probability theory
Citation for published version (APA):
Vanderperre, E. J. (1975). An extension of a theorem of Cauchy with applications to probability theory. (EUT report. WSK, Dept. of Mathematics and Computing Science; Vol. 75-WSK-05). Eindhoven University of Technology.
Document status and date: Published: 01/01/1975
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.
ONDERAFDELING DER WISKUNDE DEPARTMENT OF MATHEMATICS
An extension of a theorem of Cauchy with applications to Probability Theory
by
E.J. Vanderperre
T.H.-Report 75-WSK-05 August 1975
We use the theorem to solve an integral equation of the Pollaczek type [7J and to study a Markov chain {w , n ~ O} with state space [O,oo[ closely
relat--n ed to Queueing Theory.
I. An extension of a theorem of Cauchy
1.1. Introduction
Let S be the complex w-plane, S± := {w I 1m w ~ O} and JR := {a I _00 < a < +oo}. A. Let few) be a function, holomorphic in S and continuous in S + JR
includ-ing the point at infinity i.e., few)
=
f(oo) + 0(1) as Iwl + 00 in S + JR. Then(a) I 27fi f ( a ) - = do
{H(OO)
a -w -few) if w E S + i f w E SThe symbol C indicates the principal value (Cauchy value) of the integral. B. Let few) be a function, holomorphic in S+ and continuous in S+ + JR in-cluding the point at infinity. Then
(b)
f(a)~
a-w - 0 0=
{feW) - H(oo) -!f(oo) if w E s+ ' if w E SThe formulae (a) and (b) are called "Cauchy formulae" for the regions Sand S+ respectively.
A simple proof of (a) and (b) is gLven Ln [6J.
Remark. The condition that few)
=
f(oo) + 0(1) as Iwl + 00 in S_ + JR (or in S+ + JR) is not necessarily satisfied when few) is a Fourier-Stieltjes trans-form (F.S.T.). Therefore, we shall prove the following extension related to Probability Theory.I .2. The theorem
Let {~,B,P} be a probability space with ~ := J-oo,+oo[ the sample space, B the smallest a-algebra containing all open subsets (Borel sets) of ~ and P a pro-bability measure on B. Introduce a B-measurable mapping~: ~ +JR (a random variable) with probability distribution function (p.d.f.)
F(x) := P[x < x], x E JR. and define +00 E{ e -ia.:.} :=
f
e-iaxdF(X) , a EJR.. -00Denote by (A) de indicator of the event A, i.e.
if A occurs, i f not • Theorem 1. 1 2rri _00 = {"
~E
(,:, =-!E(!
= 0) - E{e-iw!(!
~
O)} if w E S -iwx 0) + E{ e -(!:;;; O)} i f w E S +Proof. Since
(!
< 0) + (! 0) + (x > 0) = 1, we have VwI
w E S+ u S(1. 1) I 21fi _00 + E(,.:. +00
J
E{e-ia!}~
=j
a - w +00 1~
da = 0)'2'" rr~ a - w -00provided that all principal values
We shall now prove that +00 1 2rri I +
Z-
rr~ +00I
E{ -iaxO)}~
+'j
e-(!
> a - w +00f
He -iax -(x < O)}~ a-w _00 involved exist. -iwx W E S ( 1 .2) I 21fif
E{e-ia !(! > 0)"}--2.... d = {-ne
-(x> O)}, a - w 0 -00In order to show the statement, let w E S_, i.e., put w
=
Rei8, rr <e
< 21f,o <,R < 00.
Consider a closed curve
r
consisting of a segment [-T,T] of the real line andof a semi-circle y lying in S_, with center at the origin and with radius
ie'
T
I
0 < R < T < 00. Finally, let w'=
Te , rr <e'
< 2rr. The contour integra-tion is clockwise.From the above definition of
r,
we obtain for w E S +2;i
J
-iw'x dw' He -(_x> O ) } _ -w' -w y But ex> -iw 'x E{ e -(~ > O)} =f
0+Whence, by Cauchy's theorem, we obtain for w E S
Now +T -iwx -E{e -(~ > O)} =~ I 27T~
J
E{e-icr~(_x
>O)}~
(J - W 0012~i
J
f
y 0+ -T ex> + 27T if
J
y 0+ e-iw'XdF(X) dw'1<
T w' -w -2i
T 27T 00f
f
7T 0+ 00 7T/2 ~ T - RJ J
e-Txsin~d~dF(X)
0+o
The permutation of both integrals in the right-hand side of the first ine-quality is justified by Fubini's theorem.
It may be verified that for any 2 E JO,7T/2[
7T/2
f
e -Txsin'Pd ~ :::; sec Q, e -Txsincp co s cp 'P d + ( / 2 " ) 7T - IV e -TxsinQ, •o
oAfter Some algebra we obtain (Q, fixed),
12~i
f
f
y 0+ 00 e-iw'xdF(x) dw'I
S wi -w 00 0+00
T -I T
+ T _ R tan t{F(T cosec t+) - F(O+)} + ~ sec t
f
*'
Xa (x)dF(x)0+
where a := JT -I cosec t,oo[ and
r I X (x)
=
J
l
a 0 if x E a if x ¢. a • since I X (x) :0;Tx a I, we have by a well-known dominated convergence theorem
1
27fi
J J
y 0+ 00
Finally, we remark that
ifwES i f W E S+
In a similar way, we can prove that
as T -+ +00 •
I
27fi
t
E{e~cr~ (x' > O)}~=
+ oo . , d { E{ e
.
,
~W~ (x' > O)}, W E S+ ' -00 - cr-w 0Hence, for x' = -~, we have
(I. 3) By (1.1), 1 27fi - 0 0 (I .2) 27fi -00 +00 and +00
T
E{e-icr~(x....
{ -~wx dcr E{e -(~ < O)}, < O ) } - = cr-wo
(I .3) we obtain.
{-jE(X
=
0) - E{e -~wx -(~ > E{e-~cr~}~=
--iwx cr-w 0) +~E(~ = + E{e -(~ < W E S W E S 0) }. if W E O)}, i f W E S S+Corollary I. Let x be a nonnegative random variable (r.v.) with characteris-
-.
-~wx
tic function (c.f.), E{e -}, 1m W :0; O. Then
W E S W E S +
( I .4)
00
Corollary 2. Let <~n>n=O be ~ sequence
. {-1WX 00
1ng sequence of c.f.'s <E e -n}>n=O'
of nonnegative r.v.'s with correspond-1m W ~ O.
For W € Sand z € [0, I [ we have
+00
f
00 1L
nE{ -i txn} dt 2'd z e- - =
n=O t-w -00 +00f
00 2niL
znE{eit~n}....!!!.... t-w -00 n=O 00 co II
2 n=O 00 _1L
2 n=O znp[x -n znp[x -n 00 OJ -L
znE{e-iw~n}, n=O oJ •Proof. Let <Pn>n=O be a discrete proper probability distribution, then
00 •
'i' p E{ e-1W~n}, 1m w ~ O ' f f t ' d 1
L , 1S a c •• 0 a nonnega 1ve r.v. an we may app y
n=O n
(1.4). Since p is arbitrary, we may put p = (1 - z)zn, Vz
I
z € [0,1[.n n
2. An integral equation of the Pollaczek type In this section, we use the results of section tion of the Pollaczek type [7J.
to solve an integral
equa-In order to introduce the equation, let {T , n € ~} be a family of indepen--n
dent, identically distributedr.v. '8 with p.d.f. A(T) concentrated on ~ and
-icrT
with c.f. a(cr) = E{e -}, cr €~. Define
+ ••• + T ,
-n n€JN.
{~n' n € IN} is a random walk generated by A(·). Let us introduce the
follow-ing functions for z € [0,1[
(2. 1 ) n;::1 n z E{e-iw~n(s ~ O)} n -n
- L
= e 1m w ;:: 0 ,(2.2) 1m w :S 0 •
By (2.1) and (2.2), we obtain for Z E [O,I[ and a E R
(2.3) Define
l
x -n : = max{ 0, -s 1 ' -s 2' ••• , -n-s 1 } ,~I : =
a .
From the results of Fluctuation Theory, (see e.g. Cohen [IJ, pag. 151) we have for Z E [O,I[ and 1m w :S 0
n
- L
~ ECs > 0) n -n(2.4) Z e ne ]
- Z
A similar expression can be g1ven with respect to y+. To simplify notation, we define
n a. (z)
:=!
I
~ E(s=
0) • net n -n By (2.1), (2.2) we have lim y+(w,z) = e -20. (z),
1m w-++oo (2.5) Re w=O (2.6) lim y_(w,z) = 1.
1m w-+-oo Re w=OHence by corollary 1 and 2, (2.4), (2.5) and (2.6) we obtain after simplifi-cation (2.7) 1 _ 2 21Ti da y (a,z)---- , w E S - a-w - 0 0 (2.8) = ..,,-,-21T1 - 0 0
Rewriting (2.3) in the form yields by (2.7) and (2.8), (2.9)
-
..,...,..
z 2'1T~ Define (2. ] 0) _00 do a(o)y_(o , z ) -o - w 1m w ::;° .
By (2.9) and (2.10) we have that u_(w,z) ~s the solution of the integral
equation (2.11) -00 Finally, we define Since n z Pn := -
n
00 log(] - z) , n EN, z EO ]0,1[.I
p an(o) n=
log[] - za(a)] log(1 - z)n=1
00
*
~s the c.f. of the p.d.f.
I
PnAn (t), t E R, we have for W E Sn=1 +00
(2.12)
- 0 0
do
log[1 - za(o)J---- = a(z) + log y_(w,z) •
o-w
By (2.10) and (2.12) we finally obtain for w EO S_, Z E [O,I[
sech a(z)e -00
da log[1 -
Remarks. Integral equations similar to (2.11) have been studied in [7J and [6J using different methods. Pollaczek assumed that a(w) is analytic in some strip \Im w\ < 6 for some 6 > 0, whereas Ke'ilson and Kooharian made the as-sumption that a(o) 0(1) as
\0\
+ 00 • (In this case sech a(z) = 1.)By theorem 1 and a combination of some of the techniques presented in [IJ, [2J and [7J, we have shown that the above assumptions concerning a(o) are no longer necessary to solve equations similar to (2.11).
As a second application of theorem I, we study:
3. The transient behaviour of a Markov chain {w ; n ~ O} with state space [O,oo[
-n
-3.1. Introduction
Let {~n' n ~ O} be a Markov chain with state space [O,oo[ recursively
deter-mined by ~O a,;,s.
o ,
(3.1) _ o(n)} if w 0 '-H -G ' -n = .tax{o cr(n)
::!n+1 max{O,w + o(n) _ £.~n)}, if w#:
0 -n -F -nwhere ~(n) ~G , n = 0 I 2 , , , ••• are POSl. l.ve independent r.v.'s with p.d.f. "t" r[o(n) < tJ
=
G(t) and p[o(n) ~-G -G
We also assume that o(n) o(n)
-G '-H '
" (n) (n)
OJ = (dl.tto £H '£F ).
£in) are mutually independent for each n.
Remark. The transient behaviour of the Markov chain {w } has been studied -n
by J. Loris-Teghem [4J, [5J using an algebraic method based on a theorem due to G. Baxter [5]. Our approach is based on a Wiener-Hopf integral equation of the second kind and its reduction to a functional equation onR in terms of F.S.T.'s.
3.2. A Wiener-Hopf integral equation for the Markov chain {w } -n For t > 0 and n = 0,1,2, ••• we define
(3.2) W (t)
:=
pew < tJn -n
and W (0+) := lim W (t).
For t E lR, we define AF(t) := pea (0) -F _ a (0) -G < tJ
,
~(t) := P[a(O) - cr (0) < tJ.
-R -G Whence for t E lR 00 (3.3) ~(t)=
J
F(t + ,)dG(,) 0 00 (3.4) ~(t)=
J
R(t + ,)dG(,) 0By (3.1) we have for t > 0 and n 2
(n-I) (n-I)} (~n < t) = (max{O'~R - ~G < t)(~n_1 = 0) + (n-I) (n-l)} + (max{O'~n_l + ~F - ~G < t)(~n_1 > 0) • Since (w > 0)
=
-
(w = 0) we have -n-I - n - I ' or(w < t)
=
(max{O cr(n-I) - a(n-I)} < t)(w_n_1
=
0) + -n ' - R - G pew < tJ -n { (n-I) (n-I) - (max O'~F - ~G } < t)(~n-l = 0) + { (n-I) _ cr(n-I)} + (max O'~n-l + ~F -G < t) • (n-l) (n-I)} - P[max{O'£F - £G < tJ}Wn_1(0+) (n-I) (n-I)} + P[max{O'~n_1 + £F - ~G < tJ •P[max{O'~n_1
+~~n-I)
_~~n-I)}
< tJ=
P[~n-I
+~~n-I)
~G (n-I) < tJ ,+0:>
=
J
P[~n-I
< t -TJdP[~~O)
<"T(O) ':.G < T ] •Whence for t > 0 and n ~ I we obtain
+00
(3.5)
J
Wn_l(t -T)d~(t)
Since W(O+) = I and WO(t) = UO(t) where UO(t) is the Heaviside unit-step
function at t = 0, we have that equation (3.5) is also valid for n = I
(WI (t)
=
~(t) for t > 0).Remark. If H
=
F, then (3.5) reduces to Lindley's integral equation for theMarkov chain {; , n ~ O} recursively determined by
-n
{
''''
a.~. 0 ~O;
=
max{O.; -n+1 -n3.3. Solution of the integral equation (3.5)
In order to apply a Wiener-Hopf technique to equation (3.5), we define for
+ + + +
t ~ 0, n
Em;
W (t):=
W (t) (note that W (0)=
0 or W (0) f W (0+)=
Wn(O+»,n n n n n and for t ~ 0, n E
m;
W-(t) := O. n For t ~ 0, n E m we define +00J
W n_1 (t -T)d~(T)
• -00Finally we define for t E ~
+00
0.6)
I
W:_I(t-T)d~(T)
+W~(t)
We will now reduce equation (3.6) into a functional equation on R in terms
of Fourier-Stieltjes transforms. Therefore, we define for Z E [O,I[
00 00
J
e-iwtdW+(t), w_(w,z) :=I
z n 1m w ~ 0,
n=O n 0 0 00J
w+(w,z) :=I
z n e-iwtdW-(t) , 1m w ~ 0,
n=O -00 n 00 W(z) :=L
z~ (0+), n=O n +00 aF(w) :=f
e-iwtd~(t), 1m w = 0 , -00 +00 ~(w) :=J
e-iwtd~(t),
1m w=
0 • -00Put a
=
Re w. For a E Rand z E [O,I[ we obtain by (3.6) and the abovedefi-nitions (3.7) Note that
-iaa(n) iaa(n)
aF(a) = E{e -F }E{e -G }
. (n) . (n)
-l.cr~R l.cr~G
aR(a) = E{e }E{e } •
In order to simplify (3.7),
the property that !n
~ ~~n)
by (2.1), (2.2), (2.3)
00
we introduce a sequence of r.v.'s <!n>n=1 with
(n) -iaT
~G , n E N. Since E{e _n} = aF(a), we have
y+(a,z)
Substituting in (3.7) yields
(3.8)
w+(a,z) w_(a,z)
y+(a,z) - y_(a,z) =
aF(a) - ~(a)
Since is a F.S.T. of a function of bounded variation, we have
y+(a,z) that +T , 1
J
aF(a) - ~(a)~:
2T
y + (a ,z) da -Texists and is finite. Hence we may put +T l 1m ' --2T I
J
aF(a)-~(a) ( ) da T~ y+ a,z -T M(z) • Moreover +T (c:r, z) w (c:r, z) l' 1J
w (3.9) ~m 2T y Jc:r ,z) da=
lim y_(a,z) W(z).
T~ 1m w-l-OO -T Re w=O Whence +T 1J
w+ (a, z) 1 ' ~m2T
y+(c:r,z) d a=
T~ -T W ( z ) {I + zM ( z)} • Or equivalently (3.10) w+(a,z) lim=
W(z){1 + zM(z)} • 1m W"Hoo y + ( a , z) Re w=OBy (3.8), (3.9), (3.10) and corollary (2), we obtain for W E S_,.z E [O,I[
(3.11) +00 w+(a,z) I
~
da -~W(z){ 1 + zM(z)} 21fi y+Ca,z) - = a -w _00 +00 w_(a,z) w_(a,z)j
da21fi y (a, z)
- -
a-w 4W(z)-
y_(a,z).
(3.12)By (3.11), (4.11) and (3.8) we have (3.13)
=
y_(w,z)W(z){1 + 2 z M(z) + 2~i z _00 Finally, we have Whence (3.15) lim w_(w,z)=
(1 - z)-I • w+O WES y _ (0, z) [J +1-
M (z) + 1 im w-+O WES z 2~iBy (3.13) and (3.15) we have the following:
Theorem 2. +00 aFCr) - aH(T) z z
f
+i'
M(z)+r
y +h,
z) ~~ (3.16) w_(w,z)=
; (w, z) -00 +00 I +f
M ( z) + 1 im zf
a F ( T) - aH ( T ) 2~i y+CT,Z) W+O -00 WES where n -iws- I
-
z E{ ( I - e -n)(s > O)} n -n ;_(w,z)=
e n~1 - zCorollary 3. I f aF(a), aH(a) are 0 (I) as
I
aI
+ 00, then+00 aFCT) - ~(T) Z
1
dT+r
~~ y +h,
z) T -w (3.17) w_(w,z) ;_(w,z) -00 +00 + lim zi
~(T) - ~CT) dTw+O 2~i -00 y+(T,Z) T -w
WES
-dT T -w
dT T-W
Proof. Since lim aF(o) = lim ~(O) = 0, we have by (3.7),
lol~ lol~
lim w+(o,z)
lol~ lol~ lim w_ (0, z) W(z) •
Hence M(z) =
o.
Remark I. If aF(o) and ~(o) are Holder continuous at infinity, then (3.17)
can be obtained by the methods of the classical Hilbert problem [8J, [IOJ.
Remark 2. Finally, we remark that J. Loris-Tegh~'s result [5J can be
obtain-ed from (3.13). In order to show this we
w_(w,z) w_(ie,z) +00 I
f
=~ ~ 21T 1. w_ (w, z) w_(ie,z) _00 w_(ie,z)Since lim = I, we have
e+O w_(ie,z)
References
= w _ (w , z)[ I + lim
r
e+O 1T1. _00
remark that for any e > 0,
zW(z){aF(,) - ~(l)} I
U
- zaF«)}w_(T ,z) { T - W T - I.e I }dT.zW(z){aF (l) - ~ (T)} 1 I
U -
zaF(,)}w_("Z) {-:r=w-'1'_ie}d't'[IJ J.W. Cohen,"The single server queue". North-Holland Publ. Comp., Amster-dam, 1969.
[2J W. Feller, "An introduction to probability theory and its applications
II". John Wiley & Sons, Inc., New York, 1966.
[3J J. Keilson and A. Kooherian, "On the general time dependent queue with
a single server". Ann. Math. Stat. 31, 1960.
[4J J. Loris-Teghem, "On the waiting time distribution l.n a generalized
GI/G/I queueing system". J. Appl. Probe 8, 1971.
[5J J. Loris-Teghem, "Dne application d'un theoreme de G. Baxter
a
ladeter-mination de la distribution des temps d'attente dans un modele
[6J N.T. Muskhelishvili, "Some basic problems of the mathematical theory of elasticity". P. Noordhoff Ltd. Groningen, 1962.
[7] F. Pollaczek, "Problemes stochastiques poses par Ie phenomene de for-mation d'une queue d'attente
a
un guichet et par des phenomenes apparentes". Gauthiers-Villars. Fas. CXXXVT. Paris, 1957.[8J B.W. Roos, "Analytic functions and distributions in physics and engl.-neering". John Wiley & Sons, Inc. New York, 1969.
[9J L. Takacs, "On a method of Pollaczek". Stoch. Proc. and their applica-tions. Vol. I, 1973.
[IOJ E.J. Vanderperre, "A nonhomogeneous Hilbert problem with applications
to queueing theory". Memorandum CaSaR 73-08. Eindhoven, University of Technology. Eindhoven, 1973.