2283

Lemma 66.0.13 Let Y ∈ G . Then for any stopping time τ,∫ t∧τ

0(Y,dMτ p) =

∫ t

0

(X[0,τ]Y,dMτ p

)=∫ t

0

(Y,dMτ∧τ p

)for ω off some set of measure zero.

Proof: From Lemma 66.0.9, there exists a sequence of elementary functions Y n suchthat t →

∫ t0 (Y

n,dM) converges uniformly to t →∫ t

0 (Y,dMτ p) on [0,T ] for each ω /∈ N, aset of measure zero. Then∫ t∧τ

0(Y,dMτ p) = lim

n→∞

∫ t∧τ

0(Y n,dMτ p)

= limn→∞

∫ t

0

(X[0,τ]Y

n,dMτ p)=∫ t

0

(X[0,τ]Y,dMτ p

)The last claim needs a little clarification. As shown in the above discussion proving Lemma66.0.12, while X[0,τ]Y n is no longer obviously an elementary function due to the fact thatτ has values which are not partition points, it is still the limit of a sequence of elementaryfunctions X[0,τk]Y

n and so the integral makes sense. Then from the inequality of Lemma66.0.9,

E

(∣∣∣∣∫ t

0

(X[0,τ]Y

n,dMτ p)−∫ t

0

(X[0,τ]Y,dMτ p

)∣∣∣∣2)≤ E

(∫ T

0∥Y n−Y∥2

H d [M]τ p

)and so by the same Borel Canteli argument of that lemma, there is a further subsequencefor which the convergence is uniform off a set of measure zero as n→ ∞. (Actually, thesame subsequence as in the first part of the argument works.) Therefore, the conclusionfollows.

What of the second equation? Let {Y n} be as above where uniform convergence takesplace for the stochastic integrals. Then from Lemma 66.0.12∫ t

0

(X[0,τ]Y

n,dMτ p)=∫ t

0

(Y n,dMτ∧τ p

)Hence

E

(∣∣∣∣∫ t

0

(Y n,dMτ∧τ p

)−∫ t

0

(X[0,τ]Y,dMτ p

)∣∣∣∣2)≤ E

(∫ T

0∥Y n−Y∥2

H d [M]τ p

)Now by the usual application of the Borel Canelli lemma, there is a subsequence and a setof measure zero off which

∫ t0 (Y

n,dMτ∧τ p) converges uniformly to∫ t

0 (Y,dMτ∧τ p) on [0,T ]and as n→ ∞, and also ∫ t

0

(Y n,dMτ∧τ p

)→∫ t

0

(X[0,τ]Y,dMτ p

)uniformly on t ∈ [0,T ]. Then from the above,∫ t

0

(Y n,dMτ∧τ p

)→∫ t

0

(X[0,τ]Y,dMτ p

)=∫ t∧τ

0(Y,dMτ p)

uniformly. Thus∫ t

0 (Y,dMτ∧τ p) =∫ t∧τ

0 (Y,dMτ p) .

2283Lemma 66.0.13 Let Y € Y . Then for any stopping time T,ot /\ 5 5[ “(y,dM*) = [ (Zina¥am") -/ (Y,dM*™*r)0 0 , 0for @ off some set of measure zero.Proof: From Lemma 66.0.9, there exists a sequence of elementary functions Y” suchthat t > fj (Y",dM) converges uniformly to t + {5 (Y,dM‘) on [0,7] for each w ¢ N,aset of measure zero. ThenNT tAT| (Y,aM?) = lim [| (¥",dM*)0 ne JOt t_ : n Tp\ — Tplim J, (Ziog¥". 4M") [ (Zjo.q¥,dM")The last claim needs a little clarification. As shown in the above discussion proving Lemma66.0.12, while 2j9,¥" is no longer obviously an elementary function due to the fact thatt has values which are not partition points, it is still the limit of a sequence of elementaryfunctions Lion" and so the integral makes sense. Then from the inequality of Lemma66.0.9,2 Te( ) <e( [ ¥" yam" )0and so by the same Borel Canteli argument of that lemma, there is a further subsequencefor which the convergence is uniform off a set of measure zero as n — oo. (Actually, thesame subsequence as in the first part of the argument works.) Therefore, the conclusionfollows.What of the second equation? Let {Y”} be as above where uniform convergence takesplace for the stochastic integrals. Then from Lemma 66.0.12t t[ (ioartam)— ['(aaramn)t[ (Zo,j¥".4M*) =| (Y",dM**?)0 )0Hencet t 2 Te(\/ (vam) — | (2jo.q¥.aM") )<e(/ I¥"—Y lita)0 0 0Now by the usual application of the Borel Canelli lemma, there is a subsequence and a setof measure zero off which fj (Y",dM**?) converges uniformly to {j (Y,dM*) on [0,7]and as n + ©, and also[ramen > [(2jo.q%.am)uniformly on t € [0,7]. Then from the above,t t/\[ramerr) + [(a%oqr.am'r) = [vam0 0 0uniformly. Thus fo (Y,dM**) = [" (Y,dM*”). Wf