• Aucun résultat trouvé

A CROSS SECTION THEOREM. I

N/A
N/A
Protected

Academic year: 2022

Partager "A CROSS SECTION THEOREM. I"

Copied!
20
0
0

Texte intégral

(1)

A CROSS SECTION THEOREM. I

PETER GRAY

The theory surrounding a one-parameter stochastic process(Xt)t∈R+ has appli- cations in many elds. The predictable projection (pXt)t∈R+and the predictable dual projection(Xtp)t∈R+ for the process(Xt)t∈R+ both play roles in many such applications. Several results concerning these projections are derived using the Cross Section Theorem for predictable subsets of the set×R+,which states that for any predictable setA and any >0there exists a predictable stopping timeS such that [S]A and P(π(A))P(π([S])), where [S] denotes the graph ofS, π represents the projection onto the setΩ,and P is the probability measure onΩ.The goal of this paper is to extend the Cross Section Theorem to predictable subsets of the set×R2+,thereby paving the way for the derivation of results pertaining to a two-parameter stochastic process (Xs,t)s,t∈R+.Such an extension of the Cross Section Theorem is dicult because, while the setR+ of positive real numbers is totally ordered, the set R2+ of ordered pairs of positive real numbers is not. We combat this diculty by rst collecting results using the double ltrationFs,t =Fs−∩ F.Once this is done, we will be able to consider the more general double ltrationFs,t=Fs∨t−.The main result of this paper is a cross section theorem for predictable subsets of×R2+ relative to the double ltration (Fs,t)s,t∈R+satisfyingFs,t=Fs−fors, t0.This theorem states that for any predictable subset A of ×R2+ and any > 0 there is a predictable stopping timeZ such that[Z]AandP(π(A))P(π([Z])).

AMS 2000 Subject Classication: Primary 60G20; Secondary 60G40.

Key words: two-parameter stochastic process, cross section.

1. PRELIMINARIES

Notation and Terminology 1.1. The following will be used in the sequel.

1.1a (Ω,F, P) is a probability space.

1.1b R+ is the set of non-negative real numbers. N is the set of natural numbers.

1.1cB(R+)is the Borelσ-algebra generated by the intervals (s, t] ofR+. B R2+

is the Borelσ-algebra generated by the rectangles(s, t]×(u, v]ofR2+. 1.1d A functionX: Ω×R2+→Ris called a two-parameter process, and is denoted (Xs,t). For literature pertaining to two-parameter processes, the reader is referred to [Dz], [C-W], [L], and [W].

REV. ROUMAINE MATH. PURES APPL., 53 (2008), 4, 297316

(2)

1.1e F ⊗ B(R+) is theσ-algebra generated by the semiring F × B(R+).

F ⊗ B(R2+) is theσ-algebra generated by the semiringF × B(R2+).

1.1f (Ft)t∈R+ is a ltration; that is,

• for eacht≥0,Ft is aσ-algebra contained inF, and

• Fs⊂ Ft ifs≤t.

We will simply write(Ft), and will assume that the ltration satises the usual conditions:

• F0 contains all the negligible sets (that is,(Ft)is complete), and

• Ft= T

s>t

Fs for everyt≥0 (that is,(Ft)is right continuous).

See 1.1j below for another assumption about(Ft).

1.1g A functionS : Ω→[0,∞]is a stopping time for the ltration(Ft) if {S≤t} ∈ Ft for every t≥0.

LetS, T be two stopping times. The stochastic interval (S, T]is the set {($, t)∈Ω×R+|S($)< t≤T($)},

while [S, T)is the set

{($, t)∈Ω×R+| S($)≤t < T($)}.

The stochastic intervals (S, T) and[S, T]are dened in a similar fashion.

1.1h The predictableσ-algebraP1 of subsets ofΩ×R+ is theσ-algebra generated by the sets A×(s, t] and B × {0}, where A ∈ Fs− and B ∈ F0. A stopping time S is predictable if the stochastic interval [S,∞) is a pre- dictable set.

1.1i LetS be a stopping time. FS denotes the σ-algebra {A∈ F |A∩ {S ≤t} ∈ Ft for everyt≥0}

while FS− denotes theσ-algebra generated by the sets inF0 as well as sets of the form A∩ {S > t}, wheret≥0 andA∈ Ft.

1.1j(Gs)s∈

R+ is the ltration dened by

• G0=F0 and

• Gs=Fs− for everys≥0.

We will simply write(Gs), and will assume that the ltration(Ft)is such that GS = FS− for every predictable stopping time S for the ltration (Gs).

For example, this is the case when Fs =Fbsc for every s∈ R+, where bsc is the largest integer less than or equal to s.

1.1k(Fs,t)s,t∈

R+ is a double ltration; that is,

• for eachs, t≥0,Fs,t is aσ-algebra contained inF, and

• Fs,t⊂ Fu,v ifs≤u andt≤v.

We will simply write (Fs,t), and will assume that Fs,t = Gs for every s, t≥0.

(3)

1.1l LetB ⊂Ω×R2+. B($) is the set{(s, t)∈R2+ |($, s, t)∈B}. π[B], called the projection ofB, is the set{$∈Ω|($, s, t)∈Bfor somes, t∈R+}. 1.1m The point(∞,∞)will be denoted by∞. Therefore, the inequality (s, t)<∞ means thats <∞ and t <∞.

1.1n Let g : Ω → [0,∞]×[0,∞] be a function. [g] denotes the set {($, s, t)∈Ω×R2+|g($) = (s, t)}, called the graph of g.

We now commence our study of the real-valued two-parameter process (Xs,t).

Denition 1.2. LetX: Ω×R2+→Rbe a two-parameter process.

1.2a(Xs,t) is left continuous if for everys0, t0∈R+and$∈Ω, we have

(s,t)→(slim0,t0) 0≤s≤s0 0≤t≤t0

Xs,t($) =Xs0,t0($).

1.2b (Xs,t) is adapted to the ltration (Fs,t) if for each s, t ∈ R+ the random variable Xs,t isFs,t-measurable.

1.2c The predictable σ-algebra of subsets of Ω×R2+ is the σ-algebra generated by left continuous two parameter processes(Xs,t)which are adapted to (Fs,t). We denote thisσ-algebra byP.

Proposition 1.3. P is generated by the sets (S,∞)×[0, r1] and A× {0} ×[0, r2], where r1, r2 ∈R+, S : Ω → [0,∞] is a stopping time for (Gs), and A∈G0.

Proof. Let(Xs,t)be left continuous and adapted to(Fs,t). Set Ys,tn :=X0,01{0}×{0}(s, t) +

X

k=0

X0,k

n1{0}×(k

n,k+1n ](s, t)+

+

X

m=0

Xm

n,01(m

n,m+1n ]×{0}(s, t) +

X

k,m=0

Xm n,nk1(m

n,m+1n ]×(nk,k+1n ](s, t), where 1 is the indicator function, and n ∈N. Since (Xs,t) is left continuous, Yn→X pointwise asn→ ∞.

Fixm, n,and k,and consider the process R= (Rs,t), where Rs,t :=Xm

n,kn1(m

n,m+1n ]×(kn,k+1n ](s, t).

SinceXm

n,nk isGm

n-measurable,Ris the pointwise limit of processes of the form

p

P

i=1

αi1Ai×(m

n,m+1n ]×(k

n,k+1n ],where αi ∈Rand Ai ∈ Gm

n for every index i. But

(4)

for every index iwe have Ai×(mn,m+1n ] = (Si, Ti], where Si =

m

n on Ai

∞ on Aci, which is a stopping time for (Gs), and

Ti =

( m+1

n on Ai

∞ on Aci,

which also is a stopping time for (Gs). Therefore, R is the pointwise limit of processes of the form Pp

i=1

αi1(Si,Ti]×(k

n,k+1n ], where for every index i we have αi ∈ R, and Si, Ti are stopping times for (Gs). By considering in a similar manner the processes X0,k

n1{0}×(k

n,k+1n ]andXm

n,01(m

n,m+1n ]×{0},we obtainP ⊂ σ{(S1, T1]×(r1, s1], A× {0}×(r2, s2],(S2, T2]×{0}},wherer1, r2, s1, s2 ∈R+, A ∈ G0, and S1, S2, T1, and T2 are stopping times for (Gs) ; = σ{(S,∞)× [0, r1], A× {0} ×[0, r2]}, wherer1, r2 ∈R+,A∈ G0, and S is a stopping time for(Gs). In fact we haveP =σ{(S,∞)×[0, r1], A× {0} ×[0, r2]}because the sets (S,∞)×[0, r1]and A× {0} ×[0, r2]are elements ofP.

Remark. For alternative generators ofP, see [D], p. 323.

2. STOPPING TIMES

Denition 1.4. LetZ : Ω→[0,∞]×[0,∞]be a function.

1.4a The functionZ is a stopping time for the ltration (Fs,t) if {Z ≤ z} ∈ Fz for every z∈R2+.

1.4b The stopping timeZ = (S, T) is predictable if

• S is a predictable stopping time for(Gs), and

• {S <∞} ⊂ {T <∞}.

Proposition 1.5. Let Zn= (Sn, Tn)be a sequence of stopping times for any double ltration (Fs,t0 ) that is right continuous in t. Assume that (Sn) is increasing. Then Z:= (supnSn,lim supnTn)is a stopping time for (Fs,t0 ).

Proof. Let(s, t)∈R2+. We must show that {Z ≤(s, t)} ∈ Fs,t. For each index n and each(r, u)∈R2+ we have {Zn≤(r, u)}={Sn≤r} ∩ {Tn≤u} ∈ Fr,u by hypothesis. SetTn0 :=

W

i=n

Ti.Then lim supnTn= limnTn0.We have {Z ≤(s, t)}=

n sup

n

Sn,lim

n Tn0

≤(s, t) o

=

(5)

= n

sup

n

Sn≤s o

∩n

limn Tn0 ≤t o

=

\

n=1

{Sn≤s} ∩

\

p=N

[

k=1

\

m=k

{Tm0 ≤t+p}, where N is any natural number, and (p) is any sequence of positive numbers decreasing to 0;

=

\

p=N

[

k=1

"

\

n=1

{Sn≤s} ∩

\

m=k

{Tm∨Tm+1∨ · · · ≤t+p}

#

=

\

p=N

[

k=1

"

\

n=1

{Sn≤s} ∩

\

m=k

{Tm ≤t+p} ∩ {Tm+1 ≤t+p} ∩ · · ·

#

=

\

p=N

[

k=1

\

m=k

[{Sm ≤s} ∩ {Sm+1 ≤s} ∩ · · · ∩ {Tm ≤t+p} ∩ · · ·]

since (Sn) is increasing;

=

\

p=N

[

k=1

\

m=k

[{Zm ≤(s, t+p)} ∩ {Zm+1 ≤(s, t+p)} ∩ · · ·]∈ Fs,t+0

N,

N ∈N being arbitrary. Since Fs,t0

is right continuous in t, we conclude that {Z ≤(s, t)} ∈ Fs,t0 .

Proposition 1.6. Let n ∈ N, and let Zi = (Si, Ti), i = 1, . . . , n, be a nite set of stopping times for (Fs,t) such that each Si is a stopping time for (Gs). Set Ai :={infk≤nSk=Si} ∩ {Si 6=S1} ∩ {Si6=S2} ∩ · · · ∩ {Si 6=Si−1}, i= 1, . . . , n,and set S :=

n

P

i=1

Si1Ai and T :=

n

P

i=1

Ti1Ai.

Denote (S, T) by minfi(Zi), and setZ := minfi(Zi).Then

• S is a stopping time for (Gs),

• Z is a stopping time for(Fs,t), and

• if Ti($) = ∞ ⇒ Si($) = ∞ for each index i then T($) = ∞ ⇒ S($) =∞.

Proof. We rst prove thatS is a stopping time for(Gs).

Note that the sets Ai, i = 1, . . . , n, are pairwise disjoint. Further, we have Ai ∈ GSi for each i(see [M], p. 20), and S

i=1

Ai = Ω.It follows that S is a stopping time for (Gs).

Next, we prove that Z is a stopping time for(Fs,t). Let (s, t)∈R2+.We must show that {Z ≤(s, t)} ∈ Gs.By hypothesis, {Zi ≤(s, t)} ∈ Gs for eachi.

(6)

So, we have {Z ≤(s, t)}=

n

[

i=1

{Zi ≤(s, t)} ∩Ai =

n

[

i=1

({Si≤s} ∩Ai)∩ {Zi ≤(s, t)} ∈ Gs. Lastly, we prove the third assertion of the proposition. We haveT($) =

∞ ⇒ Ti($) = ∞ for some index i such that $ ∈ Ai; ⇒ Si($) = ∞ by hypothesis ⇒S($) =∞ since $∈Ai.

3. PROJECTIONS π[A]

In this section we establish a key result concerning the projectionπ[A]of an F ⊗ B(R2+)-measurable set A. It will emerge that π[A]is F-measurable.

Proposition 1.7. Let A∈ F ⊗ B(R2+)and let g be a function such that π[[g]]∈ F, and [g]⊂A on π[A]∩π[[g]]. Then π[[g]∩A]∈ F.

Proof. Let A be the collection of sets B ∈ F ⊗ B(R2+) such that for any function h : Ω → [0,∞]×[0,∞] satisfying π[[h]] ∈ F and [h] ⊂ B on π[B]∩π[[g]] we have π[[h]∩B]∈ F. Let Rbe the ring generated by the sets (C×(s, t]×(u, v])∩Ω×R2+ andΩ×R2+, whereC∈ F and s, t, u, v∈R.

First, we show that A contains R. Let B ∈ R and let a function h be such that π[[h]]∈ F and [h]⊂B on π[B]∩π[[h]]. Without loss in generality we consider B =C×(s, t]×(u, v], where C ∈ F and s < t and u < v. We have π[[h]∩B] = (π[[h]])∩C ∈ F. Hence, B ∈ A.

Next, we show that A is a monotone class. Let (Bn) be a monotone sequence from A. Assume rst that (Bn) is increasing. Set B := S

n

Bn. Let a function h1 satisfy π[[h1]] ∈ F and [h1] ⊂ B on π[B]∩π[[h1]]. Let h0 : Ω → [0,∞]×[0,∞] be a function that satises the conditions set out below. We require that π[[h0]] = π[[h1]], h0 = h1 on (π[B])c, [h0] ⊂ B1 on (π[B1])∩π[h1], and[h0]⊂Bi on(π[Bi]\π[Bi−1])∩π[h1]fori= 2,3,4, . . .. Such a function h0 exists. Note, we haveπ[[h0]]∈ F, and[h0]⊂Bi on[π[Bi]∩π[[h0]]

for i ∈ N. Therefore, for each index i we have π[[h0]∩Bi] ∈ F (since each Bi is in A). Hence we have π[[h1]∩B] = π[[h0]∩B] from the denition of h0; = π[[h0]∩S

i

Bi] = S

i

π[[h0]∩Bi] ∈ F. Second, assume that (Bn) is decreasing. Set B0 :T

n

Bn. Let a functionh2 satisfy π[[h2]]∈ F and[h2]⊂B0 on π[B0]∩π[[h2]]. Let h00 : Ω → [0,∞]×[0,∞] be a function that satises the conditions set out below. We require that π[[h00]] = π[[h2]], h00 = h2 on (π[B1])c∪π[B0], and[h00]⊂Bi on (π[Bi]\π[Bi+1])∩π[h2] for i∈N. Such a function h00 exists. Note, we haveπ[[h00]]∈ F, and [h00]⊂Bi onπ[Bi]∩π[[h00]]

for i∈N. Therefore, for each indexi we haveπ[[h00]∩Bi]∈ F (since eachBi

(7)

is inA). Hence we haveπ[[h2]∩B0] =π[[h00]∩B0](from the denition ofh00);

=π[[h00]∩T

i

Bi] =T

i

π[[h00]∩Bi]∈ F (since[h00]is a graph).

We have shown that A is a monotone class which contains the ring of generators ofF ⊗ B(R2+). Because this ring contains the whole spaceΩ×R2+, we conclude from the Monotone Class theorem (see [D-M], p. 13-I) that A is equal to F ⊗ B(R2+). The statement of the proposition is now seen to be true.

Corollary 1.8. Let A∈ F ⊗ B(R2+). Then π[A]∈ F.

Proof. We are able to nd a function h: Ω → [0,∞]×[0,∞]such that π[[h]] = Ω and [h] ⊂ A on π[A]∩π[[h]]. From Proposition 1.7 we obtain π[[h]∩A] ∈ F. For such a function h we have π[[h]∩A] = π[A]. Hence we have π[A]∈ F.

4. SETS Kδ

We continue to prepare for the Cross Section theorem by introducing a special collection Kδ of predictable sets with compact cross sections.

Denition 1.9. LetK⊂Ω×R2+.

1.9a We say thatK has compact cross sections if for each$∈π[K]the cross section K($)⊂R2+ is compact.

1.9b The cross sectional closure ofK, denotedK, is given byK($) = (K($)) for every $ ∈ Ω, where (K($)) denotes the closure of the cross section K($)⊂R2+.

Proposition 1.10. Let N ∈N, and let (Kn)be a sequence of subsets of Ω×R2+with compact cross sections. Assume that each set Knhas the following properties: π[B∩Kn]∈ F for every F ⊗ B(R2+)-measurable setB; and there is a stopping time Zn= (Sn, Tn) such that

•Sn is a predictable stopping time for (Gs),

•[Zn]⊂Kn,

• {Zn<∞}=π[Kn],

•Tn($) =∞ ⇒Sn($) =∞ for every $∈Ω, and

•(s, t)∈Kn($)⇒s≥Sn($) for every$∈Ω. Set K :=

N

S

n=1

Kn andK0 :=

T

n=1

Kn. Then K andK0 both have compact cross sections, and K has all the above properties of the sets Kn.Further, if the sequence (Kn) is decreasing, then the set K0 also has all the above properties.

Proof. The nite union and the countable intersection of compact subsets of R2+ are compact. Therefore, K and K0 have compact cross sections.

(8)

Next, let B ∈ F ⊗ B(R2+). We have π[B ∩K] = π[B ∩ SN

n=1

Kn] = SN

n=1

π[B∩Kn] ∈ F, and π[B ∩K0] = π[B

T

n=1

Kn] =

T

n=1

π[B ∩Kn] if (Kn) is decreasing, since each setB∩Kn has compact cross sections;∈ F.

We now show thatK has the second property set out in the proposition.

Set Z := minfn=1,...,N(Zn). (See Proposition 1.6 for the denition of minf.) From Proposition 1.6 we know that Z is a stopping time, and that if we write Z = (S, T), then S is a stopping time for (Gs). Note, the stopping time S is a predictable stopping time because S is the minimum of nitely many predictable stopping times Sn,n= 1, . . . , N.Next, let$∈[Z]. ThenZ($)<

∞. There is an index n such that Z($) = Zn($). From this equality we deduce that Zn($) < ∞. By hypothesis, we have ($, Zn($)) ∈ Kn. Hence ($, Z($)) = ($, Zn($))∈Kn⊂K.

We now show that {Z < ∞} = π[K]. We have π[K] = π[

N

S

n=1

Kn] =

N

S

n=1

π[Kn] =

N

S

n=1

{Zn < ∞} by a property of the set Kn =

N

S

n=1

{Sn < ∞} ∩ {Tn < ∞}

=

N

S

n=1

{Sn <∞} by a property of the function Zn; =

N

S

n=1

[{Sn <

∞} ∩ {infn0Sn0 =Sn} ∩ T

n00<n

{Sn6=Sn00}] ={Z <∞}.

Penultimately, note that by Proposition 1.6 we have T($) = ∞ ⇒ S($) =∞ for every $∈Ω.

Lastly, we show that (s, t) ∈ K($) implies that s ≥ S($) for every

$ ∈ Ω. We have (s, t) ∈ K($) ⇒ ($, s, t) ∈ K ⇒ ($, s, t) ∈ Kn for some index n⇒s≥Sn($) by hypothesis;≥S($)since S = infn=1,...,N Sn.

To complete the proof, assume that the sequence (Kn) is decreasing.

We verify that the set K0 has all ve properties that were listed. Let Z :=

(supnSn,lim supnTn). Note that Z is a stopping time. To see this, it is enough to show (by Proposition 1.5) that the sequence (Sn) is increasing.

Let $ ∈ Ω and n0 ∈ N, and assume rst that Zn0+1($) < ∞. Then ($, Sn0+1($), Tn0+1($)) ∈ Kn0+1 by a property of the set Kn0+1 ⊂ Kn0. So,($, Sn0+1($), Tn0+1($))∈Kn0($).By a property of the setKn0, we have Sn0+1($)≥Sn0($). Next, assume thatZn0+1($) =∞. ThenSn0+1($) =∞ or Tn0+1($) = ∞. We have Sn0+1($) = ∞ by a property of the set Kn0+1. HenceSn0+1($)≥Sn0($),and we have shown thatZ is a stopping time. Now note, the stopping time supnSn is a predictable stopping time for (Gs). This follows since each Sn is a predictable stopping time for(Gs), and since(Sn)is increasing.

(9)

We now prove that [Z] ⊂ K. Let Z($) < ∞. We must show that ($, Z($)) ∈ K0. Since Z($) < ∞, we have supnSn($) < ∞. Therefore, Sn($)<∞for every indexn. It follows from a property ofKn thatZn($)<

∞ for every n. So, ($, Zn($)) ∈ Kn for every n, since [Zn] ⊂ Kn for all n. There is a subsequence (Tnk($))k of (Tn($))n (that depends on $) such that Tnk($) → lim supnTn($) ask → ∞. Since (Sn($)) is increasing, we have Snk($) → supnSn($) as k → ∞. Thus, we have ($, Znk($)) = ($, Snk($), Tnk($))→($, Z($))ask→ ∞. For eachkwe have($, Znk($))

∈ Knk. Since by assumption the sequence (Kn) is decreasing, ($, Znk($))k is a sequence that eventually belongs to each of the sets Knk, k ∈ N. By the compactness of the cross sections involved, ($, Z($)) = limk($, Znk($)) belongs to each of the sets Knk,k∈N. So, we have($, Z($))∈T

k

Knk =K0. We now show that {Z < ∞} = π[K0]. First, we establish that {Z <

∞} = T

n

{Zn < ∞}. In fact, let $ ∈ {Z < ∞}. Suppose there is a number n0 ∈ N such that Zn0($) ≮ ∞. Then either Sn0($) = ∞ or Tn0($) =

∞. Since Tn0($) = ∞ ⇒ Sn0($) = ∞, we obtain Sn0($) = ∞. Since (Sn) is increasing, we have Sm($) = ∞ for each m ≥ n0. It follows that supnSn($) =∞. Therefore,Z($) = (supnSn($),lim supnTn($))≮∞, and we have reached a contradiction. Because of this contradiction, we conclude that Zn($)<∞ for everyn∈N. Hence, we have{Z <∞} ⊂T

n

Zn<∞}. Next, let$∈T

n

{Zn<∞}. Then for eachnwe haveZn($)<∞. Note, for each n we have [Zn] ⊂ Kn. Therefore, for each n we have ($, Zn($)) ∈ Kn ⊂ K1. Since K1 has compact cross sections, there is M ∈ R2+ (that depends on$)such that(s, t)≤M ∀(s, t)∈K1($). Thus, for eachnwe have Zn($) ≤M. Consequently, we have Z($) = (supnSn($),lim supnTn($))≤ M. Accordingly, we have $ ∈ {Z < ∞}. Hence, T

n

{Zn < ∞} ⊂ {Z < ∞}. Therefore, we have {Z < ∞} = T

n

{Zn < ∞} (by the preceding lines); = T

n

π[Kn]since {Zn <∞}=π[Kn]for all n;=π[T

n

Kn]since all sets Kn have compact cross sections; =π[K0].

We now show that lim supnTn($) = ∞ ⇒ supnSn($) = ∞ for every

$∈Ω. In fact, for each$∈Ωwe havelim supnTn($) =∞ ⇒Z($)≮∞ ⇒ Zn0($) ≮ ∞ for some index n0, since {Z < ∞} = T

n

{Zn < ∞}; ⇒ either Sn0($) =∞ or Tn0($) =∞ ⇒ Sn0($) =∞ (as Tn0($) = ∞ ⇒ Sn0($) =

∞);⇒supnSn($) =∞.

Lastly, we prove that(s, t)∈K0($)⇒s≥supnSn($) for every $∈Ω. In fact, for each $∈Ω we have (s, t)∈K0($)⇒($, s, t)∈K0 =

T

n=1

Kn

(10)

($, s, t) ∈ Kn for every n; ⇒ (s, t) ∈ Kn($) for every n; ⇒ s ≥ Sn($) for everyn by a property ofKn;⇒s≥supnSn($).

Denition 1.11. We dene the setK to be the collection of nite unions

N

S

i=1

[Si+i, Ti∧ni]×[ri, r0i], where ni, N ∈N, i > 0, ri, r0i ∈ R+, and Si, Ti are stopping times for (Gs) for every index i. We dene the set Kδ to be the collection of countable intersections of sets from K.

Proposition 1.12. The set Kδis closed under nite union and countable intersection. Further, the elements of Kδ have all the properties that were presented in Proposition 1.10.

Proof. It is evident thatKδ is closed under countable intersection.

Next, we show that Kδ is closed under nite union. Let M ∈ N and let (Ki)i=1,...,M be a nite family from Kδ. For each index i we may write Ki =

T

j=1

Ki,j, where each set Ki,j is an element of K. Then we have MS

i=1

Ki=

M

S

i=1

T

j=1

Ki,j =

T

j1,j2,...,jM=1

K1,j1 ∪K2,j2 ∪ · · · ∪ KM,jM, which is a countable intersection of elements of K, since each set K1,j1∪K2,j2 ∪ · · · ∪KM,jM is an element of K. Accordingly, the set MS

i=1

Ki is an element ofKδ.

We now prove that the elements ofKδpossess all the properties that were presented in Proposition 1.10. Let S, T be stopping times for (Gs), let >0, let n ∈ N, and let r, r0 ∈ R+. In view of Proposition 1.10, it is enough to show that the set K0:= [S+,T∧n]×[r, r0] possesses all the aforementioned properties. Without loss in generality, we assume that S + ≤ T ∧n and r ≤ r0. Let B ∈ F ⊗ B(R2+). We will show that π[B∩K0] ∈ F. Let (δm) be a sequence of positive numbers decreasing to 0. For each indexm, denote by Km the set [S+−δm, T ∧n+δm]×[r−δm, r0m], and by Km the set (S +−δm, T ∧n+δm) ×(r −δm, r0m). We have π[B ∩K0] =

T

m=1

π[B ∩Km]. In fact, we have T

m=1

π[B ∩Km] ⊂

T

m=1

π[B ∩Km] since B ⊂ B; = π[B

T

m=1

Km] since (Km) is decreasing, and since each set B∩Km has compact cross sections; =π[B∩K0]. On the other hand, for each index mwe haveπ[B∗ ∩K0]⊂π[B∩Km]since K0⊂Km;=π[B∩Km] since every cross section of the set Km is an open ball when S + < ∞;

⊂π[B∩Km]sinceKm ⊂Km. Therefore,π[B∩K0]⊂

T

m=1

π[B∩Km]. So, we

(11)

haveπ[B∩K0] =

T

m=1

π[B∩Km]∈ F by Corollary 1.8, sinceKm ∈ F ⊗B(R2+) for every index m.

We now reveal the remaining properties of the setK0.SetZ := (S+, r). Note, Z is a stopping time. In fact, let (s, t) ∈ R2+. Then {Z ≤ (s, t)} = {S ≤s if r≤t

∅ otherwise . In either case we have{Z ≤(s, t)} ∈ Gs. We complete the proof by making the following ve observations. The stopping time S+ is a predictable stopping time since > 0. The graph [Z] is a subset of K0, since by assumption S+≤T ∧nandr ≤r0. We have{Z <∞}={S+ <

∞} = π[[S +, T ∧n]], since S + ≤ T ∧n = π[K0]. For each $ ∈ Ω we have r($) = ∞ ⇒ S($) + = ∞, since r < ∞. For each $ ∈ Ω we have (s, t)∈K0($)⇒s∈[S($)+, T($)∧n]andt∈[r, r0]⇒s≥(S+)($).

5. THE CROSS SECTION THEOREM

We begin with some important precursors to the Cross Section theorem for predictable subsets of Ω×R2+.

Denition 1.13. LetRbe the ring of subsets ofR2+ generated by the sets (z, z0] such that z, z0 ∈ R2+, with z < z0. The measure µ is the set function fromR intoR+that is given byµ((z, z0]) =the area of the rectangle(z, z0]for everyz < z0, and which is additively extended toR.

Remark. The measure µ is the Lebesgue measure on R, and can be extended to a sigma-additive measure on B(R2+), with values in [0,∞]. We still denote by µ this sigma-additive extension.

Lemma 1.14. Let B ∈ F ⊗ B(R2+) and $∈Ω. Then B($)∈ B(R2+). Proof. LetF ⊂Ω, z, z0 ∈R2+, and(Bn)⊂Ω×R2+. Then(F×(z, z0])($)∈ B(R2+), while(B1−B2)($) =B1($)−B2($) and(S

n

Bn)($) =S

n

Bn($).

Notation 1.15. Let A, B∈ F ⊗ B(R2+),λ∈R+, and z∈R2+. We denote by ABλ,z the set {$∈Ω|µ((A∩Ω×[0, z])($))> λµ((B∩Ω×[0, z])($))}. WhenA⊂B and 0≤λ <1, the reader might think of the setABλ,z as being the projection of the portion of A∩Ω×[0, z]that lls some Borel cross section of B∩Ω×[0, z]by a factor of100λpercent or more.

Proposition 1.16. Let A, B ∈ F ⊗ B(R2+), and let z ∈ R2+. Then for every λ∈R+ we have ABλ,z ∈ F.

(12)

Proof. Let R be the ring generated by the sets Sn

i=1

Fi ×(zi, zi0], where n ∈ N, Fi ∈ F, and zi, zi0 ∈ R2+ for every index i. Fix A0 ∈ R. Write A0∩Ω×[0, z] =

n0

S

i=1

F0,i×B0,i, where the setsF0,i∈ F are mutually disjoint, and where each set B0,i is an element of B(R2+). Let A be the collection of sets C in F ⊗ B(R2+) such that for every λ ∈ R+ the set A0Cλ,z is an element of F. We rst show that A contains the ring R. Let C1 ∈ R and let λ ∈ R+. Write C1∩Ω×[0, z] =

n1

S

i=1

F1,i×B1,i, where the sets F1,i ∈ F are mutually disjoint, and where each set B1,i is an element ofB(R2+). Then A0C1λ,z = S

(i,j)

F0,i∩F1,j, where the union is taken over all pairs (i, j) such that µ(B0,i) > λµ(B1,j). Therefore, we have A0C1λ,z ∈ F. Accordingly, we have C1 ∈ A, since λ∈R+ was arbitrary. Next, let(Cn) be an increasing sequence from A. SetC :=

S

n=1

Cn. We will show that C ∈ A. Note, for each $ ∈Ω we have

µ((C∩Ω×[0, z])($)) = lim

n→∞µ((Cn∩Ω×[0, z])($)),

since µis sigma-additive. Let λ∈R+, and let (δm) be a sequence of positive numbers decreasing to0, withλ−δ1>0. We haveA0Cλ,z =

S

m=1

T

n=1

A0Cnλ−δm,z

(since µis nite on [0, z])∈ F. Since λ∈R+ was arbitrary, we conclude that C ∈ A. Now, let(Cn0) be a decreasing sequence fromA. SetC0:=

T

n=1

Cn0. We will show that C0∈ A. Note, for each$∈Ωwe have

µ((C0∩Ω×[0, z])($)) = lim

n→∞µ((Cn0 ∩Ω×[0, z])($)),

sinceµis sigma-additive and sinceµ((C10∩Ω×[0, z])($))<∞for every$∈Ω. Let λ∈R+. We haveA0Cλ,z0 =

S

n=1

A0Cnλ,z0 . Sinceλ∈R+ was arbitrary, we conclude that C0 ∈ A. By applying the Monotone Class theorem we deduce that for everyA0 ∈R,C∈ F ⊗B(R2+), andλ∈R+we haveA0Cλ,z ∈ F. Now, letA0 be the collection of setsA0 inF ⊗ B(R2+) such that for everyλ∈R+ the setA0Bλ,zis an element ofF. By the above we haveR⊂ A0. Next, let(A0n)be an increasing sequence from A0. SetA0 :=

S

n=1

A0n. We will show thatA0 ∈ A0. Let λ ∈ R+. Since µ is sigma-additive, we have A0Bλ,z =

S

n=1

A0nBλ,z ∈ F. Since λ∈ R+ was arbitrary, we conclude thatA0 ∈ A. Lastly, let (A00n) be a

(13)

decreasing sequence from A0. SetA00 :=

T

n=1

A00n. We will show that A00 ∈ A0. Let λ∈ R+, and let (δm) be a sequence of positive numbers decreasing to 0. Since µ is sigma-additive, we have A00Bλ,z =

S

m=1

T

n=1

A00nBλ+δm,z ∈ F. Since λ ∈R+ was arbitrary, we conclude that A00 ∈ A0. We have shown that A0 is a monotone class containing R.By applying the Monotone Class theorem we deduce that for every λ∈R+ the set ABλ,z is an element ofF.

Proposition 1.17. Letz∈R2+, let(An) be an increasing sequence from F ⊗ B(R2+), and letA=

S

n=1

An. LetB be a measurable subset ofA, let >0, and let 0≤λ <1. There is an indexN ∈N such thatP(π[B∩Ω×[0, z]])− P((B∩AN)Bλ,z)≤.

Proof. Sinceµis sigma-additive, for each$∈Ω we have

µ((B∩An∩Ω×[0, z])($))%µ((B∩A∩Ω×[0, z])($)) =µ((B∩Ω×[0, z])($)).

Therefore, since λ <1 and sincez ∈R2+ is nite, for each $∈Ω there is an indexN($)such thatµ((B∩AN($)∩Ω×[0, z])($))> λµ((B∩Ω×[0, z])($)).

Hence, we have (B∩An)Bλ,z %(B∩A)Bλ,z. Since each set(B∩An)Bλ,z is an element of F (see Proposition 1.16), we have P((B∩An)Bλ,z) % P((B∩ A)Bλ,z) =P(BBλ,z) =P(π[B∩Ω×[0, z]]).

The statement of the proposition now follows.

Proposition 1.18. Let z ∈ R2+, A0 ∈ P, and 0 > 0. Let B0 be a measurable subset of A0. Then there is an element K0 of Kδ such that K0 ⊂ A0∩Ω×[0, z] and

P(π[B0∩Ω×[0, z]])−P(π[B0∩K0∩Ω×[0,z]])≤0.

Proof. Let A be the collection of sets A ∈ P such that for every >0, 0≤λ <1, andB ⊂Athere is an elementK ofKδsuch thatK ⊂A∩Ω×[0, z]

and

P(π[B∩Ω×[0, z]])−P((B∩K)Bλ,z)≤.

LetS denote the ring generated by the sets(S, T]×(s, t]andD× {0} × (u, v] such that S, T are stopping times for (Gs), D ∈ G0, and s, t, u, v ∈R+. Note, we have P = σ(S). It is our goal to show that A is a monotone class containing S. It will follow from the Monotone Class theorem that A = P. Let A∈ S. We may write

A=

n0

[

i=1

(Si, Ti]×(ri, si]∪

m0

[

j=1

[0Aj, Uj]×(vj, wj],

(14)

whereSi,Ti, andUj are stopping times for(Gs),Aj ∈ G0, and ri, si, vj, wj are real numbers for every index iand every index j. Without loss in generality, we will consider A =

n

S

i=1

(Si, Ti]×(ri, si], and we will assume that the sets Ai:= (Si,i]×(ri, si]are pairwise disjoint.

Let > 0, 0 ≤λ <1, and let B be a measurable subset of A. Let (n) be a sequence of positive numbers decreasing to 0. For each L ∈N and each index i denote by KL,i the set [Si +L, Ti∧L]×[ri+L, si]∈ Kδ. For each index iwe have KL,i % Ai asL → ∞. By Proposition 1.17, for each index i there is a number Li ∈N such that

P(π[B∩Ai∩Ω×[0, z]])−P((B∩Ai∩KLi,i)(B∩Ai)λ,z)≤ 2i. SetK:=

n

S

i=1

KLi,i. We haveK∩Ω×[0, z]∈ Kδ,K∩Ω×[0, z]⊂A∩Ω×[0, z], and

P(π[B∩[Ω×[0, z] ])−P((B∩K∩Ω×[0, z])Bλ,z)

=P(π[B∩Ω×[0, z]]\(B∩K)Bλ,z) since

(B∩K∩Ω×[0, z])Bλ,z = (B∩K)Bλ,z⊂B∩Ω×[0, z];

=P π

h B∩

n

[

i=1

Ai∩Ω×[0, z]

i

\ B∩

n

[

i=1

KLi,i

B∩

n

[

i=1

Ai

λ,z

!

=P

n

[

i=1

π[B∩Ai∩Ω×[0, z]]\

n

[

i=1

(B∩KLi,i)(B∩Ai)λ,z

!

since the sets Ai are disjoint;

≤P

n

[

i=1

(π[B∩Ai∩Ω×[0, z]]\(B∩KLi,i)(B∩Ai)λ,z)

!

n

X

i=1

P π[B∩Ai∩Ω×[0, z]]\(B∩KLi,i)(B∩Ai)λ,z

n

X

i=1

2i ≤. Thus, A∈ A. SinceA∈ S was arbitrary, we haveS ⊂ A.

Next, let (A0n) be an increasing sequence from A. Set A0 :=

S

n=1

A0n. We will show that A0 ∈ A. Let B0 be a measurable subset of A0, and let > 0 and 0 ≤ λ < 1. There are numbers λ1, λ2 such that 0 ≤ λ1, λ2 <

1 and λ1λ2 = λ. By Proposition 1.17 there is an index N ∈ N such that P(π[B0∩Ω×[0, z]])−P((B0∩A0N)Bλ,z0 )≤ 2. SinceA0N ∈ A, for the measurable subset C := [((B0∩A0N)Bλ0

1,z)×R2+]∩(B0∩A0N)∩Ω×[0, z]of A0N there is a set K ∈ Kδ such thatK⊂A0N ∩Ω×[0, z]⊂A0∩Ω×[0, z]andP(π[C∩Ω×

(15)

[0, z]])−P((C∩K)Cλ2,z)≤ 2. Butπ[C∩Ω×[0, z]] = (B0∩A0N)Bλ01,z. Hence P((B0 ∩A0N)Bλ0

1,z)−P((C∩K)Cλ2,z) ≤ 2. Note, we have (B0 ∩K)Bλ,z0 ⊃ (C∩K)Cλ2,z. Hence

P(π[B0∩Ω×[0, z] ])−P((B0∩K)B0λ,z)≤P(π[B0∩Ω×[0, z]])

−P((C∩K)Cλ2,z) =P(π[B0∩Ω×[0, z]])−P((B0∩A0N)Bλ01,z)+

+P((B0∩A0N)B0λ1,z)−P((C∩K)Cλ2,z)≤ 2 +

2 =. We conclude that A0 ∈ A.

Lastly, let(A00n) be a decreasing sequence fromA. SetA00:=

T

n=1

A00n. We will show that A00 ∈ A. Let B00 be a measurable subset of A00, and let > 0 and 0≤λ <1. Let λ0∈R+satisfy λ < λ0 <1. There is a sequence(λn) from [0,1) such that Q

n=1

λn = λ0. Note, the setB00 is a measurable subset of A001. So, there is a set K1 ∈ Kδ such thatK1 ⊂A001 ∩Ω×[0, z]and

P(π[B00∩Ω×[0, z]])−P((B00∩K1)Bλ001,z)≤ 2.

The setC1:= [((B00∩K1)Bλ001,z)×R2+]∩B00∩K1∩Ω×[0, z]is a measurable subset ofA002. So, there is a setK2∈ Kδsuch thatK2⊂A002∩Ω×[0, z]andP(π[C1∩Ω×

[0, z]])−P((C1∩K2)C2,z)≤ 22. Note thatπ[C1∩Ω×[0, z]] = (B00∩K1)B0λ,z. The set C2:= [((C1∩K2)C2,z)×R2+]∩C1∩K2∩Ω×[0, z]is a measurable subset of A003. So, there is a set K3 ∈ Kδ such that K3 ⊂ A003 ∩Ω×[0, z] and P(π[C2∩Ω×[0, z]])−P((C2∩K3)C3,z)≤ 23. Note, we haveπ[C2∩Ω×[0, z]] = (C1∩K2)C2,z. Continuing inductively, we obtain a sequence (Kn) from Kδ such that for each index nwe haveKn+1 ⊂A00n+1∩Ω×[0, z]and

P(π[Cn∩Ω×[0, z]])−P((Cn∩Kn+1)Cn+1,z)≤ 2n+1, where

Cn= [((Cn−1∩Kn)Cn−1λn,z)×R2+]∩Cn−1∩Kn∩Ω×[0, z]

and

π[Cn∩Ω×[0, z] ] = (Cn−1∩Kn)Cn−1λn,z, n= 2,3,4. . . . So, for each index nwe have

P(π[B00∩Ω×0, z]])−P((Cn∩Kn+1)Cn+1,z) =

=P(π[B00∩Ω×[0, z]])−P((B00∩K1)Bλ001,z)+

+P(π[C1∩Ω×[0, z]])−P((C1∩K2)C2,z) +· · ·+ +P(π[Cn∩Ω×[0, z]])−P((Cn∩Kn+1)Cn+1,z)≤

2+

22 +· · ·+ 2n+1 ≤.

Références

Documents relatifs

The number of words to be eliminated being equal to 3 i−1 is, as in previous section, very lesser that the total number of words and we are so ensured to find between to

Computing the Cantor rationals of given denominator In this appendix, we give an algorithm to compute the set of rational numbers in the Cantor set of given denominator q, namely

[r]

Other interesting examples are mean value operators with which we can recover the classical Liouville theorem (cf. Example 5.4), purely atomic measures with few points in the

The SCOTCH meshOrder routine computes a block ordering of the unknowns of the symmetric sparse matrix the adjacency structure of which is represented by the elements that connect

In this section, we will give some properties of these Green functions that we will need later. The proof of the following lemma is based on elementary potential theory on the

(For the definition of the diophantine notions used from now on, see Section 1.3.) One of our results will be that the sequence (G 1,N (α)) N≥1 converges for all irrational numbers

The idea of the proof of Theorem 2 is to construct, from most of the restricted partitions of n into parts in A , many unrestricted partitions of n.. In Section 4, we will prove