2466 CHAPTER 73. THE HARD ITO FORMULA, IMPLICIT CASE

Lemma 73.3.1 In the above situation, there exists a set of measure zero N ⊆Ω and a densesubset of [0,T ] , D such that for ω /∈ N, BX (t,ω) = B(X (t,ω)) for all t ∈ D.

Theorem 73.3.2 Let Z be progressively measurable and in

L2([0,T ]×Ω,L2

(Q1/2U,W

)).

Also suppose X is progressively measurable and in L2 ([0,T ]×Ω,W ). Let{

tnj

}mn

j=0be a

sequence of partitions of the sort in Lemma 73.1.1 such that if

Xn (t)≡mn−1

∑j=0

X(tn

j)X[tn

j ,tnj+1)

(t)≡ X ln (t)

then Xn→ X in Lp ([0,T ]×Ω,W ) . Also, it can be assumed that none of these mesh pointsare in the exceptional set off which BX (t) = B(X (t)). (Thus it will make no differencewhether we write BX (t) or B(X (t)) in what follows for all t one of these mesh points.)Then the expression

mn−1

∑j=0

⟨B∫ tn

j+1∧t

tnj∧t

ZdW,X(tn

j)⟩

=mn−1

∑j=0

⟨BX(tn

j),∫ tn

j+1∧t

tnj∧t

ZdW

⟩(73.3.3)

is a local martingale which can be written in the form∫ t

0

(Z ◦ J−1)∗BX l

n ◦ JdW

where

X ln (t) =

mn−1

∑k=0

X (tnk )X[tn

k ,tnk+1)

(t)

Proof: First suppose that⟨BX(tnk

),X(tnk

)⟩∈ L∞ (Ω). Then⟨

BX(tn

j),∫ tn

j+1∧t

tnj∧t

ZdW

is in L1 (Ω) for each t since both entries are in L2 (Ω). Why is this a martingale?

E

(⟨BX(tn

j),∫ tn

j+1∧t

tnj∧t

ZdW

⟩)= E

(E

(⟨BX(tn

j),∫ tn

j+1∧t

tnj∧t

ZdW

⟩|Ftn

j

))

= E

(⟨BX(tn

j),E

(∫ tnj+1∧t

tnj∧t

ZdW |Ftnj

)⟩)= E

(⟨BX(tn

j),0⟩)

= 0

because the stochastic integral is a martingale. Now let σ be a bounded stopping time.

E

(⟨BX(tn

j),∫ tn

j+1∧σ

tnj∧σ

ZdW

⟩)= E

(E

(⟨BX(tn

j),∫ tn

j+1∧σ

tnj∧σ

ZdW

⟩|Ftn

j

))

2466 CHAPTER 73. THE HARD ITO FORMULA, IMPLICIT CASELemma 73.3.1 In the above situation, there exists a set of measure zero N C Q and a densesubset of |0,T|, D such that for o € N, BX (t,@) = B(X (t,@)) for allt € D.Theorem 73.3.2 Let Z be progressively measurable and inVL ((0. T|xQ,D (o'u,w)) ,MpAlso suppose X is progressively measurable and in L? ({0,T] x Q,W). Let {ur 0 beajzsequence of partitions of the sort in Lemma 73.1.1 such that ifmn—1= he xte i) Zo.) =Xn()then X, + X in L? ({0,T] x Q,W). Also, it can be assumed that none of these mesh pointsare in the exceptional set off which BX (t) = B(X (t)). (Thus it will make no differencewhether we write BX (t) or B(X (t)) in what follows for all t one of these mesh points.)Then the expressionmn—1 th At mn~1 thty (Bf zaw.x (3) )= Y (ax (7), [2 zaw (73.3.3)t t* n < nNj=0 i j=0 ijis a local martingale which can be written in the formt| (ZoJ~!)" BX) oJdW0where ;Xi (t) = Y X (i) Zien) (0)k=0Proof: First suppose that (BX (ti) ,X (t?)) € L® (Q). ThenthtBX (ti), [| ° ZdW/ vtJis in L! (Q) for each t since both entries are in L? (Q). Why is this a martingale?tit [ ty AtE| ( BX (t?), | Zdw BX (t ZdW )\.Fnvat I. J[ras 1 = E ((BX (1?) ,0)) =because the stochastic integral is a martingale. Now let o be a bounded stopping time.th AO th {AoE | ( BX ()./ ZdW )) =E| E| ( BX ().f. ZdW )\.FnAo AG J