Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

PS: Advanced Probability Theory

Sheet 1 Solutions
lucas.teyssier@univie.ac.at

Exercise 1 (L.T.)

a) By definition of the expectation and rewriting the integral,


Z Z Z ∞
E [X] = X(ω)P(dω) = 1X(ω)≥x dxP(dω). (1)
Ω Ω 0

Then, applying Fubini-Tonelli to the nonnegative function 1X(·)≥·· (i.e. (ω, x) 7→ 1X(ω)≥x )1
gives Z ∞Z Z ∞
E [X] = 1X(ω)≥x P(dω)dx = P(X ≥ x)dx. (2)
0 Ω 0
If X has support in N, we deduce that
∞ Z
X n ∞
X
E [X] = P(X ≥ x)dx = P(X ≥ n). (3)
n=1 n−1 n=1

b) Let us first observe that for all n ≥ 0,

P(N ≥ n + 1) = P(X1 = max Xi ), (4)


1≤i≤n

and consequently, by symmetry, that


1
P(N ≥ n + 1) ≥ . (5)
n
We deduce, using Question a), that

X 1
E [N ] ≥ = ∞. (6)
n=1
n

Remark. You can think about whether the inequality in Equation (5) is always strict.
1 To be more rigorous, you could write on which space this function lives but this is not necessary.

1
Exercise 2 (Moritz Dober)
P
Take Ω = I, F = P(Ω), P = i∈I pi · δi (where δi is the Dirac measure on {i}) and X : Ω → R
given by X(i) = xi .

Then we have P(X = xi ) P = P({i}) = pi . Its law is unique since for every such X it holds
that P(X ∈ {xi | i ∈ I}) = i∈I pi = 1 and hence for A ∈ B(R),
X X
P(X ∈ A) = P(X ∈ A ∩ {xi | i ∈ I}) = P(X ∈ A ∩ {xi }) = pi · δxi (A).
i∈I i∈I

Exercise 3 (Daniel Bäumer)

a) It holds that P (− log U ≤ t) = λ(− log U ≤ t) = λ(U ≥ e−t ) = 1 − e−t . Hence,


F (t) = 1 − e−t , which is the distribution function of an exponentially distributed ran-
dom variable with parameter 1.

b) In the proof of the Lebesgue-Stieltjes theorem, we applied the cadlag-inverse of a given


function with the required properties to a uniformly distributed random variable. The
construction in (a) is almost the same, except that the inverse of F is − log(1 − U ), which
has the same distribution.

√ √
c) Computing the distribution function first, P (U 2 ≤ t) = P (U ≤ t) = t. This is seen to
Rt 1 1
have a density w.r.t. the Lebesgue measure, since FU 2 (t) = 0 2√s ds, hence fU 2 (t) = 2√ t
and the law of U 2 (which is uniquely determined by the distribution function, as seen in
Exercise 4) is Z
1
µU 2 (B) = √ dx.
B 2 x

Exercise 4, first solution (Moritz Dober)

Claim: open intervals are in σ(A)

For a, b, c, d ∈ R with a < b, c < d we have

(a, b] = (−∞, b] ∩ (a, ∞) = (−∞, b] ∩ (−∞, a]C ∈ σ(A)

and therefore
[ d−c
(c, d) = (c, d − ] ∈ σ(A).
n≥1 | {z n }
∈σ(A)

Every open set O ∈ O can be represented as union of open intervals with rational end-
points, hence as a countable union of open intervals. This implies O ⊆ σ(A) and hence
B(R) = σ(O) ⊆ σ(A).

+ n1 ).
T
Clearly A ⊆ σ(O) (and thus σ(A) ⊆ σ(O)) since (−∞, a] = n≥1 (−∞, a

2
A is a π-system since (−∞, a] ∩ (−∞, b] = (−∞, min{a, b}] ∈ A.

Now, let X, Y be two random variables with the same distribution functions FX = FY . Then
µX |A = µY |A since

µX ((−∞, a]) = FX (a) = FY (a) = µY ((−∞, a]).

Set D := {A ∈ B(R) | µX (A) = µY (A)} which is a Dynkin system with A ⊆ D.

By Dynkin’s lemma the σ-algebra generated by A is the same as the Dynkin system generated
by A, i.e. σ(A) = δ(A). Therefore we conclude

B(R) = σ(O) = σ(A) = δ(A) ⊆ δ(D) = D,

i.e. µX ≡ µY on B(R).

Exercise 4, second solution (Daniel Bäumer)

In order to prove that A generates the Borel σ-algebra on R, we need to show that A con-
tains all open sets in R. By taking complements and intersections, S it is clear that σ(A) contains

all intervals of the form (a, b]; for a sequence bn ↑ b, we also have n=1 (a, bn ] = (a, b), so σ(A)
contains all open intervals. The fact that every open set in R is a countable union of open intervals
finishes the proof.
The intersection of any two sets in A is equal to the smaller of the two, so A is intersection stable.
Now, looking at two random variables X1 , X2 with the same distribution and laws µ1 , µ2 , it
immediately follows that µ1 ((−∞, a]) = µ2 ((−∞, a]) for all a ∈ R. This just means µ1 = µ2 on
A.
Since D := {A ∈ B(R) : µ1 (A) = µ2 (A)} is a Dynkin system that contains the π-system A, we
can apply Dynkin’s lemma: D ⊃ σ(A) = B(R), so µ1 = µ2 .

You might also like