Foundations of Quantitative Finance Book V General Measure and Integration Theory 1St Edition Robert R Reitano Full Chapter

Foundations of Quantitative Finance.
Book V: General Measure and

Integration Theory 1st Edition Robert R.
Reitano
Visit to download the full and correct content document:
https://ebookmass.com/product/foundations-of-quantitative-finance-book-v-general-m
easure-and-integration-theory-1st-edition-robert-r-reitano/
Foundations of Quantitative
Finance
Chapman & Hall/CRC Financial Mathematics Series
Series Editors
M.A.H. Dempster
Centre for Financial Research
Department of Pure Mathematics and Statistics
University of Cambridge, UK
Dilip B. Madan
Robert H. Smith School of Business
University of Maryland, USA
Rama Cont
Mathematical Institute
University of Oxford, UK
Robert A. Jarrow
Ronald P. & Susan E. Lynch Professor of Investment ManagementSamuel Curtis Johnson Graduate School of
Management Cornell University
Recently Published Titles
Machine Learning for Factor Investing: Python Version
Guillaume Coqueret ad Tony Guida
Introduction to Stochastic Finance with Market Examples, Second Edition
Nicolas Privault
Commodities: Fundamental Theory of Futures, Forwards, and Derivatives Pricing, Second Edition
Edited by M.A.H. Dempster, Ke Tang
Introducing Financial Mathematics: Theory, Binomial Models, and Applications
Mladen Victor Wickerhauser
Financial Mathematics: From Discrete to Continuous Time
Kevin J. Hastings
Financial Mathematics: A Comprehensive Treatment in Discrete Time

Giuseppe Campolieti and Roman N. Makarov
Introduction to Financial Derivatives with Python
Elisa Alòs, Raúl Merino
The Handbook of Price Impact Modeling
Dr. Kevin Thomas Webster
Sustainable Life Insurance: Managing Risk Appetite for Insurance Savings & Retirement Products
Aymeric Kalife with Saad Mouti, Ludovic Goudenege, Xiaolu Tan, and Mounir Bellmane
Geometry of Derivation with Applications
Norman L. Johnson
Foundations of Quantitative Finance
Book I: Measure Spaces and Measurable Functions
Robert R. Reitano
Book II: Probability Spaces and Random Variables
Robert R. Reitano
Book III: The Integrals of Riemann, Lebesgue and (Riemann-)Stieltjes
Robert R. Reitano
Book IV: Distribution Functions and Expectations
Robert R. Reitano
Book V: General Measure and Integration Theory
Robert R. Reitano
For more information about this series please visit: https://www.routledge.com/Chapman-and-HallCRC-Financial-
Mathematics-Series/book-series/CHFINANCMTH
Foundations of Quantitative
Finance
Book V: General Measure and
Integration Theory
Robert R. Reitano
Brandeis International Business School
Waltham, MA
First edition published 2024
by CRC Press
2385 Executive Center Drive, Suite 320, Boca Raton, FL 33431
and by CRC Press

4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
CRC Press is an imprint of Taylor & Francis Group, LLC
© 2024 Robert R. Reitano
Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot as-
sume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have
attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders
if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please
write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or
utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including pho-
tocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission
from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.com or contact the
Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that are
not available on CCC please contact mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are used only for iden-
tification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Names: Reitano, Robert R., 1950- author.

Title: Foundations of quantitative finance. Book V, General measure and
integration theory / Robert R. Reitano.
Other titles: General measure and integration theory
Description: First edition. | Boca Raton, FL : CRC Press, 2024. | Series:
Chapman & Hall/CRC financial mathematics series | Includes
bibliographical references and index.
Identifiers: LCCN 2023038707 | ISBN 9781032206516 (hardback ; vol. 5) |
ISBN 9781032206509 (paperback ; vol. 5) | ISBN 9781003264576 (ebook ;
vol. 5)
Subjects: LCSH: Finance--Mathematical models. | Integration, Functional.
Classification: LCC HG106 .R4485 2023 | DDC 332.01/5195--dc23/eng/20230907
LC record available at https://lccn.loc.gov/2023038707
ISBN: 978-1-032-20651-6 (hbk)

ISBN: 978-1-032-20650-9 (pbk)
ISBN: 978-1-003-26457-6 (ebk)
DOI: 10.1201/9781003264576
Typeset in CMR10 font

by KnowledgeWorks Global Ltd.
to Michael, David, and Jeffrey
Taylor & Francis
Taylor & Francis Group
http://taylorandfrancis.com
Contents
Preface xi
Author xiii
Introduction xv
1 Measure Spaces 1
1.1 Lebesgue and Borel Spaces on R . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Starting Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 Lebesgue Measure Space . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.3 Borel Measure Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 General Extension Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Measure Space Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Finite Products of Measure Spaces . . . . . . . . . . . . . . . . . . . 11
1.3.2 Borel Measures on Rn . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.3 Infinite Products of Probability Spaces . . . . . . . . . . . . . . . . . 14
1.4 Continuity of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Measurable Functions 17
2.1 Properties of Measurable Functions . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Limits of Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Results on Function Sequences . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Approximating σ(X)-Measurable Functions . . . . . . . . . . . . . . . . . . 26
2.5 Monotone Class Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.1 Monotone Class Theorem . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5.2 Functional Monotone Class Theorem . . . . . . . . . . . . . . . . . . 34
3 General Integration Theory 36

3.1 Integrating Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Integrating Nonnegative Measurable Functions . . . . . . . . . . . . . . . . 41
3.2.1 Fatou’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 Lebesgue’s Monotone Convergence Theorem . . . . . . . . . . . . . . 49
3.2.3 Properties of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.4 Product Space Measures Revisited . . . . . . . . . . . . . . . . . . . 55
3.3 Integrating General Measurable Functions . . . . . . . . . . . . . . . . . . 57
3.3.1 Properties of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.2 Beppo Levi’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3.3 Lebesgue’s Dominated Convergence Theorem . . . . . . . . . . . . . 63
3.3.4 Bounded Convergence Theorem . . . . . . . . . . . . . . . . . . . . . 66
3.3.5 Uniform Integrability Convergence Theorem . . . . . . . . . . . . . . 67
3.4 Leibniz Integral Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
vii
viii Contents
3.4.1 Riemann Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.4.2 Lebesgue/Lebesgue-Stieltjes Integrals . . . . . . . . . . . . . . . . . 74
3.5 Lebesgue-Stieltjes vs. Riemann-Stieltjes Integrals . . . . . . . . . . . . . . . 76
3.5.1 Lebesgue-Stieltjes Integrals on R . . . . . . . . . . . . . . . . . . . . 76
3.5.2 Lebesgue-Stieltjes Integrals on Rn . . . . . . . . . . . . . . . . . . . 79
4 Change of Variables 84
4.1 Change of Measure: A Special Case . . . . . . . . . . . . . . . . . . . . . . 85
4.1.1 Measures Defined by Integrals . . . . . . . . . . . . . . . . . . . . . . 85
4.1.2 Integrals and Change of Measure . . . . . . . . . . . . . . . . . . . . 89
4.2 Transformations and Change of Measure . . . . . . . . . . . . . . . . . . . 92
4.2.1 Measures Induced by Transformations . . . . . . . . . . . . . . . . . 92
4.2.2 Change of Variables under Transformations . . . . . . . . . . . . . . 95
4.3 Special Cases of Change of Variables . . . . . . . . . . . . . . . . . . . . . 99
4.3.1 Lebesgue Integrals on R . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.2 Linear Transformations on Rn . . . . . . . . . . . . . . . . . . . . . 102
4.3.3 Differentiable Transformations on Rn . . . . . . . . . . . . . . . . . . 106
5 Integrals in Product Spaces 117

5.1 Product Space Sigma Algebras . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.1.1 Sigma Algebra Constructions . . . . . . . . . . . . . . . . . . . . . . 117
5.1.2 Implications for Chapter Results . . . . . . . . . . . . . . . . . . . . 120
5.2 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.1 Introduction to Fubini/Tonelli Theorems . . . . . . . . . . . . . . . 123
5.2.2 Integrals of Characteristic Functions . . . . . . . . . . . . . . . . . . 123
5.3 Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.3.1 Generalizing Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . 130
5.3.2 Fubini’s Theorem on σ 0 (X × Y ) . . . . . . . . . . . . . . . . . . . . 131
5.4 Tonelli’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.4.1 Tonelli’s Theorem on σ 0 (X × Y ) . . . . . . . . . . . . . . . . . . . . 135
5.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6 Two Applications of Fubini/Tonelli 140

6.1 Lebesgue-Stieltjes Integration by Parts . . . . . . . . . . . . . . . . . . . . 140
6.1.1 Functions of Bounded Variation . . . . . . . . . . . . . . . . . . . . . 141
6.1.2 Lebesgue-Stieltjes integration by parts . . . . . . . . . . . . . . . . . 145
6.2 Convolution of Integrable Functions . . . . . . . . . . . . . . . . . . . . . . 147
7 The Fourier Transform 152

7.1 Integration of Complex-Valued Functions . . . . . . . . . . . . . . . . . . . 152
7.2 Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.3 Properties of Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . 158
7.4 Fourier-Stieltjes Inversion of φF (t) . . . . . . . . . . . . . . . . . . . . . . . 165
7.5 Fourier Inversion of Integrable φF (t) . . . . . . . . . . . . . . . . . . . . . . 170
7.5.1 Integrability vs. Decay at ±∞ . . . . . . . . . . . . . . . . . . . . . 170
7.5.2 Fourier Inversion: From Integrable φF (t) to f (x) . . . . . . . . . . . 172
7.6 Continuity Theorem for Fourier Transforms . . . . . . . . . . . . . . . . . . 174
Contents ix
8 General Measure Relationships 180

8.1 Decomposition of Borel Measures on (R, B(R), m) . . . . . . . . . . . . . . 180
8.2 Decomposition of σ-Finite Measures . . . . . . . . . . . . . . . . . . . . . . 185
8.2.1 Signed Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
8.2.2 The Hahn and Jordan Decompositions . . . . . . . . . . . . . . . . . 188
8.2.3 The Radon-Nikodým Theorem . . . . . . . . . . . . . . . . . . . . . 191
8.2.4 The Lebesgue Decomposition Theorem . . . . . . . . . . . . . . . . . 200
9 The Lp Spaces 204

9.1 Introduction to Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 204
9.2 The Lp (X)-Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
9.3 Approximating Lp (X)-Functions . . . . . . . . . . . . . . . . . . . . . . . . 218
9.4 Bounded Linear Functionals on Lp (X)-Spaces . . . . . . . . . . . . . . . . 219
9.5 Hilbert Space: A Special Case of p = 2 . . . . . . . . . . . . . . . . . . . . 225
References 231
Index 235
Taylor & Francis
Preface
The idea for a reference book on the mathematical foundations of quantitative finance has
been with me throughout my professional and academic careers in this field, but the com-
mitment to finally write it didn’t materialize until completing my introductory quantitative
finance book in 2010.
My original academic studies were in pure mathematics in a field of mathematical anal-
ysis, and neither applications generally nor finance in particular were then even on my
mind. But on completion of my degree, I decided to temporarily investigate a career in ap-
plied math, becoming an actuary, and in short order became enamored with mathematical
applications in finance.
One of my first inquiries was into better understanding yield curve risk management, ulti-
mately introducing the notion of partial durations and related immunization strategies. This
experience led me to recognize the power of greater precision in the mathematical specifica-
tion and solution of even an age-old problem. From there my commitment to mathematical
finance was complete, and my temporary investigation into this field became permanent.
In my personal studies, I found that there were a great many books in finance that
focused on markets, instruments, models, and strategies, which typically provided an in-
formal acknowledgment of the background mathematics. There were also many books on
mathematical finance focusing on more advanced mathematical models and methods, and
typically written at a level of mathematical sophistication requiring a reader to have signif-
icant formal training and the time and motivation to derive omitted details.
The challenge of acquiring expertise is compounded by the fact that the field of quanti-
tative finance utilizes advanced mathematical theories and models from a number of fields.
While there are many good references on any of these topics, most are again written at
a level beyond many students, practitioners and even researchers of quantitative finance.
Such books develop materials with an eye to comprehensiveness in the given subject matter,
rather than with an eye toward efficiently curating and developing the theories needed for
applications in quantitative finance.
Thus the overriding goal I have for this collection of books is to provide a complete and
detailed development of the many foundational mathematical theories and results one finds
referenced in popular resources in finance and quantitative finance. The included topics
have been curated from a vast mathematics and finance literature for the express purpose
of supporting applications in quantitative finance.
I originally budgeted 700 pages per book, in two volumes. It soon became obvious
this was too limiting, and two volumes ultimately turned into ten. In the end, each book
was dedicated to a specific area of mathematics or probability theory, with a variety of
applications to finance that are relevant to the needs of financial mathematicians.
My target readers are students, practitioners, and researchers in finance who are quanti-
tatively literate and recognize the need for the materials and formal developments presented.
My hope is that the approach taken in these books will motivate readers to navigate these
details and master these materials.
Most importantly for a reference work, all 10 volumes are extensively self-referenced.
The reader can enter the collection at any point of interest, and then using the references
xi
xii Preface
cited, work backward to prior books to fill in needed details. This approach also works for
a course on a given volume’s subject matter, with earlier books used for reference, and for
both course-based and self-study approaches to sequential studies.
The reader will find that the developments herein are presented at a much greater level
of detail than most advanced quantitative finance books. Such developments are of necessity
typically longer, more meticulously reasoned, and therefore can be more demanding on the
reader. Thus before committing to a detailed line-by-line study of a given result, it can be
more efficient to first scan the derivation once or twice to better understand the overall logic
flow.
I hope the scope of the materials, and the additional details presented, will support your
journey to better understanding.
I am grateful for the support of my family: Lisa, Michael, David, and Jeffrey, as well as
the support of friends and colleagues at Brandeis International Business School.
Robert R. Reitano
Brandeis International Business School
Author
Robert R. Reitano is Professor of the Practice of Finance at the Brandeis International

Business School where he specializes in risk management and quantitative nance. He pre-
viously served as MSF Program Director, and Senior Academic Director. He has a PhD in
mathematics from MIT, is a fellow of the Society of Actuaries, and a Chartered Enterprise
Risk Analyst. Dr. Reitano consults in investment strategy and asset/liability risk manage-
ment, and previously had a 29-year career at John Hancock/Manulife in investment strategy
and asset/liability management, advancing to Executive Vice President & Chief Investment
Strategist. His research papers have appeared in a number of journals and have won an
Annual Prize of the Society of Actuaries and two F.M. Redington Prizes of the Investment
Section of the Society of the Actuaries. Dr. Reitano serves on various not-for-prot boards
and investment committees.
xiii
Taylor & Francis
Introduction
Foundations of Quantitative Finance is structured as follows:

Book I: Measure Spaces and Measurable Functions
Book II: Probability Spaces and Random Variables
Book III: The Integrals of Riemann, Lebesgue, and (Riemann-)Stieltjes
Book IV: Distribution Functions and Expectations
Book V: General Measure and Integration Theory
Book VI: Densities, Transformed Distributions, and Limit Theorems
Book VII: Brownian Motion and Other Stochastic Processes
Book VIII: Itô Integration and Stochastic Calculus 1
Book IX: Stochastic Calculus 2 and Stochastic Differential Equations
Book X: Classical Models and Applications in Finance
The series is logically sequential. Books I, III, and V develop foundational mathematical
results needed for the probability theory and finance applications of Books II, IV, and
VI, respectively. Books VII, VIII, and IX then develop results in the theory of stochastic
processes. While these latter three books introduce ideas from finance as appropriate, the
final realization of the applications of these stochastic and other models to finance is deferred
to Book X.
This Book V, General Measure and Integration Theory, generalizes the results of Books
I and III on these respective topics. Because the Book I development is so foundational to
the materials of this book, Chapter 1 sets out to review the key ideas and results that will be
needed. Beginning with the definitional framework of measure spaces and the Lebesgue and
Borel measure constructions, the general extension theory is then summarized and applied
to the constructions of finite product measure spaces, general Borel measure spaces on Rn ,
and finally to infinite dimensional product probability spaces.
The study of measurable functions is undertaken in Chapter 2. Following the develop-
ment of a number of basic properties, the measurability of various limits of measurable
functions is established, as is the approximation of measurable functions with simple func-
tion sequences. These approximations will be essential in the development of an integration
theory. As in Book III, deeper properties can be derived from simpler results and appropri-
ate limiting processes. An important tool is developed next, known as the monotone class
theorem. After properties of monotone classes and their relationships to sigma algebras
are developed, the important functional monotone class theorem is derived. This theorem
provides a powerful approach to verify, through a sequence of simpler steps, that a given
statement or identity applies to all bounded measurable functions.
Lebesgue-Steiltjes integration theory is developed in Chapter 3, largely following the
approach and results of the Lebesgue development of Book III in this more abstract set-
ting. Starting with simple functions, the approach is axiomatic in that the integrals of such
functions are defined and shown to be consistent. The sequential approach of the Lebesgue
theory is streamlined somewhat, omitting the development of integrals of bounded measur-
able functions and instead initiating the generalization from simple functions to nonnegative
measurable functions. Key results for bounded functions are instead developed in the exer-
cises and later in the chapter. Important “integration to the limit” results are derived for
xv
xvi Introduction
the integrals of nonnegative functions, including Fatou’s lemma and Lebesgue’s monotone
convergence theorem, which are then instrumental in the development of the properties of
such integrals. As an application of this theory, a Chapter I.7 result is generalized, address-
ing countable additivity of the set function defined on measurable rectangles of product
measure spaces.
This integration theory is then applied to derive integrals of general measurable func-
tions and their properties, along with the integration to the limit results of Beppo-Levi’s
theorem and Lebesgue’s dominated convergence theorem. The bounded convergence theo-
rem is then derived, as is an integration to the limit result not seen in Book III, called the
uniform integrability convergence theorem. The Leibniz integral rule on the derivatives of
parametrically defined integrals is then studied, generalizing the result of the Riemann the-
ory of Book III. The chapter ends with a discussion of when the Lebesgue-Stieltjes integrals
of this chapter agree with the Riemann-Stieltjes integrals of Book III.
Chapter 4 develops various results on change of variables in Lebesgue-Stieltjes integrals.
The first investigation is fundamental in probability theory and addresses integrals whose
measures are defined by integrals of measurable functions with respect to other measures.
In probability theory, for example, this measure would be the Borel measure on R induced
by the density function of a given random variable.
A more general realization of this idea is studied next, related to measurable transfor-
mations and their induced measures on range spaces. Integrals on the domain and range
spaces can then be related using this transformation and related measures. Special cases
of transformations are then studied in detail, beginning with linear transformations on Rn ,
and then turning to continuously differentiable transformations.
Integrals on product spaces are studied in Chapter 5. The key question is, echoing a Rie-
mann result of Book III, when can a multiple integral be evaluated using so-called iterated
integrals, which integrate one variable at a time? Such results prove to be related to the
product space sigma-algebra used, whereby the choices contrasted are the complete sigma
algebra and the smallest sigma algebra that contains the defining algebra. In large measure,
the results look quite similar, but there are details related to the need to qualify certain
statements with “almost everywhere.” The fundamental results here are then Fubini’s the-
orem, applicable to integrable functions, and Tonelli’s theorem, applicable to nonnegative
measurable functions. Examples from earlier books are then used to illustrate the theory,
though the next two chapters find deeper applications.
In Chapter 6, the first application of the Chapter 5 theory is to Lebesgue-Stieltjes
integration by parts. For this, the notion of a “signed measure” is introduced and seen to
possess all the properties of a measure other than nonnegativity. The integration theory is
then developed for such measures when induced by functions of bounded variation, echoing
the Riemann-Stieltjes work of Book III. The second application of the Chapter V theory
is to the study of the integrability of the convolution of integrable functions, proving that
such functions are indeed integrable.
The theory of Fourier transforms is developed in Chapter 7, a theory that will have
its most important applications in Book VI in the guise of characteristic functions. After
introducing the notion of the integral of complex-valued functions, the Fourier transform of
integrable functions and finite Borel measures is defined and various properties developed. In
particular, there are key connections between the decay at infinity of a function (or measure)
and the differentiability of its Fourier transform. Conversely, there are connections between
the existence and integrability of the derivatives of a function and the rate of decay at
infinity of its Fourier transform.
Fourier inversion is then studied, whereby one recovers the measure from its Fourier
transform, or, recovers the integrable function in the case where this transform is integrable.
Finally, a continuity theory is studied, which addresses the connection between the weak
Introduction xvii
convergence of probability measures and the pointwise convergence of their Fourier trans-
forms. An application to Poisson’s limit theorem is made, while the more powerful applica-
tions are deferred to Book VI.
Chapter 8 investigates general measure relationships and various decompositions of mea-
sures vis-a-vis other measures. The chapter begins with an example of the decomposition of
a Borel measure on R into measures that have characteristic properties relative to Lebesgue
measure. These characteristic properties are then generalized, whereby given a measure µ
we identify what it means for other measures to be absolutely continuous, or, mutually sin-
gular, relative to µ. To generalize the Borel decomposition, the signed measures introduced
in Chapter 6 are now studied in some detail. Various results are developed, such as the
Hahn decomposition, which address the positive and negative sets underlying a signed mea-
sure, and the Jordan decomposition, which decomposes a signed measure into a difference
of measures.
Perhaps the most famous and useful result on relationships between measures is the
Radon-Nikodým theorem. It states that if a measure υ is absolutely continuous with re-
spect to a measure µ, then υ is given by the µ-integral of a measurable function. This key
result will have an important application in Chapter 9, and then in the stochastic pro-
cess studies of Books VII–IX. The chapter ends with the Lebesgue decomposition theorem,
which generalizes the chapter’s Borel example to σ-finite measures.
The final Chapter 9 investigates Banach spaces, adding to their introduction in Book
III by studying the so-called Lp spaces of variously integrable or bounded measurable func-
tions, and their properties. Bounded linear functionals on Lp spaces are investigated and
characterized by the Riesz representation theorem, which is proved with the aid of the
Radon-Nikodým theorem. The special space of L2 , which is a Hilbert space, is addressed.
I hope this book and the other books in the collection serve you well.
Notation 0.1 (Referencing within FQF Series) To simplify the referencing of results
from other books in this series, we use the following convention.
A reference to “Proposition I.3.33” is a reference to Proposition 3.33 of Book I, while
“Chapter III.4” is a reference to Chapter 4 of Book III, and “II.(8.5)” is a reference to
formula (8.5) of Book II, and so forth.
Taylor & Francis
1
Measure Spaces
This chapter summarizes some of the more important results from Book I on the various
constructions of a measure space (X, σ(X), µ). The importance of such constructions is
twofold:
• The requirements for a measure space (X, σ(X), µ) are quite demanding and, in fact,
so demanding that it is not immediately clear that such objects even exist.
• Given general existence, a measure space with particular properties will often be re-
quired. What if any restrictions are needed on these properties to ensure existence? For
example, can we create a measure on R so that the measure of any interval [a, b] is given
by f (b) − f (a) for a given function f ?
For existence, the space X is generally a collection of points, sometimes with other
special properties, but any set X can in theory be used. By Definitions I.2.1 and I.2.5, the
sigma algebra σ(X) is a collection of subsets of X with the following properties:
Definition 1.1 (Algebra of sets, sigma algebra) A collection A of sets from a space
X is called a algebra of sets on X:
S
1. If A ∈ A and B ∈ A, then the union A B ∈ A where:
[
A B ≡ {x|x ∈ A or x ∈ B}.
2. If A ∈ A, then A
e ∈ A, where the complement of A is defined:
e ≡ {x ∈ X|x ∈
A / A}.
The complement of A is also denoted Ac .
A collection σ(X) of sets from a space X is called a sigma algebra on X if σ(X) is
an algebra of sets, and:
S∞
3. If {Ai }∞
i=1 ⊂ σ(X), then i=1 Ai ∈ σ(X).
Thus algebras are closed under finite unions and complementation, and also finite inter-
sections by De Morgan’s laws of Exercise I.2.2. A sigma algebra is closed under countable
unions, and thus again countable intersections. In addition, algebras, and hence also sigma
algebras,
Smust contain X and the empty set ∅. For example, A ∈ A implies A e ∈ A and thus
both A A e = X ∈ A and A T A e = ∅ ∈ A.
The existence of algebras and sigma algebras can be demonstrated starting with any
collection of sets. Such a collection then obtains both an algebra and a sigma algebra by
defining A as the smallest algebra that contains this collection, and σ(X) as the smallest
such sigma algebra. This construction works because by Proposition I.2.8, the intersection
of any collection of algebras is an algebra, and similarly for sigma algebras. To ensure that
we do not have a vacuous intersection in this construction, note that the power set P(X),
which contains all subsets of X, is one such algebra and sigma algebra.
An important example of such constructions are the Borel sigma algebras.
DOI: 10.1201/9781003264576-1 1
2 Measure Spaces
Example 1.2 (Borel sigma algebra) By Definition I.2.13, the Borel sigma algebra
n
B(R ) on Rn is the smallest sigma algebra that contains the open sets of Rn . More generally,
if X is a topological space (Definition 1.5), the Borel sigma algebra B(X) is defined as the
smallest sigma algebra on X that contains all the open sets as defined by the topology T
on X.
By De Morgan’s laws, Borel sigma algebras can also be defined as the smallest sigma
algebras that contain the closed sets, but the above formulation is conventional.
Example 1.3 (Semi-algebra, algebra, and sigma algebra generated by E) Gene-

ralizing Example 1.2, if E is any collection of subsets of a space X, then the smallest algebra
A(E) and smallest sigma algebra σ(E) that contain E, are well-defined. The same is true
for the smallest semi-algebra A0 (E). See Definition 1.14.
It is also common to refer to these collections as the semi-algebra, algebra, and sigma
algebra generated by E.
Recall the definition of open and closed sets. For more background on open and closed
sets in various contexts, see Chapter 4 of Reitano (2010) or Section III.4.3.2, and also
Dugundji (1970) or Gemignani (1967).
Definition 1.4 (Open sets in R, Rn , and metric X) A set E ⊂ R is called open if for
any x ∈ E there is an open interval containing x that is also contained in E. In other words,
there exists 1 , 2 > 0 so that (x − 1 , x + 2 ) ⊂ E. There is no loss of generality by requiring
1 = 2 .
A set E ⊂ Rn is called open if for any x ∈ E there is an open ball Br (x) about x of
radius r > 0:
Br (x) ≡ {y| |x − y| < r},
so that Br (x) ⊂ E. Here |x − y| denotes the standard metric on Rn :
hXn i1/2
2
|x − y| ≡ (xi − yi ) .
i=1
More generally, if X is a metric space with metric d, a set E ⊂ X is called open if for
any x ∈ E there is an open ball Br (x) about x of radius r > 0:
Br (x) = {y|d(x, y) < r},
so that Br (x) ⊂ E.
In all cases, a set F is called closed if Fe, the complement of F, is open.
The above notions of an open set reflect the natural metric |·| on R and Rn , where
d(x, y) ≡ |x − y| , or more generally a metric d on a space X. For more on metrics, see the
above references and Section III.4.3.1.
Open sets can also be defined without metrics, and this notion allows one to then define
a continuous function. See Proposition 2.10.
Definition 1.5 (Topology) A topology on a set X is a collection of subsets, denoted by

T and called the “open sets,” so that:
1. ∅, X ∈ T ;
2. If {Aα }α∈I S ⊂ T , where the index set I is arbitrary (finite, countably infinite, uncount-
able), then α∈I Aα ∈ T ;
Tn
3. If {Ai }ni=1 ⊂ T , then i=1 Ai ∈ T .
Measure Spaces 3
Exercise 1.6 (Natural topology on Rn ) Prove that the collection of open sets on R, as
defined in Definition 1.4, is a topology on R, and then generalize to Rn . These topologies are
often referred to as the natural topologies on R and Rn , and also the Euclidean topolo-
gies. The same result is true for the metric space (X, d), and this is called the topology
induced by the metric d. Hint: Recall Definition III.4.39 for definitional properties of
metrics.
Returning to the existence question for a measure space (X, σ(X), µ), since it is relatively
easy to construct various (X, σ(X)) pairs, the real question of existence relates to the
existence of a measure µ on σ(X). Recalling Definition I.2.23:
Definition 1.7 (Measure space) Let X be a set and σ(X) a sigma algebra of subsets of
X. A measure µ on X is a nonnegative set function defined on σ(X), taking values in
+
the nonnegative extended real numbers R ≡ R+ ∪{∞}, and which satisfies the following
properties:
1. µ(∅) = 0;
2. Countable Additivity: If {Ai }∞i=1 is aT
countable collection of pairwise disjoint sets
in the sigma algebra σ(X), meaning Aj Ak = ∅ if j 6= k, then:
[∞ X∞
µ Ai = µ (Ai ) . (1.1)
i=1 i=1
In such a case, the sets in σ(X) are said to be measurable, and sometimes µ-
measurable, and the triplet (X, σ(X), µ) is called a measure space.
Definition 1.8 (Lebesgue, Borel measure spaces) For these special measure spaces, a
third requirement is added:
• For Lebesgue measure µ ≡ m and X = R, Definition I.2.21 also adds the

following requirement:
3. For any interval I ≡ ha, bi , whether open, closed, or semi-closed,
m(I) = |I| = b − a,
the length of I.
By the construction in Section I.7.6, this criterion
Qn generalizes for Lebesgue measure on
Rn , that for any measurable rectangle R ≡ i=1 hai , bi i , whether such intervals are
open, closed, or semi-closed:
Yn
m(R) = |R| = (bi − ai ) ,
i=1
the volume of R.
n
• For a Borel measure µ, X = Rn and σ(X) = B(R ), and Definitions I.5.1 and
I.8.4 add the requirement:
n
3’. For any compact set A ∈ B(R ), µ(A) < ∞.
As noted above, while it is relatively easy to construct a sigma algebra on given set, it
is by no means apparent how one then defines a measure on this sigma algebra.
4 Measure Spaces
Remark 1.9 (Additional properties of measures) A few comments on the above def-
initions:
• Lebesgue translation invariance: For Lebesgue measure, it follows from item 3 that
m is translation invariant on intervals:
m(I) = m(I + x),
where I + x ≡ {x + y|y ∈ I} for any x. This property is then satisfied for all Lebesgue
measurable sets on R by Propositions I.2.28 and I.2.35, and analogously generalizes to
Rn by the construction of Chapter I.7.
• Borel measures: The notion of a Borel measure is not standardized, and it is not
uncommon to see other restrictions added in place of item 3’, such as the inner and
outer regularity properties of Proposition I.5.29. Measure spaces that also satisfy item
3’ are often called Radon measures, after Johann Radon (1887–1956).
• Finite additivity: Setting Ai = ∅ for i > n in (1.1), it follows from this definition that
all measures also satisfy finite additivity for pairwise disjoint sets:
[n Xn
µ Ai = µ (Ai ) . (1.2)
i=1 i=1
• Monotonicity: Measures are monotonic, which means that if A ⊂ B with both mea-
surable, then:
µ (A) ≤ µ (B) . (1.3)
S
This follows from finite additivity since B = A (B − A) as a disjoint union.
• Subadditivity: When {Ai }N i=1 is a finite (N < ∞) or countable (N = ∞) collection of

not necessarily pairwise disjoint sets in the sigma algebra σ(X), then a measure µ is
in general subadditive:
[ X
N N
µ Ai ≤ µ (Ai ) . (1.4)
i=1 i=1
This follows from

Sm Proposition
Sm I.2.20 that there exists pairwise disjoint {A0i }N
i=1 with
0 0
Ai ⊂ Ai and i=1 Ai = i=1 Ai for all m. Thus (1.4) follows from finite/countable
additivity and monotonicity.
The existence of measures is more readily justified when X has a finite or countable
number of points.
Exercise 1.10 (Finite/countable X) Verify that if X is a finite or countable set, then

one can construct a measure space (X, P(X), µ), where P(X) is the power set of X, defined
to contain all subsets of X. Hint: If X = {xi }ni=1 let µ(xi ) = 1/n, if X = {xi }∞
i=1 let
µ(xi ) = 2−n , then µ(X) = 1, and such measures are called probability measures and
(X, σ(X), µ) is called a probability space.
For uncountable sets X, the construction is of necessity more subtle and sometimes ob-
tains surprising results. In the next section, we review the special constructions of Lebesgue
and Borel measure spaces of Chapters I.2 and I.5. We then summarize the general construc-
tion theory of Chapter I.6, and in the following section, recall Book I applications of this
general framework to various special constructions.
Lebesgue and Borel Spaces on R 5
1.1 Lebesgue and Borel Spaces on R

Chapters I.2 and I.5 derived Lebesgue measure on R and then Borel measures on R. Each of
these derivations is summarized below, but suffice to say that they were fairly long, requiring
a number of both detailed and subtle steps toward the final result. While somewhat different
in their starting points and approaches, there was a certain redundancy in their development.
And frankly, that was the point of dedicating two chapters to these important derivations,
to illustrate that even with different starting points and objectives, many of the steps in
the construction of a measure were very similar.
Being the first development of its kind, the Lebesgue construction is really the special
case. Structurally, the Borel constructions introduced the framework that was generalized
and abstracted in Chapter I.6.
1.1.1 Starting Points

Both constructions started with a special collection of sets and a rudimentary definition of
the measure of such sets, and then an extension to the power sigma algebra P (R) of all
subsets of R using the notion of an outer measure.
• Lebesgue: The collection of sets is the open intervals G ≡ {(a, b)}, and the measure of
I = (a, b), denoted |I| , is defined as interval length:
|I| = b − a. (1.5)
If A ∈ P (R) , the Lebesgue outer measure of A, denoted m∗ (A), is defined in I.(2.4):

nX [ o
m∗ (A) = inf |In | A ⊂ In , , (1.6)
n n
where each In is an open interval, and |In | denotes its interval length.
• Borel: The collection of sets is the right semi-closed intervals A0 ≡ {(a, b]}, and the
measure of I = (a, b], denoted |I|F , is defined as the F -length of the interval:
|I|F = F (b) − F (a) , (1.7)
where F (x) is an increasing, right continuous function.

If A ∈ P (R) , the Borel outer measure of A, denoted µ∗F (A), is defined in I.(5.8):
nX [ o
µ∗A (A) = inf µA (An ) | A ⊂ An , (1.8)
n n
where An ∈ A and µA (An ) is F -length and defined in (1.7).

Here, A is defined as the collection of all finite disjoint unions of sets from A0 , and thus
µ∗A (A) can be equally well defined with all An ∈ A0 . In the next section, it will be seen
that for general constructions, we use the collection A.
The big question now becomes:

Can measures be created which extend the rudimentary notions of measure on G-sets
and A0 -sets to sigma algebras which contain these special sets, and if so, how?
6 Measure Spaces
1.1.2 Lebesgue Measure Space

Henri Lebesgue (1875–1941) wrote a collection of papers in the late 1800s and early 1900s
that collectively created the foundational Lebesgue measure space (R, M (R) , m) and an
associated integration theory.
When studying the Lebesgue development for the first time, many would likely expect
that the final result from the above starting point would be that m∗ is a measure on P (R) .
After all, it seems like the perfect generalization of interval length to more general sets,
defining m∗ (A) as the minimum of the total lengths of all “covers” of A by open intervals.
This is reinforced by the results of Proposition I.2.28 that m∗ is translation invariant, and
for any interval I, whether open, closed, or semi-closed, that m∗ (I) = |I| .
But this proves not to be. Remarkably, a construction was developed in 1905 by
Giuseppe Vitali (1875–1932) using the axiom of choice. Background details on this ax-
iom of set theory and Vitali’s construction can be found in Section I.2.3. He demonstrated
that the interval [0, 1] could be expressed as a countable disjoint union of translations of a
given set A0 : [∞
Aj = [0, 1],
j=0
and then by translation invariance derived that all such sets must have the same outer
measure.
This construction completely crushes any hope that m∗ is countably additive. Indeed,
it follows that depending on whether m∗ (A0 ) = 0 or m∗ (A0 ) > 0:
X∞
m∗ (Aj ) ∈ {0, ∞}.
j=1
∗
But in no case can it be that m ([0, 1]) = 1.
∗
P∞As m∗ is countably subadditive by Proposition I.2.29, it follows that m∗ (A0 ) > 0 and
∗
j=1 m (Aj ) = ∞. It then follows from this that m is not even finitely additive. So, the
problem here is even more serious than it first appears.
We are left with two possibilities:
1. There is a Lebesgue measure on P (R), but it is not m∗ ;

2. The set function m∗ is a Lebesgue measure, but on a smaller sigma algebra than P (R) .
There is no hope for item 1, since any proposal will encounter the Vitali construction.
To pursue item 2, there are two conventional approaches to identify the sets in P (R) on
which m∗ is indeed a measure, and these are discussed in Section I.2.4. The approach used
there was one that generalizes well to other constructions.
In Definition I.2.33, the approach of Constantin Carathéodory (1873–1950) was used,
an approach he developed for the general theory of outer measures.
Definition 1.11 (Lebesgue measurable set) A set A ⊂ R is said to be Lebesgue mea-

surable if it satisfies the Carathéodory criterion, that for any set E ⊂ R :
\ \
m∗ (E) = m∗ A E + m∗ A e E . (1.9)
The collection of Lebesgue measurable sets is denoted ML ≡ ML (R).
Thus by Carathéodory’s criterion, a set A will be deemed to be Lebesgue measurable

if A and its complement A e can be used to split any set E in a way that preserves finite
additivity of m∗ into two subsets. By Proposition I.2.36, the satisfaction of this criterion is
sufficient to ensure finite additivity generally, so m∗ is thus finitely additive on ML (R).
Lebesgue and Borel Spaces on R 7
It was then proved in a series of results culminating in Proposition I.2.39 that ML (R)
is a sigma algebra that contains the intervals and hence the Borel sigma algebra B(R), and
that m∗ is a Lebesgue measure on this sigma algebra. Lebesgue measure m is then defined
on ML (R) by:
m ≡ m∗ ,
and (R, ML (R), m) is a Lebesgue measure space.
It was also proved that ML (R) is a complete sigma algebra, and thus (R, ML (R), m)
is a complete measure space. See Definition 1.16. As noted above, ML (R) contains the
intervals, and m agrees with the specification in (1.5) on this collection.
An important corollary of the definition of the outer measure m∗ in (1.6) is that Lebesgue
measurable sets can be approximated well with open sets, and various classes of sets defined
using open sets. See Propositions I.2.42 for approximations, and I.2.43 for regularity of
Lebesgue measure.
1.1.3 Borel Measure Spaces

Borel measures and the associated measure spaces are named for Émile Borel (1871–1956),
an early pioneer in measure theory and probability theory. In many ways resembling the
Lebesgue construction of Chapter I.2, Borel measures on R are constructed in Chapter
I.5. Beginning with a study of such measures, it is proved in Proposition I.5.7 that every
Borel measure µ on R induces an increasing, right continuous function Fµ (x) on R with the
property that for all right semi-closed intervals (a, b]:
µ [(a, b]] = Fµ (b) − Fµ (a). (1.10)
To create a Borel measure space, it thus makes sense to investigate if any such increasing
and right continuous function F (x) can be used to induce a Borel measure µF . Indeed, the
answer is in the affirmative. Initially defining the set function F -length on right semi-closed
intervals (a, b] by (1.7), which is now compelled by (1.10), a series of results extends this
set function.
The first extension is to a set function µA on the algebra A, constructed as the collection
of all finite disjoint unions of right semi-closed intervals. This extension is defined additively
by disjointness. In Proposition I.5.13, µA proves to be a measure on this algebra.
Definition 1.12 (Measure on an algebra) A measure on an algebra, sometimes

called a pre-measure on an algebra, is a nonnegative, extended real-valued set func-
tion µ defined on an algebra A with the properties that:
1. µ(∅) = 0,
2. Countable Additivity: If {Aj }∞
j=1 is a countable collection of pairwise disjoint sets
S∞
in A with j=1 Aj ∈ A, then:
[ ∞ X∞
µ Aj = µ (Aj ) . (1.11)
j=1 j=1
An outer measure µ∗A is introduced as in (1.8), and as in the Lebesgue case, we utilize
the Carathéodory criterion in Definition I.5.16:
Definition 1.13 (Carathéodory measurability w.r.t. µ∗A ) Let F (x) be an increasing,

right continuous function and µ∗A the outer measure induced by µA as in (1.8). A set A ⊂ R
8 Measure Spaces
is Carathéodory measurable with respect to µ∗A , or simply µ∗A measurable, if for

any set E ⊂ R :
µ∗A (E) = µ∗A (A ∩ E) + µ∗A (A
e ∩ E). (1.12)
The collection of µ∗A measurable sets is denoted MµF (R).
It was proved in a series of results culminating in Proposition I.5.23 that MµF (R) is a
sigma algebra that contains the algebra A and hence the Borel sigma algebra B(R), and
that µ∗A is a Borel measure on this sigma algebra. The Borel measure µF is then defined on
MµF (R) by:
µF ≡ µ∗A ,

and R, MµF (R), µF is a Borel measure space.
It was also proved that MµF (R) is a complete sigma algebra, and thus
R, MµF (R), µF is a complete measure space. See Definition 1.16. Further, A ⊂ MµF (R)
as noted above, and µF agrees with the specification in (1.7) on A.
An important corollary of the definition of the outer measure µ∗A in (1.6) is that sets
in MµF (R) can be approximated well with various classes of sets defined using right semi-
closed intervals. See Proposition I.5.26 and Remark I.5.27 for approximations and why these
appear to differ structurally from the Lebesgue approximations, and Proposition I.5.29 for
regularity of Borel measures.
1.2 General Extension Theory

While the Lebesgue and Borel constructions had a lot of similarities, the latter development
better lends itself to generalization. The Lebesgue approach began with the class of open
intervals, so any generalization to a general set X almost certainly requires X to at least
have a topology so that the notion of “open” makes sense. The Borel approach starts with
the collection of right semi-closed intervals {(a, b]}, which superficially looks quite similar
to the collection of open intervals. However, this collection is different enough that one can
easily create an algebra A, and from this algebra, a sigma algebra is then “just” one step
away.
Chapter I.6 exploits these structures within the Borel development, first identifying the
essential property of the collection {(a, b]}, which allowed the simple step to an associated
algebra A.
Definition 1.14 (Semi-algebra of sets) A collection A0 of sets from a space X is called

a semi-algebra of sets on X:
1. If A01 , A02 ∈ A0 , then A01 A02 ∈ A0 , and thus this holds by induction for all finite
T
intersections.
f0 = Sn A0 .
2. If A0 ∈ A0 , then there exists disjoint {A0j }nj=1 ⊂ A0 so that A j=1 j
The collection A0 = A, an algebra, if in place of item 2 we have:
2’. If A ∈ A then A
e ∈ A.
The collection {(a, b]} is a semi-algebra, and by Exercise I.6.10, every semi-algebra A0
generates an algebra A defined as the collection of all finite disjoint unions of A0 -sets. This
General Extension Theory 9
implies that if we can define a set function on such a semi-algebra A0 , that the extension to
a measure on A will follow additively as in the Borel case.
Next, the notion of an outer measure is introduced. This definition captured the key
properties of the above special definitions that supported the conclusion that the collection
of Carathéodory measurable sets is a sigma algebra, and that µ∗ restricted to this
collection is a measure.
Definition 1.15 (Outer measure) Given a set X, a set function µ∗ defined on the power
sigma algebra σ(P (X)) of all subsets of X is an outer measure if:
1. µ∗ (∅) = 0.
2. Monotonicity: For sets A ⊂ B :
µ∗ (A) ≤ µ∗ (B).
3. Countable Subadditivity: Given a countable collection {Aj }∞

j=1 :
[ ∞ X∞
µ∗ Aj ≤ µ∗ (Aj ).
j=1 j=1
This definition looks nothing like the outer measure definitions in (1.6) and (1.8). But
it can be noted that the first results developed after the Book I introductions of m∗ and
µ∗A were to establish that these outer measures had the above properties. Further, these
properties were essential in the subsequent developments.
The results seen in the Borel development are then completely generalized in a series of
Chapter I.6 results, which worked backward starting with the Carathéodory extension
theorem 1 of Proposition I.6.2 and named for Constantin Carathéodory (1873–1950). It
asserts that outer measures always obtain complete measure spaces by the Carathéodory
criterion.
Definition 1.16 (Complete measure space) A measure space (X, σ(X), µ) is com-
plete if for any A ∈ σ(X) with µ(A) = 0, then B ∈ σ(X) for all B ⊂ A. It then follows by
monotonicity of measures that µ(B) = 0 for all such B.
Equivalently, if A, C ∈ σ(X) with µ(A) = µ(C), then B ∈ σ(X) for all C ⊂ B ⊂ A, and
then µ(B) = µ(A) for all such B.
Proposition 1.17 (Carathéodory extension theorem 1) Let µ∗ be an outer measure

defined on a set X. Denote by C(X) the collection of all subsets of σ(P (X)) that are
Carathéodory measurable with respect to µ∗ . That is, A ∈ C(X) if for all E ∈ σ(P (X)):
µ∗ (E) = µ∗ (A ∩ E) + µ∗ (A
e ∩ E). (1.13)
Then C(X) is a complete sigma algebra.

Further, if µ denotes the restriction of µ∗ to C(X), then µ is a measure, and thus
(X, C(X), µ) is a complete measure space.
The Hahn-Kolmogorov extension theorem of Proposition I.6.4 is named for Hans

Hahn (1879–1934) and Andrey Kolmogorov (1903–1987). It asserts that from a measure
on an algebra, one can always induce an outer measure defined exactly as in (1.8), and thus
a complete measure space by the above Carathéodory result. Specifically, given a measure
µA on an algebra A, we define the associated outer measure µ∗A on A ∈ P (X) by:
nX [ o
µ∗A (A) = inf µA (An ) | A ⊂ An , A n ∈ A . (1.14)
n n
10 Measure Spaces
Proposition 1.18 (Hahn–Kolmogorov extension theorem) Let A be an algebra of

sets on X and µA a measure on A in the sense of Definition 1.12. Then, µA gives rise to
an outer measure µ∗A on σ(P (X)) such that µ∗A (A) = µA (A) for all A ∈ A. In addition,
there exists a complete sigma algebra C(X) with A ⊂ C(X), and µ ≡ µ∗A is a measure on
C(X).
Thus, (X, C(X), µ) is a complete measure space and µ(A) = µA (A) for all A ∈ A.
The Hahn-Kolmogorov extension theorem then obtains approximations of C(X)-sets
by various collections derived from A-sets in Proposition I.6.5, exactly as in the Borel case.
The final step relates to the creation of a measure on an algebra. In the Borel case we
began with the notion of F -length in (1.7) defined on A0 , and generalized this definition to
A additively. Definition I.6.6 introduced the notion of a pre-measure on a collection of sets,
which we restrict here to a pre-measure on a semi-algebra A0 . It is an exercise to check that
F -length satisfies this definition.
Definition 1.19 (Pre-measure on a semi-algebra) Given a semi-algebra of sets A0 , a
set function µ0 : A0 → [0, ∞] is a pre-measure if:
1. Either: ∅ ∈ A0 and µ0 (∅) = 0, or:
Sn
10 . Finite additivity: If {Aj }nj=1 ⊂ A0 is a disjoint finite collection of sets and j=1 Aj ∈
A0 , then: [ n Xn
µ0 Aj = µ0 (Aj );
j=1 j=1
and:
2. Countable additivity: If {An }∞ 0
j=1 ⊂ A is a disjoint countable collection of sets and
S∞ 0
j=1 An ∈ A , then: [ ∞ X∞
µ0 Aj = µ0 (Aj ).
j=1 j=1
The final existence result of Proposition I.6.13 is again attributable to Constantin

Carathéodory (1873–1950).
Proposition 1.20 (Carathéodory extension theorem 2) Let A0 be a semi-algebra
and µ0 a pre-measure on A0 . Then, µ0 can be extended to a measure µA on the algebra
A, defined as the collection of all finite disjoint unions of sets in A0 , including ∅, if neces-
sary.
Beyond the existence theory summarized above, there is an associated uniqueness theory
of Proposition I.6.14 for sigma-finite measure spaces. Recall Definition I.5.34:
Definition 1.21 (σ-finite measure space) The measure space (X, σ(X), µ) is said to be
sigma finite, or σ-finite, if there exists a countable collection {Bj }∞ j=1 ⊂ σ(X) with
S∞
µ (Bj ) < ∞ for all j, and X = j=1 Bj . In this case, it is also said that the measure µ is
σ-finite.
A measure µA on an algebra A is sigma finite if such {Bj }∞j=1 ⊂ A.
It should be noted that there is no loss of generality in assuming that {Bj }∞

j=1 is pairwise
disjoint, recalling Proposition I.2.20 and monotonicity of measures.
The next result addresses the uniqueness of extensions from an algebra A is to σ(A),
the smallest sigma algebra that contains A. For this result, note that both σ(A) ⊂ C(X)
and σ(A) ⊂ σ(X), and thus both µ and µ0 are defined on σ(A).
Originally proved as Proposition I.6.14, see Example 2.40 for an alternative proof using
the monotone class theorem.
Measure Space Constructions 11
Proposition 1.22 (Uniqueness of Extensions to σ(A)) Let µA be a σ-finite measure

on an algebra A, and µ the extension of µA to the sigma algebra C(X) induced by the outer
measure µ∗A . Let µ0 be any other extension of µA from A to a sigma algebra σ(X).
Then for all B ∈ σ(A):
µ(B) = µ0 (B).
More generally, if µ, µ0 are σ-finite measures on a sigma algebra σ (X) , and µ(B) =
µ (B) for all B ∈ A for some algebra A ⊂ σ (X) , then µ = µ0 on σ(A).
0
1.3 Measure Space Constructions

With the powerful tools of Chapter I.6, one can create a complete measure space without
investigating all of the details seen in the Lebesgue and Borel developments.
The construction process is as follows:
1. Required Step (Choose a or b):
(a) Define a set function µ0 on a semi-algebra A0 which can be proved to be a pre-

measure.
(b) Define a set function µA on an algebra A which can be proved to be a measure.
The algebra A in 1.b can be the algebra generated by a given semi-algebra A0 or defined
independently of a semi-algebra.
2. “Free” Steps:
(a) From 1.a, if µ0 is a pre-measure on a semi-algebra A0 , then µ0 can be extended to

a measure µA on A, the algebra generated by A0 , by the Carathéodory extension
theorem 2. Alternatively from 1.b, we start with a measure µA on an algebra A.
(b) In either case from step 2.a, µA and A then generate an outer measure µ∗A
on σ(P (X)), the power set sigma algebra on X. This follows from the Hahn-
Kolmogorov extension theorem with µ∗A as defined in (1.14), and this outer measure
then satisfies the conditions specified in the Carathéodory extension theorem 1.
(c) The collection of sets that are Carathéodory measurable with respect to µ∗A , as
defined in (1.13), is then a complete sigma algebra C(X), and the restriction of µ∗A
to C(X) is a measure µ. Hence, (X, C(X), µ) is a complete measure space. Moreover,
A ⊂ C(X) and µ extends µA in the sense that µ(A) = µA (A) for all A ∈ A.
This framework was applied in Chapters I.7–I.9 to obtain measure spaces in three im-
portant contexts. While still requiring a certain amount of effort to obtain the required step,
both the reader and the author were likely equally pleased to not have to then explicitly
derive all the results obtained in the free steps.
1.3.1 Finite Products of Measure Spaces

Given measure spaces {(Xi , σ(Xi ), µi )}ni=1 , the goal of Chapter I.7 was to create a product
space and product measure. See Section 5.1 for more on product spaces.
12 Measure Spaces
Definition 1.23 (Product space and set function) Given measure spaces {(Xi , σ(Xi ),
µi )}ni=1 , the product space:
Yn
X= Xi ,
i=1
is defined as:
X = {x ≡ (x1 , x2 , ..., xn )|xi ∈ Xi }. (1.15)
A measurable rectangle in X is a set A:
Yn
A= Ai = {x ∈ X|xi ∈ Ai }, (1.16)
i=1
0
where Ai ∈ σ(Xi ). We denote the collection of measurable
Qn rectangles in X by A .
The product set function µ0 is defined on A = i=1 Ai ∈ A0 by:
Yn
µ0 (A) = µi (Ai ), (1.17)
i=1
where we explicitly define 0 · ∞ = 0.
Unsurprisingly given the notation, the collection A0 proves to be a semi-algebra, and the
set function µ0 can be extended additively to a set function µA on the associated algebra
A of finite disjoint unions of A0 -sets. The derivation that µA so defined is a measure on
this algebra is subtle. This is in part due to the complexity of A-sets, and also that the
necessary results must be derived with (1.17) and properties of the measures {µi }ni=1 .
While generalized in Section 3.2.4, the Book I proof of countable additivity of µA required
that {µi }ni=1 be σ-finite measures. This obtained σ-finiteness of µA on A and of the resulting
measure µX on the complete sigma algebra, there denoted σ(X). The product measure space
of Proposition I.7.20 is then denoted (X, σ(X), µX ) . Qn
As noted in Notation I.7.21, it is common to express µX = i=1 µi . This notation also
reflects the fact that µX Q is an extension of µA from A to σ(X) and thus extends µ0 from
n
A0 to σ(X). So, for A = i=1 Ai ∈ A0 , (1.17) can be expressed:
Yn
µX (A) = µi (Ai ). (1.18)
i=1
When {(Xi , σ(Xi ), µi )}ni=1 = {(R, MFi (R), µFi )}ni=1 are Borel measure spaces, which
n Qn
are sigma finite by item 30 of Definition 1.8,Q
the final measure space (Rn , M(R ), i=1 µFi )
n n n n
contains a Borel measure space (Rn , B(R ), i=1 µFi ) since B(R ) ⊂ M(R ), and µX proves
to be a Borel measure.
However, there are Borel measures on Rn other than these product measures, and this
is the subject we discuss next.
1.3.2 Borel Measures on Rn

Following the development of Chapter I.7, the next application of the Chapter I.6 ex-
n
tension theory is to general Borel measure spaces denoted (Rn , B(R ), µ). Generalizing the
1-dimensional case of Chapter I.5, any such Borel measure µ induces a multivariate function
Fµ : Rn → R, which is continuous from above, and n-increasing. This is seen in Proposition
I.8.10 for finite Borel measures, and Proposition I.8.12 in the general case, where:
Definition 1.24 (Continuous from above; n-increasing ) Given a function Fµ :

Rn → R:
Measure Space Constructions 13
1. Fµ is continuous from above at x = (x1 , ..., xn ) if given a sequence x(m) =

(m) (m) (m)
(x1 , ..., xn ) with xi ≥ xi for all i and m, and x(m) → x as m → ∞, then:
Fµ (x) = limm→∞ Fµ (x(m) ). (1.19)
We say that Fµ is continuous from above if the above property is true for all x.
2. Fµ satisfies
Qnthe n-increasing condition if given any bounded right semi-closed rect-
angle A = i=1 (ai , bi ] : X
sgn(x)Fµ (x) ≥ 0. (1.20)
x
Each x = (x1 , ..., xn ) in the summation is one of the 2n vertices of A, so xi = ai or
xi = bi , and sgn(x) equals −1 if the number of ai -components of x is odd, and equals
+1 otherwise.
It was derived in Propositions

Qn I.8.9 (finite Borel measures) and I.8.12 (general Borel
measures) that the measure µ [ i=1 (ai , bi ]] of a bounded right semi-closed rectangle can be
expressed in terms of this induced function:
hYn i X
µ (ai , bi ] = sgn(x)Fµ (x), (1.21)
i=1 x
where this summation is defined above.

n
Exercise 1.25 (Product functions and measures) Qn If {Fi (xi )}i=1 are increasing and
right continuous functions on R, show that F (x) ≡ i=1 Fi (xi ) is continuous from above
and n-increasing. Hint: Prove that the expression in (1.21) can be rewritten:
hYn i Yn
µ (ai , bi ] = (Fi (bi ) − Fi (ai )) .
i=1 i=1
Qn
Note that when all Fi (xi ) = xi , that µ [ i=1 (ai , bi ]] reduces to the Lebesgue measure of this
rectangle: hYn i Yn
µ (ai , bi ] = (bi − ai ) .
i=1 i=1
Given the insights of this study of general Borel measures, the Chapter I.6 extension
n
theory is then applied to investigate if a Borel measure space (Rn , B(R ), µF ) can be con-
structed from a continuous from above and n-increasing function F : Rn → R. Given such
F (x), we begin by defining the class of bounded right semi-closed rectangles:
n
n
Yn o
A0B ≡ A ∈ B(R )|A = (ai , bi ], with − ∞ < ai ≤ bi < ∞ ,
i=1
and on A0B define the set function µ0 as in (1.21). It can be checked that A0B is not a
semi-algebra (Hint: consider A),
e so this will need to be addressed later in the development.
It is then proved in Proposition I.8.13 that µ0 is finitely additive and countably subad-
ditive on A0B . Since A0B is not a semi-algebra, Carathéodory’s extension theorem 2 cannot
be directly applied. Instead, the set function µ∗F is defined on A ⊂ Rn by:
nX [ o
µ∗F (A) = inf µ0 (An ) | A ⊂ An , An ∈ A0B , (1.22)
n
and proved in Proposition I.8.14 to be an outer measure by Definition 1.15.

Carathéodory’s extension theorem 1 now assures the existence of a complete measure space
n
(Rn ,MF (R ), µF ), and in Proposition I.8.15, it is proved that µF extends the set function
n
µ0 defined on A0B by showing that A0B ⊂MF (R ) and µ∗F (A) = µ0 (A) for all A ∈ A0B .
14 Measure Spaces
Example 1.26 (Borel measures as Borel/Lebesgue product measures) When F (x)

n
is the product function of Exercise 1.25, then (Rn ,MF (R ), µF ) here is the product
n
measure space obtained from {(R, MµFi (R), µFi )}i=1 , while when all Fi (xi ) = xi , then
n
(Rn ,MF (R ), µF ) is the Lebesgue product space obtained from {(R, ML (R), mi )}ni=1 where
mi = m for all i. See Section I.7.6 for Lebesgue and Borel product spaces, and Proposition
I.8.16, for uniqueness of extensions.
1.3.3 Infinite Products of Probability Spaces

For infinite products of measure spaces, the theory essentially requires that these measure
spaces be probability spaces. And while a somewhat more general derivation is possible,
Chapter I.9 ultimately assumes that these probability spaces are defined on R, the applica-
tion of greatest interest.
Definition 1.27 (Infinite product space and set function) Given probability spaces:
{(Xi , σ(Xi ), µi )}∞

i=1 ,
Q∞
define the product space X = i=1 Xi by:
X = {(x1 , x2 , ...)|xi ∈ Xi }. (1.23)
A finite dimensional measurable rectangle A in X, also called a cylinder set, is

defined for any n and n-tuple of positive integers J = (j(1), j(2), ..., j(n)) by:
A = {x ∈ X|xj(i) ∈ Aj(i) }, (1.24)
where
Qn Aj(i) ∈ σ(Xj(i) ). The cylinder set in (1.24) is said to be defined by J and
0
i=1 j(i) , and the collection of cylinder sets in X is denoted by A .
A
0 0
Qn product set function µ0 is defined on A as follows. If A ∈ A is defined by J
The
and i=1 Aj(i) , then:
Yn
µ0 (A) = µj(i) (Aj(i) ). (1.25)
i=1
The above restriction to probability spaces stems from the need to have µ0 (A) in (1.25)
well defined. For example, if J = (1, 2, ..., n) then:
A = {xi ∈ Ai } = {xi ∈ Ai , xn+1 ∈ Xn+1 }.
For µ0 (A) to be well defined requires that:

Yn Yn
µi (Ai ) = µn+1 (Xn+1 ) µi (Ai ),
i=1 i=1
and so µn+1 (Xn+1 ) = 1 is derived unless one of these sets has infinite or zero measure.
As the notation suggests, A0 so defined is a semi-algebra. Further, µ0 extends additively
to µA on the associated algebra A, and µA proves to be finitely additive and countably
subadditive. For countable additivity, the algebra A needed to be enlarged to A+ , and all
Xi were then restricted to R.
Definition 1.28 (Product space; general cylinder sets: A+ ) Given probability spaces
∞
{(R,
Q∞ B(R), µi )}i=1 , where B(R) denotes the Borel sigma algebra, the product space R =
N
i=1 Ri is defined by:

RN = {(x1 , x2 , ...)|xi ∈ R}.
Continuity of Measures 15
A general finite dimensional measurable rectangle or general cylinder set

A ⊂ RQN
is defined
for any n-tuple of positive integers J = (j(1), j(2), ..., j(n)) and
n
H∈B i=1 j(i) by:
R
A = {x ∈ RN |(xj(1) , xj(2) , ...xj(n) ) ∈ H}. (1.26)

Qn n
Here, B i=1 Rj(i) = B (R ) denotes the finite dimensional product space Borel sigma
algebra associated with {(R, B(R), µj(i) )}ni=1 , a sigma subalgebra of the above denoted σ(X).
The cylinder set in (1.26) will be said to be defined by H and J, and the collection of
general cylinder sets in RN is denoted by A+ .
The cylinder set A can also be characterized in terms of the projection mapping:
Yn Yn
πJ ≡ π j(i) : RN → Rj(i) ,
i=1 i=1
by:
A = π −1
J (H). (1.27)
+
For A ∈ A defined by H and J, the product set function µ0 is defined by:
µ0 (A) = µJ (H), (1.28)
where µJ denotes the finite dimensional product space probability measure associated with
{(R, B(R), µj(i) )}ni=1 .
Then A+ again proves to be an algebra in Proposition I.9.17, and µ0 is a measure on the

algebra by this result and Proposition I.9.19. The Hahn-Kolmogorov extension theorem is
applied to obtain the complete probability space (RN , σ(RN ), µN ). Further, A+ ⊂ σ(RN ) and
µN (A) = µ0 (A) for all A ∈ A+ .
1.4 Continuity of Measures

One of the most important properties of all measures used in various proofs is continuity,
and specifically, continuity from above and continuity from below. These properties
identify conclusions that can be drawn on the measures of a collection of nested measurable
sets. By nested is meant that the collection {Ai }∞
i=1 satisfies:
Ai ⊂ Ai+1 , for all i,
or
Ai+1 ⊂ Ai , for all i.
In the former case, we are interested in the measure of the union, and in the latter, the
measure of the intersection.
The properties identified in this proposition are often referred to in terms of the “conti-
nuity” of measures and understood in the following sense. Given a collection of measurable
sets {Bj }∞
j=1 , define Ai by:
Si
1. Ai = j=1 Bj , or,
Ti
2. Ai = j=1 Bj , where it is assumed that µ(A1 ) < ∞.
16 Measure Spaces
Proposition 1.29 states that in both cases:

µ lim Ai = lim µ(Ai ).
i→∞ i→∞
In the first case, {Ai }∞

i=1 is an increasing sequence of sets and the result is called
continuity from below, whereas in the second case, the sequence is decreasing sequence
of sets and the result is called continuity from above.
Proposition 1.29 (Continuity of all measures) Given the measure space (X, σ (X) , µ)
and {Ai }∞
i=1 ⊂ σ (X):
1. Continuity from Below: If Ai ⊂ Ai+1 for all i:

[∞
µ Ai = lim µ(Ai ), (1.29)
i=1 i→∞
where the limit on the right may be finite or infinite.
2. Continuity from Above: If Ai+1 ⊂ Ai for all i and µ(A1 ) < ∞:

\∞
µ Ai = lim µ(Ai ). (1.30)
i=1 i→∞
Proof. To prove item 1, note that Ai ⊂ Ai+1 implies that µ(Ai ) ≤ µ(Ai+1 ) by monotonicity
of µ. Define B1 = A1 , and for i ≥ 2, let Bi = Ai − Ai−1 . Then {Bi }∞
i=1 ⊂ σ (X) are disjoint
sets, and: [∞ [∞
Ai = Bi .
i=1 i=1
By countable additivity of µ:
[∞ X∞
µ Ai = µ(Bi )
i=1 i=1
Xi
= µ(A1 ) + lim µ(Aj − Aj−1 ).
i→∞ j=2
Since Aj−1 and Aj − Aj−1 are disjoint with union Aj , finite additivity assures that:
µ(Aj − Aj−1 ) = µ(Aj ) − µ(Aj−1 ).
Thus cancellation in this telescoping summation obtains (1.29):

[∞
µ Ai = lim µ(Ai ).
i=1 i→∞
Ti
For item 2, j=1 Aj = Ai by the nesting property, while monotonicity and the assump-
tion that µ(A1 ) < ∞ yields for all i :
\
i
µ Aj = µ(Ai ) < µ(A1 ).
j=1
Again by monotonicity, {µ(Ai )}∞i=1 is a bounded, nonincreasing sequence, and thus has a
well-defined limit as i → ∞, which proves (1.30).
2
Measurable Functions
We begin with a discussion of the definition and properties of a µ-measurable function

defined on a measure space (X, σ(X), µ). For this definition, recall that for f : X → R, with
range in the extended real numbers of Definition I.3.1, the inverse f −1 is defined as a set
function on any set A ⊂ R by:
f −1 (A) ≡ {x ∈ X|f (x) ∈ A}. (2.1)
While f −1 is always defined as a set function, it is only defined pointwise if f is one-to-one.

In Book I, Lebesgue measurability of a function defined on (R, ML (R) , m) could be
characterized in Definition I.3.9 by various equivalent properties on the set function f −1 .
The ultimate goal of such measurability was derived in Proposition I.3.26, and the following
definition of a measurable function f on (X, σ(X), µ) focuses on this goal. Following this,
we investigate equivalent formulations.
Definition 2.1 (Measurable function; transformation) An extended real-valued

function f : X → R defined on the measure space (X, σ(X), µ) is said to be measur-
able, or µ-measurable, or σ(X)-measurable, if for every Borel set A ∈ B(R):
f −1 (A) ∈ σ(X).
n
An extended real-valued transformation f : X → R defined on the measure space
(X, σ(X), µ) is said to be measurable, etc., if f −1 (A) ∈ σ(X) for every Borel set A ∈
n
B(R ).
More generally, a mapping f (x) between measure spaces (X, σ(X), µX ) and (Y, σ(Y ), µY )
is measurable or σ(X)/σ(Y )-measurable, if for all A ∈ σ(Y ):
f −1 (A) ∈ σ(X).
If D ⊂ X and f (x) is defined on D, then the criterion for measurability is the same as
above, and thus of necessity, D = f −1 (R) ∈ σ(X).
n
Remark 2.2 (On A ∈ B(R) or A ∈ B(R )) With f (x) an extended real-valued function,
n
f : X → R or f : X → R , it may seem odd that the measurability criterion only addresses
Borel sets in R and Rn . But note that if f is measurable by the above definition, then:
f −1 (∞) = X − f −1 (R/Rn ),
and so f −1 (∞) ∈ σ(X). Thus f −1 (A {∞}) ∈ σ(X) for all such A.

S
Notation 2.3 (σ(X)-measurable) Given the variety of labels used above to declare mea-
surability, σ(X)-measurable is the most accurate for extended real-valued functions and
transformations because the criterion f −1 (A) ∈ σ(X) for all Borel A is a sigma algebra
restriction. Measurability has nothing to do with the measure µ since there can be many
DOI: 10.1201/9781003264576-2 17
18 Measurable Functions
measures defined on a sigma algebra, and these do not affect which functions are measur-
able and which are not. Nonetheless, the use of µ-measurable is fairly common and rarely
causes confusion when there is one sigma algebra on the space.
However, there will be instances in coming studies where we will encounter measure
spaces (X, σ i (X), µ) with various sigma-algebras σ i (X). In other words, the space X is
fixed as is the measure µ, but there can be various sigma algebras on which µ satisfies
the definition of measure. A simple example but a common one is when (X, σ(X), µ) is a
measure space and σ i (X) ⊂ σ(X) is a sigma subalgebra, then (X, σ i (X), µ) is again a
measure space. But there can also be multiple sigma algebras with no such inclusions.
In these situations, the notion of a measurable function can become ambiguous, as
can the notion of a µ-measurable function. Thus when there is more than one sigma
algebra on X, it is necessary to say that f is σ(X)-measurable, identifying the defining
sigma algebra.
When f (x) is a mapping between general measure spaces (X, σ(X), µX ) and
(Y, σ(Y ), µY ), we will always say that f is σ(X)/σ(Y )-measurable as noted above.
Although a σ(X)-measurable function or transformation could be called σ(X)/B(R)-
n
measurable, or σ(X)/B(R )-measurable, this level of formality is rarely needed.
Exercise 2.4 (Composition of measurable functions) Show that if f (x) is σ(X)/

σ(Y )-measurable between measure spaces (X, σ(X), µX ) and (Y, σ(Y ), µY ), and g(y)
is σ(Y )/σ(Z)-measurable between measure spaces (Y, σ(Y ), µY ) and (Z, σ(Z), µZ ), then
f (g(x)) is σ(X)/σ(Z)-measurable between measure spaces (X, σ(X), µX ) and (Z, σ(Z), µZ ).
Although measurability of f −1 (A) for all A in the range space sigma algebra is the
requirement, it is not necessary to verify this condition for all such A to establish measura-
bility. This was seen in Proposition I.3.4 for Lebesgue or Borel measurability, meaning where
the respective domain space was (R, ML (R) , m) or (R, B (R) , m). Then, measurability for
all y of f −1 ((−∞, y)) is equivalent to this statement for all f −1 ([y, ∞)), all f −1 ((y, ∞)), or
all f −1 ((−∞, y]). Using the same ideas, this is equivalent to measurability of all f −1 ((a, b)).
In any of these cases, Proposition I.3.26 obtains that this is equivalent to measurability of
f −1 (A) for all A ∈ B (R) .
Exercise 2.5 Generalize the prior paragraph to Lebesgue or Borel measurability Qn of transfor-
mations defined on (Rn , ML (Rn ) , m) or (Rn , B (Rn ) , m). Show that if f −1 ( i=1 (−∞, yi ))
is measurable forQall y = (y1 , ...,
Qnyn ), then this
Qnis equivalent toQmeasurability of all f −1 (A)
n n
for A defined as i=1 [yi , ∞), i=1 (yi , ∞), i=1 (−∞, yi ] or i=1 (ai , bi ).
What is clear from the Book I development and the results of Exercise 2.5 is that these
collections of sets, and there are many others, have the property that they generate the Borel
sigma algebras B (R) or B (Rn ) , the sigma algebras of the range spaces. This generalizes as
might be expected.
In the following result, we specify this special collection as A0 , which is our standard
notation for a semi-algebra. This result is true for any collection of sets that generate σ(Y ),
not just semi-algebras. But we use this notation because it will often be the case that there
is an apparent semi-algebra A0 that generates an algebra A, which in turn generates the
range space sigma algebra σ(Y ). For example, this applies when (Y, σ(Y ), µY ) is a measure
space created by the extension theory of the prior chapter.
Example 2.6 (Sigma algebras generated by collections) If A0 is a semi-algebra and

µ0 a pre-measure on A0 , then µ0 can be extended to a measure µA on the associated algebra
A of finite disjoint unions of A0 -sets by the second Carathéodory extension theorem of
Proposition 1.20.
Properties of Measurable Functions 19
Then, by the Hahn-Kolmogorov extension theorem of Proposition 1.18 (with notation

changed), if A is an algebra of sets on Y and µA a measure on A, then µA gives rise to an
outer measure µ∗A on σ(P (Y )) such that µ∗A (A) = µA (A) for all A ∈ A. In addition, there
exists a complete sigma algebra C(Y ) with A ⊂ C(Y ), and µ ≡ µ∗A is a measure on C(Y ).
Thus if σ (A) denotes the smallest sigma algebra generated by A, then (Y, σ(A), µ) is
an example of a measure space where the sigma algebra σ(Y ) ≡ σ(A) is generated by the
semi-algebra A0 , or by the algebra A.
Proposition 2.7 (Measurability test: Generating collections) If f (x) is a mapping
between measure spaces (X, σ(X), µX ) and (Y, σ(Y ), µY ), and A0 is a collection of sets that
generates σ(Y ), then f (x) is σ(X)/σ(Y )-measurable if and only if f −1 (A) ∈ σ(X) for all
A ∈ A0 .
−1 0
0 −1 0
“if ” direction, note that if f (A ) ⊂ σ(X),
Proof. “Only if” is true by definition. For the
−1
then σ f (A ) ⊂ σ(X), where σ f (A ) denotes the smallest sigma algebra generated
by this collection of sets.
It is an exercise to check that:
• f −1 (A)
e = f^−1 (A) for all A ∈ A0 , where A −1 (A) ≡ X − f −1 (A) denote
e ≡ Y − A and f^
the complements of these sets.
Sn Sn
• f −1 ( i=1 Ai ) = i=1 f −1 (Ai ) for all {Ai }ni=1 ⊂ A0 .
Thus if A ∈ σ(A0 ), the smallest sigma algebra that contains A0 , then f −1 (A) ∈
σ f −1 (A0 ) , and the proof is complete.
The final definitional result relates the notions of measurable functions and measurable
n
transformations. If f (x) : X → R is defined on the measure space (X, σ(X), µ), then f (x) =
(f1 (x), ..., f (xn )), and it is logical to inquire into the measurability of f (x) vs. measurability
of the component functions {fi (x)}ni=1 . The reader may recall a similar discussion related
to random variables and random vectors that was summarized in Proposition II.3.32.
Proposition 2.8 (Measurability test: Transformations and functions) If f (x) :
n
X → R is defined on the measure space (X, σ(X), µ) with f (x) = (f1 (x), ..., f (xn )), then
f (x) is a σ(X)-measurable transformation if and only if fi (x) is σ(X)-measurable for all i.
Proof. If f (x) is σ(X)-measurable, then f −1 (A) ∈ σ(X) for all A ∈ B (Rn ) . Fixing i and
taking A = (ai , bi ) × Rn−1 with apparent notation obtains that f −1 (A) = fi−1 ((ai , bi )) ∈
σ(X) for all (ai , bi ). Since such sets generate B (R) , fi (x) is σ(X)-measurable by Proposition
2.7.
Conversely, fi−1 ((ai , bi )) ∈ σ(X) for all i and (ai , bi ) implies that:
Yn \n
f −1 (ai , bi ) = f −1 ((ai , bi )) ∈ σ(X),
i=1 i=1
and again Proposition 2.7 completes the proof.
2.1 Properties of Measurable Functions

Recall Example 1.2 that if the measure space (X, σ(X), µ) also has a topology T , the Borel
sigma algebra on X, denoted B(X), is the smallest sigma algebra that contains the open
sets of X. A topology on X also allows one to define the notion of a continuous function.
To set the stage, we document the definition of continuous function in the more familiar
settings.
Definition 2.9 (Continuous function on metric space) The function f : R → R is

continuous at x0 if:
lim f (x) = f (x0 ). (2.2)
x→x0
That is, given > 0 there is a δ ≡ δ (x0 , ) > 0 so that:
|f (x) − f (x0 )| < if |x − x0 | < δ.
A function is said to be continuous on an interval [a, b] if it is continuous at each

x0 ∈ (a, b), and also continuous at a and b where the limit in (2.2) is understood as one-
sided, meaning for x < b or x > a. A function is said to be continuous if it is continuous
everywhere on its domain.
The same definition in (2.2) applies to a function f : Rn → R, but where |x − x0 | ≡
d(x, x0 ) is interpreted in terms of the standard metric on Rn :
hXn i1/2
2
d(x, y) = (xi − yi ) . (2.3)
i=1
More generally, this definition applies to a function f : (X1 , d1 ) → (X2 , d2 ), where Xj

is a metric space with metric dj . The limit in (2.2) then means that given > 0 there is a
δ ≡ δ (x0 , ) > 0 , so that:
d2 (f (x), f (x0 )) < if d1 (x, x0 ) < δ.
The following result is also true for f : (X1 , d1 ) → (X2 , d2 ) where, recalling Exercise 1.6,
open sets in X1 and X2 are those induced by these metrics. Details are left as an exercise
in changing notation.
Proposition 2.10 (Continuity and open sets) Let f : Rn → R be a given function.

Then, f is continuous if and only if for any open set G ⊂ R, the set f −1 (G) is open in Rn
where:
f −1 (G) ≡ {x|f (x) ∈ G}.
Proof. Assume that f is continuous, that G ⊂ R is open, and that x0 ∈ f −1 (G). To prove
that f −1 (G) is open by Definition 1.4, it must be shown that there exists r > 0 and an
open ball Br (x0 ) ⊂ Rn with Br (x0 ) ⊂ f −1 (G). As G is open there exists > 0 so that the
open ball B (y0 ) ⊂ G, where y0 = f (x0 ). Thus |y − y0 | < for y ∈ B (y0 ). By definition of
continuity, there exists δ so that |f (x) − y0 | < if |x − x0 | < δ. Translating these statements
obtains f (Bδ (x0 )) ⊂ B (y0 ), and so with r = δ:
Br (x0 ) ⊂ f −1 (B (y0 )) ⊂ f −1 (G).
Conversely, assume f −1 (G) is open for all open G ⊂ R. Let x0 ∈ Rn be given and
y0 = f (x0 ) ∈ R. Choose any open set G ⊂ R that contains y0 , for example, we could
choose G = R. By definition of open, there exists > 0 so that B (y0 ) ⊂ G. By assump-
tion f −1 (B (y0 )) is open in Rn and contains x0 . Again by definition of open, there exists
Bδ (x0 ) ⊂ f −1 (B (y0 )) and thus f (Bδ (x0 )) ⊂ B (y0 ). This now translates to the − δ
definition for continuity and the proof is complete.
This result now provides an immediate extension of the notion of continuity to functions
on topological spaces.
Definition 2.11 (Continuous function on a topological space) Given a topological

space X, a real-valued function f : X → R is continuous if f −1 (G) is open for any open
G ⊂ R.
Properties of Measurable Functions 21
It now follows from this characterization of continuity that continuous functions are
σ(X)-measurable.
Proposition 2.12 (Continuous ⇒ µ-measurable) Given a measure space (X, σ(X), µ),
assume that X is also a topological space and that σ(X) contains the open sets of X, and
hence contains the Borel sigma algebra B(X). If f : X → R is continuous, then f is σ(X)-
measurable.
Proof. Consider the open set G ≡ (a, b). Since continuous, f −1 ((a, b)) is open in X, and
thus by definition:
f −1 ((a, b)) ∈ B(X) ⊂ σ(X).
The proof is complete by Proposition 2.7.
The next result summarizes that measurability is preserved under simple arithmetic
operations. We restrict to real-valued functions with range in R. The reader is referred to
Remark I.3.34 for a discussion on generalizing to extended real-valued functions with range
in R. Furthermore, items 1 and 2 remain true for measurable real-valued transformations
and this is left as an exercise.
Proposition 2.13 Let f (x) and g(x) be real-valued σ(X)-measurable functions defined on
a measure space (X, σ(X), µ), and let a, b ∈ R. Then, the following are σ(X)-measurable:
1. af (x) + b,
2. f (x) ± g(x),
3. f (x)g(x),
4. f (x)/g(x) on {x|g(x) 6= 0}.
Proof. To simplify notation, the set {x|f (x) < r} is denoted by {f (x) < r}, and so forth.
Also, by Proposition 2.7, it is sufficient to prove that h−1 (A) ∈ σ(X) for any collection of
sets that generates B (R) , where h(x) denotes any function under consideration.
1. If a = 0, the function g(x) = b is σ(X)-measurable since g −1 (A) ∈ {∅, X} for all A. For
a > 0,
{af (x) + b < y} = {f (x) < (y − b)/a},
which is measurable since f (x) is σ(X)-measurable. A similar result applies to a < 0.
2. Consider the sum since then by part 1, −g(x) is measurable and this implies the result
for f (x) − g(x). For rational r, if f (x) < r and g(x) < y − r, then f (x) + g(x) < y.
Taking a union over all rational r:
[ h \ i
{f (x) < r} {g(x) < y − r} ⊂ {f (x) + g(x) < y}.
r
On the other hand, if f (x) + g(x) < y then f (x) < y − g(x), and by density of the
rationals, there exists rational r so that f (x) < r < y − g(x). This implies f (x) < r and
g(x) < y − r.
Hence, [ h \ i
{f (x) + g(x) < y} = {f (x) < r} {g(x) < y − r} ,
r
and this set is measurable as a countable union of intersections of measurable sets.

3. First note that both f 2 (x) and g 2 (x) are measurable. For f 2 (x), for example:
√ T √
{f (x) < y} {f (x) > − y}, y ≥ 0,
{f 2 (x) < y} =
∅, y < 0.
By parts 1 and 2, so too is [f (x) + g(x)]2 measurable, as is:
f (x)g(x) = 0.5 [f (x) + g(x)]2 − f 2 (x) − g 2 (x) .

4. First, D ≡ {g(x) 6= 0} ∈ σ(X), since:
D = g −1 (−∞, 0) g −1 (0, ∞),

S
and so 1/g(x) is real-valued and well-defined on D. Measurability of 1/g(x) on D then

follows since:
• y > 0: [
{1/g(x) < y} = {g(x) > 1/y} {g(x) < 0}.
• y = 0:
{1/g(x) < 0} = {g(x) < 0}.
• y < 0: \
{1/g(x) < y} = {g(x) > 1/y} {g(x) < 0}.
Thus 1/g(x) is measurable as is f (x)/g(x) by item 3.
The next result is a good example of when a measurability conclusion requires complete-
ness of the measure space (X, σ(X), µ). Recall that when f (x) = g(x), except on a set of
µ-measure 0, this is often written as f (x) = g(x) µ-a.e., and read, “µ almost everywhere.”
Proposition 2.14 (f (x) = g(x), µ-a.e.) Let f (x) be a σ(X)-measurable function on a

complete measure space (X, σ(X), µ), and g(x) a function with f (x) = g(x), µ-a.e.
Then, g(x) is σ(X)-measurable.
Proof. If E ∈ σ(X) is the set of µ-measure zero on which f (x) 6= g(x), then:
[
{x|g(x) < y} = {x ∈ E|g(x) < y} {x ∈ / E|g(x) < y}
[
= {x ∈ E|g(x) < y} {x ∈ / E|f (x) < y}.
The first set is a subset of a set of µ-measure zero and is hence σ(X)-measurable by com-
pleteness, wheras the second set is the intersection of measurable E,
e the complement of E,
and σ(X)-measurable {x|f (x) < y}.
2.2 Limits of Measurable Functions

In this section, we investigate various limits of measurable functions and begin by recalling
some definitions. The reader is referred to Section I.3.4.2 for a discussion of these limiting
functions.
Limits of Measurable Functions 23
Definition 2.15 (Infimum/supremum) Given a finite or countable sequence of func-

tions {fn (x)}∞ n=1 , the infimum and supremum of the sequence are defined pointwise as
follows. T∞
For each x ∈ D ≡ n=1 Dmn{fn }, where Dmn{fn } denotes the domain of the function
fn :
{fn (x)}∞

−∞, n=1 unbounded below,
inf n fn (x) = (2.4)
max{y|y ≤ fn (x) all n}, {fn (x)}∞
n=1 bounded below.
{fn (x)}∞

∞, n=1 unbounded above,
supn fn (x) = (2.5)
min{y|y ≥ fn (x) all n}, {fn (x)}∞
n=1 bounded above.
When {fn (x)}N

n=1 is a finite collection, inf n fn (x) is often denoted:
inf n fn (x) ≡ min{f1 (x), ..., fN (x)},
and supn fn (x) is denoted:
supn fn (x) ≡ max{f1 (x), ..., fN (x)}.
Definition 2.16 (Limits inferior/superior) Given a sequence of functions {fn (x)}∞ n=1 ,
the limit inferior and
T∞limit superior of the sequence are defined pointwise as follows.
For each x ∈ D ≡ n=1 Dmn{fn }:
lim inf fn (x) = supn inf fk (x), (2.6)

n→∞ k≥n
lim sup fn (x) = inf n sup fk (x). (2.7)

n→∞ k≥n
When clear from the context, the subscript n → ∞ is often dropped from the lim inf and
lim sup notation.
Notation 2.17 The limit superior of a function sequence is alternatively denoted limfn (x),
and the limit inferior denoted limfn (x), but we will use the above notation throughout these
books.
Exercise 2.18 (The lim in lim inf and lim sup) From Definition 2.16, it may not be ap-
parent where the notion of limit appears. Prove that:
lim inf fn (x) = lim inf fk (x),

n→∞ n→∞ k≥n
lim sup fn (x) = lim sup fk (x).

n→∞ n→∞ k≥n
In other words, the limit inferior is the limit of infima, wheras the limit superior is the limit
of suprema. Hint: Consider how inf k≥n fk (x) and supk≥n fk (x) vary with n.
As anticipated, these limiting functions are σ(X)-measurable when the functions in the
sequence are σ(X)-measurable. For item 7, we recall Corollary I.3.46 that lim fn (x) exists
at x if and only if:
−∞ < lim inf fn (x) = lim sup fn (x) < ∞. (2.8)
Proposition 2.19 (Measurability of functions derived from {fn (x)}∞ n=1 ) Given a se-
quence of σ(X)-measurable functions {fn (x)}∞ n=1 defined on measurable domains
{Dn }∞
n=1 of the measure space
T∞ (X, σ(X), µ), the following functions are also σ(X)-
measurable as defined on D ≡ n=1 Dn :
1. minn≤N {fn (x)}, for all N ;

2. maxn≤N {fn (x)}, for all N ;
3. inf fn (x);
4. sup fn (x);
5. lim inf fn (x);
6. lim sup fn (x);
7. If h(x) ≡ lim fn (x) exists on D0 ⊂ D, then h(x) is σ(X)-measurable on D0 .
Proof. By Proposition 2.7, it is sufficient to prove that h−1 (A) ∈ σ(X) for any collection
of sets that generate B (R) where h(x) denotes any function under consideration.
Item 1 follows from item 3, and 2 from 4, by defining fn (x) = fN (x) for n ≥ N.
For item 3, if h(x) is defined by h(x) = inf fn (x), then by (2.4):
h(x) > y ⇐⇒ fn (x) > y for all n.
Thus: \∞
{x|h(x) > y} = {x|fn (x) > y},
n=1
and is measurable as the intersection of measurable sets.
Similarly, with g(x) = sup fn (x):
\∞
{x|g(x) < y} = {x|fn (x) < y},
n=1
and this set is again measurable as the intersection of measurable sets.

Now let h(x) = lim inf fn (x), which by (2.6) means h(x) = supn inf k≥n fk (x). Then for
each n, Fn (x) ≡ inf k≥n fk (x) is measurable by 3, and hence h(x) = sup Fn (x) is measurable
by 4. The same approach proves that lim sup fn (x) is measurable using (2.7).
0
If lim fn (x) exists everywhere, then σ(X)-measurability
T −1 follows from (2.8). If D is the set
0 −1
on which lim fn (x) exists, then D = k (−∞, 0] k [0, ∞) where k(x) = lim sup fn (x) −
lim inf fn (x), and so D0 ∈ σ(X). Thus lim fn (x) = χD0 (x) lim sup fn (x) is σ(X)-measurable
on D0 by item 6 and Definition 2.1, where σ(X)-measurable χD0 (x) equals 1 on D0 and 0
elsewhere.
Corollary 2.20 (Measurability on complete (X, σ(X), µ)) Given a sequence of σ(X)-
measurable functions {fn (x)}∞ ∞
n=1 defined on σ(X)-measurable domains {Dn }n=1 of the com-
plete measure space (X, σ(X), µ), if h(x) denotes any of the functions identified in Propo-
sition 2.19, and g(x) = h(x) µ-a.e., then g(x) is σ(X)-measurable.
Proof. This is Proposition 2.14.
2.3 Results on Function Sequences

In this section, we investigate a few implications of the convergence of σ(X)-measurable
functions. The first states that pointwise convergence of measurable functions assures some-
thing more outside arbitrarily small sets. This “something more” initially resembles a uni-
form convergence result. However, as discussed below, this result does not assure uniform
convergence outside a set of measure 0, nor even outside an arbitrarily small set.
Results on Function Sequences 25
Proposition 2.21 Given (X, σ(X), µ), let {fn (x)}∞ n=1 be a sequence of real-valued σ(X)-
measurable functions defined on a measurable set D with µ(D) < ∞, and let f (x) be a
real-valued function so that fn (x) → f (x) pointwise for x ∈ D. Then given > 0 and δ > 0,
there is a measurable set A ⊂ D and an integer N , so that µ(A) < δ, and for all x ∈ D − A
and all n ≥ N :
|fn (x) − f (x)| < .
Proof. Given > 0, define:
Gn = {x| |fn (x) − f (x)| ≥ },

[∞
and DN = Gn :
n=N
DN = {x| |fn (x) − f (x)| ≥ for some n ≥ N }.
Then, {DN }∞ N =1 is a nested sequence, DN +1 ⊂ DN ⊂ D.

Since fn (x) →T∞f (x) for each x ∈ D, it follows that for every such x there is a DN with
x∈/ DN . Hence, N =1 DN = ∅, and since µ(D) < ∞, it follows from continuity from above
of µ by Proposition 1.29 that limN →∞ µ[DN ] → 0. Thus given δ > 0, there is an N with
µ[DN ] < δ.
Defining A ≡ DN , then µ(A) < δ and if x ∈ / A, then |fn (x) − f (x)| < for all n ≥ N .
Corollary 2.22 If (X, σ(X), µ) is complete, the conclusion of the above proposition re-
mains valid if fn (x) → f (x) for each x ∈ D outside a set of µ-measure 0.
Proof. Everything in the above proof remains the same, except that we can now only con-
clude that for every xT∈ D outside an exceptional set of measure 0, that there exists DN with
x∈ / DN , and hence N DN equals this set of measure 0. But then, limN →∞ µ[ DN ] → 0
again by Proposition 1.29, and the proof follows as above, with a final application of Corol-
lary 2.20.
This proposition does not imply that fn (x) converges uniformly to f (x) on D − A
because the set A depends on the given and δ. This result is close to but not equivalent to
Littlewood’s third principle of Chapter I.4, named for J. E. Littlewood (1885–1977).
To improve this result to Littlewood’s conclusion of “nearly uniform convergence,” it
must be shown that A can be chosen so that fn (x) → f (x) uniformly on D − A. That is,
we need to find a fixed set A with µ(A) < δ, so that for any > 0, there is an N, such that
|fn (x) − f (x)| < for all x ∈ D − A and all n ≥ N. See the introduction to Chapter I.4 for
more on Littlewood’s principles.
This next result formalizes Littlewood’s third principle and is known as Egorov’s the-
orem, named for Dmitri Fyodorovich Egorov (1869–1931), and sometimes phonetically
translated to Egoroff. It is also known as the Severini–Egorov theorem in recognition
of the somewhat earlier and independent proof by Carlo Severini (1872–1951).
Proposition 2.23 (Severini-Egorov theorem ) Given (X, σ(X), µ), let {fn (x)}∞ n=1 be
a sequence of σ(X)-measurable functions defined on a measurable set D with µ(D) < ∞,
and let f (x) be a σ(X)-measurable function so that fn (x) → f (x) pointwise for x ∈ D.
Then given δ > 0, there is a measurable set A ⊂ D with µ(A) < δ, so that fn (x) → f (x)
uniformly on D − A.
That is, for > 0, there is an N , so that |fn (x) − f (x)| < for all x ∈ D − A and
n ≥ N.
Proof. Given δ > 0, for each m define m = 1/m and δ m = δ/2m and apply Proposition
2.21. The result is a set Am with µ(Am ) < δ m , and
S∞an integer Nm , so that |fn (x) − f (x)| <
m for nP≥ Nm and all x ∈ D − Am . With A ≡ m=1 Am , countable subadditivity obtains
∞
µ(A) ≤ m=1 µ(Am ) = δ. We now claim that fn (x) → f (x) uniformly on D − A.
Given there is an m so that m < , and hence an Nm so that |fn (x) − f (x)| < m <
for n ≥ Nm and all x ∈ D − Am . But then this statement is also true for x ∈ D − A since
Am ⊂ A.
Corollary 2.24 (Severini–Egorov theorem) If (X, σ(X), µ) is complete, the above re-
sult remains valid if fn (x) → f (x) µ-a.e. for x ∈ D.
Proof. Left as an exercise.
Remark 2.25 (On µ(D) < ∞) To perhaps state the obvious, the above results apply with-
out the explicit need for the restriction of µ(D) < ∞ in finite measure spaces, and in par-
ticular, in probability spaces. In such a space, we can conclude that pointwise convergence
on any measurable set assures nearly uniform convergence.
2.4 Approximating σ(X)-Measurable Functions

In this section, we investigate various approximations of measurable functions with simple
functions. We begin by re-introducing the notion of a simple function, as originally seen in
Books I and III, but generalized somewhat and framed in the current context. The reader
should confirm that simple functions are σ(X)-measurable.
Definition 2.26 (Simple function) A simple function ϕ : X → R on a measure space

(X, σ(X), µ) is defined by:
Xn
ϕ(x) = ai χAi (x), (2.9)
i=1
where:
1. {Ai }ni=1 ⊂ σ(X) are disjoint measurable sets;

2. χAi (x) is the characteristic function or indicator function for Ai , defined as
χAi (x) = 1 for x ∈ Ai and χAi (x) = 0 otherwise;
3. {ai }ni=1 ⊂ R and ai ≥ 0 for all i.
A vector-valued simple function ϕ : X → Rm on a measure space (X, σ(X), µ) is

defined by:
ϕ(x) = (ϕ1 (x), ..., ϕm (x)),
where {ϕi }m
i=1 are given as in (2.9). Equivalently (see the proof of Corollary 2.30):
Xn
ϕ(x) = ai χAi (x),
i=1
where items 1 and 2 apply, with ai = (ai,1 , ..., ai,m ):
3 0. {ai }ni=1 ⊂ Rm and ai,j ≥ 0 for all i and j.

Another random document with
no related content on Scribd:
sacraments and scriptures of the Persian, Jewish, Christian, “post-
Classical” and Manichæan religions.
“Space”—speaking now in the Faustian idiom—is a spiritual
something, rigidly distinct from the momentary sense-present, which
could not be represented in an Apollinian language, whether Greek
or Latin. But the created expression-space of the Apollinian arts is
equally alien to ours. The tiny cella of the early-Classical temple was
a dumb dark nothingness, a structure (originally) of perishable
material, an envelope of the moment in contrast to the eternal vaults
of Magian cupolas and Gothic naves, and the closed ranks of
columns were expressly meant to convey that for the eye at any rate
this body possessed no Inward. In no other Culture is the firm
footing, the socket, so emphasized. The Doric column bores into the
ground, the vessels are always thought of from below upward
(whereas those of the Renaissance float above their footing), and
the sculpture-schools feel the stabilizing of their figures as their main
problem. Hence in archaic works the legs are disproportionately
emphasized, the foot is planted on the full sole, and if the drapery
falls straight down, a part of the hem is removed to show that the
foot is standing. The Classical relief is strictly stereometrically set on
a plane, and there is an interspace between the figures but no depth.
A landscape of Claude Lorrain, on the contrary, is nothing but space,
every detail being made to subserve its illustration. All bodies in it
possess an atmospheric and perspective meaning purely as carriers
of light and shade. The extreme of this disembodiment of the world
in the service of space is Impressionism. Given this world-feeling,
the Faustian soul in the springtime necessarily arrived at an
architectural problem which had its centre of gravity in the spatial
vaulting-over of vast, and from porch to choir dynamically deep,
cathedrals. This last expressed its depth-experience. But with it was
associated, in opposition to the cavernous Magian expression-
space,[190] the element of a soaring into the broad universe. Magian
roofing, whether it be cupola or barrel-vault or even the horizontal
baulk of a basilica, covers in. Strzygowski[191] has very aptly
described the architectural idea of Hagia Sophia as an introverted
Gothic striving under a closed outer casing. On the other hand, in the
cathedral of Florence the cupola crowns the long Gothic body of
1367, and the same tendency rose in Bramante’s scheme for St.
Peter’s to a veritable towering-up, a magnificent “Excelsior,” that
Michelangelo carried to completion with the dome that floats high
and bright over the vast vaulting. To this sense of space the
Classical opposes the symbol of the Doric peripteros, wholly
corporeal and comprehensible in one glance.
The Classical Culture begins, then, with a great renunciation. A
rich, pictorial, almost over-ripe art lay ready to its hand. But this
could not become the expression of the young soul, and so from
about 1100 B.C. the harsh, narrow, and to our eyes scanty and
barbaric, early-Doric geometrical style appears in opposition to the
Minoan.[192] For the three centuries which correspond to the flowering
of our Gothic, there is no hint of an architecture, and it is only at
about 650 B.C., “contemporarily” with Michelangelo’s transition into
the Baroque, that the Doric and Etruscan temple-type arises. All
“Early” art is religious, and this symbolic Negation is not less so than
the Egyptian and the Gothic Affirmation. The idea of burning the
dead accords with the cult-site but not with the cult-building; and the
Early Classical religion which conceals itself from us behind the
solemn names of Calchas, Tiresias, Orpheus and (probably)
Numa[193] possessed for its rites simply that which is left of an
architectural idea when one has subtracted the architecture, viz., the
sacred precinct. The original cult-plan is thus the Etruscan templum,
a sacred area merely staked off on the ground by the augurs with an
impassable boundary and a propitious entrance on the East side.[194]
A “templum” was created where a rite was to be performed or where
the representative of the state authority, senate or army, happened
to be. It existed only for the duration of its use, and the spell was
then removed. It was probably only about 700 B.C. that the Classical
soul so far mastered itself as to represent this architectural Nothing
in the sensible form of a built body. In the long run the Euclidean
feeling proved stronger than the mere antipathy to duration.
Faustian architecture, on the contrary, begins on the grand scale
simultaneously with the first stirrings of a new piety (the Cluniac
reform, c. 1000) and a new thought (the Eucharistic controversy
between Berengar of Tours and Lanfranc 1050),[195] and proceeds at
once to plans of gigantic intention; often enough, as in the case of
Speyer, the whole community did not suffice to fill the cathedral,[196]
and often again it proved impossible to complete the projected
scheme. The passionate language of this architecture is that of the
poems too.[197] Far apart as may seem the Christian hymnology of
the south and the Eddas of the still heathen north, they are alike in
the implicit space-endlessness of prosody, rhythmic syntax and
imagery. Read the Dies Iræ together with the Völuspá,[198] which is
little earlier; there is the same adamantine will to overcome and
break all resistances of the visible. No rhythm ever imagined radiates
immensities of space and distance as the old Northern does:
Zum Unheil werden—noch allzulange

Männer und Weiber—zur Welt geboren
Aber wir beide —bleiben zusammen
Ich und Sigurd.
The accents of the Homeric hexameter are the soft rustle of a leaf
in the midday sun, the rhythm of matter; but the “Stabreim” likes
“potential energy” in the world-pictures of modern physics, creates a
tense restraint in the void without limits, distant night-storms above
the highest peaks. In its swaying indefiniteness all words and things
dissolve themselves—it is the dynamics, not the statics, of language.
The same applies to the grave rhythm of Media vita in morte sumus.
Here is heralded the colour of Rembrandt and the instrumentation of
Beethoven—here infinite solitude is felt as the home of the Faustian
soul. What is Valhalla? Unknown to the Germans of the Migrations
and even to the Merovingian Age, it was conceived by the nascent
Faustian soul. It was conceived, no doubt, under Classic-pagan and
Arabian-Christian impressions, for the antique and the sacred
writings, the ruins and mosaics and miniatures, the cults and rites
and dogmas of these past Cultures reached into the new life at all
points. And yet, this Valhalla is something beyond all sensible
actualities floating in remote, dim, Faustian regions. Olympus rests
on the homely Greek soil, the Paradise of the Fathers is a magic
garden somewhere in the Universe, but Valhalla is nowhere. Lost in
the limitless, it appears with its inharmonious gods and heroes the
supreme symbol of solitude. Siegfried, Parzeval, Tristan, Hamlet,
Faust are the loneliest heroes in all the Cultures. Read the wondrous
awakening of the inner life in Wolfram’s Parzeval. The longing for the
woods, the mysterious compassion, the ineffable sense of
forsakenness—it is all Faustian and only Faustian. Every one of us
knows it. The motive returns with all its profundity in the Easter
scene of Faust I.
“A longing pure and not to be described

drove me to wander over woods and fields,
and in a mist of hot abundant tears
I felt a world arise and live for me.”
Of this world-experience neither Apollinian nor Magian man,

neither Homer nor the Gospels, knows anything whatever. The
climax of the poem of Wolfram, that wondrous Good Friday morning
scene when the hero, at odds with God and with himself, meets the
noble Gawan and resolves to go on pilgrimage to Tevrezent, takes
us to the heart of the Faustian religion. Here one can feel the
mystery of the Eucharist which binds the communicant to a mystic
company, to a Church that alone can give bliss. In the myth of the
Holy Grail and its Knights one can feel the inward necessity of the
German-Northern Catholicism. In opposition to the Classical
sacrifices offered to individual gods in separate temples, there is
here the one never-ending sacrifice repeated everywhere and every
day. This is the Faustian idea of the 9th-11th Centuries, the Edda
time, foreshadowed by Anglo-Saxon missionaries like Winfried but
only then ripened. The Cathedral, with its High Altar enclosing the
accomplished miracle, is its expression in stone.[199]
The plurality of separate bodies which represents Cosmos for the
Classical soul, requires a similar pantheon—hence the antique
polytheism. The single world-volume, be it conceived as cavern or as
space, demands the single god of Magian or Western Christianity.
Athene or Apollo might be represented by a statue, but it is and has
long been evident to our feeling that the Deity of the Reformation
and the Counter-Reformation can only be “manifested” in the storm
of an organ fugue or the solemn progress of cantata and mass. From
the rich manifold of figures in the Edda and contemporary legends of
saints to Goethe our myth develops itself in steady opposition to the
Classical—in the one case a continuous disintegration of the divine
that culminated in the early Empire in an impossible multitude of
deities, in the other a process of simplification that led to the Deism
of the 18th Century.
The Magian hierarchy of heaven—angels, saints, persons of the
Trinity—has grown paler and paler, more and more disembodied, in
the sphere of the Western pseudomorphosis,[200] supported though it
was by the whole weight of Church authority, and even the Devil—
the great adversary in the Gothic world-drama[201]—has disappeared
unnoticed from among the possibilities of the Faustian world-feeling.
Luther could still throw the inkpot at him, but he has been passed
over in silence by perplexed Protestant theologians long ago. For the
solitude of the Faustian soul agrees not at all with a duality of world
powers. God himself is the All. About the end of the 17th Century
this religiousness could no longer be limited to pictorial expression,
and instrumental music came as its last and only form-language: we
may say that the Catholic faith is to the Protestant as an altar-piece
is to an oratorio. But even the Germanic gods and heroes are
surrounded by this rebuffing immensity and enigmatic gloom. They
are steeped in music and in night, for daylight gives visual bounds
and therefore shapes bodily things. Night eliminates body, day soul.
Apollo and Athene have no souls. On Olympus rests the eternal light
of the transparent southern day, and Apollo’s hour is high noon,
when great Pan sleeps. But Valhalla is light-less, and even in the
Eddas we can trace that deep midnight of Faust’s study-broodings,
the midnight that is caught by Rembrandt’s etchings and absorbs
Beethoven’s tone colours. No Wotan or Baldur or Freya has
“Euclidean” form. Of them, as of the Vedic gods of India, it can be
said that they suffer not “any graven image or any likeness
whatsoever”; and this impossibility carries an implicit recognition that
eternal space, and not the corporeal copy—which levels them down,
desecrates them, denies them—is the supreme symbol. This is the
deep-felt motive that underlies the iconoclastic storms in Islam and
Byzantium (both, be it noted, of the 7th century), and the closely
similar movement in our Protestant North. Was not Descartes’s
creation of the anti-Euclidean analysis of space an iconoclasm? The
Classical geometry handles a number-world of day, the function-
theory is the genuine mathematic of night.
II
That which is expressed by the soul of the West in its
extraordinary wealth of media—words, tones, colours, pictorial
perspectives, philosophical systems, legends, the spaciousness of
Gothic cathedrals and the formulæ of functions—namely its world-
feeling, is expressed by the soul of Old Egypt (which was remote
from all ambitions towards theory and literariness) almost exclusively
by the immediate language of Stone. Instead of spinning word-
subtleties around its form of extension, its “space” and its “time,”
instead of forming hypotheses and number-systems and dogmas, it
set up its huge symbols in the landscape of the Nile in all silence.
Stone is the great emblem of the Timeless-Become; space and
death seem bound up in it. “Men have built for the dead,” says
Bachofen in his autobiography, “before they have built for the living,
and even as a perishable wooden structure suffices for the span of
time that is given to the living, so the housing of the dead for ever
demands the solid stone of the earth. The oldest cult is associated
with the stone that marks the place of burial, the oldest temple-
building with the tomb-structure, the origins of art and decoration
with the grave-ornament. Symbol has created itself in the graves.
That which is thought and felt and silently prayed at the grave-side
can be expressed by no word, but only hinted by the boding symbol
that stands in unchanging grave repose.” The dead strive no more.
They are no more Time, but only Space—something that stays (if
indeed it stays at all) but does not ripen towards a Future; and hence
it is stone, the abiding stone, that expresses how the dead is
mirrored in the waking consciousness of the living. The Faustian soul
looks for an immortality to follow the bodily end, a sort of marriage
with endless space, and it disembodies the stone in its Gothic thrust-
system (contemporary, we may note, with the “consecutives” in
Church music[202]) till at last nothing remained visible but the
indwelling depth- and height-energy of this self-extension. The
Apollinian soul would have its dead burned, would see them
annihilated, and so it remained averse from stone building
throughout the early period of its Culture. The Egyptian soul saw
itself as moving down a narrow and inexorably-prescribed life-path to
come at the end before the judges of the dead (“Book of the Dead,”
cap. 125). That was its Destiny-idea. The Egyptian’s existence is that
of the traveller who follows one unchanging direction, and the whole
form-language of his Culture is a translation into the sensible of this
one theme. And as we have taken endless space as the prime
symbol of the North and body as that of the Classical, so we may
take the word way as most intelligibly expressing that of the
Egyptians. Strangely, and for Western thought almost
incomprehensibly, the one element in extension that they emphasize
is that of direction in depth. The tomb-temples of the Old Kingdom
and especially the mighty pyramid-temples of the Fourth Dynasty
represent, not a purposed organization of space such as we find in
the mosque and the cathedral, but a rhythmically ordered sequence
of spaces. The sacred way leads from the gate-building on the Nile
through passages, halls, arcaded courts and pillared rooms that
grow ever narrower and narrower, to the chamber of the dead,[203]
and similarly the Sun-temples of the Fifth Dynasty are not “buildings”
but a path enclosed by mighty masonry.[204] The reliefs and the
paintings appear always as rows which with an impressive
compulsion lead the beholder in a definite direction. The ram and
sphinx avenues of the New Empire have the same object. For the
Egyptian, the depth-experience which governed his world-form was
so emphatically directional that he comprehended space more or
less as a continuous process of actualization. There is nothing rigid
about distance as expressed here. The man must move, and so
become himself a symbol of life, in order to enter into relation with
the stone part of the symbolism. “Way” signifies both Destiny and
third dimension. The grand wall-surfaces, reliefs, colonnades past
which he moves are “length and breadth”; that is, mere perceptions
of the senses, and it is the forward-driving life that extends them into
“world.” Thus the Egyptian experienced space, we may say, in and
by the processional march along its distinct elements, whereas the
Greek who sacrificed outside the temple did not feel it and the man
of our Gothic centuries praying in the cathedral let himself be
immersed in the quiet infinity of it. And consequently the art of these
Egyptians must aim at plane effects and nothing else, even when it
is making use of solid means. For the Egyptian, the pyramid over the
king’s tomb is a triangle, a huge, powerfully expressive plane that,
whatever be the direction from which one approaches, closes off the
“way” and commands the landscape. For him, the columns of the
inner passages and courts, with their dark backgrounds, their dense
array and their profusion of adornments, appear entirely as vertical
strips which rhythmically accompany the march of the priests. Relief-
work is—in utter contrast to the Classical—carefully restricted in one
plane; in the course of development dated by the Third to the Fifth
dynasties it diminishes from the thickness of a finger to that of a
sheet of paper, and finally it is sunk in the plane.[205] The dominance
of the horizontal, the vertical and the right angle, and the avoidance
of all foreshortening support the two-dimensional principle and serve
to insulate this directional depth-experience which coincides with the
way and the grave at its end. It is an art that admits of no deviation
for the relief of the tense soul.
Is not this an expression in the noblest language that it is possible
to conceive of what all our space-theories would like to put into
words? Is it not a metaphysic in stone by the side of which the
written metaphysics of Kant seems but a helpless stammering?
There is, however, another Culture that, different as it most
fundamentally is from the Egyptian, yet found a closely-related prime
symbol. This is the Chinese, with its intensely directional principle of
the Tao.[206] But whereas the Egyptian treads to the end a way that is
prescribed for him with an inexorable necessity, the Chinaman
wanders through his world; consequently, he is conducted to his god
or his ancestral tomb not by ravines of stone, between faultless
smooth walls, but by friendly Nature herself. Nowhere else has the
landscape become so genuinely the material of the architecture.
“Here, on religious foundations, there has been developed a grand
lawfulness and unity common to all building, which, combined with
the strict maintenance of a north-south general axis, always holds
together gate-buildings, side-buildings, courts and halls in the same
homogeneous plan, and has led finally to so grandiose a planning
and such a command over ground and space that one is quite
justified in saying that the artist builds and reckons with the
landscape itself.”[207] The temple is not a self-contained building but a
lay-out, in which hills, water, trees, flowers, and stones in definite
forms and dispositions are just as important as gates, walls, bridges
and houses. This Culture is the only one in which the art of
gardening is a grand religious art. There are gardens that are
reflections of particular Buddhist sects.[208] It is the architecture of the
landscape, and only that, which explains the architecture of the
buildings, with their flat extension and the emphasis laid on the roof
as the really expressive element. And just as the devious ways
through doors, over bridges, round hills and walls lead at last to the
end, so the paintings take the beholder from detail to detail whereas
Egyptian relief masterfully points him in the one set direction. “The
whole picture is not to be taken at once. Sequence in time
presupposes a sequence of space-elements through which the eye
is to wander from one to the next.”[209] Whereas the Egyptian
architecture dominates the landscape, the Chinese espouses it. But
in both cases it is direction in depth that maintains the becoming of
space as a continuously-present experience.
III
All art is expression-language.[210] Moreover, in its very earliest
essays—which extend far back into the animal world—it is that of
one active existence speaking for itself only, and it is unconscious of
witnesses even though in the absence of such the impulse to
expression would not come to utterance. Even in quite “late”
conditions we often see, instead of the combination of artist and
spectator, a crowd of art-makers who all dance or mime or sing. The
idea of the “Chorus” as sum total of persons present has never
entirely vanished from art-history. It is only the higher art that
becomes decisively an art “before witnesses” and especially (as
Nietzsche somewhere remarks) before God as the supreme witness.
[211]
This expression is either ornament or imitation. Both are higher

possibilities and their polarity to one another is hardly perceptible in
the beginnings. Of the two, imitation is definitely the earlier and the
closer to the producing race. Imitation is the outcome of a
physiognomic idea of a second person with whom (or which) the first
is involuntarily induced into resonance of vital rhythm (mitschwingen
im); whereas ornament evidences an ego conscious of its own
specific character. The former is widely spread in the animal world,
the latter almost peculiar to man.
Imitation is born of the secret rhythm of all things cosmic. For the
waking being the One appears as discrete and extended; there is a
Here and a There, a Proper and an Alien something, a Microcosm
and a Macrocosm that are polar to one another in the sense-life, and
what the rhythm of imitation does is to bridge this dichotomy. Every
religion is an effort of the waking soul to reach the powers of the
world-around. And so too is Imitation, which in its most devoted
moments is wholly religious, for it consists in an identity of inner
activity between the soul and body “here” and the world-around
“there” which, vibrating as one, become one. As a bird poises itself in
the storm or a float gives to the swaying waves, so our limbs take up
an irresistible beat at the sound of march-music. Not less contagious
is the imitation of another’s bearing and movements, wherein
children in particular excel. It reaches the superlative when we “let
ourselves go” in the common song or parade-march or dance that
creates out of many units one unit of feeling and expression, a “we.”
But a “successful” picture of a man or a landscape is also the
outcome of a felt harmony of the pictorial motion with the secret
swing and sway of the living opposite; and it is this actualizing of
physiognomic rhythm that requires the executant to be an adept who
can reveal the idea, the soul, of the alien in the play of its surface. In
certain unreserved moments we are all adepts of this sort, and in
such moments, as we follow in an imperceptible rhythm the music
and the play of facial expression, we suddenly look over the
precipice and see great secrets. The aim of all imitation is effective
simulation; this means effective assimilation of ourselves into an
alien something—such a transposition and transubstantiation that
the One lives henceforth in the Other that it describes or depicts—
and it is able to awaken an intense feeling of unison over all the
range from silent absorption and acquiescence to the most
abandoned laughter and down into the last depths of the erotic, a
unison which is inseparable from creative activity. In this wise arose
the popular circling-dances (for instance, the Bavarian Schuhplattler
was originally imitated from the courtship of the woodcocks) but this
too is what Vasari means when he praises Cimabue and Giotto as
the first who returned to the imitation of “Nature”—the Nature, that is,
of springtime men, of which Meister Eckart said: “God flows out in all
creatures, and therefore all created is God.” That which in this world-
around presents itself to our contemplation—and therefore contains
meaning for our feelings—as movement, we render by movement.
Hence all imitation is in the broadest sense dramatic; drama is
presented in the movement of the brush-stroke or the chisel, the
melodic curve of the song, the tone of the recitation, the line of
poetry, the description, the dance. But everything that we experience
with and in seeings and hearings is always an alien soul to which we
are uniting ourselves. It is only at the stage of the Megalopolis that
art, reasoned to pieces and de-spiritualized, goes over to naturalism
as that term is understood nowadays; viz., imitation of the charm of
visible appearances, of the stock of sensible characters that are
capable of being scientifically fixed.
Ornament detaches itself now from Imitation as something which
does not follow the stream of life but rigidly faces it. Instead of
physiognomic traits overheard in the alien being, we have
established motives, symbols, which are impressed upon it. The
intention is no longer to pretend but to conjure. The “I” overwhelms
the “Thou.” Imitation is only a speaking with means that are born of
the moment and unreproduceable—but Ornament employs a
language emancipated from the speaking, a stock of forms that
possesses duration and is not at the mercy of the individual.[212]
Only the living can be imitated, and it can be imitated only in
movements, for it is through these that it reveals itself to the senses
of artists and spectators. To that extent, imitation belongs to Time
and Direction. All the dancing and drawing and describing and
portraying for eye and ear is irrevocably “directional,” and hence the
highest possibilities of Imitation lie in the copying of a destiny, be it in
tones, verses, picture or stage-scene.[213] Ornament, on the contrary,
is something taken away from Time: it is pure extension, settled and
stable. Whereas an imitation expresses something by accomplishing
itself, ornament can only do so by presenting itself to the senses as
a finished thing. It is Being as such, wholly independent of origin.
Every imitation possesses beginning and end, while an ornament
possesses only duration, and therefore we can only imitate the
destiny of an individual (for instance, Antigone or Desdemona), while
by an ornament or symbol only the generalized destiny-idea itself
can be represented (as, for example, that of the Classical world by
the Doric column). And the former presupposes a talent, while the
latter calls for an acquirable knowledge as well.
All strict arts have their grammar and syntax of form-language,
with rules and laws, inward logic and tradition. This is true not merely
for the Doric cabin-temple and Gothic cottage-cathedral, for the
carving-schools of Egypt[214] and Athens and the cathedral plastic of
northern France, for the painting-schools of the Classical world and
those of Holland and the Rhine and Florence, but also for the fixed
rules of the Skalds and Minnesänger which were learned and
practised as a craft (and dealt not merely with sentence and metre
but also with gesture and the choice of imagery[215]), for the
narration-technique of the Vedic, Homeric and Celto-Germanic Epos,
for the composition and delivery of the Gothic sermon (both
vernacular and Latin), and for the orators’ prose[216] in the Classical,
and for the rules of French drama. In the ornamentation of an art-
work is reflected the inviolable causality of the macrocosm as the
man of the particular kind sees and comprehends it. Both have
system. Each is penetrated with the religious side of life—fear and
love.[217] A genuine symbol can instil fear or can set free from fear;
the “right” emancipates and the “wrong” hurts and depresses. The
imitative side of the arts, on the contrary, stands closer to the real
race-feelings of hate and love, out of which arises the opposition of
ugly and beautiful. This is in relation only with the living, of which the
inner rhythm repels us or draws us into phase with it, whether it be
that of the sunset-cloud or that of the tense breath of the machine.
An imitation is beautiful, an ornament significant, and therein lies the
difference between direction and extension, organic and inorganic
logic, life and death. That which we think beautiful is “worth copying.”
Easily it swings with us and draws us on to imitate, to join in the
singing, to repeat. Our hearts beat higher, our limbs twitch, and we
are stirred till our spirits overflow. But as it belongs to Time, it “has its
time.” A symbol endures, but everything beautiful vanishes with the
life-pulsation of the man, the class, the people or the race that feels it
as a specific beauty in the general cosmic rhythm.[218] The “beauty”
that Classical sculpture and poetry contained for Classical eyes is
something different from the beauty that they contain for ours—
something extinguished irrecoverably with the Classical soul—while
what we regard as beautiful in it is something that only exists for us.
Not only is that which is beautiful for one kind of man neutral or ugly
for another—e.g., the whole of our music for the Chinese, or
Mexican sculpture for us. For one and the same life the accustomed,
the habitual, owing to the very fact of its possessing duration, cannot
possess beauty.
And now for the first time we can see the opposition between
these two sides of every art in all its depth. Imitation spiritualizes and
quickens, ornament enchants and kills. The one becomes, the other
is. And therefore the one is allied to love and, above all—in songs
and riot and dance—to the sexual love, which turns existence to face
the future; and the other to care of the past, to recollection[219] and to
the funerary. The beautiful is longingly pursued, the significant instils
dread, and there is no deeper contrast than that between the house
of the living and the house of the dead.[220] The peasant’s cottage[221]
and its derivative the country noble’s hall, the fenced town and the
castle are mansions of life, unconscious expressions of circling
blood, that no art produced and no art can alter. The idea of the
family appears in the plan of the proto-house, the inner form of the
stock in the plan of its villages—which after many a century and
many a change of occupation still show what race it was that
founded them[222]—the life of a nation and its social ordering in the
plan (not the elevation or silhouette) of the city.[223] On the other
hand, Ornamentation of the high order develops itself on the stiff
symbols of death, the urn, the sarcophagus, the stele and the temple
of the dead,[224] and beyond these in gods’ temples and cathedrals
which are Ornament through and through, not the expressions of a
race but the language of a world-view. They are pure art through and
through—just what the castle and the cottage are not.[225]
For cottage and castle are buildings in which art, and, specifically,
imitative art, is made and done, the home of Vedic, Homeric and
Germanic epos, of the songs of heroes, the dance of boors and that
of lords and ladies, of the minstrel’s lay. The cathedral, on the other
hand, is art, and, moreover, the only art by which nothing is imitated;
it alone is pure tension of persistent forms, pure three-dimensional
logic that expresses itself in edges and surfaces and volumes. But
the art of villages and castles is derived from the inclinations of the
moment, from the laughter and high spirit of feasts and games, and
to such a degree is it dependent on Time, so much is it a thing of
occasion, that the troubadour obtains his very name from finding,
while Improvisation—as we see in the Tzigane music to-day—is
nothing but race manifesting itself to alien senses under the
influence of the hour. To this free creative power all spiritual art
opposes the strict school in which the individual—in the hymn as in
the work of building and carving—is the servant of a logic of timeless
forms, and so in all Cultures the seat of its style-history is in its early
cult architecture. In the castle it is the life and not the structure that
possesses style. In the town the plan is an image of the destinies of
a people, whereas the silhouette of emergent spires and cupolas
tells of the logic in the builders’ world-picture, of the “first and last
things” of their universe.
In the architecture of the living, stone serves a worldly purpose,
but in the architecture of the cult it is a symbol.[226] Nothing has
injured the history of the great architectures so much as the fact that
it has been regarded as the history of architectural techniques
instead of as that of architectural ideas which took their technical
expression-means as and where they found them. It has been just
the same with the history of musical instruments,[227] which also were
developed on a foundation of tone-language. Whether the groin and
the flying buttress and the squinch-cupola were imagined specially
for the great architectures or were expedients that lay more or less
ready to hand and were taken into use, is for art-history a matter of
as little importance as the question of whether, technically, stringed
instruments originated in Arabia or in Celtic Britain. It may be that the
Doric column was, as a matter of workmanship, borrowed from the
Egyptian temples of the New Empire, or the late-Roman domical
construction from the Etruscans, or the Florentine court from the
North-African Moors. Nevertheless the Doric peripteros, the
Pantheon, and the Palazzo Farnese belong to wholly different worlds
—they subserve the artistic expression of the prime-symbol in three
different Cultures.
IV
In every springtime, consequently, there are two definitely
ornamental and non-imitative arts, that of building and that of
decoration. In the longing and pregnant centuries before it, elemental
expression belongs exclusively to Ornamentation in the narrow
sense. The Carolingian period is represented only by its ornament,
as its architecture, for want of the Idea, stands between the styles.
And similarly, as a matter of art-history, it is immaterial that no
buildings of the Mycenæan age have survived.[228] But with the dawn
of the great Culture, architecture as ornament comes into being
suddenly and with such a force of expression that for a century mere
decoration-as-such shrinks away from it in awe. The spaces,
surfaces and edges of stone speak alone. The tomb of Chephren is
the culmination of mathematical simplicity—everywhere right angles,
squares and rectangular pillars, nowhere adornment, inscription or
desinence—and it is only after some generations have passed that
Relief ventures to infringe the solemn magic of those spaces and the
strain begins to be eased. And the noble Romanesque of
Westphalia-Saxony (Hildesheim, Gernrode, Paulinzella, Paderborn),
of Southern France and of the Normans (Norwich and Peterborough)
managed to render the whole sense of the world with indescribable
power and dignity in one line, one capital, one arch.
When the form-world of the springtime is at its highest, and not
before, the ordained relation is that architecture is lord and ornament
is vassal. And the word “ornament” is to be taken here in the widest
possible sense. Even conventionally, it covers the Classical unit-
motive with its quiet poised symmetry or meander supplement, the
spun surface of arabesque and the not dissimilar surface-patterning
of Mayan art, and the “Thunder-pattern”[229] and others of the early
Chóu period which prove once again the landscape basis of the old
Chinese architecture without a doubt. But the warrior figures of
Dipylon vases are also conceived in the spirit of ornament, and so, in
a far higher degree still, are the statuary groups of Gothic cathedrals.
“The figures were composed pillarwise from the spectator, the
figures of the pillar being, with reference to the spectator, ranked
upon one another like rhythmic figures in a symphony that soars
heavenward and expands its sounds in every direction.”[230] And
besides draperies, gestures, and figure-types, even the structure of
the hymn-strophe and the parallel motion of the parts in church
music are ornament in the service of the all-ruling architectural idea.
[231]
The spell of the great Ornamentation remains unbroken till in the
beginning of a “late” period architecture falls into a group of civic and
worldly special arts that unceasingly devote themselves to pleasing
and clever imitation and become ipso facto personal. To Imitation
and Ornament the same applies that has been said already of time
and space. Time gives birth to space, but space gives death to time.
[232]
In the beginning, rigid symbolism had petrified everything alive;
the Gothic statue was not permitted to be a living body, but was
simply a set of lines disposed in human form. But now Ornament
loses all its sacred rigour and becomes more and more decoration
for the architectural setting of a polite and mannered life. It was
purely as this, namely as a beautifying element, that Renaissance
taste was adopted by the courtly and patrician world of the North
(and by it alone!). Ornament meant something quite different in the
Egyptian Old Kingdom from what it meant in the Middle; in the
geometric period from what it meant in the Hellenistic; at the end of
the 12th Century from what it meant at the end of Louis XIV’s reign.
And architecture too becomes pictorial and makes music, and its
forms seem always to be trying to imitate something in the picture of
the world-around. From the Ionic capital we proceed to the
Corinthian, and from Vignola through Bernini to the Rococo.
At the last, when Civilization sets in, true ornament and, with it,
great art as a whole are extinguished. The transition consists—in
every Culture—in Classicism and Romanticism of one sort or
another, the former being a sentimental regard for an Ornamentation
(rules, laws, types) that has long been archaic and soulless, and the
latter a sentimental Imitation, not of life, but of an older Imitation. In
the place of architectural style we find architectural taste. Methods of
painting and mannerisms of writing, old forms and new, home and
foreign, come and go with the fashion. The inward necessity is no
longer there, there are no longer “schools,” for everyone selects
what and where it pleases him to select. Art becomes craft-art
(Kunstgewerbe) in all its branches—architecture and music, poetry
and drama—and in the end we have a pictorial and literary stock-in-
trade which is destitute of any deeper significance and is employed
according to taste. This final or industrial form of Ornament—no
longer historical, no longer in the condition of “becoming”—we have
before us not only in the patterns of oriental carpets, Persian and
Indian metal work,` Chinese porcelain, but also in Egyptian (and
Babylonian) art as the Greeks and Romans met it. The Minoan art of
Crete is pure craft-art, a northern outlier of Egyptian post-Hyksos
taste; and its “contemporary,” Hellenistic-Roman art from about the
time of Scipio and Hannibal, similarly subserves the habit of comfort
and the play of intellect. From the richly-decorated entablature of the
Forum of Nerva in Rome to the later provincial ceramics in the West,
we can trace the same steady formation of an unalterable craft-art
that we find in the Egyptian and the Islamic worlds, and that we have
to presume in India after Buddha and in China after Confucius.
Now, Cathedral and Pyramid-temple are different in spite of their

deep inward kinship, and it is precisely in these differences that we
seize the mighty phenomenon of the Faustian soul, whose depth-
impulse refuses to be bound in the prime symbol of a way, and from
its earliest beginnings strives to transcend every optical limitation.
Can anything be more alien to the Egyptian conception of the State
—whose tendency we may describe as a noble sobriety—than the
political ambitions of the great Saxon, Franconian and Hohenstaufen
Emperors, who came to grief because they overleapt all political
actualities and for whom the recognition of any bounds would have
been a betrayal of the idea of their rulership? Here the prime symbol
of infinite space, with all its indescribable power, entered the field of
active political existence. Beside the figures of the Ottos, Conrad II,
Henry VI and Frederick II stand the Viking-Normans, conquerors of
Russia, Greenland, England, Sicily and almost of Constantinople;
and the great popes, Gregory VII and Innocent III—all of whom alike
aimed at making their visible spheres of influence coincident with the
whole known world. This is what distinguishes the heroes of the Grail
and Arthurian and Siegfried sagas, ever roaming in the infinite, from
the heroes of Homer with their geographically modest horizon; and
the Crusades, that took men from the Elbe and the Loire to the limits
of the known world, from the historical events upon which the
Classical soul built the “Iliad” and which from the style of that soul we
may safely assume to have been local, bounded, and completely
appreciable.
The Doric soul actualized the symbol of the corporally-present
individual thing, while deliberately rejecting all big and far-reaching
creations, and it is for this very good reason that the first post-
Mycenæan period has bequeathed nothing to our archæologists.
The expression to which this soul finally attained was the Doric
temple with its purely outward effectiveness, set upon the landscape
as a massive image but denying and artistically disregarding the
space within as the μὴ ὄv, that which was held to be incapable of
existence. The ranked columns of the Egyptians carried the roof of a
hall. The Greek in borrowing the motive invested it with a meaning
proper to himself—he turned the architectural type inside out like a
glove. The outer column-sets are, in a sense, relics of a denied
interior.[233]
The Magian and the Faustian souls, on the contrary, built high.
Their dream-images became concrete as vaultings above significant
inner-spaces, structural anticipations respectively of the mathematic
of algebra and that of analysis. In the style that radiated from
Burgundy and Flanders rib-vaulting with its lunettes and flying
buttresses emancipated the contained space from the sense-
appreciable surface[234] bounding it. In the Magian interior "the
window is merely a negative component, a utility-form in no wise yet
developed into an art-form—to put it crudely, nothing but a hole in
the wall."[235] When windows were in practice indispensable, they
were for the sake of artistic impression concealed by galleries as in
the Eastern basilica.[236] The window as architecture, on the other
hand, is peculiar to the Faustian soul and the most significant symbol
of its depth-experience. In it can be felt the will to emerge from the
interior into the boundless. The same will that is immanent in

Foundations of Quantitative Finance Book V General Measure and Integration Theory 1St Edition Robert R Reitano Full Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Foundations of Quantitative Finance Book V General Measure and Integration Theory 1St Edition Robert R Reitano Full Chapter

Uploaded by

Copyright:

Available Formats

Foundations of Quantitative Finance.

Book V: General Measure and

Financial Mathematics: A Comprehensive Treatment in Discrete Time

and by CRC Press

CRC Press is an imprint of Taylor & Francis Group, LLC

© 2024 Robert R. Reitano

Library of Congress Cataloging-in-Publication Data

Names: Reitano, Robert R., 1950- author.

ISBN: 978-1-032-20651-6 (hbk)

Typeset in CMR10 font

3 General Integration Theory 36

3.4.1 Riemann Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5 Integrals in Product Spaces 117

6 Two Applications of Fubini/Tonelli 140

7 The Fourier Transform 152

8 General Measure Relationships 180

9 The Lp Spaces 204

Robert R. Reitano is Professor of the Practice of Finance at the Brandeis International

Foundations of Quantitative Finance is structured as follows:

Example 1.3 (Semi-algebra, algebra, and sigma algebra generated by E) Gene-

Br (x) = {y|d(x, y) < r},

Definition 1.5 (Topology) A topology on a set X is a collection of subsets, denoted by

• For Lebesgue measure µ ≡ m and X = R, Definition I.2.21 also adds the

m(I) = m(I + x),

• Subadditivity: When {Ai }N i=1 is a finite (N < ∞) or countable (N = ∞) collection of

This follows from

Exercise 1.10 (Finite/countable X) Verify that if X is a finite or countable set, then

1.1 Lebesgue and Borel Spaces on R

1.1.1 Starting Points

If A ∈ P (R) , the Lebesgue outer measure of A, denoted m∗ (A), is defined in I.(2.4):

|I|F = F (b) − F (a) , (1.7)

where F (x) is an increasing, right continuous function.

where An ∈ A and µA (An ) is F -length and defined in (1.7).

The big question now becomes:

1.1.2 Lebesgue Measure Space

1. There is a Lebesgue measure on P (R), but it is not m∗ ;

Definition 1.11 (Lebesgue measurable set) A set A ⊂ R is said to be Lebesgue mea-

The collection of Lebesgue measurable sets is denoted ML ≡ ML (R).

Thus by Carathéodory’s criterion, a set A will be deemed to be Lebesgue measurable

1.1.3 Borel Measure Spaces

µ [(a, b]] = Fµ (b) − Fµ (a). (1.10)

Definition 1.12 (Measure on an algebra) A measure on an algebra, sometimes

Definition 1.13 (Carathéodory measurability w.r.t. µ∗A ) Let F (x) be an increasing,

is Carathéodory measurable with respect to µ∗A , or simply µ∗A measurable, if for

1.2 General Extension Theory

Definition 1.14 (Semi-algebra of sets) A collection A0 of sets from a space X is called

The collection A0 = A, an algebra, if in place of item 2 we have:

3. Countable Subadditivity: Given a countable collection {Aj }∞

Proposition 1.17 (Carathéodory extension theorem 1) Let µ∗ be an outer measure

Then C(X) is a complete sigma algebra.

The Hahn-Kolmogorov extension theorem of Proposition I.6.4 is named for Hans

Proposition 1.18 (Hahn–Kolmogorov extension theorem) Let A be an algebra of

The final existence result of Proposition I.6.13 is again attributable to Constantin

It should be noted that there is no loss of generality in assuming that {Bj }∞

Proposition 1.22 (Uniqueness of Extensions to σ(A)) Let µA be a σ-finite measure

1.3 Measure Space Constructions

1. Required Step (Choose a or b):

(a) Define a set function µ0 on a semi-algebra A0 which can be proved to be a pre-

(a) From 1.a, if µ0 is a pre-measure on a semi-algebra A0 , then µ0 can be extended to

1.3.1 Finite Products of Measure Spaces

where we explicitly define 0 · ∞ = 0.

1.3.2 Borel Measures on Rn

Definition 1.24 (Continuous from above; n-increasing ) Given a function Fµ :

That is, given > 0 there is a δ ≡ δ (x0 , ) > 0 so that:

|f (x) − f (x0 )| < if |x − x0 | < δ.

d2 (f (x), f (x0 )) < if d1 (x, x0 ) < δ.

Br (x0 ) ⊂ f −1 (B (y0 )) ⊂ f −1 (G).

Gn = {x| |fn (x) − f (x)| ≥ },

DN = {x| |fn (x) − f (x)| ≥ for some n ≥ N }.