468 XI Uniform Distribution and Ergodic Theory
and hence
τ(∑x)− 1
k= 0
f ̄M(Tkx)≤
τ(∑x)− 1
k= 0
f ̃(Tkx)+τ(x)ε.
To estimate the sum
∑L− 1
k= 0 f ̄M(T
kx)for anyL>N, we partition it into blocks of
the form
τ(∑y)− 1
k= 0
f ̄M(Tky)
and a remainder block. More precisely, define inductively
n 0 (x)= 0 , nk(x)=nk− 1 (x)+τ(Tnk−^1 x)(k= 1 , 2 ,...)
and definehbynh(x)<L≤nh+ 1 (x).Then
n (^1) ∑(x)− 1
k= 0
f ̄M(Tkx)≤
n (^1) ∑(x)− 1
k= 0
f ̃(Tkx)+τ(x)ε,
n (^2) ∑(x)− 1
k=n 1 (x)
f ̄M(Tkx)≤
n (^2) ∑(x)− 1
k=n 1 (x)
f ̃(Tkx)+τ(Tn^1 x)ε,
···
nh∑(x)− 1
k=nh− 1 (x)
f ̄M(Tkx)≤
nh∑(x)− 1
k=nh− 1 (x)
f ̃(Tkx)+τ(Tnh−^1 x)ε.
Sincenh(x)<L, we obtain by addition
nh∑(x)− 1
k= 0
f ̄M(Tkx)≤
nh∑(x)− 1
k= 0
f ̃(Tkx)+Lε.
On the other hand, sinceL≤nh+ 1 (x)≤nh(x)+N,wehave
L∑− 1
k=nh(x)
f ̄M(Tkx)≤NM.
Sincef ̃≥0, it follows that
L∑− 1
k= 0
f ̄M(Tkx)≤
L∑− 1
k= 0
f ̃(Tkx)+Lε+NM.
Dividing byLand integrating overX, we obtain
∫
X
f ̄Mdμ≤
∫
X
f ̃dμ+ε+NM/L,