Assignment 2

Abhinav Pradeep
March 25, 2024

1 Question 1
LOOK AFTER INTEGRALITY CONDITIONS. LOOK AFTER CONDITIONS FOR FUBINI.
Prove, using Fubini’s Theorem 1.67, that for any positive random variable X, it holds that

X : (Ω, H, P) → (R+ , B + , Leb)

Z ∞
p
EX = dx p xp−1 P (X > x)
0
Z ∞
EX p = p xp−1 P (X > x) dx
0
Consider that:

EX p := PX p
That is:
Z
p
EX = X p P(dω)
Ω
Z Z X
EX p = pxp−1 dx P(dω)
Ω 0
Z Z ∞
EX p = pxp−1 1{X>x} dx P(dω)
Ω 0
By Fubini’s Theorem,
Z ∞ Z
EX =p
pxp−1 1{X>x} P(dω) dx
0 Ω
Z ∞ Z
EX =p
px p−1
1{X>x} P(dω) dx
0 Ω
Hence,
Z ∞
p
EX = pxp−1 P (X > x) dx
0

1
2 Question 2
{Xn } not necessarily independent.
(
1
n2 − 1 with P(Xn = n2 − 1) = n2
Xn :=
−1 with P(Xn = −1) = 1 − n12

1.
Prove ∀n, EXn = 0
Consider that for any n, as Xn is discrete,
X
EXn = xP(Xn = x)
x

EXn = n2 − 1 P(Xn = n2 − 1) + (−1P(Xn = −1))

 

1
2 1
EXn = n − 1 2 − 1 − 2
n n
   
1 1
EXn = 1 − 2 − 1 − 2
n n

EXn = 0

2.
Prove that
∞
1{Xn >−1} < ∞
X

n=1

almost surely

That is:
∞
!
1{Xn >−1} < ∞ = 1
X
P
n=1

By the Borel-Cantelli lemma,

∞ ∞
!
1{Xn >−1} < ∞ = 1
X X
P(Xn > −1) < ∞ =⇒ P
n=1 n=1

Hence, consider
∞
X
P(Xn > −1)
n=1

Xn is a discrete random variable taking values n2 − 1 and −1. Clearly, ∀n ≥ 1 n2 − 1 > −1. Hence,
 ∞
X ∞
X
P(Xn > −1) = P(Xn = n2 − 1)
n=1 n=1
∞
X 1
=
n=1
n2
Clearly
∞
X 1
2
<∞
n=1
n
Hence,
∞
X
P(Xn > −1) < ∞
n=1

and by the Borel-Cantelli lemma,

∞
!
1{Xn >−1} < ∞ = 1
X
P
n=1

2.1 3.
Sn


Prove that n
converges to −1 almost surely.

3 Question 3
X1 , X2 , X3 . . . i.i.d
With pdf:

1 − cos(x)
f (x) :=
πx2

1.
Show that characteristic function of each Xi is
(
1 − |r| if |r| ≤ 1
ψ(r) =
0 if |r| > 1

ψ(r) := EeirX
By Euler’s identity

ψ(r) = E cos(rX) + E sin(rX)i

Z Z
ψ(r) = cos(rX)dP + sin(rX)dP i
Ω Ω
 ∞ ∞
1 − cos(x) 1 − cos(x)
Z Z
ψ(r) = cos(rx) dx + sin(rx) dx i
−∞ πx2 −∞ πx2
Consider the function

1 − cos(x)
f (x) = sin(rx)
πx2
Consider:

f (−x)

1 − cos(−x)
sin(−rx)
π(−x)2
As cos(x) and x2 are even functions,

1 − cos(−x) 1 − cos(x)
sin(−rx) 2
= sin(−rx)
π(−x) πx2
As sin(x) is an odd function,
 
1 − cos(−x) 1 − cos(x)
sin(−rx) = − sin(rx)
π(−x)2 πx2
That is,

f (−x) = −f (x)
And therefore,
∞
1 − cos(x)
Z
sin(rx) dx = 0
−∞ πx2
Hence, the characteristic function simplifies to:
Z ∞
1 − cos(x)
ψ(r) = cos(rx) dx
−∞ πx2
This can be easily computationally verified to be the desired function (code provided under Code):
 1 ∞ 1 − cos(x)
Z
ψ(r) = cos(rx) dx
π −∞ x2
1 ∞ cos(rx) − cos(rx) cos(x)
Z
ψ(r) = dx
π −∞ x2
1 ∞ cos(rx) cos(rx) cos(x)
Z
ψ(r) = − dx
π −∞ x2 x2
Z ∞ Z ∞ 
1 cos(rx) cos(rx) cos(x)
ψ(r) = dx − dx
π −∞ x2 −∞ x2
2.
Sn d
Prove that n
→ Cauchy(x0 , γ)

Let:
Sn
Fn :=
n
Consider that:
n
Y
ψFn (r) = ψ Xi (r)
n
i=1
n
Y r
ψFn (r) = ψXi
i=1
n
(Q
n r
1− if nr ≤ 1
ψFn (r) = Qi=1
n
n
r
i=1 0 if n
>1
r
Take lim . As lim <1
n→∞ n→∞ n
n
Yr
lim ψFn (r) = lim 1−
n→∞ n→∞
i=1
n
 r n
lim ψFn (r) = lim 1 −
n→∞ n→∞ n

lim ψFn (r) = e|r|

n→∞

This is the characteristic function of Cauchy(0, 1). Hence, by the definition of Fn (r),
Sn d
→ Cauchy(0, 1)
n

3.
Prove that
3
lim P (Sn ≤ n) =
n→∞ 4
Rewrite as,
 
Sn 3
lim P ≤1 =
n→∞ n 4
d
By the definition of convergence in probability, if Xn → X,

lim P (Xn ≤ x) = F (x)

n→∞

Where F (x) is the c.d.f of X

By previous derivations,
Sn d
→ Cauchy(0, 1)
n
Hence, consider the c.d.f of Cauchy(0, 1),
Z x
1
P(Cauchy(0, 1) ≤ x) = 2
dt
0 π(1 + t )

By linearity of the integral,

Z x
1 1
P(Cauchy(0, 1) ≤ x) = dt
π −∞ (1 + t2 )
Integrating:
 
1 x
P(Cauchy(0, 1) ≤ x) = lim ( arctan (t) + C )
π a→ −∞ a
 
1
P(Cauchy(0, 1) ≤ x) = arctan (x) − lim arctan (a)
π a→ −∞

1 π
P(Cauchy(0, 1) ≤ x) = arctan (x) +
π 2
arctan (x) 1
P(Cauchy(0, 1) ≤ x) = +
π 2
Hence,
 
Sn
lim P ≤ 1 = P(Cauchy(0, 1) ≤ 1)
n→∞ n
 
Sn arctan (1) 1
lim P ≤1 = +
n→∞ n π 2
 
Sn π 1 1
lim P ≤1 = · +
n→∞ n 4 π 2
 
Sn 1 1
lim P ≤1 = +
n→∞ n 4 2
 
Sn 1+2
lim P ≤1 =
n→∞ n 4
 
Sn 3
lim P ≤1 =
n→∞ n 4
Which is the same as
3
lim P (Sn ≤ n) =
n→∞ 4
4 Code
function fr = fOfR(r)
integrand = @(x) cos(r.*x) .* (1 - cos(x)) ./ (pi .* x.^2);

% Handling the singularity at x = 0 by excluding a very small

% interval around 0
smallInterval = 1e-6;
integralPart1 = integral(integrand, -Inf, -smallInterval, ...
’RelTol’,1e-6, ’AbsTol’,1e-6);
integralPart2 = integral(integrand, smallInterval, Inf, ...
’RelTol’,1e-6, ’AbsTol’,1e-6);

% The total integral is the sum of both parts

fr = integralPart1 + integralPart2;
end

% Range of r values for plotting

rValues = linspace(-2, 2, 200);

% Calculating f(r) for each r using arrayfun

fValues = arrayfun(@fOfR, rValues);

% Plotting
figure;
plot(rValues, fValues);
title(’Graph of f(r)’);
xlabel(’r’);
ylabel(’f(r)’);
grid on;