62.2. KOLMOGOROV ČENTSOV CONTINUITY THEOREM 2053

for large enough n ≥ M. Just pick the first n such that T 2−(n+1) < |d−d′| . Then from62.2.4, ∣∣∣∣X (d′)−X (d)

∣∣∣∣ ≤ (2T−γ

1−2−γ

)(T 2−(n+1)

≤(

2T−γ

1−2−γ

)(∣∣d−d′∣∣)γ

Now [0,T ] is covered by 2M intervals of length T 2−M and so for any pair d,d′ ∈ D,∣∣∣∣X (d)−X(d′)∣∣∣∣≤C

∣∣d−d′∣∣γ

where C is a suitable constant depending on 2M .For γ ≤ 1, you can show, using convexity arguments, that it suffices to have C =(

2T−γ

1−2−γ

)1/γ (2M)1−γ

. Of course the case where γ > 1 is not interesting because it wouldresult in X being a constant.

The following is the amazing Kolmogorov Čentsov continuity theorem [78].

Theorem 62.2.2 Suppose X is a stochastic process on [0,T ] . Suppose also that there existsa constant, C and positive numbers, α,β such that

E(||X (t)−X (s)||α

)≤C |t− s|1+β (62.2.5)

Then there exists a stochastic process Y such that for a.e. ω, t → Y (t)(ω) is Hölder con-tinuous with exponent γ < β

αand for each t, P([||X (t)−Y (t)||> 0]) = 0. (Y is a version

of X .)

Proof: Let rmj denote j

( T2m

)where j ∈ {0,1, · · · ,2m} . Also let Dm =

{rm

j

}2m

j=1and

D = ∪∞m=1Dm. Consider the set,

[||X (t)−X (s)||> δ ]

By 62.2.5,

P([||X (t)−X (s)||> δ ])δα ≤

∫[||X(t)−X(s)||>δ ]

||X (t)−X (s)||α dP

≤ C |t− s|1+β . (62.2.6)

Letting t = rkj+1, s = rk

j ,and δ = 2−γk where

γ ∈(

0,β

α

),

this yields

P([∣∣∣∣∣∣X (rk

j+1

)−X

(rk

j

)∣∣∣∣∣∣> 2−γk])≤C2αγk

(T 2−k

)1+β

62.2. KOLMOGOROV CENTSOV CONTINUITY THEOREM 2053for large enough n > M. Just pick the first n such that T2-"*+) < |d—d'|. Then from62.2.4,2T-Y Yoe —(n+1)(5) (72 )2T-Y N\Y< (S55) e-atpNow [0,7] is covered by 2” intervals of length T2~™ and so for any pair d,d’ € D,IIx (4’) — x (4)]|lAI|X (4) -X (d’)|| <c|d—a'"where C is a suitable constant depending on 2”. §For y < 1, you can show, using convexity arguments, that it suffices to have C =1-2-7result in X being a constant.The following is the amazing Kolmogorov Centsov continuity theorem [78].1 \ 1/7 - ; . , ;( 277 ) (2” ) ’ Of course the case where y > 1 is not interesting because it wouldTheorem 62.2.2 Suppose X is a stochastic process on (0,T]. Suppose also that there existsa constant, C and positive numbers, 0, B such thatE (||X (t) —X (s)||") <€|r—s|!*8 (62.2.5)Then there exists a stochastic process Y such that for a.e. @,t — Y (t)(@) is Hélder con-tinuous with exponent Y < B and for each t, P({||X (t) —Y (t)|| > 0]) =0. (Y is a versionof X.)gmProof: Let 17" denote i(#) where j € {0,1,---,2}. Also let Dn = {rv} andD=U”m=1Dj. Consider the set,[||X (t) —X (s)|] > 6]By 62.2.5,P([||X (t) —X (s)|| > 6]) 8°IA||X (t) —X (s)||*aPFiresC\tr—s|'F. (62.2.6)IALetting t = réi+ 5 = rand 5 =2- whereve(o2)P([IPe (a) x @)]>24)) seam (2)this yields