central limit theorems for weighted sums of exchangeable random elements in banach spaces ...

17
This article was downloaded by: [Moskow State Univ Bibliote] On: 02 October 2013, At: 14:20 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Stochastic Analysis and Applications Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/lsaa20 Central limit theorems for weighted sums of exchangeable random elements in banach spaces Peter Z. Daffer Published online: 03 Apr 2007. To cite this article: Peter Z. Daffer (1984) Central limit theorems for weighted sums of exchangeable random elements in banach spaces , Stochastic Analysis and Applications, 2:3, 229-244 To link to this article: http://dx.doi.org/10.1080/07362998408809035 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/ terms-and-conditions

Upload: peter-z

Post on 14-Dec-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

This article was downloaded by: [Moskow State Univ Bibliote]On: 02 October 2013, At: 14:20Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Stochastic Analysis and ApplicationsPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/lsaa20

Central limit theorems for weightedsums of exchangeable randomelements in banach spacesPeter Z. DafferPublished online: 03 Apr 2007.

To cite this article: Peter Z. Daffer (1984) Central limit theorems for weighted sums ofexchangeable random elements in banach spaces , Stochastic Analysis and Applications, 2:3,229-244

To link to this article: http://dx.doi.org/10.1080/07362998408809035

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to orarising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms& Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Page 2: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

STOCHASTIC ANALYSIS AND APPLICATIONS, 2 ( 3 ) , 229-244 (1984)

CENTRAL LIMIT THEOREMS FOR WEIGHTED SUMS OF

EXCHANGEABLE RANDOM ELEMENTS I N BANACH SPACES(')

P e t e r Z . Daf fe r

Department o f Mathematics

Vanderb i l t U n i v e r s i t y

N a s h v i l l e , Tennessee 37235

ABSTRACT

A g e n e r a l i z a t i o n o f a c e n t r a l l i m i t theorem f o r m a r t i n g a l e d i f f e r e n c e a r r a y s due t o D . L . McLeish i s o b t a i n e d f o r random e l e - ments i n a s e p a r a b l e Banach space E . T h i s r e s u l t , wi th a t echnique o f N . C . Weber, i s used t o o b t a i n a weak convergence theorem f o r weighted sums a X o f a row-wise exchangeable a r r a y (Xnk) of random elements !nn! Rkanach space E which i s un i formly 2-smooth. C o r o l l a r i e s i n c l u d e a c e n t r a l l i m i t theorem f o r weighted sums xkankXk o f an exchangeable sequence ( X k ) , and s e v e r a l weak laws of l a r g e numbers f o r such weighted sums.

0 . INTRODUCTION

In s e c t i o n 2 a n a t u r a l e x t e n s i o n o f a m a r t i n g a l e c e n t r a l l i m i t

theorem due t o D . L . McLeish [6] t o a r r a y s o f random e lements t a k -

ing v a l u e s i n a s e p a r a b l e Banach space E i s o b t a i n e d . The random

elements i n each row form a f i n i t e sequence o f m a r t i n g a l e

d i f f e r e n c e s i n E . The a d d i t i o n a l h y p o t h e s i s t h a t i s needed i n t h e

i n f i n i t e dimensional c a s e t u r n s o u t t o be t i g h t n e s s o f t h e

sequence o f row sums.

(1) Research suppor ted by a g r a n t from t h e V a n d e r b i l t U n i v e r s i t y

Research Counci l .

Copyright O 1984 by Marcel Dekker, Inc.

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 3: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

2 30 DAFFER

I n s e c t i o n 2 t h e m a r t i n g a l e t echnique o f N . C . Weber [ l o ] i s

used t o o b t a i n a c e n t r a l l i m i t theorem f o r weighted row sums

Zkankxnk of an a r r a y (Xnk) of random elements i n E , i n which each

row forms an exchangeable s e t . To o b t a i n convergence t o a nonde-

g e n e r a t e (Gaussian) l i m i t , t h e h y p o t h e s i s t h a t t h e Banach space E

be uniformly 2-smooth i s used. C e n t r a l l i m i t theorems a r e ob ta ined

f o r weighted s u m s x a X of an i n f i n i t e exchangeable sequence k nk k

(Xk) i n E a s c o r o l l a r i e s . A s f u r t h e r c o r o l l a r i e s , s e v e r a l weak

laws o f l a r g e numbers f o r weighted sums z k a n k x k a r e ob ta ined

under s t r o n g e r c o n d i t i o n s on t h e a r r a y (ank) o f weigh ts .

1. NOTATION AND PRELIMINARIES

A l l Banach spaces i n t h i s paper a r e r e a l and s e p a r a b l e . A

p r o b a b i l i t y measure p i n a Banach space E i s g iven on t h e

a - a l g e b r a & o f Bore1 s e t s i n E. Sequences o f random elements

( r . e . ' s ) (X ) i n E a r e def ined on a ( s u f f i c i e n t l y r i c h ) p r o b a b i l i t y k space (n, F, P) . Expec ta t ions and c o n d i t i o n a l e x p e c t a t i o n s a r e

t aken i n t h e Bochner s e n s e ; f o r an e x c e l l e n t account of t h e i r

p r o p e r t i e s , used f r e e l y below, s e e S c a l o r a [ 7 ] .

We w r i t e E* f o r t h e dua l o f E . The c h a r a c t e r i s t i c func-

t i o n a l cp : E* -*C ( = {complex nos.)) o f a p r o b a b i l i t y measure p on E i s g iven by (P ( f ) = SEexpi f (x) p(dx) , f € E*, and determines

t h e measure ,IJ unique ly ( see , e . g . , Vakhania [ 9 ] , 54.1 f o r

p r o p e r t i e s ) . The covar iance o p e r a t o r R : E*-+E o f a weakly second o r d e r

2 p r o b a b i l i t y measure p on E ( i . e . , ,fEf (x)p(dx) < w , 'df E E*) is

t h e bounded, l i n e a r , symmetric and non-negat ive o p e r a t o r d e t e r -

mined by t h e b i l i n e a r form .fEf ( ~ ) ~ ( x ) p ( d x ) by posing

( Rf, g > = s E f ( ~ ) g ( x ) ~ ( d x ) , f , g e E*. I f p i s t h e d i s t r i b u t i o n 2 o f a s t r o n g l y s e c o n d ~ r d e r r , e . X i n E ( i . e . EIIX]~ .5 GW ), a s i s

always t h e c a s e below, t h e n R : E*--+ E i s compact. The c h a r a c t e r -

i s t i c f u n c t i o n a l o f a Gaussian d i s t r i b u t i o n o f E h a s t h e form

exp -; ( ~ f , f ) , f E E*.

For each n = 1, 2 . . . , l e t C z, C . . . . C Fn, kn C (

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 4: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS 231

be r - a l g e b r a s . An a r r a y (Xnk, ) , k = 1, ..., kn, o f r . e . ' s i n n , k

E , where Xnk i s ( c , k -& ) measurable and E(Xnk ~ j r ; , ~ - ~ ) = 0,

almost s u r e l y ( a . s . ) f o r each k = 1, ..., kn, i s c a l l e d a mar t in -

g a l e d i f f e r e n c e a r r a y i n E .

R denotes t h e r e a l l i n e , N = { I , 2 , ...) t h e n a t u r a l numbers,

and 1(A) t h e i n d i c a t o r f u n c t i o n of t h e s e t A : I(A) (a) = 1 i f

w c A, I (A) (a) = 0, i f w 4 A . The argument ~3 i s u s u a l l y SUP-

p ressed . Convergence i n p r o b a b i l i t y i s i n d i c a t e d by '$' o r

"p - l i m " . For r . e . I s (X,), X i n E wi th r e s p e c t i v e d i s t r i b u t i o n s W W

( p n ) , p , Xn+X o r ~ ~ d ~ i n d i c a t e s weak convergence of t h e measures

(i. e . , Ef (Xn) -4 Ef (X) f o r every bounded cont inuous f : E + R ) . For

a sequence o f c o n s t a n t s ( o r random v a r i a b l e s ) (c,), Xn = o (cn) P

means c - l x n ~ O . An i n d i c a t e d "k" i n "max k 'I, ltXktt i s always t a k e n

from 1 t o kn.

2. A CENTRAL LIMIT THEOREM FOR MARTINGALE DIFFERENCES

The fo l lowing r e s u l t due t o McLeish [6] w i l l be needed.

Theorem 2.1

Let (Xnk, q n , k ) , k = 1, . . ., kn, n E N , be an a r r a y of r e a l - 2

va lued m a r t i n g a l e d i f f e r e n c e s s a t i s f y i n g E [xnkl < 00 , f o r a l l n ,k .

Suppose t h a t

( i ) m p 1xnkl-%;

( i i ) s ~ p E {np jxnkl 2 , < m i

( i i i ) 5 x ~ ~ ~ - & ~ , cr2 k 0 non random.

2 Then Sn = Zkxnkd~ (0, u ) , a s n+ co . Reca l l t h a t a sequence (Xn) o f r . e . ' s i n a Banach space E is

s a i d t o be tight i f , t o every E > 0, t h e r e i s a compact s e t K C E

such t h a t sup P [Xn $k K] 5 c . ncN

Theorem 2.2

LET (Xnk, Fn, k), k = 1, . . . , kn, n E N , be a m a r t i n g a l e

d i f f e r e n c e a r r a y i n a Banach space E, s a t i s f y i n g :

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 5: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

( i ) mtx llxnk1l 40, a s n + m ;

DAFFER

( i i i ) (S,), where Sn = Z k ~ n k , i s t i g h t ;

2 ( i v ) tkf (Xnk) --!Ozf, $f & 0, non-random, f o r each f r E*.

W Then S n - + Z , where Z i s a cen te red Gaussian r . e . i n E . The char -

a c t e r i s t i c f u n c t i o n a l of Z i s cp(f) = exp -: ( R f , f > , f E E*,

where R : E * A E i s t h e covar iance o p e r a t o r o f Z .

Proof

Let f E E*. Then ( f (Xnk) , < , k ) , k = 1, . . . , kn, i s a mar t in -

g a l e d i f f e r e n c e a r r a y i n R , and ~ h k o r e m 2.1 y i e l d s f (Sn) 2

= I k f ( x n k ) 4 ~ ( ~ , f ) .

This means t h a t a l l weak l i m i t s o f subsequences o f (Sn) a r e 2 t h e same: t h e p r o b a b i l i t y measure wi th marg ina l s N(0, (r f ) (use

B i l l i n g s l e y [ I ] , Theorem 5.1, and t h e c o n t i n u i t y of f E E*). Now

by ( i i i ) t h e sequence (Sn) i s t i g h t , hence by Prokhorov's Theorem

( B i l l i n g s l e y [ I ] ) c o n d i t i o n a l l y compact, and s o every subsequence

of (S ) c o n t a i n s a f u r t h e r subsequence which converges weakly t o n t h i s p r o b a b i l i t y measure. By B i l l i n g s l e y [ I ] , Theorem 2 . 3 , t h e

whole sequence (Sn) converges weakly t o t h i s p r o b a b i l i t y measure,

which i s Gaussian by d e f i n i t i o n (Kuo[S], p . 153) . Let Z denote a

r . e . with t h i s Gaussian measure. I t i s e a s i l y shown t h a t f + r L f i s a q u a d r a t i c f u n c t i o n a l d e r i v e d from t h e bounded b i l i n e a r form

( f , g)-+E(f (Z)g(Z)) on E* X E X , which g e n e r a t e s t h e covar iance

o p e r a t o r R : E*+E o f Z , d e f i n e d by ( ~ f , g) = E(F(Z)g(Z)) ,

f , g E E* ( c f . Vakhania [ 9 ] , 54 .3 .3) . I t fo l lows t h a t t h e charac-

t e r i s t i c f u n c t i o n a l o f Z i s ~ ( f ) = exp - & ( R E , f ), f E E*.q.e.d.

Remark 2.1

Condit ion ( i ) o f Theorem 2 . 3 i s e q u i v a l e n t t o and could be 2 P r e p l a c e d by t h e weak Lindeberg c o n d i t i o n IIX ) I I ( lJXnklJ 7 E)+ 0, 2k nk

f o r each c 0. I f t h e sequence (Zk[(Xnk(( )neN is uniformly i n t e -

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 6: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS 233

g r a b l e t h e n ( i ) i m p l i e s t h e u s u a l Lindeberg c o n d i t i o n 2 ZkE llXnk]J I (I/Xnkll > c ) --+ 0 , which i n t u r n i m p l i e s c o n d i t i o n ( i i ) o f

Theorem 2 . 2 . Condi t ion ( i ) i s implied by t h e c l a s s i c a l Khinchin

c o n d i t i o n f o r a Gaussian l i m i t : k l p [//xnkll > E ] - 0, a s n -r m , f o r each c 1 0.

I f t h e Banach space E p o s s e s s e s a Schauder b a s i s ( e . ) . and 1 1 c N '

i f x = t z l a i e i , x E E , t h e n Pmx =Em a . e and gx = (1-P ) x = i=l 1 1 m

~ ~ m + l a i e i d e f i n e s , f o r m e N , t h e p a r t i a l sum and r e s i d u a l oper-

a t o r s on E . I t f o l l o w s from t h e proof o f Lemma 4, p . 259, of Dun-

f o r d and Schwartz [2] t h a t a s e t K C E i s c o n d i t i o n a l l y compact i f

and o n l y i f

( i ) supxcKllxll < m and

( i i ) t o every c > O t h e r e i s m E N such t h a t supxeK E 11%;1l~ .

Lemma 2 . 3

Let (X ) be a sequence o f r . e . ' s i n a Banach space E with a n Schauder b a s i s , and s a t i s f y i n g l i m supn~l lxJ / '1 = 0,

a+ 00 [ I l X J > Oc I 1 p <m. Then (Xn) is t i g h t i f and o n l y i f l i m r u p n ~ l l e ~ ~ ~ P = 0.

in+ w

Proof -- Let (X,) be t i g h t , and E > 0 given . Choose a such t h a t P

llXnll I [ 1/xnl[ > a ] < I, f o r a l l n. Next, f i n d K compact, such t h a t

PIXn 4 K] 4 I, f o r a l l n . Then E l1xn/l '1 [Xn+Kl g $P [xn& K] + 2 aP

E 1 1 ~ ~ 1 ~ 1 f + 5 = L . NOW s i n c e K i s compact, we have m-+ l i m m

P S ~ P ~ E I / Q ~ X ~ I I ~ = ~ ~ ~ s ~ P ~ E ~ / \ X ~ I [ ~ ~ ~ ~ ~ I I L C ~ U P ~ E I I X J ~ ~ I [ X ~ + ~ ( ] < Cr,

where C = supnllQmllP. S i n c e r i s a r b i t r a r y , m- l i m m S U ~ ~ E / / P X J I ~ = 0 .

Conversely, suppose ~ ~ m m s u p n E ~ I ~ X ~ / = 0. Let & > 0 be g iven .

To every k € N , t h e r e i s mk 6 N such t h a t EQ,, xn/lP < k-12-k-l k

c , f o r a l l n , and m & m k . By P [ I I x , I I ~ > ~ I ~ E

< ~ - ' E I I X ~ I I ~ ~ lixnll > a1 f i n d a such t h a t P [ ~ ~ x ~ I J ~ > a] < -, f o r a l l n. 2 1

Now l e t A. = (x € E : IIx\IP L a ) - and Ak = (X € E : l lekxl lP g k- ) , 00

k r N . Put K = nkSOAk. I t is e a s i l y checked t h a t K i s condi- * C

t i o n a l l y compact. Now P [Xn $ K] = P [Xn G U k = O ~ k ] =

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 7: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

234 DAFFER

3. LIMIT THEOREMS FOR WEIGHTED SUMS OF EXCHANGEABLE RANDOM ELEMENTS

We begin by recalling the concept of exchangeability.

Definition 3.1

A sequence (Xn) of r.e.'s in the Banach space E is said to be

exchangeable if, to each fixed n e N, the joint distribution of

(Xql, ..., Xgn) is the same as that of (XI, ..., Xn) where (~1, ..., 7m) is an arbitrary permutation of the integers

(Iy ...y n).

Clearly, exchangeable T.e.'s are identically distributed.

Definition 3.2

A Banach space E is said to be uniformly p-smooth, 1 L_ p 2, if ?(t) = 0(tP) as t 0, where the modulus of smoothness ~ ( t ) is

defined by ~ ( t ) = sup(:(lx + y l l + qllx - yII - 1 : llxll = 1,

l l ~ l l = t) . The spaces lP and L~ are uniformly 2-smooth for 2 g p <a,

and uniformly p-smooth for 1 < p L - 2.

The specific property of uniformly p-smooth spaces that will

be needed below is the following: If E is isomorphic to a uni- formly p-smooth Banach space, then E 4 C ~ ~ = ~ E I I X ~ I ~ for every martingale difference sequence XI, . . . , Xn with E l l ~ ~ l l ~ ~ m (Hoffman-JBrgensen and Pisier [3]) .

Given an array (X ), k = 1, ..., kn, of r.e.'s in E, define nk the o-algebras (9 ), k = 0, 1, . . . , kn, n € N, as follows:

n,k

tkn X .), k = 0, 1, ..., k - 1, and <,k=g(Xnl' "" 'nk9 j=k+l nl n 7mJkn = Q(Xnl, . . . , X ) . For a r.e. X, we abbreviate

nkn

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 8: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS 235

Theorem 3 .1

Let (Xnk), k = 1, ..., kn, be an a r r a y of row-wise exchange-

a b l e random elements i n a uniformly 2-smooth Banach space E . Let

= 1, ..., kn, be an a r r a y o f r e a l c o n s t a n t s , Suppose: 2

(llXnlll )ncN i s bounded s t o c h a s t i c a l l y by a random v a r i a b l e 2

X, f o r which ElXi 4 W ; - 1

maxkllxnkl l = op ( l a n k / ) ) ; 2 s"~nxkank' ;

2 2 t k a i k f 2 ( x n k ) i b f , Of 2 0 non-random, f o r each f c E* ;

(Xnl)nsN i s a t i g h t sequence.

(X ))ZZ, a cen t e r ed Gaussian random Then Zkank('nk - En, k - l nk element i n E .

Proof

(X 1 ) . Then (Ynk), k = 1, ..., Let 'nk = ank(Xnk - En, k-1 nk kn'

i s a ma r t i nga l e d i f f e r e n c e a r r a y i n E. We v e r i f y t h e hypotheses of

Theorem 2 . 2 f o r (Ynk) . For ( i ) we have

~ ~ l l ~ n k l l

f lank\ IIXnkll + mtx lankl (IE,, k-1 (Xnk)ll

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 9: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

by (1) and (3) .

For ( i v ) , w r i t e

Now

A s f o r t h e o t h e r term, we have

DAFFER

2 2 4 [ F a n k f ( x n k ~ ~ 'I2 [%ankf * (Xnk) 11 I/z -O p

by (5) and what was j u s t proved. Hence

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 10: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS

non-random, by (5).

Finally we need to show (iii) of Theorem 2.2, that

(t Y ) is tight. To accomplish this, let h : E+C[O,l] be an k nk isometric isomorphism embedding E into the universal space

Z

C [O, 11 . Then E = h(E) c C [O, 11 is a Banach space, subspace of

C[0,1], and is uniformly 2-smooth. Now h(Y ) = a (2 - .. nk nk nk (2 ) ) , where Xnk = h(Xnk) , and (h(Ynk)) , k = 1, . . . , k is En,k-l nk ,+

a martingale difference array in E. Then

.., (linearity of Qm and uniform 2-smoothness of E)

Now since (X ) is tight in E, (cl) is tight in C [0, l] . From (1) nl

] = 0'

this

and Lemma 2.3 with p = 2, then yields 2

i m E = 0. m+ca

2 This with (3), then yields lim = 0. But

m-+ w implies

lim sup~~l~J\~ll = 0, m-+m n k

and ( T ~ < ~ ) ~ ~ ~ is uniformly integrable. (Indeed,

S U P E I I ~ < ~ U n

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 11: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

DAFFER

impl ies uniform i n t e g r a b i l i t y ) . Thus Lemma 2.3, appl ied t o ,. ( Z , G ~ , ) ~ ~ , , t h i s t ime f o r p = 1, y i e l d s t h a t (JlkYnk)nN is t i g h t

i n C[0,1] . Hence (fkYnk)neN i s t i g h t i n E . Theorem 2.2 now y i e l d s W

(X ))-+Z, a cen te red Gaussian r . e . i n kYnk = z k a n k C X n k - En, k-1 nk

E . q.e.d.

Remark 3.1

Theorem 3.2 could be genera l ized by no t r e q u i r i n g ( 3 ) - - t h a t

( ~ ~ a : ~ ) be bounded--but a t t h e expense o f in t roduc ing some 2

r a t h e r unpleasant hypotheses. Write An = xlkank. For i n s t ance , i f

( 6 ) i s changed t o read Ef (Xnl)f (Xn2) = O ( A ~ ) ; and i f t h e fo l lowing 2 - 1

two hypotheses a r e in t roduced; f i r s t : E I ( X , ~ J I = O(An ) , and

second: f o r an isometry h : E--.C[O,l] we have

limm_msupnAnE~~~h(~n1)I~2 = 0 ; then t h e conc lus ion of Theorem 3.1

w i l l fol low, mu ta t i s mutandis, without (3). Indeed, i f it i s

assumed t h a t l i m m _ _supnZk (ankl P ~ l l ~ m h ( ~ n l ) ~ I P = 0 , 1 5 p < 2, t h e

requirement o f uniform 2-smoothness can be dropped t o t h a t of un i -

form p-smoothness, and t h e conc lus ion of t h e theorem w i l l fo l low.

Here, however, we must have l i m 1 la l P = Cb i f t h e l i m i t i s t o - be non-degenerate; s u p n r k l ankl < , 1 4 p ( 2 , f o r ce s

l i m n + _ x t a i k = 0, which e n t a i l s convergence i n p r o b a b i l i t y of 2 2

(X ) ) t o t h e zero element ( p [Xank f (Xnk) Ekank('nk - En, k-1 nk

> CI 4 C ' A , I I ~ ~ ~ ~ E 1x1 2- 0).

Remark 3.2

sup A = supn zka:k < m can be deduced from (5) and t h e n n

fol lowing hypothes i s : There i s an c > 0 and f E E* such t h a t

l iminf n-co P [ f ( X n l ) > c ] > 0. (See Lemma 3.2 below.) Thus, f o r a l l

s u f f i c i e n t l y l a r g e n , p o s i t i v e p r o b a b i l i t y f o r Xnl l i e s i n a h a l f -

space an E removed from t h e o r i g i n .

Remark 3 .3

Note t h a t (4) o f Theorem 3 .1 i s implied by maxk la nkl = o(log-1'2kn).

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 12: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS

Remark 3.4

I t i s no t d i f f i c u l t t o d e r i v e from ( I ) , (3) and (5) o f 2 2

Theorem 3 .1 t h a t t h e sequence ( z k a n k f (Xnk)) ncN i s un i formly

i n t e g r a b l e f o r each f E EX, and hence (5) could be r e p l a c e d by

a p p a r e n t l y s t r o n g e r c o n d i t i o n l i m 2

9'" E l ~ ~ a ~ ~ f ~ ( ~ ~ ~ ) - c r ~ 1 = 0, which i n t u r n i m p l i e s l i m AnEf (Xnl) = o f E E*.

n-m E'

239

t h e

The fo l lowing easy lemma w i l l be needed. Note t h a t it f a i l s

i f X i s no t non-negat ive. 1

Lemma 3.2

Let (X ) be i d e n t i c a l l y d i s t r i b u t e d random v a r i a b l e s wi th n X1 $ 0 and X L. 0, a s s . , and (ank) , k = I , . . .k be non-negat ive

1 - n P numbers. I f ~ : : l a n k ~ k - c , a c o n s t a n t , t h e n sup x k a n k < m .

Proof - C C

Let Sn = x kankxk and Sn = 'nl [ s n 4 c + l ] Then P[Sn # S,] 3

0. Let ~ > O a n d 6 > 0 b e s u c h t h a t P [ X 1 > e ] > S . T h e n l + c 2

sC:,Z a E I I n - k nk [Sn<c+l] [Xk> E ] '

Now P[Sn c + 1 and Xk> c ] >

6 /2 f o r a l l s u f f i c i e n t l y l a r g e n and s o , t a k i n g e x p e c t a t i o n s , -1 -1

c 8 / 2 Z k a n k L_ 1 + c , o r r k a n k 4 E 6 (1 + c) f o r a l l s u f f i c i e n t l y

l a r g e n . q .e .d .

I t fo l lows d i r e c t l y from D e f i n i t i o n 3 . 1 t h a t E n , k - l (x nk I =

1 zk2 X . k = 1, . . . , kn, i f (Xn) i s an exchangeable k - k + 1 J k n J ' n sequence ( c f . Kingman [4] ) .

A sequence (Xn) o f r . e . ' s i n a Banach space E i s s a i d t o be

bounded s t o c h a s t i c a l l y by a random v a r i a b l e X ( i n R) i f

P[llXnll 1 X] P[X 1 X I , f o r every x 1 - 0 . ( C l e a r l y X 3. - 0 , a lmost

s u r e l y ) .

For a sequence (X ) o f r . e . I s l e t = n

kn X I > x k , j k + l x j , k = 0, 1. . . - , kn -1, <,kn =

6(X1, . . > Xkn) .

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 13: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

DAFFER

Corollary 3.3

Let (X ) be a sequence of exchangeable random elements taking n values in a uniformly 2-smooth Banach space E, and satisfying (i)

~ 1 1 ~ ~ 1 1 ~ ~ m and (ii,) Ef (X )f (X2) = 0, for every f c E*. Let 1 (ank), k = 1, ..., kn, be an array of real constants satisfying:

7 L

(a) lim ank = 0; n+m2k kn - k + 1

- 1 (b) maxkllxkl = op((max lank/ ) ;

2 2 (c) xkankf (xt)Pd, 2 ). 0 non-random, for every f e E* f f -

(X ))2~, a centered Gaussian random Then Zkank('k - k element in E.

Proof

(I), (2), (4) , ( S ) , (6) and (7) of Theorem 3.1 are trivially

satisfied. If X = 0, a.s., the result is trivial. If 1 PIX1 # 0] > 0, then for some c z0 and 6 > O and some f € E* we have

L P[f (XI) > E ] > 6. Applying Lemma 3.2 to (f (X,)) , (c) then yields 2

supnxkank<m, which is (3) of Theorem 3.1. The results follows.

Remark 3.5

Referring to remark 3.4, it is seen that the hypotheses of 2 2 Corollary 3.3 imply that Ef (X )lim A = c f , SO that

2 2 1 n 4 w n Qf = const. Ef (XI), f E E*.

Remark 3.6

We do not see how to obtain a nondegenerate limit result for

the case of uniform p-smoothness, 1 f p < 2. The only place in the proof of Theorem 3.1 where uniform

2-smoothness is needed is in obtaining the tightness of the

sequence ( tkYnk)neN- If it is replaced with uniform p-smoothness

for 1 L - p < 2, and a stronger condition is placed on the array (ank) of weights, the result is a weak law of large numbers. We

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 14: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL L I M I T THEOREMS

s t a t e t h i s f o r t h e s p e c i a l c a se of a sequence (X,) o f exchange-

a b l e r . e . ! s .

Co ro l l a ry 3.4

Let (X ) be a sequence o f exchangeable r . e . ' s i n a uniformly k 2 p-smooth Banach space E, with 1 4 p < 2 , s a t i s f y i n g E / I x ~ I < . Let ( a ), k = 1, ..., k be an a r r a y of r e a l cons t an t s s a t i s f y i n g nk (b) of Co ro l l a ry 3 .3 and (d) supn xk lank 1 P< m . Then

Proof

A s before , cons ider t h e mar t inga le d i f f e r e n c e a r r a y (Ynk)

where Y nk = a nk(Xk - En,k-l(Xk)). Af t e r embedding E i s o m e t r i c a l l y

i n t o C[0,1] , a s i n t h e proof of Theorem 3.1, we have

E1~s~k<k~~P & zk tankI PEl~s'nk - sEn, k- ('nk)ll 4

c 2 P - 1 ~ k lank I * [ E I I % ' ~ ~ I I ~ + E1~sEn,k-l ('nk)(I '1 4

2 1 a E + 0 a s m , us ing (d) . As i n t h e proof of

Theorem 3.1, (t 7 ) i s uniformly i n t e g r a b l e , and an a p p l i c a t i o n k nk

of Lemma 2.3, f o r p = 1, y i e l d s t i g h t n e s s of (x Y ) . k nk 2 We need t o show t h a t (c ) o f Co ro l l a ry 3.3. ho ld s with cf = 0,

f o r each f E E*, and t h a t both ( i i ) and (a ) o f Co ro l l a ry 3 .3 a r e

unnecessary. We f i r s t no t e t h a t i f X1 = 0, a . s . , t h e r e s u l t i s

t r i v i a l , and we assume PIX1 # 01 > 0. But t h i s t oge the r with (b)

o f Co ro l l a ry 3.3 e a s i l y impl ies t h a t limn+oomaxklankl = 0. Now,

2 f o r e > 0 and f E E*, P [ t k f (Ynk) > E] ( E ~ ~ ~ E ~ ~ Y ~ ~ ) 4

2 2 2 i1 c 2 &I j2 X ~ ~ ; ~ [ E I I X ~ I I + ~ p ~ , ~ - ~ (xk)ll I 6 2 2

4 g1 f EllXll/ ,Yka;X g const.maxklankl 2-P,T+a:k4 O , by ( d ) , s i n c e

2 max / a [ 2 p - ~ , a s n . ~ h u s t ~ f C ( , ~ ) ~ O a s n - + m , f o r every

k nk

f E E X , and hypotheses ( i i ) and (a ) o f Co ro l l a ry 3 .3 were not

needed t o a r r i v e a t t h i s . Checking t h i s proof o f Theorem 3.1, it

i s seen t h a t t h e only p l ace t h a t (a) and ( i i ) o f Co ro l l a ry 3 .3 were

needed was i n v e r i f y i n g ( i v ) o f Theorem 2.2. But t h i s has been

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 15: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

242 DAFFER

v e r i f i e d above without t hese condit ions. Corollary 3 . 3 now y ie lds

the conclusion. q.e.d.

By s t rengthening t h e hypotheses on t h e a r r ay (a ), t h e cen- nk t e r i n g a t condi t ional expectat ions can be el iminated.

Corollary 3.5

Let (X ) be a sequence of exchangeable r . e . ' s i n a uniformly n 2 2-smooth Banach space E , such t h a t E(!x# < 00 . Let (a ), k = nk

1, ..., kn, be an a r r ay of r e a l cons tants s a t i s f y i n g (b) of

Corollary 3 . 3 and

(el c l a n k t -. 0 a s n + rn

m e n Sn = t k a n k x k 2 ~ , a s n--t m.

The freedom of choice of t h e r - a lgeb ras Cmk can be exploi ted

t o weaken t h e hypotheses on t h e a r r ay (ank).

Theorem 3 . 6

Let (X ) be a sequence of exchangeable r . e . ' s i n a uniformly n 2 2-smooth Banach space E , and s a t i s f y i n g ( i ) ~11x41 ( 00 and

( i i ) Ef(Xl)f(X2) = 0, f o r each f c E*. Let k 1 f o r a l l n, and n n

let (a ), k = 1, ..., 1 be an a r r ay of r e a l constants. Let (b) nk n

and (c) o f Corollary 3 . 3 be s a t i s f i e d together with

- 1 ( f ) ,,l2m&knln = 0

(X - E n , k - l ( ~ k ) ) ~ ~ , where Z i s a centered Gaussian Then x::lank k r . e . i n E.

Proof

The proof follows t h a t of Corollary 3 . 3 . (a) of t h a t

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 16: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

CENTRAL LIMIT THEOREMS 243

Coro l l a ry i s (3) o f Theorem 3.1, which was used s o l e l y t o v e r i f y

( i v ) o f Theorem 2.2. But ( i v ) can be v e r i f i e d u s ing ( f ) a s

fo l lows (Y - nk - ank(Xk - En, k-1 (xk) 1) :

L - c mex a Z [ I + log ln - nk

I f PIX1 = 01 = 1, t h e theorem i s t r i v i a l . I f we assume t h a t 2 PIX1 # 0] 7 0, then it fo l lows from Lemma 3.2 t h a t suPnxkank< 03,

from which a f o r t i o r i sup max a 2 < Oo. Now ( f ) y i e l d s n k nk 2 2

l i m n + & l t k a n k f (En, k-l (Xk)) I = 0 , from whish

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013

Page 17: Central limit theorems for weighted sums of exchangeable random elements in banach spaces               (1)

244 DAFFER

2 2 (X ) ) = 0. As in the proof of Theorem 3.1, p - limn t a n k f (En,k-l k

we obtain p - lim 2 (X ) ) = 0. Hence (iv) of n zkankf ('kIf ('n,k-1 k Theorem 2.8 follows, completing the proof. q.e.d.

Remark 3.7

Some applications call for treating the weights (ank) them-

selves as random variables (Taylor [$], Chapter VI). This poses no

problem if the weights (a ) are independent of the random elements nk

( X ) Hypotheses such as (3) of Theorem 3.1 are replaced by

expressions like sup$ ~a~ < M , involving expectations. k nk

REFERENCES

[ I ] Billingsley, P. (1968). Convergence of Probability Measures. Wiley, New York.

[2] Dunford, N. and Schwartz, J. (1958). Linear Operators I. Interscience, New York.

[3] Hoffmann-~drgensen, J. and Pisier, G. (1976). The law of large numbers and the central limit theorem in Banach spaces. Annals of Prob. 4, 587-599.

[4] Kingman, J.F.C. (1978). Uses of exchangeability. Annals of Prob. 6, 183-197. - -

[5] Kuo, H. H. (1975). Gaussian Measures in Banach Spaces. Lecture Notes in Mathematics Vol. 463. Springer-verlag, Berlin.

[6] McLeish, D. L. (1974). Dependent central limit theorems and invariance principles. Annals of Prob. 2, 620-628.

[7] Scalora, F. S. (1961). Abstract martingale convergence theorems. Pacific Jour. Math. 2, 347-374.

[8] Taylor, R. L. (1978). Stochastic Convergence of Weighted Sums of Random Elements in Linear Spaces. Lecture Notes in Mathematics Vol. 672. Springer-verlag, Berlin.

[9] Vakhania, N. N. (1981). Probability Distributions on Linear Spaces. Elsevier-North Holland.

[lo] Weber, N. C. (1980). A martingale approach to central limit theorems for exchangeable random variables. Jour. of Applied Prob. 17, 662-673. - -

Dow

nloa

ded

by [

Mos

kow

Sta

te U

niv

Bib

liote

] at

14:

20 0

2 O

ctob

er 2

013