209 The Sciellce R巴portsof the Kalla古awaUlliversity, Vol. III, No, 2, pp. 209‑212, ]uly, 1955
On the Series of Some Iudependent Random Variables. By
Shigem T AKAHASHI
(Rec巴iv巴d21 January, 1955)
1. Let { X,.! } be a s色quenceof independent random variables with a common distribution function F(吋 and{ιJ be a decreasing sequence of positive numbers. Further let us pm
and
Po =Prob.L510nX321く 十 ∞ ] P1 = Prob.
' f } , [
~ a" Xn convergesJpzニ Prob.
う [
a"( XlI ‑J X
dF(x)) conv何e S J
laJlXI三1
It is seen that each of
(1.1) PF04fob[fE
吋
IXil豆1J=land
(1.2) P'l = Prob. [
恒
101221Xzlg/も]=1 implies(1.3) Prob. [
便
Ian Xn I壬1J=1and therefore, by the Borel‑Cantelli lemma, (1.1) and also (1.2) imply (1.4)
子
Prob.[ 1
Xlll> ~去〕ニ?fdF(的く十∞
Ixl>~ an
In this note we prove that under 80me conditions on ‑ (an ,}(1.4) implies Pi = 1 (i=O, 1, 2). Especially, by Knopp's theorem in the theorγof series, Po = 1 (01' P1 = 1) imp1ies P'o = 1 (or P'l = 1).
2. THEOREM 1. If for some A
>
1 (2.1) 山一→さくall
then (1.4) implies Po = 1.
(
竺
幻 十 1 /y
n 1, 2,PROOF. By the three series theorem, it is sufficient to show that 一 (
I
1 an x I dF (X))2J
く十回k Vη =
ヱ
II
(an X)2 dF(x)n 11¥ aノ
la"x・1~三 1 九nd
S. T AKAHASI‑II
ヱ
lilnく十∞.210
3M
戸? f l G n z i
lanxl三I
On the other hand, it may be seen that
0豆 Vn豆
f l a
nxI2dF(x)豆 恥lah丸│壬1
and hence, it is sufficient to prove that
く」∞.
Xi dF(x) an 2n :
1nτ1
dF (x) =
ヱ
三
a"xlI anX l~三 1
We have
n G
削
∞ ヱ =
¥l
ノ 一
n
Z 1 一 %
/t
¥ 一tF d
く
f 1
α一 一 戸
dF〔Z〉
=zJ‑
m=1 u m∞ n
主玉玄 an 玄 J二
1 1 = τ 1 η包=)l‑f/m
where
一 一 三1 O. ao From (2.1), we have
戸
d m ( C ) A
for n 二三 mJf
豆h n z A F g t ( ; j A
豆and we have
A amm wh古reA is a constant independent of m.
dF(x)
: : s
m Im~l
百三っくlxl三
M刀1‑1 u m
Therefore, it follows that
ヱ
Mn 三玉 An=l
ニ A
三 l C JdF(約 十 J
dF(り く 十 ∞
lxl>
誌
ず二王<:lxl三 。 レ
Let F(x) be a symmetric distribution function and for some A >弘 く {~-V
¥ n十1/
THEOREM 2.
n 1, 2,
αn+1 an Then (1.4) implies Pl
=
LPROOF. It is sufficient to show社lat (2.2)
子 J
(an X)2 dF(x)く 十 ∞ lanxl,
三1and this can be proved by. the same way as the proof of L M nく+∞.
n
n 1, 2,
,1>唱
。
n+1 ̲̲̲̲ ( n ¥,1 an = ¥ n+1 ) If for someTHEOREM 3. (2.3)
。
nthe Series of Some IndelうE幻dentRandOJn Variables. 211 then (1.4') implics 1"'2 = LPI¥'OOF. Let us define random variables Y" ancl Zηas follows. if iXη1> 1
y
, , = 。 J l
X'I1, if 1 [X"l三 一
a"
anc1
if IXn
!>
1 ーZ
, , =
an0, if IXnl4 1 n = 1,2.. a}l
From (1.4) and the above definitions, it follows that (2.4)
1Dr
此 ( ミ
ia}/ Zn 1く→∞〕= I,and
DnzE[(GJM〉2]‑[E(GJM〉了
= j .
〔GFlddF〔同一(I
a!1 x dF(x))2Ml1 = E(an Y" ) =
I
a" x dF(x)lanxl でさ1
As the proof of Theorem 2, we can show that
O~ ヱD"三玉ヱ (an x)】 dF(x)く+∞.
多 n,;
lan忽 1壬,1
On the othεr hand, the random variable (an Yn ‑ 111[n ) has the mean value zero and the dispersion Dn. Hence, by Khintchine‑Kolmogoroff's theorern, we have
(2.5) Prob.
r
ヱ(酌:VYI‑1!;f" ) conv仰 ts〕
=1.From (2.4) and (2.5), we get the required result. 3 . 10 • Suppose that
( 日
P f o b ‑ B i ‑ f L │
く+∞J =
1Then, by the Borel‑Cante1li lemrna, it follows that for any constant c > 0,
手
Prob.[: XIl I >ー]く十∞.Therefore, we have
. f
1 x I dF(x)く + ∞
紅lC1the strong law oI large numb巴r8shows that
Pfob[t;1(Xjc│→
( r
x[ dF(的J =
1 On the othεr hand, fγom (3.1), we have,‑' 1 n 、
Prob. 1
云
F1│Xhl→o J
= 1.Hence, we have
212 S. T AKAHASHl
̲l, Dつ
f!xi dF(x)ニ O and this is equivalent to
(3.2)
P r o b . [Xη=oJ =
1 n 1, 2,Therefore, A in Theorem 1 can not be rε,placed by 1既 cepta trivial回 目 (3 . 20 • Let F(x) be symmetric and
(3.3)
P r o b .
1Xn
仰 1仰 初I=Ln V n Then we have, by the same way as 10,
斗じっ
x2 dF(めく+∞
and from the central 1imit theorem and (3. 3), we have
f
がdF(x)=
0This shows that .3) implies (3.2). Therefore多 Ain Theorern 2 can not be replaced by ]7白色xcepta trivial case (3
30 • Lεt
prob
i‑bL(Xη‑ I
co附 吋ω J=
1、 n Vn
!ぇ!三
vn
alld
X
n denote the symmetriz吋 randomvariable of XI' "Then we have
ω = 1
Therefore, by the discussion in 20, we have
Pyob.[Xn=OJ
ニ l This shows that(3.4)
Pr
め[X
,.,=
111J =
1where m is a constant independent of n.
n=l, 2,
n = 1,2,
Hence A in Theorem 3 can not bを replacedby弘 excepta trivial case (3.4).