Lecture 2

In this second lecture we introduce the key objects of the course: Measure-preserving
systems. We then discuss Bernoulli shifts as an important class of examples. Fi-
nally, we introduce the property of “ergodicity” and study different approaches to
subsystems of a measure-preserving system.

2.1 Measure-Preserving Systems

In ergodic theory, we are interested in the asymptotic behavior of measure-preserving
transformations. In this course, we will also assume that these are invertible, and
introduce the following notation for a probability space X.

(i) Aut(X) := {τ ∈ M(X, X) | τ invertible},

(ii) Aut(Σ(X)) := {T ∈ M(Σ(X), Σ(X)) | T bijective}, and
(iii) Aut(L2 (X)) := {U ∈ M(L2 (X), L2 (X)) | U bijective}.

Elements of these sets are called automorphisms of X, measure algebra auto-

morphisms of Σ(X), and Markov automorphisms of L2 (X), respectively.
Note that all these sets equipped with the natural compositions are groups. Theorem
1.3.7 yields a group isomorphism between Aut(Σ(X)) and Aut(L2 (X)) (cf. Remark
1.3.5). If X is a Lebesgue space, then by Theorem 1.2.8 (and Remark 1.2.4) all
three groups are isomorphic.
We could now study the behavior of the iterates of an element in any one of these
automorphism groups. However, we start from a more general definition of measure-
preserving systems.
1 n ω
L∞Γ be an arbitrary (discrete) abelian group (e.g., Γ = Z , n ≥ 1, Γ = Fp =
Let
n=1 Fp , where Fp is the finite field of prime order p) and fix this for now.

1
Some topics of these lectures can be treated in the framework of arbitrary groups (or even
semigroups). However, we restrict to the case of abelian groups throughout the entire course.

19
20 LECTURE 2.

Definition 2.1.1. An (abelian) measure-preserving system (X, T ) consists of

a probability space X and a group homomorphism
T : Γ → Aut(Σ(X)), γ 7→ Tγ ,
i.e., Tγ1 +γ2 = Tγ1 Tγ2 for γ1 , γ2 ∈ Γ.
For a measure-preserving system (X, T ) the induced group homomorphism
UT : Γ → Aut(L2 (X)), γ 7→ UTγ
is the Koopman representation of (X, T ).
The following are trivial, but important examples of measure-preserving systems.

Example 2.1.2. For any probability space X we have a system (X, Id) given by
the group homomorphism Id : Γ → Aut(Σ(X)), γ 7→ IdΣ(X) . A very special case is
the trivial system ({0}, Id) where {0} is equipped with the point measure δ0 (cf.
Example 1.1.3 (ii)).
Most examples are given by concrete measure-preserving maps. We therefore intro-
duce a second type of measure-preserving systems.

Definition 2.1.3. A concrete measure-preserving system (X, τ ) consists of a

probability space X and a group homomorphism
τ : Γ → Aut(X), γ 7→ τγ .

Remark 2.1.4. One has to be a bit careful here: If we choose a concrete measure-
preserving map representing τγ for γ ∈ Γ, the identity τγ1 +γ2 = τγ1 τγ2 only holds
outside of a nullset depending on γ1 and γ2 . In particular, notions like the orbit of
a point do not make sense in this framework.2
We immediately obtain the following fact (cf. Remark 1.3.5).

Proposition 2.1.5. Every concrete measure-preserving system (X, τ ) gives rise to

a measure-preserving system (X, τ ∗ ) as in Definition 2.1.1 by setting
(τ ∗ )γ := (τγ )∗ : Σ(X) → Σ(X), A 7→ τγ−1 (A)
for γ ∈ Γ.
If X is a Lebesgue space, then – as a consequence of Theorem 1.2.8 – every measure-
preserving system (X, T ) is of the form (X, τ ∗ ) for a uniquely determined concrete
measure-preserving system (X, τ ), see Exercise 2.6 below.
2
In the case of a countable group Γ (which suffices for most applications), one can circumvent
this problem and transition to a proper action of Γ on X by measure-preserving maps. However,
we avoid these measure-theoretic intricacies altogether by using Definition 2.1.1 and working on
the level of measure algebras.
2.1. MEASURE-PRESERVING SYSTEMS 21

Remark 2.1.6. Notice that if Γ = Z, then a measure-preserving system (X, T ) is

completely determined by T1 : Σ(X) → Σ(X) since Tm = T1m for every m ∈ Z.
Conversely, if S : Σ(X) → Σ(X) is a measure algebra automorphism, then T : Z →
Aut(Σ(X)), n 7→ S n yields a measure-preserving system with respect to the group Z.
This establishes a one-to-one correspondence between measure-preserving systems
over Z and automorphisms S : Σ(X) → Σ(X). An analogous correspondence holds
for concrete measure-preserving systems (X, τ ) over Z and elements σ ∈ Aut(X).
In view of the previous remark, we already obtain some examples of concrete
measure-preserving systems from the previous lecture. To introduce another in-
teresting and important class of examples, we need the concept of product measures
for possibly infinitely many spaces (see [HS65, Section 22]).

Proposition and Definition 2.1.7. Let I ̸= ∅ be an index set and Xi a probability

space for each i ∈ I. For a finite set {i1 , . . . , im } ⊆ I and measurable sets Aj ∈ ΣXij
for j ∈ {1, . . . , m} we call
 Y 
Z(i1 , . . . , im ; A1 , . . . , Am ) := (xi )i∈I ∈ Xi | xij ∈ Aj for all j ∈ {1, . . . , m}
i∈I

Q
a (measurable) cylinder set. The smallestNσ-algebra over i∈I Xi containing all
such cylinder the product-σ-algebra i∈I ΣXi . There is a unique probability
N sets is N
measure i∈I µXi : i∈I ΣXi → [0, 1], called the product measure, with

O
µXi (Z(i1 , . . . , im ; A1 , . . . , Am )) = µXi1 (A1 ) · · · µXim (Am )
i∈I

for each cylinder set Z(i1 , . . . , im ; A1 , . . . , Am ).

I
If I ̸= ∅ is an index set and
Q X is a probability space, we also write X for the
product probability space i∈I X (i.e., in case Xi = X for all i ∈ I). With this
notation, we introduce a new example for a measure-preserving system.

Example 2.1.8. Let X be any probability space, e.g., X = {0, . . . , k − 1} with

the probability measure defined by the probability vector p = ( k1 , . . . , k1 ), see Ex-
ample 1.1.3 (ii). Then τ : Γ → Aut(X Γ ), γ 7→ τγ with τγ (xδ )δ∈Γ = (xδ+γ )δ∈Γ for
(xδ )δ∈Γ ∈ X Γ and γ ∈ Γ defines a concrete measure-preserving system (X Γ , τ ),
called a Bernoulli shift: For γ ∈ Γ and a cylinder set Z(δ1 , . . . , δm ; A1 , . . . , Am ) we
obtain

τγ−1 (Z(δ1 , . . . , δm ; A1 , . . . , Am )) = {(xδ )δ∈Γ | xδj +γ ∈ Aj for every j ∈ {1, . . . , m}}

= Z(δ1 + γ, . . . , δm + γ; A1 , . . . , Am ),
22 LECTURE 2.

and therefore

µX Γ (τγ−1 (Z(δ1 , . . . , δm ; A1 , . . . , Am ))) = µX Γ (Z(δ1 + γ, . . . , δm + γ; A1 , . . . , Am ))

= µX (A1 ) · · · µX (Am )
= µX Γ (Z(δ1 , . . . , δm ; A1 , . . . , Am )).

Since cylinder sets form a ∩-stable generator of the σ-algebra ΣX Γ , we obtain that
the map τγ : X Γ → X Γ is measurable and measure-preserving for every γ ∈ Γ by
Proposition 1.1.4. Moreover, it is clear that τγ1 +γ2 = τγ1 ◦ τγ2 for all γ1 , γ2 ∈ Γ.
We now introduce a crucial irreducibility property for measure-preserving systems.

Definition 2.1.9. For a measure-preserving system (X, T ) we say that A ∈ Σ(X)

is invariant if Tγ (A) = A for every γ ∈ Γ. We set

Σ(X)inv := {A ∈ Σ(X) | A invariant} ⊆ Σ(X).

The system (X, T ) is called ergodic if every invariant A ∈ Σ(X) already satisfies
µX (A) ∈ {0, 1}, i.e., Σ(X)inv = {∅, X}.

Examples 2.1.10. (i) For a system with trivial dynamics (X, Id) (see Example
2.1.2) every element A ∈ Σ(X) is invariant. Hence, such a system (X, Id)
is ergodic precisely when Σ(X) = {∅, X}. Thus, (X, Id) is “almost never”
ergodic.
(ii) Consider the set X = {0, 1} with the probability measure defined by the
probability vector ( 21 , 12 ) and the measure preserving system (X, τ ∗ ) over Γ = Z
defined by the measure-preserving map τ : X → X, m 7→ 1 − m, see Example
1.1.3 (ii) and Remark 2.1.6. Then (X, τ ∗ ) is ergodic.
What about Bernoulli shifts? For the trivial group Γ = {0}, we have (X Γ , τ ∗ ) =
(X Γ , Id), and this is only ergodic if Σ(X) = {∅, X}. The same is true for any
finite abelian group Γ (see Exercise 2.4). However, for infinite groups we obtain the
following.

Proposition 2.1.11. If the group Γ is infinite, then the Bernoulli shift (X Γ , τ ∗ ) is

ergodic. Moreover, for all measurable subsets A, B ⊆ X and every ε > 0 there is a
finite subset F ⊆ Γ such that

|µX Γ (τγ−1 (A) ∩ B) − µX Γ (A) · µX Γ (B)| ≤ ε for every γ ∈ Γ \ F.

For the proof of Proposition 2.1.11 we need two statements from measure theory.
The first one is a general approximation result for measurable sets, see, e.g., [Bil95,
Theorem 11.4].
2.1. MEASURE-PRESERVING SYSTEMS 23

Lemma 2.1.12. Let X be a probability space and E ⊆ ΣX a generator of the σ-

algebra which is an algebra over X, i.e., ∅, X ∈ E and A ∪ B, A ∩ B, A \ B ∈ E
whenever A, B ∈ E. Then for every measurable subset A ⊆ X and ε > 0 there is
B ∈ E with µX (A ∆ B) ≤ ε.
The second auxiliary result is a compatibility statement for product measures. If
we decompose the index set of a product space into two parts, then the product
measure can also be written as a product of two product measures:

Lemma 2.1.13. Assume that I ̸= ∅ is an index set and Xi is a probability space

for every i ∈ I. If for ∅ =
̸ J ⊊ I we naturally identify
Y Y Y
Xi = Xi × Xi
i∈I i∈J i∈I\J

as sets, then
O O O O O O
ΣXi = ΣXi ⊗ ΣXi and µXi = µXi ⊗ µXi .
i∈I i∈J i∈I\J i∈I i∈J i∈I\J

A proof can be found in [HS65, Lemmas 22.4 and 22.12].

Proof of Proposition 2.1.11. We show the second claim first and start with partic-
ularly simple measurable sets A, B ∈ ΣX Γ . For every finite subset E ⊆ Γ we may
identify X Γ = X E × X Γ\E , hence for each C ∈ ΣX E the set C × X Γ\E defines a
measurable subset Z(E, C) of X Γ by Lemma 2.1.13. We make the following basic
observations about these measurable sets which “depend only on a finite number of
coordinates”:
(i) For disjoint finite subsets E, E ′ ⊆ Γ, C ∈ ΣX E , and C ′ ∈ ΣX E′ we have
Z(E, C) ∩ Z(E ′ , C ′ ) = Z(E ∪ E ′ , C × C ′ ).
(ii) For every finite subset E ⊆ Γ and C ∈ ΣX E we have µX Γ (Z(E, C)) = µX E (C).
(iii) If E ⊆ Γ is a finite subset and γ ∈ Γ, then (since E and E + γ have the
same cardinality) we can identify X E with X E+γ . With this identification, we
obtain τγ−1 (Z(E, C)) = Z(E + γ, C) for every C ∈ X E .
Now set E := {Z(E, C) | E ⊆ Γ finite, C ∈ ΣX E }. It is easy to check that E is
an algebra over X Γ , and, since it contains all measurable cylinder sets, it generates
the σ-algebra ΣX Γ . We now show the second statement for sets A, B ∈ E, even in
a stronger version. The idea is to shift A far enough that it “lives on coordinates
disjoint from those of B”.
To make this precise, take finite subsets E, E ′ ⊆ Γ as well as measurable sets
C ∈ ΣX E and C ′ ∈ ΣX E′ . Then F := E − E ′ = {γ − γ ′ | γ ∈ E, γ ′ ∈ E ′ } is a
24 LECTURE 2.

finite subset of Γ. For γ ∈ Γ \ F , the sets E + γ and E ′ are disjoint. Using properties
(i) – (iii) above and Lemma 2.1.13, we therefore obtain

µX Γ (τγ−1 (Z(E, C)) ∩ Z(E ′ , C ′ )) = µX Γ (Z(E + γ, C) ∩ Z(E ′ , C ′ ))

= µX Γ (Z((E + γ) ∪ E ′ , C × C ′ ) = µX (E+γ)∪E′ (C × C ′ )
= µX E+γ (C) · µX E′ (C ′ ) = µX E (C) · µX E′ (C ′ )
= µX Γ (Z(E, C)) · µX Γ (Z(E ′ , C ′ )).

For general measurable subsets A, B ⊆ X we establish the claimed inequality by

using an approximation argument. For ε > 0 we find by Lemma 2.1.12 sets A′ , B ′ ∈
E with µX Γ (A ∆A′ ) ≤ 4ε and µX Γ (B ∆B ′ ) ≤ 4ε . By the above, we find a finite subset
F ⊆ Γ with µX Γ (τγ−1 (A′ ) ∩ B ′ ) = µX Γ (A′ ) · µX Γ (B ′ ) for all γ ∈ Γ \ F . For each γ ∈ Γ
we obtain by Exercise 1.4 and Lemma 1.2.6 (iii),

|µX Γ (τγ−1 (A) ∩ B) − µX Γ (τγ−1 (A′ ) ∩ B ′ )| ≤ µX Γ (τγ−1 (A) ∆ τγ−1 (A′ )) + µX Γ (B ∆ B ′ )

ε
= µX Γ (A ∆ A′ ) + µX Γ (B ∆ B ′ ) ≤ ,
2
and, by the triangle inequality and Exercise 1.4,

|µX Γ (A)µX Γ (B) − µX Γ (A′ )µX Γ (B ′ )| ≤ |µX Γ (A) − µX Γ (A′ )| + |µX Γ (B) − µX Γ (B ′ )|
ε
≤ µX Γ (A ∆ A′ ) + µX Γ (B ∆ B ′ ) ≤ .
2
Applying the triangle inequality once more, we obtain the desired estimate.
To prove ergodicity, let A ∈ Σ(X) be invariant. For ε > 0 we find a finite subset
F ⊆ Γ with

|µX Γ ((τγ )∗ (A) ∩ A) − µX Γ (A) · µX Γ (A)| ≤ ε

for every γ ∈ Γ \ F . Since Γ is infinite, such a γ actually exists. From A being

invariant we conclude (τγ )∗ (A) = A, and therefore µX Γ ((τγ )∗ (A) ∩ A) = µX Γ (A).
This implies, |µX Γ (A) − µX Γ (A)2 | ≤ ε. Since ε > 0 was arbitrarily chosen, we obtain
µX Γ (A)2 = µX Γ (A), and therefore µX Γ (A) ∈ {0, 1}.

2.2 Subsystems and Extensions

To understand the relation between different measure-preserving systems we also
need “structure-preserving maps” between them. This leads to the following defini-
tion.
2.2. SUBSYSTEMS AND EXTENSIONS 25

Definition 2.2.1. An extension (or homomorphism) J : (Y, S) → (X, T ) of

measure-preserving systems is a measure algebra homomorphism J : Σ(Y ) → Σ(X)
such that the diagram
Tγ
Σ(X) / Σ(X)
O O
J J
Sγ
Σ(Y ) / Σ(Y )
commutes for every γ ∈ Γ. In this case, we call (Y, S) a subsystem of (X, T ).
Moreover, J is an isomorphism of measure-preserving systems if, in addition, J is
bijective.
On the level of concrete measure-preserving systems, the “arrows reverse”, and we
speak of factors instead of subsystems:

Definition 2.2.2. A factor map q : (X, τ ) → (Y, σ) of concrete measure-preserving

systems is an element q ∈ M(X, Y ) such that the identity σγ ◦ q = q ◦ τγ holds in
M(X, Y ) for all γ ∈ Γ. In this case, we call (Y, σ) a factor of (X, τ ). Moreover, q
is an isomorphism of concrete measure-preserving systems if, in addition, q is an
isomorphism of probability spaces.
A simple example is the following.

Example 2.2.3. For Γ = Z and α ∈ [0, 1) consider the system (X, τ ) defined
by the measure-preserving map [0, 1) → [0, 1), x 7→ x + α mod 1 from Example
1.1.3 (iii) (cf. Remark 2.1.6). If we pick another β ∈ [0, 1), the addition map
q : [0, 1) → [0, 1), x 7→ x + β mod 1 defines an isomorphism q : (X, τ ) → (X, τ ) of
concrete measure-preserving systems (since addition modulo 1 is commutative).
The following analogue of Proposition 2.1.5 is evident.

Proposition 2.2.4. If q : (X, τ ) → (Y, σ) is a factor map of concrete measure-

preserving systems, then q ∗ : Σ(Y ) → Σ(X), A 7→ q −1 (A) is an extension in the
sense of Definition 2.2.1. If q is an isomorphism, then so is q ∗ .
We now discuss an alternative perspective on subsystems and extensions.

Proposition and Definition 2.2.5. If J : (Y, S) → (X, T ) is an extension of

measure-preserving systems, then Λ := J(Σ(Y )) ⊆ Σ(X) is an invariant sub-σ-
algebra, i.e.,
S
(i) n∈N An ∈ Λ whenever An ∈ Λ for n ∈ N,
(ii) X \ A ∈ Λ for every A ∈ Λ,
(iii) ∅, X ∈ Λ, and
(iv) Tγ (A) ∈ Λ for all A ∈ Λ and γ ∈ Γ.
26 LECTURE 2.

Proof. Parts (i)–(iii) are a direct consequence of Lemma 1.2.6 (i), (ii), and (iv). For
part (iv) consider A = J(B) ∈ J(Σ(Y )) for some B ∈ Σ(Y ). For γ ∈ Γ we then
obtain

Tγ (A) = Tγ (J(B)) = (Tγ ◦ J)(B) = (J ◦ Sγ )(B) = J(Sγ (B)) ∈ J(Σ(Y )).

Remark 2.2.6. Note that the associated invariant sub-σ-algebra completely deter-
mines a subsystem up to an isomorphism: If J1 : (Y1 , S1 ) → (X, T ) and J2 : (Y2 , S2 ) →
(X, T ) are extensions with J1 (Σ(Y1 )) = J2 (Σ(Y2 )), then J2−1 ◦ J1 : Σ(Y1 ) → Σ(Y2 )
defines an isomorphism between (Y1 , S1 ) and (Y2 , S2 ).
We now follow the converse direction: For a given measure-preserving system (X, T )
and an invariant sub-σ-algebra Λ ⊆ Σ(X) we build a subsystem of (X, T ) in the
following way:
(i) The set ΣΛ := {A ⊆ X measurable | [A] ∈ Λ} is a σ-algebra over X and
the restriction (µX )|ΣΛ : ΣΛ → [0, 1] is a probability measure. Thus XΛ :=
(X, ΣΛ , µ|ΣΛ ) is a probability space.
(ii) The map JΛ : Σ(XΛ ) → Σ(X), A 7→ A is a measure algebra homomorphism
with range JΛ (Σ(XΛ )) = Λ.
(iii) For every γ ∈ Γ we obtain a measure algebra automorphism

(TΛ )γ := JΛ−1 ◦ Tγ |Λ ◦ JΛ : Σ(Y ) → Σ(Y ).

This gives us a group representation TΛ : Γ → Aut(Σ(Y )), γ 7→ (TΛ )γ , thus

(X, TΛ ) is a measure-preserving system.
It is then clear by construction that the following holds.

Proposition 2.2.7. Let (X, T ) be a measure-preserving system and Λ ⊆ Σ(X) an

invariant sub-σ-algebra. Then JΛ defines an extension JΛ : (XΛ , TΛ ) → (X, T ) with
JΛ (Σ(XΛ )) = Λ.
Thus, up to isomorphism, subsystems of a given measure-preserving system (X, T )
are in one-to-one correspondence with invariant sub-σ-algebras of the measure alge-
bra Σ(X). We use this observation in the following special case.

Example 2.2.8. Let (X, T ) be any measure-preserving system (X, T ). By Lemma

1.2.6, the set Λ := Σ(X)inv from Definition 2.1.9 is an invariant sub-σ-algebra of
Σ(X), hence defines an extension JΛ : (XΛ , TΛ ) → (X, T ) of measure-preserving
systems. Note that by construction (TΛ )γ = IdΣ(XΛ ) for every γ ∈ Γ, so the dynamics
on XΛ in this case are trivial. We write Jinv := JΛ and (Xinv , Id) := (XΛ , TΛ ).
2.2. SUBSYSTEMS AND EXTENSIONS 27

We also describe subsystems on a functional analytic level: If J : (Y, S) → (X, T )

is an extension of measure-preserving systems, then the induced Koopman opera-
tor UJ : L2 (Y ) → L2 (X) from Proposition 1.3.4 satisfies UJ UTγ = UTγ UJ for each
γ ∈ Γ, cf. Remark 1.3.5. Combining this observation with Proposition 1.3.3, we
immediately obtain the properties of the range of UJ .

Proposition and Definition 2.2.9. Let J : (Y, S) → (X, T ) be an extension of

measure-preserving systems. Then E = UJ (L2 (Y )) is an invariant Markov sub-
lattice of L2 (X), i.e.,
(i) E is a closed linear subspace of L2 (X),
(ii) 1 ∈ E,
(iii) |f |, Re(f ), Im(f ) ∈ E for every f ∈ E, and
(iv) UTγ f ∈ E for every f ∈ E and γ ∈ Γ.

Remark 2.2.10. As a consequence of Theorem 1.3.7, a subsystem is again uniquely

determined up to an isomorphism by the corresponding Markov sublattice. The
details are discussed in Exercise 2.7 below.
We again go the converse direction and construct a subsystem from an invariant
Markov sublattice. We use the following lemma.

Lemma 2.2.11. Assume that (X, T ) is a measure-preserving system and further

that E ⊆ L2 (X) is an invariant Markov sublattice. Then
ΛE := {A ∈ Σ(X) | 1A ∈ E} ⊆ Σ(X)
is an invariant sub-σ-algebra.
Proof. We check the properties listed in Definition 2.2.5. To obtain property (i),
note first that for A, B ∈ ΛE we also have A ∪ B ∈ ΛE since
1A + 1B + |1A − 1B |
1A∪B = sup(1A , 1B ) = ∈ E.
2
We thus obtain A1 ∪ · · · ∪ An ∈ ΛE for all A1 , . . . , An ∈SΛE and n ∈ N by in-
duction. Now if (An )n∈N is a sequence in Λ and A := n∈N An ∈ Σ(X), then
1A = limn→∞ 1A1 ∪···∪An pointwise and hence in L2 (X) by Lebesgue’s theorem. This
shows A ∈ ΛE , and hence establishes property (i) of Definition 2.2.5. For part (ii),
note that if A ∈ ΛE , then 1X\A = 1 − 1A ∈ E, hence X \ A ∈ ΛE . For (iii) observe
that 1∅ = 0 ∈ E (since E is a linear subspace) and 1X = 1 ∈ E, hence ∅, X ∈ ΛE .
Finally, (iv) of Definition 2.2.5 is obvious by part (iv) of Definition 2.2.9.

We again consider the construction in a special case. For this we need the concept
of the fixed space fix(V ) := {f ∈ E | V f = f } of a linear map V : E → E on a
complex vector space E.
28 LECTURE 2.

Example 2.2.12. Let (X, T ) be a measure-preserving system. Then, by the prop-

erties of Markov embeddings (see Definition 1.3.2 and Lemma 1.3.3), the fixed
space
\
fix(UT ) := fix(UTγ ) = {f ∈ L2 (X) | UTγ f = f for every γ ∈ Γ}
γ∈Γ

is an invariant Markov sublattice of L2 (X). Since UTγ 1A = 1Tγ (A) for A ∈ Σ(X) and
γ ∈ Γ, the corresponding sub-σ-algebra Λfix(UT ) is precisely Σ(X)inv from Examples
2.1.9 and 2.2.8.
For a measure-preserving system (X, T ) we now build subsystems from an invari-
ant Markov sublattice E ⊆ L2 (X): First construct the invariant sub-σ-algebra
ΛE := {A ∈ Σ(X) | 1A ∈ E} of Lemma 2.2.11, and from this the extension
JΛE : (XΓE , TΓE ) → (X, T ), see Proposition 2.2.7. We abbreviate (XE , TE ) :=
(XΓE , TΓE ) and JE := JΛE , and show the following result.

Proposition 2.2.13. Let (X, T ) be a measure-preserving system and E ⊆ L2 (X)

an invariant Markov sublattice. Then JE : (XE , TE ) → (X, T ) is an extension with
UJE (L2 (XE )) = E.
Proof. The inclusion “⊆” follows from the facts that E is a closed linear subspace,
UJE is a bounded linear map, and the subspace spanned by (equivalence classes of)
characteristic functions is dense in L2 (XE ) by Lemma 1.3.6.
For the converse inclusion, we make a few observations.
(i) For f, g ∈ E ∩ L2 (X, R) we have
f + g − |f − g| f + g + |f − g|
inf(f, g) = ∈ E and sup(f, g) = ∈ E.
2 2

(ii) Recall that XE is given by the set X, the σ-algebra ΣΓE = {A ⊆ X measurable |
[A] ∈ ΛE } and the restriction µX |ΣΓE : ΣΓE → [0, 1]. Moreover, the Koopman
operator UJE induced by the measure algebra homomorphism JE : Σ(XΛ ) →
Σ(X), A 7→ A (see Proposition 1.3.4) is explicitly given as UJE : L2 (XE ) →
L2 (X), f 7→ f .
(iii) If A ∈ ΣΓE , then X 1A = µX (A) = XE 1A . By the “standard measure-
R R

theoretic procedure” we also obtain that for a function

R f : XR→ [0, ∞) which
is measurable with respect to ΣΓE the integrals X f and XE f agree. In
particular, this implies that if f : X → [0, ∞) is ΣΓE -measurable and square-
integrable on the probability space X = (X, ΣX , µX ), then f is also square-
integrable on XE = (X, ΣΓE , µX |ΣΓE ) (with the same L2 -norm).
Now pick f ∈ E and show that f ∈ J(L2 (XE )). By decomposing into real and
imaginary parts we may assume that f ∈ L2 (X, R). Take a identically named
2.2. SUBSYSTEMS AND EXTENSIONS 29

representative of f . By observations (ii) and (iii) above, it suffices to check that f

is measurable with respect to ΣΓE . So for c ∈ R we show that [f > c] is an element
of ΛE , i.e., 1[f >c] ∈ E. Replacing f by f − c we may assume that c = 0. Now notice
that for every r ∈ R we have
(
1 if r > 0,
lim min(n · max(r, 0), 1) =
n→∞ 0 if r ≤ 0.

This implies 1[f >0] = limn→∞ inf(n sup(f, 0), 1) in L2 (X). By observation (i) above
and the fact that E is closed in L2 (X), we therefore obtain 1[f >0] ∈ E as claimed.

As a consequence, we obtain the following important functional analytic character-

ization of ergodic systems.

Corollary 2.2.14. Let (X, T ) be a measure-preserving system. Then the linear hull
lin{1A | A ∈ Σ(X)inv } is dense in fix(UT ). In particular, the following assertions
are equivalent.
(a) (X, T ) is ergodic.
(b) fix(UT ) = {c1 | c ∈ C}.
Proof. By Example 2.2.12 and Proposition 2.2.13 we obtain that the Markov em-
bedding UJinv : L2 (Xinv ) → L2 (X) has the range UJinv (L2 (Xinv )) = fix(UT ). Since
lin{1B | B ∈ Σ(Xinv )} is dense in L2 (Xinv ) by Lemma 1.3.6, and UJinv is a bounded
linear operator, the subspace

lin{UJinv 1B | B ∈ Σ(Xinv )} = lin{1Jinv (B) | B ∈ Σ(Xinv )} = lin{1A | A ∈ Σ(X)inv }

is dense in fix(UT ).

The last result of this lecture allows a way to build invariant Markov sublattices
(and therefore subsystems of a measure-preserving system) from certain “invariant
subalgebras”.

Proposition 2.2.15. Let (X, T ) be a measure-preserving system. Assume that F ⊆

L∞ (X) is a linear subspace with
(i) f · g ∈ F for all f, g ∈ F ,
(ii) f ∈ F for all f ∈ F ,
(iii) 1 ∈ F , and
(iv) UTγ f ∈ F for all f ∈ F and γ ∈ Γ.
Then the closure F in L2 (X) is an invariant Markov sublattice.
30 LECTURE 2.

α


For the proof we use the following identity. Recall that the binomial coefficients k
for α ∈ C are
   
α α · (α − 1) · · · (α − k + 1) α
:= for k ∈ N, and := 1.
k k! 0

Lemma 2.2.16. Let X be a probability space and f ∈ L∞ (X) with ∥f ∥∞ ≤ 1. Then

n 1
2 (−1)k (1 − |f |2 )k
X
|f | = lim
n→∞
k=0
k

in L∞ (X).
P∞ α
k
Proof. For α ∈ (0, ∞) series k=0 k z converges absolutely and uniformly to
α
(1 + z) for z ∈ C with |z| ≤ 1 (see, e.g., [AE05, Theorem V.3.10]). Observing that
1
|z| = (1 − (1 − |z|2 )) 2 for z ∈ C, we therefore find for ε > 0 some n0 ∈ N with

Xn 1
|z| − 2 (−1)k (1 − |z|2 )k ≤ ε

k=0
k

all z ∈ C with |z| ≤ 1 and for every n ≥ n0 . This implies

n 1
2 (−1)k (1 − |f |2 )k ≤ ε1
X
|f | −
k=0
k

almost everywhere for n ≥ n0 which yields the claim.

Proof of Proposition 2.2.15. We check properties (i) - (iv) of Definition 2.2.9. As the
closure of a linear subspace is again a linear subspace, we obtain (i). Property (ii) is
clear by our assumption (iv) of Proposition 2.2.15. To obtain Definition 2.2.9 (iii),
take f ∈ F with ∥f ∥∞ ≤ 1, then |f |2 = f ·f ∈ F by properties (i) and (ii). Since L∞ -
convergence implies L2 -convergence, Lemma 2.2.16 shows |f | ∈ F . If f ∈ F , we find
a sequence (fn )n∈N converging in L2 to f , and this implies |f | = limn→∞ |fn | ∈ F .
Finally, part (iv) of Definition 2.2.9 is an easy consequence of Proposition 2.2.15 (iv)
and continuity of UTγ : L2 (X) → L2 (X) for γ ∈ Γ.

To summarize this section, we can define subsystems of a measure-preserving system

(X, T ) in three different ways:
(i) via extensions J : (Y, S) → (X, T ),
(ii) via invariant sub-σ-algebras Λ ⊆ Σ(X), and
(iii) via invariant Markov sublattices E ⊆ L2 (X).
2.3. COMMENTS AND FURTHER READING 31

2.3 Comments and Further Reading

There is a plethora of textbooks on ergodic theory focusing on various different as-
pects. A few classical works related to the topics of this course are [Hal56], [Wal75],
and [Fur14]. We also mention [Gla03], [EW11] and [VO16] for contemporary intro-
ductions, and [DNP87] and [EFHN15] for an operator theoretic approach to ergodic
theory. A large part of the material of this course is based on the contents of
these monographs and two previous editions of the Internet Seminar ([EFHN09]
and [EF19]).
We remark that there are numerous further equivalent descriptions of subsystems of
a given measure-preserving systems, see [EFHN15, Theorem 13.20]. Studying exten-
sions of measure-preserving systems will play a crucial part later in this course.
32 LECTURE 2.

2.4 Exercises
Exercise 2.1. For an index set I ̸= ∅ let τi : Xi → Yi be measure-preserving maps
between probability spaces for every i ∈ I. Show that the map
Y Y Y
τi : Xi → Yi , (xi )i∈I 7→ (τi (xi ))i∈I
i∈I i∈I i∈I

between the product measure spaces is also measure-preserving.

Exercise 2.2. With the notation from Exercise 1.6 show that for a measure-preserving
system (X, T ) the following assertions are equivalent.
(a) (X, T ) is ergodic.
(b) X = sup{Tγ (A) | γ ∈ Γ} in Σ(X) for every A ∈ Σ(X) \ {∅}.

Exercise 2.3. Show that for a measure-preserving map σ : X → X on a probability

space X the following assertions are equivalent.
(a) σ ∗ (A) = A for A ∈ Σ(X) implies µX (A) ∈ {0, 1}.
(b) σ −1 (A) = A for A ⊆ X measurable implies µX (A) ∈ {0, 1}.
Hint: For a representative of A as in (a) consider ∞
T S∞ −k
n=0 k=n σ (A).

Exercise 2.4. Assume that Γ is a finite abelian group and consider the Bernoulli
shift (X Γ , τ ) from Example 2.1.8. Show that the system (X Γ , τ ∗ ) is ergodic precisely
when Σ(X) = {∅, X}.

Exercise 2.5. For Γ = Z consider the Bernoulli shift ({0, 1}Z , τ ) where the mea-
sure on {0, 1} is given by the probability vector ( 21 , 12 ), see Examples 1.1.3 (ii) and
2.1.8. Let further ([0, 1]2 , σ) be the measure-preserving system defined by the baker’s
transformation from Exercise 1.2 (cf. Remark 2.1.6). Then the map
X ∞ ∞ 
2 xn−1 X x−n
q : {0, 1} → [0, 1] , (xn )n∈Z 7→
Z
n
,
n=1
2 n=1
2n

defines an isomorphism between ({0, 1}Z , τ ) and ([0, 1]2 , σ). P

Hint: Recall that every x ∈ [0, 1) has a base 2 expansion x = ∞ n=1 xn 2
−n
for some
sequence (xn )n∈N ∈ {0, 1} . Moreover, this representation becomes unique if we
N

exclude sequences (xn )n∈N such that there is N ∈ N with xn = 1 for all n ≥ N , see,
e.g., [AE05, Topic 6 of Section II.7]).

Exercise 2.6. Show the following assertions.

(i) For every measure-preserving system (X, T ) on a Lebesgue space X there is a
unique concrete measure-preserving system (X, τ ) with (X, T ) = (X, τ ∗ ).
2.4. EXERCISES 33

(ii) If (X, τ ) and (Y, σ) are concrete measure-preserving systems on Lebesgue

spaces X and Y and J : (Y, σ ∗ ) → (X, τ ∗ ) is an extension of measure-preserving
systems, then there is a unique factor map q : (X, τ ) → (Y, σ) with J = q ∗ .
Moreover, q is an isomorphism precisely when J is an isomorphism.

Exercise 2.7. Show the following assertions.

(i) Let X be a probability space and U : Γ → Aut(L2 (X)) a representation of Γ
as Markov automorphisms. Show that there is a unique measure-preserving
system (X, T ) with U = UT .
(ii) Let (Y, S) and (X, T ) be measure-preserving systems and V : L2 (Y ) → L2 (X)
a Markov embedding with V USγ = UTγ V for all γ ∈ Γ. Then there is a unique
extension J : (Y, S) → (X, T ) with V = UJ . Moreover, J is an isomorphism
precisely when V is bijective.
(iii) Let J1 : (Y1 , S1 ) → (X, T ) and J2 : (Y2 , S2 ) → (X, T ) be extensions of measure-
preserving systems such that UJ1 (L2 (Y1 )) = UJ2 (L2 (Y2 )). Then there is a
unique isomorphism I : (Y1 , S1 ) → (Y2 , S2 ) of measure-preserving systems with
J2 = J1 ◦ I.

Exercise 2.8. For probability spaces X and Y let U : L2 (Y ) → L2 (X) be a linear

isometry with the following properties.
(i) U (L∞ (Y )) ⊆ L∞ (X),
(ii) U (f · g) = U f · U g for all f, g ∈ L∞ (Y ),
(iii) U f = U f for every f ∈ L∞ (Y ),
(iv) U 1 = 1.
Show that U is a Markov embedding.
34 LECTURE 2.
Bibliography

[AE05] H. Amann and J. Escher. Analysis I. Birkhäuser, 2005.

[Bar95] R.G. Bartle. The Elements of Integration and Lebesgue Measure. Wiley,
1995.
[BBP16] H. Bercovici, A. Brown, and C. Pearcy. Measure and Integration.
Springer, 2016.
[Bil95] P. Billingsley. Probability and Measure. Wiley, 3rd edition, 1995.
[Con85] J.B. Conway. A Course in Functional Analysis. Springer, 1985.
[DNP87] R. Derndinger, R. Nagel, and G. Palm. Ergodic Theory in the Perspective
of Functional Analysis. Unpublished book manuscript, 1987.
[EF19] T. Eisner and B. Farkas. 22nd Internet Seminar: Ergodic Theorems,
2018–2019.
[EFHN09] T. Eisner, B. Farkas, M. Haase, and R. Nagel. 12th Internet Seminar:
Ergodic Theory – An Operator Theoretic Approach, 2008–2009.
[EFHN15] T. Eisner, B. Farkas, M. Haase, and R. Nagel. Operator Theoretic Aspects
of Ergodic Theory. Springer, 2015.
[EW11] M. Einsiedler and T. Ward. Ergodic Theory with a view towards Number
Theory. Springer, 2011.
[Fur14] H. Furstenberg. Recurrence in Ergodic Theory and Combinatorial Num-
ber Theory. Princeton University Press, 2014.
[Gla03] E. Glasner. Ergodic Theory via Joinings. American Mathematical Soci-
ety, 2003.
[Haa14] M. Haase. Functional Analysis: An Elementary Introduction. American
Mathematical Society, 2014.
[Hal56] P. Halmos. Lectures on Ergodic Theory. American Mathematical Society,
1956.
[HS65] E. Hewitt and K. Stromberg. Real and Abstract Analysis. Springer, 1965.

39
40 BIBLIOGRAPHY

[JT22] A. Jamneshan and T. Tao. An uncountable Mackey-Zimmer theorem.

Studia Math., 266(3):241–289, 2022.
[JT23a] A. Jamneshan and T. Tao. Foundational aspects of uncountable mea-
sure theory: Gelfand duality, Riesz representation, canonical models, and
canonical disintegration. Fund. Math., 261(1):1–98, 2023.
[JT23b] A. Jamneshan and T. Tao. An uncountable Moore-Schmidt theorem.
Ergodic Theory Dynam. Systems, 43(7):2376–2403, 2023.
[Kal97] O. Kallenberg. Foundations of Modern Probability. Springer, 1997.
[Koo31] B. O. Koopman. Hamiltonian systems and transformation in Hilbert
space. Proc. Natl. Acad. Sci. U.S.A., 18:315–318, 1931.
[Ped89] G. K. Pedersen. Analysis Now. Springer, 1989.
[Poi90] H. Poincaré. Sur le problème des trois corps et les équations de la dy-
namique. Acta Math., 13:1–270, 1890.
[Ran02] I. K. Rana. An Introduction to Measure and Integration. American
Mathematical Society, 2nd edition, 2002.
[Rud87] W. Rudin. Real and Complex Analysis. McGraw-Hill, 1987.
[Tao11] T. Tao. An Introduction to Measure Theory. American Mathematical
Society, 2011.
[vN32] J. von Neumann. Einige Sätze über messbare Abbildungen. Ann. Math.
(2), 33:574–586, 1932.
[VO16] M. Viana and K. Oliveira. Foundations of Ergodic Theory. Cambridge
University Press, 2016.
[Wal75] P. Walters. An Introduction to Ergodic Theory. Springer, 1975.