59.7. KOLMOGOROV’S INEQUALITY 1877

=∫

|X| |Y|dP≤(∫

|X|2 dP)1/2(∫

|Y|2 dP)1/2

< ∞

and so by Fubini’s theorem,∫Ω

(X,Y)dP =∫

∑k=1

(X,ek)(Y,ek)dP =∞

∑k=1

∫Ω

(X,ek)(Y,ek)dP

=∞

∑k=1

∫Ω

(X,ek)dP∫

(Y,ek)dP =∞

∑k=1

(∫Ω

XdP,ek

)(∫Ω

YdP,ek

)dP

=

(∫Ω

XdP,∫

YdP)

Now here is Kolmogorov’s inequality.

Theorem 59.7.2 Suppose {Xk}nk=1 are independent with E (|Xk|) < ∞, E (Xk) = 0. Then

for any ε > 0,

P

([max

1≤k≤n

∣∣∣∣∣ k

∑j=1

X j

∣∣∣∣∣≥ ε

])≤ 1

ε2

n

∑j=1

E(|Xk|2

).

Proof: Let

A =

[max

1≤k≤n

∣∣∣∣∣ k

∑j=1

X j

∣∣∣∣∣≥ ε

]Now let A1 ≡ [|X1| ≥ ε] and if A1, · · · ,Am have been chosen,

Am+1 ≡

[∣∣∣∣∣m+1

∑j=1

X j

∣∣∣∣∣≥ ε

]∩

m⋂r=1

[∣∣∣∣∣ r

∑j=1

X j

∣∣∣∣∣< ε

]

Thus the Ak partition A and ω ∈ Ak means∣∣∣∣∣ k

∑j=1

X j

∣∣∣∣∣≥ ε

but this did not happen for∣∣∣∑r

j=1 X j

∣∣∣ for any r < k. Note also that Ak ∈ σ (X1, · · · ,Xk) .

Then from algebra, ∣∣∣∣∣ n

∑j=1

X j

∣∣∣∣∣2

=

(k

∑i=1

Xi +n

∑j=k+1

X j,k

∑i=1

Xi +n

∑j=k+1

X j

)

=

∣∣∣∣∣ k

∑j=1

X j

∣∣∣∣∣2

+ ∑i≤k, j>k

(Xi,X j)+ ∑i≤k, j>k

(X j,Xi)+ ∑i>k, j>k

(X j,Xi)

59.7. KOLMOGOROV’S INEQUALITY 18771/2 1/2= [ixiiviars ([ixPar) (fivPar) <=Q Q Qand so by Fubini’s theorem,[svar= LY ey vegar=¥ [eg Weyer-¥ |, (KeyAP [(W.e,)aP= Yo ( /,xare) ( [, varex) dP= ( I XaP, L va?)Now here is Kolmogorov’s inequality.Theorem 59.7.2 Suppose {X,};_, are independent with E (|X|) < 0, E (X¢) = 0. Thenfor any € > 0,k 1 2 5P x|>e|)<— E (IX ).max, | Xi] 28] ] See De (Proof: LetkA= X;| > eEmax, |Now let A; = [|X1| > €] and if A,,--- ,A», have been chosen,m+1LxAnti =yx=1 ||j=1>of)Thus the A, partition A and w € A, meanskyxj=l| Zébut this did not happen for ba x; for any r <k. Note also that Ay € 0 (X1,--- , Xx).Then from algebra,we y x. .Xi+ yx :j=k+1 i=1 j=k+1+ ¥ (X%,X)+ VY (x).x)+ YO (XX)ix<k,j>k i<k,j>k i>k,j>k