MATH 275B notes 182

27. W EAK AND STRONG MIXING

In this section we will expand on the notions of ergodicity and recurrence and show
that these are part of a hierarchy of mixing properties of measure preserving systems.
Throughout we assume that the underlying measure is a probability.
27.1 Definitions and examples.

In order to introduce the desired notions in proper context, we start by recalling that a
m.p.s. (X , G , µ, j) with µ(X ) = 1 is
(0) recurrent if
@B P G : µ( B) ° 0 ñ Dn • 1 : µ( j´n ( B) X B) ° 0 (27.1)
(1) ergodic if
n´1
1 ÿ
@A, B P G : µ j´k ( A) X B ›Ñ µ( A)µ( B) (27.2)
n nÑ8
k =0
Recurrence follows automatically for finite measure spaces from the Poincaré Recur-
rence Theorem. For (27.2) we note that the Pointwise/Mean Ergodic Theorem gives
n´1 ✓ n´1 ◆
1 ÿ ´k 1 ÿ k
µ j ( A) X B = E 1B 1A ˝ j ›Ñ E(1B 1̄ A ) (27.3)
n n nÑ8
k =0 k =0
≥
and ergodicity then forces 1̄ A = 1 A dµ = µ( A) implying (27.2). Conversely, if A is an
invariant event then (27.2) with B := A forces µ( A) = µ( A)2 ; i.e., µ( A) P t0, 1u.
We now introduce names for stronger versions of the convergence in (27.2):
Definition 27.1 (Weak and strong mixing) A m.p.s. (X , G , µ, j) with µ(X ) = 1 is
(2) weakly mixing if
1 ÿ ˇˇ ˇ
n´1
ˇ
@A, B P G : ˇ µ j´k ( A) X B ´ µ( A)µ( B)ˇ ›Ñ 0 (27.4)
n nÑ8
k =0
(3) strongly mixing (or sometimes simply mixing) if
@A, B P G : µ j´k ( A) X B ›Ñ µ( A)µ( B) (27.5)
nÑ8

For practical use it is useful to note that we only need to check the above properties
for a suitable generating class of sets:
Lemma 27.2 Let S is a semi-algebra such that G Ñ s(S). If any of (27.2), (27.4) and (27.5)
holds for all A, B P S then it holds for all A, B P G .

Proof. Let S be a semialgebra and let A be the set of all finite disjoint unions of elements
from S . Then A is an algebra. This shows that if (27.2), (27.4) and (27.5) hold for all
A, B P S , then in holds for all A, B P A. Next, by the construction of µ on s(A) via the
outer measure, for each A, B P s (A) and each e ° 0 there exists A1 , B1 P A such that
µ( A4 A1 ) † e and µ( B4 B1 ) † e. Then |µ( j´k ( A) X B) ´ µ( j´k ( A1 ) X B1 )| † 2e and

Preliminary version (subject to change anytime!) Typeset: April 18, 2025

183 MATH 275B notes

|µ( A)µ( B) ´ µ( A1 )µ( B1 )| † 2e. Using this we extend the corresponding limit claim to
all A, B P s(A), which contains G . ⇤
Yet another simple fact is the invariance under isomorphisms:
Lemma 27.3 The properties (1), (2) and (3) are isomorphism invariants: If (X , G , µ, j) and
(X 1 , G 1 , µ1 , j1 ) are m.p.s. F : X Ñ X 1 is a bi-measurable bijection such that µ1 = µ ˝ F´1
and F ˝ j = j1 ˝ F, then if any of (1-3) above hold for one m.p.s. then it holds for the other
m.p.s. as well.

Proof. This is immediate from the assumptions. ⇤

We proceed by examples of strongly mixing m.p.s.:
Lemma 27.4 Left-shifts of i.i.d. sequences are strongly mixing.

Proof. Let tXk uk•0 be i.i.d.îrandom variables defined on a suitable product space and
let j be the left shift. Since k•1 s( X0 , X1 , . . . , Xk ) is an algebra, it suffices to prove (27.5)
for A, B P s( X0 , X1 , . . . , Xk ). But this follows by noting that j´n ( A) P s ( Xn , Xn+1 , . . . )
and so j´n ( A) and B are independent once n ° k. ⇤

Corollary 27.5 The “continued fractions” m.p.s. (Example 20.10) and “Baker’s transform”
m.p.s. (Example 20.12) are strongly mixing.

Proof. Since both of these systems are isomorphic to i.i.d. sequences with the left shift,
this follows from Lemmas 27.3 and 27.4. ⇤
Another example of a strongly mixing system is an irreducible, aperiodic countable-
state Markov chain which we will discuss in the upcoming lectures. In both cases we
can infer strong mixing by showing a yet stronger property. We start with a definition:
Definition 27.6 (Tail triviality) A m.p.s. (X , G , µ, j) with µ(X ) = 1 is said to be

(4) tail trivial if µ is trivial on the tail s-field

£ (
T := j´n ( A) : A P G (27.6)
n•1

in the sense

@B P T : µ( B) P t0, 1u (27.7)

We say that j is exact if T = tH, X u. (This implies tail triviality.)

Lemma 27.7 Let (X , G , µ, j) be a m.p.s. with µ(X ) = 1. Then j is tail-trivial if and only if

ˇ ˇ
ˇ ˇ
@B P G : sup ˇ µ j´n ( A) X B ´ µ( A)µ( B)ˇ ›Ñ 0 (27.8)
nÑ8
APG

Preliminary version (subject to change anytime!) Typeset: April 18, 2025

MATH 275B notes 184

Proof. Suppose (X , G , µ, j) is tail trivial. Denote Tn := tj´n ( A) : A P G u and note

that Tn Ó T . Then
ˇ ˇ ˇ ⇣ ⌘ˇ
ˇ ˇ ˇ ˇ
ˇ µ q ´n ( A) X B ´ µ( A)µ( B)ˇ = ˇ E 1 j´n ( A) 1B ´ µ( B) ˇ
ˇ ⇣ ⌘ˇ ˇ ˇ
ˇ ˇ
= ˇ E 1 j´n ( A) E(1B |Tn ) ´ µ( B) ˇ § EˇE(1B |Tn ) ´ µ( B)ˇ. (27.9)

The Backward Lévy Theorem implies E (1B |Tn ) Ñ E (1B |T ) (both a.s. and in L1 ) and
by tail-triviality E (1B |T ) = µ( B) a.s. Hence, the right-hand side converges to zero uni-
formly in A, proving (27.8).
To prove the converse, pick B P T . Then for each n • 1 there exists An P G such that
B = j´n ( An ). Using uniformity in A of the limit (27.8), this along with µ( An ) = µ( B)
implies
µ( B) ´ µ( B)2 = µ j´n ( An ) X B ´ µ( An )µ( B) ›Ñ 0 (27.10)
nÑ8
i.e., µ( B)2 = µ( B). Hence µ is trivial on T . ⇤

27.2 Separating the mixing concepts.

It is easy to check that (4) ñ (3) ñ (2) ñ (1) ñ (0). Our next goal is to demonstrate
that each concept is different from the others. The following example shows that tail
triviality is stronger than strong mixing:
∞
Lemma 27.8 Let tXn unPZ be Bernoulli(1/2) and let Zn := m•0 2´m+1 Xn´m . Then the left
shift of tZn unPZ is strongly mixing but not tail trivial, due to
£
T := s( Zn , Zn+1 , . . . ) = s( Xn : n P Z ) = s( Zn : n P Z ) (27.11)
n•0

Proof. Since tXk uk§n is determined by the binary expansions of Zn , for each n P Z we
have s( Zn ) Ö s( Xk : k § n) and so
@n • 0 : s( Zn , Zn+1 , . . . ) Ö s( Xn : n P Z ) = s( Zn : n P Z ) (27.12)
Hence also T = s( Zn : n P Z ).
To see that the m.p.s. is strongly mixing, note that tZk • `2´n u, where `, n P N is
independent of tX j : j § ´nu and thus of tZj : j § ´nu. As Zn only depends on tXk : k §
nu, for any choices of m P N, `´m , . . . , `m P N and `1´m , . . . , `1m P N, the events
m
£ m
£
A := tZi • `i 2´n u and B := tZi • `1i 2´n u (27.13)
i =´m i =´m

obey
@k ° n + 2m : µ j´k ( A) X B = µ( A)µ( B) (27.14)
implying (27.5). Since the above events form a semialgebra that generates G , Lemma 27.2
shows that the left shift of tZn unPZ is strongly mixing. ⇤
Next we give an example of an ergodic, but non-mixing transformation:

Preliminary version (subject to change anytime!) Typeset: April 18, 2025

185 MATH 275B notes

Lemma 27.9 Irrational rotations of the unit circle are ergodic but not strongly mixing.

Proof. Ergodicity follows from the observation that all invariant functions are constant,
as established in the proof of Weil’s Equidistribution Theorem (Theorem 21.5). To inval-
idate strong mixing, let A = B := 1[0,1/4) . As ergodicity forces tx + an mod 1 : n P Nu to
be dense in [0, 1] for at least one x P [0, 1), there exists nk Ñ 8 such that nk a mod 1 Ñ 1/2.
But then j´nk ( A) is an interval eventually contained in [1/3, 2/3] and so j´nk ( A) X B = H
for k sufficiently large. As µ( A) = µ( B) = 1/4, the transformation is not strongly mixing.
(An enhanced version of this argument shows that weak mixing fails as well.) ⇤
TO BE CONTINUED . . .

Preliminary version (subject to change anytime!) Typeset: April 18, 2025