2122 CHAPTER 63. THE QUADRATIC VARIATION OF A MARTINGALE

Here is why. The sequence{

τnk (ω)

}∞

k=1 eventually equals T for all n sufficiently large.This is because if it did not, it would converge, being bounded above by T and then bycontinuity of M,

{M(τn

k (ω))}∞

k=1 would be a Cauchy sequence contrary to the requirementthat ∣∣∣∣M (τn

k+1 (ω))−M (τn

k (ω))∣∣∣∣

=∣∣∣∣M (ηn

k+1 (ω))−M (ηn

k (ω))∣∣∣∣= 2−n.

Note that if δ is any stopping time, then∣∣∣∣M (t ∧δ ∧ τnk+1)−M (t ∧δ ∧ τ

nk)∣∣∣∣

=∣∣∣∣∣∣Mδ

(t ∧ τ

nk+1)−Mδ (t ∧ τ

nk)∣∣∣∣∣∣≤ 2−n

You can see this is the case by considering the cases, t ∧ δ ≥ τnk+1, t ∧ δ ∈ [τn

k ,τnk+1), and

t ∧ δ < τnk . It is only this approximation property and the fact that the τn

k partition [0,T ]which is important in the following argument.

Now let αn be a localizing sequence such that Mαn is bounded as in Proposition 63.2.2.Thus Mαn (t) ∈ L2 (Ω) and this is all that is needed. In what follows, let δ be a stoppingtime and denote Mα p∧δ by M to save notation. Thus M will be uniformly bounded andfrom the definition of the stopping times τn

k , for t ∈ [0,T ] ,

M (t)≡ ∑k≥0

M(t ∧ τ

nk+1)−M (t ∧ τ

nk) , (63.2.4)

and the terms of the series are eventually 0, as soon as ηnk = ∞.

Therefore,

||M (t)||2 =

∣∣∣∣∣∣∣∣∣∣∑k≥0

M(t ∧ τ

nk+1)−M (t ∧ τ

nk)

∣∣∣∣∣∣∣∣∣∣2

Then this equals= ∑

k≥0

∣∣∣∣M (t ∧ τnk+1)−M (t ∧ τ

nk)∣∣∣∣2

+ ∑j ̸=k

((M(t ∧ τ

nk+1)−M (t ∧ τ

nk)),(M(t ∧ τ

nj+1)−M

(t ∧ τ

nj)))

(63.2.5)

Consider the second sum. It equals

2 ∑k≥0

k−1

∑j=0

((M(t ∧ τ

nk+1)−M (t ∧ τ

nk)),(M(t ∧ τ

nj+1)−M

(t ∧ τ

nj)))

= 2 ∑k≥0

((M(t ∧ τ

nk+1)−M (t ∧ τ

nk)),

k−1

∑j=0

(M(t ∧ τ

nj+1)−M

(t ∧ τ

nj)))

= 2 ∑k≥0

((M(t ∧ τ

nk+1)−M (t ∧ τ

nk)),M (t ∧ τ

nk))

2122 CHAPTER 63. THE QUADRATIC VARIATION OF A MARTINGALEHere is why. The sequence {ti (o)}r eventually equals T for all n sufficiently large.This is because if it did not, it would converge, being bounded above by 7 and then bycontinuity of M, {M (c? (@)) he , would be a Cauchy sequence contrary to the requirementthat(t% (@)[1M (t1(@)) — )(ng (@))|| =2"".= ||M(n() -Note that if 6 is any stopping time, thenMM|M (tA 5A th, 1) —M(tA6A t)I|= |M® (er ch,.) <P Ach)JerYou can see this is the case by considering the cases, t\6 > Ti, ,,t\6 € [T, T,,), andtA 6 < Tt}. It is only this approximation property and the fact that the t? partition [0,7]which is important in the following argument.Now let a, be a localizing sequence such that M@" is bounded as in Proposition 63.2.2.Thus M™ (t) € L? (Q) and this is all that is needed. In what follows, let 5 be a stoppingtime and denote M@»/® by M to save notation. Thus M will be uniformly bounded andfrom the definition of the stopping times T/, for t € [0,7],M(th= VM (tA) —M(tAt%), (63.2.4)k>0and the terms of the series are eventually 0, as soon as 17 = 09.Therefore,2MII? = |] PM (tA thy) —M (CA te)k>0Then this equals= Yi |[M (HA th) —M A tp)II?k>0+ ((M (tA thy) —M (tA t)), (M (tA t,) —M(tAt%))) (63.2.5)j#kConsider the second sum. It equalsk-12¥ YE ((M (Atha) Mee) (M (Atha) —M (4)lINe)—N—~Ns—<—>a>~S+—|&“—_>ars~—S”“"k>0