Visit https://ebookfinal.

com to download the full version and

explore more ebooks

Convexity from the Geometric Point of View 1st

Edition Vitor Balestro

_____ Click the link below to download _____

https://ebookfinal.com/download/convexity-from-the-
geometric-point-of-view-1st-edition-vitor-balestro/

Explore and download more ebooks at ebookfinal.com

Here are some suggested products you might be interested in.
Click the link to download

From a Heuristic Point of View Essays in Honour of Carlo

Cellucci 1st Edition Cesare Cozzo

https://ebookfinal.com/download/from-a-heuristic-point-of-view-essays-
in-honour-of-carlo-cellucci-1st-edition-cesare-cozzo/

Notions of Convexity 1st Edition Lars Hörmander (Auth.)

https://ebookfinal.com/download/notions-of-convexity-1st-edition-lars-
hormander-auth/

The View from Within Normativity and the Limits of Self

Criticism 1st Edition Menachem Fisch

https://ebookfinal.com/download/the-view-from-within-normativity-and-
the-limits-of-self-criticism-1st-edition-menachem-fisch/

View from the Fazenda A Tale of the Brazilian Heartlands

1st Edition Ellen Bromfield Geld

https://ebookfinal.com/download/view-from-the-fazenda-a-tale-of-the-
brazilian-heartlands-1st-edition-ellen-bromfield-geld/
The View from Somewhere Undoing the Myth of Journalistic
Objectivity 1st Edition Lewis Raven Wallace

https://ebookfinal.com/download/the-view-from-somewhere-undoing-the-
myth-of-journalistic-objectivity-1st-edition-lewis-raven-wallace/

Retail Strategy The View from the Bridge 1st Edition Dr

Jonathan Reynolds

https://ebookfinal.com/download/retail-strategy-the-view-from-the-
bridge-1st-edition-dr-jonathan-reynolds/

Music Direction for the Stage A View from the Podium 1st
Edition Joseph Church

https://ebookfinal.com/download/music-direction-for-the-stage-a-view-
from-the-podium-1st-edition-joseph-church/

Point counter point 1928 2014 Aldous Huxley

https://ebookfinal.com/download/point-counter-point-1928-2014-aldous-
huxley/

Requirements and Concepts for Future Automotive Electronic

Architectures from the View of Integrated Safety Chen

https://ebookfinal.com/download/requirements-and-concepts-for-future-
automotive-electronic-architectures-from-the-view-of-integrated-
safety-chen/
Convexity from the Geometric Point of View 1st Edition
Vitor Balestro Digital Instant Download
Author(s): Vitor Balestro, Horst Martini, Ralph Teixeira
ISBN(s): 9783031505072, 3031505077
Edition: 1
File Details: PDF, 11.85 MB
Year: 2024
Language: english
 Cornerstones

Vitor Balestro
Horst Martini
Ralph Teixeira

Convexity
from the Geometric
Point of View
Cornerstones
Series Editor
Steven G. Krantz, Washington University, St. Louis, MO, USA
Vitor Balestro Horst Martini
• •

Ralph Teixeira

Convexity from the

Geometric Point of View
Vitor Balestro Horst Martini
Institute of Mathematics and Statistics Faculty of Mathematics
Fluminense Federal University Chemnitz University of Technology
Niterói, Rio de Janeiro, Brazil Chemnitz, Sachsen, Germany

Ralph Teixeira
Institute of Mathematics and Statistics
Fluminense Federal University
Niterói, Rio de Janeiro, Brazil

ISSN 2197-182X ISSN 2197-1838 (electronic)

Cornerstones
ISBN 978-3-031-50506-5 ISBN 978-3-031-50507-2 (eBook)
https://doi.org/10.1007/978-3-031-50507-2

Mathematics Subject Classification: 01-99, 01A60, 11H06, 14-02, 14M25, 14P10, 26A09,
26A51, 26B25, 26D99, 28A75, 28A78, 32F17, 46-02, 46B07, 46B20, 51M04, 51M05, 51M16,
51M20, 51M25, 52-00, 52-01, 52-02, 52-03 (01A60), 52A05, 52A20, 52A21, 52A22, 52A27,
52A30, 52A35, 52A38, 52A39, 52A40, 52A41, 52B11, 52B12, 52B20, 52B45, 52B55, 52B60,
52B70, 52C07, 52C17, 52C22, 52C35, 53-02, 53C60, 53C65, 58-02, 58A99, 58B20, 60D05,
90C05, 90C25

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer
Nature Switzerland AG 2024

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

This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the reg-
istered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

If disposing of this product, please recycle the paper.

This book is dedicated to our wives Mariana, Angelika, and
Isabel.
Preface

The intuitive notion of convexity can be roughly described as follows: a set C is convex if the
segment connecting any two points of C is contained in C. This fairly simple notion gives rise to a
very extensive theory with a plethora of remarkable results known as convex geometry (or shorter
convexity). As it is known today, this field has its historical origins already in ancient times, for
example combined with names like Archimedes and Zenodorus. In more modern times, fundaments
were laid between the second half of the nineteenth century and the first half of the twentieth
century, mainly in the works of mathematicians like J. Steiner, H. Brunn, H. Minkowski, W.
Blaschke, T. Bonnesen, W. Fenchel, H. Busemann, H. Hadwiger, and A. D. Alexandrov. Since then,
convex geometry has been immensely developed, and nowadays it is a very active and fruitful
research area, having also a lot of ramifications and relations to other areas of pure and applied
mathematics (such as functional analysis, convex analysis, discrete geometry, optimization, and
location science). It is natural that an area of knowledge having such depth and finding such a
widespread interest stimulates the appearance of books with specific objectives and scopes, and
presenting different approaches and viewpoints.
We hope that this book gives a comprehensive introduction to the “common core” of convex
geometry. It is designed to be the most didactic and self-contained text that we are able to deliver, in
such a way that an interested reader can read it cover-to-cover with relatively few pre-requisites
(namely, standard analysis and linear algebra). We also tried to make it sufficiently complete such
that the reader can acquire the needed “toolkit” to later specialize in some research subfield of
convexity. Therefore, this book is mainly directed at graduate and advanced undergraduate students
who want to pursue research in convexity. However, we also believe that it can be a good reference
source for active researchers in convex geometry and in related areas, because we present proofs of
important results in a very detailed manner and give also notes to each chapter that survey results,
concepts, and methods reaching also the research front.
More than a few words are necessary regarding the existing literature in the area. As one
comprehensive starting point in the twentieth century, besides the basic work of Minkowski (see
[1805] and [1801]) and Blaschke’s popular book [358], we should mention the monograph [446] of
Bonnesen and Fenchel. On the other hand, roughly at the end of the century, we see the undoubtedly
most important reference in the field, namely Schneider’s outstanding book [2165]. This is a very
comprehensive and deep monograph on convex geometry, providing glimpses of many ramifications
of the theory. Another fundamental reference is Gruber’s book [1070], which is more oriented to the
discrete viewpoint of the theory (despite having also a very good overview of the basic theory). In
his monograph [930], Gardner presents the field of geometric tomography which is especially rich in
deep problems and results from convex geometry. Thompson gives in [2359] a comprehensive
overview of Minkowski geometry, i.e., the geometry of finite-dimensional real Banach spaces which
has also deep connections to the geometry of convex sets. Groemer’s book [1042] covers that part of
convexity which can be attacked with tools like Fourier series and spherical harmonics, and various
vii
viii Preface

surveys in the proceedings [1415], edited by Klee, were fundamental for the development of the
whole field. The field of convex polytopes, even more combined with discrete aspects, is excellently
represented by the books [1085] of Grünbaum, [2558] of Ziegler, and [242] of Barvinok (the latter
also discussing general convex bodies). Further important and nice references are Ball’s booklet
[199] and the very recent monograph [1263] of Hug and Weil, both also suitable as lecture notes.
Moreover, the books [496] by Brazitikos et al., [133] and [134] by Artstein-Avidan, Giannopoulos,
Milman, as well as [1712] by Martini, Montejano-Peimbert, and Oliveros present introductions to
convex geometry as pre-requisite to neighbouring disciplines or more specialized topics (namely,
asymptotic geometric analysis and the more narrow subfield treating bodies of constant width,
respectively). By no means, this list of important references is complete.
Our book is (as far as it is possible for us!) completely focused on being a didactic and sufficiently
complete introduction to convex geometry. In view of many thematically related books available in
the literature, the crucial point justifying a book like ours can perhaps be described as follows: It
contains the basic tools and notions of modern convex geometry presented in a way which is
comprehensive, self-contained, and sufficiently complete such that, at the same time,

(a) it can be read even without the assistance of a specialist, and without the necessity of digging
through the literature to get a clear acquaintance with concepts and proofs, and

(b) it is sufficiently broad such that the reader can have at least a glimpse of the whole breadth
of the field and, in various subfields, can see the research front due to presented notes.

Nowadays, a newcomer has often to find paths through a large amount of scattered specialist
literature in order to acknowledge the main ideas and techniques permeating a studied field, and this
holds, of course, also for convex geometry. Due to this, it is our objective to provide a “stable
bridge” which brings the reader, with a few pre-requisites, towards enough knowledge for starting to
read research papers, and think also about research problems.
Each of the 12 chapters of the book (except for the last one) has the same structure: it contains the
so-called theory part, consisting of several sections which are ordered by themes and are written in
the theorem-proof-remarks style. It is important to say that, in great parts, the presented proofs are
not original, but taken from the references mentioned above. Of course, the presentation and writing
of them are new, and we have modified many details of the original proofs. But the main ideas
behind most of the results were taken from the existing references. Of course, we give them always
the due credit. Then, after this theory part, a section with a list of exercises is added. As is usual,
some of them are simple verifications, and some of them are quite challenging. A good amount of
exercises was taken from recent research papers, and, in some cases, they can even be thought of as
very short introductions to notions which were not dealt with in the theory part (such as certain
functional extensions of some geometric inequalities). Moreover, each chapter has a notes section,
where the literature concerning the topics of the chapter is presented. We hope these comprehensive
notes may provide glimpses of where the literature stands, and how it did come to the present stage.
To give a picture that is as complete as possible, in most cases these notes are structured due to the
following priority sequence of described publications: (parts of) books and proceedings, surveys,
expository papers, and (selected) recent research papers. Which combination is chosen depends on
the existing references. For example, in some cases the existent literature is so rich that only book
chapters and surveys can be cited (bringing more would go beyond the scope of the monograph). In
the same cases (and for the same reason), here and there undefined notions occur. However, it is easy
to find their meanings. The reader should also look below at the description of the second chapter:
there is something said about the content of the notes in this chapter.
Preface ix

Also, we added an appendix containing the basic measure theory (and geometric measure theory)
that a convex geometer might find useful to know. Actually, this is not completely true because we
use a few results which are not covered there (namely, the Riesz representation theorem for mea-
sures and some results for convergence of measures). We give detailed treatment to Hausdorff and
Lebesgue measures, and we also explain why it does make sense to work with the ðn  1Þ-
dimensional Hausdorff measure in the boundary of a convex body. It is interesting to notice that all
the theories could be developed using uniquely the Hausdorff measures, and this might even be our
personal choice, since their definitions are very easy and broad, and they also give “for granted” the
notion of “volume for objects of different dimensions”. However, since a great part of the literature
also deals with the classical Lebesgue measure, we think that it is adequate to work also with it when
it is possible. It is useful to keep in mind that all concepts of area and volume appearing throughout
the text can be (equivalently) defined using only Hausdorff measures.
In Chapter 1, we give a very short introduction to Convex functions, only containing the material
that we need to study convex sets (except for the notes). To write this chapter, we were strongly
inspired by Schneider’s book [2165] and also by the book [1846] by Mordukhovich and Nam (we
thank them for kindly sending us an ahead-of-print version of the book). For the sake of a rough
description, we study early properties and regularity properties of convex functions, and also the
extension of the gradient that they admit (the so-called subgradient). More than being an intro-
duction to the related area of convex analysis, these concepts and results are useful to investigate
support and gauge functions of convex sets. We also briefly study the Legendre–Fenchel transform,
which gives a notion of duality for convex functions, and the Prékopa–Leindler inequality. Our main
reference for this inequality is Artstein-Avidan et al. [133]. The reader should be aware of the fact
that this inequality is not exactly an inequality for convex functions, but we use it later to obtain a
simple proof of the classical Brunn–Minkowski inequality, and this is the reason why we present it in
this chapter. The notes completing this chapter collect basic references (due to the large size of this
field, only books and surveys) on convex functions and their applications, and an overview of the
literature on the Prékopa–Leindler inequality is given, too.
The first steps towards the geometric approach to Convex sets are given in Chapter 2. We study
elementary concepts such as convex and affine hulls, metric projections, support (by hyperplanes)
and separation, and the basic notion of convex body, which is the most important concept in convex
geometry. We also introduce important functions related to convex sets, such as the support, gauge
and radial functions, as well as the idea of polarity. Moving forward, we introduce the normal cones
at boundary points, which are related to the notion of support by hyperplanes, and use them to
describe the boundary structure of convex bodies. A very important tool in studying convex
geometry is endowing the collection of compact subsets of Rn with a metric (namely, the so-called
Hausdorff metric), and we also introduce and explore this idea. The last section of this chapter can
be seen as an introduction to symmetrization processes. We study the Steiner symmetrization, which
is very useful in a lot of problems (for example, we use it to give a proof of Urysohn’s inequality).
For this chapter we followed multiple references, such as Artstein-Avidan [133], Gruber [1070], and
Schneider [2165]. In the notes to this chapter, all these basics are not discussed (except Hausdorff
metric and Steiner symmetrization), since they are, in many cases, folklore and very common. Thus,
instead of discussing them also in notes, we inserted here (as subsection 2.9.1) notes on special types
of convex bodies, following the idea of the handbook article [1177] by Heil and Martini. These
special types of convex bodies are simplices, cubes, Platonic and Archimedean solids, regular
polytopes and related polytope classes, Euclidean balls, ellipsoids, centrally symmetric convex
bodies, bodies of constant width, diametrically complete sets, reduced bodies, zonotopes, and
zonoids.
x Preface

One of the important subfields of the area of convex geometry is the general theory of convex
polytopes which, roughly speaking, are the extensions of polygons and (three-dimensional) poly-
hedra to higher dimensional vector spaces. Chapter 3 is presented to provide a brief introduction to
this theory. We study the basic properties of polytopes and their description in terms of faces and
normal cones. Further on, we introduce the concept of strongly isomorphic polytopes, which is used
in the proof of the classical Alexandrov–Fenchel inequality for mixed volumes. It is important to say
that this is a very introductory material, and that the general theory of convex polytopes is clearly
much more extensive than the content of this chapter. As a criterion, we have included only the ideas
which are necessary in other parts of the book (mainly in the theory of mixed volumes). The main
reference for this chapter is Schneider [2165]. The notes mainly refer to the literature on the
approximation of convex bodies by polytopes; related book parts, surveys, and recent research
articles are presented.
In Chapter 4, we discuss Volume and area, which are central concepts in convex geometry (and
this is why we dedicate an entire chapter for that, while one may consider the theory developed there
also as a pre-requisite). The way that volumes and areas are regarded in convexity is heavily
dependent on measure theory because these notions are mostly defined in terms of Lebesgue
measures and Hausdorff measures (in contrast to what is usually done in differential geometry, for
example, where volumes and areas are obtained by integration of differential forms). For this reason,
the first section of the chapter gives an overview of these concepts, fixing ideas and concepts but
without presenting formal proofs. On the other hand, for the reader who feels the necessity of a
deeper understanding, we include an appendix with “everything that a convex geometer should
know about (geometric) measure theory”. We also discuss the so-called principle of Cavalieri,
which relates the volume of a convex body to the (smaller dimensional) volumes of sections by
affine subspaces, and derive some other formulas. Further on, we study the volume and area of
polytopes and measures on the boundary of convex bodies (in particular, we prove that some
measures defined in different ways on the unit sphere coincide). The notes to this chapter contain a
survey on Cavalieri’s principle, on volumes of cubes and (sections of) simplices, and on related
general polytope theory.
Another topic which is very central for convex geometry is that of geometric inequalities. These
are inequalities which relate geometric quantities such as volume, area, in- and circumradius,
diameter, (minimal or mean) width, and so on. We dedicate Chapter 5 to the study of the classical
geometric inequalities of convex geometry, and we give multiple proofs for the classical Brunn–
Minkowski inequality, the “strongest” (because it clarifies the equality case) being based on
Artstein-Avidan et al. [133]. We follow this reference also for the proofs of the inequalities of
Blaschke–Santaló and Urysohn. Later we observe that the latter can also be obtained via the theory
of mixed volumes. For the proof of the Rogers–Shephard inequality, our main reference is Schneider
[2165]. The comprehensive notes present the literature on the Brunn–Minkowski inequality, the
isoperimetric inequality, difference bodies and the Rogers–Shephard inequality, the Mahler product
and the Blaschke–Santaló inequality, as well as mean width and Urysohn’s inequality.
The powerful theory of Mixed volumes is developed in Chapters 6, 7, and 8. The first of these
chapters is dedicated to the basic theory: we prove the polynomiality of the volume of Minkowski
combinations of compact convex sets and define mixed volumes as the coefficients of these poly-
nomials (following Gruber [1070]). After that, we also make clear that mixed volumes can be seen
as functionals defined on (Cartesian products of) collections of compact convex sets, as it is the
approach by Martini et al. [1712]. At this first moment, we also prove the Minkowski inequalities for
mixed volumes and investigate the so-called quermassintegrals. These can be seen as specializations
of mixed volumes which carry geometric information about convex bodies. They have the disad-
vantage of being “dimension-dependent” concepts, and to overcome this problem, McMullen made
Preface xi

a slight modification in their definition to introduce the intrinsic volumes. These are also (briefly)
studied in Chapter 6. We also derive some important formulas for quermassintegrals, such as
Cauchy’s surface area formula (the proof contained here is, to the best knowledge of the authors,
new). On the other hand, Chapter 7 is dedicated to introduce the Mixed surface area measures,
which are Borel measures on the unit sphere Sn1 given by mixed volumes of convex bodies. In this
chapter we also present the classical Minkowski existence theorem, which characterizes the Borel
measures of Sn1 that are surface area measures of convex bodies. Last, we show how some mixed
surface area measures are related to the differential geometric concept of curvature (in the differ-
entiable case, of course). In Chapter 8, we give a proof (which is taken from Cordero-Erausquin
et al. [659] and Hug and Weil [1263]) of the important Alexandrov–Fenchel inequality, one of the
deepest results in the theory. We also give some consequences of this inequality, and we briefly
discuss the equality case (a complete characterization is not yet settled in the literature). The notes to
the first of these three chapters refer to mixed volumes, related Minkowski inequalities, and quer-
massintegrals. In the first two cases, they are restricted to book parts surveys, and for quermass-
integrals also selected recent research papers are collected. In the second chapter notes about
Minkowski’s existence theorem and surface area and curvature measures can be found, and both
parts take book chapters, surveys, and research papers into consideration. The third chapter contains
notes on Wulff shapes and on the Alexandrov–Fenchel inequality; both parts are dominated by
research papers.
It is an important aspect of the theory to understand what happens to some concepts, objects, and
quantities related to convex bodies under affine transformations of the considered ambient vector
space (which, for us, are compositions of translations and invertible linear transformations). It is
natural to investigate this since, in some sense, affine transformations “preserve the geometry” (for
example, they always preserve parallelism and concurrence, and also preserve volume up to a
constant scalar multiplication). We dedicate Chapter 9 to this investigation. There we prove the
existence and uniqueness of the John and the Löwner ellipsoids, and we also give a characterization
of the case where the John ellipsoid is the Euclidean unit ball. This characterization is known as
John’s theorem. We present a proof based on Gruber’s book [1070], and also a beautiful partial
proof extracted from Artstein-Avidan et al. [133]. A consequence of this theorem for the so-called
Banach–Mazur distance between convex bodies is also derived. Moving forward, we investigate the
behavior of some parameters of convex bodies under affine transformations (namely, surface area,
mean width, and quermassintegrals). The notes of this chapter refer to the Löwner–John ellipsoids
(and other associated ellipsoids) as well as positions of convex bodies. In both cases expositions are
given, but also many research papers and interesting applications are discussed.
In the literature, one can find a lot of affine constructions related to convex bodies having
particular interesting features, and in Chapter 10 we introduce and study some of them. Namely, we
look at zonotopes and zonoids, as well as at projection bodies, centroid bodies, and intersection
bodies associated with a given arbitrary convex body. There is a particular importance of projection
bodies through convexity, perhaps because they are related to the very natural problem of finding
information about a convex body from information about its projections. This is a typical task of
geometric tomography. Therefore we cite here Gardner’s book [930], which is our main reference
for this chapter. Some important inequalities involving these bodies are also derived (such as the
Petty projection inequality, which is stronger than the classical isoperimetric inequality). The notes
consist of three parts referring to zonotopes, zonoids, and bodies associated with a given convex
body. For zonotopes and zonoids comprehensive surveys are given, including book parts and
expositions, but also the research development after older basic surveys.
xii Preface

The last but one chapter of the book (Chapter 11) is a collection of (relatively) recent research
topics and results which rely heavily on the “core” theory developed in the previous chapters. We
start this chapter by studying the dual mixed volumes and the Lp Brunn–Minkowski theory, which
can be seen as extensions of the classical Brunn–Minkowski theory. They were introduced in the
1970s and in the 1990s, respectively, by Lutwak (and then investigated by many other researchers).
Next we introduce the topic of valuations which, in brief, can be described as functionals that have a
kind of “geometric meaning” (and that are defined over suitable families of sets, e.g., convex sets or
polytopes). There is a vast, deep, and exciting literature on valuations, but here we restrict ourselves
to give a complete and detailed proof of the classical theorem of Hadwiger, which states that, under
certain conditions, any valuation defined on the family of compact convex sets is a combination of
intrinsic volumes (for this task, we follow the book [1263] of Hug and Weil, where this theorem is
brilliantly presented). The last topic that we deal with is the Bourgain–Milman inequality, which is
related to the classical Mahler conjecture. The proof that we follow is via isomorphic sym-
metrizations, and it is based on Artstein-Avidan et al. [133]. It is important to mention that the
choice of the topics in this chapter is guided by the knowledge and interest of the authors, and
therefore a lot of further nice and important recent developments are not covered (as a straight-
forward example we can cite the applications of Fourier analysis to convex geometry). The notes to
this chapter are dedicated to three important notions: dual mixed volumes, valuations, and the
Bourgain–Milman inequality. In the first two parts, survey-like representations, book parts, and
recent research contributions are presented, treating also generalizations, refinements, special
aspects, and applications of these two notions. Examples of respective notions are dual Brunn–
Minkowski theory, dual Orlicz–Minkowski inequality, tensor valuations, and valuations in the
theory of (lattice) polytopes, in stochastic geometry and in local stereology. In the third part, a
relatively short representation of the important publications around the Bourgain–Milman inequality
is given.
Now we come to the last chapter. While we were working on the book, one of the most
impressive experiences we had was the unexpected observation how much, especially within the last
decades, the field of convexity has grown, including also its increasing interlocking with neigh-
boring disciplines. This motivated us to show somehow this strong development, e.g., by presenting
famous contributors and their motivations, by showing a sequence of really fundamental publica-
tions and describing the interactions with neighboring areas. We observed the fast growth of
overlaps with these disciplines and wanted to present also this. But since we understood very soon
that we are not able to write a “history of convexity” (this demands wider methods of representation
and more comparative care with mathematical and historical details), we decided to present all this
knowledge at least in the style of concatenated building blocks and historical sketches. This rep-
resentation starts with Greek antiquity and important progress from the seventeenth to the nineteenth
century, and it continues (regarding the twentieth and twenty-first centuries) with information about
important contributors, essential publications (research monographs, textbooks, problem books,
proceedings, related handbooks, surveys, etc.), and closely related fields (discrete geometry, poly-
tope theory, convex algebraic geometry, differential geometry, integral and stochastic geometry,
convex analysis, and asymptotic convex geometry)
We finish this preface with some explanation about the title of this book. Despite being
undoubtedly a book on convex geometry, we believe that, in principle, geometry is rather a method
than a subject. Having this in mind, we think that studying convex geometry is one of the possible
paths to understand the idea of convexity in its full scope and strength. We hope, then, that this book
will be useful for anyone who wants to dive into this fantastic world, and not only for those who
want to pursue research in convex geometry itself. Readers with the aim to capture the powerful
Preface xiii

geometric side of convexity might certainly learn methods and get abilities which can be suc-
cessfully applied in different fields of pure and applied mathematics.
The authors wish to thank André Martini for producing the 12 plates which are placed at the
beginning of each chapter.

Niterói, Brazil Vitor Balestro

Chemnitz, Germany Horst Martini
Niterói, Brazil Ralph Teixeira
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

1 Convex functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 First steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Subgradients and subdifferentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5 The Prékopa–Leindler inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
1.7.1 Properties of convex functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.7.2 The Prékopa–Leindler inequality and related topics . . . . . . . . . . . . . . . . . 57

2 Convex sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.1 Basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2 Support and separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.3 Gauge and support functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.4 Convex cones and boundary structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
2.5 The Hausdorff metric. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
2.6 Approximations and continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
2.7 Steiner symmetrizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
2.9 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
2.9.1 Special types of convex bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
2.9.1.1 Simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
2.9.1.2 Cubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
2.9.1.3 Platonic and Archimedean solids . . . . . . . . . . . . . . . . . . . . . . . 176
2.9.1.4 Regular, semiregular, and uniform polytopes . . . . . . . . . . . . . . 188
2.9.1.5 Euclidean balls and spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
2.9.1.6 Ellipsoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.9.1.7 Centrally symmetric convex bodies . . . . . . . . . . . . . . . . . . . . . 204
2.9.1.8 Bodies of constant width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
2.9.1.9 Complete sets and completions . . . . . . . . . . . . . . . . . . . . . . . . 213
2.9.1.10 Reduced bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
2.9.1.11 Zonotopes and zonoids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
2.9.2 The Hausdorff metric in convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
2.9.3 Steiner symmetrizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

xv
xvi Contents

3 A first look into polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

3.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.2 Faces and normal cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
3.3 Strongly isomorphic polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
3.4 Approximations by polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
3.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
3.6.1 Strongly isomorphic polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
3.6.2 Approximation of convex bodies by polytopes . . . . . . . . . . . . . . . . . . . . 265

4 Volume and area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
4.2 Cavalieri’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
4.3 Some formulas for volumes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
4.4 Volumes and areas of polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
4.5 Measures on the boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
4.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
4.7.1 Cavalieri’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
4.7.2 Volumes of unit balls and related topics . . . . . . . . . . . . . . . . . . . . . . . . . 308
4.7.3 Volumes (of sections) of simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
4.7.4 Volumes and areas of polytopes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

5 Classical inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323

5.1 The Brunn–Minkowski inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
5.2 The isoperimetric inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
5.3 Difference bodies and the Rogers–Shephard inequality . . . . . . . . . . . . . . . . . . . . . 331
5.4 The Mahler product and the Blaschke–Santaló inequality . . . . . . . . . . . . . . . . . . . 334
5.5 Mean width of a convex body and Urysohn’s inequality . . . . . . . . . . . . . . . . . . . 338
5.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
5.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
5.7.1 The Brunn–Minkowski inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
5.7.2 The isoperimetric inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
5.7.3 Difference bodies and the Rogers–Shephard inequality . . . . . . . . . . . . . . 359
5.7.4 The Mahler product and the Blaschke–Santaló inequality . . . . . . . . . . . . 364
5.7.5 Mean width and Urysohn’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . 368

6 Mixed volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377

6.1 Polynomiality of the volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
6.2 Properties of mixed volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
6.3 Minkowski inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
6.4 Quermassintegrals and intrinsic volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
6.5 Formulas for quermassintegrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
6.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
6.7.1 Mixed volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
6.7.2 Minkowski’s inequalities related to mixed volumes . . . . . . . . . . . . . . . . . 426
6.7.3 Quermassintegrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
Contents xvii

7 Mixed surface area measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

7.1 Definition and properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
7.2 j-th order surface area measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
7.3 Minkowski’s existence theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
7.4 Surface area measures and curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
7.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
7.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
7.6.1 Minkowski’s existence theorem and related topics . . . . . . . . . . . . . . . . . 484
7.6.2 Surface area measures and curvature measures . . . . . . . . . . . . . . . . . . . . 487

8 The Alexandrov–Fenchel inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

8.1 Wulff shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
8.2 A proof of the inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
8.3 Other versions and consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
8.4 A brief discussion of the equality case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
8.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
8.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
8.6.1 Wulff shapes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
8.6.2 The Alexandrov–Fenchel inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566

9 Affine convex geometry – Part 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571

9.1 John and Löwner ellipsoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
9.2 John’s theorems for the centrally symmetric case . . . . . . . . . . . . . . . . . . . . . . . . . 582
9.3 Minimal positions I: surface area and mean width . . . . . . . . . . . . . . . . . . . . . . . . 594
9.4 Minimal positions II: quermassintegrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613
9.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
9.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
9.6.1 Löwner–John ellipsoids and related topics . . . . . . . . . . . . . . . . . . . . . . . . 629
9.6.2 Positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635

10 Affine convex geometry – Part 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639

10.1 Minkowski classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
10.2 Zonotopes and zonoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
10.3 Projection, centroid, and intersection bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650
10.4 Some related geometric inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667
10.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
10.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
10.6.1 Zonotopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
10.6.2 Zonoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
10.6.3 Projection, centroid, and intersection bodies . . . . . . . . . . . . . . . . . . . . . . 716

11 Further selected topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 731

11.1 Dual mixed volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 732
11.2 The Lp Brunn–Minkowski theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 741
11.3 Valuations on convex bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755
11.4 Inequalities for covering numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 779
11.5 The Bourgain–Milman inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 790
xviii Contents

11.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 802

11.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 809
11.7.1 Dual mixed volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 809
11.7.2 Valuations of convex bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 811
11.7.3 The Bourgain–Milman inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815

12 Historical steps of development of convexity as a field . . . . . . . . . . . . . . . . . . . . . . . . 817

12.1 Greek antiquity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 818
12.2 From the seventeenth to the nineteenth century . . . . . . . . . . . . . . . . . . . . . . . . . . 819
12.3 The twentieth century and the twenty-first century . . . . . . . . . . . . . . . . . . . . . . . . 821
12.3.1 Important contributors and authors of research monographs . . . . . . . . . . 821
12.3.2 Further books on convexity after 1950. . . . . . . . . . . . . . . . . . . . . . . . . . . 840
12.3.3 Problem books and problem-oriented surveys . . . . . . . . . . . . . . . . . . . . . 847
12.3.4 Related proceedings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 848
12.3.5 Handbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 852
12.3.6 Selected and collected works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855
12.3.7 Examples of classical geometry books containing sections about
convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 857
12.3.8 A collection of surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 858
12.4 Development of the overlaps with neighbouring disciplines . . . . . . . . . . . . . . . . . 864
12.4.1 Discrete (and combinatorial) geometry of convex bodies . . . . . . . . . . . . 864
12.4.1.1 Description of the field and some history . . . . . . . . . . . . . . . . 865
12.4.1.2 Geometry of numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866
12.4.1.3 Combinatorial geometry of convex bodies . . . . . . . . . . . . . . . . 868
12.4.1.4 Scissor’s congruence (Hilbert’s third problem) . . . . . . . . . . . . 871
12.4.1.5 Tilings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 872
12.4.1.6 Further books (and directly related expositions) treating
discrete geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 875
12.4.1.7 Handbooks and proceedings related to discrete geometry . . . . 879
12.4.1.8 Important further surveys on discrete geometry . . . . . . . . . . . . 880
12.4.1.9 Selected monographs from computational geometry . . . . . . . . 882
12.4.2 Polytopes, polyhedral sets, and related geometric figures . . . . . . . . . . . . 883
12.4.2.1 Monographs on polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884
12.4.2.2 Proceedings referring to polytope theory . . . . . . . . . . . . . . . . . 897
12.4.2.3 Selected surveys on polytopes . . . . . . . . . . . . . . . . . . . . . . . . . 900
12.4.3 Convex algebraic geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 909
12.4.3.1 Some history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 909
12.4.3.2 Books and book chapters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 910
12.4.3.3 Handbooks and proceedings . . . . . . . . . . . . . . . . . . . . . . . . . . . 914
12.4.3.4 Selected surveys and expository papers . . . . . . . . . . . . . . . . . . 915
12.4.4 Differential Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 918
12.4.4.1 Introduction and historical steps . . . . . . . . . . . . . . . . . . . . . . . 919
12.4.4.2 Books and book chapters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 921
12.4.4.3 Convexity-related proceedings . . . . . . . . . . . . . . . . . . . . . . . . . 923
12.4.4.4 Selected surveys between convexity and differential
geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 924
Contents xix

12.4.5 Integral geometry and stochastic geometry . . . . . . . . . . . . . . . . . . . . . . . 926

12.4.5.1 Introduction and important historical steps . . . . . . . . . . . . . . . 926
12.4.5.2 Further monographs and book chapters on integral
geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 928
12.4.5.3 Monographs and book chapters closer to geometric
probability and stochastic geometry . . . . . . . . . . . . . . . . . . . . . 933
12.4.5.4 Related proceedings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 934
12.4.5.5 Selected surveys on integral geometry . . . . . . . . . . . . . . . . . . . 935
12.4.5.6 Selected surveys on stochastic geometry . . . . . . . . . . . . . . . . . 938
12.4.5.7 Surveys on random polytopes and mosaics . . . . . . . . . . . . . . . 938
12.4.6 Convex analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 942
12.4.6.1 Historical remarks and a short description of the field . . . . . . 942
12.4.6.2 Duality and conjugacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944
12.4.6.3 Subdifferentials, subgradients, and related notions . . . . . . . . . . 946
12.4.6.4 Nonlinear analysis and calculus of variations . . . . . . . . . . . . . 950
12.4.6.5 Functionals and operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 952
12.4.6.6 Generalized convexity notions in convex analysis . . . . . . . . . . 953
12.4.6.7 Nonsmooth analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 956
12.4.6.8 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 957
12.4.6.9 Discrete convex analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 958
12.4.7 Convexity and Banach spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 959
12.4.7.1 Foundations and introductory remarks . . . . . . . . . . . . . . . . . . . 960
12.4.7.2 Convexity-related monographs in the geometry of Banach
spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 961
12.4.7.3 Proceedings devoted to the geometry of Banach spaces . . . . . 967
12.4.7.4 Selected surveys on the geometry of real Banach spaces . . . . 971

A Measure theory for convex geometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 979

A.1 Outer measures and measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 980
A.2 Measurable functions and integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 991
A.3 Lebesgue measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1011
A.4 Hausdorff measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1029
A.5 Product measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1039
A.6 Area and co-area formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1058
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1063
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1175
Chapter 1
Convex functions

“Mathematics is the gate and the key

to science.”
— Roger Bacon

Convex functions and convex sets are somewhat interdependent. When studying convex sets, some
convex functions play a very important role (mainly the support function of a convex set, or the radial
function of a star-shaped set). Conversely, some geometric concepts from the theory of convex sets
(such as support and separation) are very useful to understand convex functions. Plate 1 shows the
graphs of two types of convex functions and (below) the level curves of a convex function used in
location science (and generalizing the notion of ellipse). In this book, we are mainly interested in the
geometric aspects of convexity, and hence we only introduce some basic facts on convex functions as
a preliminary step. This is the reason why we start with convex functions.
As we will see in this chapter, convex functions have a lot of remarkable properties. In particular,
they have “good” regularity features, and even in the nonsmooth points we can define some notion of
“gradient” (the so-called subgradient). We also introduce some features of convex functions which are
important in convex geometry, such as the Jensen inequality and the Legendre transform.
Also in this chapter we present a proof of the Prékopa–Leindler inequality. It is important to observe
that this inequality is not restricted to convex functions (rather, we assume a certain “log-concavity”
property of the involved functions). This inequality can be seen as a functional version of the so-called
Brunn–Minkowski inequality, which is the most important inequality in convex geometry. We actually
use the Prékopa–Leindler inequality to give a very easy proof of the Brunn–Minkowski inequality in
Chapter 5, but we also give a proof of the latter that does not rely on the former. The Prékopa–Leindler
inequality is also a central result to understand the geometry of the so-called log-concave functions. In
some sense, and in some contexts, these functions “replace” convex sets leading to functional versions
of geometric results. Such versions are discussed in several exercises, almost at the end of this chapter
and also throughout the whole book.
The chapter ends with notes which give an overview to the literature treating the mentioned topics.
We hope that, particularly regarding convex functions, this survey is also very useful for readers
interested in convex analysis.

1.1 First steps

We work in the space Rn , for some n ∈ N, with its standard topology given by  the open balls
B(x, ρ) = BRn (x, ρ) := {z ∈ Rn : |x − z| < ρ} for x ∈ Rn and ρ > 0. Here, | · | := ·, · denotes the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 1

V. Balestro et al., Convexity from the Geometric Point of View, Cornerstones,
https://doi.org/10.1007/978-3-031-50507-2_1
2 1 Convex functions

usual Euclidean norm in Rn , given by the canonical inner product ·, · : Rn × Rn → R, defined as

(x1, . . . , xn ), (y1, . . . , yn ) := x1 y1 + . . . + xn yn .

The (closed) unit ball and unit sphere of Rn are, respectively, the sets

B n := {x ∈ Rn : |x| ≤ 1} and
Sn−1 := {x ∈ Rn : |x| = 1}.

The metric (and topology) considered on Sn−1 will always be the one given from the norm of Rn . This
means that the distance between points x, y ∈ Sn−1 is always assumed to be simply |x − y|. The closed
ball with center x ∈ Rn and radius ρ > 0 is the closure of the open ball B(x, ρ):

cl(B(x, ρ)) = {z ∈ Rn : |x − z| ≤ ρ}.

This will be often denoted as x + ρB n . This terminology comes from the operations of Minkowski sum
and scalar multiplication of sets defined as

A + B := {x + y : x ∈ A and y ∈ B} and
λA := {λx : x ∈ A},

respectively, for A, B ⊆ Rn and λ ∈ R. We will deal with these concepts in greater detail later.
For the convenience of the reader, we also recall some topological definitions that will be used
throughout the text.

(i) The interior of a set A ⊆ Rn , denoted by intA, is the set of points x ∈ A for which there is a number
ρ > 0 such that B(x, ρ) ⊆ A. We say that a set is open if it coincides with its interior.

(ii) The closure of A ⊆ Rn , denoted by clA, is the set of points x ∈ Rn for which there exists a sequence
(xm )m∈N of points of A such that xm → x as m → ∞. We say that a set is closed if it is equal to its
closure.

(iii) The boundary of A ⊆ Rn , denoted by bdA or ∂ A, is the set of points x ∈ Rn with the property that
for every ε > 0 the open ball B(a, ε) contains points of both A and Rn \ A.

Remark 1.1.1. A standard topology argument gives that for any set A ⊆ Rn we have clA = intA ∪ ∂ A.
This equality will be used throughout the text.

In general, a linear transformation (or linear map) between vector spaces V and W is a map
T : V → W such that T(u + λv) = T(u) + λT(v) for any u, v ∈ V and every λ ∈ R. A linear
transformation of Rn is a linear transformation T : Rn → Rn . A linear map is called an isomorphism
if it is invertible. An affine transformation (map) of Rn is a map A: Rn → Rn having the form
A(·) = x0 + T(·) for some x0 ∈ Rn and some isomorphism T : Rn → Rn . In other words, an affine
transformation is a composition of an isomorphism of Rn with a translation.
Before studying convex functions, we shall start by the foundational geometric idea behind convexity
theory. A subset A ⊆ Rn is called convex if

(1 − λ)x + λy ∈ A
1.1 First steps 3

for any x, y ∈ A and λ ∈ [0, 1]. Of course, A ⊆ Rn is convex if and only if (1 − λ)A + λA = A for every
λ ∈ [0, 1]. In other words, a subset of A ⊆ Rn is convex when it contains the segment seg[x, y] for any
x, y ∈ A (see Figure 1.1.1).

Fig. 1.1.1 A convex set and a non-convex set.

The epigraph of a function f : Rn → (−∞, +∞] is the set

epi( f ) = {(x, t) ∈ Rn+1 : f (x) ≤ t},

and f is said to be convex if its epigraph is a convex subset of Rn+1 . A function f : Rn → (−∞, +∞]
is said to be proper if { f = +∞} := {x ∈ Rn : f (x) = +∞}  Rn , and real-valued (or finite) if
f (x) < +∞ for any x ∈ Rn , that is, if the co-domain of f is (contained in) R. Next, we present an
equivalent definition for convexity of a function.

Lemma 1.1.1. A function f : Rn → (−∞, +∞] is convex if and only if

f ((1 − λ)x + λy) ≤ (1 − λ) f (x) + λ f (y)

for any x, y ∈ Rn and λ ∈ [0, 1]. See Figure 1.1.2 for an illustration.

Proof. First, assume that f is convex. Hence, since (x, f (x)) and (y, f (y)) are obviously points of the
convex set epi( f ) for any x, y ∈ Rn , we have that

(1 − λ)(x, f (x)) + λ(y, f (y)) ∈ epi( f )

for any x, y ∈ Rn and every λ ∈ [0, 1]. The inequality follows immediately.
For the converse, let (x, t) and (y, s) be arbitrary points of epi( f ). Then we have the inequalities
f (x) ≤ t and f (y) ≤ s. We have to prove that

(1 − λ)(x, t) + λ(y, s) ∈ epi( f )

for every λ ∈ [0, 1]. Since (1 − λ)(x, t) + λ(y, s) = ((1 − λ)x + λy, (1 − λ)t + λs), this is the same as
proving that

f ((1 − λ)x + λy) ≤ (1 − λ)t + λs.

But since (1 − λ) f (x) ≤ (1 − λ)t and λ f (y) ≤ λs, we get from the hypothesis that

f ((1 − λ)x + λy) ≤ (1 − λ) f (x) + λ f (y) ≤ (1 − λ)t + λs,

as we wished. This concludes the proof. 

Corollary 1.1.1 (Discrete Jensen’s inequality). Let f : Rn → (−∞, +∞] be a convex function. Then
we have
4 1 Convex functions

Fig. 1.1.2 How a convex function looks like.

f (λ1 x1 + . . . + λm xm ) ≤ λ1 f (x1 ) + . . . + λm f (xm )

for any m ∈ N, any points x1, . . . , xm ∈ Rn , and any numbers λ1, . . . , λm ∈ [0, +∞) such that
λ1 + . . . + λm = 1.

Proof. The case where m = 1 is trivial, and hence we assume that m ≥ 2. We proceed by induction
on m. The case m = 2 is precisely the previous lemma. Hence we assume that the inequality is
valid for m ∈ N. Let x1, . . . , xm+1 ∈ Rn be arbitrary points, and λ1, . . . , λm+1 ∈ [0, +∞) be such that
λ1 + . . . + λm+1 = 1. Assume that λm+1  1, since otherwise the desired inequality is trivial. We write


m+1 
m
λj
λ j x j = λm+1 xm+1 + (1 − λm+1 ) xj
j=1 j=1
1 − λm+1

and observe that the sum of the coefficients in the last sum equals 1:

m
λj λ1 + . . . + λm 1 − λm+1
= = = 1.
j=1
1 − λm+1 1 − λm+1 1 − λm+1

Therefore, using the inequalities for 2 and for m points, respectively, we get


m
λj 
f (λ1 x1 + . . . + λm+1 xm+1 ) ≤ λm+1 f (xm+1 ) + (1 − λm+1 ) f  xj  ≤
1 − λ m+1
 j=1
 m
λj
≤ λm+1 f (xm+1 ) + (1 − λm+1 ) f (x j ) = λ1 f (x1 ) + . . . + λm+1 f (xm+1 ),
j=1
1 − λm+1

and this finishes the proof.



Example 1.1.1 (Affine functions). An affine function is a function f : Rn → R of the form f (x) =
x, u + α, for some u ∈ Rn and α ∈ R. Any affine function is a convex proper function. Indeed, if
x, y ∈ Rn and λ ∈ [0, 1], then

f ((1 − λ)x + λy) = (1 − λ)x + λy, u + α = (1 − λ)x, u + λy, u + (1 − λ)α + λα =

= (1 − λ)(x, u + α) + λ(y, u + α) = (1 − λ) f (x) + λ f (y).
1.1 First steps 5

Proposition 1.1.1. Assume that f1, f2 : Rn → (−∞, +∞] are convex functions. Then f1 + f2 is a
convex function. Moreover, if { f j : j ∈ J} is any family of convex functions, then its superior envelope
sup f j : Rn → (−∞, +∞] defined as

sup f j (x) = sup{ f j (x) : j ∈ J}

for each x ∈ Rn is also a convex function.

Proof. The fact that the sum of convex functions is convex is obvious, and the proof is left for the
reader. Let { f j : j ∈ J} be a family of convex functions. If x, y ∈ Rn and λ ∈ [0, 1], then

sup f j ((1 − λ)x + λy) = sup{ f j ((1 − λ)x + λy) : j ∈ J} ≤

≤ (1 − λ) sup{ f j (x) : j ∈ J} + λ sup{ f j (y) : j ∈ J} = (1 − λ) sup f j (x) + λ sup f j (y),

where the inequality above comes from the fact that f j ((1 − λ)x + λy) ≤ (1 − λ) f j (x) + λ f j (y) for any
j ∈ J, and from elementary facts of the supremum.


As an easy consequence of the definition, we have that the sublevel sets { f ≤ t} := {x ∈ Rn :

f (x) ≤ t} (and also sets { f < t} defined similarly) of a convex function f : Rn → (−∞, +∞] are
convex subsets of Rn (the converse is not true). The effective domain of a function f : Rn → (−∞, +∞]
is the set

dom( f ) = {x ∈ Rn : f (x) < +∞}.

It is easy to see that if f is convex, then dom( f ) is a convex set. If x, y ∈ dom( f ) and λ ∈ [0, 1], then we
have f ((1−λ)x +λy) ≤ (1−λ) f (x)+λ f (y) < +∞. Therefore, we have indeed that seg[x, y] ⊆ dom( f ).
The convex hull of a subset A ⊆ Rn is the set of the convex combinations of the points of A:
⎧
⎨
⎪ m 
m ⎫
⎬
⎪
conv(A) := λ j x j : x j ∈ A and λ j ≥ 0 for any j = 1, . . . , m, and λj = 1 . (1.1.1)
⎪ ⎪
⎩ j=1 j=1 ⎭
An illustration of this concept is given in Figure 1.1.3. See also Figure 1.1.4.

Fig. 1.1.3 A set and its convex hull.

Lemma 1.1.2. The convex hull of any set is a convex set.

 k
Proof. Let A ⊆ Rn , and assume that x = m j=1 β j x j and y = i=1 αi yi are points of conv(A), where
all the coefficients and points are given as in (1.1.1). If λ ∈ [0, 1], then
6 1 Convex functions


m 
k
(1 − λ)x + λy = (1 − λ)β j x j + λαi yi,
j=1 i=1

and we claim that this last combination is a point of conv(A). Indeed, we have that (1 − λ)β j ≥ 0 for
each j = 1, . . . , m and λαi ≥ 0 for every i = 1, . . . , k. Further, the sum of all the coefficients is


m 
k 
m

m k

(1 − λ)β j + λαi = βj + λ  βj − αi  = 1 + λ(1 − 1) = 1.
j=1 i=1 j=1  j=1 i=1

Hence (1 − λ)x + λy ∈ conv(A), and the proof is done.



An affine combination of a set of points {x1, . . . , xm } ∈ Rn is a sum of the form


m
λ j x j,
j=1

where each λ j is a real number, and λ1 + . . . + λm = 1. We say that the points x1, . . . , xm ∈ Rn are
affinely independent if none of them is an affine combination of the others. Observe that the cardinality
of an affinely independent set in Rn is at most n + 1, because if x0, . . . , xm ∈ Rn are m + 1 affinely
independent points, then the m vectors x1 − x0, x2 − x0, . . . , xm − x0 are linearly independent (see
Proposition 2.1.3).
The convex hull of a finite set of points is called a polytope (or convex polytope). See Figure 1.1.4.
A k-simplex is the convex hull of k + 1 affinely independent points (see Figure 1.1.5). In this case,
the number k is the affine dimension of the simplex. A simplex of maximal dimension (that is, an
n-simplex in Rn ) is shortly called a simplex.

Fig. 1.1.4 The convex hull of finitely many points (that is, a polytope).

For the convenience of the reader, we recall that a function f : A ⊆ Rn → (−∞, ∞] is said to be a
Lipschitz function if there exists a (non-negative) constant c ∈ R such that for any x, y ∈ A we have
that

| f (x) − f (y)| ≤ c|x − y|.

Theorem 1.1.1. Let f : Rn → (−∞, +∞] be a convex function. Then f is continuous in the interior of
its effective domain. Moreover, f is a Lipschitz function on any compact subset of int(dom( f )).
1.1 First steps 7

Fig. 1.1.5 A 2-simplex and a 3-simplex.

Proof. Let us start by proving that f is continuous at any point x0 ∈ int(dom( f )). Choose a number
ρ > 0 and a simplex S = conv{x1, . . . , xn+1 } such that x0 ∈ int(S) ⊆ S ⊆ int(dom( f )) and
cl(BRn (x0, ρ)) ⊆ S. Thus, for each x ∈ S we can write


n+1
x= λ j x j,
j=1

with λ j ≥ 0 for each j ∈ {1, . . . , n + 1} and λ1 + . . . + λm+1 = 1. From Jensen’s inequality (Corollary
1.1.1) we get that


n+1
f (x) ≤ λ j f (x j ) ≤ max{ f (x1 ), . . . , f (xn+1 )} =: M.
j=1

Now, if y ∈ cl(BRn (x0, ρ)), we may write y = x0 + αz for some number α ∈ [0, 1] and some vector
z ∈ Rn with |z| = ρ. Since y = (1 − α)x0 + α(x0 + z), convexity implies that

f (y) ≤ (1 − α) f (x0 ) + α f (x0 + z) = f (x0 ) + α( f (x0 + z) − f (x0 )) ≤ f (x0 ) + α(M − f (x0 )),

where the last inequality follows from the fact that x0 + z ∈ cl(BRn (x0, ρ)) (⊆ S). Therefore, we have

f (y) − f (x0 ) ≤ α(M − f (x0 )).

We also obtain
1 α
x0 = y+ (x0 − z),
1+α 1+α
and hence, again by convexity, it follows that

(1 + α) f (x0 ) ≤ f (y) + α f (x0 − z) ≤ f (y) + αM,

where the last inequality comes from the fact that x0 − z ∈ cl(BRn (x0, ρ)) (⊆ S). Consequently, we have

f (x0 ) − f (y) ≤ α(M − f (x0 )),

from which we get

8 1 Convex functions

| f (y) − f (x0 )| ≤ α(M − f (x0 )),

for any y ∈ cl(BRn (x0, ρ)). Finally, we notice that α = |y − x0 |/ρ and rewrite the inequality above as

M − f (x0 )
| f (y) − f (x0 )| ≤ |y − x0 |,
ρ

and this clearly proves that f is continuous at x0 ∈ int(dom( f )).

Now let C ⊆ int(dom( f )) be a compact subset. Since its distance to the exterior of int(dom( f )) is
strictly positive (because it is the distance between a compact set and a disjoint closed set), we have
that there exists a number ρ > 0 such that

Cρ := C + cl(BRn (0, ρ)) ⊆ int(dom( f )).

Since f is continuous on int(dom( f )) and Cρ is clearly compact, we get that | f | attains a maximum
value (M0 , say) on Cρ . If x, y are any distinct points in C, then we have
ρ
z=y+ (y − x) ∈ cl(BRn (y, ρ)) ⊆ Cρ .
|y − x|

Rewriting the equality as |y − x|z = (|y − x| + ρ)y − ρx, we have that y = (1 − λ)x + λz, where

|y − x|
λ= .
|y − x| + ρ
Therefore, from the convexity of f we get that

f (y) ≤ (1 − λ) f (x) + λ f (z) = f (x) + λ( f (z) − f (x)) ≤ f (x) + 2M0 λ.

This implies

2M0 |y − x| 2M0
f (y) − f (x) ≤ 2M0 λ = ≤ |y − x|.
|y − x| + ρ ρ

Interchanging the roles of x and y, and repeating the argument, we have that f (x) − f (y) ≤ β|y − x|
for some constant β ∈ R which does not depend on x and y. It follows that f is Lipschitzian on C, as
we wished to prove.


Another important feature of convex functions is that any local minimum is a global minimum.
This is formally stated and proved next.

Proposition 1.1.2. Let f : Rn → (−∞, +∞] be a convex function. Assume that x0 ∈ Rn is a point with
the following properties:

(i) f (x0 ) < +∞,

(ii) x0 minimizes f in some non-degenerate ball BRn (x0, ρ).

Then f (x0 ) ≤ f (x) for every x ∈ Rn . In other words, if x0 is a local minimum, then x0 is a global
minimum.

Proof. From the hypothesis, we have that f (x0 ) ≤ f (x) whenever |x − x0 | ≤ ρ. Then assume that
|x − x0 | > ρ, and let
1.1 First steps 9
 
ρ ρ
y= x+ 1− x0 .
|x − x0 | |x − x0 |

We clearly have that |y − x0 | = ρ, from which it follows that f (x0 ) ≤ f (y). Hence, convexity of f
gives that
 
ρ ρ
f (x0 ) ≤ f (y) ≤ f (x) + 1 − f (x0 ),
|x − x0 | |x − x0 |

implying immediately that f (x0 ) ≤ f (x) by rearranging the terms.



Remark 1.1.2. A function f : Rn → [−∞, +∞) is said to be concave if − f is convex. It follows

immediately that any local maximum of a concave function is a global maximum.

A function f : Rn → (−∞, +∞] is lower semi-continuous or closed if its epigraph is a closed subset
of Rn+1 (see Figure 1.1.6). Below, we give several equivalent definitions for lower semi-continuity.

Fig. 1.1.6 A lower semi-continuous function.

Proposition 1.1.3. Let f : Rn → (−∞, +∞] be a function. Then the following properties are equiva-
lent:

(a) f is lower semi-continuous,

(b) for any t ∈ R the sublevel set { f ≤ t} is closed,

(c) for every x ∈ Rn and every ε > 0 there exists a neighborhood V of x such that f (y) ≥ f (x) − ε for
any y ∈ V, and

(d) for any x ∈ Rn and any sequence (xn )n∈N in Rn such that xn → x the inequality

lim inf f (xn ) ≥ f (x)

n→∞

holds.

Proof. We start with (a)⇒(b). Given an arbitrary t ∈ R, let (xn )n∈N be a sequence in the sublevel set
{ f ≤ t} such that xn → x for some x ∈ Rn . Since f (xn ) ≤ t for any n ∈ N, it follows that (xn, t)n∈N is
a sequence of points in epi( f ), and of course we have (xn, t) → (x, t). Since epi( f ) is closed, we have
that (x, t) ∈ epi( f ), yielding x ∈ { f ≤ t}.
10 1 Convex functions

Now we prove (b)⇒(c). Assume that (c) is not true. Then there exist a point x ∈ Rn and a number
ε > 0 with the property that every neighborhood V of x contains a point y such that f (y) ≤ f (x) − ε. In
particular, for any n ∈ N we can choose a point yn ∈ BRn (x, 1/n) such that f (yn ) ≤ f (x) − ε. It follows
that (yn )n∈N is a sequence in the sublevel set { f ≤ f (x) − ε}, which converges to x  { f ≤ f (x) − ε},
and hence the sublevel set is not closed. This contradiction proves the implication.
To show (c)⇒(d), notice that if xn → x and lim inf n→∞ f (xn ) < f (x), then there exists a subse-
quence (xnk ) such that f (xnk ) → f (x) − 2ε for some ε > 0. It follows that we have f (xnk ) < f (x) − ε
for k sufficiently large, and this contradicts (c).
It remains to prove that (d)⇒(a). Let (xn, tn )n∈N be a sequence in epi( f ) such that (xn, tn ) → (x, t).
Hence we have that xn → x and tn → t. From the hypothesis, we get that

f (x) ≤ lim inf f (xn ) ≤ lim inf tn = t,

n→∞ n→∞

and therefore (x, t) ∈ epi( f ). This proves that epi( f ) is closed, and hence f is lower semicontinuous.


Next we present some important properties of lower semi-continuous functions.

Proposition 1.1.4. (i) The sum of lower semi-continuous functions is lower semi-continuous.

(ii) Let { f j : Rn → (−∞, +∞] : j ∈ J} be a family of lower semi-continuous functions. Then its
superior envelope sup j ∈J f j : Rn → (−∞, +∞] is lower semi-continuous.

(iii) If f : Rn → (−∞, +∞] is lower semi-continuous and C ⊆ Rn is compact, then inf x ∈C f (x) is
attained for some point of C.

Proof. Let f , g : Rn → (−∞, +∞] be lower semi-continuous functions. If xn → x in Rn , then

lim inf ( f + g)(xn ) ≥ lim inf f (xn ) + lim inf g(xn ) ≥ f (x) + g(x) = ( f + g)(x),
n→∞ n→∞ n→∞

and this shows (i).

For (ii), observe that sup j ∈J f j (x) ≤ t if and only if f j (x) ≤ t for every j ∈ J. Therefore, we get

{sup j ∈J f j ≤ t} = { f j ≤ t}.
j ∈J

It follows that each sublevel set {sup j ∈J f j ≤ t} is an intersection of closed sets, and hence is closed.
It remains to prove (iii). Assume that f : Rn → (−∞, +∞] is lower semi-continuous and that
C ⊆ Rn is compact. First, let us show that inf x ∈C f (x) > −∞. Indeed, assume that there exists a
sequence (xn )n∈N in C such that f (xn ) → −∞. Passing to a subsequence if necessary, we can assume
that xn → x0 for some x0 ∈ C. Therefore, for sufficiently large n we have that (xn, f (x0 ) − 1) ∈ epi( f ).
On the other hand, it is clear that (xn, f (x0 ) − 1) → (x0, f (x0 ) − 1)  epi( f ), which contradicts the fact
that f is lower semi-continuous. Now assume that (xn )n∈N is a sequence in C such that

lim f (xn ) = inf x ∈C f (x).

n→∞

Again considering a subsequence if necessary, we may assume that xn → x0 for a point x0 ∈ C.

Therefore, since (xn, f (xn ))n∈N is a sequence of points in epi( f ) and

(xn, f (xn )) → (x0, inf x ∈C f (x)),

1.2 Regularity 11

it follows from the closedness of epi( f ) that f (x0 ) ≤ inf x ∈C f (x). Since the inequality f (x0 ) ≥
inf x ∈C f (x) is obvious we get the desired equality.


1.2 Regularity

In this section we will discuss some differentiability properties of convex functions. We start with the
case where the domain is one-dimensional. If f : R → R is such a function, then its lateral derivatives
at a point x ∈ R are defined (and denoted) by

f (x + t) − f (x)
f± (x) := lim± .
t→0 t
Recall that f is said to be differentiable at x ∈ R if the lateral derivatives exist and are equal.
Also, we say that f is left continuous if limx→x0− f (x) = f (x0 ) for any x0 ∈ R. Similarly, f is said
to be right continuous if limx→x0+ f (x) = f (x0 ) for every x0 ∈ R.

Theorem 1.2.1. Let f : R → (−∞, +∞] be a convex function. Then, in int(dom( f )), we have the
following:

(i) the lateral derivatives f+ and f− exist and are non-decreasing.

(ii) the inequality f− (x) ≤ f+ (x) holds for every x ∈ int(dom( f )), and equality holds except for at
most countably many points.

(iii) the map f− and f+ are left continuous and right continuous, respectively.

For an illustration of (i) and (ii) see Figure 1.2.1 below.

Fig. 1.2.1 A point where f− (x0 ) < f+ (x0 ).

Proof. Let x ∈ int(dom( f )), and let 0 < t < s. Assume also that s is small enough such that
(x − s, x + s) ∈ int(dom( f )). We write
s − t t  s−t t
f (x − t) = f x+ (x − s) ≤ f (x) + f (x − s),
s s s s
and reorganizing the terms we get
12 1 Convex functions

f (x − s) − f (x) f (x − t) − f (x)
≤ . (1.2.1)
−s −t
Analogously, if 0 < λ < μ are such that (x − μ, x + μ) ∈ int(dom( f )), then
μ−λ λ  μ−λ λ
f (x + λ) = f x+ (x + μ) ≤ f (x) + f (x + μ),
μ μ μ μ
yielding the inequality

f (x + λ) − (x) f (x + μ) − f (x)
≤ . (1.2.2)
λ μ
Roughly speaking, inequalities (1.2.1) and (1.2.2) say that the slopes of secant lines of the graph of a
convex function are monotone. These inequalities, as well as inequality (1.2.5) below, are illustrated
in Figure 1.2.2.

Fig. 1.2.2 We have (1.2.2), (1.2.1), and (1.2.5) for x = a, x = c, and x = b, respectively.

Observe that inequality (1.2.1) gives that the map

f (x − t) − f (x)
(0, ε) t → (1.2.3)
−t
is non-increasing (for small ε > 0). Since

f (x + λ) − f (x) f (x − t) − f (x)
f− (x) = lim− = lim+ ,
λ→0 λ t→0 −t
all we have to do for showing that the limit above exists is to verify that the function given in (1.2.3)
is bounded from below. On the other hand, inequality (1.2.2) gives that the map

f (x + λ) − f (x)
(0, ε) λ → (1.2.4)
λ
is non-decreasing. Thus, in order to prove that the limit

f (x + λ) − f (x)
f+ (x) = lim+
λ→0 λ
exists it suffices to show that (1.2.4) is bounded from above. We can obtain the desired bounds for (1.2.3)
and (1.2.4) with a single argument. Let λ, t > 0 be arbitrary numbers such that x−t, x+λ ∈ int(dom( f )).
1.2 Regularity 13

Then
λ t  λ t 
f (x) + f (x) = f (x) = f (x − t) + (x + λ) ≤
t+λ t+λ t+λ t+λ
λ t
≤ f (x − t) + f (x + λ).
t+λ t+λ
Rearranging the terms, we get the inequality

f (x − t) − f (x) f (x + λ) − f (x)
≤ . (1.2.5)
−t λ
It follows that f− (x) and f+ (x) exist. Actually, we can also use this inequality to prove that f− and
f+ are non-decreasing on int(dom( f )). Indeed, it follows immediately that f− (x) ≤ f+ (x), and if
x, z ∈ int(dom( f )) are such that x < z, then we have

f (z) − f (x)
f− (x) ≤ f+ (x) ≤ ≤ f− (z) ≤ f+ (z). (1.2.6)
z−x
To check the inequalities in the middle, simply notice that z = x + λ for λ = z − x > 0, and x = z − t
for t = λ (which is positive). Hence we get the desired from the fact that the maps (1.2.3) and (1.2.4)
are non-increasing and non-decreasing, respectively.
From (1.2.6), it comes straightforwardly that if f− (or f+ ) is continuous at x, then f− (x) = f+ (x).
Since f− ( f+ ) is non-decreasing, we get that f− has at most countably many discontinuity points. Hence,
the equality f− (x) = f+ (x) holds for at most countably many points of int(dom( f )). This settles (i)
and (ii).
It remains to show (iii). We prove only the case for f+ , since the other case is analogous. Let
x ∈ int(dom( f )), and first observe that the restriction of f+ values which are greater than x is
non-decreasing and bounded from below by f+ (x). Hence the limit limz→x + f+ (z) exists. Using the
continuity of f , for x < y we get that

f (y) − f (x) f (y) − f (z)

= lim+ ≥ lim+ f+ (z), (1.2.7)
y−x z→x y−z z→x

where the last inequality holds since for each x < z < y we have that

f (y) − f (z)
f+ (z) ≤ .
y−z
Observe that we are using the second inequality in (1.2.6) with other letters for the variables. Letting
y → x + in (1.2.7), we get

f (y) − f (x)
f+ (x) = lim+ ≥ lim+ f+ (z).
y→x y−x z→x

Finally, since f+ is non-decreasing, we have that f+ (x) ≤ limz→x + f+ (z). Both inequalities yield

f+ (x) = lim+ f+ (z),

z→x

and hence f+ is right continuous at any point x ∈ int(dom( f )).



Corollary 1.2.1. If f : R → (−∞, +∞] is a convex function differentiable on int(dom( f )), then f is
continuously differentiable (that is, of class C 1 ) on int(dom( f )).
14 1 Convex functions

Proof. This comes immediately from the previous theorem: if f is differentiable, then f− = f+ := f
on int(dom( f )). Since f− is left continuous and f+ is right continuous, it follows that f is both left
and right continuous, and hence continuous.


Proposition 1.2.1. Let J ⊆ R be an interval, and let f : J → R be a differentiable function. Then f is

convex if and only if f is non-decreasing.

Proof. If f is convex, then f is non-decreasing by Theorem 1.2.1 (i) (notice that f = f+ = f−

because f is differentiable).
Now assume that f is non-decreasing. Let x, y ∈ J be distinct points, and let λ ∈ (0, 1). By the
mean value theorem, there exists a number μ1 ∈ [x, (1 − λ)x + λy] such that

f ((1 − λ)x + λy) − f (x)

f (μ1 ) = ,
λ(y − x)

and there exists a number μ2 ∈ [(1 − λ)x + λy, y] for which

f (y) − f ((1 − λ)x + λy)

f (μ2 ) = .
λ(y − x)

Since f is non-decreasing, we have that f (μ2 ) ≥ f (μ1 ). This inequality gives immediately that
f ((1 − λ)x + λy) ≥ (1 − λ) f (x) + λ f (y), and hence f is convex.


Corollary 1.2.2. If J ⊆ R is an interval and f : J → R is twice differentiable, then f is convex if and

only if f ≥ 0.

We say that a line r supports a subset A ⊆ R2 if r ∩ A  ∅ and A is contained in one of the two closed
half-spaces determined by r (this notion is extended to higher dimensional spaces and hyperplanes in
the next section). In this case, if a ∈ r ∩ A, then we also say that r supports A at a. The next result is a
useful geometric interpretation of the lateral derivatives of a convex function.

Proposition 1.2.2. Let f : R → (−∞, +∞] be a convex function, and let x0 ∈ int(dom( f )). Then the
line

r( f , x0, m) := {(x, y) ∈ R2 : y − f (x0 ) = m(x − x0 )},

which is the line through the point (x0, f (x0 )) whose inclination is m, supports epi( f ) at (x0, f (x0 )) if
and only if f− (x0 ) ≤ m ≤ f+ (x0 ).
As a consequence, we have that for any x0 ∈ int(dom( f )) the epigraph of f is supported by a
non-vertical line at the point (x0, f (x0 )). If x0 ∈ dom( f ), then epi( f ) is still supported by a line at
(x0, f (x0 )), but we cannot guarantee that this line is non-vertical.

Remark 1.2.1. The lines drawn in Figure 1.2.1 represent the cases m = f− (x0 ) and m = f+ (x0 ), and for
f− (x0 ) < m < f+ (x0 ) the corresponding line lies “between” them. Hence it is geometrically clear that
these lines support the epigraph of f .

Proof. First, assume that f− (x0 ) ≤ m ≤ f+ (x0 ). Let x ∈ int(dom( f )) be such that x < x0 . Then, by
inequality (1.2.6), we have that

f (x0 ) − f (x)
≤ f− (x0 ) ≤ m.
x0 − x
Thus, we have that f (x) ≥ f (x0 ) + m(x − x0 ). On the other hand, if x > x0 , then we have
1.2 Regularity 15

f (x) − f (x0 )
≥ f+ (x0 ) ≥ m,
x − x0
from which we also get the inequality f (x) ≥ f (x0 ) + m(x − x0 ). If (x, y) ∈ epi( f ), then

y ≥ f (x) ≥ f (x0 ) + m(x − x0 ),

and hence all points of epi( f ) lie in one of the half-spaces determined by r( f , x0, m), namely in the
“upper” half-space.
For the converse, we suppose that m < f− (x0 ), because the case where m > f+ (x0 ) is analogous. Of
course, there exists x < x0 such that
f (x0 ) − f (x)
m< ≤ f− (x0 ).
x0 − x
It follows from the first inequality above that f (x) < f (x0 ) + m(x − x0 ). Now, if z > x0 , then

f (z) − f (x0 )
m < f− (x0 ) ≤ f+ (x0 ) ≤ ,
z − x0
where we again used (1.2.6). Thus, f (z) > f (x0 ) + m(z − x0 ). Consequently, the points (x, f (x)) and
(z, f (z)) are points of the epigraph of f which lie in distinct half-planes determined by r( f , x0, m).
Then this line does not support epi( f ).
For any given x0 ∈ int(dom( f )), we have that f− (x0 ) ≤ f+ (x0 ). It follows that the epigraph of f is
supported by some line in (x0, f (x0 )).
It remains to deal with the case where x0 is a non-interior point of dom( f ). Since dom( f ) is a
convex set of R, it clearly holds that dom( f ) is an interval. The only possible non-interior points are
its endpoints. So, we first write dom( f ) = [α, β), with β possibly being +∞. The epigraph of f is
obviously supported by the line x = α. The case where β is the endpoint is analogous.
A remaining comment is due. In some cases, we may guarantee that f is supported by a non-vertical
line in boundary points of dom( f ). Assuming again that dom( f ) = [α, β), observe that any sequence
(xm ) ∈ (α, β) converging to α is a sequence of points of int(dom( f )), and the lateral derivatives f− (xm )
form a non-increasing sequence. As a consequence, f− (xm ) has a limit (c, say) as m → +∞, which
may possibly be −∞. If c > −∞ and f is lower semi-continuous at α, then for any (x, t) ∈ epi( f ) we
have

t ≥ f (xm ) + f− (xm )(x − xm ),

and hence it is clear that

t ≥ lim inf f (xm ) + c(x − α) ≥ f (α) + c(x − α)

m→+∞

for any (x, t) ∈ epi( f ). Thus, the line t − f (α) = c(x − α) supports the epigraph of f at the point
(α, f (α)). On the other hand, if c = −∞, then one may easily check that epi( f ) is supported by the line
x = α at (α, f (α)).


As an immediate consequence, we have that the epigraph of a convex function f : R → (−∞, +∞]
is supported by some line at each point (x, f (x)) such that x ∈ dom( f )). This leads to a functional
version of Jensen’s inequality that we state and prove now. Recall that a probability space is a measure
space (X, Σ, μ) such that μ(X) = 1. In this case, we also say that μ is a probability measure. To fix the
notation, from now on we denote by im( f ) the image of a function f .
16 1 Convex functions

Theorem 1.2.2 (Jensen’s inequality). Let D ⊆ R be an open interval, and let ϕ : D → R be a convex
function. If X is a topological space with a Borel probability measure and f : X → D is an integrable
function, then
∫  ∫
ϕ f dμ ≤ ϕ ◦ f dμ. (1.2.8)
X X

Equality holds if and only if ϕ is affine over f (X \ Y ) for some Y ⊆ X such that μ(Y ) = 0.

Remark 1.2.2. Notice that equality holds if f is constant μ-almost everywhere. Indeed, equality con-
dition holds trivially in this case.

Proof. Let
∫
x0 := f dμ
X

be the mean of f . We claim that x0 ∈ D. Indeed, we clearly have

∫
   
c0 := inf f (x) : x ∈ X ≤ f dμ ≤ sup f (x) : x ∈ X =: c1, (1.2.9)
X

and c0, c1 ∈ cl(D). If c0, c1 ∈ D, then we are done. If c0  D and equality holds in the left-hand
inequality, then
∫
f − c0 dμ = 0,
X

and once f (x) − c0 ≥ 0 for every x ∈ X we must have f (x) = c0 for almost every x ∈ X. This is in
contradiction to the fact that c0  D. Hence if c0  D the left-hand inequality in (1.2.9) is strict. By an
analogous argument, we get that if c1  D, then the right-hand side inequality in (1.2.9) is also strict.
It follows that x0 ∈ D, even if c0 or c1 are not points of D.
Because x0 ∈ D = int(dom(ϕ)), we have that ϕ is supported by a line x → mx + b at (x0, ϕ(x0 )).
This means that

(i) mx0 + b = ϕ(x0 ) and

(ii) ϕ(x) ≥ mx + b for every x ∈ D.

Thus, ∫ ∫ ∫
ϕ ◦ f (x) dμ(x) ≥ m f (x) + b dμ(x) = m f (x) dμ(x) + b =
X X X
∫  (1.2.10)
= mx0 + b = ϕ(x0 ) = ϕ f dμ(x) ,
X
and this concludes the proof of the inequality.
Now notice that equality in (1.2.8) obviously holds if ϕ is an affine function or if f is constant.
Conversely, if equality holds, then we have from the first inequality in (1.2.10) that

ϕ ◦ f (x) = m f (x) + b

for every x ∈ X \ Y for some Y ⊆ X such that μ(Y ) = 0. This clearly happens if ϕ, over f (X \ Y ), is
the affine function ϕ(z) = mz + b.

1.2 Regularity 17

Remark 1.2.3. We can extend Jensen’s inequality (with the equality case characterization) for the case
where D ⊆ R is any interval and ϕ : D → (−∞, +∞] is a convex function which is not necessarily
real-valued. In this case, we only need to demand that
∫
x0 = f dμ ∈ int(dom(ϕ)),
X

and repeat verbatim the proof above (of course, without the part where we prove that x0 ∈ int(dom(ϕ)),
since now we assume that as a hypothesis).
It is also worth observing that if x0 is not a point of the effective domain of ϕ, then the left-hand
side of (1.2.8) is +∞, and hence there is no sense in bounding it from above. However, in this case it is
possible to have +∞ on both sides of (1.2.8) with a non-constant function f and a non-affine convex
function ϕ (see Exercise 1.11).

We go now to the n-dimensional situation. If f : Rn → (−∞, +∞] is a convex function and

x ∈ int(dom( f )), then for any non-zero vector u ∈ Rn the (positive) one-sided directional derivative

f (x + λu) − f (x)
f+ (x, u) = lim+
λ→0 λ
exists. Indeed, for small ε > 0 the function g : (−ε, ε) → R given as g(λ) = f (x + λu) is a one-variable
convex function, and
f (x + λu) − f (x) g(λ) − g(0)
f+ (x, u) = lim+ = lim+ = g+ (0).
λ→0 λ λ→0 λ
One can analogously define the negative one-sided derivative

f (x + λu) − f (x)
f− (x, u) := lim− .
λ→0 λ
However, it is more convenient to work only with f+ , since f− can be re-obtained as

f (x + λu) − f (x) f (x − tu) − f (x)

f− (x, u) = lim− = lim+ =
λ→0 λ t→0 −t
f (x + t(−u)) − f (x)
= − lim+ = − f+ (x, −u).
t→0 t
We extend the definition of the one-sided directional derivative for the zero vector by setting
f+ (x, 0) = 0. Doing this, we define the one-sided derivative map of f at x to be the map f+ (x, ·) : Rn →
R. In what follows, recall that a function g : Rn → R is positively homogeneous if g(λx) = λg(x) for
every x ∈ Rn and any λ ≥ 0.

Proposition 1.2.3. Let f : Rn → (−∞, +∞] be a convex function. For each x ∈ int(dom( f )) the one-
sided derivative of f at x satisfies the properties

i. (subadditivity) f+ (x, u + v) ≤ f+ (x, u) + f+ (x, v) for any u, v ∈ Rn .

ii. (positive homogeneity) f+ (x, λu) = λ f+ (x, u) for any λ ≥ 0 and u ∈ Rn .

Proof. For the subadditivity we simply notice that

18 1 Convex functions
 
f (x + λ(u + v)) − f (x) f 12 (x + 2λu) + 12 (x + 2λv) − f (x)
f+ (x, u + v) = lim+ ≤ lim+ ≤
λ→0 λ λ→0 λ
1
f (x + 2λu) + 12 f (x + 2λv) − 12 f (x) − 12 f (x)
≤ lim+ 2 =
λ→0 λ
f (x + 2λu) − f (x) f (x + 2λv) − f (x)
= lim+ + lim+ = f+ (x, u) + f+ (x, v),
λ→0 2λ λ→0 2λ
where in the first inequality we have used the convexity of f , and the last equality is justified by a
simple change of variables. Now we prove that f+ (x, ·) is positively homogeneous, using again a change
of variables. Let u ∈ Rn and λ > 0. Then
f (x + tλu) − f (x) f (x + su) − f (x)
f+ (x, λu) = lim+ = λ lim+ = λ f+ (x, u),
t→0 t s→0 s
where s = λt. The case λ = 0 is immediate since, by definition, f+ (x, 0) = 0.


A real-valued function (that is, a function which does not assume the value +∞) which is sub-
additive and positively homogeneous is called a sublinear function. That is, a sublinear function is a
function g : Rn → R such that

(i) g(x + y) ≤ g(x) + g(y) and

(ii) g(λx) = λg(x)

for every x, y ∈ Rn and any λ ≥ 0 (observe that, in particular, every linear function is sublinear). It
is very important to have in mind that, from this definition, every time that we refer to a function as
“sublinear”, we have that this function is real-valued.
It is immediate that every sublinear function is convex, and that any linear function is, in particular,
sublinear. Consequently, any sublinear function is a convex function whose effective domain is Rn .
Once any convex function is continuous in the interior of its effective domain, we get that every
sublinear function is continuous.
Before we continue to study the one-sided derivative of convex functions, we investigate some
properties of sublinear functions which will be important later. If g : Rn → R is a sublinear function,
then any vector u ∈ Rn for which the equality g(u) = −g(−u) holds is called a linearity direction of g.
The set of linearity directions is denoted by lin(g). What justifies this name is that if u ∈ lin(g), then g
is clearly homogeneous on span{u}, meaning that g(λu) = λg(u) for every λ ∈ R (that is, this equality
occurs even if λ is a negative number).

Proposition 1.2.4. Let g : Rn → R be a sublinear function. Then, for any x ∈ Rn , the function
g+ (x, ·) : Rn → R is sublinear and the following properties hold:

(a) g+ (x, λx) = λg(x) for every x ∈ Rn and any λ ∈ R,

(b) g+ (x, u) ≤ g(u) for every x, u ∈ Rn ,

(c) lin(g) is a vector subspace of Rn ,

(d) lin(g+ (x, ·)) ⊇ lin(g) + span{x} for every x ∈ Rn ,

(e) lin(g) = Rn if and only if g is linear.

1.2 Regularity 19

Proof. The fact that g+ (x, ·) is sublinear comes immediately from Proposition 1.2.3, since every
sublinear function is, in particular, convex. For (a), let x ∈ Rn and λ ∈ R. We simply notice that for
any λ ∈ R, if t > 0 is small enough, then 1 + tλ > 0. Hence

g(x + tλx) − g(x) (1 + tλ)g(x) − g(x) tλg(x)

g+ (x, λx) = lim+ = lim+ = lim+ = λg(x).
t→0 t t→0 t t→0 t
The proof of (b) relies only on the following calculation, where we use both the subadditivity and the
positive homogeneity of g:

g(x + tu) − g(x) g(x) + g(tu) − g(x) tg(u)

g+ (x, u) = lim+ ≤ lim+ = lim+ = g(u).
t→0 t t→0 t t→0 t
To prove (c) it is sufficient to show that u + v ∈ lin(g) whenever u, v ∈ lin(g), since we already know
that g is homogeneous in any direction of lin(g). First observe that, if u, v ∈ lin(g), then

g(u + v) ≤ g(u) + g(v) = −g(−u) − g(−v) ≤ −g(−u − v).

On the other hand,

0 = g(u + v + (−u − v)) ≤ g(u + v) + g(−u − v),

from where it follows that −g(−u−v) ≤ g(u+v). These two inequalities imply that g(u+v) = −g(−u−v),
and hence u + v ∈ lin(g).
Now we prove (d). It is already clear from (a) that span{x} ⊆ lin(g+ (x, ·)). Thus we only have to
prove that any linearity direction of g is also a linearity direction of g+ (x, ·). For any u ∈ lin(g) and
every t ∈ R we have

2g(x) = g(2x) = g(x + tu + x − tu) ≤ g(x + tu) + g(x − tu) ≤

≤ 2g(x) + t(g(u) + g(−u)) = 2g(x),

from which it follows that 2g(x) = g(x + tu) + g(x − tu). As a consequence, we get that

g(x + tu) − g(x) g(x − tu) − g(x)

+ = 0,
t t
whenever t  0. Letting t → 0+ in this equality, we obtain

g(x + tu) − g(x) g(x + t(−u)) − g(x)

g+ (x, u) + g+ (x, −u) = lim+ + lim+ =
t→0 t t→0 t
 
g(x + tu) − g(x) g(x − tu) − g(x)
= lim+ + = 0,
t→0 t t

and hence g+ (x, u) = −g+ (x, −u).

It remains to prove (e). Notice that one of the implications is obvious: if g is linear, then all directions
are linearity directions. Hence we assume that lin(g) = Rn . In this case, it is immediate by the definition
that g is homogeneous, and thus we have to prove that it is additive. For that sake, we let u, v ∈ Rn and
write

g(u) + g(v) ≥ g(u + v) = −g(−u − v) ≥ −g(−u) − g(−v) = g(u) + g(v).

This concludes the proof.


20 1 Convex functions

If f : Rn → (−∞, ∞] is a convex function, x ∈ int(dom( f )), u ∈ Rn , and f+ (x, u) = f− (x, u), then
we define

f (x, u) := f+ (x, u),

and we call this number the two-sided directional derivative of f at x in the direction of u. In this
case, we also say that the two-sided directional derivative of f at x in the direction of u exists, or that
f has a two-sided derivative at x in the direction of u. The reader has to be aware of the fact that some
authors restrict these definitions for unit vectors, but we are not doing this.
Assume that f has a two-sided derivative at x in the direction of u. Then we have that f+ (x, u) =
− f+ (x, −u), and consequently u is a linearity direction of the one-sided derivative map f+ (x, ·). The
linearity directions of the one-sided derivative map f+ (x, ·) are precisely the directions for which the
two-sided directional derivative of f at x exists.
Recall that a function f : Rn → R is said to be differentiable (or differentiable in the strong sense)
at a point x ∈ Rn if there exists a linear map dfx : Rn → R with the property that

f (x + v) − f (x) − dfx (v)

lim = 0,
v→0 |v|
where the limit is taken in Rn , of course. In this case, the map dfx is called the differential of f at
x. It is immediate from the definition that if f is differentiable at x, then it has two-sided directional
derivative at x for any u ∈ Rn , and f (x, u) = dfx (u).
A convex function is differentiable at “almost all” points of Rn . To make this sentence formal, we
need some background on basic measure theory (see Appendix A).

Theorem 1.2.3. A convex function f : Rn → R is differentiable almost everywhere in Rn (with respect

to the Lebesgue measure).

Proof. Let ρ > 0. Since the closed ball ρB n is compact, we get that f is Lipschitz in this ball. Hence
f is Lipschitz in the open ball B(0Rn , ρ). From Rademacher’s classical theorem (see [2249]), we get
that f is differentiable at almost every point of B(0Rn , ρ). Once Rn can be written as a countable union
of closed balls, it follows that the set of points where f is not differentiable has zero measure.


Let {e1, . . . , en } be the canonical basis of Rn . For a given i ∈ {1, . . . , n}, if the two-sided derivative
of f in the direction of ei exists, then

∂i f (x) := f (x, ei )

is called the (first order) i-th partial derivative of f at x. In this case, we also say that f has the i-th
partial derivative.

Proposition 1.2.5. Let f : Rn → (−∞, +∞] be a convex function, and let x ∈ int(dom( f )). Then the
following statements are equivalent:

(i) f+ (x, u) = f− (x, u) for every u ∈ Rn (in other words, f has all two-sided derivatives at x).

(ii) f has all partial derivatives at x.

(iii) f is differentiable at x.

In this case, we have that f (x, u) = dfx (u) for any u ∈ Rn .

1.2 Regularity 21

Remark 1.2.4. Observe that, from Item (e) of Proposition 1.2.4, we have that (i) is also equivalent to
f+ (x, ·) being a linear map.

Proof. First notice that (i)⇒(ii) is immediate from the definition (each partial derivative is a two-sided
derivative). The implication (iii)⇒(i) is also not too difficult. From the positive homogeneity of f+ ,
we can assume that u is unit. If f is differentiable at x, then putting v = λu yields

f (x + λu) − f (x) − dfx (λu)

0 = lim+ ,
λ→0 λ
implying immediately that dfx (u) = f+ (x, u). On the other hand, the equality

f (x + λu) − f (x) − dfx (λu)

0 = lim−
λ→0 −λ
gives that

f (x + λu) − f (x) −λdfx (u)

f− (x, u) = lim− = lim− = dfx (u).
λ→0 λ λ→0 −λ
It follows that f+ (x, u) = dfx (u) = f− (x, u), and hence f has all two-sided derivatives at x. Observe
carefully that we did not use the convexity hypothesis to prove these implications, but that we will do
so to prove the remaining implication (ii)⇒(iii).
If f has partial derivatives of first order at a given point x ∈ Rn , then we can define the gradient of
f at x to be the vector

n
∇ f (x) := ∂j f (x) e j .
j=1

Define the linear transformation T : Rn → R by

T v := ∇ f (x), v,

for every v ∈ Rn . We claim that

f (x + v) − f (x) − T v
lim = 0.
v→0 |v|

Indeed, for simplicity we write g(v) := f (x + v) − f (x) − T v. Of course, g is finite in an open ball
BRn (0, ρ), where ρ > 0 is such that BRn (x, ρ) ⊆ int(dom( f )). Also, g is convex on this open ball,
 convex function v → f (x + v) − f (x) with the affine function v → T v.
because it is the sum of the
Write in coordinates v = nj=1 α j e j , and let J = { j : α j  0}. From discrete Jensen’s inequality, and
recalling that g(0) = 0, we get

 1   1 1
n n n

g(v) = g  α j e j  = g  nα j e j  ≤ g(nα j e j ) = g(nα j e j ) =
n j=1 n j=1 n j ∈J
 j=1 
 g(nα j e j )
= αj .
j ∈J
nα j
n
Let w = j=1 β j e j , where β j = 0 if j  J, and
Exploring the Variety of Random
Documents with Different Content
voimakkaammaksi, niin voimakkaaksi, että se lopulta ahdisti, ja
silloin minussa heräsi jonkinmoinen viha kaikkea kohtaan. Tuntui
aivan siltä kuin olisin ollut suuri syyttäjä, joka vaatii yhteiskunnan
tilille. Millään yhteiskuntajärjestelmällä ei ole oikeutta kieltää
elämisen mahdollisuuksia terveeltä olennolta, vaikka se on tullut
mailmaan yhteiskunnan määräämien muotojen ulkopuolellakin. Jos
lapsen vanhemmat eivät ole noudattaneet muutamia yhteiskunnan
asettamia määräyksiä, niin onhan lapsi niiden ulkopuolella, ja kaikki
on tehtävä, jotta terveellä olennolla on terveet elämänoikeudet.
Samassa huomasin että tämä teoriia oli aivan yleinen ja oikeaksi
tunnustettu. Väittihän maailma itseään vapaamieliseksi. Mutta miksi
vapaamielisyys oli ainoastaan sanoissa eikä vastaavissa teoissa.
Kuinkahan moni, joka itseään vapaamieliseksi kutsuu, antaisi turvan
ja yhteiskunnalliset oikeudet aviottomalle lapselle ja hänen äidilleen.
Teoriioja, teoriioja, niitä on mailmalla niin runsaasti! Ne supistuvat
vaan sanoihin, mutta eivät muutu teoiksi.

Ja silloin äkkiä lämpimänä laineena tuli sieluuni ajatus siitä pyhän

korkeasta kauneudesta, joka on äiteys. Eivät suotta maailman
suuremmat henkilöt ole asettaneet äitiä ja äidin tunnetta kaiken
muun yläpuolelle. Tuo pyhä hedelmällisyys, joka on kaiken elävän
luonnon keskuksena, on kauneimman muotonsa löytänyt siinä, kun
terve äiti synnyttää terveen lapsen. Ja minusta tuntui, ikäänkuin koko
yhteiskunta, kaikki sen lait ja järjestelmät olisivat uudestaan
muodostettavat sen sanan ympärille, joka on äiti.

Tämän lapsen äiti oli hätääntynyt ja ensi hetkessä sokeana luullut

kaiken peittävänsä sen kautta, että hän kätki lapsen. Sitten oli
syyllisyyden tunto tullut ja hän oli ilmoittanut tekonsa poliisille. Oliko
hän oikeastaan syyllisempi kuin moni muu tyttöäiti, joka tahallisen
huolimattoman hoidon kautta antaa lapsensa kuolla? Todistihan
tilasto, että tällaiset kuolemantapaukset olivat hyvin yleisiä.

Mutta eihän minun sopinut ruveta filosofeeraamaan, minun oli

toimitettava tehtäväni, tarkastettava, millä tavoin lapsi oli surmattu.
Äkkiä olin päässyt tuosta hiljaisuuden lumouksesta vapaaksi. Minä
olin vain kylmä virkamies, joka täytin yhteiskunnan minulle asettamat
tehtävät. Samassa, kun tartuin leikkausveitseen, näytti tämä liikkeeni
vapauttavan molemmat toisetkin huoneessa olijat siitä jäykkyydestä,
minkä alaisina he olivat olleet, samoin kuin minäkin. Poliisi pyyhkäsi
kädellään otsaansa ja huokasi syvään. Vahtimestari riensi
leikkauspöydän luo.

— Kyllä minä teen avauksen, ja tohtori saa sitten vain tarkastaa,

sanoi hän.

Hän oli sellaiseen tottunut, sillä monasti olin antanut hänen tehdä
ruumiinavauksessa kaiken karkean työn. Olihan hän vuosien
kuluessa siinä saavuttanut hyvinkin suuren taituruuden.

En tiedä, miksi tällä kertaa kielsin häntä ryhtymästä siihen, miksi

itse tahdoin suorittaa koko työn?

— Vahtimestari saa istua rauhassa, kyllä minä tämän teen, sanoin

hänelle.

— Mitä tohtori suotta näkee sellaista vaivaa, vastasi hän. Voinhan

minä varsin hyvin auttaa.

Totuttu tapa, tuo ihmisten suuri ohjaaja ja tekojen määrääjä, oli jo

saada minussa vallan. Ojensin jo häntä kohden veitsen, kun satuin
vilkaisemaan poliisin tuomaan paperiin, joka oli viereisellä pulpetilla.
 Silmiini osui naisen nimi! Ajatusten monien sokkeloiden läpi pujahti
elävänä mieleeni olento, joka kantoi tuota nimeä. Olin nähnyt hänet,
olin puhutellut häntä, siitä olin varma, mutta missä ja milloin, sitä en
ensi hetkessä muistanut.

Pinnistin ajatuksiani, ja samassa välähti mieleeni ne olosuhteet,

joissa hänet olin nähnyt, muistin, mistä olin hänen kanssaan
puhunut, enkä silloin enää tahtonut vahtimestarin olevan läsnä
ruumiinavauksessa.

Mutta olihan minun pakko noudattaa lain asettamia määräyksiä,

poliisin ja hänen täytyy olla läsnä. Mitä oikeastaan aijoin, mitä
tahdoin tehdä, sitä en tiedä? Voimakas sisäinen vaatimus vain
pakoitti minut salamaan kaiken, mitä leikkaus voi tuoda ilmi sekä
poliisilta että vahtimestarilta. Poliisin pettäminen ei ollut vaikeata,
hänhän pysytteli koko ajan paikallaan oven luona ja sitäpaitsi eihän
hän leikkauksen tuloksista mitään ymmärtänyt. Virkavallan
palvelijana oli hänellä myöskin ehdoton usko minun auktoriteettiini,
mutta vahtimestari saattoi ymmärtää leikkauksen tulokset. Pitääkseni
hänet loitolla leikkauspöydästä, ei minulla ollut muuta keinoa kuin
antaa hänelle mahdollisimman paljon työtä ja puuhaa, jotta hän ei
voisi kaikkea tarkkaan seurata. Se ei ollut aluksi vaikeata, pyysin
milloin mitäkin, ja hän tuli siten olleeksi liikkeessä koko sen ajan,
jonka tarvitsin tehdäkseni tavanmukaiset viillokset. Mutta kohta
myöskin loppuivat minulta nämät keinot ja kun minun tuli tarkastaa
lapsen keuhkoja, nähdäkseni oliko se syntyessään hengittänyt,
seisoi hän leikkauspöydän vieressä tarkkaan katsellen työtäni.

Minun täytyi saada hänet pitemmäksi aikaa pois huoneesta.

— Onko teillä arkku valmiina?

 — On. Se on ruumishuoneessa.

— Menkää noutamaan se.

— Minä aijon vasta aamulla panna lapsen arkkuun.

— Minkätähden? Tiedättehän, etten salli ruumiiden jäävän

leikkauspöydälle avauksen jälkeen.

Vahtimestari katsoi hiukan kummastuneena minuun, sillä enhän

koskaan ennen ollut pitänyt huolta siitä, milloin hän ruumiin vie pois,
mutta, kun hän monivuotisen palveluksensa aikana oli tottunut
arvelematta noudattamaan esimiehensä määräyksiä, lausui hän
muutamia sanoja anteeksipyynnöksi ja kiiruhti pois huoneesta.

Nopeasti avasin keuhkot. Lapsi oli syntyessään elänyt, sillä se oli

hengittänyt. Minä huomasin sen keuhkojen väristä. Otin ne irti, panin
ne lähellä olevaan vesiastiaan. Ne kelluivat pinnalla. Äiti oli
epäilemättä tukehuttanut lapsensa, pitäen kättään sen suulla, koska
mitään ulkonaisia väkivallan merkkejä ei näkynyt. Otin keuhkot,
painoin ne mäsäksi ja pistin ne rintaonteloon takaisin.

Peitin ruumiin ja istuin pulpetin ääreen, tartuin poliisin tuomaan

lappuun. Lääkärivalani nimessä oli minun siihen kirjoitettava, mihin
tulokseen olin tullut ruumiinavauksessa. Minä olin yhteiskunnan
asettama suuri syyttäjä.

Kynä oli kädessäni ja kuitenkin viivyttelin. Elävänä johtui mieleeni

se hetki, jolloin tuon kuolleen lapsen äidin ensi kertaa näin.

Monta kuukautta sitten tuli eräänä iltahämäränä luokseni

palvelustyttö. Varsinainen vastaanottoaika oli jo loppunut, olin
sammuttanut valkean ja lojuin leposohvalla, antaen ajatusteni
vapaasti kulkea milloin minnekin. Silloin kuulin ovikelloani arasti
soitettavan, odotin, kukaan ei mennyt avaamaan. Kelloa soitettiin
toinen kerta. Samassa muistinkin, että olin antanut palvelijattarelleni
lomaa, hän oli pyytänyt päästä sukulaistensa luo. Menin siis itse
avaamaan. Oven takana oli nuori, ujo tyttö.

— Minuako etsitte? kysyin.

— Niin.

— Mitä teillä on asiaa?

— Saisinko puhua tohtorin kanssa?

— Minun varsinainen vastaanottoaikani on jo loppunut, tulkaa

huomenna.

— Enkö minä kumminkin saisi?

Aijoin jo käskeä häntä poistumaan, sillä itsekkäänä ajattelin, että

minulla on oikeus käyttää muutamia tuntia lepoani varten, mutta
hänen äänessään oli niin rukoileva sointu, etten hennonnut kieltää.

— Tulkaa sisään, sanoin.

Astuin edellä vastaanottohuoneeseen. Hän seurasi minua ja jäi

arkana oven suuhun seisomaan.

— Mitä on asiaa? kysyin.

Hän ei vastannut mitään. Kadulta tulevassa heikossa

valaistuksessa näin hänen painavan päänsä kumaraan. Huivinsa oli
valunut kaulaan ja hän pureskeli sen kulmaa.
 — Oletteko sairas? kysyin.

Nyyhkytys nousi hänen rinnastaan. Kuulin sen aivan selvästi

huoneen hiljaisuudessa.

Aavistin, että saisin olla enemmän sielun kuin ruumiin lääkärinä.

Suuri hellyyden tunne valtasi minut tuota ehkä elämän taistelussa
hätääntynyttä olentoa kohtaan.

— Istukaa, sanoin.

Menin hänen luokseen, kosketin hiljaa hänen käsivarteensa ja

vein hänet tuoliin istumaan.

Aijoin sytyttää lampun, olin jo tarttunut sähkölampun nappulaan,

kun jätinkin sen tekemättä. Jos hän lääkärin luo tulee aivan kuin
rippi-isän luo, niin voi hän tunnustuksensa tehdä helpommin
hämärässä. Istahdin tuolilleni ja kysyin hiljaa ja ystävällisesti:

— Mikä on nimenne?

Hän mainitsi sen. Hetkisen kuluttua kysyin taas:

— Mikä teitä vaivaa?

Hän oli vähän aikaa vaiti, mutta sitten hän sen kertoi. Kertoi
katkonaisin sanoin tuon nuoren tytön tavallisen tarinan miehestä,
joka oli uskotellut rakkautta ja sitten jättänyt tytön oman onnensa
nojaan.

Lopetettuaan kertomuksensa sanoi hän:

— Jos tohtori auttaisi minua.

 — Millä tavoin?

— Ettei lapsi synny mailmaan.

— Ettekö tiedä, ettei minulla siihen ole oikeutta?

— Kyllähän minä tiedän.

— Että laki sen kieltää, ankarasti kieltää?

— Kyllä minä sen tiedän. Mutta ehkä tohtori kuitenkin auttaa, kun
minä asian selitän.

— Se on aivan suotta, minä en sitä voi tehdä! En saa, enkä tahdo!

— Tohtori kuuntelee kuitenkin, sanoi hän rukoilevalla äänellä, ja

minä huomasin kuinka hän oli hätääntynyt ja pelkäsi, että minä
käskisin häntä menemään, ennenkuin hän oli saanut sanoa
sanottavansa.

Kun minä tiesin että jokaisella meistä ihmisistä on kerran

elämässään getsemaane-hetkensä, että me jaksamme voittaa
suurimmatkin surut, jos vain sinä hetkenä joku jaksaa meitä
kuunnella, niin annoin hänen puhua.

Hämärässä näin, miten hänen lapsensilmänsä katsoivat suoraan

minuun.

— Minä olen tätä asiaa niin paljon ajatellut. Kun minä en kerran
tahdo lasta, niin minkä tähden minun täytyy se synnyttää?

— Teidän olisi tullut ajatella sitä aikaisemmin, sanoin hiukan tylysti,

sillä en tahtonut ruveta mihinkään filosoofiseen keskusteluun
palvelustytön kanssa.
 Ehkä hän ei huomannut ääneni sointua tai olivat ajatukset tulleet
niin ylivoimaisiksi, että hänen täytyi saada ne purkaa jollekulle.

— Niin, onhan kyllä tohtori oikeassa. Enhän minä olisi antanut

mennä niin pitkälle, en koskaan, sillä ymmärsinhän minä, mikä
häpeä siinä on, mutta kun hän uuvutti minut hyväilyllään, niin en
lopulta tiennyt mitään, en osannut ajatella mitään, en osannut edes
puolustaa itseäni.

Mitä minulla oli siihen sanottavaa, tiesinhän minä, että tottunut

viettelijä voi saada valtaansa minkä tytön tahansa, ja että ihmisissä
on onnen kaipaus hetkittäin niin suuri, että he pienestä murusta
maksavat kaiken.

— Rakastitteko tuota miestä? kysyin.

— Ei kai se juuri rakkauttakaan ollut. Kun ei kukaan sitä ennen

ollut minulle niin hyvä, olin niin iloinen, kun hän minua hyväili. En
minä tiedä oliko se rakkautta. Olen minä ajatellut, että olisi se
enemmän.

— Oletteko ilmoittanut hänelle tulevanne äidiksi?

— Olen.

— Mitä hän siihen sanoi?

— Hän kiroili ja löi minua.

— Eikö' hän millään tavoin luvannut auttaa teitä?

— Ei.

— Ei luvannut mennä kanssanne naimisiin?

 — Ei. Enkä minä sitä tahtoisikaan nyt enään, sillä hän on tullut niin
toisenlaiseksi kuin ennen. Kyllä minä tiedän, ettei elämä hänen
kanssaan suinkaan helppoa olisi. Mutta vaikeampaa se olisi, jos
lapsi syntyy.

— Hän ei siis mitään tehnyt teitä auttaakseen.

— Kun hän näki minun itkevän, niin koetti hän olla hyvä, vaikka
minä huomasinkin, kuinka vihainen hän oli. Ja silloin hän käski
minua tulemaan tohtorin luo.

Hän kumartui eteenpäin, katsoi rukoilevasti minua, aivankuin lapsi,

joka oikein sydämmellisesti pyytää jotain ja sanoi:

— Jos nyt tohtori auttaisi minua, niin silloinhan olisi kaikki niin kuin
olla tulee, kukaan ei tiedä yhtään mitään.

— Te hätäilette liikoja, sanoin rauhoittaakseni häntä. Vaikka lapsi

syntyykin, niin ettehän sen kautta ole joutunut hukkaan maailmassa.
Voittehan lain kautta vaatia mieheltä elatusta lapsellenne.

— Mutta se häpeä?

— Voittehan muuttaa toiselle paikkakunnalle, jossa kukaan ei teitä

tunne.

— Kyllähän minä sitäkin olen ajatellut, herra tohtori. Minä olen niin
monelta puolelta katsellut tätä asiaa, mutta kuka minut
palvelukseensa ottaa, kun minulla on lapsi. Jos jätän sen muualle,
niin kyllä ihmiset kaikesta tiedon saavat. Meidän talon rouva kuuluu
kyllä valkonauhaan. Sen tiedän. Ja hän kyllä sanoissaan on niin
kovasti hyvä ihmisille, mutta jos menisin hänelle sanomaan että saan
lapsen, niin ajaisi hän minut heti talosta pois. Kuuluu tässä
takavuosina tehneen sillä lailla eräälle toiselle tytölle, jonka kävi
samalla lailla kuin minunkin. Se tyttö on nyt Helsingissä ja kuuluu
olevan sellainen huono nainen. Ja minä tahdon elää kunnollisena!

Hän lausui nämä viimeiset sanat melkein kirkaisten.

— Pelkäättekö, että teidän käy samalla lailla kuin tuon toisenkin?

— Minne tässä muuallekaan joutuu? Minä jään niin kovasti yksin

heti, kun tämä tulee ihmisten tietoon. Ajattelin minä kotiakin mennä
ja siellä kertoa kaikki, mutta tohtori ei tiedä, millainen isäni on. Äiti
kyllä säälisi ja antaisi minun olla siellä, mutta isä ei koskaan. Ja kun
palveluspaikassani tämä huomataan, niin millä minä sitten elän? Ja
juuri sinä kaikkein pahimpana aikana!

— Kyllä hyvät ihmiset teitä auttavat, ja jos pulaan joudutte, niin en

minäkään ole syrjässä.

— Kyllä ihmiset lupaavat niin paljon, mutta ne unohtavat niin pian.

Kyllä ihmisiltä sanoja saa, mutta ei muuta apua silloin, kun sitä oikein
tarvitsee. Enhän minä tohtoria epäile, mutta mikäs velvollisuus
tohtorilla olisi auttaa minua, yksinäistä tyttöä?

— Ajatelkaahan asiaa nyt rauhallisesti, sanoin minä. Lapsi voi olla

terve ja kaunis lapsi. Teillä voi olla hyvinkin paljon iloa hänestä
maailmassa, ja kun sitten hän varttuu, niin mikä turva teillä olisikaan
hänestä!

— Uskooko tohtori sitä?

— Uskon.
 — En minä vielä koskaan ole nähnyt, että yksinäisen naisen lapsi
tekisi muuta kuin koettaisi saada ihmiset unohtamaan, että hänellä ei
ole isää ja silloin hän karttaa äitiään. Ja silloinhan minä olisin läpi
koko elämän kaikkia toisia ihmisiä huonompi. Ja olenko minä sitten
huonompi? Toiset elävät paljoa pahemmin kuin minä ja, jollei heille
vahinkoa tapahdu, niin kukaan ei sano yhtään mitään.

Olin vaiti. Minulla ei ollut mitään hänelle sanottavaa. Olinhan

lausunut kaiken sen, mitä sellaisissa tapauksissa saattoi sanoa. Olin
käyttänyt kaikki ne valheelliset lauseet, joiden avulla koetamme
uskotella, että maailma on parempi kuin mitä se onkaan. Minä tiesin,
että minun olisi pitänyt häntä auttaa. Minun olisi pitänyt estää lasta
syntymästä maailmaan, koska sitä ei äiti tahtonut synnyttää, mutta
laki oli, siinä suhteessa selvä. Minä en uskaltanut antautua siihen
vaaraan, minkä tämänlaatuisen teon ilmi tuleminen voisi tuottaa, ja
pelkurina syydin kaiken edesvastuun tunteen pois. Huomasin hänen
odottavan vastausta. Kun olin vaiti, toivoi hän varmaankin minun
suostuneen pyyntöönsä. Sanoin senvuoksi hänelle:

— Minä en voi mitään tässä asiassa tehdä.

Hän ponnahti seisaalleen, niin odottamaton oli sittenkin tämä

vastaukseni. Näin hänen hätääntyneet silmänsä, kuulin miten hän
veti henkeä syvään tukahuttaakseen parkaisun. Hän nyyhkytti hiljaa
hetkisen ja lausui sitten kummallisen kylmästi:

— Ei kai sitten minulla ole enää mitään täällä tehtävää? Olenko

minä mitään tohtorille tästä velkaa?

— Ette, vastasin ja minua hävetti sitä sanoessani.

Hän meni ovelle, niiasi siellä kauniisti ja sanoi:

 — Kiitos kumminkin, kun tohtori on ollut niin hyvä, että on edes
kuunnellut.

Pää painui kumaraan ja hän meni.

Hän oli siis synnyttänyt lapsen ja hätääntyneenä sen surmannut.

Tämän ajan kuluessa oli vahtimestari tuonut arkun sisään, nosti

ruumiin siihen ja ruuvasi kannen kiinni. Minä tiesin, että hän ei enään
avaa arkkua, eikä kukaan muukaan sitä tee, että leikkaus,
ruumiinavauksen tulos, oli siis se, mitä minä sanoin.

Käännyin oven luona olevan poliisin puoleen kysyen:

— Tiedetäänkö mitään tämän lapsen äidistä?

Poliisi nousi nopeasti ja vastasi:

— Kyllähän minä tytön tunsin ja sääli minun tuli kovasti häntä, sillä
hän oli hiljainen ja siivo tyttö ja vasta 18 vuotias. Kuinka sattuikin niin
pahasti sekaantumaan, että meni lapsensa tappamaan?

En tiedä mistä nyt juolahti ajatus, että tuo mies tyttöä mahdollisesti
rakasti, rakasti sittenkin, vaikka tiesi hänen hairahtuneen. Ehkä sain
sen ajatuksen siitä lämpimästä sävystä, joka oli hänen äänessään.

— Missä hän synnytti lapsen? kysyin.

— Metsässä tuo kuului tapahtuneen. Tyttö parka oli siellä ollut

aivan yksin. Kun oli lapsensa tappanut, ei hän sitä edes peittänyt
vaan heti, kun jaksoi kävellä, tuli kaupunkiin ja kertoi ensimäiselle
poliisille, että niin ja niin hän oli tehnyt. Poliisi meni hakemaan, ja
sieltähän lapsi löytyi, minne tyttö sanoi sen jättäneensä.
 Katsoin terävästi häneen ja sanoin:

— Oletteko varma siitä, että hän tappoi lapsensa?

— Tottakai, kun hän itse sen sanoi.

— Mutta tiedättehän kyllä, että hätääntynyt ihminen voi usein

syyttää itseään sellaisestakin, jota ei ole tehnyt? sanoin minä.

Hänen kasvonsa aivan kuin kirkastuivat.

— Se on kyllä totta. Kyllähän se monta kertaa on huomattu, mutta

ei kukaan tässä asiassa tullut sitä ajatelleeksi.

— Sellaista on ajateltava, sanoin minä ankarasti. Minä olin

päättänyt, mitä teen ja kaikki oli minulle päivän selvää. Johtui
mieleeni tuo tytön lause: Ja minä tahdon elää kunnollisena! Jos hän
olisi ollut varakas, niin olisi hän tehnyt niiden tavalla jouduttuaan
harhateille. Olisi mennyt ajoissa ulkomaille ja palannut sieltä
vapaana. Pelastaakseen kunniansa ei rikkaan tarvitse muuta tehdä
kuin maksaa ulkomaalaiselle lääkärille hiukan rahaa. Rahaa, rahaa,
ei muuta kuin rahaa! Rahalla he paikkaavat kunniansa!

Minä en voinut todistukseen kirjoittaa, että lapsi on tukehutettu,

sillä tyttö joutuu kuritushuoneeseen. Ikäänkuin se voisi hänet
kohottaa ja tehdä hänet kelvolliseksi ihmisten seurassa olemaan.
Vaikka hän suorittaisikin tuon ihmisten vaatiman sovitusuhrin, niin
eivät ihmiset sittenkään tunnustaisi häntä vertaisekseen. Minä en
tahtonut epäröidä. Minä tahdoin pelastaa ihmisolennon, kun sen voin
tehdä. Minä en tahtonut ajatella, mikä lain mukaan on oikein, mikä
väärin.
 Tyyneesti, arkailematta kirjoitin, että lapsi oli syntyessään ollut
kuollut.

Ojensin paperin poliisille. Hän katsahti siihen ja ennen lähtöään

hän puristi lujasti kättäni. Olen melkein varma siitä, että hän lopulta
täydellisesti tiesi, mitä minä olin tehnyt, mutta minä tiesin myöskin,
että hänkin piti tekoani oikeana.

Paria viikkoa myöhemmin kävi tyttö luonani. Me emme puhuneet

koko asiasta mitään. Annoin hänelle rahoja, jotta hän pääsisi toiseen
kaupunkiin matkustamaan. Minä tiesin, että hän tahtoi ja että hän voi
myös alkaa elämänsä alusta, sillä sen osaa ihminen parhaiten tehdä
yksinään eikä toisten avulla kuritushuoneen kautta.

Toivon kuitenkin hartaasti, etten joutuisi toiseen samanlaiseen

tapaukseen. Minä en voisi toimia armottomasti, sillä vaikka olenkin
lääkäri, niin totta kai ensi sijassa saan olla ihminen.
TYLSÄMIELINEN

Hän oli varakas liikemies, harvinaisen intelligentti ihminen, joka oli

saavuttanut kaupungissa täydellisen luottamusaseman, kuuluen sen
vuoksi kaikkiin tärkeimpiin kunnallisiin komiteoihin, oli useiden
johtokuntien jäsen, kaupungin valtuuston varapuheenjohtaja. Hän oli
tunnettu laajalle ulottuvasta hyväntekeväisyydestään, jonka hän oli
hyvin järkevästi järjestänyt. Tämä seikka ehkä aiheutti sen, että
holhouslautakunta, jonka jäsenenä minäkin olen, asetti hänet
holhoamaan muuatta varakasta, tylsämielistä tyttöä, jonka äiti vuosi
sitten oli kuollut.

Me kaikki otaksuimme, että hän toimittaisi tytön johonkin

turvakotiin, koska pidimme sitä aivan luonnollisena, sillä emmehän
voineet otaksuakaan, että kukaan tavallinen ihminen ottaisi
hartioilleen sitä taakkaa, mikä tylsämielisen olennon hoitamisesta on,
vaikkakin hän tästä toimestaan saisi kohtuullisen palkkion.

Niinkauan kun tytön äiti eli, oli tämä pitänyt lasta luonaan, uhraten
koko elämänsä miehensä kuoleman jälkeen tämän onnettoman
olennon hoivaamiseksi.
 Koko holhouslautakunnan hämmästys oli jokseenkin suuri, kun
kuultiin, että liikemies oli ottanut tämän tytön omaan kotiinsa
asumaan. Hänen tekonsa oli meidän jokaisen mielestä suorastaan
suuremmoinen ja kunnioituksemme tätä miestä kohtaan kasvoi.

Liikemies asui yhdessä vanhanpuoleisen sisarensa kanssa, ja

tämä nainen oli avannut sydämensä, sulkeakseen siihen tuon
onnettoman olennon, koettaakseen kehittää hänessä nukkuvia
ominaisuuksia. Tämä vanha intelligentti nainen neuvotteli usein
minun kanssani siitä, millä keinoin hänen tuli kasvattaa tylsämielistä
tyttöä. Hän oli etevä psykoloogi, suuri ihmistuntija, ja juuri senvuoksi
hän saattoi vapautua kaikista niistä tavallisista opetussuunnitelmista,
joita tylsämielisten kasvatuksessa käytetään, ja laatia uuden, aivan
erikoisen, tähän tapaukseen sopivan menettelytavan. Tulos olikin
harvinaisen hyvä. Tyttö ei ollut montakaan kuukautta ollut hänen
hoteissaan, kun jo selvästi huomasi suuren muutoksen hänessä
tapahtuneen. Hänen äitinsä oli kohdellut lastaan, joka nyt oli 18
vuoden vanha, aivan kuin pikkulasta, pysäyttäen alkuunsa nekin
pienet mahdollisuudet, jotka hänellä oli henkiseen kehitykseen.
Kaikki oli nyt aivan toisin. Niin pian kuin tyttöä kohdeltiin täysi-
ikäisenä, asetettiin hänelle täysi-ikäisen vaatimukset, alkoi hänessä
myöskin herätä ominaisuuksia, joiden olisi luullut jo ainiaaksi
kuolleen. Hän oppi lukemaan ja kirjoittamaan, niin että ehdottomasti
otaksuin hänen oman äitinsä rakkaudellaan saaneen vähemmän
aikaan, kuin mitä vieras ihminen sai johdonmukaisella järkiperäisellä
käsittelyllä.

Minua tämä kummallinen ihmissielun herääminen hämmästytti.

Kuukausi kuukaudelta näin yhä suurempia edistysaskeleita ja iloitsin
siitä, varsinkin hänen uuden kasvatusäitinsä tähden, joka tällä tavoin
sai jonkinmoisen palkkion suuresta uhrautuvaisuudestaan. Mutta
koko ajan samalla pelkäsin sitä hetkeä, jolloin jonkinmoinen
seisahdus tapahtuu, sillä sellaisen seisahduksen täytyy tulla, sen
tiesin.

Kun eräänä päivänä saavuin tämän vanhan neidin luo ja kyselin

häneltä hoidokkinsa vointia, näin hänen tulevan alakuloiseksi.

— Onko jotain ikävää tapahtunut? kysyin.

— On.

— Mitä?

— Minun on sitä vaikea Teille sanoa, mutta olen pyytänyt veljeäni

puhelemaan siitä kanssanne.

Kun hänen veljensä vähän ajan kuluttua saapui huoneeseen,

sanoi sisar:

— Minä olen maininnut tohtorille, että jotain ikävää on tulossa

holhokkimme suhteen, ja olen sanonut, että sinä aijot siitä hänen
kanssaan puhua.

— Tietysti minä sen teen, koska sinä niin tahdot.

Kääntyen minun puoleeni sanoi hän:

— Ehkä menemme konttoorin puolelle keskustelemaan.

Varsinainen konttooriaika oli jo päättynyt. Tiesimme siis

saavamme olla aivan rauhassa. Kun olimme istuutuneet, alkoi hän:

— Te tiedätte, millä suurella rakkaudella ja innostuksella sisareni

on koettanut tätä onnetonta ihmislasta kehittää?
 — Tiedän ja olen nähnyt, että hän on saanut suorastaan ihmeitä
aikaan.

— Niin, tyttö on todellakin henkisesti odottamattoman paljon

kehittynyt. Hän osaa lukea ja kirjoittaa. Hänen keskustelussaan on
paljon hyvin järkevää, ja minä uskon, että hänestä ajan mittaan tulee
jokseenkin onnellinen ihminen, joka voi elämässä tulla suhteellisen
hyvin toimeen, sillä onhan hänellä siksi suuri perintö, jonka olen
koettanut sijoittaa mahdollisimman suuressa määrässä korkoa
tuottavaksi. Mutta tämän henkisen heräämisen ohella on tapahtunut
toinen odottamaton herääminen. Se on aistillisuus.

— Minä melkein sitä odotin.

— Ja tämä herääminen on ollut voimakkaampi, kuin mitä olemme

uskoneetkaan. Jos hän olisi täydellisesti normaali, niin silloinhan
voisimme järkevän keskustelun kautta häneen tässä suhteessa
vaikuttaa, mutta nyt toimii aistillisuus hänessä peloittavan
voimakkaana luonnonvoimana, joka mikä hetki tahansa voi tuottaa
vaarallisia seurauksia mukanaan. Olen tätä asiaa monelta kannalta
punninnut. Yksinkertaisintahan olisi, että nyt lähettäisin hänet
hulluinhuoneeseen. Mutta silloinhan hänen henkinen kehityksensä
loppuisi, sillä ette varmaankaan usko, että tällaisessa laitoksessa
hän saisi sitä hoitoa osakseen, minkä sisareni voi antaa. Tämän
tekoni kautta tuomitsisin tyttöraukan elinkautiseen vankeuteen.
Toiselta puolelta taas ajatellen, jos tämä hänessä piilevä aistillisuus,
joka näyttää olevan perintöä isältään, jatkuu niin kuin hyvin luultavaa
on, täytyy meidän pitää häntä alituisen valvonnan alaisena täällä
kodissa, ja sen kautta ehkäisemme järjen kehitystä, joka on
mahdollinen ainoastaan sen kautta, että hän tuntee itsensä vapaaksi
ja yhdenvertaiseksi toisten ihmisten kanssa. Ja olkoon valvonta
kuinka huolellinen tahansa, helposti voi sattua tapaus, jolloin hän
luonteensa pakoittamana ryhtyy tekoihin, antautuu miehelle, ja
arvaattehan mikä silloin on seuraus. Suuri edesvastuuhan lankee
silloin minun niskoilleni, mutta en ajattele niin paljoa sitä, kuin kuinka
onnettomaksi tytön elämä sen kautta muodostuu. Mikä roisto
tahansa voi siittää hänessä lapsen, ja arvaattehan, millainen sikiö
siitä tulee.

— Ja mitä vaatisitte minulta, mitä tulisi minun tehdä tässä

asiassa?

— Juuri siitä tahdoin Teidän kanssanne keskustella. Teidän täytyisi

tehdä hänessä leikkaus, joka tekisi lapsen synnyttämisen
mahdottomaksi.

— Te kai tiedätte, että minulla ei ole oikeutta siihen, että lääkäri ei

saa tehdä ihmistä steriiliksi, ei edes asianomaisen pyynnöstä, vielä
vähemmin hänen holhoojansa toivomuksesta.

— Jos tämä tapahtuisi itsekkäässä tarkoituksessa, niin silloinhan

sen ymmärtäisi, ja onhan luonnollista, että lain täytyy estää ihmistä
sillä tavoin itseään silpomasta, mutta asianlaita on nyt aivan toinen.
Tämän leikkauksen kautta voitte tehdä onnettoman ihmisen elämän
suhteellisen onnelliseksi.

— Minä pyydän, elkää jatkako enään. Minä olen vapaamielinen

lääkäri, tiedän kyllä, mitä sydämeni tässä asiassa sanoo. Mutta minä
olen sidottu! Minä en voi, minä en voi tehdä mitään!

— Antakaa minulle edes jokin neuvo.

 — Ne neuvot olette jo itse äsken lausunut. Joko panette hänet
hulluinhuoneeseen ja silloin vältätte kaiken mahdollisen vaaran tai
säilytätte häntä kodissanne ja koetatte varjella häntä kaikesta siitä,
mikä saattaa tulla.

Hän istui kalpeana paikallaan ja minä näin, että hän oli minun
suhteeni pettynyt. Se koski minuun kovasti. Mielelläni olisin tehnyt
kaiken, nauttiakseni sellaisen miehen kunnioitusta.

Tämä ajatus oli kauan aikaa mielessäni. Minä en päässyt siitä

vapaaksi.
Koetin punnita asiaa joka puolelta ja yhäti palasin takaisin samaan.
Hän oli oikeassa ja minä olin väärässä. Hän ajatteli oikein ja koko
muu
maailma väärin.

Kului jonkun aikaa. Eräänä päivänä satuin tulemaan jälleen

vanhan neidin luo ja tapasin hänet itkemässä. Kysyin syytä ja hän
mainitsi sen. Tuo hänen kasvattinsa, jonka kehitystä hän niin
suurella rakkaudella oli valvonut, oli alkanut rakkaussuhteen talon
rengin kanssa.

Raukkamaisia, niinkuin me ihmiset tavallisesti olemme vakavien

asioiden edessä, olin minäkin neuvomalla raukkamaisen keinon.

— Tämähän asia on helposti järjestettävissä, sanoin minä. Jos

aistillisuus kerran tytössä on siksi suuri, niin antakaa hänen mennä
naimisiin tuon rengin kanssa.

— Mutta sehän on aivan mahdottomuus, eihän sellainen voi tulla

kysymykseenkään, mieshän on suuri juoppo ja keuhkotautinen.
Sellainen yhdistelmä olisi suoranaista inhimillistä kurjuutta. Eikä
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookfinal.com