65.5. THE GENERAL INTEGRAL 2247

mn

∑j=0

∑k=1

(ψk(t ∧ tn

j+1)−ψk

(t ∧ tn

j))((

f nj −Φ

(tn

j)◦ J−1)(Jgk)

)The first expression does not depend on J or U1. I need only argue that the second expres-sion converges to 0 as n→ ∞. The infinite sum converges in L2 (Ω;H) and also, as in theabove diversion, the independence of the ψk implies that

E

∣∣∣∣∣ mn

∑j=0

∑k=1

(ψk(t ∧ tn

j+1)−ψk

(t ∧ tn

j))((

f nj −Φ

(tn

j)◦ J−1)(Jgk)

)∣∣∣∣∣2

H

=

mn

∑j=0

(t ∧ tn

j+1− t ∧ tnj) ∞

∑k=1

E(∣∣( f n

j −Φ(tn

j)◦ J−1)(Jgk)

∣∣2H

)

=mn

∑j=0

(t ∧ tn

j+1− t ∧ tnj)

E(∣∣∣∣ f n

j −Φ(tn

j)◦ J−1∣∣∣∣2

L2(JQ1/2U,H)

)=

∫ t

aE(∣∣∣∣Φn−Φ

n ◦ J−1∣∣∣∣2L2(JQ1/2U,H)

)ds

which is given to converge to 0 since both converge to Φ◦J−1. Consequently, the stochasticintegral defined above does not depend on J or U1.

It is interesting to note that in the above definition, the approximate problems do appearto depend on J and U1 but the limiting stochastic process does not. Since it is the case thatthe stochastic integral is independent of U1 and J, it can only be dependent on Q1/2U andU, and so we refer to W (t) as a cylindrical process on U. By Lemma 65.4.3 you can takeU1 = U and so you can consider the finite sums defining the Wiener process to be in Uitself. From the proof of this lemma, you can even have J being the identity on the spanof the first n vectors in the orthonormal basis for Q1/2U . The case where Q is trace classfollows in the next section. In this case, W is an actual Q Wiener process on U .

The following corollary follows right away from the above theorem.

Corollary 65.5.4 Let Φ,Ψ ∈ L2([a,T ]×Ω;L2

(Q1/2U,H

))and suppose they are both

progressively measurable. Then

E((∫ t

aΦdW,

∫ t

aΨdW

)H

)= E

(∫ t

a(Φ,Ψ)L2(Q1/2U,H) ds

)Also if L is in L∞ (Ω,L (H,H)) and is Fa measurable, then

L∫ t

aΦdW =

∫ t

aLΦdW (65.5.13)

and

E((

L∫ t

aΦdW,

∫ t

aΨdW

)H

)= E

(∫ t

a(LΦ,Ψ)L2(Q1/2U,H) ds

). (65.5.14)

65.5. THE GENERAL INTEGRAL 22473Msn, (Wy (CAC 1) — We (tAt7)) (CF? — ® (#7) oF!) (Jax))n.lllIThe first expression does not depend on J or U;. I need only argue that the second expres-sion converges to 0 as n — oo. The infinite sum converges in L? (Q;H) and also, as in theabove diversion, the independence of the y, implies that2EYED (Va (Ati) — We Ati) (FF — 8 (8) oF) (Fax)jJ=0k=1 H=¥ (At 100?) YE ([(f7-8 (08) oF) Vaid li)j=0 k=!= ¥ (thar eA) £ (I/F - 804) 2 Levu)J=t —1)/2[ E (||®,—o" oJ! ‘ato'rvn)) dswhich is given to converge to 0 since both converge to ®oJ~!. Consequently, the stochasticintegral defined above does not depend on J or U;.It is interesting to note that in the above definition, the approximate problems do appearto depend on J and U but the limiting stochastic process does not. Since it is the case thatthe stochastic integral is independent of U, and J, it can only be dependent on Q!/?U andU, and so we refer to W (t) as a cylindrical process on U. By Lemma 65.4.3 you can takeU; =U and so you can consider the finite sums defining the Wiener process to be in Uitself. From the proof of this lemma, you can even have J being the identity on the spanof the first n vectors in the orthonormal basis for Q'/ 2U. The case where Q is trace classfollows in the next section. In this case, W is an actual Q Wiener process on U.The following corollary follows right away from the above theorem.Corollary 65.5.4 Let ®, WY € L? ({a, 7] x0; Ly (o'/?U,H)) and suppose they are bothprogressively measurable. Thent t tE ((/ oaw, | waw) ) =E (/ (®,P) 4,(012u,) as)a a H a ,Also if L is in L® (Q,.2 (H,H)) and is #q measurable, thent tLf eaw = | Leaw (65.5.13)Ja JaandE (Gi odW, [ waw) ) =E (/ (L®,Y) (Qiu) as) - 65.5.14)