+ All Categories
Home > Documents > Operators into Lp which Factor Through lp

Operators into Lp which Factor Through lp

Date post: 05-Oct-2016
Category:
Upload: w-b
View: 245 times
Download: 16 times
Share this document with a friend
7

Click here to load reader

Transcript
Page 1: Operators into Lp which Factor Through lp

OPERATORS INTO Lp WHICH FACTOR THROUGH /,

WILLIAM B. JOHNSON

ABSTRACT

A bounded linear operator T into Lp[0,1] (2 < p < oo) factors through lp if and only if T iscompact when considered as an operator into L2 [0,1 ].

I. Introduction

In [3] E. Odell and the author proved that for 2 < p < oo a subspace of

Lp(= Lp[0, 1])

either contains a subspace isomorphic to l2 or embeds isomorphically into lp. Inother words, if T is an isomorphism from a Banach space X into Lp, and /2 is notisomorphic to a subspace of X, then T factors through some subspace of lp, i.e., thereare a subspace Y of /p and (bounded, linear) operators A : X -*• Y, B : Y -*• Lp sothat BA = T. The proof in [3], however, does not yield that T factors through lp

itself.In the present paper we classify the operators into Lp (2 < p < oo) which factor

through lp. It is well known (cf., e.g., the appendix to [9]) that every operator fromlp into L2 is compact; so a necessary condition for T : X -* Lp to factor through lp

is that T should be compact as an operator into L2. In §11 we prove that this conditionon T is also sufficient for T to factor through lp. Now it follows from [4] (cf. [3])that if T is an isomorphism from X into Lp, then T is compact as an operator into L2

if and only if X has no subspace isomorphic to /2; so the theorem stated in the abstractimproves the main result of [3] mentioned in the preceding paragraph.

We use standard Banach space theory terminology as may be found, for example,in [6].

II. Factoring operators into Lp, 2 < p < oo

Throughout this section, T is an operator from some fixed Banach space X intoLp (2 < p < oo) and T is compact when considered as an operator into L2. Wewant to prove that T factors through lp. Actually, we factor T through a space ofthe form Y = (J^(Hn, |-|n)),p, where each space (Hn, \-\n) is finite-dimensional. Weshall observe that the spaces (Hn, |-|,,) are uniformly isomorphic to uniformly comple-mented subspaces of Lp, and hence Y is isomorphic to a complemented subspace oflp. (Of course, this implies that Y is isomorphic to lp by a result of Pelczynski [8],but we do not need this fact, since it is clear that if T factors through a complementedsubspace of lp, then T factors through lp.)

The spaces (ff,,) are chosen to be a blocking of the Haar basis for Lp. That is,Hn = span (hd^^-1, where (/*,) is the Haar basis for Lp and 1 = k(l) < k(2) < ...is a suitably chosen sequence of positive integers. The operators A : X -*• Y and

Received 21 April, 1976.

Supported in part by NSF-MPS 72-04634-A03.

[J. LONDON MATH. SOC. (2), 14 (1976), 333-339]

Page 2: Operators into Lp which Factor Through lp

334 WILLIAM B. JOHNSON

B : Y -> Lp are defined in the natural way: for xeX with T* = %yn (yneHn), wedefine ,4* = {yj»ml. For j/neifn with (ytfml e 7, we define JB(yJ = 2>neLp.Obviously we have BA = T, but of course we have to show that A and B are boundedif the (Hn, |-IJ sequence is appropriately defined.

The reader who is familiar with the proof in [3] will observe that the approach in[3] for the case when T is an isomorphism was similar. In [3], an appropriate blocking(Hn) was defined, |-|n was the usual Lp norm restricted to Hn, and A was defined as inthe preceding paragraph. Now Hn is norm 2 complemented in Lp, since the Haarfunctions form a monotone basis for Lp; so Y = (Z,Hn)lp is isometric to a norm 2complemented subspace of lp. The operator A is continuous no matter how theblocking (Hn) is defined, and the hypothesis that T is compact as an operator intoL2 allowed us to define the blocking (Hn) of the Haar system to make A an isomor-phism. This means that the operator B defined in the preceding paragraph is con-tinuous when considered as an operator from AX into Lp. However, B cannot bebounded as an operator from Y to Lp when each |-|n is the usual Lp norm.

In the present paper, |-|n is defined in such a way that we must check all threethings: that Y is complemented in lp, that A is bounded, and that B is bounded.

It is convenient to define |-|B on all of Lp. For appropriate values of Mn, 1 = Mx <M2 < M3 < ..., |-|n is defined by

|/ln = max (MJ|/ | | 2 , | | / | | P) ,where

/i \i

B/lla s Ml/(OI2«M.

have their usual meaning. It is evident that each |-|n is equivalent to ||-||p on Lp, butas Mn \ oo the constant of equivalence tends to infinity.

We break the proof that T factors through Y if (Hn) and (MJ are defined appro-priately into three steps.

Step One. There is a constant K = K(p) such that {Hn, |«|J is K-isomorphic to aK-complemented subspace ofLp.

Of course, this means that Y is isomorphic to a complemented subspace of lp nomatter how Mn is defined.

Step one is easy, given a result of Rosenthal [10]. Rosenthal proved that there is aconstant A = A(p) so that for any sequence w = (wlt w2, ...) of positive numbers thespace XPt w is A-isomorphic to a A-complemented subspace of Lp. Here XPt w is thecompletion of R°° (or C00) under the norm \\-\\w defined by

It is easy to see that (Hn, |-|n) is isometric to a norm 2 complemented subspace ofXPi w for some w. Indeed, since each element of Hn is a step function and dimHn < oo,there is a sequence (even finite) of disjoint intervals (A() so that Hn c span (%Al). Let

" (= HxJI a/IIX*U

Page 3: Operators into Lp which Factor Through lp

OPERATORS INTO Lp WHICH FACTOR THROUGH lp 335

and set/, = (meas A{) llPxAi (so that ||/,||p = 1). Then, for any choice (a,) of scalars,

wt2)\

i.e., spanx^, is, in the |-|n norm, isometric to XPtW when w = (Mnwx, Mnw2, . . .).Thus, by Rosenthal's theorem, we can complete the proof of step one by observingthat (Hn, \-\n) is norm 2 complemented in (Lp, |-|n) and hence in span;^ . But theorthogonal projection P onto Hn satisfies | |P| |2 = 1 and (since the Haar functionsare a monotone, orthogonal basis for Lp) ||P||P < 2; hence \P\n < 2 by the definition

ofl-L.

Step Two. B has norm < 13 provided that, given Hlt H2, ..., Hn, Mn+2 is chosensufficiently large.

Suppose that the blocking (Hn) of the Haar functions and numbers (Mn) are given.We want to compute that for yn eHn, ||5>B||P < 13Q2 \yn\Z)l/p, as long as each Mn+2

is big relative to the modulus of uniform integrability of Ht+ ... +Hn.Let M = {n : \yn\n ^ 2nI>;J|p}. Certainly

III ym li Z ynnfM

+ I bn\

so we need check only that

and

p n e M

Vln

I y, +(Z.\yn\p)1/p>

2 n - l

Forn$M we have that Mn||yn||2 < \yn\n ^ 2n\\yn\\p) so that < 2" Mn

Now if 2" Mn * is very small, this means that yn is essentially supported on a set ofvery small measure; hence, if y is a fairly flat function in Lp, then

Thus, if Mn+2 is chosen big relative to the modulus of uniform integrability ofHx+ ... +Hn,then

y2, \2n t M

/p

This kind of reasoning is used (among other places) in [3] and [2]; we reproduce anargument in the proof of Theorem III.2 of [2] for the convenience of the reader whowishes to check the technical details.

For notational convenience, let K = {1} u {« : n $ M and n is odd}. We want toshow that

neKA Z 11*11? >

Page 4: Operators into Lp which Factor Through lp

336 WILLIAM B. JOHNSON

J

as long as Mn+2 grows fast enough. Set st = \. Given H = Hl+ ... +Hn, chooseen+! > 0 with 2en+1 < en so that iffeH and E £ [0, 1] with meas £ < e^T1 } , then

i/p

i/rl

\E /

Now assume that 2"+2 M~+2 < 2"( n + 2 ) pen+i; i.e., that Mn+2 is small with respect tothe modulus of uniform integrability of Hx 4- ... +Hn.

Set £x = [0, 1]; for neK with « > 1 set En = {t: |^,(0l > e i - l~ p ) Ibnllp). and letOO

Fn = En ~ (J £,-. (Here and in the next two paragraphs unions and sums are«=n+l

taken over indices in K.) Observe that, by the definition of En,

meas Fn ^ meas£n < e,^-i~1),

so that meas ([0, 1] ~ (J Fn) = 0 because £e p / ( p ~ 1 ) < oo. Thus

i/p

| p

•fn

n - 2

+\y*\'+i=n + 2

I /P

S y,

1/P1

= 3[a+b + c].

Observe that, by Schwartz's inequality, we have for n > 1 that

i / p l/2p

•>-%

lnMn-'\\yn\\pYIPt~n^

Letting Dt = [j Fn, we thus have that

" \\yn\\p.

1-2un = 1

C =

00

1 = 3

1/P

•( /

T yiXD,i = 3

I / P

Of course we have ft ̂ (X ll);n||p)1/p> so it remains to estimate a. Since

measF,, ^ e^r1*, and ^ 4 - ... +j/B_2eH1+ ... 4-Hn_2,

Page 5: Operators into Lp which Factor Through lp

OPERATORS INTO Lp WHICH FACTOR THROUGH lp 337

we have from the choice of en and the monotonicity of the Haar basis for Lp that

n-2v - l

n = 3

Therefore

max

Z2 n - l

A similar argument shows that

hence

3-2(1 ||^

13(Zas desired.

Recalling that the blocking Hn = spansequence 1 = A:(l) < k(2) < ..., we state

^ ^ " 1 is denned by the increasing

Step Three. A has norm ^ K\\T\\ (where K = Kp is a constant which dependsonly on p) provided that, given Mn(n > 1), k(n) is sufficiently big relative to Mn.

Let || S || 2 be the norm of the operator S when considered as an operator into L2.Let Rn be the orthogonal projection from L2 onto span (/*j)?Ln in L2. Our hypothesisthat T is compact as an operator into L2 implies that ||.Rn T| |2-> 0 as n -* oo.Suppose now that \\Rk(n)T\\2 < 2""M,,"1 | T | | for n = 2, 3, .... For xeX with

Tx= yn (yneHn\

we need to show that

Let M = {n : \yn\n = | | ^ n | | p } . Since the Haar system forms an unconditional basisfor Lp [7], there is a constant 0 < X = X(p) so that

(cf., e.g., the argument on p. 209 of [9]); thus

i/p / \ i/p/ \ 1 / p /

\neAf / \n eM

Observing that 1 e M (since M± = 1), we have that

cl lyjj?i /p

Z Mnk=n 2 n = 2

Page 6: Operators into Lp which Factor Through lp

338 WILLIAM B. JOHNSON

Thus

( 00 1/p

as desired.Of course, to complete the proof that T factors through lp, we only have to make

the obvious observation that the sufficient conditions in steps two and three for theboundedness of B and A are not mutually exclusive.

Remark 1. The ideas used in steps two and three are contained implicitly in [3]and [2]. The main new idea here involves the use of the weighted L2 norm in thedefinition of |-I,,.

Remark 2. The interesting thing to me about the argument presented here is thatRosenthaPs not-yet-classical space Xp plays a fundamental role in the proof of atheorem whose statement mentions only classical spaces.

Remark 3. It is very easy to prove (cf. [1]) that every compact operator into Lp

factors through lp. Say that an operator into Lp satisfies (*) provided that eachE £ [0,1] with meas E > 0 contains a subset F such that meas F > 0 and the operatorXF T (defined by (xF T)x = xF.(Tx)) is compact. A standard measure-theoreticexhaustion argument shows that if T satisfies (*), then T factors through lp. Ouroriginal approach to the problem discussed in this paper was to show that if T : X -+LP

is compact as an operator into L2, then T satisfies (*). This is not true, however;in fact the basis-to-basis map from the unit vector basis of lt to the normalized Haarbasis for Lp fails (*). It is even possible to construct an isomorphism from \p intoLp which fails (*). Perhaps the collection of operators into Lp which satisfy (*) is aninteresting class which is worthy of study.

Remark 4. Of course, the dual version to the result proved here is that anoperator T with domain Lq (1 < q < 2) factors through 1q if and only if T is compactwhen considered as an operator fromL2. This is false for q = 1. The proofs in [5]yield that an operator T from Lt factors through \y if and only if T is differentiate ifand only if T satisfies the dual condition to (*)—i.e., every set E £ [0, 1] of positivemeasure contains a subset F of positive measure so that the restriction of T to L^F)is compact. On the other hand, J. Uhl has observed that an operator T from Lt is aDunford-Pettis operator if and only if T is compact when considered as an operatorfrom L2, but it is known that there are Dunford-Pettis operators on Lx which arenot differentiate.

Remark 5. We conjecture that an operator T from Lp (2 < p < oo) factors throughlp if and only if T is of type p Banach-Saks; i.e., there exists a constant A suchthat every weakly null normalized sequence in L_ has a subsequence (xn) for which

Txt f o r / i =

This conjecture was verified in [2] in the case when T has closed range.

Remark 6. There is a non-separable analogue to the theorem stated in the abstract.Suppose that 2 < p < oo and T : X -> Y - (£a 6 A Lp(/ta))/p(/1) is an operator, whereeach jia is a probability measure. (It is well known that every Lp(/i) space is isometric

Page 7: Operators into Lp which Factor Through lp

OPERATORS INTO Lp WHICH FACTOR THROUGH lp 339

to such a 7.) For each a e A, let Ta be the composition of T with the projection from

y onto Lp(/ta). Then T factors through

if and only if each Ta is compact when considered as an operator into L2(na).

References

1. W. B. Johnson, " Factoring compact operators ", Israel J. Math., 9 (1971), 337-345.2. , " Quotients of Lp which are quotients of/p ", Compositio Math., to appear.3. and E. W. Odell, " Subspaces of Lp which embed into /p ", Compositio Math., 28 (1974),

37^9.4. M. I. Kadec and A. Pelczynski, " Bases, lacunary sequences, and complemented subspaces in the

spaces Lp ", Stadia Math., 21 (1962), 161-176.5. D. R. Lewis and C. Stegall, " Banach spaces whose duals are isomorphic to U (T) ", / . Functional

Analysis, 12(1973), 177-187.6. J. Lindenstrauss and L. Tzafriri, Classical Banach spaces, Lecture notes in Mathematics 338.

(Springer-Verlag, 1973).7. R. E. A. C. Paley, " A remarkable series of orthogonal functions ", Proc. London Math. Soc,

34 (1932), 241-264.8. A. Pelczynski, " Projections in certain Banach spaces ", Studia Math., 19 (1960), 209-228.9. H. P. Rosenthal, " On quasi-complemented subspaces of Banach spaces, with an appendix on

compactness of operators from Lp(n) to Lr{v)", J. Functional Analysis, 2 (1969), 176-214.10. H. P. Rosenthal, " On the subspaces of Lp (p > 2) spanned by sequences of independent random

variables", Israeli. Math., 8 (1970), 273-303.

The Ohio State University,Columbus, Ohio 43210

U.S.A.


Recommended