆ V ⊆ clτ V ⊆ U.

Thus the separation required here concerns some but not all T -closed sets.
It certainly includes all singletons as τ is Hausdorff, and so the definition is
more demanding than in the cometrizable case. It is a normality-like property
(termed binormality in the bitopology literature).
Equivalently, this may be restated with T 0 for τ as requiring that if F, F 0
disjoint are respectively T and T 0 closed, then for some disjoint G, G 0 which
are respectively T and T 0 open

F0 ⊆ G and F ⊆ G0.

(Take F := X\U disjoint from F 0 := H; then G 0 := X\ cl T 0 V is disjoint from

G := V ⊆ clτ V , also G ⊇ H and G 0 ⊇ X\U = F.)
(11) (R, D) is regular (cf. Oxt1980, Th. 22.9) – in fact cometrizable (implicitly,
by the proof just cited). More is true: D has the Lusin–Menchoff property
relative to the Euclidean topology (GofNN1961). This also implies that D is
completely regular, so has a (Hausdorff) compactification (GofNN1961).
(12) (R, D) is connected (GofW1961).
By contrast:
(13) (R, D) is not normal (GofNN1961), nor indeed pseudonormal (see Tal1978,
Th. 15), nor countably paracompact (Tal1976, Cor. 3.10).
(14) (R, D) is not second countable. Nevertheless, it does satisfy the countable
chain condition (ccc), i.e. any family of pairwise disjoint open sets is at most
 7.1 The Lebesgue Density Theorem 111

countable (more is true: it has property K (Knaster’s property) – any uncount-

able family of open sets has an uncountable subfamily of mutually intersecting
sets).
This motivates the following.

Definitions For a topology T , denote by IT the σ-ideal of meagre sets. (Thus

if T = D these are the null sets.)
For a σ-ideal I say that T is I-quasi-Lindelöf, or quasi-Lindelöf relative
to I, if every T -open family H contains a countable subfamily H 0 such that
S 0 S
H differs from H by a set in I.
When I is the σ-ideal of µ-null sets for a measure µ, we shall refer to this
property as µ-almost-Lindelöf, or almost-Lindelöf relative to µ.

(15) (R, D) is almost Lindelöf relative to Lebesgue measure. More is true.

Theorem 7.1.10 (LukMZ1986, 1.B.1a) If T satisfies the countable chain

condition and I = IT , then X is I-quasi-Lindelöf.

(16) (R, D) is not topologically complete, i.e. it is not a Gδ in its Stone–

Čech compactification, nor indeed even a Borel subset thereof (see FroN1990,
generalizing Gos1985). Nonetheless, it is pseudocomplete, hence Baire (see
§2.2), and in fact strongly α-favourable, for which see §2.3 (for proofs see
Whi1974).
We stop to consider in detail the fact that (R, D) is Baire. There is more
at work here than just the regularity of Lebesgue measure (see point (5) in
the list above). Note that in turn cometrizability of D depends on the inner
regularity of Lebesgue measure. The following theorem sheds more light on
the interconnections.

Theorem 7.1.11 (LukMZ1986, Th. 4.2) If the topology T on X is cometriz-

able relative to Tρ (for instance, has the Lusin–Menchoff property relative to
Tρ ) and Tρ is topologically complete, then any G(T )δ -subset, A, of X is a Baire
space under T . In particular, X is a Baire space under T .

Proof Suppose otherwise, and that A = i Ai with Ai nowhere dense in A

S
(under the T -subspace topology), where A = i Ui with Ui ∈ T . For X under
T
Tρ , choose a compactification Y and open sets G n in Y with X = n G n . Notice
T
that, for any T -open set U meeting A, the set A ∩ U\ cl T ( Ai ) is non-empty.
We show how to choose inductively points ai ∈ A ∩ Ui , Tρ -open sets Wi and
T -open sets Vi such that

ai ∈ Vi ⊆ clρ Vi ⊆ Ui ∩ [Wi \ cl T ( Ai )] and clY Wi ⊆ Gi .

112 Density Topology

Suppose done for i; then Vi ∩ Ui+1 ∩ A is non-empty, and so we may choose

ai+1 ∈ ( A ∩ Vi ∩ Ui+1 )\ cl T ( Ai+1 ). By regularity in Y , choose Tρ -open Wi+1
such that ai+1 ∈ Wi+1 ⊆ clY Wi+1 ⊆ Gi+1 and such that (again by regularity but
now in Tρ , and again passing to an open subset) ai+1 ∈ Wi+1 ⊆ clρ Wi+1 ⊆ Wi .
As Tρ -open is T -open, by cometrizability, choose Vi+1 ∈ T with

ai+1 ∈ Vi+1 ⊆ clρ Vi+1 ⊆ Vi ∩ Ui+1 ∩ [Wi+1 \ cl T ( Ai+1 )].

Since the sets clY (Wi ) have the finite intersection property, by compactness
∅ , i clY (Wi ) ⊆ n G n = X . So
T T
\ \ \
X∩ clY (Wi ) = clρ (Wi ) = Wi
i i i

is non-empty. Finally,
\ \ \
∅, Wn = Vn ⊆ Un = A
n n n

and
\
Vn ∩ cl T ( Ai ) = ∅ for each i.
n
S
So A\ i Ai is non-empty, a contradiction. 

Since by property (11) D has the Lusin–Menchoff property with respect to

the Euclidean topology E, and (R, E) is Baire by Baire’s Theorem, this shows
that (R, D) is Baire (recovering the statement of property (5)). Note however
that this result does not follow from any of the several versions of Baire’s
Theorem in Chapter 2. We regard Theorem 7.1.11 as the ‘Lusin–Menchoff
form of Baire’s Theorem’. For differences of non-meagre sets in a topological
group, see RaoR1975.
We close by considering two contributions of Caspar Goffman taken from
Gof1950 and Gof1975. These are stated so succinctly that we do not hesitate
to cite them verbatim.

Theorem 7.1.12 (Goffman’s Theorem) The set of points for which the metric
density of a measurable set S exists but is not equal to 0 or 1 is of measure 0
and of first category.

Proof We shall suppose all sets are contained in the open interval (0, 1). Let
T be the set of points for which the metric density of S exists, U those points of
T for which the metric density of S is 0 or 1, and Z = T\U. By the Lebesgue
density theorem, Z is of measure 0 and U is of measure 1. The metric density of
S is a function of Baire class 1 on T. For, let x be in T and let f n (x), n = 1, 2, . . .,
be the relative measure of S in the interval (x − 1/n, x + 1/n). For every n,
f n (x) is continuous and is, accordingly, continuous relative to T. Since the
 7.1 The Lebesgue Density Theorem 113

metric density of S exists at every point of T, f (x) = limn→∞ f n (x) exists, is

equal to the metric density of S, and is a function of Baire class 1 on T relative
to T. Its points of discontinuity must be a set of first category relative to T (by
the Baire Continuity Theorem, Chapter 2). On the other hand, U, as a set of
measure 1, is everywhere dense in T. Thus, every interval containing a point
of Z also contains points of U; that is, points x for which f (x) is either 0 or
1. Since, for every x in Z, f (x) is different from 0 or 1, Z must be a subset
of the set of points of discontinuity of f (x). Accordingly Z is of first category
relative to T and, therefore, relative to (0, 1). 
Katznelson and Stromberg (KatS1974) gave a relatively simple proof of the
existence of an everywhere differentiable function which is not monotonic in
any interval. We show the connection between this property and the density
topology.

Example of a Nowhere Monotonic Everywhere Differentiable Function,

cf. point (11) above. The density topology for the reals is completely regular
(GofNN1961; Zah1950). Since countable sets are closed in this topology, for
each countable S and ξ < S there is an approximately continuous f such that
0 < f (x) < 1, f (ξ) = 1 and f (x) = 0 for each x ∈ S. Let A and B be disjoint
countable sets each dense in the reals. They may be enumerated A = {an },
B = {bn }. For each n, let f n be approximately continuous, 0 < f n (x) < 1,
f n (an ) = 1, and f n (x) = 0 for each x ∈ B, and let gn be approximately
continuous, 0 < gn (x) < 1, gn (bn ) = 1, and gn (x) = 0 for each x ∈ A. The
function
X∞ X∞
f := 2−n f n − 2−n gn
n=1 n=1
is bounded, approximately continuous, positive on A and negative on B. Let F
be an indefinite integral of f . Then F is everywhere differentiable and F 0 = f .
So F is not monotonic in any interval.
We close by saying that in Euclidean space of dimension 2 and above, there
are multiple choices of density topologies, whose properties depend on the
selected basis B for the topology. See the examples considered in GofNN1961.
For other strange functions, see Kha2018.
 8

Other Fine Topologies

8.1 The Fine Topology of Potential Theory: Polar Sets

The term potential function (briefly, potential) stems from the first page of
the famous essay of 1828 by George Green (1793–1841) on electricity and
magnetism (Gre1828). Recall from physics that for a field of force (briefly,
field) which is conservative (one in which work done to arrive at a point is
independent of the path taken to get there), the field F is the gradient of the
potential, u:

F = grad u.

See, e.g., Kellogg (Doob’s favourite source) (Kell1953, III.1, pp. 48–54). There
one finds discussion of, e.g., units, force, work, potential energy, and the two
possible sign conventions here, depending whether one measures work done
by or against the field.
To begin with some background on gravitation: following two decades of
observation by Tycho Brahe (1546–1601), Johannes Kepler (1571–1630) pub-
lished his Astronomia Nova (1609). These contain Kepler’s Laws, the first of
which is: Planets move around the Sun in elliptical orbits with the Sun at one
focus. The great challenge this posed to astronomy was to explain Kepler’s
Laws, which had been arrived at empirically. It was suspected that an inverse
square law of attraction was the key. Sir Isaac Newton (1642–1727) established
this in his Principia of 1687, by linking it with the elliptical orbits of Kepler’s
First Law, by then well established experimentally (for a succinct two-page ac-
count, with references within the Principia, see Ram1951, pp. 167–168). The
counterpart of this in electromagnetism was the work of Ch. A. de Coulomb
(1736–1806), who in the period 1785–1791 established the inverse square law
of electromagnetism experimentally.

114
 8.1 The Fine Topology of Potential Theory: Polar Sets 115

In Newtonian gravitation, the potential of a unit point mass at a distance r

is 1/r; likewise for a unit point charge or magnetic pole. Differentiating this
to form the gradient above gives the inverse square law of gravity, and also
of electromagnetism. Thus already the idea of a potential has a unifying effect
on two of the four fundamental forces of nature, gravitation and electromag-
netism (despite the dissimilarity that in gravitation all matter attracts, while in
electromagnetism unlike charges attract but like charges repel). The other two,
the weak nuclear force (governing radioactivity) and the strong nuclear force
(enabling stability of matter by holding together in the nucleus the positively
charged protons that would repel each other electrostatically further apart),
emerged much later.
We speak nowadays of a fine topology as one finer than (refining, having
more open sets than) the usual (Euclidean) topology. But there is one such
topology – the first – that deserves the definite article, as the fine topology – that
arising in potential theory. As well as playing a vital role in physics and applied
mathematics, as a key component of the theory of two of the four fundamental
forces of nature, gravity and electromagnetism, potential theory also has a deep
and rich mathematical theory. It was aptly described by Pierre Jacquinot in
1964: ‘La théorie du potentiel est un véritable carrefour de la Mathématique’
(quoted by Bau1975). Our sources for this are Kellogg (Kell1953) for the early
theory, Helms (Hel1969) and the monumental Doob (Doo1984) for a more
modern, and topological, treatment. Here we work in Euclidean space R N of
dimension N ≥ 2. The case N = 1 is degenerate by comparison (Doo1984,
Ch. 1.XIV); the case N = 2 gives rise to logarithmic potentials; N = 3 to the
inverse square law above, N ≥ 3 to power-law potentials, below.
The fundamental kernel of classical potential theory on R N is (‘G for Green’)
G(x, y) (Doo1984, p. 6), while
 log(1/|x − y|), if N = 2,

G y (x) := G(x, y) := 
 1/|x − y| N −2,
 if N > 2
(with G(x, x) = +∞) gives the fundamental harmonic function (solution of
Laplace’s equation) with pole y (it is harmonic, as its value at the centre of a
ball is the average of its values over the surface of the ball). If µ is a measure
on R N ,
Z Z
G µ(x) := G(x, y) µ(dy) = G y (x) µ(dy)
RN RN
is the potential of µ. It is again harmonic, as it inherits the ‘harmonic average
property’ above. This usage is that of physics, if one thinks of µ as a distribution
of gravitational mass or electrostatic charge. With N = 3, r = |x − y|, one
recovers the 1/r above.
116 Fine Topologies

The fine topology (of potential theory, understood below) is defined as

the coarsest topology on R N making all superharmonic functions continu-
ous (Doo1984, 1.XI.1). One can conveniently use a superscript f to denote that
it is the fine topology that is being used.
As with the Euclidean and density topologies, each with its σ-ideal of negli-
gible (‘small’) sets, the meagre sets M and the (Lebesgue) null sets N , so the
fine topology of potential theory has its σ-ideal of negligible sets, the polar
sets, P, say. These are the sets each point of which has an open neighbourhood
carrying a superharmonic function which has a pole (takes the value +∞) at
each point of the set in the neighbourhood (Doo1984, Ch. 1.V.1; Hel1969,
§7.1). Equivalently, a set is polar if and only if it has no fine limit points (see
(iv) below).
From the definition, a subset of a polar set is polar. Also, P is closed under
countable unions (Doo1984, 1.V.3, p. 59). Combining, the polar sets do indeed
form a σ-ideal, P.
We note some key properties and definitions.

(i) The fine topology is strictly finer than the Euclidean topology (Doo1984,
p. 166).
(ii) A set A ⊂ R N is thin at a finite point ξ if and only if ξ < A f ; that is, ξ is
not a fine limit point of A (Doo1984, p. 167).
(iii) The Baire property holds for polar sets (ConC1972; Doo1984, pp. 167–
168). So (R N , F ) is a Baire space.
(iv) A set is polar if and only if it has no fine limit points (Doo1984, Th.
1.XI.6).
(v) The fine topology is strictly coarser than the density topology (Fug1971).

There is a hierarchy of small sets in potential theory, with the polar sets being
the smallest. To summarize (Chu1982, §3.5, p. 112; see there for definitions of
terms not already used):

polar = zero capacity ⊂ very thin ⊂ thin ⊂ semi-polar ⊂ zero potential.

Likewise, there is a hierarchy of topologies, both the inclusions (refinings)

below being strict:

Euclidean ⊂ fine ⊂ density: E ⊂ F ⊂ D.

The term polar set was introduced in 1941 by Brelot (Bre1941), and with it
the view of the polar sets as the negligible sets of potential theory. The polar
sets were shown to be the sets of capacity 0 by Cartan in 1945 (Car1945).
The Choquet theory of capacities of 1955 (Cho1953) is directly motivated by
 8.1 The Fine Topology of Potential Theory: Polar Sets 117

potential theory, as the terminology from electromagnetism suggests. For back-

ground on the links between capacity and stochastic processes, see Dellacherie
(Del1972).
The fine topology is due to Cartan (Car1946). For Cartan’s letter (30
December 1940) to Brelot on its significance in potential theory, and more on
the emergence of the term fine topology, see the Historical Notes in Doo1984,
p. 800.
Despite the comparability between the three topologies above, and the close
link each has with a σ-ideal of small sets (negligibles), these σ-ideals are not
themselves comparable. For, the three concepts of smallness involved are very
different. For instance, the real line R can be decomposed into two complemen-
tary sets, one meagre and the other (Lebesgue-)null (see Oxt1980, Th. 1.6, Th.
16.). Compare a result of Muthuvel (Mut1999): the additive group (R, +) is the
direct sum of two subgroups (one uncountable), one meagre and the other null.
The analogies between category and measure, or between the σ-ideals M and
N , have attracted great attention in descriptive set theory; see, e.g., Oxt1980;
Kec1995. The role of potential theory, or the σ-ideal P, has received less
attention; for a monograph treatment here, see LukMZ1986.

Notes
Regarding (iii) above: what is needed here is that the original topological
space be locally compact; see, e.g., LukMZ1986, p. 55.
Regarding limits and continuity in these topologies: a function f is approxi-
mately continuous at z if and only if it is continuous at z in the density topology
(LukMZ1986, p. 55; BinO2009f). Regarding links between fine and density
topologies, see LukMZ1986, 7C, pp. 267–270.
Markov Processes
We refer for background here to the standard monograph account of Markov
processes and potential theory by Blumenthal and Getoor (BluG1968). Recall
that a Markov process is one in which only the present, and not the past, is rele-
vant to predicting the future; equivalently, the past and future are conditionally
independent given the present (for background, see, e.g., Chu1982, augmented
as may be necessary by Chu1974; Chu1968).
To quote the first sentence of the introduction to Doo1984: ‘Potential theory
and certain aspects of probability theory are intimately related, perhaps most
obviously in that the transition function determining a Markov process can be
used to define the Green function of a potential theory.’ Again, to quote the
first sentence of Meyer’s classic (Mey1966, p. 1): ‘The fundamental work of
Doob and Hunt has shown, during the last ten years or so, that a certain form
of potential theory (the study of kernels which satisfy the “complete maximum
118 Fine Topologies

principle”) and a certain branch of probability theory (the study of Markov

semi-groups and processes) in reality constitute a single theory.’ (Later on the
same page, Meyer continues ‘Nothing of the theory of Markov processes itself
will be found herein’; he refers to a forthcoming ‘second volume’; this turned
out to be five, with Dellacherie, over a 17-year period (DelM1975–1992).)
In the Markov context, we can also define polar sets as follows (see, e.g.,
Haw1975). If X = {Xt } is a Markov process and B is a non-empty analytic
set, let
VB := inf {t > 0 : Xt ∈ B}

be the first-entry time into B, and call points of accessibility of B those in the set

Ac(B) := {x : Px (VB < ∞} > 0}.

Then a non-empty analytic set B is polar if and only if Ac(B) is empty.

Brownian Motion
The Markov process corresponding to the potential theory of classical elec-
tromagnetism as discussed above (the prime example of general potential the-
ory) is Brownian motion, B = {Bt : t ≥ 0} (the prime example of a Markov
process, and indeed of a stochastic process). For a monograph account of
the links between Brownian motion and classical potential theory, see, e.g.,
PortS1978. As a diffusion (path-continuous strong Markov process), B has an
infinitesimal generator, 12 ∆, with ∆ the Laplacian (or 12 D2 in one dimension);
see, e.g., Chu1982, §4.6. For more on the background here, see also the very
readable, brief, and wonderfully titled book of Chung, Chu1995.
The above link between Brownian motion and the Laplacian is actually a
link between Brownian motion (Bt ) and Laplace’s equation

∆u = 0

(elliptic). Similarly, space-time Brownian motion (Bt , t) corresponds to

1
∆u = ∂u/∂t,
2
the heat equation (parabolic). In Doo1984, Ch. 1.XV–1.XIX, Doob develops the
parabolic potential theory corresponding to this link. See also the monograph
by Constantinescu and Cornea (ConC1972), and the excellent survey by Bauer
(Bau1975) (especially §3). The related subject of harmonic spaces and ‘H-
cones’ is developed in ConC1972 and Boboc, Bucur and Cornea, BobBC1981.
Symmetric Markov Processes and Dirichlet Forms
For a Markov process X, write Pt (x, dy) for the probability of going from
 8.1 The Fine Topology of Potential Theory: Polar Sets 119

x to dy (the interval (y, y + dy)) in time t. The Markov property of X is then

expressed by the semi-group property
Pt+s f (x) = Pt (Ps f )(x)
for s, t ≥ 0, f bounded measurable. Then X is called symmetric with respect to
some reference measure m if
Z Z
g(x)Pt f (x)m(dx) = Pt g(x) f (x)m(dx)

for all such t, f , g.

We quote: if u is superharmonic on a set D, and has a subharmonic minorant,
then it has a greatest subharmonic minorant, GMD u, and this is harmonic
(Doo1984, 1.III.1). Then define
G D (y, . ) := G(y, . ) − GMD G(y, . ).
Then the function G D on D × D is called the Green function of D, and the func-
tion G D (y, . ) is called the Green function of D with pole y. Such a (Euclidean)
open set D has a Green function (‘is Greenian’).
If D is a Greenian set in R N and µ, ν are measures on D, their mutual energy
[µ, ν] is defined as
Z Z Z
[µ, ν] := G D µdν = G D (x, y) µ(dx)ν(dy);
D D D
the energy of a measure µ is defined as
k µk 2 := [µ, µ].
Note that this form is symmetric,
[µ, ν] = [ν, µ].
This symmetry corresponds to that in symmetric Markov processes, which
are the Markov processes most suited to this area. For background here, see
Silv1974; Silv1976.
It can be extended to signed measures – charges (the term is visibly derived
from electrostatics). Write E + for the set of measures on D of finite energy.
The set of charges representable as differences of positive measures in E + is
written E.
A measure of finite energy vanishes on polar sets. Conversely, if A is an
analytic subset of a Greenian set D and null for every measure on D of finite
energy supported by A, then A is polar (Doo1984, pp. 227–228).
Forms such as [. , . ] above derive from classical work of Dirichlet and are
known as Dirichlet forms. The modern period in this area dates from the
120 Fine Topologies

work of Beurling and Deny (BeuD1959) on Dirichlet spaces; see, inter alia,
BobBC1981. For a monograph treatment of the extensive links between Dirich-
let forms and Markov processes, we refer to Fukushima, Oshima and Takeda,
FukOT1994.
For more on Green functions, see, e.g., Doo1984, 1.VII. For more on energy
and capacity, see, e.g. Doo1984, 1.XIII.
Other σ-ideals of small (or ‘exceptional’) sets are those of the closed sets of
uniqueness for trigonometric series in the category and measure cases, MU
and N U say. (See Bary1964, Vol. II, p. 358 and §9.1.)

8.2 Analytically Heavy Topologies

Recall that a K -analytic subset A of a topological space (X, T ) takes the form
[
A = K (I) = K (i)
i ∈I

for K a compact-valued upper semi-continuous map from I = NN to X . We

denote by A(T ) the family of K -analytic subsets of (X, T ).
Definitions (K -Analytically Heavy Topologies)

(1) H is a topological base for X if (Eng1989, §1.1) H covers X and, for

H1, H2 ∈ H, whenever x ∈ H1 ∩ H2 there is H3 ∈ H with x ∈ H3 ⊆
H1 ∩ H2 . We write GH for the topology generated by H .
(2) B is a weak base for a topology T if, for each non-empty V ∈ T , there is
B ∈ B with ∅ , B ⊆ V . In fact, sometimes we need only a very weak base:
for each non-empty V ∈ T there is B ∈ B with ∅ , B ∩ V . See Remark (2)
below.
(3) Let (X, T ) be a regular Hausdorff space and T 0 ⊇ T a refinement topol-
ogy. We say T 0 is analytically heavy, or weakly K -analytically generated
in T , if T 0 possesses a weak base H ⊆ A(T ). That is, the weak base H
comprises sets that are K -analytic sets in T .
Remarks (1) In this bitopological context we refer to (X, T ) as the ground
space and (X, T 0 ) as the refinement.
(2) The ground space topology T itself is (weakly) K -analytically generated
if T possesses a (very weak) base H of sets that are K -analytic sets in T .
(3) If the weak base H in (1) is actually a base, then we say that T is a
generalized Gandy–Harrington topology. (See Example (2), Th. 8.2.1 and
at the end of this section.)
(4) If the base H in (2) is countable, then open sets are K -analytic.
 8.2 Analytically Heavy Topologies 121

(5) A K -analytic space, in which all open sets are K -analytic, is K -analytically
generated – take H = T .
Examples
(1) A complete separable metric ground space. For (X, Td ) with Td generated
by a complete separable metric d on X, the standard basis H of all open
(analytic) balls evidently yields GH = Td .
(2) (a) The Gandy–Harrington topology GH . For H , the countable family
of analytic subsets of R which are effective relative to a given real α (i.e.
Σ11 (α)), we obtain the Gandy–Harrington topology GH . For background on
the standard Gandy–Harrington case GH and variants, see, e.g., Lou1980,
Prop. 6 or MartK1980, §9.3.
(b) Any subfamily H of A(R) closed under intersection, including
A(R) itself, is a base for a topology in the sense of Definition (2) above.
In such circumstances the topology it generates, denoted GH , is of course
analytically generated.
(3) Density topology. For I = N , we may take H = D ∩ A(X ) as a base for
D. Here GH = D. Unlike in GH , the open sets of D are not analytic in
the ground space, although the basic sets of H are.
(4) (a) The Ellentuck topology, Ell. The points of this space lie in Cantor space
2N, the latter equipped with the Euclidean topology. The points of 2N are
interpreted as indicator functions of subsets of N. More specifically, one
considers only the points corresponding to infinite subsets of N, denoted
[N]ω . This subspace is a Gδ in 2N, so is topologically complete; indeed, if
h f n i enumerates [N]<ω, the family of all finite subsets of N, then [N]ω =
T
n {1S ∈ 2 : 1S , 1 fn }.
N

The refinement topology on [N]ω , called the Ellentuck topology after one
of its authors (Ell1974; see also Lou1976, and the more recent Rea1996),
is generated by taking for H the closed subsets
[a, A] := [N]ω ∩ {1S ∈ 2N : a ⊆ S ⊆ a ∪ A},
where a is finite and where A ⊆ N\{0, 1, . . . , max a} is infinite. Note that
A = N\{0, 1, . . . , max a} gives a set in the usual Cantor basis.
If I = {∅}, then (not unlike the case (2b) above) H is I-heavy. Recall
that a point is I-heavy if none of its neighbourhoods lie in I and a set is
I-heavy if all its points are I-heavy. The space is Choquet and so Baire
(Kec1995, 8.12 and 19.13); the latter will be confirmed in Theorem 8.2.1.
The topology yields a ‘short-cut’ for a proof of the Silver–Mathias Theorem
(Sil1971) that analytic sets (in the ground space) have the Ramsey property.
(b) Unlike GH , the Ellentuck topology is generated by a continuum of
analytic (in fact Gδ ) sets; a countable effective coarsening of significance
122 Fine Topologies

has been studied in Avi1998. These topologies are inspired by the method
of forcing used in set theory (see Chapters 14 and 16), which generate
Cohen reals, Solovay reals, and, among others, Mathias reals (associated
with the Ramsey property), and Hechler reals (for which see LabR1995).
Relations between forcing and descriptive set theory are traced in Mill1995.
(5) O’Malley’s r-topology (or resolvable-topology). To study approximate dif-
ferentiability of real-valued functions, O’Malley (OMa1977) introduced
the r-topology R on R with R ⊆ D; it is generated by taking as base
B := D ∩ G δ ∩F σ the sets of D that are ambiguously both Gδ and Fσ in
the real line. (For these, see also Sto1963, Th. 10. Recall that in a complete
space a set that is both Gδ and Fσ may be characterized as resolvable – see
Kur1966, §12. III, V.) The r-topology is a generalized Gandy–Harrington
topology, avant la lettre.
S
The argument for Theorem 8.2.1 repeatedly uses the fact that if n An ∩ B ,
∅, then An ∩ B , ∅ for some n. We view this as saying that I = {∅} has the
S
localization property and n An is I-heavy on B.
Theorem 8.2.1 (Generalized Gandy–Harrington Theorem) In a regular Haus-
dorff space, if T 0 is an analytically heavy refinement topology of T (i.e. pos-
sessing a weak base H ⊆ A(T ) ∩ T 0 whose elements are T 0-circumscribed),
then T 0 is Baire.
In particular, this applies to a Polish space, the Gandy–Harrington GH , the
density D, the Ellentuck Ell and the O’Malley r-topologies.
Proof We put In := Nn = {i | n : i ∈ I}. For each n, let Wn be dense
and open in T 0. Suppose inductively that for all m ≤ n there are upper semi-
continuous compact-valued maps Km : I → X which are T 0-circumscribed by
G m = hG m (i | n)i say, such that Km (I) ⊆ Wm with Km (I) ∈ H , σn (m) ∈ In
for m ≤ n, and
G1 (σn (1)) ∩ · · · ∩ G n (σn (n)) , ∅.
Then
Un := G1 (σn (1)) ∩ · · · ∩ G n (σn (n)) , ∅ and Un ∈ T 0 .
As Un is non-empty and open in T 0 and Wn+1 is T 0-dense, Wn+1 ∩Un , ∅. Since
H is a weak base, there is An+1 ∈ H with ∅ , An+1 ⊆ (Wn+1 ∩ Un ) ⊆ Wn+1
and in particular An+1 ∩ Un , ∅. Taking An+1 = Kn+1 (I) with Kn+1 a T 0-
circumscribed representation G m and noting that Kn+1 (I) = {G n+1 (σ) : σ ∈
S
In }, there is σn (n + 1) ∈ In such that
G1 (σn (1)) ∩ · · · ∩ G n (σn (n)) ∩ G n+1 (σn (n + 1)) , ∅.
 8.3 Other Fine Topologies 123

But G m (σn (m)) = k G m (σn (m), k). So there are extensions σn+1 (m) of
S
σn (m) for each m ≤ n + 1 such that
G1 (σn+1 (1)) ∩ · · · ∩ G n+1 (σn+1 (n)) , ∅.
This verifies the induction step. So for each m there is i(m) ∈ I with i(m) |
n = σn (m) for each n. Applying Theorem 3.1.12 in the ground space (taking
Fn = X ), we have
\ \ \
∅, G m (i(m)) ⊆ Am ⊆ Wm .
m m m
For W an arbitrary non-empty open set in T 0, as the set Wn ∩ W is T 0-dense
on W, we conclude by the preceding argument that ∅ , W ∩ m Wm . So T 0 is
T
Baire. 
Remark In the case of the Gandy–Harrington topology, the members of H
are analytic sets with representations Km such that each of the sets Km (i | n)
is also in H , so open by fiat in T 0 = GH . That is, the representations are
T 0-circumscribed.

8.3 Other Fine Topologies

The use of fine topologies in studying properties of Euclidean spaces has sev-
eral precedents. The earliest seems to be the fine topology of potential theory;
more recent examples, providing new and insightful proofs of important results,
include the Ellentuck topology, establishing the Ramsey property of analytic
sets, and the Σ11 -topology, establishing Silver’s Theorem on Π11 -equivalence re-
lations (see Kec1995; RogJ1980). Both give a Baire space. Our work, especially
in the next chapter, draws heavily not only on the density topology but also on
the z-topology (for z = hz n i a null sequence) and two other topologies, which
we term the essential topologies as they correspond to the notion of essential
point of accumulation in the sense of measure or category.
The classical prototype of a fine topology, the fine topology, F , arises in
potential theory, which as we have seen in §8.1 is the coarsest topology un-
der which all superharmonic functions bounded above are continuous; see,
e.g., Doo1984, I.XI; LukMZ1986, Ch. 10; Fug1971. In Euclidean dimension
d = 1, the fine topology coincides with the Euclidean topology on the line.
For d = 1, superharmonicity reduces to midpoint convexity, and midpoint con-
vex functions bounded above are continuous (see, e.g., BinO2008; BinO2009f;
Kucz1985, §XII.3; GerK1970). For d ≥ 2, the class of superharmonic func-
tions becomes richer: Newtonian potential theory applies, with the logarithmic
potential in the plane and the Coulomb or inverse-square potential in space.
124 Fine Topologies

For F , the small sets are the polar sets, and (R, F ) is quasi-Lindelöf relative
to the σ-ideal I of polar sets. It is Hausdorff, completely regular and locally
connected. The F -continuous functions are the superharmonic functions.

8.4 * Topologies from Functions, Base Operators or Density

Operators
The fine topology above was defined by specifying which family of real-valued
functions should be deemed continuous under the topology. As a result the space
is completely regular (see, e.g., LukMZ1986, Th. 2.3). The density topology of
Chapter 7 may also be defined in this way by reference to the approximately con-
tinuous function (as was done originally by Goffman and Waterman); O’Malley
(OMa1977) introduced a coarser topology, the r-topology, as the coarsest under
which the smaller family of approximately derivable functions is continuous.
For the topology r a base is provided by sets in D that are simultaneously Fσ
and Gδ (‘ambivalent’) in E; whilst not normal, this topology nevertheless has
the Lusin–Menchoff property relative to D (his Theorems 3.6 and 3.10). The
r-open sets differ from D-open sets by null sets (see OMa1977 for a charac-
terization). A still coarser topology (the ‘a.e.-topology’) comprises those sets
U ∈ D for which U differs from its E-interior by a null set; so the a.e.-topology
is that with open sets U ∪ M with U open in the usual topology and M a set of
density points of U. Consequently, a function that is approximately continuous
everywhere and continuous almost everywhere is a.e.-continuous.
With applications in mind, our initial approach here has been the measure-
theoretic refinement of the usual ‘limit-point property’, captured by the assertion
that A ∈ D if and only if A ∈ L and A ⊆ φ( A), where we recall that
φ( A) := {a ∈ A : D̄∗ ( A, a) > 0}. This prompts a unified approach (as we
shall make use of other kinds of limit points than in Chapter 7). The starting
point is a given topology on a set X, for instance one specified by declaring a
family of real-valued functions to be continuous. We then consider the following
constructions.

Base Operator. Given a topology T on X, one may associate with any subset
A either its closure Ā or its derived set der( A). Thus one may introduce (cf.
LukMZ1986, 1.A) a (monotone) base operator on the power set ℘(X ), i.e. a
mapping b: ℘(X ) → ℘(X ) with b(∅) = ∅ with the additivity property:

b( A ∪ B) = b( A) ∪ b(B).
 8.4 Base and Density Operators 125

Then say that F is b-closed if and only if b(F) ⊆ F. The b-closed sets define
the b-topology in which they are the closed sets, and one has b( A) ⊆ Ā. Taking
complements, the b-open sets are those sets A such that A ⊆ b( Ac ). Let us write
T (b) for the topology of b-open sets. Both closure and derived set operators
are base operators, and T (b) = T , i.e. both generate the original topology. But
they are not the only base operators, by any means. Thus the density topology
may be introduced using the base operator
b( A) := {x : D̄∗ ( A, x) > 0},
so that F is b-closed if and only if b( A) ⊆ A. This agrees with the opening
remarks as φ( A) = A ∩ b( A).
Note that if a density theorem holds, then cl T (b) ( A) = A ∪ b( A) and is a
measurable cover of A, and int T (b) ( A) = {a ∈ A : D∗ ( A, x) = 1} ∪ ( A\bX ) is
a measurable core of A.

Lower Density. Given a measure space (X, Σ, m), and denoting by Σ0 the σ-
ideal of its null sets, write A ∼ B when A4B ∈ Σ0 . A lower density (see also
p. 104) is a mapping S : Σ → Σ satisfying:

(i) (quasi-identity) S( A) ∼ A;
(ii) (neglecting property) if A ∼ B, then S( A) = S(B);
(iii) S(∅) = ∅ and S(X ) = X;
(iv) (multiplicative property) S( A ∩ B) = S( A) ∩ S(B).
For A ∈ Σ, say that A is S-open if A ⊆ S( A). This naturally reverses the
inclusion used to define the b-closed sets. In summary, the topology consists
of measurable sets and
TS := { A ∈ Σ : A ⊆ S( A)}.

Upper Density. One may take an entirely dual approach, preferring to intro-
duce the topology via closed sets. An upper density is a mapping U : Σ → Σ
satisfying:

(i) (quasi-identity) U ( A) ∼ A;
(ii) (neglecting property) if A ∼ B, then U ( A) = U (B);
(iii) U (∅) = ∅ and U (X ) = X;
(iv) (additive property) U ( A ∩ B) = U ( A) ∩ U (B).
Corresponding to a lower density S is the dual upper density U ( A) := S( Ac ) c .
Since complementation is self-inverse, any upper density corresponds to a dual
lower density via S( A) := U ( Ac ) c .
126 Fine Topologies

For an upper density U, say that A ∈ Σ is a U-closed set if and only if

A ⊇ U ( A). (Note that this yields the same topology as the dual inner density
because Acc = A ⊇ S( Ac ) c = U ( A) for A in Σ.)
For an upper density U and A ∈ ℘(X ), for H ( A) a measurable core of A
(determined up to null sets), put

b( A) := U (H ( A)).

Then b is a base operator and for A ∈ Σ

A ∩ b( Ac ) = A\S( A),

so any TS open set is b-open and conversely.

Remark A topology introduced by a lower/upper density operation may be
introduced by an abstract base operator. The corresponding closure operator
then provides a way for viewing that same topology as arising from a ‘derived
set’ operation.

Lifting. A mapping L : Σ → Σ is a lifting if it is both an upper and a lower

density (LukMZ1986, p. 223; see the characterization theorem on p. 224 –
these are maximal density topologies, and the closure of an open set is open –
and the existence theorem on p. 225).
Remark (Construction via Modifications) O’Malley’s set-wise characteriza-
tion of the topology which he introduced via the family of approximately
derivable functions (for which see Bog2007a, 5.8(v), pp. 370–373) motivates
the treatment in LukMZ1986 in considering modifications as a further tool for
a further refinement of any fine topology T , say refining a topology τ. First
there is the r-modification of a fine topology T , using as base the T -open
sets that are both Fσ and Gδ in τ. For the a.e.-topology, there are two possible
approaches. The a.e.-topology generated by T has as base the sets of the form
U ∪ {x} for U ∈ T and x ∈ X . Here V is a.e. open if and only if V \ int T (V ) is
T -discrete. A related topology is the a-modification: if b is a base operator for
T , put

a( A) := b( Ā)

and introduce the corresponding topology. Here a set V is a-open if and only if
V = G ∪ H with G a τ-open set H ⊆ G c and H ⊆ (bG c ) c . This is in general
coarser than the a.e.-topology. The a-topology inherits the Lusin–Menchoff
property with respect to a completely regular τ.
 8.5 Fine Topologies from Ideals: Base Operators 127

8.5 * Fine Topologies from Ideals I: Base Operators

Notation We write L+∗ for the sets that have positive outer Lebesgue measure
and L+ for non-null measurable sets. By analogy write Ba+∗ (X) for sets that are
non-meagre in the space X (since Ba(X) denotes its Baire sets) and Ba+ (X) for
the Baire non-meagre sets; also we write G+ (X) for the non-empty open sets
and ℘+ (X ) for the non-empty subsets of X . Of course Ba(R, E) comprises the
usual Baire sets and Ba+ (R, D) = L+ .
We quote two general results of Martin (Marti1961), albeit not quite in his
language, to introduce two fine topologies on R that we need to use later.
Definition Call a family I ⊆ ℘(X ) a proper ideal (with X < I) if
(i) I contains all singletons (atoms),
(ii) I is downwards hereditary (subsets of sets in I are in I), and
(iii) I is multiplicative, i.e. closed under intersection (if I1 , I2 ∈ I, then
I1 ∩ I2 ∈ I).
Passing to negation H = I ¬ := {H : H < I}, so that H ¬ = I, we obtain the
following dual concept used by Martin. Note that under negation (as opposed
to complementation) property (iii) does not translate into closure under union
(but rather the contrapositive of this, hence the ‘co’ below).
Definition Say that a family H ⊆ ℘+ (X ) is heavy if
(i) H is atomless (no singletons in H ),
(ii) H is upwards hereditary (supersets of sets in H are in H ), and
(iii) H is ‘additively maximal’, i.e. co-multiplicative (if H1 ∪ H2 ∈ H , then at
least one of H1 , H2 is in H ).
Note that F = I c := {X\I : I ∈ I} is a non-principal filter. A non-principal
ultrafilter F is heavy and I = F c is a corresponding maximal ideal I. The
term ‘heavy’ is borrowed from usage elsewhere (cf. in BraG1960). A heavy
point of a set is one at which the set is locally of second category, i.e. the
point is one of essential accumulation). The approach in LukMZ1986, p. 22 is
slightly different – in place of heavy sets they use the family of sets not lying
in a given ideal I and require that sets locally in I (i.e. every point of a set E
has a neighbourhood V with V ∩ E in I) are in I.
Theorem 8.5.1 (Marti1961, Th. 1, cf. LukMZ1986, §1.C) For B a base for
X = (X, T ) and H ⊆ ℘+ (X ) heavy, in particular for H = Ba+∗ or L+∗ ,
b H (E) = {x ∈ X : B ∩ E ∈ H } for every B ∈ Bx
is a base operator on X refining the topology T .
128 Fine Topologies

Definition For H = Ba+∗ (R, E) or L+∗ above, we will refer to the corre-
sponding two refinements as the essential topologies in category, respectively
in measure, and to the corresponding limit points of sets as points of essential
accumulation in category, respectively in measure.

Remarks 1. The properties (i)–(iii) of the family H formalize a sense of

largeness; thus the family H∞ of infinite subsets of X satisfies them, and the
corresponding operator gives rise to limit points that are points of accumulation
(derived points), so here it does not refine the original topology strictly. The
family H of uncountable subsets of X also satisfies them, and the corresponding
induced finer topology gives rise to limit points that are necessarily points of
condensation in the original topology.
2. Notice that b H (E) is closed in X and so Baire. If y is in its closure, then for
V ∈ By there is x ∈ V ∩ b H (E) and so there is B ∈ Bx with B ⊆ V . Then,
as x ∈ b H (E), one has B ∩ E ∈ H . By upwards closure V ∩ E ∈ H . The
set F (E) := E ∩ b H (E), which is closed relative to E, may now be viewed as
Stone’s non-locally-I kernel of A, best called the local-H kernel (see Sto1963).
The Generic Dichotomy Principle (Theorem 4.1.1) here asserts that the Stone
kernel is either empty for some non-meagre Gδ in X, or is almost all of A for
each Baire set A.
In the refinement topology A is closed if A ⊇ b H ( A).
3. As R is a topological group under addition, if H is shift-invariant one may
redefine the base operator by referring to a base at 0, say B0, and writing

b H (E) = {x ∈ X : (∀B ∈ B0 ) (B + x) ∩ E ∈ H }
= {x ∈ X : (∀B ∈ B0 ) B ∩ (E − x) ∈ H }

since H = H − x. Furthermore, if H is dilation-invariant (i.e. λH ∈ H for

H ∈ H and λ > 0), put B = (−1/n, 1/n) = I/n, with I = (−1, 1) and
n = 1, 2, . . . , and write

b H (E) = {x ∈ X : (∀n) I ∩ n(E − x) ∈ H }

\
= {x ∈ X : I ∩ n(E − x) ∈ H }.
n

We shall later need to know that b H takes values in Ba(R, E) or Ba(R, D).
For this purpose, we verify the following result.

Lemma 8.5.2 For X = R and I := {S : S < H } closed under countable

unions, the set {x : I ∩ n(E − x) ∈ H } is open and so b H (E) is a Gδ .
 8.5 Fine Topologies from Ideals: Base Operators 129

Proof Put S + (E) := {hu, vi ∈ R2 : u < v and (u, v) ∩ E ∈ H }. This is

open. Indeed, if hu, vi ⊆ ha, bi and hu, vi ∈ S + (E), then ha, bi ∈ S + (E); on
the other hand, if for each rational r > 0 we have hu + r, v − ri < S + (E), i.e.
(u + r, v − r) ∩ E < H , then
[
(u, v) ∩ E = (u + δ, v − δ) ∩ E ∈ H ¬,
0<r <(v−u)/2

a contradiction since H and H ¬ are disjoint. In turn, for J = (a, b), the set
β E (J) := {y : (J + y) ∩ E ∈ H } is also open since

{(v − b, u − a) : hu, vi ∈ S + (E) and 0 < v − u < b − a}.

[
β E (J) =

Indeed, if (u, v) ⊂ (y + a, y + b) and (u, v) ∩ E ∈ H , then v − b < y < u − a.

Finally,
! !
\ ( 1 ) \ 1
b H (E) = x∈X: I+x ∩E ∈H = βE I . 
n n n n
The lemma will enable us to show that almost all points of a set in H are
essential points of accumulation.
4a. Suppose that H is heavy but not necessarily translation invariant. Let B0
be a countable base for the neighbourhoods of 0. Put

b∗H (E) := {x ∈ X : (∀B ∈ Bx ) B ∩ E ∈ H + x}.

As B0 is countable, this defines a base operator. Indeed B ∩ E ∈ H + x if

and only if (B − x) ∩ (E − x) ∈ H , so with B 0 := (B − x) ∈ B0 we have that
B 0 ∩((E ∪F)− x) ∈ H implies that one of B 0 ∩(E − x) ∈ H or B 0 ∩(F − x) ∈ H
holds, and so for infinitely many B 0 ∈ B0 one of B 0 ∩ (E − x) ∈ H or
B 0 ∩ (F − x) ∈ H holds. Thus b∗H (E ∪ F) ⊆ b∗H (E) ∪ b∗H (F). As b∗H
is monotone, b∗H is additive. So b∗H determines a translation-invariant fine
topology.
4b. The z-topology. Let z n → 0 be a (null) sequence comprising infinitely many
distinct terms. Put Z = {z n : n ∈ ω} and take HZ to consist of sets meeting Z
in an infinite set; then HZ is heavy. Put
\ [
bz (E) = (E − zk ).
n k>n

Note that in particular bz (E) is in Ba(R, E) or Ba+ (R, D) = L+ if E is. We

claim that for H = HZ one has x ∈ bz (E) if and only if x ∈ b∗H (E). Indeed
x ∈ bz (E) if and only if there is an infinite set Mx with {x + z n : n ∈ Mx } ⊆ E
if and only if (x + {z n : n ∈ Mx }) ⊆ B ∩ E ∈ H + x if and only if x ∈ b∗H (E).
Call such an x a translator into E. Thus A is b-closed if and only if A ⊇
b H ( A) if and only if every translator into A is in A. Below we give the dual
130 Fine Topologies

statement in terms of z-open sets, which enables us to give an illuminating

example.

Theorem 8.5.3 For any null sequence z = hz n i, a set S is z-open if and only if
[ \
S⊆ (S − z n ),
k n≥k

i.e. S is z-open if and only if for each s ∈ S the sequence hs + z n i is almost

contained in S.

Proof Writing T = R\S note that T is z-closed if and only if bz (T ) ⊆ T if and

only if
\ [
(T − z n ) ⊆ T .
k n≥k

Passing to complements,
[ \ [ \
R\T ⊆ R\(T − z n ) = (S − z n ). 
k n≥k k n≥k

Example If z is any rational null sequence, then T := (0, 1) \ Q is z-closed

since Q is z-open, because any rational translate of the sequence z is rational.
However, T is not closed in the Euclidean topology.

Remarks 1. In the measure case, the essential topology is considered briefly

in Scheinberg (Sch1971, §2) as a topology in which the Borel sets coincide
with L and it is shown also that it is connected but not regular.
2. Sch1971, §3 considers a family H obtained from the family F of measurable
sets of density 1 at 0 by extending this to a maximal filter in L (an L-ultrafilter).
The family H is in particular heavy (relative to L – see his property (4) which
implies maximal additivity) and not translation invariant; here again the same
translation technique as with the z-topology may be used to generate a topology.
Scheinberg shows that the Borel sets of this topology again coincide with L
and that the topology is completely regular.
Motivated by the O’Malley characterization of his own fine topologies as
‘modifications’ of the Euclidean topology, we refer to the following result as a
‘modification’ theorem.

Theorem 8.5.4 (Modification Theorem; Marti1961, Th. 6; cf. LukMZ1986,

Prop. 1.8) For X = (X, T ) second countable and H ⊆ ℘+ (X ) heavy (in
particular for X = (R, E) and H = Ba+∗ or L+∗ ) suppose that in addition:

(iii)0 H is countably complete (if

S
i Hi ∈ H , then Hi ∈ H for some i);

(iv) H contains G+ .
 8.6 Fine Topologies from Ideals: Modes of Convergence 131

Then F is T (b H )-closed if and only if F = C ∪ Z where C is T -closed and

Z < H ; dually, V is T (b H )-open if and only if V = U\Z, where U is T -open
and Z < H .
Hence T (b H ) is neither first countable nor regular.

8.6 * Fine Topologies from Ideals II: Modes of Convergence

Our lead example here is a category analogue of the density topology, obtained
by first introducing a category analogue of density points. It is remarkable that
such a quantitatively defined notion can have a qualitative dual, and that marks
out this example as noteworthy; this was achieved by Wilczyński by altering
the mode of convergence and referring to the (Riesz) Subsequence Theorem,
1.3.1 (see §1.2).
The definition of a density point of a measurable set may be reformulated in
terms of indicator functions; by translation, it is enough to consider the case of
0 being a density point. For A any set on the line define its dilation by λ to be
λ A := {λa : a ∈ A}.
Since
| A ∩ [−1/n, 1/n]| |nA ∩ [−1, 1]|
= ,
1/(2n) 2
0 is a density point of A if and only if
|nA ∩ [−1, 1]| → 2.
Equivalently, passing to indicator functions and working in the space L 0 [−1, 1]
of measurable function on [−1, 1]:
1n A∩[−1,1] → 1 in measure (on [−1, 1]).
For fixed A, writing Jn := 1n A∩[−1,1], by the Subsequence Theorem, 1.3.1,
this is equivalent to the assertion that every subsequence hJn(k) i has a sub-
subsequence hJn(k ( j)) i converging a.e. to 1, i.e. off a null set.
The latter statement refers to the σ-ideal of null sets and to the formation
of subsequences. This opens up a number of possibilities for refining the usual
notion of density point, and hence the usual topology of the line, by varying
both the σ-ideal and the mode of convergence demanded of the sequence Jn .
Though our starting point is the base operator of the last section, modifying
the mode of convergence restricts the accumulation properties to the extent of
undermining the additive property of the operator; fortunately the monotone
operator is multiplicative and so the fine topologies here are introduced using
132 Fine Topologies

the lower density construction. We consider in detail two examples and then
comment on others. The notation is as above.
1. For H a filter, passing to the ideal I = H ¬, the definition of b H may
be read as saying that 1n(E−x)∩I (t) → 1, I-a.e. on I = [−1, 1] for each n.
A variant may be obtained by replacing this a.e.-constancy by ordinary (or
simple) a.e.-convergence, requiring instead that 1n(E−x)∩I (t) converges I-a.e.
to 1 on [−1, 1]. This leads to

H (E) = {x ∈ X : (∃k)(∀n ≥ k) I ∩ n(E − x) ∈ H }

Φord
[ \ ( 1  )
= x∈X: I+x ∩E ∈H ;
k n≥k n
this time a Gδσ monotone operator. We shall see below that Φord H
is ‘multi-
plicative’ rather than ‘additive’. Hence, if applied to the σ-algebra Ba+ is a
lower-density operator (as opposed to a base operator).
2. Switching from the simple to an alternative mode of convergence for the
indicator functions, which we call sequential (or strong), we may require now
that for each subsequence γ(n) there be a sub-subsequence γ(κ(n)) such that
1γ(κ (n))(E−x)∩I (t) converges I-a.e. to 1 on [−1, 1]. This leads to
\ \ [
seq
Φ H (E) = {x ∈ X : I ∩ γ(k)(E − x) ∈ H },
γ ∈Γ n k>n

where Γ := NN denotes the set of all subsequences. We shall presently see

seq
that Φ H is also ‘multiplicative’ (rather than ‘additive’). The set here is the
complement of the set
[ [ \
Φc (E) := {x ∈ X : I ∩ γ(k)(E − x) < H }
γ ∈Γ n k>n

and we have
[ [ \
Φc (E) = Ain,
n γ ∈Γ i

with
\
Ain := {x ∈ X : I ∩ γ(k)(E − x) < H },
n+i>k>n
as
\ \
Ain = {x ∈ X : I ∩ γ(k)(E − x) < H }.
i k>n
By Lemma 8.5.2, the sets Ain are
closed and so Φc (E) is Souslin-F . Thus, by
c seq
Nikodym’s theorem, Φ (E) and so Φ H (E) is Baire. Here again, if applied to
the σ-algebra Ba+ , this is a lower-density operator.
We need to know now that our last two monotone operators are in fact lower
densities. Here we check only that they are multiplicative. In Chapter 9, we
 8.6 Fine Topologies from Ideals: Modes of Convergence 133

shall see why when applied to a non-meagre/non-null set they are the identity
modulo their defining ideal, equivalently why quasi all points are density points
(in whichever of the two senses and the two ideals). This will follow from the
facts, just established above, that they take Baire sets to Baire sets.
seq
Lemma 8.6.1 Both Φ = Φord
H
and Φ = Φ H (E) are multiplicative, i.e.

Φ( A ∩ B) = Φ( A) ∩ Φ(B).
seq
Proof We consider first Φ = Φ H (E). By monotonicity we have Φ( A ∩ B) ⊆
Φ( A) and Φ( A ∩ B) ⊆ Φ(B), and so Φ( A ∩ B) ⊆ Φ( A) ∩ Φ(B).
For the reverse inclusion, let t 0 ∈ Φ( A) ∩ Φ(B). So 0 is a density point of
both A0 := A − t 0 and B 0 := B − t 0 .
Now consider any γ ∈ Γ, where, as before, Γ := NN denotes the set of all
0 0
subsequences. There is κ ∈ Γ such that J A (t) = limγ(κ (i)) hJiA (t)i quasi all
0 0
t. But there is also λ ∈ Γ such that J B (t) = limγ(κ (λ(i))) hJiB (t)i quasi all
0 0
t. The latter occurs simultaneously with J A (t) = limγ(κ (λ(i))) hJiA (t)i quasi
all t (since pointwise convergence is preserved under subsequences). Hence
0 0 0 0
J A ∩B (t) = limγ(κ (λ(i))) hJiA ∩B (t)i quasi all t down the subsequence κ ◦ λ of
γ. So 0 is a density point of A0 ∩ B 0 = A ∩ B − t 0, i.e. t 0 ∈ Φ( A ∩ B).
The other case Φ = Φord I
is similar but simpler. 

In summary: by switching σ-ideals and varying modes of convergence as

appropriate (ordinary-a.e., summable-a.e.) one may introduce the notion of

(i) a simple-density point, by demanding ordinary a.e.-convergence of the

original sequence Jn relative to the null sets L0 (as in WilW2007) – so a
simple-density point is a density point;
(ii) a summable-a.e. density point, better known as complete-density point, by
demanding that for each  > 0
X
limm→∞ |{t : | Jm (t) − 1| >  }| = 0,
m

or equivalently that
X
|{t : | Jn (t) − 1| >  }| < ∞
n

(as in WilW2007); this topology is coarser than the density topology and
also coarser than the simple density topology:

E ⊂ Tc ⊂ Ts ⊂ D.

(iii) a category-density point by demanding a.e.-strong (i.e. sequential) con-

vergence relative to the σ-ideal of meagre sets (as in PorWW1985).
134 Fine Topologies

Correspondingly one obtains the complete density topology and the category-
density topology. This is connected but not regular, and hence is not the coarsest
topology making I-approximately continuous functions continuous.
We are now able to complement §4.1 with the following examples.
Example 1 Our first example is a category analogue of the example in §4.1,
obtained by referring to the category analogue of density points from the com-
plete density topology above. We just saw that the topology may be introduced
cplt
by a monotone operator, which we now denote by Φ H (E), which takes Baire
sets to Baire sets and is multiplicative. So to show that almost all points of E
cplt cplt
are in Φ H (E) it is enough to demonstrate the existence of points in Φ H (E)
for E Baire non-meagre. This we now do.
Existence. For E a Baire set, if E is non-meagre there is a non-empty open
interval G with G\M ⊆ E for some meagre set M.
But every point g of G is a (simple) density point of G and so of E, as 0 is
a (simple) density point of G − g and so of E − g. That is, E has a (simple)
density point. So almost all points of E are (simple) density points.
cplt
Thus in fact Φ H (E) is an inner-density.
Example 2 Here we refer to the z-topology. Let z n → 0 and put F (T ) :=
T S
n∈ω m>n (T − z m ). Thus F (T ) ∈ Ba for T ∈ Ba and F is monotonic. Here
t ∈ F (T ) if and only if there is an infinite Mt such that {t + z m : m ∈ Mt } ⊆ T .
The Generic Dichotomy Principle (Theorem 4.1.1) asserts that once we have
proved (for which see Theorem KBD, 4.2.1) that an arbitrary non-meagre Baire
set T contains a ‘translator’, i.e. an element t which shift-embeds a subsequence
z m into T, then quasi all elements of T are translators.
Example 3 For z n = n and {Sk } a family of unbounded open sets (in the
T T S
Euclidean sense), put F (T ) := T ∩ k ∈ω n∈ω m>n (Sk − z m ). Thus F (T ) ∈
Ba for T ∈ Ba and F is monotonic. Here t ∈ F (T ) if and only if t ∈ T and,
for each k ∈ ω, there is an infinite Mtk such that {t + z m : m ∈ Mtk } ⊆ Sk . In
Kin1963, it is shown that F (V ) is non-empty for any non-empty open set V ; but
in the Bitopological Kingman Theorem, 5.1.6, we must adjust the argument to
show that F (T ) is non-empty for arbitrary non-meagre sets T ∈ Ba, hence that
quasi all members of T are in F (T ), and in particular that this is so for T = R+ .

8.7 * Coarse versus Fine Topologies

Definitions Let B be a base of open sets of a space X = (X, T ).
A regular filter base in B is a filter F ⊆ B such that any finite intersection
 8.7 * Coarse versus Fine Topologies 135

of sets in F contains the closure of an element of F . A maximal such filter

base is termed an ultrafilter base.
T
A regular filter base F is pre-convergent if F is non-empty. A regular
T
ultrafilter is called convergent if F is a single point.

Example For D the density topology, B = D, F = F0, the family of open

rational intervals about 0 is a pre-convergent (countable) regular filter base.

Definitions The space X is subcompact if for some B every regular filter base
relative to B is pre-convergent.
Equivalently, the space X is subcompact if for some B every ultrafilter base
relative to B is convergent.
The space is countably subcompact if for some B every countable regular
filter base relative to B is pre-convergent.

Theorem 8.7.1 (Gro1963) For X metrizable, (countable) subcompactness is

equivalent to topological completeness.

Theorem 8.7.2 (Gro1963) A regular (countably) subcompact space is a Baire

space.

For studying a space X = (X, T ) by reference to coarser topologies T 0 ⊆ T ,

we need two definitions. Say that X 0 = (X, T 0 ) is a cospace of X and T 0 a
cotopology if for all V closed in T and all x ∈ int T V there is a T 0-closed U
(so T -closed) such that

x ∈ int T U ⊆ U ⊆ V .

For T regular, this is equivalent to the existence of a neighbourhood base for

T consisting of T 0-closed sets. When the cotopology T 0 is metrizable, we say
that X is cometrizable. Interest in coarser topologies comes from the following:

Example If C is a family of closed sets containing a (closed) neighbourhood

base for each point of space, then {X\F : F ∈ C} is a cotopology for X .

Definition A space X is cocompact if it possesses a compact cospace X 0 .

The next theorem requires a lemma on refinement of locally finite covers

(for which see §1.2). We omit the proof of the lemma, as this would take us
too far afield; see AarGM1970b (embedded within the proof of Th. 1; this is
essentially Dow1947, Lemma 3.3).

Lemma 8.7.3 (A ‘Centred Is Finite’ Refinement) For X normal, and U a

locally finite open covering, there is an open refinement V such that every
subfamily of {V̄ : V ∈ V } with the finite intersection property is finite.
136 Fine Topologies

Theorem 8.7.4 (AarGM1970b, Th. 1) A metrizable space is topologically

complete if and only if it is cocompact.
Proof Suppose X = hX, T i is metrizable and cocompact, i.e. that X 0 =
hX, T 0i is compact. As X is regular, the family U of T 0-closed subsets of
X yields a neighbourhood base for X. Embed X densely in a complete metric
space hY, ρi (see Eng1989, Th. 4.3.14 and 4.3.15). In view of the Alexandroff–
Hausdorff Theorem (Eng1989, Th. 4.3.23; cf. p. 35), we proceed to show that
X is a Gδ subset of Y . With δ denoting the ρ-diameter, put Ui := {U ∈ U :
δ(U) ≤ 2−i } and
Gi := {V : V open in Y and V ∩ X ⊂ U for some U with δ(U) ≤ 2−i }.
So X ⊆ i Gi, since U is a neighbourhood base for X. For the reverse inclusion,
T
consider y ∈ i Gi . As y ∈ Gi, pick Vi open in Y with y ∈ Vi and Ui in U such
T
that Vi ∩ X ⊂ Ui with δ(Ui ) ≤ 2−i . As X is dense in Y, and because {Vi : i ∈ ω}
has the finite intersection property (witnessed by y), the family {Vi ∩ X : i ∈ ω}
also has the finite intersection property, and so too has {Ui : i ∈ ω}. However,
X 0 is compact, so i Ui , ∅, and, since δ(Ui ) ≤ 2−i for each i, i Ui = {x},
T T
for some x ∈ X . Working in the metric space Y, we have y ∈ Vi for all i and by
density δ(Vi ) ≤ 2−i ; on the other hand, x ∈ Ui all i, so x = y (otherwise for
large enough i, Ui ∩ Vi = ∅). Thus X = i Gi and so is a Gδ subset of Y .
T
For the converse, let ρ now denote a complete metric on X compatible with
T . Working in T , by paracompactness for each i ∈ ω there is a locally finite
open covering Ui refining the open balls of ρ-radius 2−i . By Lemma 8.7.3, there
exists for each i ∈ ω an open covering refinement Vi of Ui such that every
subfamily of {V̄ : V ∈ Vi } with the finite intersection property is finite. Put
B 0 := i {X\V̄ : V ∈ Vi } ⊆ T , and let T 0 be the (coarser) topology generated
S
by taking B 0 as a base; then X 0 = hX, T 0i is compact by the Alexander Subbase
Theorem (see Eng1989, Prob. 3.12.2 or Kel1955, Ch. 5, Th. 6). Indeed, the
family V̄ := i {V̄ : V ∈ Vi } is a subbase for the closed sets of T 0, and in any
S
infinite subfamily of V̄ with the finite intersection property there is a further
infinite subfamily with shrinking ρ-diameter (by the property of Vi ), and the
latter subfamily has non-empty intersection (by Cantor’s Theorem – §1.2). It
remains to check that T 0 is a cotopology, i.e. that any point has a neighbourhood
base for T consisting of T 0-closed sets. But this follows from the construction
above, since Vi is an open covering refining the balls of radius 2−i . 
 9

Category–Measure Duality

9.1 Introduction
The duality between measure and category emerged in the 1920s, largely in
the works of Sierpiński. See the commentary by Hartman (Hart1975), in Sier-
piński’s selected works (Sie1975; Sie1976).
We recall first that we term subsets of R non-negligible under some topology
if they have the Baire property and are non-meagre. This has the usual cate-
gory meaning for the Euclidean topology, whereas in the context of the density
topology this means they are measurable and non-null. This chapter is devoted
to some classical results in what is known as category–measure duality. These
are theorems in real analysis which, regarding negligibility or non-negligibility,
yield identical conclusions in the category and measure cases. Thus the defi-
nition of topological completeness allows us to capture a structural feature of
category–measure duality: both exhibit Gδ inner regularity, modulo sets which
we are prepared to neglect. Other instances of duality include the following:

(i) The Poincaré recurrence theorem from statistical mechanics: this occurs
in both category and measure forms; see, e.g., Oxt1980, Ch. 17.
(ii) The zero–one law of probability theory, which similarly dualizes; see, e.g.,
Oxt1980, Ch. 21.
(iii) The Erdős–Sierpiński duality principle, which gives full duality assuming
the Continuum Hypothesis (CH); see, e.g., Oxt1980, Ch. 19.

Where there are identical conclusions, the basis resides in the dual foundation
unified in two ways in BinO2009b and BinO2010h. As we had occasion to say
in Chapter 4, the former creates structural unification by way of identical
combinatorics; the latter a common single source: Baire category. Again, both
views translate immediately to Rd .

137
138 Duality

It is as well to be aware from the outset, however, that this duality has its limits;
see Chapter 14. For example, in probabilistic language, despite (ii), it does not
extend as far as the strong law of large numbers (see, e.g., Oxt1980, p. 85), nor
the theory of random series (see, e.g., Kah2000; Kah2001). Similarly, despite
(i) it does not extend as far as the (Birkhoff–Khinchin) ergodic theorem (by
above – as the ergodic theorem includes the strong law of large numbers). For
other limitations of category–measure duality, see, e.g., DouF1994; Barto2000;
BartoJS1993; Fre2008, §522.

9.1.1 Uniqueness Theorems for Trigonometric Series

A rather different example is the uniqueness theorems for trigonometric series.
For the classical theory here, see Bary1964; Zyg1988. A set P on the torus
T is called a set of uniqueness if every trigonometric series which converges
to zero outside P vanishes identically (P is a set of multiplicity otherwise).
Every (Lebesgue-)measurable set of uniqueness is null, see Bary1964, Vol. II,
Ch. XIV; Zyg1988, Vol. I, 9.6; KecL1987, 1.3. Much more recent is the corre-
sponding category result: every Baire set of uniqueness is meagre (KecL1987,
VIII; DebSR1987).

9.1.2 Cauchy Functional Equation

One of the earliest dualities concerns the Cauchy functional equation:
f (x + y) = f (x) + f (y) (x, y ∈ R), (CFE)
where the unknown function f has the Baire property (meaning it is measurable
when the topology on R is the density topology), with the conclusion that f is
continuous and so linear.
The results are linked to Steinhaus’ Theorem, 9.2.2, that the difference set,
S − S, for S non-negligible, contains an interval (Oxt1980, p. 93, Supplemen-
tary notes). The category case is known as the Piccard–Pettis Theorem. Their
generalizations are known as Steinhaus–Weil Theorems. We turn to these in
Chapter 15.

9.1.3 Kodaira’s Theorem

Yet quite another linkage between category and measure is traced in Kodaira’s
Theorem (Kod1941) which relies on a density topology arising from Haar
measure, rather like the Lebesgue density topology of Chapter 7 (cf. p. 257).
For details see BinO2010g, Th. 7.3. The theorem asserts that for X a normed
locally compact group and f : X → Y a homomorphism into a separable
 9.2 Steinhaus Dichotomy 139

normed group Y , f is Haar-measurable if and only if f is Baire under the

density topology if and only if f is continuous under the norm topology.
We regard this as further evidence for the role of category taking precedence
over that of measure.
Of a different nature is a pair of theorems asserting that a planar set is
negligible if and only if all but a negligible set of vertical sections are negligible.
In the category case, under the Euclidean topology, this is the Kuratowski–Ulam
Theorem (Oxt1980, Ch. 15) and in the measure case the Fubini Theorem for
Null sets (Oxt1980, Ch. 14), which we term the Fubini Null Theorem (see §9.5).
Unlike the theorems of the preceding paragraph, these two results do not have
a common proof. Indeed, the first of the pair does not have a generalization that
is generally valid beyond separable spaces, unlike the second. We will discuss a
restricted variant of the Kuratowski–Ulam Theorem, due to Fremlin, Natkaniec
and Recław, following FreNR2000; examples of cases of failure are given by
Pol and van Mill; see Pol1979 and MilP1986. We also include a proof of the
Fubini Null Theorem.

9.2 Steinhaus Dichotomy

We deduce the two classical motivating theorems, the Piccard–Pettis and the
Steinhaus theorems – which constitute the Steinhaus dichotomy. See Piccard
(Pic1939; Pic1942), Pettis (Pet1950) and, for the measurable case, Steinhaus
(Stei1920).
Theorem 9.2.1 (Piccard–Pettis Theorem) For S ⊆ R Baire and non-meagre,
the difference set S − S contains an interval around the origin.
Theorem 9.2.2 (Steinhaus Theorem) For S ⊆ R (Lebesgue-)measurable and
of positive measure, the difference set S − S contains an interval around the
origin.
Proof of Theorem 9.2.1 Suppose otherwise. Then, for each positive integer n
we may select z n ∈ − 1 n, + 1 n \(S − S). Since z n → 0, by Theorem KBD,
 

4.2.1, for quasi all s ∈ S there is an infinite Ms such that {s + z m : m ∈ Ms } ⊆ S.
Then, for any m ∈ Ms , we have s + z m ∈ S, i.e. z m ∈ S − S, a contradiction. 
Remark See Chapter 15 (and BinO2010g) for a derivation from here of the
more general result that for S, T Baire and non-meagre in the Euclidean topol-
ogy, the difference set S − T contains an interval.
Proof of Theorem 9.2.2 Arguing as above, here again Theorem 4.2.1 applies.

140 Duality

Just as with the Pettis extension of Piccard’s result, so also here, Steinhaus
proved that for S, T non-null measurable S − T contains an interval. For this
see again Chapter 15: the result may be derived from the Category Embedding
Theorem, 10.2.2; see BinO2010g.
These results extend to topological groups. See, e.g., Com1984, Th. 4.6,
p. 1175, for the positive statement, and the closing remarks for a negative one.

9.3 Subgroup Dichotomy

The next result gives (with its refinement, Theorem 9.3.2) the third of the sharp
dichotomies of this chapter. It concerns additive subgroups of the reals. These
may be small or large in a number of senses, to be made precise below. For
instance, a subgroup may be discrete (the integers, for example) or dense (the
rationals); we use Kronecker’s Theorem (see HarW2008, XXIII, Th. 438) to
split these two cases. In analysis one needs a stronger dichotomy, in which
‘large’ means total – the entire real line. Theorem 9.3.1 takes small in the dual
senses of category and measure, and is a direct consequence of Theorem 9.3.2
(see Remark 1 after that result).
Theorem 9.3.1 (Subgroup Theorem; cf. BGT1987, Cor. 1.1.4; Lac1998) For
an additive Baire (resp. measurable) subgroup S of R, the following are equiv-
alent:
(i) S = R;
(ii) S is non-meagre (resp. non-null).
Proof By Theorem 4.2.1, for some interval I containing 0, we have I ⊆ S−S ⊆
S, and hence R = n nI = S.
S

Interest in the theorem comes typically from examples such as the one below.
Example For an additive extended real-valued function f : R → R ∪ {∞},
D f := {x : f (x) < ∞} is a subgroup of R and is the domain of definition of the
corresponding real-valued ‘partial’ function f .
For a more intricate example, where the domain of definition of a ‘partial
function’ is a subgroup of R, see BGT1987, Lemma 3.2.1. The task there is
to give additional conditions under which the subgroup is all of R, so that
the partial function is in fact total (cf. BGT1987, Th. 3.2.5) Sometimes a
sufficient additional condition is density of the subgroup (in R), which in turn
may be reduced to the existence of two rationally incommensurable elements
in the subgroup (by Kronecker’s Theorem). Theorem 9.3.2 uses related but
 9.3 Subgroup Dichotomy 141

stronger conditions than density guaranteeing ‘totality’. See BGT1987, Th.

1.10.2, where one uses density to show that in fact the domain of definition
contains a co-countable set, so that the Subgroup Theorem applies.
Here we develop a combinatorial version, in the language of Ramsey the-
ory (TaoV2006; GraRS1990) (though this is infinite Ramsey theory, while
GraRS1990 is finite Ramsey theory – see their page 185).
Definitions 1. Say that a set S ⊆ R has the strong (weak) Ramsey distance
property if for any convergent sequence {un } there is an infinite set (a set with
two members) M such that
{un − um : m, n ∈ M with m , n} ⊆ S.
Thinking of the points of S as those having a particular colour, S has the strong
Ramsey distance property if any convergent sequence has a subsequence all of
whose pairwise distances have this colour.
2. Motivated by Lemma 9.3.4, say that S ⊆ R has the finite covering property
if there is an interval I and finite number of points {x i : i = 1, . . . , m} such that
the shifts {S + x i : i = 1, . . . , m} cover I. When S is a subgroup, these shifted
copies of S are just S-cosets.
Theorem 9.3.2 (Combinatorial Steinhaus Theorem) For an additive subgroup
S of R, the following are equivalent:
(i) S = R;
(ii) S contains a non-meagre Baire, or a non-null measurable set;
(iii) S is shift-compact;
(iv) S has the strong Ramsey distance property;
(v) S has the weak Ramsey distance property;
(vi) S has the finite covering property;
(vii) S has finite index in R.
The proof follows Lemmas 9.3.3 and 9.3.4; for an application see BinO2009b.
Theorem 9.3.2 effects a transition from topological through combinatorial to
algebraic notions bringing out different aspects of the Steinhaus Theorem. See
Laczkovich (Lac1998) for a topological study of proper subgroups of R (cf.
BinO2011a, §10, Remark 4). Taking a topological view as in §6.4, any open
covering of a shift-compact subset of R yields a finite ‘shifted-subcovering’,
i.e. one consisting of shifted copies of a finite number of members of the open
covering (BinO2010g).
Lemma 9.3.3 (Finite Index Property) An additive subgroup of R has finite
index if and only if it coincides with R.
142 Duality

Proof Suppose an additive subgroup S has finite index, n say, so that the
quotient R/S is a finite group of order n. Then, for each x ∈ R, denoting S-cosets
by [x], one has n[x/n] = [0] by Lagrange’s Theorem. That is, x = n(x/n) ∈ S
i.e. S is R itself. 
We use the above result in combination with the next observation, which is
actually an instance (by specialization to R) of the Finite Index Lemma due
to Neumann (see Neu1954a, Neu1954b; Fuc1970, Lemma 7.3): if an abelian
group can be covered by a finite number of cosets of subgroups, then one of
the subgroups has finite index. The dual reformulation, that a proper additive
subgroup of R does not have the finite covering property, is actually what we
need to prove that (vi) implies (i).
Lemma 9.3.4 (Finite-Covering Characterization) For S an additive subgroup
of R, some open interval is covered by a finite union of cosets S of R if and only
if S has finite index in R if and only if S = R.
Proof For S a subgroup of R, note first that S is either countable or dense
(or both). Indeed, if S is uncountable, then it contains two elements which
are rationally incommensurable, so there are two elements s, s 0 of S such that
ps + qs 0 is non-zero for all non-zero integers p, q. (Otherwise there are non-zero
integers p, q such that ps + qs 0 = 0 in which case s/s 0 = −q/p ∈ Q.) But then
the subgroup of S comprising the points ps + qs 0 for p, q integers is dense in
R, since s 0/s is irrational (again by Kronecker’s Theorem). So S is dense.
Suppose that a finite number of cosets of S, say {[x i ]S : i = 1, 2, . . . , m},
covers an interval (a, b) with a < b, in which case S is uncountable and so
dense. We will show that these cosets in fact cover all of R. Indeed, by density
of S, for any x ∈ R we may choose s ∈ S ∩ (x − b, x − a) and so x ∈ s + (a, b).
But [s + x i ] = [x i ], so one of these covers x. That is, {[x i ] : i = 1, 2, . . . , m}
cover R, i.e. S has finite index in R (as in Neumann’s Lemma). By the preceding
lemma, S = R. The converse is clear. 
Proof of Theorem 9.3.2 It is clear that (i) implies (ii) and from Theorem 4.2.1
that (ii) implies (iii). To see that (iii) implies (iv) observe that as S is shift-
compact there are t and an infinite M such that
{t + un : n ∈ M} ⊆ S.
As S is a subgroup, for distinct m and n in M
un − um = (t + un ) − (t + um ) ∈ S,
giving (iv). Clearly (iv) implies (v). To prove (v) implies (vi) we may assume that
S , R (otherwise there is nothing to prove) and, by aiming for a contradiction,
 9.3 Subgroup Dichotomy 143

that S does not have the finite covering property. Suppose that v0, . . . , vn−1 have
been selected with vk < 1/(k + 1) 2 and vm + · · · + vn−1 < S for each m < n − 1.
We want to select vn < 1/(n + 1) 2 such that for each m < n,
vm + · · · + vn < S, i.e. vn < S − (vm + · · · + vn−1 ). (*)
Thus we require that
\  
0, 1 (n + 1) 2 \ (S − (vm + · · · + vn−1 ))

vn ∈
m<n
  [
= 0, 1 (n + 1) 2 \ (S − (vm + · · · + vn−1 )) .

m<n
If we cannot select such a vn, then
[  
S − (vm + · · · + vn−1 ) ⊇ 0, 1 (n + 1) 2 ,

m<n
and so S does have the finite covering property, contradicting our assumptions.
(The induction can be started, as S is a subgroup, so cannot contain any interval,
in particular (0, 1).) Thus, the induction can proceed. Put un := v1 + · · · + vn ;
then {un } is convergent. By (iv), there is a set M such that for m and n in M
with m < n,
vm + · · · + vn−1 = un − um ∈ S.
This contradicts (*), so after all S does have the finite covering property, i.e.
(vi) holds. Lemma 9.3.4 shows that (vi) implies (vii) and Lemma 9.3.3 that
(vii) implies (i). 
Second Proof of Theorem 9.3.1 (Subgroup Theorem) Immediate from Theo-
rem 9.3.2. 
Remarks 1. As a consequence of the KBD Theorem, 4.2.1, on shift-compactness,
Steinhaus’ Theorem stands between Theorems 9.3.2 (which employs shift-
compactness) and 9.3.1 (which Steinhaus’ Theorem implies).
2. The role of the finite covering property may be clarified by noting that, in the
context of the theorem above, one may validly add the property ‘S is closed’
yielding amended equivalent conditions (ii)0–(vii)0, and then obtain a further
equivalent condition:
(ix)0 S is closed and contains an interval around 0.
Indeed, if S is shift-compact, then it is closed. For suppose that s n → s0 with
s n ∈ S for all n. Then z n := s0 − s n is a null sequence and, as S is shift-compact,
there is s ∈ S such that s + (s0 − s n ) ∈ S infinitely often. But s n ∈ S and s ∈ S,
so s0 ∈ S. So S is closed. In particular, if S is closed and has the finite covering
property, then by Baire’s Theorem some coset of S contains an interval and so
S itself contains an interval, (a, b) say. So its midpoint, s say, is in S. Shifting
144 Duality

by −s, the set S contains an interval about 0, say (−δ, δ), yielding (vii)0. So,
S also contains (−nδ, nδ) for each n ∈ N, so S is R (as in the first proof of
Theorem 9.3.1).

9.4 Darboux Dichotomy

We recall the following result (Dar1875 – see Kucz1985) and give its proof, as
it is short.
Theorem 9.4.1 (Darboux’s Theorem) If f : R → R is additive and locally
bounded at some point, then f is linear.
Proof By additivity we may assume that f is locally bounded at the origin.
So we may choose δ > 0 and M such that, for all t with |t| < δ, we have
| f (t)| < M. For ε > 0 arbitrary, choose any integer N with N > M/ε. Now
provided |t| < δ/N, we have
N | f (t)| = | f (Nt)| < M, or | f (t)| < M/N < ε,
giving continuity at 0. Linearity easily follows (see, e.g., BGT1987, Th. 1.1.7).

We now formulate the classical Ostrowski Theorem in what we term its
strong form, as it includes both its classical measure-theoretic version and the
Baire analogue due to Banach (see the Remarks after Th. 9.4.2 and the two
cases in Corollary 9.4.5). Below, we say that f is locally bounded (locally
bounded above, or below) on a set S if any point t has a neighbourhood I such
that f is bounded (resp. above, or below) on S ∩ I.
When null sequences can be embedded in S by an arbitrary shift (not neces-
sarily in S) we say that S is shift precompact: see Section 10.1.
Theorem 9.4.2 (Strong Ostrowski Theorem) For a shift precompact set S, if
f : R → R is additive and bounded (locally, above or below) on S, then f is
locally bounded and hence linear.
Proof Suppose that f is not locally bounded in any neighbourhood of some
point x. Then we may choose z n → 0 such that f (x + z n ) ≥ n, without loss of
generality (otherwise replace f by − f ). So f (z n ) ≥ n − f (x). Since S is shift
precompact, there are t ∈ R and an infinite Mt such that
{t + z m : m ∈ Mt } ⊆ S,
implying that f is unbounded on S locally at t (since f (t + z n ) = f (t) + f (z n )),
a contradiction. So f is locally bounded, and by Darboux’s Theorem, 9.4.1, f
is continuous and so linear. 
 9.4 Darboux Dichotomy 145

Since any non-empty interval is shift precompact, Theorem 9.4.2 embraces

Darboux’s Theorem. A weaker result, with the condition S shift precompact
strengthened to S universal (see p. 73), was given by Kestelman (Kes1947b).
Theorem 9.4.2 is the basis of the Darboux dichotomy; that in turn is connected
with the Steinhaus dichotomy, because an additive function bounded on a Baire
non-meagre (measurable non-null) set A is bounded on the difference set A − A
and so on an interval contained in A − A. That is, the Darboux dichotomy
based upon a ‘thick set’, an interval, may have its basis refined to a thinner set,
just so long as the difference set is ‘thick’. F.B. Jones (Jon1942a) refined this
basis further by observing that it is enough for A to be analytic, so long as the
subgroup generated by A is the reals (see BinO2010a).
The boundedness conditions above lead naturally to a consideration of the
level sets of a function and their combinatorial properties. The classical measure
and category contexts appeal to various forms of localization. The nub is that,
when a non-negligible set is decomposed into a countable union of nice sets,
one of these is non-negligible. This is captured in the combinatorics below.
Here we go beyond the null sequences.

Definitions 1. For the function h : R → R, the (symmetric) level sets of h are

defined by
H r := {t : |h(t)| < r }.

2. For {Tk : k ∈ ω} a countable family of sets of reals, we write NT({Tk : k ∈

ω}) to mean that, for every bounded/convergent sequence {un } in R, some Tk
contains a translate of a subsequence of {un }, i.e. there are k ∈ ω, t ∈ R and
infinite Mt ⊆ ω, such that

{t + un : n ∈ Mt } ⊆ Tk .

The NT notation and the term No Trumps in Theorem 9.4.3, a combinatorial

principle, are used in close analogy with earlier combinatorial principles, in
particular Jensen’s Diamond ^ (Jen1972) and Ostaszewski’s Club ♣ (Ost1976;
see also Douw1992). Our proof of Theorem 9.4.3 makes explicit an argument
implicit in BinGo1982a, p. 482 (and repeated in BGT1987, p. 9), itself inspired
by CsiE1964 (see also BinO2009d; BinO2010h). The intuition behind our
formulation may be gleaned from forcing arguments in Mill1989; Mill1995.
Applied to the level sets, it is equivalent to the Uniform Convergence Theorem
(UCT), as is shown in BinO2009d; it also plays a key role in the theory of
subadditive functions, for which see BinO2010a.
We shall see that the NT property is a common generalization of both
measurability and the Baire property. (Specializing to the case when Tk = S for
146 Duality

all k, we see that S is shift precompact if and only if NT(S) holds.) This allows
a formulation of when a function may be regarded as having ‘nice’ level sets.
Our next result shows that NT captures classical notions of localization. Since
R is the union of the level sets of a function, we have as an immediate corollary
of Theorem 4.2.1.
Theorem 9.4.3 (No Trumps Theorem; cf. BinO2009b) For h : R → R mea-
surable or Baire, NT({H k : k ∈ ω}) holds.
As an illustration of its usefulness, we derive a common combinatorial
generalization with a weaker hypothesis. See also BinO2010a for a common
analysis of measurable and Baire subadditivity via NT.
Theorem 9.4.4 (Generalized Fréchet–Banach Theorem) If h : R → R is
additive and its level sets H k satisfy NT({H k : k ∈ ω}), then h is locally
bounded and so continuous and linear.
Proof Suppose that h is not locally bounded at the origin. Then we may
choose z n → 0 such that h(z n ) ≥ n, without loss of generality (if not replace h
by −h). But there are s ∈ R, k ∈ ω and an infinite Ms such that
{s + z m : m ∈ Ms } ⊆ H k ,
so
h(s + z m ) = h(s) + h(z m ) > h(s) + m,
so that h is unbounded on H k , a contradiction as |h| < k on H k . Thus h is
locally bounded and additive; hence by Darboux’s Theorem (cf. the proof of
Theorem 9.4.2) we conclude that h is continuous and so linear. 
By Theorem 9.4.3 this result embraces its classical counterpart for h mea-
surable or Baire (due to Fréchet in 1914 and Banach in 1920; see Kucz1985).
Corollary 9.4.5 (Fréchet–Banach Theorem) If f : R → R is additive and
measurable or Baire, then f is continuous and so linear.
There is also a further combinatorial generalization of the classical Ostrowski
Theorem, by reference to functions with ‘nice’ level sets.
Theorem 9.4.6 (Combinatorial Ostrowski Theorem; cf. BinO2009b) For
h(x) an additive function, h(x) is continuous and h(x) = cx for some constant
c if and only if NT({H k : k ∈ ω}) holds.
Proof If NT({H k : k ∈ ω}) holds, then by Theorem 9.4.4, h is linear.
Conversely, if h(x) = cx, then H k = {t : |ct| < k} is for each k = 1, 2, . . . an
interval and hence shift precompact. 
 9.5 The Fubini Null and the Kuratowski–Ulam Theorems 147

The Subgroup Theorem may also be similarly restated. For this, we need a
variant on NT(S) in which ‘for infinitely many’ is strengthened to ‘for all but
finitely many’ (‘co-finitely many’) denoted NTcof (S).
Theorem 9.4.7 (Combinatorial Steinhaus Theorem Restated) For an additive
subgroup S of R, the following are equivalent:
(i) S = R;
(ii)0 NTcof (S);
(iii)0 NT(S).

9.5 The Fubini Null and the Kuratowski–Ulam Theorems

Theorem 9.5.1 (Theorem FN: Fubini Theorem for Null Sets) Suppose G is a
metric group G and A ⊆ G2 is measurable under µ × ν, with µ, ν ∈ M(G). If
the ‘exceptional set’ of points x for which the vertical section Ax is ν-non-null
is itself µ-null, then A is (µ × ν)-null.
Proof For µ-null N ⊆ G, the set N × G is (µ × ν)-null, so (by passing to the
complement of the null exceptional set of the theorem) we may assume without
loss of generality that the exceptional set of A is empty. By inner regularity, it
suffices to show that (µ × ν)(K ) = 0 for all compact K ⊆ A.
For K compact, denote by F the (compact) projection of K on the first axis.
Let ε > 0. By compactness, for any x ∈ F there is an open neighbourhood Ux
of x and an open Vx with ν(Vx ) < ε and
K ∩ (Ux × G) ⊆ Rx := Ux × Vx .
By compactness of F, there are U j × V j for j = 1, . . . , n, with U j , V j open and
ν(V j ) < ε such that
[ [
F⊆ Uj : K⊆ Uj × V j.
j j

To disjoin the sets U j , put

[ [ [
S j := U j \ Uj : Uj = Sj.
j<i j j

Then
[ [
F= F ∩ Sj : K⊆ Sj × V j.
j j

So
X X X
µ(K ) ≤ (µ × ν)(S j × V j ) = µ(S j )ν(V j ) ≤ µ(S j ) · ε = ε µ(F).
j j j

As ε > 0 was arbitrary, µ(K ) = 0. 

148 Duality

We do not need the group-theoretic assumptions above (in 9.5.1). The con-
verse of Theorem 9.5.1 also holds: for a µ × ν-null set A, the aforementioned
exceptional set is µ-null. For a proof (in the more general setting of σ-finite
measures), see Hal1950, §36, Th. A, p. 147.
The following is the classical category analogue in which M(X ) denotes the
family of all meagre subsets of the space X and, for E ⊆ X × Y and x ∈ X, Ex
denotes the x-section of E, i.e.
Ex := {y ∈ Y : (x, y) ∈ E}.
In its original form the Kuratowski–Ulam Theorem of 1932 asserted as
follows.
Theorem 9.5.2 (Kuratowski–Ulam Theorem; KurU1932) For X, Y metric
spaces with Y separable:
E ∈ M(X × Y ) =⇒ {x ∈ X : Ex < M(Y )} ∈ M(X ).
In fact their proof requires only that Y have a countable pseudo-base (π-
basis for short), i.e. a family U of non-empty open subsets of Y such that any
non-empty open set W in Y contains a set U ∈ U.
Thus generalized, Theorem 9.5.2 may be restated to refer to N W D, the
family of nowhere dense sets (Oxt1980, Ch. 15).
Theorem 9.5.3 (Kuratowski–Ulam Theorem Variant) For X, Y topological
spaces with Y having a countable pseudo-base
E ∈ N W D(X × Y ) =⇒ {x : Ex < N W D(Y )} ∈ M(X ).
In the non-separable realm the Kuratowski–Ulam Theorem does not hold in
general, unlike its measure counter-part. Examples of failure were by Pol in
Pol1979 and later (in a vector space setting) by van Mill and Pol (MilP1986). A
converse holds for Baire subsets E ⊆ X ×Y asserting that if Ex is meagre for all
but a meagre set of x, then E is meagre. For the proof, see (Oxt1980, Th. 15.4).

9.6 * Universal Kuratowski–Ulam Spaces

Since the conclusions of the Kuratowski–Ulam Theorem may fail in more
general circumstances, it is of interest to determine when it may nevertheless
hold. Here we follow the approach taken by Fremlin, Natkaniec and Recław
(FreNR2000) as well as Fre2002. With this aim, we will call a pair of topological
spaces (X, Y ) a Kuratowski–Ulam pair (briefly, K–U pair) if the Kuratowski–
Ulam Theorem holds in X × Y :
K–U: If E ∈ M(X × Y ), then {x ∈ X : Ex < M(Y )} ∈ M(X ).
 9.6 * Universal Kuratowski–Ulam Spaces 149

Definition A topological space Y is called a universally Kuratowski–Ulam

space (uK–U space for short) if (X, Y ) is a K–U pair for any topological space X.

So, according to the Kuratowski–Ulam Theorem, every space Y with a count-

able π-basis is a uK–U space. Note also that every space Y that is meagre in
itself is uK–U. Fremlin, Natkaniec and Recław (FreNR2000) showed that there
are uK–U Baire spaces without a countable π-basis, and further that every
Baire uK–U space satisfies the countable chain condition. Known as ‘ccc’ for
short, this means that every family of pairwise disjoint open sets in Y is at most
countable. Examples of Baire ccc spaces exist which are not uK–U.
A strengthening of the uK–U notion is particularly important (see Theorem
9.6.1). We will say that (X, Y ) is a Kuratowski–Ulam* pair (K–U∗ pair, for
short) if the variant formulation, as in Theorem 9.5.3, of the Kuratowski–Ulam
Theorem holds for the pair (X, Y ) :

K–U∗ : If E ∈ N W D(X × Y ), then{x : Ex < N W D(Y )} ∈ M(X ).

Definition A topological space Y is called a universally Kuratowski–Ulam*

space (uK–U∗ space for short) if (X, Y ) is a K–U∗ pair for any topological space
X; see Fremlin (Fre2000a) and Kucharski and Plewik (KucP2007), who give an
example of a uK–U space which is not uK–U∗ . See also KucP2007; KalK2015.

Every K–U∗ pair is a K–U pair, and so every uK–U∗ space is uK–U space.
Indeed, a space Y is uK–U if and only if its Baire part, Y \L(Y ), is uK–U∗ .
Here the light part (the analogous term heavy appears in §8.2) is defined by
S
L(Y ) := {V : V is open and meagre in Y }. Evidently, for Y a Baire space the
latter is empty, so for Y Baire: Y is K–U if and only if it is K–U∗ , and similarly
Y is uK–U if and only if it is uK–U∗ .
It emerges that a finite product of uK–U spaces is also a uK–U space (see
FreNR2000). For uK–U∗ spaces one has the stronger result.

Theorem 9.6.1 An arbitrary product of uK–U∗ spaces is uK–U∗ . In particular

2κ , for an arbitrary cardinal κ, is uK–U∗ .

We give Fremlin’s proof below, assuming the following lemma, whose proof
we will omit, as that would require further developments. The result allows the
proof to focus on compact spaces which are uK–U∗ .

Lemma 9.6.2 A topological space is uK–U∗ if and only if the Stone space of
its Boolean algebra of regular open sets is also uK–U∗ .

For the proof see Fre2002, Prop. 4(b); for background on Stone spaces, see
Joh1982.
150 Duality

Before beginning the proof of Theorem 9.6.1, we state and prove some
preliminaries.
Fix κ, S, Z and f as in the hypothesis, and let Φ := {2 A : A ∈ [κ]<ω }
S
denote the family of indicator functions φ of finite subsets of κ. Any φ ∈ Φ
defines the following basic open subset of 2κ :
U (φ) := {y ∈ 2κ : φ ⊂ y}.
We denote by U the family of all such basic open sets in 2κ . Note that U (ψ) ⊂
U (φ), for φ, ψ ∈ Φ with φ ⊂ ψ. A set U ⊂ S is basic open in S if U = Ũ ∩ S
for some basic open set Ũ ⊂ 2κ . Say that a set A ⊂ 2κ is determined by a set of
coordinates τ ⊂ κ if A = {y ∈ 2κ : y | τ ∈ A∗ } for some A∗ ⊂ 2τ .
Denote the closure and interior of A ⊆ 2κ by cl( A) and int( A), and the
closure and interior of A in S by clS ( A), and intS ( A).
We now follow Fremlin’s exposition and begin with two claims.
Claim 1 If Y = i ∈I Yi is a product of compact uK–U∗ spaces, and E ⊆ X ×Y
Q
nowhere dense, then the following is meagre:
An := {x ∈ X : V ⊆ Ex and V ∈ τn },
where τn is the family of non-empty open sets in Y determined by at most n
coordinates (from I).
Proof of Claim 1 (We proceed by induction for fixed E.) For n = 0, V = Y .
Here A0 × Y ⊆ E, so A0 is in fact nowhere dense. For the step from n − 1 to n,
suppose that An is not meagre. Then An contains a heavy point x, i.e. for each
U open in X with x ∈ U, the set U is non-meagre. As E is nowhere dense, U ×Y
contains an open set U 0 × V disjoint from E with U 0 open in X and V basic in
Y , i.e. determined by a finite set of coordinates J. For x ∈ U 0 ∩ An choose J (x)
comprising at most n members of I determining a non-empty open V (x) ⊆ Ex .
Now V (x) ∩ V = ∅ as U 0 × V is disjoint from E and V (x) ⊆ E, so V (x) and
V must disagree on some determining coordinate, and such a coordinate lies in
J (x) ∩ J. But J is finite, so there is j ∈ J with the following set non-meagre:
An0 = {x ∈ An ∩ U 0 : j ∈ J (x)}.
Fix such a j and view E as a subset of (X × Yj ) × Y j with Y j := i,j Yi .
Q
Denoting by π j projection from Y onto Yj , consider
B := {(x, u) ∈ X × Yj : x ∈ An0 and u ∈ π j [V (x)]}.
Here Yj is Baire and uK–U and Bx contains an open set for each x in the
non-meagre set An0 . So B is non-meagre. For (x, u) ∈ B, the set E(x,u) contains
V (x)u, a non-empty open set determined by coordinates in J (x)\{ j), i.e. by
 9.6 * Universal Kuratowski–Ulam Spaces 151

at most n − 1 coordinates. This contradicts that An−1 is meagre. In turn this

implies that An is meagre, establishing the inductive step. 

Claim 2 Any product of compact Hausdorff uK–U∗ spaces Yi is uK–U∗ .

S
Proof To see this, proceed as in Claim 1 and note that n An is meagre. But the
latter union is just the set {x : int Ex , ∅}, proving the present implication. 

Proof of Theorem 9.6.1 Given uK–U∗ spaces {Yi }i ∈I with topologies τi, put
Y = i ∈I Yi . Now pass to the (coarser) regular open topology σi on Yi generated
Q
by ρi the family of regular open sets, i.e. those of the form int(cl(U)) for U
open in τi . Let Zi be the Stone space of the regular open algebra of Yi . Then Zi
is a compact uK–U∗ space and so is their product Z := i ∈I Zi . Now consider
Q
the basis of the Tikhonov product corresponding to the τi and ρi topologies:
(Y )
V:= Vi : Vi ∈ τi and {i : Vi , Yi } finite ,
i ∈I
(Y )
W:= Vi : Vi ∈ ρi and {i : Vi , Yi } finite .
i ∈I

The base W generates the product topology σ of the topologies σi which,

being order isomorphic (for inclusion) to Z, makes (Y, σ) be uK–U∗ . From
here, the lemma implies that (Y, τ) is uK–U∗ . 

Theorem 9.6.3 If S is a dense subspace of 2κ , Z a regular topological space

and f : S → Z a continuous surjection, then Z is a uK–U space. If, further, Z
is Baire, then Z is a uK–U∗ space.

As a first step we prove

Lemma 9.6.4 Any regular open set W ⊂ S is a countable union of basic open
sets in S.

Proof The space 2κ has the ccc property and so likewise has S, being a dense
subspace. Choose a maximal sequence {Bn }n<ω of basic open sets in 2κ such
S S
that the complement in S of S∩ n<ω Bn is nowhere dense, so that W ∩ n<ω Bn
is a dense subset of W . Each Bn is determinedS by some τn ∈ [κ]<ω . Put
τ := n<ω τn ∈ [κ] ≤ω . Then cl(W ) = cl n<ω Bn is determined by τ. So
S
in particular its subset int(cl(W )) is also determined by τ and so, being open,
takes the form n<ω Un , for some basic open sets Un in 2κ .
S
But W is regular open in S, so W = intS (clS (W )) = S ∩ int(cl(W )), and so
W = S ∩ n<ω Un .
S


Corollary 9.6.5 If V is a non-meagre open set in Z, then there exists a basic

open set W ⊂ S such that f [W ] is non-meagre and f [W ] ⊂ V .
152 Duality

Proof Since Z is regular, there exists a non-meagre open V 0 with cl Z (V 0 ) ⊂ V ,

as follows. For each z ∈ V, by regularity, choose an open set Vz such that
z ∈ Vz ∈ cl Z (Vz ) ⊂ V . Now Vz for some z must be non-meagre, otherwise all
the sets Vz are meagre and by the Banach Category Theorem, V = x Vx is
S
meagre, contradicting that V is non-meagre.
As f is continuous, W 0 := f −1 [V 0] is non-empty open in S and so W 0 ⊆
W := intS (clS (W 0 )), with W regular open in S. Now V 0 = f (W 0 ) ⊆ f (W ) ⊆ V
and so also f (W ) is non-meagre. By Lemma 9.6.4, W = n<ω Wn with {Wn }
S
a sequence of basic open sets in S, so f (Wn ) is, for some n, non-meagre. 

Proof of 9.6.3 Returning to the proof of the theorem, consider an arbitrary

topological space X and let E ⊂ X × Z be a closed nowhere dense set. Let P
be the set of all pairs (G, I) where G is an open set in X and I ∈ [κ]<ω . Define
a relation ≺ on P by requiring (H, J) ≺ (G, I) if and only if

(i) H ⊂ G and J ⊃ I, and,

(ii) for each basic open set W ⊂ S determined by I, either
• H × f [W ] ⊂ E, or
• (H × U) ∩ E = ∅ for some basic open set W0 ⊂ W determined by J, and
open set U ⊂ Z with f [W0 ] ⊂ U.

Claim For any (G, I) ∈ P and any non-empty open set G0 ⊂ G there exists
(H, J) ∈ P such that (H, J) ≺ (G, I) and H ⊂ G0 .

Proof of the Claim Put n = |I | and let {Wi : 0 < i ≤ 2n } list all the finitely
many basic open sets that I can determine. Put J0 = I and define inductively
open sets Gi and Ji for 0 < i ≤ 2n according to which of the following two
cases is determined by Wi .
Case 1. If Gi−1 × f [Wi ] ⊂ E, set Gi = Gi−1 and Ji = Ji−1 .
Case 2. As E is nowhere dense, there are open sets Gi ⊂ Gi−1 and Ui ⊂ Z with
(Gi × Ui ) ∩ E = ∅ with Ui ∩ f (Wi ) , ∅, and so a basic open set Wi0 in S with
Wi0 ⊂ Wi with f [Wi0] ⊂ Ui . Let Ji ∈ [κ]<ω be the set which determines Wi0 .
Finally, set H = G2n ⊆ G0 and J = i ≤2n Ji ⊇ J0 = I.
S
Now choose inductively a sequence Pn ⊂ P such that

• P0 = {(X, ∅)}.
• If (H, J), (H 0, J 0 ) are distinct members of Pn , then H ∩ H 0 = ∅.
• For (H, J) ∈ Pn+1 there exists (G, I) ∈ Pn such that (H, J) ≺ (G, I).
• Pn+1 is a maximal family which satisfies the conditions above.
 9.6 * Universal Kuratowski–Ulam Spaces 153

Then all the sets G∗n = {H : (H, J) ∈ Pn } are open and dense, and so
S
T ∗
n<ω G n has meagre complement in X.
Fix any x ∈ n<ω G∗n . We will show that Ex is nowhere dense in Z. For this
T
it is suffices to consider any (non-meagre) open set V ⊂ Z and to show that
there exists a non-empty open set V 0 ⊂ V with Ex ∩ V 0 = ∅. Fix a non-meagre
open set V ⊂ Z. By Corollary 9.6.5 there exists a basic open set W0 ⊂ S such
that f [W0 ] ⊂ V is non-meagre. Let the set determining W0 be J ∈ [κ]<ω . Given
x, we may choose a sequence {(Hn, Jn )}n such that, for each n:

• (Hn, Jn ) ∈ Pn ;
• x ∈ Hn ;
• (Hn+1, Jn+1 ) ≺ (Hn, Jn ), so in particular Jn+1 ⊃ Jn .

Being the set that determines W0, J is finite and so there exists n with Jn+1 ∩ J =
Jn ∩ J. Since f [W0 ] is non-meagre and Jn determines a finite partition of S,
there exists an open basic set W determined by Jn such that f [W ∩ W0 ] is not
meagre. Hence f [W ] is not meagre.
Since Hn+1 × cl( f [W ])) ⊆ cl(Hn+1 × f [W ]) and has non-empty interior and
by assumption E is closed nowhere dense, Hn+1 × f [W ] * E. That is, Case 1
above does not occur.
So, by Case 2, there exists W 0 ⊂ W , a basic open set of S determined by
Jn+1 , and an open set U ⊂ Z such that (Hn+1 × U) ∩ E = ∅ and f [W 0] ⊂ U.
Now W 0 ∩ W0 , ∅, since Jn+1 ∩ J = Jn ∩ J : indeed, writing W0 = U (τ0 ),
W = U (τn ), and W 0 = U (τn+1 ) with τn+1 ⊃ τn, we have τn+1 | J = τn | J = τ0 | J
as W ∩ W0 , ∅; so, taking τ = τ0 ∪ τn+1 , we have U (τ) ⊆ W 0 ∩ W0 .
This implies that ∅ , f [W 0 ∩W0 ] ⊆ U ∩V as f [W0 ] ⊂ V and f [W 0] ⊂ U. So
U ∩V , ∅, and also (U ∩V )∩Ex = ∅, the latter since both (Hn+1 ×(U ∩V ))∩E =
∅ and x ∈ Hn+1 hold, as required, proving the claim and so the theorem. 

Corollary 9.6.6 There exists a uK–U∗ Baire space Y without a countable

π-basis.

Proof Consider Y = 2ω1 . By Theorem 9.6.1, Y is a uK–U∗ space. On the

other hand, it is well known that π(Y ) = ω1 . In fact, let {Un : n < ω} be a
sequence of basic open sets in Y . For each n there exists An ∈ [ω1 ]<ω and
φn : An → 2 such that Un = U (φn ). Then A = n<ω An is countable so we
S
may choose α ∈ ω1 \A and take V := {y ∈ Y : y(α) = 1}. Then V is open in Y ;
however, Un ⊆ V fails for each n since there are points y ∈ Un with y(α) = 0.
So {Un : n ∈ ω} is not a π-basis for Y . 

A compact space X is said to be dyadic if it is a continuous image of the

space 2κ for some cardinal κ (cf. Eng1989, p. 285). Theorem 9.6.3 implies
154 Duality

Corollary 9.6.7 Every dyadic space is uK–U∗ .

A topological space X is said to be quasi-dyadic if it is a continuous image

of the Tikhonov product α Xα of a family {Xα : α < κ} of metric separable
Q
spaces (see FreG1995).

Theorem 9.6.8 Every regular quasi-dyadic space is uK–U (and uK–U∗ if

Baire).

Proof We start with the following lemma.

Lemma 9.6.9 Every metric separable space is the continuous image of some
dense subset of the space 2ω .

Proof of Lemma This is a consequence of the fact that every metric separable
space is homeomorphic to a subspace of the Hilbert cube I ω (see, e.g., Kec1995,
Th. 4.14, p. 22) and that I ω is a continuous image of 2ω . Thus every metric
separable space is a continuous image of some subspace of 2ω . On the other
hand, it is easy to prove that every subset of a Cantor set is the continuous
image of some dense subset of 2ω . 

To complete the proof of Theorem 9.6.8, assume that Y is a regular space,

Xα, α < κ, are metric separable spaces, and f : α<κ Xα → Y is a continuous
Q
surjection. For every α < κ there exists a continuous surjection f α : Aα → Xα
for Aα some dense subspace of 2ω . Then the set α<κ Aα is dense in 2ωκ and
Q
Q Q
f ◦ α<κ f α is a continuous surjection from α<κ Aα onto Y . By Theorem
9.6.3, Y is a uK–U space. 

The next result refers to the additivity of M(X ), a cardinal invariant defined
by:
( [ )
add(M(X )) = min |D| : D ⊂ M(X ) and D < M(X ) .

For background, see BartoJ1995, Ch. 5, cf. the Cichoń diagram of BartoJ1995,
Ch. 2 and Chapter 16. A topological space X satisfies the κ-chain condition, in
brief: is κ-cc, if there is no family of size κ of open, pairwise disjoint sets in X.
The countable chain condition (briefly, ccc) is thus ω1 -cc.
Similarly one may define π-weight of X, denoted π(X ), to be the least
cardinal of a π-basis for X.
Remark The proof of the Kuratowski–Ulam Theorem yields the generalization
that if π(Y ) < add(M(X )), then (X, Y ) is a K–U∗ pair.

Theorem 9.6.10 Assume that X is a non-meagre space, Y a Baire space and

(X, Y ) a K–U pair. Then Y is add(M(X ))-cc.
 9.6 * Universal Kuratowski–Ulam Spaces 155

Proof Suppose that κ = add(M X )) and B = {Bα : α < κ} is a family

of open, non-empty, pairwise disjoint sets in Y . Let A = { Aα : α < κ} be
S
a family of nowhere dense sets in X with A < M (X ). In X × Y, take
S
W := α<κ Aα × Bα . Note that W is nowhere dense in X × Y . Indeed, for a
S
basic open set U × V two cases may arise. First, if V0 := V \ clY ( α<κ Bα ) , ∅,
then U × V0 is non-empty open and disjoint from W . Otherwise V0 = ∅. In this
case, as V is open, V ∩ Bα , ∅ for some α < κ and, since Aα is nowhere dense,
there is an open, non-empty set U 0 ⊂ U\Aα . Thus the set U 0 × (V ∩ Bα ) is
non-empty, open and disjoint from W . On the other hand,
[
{x : Wx < M (Y )} = A < M (X ),
and so (X, Y ) is not a K–U pair. 
Remark There exist completely regular spaces X non-meagre in themselves
with add(M (X )) = ω1 . Specifically, as is well known, X = 2ω1 has this
property: indeed, the sets Eα = {x ∈ X : x(ξ) = 0 for ξ ≥ α} are closed and
S
nowhere dense in X, but α<ω1 Eα < M (X ). Another example is provided by
the space (ωω, τd ) from Example 1 .
Since a Baire space is uK–U if and only if it is uK–U∗ , we have the following.
Corollary 9.6.11 Every Baire uK–U∗ space satisfies the ccc property.
Now we will show that the assumption of ccc for a Baire space Y is not
sufficient to make it be uK–U∗ .
For s ∈ ω <ω and f ∈ ωω with s ⊂ f , define
(s, f ) = {g ∈ ωω : s ⊂ g and f ≤ g}.
Note that the family of such pairs forms a basis for a ccc topology τd on ωω .
It is known that (ωω, τd ) is a completely regular, Baire space. Moreover, let τ
denote the standard topology on ωω . For f , g ∈ ωω we use the symbol f ≤∗ g
to mean that {n ∈ ω : f (n) > g(n)} is finite.
Example 1 ((ωω, τ), (ωω, τd )) and ((ωω, τd ), (ωω, τd )) are not K–U∗ pairs.
To see this, first define W := {( f , g) ∈ (ωω ) 2 : f ≤∗ g}. We will then prove
two claims.
Claim 1 W is meagre in the topologies τd × τd and τ × τd .
Proof of Claim 1 Put Wn = {( f , g) ∈ ωω × ωω : (∀k > n) f (k) ≤ g(k)}. We
verify that each Wn is nowhere dense in the topology τd × τd . Let (s, f )×(r, h) be
a basic set. Fix k > n such that k < dom(s) ∪dom(r). Choose s1, r 1 ∈ ω <ω such
that s ⊂ s1, r ⊂ r 1 , s1 (k) > r 1 (k), s1 ≥ f | dom(s1 ), and r 1 ≥ h | dom(r 1 ).
156 Duality

Let f 1 be any extension of s1 with f 1 ≥ f and h1 be any extension of r 1 with

h1 ≥ h.
Then (s1, f 1 ) × (r 1, h1 ) ⊂ (s, f ) × (r, h). Observe that e(k) > g(k) for each
(e, g) ∈ (s1, f 1 ) × (r 1, h1 ). Thus (s1, f 1 ) × (r 1, h1 ) ∩ Wn = ∅, so Wn is nowhere
dense, and consequently W is meagre in the topology τd × τd .
Similarly we can prove that W is meagre in the topology τ × τd . Claim 1 is
therefore proved.
Claim 2 W f < M(τd ) for each f ∈ ωω .
Proof of Claim 2 Note that W f = {h : f ≤∗ h}. Fix a basic set (s, g) and
define g1 ∈ ωω such that g1 (i) = h(i) if i ∈ dom(s) and g1 (i) = max(h(i), f (i))
otherwise. Then (s, g1 ) ⊂ (s, g) ∩ W f . Thus W f is co-meagre in the topology
τd , and Claim 2 is proved, and with it the statement of Example 9.6.
Corollary 9.6.12 The space (ωω, τd ) is not a uK–U∗ space.
We also have another better known example of a ccc space which is not
uK–U∗ . Let D denote the density topology on the real line. Recall that (R, D)
is a Baire space with the ccc property, and A ⊂ R is D-nowhere dense if
and only if it is D-meagre if and only if m( A) = 0. Here m denotes Lebesgue
measure. (See Chapter 7 for the basic properties of the topology D; cf. Oxt1980
and Tal1976.)
Example 2 For X = (R, D) the pair (X, X ) is not a uK–U∗ pair.
To see this, consider
A = {(x, y) : x − y < Q}.
As is easily seen, both A and its complement are D × D-dense (this is a
consequence of Steinhaus’ Theorem, 9.2.2, see also AnaL1985). Moreover, A is
a Gδ subset of the plane with full Lebesgue measure, so it contains a Euclidean
closed set E (so closed also in the D × D topology) with positive measure. To
finish, the set E is nowhere dense in (R2, D × D) and, by Fubini’s Theorem,
{x : Ex < M(D)} = {x : m(Ex ) > 0} < M(D).

Below we present further results concerning universal Kuratowski–Ulam∗

(uK–U∗ ) spaces, omitting routine proofs; the majority of them (except Property
7) holds for uK–U.

Applications. Recall that the product X ×Y of Baire spaces may be non-Baire.

(Some conditions for X and Y which imply that X × Y is a Baire space are
described in HawoM1977.) Note that if X and Y are Baire spaces and (X, Y ) is
 9.6 * Universal Kuratowski–Ulam Spaces 157

a K–U pair, then X × Y is a Baire space. Similarly, the product X × Y of a Baire

space X and a uK–U∗ Baire space Y is a Baire space.
Property 1 Any subspace of a uK–U∗ (uK–U) space is itself uK–U∗ (uK–U,
respectively).
Property 2 If Y0 is a dense subspace of a uK–U∗ space Y , then it is also a
uK–U∗ space.
Property 3 Assume that Y0 is a subspace of a uK–U∗ space Y such that
Y0 ⊂ intY (clY (Y0 )). Then Y0 is also a uK–U∗ space.
Example 3 There exists a subspace Y0 of a uK–U∗ space Y , which fails to be
a uK–U∗ space.
Proof Take Y0 to be the discrete space of size ω1 . As Y0 has weight ω1 , it
embeds into Y = [0, 1]ω1 (see, e.g., Eng1989, Th. 2.3.11, p. 113). By Theorem
9.6.8, Y is uK–U∗ , but Y0 is not ccc, so it is not uK–U∗ , by Corollary 9.6.11. 
We say that a set A ⊂ X is nowhere meagre in a space X if U ∩ A < M(X )
for every open, non-meagre set U ⊂ X.
Property 4 Suppose that Y0 is a uK–U∗ dense subspace of a space Y . If Y0 is
nowhere meagre in Y, then Y is a uK–U∗ space.
The assumption about Y0 cannot be omitted.
Example 4 There exists a non-uK–U∗ space Y with a dense uK–U subspace
Y0 .
Proof Let Y be any complete dense-in-itself metric space which is not ccc.
By Corollary 9.6.11, Y is not a uK–U∗ space. For every n > 0, choose a discrete
set Yn ⊂ Y which forms a 1/n-net in Y . Then Y0 = n>0 Yn is dense in Y , dense
S
in itself, and meagre in itself. Thus Y0 is a uK–U space. 
Property 5 Suppose that {Yi : i < ω} is a sequence of uK–U∗ subspaces of a
topological space Y . Then i Yi is also a uK–U∗ space.
S

Corollary 9.6.13 The topological sum of countably many uK–U∗ spaces is a

uK–U∗ space.
Example 5 The topological sum of uncountably many uK–U∗ spaces may
fail to be a uK–U∗ space.
Proof Let Y be a discrete space of size ω1 . Then Y is not ccc, so is not a
uK–U∗ space. On the other hand, every singleton is a uK–U∗ space. 
Property 6 The homeomorphic image of a uK–U∗ space is also a uK–U∗
space.
158 Duality

Property 7 The image of a uK–U∗ Baire space under an open continuous

function is a uK–U∗ space.

Note that any space Y is a continuous image of the meagre-in-itself space

Y × Q. Thus any space Y is a continuous image of a uK–U∗ space.
Remark The results above lead to the problem whether any continuous image
of a uK–U∗ Baire space is also uK–U∗ . This problem has been solved by
D. Fremlin (Fre2000a) in the negative.

9.7 * Baire Products

As Kuratowski and Ulam noted, the set E = [0, 1] N is nowhere dense in
[0, 2] N yet its projection [0, 1] is non-meagre in [0, 2]. (For any basic open set
U1 × U2 × · · · × Un × [0, 2] {n+1,n+2,... } the open subset U1 × U2 × · · · × Un ×
(1, 2) × [0, 2] {n+2,... } is disjoint from E.) In general the product of two Baire
spaces need not be Baire: Oxtoby constructed an example of failure (using the
Continuum Hypothesis, though absolute examples followed) and also identified
an additional condition, that of having a locally countable pseudo-base which
guarantees that X × Y is Baire if X is Baire and Y has such a pseudo-base. This
condition on a space is sometimes referred to by saying that the space has a
countable-in-itself π-base, as the defining property of the π-base is that each
of its members contains countably many members of the said π-base.
We have seen that if X and Y are Baire spaces and (X, Y ) is a K–U pair, then
X × Y is a Baire space. Similarly, the product of a Baire space X and a uK–U
Baire space Y is a Baire space. This leads to the notion of a space X being
almost locally uK–U when the set

W := {x ∈ X : x has an open uK–U neighbourhood}

is dense in X. Under such circumstances W is dense in X and open, i.e. its

complement is closed nowhere dense. (If it contained an open subset V, this
would meet W .)
It emerges that the property of being almost locally uK–U is a proper general-
ization of having a locally countable pseudo-base (countable-in-itself π-base).
Say that a space is almost locally ccc, provided every open set contains an
open ccc subspace, i.e. if the space has a π-base of open ccc subspaces. This
property is strictly weaker than being almost locally uK–U.
However, in the context of metrizable spaces all these concepts are equivalent,
as we shall see.
 9.7 * Baire Products 159

Below we refer to the largest closed set comprising points at which the
space is non-separable. Known as the nowhere-locally-separable kernel, it was
introduced by A. H. Stone (Sto1963).
Theorem 9.7.1 (Pol’s Product Theorem; Pol1979) For X, Y metrizable spaces,
(a) and (b) below are equivalent:
(a) If A, B are non-meagre in respectively X and Y, then A × B is non-meagre
in X × Y .
(b) At least one of X or Y has a nowhere-locally-separable kernel that is
meagre.
In particular, if (a) fails and there are non-meagre sets A, B with A × B meagre,
then both of their nowhere-locally-separable kernels are meagre.
Since the uK–U property is inherited by open subspaces and since spaces
with countable bases are uK–U (see §9.6 – cf. FreNR2000), it follows that if
X is a uK–U space or has a countable-in-itself base, then X is almost locally
uK–U. For X metrizable the converse also holds.
Corollary 9.7.2 If X is metrizable, Baire and almost locally uK–U, then X
has a countable-in-itself base.
Proof Let U ⊆ X be an open uK–U subspace with X\U meagre. Let Y be a
nowhere-locally-separable Baire metric space, so that Y is its own non-meagre
nowhere-locally-separable kernel. We claim that X has a meagre nowhere-
locally-separable kernel. For this it suffices to prove that (a) above holds. So
suppose A ⊆ X and B ⊆ Y with A × B meagre in X × Y and A, B non-meagre.
We will derive a contradiction, namely that A is meagre.
Now, U × Y ∩ A × B is a meagre subset of U × Y , so, since U is uK–U,
quasi all b ∈ B have meagre section. But B is non-meagre, so there is y ∈ B
such that U ∩ A = {x ∈ U : (x, y) ∈ A × B} is a section that is meagre in U
and hence also meagre in X, since U is open. As X\U is meagre, A is meagre
in X. Now the points in X without a separable neighbourhood form a closed
meagre subset, by Pol’s Theorem, 9.7.1. So, being a Baire space, X has a dense
open locally separable subspace. Locally separable metrizable spaces can be
partitioned into clopen separable subspaces, and so taking the union of their
respective countable bases yields a countable-in-itself base for X. 
The Krom space below allows the application of Pol’s Theorem.
Definition For a topological space (Y ,τ) let τ̃ := τ\{∅}, the non-empty open
sets, which we view as the points of a discrete (metrizable) space. We give
the space τ̃ ω of sequences the metric of first difference (just as with NN ). The
160 Duality

Krom space is the subspace of τ̃ ω corresponding to descending sequences of

open sets with non-empty intersection:
K (Y ) := f ∈ τ̃ ω :
( \ )
f (n) , ∅ and f (0) ⊇ f (1) ⊇ · · · ⊇ f (n) ⊇ · · · .
n∈ω
One writes f ∈↓ τωwhen f (0) ⊇ f (1) ⊇ · · · ⊇ f (n) ⊇ · · · , and by analogy
f ∈↓ τ n for some n < ω if f ∈ τ̃ n and f (0) ⊇ f (1) ⊇ · · · ⊇ f (n). The basic
open set generated by f ∈↓ τ n in K (Y ), denoted [ f ], comprises all infinite
extensions of f .
Theorem 9.7.3 (Krom’s Theorem; Kro1974) For X, Y Baire, X × Y is Baire
if and only if X × K (Y ) is Baire if and only if K (X ) × K (Y ) is Baire.
For a proof see Kro1974. The result extends to arbitrary products (LiZ2017,
Th. 4.1).
Theorem 9.7.4 (Zsi2004) For X, Y Baire spaces with Y almost locally ccc,
X × Y is a Baire space.
Proof First note that the Krom space K (Y ) has a countable-in-itself π-base:
indeed, let f ∈↓ τ n for some n < ω, choose U ⊂ f (n) which is ccc, and
define the extension f U = f aU. Consider a pairwise disjoint open partition
{[g] : g ∈ J} of [ f U ], for an arbitrary J ⊂ n < ω ↓ τ n . For each g ∈ J
choose ng < ω with g ∈↓ τ ng . Then {g(ng ) : g ∈ J} is a pairwise disjoint open
partition of U, hence countable, since Y is ccc and likewise {[g] : g ∈ J}. Thus,
K (Y ) is an almost locally ccc metric space, and so has a countable-in-itself
π-base, by Corollary 9.7.2.
By Krom’s Theorem K (Y ) is a Baire space, so by Oxt1960, Th. 2, X × K (Y )
is a Baire space, implying that X × Y is a Baire space, again by Krom’s
Theorem. 
This result generalizes to arbitrary products (LiZ2017, Th. 1.2).
Theorem 9.7.5 (Zsi2004) For X, Y Baire spaces with Y almost locally ccc,
X × Y is a Baire space.
Theorem 9.7.6 For {Xi : i ∈ I} an arbitrary family of almost locally ccc
Q
Baire spaces, the product i Xi is a Baire space.
 10

Category Embedding Theorem and Infinite

Combinatorics

Motivated by the Kestelman–Borwein–Ditor Theorem, 4.2.1, we focus on the

concept of shift-compactness, as it has a key role in unifying the Baire and
measurable approaches. For convenience, we recall the term here (taken, as we
have said earlier, from the probability literature – see Par1967 and BinO2024
– and adapted to our context) in tandem with two related concepts as follows,
the weaker of which is important for applications.

10.1 Preliminaries
Recall that quasi everywhere (q.e.), or for quasi all points, means for all points
off a meagre set. We will use for generically all to mean for quasi all in the
category case, and for almost all in the measure case.
Definitions 1. Say that S ⊆ R is boundedly shift-compact, resp. convergently
shift-compact (shift-compact for bounded sequences in R, resp. in S), and write
S ∈ SK R , resp. S ∈ SK , if, for any bounded/convergent sequence un , there
are t ∈ R and infinite M = Mt such that
(i) {t + um : m ∈ M} ⊆ S, and
(ii) limM (t + um ) ∈ S.
2. As earlier, say that S ⊆ R is shift-compact (shift-compact for null se-
quences), S ∈ S∗ , if, for any null sequence z n → 0, there are t ∈ S and infinite
M = Mt such that
(i) {t + z m : m ∈ M} ⊆ S, and
(ii) t = limM (t + z m ) ∈ S.
3. As in §9.4, say that S ⊆ R is shift precompact, and write S ∈ S, if, for any
null sequence z n → 0, there are t ∈ R and infinite M = Mt such that

161
162 Category Embedding

(i) {t + z m : m ∈ M} ⊆ S.

Note that Definition 1, in contrast to Definition 2, does not require t ∈ S.

The choice of the asterisk notation in Definition 2 is suggested by genericity
(cf. Kec1995, 17.26, and the Near-Closure Theorem, 10.5.1). It is clear that
SK R ⊆ S∗ and SK R ⊆ SK ; writing S for the family of closed sets in S, we
have the following proposition.

Proposition 10.1.1 If A ⊆ R is shift-compact, then A is boundedly shift-

compact and so convergently shift-compact:

S ⊆ S ∗ = SK R ⊆ SK and S∗ ⊆ S.

Proof Let an be a convergent sequence (sequence in A) with limit a0 . Then

z n := an − a0 is a null sequence, hence for some t ∈ A and infinite Mt we have
t + z n in A for n ∈ Mt . Thus with s := t − u0 we have s + un = t + z n ∈ A for
n ∈ Mt and convergence through Mt to s + u0 = t ∈ A. Thus A is boundedly
shift-compact (convergently shift-compact). 

These are forms of compactness (for a topological analysis of this insight

involving open shifted-covers, and further applications, see BinO2010g). They
generalize their forerunner universality in relation to null sequences, introduced
in a related context by Kestelman (Kes1947a), where the more demanding
requirement on the set M in Definition 3 above is that it be co-finite. The
latter concept of universality is implicit in some of Banach’s work (see, e.g.,
Ban1932).
The Near-Closure Theorem, 10.5.1, implies that a Baire non-meagre/measur-
able non-null set T is shift-compact. Although the weakest of the three concepts,
shift precompactness is the key combinatorial concept in many applications.

10.2 The Category Embedding Theorem, CET

Theorem 10.2.2 is a topological version of the Kestelman–Borwein–Ditor The-
orem, 4.2.1, from which Theorem 10.3.3 is rederived. The latter is a (home-
omorphic) embedding theorem (see, e.g., Eng1989, p. 67); Trautner uses the
term covering principle in Tra1987 in relation essentially to bounded shift-
compactness (see again BGT, p. 10). For other generalizations of a homotopic
nature, see Mille1989 (which inspired BinO2011b). We need the following
definition.
 10.2 The Category Embedding Theorem, CET 163

Definition (Category Convergence) A sequence of Baire functions hn : X →

X satisfies the category condition (cc) if, for any non-empty open set U, there
is a non-empty open set V ⊆ U such that, for each k ∈ ω,
\
V \h−1
n (V ) is meagre. (cc)
n≥k

Equivalently, for each k ∈ ω, there is a meagre set M such that, for t < M,

t ∈ V =⇒ (∃n ≥ k), hn (t) ∈ V .

We will see in Theorem 10.2.4 that this is a weak form of convergence to

the identity and indeed Theorems 10.3.1 and 10.3.10 verify that, for z n → 0,
the homeomorphisms hn (x) := x + z n satisfy (cc) in the Euclidean and in the
density topologies. However, it is not true that hn (x) converges to the identity
pointwise in the sense of the density topology; furthermore, whereas addition (a
two-argument operation) is not D-continuous (see Property 7 of D in Chapter
7), translation (a one-argument operation) is.
In Theorem 10.2.2, the topological space X may be assumed to be non-
meagre (of second category) in itself, and the Baire set T to be non-meagre, as
otherwise there is nothing to prove. To verify that X is non-meagre, one would
typically assume that X is a Baire space (cf. Chapter 2). The proof makes use
of the monotone operator (where ‘i.o.’ means ‘infinitely often’)

n (T ) = {x : x ∈ h n (T ) i.o.}
bh (T ) := lim sup h−1 −1
\ [
= h−1
n (T )
k ∈ω n≥k

associated with a sequence h of Baire functions. The following result has a

routine proof so is omitted. Base operators referred to below are discussed in
§8.5.

Lemma 10.2.1 Suppose h = hhn i is a sequence of Baire functions on X of a

space X. Then bh (T ) is additive and is in fact a base operator.

Thus bh may be used to refine the topology of X to obtain an analogue of the

z-topology on R defined in §8.5, Remark 4b, p. 129.

Theorem 10.2.2 (Category Embedding Theorem – CET; BinO2009b) Let X

be a topological space and hn : X → X be Baire functions satisfying (cc) with
preimages of meagre sets being meagre. Then, for any Baire set T, for quasi all
t ∈ T there is an infinite set Mt such that

{hm (t) : m ∈ Mt } ⊆ T .

In particular, the conclusion holds for hn homeomorphisms satisfying (cc).

164 Category Embedding

Proof Take T Baire and non-meagre. We may assume that T = U\M with U
non-empty and open and M meagre. Let V ⊆ U satisfy (cc). Since preimages
of meagre sets under hn are meagre, the set
[
M 0 := M ∪ h−1
n (M)
n

is meagre. Put
\ [
W = h(V ) := V ∩ h−1
n (V )
k ∈ω n≥k
= lim sup[h−1
n (V ) ∩ V ]
= {x : x ∈ h−1
n (V ) ∩ V i.o.}
⊆ V ⊆ U.

So for t ∈ W we have t ∈ V and

vm := hm (t) ∈ V, (*)

for infinitely many m – for m ∈ Mt , say. Now W is co-meagre in V . Indeed

[ \
V \W = V \h−1
n (V ),
k ∈ω n≥k

which by (cc) is meagre.

Take t ∈ W \M 0 ⊆ U\M = T, as V ⊆ U and M ⊆ M 0 . Thus t ∈ T . For
m ∈ Mt , we have t < h−1
m (M), since t < M and h m (M) ⊆ M ; but vm = h m (t),
0 −1 0

so vm < M. By (*), vm ∈ V \M ⊆ U\M = T . Thus {hm (t) : m ∈ Mt } ⊆ T for t

in a co-meagre subset of V .
To deduce that quasi all t ∈ T satisfy the conclusion of the theorem, put
S := T\h(T ). Then S is Baire since

m (UM) = h m (U)\h m (M),

h−1 −1 −1

both sets being Baire (the first since hm is Baire and U is open, the second as
it is meagre), and S ∩ h(T ) = ∅. If S is non-meagre, then by the preceding
argument there are s ∈ S and an infinite Ms such that {hm (s) : m ∈ Ms } ⊆ S,
i.e. s ∈ h(S) ⊆ h(T ), a contradiction. (This last step is an implicit appeal to a
generic dichotomy – see Chapter 4.) 

Corollary 10.2.3 (Quasi Identity) Suppose h = hhn i is a sequence of self-

homeomorphisms of a space X, and T is a Baire set in X. Then

bh (T ) = lim sup h−1

n (T ) ∼ T (modulo the meagre sets).

Clearly the theorem relativizes to any open subset of T; that is, the embed-
ding property is a local one. The following theorem sheds some light on the
significance of the category convergence condition (cc). The result is capable
 10.3 Kestelman–Borwein–Ditor Theorem: CET Proof 165

of improvement, by reference to more general (topological) countability condi-

tions. (Typically these lift category and measure arguments out of the classical
context of separable metric spaces; see Chapter 2, or Eng1989, §§3.9, 4.4 for
an account of Čech-completeness and metrization theory, and Arh1963, §7
for an account of p-spaces, their common generalization.) Here, for instance,
a σ-discrete family could replace the countable family B of the theorem as
the generator of the coarser topology. Such a replacement would offer a route
to Bing’s Metrization Theorem, given sufficient regularity assumptions – see
Eng1989, Th. 4.4.8, thus making X submetrizable, or even cometrizable.

Theorem 10.2.4 (Convergence to the Identity) Assume that the homeomor-

phisms hn : X → X satisfy the category convergence condition (cc) and that
X is a Baire space. Suppose there is a countable family B of open subsets of
X which generates a (coarser) Hausdorff topology on X . Then, for quasi all
(under the original topology) t, there is an infinite Nt such that

limm∈Nt hm (t) = t.

Proof For U in the countable base B of the coarser topology and for k ∈ ω,
select open Vk (U) so that Mk (U) := n≥k Vk (U)\h−1
T
n (Vk (U)) is meagre. Thus
[ [
M := Mk (U)
k ∈ω U ∈B

is meagre. Now Bt = {U ∈ B : t ∈ U} is a basis for the neighbourhoods of t.

But, for t ∈ Vk (U)\M, we have t ∈ h−1m (Vk (U)) for some m = mk (t) ≥ k, i.e.
hm (t) ∈ Vk (U) ⊆ U. Thus hmk (t) (t) → t, for all t < M. 

10.3 Kestelman–Borwein–Ditor Theorem: CET Proof

We now deduce the category and measure cases of the Kestelman–Borwein–
Ditor Theorem, 4.2.1, restated below, as two corollaries of Theorem 10.2.2 by
applying it first to the usual and then to the density topology on the reals, R.
For our first application we take X = R with the density topology, a Baire
space. Let z n → 0 be a null sequence. Put

hn (x) := x − z n, so that h−1

n (x) = x + z n .

The topology is translation-invariant, and so each hn is a homeomorphism.

To verify the category convergence of the sequence hn, consider U non-empty
and D-open; then consider any measurable non-null V ⊆ U. To verify (cc)
in relation to V, it now suffices to prove the following result, which is of
independent interest (cf. Littlewood’s First Principle, §1.1).
166 Category Embedding

Theorem 10.3.1 (Verification Theorem for D) Let V be measurable and

non-null. For any null sequence {z n } → 0 and each k ∈ ω,
\
Hk := V \(V + z n ) is null, so meagre in the D-topology.
n≥k

Proof Suppose otherwise. Then for some k, we have |Hk | > 0. Write H for
Hk . Since H ⊆ V, it follows, for n ≥ k, that ∅ = H ∩ h−1
n (V ) = H ∩ (V + z n ),
and so a fortiori ∅ = H ∩ (H + z n ).
Let u be a density point of H. Thus, for some interval Iδ (u) := (u − δ/2, u +
δ/2), we have
3
|H ∩ Iδ (u)| > δ.
4
Let E = H ∩ Iδ (u). For any z n, we have |(E + z n ) ∩ (Iδ (u) + z n )| = |E| > 43 δ.
For 0 < z n < δ/4, we have |(E + z n )\Iδ (u)| ≤ |(u + δ/2, u + 3δ/4)| = δ/4. Put
F = (E + z n ) ∩ Iδ (u); then |F | > δ/2.
But δ ≥ |E ∪ F | = |E| + |F | − |E ∩ F | ≥ 43 δ + 12 δ − |E ∩ F |. So

1
|H ∩ (H + z n )| ≥ |E ∩ F | ≥ δ,
4
contradicting ∅ = H ∩ (H + z n ). This completes the proof. 

A similar but simpler proof establishes the following result, which implies
(cc) for the Euclidean topology on R; here for given open U we may take any
open interval V ⊆ U.

Theorem 10.3.10 (Verification Theorem for E) Let V be an open interval in

R. For any null sequence {z n } → 0 and each k ∈ ω,
\
Hk := V \(V + z n ) is empty.
n≥k

We now re-state and re-prove Theorem 4.2.1. As with the Category Embed-
ding Theorem, the set T here may be assumed to be non-meagre/non-null, since
otherwise there is nothing to prove.

Theorem 10.3.3 (Theorem KBD – Kestelman–Borwein–Ditor Theorem) Let

{z n } → 0 be a null sequence of reals. If T is Baire/Lebesgue measurable, then,
for generically all t ∈ T, there is an infinite set Mt such that

{t + z m : m ∈ Mt } ⊆ T .

Proof Theorem 10.2.2 may be applied to hn (x) := x + z n in view of Theorem

10.3.1 or 10.3.10, respectively, in the category/measure cases. 
 10.4 Kestelman–Borwein–Ditor Theorem: Second Proof 167

10.4 Kestelman–Borwein–Ditor Theorem: Second Proof

Our second proof in this chapter (and third proof within the book, with Theorem
4.2.1) depends on the interplay of the z-topology (see Remark 4b in §8.5) and
the density topology D (see Chapters 7 and 8). We include this third proof
here because (**) and Lemma 10.4.1 below are needed for the results of §10.5.
We recall some salient features. The set of density points of T is denoted
by φ N (T ), where this notation refers to the σ-ideal N of Lebesgue null sets
and the fact that φ is a lower-density operator (Chapter 8). (See CieL1990
for a discussion of density topologies generated by σ-ideals.) By the Lebesgue
Density Theorem, 7.1.1, almost all points of a measurable set are density points
and so φ N (φ N (T )) = φ N (T ). See Sze2011 for a study of the exceptional points
E N (T ).
If u is a density point of both S and T it follows from the definition that u
is a density point of S ∩ T . It is this fact that justifies the introduction of the
D-topology, the density topology, on R. We thus have

D = {T ∈ L : T ⊆ φ N (T )},

introduced in GofW1961 (see also HauP1952) and studied also in GofNN1961

(cf. CieLO1994, and, for a textbook treatment, Kec1995).
For z = {z n } → 0 any null sequence, the z-topology is that defined by
reference to the base operator
\[
bz (T ) := (T − z n ). (**)
k ∈ω n≥k

We will write
z(T ) := T ∩ bz (T ).

Lemma 10.4.1 (Fundamental Lemma on Genericity) Let z = {z n } → 0 be

any null sequence. Suppose that T ∈ TN , i.e. T is density-open (measurable
and every point of T is a density point of T). Then u is a density point of bz (T )
for any u ∈ T; in symbols:

u ∈ φ N (z(T )) = φ N (T ∩ bz (T )).

Proof Let u ∈ T . Write Iδ (u) := (u − δ, u + δ). Then u is a density point of T

(since T ∈ D) and hence, for any ε > 0, there is δ < ε such that
|T ∩ Iδ (u)|
≥ (1 − ε),
2δ
i.e. nearly all of Iδ (u) is in T. Let η = 2δε. For n > N , |z n | < η and so putting
Tn = T ∩ (T − z n ),
168 Category Embedding

|Tn ∩ Iδ (u)| ≥ (1 − ε)2δ − 2η,

i.e. nearly all of Iδ (u) is in T − z n . Similarly, for k > N,

[
Tn ∩ Iδ (u) ≥ (1 − ε)2δ − 2η.
n≥k

Hence we have (cf. ErdKR1963)

\[
|z(T ) ∩ Iδ (u)| = Tn ∩ Iδ (u) ≥ (1 − 2ε)2δ,
k ∈ω n≥k

i.e. nearly all of Iδ (u) is covered by z(T ). Thus u is a density point of z(T ) and
of course z(T ) ⊂ T . 

Corollary 10.4.2 For u ∈ T arbitrarily close to u, there is a point t ∈ z(T ),

i.e. a point t such that t ∈ T and {t + z m : m ∈ Mt } ⊆ T, for some infinite Mt .

We turn to the proof of the Kestelman–Borwein–Ditor Theorem.

Proof of the Kestelman–Borwein–Ditor Theorem In the measure case, denote

the set of density points of S by T and apply the Fundamental Lemma on
Genericity, 10.4.1.
In the Baire case, if S = I\M ∪ M 0, where I is an interval and M, M 0 are
meagre, take T = I\M. We will show that for any {un } → u ∈ T, there are
v ∈ T and an infinite Mv such that

{v + um : m ∈ Mv } ⊆ S.

Select δ > 0 so that J = (u − δ, u + δ) ⊆ I. We wish to pick v, vn < M, with

the aim of later putting vn := v + un . This means we require in particular that
v + un < M. So pick v in J to avoid the meagre set
[
M∪ M − un .
n∈ω

Now, for n large enough, un ∈ J. By choice, v, vn < M. Hence, for large enough
n, we have v, vn ∈ T, as required. 

10.5 Near-Closure and No Trumps

Here we work at first in R. Recall that a set A is z-closed if and only if every
z-translator into A (i.e. t ∈ R such that t + z m ∈ A infinitely often) is in A and
that the z-topology is translation invariant (see Remark in §8.5). This notion
 10.5 Near-Closure and No Trumps 169

enables us to strengthen Theorem 10.3.3 to say that any non-meagre Baire/non-

null measurable set A is almost z-closed for any null sequence z, i.e. every
translator into A is in A, modulo a meagre/null set, depending on whether the
z-topology is regarded as refining E or D. Thus A is ‘nearly closed’ under the
z-topology.

Theorem 10.5.1 (Near-Closure Theorem) For any null sequence z, and A

a (non-meagre) Baire or (non-null) measurable set, A is almost z-closed, i.e.
modulo a meagre/null set every z translator into A is in A.

Proof By Theorem 10.3.3 T\bz (T ) is null (since t ∈ bz (T ) if and only if

t + z m ∈ T for a subsequence z m ). Recalling the definition (**) of the base
operator, the Fundamental Lemma, 10.4.1, implies that T ⊆ Φ N (T ∩ bz (T )) ⊆
Φ N (bz (T )). But Φ N (bz (T )) differs from bz (T ) by a null set (by the Density
Theorem, 7.1.1), hence bz (T )\T is also null. 

It suffices to consider the density topology case.

The theorem has two important corollaries. Our first result involves bound-
edly shift-compact sets.

Theorem 10.5.2 (Shift-Compactness Theorem) For any non-negligible Baire/

measurable set T and any bounded sequence hun i, for quasi all t ∈ T there is
an infinite Mt such that {t + (um − u0 ) : m ∈ Mt } ⊆ T for some limit u0 of the
sequence.

Proof Suppose that hun i is bounded. Passing to a subsequence, we may assume

that hun i converges, say to u0 . Then z n := un − u0 → 0 and so, for quasi all
t ∈ T, there is a subsequence hz m im∈M such that t + z m ∈ T, i.e. (t −u0 ) +um ∈ T
for m ∈ Mt . Thus, a shift of a subsequence of hun i lies in T . 

Corollary 10.5.3 The restriction to L of bz (.) is an outer density.

Further generality may be achieved by assuming somewhat less than almost

z-closure for any z. We have in mind a sense of ‘largeness’, of the following
type: for a countable sequence of sets in the class L (or Ba), if their union
is ‘large’ so is one of them. The following definition replaces null sequences
z = hz n i with convergent or bounded sequences u = hun i, and the translator t
may be arbitrary.
The origin of the NT terminology below (Jensen’s ♦ and Ostaszewski’s ♣)
was clarified in §9.4 from where we recall:

Definition (No Trumps) For a sequence {Tk : k ∈ ω} of subsets of Rd , the

property NT({Tk : k ∈ ω}) asserts that:
170 Category Embedding

for any bounded/convergent sequence u = hun i in Rd there are t ∈ Rd , an

index k ∈ ω and an infinite M ⊆ ω such that
{t + un : n ∈ M} ⊆ Tk . (NT)
In words: for every bounded/convergent sequence u = hun i in Rd , some Tk
contains a translate of a subsequence of hun i.
A localized version may be obtained by considering only convergent
sequences.
Definition (Local No Trumps) For a sequence {Tk : k ∈ ω} of subsets of Rd ,
the property NTL ({Tk : k ∈ ω}) asserts that:
for any convergent sequence hun i in Rd with limit u0 , arbitrarily close to u0
there are t ∈ Rd , an index k ∈ ω, and an infinite M ⊆ ω with
{t + un : n ∈ M} ⊆ Tk .
In words: for every bounded/convergent sequence hun i in Rd, some Tk contains
a translate of a subsequence of hun i.
Lemma 10.5.4 In words: for every convergent sequence hun i in Rd with limit
u0, some Tk contains a translate of a subsequence of hun i arbitrarily close to
u0 .
Proof As in Theorem 10.5.2 (the Shift-Compactness Theorem), if un → u0 ,
then z n := un − u0 → 0. As T − u0 is z-closed for any z, then s + z m ∈ T − u0
for m ∈ M for some infinite M and some s ∈ T − u0 . Hence s + z m ∈ T − u0,
i.e. s + um ∈ T for m ∈ M. 
Remark We also have lim(s + um ) = s + u0 ∈ T.
We will need the following result, implicit in CsiE1964.
Theorem 10.5.5 (Strong No Trumps Theorem; CsiE1964) If T is a non-
meagre Baire/non-null measurable set in any interval which it meets and T =
k ∈ω Tk with each Tk measurable/Baire, then NTL ({Tk : k ∈ ω}) holds.
S
Indeed, for every convergent sequence {un } → u0 ∈ T, any neighbourhood of
the limit u0 contains a point s for which there exist K = K (s) ∈ ω and an
infinite set M = M(s) ⊆ ω such that
s + um ∈ TK for m ∈ M.
Proof Suppose un converges to u0 . Consider any interval I = (u0 − η, u0 + η)
with η > 0. As T meets I, for some K ∈ ω, the set TK ∩ I is measurable and
non-null (resp. Baire non-meagre). Let z n := un − u. Then z n → 0 and so,
 10.5 Near-Closure and No Trumps 171

by the Kestelman–Borwein–Ditor Theorem, for almost all (resp. for quasi all)
t ∈ TK ∩ I, there is an infinite set Mt such that
{t + z m : m ∈ Mt } ⊆ TK ∩ I.
For any such t put s = t − u. Then writing M = M(s) for Mt , we have
{s + um : m ∈ Mt } ⊆ TK . 
 11

Effros’ Theorem and the Cornerstone Theorems

of Functional Analysis

11.1 Introduction
We give a short proof of a classic theorem of Effros (Eff1965) stated in a form
which holds also beyond its original separable context, namely the general
metrizable context. All that is required for the general context is a single
modification to the continuity assumptions of group action (see Definition (1)
below). Effros’ Theorem asserts that a continuous transitive action by a Polish
group G on a non-meagre space X is open: the point-evaluation g 7→ gx is an
open map from G to X . Viewed, despite its original separability, as a group-
action counterpart to one of the cornerstones of functional analysis, the Open
Mapping Theorem (OMT) (that a surjective continuous linear map between
Fréchet spaces is open – cf. Rud1973), it has come to be called the Open
Mapping Principle (OMP) – see Anc1987, §1.
The paradigm for OMP was an earlier result, due to Glimm (Gli1961),
which was restricted to locally compact groups and was directed at resolving a
conjecture of Mackey concerned with representation of a C ∗ -algebra A on (an
infinite-dimensional) Hilbert space H. A key notion was that of a quotient space
Ai /G, for Ai the irreducible representations and G the group of unitary operators
on H, being countably separated (separation of points by a countable family of
Borel sets): briefly, smooth. Group action takes (g, x) to the g-conjugate g −1 xg
of x, for g, x ∈ H, i.e. similarity, and smoothness is an equivalent of openness.
We note that the underlying (unitary) equivalence of representations, viewed
as a set in Ai × Ai , is closed (so Borel). Glimm observed that the openness
property excludes the group of those sequences of 0s and 1s which have all but
a finite number of their terms equal to 0 when this group acts on the space of
all sequences of 0s and 1s by coordinate-wise addition (modulo 2). In another
context this is the Vitali equivalence (difference modulo rationals), for which
see Kha2004. This is the nub of the Glimm–Effros dichotomy in the realm of

172
 11.1 Introduction 173

Borel equivalences: either an equivalence is smooth or it embeds in itself the

Vitali equivalence; see Harrington–Kechris–Louveau (HarrKL1990).
An altogether different area received a huge boost from the Effros result of
1965 some 10 years after its publication: the theory of homogeneous continua
(the origins of which hark back to Montgomery; Mon1950). Ungar (Ung1975)
in 1975 discovered that if (X, d) is a homogeneous compact metric space, then
the group H (X ) of self-homeomorphisms of X equipped with the compact-
open topology acts microtransitively on X, meaning that for each δ > 0 there
is an ‘Effros number’ e(δ) > 0 such that for any x, y ∈ X with d(x, y) < e(δ),
there is h ∈ H (X ) with h(x) = y and with d(h(z), z) < δ for all z ∈ X (a ‘δ-
push’). A slew of papers followed in the next decade, with early contributions
from F.B. Jones (see Jon1975) and Hagopian, J.T. Rogers, Jr, W. Lewis, Phelps,
Kennedy. See again Anc1987, §1 and CharM1966.
Not quite a decade later still, van Mill (Mil2004) offered both a general-
ization of the OMP, extending it to (separable) groups that are analytic (i.e.
continuous images of a Polish space), and a clever counterexample to a va-
riety of conjectures (Mil2008). By van Mill’s theorem of Mil2004, separable
metrizable groups that are analytic and have continuous action are microtran-
sitive (as above) on any non-meagre separable metrizable space. A co-analytic
example of a continuous group action which is not microtransitive exists under
the Axiom of Constructibility (V = L, see Chapter 16), but under the Axiom of
Determinacy (AD; see Chapters 14 and again 16) no such example can exist.
Under AC there always exist such examples. See Med2022.
To include a ‘non-separable’ context requires in place of ‘global’ countability
a more ‘local’ notion: a sequential property related to the Steinhaus-type Sum–
Set Theorem (that 0 is an interior point of A − A, for non-meagre A with BP, the
Baire property – (Pic1939; Pic1942); see Chapter 15), because of the following
argument (which goes back to Pettis, Pet1950).
Consider L : E → F, a linear, continuous surjection between Fréchet spaces,
and U a neighbourhood of the origin. Choose A an open neighbourhood of the
origin with A − A ⊆ U; as L( A) is non-meagre (since {nL( A) : n ∈ N} covers
F) and has the Baire property (see Proposition 11.4.3 in §11.4), L( A) − L( A)
is a neighbourhood of the origin by the Sum–Set Theorem. But of course
L(U) ⊇ L( A) − L( A),
so L(U) is a neighbourhood of the origin. So L is an open mapping.
The sequential approach followed here is based on Ost2015b. Throughout this
chapter, without further comment, all spaces considered will be metrizable. We
recall the Birkhoff–Kakutani Theorem (Chapter 6; cf. HewR1979, II.8.6, pp.
70, 83), that a metrizable group G with neutral element eG has a right-invariant
174 Effros’ Theorem

metric d RG . Passage to kgk := d RG (g, eG ) yields a (group) norm (invariant

under inversion, satisfying the triangle inequality), which justifies calling these
normed groups; any Fréchet space qua additive group, equipped with an F-norm
(KaltPR1984, Ch. 1, §2), is a natural example (cf. the autohomeomorphism
group Auth, see p. 176). Below we need the following.

Definitions 1. (Cf. Pet1950) For G a metrizable group, say that the group
action ϕ : G × X → X has the Nikodym property or, briefly, is Nikodym, if
for every non-empty open neighbourhood U of eG and every x ∈ X, the set
U x = ϕ x (U) := ϕ(x, U) contains a non-meagre Baire set.
2. Aq denotes the quasi-interior of A – the largest open set U with U\A meagre
(cf. Ost2011, §4).

Thanks to the Nikodym property, introduced above, we are able to state

the Effros Theorem in a form that embraces both the separable and the non-
separable contexts. We establish the property first in the simpler separable
context in §11.2 and then again in the non-separable context in Chapter 13. The
main results below are Theorems 11.1.1 and 11.1.2 (for later convenience, also
referred to symbolically as Theorems Sh and E resp.), with corollaries in §11.4
including OMT; see below for commentary.

Theorem 11.1.1 (Theorem Sh, Shift-Compactness Theorem) For T a Baire

non-meagre subset of a metric space X and G a group, Baire under a right-
invariant metric, and with separately continuous and transitive Nikodym action
on X:

for every convergent sequence x n with limit x and any Baire non-meagre
A ⊆ G with eG ∈ Aq and Aq x ∩ T q , ∅, there are α ∈ A and an integer N
such that αx ∈ T and
{α(x n ) : n > N } ⊆ T .

In particular, this is so if G is analytic and all point-evaluation maps ϕ x : g →

g(x) are base-σ-discrete.

This theorem has wide-ranging consequences, including Steinhaus’ Sum–

Set Theorem; see the survey article Ost2013a; Ost2013b and Chapter 14. See
also BinO2024.

Theorem 11.1.2 (Theorem E, Effros’ Theorem – Baire Version) If

(i) the normed group G has separately continuous and transitive Nikodym
action on X, and
(ii) G is Baire under the norm topology and X is non-meagre,
 11.2 Action, Microtransitive Action, Shift-Compactness 175

then for any open neighbourhood U of eG and any x ∈ X the set U x := {u(x) :
u ∈ U} is a neighbourhood of x, so that in particular the point-evaluation maps
g 7→ g(x) are open for each x. That is, the action of G is microtransitive.
In particular, this holds if G is Polish and all point-evaluation maps ϕ x are
continuous.
More generally, this holds if G is analytic and Baire, and all point-evaluation
maps ϕ x are continuous and base-σ-discrete (for which see Chapter 12). This
last property holds automatically when G is separable.
By Proposition 11.2.3 X, being non-meagre here, is also a Baire space.
The classical counterpart of Theorem E has G a Polish group; van Mill’s
version (Mil2004) requires the group G to be analytic (i.e. the continuous image
of some Polish space). The Baire version above improves the version given in
Ost2013b, where the group is almost complete. (The two cited sources taken
together cover the literature.)
A result due to Hoffmann-Jørgensen (Hof1980, Th. 2.3.6, p. 355) asserts that
a Baire, separable, analytic topological group is Polish (as a consequence of an
analytic group being metrizable – for which see again Hof1980, Th. 2.3.6), so
the analytic separable case of Theorem E reduces to its classical version.
Unlike the proof of the Effros Theorem attributed to Becker in Kec1994,
Th. 3.1, the one offered here does not employ the Kuratowski–Ulam Theorem
of §9.5 (the Category version of the Fubini Theorem), a result known to fail
beyond the separable context (as shown in Pol1979, cf. MilP1986, but see
FreNR2000 and Chapter 9).
For further commentary (connections between convexity and the Baire prop-
erty, relation to van Mill’s separation property in Mil2009, certain specializa-
tions) see §11.4.

11.2 Action, Microtransitive Action, Shift-Compactness

We recall some group-related notions.
A normed group G acts continuously on X if there is a continuous mapping
ϕ : G × X → X such that ϕ(eG, x) = x and ϕ(gh, x) = ϕ(g, ϕ(h, x)) for x ∈ X
and g, h ∈ G. The action ϕ is separately continuous if g : x 7→ ϕ(g, x) is
continuous for each g and ϕ x : g 7→ ϕ(g, x) is continuous for each x; in such
circumstances:
(i) the elements g ∈ G yield autohomeomorphisms of X via g : x 7→ g(x) :=
ϕ(g, x) (as g −1 is continuous); and
(ii) point evaluation of these homeomorphisms, ϕ x (g) = g(x), is continuous.
176 Effros’ Theorem

In certain situations joint continuity of action is implied by separate continuity

(see Bouz1993 and literature cited in Ost2012).
The action is transitive if for any x, y in X there is g ∈ G such that g(x) = y.
For later purposes (§11.3), say that the action of G on X is weakly microtran-
sitive if, for x ∈ X and each neighbourhood A of eG , the set
cl( Ax) = cl{ax : a ∈ A}
is a neighbourhood of x. The action is microtransitive (‘transitive in the small’
– for details see Mil2004) if for x ∈ X and each neighbourhood A of eG the set
Ax = {ax : a ∈ A}
is a neighbourhood of x. This (norm) property implies that U x is open for
U open in G (i.e. that here each ϕ x is an open mapping). We refer to Ax as
an x-orbit (the A-orbit of x). The following group action connects the Open
Mapping Theorem to the present context.
Example (Induced Homomorphic Action) A surjective, continuous homo-
morphism λ : G → H between normed groups induces a transitive action of G
on H via ϕλ (g, h) := λ(g)h (cf. Ost2012, Th. 5.1), specializing to
ϕ L (a, b) := L(a) + b
for G, H Fréchet spaces (regarded as normed, additive groups) and λ = L : G →
H linear (Anc1987; Mil2004). Of course for Fréchet spaces, by the Open
Mapping Theorem itself, ϕ L has the Nikodym property.
Definitions 1. Auth(X ) denotes the autohomeomorphisms of a metric space
(X, d X ); this is a group under composition. H (X ) comprises those h ∈
Auth(X ) of bounded norm:
khk := supx ∈X d X (h(x), x) < ∞.
2. For a normed group G acting on X, say that X has the crimping property
(property C for short) with respect to G if, for each x ∈ X and each sequence
{x n } → x, there exists in G a sequence {gn } → eG with gn (x) = x n . (This
and a variant occur in Ban1932, Ch. III; Th. 4; CharC2001; for the term see
BinO2009a.)
For a subgroup G ⊆ H (X ), say that X has the crimping property with
respect to G if X has the crimping property with respect to the natural action
(g, x) → g(x) from G × X → X . (This action is continuous relative to the left
or right norm-topology on G – cf. Dug1966, XII.8.3, p. 271.)
3. As a matter of convenience, say that the Effros property (or property E) holds
for the group G acting on X if the action is microtransitive, as above.
 11.2 Action, Microtransitive Action, Shift-Compactness 177

4. For a subgroup G ⊆ Auth(X ) say that X is G-shift-compact (or, shift-

compact under G) if, for any convergent sequence x n → x 0 , any open subset U
in X, and any Baire set T co-meagre in U, there is g ∈ G with g(x n ) ∈ T ∩ U
along a subsequence (cf. Chapter 10, p. 161). Call the space shift-compact if it
is H (X )-shift-compact (cf. MilleO2012; Ost2013c).
In such a space, any Baire non-meagre set is locally co-meagre (co-meagre
on open sets) in view of Proposition 11.2.3.

We shall prove in §11.3.1 equivalence between the Effros and crimping

properties.

Theorem 11.2.1 (Theorem EC) The Effros property holds for a group G
acting on X if and only if X has the crimping property with respect to G.

We now clarify the role of shift-compactness.

Proposition 11.2.2 For any subgroup G ⊆ H (X ), if X is G-shift-compact,

then X is a Baire space.

Proof We argue as in Mil2004, Prop 3.1(1). Suppose otherwise; then X

contains a non-empty meagre open set. By Banach’s Category Theorem, the
union of all such sets is a largest open meagre set M, and is non-empty. Thus
X\M is a co-meagre Baire set. For any x ∈ M the constant sequence x n ≡ x is
convergent and, since X\M is co-meagre in X, there is g ∈ G with g(x) ∈ X\M.
But, as g is a homeomorphism, g(M) is a non-empty open meagre set, so is
contained in M, implying g(x) ∈ M, a contradiction. 

A similar argument gives the following and clarifies an assumption in The-

orem 11.1.2 (cf. Mil2004; Hof1980, Prop. 2.2.3).

Proposition 11.2.3 If X is non-meagre and G acts transitively on X, then X

is a Baire space.

Proof As above, refer again to M, the union of all meagre open sets, which,
being meagre, has non-empty complement. For x 0 in this complement and
any non-empty open U pick u ∈ U and g ∈ G such that g(x 0 ) = u. Now
as g is continuous, g −1 (U) is a neighbourhood of x 0, so is non-meagre, since
every neighbourhood of x 0 is non-meagre. But g is a homeomorphism, so
U = g(g −1 (U)) is non-meagre. So X is Baire, as every non-empty open set is
non-meagre. 
178 Effros’ Theorem

11.2.1 Nikodym Actions: Separable Context

The following result is usually a first step in proving the weakly microtransitive
variant of the classical Effros Theorem (cf. Anc1987, Lemma 3; Ost2013a,
Th. 2). Indeed, one may think of it as giving a form of ‘very weak microtran-
sitivity’. We will later see a direct generalization to the non-separable context:
cf. Mil2004, Lemma 3.2.
Proposition 11.2.4 If G is a separable normed group acting transitively
on a non-meagre space X with each point-evaluation map ϕ x : g 7→ g(x)
continuous, then, for each non-empty open U in G and each x ∈ X, the set U x
is non-meagre in X.
In particular, if G is analytic, then G is a Nikodym action.

Proof We first work in the right norm-topology, i.e. derived from the assumed
right-invariant metric d RG (s, t) = kst −1 k. Suppose that u ∈ U, and so without
loss of generality assume that U = Bε (u) = Bε (eG )u (open balls of radius
some ε > 0): indeed,
Bε (eG )u = {xu : d RG (x, e) < ε} = {z : d RG (zu−1, e) < ε}
= {z : d RG (z, u) < ε} = U.
Now put y := ux and W = Bε (eG ). Then U x = Bε (eG )ux = W y. Next work
in the left norm-topology, derived from d LG (s, t) = ks−1 t k = d RG (s−1, t −1 ) (for
which W = Bε (eG ) is still a neighbourhood of eG ). As each set hW for h ∈ G
is now open (since now the left shift g → hg is a homeomorphism), the open
family W = {gW : g ∈ G} covers G.
As G is separable, the cover W has a countable subcover, say V. Thus
X := {V y : V ∈ V }, as X = Gy, and so V y is non-meagre for some V ∈ V,
S
say for V = V̂ . As V is a subcover, there is some ĝ ∈ G with V̂ = ĝW, so
V̂ y = ĝW y, and so ĝW y is non-meagre. As ĝ −1 is a homeomorphism of X,
W y = U x is also non-meagre in X.
If G is analytic, then as U is open, it is also analytic (since open sets are Fσ
and Souslin-F subsets of analytic sets are analytic, cf. RogJ1980), and hence
so is ϕ x (U). Indeed, since ϕ x is continuous, U x is analytic, so Souslin-F , and
so Baire by Nikodym’s Theorem. 

11.3 Effros and Crimping Properties: E, EC and C

11.3.1 Proof that E ⇐⇒ C
We first show that if the Effros property holds for the action of a group G on
X, then X has the crimping property with respect to G. Indeed, suppose that
 11.3 Effros and Crimping Properties: E, EC and C 179

x = lim x n . For each n, take U = B1/n

G
(eG ); then U x := {u(x) : u ∈ U} is an
open neighbourhood of x, and so there exists hn,m ∈ U with hn,m (x) = x m
for all m large enough, say for all m > m(n). Without loss of generality we
may assume that m(1) < m(2) < · · · . Put hm := eG for m < m(1), and for
m(k) ≤ m < m(k + 1) take hm := hk,m . Then hm ∈ B1/k G
(eG ), so hm converges
to eG and hm (eG ) = x m .
For the converse, suppose that the Effros property fails for G acting on
X . Then for some open neighbourhood U of eG and some x ∈ X, U x :=
{u(x) : u ∈ U } is not an open neighbourhood of x. So for each n there is a
point x n ∈ B1/n (x)\U x. As x n converges to x, there are homeomorphisms hn
converging to the identity eG with hn (x) = x n . As U is an open neighbourhood
of eG and since hn converges to eG , there is N such that hn ∈ U for n > N . In
particular, for any n > N, hn (x) = x n ∈ U x, a contradiction, and we are done.

11.3.2 Proof of the Shift-Compactness Theorem, Theorem Sh

We view Theorem 11.1.1 as having ‘two tasks’: to find a ‘translator of the
sequence’ τ, and to locate it in a given Baire non-meagre subset of the group –
provided that subset satisfies a consistency condition (a necessary condition).
For clarity we break the tasks into two steps – the first delivering a weaker
version of Theorem 11.1.1 in Proposition 11.3.3 . The arguments are based on
the following lemma. We note a corollary, observed earlier by van Mill in the
case of metric topological groups (Mil2008, Prop. 3.4), which concerns a co-
meagre set, but we need its refinement to a localized version for a non-meagre
set.
Lemma 11.3.1 (Separation Lemma) Let G be a normed group, with separately
continuous and transitive Nikodym action on a non-meagre space X. Then, for
any point x and any F closed nowhere dense,
Wx,F := {α ∈ G : α(x) < F}
is dense open in G. In particular, G separates points from nowhere dense closed
sets.
Proof The set Wx,F is open, being of the form ϕ−1x (X\F) with ϕ x continuous
(by assumption). By the Nikodym property, for U any non-empty open set in
G, the set U x is non-meagre, and so U x\F is non-empty, as F is meagre. But
then for some u ∈ U we have u(x) < F. 
Corollary 11.3.2 If G is a normed group, Baire in the norm topology with
transitive and separately continuously Nikodym action on a non-meagre space
X space, and T is co-meagre in X – then, for countable D ⊆ X, the set
{g : g(D) ⊆ T } is a dense Gδ .
180 Effros’ Theorem

In particular, this holds if G is analytic and each point-evaluation map

ϕ x : g → g(x) is base-σ-discrete.

Proof Without loss of generality, the co-meagre set is of the form T =

S
U\ n∈ω Fn with each Fn closed and nowhere dense, and U open. Then, by
Lemma 11.3.1 (the Separation Lemma) and as G is Baire,
\
{g ∈ G : g(D) ⊆ T } = {g : g(D) ∩ Fn = ∅}
\n∈ω
= {g : g(d) < Fn }
d ∈D,n∈ω

is a dense Gδ . For the final assertion, see p. 209, point (2). 

Proposition 11.3.3 If T is a Baire non-meagre subset of a metric space X and

G a normed group, Baire in its norm topology, acting separately continuously
and transitively on X, with the Nikodym property – then, for every convergent
sequence x n with limit x 0 , there is τ ∈ G and an integer N with τx 0 ∈ T and

{τ(x n ) : n > N } ⊆ T .
S
Proof Write T := M ∪ (U\ n∈ω Fn ) with U open, M meagre and each Fn
closed and nowhere dense in X. Let u0 ∈ T ∩ U. By transitivity there is σ ∈ G
with σx 0 = u0 . Put un := σx n . Then un → u0 . Put
\
C := {α ∈ G : α(um ) < Fn },
m,n∈ω

a dense Gδ in G; then, by the Separation Lemma, 11.3.1, as G is Baire,

{α ∈ G : α(u0 ) ∈ U} ∩ C

is non-empty. For α in this set we have α(u0 ) ∈ U\ n∈ω Fn . Now α(un ) →

S
α(u0 ), by continuity of α, and U is open. So for some N we have for n > N that
α(un ) ∈ U. Since {α(um ) : m = 1, 2, . . .} ∈ X\ n∈ω Fn, we have, for n > N,
S
that α(un ) ∈ U\ n∈ω Fn ⊆ T .
S

Finally, put τ := ασ. It then follows that τ(x 0 ) = ασ(x 0 ) ∈ T and {τ(x n ) :
n > N} ⊆ T. 

Proof of Theorem 11.1.1, Theorem Sh We work in the right norm-topology

and use the notation of the preceding proof (of Proposition 11.3.3), so that U
here is the quasi-interior of T and σx 0 = u0 . As eG ∈ Aq and A is a non-meagre
Baire set, we may without loss of generality write A = Bε (eG )\ n G n, where
S
each G n is closed nowhere dense with eG < G n and Bε (eG ) is the quasi-interior
of A.
 11.4 From Effros to the Open Mapping Theorem 181

As Aq x 0 ∩ T q is non-empty, there is α0 ∈ Bε (eG ) with α0 x 0 ∈ U (but, we

want a better α so that αx 0 ∈ T and α ∈ A). Put β0 = α0 σ −1 ; then
β0 = α0 σ −1 ∈ Bε (eG )σ −1 ∩ {α : α(x 0 ) ∈ U}σ −1
= Bε (eG )σ −1 ∩ { β : β(σx 0 ) ∈ U}
= Bε (eG )σ −1 ∩ { β : β(u0 ) ∈ U },
i.e. the open set { β : β(u0 ) ∈ U} ∩ Bε (eG )σ −1 is non-empty. So
 [ 
C\ G n σ −1 ∩ { β : β(u0 ) ∈ U} ∩ Bε (eG )σ −1 , ∅,
n

since G is a Baire space and each G n σ −1 is closed and nowhere dense in G (as
the right shift g → gσ −1 is a homeomorphism).
So there is β with β(u0 ) ∈ U such that α := βσ ∈ Bε (eG )\ n G n = A.
S
That is, αx 0 = βu0 ∈ U; so β(un ) ∈ U for large n, for n > N say, as αx 0 =
lim αx n = lim βσx n = lim βun . But { β(um ) : m = 1, 2, . . .} ∈ X\ n Fn, as
S
β ∈ C; so β(un ) ∈ U\ n Fn ⊆ T for n > N.
S
Finally, α(x 0 ) = βσ(x 0 ) ∈ T and {α(x n ) : n > N } ⊆ T . 
We recall that Theorem Sh refers to Theorem 11.1.1.
Proof that Sh =⇒ E Assume G acts transitively on X and that X is non-
meagre. Let B := Bε (eG ) and suppose that for some x the set Bx is not a
neighbourhood of x. Then there is x n → x with x n < Bx for each n. Take
A := Bε/2 (eG ) and note first that A is a symmetric open set (A−1 = A,
since kgk = kg −1 k), and second that by the Nikodym property Ax contains
a non-meagre, Baire subset T. So by Theorem Sh, as Ax meets T q, there are
a ∈ A (which, being open, has the Baire property) and a co-finite Ma such that
ax m ∈ Ax for m ∈ Ma . For any such m, choose bm ∈ A with ax m = bm x.
Then x m = a−1 bm x ∈ A2 x ⊆ Bx, contradicting x m < Bx (note that a−1 ∈ A,
by symmetry). So Bx is a neighbourhood of x. 
Remark As earlier, in the special case that G is (metrizable and) analytic,
A is analytic, since open sets are Fσ and Souslin-F subsets of analytic sets
are analytic, cf. RogJ1980, Th. 2.5.3. So by Proposition 11.3.3 Ax is Baire
non-meagre, as ϕ x is base-σ-discrete, cf. page 209, point (2).

11.4 From Effros to the Open Mapping Theorem

Definition (Anc1987) Call the map ϕ x countably covered if there exist self-
homeomorphisms hnx of X for n ∈ N such that for any open neighbourhood U
in G the sets {hnx (ϕ x (U)) : n ∈ N} cover X .
182 Effros’ Theorem

Proposition 11.4.1 (cf. Anc1987) For the action ϕ : G × X → X with X

non-meagre, if each map ϕ x is countably covered and takes open sets to sets
with the Baire property, then the action has the Nikodym property.

Proof If ϕ x is countably covered, then there exist self-homeomorphisms hnx of

X for n ∈ N such that for any open neighbourhood U in G the sets {hnx (ϕ x (U)) :
n ∈ N} cover X . Then for X non-meagre, there is n ∈ N with hnx (ϕ x (U)) non-
meagre, so U x = ϕ x (U) is itself non-meagre, being a homeomorphic copy of
hnx (ϕ x (U)). As U x is assumed Baire, the action has the Nikodym property. 

For E separable, an immediate consequence of continuous maps taking open

sets to analytic sets (which are Baire sets) and of Proposition 11.4.1 is that ϕ L
is a Nikodym action.
For the general context, one needs demi-open continuous maps, which
preserve almost completeness (absolute Gδ sets modulo meagre sets – see
Mich1991 and its antecedent Nol1990), as it is not known which linear maps
are base-σ-discrete – a delicate matter to determine, since the former include
continuous linear surjections (by Lemma 11.4.2) and preserve almost analytic-
ity as opposed to analyticity.
For present purposes, however, the monotonicity property below suffices. We
omit the proof of the following observation (for which see the opening step in
Rud1973, 2.11 or Conw1990, Ch. 3, §12.3). The open balls below refer to the
underlying translation-invariant metric of a Fréchet space.

Lemma 11.4.2 For a continuous linear map L : X → Y from a Fréchet space

X to a normed space Y , for s < t < r,

int(cl L(Bs (0)) ⊆ L(Bt (0)) ⊆ L(Br (0)).

Hence, if L(Br (a)) is convex, either it is meagre or it differs from int L(Br (a))
by a meagre set.

Proposition 11.4.3 For L a continuous linear surjection from a Fréchet space

E to a non-meagre normed space F, the action ϕ L has the Nikodym property.

Proof We first show that as in Proposition 11.2.4 for L : E → F a continuous

linear surjection, {ϕ xL : x ∈ F} are countably covered. Indeed, fixing x ∈ F,

hnx (z) := n(z − x) (n ∈ N and z ∈ F)

is, on the one hand, a self-homeomorphism satisfying hnx (ϕ x (L(V ))) = L(nV ),
since n[(L(v) + x) − x] = nL(v) = L(nv); on the other hand, the family

{hnx (L(V ) + x) : n ≥ 1}
 11.4 From Effros to the Open Mapping Theorem 183

covers F, as {nV : n ∈ N} covers E where V is any open neighbourhood of

the origin in E (by the ‘absorbing’ property, cf. Conw1990, 4.1.13; Rud1973,
1.33). In particular, nL(B1 (0)) is non-meagre for some n, and so L(Bs (0)) is
non-meagre for any s. By Lemma 11.4.2, L(Bt (0)), for any t > s, contains the
non-meagre Baire set cl L(Bs (0)). 
Corollary 11.4.4 is now immediate; it is used in Ost2012, Th. 5.1 to prove
the ‘Semi-Completeness Theorem’, an Ellis-Type Theorem (Elli1953, Cor. 2)
(cf. Ost2013d) giving a one-sided continuity condition which implies that a
right-topological group generated by a right-invariant metric is a topological
group (cf. 13.3.3).
Corollary 11.4.4 If the continuous surjective homomorphism λ between
normed groups G and H, with G analytic and H a Baire space, is base-σ-
discrete, then λ is open; in particular, for λ bijective, λ −1 is continuous.
Corollary 11.4.5 For L : E → F a continuous surjective linear map between
Fréchet spaces, the point evaluations ϕbL for b ∈ F are open, and so L is an
open mapping.
Proof By surjectivity of L, the action is transitive, and by Proposition 11.4.3
the action ϕ L has the Nikodym property. So by Theorem E, 11.1.2, above the
point-evaluations maps ϕbL are open. Hence so also is L. 
 12

Continuity and Coincidence Theorems

12.1 Continuity Theorems: Introduction

We begin with an overview of the range of results of concern to us.

Theorem 12.1.1 (Discontinuity-Set Theorem; Kur1966, 33, p. 397; Kur1924)

(i) For f : X → Y Baire-measurable, with X, Y metric, the set of discontinuity

points is meagre; in particular,
(ii) for f : X → Y Borel-measurable of class 1, with X, Y metric and Y sepa-
rable, the set of discontinuity points is meagre.

The following theorem, from somewhat more recent literature, usefully over-
laps with the last result and will be proved in the next section.

Theorem 12.1.2 (Banach–Neeb Theorem; Banach (Ban1931; Ban1932),

extended by Neeb (Nee1997))

(i) A Borel-measurable f : X → Y with X, Y metric and Y separable is Baire-

measurable.
(ii) A Baire-measurable f : X → Y with X a Baire space and Y metric is
Baire-continuous.

Remarks 1. In fact Banach shows that a Baire-measurable function is Baire-

continuous on each perfect set – see Ban1932, vol. II, p. 206. For the distinction
between Baire and Baire-measurable, see the two paragraphs below and §12.2.
2. In (i) if X, Y are completely metrizable, topological groups and f is a ho-
momorphism, Neeb’s additional assumption in Nee1997 that Y is arcwise con-
nected becomes unnecessary, as Pestov (Pes1998) remarks in his MathSciNet
commentary to Nee1997.

184
 12.1 Continuity Theorems: Introduction 185

The following ‘portmanteau theorem’ summarizes what is in the literature.

Theorem 12.1.3 concerns the Baire functions (with preimages of open sets hav-
ing the Baire property). The later Theorem 12.2.2 concerns Baire-measurable
functions (obtained from the continuous functions by iterating pointwise lim-
its). See the comment ahead of Theorem 12.1.4.

Theorem 12.1.3 (Baire Continuity Theorem – Baire Version; cf. BinO2010g,

4, Th. 11.8) A Baire function f : X → Y is Baire-continuous in the following
cases:

(i) Baire’s condition (see, e.g., Hof1980, Th. 2.2.10, p. 346): Y is a second-
countable space;
(ii) Emeryk–Frankiewicz–Kulpa (EmeFK1979): X is Čech-complete and Y
has a base of cardinality not exceeding the continuum;
(iii) Pol’s condition (Pol1976): f is Borel, X is Borelian-K and Y is metrizable
and of non-measurable cardinality, see §12.5;
(iv) Hansell’s condition (Hanse1971): f is σ-discrete and Y is metric.

We will say that the pair (X, Y ) enables Baire continuity if the spaces X, Y
satisfy either of the two conditions (i) or (ii). One might include (iii), albeit the
Borel assumption is strong.
Building on EmeFK1979, Fremlin (Fre1987, Section 10) characterizes a
space X such that every Baire function f : X → Y is Baire-continuous for all
metric Y in the language of ‘measurable spaces with negligibles’; reference
there is made to disjoint families of negligible sets all of whose subfamilies
have a measurable union. (One may term this completely additive measurable by
analogy with the established phrase completely additive analytic, which we will
meet in §13.1.3, Remark 5.) For a discussion of discontinuous homomorphisms,
especially counterexamples on C(X ) with X compact (e.g. employing Stone–
Čech compactifications, X = βN\N), see Dal1978, 10, Section 9.
Remarks Hansell’s condition, requiring the function f to be σ-discrete, is
implied by f being analytic when X is absolutely analytic (i.e. Souslin-F (Y )
in any complete metric space Y into which it embeds). Frankiewicz and Kunen
(FraK1987) study the consistency relative to ZFC of the existence of a Baire
function failing to have Baire continuity. See also Fra1982.

Theorem 12.1.4 (Hartman–Mycielski Embedding Theorem; HartM1958) Ev-

ery topological group is a closed subgroup of a group G∗ which is arcwise
connected and locally arcwise connected.
186 Continuity and Coincidence Theorems

In particular, any separable (invariantly) metrizable group G is embeddable

as a subgroup of an arcwise connected separable (invariantly) metrizable
group.

Proof Let G∗ be the collection of range-finite functions f : [0, 1) → G which

are càdlàg (continue à droite, limite à gauche), piecewise constant (i.e. there
is a partition of [0, 1) into a finite number of contiguous half-open intervals
[u, v)) with pointwise product as the group operation, so that

gh(x) = g(x)h(x), g −1 (x) = g(x) −1, e(x) ≡ 1G .

Endow G∗ with a topology in which the neighbourhoods of f take the form

BV ( f , r) := {h : |x : h(x) f (x) −1 ∈ V | < r }

for V open in G with |. | Lebesgue measure on [0, 1). An arc joining two
functions f , g ∈ G∗ may be defined by
(
f (x), x ∈ [0, t),
ht (x) :=
g(x), x ∈ [t, 1),
so that G∗ is both arcwise- and by the same token locally arcwise-connected.
Identifying G with the constant functions f g (x) ≡ g embeds G into G∗ .
If G has metric d G , then d ∗ defined below metrizes G∗ :
Z 1
d ∗ ( f , g) := d G ( f (x), g(x)) dx ( f , g ∈ G∗ ),
0

and agrees with dG on the constant functions. If d G is right/left invariant, then

so is d . If G is separable, so is G∗ .
∗ 

Remark Arcwise connectedness occurs in Dixmier’s Theorem on the structure

theory of locally compact abelian groups and embedding of infinitely divisible
probability measures on groups; see, e.g., Hey1977, pp. 8, 220. For background
see Kur1968, Ch. 6.

Theorem 12.1.5 (Banach–Mehdi Theorem) (Cf. Ban1932, 1.3.4, p. 40, al-

beit for ‘Baire-measurable’ functions, Meh1964.) An additive Baire function
between complete normed vector spaces is continuous, and so linear, provided
the image space is separable.

Proof Suppose k is a Baire function, in the sense that inverse images under k
of open sets are sets with the Baire property.
By the Baire Continuity Theorem, 12.1.3, k is continuous on some co-meagre
set D. Suppose further that k is additive. If x n → x 0 and M ⊆ N, then, since
 12.1 Continuity Theorems: Introduction 187

D is shift-compact, for some t and infinitely many m ∈ M, say m ∈ M0 ⊆ M, it

follows that t + (x m − x 0 ) ∈ D. So, by continuity on D,

k (t) = lim 0 k (t + (x m − x 0 )) = lim 0 k (t) + k (x 0 ) − k (x m ),

m∈M m∈M

so that k (x m ) → k (x 0 ) for m ∈ M0. Thus k is continuous (by the ‘subsequence

theorem’, 1.3.1). From additivity one has k (r x) = r k (x), for r rational, and so
from continuity for all real r. That is, k is linear. 

The proof above is essentially due to Banach (Ban1932), although the concept
of shift-compactness had then not been recognized.
The Souslin criterion and the next theorem together have as an immediate
corollary the classical Souslin-graph Theorem (RogJ1980, §2.10). In this con-
nection recall (see the corollary of Hof1980, Th. 2.3.6, p. 355) that a normed
topological group which is Baire and analytic is Polish. Our proof, which is for
normed groups, is inspired by the topological vector space proof in RogJ1980,
§2.10, of the Souslin-graph theorem; their proof may be construed as having
two steps: one establishing their Souslin criterion, the other the Baire homo-
morphism theorem. They state without proof the topological group analogue.
For a non-separable analogue, see Chapter 13.

Theorem 12.1.6 (Baire Homomorphism Theorem; cf. RogJ1980, §2.10) Let

X and Y be normed groups with X non-meagre and analytic (e.g. topologically
complete) and Y separable. If f : X → Y is a Baire homomorphism, then f is
continuous. In particular, if f is a homomorphism with a Souslin-F (X × Y )
graph and Y is in addition a K -analytic space, then f is continuous.

Proof For f : X → Y the given homomorphism, it is enough to prove continu-

ity at eX , i.e. that for any ε > 0 there is δ > 0 such that Bδ (eX ) ⊆ f −1 [Bε (eX )].
So let ε > 0. We work with the right norm-topology.
Being K -analytic, Y is Lindelöf (cf. RogJ1980, Th. 2.7.1, p. 36) and metric,
so separable; so choose a countable dense set {yn } in f (X ) and select an ∈
f −1 (yn ). Put T := f −1 [Bε/4 (eY )]. Since f is a homomorphism, f (T an ) =
f (T ) f (an ) = Bε/4 (eY )yn . Note also that f (T −1 ) = f (T ) −1, so

TT −1 = f −1 [Bε/4 (eY )] f −1 [Bε/4 (eY ) −1 ] = f −1 [Bε/4 (eY ) 2 ]

⊆ f −1 [Bε/2 (eY )],

by the triangle inequality.

Now
[
f (X ) ⊆ Bε/4 (eY )yn,
n
188 Continuity and Coincidence Theorems

so
[
X = f −1 (Y ) = T an .
n

But X is non-meagre, so for some n the set T an is non-meagre, and so too is

T (as right shifts are homeomorphisms). By assumption f is Baire. Thus T is
Baire and non-meagre. By the Squared Pettis Theorem, 6.3.2, (TT −1 ) 2 contains
a ball Bδ (eX ). Thus we have

Bδ (eX ) ⊆ (TT −1 ) 2 ⊆ f −1 [Bε/4 (eY ) 4 ] ⊆ f −1 [Bε (eY )]. 

Theorem 12.1.7 (Souslin-Graph Theorem; Schw1966, cf. Pet1974; RogJ1980,

p. 50) Let X and Y be normed groups with Y a K -analytic space and X
non-meagre. If f : X → Y is a homomorphism with Souslin-F (X × Y ) graph,
then f is continuous.

Proof This follows from the preceding result and the Banach–Mehdi Theo-
rem, 12.1.5. 

12.2 Banach–Neeb Theorem

We begin with a simple result.

Lemma 12.2.1 For X a Baire space and meagre Y ⊆ X, X\Y is a Baire

subspace. Moreover, each subset that is meagre in X\Y is also meagre in X.

Proof Let F ⊆ X\Y be meagre in X. Choose nowhere dense subsets Fn ⊆

X\Y with F = n Fn . Let clX , resp. clX\Y , denote the closure of a set in
S
X, resp. X\Y . Then the fact that Fn is nowhere dense in X\Y implies that
clX\Y Fn = (X\Y ) ∩ clX Fn has empty interior. We conclude that, for each open
subset U ⊆ X with U ⊆ clX Fn , we have U ∩ (X\Y ) = ∅, i.e. U ⊆ Y . Since Y is
of first category in X, the assumption that X is Baire implies that U is empty.
This shows that clX Fn has empty interior, i.e. Fn is also nowhere dense in X.
So F is meagre in X. 

The following result should be compared with Baire’s Continuity Theorem,

12.1.3, where Y is second countable and f is a Baire function in the sense that
inverse images under f have the Baire property.

Theorem 12.2.2 (Baire Continuity Theorem – Baire-Measurable Version) If

X is a Baire space, Y a metric space and f : X → Y is Baire measurable, then
there exists a meagre M ⊆ X such that f | X\M is continuous.
 12.2 Banach–Neeb Theorem 189

Proof (Nee1997, cf. Ban1932.) Since the set of Baire-measurable functions

is the smallest class of all functions containing the continuous functions closed
under pointwise limits, it is enough to show that the class of functions satisfying
the condition of the theorem is closed under pointwise limits, since the class
trivially contains the continuous functions.
Suppose that the restriction of f n : X → Y to X\Mn is continuous, where
S
Mn is meagre in X . Then M := n∈N Mn is meagre in X, and all functions f n
are continuous on X1 := X\M. Suppose that f = limn→∞ f n holds pointwise
on X. Denote the metric on Y by dY . Since the functions f n are continuous on
X1 , the sets
An,ε := {x ∈ X1 : (∀m ≥ n)dY ( f n (x), f m (x)) ≤ ε}
are closed in X1 and f = limn→∞ f n implies that X1 = n∈N An,ε . We put
S
Bε := n∈N A0n,ε , where A0 denotes the interior of A in X1 . Then Bε is open
S
and we claim that Bε is dense in X1 . In fact, let U ⊆ X1 be open. By Lemma
3.1.5 X1 is a Baire space and so, since U is open, U is also a Baire space
(Chapter 2). Hence U = n∈N (U ∩ An,ε ) implies that at least one of the sets
S
U ∩ An,ε is somewhere dense in U. But these sets are closed subsets of U, so
there exists an n ∈ N for which U ∩ An,ε has interior points in U and therefore
also in X1 , i.e. A0n,ε ∩ U , ∅. Now Bε ∩ U , ∅ entails that Bε is dense in
X1 . This means that X1 \Bε is closed and has no interior points, i.e. X1 \Bε is
nowhere dense. This proves that
[ [
J := X1 \Bε = X1 \B1/n
ε>0 n∈N
is meagre in X1 .
Fix x ∈ X1 \J and let ε > 0. Then x ∈ Bε and we can find m ∈ N with
x ∈ A0m,ε . Then
dY ( f (y), f m (y)) = lim dY ( f n (y), f m (y)) ≤ ε
n→∞
for all y ∈ Am,ε implies
dY ( f (x), f (y)) ≤ 2ε + dY ( f n (x), f m (y)),
hence that f is continuous at x because x ∈ A0m,ε . This implies that f is
continuous on X1 \J = X\(M ∪ J) and so completes the proof as J is meagre
in X (by the Lemma 3.1.5).
This proves that the class of all functions satisfying the assumptions of the
theorem is closed under pointwise limits and contains the continuous functions,
hence also contains the Baire-measurable functions. 

Theorem 12.2.3 (Banach Continuous Homomorphism Theorem) If G is a

metrizable topological group which is a Baire space and H is a metrizable
190 Continuity and Coincidence Theorems

topological group, then every Baire-measurable homomorphism f : G → H is

continuous.

Proof (Cf. Ban1932, p. 23, Th. 4; cf. Ban1920) First, Theorem 12.2.2 shows
that there exists a meagre subset M ⊆ G such that f is continuous on G\M.
Let x n → 1G in G. Then the set x n . M ⊆ G is meagre for each n ∈ N. Hence
the same holds for
[
M∪ x n . M,
n∈N

which, since G is non-meagre, implies that this set must be different from
G. Let x be in the complement of this set. Then x < M and x −1 n x < M for
all n ∈ N. Hence the continuity of f on the complement of M implies that
f (x n ) −1 f (x) = f (x −1
n x) → f (x), which in turn implies that f (x n ) → 1G .
Since G was assumed metrizable, i.e. has a countable local base in 1G , we see
that f is continuous at 1G , and so, being a homomorphism, f is continuous. 

Remark In the proof above, it would suffice to prove that lim f (x m )m∈M0 = 1 H
for some infinite subset of any infinite M ⊆ N. As G\M is shift-compact, there
is t ∈ G\M and an infinite set M0 ⊆ M with t x m ∈ G\M 0 for m ∈ M0. Since
t x m → t, by continuity on G\M,

f (t) = lim 0 f (t x m ) = f (t) lim 0 f (x m ).

m∈M m∈M

So limm∈M0 f (x m ) = 1 H .
We proceed to weaken the assumption that f is Baire measurable.

Lemma 12.2.4 (Neeb’s Lemma; Nee1997) Let X, Y be metric spaces, with Y

arcwise connected and separable, and f : X → Y a Borel function. Then f is
Baire measurable.

Proof First we show that f is the limit of a sequence ( f n )n∈N of measurable

functions with at most countably many values. Let ε > 0 and (Yn )n∈N be a
basis for the topology consisting of sets whose diameter does not exceed ε. We
put Z1 := Y1 and Zn := Yn \(Y1 ∪ · · · ∪ Yn−1 ) for n > 1. Deleting any empty
Zn , we may assume that the Zn are all non-empty. The sets Zn are Borel,
and so the sets X n := f −1 (Zn ) are Borel subsets of X . Choosing z n ∈ Zn
we define a new function f ε : X → Z by taking f ε (x) := z n for x ∈ X n .
Then dY ( f (x), f ε (x)) ≤ ε for all x ∈ X n , and f ε (X ) is countable. Hence
f is a uniform limit of functions with at most countably many values. So,
without loss of generality, we may now assume that f (X ) is countable. We
write f (X ) = {yn : n ∈ N} and, using the arcwise connectedness of Y , find
a continuous function γ : R → Y with γ(n) = yn . Next define a real-valued
 12.2 Banach–Neeb Theorem 191

function h : X → R by taking h(x) := n whenever f (x) = yn and n is minimal

with respect to this property.
Then h is Borel measurable and γ ◦ h = f .
The set of all functions u : X → R for which γ◦u : X → Y is Baire measurable
contains the continuous functions and is closed under pointwise limits. This
implies that for each Baire function u : X → R the function γ ◦ u : X → Y is
Baire. Since h is a limit of finite linear combinations of characteristic functions,
to show that h is a Baire function, it suffices to see that indicator functions χ B
of Borel B ⊆ X are Baire. In fact, the set of all subsets B ⊆ X for which χ B is
Baire contains all open subsets, because for an open subset B, we have

χ B (x) = lim min{1, n · dist(x, X\B)},

n→∞

where dist(x, C) := inf{d(x, y) : y ∈ C}. Also χ X\B = 1 − χ B and

Yn
χ∩Bn = lim χ Bn .
n→∞ k=1

So, since the Baire-measurable functions X → R form a σ-algebra, {B ⊆

X : χ B Baire measurable} is a σ-algebra containing all the Borel sets. Thus
characteristic functions of Borel sets are Baire measurable. This proves that h
is Baire measurable, and hence that f is Baire measurable. 

Theorem 12.2.5 (Banach–Neeb Theorem; Ban1932; Nee1997) Every Borel-

measurable group homomorphism f : G → H from a completely metrizable
separable topological group into a separable metrizable group is continuous.

Proof Since G is completely metrizable it is a Baire space. By the Hartman–

Mycielski Theorem, 12.1.4, any separable metrizable group embeds as a topo-
logical subgroup into an arcwise connected separable metrizable group, so
without loss of generality we may assume that H is arcwise connected. Thus
by Lemma 12.2.4, f is Baire measurable. Hence Banach’s Continuous Homo-
morphism Theorem implies that f is continuous. 

For connections between Borel functions and Baire functions, see Fos1993.
We note in passing the next result (HewR1979, Th. 22.18).

Theorem 12.2.6 Let G be a locally compact group with λ a left Haar measure
and H a topological group which is σ-compact or separable, and suppose
f : G → H is a group homomorphism for which there exists a λ-measurable
subset A ⊆ G with 0 < λ( A) < ∞ such that for each open subset U ⊆ H the
set f −1 (U) ∩ A is λ-measurable. Then f is continuous.
192 Continuity and Coincidence Theorems

12.3 Levi Coincidence Theorem

This section is inspired by Sandro Levi’s article (Lev1983) titled ‘On Baire
Cosmic Spaces’, where he derives an Open Mapping Theorem (a result of the
Direct Baire Property given below in Theorem 12.3.1) and a useful corollary on
a comparison of topologies: if one refines the other, then they must coincide on
a subspace. The treatment in this section assumes the spaces to be separable.
To demonstrate their usefulness, recall an important result of Ellis asserting
that if a metric group is endowed with a topology under which the group is
locally compact and for which inversion is continuous while multiplication is
separately continuous, then in fact multiplication is jointly continuous. In brief,
a group with a semi-topological structure has a topological group structure.
We will deduce in §12.4 a result similar to Ellis’ for groups with a metric
which is right-invariant: a right-topological group generated by a right-invariant
metric (i.e. a normed group in the terminology of Chapter 6) is a topological
group. Unlike Ellis we do not assume that the group is abelian, nor that it is
locally compact and instead a form of analyticity suffices.
In Chapter 13, we develop non-separable generalizations of Levi’s results
and with them also non-separable versions of our Ellis-Type Theorem, 12.4.1,
with a ‘one-sided’ continuity condition implying that a right-topological group
generated by a right-invariant metric (i.e. a normed group) is a topological
group. Again, unlike Ellis, we do not assume that the group is abelian, nor that
it is locally compact; the non-separable context requires some preservation of
σ-discreteness as a side-condition (see the Remarks after Corollary 13.3.3).
Given that the application in mind is metrizable, references to non-separable
descriptive theory remain, for transparency, almost exclusively in the metric
realm, though we do comment on the regular Hausdorff context in the Remark
following Theorem 13.3.8.
Levi’s work draws together two notions: BP – the Baire set property (i.e.
that a set is open modulo a meagre set, so ‘almost open’), and BS – the Baire
space property (i.e. that Baire’s Theorem holds in the space). Below we keep
the distinction clear by using the terms ‘Baire property’ and ‘Baire space’. The
connection between BP and BS is not altogether surprising, and the two are
‘almost’ the same in a precise sense, at least in the context of normed groups (cf.
Ost2013b, where this closeness is fully exploited). For, in an almost-complete
space, the terms ‘Baire set’, ‘set with the Baire property’ and ‘Baire space’ are
almost-synonyms in the sense that, for B non-meagre, B has the Baire property
if and only if B is a Baire space if and only if B is almost-complete (Ost2013b,
Th. 7.4).
We recall from the Chapter 3 (where we introduce analyticity) that a sub-
space S of a metric space X has a Souslin-F (X ) representation if there is a
 12.3 Levi Coincidence Theorem 193

‘determining’ system hF (i | n)i := hF (i | n) : i ∈ NN i of sets in F (X ) (the

closed sets) with
[ \
S= F (i | n), where I = NN
i ∈I n∈N

and i | n denotes (i 1, . . . , i n ). We will say that a topological space is classi-

cally analytic if it is the continuous image of a Polish space (Levi terms these
‘Souslin’) and not necessarily metrizable, in distinction to an (absolutely) an-
alytic space, i.e. one that here is metrizable and is embeddable as a Souslin-F
set in its own metric completion; in particular, in a complete metric space,
Gδ -subsets (being Fσδ ) are analytic. We call a Hausdorff space almost analytic
if it is analytic modulo a meagre set. Similarly, a space X 0 is absolutely Gδ , or
an absolute-Gδ , if X 0 is a Gδ in all spaces X containing X 0 as a subspace. (This
is equivalent to complete metrizability in the narrowed realm of metrizable
spaces (Eng1989, Th. 4.3.24), and to topological/Čech-completeness in the
narrowed realm of completely regular spaces, Eng1989, §3.9.) So a metrizable
absolute-Gδ is analytic; we use this fact in Lemma 12.4.3.
Levi’s results follow from the following routine observation.

Theorem 12.3.1 (Direct Baire Property; Lev1983) Let X be a classically

analytic space and Y a Hausdorff space. Every continuous map f : X → Y has
the direct Baire property: the image of any open set in X has the Baire property
in Y .

Proof Suffice to note that an open set in a metric space X is Fσ and so, being
Souslin-F in an analytic space, is itself analytic. The result is then immediate
from Nikodym’s Corollary, 3.1.3. 

The nub of the theorem is that, with X as above, continuity preserves various
analyticity properties such as that open, and likewise closed, sets are taken to
analytic sets, in brief: a continuous map is open-analytic and closed-analytic
in the terminology of Hanse1974, and so preserves the Baire property. (See
Remark 2 in §13.1.3 for a reprise of this theme.) Levi deduces the following
characterization of Baire spaces in the category of classically analytic spaces.

Theorem 12.3.2 (Levi Open Mapping Theorem; Lev1983) Let Y be a regular

classically analytic space. Then Y is a Baire space if and only if Y = f (X )
for some continuous map f on some complete separable metric space X with
the property that, for some dense metrizable absolute-Gδ subspace Y 0 ⊆ Y and
X 0 = f −1 (Y 0 ), the restriction map f | X 0 : X 0 → Y 0 is open.
194 Continuity and Coincidence Theorems

The notation in the proof below may seem inefficient; however, our purpose
is to make its later non-separable variant more intelligible, as there all the sets
appearing here need to be partitioned into σ-discrete parts. This is necessary
because use of Urysohn’s metrization theorem here needs to be replaced in
the non-separable case by Bing’s Characterization Theorem. Recall that Bing’s
Theorem (Eng1989, Th. 4.4.8) asserts that a regular space is metrizable if and
only if it has a σ-discrete base. The latter property requires the apparatus of
σ-discrete families as mentioned above. See the Remark immediately after the
proof.
Proof One direction is clear from the density statement. For the converse,
let A = { An : n = 1, 2, . . .} be an open base for X. Then E := f ( A) for
A ∈ A is analytic (see also Remark 5 in §13.1.3), so has the Baire property
(by Nikodym’s Corollary, 3.1.3, for analytic sets). Put E = { f ( A) : A ∈ A} =
{ f ( An ) : n = 1, 2, . . .}. For each E ∈ E pick an open set UE and meagre sets
NE and ME such that
E = (UE \NE ) ∪ ME ,
with ME disjoint from ME and with NE ⊆ UE . Put
[ [
M := {ME : E ∈ E} and N := {NE : E ∈ E},
both being meagre, as E is countable. Now put Y 0 := Y \(M ∪ N ) and
[
W= {UE : E ∈ E},
which is open in Y . Then, for A ∈ A with E = f ( A),
f ( A) ∩ Y 0 = E ∩ Y 0 = UE ∩ Y 0,
so that f ( A) is open relative to Y 0 .
For G ⊆ X open, since A is a base, we may write
[
G := { An : An ⊆ G}.
n
Then
[
f (G) := {En : En = f ( An ) & An ⊆ G}.
n

So, for X 0 := f −1 (Y 0 ),
[
f (G ∩ X 0 ) = Y 0 ∩ {U f ( A) : A ⊆ G & A ∈ A},
which is open in Y 0 .
Now Y 0 is second countable: the family {Y 0 ∩ UE : E ∈ E} is a countable
S
base for Y 0 . As Y is regular (Eng1989, Th. 5.1.5), the subspace Y 0 is regular
(Eng1989, Th. 2.1.6). Being regular and second countable, Y 0 is metrizable by
 12.3 Levi Coincidence Theorem 195

Urysohn’s Theorem. Finally, by replacing the meagre sets M, N by larger sets

that are unions of closed nowhere dense sets, we obtain in place of Y 0 a smaller,
metrizable, dense Gδ -subspace. 

Remark To generalize one must take A = n A n with each A n = { Atn : t ∈

S
Tn } discrete (which in the separable case reduces to the singletons A n = { An });
then E n = { f ( A) : A ∈ A n } replaces E and will have a σ-discrete base Bn
comprising sets with the Baire property. In place of the sets E above one works
with sets B in B = n Bn .
S

The result may be regarded as implying an ‘inner regularity’ property (com-

pare the capacitability property) of a classically analytic space Y : if Y is a
Baire space, then Y contains a dense absolute-Gδ subspace, so a Baire space.
Compare the result due to Roy Davies (Dav1952) that an analytic set of infinite
Hausdorff measure contains closed sets of any desired finite measure. The exis-
tence of a dense completely metrizable subspace – making Y almost complete
in the sense of Frolík (Fro1960, though the term is due to Michael, Mich1991)
– is a result that implicitly goes back to Kuratowski (Kur1966, IV.2, p. 88,
because a classically analytic set has the Baire property in the restricted sense
– Cor. 1, p. 482). Generalizations of the latter result, including the existence of
a restriction map that is a homeomorphism between a Gδ -subset and a dense
set, are given by Michael (Mich1986); but there the continuous map f requires
stronger additional properties such as openness on X (unless X is separable),
which Levi’s result delivers.
Theorem 12.3.2 has a natural extension characterizing a Baire space (in the
same way) when it is almost analytic. Indeed, with Y 0 as above, the space Y
is almost complete and so almost analytic. On the other hand, if Y is a Baire
space and almost analytic, then by supressing a meagre Fσ and passing to
an absolutely Gδ -subspace, we may assume that Y is a Baire space which
is analytic, so has the open mapping representation of the theorem, and in
particular is almost complete (for more background, see Ost2013b: cf. Cor. 1.8).
Since an analytic space is a continuous image, Theorem 12.3.2 may be
viewed as an ‘almost preservation’ result for complete metrizability under
continuity in the spirit of the classical theorem of Hausdorff (resp. Vainstein)
on the preservation of complete metrizability by open (resp. closed) continuous
mappings – see the Remarks in §13.3.3 for the most recent improvements
and the literature of preservation. We note that Michael (Mich1991, Prop.
6.5) shows that almost completeness is preserved by demi-open maps (i.e.
continuous maps under which inverse images of dense open sets are dense).
Theorem 12.3.2 has an interesting corollary on the comparison of refinement
topologies. For a discussion of refinements, see Ost2013b, §7.1 (for examples
196 Continuity and Coincidence Theorems

of completely metrizable and of analytic refinements, see Kec1995, Th. 13.6,

Th. 25.18, Th. 25.19).
Theorem 12.3.3 (Levi Coincidence Theorem; Lev1983) For T , T 0 two
topologies on a set Y with (Y, T 0 ) classically analytic (e.g. Polish) and T 0
refining T (i.e. T ⊆ T 0), if (Y, T 0 ) is a regular Baire space, then there is a
T -dense G(T ) δ -set on which T and T 0 coincide.
Proof As T 0 refines T , to prove the theorem one may pass to any dense
Gδ subset of (Y, T 0 ). We claim that we may pass to such a subset that is
also Polish. Indeed, we may pick a Polish X and g : X → (Y, T 0 ) continuous
with g(X ) = Y . Now the embedding j : (Y, T 0 ) → (Y, T ) with j (y) = y is
continuous. Taking for f the composition jg, which is continuous and gives
f (X ) = Y, apply Levi’s Open Mapping Theorem to obtain a Gδ set X 0 ⊆ X
such that f | X 0 is open and f (X 0 ) is a dense Gδ in (Y, T 0 ). By Hausdorff’s
Theorem that an open continuous image of a Polish space is open (cf. for
example Anc1987, or, for a more recent account, HoliP2010), f (X 0 ) is Polish.
So we now assume that (Y, T 0 ) is Polish and again apply Levi’s Open
Mapping Theorem, 12.3.2, this time taking X to be (Y, T 0 ), Y to be (Y, T ) and
f to be j. Then, for some Gδ subset Z of X, the map j | Z is open. Writing TZ
and TZ0 for the subspace topologies induced on Z by T and T 0 respectively,
j takes the sets of TZ0 to the sets TZ . But TZ0 refines TZ , so the two topologies
coincide (and are again Polish by Hausdorff’s Theorem). 

12.4 Semi-Polish Theorem

Below we are concerned with the join of the two metric topologies (their
coarsest joint refinement), which is generated by the symmetrized metric
d SX := max{d RX , d LX },
where d RX and d LX are respectively a right- and a left-invariant metric on X . See
Itz1972 for connections with uniform spaces.
When X is Polish/analytic under the topology of d SX we shall say that X is
semi-Polish/semi-analytic under the topology of d RX . The term was suggested
by Anatole Beck.
Notation We use the subscripts R, L, S as in x n →R x etc. to indicate
convergence in the corresponding metrics d RX , d LX , d SX .
Notice that if d SX (x, y) < r, then d RX (x, y) < r, and so BS (x; r) ⊆ BR (x; r).
Thus any R-open set is S-open. In other terms, if x n →S x, then x n →R x, and
 12.4 Semi-Polish Theorem 197

so R-closed sets are S-closed (for, if an R-closed set F were not S-closed, then
there would be a sequence x n in F with x n →S x and x < F, contradicting
x n →R x). Passing to complements shows that an R-open set is S-open. In
summary: the topology generated by d SX is finer (has more sets) than that
generated by d RX .
Its general significance comes from the theorem that, for (T, dT ) any com-
plete metric space, the group of bounded self-homeomorphisms of T is com-
plete under the symmetrization of the supremum metric (cf. §6.1; for details
see Ost2013c; Dug1966, Th. XIV.2.6, p. 296).

Theorem 12.4.1 For a group X equipped with a right-invariant metric d RX :

if the space X is non-meagre and semi-Polish (more generally, semi-analytic)
under the topology of d RX , then it is a Polish topological group (i.e. under the
d RX topology, X is completely metrizable and a topological group).

We recall that every metrizable topological group has an equivalent right-

invariant metric, by the Birkhoff–Kakutani Theorem, 6.1.4, and a Polish group
is non-meagre (by Baire’s Theorem), so this theorem covers all Polish groups.
The theorem also generalizes a result due to Hoffman-Jørgensen that a Baire
analytic topological group is Polish, because an analytic group is separable
and metrizable (see Hof1980, Th. 2.3.6, p. 355). The theorem addresses the
question: when does one-sided continuity of multiplication imply its joint con-
tinuity and further its admissibility, i.e. endowment of a topological group
structure? As noted above, that question was considered in the abelian context
by Ellis in Elli1957 (see in particular his Th. 2, where the topology is locally
compact – cf. §13.1), but otherwise the existing literature, which goes back to
Montgomery (Mon1936) and also Ellis (Elli1953) via Namioka (Nam1974),
considers some form of weak bilateral continuity, usually separate continuity,
supported by additional topological features, including some form of complete-
ness. See Bouziad’s two papers (Bouz1993; Bouz1996) for the state-of-the-art
results, deducing automatic joint continuity from separate continuity (and for
a review of the historic literature), and the later paper of Solecki and Sri-
vastava (SolS1997), where separate continuity is weakened; cf. CaoDP2010;
CaoM2004; Chr1981. For the broader context of automatic continuity, see
TopH1980 (e.g. p. 338), and for the interaction of topology and algebra, Dales
(Dal2000).
Namioka’s Theorem, cited above, giving conditions under which a separately
continuous function is jointly continuous on a dense Gδ , is reminiscent of the
coincidence themes of Chapter 12. For some developments of his theme, see
HansT1992.
198 Continuity and Coincidence Theorems

By contrast to these bilateral conditions, Theorem 12.4.1 assumes only a

particular form of one-sided continuity, supported by additional topological
properties. An advantage of this approach is to replace the use of local com-
pactness (or even subcompactness, for which see Bouz1996) by the much
weaker notion of shift-compactness in groups in the form of Theorem 6.2.3,
the Analytic Shift Theorem.
We will need two lemmas. The first is merely a sharpening appropriate for
groups equipped with a right-invariant metric of an old result of Levi. For
completeness we give the (direct) proof; our main work begins in earnest in
Lemma 12.4.3.
Lemma 12.4.2 (cf. Lev1983, Th. 2 and Cor. 4) For a group X equipped with a
right-invariant metric d RX , if (X, d SX ) is Polish, i.e. separable and topologically
complete, or more generally analytic, and (X, d RX ) non-meagre, then there is a
subset Y of X which is a dense absolute-Gδ in (X, d RX ), and on which the d SX
and d RX topologies agree.
Proof We may apply Levi’s Coincidence Theorem, taking T 0 to be the topol-
ogy generated by the metric d SX and T by the metric d RX , to obtain a dense
Gδ (X, d RX ), and on which the d SX and d RX topologies agree. Working in Y, we
have yn →R y if and only if yn →S y if and only if yn →L y. 
Observe that above, since X\Y is meagre under d RX , the space (X, d RX ) is
almost complete (see Chapter 13). We use almost completeness to extract
much more.
Lemma 12.4.3 If, in the setting of Lemma 12.4.2, the three topologies gen-
erated by d RX , d LX , d SX agree on a dense absolutely Gδ set Y of (X, d RX ), then for
any τ ∈ Y the conjugacy γτ (x) := τxτ −1 is continuous.

Proof We work in (X, d RX ). Let τ ∈ Y . We first establish the continuity on X

at e of the conjugacy x 7→ τ −1 xτ (by shifting into Y ). Let z n → e be any null
sequence in X. Fix ε > 0; then T := Y ∩ BεL (τ) is analytic, since T is dYR -open
in Y, and is non-meagre, as X is Baire. By the Analytic Shift Theorem, 6.2.3,
there is t ∈ T and t n in T with t n converging to t (in d RX , so also in d LX ) and
m z m t m : m ∈ Mt } ⊆ T . Since the three topologies
an infinite Mt such that {tt −1
agree on Y and as the subsequence tt −1 m z m t m lies in Y and converges to t in
Y under d R (see Remark on p. 97), it also converges to t under dYL . Using the
Y

identity
m z m t m, t) = d L (t m z m t m, e) = d L (z m t m, t m ),
d LX (tt −1 X −1 X

we note that
kt −1 z m t k = d LX (t, z m t) ≤ d LX (t, t m ) + d LX (t m, z m t m ) + d LX (z m t m, z m t)
≤ d LX (t, t m ) + d LX (tt −1
m z m t m, t) + d L (t m, t) → 0,
X
 12.5 * Pol’s Continuity Theorem 199

as m → ∞ through Mt . So d LX (t, z m t) < ε for large enough m ∈ Mt . Then, as

d LX (τ, t) < ε, for any such m one has

kτ −1 z m τk = d LX (z m τ, τ) ≤ d LX (z m τ, z m t) + d LX (z m t, t) + d LX (t, τ)
≤ d LX (τ, t) + d LX (t, z m t) + d LX (t, τ) ≤ 3ε.

Thus for any ε > 0 and any k there is m = m(k, ε) > k with kτ −1 z m τk ≤ 3ε.
Inductively, taking successively ε = 1/n and k (n) := m(k (n − 1), ε), one
has kτ −1 zk (n) τk → 0. By the weak continuity criterion, Lemma 6.1.10 (cf.
BinO2010g, Lemma 3.5, p. 37), γ(x) := τ −1 xτ is continuous. Since (X, d RX )
is analytic and metric, each open set U is analytic, so γτ−1 (U) = γ(U) is
analytic, so has the Baire property, by Nikodym’s Theorem, 3.1.3. So γτ (x) =
τxτ −1 = γ −1 (x) is a Baire homomorphism, and so is continuous – by the Baire
Homomorphism Theorem, 12.1.6. 

Proof of Theorem 12.4.1 Under d RX , the set ZΓ := {x : γx is continuous} is a

closed subsemigroup of X (BinO2010g, Prop. 3.43). By Lemmas 12.4.2 and
12.4.3, X = clR Y ⊆ ZΓ, i.e. γx is continuous for all x, and so (X, d RX ) is
a topological group. So x n →R x if and only if x −1 n →R x
−1 if and only if

x n →L x if and only if x n →S x. So, being homeomorphic to (X, d SX ), the

space (X, d RX ) has the structure of a Polish topological group. 

12.5 * Pol’s Continuity Theorem

This section is devoted to a sketch proof of Theorem 2.1.6 (recalled below),
namely Pol’s version of Baire’s Continuity Theorem 2.1.4. Pol’s version is
concerned with a stronger Baire property. Recall from §2.1 that A ⊆ X has
the restricted Baire property in X if A ∩ Z is Baire in any subspace Z ⊆ X:
below we will say A is hereditarily Baire. Likewise, a map is restricted Baire
(or hereditarily Baire) if preimages of open sets are restricted Baire. Below
compact is taken to imply Hausdorff. We start with a combinatorial result;
for measurable cardinals, see §16.3 for a discussion and literature. In brief:
these are cardinals κ which support {0, 1}-valued measure on ℘(κ) that vanish
on singletons and are not just σ-additive but also κ-additive, i.e. the measure
is additive over any disjoint family of cardinality less than κ. By contrast, a
cardinal κ is said to be non-measurable if the only σ-additive {0, 1}-valued
measure on ℘(κ) vanishing on singletons is trivial, i.e. identically zero (see,
e.g., KurM1968, Ch. IX.3 and Bog2007, 1.12(x), p. 79). This is not quite
the negation of the notion of measurable cardinal. However, the least cardinal
κ that is not non-measurable supports a non-trivial 2-valued measure that is
200 Continuity and Coincidence Theorems

κ -additive (see, e.g., Jec2002, Lemma 10.2; Dra1974, Ch. 6 Th. 1.4), so that κ
is a measurable cardinal. 1
Proposition 12.5.1 (Pol1976, Th. 1.) Let E be a disjoint family of meagre
subsets of a compact space X with non-meagre union. If the family E has
non-measurable cardinality, then there is a subfamily E 0 whose union fails to
be hereditarily Baire.
We will give a sketch proof below after first proving a lemma. For the
purposes of the lemma, observe that if S is a non-meagre Baire set, then
modulo a meagre set it has the form of a Gδ -set H with
∅ , H ⊆ int(cl(H)).
Indeed, there are meagre sets M, M 0 with
S = (G\M) ∪ M 0,
and without loss of generality M may be taken to be a countable union of closed
nowhere dense sets (increasing M 0 by a meagre set, if needed). Here G is non-
empty (otherwise S is meagre). Now take H = G\M, then G = int(cl(H)), as
claimed. The following lemma enables an inductive binary-tree construction in
X of a Cantor set by selecting descending Gδ -subsets with disjoint closures (so
forming a binary tree under inclusion).
Lemma 12.5.2 (Pol1976, Lemma 1) Suppose a Gδ -set H ⊆ X with ∅ ,
H ⊆ int(cl(H)) is covered by an open family U and also by a disjoint family
A, of non-measurable cardinality, comprising meagre sets. Then there is a
decomposition A = A0 ∪ A1 and two open sets Vi , refining U and with
S
disjoint closures, such that each of Vi ∩ H ∩ Ai is non-meagre. In particular,
if both unions are Baire, each covers a non-meagre Gδ -set Hi with ∅ , Hi ⊆
int(cl(Hi )).
S
Proof Key here is the family I := {B ⊆ A : H ∩ B is meagre}, which
is a σ-ideal in ℘(A), containing all the singleton subsets of A, and is proper,
since H is non-meagre. So, as the cardinality of A is non-measurable, there
is a decomposition A = A0 ∪ A1 with each Ai < I. (Otherwise, for each
B ⊆ A, the decomposition B ∪ A\B yields that either B ∈ I or A\B ∈ I.
Then F = { A ⊆ X : A < I} is an ultrafilter see, e.g., Jec2002, Ch. 10, p. 126.
So the function µ defined on A by µ( A) = 1 if A ∈ F and µ( A) = 0 if A
∈ I, is a 2-valued measure on A, as I is a σ-ideal.) By the Banach Category
1 A σ-additive prime ideal on a non-measurable cardinal κ is κ-additive; see KurM1969, Ch.
IX.3, Th.3, Bog2007, 1.12(x), p. 79. Usage of the term measurable cardinal in the early
literature followed Ulam’s original approach in Ula1930.
 12.5 * Pol’s Continuity Theorem 201

Theorem (§5.2.2), we may choose for i = 0, 1 a (heavy) point x i ∈ H ∩ Ai ,

S
S
i.e. one such that H ∩ Ai is non-meagre in any neighbourhood of x i . Finally,
by compactness, we may choose closed neighbourhoods Vi separating the two
points. 
Sketch Proof of Proposition 12.5.1 Suppose the Proposition fails for some fam-
ily E. Then E covers a non-meagre Baire set; moreover, all its subfamilies have
Baire union. Thus Lemma 12.5.2 may be used in an induction on n to con-
struct a binary-tree system of non-meagre G δ -sets H(i1,...,in ) = m G m
T
(i1,...,in )
for (i 1, . . . , i n ) ∈ {0, 1} n and corresponding subfamilies E (i1,...,in ) covering
H(i1,...,in ) . Take for i ∈ {0, 1}N
\
E (i) = E (i1,...,in ) .
n

Also, for i ∈ {0, 1}N, put

\ [
C(i) := clH (i|n) ⊆ E (i) with K := {C(i) : i ∈ {0, 1}N }.
n

The set K is compact in X and the map f : K → {0, 1}N defined by f −1 {i} = C(i)
is continuous, by construction. For A ⊆ {0, 1}N,
[ [
f −1 ( A) = K ∩ E A, with E A := E (i),
i∈A
is hereditarily Baire, since each subfamily E A has Baire union.
So arbitrary subsets of K are Baire. From here we derive a contradiction,
namely that all subsets of C are Baire.
Indeed, as K is compact, f is closed (maps closed sets to closed sets) Kel1955.
Choose a minimal closed subset M ⊆ K with f (M) = C. Then the restriction
map g := f |M is irreducible and closed, hence maps Baire sets to Baire sets
(cf. Grue1998; Sem1971, Ex. 25.2.3, p. 447). By hypothesis, f −1 ( A) ∩ M is
Baire for any A ⊆ C. Hence A = g( f −1 ( A) ∩ M) is Baire. 

Proposition 12.5.1 is used below to deduce that a certain family has meagre
union.
Theorem 12.5.3 (Pol’s Theorem; Pol1976, Th. 2.) Let f : X → Y be a
mapping from a compact space X to a metrizable space Y of non-measurable
cardinality. Then f is hereditarily Baire if and only if for each subspace Z ⊆ X
the restriction g := f | Z\M is continuous for some M ⊆ Z meagre in Z.
Proof The condition is clearly sufficent. For its necessity, it is enough to
consider only closed subspaces Z ⊆ X. So without loss of generality we may
take Z = X . Now suppose f : X → Y has the Baire property hereditarily and
Y is metrizable of non-measurable cardinality. In Y select a base B = n Bn
S
202 Continuity and Coincidence Theorems

with each Bn discrete and let Un = { f −1 (V ) : V ∈ Bn }. For U ∈ Un, as f has

the Baire property, in X there are open sets GU and disjoint meagre sets MU ,
MU0 with MU ⊆ GU , MU0 ⊆ X\GU such that
U = (GU \MU ) ∪ MU0 .
For distinct U, W ∈ Un , since GU \MU and GW \MW are disjoint, so also are
GU and GW (a non-empty open common part would be of second category, so
not covered by MU ∪ MW ). Put
[ [
Mn = {MU : U ∈ Un }, Mn0 = {MU0 : U ∈ Un }\Mn .
By the Banach Category Theorem (§5.2.2), Mn is meagre since
Mn ∩ GU = MU
for U ∈ Un . For U ⊆ Un ,
[ [ [ [ [
MU0 \Mn = MU0 \ GU = U\ GU .
U ∈U U ∈U U ∈U U ∈U U ∈U
So, given the assumptions on f , this set is hereditarily Baire. By Proposition
12.5.1, the family of meagre sets
E n := {MU0 \Mn ; U ∈ Un }
has meagre union, which is equal to Mn0 . Then M := n Mn ∪ Mn0 is meagre.
S
Consider the restriction g := f |(X\M). To see that g is continuous, for any
n and any V ∈ Bn, take U = f −1 (V ) ∈ Un and note that
g −1 (V ) = f −1 (V ) ∩ (X\M) = U\M = GU \M
is open in X\M. 
 13

* Non-separable Variants

13.1 Non-separable Analyticity and the Baire Property

This section enables some standard (separable) analyticity and category argu-
ments to be lifted from Chapter 12 to the non-separable context. The requisite
concepts and their definitions rely on various forms of countability typified
by σ-discrete families (abbreviated occasionally to σ-d): these are delayed till
after the statements of theorems and given in §13.1.2.

13.1.1 Classical Souslin Representation

We recall that a subspace S of a metric space X has a Souslin-H representation
if there is a determining system hH (i | n)i := hH (i | n) : i ∈ NN i of sets in H
with (RogJD1980; Hanse1973a)
[ \
S= H (i | n), (I := NN, i | n := (i 1, . . . , i n )).
i ∈I n∈N

A topological space S is an (absolutely) analytic space if it is embeddable as

a Souslin-F set in its own metric completion S ∗ (with F the closed sets);
in particular, in a complete metric space Gδ -subsets (being Fσδ ) are an-
alytic. For more recent generalizations, see, e.g., NamP1969. According to
Nikodym’s Theorem, 3.1.3, if H above comprises Baire sets, then also S is
Baire (RogJ1980, §2.9 or Kec1995, Th. 29.14), i.e. the Baire property is pre-
served by the Souslin operation.
Central to the needs of a non-separable context are three results: Hansell’s
Characterization Theorem, 13.1.1, yielding representation of analytic sets in the
S
form j ∈κ N H ( j) with H upper semi-continuous and compact-valued (defined
on the completely metrizable, countable product of discrete spaces of cardi-
nality κ), Nikodym’s Theorem, 3.1.3, recalled and proved below as 13.1.2,
implying their Baire property, and the conclusion that analytic base-σ-discrete

203
204 * Non-separable Variants

maps (A-σ-d maps, for short, to be defined below) are the only ones that matter
(cf. 13.1.8).
We content ourselves mostly with a metric context, though a wider one
is feasible (consult Hanse1992). A Hausdorff space S is K -analytic if S =
i ∈I K (i) for some upper semi-continuous map K from I = N to K (S),
S N

the compact subsets of S. In a separable metric space, an absolutely analytic

subset is K -analytic (RogJ1980, Cor. 2.4.3 plus Th. 2.5.3). In a non-separable
complete metric space X, it is not possible to represent a Souslin-F (X ) subset
S of X as a K -analytic set relative to I = NN . Various generalizations of
countability now enter the picture, as we now recall, referring also to two
survey papers: Sto1980 and Hanse1992.

13.1.2 Extended Souslin Representation: Consequences

Denoting by wt(X ) the weight of the space X (i.e. the smallest cardinality of
a base for the topology), and replacing I = NN by J = κ N for κ = wt(X ),
with basic open sets J ( j | n) := { j 0 ∈ J : j 0 | n = j | n} (as in Chapter 3),
consider sets S represented by the following notion of a Souslin representation,
broader than that above, namely by the extended κ-Souslin operation (briefly:
the extended Souslin operation):
[ \
S= H ( j), where H ( j) := H ( j | n),
j ∈J n∈N

applied to a determining system hH ( j | n)i := hH ( j | n) : j ∈ κ N i of sets from

a family H subject to the requirement that:

(i) {H ( j | n) : j | n ∈ κ n } is σ-discrete for each n.

We will usually also require that the determining system is shrinking, meaning:

(ii) diamX H ( j min n) < 2−n, so that H ( j) is empty or single-valued, and so

compact.

For H = F the corresponding extended Souslin-F sets reduce to the κ-

Souslin sets of Hanse1992. (This slightly refines Hansell’s terminology, and
abandons Stone’s term ‘κ-restricted Souslin’ of Sto1980.)
With X above complete (e.g. X = S ∗ , the completion of S) and for H =
F (X ), the mapping H : J → K (X ) evidently yields a natural upper semi-
continuous representation of S. We refer to it below, in relation to the Analytic
Cantor Theorem, 13.1.5, cf. 3.1.6, in the separable context of Chapter 3, and
also in Proposition 13.1.8. There the fact that C := { j : H ( j) , ∅} is closed
in κ N yields a natural representation of S as the image of C under a map h
 13.1 Non-separable Analyticity and the Baire Property 205

defined by H ( j) = {h( j)}. The map h is continuous and, as will be defined

below, index-σ-discrete with countable fibres (preimages of single points), by
(i) above, as noted in opening remarks of §13.1.2.

Theorem 13.1.1 (Hansell’s Characterization Theorem) In a metric space

X, the Souslin-F (X ) subsets of X are precisely the sets S represented by a
shrinking determining system of closed sets through the extended κ-Souslin
representation above with κ = wt(X ).

For other equivalent representations, including a weakening of σ-discreteness

in X above to σ-d relative to its union, as well as to σ-d decompositions, see
Hanse1972; Hanse1973b; Hanse1973a. Thus, working relative to J, the corre-
sponding extended Souslin sets exhibit properties similar to the K -analytic sets
relative to I. In particular of interest here is Nikodym’s Corollary, Theorem
3.1.3, which we recall and prove.

Theorem 13.1.2 (Nikodym’s Corollary) In a metric space S, analytic sets

have the Baire property.

Proof Since S ∩ F (S ∗ ) = F (S), the theorem follows immediately from the

definition of analytic sets as Souslin-F (S ∗ ) and from Nikodym’s theorem,
3.1.3, asserting that the Baire property is preserved by the usual Souslin oper-
ation, with the consequence that Souslin-F sets have the Baire property (since
a closed set differs from its interior by a nowhere dense set). 

Using Hansell’s Characterization Theorem and again Nikodym’s Theorem,

one also has the equally thematic result.

Theorem 13.1.3 In a metric space, sets with a shrinking extended Souslin-F

representation have the Baire property.

Actually, this is a direct consequence of the following result, apparently first

noted in Ost2012.

Theorem 13.1.4 In a topological space, the extended Souslin operation ap-

plied to a determining system of sets with the Baire property yields a set with
the Baire property.

Proof We follow the classical ‘separable’ proof given for the usual Souslin
operation as given in RogJ1980, Th. 2.9.2, pp. 43–44, checking that it continues
to hold mutatis mutandis for the choice Ba := Ba(X ) of the family of sets
with the Baire property and M of the meagre subsets of the metric space X.
In particular, we must interpret NN there as κ N throughout, with κ (N) denoting
finite sequences with terms in κ.
206 * Non-separable Variants

By Banach’s Category Theorem, M is closed under σ-discrete unions, and

hence so is Ba (open sets being closed under arbitrary unions).
Assume the extended Souslin operation above is applied to a determining
system of sets hB(σ | n)i in Ba, giving rise to a set
[ \
A= B( j), where B( j) := H ( j | n),
j ∈J n∈N

where {B(σ | n) : σ | n ∈ κ n } is a σ-discrete family for each n.

For σ | n ∈ κ n , put
[
A(σ | n) := {B( j) : j ∈ J, j | n = σ | n}.
j ∈J

Choose C(σ | n) ∈ Ba with A(σ | n) ⊆ C(σ | n) that is a (Baire) hull (cf.

§3.4) for A(σ | n) with the ‘approximation property’ that, for B 0 ∈ Ba, if
A(σ | n) ⊆ B 0 ⊆ C(σ | n),
then C(σ | n)\B 0 ∈ M. Write
\
D(σ | n) := [B(σ | k) ∩ C(σ | k)].
k ≤n
As the system hD(σ | n)i is a refinement of the hB(σ | n)i system, {D(σ |
n) : σ | n ∈ κ n } is also σ-discrete for each n, and so the union {D(σ | n, t) :
S
t ∈ κ} is in Ba. (Note that the sets D(σ | n) are defined as finite intersections
of sets in Ba.)
For σ | n ∈ κ n,
[ [
A(σ | n) = A(σ | n, t) ⊆ D(σ | n, t) ⊆ D(σ | n),
t ∈κ t ∈κ
and each D(σ | n) is a hull for A(σ | n) with the same approximation property
as C(σ | n).
Each set M (σ | n) := D(σ | n)\ {D(σ | n, t) : t ∈ κ} is in M, as M is
S
closed under subset formation, and again the family {M (σ | n) : σ | n ∈ κ n } is
σ-discrete for each n, as before by refinement: M (σ | n) ⊆ D(σ | n). Hence
[ [ [
L := {M (σ) : σ ∈ κ (M) } = {M (σ | n) : σ | n ∈ κ n }
n∈M
is in M, again by Banach’s Category Theorem. Denoting the empty sequence
by hi, put
D = D(hi) = C(hi).
We prove that D\L ⊆ A. Let x be any point of D that is not in L. As
[
x ∈ D(hi), but x < D(hi)\ D(t),
t ∈κ
we may choose t 1 ∈ κ so that
x ∈ D(t 1 ),
 13.1 Non-separable Analyticity and the Baire Property 207

but then
[
x < D(t 1 )\ D(t 1, t).
t ∈κ

Proceeding inductively, we can choose t 1, t 2, t 3, . . ., all in κ, so that

x ∈ D(t 1, t 2, . . . , t n ) for n ≥ 1.

Then, taking τ = (t 1, t 2, t 3, . . .) ∈ J, we have

x ∈ D(τ | n) ⊆ B(τ | n), for all n ≥ 1.

Hence x ∈ A, so D\L ⊆ A and

D\L ⊆ A ∈ M.

As M is (subset) hereditary, D\A ∈ M and

X\(D\A) ∈ M.

Now A = A(hi) ⊆ C(hi) = D so that

A = D ∩ (X\(D\A)) ∈ M.

Thus M is closed under the Souslin operation. 

A similar, but simpler, argument with Ba replaced by M, the Radon measur-

able sets, shows these to be stable under the extended Souslin operation (using
measure completeness and local determination, for which see Fre2003, 412J,
cf. 431A, and measurable hulls, Fre2001, 213L).
Evidently, the standard separable category arguments may also be applied
to σ-discrete decompositions of a set, in view again of Banach’s Category
Theorem.
Finally, since H : J → K (X ) above is upper semi-continuous (for X com-
plete and H = F ), the following theorem, used in the separable context of
Ost2011, §2 and Ost2013c, Th. AC, continues to hold in the non-separable
context (by the same proof), which permits us to quote freely some of its
consequences as established in Chapter 6.

Theorem 13.1.5 Let X be a Hausdorff space and A = K (J), with K : J →

K (X ) compact-valued and upper semi-continuous.
If Fn is a decreasing sequence of (non-empty) closed sets in X such that
Fn ∩ K (J ( j1, . . . , jn )) , ∅, for some j = ( j1, . . .) ∈ J and each n, then
T
K ( j) ∩ n Fn , ∅.

Here, beyond upper semi-continuity, we do not need properties related to the

notion of σ-discrete possessed by the mapping H (for which see HanseJR1983).
208 * Non-separable Variants

13.1.3 Supporting Notions

We will need the following definitions (see below for comments). Recall that
a Hausdorff space X is paracompact (Eng1989, Ch. 5) if every open cover
of X has a locally finite open refinement (so that a (regular) Lindelöf space
is paracompact (Eng1989, 5.1.2), as in the context of Levi’s Open Mapping
Theorem). We employ the terminology of the preceding section.

Definitions
(1) (Hanse1974, §3, cf. Hanse1971, §3.1 and Mich1982, Def. 3.3) Call f : X →
Y base-σ-discrete (or co-σ-discrete) if the image under f of any discrete family
in X has a σ-discrete base in Y . We need two refinements that are more useful
and arise in practice: call f : X → Y an analytic (resp. Baire) base-σ-discrete
map (henceforth A-σ-d , resp. B-σ-d map) if, in addition, for any discrete family
E of analytic sets in X, the family f (E) has a σ-d base consisting of analytic sets
(resp. sets with the Baire property) in Y . We explain in §13.1.4 (Theorem 13.1.6
and thereafter) why A-σ-d maps, though not previously isolated, are really the
only base-σ-discrete maps needed in practice in analytic space theory.
(2) (Hanse1974, §2) An indexed family A := { At : t ∈ T } is σ-discretely
decomposable (σ-d decomposable) if there are discrete families A n := { Atn :
t ∈ T } such that At = n Atn for each t. (The open family {(−r, r) : r ∈ R} on
S
the real line has a σ-d base, but is not σ-d decomposable – see Hanse1973b,
§3.)
(3) (Mich1982, Def. 3.3) Call f : X → Y index-σ-discrete if the image under
f of any discrete family E in X is σ-d decomposable in Y . (Note f (E) is
regarded as indexed by E, so could be discrete without being index-discrete;
this explains the prefix ‘index-’ in the terminology here.) An index-σ-discrete
function is A-σ-d (analytic base-σ-discrete): see Theorem 13.1.6.

Remarks These concepts are motivated Bing’s Theorem (see Chapter 1) that a
regular space is metrizable if and only if it has a σ-discrete base. In a separable
space discrete sets are at most countable. So all the notions above generalize
various aspects of countability; in particular, in a separable metric setting all
maps are (Baire) base-σ-discrete. We now comment briefly on their standing.
(The paper Hanse1974 is the primary source for these.)
1. In (3) above f has a stronger property than base-σ-discreteness. For a proof
see Hanse1974, Prop 3.7(i). Compare Mich1982, Prop 4.3, which shows that f
with closed fibres has (3) if and only if it is base-σ-discrete and has fibres that are
ℵ1 -compact, i.e. separable (in the metric setting). The stronger property is often
easier to work with than Baire base-σ-discreteness. In any case, the concepts are
 13.1 Non-separable Analyticity and the Baire Property 209

close, since for metric spaces and κ an infinite cardinal: X is a base-σ-discrete

continuous image of κ N if and only if X is an index-σ-discrete continuous
image of a closed subset of κ N, both equivalent to analyticity (Hanse1974, Th.
4.1). See Proposition 13.1.8, where it is shown that analytic metric spaces are
the A-σ-d continuous images of κ N .
See Hanse1992, Th. 4.2 or Hanse1974, Th. 4.1. In fact the natural continuous
index-σ-discrete representation of an analytic set has separable fibres; for a
study of fibre conditions, see Hanse1999.

2. Base-σ-discrete continuous maps (in particular, Baire base-σ-discrete and

index-σ-discrete continuous maps) preserve analyticity (Hanse1974, Cor. 4.2).

3. If X is metric and (absolutely) analytic and f : X → Y is injective and closed-

analytic (or open-analytic), then f is base-σ-discrete (Hanse1974, Prop. 3.14).
Base-σ-discreteness is key to this section just as open-analyticity is key to the
separable context of the Levi results in Chapter 12.

4. If B = n∈N Bn , with each Bn discrete, is a σ-d base for the metrizable

S
space X and each f (Bn ) is σ-d decomposable, then f is index-σ-discrete, and
so base-σ-discrete (Hanse1974, Prop. 3.9).

5. A discrete collection A = { At : t ∈ T } comprising analytic sets has the

property that any subfamily has analytic union, i.e. is completely additive ana-
lytic. It turns out that in an analytic space a disjoint (or a point-finite) collec-
tion A is completely additive analytic if and only if it is σ-d decomposable
(see KaniP1975, generalizing the disjoint case in Hanse1971, Th. 2; see also
FroH1981). By the proof of Theorem 13.1.6 the decompositions can be into
analytic sets.

It is Nikodym’s Corollary, 13.1.2, that motivates our interest in analyticity

as a carrier of the Baire property, especially as continuous images of separable
analytic sets are separable, hence Baire.
However, the continuous image of an analytic space is not in general analytic
– for an example of failure, see Hanse1974, Ex. 3.12. But analyticity does obtain
when, additionally, the continuous map is base-σ-discrete, as defined below
(Hanse1974, Cor. 4.2). This technical condition is the standard assumption
for preservation of analyticity and holds automatically in the separable realm.
Special cases include closed surjective maps and open-to-analytic injective
maps (taking open sets to analytic sets).
We may turn now to a discussion of analytic base-σ-discrete maps.
210 * Non-separable Variants

13.1.4 A-σ-Discrete Maps

Their definition requires in addition to base-σ-discreteness that, for any discrete
family E of analytic sets in X, the family f (E) has a σ-discrete base consisting
of sets with the Baire property. The remaining results in this section are gleaned
from a close reading of the main results in Hanse1974 regarding base-σ-discrete
maps, i.e. Hansell’s sequence of results 3.6–3.10 and his Th. 4.1, all of which
derive the required base-σ-discrete property by arguments that combine σ-
discrete decompositions with discrete collections of singletons. We shall see
below that all these results may be refined to the A-σ-d context.

Theorem 13.1.6 An index-σ-discrete function is Analytic base-σ-discrete

(A-σ-d).

Proof Suppose that an indexed family A := { At : t ∈ T } has a σ-discrete

decomposition using discrete families A n := { Atn : t ∈ T }, with At = n Atn
S
for each t, and that all the sets At have the Baire property. A particular case
occurs when these sets are analytic – by Nikodym’s Corollary, 13.1.2. Putting
Ãtn := At ∩ Ātn and A Hn := { Ãtn : t ∈ T }, which is discrete, we obtain a
B-σ-d decomposition for A (or an A-σ-d one in the case of an analytic σ-d
decomposition), and so a fortiori a B-σ-d base for A (or an A-σ-d one).
Thus, if E is a discrete family of analytic sets and f is index-σ-discrete, then
A := f (E) comprises analytic sets (see Remark 2 in §13.1.3); A has, for its
σ-d base, the family n A
S H
n of analytic sets, so with the Baire property, where
AHn are as just given above. 

The A-σ-d maps are closed under composition (cf. Hanse1974, Prop. 3.4);
also by the construction in Theorem 13.1.6, the conclusions of Hanse1974,
Prop. 3.7, Cors. 3.8 and 3.9 yield A-σ-discrete bases for f (E), when E com-
prises analytic sets. In similar vein are the next two refinements of results due
to Hansell – together verifying the adequacy of A-σ-d maps for analytic-sets
theory.

Theorem 13.1.7 (Hanse1974, Prop. 3.10) A closed surjective map onto a

metrizable space is Analytic base-σ-discrete.

Proof Since singletons are analytic, a base that is a discrete family of sin-
gletons is an A-σ-d base. This, combined with the construction in Theorem
13.1.6, refines the argument for Hansell’s cited Prop. 3.10 proving that a closed
surjective map onto a metrizable space is an A-σ-d map. 

Proposition 13.1.8 Analytic metric spaces are the A-σ-d continuous images
of κ N .
 13.2 Nikodym Actions: Non-separable Case 211

Proof By Remark 2 in §13.1.3 again, there is only one direction to consider. So

let S be analytic; we refine Hansell’s argument. Observe first that the argument
in his Prop. 3.5(ii) proves more: if Y is σ-discrete, then any map into Y is A-σ-d
(as in Theorem 13.1.6). Next, let us again use the notation H for analytic sets
established above (in a complete context, with H = F ). We will work in the
closed subspace C ⊆ κ N comprising those j with H ( j) , ∅, and define, as
above, the (continuous) map h on C via H ( j) = {h( j)}. Observe that h takes,
for each n, the discrete family of basic open sets J ( j | n), relativized to C, to
the σ-d family of analytic sets h(J ( j | n) ∩ C), and so h is an A-σ-d map
(by Hanse1974, Cor. 3.9, reported in Remark 4 in §13.1.3). Stone’s canonical
retraction r of κ N onto any closed subspace as applied to the closed subspace
C has σ-discrete range on κ N \C and is the identity homeomorphism on C –
for details see Eng1969 (where r is also shown to be a closed map). So, in
view of the preceding two observations, r is an A-σ-d map. Hence h ◦ r is
A-σ-d, since composition of A-σ-d maps is A-σ-d and provides the required
characterization. 

13.2 Nikodym Actions: Non-separable Case

This short section is devoted to providing a non-separable analogue for the
Effros Theorem of Chapter 11, now that we have the tools to generalize Propo-
sition 11.2.4 to the non-separable context. The wording of its statement is nearly
verbatim: we omit separability and add base-σ-discreteness to the assumption
that point-evaluation maps are continuous; but its proof is more involved.
Proposition 13.2.1 (‘Weak Microtransitivity’) Suppose G is a normed group,
acting transitively on a non-meagre space X with each point-evaluation map
ϕ x : g 7→ g(x) continuous and base-σ-discrete. Then, for each non-empty open
U in G and each x ∈ X, the set U x is non-meagre in X.
In particular, if G is analytic, then G is a Nikodym action.
Proof We first work in the right-sided topology, i.e. derived from the assumed
right-invariant metric d RG (s, t) = kst −1 k. Suppose that u ∈ U, and so without
loss of generality assume that U = Bε (u) = Bε (eG )u (open balls of radius
some ε > 0): indeed,
Bε (eG )u = {xu : d RG (x, e) < ε} = {z : d RG (zu−1, e) < ε}
= {z : d RG (z, u) < ε} = U.
Now put y := ux and W = Bε (eG ). Then U x = Bε (eG )ux = W y. Next work
in the left norm-topology, derived from d LG (s, t) = ks−1 t k = d RG (s−1, t −1 ) (for
212 * Non-separable Variants

which W = Bε (eG ) is still a neighbourhood of eG ). As each set hW for h ∈ G

is now open (since now the left shift g → hg is a homeomorphism), the open
family W = {gW : g ∈ G} covers G. As G is metrizable (and so has a σ-
discrete base), the cover W has a σ-discrete refinement, say V = n∈N Vn ,
S
with each Vn discrete. Put X n := {V y : V ∈ Vn }; then X = n∈N X n, as
S S
X = Gy, and so X n is non-meagre for some n, say n = N. Since ϕy is base-σ-
discrete, {V y : V ∈ VN } has a σ-discrete base, say B = m∈N Bm, with each
S
Bm discrete. Then, as B is a base for {V y : V ∈ VN },
[ [ 
XN = {B ∈ Bm : (∃V ∈ VN )B ⊆ V y} .
m∈N

So for some m, say m = M,

[
{B ∈ Bm : (∃V ∈ VN )B ⊆ V y}

is non-meagre. But as B M is discrete, by Banach’s Category Theorem (cf.

Proposition 11.2.2), there are B̂ ∈ B M and V̂ ∈ VN with B̂ ⊆ V̂ y such that
B̂ is non-meagre. As V refines W, there is some ĝ ∈ G with V̂ ⊆ ĝW, so
B̂ ⊆ V̂ y ⊆ ĝW y, and so ĝW y is non-meagre. As ĝ −1 is a homeomorphism of
X, W y = U x is also non-meagre in X.
If G is analytic, then as U is open, it is also analytic (since open sets are
Fσ and Souslin-F subsets of analytic sets are analytic, cf. RogJ1980), and
hence so is ϕ x (U). Indeed, since ϕ x is continuous and base-σ-discrete, U x is
analytic (Hansell’s Theorem, §13.1), so Souslin-F , and so Baire by Nikodym’s
Corollary, 13.1.2. 

13.3 Coincidence Theorems: Non-separable Case

We begin with Theorem 13.3.1, a non-separable variant of the Levi Open
Mapping Theorem of Chapter 12, a Generalized Levi Open Mapping Theorem
applicable to the broader category of (absolutely) analytic spaces – be they
separable or non-separable metric spaces. We recall that K -analytic sets were
defined in §13.1.

Theorem 13.3.1 Let Y be an analytic space (more generally, paracompact

and K -analytic). Then Y is a Baire space if and only if Y = f (X ) for some
continuous, index-σ-discrete map f on a completely metrizable space X with
the property that there exists a dense completely metrizable Gδ subset Y 0 of
Y such that, with X 0 := f −1 (Y 0 ), the restriction f |X 0 : X 0 → Y 0 is an open
mapping, with X 0 again completely metrizable.
 13.3 Coincidence Theorems: Non-separable Case 213

This is proved below (in §13.3.1). For related results on restriction maps of
other special maps, see Mich1991, §7. (Compare also Hanse1992, Ths. 6.4 and
6.25.) As immediate corollaries one has:
Theorem 13.3.2 (Generalized Levi Coincidence Theorem) If T , T 0 are two
topologies on a set X with (X, T ) a regular Baire space, and T 0 an absolutely
analytic (e.g. completely metrizable) refinement of T such that every T 0-
index-discrete collection is σ-discretely decomposable under T , then there is
a T -dense G(T ) δ -set on which T and T 0 agree.
Proof For f take the identity map from (X, T 0 ) to (X, T ), which is continuous
and index-σ-discrete. 
Corollary 13.3.3 (Almost Completeness Theorem) A space X is almost an-
alytic and a Baire space if and only if X is almost complete.
Proof If X contains a co-meagre analytic subspace A, then, by Theorem
13.3.1, A, being a Baire space, contains a dense completely metrizable subspace
X 0 which is co-meagre in X. So X is almost complete. 
For a sharper characterization in the case of normed groups, see Ost2013c,
Th. 1. Theorem 13.3.2 will enable us to prove (in §13.3.3) the automatic con-
tinuity result of the Semi-Completeness Theorem, 13.3.9, for right-topological
groups with a right-invariant metric d R (the normed groups of Chapter 6). We
write d L (x, y) := d R (x −1, y −1 ), which is left-invariant, and d S := max{d R, d L }
for the symmetrized (‘ambidextrous’) metric. The basic open sets under d S
take the form BεR (x) ∩ BεL (x), i.e. an intersection of balls of ε-radius under d R
and d L centred at x, giving the join (coarsest common refinement) of d R and
d L . Following Ost2013d, for P a topological property it is convenient to say
that the metric space (X, d R ) is topologically symmetrized-P, or just semi-P,
if (X, d S ) has property P. In particular (X, d R ) is semi-complete if (X, d S ) is
topologically complete. As d R and d L are isometric under inversion, (X, d R ) is
semi-complete if and only if (X, d L ) is.
The following result will be needed later in conjunction with Theorem 13.3.7.
Lemma 13.3.4 For a normed group X, if the continuous (identity) embedding
map, j : (X, d S ) → (X, d R ), is index-σ-discrete (resp. base-σ-discrete), then
so also is the inversion mapping from (X, d R ) to (X, d R ); i.e. i : x → x −1 .
Proof Suppose V is a family of sets that is discrete in (X, d R ). Then V −1 :=
{V −1 : V ∈ V } is a family of sets that is discrete in (X, d L ). As the d S
topology refines the d L topology, V −1 is discrete in (X, d S ). Assuming that j
is index-σ-discrete (resp. base-σ-discrete), the family j (V −1 ) = V −1 is σ-d
214 * Non-separable Variants

decomposable (has a σ-d base) in (X, d R ). So inversion maps V to a family

V −1 that is σ-d decomposable (has σ-d base) in (X, d R ). 

13.3.1 Generalized Levi Theorem

The generalized Levi characterization of Baire spaces in terms of open map-
pings in Theorem 13.3.1 is a consequence of the following result, which we
also apply in §13.3.3.
Lemma 13.3.5 (Generalized Levi Lemma) If f : X → Y is surjective, con-
tinuous and Baire base-σ-discrete (in particular, index-σ-discrete) from X
metric and analytic to Y a paracompact space, then there is a dense metriz-
able Gδ -subspace Y 0 ⊆ Y such that, for X 0 := f −1 (Y 0 ), the restriction map
f | X 0 : X 0 → Y 0 is open.
Proof Let A = n A n = { At : t ∈ T } be an open base for X with A n = { Atn :
S
t ∈ Tn } discrete. Then Et := f ( At ) is analytic (see Remarks in §13.1.2), so has
the Baire property by Nikodym’s Corollary, 13.1.2. Let E n = { f ( A) : A ∈ A n }
and let Bn be a σ-d base for E n comprising sets with the Baire property. Put
Bn = m Bnm with each Bnm discrete. Thus for each t ∈ T and Et ∈ E n one
S
has
[ [
Et = {B : B ⊆ Et and B ∈ Bnm }.
m
Put B = nm Bnm . For each B ∈ B pick an open set UB and meagre sets MB0
S
and MB such that
B = (UB \MB0 ) ∪ MB
with MB disjoint from MB0 and with MB0 ⊆ UB . As {B : B ∈ Bnm } is discrete,
the set
[ [
M := {MB : B ∈ Bnm }
mn
is meagre. By paracompactness of Y (cf. Eng1989, Th. 5.1.18), since {UB \MB0 :
B ∈ Bnm } is discrete, for B ∈ Bnm we may select open sets WB with UB \MB0 ⊆
WB with {WB : B ∈ Bnm } discrete. Without loss of generality WB ⊆ UB
(otherwise replace WB by WB ∩ UB ). So UB \MB0 ⊆ WB ⊆ UB and hence
UB \MB0 = WB \MB0 . Then
[ [
M 0 := {WB ∩ MB0 : B ∈ Bnm }
n,m

is also meagre. Now put Y 0 := Y \(M ∪ M 0 ) and W := m,n {WB : B ∈ Bnm },

S S
which is open. Then, for B ∈ Bnm , one has
B ∩ Y 0 = WB ∩ Y 0,
 13.3 Coincidence Theorems: Non-separable Case 215

so that B is open relative to Y 0 and also {Y 0 ∩ WB : B ∈ Bnm } is open and

discrete in Y 0 . Now, for t ∈ T, since Et ∈ E n for some n and Bn is a base for
E n,
[ [
Et ∩ Y 0 = {B ∩ Y 0 : B ⊆ Et & B ∈ Bnm }
n,m
[ [
= {WB ∩ Y 0 : B ⊆ Et & B ∈ Bnm }
n,m

is open in Y 0.
For G ⊆ X open, since A is a (topological) base, we may write
[
G := A nG with A nG := { A : A ⊆ G & A ∈ A n }.
n
Then
[
f (G) := f (A nG )
n
with
f (A nG ) := {E : E = f ( A) & A ⊆ G & A ∈ A n }.
So, for X 0 := f −1 (Y 0 ),
[
f (G ∩ X 0 ) = Y 0 ∩ G
Wnm,
n,m

with
G
Wnm := {WB : B ⊆ f ( A) & B ∈ Bnm & A ⊆ G & A ∈ A n },
which is open in Y 0 .
Since f is continuous and Wnm := {WB : B ∈ Bnm } is discrete for each
m, n, this also shows that the family n,m {Y 0 ∩ WB : B ∈ Bnm } is a σ-d
S
base for Y 0 . Being paracompact, Y is regular (Eng1989, Th. 5.1.5), so the
subspace Y 0 is regular (Eng1989, Th. 2.1.6), and so Y 0 is metrizable by Bing’s
Characterization Theorem (see Eng1989, Th. 4.4.8, cf. Remarks in §13.1.2).
Finally, by replacing the meagre sets M, M 0 by larger sets that are unions of
closed nowhere dense sets, we obtain, in place of Y 0, a smaller, metrizable,
dense Gδ -subspace. 
Remarks 1. For Lemma 13.3.5 we replaced each system {UB : B ∈ Bnm }
to obtain discrete systems {WB : B ∈ Bnm } to reduce the sets MB0 to the sets
MB0 ∩ WB and only then did we take unions. This circumvents the suggested
approach (in a parenthetical remark) to the proof of Th. 6.4(c) in Hanse1992
(by way of representing a Baire set as E = (G E \PE ) ∪ Q E with PE ⊆ E (sic)
for G E open and PE , Q E meagre).
2. Given an arbitrary base B one may replace each B ∈ B with a Baire hull
(§3.4) B+ such that B ⊆ B+ ⊆ B̄. Then B + = n Bn+ is σ-d and Et ⊆ Et+ :=
S
216 * Non-separable Variants

{B+ : B ⊆ Et and B ∈ Bn }, with Et+ \Et meagre. However, it is not clear

S S
n
that the union of these meagre sets is meagre.

Proof of Theorem 13.3.1 Let X be analytic of weight κ. Then for some closed
subset P of κ N there is a continuous index-σ-discrete map f : P → X . Form
P 0 and X 0 analogously to X 0 and Y 0 in the preceding lemma. As X is a Baire
space and X 0 is co-meagre, without loss of generality X 0 is a dense Gδ and is
metrizable. Also P 0 is a Gδ subspace of the complete space κ N, hence is also
topologically complete. So P 0 has the desired properties. As X 0 is metrizable,
the result now follows from Hausdorff’s Theorem that the image under an open
continuous mapping of the completely metrizable space P 0 onto a metrizable
space X 0 is also completely metrizable (for a proof see, e.g., Anc1987, or, for
a more recent account, e.g., HoliP2010).
For the converse, as X 0 is metrizable, the result again follows from Haus-
dorff’s Theorem. Thus X 0 is completely metrizable. But its complement in X
is meagre. So X is a Baire space – in fact an almost complete space. 

13.3.2 Automatic Continuity of Homomorphisms

In the proof of the Semi-Completeness Theorem, 13.3.9, we will need to know
that the inverse of a continuous bijective homomorphism is continuous. In the
separable case this follows by noting that the graph of the homomorphism is
closed and, as a consequence of Theorem 12.1.7 (the Souslin-Graph Theorem),
the inverse is a Baire homomorphism (meaning that preimages of open sets
have the Baire property) and hence continuous. However, in the non-separable
case the paradigm falls foul of the technical requirement for σ-discreteness.
We will employ a modified approach based on the following.

Theorem 13.3.6 (Open Homomorphism Theorem) For normed groups X, Y

with X analytic and Y a Baire space, let f : X → Y be a surjective, continuous
homomorphism, which is base-σ-discrete. Then f is open. So if also X = Y
and f is bijective, then f −1 is continuous.

Proof Suppose that B and A are arbitrary open balls around eX with B4 ⊆ A.
Let D be a dense set in X (e.g. X itself). Now U = {Bd : d ∈ D} is an open
cover of X . (Indeed, if x ∈ X and d ∈ D ∩ Bx, then x ∈ Bd, by symmetry.) Let
V be a σ-d open refinement of U, and say V = n Vn, with each Vn discrete.
S
Then, as X = n ( {V : V ∈ Vn }), we have Y = n ( { f (V ) : V ∈ Vn }). As
S S S S
f is base-σ-discrete each Wn := f (Vn ) = { f (V ) : V ∈ Vn } has a σ-d base
S
Bn ; write Bn := m Bnm with each Bnm discrete. So for each V ∈ Vn one has
S
f (V ) := m {B ∈ Bnm : B ⊆ f (V )}, and so
 13.3 Coincidence Theorems: Non-separable Case 217
[
Y= {B ∈ Bnm : B ⊆ f (V ) for some V ∈ Vn }.
nm

As Y is non-meagre, there are n, m ∈ N, V ∈ Vn and B ∈ Bnm such that

B ⊆ f (V ) and B is non-meagre; for otherwise, since Bnm is discrete, by
Banach’s Category Theorem, {B ∈ Bnm : B ⊆ f (V ) for some V ∈ Vn } is
meagre, implying the contradiction that also Y is meagre. Pick such m, n and
B, V such that B ⊆ f (V ) ⊆ f (Bd). Now V ⊆ Bd for some d ∈ D, as Vn
refines U, and so B ⊆ f (V ) ⊆ f (Bd) = f (B) f (d) is non-meagre. So f (B)
is non-meagre and analytic (as B is analytic). By the Squared Pettis Theorem,
6.3.1, ( f (B)( f (B)) −1 ) 2 = f (B) f (B) −1 f (B) f (B) −1 = f (BB−1 BB −1 ) is a
neighbourhood of eY contained in f ( A). 

The following corollary will be used together with Lemma 13.3.4, where not
every conjugacy will be guaranteed to be continuous.

Theorem 13.3.7 (Continuous Inverse Theorem) If, under d R , the normed

group X is an analytic Baire space and the inversion map i : x → x −1 preserves
σ-discreteness (takes discrete families to σ-discrete families), then the inverse
of any continuous conjugacy, γτ (x) := τxτ −1 , is also continuous.

Proof If {Vt : t ∈ T } is σ-d, then, for any τ ∈ X, so is {Vt τ −1 : t ∈ T }, as

right shifts are homeomorphisms. Applying our assumption about the inversion
map, for any τ the family {τVt−1 : t ∈ T } is σ-d, hence {τVt−1 τ −1 : t ∈ T } is
σ-discrete, and so again {τVt τ −1 : t ∈ T } is σ-d. This means γτ is index-σ-
discrete. By Theorem 13.3.6, if γτ is continuous, then so is its inverse. 

With some minor amendments and from somewhat different hypotheses, the
same proof as in the Open Homomorphism Theorem, 13.3.6, demonstrates the
following generalization of a separable result (given in BinO2010g, Th. 11.11),
but unfortunately without any prospect for achieving the Baire property (see
the Remark at the end of this subsection). Here again the assumed discreteness
preservation is fulfilled in the realm of separable spaces. We give the proof
for the sake of comparison and because of its affinity with a result due to Noll
(Nol1992, Th. 1) concerning topological groups (not necessarily metrizable), in
which the map f has the property that f −1 (U) is analytic for each open Fσ -set
U. In our metric setting, when preimages under the homomorphism f of open
sets are analytic, f is Baire by Nikodym’s Theorem and, since f −1 (A) is a
disjoint. Also, referring to Remark 4 on page 209, completely additive analytic
for A discrete, the σ-d decomposability condition given below is satisfied by
Hansell’s result (Hanse1971, Th. 2) cited in the Remarks of §13.1.2. Noll shows
the σ-d decomposability condition below is satisfied when X is a topologically
218 * Non-separable Variants

complete topological group (using the FroH1981 generalization of Hansell’s

result and of KaniP1975 – cf. Remark 5 on page 209).
Theorem 13.3.8 (Baire Homomorphism Theorem) For normed groups X, Y
with X analytic, a surjective Baire homomorphism f : X → Y is continuous
provided f −1 (A) is σ-discretely decomposable for each σ-discrete family A
in Y .
Proof We proceed as above but now in Y . For ε > 0, with B = Bε/4 (eY ) open
and D any dense set in Y choose ad with f (ad ) = d ∈ D. Put T := f −1 (B),
which has the Baire property (as f is Baire). As Y is metrizable, the open cover
{Bd : d ∈ D} has a σ-d refinement A = n A n with A n := { Atn : t ∈ Tn }
S
discrete. For each n, by assumption, we may write { f −1 ( Atn ) : t ∈ Tn } =
S
nm {Btnm : t ∈ Tn } with {Btnm : t ∈ Tn } discrete in X for each m and n. Now
[ [
X = f −1 (Y ) = { f −1 ( Atn ) : t ∈ Tn } = {Btnm : t ∈ Tn }.
n nm
S
As X is a Baire space, there are n, m such that {Btnm : t ∈ Tn } is non-meagre.
Again by Banach’s Category Theorem, and since {Btnm : t ∈ Tn } is discrete,
there is t with Btnm non-meagre. But Btnm ⊆ f −1 ( Atn ) ⊆ f −1 (Bd) = T ad for
some d ∈ D, as A refines {Bd : d ∈ D}. Thus T ad and so T is non-meagre,
as the right shift ρ a d (x) = xad is a homeomorphism. But T has the Baire
property and X is analytic, so T contains a non-meagre analytic subset. By the
Squared Pettis Theorem, 6.3.2, (TT −1 ) 2 contains a ball Bδ (eX ). Then
Bδ (eX ) ⊆ f −1 [(Bε/4 ) 4 ] = f −1 [Bε (eY )],
proving continuity at eX . 
Remark In the separable case, by demanding that the graph Γ of a homomor-
phism be Souslin-F (X ×Y ) one achieves the Baire property of sets f −1 (U), for
U open in Y, by projection parallel to the Y -axis of Γ ∩ (X × U), provided that
Y is a K -analytic space. For Y absolutely analytic, one has an extended Souslin
representation, and hence a representation of Y as an upper semi-continuous
image of some product space κ N . But the proof of the projection theorem in
RogJ1980, Ths. 2.6.5 and 2.6.6, now yields that the projection of a Souslin-
F (X × Y ) set has only a Souslin-F (X ) representation relative to κ N , without
guaranteeing the σ-discreteness requirement. In the non-separable context the
Baire property can be generated by a projection theorem, provided one has both
that the graph is absolutely analytic and that the relevant projection, namely
(x, f (x)) → x, is base-σ-discrete (cf. Hanse1992, Th. 4.6; Hanse1974, §6;
Hanse1971, §3.5).
These general results have their roots in much simpler contexts. See, for
instance, BinO2009e; BinO2015.
 13.3 Coincidence Theorems: Non-separable Case 219

13.3.3 Semi-Completeness Theorem: From Normed to Topological

Groups
In this section the generalized Levi result in Theorem 13.3.2 is the key ingre-
dient; we will use it and results of earlier sections to prove the following. Here
d S is as in §12.4.
Theorem 13.3.9 (Semi-Completeness Theorem; Ost2013d) For a normed
group X, if (X, d R ) is semi-complete and a Baire space, and the continuous
embedding map j : (X, d S ) → (X, d R ) is Baire base-σ-discrete (e.g. index-σ-
discrete), in particular if X is separable, then the right and left uniformities
of d R and d L coincide and so (X, d R ) is a topologically complete topological
group.
Remarks 1. Theorem 13.3.9 generalizes a classical result for abelian locally
compact groups due to Ellis (Elli1953).
2. Remarks in §13.1.2 noted that the ‘index-σ-discreteness condition’ imposed
in the theorem is a natural one from the perspective of the non-separable theory
of analytic sets, and Lemma 13.3.4 interprets this in terms of inversion, cf.
Remark 4.
3. For separable spaces, where discrete families are countable and so the embed-
ding j above automatically preserves σ-discreteness, the result here was proved
in Ost2013d (to which we refer for the literature) in the form that a semi-Polish,
normed group X, Baire in the right norm-topology, is a topologically complete
topological group. Rephrased in the language of uniformities generated by the
norm (Kel1955, Ch. 6, Pb. O), this says that a normed group, Polish in the
ambidextrous uniformity and Baire in either of the right or left uniformities,
has coincident right and left uniformities, and so is a topological group. Key to
its proof is that a continuous image of a complete separable metric space is a
classically analytic space. So the ‘index-σ-discreteness’ condition is exactly the
condition that secures preservation of analyticity. In the non-separable context
continuity is not enough to preserve analyticity, and an additional property is
needed, involving σ-d as above: see Hanse1999, Example 4.2 for a non-analytic
metric space that is a one-to-one continuous image of κ N for some uncountable
cardinal κ (so a continuous image of a countable product of discrete, hence
absolutely analytic, spaces of cardinal κ).
4. More recent work by Holický and Pol (HoliP2010), in response to Ostro-
vsky’s insights and based on Mich1986, especially §6 (which itself goes back to
GhoM1985), connects preservation of (topological) completeness under con-
tinuous maps between metric spaces to the classical notion of resolvable sets.
(The latter notion provides the natural generalization to Ostrovsky’s setting;
220 * Non-separable Variants

see also Holi2010 for non-metrizable spaces.) They find that a map f preserves
completeness if it ‘resolves countable discrete sets’, i.e. for every countable
metrically discrete set C and open neighbourhood V of C there is L with
C ⊆ L ⊆ V such that f (L) is a resolvable set.
Consider the implications for a group X with right-invariant metric d R (see
above), when, for f , one takes j, the identity embedding j : (X, d S ) → (X, d R )
and when C = {cn : n ∈ N} is a d S -discrete set (so that C and C −1 are d R -
discrete). To obtain the desired resolvability for j, it is necessary and sufficient,
for each C as above and each assignment r : N → R+ with r n → 0, that
there exist d R -resolvable sets L n ⊆ BrRn (cn ) ∩ BrLn (cn ). Since BrL (c) = {x :
d R (c−1, x −1 ) < r } = {x : d R (c−1, y) < r and y = x −1 } = {y −1 : d R (c−1, y) <
r }, this is yet another condition relating inversion to the d R -topology, via the
sets BrR (c−1 ) −1 .
We turn now to the proof of Theorem 13.3.9. The proof layout (preparatory
lemmas followed by proof) and strategy are the same as in Chapter 12, but,
as some of the details differ, it is convenient to repeat the short common part
(most of the proof of Lemma 13.3.11).

Lemma 13.3.10 For a normed group X, if (X, d S ) is topologically complete

and the continuous embedding map j : (X, d S ) → (X, d R ) is Baire base-σ-
discrete (e.g. A-σ-d, in particular index-σ-discrete), then there is a dense
absolute-Gδ subset Y 0 in (X, d R ) such that the restriction map j : (Y 0, d S ) →
(Y 0, d R ) is open, and so all the three topological spaces, (Y 0, d S ), (Y 0, d R ) and
(Y 0, d S ), are homeomorphic (and topologically complete).

Proof Take X = (X, d S ) which is metric and analytic and Y = (X, d R ) which
is paracompact; then j : X → Y is continuous, surjective and Baire base-σ-
discrete. By Lemma 13.3.5 there is a dense Gδ -subspace Y 0 of (X, d R ) such
that j | Y 0 is an open mapping from (Y 0, d S ) to (Y 0, d R ). Now j −1 (Y 0 ) is a
Gδ -subspace of (X, d S ), so it follows that (Y 0, d S ) is topologically complete, so
an absolute-Gδ . So j | Y 0 embeds Y 0 as a subset of (X, d S ) homeomorphically
into Y 0 as an absolute-Gδ subset of (X, d R ). Since x n →S x if and only if
x n →R x and x n →L x, all three topologies on Y 0 agree. 

Lemma 13.3.11 If in the setting of Lemma 13.3.10 the three topologies

d R, d L, d S agree on a dense absolutely-Gδ set Y of (X, d R ), then for any τ ∈ Y
the conjugacy γτ (x) := τxτ −1 is continuous.

Proof We work in (X, d R ) which is thus analytic (as a base-σ-discrete con-

tinuous image – Remark 2 in §13.1.3). Let τ ∈ Y . We will first show that the
conjugacy x → τ −1 xτ is continuous in X at e, and then deduce that its inverse,
 13.3 Coincidence Theorems: Non-separable Case 221

x → τxτ −1 , is continuous. So let z n → e be any null sequence in X. Fix ε > 0;

then T := Y ∩ BεL (τ) is analytic and non-meagre, since X is a Baire space
(and Y ∩ BεL (τ) is d R -open in Y with Y an absolute Gδ ). By the Analytic Shift
Theorem, 6.2.3, there are t ∈ T and t n in T with t n converging to t (in d R ) and
m z m t m : m ∈ Mt } ⊆ T . Since the three topologies
an infinite Mt such that {tt −1
agree on Y and the subsequence tt −1 m z m t m of points of Y converges to t in Y
under d R (see the Remark after Theorem 6.2.3), the same is true under d L .
m z m t m, t) = d L (t m z m t m, e) = d L (z m t m, t m ), we note
Using the identity d L (tt −1 −1

that
kt −1 z m t k = d L (t, z m t) ≤ d L (t, t m ) + d L (t m, z m t m ) + d L (z m t m, z m t)
≤ d L (t, t m ) + d L (tt −1
m z m t m, t) + d L (t m, t) → 0,

as m → ∞ through Mt . So d L (t, z m t) < ε for large enough m ∈ Mt . Then for

such m, as d L (τ, t) < ε,
kτ −1 z m τk = d L (z m τ, τ) ≤ d L (z m τ, z m t) + d L (z m t, t) + d L (t, τ)
≤ d L (τ, t) + d L (t, z m t) + d L (t, τ) ≤ 3ε.
Thus there are arbitrarily large m with kτ −1 z m τk ≤ 3ε. Inductively, taking
successively ε = 1/n and k (n) > k (n − 1) to be such that kτ −1 zk (n) τk ≤ 3/n,
one has kτ −1 zk (n) τk → 0. By the weak continuity criterion Lemma 6.1.10
(cf. BinO2010g, Lemma 3.5, p. 37), γ(x) := τ −1 xτ is continuous. Hence, by
Lemma 13.3.4 and Theorem 13.3.7, γ −1 (x) is also continuous. 
Proof of the Semi-Completeness Theorem, 13.3.9 Under d R, the set ZΓ :=
{x : γx is continuous} is a closed subsemi-group of X (BinO2010g, Prop.
3.43, but using the Open Homomorphism Theorem, 13.3.6, in place of the
Souslin-Graph Theorem, 12.1.7). So, as Y is dense, X = clR Y ⊆ ZΓ, i.e. γx is
continuous for all x, and so (X, d R ) is a topological group, by the Equivalence
Theorem, 6.1.9. So x n →R x if and only if x −1 n →R x
−1 if and only if x → x
n L
if and only if x n →S x. So (X, d R ) is homeomorphic to (X, d S ). Hence the
topological group (X, d R ) is topologically complete, being homeomorphic to
(X, d S ). 
 14

Contrasts between Category and Measure

14.1 Classical Results

We have seen above, and can also see in Oxtoby’s book (Oxt1980), that there
are extensive analogies and similarities between measure and category. For
instance, one has (Oxt1980, Ch. 19) the Sierpiński–Erdős duality principle:
that under the Continuum Hypothesis (CH) any valid statement involving only
null sets (or their complements) may be translated into its analogue involving
only meagre sets (or their complements), and conversely. Similarly (Oxt1980,
Ch. 17), in ergodic theory, duality extends to some but not all forms of the
Poincaré recurrence theorem, and, in probability theory (Oxt1980, Ch. 21),
duality extends as far as the zero–one law (RaoR1974) but not to the strong
law of large numbers. One may loosely summarize this as expressing a duality
between category theory and qualitative measure theory.
A number of distinguished authors have commented on how later develop-
ments forced a reassessment. Some quotes, in chronological order:

Bartoszyński and Judah in 1995 The Preface to their book Set theory: On the
structure of the real line (BartoJ1995) begins ‘This book reflects the
current progress in an important segment of descriptive set theory.
Its main focus is measure and category in set theory, most notably
asymmetry results.’
Bagaria and Woodin in 1997 (à propos of 1984 work by Shelah) ‘Thus, an in-
teresting asymmetry between measure and category was uncovered.
Up to these results, measure and category were regarded as symmet-
rical properties . . .’ (BagW1997, 1380).
Fremlin in 2008 on Cichón’s diagram (Fre2008, 522 Notes and comments)
(Referring forward to §14.2.) ‘For many years it appeared that “mea-
sure” and “category” on the real line, or at least the structures (R, B, N )

222
 14.1 Classical Results 223

and (R, B, M) where B is the Borel σ-algebra of R, were in a sym-

metric duality. It was perfectly well understood that the algebras
A = B/B ∩ N and C = B/B ∩ M . . . are very different, but . . .
encouraged us to suppose that anything provable in ZFC relating mea-
sure to category ought to respect the symmetry. It therefore came as
a surprise to most of us when Bartoszyński and Raisonnier & Stern
showed that add(N ) ≤ add(M) in all models of set theory.’

Of course, probability theory abounds in (quantitative) statements that have

no category (qualitative) analogue. In the classical limit theorems, for example,
one has the law of large numbers mentioned above, which involves the mean
µ, the central limit theorem, which involves the mean and the variance σ 2 ,
and similarly for the law of the iterated logarithm, etc. This illustrates a sense
in which measure (and probability) theory is stronger than category theory,
breaking the symmetry between the two that has been such a striking feature
of the results above.
We shall deal with random series, in particular random trigonometric series,
areas much influenced by Jean-Pierre Kahane (1926–2017) (see also Kahane’s
survey Kah1997). Kahane was very struck by Körner’s description (Korn1995)
of the Baire category theorem as ‘a profound triviality’, and by Kaufman’s use
of Baire category methods in Fourier analysis (Kau1967; Kau1995). Of Baire’s
theorem, he says ‘Its proof is trivial, and its use by Kaufman and Körner is
profound’ (Kah2001, §10 p. 70).
We turn now to classical instances of this asymmetry, in situations where
both category and measure theories can be used, but say different (sometimes
very different) things.
Liouville Numbers A striking classical example is that of the Liouville num-
bers (Lio1851; for textbook accounts, see Bug2012; Per1960, §46; Oxt1980,
Ch. 2): irrationals for which for each positive integer n there are integers q > 1
and p with
|z − p/q| < 1/q n .

Every Liouville number is transcendental. The set of Liouville numbers is co-

meagre (large in category) but Lebesgue-null (small in measure). Indeed, it has
s-dimensional Hausdorff measure zero for every s > 0.
Banach–Tarski (‘Paradoxical’) Decompositions The results here are partic-
ularly dramatic. The Banach–Tarski decomposition (BanT1924) is justly fa-
mous. We refer to Tomkowicz and Wagon (TomW2016) for a textbook expo-
sition. It concerns the (apparently ‘paradoxical’) possibility of decomposing a
Lebesgue-measurable set (a Euclidean ball, say) into finitely many pieces, and
224 Contrasts

reassembling these pieces into a measurable set of (say) double the measure.
Of course, the constituent pieces are necessarily non-measurable, or additivity
of measure would be violated. But it was shown by Dougherty and Foreman
(DouF1994) that they may have the Baire property.
Restriction and Continuity The condition for a function to have the Baire
property is that it be continuous off some meagre set (see, e.g., Kur1966, §28,
II; Ch. 12). The corresponding result for measurability is Lusin’s Theorem: that
the function be continuous off some sets of arbitrarily small positive measure
(Oxt1980, Th. 8.2). This condition cannot be relaxed to continuity off some
null set, as Oxtoby shows by example just after the proof of Lusin’s Theorem.
Random Series We follow Kahane’s book (Kah1985) and papers (Kah1997;
Kah2000; Kah2001).
Consider an infinite sequence of independent Rademacher random variables
{ n } – tosses of a fair coin, taking values ±1 with probability 21 each. For a
sequence ∞
P
0 cn of realsPcn , we may ask for the probability that the correspond-
ing Rademacher series  n cn – which we write more suggestively as ±cn
P
– converges. By the zero–one law (see, e.g., Kah1985, p. 7), convergence is a
tail event (invariant under deletion of finitely many terms), and so has prob-
ability 0 or 1: convergence is almost sure (a.s.), or divergence is a.s. (This is
true generally; specializing to the Rademacher case reduces the measure theory
from quantitative to qualitative, where category and measure can be ‘fairly
compared’.) As above, duality extends as far as the zero–one law; the series
either converges quasi-surely (q.s.) or diverges q.s. But in stark contrast, the
conditions on c = (cn ) are far apart:
X X
±cn converges a.s. [diverges a.s.] ⇔ c ∈ [<]` 2 : cn2 < ∞ [= ∞].
But
X X
±cn converges q.s. [diverges q.s.] ⇔ c ∈ [<]` 1 : |cn | < ∞ [= ∞].
For the first, the sufficiency of c ∈ ` 2 for a.s. convergence is due to Rademacher
(Rad1922), the necessity to Khintchin and Kolmogorov (KhiK1925). The sec-
ond, which is simpler, is due to Kahane (Kah2001, §2). To quote Kahane:
divergence is quasi-sure as soon as it may happen.
Normal Numbers In the decimal expansion of a real number x ∈ [0, 1],
∞
X
x = 0.x 1 x 2 · · · x n · · · = x n /10n,
1

x is said to be normal (to base 10) if all 10 digits occur with equal asymptotic
frequency, 1/10. We probabilize [0, 1] by imposing Lebesgue measure on it.
 14.1 Classical Results 225

By the law of large numbers, x is then normal a.s. Similarly for the expansion
to base d = 2, 3, . . .. Intersecting over all d, the real number x is normal to
all bases simultaneously, a.s. (since for the measures of the exceptional sets,
1 0 = 0). This deals with single digits, but the argument extends to pairs of
P∞
digits (including one shift), triples (including two shifts), etc. Thus all k-tuples
occur with the same asymptotic frequency d −k , even allowing for shifts. We
summarize this by saying that x is strongly normal a.s. This is Borel’s normal
number theorem (Bore1909). Apart from its intrinsic importance, the result is
of historic interest as the first example of a strong (measure-theoretic) law of
large numbers in probability theory and the beginning of the metrical theory of
numbers (for which see, e.g., Harm1988).
The category analogue of Borel’s Theorem fails in a strong sense (or other-
wise put, is completely different). It was shown by S̆alát (Sal1966) that the set of
normal numbers is meagre. One can go much further: Olsen (Ols2004) defines
and studies ‘extremely non-normal numbers’, and shows that such behaviour is
generic from a category viewpoint: the set of such numbers is residual – while
being Lebesgue-null. Hyde et al. (HydLO2010) show that this stark contrast
even persists despite the smoothing effect of repeated Cesàro averaging.
Aside on Normal Numbers Borel’s Theorem shows that normal numbers are
generic, at least in measure (though not, by above, in category). So normal
numbers exist, in great profusion. So it seems odd – it is certainly very striking
– that not a single example is known explicitly – though some partial results are
known. For instance, Champernowne’s number (Cha1933)

0.123456789101112 · · · 99100101 · · ·

is normal to all bases d that are powers of 10. If we were to seek for a specific
example to test, the prime candidate would surely be π. Its decimal expansion is
known to trillions of places (62.8 currently), and passes all tests of equidistribu-
tion with flying colours. This of course gives no information whatsoever on the
question of its normality. But decimal expansions and base 10, however useful
in ordinary life, have no claim to be natural; nor even do binary expansions and
base 2, however useful in computer science. There is only one natural way to
expand π, as a continued fraction. This was done by Brouncker in 1655; see
Khrushchev (Khr2008, §2):
4 12 32 52  (2n − 1) 2 
=1+ · · · = 1 + K∞ ,
π 2+ 2+ 2+ n=1
2
to use both the usual notations for a continued fraction (‘K for Kettenbruch’).
(Brouncker’s continued fraction for 4/π is closely related to Wallis’s product
for π, in his Arithmetica Infinitorum of 1656.) The next candidate might be e,
226 Contrasts

but here again, the only natural way to expand is by a continued fraction, due
to Euler in 1737 (and Cotes, in 1714; see Brez1991, 4.1):
1 1 1 1 1 1 1 1
e =2+ ··· .
1+ 2+ 1+ 1+ 4+ 1+ 1+ 6+
Topological and Hausdorff Dimension For topological dimension, we refer
to Hurewicz and Wallman (HurW1941) and Edgar (Edg1990, Ch. 3). Here a
theory is developed which gives a natural extension to that of Euclidean dimen-
sion. Here dimension takes values in N ∪ {−1, 0, ∞}; one needs to distinguish
between small and large inductive dimension; 0-dimensional spaces (having a
topological basis of clopen (closed and open) sets) play a major role.
For the theory of Hausdorff dimension, which is much more detailed and
quantitative, we refer to Rogers (Rog1970) and Falconer (Fal1985). Here the
dimension takes non-negative real values; for fixed dimension, a fine-detailed
comparison is possible between sets according to the size of their Hausdorff
measure functions. This gives a valuable and quantitative way to measure and
compare the size of small sets: sets such as the sample paths, zero-sets, self-
intersections, etc., of stochastic processes, such as Brownian motion (in various
dimensions), Lévy processes, etc. It suffices for us here to refer to the extensive
work by John Hawkes (1944–2001); for details and references, see his obituary
by S.J. Taylor (Tay2004).
For results in the context of normal numbers, see GolSW2000; AlbGI2017;
Sal1966.
Random Dirichlet Series We use the standard notation for Dirichlet series,
1 an n , where s = σ + it ∈ C. Convergence is in a half-plane σ > σc , to a
P∞ −s

function A(s) analytic there. Absolute convergence is in a half-plane σ > σa ,

where σa ≥ σc . One has A(σ + it) = O(|t|) for |t| → ∞. The Lindelöf (or
order) function is

µ(σ) := inf{α : A(σ + it) = O(|t| α )} (|t| → ∞).

Then
µ(σa ) = 0, µ(σc ) ≤ 1.

The Bohr abscissa σb is the infimum of σ with µ(σ) finite; then

σb ≤ σc ≤ σa .

We turn next to how cancellation effects in Dirichlet series may enable

analytic continuation. We illustrate with a classic example, the alternating zeta
function (or Dirichlet eta function). Write the Riemann zeta function
 14.1 Classical Results 227
∞
X X X X X
ζ (s) := 1/ns = + = + ,
1 n odd n even o e

say, and similarly for the alternating zeta function

∞
X X X
(−) n−1 /ns = − .
1 o e

Subtract:
∞
X X ∞
X
ζ (s) − (−) n−1 /ns = 2 =2 1/(2n) s = 21−s ζ (s) :
1 e 1

(−) n−1 /ns

P∞
1
ζ (s) = ;
1 − 21−s
Hardy (Har1922) and Titchmarsh (Tit1986, (2.2.1), p. 512). This gives (that
ζ has a simple pole at 1 of residue 1 and) the analytic continuation of ζ from
Re(s) > 1 to Re(s) > 0.
Kahane (Kah2000, §5) gives two examples with random rather than alter-
nating signs. For ‘the Riemann zeta function with random signs’, ∞ −s
P
1 ±n ,
one has
1 1
σa = 1, µ(1) = 0; σc = , µ = 0 a.s., but σc = σa = 1 q.s.
2 2
For
X∞
± ((2n − 1) −s − (2n) −s ),
1

one has again σa = 1, µ(1) = 0,

1 1  f 1 g
σb = − , σc = 0, µ(σ) = −σ on − , ∞ a.s.,
2 2 + 2
but
σb = σc = 0, µ(σ) = (1 − σ)+ on [0, ∞] q.s.

Random Taylor Series; Random Fourier Series; Nowhere Differentiable Func-

tions These matters are studied at length by Kahane (Kah1985; Kah1997).
Kahane’s interest in category aspects is developed in Kah2000; Kah2001. We
refer there for a wealth of results, comparing and contrasting the category and
measure cases. See also KahQ1997; Que1980; Korn1995.
For random Taylor series, results include those on the sense in which the
circle of convergence of the series being a natural boundary – i.e. analytic
continuation across any part of it is impossible – is generic.
228 Contrasts

Erdős–Sierpiński maps We refer to Oxt1980, Ch. 19 for background, under

the Continuum Hypothesis (CH), on the Erdős–Sierpiński Duality Theorem.
This duality may be implemented by applying an Erdős–Sierpiński map. Bar-
toszyński (Barto2000) notes an asymmetry between category and measure here,
implying in particular that there do not exist additive Erdős–Sierpiński maps.
Filters Bartoszyński et al. (BartoGJ1993) show that it is consistent with ZFC
that all filters with the Baire property are Lebesgue-measurable, but provable
in ZFC that there is a Lebesgue-measurable filter without the Baire property.
Genericity (as typicality) The category property ‘quasi everywhere’ and its
measure analogue ‘almost everywhere’ provide different (and as above some-
times contrasting) ways in which behaviour or a property can be generic.
Aspects of genericity have been developed in a number of papers by Grosse-
Erdmann; see, e.g., Gros1987; Gros1999.
The Fubini and Kuratowski–Ulam Theorems Fubini’s Theorem may be reg-
arded as having a qualitative variant, the Fubini Null Theorem (§9.5, Douw1989,
BinO2016b) stated in the vocabulary of null sets and asserting that a measur-
able planar set with the Baire property is null if and only almost all its vertical
segments are null. This has a category analogue in the Kuratowski–Ulam Theo-
rem (§9.5, Oxt1980, Ch. 15). The latter asserts that a Baire planar set is meagre
if and only if all but a meagre set of verticals have meagre vertical sections. So
much for similarity. However, despite both being qualitative statements, there
is a hidden contrast: the category result is known to fail beyond the separable
context (as shown in Pol1979, cf. MilP1986, but see FreNR2000), though not
so the Fubini Theorem (see BinO2020c, Th. FN and Chapter 13).
Lipschitz Functions and Rademacher’s Theorem Rademacher’s Theorem of
1919 states that every Lipschitz function f : Rm → Rn is a.e. differentiable
(Rad1919; Fed1969; AlbM2016). The question arises as to a category analogue.
Loewen and Wang (LoeW2000, Th. 4) construct a set in a suitably metrized
space of Lipschitz functions which is co-meagre, but each member of which
is differentiable on an at most meagre set. Thus differentiability of Lipschitz
functions is generic in (Lebesgue) measure, yet non-differentiability can be
generic in category.
Sets of Uniqueness for Trigonometric Series While sets of uniqueness for
trigonometric series provide an example of category–measure duality (§9.1),
they also provide a contrast. The category case (KecL1987, Problem C, p. 4) is
much harder to prove (KecL1987, Ch. VIII).
Natural Boundaries of Random Series Steinhaus in his classic paper (Stei1930)
showed that if a power series with radius of convergence R ∈ (0, ∞) is made
random by inserting independent random factors each uniformly distributed
 14.2 Modern Results: Forcing 229

on [0, 1] into the coefficients, then the resulting random series has the circle
of convergence as its natural boundary (no analytic continuation is possible
across it – see, e.g., Simo2015), almost surely. The dual Baire statement, with
a meagre rather than a null exceptional set, is also true but again was shown
much later, by Breuer and Simon (BreuS2011; Simo2015, p. 241, Problem 9).
For background on such random Taylor series, see, e.g., Kah1985, Ch. 4 (and
for more on the Gaussian case, Kah1985, Ch. 13).
We close this section with two insightful comments by Kahane on category–
measure contrasts:
Kah2000, p. 169: ‘If we measure the thinness of a set through the behaviour of
positive measures carried by the set and their Fourier coefficients, the tendency
of Baire’s methods is to provide sets as thin as possible, and the tendency of
probability methods is to provide measures whose Fourier coefficients tend to
zero as fast as possible.’
Kah2001, p. 61: ‘We see from these examples that the Baire approach empha-
sizes divergence, singularities and large values, while the probability approach
favors convergence, smoothing and regularizing effects.’

14.2 Modern Results: Forcing

We turn now from these ‘classical’ matters, in analysis and probability – which,
though of ongoing interest, have their roots in the work of Borel, Lebesgue,
Baire or earlier – to ‘modern’ aspects, stemming from set theory and logic, with
particular reference to forcing, following Cohen’s breakthrough in 1964 and
1966. We will not attempt a detailed account, for which we refer the interested
reader to our long survey (BinO2019a) and its extensive bibliography.
We recall the view of Judah (formerly Ihoda) and Spinas (JudSp1997):
‘Finally Cohen’s method of forcing led to a much deeper understanding and
shed light on a deep asymmetry between measure and category.’
We begin with Solovay’s paper (Solo1970). Solovay assumed that the exis-
tence of an inaccessible cardinal (see §16.3) is consistent with ZF. He showed
that then ZF is consistent with the conjunction of (1)–(5):
(1) DC, the Axiom of Dependent Choice;
(2) LM: all sets of reals are Lebesgue-measurable;
(3) PB: all sets of reals have the Baire Property;
(4) Every uncountable set of reals has a perfect subset;
(5) If A := { Ax : x ∈ R} is a class of non-empty sets of reals, then:
230 Contrasts

(a) there exists a Borel function h1 : R → R such that {x : h1 (x) < Ax } is

Lebesgue-null; and
(b) there exists a Borel function h2 : R → R such that {x : h2 (x) < Ax } is
meagre.

Cohen and forcing changed the situation regarding category and measure
with the work of Shelah (Shel1984). See also the companion paper by Raison-
nier Rai1984; the two are published contiguously, and were reviewed jointly
(MR86g:03082a,b) by F. R. Drake. Shelah’s task was to show that the inaccessi-
ble cardinal used in Solovay’s paper (Solo1970) cannot be removed. As Hrbacek
remarks in his review (MR800191) of Raisonnier and Stern (RaiS1985):
‘the classical results of descriptive set theory concerning measure and category
are completely symmetric. Thus it was a surprise when S. Shelah showed that:

(a) Lebesgue measurability of all sets of reals implies consistency of inacces-

sible cardinals,
but
(b) Models of ZF where all sets of reals have the Baire Property can be
constructed under ZF alone.’

Thus Solovay’s inaccessible cardinal is necessary for measure but not for
category.
Write Σn1 (L) as shorthand for ‘every Σn1 set is Lebesgue-measurable’, and
similarly Σn1 (B) for ‘every Σn1 set has the Baire property’. With this notation,
Bartoszyński (Barto1984) showed that

Σ21 (L) ⇒ Σ21 (B),

while Raisonnier and Stern (RaiS1985) proved that the reverse implication
fails. This shows again a striking asymmetry between category and measure,
and another sense in which measure is stronger than category, as expected:
measure theory has a quantitative side while category theory does not. One
cannot replace Σ21 by ∆12 here: Kanovei and Lyubetskii (KanoL2003, Th. 5.4).
Ihoda and Shelah (IhoS1989) give a study of ∆21 sets of reals (recall that
these are the sets particularly suited for the study of regular variation, as noted
in Chapter 3). They show (Th. 3.5) that if ZFC is consistent, then:

(i) there exists a model V in which all ∆12 sets are Lebesgue-measurable, but
that not all have the Baire property, and not all are Ramsey;
(ii) similar statements hold with Lebesgue-measurable, Baire property and
Ramsey cyclically permuted.
There is a fourth property, ‘Kσ -regularity’, for which see IhoS1989.
 14.3 Category Duality Revisited 231

They (now as Judah and Shelah, JudS1993a) follow this with a corresponding
study of ∆13 sets, with similar conclusions. This in turn is taken further by
Fischer, Friedman and Khomskii (FisFK2014), who assume ZFC or ZFC plus
the assumption of an inaccessible cardinal. They give some partial results for
∆14 .
The interesting recent monograph of Friedman and Schrittesser (FriS2020)
is on measurability without the Baire property in the projective hierarchy. They
write in the Abstract:
‘We prove that it is consistent (relative to a Mahlo cardinal) that all projective sets of
reals are Lebesgue measurable, but there is a ∆13 set without the Baire property. The
complexity of the set which provides a counterexample to the Baire property is optimal.’

Similar questions were earlier considered by Bagaria and Woodin

(BagW1997). They show that if φ is a Σ41 sentence consistent with the measur-
ability of all Σ21 sets, then it is still consistent with the measurability of all Σ21
sets and the existence of a ∆12 set without the Baire property. This asymmetry
is maximal (with respect to Σ41 ) in the sense that there exists a Π41 sentence
which implies in ZFC that all projective sets are measurable and have the Baire
property.

14.3 Category Duality Revisited

14.3.1 Practical Axiomatic Alternatives
While ZF is common ground in mathematics, the Axiom of Choice( AC) is
not, and alternatives to it are widely used, in which for example all sets are
Lebesgue-measurable (usually abbreviated to LM) and all sets have the Baire
property, sometimes abbreviated to PB (as distinct from BP to indicate indi-
vidual ‘possession of the Baire property’). One such is DC above. As Solovay
(Solo1970, p. 25), points out, this axiom is sufficient for the establishment of
Lebesgue measure, i.e. including its translation invariance and countable addi-
tivity (‘. . . positive results . . . of measure theory. . . ’), and may be assumed
together with LM. Another is the Axiom of Determinacy (AD) introduced by
Mycielski and Steinhaus (MycS1962) (see Klei1977 in relation to DC); this
implies LM, for which see MycSw1964, and PB, the latter a result, as men-
tioned in §2.4, due to Banach – see Kec1995, 38.B. Its introduction inspired
remarkable and still current developments in set theory concerned with deter-
minacy of ‘definable’ sets of reals (see ForK2010 and particularly Neem2010)
and consequent combinatorial properties (such as the partition relations) of the
alephs (see Klei1977); again see §2.4 (games of Banach–Mazur type). Others
232 Contrasts

include the (weaker) Axiom of Projective Determinacy (PD) (Kec1995, §38.B)

cf. Chapter 16, restricting the operation of AD to the smaller class of projective
sets. (The independence and consistency of DC versus AD was established
respectively by Solovay, Solo1978 and Kechris, Kec1995 – see also KecS1985;
cf. DalW1987; Ost1989.)

14.3.2 LM versus PB
In 1983 Raissonier and Stern (RaiS1983, Th. 2) (cf. Barto1984; Barto2010),
inspired by the then-current work of Shelah (circulating in manuscript since
1980) and earlier work of Solovay, showed that if every Σ21 set is Lebesgue-
measurable, then every Σ21 set has BP, whereas the converse fails – for the
latter see Ster1985 – cf. BartoJ1995, §9.3; Paw1985. This shows again that
measurability is in fact the stronger notion – see JudS1993b, §1 for a discussion
of the consistency of analogues at level 3 and beyond – which is one reason
why we regard category rather than measure as primary. For, the category
version of Berz’s Sublinearity Theorem (Berz1975; BinO2019b) implies its
measure version: the stronger hypothesis weakens the theorem/conclusion; see
BinO2017; BinO2018a; BinO2020c.
Note that the assumption of Gödel’s Axiom of Constructibility V = L, viewed
as a strengthening of AC, yields ∆21 non-measurable subsets, so that the Fenstad–
Normann result on the narrower class of provably ∆21 sets mentioned in §3.3
marks the limit of such results in a purely ZF framework (at level 2).

14.3.3 Consistency and the Role of Large Cardinals

While LM and PB are inconsistent with AC, such axioms can be consistent with
DC. Justification with scant exception involves some form of large-cardinal as-
sumption, which in turn, as in §16.3, calibrates relative consistency strengths –
see Kan2003; KoeW2010 (cf. Lar2010; KanM1978). Thus Solovay (Solo1970)
in 1970 was the first to show the consistency of ZF+DC+LM+PB with that of
ZFC+‘there exists an inaccessible cardinal’. The appearance of the inaccessible
in this result is not altogether incongruous, given its emergence in results (from
1930 onwards) due to Banach (Ban1932) (under GCH), Ulam (Ula1930) (un-
der AC), and Tarski (Tar1930), concerning the cardinalities of sets supporting
a countably additive/finitely additive [0, 1]-valued/{0, 1}-valued measure (cf.
§12.5, §16.3, Bog2007a, 1.12(x); Fre2008). Later, in 1984, Shelah (Shel1984,
5.1) showed in ZF+DC that already the measurability of all Σ13 sets implies that
ℵV1 is inaccessible in the sense of L (the symbol ℵV1 refers to the first uncount-
able ordinal of V , Cantor’s universe – cf. §16.2). As a consequence, Shelah
 14.3 Category Duality Revisited 233

(Shel1984, 5.1A) showed that ZF+DC+LM is equiconsistent with ZF+‘there

exists an inaccessible’, whereas (Shel1984, 7.17) ZF+DC+PB is equiconsistent
with just ZFC (i.e. without reference to inaccessible cardinals), so driving an-
other wedge between classical measure–category symmetries (see JudS1993b
for further, related ‘wedges’). The latter consistency theorem relies on the result
(Shel1984, 7.16) that any model of ZFC+CH has a generic (forcing) extension
satisfying ZF+‘every set of reals (first-order) defined using a real and an ordi-
nal parameter has BP’. (Here ‘first-order’ restricts the range of any quantifiers.)
For a topological proof, see Ster1985.

14.3.4 LM versus PB Continued

Raisonnier (Rai1984, Th. 5) (cf. Shel1984, 5.1B) has shown that in ZF+DC one
can prove that if there is an uncountable well-ordered set of reals (in particular
a subset of cardinality ℵ1 ), then there is a non-measurable set of reals. (This
motivated Judah and Spinas (JudSp1997) to consider generalizations including
the consistency of the ω1 -variant of DC.) See also Judah and Rosłanowski
(JudR1993) for a model (due to Shelah) in which ZF+DC+LM+¬PB holds,
and also Shel1985 where an inaccessible cardinal is used to show consistency
of ZF+LM+¬PB+‘there is an uncountable set without a perfect subset’. For a
textbook treatment of much of this material, see again BartoJ1995.
Raisonnier (Rai1984, Th. 3) notes the result, due to Shelah and Stern, that
there is a model for ZF+DC+PB+ℵ1 = ℵ1L + ‘the ordinally definable subsets of
reals are measurable’. So, in particular by Raisonnier’s result, there is a non-
measurable set in this model. Shelah’s result indicates that the non-measurable
set is either Σ31 (light-face symbol: all open sets coded effectively) or Σ12 (bold-
face); see the comments at the end of the introduction in Ster1985. Thus here
PB+¬LM holds.

14.3.5 Regularity of Reasonably Definable Sets

From the existence of suitably large cardinals flows a most remarkable result
due to Shelah and Woodin (ShelW1990) justifying the opening practical remark
about BP, which is that every ‘reasonably definable’ set of reals is Lebesgue-
measurable: compare the commentary in BeckK1996 following their Th. 5.3.2.
This is a latter-day sweeping generalization of a theorem due to Solovay (cf.
Solo1969) that, subject to large-cardinal assumptions, Σ12 sets are measurable
(and so also have BP by Barto1984 and RaiS1983).
234 Contrasts

14.3.6 Category and Measure: Qualitative versus

Quantitative Aspects
Most of the similarities between category and measure (Oxt1980) can now
be seen (BinO2018b; BinO2019b; BinO2020c) to flow from density topology
aspects. As Oxtoby points out (Oxt1980, p. 85), category–measure duality ex-
tends as far as qualitative aspects (0−1 laws) but not as far as quantitative aspects
(strong law of large numbers, etc.). The differences here can be dramatic. For
example, as in §14.1, the requirement on a series for it to converge almost surely
when ‘random signs’ are given to its terms is that it be ` 2 (Kah1985); by contrast
for convergence off a meagre set, the corresponding convergence criterion is
(minimally!) ` 1 (Kah2001). On occasion discrepancies can be engineered into
realignment by refining the metric – see CalMS2003.
As pointed out earlier, measurability is the stronger notion. Such distinctions
give rise to two streams of literature. In one, pathology (strange counterexam-
ples) is pursued; see, e.g., Cie1997. In the other, comparisons are made between
the various cardinal invariants associated with the σ-ideals of negligible sets;
these ask questions, relative to given axioms of set theory, such as: how small
may non-negligibles be (the non number), how small a family of negligibles has
non-negligible union (the additivity number), how small such a family must be
to cover the real line (the covering number), or how small if it is to be cofinal
under inclusion (the cof inality number). Two further key ingredients are b, the
bounding number, and d, the dominating number, corresponding to a smallest
unbounded family and a smallest dominating family of functions in ωω relative
to domination mod-finite. (For the connection between the latter and maximal
almost disjoint (mad) families of subsets of ω, see, e.g., Douw1984; for the
role of mad families in Ramsey properties of ultrafilters, see Mat1979, and for
recent developments, Tor2013.) A result on the cardinal invariants, memorable
for its symmetries, is summarized in the following Cichoń diagram, for which
we refer to BartoJ1995, and the somewhat more recent Buk2011.
cov(N ) −→ non(M) −→ cof(M) −→ cof(N )
↑ ↑
↑ b −→ d ↑
↑ ↑
add(N ) −→ add(M) −→ cov(M) −→ non(N )
Here the arrows → indicate ≤.
Cardinal invariants can also be studied for the σ-ideal of negligible sets
H N , the Haar null sets. See Chapter 16 and Bana2004.
 15

Interior-Point Theorems: Steinhaus–Weil Theory

In this long chapter we summarize and develop the work in our ‘Steinhaus–
Weil quartet’ (BinO2020c; BinO2020d; BinO2021a; BinO2022a) (unified in
the arXiv: BinO2016b).

15.1 The Steinhaus–Piccard Theorem

The fine topologies, T , we work with are all shift-invariant; that is, every shift
is an open mapping (takes T -open sets to T -open sets). Theorem 15.1.1 may
in principle be applied to any of them (e.g. the fine topology, F , of potential
theory, §8.1). In this section we will apply the result to the Euclidean and
density topologies.

Theorem 15.1.1 Let R be given a shift-invariant topology T under which it

is a Baire space, and suppose the homeomorphisms hn (x) = x + z n satisfy (cc),
whenever {z n } → 0 is a null sequence (in the Euclidean topology). For S Baire
and non-meagre in T , the difference set S − S contains an interval around the
origin.

Proof Suppose not: then for each positive integer n we may select

z n ∈ − 1 n, + 1 n \(S − S).
 

Since {z n } → 0 (in the Euclidean topology), the Category Embedding Theorem,

10.2.2, applies, and gives an s ∈ S and an infinite Ms such that {hm (s) : m ∈
Ms } ⊆ S. Then for any m ∈ Ms ,

s + z m ∈ S, i.e. z m ∈ S − S,

a contradiction. 

235
236 Interior-Point Theorems

We deduce two classical theorems: Piccard’s Theorem (Pic1939; Pic1942)

and Steinhaus’ Theorem (Stei1920). We give two proofs of each, both brief.

Theorem 15.1.2 (Piccard’s Theorem) For S Baire and non-meagre in the

Euclidean topology, the difference set S − S contains an interval around the
origin.

First Proof Apply Theorem 15.1.1 since, by Theorem 10.3.1, (cc) holds. 

Second Proof Suppose not: then, as before, for each positive integer n we may
select z n ∈ − 1 n, + 1 n \(S − S). Since z n → 0, by Theorem 10.3.3, for
 

quasi all s ∈ S there is an infinite Ms such that {s + z m : m ∈ Ms } ⊆ S. Then,
for any m ∈ Ms , s + z m ∈ S, i.e. z m ∈ S − S, a contradiction. 

Remark See BinO2010g for a derivation from here of the more general result
(due to Pet1950; Pet1951) that for S, T Baire and non-meagre in the Euclidean
topology, the difference set S − T contains an interval. See also §15.2.

Theorem 15.1.3 (Steinhaus’ Theorem) For measurable S of positive measure,

the difference set S − S contains an interval around the origin.

First Proof Arguing as in the first proof above, by Theorem 10.3.1 (cc) holds
and S, being measurable non-null, is Baire non-meagre under D (Property (6)
of Chapter 7, on page 108). 

Second Proof Arguing as in the second proof above, Theorem 4.2.1 applies.


Just as with the Pettis extension of Piccard’s result, so also here, Steinhaus
proved that, for S, T non-null measurable, S − T contains an interval. This too
may be derived from the CET (§10.2); see BinO2010b.
Unlike some of the results in earlier chapters, these results extend to topo-
logical groups. See, e.g., Com1984, Th. 4.6, p. 1175 for the positive statement,
and the closing remarks for a negative one.

15.2 The Common Basis Theorem

In this section we capture a further key similarity (their topological ‘common
basis’, adapting a term from logic) between the Baire and measure cases. Recall
(RogJ1980, p. 460) the usage in logic, whereby a set B is a basis for a class C
of sets whenever any member of C contains an element in B.
 15.2 The Common Basis Theorem 237

Theorem 15.2.1 (Common Basis Theorem; BinO2010b) For V, W Baire

non-meagre in the line R, equipped with either the Euclidean or the density
topology, there is a ∈ R such that V ∩ (W + a) contains a non-empty open set
modulo meagre sets common to both, up to translation. In fact, in both cases, up
to translation, the two sets share a Euclidean Gδ subset which is non-meagre
in the Euclidean case and non-null in the density case.
Proof In the Euclidean case for V, W Baire non-meagre we may suppose that
V = I\M0 ∪ M00 and W = J\M1 ∪ M10, where I, J are open intervals and Mi
and Mi0 are meagre. Take V0 = I\M0 and W0 = J\M1 . For v, w in V0, W0 , put
a := v − w. Thus v ∈ I ∩ (J + a). So I ∩ (J + a) differs from V ∩ (W + a)
by a meagre set. Since M0 ∪ M00 may be expanded to a meagre Fσ set M, we
deduce that I\M and J\M are non-meagre Gδ -sets.
In the density case, for V, W measurable non-null let V0 and W0 be the
sets of density points of V and W . For v, w in V0, W0, put a := v − w. Then
v ∈ T := V0 ∩ (W0 + a), and so T is non-null and v is a density point of T.
So for T0 := φ(T ) (the density points of T), T\T0 is null, so T0 differs from
V ∩ (W + a) by a null set. Evidently T0 contains a non-null closed, hence Gδ ,
subset (as T0 is measurable non-null, by regularity of Lebesgue measure). 
This leads to a strengthening of Theorem KBD, 10.3.3, which concerns two
sets rather than one.
Theorem 15.2.2 For V, W Baire non-meagre/measurable non-null in R, there
is a ∈ R such that V ∩ (W + a) is Baire non-meagre/measurable non-null and
for any null sequence z n → 0 and quasi all (almost all) t ∈ V ∩ (W + a) there
exists an infinite Mt such that
{t + z m : m ∈ Mt } ⊂ V ∩ (W + a).
Proof In either case applying the Common Basis Theorem, 15.2.1, for some
a, the set T := V ∩ (W + a) is Baire non-meagre/measurable non-null. We may
now apply Theorem KBD, 10.3.3, to the set T . Thus for almost all t ∈ T there
is an infinite Mt such that
{t + z m : m ∈ Mt } ⊂ T ⊂ V ∩ (W + a). 
This result concerning mutual inclusion of a non-negligible subset motivates
a further strengthening.
Definitions Let S be shift-compact.
1. Call T commutual with S if for every null sequence z n → 0 there is u ∈ S ∩T
and infinite Mu such that
{u + z m : m ∈ Mu } ⊂ S ∩ T .
238 Interior-Point Theorems

Thus S is commutual with T and both are shift-compact.

Call T weakly commutual with S if for every null sequence z n → 0 there is
s ∈ S and infinite Ms such that

{s + z m : m ∈ Ms } ⊂ T .

Thus T is shift precompact.

2. Call S self-commutual (up to reflected translation) if for some a ∈ R and
some T ⊂ S, S is commutual with a − T .
3. Call S weakly self-commutual (up to reflected translation) if for some a ∈ R
and some T ⊂ S, S is weakly commutual with a − T .

As an immediate corollary of Theorems 15.2.1 or 15.2.2, taking V = S,

W = −S, we may now formulate the following.

Theorem 15.2.3 (Self-Commutuality Theorem) Suppose that S is Baire

non-meagre/measurable non-null. Then S is self-commutual.

Proof Fix a null sequence z n → 0. If S is Baire non-meagre/measurable non-

null, then so is −S; thus we have for some a that T := S ∩ (a − S) is likewise
Baire non-meagre/measurable non-null and so for quasi all (almost all) t ∈ T
there is an infinite Mt such that

{t + z m : m ∈ Mt } ⊂ T ⊂ S ∩ (a − S),

as required. 

Commutuality is the additional feature needed to establish the Semi-Group

Theorem.

Theorem 15.2.4 (General Semi-Group Theorem) If S, T are shift-compact in

R with T (weakly) commutual with S, then S − T contains an interval about the
origin. Hence if S is shift-compact and (weakly) self-commutual, then S + S
contains an interval. Hence, if additionally S is a semi-group, then S contains
an infinite half-line.

Proof For S, T (weakly) commutual, we claim that S − T contains (0, δ) for

some δ > 0. Suppose not: then for each positive n there is z n with

z n ∈ (0, 1 n)\(S − T ).

Now −z n is null, so there is s in S and infinite Ms such that

{s − z m : m ∈ Mt } ⊂ T .
 15.3 Measure Subcontinuity and Interior Points 239

For any m in Mt pick t m ∈ T so that s − z m = t m ; then we have

s − zm = t m so z m = s − t m,
contradicting z m < S − T. Thus, for some δ > 0, we have (0, δ) ⊂ S − T .
For S self-commutual, say S is commutual with T := a − S, for some a; then
a + (0, δ) ⊂ a + (S − T ) = a + S − (a − S) = S + S, i.e. S + S contains an
interval. 
By the Common Basis Theorem, replacing T by −T, we obtain as an imme-
diate corollary of Theorems 15.2.3 and 15.2.4 a new proof of three classical
results: an extension of the Steinhaus and of the Piccard Theorems (cf. Theorem
15.1.2), and the Classical Semi-Group Theorem, 15.2.6.
The following result is due to Steinhaus (Stei1920) in the measure case, cf.
Bec1960, and to Pettis (Pet1950; Pet1951) in the Baire case, cf. Kom1971.

Theorem 15.2.5 (Sum–Set Theorem) If S, T are Baire non-meagre/measurable

non-null in R, then S + T contains an interval.

The following classical result is due to Hille and Phillips (HillP1957, Th.
7.3.2) (cf. BecCS1958, Th. 2; Bec1960) in the measurable case, and to Bing-
ham and Goldie (BinGo1982a; BinGo1982b) in the Baire case; see BGT1987,
Cor. 1.1.5. For a combinatorial form, see BinO2011a.

Theorem 15.2.6 (Classical Semi-Group Theorem) For an additive Baire

(measurable) semi-group S of R, the following are equivalent:

(i) S ⊇ (s, ∞) or S ⊇ (−∞, −s), for some s,

(ii) S is non-meagre (non-null).

15.3 Measure Subcontinuity and Interior Points

We begin by stating the Steinhaus–Weil Theorem in its simplest form (Stei1920;
for the line, see Wei1940, §11, p. 50; for a Polish locally compact group, see
Gros1989).

Theorem 15.3.1 (Steinhaus–Weil) In a locally compact Polish group G with

(left) Haar measure η G , for non-null Borel B, B −1 B (and likewise BB−1 )
contains a neighbourhood of the identity.

The context we work with in this and the next sections, unless otherwise
stated, is that groups and spaces are to be assumed separable. This both simpli-
fies the exposition and emphasizes that we need only the Axiom of Dependent
240 Interior-Point Theorems

Choices (DC – ‘what is needed to make induction work’ – see Chapter 16),
rather than the Axiom of Choice (AC) (again see Chapter 16); cf. BinO2019a.
For comments concerning non-separable settings, see the unified arXiv version
(BinO2016b, §8.1) of our four-part series on Theorem 15.3.1.
In this general context, the interior-point property of the measure-theoretically
‘non-negligible’ set B of the theorem is referred to as the Steinhaus–Weil prop-
erty, which encompasses the category variant due to Piccard (Pic1939; Pic1942)
and Pettis (Pet1950), cf. BinO2021a, Cor. 20 and Th. 1B (by reference, when
appropriate, to the quasi-interior of a set – the largest open set equivalent to
it modulo a meagre set). This important result has many ramifications; for ex-
ample, it is basic to the theory of regular variation – see, e.g., BGT1987, Th.
1.1.1.
The results below hinge on work of Solecki (Sol2006) on amenability at 1
and on an amendment of Fuller’s concept of subcontinuity (see below). These
are aimed at freeing up the classical dependency on local compactness and the
corresponding standard (Haar) reference measure. To the best of our knowledge
such aims for topological groups were last undertaken by Xia (Xia1972, Ch. 3),
where the emphasis is on (relative) quasi-invariance; cf. BinO2016b, §7.2, a
topic we pursued in the related paper, BinO2019c (cf. Bog1998, p. 64) with
tools developed here. For background on invariance of measures, see Kha1998.
For background on topological vector spaces and their negligible sets, see
Bog2018 or BogS2017.
For G a topological group with (admissible) metric d (briefly: metric group),
denote by M(G) the family of regular σ-finite Borel measures on G, with
P (G) ⊆ M(G) the probability measures (Kec1995, §17E; Par1967), by
Pfin (G) the larger family of finitely additive regular probability measures (cf.
Bin2010; Myc1979), and by Msub (G) submeasures (monotone, finitely subad-
ditive set functions µ with µ(∅) = 0). Here regular is taken to imply both inner
regularity (inner approximation by compact subsets, also called the Radon
property, as in Bog2007b, §7.1 and Schw1973), and outer regularity (outer ap-
proximation by open sets). We recall that a σ-finite Borel measure on a metric
space is necessarily outer regular (Bog2007b, Th. 7.1.7; Kall2002, Lemma 1.34;
cf. Par1967, Th. II.1.2 albeit for a probability measure) and, when the metric
space is complete, inner regular (Bog2007b, Th. 7.1.7; cf. Par1967, Ths. II.3.1
and 3.2). When G is locally compact we denote Haar measure by η G or just η.
For X metric, we denote by K = K (X ) the family of compact subsets of X (the
hyperspace of X, where we view it as a topological space under the Hausdorff
metric, or the Vietoris topology). For µ ∈ M(G) we write g µ(·) := µ(g·) and
µg (·) := µ(·g); by M(µ) we denote the µ-measurable sets of G and by M+ (µ)
those of finite positive measure and write K+ (µ) := K (G) ∩ M+ (µ). For G a
 15.3 Measure Subcontinuity and Interior Points 241

Polish group, recall that E ⊆ G is universally measurable, which we write in

brief as E ∈ U (G), if E is measurable with respect to every measure µ ∈ P (G).
For background, see, e.g., Kec1995, §21D; cf. Fre2002, 434D, 432; Sho1984;
these form a σ-algebra. Examples are analytic subsets (see, e.g., RogJD1980,
Part 1, §2.9; Kec1995, Th. 21.10; Fre2002, 434Dc) and the σ-algebra that they
generate. Beyond these are the provably ∆12 sets of FenN1973 (§3.3) – see §3.3,
cf. BinO2019a. The latter are sets whose ∆21 character can be proved from
the standard axiom system ZF (see Chapter 16) with the Axiom of Dependent
Choice(s) in place of the Axiom of Choice. See our survey article BinO2019a.
Recall that E is left Haar null, E ∈ H N , as in Solecki (Sol2005; Sol2006;
Sol2007) (following Chr1972; Chr1974; see also ShiT1998) if there are B ∈
U (G) covering E and µ ∈ P (G) with
µ(gB) = 0 (g ∈ G).
So if B ∈ U (G) is not left Haar null, then for each µ ∈ P (G) there is compact
K = Kµ ⊆ B and g ∈ G with

g µ(K ) > 0.
The question then arises whether there is also δ > 0 with g µ(Kt) > 0 for all
t ∈ Bδ, for Bδ = Bδ (1G ) the open δ-ball centred at 1G : a right-sided property
complementing the earlier left-sided property (of nullity, or otherwise). If this is
the case for some µ, then (see Corollary 15.3.7) 1G ∈ int(K −1 K ) ⊆ int(E −1 E);
indeed, one has
K ∩ Kt ∈ M+ (g µ) (t ∈ Bδ ) (∗M)
(‘M for measure’, cf. (∗B) below , ‘B for Baire’), which implies (Lemma 15.3.2):
Bδ ⊆ int(K −1 K ) ⊆ int(E −1 E);
cf. Kem1957; Kucz1985, Lemma 3.7.2; BinO2010f, Th. K; BinO2018b, Th.
1(iv). As this clearly forces local compactness of G (see Lemma 15.3.2), for
the more general context we weaken the ‘complementing right-sided property’
to hold only selectively: on a subset (cf. Bδ∆ (µ) on p. 244) of Bδ of the form
{z ∈ Bδ : | µ(K z) − µ(K )| < ε}.
We are guided by the close relation between the measure-theoretic Steinhaus–
Weil-like property (∗M) and its category version
K ∩ Kt ∈ B+ (τ), (∗B)
where the latter term B+ (τ) refers to non-meagre Baire sets of τ, a refinement
of the ambient topology TG = Td of G, the latter conveniently taken to be
242 Interior-Point Theorems

generated by a left-invariant metric d = d LG with associated group-norm (see

Chapter 6). We refer to the (left)-invariance of B+ (τ) (under translation) as the
(left) Nikodym property of τ.
Here, in the context of a metric or Polish group G, we study continuity proper-
ties of the maps m K : t 7→ µ(Kt) in the light of theorems of Solecki (Sol2006)
and of converses to Theorem 15.3.1 (see the Simmons–Mospan Theorem,
15.5.5) and related results. The key here is Fuller’s notion of subcontinuity, as
applied to the function m K (t) at t = 1G . This yields a fruitful interpretation of
Solecki’s notion of amenability at 1G via selective subcontinuity and linkage
to shift-compactness (see Theorem 15.3.8). Since commutative Polish groups
are amenable at 1 (Sol2006, Th. 1(ii)), this widens the field of applicability of
shift-compactness to non-Haar null subsets of these, as in BinO2017, and leads
to a conjecture (see Remarks preceding Theorem 15.3.8) as to whether H N
comprises the negligible sets of some refinement topology of Td .
We frequently refer for background to the extended commentaries and asso-
ciated extensive bibliography of BinO2016b, the unified arXiv version of our
four-part series on Theorem 15.3.1.
We begin with the promised adaptation of subcontinuity (of functions) due
to Fuller (Ful1968) (for which see Remark 4 below) to the context of measures.
See also Fort1949. We focus on the right-sided version of the concept. Subcon-
tinuity is a natural auxiliary in the quest for fuller forms of continuity: as one
instance, see Bouz1996 for the step from separate to joint continuity; as another
(classic) instance, note that a subcontinuous set-valued map with closed graph
(yet another relative of upper semi-continuity) is continuous – see HolN2012
for an extensive bibliography. Here its relevance to the Steinhaus–Weil Theo-
rem (which is relatively new – BinO2020c) yields Theorems 15.3.4 and 15.3.8,
linking amenability at 1 with shift-compactness, for which see Theorem 15.3.8.
Thus subcontinuity passes between local compactness and the pathology of in-
variance associated with non-local compactness: see Oxt1946 and DieS2014,
Ch. 10.

Definitions (BinO2018b) For µ ∈ Pfin (G), and (compact) K ∈ K (G), noting

that µ δ (K ) := inf{ µ(Kt) : t ∈ Bδ } is weakly decreasing in δ, put

µ− (K ) := sup inf{ µ(Kt) : t ∈ Bδ },

δ>0

and, for t = {t n } a null sequence, i.e. with t n → 1G ,

µt− (K ) := lim inf n→∞ µ(Kt n ).

 15.3 Measure Subcontinuity and Interior Points 243

Then
0 ≤ µ− (K ) ≤ µ(K ) = inf sup{ µ(Kt) : t ∈ Bδ }
δ>0

by BinO2020c, Prop. 1 (that t → Kt is upper semi-continuous). We say that a

null sequence t is non-trivial if t n , 1G infinitely often. Define as follows:
(i) µ is translation-continuous (‘continuous’ or ‘mobile’) if µ(K ) = µ− (K )
for all K ∈ K (G);
(ii) µ is maximally discontinuous at K ∈ K (G) if 0 = µ− (K ) < µ(K );
(iii) µ is subcontinuous if 0 < µ− (K ) ≤ µ(K ) for all K ∈ K+ (µ);
(iv) µ is (selectively) subcontinuous at K ∈ K+ (µ) along t if µt− (K ) > 0.
Remarks 1. m K (t) := µ(Kt) is continuous if µ is continuous, since m K (st) =
m K s (t) and K s is compact whenever K is compact; for directional continuity of
measures in linear spaces, see Bog2010, §3.1. In LiuR1968 (cf. LiuRW1970;
Gow1970; Gow1972), a Radon measure µ on a space X, on which a group
G acts homeomorphically, is called mobile if t 7→ µ(Kt) is continuous for all
K ∈ K (X ).
2. For G locally compact (i) holds for µ the left Haar measure η G , and also for
µ  η G (absolutely continuous with respect to η G ).
3. A measure µ singular with respect to Haar measure is maximally discontin-
uous for its support: this is at the heart of the analysis offered by Simmons (and
independently, much later by Mospan) – see Corollary 15.3.7.
4. Subcontinuity, in the sense of Ful1968, of a map f : G → (0, ∞) requires that,
for every t n → t ∈ G, there is a subsequence t m(n) with f (t m(n) ) convergent
in the range (i.e. to a positive value). The distinguished role of null sequences
emerges below in the Subcontinuity Theorem, 15.3.4. Null sequences should
be viewed here as a stepwise selection of an ‘asymptotic direction’ (or even, as
suggested by Tomasz Natkaniec, arcwise), since one may apply the Hartman–
Mycielski Embedding Theorem, 12.1.4, justifying the phrase ‘along t’ in (iv)
above, and allowing (iv) to be interpreted as a selective subcontinuity in ‘direc-
tion’ t. The analogous selective concept in a linear space is ‘along a vector’ as
in Bog2010, §3.1.
5. Selective versus uniform subcontinuity. Definition (iii) is equivalent to de-
manding, for K ∈ K+ (µ), that any null sequence t = {t n } have a subsequence
µ(Kt m(n) ) bounded away from 0; then (iii) may be viewed as demanding ‘uni-
form subcontinuity’: selective subcontinuity along each t for all K ∈ K+ (µ).
6. Left- versus right-sided versions. Writing µ̃(E) := µ(E −1 ) with E Borel in G
for the inverse measure captures versions associated with right-sided translation
such as µ̃_ and
244 Interior-Point Theorems

µ̃t− (K ) := lim inf n→∞ µ(t n K ).

Definition We will say that µ is symmetric if µ = µ̃; then B is null if and only
if B−1 is null for B a Borel set, or B ∈ U (G).

In Lemma 15.3.2 it suffices for µ to be a bounded, regular submeasure which

is supermodular:

µ(E ∪ F) ≥ µ(E) + µ(F) − µ(E ∩ F) (E, F ∈ U (G));

recall, however, from Bog2007a, 1.12.37 the opportunity to replace, for any
K ∈ K (G), a supermodular submeasure µ by a dominating µ0 ∈ Mfin (G), i.e.
with µ0 (K ) ≥ µ(K ).
For K ∈ K+ (µ) and δ, ∆ > 0, put

Bδ∆ = BδK,∆ (µ) := {z ∈ Bδ : µ(K z) > ∆},

0
which is monotonic in ∆ : Bδ∆ ⊆ Bδ∆ for 0 < ∆0 ≤ ∆. Note that 1G ∈ Bδ∆ for
0 < ∆ < µ(K ).
The specialization below to a mobile measure (see Definition (i) above) may
be found in Gow1970 and Gow1972 (cf. BinO2018b, Th. 2.5).

Lemma 15.3.2 Let µ ∈ Pfin (G) for G a metric group. For K ∈ K+ (µ), if
µt− (K ) > 0 for some non-trivial null sequence t, then, for ∆ ≥ µt− (K )/4 > 0,
there is δ > 0 with t n ∈ Bδ∆ for all large enough n and

∆ ≤ µ(K ∩ Kt) (t ∈ Bδ∆ ),

so that
K ∩ Kt ∈ M+ (µ) (t ∈ Bδ∆ ). (∗)

In particular,
K ∩ Kt , ∅ (t ∈ Bδ∆ ),

or, equivalently,
Bδ∆ ⊆ K −1 K, (∗∗)

so that Bδ∆ has compact closure. A fortiori, if µ− (K ) > 0, then δ, ∆ > 0 may
be chosen with ∆ < µ− (K ) and Bδ ⊆ Bδ∆ so that (∗) and (∗∗) hold with Bδ
replacing Bδ∆ , and, in particular, G is locally compact.

Proof For the first part fix a null sequence t and K ∈ K+ (µ) with µt− (K ) > 0;
take any ∆ ≥ µt− (K )/4 > 0, and, as above, write Bδ∆ for BδK,∆ . Then, for
µ(Kt) > 2∆ ≥ µt− (K )/2 and δ > 0 arbitrary, t ∈ Bδ∆ ; and so t n ∈ Bδ∆ for all
 15.3 Measure Subcontinuity and Interior Points 245

large enough n (since also t n ∈ Bδ for all large enough n). So Bδ∆ (K )\{1G } is
non-empty for t non-trivial.
Put Ht := K ∩ Kt ⊆ K. By outer regularity of µ, choose U = U (∆, K ) open
with K ⊆ U and µ(U) < µ(K ) + ∆. By upper semi-continuity of t 7→ Kt, we
may now fix δ = δ(∆, K ) > 0 so that K Bδ ⊆ U. For t ∈ Bδ∆, by finite additivity
of µ, since 2∆ < µ(Kt),

2∆ + µ(K ) − µ(Ht ) ≤ µ(Kt) + µ(K ) − µ(Ht ) = µ(Kt ∪ K )

≤ µ(U) ≤ µ(K ) + ∆.

Comparing the ends gives

0 < ∆ ≤ µ(Ht ) (t ∈ Bδ∆ ).

For t ∈ Bδ∆, as K ∩ Kt ∈ M+ (µ), take s ∈ K ∩ Kt , ∅; then s = kt for some

k ∈ K, so t = k −1 s ∈ K −1 K . Conversely, t ∈ Bδ∆ ⊆ K −1 K yields t = k −1 k 0 for
some k, k 0 ∈ K; then k 0 = kt ∈ K ∩ Kt.
By the compactness of K −1 K, Bδ∆ has compact closure.
As for the final assertions, if µ− (K ) > 0, now take ∆ := µ− (K )/2. Then
inf{ µ(Kt) : t ∈ Bδ } > ∆ for all small enough δ > 0, and so in particular
µ(Kt) > ∆ for t ∈ Bδ, i.e. Bδ ⊆ Bδ∆ . So the argument above applies for small
enough δ > 0 with Bδ in lieu of Bδ∆ , just as before. Here the compactness of
K −1 K now implies local compactness of G itself. 

As an immediate and useful corollary, we have

Lemma 15.3.3 Suppose µ ∈ Pfin (G), with G a metric group, t any null
sequence and K ∈ K (G). Then, if µt− (K ) > 0, there is m ∈ N with

0 < µt− (K )/4 < µ(K ∩ Kt n ) (n > m). (∗0)

In particular,
t n ∈ K −1 K (n > m). (∗∗0)

Proof Apply Lemma 15.3.2 to obtain ∆, δ > 0; for t ∈ Bδ∆ , we have µ(Kt) > ∆,
so, as above, t n ∈ Bδ∆ for all large enough n. 

This permits a connection with left Haar null sets H N introduced above.
Recall that a group G is amenable at 1 (Sol2006) (see below for the origin of
this term) if, given µ :=: { µn }n∈N ⊆ P (G) with 1G ∈ supp(µn ) (the support
of µn ), for n ∈ N there are σ and σn in P (G) with σn  µn satisfying

σn ∗ σ(K ) → σ(K ) (K ∈ K (G)).

246 Interior-Point Theorems

In view of Theorem 15.3.4, we term σ (or σ(µ) if context requires) a selective

measure and the measures σn , if needed, as associated measures (correspond-
ing to the sequence { µn }n∈N ).
Solecki explains (Sol2006, end of §2) the use of the term ‘amenability at 1’
as a localization (via the restriction that all supports contain 1G ) of a Reiter-like
condition (Pat1988, Prop. 0.4) which characterizes amenability: for µ ∈ P (G)
and ε > 0, there is ν ∈ P (G) with

|ν ∗ µ(K ) − ν(K )| < ε (K ∈ K (G)).

Lemma 15.3.2 and the next several results disaggregate Solecki’s Interior-
point Theorem (Sol2006, Th. 1(ii)) (Corollary 15.3.6), shedding more light on it
and in particular connecting it to shift-compactness (Theorem 15.3.4). Indeed,
we see that interior-point theorem itself as an ‘aggregation’ phenomenon. The-
orem 1 of BinO2020d identifies subgroups with a ‘disaggregation’ topology,
refining TG by using sets of the form BδK,∆ (σ), the measures σ being provided
in our first result, a result motivated by Solecki (Sol2006, Th. 1(ii)):

Theorem 15.3.4 (Subcontinuity Theorem) For G Polish and amenable at

1G and t a null sequence, there is σ = σ(t) ∈ P (G) such that, for each
K ∈ K+ (σ), there is a subsequence s = s(K ) := {t m(n) } with

limn σ(Kt m(n) ) = σ(K ) (n ∈ N), so σ_s (K ) > 0.

Proof For t = {t n } null, put µn := 2 n−1 P

m≥n 2
−m δ
−1 ∈ P (G); then 1G ∈
tm
supp(µn ) ⊇ {t −1
m : m > n}. By definition of amenability at 1G , in P (G) there
are σ and σn  µn, with σn ∗ σ(K ) → σ(K ) for all K ∈ K (G). For n ∈ N
choose α mn ≥ 0 with m≥n α mn = 1 (n ∈ N) and with σn := m≥n α mn δtm−1 .
P P
Fix K ∈ K+ (σ) and θ with 0 < θ < 1. As K is compact, σn ∗σ(K ) → σ(K );
then, without loss of generality,

σn ∗ σ(K ) > θσ(K ) (n ∈ N).

Then, for each n ∈ N,

X X
sup{σ(Kt m ) : m ≥ n} · α mn ≥ α mn σ(Kt m ) > θσ(K ).
m≥n m≥n

So for each n there is m = m(θ) ≥ n with

σ(Kt m ) > θσ(K ).

Now choose m(n) ≥ n inductively so that σ(Kt m(n) ) > (1−2−n )σ(K ); then, by
the upper semi-continuity of the maps t 7→ Kt and t 7→ σ(Kt) (cf. BinO2020c,
Prop. 1), limn σ(Kt m(n) ) = σ(K ) : σ is subcontinuous along s := {t m(n) }
on K . 
 15.3 Measure Subcontinuity and Interior Points 247

Remark The selection above of the subsequence s mirrors the role of ‘admis-
sible directions’ which we will encounter, in §15.7, in Cameron–Martin theory
(BinO2019c, §2; BinO2016b, §8.2).
We are now able to deduce Solecki’s interior-point theorem in a slightly
stronger form, which asserts that the sets Bδ∆ reconstruct the open sets of G
using the compact subsets of a ‘non-negligible set’, as follows. We recall that
K (X ) denotes the family of compact subsets of X; below we use the notation
δ(∆, K ) established in the proof of Lemma 15.3.2.

Theorem 15.3.5 (Aggregation Theorem) For G Polish and amenable at 1G ,

if E ∈ U (G) is not left Haar null and setting
[ ( gK,∆
Ê := Bδ(gK,∆) (σ(t)) : K ∈ K (E), 0 < σ(t)(gK )/4
∆>0, g∈G, t
)
≤ ∆ < σ(t)(gK ) ,

then
1G ∈ int( Ê) ⊆ Ê ⊆ E −1 E.

In particular, for E open, 1G ∈ int( Ê).

Proof Suppose otherwise; then, as in Lemma 15.3.2, for g ∈ G, any null

sequence z, compact K ⊆ E with 0 < σ(z)(gK )/4 ≤ ∆ and δ = δ(gK, ∆), we
have
gK,∆
Bδ (σ(z)) ⊆ (gK ) −1 gK = K −1 K ⊆ E −1 E,

so that Ê ⊆ E −1 E. Next, suppose there is for each n,

t n ∈ B1/n \ Ê.

Consider σ = σ(t). As E is not left Haar null, there is g with σ(gE) > 0.
Choose compact K ⊆ gE with σ(K ) > 0. Then, with h := g −1 and H :=
hK ⊆ E, it follows that σ(K ) = σ(gH) = σ s (gH) > 0 for some subsequence
s = {t m(k) }. So, again as above and as in Lemma 15.3.2, with ∆ := σ(gH)/4
for some δ = δ(K, ∆) > 0,
gH,∆
Bδ (σ(t)) ⊆ (gH) −1 gH = H −1 H ⊆ E −1 E.

Choose n with n > 1/δ. Then t n ∈ Bδ for all m > n; so for infinitely many k,
gH,∆
t m(k) ∈ Bδ (σ(t)) ⊆ Ê,

contradicting t n < Ê. As for the final assertion, for E open, D countable and
dense, G ⊆ d ∈D dE, so, for any µ ∈ P (G) (in particular for σ), µ(dE) > 0
S
for some d ∈ D, and so E is not left Haar null. 
248 Interior-Point Theorems

The immediate consequence from Sol2006, Th. 1(ii), is

Corollary 15.3.6 (Solecki’s Interior-Point Theorem) For G Polish and amen-

able at 1G , if E ∈ U (G) is not left Haar null, then 1G ∈ int(E −1 E).

Corollary 15.3.7 For G a Polish group, if E ∈ U (G) is not left Haar null
and is in M+ (µ) for some subcontinuous µ ∈ Pfin (G), then, for some δ > 0,

Bδ ⊆ int(E −1 E).

In particular, this inclusion holds for some δ > 0 in a locally compact group
G, for any Baire non-meagre set E.

Proof The first assertion is immediate from Lemma 15.3.2. As for the second,
for a non-meagre Baire set E, if Ẽ is the quasi-interior and K ⊆ Ẽ is compact
with non-empty interior, then η G (K ) > 0. Since η is subcontinuous, there is
δ > 0 with
Kt ∩ K , ∅ (kt k < δ),

and so
Ẽt ∩ Ẽ , ∅ (kt k < δ).

Then U := (Et)H∩ Ẽ , ∅, since (Et)H = Ẽt (the Nikodym property of the usual
topology of G). So since U is open and non-meagre, also Et ∩ E , ∅, and so
again (∗∗) holds. 

The next result establishes the embeddability by (left-sided) translation of

an appropriate subsequence of a given null sequence into a given target set
that (like-sidedly) is non-left Haar null. We establish this, first announced in
MilleO2012, in relation to the ideal H N of left Haar null sets. (It is a σ-ideal
for Polish G in the presence of amenability at 1: Sol2006, Th 1(i).) This leaves
open the ‘converse question’ of the existence of a refinement topology for which
H N is the associated notion of negligibility; this seems plausible under the
continuum hypothesis (CH) if one restricts attention only to Borel sets in H N
and their subsets by lifting a result concerning R in CieJ1995, Cor. 4.2, to G –
see also the Remark 1 following our next result.

Theorem 15.3.8 (Shift-Compactness Theorem for H N ) For G Polish and

amenable at 1G, if E ∈ U (G) is not left Haar null and z n is null, then there
are s ∈ E and an infinite M ⊆ N with

{sz m : m ∈ M} ⊆ E.

Indeed, this holds for quasi all s ∈ E, i.e. off a left Haar null set.
 15.3 Measure Subcontinuity and Interior Points 249

Proof Put t n := z n−1, which is null. With σ = σ(t) as in the Subcontinuity

Theorem, 15.3.4, since E is not left Haar null, there is g with σ(gE) > 0. For
this g, put µ := g σ. Fix a compact K0 ⊆ E with µ(K0 ) > 0 and then, passing
to a subsequence of t as necessary (by Theorem 15.3.4), we may assume that
µt− (K0 ) > 0. Choose inductively a sequence m(n) ∈ N, and decreasing compact
sets Kn ⊆ K0 ⊆ E with µ(Kn ) > 0 such that

µ(Kn ∩ Kn t m(n) ) > 0.

To check the inductive step, suppose Kn is already defined. As µ(Kn ) > 0, by the
Subcontinuity Theorem there is a subsequence s = s(Kn ) of t with µs− (Kn ) > 0.
By Lemma 15.3.2, there is k (n) > n such that µ(Kn ∩ Kn s k (n) ) > 0. Putting
t m(n) = s k (n) and Kn+1 := Kn ∩ Kn t m(n) ⊆ Kn completes the inductive step.
By compactness, select s with
\
s∈ Km ⊆ Kn+1 = Kn ∩ Kn t m(n) (n ∈ N);
m∈N

choosing k n ∈ Kn ⊆ K with s = k n t m(n) gives s ∈ K0 ⊆ E, and

sz m(n) = st −1
m(n) = k n ∈ K n ⊆ K0 ⊆ E.

Finally take M := {m(n) : n ∈ N}.

As for the final assertion, we follow the idea of the Generic Completeness
Principle (§4.1) (but with U (G) for Ba there): define
\ [
F (H) := H ∩ Ht m (H ∈ U (G));
n∈N m>n

then F : U (G) → U (G) and F is monotone (F (S) ⊆ F (T ) for S ⊆ T);

moreover, s ∈ F (H) if and only if s ∈ H and sz m ∈ H for infinitely many m.
We are to show that E0 := E\F (E) is left Haar null. Suppose otherwise. Then
renaming g and K0 as necessary, without loss of generality both µ(E0 ) > 0 and
K0 ⊆ E0 (and µ(K0 ) > 0). But then, as above, ∅ , F (K0 ) ∩ K0 ⊆ F (E) ∩ E0 ,
a contradiction, since F (E) ∩ E0 = ∅. 

Remarks 1. In the setting of Theorem 15.3.8 any non-empty open set U is

not left Haar null (as {dU : d ∈ D} with D countable dense covers G), hence
neither is U\H for H ∈ H N . So the (Hashimoto ideal) topology generated by
such sets includes H N among its negligible sets.
2. The special abelian case of Theorem 15.3.8 was independently established
by Banakh and Jabłońska in BanaJ2019. A similar result extends to the Haar-
meagre sets of Darji (Darj2013); cf. Jab2015. See also EleV2017, EleN2020
and BinO2016b, §8.9. We discuss these results in §§15.7 and 15.8.
250 Interior-Point Theorems

Corollary 15.3.9 For G Polish and amenable at 1G and z n null, there is

µ ∈ P (G) such that for K ∈ K+ (µ),
−1
K ∩ K zm ∈ M+ (µ) for infinitely many m ∈ N
if and only if for µ-quasi all s ∈ K there is an infinite M ⊆ N with
{sz m : m ∈ M} ⊆ K .
Proof We will refer to the function F of the preceding proof. First proceed as
in the proof of Theorem 15.3.4, taking t n := z n−1 and g = 1G (so that µ = σ).
Fix K with µ(K ) > 0. For the forward direction, continue as in the proof of
Theorem 15.3.4 with K0 = K and observe that the proof above needs only that
s k (n) ∈ Kn−1 Kn occurs infinitely often whenever µ(Kn ) > 0. This yields the
desired conclusion that µ(K\F (K )) = 0. For the converse direction, suppose
that µ(F (K )) > 0. Since for each n ∈ N,
[
F (K ) ⊆ K ∩ Kt m,
m>n
we have µ(K ∩ Kt m ) > 0 for some m > n; so
K ∩ Kt m ∈ M+ (µ) for infinitely many m. 
Remark With E as in the Shift-Compactness Theorem, 11.1.1, if z n ∈
B1/n \E −1 E, then z n is null; so, for some s ∈ E, sz m ∈ E for infinitely many m.
Then, for any such m,
z m ∈ E −1 E,
contradicting the choice of z m . So 1G ∈ int(E −1 E), i.e. E has the Steinhaus–
Weil property, as before.
The following sharpens a result due (for Lebesgue measure on R) to Mospan
(Mosp2005) by providing the converse below; it is antithetical to Lemma 15.3.2
(and so to Theorem 15.3.4).
Proposition 15.3.10 (Mospan Property) For G a metric group, µ ∈ Pfin (G)
and compact K ∈ K+ (µ):

(i) if 1G < int(K −1 K ), then µ− (K ) = 0, i.e. µ is maximally discontinuous;

equivalently, there is a null sequence t n → 1G with limn µ(Kt n ) = 0;
(ii) there is a null sequence t n → 1G with limn µ(Kt n ) = 0, and there is a
compact C ⊆ K with µ(K\C) = 0 with 1G < int(C −1 C).
Proof The first assertion follows from Lemma 15.3.2. For the converse, as
in Mosp2005, suppose that µ(Kt n ) = 0 for some sequence t n → 1G . By
passing to a subsequence, we may assume that µ(Kt n ) < 2−n−1 . Put Dm :=
 15.4 The Simmons–Mospan Converse 251

K\ n≥m Kt n ⊆ K; then µ(K\Dm ) ≤ n≥m µ(Kt n ) < 2−m, so µ(Dm ) > 0

T P
provided 2−m < µ(K ). Now choose compact Cm ⊆ Dm, with µ(Dm \Cm ) <
2−m . So µ(K\Cm ) < 21−m . Also Cm ∩ Cm t n = ∅, for each n ≥ m, as Cm ⊆ K;
but t n → 1G, so the compact set Cm−1 C contains no interior points. Hence, by
m
Baire’s Theorem, neither does C C, since C = m Cm, which differs from K
−1 S
by a null set. 
Proposition 15.3.11 A (regular) Borel measure µ on a locally compact metric
topological group G has the Steinhaus–Weil property if and only if either
(i) for each K ∈ K+ (µ), the map m K : t → µ(Kt) is subcontinuous at 1G ; or
(ii) for each K ∈ K+ (µ), there is no sequence t n → 1G with µ(Kt n ) → 0.
Remark This is immediate from Proposition 15.3.10 (cf. Mosp2005).

15.4 The Simmons–Mospan Converse

The converse to the Steinhaus–Weil Theorem for a locally compact group
G identifies exactly when a Borel measure µ on G guarantees that 1G is an
interior point of TT −1 for any non-µ-null Borel T. Simmons (Sim1975) showed
in 1975 that the measure µ has to be absolutely continuous with respect to
Haar measure on G. This (and other conditions – cf. Proposition 15.3.10)
were investigated independently by Mospan (Mosp2005) in 2005. The result
follows from their use of the Fubini Null Theorem, 9.5.1, and the Lebesgue
decomposition theorem (Hal1950, §32, p. 134, Th. C), but here we stress the
dependence on the Fubini Null Theorem and on left µ-inversion. We revert to
the Weil left-sided convention and associated K K −1 usage.
We begin with some definitions which stress the one-sided nature of trans-
lation in a non-commutative group and the relation to the switching of sides
under inversion.
Definitions (i) For µ ∈ M(G), say that N ⊆ G is left µ-null, N ∈ M0L (µ),
if it is contained in a universally measurable set B ⊆ G such that
µ(gB) = 0 (g ∈ G).
Thus a set S ⊆ G is left Haar null (Sol2006 after Chr1974) if it is
contained in a universally measurable set B ⊆ G that is left µ-null for
some µ ∈ M(G).
(ii) For µ ∈ M(G), say that N ∈ M0L (µ) is left invertibly µ-null, N ∈
M0L-inv (µ), if
N −1 ∈ M0L (µ).
252 Interior-Point Theorems

(iii) For µ, ν ∈ M (G), we say ν is left absolutely continuous with respect to

µ (i.e. ν <L µ) if ν(N ) = 0 for each N ∈ M0L (µ), and likewise for the
invertibility version: ν <L-inv µ.
(iv) For µ, ν ∈ M(G), we say ν is left singular with respect to µ (on B), i.e.
ν⊥L µ (on B), if B is a support of ν and B is in M0L (µ), and likewise
ν⊥L-inv µ.

We will take for granted the one-sided Lebesgue decomposition theorem:

Theorem 15.4.1 (Theorem LD) For G a Polish group, µ, ν ∈ M (G), there

are νa, νs ∈ M(G) with

ν = νa + νs with νa <L µ and νs ⊥L µ,

and likewise, there are νa0, νs0 ∈ M(G) with

ν = νa0 + νs0 with νa <L-inv µ and νs ⊥L-inv µ.

See BinO2020d for a pedestrian proof from DC, and its more rapid alternative
classical proof from AC.

Proposition 15.4.2 (Local Almost Nullity) For G a Polish group, µ ∈ M(G),

V ⊆ G open and K ∈ K (G) ∩ M0L-inv (µ), so that K, K −1 ∈ M0L (µ), we have,
for any ν ∈ M(G), ν(tK ) = 0 for µ almost all t ∈ V, and likewise ν(Kt) = 0.

Proof For ν left invertibly µ-absolutely continuous, the conclusion is imme-

diate from the definition (iii) above; for general ν this will follow from Theorem
LD, 15.4.1, above, once we have proved the corresponding singular version of
the assertion: that is the nub of the proof.
Thus, suppose that ν⊥L-inv µ on K . For t ∈ V let t = uw be any expression for
t as a group product of u, v ∈ G, and note that µ(uK −1 ) = 0, as K −1 ∈ M0L (µ).
Let H be the set
[
({t} × tK ),
t ∈V

here viewed as a union of vertical t-sections. We next express it as a union of

u-horizontal sections and apply the Fubini Null Theorem, 9.5.1.
Since u = t k = uwk is equivalent to w = k −1, the u-horizontal sections of H
may now be rewritten, eliminating t, as

{(t, u) : uw = t ∈ V, u ∈ tK = uwK } = {(uw, u) : uw ∈ V, uw ∈ uK −1 }.

So H may now be viewed as a union of u-horizontal sections as

[
(V ∩ (uK −1 ) × {u}),
u ∈G
 15.4 The Simmons–Mospan Converse 253

all of these u-horizontal sections being µ-null. By Theorem 9.5.1, µ-almost

all vertical t-sections of H for t ∈ V are ν-null. As the assumptions on K are
symmetric, the right-sided version follows. 
The result here brings to mind the Dodos Dichotomy Theorem (Dod2004a,
Th. A) for abelian Polish groups G: if an analytic set A is witnessed as Haar
null under one measure µ ∈ P (G), then either A is Haar null for quasi all
ν ∈ P (G) or else it is non-Haar null for quasi all such ν, i.e. if A ∈ M0 (µ)
(omitting the unnecessary superscript L), then either A ∈ M0 (ν) for quasi all
such ν or A < M0 (ν) for quasi all such ν (with respect to the Prokhorov–Lévy
metric in P (G); Dud1989, 11.3, cf. 9.2). Indeed, (Dod2004b, Prop. 5) when
A is σ-compact, A is Haar null for quasi all ν ∈ P (G). The result is also
reminiscent of Amb1947, Lemma 1.1.
Before stating the Simmons–Mospan specialization to the Haar context and
also to motivate one of the conditions in its subsequent generalizations, we cite
(and give a direct proof of) the following known result (equivalence of Haar
measure η and its inverse η̃), encapsulated in the formula
Z
η̃(K ) := η(K −1 ) = dη(t)/∆(t),
K
exhibiting the direct connection between η and η̃ via the (positive) modular
function ∆ (HewR1979, 15.14, or Hal1950, §60.5f). Such an equivalence holds
more generally between any two probability measures when one is left and the
other right quasi-invariant – see Xia1972, Cor. 3.1.4; this is related to a theorem
of Mackey (Mac1957, cf. BinO2016b, §8.16). As will be seen from the proof,
in Lemma 15.4.3, there is no need to assume the group is separable: a compact
metrizable subspace (being totally bounded) is separable.
Lemma 15.4.3 In a locally compact metrizable group G, for K compact, if
η(K ) = 0, then η(K −1 ) = 0, and, by regularity, so also for K measurable.
Proof Fix a compact K . As K is compact, the modular function ∆ of G is
bounded away from 0 on K, say by M > 0; furthermore, K is separable, so pick
{d n : n ∈ N} dense in K. Then, for any ε > 0, there are two (finite) sequences
m(1), . . . , m(n) ∈ N and δ(1), . . . , δ(n) > 0 such that {Bδ(i) d m(i) : i ≤ n}
covers K and
X X X
M η(Bδ(i) ) ≤ η(Bδ(i) )∆(d m(i) ) = η(Bδ(i) d m(i) ) < ε.
i ≤n i ≤n i ≤n
Then
X X
η(d m(i)
−1
Bδ(i) ) = η(Bδ(i) ) ≤ ε/M.
i ≤n i ≤n
−1 B
But {d m(i) −1 by the symmetry of the balls B (by the
δ(i) : i ≤ n} covers K δ
symmetry of the norm); so, as ε > 0 is arbitrary, η(K −1 ) = 0.
254 Interior-Point Theorems

As for the final assertion, if η(E −1 ) > 0 for some measurable E, then
η(K −1 ) > 0 for some compact K −1 ⊆ E −1 , by regularity; then η(K ) > 0, and
so η(E) > 0. 
Proposition 15.4.2 and Lemma 15.4.3 immediately give (cf. Saks1964, III.11;
Mosp2005; BartFF2018, Th. 7):
Theorem 15.4.4 For G a locally compact group with left Haar measure η
and ν a Borel measure on G, if the set S is η-null, then, for η-almost all t,
ν(tS) = 0.
In particular, this is so for S the support of a measure ν singular with respect
to η.
This in turn allows us to prove the locally compact separable case of the
Simmons–Mospan Theorem, 15.4.5 (Sim1975, Th. 1; Mosp2005, Th. 7, later
rediscovered in the abelian case in BartFF2018, Th. 10).
Theorem 15.4.5 (The Simmons–Mospan Converse) In a locally compact
Polish group, a Borel measure has the Steinhaus–Weil property if and only if it
is absolutely continuous with respect to Haar measure.
Proof If µ is absolutely continuous with respect to Haar measure η, then µ,
being invariant, is subcontinuous, and Lemma 15.3.3 gives the Steinhaus–Weil
property. Otherwise, decomposing µ into its singular and absolutely continuous
parts with respect to η, choose K a compact subset of the support of the
singular part of µ; then µ(K ) > µ− (K ) = 0, by Theorem 15.4.4, and so 15.3.10
(Converse part) applies. 
The next result is motivated by work of Simmons (Sim1975, Lemma) and
by BartFF2018, Th. 8.
Proposition 15.4.6 For G a Polish group, µ, ν ∈ M(G) and ν⊥L-inv µ con-
centrated on a compact left invertibly µ-null set K, there is a Borel B ⊆ K such
that K\B is ν-null and both BB−1 and B−1 B have empty interior.
Proof As we are concerned only with the subspace K K −1 ∪ K −1 K, without
loss of generality the group G is separable. By Proposition 15.4.2 Z := {x :
ν(xK ) = 0} is dense and so also
Z1 := {x : ν(K ∩ xK ) = 0},
since ν(K ∩ xK ) ≤ ν(xK ) = 0, so that Z ⊆ Z1 . Take a denumerable dense set
D ⊆ Z1 and put
[
S := K ∩ dK .
d ∈D
 15.5 The Steinhaus Property AA−1 versus AB−1 255

Then ν(S) = 0. Take B := K\S. If ∅ , V ⊆ BB−1 for some V open and

d ∈ D ∩ V, then, for some b1, b2 ∈ B ⊆ K,

d = b1 b−1
2 : b1 = db2 ∈ K ∩ dK ⊆ S,

a contradiction, since B is disjoint from S. So (K\S)(K\S) −1 has empty interior.

A similar argument based on
[
T := Kd ∩ K
d ∈D

ensures that also (K\S\T ) −1 (K\S\T ) has empty interior. 

Remark A non-separable generalization is pursued in BinO2020d.

15.5 The Steinhaus Property AA−1 versus AB−1

We clarify below the relation between two versions of the Steinhaus interior
points property: the simple (sometimes called ‘classical’) version concerning
sets AA−1 and the composite, more embracing one, concerning sets AB−1 ,
for sets from a given family H . The latter is connected to a strong form of
metric transitivity: Kominek (Kom1988) showed, for a general separable Baire
topological group G equipped with an inner-regular measure µ defined on some
σ-algebra M, that AB−1 has non-empty interior for all A, B ∈ M+ (µ), the sets
in M of positive µ-measure, if and only if for each countable dense set D and
each E ∈ M+ (µ) the set X\DE ∈ M0 (µ), the sets in M of µ-measure zero.
This is considered in Theorem 15.5.5. The composite property is thus related to
the Smítal property, for which see BartFN2011. Care is required when moving
to the alternative property for AB, since the family H need not be preserved
under inversion.
In general the simple property does not imply the composite: Matoŭsková
and Zelený (MatoZ2003) show that in any non-locally compact abelian Polish
group there are closed non-(left) Haar null sets A, B such that A + B has
empty interior. Jabłońska (Jab2016) has shown that likewise in any non-locally
compact abelian Polish group there are closed non-Haar meagre sets A, B such
that A + B has empty interior; see also BanaGJ2021. Bartoszewicz, Filipczak
and Filipczak (BartFF2018, Ths. 1, 4) analyze the Bernoulli product measure on
{0, 1}N with p the probability of the digit 1; see BinO2011a, §8.15. The product
space may be regarded as comprising canonical binary digit expansions of the
additive reals modulo 1 (in which case the measure is not invariant). Here the
(Borel) set A of binary expansions with asymptotic frequency p of the digit
1 has [0, 1) as its difference set if and only if 41 ≤ p ≤ 34 ; however, A + A
256 Interior-Point Theorems

has empty interior unless p = 12 (the base 2 simple-normal numbers case; cf.
§14.1).
Below we identify some conditions on a family of sets A with the simple AA−1
property which do imply the AB−1 property. What follows is a generalization
to a group context of relevant observations from BinO2020d from the classical
context of R. For a similar approach, see Kha2019.
The motivation for the definition below is that its subject, the space H, is
a subgroup of a topological group G from which it inherits a (necessarily)
translation-invariant (either-sidedly) topology τ. Various notions of ‘density at
a point’ give rise to ‘density topologies’ (BinO2018b), which are translation-
invariant since they may be obtained via translation to a fixed reference point:
early examples, which originate in spirit with Denjoy as interpreted by Haupt
and Pauc (HauP1952), were studied intensively in GofW1961, GofNN1961,
soon followed by Marti1961, Marti1964 and Mue1965; more recent exam-
ples include FilW2011 and others investigated by the Wilczyński school, cf.
Wil2002.
Proposition 15.5.1 embraces as an immediate corollary the case H = G
with G locally compact and σ the Haar density topology (see BinO2019c).
Proposition 15.5.4 proves that Proposition 15.5.1 applies also to the ideal
topology (in the sense of LukMZ1986) generated from the ideal of Haar null
sets of an abelian Polish group.
We recall that a group H carries a left semi-topological structure τ if the
topology τ is left invariant (ArhT2008) (hU ∈ τ if and only if U ∈ τ);
the structure is semi-topological if it is also right invariant, i.e. briefly, τ is
translation invariant. The group H is quasi-topological under τ if τ is both left
and right invariant and inversion is τ-continuous.
Definition For H a group with a translation-invariant topology τ, call a
topology σ ⊇ τ a Steinhaus refinement if:
(i) intτ ( AA−1 ) , ∅ for each non-empty A ∈ σ, and
(ii) σ is involutive-translation invariant: h A−1 ∈ σ for all A ∈ σ and all h ∈ H.
Property (ii) above (called simply ‘invariance’ in BartFN2011) apparently
calls for only left invariance, but in fact, via double inversion, delivers transla-
tion invariance, since U h = (h−1U −1 ) −1 ; then H under σ is a semi-topological
group with a continuous inverse, so a quasi-topological group. We address the
step from the simple property to the composite in the next result.
Proposition 15.5.1 If τ is translation-invariant, and σ ⊇ τ is a Steinhaus re-
finement topology, then intτ ( AB−1 ) , ∅ for non-empty A, B ∈ σ. In particular,
as σ is preserved under inversion, also intτ ( AB) , ∅ for A, B ∈ σ.
 15.5 The Steinhaus Property AA−1 versus AB−1 257

Proof Suppose A, B ∈ σ are non-empty; as B−1 ∈ σ, choose a ∈ A and

b ∈ B; then, by (ii),

1 H ∈ C := a−1 A ∩ b−1 B−1 ∈ σ.

By (i), for some non-empty W ∈ τ,

W ⊆ CC −1 = (a−1 A ∩ b−1 B) · ( A−1 a ∩ B−1 b) ⊆ (a−1 A) · (B−1 b).

As τ is translation invariant, aW b−1 ∈ τ and

aW b−1 ⊆ AB−1,

the latter since, for each w ∈ W , there are x ∈ A, y ∈ B−1 with

w = a−1 x. yb : awb−1 = xy ∈ AB−1 .

So intτ ( AB−1 ) , ∅. 

Corollary 15.5.2 In a locally compact group the Haar density topology is a

Steinhaus refinement.

Proof Property (i) follows from Weil’s Theorem since density-open sets are
non-null measurable; left translation invariance in (ii) follows from left invari-
ance of Haar measure, while involutive invariance holds, as any measurable set
of positive Haar measure has non-null inverse by Lemma 15.4.3 (HewR1979,
15.14; cf. BinO2020d, §2, Lemma H). 

A weaker version, inspired by metric transitivity, comes from applying the

following concept.

Definition Say that a group H acts transitively on a family H ⊆ ℘(G) if, for
each A, B ∈ H , there is h ∈ H with A ∩ hB ∈ H .

Thus a locally compact topological group acts transitively on its non-null

Haar measurable subsets (in fact, either-sidedly); this follows from Fubini’s
Theorem (Hal1950, 36C), via the average theorem (Hal1950, 59.F):
Z
|g −1 A ∩ B| dg = | A| · |B−1 | ( A, B ∈ M)
G

(note that g = ab−1 if and only if g −1 a = b), cf. TomW2016, §11.3 after
Th. 11.17.
In MatoZ2003 it is shown that in any non-locally compact abelian Polish
group G there exist two non-Haar null sets, A, B < H N , such that A∩hB ∈ H N
for all h; that is, G there does not act transitively on the non-Haar null sets.
258 Interior-Point Theorems

Definition In a quasi-topological group (H, τ), say that a proper σ-ideal

H has the simple Steinhaus property AA−1 if AA−1 has interior points for
universally measurable subsets A < H (cf. BartFN2011).

Proposition 15.5.3 In a group (H, τ) with τ translation-invariant, if H acts

transitively on a family of subsets H with the simple Steinhaus property, then
H has the (composite) Steinhaus property:
intτ ( AB−1 ) , ∅ for A, B ∈ H .
Furthermore, if H is preserved under inversion, then also
intτ ( AB) , ∅ for A, B ∈ H .

Proof For A, B ∈ H choose h with C := A ∩ hB ∈ H ; then

CC −1 h = ( A ∩ hB)( A−1 ∩ B−1 h−1 ) ⊆ AB−1 . 

Proposition 15.5.4 If (H, τ) is a quasi-topological group (i.e. τ is invari-

ant with continuous inversion) carrying a left-invariant σ-ideal H with the
Steinhaus property and τ ∩ H = {∅}, then the ideal topology σ with basis
B := {U\N : U ∈ τ, N ∈ H }
is a Steinhaus refinement of τ.
In particular, for (H, τ) an abelian Polish group, the ideal topology generated
by its σ-ideal of Haar null subsets is a Steinhaus refinement.

Proof If U, V ∈ B and w ∈ U ∩ V, choose M, N ∈ H and W M , W N ∈ τ such

that x ∈ (W M \M) ⊆ U and x ∈ (W N \N ) ⊆ V . Then, as M ∪ N ∈ H ,
x ∈ (W M ∩ W N )\(M ∪ N ) ∈ B.
So B generates a topology σ refining τ. With the same notation, hU =
hW M \hM ∈ σ, as hM ∈ H, and U −1 = W M −1 \M −1 . Finally, UU −1 has non-

empty τ-interior, as U < H and is non-empty.

As for the final assertion concerned with an abelian Polish group context,
note that if N is Haar null (i.e. N ∈ H N ), then µ(hN ) = 0 for some µ ∈ P (G)
and all h ∈ H, so hN ∈ H N for all h ∈ H. Furthermore, if A < H N ,
then A−1 < H N : for otherwise, µ(h A−1 ) = 0 for some µ ∈ P (G) and all
h ∈ H; then, taking µ̃(B) = µ(B−1 ) for Borel B, we have µ̃( A) = 0 and
µ̃(h A) = µ( A−1 h−1 ) = 0 for all h ∈ H, a contradiction. 
Remark A left Haar null set need not be right Haar null: for one example,
see ShiT1998, and for more general non-coincidence, see Sol2005, Cor. 6; cf.
Dod2009. So the argument in Proposition 15.5.4 does not extend to the family
 15.6 Borell’s Interior-Point Property 259

of left Haar null sets H N of a non-commutative Polish group. Indeed, Solecki

(Sol2006, Th. 1.4) showed in the context of a countable product of countable
groups that the simpler Steinhaus property holds for H N amb (involving simul-
taneous left- and right-sided translation – see BinO2021a, §1) if and only if
H N amb = H N .
Next, we reproduce a result from Kom1988, Th. 5. Recall that µ is quasi-
invariant if µ-nullity is translation invariant. The transitivity assumption (of
co-nullity) is motivated by Smítal’s Lemma, which refers to a countable dense
set – see KuczS1976.

Theorem 15.5.5 (Kominek’s Theorem) If µ ∈ P (G) is quasi-invariant and

there exists a countable subset H ⊆ G with H M co-null for all M ∈ M+ (µ),
then int( AB−1 ) , ∅ for all A, B ∈ M+ (µ).

Proof By regularity, we may assume A, B ∈ M+ (µ) are compact, so AB−1 is

compact. Fix g ∈ G; then, by quasi-invariance, µ(gB) > 0. So by the transitivity
assumption, both G\HgB and G\H A are null, and so H A ∩ HgB , ∅. Say
h1 a = h2 gb, for some a ∈ A, b ∈ B, h1, h2 ∈ H; then g = h2−1 h1 ab−1 . As g
was arbitrary,
[
G= h2−1 h1 AB−1 .
h1,h2 ∈H

By Baire’s Theorem, as H is countable, int( AB−1 ) , ∅. 

15.6 Borell’s Interior-Point Property

For completeness of this overview of the Steinhaus–Weil interior-point prop-
erty, we offer in brief here the context and statement of a (by now) classical
Steinhaus-like result in probability theory; this differs in that the Polish group
now specializes to an infinite-dimensional topological vector space and the
reference measure is Gaussian, so no longer invariant. We refer to the related
paper, BinO2019c, for further details and background literature, and to our
generalizations to Polish groups and other reference measures.
For X a locally convex topological vector space, γ a probability measure on
the σ-algebra of the cylinder sets generated by the dual space X ∗ (equivalently,
for X separable Fréchet, e.g. separable Banach, the Borel sets), with X ∗ ⊆
L 2 (γ): then γ is called Gaussian on X (‘gamma for Gaussian’, following
Bog1998) if and only if γ ◦ ` −1 defined by

γ ◦ ` −1 (B) = γ(` −1 (B)) for all Borel B ⊆ R

260 Interior-Point Theorems

is Gaussian (normal) on R for every ` ∈ X ∗ ⊆ L 2 (γ). For a monograph treat-

ment of Gaussianity in a Hilbert-space setting, see Jan1997. Write γh (K ) :=
γ(K + h) for the translate by h. Recall that relative quasi-invariance of γh and
γ means that, for all compact K,
γh (K ) > 0 if and only if γ(K ) > 0.
This property holds relative to a set of vectors h ∈ X (termed the admissible
directions) forming a vector subspace known as the Cameron–Martin space,
H (γ). Then γh and γ are equivalent, written γ ∼ γh, if and only if h ∈ H (γ).
Indeed, if γ ∼ γh fails, then the two measures are mutually singular, i.e. γh ⊥γ
(the Hajek–Feldman Theorem – cf. Bog1998, Th. 2.4.5, 2.7.2).
Continuing with the assumption above on X ∗, as X ⊆ X ∗∗ ⊆ L 2 (γ), one can
equip H = H (γ) with a norm derived from that on L 2 (γ). In brief, this is done
with reference to a natural covariance under γ obtained by regarding f ∈ X ∗
as a random variable and working with its zero-mean version, f − γ( f ); then,
γ
for h ∈ H, we can represent δh, i.e. the (shifted) evaluation map defined by
γ
δh ( f ) := f (h) − γ( f ) for f ∈ X ∗, as h f − γ( f ), ĥi L 2 (γ) for some ĥ ∈ L 2 (γ).
γ
(Here, for γ symmetric γ( f ) = 0, so δh = δh is the Dirac measure at h.) This is
followed by identifying h with ĥ (for h ∈ H), and noting that |h| H := k ĥk L 2 (γ)
is a norm on H arising from the inner product
Z
(h, k) H := ĥ(x) k̂ (x) dγ(x).
X
γ
Formally, the construction first requires an extension of the domain of δh to
Xγ∗, the closed span of {x ∗ − γ(x ∗ ) : x ∗ ∈ X ∗ } in L 2 (γ), a Hilbert subspace in
which to apply the Riesz Representation Theorem.
We may now state the Steinhaus-like property due, essentially in this form, to
Christer Borell (Borel1976). (Proposition 1 of LeP1973 offers a weaker, ‘one-
dimensional section’ form with the origin an interior point of the difference set
relative to each line of H passing through it; we may call it the H-radial form
by analogy with the Q-radial form (Kucz1985, §10.1) of Euclidean spaces: the
rational points are indeed an additive subgroup. The alternative term ‘algebraic
interior point’ is also in use, e.g. in the literature of functional equations – cf.
Brz1992.)
The following result is due to Borell (Borel1976, Cor. 4.1) – see Bog1998,
p. 64.
Theorem 15.6.1 (Borell’s Interior-Point Theorem) For γ a Gaussian measure
on a locally convex topological space X with X ∗ ⊆ L 2 (γ), and A any non-null
γ-measurable subset A of X, the difference set A − A contains a |. | H -open
neighbourhood of 0 in the Cameron–Martin space H = H (γ), above. That is,
( A − A) ∩ H contains an H-open neighbourhood of 0.
 15.7 Haar-Meagre Sets in an Abelian Polish Group 261

This follows from the continuity in h of the density of γh with respect to γ

(Bog1998, Cor. 2.4.3), as given in the Cameron–Martin–Girsanov formula:
!
1
exp ĥ(x) − k ĥk L2 2 (γ) , (CM)
2

where ĥ ‘Riesz-represents’ h, i.e. x ∗ (h) = hx ∗, ĥi, for x ∗ ∈ X ∗ , as above. Thus

here a modified Steinhaus Theorem holds: the relative interior-point theorem.

15.7 Haar-Meagre Sets in an Abelian Polish Group

We begin with Darji’s definition, in an abelian Polish group X, of the Haar-
meagre sets, which we denote H M X , and his theorem that these sets form
a σ-ideal and that they are meagre. Both results rely on the Kuratowski–
Ulam Theorem. We then proceed to give Jabłońska’s proof of a Steinhaus–
Weil–Piccard theorem for non-Haar-meagre sets and an open homomorphism
theorem for ‘Darji-measurable’ functions. (A set is Darji-measurable if it is
the union of a Borel set and a Haar-meagre set.)

Definition (Darj2013) In a normed group X, say that A ⊆ X is Haar meagre

if, for some Borel set B with A ⊆ B ⊆ X, there is a continuous function
f : K → B with K compact metric such that f −1 (B + x) is meagre for each
x ∈ X . Such a function f will be termed a witnessing function.

Remarks 1. Evidently for K compact and f a witnessing function f (K ) is

compact and this covers B ∩ (B + x).
2. For X a locally compact group and M meagre, suppose, without loss of
generality, M to be a meagre Fσ . Take any U open in X with compact closure
and put K := Ū. As (x + M) is meagre for any x, as also is K ∩ (x + M), we
may take f = idK (with idK (x) ≡ x for x ∈ K) to witness that any meagre M in
X is Haar meagre. Thus Darji’s definition extends to the non-locally compact
context an analogue of Haar null sets.
3. In a non-locally compact Polish group compact sets (and, by implication,
σ-compact sets) are Haar meagre. Indeed, for such a set K, here again idK is a
witnessing function.
4. For a witnessing function f : K → B as above, any k ∈ K and any norm
open ball U in X with f (k) ∈ U, take H := f −1 (U). Then H is non-meagre
in K, with k ∈ H, and the range of f on H is in the closed ball. Consider the
translate g : H → B with

g(h) := f (h) − f (k), for h ∈ H;

262 Interior-Point Theorems

then g −1 (B + x) = H ∩ f −1 (B − f (k) + x) is meagre for any x. Thus the range of

the witnessing function can be contained in an arbitrarily small ball around 0X .

Theorem 15.7.1 (Darj2013) The family of Haar-meagre sets in X, H M, is

a σ-ideal.

Proof It suffices to consider a sequence of Haar meagre Borel sets An and

associated witnessing functions f n : Kn → An with the range of f n lying in the
ball of radius 2−n around 0X . Define f : n Kn → X by
Q
X
f ({k n }n ) = f n (k n ),
n

which is continuous. Since, for A = j A j ,

S
[
f −1 ( A + x) = f −1 ( A j + x),
j

it suffices to show that each set f −1 ( A j + x) is meagre. To this end, fix j and
view n Kn as the product of K j and n,j Kn . For any fixed {k n }n,j , the
Q Q
following set is meagre in K j :
( X )  X 
a j ∈ K j : f j (a j ) + f n (k n ) ∈ A j + x = f j−1 A j + x − f n (k n ) .
n,j n,j

But the set f −1 ( A j + x) is Borel (as A j is Borel and f is continuous), so has

the Baire property, and
( X )
f −1 ( A j + x) = {k n } : f n (k n ) ∈ A j + x .
n

We may thus apply the Kuratowski–Ulam Theorem, 9.5.2, to conclude that

f −1 ( A j + x) is meagre, as claimed. 

Theorem 15.7.2 (Darj2013) Haar-meagre sets are meagre: H M ⊆ M.

Proof For Borel A ∈ H M and f : K → A a witnessing function, put

Σ A := {(x, y) ∈ X × K : f (y) ∈ ( A − x)}.

As A is Borel, Σ A is also Borel and so has the Baire property. Now, for each
x ∈ X,
{y ∈ K : (x, y) ∈ Σ A } = f −1 ( A − x),

so is meagre in K by assumption. Thus, by the Kuratowski–Ulam Theorem, Σ A

is meagre in X × K. Applying the Kuratowski–Ulam Theorem again, but now
the other way about, there is, in particular, y ∈ K such that

B = {x : (x, y) ∈ Σ A }
 15.7 Haar-Meagre Sets in an Abelian Polish Group 263

is meagre in X. But
B = {x : f (y) ∈ ( A − x)} = {x : x ∈ A − f (y)} = A − f (y).
That is, A − f (y) is meagre in X. However, in a Polish group, meagre sets are
translation invariant; so A is meagre in X. 
Theorem 15.7.3 (Jab2015) In an abelian Polish group, for Borel A < H M X ,
0 ∈ int( A − A).
Proof Suppose otherwise. Then, for some A < H M X , the origin 0 is a
member but not an interior point of the set
∆( A) := {x ∈ X : A ∩ ( A + x) < H M X } ⊆ A − A;
indeed, x ∈ A − A if and only if A ∩ ( A + x) , ∅. By the Birkhoff–Kakutani
Theorem, 6.1.5, X may be equipped with a group-norm k. k. So we may choose
z n → 0 with kz n kX ≤ 2−n and z n < ∆( A). Take
[
A0 = A\ A ∩ ( A + z n ).
n
Then A0 is not Haar meagre, since A ∩ ( A + z n ) ∈ H M X for each n. Give the
Cantor group K = {0, 1}N the group-norm:
X
kk k = 2−n k n,
n
so that ||en || = 2−n for the natural basis vectors, and define
X
gz (k) = k · z = kn zn .
n
Then
X X X
||gz (k)|| X ≤ kk n z n k = k n kz n k ≤ k n 2−n = kk k.
n n n

Thus g : K → X is continuous and so there exists y with g −1 ( A0 + y) not

meagre in K . By the Piccard–Pettis Theorem, 9.2.1, for some δ > 0,
B0K (δ) ⊆ g −1 ( A0 + y) −K g −1 ( A0 + y).
Choose n with ken k < δ; then, for some a, b ∈ g −1 ( A0 + y),
en = a −K b.
Thus a = b +K en ; according as bn = 0 or 1, working in X:
gz (a) − gz (b) = (1 + bn )z n − bn z n = ±z n .
That is, ±z n ∈ ( A0 + y) − ( A0 + y) = A0 − A0, a contradiction, since z n + a0 =
a00 ∈ A0 for some a0, a00 ∈ A0 . Yet A0 ∩ ( A0 + z n ) is disjoint from A0 . 
264 Interior-Point Theorems

We may now apply the preceding theorem and the fact that any abelian Polish
group is not Haar meagre to establish the next theorem, after the following
definition.

Definition Call a set D-measurable if and only if it is the union of a Borel set
and a Haar-meagre set. (‘D for Darji’.)

Since the Borel sets form a σ-algebra and the Haar-meagre sets are a σ-
ideal, the D-measurable sets also form a σ-algebra. It is thus natural to speak of
D-measurable maps, when preimages of D-measurable sets are D-measurable.

Theorem 15.7.4 (Jab2015) Any D-measurable homomorphism f : X → Y

between abelian Polish groups X and Y is continuous.

Proof We can work entirely in the closed image Z of f , as this as a subspace

will also be a Polish group. To prove continuity at 0X , consider any open
neighbourhood U = B Z (ε) of 0Y . Taking V = B Z (ε/2), we have V − V ⊆ U,
and so, by symmetry of the implied metric,

Z = f (X ) ⊆ f (X ) + V .

By separability of Z, there are points x n ∈ X with

[
f (X ) ⊆ Z ⊆ f (x n ) + V :
n
[
X ⊆ f −1 (V ) + x n ;
n

this last holds as f is a homomorphism. As X is not Haar meagre, neither is

f −1 (V ) + x n for some n (as H M X is a σ-ideal), and so neither is f −1 (V ). As
f is D-measurable, f −1 (V ) = B ∪ M for some Borel B and some M ∈ H MX .
Since B is not Haar meagre in X, by the preceding theorem, for some δ > 0,

B X (δ) ⊆ B − B ⊆ f −1 (V ) − f −1 (V ) ⊆ f −1 (V − V ) ⊆ U.

This proves continuity. 

Remark This result will also follow from the fact that non-Haar-meagre sets
in an abelian Polish group are shift-compact, our next result. Thus there are
meagre sets which, by virtue of being not Haar meagre (such sets exist in
non-locally compact groups), are shift-compact.
The following theorem is enuciated for a Borel set A; however, the proof
works just as well for any set that is universally Baire.

Theorem 15.7.5 (BanaJ2019) In an abelian Polish group, any Borel A <

H M X is shift-compact.
 15.8 Haar-Null Sets Revisited (the Abelian Case) 265

Proof Suppose otherwise. Then there is a null-sequence {z n }n such that for

any x ∈ X the set {n ∈ N : x + z n ∈ A} is finite. Passing to a subsequence, if
necessary, we can assume that kz n k < 2−n for all n. Taking

Ki := {0} ∪ {z m }m≥i,
Q
which is compact (in X ), define on K := n∈N Kn the function
X
g({x n }) := x n,
n∈N

which is continuous. Since the set A is not Haar meagre and is Borel, there
exists a point x ∈ X such that

B := g −1 (x + A)

is non-meagre and Baire in K (by continuity of g). Hence its (non-empty)

quasi-interior U ⊂ K has U ∩ B co-meagre in U. Passing to a basic subset of
U, we may, without loss of generality, assume that U := {b} × n≥J Kn, for
Q
Q
some J ∈ N and some b ∈ n<J Kn . For m ≥ J, each constituent subset of U
Q
of the form {b} × {z m } × n>J Kn is open in K, and so
( Y )
Cm := y ∈ Kn : {b} × {z m } × {y} ⊂ U ∩ B
n>J

is co-meagre in n>J Kn . Hence m≥J Cm is co-meagre in n>J Kn , and so

Q T Q
contains at least one point y for which we have

{b} × (K j \{0}) × {y} ⊂ B.

Fix such a point y. For v with {v} = {b} × {0} × {y} and m ≥ J, we have, by
considering {b} × {z m } × {y}, that

g(v) + z m ∈ g(B) = x + A.

Put u := −x + g(v); then u + z m ∈ A for all m ≥ J. So {n ∈ N : u + z n ∈ A} is

infinite, contradicting the choice of (z n )n∈N . 

15.8 Haar-Null Sets Revisited (the Abelian Case)

We return to the result in Theorem 15.3.8 where we showed that, in a Polish
group which is amenable at 1, a set that is not (left) Haar null (in the sense of
Christensen and Solecki) is shift-compact. When the group is abelian this can
be proved by a method that is highly reminiscent of a similar result concerning
sets that are not Haar meagre. The argument here again comes from BanaJ2019.
266 Interior-Point Theorems

Theorem 15.8.1 In an abelian Polish group any non-Haar null universally

measurable set is shift-compact.
Proof (BanaJ2019) Suppose otherwise, and suppose too that A is universally
measurable and not Haar null in X, but that there is a null-sequence (z n )n∈N
such that for each x ∈ X the set {n ∈ N : x + z n ∈ A} is finite. Passing to a
subsequence, without loss of generality we may assume that ||ki zi || ≤ 2−i for
n ∈ N. Now consider the compact product space of finite discrete spaces
Y
Π := {0, 1, . . . , 2n },
n∈N
endowed with the product measure λ of uniform measures. On Π define the
map
X∞
g(k) = k i zi,
i=1

where we note that ki are integers. Since kk i zi k ≤ 2i, as kzi k ≤ 4−i , the series
is convergent and the function g is well defined and continuous.
Since A is not Haar null, there exists x ∈ X such that g −1 (x + A) has positive
λ-measure, and so there is a compact K ⊆ g −1 (x + A) of positive measure. For
every n ∈ N consider the subcube
Yn−1 Y∞
Πn := {0, 1, 2, . . . , 2i } × {1, 2, . . . , 2i } ⊆ Π
i=1 i=n
and observe that λ(Πn ) → 1 as n → ∞. Replacing K by K ∩Π` for a sufficiently
large `, we can assume that K ⊂ Π` . For m ≥ ` define the ‘back-tracking’ map
from Πn to Π by
yi ,
(
i , m,
s m (y) =
yi − 1, i = m.
Claim For any compact set C ⊂ Π` with λ(C) > 0 and ε > 0,
λ(C ∩ s m (C)) > (1 − ε)λ(C)
for all large enough m > `.
Proof of Claim By the regularity of the measure λ, the set C has a neighbour-
hood O(C) ⊂ Π such that λ(O(C)\C) < ε · λ(C). For each c ∈ C there is
m(c) such that s m (c) ∈ O(C). By the compactness of C, there exists k ≥ `
such that, for any m ≥ k, the image s m (C) ⊆ O(C). Hence
λ(s m (C)\C) ≤ λ(O(C)\C) < ε · λ(C)
and so
λ(s m (C) ∩ C) = λ(s m (C)) − λ(s m (C)\C) > λ(C) − ε · λ(C) = (1 − ε)λ(C),
and the claim is proved.
 15.8 Haar-Null Sets Revisited (the Abelian Case) 267

Returning to the proof of the theorem, using the claim we can choose an
increasing sequence of integers (m(k))k ∈N with m0 > ` such that the set
\
K∞ := s m(k) (K )
k ∈N
has positive measure and so contains a point b := (b(i))i ∈N . It follows that, for
m(k) (b) ∈ K ⊂ g (x + A). Now bk (m(k)) = b(m(k)) + 1
each k ∈ N, bk := s−1 −1

and so
X
g(b) + z m(k) = b(n)z n + (b(m(k)) + 1)z m(k) = g(bk ) ∈ x + A.
n,m(k)

So −x + g(b) + z m(k) ∈ A for all k ∈ N, which contradicts the choice of the

sequence (z n )n∈N . 
 16

Axiomatics of Set Theory

16.1 The Three Elephants in the Room

We summarize here briefly some of the contents of our 2019 survey, ‘Set
Theory and the Analyst’ (BinO2019a). We begin with the three ‘elephants in
the room’. (An ‘elephant in the room’ is something which is all too obviously
there, but which no one wants to mention.)
The Canonical Status of the Reals
The first of the three elephants in BinO2019a, p. 5, concerns the canonical status
of the reals R. The rationals, Q, are canonical from any point of view. The reals
are canonical (so justifying the use of the definite article here) from many points
of view, including uniqueness as a complete Archimedean ordered field up to
isomorphism, but not including cardinality. The Continuum Hypothesis (CH)

2ℵ0 = ℵ1 (CH)

(see, e.g., Kec1995, 16D; the notation on the left derives from binary expansion
of the reals), that the cardinality of R is the smallest uncountable cardinal, is and
will always remain just that, a hypothesis. For, as Cohen (Coh1963) showed,
both CH and its negation are consistent with the ‘ordinary rules of mathematics’,
Zermelo–Fraenkel set theory (ZF).
So one can choose which cardinality of the reals one wishes to work with
(indeed, ‘which real line one wishes to work with’). As Solovay (Solo1965)
puts it, ‘it (the cardinality of the reals) can be anything it ought to be’. The only
constraint is that its cofinality be uncountable, by a result of König of 1905 –
see, e.g., Kun1983, §10.40; Kun2011, I.13.12.
If one drops the Archimedean requirement, other possibilities emerge. For
such ‘super-real’ fields, see DalW1996.

268
 16.2 Hypotheses and Axioms 269

Impossibility of Proving Consistency of Set Theory

The third elephant (BinO2019a, p. 7) was disposed of in 1931 by Gödel’s
incompleteness theorems (God1940): first, the existence of sentences that can
be neither proved nor disproved; second, the impossibility of a rich enough
axiom system being able to provide a proof of its own consistency (rich enough
to accommodate arithmetic). 1 Thus, neither ZF nor ZFC (ZF plus the Axiom
of Choice, AC) can be proved consistent in ZF, resp. ZFC. These apparently
disturbing results are facts of life, with which one must live in mathematics.
Which Sets of Reals Can We Use?
There remains the second elephant (BinO2019a, p. 6): which sets of reals
are available? Here (perhaps not surprisingly in view of the above) the answer
depends crucially on what axioms of set theory one assumes, hence this chapter.
At one extreme, one can assume AC and work with ZFC, giving a maximal
supply of sets to work with. A first advantage of this choice is that it yields the
Vitali example of a non-measurable set (see, e.g., Rud1966, §2.22; SteS2005,
pp. 24–25), albeit non-constructively.
At the other extreme, one could assume, instead of AC, the axiom of Depen-
dent Choice(s) (DC). (This is what is needed to make mathematical induction
work, a minimal requirement for a useful axiom system in mathematics.) One
could augment this with LM (the axiom that all sets are Lebesgue-measurable)
and/or PB (that all sets have the property of Baire).
Intermediate between these is PD – that all sets in the projective hierarchy
(§3.2) are determined (§2.4). This determinacy is in the sense that in the relevant
Banach–Mazur game, one player has a winning strategy. See also BinO2019a,
§7.2. For the connection between large cardinals (below in §16.3) and PD, see
JudSp1997.

16.2 Hypotheses and Axioms

Generalized Continuum Hypothesis
Much stronger than CH is the generalized continuum hypothesis (GCH): writing
κ + for the successor cardinal of an infinite cardinal κ,
2κ = κ + for every infinite cardinal κ. (GCH)
Gödel proved (God1940) that GCH is consistent with ZF. Foreman and Woodin
(ForW1991) showed that ‘the generalized continuum hypothesis can fail every-
where’.
1 There are consistent systems capable of proving their own consistency, but with restricted
arithmetic power, see, e.g., Hilbert’s axiomatization of arithmetic, denoted A0 in Kneebone
(Kne1963).
270 Axiomatics of Set Theory

Gödel’s Axiom of Constructibility, V = L

Here (see, e.g., BinO2019a, §2) one may begin with von Neumann’s 1923
definition of the natural numbers N starting naturally with 0:

0 = ∅; 1 = {0}; 2 := {0, 1}; 3 := {0, 1, 2}; ··· ;

n + 1 : = n ∪ {n} := n ∪ {0, 1, . . . , n − 1}.

The von Neumann scheme can be naturally extended: once the class of ordinals
α ∈ On (below) is established, it yields the cumulative hierarchy Vα , introduced
inductively by
[
Vα+1 := ℘(Vα ), Vλ := {Vα : α < λ},

with ℘(Vα ) the power set of Vα , where λ is a limit ordinal. The ‘class of all
sets’ is then
[
V := {Vα : α ∈ On},

known as the von Neumann universe, which one may view as the ‘Cantor
universe’. (Here we recall the dual framework of formal set theory which dis-
tinguishes sets admitted by the axioms and classes defined by arbitrary ‘prop-
erties’, expressed in the formal language of set theory; this is the established
formalization aimed at avoiding Russell’s paradox. Classes may fail to be sets.
Thus an ordinal is a set well ordered, transitively, by membership; however,
since the class of ordinals is well ordered by membership, it fails to be a set,
or it would be a member of itself; that would contradict the ZF axiom of
well-foundedness of membership.)
The ‘class of all constructible sets’, L, is defined similarly. For a set S, write
cl(S) for the closure of S under the Gödel operations which include operations
admitted by the ZF axioms as leading from sets to sets (such as the union of
two sets: for these, see, e.g., Jec1973, §3.4). Define
[
L 0 := 0, L α := Lβ if α is a limit ordinal,
β<α

L α+1 := ℘(L α ) ∩ cl(L α ∪ {L α }),

[
L:= Lα .
α∈On

This gives the constructible hierarchy L, with L ⊂ V .

Gödel’s Axiom of Constructibility (God1940) is

V=L

‘all sets are constructible’. It is stronger than both GCH and AC; see, e.g.,
Dev1977, §1. For a monograph treatment, see Dev1984.
 16.2 Hypotheses and Axioms 271

Without further axioms, one cannot say whether a given set has a non-
constructible subset; see the section on zero-sharp in §16.3.
The class L, as a subclass of V , is said to be an inner model of set theory.
Gödel’s contribution thus opened the door to the study of more general inner
models, with L the minimal such model. Thus one may expand L to include
sets not in L to enable their study. In the simplest case one takes a transitive
set x (one on which membership is a transitive relation) and starts a relative
constructible hierarchy (relative to x) with L 0 = x, applying at each stage α
the Gödel operations to L α (x) to obtain the class L(x) (see, e.g., Dra1974,
p. 149). Two examples of interest below are L(x), for x ⊆ ω (so x here is
interpreted via its characteristic function as a real number), and L(R), where
again the members of R are interpreted as subsets of ω. The axiom system ZF
holds in the latter inner model, but AC fails as the model has all sets of reals
Lebesgue-measurable and Baire: see Woo1988. These examples adjoin to L
sets from the low levels of V . Maximal inner models, known as core models,
adjoin to L structures whose elements extend all the way up to below some
‘large’ cardinal (for which see below in §16.3), enabling the study of strong
axioms of infinity. For instance, adjoin an ultrafilter U, presumed to exist on a
large enough cardinal κ, yielding a class denoted by L[U]. Here the transfinite
induction again mimics Gödel’s but constructs L α+1 [U] by reference to the set
U ∩ L α [U] (see, e.g., Dra1974, p. 151). For a textbook account, see Zem2002
or the later Handbook of Set Theory from 2010 (ForK2010).
Martin’s Axiom
A topological space X is said to satisfy the countable chain condition (ccc)
if every disjoint family of its open sets is countable. Thus R satisfies ccc (as
does any separable space); connected to this is Souslin’s hypothesis (SH) that
every (non-empty) complete, dense, linear order without first or last element
is order isomorphic with R, the study of which gave rise to Martin’s Axiom
(MA). Its statement is that (writing c for the cardinality of the continuum)
in every non-empty compact space satisfying ccc, the intersection of fewer
than c open dense sets is non-empty. In this context the statement thus asserts
Baire’s Theorem as extending from countable to fewer than c intersections. The
Continuum Hypothesis (CH) is equivalent to the same statement but without
the condition ccc. So,

CH Implies MA.

Regarding consistency, it is known that if ZFC is consistent, it remains

consistent after adding any one of the three additional axioms CH, MA and
MA + ¬CH (MA and the negation of CH).
272 Axiomatics of Set Theory

The last of these axioms, MA + ¬CH, implies SH; on the other hand, V =
L (in fact a much weaker combinatorial hypothesis, implied by this axiom)
contradicts SH, by constructing a counterexample.
Martin’s Axiom appears in the Martin–Solovay paper (MartS1970). It has
been extremely influential, as a substitute for and weakening of the CH. For a
monograph treatment of its many and fruitful consequences, see Fre1984; see
also Weis1984.
Countability, Category and Measure
As above, Martin’s Axiom may be regarded as freeing (so far as possible)
Baire category from countability. There is a measure-theoretic counterpart to
this: tau-additivity – see Fre2003 for a monograph treatment. A measure µ is
tau-additive (or τ-additive) if, for any increasing (perhaps uncountable) family
of sets, ( Aλ )λ<κ say,
[ 
supλ<κ µ( Aλ ) = µ(supλ<κ Aλ ), i.e. = µ Aλ .
λ<κ

The concept of tau-additivity should be compared with the more restrictive

property of a measure µ being κ-additive where for any disjoint family { Aλ :
λ < η} with η < κ
[  X
µ Aλ = µ( Aλ ).
λ<η λ<η

16.3 Ordinals and Cardinals

First, recall the ordinal numbers (§16.2; see, e.g., Dev1993, Ch. 3). The first
ordinal is 0, the second is 1 := {0}, and the (n + 1)th is n := {0, 1, . . . , n − 1}
(as in the von Neumann definition of N above). Then the first infinite ordinal is
ω := {0, 1, 2, . . . , n, . . .}, the second is ω + 1 := {0, 1, 2, . . . , n, . . . , ω}, etc. We
write
α + 1 := α ∪ {α},

the successor ordinal of α. Other ordinals are called limit ordinals.

Next, recall the cardinal numbers (or just cardinals) ℵα , indexed by the
ordinals, using Cantor’s notation. Call two sets equipotent if there is a bijection
between them. This is an equivalence relation on any given collection of sets.
Extending the order relation on the natural numbers N to infinite sets involves
the Schröder–Bernstein Theorem: if two sets have injections from each to the
other, there is a bijection between them. (For proof, see, e.g., BirM1953, or
 16.3 Ordinals and Cardinals 273

Cohn1989, both algebra texts; AC is not needed here, although DC is, cf.
Dev1993, pp. 77–78.) The cardinals are the initial ordinals α for which there
is no surjection f : β → α for an ordinal β < α. This introduces a second
ordinal-inspired notation for cardinals, due to von Neumann, replacing aleph
with omega so that
ℵα = ω α (α ∈ On),

with ω0 = ω. The finite cardinals n are just the finite ordinals, the number of
points in an n-point set; the smallest infinite cardinal is ℵ0 , the cardinality of
N (countable); the uncountable cardinals are ℵ1, ℵ2, . . .. We regard any ordinal
(including any cardinal) as the set of its predecessors, as above. In brief: ordinals
are absolute, cardinals are relative (cf. BinO2019a, p. 12), the latter depending
on what surjections are admitted on a given ‘putative’ cardinal.
There is a further (fourth) ‘elephant in the room’ here: the concept of a
cardinality attaching to sets presents a call on AC to choose a cardinal and a
bijection.
The Continuum Hypothesis may thus be written

c = ℵ1 . (CH)

Measurable Cardinals
We quote the following result of Ulam (Ula1930): if a finite countably additive
measure µ is defined on all subsets of a set X of cardinality ℵ1 , and vanishes
on all singletons, then µ vanishes identically.
Ulam’s Theorem leads to the related concepts of measurable cardinal and
Ulam number. A cardinal κ is called real measurable (in brief, measurable)
if there is some topological space of cardinality κ and a κ-additive proba-
bility measure µ defined on all its subsets and vanishing on singletons, non-
measurable otherwise. More restrictively, if the probability measure here takes
only the value 0 and 1, κ is called two-valued measurable (so two-valued non-
measurability is less restrictive than (real) non-measurability). Non-measurable
cardinals are called Ulam numbers. (In measure theory, measurable sets are
the ‘well-behaved’ sets; here, it is the non-measurable cardinals that are the
‘well-behaved’ cardinals. So we will say ‘two-valued non-measurable’ rather
than ‘not two-valued measurable’, etc.) The countable cardinal ℵ0 is non-
measurable. Any cardinal less than a non-measurable one is non-measurable,
so the non-measurable cardinals form an ‘initial interval’ in the cardinals. By
the Ulam–Tarski Theorem, the immediate successor of a non-measurable is
non-measurable, and the supremum of non-measurably many non-measurables
is non-measurable (see, e.g., Bog2007a, Th. 1.12.42 or Fed1969, 2.1.6).
274 Axiomatics of Set Theory

Existence of Measurable Cardinals

Write M for the axiom that there exists a measurable cardinal, ZFM for ZF +
M. In ZFM, non-constructible sets exist (Sco1961). So Gödel’s Axiom of
Constructibility V = L is incompatible with the existence of a measurable
cardinal (LevS1967, p. 240). Unfortunately, this casts no light on CH. If ZFM
is consistent, then so are both ZFM + CH and ZFM + ¬CH (LevS1967, Th. 1).
Similarly for ZFM and Souslin’s hypothesis SH (above) and its negation, and
ZFM + all projective sets are Lebesgue-measurable (LevS1967, p. 236).
Classes of Cardinals
Call a cardinal κ inaccessible if the class of all smaller cardinals has no maximal
element, and no set comprising cardinals less than κ of cardinality less than
κ has supremum κ. Furthermore, 2λ < κ for all cardinals λ < κ (the latter
condition referred to as strong inaccessibility). So if measurable cardinals exist,
the smallest one is inaccessible, and so all are. We quote (see, e.g., Bog2007a,
1.12(x)):

(a) If κ is two-valued non-measurable, so is 2κ .

(b) ℵ1 is non-measurable, being the successor of the non-measurable ℵ0 .
(c) c is two-valued non-measurable.

This follows from (a). If further one assumes Martin’s Axiom, then:

(MA) c is (real) non-measurable.

Ramsey Cardinals
The branch of infinite combinatorics relevant here, partition calculus, derives
from Erdős and Rado; see, e.g., BinO2019a, §4.1; Dra1974, Chs. 7, 8. Here, as
there, it is convenient to use Erdős’s notation: write

κ → (α)2<ω

as shorthand for: if the class [κ]<ω of finite subsets of κ is partitioned into

two classes (using two colours, say), then there is a monochromatic (‘homoge-
neous’) subset of κ of order type (isomorphism type, of linearly ordered sets)
α. Ramsey’s Theorem (Rams1930) is then written ω → (ω)22 .
For κ measurable, one has further

κ → (κ)2<ω

(see BinO2019a, §4.2), the definition of a Ramsey cardinal. For accounts of

Ramsey theory, see also, e.g., Kec1995, §19; Bol1998, Ch. VI; HinS1998, Chs.
5, 18.
 16.3 Ordinals and Cardinals 275

Supercompact Cardinals
We need three further kinds of large cardinal (see, e.g., BinO2019a, §4.3;
Mit2010, pp. 1487–1488). A cardinal κ is supercompact if it is λ-supercompact
for all λ ≥ κ; here κ is λ-supercompact if there is a (necessarily non-trivial)
elementary embedding j = jλ : V → M with M a transitive class, such that j
has critical point κ, and M λ ⊆ M, i.e. M is closed under arbitrary sequences
of length λ. Under AC, without loss of generality, j (κ) > λ.
Strong Cardinals
For κ a cardinal and λ > κ an ordinal, κ is said to be λ-strong if for some
transitive inner model (transitive class containing the ordinals), M say, there
is an elementary embedding (elementary, as below), jλ : V → M with critical
point κ, jλ (κ) ≥ λ, and
Vλ ⊆ M.
Furthermore, κ is said to be a strong cardinal if it is λ-strong for all ordinals
λ > κ.
Elementarity here concerns the requirement that any sentence referring to a
finite number of elements of the structure hV, ∈i holds in the structure if and
only if the same sentence, but rewritten so as to refer to the j images of the said
finite set, holds in hM, ∈i (BelS1969, Ch. 4).
This notion may be relativized to subsets S to yield the concept of λ-S-strong
by requiring, in place of the inclusion above, only that
j (S) ∩ Vλ = S ∩ Vλ ;
one says that j ‘preserves’ S up to λ.
Woodin Cardinals
This leads to our final definition. The cardinal δ is a Woodin cardinal (Woo1999)
if δ is strongly inaccessible, and, for each S ⊆ Vδ , there exists a cardinal θ < δ
which is λ-S-strong for every λ < δ. (So the last of these three embedding-
based definitions calls for more ‘preservation’ than the second, but less than
the first.)
The consistency strength (see the next paragraph) of various extensions of
the standard axioms, ZFC, by the addition of further axioms, may then be
compared (perhaps even assessed on a well-ordered scale) by determining
which canonical large-cardinal hypothesis will suffice to create a model for the
proposed extension. Thus, for κ supercompact, Vκ is a model for the sentence
asserting that a strong cardinal exists (abbreviated, using satisfaction |= as
below, to Vκ |= ∃µ[‘µ is strong’]) which places supercompact ‘above’ strong,
in the sense that the assumption of the existence of a supercompact cardinal is
stronger than the assumption of the existence of a strong cardinal (indeed, also
276 Axiomatics of Set Theory

of a strong cardinal below a supercompact). Likewise, for κ strong, Vκ is a model

in which the sentence asserting the existence of a measurable cardinal holds
(abbreviated to Vκ |= ∃µ[‘µ is measurable’]), placing measurability ‘below’
strong, in the same sense. (Below that in turn is the existence of a Ramsey
cardinal, as above.)
Consistency Strength
We have just met measurable and inaccessible cardinals and ZFC, whose con-
sistency we assume. But this cannot imply the existence of an inaccessible
cardinal κ, because of Gödel’s incompleteness theorem (see, e.g., BinO2019a,
p. 16), as otherwise ZF would provide a proof of its own consistency, since all
the axioms of ZF would be provably satisfied within the set Vκ . So to go further
here, one needs a stronger axiom of infinity (stronger than that which admits
N as a set) – or axioms. These involve large cardinals (see, e.g., Dra1974 or
BinO2019a, §§1,4,7,10). The axioms that we will use are comparable in their
logical strength (are ordered by implication). Using ‘>’ as shorthand for ‘exis-
tence of needs a stronger assumption than’, the categories of large cardinal we
use can be ranked, or compared, as follows (BinO2019a, p. 17):

supercompact > Woodin > strong > measurable > Ramsey.

Zero Sharp
For background on 0] and other sharps, see, e.g., Mit2010; BinO2019a, p. 21;
Dra1974, §8.5 and the brief account below. As existence of Ramsey cardinals
is in the least restrictive position in the diagram above, we note (again, see, e.g.,
Mit2010):

(i) The existence of 0] is an immediate consequence of that of a Ramsey

cardinal.
(ii) ‘Assuming the existence of a large cardinal (a Ramsey cardinal is much
more than enough), it can be shown to be consistent that 0]α exists for all
ordinals α.’

To describe 0] requires first some notation.

Recall Gödel’s use of natural numbers to code formulas of the standard
formal language of set theory (involving the symbols = and ∈). We write pϕq
for the number coding the formula ϕ. For a transitive set or class M and
formula ϕ with no more than n free variables, these to be denoted x 1, . . . , x n,
we write M |= ϕ[a1, . . . , an ] to mean that ϕ is satisfied (holds) in M when x 1
is interpreted as referring to a1 , x 2 as a2,. . . (where a1, . . . , an are in M). For
M a set, the relation of satisfaction can be defined formally by induction on the
complexity of the formula ϕ (starting with M |= (x 1 ∈ x 2 )[a, b] if and only if
 16.3 Ordinals and Cardinals 277

a ∈ b); we note the need to go outside M to describe what goes on in M, when

eliminating the operation of quantifiers, as in M |= ∃x(x ∈ x 1 )[b] if and only
if there exists a ∈ M with a ∈ b.
Define f on ω by f (i) = ωi and, following Dra1974, p. 257, put
0# := {pϕq : L |= ϕ[ f ]},
so that any free variables in any formula ϕ are assigned interpretations from
the sequence f .
Despite Tarski’s Theorem on the undefinability of truth in a structure within
itself, this may be a well-defined subset of ω assuming, for instance, the exis-
tence of a cardinal κ for which the partition relation κ → (ω1 ) <ω holds, e.g. if
κ is measurable or a Ramsey cardinal. In such circumstances, the structures L λ
for λ an uncountable cardinal constitute an elementary chain, allowing passage
from ‘language to meta-language’, so that one can define truth in L in terms of
satisfaction in L λ for large enough λ.
The phrase ‘0# exists’ is used to describe circumstances that enable satisfac-
tion in L to be definable. An insightful result due to Kunen (Kun1970) gives as
a necessary and sufficient condition the existence of a non-trivial elementary
embedding of L to itself (one that moves ordinals above a certain critical point).
A similar definition, allowing relativization to a parameter x ⊆ ω, can be
given for x # , in which L is replaced by L(x).
The proof of Kunen’s characterization of the existence of 0# involving
a further equivalence to the existence of a closed unbounded class of L-
indiscernibles (for which see our survey paper) may be found in several text-
books, including those of Jech (Jec1997), Devlin (Dev1984) and Kanamori
(Kan2003). Indeed, the definition we give, after Drake, relies on a specific
sequence, f , of indiscernibles.
 Epilogue: Topological Regular Variation

Recall from the Preface that the book grew out of our work on Topological
Regular Variation, then our intended title for the book, now that of this Epilogue,
and the use of the Prologue and Epilogue on regular variation as a framing
device for our main text, Chapters 2 to 16.
The text below may be viewed as a brief summary of (or, overview of and
introduction to) the relevant papers in our corpus:
BinO2009b: generic regular variation;
BinO2009c: the index theorem;
BinO2009d: foundations;
Ost2010: topological dynamics;
BinO2010b: regular variation without limits;
BinO2010c; BinO2010d; BinO2010e: the topological regular variation tril-
ogy;
BinO2013: Steinhaus theory and regular variation;
BinO2014: Beurling aspects.
The significant contribution regular variation plays in the classical probability
theory of random variables naturally guided the research when the context
more recently expanded to finite-dimensional vector spaces (see MeeS2001)
and further to function spaces such as C[0, 1] and the Skorokhod space D[0, 1]
of Rd -valued càdlàg functions on [0, 1] (see HulLM2005). It was realized by
Bajšanski and Karamata as early as 1969 (in BajK1969) that some of their
foundational work on regular variation can in fact be conducted in a group-
theoretic framework working with functions h : G → H between topological
groups G and H. Their investigation was brief 1 and with the sole exception of
Bal1973 that point of view was not pursued much further for three decades.
1 The notion of convergence in BajK1969 was based on filters; we have espoused a sequential
approach.

278
 Epilogue: Topological Regular Variation 279

Our purpose here is to show how the mathematical panoply assembled in the
preceding chapters can upgrade the classical theory and how its algebraic stance
provides a hitherto unknown lens through which to see and simplify classical
arguments. The principal tool marries the Category Embedding Theorem with
the Effros Theorem. The first theorem embeds into any non-negligible set S,
quasi always in S, any null sequence ψn of homeomorphisms (i.e. homeomor-
phisms converging to the identity), ψn → id X meaning that, for q.a. s ∈ S,

{ψn (s) : m ∈ Ms } ⊆ S

for some infinite set Ms ⊆ N. The second theorem provides, for any convergent
sequence z n → z0 , a null sequence of homeomorphisms with ψn (z0 ) = z n . We
may call such a sequence an Effros null sequence.
Uniform Convergence Theorem
The foundation stone of classical regular variation, and so the primary object
of study for its topological counterpart, is the Uniform Convergence Theorem
(UCT). This assumes in its multiplicative form, as we recall from the Prologue
(§P.8), a regularly varying function h : R → R; that is, one with a well-defined
limit (its kernel):

k (t) := lim h(t x)/h(x) for t ∈ R+ := (0, ∞),

x→∞

and concludes uniform convergence as t ranges over compact sets, but subject
to h being (Lebesgue) measurable or Baire (having the Baire property). The
multiplicative version of regular variation hints already that it is natural to
study a normed group X (i.e. a group with a group-norm, as in Chapter 6)
and a Baire map h : X → H with H also a normed group. The expression
t x is then interpreted in the context of a group action ϕ(t, x) on X provided
by the group H (X ) of norm-bounded autohomeomorphisms of X under the
supremum group-norm:

kαk∞ = kαk H =: sup kα(x)kX .

x ∈X

For connections to the compact-open topology on the homeomorphisms

(Eng1989, §3.4 or Kur1968, Ch. 4, §44), see MilleO2012. See also Dij2005.
The action may be restricted to some subgroup of homeomorphisms of inter-
est, T ⊆ H (X ); for instance, T might comprise a group of translations of X,
say the left translations λ x : u 7→ xu for x, u ∈ X . However, taken in as wide
a context as the premise of UCT allows (evidently, with H a normed group,
with identity element e H ), X itself may more simply be a topological space
on which a topological group T acts transitively, thus making X homogeneous.
(See also Ost2010 for a more general context.) An associated definition of
280 Epilogue: Topological Regular Variation

regular variation gives meaning to slow variation in the following sequential

convergence:

h(ϕn (x))h(ϕn (z0 )) −1 → e H (x ∈ X ).

The task is then to deduce that this convergence holds uniformly in x, i.e. for x
ranging over any compact K ⊆ X . Here z0 is a distinguished (fixed) reference
point or null point. What actions ϕn should we allow? Following Doob’s
convention we write α◦ = (α n )n for a sequence of elements and demand that it
be a (pointwise) divergent sequence: for any M > 0, ultimately all n satisfy

d X (α n (x), x) ≥ M for all x.

Shift-compactness appearing in proofs of the UCT in the main text required

us to transfer an arbitrary convergent sequence un → u in the given compact
set K to a ‘null sequence’; here we take sequences z n → z0 as our canonical
null sequences, in view of the distinguished null point. The transfer is enabled
by assuming T-homogeneity of X: we choose σ ∈ H (X ) with σz0 = u so that,
with z n := σ −1 un , we have un = σz n → σz0 . Next, the Effros property, as
mentioned above, supplies a sequence of homeomorphisms converging (in the
supremum norm of H (X )) to the identity, say ψn → idX , with z n = ψn (z0 ).
It emerges that the Category Embedding Theorem (CET) of Chapter 10 is
applicable to any sequence converging to the identity. Thus one is able to
deduce from CET, for any non-meagre Baire set S, that for quasi all s ∈ S (in
particular for some one s) and corresponding Ms ⊆ N,

{ψn (s) : m ∈ Ms } ⊆ S.

Hence α◦ -slowly varying functions h : X → H involve taking all ϕ◦ of the form

ϕn = α n0 = α n σψn .

These perturbed shifts are also divergent. We thus demand, for all x ∈ X and
all perturbed shifts α n0 of α n , that

h(α n0 (x))h(α n0 (z0 )) −1 → e H .

The UCT is then derivable in the following general form from CET:

Theorem E.1 Suppose α◦ in H (X ) divergent, T is a subgroup, and X is

non-meagre, T-transitive and T-Effros. If h : X → H is Baire and α-slowly
varying, then, for compact K ⊆ X,

h(α n (x))h(α n (z0 )) −1 →uniformly e H (x ∈ K ).

 Epilogue: Topological Regular Variation 281

For the proof, see BinO2010c. (There we refer to ‘unconditional’ slow vari-
ation emphasizing with the qualifier that the assumed convergence of slow
variation involves all α◦0 arising from a given divergent sequence α◦ .)
UCT on the L 1 Algebra of a Locally Compact Metric Group G
In the preceding contexts derivation of the UCT relied on transitive action.
But one can sometimes have an alternative approach. Here we have G a locally
compact metric group and work on L 1 (G) with its natural action of convolution;
however, this action need not be transitive. Let G be equipped with a (left)
Haar measure η. Take the domain and range of regularly varying functions to
be X = L 1 (G, η), regarded as the Banach algebra of η-integrable functions
x : G → R under convolution. Thus
Z
k xk1 = |x(g)| dη(g).

The group G defines a natural action on X, namely ∗ : G × X → X, where

(g ∗ x)(t) := x(g −1 t) (g, t ∈ G, x ∈ X ).

Definitions (1) Call the map h : X → X slowly varying (with respect to the
net x := {x δ }) if

lim kh(g ∗ x δ ) − h(x δ )k1 = 0, for each g ∈ G. (SV)

δ

(2) Say that h : X → X is Baire relative to convolution if the maps h x : G → X

defined by
h x (g) := h(g ∗ x) (g ∈ G, x ∈ X )

are Baire for all x off a meagre set, the exceptional set, Eh , of h.
(As always, the map h : X → X is Baire if h−1 takes open sets to sets with
the Baire property.)
(3) For Baire h : X → X, say that h is slowly varying with respect to regular
nets if it is slowly varying with respect to nets {x δ } with all x δ < Eh ; as in (2)
above, here Eh is the meagre exceptional set corresponding to h.

It emerges that for G separable and h : X → X Baire, there does exist a

meagre set Eh in X such that h x is Baire for each x in the co-meagre set
R := X\Eh ; when h is continuous Eh is empty. Under these circumstances we
have:

Theorem E.2 (UCT for L 1 (G); BinO2013) For G a locally compact metric
group with Haar measure η and X = L 1 (G, η):
282 Epilogue: Topological Regular Variation

(i) for G separable (i.e. σ-compact), if h : X → X is Baire and slowly varying

with respect to regular nets, then the convergence in (SV) is uniform on
compacts;
(ii) for general G, uniform convergence in (SV) holds for h continuous and
slowly varying with respect to arbitrary nets.

Here the weaker assumption on the action extracts a price: slow variation is
defined relative to regular nets, i.e. nets consisting of points avoiding a specified
meagre set – in order to secure the Baire property for the maps h x .

For G again a locally compact metric group with Haar measure η and X =
L 1 (G), denote the non-negative probability densities by
Z
P(G) := {y ∈ L 1 (G) : y ≥ 0, y dη = 1},

and say that the (weak) Reiter condition (Pat1988, Prop. 0.4) holds for a net
{x δ } in P(G) if
kg ∗ x δ − x δ k1 → 0 for each g ∈ G. (R)

Taking h(x) := x which is continuous, one may deduce in our context that the
condition (R) then holds uniformly on compact sets, that is:

k g ∗ x δ − x δ k1 → 0 on compact sets of g ∈ G. (UR)

This is a significant fact (cf. Pat1988, Prop. 4.4) in amenability theory, as the
Reiter condition (R) is equivalent to amenability. (An invariant mean can be
extracted as any weak* cluster-point of the net { x̂ν }, where x̂ represents x in the
second dual L 1 (G) 00 = L ∞ (G) 0; for background on nets, see Eng1989, §1.1.6,
or Kel1955, Ch. 2.)
Characterization Theorem
The UCT is linked to slow variation of a Baire function, h : X → H, whose
kernel is at its simplest: k (t) ≡ e H . Aiming for a general kernel, the immediate
question is its existence and characterization relative to the group of actions T
on X. Here it is helpful that the domain of the action is a group T, since

S := {t ∈ T : there exists a limit lim h(t x)/h(x)}

x→∞

is a subgroup, and so the Subgroup Dichotomy Theorem (Chapter 9) can be

called into play. We thus have:

Theorem E.3 (Characterization Theorem; cf. BGT1987, Th. 1.4.1) Let X and
H be normed groups, T ⊆ H (X ) a connected non-meagre subgroup, Baire
 Epilogue: Topological Regular Variation 283

under the (right) norm topology, acting on the group X and let h : X → H be
Baire. If the limit
k (t) = ∂X h(t) := lim h(t x)h(x) −1
x→∞

exists on a non-meagre subset of T, then ∂X h(t) exists on all of T and is a

continuous homomorphism from T to H:
k (st) = k (s)k (t) (s, t ∈ T ).
For the proof, see BinO2010d, §2. For the role of homomorphisms arising
in the associated functional equations of probability theory, see Ost2017. The
continuity assertion above is linked to continuity properties of Baire functions
described in the Discontinuity Theorem and the Banach–Neeb Theorem, for
which see Chapter 12. Matters are simpler when the groups in question are
normed spaces. We recall the following result (cf. Ban1932, 1.3.4, p. 40, albeit
for ‘Baire-measurable’ functions; Meh1964).
Theorem E.4 (Banach–Mehdi Theorem) An additive Baire function between
complete normed vector spaces is continuous, and so linear, provided the image
space is separable.
For a proof, see BinO2009c, §2 and also Chapter 12. For a locally compact
Hausdorff space T, in the case of C0 (T ), in the context of the space of real-
valued continuous functions with compact support: from this last result, the
standard Riesz representation theorem (Rud1966, Ch. 6) and reference to Φ :=
{ϕu : u ∈ C0 (T )} the group of shift homeomorphisms ϕu (x) = x + u, we have:
Theorem E.5 Let X = C0 (T ) with T a locally compact Hausdorff space.
For Φ the group of shift homeomorphisms, if h : X → R is Baire Φ-regularly
varying with distinguished null point the zero of X, then, for some measure µ
on T, we have
Z
k (x) := ∂Φ h(x) = x(t) dµ(t).
T
Real flows
A further case in which a representation is achievable is the context of real
flows where for T one takes either of T = R, the additive reals, or T = R+, the
multiplicative reals, to act on a space X. Here flow homogeneity occurs. For
the proof see BinO2009c, §3.
Theorem E.6 (Index Theorem) For a (real) flow ϕ on X and kernel k of
h : X → H, there is a flow on H associated with the kernel k
ϕk (t, z) = z ρ(t) for z ∈ H, t ∈ R
284 Epilogue: Topological Regular Variation

with ρ multiplicative (on H) R+ → H such that

k (ϕ(t, x)) = ϕk (t, k (x)), i.e. k (t x) = t ∗ k (x).
For X, H complete with H separable, there exists ρ ∈ H such that
ϕk (t, z) = z + ρt.
Illustrative Example For X = R2 = C({0, 1}), the vector flow in direction u
given by
ϕ(t, x) = ϕtu (x) = (x 0 + tu0, x 1 + tu1 ),
and h(x) = α0 x 0 + α1 x 1 , we compute its kernel to be k (x) = h(x). Indeed,
h(ϕ(t, x)) − h(ϕ(t, 0)) = α0 (x 0 + tu0 ) + α1 (x 1 + tu1 ) − (α0 u0 t + α1 u1 t)
= α0 x 0 + α1 x 1 = k (x).
So, for u = e0 = (1, 0), writing ϕhoriz (x) := (x 0 + 1, x 1 ),
k (ϕhoriz (t, x)) = α0 (x 0 + t) + α1 x 1 = k (x) + α0 t :
ρhoriz = α0 .
Similarly ρvert = α1 and ρu = α0 u0 + α1 u1 = ( ρhoriz, ρvert ) · u.
Calculus of Regular Variation: The Differential Modulus
Our main concern in this section is with products of regularly varying functions.
In the classical context of the real line, it is obvious that the product of two
regularly varying functions is regularly varying. This is also true in the current
context of functions h : X → H when the group H is abelian and the metric is
invariant. To see this observe that a regularly varying function is the product of
a multiplicative function and a slowly varying function.
What may be said if H is non-commutative? It emerges that while ϕ-slowly
varying functions have a group structure, identifying what happens when taking
a product of ϕ-regularly varying functions, f 1 and f 2 say, requires the interven-
tion of a modular function akin to that used in switching between left and right
Haar measures. The need for a switching ‘differential modulus’ (differential,
since kernels are derivatives at infinity) follows from the First Factorization
Theorem, E.8, which permits us to write f i = ki hi for corresponding kernels
k i and slowly varying factors hi , yielding:
f 1 f 2 = k 1 h1 k 2 h2 .
So h1 needs to switch position around k 2 .
Indeed, it has to be appreciated that our definition of regular variation opted
for division on the right, so to be fair the question should address one-sided
 Epilogue: Topological Regular Variation 285

multiplication (in fact on the left, see below). To guess at the answer, take
X = H and focus on the special case of two multiplicative functions k (x) and
K (x) with K (x) = x. If the (pointwise) product k (x)K (x) were to be regularly
varying, one would expect it to be multiplicative, and the latter property is
equivalent to
k (x)k (y)x y = k (x y)xy = k (xy)K (xy) = k (x)xk (y)y, i.e. k (y)x = xk (y),
for all x, y ∈ H. This asserts that each value k (y) commutes with each element
x in the group H. From this one conjectures that the range of k must lie in the
centre Z (H) of the group H. (We recall that the subgroup Z (H) = {a ∈ H :
ah = ha for all h ∈ H } is the algebraic centre.)
The conjecture is upheld as correct by the Centrality Theorem, E.13, but the
argument requires the normed group H to have a bi-invariant metric:
Definition A metric d H on H is bi-invariant if d H (ax, ay) = d H (x, y) =
d H (xa, ya) for a, x, y ∈ H.
Bi-invariance is equivalent, as Klee (Kle1952) shows, to the existence of a
metric possessing what we will term Klee’s property:
d(ab, xy) ≤ d(a, x) + d(b, y), i.e. kab(xy) −1 k ≤ kax −1 k + kby −1 k.
The bi-invariance property acts as a replacement for commutativity, and is
exactly the condition which allows a proper development of the calculus of
regularly varying functions, mimicking the non-commutative development of
the Haar integral (see, e.g., DieS2014). Note that a Klee group is a topological
group; cf. BinO2010g, Th. 3.4.
We now make our blanket assumptions that:
(i) the norm on H has the Klee property above;
(ii) h is ϕ-regular varying when the limit function below (the kernel) exists:
k (x) = ∂ϕ h(x) := lim h(ϕn (x))h(ϕn (eX )) −1 ;
n→∞

(iii) ϕ◦ is a fixed and, as earlier, divergent sequence of homomorphisms in

H (X ); and that functions g, h, k will always map from X to H.
For the proofs of results cited below, see BinO2010e.
Proposition E.7 (Group Structure of SV ) If h : X → H is ϕ-slowly varying,
then the involutory mapping h−1 : x → h(x) −1 is ϕ-slowly varying. Hence the
product of two ϕ-slowly varying functions is ϕ-slowly varying.
Theorem E.8 (First Factorization Theorem) If h : X → H is ϕ-regularly
varying, then, for k = ∂ϕ h(t),
286 Epilogue: Topological Regular Variation

(i) k (t) is ϕ-regularly varying and k (t) = ∂ϕ k (t);

(ii) h̄(t) := k (t) −1 h(t) is ϕ-slowly varying. Thus h(t) is the left product of its
limit function with a slowly varying function h̄:

h(t) = ∂ϕ h(t) h̄(t).

Theorem E.9 (Second Factorization Theorem) If g is ϕ-regularly varying

and h is slowly varying, then g(t)h(t) is regularly varying with limit ∂ϕ g.

It emerges that in the non-commutative case the set of points a of convergence

of a sequence of conjugates {hn ah−1n } is well structured. The choice of sign
in the notation below is motivated by the Modular Flow Theorem, E.11, to be
established shortly.

Theorem E.10 (Asymptotic Conjugacy Theorem) For h◦ an arbitrary se-

quence in H, the two sets of convergence defined by

D± (h◦ ) = {a : h±1 ∓1
n ah n is convergent}

are subgroups and are closed if H is complete. They support corresponding

(asymptotically inner) automorphisms between themselves:

A± (h◦, a) := lim h±1 ∓1

n ah n ∈ D∓ (h◦ ) for a ∈ D± (h◦ ).

These automorphisms are mutual inverses:

A± (h◦, A∓ (h◦, a)) = a for a ∈ D± (h◦ ).

Definition A function h : X → H generates a sequence hn =: h(ϕn (eX ))

which we may denote by h◦ without ambiguity. We may now define the forward
and backward moduli of h relative to ϕ by

∆± (h, a) = A± (h◦, a) = lim h±1 ∓1

n ah n for a ∈ D± (h◦ ).

We shall say that h : X → H is modular if all points of H are points of

convergence for h◦ ; that is, D+ (h◦ ) = H, so that one may thus say that H is
asymptotically invariant for h◦ . Put H◦ := {h ∈ H X : h is modular} and set
1 H (x) ≡ e H .

Theorem E.11 (Modular Flow Theorem) H◦ is a group with identity element

1 H , and ∆+ is an H◦ -flow on H with

∆+ (gh, a) = ∆+ (g, ∆+ (h, a)) and ∆+ (1 H , a) = a,

∆+ (h, ∆− (h, a)) = a.
 Epilogue: Topological Regular Variation 287

Theorem E.12 (Left Product Theorem) Suppose that g, h are ϕ-regularly

varying with kernel functions k and K, and with g modular. Then gh is ϕ-
regularly varying with kernel (limit)

k (x)∆+ (g, K (x)).

Theorem E.13 (Centrality Theorem) Suppose g is ϕ-regularly varying and

modular with kernel function k. Then, for every ϕ-regularly varying function h
and for all x, y, each element ∆− (h, k (y)) commutes with each element ∂ϕ h(x).
In particular, k is central, i.e. the range {k (x) : x ∈ X } is a subset of the centre
Z (H).

For H complete under its norm, modularity of h follows from the following
summability criterion for hn = h(ϕn (eX )):
∞
X
n k < ∞;
khn+1 h−1
n=1

this is a strengthened form of one of Kendall’s sequential conditions for regular

variation, namely khn+1 h−1n k → 0 (see §P.8 in the Prologue). Closeness of
modularity to centrality is indicated by the next result (in which H need not be
abelian).

Theorem E.14 For H complete and h : X → H ϕ-slowly varying and h

satisfying the summability condition:

(i) for k central, k h and hk are modular;

(ii) k h is modular if and only if k is central.

Through an Algebraic Lens: Beurling Regular Variation Reappraised

Recall from the Prologue (§P.6) the definition of Beurling slow variation:

ϕ(x + tϕ(x))/ϕ(x) → 1, as x → ∞ (t ∈ R) (BSV)

for ϕ : R → R+ with ϕ(x) = o(x); recall also its role in the Beurling Tauberian
Theorem (§P.7). This prompted the study (BinO2014) of Beurling ϕ-regular
variation, in which kernels are defined by

k (t) = lim f (x + tϕ(x))/ f (x).

x→∞

Allowing ϕ(x) = O(x) and adjusting (BSV) to have a more general limit gains
a wider remit, embracing classical as well as Beurling regular variation:

ϕ(x + tϕ(x))/ϕ(x) → η(t) as x → ∞ (t ∈ R).

288 Epilogue: Topological Regular Variation

Here ϕ is said to be self-equivarying if the convergence is locally uniform

(which will be the case for Baire ϕ; see Ost2015a). Then it emerges that the
limit η satisfies the Gołab–Schinzel
˛ equation

η(u + vη(u)) = η(u)η(v). (GS)

Imposing positivity implies that η is continuous and of the form η(t) = 1+ ρt for
some ρ ≥ 0. Significantly, the form of the equation prompts a binary operation,
termed the Popa operation (see Pop1965, cf. Jav1968, the first two papers in
which it was studied):
a ◦η b := a + bη(a).

This induces a group structure on the set Gη := {x ∈ R : η(x) , 0} and allows

η to be viewed as a homomorphism. By comparison, for fixed x, the binary
operation of ‘addition accelerated by a factor ϕ(x)’:

a ◦ϕ b = a + bϕ(x)

may properly be interpreted as an asymptotic action in which

η x (t) := ϕ(x ◦ϕ t)/ϕ(x) → η(t).

This asymptotic view enables a transparent study of regular variation in which

the symbolism

f (x ◦ϕ t)/ f (x) → k (t) as x → ∞ (t ∈ R)

is guided through the algebraic lens of topological regular variation. For details
see BinO2016a; BinO2020a. The Popa group structure yielded a simplification
to the hardest theorem of Regular Variation (see §P.6 on Quantifier Thinning in
the Prologue) and allowed a proper understanding of the Goldie equation at the
heart of that theorem. See Ost2016b; BinO2022b; BinO2022c; BinO2022d.
Cocycles and Coboundaries
It is just and fitting for the Epilogue to turn the algebraic lens briefly on the reg-
ular variation of the Prologue, only to discover the inevitable common features:
the use in algebraic topological dynamics of cocycle terminology, for which
see p. 289 (cf. Ellis’ monograph, Elli1969, and its review in MR0267561). We
take up this theme from the vantage point of the following.

Definition Let ϕ = {ϕn } be a divergent sequence. Say that h : X → H is

ϕ-regularly varying, or if context permits, just Fréchet-regularly varying, if for
some function k (· ) = ∂ϕ h(· ) and, for each t,

h(ϕn (t)h(ϕn (z0 )) −1 → k (t).

 Epilogue: Topological Regular Variation 289

See also BinO2010g, Section 5 for an integrated treatment in the context

of normed groups. For the background and relation here between Gateaux,
Fréchet, and Hadamard differentiability, see BinO2009c, §3, Remark.

Proposition E.15 (Concatenation Formula) If h is ϕ-regularly varying for

the distinguished point z = z0 , then for any w the corresponding Fréchet limit
k w (x) = limn h(ϕn (x))h(ϕn (w)) −1 exists and

k z (x) = k z (w)k w (x).

Definition Let H0 = {ϕ ∈ H (X ) : ϕ(z0 ) = z0 } be the stabilizer subgroup

(of the distinguished null point). Note that this is conjugate to the stabilizer of
any other point of the (homogeneous) space X. Thus, for σ, τ in H (X ) with
σ(z0 ) = τ(z0 ), we have σ −1 τ ∈ H0 . We will regard two homeomorphisms
σ, τ in H (X ) that are H0 -equivalent (i.e. both in the same coset of H0 , e.g.
τ ∈ σH0 ) as equal. Whenever convenient, we will denote by σ x the unique
homeomorphism (up to equivalence) taking z0 to x. This is particularly useful
when G is a topological group, where the canonical choice is

σu (g) = λ u (g) = ug,

as we then have σu σv = σuv . The following result justifies the use of H0 -

equivalence.

Proposition E.16 If h is strongly ϕ-regularly varying and σ is a bounded

homeomorphism with σ(z0 ) = z0 , then the corresponding Fréchet limit function
satisfies k (σ(t)) = k (t).

For our next result we need to observe that in the context of real-valued
functions the equation

σ f (t, x) := f (t x) f (x) −1

defines a cocycle, since

f (t x) f (st x)
σ f (st, x) = σ f (t, x)σ f (s, t x) = .
f (x) f (t x)
A cocycle σ(t, x) is a coboundary if for some f ,

f (t x) = f (x)σ(t, x).

Thus in particular the kernel k of a regularly varying function f gives rise to

a coboundary, as in the next result, which refers to Chapter 11 and its Effros
properties, Theorems 11.1.2 and 11.2.1.
290 Epilogue: Topological Regular Variation

Theorem E.17 (Continuous Coboundary Theorem) Suppose that X is a Baire

space with the Effros Crimping property, §11.3 (as in the UCT above). Let H
be a topological group such that the pair (X, H) enables Baire continuity. If
h : X → H is Baire regularly varying with limit function k, then k is Baire, has
the coboundary property
k (σy ) = k σ(x,y) k (σ x ),
equivalently,
k (σ x σy ) = k (σ x )k (σy )
and is continuous.
Recall that ‘enabling Baire continuity’ is discussed in Chapter 13. For the
proof, see BinO2021a.
This leads us to our final result, which plays a thematic role throughout the
theory.
Corollary E.18 (Continuous Homomorphism Theorem) Suppose that h : X →
H is a Baire regularly varying function defined on a Baire topological group
X with values in the topological group H, and that the pair (X, H) enables
Baire continuity. If h has a limit function (kernel) k, then k is a continuous
homomorphism, i.e.
k (x y) = k (x)k (y).
 References

[AarGM1970a] Aarts, J.M., de Groot, J. and McDowell, R.H. Cocompactness. Nieuw

Arch. Wisk. 18 (1970), 2–15. 57
[AarGM1970b] Aarts, J.M., de Groot, J. and McDowell, R.H. Cotopology for metriz-
able spaces. Duke Math. J. 37 (1970), 291–295. 57, 135, 136
[AarL1974] Aarts, J.M. and Lutzer, D.J. Completeness properties designed for
recognizing Baire spaces. Dissertationes Math. (Rozprawy Mat.) 116
(1974), 48. 37
[AczD1989] Aczél, J. and Dhombres, J. Functional Equations in Several Vari-
ables. Encycl. Math. Appl. 31. Cambridge University Press, 1989.
8
[AlbM2016] Alberti, G. and Marchese, A. On the differentiability of Lipschitz
functions with respect to measures in the Euclidean space. Geom.
Funct. Anal. 26 (2016), 1–66. 228
[AlbGI2017] Albeverio, S., Garko, I., Ibragim, M. and Torbin, G. Non-normal
numbers: Full Hausdorff dimensionality vs zero dimensionality. Bull.
Sci. Math. 141 (2017), 1–19. 226
[AloS2008] Alon, N. and Spencer, J.H. The Probabilistic Method, 3rd ed. Wiley,
2008 (2nd ed. 2000, 1st ed. 1992). 37
[Amb1947] Ambrose, W. Measures on locally compact topological groups. Trans.
Am. Math. Soc. 61 (1947), 106–121. 253
[AnaL1985] Anantharaman, R. and Lee, J.P. Planar sets whose complements do
not contain a dense set of lines. Real Anal. Exchange 11 (1985–1986),
168–179. 156
[Anc1987] Ancel, F.D. An alternative proof and applications of a theorem of
E.G. Effros. Michigan Math. J. 34 (1987), 39–55. 172, 173, 176,
178, 181, 182, 196, 216
[Arh1963] Arhangelskii, A.V. On a class of spaces containing all metric and all
locally compact spaces. Soviet Math. Dokl. 151 (1963), 751–754. 49,
165
[Arh1995] Arhangelskii, A.V. Paracompactness and metrization: The method
of covers in the classification of spaces. In A.V. Arhangelskii (ed),
General Topology III. Encyclopaedia Math. Sci. 51. Springer, 1995,
1–70. 49

291
292 References

[ArhM1998] Arhangelskii, A.V. and Malykhin, V.I. Metrizability of topological

groups (in Russian). Vestnik Moskov. Univ. Ser. I Mat. Mekh. 91
(1996), 13–16; English translation in Moscow Univ. Math. Bull. 51
(1996), 9–11. 86
[ArhR2005] Arhangelskii, A.V. and Reznichenko, E.A. Paratopological and semi-
topological groups versus topological groups. Topol. Appl. 151
(2005), 107–119. 109
[ArhT2008] Arhangelskii, A. and Tkachenko, M. Topological Groups and Related
Structures. Atlantis Press, 2008. 256
[Avi1998] Avigad, J. An effective proof that open sets are Ramsey. Arch. Math.
Logic 37 (1998), 235–240. 122
[Bae1969] Baernstein, A. A non-linear Tauberian theorem in function theory.
Trans. Am. Math. Soc. 146 (1969), 87–105. 5
[Bae1974] Baernstein, A. A generalization of the cos πλ theorem. Trans. Am.
Math. Soc. 193 (1974), 181–197. 6
[BagW1997] Bagaria, J. and Woodin, H.W. ∆1n -sets of reals. J. Symbol. Logic 62
(1997), 1379–1428. 222, 231
[Bai1899] Baire, R. Thèse: Sur les fonctions de variable réelle. Ann. di Math. 3
(1899), 1–123. 34, 36
[Bai1909] Baire, R. Sur la représentation des fonctions discontinues (2me par-
tie). Acta Math. 32 (1909), 97–176. 34
[BajK1969] Bajšanski, B. and Karamata, J. Regular varying functions and the
principle of equicontinuity. Publ. Ramanujan Inst. 1 (1969), 235–
246. 278
[Bal1973] Balkema, A.A. Monotone Transformations and Limit Laws. Mathe-
matical Centre Tracts 45. Mathematisch Centrum, 1973. v+170 pp.
278
[Ban1920] Banach, S. Sur l’équation fonctionelle f (x + y) = f (x) + f (y).
Fund. Math. 1 (1920), 123–124. Reprinted in Collected Works I,
47–48, PWN, 1967 (Commentary by H. Fast, p. 314). 8, 190
[Ban1930] Banach, S. Über additive Maßfunktionen in abstrakten Mengen.
Fund. Mathe. 15 (1930), 97–101.
[Ban1931] Banach, S. Über metrische Gruppen. Studia Math. III (1931), 101–
113. Reprinted in Collected Works II, 401–411, PWN, 1979. vii+254
pp. 184
[Ban1932] Banach, S. Théorie des Opérations Linéaires. Monog. Mat. I (1932).
Reprinted in Collected Works II, PWN, 1979. 21, 162, 176, 184,
186, 187, 189, 190, 191, 232, 283
[BanT1924] Banach, S. and Tarski, A. Sur la décomposition des ensembles de
points en parties respectivement congruents. Fund. Math. 6 (1924),
244–277. 223
[Bana2004] Banakh, T. Cardinal characteristics of the ideal of Haar null sets.
Comment. Math. Univ. Carolin. 45 (2004), 119–137. 234
[BanaGJ2021] Banakh, T., Głab, S., Jabłońska, E. and Swaczyna, J. Haar-I sets:
Looking at small sets in Polish groups through compact glasses. Diss.
Math. 564 (2021), 1–105. 255
 References 293

[BanaJ2019] Banakh, T. and Jabłońska, E. Null-finite sets in topological groups

and their applications. Israel J. Math. 230 (2019), 361–386. 249,
264, 265, 266
[Bar1955] Bartle, R.G. Implicit functions and solutions of equations in groups.
Math. Z. 62 (1955), 335–346. 86
[BartFF2018] Bartoszewicz, A., Filipczak, M. and Filipczak, T. On supports of
probability Bernoulli-like measures. J. Math. Anal. Appl. 462 (2018),
26–35. 254, 255
[BartFN2011] Bartoszewicz, A., Filipczak, M. and Natkaniec, T. On Smítal prop-
erties. Topol. Appl. 158 (2011), 2066–2075. 255, 256, 258
[Barto1984] Bartoszyński, T. Additivity of measure implies additivity of category.
Trans. Am. Math. Soc. 28 (1984), 209–213. 230, 232
[Barto2000] Bartoszyński, T. A note on duality between measure and category.
Proc. Am. Math. Soc. 128 (2000), 2745–2748. 138, 228
[Barto2010] Bartoszyński, T. Invariants of measure and category. In M. Forman
and A. Kanamori (eds), Handbook of Set Theory, vol. 1. Springer,
2010, 491–556. 232
[BartoGJ1993] Bartoszyński, T., Goldstern, M., Judah, H. and Shelah, S. All meager
filters may be null. Proc. Am. Math. Soc. 117 (1993), 515–521. 228
[BartoJS1993] Bartoszyński, T., Judah, H. and Shelah, S. The Cichoń diagram. J.
Symbol. Logic 58 (1993), 401–423. 138
[BartoJ1995] Bartoszyński, T. and Judah, H. On the Structure of the Real Line.
A.K. Peters, 1995. 154, 222, 232, 233, 234
[Barw1977] Barwise, J. (ed). Handbook of Mathematical Logic. North-Holland,
1977. 301
[Bary1964] Bary, N.K. A Treatise on Trigonometric Series, vols. I, II. Pergamon,
1964. 120, 138
[Bau1975] Bauer, F. Aspects of modern potential theory. In Proc. Int.
Conf. Math. Vancouver, 1974, vol., 41–51. Canadian Mathematical
Congress, 1975. 115, 118
[Bec1960] Beck, A. A note on semi-groups in a locally compact group. Proc.
Am. Math. Soc. 11 (1960), 992–993. 239
[BecCS1958] Beck, A., Corson, H.H. and Simon, A.B. The interior points of the
product of two subsets of a locally compact group. Proc. Am. Math.
Soc. 9 (1958), 648–652. 239
[BeckK1996] Becker, H. and Kechris, A.S. The Descriptive Set Theory of Polish
Group Actions. London Math. Soc. Lect. Note Ser. 232. Cambridge
University Press, 1996. 233
[BelS1969] Bell, J.L. and Slomson, A.B. Models and Ultraproducts: An Intro-
duction. North-Holland, 1969. Reprinted by Dover, 2006. 62, 275
[Ber1963] Berge, C. Topological Spaces, Including a Treatment of Multi-valued
Functions, Vector Spaces and Convexity. Oliver and Boyd, 1963.
Reprinted by Dover, 1997. 54
[BergHW1997] Bergelson, V., Hindman, N. and Weiss, B. All-sums sets in (0,1] –
Category and measure. Mathematika 44 (1997), 61–87. 74, 76, 79
[Bert1996] Bertoin, J. Lévy Processes. Cambridge Tracts in Math. 121. Cam-
bridge University Press, 1996. 12
294 References

[Berz1975] Berz, E. Sublinear functions on R. Aequat. Math. 12 (1975), 200–206.

232
[BeuD1959] Beurling, A. and Deny, J. Dirichlet spaces. Proc. Nat. Acad. Sci. USA
45 (1959), 208–215. 120
[Bin1981] Bingham, N.H. Tauberian theorems and the central limit theorem.
Ann. Probab. 9 (1981), 221–231. 14, 17
[Bin1984a] Bingham, N.H. Tauberian theorems for summability methods of
random-walk type. J. London Math. Soc. (2) 30 (1984), 281–287.
16
[Bin1984b] Bingham, N.H. On Valiron and circle convergence. Math. Z. 186
(1984), 273–286. 15
[Bin1988] Bingham, N.H. Tauberian theorems for Jakimovski and Karamata–
Stirling methods. Mathematika 35 (1988), 216–224. 16
[Bin2007] Bingham, N.H. Regular variation and probability: The early years. J.
Comput. Appl. Math. 200 (2007), 357–363. 3
[Bin2010] Bingham, N.H. Finite additivity versus countable additivity. Elect. J.
His. Probab. Stat. 6 (2010), 35p. 240
[Bin2014] Bingham, N.H. The worldwide influence of the work of B. V. Gne-
denko. Theory Probab. Appl. 58 (2014), 17–24. 2
[Bin2015a] Bingham, N.H. On scaling and regular variation. Publ. Inst. Math.
Beograd (NS) 97 (111) (2015), 161–174. 22
[Bin2015b] Bingham, N.H. Hardy, Littlewood and probability. Bull. London
Math. Soc. 47 (2015), 191–201. 3
[Bin2019] Bingham, N.H. Riesz means and Beurling moving averages. In P.M.
Barrieu (ed). Risk and Stochastics: Ragnar Norberg Memorial Vol-
ume. World Scientific, 2019, 159–172; arXiv:1502.07494. 17
[BinF2010] Bingham, N.H. and Fry, J.M. Regression: Linear models in Statistics.
Springer Undergraduate Series in Mathematics (SUMS), Springer,
2010. 23
[BinG2015] Bingham, N.H. and Gashi, B. Logarithmic moving averages. J. Math.
Anal. Appl. 421 (2015), 1790–1802. 17
[BinG2017] Bingham, N.H. and Gashi, B. Voronoi means, moving averages and
power series. J. Math. Anal. Appl. 449 (2017), 682–696. 17
[BinGo1982a] Bingham, N.H. and Goldie, C.M. Extensions of regular variation,
I. Uniformity and quantifiers. Proc. London Math. Soc. 44 (1982),
473–496. 10, 145, 239
[BinGo1982b] Bingham, N.H. and Goldie, C.M. Extensions of regular variation,
II. Representations and indices. Proc. London Math. Soc. 44 (1982),
497–534. 11, 12, 239
[BinGo1983] Bingham, N.H. and Goldie, C.M. On one-sided Tauberian conditions.
Analysis 3 (1983), 159–188. 15, 17
[BGT1987] Bingham, N.H., Goldie, C.M. and Teugels, J.L. Regular Variation,
Encycl. Math. Appl. 27, Cambridge University Press, 1987 (2nd ed.
1989). 6, 19, 64, 78, 80, 81, 82, 140, 141, 144, 145, 239, 240, 282
[BinI1997] Bingham, N.H. and Inoue, A. The Drasin–Shea–Jordan theorem for
Fourier and Hankel transforms. Quart. J. Math. 48 (1997), 279–307.
5
 References 295

[BinI1999] Bingham, N.H. and Inoue, A. Ratio Mercerian theorems with appli-
cations to Hankel and Fourier transforms. Proc. London Math. Soc.
79 (1999), 626–648. 5
[BinI2000a] Bingham, N.H. and Inoue, A. Abelian, Tauberian and Mercerian
theorems for arithmetic sums. J. Math. Anal. Appl. 250 (2000), 465–
493. 6
[BinI2000b] Bingham, N.H. and Inoue, A. Tauberian and Mercerian theorems for
systems of kernels. J. Math. Anal. Appl. 252 (2000), 177–197. 7, 8,
17
[BinJJ2020] Bingham, N.H., Jabłońska, E., Jabłoński. W. and Ostaszewski, A.J.
On subadditive functions bounded above on a large set. Results Math.
75:58 (2020), 12p. xiii
[BinO2008] Bingham, N.H. and Ostaszewski, A.J. Generic subadditive functions.
Proc. Am. Math. Soc. 136 (2008), 4257–4266. 63, 123
[BinO2009a] Bingham, N.H. and Ostaszewski, A.J. Very slowly varying functions
II. Colloq. Math. 116 (2009), 105–117. 176
[BinO2009b] Bingham, N.H. and Ostaszewski, A.J. Beyond Lebesgue and Baire:
Generic regular variation. Colloq. Math. 116 (2009), 119–138. 63,
73, 137, 141, 146, 163, 278
[BinO2009c] Bingham, N.H. and Ostaszewski, A.J. The index theorem of regular
variation. J. Math. Anal. Appl. 358 (2009), 238–248. 64, 278, 283,
289
[BinO2009d] Bingham, N.H. and Ostaszewski, A.J. Infinite combinatorics and the
foundations of regular variation. J. Math. Anal. Appl. 360 (2009),
518–529. 145, 278
[BinO2009e] Bingham, N.H. and Ostaszewski, A.J. Automatic continuity: Subad-
ditivity, convexity, uniformity. Aequat. Math. 78 (2009), 257–270.
218
[BinO2009f] Bingham, N.H. and Ostaszewski, A.J. Infinite combinatorics on func-
tions spaces: Category methods. Publ. Inst. Math. Beograd (NS) 86
(100) (2009), 55–73. 117, 123
[BinO2010a] Bingham, N.H. and Ostaszewski, A.J. Automatic continuity by ana-
lytic thinning. Proc. Am. Math. Soc. 138 (2010), 907–919. 10, 11,
22, 145, 146
[BinO2010b] Bingham, N.H. and Ostaszewski, A.J. Regular variation without lim-
its. J. Math. Anal. Appl. 370 (2010), 322–338. 53, 62, 63, 65, 236,
237, 278
[BinO2010c] Bingham, N.H. and Ostaszewski, A.J. Topological regular variation.
I, Slow variation. Topol. Appl. 157 (2010), 1999–2013. 278, 281
[BinO2010d] Bingham, N.H. and Ostaszewski, A.J. Topological regular variation.
II, The fundamental theorems. Topol. Appl. 157 (2010), 2014–2023.
278, 283
[BinO2010e] Bingham, N.H. and Ostaszewski, A.J. Topological regular variation.
III, Regular variation. Topol. Appl. 157 (2010), 2024–2037. 278, 285
[BinO2010f] Bingham, N.H. and Ostaszewski, A.J. Kingman, category and com-
binatorics. In N.H. Bingham and C.M. Goldie (eds), Probability and
Mathematical Genetics: Sir John Kingman Festschrift. London Math.
296 References

Soc. Lecture Notes in Mathematics 378, Cambridge University Press,

2010, 135–168. 71, 72, 81, 241
[BinO2010g] Bingham, N.H. and Ostaszewski, A.J. Normed groups: Dichotomy
and duality. Dissertationes Math. 472 (2010), 138p. 74, 86, 89, 99,
138, 139, 140, 141, 162, 185, 199, 217, 221, 236, 285, 289
[BinO2010h] Bingham, N.H. and Ostaszewski, A.J. Beyond Lebesgue and Baire II:
Bitopology and measure-category duality. Colloq. Math. 121 (2010),
225–238. 73, 137, 145
[BinO2011a] Bingham, N.H. and Ostaszewski, A.J. Dichotomy and infinite com-
binatorics: The theorems of Steinhaus and Ostrowski. Math. Proc.
Cambridge Phil. Soc. 150 (2011), 1–22. 21, 62, 99, 141, 239, 255
[BinO2011b] Bingham, N.H. and Ostaszewski, A.J. Homotopy and the Kestelman–
Borwein–Ditor theorem. Canadian Math. Bull. 54 (2011), 12–20.
162
[BinO2013] Bingham, N.H. and Ostaszewski, A.J. Steinhaus theory and regu-
lar variation: De Bruijn and after. Indag. Math. (N. G. de Bruijn
Memorial Issue) 24 (2013), 679–692. 278, 281
[BinO2014] Bingham, N.H. and Ostaszewski, A.J. Beurling slow and regular
variation. Trans. London Math. Soc. 1 (2014), 29–56. 13, 278, 287
[BinO2015] Bingham, N.H. and Ostaszewski, A.J. Cauchy’s functional equation
and extensions: Goldie’s equation and inequality, the Gołab–Schinzel
˛
equation and Beurling’s equation. Aequat. Math. 89 (2015), 1293–
1310. 218
[BinO2016a] Bingham, N.H. and Ostaszewski, A.J. Beurling moving averages and
approximate homomorphisms. Indag. Math. 27 (2016), 601–633. 17,
62, 63, 288
[BinO2016b] Bingham, N.H. and Ostaszewski, A.J. The Steinhaus–Weil property
and its converse: Subcontinuity and amenability. arXiv:1607.00049.
(See BinO2020c,d, BinO2021a, BinO2022a.) 228, 235, 240, 242,
247, 249, 253
[BinO2017] Bingham, N.H. and Ostaszewski, A.J. Category–measure duality:
Jensen convexity and Berz sublinearity. Aequat. Math. 91 (2017),
801–836. 62, 232, 242
[BinO2018a] Bingham, N.H. and Ostaszewski, A.J. Additivity, subadditivity and
linearity: Automatic continuity and quantifier weakening. Indag.
Math. 29 (2018), 687–713. 10, 11, 232
[BinO2018b] Bingham, N.H. and Ostaszewski, A.J. Beyond Lebesgue and Baire
IV: Density topologies and a converse Steinhaus theorem. Topol.
Appl. 239 (2018), 274–292. 104, 234, 241, 242, 244, 256
[BinO2019a] Bingham, N.H. and Ostaszewski, A.J. Set theory and the analyst. Eur.
J. Math. 5 (2019), 2–48. 229, 240, 241, 268, 269, 270, 273, 274, 275,
276
[BinO2019b] Bingham, N.H. and Ostaszewski, A.J. Variants on the Berz sublin-
earity theorem. Aequat. Math. 93 (2019), 351–369. 232, 234
[BinO2019c] Bingham, N.H. and Ostaszewski, A.J. Beyond Haar and Cameron–
Martin: The Steinhaus support. Topol. App. 260 (2019), 23–56. 240,
247, 256, 259
 References 297

[BinO2020a] Bingham, N.H. and Ostaszewski, A.J. General regular variation, Popa
groups and quantifier weakening. J. Math. Anal. Appl. 483 (2020),
123610. 10, 11, 17, 18, 288
[BinO2020b] Bingham, N.H. and Ostaszewski, A.J. Sequential regular variation:
Extensions to Kendall’s theorem. Quart. J. Math. 71 (2020), 1171–
1200. 18, 19
[BinO2020c] Bingham, N.H. and Ostaszewski, A.J. The Steinhaus–Weil property
and its converse. I, Subcontinuity and amenability. Sarajevo Math. J.
16 (2020), 13–32. 228, 232, 234, 235, 242, 243, 246
[BinO2020d] Bingham, N.H. and Ostaszewski, A.J. The Steinhaus–Weil property
and its converse. II, The Simmons–Mospan converse. Sarajevo Math.
J. 16 (2020), 179–186; 235, 246, 252, 255, 256, 257
[BinO2021a] Bingham, N.H. and Ostaszewski, A.J. The Steinhaus–Weil property
and its converse. III, Weil topologies. Sarajevo Math. J.17 (2021),
129–142. 235, 240, 259, 290
[BinO2021b] Bingham, N.H. and Ostaszewski, A.J. Extremes and regular variation.
In L. Chaumont and E.A. Kyprianou (eds), A Lifetime of Excursions
through Random Walks and Lévy Processes (A volume in honour
of Ron Doney’s 80th birthday). Progr. Prob. 78, 2021b, 121–137,
Birkhäuser. 3, 13
[BinO2022a] Bingham, N.H. and Ostaszewski, A.J. The Steinhaus–Weil property
and its converse. IV, Other interior-point properties. Sarajevo Math.
J. 18 (2022), 203–210. 235
[BinO2022b] Bingham, N.H. and Ostaszewski, A.J. Homomorphisms from func-
tional equations: The Goldie equation II. ArXiv:1910.05816. 288
[BinO2022c] Bingham, N.H. and Ostaszewski, A.J. Homomorphisms from func-
tional equations: The Goldie equation III. ArXiv:1910.05817. 288
[BinO2022d] Bingham, N.H. and Ostaszewski, A.J. The Gołab–Schinzel˛ and
Goldie functional equations in Banach algebras. ArXiv:2105.07794.
288
[BinO2024] Bingham, N.H. and Ostaszewski, A.J. Parthasarathy, shift-
compactness and infinite combinatorics. Indian J. Pure Appl. Math.
55 (2024), 931–948. 73
[BinS1990] Bingham, N.H. and Stadtmüller, U. Jakimovski methods and almost-
sure convergence. In G.R. Grimmett and D.J.A. Welsh (eds), Dis-
order in Physical Systems (J.M. Hammersley Festschrift). Oxford
University Press, 1990, 5–18. 16
[BinT1986] Bingham, N.H. and Tenenbaum, G. Riesz and Valiron means and
fractional moments. Math. Proc. Cambridge Phil. Soc. 99 (1986),
143–149. 15, 17
[Bir1936] Birkhoff, G. A note on topological groups. Compos. Math. 3 (1936),
427–430. 86
[BirM1953] Birkhoff, G. and Mac Lane, S. A Survey of Modern Algebra, revised
ed., Macmillan, 1953, (1st ed. 1941). 272
[Bla1984] Blass, A. Existence of bases implies the Axiom of Choice. Contemp.
Math. 31 (1984), 31–33. 20
298 References

[Blo1976] Bloom, S. A characterization of B-slowly varying functions. Proc.

Am. Math. Soc. 54 (1976), 243–250. 13
[BluG1968] Blumenthal, R.M. and Getoor, R.K. Markov Processes and Potential
Theory. Academic Press, 1968. 117
[BobBC1981] Boboc, N., Bucur, Gh. and Cornea, A. Order and Convexity in Poten-
tial Theory: H-cones. Lecture Notes in Math. 853 (1981), Springer.
118, 120
[Bog1998] Bogachev, V.I. Gaussian Measures. Math. Surveys and Monographs
62, Am. Math. Soc., 1998. 240, 259, 260, 261
[Bog2007a] Bogachev, V.I. Measure Theory, I. Springer, 2007. xiii, 24, 25, 28,
29, 30, 31, 33, 73, 104, 126, 232, 244, 273, 274
[Bog2007b] Bogachev, V.I. Measure Theory, II. Springer, 2007. 25, 33, 240
[Bog2010] Bogachev, V.I. Differentiable Measures and the Malliavin Calculus.
Math. Surveys and Monographs 164, Am. Math. Soc., 2010. 243
[Bog2018] Bogachev, V.I. Negligible sets in infinite-dimensional spaces. Anal.
Math. 44 (2018), 299–323. 240
[BogS2017] Bogachev, V.I. and Smolyanov, O.G. Topological Vector Spaces and
Their Applications, Springer, 2017. 240
[BojK1963] Bojanic, R. and Karamata, J. On a class of functions of regular
asymptotic behaviour. Math. Res. Center Tech. Rep. 436 (1963),
Madison WI. Reprinted in Kar2009, pp. 545–569. 9
[Bol1998] Bollobás, B. Modern Graph Theory, Springer, 1998. 274
[Bor1989] Border, K.C. Fixed Point Theorems with Applications to Economics
and Game Theory. Cambridge University Press, 1989. 50
[Bore1909] Borel, E. Les probabilités dénombrables et leurs applications arith-
métiques. Rend. Circ. Mat. Palermo 27 (1909), 247–271. 225
[Borel1976] Borell, C. Gaussian Radon measures on locally convex spaces. Math.
Scand. 36 (1976), 265–284. 260
[BorwD1978] Borwein, D. and Ditor, S.Z. Translates of sequences in sets of positive
measure. Canadian Math. Bull. 21 (1978), 497–498. 74
[Bou1966] Bourbaki, N. Elements of Mathematics: General Topology. Parts 1
and 2. Hermann, Paris/Addison-Wesley, 1966. 86
[Bouz1993] Bouziad, A. The Ellis theorem and continuity in groups. Topol. Appl.
50 (1993), 73–80. 49, 176, 197
[Bouz1996] Bouziad, A. Every Čech-analytic Baire semitopological group is a
topological group. Proc. Am. Math. Soc. 124 (1996), 953–959. 197,
198, 242
[BraG1960] Bradford, J.C. and Goffman, C.Metric spaces in which Blumberg’s
theorem holds. Proc. Am. Math. Soc. 11 (1960), 667–670. 40, 127
[Bre1941] Brelot, M. Sur la théorie autonome des fonctions sousharmoniques.
Bull. Sci. Math. 65 (1941), 72–98. 116
[BreuS2011] Breuer, J. and Simon, B. Natural boundaries and spectral theory. Adv.
Math. 226 (2011), 4902–4920. 229
[Brez1991] Brezinski, C. History of Continued Fractions and Padé Approxi-
mants. Springer, 1991. 226
[Bru1971] Bruckner, A.M. Differentiation of integrals. Am. Math. Monthly 78
(9) (1971), Part II, ii+51 pp. 104
 References 299

[Brz1992] Brzdek,˛ J. Subgroups of the group Zn and a generalization of the

Gołab–Schinzel
˛ functional equation. Aequat. Math. 43 (1992), 59–
71. 260
[Bug2012] Bugeaud, Y. Distribution Modulo One and Diophantine Approxima-
tion. Cambridge Tracts in Math. 193, Cambridge University Press,
2012. 223
[Buk2011] Bukovský, L. The Structure of the Real Line. Monografie Matematy-
czne (New Series) 71, Birkhäuser, 2011. 234
[BurBI2001] Burago, D., Burago, Y. and Ivanov, S. A Course in Metric Geometry.
Graduate Studies in Mathematics 33, Am. Math. Soc., 2001. 86
[Burg1983a] Burgess, J.P. Classical hierarchies from a modern standpoint. I: C-
sets. Fund. Math. 115 (1983), 81–95. 62
[Burg1983b] Burgess, J.P. Classical hierarchies from a modern standpoint. II: r-
sets. Fund. Math. 115 (1983), 97–105. 62
[CabC2002] Cabello Sánchez, F. and Castillo, J.M.F. Banach space techniques
underpinning a theory for nearly additive mappings. Dissertationes
Math. (Rozprawy Mat.) 404 (2002), 73pp.
[CalMS2003] Calude, C.S. and Marcus, S. A topological characterization of random
sequences. Inform. Proc. Lett. 88 (2003), 245–250. 234
[CaoDP2010] Cao, J., Drozdowski, R. and Piotrowski, Z. Weak continuity proper-
ties of topologized groups. Czechoslovak Math. J. 60 (2010), 133–
148. 197
[CaoM2004] Cao J. and Moors, W.B. Separate and joint continuity of homomor-
phisms defined on topological groups. New Zealand J. Math. 33
(2004), 41–45. 197
[Car1945] Cartan, H. Théorie du potentiel newtonien: énergie, capacité, suites
de potentiels. Bull. Soc. Math. France 73 (1945), 74–106. 116
[Car1946] Cartan, H. Théorie générale du balayage en potentiel newtonien. Ann.
Inst. Fourier Grenoble 22 (1946), 221–280. 117
[Cha1933] Champernowne, D. G. The construction of decimals normal in the
scale of ten. J. London Math. Soc. 8 (1933), 254–260. 225
[ChanM1952] Chandrasekharan, K. and Minakshisundaram, S. Typical Means. Ox-
ford University Press, 1952. 17
[CharC2001] Charatonik, J.J. and Charatonik, W.J. The Effros metric. Topol. Appl.
110 (2001), 237–255. 176
[CharM1966] Charatonik, J.J. and Maćkowiak, T. Around Effros’ theorem. Trans.
Am. Math. Soc. 298 (1986), 579–602. 173
[Cho1953] Choquet, G. Theory of capacities. Ann. Inst. Fourier, Grenoble 5
(1953–54), 131–295. 116
[Cho1969] Choquet, G. Lectures on Analysis. Vol. I, Benjamin, 1969. 45
[Chr1972] Christensen, J.P.R. On sets of Haar measure zero in abelian Polish
groups. Israel J. Math. 13 (1973), 255–260. 241
[Chr1974] Christensen, J.P.R. Topology and Borel Structure. Descriptive Topol-
ogy and Set Theory with Applications to Functional Analysis and
Measure Theory. North-Holland, 1974. 47, 241, 251
[Chr1981] Christensen, J.P.R. Joint continuity of separately continuous func-
tions. Proc. Am. Math. Soc. 82 (1981), 455–461. 197
300 References

[Chu1968] Chung, K.-L. A Course in Probability Theory. Academic Press, 1968.

(3rd ed., 2001.) 117
[Chu1974] Chung, K.-L. Elementary Probability Theory with Stochastic Pro-
cesses. Springer, 1974. (4th ed., 2003.) 117
[Chu1982] Chung, K.-L. Lectures from Markov Processes to Brownian Motion.
Grundlehren Math. Wiss. 249, Springer, 1982. 116, 117, 118
[Chu1995] Chung, K.-L. Green, Brown and Probability. World Scientific, 1995.
118
[Cie1997] Ciesielski, K. Set-theoretic real analysis. J. Appl. Anal. 3 (1997),
143–190. 234
[CieJ1995] Ciesielski, K. and Jasiński, J. Topologies making a given ideal
nowhere dense or meager. Topol. Appl. 63 (1995), 277–298. 248
[CieL1990] Ciesielski, K. and Larson, L. The density topology is not generated.
Real Anal. Exchange 16 (1990/1991), 522–525. 167
[CieLO1994] Ciesielski, K., Larson, L. and K. Ostaszewski, I-Density Continuous
Functions. Mem. Am. Math. Soc. 107 (1994), no. 515. 167
[Coh1963] Cohen, P. The independence of the continuum hypothesis. Proc. Natl.
Acad. Sci. USA 50 (1963), 105–110. 268
[Cohn1989] Cohn, P.M. Algebra, Vol. 2, 2nd ed. Wiley, 1989. (1st ed. 1977.) 273
[Com1984] Comfort, W.W. Topological Groups. In KunV1984, Chapter 24. 94,
140, 236
[ComN1965] Comfort, W.W. and Negrepontis, S. The ring C(X ) determines the
category of X. Proc. Am. Math. Soc. 16 (1965), 1041–1045. 40, 41
[ComN1974] Comfort, W.W. and Negrepontis, S. The Theory of Ultrafilters.
Grundlehren Math. Wiss. 211, Springer, 1974. 64
[ConC1972] Constantinescu, C. and Cornea, A. Potential Theory on Harmonic
Spaces. Grundlehren Math. Wiss. 158, Springer, 1972. 116, 118
[Conw1990] Conway, J.B. A Course in Functional Analysis. 2nd ed. Graduate
Texts in Mathematics, 96 Springer, 1990. 182, 183
[Cro1957] Croft, H.T. A question of limits. Eureka 20 (1957), 11–13. 18
[CsiE1964] Csiszár, I. and Erdős, P. On the function g(t) = lim sup x→∞ ( f (x +
t) − f (x)). Magyar Tud. Akad. Mat. Kutató Int. Kőzl. A 9 (1964),
603–606. 145, 170
[Dal1978] Dales, H.G. Automatic continuity: A survey. Bull. London Math. Soc.
10(1978), 129–183. 185
[Dal2000] Dales, H.G. Banach Algebras and Automatic Continuity. London
Math. Soc. Monog. New Series, 24, Oxford University Press, 2000.
197
[DalW1987] Dales, H.G. and Woodin, W.H. An Introduction to Independence for
Analysts. London Math. Soc. Lecture Note Series, 115. Cambridge
University Press, 1987. 232
[DalW1996] Dales, H.G. and Woodin, W.H. Super-Real Fields. Totally Ordered
Fields with Additional Structure. London Math. Soc. Monog., New
Series, 17, Oxford University Press, 1996. 268
[Dar1875] Darboux, G. Sur la composition des forces en statiques. Bull. des Sci.
Math. 9 (1875), 281–288. 144
 References 301

[Darj2013] Darji, Udayan B. On Haar meager sets. Topol. Appl. 160 (2013),
2396–2400. 249, 261, 262
[Dav1952] Davies, R.O. Subsets of finite measure in analytic sets. Nederl. Akad.
Wetensch. Proc. Ser. A. 14 (1952), 488–489. 195
[DavJO1977] Davies, Roy O., Jayne, J.E., Ostaszewski, A.J. and Rogers, C.A.
Theorems of Novikov type. Mathematika 24 (1977), 97–114. 59
[Deb1986] Debs, G. Points de continuité d’une fonction séparément continue.
Proc. Am. Math. Soc. 97 (1986), 167–176. 47
[Deb1987] Debs, G. Points de continuité d’une fonction séparément continue II.
Proc. Am. Math. Soc. 99 (1987), 777–782. 48
[DebSR1987] Debs, G. and Saint Raymond, J. Ensembles d’unicité et d’unicité au
sens large, Ann. Inst. Fourier, Grenoble 37(1987) 217–239. 138
[Del1972] Dellacherie, C. Capacités et Processus Stochastiques, Ergebnisse
Math. 67, Springer, 1972. 117
[Del1980] Dellacherie, C. Un cours sur les ensembles analytiques, Part II. In
RogJD1980, pp. 183–316. 51
[DelM1975–1992] Dellacherie, C. and Meyer, P.-A. Probabilités et Potentiel, Ch. I–IV
(1975), Ch. V–VIII (1980), Ch. IX–XI (1983), Ch. XII–XVI (1987),
Ch. XVII–XXIV (with Maisonneuve, B.) (1992). Hermann, Paris.
118
[Den1915] Denjoy, A. Sur les fonctions dérivées sommable. Bull. Soc. Math.
France 43 (1915), 161–248. 107
[Dev1973] Devlin, K.J. Aspects of Constructibility. Lecture Notes Math. 354,
Springer, 1973. 53, 64
[Dev1977] Devlin, K.J. Constructibility. In Barw1977, p. 453–489. 270
[Dev1984] Devlin, K.J. Constructibility. Perspectives Math. Logic, Springer,
1984. 270, 277
[Dev1993] Devlin, K.J. The Joy of Sets: Fundamentals of Contemporary Set
Theory, 2nd ed. Springer, 1993. (1st ed. 1979.) 272, 273
[DezDD2006] Deza, E., Deza, M.M. and Deza, M. Dictionary of Distances, Else-
vier, 2006. 86
[DieS2014] Diestel, J. and Spalsbury, A. The Joys of Haar Measure, Grad. Studies
in Math. 150, Am. Math Soc., 2014. 242, 285
[Dij2005] Dijkstra, J.J. On homeomorphism groups and the compact-open
topology. Am. Math. Monthly 112 (2005), 910–912. 279
[Dod2004a] Dodos, P. Dichotomies of the set of test measures of a Haar-null set.
Israel J. Math. 144 (2004), 15–28. 253
[Dod2004b] Dodos, P. On certain regularity properties of Haar-null sets. Fund.
Math. 191 (2004), 97–109. 253
[Dod2009] Dodos, P. The Steinhaus property and Haar-null sets. Bull. Lon. Math.
Soc. 41 (2009), 377–384. 258
[Doo1984] Doob, J.L. Classical Potential Theory and its Probabilistic Counter-
part. Grundl. Math. Wiss. 262, Springer, 1984. 115, 116, 117, 118,
119, 120, 123
[DorFN2013] Dorais, F.G., Filipów, R. and Natkaniec, T. On some properties of
Hamel bases and their applications to Marczewski measurable func-
tions. Cent. Eur. J. Math. 11 (2013), 487–508. 22
302 References

[DouF1994] Dougherty, R. and Foreman, M. Banach–Tarski decompositions us-

ing sets with the Baire property. J. Am. Math. Soc. 7 (1994), 75–124.
138, 224
[Douw1984] van Douwen, E.K. The integers and topology. In K. Kunen and J.E.
Vaughan (eds), Handbook of Set-Theoretic Topology. North-Holland,
1984, 111–167. 234
[Douw1989] van Douwen, E.K. Fubini’s theorem for null sets. Am. Math. Monthly
96(8) (1989), 718–721. 228
[Douw1992] van Douwen, E. A technique for constructing honest locally compact
submetrizable examples. Topol. Appl. 47 (1992), 179–201. 145
[Dow1947] Dowker, C.H. Mapping theorems for non-compact spaces. Am. J.
Math. 69 (1947), 200–242. 135
[Dra1974] Drake, F.R. Set Theory: An Introduction to Large Cardinals, North-
Holland, 1974. 271, 274, 276, 277
[Dras1968] Drasin, D. Tauberian theorems and slowly varying functions. Trans.
Am. Math. Soc. 133 (1968), 333–356. 4
[Dras2010] Drasin, D. Baernstein’s thesis and entire functions with negative
zeros. Mat. Stud. 34 (2010), 160–167. 5
[DrasS1969] Drasin, D. and Shea, D.F. Asymptotic properties of entire functions
extremal for the cos π ρ theorem. Bull. Am. Math. Soc. 75 (1969),
119–122. 6
[DrasS1970] Drasin, D. and Shea, D.F. Complements to some theorems of Bowen
and Macintyre on the radial growth of entire functions with negative
zeros. In H. Shankar (ed), Mathematical Essays Dedicated to A.J.
Macintyre. Ohio University Press, 1970, 101–121. 4
[DrasS1972] Drasin, D. and Shea, D.F. Pólya peaks and the oscillation of positive
functions. Proc. Am. Math. Soc. 34 (1972), 403–411. 6
[DrasS1976] Drasin, D. and Shea, D.F. Convolution inequalities, regular variation
and exceptional sets. J. Analyse Math. 29 (1976), 232–293. 4, 10, 18
[Dud1989] Dudley, R.M. Real Analysis and Probability. Cambridge Studies in
Advanced Mathematics 74. Cambridge University Press, 2002. (1st
ed. 1989.) 29, 30, 253
[Dug1966] Dugundji, J. Topology. Allyn and Bacon, 1966. 35, 85, 176, 197
[Edg1990] Edgar, G.A. Measure, Topology and Fractal Geometry. Undergrad.
Texts in Math., Springer, 1990. 226
[Edr1969] Edrei, A., Locally Tauberian theorems for meromorphic functions
of lower order less than one. Trans. Am. Math. Soc. 140 (1969),
309–332. 5
[EdrF1966] Edrei, A. and Fuchs, W.H.J. Tauberian theorems for a class of mero-
morphic functions with negative zeros and positive poles. In Contem-
porary Problems in Anal. Functions (Proc. Internat. Conf. Erevan,
1965, Russian), 339–358. Nauka, 1966. 4, 5
[Eff1965] Effros, E.G. Transformation groups and C ∗ -algebras. Ann. Math. 81
(1965), 38–55. 172
[EleN2020] Elekes, M. and Nagy, D. Haar null and Haar meagre sets: A sur-
vey and new results. Bull. Lon. Math. Soc. 52 (2020), 561–619
(arXiv:1606.06607v2) 249
 References 303

[EleV2017] Elekes, M. and Vidyánsky, Z. Naively Haar null sets in Polish groups.
J. Math. Anal. Appl. 446 (2017), 193–200. 249
[Ell1974] Ellentuck, E. A new proof that analytic sets are Ramsey. J. Symbolic
Logic 39, 163-165. 121
[Elli1953] Ellis, R. Continuity and homeomorphism groups. Proc. Am. Math.
Soc. 4 (1953). 969–973. 183, 197, 219
[Elli1957] Ellis, R. A note on the continuity of the inverse, Proc. Am. Math.
Soc. 8 (1957). 372–373. 197
[Elli1969] Ellis, R. Lectures on Topological Dynamics, Benjamin, 1969. 288
[EmeFK1979] Emeryk, A., Frankiewicz, R. and Kulpa, W. On functions having the
Baire property. Bull. Acad. Polon. Sci. Sér. Sci. Math. 27 (1979),
489–491. 185
[Eng1969] Engelking, R. On closed images of the space of irrationals. Proc. Am.
Math. Soc. 21 (1969), 583–586. 211
[Eng1989] Engelking, R. General Topology. Heldermann, 1989. xiii, 25, 26, 27,
28, 35, 36, 41, 44, 56, 58, 85, 86, 120, 136, 153, 157, 162, 165, 193,
194, 208, 214, 215, 279, 282
[ErdKR1963] Erdős, P., Kestelman, H. and Rogers, C.A. An intersection property
of sets with positive measure, Coll. Math. 11 (1963), 75–80. 168
[Ess1975] Essén, R.R. The cos πλ Theorem. Lecture Notes in Math. 467,
Springer, 1975. 6
[Fal1985] Falconer, K.J. The Geometry of Fractal Sets. Cambridge Tracts in
Math. 85, Cambridge University Press, 1985. 226
[FalGO2015] Falconer, K., Gruber, Peter M., Ostaszewski, A.J. and Stuart, Trevor.
Claude Ambrose Rogers: 1 November 1920–5 December 2005. R.
Soc. Biogr. Mem. 61 (2015), 403–435. xiii
[Fed1969] Federer, H. Geometric Measure Theory. Grundl. Math. Wiss. 153,
Springer, 1969. 228
[Fel1966] Feller, W. An Introduction to Probability Theory and Its Applications,
Vol. II, Wiley, 1966. (2nd ed. 1971.) 2
[FenN1973] Fenstad, J.E. and Normann, D. On absolutely measurable sets. Fund.
Math. 81 (1973/74), 91–98. 62, 241
[FilW2011] Filipczak, M. and Wilczyński, W. Strict density topology on the
plane. Measure case. Rend. Circ. Mat. Palermo(2) 60 (2011), 113–
124. 256
[FisFK2014] Fischer, V., Friedman, S.S. and Khomskii, Y. Cichoń diagram, reg-
ularity properties and ∆13 sets of reals. Arch. Math. Logic 53 (2014),
695–729. 231
[ForK2010] Foreman, M. and Kanamori, A. (eds). Handbook of Set Theory.
Springer, 2010. 231, 271, 312, 315, 316
[ForW1991] Foreman, M. and Woodin, W.H. The generalized continuum hypoth-
esis can fail everywhere. Ann. Math. 133 (1991), 1–35. 269
[Fort1949] Fort Jr., M.K. A unified theory of semi-continuity. Duke Math. J. 16
(1949), 237–246. 242
[Fos1993] Fosgerau, M. When are Borel functions Baire functions? Fund. Math.
143 (1993), 137–152. 191
304 References

[Fra1982] Frankiewicz, R. On functions having the Baire property, II. Bull.

Acad. Polon. Sci. Sér. Sci. Math. 30 (1982), 559–560. 185
[FraK1987] Frankiewicz, R. and Kunen, K. Solution of Kuratowski’s problem
on function having the Baire property, I. Fund. Math. 128 (1987),
171–180. 185
[Fre1913] Fréchet, M., Pri la funkcia ekvacio f(x+y) = f(x) + f(y) (in Esperanto).
Enseignement Math. 15 (1913), 390–393 21
[Fre1914] Fréchet, M. Sur la notion de différentielle d’une fonction de ligne.
Trans. Am. Math. Soc. 15 (1914), 135–161. 21
[Fre1980] Fremlin, D.H. Čech-Analytic Spaces, (1980). Available from www1
.essex.ac.uk/maths/people/fremlin/n80l08.pdf. 57
[Fre1984] Fremlin, D.H. Consequences of Martin’s Axiom. Cambridge Tracts
in Mathematics 84, Cambridge University Press, 1984. 272
[Fre1987] Fremlin, D.H. Measure-additive coverings and measurable selectors.
Dissertationes Math. (Rozprawy Mat.) 260 (1987), 116pp. 185
[Fre2000a] Fremlin, D.H. Universally Kuratowski–Ulam spaces, (2000a).
Available from www1.essex.ac.uk/maths/people/fremlin/
preprints.htm. 149, 158
[Fre2000b] Fremlin, D.H. Measure Theory, 1. The Irreducible Minimum. Tor-
res Fremlin, 2000. Available from www1.essex.ac.uk/maths/
people/fremlin/mt1.2011/index.htm. 25
[Fre2001] Fremlin, D.H. Measure Theory, 2. Broad Foundations. Torres Frem-
lin, 2001. Available from www1.essex.ac.uk/maths/people/
fremlin/mt2.2016/index.htm. 25, 207
[Fre2002] Fremlin, D.H. Measure Theory, 3. Measure Algebras. Torres Frem-
lin, 2002. Available from www1.essex.ac.uk/maths/people/
fremlin/mt3.2012/index.htm. 25, 148, 149, 241
[Fre2003] Fremlin, D.H. Measure Theory, 4. Topological Measure Spaces. Part
I, II. Torres Fremlin, 2003. Available from www1.essex.ac.uk/
maths/people/fremlin/mt4.2013/index.htm. 25, 207, 272
[Fre2008] Fremlin, D.H. Measure Theory, 5. Set-Theoretic Measure Theory.
Part I, II. Torres Fremlin, 2008. Available from www1.essex.ac
.uk/maths/people/fremlin/mt5.2015/index.htm. 25, 33, 138, 222, 232
[FreG1995] Fremlin, D.H. and Grekas, S. Products of completion regular mea-
sures. Fund. Math. 147 (1995), 27–37. 154
[FreNR2000] Fremlin, D., Natkaniec, T. and Recław, I. Universally Kuratowski–
Ulam spaces. Fund. Math. 165 (2000), 239–247. 139, 148, 149, 159,
175, 228
[FriS2020] Friedman, S.D. and Schrittesser, D. Projective measure without pro-
jective Baire. Memoirs Am. Math. Soc. 267 No. 1298, 2020. 231
[Fro1960] Frolík, Z. Generalizations of the G δ -property of complete metric
spaces. Czechoslovak Math. J. 10 (1960), 359–379. 56, 58, 195
[Fro1970] Frolík, Z. Absolute Borel and Souslin sets. Pacific J. Math. 32 (1970),
663–683. 56
[FroH1981] Frolík, Z. and Holický, P. Decomposability of completely Suslin-
additive families. Proc. Am. Math. Soc. 82 (1981), 359–365. 209,
218
 References 305

[FroN1990] Frolík, Z. and Netuka, I. Čech completeness and fine topologies in

potential theory and real analysis. Expos. Math. 8 (1990), 81–89. 111
[Fuc1970] Fuchs, L. Infinite Abelian Groups, I. Pure and Applied Mathematics
36, Academic Press, 1970. 142
[Fug1971] Fuglede, B. The quasi topology associated with a countably subaddi-
tive set function. Ann. Inst. Fourier (Grenoble) 21 (1971), 123–169.
116, 123
[FukOT1994] Fukushima, M., Oshima, Y. and Takeda, M. Dirichlet Forms and
Markov Processes. Walter de Gruyter, 1994. (1st ed., Fukushima,
M., North-Holland, 1980.) 120
[Ful1968] Fuller, R.V. Relations among continuous and various non-continuous
functions. Pacific J. Math. 25 (1968), 495–509. 242, 243
[Gao2009] Gao, S. Invariant Descriptive Theory, CRC Press, 2009. 62
[GerK1970] Ger, R. and Kuczma, M. On the boundedness and continuity of convex
functions and additive functions. Aequat. Math. 4 (1970), 157–162.
123
[GhoM1985] Ghoussoub, N. and Maurey, B. Gδ -embeddings in Hilbert space. J.
Funct. Anal. 61 (1985), 72–97. 219
[GilJ1960] Gillman, L. and Jerison, M. Rings of Continuous Functions. Van
Nostrand, 1960. (Reprinted in Graduate Texts in Mathematics 43,
Springer, 1976.) 41
[Gli1961] Glimm, J. Locally compact transformation groups. Trans. Am. Math.
Soc. 101 (1961), 124–138. 172
[GneK1954] Gnedenko, B.V. and Kolmogorov, A.N. Limit Distributions for Sums
of Independent Random Variables. Addison-Wesley, 1954. 2
[God1940] Gödel, K. The Consistency of the Axiom of Choice and of the Gen-
eralized Continuum Hypothesis. Ann. Math. Studies 3, Princeton
University Press, 1940. 269, 270
[Gof1950] Goffman, C. On Lebesgue’s density theorem. Proc. Am. Math. Soc.
1 (1950), 384–388. 112
[Gof1962] Goffman, C. On the approximate limits of a real function. Acta Sci.
Math. (Szeged) 23 (1962), 76–78. 107
[Gof1975] Goffman, C. Everywhere differentiable functions and the density
topology. Proc. Am. Math. Soc. 51 (1975), 250. 112
[GofNN1961] Goffman, C., Neugebauer , C.J. and Nishiura, T. Density topology
and approximate continuity. Duke Math. J. 28 (1961), 497–505. 107,
109, 110, 113, 167, 256
[GofW1961] Goffman, C. and Waterman, D. Approximately continuous transfor-
mations. Proc. Am. Math. Soc. 12 (1961), 116–121. 107, 110, 167,
256
[GolSW2000] Goldstern, M., Schmeling, J. and Winkler, R. Metric, fractal dimen-
sional and Baire results on the distribution of subsequences. Math.
Nachr. 219 (2000), 97–108. 226
[Gos1985] Goswami, K.C. Density topology on R is not a Borel subset of its
Stone–Čech compactification. Indian J. Pure Appl. Math. 16 (1985),
45–48. 111
306 References

[Gow1970] Gowisankaran, C. Radon measures on groups. Proc. Am. Math. Soc.

25 (1970), 381–384. 243, 244
[Gow1972] Gowisankaran, C. Quasi-invariant Radon measures on groups. Proc.
Am. Math. Soc. 35 (1972), 503–506. 243, 244
[GraRS1990] Graham, R.L., Rothschild, B.L. and Spencer, J.H. Ramsey Theory,
2nd ed. Wiley, 1990. (1st ed. 1980.) 141
[Gre1828] Green, G. An Essay on the application of mathematical analysis to
the theories of electricity and magnetism. T. Wheelhouse, Notting-
ham, 1828. (Facsimile edition, Wazäta-Melins Aktiebolag, Göteberg,
1958.) 114
[Gro1963] de Groot, J. Subcompactness and the Baire category theorem. Nederl.
Akad. Wetensch. Proc. Ser. A 66 (1963), 761–767. 135
[Gros1987] Grosse-Erdmann, K.-G. Holomorphe Monster und Universelle Funk-
tionen. Mitt. Math. Sem. Giessen 176 (1987). 228
[Gros1989] Grosse-Erdmann, K.-G. An extension of the Steinhaus–Weil theo-
rem. Colloq. Math. 57 (1989), 307–317. 239
[Gros1999] Grosse-Erdmann, K.-G. Universal families and hypercyclic opera-
tors. Bull. Am. Math. Soc. 36 (1999), 345–381. 228
[Gru1984] Gruenhage, G. Generalized metric spaces. In KunV1984, Chapter
10. 44, 45, 49
[Gru1998] Gruenhage, G. Irreducible restrictions of closed mappings. Topol.
Appl. 85 (1998), 127Ű135.
[Haa1970] de Haan, L. On regular variation and its applications to the weak con-
vergence of sample extremes. Math. Centre Tracts 32, Amsterdam,
1970. 3, 9
[Hal1950] Halmos, P.R. Measure Theory, Van Nostrand, 1950. (Reprinted as
Graduate Texts in Math. 18, Springer, 1974.) 24, 25, 75, 148, 251,
253, 257
[Ham1905] Hamel, G. Eine Basis aller Zahlen und die unstetigen Lösungen der
Funktionalgleichung f (x + y) = f (x) + f (y). Math. Ann. 60 (1905),
459–462. 20
[Han2004] Hand, D. J. Measurement Theory and Practice: The World through
Quantification. Arnold. 22
[HansT1992] Hansel, G. and Troallic, J.-P. Quasicontinuity and Namioka’s theo-
rem. Topol. Appl. 46 (1992), 135–149. 197
[Hanse1971] Hansell, R.W. Borel measurable mappings for nonseparable metric
spaces. Trans. Am. Math. Soc. 161 (1971), 145–169. 39, 185, 208,
209, 217, 218
[Hanse1972] Hansell, R.W. On the nonseparable theory of Borel and Souslin sets.
Bull. Am. Math. Soc. 78 (1972), 236–241. 205
[Hanse1973a] Hansell, R.W. On the representation of nonseparable analytic sets.
Proc. Am. Math. Soc. 39 (1973), 402–408. 203, 205
[Hanse1973b] Hansell, R.W. On the non-separable theory of k-Borel and k-Souslin
sets. General Topol. Appl. 3 (1973), 161–195. 205, 208
[Hanse1974] Hansell, R.W. On characterizing non-separable analytic and extended
Borel sets as types of continuous images. Proc. London Math. Soc.
28 (1974), 683–699. 193, 208, 209, 210, 211, 218
 References 307

[Hanse1992] Hansell, R.W. Descriptive topology. In H. HuŽek and J. van Mill

(eds). . Elsevier, 1992, 275–315. 56, 58, 204, 209, 213, 215, 218
[Hanse1999] Hansell, R.W. Nonseparable analytic metric spaces and quotient
maps. Topol. Appl. 85 (1998), 143–152. 209, 219
[HanseJR1983] Hansell, R.W., Jayne, J.E. and Rogers, C.A. K-analytic sets. Mathe-
matika 30(2) (1983), 189–221. 57, 207
[Har1904] Hardy, G.H. Researches in the theory of divergent series and divergent
integrals. Quart. J. Math. 35 (1904), 22–66. (Reprinted in Collected
Papers of G. H. Hardy, Volume VI, Oxford University Press, 1974,
37–84.) 17
[Har1922] Hardy, G.H. A new proof of the functional equation for the zeta-
function. Matematisk Tidsskrift B (1922), 71–73. Reprinted in Col-
lected Papers of G. H. Hardy, Volume II.1, Oxford University Press.
227
[Har1949] Hardy, G.H. Divergent Series. Oxford University Press, 1949. 13,
14, 15, 17
[HarL1916] Hardy, G.H. and Littlewood, J.E. Theorems concerning the summa-
bility of series by Borel’s exponential method. Rend. Circ. Mat.
Palermo 41 (1916), 36–53. (Reprinted in Collected Papers of G.H.
Hardy, Volume VI, Oxford University Press, 1974, 609–628.) 16
[HarL1924] Hardy, G.H. and Littlewood, J.E. The equivalence of certain integral
means. Proc. London Math. Soc. 22 (1924), 60–63. (Reprinted in
Collected Papers of G.H. Hardy, Volume VI, Oxford University Press,
1974, 677–680.) 12
[HarR1915] Hardy, G.H. and Riesz, M. The General Theory of Dirichlet’s Series.
Cambridge Tracts in Math. 18, Cambridge University Press, 1915.
17
[HarW2008] Hardy, G.H. and Wright, E.M. An Introduction to the Theory of
Numbers, 6th ed. (revised D.R. Heath-Brown and J.H. Silverman),
Oxford University Press, 2008. 8, 10, 140
[Harm1988] Harman, G. Metric Number Theory. LMS Monographs 18, Oxford
University Press, 1988. 225
[HarrKL1990] Harrington, L.A., Kechris, A.S. and Louveau, A. A Glimm–Effros
dichotomy for Borel equivalence relations. J. Amer. Math. Soc. 3
(1990), 903–928. 173
[Hart1975] Hartman, S. Travaux de W. Sierpiński sur la théorie des ensembles et
ses applications IV. Mesure et catégorie. Congruence des ensembles.
In Sie1975, 20–25. 137
[HartM1958] Hartman, S. and Mycielski, J. On the imbedding of topological groups
into connected topological groups. Colloq. Math. 5 (1958), 167–169.
185
[HauP1952] Haupt, O. and Pauc, C. La topologie approximative de Denjoy en-
visagée comme vraie topologie. C. R. Acad. Sci. Paris 234 (1952),
390–392. 107, 167, 256
[Haw1975] Hawkes, J. On the potential theory of subordinators. Z. Wahrschein.
33 (1975), 113–132. 118
308 References

[HawoM1977] Haworth, R.C. and McCoy, R.C. Baire spaces. Dissertationes Math.
141 (1977). 37, 156
[HayP1970] Hayes, C.A. and Pauc, C.Y. Derivation and Martingales. Ergebnisse
Math. 49, Springer, 1970. 105
[Haym1964] Hayman, W.K. Meromorphic Functions. Oxford University Press,
1964. 5, 6
[Haym1970] Hayman, W.K. Some examples related to the cos π ρ theorem. In
H. Shankar (ed), Mathematical Essays Dedicated to A.J. Macintyre.
Ohio University Press, 1970, 149–170. 6
[Hei1971] Heiberg, C. A proof of a conjecture by Karamata. Publ. Inst. Math.
Beograd (NS) 12 (26), 41–44. 10
[Hel1969] Helms, L.L. Introduction to Potential Theory. Wiley, 1969. 115, 116
[HewR1979] Hewitt, E. and Ross, K.A. Abstract Harmonic Analysis, I. Structure
of Topological Groups, Integration Theory, Group Representations,
2nd ed. Grundl. Math. Wiss. 115, Springer, 1979. 173, 191, 253, 257
[Hey1977] Heyer, H. Probability Measures on Locally Compact Groups. Ergeb-
nisse Math. 94, Springer, 1977. 186
[Hil1974] Hildenbrandt, W. Core and Equilibria of a Large Economy. Princeton
University Press, 1974. 54
[HillP1957] Hille, W. and Phillips, R.S. Functional Analysis and Semi-Groups.
Am. Math. Soc. Colloq. Publ. 31, Am. Math. Soc., 1957. 239
[HinS1998] Hindman, N. and Strauss, D. Algebra in the Stone-Čech Compact-
ification. Theory and Applications. De Gruyter Expos. Math. 27,
Walter de Gruyter, 1998 (2nd revised and extended ed., de Gruyter,
2012). 64, 274
[Hof1980] Hoffmann-Jørgensen, J. Automatic continuity. In RogJD1980, Part
3.2, 337–398. 175, 177, 185, 187, 197
[HolN2012] Holá, L. and Novotný, B. Subcontinuity. Math. Slov. 62 (2012), 345–
362. 242
[Holi2010] Holický, P. Preservation of completeness by some continuous maps.
Topol. Appl. 157 (2010), 1926–1930. 220
[HoliP2010] Holický, P. and Pol, R. On a question by Alexey Ostrovsky concerning
preservation of completeness. Topol. Appl. 157 (2010), 594–596.
196, 216, 219
[HruA2005] Hrušák, M. and Zamora Avilés, B. Countable dense homogeneity of
definable spaces. Proc. Am. Math. Soc. 133 (2005), 3429–3435. 56
[HulLM2005] Hult, H., Lindskog, F., Mikosch, T. and Samorodnitsky, G. Functional
large deviations for multivariate regularly varying random walks.
Ann. Appl. Probab. 15(4) (2005), 2651–2680. 278
[HurW1941] Hurewicz, W. and Wallman, H. Dimension Theory. Princeton Uni-
versity Press, 1941. 226
[HydLO2010] Hyde, J., Laschos, V., Olsen, L., Petrykiewicz, I. and Shaw, A. Iter-
ated Cesàro averages, convergence, frequencies of digits and Baire
category. Acta Arith. 144 (2010), 287–293. 225
[IhoS1989] Ihoda, J.I. (Judah, H.I.) and Shelah, S. ∆12 -sets of reals. Ann. Pure
Appl. Logic 42 (1989), 207–223. 230
 References 309

[IonT1961] Ionescu Tulcea, A. and Ionescu Tulcea, C. On the lifting property, I.

J. Math. Anal. Appl. 3 (1961), 537–546. 104
[IonT1969] Ionescu Tulcea, A. and Ionescu Tulcea, C. Topics in the Theory of
Lifting. Ergebnisse Math. 48, Springer, 1969. 104
[Itz1972] Itzkowitz, G.L. A characterization of a class of uniform spaces that
admit an invariant integral. Pacific J. Math. 41 (1972), 123–141. 196
[Jab2015] Jabłońska, E. Some analogies between Haar meager sets and Haar
null sets in abelian Polish groups. J. Math. Anal. Appl. 421 (2015),
1479–1486. 249, 263, 264
[Jab2016] Jabłońska, E. A theorem of Piccard’s type in abelian Polish groups.
Anal. Math. 42 (2016), 159–164. 255
[Jak1959] Jakimovski, A. A generalization of the Lototsky method of summa-
tion. Michigan Math. J. 6, 277–290. 16
[Jam1943] James, R.C. Linearly arc-wise connected topological Abelian groups.
Ann. Math. 44, (1943), 93–102. 86
[JamMW1947] James, R.C., Michal, A.D. and Wyman, M. Topological Abelian
groups with ordered norms. Bull. Am. Math. Soc. 53 (1947), 770–
774. 86
[Jan1997] Janson, S. Gaussian Hilbert Spaces. Cambridge Tracts in Mathemat-
ics 129. Cambridge University Press, 1997. 260
[Jav1968] Javor, P. On the general solution of the functional equation f (x +
y f (x)) = f (x) f (y), Aequat. Math. 1 (1968), 235–238. 288
[Jec1973] Jech, Th.J. The Axiom of Choice. North-Holland, 1973. 20, 270
[Jec1997] Jech, Th.J. Set Theory. Springer, 1977. 277
[Jen1972] Jensen, R.B. The fine structure of the constructible hierarchy. Ann.
Math. Logic 4 (1972), 229–308. 145
[JimSS2019] Jiménez-Garrido, J., Sanz, J. and Schindl, G. Indices of O-regular
variation for weight functions and weight sequences. Rev. R. Acad.
Cienc. Ex. Fac. Nat. Ser. A (RACSAM) 113 (2019), 3659–3697. 7
[Joh1982] Johnstone, P.J. Stone Spaces. Cambridge Studies in Advanced Math-
ematics 3. Cambridge University Press, 1982. 149
[Jon1942a] Jones, F.B. Connected and disconnected plane sets and the functional
equation f (x + y) = f (x) + f (y). Bull. Am. Math. Soc. 48 (1942),
115–120. 145
[Jon1942b] Jones, F.B. Measure and other properties of a Hamel basis. Bull. Am.
Math. Soc. 48 (1942), 472–481. 21
[Jon1975] Jones, F.B. Use of a new technique in homogeneous continua. Hous-
ton J. Math. 1 (1975), 57–61. 173
[Jor1974] Jordan, G.S. Regularly varying functions and convolutions with real
kernels. Trans. Am. Math. Soc. 194 (1974), 177–194. 5
[JudR1993] Judah, H. and Rosłanowski, A. On Shelah’s amalgamation. In Set
Theory of the Reals (Ramat Gan, 1991), 385–414, Israel Math. Conf.
Proc., 6, Bar-Ilan Univ., Ramat Gan, 1993. 233
[JudS1993a] Judah, H.I. (Ihoda, J.I.) and Shelah, S. ∆13 -sets of reals. J. Symbolic
Logic 58 (1993), 72–80. 231
[JudS1993b] Judah, H.I. and Shelah, S. Baire property and axiom of choice. Israel
J. Math. 84 (1993), 435–450. 232, 233
310 References

[JudSp1997] Judah, H.I. (Ihoda, J.I.) and Spinas, O. Large cardinals and projective
sets. Arch. Math. Logic 36 (1997), 137–155. 229, 233, 269
[Kah1985] Kahane, J.-P. Some Random Series of Functions, 2nd ed. Cambridge
Studies in Advanced Mathematics 5, Cambridge University Press.
(1st ed. 1968, Heath, MA.) 224, 227, 229, 234
[Kah1997] Kahane, J.-P. A century of interplay between Taylor series, Fourier
series and Brownian motion. Bull. London Math. Soc. 29 (1997),
257–279. 223, 224, 227
[Kah2000] Kahane, J.-P. Baire’s category theorem and trigonometric series. J.
Anal. Math. 80 (2000), 143–182. 138, 224, 227, 229
[Kah2001] Kahane, J.-P. Probabilities and Baire’s theory in harmonic analy-
sis. In J. S. Byrnes (ed), Twentieth Century Harmonic Analysis, A
Celebration. Kluwer, 2001, 57–72. 138, 223, 224, 227, 229, 234
[KahQ1997] Kahane, J.-P. and Queffélec, H. Ordre, convergence et sommabilités
des séries de Dirichlet. Ann. Inst. Fourier 47 (1997), 485–529. 227
[Kak1936] Kakutani, K. Über die Metrisation der topologischen Gruppen. Proc.
Imp. Acad. Tokyo 12 (1936), 82–84. (Reprinted in Selected Papers,
1, R.R. Kallman (ed), 60–62. Birkhäuser, 1986.) 86, 91
[KalK2015] Kalemba, P. and Kucharski, A. Universally Kuratowski–Ulam spaces
and open–open games. Ann. Math. Siles. 29 (2015), 421–427. 149
[Kall2002] Kallenberg, O. Foundations of Modern Probability, 2nd ed., Springer,
2002. (1st ed. 1997.) 240
[KaltPR1984] Kalton, N.J., Peck, N.T. and Roberts, J.R. An F-space Sampler, Lon-
don Math. Soc. Lect. Notes Ser. 89, Cambridge University Press,
1984. 174
[Kan2003] Kanamori, A. The Higher Infinity. Large Cardinals in Set Theory
from their Beginnings, 2nd ed. Springer, 2003. (1st ed. 1994.) 62,
232, 277
[Kan2009] Kanamori, A. Set theory from Cantor to Cohen. In A. Irvine (ed),
Handbook of the Philosophy of Scienc. Philosophy of Mathematics,
Chapter 10. North-Holland, 2009. 36
[KanM1978] Kanamori, A. and Magidor, M. The evolution of large cardinal axioms
in set theory. Higher Set Theory (Proc. Conf., Math. Forschungsinst.,
Oberwolfach, 1977), pp. 99–275, Lecture Notes in Math. 669,
Springer, 1978. 232
[KaniP1975] Kaniewski, J. and Pol, R. Borel-measurable selectors for compact-
valued mappings in the non-separable case. Bull. Acad. Polon. Sci.
Sér. Sci. Math. 23 (1975), 1043–1050. 209, 218
[KanoL2003] Kanovei, B.G. and Lyubetskii, V.A. On some classical problems of
descriptive set theory. Russian Math. Surveys 58 (2003), 839–927.
230
[Kar2009] Karamata, J. Selected Papers, V. Marić (ed). Zavod za Udz̆benike,
Beograd, 2009. 8, 298
[KatS1974] Katznelson, Y. and Stromberg, K. Everywhere differentiable,
nowhere monotone, functions. Am. Math. Monthly 81 (1974), 349–
354. 113
 References 311

[Kau1967] Kaufman, R. A functional method for linear sets. Israel J. Math. 5

(1967), 185–187. 223
[Kau1995] Kaufman, R. Thin sets, differentiable functions and the category
method. In J. Fourier Anal. Appl., Special Issue in honour of J.-P.
Kahane, 311–316. CRC Press, 1995. 223
[Kec1994] Kechris, A.S. Topology and descriptive set theory. Topol. Appl. 58
(1994), 195–222. 175
[Kec1995] Kechris, A.S. Classical Descriptive Set Theory. Graduate Texts in
Mathematics 156, Springer, 1995. xiii, 36, 38, 46, 47, 48, 61, 62, 79,
117, 121, 123, 154, 162, 167, 196, 203, 231, 232, 240, 241, 268, 274
[KecL1987] Kechris, A.S. and Louveau, A. Descriptive Set Theory and the Struc-
ture of Sets of Uniqueness. London Math. Soc. Lecture Note Series
128, Cambridge University Press, 1987. 138, 228
[KecS1985] Kechris, A.S. and Solovay, R.M. On the relative consistency strength
of determinacy hypotheses. Trans. Am. Math. Soc. 290(1) (1985),
179–211. 232
[Kel1955] Kelley, J.L. General Topology. Van Nostrand, 1955. (2nd ed.
Springer, 1975.) 36, 42, 59, 86, 136, 201, 219, 282
[Kell1953] Kellogg, O.D. Foundations of Potential Theory. Dover, 1953. (Orig-
inally published in Grundlehren Math. Wiss. 31 (1929), reprinted
1967, Springer.) 114, 115
[Kem1957] Kemperman, J.H.B. A general functional equation. Trans. Am. Math.
Soc. 86 (1957), 28–56. 74, 241
[Ken1968] Kendall, D.G. Delphic semigroups, infinitely divisible regenerative
phenomena and the arithmetic of p-functions. Z. Wahrschein vew.
Geb. 9 (1968), 163–195. (Reprinted in Stochastic Analysis, E.F. Hard-
ing and D.G. Kendall (eds), 73–114. Wiley, 1973.) 18
[Kes1947a] Kestelman, H. The convergent sequences belonging to a set. J. Lon-
don Math. Soc. 22 (1947), 130-136. 73, 162
[Kes1947b] Kestelman, H. On the functional equation f (x + y) = f (x) + f (y).
Fund. Math. 34 (1947), 144–147. 145
[Kha1998] Kharazishvili, A.B. Transformation Groups and Invariant Measures.
Set-Theoretic Aspects. World Scientific, 1998. 240
[Kha2004] Kharazishvili, Alexander. Nonmeasurable Sets and Functions. Else-
vier, 2004. 22, 172
[Kha2018] Kharazishvili, A.B. Strange Functions in Real Analysis, 3rd ed. Chap-
man and Hall, 2018. (2nd ed. 2006, 1st ed. 2000.) 113
[Kha2019] Kharazishvili, A. Some remarks on the Steinhaus property for in-
variant extensions of the Lebesgue measure. Eur. J. Math. 5 (2019),
81–90. 256
[KhiK1925] Khintchin, A.Y. and Kolmogorov, A.N. Über Konvergenz von Reihen,
deren Glieder durch den Zufall bestimmt werden. Mat. Sbornik 32
(1925), 668–677. 224
[Khr2008] Khrushchev, S. Orthogonal Polynomials and Continued Fractions,
from Euler’s Point of View. Encycl. Math. Appl. 122, Cambridge
University Press, 2008. 225
312 References

[Kin1963] Kingman, J.F.C. Ergodic properties of continuous-time Markov pro-

cesses and their discrete skeletons. Proc. London Math. Soc. 13
(1963), 593–604. 78, 82, 134
[Kin1964] Kingman, J.F.C. A note on limits of continuous functions. Quart. J.
Math. 15 (1964), 279–282. 18, 78, 82
[Kle1952] Klee, V.L. Invariant metrics in groups (solution of a problem of
Banach). Proc. Am. Math. Soc. 3 (1952), 484–487. 86, 89, 285
[Klei1977] Kleinberg, E.M. Infinitary Combinatorics and the Axiom of Deter-
minateness. Lecture Notes in Math. 612, Springer, 1977. 231
[Kne1963] Kneebone, G.T. Mathematical Logic and the Foundations of Math-
ematics. An Introductory Survey. Van Nostrand 1963. (Reprinted,
Dover, 2001). 269
[Kod1941] Kodaira, K. Über die Beziehung zwischen den Massen und den
Topologien in einer Gruppe. Proc. Phys.-Math. Soc. Japan 23 (1941),
67–119. 138
[KoeW2010] Koellner, P. and Woodin, W.H. Large Cardinals from Determinacy.
In ForK2010. 232
[Kol1934] Kolmogorov, A.N. Zur Normierbarkeit eines allgemeinen topologis-
chen linearen Raumes. Studia Math. 5 (1934), 29–33. 86
[Kom1971] Kominek, Z. On the sum and difference of two sets in topological
vector spaces. Fund. Math. 71 (1971), 165–169. 239
[Kom1981] Kominek, Z. On the continuity of Q-convex and additive functions.
Aequationes Math. 23 (1981), 146–150. 22
[Kom1988] Kominek, Z. On an equivalent form of a Steinhaus theorem, Math.
(Cluj) 30 (53)(1988), 25–27. 255, 259
[Komj1988] Komjáth, P. Large small sets. Colloq. Math. 56 (1988), 231–233. 74
[KomT2008] Komjáth, P. and Totik,V. Ultrafilters. Am. Math. Month. 115 (2008),
33–44.
[Kor2004] Korevaar, J. Tauberian Theory: A Century of Developments. Grundl.
Math. Wiss. 329, 2004, Springer. 7, 13, 14, 15, 16
[Korn1995] Körner, T.W. Kahane’s Helson curve. In J. Fourier Anal. Appl., Spe-
cial Issue in honour of J.-P. Kahane, 325–346. CRC Press, 1995.
223, 227
[Kro1974] Krom, M.R. Cartesian products of metric Baire spaces. Proc. Am.
Math. Soc. 42 (1974), 588–594. 160
[KucP2007] Kucharski, A. and Plewik, S. Game approach to universally
Kuratowski–Ulam spaces. Topol. Appl. 154 (2007), 85–92. 149
[Kucz1985] Kuczma, M. An Introduction to the Theory of Functional Equa-
tions and Inequalities. Cauchy’s Functional Equation and Jensen’s
Inequality. PWN, 1985. 20, 21, 123, 144, 146, 241, 260
[KuczS1976] Kuczma, M. and Smítal, J. On measures connected with the Cauchy
equation. Aequationes Math. 14(3) (1976), 421–428. 259
[Kun1970] Kunen, K. Some applications of iterated ultrapowers in set theory.
Ann. Math. Logic 1 (1970), 179–227. 277
[Kun1983] Kunen, K. Set Theory. An Introduction to Independence Proofs. Stud-
ies in Logic and Foundations of Mathematics 102, North-Holland,
1983. 268
 References 313

[Kun2011] Kunen, K. Set Theory. Studies in Logic (London) 34. College Publi-
cations, London, 2011. 268
[KunV1984] Kunen, K. and Vaughan, J.E. Handbook of Set-Theoretic Topology.
North-Holland, 1984. 25, 300, 306, 322
[Kur1924] Kuratowski, C. Sur les fonctions représentables analytiquement et les
ensembles de première catégorie. Fund. Math. 5 (1924), 75–86. 184
[Kur1966] Kuratowski, K. Topology, I. PWN, 1966. 38, 42, 122, 184, 195, 224
[Kur1968] Kuratowski, K. Topology, II. PWN, 1968. 186, 279
[KurM1968] Kuratowski, K. and Mostowski, A. Set Theory. North Holland, 1968.
199
[KurU1932] Kuratowski, C. and Ulam, S. Quelques propriétés topologiques du
produit combinatoire. Fund. Math. 19 (1932), 247–251. 148
[LabR1995] Łabedzki, G. and Repický, M. Hechler reals. J. Symbolic Logic 60
(1995), 444–458. 122
[Lac1998] Laczkovich, M. Analytic subgroups of the reals. Proc. Am. Math.
Soc. 126 (1998), 1783–1790. 140, 141
[Lar2010] Larson, P.B. A brief history of determinacy. In A.S. Kechris, B. Löwe,
J.R. Steel (eds), The Cabal Seminar Vol. 4. Lecture Notes in Logic,
49, Cambridge University Press, 2010, 3–60. 232
[LeP1973] LePage, R.D. Subgroups of paths and reproducing kernels. Ann.
Prob. 1 (1973), 345–347. 260
[Lev1983] Levi, S. On Baire cosmic spaces. In General Topology and Its Rela-
tions to Modern Analysis and Algebra, V, (Prague, 1981), pp. 450–
454. Sigma Ser. Pure Math. 3, Heldermann, 1983. 56, 192, 193, 196,
198
[LevS1967] Lévy, A. and Solovay, R.M. Measurable cardinals and the Continuum
Hypothesis. Israel J. Math. 5 (1967), 234–248. 274
[LiZ2017] Li, R. and Zsilinszky, L. More on products of Baire spaces. Topol.
Appl. 230 (2017), 35–44. 160
[Lio1851] Liouville, J. Sur les classes très-étendues des quantités dont la valeur
n’est ni algébrique, ni même réductible à des irrationelles algébriques.
J. Math. Pures et Appliquées 16 (1851), 133–142. 223
[LiuR1968] Liu, T.S. and van Rooij, A. Transformation groups and absolutely
continuous measures. Indag. Math. 71 (1968), 225–231. 243
[LiuRW1970] Liu, T.S., van Rooij, A. and Wang, J.-K. Transformation groups and
absolutely continuous measures II. Indag. Math. 73 (1970), 57–61.
243
[Lit1944] Littlewood, J.E. Lectures on the Theory of Functions. Oxford Uni-
versity Press, 1944. 24
[LoeW2000] Loewen, P.D. and Wang, Xianfu. Typical properties of Lipschitz
functions. Real Anal. Exchange 26 (2000), 717–726. 228
[Lou1976] Louveau, A. Une méthode topologique pour l’étude de la propriété
de Ramsey. Israel J. Math. 23 (1976), 97–116. 121
[Lou1980] Louveau, A. A separation theorem for Σ11 sets. Trans. Am. Math. Soc.
260 (1980), 363–378. 121
[Low1997] Lowen, R. Approach Spaces. The Missing Link in the Topology–
Uniformity–Metric Triad. Oxford University Press, 1997. 87
314 References

[LukMZ1986] Lukeš, J. Malý, M., and Zajíček, L. Fine Topology Methods in Real
Analysis and Potential Theory. Lecture Notes Math. 1189, Springer,
1986. 72, 101, 105, 109, 110, 111, 117, 123, 124, 126, 127, 130, 256
[Lus1917] Lusin (Luzin), N.N. Sur la classification de M. Baire. Comptes Rendus
161 (1917), 91–94. 51
[Lus1930] Lusin (Luzin), N.N. Leçons sur les Ensembles Analytiques. Gauthier-
Villars, 1930. 51
[LusS1923] Lusin (Luzin), N.N. and Sierpiński, W. Sur un ensemble non mea-
surable B. J. Math. (NS) 2 (1923), 53–72. 51
[Mac1957] Mackey, G.W. Borel structure in groups and their duals. Trans. Am.
Math. Soc. 85 (1957), 134–165. 253
[Mah1958] Maharam, D. On a theorem of von Neumann. Proc. Am. Math. Soc.
9 (1958), 987–994. 104
[Mar1935] Szpilrajn, E. (Marczewski, E.) Sur une classe de fonctions de M.
Sierpiński et la classe correspondante d’ensembles. Fund. Math. 24
(1935), 17–34. 22
[MarS1949] Marczewski, E. and Sikorski, R. Remarks on measure and category.
Colloq. Math. 2 (1949), 13–19. 33
[MartK1980] Martin, D.A. and Kechris, A.S. Infinite games and effective descrip-
tive set theory. In RogJD1980, Part 4. 61, 121
[MartS1970] Martin, D.A. and Solovay, R.M. Internal Cohen extensions. Ann.
Math. Logic 2 (1970), 143–178. 272
[Marti1961] Martin, N.F.G. Generalized condensation points. Duke Math. J. 28
(1961), 507–514. 127, 130, 256
[Marti1964] Martin, N.F.G. A topology for certain measure spaces. Trans. Am.
Math. Soc. 112 (1964), 1–18. 105, 106, 107, 256
[Mat1979] Mathias, A.R.D. Surrealist landscape with figures (a survey of recent
results in set theory). Period. Math. Hung. 10 (1979), 109–175. 234
[MatoZ2003] Matoŭsková, E. and Zelený, M. A note on intersections of non–Haar
null sets. Colloq. Math. 96 (2003), 1–4. 255, 257
[Mau1981] Mauldin, D. The Scottish Book. Birkhäuser, 1981. 46
[Med2022] Medini, A. On the scope of the Effros theorem. Fund. Math. 258
(2022), 211–223. 173
[MeeS2001] Meerschaert, M.S. and Scheffler, H.-P. Limit Distributions for Sums
of Independent Random Vectors: Heavy Tails in Theory and Practice.
Wiley, 2001. 278
[Meh1964] Mehdi, M.R. On convex functions. J. London Math. Soc. 39 (1964),
321–326. 19, 21, 186, 283
[Mey1966] Meyer, P.-A. Probability and Potentials. Blaisdell, 1988. 117
[MeyK1949] Meyer-König, W. Untersuchungen über einige verwandte Limi-
tierungsverfahren. Math. Z. 52 (1949), 257–304. 14
[Mic1940] Michal, A.D. Differentials of functions with arguments and values
in topological abelian groups. Proc. Nat. Acad. Sci. USA 26 (1940),
356–359. 86
[Mic1947] Michal, A.D. Functional analysis in topological group spaces. Math.
Mag. 21 (1947), 80–90. 86
 References 315

[Mich1982] Michael, E. On maps related to σ-locally finite and σ-discrete col-

lections of sets. Pacific J. Math. 98 (1982) 139–152. 208
[Mich1986] Michael, E. A note on completely metrizable spaces. Proc. Am. Math.
Soc. 96 (1986), 513–522. 195, 219
[Mich1991] Michael, E. Almost complete spaces, hypercomplete spaces and re-
lated mapping theorems. Topol. Appl. 41 (1991), 113–130. 182, 195,
213
[Mil2004] van Mill, J. A note on the Effros Theorem. Am. Math. Monthly 111
(2004), 801–806. 59, 173, 175, 176, 177, 178
[Mil2008] van Mill, J. Homogeneous spaces and transitive actions by Polish
groups. Israel J. Math. 165 (2008), 133–159. 173, 179
[Mil2009] van Mill, J. Analytic groups and pushing small sets apart. Trans. Am.
Math. Soc. 361 (2009), 5417–5434. 175
[MilP1986] van Mill, J. and Pol, R. The Baire category theorem in products of
linear spaces and topological groups. Topol. Appl. 22 (1986), 267–
282. 139, 148, 175, 228
[Mill1989] Miller, A.W. Infinite combinatorics and definability. Ann. Pure Appl
Math. Logic 41(1989), 179-203. (See also updated web version at:
www.math.wisc.edu/\char126\relaxmiller/res/.) 145
[Mill1995] Miller, A.W. Descriptive Set Theory and Forcing. Springer, 1995.
122, 145
[MillP2000] Miller, A.W. and Popvassilev, S.G. Vitali sets and Hamel bases that
are Marczewski measurable. Fund. Math. 166 (2000), 269–279. 22
[Mille1989] Miller, H.I. Generalization of a result of Borwein and Ditor. Proc.
Am. Math. Soc. 105 (1989), 889–893. 74, 162
[MilleO2012] Miller, H.I. and Ostaszewski, A.J. Group action and shift-
compactness. J. Math. Anal. Appl. 392 (2012), 23–39. 177, 248,
279
[MilleMO2021] Miller, H.I., Miller-Van Wieren, L. and Ostaszewski, A.J. Beyond
Erdős–Kunen–Mauldin: Singular sets with shift-compactness prop-
erties. Topol. Appl. 291 (2021), 107605. 73
[Mit2010] Mitchell, W.J. Beginning inner model theory. In ForK2010, pp. 1487–
1594. 275, 276
[Mon1935] Montgomery, D. Non-separable metric spaces. Fund. Math. 25
(1935), 527–533. 96
[Mon1936] Montgomery, D. Continuity in topological groups. Bull. Am. Math.
Soc. 42 (1936), 879–882. 197
[Mon1950] Montgomery, D. Locally homogeneous spaces. Ann. Math. 52 (1950),
261–271. 173
[Mos2009] Moschovakis, Y.N. Descriptive Set Theory, 2nd ed. Math. Surveys
Monog. 155, Am. Math. Soc., 2009. (1st ed. 1980). 61
[Mosp2005] Mospan, Y.V. A converse to a theorem of Steinhaus–Weil. Real An.
Exch. 31 (2005), 291–294. 250, 251, 254
[Mue1965] Mueller, B.J. Three results for locally compact groups connected with
the Haar measure density theorem. Proc. Am. Math. Soc. 16 (1965),
1414–1416. 256
[Mun1975] Munkres, J.R. Topology, A First Course. Prentice-Hall, 1975. 35
316 References

[Mut1999] Muthuvel, K. Application of covering sets. Colloq. Math. 80 (1999),

115–122. 117
[Myc1979] Mycielski, J. Finitely additive measures. Coll. Math. 42 (1979), 309–
318. 240
[MycS1962] Mycielski, J. and Steinhaus, H. A mathematical axiom contradict-
ing the Axiom of Choice. Bull. Acad. Polon. Sci. Sér. Sci. Math.
Astronom. Phys. 10 (1962), 1–3. 231
[MycSw1964] Mycielski, J. and Świerczkowski, S. On the Lebesgue measurability
and the axiom of determinateness. Fund. Math. 54 (1964), 67–71.
61, 231
[Nam1974] Namioka, I. Separate and joint continuity. Pacific J. Math. 51 (1974),
515–531. 197
[NamP1969] Namioka, I. and Pol, R. σ-fragmentability and analyticity. Mathe-
matika 43 (1969), 172–181. 203
[Nee1997] Neeb, K.-H. On a theorem of S. Banach. J. Lie Theory 7 (1997),
293–300. 184, 189, 190, 191, 318
[Neem2010] Neeman, I. Determinacy in L(R). In ForK2010, Chapter 21. 231
[Neu1954a] Neumann, B.H. Groups covered by finitely many cosets. Publ. Math.
Debrecen 3 (1954/5), 227–242. 142
[Neu1954b] Neumann, B.H. Groups covered by permutable subsets. J. London
Math. Soc. 29 (1954), 236–248. 142
[Nik1925] Nikodym, O. Sur une propriété de l’opération A. Fund. Math. 7
(1925), 149–154. 54
[Nol1990] Noll, D. A topological completeness concept with applications to
the open mapping theorem and the separation of convex sets. Topol.
Appl. 35 (1990), 53–69. 182
[Nol1992] Noll, D. Souslin measurable homomorphisms of topological groups.
Arch. Math. (Basel) 59 (1992), 294–301. 217
[Ols2004] Olsen, L. Extremely non-normal numbers. Math. Proc. Cambridge
Phil. Soc. 137 (2004), 43–53. 225
[OMa1977] O’Malley, R.J. Approximately differentiable functions: The r topol-
ogy. Pacific J. Math. 72 (1977), 207–222. 122, 124
[Ost1976] Ostaszewski, A.J. On countably compact perfectly normal spaces. J.
London Math. Soc. 14 (1976), 505–516. 145
[Ost1980] Ostaszewski, A.J. Monotone normality and G δ -diagonals in the class
of inductively generated spaces. In Topology II, p. 905–930. Colloq.
Math. Soc. János Bolyai 23, North-Holland, 1980. 44
[Ost1989] Ostaszewski, A.J. On how to trap a gap: “An Introduction to Indepen-
dence for Analysts by H.G. Dales and W.H. Woodin”. Bull. London
Math. Soc. 21 (1989), 197–208. 232
[Ost2010] Ostaszewski, A.J. Regular variation, topological dynamics, and the
uniform boundedness theorem. Topol. Proc. 36 (2010), 305–336.
278, 279
[Ost2011] Ostaszewski, A.J. Analytically heavy spaces: Analytic Cantor and
analytic Baire theorems. Topol. Appl. 158 (2011), 253–275. 56, 57,
60, 174, 207
 References 317

[Ost2012] Ostaszewski, A.J. Analytic Baire spaces. Fund. Math. 217 (2012),
189–210. 176, 183, 205
[Ost2013a] Ostaszewski, A.J. Almost completeness and the Effros open mapping
principle in normed groups. Topol. Proc. 41 (2013), 99–110. 174,
178
[Ost2013b] Ostaszewski, A.J. Shift-compactness in almost analytic submetriz-
able Baire groups and spaces. Topol. Proc. 41 (2013), 123–151. 174,
175, 192, 195
[Ost2013c] Ostaszewski, A.J. Beyond Lebesgue and Baire III: Steinhaus’ theo-
rem and its descendants. Topol. Appl. 160 (2013), 1144–1154. 177,
197, 207, 213
[Ost2013d] Ostaszewski, A.J. The Semi-Polish Theorem: One-sided vs joint con-
tinuity in groups. Topol. Appl. 160 (2013), 1155–1163. 183, 213, 219
[Ost2015a] Ostaszewski, A.J. Beurling regular variation, Bloom dichotomy,
and the Gołab–Schinzel
˛ functional equation. Aequationes Math. 89
(2015), 725–744. 13, 21, 288
[Ost2015b] Ostaszewski, A.J. Effros, Baire, Steinhaus and non-separability.
Topol. Appl. 195 (2015), 265–274. 173
[Ost2016a] Ostaszewski, A.J. Stable laws and Beurling kernels. Advances in
Applied Probability 48(A) (N. H. Bingham Festschrift, C.M. Goldie
and A. Mijatović, eds), (2016), 239–248. 3
[Ost2016b] Ostaszewski, A.J. Homomorphisms from functional equations: The
Goldie equation. Aequationes Math. 90 (2016), 427–448. 288
[Ost2017] Ostaszewski, A.J. Homomorphisms from functional equations in
probability. In Developments in Functional Equations and Related
Topics, pp. 171–213. Springer Optim. Appl. 124, Springer, 2017. 283
[Ostr1929] Ostrowski, A. Mathematische Miszellen XIV: Über die Funktion-
algleichung der Exponentialfunktion und verwandte Funktionalgle-
ichungen, Jahresb. Deutsch. Math. Ver. 38 (1929), 54–62. (Reprinted
in Collected Papers of Alexander Ostrowski, 4, 49–57, Birkhäuser,
1984.) 19, 21
[Ostr1976] Ostrowski, A.M. On Cauchy–Frullani integrals. Comment. Math.
Helvet. 51 (1976), 57–91. 11
[Oxt1946] Oxtoby, J.C. Invariant measures in groups which are not locally
compact. Trans. Am. Math. Soc. 60 (1946), 215–237. 242
[Oxt1960] Oxtoby, J.C. Cartesian products of Baire spaces. Fund. Math. 49
(1960/1961), 157–166. 39, 40, 160
[Oxt1980] Oxtoby, J.C. Measure and Category, 2nd ed. Grad. Texts Math. 2,
Springer 1980. (1st ed. 1971.) 1, 33, 35, 37, 42, 43, 46, 104, 109,
110, 117, 137, 138, 139, 148, 156, 222, 223, 224, 228, 234
[PalW1934] Paley, R.E.A.C. and Wiener, N. Fourier Transforms in the Complex
Domain. AMS Colloq. Publ. XIX, Am. Math. Soc., 1934. 5
[Par1967] Parthasarathy, K.R. Probability Measures on Metric Spaces. Aca-
demic Press, 1967. (Reprinted Am. Math. Soc., 2005.) 73, 161,
240
[Pat1988] Paterson, A.L.T. Amenability, Math. Surveys Monog. 29, Am. Math.
Soc., 1988. 246, 282
318 References

[Paw1985] Pawlikowski, J. Lebesgue measurability implies Baire property. Bull.

Sci. Math. (2) 109 (1985), 321–324. 232
[Per1960] Perron, O. Irrationalzahlen. Walter de Gruyter, 1960. 223
[Pes1998] Pestov, V. Review of Nee1997, MathSciNet MR1473172
(98i:22003). 184
[Pet1950] Pettis, B.J. On continuity and openness of homomorphisms in topo-
logical groups. Ann. Math. 52 (1950), 293–308. 87, 99, 139, 173,
174, 236, 239, 240
[Pet1951] Pettis, B.J. Remarks on a theorem of E.J. McShane. Proc. Am. Math.
Soc. 2 (1951), 166–171. 236, 239
[Pet1974] Pettis, B.J. Closed graph and open mapping theorems in certain
topologically complete spaces. Bull. London Math. Soc. 6 (1974),
37–41. 188
[Pic1939] Piccard, S. Sur les Ensembles de Distances des Ensembles de Points
d’un Espace Euclidien. Mém. Univ. Neuchâtel 13 (1939). 139, 173,
236, 240
[Pic1942] Piccard, S. Sur des Ensembles Parfaites, Mém. Univ. Neuchâtel 16
(1942). 139, 173, 236, 240
[PitP2016] Pitman, E.J.G. and Pitman, J.W. A direct approach to the stable
distributions. In N. H. Bingham Festschrift, C.M. Goldie and A.
Mijatović (eds), Advances in Applied Probability 48(A) 2016, 261–
282. 3
[Plo2003] Płotka, K. On functions whose graph is a Hamel basis. Proc. Am.
Math. Soc. 131 (2003), 1031–1041. 22
[Pol1976] Pol, R. Remark on the restricted Baire property in compact spaces.
Bull. Acad. Polon. Sci. Sér. Sci. Math. 24 (1976), 599–603. 39, 185
[Pol1979] Pol, R. Note on category in Cartesian products of metrizable spaces.
Fund. Math. 102 (1979), 55–59. 139, 148, 159, 175, 228
[Poly1923] Pólya, G. Bemerkungen über unendliche Folgen und ganze Funktio-
nen. Math. Ann. 88 (1923), 169–183. (Reprinted in Collected Papers
Vol. 1, R.P. Boas (ed). MIT Press, 1974.) 6
[Pop1965] Popa, C. Gh. Sur l’equation fonctionnelle f [x + y f (x)] = f (x) f (y),
Ann. Polon. Mathe. 17 (1965), 193–198 288
[PorWW1985] W. Poreda, W., Wagner-Bojakowska E. and Wilczyński, W. A cat-
egory analogue of the density topology. Fund. Math. 125 (1985),
167–173. 133
[PortS1978] Port, S.C. and Stone, C.J. Brownian Motion and Classical Potential
Theory. Academic Press, 1978. 118
[Que1980] Queféllec, H. Propriétés presque sûres et quasi-sûres des séries de
Dirichlet et des produits d’Euler. Canad. J. Math. 32 (1980), 531–
558. 227
[Rad1919] Rademacher, H. Über partielle und totale Differenzierbarkeit I. Math.
Ann. 79 (1919), 340–359. 228
[Rad1922] Rademacher, H. Einige Sätze über Reihen von allgemeinen Orthog-
onalfunktionen. Math. Ann. 87 (1922), 112–138. 224
 References 319

[Rai1984] Raisonnier, J. A mathematical proof of S. Shelah’s theorem on the

measure problem and related results. Israel J. Math. 48 (1984), 49–
56. 230, 233
[RaiS1983] Raisonnier, J. and Stern, J. Mesurabilité et propriété de Baire.
Comptes Rendus Acad. Sci. I. (Math.) 296 (1983), 323–326. 232,
233
[RaiS1985] Raisonnier, J. and Stern, J. The strength of measurability hypotheses.
Israel J. Math. 50 (1985), 337–349. 230
[Ram1951] Ramsey, A.S. Dynamics, Part I. Cambridge University Press, 1951.
114
[Rams1930] Ramsey, F.P. On a problem of formal logic. Proc. London Math. Soc.
30 (1930), 338–384. 274
[RaoR1974] Rao, K.P.S. Bhaksara and Rao, M. Bhaskara. A category analogue
of the Hewitt–Savage zero–one law. Proc. Am. Math. Soc. 44 (1974),
497–499. 222
[RaoR1975] Rao, K.P.S. Bhaksara and Rao, M. Bhaskara. On the difference of
two second category Baire sets in a topological group. Proc. Am.
Math. Soc. 47 (1975), 257–258. 112
[Rea1996] Reardon, P. Ramsey, Lebesgue, and Marczewski sets and the Baire
property. Fund. Math. 149 (1996), 191–203. 121
[Rog1970] Rogers, C.A. Hausdorff Measures. Cambridge University Press,
1970. 226
[RogJ1980] Rogers, C.A. and Jayne, J. K-Analytic sets. In RogJD1980, Part 1.
xiii, 31, 42, 51, 52, 54, 56, 61, 65, 66, 123, 178, 181, 187, 188, 203,
204, 205, 212, 218, 236
[RogJD1980] Rogers, C.A., Jayne, J., Dellacherie, C., Topsøe, F., Hoffmann-
Jørgensen, J., Martin, D.A., Kechris, A.S. and Stone, A.S. Analytic
Sets. Academic Press, 1980. 62, 203, 241, 301, 308, 314, 319, 321,
322
[RogW1966] Rogers, C.A. and Willmott, R.C. On the projection of Souslin sets.
Mathematika 13 (1966), 147–150. 52
[Roy1988] Royden, H.L. Real Analysis, 2nd ed. Prentice-Hall, 1988 (1st ed.
1963.) 24, 25
[Rud1966] Rudin, W. Real and Complex Analysis. McGraw-Hill, 1966. (3rd. ed.
1987, 2nd ed. 1974.) 269, 283
[Rud1973] Rudin, W. Functional Analysis. McGraw-Hill, 1973. (2nd ed. 1991.)
86, 90, 172, 182, 183
[Sai1983] Saint Raymond, J. Jeux topologiques et espaces de Namioka. Proc.
Am. Math. Soc. 87 (1983), 499–504. 45, 47
[Sak1956] Sakovich, G.N. A single form for the conditions for attraction to
stable laws. Th. Prob. Appl. 1 (1956), 322–325. 2
[Saks1964] Saks, S. Theory of the Integral, 2nd ed. Dover, 1964. (1st ed. 1937,
Monografie Mat. VII, Instytut Matematyczny Polskiej Akademii
Nauk.) 73, 104, 254
[Sal1966] S̆alát, T. A remark on normal numbers. Rev. Roum. Math. Pures Appl.
11 (1966), 53–56. 225, 226
320 References

[Sch1971] Scheinberg, S. Topologies which generate a complete measure alge-

bra. Adv. Math. 7 (1971), 231–239. 109, 130
[Schw1966] Schwartz, L. Sur le théorème du graphe fermé. C. R. Acad. Sci. Paris
Sér. A–B 263 (1966), 602–605. 188
[Schw1973] Schwartz, L. Radon Measures on Arbitrary Topological Spaces and
Cylindrical Measures. Tata Institute of Fundamental Research Stud-
ies in Mathematics 6, Oxford University Press, 1973. 240
[Sco1961] Scott, D. Measurable cardinals and constructible sets. Bull. Acad.
Polon. Sci. Sér. Sci. Math. Astr. Phys. 9 (1961), 521–524. 274
[Sem1971] Semadeni, Z. Banach Spaces of Continuous Functions. Monografie
Matematyczne 55, PWN. Polish Scientific Publishers, 1971. 584pp.
[Sen1973] Seneta, E. An interpretation of some aspects of Karamata’s theory
of regular variation. Publ. Inst. Math. Beograd (NS) 15 (29) (1973),
111–119. 10
[She1969] Shea, D.F. On a complement to Valiron’s Tauberian theorem for the
Stieltjes transform. Proc. Am. Math. Soc. 21 (1969), 1–9. 4, 5
[Shel1984] Shelah, S. Can you take Solovay’s inaccessible away? Israel J. Math.
48 (1984), 1–47. 230, 232, 233
[Shel1985] Shelah, S. On measure and category. Israel J. Math. 52 (1985), 110–
114. 233
[ShelW1990] Shelah, S. and Woodin, H. Large cardinals imply that every reason-
ably definable set of reals is Lebesgue measurable. Israel J. Math.
70(3) (1990), 381–394. 233
[ShiT1998] Shi, H. and Thomson, B.S. Haar null sets in the space of automor-
phisms on [0,1]. Real Anal. Exchange 24 (1998/99), 337–350. 241,
258
[Sho1984] Shortt, R.M. Universally measurable spaces: An invariance theorem
and diverse characterizations. Fund. Math. 121 (1984), 169–176. 241
[Sie1920] Sierpiński, W. Sur l’equation fonctionelle f (x + y) = f (x) + f (y).
Fund. Math. 1 (1920), 116–122. (Reprinted in Oeuvres Choisis II,
331–336, PWN, 1975.) 21, 81
[Sie1935] Sierpiński, W. Sur deux ensembles linéaires singuliers. Ann. Scuola
Norm. Super. Pisa Cl. Sci 4 (1935), 43–46. 21
[Sie1975] Sierpiński, W. Oeuvres Choisies II: Théorie des Ensembles et ses
Applications, Travaux des Années 1908–1929, PWN, 1975. 137, 307
[Sie1976] Sierpiński, W. Oeuvres Choisies III: Théorie des Ensembles et ses
Applications, Travaux des Années 1930–1966, PWN, 1976. 137
[Sil1971] Silver, J. Every analytic set is Ramsey. J. Symbolic Logic 35 (1971),
60–64. 121
[Silv1974] Silverstein, M.L. Symmetric Markov Processes. Lecture Notes in
Math. 426, Springer, 1974. 119
[Silv1976] Silverstein, M.L. Boundary Theory for Symmetric Markov Processes.
Lecture Notes in Math. 516, Springer, 1976. 119
[Sim1975] Simmons, S.M. A converse Steinhaus–Weil theorem for locally com-
pact groups. Proc. Am. Math. Soc. 49 (1975), 383–386. 251, 254
[Simo2015] Simon, B. Basic Complex Analysis. A Comprehensive Course in
Analysis, Part 2A, Am. Math. Soc., 2015. 229
 References 321

[Sol2005] Solecki, S. Size of subsets of groups and Haar null sets. Geom. Funct.
Anal. 15 (2005), 246–273. 241, 258
[Sol2006] Solecki, S. Amenability, free subgroups, and Haar null sets in non-
locally compact groups. Proc. London Math. Soc. (3) 93 (2006),
693–722. 240, 241, 242, 245, 246, 248, 251, 259
[Sol2007] Solecki, S. A Fubini theorem. Topol. Appl. 154 (2007), 2462–2464.
241
[SolS1997] Solecki, S. and Srivastava, S.M. Automatic continuity of group op-
erations. Topol. Appl. 77 (1997), 65–75. 197
[Solo1965] Solovay, R.M. 2ℵ0 can be anything it ought to be. In J.W. Addi-
son, L. Henkin and A. Tarski (eds), The Theory of Models, (Proc.
1963 Int. Symp. Berkeley). Studies in Logic and the Foundations of
Mathematics, North-Holland, 1965, 435. 268
[Solo1969] Solovay, R.M. On the cardinality of Σ21 sets of reals. In Foundations of
Mathematics (Symposium Commemorating Kurt Gödel, Columbus,
Ohio, 1966), pp. 58–73. Springer, 1969. 233
[Solo1970] Solovay, R.M. A model of set theory in which every set of reals is
Lebesgue measurable. Ann. Math. 92 (1970), 1–56. 229, 230, 231,
232
[Solo1978] Solovay, R.M. The independence of DC from AD. In The Cabal
Seminar 76–77 (Proc. Caltech–UCLA Logic Sem., 1976–77), pp.
171–183. Lecture Notes in Math. 689, Springer, 1978. 232
[Spe1994] Spencer, J. Ten Lectures on the Probabilistic Method, 2nd ed. CBMS-
NSF Reg. Conf. Ser. Appl Math. 64, SIAM, 1994 (1st ed. 1987.) 37
[Ste1970] Stein, E.M. Singular Integrals and Differentiability Properties of
Functions. Princeton University Press, 1970. 104
[SteS2005] Stein, E.M. and Shakarchi, R. Real Analysis: Measure Theory, Inte-
gration, and Hilbert Spaces. Princeton University Press, 2005. 25,
269
[Stei1920] Steinhaus, H. Sur les distances des points de mesure positive. Fund.
Math. 1 (1920), 93–104. 139, 236, 239
[Stei1930] Steinhaus, H. Über die Wahrscheinlichleit dafür, daß der Konver-
genzkreis einer Potenzreihen ihre natürliche Grenze ist. Math. Z. 31
(1930), 408–416. 228
[Ster1985] Stern, J. Regularity properties of definable sets of reals. Ann. Pure
Appl. Logic 29 (1985), 289–324. 232, 233
[Sto1963] Stone, A.H. Kernel constructions and Borel sets. Trans. Am. Math.
Soc. 107 (1963), 58–70; errata, ibid. 107 (1963), 558. 37, 122, 128,
159
[Sto1980] Stone, A.H. Analytic sets in non-separable spaces. In RogJD1980,
Part 5. 204
[Sze2011] Szenes, A. Exceptional points for Lebesgue’s density theorem on the
real line. Adv. Math. 226 (2011), 764–778. 167
[Tal1976] Tall, F.D. The density topology. Pacific J. Math. 62 (1976), 275–284.
110, 156
[Tal1978] Tall, F.D. Normal subspaces of the density topology. Pacific J. Math.
75 (1978), 579–588. 110
322 References

[TaoV2006] Tao, T. and Vu, V.N. Additive Combinatorics. Cambridge Studies in

Adv. Math., 105. Cambridge University Press, 2006. 141
[Tar1930] Tarski, A. Une contribution à la théorie de la mesure. Fund. Math.
15 (1930), 42–50. 232
[Tay2004] Taylor, S.J. John Hawkes (1944–2001). Bull. Lond. Math. Soc. 36
(2004), 695–710. 226
[Ten2015] Tenenbaum, G. Introduction to Analytic and Probabilistic Number
Theory, 3rd ed. Grad. Studies Math. 193, American Mathematical
Society, 2015. (2nd ed. Cambridge Studies Adv. Math. 46, Cambridge
University Press, 1995.) 7, 8
[Tit1986] Titchmarsh, E.C. The Theory of the Riemann Zeta-Function, Second
ed. (revised by Heath-Brown, D. R.), Oxford University Press, 1986.
(1st ed. 1951.) 227
[TomW2016] Tomkowicz, G. and Wagon, S. The Banach–Tarski Paradox, 2nd ed.
Encycl. Math. Appl. 163, Cambridge University Press, 2016. (1st ed.,
Encycl. Math. Appl. 24, Cambridge University Press, 1985.) 223,
257
[TopH1980] Topsøe, F. and Hoffmann-Jørgensen, J. Analytic spaces and their
applications. In RogJD1980, Part 3. 95, 197
[Tor2013] Törnquist, A. Σ21 and Π11 mad families. J. Symbolic Logic 78 (2013),
1181–1182. 234
[Tra1987] Trautner, R. A covering principle in real analysis. Quart. J. Math. 38
(1987), 127–130. 162
[Ula1930] Ulam, S. Zur Maßtheorie in der allgemeinen Mengenlehre. Fund.
Math. 16 (1930), 140–150. 200, 232, 273
[Ula1960] Ulam, S.M. A Collection of Mathematical Poblems. Wiley, 1960.
[Ung1975] Ungar, G.S. On all kinds of homogeneous spaces. Trans. Am. Math.
Soc. 212 (1975), 393–400. 173
[WagW2000] Wagner-Bojakowska, E. and Wilczyński, W, Cauchy condition for the
convergence in category. Proc. Am. Math. Soc. 128 (2000), 413–418.
30
[Wei1940] Weil, A. L’intégration dans les Groupes Topologiques et ses Appli-
cations, Actual. Sci. Ind. 869, Hermann, Paris, 1940. (Republished,
Princeton University Press, 1941.) 239
[Weis1984] Weiss, W. Versions of Martin’s axiom. In KunV1984, Ch. 19, pp.
827–886. 272
[Whi1974] White, H.E. Topological spaces in which Blumberg’s Theorem holds.
Proc. Am. Math. Soc. 44 (1974), 454–462. 40, 111
[Whi1975] White, H.E. Topological spaces that are α-favorable for a player with
perfect information. Proc. Am. Math. Soc. 50 (1975), 477–482. 56
[Wie1933] Wiener, N. The Fourier Integral and Certain of Its Applications.
Cambridge University Press, 1933. (Reprinted, Cambridge Mathe-
matical Library, 1988.) 17
[Wil2002] Wilczyński, W. Density topologies. In Handbook of Measure Theory,
Vol. I, 675–702, North-Holland, 2002. 256
[WilW2007] Wilczyński, W. and Wojdowski, W. Complete density topology.
Indag. Math. (NS) 18 (2007), 295–303. 133
 References 323

[Will1972] Williamson, J. Meromorphic functions with negative zeros and pos-

itive poles and a theorem of Teichmüller. Pacific J. Math. 42 (1972),
795–810. 5
[Woo1988] Woodin, W.H. Supercompact cardinals, sets of reals, and weakly
homogeneous trees. Proc. Nat. Acad. Sci. USA 85 (1988), 6587–
6591. 271
[Woo1999] Woodin, W.H. The Axiom of Determinacy, Forcing Axioms, and the
Nonstationary Ideal, de Gruyter, 1999. 275
[Xia1972] Xia, D.X. Measure and Integration Theory on Infinite Dimensional
Spaces. Abstract Harmonic Analysis. Pure and App. Math. 48 Aca-
demic Press, 1972. 240, 253
[Zah1950] Zahorski, Z. Sur la première dérivée. Trans. Am. Math. Soc. 69
(1950), 1–54. 113
[Zap1999] Zapletal, J. Terminal notions. Bull. Symbolic Logic 5 (1999), 470–
478. 64
[Zap2001] Zapletal, J. Terminal notions in set theory. Ann. Pure Appl. Logic 109
(2001), 89–116. 64
[Zel1960] Żelazko, B. A theorem on B0 division algebras. Bull. Acad. Plon.
Sci. 8 (1960), 373–375. 94
[ZelB1970] Zeller, K. and Beekmann, W. Theorie der Limitierungsverfahren, 2nd
ed. Springer, 1970. 16
[Zem2002] Zeman, M. Inner Models and Large Cardinals. De Gruyter Series in
Logic and its Applications, 5 de Gruyter, 2002. 271
[Zsi2004] Zsilinszky, L. Products of Baire spaces revisited. Fund. Math. 183
(2004) 115–12. 160
[Zyg1988] Zygmund, A. Trigonometric Series, Vols. I, II, 2nd ed. Cambridge
University Press, 1988. (3rd ed., 2002, with foreword by R. Feffer-
man, 2002.) 138
 Index

A-σ-d map, 208 analytically generated, 120

A-σ-discrete map, 210 analytically heavy, 120
a.e.-topology, 124 arcwise connected, 185, 186
abelian group-norm, 87 Arsenin–Kunugui Theorem, 52, 65
absolutely analytic subset, 32 Asymptotic Conjugacy Theorem, 286
accessibility, points of, 118 asymptotically invariant space, 286
accumulates essentially, 79 atomless, 127
acting transitively on a family, 257 autohomeomorphism, Auth(X), 176
action, 174 automatic continuity, 216
microtransitive, 175, 176 Axiom of Choice (AC), 61
Nikodym, 174, 178 Axiom of Dependent Choice (DC), 229
transitive, 176 Axiom of Determinacy (AD), 231
weakly microtransitive, 176 Axiom of Projective Determinacy (PD), 61, 232
additively maximal, 127 B-σ-d map, 208
additivity, add, 154 Baire
admissibility, 197 completely hereditarily, 37
Aggregation Theorem, 247 hereditarily, 37
Alexandroff’s Remetrization Theorem, 35 restricted, 37
almost analyticity, 193 Baire Category Theorem, 35
almost complete space, 182, 195 Baire Continuity Theorem, 38
Almost Completeness Theorem, 213 Baire version, 185
almost locally ccc space, 158 Baire-measurable, 188
almost locally uK–U space, 158 Baire function, 38
alternating zeta function, 227 Baire Homomorphism Theorem, 187
amenability at 1G , 242 non-separable, 218
Analytic Baire Theorem, 95 Baire hull, 68
Analytic Cantor Theorem, 56 Baire hull theorem, 69
non-separable, 207 Baire measurable, 38, 184, 188
analytic continuation, 227 Baire product, 158
analytic set, 52 Baire relative to convolution, 281
Analytic Shift Theorem, 97 Baire set, 71
analytic space, 55 essentially unbounded, 79
analytic topological space Baire set property, BP, 37, 192, 229
absolutely, 193 Baire space, 36
classically, 193 Baire space property, BS, 192
analytical hierarchy, 61 Baire Theorem, see Baire Category Theorem

324
 Index 325

Banach Continuous Homomorphism Theorem, cardinal, 272

189 inaccessible, 274
Banach’s Category Theorem, 42 measurable, 33, 273
variant, 43 non-measurable, 33, 273
Banach’s Localization Principle, see Banach’s Ramsey, 274
Category Theorem real measurable, 273
Banach’s Union Theorem, see Banach’s strong, 275
Category Theorem supercompact, 275
Banach–Mazur game, 45 Woodin, 275
Banach–Mehdi Theorem, 186, 283 cardinal invariants, 234
Banach–Neeb Theorem, 184, 188, 191 category and measure, relationship, 229, 230
Banach–Tarski decomposition, 223 category contrasted with measure, 222
base category convergence (cc), 163
countable-in-itself, 158 Category Embedding Theorem (CET), 162,
locally countable, 39 163, 280
point-regular, 28 category–measure, 109
pseudo-base, 39 category–measure duality, 137
regular, 28 Cauchy functional equation (CFE), 8, 138
regular filter, 135 Cauchy sequence, 30
preconvergent, 135 Čech-complete space, 56
sigma-disjoint, 40 Centrality Theorem, 287
topological, 120 Chaber’s Theorem, 44
ultrafilter, 135 chain condition, 154
convergent, 135 countable (ccc), 110, 149, 154, 271
base operator, 124 Champernowne’s number, 225
base-σ-discrete, 208 character theorem, 62
Ba(X), Baire sets of X, 71 first, 64
Beurling regular variation, 287 second, 64
Beurling slowly varying, 13 third, 64
Beurling’s Tauberian Theorem, 14 character-degradation theorem, 19
bi-invariant metric, 285 characterization theorem, 9, 282
Bing’s Theorem, 165 charge, 119
Bing’s theorem, 27, 208 Choquet game, 46
Birkhoff–Kakutani Normability Theorem, 90 strong, 46
Birkhoff–Kakutani Theorem, 90 Choquet space, 46
Bitopology, 79 strongly, 46
Blumberg’s property, 40 Christensen game, 47
locally countable, 40 Cichoń diagram, 222, 234
Bohr abscissa, 226 circumscribed map, 55, 122
Borel–Tauber Theorem, 14 cluster point, 44
Borelian hierarchy, 51 co-analytic sets, 61
Borelian-F , 32, 51 coalescing, 32
BP, all sets of reals have Baire property, 229 coboundary, 288
Brownian motion, 118 cocompact space, 135
càdlàg, 186 cocycle, 288, 289
Cameron–Martin space, 260 coded set, 62
Cameron–Martin theory, 247 Combinatorial Steinhaus Theorem, 141
Cameron–Martin–Girsanov formula, 261 cometrizable space, 135
Cantor universe, 270 Common Basis Theorem, 237
Cantor’s Nested Sets Theorem, 35 commutual, 237
capacity, 117 self-, 238
326 Index

weakly, 238 Darji-measurable, see D-measurable

self-, 238 DC, Axiom of Dependent Choice, 229
compact-valued map, 54 definable set, 233
compactness demi-open set, 182
countable, 26 density basis, 105
pseudo, 26 density point, 105
sequential, 26 category-, 133
complete complete-, 133
Čech-complete, 35 simple, 133
topologically complete, 35 density topology, 79, 101, 121
completely additive analytic, 185, 209 determining system, 204
completely additive measurable, 185 shrinking, 204
Concatenation Formula, 289 Dichotomy
condition (cc), 163 Dodos Dichotomy Theorem, 253
Consistency, 232 generic, 71
consistency strength, 275, 276 Glimm–Effros, 172
constructible hierarchy, L, 270 Steinhaus, 139
continued fraction, 225 dilation, 131
Continuity Theorem Direct Baire property, 193
Baire Continuity Theorem, 185 Dirichlet eta function, 226
Banach Continuous Homomorphism, 189 Dirichlet forms, 118, 120
Blumberg’s, 40 Dirichlet series, 226
Lusin’s, 24 Discontinuity-Set Theorem, 184
Continuous Coboundary Theorem, 290 discrete
Continuous Homomorphism Theorem, 290 σ-discrete, 27
Continuous Inverse Theorem, 217 dispersion point, 105
Continuum Hypothesis (CH), 268 Displacements Lemma, 74, 96
Generalized, GCH, 269 displacements lemma
convergence Dowker’s Theorem, 27
modes of, 29 dyadic space, 153
core models, 271 E, 79
cospace, 135 Effros null sequence, 279
cotopology, 135 Effros property, 176, 280
countability conditions, topological, 44 Effros Theorem, Theorem E, 172, 174
countable chain condition (ccc), 110, 149, crimping, 177
271 Egorov’s Theorem, 25
countably covered map, 181 elephant, 268
countably distinguished, 48 Ellentuck topology, Ell, 121
countably subcompact space, 135 enable Baire continuity, 185
covering energy, 119
almost covering, 42 ε-swelling, 88
sigma-discrete, 42 equipotent sets, 272
crible, 62 Equivalence Theorem, 93
crimping property, 176 Erdős–Sierpiński map, 228
cumulative hierarchy, 270 essential accumulation, 79
cyclically permutable, 87 essential limit point, 42
D, 79 essential topology, 123, 128, 130
D-measurable, 264 essentially unbounded, 79
Darboux Dichotomy, 144 everywhere non-meagre, 42
Darboux’s Theorem, 144 exceptional set, 281
Darji Haar-meagre sets, 249 extended Souslin operation, 204
 Index 327

family Haar null set, 265

heavy, 127 Haar null, left, 251
proper ideal, 127 Hamel basis, 21
Fechner’s Law, 22 Hamel pathology, 8
fine topology, 107, 115, 123 Hansell’s Characterization Theorem, 205
ideals, 131 Hansell’s condition, 185
finite covering property, 141 Hansell’s theorem, 39
finite index property, 141 Hardy–Littlewood maximal function, 104
finite intersection property, 57 harmonic, 115
finite-covering characterization, 142 Hartman–Mycielski Embedding Theorem, 185
First Factorization Theorem, 284, 285 Hashimoto ideal topology, 249
forcing, 229 Hausdorff dimension, 226
Fréchet-regularly varying, 288 Hausdorff space
Fréchet–Banach Theorem, 146 almost analytic, 193
Generalized, 146 heavy point, 150
Frullani integral, 11 hereditary, 127
Fubini Null Theorem, FN, 147 hierarchy
fundamental in measure sequence, 30 analytical, 61
game theory, 45 Borelian, 51
game, favourable, 47 Borelian-H , 32
game, strength of, 47 projective, 60
Gandy–Harrington Theorem homomorphism, 92
generalized, 122 hyperspace, 240
Gandy–Harrington topology, 120 I-heavy, 121, 127
Gaussian probability measure, 259 inaccessible, 229, 274
Gδ -sequence, 44 inaccessible cardinal, 229
general regular variation, 17 index, 9
generic completeness, 72 Index Theorem, 283
generic dichotomy, 71 index-σ-discrete, 208
principle, 71 indicator function
generic existence, 37 strong, 132
genericity, 167
infinite combinatorics, 71
genericity, typicality, 228
inner model, 271
G H , Gandy–Harrington topology, 121
inner regularity, 240
Glimm–Effros Dichotomy, 172
insertion property, 108
Gödel’s Axiom of Constructibility, V = L, 53,
Interior-point property
61, 232, 270
Borell, 259
Gödel’s incompleteness theorems, 269
Interior-point theorem, 235
Goffman’s Theorem, 112
Borell, 260
Gołab–Schinzel
˛ equation, 18, 288
relative, 261
Goldie equation, 288
Solecki, 248
Green function, 119
Invariance of Norm Theorem, 88
Greenian, 119
inverse measure, 243
ground space, 120
inverse square law, 114
group pre-norm, 87
group pseudo-norm, 87 Janalytic , 47
group semi-norm, 87 Jakimovski method, 16
group-norm, 86, 87 Jensen’s Diamond, ^, 145
abelian, 87 κ-additive, 199, 272
Haar density topology, 138, 257 K-analytic, 53
Haar meagre, 261 K-analytic space, 54
328 Index

K-analytic subset, 54 LM, all sets of reals Lebesgue measurable, 229

κ-chain condition (κ-cc), 154 local almost nullity, 252
K σ -regularity, 230 locally countable pseudo-base, 158
K -analytic, 53 locally meagre, 42
Karamata–Stirling method, 16 lower density, 104, 125
Kemperman’s Theorem, 74 lower-density operator, 167
Kendall’s Theorem, 18 Lusin’s Continuity Theorem, 24
Kepler’s Laws, 114 Lusin’s Theorem, 224
kernel, 115 Lusin–Menchoff property, 110
Kestelman–Borwein–Ditor Theorem (KBD), M, axiom of existence of measurable cardinal,
73, 165–167, 237 274
Kingman Theorem map
bitopological, 82 σ-discrete, 38
for Category, 84 Marczewski measurable set, 22
for Measure, 84 Marczewski null set, 22
Klee property, 89, 285 Marczewski’s Theorem and Corollary, 54
Knaster’s property (property K), 111 Markov process, 117
Kodaira’s Theorem, 138 symmetric, 119
Kominek’s Theorem, 259 Martin’s Axiom (MA), 271
Krom space, 159 maximally discontinuous measure, 243
Krom’s Theorem, 160 meagre set, 34
Kronecker’s Theorem, 10, 140 measurable cardinal, 33
Kuratowski–Ulam pair, K–U, 148 non-measurable cardinal, 33, 200, 232, 273
Kuratowski–Ulam Theorem, KU, 148 Pol’s Theorem, 200, 232, 273
Kuratowski–Ulam Theorem, Variant, 148 restricted Baire property, 200, 232, 273
Kuratowski–Ulam*, K–U∗ , pair, 149 measurable core, 105
L, class of all constructible sets, 270 measurable cover, 105
Laplace’s equation, 118 measurable set, 104
Large cardinals, 232 metric
Lebesgue Density Theorem, 101 Ky Fan, 29
Lebesgue measurable (LM), 229 left-invariant, dL (x, y), 91
left µ-null measure, L, 251 Prohorov, 29
left absolutely continuous measure, 252 right-invariant, dR (x, y), 91
left Haar null, 241 microtransitivity, 173
left invertibly µ-null measure, L-inv, 251 mobile measure, see translation-continuous
Left Product Theorem, 287 measure
left singular measure, 252 modification theorem, 130
left-invariant metric, 87 Modular Flow Theorem, 286
level set, 145 modular function, 286
Levi Coincidence Theorem, 192, 196 moduli, forward and backward, 286
Levi Coincidence Theorem, non-separable, monotonic correspondence, 71
213 Mospan property, 250
Levi Lemma, generalized, 214 M-space, 44
Levi Open Mapping Theorem, 193 Nagata–Smirnov Theorem, 27
Generalized, 212 Namioka’s Theorem, 197
lifting, 104, 126 natural boundaries, 227, 228, 229
Lindelöf function, 226 Near-closure Theorem, 162, 169
Lindelöf space, 26 Neeb’s Lemma, 190
Liouville numbers, 223 neighbourhood, 34
Lipschitz function, 228 neighbourhood base, 27
Littlewood’s principles, 24 Neumann’s Lemma, 142
 Index 329

Newton’s Principia, 114 Pol’s Product Theorem, 159

Nikodym action, 178 Pol’s Theorem, 39
Nikodym property, 174, 242 polar set, 116, 118, 124
Nikodym’s Corollary, 205 Polish topological group, 197
Nikodym’s Theorem and Corollary, 54 Pólya peaks, 6
No Trumps Theorem, 146 Popa group, 18, 288
No Trumps (NT), 145, 169 Popa operation, 288
Local, 170 potential, 114, 123
Strong, 170 probabilistic method, 37
non-negligible, 137, 169 profound triviality, x, 223
non-separability, 203 projection, 52
norm topology, 91 projective hierarchy, 60
normability, 90 property C, 176
normal numbers, 224 property E, 176
Borel’s theorem, 225 property K, 111
normed group, 87, 174 provably ∆12 , 62, 241
nowhere-locally-separable kernel, 159 pseudo-base, 148
null sequence, 242, 279 pseudocompact space, 26
Effros, 279 pseudocomplete, 111
non-trivial, 243 pseudo-normal space, 26
O’Malley topology, R, 122 qualitative, 223
OMP, Open Mapping Principle, see Open qualitative aspects of category and measure,
Mapping Theorem 234
Open Homomorphism Theorem, 216 quantitative, 223
Open Mapping Theorem, OMT or OMP, 172 quantitative aspects of category and measure,
ordinal, 272 234
initial, 273 quasi everywhere, q.e., 161
limit, 272 quasi-dyadic space, 154
successor, 272 quasi-interior, 174
Ostaszewski’s Club, ♣, 145 quasi-interior of a set, 240
Ostrowski Theorem quasi-regular varying, 19
Combinatorial, 146 quasi-topological, 256
strong, 144 r-topology, 122, 124
outer regularity, 240 Rademacher’s Theorem, 228
p-space, 48 Radon measurable set, 207
strict, 49 Radon property, see inner regularity
parabolic potential theory, 118 Ramsey cardinal, 274
paracompact, 27, 208 Ramsey distance property
paratopological group, 108 strong, 141
PB, having the property of Baire, 231, 269 weak, 141
PD, projective determinacy, 269 Ramsey theory, 141
φ-regularly varying, 288 Ramsey’s Theorem, 274
π-base, 148 random series, 223, 224
π-weight, 154 random Taylor series, 227
Piccard’s Theorem, 236 real flow, 283
Piccard–Pettis Theorem, 138, 139 refinement, 120
plumed space, 48 regular topological space
point completely, 40
heavy, 68 regularly varying, 9
light, 68 Reiter condition, 246
points of essential accumulation, 128 weak, 282
330 Index

Representation Theorem (RT), 9 shrinking determining system, 204

residual set, see set, co-meagre Sierpiński–Erdős duality principle, 222
resolvable set, 122, 219 σ-d decomposable, 208
Reuter’s example, 81 Silver–Mathias Theorem, 121
Riemann zeta function, 226 Simmons–Mospan Converse, 251
right-invariant metric, 87 locally compact separable, 254
satisfaction symbol, |=, 276 simple Steinhaus property, 258
Schröder–Bernstein theorem, 272 singleton-valued map, 54
Second Factorization Theorem, 286 slowly varying, 9
self-equivarying function, 288 Beurling, 13
self-neglecting, SN, 13 slowly varying map, 281
semi-P metric space, 213 Smítal property, 255
semi-analytic space, 196 Smítal’s Lemma, 259
semi-complete metric space, 213 Solecki’s Theorem, 248
Semi-Completeness Theorem, non-separable, Souslin operation, 31, 53
219 Souslin representation
semi-group property, 119 extended, 204
Semi-Group Theorem, 238 Souslin set, 52
semi-Polish space, 196 Souslin’s hypothesis, SH, 271
Semi-Polish Theorem, 196 Souslin’s Theorem, 53
semi-topological, 256 Souslin-graph Theorem, 188
Semigroup Theorem space
Classical, 239 almost complete, 195
Separation Lemma, 59 almost locally ccc, 158
separation properties, 26 almost locally uK–U, 158
sequence analytic, 55
Cauchy, 30 cocompact, 135
fundamental in measure, 30 countably subcompact, 135
sequential condition (for regular variation), dyadic, 153
287 everywhere non-meagre, 42
set Krom, 159
co-meagre, 34 locally meagre, 42
dense, 34 plumed, 48
meagre, 34 pseudocompact, 41
non-meagre, 34 pseudocomplete, 41
nowhere-dense, 34 quasi-dyadic, 154
set of multiplicity, 138 quasi-regular, 41
set of uniqueness, 138 semi-analytic, 196
sets of uniqueness for trigonometric series, 228 semi-Polish, 196
shift-compact, 161 shift-compact, 177
boundedly, 161 universally Kuratowski–Ulam*, 149
convergently, 161 Squared Pettis Theorem, 98, 99
for null sequences, 161 star, 44
shift-compactness, 53, 73, 169, 177 star-refinement, 44
Shift-Compactness Theorem, Theorem Sh, Steinhaus Dichotomy, 139
174, 179 Steinhaus property
for Haar null sets, 248 composite, 258
shift-covered, 99 simple, 258
compactly, 99 Steinhaus refinement, 256
strongly, 99 Steinhaus theorem, 139, 236
shift precompact, 161 combinatorial, 141, 147
 Index 331

Steinhaus–Piccard theorem, 235 topology, O’Malley, see r-topology

Steinhaus–Weil property, 240 translation-continuous measure, 243
Steinhaus–Weil theorem, 138, 239 translator, 129
Steinhaus–Weil theory, 235 trigonometric series, 120
Strong No Trumps Theorem, 170 Tychonoff space, 26
sub-subsequence property, 29 Ulam number, 273
subcompact space, 135 ultrafilter, 64, 127, 130, 200, 234, 271
countably, 135 ultrafilter base, 135
subcontinuity, 239 Uniform Convergence Theorem, UCT, 9, 279
Fuller, 242 uniqueness, 120
Subcontinuity Theorem, 246 universal, 73, 145, 162
subcontinuous measure, 243 universal Kuratowski–Ulam space, uK–U, 148
selectively, 243 universality, 162
Subgroup Theorem, 140 universally Kuratowski–Ulam* (uK–U∗ )
submetrizable, 44 space, 149
submetrizable topology, 25 upper density, 125
subsequence theorem, 29 upper semi-continuous map, 54
Sum–Set Theorem, 239 Urysohn space, 26
Σ0n, Σ1n , 32, 61, 62
superharmonic function, 123 V = L, 53, 267
supermodular measure, 244 Valiron method, 15
symmetric measure, 244 Verification Theorem for D, 166
Verification Theorem for E, 166
tail event, 224
Vitali Covering Theorem, 103
tau-additivity (τ-additivity), 272
Vitali equivalence, 172
Theorem E, 174, 181
von Neumann universe, 270
Theorem EC, 177
Theorem Sh, 174, 179, 181 weak continuity, 94
thin set, 116 weak microtransitive, non-separable, 211
thinning, 10 weakly Archimedean, 80
topological base weight, wt(X), 204
very weak, 120 Wiener’s Tauberian Theorem, 13
weak, 120 Wilczyński, 131
topological completeness, 35 witness, 253
topological dimension, 226 witnessing function, 261
topological group, 108
x-orbit, 176
topology
almost-Lindelöf, 111 Zermelo–Fraenkel set theory, ZF,
analytically heavy, 120 61, 268
cometrizable, 110 zero sharp, 0] , 276
fine, 115, 123 zeta function
quasi-Lindelöf, 111 alternating, 226
submetrizable, 109 Riemann, 226