Professional Documents
Culture Documents
Probability, Random Variables and Stochastic Processes: Solutions Manual
Probability, Random Variables and Stochastic Processes: Solutions Manual
to accompany
Probability,
Random Variables
and
Stochastic Processes
Fourth Edition
Athanasios Papoulis
Polytechnic University
S. Unnikrishna Pillai
Polytechnic University
Solutions Manual to accompany
PROBABILITY, RANDOM VARIABLES AND STOCHASTIC PROCESSES, FOURTH EDITION
ATHANASIOS PAPOULIS
Published by McGraw-Hill Higher Education, an imprint of The McGraw-Hill Companies, Inc., 1221 Avenue of the Americas,
New York, NY 10020. Copyright © 2002 by The McGraw-Hill Companies, Inc. All rights reserved.
The contents, or parts thereof, may be reproduced in print form solely for classroom use with PROBABILITY, RANDOM
VARIABLES AND STOCHASTIC PROCESSES, FOURTH EDITION, provided such reproductions bear copyright notice, but may
not be reproduced in any other form or for any other purpose without the prior written consent of The McGraw-Hill Companies, Inc.,
including, but not limited to, in any network or other electronic storage or transmission, or broadcast for distance learning.
www.mhhe.com
1
Problem Solutions for Chapter 3
3.1 (a) P (A occurs atleast twice in n trials)
= 1 P (A never occurs in n trials) P (A occurs once in n trials)
; ;
3.2
P (doublesix) = 16 16 = 361
(b) Let n represent the number of wins required so that the net gain
or loss does not exceed $5. This gives
; 5 < n (50 2 n) < 5
;
;
We need
P (B A) = PP(AB
j
(A)
) = P (BC ) = P (B ) P (C ) = r :
P (A) P (A) n
3.9 There are 52
13 ways of selecting 13 cards out of 52 cards. The
number of ways to select 13 cards of any suit (out of 13 cards) equals
13 = 1. Four such (mutually exclusive) suits give the total number
13
of favorable outcomes to be 4. Thus the desired probability is given by
4
! = 6:3 10;12
52
13
4
3.10 Using the hint, we obtain
p (Nk+1 Nk ) = q (Nk Nk;1) 1
; ; ;
Let
Mk+1 = Nk+1 Nk ;
8 k n o
>
>
> p
q M 1 1 ( q ) k p=q
< 1 p q ;
p ;
; 6
=>
>
>
: M1 kp ;p=q
This gives
i;1
X
Ni = Mk+1
k=0
8 i;1 k
X
>
M1 + 1 q i
p q p = q
>
>
>
>
< p q ;
k=0
p ;
;
6
=>
>
>
>
>
: iM1 i(i2p 1)
;
;
p=q
M1 + p 1 q = pa + qb 1 1 (q=p
; ;
q=p :
)a+b
;
;
Thus 8
> a + b 1 (q=pa)+i b i
>
p q p = q
;
>
< p q 1 (q=p)
; 6
Ni = >>
; ;
;
>
: i(a + b i) ;p=q
5
which gives for i = a
8
> a + b 1 (q=pa)+a b a
>
p q p = q
;
>
< p q 1 (q=p)
;
;
;
6
Na = >> ;
>
: ab p=q
8
>
>
> b a + b 1 (p=qa)+b b p = q:
;
< 2p 1 2p 1 1 (p=q)
; 6
=>
; ; ;
>
>
: ab p=q
6
3.11
Pn = pPn+ + qPn;
Arguing as in (3.43), we get the corresponding iteration equation
Pn = Pn+ + qPn;
and proceed as in Example 3.15.
3.12 Suppose one best on k = 1 2 6.
Then
2
p1 = P (k appears on one dice) = 31 16 56
2
p2 = P (k appear on two dice) = 32 16 65
3
p3 = P (k appear on all the tree dice) = 61
3
p0 = P (k appear none) = 65
Thus, we get
Net gain = 2p1 + 3p2 + 4p3 p0 = 0:343:
;
215
Chapter 15
15.2 Note that both the row sums and column sums are unity in this
case. Hence P represents a doubly stochastic matrix here, and
1 1 ··· 1 1
1 1 ··· 1 1
1
Pn =
.. .. .. .. ..
m+1
. . . . .
1 1 ··· 1 1
1
lim P {xn = ek } = , k = 0, 1, 2, · · · m.
n→∞ m+1
15.3 This is the “success runs” problem discussed in Example 15-11
and 15-23. From Example 15-23, we get
1 uo
ui+1 = pi,i+1 ui = ui =
i+1 (i + 1)!
= π0m
P (z) = p0 + p1 z + p2 z 2 , since pk = 0, k ≥ 3.
P (z) = z.
Notice that
P (z) − z = p0 − (1 − p1 )z + p2 z 2
= p0 − (p0 + p2 )z + p2 z 2
= (z − 1)(p2 z − p0 )
and hence the two roots of the equation P (z) = z are given by
p0
z1 = 1, z2 = .
p2
Thus if p2 < p0 , then z2 > 1 and hence the smallest positive root of
P (z) = z is 1, and it represents the probability of extinction. It follows
217
Similarly
E{xn+2 |xn } = E{E{xn+2 |xn+1 , xn }}
= E{E{xn+2 |xn+1 }|xn }
= E{µxn+1 |xn } = µ2 xn
or
x ∆
E{wn+r |wn = = w} = wn
µn
which gives
E{wn+r |wn } = wn ,
the desired result.
15.7
sn = x 1 + x 2 + · · · + x n
where xn are i.i.d. random variables. We have
sn+1 = sn + xn+1
so that
p∗ij = pij
and let
aij aij
pij = P = .
k aik ri
Then
a a a
pji = X ji = rjij = rijj
ajm
m
(i)
a
= rrji riji = rrji pij
or
ri pij = rj pji .
Hence X X X
ri pij = rj pji = rj pji = rj , (ii)
i i i
since
aji rj
P
i
X
pji = = = 1.
i rj rj
Notice that (ii) satisfies the steady state probability distribution equa-
tion (15-167) with
qi = c r i , i = 1, 2, · · ·
where c is given by
X X 1 1
c ri = qi = 1 =⇒ c = P =P P .
i i i ri i j aij
220
Thus
ri j aij
P
qi = P =P P >0 (iii)
i ri i j aij
represents the stationary probability distribution of the chain.
With (iii) in (i) we get
qi
pji = pij
qj
or
qj pji
pij = = p∗ij
qi
and hence the chain is time-reversible.
15.10 (a) M = (mij ) is given by
M = (I − W )−1
or
(I − W )M = I
M = I + WM
which gives
mij = δij + wik mkj , ei , ej ∈ T
P
k
(b) The general case is solved in pages 743-744. From page 744,
with N = 6 (2 absorbing states; 5 transcient states), and with r = p/q
we obtain
(rj − 1)(r 6−i − 1)
, j≤i
(p − q)(r 6 − 1)
mij =
(ri − 1)(r 6−i − rj−i )
, j ≥ i.
(p − q)(r 6 − 1)
where X
pij > 0, pij = 1,
j
and this may not be always possible. For example in a two state chain,
let
α 1−α
P =
1−β β
so that
α2 + (1 − α)(1 − β)
(α + β)(1 − α)
A = P2 = .
(α + β)(1 − β) β 2 + (1 − α)(1 − β)
Hence condition (i) necessary. Since 0 < α < 1, 0 < β < 1, we also
get 1 < a11 + a22 ≤ 2. Futher, the condition (i) is also sufficient in the
2 × 2 case, since a11 + a22 > 1, gives
and hence √
α+β =1± a11 + a22 − 1
and this equation may be solved for all admissible set of values 0 <
α < 1 and 0 < β < 1.
15.12 In this case the chain is irreducible and aperiodic and there are
no absorption states. The steady state distribution {uk } satisfies (15-
167),and hence we get
N
à !
X X N k N −k
uk = uj pjk = uj p j qj .
j j=0 k
222
Then if α > 0 and β > 0 then “fixation to pure genes” does not occur.
15.13 The transition probabilities in all these cases are given by (page
765) (15A-7) for specific values of A(z) = B(z) as shown in Exam-
ples 15A-1, 15A-2 and 15A-3. The eigenvalues in general satisfy the
equation X (k) (k)
pij xj = λk xi , k = 0, 1, 2, · · · N
j
cases.
However to determine the remaining eigenvalues we can exploit the
relation in (15A-7). From there the corresponding conditional moment
generating function in (15-291) is given by
N
pij sj
X
G(s) = (i)
j=0
where
{AN −1 (z) A0 (z)}N −1
λ1 = . (vii)
{AN (z)}N
Notice that (vi) can be written as
P x 1 = λ 1 x1 , x1 = [0, 1, 2, · · · N ]T
and by direct computation with A(z) = B(z) = (q + pz)2 (Example
15A-1) we obtain
{(q + pz)2(N −1) 2p(q + pz)}N
λ1 =
{(q + pz)2N }N
à !
2N
2p q N pN −1
2p{(q + pz)2N −1 }N −1 N − 1
= = = 1.
{(q + pz)2N }N
à !
2N N N
q p
N
Thus N j=0 pij j = i and from (15-224) these chains represent Martin-
P
j=0
00
{i(i − 1)Ai−2 (sz)[A0 (sz)]2 z B N −i (z) + iAi−1 (sz) A (sz)z B N −i (z)}N −1
= .
{Ai (z) B N −i (z)}N
2
{i Ai−2 (sz) B N −i (z)[(i − 1) (A0 (sz)) + A(sz) A00 (sz)]}N −2
= .
{Ai (z) B N −i (z)}N
224
N
X
pij j(j − 1) = λ2 i(i − 1) + iµ2 (viii)
j=0
where
{AN −2 (z) [A0 (z)]2 }N −2
λ2 =
{AN (z)}N
and
{AN −1 (z) A00 (z)}N −2
µ2 = .
{AN (z)}N
Eq. (viii) can be rewritten as
N
pij j 2 = λ2 i2 + (polynomial in i of degree ≤ 1)
X
j=0
N
pij j k = λk ik + (polynomial in i of degree ≤ k − 1)
X
(ix)
j=0
where
{AN −k (z) [A0 (z)]k }N −k
λk = , k = 1, 2, · · · N. (x)
{AN (z)}N
Equations (viii)–(x) motivate to consider the identities
P q k = λ k qk (xi)
(a) The transition probabilities in this case follow from Example 15A-1
(page 765-766) with A(z) = B(z) = (q + pz)2 . Thus using (ix) we
225
(b) The transition probabilities in this case follows from Example 15A-2
(page 766) with
A(z) = B(z) = eλ(z−1)
and hence
{eλ(N −k)(z−1) λk eλk(z−1) }N −k
λk =
{eλN (z−1) }N
λk {eλN z }N −k λk (λN )N −k /(N − k)!
= =
λN z
{e }N (λN )N /N !
µ ¶
N! 1 2 ··· 1 − k − 1 ,
³ ´³ ´
= = 1− N 1− N N k = 1, 2, · · · N
(N − k)! N k
(c) The transition probabilities in this case follow from Example 15A-3
(page 766-767) with
q
A(z) = B(z) = .
1 − pz
Thus
{1/(1 − pz)N +k }N −k
λk = p k
{1/(1 − pz)N }N
à ! à !
−(N + k) 2N − 1
N − k! N −k!
= (−1)k à =à , r = 2, 3, · · · N
−N 2N − 1
N N
226
where
Wik = pjk , j, k = 1, 2, · · · N − 1,
X
mi = 1 + pik mk = 1 + pi,i+1 mi+1 + pi,i−1 mi−1
k∈T
= 1 + p mi+1 + q mi−1 ,
or
mk = 1 + p mk+1 + q mk−1 .
This gives
p (mk+1 − mk ) = q (mk − mk−1 ) − 1
Let
Mk+1 = mk+1 − mk
Mk+1 = pq Mk − p1
· ¸
= pq M1 − p1 1 + pq + pq + · · · + pq
³ ´k ³ ´2 ³ ´k−1
q M − 1 n1 − ( q )k o , p 6= q
³ ´k
p 1 p−q p
=
M1 − kp , p=q
227
This gives
i−1
X
mi = Mk+1
k=0
i−1 ³ ´
³
M + 1
´ X
q k− i , p 6= q
1 p − q p p−q
k=0
=
i(i − 1)
iM1 − 2p , p=q
³ ´ 1 − (q/p)i
M + 1 i , p=
− p− 6 q
1 p−q q
1 − q/p
=
i(i − 1)
iM1 − 2p , p=q
by writing
1 − (q/p)a (q/p)a − (q/p)a+b 1 − (p/q)b
=1− =1−
1 − (q/p)a+b 1 − (q/p)a+b 1 − (p/q)a+b
(see also problem 3-10).
228
Chapter 16
ρn p , n ≤ 1
n! 0
pn =
ρn p 0 , 1 < n ≤ m
= ρ n p0 , 0≤n≤m
Thus
m m
(1 − ρm+1 )
ρn = p 0
X X
pn = p 0 =1
n=0 n=0 1−ρ
1−ρ
=⇒ p0 =
1 − ρm+1
and hence
1−ρ n
pn = ρ , 0 ≤ n ≤ m, ρ 6= 1
1 − ρn+1
16.2 (a) Let n1 (t) = X + Y , where X and Y represent the two queues.
Then
pn = P {n1 (t) = n} = P {X + Y = n}
n
X
= P {X = k} P {Y = n − k}
k=0
n
X (i)
= (1 − ρ)ρk (1 − ρ)ρn−k
k=0
= (n + 1)(1 − ρ)2 ρn , n = 0, 1, 2, · · ·
where ρ = λ/µ.
229
0
(b) When the two queues are merged, the new input rate λ =
λ + λ = 2λ. Thus from (16-102)
0
(λ /µ)n (2ρ)n
p 0 =
n! p0 , n < 2
n!
pn = 0
22 ( λ )n p = 2ρn p , n ≥ 2.
0 0
2! 2µ
Hence
∞
X ∞
X
pk = p0 (1 + 2ρ + 2 ρk )
k−0 k=2
2
= p0 (1 + 2ρ + 12ρ− ρ)
= 1 p−0 ρ ((1 + 2ρ) (1 − ρ) + 2ρ2 )
= 1 p−0 ρ (1 + ρ) = 1
1−ρ
=⇒ p0 = , (ρ = λ/µ). (ii)
1+ρ
Thus
2 (1 − ρ) ρn /(1 + ρ), n ≤ 1
pn = (iii)
(1 − ρ)/(1 + ρ), n=0
∞
0 X
E{X} = L1 = (n − 1) pn
n=2
230
1 2 ρ2
= (1 − ρ) ρ = .
(1 − ρ)2 (1 − ρ)
λi,i+1 = (m − i) λ, λi,i−1 = iµ
231
m λ p0 = µ p1 (i)
ρn1 p0 ,
n<m
=
ρm−1
1 ρ2n−m+1 p0 , n ≥ m,
where "m−1 #
∞ ∞
ρm−1
X X X
pn = p 0 ρk1 + 1 ρ2 ρn2
n=0 k=0 n=0
ρm ρ2 ρm−1
" #
1− 1 1
= p0 + =1
1 − ρ1 1 − ρ2
232
gives
!−1
1 − ρm ρ2 ρm−1
Ã
1 1
p0 = + .
1 − ρ1 1 − ρ2
(b)
∞
X
L = n pn
n=0
"m−1 ∞
#
n ρm−1 ρ2n−m+1
X X
= p0 n ρn1 + 1
n=0 n=m
!m−2
m−1
à ∞
ρ1
n ρn−1 n ρn−1
X X
= p 0 ρ1 1 + ρ1 2
n=0 ρ2 n=m
Ãm−1 ! Ã !m−2
∞
d X ρ1 d X
= p 0 ρ1 ρn1 + ρ1 ρn
dρ1 n=0 ρ2 dρ2 n=m 2
!m−2 !
1 − ρm ρm
à ! à Ã
d 1 ρ1 d
= p 0 ρ1 + ρ1
dρ1 1−ρ ρ2 dρ 1−ρ
m−1
ρ1 [1 + (m − 1)ρm ] ρ2 ρm−1
" #
1 − mρ1 1 + [m − (m − 1)ρ2 ]
= p0 2
+ .
(1 − ρ1 ) (1 − ρ2 )2
(λ/µ)n
n! p0 , n<r
pn = (λ/µ)r n−r
r! (pλ/rµ) , n ≥ r.
233
16.6
m−1
X
P {w > t} = pn P (w > t|n)
n=r
m−1 µ ¶n−r
=
X
pn (1 − Fw (t|n)) =
X
pr λ (1 − Fw (t|n))
rµ
n=r
(γµ)n−r+1 tn−r
fw (t|n) = e−γµt (see 16.116)
(n − r)!)
and
n−r
X (γµt)k −γµt
Fw (t|n) = 1 − e (see 4.)
k=0 k!
so that
n−r
X (γµt)k −γµt
1 − Fw (t|n) = e
k=0 k!
m−r−1 m−r−1
X (γµt)i X k
P {w > t} = pr e−γµt i! ρ
i=0 k=i
m−r−1
(γµt)i i
= 1 p−r ρ e−γµt
X
m−r
i! (ρ − ρ ), ρ = λ/γµ.
i=0
234
(γµρt)i
∞
P (w > t) = 1 p−r ρ e−γµt
X
i!
i=0
∞
µ
ck z k pm z m = 0
X X
−(ρ + 1) (P (z) − p0 ) + (P (z) − p0 − p1 z) + λ
z k=1 m=0
which gives
P (z)[1 − z − ρz (1 − C(z))] = p0 (1 − z)
or
p0 (1 − z)
P (z) = .
1 − z − ρz (1 − C(z))
−p0 −p0
1 = P (1) = =
−1 − ρ + ρz C (z) + ρC(z)
0 −1 + ρC 0 (1)
=⇒ p0 = 1 − ρ0 , ρ0 = ρC 0 (1).
Let
1 − C(z)
D(z) = .
1−z
Then
1 − ρL
P (z) = .
1 − ρzD(z)
(1 − ρc )
L = P 0 (1) = ρ (D(1) + D 0 (1))
(1 − ρc )2
= 1 (C 0 (1) + D 0 (1))
(1 − ρc )
C 0 (1) = E(x)
1 − C(z)
D(z) =
1−z
(1 − z) (−C 0 (z)) − (1 − C(z)) (−1)
D0 (z) =
(1 − z)2
1 − C(z) − (1 − z)C 0 (z)
=
(1 − z)2
By L-Hopital’s Rule
ρ (E(X) + E(X 2 ))
L= .
2 (1 − ρE(X))
(d)
C(z)z m E(X) = m
1−ρ
P (z) =
1−ρ mk=1 z
k
P
1 − z m m−1
zk
X
D(z) = =
1−z k=0
E(X) = m, E(X 2 ) = m2
ρ(m + m2 )
L=
2(1 − ρm)
236
(e)
qz
C(z) =
1 − Pz
1 − ρ0 qz
P (z) = , C(z) =
1 − ρzD(z) 1 − pz
qz
1 − C(z) 1 − 1−P (z) 1 − P z − (1 − P )z 1−z 1
D(z) = = = = =
1−z 1−z (1 − z)(1 − P z) (1 − z)(1 − P z) 1 − Pz
(1 − ρ0 )(1 − pz) (1 − ρ0 )(1 − pz)
P (z) = =
1 − pz − ρz 1 − (p + ρ)z
(1 − pz)q − qz(−p) q 1
C 0 (1) = = =
(1 − P z)2 q2 q
1 − C(z)
D(z) =
1−z
D(1) = C 0 (1)
1 − ρc
L = P 0 (1) = (ρ · C 0 (1) + ρ · D 0 (1))
(1 − ρc )2
−(1 − z)C 0 (z) − (1 − C(z)) (ρ − 1) 1 − C(z) − (1 − z)C 0 (1)
D0 (z) = =
(1 − z)2 (1 − z)2
−C 0 (z) − (−1)C 0 (z) − (1 − z)C 00 (z)
limz→1 D0 (z) = limz→1
2(1 − z)
−(1 − z)C 00 (z) ρ00 (z)
= = 2
−2(1 − z)
C 00 (1)
D0 (1) =
2
2
ρE(X) + ρE(X 2 )
à !
1 ρ (E(X ) − E(X))
L= ρE(X) + = .
(1 − ρc ) 2 2(1 − ρc )
∞ ∞ ∞
µ X
(λ + µ) pn z n + n+m
pn−1 z n−1 = 0
X X
− p n+m z + λz
n=1 z n n=1 n=1
237
or
m
à !
1 X
k
−(1 + ρ) (P (z) − p0 ) + m P (z) − pk z + ρzP (z) = 0
z k=0
which gives
h i m
P (z) ρ z m+1 − (ρ + 1) z m + 1 = pk z k − p0 (1 + ρ) z m
X
k=0
or
m
X
pk z k − p0 (1 + ρ) z m
k=0 N (z)
P (z) = = . (i)
ρ z m+1 − (ρ + 1) z m +1 M (z)
where
f (z) = −(1 + ρ) z m ,
g(z) = 1 + ρ z m+1 .
Notice that |f (z)| > |g(z)| in a circle defined by |z| = 1 + ε, ε > 0.
Hence by Rouche’s Theorem f (z) and f (z)+g(z) have the same number
of zeros inside the unit circle (|z| = 1 + ε). But f (z) has m zeros inside
the unit circle. Hence f (z) + g(z) = M (z) also has m zeros inside the
unit circle. Hence
M (z) = M1 (z) (z − z0 ) (ii)
where |z0 | > 1 and M1 (z) is a polynomial of degree m whose zeros are
all inside or on the unit circle. But the moment generating function
P (z) is analytic inside and on the unit circle. Hence all the m zeros
of M (z) that are inside or on the unit circle must cancel out with the
zeros of the numerator polynomial of P (z). Hence
k=0
or
m−1
X m−1
X
(1 − z m ) pk z k (z0 − 1) zk
k=0 k=0
P (z) = =
ρ z m+1 − (ρ + 1) z m + 1 m (z0 − z)
239
(1 + r + · · · + r k ) p0 ,
k ≤m−1
pn =
rn−m+1 (1 + r + · · · + r m−1 ) p0 , k ≥ m
where
1−r 1
p0 = , r= .
m z0
Finally
∞
X 0
L= n pn = Pn (1).
n=0
But
m−1
X m−1
X
k z k−1 (1 − rz) − z k (−r)
1−r
µ ¶
0 k=1 k=0
P (z) =
m (1 − rz)2
so that
1−r m−1+r
0 m − (1 − r)
L = P (1) = 2
=
m (1 − r) m (1 − r)
1 1
= − .
1−r m
This gives
ρ z −1/m = (1 + ρ) − z (ii)
ρ x = (1 + ρ) − x−m
or
ρ xm+1 − (1 + ρ) xm + 1 = 0 (iii)
1 − z[1 + ρ (1 − z)]m = 0
241
which is equivalent to
1 + ρ (1 − z) = z −1/m . (i)
[(1 + ρ) − ρ xm ] x = 1
ρ xm+1 − (1 + ρ) x + 1 = 0. (ii)
A(z) = Φs (λ (1 − z))
k
di e−λ Ti (1−z)
X
=
i=1
k
di e−λ Ti eλ Ti z
X
=
i=1
k ∞ ∞
(λ T j zj
−λ Ti i)
aj z j .
X X X
= di e j!
=
i=1 j=0 j=0
Hence
k
(λ Ti )j
di e−λ Ti
X
aj = , j = 0, 1, 2, · · · . (i)
i=1 j!
242
(m)
and let {ck } represent the m−fold convolution of the sequence {ck }
with itself. Then the steady-state probabilities are given by (16.203) as
n
∞ X
X (m)
qn = (1 − ρ) ak cn−k .
m=0 k=0
where
Ak = ”k customers arrive during a service time”
and
This gives
Z ∞ (λt)k
ak,i = e−λt dBi (t)
0 k!
(λt)k µ1 λk
Z ∞
e−λt µ1 e−µ1 t dt =
, i=0 (i)
k! (λ + µ1 )k+1
0
= Z ∞ (λt)k µ2 λk
µ2 e−µ2 t dt =
e−λt , i≥1
k! (λ + µ2 )k+1
0
243
This gives
1
, i=0
1 + ρ1 (1 − z)
∞
ak,i z k =
X
Ai (z) = 1 (ii)
k=0
, i≥1
1 + ρ2 (1 − z)
and (see(15.212))
∞
qj z j
X
Q(z) =
j=0
∞ ∞
aj,0 z j +
X X
= q0 qi aj−i+1,i
j=0 j=0 (iv)
∞ ∞
qi z i am,i z m z −1
X X
= q0 A0 (z) +
i=1 m=0
Since
h 0 0
i
q0 A0 (1) + A0 (1) − A1 (1)
Q(1) = 1 = 0
1 − A1 (1)
q0 (1 + ρ1 − ρ2 )
=
1 − ρ2
we obtain
1 − ρ2
q0 = . (viii)
1 + ρ 1 − ρ2
and
à ! ∞
à !i
1 ρ2 ρ1
zi.
X
Q2 (z) = 1− z
1 + ρ 1 − ρ2 1 + ρ1 i=0 1 + ρ1
with q0 as in (viii).
245
16.13 From (16-209), the Laplace transform of the waiting time distri-
bution is given by
1−ρ
Ψw (s) = ³
1−Φs (s)
´
1−λ s
1−ρ (i)
= ³
1−Φs (s)
´.
1 − ρµ s
Let
Z t
Fr (t) = µ [1 − B(τ )]dτ
0
· Z t ¸ (ii)
=µ t− B(τ )dτ .
0
0
ρ = A (1) > 1
where
∞
ak z k
X
A(z) =
k=0
A(z) = z (i)
has a unique positive root π0 < 1. On the other hand, the transient
state probabilities {σi } satisfy the equation (15.236). By direct substi-
tution with xi = π0i we get
∞ ∞
aj−i+1 π0j
X X
pij xj = (ii)
j=1 j=1
∞ ∞
ak π0k+i−1 = π0i−1 ak π0k
X X
k=2−i k=0
16.15 (a) The transition probability matrix here is the truncated version
of (15.34) given by
247
m−2
X
a0 a1 a2 · · · · am−2 1− ak
k=0
m−2
X
a0 a1 a2 · · · · am−2 1− ak
k=0
m−3
X
P = 0 a 0 a1 · · · · am−3 1− ak (i)
k=0
.. .. .. .. .. ..
. . . . . .
0 0 0 · · · a0 a1 1 − (a0 + a1 )
0 0 0 · · · 0 a0 1 − a0
j+1
X
qj∗ = q0∗ aj + qi∗ aj−i+1 , j = 0, 1, 2, · · · , m − 2 (ii)
i=1
qj∗ = c qj , j = 0, 1, 2, · · · , m − 1 (iv)
1
c= m−1
. (v)
X
qi
i=0
16.16 (a) The event {X(t) = k} can occur in several mutually exclusive
ways, viz., in the interval (0, t), n customers arrive and k of them
continue their service beyond t. Let An = “n arrivals in (0, t)”, and
Bk,n =“exactly k services among the n arrivals continue beyond t”,
then by the theorem of total probability
∞
X ∞
X
P {X(t) = k} = P {An ∩ Bk,n } = P {Bk,n |An }P (An ).
n=k n=k
But P (An ) = e−λt (λt)n /n!, and to evaluate P {Bk,n |An }, we argue as
follows: From (9.28), under the condition that there are n arrivals in
(0, t), the joint distribution of the arrival instants agrees with the joint
distribution of n independent random variables arranged in increasing
order and distributed uniformly in (0, t). Hence the probability that a
service time S does not terminate by t, given that its starting time x
has a uniform distribution in (0, t) is given by
Z t
pt = P (S > t − x|x = x)fx (x)dx
0
Z t 1 1Z t α(t)
= [1 − B(t − x)] dx = (1 − B(τ )) dτ =
0 t t 0 t
à !
n k
P {Bk,n |An } = p (1 − pt )n−k , k = 0, 1, 2, · · · n,
k t
249
and
!k µ ¶n−k
(λt)n n
∞
à !Ã
X
−λt α(t) 1Z t
P {X(t) = k} = e B(τ )dτ
n=k n! k t t 0
¶n−k
1Z t
µ
[λα(t)] k X∞ λt B(τ )dτ
= e−λt t 0
k! n=k (n − k)!
h R i
[λα(t)]k −λ t− 0t B(τ )dτ
= e
k!
[λα(t)]k −λ R t [1−B(τ )]dτ
= e 0
k!
[λα(t)]k −λ α(t)
= e , k = 0, 1, 2, · · ·
k!
(i)
(b) Z ∞
lim α(t) = [1 − B(τ )]dτ
t→∞ 0
(ii)
= E{s}
ρk
lim P {x(t) = k} = e−ρ (iii)
t→∞ k!
where ρ = λ E{s}.