Functional Analysis S. Kesavan Institute of Mathematical Sciences Chennai HINDUSTAN BOOK AGENCYChapter 2 introduces the notion of a normed linear space and that of continuous linear transformations between such spaces. Chapter 3 studies the analytic and geometric versions of the Hahn- Banach theorem and presents some applications of these. Chapter 4 is devoted to the famous ‘trinity’ in functional analysis - the Banach-Steinhaus, the open mapping and the closed graph theo- rems, which are all consequences of Baire's theorem for complete metric spaces. Several applications are discussed. The notion of an ‘unbounded linear mapping’ is introduced. In my opinion, most texts do not emphasize the importance of weak topologies in a course on functional analysis. That a bounded sequence in a reflexive space admits a weakly convergent sequence is the corner stone of many an existence proof in the theory of (partial) differential equations. These topologies also provide nice counter-examples to show the inadequacy of sequences in a general topological space. For instance, we will see that two topologies on a set could be different while having the same convergent sequences. We will also see an example of a com- pact topological space in which a sequence does not have any convergent subsequence. Chapter 5 deals with weak and weak* topologies and their applications to the notions of reflexivity, separability and uniform con- vexity. Chapter 6 introduces the Lebesgue spaces and also presents the the- ory of one of the simplest classes of Sobolev spaces. Chapter 7 is devoted to the study of Hilbert spaces. Chapter 8 studies compact operators and their spectra. Much of the fun in learning mathematics comes from actually doing it! Every chapter from Chapters 2 through 8 has a fairly large collection of exercises at the end. These illustrate the results of the text, show the optimality of the hypotheses of the various theorems via various ex- amples or counterexamples, or develop simple versions of theories not elaborated upon in the text. They are of varying degrees of difficulty.Texts and Readings in Mathematics Advisory Editor C. S. Seshadri, Chennai Mathematical Institute, Chennai. Managing Editor Rajendra Bhatia, Indian Statistical Institute, New Delhi. Editors R. B. Bapat, Indian Statistical Institute, New Delhi. V. S. Borkar, Tata Inst. of Fundamental Research, Mumbai. Probal Chaudhuri, Indian Statistical Institute, Kolkata. V.S. Sunder, Inst. of Mathematical Sciences, Chennai. M. Vanninathan, TIFR Centre, Bangalore. Copyright © 2009, Hindustan Book Agency (India) No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without written permission from the copyright owner, who has also the sole right to grant licences for translation into other languages and publication thereof. Allexport rights for this edition vest exclusively with Hindustan Book Agency (India). Unauthorized export is 2 violation of Copyright Law and is subject to legal action. ISBN 978-81-85931-87-6Occasionally, some hints for the solution are provided. It is hoped that the students will benefit by solving them. Since this is meant to be a first course on functional analysis, I have kept the bibliographic references to a minimum. merely citing important texts where the reader may find further details of the topics covered in this book. No claim of originality is made with respect to the material presented in this book. My treatment of this subject has been influenced by the writings of authors like Simmons (whose book was my first introduction to Functional Analysis) and Rudin. These works figure in the bibliogra- phy of this book. I would also like to mention a charming book, hardly known outside the francophonic mathematical community, viz. ‘Analyse Fonctionnelle’ by H. Brézis. The preparation of this manuscript would not have been possible but for the excellent facilities provided by the Institute of Mathematical Sciences, Chennai, and I wish to place on record my gratitude. I also thank the Hindustan Book Agency and the editor of the TRIM Series, Prof. R. Bhatia, for their kind cooperation in bringing out this volume. 1 must thank the anonymous referees who painstakingly went through the first draft of the manuscript. I have tried to incorporate many of their constructive suggestions in the current version. Finally, I thank my family for its constant support and encourage- ment. Chennai, S. Kesavan May 2008.Notations Certain general conventions followed throughout the text regarding notations are described below. All other specific notations are explained as and when they appear in the text. e The set of natural numbers is denoted by the symbol N, the in- tegers by Z, the rationals by Q, the reals by R and the complex numbers by C. @ Sets (including vector spaces and their subspaces) and also linear transformations between vector spaces are denoted by upper case latin letters, e Elements of sets (and, therefore, vectors as well) are denoted by lower case latin letters. @ Scalars are denoted by lower case greek letters. ¢ To distinguish between the scalar zero and the null (or zero) vector. the latter is denoted by the zero in boldface, 1.c. 0. This is also used to denote the zero linear transformation and the the zero linear functional. Column vectors in Euclidean space are denoted by lower case latin letters in boldface and matrices are denoted by upper case latin letters in boldface. e Elements of L? spaces (cf. Chapter 6) are equivalence classes of functions under the equivalence relation of ‘equality almost every- where’. To emphasize this fact, all elements of L? spaces (and hence of Sobolev spaces as well) are denoted by lower case latin letters in the sanserif font. A generic representative of that equiva- lence class is denoted by the same lower case latin letter (in italics). Thus, if f € L'(0,1), & generic representative of this class will be denoted by f and will feature in all computations involving this element. For instance, we have 1 ats = [scorer © The norm in a normed linear space V will be denoted by |j.\|, or by |i-lv. if we wish to distinguish it from other norms that maybe entering the argument. Similarly the innerproduct in a Hilbert space H will, in general, be denoted by (.,.) (or by (..-)i in case we wish to stress the role played by H).Preface Functional analysis studies linear spaces provided with suitable topo- logical structures and (continuous) linear transformations between such spaces. It has far reaching applications in several disciplines. For in- stance, the modern theory of (partial) differential equations and the numerical approximation of their solutions rely a lot on functional ana- lytic techniques. Functional analysis is now an integral part of the curriculum in any post graduate course in Mathematics. Ideally it should be taught after having covered courses in linear algebra, real and complex analysis, and topology. An introduction to the theory of measure and integration is also helpful since the richest examples in functional analysis come from function spaces whose study demands such a knowledge. The present book grew out of notes prepared by myself while lectur- ing to graduate students at the Tata Institute of Fundamental Research (Bangalore Centre) and the Institute of Mathematical Sciences, Chen- nai. The material presented in this book is standard and is ideally suited for a course which can be followed by masters students who have cov- ered the necessary prerequisites mentioned earlier. While covering all the standard material, | have also tried to illustrate the use of various theorems via examples taken from differential equations and the calcu- lus of variations, either through brief sections or through the exercises. In fact, this book is well suited for students who would like to pursue a research career in the applications of mathematics. In particular, famil- iarity with the material presented in this book will facilitate studying my earlier book, published nearly two decades ago, ‘Topics in Func- tional Analysis and Applications’ (Wiley Eastern, now called New Age International), which serves as a functional analytic introduction to the theory of partial differential equations. The first chapter of the present book gives a rapid revision of linear algebra, topology and measure theory. Important definitions, examples and results are recalled and no proofs are given. At the end of each section, the reader is referred to a standard text on that topic. This chapter has been included only for reference purposes and it is not in- tended that it be covered in a course based on this book.Contents 1 Preliminaries 1 1.1 Linear Spaces... 0.2 ee ee 1 1.2 Topological Spaces tae u 1.3 Measure and Integration . . . . . sree eereres 18 2 Normed Linear Spaces 26 21 The Norm Topology ...........-...-- L.. 26 22 Examples... 0.0.0.0. 00002 eee eee eee 28 2.3. Continuous Linear Transformations... .........- 36 24 Applications to Differential Equations .......... . 52 25 Exercises 2... ccc ccc ccc cece eres cccese 57 3 Hahn-Banach Theorems BL Analytic Versions 2... eee 3.2 Geometric Versions... 6. eee 3.3. Vector Valued Integration . . 3.4 An Application to Optimization Theory .... 2.2... 85 SG Enercon’ 20s ccc ce svccneseccssveses 92 4 Baire’s Theorem and Applications 41 42 43 44 45 46 47 48 Baire's Theorem .........-. Principle of Uniform Boundedness * Application to Fourier Series .. 2... ...-. =e The Open Mapping and Closed Graph Theorems . . . . . 104 Annibllators. o:05 0 oes ccc ceeeeeesencare 108 Complemented Subspaces .. 2.2... 2000-0000. 113 Unbounded Operators, Adjoints. .........-..-.. 116 TEWRRCWER. ose os 6 oe oo es cesses eessns 1255 Weak and Weak*™ Topologies 51 5.2 5.3 54 5.5 5.6 5.7 6 L” Spaces 7 Hilbert Spaces Bibliography Index The Weak Topology ......-...--0-.--000- The Weak* Topology... -- 2-200 ee ee eee ee Reflexive Spaces Separable Spaces . Uniformly Convex Spaces . . Application: Calculus of Variations Exercises 2-00.22 ee eee ee Basic properties... 2... .--0-.-.0-0-00-- Duals of L? Spaces or The Spaces (2) ...... The Spaces W’! (a,b) a a Exercises . 2... 0.2. eee eee ee ee eee eee Basic Properties ....... The Dual of a Hilbert Space . .. Application: Variational Inequalities =) Orthonormal Sets . sae eee oe . Exerciass 2.0... cc ccc cece wee cee ceee Basic Properties . . Riesz-Fredhélm Theory eee Spectruin of a Compact Operator . . . . Compact Self-Adjoint Operators .. . . saueee EXON Cc ac cers cree ees sees ces2 1 Preliminaries Scalar Multiplication: (a,x) € F x V+ ax € V such that (v) for every x € V, lz =z where 1 ts the multiplicative identity in F; (vi) for alla and 9 € F and for everyre V, a(Szx) = (af)z; (vii) for all a and 3 € F and for everyre V, (a+8)r = ar+ Gr; (viii) for every a € F and for all x and y € V, a(r+y) = ar+ay.8 Remark 1.1.1 The conditions (i) - (iv) above imply that V is an abelian group with respect to vector addition. The conditions (vii) and (viii) above are known as the distributive laws. Example 1.1.1 Let N > 1 be a positive integer. Define R® = {z= (2;,..,.2n) | 2,€R forall b 1, forms a basis for P. Thus, P is an infinite dimen- sional vector space. We will come across numerous examples of infinite dimensional vec- tor spaces in the sequel (cf. Section 2.2, for instance). Let V be a vector space and let Wj, 1 < i dim(W). Thus, if T is invertible, then the two spaces must have the same dimension. We now focus our attention on finite dimensional spaces. Let V be a space of dimension n with basis {v),...,Un} and W a space of dimension m with basis {w;,...,wWm}. A linear map T: V — W is completely defined, once it is defined on a basis. So let us write = Ty) = Sty, 1 sj san. (1.1.1) The coefficients (t,,) in the above relation form a matrix with m rows and n columns. Such a matrix is referred to as an m x n matrix. The j-th column of the matrix represents the coefficients in the expansion of T(v;) in terms of the basis {w,}7, of W. Of course, if we change the bases for V and W, the same linear transformation will be given by another matrix. In particular, let dim(V) = n and let T: V — V bea linear operator. Let T be represented by the n x n matrix (also known as a square matrix of order n) T = (t,;) with respect to a given basis. If ge change the basis, then T will be represented by another n x n matrix = (ty) and the two will be connected by a relation of the form: T = PTP'6 1 Preliminaries where P is called the change of basis matrix and represents the lincar transformation which maps onc basis to another. The matrix P~! rep- resents the inverse of this change of basis mapping and is the inverse matrix of P. In this case, the matrices T and T are said to be simi- lar. The identity matrix I represents the identity mapping z — z for all x € V for any fixed basis of V. For a given basis, if T: V — V is invertible, then the matrix representing T~! will be the inverse of the matrix representing T. A square matrix is said to be diagonal if all its off-dizgonal entries are zero. A n x n square matrix A = (q,,) is said to be upper trian- gular (repectively. lower triangular) if a,, = 0 for all 1 0. The matriz T is said to be positive definite if, in addition, the above inequality is strict if x #0. @ Remark 1.1,2 A hermitian matrix is equal to its adjoint and the inverse of a unitary matrix is its adjoint. A matrix T, with real entries, which is equal to its transpose is called symmetric and one whose inverse is its own transpose is called orthogonal. In case the matrix T has real entries, we can still define positive semi-definiteness (or positive definite. ness) by considering real column vectors x in the above definition.8 1 Preliminaries ‘We now introduce an important notion, viz. that of the determinant. Before we do this, we need some notation. Let S, denote the set of all permutations of n symbols, A transposition is a permutation wherein two symbols exchange places with each other and all other symbols are left invariant. It is known that every permutation is the product (in the sense of composition of mappings) of transpositions. A permutation is even if it is the product of an even number of transpositions and odd if it is the product of an odd number of transpositions. The signature of a permutation ¢, denoted sgn a, is +1 if the permutation is even and —1 if it is odd. Definition 1.1.16 Let T = (tij) be annxn matriz, The determinant of T. denoted det(T), is given by the formula det(T) = > (sen o)ty.6(1)--tno(ny: e€S. We list below the important properties of the determinant. Proposition 1.1.4 (i) //1 is the identity matrix of order n, then det(I) = 1. (ii) If T and S are two n x n matrices, then det(ST) = det(S).det(T). In particular, T is invertible if, and only if, det(T) #0 and det(T“') = (det(T))7'. (iii) IfT is ann xn matriz, then det(T’) = det(T). @ Definition 1.1.17 An invertible matriz is also said to be a non-singular matriz. Otherwise, it is said to be singular. @ If T is a non-singular matrix of order n, then, given any nx | column vector b, there exists a unique n x 1 column vector x such that Tx =b. This is because the corresponding linear transformation is invertible and hence onto (which gives the existence of the solution) and one-to-one (which gives the uniqueness of the solution). If T is singular, this is no longer the case and we have the following result.10 1 Preliminaries Proposition 1.1.6 The geometric multiplicity of an eigenvalue does not exceed its algebraic multiplicity. Definition 1.1.20 The set of all eigenvalues of a matriz is called its spectrum. The mazimum of the absolute values of the eigenvalues is called its spectral radius. @ The sum of the diagonal entries of a square matrix T is called its trace and is denoted by tr(T). It is easy to see from the characteristic equation that the trace is the sum of all the eigenvalues taking into account their multiplicities. Similarly the determinant of a matrix is the product of all its eigenvalues (again counting multiplicity). Proposition 1.1.7 (i) The eigenvalues of the adjoint of a matriz are the complex conjugates of the eigenvalues of the original matriz. (ii) The eigenvalues of a hermitian matriz are all real. (iii) The eigenvalues of a hermitian and positive definite matriz are pos- itive. (iv) The eigenvalues of a unitary matriz lie on the unit circle of the compler plane. The eigenvalues of a hermitian matrix admit a ‘variational charac- terization’. Let T be a hermitian matrix of order n. Its eigenvalues are all real and let them be numbered in increasing order as follows: MS A2 See S Ane Let u,, 1 ¢ i 0 such that Biz;r) © {ye X | dlz.y) N. In this case, we write rq z in X.4 1 Preliminaries Definition 1.2.7 Let {X,,7;}, i = 1,2, be two topological spaces and let f : X, + X2 be a given function. We say that f is continuous if {-"(U) is an open set in X; for every open set U in Xp. If f is a bijection such that both f and f~' are continuous, then f is said to be a homeomorphism and the two topological spaces are said to be homeomorphic to each other. @ The following propositions are easy to prove. Proposition 1.2.1 Let {X,d} be a metric space and let {r,} be a se- quence in X. Then x, — x in X if, and only if, for every € > 0, there exists a positive integer N such that d(zy.z) < € forevery k>N.B In particular, every convergent sequence is bounded, i-e., it can be contained in a (sufficiently large) ball. Proposition 1.2.2 Let {X,,d;},i = 1,2, be metric spaces and let f : X1 — Xz be a given function. The following are equivalent: (i) f is continuous. (ii) For every x € X and for every ¢ > 0, there exists 5 > 0 such that whenever di(x,y) < 6, we have do( f(r). f(y)) < ¢. (iii) If tn — x in Xj, then f(r) — f(z) in Xo. Remark 1.2.2 When property (ii) (or, equivalently, (iii)) holds for a particular point x € X, we say that f is continuous at x. Thus, f is continuous if, and only if, it is continuous at each point of X. If f is continuous and, for a given ¢ > 0, the 6 > 0 described in statement (ii) above does not depend on the point z, then the function is said to be uniformly continuous on X. Let {X,d} be a metric space and let E be an arbitrary subset of X. Let r€ X. Define d(z,E) = nf dtz.y). TT called the distance of the point x from the set E. The following proposition is easy to prove. Proposition 1.2.3 Let {X,d} be a metric space and let EC X. Then (i) for all x and y € X, we have \d(z,E) —d(y.E)| < d(z,y).1.2 Topological Spaces 15 Thus, the function x - d(x, E) is a uniformly continuous function on xX; (ii) if E is a closed set, then, d(z,E) = 0 if, and only if, x € E; more generally, if E is any subset of X, we have E = {x€ X | d(z,E) =0}. Definition 1.2.8 Let {X,d} be a metric space. A sequence {zn} in X is said to be Cauchy if, for every > 0, there exists a positive integer N such that d(z,,2) < € for everyk >NI>N. It is simple to verify that every Cauchy sequence is bounded. It is also easy to see that every convergent sequence in a metric space is Cauchy. The converse is not true and this leads to the following important defi- nition. Definition 1.2.9 A metric space is said to be complete if every Cauchy sequence is convergent. Hi With their usual metric, the spaces R and C are complete. We now introduce an important notion which we will study in detail in later chapters. Definition 1.2.10 Let I be an arbitrary indexing set. Let X be a set and {X,, T,}er be topological spaces. Let f,: X — X, be given functions. The weak topology generated by the functions f, i € T, is the smallest topology on X such that all the f, are continuous. From the above definition it follows that a subbase for the weak topology generated by the f; is the collection of all sets of the form S7'(U) where U is an arbitrary open set in X, and the index i ranges over the indexing set Z. A typical neighbourhood of a point x € X will therefore be a finite intersection of sets of the form f-'(U;) where U, is a neighbourhood of f,(z) in X;. Definition 1.2.11 Let I be an indezing set and let {X,,7j}, i € I be topological spaces. Set X = MyerX;. Let x = (z,)er- Let pir € X 2; € X, be the i-th coordinate projection. The product topology on X is the weak topology generated by the coordinate projections, i.e. it is the smallest topology such that the projections are all continuous. Ml16 1 Preliminaries Thus, sets of the form IjczU,;, where U; = X;, for all i # ty (an arbitrary element of Z) and Uj, is open in X,,, form a subbase for the product topology. A base for the topology is the collection of all sets of the form Il,¢zU, where U, = X, for all but a finite number of indices and, for those indices, U, is an open set in Xj. Definition 1.2.12 Let {X,T} be a topological space and let 0 # K C X. A collection of open sets F is said to be an open cover of K if the union of the members of F contains K. A subcover of F is a subcollection of members of F which is also an open cover of K. Definition 1.2.13 Let {X,7} be a topological space and let 8 # K C X. The set K is said to be a compact set if every open cover of K admits a finite subcover. If X is itself a compact set, we say that it is a compact space. We can also describe compactness via closed sets. Definition 1.2.14 A collection A of subsets of a set X is said to have finite intersection property if every finite subcollection has non- empty intersection. The following proposition is easily proved. Proposition 1.2.4 A non-empty subset K of a topological space is com- pact if, and only if, every collection of closed sets in K having finite intersection property has non-empty intersection. Definition 1.2.15 Let {X,T7} be a topological space and let @# KC X. The set K is said to be sequentially compact if every sequence in K has a convergent subsequence. We list important facts about compact sets in the following propo- sition. Proposition 1.2.5 (i) Every continuous image of a compact set is com- pact. (ii) The product of compact sets is compact. (iii) Compact subsets of Hausdorff spaces are closed. (iv) A compact subset of a metric space is closed and bounded. (v) For the usual topology on R%, a subset is compact if, and only if, it is closed and bounded.1.2 Topological Spaces 17 (vi) Every continuous real valued function on a compact space is bounded and attains its marimum and minimum values in that set. (vii) Every continuous real valued function on a compact metric space is uniformly continuous. Compact metric spaces are very special. In order to characterize them, we need the following notion. Definition 1.2.16 A metric space {Xd} is said to be totally bounded if, for every € > 0, there exists finite sct of points {z,}M*) such that X c UNO B(z:6). Proposition 1.2.6 Let {X,d} be a metric space. The following state- ments are equivalent: (i) X is compact. (ii) X ts sequentially compact. (iii) Every infinite subset of X has a limit point. (iv) X is complete and totally bounded. We conclude this section with one final important topological notion. Definition 1.2.17 Let {X,T} be a topological space. We say that X is connected if there do not exist non-empty open sets U and V such that X = UUV and UNV = 0. A subset A C X is said to be connected if there do not exist disjoint open sets U and V such that A=AN(UUV). ANU #0, ANV £0. Definition 1.2.18 A non-empty subset A of a topological space is said to be path connected if given any patr of points x and y in A, there erists a continuous function y : [0,1] + A such that 7(0) = x and V1)=y.# Proposition 1.2.7 (i) Every continuous image of a connected set is connected. (ii) The product of connected sets is connected. (iii) Every path connected set is connected. In particular, every ball in a metric space is path connected, and hence, connected. The only connected sets in R are intervals. For a detailed study of topological spaces, the reader is referred to, for instance, Simmons [9].18 1 Preliminaries 1.3. Measure and Integration In this section, we will recall basic facts and results of Lebesgue’s theory of measure and integration. Definition 1.3.1 Let X be a set. A o-algebra is a collection S of subsets of X such that (i) X ES; (ii) if AES, then AS ES, where AS denotes the complement of A in X; (iii) if A, € S fori €N, then U7, A € S. The pair (X,S) is then called a measurable space. The members of S are called measurable sets. & In other words, a o-algebra on a set X is collection of subsets of Xcontaining X and which is closed under the set theoretic operations of complementation and countable unions. Definition 1.3.2 Let (X.S) be a measurable space. Let f : X —+ R be a given function. It is said to be a measurable function if, for every a@e€R, we have I-\((a,90)) = {rE X | f(z) >a} ES. A complex valued function is said to be measurable if its real and imag- tnary parts are measurable. @ Proposition 1.3.1 Let (X,S) be a measurable space and let f : X —-R be a given function. Then, the following are equivalent: (i) £7 '((a,00)) € S, for alla ER. (ii) f-"([a,00)) € S, for alla ER. (iii) f-'((—9¢,a)) € S, for alla c R. (iv) f-"((-00, a) € S, for alae R. Given any collection of subsets of a set X, there is a smallest o- algebra containing the collection and it is called the o-algebra generated by the given collection of sets. If (X,7) is a topological space , the the o-algebra generated by the open sets is called the Borel o-algebra and its members are called Borel sets.1,39 Measure and Integration 19 Definition 1.3.3 Let (X.S) be a measurable space. A measure on X is a function js ; S — [0,00] such that, if Aji € N are members of S which are pairwise disjoint,i.e. AOA, =@ if i # j, then x w(UZ.A) = So aAd- ro The triple (X.S.) is called a measure space. H Definition 1.3.4 Let (X,S,11) be a measure space. If u(X) < 00, then i is said to be a finite measure. If X can be covered by a countable union of measurable sets, each with finite measure, then 41 is said to be ao-finite measure. Example 1.3.1 Let X be any non-empty set and let S be the collection of all subsets of X, which is obviously a o-algebra. If A C X, define H(A) to be the number of elements in A, if A is a finite set and to be 00, otherwise. This defines a measure and is called the counting measure on X. Example 1.3.2 Let (X,S) be as in the preceding example and let 29 € X be a given point. let A CX. Define _[. if med uta) = {0 otherwise. This defines a measure on X which is called the Dirac measure con- centrated at the point ro. @ The most important example of a measure is the Lebesgue mea- sure defined on R'. Consider the real line R endowed with its usual topology. Then, it is possible to construct a o-algebra M on R which contains all the Borel sets and to define a measure y such that for any interval of the form [a, 6) (or [a,b], (a, 6], (a,6)), where a,b € R, we have w([a,b)) = b-a. This measure also has the following additional properties: (i) (Completeness) If E € M and y(E) = 0, then, for any F C E, we have F € M and, a fortiori, u(F) = 0. (ii) (Translation Invariance) If E € M and if a € R, then a+E = {a+zr|reE}emM