CHAPTER VII

TOPOLOGICAL ENTROPY II:

HOMOGENEOUS MEASURES

In this section we show how the topological entropy of simple examples may
be computed explicitly, and then show that in good situations certain measure–
theoretic entropies may be deduced.

A topological Kolmogorov–Sinai type theorem

We have seen that if {αn } is a sequence of open covers with diam(αn ) → 0, then
h(T ) = limn→∞ h(T, αn ). This is a topological analogue of Theorem 4.10. We now
give a topological analogue of Theorem 4.6. Let X denote a compact metric space.
A homeomorphism T : X → X is expansive if there is an expansive constant
δ > 0 with the property that if x 6= y then there is an n ∈ Z with d(T n , T n y) > δ.
T∞A finite−nopen cover α is a generator for T if for each map φ : Z → α, the set
¯
n=−∞ T φ(n) contains at most one point of X.
ATfinite open cover α is a weak generator for T if for each map φ : Z → α, the
∞
set n=−∞ T −n φ(n) contains at most one point of X.
Lemma 7.1. T has a generator if and only if T has a weak generator.
Proof. If α is a generator, then α is certainly a weak generator.
Suppose therefore that β = {B1 , . . . , Bs } is a weak generator for T . Let δ be a
Lebesgue number for β, and let α = {Ai } be a finite open cover with diam(α) ≤ δ.
Given a map φ : Z → α, then for any n, there is a j(n) with φ(n) ⊂ Bj(n) . Therefore
∞ ∞
¯ ⊂
\ \
−n
T φ(n) T −n Bj(n) ,
n=−∞ n=−∞

so that α is a generator. 
Lemma 7.2. T is expansive if and only if T has a generator.
Proof. Let δ be an expansive constant T∞ for T , and¯let α be an nopenncover with
diam(α) < δ. If φ : Z → α, and x, y ∈ n=−∞ T −n φ(n), then (T x, T y) ≤ δ for
all n, so by expansiveness x = y. Therefore α is a generator.
Conversely, let α be a generator, and let δ be a Lebesgue number for α. If
d(T n x, T n y) ≤ δ for all n, then for each n there is a φ(n) ∈ α with T n x, T n y ∈ φ(n),
so
∞
\
x, y ∈ T −n φ(n).
n=−∞

Since α is a generator, this implies that x = y, so T is expansive. 

Typeset by AMS-TEX
43
44 CHAPTER 7. TOPOLOGICAL ENTROPY II: HOMOGENEOUS MEASURES

Theorem 7.3. Let T : X → X be an expansive homeomorphism of the compact

metric space (X, d). (a) If α is a generator for T , then h(T ) = h(T, α). (b) If δ is
an expansive constant for T , then h(T ) = r(δ 0 , T ) = s(δ 0 , T ) for all δ 0 < δ/4.
Proof. (a) Let β be an open cover,
W and letδ be a Lebesgue number for β. We
N −n
claim first that for some N , diam −N T α < δ.
If not, then for all j there are points xj , yj with d(xj , yj ) > δ and a φj (i) ∈ α =
Tj
{A` }, −j ≤ i ≤ j with xj , yj ∈ i=−j T −i φj (i). Choose a subsequence {jk } with
xjk → x and yjk → y. Then x 6= y since for all J, d(xj , yj ) > δ. Consider the sets
φj (0); since α is finite infinitely many of them must coincide, so xjk , yjk ∈ A0 say,
for infinitely many k, hence x, y ∈ Ā0 . Similarly, for each n, infinitely many φjk (n)
coincide, giving An ∈ α with x, y ∈ T −n A¯n . Thus,

∞
\
x, y ∈ T −n A¯n ,
n=−∞

which is impossible since α is a generator.

WN
Choose therefore N with β < −N T −n α. Then

N
!
_
h(T, β) ≤ h T, T −n α
−N
k−1 N
!!
1 _ _
= lim H T −i T −n α
k→∞ k
i=0 −N
N +k−1
!
1 _
= lim H T −n α
k→∞ k
n=−N
2N_
+k−1
!
1
= lim H T −n α
k→∞ k
n=0
2N_
+k−1
!
2N + k − 1 1 −n
= lim H T α
k→∞ k 2N + k − 1 n=0
= h(T, α).

That is, h(T, β) ≤ h(T, α) for all open covers β so h(T ) = h(T, α).
(b) Let δ 0 < δ/4, and choose x1 . . . , xk such that X is covered by the sets
B(xi , 12 δ − 2δ 0 ). Then the open cover α = {B(xi , 12 δ)} has 2δ 0 for a Lebesgue
number so by Lemma 6.11, h(T, α) ≤ r(δ 0 , X) ≤ s(δ 0 , X) ≤ h(T ). Apply (a) to
obtain the result (since α is a generator). 
Corollary 7.4. An expansive homeomorphism has finite topological entropy.

Examples
(1)Q
The full shift. Let T be the shift map on the full two–sided shift space
∞
X = −∞ {0, 1, . . . , k − 1}. Then α = {Aj = {x ∈ X | x(0) = j}} is a natural
 EXAMPLES 45

generator for T . By Theorem 7.3 we therefore have

n−1
!
1 _
h(T ) = h(T, α) = lim log N T −i α
n→∞ n
i=0
1
= lim log(k n ) = log k.
n→∞ n

(2) Subshifts. If T on X is the full shift on k symbols as above, then a closed

subset Z ⊂ Z = X with T Z = Z is called a subshift. Let tn (Z) denote the number
of ordered n–tuples [i0 , . . . , in−1 ] such that the set {x ∈ Z | x0 = i0 , . . . , xn−1 =
in−1 } is non empty. Then

1
h(T |Z ) = lim log tn (Z).
n→∞ n

This is clear: let α be the natural generator, and notice that tn (Z) is equal to
Wn−1
N ( i=0 T |−i
Z α).
(3) Subshifts of finite type. A special case of (2) is given by subshifts of
finite type or topological Markov shifts. Let A = (aij ) be a k × k matrix with
entries in {0, 1}. Define XA ⊂ X by

XA = {x ∈ X | ax(n),x(n+1) = 1∀n ∈ Z}.

Then XA is a closed shift–invariant subset of the full shift on k symbols. Let TA

denote T |XA ; TA is the subshift of finite type determined by the matrix A. Notice
that if aij = 1 for all i, j, then XA = X. The matrix A is called irreducible if for
any pair i, j there is some n > 0 such that the (i, j)–th element in An is positive.
If A is an irreducible {0, 1} matrix, then h(TA ) = log λ, where λ is the largest
positive eigenvalue of A. To prove this, use (2) as follows. The set {x ∈ Z | x0 =
i0 , . . . , xn−1 = in−1 } is non–empty if and only if

ai0 i1 ai1 i2 . . . ain−2 in−1 = 1.

It follows that
k−1
X k−1
X
tn (XA ) = ai0 i1 ai1 i2 . . . ain−2 in−1 = (An−1 )i0 in−1 .
i0 ,...,in−1 =0 i0 ,in−1 =0

P
Introduce a norm on the set of k×k matrices by k(bij )k = i,j |bij |. Then tn (XA ) =
kAn−1 k, so
1
log tn (XA ) = log kAn−1 k1/n → log λ,
n
by the spectral radius formula.
(4) Smooth maps. Let T : M → M be a differentiable transformation of a
p–dimensional Riemannian manifold M . Let d be the metric on M induced by the
Riemannian metric. Then

hd (T ) ≤ max{0, p log( sup kτx T k)}

x∈M
46 CHAPTER 7. TOPOLOGICAL ENTROPY II: HOMOGENEOUS MEASURES

where τx T : τx M → τx M is the tangent to T at x ∈ M , and k · k is the norm

induced on τx M by the Riemannian metric.
Let a = supx∈M kτx T k and assume that a < ∞. If a ≤ 1 then the mean value
theorem implies that d(T x, T y) ≤ ad(x, y), so that h(T ) = 0.
Assume therefore that 1 < a < ∞. Again, the mean value theorem implies that
d(T x, T y) ≤ ad(x, y). Let K ⊂ M be compact, and fix  > 0. Let k · kp be the
max norm on Rp , and let B(0, r) be the ball centred at 0 with radius r under this
norm. Choose differentiable maps fj : B(0, 12 ) → M , 1 ≤ j ≤ q, such that the sets
fj (B(0, 12 )) cover K. Let b > 0 have d(fj (u), fj (v)) ≤ bku−vkp for all u, v ∈ B(0, 12 )
and any j. For any δ, 0 < δ < 1, let E(δ) = {(k1 δ, . . . , kp δ) ∈ Rp | ki ∈ Z}∩B(0, 21 ).

p
Then |E(δ)| ≤ 1δ . Each point of B(0, 12 ) is within a distance δ of some point in
Sq
E(δ). Consider F (δ) = j=1 fj E(δ); this set is clearly a (n, an bδ)–spanning set for
K with respect to T . Put δ = (an b)−1 , and deduce that rn (, K) ≤ q(an b−1 )p =
anp (qb−p −p ), so r(, K, T ) ≤ p log a.
(5) As an application of (4), if T : Rn → Rn is a linear map, then h(T ) is
bounded above by max{0, n log |λ|} where λ is an eigenvalue of T with maximum
modulus. To see this, apply (4) to T m as follows:

1 1
h(T ) = h(T m ) ≤ max{0, n log kT m k} ≤ max{0, n log kT m k1/m },
m m

and then recall that kT m k1/m → |λ| as m → ∞.

Homogeneous measures
The next goal is to replace the estimate in Example (5) above with an exact
formula. To do this, we need to make some preliminary steps towards the following
interesting problem: given a homeomorphism T of a compact metric space X, which
(if any) T –invariant Borel probability measures µ on X may be expected to have
h(T ) = hµ (T )? The main result in this area is the variational principle, which states
that
h(T ) = sup{hµ (T ) | µ ∈ M (X, T )},
where M (X, T ) is the space of T –invariant Borel probability measures on X. This
result is more than we need, so we will prove a much more elementary result suffi-
cient for our needs. Recall that X has covering dimension at most m if every open
cover α of X has a refinement β such that each point of X lies in at most (m + 1)
elements of β. For example, an m–dimensional manifold has this property.
Lemma 7.4. Let (X, d) be a compact metric space with covering dimension at
most m for some m, and let T : X → X be continuous. Then, if µ is a T –invariant
Borel probability measure on X,

hµ (T ) ≤ hd (T ).

Proof. Let β = {B1 , . . . , Br } be an open cover of X with diam(β) <  and

with each point of X in at most m + 1 members of β . Let α = {A1 , . . . , Ar } be
a measurable partition of X with Āi ⊂ Bi for each i. For each x ∈ X choose an
open neighbourhood Cx of x intersecting at most (m + 1) Ai ’s (if x ∈/ Bi then we
 HOMOGENEOUS MEASURES 47

can assume that x ∈

/ Āi ). Let γ = {Cx1 , . . . , Cxs } cover X, and let δ be a Lebesgue
number for γ, and let C(y) ∈ γ be the element of γ with Bδ¯(y) ⊂ C(y).
Let Sn be an (n, δ)–spanning set for X with respect to T . Let
Tn = {(i0 , . . . , in−1 ) | y ∈ Sn , Aij ∩ C(T j y) 6= ∅∀0 ≤ j < n}.
Tn−1
Now suppose that x ∈ A(i0 , . . . , in−1 ) = j=0 T −j Aij . Pick y ∈ Sn so that
d(T j x, T j y) ≤ δ for all 0 ≤ j < n. Then T j (x) ∈ Aij ∩C(T j y) and so (i0 , . . . , in−1 ) ∈
Tn . Let Un = {(i0 , . . . , in−1 ) | A(i0 , . . . , in−1 ) 6= ∅}. Notice that Un ⊂ Tn . Since
C(T j y) can intersect at most (m + 1) of the Ak ’s,
|Tn | ≤ (m + 1)n |Sn |.
It follows that
|Un | ≤ |Tn | ≤ rn (δ, X, T )(m + 1)n .
Now we know that (Lemma 3.5)
n−1
!
_
H T −i α ≤ log |Un |,
i=0
so
n−1
!
1 _
−i 1
hµ (T, α ) = lim H T α ≤ lim sup log(m + 1)n rn (δ, X)
n→∞ n n→∞ n
i=0
= log(m + 1) + r(δ, X) ≤ log(m + 1) + hd (T ).
Let  → 0 to get
hµ (T ) = sup hµ (T, α ) ≤ log(m + 1) + hd (T ).


Applying this inequality to T n ,

1 1 1 1
hµ (T n ) ≤ log(m + 1) + hd (T n ) = log(m + 1) + hd (T ).
hµ (T ) =
n n n n
Letting n → ∞ gives hµ (T ) ≤ hd (T ). 
Definition 7.5. Let T : X → X be a uniformly continuous map of the metric
space (X, d). A Borel measure µ on X is T –homogeneous if
(1) µ(K) < ∞ for all compact K ⊂ X;
(2) µ(K) > 0 for some compact K;
(3) for each  > 0 there exists δ > 0 and c > 0 such that
n−1
! n−1
!
\ \
−k k −k k
µ T Bδ (T x) ≤ cµ T B (T y)
k=0 k=0

for all n ≥ 0 and x, y ∈ X.

Given such a measure, define
n−1
!
1 \
h∗ (µ, T ) = lim lim sup − log µ T −k B (T k x) .
→0 n→∞ n
k=0

By property (3) this definition does not depend on the point y chosen. See Exercise
7.1.
48 CHAPTER 7. TOPOLOGICAL ENTROPY II: HOMOGENEOUS MEASURES

Theorem 7.6. Let T be a uniformly continuous map on X (a locally compact

metric space), and let µ be a T –homogeneous measure. Then hd (T ) = h∗ (µ, T ).
If X is compact and of finite covering dimension, and µ is a T –invariant T –
homogeneous probability measure, then hµ (T ) = h∗ (µ, T ).
Tn−1
Proof. Let Dn (x, , T ) = k=0 T −k B (T k x).
Consider a compact subset K ⊂ X. Using (1) and local compactness, choose a
neighbourhood U of K with µ(U ) < ∞. Let  > 0 be so small that B (K) ⊂ U . If
E ⊂ K is (n, )–separated, then
[
Dn (x, 21 , T ) ⊂ U
x∈E

is a disjoint union. Choose δ, c so that

µ (Dn (y, δ, T )) ≤ cµ Dn (x, 21 , T )

for all x, y. Then µ (Dn (y, δ, T )) sn (, K) ≤ cµ(U ) and so

1
s(, K) ≤ lim sup − log µ (Dn (y, δ, T )) .
n→∞ n

Letting  → 0 shows that hd (T, K) ≤ h∗ (µ, T ), so hd (T ) ≤ h∗ (µ, T ).

Now consider
S K with µ(K) > 0. If F is an (n, δ)–spanning set for K under
T , then x∈F Dn (x, 2δ, T ) ⊃ K. Given  > 0, choose δ > 0 and c > 0 such that
µ (Dn (x, 2δ, T )) ≤ cµ (Dn (y, , T )) for all x, y and n ≥ 0. Then

cµ (Dn (y, , T )) rn (δ, K) ≥ µ(K) > 0

and so
1
r(δ, K) ≥ lim sup − log µ (Dn (y, , T )) .
n→∞ n
Letting  → 0, hd (T, K) ≥ h∗ (µ, T ), so hd (T ) = h∗ (µ, T ).
Now assume that X is compact, and µ is a T –invariant T –homogeneous proba-
bility measure. By Lemma 7.4,

hµ (T ) ≤ hd (T ) = h∗ (µ, T ).

Let  > 0, and choose δ > 0 and c > 0 with µ (Dn (x, δ, T )) ≤ cµ (Dn (y, , T )) for
all x, y and n ≥ 0. Let α = {A1 , . . . , Ar } be a measurable partition of X with
Tn−1
diam(α) < δ. If x ∈ A(i0 , . . . , in−1 ) = k=0 T −k Aik . Then A(i0 , . . . , in−1 ) ⊂
Dn (x, δ, T ) and so µ(A(i0 , . . . , in−1 )) ≤ cµ(Dn (y, , T )). It follows that

1
hµ (T ) ≥ hµ (T, α) ≥ lim sup − log µ (Dn (y, , T )) .
n→∞ n

Letting  → 0, hµ (T ) ≥ h∗ (µ, T ). 
 THE VARIATIONAL PRINCIPLE 49

Lemma 7.7. Let G be a locally compact metrizable group, with right invariant
metric d and a right Haar measure µ, and let T : G → G be a surjective affine map
T (x) = A(x)g. Then µ is T –homogeneous and h∗ (µ, T ) = h∗ (µ, A).
Proof. It is clear that T is uniformly continuous. We prove by induction that

(7.1) T −k B (T k x) = (A−k B (e))x.

For k = 0 this is clear, since d is right–invariant. Assume (7.1); then

T −(k+1) B (T k+1 x) = T −1 T −k B (T k (T x)) = T −1 [(A−k B (e))T x]

= A−1 ([(A−k B (e))A(x)g]g −1 ) = A−1 [(A−k B (e))A(x)]
= (A−(k+1) B (e))x,

where for the last step we have used the fact that A−1 (CA(x)) = A−1 (C)x for any
set C ⊂ G since A is onto. By intersecting (7.1) over all k, 0 ≤ k < n, we see that

Dn (x, , T ) = Dn (e, , A)x.


Theorem 7.8. Let G be a compact metrizable group (of finite covering dimen-
sion), T : G → G a surjective affine transformation, T (x) = A(x)g for an auto-
morphism A, and let µ be normalized right Haar measure on G. Then

hµ (T ) = hµ (A) = h∗ (µ, T ) = hd (A).

Proof. By compactness, µ is T –invariant and A–invariant, so we may apply

Lemma 7.7 and Theorem 7.6. 

The variational principle

Let X be a compact metric space, and T : X → X a continuous map. We have
seen in Lemma 7.4 that under certain special circumstances, the topological entropy
h(T ) exceeded that measure–theoretic entropy of T with respect to a “natural”
measure. To explain the modern context for this result, let M (X, T ) denote the
space of T –invariant Borel probability measures on X.
Theorem 7.9. h(T ) = sup{hµ (T ) | µ ∈ M (X, T )}.
Lemma 7.4 is sufficient for our purposes, so Theorem 7.9 will not be proved here.
The variational principle gives a criterion to select certain distinguished invariant
measures.
Definition 7.10. A measure µ ∈ M (X, T ) is a measure of maximal entropy if
hµ (T ) = h(T ).
Notice that a given map T may not have any maximal measures. Also, Theorem
7.8 states the following: Haar measure is a measure of maximal entropy for compact
group endomorphisms. In fact more is true: if T is an ergodic group automorphism,
then Haar measure is the unique measure of maximal entropy for T .
50 CHAPTER 7. TOPOLOGICAL ENTROPY II: HOMOGENEOUS MEASURES

Linear maps and covering spaces

Theorem 7.11. Let (X, d) and (X 0 , d0 ) be metric spaces with a continuous sur-
jection ρ : X 0 → X such that there exists δ > 0 with

ρB(x0 ,δ) : B(x0 , δ) → B(ρ(x0 ), δ)

an isometric surjection for all x0 ∈ X 0 . If T 0 : X 0 → X 0 is uniformly continuous,

and T : X → X is uniformly continuous with ρT 0 = T ρ, then

hd (T ) = hd0 (T 0 ).

Proof. If K 0 ⊂ X 0 is compact and diam(K 0 ) < δ then ρ(K 0 ) ⊂ X is compact

and diam(ρ(K 0 )) < δ, and any compact subset in X with diameter smaller than δ
is of this form. Choose  ∈ (0, δ) with the property that d0 (x0 , y 0 ) <  implies that
d(T 0 x0 , T 0 y 0 ) < δ.
Let E 0 ⊂ K 0 be an (n, )–separating set for T 0 . Let x0 6= y 0 belong to E 0 ;
i i
then ρ(x0 ) 6= ρ(y 0 ). Let j be chosen so that d0 (T 0 x0 , T 0 y 0 ) ≤  for i ≤ j, and
j+1 j+1 j j
d0 (T 0 x0 , T 0 y 0 ) > . Since d0 (T 0 x0 , T 0 y 0 ) ≤ , we also have

j+1 0 j+1 0
d0 (T 0 x ,T0 y)<δ

by choice of . So
j+1 0 j+1 0
d(T j+1 ρ(x0 ), T j+1 ρ(y 0 )) = d0 (T 0 x ,T0 y ) > .

It follows that ρ(E 0 ) is an (n, )–separated set for T , so

sn (, K 0 , T 0 ) ≤ sn (, ρ(K 0 ), T ).

Coversely, let E be an (n, )–separated set in ρ(K 0 ) for T , where K 0 is compact

with diameter less than δ. Let E 0 = ρ−1 (E) ∩ K 0 ; then E 0 is an (n, )–separated
set for T 0 since ρ is an isometry on K 0 . Therefore

sn (, ρ(K 0 ), T ) ≤ sn (, K 0 , T 0 ),

so h0d (T 0 , K 0 ) = hd (T, ρ(K 0 )). It follows that hd0 (T 0 ) = hd (T ). 

Recall Example 5 above: if T : Rn → Rn is a linear map, and d is any metric
giving the norm on Rn , then

(7.2) hd (T ) ≤ max{0, n log |λ|},

where λ is an eigenvalue of T with maximum absolute value.

Theorem 7.12. If T : Rm → Rm is a linear map, then
X
hd (T ) = log |λi |
|λi |>1
 LINEAR MAPS AND COVERING SPACES 51

where λ1 , . . . , λm are the eigenvalues of T .

Proof. Decompose T into T1 ⊕· · ·⊕Ts acting on Rm = E1 ⊕· · ·⊕Es , where the
Ei are linear subspaces of Rm with T (Ei ) ⊂ Ej , Ti = T |Ei , and all the eigenvalues
of Ti have fixed absolute value αi . Thus
X X
log |λi | = dim(Ei ) log αi .
|λi |>1 αi >1

By Lemma 6.4 and (7.2) we therefore have

s
X s
X X
(7.3) hd (T ) ≤ hd (Ti ) ≤ max{0, dim(Ei ) log αi } = log |λi |.
i=1 i=1 |λi |>1

For the reverse inequality, decompose T into A1 ⊕ A2 , Rm = V1 ⊕ V2 , Ai = T |Vi ,

A1 has all eigenvalues with modulus greater than 1, and A2 has all eigenvalues with
modulus less than or equal to 1. Let µi be Lebesgue measure on Vi , and let di be
the Euclidean metric on Vi . Then d = max{d1 , d2 } is a metric on Rm uniformly
equivalent to the usual one, so we may use it to evaluate the entropy of T . By
Theorem 7.6, hd (T ) is given by
    
n−1 n−1
1  − 1 log µ2 
A−j A−j
\ \
lim lim sup − log µ1  1 Bd1 (0, ) 2 Bd2 (0, )
 .
→0 n→∞ n j=0
n j=0

Now
 
n−1  
−(n−1)
A−j
\
µ1  1 B d1 (0, ) ≤ µ1 A1 B d1 (0, )
j=0

= | det(A1 )|−(n−1) µ1 (Bd1 (0, )),

while  
n−1
A−j
\
µ2  2 Bd2 (0, )
 ≥ µ2 (Bd2 (0, )) .
j=0

It follows that X
h(T ) ≥ log | det A1 | = log |λi |.
|λi |>1

This with (7.3) proves the theorem. 

Corollary 7.13. Let T : Tm → Tm be an affine map, T (x) = g + A(x), with
A onto. Then
X
h(T ) = hµ (T ) = h(A) = hµ (A) = log |λi |,
|λi |>1

where µ is Haar measure on the torus, and λ1 , . . . , λm are the eigenvalues of the
matrix determined by A.
Proof. The first four equalities have been proved already. The last follows by
applying Theorem 7.11 and Theorem 7.12. 
52 CHAPTER 7. TOPOLOGICAL ENTROPY II: HOMOGENEOUS MEASURES

Notes. The material on expansiveness is due mainly to Keynes and

Robertson, [10] (exact references are in [35]). The notion of homogeneous
measure, and the formulation of topological entropy in terms of the
decay of volume under a homogeneous measure, is due to Bowen, [3].
A complete proof of the variational principle may be found in [34] and
[35].

Exercises
7.1 Show that the property of being T –homogeneous, together with the quan-
tity h∗ (µ, T ) for a homogeneous measure µ, depend only on the uniform
equivalence class of the metric d.
7.2 Use Corollary 7.13 to exhibit countably many toral automorphisms that are
not measurably isomorphic to each other.
7.3 Let X be a compact metric space, and let M (X) be the set of all Borel
probability measures on X; write B for the σ–algebra of Borel sets. Recall
that any m ∈ M (X) is regular, so for any Borel set B,

m(B) = sup m(C),

C⊂B

where the supremum is taken over all closed sets C, and

m(B) = inf m(U ),

U ⊃B

where the infimum is taken over all open sets U . Recall also the Riesz Rep-
resentation Theorem, stating that any continuous linear positive operator
J : C(X) R → C with J(1) = 1 has a corresponding mJ ∈ M (X) such that
J(f ) = X f dm.
a. Show that each m ∈ M (X) is characterised by Rhow it integrates R the
continuous functions: if m1 , m2 ∈ M (X) have X f dm1 = X f dm2
for all f ∈ C(X), then m1 = m2 .
b. Let the weak∗ –topology
R on M (X) be the smallest topology making
each map m → X f dm, f ∈ C(X), continuous. A basis for this
topology
R isRtherefore given by all sets of the form U (µ) = {m ∈ M (X) |
| fi dµ − fi dm| < , i = 1, . . . , n}. Prove that M (X) is metrisable
in the weak∗ –topology by showing that
∞ R R
−k | fn dµ − fn dm
X
d(µ, m) = 2
kfn k
k=1

is a metric giving this topology (here {fn } is a dense subset of C(X),

and k · k is the sup norm.)
c. Prove that (i) =⇒ (ii) =⇒ (iii) =⇒ (iv) for the following statements.
(i) µn → µ weak∗ ; (ii) for F closed, lim supn→∞ µn (F ) ≤ µ(F ); (iii)
for O open, lim inf n→∞ µn (O) ≥ µ(O); (iv) for A ∈ B with µ(δA) = 0,
µn (A) → µ(A). Show by example that µn → µ does not imply that
µn (A) → µ(A) for all A ∈ B.
d. Prove directly the following standard fact from functional analysis:
M (X) is weak∗ –compact.
 LINEAR MAPS AND COVERING SPACES 53

7.4 This problem continues 7.3 above. Let T : X → X be continuous. Such a

T is automatically B–measurable, so there is a map T̃ : M (X) → M (X)
defined by (T̃ µ)(B) =R µ(T −1 B). R
a. Show that: (i) f (x)d(T̃ µ(x)) = f (T (x))dµ(x) for any f ∈ C(X).
Let M (X, T ) ⊂ M (X) denote the subset of TR–invariant measures.
R Use
(i) to show (ii) µ ∈ M (X, T ) if and only if f ◦ T dµ = f dµ for all
f ∈ C(X).
b. Show that the map T̃ : M (X) → M (X) is continuous and affine.
(Then T̃ is a continuous affine map of a compact convex set, so by the
Kakutani–Markov theorem T̃ has a fixed point.)
c. Prove directly that T̃ has a fixed point as follows. Let {νn }n∈N be
Pn−1
any sequence in M (X). Let µn = n1 i=0 T̃ i νn . Prove that any limit
point of {µn }n∈N is a member of M (X, T ). Deduce that T̃ has a
fixed point in M (X), so M (X, T ) is always non–empty (this is the
Krylov–Bogolioubov Theorem).
d. Show that M (X, T ) is a compact convex subset of M (X).
e. Show that µ ∈ M (X, T ) is an extreme point if and only if T is an
ergodic measure–preserving transformation of (X, B, µ). (Such a mea-
sure is called an ergodic measure).
f. If µ and ν are ergodic members of M (X, T ), show that either µ = ν
or µ and ν are mutually singular.
7.5 Prove the following (used in Theorem 7.6). Let α = {A1 , . . . , Ar } be a
measurable partition of X, and  assume that there
 is a constant c and a
Tn−1 −k
positive sequence an such that µ k=0 T Aik ≤ can . Then hµ (T, α) ≥
lim supn→∞ − n1 log an .