Homework 4
November 9, 2023
1. Define Bn = ∪∞ m=n Am . Then, it follows that Bn+1 ⊂ Bn and ∩n Bn = limn→∞ Bn = lim sup An .
And since for k ≥ n, Ak ⊂ ∪∞m=n Am = Bn , we have supk≥n P (Ak ) ≤ P (Bn ). Hence,
P (lim sup An ) = lim P (Bn ) ≥ lim sup P (Ak ) = lim sup P (An ).
n→∞ n→∞ k≥n
The desired result is proved.
Similarly, to prove P (lim inf An ) ≤ lim inf P (An ). Define Cn = ∩∞
m=n Am . Then, it follows that
Cn ⊂ Cn+1 and ∪n Cn = limn→∞ Cn = lim inf An . And since for k ≥ n, Cn = ∩∞ m=n Am ⊂ Ak , we
have P (Cn ) ≤ inf k≥n P (Ak ). Hence,
P (lim inf An ) = lim P (Cn ) ≤ lim inf P (Ak ) = lim inf P (An ).
n→∞ n→∞ k≥n
The desired result is proved.
2. Since A1 , A2 , . . . are independent, Ac1 , Ac2 , . . . are independent. We have
Y Y
P (∪An ) = 1 − P ((∪An )c ) = 1 − P (∩Acn ) = 1 − P (Acn ) = 1 − (1 − P (An )).
Q
Hence, (1 − P (An )) = 1 − P (∪An ) = 0.
P P
We prove n P (An ) = ∞ by contradiction. Suppose n P (An ) < ∞, then P (An ) → 0. Then
1 1 1
for ϵ = , there exists an N , such that for n ≥ N , P (An ) < ϵ = . When 0 ≤ x < ,
2 2 2
log(1 − x) ≥ −2x, then we have
Y X X
(1 − P (An )) = exp( log(1 − P (An ))) ≥ exp(−2 P (An )) > 0.
n≥N n≥N n≥N
Since P (An ) < 1, tihs implies Y
(1 − P (An )) > 0.
P
which is a contraction. Hence, P (∪An ) = 1 implies n P (An ) = ∞, and by the second Borel-
Cantelli lemma, P (An i.o.) = 1.
3. (i) For 0 < ϵ < 1, we have P (|Xn | > ϵ) = pn . And for ϵ ≥ 1, P (|Xn | > ϵ) = 0. Hence, Xn → 0
in probability if and only if pn → 0.
(ii) Define An = {|Xn (ω)| > ϵ}. Since X1 , X2 . . . . are independent,
P A1 , A2 , . . .Pare independent.
When 0 < ϵ < 1, P (An ) = pn , P (An i.o.) = 0 if and onlyPif P (An ) = pn < ∞. When
ϵ > 0, P (An i.o.) = 0. Hence, Xn → 0 a.s. if and only if pn < ∞.
P
4. (i) Suppose n P (Xn > A) < ∞ for som A, by Borel-Cantelli lemma, P ({Xn > A i.o.}) = 0.
So for every ω ∈ {Xn > A i.o.}c , we have that there are only finite many n such that
Xn (ω) > A. Hence, sup Xn (ω) < ∞. Since P ({Xn > A i.o.}) = 0, we have sup Xn < ∞ a.s.
1
P
(ii) Suppose it is not true, then for ∀K, we have n P (Xn > K) = ∞. By the second Borel-
Cantelli lemma, we have P ({Xn ≥ K i.o.}) = 1, which means that lim sup Xn (ω) ≥ K for
ω ∈ ΩK where P (ΩK ) = 1. Then for ω ∈ ∩K ΩK , lim sup Xn (ω) = ∞. Since P (∩K ΩK ) =
limK→∞ P (ΩK ) = 1, lim sup Xn (ω) = ∞ a.s.
P
Hence, sup Xn < ∞ a.s. if and only if n P (Xn > A) < ∞ for some A.
5. If λn > 1, we can decompose PX n into Xn = Yn1 + Yn2 + · · · Ynm where Yni ’s are independent and
m
Yi ∼ P oisson(λni ), λni ≤ 1, i=1 λni = λn . Hence, without loss of generality, we can assume that
λn ≤ 1 for ∀n.
For δ > 0, Chebyshev’s inequality implies
P
Sn − ESn var(Sn ) λn 1
P > δ = P (|Sn − ESn | > δESn ) ≤ 2
= 2 Pn 2
= 2P →0
ESn (δESn ) δ ( n λn ) δ n λn
Sn
as n → ∞. Hence → 1 in probability.
ESn
To get almost sure converge, we have to take subsequences. Let nk = inf{n : ESn ≥ k 2 }. Let
Tk = Snk and note that EXm ≤ 1 implies k 2 ≤ ETk ≤ k 2 + 1. By Chebyshev’s inequality,
Tk − ETk var(Tk ) 1 1
P > δ = P (|Tk − ETk | > δETk ) ≤ 2
= 2 ≤ 2 2.
ETk (δETk ) δ ETk δ k
P Tk − ETk Tk − ETk
So k P >δ < ∞, and the Borel-Cantelli lemma imples P > δ i.o. =
ETk ETk
Tk
0. Since δ is arbitrary, it follows that → 1 a.s.
ETk
Sn Tk (ω)
To show → 1 a.s., pick an ω so that → 1 and observe that of nk ≤ n < nk+1 then
ESn ETk
Tk (ω) Sn (ω) Tk+1 (ω)
≤ ≤ .
ETk+1 ESn ETk
To show that the terms at the left and right end → 1, rewrite the inequalities as
ETk Tk (ω) Sn (ω) Tk+1 (ω) ETk+1
· ≤ ≤ · .
ETk+1 ETk ESn ETk ETk
Following from that k 2 ≤ ETk ≤ ETk+1 ≤ (k + 1)2 + 1, we have
ETk+1 (k + 1)2 + 1 2k + 2
≤ 2
=1+ → 1.
ETk k k2
Sn (ω) Sn
Hence, → 1, and therefore → 1 a.s.
ESn ESn
