33.6. THE CASE OF f ∈ L2 (Ω× [0,T ] ;L2 (U,H)) 907

and by completeness of M 2T (H) ,

∫ t0 gndM converges to a continuous martingale which I

can call∫ t

0 f dM. It is clear that any two sequences give the same result from the inequalitysatisfied. Therefore, the stochastic integral

∫ t0 f dM is well defined. Also from the theory of

M 2T (H) , there is a subsequence for which

∫ t0 gndM converges uniformly in t to

∫ t0 f dM off

some set of measure zero.In case M is only a local martingale, we see from approximating f with continuous in t

and adapted and bounded functions gn as above that the appropriate way to define∫ t

0 f dMis as

∫ t∧σn0 f dM ≡

∫ t0 f dMσn where {σn} is a localizing sequence for M.

The quadratic variation of Mσn is no more than the quadratic variation of M and so

limm,n→∞

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t∧σn

0(gn−gm)dM

∥∥∥∥2)

≤ limm,n→∞

∫Ω

∫ T

0∥gn−gm∥2 dtdP = 0

Thus we can obtain∫ t

0 f dMσn as a limit in M 2T (H) as just done. Then one can define∫ t

0 f dM ≡ limn→∞

∫ t0 f dMσn where σn is a localizing sequence for M. Also, we can pass to

a limit as n→ ∞ using the monotone convergence theorem in the inequality

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t∧σn

0f dMσn

∥∥∥∥2)≤∫

∫ T

0∥ f∥2 dtdP

The stopping time has the effect of restricting the time interval, so as σn increases, one istaking sup over a larger set. That is why the monotone convergence theorem applies on theleft side. Then

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤∫

∫ T

0∥ f∥2 dtdP

As to the last claim about stopping times, it works for f bounded and continuous in tand adapted and M a martingale. Therefore, it also works for

f ∈ L2 (Ω× [0,T ] ;L2 (U,H)) .

In general, when M is only a local martingale with localizing sequence stopping times{τn} , then Mσ is also a local martingale with localizing sequence {τn}. I need to showthat ∫ t∧σ

0f dM =

∫ t

0f dMσ

as local martingales. I need to show that∫ t∧σ∧τn

0f dM =

∫ t

0f d (Mσ )τn

The equation is true because of the definition of∫ t∧σ

0 f dM in terms of the stopping times.Therefore,

∫ t∧σ

0 f dM =∫ t

0 f dMσ as local martingales. Of course if M is a martingale, thisis true also. ■

33.6. THE CASE OF f € L? (Q x [0,T];-A (U,H)) 907and by completeness of 47 (H), Jj 8.dM converges to a continuous martingale which Ican call fo fdM. It is clear that any two sequences give the same result from the inequalitysatisfied. Therefore, the stochastic integral fo JfdM is well defined. Also from the theory ofM(H), there is a subsequence for which {5 g,dM converges uniformly int to {} fdM offsome set of measure zero.In case M is only a local martingale, we see from approximating f with continuous in tand adapted and bounded functions g,, as above that the appropriate way to define Jo fdMis as f°" fdM = Jy fdM™ where {o,} is a localizing sequence for M.The quadratic variation of M® is no more than the quadratic variation of M and so)Thus we can obtain {j fdM° as a limit in #7 (H) as just done. Then one can defineJo SdM = limy-+. Jo fdM°* where ©, is a localizing sequence for M. Also, we can pass toa limit as n + co using the monotone convergence theorem in the inequality1 t\On 2 T 2“e( sup | fame|| \ < | | \| fata2° \refo,rj||40 aJoThe stopping time has the effect of restricting the time interval, so as oO, increases, one istaking sup over a larger set. That is why the monotone convergence theorem applies on theleft side. Then1 t 2 T-e( sup [ fan <[ | || f\|2ataP2 t€(0,7] ||40 Q/0As to the last claim about stopping times, it works for f bounded and continuous in t¢and adapted and M a martingale. Therefore, it also works fort\On[ (8n _ 8m) dM1lim ~E | supmn—o 2 [sueT< lim Ef ll@n — @m||2dtdP =0Q/0m,n—cofel’ (Qx [0,7]; (U,H)).In general, when M is only a local martingale with localizing sequence stopping times{t,}, then M®° is also a local martingale with localizing sequence {t,}. I need to showthat[pau = [raneas local martingales. I need to show that—_— fdM _ [ fd (m?)™0The equation is true because of the definition of Jo? fdM in terms of the stopping times.Therefore, {5° fdM = fj fdM° as local martingales. Of course if M is a martingale, thisis true also.