73.3. PRELIMINARY RESULTS 2467

= E

(⟨BX(tn

j),

(E∫ tn

j+1∧σ

tnj∧σ

ZdW |Ftnj

)⟩)

= E(⟨

BX(tn

j),E(∫ tn

j+1∧σ

0ZdW −

∫ tnj∧σ

0ZdW |Ftn

j

)⟩)= E

(⟨BX(tn

j),0⟩)

= 0

and so this is a martingale by Lemma 63.1.1. I want to write the formula in 73.3.3 as astochastic integral. First note that W has values in U1.

Consider one of the terms of the sum more simply as⟨B∫ b

aZdW,X (a)

⟩, a = tn

k ∧ t, b = tnk+1∧ t.

Then from the definition of the integral, let Zn be a sequence of elementary functions con-verging to Z ◦ J−1 in L2

([a,b]×Ω,L2

(JQ1/2U,W

))and∥∥∥∥∫ t

aZdW −

∫ t

aZndW

∥∥∥∥L2(Ω,W )

→ 0

Using a maximal inequality and the fact that the two integrals are martingales along with theBorel Cantelli lemma, there exists a set of measure 0 N such that for ω /∈N, the convergenceof a suitable subsequence of these integrals, still denoted by n, is uniform for t ∈ [a,b]. Itfollows that for such ω,⟨

B∫ t

aZdW,X (a)

⟩= lim

n→∞

⟨B∫ t

aZndW,X (a)

⟩. (73.3.4)

Say

Zn (u) =mn−1

∑k=0

Znk X[tn

k ,tnk+1)

(u)

where Znk has finitely many values in L (U1,W )0 , the restrictions of maps in L (U1,W ) to

JQ1/2U, and the tnk refer to a partition of [a,b]. Then the product on the right in 73.3.4 is of

the formmn−1

∑k=0

⟨BZn

k(W(t ∧ tn

k+1)−W (t ∧ tn

k )),X (a)

⟩W ′,W

Note that it makes sense because Znk is the restriction to J

(Q1/2U

)of a map from U1 to W

and so BZnk is a map from U1 to W ′. Then the Wiener process has values in U1 so when you

apply BZnk to W

(t ∧ tn

k+1

)−W

(t ∧ tn

k

), you get something in W ′ and so the duality pairing

is between W ′ and W as shown. Also, Znk

(W(t ∧ tn

k+1

)−W

(t ∧ tn

k

))gives something in W

because the Wiener process has values in U1 and Znk acts on these things to give something

in W . Thus the above equals

=mn−1

∑k=0

⟨BX (a) ,Zn

k(W(t ∧ tn

k+1)−W (t ∧ tn

k ))⟩

W ′,W

73.3. PRELIMINARY RESULTS 2467th, AO=E|( BX(t?),{E [°° ZdW\| FnJ "ho /Jti NO tA= B ( (ax (17) ,E (/ zaw ~ | zaW\Fx))0 0 /E ((BX (#7) ,0)) =0and so this is a martingale by Lemma 63.1.1. I want to write the formula in 73.3.3 as astochastic integral. First note that W has values in U).Consider one of the terms of the sum more simply asb(a / zaW,X (a) ). a=1 Mt, b=A,, At.aThen from the definition of the integral, let Z,, be a sequence of elementary functions con-verging to ZoJ~! in L? ([a,b] x Q,.Z (JO'/?U,W)) andt t| [zw | ZndWJa JaUsing a maximal inequality and the fact that the two integrals are martingales along with theBorel Cantelli lemma, there exists a set of measure 0 N such that for @ ¢ N, the convergenceof a suitable subsequence of these integrals, still denoted by n, is uniform for ¢ € [a,b]. Itfollows that for such @,(8 [ zaW.X (a) ) = lim (2 | “ZadW.X(a)). (73.3.4)>0L2(Q,W)Saymn—1Zn (u) = y Zi Xin at) (u)k=0where Z;' has finitely many values in # (U;,W),, the restrictions of maps in # (U;,W) toJo'/?U, and the t refer to a partition of [a,b]. Then the product on the right in 73.3.4 is ofthe formmn—1py (BZE (W(t Athy) —W (tg) X (2) wwNote that it makes sense because Z; is the restriction to J (ol/ 2U ) of a map from U; to Wand so BZ; is a map from U, to W’. Then the Wiener process has values in U; so when youapply BZ? to W (t At, ,) —W (ttf) , you get something in W’ and so the duality pairingis between W’ and W as shown. Also, Z? (W (t A1Z,,) — W (t At?) gives something in Wbecause the Wiener process has values in U; and Z; acts on these things to give somethingin W. Thus the above equalsmy—1= py (BX (a) Zi (W (tA th.) —W(tAtt)) wow