732 CHAPTER 26. INDEPENDENCE

Theorem 26.8.3 Let {Xk}∞

k=1 be independent random vectors having values in aseparable real Hilbert space and suppose E (|Xk|)< ∞ for each k and E (Xk) = 0. Sup-

pose also that ∑∞j=1 E

(∣∣X j∣∣2)< ∞.Then ∑

∞j=1X j converges a.e.

Proof: Let ε > 0 be given. By Kolmogorov’s inequality, Theorem 26.8.2, it followsthat for p≤ m < n

P

([max

m≤k≤n

∣∣∣∣∣ k

∑j=mX j

∣∣∣∣∣≥ ε

])≤ 1

ε2

n

∑j=p

E(∣∣X j

∣∣2)≤ 1ε2

∑j=p

E(∣∣X j

∣∣2) .Therefore, letting n→ ∞ it follows that for all m,n such that p≤ m≤ n

P

([max

p≤m≤n

∣∣∣∣∣ n

∑j=mX j

∣∣∣∣∣≥ ε

])≤ 1

ε2

∑j=p

E(∣∣X j

∣∣2) .It follows from the assumption ∑

∞j=1 E

(∣∣X j∣∣2) < ∞ there exists a sequence, {pn} such

that if m≥ pn

P

([max

k≥m≥pn

∣∣∣∣∣ k

∑j=mX j

∣∣∣∣∣≥ 2−n

])≤ 2−n.

By the Borel Cantelli lemma, Lemma 26.1.2, there is a set of measure 0, N such that forω /∈ N, ω is in only finitely many of the sets,[

maxk≥m≥pn

∣∣∣∣∣ k

∑j=mX j

∣∣∣∣∣≥ 2−n

]

and so for ω /∈ N, it follows that for large enough n,[max

k≥m≥pn

∣∣∣∣∣ k

∑j=mX j (ω)

∣∣∣∣∣< 2−n

].

However, this says the partial sums{

∑kj=1X j (ω)

}∞

k=1are a Cauchy sequence. Therefore,

they converge. ■With this amazing result, there is a simple proof of the strong law of large numbers but

first is an elementary lemma. In the following lemma, sk and a j could have values in anynormed linear space.

Lemma 26.8.4 Suppose sk→ s. Then limn→∞1n ∑

nk=1 sk = s. Also if ∑

∞j=1

a jj converges,

then limn→∞1n ∑

nj=1 a j = 0.

Proof: Consider the first part. Since sk → s, it follows there is some constant, C suchthat |sk|<C for all k and |s|<C also. Choose K so large that if k ≥ K, then for n > K,

|s− sk|< ε/2.∣∣∣∣∣s− 1n

n

∑k=1

sk

∣∣∣∣∣≤ 1n

n

∑k=1|sk− s|= 1

n

K

∑k=1|sk− s|+ 1

n

n

∑k=K|sk− s|

732 CHAPTER 26. INDEPENDENCETheorem 26.8.3 Let {Xx};_, be independent random vectors having values in aseparable real Hilbert space and suppose E (|X|) < °° for each k and E(X,) =0. Sup-pose also that Yi, E (|x,”) < 00.Then Yi X j converges a.e.Proof: Let € > 0 be given. By Kolmogorov’s inequality, Theorem 26.8.2, it followsthat for p<m<nJeo be(ar)<a be (ot),° (Therefore, letting n — © it follows that for all m,n such that p<m<n1 << 2P( | sm, Ex ‘|) <a he (|X).= J=pPIt follows from the assumption ))j"_, E (|x il’) < oo there exists a sequence, {p,} suchyx;that if m > pyr( >>) <0By the Borel Cantelli lemma, Lemma 26.1.2, there is a set of measure 0, N such that foro € N, @ is in only finitely many of the sets,maxm<k<ny x5j=mkmaxk>m=pnj=mkmax yk>m=Pn j=mXxX;and so for @ ¢ N, it follows that for large enough n,< <2" .y Xj (However, this says the partial sums {rk Xj (@) \ ' are a Cauchy sequence. Therefore,maxk>m=>pn jamthey converge.With this amazing result, there is a simple proof of the strong law of large numbers butfirst is an elementary lemma. In the following lemma, s; and a; could have values in anynormed linear space.Lemma 26.8.4 Suppose sy > s. Then lity 5.04 Y}_ | 5k = 8. Also if a 4 converges,then limy +0 4 Yi, a; = 0.Proof: Consider the first part. Since s, — s, it follows there is some constant, C suchthat |s,| < C for all k and |s| < C also. Choose K so large that if k > K, then forn > K,|s — sz] << €/2.<b Ly ‘y—) |sx-sl=—) |se—s|+— )) |sx—s|Mr] Nn f=] K=Kadssi