ADVANCED PROBABILITY HOMEWORK 2

GROUP 4
(1120240065)
(2120240153)
(2120240161)
(1120240072)
(2120240160)

2.2.2 The L2 weak law generalizes immediately to certain dependent sequences. Suppose EXn = 0 and
EXn Xm ≤ r(n − m) for m ≤ n (no absolute value on the left-hand side!) with r(k) → 0 as k → ∞. Show
that (X1 + . . . + Xn ) /n → 0 in probability.

Proof. We assume that

!" r(0) is finite. Let Sn = X1 + · · · + Xn , since E[Xn ] = 0, we have E[Xn ] = 0.
# $
Sn 2
We first compute E n :

%& '2 ( & ' n n

Sn Sn 1 1 ))
E = Var = Var(S n ) = E[Xi Xj ].
n n n2 n2
i=1 j=1

Expanding the double sum, we get:

n n
* n n−1 n−2
+
1 )) 1 ) ) )
E[Xi Xj ] ≤ 2 E[Xi2 ] +2 E[Xi Xi+1 ] + 2 E[Xi Xi+2 ] + · · · + 2E[X1 Xn ] .
n2 n
i=1 j=1 i=1 i=1 i=1

Now, since r(k) → 0 as k → ∞, for given ε > 0, there exists N such that for n ≥ N , |r(n)| ≤ ε. Moreover,
since E[Xi2 ] = r(0) < ∞, by the Cauchy-Schwarz Inequality, we have E[|Xi Xj |] ≤ (E[Xi2 ])1/2 (E[Xj2 ])1/2 ≤

Date: 2024.10.18.
1
2 GROUP 4 (1120240065) (2120240153) (2120240161) (1120240072) (2120240160)

r(0) < ∞. Then, for n ≥ N + 1, we have

%& ' ( * n n−1 n−N
Sn 2 1 ) ) )+1
E ≤ 2 E[Xi2 ] + 2 E[|Xi Xi+1 |] + · · · + 2 E[|Xi Xi+N −1 |]
n n
i=1 i=1 i=1
n−N
+
)
+2 E[Xi+N Xi ] + · · · + 2E[X1 Xn ]
i=1
* n−N
1 )
≤ 2 nr(0) + 2(n − 1)r(0) + · · · + 2(n − N + 1)r(0) + 2 r(N )
n
i=1
+
)−1
n−N
+2 r(N + 1) + · · · + 2r(n − 1)
i=1
* +
1
n−N
) )−1
n−N
≤ 2 2N nr(0) + 2 |r(N )| + 2 |r(N + 1)| + · · · + 2|r(n − 1)|
n
i=1 i=1
* n−N
+
1 )
≤ 2 2N nr(0) + 2 iε
n
i=1
* n
+
1 )
≤ 2 2N nr(0) + 2 iε
n
i=1
2N r(0) n(n + 1)
≤ + ε.
n n2
!" # 2 $
Since ε > 0 is arbitrary, we conclude that E Snn → 0, which means that Sn /n → 0 in L2 . Therefore,
Sn /n → 0 in probability. !

" #
2.2.4 Let X1 , X2 , . . . be i.i.d. with P Xi = (−1)k k = C/k 2 log k for k ≥ 2 where C is chosen to make
the sum of the probabilities = 1. Show that E |Xi | = ∞, but there is a finite constant µ so that Sn /n → µ
in probability.

Proof.
∞
) ) 1 ∞
C k
E[Xi ] = (−1) 2 =C
k log k k log k
k=2 k=2
Suppose that f (x) = 1
x log x for x ∈ [2, ∞). f (x) is a strictly monotone decreasing and non-negative function.
Since
, ∞ , ∞
-∞
1 -
f (x) dx = dx = log log x-- = ∞,
2 2 x log x 2
.∞
we have that 1
k=2 k log k = ∞. Thus,

∞
) 1
E[Xi ] = C = ∞.
k log k
k=2
 ADVANCED PROBABILITY HOMEWORK 2 3

On the other hand,

∞
) ∞
) C
nP(|Xi | > n) = n P(|Xi | = k) = n
k 2 log k
k=n+1 k=n+1
∞ ∞ & '
Cn ) 1 Cn ) 1 1 C
≤ = − = → 0 as n → ∞
log n k(k − 1) log n k−1 k log n
k=n+1 k=n+1
and

n
) n
) (−1)k
k C
µn = E[Xi I{|Xi |≤n} ] = (−1) 2 =C
k log k k log k
k=2 k=2
.∞ .∞ (−1)k .∞ (−1)k
Since k=2 (−1) is a Leibniz series,
k
k=2 k log k < ∞. We denote by µ =C k=2 k log k , and µn → µ as
n → ∞. By WLLN, n − µn → 0 in probability. For ϵ > 0, P(| Snn − µn | > ϵ)
Sn
→ 1 as n → ∞ and |µn − µ| ≤ ϵ
2
when n is sufficiently large, so we have
&- - ' &- - '
- Sn - - Sn - ϵ ϵ
1 ≥ P -- − µ-- > ϵ ≥ P -- − µn -- < , |µn − µ| < → 1 as n → ∞.
n n 2 2
Hence, Sn
n − µ → 0 in probability.
!

2.2.5 Let X1 , X2 , . . . be i.i.d with P (Xi > x) = e/x log x for x ≥ e. Show that E|Xi | = ∞, but there is a
sequence of constants µn → ∞ so that Sn /n − µn → 0 in probability.

Proof. Check the conditions for Theorem 2.2.12 to hold:

Condition(1): we need to verify that xP (|Xi | > x) → 0 as x → ∞. According to the problem statement,
we obtain
xP (|Xi | > x) = xP (Xi > x) = e/ log x → 0 as x → ∞
Condition(2): we need to verify that µn = E(X1 1(|X1 |≤n) ) are exist. Since
, n , n
e
µn = E(X1 1(|X1 |≤n) ) = P (Xi > x)dx = dx = e log(log n)
e e x log x

Therefore, by Theorem 2.2.12, we have

Sn
− µn → 0
n
Moreover, we can show that E|Xi | = ∞ by the following inequality:
, ∞ , n
E|Xi | = EXi = P (Xi > x)dx > P (Xi > x)dx = µn
e e

Since µn → ∞ as n → ∞, so we obtain that E|Xi | = ∞. !

4 GROUP 4 (1120240065) (2120240153) (2120240161) (1120240072) (2120240160)

2.3.5 Dominated convergence. Suppose Xn → X in probability and (a) |Xn | ≤ Y with EY < ∞ or
(b) there is a continuous function g with g(x) > 0 for large x with |x|/g(x) → 0 as |x| → ∞ so that
Eg (Xn ) ≤ C < ∞ for all n . Show that EXn → EX .

Proof (a). By Theorem 2.3.2 in text, for every subsequence Xn(m) , there is a further subsequence Xn(mk ) that
convergence almost surely to X. Since |Xn(mk ) | < Y for all n(mk ), and EY < ∞ so Y is integrable. Therefore
by DCT, EXn(mk ) → EX. That means that every subsequence (EXn(m) ) of (EXn ) has a sub-subsequence
EXn(mk ) converging to EX. Therefore, EXn → EX.
(b) By Theorem 2.3.2 in text, for every subsequence Xn(m) , there is a further subsequence Xn(mk ) that
convergence almost surely to X. Let h(x) = x. By Theorem 1.6.8 in text, we can get EXn(mk ) → EX.
That means that every subsequence (EXn(m) ) of (EXn ) has a sub-subsequence EXn(mk ) converging to EX.
Therefore, EXn → EX.
!

2.3.8 Let An be a sequence of independent events with P (An ) < 1 for all n. Show that P (∪An ) = 1 implies
.
n P (An ) = ∞ and hence P (An i.o. ) = 1.

Proof. Since P (∪n An ) = 1, we can gain that P (∩n Acn ) = P {(∪n An )c } = 1 − P (∪n An ) = 0 according to
De Morgan’s laws.
Then, since An ’s are a sequence of independent events, Acn ’s are also a sequence of independent events. So
we have

* ∞
+ ∞ ∞
/ 0 0
P Acn = P (Acn ) = (1 − P (An )) = 0
n=1 n=1 n=1
1m−1
Consider that P (An ) < 1 for all n, we have n=1 (1 − P (An )) > 0 for m ≥ 2. Then, we have for all
m ∈ N,

* ∞
+
/
P Acn =0
n=m
2∞ 3∞
Note that { lim sup n→∞ Acn } = m=1 n=m An ,
c

* ∞ /
∞
+
4
P ( lim sup n→∞ Acn ) = P Acn
m=1 n=m
∞
* ∞ +
) /
≤ P Acn =0
m=1 n=m
.∞
Then, P (An i.o. ) = 1. Finally, according to Theorem 2.3.1 , we have n=1 P (An ) = ∞. !
 ADVANCED PROBABILITY HOMEWORK 2 5

2.3.11 Let X1 , X2 , . . . be independent with P (Xn = 1) = pn and P (Xn = 0) = 1 − pn . Show that (i)
.
Xn → 0 in probability if and only if pn → 0, and (ii) Xn → 0 a.s. if and only if pn < ∞.

Proof. (i) Necessity: Assume Xn → 0 in probability. Then for any 0 < ϵ < 1, we have

P (|Xn − 0| ≥ ϵ) → 0.

Since Xn takes values only in {0, 1}, we get

P (|Xn − 0| ≥ ϵ) = P (Xn = 1) = pn .

Thus, if Xn → 0 in probability, then pn → 0.

Sufficiency: If pn → 0, then for any 0 < ϵ < 1,

P (|Xn − 0| ≥ ϵ) = pn → 0.

This implies that Xn → 0 in probability. Thus, we conclude that:

Xn → 0 in probability if and only if pn → 0.

.∞
(ii) Sufficiency: If n=1 pn < ∞, then by the Borel-Cantelli lemma, for the independent events {Xn = 1}, we
have
∞
)
P (Xn = 1) < ∞ =⇒ P (Xn = 1 i.o.) = 0.
n=1
Thus, Xn → 0 a.s.
Necessity: If Xn → 0 a.s., then
P (Xn = 1 i.o.) = 0.
. .∞
By the second Borel-Cantelli lemma, if Xn are independent and ∞ n=1 P (Xn = 1) = n=1 pn = ∞, then
.∞
P (Xn = 1 i.o.) = 1. Therefore, if Xn → 0 a.s., it must be that n=1 pn < ∞.
Thus, we conclude that:
∞
)
Xn → 0 a.s. if and only if pn < ∞.
n=1
!

2.3.20 Show that if Xn is the outcome of the nth play of the St. Petersburg game (Example 2.2.16) then
lim supn→∞ Xn / (n log2 n) = ∞ a.s. and hence the same result holds for Sn . This shows that the conver-
gence Sn / (n log2 n) → 1 in probability proved in Section 2.2 does not occur a.s.

Proof. For fixed M > 0, Let Tn = ⌊log2 (M · n log2 n)⌋ + 1. Then, for n ≥ 2,

∞
) 1 1 1 1
P (X1 ≥ M · n log2 n) = = Tn −1 = ⌊log (M ·n log n)⌋ ≥
2j 2 2 2 2 M · n log2 n
j=T n

Then, according to Exercise 2.2.4 , we have

6 GROUP 4 (1120240065) (2120240153) (2120240161) (1120240072) (2120240160)

∞
) & ' ∞
)
X1 1
P ≥M ≥ =∞
n log2 n M · n log2 n
n=2 n=2
5
Note that Xn ’s are independent, according to Borel-Cantelli Lemma 2, we get P Xn
n log2 n ≥ M i.o. ) =
1. !

2.4.2 Let X0 = (1, 0) and define Xn ∈ R2 inductively by declaring that Xn+1 is chosen at random from the
ball of radius |Xn | centered at the origin, i.e., Xn+1 / |Xn | is uniformly distributed on the ball of radius 1 and
independent of X1 , . . . , Xn . Prove that n−1 log |Xn | → c a.s. and compute c.

Proof. Let Un = Xn
|Xn−1 | for n = 1, 2, . . .. Then Un is a uniform distribution on the unit ball and U1 , U2 , . . . are
i.i.d.. Then for 0 ≤ x ≤ 1, using polar coordinates,
, 2π , x
1
P(|Un | ≤ x) = r dr dθ = x2 .
0 0 π
Thus, the density function of Un is 2x for 0 ≤ x ≤ 1, and
, 1
1
E[log |Un |] = 2x log x dx = − .
0 2
It is obvious that E[| log |Un ||] = 1/2 < ∞, then by the SLLN, we have
n
) 1
n−1 log |Xn | = n−1 log |Uk | → E[log |Un |] = − a.s.
2
k=1

2.5.2 The converse of Theorem 2.5.12 is much easier. Let p > 0. If Sn /n1/p → 0 a.s. then E|X1 |p < ∞.

Proof. Assume that E|X1 |p = ∞, then

, ∞
p
E|X1 | = P (|X1 |p > y)dy
0
∞
)
≤ P (|X1 |p > n)
n=0
)∞
= P (|X1 | > n1/p )
n=0

Since E|X1 | = ∞ and X1 , X2 , . . . are i.i.d., it follows from the second Borel-Cantelli lemma that P (|Xn | >
p

n1/p i.o.) = 1, which contradicts Sn /n1/p → 0 a.s. !

 ADVANCED PROBABILITY HOMEWORK 2 7

2.5.3Let X1 , X2 , . . . be i.i.d. standard normals. Show that for any t

∞
) sin(nπt)
Xn · converges a.s.
n
n=1
We will see this series again at the end of Section 8.1 [REF].

sin(nπt) sin(nπt) .
proof. Let Yn = Xn · n . EYn = · E(Xn ) = 0, so ∞
n n=1 E(Yn ) = 0 < ∞. V ar(Yn ) =
(sin(nπt))2 .∞ .∞ 1 π2
n2
· V ar(Xn )≤ n2 .
1
ar(Yn ) ≤
n=1 V n=1 n2 = 6 < ∞. Hence by Kolmogorov’s two-series
.∞ .∞ sin(nπt)
theorem, n=1 Yn converges a.s. i.e n=1 Xn · n converges a.s..
!