Selected Titles in This Series 2 20 19 18 16 15, 14 13 12 cP 10 weaaree John B. Conway, A course in operator theory, 2000 Robert E. Gompf and Andrés I. Stipsiez, 4-manifolds and Kirby caleulus, 1999 Lawrence C. Evans, Partial differential equations, 1998 . Winfried Just and Martin Weese, Discovering modern sct theory’ Il: Set-theoretic tools for every mathematician, 1997 Henryk Iwaniec, Topics in classical automorphic forms, 1997 Richard V. Kadison and John R. Ringrose, Fundamentals of the theory of operator algebras. Volume II: Advanced theory, 1997 Richard V. Kadison and John R. Ringrose, Fundamentals of the theory of operator algebras. Volume I: Elementary theory, 1997 Elliott H. Lieb and Michael Loss, Analysis, 1997 Paul C. Shields, The ergodic theory of discrete sample paths, 1996 N. V. Krylov, Lectures on elliptic and parabolic equations in Hilder spaces, 1996 Jacques Dixmier, Enveloping algebras, 1996 Printing Barry Simon, Representations of finite and compact groups, 1996 Dino Lorenzini, An invitation to arithmetic geometry, 1996 Winfried Just and Martin Weese, Discovering modern set theory. I: The basics, 1996 Gerald J. Janusz, Algebraic number ficlds, second edition, 1996 Jens Carsten Jantzen, Lectures on quantum groups, 1996 Rick Miranda, Algebraic curves and Riemann surfaces, 1995 Russell A. Gordon, The integrals of Lebesgue, Denjoy, Perron, and Henstock, 1994 William W. Adams and Philippe Loustaunau, An introduction to Grabner bases, 1994Bey oc A Course in Operator Theorypoole us oF wv pry ¥4 Cle teers A Course in Operator Theory John B. Conway Graduate Studies in Mathematics Volume 21 American Mathematical Society Providence, Rhode IslandEditorial Board James E. Humphreys (Chair) David J. Saltman David Sattinger Ronald J. Stern 1991 Mathematics Subject Classification. Primary 47A99. Library of Congress Cataloging-in-Publication Data Conway, John B. A course in operator theory / John B. Conway. p.m. — (Graduate studies in mathematics ; v. 21) Includes bibliographical references and index. ISBN 0-8218-2065-6 (alk. paper) 1, Operator theory. 1. Title. II. Series. QA329.C7 1999 818/.724-de21 99-41229 cIP Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting for them, are permitted to make fair use of the material, such as to copy a chapter for use in teaching or research. Permission is granted to quote brief passages from this publication in reviews, provided the customary acknowledgment of the source is given. Republication, systematic copying, or multiple reproduction of any material in this publication is permitted only under license from tlic American Mathematical Society. Requests for such permission should be addressed to the Assistant to the Publisher, American Mathematical Society, P.O. Box 6248, Providence, Rhode Island 02940-6248. Requests can also be made by e-mail to reprint-pernission®ans.org, © 2000 by the American Mathematical Society. All rights reserved. Printed in the United States of America. © The paper used in this book is acid-free and falls within the guidelines established to ensure permanence and durability. Visit the AMS home page at URL: bttp: //wwv.ans.org/ 10987654321 050403020100For Ann, my source for happinessContents Preface xiii Chapter 1. Introduction to C*-Algebras 1 §1. Definition and examples 1 §2. Abelian C*-algebras and the Functional Calculus 7 §3. The positive elements in a C*-algebra 12 §4. Approximate identities 17 §5. Ideals in a C*-algebra 21 §6. Representations of a C*-algebra 24 §7. Positive linear functionals and the GNS construction 29 Chapter 2. Normal Operators 37 §8. Some topologies on B(71) 37 §9. Spectral measures 41 §10. The Spectral Theorem 47 §11. Star-cyclic normal operators 51 §12. The commutant 55 §13. Von Neumann algebras 60 §14. Abelian von Neumann algebras 62 §15. The functional calculus for normal operators 65 Chapter 3. Compact Operators 7 §16. C*-algebras of compact operators 7§17. §18. §19. §20. §21. Contents Ideals of operators Trace class and Hilbert-Schmidt operators The dual spaces of the compact operators and the trace class The weak-star topology Inflation and the topologies Chapter 4. Some Non-Normal Operators §22. §23. §24, §25. §26. §27. §28. §29 §30. §31. Chapter §32. §33. §34. §35. §36. Chapter §37. §38. §39. §40. §41. §42. Chapter §43. §44. Algebras and lattices Isometries Unilateral and bilateral shifts Some results on Hardy spaces The functional calculus for the unilateral shift Weighted shifts The Volterra operator Bergman operators Subnormal operators Essentially normal operators 5. More on C*-Algebras Irreducible representations Positive maps Completely positive maps An application: Spectral sets and the Sz.-Nagy Dilation Theorem Quasicentral approximate identitites 6. Compact Perturbations Behavior of the spectrum under a compact perturbation By perturbations of hermitian operators The Weyl-von Neumann-Berg Theorem Voiculescu’s Theorem Approximately equivalent representations Some applications 7. Introduction to Von Neumann Algebras Elementary properties and examples The Kaplansky Density Theorem 82 86 93, 95 99 105 105 11 118 126 132 136 143 147 157 170 181 181 187 190 198 204 207 207 211 214 220 229 236 241 242 250Contents xi §45. The Pedersen Up-Down Theorem 253 §46. Normal homomorphisms and ideals 258 §47. Equivalence of projections 265 §48. Classification of projections 270 §49. Properties of projections 278 §50. The structure of Type I algebras 282 §51. The classification of Type I algebras 289 §52. Operator-valued measurable functions 294 §53. Some structure theory for continuous algebras 301 §54. Weak-star continuous linear functionals revisited 305 §55. The center-valued trace 311 Chapter 8. Reflexivity 319 §56. Fundamentals and examples 319 §57. Reflexive operators on finite dimensional spaces 323 §58. Hyperreflexive subspaces 327 §59. Reflexivity and duality 335 §60. Hypereflexive von Neumann algebras 342 §61. Some examples of operators 348 Bibliography 355 Index 367 List of Symbols 371Preface The genesis of this book is a pair of courses I taught, one at Indiana University and another at the University of Tennessee. Both of these followed a standard two semester course in functional analysis, though this book is written with only a one semester course in functional analysis as its prerequisite. The aim is to cover with varying depth a variety of subjects that are central to operator theory. Many of these topics have treatises devoted to their explication, and some care has been taken to alert the reader to additional sources for deeper study. So you can think of this book as a sequel to a basic functional analysis course or as a foundation for the study of operator theory. ‘The prerequisites for this book are a bit fuzzy. The reader is assumed to know fundamental functional analysis. Specifically it will be assumed that the reader knows the material in the first seven chapters of [ACFA]. (Repeated reference to [ACFA] is made in this text and it is referenced in this way rather than the form of other references.) The desire is to make the book as accessible as possible. It is in- cumbent on me to avoid the arrogance of assuming or requiring familiarity with one of my previous books. But I have to set the standard somewhere, so I will assume the reader knows what I believe to be the material common to all basic courses in functional analysis: the three basic principles of Banach spaces, the definition and elementary properties of locally convex spaces, the foundations of Banach algebras, including the Riesz functional calculus, and the rudiments of operator theory, i cluding the spectral theory of compact operators. This constitutes the first seven chapters of [ACFA]. This book starts with an introduction to C*-algebras followed by a chapter on normal operators that culminates with the Spectral Theorem and the functional calculus. There is considerable overlap between these first two chapters and Chap- ters VIII and IX in [ACFA]. If the reader really knows these first nine chapters of [ACFA], he/she will be able to fly through the first two chapters here. Such a reader could realistically begin reading Chapter 3 of the present book. There is,xiv Preface however, material in the first two chapters of this book, specifically §4, that does uot appear in [ACFA] Chapter 3 examines compact operators. Again there is some overlap with [ACFA], but most of this material does not appear there. Chapter 4 begins the study of non-normal operators, which has seen so much significant progress in recent times. In particular the reader is introduced to some rather deep connections between operator theory and analytic functions. This is a hallmark of much that has been done in recent years. The Fredholm index appears here. This is not part of the stated prerequisites and is the subject of Chapter XI in [ACFA]. On the other hand. the need for the index is not that substantial that the reader should be dismayed by encountering it. Indeed, Fredholm theory is discussed later in §37, where the reader will see statements of the pertinent results from this topic, some proofs, and references to |ACFA] for the omitted proofs. The reader who feels some insecurity on this point may examine §37 during tle study of Chapter 4. Chapter 5 returns to the theory of C*-algebras and examines irreducible rep- resentations as well as completely positive maps. As an application, a proof is presented of the Sz.-Nagy Dilation Theorem. which is the basis for a large portion of modern operator theory. The chapter closes with the existence of a quasicentral approximate identity in a C*-algebra. This is used in the following chapter. Chapter 6 explores the general topic of compact perturbations. After an ab- breviated treatment of Fredholm theory, the Werl-von Neumann-Berg Theorem is proved as is Voiculescu’s Theorem on the approximation of representations of a separable C’-algebra. The chapter concludes with some applications of these ideas to single operators. . Chapter 7 is a rather extended introduction to von Neumann algebras. The classification scheme is obtained as is complete information on Type I algebras. This is used to recapture the multiplicity theory of normal operators. The chapter includes a proof that a von Neumann algebra is finite if and only if it has a faithful, centered-valued trace. Chapter 8 the last chapter, gives an introduction to reflexive subspaces of operators. Here the word “reflexive” is used differently from its meaning in Banach space theory, Reflexive subspaces are spaces of operators that are determined by their invariant subspaces. An operator is reflexive if and only if the weakly closed algebra it gencrates is a reflexive subspace. This. together with the related notion of a hyperrefiexive subspace, is a still developing area of research. In many ways this subject is one of the more successful episodes in the modern exploration of asyminetric algebras. There are many topics in operator theory that are not included here. Given the vastness of the subject, that is no surprise. Many important topics have been omitted because good treatments already exist in the literature. For example, nothing is in this book on the Brown-Douglas-Fillmore theory, which would have been a natural sequel to the chapter on compact perturbations. But whatever approach ] would have used would not have differed in any substantial way from that in Davidson [1996]. Also the theory of dual algebras is not presented, but this is due more to space and time limitations. So much has happened since Bercovici, Foias, and Pearcy [1985] that the area is ripe for a further exposition. Such anPreface xv exposition would have almost doubled the size of the present book. This list of omissions could continue. Like all books, especially more advanced ones, thé material is not linearly dependent. The reader can skip around, especially after the first three chapters. For example, Chapter 8 on reflexivity does not depend on the preceding three chapters in any substantial way. Developing a dependency chart was a temptation, but instead I'll encourage readers to skip around, covering the topics that interest them and filling in the gaps as necessary. One final caveat. Be aware that I ama bit schizophrenic about separability. On the one hand, I don’t wish to present anything in that setting as though it depends on the assumption of separability. On the other hand, in Hilbert space this is where the interest lies. There are also parts of operator theory that really only hold in a reasonable way when the underlying Hilbert space is separable. There are others that are connected to measure theory and I did not want to get into discussing non-separable measure spaces. So I start out with no assumption of separability, but occasionally giving a result that does depend on this. Later, in §51. all Hilbert spaces are assumed to be separable for the remainder of the book. Throughout the text I give references for further study. I frequently also will cite the source of some, but not all, of the results. I have confidence in the attributions I give, but I certainly am not infallible. I'll maintain a list of corrections and updates on this book linked to my web page (http://www-.math.utk.edu/~conway), and any corrections or changes in attribution will be found there as well in future printings should they come to be. I have many people to thank for their assistance during the preparation of this book. My former student Nathan Feldman read various editions of the manuscript and made many helpful suggestions. Similar help came from my current students Gabriel Prajitura and Sherwin Kouchekian. who. with Nathan. were in onc of the courses that eventually led to this book. In the final analysis, of course, I am the ‘one who is responsible for any errors.Chapter 1 Introduction To C*-Algebras §1. Definition and examples 1.1 Definition. If A is a Banach algebra, an involution is a map a —> a* of A into itself such that for all a and b in A and all scalars a the following conditions hold: (i) (a*)* = a; (ii) (ab)* = b°a’; (iii) (aa + 6)* = aa* +b". If A has an involution as well as an identity, then 1"a = (1*a)"* = (a°1)* =a. Thus 1" = 1 by the uniqueness of the identity. It also follows that a” = & for any multiple of the identity. An algebra that has an identity is called unital, 1.2 Definition. A C*-algebra A is a Banach algebra with involution such that lla" al] = [lal]? for every a in A. This seems a rather simple requirement and it has always impressed the author how much follows from this condition. Here are a few examples. 1.3 Examples. (2) B = B(2), the algebra of bounded operators on a Hilbert space, is a C*-algebra, where for each operator A, A* is the adjoint of A. (See {ACFA], Proposition 11.2.7.) In many ways this algebra is the subject of study of this book, though not just as an algebra. 12 1. Introduction to C*-Algebras _(b) For any Hilbert space 1, By = Bo(#), the algebra of compact operators on the Hilbert space #1, is a C*-algebra, though it lacks an identity when H is not finite dimensional. -(c) If X is a compact space. C(X). the collection of all continuous functions from X into C, is a C*-algebra if for each continuous function f on X, f*(z) = f(z). Here C(X) is an abelian C*-algebra with an identity. In the next section it is shown that every abelian C*-algebra with identity is isomorphic to one of this type. (d) If X is locally compact but not compact, then Co(X), the algebra of continuous functions on X that vanish at infinity. is an abelian C*-algebra without identity. The next section shows that every abelian C*-algebra without identity is isomorphic to one of this type. The starting point is a collection of basic properties of the norm for a C*-algebra. Throughout this section, A will be a C*-algebra. Recall that for any Banach space +’, ball 4 denotes the closed unit ball. 1.4 Proposition. Ifa € A. then: (a) lla"ll = llall: (b) flaa*] = llall?; \lal] = sup{ flax|] : x € ball A} = sup{ |lzal} : x € ball.A}. Proof. (a) |lal|? = |ja"al] < |Ja"|| [lal]. So |Ja"|| < lel. Since a** = a, substituting a* for a in this inequality gives the reverse inequality. (b) Combine (a) with the definition. (c) Let & be the first supremum in (c). Clearly a < |lal] with equality ifa=0. Ifa £0, let x =a"/Ilal}. So {||| = 1 by part (a) and so a > Jal). The proof of the other equality is similar. = When A is a C’-algebra without an identity. there is a unique way of adjoining an identity. For the algebras Bo(H) and Co(X) above, the way to do this is clear. Doing this in general takes a bit more work. If A and C are two C*-algebras, a map p: A — C is called a *-homomorphism or homomorphism if p is an algebraic homomorphism such that p(a*) = p(a)* for all a in A. A +-isomorphism or isomorphism between C*-algebras is a bijective homomorphism. A monomorphism or epimorphism between C*- algebras is defined similarly, It is easy to check that if p is an isomorphism, so is p~!. This is the natural means of identification in the category of C*-algebras and the word “unique.” as used in the first sentence of this paragraph, means unique up to an isomorphism. Similarly when, as in the§1. Definition and examples 3 next proposition, it is stated that the algebra A is contained in the algebra Ai, this means that A is isomorphic to a subalgebra of Ay. 1.5 Proposition. If A is a C*-algebra without identity, then there is a unique C*-algebra with identity A, containing A as an ideal and such that Ai/A is one dimensional. If p: A — C is a +-homomorphism, then pi(a+ a) = p(a) + a defines a *-homomorphism py: Ay > Ci. Proof. Let B(A) be all the bounded operators on the algebra A considered as a Banach space. For a in A define L, in B(A) by La(x) = ax. The reader is asked to verify that the map A: A — B(A) defined by A(a) = Lo is an algebraic homomorphism. By part (c) of the preceding proposition \ is an isometry. Define 4, = (A) + C; so A; 1s a Banach subalgebra of B(A). On A; define [La + a]* = Lg- + &. It is left to the reader to verify that this defines an involution on A). The claim is that with this involution Aj is a C*-algebra. If a € A and ais a scalar, for any € > 0, there is an z in ball.A such that ||La + all? —€ < |jax + az|l?. Thus LN Lb TSX [La + al? —€ < (a*a* + da*)(ax + ax)|] . = ||2*(La + a)" (La + a)z"\| S Il(La + @)*(Le +0)}). Therefore ||La + al? < ||(Le + @)*(La + @)||. On the other hand, because Ay is a Banach algebra, ||(Le + @)*(La + @)|| < ||(Lo +.4)"I| [La +l]. So the proof will be complete if it can be shown that ||(La +4)'I] < [Le + all. But using Proposition 1.4.c twice gives (Lo +a)" = supf{ |la"x + dell : ¢ € ball A} = sup{ lly’a'a + Gy"a|| :2,y € ball A} = sup{ ||z"ay + az"yl| : x,y € ball A} = sup{ ||z"(La + a)yl| : 2,y € ball A} S lla + a]. Thus A; is a C*-algebra and A is a *-monomorphism. Identify A and its image \(A). Clearly A is an ideal in A that has codimension 1. The proof of the statement about the homomorphism is left to the reader. It remains to prove the uniqueness of 4). But if Aj is another C’-algebra with identity that contains A as an ideal of codimension 1, it follows that Ai = .A+C. It is easy to verify that A, and Aj are isomorphic by means of an isomorphism that fixes the elements of A. =4 1. Jituductiva to C*-Aigevras If A is a C*-algebra with identity and a € A, then the spectrum of a, o(a), is defined in the usual way. If A does not have an identity, the spectrum of a is defined as its spectrum when it is considered as an element of the algebra 4; obtained by adjoining an identity to A. 1.6 Definition. If A is a C*-algebra and a € A, then: (a) a is hermitian if a; (b) a is normal if a"a = aa"; (c) when A bas an identity, a is unitary The reader can check the examples in (1.3) to see when an element of these C*-algebras is hermitian, normal, or unitary. Of course every element of an abelian C*-algebra is normal. If A is any C*-algebra and a is a normal element, then the C*-algebra generated by a is abelian. 1.7 Proposition. Ifa € A. the following hold. (a) Ifa is invertible, then a” is invertible and (a*)-} = (a~})*. (b) a=2-+ iy, where x and y are hermitian elements of A. (c) Ifwis a unitary in A. then ful] = 1. (d) Ifa is hermitian, then |Jal] = r(a). the spectral radius of a. (c) If B is another C*-algebra and p: A — B is a +-homomorphism, then llo(@)Il < lial] for alla in A. Proof. The proofs of (a), (b), and (c) are left as exercises. To prove (d), it suffices to assume that A is unital. If a is a hermitian element of A, then |la\|? = |Ja"a|] = |Ja?||. By induction, |Ja/]?" = |ja?"|] for all n > 1. Thus r(a) = lim, []a?"|}!/2" = jal). (e) First note that by adjoining an identity if necessary, it can be assumed that Ais unital. Now the definition of a -homomorphism does not assume that. p(1) is the identity of B. However it is easy to see that p(1) is an identity for p(.A). So there is no loss of generality in assuming that B has an identity and p(1) = 1. For a in A, it is easy to check that o((a)) C (a). From (d) it follows that [lp(a)||? = llo(a*a)|] = r(p(a"a)) < r(a*a) < jlal)?. = Since the inverse of a +-isomorphism is also a +-isomorphism, the follow- ing corollary is immediate. 1.8 Corollary. If A and B are C*-algebras and p: A — B is a +-isomor- phism, then p is an isometry. For any C*-algebra A. Re A will denote the collection of hermitian ele- ments of A. It is worth noting that Re A is a real Banach space.§1. Definition and examples 5 1.9 Proposition. If h : A — C is an algebraic homomorphism, then the following hold. ‘ (a) Ifae ReA, h(a) ER. (b) For an arbitrary element a of A, h(a") = h(a). (c) h(a"a) > 0 for alla in A. (d) If A is unital and u is a unitary in A, then |h(u)| = 1. Proof. By Proposition 1.5 it can be assumed that .4 has an identity and A(1) = 1. By Proposition VII.8.4 of [ACFA], the restriction of h to any abelian subalgebra of A has norm 1. Let a € ReA and note that for any real number t, a + it is normal; hence the C*-algebra generated by a + it and the identity is abelian. Thus [h(a + it)? < lla + ee]? I|(a + it)*(a + it)|] = II(o- it)(a + it)]] = lo? +7] S lla? +2. Suppose h(a) = a +18, a, 8 € R. So the above inequalities yield lla? +0? > a+ i(8 +t)? =a? +(B+t)? =a? +6? +26t+2; hence |lal|? > a? + 6? + 2(t for every real number t. Therefore 3 = 0, proving (a). Let a = x + iy, where x and y are hermitian. Because A(z), h(y) € R and a” = z— iy, part (b) follows. Also h(a*a) = h(a*)h(a) = |h(a)/? > 0, yielding (c). Finally, if u is unitary, |A(u)|? = A(u*u) = A(1 . The next corollary is a restatement of part (b) of the preceding propo- sition. 1.10 Corollary. Every algebraic homomorphism from a C*-algebra into the real numbers is a *-homomorphism. Suppose .A is an abelian C*-algebra and a € A. From basic Banach algebra theory (see [ACFA], Theorem VII.8.6) it is known that the spec- trum of a is {h(a) : A is a homomorphism from A into C}. In light of the6 1. Introduction to C*-Algebras preceding proposition, this implies that o(a) C R whenever A is abelian and a € Re.A. Now suppose A is not abelian, but a € Re A. Let C be the C’-algebra generated by a and the identity: so C is abelian. Thus ¢(a), the spectrum of a as an element of the algebra C, is a subset of R. According to Theorem VII5.4 in [ACFA], .4(a) © o¢(a) and do¢(a) € Ao.4(a). But because o¢(a) C R, it equals its boundary. Thus o.4(a) = oc(a) CR for a in ReA. This proves ancther corollary of Proposition 1.9 and sets the stage for the final theorem of this section. 1.11 Corollary. If a € Re A, o(a) CR. 1.12 Theorem. Jf A and B are two C*-algebras with a common identity such that BC A, then og(a) = 0.4(a) for anya in B. Proof. From the argument preceding the last corollary, it follows that op(a) = o.4(a) if a € ReB. Now let a be an arbitrary element of B. Since o6(a) 2 0.4(a) by Theorem VII.5.4 of [ACFA], it suffices to show that if @ is invertible in A. it is invertible in B. So suppose there is an x in A such that ax = za = 1. Thus (a°a)(z2*) = (x2")(a*a) = 1. But a*a € ReB, so the first step in the proof implies that aa is invertible in B. Since inverses are unique, rz" € B. Hence x = 2(z"a") = (z2")a" € B. m Exercises. clA and int A are used to denote the closure and interior of a set A in a topological space. 1. Let A={f € C(clD) : f is analytic on D} and for f in A define f*(z) = F(@). Show that f — f° is an involution on A and || f*|] = || fl] for all f in A, but A is not a C*-algebra. 2. Let A be a C*-algebra and let X be a locally compact space. Show that Co(X,.A), the set of all continuous functions f: X — A such that for every € > 0, {x : ||f(x)l| 2 €} is compact, is a C*-algebra, where all the operations are defined pointwise and the norm is the sup norm: fll = supf |f(@)Il + 2 € X }. 3. With the notation of the preceding exercise, show that if X and Y are locally compact spaces, then there is a natural *-isomorphism from Co(X,Co(¥)) onto Co(X x Y). 4. Let {Aj : i € J} be a collection of C*-algebras. Show that @{A: : #€ 1) = {{a,} € J], A, with sup, lla:|| < 00} is a C’-algebra if the opera- tions are defined coordinatewise. Show that if @po{ Ai :i € J} is the set of those {a,} in [], A; such that for every € > 0. {7: llail] > €} is finite, then @ol Ai: t€ 1} is a C*-algebra.§2. Abelian C*-algebras and the Functional Calculus 7 5. Show that the C*-algebra obtained by adjoining an identity to Co(X) is isomorphic to C(X), where X is the one-point compactification of X. §2. Abelian C*-algebras and the Functional Calculus Recall a few facts concerning the Gelfand transform. Let A be an abelian Banach algebra with identity and let D be its maximal ideal space. © is identified with the set of non-zero homomorphisms from A into the complex numbers. Each such homomorphism has norm 1 ([ACFA], VII.8.4) and so EC ball.A’. Furnish E with the relative weak” topology, and E becomes a compact Hausdorff space. If a € A, define @: 2 — C by @(h) = h(a). Thus @ € C(£) and this function @ is called the Gelfand transform of a. The m 7: A= C(E) defined by y(a) = @ is the Gelfand transform for the algetfs, It is known that + is an algebraic homomorphism having norm 1. Whe Ais an abelian C*-algebra, much more is true. 3 2.1 Theorem. [fA is an abelian unital C*-algebra and 5 its maximal ideal space, then the Gelfand transform y : A — C() és a *-isomorphism. Proof. For each a in A, |[@lloo < [lal] By Proposition 1.7.d, ||7(2) [loo = llall whenever a € ReA. In particular, ||7(a"a)|loo = ||a*al| for any a in A. By Proposition 1.9.b, @* = y(a*) = @ for all ain A. That is, y(a*) = y(a)* and y is a +-homomorphism. Therefore ||-(a)||2, = {|7(a)*7(2) loo = [I7(a*@)lloo = \|a*al| = |lal|?. Therefore + is an isometry. It remains to show that 7 is surjective; this is done with the aid of the Stone-Weierstrass Theorem. Because ‘y is a *-homomorphism and an isometry, 7(A) is a closed subalgebra of C() containing the constants and closed under complex conjugation. So to show that + is surjective, it. suffices to prove that (A) separates points of E. But if hy and hg are distinct homomorphisms on .A, the definition of homomorphism implies there is an a in A such that hy(a) # ho(a). Thus @ = 7(a) separates the points hy and ho. # Now let A be an abelian C*-algebra without. an identity and let A; be the algebra obtained by adjoining an identity to A. Let E and 5 be the two maximal ideal spaces. Now A is a maximal ideal in A) so if ho : A, — C is the unique homomorphism with ker ho = A, then © = ¥) \ {ho} in a natural manner. Combining this observation with the preceding theorem and a little work produces the following corollary. 2.2 Corollary. If A is an abelian C*-algebra without an identity and Z is ils mazimal ideal space, then the Gelfand transform y : A — Co(Z) is a +-isomorphism.8 1, Introduction to C*-Algebras If (X,2, 4) is a o-finite measure space, L®(y) is an abelian C*-algebra. Thus L(y) is *-isomorphic to C(Z). the «lgebra of continuous functions on its maximal ideal space. See Exercise 5 for more. A bit of notation will prove useful here and later. If A is a C*-algebra and SC A, let C3(S) be the C*-algebra generated by S. That is, C3(S) is the intersection of all the C*-algebras contained in A that contain S. It is trivial to see that Cj(S) is a C*-algebra, though it may not have an identity. Let C*(S) = C3(S U {1}). In particular, if a € A, C*(a) is the closure of all words in a, a*, and 1. The case of an abelian C*-algebra with a single generator is worthy of special consideration. I like to call the next result the “a to z theorem.” 2.3 Theorem. /f A is an abelian unital C*-algebra such that there is an element a with A= C"(a). then there is a unique «-isomorphism 9: A — C(o(a)) such that p(a) = z. Proof. The hypothesis of the proposition implies that A is the closure of {p(a.a*) : p(a.a) is a polynomial in a and a*}. Define r: £ — o(a) by 1(h) = h(a). Clearly 7 is continuous and Theorem VII.8.6 in [ACFA] implies r is surjective. If hi,he € D and r(hi) = 7(ha), then algebraic manipulations show that hy(p(a,a*)) = ho(p(a,a*)) for every polynomial p(a,a*) in @ and a’. By the observation at the beginning of the proof, this implies hy = hg. Thus 7 is injective and must be a homeomorphism. Define p: A — C(o(a)) by p(z) = F077}. It is a routine exercise to check that p is a *-isomorphism. The fact. that. p is the unique *-isomorphism mapping a to z follows directly from the fact that a generates the C*-algebra Cr(a). = Now to exploit this and define f(a) for a normal element a and a contin- uous function f on o(a). Let B be any C*-algebra with identity and let a be a normal element of B. Put A = C"(a). If r: £ — o(a) is the homeomor- phism defined in the last proposition, let 7# » C(o(a)) + C(3) be defined by 7#(f) = for. is left to the reader to check that r# is a +-homomorphism. In fact, since r is a homeomorphism, 7* is a +-isomorphism. Thus 77! 0 r# : C(e(a)) — A is a *-isomorphism. If p(z, 2) is a poly- nomial, 77} 0 r# = p(a,a*). This is the functional calculus for the normal element. a. Since this map f — f(a) takes the identity to the identity and the function z to a, it extends the Riesz Functional Calculus by the uniqueness of that functional calculus. (See Theorem VII.4.7 in [ACFA].) Summarize this as follows.§2. Abelian C*-algebras and the Functional Calculus 9 2.4 Theorem. If B is any C*-algebra and a is a normal element of B, then there is a +-monomorphism f —+ f(a) from C{o(a)) into B having the following properties: (a) IF(@)Il = IIflleo for all f in C(o(a)); (b) f = f(a) extends the Riesz Functional Calculus for a; (c) the range of this map is C*(a). Moreover this map is unique in the following sense. If p: C(o(a)) — C*(a) is a *-homomorphism such that p(1) = 1 and p(z) = a, then p(f) = f(a) for all f in C(o(a)). Proof. The only part not proved yet is the uniqueness statement. Let p be as described. Algebra implies that p(p(z,2)) = p(a,a”) for any polynomial in z and 3. If f € C(o(a)), there is a sequence of polynomials {pn(z, 2)} that converges to f uniformly on o(a). But pp(a,a") = p(pa) — p(f) and also pn(a,a*) + f(a). This referred to as the functional calculus of the normal element a. Be- cause of the uniqueness of the functional calculus, it is not. necessary to remember its exact form but only that it constitutes a *-monomorphism that extends the Riesz Functional calculus. This will be illustrated shortly, but the stage needs to be set. The next two results might be considered examples, but they are truly more than that. The first is an example of a normal operator that is the building block for all normal operators on a separable Hilbert space. The second explores an abelian C*-algebra that is the prototype for all maximal abelian C*-algebras. The second result is actually a generalization of the first, so only that one will be proved. Deriving the first from the second is left as an exercise. The support of a measure y: is defined as the complement of U{U : U is open and u(U) = 0} and is denoted by suppy. Another standard notion from measure theory is the essential range of a measurable function, ess-ran (#) = (){ cl (4(A)) : A € @ and p(X \ A) = 0}. 2.5 Proposition. If y is a compactly supported, regular Borel measure on C, and Ny: L2(u) + L2(y) is defined by N, f(z) = zf(z) for all fn L2(u). then N,, is a normal operator with the following properties: (a) Nu"S(2) = 2f(z) for all f in L7(u); (b) o(Ny) is the support of the measure p1.10 1. Introduction to C*-Algebras 2.6 Proposition. If (X,9,) is a o-finite measure space, for each @ in L(y), define Mg on L?() as multiplication by ¢. If ¢ € L&(u), the following statements hold: (a) the operator Mg is normal and Mg* = M3; (b) @— Mg is a *-homomorphism from L(x) into B(L?(u)); (c) Mell = ll¢lleos (d) o(Mg) = ess-rang. Proof. The proofs of (a) and (b) are routine. If f € L(y), |IMg fll? = JId#P du < [ll IVI?. so llAfell < IIdlloo- Ie > 0, let A € 2 such that A has full measure and sup{ |¢(z)| : z € A} > |\dlloo — €. There is a Borel subset A of A such that 0 < y(A) < oo and |4(z)| > [Idlloc — € for all x in A. If f = [u(A)]-'/2xq, then f is a unit vector in L?(u) and WM SI? = [u(A)}“? fy 101? dit > (Ildlloo — €). This proves (c) (d) Assume that A ¢ ess-ran(¢). So there is a set A in © having full measure such that \ ¢ cl[{é(A)]. Thus there is a 6 > 0 with |d(x) — \| > 6 for all z in A. Hence y = (¢— A)~! € L™(u). From (b) it follows that My = (Mg—d)7)- Conversely, assume that \ € ess-ran (¢). Thus for every integer n there is a set A, in 2 with 0 < u(An) < 00 and |6(z) — A] < n™? for all z in An. If fn = (u(An))7/2xa,5 then {fn} is a sequence of unit vectors in L?(u) and || [Ms — Alfnll < 1/n. Hence \ € o(Mg). (Actually this shows that \ belongs to the approximate point spectrum of Mg. But for a normal operator, the spectrum and approximate point spectrum coincide. A review of these notions is warranted.) In the preceding proof. where is use made of the fact that the measure space is o-finite? 2.7 Example. Continue with the notation as in the last proposition. If $ € L®(u) and f € C(o(Mg), then (Mg) = Myog. In fact, since o(Mg) is the essential range of ¢, f o@ makes sense and is seen to also belong to L(y). Thus Myog is well defined. The reader is asked to check that the map f + Myog is a +-monomorphism. Clearly the function z is mapped to Mg. Thus by the uniqueness of the functional calculus, f(Mg) = Myo. 2.8 Example. If X is a compact space and g € C(X), then o(g) = 9(X) and the functional calculus for g is given by f —+ fog for all f in C(g(X)). These examples illustrate that computing the functional calculus is more a matter of making an educated guess and then verifying that the unique- ness criteria are satisfied for the guess. Also logic says that anything that§2. Abelian C*-algebras and the Functional Calculus ae must be proved about the functional calculus can be proved by using its determining characteristics. There may be an occasion when the exact form of the functional calculus is useful in executing a proof, but my experience is that usually this precise form just gets in the way of understanding. The next, and final, result of the section is the answer to a fundamental question that should be addressed whenever there is a functional calculus. What is o(f(a))? 2.9 Spectral Mapping Theorem. If A is a C*-algcbra, a is a normal element in A, and f € C(o(a)), then (f(a) = f(o(a)). Proof. This is so simple it may confuse you. The map f — f(a) is a +-isomorphism from C(o(a)) onto C*(a). Thus o(f(a)) = o¢o(ay(f) = S(o(a)). = Exercises. 1. Prove a converse to Theorem 2.3 by showing that when X is a compact subset of the plane, C(X) is a C*-algebra with a single generator. 2. If A is an abelian C*-algebra with a finite number of generators Q},..-,@n, then there is a compact subset X of C” and a #-isomorphism p:A—C(X) such that p(a,) = 24, the k-th coordinate function. 3. If A is an abelian C*-algebra, show that A is separable if and only if its maximal ideal space is metrizable. (Theorem V.6.6 in [ACFA] has a more general result.) 4. A projection in a C*-algebra is a hermitian idempotent. If X is compact, show that X is totally disconnected if and only if C(X), as a C*-algebra, is generated by its projections. 5. Using the preceding exercise, show that for a o-finite measure space (X,9, n), the maximal ideal space of L(y) is totally disconnected. 6. If A is a unital C*-algebra and a € ReA, show that u = exp(ia) is unitary. Is the converse true? 7. Assume that (X,9, 2) is a finite measure space and ¢ is a bounded measurable function. Using the notation of Proposition 2.6, show that the essential range of ¢ is the support of the measure po ¢7). 8. Verify the statements in Example 2.8. 9. Let X be a compact space, ro € X, and define A = {{fa}: fn € C(X), sup, [Ifnll < 00, and {fn(zo)} is a convergent sequence }. If II{ fa} ll is defined as sup, || fn|| and the operations on A are defined entrywise, show that .A is an abelian C*-algebra and find its maximal ideal space.12 1. Introduction to C*-Algebras 10. For a completely regular space X, let C(X) be the C*-algebra of bounded continuous functions from X into C. Show that the maximal ideal space of Cy(X) is homeomorphic to X, the Stone-Cech compactification of X. §3. The positive elements in a C’-algebra In this section the functional calculus developed in the last section will be exploited to study a fundamental concept. the positive elements in a C*- algebra. 3.1 Definition. If A is a C*-algebra, an element a of A is positive if a € ReA and o(a) C Ry, the set of non-negative real numbers. This is denoted by a > 0 and Ay denotes the collection of all positive elements in A. Say that an element a is negative if —a € Ay. Write this as a < 0 and let A_ be the collection of all negative elements in A. It is easy to see that f is a positive element of the C*-algebra C(X) if and only if f(x) > 0 for all z in X. A function ¢ in L®(j2) is positive if and only if ¢(x) > 0 ae. [uJ]. Clearly the identity is a positive element of any C*-algebra. Also Ay A = (0). In fact, if a € Ay MA-, then o(a) = {0}. But a is hermitian so that |lal] = r(a) =0. 3.2 Proposition. Ifa € ReA, then there are unique positive elements u and v in A such thata = u-—v and uv = vu=0. Proof. Let f(t) = max(t,0) and g(t) = —min(t,0). So f,g € C(R)4 and f(t) — g(t) = t. Using the functional calculus for a, let u = f(a) and v = g(a). From the Spectral Mapping Theorem it follows that u,v € Ay. Also a = u—v and uv = vu = 0 since fg = 0 in C(a(a)). This proves existence. To prove uniqueness, assume u;, 0; € Ay satisfy a = uv; —v; and uv) = 1u) = 0. Note that this implies that u; and v, commute with a. Let {pn} be a sequence of polynomials such that p,(0) = 0 and p,(t) — f(t) uniformly on the spectrum of a. So pa(a) + f(a) =u in A. Because ua = auy, 11Pa(a) = pu(a)y for all n, so that uu = wy. Similar reasoning shows that the elements a,u.v.ui.v1 are pairwise commuting hermitian elements of A. let B = C*(a.u.v.u1.11). So B is abelian and, therefore, B = C(X) for some compact space X. Thus the proof of uniqueness for the general C*-algebra will be proved if it is proved for C(X).§3. Positive elements in a C*-algebra 13 So assume that a, u,v are continuous functions oy X having the relations listed above. It is left to the reader to show that for any x in X, u(r) = max{a(z), 0} and is, therefore, unique. Similarly, v(x) = —min{a(z),0}. = The elements u and v in the preceding proposition are called the positive part and the negative part of the hermitian element a and are denoted by u=a,andv=a_. The proof of the next result is similar and is left to the reader as an exercise. 3.3 Proposition. Ifa € Ay andn €N, there is a unique b in Ay such thata =". The positive element 6 in this proposition is called the n-th root of the positive element a and is denoted by b = a/". Now for some alternate characterizations of positive elements of a C*- algebra. Some of these are useful for verifying that examples are positive and others are only useful at a theoretical level. Observe that if f € C(X),, then | f(x) — t] < t for all real numbers t > |[f||. Conversely if f € C(X) and |f(z) — t| < t for some t > ||f||, then f(x) > 0 for all x in X. These observations will be used in the next proof. 3.4 Theorem. Jf A is a C*-algebra, the following statements are equiva- lent. (a) @20. (b) a= 0? for some b in Ay (c) a=2*z for some x in A. (d) @€ReA and ||t - al] < t for all t > |lal]. (e) @€ ReA and ||t — al] < t for some t > |lal]. Proof. Clearly (b) implies (c), and (d) implies (e). Proposition 3.3 says that (a) implies (b). . (e) implies (a). Since a is hermitian, C*(a) is abelian. Let. X = o(a), so that X C R. Transferring the inequality in (e) to C(X), there is a t > |lal] = sup|X| such that |t — z| <¢ for all z in X. As was pointed out before the statement of the theorem, this implies X¥ C Ry. (a) implies (d). This follows along the same line as the proof that (e) implies (a) and is left to the reader. Note that at this point, conditions (a), (d), and (e) are equivalent. This is used to prove the next proposition, which, in turn, is used to establish the14 1. Introduction to C*-Algebras remaining implication. No slight of hand here, just an economy of expres- sion. (c) implies (a). The condition that a = «*z for some z in A implies that aéReA. Let a = u—v as in Proposition 3.2 It must be shown that v = 0. Put zv'/? = b+ ic, bc € ReA. So (xv!/2)*(av!/2) = (b — ic)(b + ic) = b?+¢?+i(be—cb). But performing the computation the other way shows that (xv¥/?)* (aul?) = vM/2g*zyl/? = yl/2(u — v)vl/? = -v?, since uv = vu = 0. Hence i(bc — cb) = —v? — b? — c?. By the next proposition, i(be — cb) < 0. Also (xv!/?)*(xv!/2) = —v? < 0 by the Spectral Mapping Theorem. But the spectrum of (xv!/?)*(xv1/?) and (xv!/?)(xv!/)" differ by at most 0 (Ex- ercise VII.3.7 in [ACFA]). Thus (zv1/?)(zv"/?)* < 0. Put (xv'/?)(xv1/2)" = —y, where y € Ay. Hence —y = (b+ ic)(b— ic) = b? + c? — i(be — cb), so that i(be — cb) = y + b? + c? > 0 by the next proposition. Therefore i(be — cb) € Ay NA_ = (0). But this implies that —v? = (xv'/?)*(xv!/2) = b? +c? € Ay MA = (0). so that v? = 0. Since v? > 0, the uniqueness of the square root says that v = 0. 3.5 Proposition. A, is a closed cone in A. Proof. First show that Ay is closed. Let {an} be a sequence of positive elements and suppose |lan — a|| + 0. Clearly a € ReA. By condition (d) in the preceding theorem, |lan ~ [lan[||] < [lanl] for all n 2 1. Taking limits shows that |Ja — |Jalja|| < |lal|. By (3.4.e). a> 0. From the definition, aa > 0 when a € Ay and a € R,. Now let a,b € Ax. To show that a+b € Ax, it suffices to also assume that |Jall, [bl] < 1. Applying (3.4.d) shows that ||] — 3(a + 6)|] = }I|(1 —@) + (1-8)|| $1. By (3.4), $(a +b) > 0, so thata+bE Ay.» Because A, is a cone. Re A is a real ordered vector space. If a,b € Re A, say that a :: b if b— a > 0. This has a pleasing aspect in that the theory of ordered spaces can be applied. But usually it is easier to just prove what is needed about the ordering directly rather than appealing to the theory of ordered spaces. It is useful, especially when making certain estimates, to note that if'a € Re A. then —|jal] < @ < |ja||. In particular, if a is any element of A, then 0 < aa < |ja||?. The remainder of the section is devoted to an examination of positivity for the C*-algebra B(2) and to giving an important application to operator theory. 3.6 Proposition. If H is a Hilbert space and A € B(H), then A is positive if and only if (Ah, h) > 0 for every vector h.§3. Positive elements in a C*-algebra 15 Proof. If A > 0, then Theorem 3.4.c implies there is a T in B(#1) such that A=T"T. Thus (Ah, h) = ||Th||? > 0. Conversely, assume that (Ah, h) > 0 for all h in H. By Proposition 1.2.12 of [ACFA], A = A*. If \ < 0 and he, the fact that (Ah, h) > 0 implies II(A — A)h||? = |]Ah|)? — 24(Ah, h) + 0? I)? 2 —2A(Ah, h) + 7 |||? > |AIP. Thus \ ¢ oap(A), the approximate point spectrum of A. This says that A Dis left invertible. (See Exercise VII.6.5 of |ACFA].) Since A— ) is hermitian, it must also be right invertible, That is, ¢ ¢(A) and so A > 0. / 3.7 Definition. For any element a in a C*-algebra, define the absolute value of a by |a| = (a"a)/?, If a is hermitian, then it is not difficult to show that |a| Exercise 4. =a;,+a_. See 3.8 Definition. A partial isometry is an operator IW on a Hilbert space H such that ||WAl] = ||All for all vectors h in (ker W)+. The space (Iker 17’) is called the initial space of W and ranW is called the final space. The exercises at the end of this section contain a lot more information about partial isometries. Later, when the discussion turns to von Neumann algebras (Chapter 7), they will play a crucial role. 3.9 Polar Decomposition. If A € B(7), then there is a partial isometry W with initial space (ker A)+ and final space cl (ran A) such that A = W’|A| Moreover, if A= UP, where P > 0 and U is a partial isometry with ker U = ker P, then P =|A| and U =W. Proof. If h €H, then || Ah||? = (A*Ah, A) = (Ala, |Alh). Thus HAAl| = Ala tl for all vectors h. Thus permits the definition of IV : ran|A| — ran A by W(|Alh) = Aa. The preceding equation shows that WW is an isometry on ran A and, as such, can be extended to map cl [ran|A]] onto cl [ran A]. Extend 1’ to all of H by letting it be 0 on (ran|A|)+. This makes W a partial isometry and. clearly W|A| = A. It remains to check that it has the correct. initial space.16 1. Introduction to C*-Algebras That is, it is necessary to show that cl [ran|Al] = (ker A)4. If f € ran At, then f = A*g for some g in (ker A*)+ = cl[ran A]. Therefore ran A*A is dense in cl fran A*) = (ker A)4. But A*Ak = |A|?k = |A|h, where h = |Alk. Thus ran |A| is dense (ker A) as asserted. To show uniqueness, note that A"A = PU"UP. But U"U = E, the projection onto the initial space of U, (kerU)+ = (ker P)+ = cl [ran P). (See Exercise 10.) Thus A*A = P?, so that P = [Al by the uniqueness part of Proposition 3.3. If h € H, WIAlh = Ah = UlAlh. That is, U and W agree on a dense manifold in their common initial space. Hence U = W. = When reference is made to the polar decomposition of an operator, this means the unique decomposition obtained in the preceding theorem. Exercises. 1. Using the notation of Proposition 2.6 give necessary and sufficient conditions that Mz be positive. If M, > 0, find Mg!/". If Mg is hermitian, find its positive and negative parts. 2. Find an example of a positive operator that has a non-hermitian square root. Can you find one on a 2-dimensional space? 3. If {A,: 4 € T} is a collection of C*-algebras and A = @, Aj as in Exercise 1.4, show that an element {a,} in is positive if and only if a; > 0 for all i. 4. Show that if a € Re A. then |a] = a, +a_. 5. Show that each element a in ball ReA is the sum of two unitaries in A. (Hint: First show this when A = C.) 6. If a € Ay, show that a has a positive logarithm if and only if a is invertible. (a has a logarithm if there is an element b in A such that e = a.) Show that an invertible hermitian element has a logarithm. Is this logarithm hermitian? 7. Give an example of a C’-algebra A and two positive elements a and bin A such that ab is not positive. If ab = ba. show that ab € Ay. 8. Show that for a hermitian operator, the spectrum and the approxi- mate spectrum are the same. 9. Show that the following statements are equivalent. for an operator W’ in B(X). (a) W is a partial isometry. (b) W* is a partial isometry. (c) W"W is a projection. (d) WIV" is a projection. (e) WW"W = W. (f) Weviy = We. 10. If W’ is a partial isometry, show that IV*IV is the projection onto the initial space of 1” and IW'I¥"* is the projection onto the final space of W.§4. Approximate identities 17 11, For partial isometries W and We, define W1 < W2 to mean that WW, < WzW, WW} < WeW}. and Wyh = Woh for all vectors h in the initial space of W,. Show that < is a partial ordering on the set of partial isometries and that a partial isometry is a maximal element relative to this ordering if and only if either IV or W* is an isometry. 12. Using the terminology of the preceding exercise, show that the ex- treme points of ball B(H) are the maximal partial isometries. 13. Find the polar decomposition of the operator My from Proposition 2.6. 14. If the operator A has polar decomposition A = WA] and a € C, find the polar decomposition of A @ a. 15. For a Hilbert space H, define K = H ®H--- and define $:K = K by S(Aj,h2,...) = (0,h1,h2,...). What is the polar decomposition of 5? If A is a positive operator on H and T is defined on K by T(hy,ha,-.-) = (0, Ahj, Ahg,...), find the polar decomposition of T. 16. Define T : C™ — C” by Tex = ex41 for 1 < k < n—-1 and Te, = 0, where {€},...,€n} is the standard basis for C”. Find the polar decomposition of T. 17. Show that the parts of the polar decomposition of a normal operator commute. 18. If A € B(H), show that there is a positive operator P and a partial isometry W such that A= PW. Discuss the uniqueness of P and 1’. 19. If A is an invertible normal operator, show that the parts of the polar decomposition of A belong to C*(A). Give an example of a normal operator A for which the parts of the polar decomposition do not belong to C*(A). §4. Approximate identities Most of the C*-algebras encountered in this book will have an identity. There are many notable exceptions, the most prominent of which is the C*-algebras of compact operators examined in detail in §16 below. But it is necessary to consider ideals in a C*-algebra, and an ideal constitutes a C-algebra without identity. (See §5.) Of course, as seen, it is always possible to adjoin an identity, and this helps in many cases. In this section approximate identities are examined, and this often provides a better remedy than adjoining an identity. Let A be a C*-algebra and, in this section, assume that A does not. have an identity. Let 4; be the C*-algebra obtained by adjoining an identity to18 1. Introduction to C*-Algebras A. A consideration of A; = A+ C is not required for what is done below, but it is a notational convenience. 4.1 Definition. An approximate identity for A is a net {e;} in A such that: (i) 0 Se <1 for all §; (ii) e, < es when i < j; (iii) lim; ||ze; — zl] = 0 for all z in A. If the net is a sequence, call it a sequential approximate identity. First note that by taking adjoints, condition (iii) in the definition implies that the symmetric condition lim; ||e;z — || = 0 also holds for all x in A. A second fact to note is that if a net { ¢; } satisfies conditions (i) and (ii), then, due to its boundedness, to prove it is an approximate identity it suffices to show that it satisfies (iii) for z belonging to a total subset of A. If H is separable and { Py} is an increasing sequence of finite rank projec- tions that increases to the identity, then {P,} is a sequential approximate identity for Bo(H). (Note that for a non-separable Hilbert space, Bo(H) cannot have a sequential approximate identity.) Start with a simple lemma. (It might be mentioned that the displayed inequalities in this lemma illustrate the convenience of having an identity lurking in the background.) 4.2 Lemma. If A is a C*-algebra and a,b € A withO 0. Put c= d(a) +4(b) and d = ¥(c). Now for any e in €, (e) = e(1—e)"! € Z,. Thus c € Zy. Similarly, d=c(1+c)"! €Z. Because 0 < u(t) < 1 for all positive t, d € €. Now x = (a) (1 +¢)-!. Since (G(t)) = t, this implies that§4. Approximate identities 19 a= (2) =1-(1+2)-! < Y(c) =d. Similarly, b <‘W. Thus d is the sought for element of € that dominates both a and b, and € is directed. If x € Zy, let ay = Y(nz); so a, € E. Define n(t) = #2(1 — w(nt)) = (1+ nt)7) < t/n. Thus n(z) = 2?(1 — (nz)) = x(1 — an) and so ll2(Q1 = @n)zIl < [lnloce) $ llell/n- If € > 0 and n is chosen larger than e~!||z||, the preceding lemma implies that for e in € and e > ap, |x — rell? < €. That is li - =0 ti re ~ 2 whenever r € Zy. If r € A}, let {yn} be a sequence in Z such that lyn — 2'/2|| + 0. Thus tn = ytyn > @ and pq € Ty. Since € is bounded, the above limit holds for every x in A. But each element of A can be written as a linear combination of four elements in A}, so this limit is 0 for allzin A. = 4.4 Corollary. A separable C*-algebra has a sequential approximate iden- tity. Proof. Let {an} be a countable dense subset of A and let {e;} be an approximate identity. A simple induction argument shows that there are iy Sig S--+ such that |laze;— axl] < 27" for 1 < k ip. Letting Zn = €iqs it follows that for all k > 1, |laxtn — axl] + 0 as n — oo. Since {an} is bounded, |lazn — al] > 0 for all a in A. m In general, there is no polar decomposition in a C*-algebra. For example, if A = Bo(H) + C, every projection in A has finite dimensional range or kernel. It is left to the reader to find a compact operator K’ that cannot be expressed as K = W|K'|, where W is a partial isometry in A. What happens if the requirement that one of the factors be a partial isometry is relaxed? In this case such a factorization is possible using the techniques developed in this section. 4.5 Lemma. Let a and @ be positive scalars witha+f > 1. If A is a C*-algebra, x,y € A, a € Ay such that x*x < a® and y*y < a%, and un = a(n) + a)-1/?y, then {un} converges to an element u in A with Itul] < Jado*2-Dy, Proof. If dam = (n~} + a)7¥/? — (m=) + a)-¥/?, then20 1. Introduction to C*-Algebras [lin — uml? = |lzdnmyll? = [ly"dama*zdnmyll < [ly"dnma%damyll = la? damull? = la? damyy"dnma? || 2 2 o+8 S |la? dnma?dnma? || = IId2m2°*?l| = |Idnma 2 I. But {(n7? + 1)7/¢°S"} is an increasing sequence of functions on the a positive real axis. Thus {(n! + a)-1/2a°#"} is an increasing sequence in ‘Ay. Applying Dini’s Theorem from real variables shows that (nt + a)-1/2g#(o+9) a ghlet6-1) in A. Applying this to the above inequalities it follows that {un} is a Cauchy sequence in A. Let u = limp tu, € A. Using the same techniques as above gives that, Iltall = [la(n7? + a)" "/2yl] < l(a“? + a)“ a o*8) |) 7 |Jaz(o+8-0))). 4.6 Proposition. Ifa € A;. 2 € A such that x"z < a, and0 0, show that lim; ||A!/?E,A'/h — Ah|| = 0 for all h in H. §5. Ideals in a C*-algebra Start by examining the behavior of the functional calculus relative to a closed one-sided ideal. A will always denote a C*-algebra in this section. 5.1 Proposition. IfTZ is a closed left or right ideal of A,a € I witha =a*, and f € C(o(a)) such that f(0) = 0, then f(a) € Z. Consequently, for a hermitian element a of I, a4,a-,|a|, and |a|'/? € Z. Proof. Note that if Z is proper, 0 € (a). Since o(a) CR, if f € C(o(a)) and (0) = 0, there is a sequence of polynomials pp such that pa(0) = 0 and pn(t) + f(t) uniformly in o(a). Clearly pa(a) € Z for each n > 1, so f(a) eZ. 0 5.2 Definition. If A is a C*-algebra, then a hereditary subalgebra of Ais a C*-subalgebra B of A such that if a € By and x € A with 0 < x < a, then res. ‘The compact operators form a hereditary subalgebra of B(H). If P is a projection on H and B = {T € B(H): T = PT = TP}, then B is a hereditary subalgebra of B(H). The next proposition defines an order preserving bijection between the collection of all hereditary subalgebras of A and the collection of closed left ideals. 5.3 Proposition. Let A be a C*-algebra. (a) If L is a closed left ideal of A and B = LOL", then B is a hereditary subalgebra of A. (b) If Bis a hereditary subalgebra of A and £ = {xz€ A: 2*z € B}, then L is a closed left ideal of A. (c) If Lis a closed left ideal of A and B= LAL", then £L={ x: 2" € B}. (d) If B is a hereditary subalgebra of A and L = {x : 2*z € B}, then B=L£nL.22 1. Introduction to C*-Algebras Proof. (a) Clearly B is a C*-algebra. Let a € B, and suppose z € A with 0<2 a. =§5. Ideals in a C*-algebra 23 5.6 Theorem. Let A be a C*-algebra and let Z be a closed ideal of A. For a in A define (a+TZ)* =a" +I. With this definition and the quotient norm, A/T is a C*-algebra. Proof. From the general theory of Banach algebras, it is known that A/Z is a Banach algebra. It must be shown that A/T satisfies the C* identity. That is, show that ||a +Z||? = ||a"a+Z|| for all a in A. By Proposition 5.4, |la* + Z|] = lla + || for all a in A. Hence, ilove + ZI| = Ia" +Z)(a +2) $ lla” + Zilia + Zi = lla+ ZIP. On the other hand, the preceding lemma implies that lla + Z|]? = inf{ |la — az|l? : x € (ball Z) 4 } inf{ |]a(1 — 2)||? : & € (ball Z)4. } inf { |\(1 — )a"a(1 — z)|] : 2 € (ball Z), } < inf{ |Ja*a(1 — x)] : x € (ball Z) 4. } inf{ ja"a — a*az|) : x € (ballZ),} = |la*o+Z\|. 0 5.7 Corollary. If p : A — B is a +-homomorphism, then ran p is a C*- algebra, and the induced map j : A/ker p —+ ran p is a *-isomorphism. Proof. The only thing that needs to be proved is that ranp is closed. By Proposition 1.7, p is bounded. Thus Z = ker p is a closed ideal of A. Thus the induced map : J — B is a «monomorphism. By Corollary 1.8, 9 is an isometry, so that ran? = ran p is closed. = This section closes with an illustration of the preceding results for the case of an abelian C*-algebra. 5.8 Proposition. if X is a compact space and I is a closed ideal of C(X), then there is a unique closed subset F of X such that I = {f € O(X): f(x) =0 for allz in F}. The C*-algebra C(X)/T is isomorphic to C(F) by means of the map induced by the restriction map from C(X) to C(F). Proof. Let F = {x € X: f(x) = 0 forall f inZ}; clearly F is a closed subset of X. If w € M(X) and 1 Z, then for each f in Z. f|f[?du = 0 since |f|? € Z. Thus each f in Z must vanish on the support of 4. Hence24 1. Introduction to C*-Algebras suppu C F or |yu|(X\ F) = 0. Conversely, ifs € M(X) and |u|(X\ F) =0, then f f du = 0 for all f in Z. Therefore Z+ = {2 € M(X): |u|(X\F) = 0}. Duality in Banach spaces now implies that J = (I+), = {yu € M(X): lwl(X \F) =0}. ={f €C(X): f(x) =0 for all x in F}. Let p: C(X) — C(F) be the restriction map: p(f) = f|F. Clearly p is a homomorphism and ker p = Z. By the Tietze Extension Theorem p is surjective. Thus the induced map j: C(X)/I + C(F) is a +-isomorphism. . Exercises. 1. Show that My, the C*-algebra of all n x n matrices over the complex numbers, has no proper ideals. 2. If Ais a C*-algebra, I is a closed ideal of A, and B is a C*-subalgebra of A, show that the C*-algebra generated by [UB is I +B. 3. If A is a C*-algebra, let Z be the collection of all closed left ideals of A and let B be the collection of hereditary subalgebras of A. Define L: BTL by L(B) = {x € A: 2*z € B} and define B: L — B by B(C) = £AL*. Show that L and B are order preserving bijections and B= L-}, What are the fixed points of L and B? 4. If Ais a C*-algebra and J and J are closed ideals of A, show that 1 + J is a closed ideal of A. 5. Give an example of a non-closed ideal of C(clD) that is not self adjoint. 6. Find all the left ideals and hereditary subalgebras of Mn. §6. Representations of a C'-algebra A representation is just a +-homomorphism with a specific range. 6.1 Definition. If A is a Ct-algebra. a representation of A is a pair (7,14), where #1 is a Hilbert space and 7 : A — B(H) is a homomorphism. If A is unital, then it is required that (1) = The role here of 1 will often be suppressed and it will be stated that a is a representation. If A does not have an identity and A; = A+C, then whenever 7 : A — B(H) is a representation, it can be extended to a representation # of A; by letting #(a+a) = (a)+a for @ in A and a in C. Note that in light of Proposition 1.7.e. every representation is contractive. By Corollary 5.7, the range of a representation is closed. 6.2 Examples. (a) If A is a C*-subalgebra of B(H), then the inclusion map A — B(H) is a representation.§6. Representations of a C*-algebra 25 (b) If (X,Q,,) is a o-finite measure space, then Ri L™(u) > B(L?(u)) defined by (4) = Mg is a representation. (c) If X is compact and y is a positive Borel measure on X, then m, : C(X) — B(L?(x1)) defined by z,(f) = My is a representation of C(X). The notation set in part (c) of the last example will be used in the sequel. 6.3 Definition. Let A be a C*-algebra. (a) If d is a cardinal number and 4 is a Hilbert space, let 1 denote the direct sum of H with itself d times. If A € B(H), AM is the direct sum of A with itself d times. The operator A js called the d-fold inflation of A, or simply the inflation of A. If t : A — B(2) is a representation, then (4 > 4 — B(H() defined by 7((a) = x(a) is a representation. The representation (4 is called the inflation of m. For convenience, the inflations 10) and 7") are denoted by #{'©) and 1). (b) If {(m;,H;)} is a collection of representations of A, then the direct sum of these representations is the representation 7 = @,7 : A > B(@,; 1.) defined by 1(a) = Q, m(a). Note that in the definition of the direct sum of representations the fact that ||7:(a)|] < lal] for all i implies that the direct sum of operators @; ™(a) is a well-defined bounded operator. It is easily checked that if 7 is an isometry, then so is x. Also m : B(H) + B(H) defined by x(T) = T is an isometric representation of B(). 6.4 Example. Let X be a compact space and let {un} be a sequence of positive Borel measures on X. For each n > 1 define tm: C(X) + B(L?(stn)) as in Example 6.2.c. Then @, 7m is a representation. 6.5 Definition. Two representations of the C*-algebra A, (m,24:) and (m2, Ho), are equivalent if there is an isomorphism U: Hy + Ho such that 12(a) = Um(a)U~? for all a in A. In symbols this is denoted by m * 72. (The terminology of [ACFA] is maintained here. An isomorphism between Hilbert spaces is an isomorphism in that category. That is, an isomorphism is an inner product preserving linear bijection. The term “unitary” is re- served for an isomorphism from a Hilbert space onto itself.) 6.6 Example. (a) In Example 6.4, if the measures {jin}: are pairwise sin- gular and »p = 5>,,(2"[unll)~? un, then the representation 7, : C(X) — B(L?()) is equivalent to the direct sum of the representations {7,,, }. (b) Let p and v be two positive measures on the compact space X and mt, and 7, the two representations defined in Example 6.2.c. Then 7, & m, if26 1. Introduction to C*-Algebras and only if » and v are mutually absolutely continuous; that is, if and only if the two measures have the same sets of measure zero To see part (b), first assume that that the measures are mutually abso- lutely continuous. If ¢ is the Radon-Nikodym derivative du/dv, then define U: L(y) + L(v) by Uh = hy@. It is left to the reader to check that U is the required isomorphism. For the converse, assume that an isomorphism U : L?(u) + L?(v) is given such that for each f in C(X) and each h in L?(u), U(fh) = fUh. Letting J = U(1), this implies that U(f) = U(f-1) = fv for each f in C(X). Since U is an isometry, f |fPlv/?dv = f|f|? dp. But this implies that f f|v|? dv = f f du for every positive continuous function fon X. Thus p = |w|?v and so p << v. The same argument applied to the isomorphism U~} shows that v << p. This high ratio of definitions to results must continue a bit longer. 6.7 Definition. A representation 7 of a C*-algebra A is cyclic if there is a vector e in H such that {(a)e: a € A} is dense in H. Any vector e that satisfies this condition is called a cyclic vector for the representation . The representation 7 is called faithful if it is injective. A subrepresentation of 7 is a representation of A of the form a — 7(a)|M. where M is a closed subspace of 1 that reduces (A). It is easy to see that if X is compact and y is a positive Borel measure on X, then m,. the representation of C(X) defined in (6.2.c), is cyclic with cyclic vector 1. In fact. any function in L?() that does not vanish on a set of positive jr measure is cyclic for this representation. (See Exercise 1.) Also the identity representation of B(H) is cyclic with any non-zero vector as a cyclic vector. It is also easy to see that the representation 7, is faithful if and only if the support of yz is all of X. Note that a faithful representation is an isometry. There is more about faithful representations in the next section. The reader might find some subrepresentations of the representations in Examples 6.2.b and 6.2.c. 6.8 Theorem. Every representation is equivalent to the direct sum of cyclic representations. If the representation acts on a separable Hilbert space, the representation is equivalent to the direct sum of a sequence of cyclic repre- sentations. Proof. The proof is a rather simple extrapolation from an elementary ob- servation. Fix a representation s : A — B(H). For any vector e in H, put M = [r(A)e]. (For any collection S of vectors in a Hilbert space, [S] denotes the closed linear span of the set S.) Since A is a C*-algebra, M is invariant for every operator (a) as well as its adjoint. That is, M reduces§6._ Representations of a C*-algebra 27 (A). So if m: A — B(M) is the subrepresentation (a) = m(a)|M and 172: A B(M?) is defined by 72(a) = 1(a)|M+,‘r and mp are representa- tions and 7 is cyclic. Now use Zorn's Lemma to obtain a maximal family of vectors {e;} such that [r(A)ei] 1 [(A)e,] for i # j. If Hs = [r(A)ey], then maximality implies that H = @; Mi. If m(a) = 7(a)|7,, then 7 = Q, 73. If H is separable, this family must be countable. = A natural question to ask is whether a representation can be extended from a C*-subalgebra of the C*-algebra A. The answer is no. For example consider the C*-algebra of all n x n matrices, M,. Let B be the subalge- bra consisting of all diagonal matrices and define p : B — C = B(C) by p(diag(a1,...,an)) = oy. It is left to the reader to show that p is a rep- resentation. But there can be no representation « : M, — C other than the 0 representation. In fact, if there were, then ker « is a proper ideal in Mn, a contradiction (Exercise 5.1). Therefore p has no extension. If the representation is defined on an ideal of A, however, the story has a different conclusion. 6.9 Definition. If A is a C*-algebra, say that a representation p : A B(H) is non-degenerate if [p(A)H] = 1 For any representation p, [p(A)H] is a reducing subspace for p(A) and this is where the representation “lives.” So to say that a representation is non-degenerate is to say that it is living on the entire Hilbert space and not in some corner. Equivalently, a representation p : A — B(H) is non- degenerate if the only vector g in H satisfying Ag = 0 for every A in A is g =0. Also note that if p is a representation of A and p = Q, p,. then p is non-degenerate if and only if each p, is non-degenerate. 6.10 Proposition. If A is a C*-algebra and Z is an ideal of A, then every representation p : I + B(H) can be extended to a representation p : A — B(H). If the representation p is non-degenerate, the extension p is unique. Proof. First assume that p is non-degenerate. Claim. For every @ in A there is a unique (a) in B(71) such that for every z in Z, A(a)p(x) = p(az). By means of Theorem 6.8, it suffices to prove the claim for a cyclic representation. So assume p is cyclic with cyclic vector ho. Fix a in A and let {e;} be an approximate identity for the ideal J. Thus for every x in I28 4. introduction to U*-Algebras le(azx)holl = lim |l(aeiz)ho|l = lim [ip(ae,)o(z)hol S sup [lacs llo(z)holl S lle o()roll- This shows that p(x)ho — p(az)ho is a well defined bounded operator on a dense manifold in H. Extend this to a bounded operator on all of and denote this operator by p(a). If x,y € Z, then p(a)p(x)p(y)ho = Ala)p(xy)ho = p(axy)ho = p(azr)p(y)ho. Since H = [p(Z)ho], this shows that A(a)p(xz) = p(az) for all x in Z and establishes the existence part of the claim. To show the uniqueness statement in the claim, observe that if T is any other operator on # such that Tp(x) = p(ar) for all x in Z, then Tp(z)h = p(a)p(x)h for all x in Z and all h in H. Since p is non-degenerate, T =7(a). Therefore there is a well defined map p: A — B(H). Ifa,b€ A, x € TZ, and h € H, then p(ab)p(x)h = p(abr)h = pla)p(bx)h = p(a)p(b)p(x)h. Thus 2(ab) = B(a)p(b). Similar arguments show that 3 is a representation of A and it is clear that it extends p. Now assume that p is degenerate. Put Ho = [o(Z)H] and let po: I + B(Ho) be the subrepresentation po(x) = p(z)|Ho. Since po is non-degenerate, there is a unique extension fo : A > B(Ho). If: A > B(H@) is any repre- sentation (like the 0 representation, for example). then fo @x is an extension of p. # It’s easy to see from the preceding proof why uniqueness fails unless the representation is non-degenerate. The proof also identifies all the possible representations in the degenerate case. Since cyclic representations are non- degenerate, the following is immediate. 6.11 Corollary. If A is a C*-algebra and I is an ideal of A, then ev- ery cyclic representation p : I + B(H) has a umigue extension to a cyclic representation of A 6.12 Corollary. If A is a C*-algebra, I is an ideal of A. and (p,H) and («,K) are two representations of A such that: (a) pIZ is a non-degenerate representation of the ideal T: (b) plZ = «IZ; then pr§7. Positive linear functionals and the GNS construction 29 Proof. The hypothesis implies there is an isomorphism U : # — K such that U*x(x)U = p(z) for all x in Z. Now p(-) antl U*x(-)U are both ex- tensions of p|Z. By Proposition 6.10 and condition (a), p[Z has a unique extension to A. Thus U*x(a)U = p(a) for all a in A, so that p& x. Exercises. 1. Show that the representation in Example 6.2.b is cyclic and that a function f in L?(,) is a cyclic vector for this representation if and only if it does not vanish on a set of positive ys measure. 2. Let X be a compact space, p a positive Borel measure on X, and (7, L?(u)) the representation from Example 6.2.c. Determine the kernel of ty. If X is metrizable, show that the measure pz can be chosen so that ™, is injective. 3. Consider the two positive Borel measures y and v on the compact. space X and the representations 7, and 7, defined in Example 6.2.c. Show that 7m, is cyclic if and only if Lv. If 1 v, show that 7, @my ¥ yy: Note that this implies that 7! is not cyclic for d > 2. Can you show this directly? 4. Verify the statements in Example 6.6. 5. Let A be any C*-algebra contained in B(H) that contains the compact operators. If 7: A — B(H) is the identity representation, show that 7°) is a cyclic representation. 6. Let p : A — B(H) be a representation and let I be an ideal of A. Define Hz = [p(Z)H] and let pz: I > B(Hz) by pz(z) = p(x)|Hz. Show that pz is a non-degenerate representation of Z. What is its extension? §7. Positive linear functionals and the GNS construction Suppose A is a unital C*-algebra and 7: A — B(H) is a representation. If is a unit vector in H and $: A — C is defined by @(a) = (m(a)e,e), then ¢ is a positive linear functional. That is, ¢(a) > 0 when a € Ay. Also (1) =1 since |lel] = 1. 7.1 Definition. A state on a C*-algebra is a positive linear functional of norm 1. Let E = E(A) be the collection of all states on A. D(A) is called the state space of A. A state on a C*-algebra without identity is extended to the algebra obtained by adjoining an identity if we set #(a + a) = ¢(a) +a. The paragraph preceding the definition shows that whenever there is a representation, this produces a state. The main result of this section is30 1. Introduction to C*-Algebras the converse. Whenever a state is present, it is possible to manufacture a representation. Indeed this is done in such a manner that these two processes are the inverses of each other. The first step in carrying out this plan is deriving some elementary properties of positive linear functionals. The first is that the Cauchy-Schwarz~Bunyakowskii Inequality holds. 7.2 Proposition. If ¢ is @ positive linear functional on a C*-algebra A, then for any x,y in A I(y°2)P < d(y"y)o(e" Proof. If [z,y] = ¢(y*z) for x,y in A. then it is easily checked that [-,-] is a semi-inner product on A. Thus the inequality follows from the CBS Inequality. = i ° The preceding inequality is also called the CBS Inequality. 7.3 Corollary. If ¢ is a positive linear functional on A. then ¢ is bounded. If {es} is an approximate identity for A. then |jol] = lim; 4(es). Proof. The proof of the case where A is unital is left to the reader; it follows from the CBS Inequality. Note that in this case ||¢|| = (1). Assume A does not have an identity and let {e;} be an approximate identity. First show that ¢ is bounded on (ball.A)+. If not. then for every integer k > 1 there is an a, in (ball), with (ay) > 2%. Ifa = ,27*ay, then (a) > Oh 27*ax) = DRL, o(27*a,) > n. an impossibility. Thus a = sup{ (a) : a € (ball.A)4 } < 00. Since each element of A can be written as the linear combination of 4 positive elements, o is bounded with ||¢|| < 4a. Note also that a > sup, d(e,) = lim; ¢(e:) = 8, since the approximate identity is increasing. Now if a € ballA, then 0 < a*a < 1 and so |d(a)|? = lim |¢(e,2)|? < lim 4(€?) o(a"a) < BI|d})- This imnplics that [|¢| < B 0. * Proof. If A = C(X), then ¢ corresponds to a measure : on X and the hypothesis is the statement that 4(X) = |lu||. This implies that 2 > 0. For an arbitrary C*-algebra A , let a € Ay and consider C = C*(a). So C = C(o(a)). If do = SIC, then ¢o(1) < |looll < IIdll = (1) = go(1). So the result from measure theory implies that ¢(a) > 0. Hence ¢ is positive. = When mention is made of a “subalgebra” of A and the subalgebra has an identity, then it is assumed that it has the same identity as A. However every subalgebra of an algebra with identity need not have one. 7.6 Corollary. If A is a C*-algebra and B is a C*-subalgebra of A, then any state on B has an extension to a state on A. Proof. Without loss of generality it may be assumed that A has an identity. Let ¥: B— C be a state. If B has an identity, let ¢ be a norm preserving extension to A. Then 1 = |/¢l] = [vl] = ¥(1) = (1). Thus ¢ is a state on A by the preceding proposition. If B does not have an identity, let By = B+C, a subalgebra of A. If vy : By) — C is defined by 4u(b-+ 8) = ¥(b) +8, then 1), is a state on By, and the first part of the proof applies. = Now for the promised converse to the observation made at the beginning of the section. Before stating and proving the result, examine what happens when A is abelian. If ¢ ; C(X) — C is a state, then there is a probability measure 1 such that ¢(f) = f f dy. It seems clear that the representation that should be associated with the state ¢ is the representation m,, : C(.Y) — B(L?(u)). Note that L?() is the completion of C(X) with respect to the norm defined by the inner product (f,9) = J f9du = $(f9). This furnishes the idea of how to prove the next theorem. The reader’s comprehension might benefit by examining for the abelian case each step of the next proof. 7.7 The Gelfand-Naimark-Segal Construction. Let A he a C*-alge- bra with identity. (a) If ¢ is a state on A, then there is a cyclic representation (m,He) of A with unit cyclic vector e such that $(a) = (mg(a)e,e) for alla in A. (b) If (7, H) is a cyclic representation with unit cyclic vector e, @ is the state on A defined by ¢(a) = (x(a)e,e), and (74, Hy) is the representation obtained in (a), then 7 ¥ 7. Proof. Let £ = {x € A: o(2"z) = 0}. It is easy to see that L is closed in A. Also if € A and z € L, then 0 < ¢((az)*(az)) = ¢(z*a"az) <