On the independence of integer and fractional parts
Citation for published version (APA):
Steutel, F. W., & Thiemann, J. G. F. (1987). On the independence of integer and fractional parts. (Memorandum COSOR; Vol. 8729). Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/1987
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
EINDHOVEN UNIVERSITY OF TECHNOLOGY
Department of Mathematics and Computing Science
Memorandum COSOR 87-29
On the independence of integer and fractional parts
by
F.W. Steutel and J.G.F. Thiemann
Eindhoven, Netherlands October 1987
On the independence of integer and fractional parts
by F.W. Steutel and J.G.F. Thiemann Eindhoven University of Technology,
Department of Mathematics and Computing Science, P.O. Box 513,
5600 MB Eindhoven, The Netherlands.
ABSTRACT
We use the independence of the integer and fractional parts of exponentially distributed random variables to obtain expressions for the order statistics from a geometric distribution. As our main result we show that a strong form of this independence characterizes the exponential distribution.
1.
Introduction
It is easily verified that for an exponentially distributed random variable the integer part and the fractional part are independent. We formulate this property as a lemma.
Lemma 1. Let Y have an exponential distribution, let [Yl denote the largest integer not
exced-ing Y (the integer part) and let {Y} := Y - [Y] (the fractional part). Then
[YJ and {Y} are independent .
It is also easily verified that there is the following (well-known?) relation between exponential
and geometric random variables; this relation was pointed out to me long ago by L. de Haan
(1978) in the context of problems 56 and 57 in Statistica Neerlandica, where nobody had used it (see Jagers (1978)).
2
-Lemma 2. If X has a geometric distribution, Le., if
P(X =k)=p(1-pi' (k =0,1,'" ;O<p
< 1) ,
thend
X = [YJ ,
where Y has an exponential distribution with density
fy(y)
=
'A.e-'A.y
(y > 0; 'A.=
-log(1 -p» .
In section 2 of this note we return to order statistics of geometric random variables, and in
sec-tion 3 we prove that a random variable Y has an exponential distribution if and only if [cYJ
and {cY} are independent for every c
>
O.2. Order statistic from the geometric distribution
It is well known and not hard to prove that the order statistics Y l;n S; Y 2;n S; • . . S; Yn;n from an exponential distribution satisfy (cf. Feller (1971), p. 19)
d Ie-I Y . n-}
YIe;n
=
L --.
(k=
1,2, ... ,n) ,j=O n-j
(1)
where Y l' . . . , Yn are iid exponential random variables. This is most easily seen as follows:
waiting for YIe;n to finish we first wait for Y l;n' after which the remaining Y's start anew by
the lack-of-memory property, so
d Ie-I
YIe;n
=
L
Y l;n-j ,j=O
d
with the Y l;n-j independent and (cf. (1» Y l;n-j
=
Yn- j I (n - j) , Since the geometric distribution has a similar lack-of-memory property:(2)
P (X ~ k
+
m I X ~ m)=
P (X ~ k). it is tempting to conjecture that a relation similar to (2) would hold for geometric order statistics. It is easily seen. however, that this is not so. Instead of (2) we have the following result.Theorem 1. Let X";lI (k
=
1, 2, ...• n) denote the order statistics from a geometric distribu-tion on {O,l.2 •... }. Then3 -d k-l k-1 Y . n-J Xk;n
=
I:X1;n-j+
[I: {--. }] ,
j=O j=O n-} (3) dwhere Xj and Yj are related by Xj
=
[Yj ] as in Lemma 2, and all variables in the right-handside of (3) are independent.
d
Proof. Using Lemmas 1 and 2, and the fact that X l;m
=
[Y 11m], one easily hasd d d k-l y . k-l y . k-l Y .
Xk;n
=
[Y]k;n = [Yk;n]=
[I:
n-~]=
[I: [
n-~]+
I: {
n-~}] =j=O n-) j=O n-) j=O n-}
k-l y . k-l y . d k-1 k-l Y .
=
I: [
n-~]+
[I: {
n-J.}]=
I:
X l;n-j+
[I:
{~}].
j=O n-) j=O n-) j=O j=O n-j
where all random variables in the last expression are independent.
Corollary. For the probability generating function of Xk;n one has
k-l 1-(1-p )n-j
PXk;tt(z)
=
IT
1-(1 )n-j • Qn,k(z) ,j=O -p Z
where Qn,k is a probability generating function and a polynomial of degree k - 1.
o
(4)
Remark 1. For the order statistics of exponentially distributed random variables mentioned
ear-lier we have for the Laplace transfonn
_ k-l 'J..(n-j)
C\>Yk;n (s) -
IT
'J..( '}i;j=O n-j S
here the extra factor occurring in (4) is missing.
Remark 2. A special case of this Corollary occurs in Jagers (1979). The explicit fonn of Qn,k
seems intractable.
Remark 3. From Theorem 1 it follows that
k-l k-l
I:
X 1;n-j S Xk;n SI:
X1;n-j+
n - 1 •j=O j=O
(5)
where the inequalities should be read in distribution (i.e. in the sense of stochastic ordering);
the inequality on the right in (5) holds, in fact, with probability one on the probability space
4 -3. A characterization of the exponential distribution
Let Y be a random variable with distribution function F. For [Y] and {Y} (cf. Lemma 1) to be
independent it is necessary and sufficient that (F(n
+
y) - F«n»/(F(n+
1) - F(n» isindependent of n E N for y E (0,1]. There are many F's with this property, but if F is
required to have a continuous density
f
then it follows thatf
must satisfyfor some (X, i.e. that
f
is "globally exponential",The object of this section is to prove the following theorem,
Theorem 2. If a random variable Y is non-constant and such that [eY] and {eY} are
indepen-dent for every e > 0, then Y has an exponential distribution on (0,00) or on (--00,0).
We first prove three lemmas. The first lemma is a direct consequence of the independence of
[eYJ and leY}.
Lemma 3. Let Y satisfy the conditions of Theorem 2 and let k E
.z,
a,b E [0,1], e > 0 besuch that
Peke S;; Y < (k + l)e) > 0 and peek +a)c < Y < (k +b)c)
=
0 .Then, for each 1 E Z,
P((l +a)e
<
Y<
(l +b)e)=
O.Proof: P([Y/e] = k) = P(ke S;; Y < (k + l)c)
>
0, soP«l+a)c <Y
<
(l+b)c)=P([Ylc]=1)P(a<
{Ylc} <b)== P([Y/e] = l)P([Ylc] = krl peek +a)c
<
Y<
(k +b)e)=
O.Lemma 4. Let Y satisfy the conditions of Theorem 2. Then
i) P (Y < 0)
=
1 or P (Y ~ 0)=
1 ,ii) P ( I Y I < 'Y)
>
0 and P ( I Y I> 'Y)
>
0 for each 'Y > 0 .Proof.
i) Suppose P(Y < 0) < 1 and P(Y ~ 0) < 1. Then P(Y ~ 0) > 0 and P(Y < 0) > 0 and therefore, for a large enough. we have
P(OS Y
<
a) >t
P(OS Y) and P(-«S Y < 0) >t
P(Y<
0) From this it follows thatp({L} <.11 [L]=-I)= P(-2aS Y <-a) = 2a 2 2a P (-2a S Y < 0)
=
1 _ P (-a S Y < 0) < 1 _ 1I2P (Y < 0) =.1 ;P(-2a S Y < 0) P(Y < 0) 2
P ({ L } <.1 1 [L]
=
0)=
P (0 S Y < a) > 112 P (0 S Y) =.12a 2 2a P(O S Y < 2a) P(O S Y) 2
The inequality of these conditional probabilities contradicts the independence of
[:a.]
and{ :a. },
and so (i) holds.ii) We consider the case P(Y ~ 0) = 1; the case P(Y < 0) = 1 can be handled in a similar way. Again, we argue by contradiction.
First suppose that P (Y < y)
=
0 for some y > O. Then3 := sup {y > 0 I P (0 S Y < y)
=
O} < ex> and P (0 S Y < 3)=
0 .Also, for each a > 3. we have P (0 S Y < a) > 0 and hence. by Lemma 3,
P(a < Y < a+3)
=
O.So, for all a> 3 we have P(a < Y < a
+
3) = 0, and therefore P (Y > 3) =o.
As we also have P(O S Y < 3)=
0, we conclude that P(Y=
3)=
1, i.e. Y is constant. This contradiction proves that P (Y < y) > 0 for every y > O.Next suppose that P(Y > y) = 0 for some y> O. Then 3:= inf{y> 0 I P(Y >
1>
= O} > 0, since 3 = 0 would imply P(Y = 0) = I, i.e. Y is constant. Moreover P(Y > 3) =o.
Now let
a
E(~
3, 3). Then, by the definition of 3,P(3 < Y < 2a)
=
0 and peaS Y < 2a) > 0 . So, by Lemma 3,P(3-a < Y < a)
=
0 . (7)Now suppose that P (0 S Y S 3 - a) > O. Then, since 3 - a < ; 3 < a, we have
P(O S Y < ; 3) > 0 and, by (7), P(3-a < Y < ; 3) = 0, hence, by Lemma 3,
5 4 5
- 6 ~ definition of O. So (P (0 S Y S 0 - a)
=
O.All this implies P (0 S Y
<
a) = 0 for each a E(~
0, 0), so P (0 S Y<
0)=
O.Since also P(O < Y)
=
O. we have P(Y=
0)=
1, i.e. Y is constant So, again we obtain acon-tradiction, and hence P (Y
>
y)>
0 for every y>
O.0
Lemma
5.
Let Y satisfy the conditions of Theorem 2. then [Y] has a geometric distributioneither on {O, 1,2 ... } or on {-1,-2,-3, ... }.
Proof. We first consider the case P(Y ~ 0) = 1 (cf. Lemma 4).
Let
Pi := P([Y] = i) (i = 0,1,2, ... )
Then, for each i • j E N with i < j, we have
PiPj =P(i S Y < i+l)PU S Y <j+l)=
Y i Y i+l Y Y 1
= P ([ -;- ] = 0) P (-: S {-;-} < - . ) P ([ -;- ] = 1) P (0 S {-;-} < -;- ) =
} } } } } J J
=P(OS Y < l)PCi+jS Y <i+j+l)=pcj1j+j'
Now Po> 0 by Lemma 4 ii), So, taking i = 1. we get
hence
E1. .
Pj = Pj+l U = 2,3, ... ) ;Po
Also by Lemma 4 ii) we have P (Y ~ 3) > O.
00 00 P
So 0
<
L
PI=
L
(_1),= P3. which implies P3*'
O.1=3 k=O Po
The foregoing now gives, for 1
=
2,3,4, ... ,PI )1-2 PI 1-2 PI 1
P3Pl
=
P3(- P2= (-)
PsPo= (-)
P3PoPo Po Po
so PI =
(E1./
Po. As the last equality holds also for I = 0,1, the random variable [Y] isPo
7
-For the case P (Y < 0)
=
1 we defineqj :=P([y] =-I-i) (i =0,1,2,"')
and prove the relations qjqj
=
qi+jqO (0 < i < j) by considering the random variable Y Ij.The rest of the proof is similar to the proof given above. []
Proof of Theorem 2
We consider the case P (Y ;?! 0) = 1; the case P (Y < 0) can be treated in the same way.
Let n E IV. Then the random variable nY satisfies the conditions of Theorem 2 and therefore
[ny] is geometrically distributed on {O, 1,2 •... }, by Lemma 5. So p E (0, 1) exists such that, for each m e {O, 1, 2, . . . }, we have
pm = P([nY];?! m) = P(nY;?! m) = P(Y;?!.!!!.) , n
and, in particular, pI!
=
P (Y;?! 1); hencem
P (Y ;?! .!!!.)
=
P (Y ;?! 1)-; .n
So P (Y ;?! r)
=
P (Y;?! 1)' for each rational r E [0, (0), Since both members of the equalityare non-increasing functions of
r
on [0,(0), the equality holds for eachr
e [0,(0). Hence Y isexponentially distributed on [0,(0). []
References
Feller, W. (1971), The theory of probability and its applications, Vol. 2, 2-nd ed., Wiley, New York.
de Haan, L. (1978), Personal communication.
Jagers, A.A. (1978), Solution of Problem 57 in Statistica Neerlandica, Vol. 33, Dr. 3.
Eindhoven University of Technology,
Department of Mathematics and Computing Science, P.O. Box 513,
5600 MB Eindhoven, The Netherlands.