Approximation And Optimization Of Discrete And

Differential Inclusions 1st Edition Elimhan N

Mahmudov download

https://ebookbell.com/product/approximation-and-optimization-of-
discrete-and-differential-inclusions-1st-edition-elimhan-n-
mahmudov-2405330

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Optimization And Approximation On Systems Of Geometric Objects Diss

Van Leeuwen E

https://ebookbell.com/product/optimization-and-approximation-on-
systems-of-geometric-objects-diss-van-leeuwen-e-1440676

Approximation And Optimization Algorithms Complexity And Applications

1st Ed Ioannis C Demetriou

https://ebookbell.com/product/approximation-and-optimization-
algorithms-complexity-and-applications-1st-ed-ioannis-c-
demetriou-10486020

Duality For Nonconvex Approximation And Optimization Singer I

https://ebookbell.com/product/duality-for-nonconvex-approximation-and-
optimization-singer-i-997768

Optimization And Approximation Pedregal Pablo

https://ebookbell.com/product/optimization-and-approximation-pedregal-
pablo-6753464
Complexity And Approximation Combinatorial Optimization Problems And
Their Approximability Properties Harcdr G Ausiello

https://ebookbell.com/product/complexity-and-approximation-
combinatorial-optimization-problems-and-their-approximability-
properties-harcdr-g-ausiello-1529782

Exploring Mathematical Analysis Approximation Theory And Optimization

270 Years Since Am Legendres Birth 1st Edition Nicholas J Daras

https://ebookbell.com/product/exploring-mathematical-analysis-
approximation-theory-and-optimization-270-years-since-am-legendres-
birth-1st-edition-nicholas-j-daras-54809334

Approximation Optimization And Mathematical Economics 1st Edition Jan

Andres

https://ebookbell.com/product/approximation-optimization-and-
mathematical-economics-1st-edition-jan-andres-4210280

Approximation Randomization And Combinatorial Optimization Algorithms

And Techniques 11th International Workshop Approx 2008 And 12th
International Workshop Random 2008 Boston Ma Usa August 2527 2008
Proceedings 1st Edition Micah Adler
https://ebookbell.com/product/approximation-randomization-and-
combinatorial-optimization-algorithms-and-techniques-11th-
international-workshop-approx-2008-and-12th-international-workshop-
random-2008-boston-ma-usa-august-2527-2008-proceedings-1st-edition-
micah-adler-1225962

Nonlinear Analysis Approximation Theory Optimization And Applications

1st Edition Qamrul Hasan Ansari Eds

https://ebookbell.com/product/nonlinear-analysis-approximation-theory-
optimization-and-applications-1st-edition-qamrul-hasan-ansari-
eds-4698488
Approximation and Optimization of Discrete and
Differential Inclusions
Approximation and
Optimization of Discrete
and Differential Inclusions

Elimhan N. Mahmudov
Industrial Engineering Department
Faculty of Management
Istanbul Technical University
Macka, Istanbul, Turkey
Azerbaijan National Academy of Sciences
Institute of Cybernetics, Azerbaijan

AMSTERDAM BOSTON HEIDELBERG LONDON NEW YORK OXFORD

G G G G G

PARIS SAN DIEGO SAN FRANCISCO SINGAPORE SYDNEY TOKYO

G G G G G
Elsevier
32 Jamestown Road London NW1 7BY
225 Wyman Street, Waltham, MA 02451, USA
First edition 2011
Copyright r 2011 Elsevier Inc. All rights reserved
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our
arrangement with organizations such as the Copyright Clearance Center and the Copyright
Licensing Agency, can be found at our website: www.elsevier.com/permissions
This book and the individual contributions contained in it are protected under copyright by
the Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and
experience broaden our understanding, changes in research methods, professional practices,
or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described
herein. In using such information or methods they should be mindful of their own safety and
the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of
products liability, negligence or otherwise, or from any use or operation of any methods,
products, instructions, or ideas contained in the material herein.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the Library of Congress
ISBN: 978-0-12-388428-2

For information on all Elsevier publications

visit our website at elsevierdirect.com

This book has been manufactured using Print On Demand technology. Each copy is
produced to order and is limited to black ink. The online version of this book will show
color figures where appropriate.
Dedication

This book is dedicated to the memory of my doctoral thesis advisor B.N. Pshenichnyi,
Academician of the Academy of Sciences of Ukraine, from whom I learned the theory
of extremal problems, numerical methods in optimal systems theory, optimal control,
differential game theory, and much more.
This book is also dedicated to memory of my lovely son Elshad (1977 2000),
who was a selfless manager for the American petrol company Exxon.
Besides this book is dedicated to the memory of Khojali massacre’s martyrs
(25-26.02.1992).
Preface

Give me a place to stand on, and I will move the Earth.

Archimedes

Mathematics is the language with which God has written the universe.
Galileo Galilei

As a well spent day brings happy sleep, so life well used brings happy death
Leonardo da Vinci

Scientists investigate that which already is;engineers create that which has never
been
Albert Einstein

The primary goals of this book are to present the basic concepts and principles of
mathematical programming in terms of set-valued analysis (Chapters 2 and 3) and
on the basis of the method of approximation, to develop a comprehensive optimal-
ity theory of problems described by ordinary and partial differential inclusions
(DFI) (Chapters 4 6). This book consists of six chapters divided into sections and
subsections, and contains many results that have not been published in the mono-
graphic literature.
In Chapter 1, convex sets and convex functions are studied in the setting of n-
dimensional Euclidean space. However, the reader familiar with functional analysis
can generalize the main results to the case of infinite-dimensional functional
spaces. In spite of the fact that the stated notions and results are known, they play a
decisive role for obtaining the main results in the next chapters of the book. The
key issues of convex analysis in finite-dimensional spaces have been addressed in
the books Convex Analysis by Rockafellar and Convex Analysis and Extremum
Problems by B.N. Pshenichnyi. The identifications of convex functions and their
epigraphs make it easy to pass back and forth between the geometric and analytical
approaches. It is shown that convex sets and functions form classes of objects pre-
served under numerous operations of combination; pointwise addition, pointwise
supremum, and infimal convolution of convex functions are convex.
In Chapter 2, the apparatus of locally adjoint mappings (LAM) (which is new)
is studied in the light of convex analysis. It is the fundamental concept in what fol-
lows, and it is used to obtain the optimality conditions for the problems posed in
this book. We give the calculus of LAM on different multivalued mappings, such
as the sum, composition, and inverse. We introduce the adjoint (not locally adjoint)
x Preface

mapping, using the recession cone, and the connection between the adjoint and
LAM is established. Based on the adjoint mapping, the duality theorems for convex
set-valued mappings are proved.
Chapter 3 is devoted to applications of these basic tools to the study of mathemati-
cal programming with possibly nonsmooth data. Starting with problems of mathemati-
cal programming under functional and geometric constraints, we then consider
various problems of constrained optimization, minimax problems and equilibrium
constraints, infimal convolution of convex functions, duality in convex programming,
and duality relations. In order to formulate a necessary condition for the existence of
an extremum, of course, some special condition by function taking part in the given
problem is required. In particular, in some neighborhood of a point minimizing our
objective function, we deal with the comparatively easily computable functions. As is
known, a smooth function admits a linear approximation. On the other hand, a convex
function can be approached by positively homogeneous functions. However, a non-
smooth and nonconvex function cannot be approximated in a neighborhood of a point
by positively homogeneous functions. Precisely this class of functions is required for
introducing the concept of convex upper approximations (CUAs). The key tools of
our analysis are based on the extremal principle and its modifications together with
the LAM calculus.
Chapters 4 and 5 deal mostly with optimal control problems of the Bolza type
described by ordinary differential, high-order differential, delay-differential, and
neutral functional-differential inclusions. The development and applications of the
LAM are demonstrated in these problems with ordinary discrete and differential
inclusions. In particular, for polyhedral DFI, under the corresponding condition for
generality of position, the theorem of the number of switchings is proved. The cor-
responding results are obtained for linear optimal control problems in linear mani-
folds. For a nonautonomous polyhedral DFI, a special condition for generality of
position is formulated. Moreover, for problems described by ordinary nonconvex
DFI under the specially formulated monotonicity and t1-transversality conditions,
sufficient conditions for optimality are proved.
In Chapter 6, we continue the study of optimal control problems governed by dis-
crete and differential inclusions with distributed parameters, which during the past
15 20 years has been a basic source of inspiration for analysis and applications. Using
LAM and the discrete-approximation method in Hamiltonian and Euler Lagrange
forms, we derive necessary and sufficient optimality conditions for various boundary
values (Dirichlet, Neumann, Cauchy) problems for first-order, elliptic, parabolic, and
hyperbolic types of discrete and partial DFI. One of the most characteristic features of
such approaches with partial DFI is peculiar to the presence of equivalents to the LAM.
Such problems have essential specific features in comparison with the ordinary differ-
ential model considered in the second part of the book. For every concrete problem
with partial DFI, we establish rather interesting equivalence results that shed new light
on both qualitative and quantitative relationships between continuous and discrete
approximation problems.
Preface xi

In Chapter 5 and the second part of Chapter 6, we construct the dual problem of
convex problems for ordinary and partial differential inclusions of hyperbolic, para-
bolic, and elliptic types. We study separately the duality problems with first-order
partial differential inclusions. As is known, duality problems have always been at
the center of convex optimality theory and its applications. In this book, we formu-
late duality results and search for the conditions under which primary and dual pro-
blems are connected by such duality relations. For duality constructions of convex
problems, we use the duality theorems concerning infimal convolution and the
addition of convex functions.
Thus, we can list the major features of our book that make it unique:
G
The introduction of a new concept of LAM and its calculus.
G
The connection between LAM and adjoint (not locally) mappings defined in terms of the
recession cone.
G
Duality theorems for convex multivalued mappings established in terms of a Hamiltonian
function.
G
The basic results of mathematical programming in terms of Hamiltonian functions.
G
Under a suitable condition for generality of position, the theorem of the finiteness of
switching numbers for optimal control of polyhedral differential inclusions.
G
Under a special t1-transversality condition, new sufficient conditions for optimality in
terms of extended Euler Lagrange inclusions for Bolza-type problems with ordinary dif-
ferential inclusions and state constraints.
G
A new class of optimal control problems for higher-order differential inclusions.
G
Duality relations in mathematical problems with equilibrium constraints via recession
cones. Major features using the method of discrete approximation.
G
Optimization of first-order discrete and partial differential inclusions. Note that one of the
characteristic features of optimization of Cauchy for first-order discrete inclusions is the
intrinsic presence of the infinite dimensional self-adjoint Hilbert space l2.
G
Optimization of Darboux-type partial differential inclusions and duality.
G
Optimization of elliptic, hyperbolic, and parabolic types of discrete and partial differential
inclusions and duality.
G
Optimization of partial differential inclusions with a second-order elliptic operator.
G
Equivalence results that facilitate making a bridge between discrete and corresponding
discrete-approximation problems.
Throughout this book, a proof is marked with an empty square & and its end is
marked with a Halmos box, ’. Since many problems in engineering reduce to
such problems, the book will be of interest to mathematicians and nonmathemati-
cian specialists whose study involves the use of ordinary and partial differential
equations (inclusions) and approximation methods and its applications, as well as
to undergraduate, graduate, and postgraduate students at universities and technical
colleges. In other words, the book is intended for a broad audience—students of
universities and colleges with comprehensive mathematical programs, engineers,
economists, and mathematicians involved in the solution of extremal problems.
Basic material has also been incorporated into many lectures given by the author at
various international conferences in London, UK; Zurich, Switzerland; and Leipzig,
xii Preface

Germany, and the Banach international mathematical center in Warsaw, Poland,

during recent years.
This book includes an index of symbols and notation, and a subject index. Using
the subject index, the reader can easily find the page where some notion or notation
is introduced. Our notation and terminology are generally consistent with those
used by Rockafellar, Mordukhovich, and Pshenichnyi in their writings. For the
reader’s convenience, an introduction in each chapter of the book describes the
contents and commentaries, and outlines a selection of material that would be
appropriate for the subject. This book is also accompanied by an abundant bibliog-
raphy. Parts of this book have been used by me in teaching graduate and postgradu-
ate students on Convex Analysis, Optimal Control Theory, and Nonlinear Analysis
and Its Applications at Azerbaijan State University and Istanbul Technical
University.

Prof. Elimhan N. Mahmudov

Baku and Istanbul
August 2011
Acknowledgments

My first thanks go to Boris S. Mordukhovich, Professor in the Department of

Mathematics at Wayne State University, who inspired me during the work on this
book. I am particularly indebted to Alex Rubinov, V.F. Demyanov, and A.V.
Kryazhimskii at all stages of this book for recognizing and supporting my first
efforts in the field of locally adjoint mappings and optimization. I am indebted to
the Institute of Cybernetics, headed by T.A. Aliev, Academician of the National
Academy of Sciences of Azerbaijan. I would like to thank my genial colleagues C.
Kahraman, N. Serarslan, and S. Polat of Istanbul Technical University for their sup-
port during the preparation of this book. I am very grateful to the supportive editors
at Elsevier for their help during the preparation and publishing of this book; the
timely publication of this book is a direct consequence of the help of Lisa Tickner,
Publishing Director at Elsevier. Especially valuable help was provided by my doc-
toral students Özkan Değer, Murat Ünal, and Kutay Tinç. And I thank my pencil
sharpener, my younger daughter Elmira, for her interest in my book.
Prof. Elimhan N. Mahmudov
Baku and Istanbul
August 2011
About the Author

Elimhan N. Mahmudov was born in Kosali, Karayazi, and after graduating the secondary
school with a gold medal for being the most successful student, he attended Azerbaijan
State University as part of the Mechanical Mathematical Department. From 1973 to
1977, he was a PhD student at the Cybernetics Institute of Academy of Sciences of
Ukraine advised by Prof. B.N. Pshenichnyi—Academician, one of the most prominent
authorities in the fields of theory of extremal problems, numerical methods, and game
theory. E.N. Mahmudov defended his thesis in Ukraine in 1980. He then worked as a
Ph.D. and senior scientific associate at the Cybernetics Institute of Academy of Sciences
of Azerbaijan until 1995. He also taught part time at the Azerbaijan State University and
was a permanent member of the Physical Mathematical Doctoral Sciences Committee
in Azerbaijan. On the recommendation of the Physical Mathematical Doctoral
Committee of the Taras Shevchenko’s Kiev State University, in 1992 he became a Doctor
of Physical Mathematical Sciences, and he has earned both a Ph.D. and a Doctorate of
Sciences from Moscow. In 1992, he received a Grant Support for Mathematics by
National Science Foundation in Washington, DC, and became a member of the American
Mathematical Society.
Prof. Mahmudov has devoted numerous research papers to Convex Analysis,
Approximation Theory, Optimal Control Theory, Dual Problems, and Mathematical
Economy, which have been published in high-level Science Citation Index (SCI) jour-
nals in the former Soviet Union, the United States, and Europe. His textbooks include
Mathematical Analysis and Its Applications, published in 2002 by Papatya, Istanbul,
and Single Variable Differential and Integral Calculus (in press). He has lectured on
the problems of Optimal Control at International Conferences in England, France,
Switzerland, Germany, Poland, Russia, and Turkey, and at the World Mathematical
Banach Center in Warsaw, Poland. He is an Editor of the journal Advances in Pure
Mathematics (APM).
Prof. Mahmudov currently teaches Nonlinear Programming, Game Theory, and
applications in Economics, Nonlinear Optimization, and Numerical Analysis at
Istanbul Technical University. He enjoys playing chess, playing Tar and Saz,
Azerbaijan folk instruments, and traveling and is an avid painter.
1 Convex Sets and Functions
1.1 Introduction
Convexity is an attractive subject to study, for many reasons; it draws upon geome-
try, analysis, linear algebra, and topology, and it has a role to play in such topics as
classical optimal control theory, game theory, linear programming, and convex pro-
gramming. Convex sets and convex functions are studied in this chapter in the set-
ting of n-dimensional Euclidean space ℝn. (However, if you are familiar with
functional analysis, you will be able to generalize the main results to the case of
infinite dimensional functional spaces.) These results play a decisive role in obtain-
ing the main results in the next chapters of this book. In the field of convex analy-
sis, you can consult Rockafellar [228] and Pshenichnyi [226] for related and
additional material (most of the familiar results on convex analysis presented in
this chapter are taken from Pshenichnyi [226]). The basic idea in convexity is that
a convex function on ℝn can be identified with a convex subset of ℝn11, which is
called the epigraph of the given function. This identification makes it easy to move
back and forth between geometrical and analytical approaches. It is shown that
pointwise addition of functions, pointwise supremum, and infimal convolution of
convex functions are convex, in fact, convex sets and functions are classes of
objects that are preserved under numerous operations of combination. A function is
closed if its epigraph is closed. The latter is equivalent to lower semicontinuity of
functions. This leads to the notion of the closure operation for proper convex func-
tions, which corresponds to the closure operation for epigraphs.
In Section 1.2, we study some topological properties of sets and their convex
hull, and consider how a convex set can be characterized by both Minkowski’s
method and support functions. The role of dimensionality in the generation of con-
vex hulls is explored in Carathéodory’s theorem (Theorem 1.1). It is interesting
that Theorem 1.2 (Gauss
Lucas) says that the roots of the derivative of a noncon-
stant complex polynomial belong to the convex hull of the set of roots of the poly-
nomial itself. The foundations for extremal theory are laid in the Separation
Theorems 1.5
1.7.
In Section 1.3, we discuss the convex cone, which is one of the important con-
cepts in convex analysis and extremal theory. The investigation of its properties is
connected with the calculation of the dual cone.
The cones K1, . . . , Km are called separable if there exist not all zero vectors
xi AKi ; such that x1 1 ? 1 xm 5 0: By Theorem 1.12, if K 5 K1\?\Km, then
either K  5 K1 1 ? 1 Km or the cones K1, . . . , Km are separable. By Lemma 1.17,

Approximation and Optimization of Discrete and Differential Inclusions. DOI: 10.1016/B978-0-12-388428-2.00001-1

© 2011 Elsevier Inc. All rights reserved.
2 Approximation and Optimization of Discrete and Differential Inclusions

ðK1 \ ? \ Km Þ 5 K1 1 ? 1 Km : But since for polyhedral cones, K1 1 ? 1 Km is
also a polyhedral cone, this sum of cones is closed and the bar above it can be
removed. One of the remarkable properties of a polyhedral set is that it can be repre-
sented as a sum of polytope (polyhedron) and polyhedral cone. Conversely, the sum
of any polytope and polyhedral cone is a polyhedral set (Theorem 1.14). The reces-
sion cone of a nonempty convex set M, i.e., the set of vectors x such that M 1 x C M
is denoted by 01 M and for a bounded set 01 M 5 {0}.
In Section 1.4, we develop the main properties of convex functions. Recall that
by Definition 1.20 a function f is said to be proper, if f(x) , 1N for at least one x
and f(x) . 2N for every x. A function that is not proper is improper. It follows
from Lemma 1.24 that dom f is convex, even if f is an improper function. Besides,
an improper convex function may have a finite values only at points of the relative
boundary of dom f. The sum of proper convex functions fi, i 5 1, . . . , m with non-
negative coefficients is convex (Lemma 1.27).
It is known that the indicator function δM(  ) of M is useful as a correspondence
between convex sets and convex functions. Then note that the sum of proper con-
vex functions fi, i 5 1, . . . , m with nonnegative coefficients may not be a proper
function (Lemma 1.27). For example, for the disjoint sets M1 and M2, the sum of
indicator functions δM1 1 δM2 is identically 1N.
We shall denote the gradient of f at x by f 0 (x) and the Hessian matrix of f at x by
fv(x), whose (i,j)th element is @2f/@xi@xj. Then if f is twice differentiable, the con-
vexity of f and the positive semidefiniteness of fv(x) are equivalent. Of course, the
latter is an important result not only in analysis but also in nonlinear programming.
Lemma 1.29 implies that properness of convex functions is not always preserved
by infimal convolution f1"f2, which is commutative, associative, and convexity-
preserving. Indeed, if f1 and f2 are linear functions not equal to each other, then
their infimal convolution identically is 2N.
The convex hull conv g of a nonconvex function g, defined as the greatest con-
vex function majorized by g, is used in establishing the dual problem governed by
polyhedral maps in Section 5.2. In Theorems 1.16 and 1.17, the continuity and
Lipschitz properties of convex functions are shown. By Theorem 1.18, f, a proper
convex function, is necessarily continuous on ri dom f. As is seen from this theo-
rem, a convex function is continuous in dom f and may have a point of discontinuity
only in its boundary. In order to characterize the case in which there is no such dis-
continuity, it is convenient to introduce the closure function concept (a function f is
said to be a closure if its epigraph epi f is a closed set in ℝn11). By Definition 1.26,
the recession function is denoted by f 01 and defined as epi (f 01 ) 5 01 (epi f ).
Obviously, if f is a proper convex function, then the recession function f 01 of f is a
positively homogeneous proper convex function.
Section 1.5 is devoted to the conjugate of a convex function, which is one of the
basic concepts both of convex analysis and of duality theory. The definition of the
conjugate of a function grows naturally out of the fact that the epigraph of a closed
proper convex function on ℝn is the intersection of the closed half-spaces in ℝn11
that contain it. The function defined as f  ðx Þ 5 supx fhx; x i 2 f ðxÞg is called the
conjugate of f. It is closed and convex. It is useful to remember, in particular, that
Convex Sets and Functions 3

Young’s inequality f(x) 1 f  (x ) $ hx,x i holds for any function. If here f is a proper
convex function, then we shall refer to this relation as Fenchel’s inequality. Taking
conjugates clearly reverses functional inequalities: f1 $ f2 implies f1 # f2 :
The polar of a set M is denoted by MO and defined as MO 5 {x : HM(x ) # 1},
where HM(x ) is a support function of M. Note that if K is a convex cone, then
KO 5 {x :hx,x i # 0, ’xAK} and so K 5 2KO. Then rM  
ðx Þ 5 δMO ðx Þ; where
rM(  ) is Minkowski’s function. If f is a closed proper convex function, then f 5 f 
(Theorem 1.21). Thus, the conjugacy operation f-f  induces a symmetric one-
to-one correspondence in the class of all closed proper convex functions on ℝn. By
Theorem 1.23, if f is a closed proper convex positively homogeneous function, then
f  (x ) 5 δdom f  (x ) and dom f  is a closed set. The conjugate function of the
support function of a closed convex set is the indicator function of this set
(Theorem 1.25).
In Section 1.6, we analyze directional derivatives and subdifferentials of convex
functions. The theory of subdifferentiation is a fundamental tool in the analysis of
extremum problems. It is known that, in general, convex functions are not differen-
tiable. Nevertheless, these functions have many useful differential properties, and
one of them is the fact that directional derivatives exist universally. Moreover, for
convex function can be defined as the notion of subgradient and the set of subgra-
dients is the subdifferential conception. If f is a proper convex function and
x0Adom f, then the value of the directional derivative f 0 (x0,p), finitely or not,
always exists for all p (Lemma 1.31). Obviously, the subdifferential @f(x0) is a
closed convex set. The directional derivative f 0 (x0,p) is a positively homogeneous
convex function of p. Theorem 1.27 asserts that x A@f(x0) if and only if hx0,x i 2
f(x0) 5 f  (x ). This simple fact will be used in the next investigations. Also, if f is a
closed proper convex function, then x A@f(x0) if and only if x0A@f  (x ) (Corollary
1.2). In Section 1.4, we will see that multiplication by positive constants of convex
functions, their addition, and the pointwise supremum of convex functions are
again convex. So it is important to calculate the subdifferentials of such functions.
We have that for f(x) 5 α f0(x), α . 0, @f(x0) 5 α@f0(x0). Let f1, f2 be proper convex
functions and f 5 f1 1 f2, x0Adom f1\dom f2 and suppose either (1) there is a point
x1Adom f1\dom f2, where f1 is continuous, or (2) ri dom f1\ri dom f2 6¼ [. Then @f
(x0) 5 @f1(x0) 1 @f2(x0) (Moreau
Rockafellar). In Theorem 1.31, we calculate the
subdifferential of functions of the form f ðxÞ 5 supα f f ðx; αÞ: αAAg:
If a set M is given by the inequality of the form M 5 {x: f(x) # 0}, where f is a
convex function, then under the existence of an interior point it is proved that the
dual cone is expressed by cone @f(x0), i.e., [cone(M 2 x0)] 5 2cone @f(x0)
(Theorem 1.33).

1.2 Some Basic Properties of Convex Sets

We consider the vector x, belonging to n-dimensional real Euclidean space X 5 ℝn.
We also consider its dual space X 5 ℝn. In this book, the inner product of two
4 Approximation and Optimization of Discrete and Differential Inclusions

P  
vectors xAX and x AX is defined by hx; x i 5 ni51 xi xi ; where xi and xi are the
components of x and x , respectively. Let us explain the difference between the
two spaces X and X . The point is that each element of X plays the role of a linear
function defined on X. It is well known that any linear function l(x) defined on X
can be given by some x so that l(x) 5 hx, x i. Thus, in order to emphasize the fact
that x defines a linear function, we introduced such notation.

Definition 1.1. A subset M is said to be convex if

λ1 x1 1 λ2 x2 AM; λ1 ; λ2 $ 0; λ1 1 λ2 5 1:

Whenever x1, x2AM, it follows that [ and ℝn are convex.

Let us give some important properties of convex sets.

Lemma 1.1. Let {Mi:iAI} be a collection of nonempty convex sets in ℝn, where
I is an arbitrary index set. Then the intersection
M 5 \ Mi
iAI

is also convex.
& By definition, if x1, x2AM, then x1, x2AMi, iAI, and so λ1x1 1 λ2x2AMi for all
iAI and hence λ1x1 1 λ2x2AM’

Lemma 1.2. If M1 and M2 are convex sets in ℝn, then so is the sum c1M1 1 c2M2,
where c1, c2 are real numbers.
& Recall that by definition, c1M1 1 c2M2 is the set of points of the form
c1x1 1 c2x2, where x1AM1, x2AM2. Now, the proof of the lemma follows from the
definition of a convex set.’

Lemma 1.3. If M is a convex set and c1 $ 0, c2 $ 0, then

ðc1 1 c2 ÞM 5 c1 M 1 c2 M:

& The inclusion C would be true whether M is convex or not. The reverse inclu-
sion * follows from the convexity relation
c1 c2
M* M1 M
c1 1 c2 c1 1 c2

on multiplying through by c1 1 c2, provided c1 1 c2 . 0. If c1 5 c2 5 0, the asser-

tion of the theorem is trivial.’

Definition 1.2. A point x is called a convex combination of points x1, . . . , xm if

there are nonnegative numbers λ1, . . . , λm such that
x 5 λ1 x1 1 ? 1 λm xm
ð1:1Þ
λi $ 0; i 5 1; . . . ; m; λ1 1 ? 1 λm 5 1:
Convex Sets and Functions 5

Lemma 1.4. If M is a convex set and xiAM, i 5 1, . . . , m, then the set M contains
all the convex combinations of elements x1, . . . , xm.
& It is convenient to use mathematical induction; for m 5 2, the assertion follows
from the definition. Make the induction hypothesis that M contains all the convex
combinations of elements x1, . . . , xm, where m # k. We show that
x 5 λ1 x1 1 λ2 x2 1 ? 1 λk xk 1 λk11 xk11 AM

for all λi, such that λi $ 0, λ1 1 ? 1 λk 1 1 $ 1. For convenience, suppose λi . 0.

Thus, 1 2 λk 1 1 5 λ1 1 ? 1 λk . 0. Since
X
k
λi 1 Xk
5 λi 5 1;
i51
1 2 λk11 1 2 λk11 i51

by the induction hypothesis

λ1 λk
y5 x1 1 ? 1 xk AM:
1 2 λk11 1 2 λk11

Therefore, by the definition of convex set

x 5 ð1 2 λk11 Þy 1 λk11 xk11 AM:’

Note that in many situations in which convex combinations occur in applied mathe-
matics, λ1, . . . , λm can be interpreted as proportions or probabilities. For example,
if m particles with masses α1, . . . , αm are located at points x1, . . . , xm of ℝ3,
the center of gravity of the system is the point x 5 λ1x1 1 ? 1 λmxm, where
λi 5 αi/(α1 1 ? 1 αm). In this convex combination, λi is the proportion of the
weight, which is at xi.

Definition 1.3. The intersection of all convex sets containing M is called the con-
vex hull of a set M and denoted by conv M.
We show that conv M is the smallest convex set containing M, characterizing it
as the set of all convex combinations of points of M. Using this characterization,
we prove two combinatorial results, one due to Carathéodory, the other to
Gauss
Lucas on the location of the roots of the derivative of complex polynomial.
According to Lemma 1.1, conv M is a convex set. Since MDconv M, by
Lemma 1.4 it contains all the convex combinations of elements x1, . . . , xm:
x 5 λ1 x1 1 ? 1 λm xm ; xi AM; i 5 1; . . . ; m
λi $ 0; i 5 1; . . . ; m; λ1 1 ? 1 λm 5 1:

Moreover, it is not hard to see that the set of such points itself is convex. Thus,
on the one hand, conv M contains all the convex combinations (Eq. 1.1) of ele-
ments x1, . . . , xm; on the other hand, by Definition 1.3, conv MDM.
Thus, we have obtained the following result.
6 Approximation and Optimization of Discrete and Differential Inclusions

Lemma 1.5. Let M be a set in X 5 ℝn. Then conv M is the set of all convex combi-
nations of points of M.
Carathéodory’s theorem, which is proved below, states that each point of the
convex hull conv M can be expressed as a convex combination of n 1 1 or fewer
points of the set. Thus, a point in the convex hull of a set in ℝ3 is either a point of
the set or belongs to a line segment, a triangle, or a tetrahedron with vertices in
the set.

Theorem 1.1. (Carathéodory) Let M be any set in X 5 ℝn and x be any point in

conv M. Then x can be expressed as a convex combination of n 1 1 or fewer points
of M; i.e., for every xAconv M, there are points x1, . . . , xrAM, such that
x 5 λ1 x1 1 ? 1 λr xr ;
ð1:2Þ
λ1 1 ? 1 λr 5 1; λ1 $ 0; . . . ; λr $ 0;

where r # n 1 1.
& In fact, we must show that r # n 1 1. Take a point of the form of Eq. (1.1) and
show that the number of nonzero terms in Eq. (1.1) can be reduced if m . n 1 1. It
is sufficient to assume that λi . 0. Consider the (n 1 1)-dimensional vectors (xi,1),
i 5 1, . . . , m. Since by assumption, the number of such vectors is m . n 1 1, there
is a linear dependence. Hence, there exist scalars αi, i 5 1, . . . , m, not all zero such
that
X
m
αi xi 5 0 ð1:3Þ
i51

and
X
m
αi 5 0: ð1:4Þ
i51

Of course, by Eq. (1.4) there are necessarily positive αi. Let us denote
 
λi
ε0 5 min : αi . 0; i 5 1; . . . ; m :
αi

Suppose the minimum is attained at i 5 i0. Then it is clear that for any αi,
i 5 1, . . . , m
λi 5 λi 2 ε0 αi $ 0:

Obviously by (1.3) and (1.4),

!
X
m X
m X
m
λi 5 λi xi 2 ε0 αi xi 5 x
i51 i51 i51 !
Xm Xm Xm
λi 5 λi 2 ε0 αi 5 1:
i51 i51 i51
Convex Sets and Functions 7

By the choice of ε0, the new coefficients λi are nonnegative, and at least one of
them is 0. We therefore have an expression of x as a nonnegative linear combina-
tion of fewer than m elements. This reduction is possible until m . n 1 1.’

Theorem 1.2. (Gauss
Lucas) The roots of the derivative of a nonconstant complex

polynomial belong to the convex hull of the set of roots of the polynomial itself.
& Let P be the complex polynomial defined by the equation
PðzÞ 5 a0 zn 1 ? 1 an 21 z 1 an ;

where n $ 1 and a0, . . . , an are complex numbers with a0 6¼ 0. Then

PðzÞ 5 a0 ðz 2 z0 Þ?ðz 2 zn Þ;

where z1, . . . , zn are the roots of P, each being repeated according to its multiplic-
ity. A routine verification shows that for z 6¼ z1, . . . , zn,
P0 ðzÞ 1 1 z 2 z1 z 2 zn
5 1? 5 1?1 ; ð1:5Þ
PðzÞ z 2 z1 z 2 zn jz 2 z1 j 2
jz 2 zn j2

where z is the conjugate of z. Now, let z be a root of P0 . We establish the theorem

by showing z as a convex combination of z1, . . . , zn. This can be done trivially if z
is one of z1, . . . , zn, so we assume that this is not the case. Setting P0 (z) 5 0 in
Eq. (1.5), we find that
1 1
z1 1 ? 1 zn
jz 2 z1 j2 jz 2 zn j2
z5 ;
1 1
1 ? 1
jz 2 z1 j2 jz 2 zn j2

which expresses z as a convex combination of z1, . . . , zn. Here we have used the
fact that if P0 (z) 5 0, then P0 ðzÞ 5 0:’
Note that the open unit ball B in ℝn centered at the origin and radius 1 is the set
of all points of ℝn whose distance from the origin is less than 1; i.e.,

B 5 fx : kxk , 1g; kxk 5 hx; xi1=2 :

Definition 1.4. A point x of a set M in ℝn is said to be an interior point of M, if

there exists a number ε . 0 such that x 1 εBDM.
The set of interior points of M is called the interior of M and denoted by int M.
Clearly, int M1Dint M2 when M1DM2. A set in ℝn, each of whose points is an
interior point of the set, is said to be open. Since int MDM is always true, M is
open if and only if int M 5 M. Clearly, the sets [ and ℝn are open.

Definition 1.5. A point x is said to be a limit point of M if there exists a sequence

xkAM that converges to x. The set of all limit points of M is called the closure of
M and is denoted by M:
8 Approximation and Optimization of Discrete and Differential Inclusions

Lemma 1.6. The closure and interior of a convex set M are convex sets.
& If M is convex, then x1Aint M and x2Aint M imply that x1 1 ε1BDM,
x2 1 ε2BDM. Let λ1xl 1 λ2x2 be the convex combination of points x1 and x2. Thus,

λ1 ðx1 1 ε1 BÞ 1 λ2 ðx2 1 ε2 BÞ 5 λ1 x1 1 λ2 x2 1 ðλ1 ε1 1 λ2 ε2 ÞBDM;

i.e., λ1x1 1 λ2x2 is a point of int M.

Now, if x1 ; x2 AM; then by definition of a closure set there exist sequences
x1k, x2kAM, such that x1k-x1, x2k-x2. Let λ1x1k 1 λ2x2k be the convex combina-
tion of points x1k, x2k. Therefore, since λ1x1k 1 λ2x2kAM, we have λ1 x1 1 λ2 x2 5
limk-N ðλ1 x1k 1 λ1 x2k ÞAM:’

Definition 1.6. Let M be a convex set in ℝn and x0 be any point of M. Then the
intersection of all possible underlying (carrier) subspaces containing M 2 x0
is called the linear hull for a set M and denoted by Lin M. In other words, Lin
M is the smallest subspace of ℝn that contains M. The affine hull Aff M of a set M
is the intersection of all affine sets (synonyms used by other authors for affine set
include affine manifold, affine variety, linear variety, flat) in ℝn containing M.
Thus, we may refer to Aff M as the smallest affine set containing M. Then we
can define the affine hull as Aff M 5 x0 1 Lin M.

Definition 1.7. The relative interior of a convex set M in ℝn, which we denote by
ri M, is defined as the interior that results when M is regarded as a subset of its
affine hull Aff M. That is, ri M consists of the points x1AAff M for which there
exists an ε . 0, such that x2AM whenever x2AAff M and kx1 2 x2k # ε; i.e.,

ri M 5 fxAaff M : 'ε . 0; ðx 1 εBÞ \ ðaff MÞCMg;

where B is a closed unit ball. Observe that ri MCMCM: Naturally, M is said to be

relatively open if ri M 5 M.

Definition 1.8. The dimension of a convex set M is the dimension of Lin M and is
denoted by dim M.
A set of m 1 1 points x0, x1, . . . , xm is said to be affinely independent if Aff {x0,
x1, . . . , xm} is m-dimensional. If {x0, x1, . . . , xm} is affinely independent, its con-
vex hull is called an m-dimensional simplex, and x0, x1, . . . , xm are called the verti-
ces of the simplex.
For an n-dimensional convex set, Aff M 5 ℝn by definition, so ri M 5 int M.
It should be pointed out that although the inclusion M1CM2 implies M1 CM2
and int M1Cint M2, it does not generally imply ri M1Cri M2. For example, if M2
is a square in ℝ2 and M1 is one of the sides of M2, then ri M1 and ri M2 are both
nonempty but disjoint.
Convex Sets and Functions 9

Lemma 1.7. ri M 5 ri M:
& Observe that Lin M is a closed set containing M; i.e., Lin M+M: On the other
hand, Lin M 5 Lin M: It is also clear that ri MDri M: Let us show the reverse
inclusion. Let xAri M and suppose e1, . . . , er is a basis for Lin M. Then, for small ε
we have
0 1
1
yk 5 x 1 ε @ek 2 eAA M; k 5 1; . . . ; r;
r11
ε
y0 5 x 2 e A M;
r11

where e 5 e1 1 ? 1 er. The vectors yk 2 y0 5 εek are linearly independent and

1 1
x5 y0 1 ? 1 yr :
r11 r11
This equality means that x is an interior point of an r-dimensional simplex with
vertices y0, . . . , yr. Taking points yk AM sufficiently close to yk, we see that x is an
interior point of a simplex with vertices yk : It follows that x is also an interior point
of M. Thus, ri M+ri M:’

Definition 1.9. The intersection of all closed convex sets containing M is called the
closed convex hull of a set M and denoted by conv M:

Lemma 1.8. conv M 5 conv M:

& Since conv M is the intersection of all convex sets containing M, we have the
inclusion conv M + conv M: It follows that conv M + conv M: Conversely,
conv M is a closed convex set. Hence, conv M D conv M: This ends the proof of
lemma.’

Theorem 1.3. The convex hull of a compact set is itself compact.

& Remember that a set in ℝn is compact if and only if it is both closed and
bounded (or, equivalently, if each sequence of its points contains some subse-
quence that converges to a point of the given set).
If M is compact and xAconv M, then by Theorem 1.1 we can write:
X
n11 X
n11
x5 λi xi ; xi AM; λi $ 0; λi 5 1:
i51 i51

Therefore,
X
n11
kxk # λi kxi k # c;
i51
10 Approximation and Optimization of Discrete and Differential Inclusions

where c is a constant such that kxk # c for all xAM. Thus, conv M is a bounded
set. We show that M is a closed set. Let
X
n11 X
n11
xk 5 λik xik ; xik AM; λik 5 1: ð1:6Þ
i51 i51

Since the sequences λik and xik are bounded, they contain convergent subsequences.
Without loss of generality, assume that λik-λi0 and xik-xi0AM (remember that M
is compact). Then by passing to the limit in Eq. (1.6), we have
X
n11 X
n11
x0 5 λi0 xi0 ; xi0 AM; λi0 5 1:
i51 i51

It follows that x0Aconv M, so conv M is closed.’

The following property of closures and relative interiors of convex sets is
fundamental.

Theorem 1.4. Let M be a convex set in ℝn. If x1 AM; x1 Ari M; then (1 2 λ)

x1 1 λx2Ari M for all λ in (0,1). In addition, M 5 ri M:
& If x2Ari M, then x2 1 Lin M\(εB)DM. Hence, it follows that
ð1 2 λÞx1 1 λðx2 1 Lin MðεBÞÞ 5 ð1 2 λÞx1 1 λx2 1 Lin MðλεBÞDM:

That is, since ri M 5 ri M by Lemma 1.7, we have (1 2 λ)x1 1 λx2Ari M. Now, let
x0 AM; xkAM, xk-x0, λk-0. Thus, (1 2 λk)xk 1 λkyAri M if yAri M.
Therefore, x0 is a limit point for ri M. Consequently, MDri M: The reverse
inclusion is trivial.’

Lemma 1.9. Let M be a convex set and x0 AM; but x0 2

= M. Then in any neighbor-
hood of x0 there are points not contained in M:
& Take a point yAri M. Then the points of the ray y 1 λ(x0 2 y), λ . 1, do not
belong to M: We argue by contradiction. Suppose that if λ . 1, then
x1 5 y 1 λðx0 2 yÞAM: Thus, by Theorem 1.4, we deduce that
 
1 1
x0 5 x1 1 1 2 yAri M:
λ λ

But by hypothesis, x0 2= M. This contradiction proves the lemma.’

The results on the separation of convex sets that we establish below are among
the most important ones in convexity. They are based on the fact that a hyperplane
in ℝn divides ℝn into two, in the sense that the complement of the hyperplane is
the union of two disjoint open convex sets—the open half-spaces associated
with the hyperplane. A hyperplane H is said to separate the sets M1 and M2 if M1
lies in one of the closed half-spaces determined by H, and M2 lies in other. Some
kinds (properly, strongly) of separation the reader can find, for example, in Refs.
[111, 226]. Obviously, if a hyperplane separates two sets, then it also separates
their convex hulls; for this reason, we consider only the separation of convex sets.
Convex Sets and Functions 11

Theorem 1.5. Let M be a nonempty convex set and assume that x0 2

= M: Then there
are x and ε . 0, such that

hx; x i # hx0 ; x i 2 ε ’xAM:

& Let y be the point of M closest to x0 (since M is closed, a point y exists) such
that

kx 2 x0 k $ ky 2 x0 k; ’xAM:

The convexity of M implies that λx 1 ð1 2 λÞy 5 y 1 λðx 2 yÞAM for all λA[0,1].
Thus,

kλx 1 ð1 2 λÞy 2 x0 k2 5 hy 2 x0 1 λðx 2 yÞ; y 2 x0 1 λðx 2 yÞi

5 ky 2 x0 k2 1 2λhx 2 y; y 2 x0 i 1 λ2 kx 2 yk2 $ ky 2 x0 k2 :

As a result of a simple transformation we obtain

2hx 2 y; y 2 x0 i 1 λkx 2 yk2 $ 0; λA½0; 1:

In particular, if λ 5 0 it follows that

hx 2 y; y 2 x0 i $ 0: ð1:7Þ

Let x 5 x0 2 y, ε 5 kx k2. Here, y 6¼ x0, because x0 2

= M by hypothesis.
This implies that x 6¼ 0 and so ε . 0. Now, the inequality in Eq. (1.7) can be
rewritten as follows:

hx; x i # hy; x i 5 hx0 ; x i 2 hx ; x i 5 hx0 ; x i 2 ε:

for arbitrary xAM.’

Theorem 1.6. Let M be a convex set not containing x0. Then, there is a point
x 6¼ 0 such that

hx; x i # hx0 ; x i ’xAM:

& The assertion follows from the previous theorem if x0 2 = M: So assume that
x0 2
= M: Then, by Lemma 1.9, there exists a sequence xk-x0, xk 2
= M: Then, accord-
ing to Theorem 1.5,

hx; xk i # hxk ; xk i 2 εk ; εk . 0:

12 Approximation and Optimization of Discrete and Differential Inclusions

Since εk . 0, it follows that xk 6¼ 0: Taking into account that εk is arbitrarily small
and xk 6¼ 0; we can write
   
x x
x; k # xk ; k : ð1:8Þ
kxk k kxk k

The sequence xk =kxk k is bounded and so contains a convergent subsequence.

Without loss of generality, suppose that
xk
-x0 ; kx0 k 5 1: ð1:9Þ
kxk k

Thus, by passing to the limit in Eq. (1.8), we have hx; x0 i # hx0 ; x0 i:’

Theorem 1.7. If M1 and M2 are nonempty disjoint convex sets, then there exists
x 6¼ 0 such that
hx1 ; x i # hx2 ; x i

for all x1AM1, x2AM2.

& Consider the convex set M 5 M1 2 M2. Since M1 and M2 are nonempty disjoint
convex sets, the point x0 5 0 does not belong to M. Then by the preceding theorem
we obtain the existence of x 6¼ 0 such that
hx; x i # h0; x i ð1:10Þ

for all xAM; i.e., for x 5 x1 2 x2, x1AM1, x2AM2. Then by substituting x 5 x1 2 x2
into Eq. (1.10), we obtain
hx1 2 x2 ; x i # 0; x1 AM1 ; x2 AM2 :’

Theorem 1.8. Let M1 and M2 be nonempty disjoint closed convex sets, one of
which is compact. Then there exist x 6¼ 0 and ε . 0 such that

hx1 ; x i # hx2 ; x i 2 ε ’x1 AM1 ; x2 AM2 :

& Observe that, as in the proof of the preceding theorem, the point x0 5 0 does not
belong to M 5 M1 2 M2. But since M1 and M2 are closed and one of them is com-
pact, it is not hard to see that M is a closed set. Thus, 0 2
= M. Applying the preced-
ing theorem ends the proof of this theorem.’

Definition 1.10. Let M be a convex set and 0Aint M. Then a function

n x o
rM ðxÞ 5 inf α : α . 0; AM
α

defined on X 5 ℝn is called a Minkowski (or gauge) function.

Convex Sets and Functions 13

Since 0Aint M, α21xAM for any x, when α is sufficiently large and rM(x) is
always a finite number. Besides, by definition, rM(x) $ 0.

Lemma 1.10. Let M be a convex set and 0Aint M. Then:

a. rM(λx) 5 λrM(x) if λ $ 0;
b. rM(x) # 1, if xAM and rM(x) $ 1, if x 2
= M;
c. rM(x 1 y) # rM (x) 1 rM(y).
& a. Let λα1 5 α. We have

8 9
< λx =
rM ðλxÞ 5 inf α : α . 0; AM 5 inf fλα1 : α1 . 0; α21
1 xAMg
α : α ; α1

5 λ inf fα1 : α1 . 0; α21

1 xAMg 5 λrM ðxÞ:
α1

b. If xAM, then x/1AM, so that rM(x) # 1. Now, let x 2

= M. On the contrary, if rM(x) , 1,
then there is α , 1 such that α21xAM. Because 0AM and M is convex, we can write

x 5 ð1 2 αÞ0 1 αðα21 xÞAM;

which contradicts the fact that x 2

= M. Thus, rM(x) $ 1 for all x 2
= M.
c. Let γ . rM(x) 1 rM(y). Then there are α and β so that γ 5 α 1 β, α . rM(x),
β . rM(y). The inequality α . rM(x) implies that α21xAM. Indeed, by definition of
rM(x), there is a number α1 such that α . α1 . rM(x) and α21 1 xAM: Because M is con-
vex and 0AM, it follows that α21 x 5 ð1 2 α21 α1 Þ0 1 α21 α1 ðα21 1 xÞAM: Similarly, we
can prove that β 21yAM. Thus, we have
x1y x1y α β
5 5 ðα21 xÞ 1 ðβ 21 yÞAM:
γ α1β α1β α1β

Consequently, rM(x 1 y) # γ. But since γ is an arbitrary number satisfying

γ . rM(x) 1 rM (y), finally we obtain
rM ðx 1 yÞ # rM ðxÞ 1 rM ðyÞ:

We now associate with each nonempty convex set a convex function known as its
support function.’

Definition 1.11. The support function HM of a nonempty set M in ℝn is defined by:

HM ðx Þ 5 supfhx; x i : xAMg:
x

Theorem 1.9. Let M be a closed convex set. Then xAM if and only if
hx; x i # HM ðx Þ ð1:11Þ

for all x .
14 Approximation and Optimization of Discrete and Differential Inclusions

& If xAM, then according to the definition of support function, the inequality in
Eq. (1.11) is satisfied.
Conversely, let Eq. (1.11) be satisfied for x0 and x0 2
= M. Then by Theorem 1.5,
there exist x and ε . 0 such that

hx; x i # hx0 ; x i 2 ε

for all xAM. Thus, the least upper bound (supremum) of the left-hand side of this
inequality is HM(x ); i.e.,

HM ðx Þ # hx0 ; x i 2 ε;

which contradicts Eq. (1.11).’

Thus, a convex set can be characterized by both Minkowski’s function and sup-
port functions.
By way of an example, we find the support function HM of the set M defined as

M 5 fx 5 ðx1 ; . . . ; xn ÞAℝn : jx1 j 1 ? 1 jxn j # 1g:

Let x 5 ðx1 ; . . . ; xn Þ. Then

HM ðx Þ 5 sup hx ; xi 5 supfx1 x1 1 ? 1 xn xn : jx1 j 1 ? 1 jxn j # 1g

xAM x
# supfjx1 jjx1 j 1 ? 1 jxn jjxn j : jx1 j 1 ? 1 jxn j # 1g
x
# supf½maxðjx1 j; . . . ; jxn jÞðjx1 j 1 ? 1 jxn jÞ : jx1 j 1 ? 1 jxn j # 1g
x
5 maxfjx1 j; . . . ; jxn jg:

Let kA{1, . . . , n} be such that jxk j 5 maxfjx1 j; . . . ; jxn jg: Define a point x 5
(x1, . . . , xn) of M by the conditions xi 5 0 when i 6¼ k and xk is 1 or 21 according
to whether xk is nonnegative or negative. Then

hx ; xi 5 jxk j 5 maxfjx1 j; . . . ; jxn jg;

and, therefore, HM ðx Þ $ maxfjx1 j; . . . ; jxn jg: We have thus shown that

HM ðx Þ 5 maxfjx1 j; . . . ; jxn jg:

1.3 Convex Cones and Dual Cones

A convex cone is one of the important concepts in the theory of extremal problems.
To investigate its properties, first we must calculate the dual cone.
Convex Sets and Functions 15

Definition 1.12. A convex set K is said to be a cone if xAK and λ . 0 imply that
λxAK.

Lemma 1.11. If x1, . . . , xmAK, λ1 . 0, . . . , λm . 0, then

λ1 x1 1 λ2 x2 1 ? 1 λm xm AK:

& The proof of the lemma follows immediately from the formula:
 
λ1 λm
λ1 x1 1 ? 1 λm xm 5 λ x1 1 ? 1 xm ;
λ λ

where λ 5 λ1 1 ? 1 λm, which is Lemma 1.4, and the definition of a cone.’

Thus, a convex cone (or, briefly, a cone) contains any linear combination with posi-
tive coefficients of its elements.
We assume, unless otherwise stated, that the cones being considered are convex.

Definition 1.13. The set of vectors x AX 5 ℝn for which hx,x i $ 0 holds for all
xAK is called the dual cone to the cone K and is denoted by K . Briefly, it can be
written as follows:

K  5 fx AX  : hx; x i $ 0 ’xAKg:

Obviously, K is a convex cone.

Some important properties of cones follow.

Lemma 1.12. K is a closed cone.

& If xk AK  ; then hxk ; xi $ 0 for all xAK. By passing to the limit in this inequality
as xk -x0 ; k-N, we have

hx; x0 i $ 0 ’xAK;

which means that x0 AK  :’

Lemma 1.13. K and K have the same dual cones; i.e., K  5 ðK Þ :

& If x AðK Þ ; then hx,x i $ 0 for all xAK and so for all xAK. Hence, x AK .
Conversely, if x AK , then by continuity, hx,x i is nonnegative, because elements
of K are the limit points of K.’

Lemma 1.14. If K is a closed cone and hx,x i $ 0 for all x AK , then xAK.
& We argue by contradiction. Suppose x0 2 = K, but hx0,x i $ 0 for all x AK . Since

x0 2
= K, by Theorem 1.8, there exist x and ε . 0 such that

hx0 ; x  i # hx; x  i 2 ε ð1:12Þ

16 Approximation and Optimization of Discrete and Differential Inclusions

for all xAK. We show that

inf hx; x  i 5 0 ð1:13Þ
xAK

Indeed, since by definition xAK implies λxAK, λ . 0, any cone K contains ele-
ments sufficiently close to zero. In particular, a closed cone always contains the
zero point, so the infimum of hx; x i cannot be greater than zero. On the other
hand, according to Eq. (1.12), the inner product hx; x i is bounded below. We prove
that hx; x i $ 0: On the contrary, suppose that there exists a point x1AK such that
hx; x i , 0: Then denoting x 5 λx1 and by passing to the limit in hx; x i; as
λ- 1N, we observe that hx; x i-2N: This contradiction proves that hx; x i $ 0:
It follows that the formula in Eq. (1.13) is true. Thus, x AK  : Setting x 5 0 in the
inequality in Eq. (1.12), we have hx; x i , 2ε; which contradicts that hx0,x i $ 0
for all x AK . Consequently, x0AK.’
Naturally, later on we denote K 5 (K ) .

Lemma 1.15. If K is a closed cone, then K 5 K.

& According to Definition 1.13, we can write
K  5 fx : hx; x i $ 0; x AK  g:

And if xAK, then hx,x i $ 0 for x AK . Conversely, if K is closed, then by the pre-
vious lemma, from the inequality, hx,x i $ 0 for all x AK , it follows that xAK.’
Note that in the more general case for any convex cone, K  5 K : This follows
from Lemma 1.13 because K  5 ðK Þ and so K  5 ðK Þ 5 K :

Lemma 1.16. Let K1 and K2 be convex cones. Then K1 1 K2 is also a convex cone
and
ðK1 1 K2 Þ 5 K1 \ K2 :

& Note that the convexity of K1 1 K2 follows immediately from Lemma 1.2. On
the other hand, it is easy to see that if K1 1 K2 contains a point x 5 x1 1 x2, x1AK1,
x2AK2, then it also contains an element λx 5 λx1 1 λx2, λ . 0. Moreover, x A
(K1 1 K2) if and only if

hx1 1 x2 ; x i $ 0; x1 AK1 ; x2 AK2 :

Since x1 and x2 change independently and one of them can tend to zero, if neces-
sary, the latter inequality is equivalent to the inequalities
hx1 ; x i $ 0; x1 AK1 ;
hx2 ; x i $ 0; x2 AK2 ;

i.e., x AK1 and x AK2 : Thus, it is proved that x A(K1 1 K2) if and only if
x AK1 \ K2 :’
Convex Sets and Functions 17

Lemma 1.17. If the cones K1 and K2 are closed, then

ðK1 \ K2 Þ 5 K1 1 K2 : ð1:14Þ

& In fact, by using the previous results, we can write

ðK1 \ K2 Þ 5 ðK1 \ K2 Þ 5 ððK1 1 K2 Þ Þ 5 ðK1 1 K2 Þ 5 K1 1 K2 :’

Remark 1.1. We mention that a bar denotes the closure of a set. It should be
pointed out that, in general, the formula in Eq. (1.14) is not true without the clo-
sure: the left-hand side of Eq. (1.14) is always closed, but the right-hand side may
or may not be closed.

Lemma 1.18. If hx,x i is bounded below for all xAK, then x AK . Moreover, if
xAint K, then
hx; x i . 0

for all x AK , x 6¼ 0.

& In fact, the first assertion is proved in Lemma 1.14. We show the second asser-
tion. If xAint K, then there is an ε . 0 such that x 1 εBDK, where B is the
Euclidean unit ball. Hence,

hx 1 εz; x i $ 0

for all x AK and zAB. Then it follows that

  
x
hx; x i $ ε subh2z; x i $ ε ; x 
5 εkx k . 0;
zAB kx k

i.e., hx,x i . 0 for all x AK , x 6¼ 0.’

Theorem 1.10. Let K1, . . . , Km be convex cones. In order that their intersection be
empty, it is necessary and sufficient that there exist xi AKi ; i 5 1; . . . ; m not all
zero, such that

x1 1 ? 1 xm 5 0:

& Consider the Cartesian product Xm 5 X 3 X 3 ? 3 X 5 ℝmn. The elements of

Xm have the form (x1, x2, . . . , xm), where xiAX(i 5 1, . . . , m). Obviously, the scalar
product of vectors (x1, . . . , xm) and ðx1 ; . . . ; xm Þ can be written in the form of the
sum:

hx1 ; x1 i 1 hx2 ; x2 i 1 ? 1 hxm ; xm i:

18 Approximation and Optimization of Discrete and Differential Inclusions

Consider two cones in Xm of the form

K 5 K1 3 ? 3 Km 5 fðx1 ; . . . ; xm Þ : x1 AK1 ; . . . ; xm AKm g;

P 5 fðx; . . . ; xÞ : xAXg:

Since \mi 5 1 Ki 5 [; it is not hard to see that the cones K and P are disjoint, i.e.,
K\P 5 [. Applying Theorem 1.6, we determine that there is a vector ðx1 ; . . . ; xm Þ
such that

hx; x1 i 1 ? 1 hx; xm i # hx1 ; x1 i 1 ? 1 hxm ; xm i ð1:15Þ

for all xAX, x1AK1, . . . , xmAKm.

The inequality in Eq. (1.15) implies that hxi ; xi i is bounded on Ki, i 5 1, 2, . . . , m.
By Lemma 1.18, it follows that

xi AKi ; i 5 1; . . . ; m ð1:16Þ

By Eq. (1.15), hx; x1 1 ? 1 xm i is bounded above for all xAX. Since a nonzero lin-
ear function takes infinitely large values, the boundedness of hx; x1 1 ? 1 xm i
implies that

x1 1 ? 1 xm 5 0: ð1:17Þ

Now taking into account Eqs. (1.16) and (1.17), we have the desired result.’

Theorem 1.11. Let K1, K2, . . . , Km be convex cones. If K 5 K1\K2\?\Km and

K1\int K2\?\int Km 6¼ [, then

K  5 K1 1 ? 1 Km :

& The validity of the inclusion

K  +K1 1 ? 1 Km

is verified immediately, starting from the definition of a dual cone. We prove the
inverse inclusion. Let x AK , x 6¼ 0 denote

K0 5 fx : hx; x i , 0g:

We show that the cones K0 and K are disjoint. On the contrary, suppose that there
exists an element x1AK0\K 6¼ [. Since x1AK0, we have:

hx1 ; x i , 0:
Convex Sets and Functions 19

On the other hand, since x1AK, x AK , by definition of dual cone,

hx1 ; x i $ 0:

Therefore, the last two inequalities contradict each other, so K0\K 6¼ [. Consider
the dual cone K0 : Suppose that hx,x i , 0 implies that hx,y i $ 0; i.e., y AK0 ,
y 6¼ 0. Then x and y are linearly dependent vectors. This means that there are
numbers α1 and α2 (at least one of them is nonzero) such that

α1 x 2 α2 y 5 0:

Because x 6¼ 0 and y 6¼ 0, we have α2 6¼ 0 and

y 5 λx ; λ 5 α1 =α2 :

Moreover,
0 # hx; y i 5 λhx; x i

for all xAK0. Since hx,x i , 0, it follows from the last inequality that λ , 0. If
y 5 0, we put y 5 0  x . Thus, we find that

K0 5 fy : y 5 λx ; λ # 0g:

Because the cones K0 and K are disjoint (i.e., K0, K1, . . . , Km are disjoint), accord-
ing to preceding theorems, there exist vectors y AK0 ; xi AKi ; i 5 1; . . . ; m such
that

y 1 x1 1 ? 1 xm 5 0; ð1:18Þ

where not all terms are zero. By using y 5 λx and λ # 0, we rewrite Eq. (1.18) in
the form

2λx 5 x1 1 ? 1 xm : ð1:19Þ

Observe that if λ , 0, then

     
1 1 1
x 5 2 x1 1 2 x2 1 ? 1 2 xm AK1 1 ? 1 Km :
λ λ λ

Here, note that xi AKi ; i 5 1; . . . ; m:

Next, we show that λ 6¼ 0. On the contrary, suppose that λ 5 0. Then, it follows
from Eq. (1.19) that

x1 1 ? 1 xm 5 0 ð1:20Þ

20 Approximation and Optimization of Discrete and Differential Inclusions

Here, not all xi are zero and so at least two of them, say x1 and x2 ; are nonzero. By
the hypotheses of the theorem, there exists a point x0AK1\int K2?\int Km so that
by Lemma 1.18

hx0 ; x2 i . 0

and hx0 ; xi i $ 0 for i 6¼ 2. Then taking into account Eq. (1.20), we establish the fol-
lowing contradiction:

0 5 hx0 ; x1 1 x2 1 ? 1 xm i 5 hx0 ; x1 i 1 ? 1 hx0 ; xm i . 0;

i.e., 0 . 0.’

Theorem 1.12. Let K1, . . . , Km be convex cones and K 5 K1\?\Km. Then either

K  5 K1 1 ? 1 Km ð1:21Þ

or there exist xi AKi ; not all zero, such that

x1 1 ? 1 xm 5 0:

& If for all x AK it occurs that λ , 0 in Eq. (1.19), then x can be represented as
a sum of xi AKi : Clearly, this representation implies Eq. (1.21). If λ 5 0 for some
x , then the formula in Eq. (1.20) is satisfied.’
As is shown in Theorem 1.11, if the intersection of cones is empty, then the
equality in Eq. (1.20) is true.

Definition 1.14. The cones K1, . . . , Km are called separable if there exist vectors
xi AKi ; not all zero, such that

x1 1 ? 1 xm 5 0:

Definition 1.15. Let M be an arbitrary nonempty set in ℝn. The cone defined by

cone M 5 fx : x 5 λx1 ; x1 AM; λ . 0g

is called the cone generated by M.

Lemma 1.19. If M is a convex set, then cone M is a convex cone.

& Take x1Acone M and x2Acone M. Then

x1 5 λx1 ; x2 5 λx2 ; x1 ; x2 AM; λ1 ; λ2 . 0:

Convex Sets and Functions 21

Let α1 . 0 and α2 . 0 be arbitrary numbers. Then we can write:

α1 x1 1 α2 x2 5 α1 λ 1 x1 1 α2 λ 2 x2

α1 λ 1 α2 λ 2
5 ðα1 λ1 1 α2 λ2 Þ x1 1 x2 Acone M:
α1 λ1 1 α2 λ2 α1 λ1 1 α2 λ2

Here, we have used the fact that M is convex, and so


α1 λ 1 α2 λ 2
x1 1 x2 AM:’
α1 λ1 1 α2 λ2 α1 λ1 1 α2 λ2

Lemma 1.20. Let M be a convex set. Then x A(cone M) if and only if hx,x i $ 0
for all xAM.
& In fact, x A(cone M) if and only if

hλx; x i $ 0

for all λ $ 0 and all xAM. Since λ $ 0, it follows that hx,x i $ 0 for all xAM.’

Definition 1.16. Let M be a nonempty convex set in ℝn. The recession cone of M
is denoted by 01 M and is defined as

0 1 M 5 fx 6¼ 0 : x 1 λxAM; ’xAM; ’λ $ 0g [ f0g:

Observe that if M is bounded, it certainly contains no half-lines, so that

01 M 5 {0}.
As examples of recession cones in ℝ2, for
8 9
< 1=
M1 5 ðx1 ; x2 Þ : x1 . 0; x2 $ ;
: x1 ;
M2 5 fðx1 ; x2 Þ : x1 . 0; x2 . 0g [ fð0; 0Þg;

one has

0 1 M1 5 fðx1 ; x2 Þ : x1 $ 0; x2 $ 0g;
0 1 M2 5 fðx1 ; x2 Þ : x1 . 0; x2 . 0g [ fð0; 0Þg 5 M2 :

Moreover, if

M 5 fx : hx; xk i # β k ; kAIg;

22 Approximation and Optimization of Discrete and Differential Inclusions

where xk Aℝn ; β k Aℝ1 and I is an arbitrary index set, then

0 1 M 5 fx : hx; xk i # 0; kAIg:

Definition 1.17. A polyhedral convex set in ℝn is a set that can be expressed as the
intersection of some finite family of closed half-spaces, that is, as the set of solu-
tions to some finite system of inequalities of the form

hx; xk i # β k ; k 5 1; . . . ; l: ð1:22Þ

The definition of a polyhedral set immediately makes it clear why such sets play a
leading role in linear programming.
In particular, if the finite system of inequalities in Eq. (1.22) is homogeneous
(β k 5 0, k 5 1, . . . , l); i.e.,

hx; xk i # 0; k 5 1; . . . ; l; ð1:23Þ

then the set of solutions to this finite system of inequalities is called a polyhedral
convex cone. A bounded polyhedral convex set is a polytope (polyhedron), which
is the convex hull of finitely many points. Clearly, a polyhedral set is closed.

Theorem 1.13. (Farkas) If K is a polyhedral cone, then the dual cone K is polyhe-
dral and consists of the elements of the form

X
l
x 5 γ k xk ; γ k $ 0:
k51

& Consider a polyhedral cone

( )
X
l
K~ 5 x : x 5  
γ k xk ; γk $ 0 :
k51

By the definition of a dual cone, xAðK~ Þ ; if

* +
X
l
x; γ k xk $ 0; γ k $ 0; k 5 1; . . . ; l:
k51

It follows that

hx; xk i $ 0; k 5 1; . . . ; l;
Convex Sets and Functions 23

i.e., xAK. Thus, K 5 ðK~ Þ : Consequently, since K is closed as a polyhedral cone,

we have

K  5 ðK~ Þ 5 K~ :’

Lemma 1.21. The sum of polyhedral cones is a polyhedral cone.

& Let K1 and K2 be polyhedral cones. Then there exists a collection of vectors xi1,
i 5 1, . . . , m and xj2, j 5 1, . . . , l such that all x1AK1, x2AK2 can be represented as

X
m X
l
x1 5 λi xi1 ; λi $ 0; x2 5 γ j xj2 ; γ j $ 0:
i51 j51

If now xAK1 1 K2, then

X
m X
l
x 5 x1 1 x 2 5 λi xi1 1 γ j xj2 ; λi $ 0; γ j $ 0;
i51 j51

i.e., K1 1 K2 is a polyhedral cone.’

Lemma 1.22. If K1, . . . , Km are polyhedral cones, then

ðK1 \ ? \ Km Þ 5 K1 1 ? 1 Km :

& According to Lemma 1.17, we have

ðK1 \ ? \ Km Þ 5 K1 1 ? 1 Km :

Since Ki ; i 5 1; . . . ; m are polyhedral cones, by Lemma 1.21, K1 1 ? 1 Km is also
a polyhedral cone. Thus, this sum of cones is closed and the bar above it can be
removed.’

Theorem 1.14. A polyhedral set can be represented as a sum of a polytope and a

polyhedral cone. Conversely, the sum of any polytope and a polyhedral cone is a
polyhedral set.
& Let us introduce an additional coordinate x0 and consider a system of homoge-
neous inequalities

hx; xk i 2 x0 β k # 0; k 5 1; . . . ; l; x0 $ 0 ð1:24Þ

24 Approximation and Optimization of Discrete and Differential Inclusions

Clearly, the set of solutions to this system is a polyhedral cone in ℝn11, the ele-
ments of which can be represented as follows:
 0 X m  0
x x
5 λj j ; λj $ 0; ð1:25Þ
x j51
xj

0
where xx is an (n 1 1)-dimensional vector in ℝn11. Observe that λj $ 0 is arbi-
trary. In particular, taking λj 5 1, λi 5 0, i 6¼ j, the inequality in Eq. (1.25) implies
that x0j $ 0:
Let

I 0 5 f j : x0j 5 0; j 5 1; . . . ; mg

and

I1 5 fj : x0j . 0; j 5 1; . . . ; mg:

Denoting yj 5 xj =x0j ; γ j 5 λj x0j ; jAI1; we rewrite the inequality in Eq. (1.25) in the
forms:
X X
x0 5 λj x0j 5 γj ; γj $ 0 ð1:26Þ
jAI1 jAI1

and
!
X X xj X X
x5 λj xj 1 λj x0j 5 λj x j 1 γ j yj ; γj $ 0
jAI 0 jAI1
x0j jAI 0 jAI1

Thus, every solution of Eq. (1.24) can be represented in the form of Eq. (1.26).
Remember that setting x0 5 1, a solution to Eq. (1.22) is obtained from a solution
to Eq. (1.24). Consequently, every solution of Eq. (1.22) can be represented in the
form
X X
x5 λj xj 1 γ j yj ; λj $ 0 ð1:27Þ
jAI 0 jAI1

and
X
γ j 5 1; γ j $ 0; jAI1: ð1:28Þ
jAI1

Note that the first term on the right-hand side of Eq. (1.27) represents a point of
a polyhedral cone, while the second term is a point of some polytope. Similarly,
taking into account Eqs. (1.27) and (1.28) in reverse direction, it can be shown that
the sum of any polytope and polyhedral cone is a polyhedral set.’
Convex Sets and Functions 25

1.4 The Main Properties of Convex Functions

Let f be a function whose values are real or 6N and whose domain is a subset of
ℝn. The effective domain of a function f (dom f) is the set of points of ℝn on which
the values of f are real or 2N:
dom f 5 fx : f ðxÞ , 1Ng:

The epigraph of f is the set of pairs x0Aℝ1 and xAX 5 ℝn such that x0 $ f(x):
epi f 5 fðx0 ; xÞ : x0 $ f ðxÞg:

It should be noted that a point (x0,x)Aℝn11 belongs to epi f, if xAdom f, because in

the case f(x) 5 1N, there is no real number x0 such that x0 $ f(x). It is not hard to
see that the function f can be defined by epi f as follows:

f ðxÞ 5 inf fx0 : ðx0 ; xÞAepi f g: ð1:29Þ

x0

Thus, there is a close connection between subsets of ℝn11 and functions defined
on ℝn.

Definition 1.18. A function f is said to be convex if epi f is a convex set.

Definition 1.19. A function f is said to be proper if f(x) , 1N for at least one

x and f(x) . 2N for every x. A function that is not proper is improper.
Thus, f is proper if and only if dom f 6¼ [ and f(x) is finite for xAdom f.

Lemma 1.23. Let f be a proper function. Then f is convex if and only if

f ðλx1 1 λx2 Þ # λ1 f ðx1 Þ 1 λ2 f ðx2 Þ;

ð1:30Þ
λ1 $ 0; λ2 $ 0; λ1 1 λ2 5 1:

& If f is convex, then, by Definition 1.18, for all λ1, λ2 $ 0, λ1 1 λ2 5 1,

λ1 ðx01 ; x1 Þ 1 λ2 ðx02 ; x2 Þ 5 ðλ1 x01 1 λ2 x02 ; λ1 x1 1 λ2 x2 ÞAepi f

if ðx01 ; x1 ÞAepi f ; ðx02 ; x2 ÞAepi f or, equivalently,

f ðλ1 x1 1 λ2 x2 Þ # λ1 x01 1 λ2 x02 :

In particular, if x01 5 f ðx1 Þ; x02 5 f ðx2 Þ; then

f ðλ1 x1 1 λ2 x2 Þ # λ1 f ðx1 Þ 1 λ2 f ðx2 Þ;

λ1 1 λ2 5 1; λ1 ; λ2 $ 0
26 Approximation and Optimization of Discrete and Differential Inclusions

Conversely, it is not hard to see that if f is a proper function satisfying

Eq. (1.30), then f is convex.’

Lemma 1.24. If f is convex, then its dom f is a convex set.

& Let x1, x2Adom f. Then there exist x01 ; x02 such that f ðx1 Þ # x01 ; f ðx2 Þ # x02 : Thus,
ðx01 ; x1 ÞAepi f ; ðx02 ; x2 ÞAepi f ; and so

λ1 ðx01 ; x1 Þ 1 λ2 ðx02 ; x2 ÞAepi f ;

λ1 1 λ2 5 1; λ1 ; λ2 $ 0

or

f ðλ1 x1 1 λ2 x2 Þ # λ1 x01 1 λ2 x02 :

Consequently, f(λ1x1 1 λ2x2) is not equal to 1N, and so λ1x1 1 λ2x2Adom f.’
It follows from Lemma 1.24 that dom f is convex, even if f is an improper func-
tion. Consider such a function. Let yAdom f, for which f(y) 5 2N, but xAri dom f.
Then (x 2 y)ALin dom f and so x1 5 x 1 ε(x 2 y)Adom f for a sufficiently small
ε . 0. It is clear that

1 ε
x5 x1 1 y:
11ε 11ε

If y0 is any number and x01 $ f ðx1 Þ; then (y0,y)Aepi f [remember that f(y) 5 2N],
ðx01 ; x1 ÞAepi f ; and so by virtue of the convexity of epi f,
0 1
@ 1 ε 1 ε
x0 1 y0 ; x1 1 yAAepi f ;
11ε 1 11ε 11ε 11ε
0 1
1 ε A 1 0 ε 0
f ðxÞ 5 f @ x1 1 y # x 1 y :
11ε 11ε 11ε 1 11ε

In the latter inequality, by passing to the limit as y0- 2N, we have f(x) 5 2N.
Therefore, if f is an improper convex function, then f(x) 5 2N for every xAri dom f.
It follows that an improper convex function may take finite values only at points of
the relative boundary of dom f.

Lemma 1.25. Let fi, iAI be a collection of convex functions, where I is an arbitrary
index set. Then their pointwise supremum

f ðxÞ 5 sup fi ðxÞ

iAI

is a convex function.
Convex Sets and Functions 27

& It is not hard to see that

epi f 5 \ epi fi ;
iAI

which is a convex set by Lemma 1.1.’

Lemma 1.26. (Jensen’s inequality) Let f be a proper convex function. Then

f ðλ1 x1 1 ? 1 λm xm Þ # λ1 f ðx1 Þ 1 ? 1 λm f ðxm Þ;

λi $ 0; i 5 1; . . . ; m; λ1 1 ? 1 λm 5 1:

& Without loss of generality, let λi . 0, iAI. If xi 2 = dom f, then f(xi) 5 1N,
λi f(xi) 5 1N and the inequality is fulfilled trivially, because its right-hand side is
equal to 1N.
Now, let xiAdom f, i 5 1, . . . , m. Since epi f is a convex set, by Lemma 1.4, the
inclusion ( f(xi),xi)Aepi f implies that

ðλ1 f ðx1 Þ 1 ? 1 λm f ðxm Þ; λ1 x1 1 ? 1 λm xm ÞAepi f :

Thus,

f ðλ1 x1 1 ? 1 λm xm Þ # λ1 f ðx1 Þ 1 ? 1 λm f ðxm Þ:’

Lemma 1.27. The sum of proper convex functions fi, i 5 1, . . . , m with nonnega-
tive coefficients is convex.
& Since fi, i 5 1, . . . , m are proper convex functions, by Lemma 1.23,

fi ðλ1 x1 1 λ2 x2 Þ # λ1 fi ðx1 Þ 1 λ2 fi ðx2 Þ;

where λ1 $ 0, λ2 $ 0, and λ1 1 λ2 5 1. Thus, multiplying these inequalities by

αi $ 0 and then adding them, we have

X
m X
m X
m
αi fi ðλ1 x1 1 λ2 x2 Þ # λ1 αi fi ðx1 Þ 1 λ2 αi fi ðx2 Þ:’
i51 i51 i51

There are some useful correspondences between convex sets and convex func-
tions. The simplest associates with each set M in ℝn the indicator function δM (  )
of M, where

0; xAM;
δM ðxÞ 5
1N; x2
= M:

Clearly, M is convex if and only if δM (  ) is a convex function on ℝn.

28 Approximation and Optimization of Discrete and Differential Inclusions

Note that the resulting function of Lemma 1.27 may not be a proper function.
For example, for the disjoint sets M1 and M2, the sum of indicator functions
δM1 1 δM2 is identically 1N.
Now, we show that on the basis of Lemmas 1.25
1.27 a new class of convex
functions can be constructed. Let f be a proper function. Choose a point pAℝn and
construct a function
gx;p ðαÞ 5 f ðx 1 αpÞ:

Lemma 1.28. A function f is convex if and only if gx,p(α) is convex in α for all x
and p.
& If f is convex, then for all λ1 $ 0, λ2 $ 0, λ1 1 λ2 5 1, we have
gx;p ðλ1 α1 1 λ2 α2 Þ 5 f ðx 1 ðλ1 α1 1 λ2 α2 ÞpÞ
5 f ðλ1 ðx 1 α1 pÞ 1 λ2 ðx 1 α2 pÞÞ
# λ1 f ðx 1 α1 pÞ 1 λ2 ðx 1 α2 pÞ
5 λ1 gx;p ðα1 Þ 1 λ2 gx;p ðα2 Þ:

Conversely, let gx,p(α) be convex on α for all x and p. Then

f ðλ1 x1 1 λ2 x2 Þ 5 f ðx2 1 ðλ1 U1 1 λ2 U0Þðx1 2 x2 ÞÞ

5 gx2 ; x12x2 ðλ1 U1 1 λ2 U0Þ
# λ1 gx2 ; x12x2 ð1Þ 1 λ2 gx2 ; x12x2 ð0Þ
5 λ1 f ðx1 Þ 1 λ2 f ðx2 Þ:’

We shall denote the gradient of f at point x by f 0 (x) and the Hessian matrix of
f at x whose (i,j)th element is @2f/@xi@xj by f v(x). Thus, f 0 (x) is a column vector
with components @f(x)/@xi, i 5 1, . . . , n, and
 
@2 f ðxÞ
f vðxÞ 5 ; i 5 1; . . . ; n; j 5 1; . . . ; n:
@xi @xj

Let denote g(α) 5 f(x 1 αp). It is not hard to verify that

g0 ðαÞ 5 hp; f 0 ðx 1 αpÞi;

gvðαÞ 5 hp; f vðx 1 αpÞpi:

Theorem 1.15. Let f be a differentiable function. Then the following assertions are
equivalent:
1. f is convex.
2. f(x2) 2 f(x1) $ hx2 2 x1, f 0 (x1)i for all x1 and x2.
3. hp,f 0 (x 1 αp)i is nondecreasing function on α for all x and p.
4. If f is twice differentiable, then fv(x) is positive semidefinite matrix.
Convex Sets and Functions 29

& We show that (1) implies (2). In fact, since

f ðð1 2 λÞx1 1 λx2 Þ # ð1 2 λÞf ðx1 Þ 1 λf ðx2 Þ; 0 , λ , 1;

then
f ðx1 1 λðx2 2 x1 ÞÞ 2 f ðx1 Þ
# f ðx2 Þ 2 f ðx1 Þ:
λ
By passing to the limit as λ-0, we have:
hx2 2 x1 ; f 0 ðx1 Þi # f ðx2 Þ 2 f ðx1 Þ: ð1:31Þ

Now we show that (2)-(3). Taking x1 5 x 1 α1p, x2 5 x 1 α2p in Eq. (1.31), we

get
gx;p ðα2 Þ 2 gx;p ðα1 Þ $ ðα2 2 α1 Þhp; f 0 ðx 1 α1 pÞi:

Similarly,
gx;p ðα1 Þ 2 gx;p ðα2 Þ $ ðα1 2 α2 Þhp; f 0 ðx 1 α2 pÞi

if x1 5 x 1 α2p, x2 5 x 1 α1p. From the last two inequalities, it follows that (say
α2 . α1)
gx;p ðα2 Þ 2 gx;p ðα1 Þ
hp; f 0 ðx 1 α1 pÞi # # hp; f 0 ðx 1 α2 pÞi:
α2 2 α 1
0
We show that (3)-(1). Let gx;p ðαÞ 5 hp; f 0 ðx 1 αpÞi be a nondecreasing function of
0 0
α. Hence, gx;p ðα1 Þ # gx;p ðα2 Þ if α2 $ α1.
Suppose that 0 , μ , 1. Then
ð1
0 0
0 # μðα2 2 α1 Þ ½gx;p ðα1 1 τðα2 2 α1 ÞÞ 2 gx;p ðα1 1 τμðα2 2 α1 ÞÞdτ
0
5 ð1 2 μÞgx;p ðα1 Þ 1 μgx;p ðα2 Þ 2 gx;p ðð1 2 μÞα1 1 μα2 Þ;

i.e., gx,p(α) is convex. On the basis of Lemma 1.28, we conclude that f is convex.
0
Let f be a twice differentiable function. We show that (4)$(3). Since gx;p ðαÞ is
nondecreasing,
gvx;p ðαÞ 5 hp; f vðx 1 αpÞpi $ 0; ð1:32Þ

and so the matrix f v(x) is positive semidefinite. Conversely, if Eq. (1.32) is satis-
fied, then gvx;p ðαÞ is nonnegative and hence gvx;p ðαÞ 5 hp; f vðx 1 αpÞpi is
nondecreasing.
Consequently, we have that (1)-(2)-(3)-(1) and (4)$(3).’
It is not hard to see that a quadratic function
1
f ðxÞ 5 hx; Axi 1 hx; bi;
2
30 Approximation and Optimization of Discrete and Differential Inclusions

where A is a symmetric n 3 n matrix, is convex on ℝn if and only if A is positive

semidefinite; i.e.,

hz; Azi $ 0 for every zAℝn :

Lemma 1.29. Let f1, f2, . . . , fm be proper convex functions on ℝn, and let
( )
X
m
f ðxÞ 5 inf f1 ðx1 Þ 1 f2 ðx2 Þ 1 ? 1 fm ðxm Þ : xm Aℝ ; n
xi 5 x :
i51

Then f is a convex function on ℝn.

& Let Ei 5 epi fi and E 5 E1 1 ? 1 Em. Then E is a convex set in ℝn11. By defini-
tion, (x,x0)AE if and only if there exist xiAℝn, x0i Aℝ so that x0i $ f ðxi Þ;
x0 5 x01 1 ? 1 x0m and x 5 x1 1 ? 1 xm. Thus, the f defined in the theorem is the
convex function obtained from E by the construction in formula in Eq. (1.29).’

Definition 1.20. The function f in Lemma 1.29 will be denoted by f1"f2"?"fm

and the operation " is called infimal convolution. The terminology arises from the
fact that in the case of two functions,

ð f1 "f2 ÞðxÞ 5 inf f f1 ðx 2 yÞ 1 f2 ðyÞg

y

and this is analogous to the classical formula for integral convolution.

Let f1 be the Euclidean norm and f2 be the indicator function of a convex set M.
Then we get

ð f1 "f2 ÞðxÞ 5 inf fkx 2 yk 1 δM ð yÞg 5 inf kx 2 yk 5 dM ðUÞ;

y yAM

which establishes the convexity of the distance function dM(  ).

If f2 5 δM, M 5 {b} for some fixed point bAℝn, then ( f1"f2)(x) 5 f1(x 2 b). Hence,
f1"δb is the function whose graph is obtained by translating the graph of f1 hori-
zontally by b.
Properness of convex functions is not always preserved by infimal convolution,
because the infimum in the formula in Lemma 1.29 may be 2N. Indeed, if f1 and
f2 are linear functions not equal to each other, then their infimal convolution is
identically 2N. Note that f1"f2 can be defined for any functions f1, f2:
ℝn-[2N, 1N] in terms of addition of epigraphs:

ð f1 "f2 ÞðxÞ 5 inffx0 : ðx0 ; xÞAðepi f1 1 epi f2 Þg:

As an operation on the family of all functions from ℝn to [2N, 1N], infimal

convolution is commutative, associative, and convexity-preserving.
Convex Sets and Functions 31

Definition 1.21. The convex hull of a nonconvex function g is the function

f 5 conv g, defined by
f ðxÞ 5 inf fx0 : ðx0 ; xÞAconvðepi gÞg:
x0

It is the greatest convex function majorized by g.

The convex hull of an arbitrary collection of functions {fi: iAI} on ℝn, denoted
by conv {fi: iAI}, is the convex hull of the pointwise infimum of the collection:
  
convffi : iAIgðxÞ 5 inf x Aℝ : ðx ; xÞAconv [ epi fi
0 0
:
iAI

Lemma 1.30. Let g be a convex function and α0 , α1 , α2; α0, α1, α2Adom g.
Then
gðα2 Þ 2 gðα0 Þ gðα1 Þ 2 gðα0 Þ gðα1 Þ 2 gðα0 Þ gðα2 Þ 2 gðα1 Þ
$ ; # :
α2 2 α0 α1 2 α0 α1 2 α0 α2 2 α1

& Setting
α1 2 α0 α2 2 α1
λ1 5 ; λ2 5 1 2 λ1 5 ;
α2 2 α0 α2 2 α0
α1 2 α0 α2 2 α1
λ1 α2 1 λ 2 α0 5 α2 1 α0 5 α1 :
α2 2 α0 α2 2 α0

Hence,
α1 2 α0 α2 2 α1
gðα1 Þ 5 gðλ1 α2 1 λ2 α0 Þ # gðα2 Þ 1 gðα0 Þ:
α2 2 α0 α2 2 α0

Subtracting g(α0) from this inequality, we have

α1 2 α0
gðα1 Þ 2 gðα0 Þ # ðgðα2 Þ 2 gðα0 ÞÞ:
α2 2 α0

Deleting α1 2 α0 from this inequality, we obtain the first inequality of the lemma.
On the other hand, since λ1 1 λ2 5 1,

α1 2 α0 α2 2 α1 α1 2 α0 α2 2 α1
gðα1 Þ 1 gðα1 Þ # gðα2 Þ 1 gðα0 Þ;
α2 2 α0 α2 2 α0 α2 2 α0 α2 2 α0

or

α2 2 α1 α1 2 α0
½gðα1 Þ 2 gðα0 Þ # ½gðα2 Þ 2 gðα1 Þ:
α2 2 α0 α2 2 α0

This is the desired inequality.’

32 Approximation and Optimization of Discrete and Differential Inclusions

Note that from the first inequality it can be deduced that (g(α) 2 g(α0))/(α 2 α0)
is a monotonically nondecreasing function in α, α . α0.
It seems that the convexity of a function is closely connected with its continuity.

Theorem 1.16. Let a proper convex function f be bounded above in a neighbor-

hood of some point x0Adom f. Then f is continuous at x0.
& Without loss of generality, suppose that x0 5 0. Let Ω be an open ball with the
center the origin such that f(x) # c1 for all xAΩ. Consider a function g(α) 5 f(αx)
for fixed xAΩ. Taking in the first inequality of Lemma 1.30, α0 5 0, α1 5 α . 0,
α2 5 1, we have
gðαÞ 2 gð0Þ gð1Þ 2 gð0Þ
# :
α 1
Because
gð1Þ 5 f ðxÞ # c1 ; gð0Þ 5 f ð0Þ # c1 ;

we also have
f ðαxÞ 2 f ð0Þ # 2c1 α: ð1:33Þ

Moreover, setting α0 5 21, α1 5 0, α2 5 α in the second inequality of Lemma

1.30, we have
gð0Þ 2 gð21Þ gðαÞ 2 gð0Þ
# ;
0 2 ð21Þ α

whence
22c1 α # f ðαxÞ 2 f ð0Þ: ð1:34Þ

Thus, it follows from the inequalities (1.33) and (1.34) that

9f ðαxÞ 2 f ð0Þ9 # 2c1 α: ð1:35Þ

Take ε . 0 and set δ 5 ε/(2c1) , 1, Ωδ 5 δΩ. Let yAΩδ. Then there is xAΩ such
that y 5 δx. Thus, according to Eq. (1.35), we have
9 f ðyÞ 2 f ð0Þ9 5 9 f ðδxÞ 2 f ð0Þ9 # 2δc1 5 ε;

i.e., f is continuous at x0 5 0.’

Theorem 1.17. If a convex function f is continuous at x0, then f satisfies the

Lipschitz condition for neighborhood points of x0:
9 f ðxÞ 2 f ðx0 Þ9 5 Lkx 2 x0 k;

where L is some constant.

Convex Sets and Functions 33

& Suppose that x0 5 0. Let Ω be an open ball of radius r centered at the origin.
Take yAΩ, satisfying the inequality kyk , r/2, and set x 5 (r/2)(y/kyk). By using
Eq. (1.35), we obtain
 
2kyk
9 f ðyÞ 2 f ð0Þ9 5 f x 2 f ð0Þ # Lkyk;
r

where L 5 4c/r.’

Theorem 1.18. Let f be a proper convex function. Then f is continuous on ri dom f.

& Any point, x0Ari dom f can be made an interior point of some simplex with ver-
tices y0, . . . , ykAdom f, where k 5 dim dom f. Every point of this simplex has the
form:

x 5 λ0 y0 1 ? 1 λk yk ; λi $ 0; i 5 0; 1; . . . ; k;
λ0 1 λ1 1 ? 1 λk 5 1:

Thus,

f ðxÞ # λ0 f ðy0 Þ 1 ? 1 λk f ðyk Þ # max f ðyi Þ;

i

i.e., f is bounded in some neighborhood of x0. Consider a convex function of

y 5 x 2 x 0:

gðyÞ 5 f ðy 1 x0 Þ; yALin dom f :

Now apply Theorem 1.15.’

As is seen from this theorem, a convex function is continuous in dom f and may
have a point of discontinuity only in its boundary. In order to characterize the case
in which there is no such discontinuity, it is convenient to introduce the closure
function concept.

Definition 1.22. A function f is said to be a closure if its epigraph epi f is a closed

set.

Theorem 1.19. The following three conditions are equivalent:

1. f is a closure.
2. The level sets Cα 5 {x: f(x) # α} are convex for every αAℝ1.
3. f is lower semicontinuous throughout ℝn.

& We show that (1)-(2). Let xk-x0 and f(xk) # α. Without loss of generality, we
suppose that f(xk)-μ # α, where μ is either finite or 2N.
We show that f(x0) 5 2N if μ 5 2N. In fact, if f(x0) 5 μ0 is finite, then
f(xk) , μ0 2 ε for large k. Consider the sequence of points (μ0 2 ε, xk)Aepi f. Since
34 Approximation and Optimization of Discrete and Differential Inclusions

(μ0 2 ε, xk)-(μ0 2 ε, x0) and epi f is closed, then (μ0 2 ε,x0)Aepi f; i.e., μ0 2 ε $
f(x0) 5 μ0. This contradiction proves that f(x0) 5 2N, i.e., x0ACα.
If μ is finite, then ( f(xk), xk)-(μ, x0)Aepi f, so α $ μ $ f(x0); i.e., again x0ACα.
Thus, Cα is closed.
Suppose now that (2) is satisfied. If xk-x0 and f(xk)-α, then for all ε . 0 and
sufficiently large k we have f(xk) # α 1 ε and hence f(x0) # α 1 ε (Cα 1 ε is closed).
Since ε is arbitrarily small, then f(x0) # α. Thus, (2)-(3).
Finally we show that (3)-(1). Remember that f is lower semicontinuous at a
point x0 if for every sequence xk-x0,
lim inf f ðxk Þ $ f ðx0 Þ:
k

Thus, if x0k $ f ðxk Þ; f ðx0k ; xk Þ-ðx0 ; x0 Þ; then

x0 $ lim inf f ðxk Þ $ f ðx0 Þ;

k

i.e., (x0,x0)Aepi f, which implies that epi f is closed.’

Theorem 1.20. If f is a closed convex function and f(x0) is finite at x0, then
f(x) . 2N everywhere.
& Above, it was shown that if f takes the value 2N, then f(x) 5 2N for all xAri
dom f. But by Theorem 1.4, if xAdom f then (1 2 λ)x 1 λx1Ari dom f for all λ and
x1Ari dom f. Because f is closed,

f ðxÞ # lim f ðð1 2 λÞx 1 λx1 Þ 5 2N;

λk0

i.e., f(x) 5 2N for all xAdom f. Thus, if f is finite at even one point, then
f(x) . 2N for all x.’

Definition 1.23. If f is a convex function, then the function f defined by

epif 5 epi f
f ðxÞ 5 inf fx0 : ðx0 ; xÞAepi f g
x0

is said to be the closure of f. Since the closure of a convex set is convex (Lemma
1.6), it follows that f is a convex function.
A convex function is said to be closed if f 5 f :

Definition 1.24. A function f on ℝn is said to be positively homogeneous (of degree

one) if for every x one has

f ðλxÞ 5 λf ðxÞ; λ . 0:
Convex Sets and Functions 35

Clearly, positive homogeneity is equivalent to the epigraph’s being a cone in

ℝn11. An example of a positively homogeneous convex function that is not simply
a linear function is f(x) 5 kxk.
Furthermore, if f is convex, then
    
1 1 1 1
f ðx1 1 x2 Þ 5 f 2 x1 1 x2 5 2f x1 1 x2 # f ðx1 Þ 1 f ðx2 Þ
2 2 2 2

or, more generally,

f ðx1 1 ? 1 xm Þ # f ðx1 Þ 1 ? 1 f ðxm Þ:

Setting in the previous inequality x1 5 0, x2 5 x, we obtain f(x) # f(0) 1 f(x); i.e.,

f(0) $ 0.
Let f be a positively homogeneous closed proper convex function. By virtue of the
closure of f,

lim f ðλxÞ 5 lim λf ðxÞ 5 0 $ f ð0Þ:

λk0 λk0

Thus, f(0) 5 0 for homogeneous closed proper convex functions.

Now, let f be an arbitrary convex function on ℝn not identically 1N. The epi-
graph of f as a nonempty convex set in ℝn11 has a recession cone 01 (epi f ). By
definition, (y0, y)A01 (epi f ) if and only if

ðx0 ; xÞ 1 λðy0 ; yÞ 5 ðx0 1 λy0 ; x 1 λyÞAepi f

for every (x0,x) epi f and λ $ 0. This means that

f ðx 1 λyÞ # f ðxÞ 1 λy0

for every x and λ $ 0. By definition, the latter inequality holds for every x and
every λ $ 0 as long as it holds for every x with λ 5 1. In any case, for a given y,
the values of y0 for which (y0, y)A01 (epi f ) will form either a closed interval of ℝ
unbounded above or the empty interval. Hence, 01 (epi f ) is the epigraph of some
function.

Definition 1.25. The recession function is denoted by f 01 and defined as

epi ð f 01 Þ 5 01 ðepi f Þ:

Obviously, if f is a proper convex function, then the recession function f 01 of f

is a positively homogeneous proper convex function.
36 Approximation and Optimization of Discrete and Differential Inclusions

1.5 Conjugate of Convex Function

The definition of the conjugate of a function grows naturally out of the fact that the
epigraph of a closed proper convex function on ℝn is the intersection of the closed
half-spaces in ℝn11 that contain it. Remember that a closed convex set can be char-
acterized completely by its support function
HM ðx Þ 5 supfhx; x i : xAMg
x

We now calculate the support function of the epigraph of a closed convex func-
tion f. Taking into account that epi fCℝn11, we can write
 
Hepi f ðx0 ; x Þ 5 sup fhx; x i 1 x0 x0 : ðx0 ; xÞAepi f g:
ðx0 ;xÞ


If x0 . 0, then for a given x the number x0 may be an arbitrary number, larger
 
than f(x). Hence Hepi f (x0 , x ) 5 1N. It remains to calculate Hepi f when x0 # 0.
0
Below, we will show that for the evaluation of Hepi f, x , 0 when f is a closed

convex function, it is sufficient to find Hepi f ( 2 1, x ). If x0 5 21, then
Hepi f ð 21; x Þ 5 sup fhx; x ix0 : x0 $ f ðxÞg 5 supfhx; x i 2 f ðxÞg:
ðx0 ;xÞ x

Definition 1.26. The function

f  ðx Þ 5 supfhx; x i 2 f ðxÞg
x

is called the conjugate of f.

Lemma 1.31. The conjugate function is closed and convex.

& For a fixed x the function hx,x i 2 f(x) is linear in x , so it is closed and convex
in x . The epigraph of f  (x ) is the intersection of the epigraphs of the closed con-
vex functions hx,x i 2 f(x). Therefore, epi f  is closed and convex.’
For example, consider the function f(x) 5 (1/2)x2, xAℝ. Then, by definition,
   
1 1 2 1
f  ðx Þ 5 sup xUx 2 x2 5 sup x 2 ðx 2 x Þ2 ;
x 2 x 2 2

i.e.,
1 2
f  ðx Þ 5 x ; x Aℝ:
2
It is useful to remember, in particular, that Young’s inequality,

f ðxÞ 1 f  ðx Þ $ hx; x i;

Convex Sets and Functions 37

holds for any function. If here, f is proper and convex, then we shall refer to this
relation as Fenchel’s inequality.
Taking conjugates clearly reverses inequalities: f1 $ f2 implies f1 # f2 .
The polar of a set M is denoted by MO and defined as

M O 5 fx : HM ðx Þ # 1g:

Note that if K is a convex cone, then

K O 5 fx : hx; x i # 0 ’xAKg;

and so K 5 2KO. Let us calculate the conjugate of Minkowski’s function rM(  ):


rM ðx Þ5supfhx;x i2rM ðxÞg5supfhx;x i2inf½α.0 : α21 xAMg
5supfhx;x i2α : α.0;α21 xAMg5supfsup½hx;x i : xAαM2α : α.0g
5 sup fα½sup hx;x i21g5δM O ðx Þ;
α.0 xAM

where δMO ðUÞ is the indicator function of MO.

Theorem 1.21. If f is a closed proper convex function, then

f 5 f  ;

where, according to Definition 1.22,

f  ðxÞ 5 supfhx; x i 2 f  ðx Þg:

x

& By Fenchel’s inequality,

f ðxÞ $ hx; x i 2 f  ðx Þ;

and so

f ðxÞ $ f  ðxÞ:

In particular, it follows that dom fDdom f  . Let x0Adom f. Choose γ , f(x0) and
consider in ℝn11 the following set:

Pε 5 fðx0 ; xÞ : x 2 x0 AεB; x0 # γg;

where B is the unit ball in ℝn. Since epi f is closed, then for sufficiently small ε the
sets Pε and epi f are disjoint. We argue by contradiction. Suppose that there exists
a sequence ðx0k ; xk ÞAPεk ; εk -0; x0k # γ such that ðx0k ; xk ÞAepi f : Since εk-0,
38 Approximation and Optimization of Discrete and Differential Inclusions

xk-x0, limk-N x0k # γ; and by virtue of the fact that epi f is closed,
γ $ limk-N x0k $ f ðx0 Þ: This contradiction proves that epi f\Pε 5 [. Thus we can

separate epi f and Pε. Then there is a nonzero vector (x0 , x ) such that
 
x0 y0 1 hy; x i $ x0 x0 1 hx; x i
ð1:36Þ
ðy0 ; yÞAPε ; ðx0 ; xÞAepi f :

Since y0 can be chosen as any number less than γ, from Eq. (1.36), it follows
  
that x0 # 0. Actually, we show that x0 , 0. If x0 5 0, then Eq. (1.36) implies that
hy; x i $ hx; x i; y 2 x0 AεB; xAdom f :

Setting x 5 x0, we have

hy 2 x0 ; x i $ 0; y 2 x0 AεB:

By virtue of the arbitrariness of y 2 x0AεB, this inequality is satisfied only if


x 5 0. In turn, this contradicts the fact that (x0 , x ) is nonzero.
0
Let now x 5 21. Then Eq. (1.36) takes the form
2y0 1 hy; x i $ 2x0 1 hx; x i;
ð1:37Þ
y # γ; y 2 x0 AεB; x0 $ f ðxÞ:
0

Let y0 5 γ, x0 5 f(x), y 5 x0. Thus, from Eq. (1.37) we obtain

hx0 ; x i 2 γ $ supfhx; x i 2 f ðxÞg 5 f  ðx Þ:
x

or
γ # hx0 ; x i 2 f  ðx Þ;

whence
γ # supfhx0 ; x i 2 f ðx Þg 5 f  ðx0 Þ: ð1:38Þ
x

Since γ is an arbitrary number less than f(x0), Eq. (1.36) implies that
f ðx0 Þ # f  ðx0 Þ:

Comparing this inequality with f(x0) $ f  (x0), we conclude that f(x0) 5 f  (x0).
Now assume that x0 2 = dom f; i.e., f(x0) 5 1N. In just the same way as before,
taking arbitrary γ it can be shown that epi f\Pε 5 [ for a small ε and Eq. (1.36)
holds.

If for arbitrarily large γ we have x0 , 0, then Eq. (1.38) holds for γ, and so
f  (x0) 5 1N.

Now let x0 5 0 for some γ. Then from Eq. (1.36), we obtain

hy; x i $ hx; x i; y 2 x0 AεB; xAdom f ;

Convex Sets and Functions 39

or
hy 2 x0 ; x i $ hx 2 x0 ; x i; y 2 x0 AεB; xAdom f :

Thus,
2εkx k $ sup hx; x i 2 hx0 ; x i ð1:39Þ
xAdom f

Remember that there exists a point x1 such that f(x1) is a finite number. Hence,
by substituting γ , f(x1), y 5 x1 into Eq. (1.37), we derive:

2γ 1 hx1 ; x i $ 2 f ðxÞ 1 hx; x i;

i.e., f  (x ) # 2γ 1 hx1,x i. Thus, there is a vector x1 with finite f ðx1 Þ: It follows
that
f  ðx1 1 αx Þ 5 supfhx; x1 1 αx i 2 f ðxÞg # f  ðx1 Þ 1 α sup hx; x i
x xAdom f

and hence, taking into account Eq. (1.39),

f ðxÞ 5 f  ðx0 Þ $ hx0 ; x1 1 αx i 2 f  ðx1 1 αx Þ
$ hx0 ; x1 i 2 f  ðx1 Þ 1 α½hx0 ; x i 2 sup hx; x i
xAdom f
$ hx0 ; x1 i 2 f  ðx1 Þ 1 αεkx k- 1N as α- 1N:’

Theorem 1.22. If f is closed, proper, and convex, then f  is closed, proper, and
convex.
& By Lemma 1.31 f  is closed and convex. It remains to prove that it is proper.
Because f(x1) is a finite number for some x1Adom f, then

f  ðx Þ $ hx1 ; x i 2 f ðx1 Þ;

and so f  is not equal to 2N. In the proof of the preceding theorem, we saw that
there exists x1 with f  ðx1 Þ finite. This means that f is proper.’

Corollary 1.1. If f is lower semicontinuous at x0Adom f, then f(x0) 5 f  (x0).

The conjugate function of f, a closed proper convex positively homogeneous func-
tion, is

f  ðx Þ 5 supfhx; x i 2 f ðxÞg $ 2 f ð0Þ 5 0:

x

If there exists a point x1 such that

hx1 ; x i 2 f ðx1 Þ . 0;
Other documents randomly have
different content
giving effect to it.... In my opinion it is clear that if the Crown desires to set up its
title, as a bar to the investigation by the Native Land Court in its ordinary
jurisdiction of claims by natives, it must either be prepared to prove its title, or it
must be able to rely upon a proclamation in accordance with the terms of the 85th
section of "The Native Land Act, 1909."

COOPER, J.—I have had the advantage of reading and considering the judgment
of His Honour, the Chief Justice, and, upon substantially the same grounds as are
expressed by His Honour in that judgment, I have arrived at the same conclusion.
I have very little to add.
"The Land Act, 1908," contains the statutory provisions regulating the
administration of Crown lands in New Zealand. In respect of Native lands, section
2 brings within the category of Crown lands only those "Native lands which have
been ceded to His Majesty by the Natives on behalf of His Majesty, or otherwise
acquired in freehold from the Natives on behalf of His Majesty, or have become
vested in His Majesty by right of his prerogative."
Customary lands owned by natives, which have not been ceded to His Majesty
or acquired from the native owners on behalf of His Majesty, cannot in my opinion
be said to be land vested in His Majesty by right of his prerogative. It is true that,
technically, the legal estate is in His Majesty, but this legal estate is held subject to
the right of the natives, recognised by the Crown to the possession and ownership
of the customary lands, which they have not ceded to the King, and which His
Majesty has not acquired from them.
Section 338 of the Act draws a clear distinction between Native lands and Crown
lands. Under subsection (1) of that section, where the Governor is satisfied that
Native lands have been acquired by the Crown, he shall by proclamation declare
such lands to be Crown lands, and under subsection (2), when such lands have
been so acquired, he may give effect to any stipulation in the instrument of sale or
transfer to His Majesty for the reservation to the natives of any part of such lands,
and may reserve or grant such portions in manner required by the natives.
Section 88 of "The Native Land Act, 1909," which has been referred to by His
Honour also, recognises that these lands are not "Crown lands." They are only
deemed to be Crown lands for the one purpose, namely, that if any person is in
possession of or trespassing upon, or injuring the lands against the interest of the
Native owners, then, for the purpose of protecting the Native owners the Crown
may under the Land Act take proceedings against such wrongdoers. Even this
provision is a guarded one, for subsection (3) of that section expressly provides
that nothing in it contained shall take away or affect any jurisdiction conferred
upon the Native Land Court.
 CHAPMAN, J.—I agree with the judgments which have been read. It has been
argued that the Treaty of Waitangi was an international treaty entered into with
chiefs having the sovereignty. The contrary opinion was pronounced by the
Supreme Court in Wi Parata v. The Bishop of Wellington (3 N.Z. Jur. N.S. 72). The
terms employed and the mode of execution of the treaty leave it at least an open
question whether it was so regarded at the time. It professes to be made with
certain federated chiefs and certain chiefs who are not federated, but it does not
state over what territories they exercised authority, though the text of the treaty
seems to suggest that it was contemplated that it should be made with several
chiefs who might possibly be regarded, and were provisionally and hypothetically
treated as sovereigns of their respective territories. Later it became a matter of
general knowledge, derived, I presume, from maps prepared pursuant to section
21 of "The Native Land Act, 1873," that there are eighteen or twenty tribes in New
Zealand. If that be so the numerous signatories of the Treaty of Waitangi can
hardly be described as sovereign chiefs. I agree that if they had been explicitly so
declared by Her Majesty's government, or had been so treated in a course of
political transactions that would have been sufficient to make them so, and that
their numbers and their individual unimportance would not have rendered this
impossible, provided that in each case there was a sovereign to a territory.
Hemchand Devchand v. Azam Sakaral Chhotamlal (1906, A.C. 212). The whole
current of authorities shows, however, that the question of the origin of the
sovereignty is immaterial in connection with the rights of private persons
professing to claim under the provisions of the treaty of cession. Cook v. Sprigg
(1899, A.C. 572). Such a treaty only becomes enforceable as part of the municipal
law if and when it is made so by legislative authority. That has not been done. The
sense in which the treaty has received legislative recognition I will refer to later....
From the earliest period of our history, the rights of the natives have been
conserved by numerous legislative enactments. Section 10 of 9 and 10 Vict. cap.
103, called an Act to make further provision for the Government of the New
Zealand Islands (Imperial, 1846), recognises the laws, customs, and usages of the
natives which necessarily include their customs respecting the holding of land.
Section 1 of 10 and 11 Vict. cap. 112, called an Act to promote colonisation in New
Zealand and to authorise a loan to the New Zealand Company (Imperial, 1847),
recognises the claims of the aboriginal inhabitants to the land. To the same effect
is the whole body of colonial legislation. The expressions "land over which the
Native title has not been extinguished" and "land over which the Native title has
been extinguished" (familiar expressions in colonial legislation), are both pregnant
with the same declaration. In the judgment of the Privy Council in Nireaha Tamaki
v. Baker (1901, A.C. 561), importance is attached to these and similar declarations
in considering the effect of colonial legislation. There the whole of the legislation
from the date of the constitution is summarised. This summary includes the
principal colonial Acts. Referring to section 5 of "The Native Rights Act, 1865,"
their Lordships say: "The Legislation, both of the Imperial Parliament and of the
Colonial Legislature is consistent with this view of the construction of 'The Native
Rights Act,' and one is rather at a loss to know what is meant by such expressions
as 'Native title,' 'Native lands,' 'owners,' and 'proprietors,' or the careful provision
against sale of Crown lands until the Native title has been extinguished, if there be
no such title cognizable by the law, and no title therefore to be extinguished." I
might refer further to less precise but equally important expressions, such as
"tribal lands," in "The Native Land Act, 1873," section 21. The various statutory
recognitions of the Treaty of Waitangi mean no more, but they certainly mean no
less than these recognitions of native rights.

FOOTNOTES

[195] Tawhaki, the God-man, whose name frequently occurs in all

the ancient mythology of the Maori race.
[196] Rain and wind—figurative expressions denoting wars and
tumults.
[197] Te Wherowhero Potatau, the first Maori king.
[198] Tawhiao, the second Maori king.
[199] A Total Abstinence organisation.

This relic of the treaty came into the AUTHOR'S possession after the volume was
printed and is now inserted as a supplement.

No te 30 o nga ra o Hanuere, 1840

E taku hoa aroha,
Tenei ano taku ki a koe; na, tenei ano tetahi kaipuke manawa kua u mai nei, me
tetahi Rangatira ano kei runga, no te Kuini o Ingarani ia, hei Kawana hoki mo
tatou. Na, e mea ana ia, kia huihuia katoatia mai nga Rangatira o te Wakaminenga
o Nu Tireni, a te Wenerei i tenei wiki tapu e haere ake nei, kia kitekite ratou i a ia.
Koia ahau ka mea atu nei ki a koe, e hoa, kia haere mai koe ki konei ki Waitangi,
ki taku kainga ano, ki tenei huihuinga. He Rangatira hoki koe no taua
Wakaminenga tahi. Heoi ano, ka mutu taku.
Naku,
Na tou hoa aroha,
Na te PUHIPI.
KI A TAMATI WAKA NENE.
 Text of the invitation sent toTamati Waka Nene to attend the meeting of chiefs at
Waitangi, 5th February, 1840.
(Original in the Auckland Museum).

TRANSLATION
(By H. M. STOWELL).

Of the 3Oth day of January, 1840.

O my dear friend.
Herewith something special from me to you. A Man-o-war vessel has called here
particularly, with a certain Chief on board. He is sent by the Queen of England, as
a Governor for us.
Now, he desires that as many representative New Zealand chiefs as possible be
gathered together by Wednesday of next week to see him.
I now therefore say unto you, O friend, make your arrangements to come along
here to Waitangi, to my home here, to this gathering and meeting. For you
yourself are a typical representative chief and will be welcomed to the meeting as
such.
That is all, mine ends here,
From your sincere friend,
From Te PUHIPI (Mr. Busby)
To Tamati Waka Nene.
 BIBLIOGRAPHY
The following authorities have been consulted during the compilation
of this work, and will serve as a useful bibliography of the treaty:—
A Chapter in the History of New Zealand. Sir W.
Fox. 1883.
Adventure in New Zealand. E. J. Wakefield. 1845.
An Appeal on Behalf of the Ngatiraukawa Tribe. T.
C. Williams. 1873.
Ao Tea Roa (Long White Cloud). W. P. Reeves.
1898.
Appendix to Twelfth Report of New Zealand
Company. 1844.
Aureretanga. G. W. Rusden. 1888.
Australia and New Zealand. A. Trollope. 1873.
Authentic and Genuine History of the Signing. Rev.
W. Colenso. 1890.
Britain of the South. C. Hursthouse. 1857.
British Colonisation of New Zealand. Aborigines
Protection Society. 1846.
Christian Mission to the Aborigines of New
Zealand. Rev. J. Warren. 1863.
Correspondence with Earl Grey. Wesleyan Mission
Committee. 1848.
Early History of New Zealand. Brett Publishing Co.
1890.
Early History of the Catholic Church in Oceania.
Bishop Pompallier. 1888.
England and the New Zealanders. Sir W. Martin.
1847.
Extracts from Final Report. G. Clarke. 1846.
Facsimiles of Treaty of Waitangi. H. H. Turton.
1877.
Forty Years in New Zealand. Rev. J. Buller. 1878.
Governor Gordon and the Maori. Sir R. Stout.
1883.
Handbook for Emigrants. Dr. J. Bright. 1841.
Handbook to New Zealand. E. J. Wakefield. 1848.
Heke's War. Rev. R. Burrows. 1886.
History of Early Colonising of New Zealand. H. T.
Kemp. 1901.
History of New Zealand. Alfred Saunders. 1896.
History of New Zealand. G. W. Rusden. 1883.
Judgments of Court of Appeal in re Tamihana
Korokai v. The Solicitor-General. 1912.
Judgments of Mr. Justice Chapman and Sir William
Martin, C. J., in re Regina v. Symonds. 1847.
King Country, The. J. H. Kerry-Nicholls. 1884.
Kohimarama Conference, Proceedings of. 1860.
Letters from New Zealand. Dr. Martin. 1845.
Life and Times of Patuone. C. O. Davis. 1876.
Life of Archdeacon Henry Williams. H. Carleton.
1874.
Life of Rev. J. H. Bumby. Rev. A. Barrett. 1852.
Life of Bishop Selwyn. Rev. H. W. Tucker. 1879.
Life of Captain James Cook. Rev. A. Kippis. 1788.
Life of Lord John Russell. J. Reid-Stuart.
Life of Rev. Samuel Leigh. Rev. A. Strachan. 1863.
Life of Sir George Grey. W. L. and L. Rees. 1892.
Manawatu Purchase Completed. T. C. Williams.
1867.
Maori English Tutor. H. M. Stowell. 1913.
Maori History. Lieut.-Col. M'Donnell. 1887.
Maori King, The. Sir John Gorst. 1864.
Maori Record (Newspaper). 1906.
Memoir of Rev. R. Davis. Rev. J. D. Coleman.
1865.
Narrative of United States Exploring Expedition.
Commander Wilkes. 1845.
New Zealand. Alexander Kennedy. 1873.
New Zealand. Charles Terry. 1842.
New Zealand. Dr. R. G. Jameson. 1841.
New Zealand and its Aborigines. W. Brown. 1845.
New Zealand and its Colonisation. W. Swainson.
1859.
New Zealand and the War. W. Swainson. 1867.
New Zealand Gazette (Newspaper). 1840.
New Zealand Journal. 1840-1848.
New Zealand Revisited. Sir John Gorst. 1908.
New Zealand Year Book. 1912.
Notes on Early Life in New Zealand. Rev. G.
Clarke. 1903.
Notes on Maori Matters. Mr. Justice Johnston.
1860.
Old New Zealand. F. E. Manning. 1863.
Parliamentary Debates (English), vols. 81-82.
1845.
Parliamentary Papers (English). 1839-1848.
Personal Narrative of Visits to New Zealand. Dr.
Marshall. 1836.
Proceedings of the Royal Colonial Institute.
January 1883.
Remarks on New Zealand. Captain R. Fitzroy.
1846.
Remarks on New Zealand. W. Brodie. 1845.
Reminiscences of a Veteran. Col. T. Bunbury.
1861.
Reminiscences of the War in New Zealand. Lieut.
Gudgeon. 1879.
Romance of a Pro-Consul. James Milne. 1899.
Sketches in New Zealand. James Cowan. 1901.
Story of New Zealand. Dr. A. S. Thompson. 1859.
Te Ika a Maui. Rev. Richard Taylor. 1855.
The Aborigines' Friend, pp. 139-157.
The New Zealanders and their Lands. D. Coates.
1844.
The New Zealand Question. L. A. Chamerovzow.
1848.
The War in New Zealand. Sir W. Fox. 1860.
Voyage to the South Pole, vols. viii. and ix.
Dumont D'Urville.
Waitara War—Numerous Pamphlets relating
thereto.
Printed by R. & R. Clark, Limited, Edinburgh.
Map of the Bay of Islands
Map of Cloudy Bay
*** END OF THE PROJECT GUTENBERG EBOOK THE TREATY OF
WAITANGI; OR, HOW NEW ZEALAND BECAME A BRITISH COLONY
***

Updated editions will replace the previous one—the old editions

will be renamed.

Creating the works from print editions not protected by U.S.

copyright law means that no one owns a United States
copyright in these works, so the Foundation (and you!) can copy
and distribute it in the United States without permission and
without paying copyright royalties. Special rules, set forth in the
General Terms of Use part of this license, apply to copying and
distributing Project Gutenberg™ electronic works to protect the
PROJECT GUTENBERG™ concept and trademark. Project
Gutenberg is a registered trademark, and may not be used if
you charge for an eBook, except by following the terms of the
trademark license, including paying royalties for use of the
Project Gutenberg trademark. If you do not charge anything for
copies of this eBook, complying with the trademark license is
very easy. You may use this eBook for nearly any purpose such
as creation of derivative works, reports, performances and
research. Project Gutenberg eBooks may be modified and
printed and given away—you may do practically ANYTHING in
the United States with eBooks not protected by U.S. copyright
law. Redistribution is subject to the trademark license, especially
commercial redistribution.

START: FULL LICENSE

THE FULL PROJECT GUTENBERG LICENSE
 PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK

To protect the Project Gutenberg™ mission of promoting the

free distribution of electronic works, by using or distributing this
work (or any other work associated in any way with the phrase
“Project Gutenberg”), you agree to comply with all the terms of
the Full Project Gutenberg™ License available with this file or
online at www.gutenberg.org/license.

Section 1. General Terms of Use and

Redistributing Project Gutenberg™
electronic works
1.A. By reading or using any part of this Project Gutenberg™
electronic work, you indicate that you have read, understand,
agree to and accept all the terms of this license and intellectual
property (trademark/copyright) agreement. If you do not agree
to abide by all the terms of this agreement, you must cease
using and return or destroy all copies of Project Gutenberg™
electronic works in your possession. If you paid a fee for
obtaining a copy of or access to a Project Gutenberg™
electronic work and you do not agree to be bound by the terms
of this agreement, you may obtain a refund from the person or
entity to whom you paid the fee as set forth in paragraph 1.E.8.

1.B. “Project Gutenberg” is a registered trademark. It may only

be used on or associated in any way with an electronic work by
people who agree to be bound by the terms of this agreement.
There are a few things that you can do with most Project
Gutenberg™ electronic works even without complying with the
full terms of this agreement. See paragraph 1.C below. There
are a lot of things you can do with Project Gutenberg™
electronic works if you follow the terms of this agreement and
help preserve free future access to Project Gutenberg™
electronic works. See paragraph 1.E below.
1.C. The Project Gutenberg Literary Archive Foundation (“the
Foundation” or PGLAF), owns a compilation copyright in the
collection of Project Gutenberg™ electronic works. Nearly all the
individual works in the collection are in the public domain in the
United States. If an individual work is unprotected by copyright
law in the United States and you are located in the United
States, we do not claim a right to prevent you from copying,
distributing, performing, displaying or creating derivative works
based on the work as long as all references to Project
Gutenberg are removed. Of course, we hope that you will
support the Project Gutenberg™ mission of promoting free
access to electronic works by freely sharing Project Gutenberg™
works in compliance with the terms of this agreement for
keeping the Project Gutenberg™ name associated with the
work. You can easily comply with the terms of this agreement
by keeping this work in the same format with its attached full
Project Gutenberg™ License when you share it without charge
with others.

1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside
the United States, check the laws of your country in addition to
the terms of this agreement before downloading, copying,
displaying, performing, distributing or creating derivative works
based on this work or any other Project Gutenberg™ work. The
Foundation makes no representations concerning the copyright
status of any work in any country other than the United States.

1.E. Unless you have removed all references to Project

Gutenberg:

1.E.1. The following sentence, with active links to, or other

immediate access to, the full Project Gutenberg™ License must
appear prominently whenever any copy of a Project
Gutenberg™ work (any work on which the phrase “Project
Gutenberg” appears, or with which the phrase “Project
Gutenberg” is associated) is accessed, displayed, performed,
viewed, copied or distributed:

This eBook is for the use of anyone anywhere in the United

States and most other parts of the world at no cost and
with almost no restrictions whatsoever. You may copy it,
give it away or re-use it under the terms of the Project
Gutenberg License included with this eBook or online at
www.gutenberg.org. If you are not located in the United
States, you will have to check the laws of the country
where you are located before using this eBook.

1.E.2. If an individual Project Gutenberg™ electronic work is

derived from texts not protected by U.S. copyright law (does not
contain a notice indicating that it is posted with permission of
the copyright holder), the work can be copied and distributed to
anyone in the United States without paying any fees or charges.
If you are redistributing or providing access to a work with the
phrase “Project Gutenberg” associated with or appearing on the
work, you must comply either with the requirements of
paragraphs 1.E.1 through 1.E.7 or obtain permission for the use
of the work and the Project Gutenberg™ trademark as set forth
in paragraphs 1.E.8 or 1.E.9.

1.E.3. If an individual Project Gutenberg™ electronic work is

posted with the permission of the copyright holder, your use and
distribution must comply with both paragraphs 1.E.1 through
1.E.7 and any additional terms imposed by the copyright holder.
Additional terms will be linked to the Project Gutenberg™
License for all works posted with the permission of the copyright
holder found at the beginning of this work.

1.E.4. Do not unlink or detach or remove the full Project

Gutenberg™ License terms from this work, or any files
 containing a part of this work or any other work associated with
Project Gutenberg™.

1.E.5. Do not copy, display, perform, distribute or redistribute

this electronic work, or any part of this electronic work, without
prominently displaying the sentence set forth in paragraph 1.E.1
with active links or immediate access to the full terms of the
Project Gutenberg™ License.

1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form,
including any word processing or hypertext form. However, if
you provide access to or distribute copies of a Project
Gutenberg™ work in a format other than “Plain Vanilla ASCII” or
other format used in the official version posted on the official
Project Gutenberg™ website (www.gutenberg.org), you must,
at no additional cost, fee or expense to the user, provide a copy,
a means of exporting a copy, or a means of obtaining a copy
upon request, of the work in its original “Plain Vanilla ASCII” or
other form. Any alternate format must include the full Project
Gutenberg™ License as specified in paragraph 1.E.1.

1.E.7. Do not charge a fee for access to, viewing, displaying,

performing, copying or distributing any Project Gutenberg™
works unless you comply with paragraph 1.E.8 or 1.E.9.

1.E.8. You may charge a reasonable fee for copies of or

providing access to or distributing Project Gutenberg™
electronic works provided that:

• You pay a royalty fee of 20% of the gross profits you derive
from the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
 payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”

• You provide a full refund of any money paid by a user who

notifies you in writing (or by e-mail) within 30 days of receipt
that s/he does not agree to the terms of the full Project
Gutenberg™ License. You must require such a user to return or
destroy all copies of the works possessed in a physical medium
and discontinue all use of and all access to other copies of
Project Gutenberg™ works.

• You provide, in accordance with paragraph 1.F.3, a full refund of

any money paid for a work or a replacement copy, if a defect in
the electronic work is discovered and reported to you within 90
days of receipt of the work.

• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.

1.E.9. If you wish to charge a fee or distribute a Project

Gutenberg™ electronic work or group of works on different
terms than are set forth in this agreement, you must obtain
permission in writing from the Project Gutenberg Literary
Archive Foundation, the manager of the Project Gutenberg™
trademark. Contact the Foundation as set forth in Section 3
below.

1.F.

1.F.1. Project Gutenberg volunteers and employees expend

considerable effort to identify, do copyright research on,
transcribe and proofread works not protected by U.S. copyright
law in creating the Project Gutenberg™ collection. Despite these
efforts, Project Gutenberg™ electronic works, and the medium
on which they may be stored, may contain “Defects,” such as,
but not limited to, incomplete, inaccurate or corrupt data,
transcription errors, a copyright or other intellectual property
infringement, a defective or damaged disk or other medium, a
computer virus, or computer codes that damage or cannot be
read by your equipment.

1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except

for the “Right of Replacement or Refund” described in
paragraph 1.F.3, the Project Gutenberg Literary Archive
Foundation, the owner of the Project Gutenberg™ trademark,
and any other party distributing a Project Gutenberg™ electronic
work under this agreement, disclaim all liability to you for
damages, costs and expenses, including legal fees. YOU AGREE
THAT YOU HAVE NO REMEDIES FOR NEGLIGENCE, STRICT
LIABILITY, BREACH OF WARRANTY OR BREACH OF CONTRACT
EXCEPT THOSE PROVIDED IN PARAGRAPH 1.F.3. YOU AGREE
THAT THE FOUNDATION, THE TRADEMARK OWNER, AND ANY
DISTRIBUTOR UNDER THIS AGREEMENT WILL NOT BE LIABLE
TO YOU FOR ACTUAL, DIRECT, INDIRECT, CONSEQUENTIAL,
PUNITIVE OR INCIDENTAL DAMAGES EVEN IF YOU GIVE
NOTICE OF THE POSSIBILITY OF SUCH DAMAGE.

1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you

discover a defect in this electronic work within 90 days of
receiving it, you can receive a refund of the money (if any) you
paid for it by sending a written explanation to the person you
received the work from. If you received the work on a physical
medium, you must return the medium with your written
explanation. The person or entity that provided you with the
defective work may elect to provide a replacement copy in lieu
of a refund. If you received the work electronically, the person
or entity providing it to you may choose to give you a second
opportunity to receive the work electronically in lieu of a refund.
If the second copy is also defective, you may demand a refund
in writing without further opportunities to fix the problem.

1.F.4. Except for the limited right of replacement or refund set

forth in paragraph 1.F.3, this work is provided to you ‘AS-IS’,
WITH NO OTHER WARRANTIES OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.

1.F.5. Some states do not allow disclaimers of certain implied

warranties or the exclusion or limitation of certain types of
damages. If any disclaimer or limitation set forth in this
agreement violates the law of the state applicable to this
agreement, the agreement shall be interpreted to make the
maximum disclaimer or limitation permitted by the applicable
state law. The invalidity or unenforceability of any provision of
this agreement shall not void the remaining provisions.

1.F.6. INDEMNITY - You agree to indemnify and hold the

Foundation, the trademark owner, any agent or employee of the
Foundation, anyone providing copies of Project Gutenberg™
electronic works in accordance with this agreement, and any
volunteers associated with the production, promotion and
distribution of Project Gutenberg™ electronic works, harmless
from all liability, costs and expenses, including legal fees, that
arise directly or indirectly from any of the following which you
do or cause to occur: (a) distribution of this or any Project
Gutenberg™ work, (b) alteration, modification, or additions or
deletions to any Project Gutenberg™ work, and (c) any Defect
you cause.

Section 2. Information about the Mission

of Project Gutenberg™
Project Gutenberg™ is synonymous with the free distribution of
electronic works in formats readable by the widest variety of
computers including obsolete, old, middle-aged and new
computers. It exists because of the efforts of hundreds of
volunteers and donations from people in all walks of life.

Volunteers and financial support to provide volunteers with the

assistance they need are critical to reaching Project
Gutenberg™’s goals and ensuring that the Project Gutenberg™
collection will remain freely available for generations to come. In
2001, the Project Gutenberg Literary Archive Foundation was
created to provide a secure and permanent future for Project
Gutenberg™ and future generations. To learn more about the
Project Gutenberg Literary Archive Foundation and how your
efforts and donations can help, see Sections 3 and 4 and the
Foundation information page at www.gutenberg.org.

Section 3. Information about the Project

Gutenberg Literary Archive Foundation
The Project Gutenberg Literary Archive Foundation is a non-
profit 501(c)(3) educational corporation organized under the
laws of the state of Mississippi and granted tax exempt status
by the Internal Revenue Service. The Foundation’s EIN or
federal tax identification number is 64-6221541. Contributions
to the Project Gutenberg Literary Archive Foundation are tax
deductible to the full extent permitted by U.S. federal laws and
your state’s laws.

The Foundation’s business office is located at 809 North 1500

West, Salt Lake City, UT 84116, (801) 596-1887. Email contact
links and up to date contact information can be found at the
Foundation’s website and official page at
www.gutenberg.org/contact
 Section 4. Information about Donations to
the Project Gutenberg Literary Archive
Foundation
Project Gutenberg™ depends upon and cannot survive without
widespread public support and donations to carry out its mission
of increasing the number of public domain and licensed works
that can be freely distributed in machine-readable form
accessible by the widest array of equipment including outdated
equipment. Many small donations ($1 to $5,000) are particularly
important to maintaining tax exempt status with the IRS.

The Foundation is committed to complying with the laws

regulating charities and charitable donations in all 50 states of
the United States. Compliance requirements are not uniform
and it takes a considerable effort, much paperwork and many
fees to meet and keep up with these requirements. We do not
solicit donations in locations where we have not received written
confirmation of compliance. To SEND DONATIONS or determine
the status of compliance for any particular state visit
www.gutenberg.org/donate.

While we cannot and do not solicit contributions from states

where we have not met the solicitation requirements, we know
of no prohibition against accepting unsolicited donations from
donors in such states who approach us with offers to donate.

International donations are gratefully accepted, but we cannot

make any statements concerning tax treatment of donations
received from outside the United States. U.S. laws alone swamp
our small staff.

Please check the Project Gutenberg web pages for current

donation methods and addresses. Donations are accepted in a
number of other ways including checks, online payments and
credit card donations. To donate, please visit:
www.gutenberg.org/donate.

Section 5. General Information About

Project Gutenberg™ electronic works
Professor Michael S. Hart was the originator of the Project
Gutenberg™ concept of a library of electronic works that could
be freely shared with anyone. For forty years, he produced and
distributed Project Gutenberg™ eBooks with only a loose
network of volunteer support.

Project Gutenberg™ eBooks are often created from several

printed editions, all of which are confirmed as not protected by
copyright in the U.S. unless a copyright notice is included. Thus,
we do not necessarily keep eBooks in compliance with any
particular paper edition.

Most people start at our website which has the main PG search
facility: www.gutenberg.org.

This website includes information about Project Gutenberg™,

including how to make donations to the Project Gutenberg
Literary Archive Foundation, how to help produce our new
eBooks, and how to subscribe to our email newsletter to hear
about new eBooks.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com