2020 Book TheDualOfLXLΛFinitelyAdditiveM PDF

SPRINGER BRIEFS IN MATHEMATICS
John Toland
The Dual of L∞(X, L, λ), ),

Finitely Additive Measures
and Weak Convergence
A Primer
SpringerBriefs in Mathematics
Series Editors
Nicola Bellomo, Torino, Italy
Michele Benzi, Pisa, Italy
Palle Jorgensen, Iowa, USA
Tatsien Li, Shanghai, China
Roderick Melnik, Waterloo, Canada
Otmar Scherzer, Linz, Austria
Benjamin Steinberg, New York, USA
Lothar Reichel, Kent, USA
Yuri Tschinkel, New York, USA
George Yin, Detroit, USA
Ping Zhang, Kalamazoo, USA
SpringerBriefs in Mathematics showcases expositions in all areas of mathematics

and applied mathematics. Manuscripts presenting new results or a single new result
in a classical field, new field, or an emerging topic, applications, or bridges between
new results and already published works, are encouraged. The series is intended for
mathematicians and applied mathematicians.
More information about this series at http://www.springer.com/series/10030

SpringerBriefs present concise summaries of cutting-edge research and practical
applications across a wide spectrum of fields. Featuring compact volumes of 50 to
125 pages, the series covers a range of content from professional to academic.
Briefs are characterized by fast, global electronic dissemination, standard publish-
ing contracts, standardized manuscript preparation and formatting guidelines, and
expedited production schedules.
Typical topics might include:
• A timely report of state-of-the art techniques

• A bridge between new research results, as published in journal articles, and a
contextual literature review
• A snapshot of a hot or emerging topic
• An in-depth case study
• A presentation of core concepts that students must understand in order to make
independent contributions
Titles from this series are indexed by Web of Science, Mathematical Reviews, and
zbMATH.
John Toland
The Dual of L1ðX; L; ‚Þ,

and Weak Convergence
A Primer
123
John Toland
Department of Mathematical Sciences
University of Bath
Bath, UK
ISSN 2191-8198 ISSN 2191-8201 (electronic)

SpringerBriefs in Mathematics
ISBN 978-3-030-34731-4 ISBN 978-3-030-34732-1 (eBook)
https://doi.org/10.1007/978-3-030-34732-1
Mathematics Subject Classification (2010): 46E30, 28C15, 46T99, 26A39, 28A25, 46B04
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2020

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Assuming some familiarity with Lebesgue measure, integration and related func-
tional analysis summarised in Chap. 2, this is an exposition of topics that arise
when identifying elements of the dual space of L1 ðX; L; ‚Þ with finitely additive
measures on a r-algebra L when the measure ‚ is complete and r-finite. Such a
representation has its origins in the independent work of Fichtenholz and
Kantorovitch [14] and Hildebrandt [20] and culminated in a general abstract theory
due to Yosida and Hewitt [35] in 1952. However, even now it is not unusual ([16] is
an exception) for books on measure theory to give a detailed account of the dual
space of Lp ðX; L; ‚Þ, 1 6 p\1, while relegating the case p ¼ 1 to references, e.g.
[12, p. 296] which invokes the theory of finitely additive measures on algebras such
as is developed in [35].
An explanation may be that a study of finitely additive measures on algebras
necessitates a possibly unwelcome diversion from the mainstream theory of
countably additive measures that suffices for p 2 ½1; 1Þ. Whatever the reason, a
consequence is that L1 ðX; L; ‚Þ has acquired an aura of mystery, to the extent that
it is often not very clear beyond the mere definition what is meant by saying that a
bounded sequence in L1 ðX; L; ‚Þ is weakly convergent.
The aim here is to take Yosida and Hewitt theory on r-algebras beyond the
representation theorem for L1 ðX; L; ‚Þ , pointing out some of its consequences for
measurable functions generally and in particular for weak convergence of
sequences in L1 ðX; L; ‚Þ. The target audience is anyone who feels nervous about
representing the dual of L1 ðX; L; ‚Þ by finitely additive measures in the knowledge
that there exist uncountably many, linearly independent, finitely additive measures
m > 0 defined on the Lebesgue r-algebra of ð0; 1Þ with the property that
Z 11k Z 1
u dm ¼ 0 for all u 2 L1 ð0; 1Þ and k 2 N; but 1 dm ¼ 1: ð†Þ
0 0
An essential goal will be to come to terms with observations such as this one.
v
vi Preface
In their seminal work, Yosida and Hewitt [35] studied general Banach spaces
L1 ðX; M; N Þ of essentially bounded measurable functions, where measurability is
determined by an algebra M (closed under complementation and finite unions) and
essential boundedness is defined in terms of a family N M (closed under
countable unions with the added property that A B 2 N implies A 2 N ) that
mimics null sets. Obviously, L1 ðX; L; ‚Þ is a special case of L1 ðX; M; N Þ but in
general no measure of any kind is involved in the definition of L1 ðX; M; N Þ.
However, although [35] shows that the dual of L1 ðX; M; N Þ can be expressed in
terms of finitely additive measures, the exposition here is restricted to L1 ðX; L; ‚Þ
because
properties of finitely additive measures on r-algebras are less circumscribed by hypotheses

than on algebras, and replacing the algebra M by a r-algebra L and N by the family of
null sets fE 2 L : ‚ðEÞ ¼ 0g, where ‚ is complete and r-finite, yields a theory which is
relevant in applications, including when X is a Lebesgue measurable subset of Rn or a
differentiable manifold, or when X ¼ N with counting measure.
For a r-finite measure space the ultimate aim is to develop theory sufficient to
characterise weakly convergent sequences in L1 ðX; L; ‚Þ in terms of their ‚-
almost-everywhere pointwise behaviour. However, in the process, when ðX; ‰Þ is a
locally compact Hausdorff topological space and ðX; B; ‚Þ is a corresponding Borel
measure space, there emerges a natural way to localise weak convergence.
A sequence is weakly convergent in L1 ðX; B; ‚Þ if and only if it is weakly con-
vergent at every point x0 in the one-point compactification of ðX; ‰Þ. Here, weak
convergence at x0 is defined in terms of functionals which are zero outside every
neighbourhood of x0 ; for an example of such, see (†).
The essential range RðuÞðx0 Þ1 of a Borel measurable function u at x0 is similarly
defined in terms of those elements of L1 ðX; B; ‚Þ which are localised at x0 . Since
it need not be a singleton, RðuÞðx0 Þ can be interpreted as a multivalued repre-
sentation of the fine structure at x0 of u 2 L1 ðX; B; ‚Þ which is intimately related to
weak convergence.
The Literature
In her Foreword to the monograph by Bhaskara Rao and Bhaskara Rao [6], Dorothy
Maharam Stone cites Salomon Bochner as having said that “contrary to popular
mathematical opinion finitely additive measures were more interesting, more dif-
ficult to handle, and perhaps more important than countably additive measures”.
Oxtoby [25] described [6] as a comprehensive account of finitely additive measures
which effectively organises a large body of material that is widely scattered in the
literature and deserves to be better known, and in their preface the authors them-
selves described it as a reference book as well as a textbook.
RðuÞðx0 Þ is sometimes referred to as the cluster set of u at x0 .

1
Preface vii
The origins of this theory are to be found in the early days of modern integration
theory when there were many contributors: see [12, §III.15, p. 233 and §IV.16,
p. 3882] and the comprehensive bibliography with notes in [6]. However, pre-
sumably because they could not match the versatility of Lebesgue’s theory of
integration and the power of its convergence theorems, finitely additive measures
seem to have fallen out of fashion. Nevertheless, they continue to have significant
roles in, for example, mathematical economics, probability, statistics, optimization,
control theory and analysis [7, 9, 25, 35].
In a series of three papers on additive set functions on abstract topological
spaces, A. D. Alexandroff [2] studied bounded regular finitely additive measures
that represent linear functionals on spaces of continuous functions. On a similar
theme, but in a more general setting, a much-cited reference for the dual of
L1 ðX; L; ‚Þ is Dunford and Schwartz [12, p. 296] which covers the theory of
finitely additive set functions on algebras and includes extensive historical notes.
For a recent account, see Fonseca and Leoni [16, Theorem 2.24], and Aliprantis and
Border [3] for the abstract theory in which it is embedded.
It will soon be apparent that key results, including (†), rely on the axiom of
choice. For a discussion of the role of the axiom of choice, geometrical and
paradoxical aspects of finitely additive measures, and their invariance under group
actions on X, see Tao [32]. Oxtoby’s commentary [25] is of independent interest.
A key role is played throughout by the set G of finitely additive measures that
take only the values 0 and 1 and the observation that L1 ðX; L; ‚Þ is isometrically
isomorphic to a space of real-valued continuous functions on ðG; ¿Þ with the
maximum norm, where ¿ is a compact Hausdorff topology. Further analysis of G in
a Borel setting then leads to localization, and to other developments mentioned
above and outlined in the Introduction.
What follows is in large part an extension of a simplified version of Yosida and
Hewitt [35], set out in the notation and terminology of Chap. 2.
Bath, UK John Toland
Acknowledgements I am indebted to Charles Stuart (Lausanne) for many things including his
encouragement of this project. I am grateful to Anthony Wickstead (Belfast) who obtained for me
a copy of [33] and drew my attention to [34], and to Geoffrey Burton (Bath) and Eugene
Shargorodsky (King’s London) for their interest and many comments. In addition, I would like to
thank Eugene Shargorodsky who contributed Sect. 9.4 and Mauricio Fernández (Stuttgart) who on
a visit to Cambridge asked a question that the account that follows attempts to answer.
2
The reference to Theorem 8.15 on p. 388 is a misprint; 8.15 is a Definition and obviously
Theorem 8.16 was intended.
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Notation and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Partial Ordering and Zorn’s Lemma . . . . . . . . . . . . . . . . . . . . . 7
2.2 Algebras and r-Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Measurable Sets and Measurable Functions . . . . . . . . . . . . . . . 9
2.4 Measures and Real Measures . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Splitting Families of Measurable Sets . . . . . . . . . . . . . . . . . . . . 14
2.6 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Dual Spaces and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.9 Functional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.10 Point Set Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3 L1 and Its Dual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 Finitely Additive Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1 Definition, Notation and Basic Properties . . . . . . . . . . . . . . . . . 31
4.2 Purely Finitely Additive Measures . . . . . . . . . . . . . . . . . . . . . . 36
4.3 Canonical Decomposition: baðLÞ ¼ RðLÞ PðLÞ . . . . . . . . . . 38
4.4 L1 ðX; L; ‚Þ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5 G: 0–1 Finitely Additive Measures . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.1 G and Ultrafilters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 G and the ‚-Finite Intersection Property . . . . . . . . . . . . . . . . . 44
ix
x Contents
6 Integration and Finitely Additive Measures . . . . . . . . . . . . . . . . . . 47

6.1 The Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Yosida–Hewitt Representation: Proof of Theorem 3.1 . . . . . . . . 50
6.3 Integration with Respect to ! 2 G . . . . . . . . . . . . . . . . . . . . . . 51
6.4 Essential Range of u 2 L1 ðX; L; ‚Þ . . . . . . . . . . . . . . . . . . . . . 52
6.5 Integrating u 2 ‘1 ðNÞ with Respect to G . . . . . . . . . . . . . . . . 52
6.6 The Valadier–Hensgen Example . . . . . . . . . . . . . . . . . . . . . . . 54
7 Topology on G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.1 The Space ðG; ¿Þ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.2 L1 ðX; L; ‚Þ and CðG; ¿Þ Isometrically Isomorphic . . . . . . . . . . 58
7.3 Properties of G and ¿ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.4 G and the Weak* Topology on L1 ðX; L; ‚Þ . . . . . . . . . . . . . . 63
7.5 G as Extreme Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8 Weak Convergence in L1 (X, L; ‚) . . . . . . . . . . . . . . . . . . . . . . . . . 67
8.1 Weakly Convergent Sequences . . . . . . . . . . . . . . . . . . . . . . . . 67
8.2 Pointwise Characterisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.3 Applications of Theorem 8.7 . . . . . . . . . . . . . . . . . . . . . . . . . . 72
9 L1 When X is a Topological Space . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.1 Localising G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.2 Localising Weak Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.3 Fine Structure at x0 of u 2 L1 ðX; B; ‚Þ . . . . . . . . . . . . . . . . . . 81
9.4 A Localised Range from Complex Function Theory . . . . . . . . . 84
10 Reconciling Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.1 ðG; ¿Þ Versus L1 ðX; L; ‚Þ . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.2 Restriction to C0 ðX; .Þ of Elements of L1 ðX; B; ‚Þ . . . . . . . . . . 89
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Chapter 1
Introduction
Overview
In a normed linear space V , a sequence {vk } converges weakly to v (vk v) if
v ∗ (vk ) → v ∗ (v) for all v ∗ ∈ V ∗ , the dual space of V and, from the uniform bound-
edness principle, weakly convergent sequences are bounded in norm. However, it
has been known since the work of Banach that when V is a complete normed linear
space it may not be necessary to use all elements of V ∗ when testing for weak con-
vergence. Indeed, when C(Z ) denotes the space of real-valued continuous functions
on a compact metric space Z with the maximum norm, he showed that vk v in
C(Z ) if and only if vk (z) → v(z) for all z ∈ Z and {vk } is bounded. To do so he
observed [5, Annexe, Thm. 7] that Dirac δ-functions satisfy conditions for a set W ∗
in the dual space of a Banach space to have the property that
{vk } bounded and w ∗ (vk ) → 0 for all w ∗ ∈ W ∗ imply vk 0. (W)
When (X, ) is a locally compact Hausdorff space and (C0 (X, ), · ∞ ) is the
Banach space of real-valued continuous functions on X that vanish at infinity (see
(2.9)), weakly convergent sequences are pointwise convergent because δ-functions
belongs to the dual space of C0 (X, ), and bounded by the uniform boundedness
principle. Conversely, from Theorem 2.37 (Riesz) and Lebesgue’s Dominated Con-
vergence Theorem [15, Thm. 2.24], sequences that are norm-bounded and pointwise
convergent on X are weakly convergent.
In particular, when Z is a compact Hausdorff space, for {vk } ⊂ C(Z) (the space
of real-valued continuous functions on Z with the maximum norm)
vk v0 in C(Z) ⇔ sup vk < ∞ and vk (z) → v0 (z) for all z ∈ Z. (V)
k
The possibility of usefully extending these observations to L ∞ (X, L, λ) (the real

Banach space of essentially bounded real-valued functions defined by (2.8)) at first
appears limited because, for example, in an open set ⊂ Rn with Lebesgue measure,
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2020 1
J. Toland, The Dual of L ∞ (X, L, λ), Finitely Additive Measures
and Weak Convergence, SpringerBriefs in Mathematics,
https://doi.org/10.1007/978-3-030-34732-1_1
2 1 Introduction
u k u in L ∞ () implies that u k (x) → u(x) almost everywhere and {u k ∞ } is

bounded, but the converse is false (Lemma 8.4 and Example 8.6). Nevertheless, for
bounded sequences it is shown in Theorem 8.7 that weak convergence is equivalent
to strong convergence in L ∞ (X, L, λ) of elements of a family of related sequences
and the resultant test is illustrated by several examples in Sect. 8.3.
The proof, which inter alia is developed from first principles in the follow-
ing pages, depends on the construction a compact Hausdorff topological space
(G, τ ) with the property that L ∞ (X, L, λ) and C(G, τ ) are isometrically isomorphic
(Theorem 7.4). Here G (Gothic G) is the set of the finitely additive measures that
take only values 0 and 1 on L, and are 0 on null sets in a general measure space
(X, L, λ).
When the essential range of u ∈ L ∞ (X, L, λ) is defined as

R(u) = α ∈ R : λ({x : |u(x) − α| < }) > 0 for all > 0 ,
it is shown that

R(u) = u dω : ω ∈ G
X
and its relation to the isometric isomorphism is given by (7.2) and (7.3). If the setting
is a Borel measure space (X, B, λ) corresponding to a locally compact Hausdorff
spaces (X, ), these considerations can be localised to points x0 in the one-point
compactification X ∞ of X . This in turn leads to the definition of the essential range
of u ∈ L ∞ (X, B, λ) localised at x0 ∈ X ∞ and the possibility of regarding u as multi-
valued at every point while at the same time being single-valued almost everywhere.
These ideas are closely related to the dual of L ∞ (X, B, λ) and weak convergence.
The fact that L ∞ (X, L, λ) and C(G, τ ) are isometrically isomorphic appears to
be at variance with L ∞ (X, L, λ)∗ being represented by finitely additive measures
while C(G, τ )∗ is represented by σ-additive measures. Moreover, when (X, ) is
a locally compact Hausdorff space, elements of L ∞ (X, B, λ)∗ are represented by
finitely additive measures but their restrictions to C0 (X, ) are also represented by
Borel measures. Both these issues will be addressed.
As explained in the Preface, finitely additive measures will be considered only on
σ-algebras. The classical text [12] and the exhaustive monograph [6] both consider
more general situations from the outset, as does Yosida and Hewitt [35] which is the
motivation and main source for this account of the simplified theory. The approach
here is referred to as Yosida–Hewitt theory.
Layout
Chapter 2 is a brief survey of background material, collected with references from
different sources and organised in a consistent notation that will be used in the chap-
ters that follow. Although most of the material is entirely standard and does not need
to be digested unless a need arises in later chapters, some of it may be less famil-
iar. For example, Sect. 2.5 is a slight extension of Kirk’s [22] application of Baire’s
1 Introduction 3
category theorem to show that if (X, L, λ) has no atoms and G is a denumerable fam-
ily of sets of positive
measure, there exists an infinite family of disjoint sets A ∈ L
such that λ(G A) > 0 and λ(G \ A) > 0 for all G ∈ G. This construction is used
in Example 9.11 when considering possibilities for the localised essential range of
u ∈ L ∞ (X, L, λ) at x0 defined in (9.4).
Chapter 3 begins by stating Theorem 3.1, which is the Yosida–Hewitt representa-
tion of the dual space L ∞ (X, L, λ)∗ as the space of finitely additive measures on L
which are zero on sets of zero λ-measure. These finitely additive measures will be
denoted by L ∗∞ (X, L, λ), which should therefore be identified with, but not confused
with, L ∞ (X, L, λ)∗ , the bounded, real-valued, linear functionals on L ∞ (X, L, λ).
From Theorem 3.1, it follows intuitively that if {Ak } ⊂ L is any sequence of mutu-
ally disjoint sets, then χ Ak 0 in L ∞ (X, L, λ) as k → ∞, where χ Ak denotes the
characteristic function of Ak . Moreover, there are finitely additive measures that rep-
resent non-zero bounded linear functionals f on L ∞ (0, 1) with f (v) = 0 when v
is continuous. Since no such countably additive measure can exist because of the
dominated convergence theorem and Theorem 2.35 (Lusin), observations such as
these reflect the delicacy of Theorem 3.1 and the differences between σ-additive and
finitely additive measures.
Chapter 4 introduces basic notation and definitions, for example, of partial order-
ing, lattice operations, absolute continuity and singularity, for finitely additive mea-
sures. The important notion of pure finite additivity of measures follows, and it is
shown that every finitely additive measure is uniquely the sum of a σ-additive and a
purely finitely additive measure.
Chapter 5 introduces the set G of elements in L ∗∞ (X, L, λ) which take only the
values 0 or 1 on L and explains the sense in which every u ∈ L ∞ (X, L, λ) is constant
ω-almost everywhere when ω ∈ G. Elements ω ∈ G give rise to families of sets
Uω = {E ∈ L : ω(E) = 1} which are ultrafilters (Definition 5.3). The collection of
ultrafilters is denoted by U (Gothic U), and there is a one-to-one correspondence
(Theorem 5.4) between G and U. The existence of ultrafilters, and consequently of
elements of G with certain properties, follow from Zorn’s lemma. G will dominate
subsequent developments.
Chapter 6 defines the integral of essentially bounded measurable functions with
respect to the finitely additive measures that featured in Theorem 3.1 (the Yosida–
Hewitt representation theorem).
In Theorem 6.2, it is noted that for all ω ∈ G and u ∈
L ∞ (X, L, λ), u(x) = X u dω, ω-almost everywhere in the sense of finitely additive
measures (Remark 5.2). Observation (†) in the Preface is justified in Remark 6.1.
The chapter ends with an account of the Valadier–Hensgen example [19, 33] of
purely finitely additive measures on [0, 1] that are definitely not σ-additive but which
coincide with σ-additive Lebesgue measure when integrating continuous functions.
Chapter 7 introduces a compact Hausdorff topology τ on G and from theory
already developed, derives the existence, Theorem 7.4, of an isometric isomorphism
between the Banach algebra L ∞ (X, L, λ) and the space of real-valued continuous
functions C(G, τ ). It is immediate that the functionals corresponding to G have
property (W) in the Introduction.
4 1 Introduction
Independently, it is shown that τ coincides with the restriction to G of the weak*

topology on L ∗∞ (X, L, λ) and that G is a closed subset of L ∗∞ (X, L, λ) with the
weak* topology. Also, it is shown that ±G coincide with the extreme points of the
closed unit ball in L ∗∞ (X, L, λ) and consequently Theorem 2.55 (Rainwater) yields
an alternative proof that G has property (W) in the Preface.
Chapter 8 opens with the observation, based of the duality between Dirac mea-
sures acting on C(G, τ ) and elements of G acting on L ∞ (X, L, λ), that u k u 0 in
L ∞ (X, L, λ) if and only if

u k ∞ M and u k dω → u 0 dω as k → ∞ for all ω ∈ G.
X X
It follows that when F : R → R is continuous, u → F(u) is sequentially weakly

continuous [4] on L ∞ (X, L, λ). Necessary pointwise conditions for a sequence to
be weakly convergent in L ∞ (X, L, λ) are given but examples show that they are not
sufficient. However, a necessary and sufficient pointwise condition for a sequence to
be weakly convergent in L ∞ (X, L, λ), Theorem 8.7, follows from Theorem 5.6 and
the fact that any u ∈ L ∞ (X, L, λ) is a constant ω-almost everywhere in the sense
of finitely additive measures when ω ∈ G. In Sect. 8.3, some quite subtle questions
about the weak convergence of specific sequences are settled using Theorem 8.7.
Chapter 9 deals with refinements of the theory to measure spaces (X, B, λ) where
(X, ) is a locally compact Hausdorff space with measure λ on its Borel σ-algebra
B. Prototypical examples of this setting are when X is an open subset of Rn with
the Euclidian metric and Lebesgue measure and X = N with the discrete topology
and counting measure. The key observation is that G = x0 ∈X ∞ G(x0 ) where, for
distinct points x0 ∈ X ∞ (the one-point compactification of X ), the sets G(x0 ) are
closed in (G, τ ) and disjoint, and elements of G(x0 ) are zero outside every open
neighbourhood of x0 ∈ X ∞ .
In Sect. 9.2, this localisation result leads to a characterization of weakly conver-
gent sequences in terms of the pointwise behaviour of related sequences of functions
in neighbourhoods of points of X ∞ . (Here “pointwise” has its usual λ-almost every-
where meaning which is familiar in any measure space, whereas “localisation” refers
to the behaviour of Borel measures and functions restricted to open neighbourhoods
of points in a topological space.)
Writing R(u), u ∈ L ∞ (X, B, λ), as a union of disjoint compact sets

R(u) = R(u)(x0 ) where R(u)(x0 ) = u dω, ω ∈ G(x0 )
x0 ∈X ∞ X
localises the essential range and reflects the fine structure of u at x0 .

As a consequence of Theorem 2.25, if (X, ) is completely separable and (X, B, λ)
has no atoms, there exist u ∈ L ∞ (X, B, λ) such that
u(x) ∈ Q (the rational numbers)
for all x ∈ X yet R(u)(x) is the closed interval 0, u for all x. The chapter ends
1 Introduction 5
with an example due to Shargorodsky of a similar phenomenon occurring naturally

in the theory of complex Hardy spaces, see Sect. 9.4.
Chapter 10 first reconciles the general fact that L ∞ (X, L, λ) and C(G, τ ) are
isometrically isomorphic while the dual of L ∞ (X, L, λ) is represented by finitely
additive measures, whereas the dual of C(G, τ ) is represented by regular, real Borel
measures which are σ-additive. It goes on to consider the special case L ∞ (X, B, λ)
where (X, ) is a locally compact Hausdorff space and B denotes the Borel subsets
of X . In that case, elements of L ∞ (X, B, λ)∗ are represented by finitely additive
measures but when restricted to C0 (X, ) they are also represented by Borel measures.
As seen in Chap. 6, Valadier and Hensgen independently noted that Riemann
sums have Banach limits (Definition 2.41) which on L ∞ [0, 1] are represented by
purely finitely additive measures which are not σ-additive, but which coincide with
the Lebesgue integral for continuous functions. That observation is the motivation
for Sect. 10.2 which considers the relation between the finitely additive measures ν
that yield elements of L ∞ (X, B, λ)∗ (Theorem 3.1) and the Borel measures ν̂ that
by Theorem 2.37 (Riesz) represent their restrictions to C0 (X, ). In particular, those
ν for which ν̂ is singular with respect to λ are characterised in Corollary 10.10 and
a minimax formula for ν̂ in terms of ν is given in Theorem 10.11. It follows that ν̂
may be zero when ν 0 if (X, ) is not compact.
When ω ∈ G, either ω̂ ∈ D (a Dirac measures on X ) or ω̂ may be zero if (X, )
is not compact; if (X, ) is compact ω̂ ∈ D. Note from Remark 6.7 that an arbitrary
Hahn–Banach extension to L ∞ (X, B, λ) of a Dirac δ-function acting on C0 (X, )
need not be in G and from Chap. 9 there may be infinitely many extensions that are
in G.
Starting from (and referring to Chap. 2 only when necessary), a self-contained
approach to the pointwise characterisation of weakly convergent sequences is pre-
sented by
• Chapter 3
• Section 4.1
• Chapter 5, up to Corollary 5.7(a)
• Chapter 6 up to Sect. 6.3
• Chapter 7 up to Sect. 7.2
• Chapter 8
To begin, Chap. 2 reviews background material, cites references and fixes notation.
Chapter 2
Notation and Preliminaries

The set of natural numbers {1, 2, · · · , } is denoted by N, N0 = N {0}, R is the real
numbers, C is the complex numbers, Q is the rational numbers, and extended real
numbers, R {+∞, −∞}, are denoted by R. The empty set is denoted by ∅, and a
set with only one element is called a singleton.
For an arbitrary set S, ℘ (S) denotes the collection of all its subsets, including
the empty set ∅ and S itself. A set S is said to be denumerable or countable if there
is an injection from S into N, and uncountable otherwise. If Sk is denumerable for
all k ∈ N, then k Sk is denumerable. If there is an injection from S into the set
{1, 2, · · · , K } for some K ∈ N, S is said to be finite, and infinite otherwise. If S is
an infinite denumerable set, there is a bijection from S onto N.
2.1 Partial Ordering and Zorn’s Lemma
For a non-empty set S and R ⊂ S × S, a relation on S is defined by writing x y

if (x, y) ∈ R.
Definition 2.1 (Partial and Total Ordering) A partial ordering on S is a relation

which satisfies the following axioms [12, Sect. I.2]:
(i) x x for all x ∈ S,
(ii) x y and y z implies that x z for all x, y, z ∈ S,
(iii) x y and y x implies x = y for all x, y ∈ S.
If in addition at least one of x y or y x holds for every (x, y) ∈ S × S, the partial
ordering is said to be a total ordering on S. If x y but x = y, write x ≺ y.
Example 2.2 The usual ordering ≤ on R is a partial ordering which is a total ordering
on every subset S of R.
https://doi.org/10.1007/978-3-030-34732-1_2
8 2 Notation and Preliminaries
Any collection S of subsets of a set is partially ordered by set inclusion, i.e. A B

if and only if A ⊂ B, but if S is not a singleton it may not be totally ordered.
Definition 2.3 (Upper Bounds and Maximal Elements) If is a partial ordering on

S and A ⊂ S,
(i) u ∈ S is an upper bound of A if a u for all a ∈ A,
(ii) m ∈ A is a maximal element of A if m ⊀ a for all a ∈ A.
Lemma 2.4 (Zorn) Suppose is a partial ordering on S and every totally ordered
subset A ⊂ S has an upper bound. The S has a maximal element.
The proof depends on an assumption that for any collection T of subsets of S there
is a set which contains a point from each of the sets in T . More precisely:
The Axiom of Choice
Let S be non-empty and T a non-empty subset of ℘ (S) \ {∅}. Then there exists a
function f : T → S such that f (T ) ∈ T for all T ∈ T .
Conversely, the Axiom of Choice can be proved using Zorn’s lemma. So these two
seemingly different statements are in fact equivalent formulations of a fundamental
axiom of set theory.
Definition 2.5 (Equivalence Relation) A relation ∼ on a set S is called an equiv-
alence relation if for all x, y, z ∈ S, (i) x ∼ x; (ii)x ∼ y ⇔ y ∼ x; (iii) x ∼ y
and y ∼ z ⇒ x ∼ z. The set [x] = y ∈ S : x ∼ y is the equivalence class which
contains x.
2.2 Algebras and σ-Algebras
For an arbitrary set X , a collection M ⊂ ℘ (X ) is an algebra if

∅ and X are in M,
A ∈ M if and only X \ A ∈ M,
K
k=1 Ak ∈ M when Ak ∈ M, k = 1, 2, · · · , K , K ∈ N.

If M is an algebra and A, B ∈ M, it follows that A B ∈ M, A \ B ∈ M and M
is closed under finite unions and intersections.
A σ-algebra, denoted by L, is an algebra with the additional property that

Ak ∈ L, k ∈ N, implies k∈N Ak ∈ L.
Hence, a σ-algebra is closed under complementation, and countable unions and
intersections.
Examples.
For any set X , {∅, X } and ℘ (X ) are σ-algebras.
With X = [0, 1), the collection of unions of finitely many intervals of the form
[a, b), 0 ≤ a ≤ b ≤ 1, is an algebra, but not a σ-algebra.
2.2 Algebras and σ-Algebras 9
M ⊂ ℘ (N), defined by A ∈ M if and only if either A or N \ A is finite, is an

algebra. Clearly,
the singletons {2n + 1} ∈ M for all n ∈ N0 . So M is not a σ-algebra
because n∈N0 {2n + 1} ∈ / M.
On the other hand, saying A ∈ L if and only if A or X \ A is denumerable defines
a σ-algebra L on any set X. (When X is uncountable, L = ℘ (X ).)
It is obvious that if Lκ , κ ∈ K, where K is an arbitrary set, are σ-algebras, L :=
κ∈K Lκ is also a σ-algebra. Therefore, since ℘ (X ) is a σ-algebra, for any V ⊂ ℘ (X )
the intersection of all σ-algebras which contain V is the smallest σ-algebra containing
V, called the σ-algebra generated by V.
In particular, an algebra M generates a σ-algebra, denoted by M in [35]. Another
special case arises when (X, ) is a topological space, ⊂ ℘ (X ) being the collection
of open sets in X . Then the σ-algebra generated by is called the Borel σ-algebra
of (X, ), often denoted by B and elements of B are called Borel subsets of X .
2.3 Measurable Sets and Measurable Functions
A measurable space is a pair (X, L) where X = ∅ is arbitrary and L ⊂ ℘ (X ) is

a σ-algebra. Elements of L are referred to as measurable sets and a function u :
X → R is said to be measurable if

x : u(x) > α ∈ L for all α ∈ R.
The set of measurable functions has the following properties.

(i) If u n , n ∈ N, is measurable and u n (x) → u(x) for all x ∈ X ,
u is measurable.
(ii) If {u n } is a sequence of extended real-valued measurable functions,
u(x) := inf u n (x) and u(x) := sup u n (x), x ∈ X, are measurable.

n n
(iii) If u is measurable, u ± are measurable where u ± (x) := sup{±u(x), 0}, so that

u ± (x) ≥ 0, x ∈ X , and
u = u + − u − and |u| = u + + u − . (2.1)
(iv) If u, v are real-valued and measurable, c ∈ R and g : R → R is continuous,

then cu, g(u) = g ◦ u, u ± , uv, u + v are measurable.
(v) If u is non-negative and measurable, there is a sequence {ϕn } of non-negative
measurable functions on X with
(a) 0 ≤ ϕn (x) ≤ ϕn+1 (x) ≤ u(x), x ∈ X, n ∈ N,
(b) u(x) = limn ϕn (x), x ∈ X ,
(c) ϕn is real-valued with only finitely many values for each n ∈ N.
The characteristic function χ A of a set A ∈ L, defined by

1 if x ∈ A
χ A (x) = ,
0 otherwise
is a measurable function. Finite linear combinations of characteristic functions

K
ϕ(x) = ak χ Ak , ak ∈ R, Ak ∈ L, K ∈ N, (2.2)
k=1
are called simple functions and are measurable. The functions ϕn in (v) above are
non-negative simple functions.
2.4 Measures and Real Measures
In the measure theory literature, subtly different meanings are often assigned by
different authors to the same terminology. For example, in [15, Chap. 3, p. 85] a
signed measure may have infinite values, whereas in [28, Sect. 6.6, p. 119] it may
not. See also Remark 2.38. For this reason, the terminology chosen for subsequent
chapters is described in some detail below.
Definition 2.6 (Measures) In a measurable space (X, L), a measure λ is an extended
real-valued function on L with
λ(X ) > 0, λ(∅) = 0, λ(A) ≥ 0 for all A ∈ L, and
when E k ∈ L, k ∈ N, and E i E j = ∅, i = j,

λ Ek = λ(E k ). (2.3)
k∈N k∈N
When L is a σ-algebra this identity is referred to as the σ-additivity of λ. Some texts

use the term “positive measure” for what here is referred to as a measure.
The triple (X, L, λ) is called a measure space. A measure is said to be finite if
λ(X ) < ∞, and σ-finite if X = n∈N X n where X n ∈ L and λ(X n ) < ∞ for all n.
In a measure space a set E ∈ L is null, written as E ∈ N , if λ(E) = 0, and a property
is said to hold λ-almost everywhere if the exceptional set where it does not hold is
null. A measure space is said to be complete if A ⊂ B ∈ N implies that A ∈ N .
Definition 2.7 In a measure space (X, L, λ), a sequence {u k } of measurable func-
tions is said to converge in measure to a measurable function u if for all α > 0

λ {x : |u k (x) − u(x)| > α} → 0 as k → ∞.
A sequence of measurable functions which converges in measure has a subsequence

which converges pointwise λ-almost everywhere [15, Thm. 2.30].
2.4 Measures and Real Measures 11
Definition 2.8 (Atom) A ∈ L is an atom if A ∈

/ N and A ⊃ B ∈ L implies that
either λ(B) = 0 or λ(B) = λ(A).
Remark 2.9 An atom A in a σ-finite measure space has finite measure. If λ(A) < ∞
and u is a bounded measurable function let

a = inf b ∈ R : λ({x ∈ A : u(x) ≤ b}) = λ(A) .

Then λ {x ∈ A : u(x)
≤ a} = λ(A) and λ {x ∈ A : u(x) < a} = 0, since λ
is σ-additive and λ {x ∈ A : u(x) ≤ a − 1/k} = 0 for all k. In other words u = a,
a constant, λ-almost everywhere on A. The next result is used in Sect. 2.5.
Lemma 2.10 If (X, L, λ) has no atoms, for all G ∈ L with λ(G) > 0 and > 0
there exists a subset E ∈ L of G with λ(E) ∈ (0, ].
Proof Since there are no atoms, the given G can be replaced by a subset, also denoted
by G ∈ L, with λ(G) ∈ (0, ∞). Let > 0. Now since G is not an atom, there exists
E 1 ⊂ G, E 1 ∈ L, with λ(E 1 ) ∈ (0, λ(G)). If λ(E 1 ) > for all such E 1 it follows that
λ(E 1 ) > and λ(G \ E 1 ) > . Since there are no atoms,
by the same argument there
exists E 2 ⊂ G \ E 1 with λ(E 2 ) > and λ(G \ (E 1 E 2 )) > . By induction there
is a sequence {E k } ⊂ L of disjoint subsets of G with λ(E k ) > . Since λ(G) < ∞
this is false. Hence, there exists E ⊂ G, E ∈ L with λ(E) ∈ (0, ].
Remark 2.11 Lemma 2.10 implies the Darboux property of atomless measures [8,
Cor. 1.12.10] which is apparently stronger: If (X, L, λ) has no atoms and λ(F) > 0
for some F ∈ L, then for any a ∈ (0, λ(F)) there exists E ∈ L with E ⊂ F and
λ(E) = a. (The prototype is due independently to Fichtenholz [13] and Sierpiński
[30].)
Definition 2.12 (Regular Borel Measure) When B is a Borel σ-algebra on a topo-
logical space (X, ), a measure λ on B is called a Borel measure. For B ∈ B, a Borel
measure is said to be

outer regular on B if λ(B) = inf λ(U ) : B ⊂ U ∈ ,

inner regular on B if λ(B) = sup λ(K ) : K ⊂ B, K compact .
A Borel measure which is both outer and inner regular on every B ∈ B is called
regular.
Example 2.13 (Dirac Measures) For any measurable space (X, L) and x ∈ X , let
δx be defined on L by δx (E) = 1 if x ∈ E and δx (E) = 0 otherwise. Then δx is a
measure, called a Dirac measure at x, and (X, L, δx ) is a finite, complete measure
space in which {x} is an atom.
Example 2.14 (Counting Measure on N) Let X = N, L = ℘ (N) and define λ(E) as
the number of elements in E ⊂ N if E is finite, and λ(E) = +∞ if E is infinite. Then
(N, ℘ (N), λ) is a σ-finite, complete measure space, λ is called counting measure and
every singleton {n}, n ∈ N, is an atom.
Example 2.15 (Lebesgue Measure on Rn ) Let B denote the Borel subsets of X = Rn

with the standard metric and (Rn , B) the corresponding measurable space. Then
although there is a unique σ-finite measure λ on B which coincides with the
n-dimensional volume of balls in Rn , it is not complete. However, a complete, σ-finite
measure space (Rn , L, λ) is defined as follows:

L= B A : B ∈ B, A ⊂ B ∈ B, λ (B ) = 0 ,

λ B A = λ (B), B A ∈ L.
This is the classical Lebesgue measure space and λ is Lebesgue measure on Rn .

Thus every Borel measurable function is Lebesgue measurable, and every Lebesgue
measurable function is equal almost everywhere to a Borel measurable function.
To see that Rn with Lebesgue measure has no atoms, suppose E ∈ L and
λ(E) > 0. Now define a function f : [0, ∞) → R by f (r ) = λ E Br where
Br denotes the ball of radius r centred at 0 in Rn . From the σ-additivity of λ, it
follows that f is continuous with f (0) = 0 and f (r ) → λ(E) > 0 as r → ∞. So,
by the intermediate value theorem, there exists r̂ with

λ E Br̂ = f (r̂ ) = λ(E)/2 ∈ (0, λ(E)),
which shows that E is not an atom.

If Ω ∈ L, a typical case
being when Ω is an open subset of R , the restriction
n
of λ to the σ-algebra {Ω E : E ∈ L} creates a complete measure space which is

denoted by (Ω, L, λ).
Definition 2.16 (Real Measures) In a measurable space (X, L), a real measure [28,
Sects. 1.8 and 6.6] is a real-valued function μ (not necessarily one-signed) on L with
μ(∅) = 0 and

μ Ei = μ(E i ) when E i ∈ L and E i E j = ∅, i = j ∈ N.
i∈N i∈N
Since the left side is independent of the ordering of i ∈ N, the sum of the series on
the right is the same if i is replaced by σ(i), where σ is a permutation of N. Hence,
the series is absolutely convergent.
For a real measure μ and E ∈ L let

∞ ∞

|μ|(E) = sup |μ(E i )| : E = E i , E i ∈ L, E j E j = ∅, i = j ∈ N .
i=1 i=1
Then |μ(E)| ≤ |μ|(E), E ∈ L, and |μ| is a measure in the sense of Definition 2.6
[28, Thm. 6.2] with |μ|(X ) < ∞ [28, Thm. 6.4]. Hence μ± are real measures where
2.4 Measures and Real Measures 13
0 ≤ μ+ := 21 (|μ| + μ) and 0 ≤ μ− := 21 (|μ| − μ), (2.4)
whence
μ = μ+ − μ− and |μ| = μ+ + μ− , 0 ≤ μ± (E) ≤ |μ|(X ) < ∞, E ∈ L.
|μ| is called the total variation of μ, μ± are the positive and negative parts of μ, and
μ = μ+ − μ− is the Jordan decomposition of μ.
Theorem 2.17 (Hahn Decomposition [28, Thm 6.14]) Let μ be a real
measure on a
σ-algebra L. Then there exist A± ∈ L such that A+ A− = X , A+ A− = ∅ and
for E ∈ L,
μ+ (E) = μ(A+ E), μ− (E) = −μ(A− E)
where μ± are defined in (2.4).

Definition 2.18 (Absolute Continuity and Singularity) A real measure μ is abso-
lutely continuous with respect to a measure λ, written as
μ λ if and only if λ(E) = 0 implies |μ|(E) = 0
and singular with respect to λ, written as
λ ⊥ μ if and only if λ(E) + |μ|(X \ E) = 0 for some E ∈ L.
This notation will be generalised in Definition 4.6 and Remark 4.7 to accommodate
finitely additive measures that are not σ-additive. In Chap. 4, real measures on L are
seen as examples of finitely additive measures, in which context they are denoted by
Σ(L).
Remark 2.19 Note that in [15, Chap. 7] the term “signed measure” allows μ to take
one, but not both, of the values ±∞. Thus in that terminology a real measure would
be a signed measure but a signed measure might not be a real measure.
Example 2.20 Any finite linear combination of finite measures is a real measure. A
class of real measures that are absolutely continuous with respect to λ will be defined
in Remark 2.27.
Definition 2.21 (Regular Real Borel Measures) Let B denote the Borel σ-algebra
of a locally compact Hausdorff space (X, ). Then a real Borel measure μ is said to
be regular if both the measures μ± are regular in the sense of Definition 2.12: for all
B∈B

μ± (B) = inf μ± (U ) : B ⊂ U ∈ = sup μ± (K ) : K ⊂ B, K compact ,
where μ± are defined by (2.4).

2.5 Splitting Families of Measurable Sets
This section concerns measurable sets which in an arbitrary measure space (X, L, λ)
split families of measurable sets in the following sense.

Definition 2.22 (Splitting Measurable Sets) A ∈ L splits G ⊂ L \ N if λ(G A) >
0 and λ(G \ A) > 0 for all G ∈ G.
Remark 2.23 If A splits G, then X \ A also splits G. Moreover, if G1 , G2 ⊂ L \ N
are such that for each G 1 ∈ G1 there exists G 2 ∈ G2 with G 2 ⊂ G 1 , then A splits G1
if A splits G2 .
The main result, Theorem 2.25, depends on Baire’s category theorem. To set the
scene define an equivalence relation ∼ on L by writing
E ∼ F if an only if EΔF is
null, i.e. the symmetric difference EΔF = (E \ F) (F \ E) ∈ N . For E, F ∈ L
let
−1
−1
∂(E, F) = tan λ(EΔF) = tan |χ E − χ F | dλ , (2.5)
X
where χ A is the characteristic function of A ∈ L and tan−1 maps R onto [−π/2, π/2].
Since L 1 (X, L, λ) is a Banach space and since a convergent sequence in L 1 (X, L, λ)
has a subsequence that converges pointwise λ-almost everywhere,
it is immediate
that the space (L, ∂) is a complete metric space. Let L̂ = G ∈ L : λ(G) ∈ (0, ∞)
and for G ∈ L̂ let

F(G) = E ∈ L : λ(E G) ∈ {0, λ(G)} .
The following simple adaption of Kirk’s argument [22] yields Theorem 2.25, Lemma
7.9 and Example 9.11.
Lemma 2.24 For any G ∈ L̂ the set F(G) is closed with empty interior in (L, ∂).
In other words F(G) is nowhere dense in (L, ∂).

Proof For G ∈ L̂ let g : L → R be given by g(E) = λ(G E). Then g is continuous
because for E 1 , E 2 ∈ L,

|g(E 1 ) − g(E 2 )| = χG E1 − χG E2 dλ
X

≤ |χ E1 − χ E2 | dλ = tan ∂(E 1 , E 2 ) .
X
So F(G) = g −1 {0, λ(G)} is closed in (L, ∂). To see that F(G) is nowhere dense let
E ∈ F(G) and > 0 be arbitrary. By Lemma 2.10 there exists E ⊂ G with E ∈ L
) < min{tan , λ(G)}.
and 0 < λ(E
If λ(E G) = 0 let E + = E E , so that 0 < λ(E + G) = λ(E ) < λ(G)
and ∂(E, E + ) = tan−1 λ(E \ E) ≤ tan−1 λ(E ) < .
2.5 Splitting Families of Measurable Sets 15

In the other case, when λ(E G) = λ(G), let E − = E \ E , so that

λ(E − G) = λ((E G) \ E )) = λ(G \ E ) = λ(G) − λ(E ) ∈ (0, λ(G))

and ∂(E, E − ) = tan−1 λ(E \ E − ) = tan−1 λ(E E ) < .
In both cases E ± ∈ L with ∂(E ± , E) < but E ± ∈
/ F(G). Hence, F(G) is closed
with empty interior when G ∈ L̂, as required.
Since Lebesgue measure on Rn has no atoms and counting measure on N has

infinitely many atoms, the following result applies to the former but not the latter.
Theorem 2.25 If (X, L, λ) has no atoms and G is a denumerable subset of L \ N ,

there exists an infinite sequence {Ai } in L of mutually disjoint sets such that Ai splits
G for all i ∈ N.
Proof Let G = {G j : j ∈ N}. Since there are no atoms, each G j has a subset G̃ j
with λ(G̃ j ) ∈ (0, ∞). So, by Remark 2.23, it suffices to assume that G j ∈ L̂ for
all j. Then, by Baire’s category theorem, L = j∈N F(G j ) ⊂ L since, by Lemma
2.24,each F(G j ) is nowhere dense in the complete metric space (L, ∂). If A1 ∈
L \ j∈N F(G j ) then, from the definition of F(G j ), A1 splits G.
Now let X 1 = X \ A1 , L1 = {E \ A1 : E ∈ L) and G1 = {G \ A1 : G ∈ G). Then
G1 ⊂ L1 \ N satisfy the hypotheses of the theorem in the measure space (X 1 , L1 , λ).
So there exists A2 ⊂ X \ A1 , A2 ∈ L1 , which splits G1 . By construction A2 A1 = ∅
and A2 splits G by Remark 2.23. Iteration of this argument yields {Ai : i ∈ N} ⊂ L
as in the statement of the theorem.
2.6 Integration
The Lebesgue integral (which may be infinite) of a non-negative simple function

(2.2) with respect to a measure γ on L is defined in an obvious way by

K
ϕ dγ = ak γ(Ak ),
X k=1
where ak ≥ 0 and γ(A) ≥ 0 for all A ∈ L. Since the functions ϕn in Sect. 2.3 (v) are
non-negative simple functions, the integral of a non-negative measurable function u
with respect to non-negative γ may be defined by

u dγ = lim ϕn dγ. (2.6)
X n→∞ X
Because γ is σ-additive it is easily seen that the limit on the right is independent of
the sequence {ϕn } which satisfies (v)(a)–(c).
Remark 2.26 Note that for u ∈ L ∞ (X, L, λ) this definition coincides with the
definition in Sect. 6.1 when γ is σ-additive, but not when γ is only finitely addi-
tive. This is because
χ(0,1−1/k) (x) 1 for all x ∈ (0, 1),
but (†) in the Preface (established in Remark 6.1) contradicts (2.6). Indeed (†) shows
that the limit in (2.6) may not be independent of the sequence {ϕn } when γ is not
σ-additive.
Finally, if γ is either a measure (in which case γ − = 0) or an arbitrary real measure,

a measurable function u is said to be Lebesgue integrable with respect to γ if and
only if

u + dγ + , u − dγ + , u + dγ − , u − dγ − are all finite,
X X X X
in which case

u dγ := u + dγ + − u − dγ + − u + dγ − + u − dγ − .
X X X X X
It follows that u is integrable with respect to γ if and only if |u| is integrable with
respect to |γ| and

u dγ ≤ |u| d|γ| < ∞.

X X
Moreover

|u| dλ = 0 if and only if u is zero λ-almost everywhere. (2.7)
X
Remark 2.27 When (X, L, λ) is a measure space and h : X → R is any integrable

function, a real measure μh is defined on L by

μh (E) = hχ E dλ, E ∈ L, and μh λ.
X
In fact part (b) of the next result indicates that all real measures μ that are absolutely
continuous with respect to λ arise in this way.
Theorem 2.28 (Lebesgue–Radon–Nikodym [15, Thm. 3.8], [28, Sect. 6.10]) For a
σ-finite measure λ and a real measure μ on L there is
(a) a unique pair μa , μs of real measures with μ = μa + μs , μa λ and μs ⊥ λ
(Definition 2.18), and if μ is positive then so are μa and μs , and
2.6 Integration 17
function h which is integrable with respect to λ with the property

(b) a unique
that μa (E) = E h dλ for all E ∈ L.
Part (a) is commonly referred to as the Lebesgue decomposition of μ relative to λ

and (b) is the Radon–Nikodym theorem.
2.7 Function Spaces
For spaces of pth-power integrable functions, the notation is that of [6, 12, 35],
namely, L p rather than L p , because for dual spaces L ∗p is easier to manage than
L p ∗ . With the exception of Sect. 9.4 only spaces of real-valued functions will be
considered.
L p -Spaces. When (X, L, λ) is a measure space an equivalence relation ∼ is defined
on measurable functions by u ∼ v if and only if u = v λ-almost everywhere. For
convenience with notation, it is usual to denote the equivalence class [u] (Definition
2.5) simply as u and to refer to it as a function rather than an equivalence class of
functions.
Let L p (X, L,λ), p ∈ [1, ∞), denote the (equivalence classes of) measurable
functions u with |u| p dλ < ∞. Then it is well known that with norm
1/ p
u p := |u| p dλ , p ∈ [1, ∞),
L p (X, L, λ) is a real Banach space. It is obvious that u k − u p → 0 as k → ∞

for any p ∈ [1, ∞) implies that u k → u in measure (Definition 2.7), and hence for
a subsequence that u k j (x) → u(x) λ-almost everywhere.
Example 2.29 Let X = (0, 1) with the usual metric and Lebesgue measure and con-
sider the intervals

I1 = 0, 1 ;

1
1
I2 = 0, ; I3 = ,1 ;
2 2

1
1 2
2
I4 = 0, ; I5 = , ; I6 = ,1 ;
3 3 3 3

1
1 1
1 3
3
I7 = 0, ; I8 = , ; I9 = , ; I10 = ,1 ;
4 4 2 2 4 4
and so on. Note that λ(Ik ) ≤ 1/(n + 1) if k > n(n + 1)/2 (the sum of the first n
natural numbers). Let u k = χ Ik . Then u k → 0 in L p (0, 1) for all p ∈ [1, ∞), from
which it follows that u k → 0 in measure, and hence there exists a subsequence with
u k j (x) → 0 for λ-almost all x ∈ (0, 1). However, it is obvious from the construction
that
lim sup u k (x) = 1 for λ-almost all x ∈ (0, 1).

k→∞
A measurable function u is said to be essentially bounded if |u(x)| ≤ c, a constant,

λ-almost everywhere. With norm

u∞ = inf c > 0 : |u(x)| ≤ c λ-almost everywhere (2.8a)
the space L ∞ (X, L, λ) of (equivalence classes of) essentially bounded measurable

functions is a real Banach space. Let

[u]+
∞ = inf c : u(x) ≤ cλ-almost everywhere , (2.8b)

[u]−
∞ = sup c : u(x) ≥ cλ-almost everywhere . (2.8c)
Then [u]− ≤ u(x) ≤ [u]+ for λ-almost all x ∈ X .
Definition 2.30 (Local Compactness [28, Defn. 2.3]) A topological space (X, ) is
locally compact if for every x ∈ X there is an open set G with x ∈ G and the closure
G of G is compact.
Definition 2.31 (σ-Compactness

[28, Defn. 2.16]) A topological space (X, )
is σ-compact if X = k∈N K k for a denumerable family of compact sets K k .
Definition 2.32 (One-point Compactification [21, p. 150]) The one-point compact-

ification of a topological
space (X, ) is the compact space (X ∞ , ∞ ) defined by
putting X ∞ = X {x∞ } where x∞ ∈ / X (x∞ is referred to as the point at infinity)
and saying G ⊂ X
∞ is open, i.e. G ∈ ∞ , if either G ⊂ X and G ∈ , or x∞ ∈ G
and G = {x∞ } (X \ K ) for a compact set K in (X, ).
Remark 2.33 (X ∞ , ∞ ) is Hausdorff if and only if (X, ) is locally compact and

Hausdorff, and (X, ) is compact if and only if {x∞ } is isolated (both open and
closed) in (X ∞ , ∞ ) [21, loc. cit.].
Continuous Function Spaces

For a locally compact Hausdorff space (X, ), let C0 (X, ) denote the linear space
of real-valued continuous functions v with the property that for all > 0 there exists
a compact set K ⊂ X such that |v(x)| < for all x ∈ X \ K . With the maximum
norm
v∞ = max |v(x)|, v ∈ C0 (X, ), (2.9)
x∈X
C0 (X, ) is a real Banach space. When X is compact C0 (X, ) consists of all the real-
valued continuous functions on X . When (X, ) is not compact C0 (X, ) consists
of the restrictions to X of real-valued functions that are continuous on the compact
space (X ∞ , ∞ ) and zero at x∞ .
2.7 Function Spaces 19
Remark 2.34 In a Hausdorff topological space (X, ) with Borel σ-algebra B every
singleton {a} is closed, and hence χ{a} , a ∈ X , is Borel measurable. Now suppose
that all Borel measurable functions are continuous. Then {a} = χ−1 {a} (0, 2) is open,
because χ{a} is continuous. Therefore, all subsets of X are open, and hence = ℘ (X )
( is the discrete topology) when all measurable functions are continuous. Conversely,
if is the discrete topology every function is continuous. An important case is given
by Example 2.14.
The following observation is a corollary of Lusin’s theorem.
Theorem 2.35 ([28, Cor. 2.24]) Suppose (X, ) is a locally compact Hausdorff
space, λ(X ) < ∞ and λ is a complete regular real Borel measure. Then for u ∈
L ∞ (X, B, λ) there exists a sequence {gk } of continuous functions on (X, ) such that
each gk has compact support (the closure of {x : gk (x) = 0} is compact), gk ∞ ≤
u∞ and u(x) = limk→∞ gk (x) for λ-almost all x ∈ X .
2.8 Dual Spaces and Measures
Given a normed linear space (V, · ), a metric is defined on V by ρ(x, y) = x −

y, x, y ∈ V , and V is a Banach space if (V, ρ) is complete (i.e. Cauchy sequences
converge). A linear function f : V → R is continuous if and only if there is a constant
M such that | f (x)| ≤ Mx for all x ∈ V . The set of all such continuous linear
functions, which are usually called bounded linear functionals, forms a linear space
V ∗ over R, and V ∗ , the dual space of V , is a Banach space (whether or not V is a
Banach space) when endowed with the norm

f V ∗ = sup | f (x)| : x ≤ 1, x ∈ V .
In the special case when V is a space of real-valued functions v on X there has

historically been a great deal of interest in describing the action of f ∈ V ∗ on v ∈ V
as an integral of v over X . Indeed much of the subsequent chapters concern the
consequences of such a description of L ∞ (X, L, λ)∗ . The classical result for the
dual space L p (X, L, λ)∗ , p ∈ [1, ∞), a Banach space with norm

f L p (X,L,λ)∗ = sup | f (u)| : u p ≤ 1, u ∈ L p (X, L, λ) , (2.10)
is the following.
Theorem 2.36 ([15, Thm. 6.15]) For 1 ≤ p < ∞ let f ∈ L p (X, L, λ)∗ . Then for
p > 1 there exists a unique g ∈ L q (X, L, λ) such that for all u ∈ L p (X, L, λ)

p
f (u) = ug dλ where q = and f L p (X,L,λ)∗ = gq .
X p−1
If (X, L, λ) is σ-finite, for p = 1 there exists g ∈ L ∞ (X, L, λ) such that

f (u) = ug dλ for all u ∈ L 1 (X, L, λ) and f L ∞ (X,L,λ)∗ = g∞ .
X
Therefore, L p (X, L, λ)∗ and L q (X, L, λ) are isometrically isomorphic when p ∈

(1, ∞) and p −1 + q −1 = 1, and also in the limiting case p = 1 and q = ∞ if
(X, L, λ) is σ-finite. For a necessary and sufficient for L ∞ (X, L, λ) to be the dual
space of L 1 (X, L, λ), see [16, Cor. 2.41].
The dual space C0 (X, )∗ is a Banach space with norm

f C0 (X,)∗ = sup | f (v)| : v∞ ≤ 1, v ∈ C0 (X, ) . (2.11)
Theorem 2.37 (Riesz [28, Thm. 6.19]) For f ∈ C0 (X, )∗ there exists a unique,
regular, real Borel measure μ on X (Definition 2.21) with

f (v) = v dμ for all v ∈ C0 (X, ). (2.12)
X
Moreover, |μ|(X ) = f C0 (X,)∗ < ∞ and f (v) ≥ 0 when 0 ≤ v ∈ C0 (X, ) implies
that μ ≥ 0 on Borel sets.
The space of all regular, real Borel measures on B is denoted by C0∗ (X, τ ) and
for every μ ∈ C0∗ (X, ), (2.12) defines f ∈ C0 (X, )∗ .
Remark 2.38 Although the statements are different because the nomenclature is
different, the Riesz representation theorems in [15, Thm. 7.17] and [28, Thm. 6.19]
are equivalent to Theorem 2.37.
2.9 Functional Analysis
Theorem 2.39 (Hahn–Banach [12, p. 62, II.3.10]) On a real linear space V let
p : V → R be sublinear:
p(x + y) ≤ p(x) + p(y), p(αx) = α p(x) for all x, y ∈ V, α ∈ [0, ∞),
and suppose f : Y → R is linear with f (y) ≤ p(y) for all y ∈ Y , where Y is a

linear subspace of V . Then there is a linear function f˜ : V → R with
f˜(y) = f (y) for all y ∈ Y and f˜(x) ≤ p(x) for all x ∈ V.
It follows that − p(−x) ≤ f˜(x) ≤ p(x) for all x ∈ V .

f˜ is known as as a Hahn–Banach extension of f . The following is immediate from
the fact that p(x) = x is sublinear on a normed linear space.
2.9 Functional Analysis 21
Corollary 2.40 Let Y be a closed linear subspace of a normed linear space V and
let v0 ∈ V \ Y . Then there exists f ∈ V ∗ such that f (v0 ) = 1 and f (y) = 0 for all
y ∈ Y.
Proof Let Z = {y + αv0 : y ∈ Y, α ∈ R}, a linear subspace of V . Since v0 = 0,

z ∈ Z determines a unique α ∈ R and y ∈ Y such that z = y + αv0 and since Y is
closed, a bounded linear functional g is defined on Z by
g(z) = g(y + αv0 ) = α, z ∈ Z .
By Theorem 2.39 (Hahn–Banach), there exists f ∈ V ∗ which coincides with g on

Z . Since g(v0 ) = 1, f has the required properties.
Definition 2.41 (Banach Limits [5, p. 34, Chap. II]) As in Example 2.14 and Remark
2.34, let ∞ (N) denote the space of bounded, real sequences indexed by N. Now
let c(N) be the subspace of convergent sequences and let l(u) denote the limit of
{u(k)} ∈ c(N). Since l : c(N) → R is positive, bounded and linear, there exists [11,
Thm. III.7.1] a positive bounded linear functional L on ∞ (N) such that for u ∈
∞ (N),
L(u) = l(u), u ∈ c(N), L(u) = L(u n ) where u n (k) = u(k + n), n ∈ N,
and L∞ (N) = lc(N) = 1. L : ∞ (N) → R is called a Banach limit; see [31] for
development of the theory. For an account based on finitely additive measures, see
Remark 6.5.
Let C(Z) with the maximum norm denote the Banach space of real-valued continuous
functions on a compact Hausdorff space (Z, ).
Theorem 2.42 (Stone–Weierstrass [11, Thm. V.8.1], [15, Thm. 4.45]) Suppose A ⊂
C(Z) is a linear space in which (i) f, g ∈ A implies the product f g ∈ A; (ii) the
constant function 1 ∈ A; and (iii) for z 1 = z 1 ∈ Z there exists f ∈ A such that
f (z 1 ) = f (z 2 ). Then A = C(Z), where A is the closure of A in C(Z).
Theorem 2.43 (Weierstrass) For a continuous real-valued function f on [a, b] and

> 0 there exists a polynomial p on [a, b] with | f (x) − p(x)| < for all x ∈ [a, b].
Definition 2.44 (Weak Convergence) In a normed linear space V a sequence {yk }

converges weakly to y0 in V , written as yk y0 , if v ∗ (yk ) → v ∗ (y0 ) for all
v∗ ∈ V ∗.
By the uniform boundedness principle (also known as the Banach–Steinhaus theo-

rem) [28, Thm. 5.8], yk y0 in V implies {yk } is bounded.
Theorem 2.45 (Mazur [23, Chap. 10, Thm. 6]) If {yk } ⊂ K , where K is closed and
convex in a normed linear space, and yk y0 , then y0 ∈ K .
Corollary 2.46 If yk y0 in a normed linear space V and {k j } ⊂ N is strictly

increasing, there exists {y i } with y i − y0 → 0 as i → ∞, where

mi
mi
yi = γ ij yk j , γ ij ∈ [0, 1] and γ ij = 1, for some m i ∈ N.
j=1 j=1
Proof For the sequence {k j } ⊂ N let K be the closure in V of the set of convex
combinations of elements of {yk j : j ∈ N}. Since yk j y0 and yk j ∈ K , which is
closed and convex, y0 ∈ K follows from Theorem 2.45.
Definition 2.47 (Weak* Topology on V ∗ [11, Chap. IV, Example 1.8]) For a normed
linear space V a set U ⊂ V ∗ is open in the weak* topology if and only if for every
u ∗ ∈ U there exists v1 , · · · , vn ∈ V and > 0 such that
n

v ∗ ∈ V ∗ : |(v ∗ − u ∗ )(v j )| < ⊂ U.
j=1
The collection

V ,v,u ∗ = v ∗ ∈ V ∗ : |(v ∗ − u ∗ )(v)| < , > 0, v ∈ V, u ∗ ∈ V ∗
is said to form a sub-base for the weak* topology on V ∗ .
Theorem 2.48 (Alaoglu [11, Thm. V.3.1]) For a normed linear space V , B ∗ =
{v ∗ ∈ V ∗ : v ∗ V ∗ ≤ 1}, the closed unit ball in V ∗ , is weak* compact.
Definition 2.49 (Weak* Convergence in V ∗ ) For a normed linear space V a sequence

∗
{vk∗ } ⊂ V ∗ converges weak* to v0∗ ∈ V ∗ , written as vk∗ v0∗ , if vk∗ (v) → v0∗ (v) for
all v ∈ V .
Note that when V is a Banach space, weak* convergent sequences in V ∗ are bounded,
but not necessarily if V is not a Banach space.
Definition 2.50 (Weak* Sequential Compactness in V ∗ ) When V is a normed linear

space K ⊂ V ∗ is weak* sequentially compact if every sequence {vk∗ } ⊂ K has a
subsequence which converges weak* to a point v0∗ ∈ K .
Remark 2.51 If V is separable (has a denumerable dense subset) the closed unit
ball in V ∗ with the weak* topology is metrizable [11, Thm. V.5.1], and hence [23,
Chap. 10, Thm. 12] weak* sequentially compact.
Definition 2.52 (Extreme Points) If C is a convex subset of a linear space, c ∈ C is

an extreme point of C if c = αc1 + (1 − α)c2 for c1 , c2 ∈ C and α ∈ (0, 1) implies
that c = c1 = c2 .
2.9 Functional Analysis 23
For example, the four vertices of a closed square in the plane are its extreme points,
and there are no others. Every point of the unit circle is an extreme point of the closed
unit disc in the plane.
Theorem 2.53 (Krein–Milman [11, Thm. V.7.4]) If K is non-empty, compact and
convex in a locally convex space, K is the closed convex hull of its extreme points.
In other words, K is the closure of the intersection of all convex sets which contain
the extreme points of K . (Equivalently, K is the intersection of all closed convex sets
which contain the extreme points of K .)
The next result is a consequence of Theorem 2.48 (Alaoglu) and Theorem 2.53
(Krein–Milman), since V ∗ with the weak* topology is a locally convex topological
space when V is a normed linear space.
Corollary 2.54 When V is a normed linear space the closed unit ball B ∗ in V ∗ is
the weak* closed convex hull of its extreme points (i.e. B ∗ is the intersection of all
weak* closed convex set which contains the extreme points of B ∗ ). In particular, the
set of extreme points of B ∗ is non-empty.
The final theorem of this section says that when V is a Banach space the extreme
points of B ∗ satisfy property (W) in the Introduction.
Theorem 2.55 (Rainwater1 [26]) In a Banach space, xk x if and only if {xk } is
bounded and x ∗ (xk ) → x ∗ (x) for all extreme points x ∗ in the closed unit ball of the
dual space.
2.10 Point Set Topology
Definition 2.56 (Base and Sub-base for a Topology) In a topological space (X, )
a collection G of open sets such that every non-empty open set is a union of sets in
G is called a base for the topology . A sub-base for is a collection G of open sets
with the property that the finite intersections of elements of G form a base for .
Remark
2.57 Note that for any collection T ⊂ ℘ (X ) with the property that X =
G∈T G, there is a unique topology τ for which T is a sub-base.
Definition 2.58 (Separable) A topological space is separable if it has a denumerable
dense subset.
Definition 2.59 (First Axiom of Countability) A topological space (X, ) satisfies
the first axiom of countability if for every x ∈ X there is a denumerable collection
Gx of open sets G with x ∈ G such that x ∈ U ∈ implies that x ∈ G ⊂ U for some
G ∈ Gx . The collection Gx is called a local base at x for the topology .
1 Many internationally renowned mathematicians have published under the name John Rainwater,
see https://en.wikipedia.org/wiki/John_Rainwater.
Definition 2.60 (Complete Separability or Second Axiom of Countability) A space

is said to be completely separable (or to satisfy the second axiom of countability) if
its topology has a denumerable base.
A completely separable space satisfies the first axiom of countability and is separable.
In metric spaces separability and completely separability are equivalent.
If, for any pair of disjoint closed sets Fi , i = 1, 2, in a topological space there
exists a pair of disjoint open sets G i with Fi ⊂ G i , the space is normal; if this
holds for any closed F1 when F2 is a singleton the space is regular, and if it holds
when both Fi are singletons the space is Hausdorff. Thus normal implies regular
implies Hausdorff and a compact Hausdorff space is normal. Metric spaces are normal
because, with {i, j} = {1, 2}, Fi ⊂ G i = {x : dist(x, Fi ) < dist(x, F j )}. A famous
characterisation of normality is the following.
Lemma 2.61 (Urysohn [21, Chap. 4, Lemma 4]) A topological space is normal if
and only if for every pair of disjoint closed sets F1 and F2 there is a continuous
function from the space into [0, 1] with f (F1 ) = 0 and f (F2 ) = 1.
This is not to say that f −1 {0} = F1 and f −1 {1} = F2 . To obtain that level of precision,
see Lemma 2.62(b).
A set is a G δ -set if it is the intersection of a denumerable family of open sets.
Closed G δ -sets feature in Corollary 7.11, Theorems 9.4, 10.2, 10.4, and the following
lemma.
Lemma 2.62 (a) For A = ∅ in a normal topological space X there exists a contin-
uous function f : X → [0, 1/2] with f (x) = 0 for x ∈ A and f (x) > 0 otherwise,
if and only if A is a closed G δ -set.
(b) For every disjoint pair, F1 and F2 , of closed G δ -sets in a normal space X there
is a continuous function f : X → [0, 1] such that F1 = f −1 {0} and F2 = f −1 {1}.
Proof (a) For the “only if” part suppose such an f exists. Then A = f −1 {0} is closed.
Also
A = f −1 {0} = f −1 [0, 1/k),
k∈N

and so A is a G δ -set. Conversely, suppose A = ∅ is closed and A = k∈N Uk where
Uk is open. Since X \ Uk and A are disjoint closed sets, by Lemma 2.61 (Urysohn),
there exists a continuous function f k : X → [0, 1] suchthat f k (A) = {0} and f k (X \
Uk ) = {1}. Now define f : X → [0, 1/2] by f (x) = k∈N 2−(k+1) f k (x). Since the
series is uniformly convergent f is continuous, f (x) = 0 on A and f (x) ∈ (0, 1/2]
elsewhere.
(b) By Urysohn’s lemma there exists a continuous h 0 : X → [0, 1] with
h 0 (F1 ) = 0 and h 0 (F2 ) = 1. Since X is a G δ -space, by part (a) there exists con-
tinuous h 1 : [0, 1] → [0, 1/2] with h −1 1 {0} = F1 and h 2 : [0, 1] → [1/2, 1] with
h −1
2 {1} = F2 . Now f : X → [0, 1] defined by
2.10 Point Set Topology 25

f (x) = inf h 2 (x), sup h 0 (x), h 1 (x)
X X
is continuous and has the required property.

A topological space is a G δ -space if all its closed
sets are G δ -sets. Metric spaces are
G δ -spaces because, for every closed set, F = k G k where G k = {x : dist {x, F} <
1/k}, k ∈ N. However, a G δ -space need not be normal and a normal space need not
be a G δ -space. In this context, the following characterisation of normality is useful.
Theorem 2.63 A topological space is normal if and only if for all
closed F and open
G with F ⊂ G there exist open sets Wn , n ∈ N, such that F ⊂ n Wn and Wn ⊂ G
for all n ∈ N.
Proof By definition of normality, for given F (closed) ⊂ G (open) in a normal space
there is an open set, W1 say, with F ⊂ W1 ⊂ W1 ⊂ G.
Conversely, let A and B be disjoint
closed sets inX . Then, by hypothesis, there
exist open sets Wn such that A ⊂ n W n and
W n B = ∅. Similarly, there exist
open sets Vn such that B ⊂ n Vn and Vn A = ∅.
Now define open sets G n , Hm , W and V by

G n = Wn \ Vi , Hm = Vm \ Wj, W = Gn, V = Hm .
i≤n j≤m n m
⊂ W and B ⊂ V , and W and V are open.

Then A To see they are disjoint suppose
x ∈ W V . Then for some m ∗ and n ∗ , x ∈ G n ∗ Hm ∗ . But then

x ∈ Wn ∗ \ Vi ) Vm ∗ \ W j ) = ∅.
i≤n ∗ j≤m ∗
Thus V and W are disjoint, and hence the space is normal.

Lemma 2.64 A completely separable, regular topological space X is a normal G δ -
space.
Proof Let F (closed) ⊂ G (open) ⊂ X . By regularity, for every x ∈ F there exist
disjoint open sets G x and G x with x ∈ G x , X \ G ⊂ G x and G x G x = ∅. There-
fore
x ∈ G x ⊂ X \ G x ⊂ G which implies that x ∈ G x ⊂ G x ⊂ G for all x ∈ F.
Since there is no loss in assuming that G x ∈ G, the denumerable base for the topology,
the normality of X follows from Theorem 2.63.
To see that X is a G δ -space, let F be closed. Then for every x ∈ / F there exists

open
sets G x and G x with F ⊂ G x , x ∈ G x , G x G x = ∅ and G x ∈ G. Since F =

/ G x and at most countably many G x are needed, F is a G δ -set and so X is a
x ∈F
normal G δ -space.
Lemma 2.65 A locally compact Hausdorff space is regular. A completely separable,

locally compact Hausdorff space is a normal G δ -space.
Proof For F closed and x ∈ / F,let x ∈ G where G is open and G is compact.

Since X is Hausdorff and x∈/ F G which is compact, there are disjoint open sets
V, W with x ∈ V and F G ⊂W . Since G is compact in a Hausdorff space it
is closed,
and so V G and W (X \ G) areopen.Therefore, X is regular since
x ∈ V G, F ⊂ W (X \ G) and (V G) (W (X \ G)) = ∅. Lemma 2.64
now implies that a completely separable, locally compact Hausdorff space is a normal
G δ -space.
Chapter 3
L ∞ and Its Dual
Henceforth, (X, L, λ) is a complete σ-finite measure space and L ∞ (X, L, λ) is

defined by (2.8); similarly for (X, B, λ) and L ∞ (X, B, λ). The general terminol-
ogy will be that of Chap. 2.
In notation for finitely additive measures that will be considered in detail in
Chap. 4, the representation theorem for the dual space L ∞ (X, L, λ)∗ , which is anal-
ogous to the Riesz Theorem 2.37 for continuous functions, is the following special
case of [35, Thm. 2.3], see also [12, Thm. IV.8.16] and the references therein. This
result will be referred to as the Yosida–Hewitt representation of L ∗∞ (X, L, λ).
Theorem 3.1 For every bounded linear functional on L ∞ (X, L, λ) there exists a
finitely additive measure (Definition 4.1) ν on L such that
ν(N ) = 0 for all N ∈ N , (3.1a)

f (u) = u dν for all u ∈ L ∞ (X, L, λ), (3.1b)
X
|ν|(X ) = f ∞ < ∞. (3.1c)
Conversely if ν is a finitely additive measure on X with ν(N ) = 0 for all N ∈ N ,

then f in (3.1b) is a well-defined bounded linear functional on L ∞ (X, L, λ). The
set of finitely additive measure on X with ν(N ) = 0 for N ∈ N will be denoted by
L ∗∞ (X, L, λ). (Note that |ν|(N ) = 0 if N ∈ N and ν ∈ L ∗∞ (X, L, λ).)
Once the integral with respect to finitely additive measures in Theorem 3.1 has been
defined, the proof in Sect. 6.2 of the theorem is straightforward (indeed, an interested
reader might like to have a quick look now). However, there are differences from
standard theory, see Remark 2.26, and the development needs to be handled carefully.
In the next few chapters, notation and terminology for finitely additive measures,

https://doi.org/10.1007/978-3-030-34732-1_3
28 3 L ∞ and Its Dual
their existence and basic properties on σ-algebras will be established. However, the
following implications of Theorem 3.1 are more or less immediate.

Corollary 3.2 If {Ak } ⊂ L is a sequence in L with λ Ai A j = 0, i = j, then
χ Ak 0 as k → ∞ in L ∞ (X, L, λ), where χ Ak denotes the characteristic function
of Ak .
Proof By Theorem 3.1, for any f ∈ L ∞ (X, L, λ)∗ there is a finitely additive measure
ν with
f (χ Ak ) = χ Ak dν = ν(Ak ) = ν + (Ak ) − ν − (Ak ),
X
where ν ± (the positive and negative parts of ν, see (4.1c)) are non-negative finitely
additive measures. Since, by finite additivity and (6.2d),

K
K
0 ν ± (Ak ) = ν ± Ak ν ± (X ) < ∞, for all K ∈ N,
k=1 k=1
ν ± (Ak ) → 0 and the result follows.
When combined with the Hahn–Banach theorem, Theorem 3.1 yields a wide variety
of finitely additive measures.
Corollary 3.3 If (X, ) is a Hausdorff topological space and is not the discrete
topology,
there exists a finitely additive measure ν = 0 which is zero on N and

X v dν = 0 for all v ∈ L ∞ (X, B, λ) which are continuous on X .
Proof From the hypothesis and Remark 2.34, elements of L ∞ (X, B, λ) which are
continuous on X form a proper closed linear subspace of L ∞ (X, B, λ). The result
is then immediate from Theorem 3.1 and Corollary 2.40 of the Hahn–Banach
theorem.
The existence of finitely additive measures with particular properties will be consid-
ered again in Chap. 5.
Corollary 3.4 Suppose B is the Borel σ-algebra of a locally compact Hausdorff

space (X, ), and (X, B, λ) is a corresponding measure space. Then for every ν ∈
L ∗∞ (X, B, λ) there exists a unique, regular, real Borel measure ν̃ such that

v dν = v d ν̃ for all v ∈ C0 (X, ), |ν̃|(X ) |ν|(X ), (3.2)
X X
and ν̃ 0 on Borel sets when ν 0 on L.
Proof For ν ∈ L ∗∞ (X, B, λ), define f ∈ L ∞ (X, B, λ)∗ by (3.1b). The restriction
f˜ of f to C0 (X, ) has f˜C0 (X,)∗ f L ∞ (X,B,λ)∗ = |ν|(X ) < ∞, and the result
follows from Theorem 2.37 (Riesz).
3 L ∞ and Its Dual 29
It follows from Corollary 3.3 that the inequality in (3.2) may be strict and, from the
Valadier–Hensgen example in Sect. 6.6, that for a given regular real Borel measure
ν̃ there may be uncountably many finitely additive measures ν which satisfy (3.2).
Chapter 4
Finitely additive measures are naturally defined on algebras (collections of sets which
are closed under complementation and finite unions), but here they are considered
on σ-algebras (closed under complementation and countable unions) because L in
Theorem 3.1 is a σ-algebra. Although the terminology “finitely additive measures”
may suggest only a slight modification of familiar measure theory, intuition based
on experience with σ-additivity can seriously mislead, see (†) in the Preface which
is justified in Remark 6.1.
4.1 Definition, Notation and Basic Properties
Definition 4.1 (Finitely Additive Measures [35, Sects. 1.2–1.7]) A finitely additive
measure ν on L is a mapping from L into R with
ν(∅) = 0 and sup |ν(A)| < ∞;

A∈L

ν(A B) = ν(A) + ν(B) for all A, B ∈ L with A B = ∅.
A finitely additive measure ν is σ-additive if and only if

ν Ek = ν(E k ) for all {E k } ⊂ L with E j E k = ∅, j = k.
k∈N k∈N
In the literature, it is usual for ba(L) (bounded and additive on L) to denote the
set of finitely additive measures on L. Let Σ(L) ⊂ ba(L) denote the σ-additive
measures.

https://doi.org/10.1007/978-3-030-34732-1_4
32 4 Finitely Additive Measures
Obviously, the set Σ(L) coincides with the set of real measures in Sect. 2.4. However
λ ∈ ba(L) if and only if λ(X ) < ∞. Hence λ ∈ / Σ(L) when λ(X ) = ∞ although it
is σ-additive on L. However, because by hypothesis λ is σ-finite on L, there exists
[28, Lemma
6.9] an integrable function f which is positive everywhere on X . Hence
γ(E) := E f dλ defines γ ∈ Σ(L) which is positive on L \ N and can be used as
a surrogate for λ in arguments involving finitely additive measures, for example, in
the proof of Theorem 10.4.
There follows a review of standard theory sufficient for our purposes. For a com-
prehensive account see [35], [12, Chap. III] or [6, Chap. 4].
Lemma 4.2 Suppose ν ∈ ba(L).
(a) Then ν ∈ Σ(L) if and only if ν(Ak ) → 0 as k → ∞ for any nested sequence
Ak+1 ⊂ Ak , Ak ∈ L with k Ak = ∅.
(b) If 0 ν γ for some γ ∈ Σ(L), then ν ∈ Σ(L).

Proof (a) Suppose ν ∈ Σ(L) and, for {Ak } ⊂ L nested with k Ak =∅, let E k =
Ak \ Ak+1 ∈ L. Then {E k } is a sequence of mutually disjoint sets with k E k = A1 .
Since ν(A1 ) is finite and ν is σ-additive
∞
∞

ν(A1 ) = ν(E k ) = ν(Ak \ Ak+1 ) = lim ν(A1 ) − ν(Ak ) ,
k→∞
k=1 k=1
whence limk→∞ ν(Ak ) = 0.

converse suppose ν ∈ ba(L) and limk→∞ ν(Ak ) = 0 for all nested {Ak } ⊂
For the
L with k Ak = ∅. Then for any sequence of mutually disjoint
sets {E j } in L let
Ak = ∞ k E j . By hypothesis lim k→∞ ν(Ak ) = 0 because k Ak = ∅ and by finite
additivity,

k−1
∞
∞
ν(E j ) = ν E j − ν(Ak ) → ν E j as k → ∞.
j=1 j=1 j=1
Hence ν ∈ Σ(L).
Part (b) is immediate because its hypothesis implies that ν satisfies the criterion
for σ-additivity in (a).
The next result, which is a simple case of Nikodym’s convergence theorem [8, Thm.
4.6.3(i)], depends on the following trivial observation. Suppose
a jk ∈ R, j, k ∈ N, 0 a jk a j k M when j j , k k .
Then
lim lim a jk = sup sup a jk = sup sup a jk = lim lim a jk .

k→∞ j→∞ k j j k j→∞ k→∞
4.1 Definition, Notation and Basic Properties 33
Lemma 4.3 Suppose {γk } ⊂ Σ(L) and
0 γk (E) γk+1 (E) M < ∞ for all E ∈ L, k ∈ N.
Then γ ∈ Σ(L) where γ(E) = limk→∞ γk (E), E ∈ L.

Proof It is obvious that γ is finitely additive. To show j γ ∈ Σ(L), let {E } ⊂ L be
a sequence of mutually disjoint sets and put a jk = =1 γk (E ) so that 0 a jk
a j k M if j j , k k . Now since γk ∈ Σ(L),

∞
∞
j
γ E = lim γk E = lim lim γk (E )
k→∞ k→∞ j→∞
1 1 =1
j

j
= sup sup γk (E ) = sup sup γk (E )
k j =1 j k =1

j

j ∞

= lim lim γk (E ) = lim γ(E ) = γ(E ),
j→∞ k→∞ j→∞
=1 =1 =1
which shows γ ∈ Σ(L).

Finitely additive measures form a real linear space with, for E ∈ L,
(α1 ν1 + α2 ν2 )(E) = α1 ν1 (E) + α2 ν2 (E), νi ∈ ba(L), αi ∈ R, i = 1, 2,
and a partial ordering ν1 ν2 is defined by
ν1 (E) ν2 (E) for all E ∈ L.
For ν1 , ν2 ∈ ba(L) and E ∈ L let

(ν1 ∨ ν2 )(E) = sup ν1 (F) + ν2 (E \ F) ,
{F∈L:F⊂E}
(4.1a)
(ν1 ∧ ν2 )(E) = inf ν1 (F) + ν2 (E \ F) ,
{F∈L:F⊂E}
from which it is obvious that

ν1 ∧ ν2 = − (−ν1 ) ∨ (−ν2 ) and ν1 ∨ ν2 = − (−ν1 ) ∧ (−ν2 ) . (4.1b)
Theorem 4.4 (a)
ν1 , ν2 ∈ ba(L) ⇒ ν1 ∧ ν2 , ν1 ∨ ν2 ∈ ba(L),
ν1 , ν2 ∈ Σ(L) ⇒ ν1 ∧ ν2 , ν1 ∨ ν2 ∈ Σ(L).
(b) For ν ∈ ba(L), let ν + = ν ∨ 0, ν − = (−ν) ∨ 0. Then
ν ± 0, ν = ν + − ν − and ν + ∧ ν − = 0. (4.1c)
(c) If ν ∈ ba(L) and γ1 , γ2 ∈ Σ(L) with γ1 ν γ2 , then ν ∈ Σ(L).

(d) If ν ∈ ba(L) and ν(E) = 0 for all E ∈ N , then ν ± (E) = 0 for all E ∈ N .
Remark 4.5 When ν ∈ Σ(L) (equivalently when ν is a real measure) (4.1c) is

equivalent to (2.4) in Sect. 2.4 because of the Hahn Decomposition Theorem 2.17.
For ν ∈ ba(L), ν ± are referred to as the positive and negative parts of ν, and
|ν| := ν + + ν − as the total variation of ν (see Theorem 3.1).

(a) Let E 1 , E 2 ∈ L where E 1 E 2 = ∅ and for F ⊂ E 1 E 2 let Fi =
Proof
E i F, i = 1, 2. Then by finite additivity,

(ν1 ∨ ν2 )(E 1 E2 ) = sup
ν1 (F) + ν2 (E 1 E2 ) \ F
{F∈L:F⊂E 1 E2 }

= sup
ν1 (F1 ) + ν1 (F2 ) + ν2 (E 1 \ F1 ) + ν2 (E 2 \ F2 )
{F∈L:F⊂E 1 E2 }

= sup ν1 (F1 ) + ν2 (E 1 \ F1 )
{F1 ∈L:F1 ⊂E 1 }

+ sup ν1 (F2 ) + ν2 (E 2 \ F2 ) = (ν1 ∨ ν2 )(E 1 ) + (ν1 ∨ ν2 )(E 2 ).
{F2 ∈L:F2 ⊂E 2 }

So ν1 ∨ ν2 ∈ ba(L), and hence ν1 ∧ ν2 = − (−ν1 ) ∨ (−ν2 ) ∈ ba(L) by (4.1b).
Now suppose ν1 , ν2 ∈ Σ(L) and {E k } ⊂ L is a sequence of mutually disjoint sets.
Then, for ∈ (0, 1) and k ∈ N, there exists Fk ⊂ E k , Fk ∈ L, such that
ν1 (Fk ) + ν2 (E k \ Fk ) (ν1 ∧ ν2 )(E k ) + k .

Then, by countable additivity, with F := k Fk ⊂ k E k =: E,
∞

(ν1 ∧ ν2 )(E) ν1 (F) + ν2 (E \ F) = ν1 (Fk ) + ν2 (E k \ Fk )
k=1
∞
∞

(ν1 ∧ ν2 )(E k ) + k = (ν1 ∧ ν2 )(E k ) + ,
k=1 k=1
1−
for all ∈ (0, 1). Hence

∞ ∞

(ν1 ∧ ν2 ) E k = (ν1 ∧ ν2 )(E) (ν1 ∧ ν2 )(E k ).
k=1 k=1
4.1 Definition, Notation and Basic Properties 35

For the opposite inequality note that, with F ∈ L and F ⊂ E = k Ek ,
∞
∞
ν1 (F) + ν2 (E \ F) = ν1 (F E k ) + ν2 (E k \ F) (ν1 ∧ ν2 )(E k ),
k=1 k=1
whence

∞ ∞

(ν1 ∧ ν2 ) E k = (ν1 ∧ ν2 )(E) (ν1 ∧ ν2 )(E k ).
k=1 k=1
Hence ν1 ∧ ν2 ∈ Σ(L), and it follows that ν1 ∨ ν2 = −((−ν1 ) ∧ (−ν2 )) ∈ Σ(L).

(b) By definition ν ± 0 and
inf ν(F) = −ν − (E), sup ν(F) = ν + (E). (4.2)

{F∈L:F⊂E} {F∈L:F⊂E}
Since ν(E) = ν(F) + ν(E \ F) for E, F ∈ L with F ⊂ E, it follows that
ν(F) − ν − (E) ν(E) ν(F) + ν + (E), (4.3)
and ν = ν + − ν − follows by taking the supremum on the left and the infimum on
the right of (4.3), over F ∈ L with F ⊂ E, using (4.2).
Finally suppose ν + ∧ ν − = 0. Then, since ν ± 0 are finitely additive by part (a)
there exists E ∈ L and α > 0 such that for any F ⊂ E, F ∈ L,
0 < α ν + (E \ F) + ν − (F) = ν + (E) − ν + (F) + ν − (F) = ν + (E) − ν(F).
Hence for all F ⊂ E, F ∈ L, ν(F) ν + (E) − α. Since α > 0 this contradicts (4.2)
and it follows that ν + ∧ ν − = 0.
(c) Since 0 ν − γ1 γ2 − γ1 and γ2 − γ1 ∈ Σ(L), Lemma 4.2 implies ν −
γ1 ∈ Σ(L), and hence ν ∈ Σ(L) as required.
(d) The result follows from the hypothesis and (4.2) since F ∈ N when F ∈ L
and F ⊂ E ∈ N , because (X, L, λ) is complete.
Definition 4.6 (Absolute Continuity and Singularity [6, Chap. 6]) For ν1 , ν2 ∈
ba(L) write ν1 ν2 (ν1 is absolutely continuous with respect to ν2 ), if for all
> 0 there exists δ such |ν1 (E)| < when |ν2 |(E) < δ, and write ν1 ⊥ ν2 (ν1
and ν2 are mutually singular) if for every > 0 there exists E ∈ L such that
|ν1 |(E) + |ν2 |(X \ E) < .
Remark 4.7 In the special case when ν1 , ν2 ∈ Σ(L) the above definitions are equiv-
alent to simpler statements [6, Thms. 6.1.6 & 6.1.17]:
ν1 ν2 if and only if |ν2 |(E) = 0 implies ν1 (E) = 0;

ν1 ⊥ ν2 if and only if |ν1 |(E) + |ν2 |(X \ E) = 0 for some E ∈ L.
However, it is important to remember that a finitely additive measure ν which vanishes

on N need not satisfy ν λ in Definition 4.6 if ν ∈ / Σ(L).
4.2 Purely Finitely Additive Measures
Definition 4.8 (Purely Finitely Additive Measures) μ ∈ ba(L) is purely finitely addi-
tive if

γ ∈ Σ(L) : 0 γ μ+ = 0 = γ ∈ Σ(L) : 0 γ μ− .

+ −
Clearly, μ is purelyfinitely additive if and only if μ and μ are purely finitely
additive and Π (L) Σ(L) = {0}, where Π (L) denotes the set of purely finitely
additive measures.
Remark 4.9 When μ1 , μ2 ∈ Π (L) and μ1 ν μ2 for some ν ∈ ba(L), it is imme-
diate that ν ∈ Π (L) since ν + μ+ − −
2 , ν μ1 . Similarly, |ν| μ ∈ Π (L) implies
ν ∈ Π (L).
Theorem 4.10 (a) ν ∈ ba(L) is purely finitely additive if and only if
ν ± ∧ γ = 0 for all 0 γ ∈ Σ(L).
(b) For μ, μ1 , μ2 ∈ Π (L) and α ∈ R,
αμ, μ1 + μ2 , |μ|, μ1 ∧ μ2 , μ1 ∨ μ2 ∈ Π (L).
Proof (a) Suppose ν ∈ Π (L). Then 0 ν ± ∧ γ γ for any 0 γ ∈ Σ(L), which

implies, by Theorems 4.4(a) and 4.2(b), that ν ± ∧ γ ∈ Σ(L) and hence by Defini-
tion 4.8 that ν ± ∧ γ = 0. Conversely, for ν ∈ ba(L) suppose that ν ± ∧ γ = 0 for all
0 γ ∈ Σ(L). Then 0 γ ν ± , γ ∈ Σ(L), implies that γ = ν ± ∧ γ = 0. Hence
ν ∈ Π (L) by Definition 4.8.
(b) That αμ ∈ Π (L), α ∈ R, is immediate from Definition 4.8 since (−μ)± =
∓
μ . To show that μ1 + μ2 ∈ Π (L) consider first the case when μ1 and μ2 are non-
negative. By part (a) it suffices to show that (μ1 + μ2 ) ∧ γ = 0 for all 0 γ ∈ Σ(L).
In other words, it suffices to show, for all E ∈ L and all 0 γ ∈ Σ(L), that
inf {μ1 (F) + μ2 (F) + γ(E \ F)} = 0.

{F∈L:F⊂E}
Since μi ∧ γ = 0, for any > 0 there exist E ⊃ Fi ∈ L, with

4.2 Purely Finitely Additive Measures 37
μi (Fi ) + γ(E \ Fi ) , i = 1, 2.
Therefore since μi 0,

μ1 (F1 F2 ) + μ2 (F1 F2 ) + γ(E \ (F1 F2 ))
μ1 (F1 ) + μ2 (F2 ) + γ(E \ F1 ) + γ(E \ F2 ) 2. (4.4)
This proves μ1 + μ2 ∈ Π (L) for non-negative μi ∈ Π (L) and hence |μ| = μ+ +

μ− ∈ Π (L) for any μ ∈ Π (L).
More generally,
|μ1 + μ2 | μ+ + − −
1 + μ2 + μ1 + μ2 ∈ Π (L) when μi ∈ Π (L), i = 1, 2,
and μ1 + μ2 ∈ Π (L) follows by Remark 4.9.

Similarly
|μ1 ∧ μ2 | + |μ1 ∨ μ2 | 2(|μ1 | + |μ2 |), μ1 , μ2 ∈ Π (L),
implies that μ1 ∧ μ2 and μ1 ∨ μ2 are in Π (L).
By (4.1a) and Theorem 4.10(a), a non-negative ν ∈ ba(L) is purely finitely additive

if and only if for every 0 γ ∈ Σ(L), E ∈ L and > 0 there exists
F ∈ L, F ⊂ E with ν(E \ F) + γ(F) < . (4.5)
Since L is a σ-algebra this remark can be significantly refined.
Lemma 4.11 For 0 γ ∈ Σ(L), 0 μ ∈ Π (L) and > 0 there exists F ∈ L with
γ(F) and μ(F) = μ(X ).
Proof If μ(X ) = 0 the result holds with F = ∅.

So suppose μ(X ) > 0 and for given > 0 let ek = 2−k . Then with E 0 = X ,
by (4.5) there exists F1 ∈ L with μ(E 0 \ F1 ) + γ(F1 ) < e1 . With E 1 = E 0 \ F1 , in
L there exists F2 ⊂ E 1 , F2 ∈ L, with μ(E 1 \ F2 ) + γ(F2 ) < e2 . Then with E 2 =
E 1 \ F2 , there exists F3 ⊂ E 2 , F3 ∈ L, with μ(E 2 \ F3 ) + γ(F3 ) < e3 . Proceeding
by induction, for k ∈ N0 ,
μ(E k \ Fk+1 ) + γ(Fk+1 ) < ek+1 , E k ⊃ Fk+1 ∈ L, E k+1 = E k \ Fk+1 .
It follows that for K ∈ N,

K
K K
0 μ E0 \ Fk e K and γ Fk ek < .
k=1 k=1 k=1

Now since L is a σ-algebra, F = ∞ k=1 Fk ∈ L with γ(F) . Moreover, μ(X ) =
μ(F) since μ(X \ F) K for all K and μ is finitely additive. This completes the
proof.
The sense in which a purely finitely additive measure on a σ-algebra is singular with
respect to any σ-additive measure is captured by the following.
Theorem 4.12 Let 0 ν ∈ ba(L). Then ν ∈ Π (L) if and only if for every non-
negative γ ∈ Σ(L) there exists a sequence {E k } ⊂ L with
E k+1 ⊂ E k , ν(E k ) = ν(X ) for all k and γ(E k ) → 0 as k → ∞.
Proof Suppose 0 ν ∈ Π (L) and 0 γ ∈ Σ(L). Then by the preceding lemma,

k n ∈ N there exists Fn ∈ L with ν(Fn ) = ν(X ) and 0 γ(Fn ) < 1/n.
for Let
k
Ek =
n=1 F n . Then ν(E k ) = ν(X ) because, by finite additivity, ν(X \ E k ) n=1 ν(X \
Fn ) = 0, and 0 γ(E k ) ν(Fk ) 1/k. Conversely, if for every non-negative
γ ∈ Σ(L) such a sequence {E k } exists, then γ ∧ ν = 0. Hence ν ∈ Π (L) by Theo-
rem 4.10.
4.3 Canonical Decomposition: ba(L) = Σ(L) ⊕ Π(L)
Theorem 4.13 Any ν ∈ ba(L) can be written uniquely as ν = μν + γν where μν ∈

Π (L), γν ∈ Σ(L). Moreover μν , γν 0 if ν 0.
Proof To show that such a decomposition, if it exists, is unique let ν = μi + γi ,

μi ∈ Π(L), γi ∈ Σ(L), i = 1, 2. Then, by Theorem 4.10, μ1 − μ2 = γ2 − γ1 ∈
Π (L) Σ(L) and uniqueness follows since Π (L) Σ(L) = {0} as noted after
Definition 4.8.
Since, by (4.1c), ν = ν + − ν − and, by Theorem 4.10, Σ(L) and Π (L) are linear
spaces, it suffices to show the decomposition exists for non-negative ν ∈ ba(L). Let

ς := sup γ(X ) : 0 γ ν, γ ∈ Σ(L)
and let γk ∈ Σ(L) be such that 0 γk ν and limk→∞ γk (X ) = ς.

Replacing γk by ∨kj=1 γ j , by Theorem 4.4(a) there is no loss in assuming that γk
γk+1 for all k ∈ N. Consequently γν (E) := limk→∞ γk (E) ν(E) ν(X ) exists
for all E ∈ L, and γν ∈ Σ(L) by Lemma 4.3. Clearly 0 μν := ν − γν ∈ ba(L)
and γν (X ) = ς.
To show that μν ∈ Π (L), suppose 0 γ μν for some γ ∈ Σ(L). Then 0 γ +
γν ν which implies that γ(X ) + γν (X ) ς = γν (X ). Hence γ(X ) = 0, which, by
Definition 4.8, implies 0 μν ∈ Π (L). Obviously from the construction, γν and μν
are non-negative when ν 0. This completes the proof.
4.4 L ∗∞ (X, L, λ) 39
4.4 L ∗∞ (X, L, λ)
This chapter closes by specialising briefly to the set L ∗∞ (X, L, λ) of finitely additive
measures that feature in Theorem 3.1 (Yosida–Hewitt).
Definition 4.14 (L ∗∞ (X, L, λ)) ν ∈ L ∗∞ (X, L, λ) if ν ∈ ba(L) and ν(E) = 0 for

all E ∈ N .
Theorem 4.15 Suppose ν ∈ L ∗∞ (X, L, λ). Then ν ∈ Π (L) if and only if ν + ∧ γ =

ν ∧ γ = 0 for all non-negative γ ∈ Σ(L) L ∗∞ (X, L, λ).
−

Proof Suppose ν ∈ L ∗∞ (X, L, λ) and ν ± ∧ γ = 0 for all non-negative γ ∈ Σ(L)
L ∗∞ (X, L, λ). Let γ̂ ∈ Σ(L) be arbitrary with
0 γ̂ ν ± . Since, by Theorem 4.4(d),
± ∗
ν = 0 on N it follows that γ̂ ∈ Σ(L) L ∞ (X, L, λ). Hence, by hypothesis,
ν ± ∧ γ̂ = 0 and, by Definition 4.8, ν ∈ Π (L).
Conversely, if ν ∈ Π (L), by Definition 4.8 ν ± ∧ γ = 0 for all 0 γ ∈ Σ(L)
L ∗∞ (X, L, λ).
Theorem 4.16 Any ν ∈ L ∗∞ (X, L, λ) can be written uniquely as

ν = μν + γν ∈ L ∗∞ (X, L, λ) Π (L) ⊕ L ∗∞ (X, L, λ) Σ(L) . (4.6)
Proof If 0 ν ∈ L ∗∞ (X, L, λ) in Theorem 4.13, 0 γν ν implies that

γν (E) = 0, E ∈ N . Hence γν ∈ L ∗∞ (X, L, λ) and since μν = ν − γν , so is μν .
The result for general ν ∈ Π (L) follows since ν ∈ L ∗∞ (X, L, λ) implies ν ± ∈
L ∗∞ (X, L, λ), see Theorem 4.4(d).
Chapter 5
G: 0–1 Finitely Additive Measures
In the notation of Definition 4.14, this chapter concerns the set

G = ω ∈ L ∗∞ (X, L, λ) : ω(X ) = 1, ω(E) ∈ {0, 1}, E ∈ L (5.1)
of finitely additive measures, each element of which would, by Theorem 3.1, corre-
spond to an element f ∈ L ∞ (X, L, λ)∗ with the property that
f (χ X ) = 1, f (χ E ) ∈ {0, 1}, E ∈ L; f (χ E ) = 0, E ∈ N .
Much of the importance of G in the sequel can be traced back to the next theorem and
subsequent remark which suggest that the action of elements of G on L ∞ (X, L, λ)
is reminiscent of Dirac measures acting on continuous functions.

Theorem
5.1 For u ∈ L ∞ (X, L, λ) and ω ∈ G there is a unique α ∈ I := [u]− ∞,
[u]+∞ (see (2.8)) such that

ω x ∈ X : |u(x) − α| < = 1 for all > 0. (5.2)
This means (5.2) holds for a unique α ∈ R, independent of the function chosen to
represent the equivalence class u ∈ L ∞ (X, L, λ).
Proof For i = 1, 2 suppose u i is essentially bounded, αi ∈ R and u 1 (x) = u 2 (x) for

λ-almost all x ∈ X . If (5.2) is satisfied by both (u i , αi ), i = 1, 2, and if α1 − α2 =:
20 > 0, then

x ∈ X : |u 1 (x) − α1 | < 0 x ∈ X : |u 2 (x) − α2 | < 0 ∈ N .
Hence by finite additivity the ω-measure of their union would be 2, contradicting

ω ∈ G. Thus (5.2) cannot be satisfied by both (u 1 , α1 ) and (u 2 , α2 ) simultaneously

https://doi.org/10.1007/978-3-030-34732-1_5
42 5 G: 0–1 Finitely Additive Measures
if α1 = α2 when u 1 = u 2 λ-almost everywhere. On the other hand if α1 = α2 and

u 1 = u 2 λ-almost everywhere, either they both satisfy (5.2) or neither satisfies (5.2).
This shows that the value of α (if any) for which (5.2) holds is independent of the
function used to represent the equivalence class u ∈ L ∞ (X, L, λ) and, for given
u ∈ L ∞ (X, L, λ), (5.2) holds for at most one α. By a similar argument α ∈ I if
(5.2) holds.
Now suppose for given u there is no
such α. Then for all α ∈ I there exists
K
α > 0 with ω {x ∈ X : |u(x) − α| < α } = 0. By compactness, I ⊂ k=1 (αk −
αk , αk + αk ) and consequently

K
ω(X ) = ω x : u(x) ∈ (αk − αk , αk + αk )
k=1

K
ω x : u(x) ∈ (αk − αk , αk + αk ) = 0.
k=1
Since ω(X ) = 1, (5.2) holds for a unique α.
Remark 5.2 If ω ∈ G but ω ∈ / Σ(L) (which often happens, see Theorem 5.9 below)
it does not follow from (5.2) that ω ({x ∈ X : u(x) = α}) = 1. Nevertheless (5.2)
is what is meant by saying that in the sense of finitely additive measures u ∈
L ∞ (X, L, λ) is a constant ω-almost everywhere.
The important fact that elements of G proliferate is a consequence of Zorn’s

lemma in the light of a well-known one-to-one correspondence between elements of
G and families of sets in L \ N called ultrafilters in (X, L, λ). This correspondence
leads to Theorem 5.6 on the existence of elements of G with certain properties and
ultimately to (†) in the Preface.
5.1 G and Ultrafilters
Definition 5.3 (Ultrafilter) Given (X, L, λ), F ⊂ L is a filter if

(i) X ∈ F and N F = ∅,

(ii) E 1 , E 2 ∈ F ⇒ E 1 E 2 ∈ F,
(iii) E 2 ⊃ E 1 ∈ F ⇒ E 2 ∈ F.
A filter U which is maximal with respect to set inclusion, i.e. for a filter F
(iv) U ⊂ F ⇒ F = U,
5.1 G and Ultrafilters 43
is called an ultrafilter in (X, L, λ). The set of ultrafilters is denoted by U.
Theorem 5.4 (a) For U ∈ U, an element ωU ∈ G is defined by

1 if E ∈ U
ωU (E) := . (5.3a)
0 otherwise
(b) For ω ∈ G, an element Uω ∈ U is defined by

Uω := E ∈ L : ω(E) = 1 . (5.3b)
Note that ωUω = ω when ω ∈ G and UωU = U when U ∈ U.
Proof (a) To show that ωU , U ∈ U, is well defined on L it suffices to show that exactly
one of E and E
= X \ E is in U for every E ∈ L. To begin note that at most one of
E and E
is in U, for otherwise E, E
∈ U implies ∅ = E E
∈ U, a contradiction
since ∅ ∈ N .
To see that one of E, E
is in U, suppose E ∈/ U. Then suppose that E F ∈ /N
for all F ∈ U and let

F =U G∈L:E F ⊂ G for some F ∈ U .
Clearly F is a filter with U ⊂ F and E ∈ F \ U. Since this contradicts the maximality

of U in (iv) it follows that E F̂ ∈ N for some F̂ ∈ U.

It now follows that E
∈ / N , since E F̂ ∈ N and F̂ ∈ / N by (i). Thus E ∈ /U

implies that E ∈ / N , equivalently N ∈ N implies N ∈ U. In particular, E F̂ ∈ N

implies that E
F̂
∈ U, and hence by (ii) E
F̂ = (E
F̂
) F̂ ∈ U. Finally

E
F̂ ⊂ E
implies that E
∈ U by (iii). This shows that E ∈ / U implies X \ E ∈ U,
and hence U ω is well defined by (5.3a). It remains to check that ω ∈ ba(L).
If A B = ∅, A, B ∈ L, it follows from (i) and (ii) that at most one of ωU (A)
andωU (B) is 1, and if one of them is 1, then ωU (A B) = 1 by (iii). If both are zero,
A
B
∈ U by (ii), and hence ωU (A B) = 0. Hence ωU ∈ ba(L) and it follows
that ωU ∈ G.
(b) Now suppose Uω is given by (5.3b) where ω ∈ G. Then Uω satisfies (i) and
(iii), by (5.1) and finite additivity. Also, for E 1 , E 2 ∈ Uω ,

ω(E 1 \ E 2 ) + ω(E 2 ) = ω(E 1 \ E 2 ) + 1
1 = ω(E 1 E2 ) = .
ω(E 2 \ E 1 ) + ω(E 1 ) = ω(E 2 \ E 1 ) + 1

Hence ω(E 1 \ E 2 ) = 0 = ω(E 2 \ E 1 ) which implies ω(E 1 E 2 ) = 1. So E 1 E 2
∈ Uω and Uω satisfies (ii).
To see that U ω satisfies (iv), suppose it does not and let
F denote thecollection
of filters which contain Uω . If F ⊂ F is totally ordered by inclusion, F ∈F F ∈

F is an upper bound for F. Hence, by Zorn’s lemma, F has a maximal element,
∈ U say. Since, by assumption, Uω is not maximal, there exists F ∈ U

U \ Uω with
ωU (F) = 1 and ω(F) = 0. Hence ω(F
) = 1 which implies that F
∈ Uω and hence
that ωU (F
) = 1 and ωU (F) = 0. This contradiction proves Uω satisfies (iv).
The observation that G is determined entirely by the ultrafilter structure of L \ N is

considered further in Lemma 9.1 and Corollary 9.2. For example, in a locally compact
Hausdorff space X , each ω ∈ G may be identified with a unique point of X just as a
Dirac measure is identified with a singleton (see Remark 9.5). However unlike Dirac
measures, infinitely many elements of G may be identified with the same point and
some elements of G are identified with the point at infinity.
A simple consequence of Theorem 5.4 is that

ω(A B) = ω(A)ω(B) for all A, B ∈ L, ω ∈ G. (5.4)
5.2 G and the λ-Finite Intersection Property
A finitely additive measure on L belongs to L ∗∞ (X, L, λ) if ν(N ) = 0 for all N ∈ N

and Theorem 3.1 combined with the Hahn–Banach theorem yields the existence of a
wide variety of elements of L ∗∞ (X, L, λ). The following complementary approach,
also based on Zorn’s lemma, concerns the existence of elements of G which are
determined (non-uniquely) by families of sets with the following property.
E ⊂
Definition 5.5 (λ-finite intersection property) L has the λ-finite intersection
K
property if E k ∈ E, 1 k K , implies λ k=1 k > 0.
E
Theorem 5.6 If E has the λ-finite intersection property, there exists ω ∈ G with
ω(E) = 1 for all E ∈ E.
Proof Consider the collection F of filters F with E ⊂ F. Then F = ∅ since

K
A∈L: A⊃ E k , E k ∈ E, K ∈ N ∈ F.
k=1
Now suppose that T ⊂ F is a family of filters which is totally ordered by set inclusion
and let F ∗ = F ∈T F. Then F ∗ ∈ F is an upper bound for T. So by Zorn’s lemma,
F has a maximal element U which is an ultrafilter. By Theorem 5.4 (a), ωU ∈ G
satisfies the theorem.
The next result shows that for given E there may be uncountably many ω ∈ G that
satisfy Theorem 5.6.
Corollary 5.7 (a) For A ∈ L \ N there exists ω∈ G with ω(A) = 1.

(b) Suppose, for j ∈ N, that F j ∈ L \ N and Fi F j ∈ N when i = j. Let
5.2 G and the λ-Finite Intersection Property 45

E = E k : k ∈ N where E k = Fj .
jk
Then for uncountably many distinct ω ∈ G \ Σ(L), ω(E k ) = 1 for all k.

Proof (a) This follows from Theorem 5.6 with E = {A}.
(b) For a ∈ (0, 1) let q j (a) ∈ Q, j ∈ N, be suchthat q j−1 (a) < q j (a) a as
j → ∞. Let Q a = {q j (a) : j ∈ N} and note that Q a Q b is at most finite if a = b.
Now let Na = η(Q a ) where η : Q → N is a bijection. So {Na : a ∈ (0,1)} is an
uncountable collection of infinite subsets of N with the property that Na Nb is at
most finite if a = b. In particular, for a = b ∈ (0, 1) there exists K (a, b)∈ N such
that { j ∈ Na Nb : j K (a, b)} = ∅. Consequently a = b implies Fi F j ∈ N
for all i ∈ Na , j ∈ Nb with i, j K (a,b).
Now for a ∈ (0, 1) note that E ka := k j∈Na F j ∈ L, since L is a σ-algebra, and
let Ea = {E ka : k ∈ N}. Then by Theorem 5.6there exists ωa ∈ G with ωa (E ka ) = 1
for all k ∈ N. However, by construction, E ka E kb ∈ N for all k K (a, b), a = b.
It follows that ωa = ωb since 1 = ωa (E ka ) = ωb (E ka ) = 0 for k sufficiently large.
Moreover, for all a, E ka ⊂ E k , k ∈ N, which implies ωa (E k ) = 1 for all a ∈ (0, 1)
and k ∈ N.
Finally, that ωa ∈/ Σ(L) follows from Lemma 4.2(a) since k E ka = ∅ but E k+1 a
⊂
E k and ω(E k ) = 1 for all k. This completes the proof.
a a

Example 5.8 As in Example 2.14, the space of all bounded sequences of real num-
bers indexed by N, usually denoted by ∞ , coincides with the space of bounded
functions u : N → R. With ∞ (N) = L ∞ (X, L, λ), where X = N, L = ℘ (N) and
λ is counting measure, let E = {E k : k ∈ N}, where E k = k + N. Then E has the
λ-finite intersection property and E k satisfies the hypotheses of Corollary 5.7 with
F j = { j}. Hence, there exist uncountably many ω ∈ G with ω(E k ) = 1 for all k.
This example underlies Sect. 6.5 on integration of sequences in ∞ (N), and leads to
Remark 9.13.

Theorem 5.9 For ω ∈ G
(a) either ω ∈ Π (L) or ω ∈ Σ(L).
(b) ω ∈ Σ(L) if and only if for an atom E ω ∈ L (Definition 2.8)

λ(E E ω )
ω(E) = for all E ∈ L.
λ(E ω )
Proof (a) If this is false, by Theorem 4.13 ω = μ + γ where μ ∈ Π (L) and γ ∈

Σ(L) are non-negative, ω(X ) = 1, μ(X ) ∈ (0, 1), and by Theorem 4.12 there exists
{E k } ⊂ L with μ(E k ) = μ(X ) for all k and γ(E k ) → 0 as k → ∞.
If ω(E k ) = 0 for some k then 0 = ω(E k ) μ(E k ) = μ(X ), whence μ(X ) = 0
which is false. Therefore, since ω ∈ G, ω(E k ) = 1 for all k and it follows that
1 = ω(E k ) = μ(E k ) + γ(E k ) = μ(X ) + γ(E k ) → μ(X ) as k → ∞.

Hence μ(X ) = 1 which

is also false. This proves (a).
(b) Suppose ω ∈ G Σ(L). Then ω is a real measure and both ω and λ are
σ-additive with ω λ. Since by hypothesis λ is σ-finite, by Theorem 2.28(b)
(Radon–Nikodym) there exists g ∈ L 1 (X, L, λ) with

ω(E) = g dλ for all E ∈ L.
E
So g is non-negative λ-almost everywhere on X , λ ({x ∈ X : g(x) n}) → 0 as

n → ∞, and hence
g dλ → 0 as n → ∞.
{x∈X :g(x)n}
In particular,

ω x ∈ X : g(x) n = g dλ → 0 as n → ∞.
{x∈X :g(x)n}
Now ω ∈ G implies that for some N ∈ N, ω ({x ∈ X : g(x) N }) = 0 and it follows

that λ ({x ∈ X : g(x) N }) = 0. Thus g ∈ L ∞ (X, L, λ) and by (5.2), for a unique
α ∈ R,
ω x ∈ X : |g(x) − α| < = 1 for all > 0.

Since ω ∈ Σ(L) it follows that ω E ω = 1 where E ω = {x ∈ X : g(x) = α}. More-

over λ(E ω ) > 0 since ω(E ω ) = 1 and ω ∈ G. Therefore

ω(E) = ω(E E ω ) = α dλ = αλ(E E ω ) for all E ∈ L.
E Eω
With α = 1/λ(E ω ), E ω is an atom with the required

properties since ω ∈ G.
Conversely, when A is an atom ω(E) := λ(E A)/λ(A), E ∈ L, defines ω ∈
G Σ(L) since λ is σ-additive.
Remark 5.10 This shows that all elements of G are purely finitely additive when
L has no atoms. In any case if X has infinitely many mutually disjoint subsets
with positive λ-measure, by Corollary 5.7(b) there are uncountably many distinct
ω ∈ G which are purely finitely additive. For example, in ∞ (N) G has countably
many σ-additive elements given by the atoms {n}, n ∈ N, and uncountably many
ω ∈ G which are purely finitely additive are given by Corollary 5.7 and Theorem 5.9.
Further statements can be made about G, e.g. Lemmas 7.7, or 9.1 in a locally compact
Hausdorff setting.
Chapter 6
Integration and Finitely Additive
Measures
The integral of u ∈ L ∞ (X, L, λ) with respect to ν ∈ L ∗∞ (X, L, λ) will been defined

and sufficient of its properties established to prove the representation Theorem 3.1
(Yosida–Hewitt). Then the special case of integration with respect to ω ∈ G will
be examined and the essential range of u ∈ L ∞ (X, L, λ) introduced. The chapter
ends with an account of the Valadier–Hensgen example. For a more comprehensive
account of integration with respect to finitely additive measures, see [6, Chap. 4],
[12, Chap. III] and [35].
6.1 The Integral
To begin note that if u ∈ L ∞ (X, L, λ) and w(x) = u(x) for λ-almost all x ∈ X , it
follows that |ν|({x : w(x) = u(x)} = 0 for all ν ∈ L ∗∞ (X, L, λ) since |ν|(N ) = 0
when λ(N ) = 0. Therefore, by finite additivity the following construction is inde-
pendent of whether u or w is used to represent an equivalence class in L ∞ (X, L, λ).
Since, by (4.1c), ν = ν + − ν − and by Theorem 4.4(d) ν ± ∈ L ∗∞ (X, L, λ) when
ν ∈ L ∗∞ (X, L, λ), it suffices first to define integration with respect to non-negative
ν ∈ L ∗∞ (X, L, λ) and to extend by linearity.
For u ∈ L ∞ (X, L, λ) and p > u∞ , let
P: − p = y0 < y1 < · · · < y K < y K +1 = p, K ∈ N, (6.1a)
be a partition of [− p, p] and let

K
K
S P (u) := yk ν(E k (u)) yk+1 ν(E k (u)) =: S P (u), (6.1b)
k=0 k=0
where E k (u) = {x : yk u(x) < yk+1 }, 0 k < K . As remarked above, S P (u)

and S P (u) are independent of which function u is used to represent an element of
https://doi.org/10.1007/978-3-030-34732-1_6
48 6 Integration and Finitely Additive Measures
L ∞ (X, L, λ). Since ν is non-negative, finitely additive and ν(X ) < ∞,
0 S P (u) − S P (u) < ρ(P)ν(X ),
where ρ(P) = max{yk+1 − yk : 0 k K }. Furthermore, if a partition P is aug-

mented by adding an additional point then S P (u) is not decreased and S P (u) is
not increased. It follows that over all partitions P, inf P S P (u) = supP S P (u). With
these observations, the integral of u ∈ L ∞ (X, L, λ) with respect to ν 0 is defined
as

inf S P (u) = u dν = sup S P (u),
P X P

or equivalently lim S P (u) = u dν = lim S P (u). (6.2a)
ρ(P)→0 X ρ(P)→0
It follows that when ν 0 and u v λ-almost everywhere on X ,

u dν v dν. (6.2b)
X X
Now the integral of u ∈ L ∞ (X, L, λ) with respect to general ν ∈ L ∗∞ (X, L, λ) can

be defined unambiguously by

u dν := u dν + − u dν − , (6.2c)
X X X
from which it follows that

χ E dν = ν(E), E ∈ L, ν ∈ L ∗∞ (X, L, λ). (6.2d)
X
For simplicity of notation write

u dν := χ E u dν, E ∈ L. (6.3a)
E X
Note that for E ∈ L,

u dν u∞ |ν|(E) and u dν = 0 if E ∈ N . (6.4)

E E
N
Moreover, when E = n=1 E n , a finite union of sets in L with E i Ej ∈ N
when i = j,
6.1 The Integral 49
N

u dν = u dν. (6.5)
E n=1 En
Since (−ν)± = ν ∓ , it follows from (6.2c) that

αu dν = α u dν, u ∈ L ∞ (X, L, λ), α ∈ R. (6.6)
X X
Therefore, the integral is linear on L ∞ (X, L, λ) if it is shown that

(u + w) dν = u dν + w dν, u, w ∈ L ∞ (X, L, λ), (6.7)
X X X
and by (6.2c) it is enough to consider ν 0.

Let p = 1 + u∞ + w∞ and for > 0 let P be a partition of [− p, p] for
which S P (u) − S P (u) < . Then, by (6.5),
K

(u + w) dν = K
(u + w) dν = (u + w) dν
X k=0 E k (u) k=0 E k (u)
K
K
k=0 (yk + w) dν = S P (u) + k=0 w dν

K
Ek (u) K
Ek (u)
k=0 Ek (u) (yk+1 + w) dν = S P (u) + k=0 E k (u) w dν
= S P (u) +
X w dν
X u dν +
X w dν −
.
= S P (u) + X w dν X u dν + X w dν +
Since > 0 was arbitrary, (6.7) follows when 0 ν ∈ L ∗∞ (X, L, λ). The general
case is immediate from (6.2c).

Remark 6.1 When γ ∈ Σ(L) L ∗∞ (X, L, λ), it is obvious that integration of u ∈
L ∞ (X, L, λ) with respect to γ coincides with Lebesgue integration. However, if ν is
not σ-additive, familiar results such as the Monotone convergence theorem do not
hold for this integral. For example, by Corollary 5.7(b), on the interval (0, 1) there
are uncountably many finitely additive measures ν ∈ G such that ν(0, 1 − 1/k] = 0
and ν(1 − 1/k, 1) = 1 for all k > 1, k ∈ N. It follows from (6.2a) that for all such ν
1 1− k1
1 dν = 1 but u dν = 0 for all k ∈ N and all u ∈ L ∞ (X, L, λ).
0 0
This establishes (†) in the Preface.

∗
be written as ν = μ + γ where μ ∈ L ∗∞
4.16, ν ∈ L ∞ (X, L, λ) can
By Theorem
(X, L, λ) Π (L) and λ γ ∈ Σ(L) L ∗∞ (X, L, λ). Since λ is a σ-finite
measure, by Theorem 2.28(b) (Radon–Nikodym) there exists g ∈ L 1 (X, L, λ) with

u dγ = ug dλ for all u ∈ L ∞ (X, L, λ). (6.8)
X X
So (4.6) can be rewritten as

ν = μ + gλ, μ ∈ Π (L) L ∗∞ (X, L, λ), g ∈ L 1 (X, L, λ). (6.9)
The relation of (6.9) to Theorem 2.28(a) when λ is Borel measure is examined in

Chap. 10.
6.2 Yosida–Hewitt Representation: Proof of Theorem 3.1
Having prepared the ground in Sects. 4.1 and 6.1, the proof of Theorem 3.1, which
identifies elements of L ∞ (X, L, λ)∗ with elements of L ∗∞ (X, L, λ), and vice versa,
is straightforward.
In Sect. 6.1, it was shown that when ν ∈ L ∗∞ (X, L, λ) a bounded linear functional
f ∈ L ∞ (X, L, λ)∗ is well defined on L ∞ (X, L, λ) by

f (u) = u dν and | f (u)| u∞ |ν|(X ).
X
Hence f ∞ = sup{| f (u)| : u∞ = 1} |ν|(X ). To see that equality holds note
that, since ν + ∧ ν − = 0, by (4.1c) for any > 0 there exists a set A ∈ L with
ν + (A) + ν − (A ) < , where A = X \ A. Let u = χ A − χ A ∈ L ∞ (X, L, λ). Then
u∞ = 1 and

f (u) = u dν = u dν + + u dν + − u dν − − u dν −
X A A A A
= −ν (A) + ν (A ) + ν (A) − ν (A ) ν (A ) + ν − (A) −
+ + − − +
= ν + (X ) − ν + (A) + ν − (X ) − ν − (A ) − (|ν|(X ) − 2)u∞ .
Therefore f ∞ |ν|(X ) and equality follows.

For the converse, given f ∈ L ∞ (X, L, λ)∗ define ν on L by ν(E) = f (χ E ), E ∈
L. The goal is to show that ν ∈ L ∗∞ (X, L, λ) and (3.1b) holds.
First note that ν(E) = 0 if E ∈ N because χ E ∞ = 0. Moreover,
ν is finitely
additive because f is linear and χ E1 E2 = χ E1 + χ E2 when E 1 E 2 = ∅. It remains
only to show that (3.1b) holds.
For u ∈ L ∞ (X, L, λ) let P in (6.1a) be a partition into K + 1 equal intervals, and
K
let u K := k=0 yk χ Ek (u) ∈ L ∞ (X, L, λ). Since u K → u in L ∞ (X, L, λ) and f is a
bounded linear functional
6.2 Yosida–Hewitt Representation: Proof of Theorem 3.1 51

K
K
f (u) = lim f (u K ) = yk f χ Ek (u) = yk ν(E k (u)) → u dν
K →∞ X
k=0 k=0
by (6.2a) and (6.2c). Thus (3.1b) holds and the proof is complete.
6.3 Integration with Respect to ω ∈ G
Theorem 6.2 (a) For u ∈ L ∞ (X, L, λ) and ω ∈ G

u dω = α and |u| dω = |α| (6.10a)
X X
where, as in Theorem 5.1, α is such that

ω x ∈ X : |u(x) − α| < = 1 for all > 0.
(b) For u, v ∈ L ∞ (X, L, λ) and ω ∈ G

uv dω = u dω v dω . (6.10b)
X X X
(c) If F : R → R is continuous, u ∈ L ∞ (X, L, λ) and ω ∈ G,

F(u) dω = F u dω . (6.10c)
X
Proof (a) With α given by Theorem 5.1, the first part of (6.10a) follows because

u dω − α = (u − α) dω = (u − α) dω <

X X {x∈X :|u(x)−α|<}
for all > 0. The second part follows because, for all > 0

ω x ∈ X : |u(x)| − |α| < ω x ∈ X : |u(x) − α| < = 1.
(b) If
u dω = α and v dω = β,
X X
it follows that for all > 0,

ω x ∈ X : |u(x) − α| < = 1 and ω x ∈ X : |v(x) − β| < = 1
and hence, by (5.4),

ω x ∈ X : |u(x)v(x) − αβ| < = 1, for all > 0.
By (6.10a) the proof of (b) is complete.

(c) It follows from (6.10b) that (6.10c) holds when F is a polynomial, and therefore
for continuous F by approximation using Theorem 2.43 (Weierstrass).
6.4 Essential Range of u ∈ L ∞ (X, L, λ)
In a general measure space (X, L, λ), the essential range [28, Chap.3, Ex.19] of
u ∈ L ∞ (X, L, λ) is defined as

R(u) = α ∈ R : λ {x : |u(x) − α| < } > 0 for all > 0 . (6.11a)
Lemma 6.3 For u ∈ L ∞ (X, L, λ),

R(u) = u dω : ω∈G . (6.11b)
X
Proof From Theorems 5.6 and 6.2, the right side contains the left. Since ω(E) = 1
implies λ(E) > 0, by Theorem 6.2 the left contains the right.
For another viewpoint on the essential range, see (7.3). In a Borel setting the essential
range can be localised, see Sect. 9.3.
6.5 Integrating u ∈ ∞ (N) with Respect to G
Recall from Example 5.8 that bounded real sequences {u j } ∈ ∞ can be identified
with the essentially bounded functions u ∈ ∞ (N) defined by u( j) = u j , j ∈ N, and
there are uncountably many ω̂ ∈ G, each with the property that ω̂(E k ) = 1 for all k,
where E k = k + N.
Theorem 6.4 A sequence {u j } ∈ ∞ has a subsequence that converges
to α ∈ R if
and only if there exists ω ∈ G with ω(E k ) = 1 for all k ∈ N and N u dω = α.

Proof Let ω̂ ∈ G have the property that ω̂(E k ) = 1 for all k and suppose α = N u d ω̂
where u ∈ ∞ (N) is defined by u( j) = u j . Then by (5.4) and Theorem 6.2(a)

ω̂ j ∈ E k : |u j − α| < = 1 for any k ∈ N and any > 0.
6.5 Integrating u ∈ ∞ (N) with Respect to G 53
Hence, for any k ∈ N and > 0, there exists j k such that |u j − α| < . In particu-
lar, there exists j1 ∈ N such that |u j1 − α| < 1 and j2 > j1 such that |u j2 − α| < 1/2.
Proceeding by induction,
for each n ∈ N there exists jn > jn−1 such that |u jn − α| <
1/n. This shows that N u d ω̂ = α, ω̂ ∈ G, implies a subsequence u jn → α as
n → ∞.
Now suppose {u j } has a subsequence u jn → α, let E = { Ẽ k : k ∈ N} where Ẽ k =
{ jn : n k} and let ω̃ ∈ G be a finitely additive measure with ω̃( Ẽ k ) = 1 for all k ∈ N
given by Theorem 5.6. Then for > 0 there exists k ∈ N such that

Ẽ k ⊂ { j : |u j − α| < }, and hence 1 = ω̃( Ẽ k ) ω̃ { j : |u j − α| < } .

By Theorem 6.2 this implies N u d ω̃ = α.
In Remark 9.13, the preceding observation is related to the essential range at infinity
of u ∈ ∞ (N).
Remark 6.5 For u ∈ ∞ it is immediate that

u j → α as j → ∞ ⇒ u dω = α
N
for all ω ∈ G with ω(E k ) = 1 for all k. Let S : ∞ (N) → ∞ (N) be the shift operator
defined by (Su)( j) = u( j + 1). Then by (6.10b), for ω ∈ G,

u S(u) dω = u dω S(u) dω , u ∈ ∞ (N).
N N N
In particular, when u( j) = (−1) j it follows that u S(u) ≡ −1 and hence that

u dω S(u) dω = −1
N N

In other words, u → N u dω, u ∈ ∞ (N), is not shift invariant, and therefore is not
a Banach limit (Definition 2.41 and [31]) for any ω ∈ G.
However, a Banach limit can be constructed as follows. For u ∈ ∞ (N) let T u ∈
∞ (N) be defined by
1
j
(T u)( j) = u(i), i, j ∈ N.
j i=1
Then T : ∞ (N) → ∞ (N) is a bounded linear operator with T = 1 and

T u( j) − T (Su)( j)| = 1 |u( j + 1) − u(1)| → 0 as j → ∞.
j

It follows that if ω ∈ G and ω(E k )=1 for all k then ω j : |T u( j) − α| < =1
for all > 0 if and only if ω j : |T (Su)( j) − α| < = 1 for all > 0. Hence, by
Theorem 6.2(a),

T u dω = T (Su) dω if ω ∈ G and ω(E k ) = 1 for all k.
N N

Since u( j) → α implies T u( j) → α as j → ∞, it follows that u → N T u dω,

u ∈ ∞ (N), is a Banach limit.
6.6 The Valadier–Hensgen Example
Independently, Valadier [33] and Hensgen [19] made the following observation
which contradicts a claim in [35].
When L and λ refer to Lebesgue measure on X = [0, 1], there exists

Theorem 6.6
μ ∈ Π (L) L ∗∞ (X, L, λ) such that
1 1
v dμ = v dλ for all continuous v : [0, 1] → R. (6.12)
0 0
For x ∈ [0,
Proof 1] let μx ∈ G be given by Theorem 5.6 with the property that
μ (x − , x + ) [0, 1]) = 1 for all > 0, and note from Theorem 6.2 that
x
1
v dμx = v(x) for all v ∈ C[0, 1].
0
For n ∈ N, let x nj = j/2n , 0 j 2n , j ∈ N0 , and for v ∈ C[0, 1] let

1 1 x nj
2n 1 2n
snv = n v(x nj ) = v dμn , where μn = n μ . (6.13)
2 + 1 j=0 0 2 + 1 j=0
n
For all n ∈ N, μn ∈ L ∗∞ (X, L, λ) and μn (E) ∈ [0, 1], E ∈ L, since each μx j ∈ G.
Moreover, snv is an n th Riemann sum of v ∈ C[0, 1] corresponding to partitioning
[0, 1] with 2n + 1 equally spaced points. Hence s v := {snv } ∈ ∞ is convergent and
1
lim s v = v dλ, v ∈ C[0, 1],
n→∞ n 0
or, in the notation of Definition 2.41,

6.6 The Valadier–Hensgen Example 55
1
v v
s ∈ c(N) and l(s ) = v dλ for all v ∈ C[0, 1].
0
Now define μ on L by
μ(E) = L {μn (E)} , E ∈ L, (6.14)
where L : ∞ (N) → R is a Banach limit, Definition 2.41. Clearly μ 0 is finitely

additive. Also μ(E) = 0 for all E ∈ N since μn is zero on N .
To see μ is purely finitely additive, for k ∈ N let
n 1 1
E k = 0, 1 x j − 2n , x nj + 2n .
{ j,n:
2 k 2 k
0 j2n
n∈N}
Note that since μn (E k ) = 1 for all n ∈ N, it follows that μ(E k ) = 1 = μ(X ) for all k.
Moreover, since {Ek } is a nested sequence of open subsets
of [0, 1] with λ(E k ) 2/k,
it follows that λ k E k = 0. Hence γ(E k ) → γ k E k =
0 if 0 γ ∈ Σ(L). S-
ince μ 0 this shows that μ ∧ γ = 0 for any 0 γ ∈ Σ(L) L ∗∞ (X, L, λ) whence,
by Theorem 4.15, μ is purely finitely additive. It remains
M to prove (6.12).
When g is a simple function on [0, 1] (i.e. g = m=1 gm χG m , gm ∈ R, G m ∈ L)
it follows from (6.2d) and (6.14) that
1 1
L g dμn = g dμ. (6.15)
0 0
Now for an arbitrary u ∈ L ∞ ([0, 1], L, λ) let u k → u uniformly on [0, 1] where {u k }

is a sequence of simple functions. Then for every k ∈ N
1 1 1
1
L u dμn − u dμ L u dμn − u k dμn

0 0 0 0
1 1 1 1

+ L u k dμn −
u k dμ + u k dμ − u dμ .
0 0 0 0
Now the first and third terms on the right converge to 0 as k → ∞ and by (6.15) the
middle term is zero for all k. Hence,
1 1
L u dμn = u dμ for all u ∈ L ∞ ([0, 1], L, λ).
0 0
In particular, for v ∈ C[0, 1]

1 1 1
v dμ = L v dμn = lim snv = v dλ.
0 0 n→∞ 0
Hence (6.12) is proved.
Remark 6.7 Yosida and Hewitt [35, Thm. 3.4] claimed that
1
v dν = 0 for all v ∈ C[0, 1] (6.16)
0
implies ν is purely finitely additive but omitted the proof because it was consid-
ered straightforward. However, Valadier [33], and later Hensgen [19], independently
showed that this claim is wrong. In Theorem 6.6 μ ∈ Π (L), λ ∈ Σ(L) and so, by the
uniqueness statement in Theorem 4.13, ν := μ − λ ∈ L ∗∞ (X, L, λ) is neither purely
finitely additive nor σ-additive yet satisfies (6.16).
Different choices of the points x nj lead to different μ. So if {y nj : j ∈ N} {x nj :
j ∈ N} = ∅, there are two measures μ1 , μ2 ∈ Π (L) which satisfy (6.12). Hence
0 = μ1 − μ2 = ν ∈ Π (L) also satisfies (6.16). For further discussion of these issues,
see Sect. 10.2.
Chapter 7
Topology on G
7.1 The Space (G, τ )
For A ∈ L, let Δ A = {ω ∈ G : ω(A) = 1} and note that Δ A ⊂ Δ B if A ⊂ B and G \

Δ A = Δ X \A . More generally, from finite additivity it follows that for finite families,

K
K
Δ Ak = Δ1K Ak and Δ Ak = Δ1K Ak , Ak ∈ L, (7.1a)
1 1
whereas for arbitrary families only inclusions hold:

Δ Aα ⊂ Δα Aα and Δα Aα ⊂ Δ Aα . (7.1b)
α α
For further developments, see Theorem 7.10 and Corollary 7.11.

Lemma 7.1 Suppose A, B ∈ L. Then Δ A Δ B = ∅ if and only if A B ∈ / N.
Moreover, B \ A ∈ Δ B ⊂ Δ A . Consequently, Δ A = Δ B if and only
/ N if and only if
if the symmetric difference (A \ B) (B \ A) ∈ N .
Proof The first part is immediate

from (7.1a), since elements of G are 0 on N , and
from Theorem 5.6 since A B ∈ / N implies Δ A B = ∅. It follows that B \ A ∈
/N
if and only if Δ B \ Δ A = ∅ and the observation on the symmetric difference is
immediate.
Definition 7.2 ( (G, τ ) ) Let T := {Δ A : A ∈ L} be a sub-base for the topology τ on

G. (See Remark 2.57.) Equivalently, by (7.1a), T is a base for τ , see Definition 2.56.
Note that Δ A , A ∈ L, is open and closed because G \ Δ A = Δ X \A .
Theorem 7.3 (G, τ ) is a compact Hausdorff space.

https://doi.org/10.1007/978-3-030-34732-1_7
58 7 Topology on G
Proof Let τ be the topology defined in Definition 7.2. To see that (G, τ ) is Hausdorff
suppose that ω1 = ω2 . Then there exists A ∈L such that ω1 (A) = 1 and ω2 (A) = 0.
Hence ω1 ∈ Δ A and ω2 ∈ Δ X \A where Δ A Δ X \A = ∅. Thus (G, τ ) is Hausdorff.
To show
that (G, τ ) is compact suppose that G = A∈A Δ A for some A ⊂ L and
G = A∈A Δ A for any finite subset A
that of A. Since Δ X \A = G \ Δ A , it follows
that A∈A Δ X \A = ∅ but, by (7.1a),

Δ( = ⊂ A is finite.
Δ X \A = ∅ when A
X \A)
A∈A

A∈A

By Lemma 7.1 this implies that A∈A(X \ A) ∈ / N for any finite A ⊂ A.
Since E := {X \ A : A ∈ A} satisfies the hypotheses of Theorem 5.6, there exists
ω ∈ G such thatω(X \ A) = 1, and consequently ω(A) = 0, for all A ∈ A. Since
ω ∈ G and ω ∈ / A∈A Δ A , this is a contradiction. It follows that G ⊂ A∈A Δ A for
⊂ A and hence, since T is a base for τ , (G, τ ) is compact.
some finite A
7.2 L ∞ (X, L, λ) and C(G, τ ) Isometrically Isomorphic
The next observation is that L ∞ (X, L, λ) and C(G, τ ) are isometrically isomorphic
with the consequence that their duals are isometrically isomorphic, even though one
is represented by finitely additive measures, Theorem 3.1, and the other by σ-additive
measures, Theorem 2.37 (Riesz).
For u ∈ L ∞ (X, L, λ) let L[u] : G → R be defined by

L[u](ω) = u dω for all ω ∈ G. (7.2)
X
Note from Lemma 6.3 that the essential range of u ∈ L ∞ (X, L, λ) is the range of
L[u] on G:
R(u) = L[u](G). (7.3)
Theorem 7.4 L[u] is continuous on (G, τ ),
u∞ = L[u]C(G,τ ) =: sup |L[u](ω)|, (7.4)

ω∈G
and the mapping u → L[u], from L ∞ (X, L, λ) to C(G, τ ) is linear. Moreover, for
u, v ∈ L ∞ (X, L, λ),
L[u](ω)L[v](ω) = L[uv](ω) for all ω ∈ G. (7.5)

7.2 L ∞ (X, L, λ) and C(G, τ ) Isometrically Isomorphic 59
Conversely, for every real-valued continuous function u on (G, τ ) there exists u ∈

L ∞ (X, L, λ) with u = L[u]. In other words, L is an isometric isomorphism between
the Banach algebras L ∞ (X, L, λ) and C(G, τ ).

Proof For u 0 ∈ L ∞ (X, L, λ) and ω0 ∈ G let α0 = L[u 0 ](ω0 ) = X u 0 dω0 and for
0 > 0 put A0 = {x ∈ X : |u 0 (x) − α0 | < 0 /2}. Then ω0 ∈ Δ A0 by Theorem 6.2
and {x ∈ X : |u 0 (x) − α̂| < 0 /2} A0 ∈ N if |α̂ − α0 | 0 . Therefore ω({x ∈
X : |u 0 (x) − α̂| < 0 /2}) = 0, and consequently X u 0 dω = α̂, if ω ∈ Δ A0 and
|α̂ − α0 | 0 . It follows that

L[u 0 ](ω) − α 0
=
u 0 dω − α0
< 0 if ω ∈ Δ A0 .
X
Hence, for any u 0 ∈ L ∞ (X, L, λ), L[u 0 ] : (G, τ ) → R is continuous at any ω0 ∈ G,

as required.
Identity (7.5) is a restatement of (6.10b) in the present notation.
It follows from (6.4), (6.6) and (6.7) that u → L[u] is linear from L ∞ (X, L, λ) to
C(G, τ ) and supω∈G |L[u](ω)| u∞ . To prove equality, suppose u∞ = [u]+
(see (2.8), if not replace u with −u). Then for > 0 let D = {x : u∞ − u(x) <
/ N . By Theorem 5.6 there exists ω ∈ G such that ω(D ) = 1. Hence, for any
} ∈
>0

u∞ − L[u](ω) = u∞ − u dω = (u∞ − u) dω .
X D
Therefore (7.4) holds.

Now let L = L[u] : u ∈ L ∞ (X, L, λ) ⊂ C(G, τ ). Then L[1] ∈ L is the con-
stant function 1 and by (7.5) the product of two elements of L is in L. If ω1 = ω2
there
is a set E ∈ L with ω1 (E) = 1 and ω2 (E) = 0 and, since X χ E dω1 = 1 = 0 =
X χ E dω2 , L separates points in G. Moreover, by (7.4) L is closed since L ∞ (X, L, λ)
is a Banach space. Therefore, by Theorem 2.42 (Stone–Weierstrass), L = C(G, τ )
since (G, τ ) is a compact Hausdorff space. This completes the proof.
7.3 Properties of G and τ
Lemma 7.5 H is open and closed in (G, τ ) if and only if H = Δ A , A ∈ L.

Proof First recall that Δ A is open and closed because Δ A and its complement G \
Δ A = Δ X \A are open. Conversely, if H is closed in (G, τ ) it is compact. Hence, if it
is also open, there exists Ak ∈ L, 1 k K , such that, by (7.1a),

K
K
H= Δ Ak = Δ A where A = Ak ∈ L,
k=1 k=1
and the result follows.

60 7 Topology on G
Theorem 7.6 (G, τ ) is totally disconnected, meaning all its connected sets are sin-
gletons.
Proof Suppose H ⊂ G and ω1 , ω2 ∈ H are distinct. Then, for some A ∈ L, ω1 (A)=1

and ω2 (A) = 0, whence ω1 ∈ Δ A and ω2 ∈ G \ Δ A = Δ X \A . Therefore, H is not
connected since Δ X \A and Δ A are disjoint open sets the union of which is G. Clearly
a singleton {ω} is connected.
Lemma 7.7 For A ∈ L, Δ A is a singleton if and only if A is an atom in (X, L, λ).

Proof If A A = A1 A2 where A1 A2 ∈ N , A1 , A2 ∈
is not an atom, / N . Hence
Δ A = Δ A1 Δ A2 , Δ A1 Δ A2 = ∅ and Δ A1 , Δ A2 = ∅, and so Δ A is not a singleton.
Conversely, ω1 , ω2 ∈ Δ A , ω1 = ω2 , implies there exist A1 , A2 ⊂ A with ω1 (A1 ) =
1 =ω2 (A2 ), ω1 (A2 ) = 0 = ω2 (A1 ). It follows from Lemma 7.1 that A1 , A2 , (A1 \
A2 ) (A2 \ A1 ) ∈ / N , and so A is not an atom if Δ A is not a singleton.
Corollary 7.8 When (X, L, λ) has no atoms every element of G is purely finitely
additive and no element of G is isolated.
Proof Since by definition ω is isolated in (G, τ ) if and only if {ω} = Δ A for some
A ∈ L, by Lemma 7.7 no element of G is isolated if there are no atoms. The pure
finite additivity of ω ∈ G follows by Theorem 5.9.
In a Borel measure space (X, B, λ), a further implication of the absence of atoms is
the following. Recall from Sect. 2.10 that a topological space (X, ) is completely
separable and it has a denumerable basis (a denumerable collection T = {Ti } of
non-empty open sets such that every open set is a union of elements of T ).
Lemma 7.9 If B has no atoms, (G, τ ) is not completely separable.
Proof Let {Ti : i ∈ N} be an arbitrary denumerable family of non-empty sets in τ and

let ∅ = ΔG i ⊂ Ti , G i ∈ L. Since λ(G i ) > 0 for all i, by Theorem 2.25 there exists
A ∈ L which splits G = {G i : i ∈ N}. Since G i \ A ∈ / N it follows from Lemma 7.1
that ΔG i ⊂ Δ A , and hence that Ti ⊂ Δ A , for all i ∈ N. Since Δ A ∈ τ , it follows that
{Ti : i ∈ N} is not a base for τ . This completes the proof.
This discussion of (G, τ ) is continued in Sect. 9.1, but for that the following
refinement of (7.1b) is needed.

Theorem 7.10 (i) If Γ = k∈N Δ Ak , Ak ∈ L:

(a) Γ ◦ = Δ A0 , where Γ ◦ is the interior of Γ in (G, τ ) and A0 = k∈N Ak .
(b) Both Γ and Γ ◦ are closed.
(c) The boundary ∂Γ = Γ \ Γ ◦ of Γ is characterised by

∂Γ = ω ∈ G : ω(Ak ) = 1 for all k ∈ N but ω(A0 ) = 0 .
7.3 Properties of G and τ 61
(d) If in addition Ak+1 ⊂ Ak , k ∈ N, Ak \ Ak+1 ∈

/ N for infinitely many k, then ∂Γ
is uncountable.
(ii) If Λ = k∈N Δ Bk , Bk ∈ L:

(a) Λ = Δ B 0 , where Λ is the closure of Λ in (G, τ ) and B 0 = k∈N Bk .
(b) Both Λ and Λ are open.
(c) The boundary ∂Λ = Λ \ Λ◦ of Λ is characterised by

∂Λ = ω : ω(Bk ) = 0 for all k ∈ N but ω(B 0 ) = 1 .
(d) If, in addition, Bk ⊂ Bk+1 , k ∈ N, and Bk+1 \ Bk ∈

/ N for infinitely many k, then
∂Λ is uncountable.
Proof (i)(a) Since Δ A0 is open, it follows from (7.1b) that

◦
Δ A0 = Δk Ak ⊂ Δ Ak = Γ ◦.
k
Suppose ω ∈ Γ ◦ \ Δ A0 and let E ∈ L \ N be such that ω ∈ Δ E ⊂ Γ . It follows that

E \ A k0 ∈
/ N for some k0 ∈ N, for otherwise

E\ Ak = (E \ Ak ) ∈ N ,
k k
whence

1 = ω(E) = ω(E \ A0 ) + ω(E A0 ) = ω E \ Ak + ω(E A0 ) = 0,
k
a contradiction. Since E \ Ak0 ∈ / N , it follows from Theorem 5.6 that there exists
ω0 ∈ G with ω0 (E \ Ak0 ) = 1. This implies that ω0 (Ak0 ) = 0 and hence that ω0 ∈/ Γ.
But ω0 (E \ Ak0 ) = 1 also implies that ω0 ∈ Δ E ⊂ Γ . This contradiction implies that
Γ ◦ \ Δk Ak is empty and hence that Δ A0 = Γ ◦ .
(i)(b, c) Since Γ is an intersection of closed sets, Γ = Γ is closed, Γ ◦ is closed
by Lemma 7.5 and the formula for the boundary ∂Γ = Γ \ Γ ◦ follows.
(i)(d) Now suppose Ak+1 ⊂ Ak , k ∈ N, and Ak \ Ak+1 ∈ / N for infinitely many
k. Then there is no loss since, by Lemma 7.1, Ak \ Ak+1 ∈ N implies Δ Ak = Δ Ak+1 ,
in assuming that F k := Ak \ Ak+1 ∈ / N for all k. Since the Fk are non-null and
disjoint, and since jk F j ⊂ Ak , Corollary 5.7 implies the existence of a family

{ωa : a ∈ (0, 1)} ⊂ G of distinct finitely additive measures with ωa jk j =
F
ωa (Ak ) = 1, k ∈ N. Hence ωa ∈ Γ for all a ∈ (0, 1). However

A0 F j = A0 Ak \ A0 = ∅.
jk
62 7 Topology on G
Therefore, ωa (A0 ) = 0 for all a ∈ R and hence ωa ∈ Γ \ Γ0 = ∂Γ for all a ∈ (0, 1),
as required.
(ii) For the statements about Λ, let Ak = X \ Bk and Γ = G \ Λ in part (i). This
completes the proof.
The next result generalises the first part of Theorem 7.10 since Γ is a closed G δ -set
(a closed intersection of countably many open sets).
Corollary 7.11If H ⊂ G is a closed G δ -set, there exists

{Ak } ⊂ L with Ak+1 ⊂ Ak
such that H = k∈N Δ Ak and H◦ = Δ A0 where A0 = k Ak . The boundary ∂H is
uncountable if it is non-empty, equivalently if H ∈
/ τ.
Proof Since (G, τ ) is compact and Hausdorff, it is normal (see Sect. 2.10) and since
H is a closed G δ -set, by Lemma 2.62 there exists a non-negative continuous function
f : G → [0, 1/2] with H = f −1 ({0}). By Theorem 7.4, for some non-negative u ∈
L ∞ (X, L, λ),
f (ω) = u dω for all ω ∈ G.
X
Let Ak = {x ∈ X : u(x) 1/k}, k ∈ N. Then Ak+1 ⊂ Ak and if ω ∈ H,

0 = f (ω) = u dω = u dω + u dω k −1 ω(X \ Ak ) 0,
X Ak X \Ak
and it follows that ω(Ak ) = 1 for all k ∈ N. If ω ∈

/ H let k ∈ N be sufficiently large
that 1/k < f (ω). Then

1 1
0 < f (ω) − u dω − + u dω u∞ ω(X \ Ak )
k Ak k X \Ak
and since ω ∈ G it follows that ω(Ak ) = 0 for all k sufficiently large if ω ∈

/ H. Hence

H = ω ∈ G : ω(Ak ) = 1 for all k ∈ N = Δ Ak . (7.6)
k∈N

Since H = k∈N Δ Ak , by Theorem 7.10(i) H◦ = Δk∈N Ak .
Finally, note that since H is closed its boundary ∂H = H \ H◦ is non-empty if
and only if H is not open. If H is not open it follows that Ak \ Ak+1 ∈
/ N for infinitely
many k, for otherwise, by Lemma 7.1, the collection of sets {Δ Ak } is finite and so
H = k Δ Ak is open. Thus when ∂H is non-empty {Ak } satisfies the hypothesis of
Theorem 7.10(i)(d) and ∂H is uncountable.
7.4 G and the Weak* Topology on L ∗∞ (X, L, λ) 63
7.4 G and the Weak* Topology on L ∗∞ (X, L, λ)
Recall from Definition 2.47 that a set U ⊂ L ∞ (X, L, λ)∗ is open in the weak* topol-
ogy if and only if for every f 0 ∈ U there exists u 1 , · · · , u n ∈ L ∞ (X, L, λ) and > 0
such that
n
f ∈ L ∞ (X, L, λ)∗ : |( f − f 0 )(u j )| < ⊂ U.
j=1
By analogy say W ⊂ L ∗∞ (X, L, λ) is weak* open in L ∗∞ (X, L, λ) if and only if

the corresponding set of elements of L ∞ (X, L, λ)∗ , defined by Theorem 3.1, is open
in the weak* topology of L ∞ (X, L, λ)∗ . In other words

W := W0 ,u 0 ,ν0 : 0 > 0, u 0 ∈ L ∞ (X, L, λ), ν0 ∈ L ∗∞ (X, L, λ) , (7.7a)
where

W0 ,u 0 ,ν0 = ν ∈ L ∗∞ (X, L, λ) :
u 0 dν − u 0 dν0
< 0 , (7.7b)
X X
is a sub-base for the weak* topology on L ∗∞ (X, L, λ).
Lemma 7.12 The topology τ on G coincides with the restriction to G of the weak*
topology on L ∗∞ (X, L, λ).
Proof It suffices to show that when ω0 ∈ Δ A0 ∈ T (Definition 7.2),

there exists
W0 ∈ W (see (7.7)) with ω0 ∈ W0 G ⊂ Δ A0 , and when ω0 ∈ W0 G, W0 ∈ W,
there exists A0 ∈ L with ω0 ∈ Δ A0 ⊂ W0 G.
First suppose ω0 ∈ Δ A0 and let

1
W0 = ν ∈ L ∗∞ (X, L, λ) :
χ A0 dω0 − χ A0 dν
< ∈ W.
X X 2

Then ω ∈ W0 G implies that |ω 0 (A0 ) − ω(A0 )| < 2 . Hence ω(A0 ) = 1 since
1
ω0 (A0 ) = 1, and therefore ω0 ∈ W0 G ⊂ Δ A0 .

Now let ω0 ∈ G and, for any u 0 ∈ L ∞ (X, L, λ) and 0 > 0, let

W0 = ν ∈ L ∗∞ (X, L, λ) :
u 0 dν − u 0 dω0
< 0 ∈ W.
X X

Let α0 = X u 0 dω0 and put A0 = {x ∈ X : |u 0 (x) − α0 | < 0 /2}. Then ω0 ∈
Δ A0 by Theorem 6.2 and {x ∈ X : |u 0 (x) − α̂| < 0 /2} A0 ∈ N if |α̂ − α0 | 0 .
Therefore ω({x ∈ X : |u 0 (x) − α̂| < 0 /2}) =
0, and consequently

X u 0 dω =
α0 | 0 . It follows that
X u 0 dω − α0
< 0 if ω ∈ Δ A0 .
α̂, if ω ∈ Δ A0 and |α̂ −
Hence ω0 ∈ Δ A0 ⊂ W0 G, as required.
64 7 Topology on G
Lemma 7.13 G is closed in the weak* topology of L ∗∞ (X, L, λ).
Proof It suffices to show that L ∗∞ (X, L, λ) \ G is open. If ν0 ∈ L ∗∞ (X, L, λ) \ G

there exists A ∈ L with ν0 (A) = a ∈/ {0, 1}. Let 2 = min{|a|, |1 − a|} > 0 and sup-
pose ν ∈ L ∗∞ (X, L, λ) is in the weak* open neighbourhood of ν0 given by

V = ν ∈ L ∞ (X, L, λ) :
χ A dν −
χ A dν0
< .
X X
Then (see (6.2d)) ν ∈ V implies that |ν(A) − ν0 (A)| < , which means ν(A) ∈ /
/ G. Therefore L ∗∞ (X, L, λ) \ G is weak* open, and G is weak*
{0, 1}, and hence ν ∈
closed.
It follows from Theorem 2.48 (Alaoglu) that (G, τ ) is compact. For a direct proof,
see Theorem 7.3.
7.5 G as Extreme Points
By Definition 2.52 f is an extreme point of B ∗ , the closed unit ball in L ∞ (X, L, λ)∗ ,
if and only if for f 1 , f 2 ∈ B ∗ and α ∈ (0, 1)
f (u) = α f 1 (u) + (1 − α) f 2 (u) for all u ∈ L ∞ (X, L, λ)
implies that f = f 1 = f 2 . Let
ν, ν1 , ν2 ∈ U ∗ = {ν ∈ L ∗∞ (X, L, λ) : |ν|(X ) 1}
be defined in terms of f, f 1 , f 2 ∈ B ∗ by Theorem 3.1. Then a necessary and sufficient

condition for f to be an extreme point of B ∗ is that for α ∈ (0, 1)

u dν = α u dν1 + (1 − α) u dν2 for all u ∈ L ∞ (X, L, λ) (7.8a)
X X X
implies that ν = ν1 = ν2 . Since simple functions are dense in L ∞ (X, L, λ), it is

immediate that (7.8a) holds if and only if
ν(E) = αν1 (E) + (1 − α)ν2 (E) for all E ∈ L. (7.8b)
It is obvious that f is on the boundary of B ∗ , and so f = |ν|(X ) = 1, if f is

extreme. But more can be said.
Lemma 7.14 Let f ∈ L ∞ (X, L, λ)∗ and ν ∈ L ∗∞ (X, L, λ) be related by (3.1b) in

Theorem 3.1. Then f is an extreme point of B ∗ if and only if ν or −ν is in G.
7.5 G as Extreme Points 65
Proof First suppose f is an extreme point of B ∗ . Then f = |ν|(X ) = 1. If ν is not

one-signed then |ν| = ν + + ν − where ν + ∧ ν − = 0 and ν ± (X ) ∈ (0, 1). Let 0 <
0 = 21 min{ν + (X ), ν − (X )} and, by (4.1c), choose A ∈ L such that ν + (X \ A) +
ν − (A) = < 0 . If ν(A) = 0 then ν + (X ) = ν + (X \ A) + ν + (A) = ν + (X \ A) +
ν − (A) = < 0 , which is false. So ν(A) = 0 and hence |ν|(A) > 0. If |ν|(A) = 1
then
ν − (X ) = 1 − ν + (X ) = ν − (A) + ν + (A) − ν + (X )
= ν − (A) − ν + (X \ A) ν − (A) + ν + (X \ A) = < 0
which is false. So |ν|(A) ∈ (0, 1). Let

ν(A E) ν((X \ A) E)
ν1 (E) = , ν2 (E) = for all E ∈ L.
|ν|(A) |ν|(X \ A)
Then ν1 , ν2 ∈ U ∗ and, for all E ∈ L,
ν(E) = αν1 (E) + (1 − α)ν2 (E), where α = |ν|(A), (1 − α) = |ν|(X \ A).
Since α ∈ (0, 1), ν1 (A) = ν(A)/|ν|(A) = 0 and ν2 (A) = 0, this shows that f is not
an extreme element of B ∗ if ν is not one-signed.
So suppose ν is one-signed, say 0 ν ∈ U ∗ (for ν 0 replace ν with −ν), but
ν∈/ G. Then there exists A ∈ L with ν(A) ∈ (0, 1). Let

ν(A E) ν((X \ A) E)
ν1 (E) = , ν2 (E) = for all E ∈ L.
ν(A) ν(X \ A)
Then ν1 , ν2 ∈ U ∗ , ν(E) = αν1 (E) + (1 − α)ν2 (E) for all E ∈ L, where α =

ν(A), (1 − α) = ν(X \ A). Since ν1 (A) = 1 = 0 = ν2 (A), ν is not extreme in U ∗ .
Hence ±ν ∈ G if f is an extreme point of B ∗ .
Now to show that f is an extreme point of B ∗ if ν ∈ G, suppose that ν1 , ν2 ∈ U ∗
and for all E ∈ L,
ν(E) = αν1 (E) + (1 − α)ν2 (E), α ∈ (0, 1), ν1 , ν2 ∈ U ∗ .
Then ν 0 and if ν(E) = 1,
1 = ν(E) = αν1 (E) + (1 − α)ν2 (E) α|ν1 |(X ) + (1 − α)|ν2 |(X ) 1
which implies that ν1 (E) = ν2 (E) = ν(E) = 1. In particular ν1 (X ) = ν2 (X ) = 1. If

ν(E) = 0 then ν(X \ E) = 1 and so ν1 (X \ E) = ν2 (X \ E) = 1, whence ν1 (E) =
ν2 (E) = ν(E) = 0. Thus ν = ν1 = ν2 which shows that f is an extreme point of
B ∗ if ν ∈ G.
66 7 Topology on G
Remark 7.15 Because of Theorem 2.55 (Rainwater), Lemma 7.14 implies that G has
property (W) in the Introduction. In Sect. 8, this conclusion is derived independently,
from the isometric isomorphism in Theorem 7.4.
Let (X, ) be a Hausdorff space, B its Borel σ-algebra and let D denote the set of
regular Borel measures (Definition 2.12) δ that take only values 0 and 1 on B with
δ(X ) = 1.
Lemma 7.16 For δ ∈ D there exists a unique x0 ∈ X such that δ = δx0 , the Dirac
measure introduced in Sect. 2.4.
Proof Since δ ∈ D is regular, there is a compact K ⊂ X with δ(K ) = 1. Let K 0

denote the intersection of all such compact K and note that K 0 = ∅ because these
compact sets have the finite intersection property. If x0 , y0 ∈ K 0 are distinct, then at
least one of {x0 }, {y0 } has zero measure. If δ({y0 }) = 0, by the regularity of δ there is
an open set G y0 with y0 ∈ G y0 and δ(G y0 ) = 0. So δ(K \ G y0 ) = 1 for all compact
K with δ(K ) = 1. This contradicts the fact that y0 ∈ K 0 . Hence K 0 is a singleton,
{x0 } say. If δ({x0 }) = 0, the same argument implies that x0 ∈ / K 0 . So K 0 = {x0 } and
δ = δx0 .
Chapter 8
Weak Convergence in L ∞ (X, L, λ)
In this chapter, theory so far developed yields a necessary and sufficient condition in
terms of behaviour λ-almost everywhere for a bounded sequence that is pointwise
convergent to be weakly convergent in L ∞ (X, L, λ). The resulting test is illustrated
by examples.
8.1 Weakly Convergent Sequences
Since, by Sect. 7.2, L ∞ (X, L, λ) and C(G, τ ) are isometrically isomorphic, their
duals are isometrically isomorphic and u k u 0 in L ∞ (X, L, λ) if and only if
L[u k ] L[u 0 ] in C(G, τ ). Since (G, τ ) is a compact Hausdorff space, by (V) in the
Introduction L[u k ] L[u 0 ] in C(G, τ ) if and only if {L[u k ]C(G,τ ) } is bounded
and L[u k ] → L[u 0 ] pointwise on G. In other words, u k u 0 in L ∞ (X, L, λ) if and
only if for some M

u k ∞ M and u k dω → u 0 dω as k → ∞ for all ω ∈ G. (8.1)
X X
Thus the collection G of finitely additive measures plays a rôle for weak convergence
in L ∞ (X, L, λ) analogous to that of σ-additive Dirac measures for weak convergence
in spaces of continuous functions. The sequential weak continuity of composition
operators is an immediate consequence, in contrast to what happens for weak conver-
gence in L p (X, L, λ), 1 p < ∞. (When u k (x) = sin(kx) in L p (0, 2π), u k 0
but |u k | 0.)
Theorem 8.1 If u k u 0 in L ∞ (X, L, λ) as k → ∞ and F : R → R is continuous,
then F(u k ) F(u 0 ) in L ∞ (X, L, λ).
Proof This is immediate from Theorem 6.2(c) and (8.1).

https://doi.org/10.1007/978-3-030-34732-1_8
68 8 Weak Convergence in L ∞ (X, L, λ)
Corollary 8.2 In L ∞ (X, L, λ), u k 0 if and only if |u k | 0.

Proof “Only if” follows from Theorem 8.1 and “if” from (6.10a) and (8.1).
Recall from Remark 2.9 that if A is an atom, λ(A) ∈ (0, ∞) and u ∈ L ∞ (X, L, λ)
is constant on A. Recall also from Theorem 5.9 that ω ∈ G is either purely finitely
additive or, for all E ∈ L, ω(E) is a constant multiple of λ(E ω E) where E ω is an
atom. Hence from (8.1) if {u k } is bounded in L ∞ (X, L, λ), u k u in L ∞ (X, L, λ)
if and only if
u k (A) → u(A) for all atoms A ∈ L, (8.2a)
and

u k dμ → u dμ for all purely finitely additive μ ∈ G. (8.2b)
X X
Of course many familiar measure spaces have no atoms (Example 2.15), but when
they do (8.2a) is equivalent to

χ A u k dλ → χ A u dλ for all atoms A.
X X
∗
Hence for bounded sequences (8.2a) holds if u k u, and in particular if u k (x) →
u(x) for λ-almost all x ∈ X .
Since (X, L, λ) is σ-finite, L ∞ (X, L, λ) is the dual of L 1 (X, L, λ) [8, Thm.
4.4.1] and hence by Theorem 2.48 (Alaoglu) the closed unit ball in L ∞ (X, L, λ)
is compact in the weak* topology of L 1 (X, L, λ)∗ (Definition 2.47). If in addition
L 1 (X, L, λ) is a separable metric space, by Remark 2.51 the closed unit ball in
L ∞ (X, L, λ) with the weak* topology is metrisable, and hence weak* sequentially
compact (Definition 2.49). Thus a bounded sequence in L ∞ (X, L, λ) has a weak*
convergent subsequence which therefore satisfies (8.2a). (For example (X, L, λ) is
σ-finite and L 1 (X, L, λ) is separable when X ⊂ Rn and λ is any Borel measure on
Rn which is finite on bounded sets [8, Cor. 4.2.2].)
In circumstances such as these, deciding whether a bounded sequence is weakly
convergent or has a weakly convergent subsequence is mainly a question of deciding
whether (8.2b) is satisfied. Theorem 8.7 settles this issue by characterising sequences
that converge weakly to 0 in terms of their λ-almost everywhere pointwise behaviour.
To set the scene, recall some necessary conditions for the weak convergence of
sequences in L ∞ (X, L, λ).
8.2 Pointwise Characterisation
Lemma 8.3 When u k u 0 in L ∞ (X, L, λ), there is a subsequence with u k j (x) →

u 0 (x) for λ-almost all x ∈ X .
8.2 Pointwise Characterisation 69
Proof Since (X, L, λ) is σ-finite there exists f ∈ L 1 (X, L, λ) which is positive λ-

almost everywhere and, by Corollary 8.2, |u k − u 0 | f → 0 in L 1 (X, L, λ). Hence a
subsequence u k j (x) → u 0 (x) for λ-almost all x.
In a metric space, let B(x, r ) denote the ball with centre x and radius r .
Lemma 8.4 Suppose (X, ρ) is a metric space and for u ∈ L ∞ (X, B, λ), there exists
a set E(u) ∈ L such that λ(X \ E(u)) = 0,

1
u(x) := lim u dλ exists for all x ∈ E(u), (8.3)
0<r →0 λ(B(x, r )) B(x,r )
and u(x) = u(x), x ∈ E(u). Then u k u 0 in L ∞ (X, B, λ) implies that u k (x) →

u 0 (x) for λ-almost all x ∈ X .
Remark 8.5 Hypothesis (8.3) means that λ-almost all x ∈ X is a Lebesgue point of
u [28, 7.6 & Th. 7.7]. According to Heinonen [18, Thm. 1.8 & Remark 1.13] this is
not very restrictive. For example, it holds when λ is a regular Borel measure on B
which is doubling on (X, ρ) (i.e. balls have finite positive measure and there exists a
constant C such that λ(B(x, 2r )) Cλ(B(x, r )), or on Rn with the standard metric
when λ is a Radon measure, finite on compact sets and positive on balls.

Proof Let E = ∞ 0 E(u k ), which has full measure, and let V denote the linear span
in L ∞ (X, B, λ) of {u k : k ∈ N0 }. Now for fixed x ∈ E define a linear functional x
on V by
x (v) = v(x), v ∈ V.
Since |x (v)| v∞ , by the Hahn–Banach theorem there exists L x ∈ L ∞ (X, B, λ)∗
with L x (v) = x (v) for all v ∈ V . Therefore u k u 0 implies that u k (x) = x (u k ) =
L x (u k ) → L x (u 0 ) = u 0 (x) for all x ∈ E, as required.
Example 8.6 To see that the converse is false let X = (0, 1) and, for k ∈ N, define
u k on (0, 1) by u k (x) = 0, x ∈ [1/2k, 1); u k (x) = 1, x ∈ (0, 1/4k]; u k linear on
[1/4k, 1/2k]. Then {u k ∞ } is bounded, u k is continuous on X and u k (x) → 0
monotonically for all x ∈ X as k → ∞. But by Theorem 5.6 there exists ω ∈ G with
ω(E ) = 1, where E = (0, 1/4) for each . Therefore, by Theorem 6.2, X u k dω =
1 for all k and hence u k 0 in L ∞ (X, L, λ). Hence u k is not weakly convergent to
0 (indeed not weakly convergent to any limit) in L ∞ (X, L, λ) .
If {k j } ⊂ N is strictly increasing and u k 0 in L ∞ (X, L, λ), by Corollary 2.46

of Mazur’s theorem and Corollary 8.2 there exists {u i } in the convex hull of {|u k j | :
j ∈ N} with

mi
mi
u i ∞ → 0 as i → ∞, u i = γ ij |u k j |, γ ij ∈ [0, 1] and γ ij = 1, m i ∈ N.
j=1 j=1
Since γ ij may be zero, there is no loss in assuming that {m i } is increasing. Therefore,

for any strictly increasing sequence {k j },

0 wi (x) := min |u k j (x)| : j ∈ {1, · · · , m i } u i (x), x ∈ X,
defines a non-increasing sequence of non-negative functions in L ∞ (X, L, λ) with

wi ∞ → 0. It follows that if u k 0 and {k j } ⊂ N is any strictly increasing
sequence, the corresponding sequence defined on X by

0 v J (x) := min |u k j (x)| : j ∈ {1, · · · , J } (8.4)
is pointwise non-increasing in L ∞ (X, L, λ) and v J ∞ → 0 as J → ∞. To show

that this condition is sufficient as well as necessary for a sequence to be weakly
convergent to 0 in L ∞ (X, L, λ), let

Aα (u) = x ∈ X : |u(x)| > α , u ∈ L ∞ (X, L, λ), α > 0.
The proof is independent of Mazur’s theorem.

Theorem 8.7 In L ∞ (X, L, λ) bounded sequence {u k } converges weakly to zero if
and only if for every α > 0 and every strictly increasing sequence {k j } in N there
exists J ∈ N with
J
λ Aα (u k j ) = 0. (8.5)
j=1
This criterion is equivalent to saying that for all strictly increasing sequences {k j },
the sequence {v J } in (8.4) converges strongly to zero in L ∞ (X, L, λ).
Remark 8.8 In the Borel setting of the next chapter, this criterion for weak conver-
gence in L ∞ (X, B, λ) will be localised to arbitrary open neighbourhoods of points
in the one-point compactification of X .
Proof Suppose, for a strictly increasing sequence {k j } and α > 0, that (8.5) is false
for all J ∈ N. Then E = {Aα (u k j ) : j ∈ N} satisfies the hypothesis of Theorem 5.6
and hence there exists ω ∈ G such that ω(Aα (u k j )) = 1 for all j. It follows that

|u k j | dω |u k j | dω α > 0 for all j.
X Aα (u k j )
Hence |u k | 0 by (8.1) and so, by Corollary 8.2, u k 0.

Conversely, suppose u k 0. Then by (8.1) and Corollary 8.2 there exists α > 0,
a strictly increasing sequence {k j } ⊂ N and ω ∈ G such that

|u k j | dω =: α j > α > 0 for all j ∈ N. (8.6)
X
8.2 Pointwise Characterisation 71
Now α j − α > 0 for all j and

x : |u k j (x)| − α j < α j − α ⊂ x : |u k j (x)| > α = Aα (u k j ).
From (8.6) and Theorem 6.2 it follows that

1=ω x : |u k j (x)| − α j < α j − α ω Aα (u k j ) ,
J
and so ω(Aα (u k j )) = 1 for all j. Hence ω j=1 Aα (u k j ) = 1, for all J , since ω ∈
G. Since ω is zero on N because ω ∈ G, it follows that (8.5) is false for all J . Finally,
note that for a strictly increasing sequence {k j } and α > 0,

λ x : v J (x) > α =λ x : |u k j (x)| > α for all j ∈ {1, · · · , J }

J
=λ Aα (u k j ) .
j=1
Since v J (x) v J +1 (x) 0 it follows that v J → 0 in L ∞ (X, L, λ) if and only if,

for every α > 0, (8.5) holds for some J . This completes the proof.
Dini’s theorem [27, Thm. 7.13] says that on compact spaces monotone pointwise
convergence of a sequence of continuous functions to a continuous function is uni-
form; equivalently, by (V) in the Introduction, weak and strong convergence coin-
cide for bounded monotone sequences of continuous functions on compact spaces.
An analogue in L ∞ (X, L, λ) now follows either from Dini’s theorem in the light of
Theorem 7.4, or directly from Theorem 8.7. The sequence in Example 8.6 converges,
pointwise everywhere and monotonically, to zero, but not weakly in L ∞ (X, L, λ).
Corollary 8.9 Suppose {u k } is bounded in L ∞ (X, L, λ) and |u k (x)| |u k+1 (x)|,

k ∈ N, for λ-almost all x ∈ X . Then u k 0 if and only if u k → 0 in L ∞ (X, L, λ).
Proof Suppose u k 0 in L ∞ (X, L, λ). Since the monotonicity in k of {|u k (x)|}

implies that v J = |u J |, by Theorem 8.7, u J → 0 in L ∞ (X, L, λ) as J → ∞. The
converse is obvious.

Remark 8.10 Let u k = χ Ak where {Ak } ⊂ L and λ Ai A j = 0, i = j. Then,
defined in terms of any increasing sequence {k j }, v J = 0 for all J 2. Hence
u k 0 as k → ∞ in L ∞ (X, L, λ) by Theorem 8.7. This is a new perspective on
Corollary 3.2.
With ∞ (N) as in Example 5.8 and Sect. 6.5, for every j ∈ N a bounded linear func-
tional δ j is defined on ∞ (N) by δ j (u) = u( j), u ∈ ∞ (N). Hence u k 0 in ∞ (N)
implies that u k ( j) → 0 as k → ∞ for each j ∈ N. However, pointwise convergence
does not imply weak convergence.
Corollary 8.11 Let {u k } be a sequence in ∞ (N) with u k (i) → 0 as k → ∞ for

each i ∈ N. Then u k 0 if and only if there exist α > 0 and a strictly increasing
sequence {k j } for which there exist strictly increasing sequences {i n }, {Jn } such that
for all n ∈ N
|u k j (i n )| > α for all j ∈ {1, 2 · · · · · · , Jn }. (8.7)
Proof Suppose u k 0. Then by Theorem 8.7 there exists α > 0 and a strictly
increasing sequence {k j } such that for all J ∈ N

v J ∞ > α where v J (i) = min |u k j (i)| : j ∈ {1, · · · , J } for i ∈ N. (8.8)
Let J1 = 1 and let i 1 denote the smallest i such that |u k1 (i)| > α. Since u k j converges
pointwise to 0 as j → ∞, there exists a smallest J2 > J1 such that |u k j (i 1 )| α for
all j J2 . Therefore by (8.8) there exists a smallest i > i 1 , denoted by i 2 , such that

min |u k j (i 2 )| : j ∈ {1, · · · , J2 } > α.
Again there exists a smallest J3 > J2 such that |u k j (i 2 )| α for all j J3 . Let i 3
denote the smallest i > i 2 such that

min |u k j (i 3 )| : j ∈ {1, · · · , J3 } > α.
By induction, for the strictly increasing sequence {k j } satisfying (8.8), this process
yields strictly increasing sequences {i n } and {Jn } with the required properties.
Conversely, if (8.7) holds it is immediate that v J ∞ → 0 as J → ∞ and hence
u k 0 as k → ∞ by Theorem 8.7. This completes the proof.
Example 8.12 A sequence {u j } in ∞ (N) with

u i j → 0 as j → ∞, i < j
u j (i) = , i ∈ N.
α > 0, i j
is pointwise convergent to 0, but not weakly convergent in ∞ (N) by Corollary 8.11

with i n = Jn = n.
8.3 Applications of Theorem 8.7
Example 8.13 In Example 2.29 u k → 0 strongly in L p (0, 1) for all finite p, and
therefore {u k } has a subsequence (see Sect. 2.27) which converges almost everywhere
to 0, yet lim supk→∞ u k (x) = 1 for all x ∈ (0, 1). To see that u k 0 in L ∞ (0, 1) let
j ( j + 1) 1
kj = 1 + and put Ik j = 0, .
2 j +1
8.3 Applications of Theorem 8.7 73
Then 1
u k j (x) 1, x ∈ 0, when j ∈ {1, 2, · · · , J },

J +1
and the conclusion follows from Theorem 8.7. However

2 j − 1 2 j+1 − 1
u k j = χ Ik where Ik j = , ,
j 2j 2 j+1
defines a subsequence for which u k j 0 by Remark 8.10, since {Ik j } is a sequence
of disjoint intervals.
Example 8.14 (Step Functions) In this example X = (−1, 1) with Lebesgue mea-
sure and 1 1
u k = χ Ak where Ak = k+1 , k .
2 2
Let
1
1
u+ (x) = u k x + = χ A+
, where A +
= 0, ,
k
2(k+1) k
k
2(k+1)
1
1 3
u−
k (x) = u k x − (k+1) = χ A− , where Ak =
−
k
, (k+1) .
2 k 2 2
So u k = u ± −
k = 1, and u k 0 and u k 0 by Remark 8.10 since {Ak } and {Ak }
−
are sequences of disjoint measurable sets.

However v J , defined in (8.4) by {u +
k }, is 1 on (0, 2
−(J +1)
) for all J ∈ N, and hence
+
u k 0 in L ∞ (X, L, λ) by Theorem 8.7. The key observation is that χ Ak 0 for
any family {Ak } of disjoint sets in L. For another illustration, see Example 9.9.
∞
Example 8.15 (Simple Functions) In L ∞ (X, L, λ) let u k = i=1 αi χ Ai , where
k
∞
Σi=1 |αi | < ∞ and, for each i ∈ N, {Aik }k∈N is a family of mutually disjoint measur-
able sets. Then u k 0.
To see this, note that for each x ∈ X and i ∈ N there exists at most one k ∈ N,
denoted, if it exists, by κ(x, i), such that x ∈ Aik if and only if k = κ(x, i). Note
also that for > 0 there exists I ∈ N such that Σ I∞ +1
|αi | < . Hence, for any given
k ∈ N and x ∈ X ,

I
|u k (x)| |αi |χ Aik (x) + = |αi | + .
i=1 i∈{1,··· ,I }
κ(x,i)=k
Since {κ(x, i) : i ∈ {1, · · · , I }} has at most I elements, there exists k ∈ {1, · · · , I +

1} such that k = κ(x, i) for any i ∈ {1, · · · , I }. Consequently, inf{|u k (x)| : 1 k
I + 1} , independent of x ∈ X . Since this argument can be repeated with k ∈ N
replaced by any strictly increasing subsequence {k j }, {v J } defined in terms of that
subsequence by (8.4) has v J ∞ → 0 in L ∞ (X, L, λ), and u k 0 follows.
Example 8.16 (Translations) Let u : R → R be essentially bounded and measur-

able with |u(x)| → 0 as |x| → ∞ and let u k (x) = u(x + k). Then u k 0 in
L ∞ (R, L, λ) where λ is Lebesgue measure on R. To see this, for > 0 suppose
that |u(x)| < if |x| > K . Then for any strictly increasing {k j } ⊂ N, v J ∞ <
for all J 2K where {v J } is defined in terms of {u k j } by (8.4), and the result follows.
For u : R → R essentially bounded and measurable with u(x) → 0 as x → ∞
and u(x) → 1 as x → −∞, let u k (x) = u(x + k). Then u k (x) → 0 as k → ∞ for
all x ∈ R, but u k is not weakly convergent to 0 because of Theorem 8.7. However,
in the notation of Definition 9.6, u k 0 at every point of R but not, as will be seen
in Chap. 9, at the point at infinity in its one-point compactification.
Example 8.17 (Oscillatory Functions) With X = (0, 2π) and Lebesgue measure, let
u k (x) = sin(1/(kx)). Clearly |u k (x)| → 0 as k → ∞ uniformly on ( , 2π) for any
∈ (0, 2π). Therefore, if a subsequence {u k j } is weakly convergent, its weak limit
is zero.
To see that no subsequence of {u k } is weakly convergent to 0, consider first a
strictly increasing sequence {k j } of natural numbers for which there exists a prime
power p m which does not divide k j for all j. Then, for J ∈ N sufficiently large, let
−1
π
xJ = lcm {k1 , · · · , k J } ∈ (0, 2π),
pm
where lcm denotes the least common multiple. Then, for j ∈ {1, · · · , J }, since p m
k j and p is prime,
1 lcm {k1 , · · · , k J }
= π where
k j xJ pm k j
lcm {k1 , · · · , k J }
= r mod p m , r ∈ {1, · · · , p m − 1}.
kj
It follows that |u k j (x J )| | sin(π/ p m )| > 0, independent of j ∈ {1, · · · , J } and,

since u k j is continuous at x J , that v J L ∞ (X,L,λ) | sin(π/ p m )| > 0 for all J suf-
ficiently large. By Theorem 8.7, this shows that u k j 0 if {k j } has a subsequence
{k j } for which p m k j for all j ∈ N.
Note that if {k j } has no such subsequence for any prime p and m ∈ N, then every
K ∈ N is a divisor of k j for all j sufficiently large, how large depending on K .
Consequently, if u k j 0, {k j } has subsequence {k j } such that

j
2k j divides k j+1 and k j+1 2 ki for all j ∈ N.
i=1
Now for J ∈ N let

2
−1
J
xJ = k .
π i=1 i
8.3 Applications of Theorem 8.7 75
Then the properties of {k j } imply that for j ∈ {1, 2, · · · , J },

+ k j +
j−1
1
J
i=1 ki π
J
i= j+1 ki i=1 ki π
= =
k j xJ k j 2 k j 2

π
1
= 2N + 1 + z where N ∈ N and |z| < .
2 2
Hence π
π
1
|u k j (x J )| ∈ sin , sin = √ , 1 , j ∈ {1, · · · , J }.
4 2 2
It follows from Theorem 8.7 that u k j 0 as j → ∞ since it has a subsequence {u k }

√ j
which generates a sequence {v J } with v J ∞ 1/ 2 for all J .

Chapter 9
L ∗∞ When X is a Topological Space
This chapter considers what more can be said when (X, ) is a locally compact
Hausdorff space, with B the corresponding Borel σ-algebra and λ a measure on B as
described in Chap. 3. In addition, here λ is assumed regular, finite on compact sets
and positive on open sets. Recall from Lemma 7.16 that a regular Borel measure that
takes only values 0 or 1 is a Dirac measure concentrated at a point x0 ∈ X . Now it
will be shown that G can be partitioned into a union of disjoint closed subsets G(x0 ),
x0 ∈ X ∞ , where X ∞ is the one-point compactification of X . Because of (8.1), this
yields a notion of weak convergence at points of X ∞ and leads to a related idea, the
essential range at x0 of u ∈ L ∞ (X, B, λ) (sometimes called the set of cluster values
of u at x0 ) which is parameterized by elements of G(x0 ). The pointwise criterion for
weak convergence in the previous chapter can then be localised and related to the
pointwise essential range.
The distinction between the essential range at x0 of u ∈ L ∞ (X, B, λ) and its
actual range can be rather striking. When Ω ⊂ Rn with positive Lebesgue measure,
exists u ∈ L ∞ (Ω, B, λ) for which the range is denumerable, {u(x) : x ∈ Ω} =
there
Q [0, 1], while at every x0 ∈ Ω the essential range of u is the continuum [0, 1].
The essential range of u ∈ ∞ (N) at the point at infinity in N∞ is the set of limits of
convergent subsequences {u(k j )} of {u(k)}.
9.1 Localising G
When (X, ) is a locally compact Hausdorff space, by (5.1),

G = ω ∈ L ∗∞ (X, B, λ) : ω(X ) = 1, ω(A) ∈ {0, 1}, A ∈ B .
Lemma 9.1 Let (X, ) be locally compact and Hausdorff.

(a) For ω ∈ G there exists a compact set K with ω(K ) = 1 if and only if there exists
x0 ∈ K such that
https://doi.org/10.1007/978-3-030-34732-1_9
78 9 L ∗∞ When X is a Topological Space
ω(G) = 1 for all open sets G with x0 ∈ G. (9.1a)
For all ω ∈ G there is at most one x0 ∈ X satisfying (9.1a) and when (X, ) is
compact there is exactly one such x0 .
(b) For x0 ∈ X at least one ω ∈ G satisfies (9.1a). If (X, ) is not compact, at
least one ω ∈ G satisfies,
ω(X \ K ) = 1 for all compact K ⊂ X. (9.1b)
Proof (a) Suppose that ω(K ) = 1 for some compact K and (9.1a) is false. Then for
each x∈ K there is an open set G x with x ∈ G x and ω(G x ) = 0. Since K is compact,
K
K ⊂ i=1 G xi . Since ω(G xi ) = 0, 1 i K and ω is finitely additive, ω(K ) = 0
which is false. Hence if ω(K ) = 1 and K is compact, there exists x0 ∈ K satisfying
(9.1a). Suppose there is another x1 ∈ X satisfying (9.1a). Since X is Hausdorff, there
exist open sets with x0 ∈ G x0 , x1 ∈ G x1 and G x0 G x1 = ∅. But
this is impossible
because, by finite additivity, it would imply that ω G x0 G x1 = 2.
Now suppose ω ∈ G and ω(K ) = 0 for all compact sets K . By local compactness,
for any x ∈ X there is an open set G x with x ∈ G x and its closure G x is compact.
Hence, since ω(G x ) ω(G x ) = 0, there is no x ∈ X with the required property.
Finally, the existence of x0 when X is compact follows because ω(X ) = 1.
(b) For x0 ∈ X , let E(x0 ) = {G ∈ : x0 ∈ G}. Since by hypothesis λ(G) > 0
when G is non-empty and open, by Theorem 5.6 there exists ω ∈ G with ω(G) = 1
for all G ∈ E(x0 ).
Finally if (X, ) is not compact let E∞ = {X \ K : K compact}. Since sets in E∞
are non-empty and open (because compact sets in Hausdorff spaces are closed), the
existence of ω ∈ G follows from Theorem 5.6 as before. This completes the proof.
For ω ∈ G, let ω∞ be defined on Borel subsets E

of X ∞ , the one-point compactifica-
tion of X (Definition 2.32), by ω∞ (E) = ω(E X ). Then ω∞ is the unique finitely
additive measure on X ∞ which takes only values 0 and 1 and which coincides with
ω on Borel sets in X . Lemma 9.1 can then be re-stated as follows.
Corollary 9.2 (a) For ω ∈ G there is a unique x0 ∈ X ∞ such that ω∞ (G) = 1 when
x0 ∈ G and G is open in X ∞ . Moreover x0 = x∞ if and only if ω(K ) = 0 for all
compact K ⊂ X .
(b) For x0 ∈ X ∞ there exists ω ∈ G such that ω(G) = 1 when x0 ∈ G and G is
open in X ∞ .
Definition 9.3 For x0 ∈ X ∞ , let G(x0 ) ⊂ G denote the set of ω ∈ G for which the
conclusion of Corollary 9.2(b) holds, and let U(x0 ) ⊂ U denote the corresponding
family of ultrafilters (Definition 5.3).
Properties of G(x0 ), x0 ∈ X ∞
(a) G(x0 )
= ∅ by Corollary 9.2(b);
(b) G(x1 ) G(x2 ) = ∅ if x1 = x2 ∈ X ∞ , because (X ∞ , ∞ ) is Hausdorff;
9.1 Localising G 79
(c) G(x0 ) is compact because (G, τ ) is compact by Theorem 7.3 and

G(x0 ) = ΔG , where ΔG is closed.
{G:x0 ∈G∈∞ }
Theorem 9.4 Let (X, ) be locally compact and Hausdorff.

(a) Suppose x0 ∈ X and {G k }k∈N is a nested, denumerable local base for at
x0 (i.e. x0 ∈ G k+1 ⊂ G k ∈ for all k and x0 ∈ G ∈ implies G k ⊂ G for some k).
Then
(i) G(x0 ) is a compact G δ -set in (G, τ );
(ii) G(x0 )◦ = ∅ if and only if {x0 } is an atom and G(x0 )◦ = {δx0 }, a Dirac measure;
(iii) G(x0 )◦ = ∅ and G(x0 ) is uncountable if {x0 } is not an atom.

(b) Suppose (X, ) is not compact and {x∞ } (X \ K k ) k∈N is a nested, denu-
merable local base for ∞ at x∞ (i.e. K k is compact, K k ⊂ K k+1 and for every
compact K there exists k ∈ N with K ⊂ K k ). Then
(i) G(x∞ ) is a closed G δ -set in (G, τ ),
(ii) G(x∞ )◦ = ∅ and G(x∞ ) is uncountable.

Proof (a) (i) For x0 ∈ X it is immediate that G(x0 ) = k ΔG k and since ΔG k is both
open and closed, G(x0 ) is a closed G δ -set in (G, τ ), and (G, τ ) is compact.
(ii) Since (X, ) is Hausdorff, ∩k∈N G k = {x0 } and by Theorem 7.10 the interior
G(x0 )◦ = Δ{x0 } . Hence G(x0 )◦ = ∅ if and only if λ({x0 }) > 0, i.e. {x0 } is an atom
and by Lemma 7.7 G(x0 )◦ is the singleton Δ{x0 } .
(iii) Since x0 ∈ G k , by hypothesis λ(G k ) > 0 because G k is non-empty and open,
and λ(G k ) → 0, since λ is σ-additive and λ({x0 }) = 0. That G(x0 ) is uncountable
and G(x0 )◦ = ∅ now follows from Theorem 7.10.
(b) Since a singleton
is compact, the hypothesis implies that X = k∈N K k , and
since G(x∞ ) = k Δ(X \Kk ) the results follow from Theorem 7.10 as in part (a).

The hypothesis of part (b) implies that X = k K k ; in other words, (X, ) is
σ-compact (Definition 2.31). However, (X, ) being σ-compact does not imply that
the hypotheses of part (b) are satisfied.
Remark 9.5 Given the one-to-one correspondence between the set G of 0–1 finitely
additive measures and ultrafilters U in (X, B, λ) in Theorem 5.4, there is an obvious
similarity with Lemma 7.16, but there are important differences. By Lemma 7.16
there is a one-to-one correspondence between points x0 ∈ X and Dirac measures
δx0 ∈ D, and no Dirac measure is concentrated at infinity. By contrast, there is a one-
to-one correspondence between points x0 ∈ X ∞ and the disjoint sets G(x0 ) ⊂ G
which by Theorem 9.4(iii) may be uncountably infinite.
9.2 Localising Weak Convergence
In the notation of Definition 9.3

G= G(x0 ), U= U(x0 ), (9.2)
x0 ∈X ∞ x0 ∈X ∞
which leads to the following definition of pointwise weak convergence.
Definition 9.6 (Weak Convergence at a Point) A bounded sequence {u k } in

L ∞ (X, B, λ) converges weakly to u at x0 ∈ X ∞ (written as u k u at x0 ) if and
only if
u k dω → u dω as k → ∞ for all ω ∈ G(x0 ).
X X
Equivalently, for all open sets G with x0 ∈ G,

u k dω → u dω for all ω ∈ G(x0 ).
G G
Theorem 9.7 A bounded sequence converges weakly to u in L ∞ (X, B, λ) if and

only if it converges weakly to u at every x0 ∈ X ∞ .
Proof This is immediate from (8.1), (9.2) and Definition 9.6.
For u ∈ L ∞ (X, B, λ), α > 0 and G open in X ∞ , let

Aα (u|G ) = x ∈ G X : |u(x)| > α .
Theorem 9.8 For bounded {u k } in L ∞ (X, B, λ) the following are equivalent:

(a) u k 0 at x0 ∈ X ∞ ;
(b) there exists α0 > 0 and a strictly increasing sequence {k j } ⊂ N such that for all
open G ⊂ X ∞ with x0 ∈ G and every J ∈ N

J
λ Aα0 (u k j |G ) > 0; (9.3)
j=1
(c) there exists a strictly increasing {k j } such that, for all open G ⊂ X ∞ with x0 ∈ G,
the non-increasing sequence {v J } defined by (8.4) satisfies

v J |G X → 0 in L ∞ (G X, B, λ) as J → ∞.
9.2 Localising Weak Convergence 81
Proof That (b) and (c) are equivalent is immediate from the definitions. If (a) holds
there exists α0 > 0, ω0 ∈ G(x0 ) and a strictly increasing sequence {k j } ⊂ N, such
that for any open G with x0 ∈ G

0 < α0 |u k j | dω0 = |u k j | dω0 for all j.
X G
Now (b) follow as in the proof of Theorem 8.7. If (b) holds then (9.3) implies that

E0 := Aα (u k j |G ) : j ∈ N, G open, x0 ∈ G
satisfies the hypothesis of Theorem 5.6. Hence, there exists ω0 ∈ G with ω0 (A) = 1
for all A ∈ E0 . It follows that ω0 ∈ G(x0 ) and from (9.3),

|u k j | dω0 = |u k j | dω0 α0 > 0 for all j,
G X
and (a) follows.
Example 9.9 Suppose (X, B, λ) has no atoms and (X, ) is completely separable
(i.e. has a denumerable base, G say, see Sect. 2.10). By Theorem 2.22 there exists
{Ak }, a sequence of mutually disjoints sets in B, each of which splits G.
Hence, for all k ∈ N and for every open set G ⊂ X the function χ Ak takes the value
1 on a subset of G with positive measure. However, since the sets Ak are mutually
disjoint it follows from Example 8.14 that χ Ak 0 in L ∞ (X, B, λ), and therefore
χ Ak 0 at every point of X . Note that {χ Ak } does not satisfy (9.3) for any {k j } ⊂ N
j
and x0 ∈ X .
9.3 Fine Structure at x0 of u ∈ L ∞ (X, B, λ)
By analogy with (6.11), in a Borel measure space (X, B, λ) the essential range of u
localised at x0 ∈ X ∞ is defined as

R(u)(x0 ) = α:λ x ∈G X : |α − u(x)| <
> 0 (9.4)
{
>0, G: x0 ∈G∈}

= u dω : ω ∈ G(x0 ) .
X
From the first line of (9.4), the multivalued mapping x0 → R(u)(x0 ) reflects the
measure-theoretic fine structure at x0 of u ∈ L ∞ (X, B, λ) and, since

R(u)(x0 ) = L[u] G(x0 ) , (9.5)
the second line relates that fine structure to the isometric isomorphism L :
L ∞ (X, B, λ) → C(G, τ ) in Sect. 7.2. R(u)(x0 ) is closed because G(x0 ) is compact
and L[u] is continuous.
A point α ∈ R(u)(x0 ) given by α = X u dω, ω ∈ G(x0 ), may be thought of as
the directional limit of u at x0 , the “direction” being determined by the ultrafilter
Uω ∈ U(x0 ) (see (5.3b). Thus, by Theorem 9.7, weak convergence in L ∞ (X, B, λ) is
equivalent to convergence, for each U ∈ U(x0 ), of the directional limits of u k at x0 to
corresponding directional limits of u at x0 , for each x0 ∈ X ∞ . In other words, u k u
implies that for every x0 ∈ X ∞ and every α ∈ R(u)(x0 ) there exist αk ∈ R(u k )(x0 )
such that αk → α as k → ∞. This is not equivalent to αk → α when αk ∈ R(u k )(x0 )
and α ∈ R(u)(x0 ) because of the possibility that

αk = u k dωk and α = u dω, but ωk = ω.
X X
Thus the multivalued R(u)(x0 ) may be interpreted as representing the fine structure at
x0 of u ∈ L ∞ (X, B, λ) which is intimately related to weak convergence. A sufficient
condition for u k u is that for every x0 ∈ X ∞

sup |α| : α ∈ R(u k − u)(x0 ) → 0 as k → ∞.
As already noted, the necessary condition is not sufficient. The following example
shows that the sufficient condition is not necessary.
Example 9.10 With {Ak } in Example 9.9, u k = χ Ak has the property that R(u k )(x) =
{0, 1} for all x ∈ X while u k 0 in L ∞ (X, B, λ) by Corollary 3.2. For a simpler
example, let u k = χ Ak where {Ak } is a sequence of disjoint open segments cen-
in R . Then R(u k )(0) = {0, 1} but u k 0 by
2
tred on the origin 0 of the unit disc
Corollary 3.2. In these examples X u k dω → 0, but not uniformly for ω ∈ G(0).
The next example illustrates the strong distinction between the pointwise value u(x)
and the essential range R(u)(x) at a point x.
Example 9.11 Suppose (X, B, λ) has no atoms and (X, ) is completely separable
with denumerable base G. Let {Ai : i ∈ N} be the disjoint family of measurable
sets,
each of which splits G according to Theorem 2.25. Let q : N → Q (0, 1) be a
bijection and define a measurable function by
∞

q(i) if x ∈ Ai , i ∈ N
u(x) = q(i)χ Ai = .
0 otherwise
i=1
From Theorem 2.25 and (9.4) it follows that

Q (0, 1) ⊂ R(u)(x0 ) ⊂ [0, 1] for all x0 ∈ X.
9.3 Fine Structure at x0 of u ∈ L ∞ (X, B, λ) 83
Since R(u)(x0 ) is closed, R(u)(x0 ) = [0, 1], even though u(x0 ) ∈ Q, for all
x0 ∈ X .
Lemma 9.12 Suppose (X, ) is completely separable and u ∈ L ∞ (X, B, λ). Then
u(x) ∈ R(u)(x) for λ-almost all x ∈ X .
K
Proof Consider first the case when u is simple, say u = k=1 αk χ Ak where the αk
are distinct and Ak = {x : u(x) = αk } ∈ B. Then let

Bk = x ∈ X : u(x) = αk ∈
/ R(u)(x) , k ∈ {1, · · · , K },
K
so that x ∈ X : u(x) ∈
/ R(u)(x) = Bk .
k=1

Let y ∈ Bk . Then u(y) = αk and there exists an open set G with y ∈ G and λ {x ∈
G : u(x) = αk } = 0. Since there is no loss in assuming that G ∈ G where G is
the denumerable base for , Bk is covered bya denumerable family of sets of zero
measure. Hence λ(Bk ) = 0 for all k, and so λ {x : u(x) ∈ / R(u)(x)} = 0.
For the general case, let u ∈ L ∞ (X, B, λ) be arbitrary, let {u k } be a sequence
of simple functions with u k − u∞ → 0 in L ∞ (X, B, λ) and for fixed k let
u k (x)∈ R(u k )(x) for all x ∈ E k where λ(X \ E k ) = 0. Then λ(X \ E) = 0 where
E = k∈N E k ,
u k (x) ∈ R(u k )(x) for all k ∈ N and x ∈ E,
and without loss of generality suppose u k (x) → u(x) for all x ∈ E.
Now for x ∈ E and k ∈ N let ωk ∈ G(x) be such that u k (x) = X u k dωk . Then
for x ∈ E,

u(x) = lim u k (x) = lim u k dωk
k→∞ k→∞ X

= lim (u k − u) dωk + u dωk = lim u dωk ∈ R(u)(x),
k→∞ X X k→∞ X

since X u dωk ∈ R(u)(x) and R(u)(x) is closed. The proof is complete.
Remark 9.13 As in Example 5.8, ∞ (N) = L ∞ (X, B, λ) where X = N, B = ℘ (N)

and λ is counting measure. Then G(k), k ∈ N, has only one element, the σ-additive
Dirac measure δk (Example 2.13). Moreover, {E k = k + N : k ∈ N} is a local base
of the open neighbourhoods of ∞ in N∞ . In this notation, the results of Sect. 6.5 on
integrating u ∈ ∞ (N) are examples of the formula

R(u)(∞) = u dω : ω ∈ G(∞) , u ∈ (N).
N
Thus when u( j) = u j ∈ ∞ , the essential range of u at k ∈ N is the singleton {u k },

while at infinity the essential range is a singleton {α} if and only if u j → α as
j → ∞.
Lorentz [24] proved (see [31] for an elementary proof) that a sequence {xk } ∈ ∞
has the property that all its Banach limits coincide and equal s if and only if as
n→∞
1
n
yn → y0 in ∞ (N), where yn ( j) = xi+ j and y0 ( j) = s for all j.
n i=1
This does not imply that limk→∞ xk exists.
9.4 A Localised Range from Complex Function Theory
The description of a localised essential range in (9.4) is reminiscent of the cluster

set C D ( f, z 0 ) [10] of an analytic functions f at a boundary point z 0 of its domain of
definition. In Corollary 9.15, Shargorodsky [29] shows that the phenomenon illus-
trated in Example 9.11 arises naturally in the theory of complex Hardy spaces. Let
T and D denote the unit circle and the unit disc, respectively, and let

Bρ (α) := z ∈ C : |z − α| < ρ , α ∈ C, ρ > 0.
Let f ∈ H ∞ (D). We say that ζ ∈ T is a singularity of f if f cannot be extended

analytically to ζ (see, e.g. [17, Chap. II, Sect. 6]).
At almost every ζ ∈ T, f has a non-tangential limit, denoted by f (ζ), at ζ. The
following notation differs from that of [17, Chap. II, Sect. 6]:

R( f )(ϑ0 ) := α ∈ C : λ ϑ ∈ R : |ϑ − ϑ0 | < δ, f (eiϑ ) − α < ε > 0

for all ε, δ > 0 , ϑ0 ∈ (−π, π],
where λ denotes the standard Lebesgue measure on R.
Theorem 9.14 Let f ∈ H ∞ (D) be an inner function and let eiϑ0 ∈ T be its singu-
larity. Then R( f )(ϑ0 ) = T.
Proof Since | f | = 1 almost everywhere on T, one has R( f )(ϑ0 ) ⊆ T.

Suppose α ∈ T does not belong to R( f )(ϑ0 ). Then there exist ε, δ ∈ (0, 1) with

λ ϑ ∈ R : |ϑ − ϑ0 | < δ, f eiϑ − α < ε = 0. (9.6)
Let α±ε be the two points where the circle {ζ ∈ C : |ζ− α| = ε} intersects T and let
T(α, ε) and D(α, ε) be the intersections of T and D T with the closed half-plane
9.4 A Localised Range from Complex Function Theory 85
containing 0 and bounded by the straight line through α±ε :

T(α, ε) := ζ ∈ T| Re αζ ≤ Re (αα±ε ) ,

D(α, ε) := ζ ∈ D T| Re αζ ≤ Re (αα±ε ) .
Let > 0 be the distance from α to D(α, ε). (It is not difficult to see that is the
distance from α to the midpoint of the chord [α−ε , αε ] and = 1 − cos(2 arcsin 2ε ).)
For the Poisson kernel
1 − r2
Pr (ϑ) := , 0 ≤ r < 1, ϑ ∈ R,
1 − 2r cos ϑ + r 2
there exists δ0 > 0 such that

Pr (θ − ϑ) dϑ < /2, for all z = r eiθ ∈ Bδ0 (eiϑ0 ). (9.7)
δ≤|ϑ−ϑ0 |≤π

It follows from (9.6) that f eiϑ ∈ T(α, ε) for almost all ϑ with |ϑ − ϑ0 | < δ. Since
Pr ≥ 0,
Pr (θ − ϑ) dϑ < Pr (θ − ϑ) dϑ = 1,
|ϑ−ϑ0 |<δ |ϑ−ϑ0 |≤π
T(α, ε) ⊂ D(α, ε) and the latter is a closed convex set containing 0, it follows that

Pr (θ − ϑ) f (eiϑ ) dϑ ∈ D(α, ε). (9.8)
|ϑ−ϑ0 |<δ
On the other hand, it follows from (9.7) that

P (θ − ϑ) f (e iϑ
) dϑ ≤ Pr (θ − ϑ) dϑ < /2 (9.9)
r
δ≤|ϑ−ϑ0 |≤π δ≤|ϑ−ϑ0 |≤π
for all r eiθ ∈ Bδ0 (eiϑ0 ). Then it follows from the definition of and from (9.8), (9.9)
that for z = r eiθ ∈ Bδ0 (eiϑ0 ),

f (z) = Pr (θ − ϑ) f (eiϑ ) dϑ
|ϑ−ϑ0 |≤π

= Pr (θ − ϑ) f (eiϑ ) dϑ + Pr (θ − ϑ) f (eiϑ ) dϑ,
|ϑ−ϑ0 |<δ δ≤|ϑ−ϑ0 |≤π
cannot belong to B/2 (α). Since this contradicts [17, Chap. II, Sect. 6, Theorem
6.6], every α ∈ T must belong to R( f )(ϑ0 ). So, T ⊆ R( f )(ϑ0 ) ⊆ T, i.e. R( f )
(ϑ0 ) = T.
If f is an infinite Blaschke product, such that the closure of its zero set contains T, or
the singular inner function defined by a singular measure with closed support equal
to T, then the set of its singularities equals T (see [17, Chap. II, Sect. 6, the paragraph
after Theorem 6.6]).
Corollary 9.15 Let f ∈ H ∞ (D) be an inner function whose set of singularities

equals T. Then R( f )(ϑ) = T for every eiϑ ∈ T.
Remark 9.16 Corollary 9.15 provides an example of a function w1 defined on

(−π, π) and such that R(w1 )(x) = T for every x ∈ (−π, π). Let w(x) := w0 (x1 ),
x = (x1 , . . . , xn ) ∈ (−π, π)n . It is easy to see that R(w)(x) = T for every x ∈
(−π, π)n .
Chapter 10
Reconciling Representations
Theorem 3.1 identifies elements of L ∞ (X, L, λ)∗ with finitely additive measures
in L ∗∞ (X, L, λ) and Theorem 2.37 (Riesz) identifies bounded linear functionals on
C(G, τ ) with regular real Borel measures, denoted by C ∗ (G, τ ), on (G, τ ). However,
the isometric isomorphism in Theorem 7.4 means that these representations must be
related to one another and the finite additivity of one should be reconciled with the
σ-additivity of the other. This issue will be addressed first.
Then, in a topological space setting, for a functional f ∈ L ∞ (X, B, λ)∗ let ν 0
be the finitely additive measure in the Yosida–Hewitt representation of f and ν̂ 0
the regular real Borel measure in the classical Riesz representation of f restricted
to the space C0 (X, ) of continuous functions that vanish at infinity. A minimax
formula for ν̂ in terms ν is obtained in Sect. 10.2 when ν is non-negative.
10.1 (G, τ ) Versus L ∞ (X, L, λ)
In this section, (X, L, λ) is the general measure space introduced in Chap. 3. For
a bounded linear functional f on L ∞ (X, L, λ), let f˜ ∈ C(G, τ )∗ be defined by
f˜ ◦ L = f , where L is as in Theorem 7.4. Then with ν ∈ L ∗∞ (X, L, λ) representing
f ∈ L ∞ (X, L, λ)∗ let ν̃ ∈ C ∗ (G, τ ) denote the regular real Borel measure on G that
represents f˜ ∈ C(G, τ )∗ . Since, for any A ∈ L and ω ∈ G,

L[χ A ](ω) = χ A dω = ω(A) = χΔ A (ω),
X
it follows that χΔ A = L[χ A ]. Hence for A ∈ L and ν ∈ L ∗∞ (X, L, λ),

ν̃(Δ A ) = χΔ A d ν̃ = f˜(χΔ A ) = f˜ L[χ A ] = f (χ A ) = χ A dν = ν(A).
G A
Thus the key relation is

https://doi.org/10.1007/978-3-030-34732-1_10
88 10 Reconciling Representations
ν̃(Δ A ) = ν(A) for all A ∈ L, ν ∈ L ∗∞ (X, L, λ), (10.1)
which is compatible with (7.1b) and the additivity

properties of ν and ν̃. To
illustrate this suppose {Ai : i ∈ N} ⊂ L, Ai A j ∈ N when j = j and 0 ν ∈
L ∗∞ (X, L, λ). Then the σ-additivity of ν̃, the finite additivity of ν, (7.1b) and
Theorem 7.10 yield

∞ ∞
∞

ν̃ Δ Ai = ν̃ Δ Ai = ν Ai
i=1 i=1 i=1

∞
∞
ν Ai = ν̃ Δ
∞ = ν̃ Δ Ai .
i=1 Ai
i=1 i=1
Remark 10.1 From (10.1) it follows that ν 0 on L if ν̃ 0 on the Borel σ-algebra

in G. Conversely, if ν 0 on L then ν̃ 0 by Corollary 3.4 of the Riesz represen-
tation theorem.
Theorem 10.2 γ ∈ L ∗∞ (X, L, λ) is σ-additive if and only if γ̃ ∈ C ∗ (G, τ ) is zero
on every closed, nowhere dense G δ -set in (G, τ ).
Proof If H ⊂ G is a closed G δ -set which is nowhere dense, H◦ = ∅ in (G, τ ).
∞ ∞ ◦
∞ of sets Ak+1 ⊂ Ak in L with H =
By Corollary 7.11, there is a nested sequence

k=1 Δ Ak and Δ k=1 Ak
= H = ∅. Hence k=1 Ak ∈ N by Theorem 5.6.
Now suppose that γ ∈ L ∗∞ (X, L, λ) is σ-additive. Then the σ-additivity of both
γ̃ and γ, and the fact that γ λ, imply that

∞
γ̃(H) = lim γ̃(Δ Ak ) = lim γ(Ak ) = γ Ak = 0.
k→∞ k→∞
k=1
Conversely, suppose that γ̃(H) = 0 for every closed, nowhere dense G δ -set H in
(G, τ ). Let {E k } ⊂ L
with E k+1 ⊂ E k and k∈N E k = ∅. Since the nested sets Δ Ek
are open and closed, k Δ Ek is a closed G δ -set which is nowhere dense since, by
Corollary 7.11, its interior is empty. Therefore
∞

lim γ(E k ) = lim γ̃(Δ Ek ) = γ̃ Δ Ek = 0,
k→∞ k→∞
k=1
which, by Lemma 4.2(a), proves γ is σ-additive.

Definition 10.3 (Concentration of Measures) A real measure ν̃ is concentrated on
a set H ⊂ G if ν̃(B) = 0 for all Borel sets B ⊂ G \ H.
Theorem 10.4 μ ∈ L ∗∞ (X, L, λ) is purely finitely additive if and only μ̃ is concen-
trated on a closed, nowhere dense G δ -set H ⊂ G.
10.1 (G, τ ) Versus L ∞ (X, L, λ) 89
Proof To begin, suppose μ ∈ L ∗∞ (X, L, λ) is non-negative

with μ̃ concentrated on a
closed, nowhere dense G δ -set H and let γ ∈ (L) L ∗∞ (X, L, λ) with 0 γ μ.
Then 0 γ̃ μ̃ by Remark 10.1 and hence γ̃ is concentrated on a closed, nowhere
dense G δ -set. It follows from Theorem 10.2 that γ̃ is zero. Hence by (10.1) γ is
zero and by Definition 4.8 μ 0 is purely finitely additive. The result for general μ
follows because μ̃ = μ̃+ − μ̃− , by (4.1c), and μ̃+ and μ̃− are concentrated on H if
μ̃ is concentrated on H.
Since λ is σ-finite there exists f ∈ L 1 (X,
L, λ) which is strictly positive on
X . Let γ f ∈ (L) be defined by γ f (E) = X f dλ, E ∈ L, and suppose 0 μ ∈
L ∗∞ (X, L, λ) is purely finitely additive.
By Corollary 4.12, there exists a nested sequence E k+1 ⊂ E k in L such that
μ(E k ) = μ(X ) for all k and γ f (E k ) → 0 as k → ∞. Since γ f is σ-additive,
γ f (E) = limk γ f (E k ) = 0 where E := k E k . Therefore λ(E) = 0 since f > 0
λ-almost everywhere. Moreover μ̃(Δ Ek ) = μ(E k ) = μ(X ) = μ̃(G)
for all k.
Since E ∈ N it follows from Theorem 7.10 that H := k Δ Ek is a closed,
nowhere dense G δ -set in (G, τ ) and μ̃ (H) = μ̃(G) since μ̃ is σ-additive. There-
fore B ⊂ G \ H implies
μ̃(G) = μ̃(B) + μ̃(G \ B) μ̃(B) + μ̃(H) = μ̃(B) + μ̃(G),
and hence μ̃(B) = 0, as required. The general case is immediate since μ = μ+ − μ−

by (4.1c).
10.2 Restriction to C0 (X, ) of Elements of L ∗∞ (X, B, λ)
When (X, ) is a locally compact Hausdorff space, a bounded linear functional on

L ∞ (X, B, λ), when restricted to the Banach space C0 (X, ), has a representation as
a regular real Borel measure ν̂ on X . This section explores the relationship between ν
and ν̂. To be precise, by Corollary 3.4 for ν ∈ L ∗∞ (X, B, λ), there is a unique regular
real Borel measure ν̂ with

v dν = v d ν̂ for all v ∈ C0 (X, ). (10.2)
X X
When X is compact it follows that ν(X ) = ν̂(X ), because in that case non-zero
constant functions belong to C0 (X, ). However, when (X, ) is not compact ν̂(X )
may be zero when ν is non-negative and ν(X ) = 1.
Since ν = ν + − ν − , without loss of generality attention will henceforth be
restricted to non-negative ν ∈ L ∗∞ (X, B, λ) with the goal of understanding how
ν̂ 0 depends on ν 0. Note:
(i) from the Yosida–Hewitt decomposition (6.9), ν = μ + gλ where μ ∈
L ∗∞ (X, B, λ) is purely finitely additive and gλ, g ∈ L 1 (X, B, λ), is σ-additive.
(ii) from Theorem 2.28 (Lebesgue–Radon–Nikodym), ν̂ = ρ + kλ where ρ and kλ,

k ∈ L 1 (X, B, λ), are σ-additive, ρ singular and kλ is absolutely continuous with
respect to λ (see Definition 2.18).
As a consequence, for v ∈ C0 (X, ),

v dμ + vg dλ = v dν = v d ν̂ = v dρ + vk dλ. (10.3a)
X X X X X X
As Valadier and Hensgen noted (see Sect. 6.6), the relations between μ and ρ, and
g and k are not straightforward: for Lebesgue measure λ on [0, 1], there is a non-
negative ν ∈ Π (B) with
1 1
v dν = v dλ for all v ∈ C[0, 1], (10.3b)
0 0
which means 0 = μ ∈ Π (B) and g = 0 but ρ = 0 and k ≡ 1. Subsequently,

Abramovich and Wickstead [1] provided wide a ranging generalisation of this obser-
vation. Recently, Wrobel [34] gave a sufficient condition on a purely finitely additive
ν on L ∞ [0, 1] for ν̂ to be singular with respect to Lebesgue measure. His result is
essentially a special case of the “if” part of Corollary 10.10 when X is compact.
To find a formula for ν̂ satisfying (10.2) for a given non-negative ν ∈ L ∗∞ (X, B, λ),
and to characterise those ν for which ν̂ has a singularity, recall the following version
of Urysohn’s Lemma 2.61 .
Lemma 10.5 ([28, Sect. 2.12]) If X is a locally compact Hausdorff space and
K ⊂ G ⊂ X where K is compact and G is open, there exists a continuous function
f : X → [0, 1] such that f (K ) = 1 and {x : f (x) > 0} is a compact subset of G.
In particular, f ∈ C0 (X, ).
Lemma 10.6 Suppose (X, ) is a locally compact Hausdorff space, 0 ν ∈

L ∗∞ (X, B, λ) and B ∈ B. Then for compact K and open G with K ⊂ B ⊂ G,
ν(K ) ν̂(B) ν(G),

ν(K ) ν̂(K ) and ν̂(G) ν(G) for compact K and open G,
and ν(F) ν̂(F) + ν(X ) − ν̂(X ) for closed F.
In particular, ν̂(X ) = ν(X ) implies ν(F) ν̂(F) for all closed F ⊂ X . (ν(X ) =
ν̂(X ) when (X, ) is compact was noted following (10.2).)
Proof For a given Borel set B and K ⊂ B ⊂ G as in the statement, let f be the
continuous function determined in Lemma 10.5 by K and G. Then

ν(K ) f dν ν(G) and ν̂(K ) f d ν̂ ν̂(G).
X X
10.2 Restriction to C0 (X, ) of Elements of L ∗∞ (X, B, λ) 91
From (10.2), it follows that ν(K ) ν̂(G) and ν̂(K ) ν(G) whence, since ν̂ is
a regular real Borel measure (Corollary 3.4), ν(K ) ν̂(B) ν(G). In particular,
if B = K is compact, ν(K ) ν̂(K ), and if B = G is open, ν̂(G) ν(G). That
ν(F) ν̂(F) + ν(X ) − ν̂(X ) when F is closed follows by finite additivity since
X \ F is open and 0 ν(X ), ν̂(X ) < ∞.
Remark 10.7 A non-negative finitely additive set function ν on B is said to be regular

[12, III.5.11] if for all E ∈ B and > 0 there are sets F ⊂ E ⊂ G with F closed, G
open and ν(G \ F) < . If X is compact and ν is regular in this sense, by a theorem
of Alexandroff [2, Pt. I, p. 590], [12, III.5.13] ν is σ-additive and hence ν̂ = ν. By
Lemma 10.6, if ν(X ) = ν̂(X ) and F ⊂ E ⊂ G, where F is closed and G is open,
ν(F) ν̂(F) ν̂(E) ν̂(G) ν(G).
Hence, ν 0 regular implies ν = ν̂ is σ-additive on B if ν(X ) = ν̂(X ).
Theorem 10.8 Suppose that (X, ) is a locally compact Hausdorff space, K ⊂ G

where K is compact, G is open and ν ∈ L ∗∞ (X, B, λ) is non-negative. Then for n ∈ N
there exists a compact set K n and an open set G n with
K ⊂ G n ⊂ K n ⊂ G, G n ⊂ G n−1 , K n ⊂ K n−1 ,
ν̂(K ) ν(K n ), ν̂(G) ν(G n ) and λ(K n ) < λ(K ) + 1/n.
Proof Since λ is a regular Borel measure that is finite on compact sets there exist
open sets G k with K ⊂ G k ⊂ G and λ(G k ) < λ(K ) + 1/k for k ∈ N.
Since K ⊂ G k , for k ∈ N there exists, by Lemma 10.5, a continuous function
f k : X → [0, 1] such that f k (K ) = 1 and {x : f k (x) > 0} is a compact subset of G k .
For x ∈ X , let gn (x) = min{ f k (x) : k n} so that gn gn−1 , gn is continuous on
X , gn (K ) = 1 and {x : gn (x) > 0} ⊂ G n is compact.
Let G n = {x : gn (x) > 0} and K n = {x : gn (x) > 0}. Then K ⊂ G n ⊂ K n ⊂
G n ⊂ G and, by Lemma 10.6,
ν̂(K ) ν̂(G n ) ν(G n ) ν(K n ), ν̂(G) ν̂(K n ) ν(K n ) ν(G n ),
and λ(K n ) < λ(K ) + 1/n because K n ⊂ G n . Now {G n } and {K n } are nested
sequences of open and compact sets, respectively, because gn (x) is decreasing in
n, with the required properties. This completes the proof.
Corollary 10.9 For G open, K compact and ν ∈ L ∗∞ (X, B, λ) non-negative,
ν̂(G) = sup{ν(K ) : K ⊂ G, K compact},

ν̂(K ) = inf{ν(G) : K ⊂ G, G open}.
Proof From Lemma 10.6, for any open set G and all compact K ⊂ G, ν(K )
ν̂(K ) ν̂(G). Since ν̂ is a regular real Borel measure, for any > 0 there exists
compact K ⊂ G with ν̂(K ) > ν̂(G) − . Now by Theorem 10.8 there exists compact
K 1 with K ⊂ K 1 ⊂ G and ν(K 1 ) ν̂(K ) > ν̂(G) − . This establishes the first
identity. For a given compact set K and any open set G with K ⊂ G, ν̂(K ) ν̂(G)
ν(G) and for > 0, there exists an open G with K ⊂ G and ν̂(G) < ν̂(K ) + . By
Theorem 10.8, there exists an open set G 1 with K ⊂ G 1 ⊂ G with ν̂(K ) + >
ν̂(G) ν(G 1 ), and the result follows.
Say that ν̂ has a singularity with respect to λ if ν̂(E) = 0 (equivalently ρ(E) = 0)

for some E ∈ N .
Corollary 10.10 Let (X, ) be a locally compact Hausdorff space and 0 ν ∈

L ∗∞ (X, B, λ). Then 0 ν̂ ∈ (B) has a singularity if and only if there exists
α > 0 and a sequence of compact sets with ν(K n ) α, K n+1 ⊂ K n for all n, and
λ(K n ) → 0 as n → ∞.
Proof If α > 0 and such a sequence exists, by Lemma 10.6, ν̂(K n ) α for all n.
Since {K n } is nested and ν̂ is σ-additive it follows that ν̂(K ) α where K = n K n .
Since K ∈ N , because limn→∞ λ(K n ) = 0 and λ is σ-additive, ν̂ has a singularity.
Conversely, if ν̂ 0 has a singularity there exists E ∈ N and α > 0 with ν̂(E) =
2α. Since ν̂ is a real Borel measure which is regular (Definition 2.21), there exists a
compact K ⊂ E with ν̂(K ) α > 0. Now since λ(K ) = 0 because K ⊂ E ∈ N ,
the existence of a sequence {K n } of compact sets with ν(K n ) ν̂(K ) α, K n+1 ⊂
K n for all n, and λ(K n ) → 0 as n → ∞ follows from Theorem 10.8.
Theorem 10.11 For (X, ) locally compact and 0 ν ∈ L ∗∞ (X, B, λ),
⎧ ⎫ ⎧ ⎫
⎪
⎨ ⎪
⎬ ⎨ ⎬
ν̂(B) = inf sup ν(K ) = sup inf ν(G) ,
G open ⎪
⎩ K compact ⎪
⎭ K compact
⎩ G open ⎭
B⊂G K ⊂G K ⊂B K ⊂G
for all B ∈ B.
Proof Since ν̂ is a regular real Borel measure,
ν̂(B) = inf{ν̂(G) : B ⊂ G, G open} = sup{ν̂(K ) : K ⊂ B, K compact},
and the formulae follow from Corollary 10.9.
Corollary 10.12 For a locally compact Hausdorff space (X, ) and ω ∈ G,
(a) either ω̂ is zero or ω̂ is a Dirac measure δx0 ;

(b) both possibilities may occur when (X, ) is not compact;
(c) if ω̂ = δx0 ∈ D, then ω ∈ G(x0 ), i.e.(9.1a) holds .
Proof (a) By Lemma 9.1, either ω(K ) = 0 for all compact K , in which case ω̂ = 0
by the first formula for ω̂(B), or ω(K ) = 1 for some compact K . In the latter case,
10.2 Restriction to C0 (X, ) of Elements of L ∗∞ (X, B, λ) 93
there is a unique x0 ∈ X for which ω(G) = 1 if x0 ∈ G and G is open. From the

second formula for ω̂(B) it is immediate that ω̂(B) = 1 if and only if x0 ∈ B. Hence
ω̂ = δx0 ∈ D.
(b) For an example of both possibilities let X = (0, 1) with the standard (locally
compact but not compact) topology and Lebesgue measure. Let ω ∈ G be defined by
Theorem 5.6 with E = (0, 1/), ∈ N. Then ω(K ) = 0 for all compact K ⊂ (0, 1)
and hence ω̂ = 0. On the other hand, if E = (1/2 + 1/, 1/2) in Theorem 5.6,
ω ∈ G with ω([1/2 + 1/, 1/2]) = 1 for all and hence ω̂ = δ1/2 ∈ D.
(c) If ω̂ = δx0 let G an open set with x0 ∈ G. Since {x0 } is compact, there exists
v ∈ C0 (X, ) with v(X ) ⊂ [0, 1], v(x0 ) = 1, v(X \ G) = 0 and

1 ω(G) v dω = v dω = v d ω̂ = v(x0 ) = 1,
G X X
Hence, ω(G) = 1 for every open set with x0 ∈ G.

References
1. Y.A. Abramovich, A.W. Wickstead, Singular extensions and restrictions of order continuous
functionals. Hokkaido Math. J. 21, 475–482 (1992)
2. A.D. Alexandroff, Additive set-functions in abstract spaces I, II, III. Rec. Math. (Mat. Sbornik)
N.S. 8(50), 307–348 (1940); 9(51), 563–628 (1941); 13(55), 169–238 (1943)
3. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis—A Hitchiker’s Guide (Springer,
Berlin, 2006)
4. J.M. Ball, Weak continuity properties of mappings and semigroups. Proc. Roy. Soc. Edin.
Sect. A 72(4), 275–280 (1975)
5. S. Banach, Théorie des Opérations Linéaires. Monografje Matematyczne, Tom I, Z Subwe-
hcji Fundusku Kultryy Narodowej, Warszawa, 1932; reprinted with corrections and added
material, Chelsea Publishing Co. New York, 1955; Translated from the French by F. Jellett,
with comments by A. Pelczyński and Cz. Bessaga. North-Holland Mathematical Library, vol.
38. North-Holland Publishing Co., Amsterdam, 1987 & Dover Books
6. K.P.S. Bhaskara Rao, M. Bhaskara Rao, Theory of Charges - A Study of Finitely Additive
Measures, vol. 109, Pure and Applied Mathematics (Academic Press, London, 1983)
7. N.H. Bingham, Finite additivity vs countable additivity: De Finetti and Savage. J. Électron.
Hist. Probab. Stat. 6(1), 35 (2010)
8. V.I. Bogachev, Measure Theory, vol. 1 (Springer, Berlin, 2007)
9. A.G. Chentsov, Finitely Additive Measures and Relaxations of Extremal Problems. Mono-
graphs in Contemporary Mathematics (Plenum, New York, 1996)
10. E.F. Collingwood, A.J. Lohwater, The Theory of Cluster Sets (Cambridge University Press,
Cambridge, 1966)
11. J.B. Conway, A Course in Functional Analysis, 2nd edn. (Springer, New York, 2007)
12. N. Dunford, J.T. Schwartz, Linear Operators, vol. I (Wiley Interscience, New York, 1958)
13. G. Fichtenholz, Sur les fonctions d’ensemble additives et continues. Fund. Math. 7(1), 296–
301 (1925)
14. G. Fichtenholz, L. Kantorovitch, Sur les opérations lináires dans lespace des fonctions bornés.
Stud. Math. 5, 69–98 (1934)
15. G.B. Folland, Real Analysis, 2nd edn. (Wiley, New York, 1999)
16. I. Fonseca, G. Leoni, Modern Methods in the Calculus of Variations: L p Spaces, Springer
Monographs in Mathematics (Springer, New York, 2007)
17. J.B. Garnett, Bounded Analytic Functions (Revised First Edition, Springer, New York, 2007)
18. J. Heinonen, Lectures on Analysis on Metric Spaces (Springer, New York, 2001)

https://doi.org/10.1007/978-3-030-34732-1
96 References
19. W. Hensgen, An example concerning the Yosida-Hewitt decomposition of finitely additive

measures. Proc. Amer. Math. Soc. 121, 641–642 (1994)
20. T.H. Hildebrandt, On bounded functional operations. Trans. Am. Math. Soc. 36, 868–875
(1934)
21. J.L. Kelley, General Topology (Van Nostrand Reinhold, Toronto, 1955)
22. R.B. Kirk, Sets which split families of measurable sets. Amer. Math. Monthly 79, 884–886
(1972)
23. P.D. Lax, Functional Analysis (Wiley, Hoboken, 2002)
24. G.G. Lorentz, A contribution to the theory of divergent sequences. Acta. Math. 80, 167–190
(1948)
25. J.C. Oxtoby Review: K.P.S. Bhaskara Rao, M. Bhaskara Rao, Theory of charges, a study of
finitely additive measures. Bull. Am. Math. Soc. 11(1), 221–223 (1984)
26. J. Rainwater, Weak convergence of bounded sequences. Proc. Amer. Math. Soc. 14(6), 999
(1963)
27. W. Rudin, Principles of Mathematical Analysis, 3rd edn. (McGraw-Hill, New York, 1976)
28. W. Rudin, Real and Complex Analysis, 3rd edn. (McGraw-Hill, New York, 1986)
29. E. Shargorodsky. Personal Communication
30. W. Sierpiński, Sur les fonctions d’ensemble additives et continues. Fund. Math. 3(1), 240–246
(1922)
31. L. Sucheston, Banach limits. Amer. Math. Monthly 74, 308–311 (1967)
32. T. Tao, Amenability, the ping-pong lemma, and the Banach-Tarski paradox, https://terrytao.
wordpress.com/2009/01/08/245b-notes-2-amenability-the-ping-pong-lemma-and-the-
banach-tarski-paradox-optional/#more-1354
33. M. Valadier, Une singulière forme linéaire sur L ∞ . Sém. d’Analyse Convexe Montpelier
17(4) (1987)
34. A.J. Wrobel, A sufficient condition for a singular functional on L ∞ [0, 1] to be represented
on C[0, 1] by a singular measure. Indag. Math. 29(2), 746–751 (2018)
35. K. Yosida, E. Hewitt, Finitely additive measures. Trans. Amer. Math. Soc. 72, 46–66 (1952)
Index
Symbols G(x0 ): G localised at x0 ∈ X ∞ , 78

(X, ) a topological space, 9 U ultrafilters, 3
(X ∞ , ∞ ) one-point compactification, 18 D, 84
((G, τ ), 57 T, 84
Aα (u), 70 D Dirac measures, 5, 66
B ∗ , 22 U, 42
C(G, τ ), 3 ν1 ν2 , partial ordering, 33
C ∗ (G, τ ), C(G, τ )∗ , 87 ν + , ν − , |ν|, 34
C0 (X, ), 18 ν1 ν2 , absolute continuity, 35
C D ( f, z 0 ), 84 ν1 ⊥ ν2 , 35
EF, 14 ν1 ∨ ν2 , ν1 ∧ ν2 , sup and inf, 33
Gδ M, 9
closed, 24 ω-almost everywhere, 42
set, 24 ωU , Uω , 43
space, 24 ∂(E, F), 14
G δ -space, 25
≺, 7
H ∞ (D), 84
, 7
L p -spaces, 17
B, 9
L p (X, L, λ), 17
L, 8
X ∞ , one-point compactification, 2
M, 8
L ∞ (X, L, λ), 18
N null sets, 10
L ∞ (X, L, λ)∗ , 27
T , 57
L ∗∞ (X, L, λ), 3, 27, 39, 44
W , 63
(L) purely finitely additive measures, 36
σ -additive measures, 31
(L), 13, 31
ba(L) finitely additive measures, 31 σ -additivity, 10
ba(L) = (L) ⊕ (L), 38 σ -algebra, 8
χ A characteristic function, 28 σ -compact, 18
∞ , 45 ∼, 14
∞ (N), 21, 45, 46, 71, 83 collection of open sets, 9
(†), v, 3, 16, 31, 42 , weak convergence, 1
L̂, 14 ℘ (S) the power set of S, 7
∗
λ-finite intersection property, 44 , 22
G 0-1 finitely additive measures, 3, 41 x∞ , point at infinity in (X ∞ , ∞ ), 18
Hausdorff topology, 57 (V), 1
weak* closed in L ∗∞ (X, L, λ), 64 (W), 1

https://doi.org/10.1007/978-3-030-34732-1
98 Index
A F
Absolute continuity, 35 Filter, 42
Algebra, 8 Fine structure at x0 of u ∈ L ∞ (X, B, λ), 81
Algebras and σ -algebras of sets, 8 Finitely additive measures, 31
Atom, 11 First axiom of countability, 23
Axiom of choice, 8 Functionals, bounded, 19
Function spaces, 17
B
H
Banach limit, 21, 53, 84
Hahn decomposition, 13
Banach space, 19
Hardy spaces, complex, 84
Base, 23
Hausdorff topology, 24
Blaschke product, 86
Borel
σ -algebra, 9, 77
I
measure, 77
Inner function, 84
sets, 9 Integration, 47
with respect to ν ∈ G, 51
Isolated, 60
C Isometric isomorphism, 58
Canonical decomposition, 38
Characteristic function, 10, 14
Closed G δ -set, 62, 79 J
Cluster set, 84 Jordan decomposition, 13
Cluster value, 77
Completely separable, 24, 60, 81
Complex function theory, 84 L
Convergence Lebesgue
in measure, 10 σ -algebra, 12
λ-almost everywhere, 10 decomposition, 17
Countable, 7 integrable, 16
Counting measure on N, 11 integral, 15
measure, 12
point, 69
Lemma
D
Urysohn, 24, 90
Darboux property, 11
Zorn, 3, 7, 8, 42–44
Denumerable, 7
Local base, 23
Dirac measure, 11
Localising
Directional limit of u at x0 , 82
G, 77
Discrete topology, 19
complex function theory, 84
Dual space, 1, 19
essential range, 81
weak convergence, 80
Locally compact, 18, 77
E
Equality almost everywhere, 42
Equivalence M
class, 8 Maximal, 8, 42
relation, 8 Maximal element, 8
Essential range, 52, 77 Maximum norm, 18
localised, 81 Measurable
Extended real numbers, 7 functions, 9
Extreme point, 22, 64 sets, 9
Index 99
space, 9 T
Measure, 10 Theorem
σ -finite, 10 Alaoglu, 22, 64, 68
Borel outer, inner regular, 11 Alexandroff, 91
finite, 10 Baire’s category, 3, 14, 15
Lebesgue, 12 Banach–Steinhaus, 21
space, 10 Dini, 71
Measure space Dominated convergence, 1
complete, 10 Hahn–Banach, 5, 20, 28, 44, 69
Krein–Milman, 23
Lebesgue–Radon–Nikodym, 16, 90
N Lusin, 19
Normal topology, 24 Mazur, 21, 69
Null sets, 10
Monotone convergence, 49
Nikodym’s convergence, 32
Radon–Nikodym, 17, 50
O
One-point compactification, 18 Rainwater, 23
Oscillatory functions, 74 Riesz representation, 1, 20, 27, 87
Stone–Weierstrass, 21, 59
Valadier–Hensgen, 3, 5, 54, 56
P Weierstrass approximation, 21
Partial ordering, 33 Yosida–Hewitt representation, v, 3, 27,
Poisson kernel, 85 50, 87
Positive and negative parts, 13 Totally disconnected, 60
Purely finitely additive measures, 36 Total ordering, 7
Total variation, 13
Translations, 74
R
Real measure, 12
Reconciling representations, 87
U
Regular
Ultrafilter, 42, 43
Borel measure, 11
Upper bound, 8
inner, 11
outer, 11
real Borel measure, 13
Regular topology, 24 W
Weak*
closed convex hull, 23
S convergence, 22
Second axiom of countability, 24 sequential compactness, 22
Separable, 22, 23 topology, 22
Sequential weak continuity, 67 topology on L ∗∞ (X, L, λ), 63
Shargorodsky, 5, 84 Weak convergence
Shift operator, 53 at a point, 77, 80
Simple function, 10, 73 criterion, 70
Singleton, 7
of sequences, 21, 67
Singular, 35
Splitting measurable sets, 14
Step functions, 73
Sub-base, 22, 23, 57, 63 Y
Symmetric difference, 14 Yosida–Hewitt decomposition, 89

2020 Book TheDualOfLXLΛFinitelyAdditiveM PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

2020 Book TheDualOfLXLΛFinitelyAdditiveM PDF

Hochgeladen von

Copyright:

Verfügbare Formate

SPRINGER BRIEFS IN MATHEMATICS

The Dual of L∞(X, L, λ), ),

SpringerBriefs in Mathematics showcases expositions in all areas of mathematics

More information about this series at http://www.springer.com/series/10030

Typical topics might include:

• A timely report of state-of-the art techniques

The Dual of L1ðX; L; ‚Þ,

ISSN 2191-8198 ISSN 2191-8201 (electronic)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2020

properties of ﬁnitely additive measures on r-algebras are less circumscribed by hypotheses

RðuÞðx0 Þ is sometimes referred to as the cluster set of u at x0 .

Bath, UK John Toland

6 Integration and Finitely Additive Measures . . . . . . . . . . . . . . . . . . 47

{vk } bounded and w ∗ (vk ) → 0 for all w ∗ ∈ W ∗ imply vk 0. (W)

The possibility of usefully extending these observations to L ∞ (X, L, λ) (the real

u k u in L ∞ () implies that u k (x) → u(x) almost everywhere and {u k ∞ } is

Independently, it is shown that τ coincides with the restriction to G of the weak*

It follows that when F : R → R is continuous, u → F(u) is sequentially weakly

localises the essential range and reflects the fine structure of u at x0 .

with an example due to Shargorodsky of a similar phenomenon occurring naturally

2.1 Partial Ordering and Zorn’s Lemma

For a non-empty set S and R ⊂ S × S, a relation  on S is defined by writing x  y

Definition 2.1 (Partial and Total Ordering) A partial ordering on S is a relation

Any collection S of subsets of a set is partially ordered by set inclusion, i.e. A  B

Definition 2.3 (Upper Bounds and Maximal Elements) If  is a partial ordering on

2.2 Algebras and σ-Algebras

For an arbitrary set X , a collection M ⊂ ℘ (X ) is an algebra if

M ⊂ ℘ (N), defined by A ∈ M if and only if either A or N \ A is finite, is an

2.3 Measurable Sets and Measurable Functions

A measurable space is a pair (X, L) where X = ∅ is arbitrary and L ⊂ ℘ (X ) is

The set of measurable functions has the following properties.

u(x) := inf u n (x) and u(x) := sup u n (x), x ∈ X, are measurable.

(iii) If u is measurable, u ± are measurable where u ± (x) := sup{±u(x), 0}, so that

u = u + − u − and |u| = u + + u − . (2.1)

(iv) If u, v are real-valued and measurable, c ∈ R and g : R → R is continuous,

The characteristic function χ A of a set A ∈ L, defined by

is a measurable function. Finite linear combinations of characteristic functions

2.4 Measures and Real Measures

When L is a σ-algebra this identity is referred to as the σ-additivity of λ. Some texts

A sequence of measurable functions which converges in measure has a subsequence

Definition 2.8 (Atom) A ∈ L is an atom if A ∈

Example 2.15 (Lebesgue Measure on Rn ) Let B denote the Borel subsets of X = Rn

This is the classical Lebesgue measure space and λ is Lebesgue measure on Rn .

which shows that E is not an atom.

of λ to the σ-algebra {Ω E : E ∈ L} creates a complete measure space which is

For a real measure μ and E ∈ L let

0 ≤ μ+ := 21 (|μ| + μ) and 0 ≤ μ− := 21 (|μ| − μ), (2.4)

μ = μ+ − μ− and |μ| = μ+ + μ− , 0 ≤ μ± (E) ≤ |μ|(X ) < ∞, E ∈ L.

where μ± are defined in (2.4).

μ  λ if and only if λ(E) = 0 implies |μ|(E) = 0

and singular with respect to λ, written as

λ ⊥ μ if and only if λ(E) + |μ|(X \ E) = 0 for some E ∈ L.

where μ± are defined by (2.4).

2.5 Splitting Families of Measurable Sets

Since Lebesgue measure on Rn has no atoms and counting measure on N has

Theorem 2.25 If (X, L, λ) has no atoms and G is a denumerable subset of L \ N ,

The Lebesgue integral (which may be infinite) of a non-negative simple function

Finally, if γ is either a measure (in which case γ − = 0) or an arbitrary real measure,

Remark 2.27 When (X, L, λ) is a measure space and h : X → R is any integrable

 function h which is integrable with respect to λ with the property

Part (a) is commonly referred to as the Lebesgue decomposition of μ relative to λ

2.7 Function Spaces

L p (X, L, λ) is a real Banach space. It is obvious that u k − u p → 0 as k → ∞

{vk } bounded and w ∗ (vk ) → 0 for all w ∗ ∈ W ∗ imply vk 0. (W)

u k u in L ∞ () implies that u k (x) → u(x) almost everywhere and {u k ∞ } is

For a non-empty set S and R ⊂ S × S, a relation on S is defined by writing x y

Any collection S of subsets of a set is partially ordered by set inclusion, i.e. A B

Definition 2.3 (Upper Bounds and Maximal Elements) If is a partial ordering on

A measurable space is a pair (X, L) where X = ∅ is arbitrary and L ⊂ ℘ (X ) is

μ λ if and only if λ(E) = 0 implies |μ|(E) = 0

function h which is integrable with respect to λ with the property

L p (X, L, λ) is a real Banach space. It is obvious that u k − u p → 0 as k → ∞

Remark 2.33 (X ∞ , ∞ ) is Hausdorff if and only if (X, ) is locally compact and

Given a normed linear space (V, · ), a metric is defined on V by ρ(x, y) = x −

Proof Let Z = {y + αv0 : y ∈ Y, α ∈ R}, a linear subspace of V . Since v0 = 0,

Corollary 2.46 If yk y0 in a normed linear space V and {k j } ⊂ N is strictly

x ∈ G x ⊂ X \ G x ⊂ G which implies that x ∈ G x ⊂ G x ⊂ G for all x ∈ F.

Proof For F closed and x ∈ / F,let x ∈ G where G is open and G is compact.

there exists a finitely additive measure ν = 0 which is zero on N and

ν1 (Fk ) + ν2 (E k \ Fk ) (ν1 ∧ ν2 )(E k ) + k .

for all ∈ (0, 1). Hence