A short note on the proof of the ergodic theorem

Norbert Kusolitsch

15. 9. 2013

Abstract
It is shown that Birkhoff’s Pointwise Ergodic Theorem follows al-
most immediately from a slight generalization of the Maximal Ergodic
Inequality and the proof of this generalization is simplified.

Let (Ω, S, P ) be a probability space, T : Ω → Ω a measure-preserving

transformation, X an integrable random variable, I the σ-field of invariant
k−1
X ◦ T i , Mm := max Sk , M := sup Sk , U (X) := lim sup Skk
P
sets, Sk :=
i=0 1≤k≤m k∈N k
Sk
and L(X) := lim inf k .
k
With these notations the Mean Ergodic Inequality states
Z
X dP ≥ 0 . (1)
[ M >0]

Therefrom easily follows

Corollary (Generalized Maximal Ergodic Inequality) If Y is an in-
variant integrable random variable and A ∈ I, then
Z Z
X dP ≥ Y dP . (2)
   
Sk Sk
A∩ sup k
>Y A∩ sup k
>Y
k∈N k∈N

Proof : Just apply (1) to 1A (X − Y) and take into consideration that
k−1 k−1
 
1A (X − Y ) ◦ T i > 0 = A ∩ sup k1 X ◦ T i > Y because of
P P
sup
k∈N i=0 k∈N i=0
Y ◦ T i = Y and 1A ◦ T i = 1A ∀ i ∈ N0 .  
1 Sk 1
 
Now An := U (X) > E(X|I) + n is a subset of sup k > E(X|I) + n
k∈N
and obviously invariant for all n ∈ N and so the corollary above applied to
A := An and Y := E(X|I) + n1 yields
Z Z   Z
1 1
X dP ≥ E(X|I) + dP = X dP + P (An ) (3)
n n
An An An

1
thus implying P (An ) = 0 i.e. U (X) ≤ E(X|I) + n1 P -a.s. ∀ n ∈ N.
Therefrom follows U (X) ≤ E(X|I) P -a.s. immediately. Replacing X by
−X above gives L(X) ≥ E(X|I) P -a.s. Thus the Pointwise Ergodic The-
orem is proved.
Keane - Petersen [2] used a similar modification of the Maximal Ergodic
Inequality especially
Z Z
X dP ≥ Y dP. (4)
   
Sk Sk
sup k
>Y sup k
>Y
k∈N k∈N

By replacing Y through a truncated version of U (X + ) respectively U (X − ),

by the inequalities U (X)+ ≤ U (X + ), U (X)− ≤ U (X − ) and the implicit
use of the Monotone Convergence Theorem they proved the integrability of
U (X) in a first step (a necessity that is avoided in the approach described
above). Then they applied (4) once more to Y := U (X) − ε, ε > 0 and
used the Monotone Convergence Theorem a second time to get the final
result i.e. the Pointwise Ergodic Theorem.
Keane and Petersen proved their inequality directly and not by means of
inequality (1). But their approach also provides an interesting alternative to
the elegant proof of this inequality given by Garsia [1] or the classical proof
due to F. Riesz [3]. It can be slightly simplified as one need not restrict the
proof to bounded random variables at first. This will be shown now.
Let m ∈ N and Bm := [ Mm > 0 ] then
−|X| ≤ X ≤ X 1Bm . (5)
The left inequality above is trivial, the right one follows from the fact that
ω∈/ Bm implies X(ω) = S1 (ω) ≤ 0 = X 1Bm (ω).
T being measure-preserving furthermore we have
Z Z
X 1Bm ◦ T i dP = X 1Bm dP ∀ i ∈ N0 . (6)

n−1
X 1Bm ◦ T i then we want to show
P
Let n ∈ N be arbitrary and Rn :=
i=0
n+m−1
X
Rn+m ≥ − |X| ◦ T i . (7)
i=n

IfT iω ∈
/ Bm ∀ 0 ≤ i ≤ n − 1 then Rn (ω) = 0 and so (5) implies (7).
Otherwise there exists an index i1 := i1 (ω) := min{ i ≤ n − 1 : T i ω ∈ Bm }.
T i1 ω being in Bm there exists as well m1 := m1 (ω) ≤ m : Sm1 (T i1 ω) > 0.
and (5) implies
X1 −1
i1 +m
X 1Bm ◦ T i ω ≥ Sm1 (T i1 ω) > 0. (8)
i=i1

2
Now define indices ij := ij (ω) := min{ ij−1 +mj−1 ≤ i ≤ n−1 : T i ω ∈ Bm }
and their corresponding mj -s recurrently as long as possible. Let us assume
that i1 < . . . < ik are these indices, then according to the construction
X 1Bm ◦ T i ω = 0 for all i < i1 , ij + mj ≤ i < ij+1 or ik + mk ≤ i < n. This
together with (8) and (5) induces

X Xj −1
k ij +m n+m−1
X
i
Rn+m (ω) = X 1Bm ◦ T ω + X 1Bm ◦ T i ω
j=1 i=ij i=(ik +mk )∨n
k
X n+m−1
X n+m−1
X
ij i
≥ Smj (T ω) − |X| ◦ T ω ≥ − |X| ◦ T i ω. (9)
j=1 i=(ik +mk )∨n i=n

So we have proved (7) for every ω ∈ Ω and arbitrary n ∈ N. Therefor

integrating both sides of (7) and taking into account (6) implies

m E |X|
Z Z
(n + m) X dP ≥ −m E |X| ∀ n ∈ N ⇒ X dP ≥ − lim = 0.
n→∞ n + m
Bm Bm

R (10)
Finally from X dP ≥ 0 ∀ m ∈ N and Bm % [ M > 0 ] inequality (2)
Bm
follows in the usual way by means of the Dominated Convergence Theorem.

References
[1] Garsia, A. (1965). A simple proof of Eberhard Hopf ’s maximal ergodic
theorem. J. Math and Mech, 1965, Vol. 14, pp 381-382.

[2] Keane, M. - Petersen, K. (2006). Easy and nearly simultaneous proofs

of the Ergodic Theorem and Maximal Ergodic Theorem. IMS Lecture
Notes-Monograph Series Dynamics & Stochastics, 2006, Vol. 48, pp
248-251.

[3] Riesz, F. (1945). Sur la thorie ergodique. Comment. Math. Helv, 1945,
Vol. 17, pp 221-239.