902 CHAPTER 33. QUADRATIC VARIATION AND STOCHASTIC INTEGRATION

such that if m≥ n, then for a given ω , τm (ω) ,σm (ω) are both ∞. It follows upon using ap-proximation with elementary functions and passing to a limit using a suitable subsequenceof elementary functions that for such τm,σm,

∫ t∧τm

0f dM (ω) ,

∫ t∧σm

0f dM (ω) = lim

p→∞

p−1

∑k=0

f(t pk

)(M(t ∧ t p

k+1

)−M

(t ∧ t p

k

))(ω)

this limit being independent of the localizing sequence used. The reason it is a local con-tinuous martingale is that Mτn is a bounded continuous martingale and so(∫ t

0f dM

)τn

≡∫ t∧τn

0f dM ≡

∫ t

0f dMτn

is a martingale. ■The idea is that if you know it at t ∧ τn for all τn where τn→ ∞, then you know it at t

because you can simply pick τn larger than t and from the above, it doesn’t matter whichlocalizing sequence you use.

Now suppose M is just a local martingale so there is a localizing sequence of stoppingtimes {τn} such that Mτn is a bounded martingale and suppose [M] (T ) ∈ L1 (Ω) . Let fnbe elementary functions converging to f a bounded, adapted, continuous in t function,convergence uniform in t for each ω . Then from the fundamental inequality above andusing the formulas for stopping times,

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t∧τ p

0fndM

∥∥∥∥2)

=12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0fndMτ p

∥∥∥∥2)

≤∫

∫ T

0∥ fn∥2 d [M]τ p (t)dP≤

∫Ω

∫ T

0∥ fn∥2 d [M] (t)dP

Since ∥ f∥ is assumed bounded and [M] (T ) ∈ L1, it follows that there is a dominatingfunction on the right, namely K2 [M] (T )where K ≥ ∥ f (t,ω)∥ for all (t,ω). Let n→ ∞

and use the dominated convergence on the right and either Fatou’s lemma or monotoneconvergence theorem on the left to obtain

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤∫

∫ T

0∥ f∥2 d [M] (t)dP

The reason the monotone convergence theorem applies is that on the left, the stopping timeeffectively restricts the values of t over which the sup is taken until τ p ≥ t. You could alsoapply Fatou’s lemma.

This has shown the following proposition.

Proposition 33.4.3 Let f be adapted, continuous in t and bounded. Also let M be alocal martingale with [M] (T ) ∈ L1 (Ω). Then

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤∫

∫ T

0∥ f∥2 d [M] (t)dP

902 CHAPTER 33. QUADRATIC VARIATION AND STOCHASTIC INTEGRATIONsuch that if m > n, then for a given @, T,(@) , Om (@) are both ce. It follows upon using ap-proximation with elementary functions and passing to a limit using a suitable subsequenceof elementary functions that for such T,, Om,t\Tm t\Om a.[game [fam (oo) = tim YF) (M (OAIf,3) —M (Cah) (0)0 0 Por? R=this limit being independent of the localizing sequence used. The reason it is a local con-tinuous martingale is that M™ is a bounded continuous martingale and sot Th t\Tp t( [ fast) = [ fdM = | fdM™0 0 0is a martingale.The idea is that if you know it at t AT, for all t, where T,, — o-, then you know it at tbecause you can simply pick Tt, larger than ¢ and from the above, it doesn’t matter whichlocalizing sequence you use.Now suppose M is just a local martingale so there is a localizing sequence of stoppingtimes {t,,} such that M™ is a bounded martingale and suppose [M](T) € L!(Q). Let fy,be elementary functions converging to f a bounded, adapted, continuous in ¢ function,convergence uniform in ¢ for each w. Then from the fundamental inequality above andusing the formulas for stopping times,2I T<E| sup sup [ fndM"?2 te[0,7] te[0,7]<f [umrawrnars [ [" imiann@arSince ||/|| is assumed bounded and [M](T) € L’, it follows that there is a dominatingfunction on the right, namely K* [M](T)where K > ||f (t,@)|| for all (t,@). Let n > ©and use the dominated convergence on the right and either Fatou’s lemma or monotoneconvergence theorem on the left to obtainPy ay2ef [ineaunmaralosupte[0,7]The reason the monotone convergence theorem applies is that on the left, the stopping timeeffectively restricts the values of t over which the sup is taken until t, >t. You could alsoapply Fatou’s lemma.This has shown the following proposition.tATpfndMProposition 33.4.3 Let f be adapted, continuous in t and bounded. Also let M be alocal martingale with [M](T) € L' (Q). ThenTsup <[ f iri?atm arte [0,7] QI0