205A Homework #1, due Tuesday 9 September.

1. [Bill. 2.4] Let Fn be classes of subsets of S. Suppose each Fn is a field,

and Fn Fn+1 for n = 1, 2, . . .. Define F =
n=1 Fn . Show that F is a
field. Give an example to show that, if each Fn is a -field, then F need not
be a -field.
2. [Bill. 2.5(b)] Given a non-empty collection A of sets, we defined F(A)
as the intersection of all fields containing A. Show that F(A) is the class
ni
of sets of the form m
i=1 j=1 Aij , where for each i and j either Ai,j A or
c
i
Aij A, and where the m sets nj=1
Aij , 1 i m are disjoint.
3. [Bill. 2.8] Suppose B (A), for some collection A of subsets. Show
there exists a countable subcollection AB of A such that B (AB ).
4. Show that the Borel -field on Rd is the smallest -field that makes all
continuous functions f : Rd R measurable.
5. [Durr. 1.3.5] A function f : Rd R is lower semicontinuous (l.s.c.) if
lim inf yx f (y) f (x) for all x. A function is upper semicontinuous (u.s.c.)
if lim supyx f (y) f (x) for all x. Show that, if f is l.s.c. or u.s.c., then f
is measurable.

205A Homework #2, due Tuesday 16 September.

1. [similar Bill. 2.15] Let B be the Borel subsets of R. For B B define
(B) = 1
=0

if (0, ) B for some > 0

if not

(a) Show that is not finitely additive on B.

(b) Show that is finitely additive but not countably additive on the
field B0 of finite disjoint unions of intervals (a, b].
2. Show that, in the definition of a probability measure on a measurable
space (S, S), we may replace countably additive by finitely additive,
and satisfies
if An then (An ) 0 .
3. [similar Durr. A.1.1] Give an example of a measurable space (S, S), a
collection A and probability measures and such that
(i) (A) = (A) for all A A
(ii) S = (A)
(iii) 6= .
Note: this can be done with S = {1, 2, 3, 4}
4. [similar Durr. Lemma A.2.1] Let be a probability measure on (S, S),
where S = (F) for a field F. Show that for each B S and > 0 there
exists A F such that (BA) < .
5. Let g : [0, 1] R be integrable w.r.t. Lebesgue measure. Let >
0.
R Show that there exists a continuous function f : [0, 1] R such that
|f (x) g(x)| dx .

205A Homework #3, due Tuesday 23 September.

1. Use the monotone convergence theorem to prove the following.
(i) If Xn 0, Xn X a.s. and EXn < for some n then EXn EX.
(ii) If E|X| < then E|X|1(|X|>n) 0 as n .
(iii) If E|X1 | < and Xn X a.s. then either EXn EX < or else
EXn and E|X| = .
(iv) If X takes values in the non-negative integers then
EX =

P (X n).

n=1

P
2. (i) For a counting r.v. X = ni=1 1Ai , give a formula for the variance of
X in terms of the probabilities P (Ai ) and P (Ai Aj ), i 6= j.
(ii) If k balls are put at random into n boxes, what is the variance of
X = number of empty boxes?
3. (i) Suppose EX = 0 and var(X) = 2 < . Prove
P (X a)

2
, a > 0.
2 + a2

(ii) Suppose X 0 and EX 2 < . Prove

P (X > 0)

(EX)2
.
EX 2

4. Chebyshevs other inequality.

Let f : R R and g : R R be bounded and increasing functions.
Prove that, for any r.v. X,
E(f (X)g(X)) (Ef (X))(Eg(X)).
[In other words, f (X) and g(X) are positively correlated. This is intuitively
obvious, but a little tricky to prove. Hint: consider an independent copy
Y of X. For this and the next question you may need the product rule for
expectations of independent r.v.s]
5. Let X have Poisson() distribution and let Y have Poisson(2) distribution.

(i) Prove P (X Y ) exp((3 8)) if X and Y are independent.

(ii) Find constants A < , c > 0, not depending on , such that, without
assuming independence, P (X Y ) A exp(c).
3

205A Homework #4, due Tuesday 30 September.

1. Monte
Carlo integration [cf. Durr. 2.2.3] Let f : [0, 1] R be such
R1
that 0 f 2 (x) dx < . Let (Ui ) be i.i.d. Uniform(0, 1). Let
Dn := n

n
X

Z
f (Ui )

f (x) dx.
0

i=1

(i) Use Chebyshevs inequality to bound P (|Dn | > ).

(ii) Show this bound remains true if the (Ui ) are only pairwise independent.
2. Let X 0 and Y 0 be independent r.v.s with densities f and g.
Calculate the densities of XY and of X/Y .
Note: this is just to remind you of undergraduate results.
3. [Durr. 2.2.2.] Let (Xi ) be r.v.s with EXi = 0 and EXi Xj r(j i), 1
i j < , where
P r(n) is a deterministic sequence with r(n) 0 as n .
Prove that n1 ni=1 Xi 0 in probability.
4. [Durr. 2.3.11] Suppose events An satisfy P (An ) 0 and

P (Acn An+1 ) < .

n=1

Prove that
P (An occurs infinitely often ) = 0.
5. (a) Let Z have standard Normal distribution. Show
P (Z > z) z 1 (2)1/2 exp(z 2 /2) as z .
(b) Let (Z1 , Z2 , . . .) be independent with standard Normal distribution.
Find constants cn such that
lim sup Zn /cn = 1 a.s.
n

205A Homework #5, due Tuesday 7 October.

1. Let (Xn ) be i.i.d. with E|X1 | < . Let Mn = max(X1 , . . . , Xn ). Prove
that n1 Mn 0 a.s.
2. [Durr. 2.3.2] Let 0 X1 X2 . . . be r.v.s such that EXn an and
var(Xn ) Bn , where 0 < a, B < and 0 < < 2 < . Prove that
n Xn a a.s.
3. Prove that the following are equivalent.
(i) Xn X in probability.
(ii) There exist n 0 such that P (|Xn X| > n ) n .
(iii) E min(|Xn X|, 1) 0.
4. Durr. exercise 2.4.4 (An Investment Problem).
5. Prove the deterministic lemma we used in the proof of the GlivenkoCantelli Theorem.
Lemma. If F1 , F2 , . . . , F are distribution functions and
(i) Fn (x) F (x) for each rational x
(ii) Fn (x) F (x) and Fn (x) F (x) for each atom x of F
then supx |Fn (x) F (x)| 0.

205A Homework #6, due Tuesday 14 October.

1. [Durr. 2.5.9] Let (Xi ) be independent, Sn =
Prove that
P (Sn > 2a)

Pn

i=1 Xi ,

Sn = maxin |Si |.

P (|Sn | > a)
, a > 0.
minjn P (|Sn Sj | a)

[Hint. If |Sj | > 2a and |Sn Sj | a then |Sn | > a.]

2. [Durr. 2.5.10 and 11] In the setting of the previous question, prove
(i) if limn Sn exists in probability then the limit exists a.s.
(ii) if the (Xi ) are identically distributed and if n1 Sn 0 in probability
then n1 maxmn Sm 0 in probability.
3. [cf. Durr 2.2.8] Let (Xi ) be i.i.d. taking values in {1, 1, 3, 7, 15, . . .},
such that
1
, k1
P (X1 = 2k 1) =
k(k + 1)2k
(which implicitly specifies P (X1 = 1)).
(a) Show EX1 = 0.
(b) Show that for all < 1,


n
P Sn <
1.
log2 n
Comment. This is sometimes described as an unfair, fair game. It
shows that the conclusions of the SLLN and the recurrence of sums theorem cant be strengthened much.

205A Homework #7, due Tuesday 21 October.

1. Suppose S and T are stopping times. Are the following necessarily
stopping times? Give proof or counter-example.
(a) min(S, T )
(b) max(S, T )
(c) S + T .
Pn
2. Let (Xi ) be i.i.d. with EXi2 < . Let Sn =
i=1 Xi . Let T be a
bounded stopping time. Is it true in general that
var(ST ) = (var(X1 ))(ET )?
If not, is it true in the special case EX1 = 0?
3. Let (XP
i ) be a sequence of random variables, and let T be its tail -field.
Let Sn = ni=1 Xi . Let bn be constants. Which of the following events
must be in T ? Give proof or counter-example.
(i) {Xn 0}
(ii) {Sn converges }
(iii) {Xn > bn infinitely often }
(iv) {S
n > bn infinitely often }
(v) {

Pn

i=1

Xi2

Sn

0}.

Pn

4. Let Sn = i=1 Xi , where (Xi ) are i.i.d. with exponential(1) distribution.

Use the large deviation theorem to get explicit limits for
n1 log P (n1 Sn a), a > 1 and n1 log P (n1 Sn a), a < 1.
5. Oriented first passage percolation. Consider the lattice quadrant
{(i, j) : i, j 0} with directed edges (i, j) (i + 1, j) and (i, j) (i, j + 1).
Associate to each edge e an exponential(1) r.v. Xe , independent for different
edges.
For each directed path of length d started at (0, 0), let S =
P
edges e in path Xe . Let Hd be the minimum of S over all such paths
of length d. It can be shown that d1 Hd c a.s., for some constant c. Give
explicit upper and lower bounds on c.
[Hint: use result of previous question for lower bound.]

205A Homework #8, due Tuesday 4 November.

[Theorem 7 and Corollary 8 refer to the notes linked from the week 8
row of the schedule.]
1. Suppose probability measures satisfy   . Show that
d
d d
=

.
d
d
d
2. In the setting of Theorem 7 [hard part], where S2 is nice, show that Q is
unique in the following sense. If Q is another conditional probability kernel
for , then
1 {x : Q (x, B) = Q(x, B) for all B S2 } = 1.
3. Let F be a distribution function. Let c > 0. Find a simple formula for
Z
(F (x + c) F (x)) dx.

4. In the proof of Corollary 8 we used the inverse distribution function

f (x, u) = inf{y : u Q(x, (, y])}
associated with the kernel Q. Show that f is product measurable.
5. Given a triple (X1 , X2 , X3 ), we can define 3 p.m.s 12 , 13 , 23 on R2 by
ij is the distribution of (Xi , Xj ).

(1)

These p.m.s satisfy a consistency condition:

the marginal distribution 1 obtained from 12 must coincide
with the marginal obtained from 13 , and similarly for 2 and 3 .

(2)

Give an example to show that the converse is false. That is, give an example
of 12 , 13 , 23 satisfying (2) but for which there does not exist a triple
(X1 , X2 , X3 ) satisfying (1).

205A Homework #9, due Thursday 13 November

1. Let X, Y be random variables, and suppose Y is measurable with respect
to some sub--field G. Let (, ) be a regular conditional distribution for
X given G. Prove that, for bounded measurable h,
Z
E(h(X, Y )|G)() = h(x, Y ())(, dx) a.s.
2. For i = 1, 2 let Xi be a r.v. defined on (, F, P ) taking values in (Si , Si ).
Let G be a sub--field of F. Prove that assertions (a),(b) and (c) below
are equivalent. When these assertions hold, we say call X1 and X2 are
conditionally independent given G.
(a) P (X1 A1 , X2 A2 |G) = P (X1 A2 |G)P (X2 A2 |G) for all
Ai Si .
(b) E(h1 (X1 )h2 (X2 )|G) = E(h1 (X1 )|G) E(h2 (X2 )|G) for all bounded
measurable hi : Si R.
(c) E(h1 (X1 )|G, X2 ) = E(h1 (X1 )|G) for all bounded measurable h1 :
S1 R.
3. Suppose X and Y are conditionally independent given Z. Suppose X
and Z are conditionally independent given F, where F (Z). Prove that
X and Y are conditionally independent given F.
4. Let (Xn ) and (Yn ) be submartingales w.r.t. (Fn ). Show that (Xn + Yn )
and that (max(Xn , Yn )) are also submartingales w.r.t. (Fn ).
5. Give an example where
(Xn ) is a submartingale w.r.t. (Fn )
(Yn ) is a submartingale w.r.t. (Gn )
(Xn + Yn ) is not a submartingale w.r.t. any filtration.

205A Homework #10, due Tuesday 18 November.

P
Ei = 0 and var i <
1. Let Sn = ni=1
Pni , where the (i ) are independent,
2
2
2
. Let sn =
i=1 var i . So we know that (Sn sn ) is a martingale.
Suppose also that |i | K for some constant K. Show that


2
P max |Sm | < x s2
x > 0.
n (K + x) ,
mn

2. Let (Xn ) be a martingale with X0 = 0 and EXn2 < . Using the fact
that (Xn + c)2 is a submartingale, show that


EXn2
P max Xm x 2
, x > 0.
mn
x + EXn2
3. Let (Xn ) and (Yn ) be martingales with E(Xn2 + Yn2 ) < . Show that
EXn Yn EX0 Y0 =

n
X

E(Xm Xm1 )(Ym Ym1 ).

m=1

4. Let (Xn , Fn ), n 0 be a positive submartingale with X0 = 0. Let Vn be

random variables such that
(i) Vn Fn1 , n 1
(ii) B V1 V2 . . . 0, for some constant B.
Prove that for > 0
P ( max Vj Xj > ) 1
1jn

n
X

E[Vj (Xj Xj1 )].

j=1

5. Prove Dubins inequality. If (Xn ) is a positive martingale then the

number U of upcrossings of [a, b] satisfies
P (U k) (a/b)k E min(X0 /a, 1).
[if you follow sketch in Durrett then prove the quoted exercise]

10

205A Homework #11, due Tuesday 25 November.

In each question, there is some fixed filtration (Fn ) with respect to which
martingales are defined.
1. Let (Xn ) be a submartingale such that supn Xn < a.s. and E supn (Xn
Xn1 )+ < . Show that Xn converges a.s.
2. For a sequence (An ) of events, show that

P (An | n1
m=1 Am ) = implies P (m=1 Am ) = 1.

n=2

3. Let (Xn )Pbe a martingale and write n = Xn Xn1 , Suppose that

2
2
bm and
m=1 bm Em < . Prove that Xn /bn 0 a.s.
4. Let (Xn ) be a martingale with supn E|Xn | < . Show that there
is a representation Xn = Yn Zn where (Yn ) and (Zn ) are non-negative
martingales such that supn EYn < and supn EZn < .
5. Let (Xn ) be adapted to (Fn ) with 0 Xn 1. Let , > 0 with
+ = 1. Suppose X0 = x0 and
P (Xn+1 = + Xn |Fn ) = Xn , P (Xn+1 = Xn |Fn ) = 1 Xn .
Show that Xn X a.s., where P (X = 1) = x0 and P (X = 0) = 1x0 .
6. Suppose Fn F and Yn Y in L1 . Show that E(Yn |Fn )
E(Y |F ) in L1 .
7. Let Sn be the total assets of an insurance company at the end of year n.
Suppose that in year n the company receives premiums of c and pays claims
totaling n , where n are independent with Normal(, 2 ) distribution, where
0 < < c. The company is ruined if its assets fall to 0 or below. Show
P (ruin) exp(2(c )S0 / 2 ).

11