33.3. MORE ON STOPPING TIMES 899

fnX[0,σn] → f X[0,σ ] in L2 (Ω× [0,T ] ;L2 (U,H)). Here fn (t) is the elementary function

∑mn−1k=0 f

(tnk

)X(tn

k ,tnk+1]

(t) where{

tmnk

}is a partition Pn with ∥Pn∥→ 0 so that fn converges

uniformly to f for each ω .

Proof: Note that |σn (ω)−σ (ω)| ≤ ∥Pn∥ . Then∥∥ fnX[0,σn]− f X[0,σ ]

∥∥ ≤∥∥( fn− f )X[0,σn]

∥∥+∥∥ f X[σn,σ ]

∥∥∥∥ fnX[0,σn]− f X[0,σ ]

∥∥2 ≤ 2(∥∥( fn− f )X[0,σn]

∥∥2+∥∥ f X[σn,σ ]

∥∥2)

Thus ∫Ω

∫ T

0

∥∥ fnX[0,σn]− f X[0,σ ]

∥∥2 d [M]dP

≤ 2∫

∫ T

0∥ fn− f∥2 d [M]dP+2K

∫Ω

∫ T

0X[σn,σ ]d [M]dP

≤ 2∫

∫ T

0∥ fn− f∥2 d [M]dP+2K

∫Ω

[M] (σ)− [M] (σn)dP (33.27)

the integrands∫ T

0 ∥ fn− f∥2 d [M] and [M] (σ)− [M] (σn) both converge to 0 a.e. ω asn→∞ thanks to continuity of [M]. Also the assumption that [M] (T ) is in L1 along with theboundedness of f imply these integrands are uniformly integrable. Hence we can use theVitali convergence theorem and conclude that the limit of 33.27 is 0. ■

Now note that if you fix ω, fnX[0,σn] (t)→ f X[0,σ ] (t) for each t < σ (ω) . Thus, iff is of bounded variation as well as being continuous in each t, standard Stieltjes in-tegral considerations involving continuity of f and M show that

∫ t0 fnX[0,σn] (t)dM →∫ t

0 f X[0,σ ] (t)dM with no probabilistic complications at all.

Definition 33.3.3 Let f be adapted, continuous in t, and bounded. Let M be acontinuous bounded martingale. Also let σ be a stopping time and let σn be the discreetapproximation above relative to partitions Pn =

{tnk

}mnk=0 where ∥Pn∥ → 0 and { fn} the

sequence of elementary functions approximating f ,

fn (t)≡mn−1

∑k=0

f (tnk )X(tn

k ,tnk+1]

(t) ,

Then there exists a martingale denoted as∫ t

0 f X[0,σ ]dM such that∫ t

0f X[0,σ ]dM = lim

n→∞

∫ t

0fnX[0,σn]dM in M2

T (H)

This is something new. Earlier we had∫ t

0 f dM defined where f is continuous on [0,T ] .We also have

∫ t∧σ

0 f dM defined where σ is a stopping time and f is continuous on [0,T ] .However, f X[0,σ ] is not necessarily continuous on [0,T ] . It is continuous on [0,T ∧σ ] sothe time intervals are changing as a function of ω .

To begin with, we can stop the martingale∫ t

0 fndM with the stopping time σ , this de-noted as

∫ t∧σ

0 fndM. Then∫ t∧σ

0fndM−

∫ t∧σn

0fndM

=mn−1

∑k=0

f (tnk )

((M (t ∧σ ∧ tk+1)−M (t ∧σ ∧ tk))−(M (t ∧σn∧ tk+1)−M (t ∧σn∧ tk))

)

33.3. MORE ON STOPPING TIMES 899Sn 20,0) + f 20,0} in L? (Qx [0,T];-A(U,H)). Here fy (t) is the elementary functionan 'f (t?) 2 (ip ap,) (0) where {2"} is a partition P, with ||P,|| + 0 so that fy convergesuniformly to f for each @.Proof: Note that |o,,(@)— o(@)| < ||P,||. ThenI|fn 20.0.) -~FZ%io0)|| < \n-f) 2o.0,|| + || fF ic,.01|lf Zoon-FZoail < 2(|lGe— A) Ziouill’ +f Zion0ill’)ThusTLF Wn%ne~F%io0\|aimlar2 [iim siPalmiar-+2K [ [* 2%, .\¢lMlar2 [lite fiPalm)ae+2n fim (o)—[M](o,)dP (33.27)IAIAthe integrands fo || fr —f||’d[M] and [M](o) — [M](on) both converge to 0 ae. @ asn—> oo thanks to continuity of [M]. Also the assumption that [M] (T) is in L' along with theboundedness of f imply these integrands are uniformly integrable. Hence we can use theVitali convergence theorem and conclude that the limit of 33.27 is 0.Now note that if you fix @, fn? jo,0,] (4) + 20,9] (¢) for each t < o(@). Thus, iff is of bounded variation as well as being continuous in each f¢, standard Stieltjes in-tegral considerations involving continuity of f and M show that fj fh, Xio,o,) (t)dM >Jof 20,9) (t) dM with no probabilistic complications at all.Definition 33.3.3 Let f be adapted, continuous in t, and bounded. Let M be acontinuous bounded martingale. Also let o be a stopping time and let Oy, be the discreetapproximation above relative to partitions P, = {t, co where ||P,|| > 0 and {f,} thesequence of elementary functions approximating f,my—1= 2b fia) Kann y(t),Then there exists a martingale denoted as Jo f 21,6)4M such thatt t| f 2o,o)dM = lim [ Jn %0,6,)4M in Mz (H)0 ; no JO) :This is something new. Earlier we had {j fdM defined where f is continuous on [0,7].We also have f° fdM defined where o is a stopping time and f is continuous on {0,7].However, f %j,5] is not necessarily continuous on [0,7]. It is continuous on [0,7 A o] sothe time intervals are changing as a function of o.To begin with, we can stop the martingale fo JndM with the stopping time o, this de-noted as {j/\° f,dM. ThentAt\Onf,dM — [ f,dMmel (M(t GAtk,1) —M(tAG At)= d f(t) (_ (M(tAGyA tet) — Monn )