778 CHAPTER 28. THE NORMAL DISTRIBUTION

Thus ψk (x) ∈ [0,1] ,ψk = 1 if dist(x,VC

)≥ 1/k, and ψk = 0 on VC. Since V is open,

it follows limk→∞ ψk (x) = XV (x). Choose k large enough that∫

ψkdλX ≥ λX (V )− ε.Then by Lemma 28.5.3,

lim infn→∞

λXn (V )≥ lim infn→∞

∫ψk (x)dλXn =

∫ψk (x)dλX ≥ λX (V )− ε

and since ε is arbitrary, liminfn→∞ λXn (V )≥ λX (V ). Now let λX (∂A) = 0 for A a Borelset.

λX (int(A)) ≤ lim infn→∞

λXn (int(A))≤ lim infn→∞

λXn (A)≤

lim supn→∞

λXn (A) ≤ lim supn→∞

λXn

(A)≤ λX

(A).

But λX (int(A)) = λX

(A)

by assumption and so limn→∞ λXn (A) = λX (A) as claimed.■

As an application of this theorem the following is a version of the central limit theoremin the situation in which the limit distribution is multivariate normal. It concerns a sequenceof random vectors, {Xk}∞

k=1, which are identically distributed, have finite mean m, and

satisfy E(|Xk|2

)< ∞.

Definition 28.5.6 ForX a random vector with values in Rp, let

FX (x)≡ P({

X j ≤ x j for each j = 1,2, ..., p})

.

A different proof of the central limit theorem is in [48].

Lemma 28.5.7 If all the zi and wi have absolute value no more than 1, then|∏n

i=1 zi−∏ni=1 wi| ≤ ∑

nk=1 |zk−wk| .

Proof: It is clearly true if n = 1. Suppose true for n. Then∣∣∣∣∣n+1

∏i=1

zi−n+1

∏i=1

wi

∣∣∣∣∣≤∣∣∣∣∣n+1

∏i=1

zi− zn+1

n

∏i=1

wi

∣∣∣∣∣+∣∣∣∣∣zn+1

n

∏i=1

wi−n+1

∏i=1

wi

∣∣∣∣∣≤

∣∣∣∣∣ n

∏i=1

zi−n

∏i=1

wi

∣∣∣∣∣+ |zn+1−wn+1|

∣∣∣∣∣ n

∏i=1

wi

∣∣∣∣∣≤ n+1

∑k=1|zk−wk| ■

Theorem 28.5.8 Let {Xk}∞

k=1 be random vectors satisfying E(|Xk|2

)< ∞ which

are independent and identically distributed with mean m and positive definite covarianceΣ≡ E

((X−m)(X−m)∗

). Let Zn ≡ ∑

nj=1

X j−m√n . Then for Z ∼ Np (0,Σ) ,

limn→∞ FZn (x) = FZ (x) for all x.

Proof: The characteristic function of Zn is given by

φZn(t) = E

(eit·∑n

j=1X j−m√

n

)=

n

∏j=1

E

(e

it·(

X j−m√n

)).

7718 CHAPTER 28. THE NORMAL DISTRIBUTIONThus y;, (x) € [0,1], wy, = 1 if dist (a,V°) > 1/k, and y, =0 on V©. Since V is open,it follows limg_,.o W, (w) = Zy (a). Choose k large enough that [ yw,dax >Ax (V)—e.Then by Lemma 28.5.3,lim inf Ax, (V) > lim inf | (a)dax, = | (w)ddx >Ax(V)—e€nooand since € is arbitrary, liminf,,.Ax, (V) > Ax (V). Now let Ax (0A) =0 for A a Borelset.Ax (int(A)) < lim inf Ax, (int(A)) <lim inf Ax, (A) <n—co n—o ~~Alim sup Ax,(A) <_ limsupAx, (A) <Ax (A).n—y00 n—-ooBut Ax (int(A)) = Ax (A) by assumption and so limy4..Ax,, (A) = Ax (A) as claimed.|As an application of this theorem the following is a version of the central limit theoremin the situation in which the limit distribution is multivariate normal. It concerns a sequenceof random vectors, {X kheet which are identically distributed, have finite mean m, andsatisfy E (xx!) <0,Definition 28.5.6 For X a random vector with values in RP , letFx (x) = P({X; <x; for each j = 1,2,...,p}) .A different proof of the central limit theorem is in [48].Lemma 28.5.7 [f all the z; and w; have absolute value no more than 1, thenTie 2 — We wil S Leer [ze — wel -Proof: It is clearly true if n = 1. Suppose true for n. Thenn+l n+lTe-[i=1i=1nt1 n[l« —Zn+1 I]i=1i=ln n+l< + |znui [wi — [] wii=l iadn n+l< + [Zn41 Wail |[]wil < ¥ kee —weli=l k=ln n[lz _ [ii=l i=lTheorem 28.5.8 Let {Xx}i_, be random vectors satisfying E (|xx1”) < co whichare independent and identically distributed with mean m and positive definite covarianceHSE ((X—m)(X—m)"). Let Z, = y, aE Then for Z ~N,(0,E),limp +00 Fz, (@) = Fz (x) for all x.Proof: The characteristic function of Z, is given byit-y” xXj-m n i-(~")painne( 58%) fy (*C))j=l