You are on page 1of 7

Math. Ann.

24l, 35 41 (1979)
Am
© by Springer-Verlag 1979

Dunford-Pettis Sets in the Space


of Bochner Integrable Functions*
Kevin T. Andrews
Department of Mathematics, University of Illinois, Urbana, IL 61801, USA

One question of long standing interest is whether the space L,(#,X) of Bochner
integrable functions has the Dunford-Pettis property if the Banach space X has the
Dunford-Pettis property. The lack of a characterization of weakly compact subsets
of LI(#,X ) has impeded a successful solution to this problem. In an attempt to
circumvent this difficulty we shall study those sets in LI(#,X) that are mapped into
norm compact sets by weakly compact operators on LI(#,X ). The paper opens
with a study of those sets in an arbitrary Banach space X that mapped into
relatively norm compact sets by all weakly compact operators on X. These sets are
called Dunford-Pettis sets. An internal characterization of Dunford-Pettis sets is
given. In the second section we give a rather general sufficient condition for a
subset of LI(#,X) to be a Dunford-Pettis set. It is then deduced that if X is a
Banach space with the Dunford-Pettis property and contains no copy of E1, then
LI(#,X) has the Dunford-Pettis property for all finite measures #. The third section
deals with specific sets in L~(#,X) that are known to be weakly compact and shows
that if X has the Dunford-Pettis property, then these specific sets are Dunford-
Pettis sets.
Throughout this paper (D, X, #) is a finite measure space and X and Y are
Banach spaces with duals X* and Y* respectively. The space of all bounded linear
operators from X to Y under the usual operator norm will be denoted by L(X, Y).
The space of #-Bochner integrable functions on D with values in X will be denoted
by L~(#,X). A subset M of L,(#,X) is called uniformly integrable if

~(~)-o
lim ! IIflld/z=0

uniformly in f s M.

* This work will constitute a portion of the author's Ph.D. thesis now in preparation at the
University of Illinois under the direction of Professor J. J. Uhl, Jr.

0025-5831/79/0241/0035/$01.40
36 K.T. Andrews

L Dunford-Pettis Sets
Definition. A bounded subset A of a Banach space X is called a Dunford-Pettis set
if for all Banach spaces Y every weakly compact operator from X to Y maps A into
a relatively norm compact set.
Since a Banach space X has the Dunford-Pettis property if and only if every
weakly compact operator from X to any other Banach space maps weakly
compact sets into norm compact sets, a Banach space has the Dunford-Pettis
property if and only if each of its weakly compact subsets is a Dunford-Pettis set.
One often reads in the literature that a space has the strict Dunford-Pettis property
if each of its weakly conditionally compact sets (i.e. a set A such that any sequence
in A has a weakly Cauchy subsequence) is a Dunford-Pettis set. Actually as is
pointed out in [-5, p. 177] and elsewhere, the strict Dunford-Pettis property and the
Dunford-Pettis property are identical properties. We shall use this fact freely. In
this regard we shall also use the following theorem of Odell and Stegall (see [8, p.
377]). Every Dunford-Pettis set is weakly conditionally compact.
A direct consequence of the Odell-Stegall theorem is the following fact.
A Dunford-Pettis set in LI(#,X) is bounded and uniformly integrable. To see why
this is true, note that ifMC=L~(p,X) is a Dunford-Pettis set, then it is bounded and
if it is not uniformly integrable an appeal to [5, p. 104] shows that it contains a
copy of the (~-unit vector basis, a set which is not weakly conditionally compact.

Theorem 1. Any one of the following statements about a bounded subset A of a


Banach space X implies all the others:
a) The set A is a Dunford-Pettis set.
b) Every weakly compact operator from X into c o maps A into a relatively norm
compact set.
c) I f (x*) is a weakly null sequence in X* and (x,) is a sequence in A, then
lim x*(x,)=0.

Proof Obviously a) implies b); to see that b) implies c), let (x*) be a weakly null
sequence in X*. Define an operator T :X-~c 0 by Tx = (x'x). Notice that the adjoint
T :~1 -~X* of T has action T*(c0 = ~ e,,x* for each c~= (~,) in fl. Accordingly T*
n=l
maps the closed unit ball of E1 into the absolutely convex null of the weakly
compact set { x * : n e N } . Hence T*, and therefore T, is a weakly compact operator.
Thus by hypothesis T(A) is relatively norm compact. By the well-known charac-
terization of relatively norm compact subsets of c o, this means
lim sup [x*(x)[ = 0 .
n x~A

This shows that b) implies c).


To prove that c) implies a) suppose there exists a Banach space Y and a weakly
compact operator T 9(--* Y such that T(A) is not relatively norm compact. By the
Davis-Figiel-Johnson-Pelczynski factorization lemma [3, p. 314] we can and do
assume that Y is reflexive. Now since T(A) is not relatively norm compact but is
relatively weakly compact there exists a sequence (x,) in A, a y e Y and a 6 > 0 such
Dunford-Pettis Sets 37

that limt~
7 k , = y weakly but HTx.-y][ > 6 for all positive integers n. For each
positive integer n choose y* ~ Y* with ][y* t] < 1 such that y,*(Tx, - y ) > 6. Since Y* is
reflexive, by passing to a subsequence if necessary we can and do assume that (y*)
converges weakly to some y*e Y*. Hence lim T * ( y * - y * ) = O weakly and so by
n

hypothesis
0 = lim T*(y* - y*)(x,) = lim (y* - y*)T%.

But l i m y * ( T k , - y ) = O and l i m ( y * - y * ) y = O . Hence

0 = li~n [(y* -- y*)Tx, + y*(Tx, - y) + (y* - y*)y]

= lim y * ( T k , - y ) > 6 > 0 ,

a contradiction which completes the proof of the theorem.

II. Dunford-Pettis Sets in Ll(p, X)


The main theorem of this paper is
Theorem 2. Let M be a bounded uniformly integrable subset of LI(p,X ). Then M is
a Dunford-Pettis set if for each 6 > 0 there is a set E~X with p(E) <6 such that for
each co(~E there is a Dunford-Pettis set D(o))~X with f(e~)~ D(co) Jor all f s M .
Proof According to Theorem 1 it suffices to show that if T : L I ( p , X ) ~ c o is a
weakly compact operator, then T(M) is relatively norm compact. Let T be such a
weakly compact operator and note that for each x ~ X {T(x~b):~b is a simple
function in LI(#) and [[q~[[,< 1} is weakly compact in c 0. Thus by a theorem of
Dinculeanu and Ionescu-Tulcea [6, Theorem 8, pp. 279 280] there is a function
g :Y2~L(X, Co) so that
(Tf, a) = ~ (gfo~Fd# (,)

for all ~z6#1 and for all f ~ LI(P,X ). Consulting the proof of this theorem in [6, pp.
263-280] one finds that without loss of generality we can and do assume
g(o~)(x)C=T { f : f e LI(/2,X ), }lfl/1 =<1 } (1)
for all (o~2 and x ~ X with Hxt[< 1

]]g(" )NLCX,co)EL~(I~) and


ess sup IIg( ' )]lL(X,co)=]1T]I • (2)
Next notice by (*) that the vector valued function g( • )x is scalarly integrable
for each x~X. In particular g(- )x is weakly measurable [5, p. 41] for each xaX.
Since for each x 6 X , g(. )x is separably valued (because it has values in co), the
Pettis measurability theorem [5, p. 42] guarantees that g(.)x is (strongly)
measurable for each x~X. From this and (2) we have that (*) takes the form
T(f) = Bochner - I fgdt~
I2
38 K.T. Andrews

for all simple functions feLl(,u,X ). It follows that


T(f) = Bochner - ~ fgdla
fl

for all f eLl(la, X ).


For future use write for each xeX and e)eO
~(~o)x=(~.(o~)x),
where each 9. :~-+X* is weak*-measurable [5, p. 41] and sup Ilo.II ~ < {ITI{. Notice
that this means lim
n
9,(o))=0 in the weak*-topology of X* for each ~ o ~ . In fact
lim 9.((o)= 0 in the weak topology of X* for each me ~. To see why this is true fix
~0e Q and note that (1) implies that 9((o):X--+Co is a weakly compact operator. If
(e.) is the unit vector basis of (~, then g(o~)*(e.)=g.(eo). Hence {g,(o):neN} is
relatively weakly compact in X*. Since (9,((o)) tends to zero in the weak*-topology,
it tends to zero in the weak topology.
Now assume by way of contradiction that T(M) is not relatively norm
compact ; i.e.
lim sup ](Tf).] > 0.
n f~M

Since ( T f ) , = ~ #,,fdfa for f e M , there are positive integers n 1 < n 2 < ..,-<nj < . . .
12
and a sequence (fj) in M such that

5ag.,fjd # >e
for some e > 0 and all j = 1,2 ..... By the uniform integrability of M choose 3 > 0
such that/,(F) <6 implies

5 I[flld/~< e
F 211TII
for all feM. Choose a set EeS with #(E)<6 such that for each co$E there is a
Dunford-Pettis set D(c0) with f(co)eD(co) for all f e M . Then since limg.,(o))=0
J
weakly Theorem 1 guarantees that lim g.j(a))fj(o)=0 for o)¢E. Since the sequence
J
(g,,(c0)fj(o)) is a uniformly integrable sequence, the Vitali convergence theorem
ensures that lira I [g,,fjld#=0. Hence for j sufficiently large I Ig.Jjld#<E/2.
J ~\E ~\E
But then

fg./,d, =<!lg.,s, la + ~ f\ E
< sup IIg.,ll~ sup 5 tlfjl[d/~+ 5 Jg.,fjld#
J J E ~\E

< IlZtl ~ t l ~ + ~ / 2 = g ,

a contradiction which completes the proof.


Dunford-Pettis Sets 39

The first corollary is reminiscent of a theorem of Diestel's [4, p. 89] dealing


with weak compactness in LI(p,X ).
Corollary 3. Let X have the Dunford-Pettis property. Let M be a subset of LI(p,X )
with the property that for every ~ > 0 there is a weakly conditionally compact subset
W~,of X such that for each f in M there is a subset A~,I of g2 such that
[[ftld#<e (1)

and
f(A~, s) C 1,1,;. (2)
Then M is a Dunford-Pettis set.
Proof. Consider the set
M~ = {f)~A,., :fG M } .
Since M s is L~(#,X) bounded it is uniformly integrable and if geM~, then
g(f2)c= W~w{0}, a set which is evidently weakly conditionally compact. Since X has
the Dunford-Pettis property, the set W~w{0} is a Dunford-Pettis set. By Theorem
2, the set M s is a Dunford-Pettis set. Since l l f - f)~A~~llx <e for all f in M and e > 0
is arbitrary it follows quickly that M is a Dunford'Pettis set. This completes the
proof.
The problem of whether La(#,X) has the Dunford-Pettis property whenever X
has the Dunford-Pettis property is open. Bourgain [1] has shown that for an
arbitrary Banach space X the sets M in L:(Iz, X) having the property described in
the hypothesis of Corollary 3 are weakly conditionally compact. He also shows
that the converse fails for some Banach spaces but holds for spaces not containing
{1. In any case Corollary 3 is in harmony with the possibility that LI(p,X) has the
Dunford-Pettis property whenever X has the Dunford-Pettis property. For
Banach spaces containing no copy of f~ we can make a contribution to this
problem.
Corollary 4. Let X have the Dunford-Pettis property and suppose X contains no
copy of E:. A subset of Ll(p,X ) is a Dunford-Pettis set if and only if it is bounded
and uniformly integrable. Consequently the space L:(p,X) has the Dunford-Pettis
property.
Proof We have already seen that a Dunford-Pettis set in La(p,X) is bounded and
uniformly integrable.
To prove the converse suppose MCLI(p,X) is bounded and uniformly
integrable. It follows that

lira o[ IlflIdp=O
n [llfi} > n ]

uniformly in f e M . Let e > 0 and choose n o such that

.[ llflld~<e
[lloqt> not
for all f e M .
40 K.T. Andrews

Put W~:{xeX:llxll~n o} and A~,s={co:llf(~o)N<=no}. Then [. Ilfl[dl~<e


I?\A~,y-
and f(A~,i)c W~.for all f e M . Now since X contains no copy of El, the set W~ is
weakly conditionally compact by a theorem of Rosenthal's [7, p. 2411]. An appeal
to Corollary 3 completes the proof.
It is worth remarking that Corollary 4 cannot be improved. Indeed if X is a
Banach space such that the bounded uniformly integrable subsets of Li(la, X) are
Dunford-Pettis sets, then the set {xza:xeX, Ilxll < 1} is a Dunford-Pettis set. By
the Odell-Stegall theorem we see that this set is weakly conditionally compact. It
follows immediately that the unit ball of X is weakly conditionally compact ; hence
X contains no copy of ~1 by Rosenthal's theorem [7, p. 2411].

IlL Weakly Compact Sets in L 1(p, X) that are Dunford-Pettis Sets


In principle the complete lack of characterizations of weakly compact and weakly
conditionally compact sets in LI(I~,X ) is a barrier to the solution of the question of
whether L~(/~,X) has the Dunford-Pettis property if X has the Dunford-Pettis
property. Nevertheless we can show that for an arbitrary Banach space X with the
Dunford-Pettis property certain known weakly compact sets in LI(#,X) are
Dunford-Pettis sets. For example Diestel [4, p. 89] has shown that if M is a
bounded uniformly integrable subset of L~(II,X) such that for each e > 0 there is a
set A~ with p(A~)<e and a weakly compact set W~ in X with f(£2\A~)C=W~ for all
f ~ M , then M is a relatively weakly compact set. Since X has the Dunford-Pettis
property, the set W~is a Dunford-Pettis set. A glance at Corollary 3 shows that this
type of weakly compact set in LI(#,X ) is a Dunford-Pettis set.
Brooks and Dinculeanu [2, p. 174] have shown that if M is a bounded
uniformly integrable of L I(/~,X) then M is relatively weakly compact provided that
for each E ~ Z the set {! f dl~ :f ~ M} is relatively weakly compact in X and for every
countable subset M o of M there is a sequence (H,) of partitions of O into disjoint
members of Z such that lim NEn, f - f N x = 0 uniformly in f ~ M o [here i f / 7 is a

partition, then En(f)=r~(([fdl~)/l~(E)lzE 1. To show that such a set M is a


Dunford-Pettis set if X has the Dunford-Pettis property, it suffices to show that
every countable subset of M is a Dunford-Pettis set. To this end let M o be a
countable subset of M and e >0. Choose a partition/7 such that HEn(f )-fill <e
for all f in M o. Let
M~o= {En(f):f eMo} •
For each coef2 select the EeII such that e)eE and write

/!i..
Then since X has the Dunford-Pettis property, each of the relatively weakly
compact sets D(~o) is a Dunford-Pettis set. An appeal to Theorem 2 reveals that M;
is a Dunford-Pettis set.
Dunford-Pettis Sets 41

Acknowledgement. The author is grateful to Professors N. J. Kalton and H. P. Rosenthal for helpful
discussions concerning this paper and to Professor J. J. Uhl, Jr. for his constant advice and
encouragement during its preparation.

References

1. Bourgain, J. : An averaging result for El-sequences and application to weakly conditionally compact
sets in LI(#,X ) (preprint)
2. Brooks, J.K., Dinculeanu, N. : Weak compactness in the space of Bochner integrable functions and
applications. Advances in Math. 24, t72-188 (1977)
3. Davis, W.J, Figiel, T., Johnson, W.B., Pelczynski, A. : Factoring weakly compact operators. J. Funct.
Anal. 17, 311-327 (1974)
4. Diestel, J. : Remarks on weak compactness in LI(#,X ). Glasgow Math. J. 18, 87-91 (1977)
5. Diestel, J., Uht, J.J.,Jr. : Vector measures. Math. Surveys No. 15, American Mathematics Society,
Providence 1977
6. Dinculeanu, N. : Vector measures. New York : Pergamon Press 1967
7. Rosenthal, H.P.: A characterization of Banach spaces containing/1. Proc. Nat. Acad. Sci. USA 71,
2411-2413 (1974)
8. Rosenthal, H.P.: Pointwise compact subsets of the first Baire class. Amer. J. Math. 99, 362 378
(1977)

Received August 8, 1978

Added in prooL S. S. Khurana has recently obtained an independent proof of the final statement of
Corollary 4. We also note that an easy application of Theorem 2 yields the result that if X has the
Dunford-Pettis property then the space of absolutely summable X-valued sequences (1(X) has this
property.

You might also like