33.5. THE STOCHASTIC INTEGRAL AND THE QUADRATIC VARIATION 903

If f is bounded, continuous, and of bounded variation, and adapted and M is a localmartingale, this shows that the Stieltjes integral

∫ t0 f dM is a local martingale.

It is clear that if M is a local martingale with localizing sequence τ p and if σ is astopping time, then Mσ is also a local martingale with localizing sequence τ p because(Mσ )τ p = (Mτ p)σ , the latter being a bounded martingale. Now for f continuous in t andbounded and adapted, ∫ t∧σ∧τ p

0f dM ≡

∫ t

0f dMσ∧τ p ≡

∫ t∧τ p

0f dMσ

Therefore, from the definition, whenever M is a local martingale and f bounded and con-tinuous in t and adapted, ∫ t∧σ

0f dM =

∫ t

0f dMσ (33.30)

33.5 The Stochastic Integral and the Quadratic Varia-tion

In this simple case of the above, you have a bounded martingale M with values in U andyou have f ∈ U ′. Thus f ∈ L (U,R). Is f actually in L2 (U,R)? By Riesz represen-tation theorem, there is x ∈ U such that Rx = f . Then if {gk} is an orthonormal basisfor U,∑k | f (gk)|2 = ∑k |(gk,Rx)|2 = ∑k |(gk,x)|2 = ∥x∥2 because it is just the sum of thesquares of the Fourier coefficients of x. Thus f ∈L2 (U,R).

Now an example of a continuous in t, adapted, bounded function in L2 (U,R) is justRM (t) ≡ f (t). Therefore, it makes perfect sense to consider

∫ t0 (RM)dM. Let ∥Pn∥ → 0

and let the stochastic integral of an elementary function fn (t) =∑mn−1k=0 RM

(tnk

)X(tn

k ,tnk+1]

(t)be of the form

mn−1

∑k=0

RM (tnk )(M(t ∧ tn

k+1)−M (t ∧ tn

k ))

where{

tnk

}mnk=0 is this partition Pn.

ALWAYS assume in this that the partitions are nested, Pn ⊆ Pn+1.

Qn (t) ≡mn−1

∑k=0

∥∥M(t ∧ tn

k+1)−M (t ∧ tn

k )∥∥2

U

=mn−1

∑k=0

∥∥M(t ∧ tn

k+1)∥∥2

+∥M (t ∧ tnk )∥

2−2(M(t ∧ tn

k+1),M (t ∧ tn

k ))

=mn−1

∑k=0

∥∥M(t ∧ tn

k+1)∥∥2−∥M (t ∧ tn

k )∥2−2

(M (t ∧ tn

k ) ,M(t ∧ tn

k+1)−M (t ∧ tn

k ))

= ∥M (t)∥2−2∫ t

0(RMn)dM

Then passing to a limit, then Qn (t)→ Q(t) in L2 (Ω) because 2∫ t

0 (RMn)dM converges inM 2

T (R) . Using a subsequence, we can also get uniform convergence in t for all ω off a setof measure zero. Thus Q is increasing. It follows

Q(t) = ∥M (t)∥2−2∫ t

0(RM)dM = [M] (t)+N (t)−2

∫ t

0(RM)dM

33.5. THE STOCHASTIC INTEGRAL AND THE QUADRATIC VARIATION 903If f is bounded, continuous, and of bounded variation, and adapted and M is a localmartingale, this shows that the Stieltjes integral Jo dM is a local martingale.It is clear that if M is a local martingale with localizing sequence T, and if o is astopping time, then M° is also a local martingale with localizing sequence t, because(M°)*? = (M*)° , the latter being a bounded martingale. Now for f continuous in f andbounded and adapted,tAOATp t tATp| fdM = [ fame» = [” fame0 0 0Therefore, from the definition, whenever M is a local martingale and f bounded and con-tinuous in ¢ and adapted,| i O° tdM = | i ' faM® (33.30)33.5 The Stochastic Integral and the Quadratic Varia-tionIn this simple case of the above, you have a bounded martingale M with values in U andyou have f € U’. Thus f € “(U,R). Is f actually in %(U,R)? By Riesz represen-tation theorem, there is x € U such that Rx = f. Then if {g,} is an orthonormal basisfor U,Y x |f (ge) > = Le lees Rx)? = Le |(ge.x)|° = ||x||* because it is just the sum of thesquares of the Fourier coefficients of x. Thus f € 2 (U,R).Now an example of a continuous in t, adapted, bounded function in % (U,R) is justRM (t) = f (t). Therefore, it makes perfect sense to consider {j (RM) dM. Let ||P,|| > 0and let the stochastic integral of an elementary function f;, (t) = ml RM (1?) Keng] (t)be of the formmn—1Yo RM (Hf) (M(t Ath.) —M(AM))k=0where {t/})."9 is this partition P,.ALWAYS assume in this that the partitions are nested, P, C P41.mn—1a) = ¥ IM (CAtta) Mtg) IIemn—1Ye [lM (tag) I+ [M(t Ath) |? —2 (M(t Arh.) M(t Are)k=0mn—1= ¥ [im (enag)|l? = eaegy|? —2 (Mer) M (Ads) —M C0)k=0t= | ()|? -2 [ (RM, amThen passing to a limit, then Q, (t) + Q(t) in L? (Q) because 2 5 (RM,) dM converges inM; (R). Using a subsequence, we can also get uniform convergence in ¢ for all @ off a setof measure zero. Thus Q is increasing. It followsQ(t) =||M(|P-2 [ (RM)am = (M\(n) +0) -2 | (RM) ae