Banach Spaces of Analytic Functions and Absolutely Summing Operators (Regional Conference Series in Mathematics ; No. 30)

conference board of the mathematical sciences regional conference series in mathematics number

30

Aleksander Pefczynski BANACH SPACES of ANALYTIC FUNCTIONS and ABSOLUTELY SUMMING OPERATORS

supported by the national science foundation published by the american mathematicai society

Conference Board of the Mathematical Sciences REGIONAL CONFERENCE SERIES IN MATHEMA TICS

supported by the National Science Foundation

Number 30

BANACH SPACES OF ANALYTIC FUNCTIONS AND ABSOLUTELY SUMMING OPERATORS Aleksander Petczynski

Published for the Conference Board of the Mathematical Sciences by the American Mathematical Society Providence Rhode Island I

Expository Lectures from the CBMS Regional Conference held at Kent State University July 11-16, 1976 AMS (MOS) subject classifications (1970). Primary 46E15, 46J15; Secondary 30A98, 32E25.

Library of Congress Cataloging-in-Publication Data

Pelczynski, Aleksander. Banach spaces of analytic functions and absolutely summing operators. (Regional conference series in mathematics, no. 30) "Expository lectures from the CBMS regional conference held at Kent State University, July 11-16, 1976." Includes bibliographical references. 1. Banach spaces. 2. Banach algebras. 3. Operator theory. 4. Holomorphic functions. I. Conference Board of the Mathematical Sciences. II. Title. III. Series. QAl.R33 no. 30 [QA322.2] 51O'.8s [515'.73] ISBN 0-8218-1680-2 77-9884

Copying and reprinting. IndIvidual readers of thIS pubhcation, and nonprofit hbranes acting for them. are permItted to make fair use of the material, such as to copy an artIcle for use In teachIng or research. PermIssion is granted to quote bnef passages from thIS publication in reviews. provIded the customary acknowledgment of the source is gIVen. Repubhcation, systematIc copyIng. or multIple reproduction of any matenal In this pubhcation (including abstracts) IS permitted only under hcense from the American MathematIcal SocIety. Requests for such permIssIOn should be addressed to the Executive Director, Amencan Mathematical Society. P.O. Box 6248, Providence. Rhode Island 02940-6248 The owner consents to copyIng beyond that permltted by SectIOns 107 or 108 of the U.S. Copyright Law. provided that a fee of $1.00 plus $.25 per page for each copy be paId dIrectly to the Copyright Clearance Center. Inc, 27 Congress Street. Salem, Massachusetts 01970. When paying thIS fee please use the code 0160-7642/90 to refer to thiS pubhcatlOn. ThiS consent does not extend to other kinds of copyIng, such as copyIng for general distnbutlOn, for advertisIng or promotIOnal purposes, for creating new collective works, or for resale Copynght

© 1977

by the Amencan MathematIcal SOCIety. All nghts reserved. Pnnted In the Untted States of Amenca The Amencan Mathematical SocIety retaInS all nghts except those granted to the United States Government. The paper used In thiS book IS aCId-free and falls WIthIn the guidelmes established to ensure permanence and durabIlity. €I 109876543

95 94 93 92 91 90

CONTENTS 1

Preface O. Preliminaries . • . . . . . . . • . . • . • • • . . . . • • • . . • • • . . . • • • • •

0.1. e(/1)-spaces and C(S)-spaces. • . . . . • • • • • • . • . . . . • . • • • . 0.11. Classical spaces of analytic functions and the Hilbert transform. . • 0.I1I. Absolutely summing operators and their relatives. . . • • • . . • • • 1. The F. and M. Riesz Theorem and Duals of the Disc Algebra. . . • • . • • 2. Absolutely Summing Operators from the Disc Algebra. . • • . • . • . . . • 3. Absolutely Summing Operators from the Disc Algebra into HJlbert Space. 4. The Nonexistence of Local Unconditional Structure for the Disc Algebra and for its Duals. • • • . • . • . . • • • . . . • . . • . . . . . • . . . . . • . . . 5. Application to Uniform Algebras. • • • • • • . . • . . . • . • • . . . . • • • . 6. Uniformly Peaking Families of Functions in A and H~. The Havin Lemma. 7. Characterizations of Weakly Compact Sets in L I /H~ and in A *. • • • . • . 8. Weakly Compact Operators from A, L I /H~ and A * and Complemented Subspaces of These Spaces. . . • • . • . . . . • • . • • . • . . • • • . • • . • • 9. Complementation of Finite Dimensional Subspaces in A, L I /II~ and H~. 10. Bases and the Approximation Property in Some Spaces of Analytic Functions. . • . • • • . . • • • • • . • • . . • • • • • • • • . . • • . . • . . • • • • • 11. The Polydisc Algebra and the n-Ball Algebra, and Their Duals. . • • . • • References •• . . . • • • • • • • . • . . • • . • . • . • • . • • • • • • • • • • • . • . . • . • •

3 3 4 7 10 13 19 24 28 36 43 50 56 65 71 85

Other Monographs in this S('ril's

No. 1. Irving Kaplansky: Algebraic and analytic aspects of operator algebras 2. Gilbert Baumslag: Lecture notes on nilpotent groups 3. Lawrence Markus: Lectures

differentiable dynamics 4. H.S.M. Coxeter: TWisted honeycombs In

5. George W. Whitehead: Recent advances in homotopy theory 6. Walter Rudin: Lectures on the edge-of-the-wedge theorem 7. Yozo Matsushima: Holorr,orphlc vector fIelds on compact Kahler ma7lifolds 8. Peter Hilton: Lectures in homological algebra 9. I. N. Herstein: Notes from a rmg theory conference 10. Branko Grunbaum: Arrangements and spreads 11. Irving Glicksberg: Recent results on functIOn algebras 12. Barbara L. Osofsky: Homological dlme7lSIons of modules 13. Michael Rabin: Automata on infmlte obJects and Church's problem 14. Sigurdur Helgason: AnalYSIS on Lie groups and homogeneous spaces 15. R. G. Douglas: Banach algebra techniques m the theory of Toeplitz operators 16. Joseph L. Taylor: Measure algebras 17. Louis Nirenberg: Lectures on lmear partial di//erentlal equations 18. Avner Friedman: Dlf/erentlal games 19. B~la Sz.-Nagy: Umtary dzlations of Hilbert space operators and related topics 20. Hyman Bass: Introduction to some methods of algebraic K-theory 21. Wilhelm Stoll: Holomorphic functions

0/

fimte order m several complex variables

22. O. T. O'Meara: Lectures on linear groups 23. Mary Ellen Rudin: Lectures on set theoretic topology 24. Melvin Hochster: TopIcs m the homologIcal theory

0/ modules

over

commutatIVe rmgs 2'1. Karl. W. Gruenberg: RelatIOn modules

0/

fmlte groups

26. Irving Reiner: Class groups and Picard groups of group rmgs and orders 27. H. Blaine Lawson, Jr.: The quantltotlVe theory 0/ /olzatlons 28. T. A. Chapman: Lectures on lIz/hert (uhe mam/olds 29. Alan Weinstein: Lectures on symplectic mamfolds 30. Aleksander Pdczydski: Banach spaces of analytic functzons and absolutely summing operators 31. Ronald R. Coifman and Guido Weiss: Trans/erence methods in analYSIS 32. Wolfgang M. Schmidt: Small fractional parts of polynomials

Preface This is a revised version of my ten-lecture marathon at the American Mathematical Society Regional Conference "On Banach spaces of analytic functions" held at Kent State University, July 11-16,1976. It is based on the course which I taught at the Ohio State University during the fall and winter quarters of the academic year 1975-1976. In the past, the "classical" spaces of analytic functions, the disc algebra A and the Hardy spaces HP, have been studied mainly from the point of view of the theory of Banach algebras and harmonic analysis. However, several results obtained more recently can be naturally formulated in the language of general Banach spaces and operator theory, although they are still closely related to some questions in hard analysis and often use analytic tools. An example is the Fefferman-Stein duality between BMO and HI (cf. [F-S]).

Another one

is Henkin's theory of analytic measures for spaces of analytic functions of several complex variables in strictly pseudoconvex domains (cf. [He3)).

Also the techniques of p-absolutely

summing operators and related ideas which have been recently the object of intense study in the theory of Banach spaces seem to offer a new approach to spaces of analytic functions. To explain this, note that the natural injection A --+ HP is an example of a p-absolutely summing operator, and the fact that this operator is p-integral for p

>I

is equivalent to the

M. Riesz theorem on the boundedness in LP -norm of the orthogonal projection from LP onto HP _ More generally considering the completion in LP(J.1)-norm of the uniformly continuous holomorphic functions in a fixed bounded domain of holomorphy, where J.1 is a measure on the boundary of the domain, is nothing more than studying certain properties of a certain p-absolutely summing operator: "natural injection". A more sophisticated example is a map [--+ (/(2 n says that

»;=

I

from A into

[2.

An inequality discovered by Paley around 1932

for [EA. It means exactly that the map in question is absolutely summing. Paley's inequality depet}ds heavily on the analyticity of f. There is no counterpart for continuous functions on the circle and this reflects the fact that every absolutely summing operator from a C(S)-space into a Hilbert space is nuclear, and therefore compact. This produces a linear topological invariant which enables us to show that the disc algebra (and more generally a uniform algebra with a nontrivial Gleason part) is not isomorphic, as a Banach space, to any quotient of a C(S)-space (cf. § §4 and 5). On the other hand, Paley's inequality can be applied to prove the basic fact in the theory of absolutely summing operators, discovered by Grothendieck, that every operator from an L I (J.1)-space to a Hilbert space is absolutely summing (cf. §3 for details). Some words about the organization of the paper: Copynght © 1977, Amencan Mathematical Society

1

ALEKSANDER paCZYNSKI

2

§ §O and I have preliminary character. § § 2 and 3 are devoted to studying properties of p-absolutely summing operators from the disc algebra. The main result proved there says that if p > I then every p-absolutely summing operator from the disc algebra is p-integral. § §4 and 5 should convince the reader that the disc algebra is a pathological space from the point of view of Banach space theory, or at least that it is very different from C(S)-spaces, and that the same is true for a large class of uniform algebras. On the other hand in § § 7 and 8 we show that the disc algebra shares various properties of C(S)-spaces like the DunfordPettis property, characterizations of weakly compact operators, weak sequential completeness of the dual, etc. The basic analytic tool which leads to these results is the Havin lemma which is presented in §6. Roughly speaking, the Havin lemma is a quantitative analytic refinement of Fatou's theorem that closed sets in the unit circle, of Lebesgue measure zero, are the peak sets for the disc algebra. In §9 we apply the main result of §2 to study asymptotic behaviour of norms of projections from the disc algebra onto its finite dimensional subspaces as the dimensions of these subspaces tend to infinity. We prove that the dual of the disc algebra has Rademacher cotype p for some p

<

00.

The crucial question, whether

this dual has cotype 2, still remains open.

§ 10 is loosely related to other sections. It is a survey (without proofs) of results concerning bases and various approximation properties in the classical spaces of analytic functions. In the last § II we consider the problem of isomorphic classification of spaces of analytic functions of different numbers of variables. Most of the section is devoted to an elementary presentation of Henkin's generalization of the M. and F. Riesz theorem for the ball algebras. The material presented here mostly concerns the disc algebra and its generalization to several complex variables. The Banach space properties of other HP spaces are almost untouched. The reader interested in this subject is referred to the excellent survey by Coifman and Weiss [C-W] and to the paper [Kw-P]. Finally I would like to express my gratitude to everybody who helped me to complete this work: to Joseph Diestel who got the idea of organizing the Regional Conference in Kent and who was able to transform his dream into reality, to Peter Q)rno whose permanent interest in the subject and sharp criticism constantly stimulated my work; I am particularly indebted to him for permission to include several of his unpublished results, to W. J. Davis, G. A. Edgar, T. Figiel, W. B. Johnson, G. Pisier and P. Wojtaszczyk who are responsible for many improvements in the text. My sincere thanks go to Pam Walsh for her typing job and patience in preparing the manuscript. I would like to also acknowledge the Ohio State University and the National Science Foundation for their financial support. ALEKSANDER PEt,CZYNSKI POLISH ACADEMY OF SciENCES OHIO STATE UNIVERSITY

o.

Preliminaries

Our terminology and notation for Banach spaces and linear operators is standard; it is similar to that employed by Dunford and Schwartz [D-SI], and Lindenstrauss and Tzafriri

[L-T2]. 0.1. LP(J.1}spaces and C(S}spaces. Let J.1 be a nonnegative (not necessarily sigma finite) measure defined on a sigma field of subsets of a set D.. Let 0 < p ,.;;; 00. By LP(J.1) (resp. L~(J.1)) we denote the space of J.1-equivalence classes of complex-valued (resp. real-valued) functions f such that IIfllp

<

where

00,

In

Ifl P dJ.1

for 0

< p < 1,

for 1 ,.;;; p ess sup If(s) I

for p =

<

00,

00.

sES

For 1 ,.;;; p ,.;;;

00,

LP (J.1) under the norm II . lip is a Banach space.

By C(S) (resp. CR(S)) we denote the Banach space of all the continuous complex-valued (resp. real-valued) functions on a compact Hausdorff space S with the norm Ilfll = sUPsEslf(s)l. Given an f E C(S) we define the functions If I and 1 by Ifl(s) = If(s)I,1(s) = f(s) for s E S. The constant functions are identified with the scalars. We identify the dual space [C(S)] * via the Riesz representation theorem with the space of all complex Borel measures on S with the norm 11J.111 = the total variation of J.1. We put [C(S)] = {J.1 E [C(S)] *: J.1 nonnegative}. A J.1 E [C(S)] with J.1(S) = 11J.111 = 1 is called a probability measure. Given a J.1 E [C(S)] and a v E [C(S)] * we write v «J.1 if v is absolutely continuous with respect to J.1, and v 1 J.1 if v is singular with respect to J.1. Let J.1 E [C(S)] Then there is a nat ural isometric isomorphism from L 1 (J.1) onto the subspace {v E [C(S)] *: v «J.1} which assigns to each gEL 1 (J.1) the measure v defined by

+

+

+

+.

(0.1)

v(A) =

L

g dJ.1

for every Borel set A C S.

Given v «J.1, the unique g satisfying (0.1) is denoted by dv/dJ.1 and it is called the RadonNikodym derivative of v with respect to J.1. For a v E [C(S)] * we denote by Ivl the unique element of [C(S)]

t

such that v

nally given a v E [C(S)]

* and

«

Ivl and Idv(s)/dlvll

= 1 lvi-almost everywhere on S.

agE L 1 (Iv!) we denote by g . v the unique measure in

[C(S)] * whose Radon-Nikodym derivative with respect to Ivl is g . (dv/dlvl)-l .

3

Fi-

ALEKSANDER PEl.CZyNSKI

4

OJ!. Classical spaces of analytic functions and the Hilbert transform. General references to this part are the books of Duren [Du) , Hoffman [H), and the treatise of Zygmund [Z). In the sequel C-denotes the complex plane, D = {z E C : Izl .;:;; l}-the closed unit disc, "OD = {z E C: Izl = l}-the unit circle = the boundary of D, m-denotes (except §11) the normalized Lebesgue measure on "OD, LP = LP(m) and L~ = L~(m) (0 < p .;:;; 00), A -stands for the Disc Algebra which is defined to be the smallest closed (in the topology of uniform convergence) subspace of C("oD) which contains all the polynomials in z, P(z) = I;~=OCkZk (ck-arbitrary complex numbers, k

< 00 we

= 0,

1, ... ,n; n

= 0,

1, ... ).

J!P to be the subspace of LP consisting of all the functions f which are limits of a sequence of polynomials (Ps ) in the norm II . lip' For 0 .;:;; p

i.e., limsfaDlPs - flP dm For p = 00 we put

define the Hardy space

= O.

f i = {fE L 00: f'lDf(z)zn m(dz) = 0

for n

= 1,2,

... }.

Next we put

Ao

= zA =

{f E A: f

J--Ig = zlfP =

= zg

{f E 1fP: f

for some g E A},

= zg

for some g E HP}.

(Here z denotes the identity function on "oD.) Obviously A and HP for 1 .;:;; p .;:;; 00 are complex Banach spaces; if 0

1, then LP

and fJP are complete complex linear metric spaces. Clearly A can be identified with a closed linear subspace of f i via the map which assigns to each f E A its m-equivalence class. Let g be an analytic function in the open unit disc D\"oD such that for m-almost all

z E "OD there exists the radial limit, limrt 19(rz). Then the measurable function z limrt Ig(rz) (defined m-almost everywhere on "oD) is called the boundary value function of g. Recall (cf. [Du, Chapters 2 and 3) that every f E IfP (0 < p .;:;; 00) is a boundary value function of a unique analytic function in D\3D; which is called the analytic extension of f; we denote this function, unless otherwise stated, also by f. It satisfies the inequality (0.2)

sup

MpCr, f)

<

00

O
where Mp(r, f) = faD If(rz)I P m(dz) for 0 < p < 00 and Moo(r, f) = ess sUPzEaD If(rz)l. Conversely, if f is an analytic function in D\3D which satisfies (0.2) for some p with

o < p .;:;; 00, then An

f

E

there exists a boundary value function of f and it belongs to HP.

A iff f extends to a continuous function on the disc D which is analytic in

D\"oD. For fE LI we put

BANACH SPACES

fen)

=

If f E HI and if we identify

fan f(z)z-n m(dz) f

(n

5

= 0,

± 1, ... ).

with its analytic extension, then

(0.3)

/(n)

for n

fn)(O)

=

n!

< o.

Clearly if f E HI , then fen) = 0 for n

~ O.

The converse implication is also true; it follows

from 0.1

THEOREM

Iff E LP (l ~ P

(THE FEJER THEOREM).

f

lim n an

fez)

f n k=-n+

then

P

_.l

For the proof cf. [H, p. 23].

< 00),

(n - Ikl)/(k)zk

m(dz) =

o.

I

0

We introduce now the operator H called the Hilbert transform which plays the crucial role in the study of "classical" spaces of analytic functions. For u E L

1 we define the Hilbert transform of u to be the unique function v =

+ iv is a boundary = Janu dm and

such that u

f(O)

Re f(re i {})

= (21T)-1

f

value function of an analytic function

27T l-r2 0 u(e i {}) - - - - - - - 2 r cos {} r2

+

d{}

f

H(u)

in D\aD such that

for re i {} E D\aD.

For gEL I the Hilbert transform H(g) is defined by

M(g) = H(Re g) where Re g

+ i H(Im g)

= \,1(g + g). Img = (g - i)/2i.

The next result summarizes the most important properties of the Hilbert transform (for the proof cf. [Du, Chapter 4]).

0.2. (i) If 00 > p > 1, then H(f) E LP whenever f E LP; moreover there exists an absolute constant a independent of p such that THEOREM

IIH(!)llp ~ a

p2

p-=r IIfllp

for every f E LP

and

IIH(f)llp

> a-I

p2

p=-IlIfllp

for some f= fp E LP.

(ii) (the Kolmogorov theorem) Iff ELI, then H(!) E LP for every p there exists a constant Cp such that

IIH(f)lI p

~ Cpllfll~

< 1.

Hence

for every f ELI.

(iii) The operator f - H(f) is of weak type (1-1), i.e., there is an absolute constant CI such that for every K > 0

ALEKSANDER PEt.CZyNSKI

6

O.III. Absolutely summing operators and their relatives. General references to this part are the papers by Grothendieck [Grl], [Gr3], Pietsch [Pi], Persson and Pietsch [P-P] and Seminaire Maurey-Schwartz 1972-1973, 1973-1974, 1974-1975 [M-S].

B(X. Y) stands for the space of all the bounded linear operators from a Banach space X into a Banach space Y. Let 1 ,;;;; p

<

00.

Let S be a compact Hausdorff space and let /1 E [C(S)lt. By i

Il.P

we denote the natural injection of C(S) into LP(/1), i.e., the operator which assigns to each

fE C(S) its /1-equivalence class regarded as an element of LP(J1). If E is a closed linear subspace of C(S), then (unless otherwise stated) E Il •p denotes the subspace of LP(J1) being the closure in LP(/1)-norm of ill.p(E) and i: p denotes the restriction of ill .p to E regarded as an operator to E Il.P' DEFINITION

0.1. Let I ,;;;; p

<

00.

A bounded linear operator T: X -

Y is p-integra!

(resp. strictly p-integral) if there are a compact Hausdorff space S, a /1 E [C(S)] bounded linear operators U: X -

C(S) and V: LP(J1) -

t

y** (resp. V: LP(J1) -

and

Y) such

that

Ky' T= Vill.pU

(0.4) where Ky: Y -

(resp. T= Vill.pU)

y** denotes the canonical embedding of Y into its second dual Y**.

A triple (U, V, ill,p) satisfying (0.4) is called a p-integral factorization of T

The p-

integral nonn of T is the quantity

where the infimum is extended over all p-integral representations of T. DEFINITION

0.2. Let 1 ,;;;; p

<

00.

A bounded linear operator T: X -

summing if there are a compact Hausdorff space S, a /1 E [C(S)] E of C(S), and bounded linear operators U: X -

Y is p-absolutely

t, a closed linear subspace

E and V: E Il •p -

Y such that

(0.5)

T

) satisfying (0.5) is called a p-abso!utely summing factorization of A triple (U, V. i~ r-.P The p-dbsolutely summing norm of T is the quantity

where the infimum is extended over all p-absolutely summing factorizations of T. We shall use the term "absolutely summing" instead of "I-absolutely summing". The following result is a slightly improved version of the so-called Grothendieck-Pietsch theorem (cf. [Pi], [P-P], [Mt-P], [P8]). 0.4. Let 1 ,;;;; p < 00 and let T E B(X, Y). (i) If T is p-integral and j: X - C(So) is a fIXed isometric isomorphic embedding of

THEOREM

X into C(So)' So arbitrary compact Hausdorff space, then there exists a /1 E [C(So)] t and a linear operator V: LP(J1) ipCT)

y** such that

= 11/111 I / p , IIVII = 1,

KyT= Vill,pj.

BANACH SP<\CES

(ii) If Tis p-absolutely summing and i: X

----'>

7

C(So) is an isometric embedding of X

into C(So)' So arbitrary compact Hausdorff space, then there exists a !J. E [C(So)] linear operator V: E IJ.,P

----'>

Y where E = i(X) such that

(0.6)

IWII (iii) If there is a constant C n

(0.7)

L

+and a

<

00

=

1, T

= v,/.E IJ..P " .

such that

n

IITxjll P :s;;; CP

j= 1

sup

L

II x * 11<;; 1 i= 1

Ix*(xj)iP for arbitrary

then T is p-absolutely summing with

'IT peT)

Xl' '"

= inf {C:

,xn in X(n = 1,2, ... ),

C satisfies (0.7)}. Conversely, if T is

p-absolutely summing, then T satisfies (0.7) with C = 'lTp(T).

0

Clearly every p-integral operator, say T, is p-absolutely summing and

'IT peT)

:s;;; ip(T).

Using the fact that every closed linear subspace of a Hilbert space is a range of the norm one projection from the whole space, it can be easily seen that every 2-absolutely summing oper· ator is (strictly) 2-integral. By Theorem O.S(ii), every p·absolutely summing operator from a C(S)-space is (strictly) p-integral. Next observe that if Y is a dual Banach space (in particular Y is reflexive) or more generally if Y is complemented in its second dual, then every pintegral operator is strictly p-integral. A discrete analogue of strictly p-integral operators are the p-nuclear operators which we define next. DEFINITION

0.3.

Let 1 :s;;; p

<

00.

A bounded linear operator T: X

if there are bounded linear operators U: X

(/\j) with :2:IAjIP

<

00

----'>

l~, V:

IP

----'>

----'>

Y is p-nuclear

Y, and a sequence of scalars

such that

T= VApU where Ap: l~

----'>

IP is defined by Ap((~i)) = (A;~) for

(U E r.

The triple (U, V, Ap) is

called a p-nuclear factorization of T. The p-nuclear norm of T is the quantity

where the infimum is extended over all p-nuclear factorizations of T.

It is easy to see that every p-nuclear operator is compact. For p

=

1 we write neT) instead of n 1 (T) and we shall use the term "nuclear" instead

of "I-nuclear". The set of nuclear operators from X into Y is denoted by N(X, Y). It is a Banach space under the norm n( . ). We shall need the following result due to Grothendieck [Gr3] , cf. also [D-U] , and

[Per] for the reflexive case. THEOREM

0.5. If Y is a Banach space with a Radon-Nikodym property, then every

absolutely summing operator from a C(S)-space into Y is nuclear and therefore compact.

<\LEKSANDER PEt.CZyNSKI

8

~cIK-li

m{zEaD: I(Hf(z))I;;;'K}

aD

Ifldm

forfELI.

(iv) Iff satisfies the Lipschitz condition, then H(!) is continuous; if f is a COO -smooth

function so is H(!). (v) If H(f) ELI, then

H(f)(n) =

I

> 0,

-i/(n)

for n

/:0)

0, for n < o.

if(n)

for n =

0

The close relatives of the Hilbert transform are: the Riesz projection

R and the Boas

isomorphism B. They are defined via the Hilbert transform by

R(f) 2B(f)(z)

= rl(f+ ;H(!) + =

(I - i)/(O))

(fE LI),

+ z-I [f- iH(f)](z-2) + (I - i)/(O)(I -

[f+ iH(!)](z2)

rl)

(f ELI,

Z

E aD).

Theorem 0.2 yields (for the proof of (jjjj) cf. Boas [80] and [K-P].

0.3. (j) Let 1

Then R is a bounded projection from LP onto its subspace HP such that ker R = {f E LP: 1 E ff.Po}. Moreover, if II Rllp denotes the norm of this projection regarded as an operator on LP, then THEOREM

00.

2

2

b- I - p - ~ IIRII ~ b - p p-I P p-I

where b is an absolute constant independent of p. (jj) (the Kolmogorov theorem) If f ELI, then Rf E H" for every p with 0 Hence there is a constant Cp such that

(0

< p < I.

< p < I).

(jjj) The operator f -+ R(f) is of weak type (I-I). (jjjj) Let I < p < 00. Then B maps isomorphically LP onto H"; we have 00

B(f)

= /(0) +

L

[/(n)z2n

+ j(-n)z2n-l]

n=1

(the series on the right-hand side converges in LP). Moreover

4- l lIfllp ~ IIB(!)lIp

2

~ c p"-

I IIfllp

(fE LP)

where c is an absolute constant independent of p. Finally observe that sometimes it is more convenient to deal with the models of LP and

H"

which consist of 21T-periodic functions on the real line; in these models the Hilbert

transform is given by the formula

H(f)(t)

= - -I

1T

for m-almost all t.

lim 1: ..... 0

Sff I:

[get

+ s) -

get - s)] ds 2tg(s/2)

BANACH SPACES

9

Recall that a Bana;;h space Y has the Radon-Nikodym property if for every finite measure 11 on a sigma field of subsets of a set [2 every V E B(L t (11), Y) admits a repre~entation (0.8)

V(f) =

fn f(w)y(w)ll(dw),

fE Lt{J1),

where y( . ): [2 -+ Y is a Il-Bochner integrable Il-essentially bounded function. While the proof of Theorem 0.5 in full generality is slightly complicated, the fact that the operators in question are compact is easy. Indeed, if T: C(S) -+ Y is absolutely summing then, by Theorem O.4(ii), it admits a factorization

L 1 (11)

i,i \ C(S)

l

Y

T for some 11 E [C(S)]

t

and a V E B(L 1 (11), Y). Since V admits a representation (0.8), it

takes weakly convergent sequences in L t (11) into norm convergence sequences in Y (cL [D-SI, Chapter VI, §8, Theorem 14]). On the other hand,

ill,l

is a weakly compact operator.

Hence Vill,t = T is compact. 0 Finally recall that if X is a finite dimensional Banach space and T E B(X, X), then the

trace of T is defined by dimX

tr(T)

=

L

el(Tej )

j=l

where

(ej)t<;;j<;;dim x

is any basis for X and

(ej)t<;;j<;;dim x

the sequence of coefficient func-

tionals of this basis. The quantity tr(T) is independent of the choice of a basis. It satisfies the inequality Itr(T)1 ~ neT).

1. The F. and M. Riesz Theorem and Duals of the Disc Algebra This section has an introductory character. We outline a proof of the classical M. and F. RIesz theorem and use it to describe the duals of the disc algebra A and the quotient

qaD)/A. Let us regard the disc algebra A as a subspace of C(aD). The annihilator

Al

=

{p E [qaD)] *: faDfdP

=0

for fE A}

is completely described by the classical THEOREM 1.1 (F. AND M. RIEsz). Let m be Lebesgue measure on the unit circle aD

{z

E

=

C: Izl = I}. Then Al

=

{p E [qaD)]

* :p = h

. m for some h E H~}

where H ~ = (the closure in L 1 = L 1 (m) of all the polynomials

r.'J= 1 cjz j (n

= I, 2, ... )} .

Briefly Al = H~ if we identify (via the Radon·Nikodym theorem) the space L 1 with the subspace {p E [C(S)]

* : p «m}.

SKETCH OF THE PROOF. Step I. A closed set Fe aD with m(F)

A, i.e., there is an fF E A such that fF(z) We put fez)

=1-

exp

=

I iff z E F, IfF(z)1

1 r21T z + e it 2), - - . log wet) dt rr 0 z _ e,t

=0

is a peak set of

< 1 for z t{: F.

for

Izl


where w( . ) is a real 2rr·periodic continuous function such that 0 ~ w ~ 1, f~1Tlog wet) dt

>

-00,

wet) = 0 iff e it E F, and w is infinitely many times differentiable at each point t t{: F. fF(z) = lim f(pz) for z E aD.

with e it

p->l

The

es~ential

technical difficulty is to show that the limit exists and that

f

is uniformly con·

Izl < I (cf. [H, p.80] , [St, p. 205] for details). Step II. If pEAl, then p «m. Pick any closed Fe aD with m(F) = 0 and define

tinuous for

fF as in Step I. Then

o = f f'j.. dp Hence p(F)

= limnff~ dp = O.

Thus p

«

for n

=

I, 2, . "

.

m, being a regular Borel measure which vanishes

on every closed subset of aD of Lebesgue measure zero. 10

11

BANACH SPACES

Step III. Ai = H~. By Step II and the Radon-Nikodym theorem, if p. EA 1 , then p. = h . m for some h E L I(m). Let f(t) = h(e it ). Then p. E Ai yields A

f(-n)

I f2IT . t = 2; Jo f(t)e 1n

dt

=0

for n

= 0,

I, 2, ...

Thus, by the FeJer theorem, f is the limit in L I of polynomials of the form ~:= I cie iit . Equivalently h E H~. Thus H~ J Ai. Obviously zi E Ai for j = 1,2, . . . . Hence H~ C Ai. 0 The description of the dual of A. If Y is a subspace of a Banach space X, then the

+

map which assigns to the coset {x* phism from

X*/y1

onto Y*.

Y 1 } the restriction of x* to Y is an isometric isomor-

Thus, by the M. and F. Riesz theorem, A* is isometrically iso-

morphic to [C(aD») */H~. By the Lebesgue decomposition theorem

(1.1 )

[C(aD)] *

= L1

$1

Vsing .

(The equality in (1.1) and (1.2) below means "isometric isomorphism" and the symbol X $1 Z means the [I -sum of X and Z.) Thus we have

(I.2)

A*

= LI/H~

$1 Vsing .

It follows from the definition of 11 that

(H~)l = {fE L 00 = LOO(m): f fgdm = 0 for g E H~} = 11. Thus a standard duality argument gives (L I /H ~)*

(1.3)

A ** = 11

= H oo •

Now (I .2) yields

$00 V:ing·

(The symbol X $00 Z means the 1 -sum of X and Z.) Observe that Vsing is an L 1 (v)-space. 00

Hence

V:

i ng is a C(K)-space for an extremely disconnected compact Hausdorff space K. Our next aim is to state some consequences of the F. and M. Riesz theorem and (1.2)

in terms of isomorphic invariants of Banach spaces. A subspace £ of a C(S)-space is said to have a small annihilator if there is a v E [C(S») such that £1 C L I (v). Obviously if £1 is a norm separable subspace of [C(S») *, then £ has a small anmhilator. Conversely if S is metrizable and £ C C(S) has a small annihilator then £1 is norm separable. We will distinguish two properties of a Banach space X. Property I. X is isomorphic to a space with a small annihilator. Property II. X* is a separable distortion of an L 1 (v), i.e., X* is isomorphic to a product M $ V where M is a separable space and V is an L I (v) space.

Observe the following: 1°. The disc algebra has both Properties I and II. 2°. If X is separable then Property I implies Property II. 3°. If n ;;:. 2 then n-dimensional Hilbert space is not isometrically isomorphic to any subspace of a C(S)-space with a small annihilator. However, every finite dimensional Banach space (being isomorphic to a C(S)-space) has Property I.

+

ALEKSANDER PH.CZyNSKI

12

We end this section by stating a description of a predual and the dual of HI which is also a simple consequence of the F. and M. Riesz theorem and the following general fact. If Y is a subspace of a Banach space X then the dual of X/Y is naturally isometrically isomorphic to yl and (yl)* is naturally isometrically isomorphic to X**/Y**. We put Ao

= zA

and H~ = zH"" . THEOREM

1.2. (i) The map f ~ x; where x;( {g

+ Ao} )

=

f <wfg dm for g

E q3D)

is an isometric isomorphism between HI and the dual (q3D)/A o )*'

(ii) The map {g + H~} ~ IP; for gEL"" where isometric isomorphism from L "" /H~ onto (HI )*.

+ A o}

~ {f

+ H~}

1P;(n = f il Dfg dm for f E HI

for f E q3D) can be identified with the canonical embedding of q3D)/A o into its second dual identified with L "" /H~. 0 (iii) The map {f

is an

2. Absolutely Summing Operators from the Disc Algebra The main result of this section is Theorem 2.3 describing p-absolutely summing operators from the disc algebra. It is preceded by a description of p-absolutely summing operators from a subspace of a C(S)-space with a small annihilator (Theorem 2.2). The technique is based upon Bishop's generalized Rudin-Carleson theorem [BI], [Gil] which with the improvement of [Gm2], [P2], [P3], [M-P] (cf. also [Gm 1], [St]) can be stated as follows. THEOREM 2.1. Let X be a subspace of a C(S)-space. Let F be a closed subset of S such that

(2.1)

p(F 1 ) = 0

for every closed Fl C F and every p E Xl.

Then for every u E C(F), every fuCs)

(2.2)

> 0 and every open = u(s) for s E F,

E

Ifu(s)1

<E

G :J F there is an fu E X such that

for s $. G and IIJlI

Moreover, if X is separable then the map u linear isometry.

-+

= lIuli.

fu from C(F) into X can be chosen to be a

Let us observe that if X has a small annihilator in C(S), say Xl eLl (A) for some X E [C(S)]

t, then

a closed F C S satisfies (2.1) whenever X(F)

= O.

Now we are ready to prove THEOREM 2.2 (MITJAGIN-PELCZyNSKI). Let X be a subspace of a C(S)-space with Xl eLl (X) for some probability measure X E [C(S)] *.

Let 1 .;;;; p space

<

00

and let T: X

-+

E be a p-absolutely summing operator into a Banach

E.-

Then there exists a strictly p-integral operator V: X such that

II(T - V)(nll .;;;; (fslf(S)IPh(S) dA) IIp

(2.3)

-+

E and a nonnegative h ELI (A)

for f EX,

(2.4) PROOF. Smce Tis p-absolutely summing, there exists apE [C(S)]

1rp(T)

=

IIpllllP

and IITfll .;;;;

(fslf(s)I P

unique bounded linear operator B: X IJ.

t

such that

dp)l/p for f EX (cf. §O.III). Hence there is a -+

E such that B(ilJ.,p(n)

13

= T(f)

for f E X.

For a

ALEKSANDER PEl.CZyNSKI

14

r E IC(S)]

+we denote by XT,p the closure of X (more precisely, of iT,p(X» = a + v with a «A, v 1 A and

Lebesgue decomposition theorem yields p

Ilpll

in LP(r). The

=

lIall

+ Ilvll.

We shall identify e(p) with the [P-direct sum e(a) EBp LP(v). The crucial point of the proof is to establish

(2.5)

X/1-,P

= Xa,p

EBp LP(v).

Assume that (2.5) has been proved. We then define V: X ----> E by BivQvi/1-,P where Q v : e(a) EBp e(v) ----> e(v) is the natural projection,i v : e(v) ----> e(a) EBp LP(v) the natural embedding (i.e., iv(b)

= (0, b)

for bE e(v». By (2.5), ivQvi/1-,p(X) C X/1-,P; thus V is well

defined. Clearly Qvi/1-,P = iv,p' and IIBiv ll .:;;; IIB1i1liv11 .:;;; 1. Thus V is strictly p-integral with ip(V)':;;; Ilvll l / P . Next for eachfEX

II(T - VXf)1I

= IIBi/1-,p(f) -

BivQ)/1-,p(f)1I

.:;;; lIi/1-,pCf) - i vQv i /1-,p(f)II.

e

If h E LP(p) = e(a) EBp (v) then h = i aQa(h) + ivQv(h) where Q a : L I (J1) ----> L I (a) and i a: L I (a) ----> L 1 (p) are the natural projection and the natural embedding respectively. Thus for h

= i/1-,p(f), we

have (f) - ivQvi/1-,p(f)

II(T- V)(f) II .:;;; lIia,pCf)1I Hence we get (2.3) with h

= da/dA.

= i aQ ai/1-,p(f) = ia,p(f).

= (fs lflP do rIP

It follows that

for IE X

Finally

(1Tp(T»)P - (ipCV»P

~

Ilpll - Ilvll

=

lIall

=

Ish dA

which proves (2.4).

(2.5). The inclusion X/1-,P C Xa,p EBp LP(v) is trivial. For proving the reverse one, pick a z in Xa,p EBp LP(v), i.e., pick an a E Xa,p and bE LP(v) so that z = a + b PROOF OF

and

(2.6) Next fix finite M

> °and pick an x E X so that f sla - xl P do < EP. Since b ELP(v), there is a > 1 such that if Z = {s E S: Ib(s)IP .:;;; MP} , then

E

f

(2.7)

JS\Z

IblP dv

< EP

and

v(S\Z)' MP

< EP.

Combining (1.6), (1.7) with the Lusin theorem and the fact that v, a, A are regular Borel measures with v 1 A, and a A and using the fact that the measure B ----> f B Ib(s)I P dv is

«

absolutely continuous with respect to v, we construct a compact FeZ and an open G :J F such that

b restricted to F is continuous, A(F)

= 0,

(2.8) max{v(S\F), a(G), fZ\FlbiP dV}

< EP/MP.

Now we apply Theorem 2.1 for u E C(F) defined by u(s)

= b(s) -

xes) for s E F.

15

BANACH SPACES

Let fu E X satisfy

(2.2). Then IIfull

« sUPSEFlb(s)1 + IIxll « M + Ilxll because Fez.

Hence, we get

« (fsla -

(fslZ - (x + f)IP dp yIP

x - fu lP da yIP + (fs Ib - x - fui P dv YIP.

Estimating each of the terms separately we get

(fsla - x - fui P da yIP

« (fsla -

xl P da )I IP + (fslfu lP da rIP

«€+ (fS\GlfulPdaYIP + (fGlfuIPda)"P

« € + a(S)I/P€ + (a(G)llfull)l lp « €(1 + a(S)l/p + 1 + IIxli/M); = (fS\F Ib -x - f.IP dV)I IP 5SIb -x - f.IPdv u u « (fS\F IbI P)l IP + (JS\F Ix + f.u IP dV)I/P

«

(J

Z\F

Ibl P dv)l lP

+

(5

S\Z

Ibl P dv)I/P

+

[V(S\F)]'/Plix

+ f.u II

« (c + € + €(21Ixll + M))M- I « €(2 + 211xll + M)M-I. Letting

€

tend to zero we infer that

Z

E X p."

0

COROLLARY 2.1. Let X satisfy the assumption of Theorem 2.1. If E has the Radon-

Nikodym property and p = 1 then the operator V satisfying (2.3) and (2.4) is nuclear with n(V) = i l (V) and therefore compact. PROOF. Use the fact that every strictly integral operator to a space with the RadonNikodym property is nuclear (cf. §O.III). REMARK. or E

= HI,

The assumption of Corollary 2.1 is satisfied if E is either a Hilbert space

because HI being a separable dual (by Theorem 1.2) has the Radon-Nikodym

property. For the disc algebra Theorem

2.2 can be strengthened as follows:

THEOREM 2.3. Let E be a Banach space, I

«p <

and let TA --+ E be a p-absolutely summing operator. Then there is a strictly p-integral operator V: A --+ E and an outer function

(2.9)

(2.1 0)

I
< 1,

16

ALEKSANDER paCZYNSKI

(2.11) Let V and hELl be defined to satisfy (2.3) and (2.4) (for S Let

PROOF.

X

= A).

.p(z)

1

= exp

-2

r27Teit+z . log [h(e 't ) 0 e't-z

J

11

I

•

+

1] dt

for Izl

= aD,

l\

= m,

< 1.

It is well known [Ou, Chapter 2] that

l.p(eit)1

=

lim l.p(pe it) 1 = h(e it )

+

1

almost everywhere

p-+I

which implies (2.9) and (2.11).

< 1 we

Since h ;;;, 0, for z = re u with r 1 S27T I.p(z) 1 = exp -2 11

= exp -

0

Re«te,t +- zz) .10g(I + h(e

1 J27T

211

0

have

1 - r2 1

+ r2

- 2rcos(t - argz)

it

» dt

10g(I

.

+ h(e 't » dt

;;;, 1, which proves (2.1 0). THEOREM 2.4 (MITJAGIN-PEt.CZyNSKI). Let 1 < p < 00. Then every p-absolutely summing operator from A is strictly p-integral. More precisely, there is a constant Cp with p2

(2.12)

Cp .;;;

--p=I" b

(where b is an absolute positive constant independent ofp) such that ip(T)';;; Cp11p(T) for every p-absolutely summing operator from A. PROOF.

Let

restriction of the natural injection i lIP 1m ,p

:

C(aD) ~ LP (I
from A into HfIP'm where HfIPlm is the closure of iIIPlm,p(A) in LP(I
(2.13)

A ip(iIIPlm,p) .;;; IIRllp

(5avl
Here IIRllp denotes the norm of the orthogonal (Riesz) projection from LP onto HP regarded as an operator on LP. Let us consider the diagram C(aD)

(2.14)

rj A

ilIPlm,p

)

LP(I
where j denotes the natural inclusion,

MtJ;

~

uL

IfP

1

MtJ;-l ) HP IIPI'm

VI is the outer function defined by

BANACH SPACES

I/I(z) = exp ( -

1

p

1

(21T Z

-2 J, 1r

0

17

+ e.it log Re .,o(t) dt)

Z _

for Izl

e,t

and M!J; and M!J;-I the operators of multiplications by 1/1 and 1/1

M!J;(f) = 1/1 - I

and

-I

< 1, respectively, i.e.,

M r 1 (f) = 1/I- 1f.

It is known that II/II P = I'PI and 1/1 E HP (because 'P E HI). Hence M!J; is an isometric isomorphism from LP(I'Plm) onto LP. We have to check that

(2.15)

M!J; carries Hf
This will show that the diagram (2.14) is commutative and therefore i~lm,p is strictly pintegral. To prove (2.15) observe that obviously M!J; (Hf
(2.16)

if g E HP, then g/1/l E Hf
= I/I(rz) for Izl ~ 1 and 0 < r < 1. Clearly I/I r E A and, by (2.10), 1I/I,(z)1 1lp l.,o(rz)1 ;;;. 1 for Izl ~ 1. Observe that if lEA then IN EHf
=

lim

r= 1

J ~ 1I

-

'Y r

'Y

Ipl'Pl dm

~ IIfll 00

lim} II/I r - I/IIP dm 11/1 riP

r= 1

~ IIflloo r=lim1 SII/I r -

I/IIP dm = 0

(we use here that 1/1 E HP (cf. [Dll, Chapter 2] )). Next given g E HP, there is a sequence

(In) in A with 0 = limnfaD Ig - In IP dm = limnfaD IgN - InN IP I'PI dm. Since InN E Hf
i~lm,p = M!J;-IRM!J;i l
= IIRllp(fI'P1 dm)l Ip _ This

= 11M!J;-111 = lUll = 1, we have completes the proof of (2.13).

Now, let E be an arbitrary Banach space, 1 lutely summing operator with 1rp(T)

= 1.

00,

ip(i~lm,p) ~

and let T: A ~ E be a p-abso-

By Theorem 2.3, there is a strictly p-integral

operator V: A ~ E and an outer 'P E HI satisfying the conditions (2.9)-(2.11). It follows from (2.9) that there is a unique bounded linear contraction B: Hf
= (T-

VXI) for lEA, equivalently B· i~lm,p ip(T - V)

Since 1rpCT)

=

~ IIBllip(il!,lp) "

= T-

V. Thus, by (2.13)

IIRllp(f I'PI dm riP.

1, (2.11) yields (f1'Pldm)l /p " 21/p

< 2.

Hence, again using (2.11), ip(T)

+ ip(V) ~ 211Rllp + 2 = Cpo This proves the last assertion of Theorem 2.4. Finally (2.12) follows from the estima-

"ip(T - V)

tion for IIRllp (cf. §O.lI). REMARK.

0

Let us observe that the estimation for Cp given in (2.12) is the best possi-

ble in the following sense infip(T) ~ ap2/(p - 1)

for 1 < p

<

00

where a is an absolute constant independent of p and inf is extended over all operators

18

ALEKSANDER PEI.CZyNSKI

trom A with 1fpCT)

1f p(i!,p)

=I

=

1. In fact for the natural embedding i!,p: A

---+

HP we have

and

(2.17)

i/i!.p)

= IIRll p'

(2.17). Since i!,p = R im,pj we have IIRllp ~ ip(i!,p) (here j: A ---+ qaD) denotes the natural inclusion). To prove the reverse inequality we use the standard averaging PROOF OF

technique (cf. [PlO] for a more general treatment). Pick fJ. E [qaD)]

+with 1IfJ.1I = I so

that i!,p = BilJ.,pj for some operator B: LP(fJ.) ---+ HP with IIBII = ip(i!,p)' (The existence of B follows from the definition of p-integral operators (cf. §O.III) and the fact that HP is reflexive for 1

00

and therefore every integral operator to HP is strictly integral.) Let

us set

(Saf)(z) =/(exz)

B(n =

for IELI and

f SaBSa-I I m(dex)

iexl = Izl = 1,

for IE qaD).

Then, we have, for every IE qaD),

IIBfll~ ~ fIlSaBSa-Ifll~ m(dex) = fIlBSa-Ifll~ m(dex)

~ IIBIIJ (JISa-1/IP dP.) m(dex) = IIBllfl/lP dm. (The last equality follows from the observation that because of the uniqueness of the Haar measure JSa-Ih dp.m(dex) = Jh dm for every probability measure fJ. E qaD)* and every nonnegative h ELI.) Thus extends to a bounded linear operator from LP into ffP which co-

B

incides with R because BI = I for lEA.

TIlUS

IIRllp ~ IIBII

= ip(i!,p)'

0

3. Absolutely Summing Operators from the Disc Algebra into Hilbert Space The operator f - * (/(2 n )) is an example of an absolutely summing map from A onto [2. The existence of this phenomenon leads to a new proof of Grothendieck's theorem that

every bounded operator from [I into [2 is absolutely summing. It also emphasizes the difference between the disc algebra and C(S)-spaces, because every absolutely summing operator from a C(S)-space into [2 is nuclear, hence compact. The main result of this section- Theorem 3.2-gives a description of absolutely summing operators with a closed range from A into 12. At the end of this section we discuss the problem whether every bounded operator from A into [2 is 2-absolutely summing. The following concept plays an important role throughout this section. DEFINITION 3.1. is a K

> 0 such

An orthogonal projection P: H2 - * H2 is a Paley projection if there

that

IIPfl12 ~ Kllflll

(3.1)

for f E H2.

EXAMPLE 3.1. Let (n k ) be a sequence of nonnegative integers such that lim n k + 1 /n k

> 1.

k

Then the orthogonal projection Pf = 'Lk/(n k )zn k, f E H2 (where /(n)

= JaDf(z )z-n m(dz) I, ... ) is a Paley projection. This fact is simply a restatement of a classical result of Paley (cf. [Z], [Du, Chapter 6], [Pal]). If P: H2 - * H2 is an operator then PA denotes the restriction of P to A regarded as an operator from the disc algebra A into H2, formally PA = Pi~. 2' Our next result shows that every Paley projection induces an absolutely summing surjection from A onto a Hilbert for n

= 0,

space. Precisely we have PROPOSITION 3.1. If P is a Paley projection, then PAis an absolutely summing surjection from A onto P(H 2 ). PROOF. Clearly (3.1) implies (in fact is equivalent to) (3.2)

liPA (f)1I

Hence PAis absolutely summing and

~K

S

aD

IfI dm

1T 1 (P A) ~

for

f

E A.

K.

To prove that PAis onto P(H2) we show that PA: [PA (H2)]

* - * A * is

an isomorphic

embedding. Clearly we may identify the Hilbert space [PA (H 2 )] * with P(H2) and the adjoint PA with the operator which assigns to each f E P(H 2 ) the functional x; defined by 19

A.LEKSA.NDER PELCZYNSKI

20

xf
for g EA.

Here we used the fact that the projection P is orthogonal. A*

By (1.2), the functional

x; E

= L I /H~ x Vsing can be identified with the coset {f + H~} or, more precisely, with the + H ~}, 0). Therefore to prove that PA is an isomorphic embedding it suffices to

pair ( {f

show that there is K I

I~A (nil

(3.3)

=

> 0 such that inf

S

o aD

hEHI

If(-z)

+ h(z)1 m(dz) ~ K IIIfll2

for f E p(H2).

By the Kolmogorov theorem the "orthogonal projection" annihilating H~ is a bounded linear operator from L I into L 1/2 (cf. §O.II). Hence there is a c

> 0 such

that

(3.4) Next, combining (3.1) with the Schwarz inequality applied to the product Ifl l / 4 IfI 3 / 4 = If I, we get

Hence

(3.5) Combining (3.4) with (3.5) we get (3.3) with K 1 = cK- 4 .

0

COROLLARY 3.1. There exists a map from the disc algebra onto 12 which is absolutely summing. PROOF. Let (n k ) be a sequence of positive integers with limkn k + link> 1. Then the operator T: A - ) 0 12 defined by Tf = (/(n k » for f E A has the desired properties. This follows immediately upon combining Example 3.1 with Proposition 3.1. Corollary 3.1 has a surprising application. THEOREM 3.1 (GROTHENDIECK [Grl]). Every bounded linear operator from [I into /2 is absolutely summing. PROOF. Let S: [I

-)0

[2

be any bounded linear operator and let T: A

-)0

[2

be an

absolutely summing surjection. By the open mapping principle, there is a constant K 1 such that given x E [2 there is an f E A with IIfll ~ K Illxll and Tf

= x.

>0

Hence if ej is the

jth unit vector of [I ' then there is an /j E A such that T/j = Se. and 11[,·11 ~ K IIiSIi. Now --1 "" I J,....., ,. . . " we define the lifting S: [ - ) 0 A by S((tj» = r.tj/j for (tj ) E [I. Then S = TS and liS II ~

KIIiSIi. Hence

1T 1 (S)

~

K 11T I (T)IISII.

0

BANACH SPACES

21

Our next result shows that up to a Banach space automorphism of a Hilbert space and of the disc algebra every absolutely summing surjection from A into a Hilbert space is a nuclear perturbation of a Paley projection. Instead of discussing "surjection onto Hilbert spaces" it will be more convenient for us to work with operators into 12 with a closed range.

*"

0 for Izl < 1 then the operator M.p defined by M/t) = If IP is a function in A with
12 be an absolutely summing operator with a closed range. Then there is a nuclear operator V: A - 12 such that for every E > 0 there exist a multiplication automorphism M.p: A - A, a Paley projection P: H2 - H2 and an invertible operator U: P(H 2 ) - T(A) such that 1T 1(T-(UP A M.p + V) < E. PROOF. By Corollary 2.1, the remark after Corollary 2.1 and Theorem 2.2, there exists a nuclear operator V: A -

(2.11) for p

=

1. Let Tl

[2 and an outer function

= T - V.

IP satisfying the conditions (2.9)-

Then the inequality (2.9) can be rewritten

II Tl fll ,;;;; flfl IIPI dm

for f E A.

Hence there exists a unique linear operator Q: H~.plm -

Qi~lm,l. Next fix r with 0 (3.6) where IPrCz)


Qy(h)

= IP(rz)

for

=

12 with IIQII ,,;;; 1 such that Tl 1 and define Qy: H11.pylm ---+ 12 by

Q(h . IPy/IP)

for h E Hr.pylm

Izl ,,;;; 1. To show that Qy is "well defined" we have to check that

if h E Hf.py,m then hIPy/IP E H ~.plm. To this end fix 8

~ > flh

- fllIPyl dm = flhIP y - fIPyl dm

In view of [Du, Chapter 2] there exists an R

< 1 so

>0

=

f

and pick f E A so that

Ih~ - f~IIIPldm.

close to 1 that

Then, by (2.10),

I

I

IPy - f IPR IPy IIPI dm ,,;;; IlfIPyll~ f lIP - IPR I dm f f-;

8 < 2"

Hence

Since fIPy/IPR E A, the last inequality yields that hIPy/IP E H~.plm. Hence Qy is well defined. Now put Ty

= Qyif!,ylm,l:

A _/2. Then, by (3.6), for fEA,

II(T. - Ty)(t) II = IIQ(f -

f·~ ) II ,;;;; f

,,;;; flfllIP - IPrl dm.

\t - ~ f

IIIPI dm

ALEKSANDER PEI:.CZyNSKI

22

Thus in view of [Du, Chapter 2] , lim

(3.7)

r= 1

1T 1 (T 1 -

It follows from (2.1 0) that for 0

Tr)

<, < I

=

lim flIP - IPrl dm

r= 1

= O.

the operator M'Pr is a multiplication automorphism.

Hence, by the definition of Tr ' we have

IITrM

0~)

-1

~

(f)11 = IIQ(M

-1 (f)

~

. IPrllPll

= IIQ(fIP- 1 )11 ..;; Jill dm

..;; (J1/12 dm)t/2.

Hence there exists a linear contraction, say Sr: H2 - [2, such that Sr(f) = TrM -1 (f) for 'Pr lEA. Since A is dense in H2 in the norm II . Ib, Sr is uniquely determined and closure Sr(H2) = closure (TrM -l(A». Let P = [l..r>: H2 - H2 be the orthogonal projection with 'Pr ker P = ker Sr' Then there exists a one-to-one bounded linear operator Ur = U: p(H2) = [2 such that Sr = UP. Now assume that U has a bounded inverse, i.e., there exists a bounded linear operator V: U(p(H2» - p(H2) such that VU is the identity on P(H 2). Then, by (3.8) for I E A, 11P(f)11";; 1IV11I1UP(f)1I = IIVllllTrM -1(f)1I 'Pr ..;; II VII fill dm. Thus 1IP(f)1I ..;; II VIIJl/I dm for IE H2 (because for every IE H2 there is a sequence (fn) in A such that limnilln - 1112 = limnil/n - 1111 = 0). Hence, if U has a bounded inverse, then P is a Paley projection and

,

Tr and given e

= TrM'Pr_lM.~Tr = UPAM'P r

> 0, 1T 1 (T-

(V

+ UPAM'Pr» = 1T 1 (T 1

-

Tr)

<e

for, sufficiently close to 1 (by (3.7». We complete the proof by showing that there is an '0 with 0 < '0 < 1 such that if '0 < , < 1, then U = Ur does have a bounded inverse. To this end observe that Tl = T - V has a closed range because T does and V being nuclear is compact. Hence there is an e > 0 such that if T: A - [2 is an operator with IITl - Til < €, then T has a closed range. Thus, by (3.7), there is an '0 with 0 < '0 < I such that if '0 <, < 1, then Tr has a closed range. Therefore, TrM -1 also has a closed range and so does Sr' This implies that U has a closed 'Pr range because Sr = UP and P is a projection onto the domain of U. Thus U being one-toone has a bounded inverse. 0 The results of this section show that absolutely summing operators from the disc algebra into a Hilbert space are in general quite different in nature than absolutely summing operators from a C(S)-space into a Hilbert space. The situation seems to be different in the case of 2-absolutely summing operators. Prob[em 3.1. Is it true that (i) every bounded linear operator from A into [2 is 2-absolutely summing?

BANACH SPACES

23

Let us observe that (i) is equivalent to each of the following properties:

(ii) every operator from

{2

into

{2

which factors through A is Hilbert-Schmidt;

(iii) every bounded linear operator from A into

{2

extends to a bounded linear opera-

tor from C(aD) into /2; (iv) every operator from L I IH~ into (v) every operator from

{2

into

{2

{2

is 2-absolutely summing;

which factors through L I IH~ is Hilbert-Schmidt;

(vi) for every linear operator T: L I IH~ -+

= T where q:

{2

there exists an operator

T:

LI

-+ {2

L I IH~ denotes the quotient map. The equivalences (i) <== (ii) and (iv) <== (v) follow immediately from the definition of

such that Tq

LI

-+

a 2-absolutely summing map. The equivalence of (i) and (iii) follows from the extension property of 2-absolutely summing operators and the fact that every bounded linear operator from a C(S) space into

{2

is 2-absolutely summing (cL [Grl] , [L-PI]). The equivalences

(ii) <== (v) and (iii) <== (vi) follow by a standard duality argument (we apply formula (1.2) and use the fact that every bounded linear operator from L I (resp. C(aD)) into

{2

is 2·sum·

mingo

Notes and remarks to § §2 and 3. The results of §2 are due to Mitjagin and Pelczynski. Theorem 2.2 generalizes Proposition I of [Mt-P] where the case p = 1 is considered. The proofs of Theorems 23 and 2.4 have not been published previously. The following is open (the answer is "yes" if H~ has the bounded approximation property):

Problem 3.2. Is every p·absolutely summing operator from H~ into an arbitrary Banach space p-integral? The results of §3 are due to Peiczynski and Wojtaszczyk. The fact that if

~(nk+ 1 Ink) > I then the operator f -+ (/(n k )) maps the disc algebra onto (cL, e.g., [Fourl], [Four2] and [Vinl]).

{2

is not new

4. The Nonexistence of Local Unconditional Structure for the Disc Algebra and for its Duals In this section we shall show that the disc algebra A as a Banach space is in a certain sense very different from a C(S)-space. A is not isomorphic to a quotient of any C(S)-space, nor is it complemented in any Banach space with a local unconditional structure. Most of the argument which leads to the above results is based upon the observation that the natural embedding

i!

I: A ~ HI does not factor through any L I-space. To illustrate our approach

we begin with the following simple PROPOSITION 4.1. The disc algebra and H oo are not isomorphic to quotients of C(S)-

spaces. PROOF. Let us consider the following property of a Banach space X

(4.1)

there is a noncom pact absolutely summing operator from X into a separable duai Banach space. Note that 10. If X satisfies (4.1), then so does Y whenever there is an operator from

Y onto X. (Hint. Use the Banach open mapping principle.) 2°. A and H"" do have (4.1); the desired operator is i~,I: A ~ HI, resp., itf,':'t: H"" ~ HI (HI is a separable dual by Theorem 1.2). 3°. Every C(S)-space fails to have (4.1). Every absolutely summing operator from a C(S)-space to a separable dual is nuclear and therefore compact (cL §O.III).

0

COROLLARY 4.1. Neither the disc algebra nor H oo is isomorphic to a complemented

subspace of a C(S)-space (Pe[czynski [P2] , Rosenthal [Rl]); in particular A is uncomplemented in C(aD) and H"" is uncomplemented in L"" (Rudin [Rul]). We shall strengthen Proposition 4.1 in various directions. To state our results we recall some concepts and facts. An operator T: X ~ Y is LI-factorable if there is an LI(v)-space and operators U: X ~ L I (v) and V: L I (v) ~ Y with VU = T.

A Banach space X is said to have C-L l.u.st

(= Gordon-Lewis local unconditional structure) if there is a K

> 0 and

a function E ~

(UE • VE• FE) which assigns to every finite dimensional subspace E of X a finite dimensional space FE with a basis (I;) satisfying the condition II~t;1; II = II~ It; 11;11 for all scalars t I' t 2 , .. tn' and operators UE : E ~ FE' VE : FE ~ X such that VEUE(e) = e for e E E and IIVE"IIUEII < K. X is said to have l.u.st if moreover the above UE and VE are isomorphisms from E onto FE and from FE onto E respectively [D-P-R]. 24

BANACH SPACES

25

REMARKS. R.1. If Tis L I-factorable so is T**. The converse is true assuming that the range of T is isomorphic to a complemented subspace of a dual Banach space [G·! ] .

[F-J-T]. R.Il. X has G-L l.u.st iff X** is isomorphic to a complemented subspace of a (complex) Banach lattice. Hence if X has G-L l.u.st so do X* and complemented subspaces of X

[F-J·T]. R.1l1. If a Banach lattice Z does not contain an isomorph of

Co

then Z is comple-

mented in Z** (cf. [Sch, Theorem 2.10 and Proposition 5.15]). The following important result due to Gordon and Lewis [G-L] links Banach lattices and spaces with l.u.st with absolutely summing operators. THEOREM 4.1. Let X have G-L I.u.st.

Then every absolutely summing operator from

X into a dual Banach space is L I-factorable. COROLLARY 4.2. If X is a domain of a non-L I-factorable absolutely summing opera-

tor whose range is a dual Banach space then X and all the duals of X do not have G-L I.u.st. PROOF. The second adjoint of an absolutely summing operator is absolutely summing. Combine this fact with Theorem 4.1, R.l and R.Il.

0

Now we are ready to state the main result of this section. THEOREM 4.2. (i) Every L I-factorable operator from A into HI is compact.

(ii) A and all the duals of A do not have G-L l.u.st. (iii) If either A or H~ is a quotient of a Banach space Y with G-L I.u.st, then Y contains a complemented isomorph of 11. (iv) If L 1 /H~ is a subspace of a Banach space Z with G-L I.u.st, then Z contains I~

uniformly, i.e., given any positive integer nand

E

> 0 there is isomorphic embedding

Un:

I~ ~ Z with IIUnllllU;;111 < I + E. (v) If L 1 /H~ is isomorphic to a subspace of a Banach lattice Z then Z contains an

isomorph of co' PROOF. (i) Let T

= VU:

A ~ HI with V: L 1 (v) ~ HI and U: A ~ L 1 (v). Since

HI is a separable dual, V takes weakly compact sets in Ll(V) into a compact set in HI (cf. [D-SI, Chapter VI, 8]). To complete the proof it is enough to show that U is weakly compact or equivalently that U*: [L 1 (v)]

* ~ L 1 /H~

x Vsing has the same property. Recall

that every operator from a C(S)-space into a Banach space which does not contain isomorphs of Co is weakly compact [PI]. Thus every operator from [L 1 (v)]

*

into L 1 /Ho is weakly

compact. L 1 /H~ does not contain isomorphs of co' because L 1 /H~ is separable and complemented in a dual Banach space A *, isomorphs of

Co

are complemented in every separable

Banach space in which they are embedded (Sobczyk [S], cf. also [Ve]), and isomorphs of Co

are never complemented in dual Banach spaces [B-P]. Since Vsing is an L 1 (p)·space ev-

ery operator from a C(S)-space [L 1 (v)] ator from [L 1 (v)]

*

* into

Vsing is weakly compact. Hence every oper-

into the product L 1 /H~ x Vsing is weakly compact.

(ii) The natural embedding i~,I: A ~ HI is absolutely summing but not compact.

ALEKSANDER PELCZYNSKI

26

A I is not L 1 -factorable. Now we use Corollary 4.2. Hence by (i), i m, (iii) Let T = iHe<>1 q: Y - 4 HI where q: Y - 4 He<> is the quotient map. Assume that m,

Tis L I-factorable. (By Theorem 4.1, this is the case where Y has G-L I.u.st.) Consider the diagram

By the open mapping principle, there is a C > 0 and Y n in Y with llYn II < c and Ty n = zn forn = 1,2, . . . . Clearly Ilz n -zmIlHI): 1 forn,*m. Hencethesequence(Uyn)CLI(v) does not contain weak Cauchy subsequences, because V: L I (v) - 4 HI takes weak Cauchy sequences into norm Cauchy sequences [D-SI, Chapter VI.8]. (HI is a separable dual.) Therefore, by a result of [K-PJ and [P-RJ, there is an infinite sequence (UYnk) which is equivalent to the unit vectors of /1 such that there is a projection P from L1 (v) onto the closed linear span of the UYnk's. Hence PU maps Y onto /1, which yields the desired conclusion for HOC because [I has the lifting property. The proof for A is the same. (iv) If L 1 IH~ C Z and Z has G-L I.u.st, then He<>

= (L I IH~)*

is a quotient of Z*.

By R.Il, Z* has G-L l.u.st. Hence, by (iii), Z* contains a complemented isomorph of [I which implies that Z contains

1';;

uniformly [G-J].

(v) Assume to the contrary that there is an isomorphism j: L I IH~

-4

Z where Z is

a Banach lattice which does not contain isomorphs of co' By R.Ill, there is a projection

P: Z**

~

onto

Z_ Let us consider the diagram

CIAo

~ LIIH~ ~Z

~

(4.2)

.Ip C(S)~

Z**

where i*: CIAo

-4

L I IH~ is defined by i*( {[ + Ao})

= {[ + H~}.

Clearly (i*)*

= ifh~.

Thus the assumption that Z has G-L l.u.st combined with R.Il and Theorem 4.1 yields that

(ji*)*: Z* - 4 HI factors through an L I (v) space. Let (ji*)* = VU be the factorization. Then (ji*)** = U*V*: HOC - 4 Z** factors through a C(S}-space (L I (v))*. Let V: denote the restriction of V*: L IH'; - 4 C(S) into a subspace CIAo identified with its canonical image in C" IH'; (cL Theorem 1.2(iii)). It is easy to check that the diagram (4.2) commutes, i.e.,ji* = PU*V:_ Since Z does not contain co' PU* is weakly compact [PI] and therefore PU* takes weak Cauchy sequences in C(S) into norm Cauchy sequences in Z [D-SI, Chapter VI] _ Since the space CIAo has a separable dual (namely, HI), every bounded sequence in CIAo contains a weak Cauchy sUbsequence. This implies that PU*V: is compact. On the other hand, ji* is not compact because j is an isomorphic embedding and i* is not compact (we have Ili*( {z-n + Ao} - {z-m + Ao} )II L I /H~ ~ I for n m). Thus PU*V: ji*, a 00

'*

contradiction.

0

'*

BANACH SPACES

27

The argument of Theorem 4.2(i) shows, in fact, the following: Let X be a Banach space whose dual is a separable distortion of L 1 (v) (cf § I). Then REMARK.

(a) Every operator from X into an L 1 (v) is weakly compact. (b) Every L I-factorable operator from X into a separable dual Banach space is compact.

5. Application to Unifonn Algebras In this section we extend results of §4 and some results of §3 to uniform algebras with nontrivial Gleason parts, and to -uniform algebras with a separable annihilator as well. We close by discussing GJicksberg's conjecture on uniform algebras on S complemented in C(S) and related problems. Recall that a uniform algebra on a (compact Hausdorff) space S is a closed subalgebra which separates the points of S and contains the constant functions. We begin with the case of a uniform algebra whose annihilator is norm separable. Our first proposition combines a result of [P4] with a recent observation due to Wojtaszczyk

[WI] . For p E Xl = {v E [C(S)] *: Ix dv = 0 for x E X} we define Til: X --+ L 1 (Ipi) by Tix) = xg for x E X, where g = dp/dlpl is the Radon-Nikodym derivative of p with respect to Ipl.

5.1. Let X be a uniform algebra on S with a norm separable annihilator Then the following conditions are equivalent:

PROPOSITION

Xl

c

[C(S)]

*.

(i) X = C(S). (ii) If p E Xl, then fJ. is purely atomic. (iii) If p E Xl, then Til is compact.

(i) ~ (iii) obvious. (iii) ~ (ii) (Wojtaszczyk). Assume that there is a non-purely-atomic p E Xl, so that PROOF.

Til is compact. Then TZ: L DO (1fJ.i) --+ X* is also compact. Since fJ. is not purely atomic, there is a sequence (rn) in L DO(IfJ.i) which has the same distribution as the sequence of Rademacher functions in L 1 [0, I]. Observe that (5.1)

fr

n

-w*

0

as n -+

00

for every fE LDO(IfJ.i)

(here "~" denotes convergence in the a(L" (luI), L 1 (Lui)-topology). Since T! is weakstar continuous and compact, (5.1) yields

(5.2)

limIlT~(frn)lIp n

=0

for every fELDO(IfJ.I).

Using (5.2) we define inductively an increasing sequence of the indices n(I)

so that (5.3)

IIT.~(w,· " " )lI • ,1'2"·"$ x

< 1/3 k

28

for j.~

=k

(k

= 1,2, ...

)

< n(2) < ...

29

BANACH SPACES

where

s

Wilh.···.is

kQ rn(lk)

for any finite increasing sequence of positive integers

Uk)l<;;k<;;S'

Since X* can be identified

with [C(S)] */Xl. it follows from (5.3) that for every wil.i2 ..... i s there is a Vil.h ..... i s E Xl such that Ilw.

(5.4)

. glpl - v·

. IllC(S)I' 'I' .. ·.'s

'I····"s

(because T~( wi 1 '

. . . , is)

< 3 -i s,

= {w'l' ... 'i sglp I + Xl}). Let F del ~te the closed linear subspace

of L 1 (Ipl) spanned by all the "like-Walsh" functions wiJ h .... ,is. It is easy to see that E is isometrically isomorphic to L 1 [0, I], by the isometry which assigns to the Walsh functions (in the Paley order) the functions (w'lJ .....n k)' Since g = dp/dlpl is a unimodular function, the sequence (gw n I .... ,n k) is also equivalent in L 1 (Ipl) to the Walsh-Paley sequence in L 1 [0, I] and the subspace gE is isometric to L 1 [0, I]. Now (5.4) yields that the perturbed sequence

(vi 1.i2 , .... is) is also equivalent to the Walsh-Paley sequence in L 1 [0, I], i.e., the map wi! ,.... i s --+ Vil ..... i s extends to an isomorphic embedding of E into Xl (because the subspaces Ek = span {wil ,i2 ,.... is with js = k} form a Schauder decomposition of 1:' and dim Ek = 2 k - I ). Hence Xl contains a subspace isomorphic to L 1 [0, I] and therefore, by a result of [PB] , is not isomorphic to a separable dual. On the other hand, Xl is isometrically isomorphic to the dual of C(S)/ X. Thus Xl is not separable, a contradiction. (ii) => (i) [P4]. Case I: S metrizable. If p E xl, then, by (ii), p =

L

as(p)1]s

sES

where 1]s denotes the unit mass at the point sand asCp) are the complex numbers with Iipli = ~sEslaip)l.

Let M(X) denote the Choquet boundary of X. Then, for every s E M(X), there is an

Xs E X such that xis) = I and Ixs(t) I < 1 for t

°= f lim n

Therefore if p E

xl,

S

x: dp

then Ipl(M(X))

= lim n

= 0.

"* s (cf. [Ph I]).

L

tES

aip)xn(t)

Hence

= aip)·

(Here we use the fact that if S is metrizable then

M(X) is a Go and therefore measurable.) To complete the proof in Case I, it is enough to show that M(X) = S. This combined with the previous remark will imply Xl = {o}, i.e., X = C(S). Suppose to the contrary that there is an s in S\M(X). By the Choquet-Bishop-de Leeuw representation theorem [PhI] there is a v E [C(S)] * concentrated on M(X) such that 1]/x)

= xes) =

f x dv = ( S

JM(X)

x dv

for x EX

.

1]s E Xl, and, by the preceding remark, v(M(X)) = p(M(X)) + 1]/M(X)) = f M(X) I dv = I, a contradiction. Case II: S arbitrary compact Hausdorff space. Then C(S)/X is separable because

Hence, p

=v-

On the other hand v(M(X))

= 0.

30

ALEKSANDER PELCZYNSKI

(C(S)/X)*

Xl is separable. Hence a standard construction yields the existence of a compact metric space SI and a continuous map h: S~ SI such that if he: C(SI) ~ C(S) is defined by ho(n = f ° h and if XI = {fE C(SI): hO(n E X} then the map H: C(Sl)/XI ~ C(S)/X defined by H( {f + XI}) = {hOCn + X} is an isometric isomorphism onto C(S)/X. Next we check that the operator (ho)*: [C(S)] '" ~ C(SI) takes purely atomic measures on S onto purely atomic measures on S I and restricted to the annihilator of Xl coincides with the isometric isomorphism H*: Xl ~ xt. Thus XI is an algebra which satisfies the assumption of Case I. Hence X = {O} and therefore Xl = {O}. 0 REMARK. Chevalier [Chv] has extended the argument of Case 1. He showed that for every compact Hausdorff space S, C(S) is the only uniform algebra on S with a purely atomic =

t

annihilator. Proposition S.l allows us to extend Theorem 4.2 to the case of an arbitrary proper uniform algebra X with a separable annihilator. We replace A by X, H~ by Xl ( a separable dual!), L I (m)/H~ by L I (A)/Xl where A E C(S)t is any separable probability measure such that Xl eLI (A) and such that whenever /1 ELI (A) and /1 is singular to then /1

= 0; H""

Ivl for every v e Xl,

is replaced by (Xl)l C L "" (A). The natural embedding i~,1 is replaced by

any operator i:,I' where the latter is defined to be TJ1. regarded as an operator from X into

Xl whenever /1 E Xl is chosen so that TJ1. is not compact; this is possible, by Proposition S.1. In particular we have COROLLARY S.l. Let X be a uniform algebra on a compact Hausdorff space S with a norm separable annihilator. Then (a) There is a measure /1 E Xl such that the operator i:,1 : X ~ Xl is absolutely summing, noncompact and non-L I-factorable. (b) X and all duals of X do not have G-L l.u.st. (c) X is not isomorphic to a quotient of any C(K)-space; in particular, X is not complemented in C(S).

Next we shall show that the properties of the disc algebra exhibited in §4 are shared by a large class of uniform algebras; every uniform algebra with a nontrivial Gleason part belongs to this class. Unfortunately the technical condition which characterizes the class is slightly complicated. DEFINITION

S.1.

Let X be a uniform algebra on S. Then p is called a Hardy measure

for X if there is a triplet (p, f n , F) such that (S.5)

p is a probability Borel measure on S,

(S.6) (S.7)

fn EX,

F(s)

= lim fn(s)

IIfnll <; 1 and

for n = 1,2, ... ,

IF(s) I = 1

for p-almost all s,

n

(S.8) Our next proposition explains the role of Hardy measures. It is the main technical result of the present section.

BANACH SPACES

31

PROPOSITION 5.2. Let X be a uniform algebra on S. Assume that there exists a Hardy measure p for X. Then there are operators T: X ~ Xl and Q: Xl ~ H~ such that

(i) Q is a contractive projection, that is, IIQII ,,;; I and there is an isometric embedding

J: H~ ~ Xl such that QJ = idHI. (ii) The natural embeddini i~ ~I : Ao ~ H~ factors through (QD**, i.e., there are operators U: Ao ~ X** and 1T: (Hb)** ~ H~ such that i~?l = 1T(QD**U. (iii) T and QT are absolutely summing, noncompact and non-L I-factorable operators. PROOF. Let (p, fn' F) be as in Definition 5.1. Put T(x) = xFp for x E X. T(x) E Xl. Clearly T: X ~ Xl is absolutely summing and IITII ,,;; 1T(T)

=

By (5.8),

1. The construc-

tion of Q is more complicated. Given f E C(aD),! denotes the continuous function on the unit disc D which extends f and is harmonic for Iz I < I. Now fix n = I, 2, . . . and assign to every v E Xl the unique measure a E [C(aD)) * which represents the linear functional

faD fda Note that, by (5.6),1

0

=

(!Un(s))v(ds)

Js

for fE C(aD).

fn is well defined and if f E A [regarded as a subspace of C(aD)),

then! is the analytic extension of f onto the interior of the unit disc; hence it admits uniform approximation by polynomials in z. This implies that!

0

fn E X; hence f aDf da = 0

(because v E Xl). Thus by the M. and F. Riesz theorem, a = h . m for some h E H~. We put Qn(v)

= h.

Clearly Q n : Xl ~ H ci is a linear operator with IIQ n II ,,;; I, for n

=

1, 2, ....

Let us put Q(v)

= LimQn(v)

for v E Xl.

n

Here Limnh n denotes a fixed Banach limit of a bounded sequence (h n ) with respect to the a(H~, C(aD)/A)-topology. Since H~ is the dual of C(aD)/A, the Banach limit exists and limnh n is a cluster point of the set U;=I {h n } in the a(H~, C(aD)/A)-topology. Clearly Q: Xl ~ H~ is a linear operator with IIQII ,,;; 1. Moreover if v E Xl

n LI(p),

then Qn(v) ~ Q(v)

in the a(H~, C(aD)/A)-topology;

(5.9)

and if g

= dv/dp,

f aD Q(v)· f dm =

then, for every f E C(aD),

fs f(F(s))g(s)p(ds).

Indeed, given f E C(aD), by (5.7), we have limnfUn(s))

= !(F(s)) = f(F(s))

for p-al-

most all s. Hence, by the Lebesgue theorem,

Ii:

faD Qn(V) . fdm

Thus f aDQ(v)f dm

=

Ii:

fs!(fn(s))g(S)P(dS)

= f sf(F(s))g(s)p(ds) because

= fsf(F(s)lg-(s)p(ds).

a w*-convergent sequence has a unique w*-

cluster point. Next we construct an isometric embedding J: H~ ~ Xl with QJ = id I

first that q E Ho is a polynomial, say q chJck that

k'

= r,j=laj . z'.

We put J(q)

= q(F)

I. HO

Assume

. p. We shall

32

ALEKSANDER PH.CZyNSKI

J(q) E

(5.10)

xl,

(5.11)

(F k )k=O.±1.±2•... is an orthonormal sequence in L 2 (p).

(5.12)

IIJ(q)1I 1 ~ IIq1l1.

E X, f~Fk dp = limnfsxl~-1Fdp = o. This proves (5.10). By (5.7), F(s)-1 = F(s) for p-almost all s. Thus, if k = ±I, ±2, ... , then fF- k dp = fFk dp = O. This yields (5.11) because by (5.5) fFkF- k dp = fl dp = 1. The verification of (5.12) is a little bit tricky. By (5.11), J extends to an isometric embedding, say], of H~ into Xl. By the factorization theorem (cf. [Du, Chapter 2]), given a By (5.8), Fp E Xl; if k

> 1, then, by (5.7), given x

polynomial q E H~ there are 11 and 12 in H~ with q

~ (fI11(F)12 dp

IIJ(q)lIl = flq(F)1 dp

= 1112

r

and IIqll1

= 111111 2111211 2.

/2 (II12(F)1 2 dp

r

Thus

/2

= 11]([1)11 211]([2)11 2 = 11111121112112 = Ilqll1· This proves (5.12). It follows from (5.12) that J extends to an operator (which will be denoted also by J) of norm ~ 1 from H~ into Xl.. To complete the proof of (i) it is enough to show that QJ(zk)

= zk

for k

=

1, 2, . . . . This will imply that QJ(h)

=h

for h E H~

and therefore IIJ(h)11 ;;;. IIhll, which together with (5.12) yields IIJ(h)1I = IIhll. Clearly J(Zk)

= Fkp for k

= 1,2, . . . .

Letl(z)

= "i:,j=_rajzj

and let r >k. By (5.9) and (5.11) we get

faD Q(Fkp)ldm =a- k = f aD l(z)Zkm(dz). Hence QJ(zk) = zk for k = 1, 2, . . . . This proves (i). We identify X with its canonical image in X* * .

1*

Let 1* =

Z -1 I

and Un (t) =

In for IE Ao. Since the disc algebra operates on every uniform algebra, Un: Ao X is a well-defined linear operator with II Un II ~ 1, for n = 1, 2, . .. . Let us put U(t) = Limn Un (f) for I E Ao where "Limn" denotes here the Banach limit in the a(X**, X*)topology. Finally define "IT to be a contractive projection from (H~)** onto H~. (Since H~ is a dual Banach space, the "IT in question exists.) Since T is weakly compact, (QT)**(X**) C H~. Thus "IT(QT)**(X**) = (QT)**(X**) E X**. To complete the proof of (ii) it suffices to show that (QT)**U(zk) = zk for k = 1, 2, . .. . Fix k. By construction, U(zk) is a a(X**, X*)-cluster point of the sequence (Ur(zk». Thus (QT)**(U(zk» is a a((H~)**, (H~)*)-cluster point of the sequence (QT**(Ur(zk». Since (QT)**(X**) :::> H~, it is in fact a a(H~, qaD)/A)-cluster point. We have (QT)**(Ur(Zk» = QT(f~-1) = Q(fr-1 Fp). Hence, by (5.9), for every IE qaD), 0

----l-

f

JaD

Q(f~-1Fp)ldm =

J.S1~-1FI(F)dp

for r

= 1,2, . . . .

Now, by (5.7), we infer that the sequence (Q(f~-1 Fp» converges in the a(H~, qa)/A)-topology to the functional zk (we have limrfsl~-1 FI(F) dp = fsFkl(F) dp = faDj(z)zk m(dz); the last equality has been verified in the proof of (i». This proves (ii).

BANA.CH SPACES

33

Since T is absolutely summing and i~?1 is a noncompact and non-L I-factorable operator, (iii) follows directly from (ii), Remark R.I in §4 and the fact that a second adjoint of a compact operator is compact.

0

COROLLARY 5.2. If a uniform algebra X admits a Hardy measure, then there is an absolutely summing surjection from X onto 12. PROOF. Let P: H2 ~ H2 be a Paley projection such that P(H 2 ) C H~. It follows easily from Definition 3.1 that P(H2) is a closed linear subspace isomorphic to 12 and P extends to a bounded linear projection, say P, from HI onto P(H 2 ). The desired operator is

PQT. It is clearly absolutely summing. By (ii), (PQT)** is a surjection. Since P and Tare weakly compact and the unit ball of X is dense in the unit ball of X** in the a(X**, X*)topology, it follows that PQT maps the unit ball of X onto a set whose norm closure contains a ball of P(H~). Hence PQT is surjective. 0 Combining Proposition 5.2 with an obvious adaptation of the argument used in the proof of Theorem 4.2 we obtain THEOREM 5.1. Let X be t1 uniform algebra on a compact Hausdorff space S. Assume that X admits a Hardy measure. Then G) X and all the duals of X do not have G-L l.u.st (jj) if X is a quotient of a Banach space Y with G-L l.u.st, then Y contains a complemented isomorph of 11 ; (jjj) if C(S)*/X1 is a subspace of a Banach space Z with G-L l.u.st, then Z contains

I; uniformly; (jjjj) if C(S)*/X1 is isomorphic to a subspace of a Banach lattice Z, then Z contains an isomorph of co. 0 COROLLARY 5.3. If X is a uniform algebra on S which admits a Hardy measure, then X is not complemented in C(S). 0 Our last result gives a sufficient condition in order that a uniform algebra have a Hardy measure, in particular this is the case if a uniform algebra X has a non-one-point Gleason part, i.e., if there are two different homomorphisms of X into complex numbers, say ..p and "', such that sup IIxll=l,xEX

I..p(x) - 1/J(x) I < 2.

PROPOSITION 5.3. Let X be a uniform algebra on a compact Hausdorff space S. Let us consider the following three properties: (l) X has a Gleason part which contains at least two different points. (2) There exists an ideal leX and a multiplicative linear functional ..p on X such

that if a = 11..pllll = SUPtEJ, IItll = 1 1..p(f)1, then 0 (3) There is a Hardy measure for x. Then (l) => (2) => (3).

< a < 1.

34

ALEKSANDER PEtCZYNSKI

(I) => (2). If

PROOF.

J

= ker t/I

t/I, then

(2) => (3). Let 11 be a Borel measure on S such that If dll =
a

= lim
lim I fn dll· n

To complete the proof we shall check that the triplet (p, fn' F) satisfies the conditions (5.5)(5.8) of Definition 5.1.

+ a 11l1-aimost everywhere. Hence 1111 IF(s) I = 1 p-almost everywhere. Since IIfn II

Clearly IF(s) I = 1 and I - a ~ 11 - aF(s) I ~ 1 is absolutely continuous with respect to P and 1 and

~

0< a

= 111111 = lim Ifn dll = lim n

n

f ;!, dllll,

it follows that limnfn(s) = F(s) 11l1-almost everywhere and therefore p-almost everywhere. Finally given f E X we have

IIplli IfF dp

=

IfFll - aFl 2 dill I = I jF211 - aFI2 dll

= I fF2(1

- aF){l - aF) dll

= lim I f(f~{l + a2 ) - af~

- afn] dll

n

n

Thus Fp E Xl. This completes the proof.

0

Notes and remarks to § §4 and 5. The results of § §4 and 5 are closely related to the following problem due to Glicksberg [GI2].

Problem 5.1. Is C(S) the unique uniform algebra on a compact Hausdorff space S which is complemented in C(S)? Glicksberg (using Rudin's averaging technique [Rul]) proved that the answer is "yes" for S being a homogeneous space of a compact topological group G and for uniform algebras on S which are invariant under the action of G on C(S): (Tgf)(s) g E G,

= f(gs)

for f E C(S),

s E S. For details and generalization to locally compact groups cf. [G12] and [RI].

It was shown in [P4] that the answer is "yes" for an arbitrary compact Hausdorff space S and for uniform algebras on S with norm separable annihilators (cf. our Corollary 5.1). For uniform algebras with nontrivial Gleason parts the same fact (cf. our Corollary 5.3) was proved by Kisliakov [Kis 1] and with some additional assumptions by Etcheberry [Etch]. While Glicksberg's problem is still open, it seems to be more natural to ask

BAN I\CH SPACES

35

Problem 5.2. Conjecture. Let X be a proper uniform algebra on a compact Hausdorff space S (i.e., C(S)\X is nonempty). Then (a) X as a Banach space is not isomorphic to any C(K)-space (b) X is not isomorphic to any complemented subspace of a C(K)-space (c) X is not isomorphic to any quotient space of a C(K)-space (d) X does not have G-L l.u.st. It was observed in [P5] that if Sis metrizable and X is a uniform algebra on S which

is complemented in a C(K)-space then X is isomorphic as a Banach space to C(S). Varopoulos [Va] proved recently that if a semisimple algebra X is isomorphic as a Banach space to a C(K)-space, then X is an operator algebra. Bya result of Bishop (cf. [B2], [Gml, pp. 6062]), it follows that every uniform algebra has a quotient antisymmetric algebra; thus in (c) one may assume without loss of generality that X is antisymmetric. It is not clear whether the criterion of Proposition 5.2 can be applied for uniform algebras with norm separable annihilators. This suggests the following

Problem 5.3. Let X be a uniform algebra with a norm separable annihilator. Does X admit a Hardy measure? Does X have a non-one-point Gleason part? Another approach to attacking Problem 5.2 is to eliminate the assumption of the norm separability of the annihilator in Proposition 5.1 (cf. the Remark after the proof of Proposition 5.1). This suggests Problem 5.4. Let X be a proper uniform algebra on S. Does there exist a /1 E Xl such that the operator Til: X - L I (1/11) is not compact where T/x) == xdp/dipi for x EX? Theorem 4.2(i) was proved in [P7] while (ij) and (iii) were observed in [P6], (iv) was first proved using a different technique of [Pisl] by Pisier, (v) is due to W. B. Johnson (oral communication). The equivalence (iii) <=> (i) of Proposition 5.1 is due to Wojtaszczyk [W I] , while (i) <=> (ii) is taken from [P4]. The concept of a Hardy measure is implicitly contained in Kisliakov's work [Kis I]. Proposition 5.2 in the present form was stated and proved by Wojtaszczyk. A slightly weaker fact was established in [P7]. Proposition 5.3 goes back to Bishop and Wermer (cf. [Gm 1, Chapter VI, proof of Theorem 7.1]). ADDED IN PROOF.

[Kis3].

A similar result to Proposition 5.2 has been obtained by Kisliakov

6. Uniformly Peaking Families of Functions in A and

H~ .

The Havin Lemma. This section has a preparatory character for the next one. The main result is a very technical Proposition 6.1 due to Havin [Hvl]. This is a generalization of a quantitative character of the fact that every closed subset of the unit circle of Lebesgue measure zero is a peak set for the disc algebra (cf. § 1, Step I in the proof of the F. and M. Riesz theorem). Roughly speaking it is shown that if a closed subset e of the unit circle has a small Lebesgue measure then there is a function in the disc algebra of norm one whose values are close to zero except for a set of small measure and are close to one at each point of e. Moreover all the estimations on the measure of the exceptional set and how the values of the function are close either to zero or to one depend on the measure of e only. A similar construction leads to a result of Amar and Lederer [A-L] on peak sets for Hoc (Proposition 6.2) and to a characterization of exposed points of Hoc (Theorem 6.1). Recall that m denotes the normalized Lebesgue measure on the unit circle aD = {z E C: Izl = I}. Given a measurable function u: aD ---'> C, we denote (if it exists) by H(u) the Hilbert transform of u (cf. §O.II) and by Ii the harmonic extension of u onto D, i.e.,

PROPOSITION 6.1 (THE HAVIN LEMMA). There exists a positive function 8 ---'> €(8) for 0<8 < 1 with lim.s=o€(8) = 0 such that given a measurable subset e of aD with 0 < m(e) < 8 there exist fe and ge in HOC with the following properties: (6.1 )

(6.2)

(6.3) (6.4)

Ife(z)1

sup Ife(z) -

+ Ig/z)1 = 1

for z E aD,

11 = 0(8 1 / 2 ),

zEe

sup Ige(z)1

zEe

= 0(8 1 / 2 ),

f

aD

Ige(z) -

11 dz < €(8),

if e is closed, then fe and ge are in A.

PROOF. Step I. Construction. Fix e C aD measurable with m(e) < 8 B = 8- 1 / 2 , b = 8 1 / 2 . If e is not closed we define we = w by 36

~

1 and put

BANACH SPACES

w(z)

-B

={

-b

37

if z E e, if z f- e.

If e is closed we first pick finite families of closed intervals (.6.;)1 ~i~N and (.6./')1 ~i~N such that if e' = UN .6.~ and e" = UN .6.~' then J= 1 J J= 1 J

e C e',

(6.5)

.6.~

m(e")

< 0,

.6.; C int.6./'

n .6.; = .6.~ n .6./, = ¢

and

= 1, 2,

for k =1= j (k, j

... , N),

and we define w to be a C~ -function on aD such that -B ,,;;; w(z) ,,;;; -b for all z E aD and

w(z)

=

{

-B

for z E e'

-b

for z

' tf. e".

Next we put

v

= H(w);

h

= w + iv;

fe

= exp h- I .

We shall check

h(z)

=1=

for zED,

0

If/z)1 ,,;;; 1

for z E aD,

(6.6)

faD 10g(1

- Ifel) dm

>

-00.

Then we define ge to be the outer function with Ige(z)1

=

1 - Ife(z)1 for z E aD almost

everywhere. Precisely we put -

(6.7)

1 rr

ge(z) = exp - 2 ge(z)

=

S21T 0

lim ge(Pz)

tt. + z 10g(I £It _ z

.

- Ife(e't)1 dt

for Izl

< 1,

for z E aD almost everywhere.

p=1

Step II. Verification of (6.6). By the maximum principle for harmonic functions, -B";;; Re h(z) = w(z) ,,;;; -b

for zED.

Hence h(z) =1= 0 for zED and Re

h- 1 =

_-,-w'--_ ";;;-b <0.

W2+lJ2

Since wE L ~, v = H(w) exists and v E LP for all p (cf. §O.II). Thus

< 1.

11 and h -I

exist and

I; hence fe E H~ and Ife(z)I";;; 1 for z E aD almost everywhere. It remains to show that f aD(I - Ife I) dm > -00; this will imply that ge defined by (6.7) does exist and belongs to H~ (cf. [Du, Chapter 2]). Let Zo = {z E aD: V(z)2 ,,;;; 2} and Zk = {z E aD: 2k ,,;;; V2(z) < 2k+1} for k = 1, 2, . . . . Observe that there is a C> 0 such that m(Zk) < CT k for k = 0, I, ... because vEL 2 • If z E Z k' then are analytic for Izl

Thus

Furthermore leh-II ,,;;; e Reh -

1 ,,;;;

ALEKSANDER PEl.CZyNSKI

38

flog(l - If I) dm

=

£. f

k=O

log(l - lexp h-l(z)l) dz

Zk

Step Ill. Verification of (6.4). If e is closed, then wand therefore v are COO -functions < Ife(eit)1 < 1 is a 21T-periodic C oo _ function, hence ge E A (cf. [Du, Chapter 5]). Step IV. Verification of (6.1) and (6.2). The formula (6.7) yields that Ige(z)1 = 1 Ife(z)1 for z E aD almost everywhere (cf. [Du, Chapter 2]). This yields (6.1). If z E e, then

(cf. §O.II). Hence h, h- I and fe are in A. Moreover 0

_1__ Ih(z)1 Hence remembering that B 11 - fiz)1

~

1

v'w

2 (z)

+

v2 (z)

_1_ _ - I "'" lw(z)1 - B

> 1, we get = 11

- exp h-l(z)1 ~ explh-l(z)1 - 1

~ Ih-l(z)lexplh-l(z)1 ~ B-Iexp(I)

= 0(0 1 / 2 ).

The proof of the second inequality of (6.2) is slightly harder. To treat simultaneously the case where e is a closed set and e is not closed we put e* = e" if e is closed, and e* = e otherwise. Let X

=

{z E aD: Iv(t)1 ~ 77} for some 77

> O.

Since the Hifbert transform is of

the weak type (1.1) there is an absolute constant C I such that

m(X)~Cl17-1f

aD

because faDlwldm

Iwldm~ClBo17-1 +C1 b77- 1 ,

= faD\e.lwldm + fe.lwldm

~Bo

+ b.

Thus, taking into account that

if z ($ X U e* then

Ife(z)1

= exp Reh-I(z) = exp -b

~exp

b2

we get

faDlfel dm

w 2(z)

w(z)

+ V2(z)

~b+172b-l,

+ 17 2

= f)fel dm + fx\e.lfel dm + faD\Xue.lfel dm ~ m(e*)

+ CI(B77-IO + b77- I ) + b + 17 2 b- 1

~ 0(0 1 / 6 ), provided we put 77 = This completes Step IV. Step V. Verification of (6.3). Let ue = 1l0g(J - Ifel)l. Then ge = exp(-u e - iH(u e

01/ 3 .

»·

The crucial point is to show that there is a constant M such that

BANACH SPACES

JaD U; dm < M r

(6.8)

39

for all measurable e

< 3D.

Having done this we complete the proof as follows: By (6.1) and (6.2), ue whenever m(e) ---+ O. Thus, by (6.10), faDlueldm ---+ 0 whenever m(e) Hilbert transform is of weak type (1.1), the last relation yields

m{z E 3D: IH(ue)(z) I

---+

---+

0 in measure

O. Since the

~ (faDIUel dm y/2} .; ; C1 (IaD IUel dm y/2

where C 1 is an absolute constant (cf. §O.lI). Thus if m(e)

---+

0 then H(u e )

---+

0 in measure

1 in measure. Since Ilgell"" .;;;; 1 for all measurable e < 3D, we conclude that flg e - 11 dm ---+ 0 whenever m(e) ---+ 0 which proves the second inequality of (6.3). The

and ge

---+

first formula of (6.3) follows directly from (6.1) and the first formula of (6.2). To establish (6.8) put

=

y

Clearly 0


00

.ll

IRe

h

and

ue

= log

Iwl

+ v2

.

(1)1/6

+ y-l)1/3

Jy-I/3 dm

w2

1 1 - exp(-y) .;;;; 6 1 - exp(-y)

=6(1+ Therefore u; .;;;; 36 (I

=

1

exp(y) - 1

.;;;; 36 (I

)1/6';;;;6(l+y-l)I/6.

+ y-l/3).

Finally

= J(lv2 1 + IWI)I/3 dm .;;;;J~dm + Jlwll/3 dm. Iwll/3

w

We estimate each term separately. By Kolmogorov's theorem (cf. §O.II), there is an absolute constant C2/3 such that flvl 2/3 dm .;;;; C2 / 3(flwl dm)2/3. Hence

2/3 dm';;;; b-1/3C2/3 (JIwl dm )2 / 3 J-l/3 w V

~ C ~-1/6(2d/2)2/3 "'" 2/3 u U

while flwl 1/3 dm ';;;;B 1/ 3S

+ bl /3

.;;;;

2.

~

22/3 C

2/3'

"'"

0

€(S) which follows from the argument in Step V seems to be unsatisfactory. We get only €(S) .;;;; Cilogil - eoal 12a REMARK.

for a

The estimation for the order of the function S

---+

< 1/2. The next result, due to Amar and Lederer [A-L], uses a similar technique to that

of Havin's lemma. It is, however, formulated in a different language, which we now recall. Let A be the maximal ideal space of L"" and let

f

---+

f be the algebraic isometric iso-

morphism from L"" onto C(A). Recall that A is a compact Hausdorff space. If Xe is the characteristic function of a measurable subset e of 3D, then e = (t E A: Xe(t) = I} is a

40

-\LEKSANDER PELCZYNSKI

e

m

clopen set; all the sets form a basis for the topology of Ll. By we denote the unique regular Borel measure on Ll which corresponds via the RIesz representation theorem to the functional

f ~ II dm, i.e., mis defined by

f j dm = f I dm

for

IE L

00 •

H oo can be regarded as the uniform algebra on Ll, in fact Ll is the Shilov boundary for ~ (cf. [H, Chapter 10)). Now we are ready to state PROPOSITION 6.2. Let F be a closed Go in Ll with m(F)

= O.

Then F is contained in a nontrivial peak set lor H , i.e., there exists a nonconstant I E ~ such that 11/11 = 1, F C j-I(1) = {t E Ll: f(t) = I} and j-I(1) = {t E Ll: I/(t) I = I}. oo

PROOF. The assumption on F yields the existence of a decreasing sequence (en) of

n;=

measurable subsets of aD such that F = len and m(en ) = m(en ) ~ 0 as n ~ 00. Passing to a subsequence, if necessary, we may assume without loss of generality that e1

m(en ) <

n- 3

for n

= 1,2, ....

= Ll,

Now we put 00

L - n Xe n\en+1 '

w

=

h

= w + iv,

v = H(w),

n=1

1= exp(1/h).

Since wE LP for p < 2, h E 1fP. Thus h is a boundary value of an analytic function, say h in the unit disc. Furthermore, Re h . ;;; -1 , because Re h ..;;; -1. Thus h

"* 0 and I is a nonconstant

~ function with 11/1100 ..;;; 1. Furthermore if t E F then f(t) = 1 because for z I/(t) - 11 = lexp(1/h) - 11";;; lexp{l/Ihl) - 11";;; e lln - 1. Hence

00

[-1(1)

=

n

E en'

00

{t E Ll: If(t) - 11 ..;;; e lln - I} ::)

n=1

n

en

= F.

n=1

Finally replacing if necessary I by Yz(f 6.2. 0

+ 1), we

obtain the last assertion of Proposition

It is interesting to compare Proposition 6.2 with the next one. PROPOSITION 6.3. There is a closed nonempty G/j-set Fell with m(F) not a peak set lor H oo •

= 0 which

is

PROOF. Suppose that the assertion of Proposition 6.3 is false. Then given a closed G/j -set Fell with m(F)

= 0 there exists an IF E H

oo

such that

iF

peaks exactly on F.

Hence

(6.9) Indeed, by regularity of J.l, it is enough to observe that J.l(F) C Ll with m(F)

= 0".

=0

for every closed G/j -set F

The last follows from

J.l(F)

f

= lim j'j. dJ.l = 0 n

(because fr;. converges pointwise to the characteristic function of F). The map I ~ f for induces an isometric isomorphism from L l(m) onto L I (m) which carries H~ onto

IE L

00

41

BANACH SPACES

LI(m) n (H~)l. Now (6.9) implies that the annihilator (H~l coincides with LI(m) n (~)1; hence (H~)l, being isometrically isomorphic to Hb, is norm separable. Since (L ~ /~)* is isometrically isomorphic to (H~)l, we infer that (L ~ /H~) is separable. This leads to a contradiction because (Hb)* is isometrically isomorphic to L ~ /H~ and Hb contains a subspace isomorphic to II. 0 Problem 6.1. Characterize the subsets of the Shilov boundary of H~ which are peak sets for H~. We end this section with an application of Proposition 6.2 of a geometric character. We shall characterize the exposed points of the unit ball of H~. Recall that if X is a Banach space then an x E X with IIxll = I is an exposed point of the unit ball B X = {y EX: Ilyll <: I} if there is an x* E X* with Ilx*11 = 1 which strictly supports Bx at x, i.e., x*(x) = 1 and Rex*(y) < 1 for every y E Bx\{x}. BH~

THEOREM 6.1. A function f E iff

(6.10)

m(£)

>0

H~

with Ilfll

= 1 is

an exposed point of the unit ball

where E == {z E aD : If(z)1 == l}.

PROOF. Let f be an exposed point for B H~ and let x* E (H~)* be a linear functional which strictly supports B H"" at f. Let <1>* E (L ~)* be a norm preserving extension of x*. Given measurable e c aD we denote by <1>: the linear functional on L ~ defined by :(g) = *(Xeg) for gEL ~. Clearly <1>* = <1>: + ~D\e and 11<1>*11 = 11<1>:11 + 1I~D\ell. Furthermore if e satisfies the condition If(z) I <: a < 1 for z E aD\e almost everywhere, then <1>: = <1>*. Indeed, otherwise 1I~D\ell > 0 and

a contradiction. Now let en = {z E aD: 1 - If(z) I < n- I }. By the previous remark :n = <1>* for n = 1,2, . . . . Clearly e l :J e 2 :J ... . Let e~ = new To show thatfsatisfies (6.10) it is enough to establish that m(e~) > O. Suppose that m(e~) = 0; then F = n;=len is a closed Go-subset of ~ with m(F) = O. Thus, by Proposition 6.2, there is a nonconstant fF E H"" with IlfF11 = 1 such that iFI (I) :J F. For n = 1, 2, ... , we have 1*(n - *(fFnl == <1>;

n

«(1 - fF)n

<: Ilill sup 11 - fF(z)l. zEe n

n;

Taking into account thatel:J e2 :J ... :J F = = I en and iFI (I) :J F, we infer that limnsuPzEen 11 - fF(z) I = O. Hence *(n = *(f. f F ) which contradicts the definition of the exposed point, becausefF being nonconstant implies thatf· fF EBH",,\{f}. Hence m(e",,) > o. Now assume that anfEH"" with Ilfll == 1 satisfies (6.10). Let <1>* E (L"")* be defined by

*(g)

=

m(E)-l fEg(z)signf(z)m(dz) for gEL"".

ALEKSANDER PElCZYNSKI

42

Let x* be the restriction of <1>* onto H~. Clearly IIx*1I = Ix*(f)1 = 11<1>*11 = 1. Hence Re x*(g) :(; 1 for g E B H~' Finally if, for some g E B~, *(g) = x*(g) = I, then Ig(z)1 = 1 and g(z)sign/(z) = 1 for z E E almost everywhere. Hence g(z) = I(z) for z E E almost everywhere. Since m(£)

> 0, we

infer that

I

= g.

0

REMARK. Note that if I is an exposed point for B H~ then a linear functional which strictly supports B H~ at I can be chosen from L 1/H ~, a predual of H~.

Notes and Remarks to §6. The proof of Havin's lemma is taken with some modification from [Hvl]. The condition (6.4) of the lemma was observed by Kisliakov [Kis2] and independently by Pe1czynski (unpublished). Wojtaszczyk [W2] observed recently that Havin's lemma can be generalized to uniform algebras with unique representing measures for linear multiplicative functionals. The proof of Proposition 6.2 is taken from [A-L]. The construction is classical (cf. [Z, Vol. I, p. 105]). Proposition 6.3 shows that the "Remarque" in

[A-L] is false. Theorem 6.1 is due to Amar and Lederer [A-L] and Fisher [Fi]. An analogous result for the disc algebra is due to Phelps [Ph2]. Let us recall that an IE BH~ is an extreme point of B H~ iff I aDlog(l - III) dm = --00 (cf. [H, p. 138]). Finally observe that no point of BH~ is strongly exposed, i.e., there is no IE B H~ and x* E (H~)* with x*(f) = 11/11 = IIx*1I = 1 and such that for every sequence

(fn) in B H"" if x*(fn) --+ x*(f) then II/n - III --+ 0. Indeed, regard H~ as a subspace of C(Ll) and note that for every f.1 E C(Ll)* there exists a peak set, say F, for H~ with f.1(F) = 0. Suppose IF E H~ peaks exactly at F, i.e., {s E Ll: iF(s) = I} = {s Ell: liF(s) I = I} = F. E BH~ put

Given

I

for n

= 1, 2,

In

= 1(1 - f~·)/ill

-

I~II for n = 1,2, . . . . Then

limnilln -

III = 11/11

... while limnI/~ df.1 = Iidf.1. Various interesting properties of H~ and L 1 /H~ related to the material of this section

are discussed in the expository paper by Havin [Hv2]. ADDED IN PROOF.

Delbaen [De3] recently generalized Havin's Lemma for a represent-

ing measure of a linear multiplicative functional on a uniform algebra such that the representing measures of the functional form a weakly compact set.

7. Characterizations of Weakly Compac(Sets in L 1 /Hb and in A * In this and in the next section we shall show that, despite the results of §4, the spaces

A. A * and L 1 /Hb share various properties of C(aD), C(aD)*, and L 1. In particular a specific characterization of weakly compact sets in L 1 -spaces [Gr2] , [P9] as those sets on which weakly unconditional series of functionals converge uniformly can be carried out to L 1 /Hb and A *. Another peculiar property of a weakly compact set in L 1 /Hb says that the set is an image under the quotient map of a weakly compact set in L 1. The main result of this section, Theorem 7.1, was proved recently by Delbaen [D I] and independently by Kisliakov

[Kis2] . Recall that a sequence (xn) of elements of a Banach space X is w.u.s. (weakly uncon-

ditionally summable) if :E:=llx*(xn)1 < 00 for every x* E X*. Clearly if (x n ) is w.u.s. then there is a K > 0 such that :Elx*(xn)1 ~ Kllx*1I for x* E X*. A sequence (yn) in X is weakly Cauchy if limnx*(y n) exists for every x* E X*. The following classical result goes back to Orlicz. LEMMA

7.1. Let W be a subset of X such that every sequence of elements of W con-

tains a weak Cauchy subsequence. Then (7.1)

lim n

PROOF.

sup Ix~(x)1 xEW

=0

for every w.u.s. sequence (x~) in X*.

Define T: X -+ 11 by Tx

= (x~(x».

Then T is a bounded linear operator

which carries W into a totally bounded subset of 11 , because in 11 every weakly Cauchy sequence is norm convergent. Hence limN:E:=Nlx~(x)1 (7.1). 0

=0

uniformly for x E W. This yields

In view of formulas (1.1) and (1.2) of §1, in this and in the next section we identify L 1 /Hb with a subspace of A * and A with a subspace of (L 1 /Hb)*. The duality is given by the bilinear form (BE denotes the unit ball of a Banach space E)

(w, x)

-+

faDxwdm

Moreover, SUPXEBAJaDxwdm

for x EA and w

= Ilw Il L 1/H A and

=

{w

+ H~}

E

Ll/H~.

SUPWEBL1/HAJaDxwdm

= IIxII A .

Let Y be a closed linear subspace of a Banach space X. A map (in general nonlinear) r: X/Y -+ X is called the nearest point cross-section if r(w) E wand

IIwllx/y

=

inf lIy yEY

+ r(w)lI x = IIr(w)lIx

for w E X/Yo

The nearest point cross-section need not exist. It does in the case where Y is 43

Hb

and X is

ALEKSANDER PELCZYNSKI

44

either L I or X is qaD)*, because by the F. and M. Riesz theorem, H ~ is a closed subspace of qaD)* in the a(qaD)*, qaD»-topology. In fact in these cases it is unique (cf. [Kh 1] , [Hv2]). Now we are ready to state THEOREM 7.1. tet W be a subset of L I /H~. Then the' following conditions are equivalent:

(1) There is a weakly compact subset V of L I such that q(V)::::l W where q: LI

-->

L I /H~ is the quotient map. More precisely, for V one can take the weak closure of the image of W under the nearest point cross-section.

(2) The weak closure of W in L 1 /H~ is weakly compact. (3) Every sequence of elements of W contains a weak Cauchy subsequence.

o for every w.u.s.

(4) Iimn(sup{w+HA}EwIJaD
PROOF. Clearly for arbitrary Banach space (I)

~

(2)

=\>

(3)

sequence (

(4). (Use Lemma 7.1

and the "easier" implication of the Eberlein-Smulian theorem.) To prove that (4) ~ (1), we observe first that without loss of generality one may assume W is an infinite countable set,

(7.2)

+ H ~}

every {w (7.3)

E W regarded as a linear functional on

A attains its norm on the unit ball of A.

For (7.2) note that L 1/H ~ is separable and observe that if a subset of L I / H ~ satisfies (1) so does its closure. By the open mapping principle, if ({w~

1

For (7.3) note:

0

+ H~}) and ({w~ + H~}) + H~} -

are enumeration of countable sets W' and W" respectively and if II {w~

{w~

+

<

H~} IlL I/HI

the Bishop-Phelps

Tn for n

t~eorem

=

1, 2, . . . , then W' satisfies (I) iff W" does.

20

[B-Ph], the subset A ZP of A * consisting of the linear functionals

which attain their norms on the unit ball of A is dense in A *. Since A * is the II-sum of

L I /H~ and Vsing (cf. § 1), the natural projection of A * onto L 1 /H~ which annihilates Vsing maps A~p into itself. Hence A~p

+ H~})

Now let ({w n Yn E A so that (7.4)

faD WnYn dm

=

n L 1 /H~

is dense in L 1 /H~.

be an enumeration of W. It follows from (7.3) that there is a

inf

l hEHO

faD IWn

+ hi dm

and IIYnl1

=

1 for n

=

1,2, ... -

By the F. and M. Riesz theorem, H~ is a closed subspace of qaD)* in the a(qaD)*, qaD»-topology; hence there are h n E H~ such that

(7_5)

inf

1

f

aD

Iw n

+ hi dm

=

f

aD

Iw n

+ h n I dm

for n = 1, 2, . .. .

hEHO

Let Vn

= wn + h n

for n

=

1,2, ... and let V

=

U~=l {vn}' Our goal is to show

By

45

BANACH SPACES

(7.6)

the weak closure of V in L 1 is weakly compact.

Since q(V) = Wand T( {w n + H~}) = vn for n = 1,2, ... , this will complete the proof. We left to the reader a simple checking that if W satisfies (4) then it is norm bounded. This implies that sUPnllvnll = M < 00. To apply the well-known criterion of weak compact· ness in L 1 it remains to show that the vn's are uniformly integrable or equivalently that the un's are uniformly integrable, where un = Y nVn (or n = I, 2, . .. . It follows from (7.4) and (7.5) that

[

)aD

= )aD [ lu n I dm

un dm

for n

= 1,2, . . . .

Thus un ~ 0 for n = 1,2, . . . . Suppose that the functions un's are not uniformly integrable. Then using a standard gliding hump procedure (as in [K-P, Lemma 2]) we construct a subsequence (uic) of (un) and a sequence (e k ) of mutually disjoint closed subsets of 3D such that for some l) > 0 (7.7)

for k

(7.8)

the functions XaD\e

Since sUPnllunlll

= sUPnllvnlll <

00,

k

=

1, 2, ... ,

. Uic are uniformly integrable.

we can choose the sequence (uic) so that

there exists a tJ E C(3D)* which is the limit (7.9)

Let tJ

of the sequence (uic) in the a(C(3D) *, C(3D))topology, i.e., limkf aDxuic dm x E C(3D).

= U . m .+ v where U ~ L 1

= f aDx dtJ for

and v is singular with respect to m. By (7.7), U ~ 0, v ~ 0

and IItJlI ~ l) > O. We shall denote by (yic) and (vic) subsequences of (yn) and (v n ), respectively corresponding to the subsequence (uic) of (un); we have uic = Yicvic for k = 1,2, . . . . The crucial part of the proof is, exploiting the fact that tJ> 0, to construct in A a w.u.s. sequence (x,,) and a subsequence (Uk) so that the integrals fX"Uk s+ 1 dm = f(XsYk s+ 1 )vks+ 1 dm stay uniformly away from zero. This will contradict (4), because (x s ) w.U.S. yields (X sYk S +1) w.U.S. We shall consider two cases separately. The first one heavily depends on Havin's lemma (Proposition 6.1). In the second one we use only the fact that the annihilator of A is norm separable. Case I: v = O. Then a = f aDu dm = IltJlI > l). Let fk = fek and gk = gek for k = 1,2, ... ,where the functionsfek and gek are constructed in Havin's lemma (Proposition 6.1). Since ek are closed,fk and gk belong to A. Oearly (6.1)-(6.3) yield (7.10) (7.11)

limJ k

aD

hgkdm=f

aD

hdm

lim sup Ifk(z) k

zEek

forhELl,

11 = o.

46

-\LEKSANDER PH.CZyNSKI

Let us fix '7 > 0 so that 0 - a1/ > 0/2. Using (7.9)-(7.11) we construct an increasing sequence of the indices k j < k2 < ... so that if Go = 1 and Gs = nj== Igkj for s ;;;. 1, then for s = 1, 2, ... (7.12)

(7.13)

(7.14) Now we define a w.u.s. sequence (x s ) in A by

xs = Gsfk

for s = 1,2, ...

s+1

The sequence (x s ) is w.u.s. because, by (6.1), s s+ 1 Ix,iz) I ~ (lfkJz)1 + IgkJz)1) = 1

n

L

j== 1

j== 1

J

for z E

aD and for s = 1,2, . . . .

J

Combining (7.12) with (7.13) we get

IfaD GSUk

s+1

dm

I;;;. IJ-aD Gsu dm 1- 2-s- 2a1/ ; ;. f aD udm - t

(7.15)

j==1

s

;;;. a -

L

If

aD

Gjudm - (

JaD

T~-2a1/ - 2-s- 2a1/

> a(1

Gj_1udm

1-

2-S- 2a1/

- 1/).

j== 1

Clearly IIGslI"" ~ 1 (by (6.1», SUPzEeks+llfks+l(z) - 11- 0 as s sUPkllu';1I1 ~ sUPnllunli

=M < lim 8==""

00.

lim sup 8==00

f.eks + 1 11 - fk

IIGslu k

s+1 _

= faDudm = a, we

f 3D 11 - fk

8+ 1

(by 7.11), and

Thus

and (by (6.2» lim8==",,!aD\eks+lI1- fk8+1ldm

lims==""!oDuks+1 dm

00

IIGslu k

8+ 1

(7.16)

dm

dm

=0

Hence, by (7.8) and the relation

get

f aD\ek 11 - fk IIGslu ks+ s+ ~ lim sup f Uk dm 8==00 aD\ek s+ ~ a - lim inf f. uk dm s==oo e ks + 1 8+1

= lim sup 8==00

-0.

8+ 1

1

1

~a

Combining (7.15) with (7.16) we get

= 1.

s+1

8+ 1

1

dm

47

BANACH SPACES

lim inf I JaD ( xu' dm I 11=00 II k.l'+1

IJaD (G.I'u/c

?o lim inf .1'=00

dm I-lim sup (

.1'+1

JaD

.1'=00

11 - Ik

.1'+1

IIG.I'luic

.1'+1

dm ?o Q.2·

Finally we put

4fJ.I' Since

(x.l')

='Yk .1'+ I . x.l'

for s

=

1, 2, ....

is w.u.s. and IlYic II = 1 for k = 1, 2, ... , the sequence ( 4fJ.I') is w.u.s, Clearly we have

(

4fJ

laD .I'

Uk'

dm = (

laD

.1'+1

x

II

Uk .1'+1 dm

for s = 1,2, ....

Thus lim sup .1'=00

{

sup I}

I(

laD

w+Ho EW

4fJ.I'w dm I ?o lim inf I (xllu k

laD

.1'=00

11+1

dm I ?o Q.2·

This contradicts (4) and completes the proof in Case I.

Case II: v> O. Since the measure v is regular and singular with respect to m, there is a closed set e C aD with m(e) = 0 such that vee) ?o 711vll/8 = 7a/8. Using Bishop's general Rudin-Carleson theorem (cf. §2, Theorem 2.1) we construct inductively a sequence (fn)n;;.O in A and a sequence (e n )n;;'1 of closed subsets of aD such that

10

(7.17)

=0,

11

In(z)

= Il/nll =

en + I

= en n

{

e 2 = {zEaD: Iz-wl~TI for wEe},

e l = aD,

= 1,

"itn(z) I < Tn

1 for z E e, z E aD: sup Iz -

wi

wEe

for z

tt- en

~ Tn and j~ Ifj(z) - fj_1 (z)1 ..;; Tn f (n

= 2, 3, ... ).

Observe that

S(z)

=L

Ifj(z) - fj_1 (z)1 ~ 3

for z E aD.

;=1

Indeed, if z E e, then S(z)

= 1.

If z

tt- e,

then z E e k - 1 \e k for some k

= 2, 3, ...

2, then 00

S(z) ~ 2

L

00

Ifj(z)1 ~ 2

+2

j=1

if k

L

Tn ~ 3;

n=2

> 2, then 00

k-2

S(z)

=

L

If;(z) - fj_l(z)1

+ I/k_l(z)1 + 2

;=1

j=k 00

~ 2 -k+ 2

L

+ 1+2

L j=k

T;

< 3.

Ifj(z) I

; if k

=

48

-\LEKSANDER PELCZYNSKI

Clearly n;:lej = e. Hence using (7.9) and (7.17) we construct inductively increasing sequences of the indices (n s ) and (k s ), so that for s = 1,2, ... ,

If

aD

fn Uk. dm s

/

I~

2-sa,

for i

~ s,

r f.ns dIJ. I "'",.; : 2 -s-2 a. IJ'"aD f.nsu'kS+l dm - JaD Let us put no

= 0 and Xs = fns - fns+ 1 (s = 1,2, ... ). 00

L

IxsCz)I ~

s=1

L

Ifn/ z).- f nS_ 1 (z)1 ~ ~ Ifj(z) - fj-l (z)1 ~ 3.

s=1

/=1

Furthermore, for s = 1, 2, ... ,

I

Then (x s) is w.u.s. because, for z E aD,

00

faD X,U kS+ 1 dm

!~ IfaD f n, UkS+

1

dm

I-I

~ IlaD r f.n, dlJ.l- 2-9- 2a -

faD fns+ 1U,k s+ 1 dm

I

2-s- 1 a

~

IfaDfn,dvl-lfaDfn9udml-rS-2a-rs-la

~

If.e f.n, dv

~

I-I f

aD\e

vee) - v(aD\e) - T'a

f.n, dv

1-

2-s- 2a - 2-9- 2a - 2-,-l a

~ ~a,

Now we put I{J, = XsYk,+ 1 for s = 1, 2, . . . and we complete the proof as in Case I.

0

COROLLARY 7.1. The assertion of Theorem 7.1 remains valid if L 1 /H~ is rep/aced by A*, Ll by qaD)* and the condition (4) by

(4a) limn (sup {/J+H6}EW IfaDl{Jn dlJ.l)

= 0 for every w.u.s.

sequence (I{Jn) in A.

PRobF. To derive the implication (4a) ~ (1) from Theorem 7.1, denote by PI and P2

the natural projections from A* = qaD)*/H~ = Ll/H~ Ea1 Vsing onto LI/H~ and Vsing respectively, and let T: qaD)*/H~ ~ qaD)* and TL : L I /H~ ~ L I to be the nearest point cross-sections. Now if W C A * satisfies (4a), then both of the sets PI (W) and P2 (W) have the same property. For subsets of Ll/H~, (4a) is equivalent to (4). Hence the weak closure of T(Pl(W» = TL(Pt(W» is weakly compact. Furthermore, for subsets of Vsing ' (4a) combined with the general Rudin-Carleson theorem and the identity TP2 = P2 yields lim n(sup/JEP2(W)lfaDl{Jn dlJ.l) = 0 for every w.u.s. sequence (I{Jn) in C(aD). Thus, by a result of [P9], the weak closure of the set T(P2(W» = P2(W) is weakly compact. Since

49

BANACH SPACES

we conclude that the weak closure of r(W) is weakly compact.

0

7.2. A weak closure of a uniformly bounded set We L 1 /H~ is weakly compact iff there exists a function e ---+ K(e) for e > 0 such that for every {w + H~} E W there exists agE 11 with IlglI"" "" K(e) such that Ilq(w - g)II L 1 /HA < e(q: L 1 ---+ L 1 /H~ denotes the quotient map). COROLLARY

Hint. Use Theorem 7.1 and the uniform integrability criterion for weak compactness of subsets of L 1. 0

8. Weakly Compact Operators from A, L I /H~ and A* and Complemented Subspaces of These Spaces In this section we reap the benefits of Theorem 7.1. We establish further similarities between A and C(aD), and between L I and L I /H~. We show that A, A * and L I /H~ are weakly complete and have the Dunford-Pettis property. We also show that weakly compact operators from A into an arbitrary Banach space are characterized by their behavior on subspaces isomorphic to co' exactly in the same way as the weakly compact operators from a C(S)-space. In the second part of the section, we deal with complemented subspaces of A I and LI/H~. We show, in particular, that LI/H~ does not contain any complemented subspace isomorphic to L I. At the end of the section we discuss v~rious related open problems. Recall that a Banach space X is weakly complete if every weak Cauchy sequence in X converges weakly to some element of X. We say that X has the Dunford-Pettis property if every weakly compact operator from X into arbitrary Banach space takes weak Cauchy sequences into norm convergent sequences. The proofs of the Corollaries 8.1-8.5 below are, "modulo Theorem 7.1", easy modifications of analogous results for C(S)-spaces and L I-spaces. COROLLARY 8.1. (a) The spaces L I /H~ and A * are weakly complete. (b) The space L I /H~, A * and A have the Dunford-Pettis property. PROOF. (a) follows immediately from the equivalence of the conditions (2) and (3) of Theorem 7.1 (resp. Corollary 7.1). To prove (b) for L I /H ~, resp. for A *, observe that the equivalence of conditions (1) and (3) of Theorem 7.1 (resp. Corollary 7.1) implies that every weak Cauchy sequence in L I /H~ (resp. in C(aD)*) is the image under a quotient map of a weak Cauchy sequence in L I (resp. in C(aD)*). Now we use the fact that L I (resp.

[C(aD)) *) has the Dunford-Pettis property (cf. [D-SI, Chapter VI)). The assertion (b) for A follows from Grothendieck's observation that if X* has the Dunford-Pettis property then X does (cf. [Gr2), [P9]). 0 COROLLARY 8.2. If E is one of the spaces A, A*, LI /H~ and T: E -+ E a weakly compact operator, then T2 is compact. 0 COROLLARY 8.3. Let Y be a Banach space and let T: A -+ Y be a bounded linear operator. Then the following conditions are equivalent: (i) T is weakly compact. (ii) T takes weak Cauchy sequences into convergent sequences.

50

BA.NACH SPACES

51

(iii) T restricted to any isomorph of Co is not invertible. (iv) T takes every W.U.s. sequence in A into a sequence which converges to zero. PROOF. (i)

'* (ii).

Apply Corollary 8.1 (b) for A.

(ii) '* (iii). Obvious, because there are weak Cauchy sequences in Co which do not converge in norm. (iii) '* (iv). Note that a bounded linear operator takes W.U.s. sequences into w.u.s. ones and use the result of [B·P] which says that if (xn) is a w.u.s. sequence in a Banach space X which does not converge to zero in norm, then there is an isomorphic embedding from Co into X which takes the unit vectors of Co onto a subsequence of the sequence (xn)' (iv) '* (i). Observe that if T satisfies (iv), then W = T*(B A) satisfies the condition (4a) of Corollary 7.1. Hence, by this corollary, the weak closure of W in C(3D)* is weakly compact. Hence T* is weakly compact and therefore T is weakly compact (cf. [D-SI, Chapter

VI]).

0

COROLLARY 8.4. Let W be a bounded subset of A * (resp. L 1 /H~) whose weak closure is not weakly compact. Then there is a sequence (x~) of elements of W which is equiv-

alent to the unit vector basis of {1 and spans a complemented subspace of A * (resp. L 1 /H~). More precisely, there are operators T: {1 ---+ [C(3D)] * (resp. T: 11 ---+ L 1) and S: A * ---+ {1 (resp. S: L1 /H~ ---+ {1) such that SqT = id/ 1 and qTe n = x~ for n = 1,2, ... where q: [C(3D)] * ---+ A * (resp. q: L 1 ---+ L 1 /H~) denotes the restriction (the quotient) map and en denotes the nth unit vector of Z1 . PROOF. The assumption on W combined with Corollary 7.1 yields the existence of a W.U.s. sequence (IPs) in A and a sequence (y:) in W such that ly:(IPs) 1 ~ 1 for s

= 1,2, ....

Let M

~ max (sup IlY:II,' sup L s lIy·II=1 s

IY*('Ps)I).

< +00. Passing to a subsequence, if necessary, we may also assume that ~~=s+ 1 IY:(IPr) 1< 2- s- 3 for s = 1, 2, .. , . Next we define inductively a sequence of the

Clearly M

indices (k(n));=o and a sequence of infinite subsets of the integers (Nn);=o such that No - the set of all positive integers, ko = 0, (8.1)

(n

(8.2)

if j E N n , then j > k(n) and 1xt(IPk(n») 1 < 2- n-

= 1,2, ... ), 3

(n=1,2, ... ).

Suppose that for some q ~ 0 and for 0 <: r <: q the indices k(r) and the infinite sets N(r) have been constructed to satisfy (8.1) and (8.2). Fix an integer m > M2 q +4 and let Aq be a set of distinct m indices in N q which are> k(q). Then for each j

= 1, 2,

. .. there is an

index tV) E Aq such that Ixt(IPrU»)1 < 2 -q-4, because otherwise ~~= 1Ixt(IPt)1 ~ ~tEAq Ixt(IPr)1 > M which leads to a contradiction. Since N q is an infinite set and Aq is a finite one, we infer that there is an infinite subset Nq+ 1 of the set N q n {k(q) + 1, k(q) + 2, ... } and an index k(q + 1) E Aq such that if j E Nq+ 1> then Ixt(IPk(q + 1»)1 < q- 4 .

r

This completes the induction.

ALEKSANDER PH.CZyNSKI

52

Now put 1/I n =
n»

every finite sequence of scalars c l , c 2 ,

. •• ,

cm (m = 1,2, ... )

(1+ *)k~1 Icjl ~ Ilj~1 cjS(xf) II ~ (1 - *)j~ ICj I. Applying a standard stability argument (cf. [B-P)) we construct an isomorphism V: II ~ such that VS(x!) ,,; en for n = 1,2, . . . . Finally define T: [I ~ [C(aD)] * by T(e n ) = T(X~) for n = 1, 2, ... where T: A * ~ [C(aD)] * is the nearest point cross-section. The proof for L I /H~ is the same. 0 Next we shall discuss properties of complemented subspaces of A, LI /H~ and H*. Our first corollary generalizes results for C(S) and L I (v) spaces (cf. [PI] and [R2]); it shows in particular that A, L I /H~ and A * do not have complemented infinite dimensional reflexive subspaces.

II

COROLLARY 8.5. Let X be an infinite dimensional Banach space. (a) If X is complemented in A, then X contains an isomorph of co. (b) If X is complemented in A and X* is not separable, then X contains an isomorph ofC(aD). (c) If X is complemented either in A * or in L I /H~, then X contains an isomorph of

(d) If X is a separable dual and is complemented either in A * or in L I /H~ then weak and norm convergence of sequences in X coincide. (e) If X has a G-L l.u.st and X is complemented in LI /H~ then weak and norm convergence of sequences in X coincide. PROOF. (a) Apply Corollaries 8.l(b) and 8.2. (c) Combine Corollary 8.l(b) with Corollaries 8.2 and 8.4. (d) Use Corollary 8.I(b) combined with a simple observation that a complemented subspace of a Banach space with the Dunford-Pettis property has the Dunford-Pettis property, and apply a result of Grothendieck (cf. [Gr2] and [P4]) that if a separable dual space has the Dunford-Pettis property, then weak and norm convergence of sequences coincide in the space. PROOF OF (b). Let Y be a complementary subspace to X in A. Since A/Y is isomorphic to X, the hypothesis implies that (A/Y)* is not separable. Thus (C(aD)/Y)* is not separable and, by a result of Rosenthal [R2] , C(aD)/Y contains a subspace isomorphic to C(aD). By a result of [L-P2] (cf. also [R2]), this implies that A/Y contains a subspace isomorphic to C(aD) because, by the F. and M. Riesz theorem the space (C(aD)/y)/(A/Y) has a separable dual. This completes the proof because A/Y is isomorphic to X. REMARK. In fact (b) is true for every Banach space which is isomorphic to a subspace of a C(S)-space with separable annihilator, provided S is an uncountable compact metric space. PROOF OF (e). The argument is similar to that of Theorem 4.2(v). We consider the diagram

BA.NACH SPACES

53

where P: LI /H~ ~ X is a projection, q: LI ---+ LI /H~ is the quotient map, and i*:

L 1 /H~ is defined by i*( {[ + Ao})

= {[ + H~}

for [E C(aD). To define oo operators Sand T we first note that (i*)* = iM~ : H ---+ HI is an absolutely summing operator. Hence (Pi*)* is absolutely summing with values in the separable dual space HI. Since X has G-L l.u.st, Theorem 4.1 combined with Remark R.IIl in §4 yields that itp* is L I-factorable. Hence (Pi*)* = VU for some L 1 (v)-space and operators U: X* ---+ L 1 (v) and V: Ll(v) ---+ HI. We now define C(K) to be the dual of LI(v), S to be the restriction of V* to C(aD)/A o [identified with its canonical image in (C(aD)/A o )**] and we put T = QU*, where Q is a projection from X** onto X. Note that L I /H~ is complemented in the dual Banach space A * and therefore every complemented subspace of L I /H~ is complemented in the second dual of the subspace. Next we observe that Pi* = TS is a compact operator. Indeed note that T: C(S) ---+ X takes weakly Cauchy sequences into convergent sequences because C(S) has the DunfordPettis property (cL [D-SI, Chapter VI)) and, by a result of [PI], T is weakly compact. (The range of T, being contained in the weakly complete space L 1 /H~, does not contain subspaces isomorphic to co.) Furthermore every sequence in ~e unit ball BC(aD)/Ao has a weak Cauchy subsequence because (C(aD)/A o )* = HI is separable. Hence TS(BC(aD)/Ao) is a totally bounded subset of X. Now suppose that (x n ) is a sequence which is weakly convergent to zero in X. By Theorem 7.1, there is a sequence (fn) in LI which consists of uniformly bounded and uniC(aD)/A o ---+

formly integrable functions, and q(fn) = xn for n = 1, 2, . " . The properties of (fn) combined with the density in L 1 of all the continuous functions yield that for every € > 0 there is an M = M(€) < +00 and a sequence (gn) in C(aD) such that Iign lioo ~ M and Iign - [nlll < € for n = 1,2, . . . . Therefore the elements of the sequence (Pi*( {gn form a totally bounded subset of X because Pi* is compact and sUPnll{gn + Ao}ll ~ sUPnlignll ~ M. On the other hand we have, for n = 1,2, ... ,

11Pi*( {gn + Ao}) - xn II

= lIP( {gn + H~ }) -

+ Ao}))

Pq(fn)1I

Since € > 0 has been chosen arbitrarily, we infer that the set Un {x n} is totally bounded in X. Hence the weakly convergent sequence (xn) converges in norm. 0

ALEKSANDER PEl.CZyNSKI

54

Our next corollary which is an immediate consequence of Corollary 8.5 shows that L 1 does not have "the strong primary property" analogous to the property of C( [0, I]) discovered in [L-P2]. Note that H~, being separable dual, does not contain isomorphs of Ll. COROLLARY

8.6. The space L 1 /H~ does not contain a complemented isomorph 01 L 1.

Problem 8.1. Does there exist a subspace of L 1 /H~ which is isomorphic to L I? Notes and remarks to §§7 and 8. In 1967 Piranian, Shields and Wells [P-S-W] raised the question whether L 1 /H~ is weakly complete in the following form. Given a bounded sequence (gn) in L 1 such that the limit limnfaDlgn dm = (f) exists for every 1 E H~, does there exist agE L 1 such that (f) = f oDlg dm for 1 E H~? Kahane [Kh2] observed that there is a g in question such that <1>(1) = f oDlg dm for 1 EA. The weak completeness of L 1 /H~ and therefore of A * (our Corollary 8.1(a)) was proved by Mooney [Moo] and independently by Havin [Hvl]. Applying the result of [A-L] on peak sets in H~, Amar [A] adopted the method of [Kh2] to get an elegant proof of the Mooney-Havin result. Chaumat [Cbm] proved that L 1 /H~ and therefore A * and A have the Dunford-Pettis property. A \different proof of this result and the "liftability" of weakly compact sets in L 1 /H~ to weakly compact sets in L 1 was obtained by Cnop and Delbaen [C-O]. Chaumat's approach generalizes to the case of a predual of a uniform algebra in which an abstract analogue of the Amar-Lederer lemma (our Proposition 6.2) holds, while the Delbaen and Cnop method generalizes to uniform algebras satisfying some conditions imposed on sets of representing measures (cf. [C-O] and [02]). Theorem 7.1 in the present form and most of the corollaries in §8 have been obtained independently by Delbaen [01] and Kisliakov [IGs2]. Wojtaszczyk [W2] recently generalized Theorem 7.1 to certain planar uniform algebras. The proof of Theorem 7.1 presented in the text follows with some modifications the argument in [Kis2]. The construction in Case II goes back to Chaumat [Chm] ; our argument is different. Corollary 8.5(e) and Corollary 8.6 are due to W. B. Johnson and are published here with his permission. Several questions remain open. Very little is known about the nature of the dual of H~. Kisliakov in [IGs2] claims that the assertion of Theorem 7.1 is also true in the case ot H~. Unfortunately his argument contains a gap which he has recognized and communicated to me. Let us consider the following properties of a Banach space X: (a) the Dunford-Pettis property of X, (b) the Dunford-Pettis property of the dual of X, (c) the weak completeness of the dual of X, (d) a set We X* is weakly compact if it is weakly closed and sUPW*EW1w*(fn)1 = 0 for every w.u.s. sequence (in) in X, (e) every sequence in X* which converges in the a(X*, X)-topology converges weakly, (f) every complemented subspace of X is either finite dimensional or contains l~. Problem 8.2. Does H~ have the properties (a)-(f)? Problem 8.3. Let S be a compact metric space and let X be a subspace of C(S) with

BANACH SP-\CES

55

a norm separable annihilator. Does X have the properties (a)-(d)? How about the special case in which it is assumed in addition that X is a uniform algebra on S? Problem 8.4. Does the n-disc algebra A(Dn) (resp. the n-ball A(Bn)) have the properties (a)-(d) for n ~ 2? (For the definition of A(Dn) and A(Bn) see § 11.) Problem 8.S. What translation invariant subspaces of C(3Dn) have some of the properties (a)-(d)? Our next problem concerns complemented subspaces of A and L 1 /H~. Problem 8.6. (a) Let E be an infinite dimensional complemented subspace of A. Assume that E has an unconditional basis (resp. E has a G-L l.u.st). Is E isomorphic to Co (resp. is E isomorphic to a complemented subspace of C(3D))? (b) Let E be an infinite dimensional complemented subspace of L1/H~. Assume that E has one of the following properties: (1) E has an unconditional basis, (2) E has G-L l.u.st, (3) E is a separable dual, (4) E has the Radon-Nikodym property. Is E isomorphic to 11? Our last problem is closely related to the fact that every weakly compact set in L 1 /H~ can be lifted to a weakly compact set in L 1 (cf. Theorem 7.1) .. It is also related to Problem 3.1. Problem 8.7. Let E be a reflexive subspace of L 1 /H~. Does there exist a reflexive subspace E 1 of L 1 such that the quotient map q: L 1 ~ L 1 / H ~ restricted to E 1 is an isomorphism from E 1 onto E? How about the special case when E is isomorphic to a Hilbert space? ADDED IN PROOF. 10 Delbaen [De3] generalized Theorem 7.1 for uniform algebras such that every linear multiplicative functional has a weakly cqmpact set of representing measures and there is no nontrivial measure which is Singular with respect to all the representing measures. 2 0 Kisliakov announces in [Kis4] that if E is a reflexive subspace of A * then there exists a bounded linear operator T: A * ~ A * such that TIE is an isomorphism and there exists an E1 C [C(3D)] * such that the quotient map maps isomorphically E1 onto E. 3 0 The answers on Problems 8.6(b)(3) and 8.6(b)(4) are negative (Delbaen private communication).

9. Complementation of Finite Dimensional Subspaces in A, L 1/H ~ and I I The technique of Theorem 7.1 provides no information on the degree of complementation of finite dimensional subspaces of A (resp. L 1 /H~). For instance Corollary 8.5(a) obviously implies that infinite dimensional hilbertian subspaces are un complemented in A but it yields no estimate of the growth of the norm of the best projection onto n-dimensional Hilbert subspaces of A. Our approach to handle the above problem and similar questions concerning the "local" structure of A and L 1 /H~ is based upon Theorem 2.4. This theorem and the duality relation between A, L 1 /H~, and H"" are the only facts on spaces of analytic I functions which are used in this section, the rest is a good example of "purely formal" technique of the local theory of Banach spaces and the theory of absolutely summing operators and related Banach ideals. We shall show (cf. Theorems 9.1 and 9.2) that A, L 1 /H~, H"" do not contain nicely complemented l~'s for I < p < 00; A and H"" do not contain nicely complemented l~'s, equivalently L 1 /H~ does not contain uniformly (cf. Theorem 4.2(iv) for definition). We begin with introducing two concepts; the first one plays an important role in the local theory of Banach spaces, the second is a specific invariant suggested by Theorem 2.4. DEFINITION 9.1. Let X and Y be Banach spaces; a pair (S, T) of linear operators S: X --* Y, T: Y --* X with TS = id x is said to be a factorization of X through Y. The factorization constant of X through Y is the quantity

r;:

ry(X)

=

inf IISIlIITIl

1+00

if there are factorizations of X through Y, otherwise.

Here "inf" is extended over all factorization of X through Y. DEFINITION

9.2.

Let X be a Banach space and let I

< p < 00.

The ip -

7rP

ratio of

X is the quantity

where the supremum extends over all operators T from X into arbitrary Banach spaces with = 1. The concepts defined above are related between themselves and with projections. We have

7rp (T)

PROPOSITION

(a) kp(X)

9.1. Let

X. Yand Xl be Banach spaces and let I < p < 00. Then

< ry(X)kp(Y); 56

BANACH SP<\CES

(b) if X I is a subspace of Y and P: Y

~

57

X I a projection then

IIPII ;;. r y(X)h x I (X) provided min(ry(X), rXI(X)) < +00; (c) if X is finite dimensional and Xl is isomorphic to X then rX I (X) = d(X, Xl) where d(X, Xl) = inf {IISIlIiS-IIlIS: X ~ Xl isomorphism} is the Banach-Mazur distance between X and X I ; (d) if X is finite dimensional then ry(X)

= ry •• (X).

PROOF. (a) Let us fix a linear operator U: X with 1rp(U)

=

E (E an arbitrary Banach space) I and let (S, T) be a factorization of X through Y. Then ipCU) ;::: ip(UTS)

< IISlIip(UT) < IISlIkpCy)1rpCUT)

< liS II II TllkpC Y)1rp (U). Thus

which yields the desired conclusion. We left to the reader a routine proof of (b) and (c). (d) Clearly ry(X) ;;. ry .. (X). To prove the reverse inequality fix e

> 0 and pick a

factorization (S, T) of X through y** so that IISIlIiTIl < ry .. (XXl + e). Let F = SeX). Next note that for a finite dimensional X the space [B(Y, X)] ** can be identified with the space B(Y**, X). Moreover the canonical embedding of Y into y** induces the canonical embedding of B(Y, X) into S(Y**, X)

= (B(Y,

X))**. A simple restatement of the Gold-

stine theorem that canonical embedding of the unit ball of the space is weak-star dense in the unit ball of the second dual implies that there exists aTE B(Y, X) such that IITII

r,i=

<

(1 + e)IITil and T**Y** = Ty** for y** E F. Let T(y) = lyt(Y)X; for y E Y. Now we apply the Lindenstrauss-Rosenthallocal reflexivity principle [L-R] in the improved form due to Johnson-Rosenthal-Zippin [J-R-Z]. Hence there exists an isomorphic embedding U:

n

Y such that lIy**1I < II UY**11 < (1 + e)lIy**1I for y** E F and y**(y = y t(Uy**) for y** E F and for j = I, 2, . .. . Then (US, T) is a factorization of X through Y with

F-

II USIlIiTIl

< (l + e)2I1SIIIITil < ry •• (XXI + e)3.

Letting e -

0 we get ry(X) <

ry • • (X). 0 We shall also need the following result on absolutely summing operators. PROPOSITION 9.2 (B. MAUREY [Maul]). Let X be a finite dimensional Banach space,

let p with 1 < p < 00 be fixed, and let k ;;. I. Then the following conditions are equivalent: (i) ip(T) < k p1rp(T) for every bounded linear operator T from X into arbitrary Banach space. (ii) For every space LP.·(v) with p* = p/(P - I) and every subspace Y of LP·(v) every operator u: Y - X extends to a linear operator U: LP·(v) - X with IIUII < kpllViI.

ALEKSANDER PELCZYNSKI

58

PROOF. (i) ~ (ii). Recall that if X is a finite dimensional Banach space and Z a reflexive one, then the space N(X, Z) of all nuclear operators from X into Z is reflexive and its dual is isometrically isomorphic to the space B(Z, X) of all bounded operators from Z into X; the duality is given by the trace

(U, V)

~

for U E N(X, Z), V E B(Z, X).

tr(UV)

The condition (ii) says that the restriction operator B(LP * (v), X) ~ B(Y, X) takes the ball in B(LP * (v), X) of radius kp and with the center at the origin onto the unit ball of B(Y, X). Hence a standard duality argument yields that (ii) is equivalent to the following condition: (ii)* n(V):S;;; kpn(JV) for every operator

v:

X

~

Y and every closed linear subspace

Y of every LP* (v)-space. Here n( . ) denotes the nuclear norm of an operator and J: Y ~ LP * (v) denotes the inclusion map. We shall show that (i) ~ (ii)*. Pick V: X ~ Y with n(JV) = 1 and consider the best nuclear factorization diagram for JV, i.e., we write JV = Ti\S where T: X ~ [00 and S: [I ~ LP*(v) are operators of norm one and A: 1 ~ [I is the operator of ml.\ltipli00

cation by a sequence (A;) with A; ;;;;. 0 for j = 1, 2, . . . and 'i\ = 1. (The best nuclear factorization for JV does exist because X is finite dimensional and LP * (v) is reflexive.) Clearly A = Ap * Ap where Ap: [00 ~ [p and Ap *: [p ~ [I are the operators of multiplication by the sequences (A} Ip) and (A} Ip·) respectively. Let us consider the diagram

r

A p)

IP

Ap •

)

[I

U

S

Here Ep is the closure of ApS(X) in [p, and Wp

= ApS is regarded

as an operator from X

into Ep. Clearly TAp.(Ep) C Y. Let Wp' denote the restriction of TAp' to Ep regarded as an operator to Y. Obviously Wp is a p-absolutely summing operator with 1TpCWp) = ('i(A.I 1P f)l lp = 1. Hence, by (i), Wp is p-integral and ip(Wp) :s;;; k p1Tp(Wp ) = k p . By the Schwartz duality theorem (cL [Kw, Theorems 1 and 2]) TAp. is p*-absolutely summing because (TA p .)* is p*. integral. Thus Wp.' being restriction of a p*-absolutely summing operator, is p*-absolutely summing and 1T p *(W p .) = 1T p .(TAp ') = ip ' [(TA p ')*] = ('iA;)I IP * = 1. Now we use a result of Persson and Pietsch [poP] or a result of Persson [Per] to conclude that V = Wp' Wp is integral and therefore nuclear (because X is finite dimensional), and

n(V)

= i(V) :s;;; 1T p *(Wp .)ip(W p) :s;;; k p .

(ii) ~ (i). By the Persson·Pietsch duality theory or [P-P] , condition (i) is equivalent to the following

BANACH SPACES

59

(i)* npo(V) ~ kpnpo(jV) for every Banach space E and every linear operator V:E

-x. Here npo( . ) denotes the p*-nuclear norm of the operator and j: X ear isometric embedding. To show that (ii) '* (i)* fix e > 0 and an operator V: E consider the p*-nuclear factorization diagram for jV

["'" a fixed lin-

X with npo(jV)

= 1 and

Apo

E - - - -.....V.,-----+l X -----,;J;----+l t°

where Ap 0: 1

00

-

IP ° is the multiplication operator by the sequence

Ol} /po) with Aj ~ 0

and 1 ~ "i:.Aj ~ 1 + e, and Sand T are operators of norms one. Let Epo be the closure of ~oS(X) in IPo, Vpo = ApoS regarded as an operator from E into Epo, and Tpo the restriction of T to E pO regarded as an operator to X. By (ii), there is an operator T po: IP 0 - X which extends Tpo with IITpoll ~ kpllTpoll = k p . Thus V = TpoApoS and npo(V) ~ IISIlIITp.llnpo(Apo) ~kp(l + e)l/pO. Letting e - 0, we get np.(V) ~ kp = kpn(V). Now we are ready to prove

0

THEOREM 9.1. If Yis one of the spaces A, A *, L I /H~, H oo then

nl / 2 'Yy(l~) ~ C 10g(n + 1)

for n

=

1,2, ...

where C is an absolute const~nt. Consequently if Xl is an n-dimensional subspace of Y and P: tion, then IIPlI ~ C(n l / 2 /log(n + 1»· d(X I , I~)-l.

Y~

Xl is a projec-

PROOF. The second part of the theorem is an obvious consequence of the first one and Proposition 9.1(b) and (c). Next observe that by Proposition 9.1(a) and the standard duality relations between A, A *, L I /H~ and H"" (cf. §2) it is enough to consider the case Y = A only. Let CI • C2 , ••• denote positive constants independent of p and n. By Theorem 2.4, kp(A) ~ Cip for p ~ 2. Next we shall show (9.1)

kp(l~) ~ C2nl/2-I/P

for p

> 2 and for n = 2, 3, ...

Assuming (9.1) and combining it with Proposition 9.2(a), we get k (12) "V (12) ~ ~ ~ C n l /2- 1 / p • p-l for p ~ 2. 'A n P kp(A) 3 Specifying p = (Iog(n + 1))-1, we get 'YA (l~) ~ C4 n 1 / 2 /log(n + 1). To establish (9.1) it is enough, in view of Proposition 9.2, to show that there exists a

.\LEKSANDER PEl.CZyNSKI

60

constant Cs such that the condition (ii) of Proposition 9.1 is violated for X = l~ and for every constant kp < Csnl/2-1IP. By a recent result of Figiel, Undenstrauss and Milman [F-L-M] , there exists a constant C6 such that there exists an n-dimensional subspace Y of and an isomorphism U: Y ~ I~ such that IIUII = 1 and IIU- 1 11 0:;;;; 2 (note that p* = I) < 2). Let U: Ig: n ~ I~ be any extension of U. Then (U- 1 , U) is a factorization

z'f;* n

p/& -

of I~ through l~:n. Let I: Ib 6 n ~ Ig: n be the natural embedding. It is easy to see that 11111 = 1 and IIrlll 0:;;;; (C6 n)I IP = C7 n 11P . Clearly (I- 1U- 1, UI) is a factorization of I~.

It is well known that for every Ll(lI) space, rLl(V)(I~);;;;' Can 1/2 . (This is for instance an easy consequence of Grothendieck's result that every bounded linear operator in 12 which is L I-factorable is Hilbert-Schmidt (cf. [GRI] , [L-PI] ).) Thus

Cs n 1/2 Hence IIUII;;;;' C9nI/2-1IP.

IIr 1U- 111I1U· III

0:;;;;

0:;;;;

2n 11P IIUIi.

0

r A (l~) ;;;;. Cn 1/2? Observe that the positive answer to Problem 3.1 implies the positive answer to Problem 9.1. On the other hand Theorem 9.1 yields that operators on 12 which factor through A are "almost" of Hilbert-Schmidt type. We have Problem 9.1. Does there exist a C> 0 such that

COROLLARY 9.1. If an operator T: 12 ~ 12 factors through A, then T is compact

and if (Sj) is an enumeration in a non increasing sequence of nonzero eigenvalues of ITI, each eigenvalue repeated according to its multiplicity, then

L

s~

[log(j

~

1)) C

< +00

for every c> 3.

PROOF. Let (U, V) be a factorization of T through A. Since P is reflexive and A has the Dunford-Pettis property (cf. §8), T is compact. Next observe that the polar formula implies that an operator T factors through a Banach space iff its absolute value ITI factors through the same Banach space. Hence without loss of generality one may assume that T = ITI and that the sequence (Sj) is infinite (otherwise there is nothing to prove). Let (ej ) be the orthonormal sequence of eigenvectors of T corresponding to the sequence (Sj) of the eigenValues. Let us fix a positive integer n and denote by Pn the orthogonal projection from 12 onto the span (e j )l<;;;j<;;;n. Next define Sn: Pn{l2) ~Pn{l2) by Sn(e j ) = Sj-l ej for

= 1,2, ... ,n and let Un be the restriction of U to Pn(l2). Clearly SnPn VUn Hence Theorem 9.1 yields

j

IISnllllPnllllVlIlIUnll ;;;;. C log(n

+ 1)

for n

=

= idPn(l2).

1,2, ... ;

since IIPnll = 1 and IlUnll 0:;;;; IIVlI, for C1 = C/IIUIIIIVIl we have IISnll ;;;;. C1 n 1/2 /10g(n + I) for n = 1,2, . . . . Using the relation Sl ;;;;. S2 ;;;;. ... ;;;;. sn' we easily obtain IISnl1 = S;;l. Thus

sn

0:;;;;

C~110g(n

+ 1)/n I/2 .

Hence

..

L n=l for every c > 3.

0

s2 [log(n

n

+

..

I)]C

0:;;;;

L

C- 1 1 n=l

n[log(n

1

+ 1)]C

2

< +00

BANACH SPA..CES

61

Another consequence of Theorem 9.1 is COROLLAR Y 9.2. Let 1 < p < 00 and let p* = p/(P - I). Then there exist an absolute constant C > 0 and a number ex(p) with 0 < ex(p) <; \6 such that

I'A(l~)

= I'H",,(/~) = I'LI/H~{l~·);;' CnG:(p)

for n

= 1,2, ...

PROOF. Combine Theorem 9.1 with a result of Guran! Kadec and Macaev [G-K-M] that d(l~, I~) <; nII/2-I/pl. 0 Next observe that Davis and Johnson [D-J] showed that if X is an infinite dimensional Banach space which does not contain I ~'s uniformly, then there is an ex with 0 <; ex <; \6 such that l' x(l~) <; CnG: for some absolute constant C. Combining this fact with Theorem 9.1 one can prove COROLLARY 9.3. If (Xn) is an increasing sequence of finite dimensional subspaces of

a space X with the above property, then lim I'A (Xn ) = lim l' ",,(Xn ) n

n

H

= +00.

Our next result extends Corollary 9.2 on the case p THEOREM 9.2. There is an ex with 0 that, for n = 1, 2, ... ,

CnG: <; I'A (l~)

=

1.

< ex <; \6 and an absolute constant C> 0 such

= 'YH""(/~) = inf{d(/;;',

£)1 E C LI /H~; dim E

= n}.

PROOF. It is enough to prove the left-hand inequality. The equalities follow from Proposition 9.1 (d) and the duality relations between A, H"" and L 1 /H ~ established in § 1. (Note that H"" is norm one complemented in A **.) By a result of Maurey and Pisier [MauPis] , to prove the inequality it is enough to show that limn l' A (l~)

= 00; hence

Theorem 2.4 and Proposition 9.1(a), it suffices to show that, for some p +00. Assume to the contrary that sUPnkp(l~)

= M < 00.

in view of

> 2, limnkpC/~) =

(Note that the sequence

(kp(l ~)n= 1,2,... is increasing.) Then, Proposition 9.2 would imply that for every subspace Y of LP*, every operator U: Y --* I~ admits an extension V: LP· --* 11 with /lV/I <; MlIU/I. Since L 1 contains an increasing sequence of subspaces isometrically isomorphic to I~'s whose union is dense in L 1 and since L 1 is complemented in its second dual, a standard compactness argument (for instance involving Banach limits; cf. § 5, proof of Proposition 5.2) would yield that every bounded operator from yeLP· into L 1 extends to a bounded operator from LP· into L 1. Hence to complete the proof it is enough to establish the following LEMMA 9.1. For some p

> 2 (in fact for every p > 2) there exists a bounded linear

operator from a subspace of LP· into L 1 which has no extension to a bounded linear operator from LP * into L 1 . PROOF. Let Y be a noncomplemented subspace of LP* which is isomorphic to 12 (for instance for every p

> 4 one can take

as Y a closed linear subspace of LP * spanned by

the characters {zn: n E Rud} where Rud is a A4 -set constructed by Rudin, cf. [Ru2] and

ALEKSANDER PELCZYNSKI

62

[Rl] ; for 4 ;;. p > 2 the existence of Y in question follows from the recent result of five authors [B-D-G-J-N]). Let Eq (I ~ q

< 00) denote

the closed linear space in L q of the char-

acters {z2 n : n = 1,2, ... }. It is well known (cf. [Z, Vol. I, Chapter V, §6]) that for every 1 ~ q 1

< q2 <

00

the natural embedding iq2Q 1: L q2 -+ L q 1 maps isomorphically Eq 2

onto Eq 1 (in other words all sets of functions E q coincide with E 2 and the norms

II IIq 2 and

II IIq 1 are equivalent on E 2 ). This implies easily For every q ;;. 1 and

€

> 0 there exists aD> 0

such that if e is a measurable subset of aD with

(9.2)

m(e) < 0 and if f E E2 with IIfll2 felfIPdm<€for 1 ~p~q.

= 1 then

Since both Y and El are isomorphic to 12 , there exists an isomorphism U: Y ~ E 1 . We shall show that U is the desired operator. Suppose to the contrary that there exists a bounded linear operator

U:

LP *

-+

L 1 which extends U. Then, by a result of Maurey

[Mau2] , for every fixed r with p* > r > 1, there exists an operator V: LP * -+ L r and a multiplication operator Mg : Lr -+ L 1 defined, for some g E Lr*, by Mil)

= gf for f

E Lr

such that U = Mg V. Next fix q with r > q > 1 and using (9.2) pick 0 > 0 so small that if e is a measurable subset of aD with m(aD\e) < 0, then the operator Me: L S -+ L S of multiplication by the characteristic function Xe of e restricted to Es is an isomorphism for 1 ~ s ~ q*. Now we pick K > 0 so large that if e = {z E aD: Ig(z) I ~ K} then m(aD\e) < D. By a result of [K-P] , the orthogonal projection Q: L q * -+ MAEq *) is bounded as an operator on Lq*; hence Q being orthogonal is also bounded as an operator from Lq into Me(Eq). Let ge = g . Xe' Clearly the operator Mge of multiplication by ge is bounded as an operator from L r into Lq. Since the restrictions to Eq of Me and of iq,l are isomorphisms, there is an isomorphism W: Me(Eq) ~ El such that WMe(f) = iq,1 (I) for f E Eq . Clearly WMge V(LP*) eEl and P: U- 1 WMge V is a projection from LP· onto Y, a contradiction. 0 Before we pass to a discussion of possible strengthenings of Theorem 9.1 we would like to state a corollary of "global" character which is related to Problem 8.2. COROLLARY

9.4. Let X be a subspace oJ H oo satisfying one of the following proper-

ties: (a) X is separable and contains an isomorph of co. (b) X does not contain I ~ uniformly (in particular X is superrejlexive), dim X = 00.

(c) X contains I~ uniformly complemented. Then X is uncomplemented in H oo • PROOF.

(a) Use Sobczyk's theorem [S] and a result of [B-P] that isomorphs of Co are

uncomplemented in dual Banach spaces. (b) Use Theorem 9.1 and the result of [D-J] mentioned before Corollary 9.3. (c) follows from Theorem 9.2. REMARK.

Using slightly more carefully results of [B-D-G-J-N] and [B-G-N] one can

show that the exponents a in Theorem 9.2 can be made any number feel that the "real value" is a

=

< 1/8.

However, we

1/2.

Problem 9.2. (a) Does there exist a C> 0 such that 'YA(l~);;' Cn 1 / 2 for every n?

BANACH SPACES

63

(b) Does every operator from A into [I factor through [2? (c) Does L I /H~ have cotype 2, i.e., there is a C> 0 such thatJ~II~7=1 w;,/t)IIL I/HI dt ;;;. C(~f=lllwjllil/HA)I/2 for every wI' w 2 , . . . , wn in LI /H~ and for n = 1,2, ... ? 0 (Here (rj ) denotes the Rademacher functions.) It follows from a theorem of Maurey (cf. Tomczak Jaegermann [TJ]) that the positive answer on (c) implies the positive answer on (b); obviously the positive answer to (b) implies the positive answer on (a).

Recall that a scalar sequence (c n ) is an (A, [1) multiplier if there exists a (unique) bounded linear operator T(cnf A ---+ [1 such that T(zn) = cno n for n = 0, 1, ... where on denotes the nth unit vector of [1. It is worthwhile to mention that the answer on Prob-

lem 9 .2(b) is positive for (A, [1) multipliers. It is in fact a restatement of an old result of Paley [Pa2]. Precisely we have PROPOSITION 9.3 (PALEY). If (c n ) is an (A, [1)-mu[tiplier then ~lcnl2

< +00; hence

T(c n ) factors through [2.

PROOF. The second part of the proposition is an obvious consequence of the first. We define U: A ---+ [2 by U(f) = (f(n» for f E A and V: [2 ---+ [1 by V«d n = (dnc n )

»

for (d n ) E [2. Then clearly T = Vu. To prove the first assertion denote by tional on

[1

(n

= 0,

the coset {cnz- n

°

~

the nth unit vector of Co regarded as a func-

°

1, ... ) and note that the adjoint operator T*: ["" ---+ A * takes ~ into

+ Ll /H~}

for n

= 0,1,

... (remember that A*

= Ll /H~

ffi~ Vsing ).

Fix a positive integer N. By the Kolmogorov theorem (cf. §O.lI) there is a constant C> 0 such that for an arbitrary sequence (f) with

IIT*II ;;;.IIT*(.f 1=0

€j

= ± 1 (j =

€/jj) II = hEHl inf SaD o

;;;. C

(faD \jt

1, 2, ... , N) we have

I.i ClfZ- j

+ h(z)

1-0

\m (dz)

j

c/jz- \1/2 m(dZ») 2.

Averaging over all ± 1 sequences, i.e., integrating against the Rademacher functions and applying the Khinchine inequality for L 1/2 we get

C- 1 1IT*II;;;'

;;;.

m(dz) dt SI(rJan \j~N ClP)z-1.\1/2)2 0

(SIII ~ 0

aD

\N j~

= ( r J-liN L JaD 0 j=O ;;;'C1 / 2 (

I

aD

(

LN

j=O.

Icjl

Cllt)z-I.\1/2 m(dz}:1t)2

)2

Cllt)z-I.\1/2 m(dz) dt

2)1/2 m(dz))2 =C (.LN 1/2

1=0

Icjl 2)1/2

ALEKSANDER PEt.CZyNSKI

64

where C1 / 2 is the constant in the Khinchine inequality. Letting N -

<

00.

00

we get ~;:olc/

0

A positive answer on our last problem will provide a "local" analogue of Corollary 8. Problem 9.3. Does there exist for k = 1, 2, ... and for C ~ 1 an integer n = n(k. C) such that if X is a subspace of A with dim X = n which is a range of a projection from A of norm";; C, then X contains a k-dimensional subspace, say E with deE, 1';) < 2? Notes and remarks to §9. Most of the material of this section was obtained jointly by w. B. Johnson, G. Pisier and the author and is published here for the first time. Proposition 9.2 is due to Maurey [Maul]. The proof in the text is due to W. B. Johnson and G. Pisier and is published here with their permission. Proposition 9.3 is due to Paley [Pa2]. For other proofs see Helson [Hel] and Rudin [Ru3] _ The proof in the text is similar to Rudin's proof. ADDED IN PROOF. Every multiplier from L 1 /H~ into P is absolutely summing (cf. [Kw-P2]). This easily yields Proposition 7.3. It is unknown whether every bounded linear operator from L 1 /H~ into [2 is absolutely summing.

10. Bases and the Approximation Property in Some Spaces of Analytic Functions In this section we review results on the approximation property and bases in HP spaces (I ~ p ~ 00) and in the disc algebra. Most of the results are stated without proofs. We discuss several open problems. Recall that a Banach space X has the approximation property (shortly A.P.) if for every compact set K in X and € > 0 there exists a finite rank operator T: X -- X with IIT(x) - xII < € for x E K; moreover, if there exists a A ~ 1 (which depends on X only but not on K) such that the operator T in question can be chosen with II Til ~ A, then X is said to have the A-bounded approximation property (shortly the A-B.A.P.). X has the bounded approximation property (shortly the B.A.P.) if X has the A-B.A.P. for some A ~ 1. A sequence (x n ) of elements of X is called a basis for X if for every x E X there is a unique sequence of scalars (en) such that x = ~cnxn. The basis (x n ) is unconditional if the series ~cnxn converges unconditionally to X for every x EX. Each of the above properties is stronger than the preceding one, i.e., unconditional basis

=>

basis

=>

B.A.P.

=>

A.P.

The table below shows what is known about the approximation property and the existence of bases for "classical" spaces of analytic functions. A.P. I-B.A.P. basis unconditional basis

If'''

?

?

-

-

A

+ + +

+ + +

+ + +

+ ?

HP (l HI

< p < 00)

L oo/H";

?

?

-

CIAo = qaD)IA o

+ + +

+ + +

+ + +

LP /H~ (I LI/HA

< p < 00)

?

+ -

We pass to a discussion of the results contained in the table. The nonexistence of bases in H oo and L IH'; is trivial because these spaces are unseparable. The nonexistence of unconditional bases in A and L I /H~ follows for instance from the results of §4 because those spaces do not have any G-L I.u.st and the existence of an unconditional basis in a Banach space implies the existence of a G-L I.u.st. All the separable spaces in the table have I-B.A.P. To this end first recall that a separable Banach space X has the A-B.A.P. iff there is a sequence Tn: X -- X of finite rank operators such that II Tn (x) -xll-- 0 for every x E X and sUPnllTnll ~ A. For fELl we put 00

65

.\LEKSANDER PH.CZyNSKI

66

1

Tn (f)

= -n

n-l

"

(n

L..

= 1,2, ... ).

k=-n+l

It is easy to see that the Tn's take HP into HP for I ~ p ~ 00, A into A, and induce finite

rank operators on LP IHg (I ~ p ~ 00) and CIAo. The Tn's regarded as operators in each of the above spaces are of norm one. By the Fejer theorem (cf. [8, p. 23]) IITn(f) - flip -

0

for fE HP;

IlqTn(f) - q(f)II LP /Hg -

0

for f E LP;

Here q: LP -

LP IHg (resp. q: qaD) -

IITn(f) - flloo IIqTn(f) - q(f)IICfA 0

for fE A,

0

for fE qaD).

CIAo) denotes the quotient map. Thus

10.1. Each of the spaces 11', LP /Hg (l ~ p

PROPOSITION

-

0

< 00), A, CIAo

has I-B.A.P.

By a result of Boas [Bo] (cf. also [Kw-P] and §O.Il), the spaces HP and LP /Hg are isomorphic to LP for I < p < 00. By a result of Marcinkiewicz and Paley (cf., e.g., Burkholder and Gundi [B-G)), the Haar system is an unconditional basis for LP (l < p < 00). Thus

10.2. If I

PROPOSITION

< p < 00,

then the spaces HP and LP IHg have unconditional

bases. The situation in the "limit cases" of HI, A, L I IH~ and CIAo is much more delicate. The existence of bases in these cases is essentially due to Billard [Bi] and Bockarev [Bt]. We shall briefly describe their approach. Let (gj) be an orthonormal system in the real space L~ [0, 27T] with go == 1/..[I1r. We define a new orthonormal system consisting of 27T-periodic real even functions in (""11',7T) by

L1

g~(t) 1

= ~ gj(-2t) g,.(2t)

for t E [0, 7T] ,

(j

for t E [-7T, 0] ,

= 0,

I, ... )_

Next we put g# -

1

0-

+i

g,~ =

vfS'

_1_

V2

(g~ + iH(g~)) 1

/

(j

= 1,2, ... )

where H(g~)(t) ,

=

lim e-HO

1 - 7T

f.7r e

gj(t

+ s) - gj(t - s) 2 tg(sI2)

ds

is the Hilbert transform of gj. It is easy to see that (gt) is an orthonormal sequence in the complex space L 2 (-7T, 7T), in fact in H2. (We identify H2 with the subspace of L 2 (-7T, 7T) spanned by the characters (eint)n;;.o). Moreover, (gf) is an orthonormal basis for H2 if (gj) is a complete orthonormal

1.

system in L Starting with "nice" orthonormal systems in L ~ one can construct in that way orthonormal systems which form bases for HP (00 > P ~ l) and for A. In particular we have

BANACH SPACES PROPOSITION lO.3.(BILLARD c[Bi]).

67

1

Let (hn):=o be the Haar system in L [0, 21T] ,

i.e.,

~ ~ t < (2r + 1) ~,

(21T)-1/2,2 k / 2 for 2r

2k + 1

-(21T)-1/22 k / 2 for (2r

2k + 1

+ 1) ~ ~ t < 2(r + 1) ~, 2k +1

o

2k + 1

otherwise, (r

= 0,

1, ... , 2k - 1; k

= 0,

1, .. , ).

Then (h!):=o is a basis for HI . Let (fn) be the Franklin system, i.e., the orthonormal system in

L1 [0, 2'lT]

which is

obtained by the Gram-Schmidt orthogonalization of the sequence 1, J~ho(s) ds, J~hl (s) ds, where (hn):=o is the Haar system. Then we have PROPOSITION

10.4 (BOCKAREV [Bt]).

The system (f!) is a basis for A.

Here we identify A with a subspace of C [-1T, 1T] spanned by the characters (eint)n?>o. A few words on the proof of Proposition 10.4. Since each fn is a piecewise linear function, H(f~) is continuous and therefore ff: E A for n = 0, 1, . .. . Next recall that the Franklin system is a basis for CR 10, 21T] (for an elegant proof of this fact cf. [Cl]) and therefore the ~equence (j~) is a basis for the subspace of (10.1 ) CR [-1T, 1T] consisting of all the even functions. The crucial point of the proof is to establish the inequality (10.2)

IIHCto f~}f~ )II~ ~K(llfll~ + IIH(f)II~)
for every even fE CR [-1T, 1T] and for N

= 1,2,

...

where K is an absolute constant. The proof of (10.2) is difficult. It uses a delicate real variable technique involving an analysis of the behavior of the Hilbert transform of piecewise linear functions with nodes in the "first 2k

+ r"

dyadic points of the interval [-1T, 1T] and

some results of Ciesielski [C2]. Next it is convenient to introduce the real space AR

=

{(t, g) E CR [-1T, 1T] x CR [IT, 1T] : g

= H(f)}.

Using (10.1) and (10.2) we easily check that the sequence (f~, H(f~)) is a basis for the closed linear subspace of A R which it spans. This subspace consists of all the pairs (f, g) E

AR with f even. (The last fact is not obvious but not hard to check.) Thus for each fE A with Re(f) even there is a real sequence (c/f))j?> 0 such that

ALEKSANDER PEl,CZYNSKI

68

f=

{I 0.3)

L clf)ff

;==0

(the series converges uniformly on [-1T, 1T], i.e., in the norm topology of A). Hence if Ref is an odd function, then 00

f

=

L

-ic;(if)fr

;==0

because if Re(f) is odd then Im(f) is even. Finally using the identity f(t)

= ~ [(f(t) + f(-t)) + (f(t) -

fe-f»~]

we infer that for every f E A there is a sequence of complex numbers (Cj(f»r~o such that (10.3) holds. The sequence (cj(f)j<;.o is unique because the sequence (fr) is orthonormal. The proof of Proposition 10.3, being similar to that of Proposition 10.4, is even technically simpler. There are several consequences of Propositions 10.3 and 10.4. First we have COROLLARY 10.1. The spaces CIAo and L I IH~ have bases. PROOF. The space CIAo is a predual of HI (cf. § 1). Therefore the existence of a basis in CIAo follows from Proposition 10.3 and the result of Johnson, Rosenthal and Zippin [J-R-Z] that if X* has a basis then X does. A strengthening of this general result can be also used to show that L I /H~ has a basis. There exists however a less formal argument. The coefficient functionals of the Bockarev basis (fr)j;' 0 form a basis for the subspace of A * which they span. Since (fr)j;'o is an orthonormal system, the jth coefficient functionals can be identified with the coset {1l + H~} for j = 0,1, ... (under the natural identification of L I /H~ with a subspace of A *; cf. § 1). To complete the proof it is enough to show that the closed linear span E of the cosets ({1r + H~} )j~O coincides with LIIH~. To this end, fix n ~ 0 and let e int = r:;oc/fl(f) uniformly on [-1T, 1T]. Then e- int uniformly on [-1T,1T] and therefore also in LI-norm. Hence {e- int + H~} =

= r:;oCjl!(f)

r:;oc/ if! + H~} in the norm topology of L I IH~.

This shows that each coset {e- int + H~} belongs to E for n = 0, 1, .... Combining the last fact with the Fejer theorem we get the desired conclusion. 0 The space A(Dn) of all continuous functions on the n-disc analytic in its interior can be regarded as the n weak tensor product power of the disc algebra. By a result of Gelbaum and Gil de Lam ad rid [G-GL] , the weak tensor product of Banach spaces with bases has a basis. Hence CoROLLARY 102. The space A(Dn) has a basis n = 1,2, .... Very little is known on the existence of bases in the spaces A(U) for other bounded closed domains of holomorphy U in (cf. § 11 for the. definition). In particular we have Problem 10.1. Does the spaceA(Bn ) have a basis? Here Bn = {A = (Zj) E C": 1:IZj 12 :e;;; I}. Let us observe that if U is a bounded closed

en

BANACH SP!\CES

69

circled domain in en then A(U) has the I-B.A.P. For, define Tn: A(U) ~ A(U) to be the nth Fejer sum operator, i.e.,

Tn(f) =

f

n n+ ~ ; j [In]

for [EA(U) (n = 0,1, ... )

;=0

where [I n] denotes the homogeneous part of degree n of the Taylor expansion of [at zero. Next we turn to a discussion of the interrogation marks in the table. First we state Problem 10.2. (a) Does the space CIAo have an unconditional basis?

(b) Does HI have an unconditional basis? (c) Is H I isomorphic to a subspace of a space with an unconditional basis? (d) Does HI have any local unconditional structure? Obviously the positive answer to (a) implies the positive answer to (b). Furthermore, it has been observed in [Kw-P] and independently by Burkholder (unpublished) that the Billard basis (ht) for HI is not unconditional. We do not know whether the space HI is isomorphic to the space "martingale HI" - Mart HI where the last one is defined to be the space of all the functions [in L I whose Fourier-Haar expansion "'if=oc;(f)f; converges unconditionally in L I ; we admit sup 1e;1=1

If

;=0

E;C;(f)!j

III· L

Finally in connection with (b) and (d) it is interesting to mention that no example of a subspace of L I without any local unconditional structure is known. Next we pass to the Problem 10.3. Does H"" have the approximation property (resp. the bounded approxi-

mation property)? The same problem is open for L"" IH';. The spaces H"", L "" IH'; and the space of all the bounded linear operators on an infinite dimensional Hilbert space are the most common Banach spaces naturally appearing in analysis for which the approximation property is unknown. All those spaces are nonseparable. However Problem 10.3 can be reduced to some questions on separable function algebras. It is known (cf., e.g., [J]) that a Banach space X has A.P. (resp. A-B.A.P. for some A ~ 1) iff every separable subspace of X is contained in another separable subspace of X with A.P. (resp. with A-B.A.P. for the same A). On the other hand, by a recent result of Marshall [M] the Blaschke products are linearly dense in H"". Thus the positive answer on Problem 10.3 would follow from the positive answer on

the following Problem 10.4. Does every closed sub algebra of H"" generated by a countable set of

Blaschke products have A.P. (resp. A-B.A.P. for some absolute constant A)? We do not know the answer to Problem 10.4 even in the case of algebras generated by finitely many Blaschke products. Observe that, by a result of Milne [Mi] , there is a separable uniform algebra which fails to have the A.P. Another way of reducing Problem 10.3 to a question on the disc algebra involves the concept of the uniform approximation property introduced in [P-R]. Recall that a Banach

-\LEKSANDER PELCZYNSKI

70

space X has the uniform approximation property (V.A.P. for short) if there exists a A < 00 and a function k -+ N(k) (k = 1, 2, ... ) such that for every subset F C X of cardinality k there is an operator T: X

-+

X such that T(f)

=f

for

f

E

F, II Til

~

A, dim T(X) ~ N(k).

Obviously the V.A.P. with a constant A implies A-B.A.P. Furthermore, a complemented subspace of a Banach space with V.A.P. has the V.A.P. Lindenstrauss and Tzafriri [L- TI] proved that a Banach space X has the V.A.P. iff X** does. From the results of § 1 it follows that 11 is a complemented subspace of A **. Thus Problem 10.5. Does the disc algebra A have the uniform approximation property? ADDED IN PROOF. Every finite dimensional in L 1 /H~ of norm one is one dimensional;

there is no sequence in A and L 1 /H~ finite dimensional projections of norm one which tends strongly to the identity operator; in fact the range of every finite dimensional projection in A of norm one is isometrically isomorphic to

t;; (cf. [W3].

11. The Polydisc Algebra and the n-BaU Algebra, and Their Duals Let V be a bounded closed domain of holomorphy in the n-dimensional complex vector space en, let av denote the boundary of U. By A(U) we denote the Banach space of all continuous complex valued functions on V which are holomorphic in V\aV-the interior of

U; the norm in A(V) is defined by IIfll = suPAEulf(A)1 = sUPAEa ulf(A)1 for f E A(V). In this section we shall primarily deal with the following domains: the n-disc, D n = {A = (Zj) E en: maxI <;;j<;;n IZjl ~ I}; the n-ball, B n = {A = (Zj) E en: ~j=llzl ~ I}. Clearly DI = B I = D is the unit disc. The spaces A(D n ) and A(Bn) are called the n-disc algebra and the n-ball algebra respectively. They are the most natural analogue of the disc algebra among the spaces of analytic functions of several complex variables. Very little is known about the linear topological classification of the spaces A(V). Clearly if there is a biholomorphic map from the interior of a domain VI onto the interior of a domain U2 which extends to a homeomorphism from VI onto V 2 , then the spaces A(VI ) and A(V2 ) are isomorphic (= linearly homeomorphic). This condition is not necessary as the example shows: VI = D-the unit disc in e l ; V 2 = D U {z E e l : Iz - 31 ~ I} -the union of two disjoint discs. A reasonable conjecture says that the topological dimension of V, i.e., the number of variables, is a linear topological invariant of A(U). More precisely, if VI and V 2 are bounded closed domains of holomorphy in en and em respectively and if n =1= m, then the Banach spaces A(V I ) and A(V2 ) are not isomorphic. On the other hand a beautiful result due to Henkin [He2], [He3] shows that the spaces A(V) of the same number of variables may not be isomorphic; in particular, the n-disc algebra is not isomorphic as a Banach space to the nball algebra whenever n ;;;. 2 (cf. Theorem 11.1 below). A concrete open question related to the "dimension conjecture" mentioned above is the following Problem 11.1. Are the spaces A(Dn) and A(Dm ) (resp. A(Bn) and A(Bm)) not isomorphic whenever n =1= m? The next theorem summarizes what is known in the linear topological classification of the Banach spaces A(Dn) and A(Bm). THEOREM 11.1. (a) (Henkin [HeI]' [He2l). If n then A(Bn) is not isomorphic to A(Dm).

71

=

1,2, ... and m

= 2,

3, ...

ALEKSANDER PEI:.CZyNSKI

72

(b) (Mitjagin-Pelczynski [Mt-P]). The disc algebra A phic to A(Bn) for n ~ 2.

= A(Dl) = A(B l ) is not isomor-

The isomorphic invariants which enable us to obtain Theorem 11.1 are expressed in terms of the duals of the spaces A(Dn ) and A(Bm). and they have been introduced in § I. We recall those invariants. Let X be a separable Banach space. We say that X has Property I if it is isomorphic to a subspace of a C(S)-space whose annihilator in [C(S)] * is norm separable; X has Property II if X* is a separable distortion of L 1 (v), i.e., X* is isomorphic to the Cartesian product M ffi V where M is a separable Banach space and V is an L 1 (v)-space. Clearly the possessing by a Banach space X of Property I (resp. Property II) is an isomorphic invariant of X. Hence our Theorem 11.1 is an immediate consequence of the following. THEOREM 11.2. We have

A

= A(Dl) = A(B l )

Property I

Property II

Yes

Yes

A(Bn) (n

~

2)

No

Yes

A(Dn) (n

~

2)

No

No

Most of this section is devoted to the proof of Theorem 11.2. We begin with the "yes" entries of the table. The first row follows easily from the F. and M. Riesz theorem (cL § 1). The last "yes" in the second row is a consequence of Henkin's generalization of the F. and M. Riesz theorem which we present next. We begin with introducing the concept of an analytic measure (due to Henkin) which plays now an important role in the theory of analytic functions of several complex variables. Let U be a closed bounded domain of holomorphy in Cn . A sequence ~) in A(U) is a Montel sequence provided sup;lIfjll < 00 and lim;fj(A) = 0 for every A in the interior of U. DEFINITION 11.1. An analytic measure on aD is a complex Borel measure p. on au such that lim;foufjdp. = 0 for every Montel sequence (fj) in A(U). We shall denote by AM(U) the set of all the analytic measures on au. Clearly we have PROPOSITION 11.1. (i) AM(U) is a norm closed subspace of [C(au)] *. (ii) [A(U)] 1 C AM(U). (iii) If J.1. E AM(U) and f E A(U) then fp. E AM(U). (iv) If v is a representing measure for a point A E U\au, i.e., fa uf dv = f(A) for f E A(U), then v E AM(U). 0 The evaluation at a point AE U\au regarded as an element of [A(U)]

'Pr.

noted by Now we are ready to state Henkin's result.

* will be de-

BANACH SPACES

73

THEOREM 11.3. (a) If II E [C(3Bn)] * and II « iJ.Li with J.L E AM(3Bn ), then II E AM(3Bn )· (b) The dual [A(Bn)] * is isometrically isomorphic to the product AM(Bn)/[A(Bn)] 1 $1 V where V is the L I-space consisting of all measures in [C(3Bn)] * which are singular with respect to all positive analytic measures on 3Bn . (c) The space AM(Bn)/[A(B n)] 1 is separable and is isometrically isomorphic to the closed linear span of the IP~ for A E Bn \3Bn" PROOF. The difficult part of the Theorem is (a). Assuming that (a) has been established we deduce (b) and (c) as follows. Let V = {v E [C(3Bn)] *. v 1 iJ.Li for every J.L E AM(3B n)}. Then V is a band (= a complex sub lattice ) in [C(3B n)] *. Hence V is an L 1_ space. Furthermore by (a) and the Lebesgue decomposition theorem

Factoring through the annihilator [A(Bn)] 1 and taking into account that the dual of [A(Bn)] * is naturally isometrically isomorphic (via the restriction map) with the quotient

[C(3Bn)] */[A(Bn)] 1 and that [A(Bn)] 1 C AM(Bn ), we get [A(Bn)] *

= [C(3Bn )] */[A(bn )] 1 = (AM(Bn)/[A(B n)] 1) $1

This completes the proof of (b). To prove (c) we assume (b) and we shall identify the dual [A(Bn)]

V.

* with the

product

AM(Bn)/[A(Bn)] 1 $1 V and the quotient AM(Bn)/[A(Bn)] 1 with the subspace AM(Bn)/A(Bn)l $1 {OJ of that product. Let E be the norm closed subspace of [A(Bn)] * spanned by the IPt's for A E Bn \3Bn . Under the above identification each IPt corresponds to the coset {J.L + [A(Bn)]l} where J.L is any representing measure for the point A EBn\3Bn. Since J.L E AM(Bn ), IP: E AM(Bn)/[A(Bn)] 1. Hence E C AM(Bn)/[A(Bn )] 1. Now let x* E [A(Bn)] *\E. Then, by the Hahn-Banach theorem, there exists an x** in [A(Bn)] ** such that x**(x*) = 1 and x**(IP1) = 0 for AE Bn \3Bn . Thus, by Helly's theorem, for every sequence (Aj) in Bn \3Bn there is a sequence (fj) in A(Bn) such that IP,/fk ) = fk(Aj) = 0 for j = I, 2, ... , k; x*(fk) = 1; IIfkii :e;;; 2I1x**11 (k = I, 2, ... ). If we have chosen the sequence (Aj) to be dense in Bn \3Bn , the Montel theorem then yields that (fj) is a Montel sequence. Thus, if J.L is any extension of x* to a linear functional on [C(3Bn)] *, then J.L is not an analytic measure because lil!l I

f

aBn

fj dJ.L

= li~ I

x*(fj)

= 1.

This shows that E = AM(Bn)J [A (B n) ]1. lt remains to prove that E is separable. To this end fix r with 0

< r < 1 and let Qr :

C(rB n ) be the restriction map, where rBn = {A E Bn: r -I A E Bn}. By the Montel theorem, Qr is compact. Hence Q:: [C(rBn)] * - [A(Bn)] * is also.compact. Thus the set {IPt: A E rB n} is totally bounded being contained in the image under of the unit ball of [C(rBn)] *. Therefore, the set

A(Bn) -

Q:

I\LEKSANDER PELCZYNSKI

74

is separable and so is E. This completes the proof of (c). PROOF OF (a). It is enough to show if p. E AM(Bn ), then ~p. E AM(Bn)

(11.1 )

for k

= 1,2, ...

, n.

Indeed (11.1) together with Proposition 11.1(iii) yields Pp. E AM(Bn) for every polynomial P in the 2n variables zl' z2' ... ,zn' zl' z2' ... ,zn- By the Stone-Weierstrass theorem the polynomials are dense in L1(1p.1). Thus Ll(Ip.1) C AM(Bn), because AM(Bn) is a closed subspace of C{aB n)*. This proves (a). Now let m denote the Lebesgue measure on aB n' Before proving (11.1) we need the observation that the assertion of Theorem 11.3(a) holds for p. = m. To this end recall the Cauchy formula for Bn (11.2) where cn is a numeric factor independent of f. From (11.2) we derive

f(O)

= cn

ak1 +k2+"'+kn ~---k-f(O) aZ kl .•. aZn n

= kl!

l

(

JaBn

f(w)m(dw),

k2! ... k n ! cn (

JaB n

few)

n -k. n wi" .

m(dw)

,= 1

for arbitrary multi-index (k 1 • k 2 • ... ,kn ) and for fE A(Bn)' (Here w

= (wI'

w 2 ' ... ,

wn ).) Hence the Lebesgue measure m is analytic being a representing measure for the origin. By the Montel theorem, lim;
= 0 for every

Montel sequence

ni=

every multi-index (k 1 • k2' ... ,kn )· Therefore, the measures 1Zfi . m are analytic and the same application of the Stone-Weierstrass theorem as in the beginning of the proof of (a) yields

Uj) in A(Bn} and

if (11.3)

1/

is a complex Borel measure on aBn which is absolutely

continuous with respect to the Lebesgue measure m, then

EAM(Bn}' Next we shall show how to derive (11.1) from (11.2), (11.3) and the following 1/

MAIN LEMMA.

Let

Then Tf extends to a continuous function on C{Bn} for each f E A(Bn }, and moreover the operator T: A(Bn) - C(Bn) is compact. PROOF OF (11.1). Without loss of generality one may assume that k

Sf= zlf- Tf for ffiC.A(B n }.

= 1.

Let us set

75

BANACH SPACES

Then, by the first part of the Main Lemma, Sf E C(Bn). Furthermore, by (11.2), (SfX~)

r

wlf(w)

= en Ji

aBn [1 - (a, w)}n

m(dw)

for

a E Bn \aBn ·

Therefore, (Sf Xa) depends analytically on & in the interior of the ball Bn; hence Sf is holomorphic in Bn \aBn , equivalently Sf E A(Bn). Thus S: A(Bn) --+ A(Bn) is a bounded linear operator. Next observe that for every fixed a E Bn \aBn the measure vI E [C(aBn)] * defined by VI (A) = JA (wd[1 - (A, w)]n)m(dw) for measurable A c aBn is absolutely continuous with respect to m. Hence, by (11.3), vA E AM(Bn ). Therefore, for every Montel sequence

a E Bn \aBn

(fj) in A(Bn) and for every (11.4) and (11.5)

By the second part of the Main Lemma, (Tfj) is a sequence of equicontinuous functions (because sup;llfjll < 00) which, by (11.5), converges to zero on a dense subset of Bn. Hence

Tfj( a)

(11.6)

--+

as i --+ 00 uniformly for

0

= o.

Thus, for every J.I. E [C(aB n )]*, lim;JaBnTfjdJ.l. sure J.I.,

r

lim

; JaBn

(Sfj) dJ.l.

aE Bn·

Furthermore, for every analytic mea-

=0

because, by (11.4), if (fj) is a Montel sequence so is (Sfj). Therefore, for every Montel sequence (fj) in A(Bn) lim

;

f aBn fj(w)w

t

J.I.(dw)

= lim C· ; JaBn

Tfj dJ.l.

+ lim ;

JaBn Sfj dJ.l. = 0

which completes the proof of (11.1). PROOF OF THE

w

= (Zt - l )/[1- (A' w)]n for .4 = E aBn and A -=1= w; K(a, w) = 0 for a = wE

MAIN LEMMA. Let us set K(a, w)

= (WI' ...

(Zt' ... ,zn) E Bn and w

,

W n)

aBn. First observe that it is enough to show lim cx; = 0

where cx; = sup

;

IEBn

f

IK( A' w )Im(dw)

A;(I)

(11.7) and A/A)

= {w E

aBn: Re(A, w)

Assuming (11.7) we complete the proof as follows. Let tinuous function defined by 1/I(t)

h

~

t

< 1.

for 0 ~ t

< ~,

l/i}.

1/1: [0,00]

--+

= K(a' w)1/I(j{1

[0, 1] be a con-

1/I(t) ~ 1 for t ~ 1, 1/I(t) linear for

Let

K;Ca' w) and let

=0

~ 1-

- Re(A' w)))

for

i = 1,2, ...

76

ALEKSANDER PELCZYNSKI

Since the kernels K j are continuous on Bn x aBn • T/ E C(Bn) for f E C(Bn) and the Opt tors Tj are compact. We have also

(11.8) Hence, by (11.7), the sequence (Tjt) tends uniformly to Tf on Bn. Thus Tf is continuous for every fE C(Bn) and T is a bounded linear operator on C(Bn). Moreover, limjllTj - TIl = 0; hence T is compact.

The proof of (11.7) is easy for n

= 1 because

= 1 for A"* w.

then IK( A' w)1

If n ~ 2

we observe first that IK(A, w)1 ~ II~ - wll/ll - (A, w)l n and Aj(A) C Aj(o/IIAII) for 0"* 0 and j = l, 2, .... Clearly, by the rotation invariantness of the Lebesgue measure m on aB n' the integral JAj(AIIAII-I)ll o- willI! - (A, w)l n m(dw) is a function of 110 11 only. Therefore, we may assume without loss of generality that 0 = ae 1 where e 1 = (1, 0, 0, ... , 0) and 0
=0

where (3j

=

j=oo

Ilae 1 sup

(

O":a":l

11 -

JA j (el)

wll _ m(dw).

-

aWl

In

This will follow if we show (11.9)

wll _ ~ 2 n + 1/ 2 11 - w11(1-2n)/2 11 -aw 1 ln

lIae l

-


(0

1 and w E aNn)

and (11.10) To establish (11.9) we need elementary inequalities 1 - az \ 1a-Z\~I; - az 1 ~1~2

The second one yields (for z

(11.11)

(0


1, zED).

= WI) 2n -

1

1

----=---~---=---

11 - aWl In-l

11-wlln-l

Applying both inequalities we get

la - w112 + 1 - Iw I I2 11-aw112 11 - aw112

+ 1 - Iw112

4(1 - IW112) ~1+

(11.12)

11 - aw 1 12 5 - 2 Re WI

-

11 - wl12

31wl12

11 - w112 ~

8(1 - Re

WI)

II-w 1 12

~

-1

811 - wll

BANACH SPACES

77

Combining (11.11) with (11.12) we get (11.9). The proof of (11.10) is routine. After expressing the integral in terms of spherical coordinates (in the unit sphere in R2n) the problem reduces to the finiteness of the integral 1T

sin 2n - 2a sin 2n - 3 /3 da d/3

1T

So So . [(I - cosa)2 + sin2a cos2i3]

(2n-l)/4 '

whose integrand is ~ C/(a 2 + (13 - Tr/2)2)1/2 for some absolute constant C. 0 Our next goal is to show that if n ;;;;. 2 then the spaces A(Dn) and A(Bn) are not isomorphic to complemented subspaces of a C(S)-space with a separable annihilator. This will establish the "no" entries in the first row of the table. We shall need the following concept. DEFINITION 11.2. A subset F of a closed bounded domain U C Cn is called a support for the disc algebra if there exists a map
'PF(D)

(11.14)

gF

0

= F,


=Z

<1

for every wE U\(F

(11.15)

~F(w)1

(11.16)

I F (!) =fo
for every zED,

n

aU),

for every fEA(U).

Our next result provides a useful criterion for nonisomorphism of some A(U)-spaces to a disc algebra. THEOREM 11.4. Let U C Cn be a closed bounded domain of holomorphy. Suppose that

(*) there exists in U an uncountable family (F'Y)'YEr of supports of the disc algebra such that

Fa

n F~ n

au =

¢ whenever a =1= 13

(a, 13 E r).

Then the Banach space A(U) is not isomorphic to any subspace ofC(S)-space with a separable annihilator.

PROOF. Let us set I",I

= IF' Y ,g", = gF I 'Y

for 'Y E

r, where the operators IF 'Y and the = F'Y' Let PA : A ~ 12 be the

functions gF'Y are the IF and gF of Definition 11.2 for F Paley operator defined by

and let P'Y = PAI'Y for 'Y E r. Since the operator PA is absolutely summing (cf. §3, Example 3.1), so are P'Y and Tr1(P'Y) ~ Tr1(PA ) = C for every 'Y E r. Now assume to the contrary that there is a closed linear subspace X of a C(S)-space (S compact Hausdorff) such that the annihilator Xl eLl ()I.) for some Borel probability measure A on S, and there are bounded linear operators R: A(U) ~ X and Q: X ~ A(U) such

78

ALEKSANDER PE-lCZYNSKI

that QR = idA(u) is the identity on A(U). Then, by Corollary 2.1, for every 'Y E r, there exist a nonnegative function h-y and a compact (even nuclear) operator V-y: X _[2 such that (11.17) Next fix € with 0 < € < (Vi - l)/(IIRII tion b-y E L (X) so that

+ 1) and, for every

'Y E

r, pick a nonnegative

func-

00

(11.18) Let (m

= 1,2, ... ).

U;;;

Since =1 r m = r and since r is uncountable, at least one of the sets r m' say r m 0' is infmite. Now fix a positive integer M > 411Q IIm~ and distinct indices 'Y 1 ' 'Y 2' ... , 'YM in r mo. For simplicity we shall write in the sequel the index i instead of 'Yj; for instance Pj instead P-Yj' bj instead of b-Yj' etc. Let us set Yj,r =R«(gyr-l)

fori

= 1,2, ... ,M;r = 1,2, . . . .

Clearly IIYJ• rll ~ 1IRIIIIg~r-llloo ~ IIRII because by (11.13), (11.14) and (11.15), IIg~r-lll ,

J

/

=

IIgjll = 1. Thus, using the fact that the operators Vj are compact, we extract an infinite increasing subsequence (r(k)) of the indices such that IlVj(Xj,k)11

<€

for i

= 1,' 2, ...

,M; k

= 1, 2, ...

where Xj,k = Yj,r(2k-l) - Yj,r(2k)' Thus, by (1Ll7) and (11.18),

(11.19)

because IIxj,kll ~ IIYj,r(2k-1)11 + IIYj,r(2k)1I ~ 211RII (j = 1,2, ... ,M; k = 1,2, ... ). On the other hand, using the definition of PA and remembering that QR = id x , we get PjQ(Yj,r)

= Pj(glr-l) = PA ( z2r-l) = er

where er denotes the rth unit vector of [2 (r

(j

= 1,2, ...

,M)

= 1, 2, ... ).

Hence IIPjQ(xj,k)1I = Vi because PjQ(Xj,k) is a difference of two orthogonal vectors each of norm one. Thus (11.20)

Is

IXj,k1bjdA

~Vi- €(lIRIl + 1) ~ 1·

(j

= 1,2, ... ,M;k =

1,2, ... ).

79

BANACH SPACES

It follows from the assumption (*) that the sets FI n au, F2 n au, ... , FM n au are mutually disjoint. Combining this with (I 1.13), (I 1.14) and (11.15), we infer that there

exists an index ro such that, for r > ro' ~~llgr-l(w)1 ~ 2 for w E U. Thus there exists an index k such that for arbitrary complex numbers c I' c 2 , . . . , cM , we have

IIjtl CjXj,k

II

~ IIj~

cj Yj,r<2k-l)

II +

Ilj~ c

j Yj,r(2k)

II

qQII(t~ c;g(2HH I + t~ c;g(2kH II) ~

411QII

max

l';;;'j';;;'M

ICjl.

Hence ~~ 1 Ix/s) I ~ 411QII for every s E U. Therefore, M

L

;=1

Ix/s)lb/s) ~ 411Q11mo

because "I; E rmo (equivalently IIb;lIoo ~

which contradicts with the choice of M

for s E U

mol. Thus, by (11.14),

0

Now we are ready for COROLLARY 11.1. If n ;;;. 2 then each of the spaces (i) A(Bn), (ii) A(Dn) is not iso-

morphic to any complemented subspace of a C(S)-space with a separable annihilator. PROOF. It is enough to show that both domains Bn and D n (n ;;;. 2) satisfy the assumption of Theorem 11.4. (i) Let r

= (wI' w 2 , ... , wn ) E aBn: WI = Re WI > OJ. For w E r define = zw for Z ED; next defineg w EA(Bn) by gw(~) = (~, w) = = (zl' z2"'" zk) EBn, and put Fw = I{Jw(D)· Clearly each triple

=

{w

I{Jw: D -Bn by I{Jw(z) ~~=lZkWk for ~

(F w' I{Jw' g w) satisfies the conditions (11.13)-(11.16) of Definition 11.2, and Bn together with the family (Fw)wEr satisfies (*). (ii) Let r

=

{w

= (WI'

W 2 ' ... , W n - 1 )

E Cn-

1:

IWll

= IW21 = ... =

Iwn_11

=

I}.

For w E r define I{Jw: D - D n by I{Jw(z) = (wI' w 2 ' ... , w n- 1 ' z) for zED, and gw E A(Dn) by gw(~) = znn~:!l rl(Zk + w k ) for ~ = (Zl' Z2' ... , zm) E Dn, and put Fw = I{JjD). Clearly each triple (Fw, I{JW' gw) satisfies the conditions (11.13)-(11.16) of Definition 11.2, and D n together with the family (FW)WEr satisfies (*). 0 We turn to the last "no" entry in the table of Theorem 11.2. Again we shall prove more than is stated there; namely, we have THEOREM 11.5. If m ;;;. 2 then the dual [A(Dm)]

* is not isomorphic to any

subspace

<\LEKSANDER PELCZYNSKI

80

of the Cartesian product M EBI V where M is an arbitrary separable Banach space and V is an L I (v}-space.

This stronger result immediately yields COROLLARY 11.2. If n

=

1, 2, ... and m

= 2,

3, ... then A(Dm) is not isomorphic

to any quotient space of A(Bn).

Theorem 11.5 itself is an immediate consequence of the following three lemmas stated below. If E is a Banach space and r a set of indices then by 1~(E) we denote the Banach space of all functions (X'Y)'YEr: r ----+ E such that II(x'Y)'YEr ll = sUProcr~'YErollx'Yll where the supremum is extended over all finite subsets roof r.

<

00

LEMMA

11.1. The space L I /H~ is not isomorphic to any subspace of an L I (v}-space.

LEMMA

11.2. If m ;;;;. 2 then the space [A(Dm)]

isomorphic to I~(L I /H~) for some uncountable set LEMMA

* contains a subspace isometrically

r.

11.3. Let E be a Banach space. Assume that for some uncountable set

r

the

space I ~(E) embeds isomorphically into a Cartesian product M EBI V where M is a separable Banach space and V is an L 1 (v}-space. Then E is isomorphic to a subspace of some other L I (p. }-space.

PROOF OF LEMMA 11.1. Use Corollary 4.1, and the fact that the dual of L I /H~ is isometrically isomorphic to H oo (cf. § 1). 0 PROOF OF LEMMA 11.2. Let r = aD. To each w E aD we assign the subspace E w of [A(Dn)] * consisting of the functionals

(11.21)

x*(f) = J( aD

x*

for which there is a gx. ELI such that

f(z, w, 0,0, ... , O)gx.(z)m(dz)

for fEA(D n ).

It is clear that the map x* ----+ {H~ + gx.} ts, for every fixed w, an isometric isomorphism from Ew onto L I /H~. We shall show that the family of these isometric isomorphisms "extends" to an isometric isomorphism from the smallest closed linear subspace of [A(Dm)] * which contains all the subspaces E w ' for w E r = aD, onto I~(L I /H~). To this end it is clearly enough to show that if {WI' w2 ' ..• ,wM } is an arbitrary finite subset of r then

(I 1.22)

for every

Let gj = g x! ELI satisfy (I 1.21) with

x*

=

xt

xt

E Ew. (j /

= 1,2, ...

,M).

for j = 1, 2, . . . ,M. For j = 1, 2, . . . ,M,

we define an h j E A (D) so that hj(w k ) = 0 for k =1= j, hj(wj) = 1, Ihiw) I < 1 for w =1= Wj. The existence of the h/s with the above properties follows immediately from the RudinCarleson theorem (cf. Theorem 2.1). By the definition of the xl's, for f E A(D n ), for j = 1, 2, ... ,M, and for r = 1, 2, . .. we have (11.23)

xt(f) = xtUh'j);

xt(fh'k) = 0

whenever k =1= j;

IIfhjll

~

Ilfli.

81

BANACH SPA.CES

Here we identify hj with functions in A(B m ) whose value at a point (z\, Z2' ... , zn) is

[h j (Z2)r. Now fix E > 0 and pick fj E A(Dm) with Ilfjll = 1 so that xt(fj) - ElM for j = I, 2, ... , M. Next pick an integer r so large that

= Re xt(fj) ~ Ilx!11

(11.24)

Then, by (11.23), putting

x'(.~,

x*

=

hVk) ~

~f!= 1 xt, we have

J, x'(hUk) ~ k~' it. xWkh~) M

= L xt(fjhj) = j=

1

M

L j=

1

xt(fj) ~

M

L j=

1

Ilxtll - E,

while, by (11.23) and (11.24),

L~1 h~fk II ~ :~b 1;1 Ih~(z)lllfkll ~ 1 + E. Thus,

j~ Letting

IIx!11

E

~ IIx*1I ~ (1 + E)-1 Ix*C~1 fkh~) I ~ C~

tend to 0 we get (11.22).

Ilxtll- E)/(1 + E).

0

PROOF OF LEMMA 11.3. By [L-PI, Proposition 7.1] , it is enough to show that there is a C;:' 1 such that for every finite dimensional subspace F of E there is a subspace Fl in

V and an isomorphism Q: F -+ F 1 with IIQIIIIQ- 111 ~ C Let qM and q v denote the natural projection of M 611 V onto M and V respectively. Let, for a E r,

Clearly we may identify each EOI with E via the map (x')') -+ x Oi ' Let T be an isomorphic embedding of 1~(E) into M 611 V, let T- 1 : Tl~(E) -+ 1~(E) denote the inverse of T and let TOI denote the restriction of T to E OI . Let us fix a finite dimensional subspace F of E and consider the space B(F, M) of all bounded linear operators from F into M. Since M is separable and F finite dimensional, B(F, M) is separable. Thus there is a pair of distinct indices 'Yl and h such that IlqMT')'1 - qMT')'211 < (211 T- 111)-1 because the set {qMT')': 'Y E r} after the identification of each E')' with E is an uncountable subset of B(F, M). Next define W: F -+ 1~(E) by W(F)

for'Y

*

let Q

= q v TW:

'Y 1 and'Y

*

= (X~)')'Er

where

X?1 = -X?2 = Yzf and x~ = 0

'Y2 (fE F). Clearly W is an isometric isomorphism embedding. Now,

F -+ V. Then for every

f

E

F we have

82

ALEKSANDER PEl:.CZyNSKI

IIQfII = IITWfll-lIqMTWfli

while II Qfll .;;; IIq villi Til IIWIIIIfli = II Tllllfli. Hence Q: F ---+ V is an isomorphic embedding with II Q 1111 Q-III .;;; 211 Til II T-11I where Q-l: Q(f) ---+ F denotes the inverse of Q. This completes the proof.

0

Notes and remarks to § 11. Theorem 11.3 is due to Henkin [He2]. The proof presented in the text is a simplified version of the proof due to Cole and Range [C-R]. The proof of the Main Lemma is due to T. Figiel. Similar operators and integrals have been considered in [Ru4]. Theorem 11.3 is a special case of a more general result for strictly pseudoconvex domains. Recall that a bounded closed domain U in Cn is strictly pseudoconvex with C 2 _ smooth boundary if there is a twice continuously differentiable real function p defined on an open neighborhood G of U such that U = {A E G: p( A) .;;; O}, au = {A E G: peA) = O}, grad p 0 at each point A E au, and, for every AE au, n a2 p(A) _ WjWk > 0

*

L

j.k=l

whenever 0

*

(WI'

w2 ,

... ,

azjazk

w n ) E Cn satisfies the condition

-£ j= I

apO)

Regarding p as a function of 2n real variables admit

:: ~ +(:~ -, ;;)

Wj

= O.

aZj XI' YI' . . . ,

x n • Y n where

:~ ~ ~'(::, +, :;')

Zj

= Xj + iYj we

(j=1,2, ... ,n).

We have

l1.3a (CF. HENKIN [He3]). The assertions (a), (b), (c) of Theorem 11.3 remain valid if Bn is replaced by arbitrary bounded closed strictly pseudoconvex domain with C 2 -smooth boundary. THEOREM

Theorem 11.3a can be proved along the same line as Theorem 11.3. The main technical difficulty is a construction of an analogue of the Cauchy formula (11.2), i.e., an integral representation

83

BANACH SPA.CES

f(~)

fa

=

r/(w)K(w, a)a(dw)

(~

E u\au,[ EA(w))

where a is the (2n - I)-dimensional surface Lebesgue measure and K: au x (U\au)

---+

Cis

a kernel which is analytic in the first variable, and such that for the operator induced by this kernel the analogue of the Main Lemma is true (at this point the "geometry" of the boundary of the strictly pseudoconvex domain is exploited). The desired integral formula has been first constructed by Henkin [He3] and by Ramirez de Arellano [Rrn] for strictly pseudoconvex domains with C 3 -smooth boundaries, and next, due to improvement by qlverlid [~], for strictly pseudo convex domains with C 2 -smooth boundaries. The reader is referred to the excellent survey [C-He] for further information concerning the integral representations and analytic measures. Also other results of this section can be extended to the case of domains of holomorphy with C 2 -smooth boundaries. In particular Corollary 11.1 to Theorem 1104 admits the following generalization (cL Mitjagin-Pelczynski [Mt-P, Theorem I]). 11.6. If n ~ 2 and if U

THEOREM

C

Cn is a bounded closed domain of holomorphy

with the C 2 -smooth boundary, then the Banach space A(U) is not complemented in any subspace of a C(S)-space with a separable annihilator. Before stating a generalization of Theorem 11.5 let us observe that the Shilov boundary of the uniform algebra A(D m ) is the m-torus (aDr

={~ =(Zj) E em: Iz I =IZ21 =... =IZm I =I}, J

while the topological boundary of D m is aD m

={~= (Zj)

E Cm:

max IZjl l";;',";;'m

=I}.

Hence, if m ~ 2, then the Shilov boundary of A(Dm) is a proper subset of the topological boundary of D m . In general we have THEOREM

11.7. Let U

C

Cm (m ~ 2) be a bounded closed domain of holomorphy

C 2 -smooth

with boundary. Assume that the Shilov boundary of the algebra A(U) is a proper subset of the topological boundary of U. Then [A(U)] * is not a separable distortion of an L 1 (v)-space. Hence A(U) is not isomorphic to A(U1 ) for any strictly pseudo convex domain U 1 in Cn (n = 1,2, ... ) with C 2 -smooth boundary. Theorem 11.7 is also due to Henkin (unpublished); the second part of Theorem 11.7 has been announced in a slightly weaker form in [He3] (cf. Theorem 1.6 of [He3]). Theorem 11.5 and Corollary 11.2 are essentially due to Henkin [He 1]. The proof presented in the text seems to be new. Let us outline briefly another proof that A (D) is not isomorphic to A(Dn) for n ~ 2. Let mn denote the normalized Haar measure on [aD] n and let f1P([aD] n) denote the closure of A(D n ) in If([aD]n), i.e., the completion of A(Dn) in the norm U[aD]lIIfI Pdm n )l/P. Let i~n): A(Dn) ---+ HP( [aD] n) be the natural embedding. Clearly i~n) is a p-absolutely

84

ALEKSANDER PRCZYNSKI

summing operator with 'ITpu~n») = 1. Next, for p > 1, we evaluate the p-integral norm of i~n) using the in variance of i~n) with respect to action of the group (aDt on A(Dn) and

HP([aD] n) and the averaging technique exactly in the same way as in the proof of formula (2.17) of §2. We get ipu~n») = IIR(n)lI p where R(n) denotes the orthogonal projection from L2([aD]n) onto H2([aD]n) regarded as an operator from LP([aD]n) onto HP([aD]n). On the other hand a straightforward argument gives IIR(n)lI p = IIRII; where R is the Riesz projection. Thus ip(i~n»);;;. C 1 [p2/(p - 1)]n where C 1 is an absolute constant (cf. §O.II). The above arguement combined with Theorem 2.4 shows that for n ;;;. 2, the Up - 'lTp)-ratio of A (D n) (cf. Definition 9.2) has a different behavior at 00 than the Up - 'ITP )-ratio of A (D n). In fact we have

lim kpCA(Dn))/k (A (D))

p=oc

p

=

00

which, in view of Proposition 9.l(a), yields that the Banach space A(D) is not isomorphic to

A(D n ) for n ;;;. 2.

0

REMARK. (1) A similar technique enables us to show HOC (aD) is not isomorphic to H""([aD]n) for n ;;;. 2. (2) However this approach does not seem to work to distinguish

between A(Bn) and A(D). The reason is that the norm of the analogue of the Riesz projection-the rotation invariant projection from LP(aB n ) onto HP(aBn)-is of order Cp2/(p - 1). This can be deduced from the proof due to Koranyi and Vagi [K-V] of the boundedness of this projection [B. S. Mitjagin, private communication]. (3) We do not know whether for n ;;;. 2 and for 1 < p *- 2 < 00 the Up - 'lTp)-ratio of A(Dn) (resp. A(Bn)) is finite. If it is finite, we shall be able at least for the polydisc algebras to use this fact to solve Problem 11.1. Thus Problem 11.2. Let 1 < p < 00. (a) Is every p-absolutely summing operator from n A(D ) (resp. A(Bn)) p-integral? (b) Is it true that for n = 2, 3, ... , there exists an absolute constant Cn such that, for every finite rank operator T: A(D n ) --')- E (E arbitrary Banach space), ip(T) ~ Cn(p2/(p - l)t'IT peT)? (c) Is it true that there is an absolute constant C such that for every finite rank operator T: A(Bn) ~ E (E arbitrary Banach space), ipCT) ~ Cp2/(p - I)?

REFERENCES

[A]

E. Arnar, Sur un theoreme de Mooney relatif awe fonctions analytiques bornees, Pacific J. Math. 49 (1973), 311-314. MR 49 #9601. [A-L] E. Amar and Aline Lederer,Points exposes de la boule unite de H~(D), C. R. Acad. Sci. Paris Ser. A-B 272 (1971), AI449-AI452. MR 44 #788. [Bl] E. Bishop, A general Rudin-Carleson theorem, Proc. Arner. Math. Soc. 13 (1962), 140-143. MR 24 #A3293. [B2] , A generalization of the Stone Weierstrass theorem, Pacific J. Math. 11 (1961),777-783. MR 24 #A3502. [B-D-G-J-N] G. Bennett, L. E. Dor, V. Goodman, W. B. Johnson and C. M. Newman, On uncomplemented subspace~ of LP (1 < p < 2), Israel J. Math. (to appear). [B-G] D. L. Burkholder and R. F. Gundy, Extrapolation and interpolation of quasi linear operators on martingales, Acta Math. 124 (1970), 249-304. [B-G-N] G. Bennett, V. Goodman and C. M. Newman, Norms of random matrices, Pacific J. Math. 59 (1975), 359-365. [Bi] P. Billard, Bases dans H et bases de so us espaces de dimension finie dans A, Proc. Conf.,Oberwolfach (August 14-22, 1971), ISNM Vol. 20, Birkhauser, Basel and Stuttgart, 1972. [Bo] R. P. Boas, Jr., Isomorphism between HP and LP, Arner. J. Math. 77 (1955), 655-656. MR 17, 1080. [B-P] C. Bessaga and A. Peiczyriski, On bases and unconditional convergence of series in Banach spaces, Studia Math. 17 (1958), 151-164. MR 22 #5872. [B-Ph] E. Bishop and R. R. Phelps, A proof that every Banach space is subreflexive, Bull. Arner. Math. Soc. 67 (1961),97-98. MR 23 #A503. [Bt] S. V. Bockarev, Existence of a basis in the space of functions in the disk, and some properties of the Franklin system, Mat. Sb. (N.S.) 95 (137) (1974), 3-18 = Math. USSR Sbornik 24 (1974), 1-16. [Cl] Z. Ciesielski, Properties of the orthonormal Franklin system, Studia Math. 23 (1963), 141-157. MR 28 #419. [C2] , Properties of the orthonormal Franklin system. II, Studia Math. 27 (1966),289-323. MR 34 #3202. [C-D] I. Cnop and F. Delbaen, A Dunford-Pettis theorem for Ll /H~1, J. Functional Analysis 24 (1967).

85

86

ALEKSANDER PEl.CZyNSKI

[C-H] E. M. Cirka and G. M. Henkin, Boundary-values properties of holomorphic functions of several complex variables, Sovremennye Problemy Mat. 4 (1975),13-142. (Russian) [Chm] J. Chaumat, Une generalisation d'un theoreme de Dunford-Pettis, Univ. Paris XI, U.E.R. Mathematique no. 85, 1974 (preprint). [Chv] L. Chevalier, Algebres de fonctions orthogonal purement atomique, Studia Math. 40 (1971), 81-84. MR 46 #6036. [CoR] B. Cole and R. M. Range, A-measures on complex manifolds and some applications, J. Functional Analysis 11 (1972), 393-400. MR 49 #5398. [CoW] R. R. Coifman and G. Weiss, HP spaces and harmonic analysis, Bull. Amer. Math. Soc. (to appear). [Del] F. Delbaen, Weakly compact operators on the disc algebra, 1. Algebra (to appear).

a

[De2] - - - , The Dunford-Pettis property for certain uniform algebras, Pacific J. Math. 65 (1976), 29-33. [D-J] W. J. Davis and W. B. Johnson, Compact non-nuclear operators, Studia Math. 51 (1974),81-85. MR 50 #5514. [D-P-R] E. Dubinsky, A. Petczynski and H. P. Rosenthal, On Banach spaces X for which n2 (L oa , x) = B(L oa , x), Studia Math. 44 (1972), 617-648. MR 51 #1350. [D-SI] N. Dunford and J. T. Schwartz, Linear Operators. Vol. I, Interscience, New York, 1958. MR 22 #8302. [Du] P. L. Duren, Theory of HP spaces, Academic Press, New York and London, 1970. MR 42 #3552. [D-U] J. Diestel and J. J. Uhl, Jr., The Radon-Nikodym theorem for Banach space valued measures, Rocky Mountain J. Math. 6 (1976), 1-46. [Etch] A. Etcheberry, Some uncomplemented uniform algebras, Proc. Arner. Math. Soc. 43 (1974), 323-325. MR 49 #1123. [Fi] S. D. Fisher, Exposed points in spaces of bounded analytic functions, Duke Math. J. 36 (1969), 479-484. MR 41 #4247. [F-J-T] T. Figiel, W. B. Johnson and L. Tzafriri, On Banach lattices and spaces having local unconditional structures with applications to Lorentz function spaces, J. Approximation Theory 13 (1975), 395-412. MR 51 #3866. [F-L-M] T. Figiel, J. Lindenstrauss and V. Milman, The dimension of almost spherical sections of convex bodies (to appear). - - - , The dimension of almost spherical sections of convex bodies, Bull. Arner. Math. Soc. 83 (1976), 575-578. [Fourl] J. Fournier, Fourier coefficients after gaps, J. Math. Anal. Appl. 42 (1973), 255-270. MR 48 #795. [Four2] - - - , An interpolation problem for coefficients of Hoa functions, Proc. Arner. Math. Soc. 42 (1974), 402-408. MR 48 #8806. [F-S] C. Fefferman and E. M. Stein, HP spaces of several variables, Acta Math. 129 (1972), 137-193.

BANACH SPACES

87

[G-GL] B. Gelbaum and J. Gil de Lamadrid, Bases of tensor products of Banach spaces, Pacific J. Math. 11 (1961), 1281-1286. MR 26 #5394. [G-J] D. P. Giesy and R. C. James, Uniformly non-II and B-convex Banach spaces, Studia Math. 48 (1973), 61-69. MR 48 #11994. [G-K-M] V. I. Gurarii, M. I. Kadec and V. I. Macaev, On distances between finite-dimensional analogues of Lp-spaces, Mat. Sb. 70 (112) (1966), 481-489; English trans!., Amer. Math. Soc. Trans!. (2) 76 (1968),207-216. MR 33 #4649. [G-L] Y. Gordon and D. R. Lewis, Absolutely summing operators and local unconditional structures, Acta Math. 133 (1974),27-48. [GIl] I. Glicksberg, Measures orthogonal to algebras and sets of antisymmetry, Trans. Amer. Math. Soc. 105 (1962),415-435. MR 30 #4164. [GI2] , Some uncomplemented function algebras, Trans. Amer. Math. Soc. 111 (1964),121-137. MR 28 #4383. [Gml] T. W. Gamelin, Uniform algebras, Prentice-Hall, Englewood Cliffs, N. J., 1969. [Gm2] , Restriction of subspaces of C(X), Trans. Amer. Math. Soc. 112 (1964), 278-286. MR 28 #5331. [Grt] A. Grothendieck, Resume de la theorie metrique des produits tensoriels topologiques, Bo!. Soc. Mat. Sao Paulo 8 (1953),1-79. MR 20 #1194. [Gr2] , Sur les applications lineaires faiblement compactes d'espaces du type C(K), Canad. J. Math. 5 (1953),129-173. MR 15,438. [Gr3] , Produits tensoriels topologiques et espaces nuc/eaires, Mem. Amer. Math. Soc. No. 16, 1955. MR 17, 763. [H] K. Hoffman, Banach spaces of analytic functions, Prentice-Hall, Englewood Cliffs, N. J., 1962. MR 24 #A2844. [He 1] G. M. Henkin, Nonisomorphism of certain spaces of functions of different numbers of variables, Funkcional Anal. i Prilozen 1 (1967), no. 4, 57-68 = Functional Ana!. Appl 1 (1967), no. 4, 306-315. MR 36 #4331. [He2] , Banach spaces of analytic functions in a sphere and in a bicylinder are not isomorphic, Funkcional Ana!. i Prilozen 2 (1968), no. 4, 82-91 = Functional Anal. App!. 2 (1968), no. 4, 334-341. [He3] , Integral representations of functions holomorphic in strictly pseudoconvex domains and some applications, Mat. Sb. 78 (120) (1969), no. 4, 611-632 = Math. USSR Sbornik 7 (1969), no. 4, 597-616. MR 40 #2902. [Hel] H. Helson, Conjugate series and a theorem of Paley, Pacific J. Math. 8 (1958), 437-446. MR 20 #5397. [Hvl] V. P. Havin, Weak completeness of the space L I /H~, Vestnik Leningrad Univ. 13 (1973), 77-81. (Russian) [Hv2] , The spaces HOC and L I /H~, Zap. Nauen. Sem. Leningrad. Otde!. Mat. Inst. Steklov. (LOMI) 39 (1974), 120-148. (Russian) MR SO #965. [J] W. B. Johnson, A complementably universal conjugate Banach space and its relation to the approximation problem, Israel J. Math. 13 (1972),301-310. MR 48 #4700.

88

.\LEKSA.NDER PH..CZyNSKI

(J-R-Z] W. B. Johnson, H. P. Rosenthal and M. Zippin, On bases, finite dimensional decompositions and weaker structures in Banach spaces, Israel J. Math. 9 (1971), 488-506. MR 43 #6702. [Khl] J.-P.Kahane,Best approximation in Lt(T), Bull. Amer. Math. Soc. 80 (l974), 788-804. MR 52 #3845. [Kh2] , Another theorem on bounded analytic functions, Proc. Amer. Math. Soc. 18 (1967),827-831. MR 35 #5912. [Kisl] S. V. Kisljakov, Uncomplemented uniform algebras, Mat. Zametki 18 (1975), 91-96. (Russi till) [Kis2] - - - , On the conditions of Dunford-Pettis, Pefczynski, and Grothendieck, Dokl. Akad. Nauk SSSR 225 (1975),1252-1255 = Soviet Math. Dokl. 16 (1975),16161621. [K-P] M. I. Kadec and A. Petczyriski, Bases, lacunary sequences and complemented subspaces in the spaces L p ' Studia Math. 21 (1961/62), 161-176. MR 27 #2851. [K-V] A. Koranyi and S. Vagi, Singular integrals on homogeneous spaces and some problems of clas~ical analysis, Ann. Scuola. Norm. Sup. Pisa Sci. Fis. Mat. 25 (1972), 575648. [Kw] S. Kwapieri, On a theorem of L. Schwartz and its applications to absolutely summing operators, Studia Math. 38 (I970), 193-201. MR 43 #3822. [Kw-P] S. Kwapieri and A. Petczyriski, Some linear topological properties of the Hardy spaces HP, Compositio Math. 33 (l976), 261-288. [L-Pl] J. Lindenstrauss and A. Petczyriski, Absolutely summing operators in Lpspaces and their applications, Studia Math. 29 (I968), 275-326. MR 37 #6743. [L-P2] - - - , Contributions to the theory of the classical Banach spaces, J. Functional Analysis 8 (I971), 225-249. MR 45 #S63. [L-R] J. Lindenstrauss and H. P. Rosenthal, The Lp-spaces, Israel J. Math. 7 (I969), 325-349. MR 42 #5012. [L-Tl] J. Lindenstrauss and L. Tzafriri, The uniform approximation property in Orlicz spaces, Israel J. Math. 23 (1976),142-155. [L-T2] , Classical Banach spaces, Lecture Notes in Math. vol. 338, SpringerVerlag, Berlin and New York, 1973. [M] D. E. Marshall, Blaschke products generate He<>, Bull. Amer. Math. Soc. 82 (1976), 494-496. [Maul] B. Maurey, Espaces de cotype (p, q) et theoremes de reievement, C. R. Acad. Sci. Paris Ser. A-B 275 (l972), A785-A788. MR 49 #11235. [Mau2] , Theoremes de factorisation pour les operateurs lineaires ti valeurs dans les espaces LP, Asterisque No. 11, Soc. Math. de France, Paris, 1974. MR 49 #9670. [Mau-Pis] B. Maurey and G. Pisier, Caracterisation d'une classe d'espaces de Banach par des proprihes de series ateatoires vectorielles, C. R. Acad. Sci. Paris Ser. A-B 277 (l973), A687-A69t l . MR 48 #9352. [Mi] H. r.'illle, Banach space properties of uniform algebras, Bull. London Math. Soc. 4 (1972), 323-326. MR 48 #S47.

BANACH SPACES

89

[Moo] M. C. Mooney, A theorem on bounded analytic functions, Pacific 1. Math. 43 (1972),457-463. MR 47 #2374. [M-P] E. Michael and A. Perczynski, A linear extension theorem, Illinois 1. Math. 11 (1967),563-579. MR 36 #67l. [Mt-P] B. S. Mitjagin and A. Petczyriski, On the nonexistence of linear isomorphisms between Banach spaces of analytic functions of one and several complex variables, Studia Math. 56 (1975), 85-96. [M-S] B. Maurey et L. Schwartz, Espaces LP et applications radonifiantes, Seminaire Maurey-Schwartz, Annee 1972-1973, Exposes 1-27, Ecole Poly technique, Paris, 1973. - - - , Espaces LP, applications radonifiantes et geomerrie des espaces de

Banach, Seminaire Maurey-Schwartz, Annee 1973-1974; Exposes 1-26; ibid., Annee 19741975, Exposes 1-27, Ecole Poly technique, Paris, 1974, 1975. [<1>] N. <1>verlid, Integral representation formulas and LP estimates for the 3-equation, Math. Scand. 29 (1971), 137-160. MR 48 #2425. [PI] A. Petczyriski, Projections in certain Banach spaces, Studia Math. 19 (1960), 209-228. MR 23 #A344l. [P2] , On simultaneous extension of continuous functions. A generalization of the theorems of Rudin-Carleson and Bishop, Studia Math. 24 (1964), 285-304. MR 30 #5184a. ,Supplement to my paper "On simultaneous extension of continuous [P3] functions", Studia Math. 25 (1964), 157-161. MR 30 #5184b. [P4] , Uncomplemented function algebras with separable annihilator, Duke Math. 1. 33 (1966), 605-612. MR34 #3373. [P5] , Some linear topological properties of separable function algebras, Proc. Amer. Math. Soc. 18 (1967), 652-660. MR 35 #4737. [P6] , Sur certaines proprieres isomorphiques nouvelles des espaces de Banach de fonctions holomorphes A et H oo , C. R. Acad. Sci. Paris Ser. A 279 (1974), 9-12. MR 51 #1359. [P7] , On Banach space properties of uniform algebras, Spaces of Analytic Functions (Kristiansand, 1975), Lecture Notes in Math., vol. 512, Springer-Verlag, Berlin and New York, 1976, pp. 109-116. [P8] , On the impossibility of embedding of the space L in certain Banach spaces, Colloq. Math. 8 (1961), 199-203. MR 24 #AI008. [P9] , Banach spaces on which every unconditionally converging operator is weakly compact, Bull. Acad. Polon. Sci. Ser. Math. Astronom. Phys. 10 (1962), 265-270. MR 26 #6785. [PI0] , p-integral operators commuting with group representations and examples of quasi p-integral operators which are not p-integral, Studia Math. 33 (1969), 63-70. MR 39 #6125. [Pal] R. E. A. C. Paley, On the lacunary coefficients of power series, Ann. of Math. (2) 34 (1933), 615-616.

90

ALEKSANDER PElCZYNSKI

[Pa2] R. E. A. C. Paley, A note on power series, J. London Math. Soc. 7 (1932), 122-130. [Per] A. Persson, On some properties of p-nuclear and p-integral operators, Studia Math. 33 (1969), 213-222. MR 40 #769. [PhI] R. R. Phelps, Lectures on Choquet's theorem, Van Nostrand, Princeton, N. J., 1966. MR 33 #1690. [Ph2] , Extreme points in function algebras, Duke Math. J. 32 (1965), 267277. MR 31 #3890. [Pis] G. Pisier, Embedding spaces of operators into certain Banach lattices, Univ. of Illinois at Urbana-Champaign, 1976 (preprint). [Pi) A. Pietsch, Absolut p-summierende Abbildungen in normierten Raumen, Studia Math. 28 (1966/67), 333-353. MR 35 #7162. [P-P] A. Persson und A. Pietsch, p-nukleare und p-integrale Abbildungen in Banachriiumen, Studia Math. 33 (J 969), 19-62. MR 39 #4645. [P-R] A. Pclczyriski and H. P. Rosenthal, Localization techniques in LP -spaces, Studia Math. 52 (1974/75), 263-289. MR 50 #14l74. [P-S-W] G. Piranian, A. Shields and J. H. Wells, Bounded analytic functions and absolutely continuous measures, Proc. Amer. Math. Soc. 18 (1967),818-826. MR 35 #5911. [RI] H. P. Rosenthal, Projections onto translation-invariant subspaces of Lp(G), Mem. Amer. Math. Soc.No. 63 (1966). MR 35 #2080. [R2] , On factors ofC[O, 1] with nonseparable dual, Israel J. Math. 13 (1972),513-522. [Rm] E. Ramirez de Arellano, Ein Divisionsproblem und Randintegraldarstellungen in der komplexen Analysis, Math. Ann. 184 (1969/70), 172-187. MR 42 #4767. [Rul] W. Rudin, Projections on invariant subspaces, Proc. Amer. Math. Soc. 13 (1962), 429-432. MR 25 #1460. [Ru2] - - - , Trigonometric series with gaps, J. Math. Mech. 9 (1960), 203-227. MR 22 #6972. [Ru3] - - - , Fourier analysis on group§, Interscience, New York, 1962. MR 27 #2808. [Ru4] - - , Spaces of type H oo + C. Ann. Inst. Fourier (Grenoble) 25 (1975), 99-125. MR 51 #13692. [S] A. Sobczyk, Projection of the space (m) on its subspace (co), Bull. Amer. Math. Soc. 47 (1941), 938-947. MR 3,205. [Sch] H. H. Schaefer, Banach lattices and positive operators, Grundlehren Math. Wiss., Band 215, Springer-Verlag, Berlin and New York, 1974. [St] E. L. Stout, The theory of uniform algebras, Bogden Quigley, New York, 1971. [TJ] N. Tomczak-Jaegermann, The moduli of smoothness and convexity and Rademacher averages of the trace classes Sp (1 ,;;;;; p <00), Studia Math. 50 (1974), 163-182. MR 50 #8141. [Va] N. Th. Varopoulos, A theorem on operator algebras, Math. Scand. 37 (1975), 173-182.

BANACH

w.

[Vel

SP~CES

91

A. Veech, Short proof of Sobczyk's theorem, Proc. Amer. Math. Soc. 28

(1971),627-628.

MR 43 iffl79.

[Vin] S. A. Vinogradov, Interpolation theorems of Banach-Rudin-Carleson and norms of embedding operators for certain classes of analytic functions, Zap. Nauen. Sem. Leningrad Otdel. Mat. Inst. Steklov. (LOMI) 19 (1970),6-54. (Russian) MR 45 #4137 . [WI] P. Wojtaszczyk, On Banach space properties of uniform algebras with separable

annihilator, Bull. Acad. Polon. Sci. Ser. Math. Astronom. Phys. 25 (1977), No. I. [W2] , Weak(v compact operators from some uniform algebras, Studia Math. (to appear). [Z]

A. Zygmund, Trigonometric series. I, II, Cambridge Univ. Press, London and New

York, 1959. MR 21 #6498. [De3] F. Delbaen, The Pelczynski property for some uniform algebras, Studia Math. (to appear). [Kis3]

S. V. Kisljakov, Umform algebras as Banach spaces, Zap. Nauen. Sem. lenin-

grad Otdel. Mat. Inst. Steklov (LOMI) 65 (1976), 80-89. (Russian). [Kis4]

, On spaces with a "small" annihilator, Zap. Nallen. Sem. Leningrad

Otde!. Mat. Inst. Steklov (LOM!) 65 (1976),192-195. (Russian). [Kw-P2]

S. Kwapien and A. Pelczynski, Remarks on absolutelv slimming translation

invariant operators from the disc algebra and its dual into a Hilbert space (to appear). [W3] P. Wojtaszczyk, On projections in spaces of bounded analytic functions with applications, Studia Math. (to appear).