• 検索結果がありません。

On the convergence of certain sums of independent random elements

N/A
N/A
Protected

Academic year: 2022

シェア "On the convergence of certain sums of independent random elements"

Copied!
5
0
0

読み込み中.... (全文を見る)

全文

(1)

On the convergence of certain sums of independent random elements

J.C. Ferrando

Abstract. In this note we investigate the relationship between the convergence of the sequence {Sn} of sums of independent random elements of the formSn=

Pn i=1εixi

(whereεitakes the values±1 with the same probability andxibelongs to a real Banach spaceX for each i ∈ N) and the existence of certain weakly unconditionally Cauchy subseries ofP∞n=1xn.

Keywords: independent random elements, copy ofc0, Pettis integrable function, perfect measure space

Classification: 46B15, 46B09

1. Preliminaries

Our notation is standard ([1], [3], [4], [9]). Throughout this note ∆ will denote the Cantor space {−1,1}N, Σ the σ-algebra of subsets of ∆ generated by the n-cylinders of ∆ for eachn∈N, andν the Borel probability⊗∞i=1νi on Σ, where νi : 2{−1,1} → [0,1] is defined by νi(∅) = 0, νi({−1}) = νi({1}) = 1/2 and νi({−1,1}) = 1 for eachi ∈ N. In what follows X will be a real Banach space andL0(ν, X) will stand for the (F)-space overRof all [classes of] ν-measurable X-valued functions equipped with the (F)-norm

kfk0= Z

∆

kf(ε)k

1 +kf(ε)kdν(ε)

of the convergence in probability. We shall represent by P1(ν, X) the (real) normed space consisting of all those [classes of] ν-measurable X-valued Pettis integrable functionsf defined on ∆ provided with the semivariation norm

kfkP1(ν,X)= sup Z

∆

|x∗f(ω)| dν(ω) :x∗∈X∗,kx∗k ≤1

.

As it is well known, P1(ν, X) is not a Banach space whenever X is infinite- dimensional. In the sequel we shall shorten by wuC the sentence ‘weakly uncon- ditionally Cauchy’.

Supported by DGESIC PB97-0342 and Presidencia de la Generalitat Valenciana.

(2)

In [5] we have shown that if a series of independent random elements of the formP∞

n=1fn, withfn(ω) =ωnxnforω∈∆ and{xn} ⊆X, convergesν-almost surely inX, thenP∞

n=1xnhas a subseries which is unconditionally convergent in norm. In this note we continue the investigation on the relationship among the convergence of the functional series P∞

n=1fn under different topologies and the existence of certain wuC subseries ofP∞

n=1xn.

2. On certain weakly unconditionally Cauchy subseries

Lemma 2.1. If there are a closed setA in∆ withν(A)>1/2and a nonempty setS ⊆X∗ such that P∞

i=1x∗fi(ω)converges forω∈Aand x∗ ∈S, then there exists a subsequence{xni}of{xn}such thatP∞

i=1|x∗xni|<∞for eachx∗ ∈S.

Proof: The following fact is contained in the proof of [8, Proposition] (see also [5, Claim]). We shall denote byCi1i2...ik or Ci1i2...ik(ε) any rectangle of ∆ with fixed coordinatesi1, i2,. . .,ik, i.e.,Ci1i2...ik(ε) ={ω∈∆ :ωij =εj,1≤j≤k}

for someε∈∆. On the other hand, given a strictly increasing sequenceQ={ni: i∈ N} of positive integers, for each ω ∈∆ we shall design byω′ (as in [8]) the element of ∆ defined byωi′=ωi ifi∈Qandω′i=−ωi ifi /∈Q.

Fact. Let A∈Σ. Ifν(A)>1/2, there is a strictly increasing sequence{ni} of positive integers such thatA∩A′∩Cn1n2...nk 6=∅ for each Cn1n2...nk and each k∈N.

By hypothesis there is a closed setAin ∆ withν(A)>1/2 such thatP∞

n=1ωnx∗xn converges for ω∈ Aand x∗ ∈S. According to the preceding fact there exists a strictly increasing sequenceQ={ni}of positive integers such that, givenε∈∆, then A∩A′∩Cn1n2...nk(ε)6=∅ for each k ∈N. Since{A∩A′∩Cn1n2...nk(ε) : k ∈ N} is a decreasing sequence of nonempty closed sets in the compact space

∆, there is a pointζ (which depends ofε) in ∆ which belongs to the intersection T∞

k=1A∩A′∩Cn1n2...nk(ε). Hence, for eachx∗∈Sand each pair (r, s) of positive integers, withs > r, one has

s

X

i=r+1

εix∗xni

=

s

X

i=r+1

ζnix∗xni

≤ 1 2

ns

X

i=nr+1

x∗fi(ζ)

+

ns

X

i=nr+1

x∗fi ζ′

! .

Since ζ, ζ′ ∈ A and x∗ ∈ S, both series P∞

i=1x∗fi(ζ) and P∞

i=1x∗fi(ζ′) are convergent. So, for a givenǫ >0 there is ak∈Nsuch that

Pns

i=nr+1x∗fi(ζ) < ǫ and

Pns

i=nr+1x∗fi(ζ′)

< ǫfors > r≥k, which implies that Ps

i=r+1εix∗xni

≤ǫ for s > r ≥ k. Hence the numerical seriesP∞

i=1εix∗xni converges. Given that this is true for eachε∈∆, it follows thatP∞

i=1|x∗xni|<∞for eachx∗∈S

and we are done.

(3)

Theorem 2.2. Assume thatkxnk= 1for eachn∈NandX has a dual unit ball with countably many extreme points. If

sup

n∈N

Z

∆

|x∗Sn(ω)|dν(ω)<∞ for eachx∗∈ExtBX∗, thenX contains a copy ofc0.

Proof: By hypothesis, for eachx∗ ∈ExtBX∗ there existsCx∗>0 such that

(2.1) sup

n∈N

Z

∆

n

X

i=1

x∗fi(ω)

dν(ω)< Cx∗.

Hence, givenx∗∈ExtBX∗, as a consequence of (2.1) and of Khinchine’s inequal- ities there exists aK >0 such that

(2.2)

( n X

i=1

σ2(x∗fi) )1/2

= ( n

X

i=1

(x∗xi)2 )1/2

≤K Z

∆

n

X

i=1

x∗fi(ω)

dν(ω)< KCx∗ for each n ∈ N. Considering that the sequence {x∗fi} consists of independent random variables such that

E(x∗fi) = Z

∆

x∗fi(ω)dν(ω) = 0

for eachi∈N, according to [7, Section 46, Theorem B] equation (2.2) ensures that P∞

i=1x∗fi(ω) converges almost surely forω∈∆. Since ExtBX∗ is countable, it follows that there exists aν-null setN such thatP∞

i=1x∗fi(ω) converges for each ω∈∆−N and eachx∗ ∈ExtBX∗. So, using inner regularity we may choose a closed setAwithA⊆∆−N andν(A)>1/2 such thatP∞

i=1x∗fi(ω) converges for eachω ∈Aand eachx∗∈ExtBX∗. On the basis of Lemma 2.1, this implies that there exists a subsequence{xni} of {xn} such that P∞

i=1|x∗xni| <∞ for eachx∗∈ExtBX∗. SinceP∞

n=1xndiverges, Elton’s theorem guarantees thatX

contains a copy ofc0.

Proposition 2.3. If the sums {Sn} are bounded inside of a complete linear subspaceLof P1(ν, X), thenP∞

n=1xnhas a wuC subseries.

Proof: Since{Sn}is bounded inside of a complete linear subspaceLofP1(ν, X) and given that the canonical inclusion map fromP1(ν, X) intoL0(ν, X) has closed graph ([6, Lemma 4]), then Banach-Schauder’s theorem guarantees that{Sn} is stochastically bounded. So, according to [9, Section 5.2.3, Theorem 2.2] the sums {Sn} are bounded almost surely, i.e. ν({ω ∈ ∆ : supn∈N

Pn

i=1fi(ω)

= ∞})

= 0. Hence Kwapie´n’s theorem [8, Proposition] assures the existence of a wuC subseries ofP∞

n=1xn.

(4)

Corollary 2.4. Assume that{fn} is a basic sequence inP1\(ν, X)equivalent to the unit vector basis of c0. If [fn] is contained inP1(ν, X), then there exists a subsequence{fni} such that[fni] is isomorphic to a complemented copy of c0. Proof: Since the seriesP∞

i=1fi is wuC inP1(ν, X), there isK >0 such that sup

n∈N

n

X

i=1

ξifi P

1(ν,X)

< Kkξk∞

for each ξ ∈ ℓ∞. Hence the sums {Sn} are bounded in the complete linear subspace [fi] of P1(ν, X) and Proposition 2.3 guarantees that P∞

n=1xn has a wuC subseries. Sincekxnk =kfnkP1(ν,X) for eachn∈N, then infn∈Nkxnk>0 and the classic Bessaga-Pe lczy´nski allows us to conclude that {xn} contains a subsequence {xni} equivalent to the unit vector basis of c0. Therefore, there exists a bounded sequence{yi∗} in X∗ such that y∗ixnj =δij for each i, j ∈ N. Assuming without loss of generality thatyi∈BX∗, setgi(ε) =εiy∗i for eachi∈N and define

hgi, fi= Z

∆

εiy∗if(ε)dν(ε)

for eachf ∈P1(ν, X). So we havehgi, fnji=δij for eachi, j∈N. On the other hand, denoting byCn the rectangle of ∆ formed by all thoseε∈∆ withεn= 1 and noting thatν(E∩Cn)→ν(E)/2 for allE∈Σ, it follows that

ECn(ϕ) = 1 ν(Cn)

Z

Cn

ϕ dν→ Z

∆

ϕ dν=E(ϕ)

for each ν-simple function ϕ : ∆ → R. This implies that ECn(ϕ) → E(ϕ) for each ϕ ∈ L1(ν), which leads to R

∆εiϕ(ε)dν → 0 for each ϕ ∈ L1(ν). Since, in addition, (∆,Σ, ν) is a perfect measure space, it can be shown as in [2] that hgi, fi →0 for eachf ∈P1(ν, X). Consequently the mapP :P1(ν, X)→P1(ν, X) defined by

P f =

∞

X

i=1

hgi, fifni

is a bounded linear projection operator from the barreled space P1(ν, X)

onto [fni].

Proposition 2.5. If there exists a complete linear subspaceLin P1(ν, X)such that{fi} ⊆LandP∞

i=1ficonverges inP1(ν, X)to some separably-valuedf ∈L, then there exists a subseries of P∞

i=1xiwhich is unconditionally convergent inX.

(5)

Proof: Given thatP∞

i=1fi=finP1(ν, X) andLis complete, thenP∞ i=1fi=f inL. Then, using the fact that the inclusion map fromP1(ν, X) intoL0(ν, X) has closed graph together with the Banach-Schauder theorem, we get thatP∞

i=1fi = f in probability. Since the range off is separable in norm, then [9, Section 5.2.3, Theorem 2.1] guarantees that the seriesP∞

i=1fi(ω) converges inX tof(ω) almost surely forω∈∆. Hence [5, Theorem 2.1] establishes the existence of a subseries ofP∞

n=1xn which is unconditionally convergent inX.

Question. We do not know whether the statement of Theorem2.2is true without the assumption thatBX∗ has countable many extreme points.

References

[1] Cembranos P., Mendoza J.,Banach Spaces of Vector-Valued Functions, LNM1676, Springer, 1997.

[2] D´ıaz S., Fern´andez A., Florencio M., Pa´ul P.J.,Complemented copies ofc0 in the space of Pettis integrable functions, Quaestiones Math.16(1993), 61–66.

[3] Diestel J.,Sequences and series in Banach spaces, GTM92, Springer-Verlag, New York- Berlin-Heidelberg-Tokyo, 1984.

[4] Diestel J., Uhl J.,Vector measures, Math Surveys15, Amer. Math. Soc., Providence, 1977.

[5] Ferrando J.C.,On a theorem of Kwapie´n, Quaestiones Math.24(2001), 51–54.

[6] Freniche F.J.,Embeddingc0in the space of Pettis integrable functions, Quaestiones Math.

21(1998), 261–267.

[7] Halmos P.R.,Measure Theory, GTM18, Springer, New York-Berlin-Heidelberg-Barcelona, 1950.

[8] Kwapie´n S.,On Banach spaces containingc0, Studia Math.52(1974), 187–188.

[9] Vakhania N.N., Tarieladze V.I., Chobanian S.A., Probability Distributions on Banach Spaces, D. Reidel Publishing Company, Dordrecht, 1987.

Centro de Investigaci´on Operativa, Universidad Miguel Hern´andez, E-03202 Elche (Alicante), Spain

E-mail: [email protected]

(Received June 27, 2001)

参照

関連したドキュメント