4 Applications 479
since
B\(C∩σ−nC)⊆(B\C)∪(B\σ−nC)⊆(B\C)∪σ−n(B\C),
and similarly
μ((C∩σ−nC)\B)≤ 2 μ(C\B).
Hence
|μ(B)−μ(C∩σ−nC)|≤μ(B\(C∩σ−nC))+μ((C∩σ−nC)\B)< 2 ε.
Thus
0 ≤μ(B)−μ(B)^2 =μ(B)−μ(C∩σ−nC)+μ(C∩σ−nC)−μ(B)^2
< 2 ε+μ(C)^2 −μ(B)^2 < 4 ε.
Sinceεis arbitrary, we conclude thatμ(B)=μ(B)^2. Henceμ(B)=0or1,andσis
ergodic.
Similarly, ifYis the set of all infinite sequencesy=(y 1 ,y 2 ,y 3 ,...)withyi∈A
for everyi∈N, then theone-sided Bernoulli shift B+p 1 ,...,pr,i.e.themapτ:Y→Y
defined byτy=y′,wherey′i=yi+ 1 for everyi∈N, is a measure-preserving transfor-
mation of the analogously constructed probability space (Y,B,μ). It should be noted
that, althoughτY=Y,τis not invertible. In the same way as for the two-sided shift,
it may be shown that the one-sided Bernoulli shiftB+p 1 ,...,pris always ergodic.
(iv) An example of some historical interest is the ‘continued fraction’ orGaussmap.
LetX=[0,1] be the unit interval andT:X→Xthe map defined (in the notation of
§1) by
Tξ={ξ−^1 } ifξ∈( 0 , 1 ),
=0ifξ=0or1.
ThusTacts as the shift operator on the continued fraction expansion ofξ:if
ξ=[0;a 1 ,a 2 ,...]=
1
a 1 +
1
a 2 +···
,
thenTξ=[0;a 2 ,a 3 ,...]. (In the terminology of Chapter IV, the complete quotients
ofξareξn+ 1 = 1 /Tnξ.)
It is not difficult to show thatTis a measure-preserving transformation of the prob-
ability space (X,B,μ), whereBis the family of Borel subsets ofX=[0,1] andμ
is the ‘Gauss’ measure defined by
μ(B)=(log 2)−^1
∫
B
( 1 +x)−^1 dx.
It may further be shown thatTis ergodic. Hence, by Birkhoff’s ergodic theorem, iff
is an integrable function on the intervalXthen, for almost allξ∈X,
nlim→∞n−^1
n∑− 1
k= 0
f(Tkξ)=(log 2)−^1
∫
X
f(x)( 1 +x)−^1 dx.