31.3. DOOB OPTIONAL SAMPLING CONTINUOUS CASE 837

1. Fτ ⊆Fτk and if Ft is normal, then if A ∈Fτk for all k, it follows that A ∈Fτ .

2. A ∈Fτk if and only if A∩[τk = n2−k

]∈Fn2−k for all n.

3. If τ ≤ σ , then τk ≤ σ k.

4. More generally, if τ ≤ σ are two stopping times, then Fτ ⊆Fσ .

5. If X (t) is a right continuous process adapted to the normal filtration Ft and τ is astopping time, then ω→ X (τ (ω)) is Fτ measurable. Here X has values in a Banachspace.

Proof: 1.) Let A ∈Fτ . Then for t ∈ (n2−k,(n+1)2−k], if t < (n+1)2−k,

A∩ [τk ≤ t] = A∩[τk ≤ n2−k

]= A∩

[τ ≤ n2−k

]∈Fn2−k ⊆Ft

so A∈Fτk . If t = (n+1)2−k then A∩ [τk ≤ t]∈F(n+1)2−k =Ft . Thus A∈Fτk and Fτ ⊆Fτk . Now for the other part, Consider A∩ [τ ≤ t] . If t ∈ (n2−k,(n+1)2−k], then [τ ≤ t] =[τk ≤ (n+1)2−k

]and so A∩ [τ ≤ t] =A∩

[τk ≤ (n+1)2−k

]∈F(n+1)2−k ⊆Ft+2−k . Since

the filtration is normal, and for all k, A∩ [τ ≤ t] ∈Ft+2−k , it follows that A∩ [τ ≤ t] ∈Ftand so A ∈Fτ as claimed.

2.) Note [τk = n2−k

]=

[τk ≤ n2−k

]\[τk ≤ (n−1)2−k

]=

[τ ≤ 2−kn

]\[τ ≤ (n−1)2−k

]If A ∈Fτk then by definition,

A∩[τk = n2−k

]= A∩

([τk ≤ n2−k

]\[τk ≤ (n−1)2−k

])= A∩

([τ ≤ 2−kn

]\[τ ≤ (n−1)2−k

])∈Fn2−k

Conversely, ifA∩[τk = n2−k

]∈Fn2−k

for all n, then consider A∩ [τk ≤ t] for t ∈ (n2−k,(n+1)2−k]. Is A∩ [τk ≤ t] ∈ Ft? Ift < (n+1)2−k, then A∩ [τk ≤ t] = ∑

nj=1 A∩

[τk = j2−k

]∈Fn2−k ⊆Ft . If t = (n+1)2−k,

A∩ [τk ≤ t] = ∑n+1j=1 A∩

[τk = j2−k

]∈Ft and so A ∈F(n+1)2−k ⊆Fτk . This proves 2.).

3.) The values of both τk and σ k are n2−k for some nonnegative integer n. τk equalsn2−k on τ−1

((n−1)2−k,n2−k]

). Thus on this set, σ cannot be smaller than or equal to

(n−1)2−k. Hence σ k is at least n2−k.4.) Let τ ≤ σ and let A ∈ Fτ . A∩ [σ ≤ t] = A∩ [τ ≤ t]∩ [σ ≤ t] because τ ≤ σ .

However, A∩ [τ ≤ t]∈Ft and [σ ≤ t]∈Ft so the right side is in Ft which means A∈Fσ .5.)Let U be an open set.

X (τk)−1 (U)∩ [τk < t] = ∪ j2−k≤{t}k X

(j2−k

)−1(U) ∈F{t}k .

Now say t ∈((n2−k,(n+1)2−k]

). If t < (n+1)2−k, then X (τk)

−1 (U) ∩ [τk ≤ t] wasjust given. It is in Fn2−k = F{t}k ⊆ Ft . Otherwise t = (n+1)2−k and in this case,

31.3. DOOB OPTIONAL SAMPLING CONTINUOUS CASE 837l. #, C F;, and if F; is normal, then if A € ¥;, for all k, it follows that A € F;.. A€ F,, if and only if AN [t, =n2~*] € F,y-« for all n.. If T<o, then tT; < Ox.. More generally, if tT < o are two stopping times, then ¥; C Fg.nan FF WwW NY. If X (t) is a right continuous process adapted to the normal filtration Y, and T is astopping time, then @ — X (t(@)) is -#, measurable. Here X has values in a Banachspace.Proof: 1.) Let A € ¥-. Then fort € (n2-*, (n+1)2, ift < (n+1)2-*,AN [tT <t]=AN lr < n2*] =AN k < n2*] CF 4 CF;soA € Fz. Ift=(n+1)2-* thenAn[t; <t] € F ny iy-k = Fy. Thus A € Fz, and Fz C¥,,. Now for the other part, Consider AN [t < ¢]. If t € (n2~*, (n +1) 2), then [t < t] =[te < (n+1)2~] andsoAN[t <1] =AN[t; < (n+1)2~] © Feng tyne © F,49-- Sincethe filtration is normal, and for all k, AN [t <t] € F,,5-«, it follows that AN [t < t] © F,and so A € F#; as claimed.2.) Note[ze = no] = lr < no] \ Ir < (n— 2*|= i < 2%] \ E <(n— 2"If A € Fz, then by definition,An It =n2*| = AN (|x < no \ lz < (n— 1)2*])= AN (|< < 2*n| \ lz <(n— 1)2*]) EF»Conversely, ifAN lr = no] EF ikfor all n, then consider AN [t, < ft] for t € (n2~*, (n+ 1) 2-4]. Is AN [ty <t] € FA? Ift<(n+1)2~, then AN [ty <t]=L"_, AN [te = j2-*] © Fyp-e C.F. Met = (n+ 1)2%,AN[t% <t]= retan [t, = j2*] € F, and soA € F ny 1)0-k © Fz,. This proves 2.).3.) The values of both t; and o; are n2~* for some nonnegative integer n. tT; equalsn2~* on t~! ((n—1)2~*,n2~*]) . Thus on this set, o cannot be smaller than or equal to(n—1)2~*. Hence 0; is at least n2~*.4.) Let t< o and let A € F,. AN[o <t] =AN[t<t]N[o <t] because tT < o.However, AN |t < t] € FY, and [o < t] € .F; so the right side is in ¥, which means A € Fg.5.)Let U be an open set.-1X(t) (UW) N[tk <1] =Up-reiy,X (2%) UW) Fey,Now say ¢ € ((n2-*,(n+1)2-"}). If t < (n+1)2~, then X (t;)~' (U) A [te <1] wasjust given. It is in Fy-% = Fy, C F,. Otherwise ¢ = (n+ 1)2-* and in this case,