894 CHAPTER 33. QUADRATIC VARIATION AND STOCHASTIC INTEGRATION

Note that for tr < s≤ t,

E ( fr (M (t ∧ tr+1)−M (t ∧ tr)) |Fs) = frE (M (t ∧ tr+1)−M (t ∧ tr) |Fs)

= fr (M (s∧ tr+1)−M (s∧ tr))

However, for t ≤ tr the term in the sum equals 0. Thus each term in that sum is a martingale.It follows that

∫ t0 f dM is a martingale for f an elementary function. It is also a continuous

martingale because each term is continuous. By Theorem 30.5.3, the maximal estimate andthe theorem about the quadratic variation,

E

(sup

t∈[0,T ]

∥∥∥∥∫ t

0f dM

∥∥∥∥2)≤ 2E

(∥∥∥∥∫ T

0f dM

∥∥∥∥2)

Now from the definition of the integral given above, E(∥∥∥∫ T

0 f dM∥∥∥2)=

E

(mn−1

∑r=0

fr (M (tr+1)−M (tr)) ,mn−1

∑r=0

fr (M (tr+1)−M (tr))

)H

(33.17)

because T ≥ tr for each tr in the partition of the interval. Consider a mixed term in theabove product in which j < k

E ( fk (M (tk+1)−M (tk)) , f j (M (t j)−M (t j)))

= E(E[

fk (M (tk+1)−M (tk)) , f j (M (t j)−M (t j)) |Ftk

])= E

(E[(M (tk+1)−M (tk)) , f ∗k f j (M (t j)−M (t j)) |Ftk

])= E

(f ∗k f j (M (t j)−M (t j)) ,E

[(M (tk+1)−M (tk)) |Ftk

])= 0

Now from Lemma 22.5.15 on Hilbert Schmidt operators,

∥ frM∥2 ≤ ∥ fr∥2 ∥M∥2 ≤ ∥ fr∥2L2∥M∥2 .

Therefore, from 33.17,

E

(mn−1

∑r=0

fr (M (tr+1)−M (tr)) ,mn−1

∑r=0

fr (M (tr+1)−M (tr))

)H

=

mn−1

∑r=0

E(∥ fr (M (tr+1)−M (tr))∥2

)≤

mn−1

∑r=0

E(∥ fr∥2 ∥(M (tr+1)−M (tr))∥2

)=

mn−1

∑r=0

E(∥ fr∥2

L2∥M (tr+1)∥2

)+

mn−1

∑r=0

E(∥ fr∥2

L2∥M (tr)∥2

)−2

mn−1

∑r=0

E(∥ fr∥2

L2(M (tr+1) ,M (tr))

)(33.18)

Consider the mixed term on the end.

E(∥ fr∥2

L2(M (tr+1) ,M (tr))

)= E

(E(∥ fr∥2

L2(M (tr+1) ,M (tr)) |Ftr

))

894 CHAPTER 33. QUADRATIC VARIATION AND STOCHASTIC INTEGRATIONNote that fort, <s <t,E(fr(M(tAtrs1) -M(tAt,))|Fs) = frE(M(tAtra1) -—M (tt) |Fs)= fr(M(sAt-+1) -M(sAt,))However, for t < t, the term in the sum equals 0. Thus each term in that sum is a martingale.It follows that fo fdM is a martingale for f an elementary function. It is also a continuousmartingale because each term is continuous. By Theorem 30.5.3, the maximal estimate andthe theorem about the quadratic variation,*) <2e(|f )supte [0,7]2Now from the definition of the integral given above, E (| fo fdM | ) =my—1 >(E y f.(M (tr41) — yd Sf (M (tr1) — mw) (33.17)Hbecause T > t, for each ¢, in the partition of the interval. Consider a mixed term in theabove product in which j < kE (fe (M (that) —M (te) £7 (M (Gj) — M (t;)))= E(E [fc (M (tev) —M (tx) 7 (M (t7) — M(t) | Fn])E(E [(M (t.41)—M (te) fe fi (M (tj) —M (t))) |F])E (fe. f)(M (tj) —M (ti) .E [(M (te+1) —M (tu) |Fu]) = 0Now from Lemma 22.5.15 on Hilbert Schmidt operators,2 2 2 2 2AMP < [Fell IMIS S I frllee, MAPTherefore, from 33.17,my—1| "S(E y f-(M (t-+1) »d f-(M (tr41) ~My) =Hmyn—1EF (i (oes) MIP) <"S (Wil later) MI)r=0="F (Wl2slIM Gun)mn—1 Myn—+S (Wl i IP) -2°S & (ls (ean), (33.18)Consider the mixed term on the end.E (I\frly (M (test) 5M (6,))) = E (E (fell (M (trot) M (6) Lx.)