33.2. THE STOCHASTIC INTEGRAL WHEN f (s) ∈L2 (U,H) 895

= E(∥ fr∥2

L2M (tr) ,E

(M (tr+1) |Ftr

))= E

(∥ fr∥2

L2∥M (tr)∥2

)Thus 33.18 reduces to

mn−1

∑r=0

E(∥ fr∥2

L2∥M (tr+1)∥2

)−E

(∥ fr∥2

L2∥M (tr)∥2

)=

∫Ω

mn−1

∑r=0∥ fr∥2

L2

(∥M (tr+1)∥2−∥M (tr)∥2

)dP

=∫

mn−1

∑r=0∥ fr∥2

L2([M (tr+1)]− [M (tr)])dP

because the martingales from the quadratic variation have expectation 0. This has provedthe following theorem.

Theorem 33.2.2 Let f be an elementary function corresponding to the partitionP = {tk}mn

k=0 . Then

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤

∫Ω

mn−1

∑r=0∥ fr∥2

L2([M (tr+1)]− [M (tr)])dP

=∫

∫ T

0∥ f∥2

L2d [M] (t)dP (33.19)

That integral on the right end is just the conventional Stieltjes integral of a step function.Recall that [M] is increasing and continuous.

Theorem 33.2.3 Let f be uniformly bounded, ∥ f (t,ω)∥L2(U,H) ≤K, continuous int, and adapted. Also let M (t) be a bounded continuous martingale with values in U andsuppose [M] (T ) ∈ L1 (Ω,P). Let fn be an elementary function approximating f

fn (t)≡mn−1

∑k=0

f (tnk )X(tn

k ,tnk+1]

(t)

where Pn ={

tnk

}mnk=0 is a partition of [0,T ] . Assume

∥Pn∥ ≡max{∣∣tn

k+1− tnk

∣∣ : k ≤ mn}.

Then there exists a unique continuous bounded martingale denoted as∫ t

0 f dM which satis-fies

lim∥Pn∥→0

∫ t

0fndM =

∫ t

0f dM in M 2

T (H)

where this means: For every ε > 0 there is δ > 0 such that if Pn is a partition having∥Pn∥< δ , then ∥∥∥∥∫ (·)

0fndM−

∫ (·)

0f dM

∥∥∥∥M 2

T (H)

< ε.

For any such sequence of partitions and approximating elementary functions, there is aset of measure zero N such that if ω /∈ N, then

∫ t0 fndM (ω)→

∫ t0 f dM (ω) uniformly in

t ∈ [0,T ] . Also, for such f just described,

12

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤∫

∫ T

0∥ f∥2

L2d [M] (t)dP (33.20)

33.2. THE STOCHASTIC INTEGRAL WHEN f (s) € %2(U,H) 895HE (fell M(t) ,E (M (test) Lx,)) = (Well IM (oI)Thus 33.18 reduces tomn—1E(\f-ll2 IM (tI?) —E (WI fell2,, IM (t,) 117¥ E (Il lM Gras?) ~& (Ive IM IP)mn—1LX Wl (IM Ges? — eI?)» Mn—1LX Well, (Gro) lM gharbecause the martingales from the quadratic variation have expectation 0. This has provedthe following theorem.Theorem 33.2.2 Lez f be an elementary function corresponding to the partitionP= {ty}. Thenmn—|) [YE Wl (or enn] en)yarse ( supte|0,T]T2[ff Wti,ata war (33.19)That integral on the right end is just the conventional Stieltjes integral of a step function.Recall that [M] is increasing and continuous.IA(t,@)|| g(u,n) SK, continuous int, and adapted. Also let M(t) be a bounded continuous martingale with values in U andsuppose [M](T) € L' (Q,P). Let fy be an elementary function approximating fmyn—1=) f(t) 2 (ar) | (t)where P, = {th eG is a partition of [0,T]. Assume||P, || = max {|fj), — | :k<m}.Then there exists a unique continuous bounded martingale denoted as fo fdM which satis-fieslim I f,dM = [ fdM in 2 (H)\|Pul| +0 70where this means: For every € > 0 there is 6 > 0 such that if P, is a partition having||Pal| < 6, then(-) (-)l/ fram — | ram <€é° ° AH)For any such sequence of partitions and approximating elementary functions, there is aset of measure zero N such that if @ ¢ N, then J} fudM(@) — fj fdM (@) uniformly int € [0,7]. Also, for such f just described,T) < [| Wiz, oar (33.20)supte [0,7]