1 Lecture: Measure Theory (solutions)

1.
S1
(a) =)) Let fAn gn2N F be an increasing sequence and let A := n=1 An . Then
1
X
S
1 (1) U
1 (2)
(A) = An = (An An 1) = (An An 1)
n=1 n=1 n=1
1
X
(3) (4)
= ( (An ) (An 1 )) = lim (An ) (A0 ) = lim (An ) :
n!1 n!1
n=1
U
(1) denotes the disjoint union of sets. We de…ne A0 = ;.
(2) We use the -additivity of .
(3) We use the …nite additivity of . If (An ) (An 1 ) = 1 1 for some n 2 N, then
the sequence f (Ak )gk n would be identically 1. The monotony property of would
then imply that (A) = 1 and the result would be also true.
(4) Telescopic series.
S1
(=) Let fAn gn2N F be a disjoint increasing sequence of sets and let A := n=1 An .
Then
n
X 1
X
(1) S
1 S
n (2) S
n (3)
(A) = Ak = lim Ak = lim (Ak ) = (Ak ) :
n=1 k=1 n!1 k=1 n!1
k=1 k=1

(1) Standard
Sn trick to transform a countable union into a limit of an increasing sequence
of sets. f k=1 Ak gn2N is increasing.
(2) We apply the hypothesis.
(3) Because of the …nite additivity.

(b) T
Let fAn gn2N F be an decreasing sequence such that (A1 ) < 1 and let A :=
1
n=1 An . Then

T
1 (1) S
1
(A1 ) (A) = (A1 A) = A1 An = (A1 An )
n=1 n=1
(2)
= lim (A1 An ) = (A1 ) lim (An ) :
n!1 n!1

(1) We use that the complement of an intersection of sets is the union of the corresponding
complement sets.
(2) We apply (a) since fA1 An gn2N is an increasing sequence.
S1
(c) Let fAn gn2N F be a disjoint countable sequence of sets and let A := n=1 An . Observe
that
S
n S
n
(A) = Ak + A Ak ; 8n: (1)
k=1 k=1
Therefore
n
X 1
X
(1) S
n
(A) = lim Ak = lim (Ak ) = (Ak ) :
n!1 k=1 n!1
k=1 k=1

1
 T1 Sn
(1) We take limits: n=1 (A k=1 Ak ) = ; so the second term of (1) is zero because
by hypothesis.
1
4. Let A := A R:f (A) 2 F . It is easy to prove that A is a -algebra.
Since C A and A is a -algebra, (C) A, because the -algebra generated by C is the
smallest -algebra containing C. But B (R) = (C), therefore B (R) A and f is measurable.
T1 S1 c c
5. (a) fsupn2N fn < ag = n=1 ffn < ag = ( n=1 ffn < ag ) 2 F, because any ffn < ag 2 F.
We have used that the complement of an intersection of sets is the union of the complement
sets.
(b) inf n2N fn = supn2N ( fn ). The result follows from (a).
(c) lim fn = inf n2N supm n fm by de…nition. We apply (a) and (b).
n!1
(d) lim fn = supn2N inf m n fm . Again, the results follows from (a) and (b).
n!1
R
7. First of all, let Y := inf n2N fn . Since X Y d exists, it is a …nite number. The sequence
ffn Y gn2N is a sequence of positive functions so that
Z Z
limn!1 (fn Y ) d limn!1 (fn Y ) d
X X

by hypothesis. Since Y does not depend on n and using the linearity properties of the Lebesgue
integral, we have
Z Z Z Z
limn!1 fn d Yd limn!1 fn d Yd :
X X X X
R
Finally, we can subtract X
Yd .
8. Since jf j g and f is measurable, f 2 L1 (X; ). Since jfn fj 2g, Fatou’s Lemma applies
to the functions 2g jfn f j and yields
Z Z
2gd limn!1 (2g jfn f j) d
X X

because limn!1 (2g jfn f j) = 2g due to the pointwise convergence limn!1 fn = f . Now,
Z Z Z
limn!1 (2g jfn f j) d = 2gd + limn!1 jfn f j d
X X X
Z Z
= 2gd limn!1 jfn f j d :
X X
R
Since X
2gd is …nite, we may subtract it and obtain
Z
limn!1 jfn f j d 0
X
R
which clearly implies limn!1 jfn f j d = 0. Finally, since
Z Z Z
fn d fd jfn f j d ;
X X X

the result follows.

2
2 Lecture: Invariant Measures
1. Let U Rn be an open set and let F (t; x) a di¤erentiable function of I U , where I R is
an open neighbourhood of 0. Let us try to compute the derivative
Z
d
F (t; x) dx:
dt t=s 't (U )

Applying the change of variables formula, we have

Z Z
F (t; x) dx = 't (F (t; x) dx)
't (U ) U

where
't (F (t; x) dx) = F (t; 't (x)) 't (dx) = F (t; 't (x)) Jt (x) dx
and
@'t
Jt (x) = det (x)
@xi
is the Jacobian of the di¤eomorphism 't . Therefore,
Z Z
d d
F (t; x) dx = F (t; 't (x)) Jt (x) dx
dt t=s 't (U ) dt t=s U
Z
d
= F (t; 't (x)) Jt (x) dx: (2)
U dt t=s

Now, if div v = 0 then, by Liouville’s formula,

Z t
Jt (x) = exp div v ('u (x)) du = 1:
0

On the other hand, the volume of 't (U ) is

Z
1dx
't (U )

so F (t; x) 1 and the integrand of (2)reduces to

d
F (t; 't (x)) Jt (x) dx = 0:
dt t=s

d
R
Consequently, dt t=s 't (U ) 1dx = 0 and the volume of 't (U ) is constant.
1
3. Let X := (0; 1) and suppose that is an invariant Borel probability. Since T ((0; 1=2)) =
(0; 1) we conclude that
((0; 1=2)) = ( (0; 1)) = 1
and is supported on (0; 1=2). In general,

1 1 1
T 0; = 0; ; n 1;
2n+1 2n

3
and, by induction,
1 1
0; = 0; =1
2n+1 2
Therefore, the sequence
c
1 1
An := 0; = ;1
2n+1 2n+1
is increasing and such that (An ) = 0 for any n 2 N. Moreover,
S
An = (0; 1)
n 1

and since is -additive,

!
S
1= ((0; 1)) = An = lim (An ) = lim 0 = 0
n n!1 n!1

by Lecture 1 Problem 1 (a), which is a contradiction.

4
3 Lecture: Birkho¤’s ergodic theorem
S n
1. Let U 2 F such that (U ) > 0 and B := n 0 T (U ). Then
[
1 n
T (B) = T (U ) B
n 1

and T 1 (B) = (B) (T is measure preserving). Consequently, 1B is T -invariant a.s.,

which implies that 1B is constant a.s. because T is ergodic. In other words, since (B) > 0,
1B = 1X and (B) = 1.
3. Let f : X ! R be a G-measurable. Since f is G-measurable, there exists a sequence f n gn2N
of elementary functions such that n ! f as n ! 1. Suppose that
kn
X (n) (n) (n)
n = ci 1A(n) , where ci 2 R and Ai 2 G:
i
i=1

Then
kn
X kn
X kn
X
(n) (n) (n)
n T = ci 1A(n) T = ci 1T 1 (n)
Ai
= ci 1A(n) = n
i i
i=1 i=1 i=1

1 (n) (n)
because T (Ai ) = Ai for any n 2 N and any i = 1; :::; kn . That is, n are T -invariant.
However, n T !f T as n ! 1 (because n ! f ). Therefore,

n T ! f T
n!1
k
n ! f:
n!1

Since the limit is unique, this means that f T = f , so f is T -invariant.

4.

(a) By de…nition, given an arbitrary -algebra F, L2C (X; F; ) is a Hilbert space. In par-
ticular a Banach space: any Cauchy sequence ffn gn2N L2C (X; F; ) converges to an
2 2
element f 2 LC (X; F; ). This means that LC (X; F; ) is closed. Finally, take F = G.
(b) If H is a Hilbert space with Hermitian product h ; i and E H is a closed subspace, for
any y 2 H, p
x = min ky xk = min hy x; y xi
x2E x2E

exists and belongs to E. x is called the orthogonal projection of y onto E. Moreover,

x has is characterised by the following property:

hy x ; xi = 0 for any x 2 E. (3)

Indeed, if (3) holds for some x 2 E,

2
ky xk = hy x; y xi = h(y x ) + (x x); (y x ) + (x x)i
= hy x ; y x i + hy x ; x xi + hx x; y x i + hx x; x xi
2 2
= ky x k + kx x k

5
 2
where we have used that x x 2 E. Consequently, since kx x k 0,
2 2
ky xk ky x k for any x 2 E

and x is the orthogonal projection.

We can now prove that E [f jG] is the orthogonal projection of f onto L2C (X; G; ). Ob-
viously E [f jG] 2 L2C (X; G; ) and if g 2 L2C (X; G; ) is a characteristic function, g = 1A ,
A 2 G,
Z Z Z
hg; f E [f jG]i = g (f E [f jG]) d = gf d g E [f jG] d
Z Z
= fd E [f jG] d = 0
A A

by the de…nition of the conditional expectation. The result is also true for elementary
functions. For an arbitrary g 2 L2C (X; G; ), we take an increasing sequence f n gn2N of
elementary functions such that n ! g. Then, for any n 2 N,
Z
0= n (f E [f jG]) d :
R R
But n (f E [f jG]) d ! g (f E [f jG]) d because the integrand is dominated.
Therefore Z
0= g (f E [f jG]) d :

5.
R
(a) By induction. If , then there exists f 2 L1 (X; ) such that (A) = A
f d for
any A 2 F. On the other hand,
Z Z
T (A) = T 1 (A) = fd = P (f )d
T 1 (A) A

where we have applied the de…nition of the Frobenius-Perron operator. Therefore, T

is absolutely continuous with Radon-Nikodym derivative P (f ).
T = () P (f ) = f .
R R
=)) T = means that, for any A 2 F, A P (f ) d = A f d or, what is the same,
Z
(P (f ) f ) d = 0:
A

+
In particular, take A+ := supp (P (f ) f ) = fx : P (f ) (x) f (x) > 0g and A :=
supp (P (f ) f ) = fP (f ) (x) f (x) < 0g. Then
Z Z
+
0= (P (f ) f ) d = (P (f ) f ) d
A+ A+

+ +
which implies (P (f ) f ) 0 because (P (f ) f ) is a nonnegative function. One
can equally prove that (P (f ) f ) 0 so P (f ) = f .
(=) Obvious.

6
(b) From the de…nition of the Frobenius-Perron operator,
Z Z
1T 1 (A) f d = 1A P (f )d ;
X X

It is clear that P (f )(x) = 0 on A implies f (x) = 0 on T 1 (A) and vice versa. Let
A := Xn supp P (f ). Then f (x) = 0 for any x 2 T 1 (A) so supp f XnT 1 (A) =
1
T (XnA). But XnA = supp P (f ) and the result follows.
(c) Let f1 and f2 be two di¤erent stationary densities. That is, P (f1 ) = f1 , P (f2 ) = f2 and
kf1 kL1 (X; ) = kf2 kL1 (X; ) = 1. Let g = f1 f2 . Then P (g) = g and P (g + ) = g + and
P (g ) = g as well. Since f1 6= f2 and both are densities, g + and g are not identically
zero a.s.. De…ne A = supp g + and B = supp g , which are strictly positive disjoint sets.
By (b), A T 1 (A) and B T 1 (B), where T 1 (A) and T 1 (B) are also disjoint. We
have
1 2
A T (A) T (A) :::
1 2
B T (B) T (B) :::
n n
where T (A) and T (B) are disjoint for any n 2 N. De…ne
1
[ 1
[
n n
A= T (A) and B = T (B) :
n=0 n=0

These sets are invariant. Indeed, consider for instance A,

1
[ 1
[
1 n n
T A = T (A) = T (A) = A
n=1 n=0

because A T 1 (A). Equivalently, T 1 B = B. Since A and B are invariant disjoint

sets with strictly positive measure, the map is not ergodic. Contradiction.
(d) Suppose that T is not ergodic. Then 9A 2 F such that T 1 (A) = A and 0 < (A) < 1.
B := XnA is also invariant with strictly positive measure. Observe that
1A + 1B = 1X = P (1X ) = P (1A + 1B ) = P (1A ) + P (1B )
where we have used the linearity of P and the fact that P (1X ) = 1X . Since 1A is zero
on B (resp. 1B is zero on A), by (b), P (1A ) is zero on B as well (resp. P (1B ) is zero
on A). Therefore,
P (1A ) = 1A and P (1B ) = 1B
which is a contradiction because neither 1A nor 1B are constants.
(e) Suppose that T is ergodic.
From previous results, we have
Z Z Z
gP n (f ) d = gd (T n ) = g T nd :
X X X

Now,
N 1Z Z Z
1 X n 1 XN 1 n 1 XN 1
gP (f ) d = g T d = g T n fd :
N n=0 X X N n=0 X N n=0

7
Since g 2 L1 (X; ) and , g 2 L1 (X; ) and

1 XN 1 1 XN 1
g Tn kgkL1 (X; ) = kgkL1 (X; )
N n=0
L1 (X; ) N n=0

so
f XN 1
g Tn jf j kgkL1 (X; ) for any N 2 N:
N n=0

On the other hand, since T is ergodic,

Z
1 XN 1
g Tn ! gd -a.s..
N n=0 X

By the Dominated Convergence Theorem,

Z Z Z Z Z
1 XN 1 n
g T fd ! gd fd = gd fd :
X N n=0 n!1 X X X X

Now suppose that

N 1Z Z Z
1 X
lim P n (f )gd = f d gd (4)
N !1 N
n=0

for any f 2 L1 (X; ) and any g 2 L1 (X; ). We want to prove that T is ergodic. So
take A 2 F and invariant set, T 1 (A) = A. Since A is invariant, P (1A ) = 1A . Indeed,
for any B 2 F,
Z Z Z
P (1A ) d = 1A d = 1A\T 1 (B) d
B T 1 (B) X
Z Z
= 1T 1 (A\B) d = 1A\B T d
ZX ZX Z
= 1A\B d(T ) = 1A\B d = 1A d ;
X X B

which implies P (1A ) = 1A . Taking f = 1A in Eq. (4),

Z Z Z Z
1A gd = 1A d gd = (A) gd : (5)

2
Now, taking g = 1A , (5) we have (A) = (A) which can only be satis…ed either if
(A) = 0 or (A) = 1. So T is ergodic.