This action might not be possible to undo. Are you sure you want to continue?
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
Advanced Probability Theory
for Biomedical Engineers
i
Copyright © 2006 by Morgan & Claypool
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations
in printed reviews, without the prior permission of the publisher.
Advanced Probability Theory for Biomedical Engineers
John D. Enderle, David C. Farden, and Daniel J. Krause
www.morganclaypool.com
ISBN10: 1598291505 paperback
ISBN13: 9781598291506 paperback
ISBN10: 1598291513 ebook
ISBN13: 9781598291513 ebook
DOI 10.2200/S00063ED1V01Y200610BME011
A lecture in the Morgan & Claypool Synthesis Series
SYNTHESIS LECTURES ON BIOMEDICAL ENGINEERING #11
Lecture #11
Series Editor: John D. Enderle, University of Connecticut
Series ISSN: 19300328 print
Series ISSN: 19300336 electronic
First Edition
10 9 8 7 6 5 4 3 2 1
Printed in the United States of America
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
Advanced Probability Theory
for Biomedical Engineers
John D. Enderle
Program Director & Professor for Biomedical Engineering,
University of Connecticut
David C. Farden
Professor of Electrical and Computer Engineering,
North Dakota State University
Daniel J. Krause
Emeritus Professor of Electrical and Computer Engineering,
North Dakota State University
SYNTHESIS LECTURES ON BIOMEDICAL ENGINEERING #11
M
&C
Morgan
&
Claypool Publishers
iii
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
iv
ABSTRACT
This is the third in a series of short books on probability theory and random processes for
biomedical engineers. This book focuses on standard probability distributions commonly en
countered in biomedical engineering. The exponential, Poisson and Gaussian distributions are
introduced, as well as important approximations to the Bernoulli PMF and Gaussian CDF.
Many important properties of jointly Gaussian random variables are presented. The primary
subjects of the ﬁnal chapter are methods for determining the probability distribution of a func
tion of a random variable. We ﬁrst evaluate the probability distribution of a function of one
random variable using the CDF and then the PDF. Next, the probability distribution for a
single random variable is determined from a function of two random variables using the CDF.
Then, the joint probability distribution is found from a function of two random variables using
the joint PDF and the CDF.
The aim of all three books is as an introduction to probability theory. The audience
includes students, engineers and researchers presenting applications of this theory to a wide
variety of problems—as well as pursuing these topics at a more advanced level. The theory
material is presented in a logical manner—developing special mathematical skills as needed.
The mathematical background required of the reader is basic knowledge of differential calculus.
Pertinent biomedical engineering examples are throughout the text. Drill problems, straight
forward exercises designed to reinforce concepts and develop problem solution skills, follow
most sections.
KEYWORDS
Probability Theory, Random Processes, Engineering Statistics, Probability and Statistics for
Biomedical Engineers, Exponential distributions, Poisson distributions, Gaussian distributions
Bernoulli PMF and Gaussian CDF. Gaussian random variables
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
v
Contents
5. Standard Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
5.1 Uniform Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
5.2 Exponential Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
5.3 Bernoulli Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.3.1 Poisson Approximation to Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.3.2 Gaussian Approximation to Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
5.4 Poisson Distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.4.1 Interarrival Times. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.5 Univariate Gaussian Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.5.1 Marcum’s Q Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.6 Bivariate Gaussian Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.6.1 Constant Contours. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6. Transformations of RandomVariables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1 Univariate CDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.1 CDF Technique with Monotonic Functions . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.2 CDF Technique with Arbitrary Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.2 Univariate PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.1 Continuous Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.2 Mixed Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2.3 Conditional PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 One Function of Two Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.4 Bivariate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.1 Bivariate CDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.2 Bivariate PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
vi
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
vii
Preface
This is the third in a series of short books on probability theory and random processes for
biomedical engineers. This text is written as an introduction to probability theory. The goal
was to prepare students at the sophomore, junior or senior level for the application of this
theory to a wide variety of problems  as well as pursue these topics at a more advanced
level. Our approach is to present a uniﬁed treatment of the subject. There are only a few key
concepts involved in the basic theory of probability theory. These key concepts are all presented
in the ﬁrst chapter. The second chapter introduces the topic of random variables. The third
chapter focuses on expectation, standard deviation, moments, and the characteristic function.
In addition, conditional expectation, conditional moments and the conditional characteristic
function are also discussed. The fourth chapter introduces jointly distributed random variables,
along with joint expectation, joint moments, and the joint characteristic function. Convolution
is also developed. Later chapters simply expand upon these key ideas and extend the range of
application.
This short book focuses on standard probability distributions commonly encountered in
biomedical engineering. Here in Chapter 5, the exponential, Poisson and Gaussian distributions
are introduced, as well as important approximations to the Bernoulli PMF and Gaussian CDF.
Many important properties of jointly distributed Gaussian random variables are presented.
The primary subjects of Chapter 6 are methods for determining the probability distribution of
a function of a random variable. We ﬁrst evaluate the probability distribution of a function of
one random variable using the CDF and then the PDF. Next, the probability distribution for a
single random variable is determined from a function of two random variables using the CDF.
Then, the joint probability distribution is found from a function of two random variables using
the joint PDF and the CDF.
Aconsiderable effort has been made to develop the theory in a logical manner  developing
special mathematical skills as needed. The mathematical background required of the reader is
basic knowledge of differential calculus. Every effort has been made to be consistent with
commonly used notation and terminology—both within the engineering community as well as
the probability and statistics literature.
The applications and examples given reﬂect the authors’ background in teaching prob
ability theory and random processes for many years. We have found it best to introduce this
material using simple examples such as dice and cards, rather than more complex biological
P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
MOBK042FM MOBK042Enderle.cls October 30, 2006 19:55
viii PREFACE
and biomedical phenomena. However, we do introduce some pertinent biomedical engineering
examples throughout the text.
Students in other ﬁelds should also ﬁnd the approach useful. Drill problems, straightfor
ward exercises designed to reinforce concepts and develop problem solution skills, follow most
sections. The answers to the drill problems follow the problem statement in random order.
At the end of each chapter is a wide selection of problems, ranging from simple to difﬁcult,
presented in the same general order as covered in the textbook.
We acknowledge and thank WilliamPruehsner for the technical illustrations. Many of the
examples and end of chapter problems are based on examples from the textbook by Drake [9].
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
1
C H A P T E R 5
Standard Probability Distributions
A surprisingly small number of probability distributions describe many natural probabilistic
phenomena. This chapter presents some of these discrete and continuous probability distribu
tions that occur often enough in a variety of problems to deserve special mention. We will see
that many random variables and their corresponding experiments have similar properties and
can be described by the same probability distribution. Each section introduces a new PMF or
PDF. Following this, the mean, variance, and characteristic function are found. Additionally,
special properties are pointed out along with relationships among other probability distribu
tions. In some instances, the PMF or PDF is derived according to the characteristics of the
experiment. Because of the vast number of probability distributions, we cannot possibly discuss
them all here in this chapter.
5.1 UNIFORMDISTRIBUTIONS
Deﬁnition 5.1.1. The discrete RV x has a uniform distribution over n points (n > 1) on the
interval [a, b] if x is a lattice RV with span h = (b −a)/(n −1) and PMF
p
x
(α) =
_
1/n, α = kh +a, k = 0, 1, . . . , n −1
0, otherwise.
(5.1)
The mean and variance of a discrete uniform RV are easily computed with the aid of
Lemma 2.3.1:
η
x
=
1
n
n−1
k=0
(kh +a) =
h
n
γ
[2]
n
2
+a =
1
n
b −a
n −1
n(n −1)
2
+a =
b +a
2
, (5.2)
and
σ
2
x
=
1
n
n−1
k=0
_
kh −
b −a
2
_
2
=
(b −a)
2
n
n−1
k=0
_
k
2
(n −1)
2
−
k
n −1
+
1
4
_
. (5.3)
Simplifying,
σ
2
x
=
(b −a)
2
12
n +1
n −1
. (5.4)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
2 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
0 .5 1
x
p ( )
.05
(a)
1
x
(t )
t
10 30 50
(b)
φ
α
α
FIGURE5.1: (a) PMF and (b) characteristic function magnitude for discrete RV with uniform distri
bution over 20 points on [0, 1].
The characteristic function can be found using the sum of a geometric series:
φ
x
(t) =
e
jat
n
n−1
k=0
(e
jht
)
k
=
e
jat
n
1 −e
jhnt
1 −e
jht
. (5.5)
Simplifying with the aid of Euler’s identity,
φ
x
(t) = exp
_
j
a +b
2
t
_
sin
_
b−a
2
n
n−1
t
_
n sin
_
b−a
2
1
n−1
t
_. (5.6)
Figure 5.1 illustrates the PMFand the magnitude of the characteristic function for a discrete RV
which is uniformly distributed over 20 points on [0, 1]. The characteristic function is plotted
over [0, π/h], where the span h = 1/19. Recall from Section 3.3 that φ
x
(−t) = φ
∗
x
(t) and that
φ
x
(t) is periodic in t with period 2π/h. Thus, Figure 5.1 illustrates onehalf period of φ
x
(·).
Deﬁnition 5.1.2. The continuous RV x has a uniform distribution on the interval [a, b] if x has
PDF
f
x
(α) =
_
1/(b −a), a ≤ α ≤ b
0, otherwise.
(5.7)
The mean and variance of a continuous uniform RV are easily computed directly:
η
x
=
1
b −a
b
_
a
αdα =
b
2
−a
2
2(b −a)
=
b +a
2
, (5.8)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 3
0 .5 1
1
(a)
1
x
(t ) φ
t
10 30 50
(b)
x
f (α)
FIGURE 5.2: (a) PDF and (b) characteristic function magnitude for continuous RV with uniform
distribution on [0, 1].
and
σ
2
x
=
1
b −a
b
_
a
_
α −
b +a
2
_
2
dα =
(b −a)
2
12
. (5.9)
The characteristic function can be found as
φ
x
(t) =
1
b −a
b
_
a
e
jαt
dα =
exp
_
j
b+a
2
t
_
b −a
(b−a)/2
_
−(b−a)/2
e
jαt
dα.
Simplifying with the aid of Euler’s identity,
φ
x
(t) = exp
_
j
a +b
2
t
_
sin
_
b−a
2
t
_
b−a
2
t
. (5.10)
Figure 5.2 illustrates the PDF and the magnitude of the characteristic function for a continuous
RV uniformly distributed on [0, 1]. Note that the characteristic function in this case is not
periodic but φ
x
(−t) = φ
∗
x
(t).
Drill Problem 5.1.1. A pentahedral die (with faces labeled 0,1,2,3,4) is tossed once. Let x be a
random variable equaling ten times the number tossed. Determine: (a) p
x
(20), (b) P(10 ≤ x ≤ 50),
(c) E(x), (d) σ
2
x
.
Answers: 20, 0.8, 200, 0.2.
Drill Problem5.1.2. Random variable x is uniformly distributed on the interval [−1, 5]. Deter
mine: (a) F
x
(0), (b) F
x
(5), (c) η
x
, (d) σ
2
x
.
Answers: 1, 1/6, 3, 2.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
4 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
5.2 EXPONENTIAL DISTRIBUTIONS
Deﬁnition 5.2.1. The discrete RV x has a geometric distribution or discrete exponential distri
bution with parameter p(0 < p < 1) if x has PMF
p
x
(α) =
_
p(1 − p)
α−1
, α = 1, 2, . . .
0, otherwise.
(5.11)
The characteristic function can be found using the sum of a geometric series (q = 1 − p):
φ
x
(t) =
p
q
∞
k=1
_
qe
j t
_
k
=
pe
j t
1 −qe
j t
. (5.12)
The mean and variance of a discrete RV with a geometric distribution can be computed using
the moment generating property of the characteristic function. The results are
η
x
=
1
p
, and σ
2
x
=
q
p
2
. (5.13)
Figure 5.3 illustrates the PMF and the characteristic function magnitude for a discrete RV with
geometric distribution and parameter p = 0.18127.
It can be shown that a discrete exponentially distributed RV has a memoryless property:
p
xx>
(αx > ) = p
x
(α −), ≥ 0. (5.14)
Deﬁnition5.2.2. The continuous RV x has an exponential distribution with parameter λ(λ > 0)
if x has PDF
f
x
(α) = λe
−λα
u(α), (5.15)
where u(·) is the unit step function.
0
α
10 20
x
p (α)
.18
.09
(a)
1
x
(t ) φ
t
0 1 2 3
(b)
FIGURE 5.3: (a) PMF and (b) characteristic function magnitude for discrete RV with geometric dis
tribution [ p = 0.18127].
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 5
0 10 5
.2
x
f (α)
.1
1
x
(t ) φ
t
10 30 50
(a) (b)
15 α
FIGURE 5.4: (a) PDF and (b) characteristic function magnitude for continuous RV with exponential
distribution and parameter λ = 0.2.
The exponential probability distribution is also a very important probability density function
in biomedical engineering applications, arising in situations involving reliability theory and
queuing problems. Reliability theory, which describes the time to failure for a system or compo
nent, grewprimarily out of military applications and experiences with multicomponent systems.
Queuing theory describes the waiting times between events.
The characteristic function can be found as
φ
x
(t) = λ
∞
_
0
e
α( j t−λ)
dα =
λ
λ − j t
. (5.16)
Figure 5.4 illustrates the PDF and the magnitude of the characteristic function for a continuous
RV with exponential distribution and parameter λ = 0.2.
The mean and variance of a continuous exponentially distributed RV can be obtained
using the moment generating property of the characteristic function. The results are
η
x
=
1
λ
, σ
2
x
=
1
λ
2
. (5.17)
A continuous exponentially distributed RV, like its discrete counterpart, satisﬁes a memoryless
property:
f
xx>τ
(αx > τ) = f
x
(α −τ), τ ≥ 0. (5.18)
Example 5.2.1. Suppose a system contains a component that has an exponential failure rate. Reli
ability engineers determined its reliability at 5000 hours to be 95%. Determine the number of hours
reliable at 99%.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
6 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. First, the parameter λ is determined from
0.95 = P(x > 5000) =
∞
_
5000
λe
−λα
dα = e
−5000λ
.
Thus
λ =
−ln(0.95)
5000
= 1.03 ×10
−5
.
Then, to determine the number of hours reliable at 99%, we solve for α from
P(x > α) = e
−λα
= 0.99
or
α =
−ln(0.99)
λ
= 980 hours.
Drill Problem 5.2.1. Suppose a system has an exponential failure rate in years to failure with
λ = 0.02. Determine the number of years reliable at: (a) 90%, (b) 95%, (c) 99%.
Answers: 0.5, 2.6, 5.3.
Drill Problem 5.2.2. Random variable x, representing the length of time in hours to complete an
examination in Introduction to Random Processes, has PDF
f
x
(α) =
4
3
e
−
4
3
α
u(α).
The examination results are given by
g(x) =
⎧
⎪
⎨
⎪
⎩
75, 0 < x < 4/3
75 +39.44(x −4/3), x ≥ 4/3
0, otherwise.
Determine the average examination grade.
Answer: 80.
5.3 BERNOULLI TRIALS
A Bernoulli experiment consists of a number of repeated (independent) trials with only two
possible events for each trial. The events for each trial can be thought of as any two events which
partition the sample space, such as a head and a tail in a coin toss, a zero or one in a computer
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 7
bit, or an even and odd number in a die toss. Let us call one of the events a success, the other
a failure. The Bernoulli PMF describes the probability of k successes in n trials of a Bernoulli
experiment. The ﬁrst two chapters used this PMF repeatedly in problems dealing with games
of chance and in situations where there were only two possible outcomes in any given trial.
For biomedical engineers, the Bernoulli distribution is used in infectious disease problems and
other applications. The Bernoulli distribution is also known as a Binomial distribution.
Deﬁnition 5.3.1. A discrete RV x is Bernoulli distributed if the PMF for x is
p
x
(k) =
⎧
⎪
⎨
⎪
⎩
_
n
k
_
p
k
q
n−k
, k = 0, 1, . . . , n
0, otherwise,
(5.19)
where p =probability of success and q = 1 − p.
The characteristic function can be found using the binomial theorem:
φ
x
(t) =
n
k=0
_
n
k
_
( pe
j t
)
k
q
n−k
= (q + pe
j t
)
n
. (5.20)
Figure 5.5 illustrates the PMF and the characteristic function magnitude for a discrete RV with
Bernoulli distribution, p = 0.2, and n = 30.
Using the moment generating property of characteristic functions, the mean and variance
of a Bernoulli RV can be shown to be
η
x
= np, σ
2
x
= npq. (5.21)
0
α
10 20
x
p (α)
.18
.09
(a)
1
x
(t ) φ
t
0 1 2
(b)
FIGURE5.5: (a) PMF and (b) characteristic function magnitude for discrete RV with Bernoulli distri
bution, p = 0.2 and n = 30.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
8 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Unlike the preceding distributions, a closed form expression for the Bernoulli CDF is not
easily obtained. Tables A.1–A.3 in the Appendix list values of the Bernoulli CDF for p =
0.05, 0.1, 0.15, . . . , 0.5 and n = 5, 10, 15, and 20. Let k ∈ {0, 1, . . . , n −1} and deﬁne
G(n, k, p) =
k
=0
_
n
_
p
(1 − p)
n−
.
Making the change of variable m = n − yields
G(n, k, p) =
n
m=n−k
_
n
n −m
_
p
n−m
(1 − p)
m
.
Now, since
_
n
n −m
_
=
n!
m! (n −m)!
=
_
n
m
_
,
G(n, k, p) =
n
m=0
_
n
m
_
p
n−m
(1 − p)
m
−
n−k−1
m=0
_
n
m
_
p
n−m
(1 − p)
m
.
Using the Binomial Theorem,
G(n, k, p) = 1 − G(n, n −k −1, 1 − p). (5.22)
This result is easily applied to obtain values of the Bernoulli CDF for values of p > 0.5 from
Tables A.1–A.3.
Example 5.3.1. The probability that Fargo Polytechnic Institute wins a game is 0.7. In a 15 game
season, what is the probability that they win: (a) at least 10 games, (b) from 9 to 12 games, (c) exactly
11 games? (d) With x denoting the number of games won, ﬁnd η
x
and σ
2
x
.
Solution. With x a Bernoulli randomvariable, we consult Table A.2, using (5.22) with n = 15,
k = 9, and p = 0.7, we ﬁnd
a) P(x ≥ 10) = 1 − F
x
(9) = 1.0 −0.2784 = 0.7216,
b) P(9 ≤ x ≤ 12) = F
x
(12) − F
x
(8) = 0.8732 −0.1311 = 0.7421,
c) p
x
(11) = F
x
(11) − F
x
(10) = 0.7031 −0.4845 = 0.2186.
d) η
x
= np = 10.5, σ
2
x
= np(1 − p) = 3.15.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 9
We now consider the number of trials needed for k successes in a sequence of Bernoulli trials.
Let
p(k, n) = P(k successes in n trials) (5.23)
=
⎧
⎪
⎨
⎪
⎩
_
n
k
_
p
k
q
n−k
, k = 0, 1, . . . , n
0, otherwise,
where p = p(1, 1) and q = 1 − p. Let RV n
r
represent the number of trials to obtain exactly
r successes (r ≥ 1). Note that
P(success in th trial r −1 successes in previous −1 trials) = p; (5.24)
hence, for = r, r +1, . . . , we have
P(n
r
= ) = p(r −1, −1) p. (5.25)
Discrete RV n
r
thus has PMF
p
n
r
() =
⎧
⎪
⎨
⎪
⎩
_
−1
r −1
_
p
r
q
−r
, = r, r +1, . . .
0, otherwise,
(5.26)
where the parameter r is a positive integer. The PMF for the RV n
r
is called the negative
binomial distribution, also known as the P´ olya and the Pascal distribution. Note that with
r = 1 the negative binomial PMF is the geometric PMF.
The moment generating function for n
r
can be expressed as
M
n
r
(λ) =
∞
=r
( −1)( −2) · · · ( −r +1)
(r −1)!
p
r
q
−r
e
λ
.
Letting m = −r , we obtain
M
n
r
(λ) =
e
λr
p
r
(r −1)!
∞
m=0
(m +r −1)(m +r −2) · · · (m +1)(qe
λ
)
m
.
With
s (x) =
∞
k=0
x
k
=
1
1 − x
, x < 1,
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
10 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
we have
s
()
(x) =
∞
k=
k(k −1) · · · (k − +1)x
k−
=
∞
m=0
(m +)(m + −1) · · · (m +1)x
m
=
!
(1 − x)
+1
.
Hence
M
n
r
(λ) =
_
pe
λ
1 −qe
λ
_
r
, qe
λ
< 1. (5.27)
The mean and variance for n
r
are found to be
η
n
r
=
r
p
, and σ
2
n
r
=
r q
p
2
. (5.28)
We note that the characteristic function is simply
φ
n
r
(t) = M
n
r
( j t) = φ
r
x
(t), (5.29)
where RV x has a discrete geometric distribution. Figure 5.6 illustrates the PMF and the
characteristic function magnitude for a discrete RV with negative binomial distribution, r = 3,
and p = 0.18127.
1
x
(t ) φ
t
0 1 2
(b)
0
α
20 40
x
p (α)
.06
.03
(a)
FIGURE5.6: (a) PMFand (b) magnitude characteristic function for discrete RVwith negative binomial
distribution, r = 3, and p = 0.18127.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 11
5.3.1 Poisson Approximation to Bernoulli
When n becomes large in the Bernoulli PMF in such a way that np = λ = constant, the
Bernoulli PMF approaches another important PMF known as the Poisson PMF. The Poisson
PMF is treated in the following section.
Lemma 5.3.1. We have
p(k) = lim
n→∞,np=λ
_
n
k
_
p
k
q
n−k
=
⎧
⎨
⎩
λ
k
e
−λ
k!
, k = 0, 1, . . .
0, otherwise,
(5.30)
Proof. Substituting p =
λ
n
and q = 1 −
λ
n
,
p(k) = lim
n→∞
1
k!
_
λ
k
_
k _
1 −
λ
n
_
n−k k−1
i =0
(n −i ).
Note that
lim
n→∞
n
−k
_
1 −
λ
n
_
−k k−1
i =0
(n −i ) = 1,
so that
p(k) = lim
n→∞
λ
k
k!
_
1 −
λ
n
_
n
.
Now,
lim
n→∞
ln
_
1 −
λ
n
_
n
= lim
n→∞
ln
_
1 −
λ
n
_
1
n
= −λ
so that
lim
n→∞
_
1 −
λ
n
_
n
= e
−λ
,
from which the desired result follows.
We note that the limiting value p(k) may be used as an approximation for the Bernoulli
PMF when p is small by substituting λ = np. While there are no prescribed rules regarding the
values of n and p for this approximation, the larger the value of n and the smaller the value of p,
the better the approximation. Satisfactory results are obtained with np < 10. The motivation
for using this approximation is that when n is large, Tables A.1–A.3 are useless for ﬁnding
values for the Bernoulli CDF.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
12 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Example 5.3.2. Suppose x is a Bernoulli random variable with n = 5000 and p = 0.001. Find
P(x ≤ 5).
Solution. Our solution involves approximating the Bernoulli PMF with the Poisson PMF
since n is quite large (and the Bernoulli CDF table is useless), and p is very close to zero.
Since λ = np = 5, we ﬁnd from Table A.5 (the Poisson CDF table is covered in Section 4) that
P(x ≤ 5) = 0.6160.
Incidentally, if p is close to one, we can still use this approximation by reversing our deﬁnition of
success and failure in the Bernoulli experiment, which results in a value of p close to zero—see
(5.22).
5.3.2 Gaussian Approximation to Bernoulli
Previously, the PoissonPMFwas usedtoapproximate a Bernoulli PMFunder certainconditions,
that is, when n is large, p is small and np < 10. This approximation is quite useful since the
Bernoulli table lists only CDF values for n up to 20. The Gaussian PDF (see Section 5.5)
is also used to approximate a Bernoulli PMF under certain conditions. The accuracy of this
approximation is best when n is large, p is close to 1/2, and npq > 3. Notice that in some
circumstances np < 10 and npq > 3. Then either the Poisson or the Gaussian approximation
will yield good results.
Lemma 5.3.2. Let
y =
x −np
√
npq
, (5.31)
where x is a Bernoulli RV. Then the characteristic function for y satisﬁes
φ(t) = lim
n→∞
φ
y
(t) = e
−t
2
/2
. (5.32)
Proof. We have
φ
y
(t) = exp
_
−j
np
√
npq
t
_
φ
x
_
t
√
npq
_
.
Substituting for φ
x
(t),
φ
y
(t) = exp
_
−j
_
np
q
t
__
q + p exp
_
j
t
√
npq
__
n
.
Simplifying,
φ
y
(t) =
_
q exp
_
−j t
_
p
qn
_
+ p exp
_
j t
_
q
np
__
n
.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 13
Letting
β =
_
q
p
, and α =
_
1
n
,
we obtain
lim
n→∞
ln φ
y
(t) = lim
α→0
ln( pβ
2
e
−j tα/β
+ pe
j tβα
)
α
2
.
Applying L’Hˆ ospital’s Rule twice,
lim
α→0
ln φ
y
(t) = lim
α→0
−j tpβe
−j tα/β
+ j tpβe
j tβα
2α
=
−t
2
p −t
2
β
2
p
2
= −
t
2
2
.
Consequently,
lim
n→∞
φ
y
(t) = exp
_
lim
n→∞
ln φ
y
(t)
_
= e
−t
2
/2
.
The limiting φ(t) in the above lemma is the characteristic function for a Gaussian RV
with zero mean and unit variance. Hence, for large n and a < b
P(a < x < b) = P(a
< y < b
) ≈ F(b
) − F(a
), (5.33)
where
F(γ ) =
1
√
2π
γ
_
−∞
e
−α
2
/2
dα = 1 − Q(γ ) (5.34)
is the standard Gaussian CDF,
a
=
a −np
√
npq
, b
=
b −np
√
npq
, (5.35)
and Q(·) is Marcum’s Q function which is tabulated in Tables A.8 and A.9 of the Appendix.
Evaluation of the above integral as well as the Gaussian PDF are treated in Section 5.5.
Example 5.3.3. Suppose x is a Bernoulli random variable with n = 5000 and p = 0.4. Find
P(x ≤ 2048).
Solution. The solution involves approximating the Bernoulli CDF with the Gaussian CDF
since npq = 1200 > 3. With np = 2000, npq = 1200 and b
= (2048 −2000)/34.641 =
1.39, we ﬁnd from Table A.8 that
P(x ≤ 2048) ≈ F(1.39) = 1 − Q(1.39) = 0.91774.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
14 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
When approximating the Bernoulli CDF with the Gaussian CDF, a continuous distribution is
used to calculate probabilities for a discrete RV. It is important to note that while the approxi
mation is excellent in terms of the CDFs—the PDF of any discrete RV is never approximated
with a continuous PDF. Operationally, to compute the probability that a Bernoulli RV takes an
integer value using the Gaussian approximation we must round off to the nearest integer.
Example 5.3.4. Suppose x is a Bernoulli random variable with n = 20 and p = 0.5. Find
P(x = 8).
Solution. Since npq = 5 > 3, the Gaussian approximation is used to evaluate the Bernoulli
PMF, p
x
(8). With np = 10, npq = 5, a
= (7.5 −10)/
√
5 = −1.12, and b
= (8.5 −10)/
√
5 = −0.67, we have
p
x
(8) = P(7.5 < x < 8.5) ≈ F(−0.67) − F(−1.12) = 0.25143 −0.13136;
hence, p
x
(8) ≈ 0.12007. From the Bernoulli table, p
x
(8) = 0.1201, which is very close to the
above approximation.
Drill Problem 5.3.1. A survey of residents in Fargo, North Dakota revealed that 30% preferred a
white automobile over all other colors. Determine the probability that: (a) exactly ﬁve of the next 20
cars purchased will be white, (b) at least ﬁve of the next twenty cars purchased will be white, (c) from
two to ﬁve of the next twenty cars purchased will be white.
Answers: 0.1789, 0.4088, 0.7625.
Drill Problem 5.3.2. Prof. Rensselaer is an avid albeit inaccurate marksman. The probability she
will hit the target is only 0.3. Determine: (a) the expected number of hits scored in 15 shots, (b) the
standard deviation for 15 shots, (c) the number of times she must ﬁre so that the probability of hitting
the target at least once is greater than 1/2.
Answers: 2, 4.5, 1.7748.
5.4 POISSONDISTRIBUTION
A Poisson PMF describes the number of successes occurring on a continuous line, typically a
time interval, or within a given region. For example, a Poisson random variable might represent
the number of telephone calls per hour, or the number of errors per page in this textbook.
In the previous section, we found that the limit (as n →∞and constant mean np) of a
Bernoulli PMF is a Poisson PMF. In this section, we derive the Poisson probability distribution
from two fundamental assumptions about the phenomenon based on physical characteristics.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 15
The following development makes use of the order notation o(h) to denote any function
g(h) which satisﬁes
lim
h→0
g(h)
h
= 0. (5.36)
For example, g(h) = 15h
2
+7h
3
= o(h).
We use the notation
p(k, τ) = P(k successes in interval [0, τ]). (5.37)
The Poisson probability distribution is characterized by the following two properties:
(1) The number of successes occurring in a time interval or region is independent of the
number of successes occurring in any other nonoverlapping time interval or region. Thus, with
A = {k successes in interval I
1
}, (5.38)
and
B = { successes in interval I
2
}, (5.39)
we have
P(A ∩ B) = P(A)P(B), if I
1
∩ I
2
= . (5.40)
As we will see, the number of successes depends only on the length of the time interval
and not the location of the interval on the time axis.
(2) The probability of a single success during a very small time interval is proportional to
the length of the interval. The longer the interval, the greater the probability of success. The
probability of more than one success occurring during an interval vanishes as the length of the
interval approaches zero. Hence
p(1, h) = λh +o(h), (5.41)
and
p(0, h) = 1 −λh +o(h). (5.42)
This second property indicates that for a series of very small intervals, the Poisson process is
composed of a series of Bernoulli trials, each with a probability of success p = λh +o(h).
Since [0, τ +h] = [0, τ] ∪ (τ, τ +h] and [0, τ] ∩ (τ, τ +h] = , we have
p(0, τ +h) = p(0, τ) p(0, h) = p(0, τ)(1 −λh +o(h)).
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
16 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Noting that
p(0, τ +h) − p(0, τ)
h
=
−λhp(0, τ) +o(h)
h
and taking the limit as h →0,
dp(0, τ)
dτ
= −λp(0, τ), p(0, 0) = 1. (5.43)
This differential equation has solution
p(0, τ) = e
−λτ
u(τ). (5.44)
Applying the above properties, it is readily seen that
p(k, τ +h) = p(k −1, τ) p(1, h) + p(k, τ) p(0, h) +o(h),
or
p(k, τ +h) = p(k −1, τ)λh + p(k, τ)(1 −λh) +o(h),
so that
p(k, τ +h) − p(k, τ)
h
+λp(k, τ) = λp(k −1, τ) +
o(h)
h
.
Taking the limit as h →0
dp(k, τ)
dτ
+λp(k, τ) = λp(k −1, τ), k = 1, 2, . . . , (5.45)
with p(k, 0) = 0. It can be shown ([7, 8]) that
p(k, τ) = λe
−λτ
τ
_
0
e
λt
p(k −1, t)dt (5.46)
and hence that
p(k, τ) =
(λτ)
k
e
−λτ
k!
u(τ), k = 0, 1, . . . . (5.47)
The RV x = number of successes thus has a Poisson distribution with parameter λτ and PMF
p
x
(k) = p(k, τ). The rate of the Poisson process is λ and the interval length is τ.
For ease in subsequent development, we replace the parameter λτ with λ. The characteris
tic function for a Poisson RV x with parameter λ is found as (with parameter λ, p
x
(k) = p(k, 1))
φ
x
(t) = e
−λ
∞
k=0
(λe
j t
)
k
k!
= e
−λ
exp(λe
j t
) = exp(λ(e
j t
−1)). (5.48)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 17
0 10 20
x
p (α)
.12
.06
(a)
1
x
(t )
t
0 1 2
(b)
φ
α
FIGURE 5.7: (a) PMF and (b) magnitude characteristic function for Poisson distributed RV with
parameter λ = 10.
Figure 5.7 illustrates the PMF and characteristic function magnitude for a discrete RV with
Poisson distribution and parameter λ = 10.
It is of interest to note that if x
1
and x
2
are independent Poisson RVs with parameters λ
1
and λ
2
, respectively, then
φ
x
1
+x
2
(t) = exp((λ
1
+λ
2
)(e
j t
−1)); (5.49)
i.e., x
1
+ x
2
is also a Poisson with parameter λ
1
+λ
2
.
The moments of a Poisson RV are tedious to compute using techniques we have seen so
far. Consider the function
ψ
x
(γ ) = E(γ
x
) (5.50)
and note that
ψ
(k)
x
(γ ) = E
_
γ
x−k
k−1
i =0
(x −i )
_
,
so that
E
_
k−1
i =0
(x −i )
_
= ψ
(k)
x
(1). (5.51)
If x is Poisson distributed with parameter λ, then
ψ
x
(γ ) = e
λ(γ −1)
, (5.52)
so that
ψ
(k)
x
(γ ) = λ
k
e
λ(γ −1)
;
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
18 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
hence,
E
_
k−1
i =0
(x −i )
_
= λ
k
. (5.53)
In particular, E(x) = λ, E(x(x −1)) = λ
2
= E(x
2
) −λ, so that σ
2
x
= λ
2
+λ −λ
2
= λ.
While it is quite easy to calculate the value of the Poisson PMF for a particular number
of successes, hand computation of the CDF is quite tedious. Therefore, the Poisson CDF is
tabulated in Tables A.4A.7 of the Appendix for selected values of λ ranging from 0.1 to 18.
Fromthe Poisson CDFtable, we note that the value of the Poisson PMFincreases as the number
of successes k increases from zero to the mean, and then decreases in value as k increases from
the mean. Additionally, note that the table is written with a ﬁnite number of entries for each
value of λ because the PMF values are written with six decimal place accuracy, even though an
inﬁnite number of Poisson successes are theoretically possible.
Example 5.4.1. On the average, Professor Rensselaer grades 10 problems per day. What is the
probability that on a given day (a) 8 problems are graded, (b) 8–10 problems are graded, and (c) at
least 15 problems are graded?
Solution. With x a Poisson random variable, we consult the Poisson CDF table with λ = 10,
and ﬁnd
a) p
x
(8) = F
x
(8) − F
x
(7) = 0.3328 −0.2202 = 0.1126,
b) P(8 ≤ x ≤ 10) = F
x
(10) − F
x
(7) = 0.5830 −0.2202 = 0.3628,
c) P(x ≥ 15) = 1 − F
x
(14) = 1 −0.9165 = 0.0835.
5.4.1 Interarrival Times
In many instances, the length of time between successes, known as an interarrival time, of a
Poisson random variable is more important than the actual number of successes. For example,
in evaluating the reliability of a medical device, the time to failure is far more signiﬁcant to the
biomedical engineer than the fact that the device failed. Indeed, the subject of reliability theory
is so important that entire textbooks are devoted to the topic. Here, however, we will brieﬂy
examine the subject of interarrival times from the basis of the Poisson PMF.
Let RV t
r
denote the length of the time interval from zero to the r th success. Then
p(τ −h < t
r
≤ τ) = p(r −1, τ −h) p(1, h)
= p(r −1, τ −h)λh +o(h)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 19
so that
F
t
r
(τ) − F
t
r
(τ −h)
h
= λp(r −1, τ −h) +
o(h)
h
.
Taking the limit as h →0 we ﬁnd that the PDF for the r th order interarrival time, that is, the
time interval from any starting point to the r th success after it, is
f
t
r
(τ) =
λ
r
τ
r −1
e
−λτ
(r −1)!
u(τ), r = 1, 2, . . . . (5.54)
This PDF is known as the Erlang PDF. Clearly, with r = 1, we have the exponential PDF:
f
t
(τ) = λe
−λτ
u(τ). (5.55)
The RV t is called the ﬁrstorder interarrival time.
The Erlang PDF is a special case of the gamma PDF:
f
x
(α) =
λ
r
α
r −1
e
−λα
(r )
u(α), (5.56)
for any real r > 0, λ > 0, where is the gamma function
(r ) =
∞
_
0
α
r −1
e
−α
dα. (5.57)
Straightforward integration reveals that (1) = 1 and (r +1) = r (r ) so that if r is a positive
integer then (r ) = (r −1)!—for this reason the gamma function is often called the factorial
function. Using the above deﬁnition for (r ), it is easily shown that the moment generating
function for a gammadistributed RV is
M
x
(η) =
_
λ
λ −η
_
r
, for η < λ. (5.58)
The characteristic function is thus
φ
x
(t) =
_
λ
λ − j t
_
r
. (5.59)
It follows that the mean and variance are
η
x
=
r
λ
, and σ
2
x
=
r
λ
2
. (5.60)
Figure 5.8 illustrates the PDF and magnitude of the characteristic function for a RV with
gamma distribution with r = 3 and λ = 0.2.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
20 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
0 20 40
x
f (α)
.06
.03
(a)
1
x
(t )
t
0 1 2
(b)
φ
α
FIGURE5.8: (a) PDF and (b) magnitude characteristic function for gamma distributed RV with r = 3
and parameter λ = 0.2.
Drill Problem 5.4.1. On the average, Professor S. Rensselaer makes ﬁve blunders per lecture.
Determine the probability that she makes (a) less than six blunders in the next lecture: (b) exactly ﬁve
blunders in the next lecture: (c) from three to seven blunders in the next lecture: (d) zero blunders in
the next lecture.
Answers: 0.6160, 0.0067, 0.7419, 0.1755.
Drill Problem 5.4.2. A process yields 0.001% defective items. If one million items are produced,
determine the probability that the number of defective items exceeds twelve.
Answer: 0.2084.
Drill Problem5.4.3. Professor S. Rensselaer designs her examinations so that the probability of at
least one extremely difﬁcult problem is 0.632. Determine the average number of extremely difﬁcult
problems on a Rensselaer examination.
Answer: 1.
5.5 UNIVARIATEGAUSSIANDISTRIBUTION
The Gaussian PDF is the most important probability distribution in the ﬁeld of biomedical
engineering. Plentiful applications arise in industry, research, and nature, ranging from instru
mentation errors to scores on examinations. The PDF is named in honor of Gauss (1777–1855),
who derived the equation based on an error study involving repeated measurements of the same
quantity. However, De Moivre is ﬁrst credited with describing the PDF in 1733. Applications
also abound in other areas outside of biomedical engineering since the distribution ﬁts the
observed data in many processes. Incidentally, statisticians refer to the Gaussian PDF as the
normal PDF.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 21
Deﬁnition 5.5.1. A continuous RV z is a standardized Gaussian RV if the PDF is
f
z
(α) =
1
√
2π
e
−
1
2
α
2
. (5.61)
The moment generating function for a standardized Gaussian RV can be found as follows:
M
z
(λ) =
1
√
2π
∞
_
−∞
e
λα−
1
2
α
2
dα
=
1
√
2π
∞
_
−∞
e
−
1
2
((α−λ)
2
−λ
2
)
dα.
Making the change of variable β = α −λ we ﬁnd
M
z
(λ) = e
1
2
λ
2
∞
_
−∞
f
z
(β)dβ = e
1
2
λ
2
, (5.62)
for all real λ. We have made use of the fact that the function f
z
is a bona ﬁde PDF, as treated
in Problem 42. Using the Taylor series expansion for an exponential,
e
1
2
λ
2
=
∞
k=0
λ
2k
2
k
k!
=
∞
n=0
M
(n)
x
(0)λ
n
n!
,
so that all moments of z exist and
E(z
2k
) =
(2k)!
2
k
k!
, k = 0, 1, 2, . . . , (5.63)
and
E(z
2k+1
) = 0, k = 0, 1, 2, . . . . (5.64)
Consequently, a standardized Gaussian RV has zero mean and unit variance. Extending the
range of deﬁnition of M
z
(λ) to include the ﬁnite complex plane, we ﬁnd that the characteristic
function is
φ
z
(t) = M
z
( j t) = e
−
1
2
t
2
. (5.65)
Letting the RV x = σz +η we ﬁnd that E(x) = η and σ
2
x
= σ
2
. For σ > 0
F
x
(α) = P(σz +η ≤ α) = F
z
((α −η)/σ),
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
22 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
so that x has the general Gaussian PDF
f
x
(α) =
1
√
2πσ
2
exp
_
−
1
2σ
2
(α −η)
2
_
. (5.66)
Similarly, with σ > 0 and x = −σz +η we ﬁnd
F
x
(α) = P(−σz +η ≤ α) = 1 − F
z
((η −α)/σ),
so that f
x
is as above. We will have occasion to use the shorthand notation x ∼ G(η, σ
2
) to
denote that the RVhas a Gaussian PDFwith mean η and variance σ
2
. Note that if x ∼ G(η, σ
2
)
then (x = σz +η)
φ
x
(t) = e
j ηt
e
−
1
2
σ
2
t
2
. (5.67)
The Gaussian PDF, illustrated with η = 75 and σ
2
= 25, as well as with η = 75 and σ
2
= 9
in Figure 5.9, is a bellshaped curve completely determined by its mean and variance. As can
be seen, the Gaussian PDF is symmetrical about the vertical axis through the expected value.
If, in fact, η = 25, identically shaped curves could be drawn, centered now at 25 instead of
75. Additionally, the maximum value of the Gaussian PDF, (2πσ
2
)
−1/2
, occurs at α = η. The
PDF approaches zero asymptotically as α approaches ±∞. Naturally, the larger the value of the
variance, the more spread in the distribution and the smaller the maximum value of the PDF.
For any combination of the mean and variance, the Gaussian PDF curve must be symmetrical
as previously described, and the area under the curve must equal one.
Unfortunately, a closed form expression does not exist for the Gaussian CDF, which
necessitates numerical integration. Rather than attempting to tabulate the general Gaussian
CDF, a normalization is performed to obtain a standardized Gaussian RV (with zero mean
α
x
f (α)
.12
65 75 85
.08
.04
FIGURE5.9: Gaussian probability density function for η = 75 and σ
2
= 9, 25.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 23
and unit variance). If x ∼ G(η, σ
2
), the RV z = (x −η)/σ is a standardized Gaussian RV:
z ∼ G(0, 1). This transformation is always applied when using standard tables for computing
probabilities for Gaussian RVs. The probability P(α
1
< x ≤ α
2
) can be obtained as
P(α
1
< x ≤ α
2
) = F
x
(α
2
) − F
x
(α
1
), (5.68)
using the fact that
F
x
(α) = F
z
((α −η)/σ). (5.69)
Note that
F
z
(α) =
1
√
2π
α
_
−∞
e
−
1
2
τ
2
dτ = 1 − Q(α), (5.70)
where Q(·) is Marcum’s Q function:
Q(α) =
1
√
2π
∞
_
α
e
−
1
2
τ
2
dτ. (5.71)
Marcum’s Qfunction is tabulated in Tables A.8 and A.9 for 0 ≤ α < 4 using the approximation
presented in Section 5.5.1. It is easy to show that
Q(−α) = 1 − Q(α) = F
z
(α). (5.72)
The error and complementary error functions, deﬁned by
erf(α) =
2
π
α
_
0
e
−t
2
dt (5.73)
and
erfc(α) =
2
π
∞
_
α
e
−t
2
dt = 1 −erf(α) (5.74)
are also often used to evaluate the standard normal integral. A simple change of variable reveals
that
erfc(α) = 2Q(α/
√
2). (5.75)
Example 5.5.1. Compute F
z
(−1.74), where z ∼ G(0, 1).
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
24 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. To compute F
z
(−1.74), we ﬁnd
F
z
(−1.74) = 1 − Q(−1.74) = Q(1.74) = 0.04093,
using (5.72) and Table A.8.
While the value a Gaussian randomvariable takes on is any real number between negative
inﬁnity and positive inﬁnity, the realistic range of values is much smaller. From Table A.9,
we note that 99.73% of the area under the curve is contained between −3.0 and 3.0. From
the transformation z = (x −η)/σ, the range of values random variable x takes on is then
approximately η ±3σ. This notion does not imply that randomvariable x cannot take on a value
outside this interval, but the probability of it occurring is really very small (2Q(3) = 0.0027).
Example 5.5.2. Suppose x is a Gaussian random variable with η = 35 and σ = 10. Sketch the
PDF and then ﬁnd P(37 ≤ x ≤ 51). Indicate this probability on the sketch.
Solution. The PDF is essentially zero outside the interval [η −3σ, η +3σ] = [5, 65]. The
sketch of this PDF is shown in Figure 5.10 along with the indicated probability. With
z =
x −35
10
we have
P(37 ≤ x ≤ 51) = P(0.2 ≤ z ≤ 1.6) = F
z
(1.6) − F
z
(0.2).
Hence P(37 ≤ x ≤ 51) = Q(0.2) − Q(1.6) = 0.36594 from Table A.9.
x
f (α)
.03
30 60
.02
.01
.04
0
α
FIGURE5.10: PDF for Example 5.5.2.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 25
Example 5.5.3. A machine makes capacitors with a mean value of 25 μF and a standard deviation
of 6 μF. Assuming that capacitance follows a Gaussian distribution, ﬁnd the probability that the value
of capacitance exceeds 31 μF if capacitance is measured to the nearest μF.
Solution. Let the RV x denote the value of a capacitor. Since we are measuring to the nearest
μF, the probability that the measured value exceeds 31 μF is
P(31.5 ≤ x) = P(1.083 ≤ z) = Q(1.083) = 0.13941,
where z = (x −25)/6 ∼ G(0, 1). This result is determined by linear interpolation of the CDF
between equal 1.08 and 1.09.
5.5.1 Marcum’s QFunction
Marcum’s Q function, deﬁned by
Q(γ ) =
1
√
2π
∞
_
γ
e
−
1
2
α
2
dα (5.76)
has been extensively studied. If the RV z ∼ G(0, 1) then
Q(γ ) = 1 − F
z
(γ ); (5.77)
i.e., Q(γ ) is the complement of the standard Gaussian CDF. Note that Q(0) = 0.5, Q(∞) = 0,
and that F
z
(−γ ) = Q(γ ). A very accurate approximation to Q(γ ) is presented in [1, p. 932]:
Q(γ ) ≈ e
−
1
2
γ
2
h(t), γ > 0, (5.78)
where
t =
1
1 +0.2316419γ
, (5.79)
and
h(t) =
1
√
2π
t(a
1
+t(a
2
+t(a
3
+t(a
4
+a
5
t)))). (5.80)
The constants are
i a
i
1 0.31938153
2 −0.356563782
3 1.781477937
4 −1.821255978
5 1.330274429
The error in using this approximation is less than 7.5 ×10
−8
.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
26 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
A very useful bound for Q(α) is [1, p. 298]
_
2
π
e
−
1
2
α
2
α +
√
α
2
+4
< Q(α) ≤
_
2
π
e
−
1
2
α
2
α +
√
α
2
+0.5π
. (5.81)
The ratio of the upper bound to the lower bound is 0.946 when α = 3 and 0.967 when α = 4.
The bound improves as α increases.
Sometimes, it is desired to ﬁnd the value of γ for which Q(γ ) = q. Helstrom [14] offers
an iterative procedure which begins with an initial guess γ
0
> 0. Then compute
t
i
=
1
1 +0.2316419γ
i
(5.82)
and
γ
i +1
=
_
2 ln
_
h(t
i
)
q
__
1/2
, i = 0, 1, . . . . (5.83)
The procedure is terminated when γ
i +1
≈ γ
i
to the desired degree of accuracy.
Drill Problem5.5.1. Students attend Fargo Polytechnic Institute for an average of four years with a
standard deviation of onehalf year. Let the randomvariable x denote the length of attendance and as
sume that x is Gaussian. Determine: (a) P(1 < x < 3), (b)P(x > 4), (c )P(x = 4), (d)F
x
(4.721).
Answers: 0.5, 0, 0.02275, 0.92535.
Drill Problem 5.5.2. The quality point averages of 2500 freshmen at Fargo Polytechnic Institute
follow a Gaussian distribution with a mean of 2.5 and a standard deviation of 0.7. Suppose grade
point averages are computed to the nearest tenth. Determine the number of freshmen you would expect
to score: (a) from 2.6 to 3.0, (b) less than 2.5, (c) between 3.0 and 3.5, (d) greater than 3.5.
Answers: 167, 322, 639, 1179.
Drill Problem5.5.3. Professor Rensselaer loves the game of golf. She has determined that the distance
the ball travels on her ﬁrst shot follows a Gaussian distribution with a mean of 150 and a standard
deviation of 17. Determine the value of d so that the range, 150 ±d, covers 95% of the shots.
Answer: 33.32.
5.6 BIVARIATEGAUSSIANRANDOMVARIABLES
The previous section introduced the univariate Gaussian PDF along with some general char
acteristics. Now, we discuss the joint Gaussian PDF and its characteristics by drawing on our
univariate Gaussian PDF experiences, and signiﬁcantly expanding the scope of applications.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 27
Numerous applications of this joint PDF are found throughout the ﬁeld of biomedical engi
neering and, like the univariate case, the joint Gaussian PDF is considered the most important
joint distribution for biomedical engineers.
Deﬁnition 5.6.1. The bivariate RV z = (x, y) is a bivariate Gaussian RV if every linear combi
nation of x and y has a univariate Gaussian distribution. In this case we also say that the RVs x and
y are jointly distributed Gaussian RVs.
Let the RV w = ax +by, and let x and y be jointly distributed Gaussian RVs. Then
w is a univariate Gaussian RV for all real constants a and b. In particular, x ∼ G(η
x
, σ
2
x
) and
y ∼ G(η
y
, σ
2
y
); i.e., the marginal PDFs for a joint Gaussian PDF are univariate Gaussian. The
above deﬁnition of a bivariate Gaussian RV is sufﬁcient for determining the bivariate PDF,
which we now proceed to do.
The following development is signiﬁcantly simpliﬁed by considering the standardized
versions of x and y. Also, we assume that ρ
x,y
 < 1, σ
x
= 0, and σ
y
= 0. Let
z
1
=
x −η
x
σ
x
and z
2
=
y −η
y
σ
y
, (5.84)
so that z
1
∼ G(0, 1) and z
2
∼ G(0, 1). Below, we ﬁrst ﬁnd the joint characteristic function
for the standardized RVs z
1
and z
2
, then the conditional PDF f
z
2
z
1
and the joint PDF f
z
1
,z
2
.
Next, the results for z
1
and z
2
are applied to obtain corresponding quantities φ
x,y
, f
yx
and f
x,y
.
Finally, the special cases ρ
x,y
= ±1, σ
x
= 0, and σ
y
= 0 are discussed.
Since z
1
and z
2
are jointly Gaussian, the RV t
1
z
1
+t
2
z
2
is univariate Gaussian:
t
1
z
1
+t
2
z
2
∼ G
_
0, t
2
1
+2t
1
t
2
ρ +t
2
2
_
.
Completing the square,
t
2
1
+2t
1
t
2
ρ +t
2
2
= (t
1
+ρt
2
)
2
+(1 −ρ
2
)t
2
2
,
so that
φ
z
1
,z
2
(t
1
, t
2
) = E(e
j t
1
z
1
+j t
2
z
2
) = e
−
1
2
(1−ρ
2
)t
2
2
e
−
1
2
(t
1
+ρt
2
)
2
. (5.85)
From (6) we have
f
z
1
,z
2
(α, β) =
1
2π
∞
_
−∞
I(α, t
2
)e
−jβt
2
dt
2
, (5.86)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
28 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
where
I(α, t
2
) =
1
2π
∞
_
−∞
φ
z
1
,z
2
(t
1
, t
2
)e
−j αt
1
dt
1
.
Substituting (5.85) and letting τ = t
1
+t
2
ρ, we obtain
I(α, t
2
) = e
−
1
2
(1−ρ
2
)t
2
2
1
2π
∞
_
−∞
e
−
1
2
τ
2
e
−j α(τ−ρt
2
)
dτ,
or
I(α, t
2
) = φ(t
2
) f
z
1
(α),
where
φ(t
2
) = e
j αρt
2
e
−
1
2
(1−ρ
2
)t
2
2
.
Substituting into (5.86) we ﬁnd
f
z
1
,z
2
(α, β) = f
z
1
(α)
1
2π
∞
_
−∞
φ(t
2
)e
−jβt
2
dt
2
and recognize that φ is the characteristic function for a Gaussian RVwith mean αρ and variance
1 −ρ
2
. Thus
f
z
1
,z
2
(α, β)
f
z
1
(α)
= f
z
2
z
1
(βα) =
1
_
2π(1 −ρ
2
)
exp
_
−
(β −ρα)
2
2(1 −ρ
2
)
_
, (5.87)
so that
E(z
2
 z
1
) = ρz
1
(5.88)
and
σ
2
z
2
z
1
= 1 −ρ
2
. (5.89)
After some algebra, we ﬁnd
f
z
1
,z
2
(α, β) =
1
2π(1 −ρ
2
)
1/2
exp
_
−
α
2
−2ραβ +β
2
2(1 −ρ
2
)
_
. (5.90)
We now turn our attention to using the above results for z
1
and z
2
to obtain similar results for
x and y. From (5.84) we ﬁnd that
x = σ
x
z
1
+η
x
and y = σ
y
z
2
+η
y
,
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 29
so that the joint characteristic function for x and y is
φ
x,y
(t
1
, t
2
) = E(e
j t
1
x+j t
2
y
) = E(e
j t
1
σ
x
z
1
+j t
2
σ
y
z
2
)e
j t
1
η
x
+j t
2
η
y
.
Consequently, the joint characteristic function for x and y can be found from the joint charac
teristic function of z
1
and z
2
as
φ
x,y
(t
1
, t
2
) = φ
z
1
,z
2
(σ
x
t
1
, σ
y
t
2
)e
j η
x
t
1
e
j η
y
t
2
. (5.91)
Using (4.66), the joint characteristic function φ
x,y
can be transformed to obtain the joint PDF
f
x,y
(α, β) as
f
x,y
(α, β) =
1
(2π)
2
∞
_
−∞
∞
_
−∞
φ
z
1
,z
2
(σ
x
t
1
, σ
y
t
2
)e
−j (α−η
x
)t
1
e
−j (β−η
y
)t
2
dt
1
dt
2
. (5.92)
Making the change of variables τ
1
= σ
x
t
1
, τ
2
= σ
y
t
2
, we obtain
f
x,y
(α, β) =
1
σ
x
σ
y
f
z
1
,z
2
_
α −η
x
σ
x
,
β −η
y
σ
y
_
. (5.93)
Since
f
x,y
(α, β) = f
yx
(βα) f
x
(α)
and
f
x
(α) =
1
σ
x
f
z
1
_
α −η
x
σ
x
_
,
we may apply (5.93) to obtain
f
yx
(βα) =
1
σ
y
f
z
2
z
1
_
β −η
y
σ
y
¸
¸
¸
¸
α −η
x
σ
x
_
. (5.94)
Substituting (5.90) and (5.87) into (5.93) and (5.94) we ﬁnd
f
x,y
(α, β) =
exp
_
−
1
2(1−ρ
2
)
_
(α−η
x
)
2
σ
2
x
−
2ρ(α−η
x
)(β−η
y
)
σ
x
σ
y
+
(β−η
y
)
2
σ
2
y
__
2πσ
x
σ
y
(1 −ρ
2
)
1/2
(5.95)
and
f
yx
(βα) =
exp
_
−
_
β−η
y
−ρσ
y
α−η
x
σ
x
_
2
2(1−ρ
2
)σ
2
y
_
_
2πσ
2
y
(1 −ρ
2
)
. (5.96)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
30 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
It follows that
E(yx) = η
y
+ρσ
y
x −η
x
σ
x
(5.97)
and
σ
2
yx
= σ
2
y
(1 −ρ
2
). (5.98)
By interchanging x with y and α with β,
f
xy
(αβ) =
exp
⎛
⎝
−
_
α−η
x
−ρσ
x
β−η
y
σ
y
_
2
2(1−ρ
2
)σ
2
y
⎞
⎠
_
2πσ
2
x
(1 −ρ
2
)
, (5.99)
E(xy) = η
x
+ρσ
x
y −η
y
σ
y
, (5.100)
and
σ
2
xy
= σ
2
x
(1 −ρ
2
). (5.101)
A threedimensional plot of a bivariate Gaussian PDF is shown in Figure 5.11.
The bivariate characteristic function for x and y is easily obtained as follows. Since x and
y are jointly Gaussian, the RV t
1
x +t
2
y is a univariate Gaussian RV:
t
1
x +t
2
y ∼ G
_
t
1
η
x
+t
2
η
y
, t
2
1
σ
2
x
+2t
1
t
2
σ
x,y
+t
2
2
σ
2
y
_
.
x , y
f (α, β )
.265
α = −3
α = 3
β = −3
β = 3
FIGURE5.11: Bivariate Gaussian PDF f
x,y
(α, β) with σ
x
= σ
y
= 1, η
x
= η
y
= 0, and ρ = −0.8.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 31
Consequently, the joint characteristic function for x and y is
φ
x,y
(t
1
, t
2
) = e
−
1
2
(t
2
1
σ
2
x
+2t
1
t
2
σ
x,y
+t
2
2
σ
2
y
)
e
j t
1
η
x
+j t
2
η
y
, (5.102)
which is valid for all σ
x,y
, σ
x
and σ
y
.
We now consider some special cases of the bivariate Gaussian PDF. If ρ = 0 then (from
(5.95))
f
x,y
(α, β) = f
x
(α) f
y
(β); (5.103)
i.e., RVs x and y are independent.
As ρ →±1, from (5.97) and (5.98) we ﬁnd
E(yx) →η
y
±σ
y
x −η
x
σ
x
and σ
2
yx
→0. Hence,
y →η
y
±σ
y
x −η
x
σ
x
in probability
1
. We conclude that
f
x,y
(α, β) = f
x
(α)δ
_
β −η
y
±σ
y
α −η
x
σ
x
_
(5.104)
for ρ = ±1. Interchanging the roles of x and y we ﬁnd that the joint PDF for x and y may also
be written as
f
x,y
(α, β) = f
y
(β)δ
_
α −η
x
±σ
x
β −η
y
σ
y
_
(5.105)
when ρ = ±1. These results can also be obtained “directly” fromthe joint characteristic function
for x and y.
A very special property of jointly Gaussian RVs is presented in the following theorem.
Theorem5.6.1. The jointly Gaussian RVs x and y are independent iff ρ
x,y
= 0.
Proof. We showed previously that if x and y are independent, then ρ
x,y
= 0.
Suppose that ρ = ρ
x,y
= 0. Then f
yx
(βα) = f
y
(β).
Example 5.6.1. Let x and y be jointly Gaussian with zero means, σ
2
x
= σ
2
y
= 1, and ρ = ±1.
Find constants a and b such that v = ax +by ∼ G(0, 1) and such that v and x are independent.
1
As the variance of a RV decreases to zero, the probability that the RV deviates from its mean by more than an
arbitrarily small ﬁxed amount approaches zero. This is an application of the Chebyshev Inequality.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
32 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. We have E(v) = 0. We require
σ
2
v
= a
2
+b
2
+2abρ
2
x,y
= 1
and
E(vx) = a +bρ
x,y
= 0.
Hence a = −bρ
x,y
and b
2
= 1/(1 −ρ
2
x,y
), so that
v =
y −ρ
x,y
x
_
1 −ρ
2
x,y
is independent of x and σ
2
v
= 1.
5.6.1 Constant Contours
Returning to the normalized jointly Gaussian RVs z
1
and z
2
, we now investigate the shape of
the joint PDF f
z
1
,z
2
(α, β) by ﬁnding the locus of points where the PDF is constant. We assume
that ρ < 1. By inspection of (5.90), we ﬁnd that f
z
1
,z
2
(α, β) is constant for α and β satisfying
α
2
−2ραβ +β
2
= c
2
, (5.106)
where c is a positive constant.
If ρ = 0 the contours where the joint PDF is constant is a circle of radius c centered at
the origin.
Along the line β = qα we ﬁnd that
α
2
(1 −2ρq +q
2
) = c
2
(5.107)
so that the constant contours are parameterized by
α =
±c
_
1 −2ρq +q
2
, (5.108)
and
β =
±c q
_
1 −2ρq +q
2
. (5.109)
The square of the distance from a point (α, β) on the contour to the origin is given by
d
2
(q) = α
2
+β
2
=
c
2
(1 +q
2
)
1 −2ρq +q
2
. (5.110)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 33
Differentiating, we ﬁnd that d
2
(q) attains its extremal values at q = ±1. Thus, the line β = α
intersects the constant contour at
±β = α =
±c
_
2(1 −ρ)
. (5.111)
Similarly, the line β = −α, intersects the constant contour at
±β = −α =
±c
_
2(1 −ρ)
. (5.112)
Consider the rotated coordinates α
= (α +β)/
√
2 and β
= (β −α)/
√
2, so that
α
+β
√
2
= β (5.113)
and
α
−β
√
2
= α. (5.114)
The rotated coordinate system is a rotation by π/4 counterclockwise. Thus
α
2
−2ραβ +β
2
= c
2
(5.115)
is transformed into
α
2
1 +ρ
+
β
2
1 −ρ
=
c
2
1 −ρ
2
. (5.116)
The above equation represents an ellipse with major axis length 2c /
√
1 −ρ and minor axis
length 2c /
√
1 +ρ. In the α −β plane, the major and minor axes of the ellipse are along the
lines β = ±α.
From (5.93), the constant contours for f
x,y
(α, β) are solutions to
_
α −η
x
σ
x
_
2
−2ρ
_
α −η
x
σ
x
__
β −η
y
σ
y
_
+
_
β −η
y
σ
y
_
2
= c
2
. (5.117)
Using the transformation
α
=
1
√
2
_
α −η
x
σ
x
+
β −η
y
σ
y
_
, β
=
1
√
2
_
β −η
y
σ
y
−
α −η
x
σ
x
_
(5.118)
transforms the constant contour to (5.116). With α
= 0 we ﬁnd that one axis is along
α −η
x
σ
x
= −
β −η
y
σ
y
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
34 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
with endpoints at
α =
±c σ
x
√
2
√
1 +ρ
+η
x
, β =
±c σ
y
√
2
√
1 +ρ
+η
y
;
the length of this axis in the α −β plane is
√
2c
_
σ
2
x
+σ
2
y
1 +ρ
.
With β
= 0 we ﬁnd that the other axis is along
α −η
x
σ
x
=
β −η
y
σ
y
with endpoints at
α =
±c σ
x
√
2
√
1 −ρ
+η
x
, β =
±c σ
y
√
2
√
1 −ρ
+η
y
;
the length of this axis in the α −β plane is
√
2c
_
σ
2
x
+σ
2
y
1 −ρ
.
Points on this ellipse in the α −β plane satisfy (5.117); the value of the joint PDF f
x,y
on this
curve is
1
2πσ
x
σ
y
_
1 −ρ
2
exp
_
−
c
2
2(1 −ρ
2
)
_
. (5.119)
A further transformation
α
=
α
√
1 +ρ
, β
=
β
√
1 −ρ
transforms the ellipse in the α
−β
plane to a circle in the α
−β
plane:
α
2
+β
2
=
c
2
1 −ρ
2
.
This transformation provides a straightforward way to compute the probability that the joint
Gaussian RVs x and y lie within the region bounded by the ellipse speciﬁed by (5.117). Letting
A denote the region bounded by the ellipse in the α −β plane and A
denote the image (a circle)
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 35
in the α
−β
plane, we have
_
A
_
f
x,y
(α, β) dα dβ =
_ _
1
2πσ
x
σ
y
_
1 −ρ
2
e
−
1
2
(α
2
+β
2
)
 J (α, β)
dα
dβ
,
where the Jacobian of the transformation is
J (α, β) =
¸
¸
¸
¸
¸
¸
¸
¸
∂α
∂α
∂α
∂β
∂β
∂α
∂β
∂β
¸
¸
¸
¸
¸
¸
¸
¸
.
Computing the indicated derivatives, we have
J (α, β) =
¸
¸
¸
¸
¸
¸
¸
¸
¸
1
σ
x
√
2
√
1 +ρ
1
σ
y
√
2
√
1 +ρ
−1
σ
x
√
2
√
1 −ρ
1
σ
y
√
2
√
1 −ρ
¸
¸
¸
¸
¸
¸
¸
¸
¸
,
so that
J (α, β) =
1
σ
x
σ
y
_
1 −ρ
2
.
Substituting and transforming to polar coordinates, we ﬁnd
_
A
_
f
x,y
(α, β) dα dβ =
_
A
_
1
2π
e
−
1
2
(α
2
+β
2
)
dα
dβ
=
1
2π
2π
_
0
c /
√
1−ρ
2
_
0
r e
−
1
2
r
2
dr dθ
=
c
2
/(2−2ρ
2
)
_
0
e
−u
du
= 1 −e
−c
2
/(2−2ρ
2
)
.
This bivariate Gaussianprobability computationis one of the fewwhichis “easily” accomplished.
Additional techniques for treating these computations are given in [1, pp. 956–958].
Drill Problem5.6.1. Given that x and y are jointly distributed Gaussian random variables with
E(yx) = 2 +1.5x, E(xy) = 7/6 + y/6, and σ
2
xy
= 0.75. Determine: (a) η
x
, (b) η
y
, (c) σ
2
x
, (d)
σ
2
y
, and (e) ρ
x,y
.
Answers: 0.5, 1, 9, 2, 5.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
36 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Drill Problem 5.6.2. Random variables x and y are jointly Gaussian with η
x
= −2, η
y
= 3,
σ
2
x
= 21, σ
2
y
= 31, and ρ = −0.3394. With c
2
= 0.2212 in (154), ﬁnd: (a) the smallest angle
that either the minor or major axis makes with the positive α axis in the α −β plane, (b) the length
of the minor axis, (c) the length of the major axis.
Answers: 3, 2, 30
◦
.
Drill Problem 5.6.3. Random variables x and y are jointly Gaussian with η
x
= −2, η
y
= 3,
σ
2
x
= 21, σ
2
y
= 31, and ρ = −0.3394. Find: (a) E(y  x = 0), (b)P(1 < y ≤ 10  x = 0),
(c )P(−1 < x < 7).
Answers: 0.5212, 0.3889, 2.1753.
5.7 SUMMARY
This chapter introduces certain probability distributions commonly encountered in biomedical
engineering. Special emphasis is placed on the exponential, Poisson and Gaussian distributions.
Important approximations to the Bernoulli PMF and Gaussian CDF are developed.
Bernoulli event probabilities may be approximated by the Poisson PMF when np < 10
or by the Gaussian PDF when npq > 3. For the Poisson approximation use η = np. For the
Gaussian approximation use η = np and σ
2
= npq.
Many important properties of jointly Gaussian random variables are presented.
Drill Problem 5.7.1. The length of time William Smith plays a video game is given by random
variable x distributed exponentially with a mean of four minutes. His play during each game is
independent from all other games. Determine: (a) the probability that William is still playing after
four minutes, (b) the probability that, out of ﬁve games, he has played at least one game for more than
four minutes.
Answers: exp(−1), 0.899.
5.8 PROBLEMS
1. Assume x is a Bernoulli random variable. Determine P(x ≤ 3) using the Bernoulli
CDF table if: (a) n = 5, p = 0.1; (b) n = 10, p = 0.1; (c) n = 20, p = 0.1; (d) n = 5,
p = 0.3; (e) n = 10, p = 0.3; (f ) n = 20, p = 0.3; (g) n = 5, p = 0.6; (h) n = 10,
p = 0.6; (i) n = 20, p = 0.6.
2. Suppose you are playing a game with a friend in which you roll a die 10 times. If the
die comes up with an even number, your friend gives you a dollar and if the die comes
up with an odd number you give your friend a dollar. Unfortunately, the die is loaded
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 37
so that a 1 or a 3 are three times as likely to occur as a 2, a 4, a 5 or a 6. Determine:
(a) how many dollars your friend can expect to win in this game; (b) the probability of
your friend winning more than 4 dollars.
3. The probability that a basketball player makes a basket is 0.4. If he makes 10 attempts,
what is the probability he will make: (a) at least 4 baskets; (b) 4 baskets; (c) from 7 to
9 baskets; (d) less than 2 baskets; (e) the expected number of baskets.
4. The probability that Professor Rensselaer bowls a strike is 0.2. Determine the probability
that: (a) 3 of the next 20 rolls are strikes; (b) at least 4 of the next 20 rolls are strikes;
(c) from 3 to 7 of the next 20 rolls are strikes. (d) She is to keep rolling the ball until
she gets a strike. Determine the probability it will take more than 5 rolls. Determine
the: (e) expected number of strikes in 20 rolls; (f ) variance for the number of strikes in
20 rolls; (g) standard deviation for the number of strikes in 20 rolls.
5. The probability of a man hitting a target is 0.3. (a) If he tries 15 times to hit the target,
what is the probability of him hitting it at least 5 but less than 10 times? (b) What
is the average number of hits in 30 tries? (c) What is the probability of him getting
exactly the average number of hits in 30 tries? (d) How many times must the man try
to hit the target if he wants the probability of hitting it to be at least 2/3? (e) What is
the probability that no more than three tries are required to hit the target for the ﬁrst
time?
6. In Junior Bioinstrumentation Lab, one experiment introduces students to the transistor.
Each student is given only one transistor to use. The probability of a student destroying
a transistor is 0.7. One lab class has 5 students and they will perform this experiment
next week. Let random variable x show the possible numbers of students who destroy
transistors. (a) Sketch the PMF for x. Determine: (b) the expected number of destroyed
transistors, (c) the probability that fewer than 2 transistors are destroyed.
7. On a frosty January morning in Fargo, North Dakota, the probability that a car parked
outside will start is 0.6. (a) If we take a sample of 20 cars, what is the probability that
exactly 12 cars will start and 8 will not? (b) What is the probability that the number of
cars starting out of 20 is between 9 and 15.
8. Consider Problem 7. If there are 20,000 cars to be started, ﬁnd the probability that: (a)
at least 12,100 will start; (b) exactly 12,000 will start; (c) the number starting is between
11,900 and 12,150; (d) the number starting is less than 12,500.
9. A dart player has found that the probability of hitting the dart board in any one throw
is 0.2. How many times must he throw the dart so that the probability of hitting the
dart board is at least 0.6?
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
38 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
10. Let random variable x be Bernoulli with n = 15 and p = 0.4. Determine E(x
2
).
11. Suppose x is a Bernoulli random variable with η = 10 and σ
2
= 10/3. Determine:
(a) q, (b) n, (c) p.
12. An electronics manufacturer is evaluating its quality control program. The current
procedure is to take a sample of 5 from 1000 and pass the shipment if not more than 1
component is found defective. What proportion of 20% defective components will be
shipped?
13. Repeat Problem 1, when appropriate, using the Poisson approximation to the Bernoulli
PMF.
14. Acertain intersection averages 3 trafﬁc accidents per week. What is the probability that
more than 2 accidents will occur during any given week?
15. Suppose that on the average, a student makes 6 mistakes per test. Determine the
probability that the student makes: (a) at least 1 mistake; (b) from 3 to 5 mistakes;
(c) exactly 2 mistakes; (d) more than the expected number of mistakes.
16. On the average, Professor Rensselaer gives 11 quizzes per quarter in Introduction to
Random Processes. Determine the probability that: (a) from 8 to 12 quizzes are given
during the quarter; (b) exactly 11 quizzes are given during the quarter; (c) at least 10
quizzes are given during the quarter; (d) at most 9 quizzes are given during the quarter.
17. Suppose a typist makes an average of 30 mistakes per page. (a) If you give him a one
page letter to type, what is the probability that he makes exactly 30 mistakes? (b) The
typist decides to take typing lessons, and, after the lessons, he averages 5 mistakes per
page. You give him another one page letter to type. What is the probability of him
making fewer than 5 mistakes. (c) With the 5 mistakes per page average, what is the
probability of him making fewer than 50 mistakes in a 25 page report?
18. On the average, a sample of radioactive material emits 20 alpha particles per minute.
What is the probability of 10 alpha particles being emitted in: (a) 1 min, (b) 10 min?
(c) Many years later, the material averages 6 alpha particles emitted per minute. What
is the probability of at least 6 alpha particles being emitted in 1 min?
19. At Fargo Polytechnic Institute (FPI), a student may take a course as many times as
desired. Suppose the average number of times a student takes Introduction to Random
Processes is 1.5. (Professor Rensselaer, the course instructor, thinks so many students
repeat the course because they enjoy it so much.) (a) Determine the probability that a
student takes the course more than once. (b) The academic vicepresident of FPI wants
to ensure that on the average, 80% of the students take the course at most one time. To
what value should the mean be adjusted to ensure this?
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 39
20. Suppose 1% of the transistors in a box are defective. Determine the probability that
there are: (a) 3 defective transistors in a sample of 200 transistors; (b) more than 15
defective transistors in a sample of 1000 transistors; (c) 0 defective transistors in a
sample of 20 transistors.
21. A perfect car is assembled with a probability of 2 ×10
−5
. If 15,000 cars are produced
in a month, what is the probability that none are perfect?
22. FPI admits only 1000 freshmen per year. The probability that a student will major in
Bioengineering is 0.01. Determine the probability that fewer than 9 students major in
Bioengineering.
23. (a) Every time a carpenter pounds in a nail, the probability that he hits his thumb is
0.002. If in building a house he pounds 1250 nails, what is the probability of himhitting
his thumb at least once while working on the house? (b) If he takes ﬁve extra minutes
off every time he hits his thumb, how many extra minutes can he expect to take off in
building a house with 3000 nails?
24. The manufacturer of Leaping Lizards, a bran cereal with milk expanding (exploding)
marshmallow lizards, wants to ensure that on the average, 95% of the spoonfuls will
each have at least one lizard. Assuming that the lizards are randomly distributed in the
cereal box, to what value should the mean of the lizards per spoonful be set at to ensure
this?
25. The distribution for the number of students seeking advising help fromProfessor Rens
selaer during any particular day is given by
P(x = k) =
3
k
e
−3
k!
, k = 0, 1, . . . .
The PDF for the time interval between students seeking help for Introduction to
Random Processes from Professor Rensselaer during any particular day is given by
f
t
(τ) = e
−τ
u(τ).
If random variable z equals the total number of students Professor Rensselaer helps
each day, determine: (a) E(z), (b) σ
z
.
26. This year, on its anniversary day, a computer store is going to run an advertising cam
paign in which the employees will telephone 5840 people selected at random from the
population of North America. The caller will ask the person answering the phone
if it’s his or her birthday. If it is, then that lucky person will be mailed a brand
new programmable calculator. Otherwise, that person will get nothing. Assuming that
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
40 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
the person answering the phone won’t lie and that there is no such thing as leap year,
ﬁnd the probability that: (a) the computer store mails out exactly 16 calculators, (b) the
computer store mails out from 20 to 40 calculators.
27. Random variable x is uniform between −2 and 3. Event A = {0 < x ≤ 2} and B =
{−1 < x ≤ 0} ∪ {1 < x ≤ 2}. Find: (a) P(−1 < x < 0), (b) η
x
, (c) σ
x
, (d) f
x A
(α A),
(e) F
x A
(α A), (f ) f
xB
(αB), (g) F
xB
(αB).
28. The time it takes a runner to run a mile is equally likely to lie in an interval from 4.0 to
4.2 min. Determine: (a) the probability it takes the runner exactly 4 min to run a mile,
(b) the probability it takes the runner from 4.1 to 4.15 min.
29. Assume x is a standard Gaussian random variable. Using Tables A.9 and A.10, de
termine: (a) P(x = 0), (b) P(x < 0), (c) P(x < 0.2), (d) P(−1.583 ≤ x < 1.471),
(e) P(−2.1 < x ≤ −0.5), (f ) P(x is an integer).
30. Repeat Problem 1, when appropriate, using the Gaussian approximation to the
Bernoulli PMF.
31. A light bulb manufacturer distributes light bulbs that have a length of life that is
normally distributed with a mean equal to 1200 h and a standard deviation of 40 h.
Find the probability that a bulb burns between 1000 and 1300 h.
32. A certain type of resistor has resistance values that are Gaussian distributed with a
mean of 50 ohms and a variance of 3. (a) Write the PDF for the resistance value. (b)
Find the probability that a particular resistor is within 2 ohms of the mean. (c) Find
P(49 < r < 54).
33. Consider Problem 32. If resistances are measured to the nearest ohm, ﬁnd: (a) the
probability that a particular resistor is within 2 ohms of the mean, (b) P(49 < r < 54).
34. A battery manufacturer has found that 8.08% of their batteries last less than 2.3 years
and 2.5% of their batteries last more than 3.98 years. Assuming the battery lives are
Gaussian distributed, ﬁnd: (a) the mean, (b) variance.
35. Assume that the scores on an examination are Gaussian distributed with mean 75 and
standard deviation 10. Grades are assigned as follows: A: 90–100, B: 80–90, C: 70–80,
D: 60–70, and F: below 60. In a class of 25 students, what is the probability that grades
will be equally distributed?
36. A certain transistor has a current gain, h, that is Gaussian distributed with a mean of 77
and a variance of 11. Find: (a) P(h > 74), (b) P(73 < h ≤ 80), (c) P(h −η
h
 < 3σ
h
).
37. Consider Problem 36. Find the value of d so that the range 77 +d covers 95% of the
current gains.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 41
38. A250 question multiple choice ﬁnal examis given. Each question has 5 possible answers
and only one correct answer. Determine the probability that a student guesses the correct
answers for 20–25 of 85 questions about which the student has no knowledge.
39. The average golf score for Professor Rensselaer is 78 with a standard deviation of 3.
Assuming a Gaussian distribution for random variable x describing her golf game,
determine: (a) P(x = 78), (b) P(x ≤ 78), (c) P(70 < x ≤ 80), (d) the probability that
x is less than 75 if the score is measured to the nearest unit.
40. Suppose a system contains a component whose length is normally distributed with a
mean of 2.0 and a standard deviation of 0.2. If 5 of these components are removed from
different systems, what is the probability that at least 2 have a length greater than 2.1?
41. A large box contains 10,000 resistors with resistances that are Gaussian distributed. If
the average resistance is 1000 ohms with a standard deviation of 200 ohms, how many
resistors have resistances that are within 10% of the average?
42. The RV x has PDF
f
x
(α) = a exp
_
−
1
2σ
2
(α −η)
2
_
.
(a) Find the constant a. (Hint: assume RV y is independent of x and has PDF f
y
(β) =
f
x
(β) and evaluate F
x,y
(∞, ∞).) (b) Using direct integration, ﬁnd E(x). (c) Find σ
2
x
using direct integration.
43. Assume x and y are jointly distributed Gaussian random variables with x ∼ G(−2, 4),
y ∼ G(3, 9), and ρ
x,y
= 0. Find: (a) P(1 < y < 7  x = 0), (b) P(1 < y < 7),
(c) P(−1 < x < 1, 1 < y < 7).
44. Suppose x and y are jointly distributed Gaussian random variables with E(yx) =
2.8 +0.32x, E(xy) = −1 +0.5y, and σ
yx
= 3.67. Determine: (a) η
x
, (b) η
y
, (c) σ
x
,
(d) σ
y
, (e) ρ
x,y
, (f ) σ
x,y
.
45. Assume x ∼ G(3, 1), y ∼ G(−2, 1), and that x and y are jointly Gaussian with ρ
x,y
=
−0.5. Draw a sketch of the joint Gaussian contour equation showing the original and
the translatedrotated sets of axes.
46. Consider Problem45. Determine: (a) E(yx =0), (b) f
yx
(β0), (c) P(0 < y <4x =0),
(d) P(3 < x < 10).
47. Assume x and y are jointly Gaussian with x ∼ G(2, 13), y ∼ G(1, 8), and ρ
x,y
=
−5.8835. (a) Draw a sketch of the constant contour equation for the standardized RVs
z
1
and z
2
. (b) Using the results of (a), Draw a sketch of the joint Gaussian constant
contour for x and y.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
42 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
48. Consider Problem 47. Determine: (a) E(y  x =0), (b) f
y  x
(β  0), (c) P(0 <
y <4x =0), (d) P(3 < x < 10).
49. Assume x and y are jointly Gaussian with x ∼ G(−1, 4), y ∼ G(1.6, 7), and ρ
x,y
=
0.378. (a) Draw a sketch of the constant contour equation for the standardized RVs z
1
and z
2
. (b) Using the results of (a), Draw a sketch of the joint Gaussian contour for x
and y.
50. Consider Problem 49. Determine: (a) E(y  x =0), (b) f
yx
(β  0), (c) P(0 <
y <4x =0), (d) P(−3 < x < 0).
51. A component with an exponential failure rate is 90% reliable at 10,000 h. Determine
the number of hours reliable at 95%.
52. Suppose a system has an exponential failure rate in years to failure with η = 2.5. De
termine the number of years reliable at: (a) 90%, (b) 95%, (c) 99%.
53. Consider Problem 52. If 20 of these systems are installed, determine the probability
that 10 are operating at the end of 2.3 years.
54. In the circuit shown in Figure 5.12, each of the four components operate independently
of one another and have an exponential failure rate (in hours) with η = 10
5
. For suc
cessful operation of the circuit, at least two of these components must connect A with
B. Determine the probability that the circuit is operating successfully at 10,000 h.
55. The survival rate of individuals with a certain type of cancer is assumed to be exponential
with η = 4 years. Five individuals have this cancer. Determine the probability that at
most three will be alive at the end of 2.0433 years.
56. Random variable t is exponential with η = 2. Determine: (a) P(t > η), (b) f
z
(α) if
z = t − T, where T is a constant.
57. William Smith is a varsity wrestler on his high school team. Without exception, if he
does not pin his opponent with his trick move, he loses the match on points. William’s
3
1 2
4
A
B
FIGURE5.12: Circuit for Problem 54.
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
STANDARDPROBABILITY DISTRIBUTIONS 43
trick move also prevents him from ever getting pinned. The length of time it takes
William to pin his opponent in each period of a wrestling match is given by:
period 1 : f
x
(α) = 0.4598394 exp(−0.4598394α)u(α),
period 2 : f
x A
(α A) = 0.2299197 exp(−0.2299197α)u(α),
period 3 : f
x A
(α A) = 0.1149599 exp(−0.1149599α)u(α),
where A = {Smith did not pin his opponent during the previous periods}. Assume
each period is 2 min and the match is 3 periods. Determine the probability that William
Smith: (a) pins his opponent during the ﬁrst period: (b) pins his opponent during the
second period: (c) pins his opponent during the third period: (d) wins the match.
58. Consider Problem 57. Find the probability that William Smith wins: (a) at least 4 of
his ﬁrst 5 matches, (b) more matches than he is expected to during a 10 match season.
59. The average time between power failures in a Coop utility is once every 1.4 years.
Determine: (a) the probability that there will be at least one power failure during the
coming year, (b) the probability that there will be at least two power failures during the
coming year.
60. Consider Problem59 statement. Assume that a power failure will last at least 24 h. Sup
pose Fargo Community Hospital has a backup emergency generator to provide auxilary
power during a power failure. Moreover, the emergency generator has an expected time
between failures of once every 200 h. What is the probability that the hospital will be
without power during the next 24 h?
61. The queue for the cash register at a popular package store near Fargo Polytechnic
Institute becomes quite long on Saturdays following football games. On the average,
the queue is 6.3 people long. Each customer takes 4 min to check out. Determine: (a)
your expected waiting time to make a purchase, (b) the probability that you will have
less than four people in the queue ahead of you, (c) the probability that you will have
more than ﬁve people in the queue ahead of you.
62. Outer Space Adventures, Inc. prints brochures describing their vacation packages as
“Unique, Unexpected and Unexplained.” The vacations certainly live up to their ad
vertisement. In fact, they are so unexpected that the length of the vacations is random,
following an exponential distribution, having an average length of 6 months. Suppose
that you have signed up for a vacation trip that starts at the end of this quarter. What
is the probability that you will be back home in time for next fall quarter (that is, 9
months later)?
P1: IML/FFX P2: IML
MOBK04205 MOBK042Enderle.cls October 30, 2006 19:51
44 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
63. A certain brand of light bulbs has an average lifeexpectancy of 750 h. The failure rate
of these light bulbs follows an exponential PDF. Sevenhundred and ﬁfty of these bulbs
were put in light ﬁxtures in four rooms. The lights were turned on and left that way for
a different length of time in each room as follows:
ROOM TIMEBULBS LEFTON, HOURS NUMBEROF BULBS
1 1000 125
2 750 250
3 500 150
4 1500 225
After the speciﬁed length of time, the bulbs were taken from the ﬁxtures and placed
in a box. If a bulb is selected at random from the box, what is the probability that it is
burnt out?
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
45
C H A P T E R 6
Transformations of RandomVariables
Functions of random variables occur frequently in many applications of probability theory. For
example, a full wave rectiﬁer circuit produces an output that is the absolute value of the input.
The input/output characteristics of many physical devices can be represented by a nonlinear
memoryless transformation of the input.
The primary subjects of this chapter are methods for determining the probability dis
tribution of a function of a random variable. We ﬁrst evaluate the probability distribution of
a function of one random variable using the CDF and then the PDF. Next, the probability
distribution for a single random variable is determined from a function of two random variables
using the CDF. Then, the joint probability distribution is found from a function of two random
variables using the joint PDF and the CDF.
6.1 UNIVARIATECDF TECHNIQUE
This section introduces a method of computing the probability distribution of a function of
a random variable using the CDF. We will refer to this method as the CDF technique. The
CDF technique is applicable for all functions z = g(x), and for all types of continuous, discrete,
and mixed random variables. Of course, we require that the function z : S → R
∗
, with z(ζ ) =
g(x(ζ )), is a random variable on the probability space (S, , P); consequently, we require z to
be a measurable function on the measurable space (S, ) and P(z(ζ ) ∈ {−∞, +∞}) = 0.
The ease of use of the CDF technique depends critically on the functional form of g(x).
To make the CDF technique easier to understand, we start the discussion of computing the
probability distribution of z = g(x) with the simplest case, a continuous monotonic function
g. (Recall that if g is a monotonic function then a onetoone correspondence between g(x)
and x exists.) Then, the difﬁculties associated with computing the probability distribution of
z = g(x) are investigated when the restrictions on g(x) are relaxed.
6.1.1 CDF Technique with Monotonic Functions
Consider the problem where the CDF F
x
is known for the RV x, and we wish to ﬁnd the CDF
for random variable z = g(x). Proceeding from the deﬁnition of the CDF for z, we have for a
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
46 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
monotonic increasing function g(x)
F
z
(γ ) = P(z = g(x) ≤ γ ) = P(x ≤ g
−1
(γ )) = F
x
(g
−1
(γ )), (6.1)
where g
−1
(γ ) is the value of α for which g(α) = γ . As (6.1) indicates, the CDF of random
variable z is written in terms of F
x
(α), with the argument α replaced by g
−1
(γ ).
Similarly, if z = g(x) and g is monotone decreasing, then
F
z
(γ ) = P(z = g(x) ≤ γ ) = P(x ≥ g
−1
(γ )) = 1 − F
x
(g
−1
(γ )
−
). (6.2)
The following example illustrates this technique.
Example 6.1.1. Random variable x is uniformly distributed in the interval 0 to 4. Find the CDF
for random variable z = 2x +1.
Solution. Since random variable x is uniformly distributed, the CDF of x is
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < 0
α/4, 0 ≤ α < 4
1, 4 ≤ α.
Letting g(x) = 2x +1, we see that g is monotone increasing and that g
−1
(γ ) = (γ −1)/2.
Applying (6.1), the CDF for z is given by
F
z
(γ ) = F
x
_
γ −1
2
_
=
⎧
⎪
⎨
⎪
⎩
0, (γ −1)/2 < 0
(γ −1)/8, 0 ≤ (γ −1)/2 < 4
1, 4 ≤ (γ −1)/2.
Simplifying, we obtain
F
z
(γ ) =
⎧
⎪
⎨
⎪
⎩
0, γ < 1
(γ −1)/8, 1 ≤ γ < 9
1, 9 ≤ γ,
which is also a uniform distribution.
6.1.2 CDF Technique with Arbitrary Functions
In general, the relationship between x and z can take on any form, including discontinuities.
Additionally, the function does not have to be monotonic, more than one solution of z = g(x)
can exist—resulting in a manytoone mapping from x to z. In general, the only requirement
on g is that z = g(x) be a random variable. In this general case, F
z
(γ ) is no longer found by
simple substitution. In fact, under these conditions it is impossible to write a general expression
for F
z
(γ ) using the CDF technique. However, this case is conceptually no more difﬁcult than
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 47
the previous case, and involves only careful book keeping.
Let
A(γ ) = {x : g(x) ≤ γ }. (6.3)
Note that A(γ ) = g
−1
((−∞, γ ]). Then
F
z
(γ ) = P(g(x) ≤ γ ) = P(x ∈ A(γ )). (6.4)
Partition A(γ ) into disjoint intervals {A
i
(γ ) : i = 1, 2, . . .} so that
A(γ ) =
∞
_
i =1
A
i
(γ ). (6.5)
Note that the intervals as well as the number of nonempty intervals depends on γ . Since the
A
i
’s are disjoint,
F
z
(γ ) =
∞
i =1
P(x ∈ A
i
(γ )). (6.6)
The above probabilities are easily found from the CDF F
x
. If interval A
i
(γ ) is of the form
A
i
(γ ) = (a
i
(γ ), b
i
(γ )], (6.7)
then
P(x ∈ A
i
(γ )) = F
x
(b
i
(γ )) − F
x
(a
i
(γ )). (6.8)
Similarly, if interval A
i
(γ ) is of the form
A
i
(γ ) = [a
i
(γ ), b
i
(γ )], (6.9)
then
P(x ∈ A
i
(γ )) = F
x
(b
i
(γ )) − F
x
(a
i
(γ )
−
). (6.10)
The success of this method clearly depends onour ability to partition A(γ ) into disjoint intervals.
Using this method, any function g and CDF F
x
is amenable to a solution for F
z
(γ ). The
following examples illustrate various aspects of this technique.
Example 6.1.2. Random variable x has the CDF
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < −1
(3α −α
3
+2)/4, −1 ≤ α < 1
1, 1 ≤ α.
Find the CDF for the RV z = x
2
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
48 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. Letting g(x) = x
2
, we ﬁnd
A(γ ) = g
−1
((−∞, γ ]) =
_
∅, γ < 0
[−
√
γ ,
√
γ ], γ ≥ 0.
so that
F
z
(γ ) = F
x
(
√
γ ) − F
x
((−
√
γ )
−
).
Noting that F
x
(α) is continuous and has the same functional form for −1 < α < 1, we obtain
F
z
(γ ) =
⎧
⎪
⎨
⎪
⎩
0, γ < 0
(3
√
γ −(
√
γ )
3
)/2, 0 ≤ γ < 1
1 1 ≤ γ.
Example 6.1.3. Random variable x is uniformly distributed in the interval −3 to 3. Random
variable z is deﬁned by
z = g(x) =
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
−1, x < −2
3x +5, −2 ≤ x < −1
−3x −1, −1 ≤ x < 0
3x −1, 0 ≤ x < 1
2, 1 ≤ x.
Find F
z
(γ ).
Solution. Plots for this example are given in Figure 6.1. The CDF for random variable x is
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < −3
(α +3)/6, −3 ≤ α < 3
1, 3 ≤ α.
Let A(γ ) = g
−1
((−∞, ∞)). Referring to Figure 6.1 we ﬁnd
A(γ ) =
⎧
⎪
⎪
⎨
⎪
⎪
⎩
∅, γ < −1
_
−∞,
γ −5
3
_
∪
_
−γ −1
3
,
γ +1
3
_
, −1 ≤ γ < 2
R 2 ≤ γ.
Consequently,
F
z
(γ ) =
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
0, γ < −1
F
x
_
γ −5
3
_
+ F
x
_
γ +1
3
_
− F
x
_
_
−γ −1
3
_
−
_
, −1 ≤ γ < 2
1 2 ≤ γ.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 49
−3 −1 1 3
1
z
F (γ)
−3 −1 1 3
1
x
F (α)
−3 −1 1 3
4
x
g(x)
2
−1
γ
α
FIGURE6.1: Plots for Example 6.1.3.
Substituting, we obtain
F
z
(γ ) =
⎧
⎪
⎪
⎨
⎪
⎪
⎩
0, γ < −1
1
6
_
γ −5
3
+3 +
γ +1
3
−
−γ −1
3
_
=
γ +2
6
, −1 ≤ γ < 2
1 2 ≤ γ.
Note that F
z
(γ ) has jumps at γ = −1 and at γ = 2.
Whenever random variable x is continuous and g(x) is constant over some interval or
intervals, then random variable z can be continuous, discrete or mixed, dependent on the CDF
for random variable x. As presented in the previous example, random variable z is mixed due to
the constant value of g(x) in the intervals −3 ≤ x < −2 and 1 ≤ x < 3. In fact, z is a mixed
random variable whenever g(α) is constant in an interval where f
x
(α) = 0. This results in a
jump in F
z
and an impulse function in f
z
. Moreover, z = g(x) is a discrete random variable if
g(α) changes values only on intervals where f
x
(α) = 0. Random variable z is continuous if x is
continuous and g(x) is not equal to a constant over any interval where f
x
(α) = 0.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
50 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Example 6.1.4. Random variable x has the PDF
f
x
(α) =
_
(1 +α
2
)/6, −1 < α < 2
0, otherwise.
Find the PDF of random variable z deﬁned by
z = g(x) =
⎧
⎪
⎨
⎪
⎩
x −1, x ≤ 0
0, 0 < x ≤ 0.5
1, 0.5 < x.
Solution. To ﬁnd f
z
, we evaluate F
z
ﬁrst, then differentiate this result. The CDF for random
variable x is
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < −1
(α
3
+3α +4)/18, −1 ≤ α < 2
1, 2 ≤ α.
Figure 6.2 shows a plot of g(x). With the aid of Figure 6.2, we ﬁnd
A(γ ) = {x : g(x) ≤ γ } =
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
(−∞, γ +1], γ ≤ −1
(−∞, 0], −1 ≤ γ < 0
(−∞, 0.5], 0 ≤ γ < 1
(−∞, ∞), 1 ≤ γ.
Consequently,
F
z
(γ ) =
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
F
x
(γ +1), γ ≤ −1
F
x
(0), −1 ≤ γ < 0
F
x
(0.5), 0 ≤ γ < 1
1, 1 ≤ γ.
−1 1 2
x
1
−1
g( x)
−2
FIGURE6.2: Transformation for Example 6.1.4.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 51
Substituting,
F
z
(γ ) =
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
0, γ < −2
((γ +1)
3
+3γ +7)/18, −2 ≤ γ < −1
2/9, −1 ≤ γ < 0
5/16, 0 ≤ γ < 1
1, 1 ≤ γ.
Note that F
z
(γ ) has jumps at γ = 0 and γ = 1 of heights 13/144 and 11/16, respectively.
Differentiating F
z
,
f
z
(γ ) =
3γ
2
+6γ +6
18
(u(γ +2) −u(γ +1)) +
13
144
δ(γ ) +
11
16
δ(γ −1).
Example 6.1.5. Random variable x is uniformly distributed in the interval from 0 to 10. Find the
CDF for random variable z = g(x) = −ln(x).
Solution. The CDF for x is
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < 0
α/10, 0 ≤ α < 10
1, 10 ≤ α.
For γ > 0, we ﬁnd
A(γ ) = {x : −ln(x) ≤ γ } = (e
−γ
, ∞),
so that F
z
(γ ) = 1 − F
x
((e
−γ
)
−
). Note that P(x ≤ 0) = 0, as required since g(x) = −ln(x) is
not deﬁned (or at least not real–valued) for x ≤ 0. We ﬁnd
F
z
(γ ) =
_
0, γ < ln(0.1)
1 −e
−0.1γ
, ln(0.1) ≤ γ.
The previous examples illustrated evaluating the probability distribution of a function
of a continuous random variable using the CDF technique. This technique is applicable for
all functions z = g(x), continuous and discontinuous. Additionally, the CDF technique is
applicable if random variable x is mixed or discrete. For mixed random variables, the CDF
technique is used without any changes or modiﬁcations as shown in the next example.
Example 6.1.6. Random variable x has PDF
f
x
(α) = 0.5(u(α) −u(α −1)) +0.5δ(α −0.5).
Find the CDF for z = g(x) = 1/x
2
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
52 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. The mixed random variable x has CDF
F
x
(α) =
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
0, α < 0
0.5α, 0 ≤ α < 0.5
0.5 +0.5α, 0.5 ≤ α < 1
1, 1 ≤ α.
For γ < 0, F
z
(γ ) = 0. For γ > 0,
A(γ ) = {x : x
−2
≤ γ } =
_
−∞, −
1
√
γ
_
∪
_
1
√
γ
, ∞
_
,
so that
F
z
(γ ) = F
x
(−1/
√
γ ) +1 − F
x
((1/
√
γ )
−
).
Since F
x
(−1/
√
γ ) = 0 for all real γ , we have
F
z
(γ ) = 1 −
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
0, (1/
√
γ )
−
< 0
0.5γ
−1/2
, 0 ≤ (1/
√
γ )
−
< 0.5
0.5 +0.5γ
−1/2
, 0.5 ≤ (1/
√
γ )
−
< 1
1, 1 ≤ (1/
√
γ )
−
.
After some algebra,
F
z
(γ ) =
⎧
⎪
⎨
⎪
⎩
0, γ < 1
0.5 −0.5γ
−1/2
, 1 ≤ γ < 4
1 −0.5γ
−1/2
, 4 ≤ γ.
Drill Problem6.1.1. Randomvariable x is uniformly distributed in the interval −1 to 4. Random
variable z = 3x +2. Determine: (a) F
z
(0), (b)F
z
(1), (c ) f
z
(0), (c ) f
z
(15).
Answers: 0, 2/15, 1/15, 1/15.
Drill Problem6.1.2. Random variable x has the PDF
f
x
(α) = 0.5α(u(α) −u(α −2)).
Random variable z is deﬁned by
z =
⎧
⎪
⎨
⎪
⎩
−1, x < 1
x, −1 ≤ x ≤ 1
1, x > 1.
Determine: (a) F
z
(−1/2), (b)F
z
(1/2), (c )F
z
(3/2), (d) f
z
(1/2).
Answers: 0, 1, 1/16, 1/4.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 53
Drill Problem6.1.3. Random variable x has the PDF
f
x
(α) = 0.5α(u(α) −u(α −2)).
Random variable z is deﬁned by
z =
⎧
⎪
⎨
⎪
⎩
−1, x ≤ 0.5
x +0.5, 0.5 < x ≤ 1
3, x > 1.
Determine: (a) F
z
(−1), (b)F
z
(0), (c )F
z
(3/2), (d)F
z
(4).
Answers: 1/4, 1/16, 1/16, 1.
Drill Problem6.1.4. Random variable x has PDF
f
x
(α) = e
−α−1
u(α +1).
Random variable z = 1/x
2
. Determine: (a) F
z
(1/8), (b)F
z
(1/2), (c )F
z
(4), (d) f
z
(4).
Answers: 0.0519, 0.617, 0.089, 0.022.
6.2 UNIVARIATEPDF TECHNIQUE
The previous sectionsolvedthe problemof determining the probability distributionof a function
of a random variable using the cumulative distribution function. Now, we introduce a second
method for calculating the probability distribution of a function z = g(x) using the probability
density function, called the PDF technique. The PDF technique, however, is only applicable
for functions of randomvariables in which z = g(x) is continuous and does not equal a constant
in any interval in which f
x
is nonzero. We introduce the PDF technique for two reasons. First,
in many situations it is much simpler to use than the CDF technique. Second, we will ﬁnd the
PDF method most useful in extensions to multivariate functions. In this section, we discuss
a wide variety of situations using the PDF technique with functions of continuous random
variables. Then, a method for handling mixed random variables with the PDF technique is
introduced. Finally, we consider computing the conditional PDF of a function of a random
variable using the PDF technique.
6.2.1 Continuous RandomVariable
Theorem 6.2.1. Let x be a continuous RV with PDF f
x
(α) and let the RV z = g(x) . Assume g
is continuous and not constant over any interval for which f
x
= 0 . Let
α
i
= α
i
(γ ) = g
−1
(γ ), i = 1, 2, . . . , (6.11)
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
54 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
denote the distinct solutions to g(α
i
) = γ . Then
f
z
(γ ) =
∞
i =1
f
x
(α
i
(γ ))
g
(1)
(α
i
(γ ))
, (6.12)
where we interpret
f
x
(α
i
(γ ))
g
(1)
(α
i
(γ ))
= 0,
if f
x
(α
i
(γ ) = 0.
Proof. Let h > 0 and deﬁne
I(γ, h) = {x : γ −h < g(x) ≤ γ }.
Partition I(γ, h) into disjoint intervals of the form
I
i
(γ, h) = (a
i
(γ, h), b
i
(γ, h)), i = 1, 2, . . . ,
such that
I(γ, h) =
∞
_
i =1
I
i
(γ, h).
Then
F
z
(γ ) − F
z
(γ −h) =
∞
i =1
(F
x
(b
i
(γ, h)) − F
x
(a
i
(γ, h)))
By hypothesis,
lim
h→0
a
i
(γ, h) = lim
h→0
b
i
(γ, h) = α
i
(γ ).
Note that (for all γ with f
x
(α
i
(γ )) = 0)
lim
h→0
b
i
(γ, h) −a
i
(γ, h)
h
= lim
h→0
b
i
(γ, h) −a
i
(γ, h)
g(b
i
(γ, h)) − g(a
i
(γ, h))
=
1
g
(1)
(α
i
(γ ))
,
and that
lim
h→0
F
x
(b
i
(γ, h)) − F
x
(a
i
(γ, h))
b
i
(γ, h) −a
i
(γ, h)
= f
x
(α
i
(γ )).
The desired result follows by taking the product of the above limits. The absolute value ap
pears because by construction we have b
i
> a
i
and h > 0, whereas g
(1)
may be positive or
negative.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 55
Example 6.2.1. Randomvariable x is uniformly distributed in the interval 0–4. Find the PDF for
random variable z = g(x) = 2x +1.
Solution. In this case, there is only one solution to the equation γ = 2α +1, given by α
1
=
(γ −1)/2. We easily ﬁnd g
(1)
(α) = 2. Hence
f
z
(γ ) = f
x
((γ −1)/2)/2 =
_
1/8, 1 < γ < 9
0, otherwise.
Example 6.2.2. Random variable x has PDF
f
x
(α) = 0.75(1 −α
2
)(u(α +1) −u(α −1)).
Find the PDF for random variable z = g(x) = 1/x
2
.
Solution. For γ < 0, there are no solutions to g(α
i
) = γ , so that f
z
(γ ) = 0 for γ < 0. For
γ > 0 there are two solutions to g(α
i
) = γ :
α
1
(γ ) = −
1
√
γ
, and α
2
(γ ) =
1
√
γ
.
Since γ = g(α) = α
−2
, we have g
(1)
(α) = −2α
−3
; hence, g
(1)
(α
i
) = 2/α
i

3
= 2γ 
3/2
, and
f
z
(γ ) =
f
x
(−γ
−1/2
) + f
x
(γ
−1/2
)
2γ
3/2
u(γ ).
Substituting,
f
z
(γ ) =
0.75(1 −γ
−1
)(u(1 −γ
−1/2
) −0 +1 −u(γ
−1/2
−1))
2γ
3/2
u(γ ).
Simplifying,
f
z
(γ ) =
0.75(1 −γ
−1
)(u(γ −1) −0 +1 −u(1 −γ ))
2γ
3/2
u(γ ),
or
f
z
(γ ) = 0.75(γ
−
3
2
−γ
−
5
2
)u(γ −1).
Example 6.2.3. Random variable x has PDF
f
x
(α) =
1
6
(1 +α
2
)(u(α +1) −u(α −2)).
Find the PDF for random variable z = g(x) = x
2
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
56 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Solution. For γ < 0 there are no solutions to g(α
i
) = γ , so that f
z
(γ ) = 0 for γ < 0. For
γ > 0 there are two solutions to g(α
i
) = γ :
α
1
(γ ) = −
√
γ , and α
2
(γ ) =
√
γ .
Since γ = g(α) = α
2
, we have g
(1)
(α) = 2α; hence, g
(1)
(α
i
) = 2α
i
 = 2
√
γ , and
f
z
(γ ) =
f
x
(−
√
γ ) + f
x
(
√
γ )
2
√
γ
u(γ ).
Substituting,
f
z
(γ ) =
1 +γ
12
√
γ
(u(1 −
√
γ ) −u(−2 −
√
γ ) +u(
√
γ +1) −u(
√
γ −2))u(γ ).
Simplifying,
f
z
(γ ) =
1 +γ
12
√
γ
(u(1 −γ ) −0 +1 −u(γ −4))u(γ ),
or
f
z
(γ ) =
⎧
⎪
⎨
⎪
⎩
(γ
−1/2
+γ
1/2
)/6, 0 < γ < 1
(γ
−1/2
+γ
1/2
)/12, 1 < γ < 4
0, elsewhere.
6.2.2 Mixed RandomVariable
Consider the problem where random variable x is mixed, and we wish to ﬁnd the PDF for
z = g(x). Here, we treat the discrete and continuous portions of f
x
separately, and then combine
the results to yield f
z
. The continuous part of the PDF of x is handled by the PDF technique.
To illustrate the use of this technique, consider the following example.
Example 6.2.4. Random variable x has PDF
f
x
(α) =
3
8
(u(α +1) −u(α −1)) +
1
8
δ(α +0.5) +
1
8
δ(α −0.5).
Find the PDF for the RV z = g(x) = e
−x
.
Solution. There is only one solution to g(α) = γ :
α
1
(γ ) = −ln(γ ).
We have g
(1)
(α
1
) = −e
ln(γ )
= −γ . The probability masses of 1/8 for x at −0.5 and 0.5 are
mapped to probability masses of 1/8 for z at e
0.5
and e
−0.5
, respectively. For all γ > 0 such that
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 57
α
1
(γ ) ±0.5 > 0 we have
f
z
(γ ) =
f
x
(−ln(γ ))
γ
.
Combining these results, we ﬁnd
f
z
(γ ) =
3
8γ
(u(γ −e
−1
) −u(γ −e )) +
1
8
δ(γ −e
0.5
) +
1
8
δ(γ −e
−0.5
).
6.2.3 Conditional PDF Technique
Since a conditional PDF is also a PDF, the above techniques apply to ﬁnd the conditional PDF
for z = g(x), given event A. Consider the problem where random variable x has PDF f
x
, and
we wish to evaluate the conditional PDF for random variable z = g(x), given that event A
occurred. Depending on whether the event A is deﬁned on the range or domain of z = g(x),
one of the following two methods may be used to determine the conditional PDF of z using
the PDF technique.
(i) If A is an event deﬁned for an interval on z, the conditional PDF, f
z A
, is computed by ﬁrst
evaluating f
z
using the technique in this section. Then, by the deﬁnition of a conditional
PDF, we have
f
z A
(γ  A) =
f
z
(γ )
P(A)
, γ ∈ A, (6.13)
and f
z A
(γ  A) = 0 for γ ∈A.
(ii) If A is an event deﬁned for an interval on x, we will use the conditional PDF of x to evaluate
the conditional PDF for z as
f
z A
(γ  A) =
∞
i =1
f
z A
(α
i
(γ ) A)
g
(1)
(α
i
(γ ))
. (6.14)
Example 6.2.5. Random variable x has the PDF
f
x
(α) =
1
6
(1 +α
2
)(u(α +1) −u(α −2)).
Find the PDF for random variable z = g(x) = x
2
, given A = {x : x > 0}.
Solution. First, we solve for the conditional PDF for x and then ﬁnd the conditional PDF for
z, based on f
x A
. We have
P(A) =
1
6
_
2
0
(1 +α
2
)dα =
7
9
,
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
58 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
so that
f
x A
(α A) =
3
14
(1 +α
2
)(u(α) −u(α −2)).
There is only one solution to γ = g(α) = α
2
on the interval 0 < α < 2 where f
x A
= 0. We
have α
1
(γ ) =
√
γ and g
(1)
(α
1
(γ )) = 2
√
γ . Consequently,
f
z A
(γ  A) =
3
28
_
√
γ +
1
√
γ
_
(u(γ ) −u(γ −4)).
Drill Problem6.2.1. Randomvariable x has a uniformPDFinthe interval 0–8. Randomvariable
z = 3x +1 .Use the PDF method to determine: (a) f
z
(0), (b) f
z
(6), (c )E(z), (d)σ
2
z
.
Answers: 13, 48, 0, 1/24.
Drill Problem6.2.2. Random variable x has the PDF
f
x
(α) =
⎧
⎪
⎨
⎪
⎩
9α
2
, 0 ≤ α < 0.5
3(1 −α
2
), 0.5 ≤ α ≤ 1
0, otherwise.
Random variable z = x
3
.Use the PDF method to determine: (a) f
z
(1/27), (b) f
z
(1/4), (c ) f
zz>1/8
(1/4z > 1/8), (d) f
z
(2).
Answers: 1.52, 0, 3, 2.43.
Drill Problem6.2.3. Random variable x has the PDF
f
x
(α) =
2
9
α(u(α) −u(α −3)).
Random variable z = (x −1)
2
. Use the PDF method to determine: (a) f
z
(1/4), (b) f
z
(9/4), (c )
f
zz≤1
(1/4z ≤ 1), (d)E(zz ≤ 1).
Answers: 4/9, 5/27, 1/3, 1.
Drill Problem6.2.4. Random variable x has the PDF
f
x
(α) =
2
9
(α +1)(u(α +1) −u(α −2)).
Random variable z = 2x
2
and event A = {x : x ≥ 0}. Determine: (a) P(A), (b) f
x A
(1 A), (c )
f
z A
(2 A),(d) f
z A
(9 A).
Answers: 0, 1/2, 1/8, 8/9.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 59
6.3 ONEFUNCTIONOF TWORANDOMVARIABLES
Consider a random variable z = g(x, y) created from jointly distributed random variables x
and y. In this section, the probability distribution of z = g(x, y) is computed using a CDF
technique similar to the one at the start of this chapter. Because we are dealing with regions in
a plane instead of intervals on a line, these problems are not as straightforward and tractable as
before.
With z = g(x, y), we have
F
z
(γ ) = P(z ≤ γ ) = P(g(x, y) ≤ γ ) = P((x, y) ∈ A(γ )), (6.15)
where
A(γ ) = {(x, y) : g(x, y) ≤ γ }. (6.16)
The CDF for the RV z can then be found by evaluating the integral
F
z
(γ ) =
_
A(γ )
dF
x,y
(α, β). (6.17)
This result cannot be continued further until a speciﬁc F
x,y
and g(x, y) are considered. Re
member that in the case of a single random variable, our efforts primarily dealt with algebraic
manipulations. Here, our efforts are concentrated on evaluating F
z
through integrals, with the
ease of solution critically dependent on g(x, y).
The ease in solution for F
z
is dependent on transforming A(γ ) into proper limits of
integration. Sketching the support region for f
x,y
(the region where f
x,y
= 0, or F
x,y
is not
constant) and the region A(γ ) is often most helpful, even crucial, in the problem solution. Pay
careful attention to the limits of integration to determine the range of integration in which the
integrand is zero because f
x,y
= 0. Let us consider several examples to illustrate the mechanics
of the CDF technique and also to provide further insight.
Example 6.3.1. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
1/4, 0 < α < 2, 0 < β < 2
0, otherwise.
Find the CDF for z = x + y.
Solution. We have A(γ ) = {(α, β) : α +β ≤ γ }. We require the volume under the surface
f
x,y
(α, β) where α ≤ γ −β:
F
z
(γ ) =
_
∞
−∞
_
γ −β
−∞
f
x,y
(α, β)dα dβ.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
60 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
( a) 0 < γ < 2 ( b) 2 < γ < 4
α 0
β
2
2
γ
γ
2
2 − γ
β
0 α
2
2
γ
γ
− γ
FIGURE6.3: Plots for Example 6.3.1.
For γ < 0 we have F
z
(γ ) = 0. For 0 ≤ γ < 2, with the aid of Figure 6.3(a) we obtain
F
z
(γ ) =
_
2
0
_
γ −β
0
1
4
dαdβ =
1
8
γ
2
.
For 2 ≤ γ < 4, referring to Figure 6.3(b), we consider the complementary region (to save some
work):
F
z
(γ ) = 1 −
_
2
γ −2
_
2
γ −β
1
4
dαdβ = 1 −
1
8
(4 −γ )
2
.
Finally, for 4 ≤ γ , F
z
(γ ) = 1.
Example 6.3.2. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
1, 0 < α < 1, 0 < β < 1
0, otherwise.
Find the PDF for z = x − y.
Solution. We have A(γ ) = {(α, β) : α −β ≤ γ }. We require the volume under the surface
f
x,y
(α, β) where α ≤ γ +β:
F
z
(γ ) =
_
∞
−∞
_
γ +β
−∞
f
x,y
(α, β)dαdβ.
For γ < −1 we have F
z
(γ ) = 0 and f
z
(γ ) = 0. With the aid of Figure 6.4(a), for −1 ≤ γ < 0,
F
z
(γ ) =
_
1
−γ
_
γ +β
0
dαdβ =
1
2
(1 +γ )
2
,
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 61
β
0 α
1
1
1
γ
γ
β
0
α
1
1 1+
−γ − γ
(a) −1 < γ < 0 (b) 0 < γ < 1
FIGURE6.4: Plots for Example 6.3.2.
so that f
z
(γ ) = 1 +γ . For 0 ≤ γ < 1, we consider the complementary region shown in Fig
ure 6.4(b) (to save some work):
F
z
(γ ) = 1 −
_
1−γ
0
_
1
γ +β
dαdβ = 1 −
1
2
(1 −γ )
2
,
so that f
z
(γ ) = 1 −γ . Finally, for 1 ≤ γ , F
z
(γ ) = 1, so that f
z
(γ ) = 0.
Example 6.3.3. Find the CDF for z = x/y, where x and y have the joint PDF
f
x,y
(α, β) =
_
1/α, 0 < β < α < 1
0, otherwise.
Solution. We have A(γ ) = {(α, β) : α/β ≤ γ }. Inside the support region for f
x,y
, we have
α/β > 1; hence, for γ < 1 we have F
z
(γ ) = 0. As shown in Figure 6.5 it is easiest to integrate
with respect to β ﬁrst: for 1 ≤ γ ,
F
z
(γ ) =
_
1
0
_
α
α/γ
1
α
dβdα = 1 −
1
γ
.
β
γ
0 α
1
1
1
FIGURE6.5: Integration region for Example 6.3.3.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
62 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
Example 6.3.4. Find the CDF for z = x
2
+ y
2
, where x and y have the joint PDF
f
x,y
(α, β) =
_
3α, 0 < β < α < 1
0, otherwise.
Solution. We have A(γ ) = {(α, β) : α
2
+β
2
≤ γ }. For γ < 0, we obtain F
z
(γ ) = 0. Trans
forming to polar coordinates: α = r cos(θ), β = r sin(θ),
F
z
(γ ) =
_
2π
0
_
r
2
≤γ
f
x,y
(r cos(θ), r sin(θ))r dr dθ.
Referring to Figure 6.6(a), for 0 ≤ γ < 1,
F
z
(γ ) =
_
π/4
0
_
√
γ
0
3r
2
cos(θ)dr dθ =
1
√
2
γ
3
2
.
For 1 ≤ γ < 2, we split the integral into two parts: one with polar coordinates, the other using
rectangular coordinates. With
sin(θ
1
) =
√
γ −1
√
γ
,
we ﬁnd with the aid of Figure 6.6(b) that
F
z
(γ ) =
_
π/4
θ
1
_
√
γ
0
3r
2
cos(θ)dr dθ +
_
1
0
_
α
√
γ −1
0
3αdβ dα,
or
F
z
(γ ) =
1
√
2
γ
3
2
−(γ −1)
3
2
.
Finally, we have F
z
(γ ) = 1 for 2 ≤ γ .
(a) 0 < γ < 1 (b) 1 < γ < 2
β
0 α
1
1
γ
γ
α 0
β
1
1
γ
γ
1 − γ
FIGURE6.6: Integration regions for Example 6.3.4.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 63
Drill Problem6.3.1. Random variables x and y have joint PDF
f
x,y
(α, β) = e
−α
e
−β
u(α)u(β).
Random variable z = x − y . Find (a) F
z
(−1/3), (b) f
z
(−1), (c )F
z
(1), and (d) f
z
(1).
Answers:
1
4
e
−1
,
3
4
e
−1
, 1 −
3
4
e
−1
,
3
4
e
−1
.
6.4 BIVARIATETRANSFORMATIONS
In this section, we ﬁnd the joint distribution of random variables z = g(x, y) and w = h(x, y)
fromjointly distributed randomvariables x and y. First, we consider a bivariate CDF technique.
Then, the joint PDF technique for ﬁnding the joint PDF for random variables z and w formed
from x and y is described. Next, the case of one function of two random variables is treated by
using the joint PDF technique with an auxiliary random variable. Finally, the conditional joint
PDF is presented.
6.4.1 Bivariate CDF Technique
Let x and y be jointly distributed RVs on the probability space (S, , P), and let z = g(x, y)
and w = h(x, y). Deﬁne A(γ, ψ) to be the region of the x − y plane for which z = g(x, y) ≤ γ
and w = h(x, y) ≤ ψ; i.e.,
A(γ, ψ) = {(x, y) : g(x, y) ≤ γ, h(x, y) ≤ ψ}. (6.18)
Note that
A(γ, ψ) = g
−1
((−∞, γ ]) ∩ h
−1
((−∞, ψ]). (6.19)
Then
F
z,w
(γ, ψ) =
_
A(γ,ψ)
_
dF
x,y
(α, β). (6.20)
It is often difﬁcult to perform the integration indicated in (6.20).
Example 6.4.1. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
0.25, 0 ≤ α ≤ 2, 0 ≤ β ≤ 2
0, otherwise.
With z = g(x, y) = x + y and w = h(x, y) = y, ﬁnd the joint PDF f
z,w
.
Solution. We have
g
−1
((−∞, γ ]) = {(x, y) : x + y ≤ γ }
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
64 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
y
x
ψ
γ
γ ψ −
FIGURE6.7: Integration region for Example 6.4.1.
and
h
−1
((−∞, ψ]) = {(x, y) : y ≤ ψ}.
The intersection of these regions is
A(γ, ψ) = {(x, y) : y ≤ min(γ − x, ψ)},
which is illustrated in Figure. 6.7. With the aid of Figure 6.7 and (6.20) we ﬁnd
F
z,w
(γ, ψ) =
_
ψ
−∞
_
γ −β
−∞
d F
x,y
(α, β).
Instead of carrying out the above integration and then differentiating the result, we differentiate
the above integral to obtain the PDF f
z,w
directly. We ﬁnd
∂ F
z,w
(γ, ψ)
∂γ
= lim
h
1
→0
1
h
1
_
ψ
−∞
_
γ −β
γ −h
1
−β
d F
x,y
(α, β),
and
∂
2
F
z,w
(γ, ψ)
∂ψ ∂γ
= lim
h
2
→0
lim
h
1
→0
1
h
1
h
2
_
ψ
ψ−h
2
_
γ −β
γ −h
1
−β
dF
x,y
(α, β).
Performing the indicated limits, we ﬁnd that f
z,w
(γ, ψ) = f
x,y
(γ −ψ, ψ); substituting, we
obtain
f
z,w
(γ, ψ) =
_
0.25, 0 < γ −ψ < 2, 0 < ψ < 2
0, otherwise.
When the RVs x and y are jointly continuous, it is usually easier to ﬁnd the joint PDF
f
z,w
than to carry out the integral indicated in (6.20).
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 65
6.4.2 Bivariate PDF Technique
A very important special case of bivariate transformations is when the RVs x and y are jointly
continuous RVs and the mapping determined by g and h is continuous with g and h having
continuous partial derivatives. Let h
1
> 0, h
2
> 0 and deﬁne
I(γ, h
1
, ψ, h
2
) = g
−1
((γ −h
1
, γ ]) ∩ h
−1
((ψ −h
2
, ψ]). (6.21)
Partition I into disjoint regions so that
I(γ, h
1
, ψ, h
2
) =
_
i
I
i
(γ, h
1
, ψ, h
2
). (6.22)
Let (α
i
, β
i
) denote the unique element of
lim
h
1
→0
lim
h
2
→0
I
i
(γ, h
1
, ψ, h
2
).
Then for small h
1
and small h
2
we have
h
1
h
2
f
v,w
(γ, ψ) ≈
i
f
x,y
(α
i
, β
i
)
_
I
i
(γ,h
1
,ψ,h
2
)
dα dβ. (6.23)
Dividing both sides of the above by h
1
h
2
and letting h
1
→ 0 and h
2
→ 0, the approximation
becomes an equality. The result is summarized by the theorem below.
Theorem 6.4.1. Let x and y be jointly continuous RVs with PDF f
x,y
, and let z = g(x, y) and
w = h(x, y) . Let
(α
i
(γ, ψ), β
i
(γ, ψ)), i = 1, 2, . . . , (6.24)
be the distinct solutions to the simultaneous equations
g(α
i
, β
i
) = γ (6.25)
and
h(α
i
, β
i
) = ψ. (6.26)
Deﬁne the Jacobian
J (α, β) =
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂g(α, β)
∂α
∂g(α, β)
∂β
∂h(α, β)
∂α
∂h(α, β)
∂β
¸
¸
¸
¸
¸
¸
¸
¸
¸
, (6.27)
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
66 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
where the indicated partial derivatives are assumed to exist. Then
f
z,w
(γ, ψ) =
∞
i =1
f
x,y
(α
i
, β
i
)
¸
¸
J (α
i
, β
i
)
¸
¸
. (6.28)
The most difﬁcult aspects to ﬁnding f
z,w
using the joint PDF technique are solving the si
multaneous equations γ = g(α
i
, β
i
) and ψ = h(α
i
, β
i
) for α
i
and β
i
, and determining the
support region for f
z,w
. Let us consider several examples to illustrate the mechanics of this
PDF technique.
Example 6.4.2. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
1/4, 0 < α < 2, 0 < β < 2
0, otherwise.
Find the joint PDF for z = g(x, y) = x + y and w = h(x, y) = y.
Solution. There is only one solution to γ = α +β and ψ = β: α
1
= γ −ψ and β
1
= ψ. The
Jacobian of the transformation is
J =
¸
¸
¸
¸
¸
1 1
0 1
¸
¸
¸
¸
¸
= 1 · 1 −1 · 0 = 1.
Applying Theorem 1, we ﬁnd
f
z,w
(γ, ψ) =
f
x,y
(α
1
, β
1
)
 J 
= f
x,y
(γ −ψ, ψ).
Substituting, we obtain
f
z,w
(γ, ψ) =
_
1/4, 0 < γ −ψ < 2, 0 < ψ < 2
0, otherwise.
The region of support for f
z,w
is illustrated in Figure 6.8.
Example 6.4.3. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
12αβ(1 −α), 0 < α < 1, 0 < β < 1
0, otherwise.
Find the joint PDF for z = g(x, y) = x
2
y and w = h(x, y) = y.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 67
ψ
1 2 3 4
γ 0
1
2
FIGURE6.8: Support region for Example 6.4.2.
Solution. Solving γ = α
2
β and ψ = β, we ﬁnd α
1
=
√
γ/ψ and β
1
= ψ. The other solution
α
2
= −α
1
is not needed since f
x,y
(α, β) = 0 for α < 0. The Jacobian is
J
1
=
¸
¸
¸
¸
¸
2α
1
β
1
α
2
1
0 1
¸
¸
¸
¸
¸
= 2α
1
β
1
= 2
_
γ ψ.
Applying Theorem 1, we ﬁnd
f
z,w
(γ, ψ) =
f
x,y
(
√
γ/ψ, ψ)
2
√
γ ψ
=
_
6(1 −
√
γ/ψ), 0 <
√
γ/ψ < 1, 0 < ψ < 1
0, otherwise.
Example 6.4.4. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
0.25(α +1), −1 < α < 1, −1 < β < 1
0, otherwise.
Find the joint PDF for z = g(x, y) = xy and w = h(x, y) = y/x.
Solution. Solving γ = αβ and ψ = β/α, we ﬁnd α = γ/β = β/ψ, so that β
2
= γ ψ. Let
ting β
1
=
√
γ ψ we have α
1
=
√
γ/ψ. Then β
2
= −
√
γ ψ and α
2
= −
√
γ/ψ. Note that the
solution (α
1
, β
1
) is in the ﬁrst quadrant of the α −β plane; the solution (α
2
, β
2
) is in the third
quadrant of the α −β plane. We ﬁnd
J =
¸
¸
¸
¸
¸
β α
−β/α
2
1/α
¸
¸
¸
¸
¸
= 2
β
α
,
so that J
1
= 2ψ = J
2
. Hence
f
z,w
(γ, ψ) =
f
x,y
(
√
γ/ψ,
√
γ ψ)
2ψ
+
f
x,y
(−
√
γ/ψ, −
√
γ ψ)
2ψ
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
68 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
In the γ −ψ plane,
f
x,y
(
√
γ/ψ,
√
γ ψ)
2ψ
has support region speciﬁed by
0 <
_
γ/ψ < 1 and 0 <
_
γ ψ < 1,
or
0 <
γ
ψ
< 1 and 0 < γ ψ < 1.
Similarly,
f
x,y
(−
√
γ/ψ, −
√
γ ψ)
2ψ
has support region speciﬁed by
−1 < −
_
γ/ψ < 0 and −1 < −
_
γ ψ < 0,
or
0 <
γ
ψ
< 1 and 0 < γ ψ < 1.
Consequently, the two components of the PDF f
z,w
have identical regions of support in
the γ −ψ plane. In the ﬁrst quadrant of the γ −ψ plane, this support region is easily seen to be
0 < γ < ψ < γ
−1
. Similarly, in the third quadrant, the support region is γ
−1
< ψ < γ < 0.
This support region is illustrated in Figure 6.9. Finally, we ﬁnd
f
z,w
(γ, ψ) =
_
1/(4ψ), 0 < γ < ψ < γ
−1
, or γ
−1
< ψ < γ < 0
0, otherwise.
Auxiliary RandomVariables
The joint PDF technique can also be used to transform random variables x and y with joint
PDF f
x,y
to randomvariable z = g(x, y) by introducing auxiliary randomvariable w = h(x, y)
to ﬁnd f
z,w
and then ﬁnding the marginal PDF f
z
using
f
z
(γ ) =
_
∞
−∞
f
z,w
(γ, ψ)dψ. (6.29)
It is usually advisable to let the auxiliary random variable w equal a quantity which allows
a convenient solution of the Jacobian and/or the inverse equations. This method is an alternative
to the CDF technique presented earlier.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 69
3
2
1
−1
−2
−3
−3 −2 −1
1 2 3
ψ
γ
FIGURE6.9: Support region for Example 6.4.4.
Example 6.4.5. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
4αβ, 0 < α < 1, 0 < β < 1
0, otherwise.
Find the PDF for z = g(x, y) = x
2
.
Solution. Let auxiliary variable w = h(x, y) = y. Solving γ = α
2
and ψ = β, we ﬁnd α =
±
√
γ and β = ψ. The only solution inside the support region for f
x,y
is α =
√
γ and β = ψ.
The Jacobian of the transformation is
J =
¸
¸
¸
¸
¸
2α 0
0 1
¸
¸
¸
¸
¸
= 2α,
so that
f
z,w
(γ, ψ) =
f
x,y
(
√
γ , ψ)
2
√
γ
=
⎧
⎨
⎩
4
√
γ ψ
2
√
γ
= 2ψ, 0 < γ < 1, 0 < ψ < 1
0, otherwise.
We ﬁnd the marginal PDF for z as
f
z
(γ ) =
_
∞
−∞
f
z,w
(γ, ψ)dψ =
_
1
0
2ψ dψ.
for 0 < γ < 1 and f
z
(γ ) = 0, otherwise.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
70 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
1 2 0
1
γ
ψ
FIGURE6.10: Support region for Example 6.4.6.
Example 6.4.6. Random variables x and y have the joint PDF
f
x,y
(α, β) =
_
4αβ, 0 < α < 1, 0 < β < 1
0, otherwise.
Find the PDF for z = g(x, y) = x + y.
Solution. Let auxiliary variable w = h(x, y) = y. Solving γ = α +β and ψ = β, we ﬁnd
α = γ −ψ and β = ψ. We ﬁnd
J =
¸
¸
¸
¸
¸
1 1
0 1
¸
¸
¸
¸
¸
= 1,
so that
f
z,w
(γ, ψ) = f
x,y
(γ −ψ, ψ) =
_
4(γ −ψ)ψ, 0 < γ −ψ < 1, 0 < ψ < 1
0, otherwise.
The support region for f
z,w
is shown in Figure 6.10. Referring to Figure 6.10, for 0 < γ < 1,
f
z
(γ ) =
_
γ
0
4(γ −ψ)ψ dψ =
2
3
γ
3
.
For 1 < γ < 2,
f
z
(γ ) =
_
1
γ −1
4(γ −ψ)ψ dψ = −
2
3
γ
3
+4γ −
8
3
.
Otherwise, f
z
= 0.
Conditional PDF Technique
Since a conditional PDF is also a PDF, the above techniques apply to ﬁnd the conditional PDF
for z = g(x, y), given event A. Consider the problemwhere randomvariables x and y have joint
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 71
PDF f
x,y
, and we wish to evaluate the conditional PDF for random variable z = g(x, y), given
that event A occurred. Depending on whether the event A is deﬁned on the range or domain of
z = g(x, y), one of the following two methods may be used to determine the conditional PDF
of z using the bivariate PDF technique.
i. If A is an event deﬁned for an interval on z, the conditional PDF, f
z A
, is computed by ﬁrst
evaluating f
z
using the technique in this section. Then, by the deﬁnition of a conditional
PDF, we have
f
z A
(γ  A) =
f
z
(γ )
P(A)
, γ ∈ A. (6.30)
ii. If A is an event deﬁned for a region in the x − y plane, we will use the conditional PDF
f
x,y A
to evaluate the conditional PDF f
z A
as follows. First, introduce an auxiliary random
variable w = h(x, y), and evaluate
f
z,w A
(γ, ψ A) =
∞
i =1
f
x,y A
(α
i
(γ, ψ), β
i
(γ, ψ) A)
 J (α
i
(γ, ψ), β
i
(γ, ψ))
, (6.31)
where (α
i
, β
i
), i = 1, 2, . . ., are the solutions to γ = g(α, β) and ψ = h(α, β) in region
A. Then evaluate the marginal conditional PDF
f
z A
(γ  A) =
_
∞
−∞
f
z,w A
(γ, ψ A)dψ. (6.32)
Example 6.4.7. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
3α, 0 < β < α < 1
0, otherwise.
Find the PDF for z = x + y, given event A = {max(x, y) ≤ 0.5}.
Solution. We begin by ﬁnding the conditional PDF for x and y, given A. From Figure 6.11(a)
we ﬁnd
P(A) =
_
A
_
f
x,y
(α, β)dα dβ =
_
0.5
0
_
α
0
3α dβ dα =
1
8
;
consequently,
f
x,y A
(α, β A) =
⎧
⎨
⎩
f
x,y
(α, β)
P(A)
= 24α, 0 < β < α < 0.5
0, otherwise.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
72 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
FIGURE6.11: Support region for Example 6.4.7.
Letting the auxiliary random variable w = h(x, y) = y, the only solution to γ = g(α, β) =
α +β and ψ = h(α, β) = β is α = γ −ψ and β = ψ. The Jacobian of the transformation is
J = 1 so that
f
z,w A
(γ, ψ A) =
f
x,y A
(γ −ψ, ψ A)
1
.
Substituting, we ﬁnd
f
z,w A
(γ, ψ A) =
_
24(γ −ψ), 0 < ψ < γ −ψ < 0.5
0, otherwise.
The support region for f
z,w A
is illustrated in Figure 6.11(b). The conditional PDF for z, given
A, can be found from
f
z A
(γ  A) =
_
∞
−∞
f
z,w A
(γ, ψ A)dψ.
The integration is easily carried out with the aid of Figure 6.11(b). For 0.5 < γ < 1 we have
f
z A
(γ  A) =
_
0.5γ
0
24(γ −ψ)dψ = 9γ
2
.
For 0.5 < γ < 1 we obtain
f
z A
(γ  A) =
_
0.5γ
γ −0.5
24(γ −ψ)dψ = 3(1 −γ
2
).
For γ < 0 or γ > 1, f
z A
(γ  A) = 0.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 73
Drill Problem6.4.1. Random variables x and y have joint PDF
f
x,y
(α, β) = e
−α−β
u(α)u(β).
Random variable z = x and w = xy. Find: (a) f
z,w
(1, 1), (b) f
z,w
(−1, 1).
Answers: 0, e
−2
.
Drill Problem6.4.2. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
4αβ, 0 < α < 1, 0 < β < 1
0, otherwise.
Random variable z = x + y and w = y
2
. Determine: (a) f
z,w
(1, 1/4), (b) f
z
(−1/2), (c ) f
z
(1/2),
(d) f
z
(3/2).
Answers: 0, 1/12, 1, 13/12.
Drill Problem6.4.3. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
1.2(α
2
+β), 0 < α < 1, 0 < β < 1
0, otherwise.
Random variable z = x
2
y. Determine: (a) f
z
(−1), (b) f
z
(1/4), (c ) f
z
(1/2), (d) f
z
(3/2).
Answers: 0.7172, 0, 0, 1.3.
Drill Problem6.4.4. Random variables x and y have joint PDF
f
x,y
(α, β) =
_
2β, 0 < α < 1, 0 < β < 1
0, otherwise.
Random variable z = x − y and event A = {(x, y) : x + y ≤ 1}. Determine: (a) f
z A
(−3/
2 A), (b) f
z A
(−1/2 A), (c )F
z A
(0 A), (d) f
z A
(1/2 A).
Answers: 0, 3/4, 15/16, 3/16.
6.5 SUMMARY
This chapter presented a number of different approaches to ﬁnd the probability distribution of
functions of random variables.
Two methods are presentedto ﬁndthe probability distributionof a functionof one random
variable, z = g(x). The ﬁrst, and the most general approach, is called the CDF technique. From
the deﬁnition of the CDF, we write
F
z
(γ ) = P(z ≤ γ ) = P(g(x) ≤ γ ) = P(x ∈ A(γ )), (6.33)
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
74 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
where A(γ ) = {x : g(x) ≤ γ }. By partitioning A(γ ) into disjoint intervals, the CDF F
z
may be
found using the CDF F
x
. The second approach, called the PDF technique, involves evaluating
f
z
(γ ) =
∞
i =1
f
x
(α
i
(γ ))
¸
¸
g
(1)
(α
i
(γ ))
¸
¸
, (6.34)
where
α
i
= α
i
(γ ) = g
−1
(γ ), i = 1, 2, . . . , (6.35)
denote the distinct solutions to g(α
i
) = γ . Typically, the PDF technique is much simpler to
use than the CDF technique. However, the PDF technique is applicable only when z = g(x)
is continuous and does not equal a constant in any interval in which f
x
is nonzero.
Next, we evaluated the probability distribution of a random variable z = g(x, y) created
from jointly distributed random variables x and y using two approaches. The ﬁrst approach, a
CDF technique, involves evaluating
F
z
(γ ) = P(z ≤ γ ) = P(g(x, y) ≤ γ ) = P((x, y) ∈ A(γ )), (6.36)
where
A(γ ) = {(x, y) : g(x, y) ≤ γ }. (6.37)
The CDF for the RV z can then be found by evaluating the integral
F
z
(γ ) =
_
A(γ )
dF
x,y
(α, β). (6.38)
The ease of solution here involves transforming A(γ ) into proper limits of integration. We wish
to remind the reader that the special case of a convolution integral is obtained when random
variables x and y are independent and z = x + y. The second approach involves introducing
an auxiliary random variable and using the PDF technique applied to two functions of two
random variables.
To ﬁnd the joint probability distribution of randomvariables z = g(x, y) and w = h(x, y)
from jointly distributed random variables x and y, a bivariate CDF as well as a joint PDF
technique were presented. Using the joint PDF technique, the joint PDF for z and w can be
found as
f
z,w
(γ, ψ) =
∞
i =1
f
x,y
(α
i
, β
i
)
¸
¸
J (α
i
, β
i
)
¸
¸
, (6.39)
where
(α
i
(γ, ψ), β
i
(γ, ψ)), i = 1, 2, . . . , (6.40)
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 75
are the distinct solutions to the simultaneous equations
g(α
i
, β
i
) = γ (6.41)
and
h(α
i
, β
i
) = ψ. (6.42)
The Jacobian
J (α, β) =
¸
¸
¸
¸
¸
¸
¸
¸
∂g(α, β)
∂α
∂g(α, β)
∂β
∂h(α, β)
∂α
∂h(α, β)
∂β
¸
¸
¸
¸
¸
¸
¸
¸
, (6.43)
where the indicated partial derivatives are assumed to exist. Typically, the most difﬁcult aspect
of the joint PDF technique is in solving the simultaneous equations. The joint PDF technique
can also be used to ﬁnd the probability distribution for z = g(x, y) from f
x,y
by introducing an
auxiliary random variable w = h(x, y) to ﬁnd f
z,w
, and then integrating to obtain the marginal
PDF f
z
.
6.6 PROBLEMS
1. Let random variable x be uniform between 0 and 2 with z = exp(x). Find F
z
using the
CDF technique.
2. Given
f
x
(α) =
_
0.5(1 +α), −1 < α < 1
0, otherwise,
and
z =
_
x −1, x < 0
x +1, x > 0.
Use the CDF technique to ﬁnd F
z
.
3. Suppose random variable x has the CDF
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < 0
α
2
, 0 ≤ α < 1
1, 1 ≤ α
and z = exp(−x). Using the CDF technique, ﬁnd F
z
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
76 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
4. The PDF of random variable x is
f
x
(α) =
1
2
e
−
1
2
(α+2)
u(α +2).
Find F
z
using the CDF technique when z = x.
5. Suppose random variable x has the PDF
f
x
(α) =
_
2α, 0 < α < 1
0, otherwise.
Random variable z is deﬁned by
z =
⎧
⎪
⎨
⎪
⎩
−1, x ≤ −1
x, −1 < x < 1
1, x ≥ 1.
Using the CDF method, determine F
z
.
6. Random variable x represents the input to a halfwave rectiﬁer and z represents the
output, so that z = u(x). Given that x is uniformly distributed between −2 and 2, ﬁnd:
(a) E(z) using f
x
, (b) F
z
using the CDF method, (c) f
z
, (d) E(z) using f
z
(compare
with the answer to part a).
7. Random variable x represents the input to a fullwave rectiﬁer and z represents the
output, so that z = x. Given that x is uniformly distributed between −2 and 2, ﬁnd:
(a) E(z) using f
x
, (b) F
z
using the CDF method, (c) f
z
, (d) E(z) using f
z
(compare
with the answer to part a).
8. Random variable x has the PDF f
x
(α) = e
−α
u(α). Find F
z
for z = x
2
using the CDF
method.
9. Given
f
x
(α) =
a
1 +α
2
and
z =
⎧
⎪
⎨
⎪
⎩
−1, x < −1
x, −1 ≤ x ≤ 2
2, x > 2.
Determine: (a) a, (b) F
z
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 77
10. Random variables x and z are the input and output of a quantizer. The relationship
between them is deﬁned by:
z =
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
0, x < 0.5
1, 0.5 ≤ x < 1.5
2, 1.5 ≤ x < 2.5
3, 2.5 ≤ x < 3.5
4, x ≥ 3.5.
Given that the input follows a Gaussian distribution with x ∼ G(2.25, 0.49), ﬁnd f
z
using the CDF method.
11. Random variable x has the PDF f
x
(α) = e
−α
u(α). With z = 100 −25x, ﬁnd: (a) F
z
using the CDF technique, (b) F
z
using the PDF technique.
12. Random variable x has the following PDF
f
x
(α) =
_
4α
3
, 0 < α < 1
0, otherwise.
Find the PDFs for the following random variables: (a) z = x
3
, (b) z = (x −1/2)
2
, (c)
z = (x −1/4)
2
.
13. Random variable x has the CDF
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < −1
(α
2
+2α +1)/9, −1 ≤ α < 2
1, 2 ≤ α
and
z =
⎧
⎪
⎨
⎪
⎩
x, x < −0.5
0, −0.5 ≤ x ≤ 0
x, 0 < x.
Determine F
z
.
14. Suppose
f
x
(α) =
_
(1 +α
2
)/6, −1 < α < 2
0, otherwise.
Let z = 1/x
2
. Use the CDF technique to determine F
z
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
78 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
z g( x ) =
x
3
1
−1
2 1 −1 −2
FIGURE6.12: Plot for Problem 15.
15. Random variable x has the CDF
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < −1
0.25(α +1), −1 ≤ α < 3
1, 3 ≤ α.
Find the CDF F
z
, with RV z = g(x), and g(x) shown in Figure 6.12. Assume that
g(x) is a second degree polynomial for x ≥ 0.
16. The voltage x in Figure 6.13 is a random variable which is uniformly distributed from
−1 to 2. Find the PDF f
z
. Assume the diode is ideal.
17. The voltage x in Figure 6.14 is a Gaussian random variable with mean η
x
= 0 and
standard deviation σ
x
= 3. Find the PDF f
z
. Assume the diodes are ideal.
18. The voltage x in Figure 6.15 is a Gaussian random variable with mean η
x
= 1 and
standard deviation σ
x
= 1. Find the PDF f
z
. Assume the diode and the operational
ampliﬁer are ideal.
19. Randomvariable x is Gaussian with mean η
x
= 0 and standard deviation σ
x
= 1. With
z = g(x) shown in Figure 6.16, ﬁnd the PDF f
z
.
+
z
−
+
x
−
3KΩ
1K Ω
FIGURE6.13: Circuit for Problem 16.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 79
+
z
−
+
x
−
R
2 V 3 V
FIGURE6.14: Circuit for Problem 17.
20. Random variable x has the following PDF
f
x
(α) =
_
α +0.5δ(α −0.5), 0 < α < 1
0, otherwise.
Find F
z
if z = x
2
.
21. Random variable x is uniform in the interval 0 to 12. Random variable z = 4x +2.
Find f
z
using the PDF technique.
22. Find f
z
if z = 1/x
2
and x is uniform on −1 to 2.
23. Random variable x has the PDF
f
x
(α) =
_
2(α +1)/9, −1 < α < 2
0, otherwise.
Find the PDF of z = 2x
2
using the PDF technique.
24. Let z = cos(x). Find f
z
if:
(a) f
x
(α) =
_
1/π, α < π/2
0, otherwise.
1K Ω
5K Ω
x
z
FIGURE6.15: Circuit for Problem 18.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
80 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
−2 −1 1
x
1.4
z g( x ) =
2
−0.5
FIGURE6.16: Transformation for Problem 19.
(b) f
x
(α) =
_
8α/π
2
, 0 < α < π/2
0, otherwise.
25. Given that x has the CDF
F
x
(α) =
⎧
⎪
⎨
⎪
⎩
0, α < 0
α, 0 ≤ α < 1
1, 1 ≤ α.
Find the PDF of z = −2 ln(x) using the PDF technique.
26. Random variable x has the PDF
f
x
(α) =
_
2α/9, 0 < α < 3
0, otherwise.
Random variable z = (x −1)
2
and event A = {x : x ≥ 1/2}. Find the PDF of random
variable z, given event A.
27. Random variable x is uniform between −1 and 1. Random variable
z =
_
x
2
, x < 0
x, x ≥ 0.
Using the PDF technique, ﬁnd f
z
.
28. A voltage v is a Gaussian random variable with η
v
= 0 and σ
v
= 2. Random variable
w = v
2
/R represents the power dissipated in a resistor of R with v volts across the
resistor. Find (a) f
w
, (b) f
w A
if A = {v ≥ 0}.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 81
29. Random variable x has an exponential distribution with mean one. If z = e
−x
, use
the PDF technique to determine: (a) f
z A
(γ  A) if A = {x : x > 2}, (b) f
zB
(γ B) if
B = {z : z < 1/2}.
30. Find f
z
if z = 1/x and
f
x
(α) =
1
π(α
2
+1)
.
31. Given that random variable x has the PDF
f
x
(α) =
⎧
⎪
⎨
⎪
⎩
α, 0 < α < 1
2 −α, 1 < α < 2
0, otherwise.
Using the PDF technique, ﬁnd the PDF of z = x
2
.
32. Suppose
f
x
(α) =
_
2α, 0 < α < 1
0, otherwise,
and z = 8x
3
. Determine f
z
using the PDF technique.
33. Given
f
x
(α) =
⎧
⎪
⎨
⎪
⎩
9α
2
, 0 < α < 0.5
3(1 −α
2
), 0.5 < α < 1
0, otherwise,
and z = −2 ln(x). Determine: (a) f
z
, (b) F
z
, (c) E(z), (d) E(z
2
).
34. Using the PDF technique, ﬁnd f
z
if z = x and x is a standard Gaussian random
variable.
35. Suppose RV x has PDF f
x
(α) = 0.5α(u(α) −u(α −2)). Find a transformation g such
that z = g(x) has PDF
f
z
(γ ) = c γ
2
(u(γ ) −u(γ −1)).
36. Let
f
x
(α) =
_
0.75(1 −α
2
), −1 < α < 1
0, otherwise,
and z = x
2
. Determine f
z
using the PDF technique.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
82 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
37. Find f
z
if z = sin(x) and
f
x
(α) =
_
1/(2π), 0 ≤ α < 2π
0, otherwise.
38. Random variable x has the PDF
f
x
(α) =
⎧
⎪
⎨
⎪
⎩
0.5, −1 < α < 0
0.5 −0.25α, 0 < α < 2
0, otherwise.
(a) Find the transformation z = g(x) so that
f
z
(γ ) =
⎧
⎪
⎨
⎪
⎩
1 −0.25γ, 0 < γ < 1
0.5 −0.25γ, 1 < γ < 2
0, otherwise.
(b) Determine f
z
if z = (2x +2)u(x).
39. Let random variable x have the PDF
f
x
(α) =
_
1/12, −2 < α < 1
1/4, 1 < α < 2;
in addition, P(x = −2) = P(x = 2) = 0.25. If z = 1/x, ﬁnd f
z
.
40. Random variable x has PDF
f
x
(α) =
1
4
δ(α +1) +
1
4
δ(α) +(u(α) −u(α −2)).
Random variable z = g(x), with g shown in Figure 6.17. Find the PDF f
z
.
41. Random variable x has PDF
f
x
(α) =
α
2
3
(u(α +1) −u(α −2)).
−1 1
x
2
z = g(x)
2
1
0
FIGURE6.17: Transformation for Problem 40.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 83
−1 1
2
z = g( x)
2
1
0
x
FIGURE6.18: Transformation for Problem 41.
Random variable z = g(x), with g shown in Figure 6.18. Use the PDF technique to
ﬁnd: (a) f
z
, (b) f
z A
with A = {x > 0}.
42. The PDF for random variable x is f
x
(α) = e
−α
u(α). With z = e
−x
, determine: (a)
f
z A
(γ  A) where A = {x : x > 2}, (b) f
zB
(γ B) where B = {z : z < 1/2}.
43. The joint PDF of x and y is
f
x,y
(α, β) =
_
e
−α
, 0 < β < α
0, otherwise.
With z = x + y, write an expression(s) for F
z
(γ ) (do not solve, just write the integral(s)
necessary to ﬁnd F
z
.
44. If z = x/y, ﬁnd F
z
when x and y have the joint PDF
f
x,y
(α, β) =
_
0.25α, 0 < β < α < 1
0, otherwise.
45. Resistors R
1
and R
2
have values r
1
and r
2
which are independent RVs uniformly
distributed between 1 and 2 . With r denoting the equivalent resistance of a series
connection of R
1
and R
2
, ﬁnd the PDF f
r
using convolution.
46. Resistors R
1
and R
2
have values r
1
and r
2
which are independent RVs uniformly
distributed between 1 and 2 . With g denoting the equivalent conductance of a
parallel connection of R
1
and R
2
, ﬁnd the PDF f
g
using convolution.
47. Suppose the voltage v across a resistor is a random variable with PDF
f
v
(α) =
_
6α(1 −α), 0 < α < 1
0, otherwise,
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
84 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
and that resistance r is a random variable with PDF
f
r
(β) =
_
1/12, 94 < β < 106
0, otherwise.
Moreover, suppose that v and r are independent random variables. Find the PDF for
power, p = v
2
/r .
48. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
α +β, 0 < α < 1, 0 < β < 1
0, otherwise.
Let z = 2x + y and w = x +2y. Find f
z,w
.
49. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
4αβ, 0 < α < 1, 0 < β < 1
0, otherwise.
Let z = x
2
and w = xy. Find: (a) f
z,w
, (b) f
z
.
50. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
3(α
2
+β
2
), 0 < β < α < 1
0, otherwise.
Find the joint PDF for random variables z = x + y and w = x − y.
51. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
0.5βe
−α
, 0 < α, 0 < β < 2
0, otherwise.
With z = y/x, ﬁnd f
z
.
52. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
8αβ, 0 < α
2
+β
2
< 1, 0 < α, 0 < β
0, otherwise.
Let A = {(x, y) : x > y}, z = x and w = x
2
+ y
2
. Find: (a) f
z,w A
, (b) f
z A
.
53. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
1/α, 0 < β < α < 1
0, otherwise.
Let z = x/y and w = y. Find f
z,w
.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
TRANSFORMATIONS OF RANDOMVARIABLES 85
54. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
α sin(β), 0 < α < 1, 0 < β < π
0, otherwise.
Let z = x
2
and w = cos(y). Find f
z,w
.
55. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
12αβ(1 −α), 0 < α < 1, 0 < β < 1
0, otherwise.
Let z = x
2
y. Find f
z
.
56. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
e
−α−β
, 0 < α, 0 < β
0, otherwise.
Let z = x + y and w = x/(x + y). Find: (a) f
z,w
, (b) f
z
.
57. The joint PDF for random variables x and y is
f
x,y
(α, β) =
_
0.25(α +β), α < 1, β < 1
0, otherwise.
Let z = xy and w = y/x. Find f
z,w
using the joint CDF technique.
P1: IML/FFX P2: IML
MOBK04206 MOBK042Enderle.cls October 30, 2006 19:53
86
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
87
A P P E N D I X A
Distribution Tables
TABLEA.1: Bernoulli CDF for n = 5 and n = 10
n = 5
p k +0 +1 +2 +3 +4
0.05 0 0.773781 0.977408 0.998842 0.999970 1.000000
0.1 0 0.590490 0.918540 0.991440 0.999540 0.999990
0.15 0 0.443705 0.835210 0.973388 0.997773 0.999924
0.2 0 0.327680 0.737280 0.942080 0.993280 0.999680
0.25 0 0.237305 0.632813 0.896484 0.984375 0.999023
0.3 0 0.168070 0.528220 0.836920 0.969220 0.997570
0.35 0 0.116029 0.428415 0.764831 0.945978 0.994748
0.4 0 0.077760 0.336960 0.682560 0.912960 0.989760
0.45 0 0.050328 0.256218 0.593127 0.868780 0.981547
0.5 0 0.031250 0.187500 0.500000 0.812500 0.968750
n = 10
p k +0 +1 +2 +3 +4
0.05 0 0.598737 0.913862 0.988496 0.998971 0.999936
5 0.999997 1.000000 1.000000 1.000000 1.000000
0.1 0 0.348678 0.736099 0.929809 0.987205 0.998365
5 0.999853 0.999991 1.000000 1.000000 1.000000
0.15 0 0.196874 0.544300 0.820197 0.950030 0.990126
5 0.998617 0.999865 0.999991 1.000000 1.000000
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
88 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.1: Bernoulli CDF for n = 5 and n = 10 (Continued)
n = 10
p k +0 +1 +2 +3 +4
0.2 0 0.107374 0.375810 0.677800 0.879126 0.967206
5 0.993631 0.999136 0.999922 0.999996 1.000000
0.25 0 0.056314 0.244025 0.525593 0.775875 0.921873
5 0.980272 0.996494 0.999584 0.999970 0.999999
0.3 0 0.028248 0.149308 0.382783 0.649611 0.849732
5 0.952651 0.989408 0.998410 0.999856 0.999994
0.35 0 0.013463 0.085954 0.261607 0.513827 0.751495
5 0.905066 0.973976 0.995179 0.999460 0.999972
0.4 0 0.006047 0.046357 0.167290 0.382281 0.633103
5 0.833761 0.945238 0.987705 0.998322 0.999895
0.45 0 0.002533 0.023257 0.099560 0.266038 0.504405
5 0.738437 0.898005 0.972608 0.995498 0.999659
0.5 0 0.000977 0.010742 0.054688 0.171875 0.376953
5 0.623047 0.828125 0.945313 0.989258 0.999023
TABLEA.2: Bernoulli CDF for n = 15
n = 15
p k +0 +1 +2 +3 +4
0.05 0 0.463291 0.829047 0.963800 0.994533 0.999385
5 0.999947 0.999996 1.000000 1.000000 1.000000
10 1.000000 1.000000 1.000000 1.000000 1.000000
0.1 0 0.205891 0.549043 0.815939 0.944444 0.987280
5 0.997750 0.999689 0.999966 0.999997 1.000000
10 1.000000 1.000000 1.000000 1.000000 1.000000
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 89
TABLEA.2: Bernoulli CDF for n = 15 (Continued)
n = 15
p k +0 +1 +2 +3 +4
0.15 0 0.087354 0.318586 0.604225 0.822655 0.938295
5 0.983190 0.996394 0.999390 0.999919 0.999992
10 0.999999 1.000000 1.000000 1.000000 1.000000
0.2 0 0.035184 0.167126 0.398023 0.648162 0.835766
5 0.938949 0.981941 0.995760 0.999215 0.999887
10 0.999988 0.999999 1.000000 1.000000 1.000000
0.25 0 0.013363 0.080181 0.236088 0.461287 0.686486
5 0.851632 0.943380 0.982700 0.995807 0.999205
10 0.999885 0.999988 0.999999 1.000000 1.000000
0.3 0 0.004748 0.035268 0.126828 0.296868 0.515491
5 0.721621 0.868857 0.949987 0.984757 0.996347
10 0.999328 0.999908 0.999991 1.000000 1.000000
0.35 0 0.001562 0.014179 0.061734 0.172696 0.351943
5 0.564282 0.754842 0.886769 0.957806 0.987557
10 0.997169 0.999521 0.999943 0.999996 1.000000
0.4 0 0.000470 0.005172 0.027114 0.090502 0.217278
5 0.403216 0.609813 0.786897 0.904953 0.966167
10 0.990652 0.998072 0.999721 0.999975 0.999999
0.45 0 0.000127 0.001692 0.010652 0.042421 0.120399
5 0.260760 0.452160 0.653504 0.818240 0.923071
10 0.974534 0.993673 0.998893 0.999879 0.999994
0.5 0 0.000031 0.000488 0.003693 0.017578 0.059235
5 0.150879 0.303619 0.500000 0.696381 0.849121
10 0.940765 0.982422 0.996307 0.999512 0.999969
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
90 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.3: Bernoulli CDF for n = 20
n = 20
p k +0 +1 +2 +3 +4
0.05 0 0.358486 0.735840 0.924516 0.984098 0.997426
5 0.999671 0.999966 0.999997 1.000000 1.000000
10 1.000000 1.000000 1.000000 1.000000 1.000000
15 1.000000 1.000000 1.000000 1.000000 1.000000
0.1 0 0.121577 0.391747 0.676927 0.867047 0.956825
5 0.988747 0.997614 0.999584 0.999940 0.999993
10 0.999999 1.000000 1.000000 1.000000 1.000000
15 1.000000 1.000000 1.000000 1.000000 1.000000
0.15 0 0.038760 0.175558 0.404896 0.647725 0.829847
5 0.932692 0.978065 0.994079 0.998671 0.999752
10 0.999961 0.999995 0.999999 1.000000 1.000000
15 1.000000 1.000000 1.000000 1.000000 1.000000
0.2 0 0.011529 0.069175 0.206085 0.411449 0.629648
5 0.804208 0.913307 0.967857 0.990018 0.997405
10 0.999437 0.999898 0.999985 0.999998 1.000000
15 1.000000 1.000000 1.000000 1.000000 1.000000
0.25 0 0.003171 0.024313 0.091260 0.225156 0.414842
5 0.617173 0.785782 0.898188 0.959075 0.986136
10 0.996058 0.999065 0.999816 0.999970 0.999996
15 1.000000 1.000000 1.000000 1.000000 1.000000
0.3 0 0.000798 0.007637 0.035483 0.107087 0.237508
5 0.416371 0.608010 0.772272 0.886669 0.952038
10 0.982855 0.994862 0.998721 0.999739 0.999957
15 0.999994 0.999999 1.000000 1.000000 1.000000
0.35 0 0.000181 0.002133 0.012118 0.044376 0.118197
5 0.245396 0.416625 0.601027 0.762378 0.878219
10 0.946833 0.980421 0.993985 0.998479 0.999689
15 0.999950 0.999994 0.999999 1.000000 1.000000
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 91
TABLEA.3: Bernoulli CDF for n = 20 (Continued)
n = 20
p k +0 +1 +2 +3 +4
0.4 0 0.000037 0.000524 0.003611 0.015961 0.050952
5 0.125599 0.250011 0.415893 0.595599 0.755337
10 0.872479 0.943474 0.978971 0.993534 0.998388
15 0.999683 0.999953 0.999995 1.000000 1.000000
0.45 0 0.000006 0.000111 0.000927 0.004933 0.018863
5 0.055334 0.129934 0.252006 0.414306 0.591361
10 0.750711 0.869235 0.941966 0.978586 0.993566
15 0.998469 0.999723 0.999964 0.999997 1.000000
0.5 0 0.000001 0.000020 0.000201 0.001288 0.005909
5 0.020695 0.057659 0.131588 0.251722 0.411901
10 0.588099 0.748278 0.868412 0.942341 0.979305
15 0.994091 0.998712 0.999799 0.999980 0.999999
TABLEA.4: Poisson CDF for λ = 0.1, 0.2, . . . , 1, 1.5, 2, . . . , 4.5
λ k +0 +1 +2 +3 +4
0.1 0 0.904837 0.995321 0.999845 0.999996 1.000000
0.2 0 0.818731 0.982477 0.998852 0.999943 0.999998
0.3 0 0.740818 0.963064 0.996400 0.999734 0.999984
0.4 0 0.670320 0.938448 0.992074 0.999224 0.999939
0.5 0 0.606531 0.909796 0.985612 0.998248 0.999828
5 0.999986 0.999999 1.000000 1.000000 1.000000
0.6 0 0.548812 0.878099 0.976885 0.996642 0.999605
5 0.999961 0.999997 1.000000 1.000000 1.000000
0.7 0 0.496585 0.844195 0.965858 0.994247 0.999214
5 0.999910 0.999991 0.999999 1.000000 1.000000
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
92 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.4: Poisson CDF for λ = 0.1, 0.2, . . . , 1, 1.5, 2, . . . , 4.5 (Continued)
λ k +0 +1 +2 +3 +4
0.8 0 0.449329 0.808792 0.952577 0.990920 0.998589
5 0.999816 0.999979 0.999998 1.000000 1.000000
0.9 0 0.406570 0.772482 0.937143 0.986541 0.997656
5 0.999656 0.999957 0.999995 1.000000 1.000000
1 0 0.367879 0.735759 0.919699 0.981012 0.996340
5 0.999406 0.999917 0.999990 0.999999 1.000000
1.5 0 0.223130 0.557825 0.808847 0.934358 0.981424
5 0.995544 0.999074 0.999830 0.999972 0.999996
2 0 0.135335 0.406006 0.676676 0.857123 0.947347
5 0.983436 0.995466 0.998903 0.999763 0.999954
2.5 0 0.082085 0.287297 0.543813 0.757576 0.891178
5 0.957979 0.985813 0.995753 0.998860 0.999723
10 0.999938 0.999987 0.999998 1.000000 1.000000
3 0 0.049787 0.199148 0.423190 0.647232 0.815263
5 0.916082 0.966491 0.988095 0.996197 0.998897
10 0.999708 0.999929 0.999984 0.999997 0.999999
3.5 0 0.030197 0.135888 0.320847 0.536633 0.725445
5 0.857614 0.934712 0.973261 0.990126 0.996685
10 0.998981 0.999711 0.999924 0.999981 0.999996
4 0 0.018316 0.091578 0.238103 0.433470 0.628837
5 0.785130 0.889326 0.948866 0.978637 0.991868
10 0.997160 0.999085 0.999726 0.999924 0.999980
4.5 0 0.011109 0.061099 0.173578 0.342296 0.532104
5 0.702930 0.831051 0.913414 0.959743 0.982907
10 0.993331 0.997596 0.999195 0.999748 0.999926
15 0.999980 0.999995 0.999999 1.000000 1.000000
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 93
TABLEA.5: Poisson CDF for λ = 5, 5.5, . . . , 8.5
λ k +0 +1 +2 +3 +4
5 0 0.006738 0.040428 0.124652 0.265026 0.440493
5 0.615961 0.762183 0.866628 0.931906 0.968172
10 0.986305 0.994547 0.997981 0.999302 0.999774
15 0.999931 0.999980 0.999995 0.999999 1.000000
5.5 0 0.004087 0.026564 0.088376 0.201699 0.357518
5 0.528919 0.686036 0.809485 0.894357 0.946223
10 0.974749 0.989012 0.995549 0.998315 0.999401
15 0.999800 0.999937 0.999981 0.999995 0.999999
6 0 0.002479 0.017351 0.061969 0.151204 0.285057
5 0.445680 0.606303 0.743980 0.847238 0.916076
10 0.957379 0.979908 0.991173 0.996372 0.998600
15 0.999491 0.999825 0.999943 0.999982 0.999995
6.5 0 0.001503 0.011276 0.043036 0.111850 0.223672
5 0.369041 0.526524 0.672758 0.791573 0.877384
10 0.933161 0.966120 0.983973 0.992900 0.997044
15 0.998840 0.999570 0.999849 0.999949 0.999984
7 0 0.000912 0.007295 0.029636 0.081765 0.172992
5 0.300708 0.449711 0.598714 0.729091 0.830496
10 0.901479 0.946650 0.973000 0.987189 0.994283
15 0.997593 0.999042 0.999638 0.999870 0.999956
7.5 0 0.000553 0.004701 0.020257 0.059145 0.132062
5 0.241436 0.378155 0.524639 0.661967 0.776408
10 0.862238 0.920759 0.957334 0.978435 0.989740
15 0.995392 0.998041 0.999210 0.999697 0.999889
20 0.999961 0.999987 0.999996 0.999999 1.000000
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
94 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.5: Poisson CDF for λ = 5, 5.5, . . . , 8.5 (Continued)
λ k +0 +1 +2 +3 +4
8 0 0.000335 0.003019 0.013754 0.042380 0.099632
5 0.191236 0.313374 0.452961 0.592547 0.716624
10 0.815886 0.888076 0.936203 0.965819 0.982743
15 0.991769 0.996282 0.998406 0.999350 0.999747
20 0.999906 0.999967 0.999989 0.999996 0.999999
8.5 0 0.000203 0.001933 0.009283 0.030109 0.074364
5 0.149597 0.256178 0.385597 0.523105 0.652974
10 0.763362 0.848662 0.909083 0.948589 0.972575
15 0.986167 0.993387 0.996998 0.998703 0.999465
20 0.999789 0.999921 0.999971 0.999990 0.999997
TABLEA.6: Poisson CDF for λ = 9, 9.5, 10, 11, 12, 13
λ k +0 +1 +2 +3 +4
9 0 0.000123 0.001234 0.006232 0.021226 0.054964
5 0.115691 0.206781 0.323897 0.455653 0.587408
10 0.705988 0.803008 0.875773 0.926149 0.958534
15 0.977964 0.988894 0.994680 0.997574 0.998944
20 0.999561 0.999825 0.999933 0.999975 0.999991
9.5 0 0.000075 0.000786 0.004164 0.014860 0.040263
5 0.088528 0.164949 0.268663 0.391824 0.521826
10 0.645328 0.751990 0.836430 0.898136 0.940008
15 0.966527 0.982273 0.991072 0.995716 0.998038
20 0.999141 0.999639 0.999855 0.999944 0.999979
10 0 0.000045 0.000499 0.002769 0.010336 0.029253
5 0.067086 0.130141 0.220221 0.332820 0.457930
10 0.583040 0.696776 0.791556 0.864464 0.916542
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 95
TABLEA.6: Poisson CDF for λ = 9, 9.5, 10, 11, 12, 13 (Continued)
λ k +0 +1 +2 +3 +4
15 0.951260 0.972958 0.985722 0.992813 0.996546
20 0.998412 0.999300 0.999704 0.999880 0.999953
11 0 0.000017 0.000200 0.001211 0.004916 0.015105
5 0.037520 0.078614 0.143192 0.231985 0.340511
10 0.459889 0.579267 0.688697 0.781291 0.854044
15 0.907396 0.944076 0.967809 0.982313 0.990711
20 0.995329 0.997748 0.998958 0.999536 0.999801
25 0.999918 0.999967 0.999987 0.999995 0.999998
12 0 0.000006 0.000080 0.000522 0.002292 0.007600
5 0.020341 0.045822 0.089505 0.155028 0.242392
10 0.347229 0.461597 0.575965 0.681536 0.772025
15 0.844416 0.898709 0.937034 0.962584 0.978720
20 0.988402 0.993935 0.996953 0.998527 0.999314
25 0.999692 0.999867 0.999944 0.999977 0.999991
13 0 0.000002 0.000032 0.000223 0.001050 0.003740
5 0.010734 0.025887 0.054028 0.099758 0.165812
10 0.251682 0.353165 0.463105 0.573045 0.675132
15 0.763607 0.835493 0.890465 0.930167 0.957331
20 0.974988 0.985919 0.992378 0.996028 0.998006
25 0.999034 0.999548 0.999796 0.999911 0.999962
TABLEA.7: Poisson CDF for λ = 14, 15, 16, 17, 18
λ k +0 +1 +2 +3 +4
14 0 0.000001 0.000012 0.000094 0.000474 0.001805
5 0.005532 0.014228 0.031620 0.062055 0.109399
10 0.175681 0.260040 0.358458 0.464448 0.570437
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
96 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ k +0 +1 +2 +3 +4
15 0.669360 0.755918 0.827201 0.882643 0.923495
20 0.952092 0.971156 0.983288 0.990672 0.994980
25 0.997392 0.998691 0.999365 0.999702 0.999864
30 0.999940 0.999974 0.999989 0.999996 0.999998
15 0 0.000000 0.000005 0.000039 0.000211 0.000857
5 0.002792 0.007632 0.018002 0.037446 0.069854
10 0.118464 0.184752 0.267611 0.363218 0.465654
15 0.568090 0.664123 0.748859 0.819472 0.875219
20 0.917029 0.946894 0.967256 0.980535 0.988835
25 0.993815 0.996688 0.998284 0.999139 0.999582
30 0.999803 0.999910 0.999960 0.999983 0.999993
16 0 0.000000 0.000002 0.000016 0.000093 0.000400
5 0.001384 0.004006 0.010000 0.021987 0.043298
10 0.077396 0.126993 0.193122 0.274511 0.367527
15 0.466745 0.565962 0.659344 0.742349 0.812249
20 0.868168 0.910773 0.941759 0.963314 0.977685
25 0.986881 0.992541 0.995895 0.997811 0.998869
30 0.999433 0.999724 0.999869 0.999940 0.999973
17 0 0.000000 0.000001 0.000007 0.000041 0.000185
5 0.000675 0.002062 0.005433 0.012596 0.026125
10 0.049124 0.084669 0.135024 0.200873 0.280833
15 0.371454 0.467738 0.564023 0.654958 0.736322
20 0.805481 0.861466 0.904728 0.936704 0.959354
25 0.974755 0.984826 0.991166 0.995016 0.997273
30 0.998552 0.999253 0.999626 0.999817 0.999913
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 97
TABLEA.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ k +0 +1 +2 +3 +4
18 0 0.000000 0.000000 0.000003 0.000018 0.000084
5 0.000324 0.001043 0.002893 0.007056 0.015381
10 0.030366 0.054887 0.091669 0.142598 0.208077
15 0.286653 0.375050 0.468648 0.562245 0.650916
20 0.730720 0.799124 0.855090 0.898890 0.931740
25 0.955392 0.971766 0.982682 0.989700 0.994056
30 0.996669 0.998187 0.999040 0.999506 0.999752
TABLEA.8: Marcum’s Q function for α = 0, 0.01, . . . , 1.99
α +0.00 +0.01 +0.02 +0.03 +0.04
0.00 0.500000 0.496011 0.492022 0.488033 0.484046
0.05 0.480061 0.476078 0.472097 0.468119 0.464144
0.10 0.460172 0.456205 0.452242 0.448283 0.444330
0.15 0.440382 0.436441 0.432505 0.428576 0.424655
0.20 0.420740 0.416834 0.412936 0.409046 0.405165
0.25 0.401294 0.397432 0.393580 0.389739 0.385908
0.30 0.382089 0.378281 0.374484 0.370700 0.366928
0.35 0.363169 0.359424 0.355691 0.351973 0.348268
0.40 0.344578 0.340903 0.337243 0.333598 0.329969
0.45 0.326355 0.322758 0.319178 0.315614 0.312067
0.50 0.308538 0.305026 0.301532 0.298056 0.294598
0.55 0.291160 0.287740 0.284339 0.280957 0.277595
0.60 0.274253 0.270931 0.267629 0.264347 0.261086
0.65 0.257846 0.254627 0.251429 0.248252 0.245097
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
98 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.8: Marcum’s Q function for α = 0, 0.01, . . . , 1.99 (Continued)
α +0.00 +0.01 +0.02 +0.03 +0.04
0.70 0.241964 0.238852 0.235762 0.232695 0.229650
0.75 0.226627 0.223627 0.220650 0.217695 0.214764
0.80 0.211855 0.208970 0.206108 0.203269 0.200454
0.85 0.197662 0.194894 0.192150 0.189430 0.186733
0.90 0.184060 0.181411 0.178786 0.176186 0.173609
0.95 0.171056 0.168528 0.166023 0.163543 0.161087
1.00 0.158655 0.156248 0.153864 0.151505 0.149170
1.05 0.146859 0.144572 0.142310 0.140071 0.137857
1.10 0.135666 0.133500 0.131357 0.129238 0.127143
1.15 0.125072 0.123024 0.121001 0.119000 0.117023
1.20 0.115070 0.113140 0.111233 0.109349 0.107488
1.25 0.105650 0.103835 0.102042 0.100273 0.098525
1.30 0.096801 0.095098 0.093418 0.091759 0.090123
1.35 0.088508 0.086915 0.085344 0.083793 0.082264
1.40 0.080757 0.079270 0.077804 0.076359 0.074934
1.45 0.073529 0.072145 0.070781 0.069437 0.068112
1.50 0.066807 0.065522 0.064256 0.063008 0.061780
1.55 0.060571 0.059380 0.058208 0.057053 0.055917
1.60 0.054799 0.053699 0.052616 0.051551 0.050503
1.65 0.049471 0.048457 0.047460 0.046479 0.045514
1.70 0.044565 0.043633 0.042716 0.041815 0.040929
1.75 0.040059 0.039204 0.038364 0.037538 0.036727
1.80 0.035930 0.035148 0.034379 0.033625 0.032884
1.85 0.032157 0.031443 0.030742 0.030054 0.029379
1.90 0.028716 0.028067 0.027429 0.026803 0.026190
1.95 0.025588 0.024998 0.024419 0.023852 0.023295
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
DISTRIBUTIONTABLES 99
TABLEA.9: Marcum’s Q function for α = 3, 3.01, . . . , 3.99
α +0.00 +0.01 +0.02 +0.03 +0.04
2.00 0.022750 0.022216 0.021692 0.021178 0.020675
2.05 0.020182 0.019699 0.019226 0.018763 0.018309
2.10 0.017864 0.017429 0.017003 0.016586 0.016177
2.15 0.015778 0.015386 0.015003 0.014629 0.014262
2.20 0.013903 0.013553 0.013209 0.012874 0.012545
2.25 0.012224 0.011911 0.011604 0.011304 0.011011
2.30 0.010724 0.010444 0.010170 0.009903 0.009642
2.35 0.009387 0.009137 0.008894 0.008656 0.008424
2.40 0.008198 0.007976 0.007760 0.007549 0.007344
2.45 0.007143 0.006947 0.006756 0.006569 0.006387
2.50 0.006210 0.006037 0.005868 0.005703 0.005543
2.55 0.005386 0.005234 0.005085 0.004940 0.004799
2.60 0.004661 0.004527 0.004397 0.004269 0.004145
2.65 0.004025 0.003907 0.003793 0.003681 0.003573
2.70 0.003467 0.003364 0.003264 0.003167 0.003072
2.75 0.002980 0.002890 0.002803 0.002718 0.002635
2.80 0.002555 0.002477 0.002401 0.002327 0.002256
2.85 0.002186 0.002118 0.002052 0.001988 0.001926
2.90 0.001866 0.001807 0.001750 0.001695 0.001641
2.95 0.001589 0.001538 0.001489 0.001441 0.001395
3.00 0.001350 0.001306 0.001264 0.001223 0.001183
3.05 0.001144 0.001107 0.001070 0.001035 0.001001
3.10 0.000968 0.000936 0.000904 0.000874 0.000845
3.15 0.000816 0.000789 0.000762 0.000736 0.000711
3.20 0.000687 0.000664 0.000641 0.000619 0.000598
3.25 0.000577 0.000557 0.000538 0.000519 0.000501
3.30 0.000483 0.000467 0.000450 0.000434 0.000419
(Continued )
P1: IML/FFX P2: IML
MOBK042Appe MOBK042Enderle.cls October 30, 2006 19:54
100 ADVANCEDPROBABILITY THEORY FORBIOMEDICAL ENGINEERS
TABLEA.9: Marcum’s Q function for α = 3, 3.01, . . . , 3.99 (Continued)
α +0.00 +0.01 +0.02 +0.03 +0.04
3.35 0.000404 0.000390 0.000376 0.000362 0.000350
3.40 0.000337 0.000325 0.000313 0.000302 0.000291
3.45 0.000280 0.000270 0.000260 0.000251 0.000242
3.50 0.000233 0.000224 0.000216 0.000208 0.000200
3.55 0.000193 0.000185 0.000179 0.000172 0.000165
3.60 0.000159 0.000153 0.000147 0.000142 0.000136
3.65 0.000131 0.000126 0.000121 0.000117 0.000112
3.70 0.000108 0.000104 0.000100 0.000096 0.000092
3.75 0.000088 0.000085 0.000082 0.000078 0.000075
3.80 0.000072 0.000070 0.000067 0.000064 0.000062
3.85 0.000059 0.000057 0.000054 0.000052 0.000050
3.90 0.000048 0.000046 0.000044 0.000042 0.000041
3.95 0.000039 0.000037 0.000036 0.000034 0.000033
Copyright © 2006 by Morgan & Claypool
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations in printed reviews, without the prior permission of the publisher. Advanced Probability Theory for Biomedical Engineers John D. Enderle, David C. Farden, and Daniel J. Krause www.morganclaypool.com ISBN10: 1598291505 ISBN13: 9781598291506 ISBN10: 1598291513 ISBN13: 9781598291513 paperback paperback ebook ebook
DOI 10.2200/S00063ED1V01Y200610BME011 A lecture in the Morgan & Claypool Synthesis Series SYNTHESIS LECTURES ON BIOMEDICAL ENGINEERING #11 Lecture #11 Series Editor: John D. Enderle, University of Connecticut
Series ISSN: 19300328 Series ISSN: 19300336 First Edition 10 9 8 7 6 5 4 3 2 1
print electronic
Printed in the United States of America
Advanced Probability Theory for Biomedical Engineers
John D. Enderle
Program Director & Professor for Biomedical Engineering, University of Connecticut
David C. Farden
Professor of Electrical and Computer Engineering, North Dakota State University
Daniel J. Krause
Emeritus Professor of Electrical and Computer Engineering, North Dakota State University
SYNTHESIS LECTURES ON BIOMEDICAL ENGINEERING #11
M &C
Mor gan
& Cl aypool
Publishers
the probability distribution for a single random variable is determined from a function of two random variables using the CDF. Random Processes. Engineering Statistics. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. Drill problems. The audience includes students. The theory material is presented in a logical manner—developing special mathematical skills as needed. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. as well as important approximations to the Bernoulli PMF and Gaussian CDF. KEYWORDS Probability Theory. Exponential distributions. follow most sections. Gaussian random variables . The mathematical background required of the reader is basic knowledge of differential calculus. Poisson distributions. The exponential. Gaussian distributions Bernoulli PMF and Gaussian CDF. Pertinent biomedical engineering examples are throughout the text. engineers and researchers presenting applications of this theory to a wide variety of problems—as well as pursuing these topics at a more advanced level. Poisson and Gaussian distributions are introduced. straightforward exercises designed to reinforce concepts and develop problem solution skills. Many important properties of jointly Gaussian random variables are presented. Probability and Statistics for Biomedical Engineers. The aim of all three books is as an introduction to probability theory. The primary subjects of the ﬁnal chapter are methods for determining the probability distribution of a function of a random variable.iv ABSTRACT This is the third in a series of short books on probability theory and random processes for biomedical engineers. This book focuses on standard probability distributions commonly encountered in biomedical engineering. Next. Then.
.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 5. .3 Conditional PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 5. . . . . . . . . .1 Marcum’s Q Function .1 Poisson Approximation to Bernoulli . . . . . . . . . . . . . . . . . . 4 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 6. . .2 Mixed Random Variable . . . . . . . 12 5. . . . . . . . . . . . . . . 45 6. .5. . . . . . . . . . . . . . . . . . .18 5. . . 59 6. . . . . . . . . . . . 36 5. . . . . . . . . . . . . . . . . . . . . .2. . . 11 5. . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6. . . . . . . . . . . . 1 5. . . . . . . . . 25 5. . . . .3 Bernoulli Trials . . . . . . . . . . . . . . . .4 Poisson Distribution . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . .6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 6. . . . . . . . . . .7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Standard Probability Distributions . . . . . . . . . . . . . . . . . . .1 CDF Technique with Monotonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 6. . . .4. . . . . . . . . . . . . . . . . . . . 73 6. . . . . . . . . . . . . 20 5. . . . . . . . . . . . . . . . . . . . . . . . .v Contents 5. . . . . . . 36 Transformations of Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 5. . . . . . . . . . . . . . .1 Uniform Distributions . . . . . . . . . . . . . . . . . . 53 6. . . . . . . . . . . . .3 One Function of Two Random Variables . . . . . . . . . . . . . . . .1 Continuous Random Variable . . . . . . . . . . . . .4 Bivariate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 CDF Technique with Arbitrary Functions . . . . . . . .6 Bivariate Gaussian Random Variables . . .1 Bivariate CDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Bivariate PDF Technique . . .5 Univariate Gaussian Distribution . . . . . . . . . . . . . . . . . .2 Gaussian Approximation to Bernoulli . . . . . . . . . . . . . .8 Problems . . . . . . . . . . . . . . . . .1 Constant Contours. . . . . . . . . . . . . . . 65 6.1 Univariate CDF Technique . . . . . . . . . . . . . . . . 45 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Exponential Distributions . .6. . . . . . . . . . . . . . . . . . . . . 75 6. . . . . . . . . 1 5. .5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . .32 5. . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . 56 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Interarrival Times. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 6. . . . .2. .2 Univariate PDF Technique . . . . . . . . . . .1. 63 6. . . . . . . . .
.
These key concepts are all presented in the ﬁrst chapter. Many important properties of jointly distributed Gaussian random variables are presented. Next. The fourth chapter introduces jointly distributed random variables. conditional expectation. The primary subjects of Chapter 6 are methods for determining the probability distribution of a function of a random variable. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. Later chapters simply expand upon these key ideas and extend the range of application. We have found it best to introduce this material using simple examples such as dice and cards. and the joint characteristic function. Our approach is to present a uniﬁed treatment of the subject. There are only a few key concepts involved in the basic theory of probability theory. along with joint expectation. standard deviation. A considerable effort has been made to develop the theory in a logical manner . the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF.developing special mathematical skills as needed. and the characteristic function. conditional moments and the conditional characteristic function are also discussed.vii Preface This is the third in a series of short books on probability theory and random processes for biomedical engineers. The third chapter focuses on expectation. This text is written as an introduction to probability theory. The mathematical background required of the reader is basic knowledge of differential calculus. The goal was to prepare students at the sophomore.as well as pursue these topics at a more advanced level. joint moments. junior or senior level for the application of this theory to a wide variety of problems . the exponential. Poisson and Gaussian distributions are introduced. In addition. Here in Chapter 5. The applications and examples given reﬂect the authors’ background in teaching probability theory and random processes for many years. as well as important approximations to the Bernoulli PMF and Gaussian CDF. Convolution is also developed. Every effort has been made to be consistent with commonly used notation and terminology—both within the engineering community as well as the probability and statistics literature. This short book focuses on standard probability distributions commonly encountered in biomedical engineering. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. moments. rather than more complex biological . The second chapter introduces the topic of random variables. Then.
presented in the same general order as covered in the textbook. Many of the examples and end of chapter problems are based on examples from the textbook by Drake [9]. At the end of each chapter is a wide selection of problems. However. We acknowledge and thank William Pruehsner for the technical illustrations. ranging from simple to difﬁcult. . straightforward exercises designed to reinforce concepts and develop problem solution skills. Drill problems. we do introduce some pertinent biomedical engineering examples throughout the text. The answers to the drill problems follow the problem statement in random order. follow most sections. Students in other ﬁelds should also ﬁnd the approach useful.viii PREFACE and biomedical phenomena.
1.1) The mean and variance of a discrete uniform RV are easily computed with the aid of Lemma 2. 5. .1: ηx = and 2 σx = 1 n n−1 (kh + a) = k=0 [2] 1 b − a n(n − 1) b +a h γn +a = +a = .3) Simplifying. . 0.3. (5. n 2 nn−1 2 2 (5. . This chapter presents some of these discrete and continuous probability distributions that occur often enough in a variety of problems to deserve special mention. . Because of the vast number of probability distributions. b] if x is a lattice RV with span h = (b − a)/(n − 1) and PMF p x (α) = 1/n.2) 1 n n−1 kh − k=0 b −a 2 2 = (b − a)2 n n−1 k=0 k 1 k2 − + . 2 (n − 1) n−1 4 (5. We will see that many random variables and their corresponding experiments have similar properties and can be described by the same probability distribution. k = 0.4) .1 UNIFORM DISTRIBUTIONS Deﬁnition 5. 1. α = kh + a. Each section introduces a new PMF or PDF. 12 n − 1 (5. Following this. The discrete RV x has a uniform distribution over n points (n > 1) on the interval [a. we cannot possibly discuss them all here in this chapter. n − 1 otherwise. and characteristic function are found. Additionally. 2 σx = (b − a)2 n + 1 . variance.1 CHAPTER 5 Standard Probability Distributions A surprisingly small number of probability distributions describe many natural probabilistic phenomena. special properties are pointed out along with relationships among other probability distributions. the PMF or PDF is derived according to the characteristics of the experiment.1. the mean. In some instances.
a ≤α≤b otherwise.1. b] if x has PDF f x (α) = 1/(b − a). n 1 − e jht (5.8) .2 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS px ( α ) . Recall from Section 3.05 1 φ x (t ) 0 . φx (t) = exp j a +b t 2 sin n sin b−a n t 2 n−1 b−a 1 t 2 n−1 . The characteristic function is plotted ∗ over [0. 1].7) The mean and variance of a continuous uniform RV are easily computed directly: 1 ηx = b −a b αd α = a b2 − a2 b +a = .1: (a) PMF and (b) characteristic function magnitude for discrete RV with uniform distribution over 20 points on [0.5 (a) 1 α 10 30 (b) 50 t FIGURE 5. 2(b − a) 2 (5. 0. π /h]. Figure 5. Thus. The characteristic function can be found using the sum of a geometric series: φx (t) = e jat n n−1 (e jht )k = k=0 e jat 1 − e jhnt .1 illustrates onehalf period of φx (·).2.6) Figure 5. (5. Deﬁnition 5.1 illustrates the PMF and the magnitude of the characteristic function for a discrete RV which is uniformly distributed over 20 points on [0. where the span h = 1/19.5) Simplifying with the aid of Euler’s identity.3 that φx (−t) = φx (t) and that φx (t) is periodic in t with period 2π /h. The continuous RV x has a uniform distribution on the interval [a. (5. 1].
1]. Note that the characteristic function in this case is not ∗ periodic but φx (−t) = φx (t). 5].2. (b) Fx (5). −(b−a)/2 Simplifying with the aid of Euler’s identity.10) Figure 5.1. φx (t) = exp a +b j t 2 sin b−a t 2 b−a t 2 . . Determine: (a) p x (20).2 illustrates the PDF and the magnitude of the characteristic function for a continuous RV uniformly distributed on [0.5 (a) 1 10 30 (b) 50 t FIGURE 5. 3.2: (a) PDF and (b) characteristic function magnitude for continuous RV with uniform distribution on [0. 2. 1/6. 1]. Drill Problem 5.2.1. 2 (c) E(x). Answers: 20. (d) σx . 200.8.4) is tossed once. 0. (d) σx .1.1. Random variable x is uniformly distributed on the interval [−1. Let x be a random variable equaling ten times the number tossed. (c) ηx . and 2 σx 1 = b −a b α− a b +a 2 2 dα = (b − a)2 . Drill Problem 5.2. (b) P(10 ≤ x ≤ 50). Answers: 1. A pentahedral die (with faces labeled 0. 0.STANDARD PROBABILITY DISTRIBUTIONS f x (α) 1 1 3 φx (t ) 0 . Deter2 mine: (a) Fx (0). (5.9) The characteristic function can be found as 1 φx (t) = b −a b a exp j b+a t 2 e dα = b −a jαt (b−a)/2 e jαt d α. 12 (5.3.
15) φx (t ) .13) Figure 5.2. 0. otherwise. ≥ 0. 2. The discrete RV x has a geometric distribution or discrete exponential distribution with parameter p(0 < p < 1) if x has PMF p x (α) = p(1 − p)α−1 . p and 2 σx = q . . The continuous RV x has an exponential distribution with parameter λ(λ > 0) if x has PDF f x (α) = λe −λα u(α).18 1 (5.09 0 10 (a) 20 α 0 1 (b) 2 3 t FIGURE 5.3 illustrates the PMF and the characteristic function magnitude for a discrete RV with geometric distribution and parameter p = 0.12) The mean and variance of a discrete RV with a geometric distribution can be computed using the moment generating property of the characteristic function. (5. α = 1.4 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 5. where u(·) is the unit step function.2. . . .2.3: (a) PMF and (b) characteristic function magnitude for discrete RV with geometric distribution [ p = 0.14) Deﬁnition 5.11) The characteristic function can be found using the sum of a geometric series (q = 1 − p): p φx (t) = q ∞ k=1 q e jt k = pe j t .18127]. p2 (5. (5. px (α) . 1 − q e jt (5. It can be shown that a discrete exponentially distributed RV has a memoryless property: p xx> (αx > ) = p x (α − ).2 EXPONENTIAL DISTRIBUTIONS Deﬁnition 5.18127.1. The results are ηx = 1 .
18) Example 5. λ 2 σx = 1 . Suppose a system contains a component that has an exponential failure rate.16) Figure 5.1 0 5 10 (a) 15 α 10 30 (b) 50 t FIGURE 5. like its discrete counterpart.17) A continuous exponentially distributed RV.4: (a) PDF and (b) characteristic function magnitude for continuous RV with exponential distribution and parameter λ = 0. Reliability engineers determined its reliability at 5000 hours to be 95%. λ − jt (5. . (5. τ ≥ 0. grew primarily out of military applications and experiences with multicomponent systems.2. The exponential probability distribution is also a very important probability density function in biomedical engineering applications.2. arising in situations involving reliability theory and queuing problems.STANDARD PROBABILITY DISTRIBUTIONS f x (α) . λ2 (5.1.2. Queuing theory describes the waiting times between events. which describes the time to failure for a system or component. The characteristic function can be found as ∞ φx (t) = λ 0 e α( j t−λ) d α = λ . Determine the number of hours reliable at 99%. Reliability theory. The mean and variance of a continuous exponentially distributed RV can be obtained using the moment generating property of the characteristic function.2 1 5 φx (t ) . satisﬁes a memoryless property: f xx>τ (αx > τ ) = f x (α − τ ).4 illustrates the PDF and the magnitude of the characteristic function for a continuous RV with exponential distribution and parameter λ = 0. The results are ηx = 1 .
the parameter λ is determined from ∞ 0. Determine the average examination grade. to determine the number of hours reliable at 99%. Drill Problem 5.2. Random variable x.44(x − 4/3).1. 0 < x < 4/3 x ≥ 4/3 otherwise. Thus λ= − ln(0.5. Answers: 0.6. 5. First.2.2. λ Drill Problem 5.99) = 980 hours. Determine the number of years reliable at: (a) 90%. Answer: 80. 5. (c) 99%. Suppose a system has an exponential failure rate in years to failure with λ = 0. we solve for α from P (x > α) = e −λα = 0.3 BERNOULLI TRIALS A Bernoulli experiment consists of a number of repeated (independent) trials with only two possible events for each trial. 3 The examination results are given by ⎧ ⎪ 75.95 = P (x > 5000) = 5000 λe −λα d α = e −5000λ .3. 5000 Then. ⎪ ⎩ 0. ⎨ g (x) = 75 + 39. representing the length of time in hours to complete an examination in Introduction to Random Processes. 2. such as a head and a tail in a coin toss. has PDF 4 4 f x (α) = e − 3 α u(α). The events for each trial can be thought of as any two events which partition the sample space. a zero or one in a computer .03 × 10−5 .95) = 1. (b) 95%.6 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution.99 or α= − ln(0.02.
and n = 30. 1. where p = probability of success and q = 1 − p. Using the moment generating property of characteristic functions. The ﬁrst two chapters used this PMF repeatedly in problems dealing with games of chance and in situations where there were only two possible outcomes in any given trial. . . k = 0. Deﬁnition 5. . .2 and n = 30. p = 0.2. (5.18 1 2 σx = npq .20) Figure 5. The Bernoulli PMF describes the probability of k successes in n trials of a Bernoulli experiment. . A discrete RV x is Bernoulli distributed if the PMF for x is ⎧ ⎪ n ⎨ p k q n−k .3.09 0 10 (a) 20 α 0 1 (b) 2 t FIGURE 5. For biomedical engineers. px (α) .1. Let us call one of the events a success. The characteristic function can be found using the binomial theorem: n (5.STANDARD PROBABILITY DISTRIBUTIONS 7 bit.21) φx (t ) . the Bernoulli distribution is used in infectious disease problems and other applications.5 illustrates the PMF and the characteristic function magnitude for a discrete RV with Bernoulli distribution.19) φx (t) = k=0 n ( pe j t )k q n−k = (q + pe j t )n . the mean and variance of a Bernoulli RV can be shown to be ηx = np.5: (a) PMF and (b) characteristic function magnitude for discrete RV with Bernoulli distribution. p = 0. otherwise. k (5. The Bernoulli distribution is also known as a Binomial distribution. the other a failure. or an even and odd number in a die toss. n p x (k) = k ⎪ ⎩ 0.
and 20. we consult Table A. n − k − 1.7421.0 − 0. G(n. . 0.05.1–A.5 and n = 5. and p = 0. what is the probability that they win: (a) at least 10 games.22) This result is easily applied to obtain values of the Bernoulli CDF for values of p > 0. k. n G(n. 1.3 in the Appendix list values of the Bernoulli CDF for p = 0.5.2784 = 0.4845 = 0. . .15. since n n−m = n! = m! (n − m)! n m .7. k. . Example 5.1.15.3. Tables A. Using the Binomial Theorem. p) = 1 − G(n. p) = m=n−k n n−m p n−m (1 − p)m .1311 = 0. 10. (5.5 from Tables A. (c) exactly 2 11 games? (d) With x denoting the number of games won.7216. a closed form expression for the Bernoulli CDF is not easily obtained. using (5. we ﬁnd a) P (x ≥ 10) = 1 − Fx (9) = 1. k = 9. The probability that Fargo Polytechnic Institute wins a game is 0.1–A. Making the change of variable m = n − yields n G(n. With x a Bernoulli random variable. p) = =0 n p (1 − p)n− . Let k ∈ {0. c) p x (11) = Fx (11) − Fx (10) = 0. k. In a 15 game season. n − 1} and deﬁne k G(n.3. Solution. .2. p) = m=0 n m n−k−1 p n−m (1 − p) − m m=0 n m p n−m (1 − p)m . 1 − p). (b) from 9 to 12 games.7031 − 0. 0. Now.1.2186. 15. b) P (9 ≤ x ≤ 12) = Fx (12) − Fx (8) = 0. σx = np(1 − p) = 3. . 2 d) ηx = np = 10.22) with n = 15. 0. .8 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Unlike the preceding distributions.8732 − 0. ﬁnd ηx and σx . . k.7. .
Let RV nr represent the number of trials to obtain exactly r successes (r ≥ 1). − 1) p. . k = 0. r + 1. . (5. . . The moment generating function for nr can be expressed as Mnr (λ) = Letting m = − r .24) (5. . e λr p r ∞ (m + r − 1)(m + r − 2) · · · (m + 1)(q e λ )m . . . hence. Let p(k. 1−x x < 1. otherwise. . Note that P (success in th trial r − 1 successes in previous − 1 trials) = p. we obtain Mnr (λ) = With s (x) = ∞ k=0 ∞ =r ( − 1)( − 2) · · · ( − r + 1) r p q (r − 1)! −r λ e . pr q −r (5. 1) and q = 1 − p. Discrete RV nr thus has PMF p nr ( ) = ⎧ ⎪ ⎨ ⎪ ⎩ −1 r −1 0. we have P (nr = ) = p(r − 1.25) . . The PMF for the RV nr is called the negative binomial distribution. = r.STANDARD PROBABILITY DISTRIBUTIONS 9 We now consider the number of trials needed for k successes in a sequence of Bernoulli trials. . n) = ⎧(k successes in n trials) P ⎪ n ⎨ p k q n−k . otherwise. (5. also known as the P´ lya and the Pascal distribution.26) where the parameter r is a positive integer. . (r − 1)! m=0 xk = 1 . Note that with o r = 1 the negative binomial PMF is the geometric PMF. for = r. .23) where p = p(1. n = k ⎪ ⎩ 0. r + 1. 1.
p and 2 σn r = rq .06 . (5. (5.10 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS we have s (x) = ( ) ∞ k= ∞ m=0 k(k − 1) · · · (k − + 1)x k− (m + )(m + − 1) · · · (m + 1)x m ! +1 = = Hence (1 − x) . φx (t ) 1 px (α) . . and p = 0. Mnr (λ) = pe λ 1 − qeλ r .18127.28) We note that the characteristic function is simply r φnr (t) = Mnr ( j t) = φx (t).03 0 20 (a) 40 α 0 1 (b) 2 t FIGURE 5.29) where RV x has a discrete geometric distribution. q e λ < 1. r = 3.6: (a) PMF and (b) magnitude characteristic function for discrete RV with negative binomial distribution.27) The mean and variance for nr are found to be ηn r = r .6 illustrates the PMF and the characteristic function magnitude for a discrete RV with negative binomial distribution. and p = 0. r = 3. p2 (5.18127. Figure 5.
. . We have p(k) = lim n k p k q n−k ⎧ k −λ ⎨λ e . n→∞.30) Proof. We note that the limiting value p(k) may be used as an approximation for the Bernoulli PMF when p is small by substituting λ = np. from which the desired result follows.3. The Poisson PMF is treated in the following section. Satisfactory results are obtained with np < 10. Substituting p = and q = 1 − λ .np=λ λ n (5. the better the approximation. λ lim ln 1 − n→∞ n so that n→∞ n λk n→∞ k! 1− λ n n .1 Poisson Approximation to Bernoulli When n becomes large in the Bernoulli PMF in such a way that np = λ = constant. 1. The motivation for using this approximation is that when n is large. . = lim ln 1 − 1 n λ n n→∞ = −λ lim 1− λ n n = e −λ . the larger the value of n and the smaller the value of p.3 are useless for ﬁnding values for the Bernoulli CDF. = k! ⎩ 0.3. i=0 n→∞ lim n−k 1 − λ n −k k−1 i=0 (n − i) = 1. so that p(k) = lim Now. the Bernoulli PMF approaches another important PMF known as the Poisson PMF. otherwise. Lemma 5. n λ k k 1 p(k) = lim n→∞ k! Note that 1− λ n n−k k−1 (n − i). While there are no prescribed rules regarding the values of n and p for this approximation.1–A. Tables A.STANDARD PROBABILITY DISTRIBUTIONS 11 5.1. k = 0. .
Incidentally. when n is large. that is.001.2 Gaussian Approximation to Bernoulli Previously. φ y (t) = q exp − j t p qn + p exp jt q np n t √ npq . 5.5) is also used to approximate a Bernoulli PMF under certain conditions.5 (the Poisson CDF table is covered in Section 4) that P (x ≤ 5) = 0.2.3. Suppose x is a Bernoulli random variable with n = 5000 and p = 0. p is close to 1/2. Solution. npq where x is a Bernoulli RV. np t q q + p exp t j√ npq n . Since λ = np = 5. Then the characteristic function for y satisﬁes φ(t) = lim φ y (t) = e −t n→∞ 2 (5. The accuracy of this approximation is best when n is large. (5.3. and p is very close to zero.3. Then either the Poisson or the Gaussian approximation will yield good results.32) Proof. and npq > 3. Lemma 5. This approximation is quite useful since the Bernoulli table lists only CDF values for n up to 20. We have np t φx φ y (t) = exp − j √ npq Substituting for φx (t). we ﬁnd from Table A. if p is close to one.2. we can still use this approximation by reversing our deﬁnition of success and failure in the Bernoulli experiment. Notice that in some circumstances np < 10 and npq > 3. Our solution involves approximating the Bernoulli PMF with the Poisson PMF since n is quite large (and the Bernoulli CDF table is useless).31) /2 . φ y (t) = exp − j Simplifying. p is small and np < 10. . .22). which results in a value of p close to zero—see (5. Let x − np y= √ .6160.12 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 5. Find P (x ≤ 5). The Gaussian PDF (see Section 5. the Poisson PMF was used to approximate a Bernoulli PMF under certain conditions.
9 of the Appendix. o α→0 lim ln φ y (t) = lim −t 2 p − t 2 β 2 p t2 − j tpβe − j tα/β + j tpβe j tβα = =− . The solution involves approximating the Bernoulli CDF with the Gaussian CDF since npq = 1200 > 3.34) and Q(·) is Marcum’s Q function which is tabulated in Tables A. n lim ln φ y (t) = lim ln( pβ 2 e − j tα/β + pe j tβα ) .STANDARD PROBABILITY DISTRIBUTIONS 13 Letting β= we obtain n→∞ q . where 1 F(γ ) = √ 2π is the standard Gaussian CDF. .8 and A. npq (5.8 that P (x ≤ 2048) ≈ F(1. we ﬁnd from Table A. Example 5.33) e −α −∞ 2 /2 d α = 1 − Q(γ ) (5. The limiting φ(t) in the above lemma is the characteristic function for a Gaussian RV with zero mean and unit variance.5. p and α= 1 . Find P (x ≤ 2048).91774.3. for large n and a < b P (a < x < b) = P (a < y < b ) ≈ F(b ) − F(a ). Solution. With np = 2000. Suppose x is a Bernoulli random variable with n = 5000 and p = 0. Evaluation of the above integral as well as the Gaussian PDF are treated in Section 5. α→0 2α 2 2 lim φ y (t) = exp lim ln φ y (t) = e −t n→∞ /2 Consequently.3.641 = 1. npq b − np b = √ .39. a − np a = √ . npq = 1200 and b = (2048 − 2000)/34. 2 n→∞ .4.39) = 0. α→0 α2 Applying L’Hˆ spital’s Rule twice. Hence.35) γ (5.39) = 1 − Q(1.
npq = 5. Answers: 0. North Dakota revealed that 30% preferred a white automobile over all other colors. From the Bernoulli table. Answers: 2. In the previous section. With np = 10.5 < x < 8. Solution. (b) the standard deviation for 15 shots. 0. p x (8) ≈ 0. 0. Suppose x is a Bernoulli random variable with n = 20 and p = 0. (b) at least ﬁve of the next twenty cars purchased will be white. 5. a Poisson random variable might represent the number of telephone calls per hour. 1. which is very close to the above approximation.5. (c) from two to ﬁve of the next twenty cars purchased will be white. and b = (8.7625. typically a time interval. Prof.13136. For example.3. In this section.67) − F(−1.1201. Example 5.5 − 10)/ √ 5 = −0. p x (8) = 0. (c) the number of times she must ﬁre so that the probability of hitting the target at least once is greater than 1/2. hence.12007. Since npq = 5 > 3. to compute the probability that a Bernoulli RV takes an integer value using the Gaussian approximation we must round off to the nearest integer.1789.67.5 − 10)/ 5 = −1. The probability she will hit the target is only 0. a = (7. or the number of errors per page in this textbook.12) = 0. Determine the probability that: (a) exactly ﬁve of the next 20 cars purchased will be white. a continuous distribution is used to calculate probabilities for a discrete RV.5.3. Drill Problem 5. Find P (x = 8). Rensselaer is an avid albeit inaccurate marksman. we derive the Poisson probability distribution from two fundamental assumptions about the phenomenon based on physical characteristics. A survey of residents in Fargo. we found that the limit (as n → ∞ and constant mean np) of a Bernoulli PMF is a Poisson PMF. Drill Problem 5. Operationally. p x (8).12.4. 4. we have p x (8) = P (7.14 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS When approximating the Bernoulli CDF with the Gaussian CDF. .3.25143 − 0.4088.2.5) ≈ F(−0. Determine: (a) the expected number of hits scored in 15 shots.7748.4 POISSON DISTRIBUTION A Poisson PMF describes the number of successes occurring on a continuous line. the Gaussian approximation is used to evaluate the Bernoulli √ PMF. or within a given region.1.3. It is important to note that while the approximation is excellent in terms of the CDFs—the PDF of any discrete RV is never approximated with a continuous PDF.
(2) The probability of a single success during a very small time interval is proportional to the length of the interval. We use the notation p(k. τ + h] = [0. and p(0. τ ] ∩ (τ. (5. τ + h] and [0. τ )(1 − λh + o (h)). τ ]). τ + h] = . and B = { successes in interval I2 }. the greater the probability of success. τ + h) = p(0. τ ] ∪ (τ. with A = {k successes in interval I1 }. we have P (A ∩ B) = P (A)P (B). (5. τ ) p(0. if I1 ∩ I2 = . . the Poisson process is composed of a series of Bernoulli trials.40) (5. (5.37) (5. Hence p(1.39) (5.36) The Poisson probability distribution is characterized by the following two properties: (1) The number of successes occurring in a time interval or region is independent of the number of successes occurring in any other nonoverlapping time interval or region. the number of successes depends only on the length of the time interval and not the location of the interval on the time axis. h) = 1 − λh + o (h).41) This second property indicates that for a series of very small intervals.38) As we will see. τ ) = P (k successes in interval [0. g (h) = 15h 2 + 7h 3 = o (h).STANDARD PROBABILITY DISTRIBUTIONS 15 The following development makes use of the order notation o (h) to denote any function g (h) which satisﬁes g (h) = 0. h) = p(0. h→0 h lim For example. h) = λh + o (h). each with a probability of success p = λh + o (h). Thus.42) (5. we have p(0. The probability of more than one success occurring during an interval vanishes as the length of the interval approaches zero. Since [0. The longer the interval.
(5. τ ) p(0. h) + o (h). Applying the above properties. it is readily seen that p(k. The rate of the Poisson process is λ and the interval length is τ .47) The RV x = number of successes thus has a Poisson distribution with parameter λτ and PMF p x (k) = p(k. τ ) = −λτ 0 e λt p(k − 1. so that o (h) p(k. (5. τ ) = λe and hence that p(k. τ ) −λh p(0. 0) = 0. τ ) p(1. τ ) = λp(k − 1. τ + h) − p(0. (5.43) (5. τ ) = e −λτ u(τ ). dτ with p(k. t)d t (5. The characteristic function for a Poisson RV x with parameter λ is found as (with parameter λ. 2. k! (5. τ )(1 − λh) + o (h). . τ )λh + p(k. 1)) φx (t) = e −λ ∞ k=0 (λe j t )k = e −λ exp(λe j t ) = exp(λ(e j t − 1)). 0) = 1. . For ease in subsequent development. τ ) + λp(k. τ + h) − p(k. τ ) + .44) k = 1.45) p(k. d p(0. . τ ) + λp(k.46) (λτ )k e −λτ u(τ ). . . τ ). p x (k) = p(k. dτ This differential equation has solution p(0. τ + h) = p(k − 1. or p(k. τ ). τ ) = λp(k − 1. . . . It can be shown ([7. h h Taking the limit as h → 0 d p(k. τ ). h) + p(k. we replace the parameter λτ with λ. 1. τ ) + o (h) = h h and taking the limit as h → 0. τ + h) = p(k − 1. τ ) = −λp(0. 8]) that τ p(0.16 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Noting that p(0. k! k = 0.48) .
x1 + x2 is also a Poisson with parameter λ1 + λ2 . respectively. Figure 5. It is of interest to note that if x1 and x2 are independent Poisson RVs with parameters λ1 and λ2 .51) If x is Poisson distributed with parameter λ. so that (k) ψx (γ ) = λk e λ(γ −1) .7 illustrates the PMF and characteristic function magnitude for a discrete RV with Poisson distribution and parameter λ = 10.50) (x − i) . then ψx (γ ) = e λ(γ −1) .12 1 φx (t ) 17 .e. Consider the function ψx (γ ) = E(γ x ) and note that k−1 (k) ψx (γ ) = E γ x−k i=0 (5.06 0 10 (a) 20 α 0 1 (b) 2 t FIGURE 5.7: (a) PMF and (b) magnitude characteristic function for Poisson distributed RV with parameter λ = 10.49) i. so that k−1 E i=0 (k) (x − i) = ψx (1)..STANDARD PROBABILITY DISTRIBUTIONS px (α) . (5. then φx1 +x2 (t) = exp((λ1 + λ2 )(e j t − 1)).52) . (5. The moments of a Poisson RV are tedious to compute using techniques we have seen so far. (5.
of a Poisson random variable is more important than the actual number of successes.18 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS hence. On the average. c) P (x ≥ 15) = 1 − Fx (14) = 1 − 0. (b) 8–10 problems are graded. E(x(x − 1)) = λ2 = E(x 2 ) − λ. b) P (8 ≤ x ≤ 10) = Fx (10) − Fx (7) = 0. in evaluating the reliability of a medical device.53) 2 In particular. Here. hand computation of the CDF is quite tedious. E(x) = λ. τ − h) p(1.7 of the Appendix for selected values of λ ranging from 0.4. the Poisson CDF is tabulated in Tables A. Example 5. τ − h)λh + o (h) . we consult the Poisson CDF table with λ = 10. we will brieﬂy examine the subject of interarrival times from the basis of the Poisson PMF. the time to failure is far more signiﬁcant to the biomedical engineer than the fact that the device failed.1126.0835. Therefore. (5.1.4.2202 = 0.2202 = 0.3328 − 0. and then decreases in value as k increases from the mean. What is the probability that on a given day (a) 8 problems are graded.3628.4A. we note that the value of the Poisson PMF increases as the number of successes k increases from zero to the mean. so that σx = λ2 + λ − λ2 = λ. From the Poisson CDF table. known as an interarrival time. k−1 E i=0 (x − i) = λk . 5. With x a Poisson random variable. h) = p(r − 1.5830 − 0. even though an inﬁnite number of Poisson successes are theoretically possible. While it is quite easy to calculate the value of the Poisson PMF for a particular number of successes. Additionally. Let RV tr denote the length of the time interval from zero to the r th success. Professor Rensselaer grades 10 problems per day. the length of time between successes. and ﬁnd a) p x (8) = Fx (8) − Fx (7) = 0.9165 = 0. the subject of reliability theory is so important that entire textbooks are devoted to the topic. Then p(τ − h < tr ≤ τ ) = p(r − 1. and (c) at least 15 problems are graded? Solution.1 to 18. however. note that the table is written with a ﬁnite number of entries for each value of λ because the PMF values are written with six decimal place accuracy.1 Interarrival Times In many instances. For example. Indeed.
STANDARD PROBABILITY DISTRIBUTIONS
19
so that o (h) Ftr (τ ) − Ftr (τ − h) = λp(r − 1, τ − h) + . h h Taking the limit as h → 0 we ﬁnd that the PDF for the r th order interarrival time, that is, the time interval from any starting point to the r th success after it, is f tr (τ ) = λr τ r −1 e −λτ u(τ ), (r − 1)! r = 1, 2, . . . . (5.54)
This PDF is known as the Erlang PDF. Clearly, with r = 1, we have the exponential PDF: f t (τ ) = λe −λτ u(τ ). The RV t is called the ﬁrstorder interarrival time. The Erlang PDF is a special case of the gamma PDF: f x (α) = for any real r > 0, λ > 0, where λr αr −1 e −λα u(α), (r ) (5.56) (5.55)
is the gamma function
∞
(r ) =
0
αr −1 e −α d α.
(5.57)
Straightforward integration reveals that (1) = 1 and (r + 1) = r (r ) so that if r is a positive integer then (r ) = (r − 1)!—for this reason the gamma function is often called the factorial function. Using the above deﬁnition for (r ), it is easily shown that the moment generating function for a gammadistributed RV is Mx (η) = The characteristic function is thus φx (t) = It follows that the mean and variance are r ηx = , λ λ λ − jt
r
λ λ−η
r
,
for η < λ.
(5.58)
.
(5.59)
and
2 σx =
r . λ2
(5.60)
Figure 5.8 illustrates the PDF and magnitude of the characteristic function for a RV with gamma distribution with r = 3 and λ = 0.2.
20
ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS
f x (α) .06 1 φx (t )
.03
0
20 (a)
40
α
0
1 (b)
2
t
FIGURE 5.8: (a) PDF and (b) magnitude characteristic function for gamma distributed RV with r = 3 and parameter λ = 0.2.
Drill Problem 5.4.1. On the average, Professor S. Rensselaer makes ﬁve blunders per lecture. Determine the probability that she makes (a) less than six blunders in the next lecture: (b) exactly ﬁve blunders in the next lecture: (c) from three to seven blunders in the next lecture: (d) zero blunders in the next lecture. Answers: 0.6160, 0.0067, 0.7419, 0.1755. Drill Problem 5.4.2. A process yields 0.001% defective items. If one million items are produced, determine the probability that the number of defective items exceeds twelve. Answer: 0.2084. Drill Problem 5.4.3. Professor S. Rensselaer designs her examinations so that the probability of at least one extremely difﬁcult problem is 0.632. Determine the average number of extremely difﬁcult problems on a Rensselaer examination. Answer: 1.
5.5
UNIVARIATE GAUSSIAN DISTRIBUTION
The Gaussian PDF is the most important probability distribution in the ﬁeld of biomedical engineering. Plentiful applications arise in industry, research, and nature, ranging from instrumentation errors to scores on examinations. The PDF is named in honor of Gauss (1777–1855), who derived the equation based on an error study involving repeated measurements of the same quantity. However, De Moivre is ﬁrst credited with describing the PDF in 1733. Applications also abound in other areas outside of biomedical engineering since the distribution ﬁts the observed data in many processes. Incidentally, statisticians refer to the Gaussian PDF as the normal PDF.
STANDARD PROBABILITY DISTRIBUTIONS
21
Deﬁnition 5.5.1. A continuous RV z is a standardized Gaussian RV if the PDF is 1 1 2 f z(α) = √ e − 2 α . 2π (5.61)
The moment generating function for a standardized Gaussian RV can be found as follows: 1 Mz(λ) = √ 2π 1 =√ 2π
∞
e λα− 2 α d α
1 2
−∞ ∞
e − 2 ((α−λ) −λ ) d α.
1 2 2
−∞
Making the change of variable β = α − λ we ﬁnd
∞
Mz(λ) = e
1 2 2λ
f z(β)dβ = e 2 λ ,
1 2
(5.62)
−∞
for all real λ. We have made use of the fact that the function f z is a bona ﬁde PDF, as treated in Problem 42. Using the Taylor series expansion for an exponential, e 2λ =
1 2
∞ k=0
λ2k = 2k k!
∞ n=0
Mx(n) (0)λn , n!
so that all moments of z exist and E(z 2k ) = and E(z 2k+1 ) = 0, k = 0, 1, 2, . . . . (5.64) (2k)! , 2k k! k = 0, 1, 2, . . . , (5.63)
Consequently, a standardized Gaussian RV has zero mean and unit variance. Extending the range of deﬁnition of Mz(λ) to include the ﬁnite complex plane, we ﬁnd that the characteristic function is φz(t) = Mz( j t) = e − 2 t .
1 2
(5.65)
2 Letting the RV x = σ z + η we ﬁnd that E(x) = η and σx = σ 2 . For σ > 0
Fx (α) = P (σ z + η ≤ α) = Fz((α − η)/σ ),
25. is a bellshaped curve completely determined by its mean and variance. 1 2 2 (5. the larger the value of the variance. the maximum value of the Gaussian PDF. the more spread in the distribution and the smaller the maximum value of the PDF. occurs at α = η.12 . with σ > 0 and x = −σ z + η we ﬁnd Fx (α) = P (−σ z + η ≤ α) = 1 − Fz((η − α)/σ ). The PDF approaches zero asymptotically as α approaches ±∞. η = 25.9: Gaussian probability density function for η = 75 and σ 2 = 9. in fact. (2πσ 2 )−1/2 . Note that if x ∼ G(η. . which necessitates numerical integration.04 65 75 85 α FIGURE 5.08 . so that f x is as above. We will have occasion to use the shorthand notation x ∼ G(η. Unfortunately. illustrated with η = 75 and σ 2 = 25.66) Similarly. a closed form expression does not exist for the Gaussian CDF. identically shaped curves could be drawn. and the area under the curve must equal one. σ 2 ) to denote that the RV has a Gaussian PDF with mean η and variance σ 2 .67) The Gaussian PDF. 2σ 2 (5. As can be seen. If. Additionally. the Gaussian PDF curve must be symmetrical as previously described. as well as with η = 75 and σ 2 = 9 in Figure 5. the Gaussian PDF is symmetrical about the vertical axis through the expected value. Rather than attempting to tabulate the general Gaussian CDF.9. a normalization is performed to obtain a standardized Gaussian RV (with zero mean f x (α) . centered now at 25 instead of 75. For any combination of the mean and variance. σ 2 ) then (x = σ z + η) φx (t) = e j ηt e − 2 σ t . Naturally.22 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS so that x has the general Gaussian PDF f x (α) = √ 1 2πσ 2 exp − 1 (α − η)2 .
the RV z = (x − η)/σ is a standardized Gaussian RV: z ∼ G(0. If x ∼ G(η.74). deﬁned by 2 erf(α) = π and 2 erfc(α) = π ∞ α (5. 1 2 (5. It is easy to show that Q(−α) = 1 − Q(α) = Fz(α). This transformation is always applied when using standard tables for computing probabilities for Gaussian RVs.8 and A.5.75) Example 5. 1). A simple change of variable reveals that √ erfc(α) = 2Q(α/ 2). where z ∼ G(0.73) 0 e −t d t = 1 − erf(α) 2 (5.72) e −t d t 2 (5. The error and complementary error functions.74) α are also often used to evaluate the standard normal integral. using the fact that Fx (α) = Fz((α − η)/σ ). 1 2 (5.69) e − 2 τ d τ = 1 − Q(α). Note that 1 Fz(α) = √ 2π where Q(·) is Marcum’s Q function: 1 Q(α) = √ 2π ∞ α (5.9 for 0 ≤ α < 4 using the approximation presented in Section 5.STANDARD PROBABILITY DISTRIBUTIONS 23 and unit variance).68) (5. The probability P (α1 < x ≤ α2 ) can be obtained as P (α1 < x ≤ α2 ) = Fx (α2 ) − Fx (α1 ).1. (5. Compute Fz(−1. 1).70) −∞ e − 2 τ d τ. σ 2 ).1.71) α Marcum’s Q function is tabulated in Tables A. .5.
5. To compute Fz(−1.0 and 3.9. η + 3σ ] = [5.8.01 0 30 60 α FIGURE 5.2. we note that 99. The PDF is essentially zero outside the interval [η − 3σ.6) − Fz(0. Suppose x is a Gaussian random variable with η = 35 and σ = 10. Hence P (37 ≤ x ≤ 51) = Q(0.04093. With z= we have P (37 ≤ x ≤ 51) = P (0. Example 5. Sketch the PDF and then ﬁnd P (37 ≤ x ≤ 51).0027).03 .5.74) = 0.02 .2 ≤ z ≤ 1.9.10 along with the indicated probability.0. From Table A.73% of the area under the curve is contained between −3.2. 65].36594 from Table A. the range of values random variable x takes on is then approximately η ± 3σ . but the probability of it occurring is really very small (2Q(3) = 0. This notion does not imply that random variable x cannot take on a value outside this interval.74) = Q(1. we ﬁnd Fz(−1.2).10: PDF for Example 5. From the transformation z = (x − η)/σ . The sketch of this PDF is shown in Figure 5.74) = 1 − Q(−1.2) − Q(1.74). using (5.72) and Table A.04 . Indicate this probability on the sketch.6) = 0. x − 35 10 f x (α) . the realistic range of values is much smaller. While the value a Gaussian random variable takes on is any real number between negative inﬁnity and positive inﬁnity.24 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. Solution.6) = Fz(1. .
78) where t= and 1 h(t) = √ t(a 1 + t(a 2 + t(a 3 + t(a 4 + a 5 t)))). where z = (x − 25)/6 ∼ G(0. Q(γ ) is the complement of the standard Gaussian CDF. 1 + 0.STANDARD PROBABILITY DISTRIBUTIONS 25 Example 5. 2π The constants are i 1 2 3 4 5 ai 0.356563782 1.3.330274429 (5. If the RV z ∼ G(0.5.. 932]: Q(γ ) ≈ e − 2 γ h(t).31938153 −0. and that Fz(−γ ) = Q(γ ).09. Let the RV x denote the value of a capacitor. (5. p.5. This result is determined by linear interpolation of the CDF between equal 1. 5. Assuming that capacitance follows a Gaussian distribution.1 Marcum’s Q Function ∞ Marcum’s Q function. A very accurate approximation to Q(γ ) is presented in [1. 1 2 γ > 0.80) 1 . (5. 1). Solution.13941.083 ≤ z) = Q(1.e.2316419γ (5.5 × 10−8 .821255978 1.08 and 1. Since we are measuring to the nearest μF.76) γ has been extensively studied.781477937 −1. ﬁnd the probability that the value of capacitance exceeds 31 μF if capacitance is measured to the nearest μF. Q(∞) = 0. the probability that the measured value exceeds 31 μF is P (31.5 ≤ x) = P (1.79) The error in using this approximation is less than 7. . deﬁned by 1 Q(γ ) = √ 2π e−2α dα 1 2 (5.083) = 0. 1) then Q(γ ) = 1 − Fz(γ ).77) i.5. Note that Q(0) = 0. A machine makes capacitors with a mean value of 25 μF and a standard deviation of 6 μF.
82) .5. Drill Problem 5.5. 298] e−2α 2 < Q(α) ≤ √ π α + α2 + 4 1 2 e−2α 2 .946 when α = 3 and 0. .26 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS A very useful bound for Q(α) is [1. 0.967 when α = 4.81) The ratio of the upper bound to the lower bound is 0. 1. The bound improves as α increases.5. we discuss the joint Gaussian PDF and its characteristics by drawing on our univariate Gaussian PDF experiences. i = 0.5. Answers: 167. 1179. Let the random variable x denote the length of attendance and assume that x is Gaussian. .92535.5. 0.721). Students attend Fargo Polytechnic Institute for an average of four years with a standard deviation of onehalf year. . 639. √ π α + α 2 + 0.32.83) The procedure is terminated when γi+1 ≈ γi to the desired degree of accuracy.5 and a standard deviation of 0. (d) greater than 3. Then compute ti = and γi+1 h(ti ) = 2 ln q 1/2 1 1 + 0.02275. 322. (d )Fx (4.6 to 3. Drill Problem 5. 0. (c )P (x = 4). Drill Problem 5. Now.5π 1 2 (5. 150 ± d .3.0. (b)P (x > 4). 5. Suppose grade point averages are computed to the nearest tenth. . it is desired to ﬁnd the value of γ for which Q(γ ) = q . covers 95% of the shots.6 BIVARIATE GAUSSIAN RANDOM VARIABLES The previous section introduced the univariate Gaussian PDF along with some general characteristics. and signiﬁcantly expanding the scope of applications.2316419γi (5.5. Sometimes. Helstrom [14] offers an iterative procedure which begins with an initial guess γ0 > 0. Answer: 33.1. p. Professor Rensselaer loves the game of golf. (b) less than 2.2. (c) between 3. Determine the value of d so that the range.0 and 3.5. . (5.7. She has determined that the distance the ball travels on her ﬁrst shot follows a Gaussian distribution with a mean of 150 and a standard deviation of 17. Determine the number of freshmen you would expect to score: (a) from 2. Answers: 0. The quality point averages of 2500 freshmen at Fargo Polytechnic Institute follow a Gaussian distribution with a mean of 2. Determine: (a) P (1 < x < 3).
we ﬁrst ﬁnd the joint characteristic function for the standardized RVs z1 and z2 . x ∼ G(ηx .1. Finally. and let x and y be jointly distributed Gaussian RVs. σy (5. Let z1 = x − ηx σx and z2 = y − ηy . Completing the square. Next. σx ) and 2 y ∼ G(η y . β) = 2π ∞ I (α.z2 (t1 . σx = 0.86) .85) From (6) we have 1 f z1 . −∞ (5.y . Below.z2 (α. so that φz1 . Then 2 w is a univariate Gaussian RV for all real constants a and b.STANDARD PROBABILITY DISTRIBUTIONS 27 Numerous applications of this joint PDF are found throughout the ﬁeld of biomedical engineering and. In this case we also say that the RVs x and y are jointly distributed Gaussian RVs. Also.z2 . and σ y = 0. The following development is signiﬁcantly simpliﬁed by considering the standardized versions of x and y. the marginal PDFs for a joint Gaussian PDF are univariate Gaussian. σ y ). Let the RV w = ax + by. t2 ) = E(e j t1 z1 + j t2 z2 ) = e − 2 (1−ρ 1 2 2 )t2 − 1 (t1 +ρt2 )2 2 e .y = ±1. f yx and f x. like the univariate case. 1). In particular.y . which we now proceed to do. then the conditional PDF f z2 z1 and the joint PDF f z1 . Deﬁnition 5. σx = 0.e.. The above deﬁnition of a bivariate Gaussian RV is sufﬁcient for determining the bivariate PDF. and σ y = 0 are discussed. 2 2 2 t1 + 2t1 t2 ρ + t2 = (t1 + ρt2 )2 + (1 − ρ 2 )t2 . the RV t1 z1 + t2 z2 is univariate Gaussian: 2 2 t1 z1 + t2 z2 ∼ G 0. 1) and z2 ∼ G(0.84) so that z1 ∼ G(0. the results for z1 and z2 are applied to obtain corresponding quantities φx. i.y  < 1. we assume that ρx. y) is a bivariate Gaussian RV if every linear combination of x and y has a univariate Gaussian distribution. Since z1 and z2 are jointly Gaussian. The bivariate RV z = (x. t1 + 2t1 t2 ρ + t2 .6. t2 )e − jβt2 d t2 . the joint Gaussian PDF is considered the most important joint distribution for biomedical engineers. the special cases ρx. (5.
88) (5. β) = f z2 z1 (βα) = f z1 (α) so that E(z2  z1 ) = ρz1 and 2 σz2 z1 = 1 − ρ 2 .z2 (α.87) (5.90) We now turn our attention to using the above results for z1 and z2 to obtain similar results for x and y. β) = f z1 (α) 2π ∞ φ(t2 )e − jβt2 d t2 −∞ and recognize that φ is the characteristic function for a Gaussian RV with mean αρ and variance 1 − ρ 2 . we ﬁnd f z1 . 1 2 −∞ . Substituting into (5. From (5. −∞ Substituting (5.z2 (α. t2 ) = e or I (α. we obtain I (α. β) = 1 α 2 − 2ραβ + β 2 exp − . 2π (1 − ρ 2 )1/2 2(1 − ρ 2 ) (5. . t2 ) = 2π ∞ φz1 .z2 (α. t2 ) = φ(t2 ) f z1 (α). t2 )e − j αt1 d t1 . Thus f z1 . 2(1 − ρ 2 ) (5. 1 2π (1 − ρ 2 ) exp − (β − ρα)2 .28 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where 1 I (α.89) After some algebra.z2 (t1 .86) we ﬁnd 1 f z1 . where φ(t2 ) = e j αρt2 e − 2 (1−ρ 1 2 2 )t2 2 − 1 (1−ρ 2 )t2 2 1 2π ∞ e − 2 τ e − j α(τ −ρt2 ) d τ.85) and letting τ = t1 + t2 ρ.84) we ﬁnd that x = σx z1 + ηx and y = σ y z2 + η y .
β) = Since f x.94) we ﬁnd f x. β) = and exp − f yx (βα) = β−η y −ρσ y α−ηx σx 2 2(1−ρ 2 )σ y 2 1 exp − 2(1−ρ 2 ) (α−ηx )2 2 σx − 2ρ(α−ηx )(β−η y ) σx σ y + (β−η y )2 2 σy 2πσx σ y (1 − ρ 2 )1/2 (5. β) = (2π)2 ∞ ∞ φz1 .STANDARD PROBABILITY DISTRIBUTIONS 29 so that the joint characteristic function for x and y is φx.z2 (σx t1 . the joint characteristic function for x and y can be found from the joint characteristic function of z1 and z2 as φx. the joint characteristic function φx.95) 2 2πσ y (1 − ρ 2 ) .y (α. 1 f z . (5. σ y t2 )e − j (α−ηx )t1 e − j (β−η y )t2 d t1 d t2 .z2 (σx t1 .z σx σ y 1 2 α − ηx β − η y . Consequently. β) as 1 f x. σ y t2 )e j ηx t1 e j η y t2 .92) Making the change of variables τ1 = σx t1 .96) . we obtain f x.87) into (5. (5. (5.y can be transformed to obtain the joint PDF f x.93) and (5. τ2 = σ y t2 .94) 1 fz σx 1 α − ηx σx .y (α.93) to obtain f yx (βα) = 1 f z z σy 2 1 β − η y α − ηx σy σx .90) and (5. t2 ) = E(e j t1 x+ j t2 y ) = E(e j t1 σx z1 + j t2 σ y z2 )e j t1 ηx + j t2 η y .y (t1 .y (α. (5.91) Using (4.93) Substituting (5.y (t1 . −∞ −∞ (5.y (α. β) = f yx (βα) f x (α) and f x (α) = we may apply (5.y (α. t2 ) = φz1 . σx σy .66).
ηx = η y = 0. The bivariate characteristic function for x and y is easily obtained as follows.30 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS It follows that E(yx) = η y + ρσ y and 2 2 σ yx = σ y (1 − ρ 2 ).97) (5. the RV t1 x + t2 y is a univariate Gaussian RV: 2 2 2 2 t1 x + t2 y ∼ G t1 ηx + t2 η y .y (α.y + t2 σ y . x − ηx σx (5.101) A threedimensional plot of a bivariate Gaussian PDF is shown in Figure 5.98) ⎞ ⎠ . (5.11: Bivariate Gaussian PDF f x.265 β =3 α = −3 β = −3 α =3 FIGURE 5. β) with σx = σ y = 1. . Since x and y are jointly Gaussian.99) By interchanging x with y and α with β. σy (5. fx . (5.11. β ) . t1 σx + 2t1 t2 σx. ⎛ exp ⎝− f xy (αβ) = α−ηx −ρσx β−η y σy 2 2(1−ρ 2 )σ y 2 2 2πσx (1 − ρ 2 ) E(xy) = ηx + ρσx and y − ηy .8. and ρ = −0.y (α.100) 2 2 σxy = σx (1 − ρ 2 ).
Proof. If ρ = 0 then (from (5. We conclude that x − ηx σx α − ηx σx f x.6.1. and ρ = ±1.y (t1 . t2 ) = e − 2 (t1 σx +2t1 t2 σx. β) = f y (β)δ α − ηx ± σx β − ηy σy (5.98) we ﬁnd E(yx) → η y ± σ y 2 and σ yx → 0. the joint characteristic function for x and y is φx. 1 2 2 2 2 (5. We now consider some special cases of the bivariate Gaussian PDF. from (5. Find constants a and b such that v = ax + by ∼ G(0.1.y (α. then ρx. σx and σ y .y +t2 σ y ) e j t1 ηx + j t2 η y . 1) and such that v and x are independent. Interchanging the roles of x and y we ﬁnd that the joint PDF for x and y may also be written as f x. As ρ → ±1. Theorem 5. the probability that the RV deviates from its mean by more than an arbitrarily small ﬁxed amount approaches zero. These results can also be obtained “directly” from the joint characteristic function for x and y. Let x and y be jointly Gaussian with zero means. β) = f x (α) f y (β). The jointly Gaussian RVs x and y are independent iff ρx. σx = σ y = 1.6. Then f yx (βα) = f y (β).102) which is valid for all σx.95)) f x. (5.y (α. Hence. A very special property of jointly Gaussian RVs is presented in the following theorem. 2 2 Example 5.STANDARD PROBABILITY DISTRIBUTIONS 31 Consequently.105) when ρ = ±1.y = 0.97) and (5.104) for ρ = ±1. .y = 0. RVs x and y are independent.y = 0.103) x − ηx σx y → ηy ± σy in probability1 .y (α. β) = f x (α)δ β − η y ± σ y (5.e. i. Suppose that ρ = ρx. We showed previously that if x and y are independent. 1 As the variance of a RV decreases to zero.y . This is an application of the Chebyshev Inequality..
We assume that ρ < 1.y ).y = 1 and E(vx) = a + bρx. β) on the contour to the origin is given by d 2 (q ) = α 2 + β 2 = c 2 (1 + q 2 ) .6.1 Constant Contours Returning to the normalized jointly Gaussian RVs z1 and z2 . 1 − 2ρq + q 2 (5. (5.107) The square of the distance from a point (α.109) ±c 1 − 2ρq + q 2 .z2 (α. β) by ﬁnding the locus of points where the PDF is constant. we now investigate the shape of the joint PDF f z1 . Along the line β = q α we ﬁnd that α 2 (1 − 2ρq + q 2 ) = c 2 so that the constant contours are parameterized by α= and β= ±c q 1 − 2ρq + q 2 .110) . we ﬁnd that f z1 .y and b 2 = 1/(1 − ρx.90). so that v= 2 is independent of x and σv = 1. We require 2 2 σv = a 2 + b 2 + 2abρx.y 5. We have E(v) = 0. If ρ = 0 the contours where the joint PDF is constant is a circle of radius c centered at the origin. 2 Hence a = −bρx. y − ρx. (5.108) (5.106) where c is a positive constant.32 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution.y = 0. By inspection of (5. (5.z2 (α.y x 2 1 − ρx. β) is constant for α and β satisfying α 2 − 2ραβ + β 2 = c 2 .
114) (5.y (α.112) √ √ Consider the rotated coordinates α = (α + β)/ 2 and β = (β − α)/ 2.117) transforms the constant contour to (5. 1 β =√ 2 β − ηy α − ηx − σy σx (5. From (5. the constant contours for f x.113) √ The above equation represents an ellipse with major axis length 2c / 1 − ρ and minor axis √ length 2c / 1 + ρ.116) (5. Thus α 2 − 2ραβ + β 2 = c 2 is transformed into β2 c2 α2 + = . (5. β) are solutions to α − ηx σx Using the transformation 1 α =√ 2 β − ηy α − ηx + σx σy . the major and minor axes of the ellipse are along the lines β = ±α.116). With α = 0 we ﬁnd that one axis is along β − ηy α − ηx =− σx σy . Thus. 1+ρ 1−ρ 1 − ρ2 (5.93). (5. the line β = −α.118) 2 − 2ρ α − ηx σx β − ηy σy + β − ηy σy 2 = c 2. √ 2 The rotated coordinate system is a rotation by π/4 counterclockwise.115) (5.111) 2(1 − ρ) Similarly. the line β = α intersects the constant contour at ±c ±β = α = .STANDARD PROBABILITY DISTRIBUTIONS 33 Differentiating. intersects the constant contour at ±β = −α = ±c 2(1 − ρ) . (5. In the α − β plane. we ﬁnd that d 2 (q ) attains its extremal values at q = ±1. so that α +β =β √ 2 and α −β = α.
α =√ 1+ρ β =√ β 1−ρ exp − c2 . 2 1−ρ the length of this axis in the α − β plane is √ 2c 2 2 σx + σ y ±c σ y β=√ √ + ηy .117). 2 1−ρ 1−ρ .y on this curve is 1 2πσx σ y 1 − ρ 2 A further transformation α . 2(1 − ρ 2 ) (5. 2 1+ρ the length of this axis in the α − β plane is √ 2c 2 2 σx + σ y ±c σ y β=√ √ + ηy .119) transforms the ellipse in the α − β plane to a circle in the α − β plane: α 2 +β 2 = c2 . 2 1+ρ 1+ρ . Letting A denote the region bounded by the ellipse in the α − β plane and A denote the image (a circle) . 1 − ρ2 This transformation provides a straightforward way to compute the probability that the joint Gaussian RVs x and y lie within the region bounded by the ellipse speciﬁed by (5.34 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS with endpoints at ±c σx α=√ √ + ηx .117). With β = 0 we ﬁnd that the other axis is along β − ηy α − ηx = σx σy with endpoints at ±c σx α=√ √ + ηx . the value of the joint PDF f x. Points on this ellipse in the α − β plane satisfy (5.
β) d α dβ = A 1 2πσx σ y e − 2 (α +β ) d α dβ . β) = ∂β ∂α Computing the indicated derivatives. Substituting and transforming to polar coordinates. (b) η y .6. and (e) ρx. Additional techniques for treating these computations are given in [1. we have 1 √ √ σx 2 1 + ρ J (α. we ﬁnd f x. β) = −1 √ √ σx 2 1 − ρ so that J (α. (c) σx . (d) 2 σ y . . This bivariate Gaussian probability computation is one of the few which is “easily” accomplished.y . Given that x and y are jointly distributed Gaussian random variables with 2 2 E(yx) = 2 + 1. Drill Problem 5. Answers: 0.STANDARD PROBABILITY DISTRIBUTIONS 35 in the α − β plane.y (α. we have f x. E(xy) = 7/6 + y/6.y (α. and σxy = 0. 1.1. 956–958]. β) = 1 σx σ y 1 − ρ 2 .5. Determine: (a) ηx . 9. 1 √ √ σy 2 1 − ρ ∂α ∂β ∂β ∂β . β) d α dβ = A A 1 = 2π = 2π c / 1 − 1 (α 2 +β 2 ) e 2 d α dβ 2π √ 2 1−ρ r e − 2 r dr d θ 1 2 0 c 2 /(2−2ρ 2 ) 0 e −u d u 0 2 = 1 − e −c /(2−2ρ 2 ) . 5.75. 1 √ √ σy 2 1 + ρ . β) 1 2 2 where the Jacobian of the transformation is ∂α ∂α J (α. 1 − ρ 2 J (α.5x. pp. 2.
(c )P (−1 < x < 7).6. Important approximations to the Bernoulli PMF and Gaussian CDF are developed. 2. (b) the probability that.6. Drill Problem 5.1. 5. (g) n = 5. your friend gives you a dollar and if the die comes up with an odd number you give your friend a dollar. 0.6.3. Determine P (x ≤ 3) using the Bernoulli CDF table if: (a) n = 5.3. 2 2 σx = 21. With c 2 = 0. (e) n = 10. The length of time William Smith plays a video game is given by random variable x distributed exponentially with a mean of four minutes.1. (i) n = 20. Poisson and Gaussian distributions. p = 0. Determine: (a) the probability that William is still playing after four minutes. 2.3. he has played at least one game for more than four minutes.1. Bernoulli event probabilities may be approximated by the Poisson PMF when np < 10 or by the Gaussian PDF when npq > 3. (b) the length of the minor axis. p = 0.3889.6. His play during each game is independent from all other games. (h) n = 10. Answers: 3. p = 0.3. σ y = 31.2.36 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Drill Problem 5.2212 in (154).5212. (f ) n = 20. 0. (d) n = 5. 2.7 SUMMARY This chapter introduces certain probability distributions commonly encountered in biomedical engineering.3394. Suppose you are playing a game with a friend in which you roll a die 10 times. p = 0. p = 0. Random variables x and y are jointly Gaussian with ηx = −2. Answers: 0. σ y = 31. Find: (a) E(y  x = 0). η y = 3. (c) the length of the major axis. p = 0.1. Many important properties of jointly Gaussian random variables are presented.7. ﬁnd: (a) the smallest angle that either the minor or major axis makes with the positive α axis in the α − β plane. For the Poisson approximation use η = np. Special emphasis is placed on the exponential. Unfortunately. 2 2 σx = 21. and ρ = −0. (b)P (1 < y ≤ 10  x = 0). p = 0. (c) n = 20.899. If the die comes up with an even number. Random variables x and y are jointly Gaussian with ηx = −2. out of ﬁve games. p = 0. and ρ = −0. For the Gaussian approximation use η = np and σ 2 = npq . η y = 3.8 PROBLEMS 1. the die is loaded . (b) n = 10.1753.6. Assume x is a Bernoulli random variable.3394. 30◦ . Answers: exp(−1). p = 0. 5. Drill Problem 5.
(d) She is to keep rolling the ball until she gets a strike.150. In Junior Bioinstrumentation Lab. North Dakota. (c) from 7 to 9 baskets. (c) from 3 to 7 of the next 20 rolls are strikes. 9. a 5 or a 6. 8.000 cars to be started. The probability of a man hitting a target is 0. 4. (b) at least 4 of the next 20 rolls are strikes. Determine the: (e) expected number of strikes in 20 rolls. A dart player has found that the probability of hitting the dart board in any one throw is 0.7. Determine the probability it will take more than 5 rolls. Determine the probability that: (a) 3 of the next 20 rolls are strikes. If there are 20. ﬁnd the probability that: (a) at least 12. (b) exactly 12.2. 7.900 and 12. a 4. the probability that a car parked outside will start is 0. (g) standard deviation for the number of strikes in 20 rolls. On a frosty January morning in Fargo. one experiment introduces students to the transistor. what is the probability he will make: (a) at least 4 baskets. 3. The probability that Professor Rensselaer bowls a strike is 0. Determine: (b) the expected number of destroyed transistors. (b) 4 baskets. (a) If he tries 15 times to hit the target. The probability of a student destroying a transistor is 0. what is the probability that exactly 12 cars will start and 8 will not? (b) What is the probability that the number of cars starting out of 20 is between 9 and 15.100 will start. (b) the probability of your friend winning more than 4 dollars. what is the probability of him hitting it at least 5 but less than 10 times? (b) What is the average number of hits in 30 tries? (c) What is the probability of him getting exactly the average number of hits in 30 tries? (d) How many times must the man try to hit the target if he wants the probability of hitting it to be at least 2/3? (e) What is the probability that no more than three tries are required to hit the target for the ﬁrst time? 6. (c) the probability that fewer than 2 transistors are destroyed. If he makes 10 attempts. (c) the number starting is between 11.6.3.000 will start.6? . (e) the expected number of baskets. (d) less than 2 baskets.STANDARD PROBABILITY DISTRIBUTIONS 37 so that a 1 or a 3 are three times as likely to occur as a 2.4. The probability that a basketball player makes a basket is 0. Consider Problem 7. Each student is given only one transistor to use. (d) the number starting is less than 12. Determine: (a) how many dollars your friend can expect to win in this game. How many times must he throw the dart so that the probability of hitting the dart board is at least 0.500. Let random variable x show the possible numbers of students who destroy transistors. (a) Sketch the PMF for x. (f ) variance for the number of strikes in 20 rolls. One lab class has 5 students and they will perform this experiment next week.2. 5. (a) If we take a sample of 20 cars.
Determine: (a) q.) (a) Determine the probability that a student takes the course more than once. What is the probability of at least 6 alpha particles being emitted in 1 min? 19. 17. You give him another one page letter to type. (b) 10 min? (c) Many years later. Suppose x is a Bernoulli random variable with η = 10 and σ 2 = 10/3. 11. Professor Rensselaer gives 11 quizzes per quarter in Introduction to Random Processes. Suppose the average number of times a student takes Introduction to Random Processes is 1. (c) p. the material averages 6 alpha particles emitted per minute. 16. the course instructor. (b) exactly 11 quizzes are given during the quarter. Let random variable x be Bernoulli with n = 15 and p = 0.38 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 10. a sample of radioactive material emits 20 alpha particles per minute. Repeat Problem 1. (d) at most 9 quizzes are given during the quarter. Determine the probability that the student makes: (a) at least 1 mistake.4. (b) n. what is the probability that he makes exactly 30 mistakes? (b) The typist decides to take typing lessons. (b) The academic vicepresident of FPI wants to ensure that on the average. (a) If you give him a one page letter to type. 12. An electronics manufacturer is evaluating its quality control program. he averages 5 mistakes per page. On the average. A certain intersection averages 3 trafﬁc accidents per week. What is the probability of him making fewer than 5 mistakes. What proportion of 20% defective components will be shipped? 13. what is the probability of him making fewer than 50 mistakes in a 25 page report? 18. using the Poisson approximation to the Bernoulli PMF. 14. The current procedure is to take a sample of 5 from 1000 and pass the shipment if not more than 1 component is found defective. (b) from 3 to 5 mistakes. thinks so many students repeat the course because they enjoy it so much. Determine the probability that: (a) from 8 to 12 quizzes are given during the quarter.5. a student may take a course as many times as desired. At Fargo Polytechnic Institute (FPI). (c) at least 10 quizzes are given during the quarter. To what value should the mean be adjusted to ensure this? . when appropriate. and. On the average. What is the probability of 10 alpha particles being emitted in: (a) 1 min. 80% of the students take the course at most one time. Suppose that on the average. (c) With the 5 mistakes per page average. a student makes 6 mistakes per test. (c) exactly 2 mistakes. (Professor Rensselaer. after the lessons. Suppose a typist makes an average of 30 mistakes per page. (d) more than the expected number of mistakes. What is the probability that more than 2 accidents will occur during any given week? 15. Determine E(x 2 ).
(b) more than 15 defective transistors in a sample of 1000 transistors.STANDARD PROBABILITY DISTRIBUTIONS 39 20. determine: (a) E(z). (a) Every time a carpenter pounds in a nail.01. The manufacturer of Leaping Lizards. This year. k! k = 0. (b) σz. The distribution for the number of students seeking advising help from Professor Rensselaer during any particular day is given by P (x = k) = 3k e −3 . what is the probability that none are perfect? 22. 23. a bran cereal with milk expanding (exploding) marshmallow lizards.002. A perfect car is assembled with a probability of 2 × 10−5 . If in building a house he pounds 1250 nails.000 cars are produced in a month. Assuming that . . 95% of the spoonfuls will each have at least one lizard. If random variable z equals the total number of students Professor Rensselaer helps each day. 21. what is the probability of him hitting his thumb at least once while working on the house? (b) If he takes ﬁve extra minutes off every time he hits his thumb. Determine the probability that there are: (a) 3 defective transistors in a sample of 200 transistors. (c) 0 defective transistors in a sample of 20 transistors. 1. how many extra minutes can he expect to take off in building a house with 3000 nails? 24. to what value should the mean of the lizards per spoonful be set at to ensure this? 25. that person will get nothing. wants to ensure that on the average. Assuming that the lizards are randomly distributed in the cereal box. If it is. The PDF for the time interval between students seeking help for Introduction to Random Processes from Professor Rensselaer during any particular day is given by f t (τ ) = e −τ u(τ ). . Suppose 1% of the transistors in a box are defective. . the probability that he hits his thumb is 0. on its anniversary day. The caller will ask the person answering the phone if it’s his or her birthday. Determine the probability that fewer than 9 students major in Bioengineering. The probability that a student will major in Bioengineering is 0. FPI admits only 1000 freshmen per year. 26. then that lucky person will be mailed a brand new programmable calculator. Otherwise. a computer store is going to run an advertising campaign in which the employees will telephone 5840 people selected at random from the population of North America. . If 15.
ﬁnd the probability that: (a) the computer store mails out exactly 16 calculators. 27. Find: (a) P (h > 74). The time it takes a runner to run a mile is equally likely to lie in an interval from 4. A certain transistor has a current gain. (f ) P (x is an integer).2). (d) f xA (αA). (g) FxB (αB).1 < x ≤ −0.1 to 4. ﬁnd: (a) the mean. when appropriate. C: 70–80. 37. Determine: (a) the probability it takes the runner exactly 4 min to run a mile. (e) FxA (αA). Assuming the battery lives are Gaussian distributed. Repeat Problem 1. (b) the computer store mails out from 20 to 40 calculators. what is the probability that grades will be equally distributed? 36.98 years. (b) variance. Event A = {0 < x ≤ 2} and B = {−1 < x ≤ 0} ∪ {1 < x ≤ 2}. Find the probability that a bulb burns between 1000 and 1300 h. 34. (d) P (−1. 28. . 35. Random variable x is uniform between −2 and 3. Consider Problem 36. B: 80–90.5% of their batteries last more than 3. A light bulb manufacturer distributes light bulbs that have a length of life that is normally distributed with a mean equal to 1200 h and a standard deviation of 40 h. 33. (b) P (49 < r < 54). ﬁnd: (a) the probability that a particular resistor is within 2 ohms of the mean. Assume x is a standard Gaussian random variable. (b) ηx . (b) P (x < 0). 29. (b) P (73 < h ≤ 80). (b) the probability it takes the runner from 4. Consider Problem 32.10. Grades are assigned as follows: A: 90–100. D: 60–70.583 ≤ x < 1. using the Gaussian approximation to the Bernoulli PMF. (b) Find the probability that a particular resistor is within 2 ohms of the mean.9 and A. A certain type of resistor has resistance values that are Gaussian distributed with a mean of 50 ohms and a variance of 3.471).0 to 4.40 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS the person answering the phone won’t lie and that there is no such thing as leap year. (c) σx . A battery manufacturer has found that 8.3 years and 2.15 min. Assume that the scores on an examination are Gaussian distributed with mean 75 and standard deviation 10. 30. In a class of 25 students. (c) P (h − ηh  < 3σh ). h.08% of their batteries last less than 2. (a) Write the PDF for the resistance value. (f ) f xB (αB). and F: below 60. (c) P (x < 0. determine: (a) P (x = 0). that is Gaussian distributed with a mean of 77 and a variance of 11.2 min. Find: (a) P (−1 < x < 0). Using Tables A. (e) P (−2. 32. If resistances are measured to the nearest ohm. Find the value of d so that the range 77 + d covers 95% of the current gains.5). 31. (c) Find P (49 < r < 54).
A 250 question multiple choice ﬁnal exam is given. 44.8835. (c) Find σx using direct integration. ∞). A large box contains 10. (b) Using the results of (a).0 and a standard deviation of 0. y ∼ G(3.67. Find: (a) P (1 < y < 7  x = 0). how many resistors have resistances that are within 10% of the average? 42. Assume x ∼ G(3. Assume x and y are jointly Gaussian with x ∼ G(2. Consider Problem 45.y (∞. Determine the probability that a student guesses the correct answers for 20–25 of 85 questions about which the student has no knowledge.y . and ρx.y = −5. Draw a sketch of the joint Gaussian constant contour for x and y. The RV x has PDF f x (α) = a exp − 1 (α − η)2 . (c) P (−1 < x < 1. y ∼ G(1. 43. 8). 39. 1). 46. 45. and ρx. (e) ρx. 47.) (b) Using direct integration. determine: (a) P (x = 78). 40.y = −0. 9). (b) f yx (β0). Suppose a system contains a component whose length is normally distributed with a mean of 2. ﬁnd E(x). (b) P (1 < y < 7). Assuming a Gaussian distribution for random variable x describing her golf game. (Hint: assume RV y is independent of x and has PDF f y (β) = 2 f x (β) and evaluate Fx. Assume x and y are jointly distributed Gaussian random variables with x ∼ G(−2. (d) P (3 < x < 10). 1). Determine: (a) ηx . Determine: (a) E(yx = 0). If the average resistance is 1000 ohms with a standard deviation of 200 ohms.5.2. 1 < y < 7). (b) P (x ≤ 78). (c) σx . (b) η y . (c) P (70 < x ≤ 80). and σ yx = 3.5y. (c) P (0 < y < 4x = 0).32x. and that x and y are jointly Gaussian with ρx. . (d) σ y . Suppose x and y are jointly distributed Gaussian random variables with E(yx) = 2.8 + 0. (d) the probability that x is less than 75 if the score is measured to the nearest unit.1? 41.000 resistors with resistances that are Gaussian distributed. Each question has 5 possible answers and only one correct answer.STANDARD PROBABILITY DISTRIBUTIONS 41 38. 4). The average golf score for Professor Rensselaer is 78 with a standard deviation of 3. 2σ 2 (a) Find the constant a. y ∼ G(−2. (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 .y . (f ) σx. 13). what is the probability that at least 2 have a length greater than 2. E(xy) = −1 + 0. If 5 of these components are removed from different systems.y = 0. Draw a sketch of the joint Gaussian contour equation showing the original and the translatedrotated sets of axes.
Assume x and y are jointly Gaussian with x ∼ G(−1.5. 56. . 7). William Smith is a varsity wrestler on his high school team. and ρx. Without exception. 51. (d) P (−3 < x < 0). (c) 99%. Determine the probability that at most three will be alive at the end of 2. Determine: (a) E(y  x = 0). 52. (b) 95%.3 years. (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 . If 20 of these systems are installed. Determine the probability that the circuit is operating successfully at 10. Random variable t is exponential with η = 2.0433 years. A component with an exponential failure rate is 90% reliable at 10.12.6. 4). Consider Problem 47. 54.000 h. 55. determine the probability that 10 are operating at the end of 2. (c) P (0 < y < 4x = 0). William’s 1 A 3 2 B 4 FIGURE 5. (b) f yx (β  0). For successful operation of the circuit. Suppose a system has an exponential failure rate in years to failure with η = 2. Consider Problem 49. Determine the number of years reliable at: (a) 90%. y ∼ G(1. 50. where T is a constant. Determine: (a) P (t > η). (b) Using the results of (a). (c) P (0 < y < 4x = 0). In the circuit shown in Figure 5. at least two of these components must connect A with B. each of the four components operate independently of one another and have an exponential failure rate (in hours) with η = 105 . if he does not pin his opponent with his trick move.378.y = 0. 57. (b) f y  x (β  0). Five individuals have this cancer.000 h.12: Circuit for Problem 54.42 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 48. 49. Determine: (a) E(y  x = 0). (d) P (3 < x < 10). The survival rate of individuals with a certain type of cancer is assumed to be exponential with η = 4 years. Draw a sketch of the joint Gaussian contour for x and y. Consider Problem 52. (b) f z(α) if z = t − T. he loses the match on points. 53. Determine the number of hours reliable at 95%.
STANDARD PROBABILITY DISTRIBUTIONS 43 trick move also prevents him from ever getting pinned. The average time between power failures in a Coop utility is once every 1. Determine: (a) the probability that there will be at least one power failure during the coming year. Determine: (a) your expected waiting time to make a purchase. Each customer takes 4 min to check out.4 years.3 people long. Suppose Fargo Community Hospital has a backup emergency generator to provide auxilary power during a power failure. Determine the probability that William Smith: (a) pins his opponent during the ﬁrst period: (b) pins his opponent during the second period: (c) pins his opponent during the third period: (d) wins the match. prints brochures describing their vacation packages as “Unique. (c) the probability that you will have more than ﬁve people in the queue ahead of you. period 3 : f xA (αA) = 0. (b) the probability that there will be at least two power failures during the coming year. Find the probability that William Smith wins: (a) at least 4 of his ﬁrst 5 matches.1149599α)u(α). 58. Consider Problem 59 statement.” The vacations certainly live up to their advertisement.4598394 exp(−0. On the average.2299197α)u(α). In fact. Consider Problem 57. The length of time it takes William to pin his opponent in each period of a wrestling match is given by: period 1 : f x (α) = 0. Assume that a power failure will last at least 24 h. Outer Space Adventures. (b) more matches than he is expected to during a 10 match season.2299197 exp(−0. (b) the probability that you will have less than four people in the queue ahead of you. the emergency generator has an expected time between failures of once every 200 h. the queue is 6. following an exponential distribution. Moreover. 62.1149599 exp(−0. 60. Assume each period is 2 min and the match is 3 periods. 59. What is the probability that the hospital will be without power during the next 24 h? 61. 9 months later)? . Suppose that you have signed up for a vacation trip that starts at the end of this quarter.4598394α)u(α). having an average length of 6 months. Inc. Unexpected and Unexplained. period 2 : f xA (αA) = 0. The queue for the cash register at a popular package store near Fargo Polytechnic Institute becomes quite long on Saturdays following football games. they are so unexpected that the length of the vacations is random. What is the probability that you will be back home in time for next fall quarter (that is. where A = {Smith did not pin his opponent during the previous periods}.
the bulbs were taken from the ﬁxtures and placed in a box. The failure rate of these light bulbs follows an exponential PDF. HOURS 1000 750 500 1500 NUMBER OF BULBS 125 250 150 225 After the speciﬁed length of time. Sevenhundred and ﬁfty of these bulbs were put in light ﬁxtures in four rooms.44 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 63. If a bulb is selected at random from the box. A certain brand of light bulbs has an average lifeexpectancy of 750 h. what is the probability that it is burnt out? . The lights were turned on and left that way for a different length of time in each room as follows: ROOM 1 2 3 4 TIME BULBS LEFT ON.
+∞}) = 0. ) and P (z(ζ ) ∈ {−∞. a full wave rectiﬁer circuit produces an output that is the absolute value of the input. we start the discussion of computing the probability distribution of z = g (x) with the simplest case.1. with z(ζ ) = g (x(ζ )). the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. (Recall that if g is a monotonic function then a onetoone correspondence between g (x) and x exists. The input/output characteristics of many physical devices can be represented by a nonlinear memoryless transformation of the input. For example. consequently.) Then. Proceeding from the deﬁnition of the CDF for z. . and we wish to ﬁnd the CDF for random variable z = g (x). The ease of use of the CDF technique depends critically on the functional form of g (x).1 CDF Technique with Monotonic Functions Consider the problem where the CDF Fx is known for the RV x. P ). is a random variable on the probability space (S. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. and mixed random variables. a continuous monotonic function g . we require z to be a measurable function on the measurable space (S. 6. and for all types of continuous. To make the CDF technique easier to understand. Next. we have for a . discrete. The CDF technique is applicable for all functions z = g (x). The primary subjects of this chapter are methods for determining the probability distribution of a function of a random variable. the difﬁculties associated with computing the probability distribution of z = g (x) are investigated when the restrictions on g (x) are relaxed. Of course. Then. We will refer to this method as the CDF technique.1 UNIVARIATE CDF TECHNIQUE This section introduces a method of computing the probability distribution of a function of a random variable using the CDF.45 CHAPTER 6 Transformations of Random Variables Functions of random variables occur frequently in many applications of probability theory. 6. we require that the function z : S → R∗ .
we obtain Fz(γ ) = which is also a uniform distribution. (γ − 1)/8. Additionally. if z = g (x) and g is monotone decreasing. Example 6. In general. 6.46 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS monotonic increasing function g (x) Fz(γ ) = P (z = g (x) ≤ γ ) = P (x ≤ g −1 (γ )) = Fx (g −1 (γ )). more than one solution of z = g (x) can exist—resulting in a manytoone mapping from x to z. Similarly. Since random variable x is uniformly distributed. (γ − 1)/2 < 0 ⎨ γ −1 Fz(γ ) = Fx = (γ − 1)/8. In this general case. ⎧ ⎪ ⎨ (6. In fact.1). 4 ≤ α.1. the relationship between x and z can take on any form. (6. with the argument α replaced by g −1 (γ ).1.1) where g −1 (γ ) is the value of α for which g (α) = γ . Find the CDF for random variable z = 2x + 1. we see that g is monotone increasing and that g −1 (γ ) = (γ − 1)/2. the function does not have to be monotonic.1) indicates. the CDF of x is ⎧ ⎪ 0. then Fz(γ ) = P (z = g (x) ≤ γ ) = P (x ≥ g −1 (γ )) = 1 − Fx (g −1 (γ )− ). this case is conceptually no more difﬁcult than . The following example illustrates this technique. Random variable x is uniformly distributed in the interval 0 to 4. the CDF for z is given by ⎧ ⎪ 0. the only requirement on g is that z = g (x) be a random variable. 4 ≤ (γ − 1)/2. including discontinuities. Fz(γ ) is no longer found by simple substitution. 0 ≤ (γ − 1)/2 < 4 ⎪ 2 ⎩ 1. Letting g (x) = 2x + 1. γ <1 1≤γ <9 9 ≤ γ. As (6. α<0 ⎨ Fx (α) = α/4. the CDF of random variable z is written in terms of Fx (α).2) 0. Applying (6. 0 ≤ α < 4 ⎪ ⎩ 1. under these conditions it is impossible to write a general expression for Fz(γ ) using the CDF technique. ⎪ ⎩ 1. Simplifying.1. Solution. However.2 CDF Technique with Arbitrary Functions In general.
Note that A(γ ) = g −1 ((−∞.5) Note that the intervals as well as the number of nonempty intervals depends on γ . Partition A(γ ) into disjoint intervals {Ai (γ ) : i = 1.9) (6. Find the CDF for the RV z = x 2 . Fz(γ ) = ∞ i=1 P (x ∈ Ai (γ )).8) (6.1. 2.6) The above probabilities are easily found from the CDF Fx .10) (6. α < −1 −1 ≤ α < 1 1 ≤ α.3) (6. γ ]). Similarly. b i (γ )]. If interval Ai (γ ) is of the form Ai (γ ) = (a i (γ ). (6. then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )− ). Random variable x has the CDF ⎧ ⎪ 0. b i (γ )]. Since the Ai ’s are disjoint. i=1 (6. Using this method. . (6. . then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )). . ⎪ ⎩ 1. The following examples illustrate various aspects of this technique. Then Fz(γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )).} so that A(γ ) = ∞ (6. if interval Ai (γ ) is of the form Ai (γ ) = [a i (γ ). Example 6. .TRANSFORMATIONS OF RANDOM VARIABLES 47 the previous case.7) The success of this method clearly depends on our ability to partition A(γ ) into disjoint intervals. Let A(γ ) = {x : g (x) ≤ γ }. ⎨ Fx (α) = (3α − α 3 + 2)/4. and involves only careful book keeping. any function g and CDF Fx is amenable to a solution for Fz(γ ).4) Ai (γ ).2.
α < −3 ⎨ Fx (α) = (α + 3)/6. we ﬁnd A(γ ) = g −1 ((−∞. Random variable x is uniformly distributed in the interval −3 to 3. γ ]. ⎪ ⎪ ⎪ 3x + 5. γ +1 − Fx 3 1 −γ − 1 3 − γ < −1 −1 ≤ γ < 2 2 ≤ γ. ∪ . √ √ Fz(γ ) = Fx ( γ ) − Fx ((− γ )− ). The CDF for random variable x is ⎧ ⎪ 0. Random variable z is deﬁned by ⎧ x < −2 ⎪ −1. γ ]) = so that ∅. ⎧ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎩ Fx γ −5 3 0.1. γ < −1 . −1 ≤ γ < 2 2 ≤ γ. −3 ≤ α < 3 ⎪ ⎩ 1. Let A(γ ) = g −1 ((−∞. Noting that Fx (α) is continuous and has the same functional form for −1 < α < 1. 1 ≤ x. . Letting g (x) = x 2 . √ √ [− γ . Solution. we obtain ⎧ ⎪ 0. γ <0 ⎨ √ √ Fz(γ ) = (3 γ − ( γ )3 )/2. γ <0 γ ≥ 0. −1 ≤ x < 0 ⎪ ⎪ 3x − 1. ∞)).1 we ﬁnd ⎧ ∅. Fz(γ ) = + Fx . ⎪ 0≤x<1 ⎪ ⎪ ⎩ 2. ⎪ 3 3 3 ⎪ ⎩ R Consequently. Example 6. Find Fz(γ ). Plots for this example are given in Figure 6.48 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. 3 ≤ α. ⎪ ⎪ ⎨ γ −5 −γ − 1 γ + 1 A(γ ) = −∞. Referring to Figure 6.1.3. ⎪ −2 ≤ x < −1 ⎨ z = g (x) = −3x − 1. 0≤γ <1 ⎪ ⎩ 1 1 ≤ γ.
Fz(γ ) = Note that Fz(γ ) has jumps at γ = −1 and at γ = 2. As presented in the previous example. 6 γ < −1 −1 ≤ γ < 2 2 ≤ γ.3. dependent on the CDF for random variable x.TRANSFORMATIONS OF RANDOM VARIABLES g(x) 4 2 x 49 −3 −1 −1 Fx (α) 1 1 3 −3 −1 Fz (γ) 1 1 3 α −3 −1 1 3 γ FIGURE 6. then random variable z can be continuous. Moreover. Substituting. Random variable z is continuous if x is continuous and g (x) is not equal to a constant over any interval where f x (α) = 0. Whenever random variable x is continuous and g (x) is constant over some interval or intervals. . z = g (x) is a discrete random variable if g (α) changes values only on intervals where f x (α) = 0. we obtain ⎧ ⎪ ⎪ ⎨1 ⎪6 ⎪ ⎩ 0. This results in a jump in Fz and an impulse function in f z. z is a mixed random variable whenever g (α) is constant in an interval where f x (α) = 0.1. random variable z is mixed due to the constant value of g (x) in the intervals −3 ≤ x < −2 and 1 ≤ x < 3. γ −5 γ + 1 −γ − 1 +3+ − 3 3 3 1 γ +2 = . In fact. discrete or mixed.1: Plots for Example 6.
⎪ ⎪ ⎩ 1. −1 ≤ α < 2 ⎪ ⎩ 1.5). 0].4.5]. we ﬁnd ⎧ ⎪ (−∞. .1. ⎪ ⎪ ⎨ F (0). Random variable x has the PDF f x (α) = (1 + α 2 )/6. 0. ⎪ ⎩ 1. α < −1 ⎨ 3 Fx (α) = (α + 3α + 4)/18. 0≤γ <1 ⎪ ⎪ ⎩ (−∞.2. γ + 1]. x≤0 0 < x ≤ 0. −1 ≤ γ < 0 A(γ ) = {x : g (x) ≤ γ } = ⎪ (−∞. Find the PDF of random variable z deﬁned by ⎧ ⎪ x − 1.5 < x. γ ≤ −1 ⎪ ⎪ ⎨ (−∞. 2 ≤ α. −1 < α < 2 otherwise.2: Transformation for Example 6. Consequently. −1 −1 −2 1 2 FIGURE 6.50 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6. 0. Solution.2 shows a plot of g (x). we evaluate Fz ﬁrst. To ﬁnd f z. With the aid of Figure 6. The CDF for random variable x is ⎧ ⎪ 0. ⎧ ⎪ Fx (γ + 1). x Fz(γ ) = ⎪ Fx (0. Figure 6.4. then differentiate this result.1. 1 ≤ γ. g( x ) 1 x γ ≤ −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ.5 0. ∞). ⎨ z = g (x) = 0.
continuous and discontinuous. γ < ln(0. This technique is applicable for all functions z = g (x).6. ∞).1) ≤ γ . ⎪ ⎪ ⎩ 1. The previous examples illustrated evaluating the probability distribution of a function of a continuous random variable using the CDF technique. 1 − e −0. Note that P (x ≤ 0) = 0.5). ⎨ Fx (α) = α/10. For γ > 0. f z(γ ) = 13 11 3γ 2 + 6γ + 6 (u(γ + 2) − u(γ + 1)) + δ(γ ) + δ(γ − 1). Additionally. Differentiating Fz. ⎪ ⎨ Fz(γ ) = 2/9. the CDF technique is used without any changes or modiﬁcations as shown in the next example.5(u(α) − u(α − 1)) + 0. We ﬁnd Fz(γ ) = 0. 18 144 16 Example 6. ⎪ ⎪ ⎪ 5/16.1γ . as required since g (x) = − ln(x) is not deﬁned (or at least not real–valued) for x ≤ 0.5.5δ(α − 0.1) ln(0. respectively.1. For mixed random variables. The CDF for x is ⎧ ⎪ 0. Find the CDF for z = g (x) = 1/x 2 . Random variable x has PDF f x (α) = 0.TRANSFORMATIONS OF RANDOM VARIABLES 51 Substituting. the CDF technique is applicable if random variable x is mixed or discrete. α<0 0 ≤ α < 10 10 ≤ α. Note that Fz(γ ) has jumps at γ = 0 and γ = 1 of heights 13/144 and 11/16. ⎪ ⎪ ⎪ ⎪ ((γ + 1)3 + 3γ + 7)/18. γ < −2 −2 ≤ γ < −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ. we ﬁnd A(γ ) = {x : − ln(x) ≤ γ } = (e −γ . so that Fz(γ ) = 1 − Fx ((e −γ )− ). . Random variable x is uniformly distributed in the interval from 0 to 10. ⎧ 0. ⎪ ⎩ 1. Solution. Example 6. Find the CDF for random variable z = g (x) = − ln(x).1.
The mixed random variable x has CDF ⎧ ⎪ 0.5α. γ <1 −1/2 Fz(γ ) = 0. γ √ Since Fx (−1/ γ ) = 0 for all real γ . 0 ≤ α < 0. Drill Problem 6. Random variable x is uniformly distributed in the interval −1 to 4.5 √ 0. we have ⎧ ⎪ 0.5 + 0. (d ) f z(1/2).∞ . x<1 −1 ≤ x ≤ 1 x > 1. Random variable x has the PDF f x (α) = 0. After some algebra.1.1.5γ .5α. Fz(γ ) = 0. 1/4. Answers: 0. .5γ −1/2 . ⎪ ⎪ ⎨ 0. 1/16. 4 ≤ γ. 1 A(γ ) = {x : x −2 ≤ γ } = −∞. √ (1/ γ )− < 0 √ 0 ≤ (1/ γ )− < 0. 1 ≤ α. 1≤γ <4 ⎪ ⎩ 1 − 0. ⎧ ⎪ ⎨ √ √ Fz(γ ) = Fx (−1/ γ ) + 1 − Fx ((1/ γ )− ).1. Determine: (a) Fz(0). ⎪ ⎪ ⎩ 1.5 − 0. (c ) f z(0). α<0 ⎪ ⎪ ⎨ 0. ⎪ ⎩ 1.2.5α(u(α) − u(α − 2)). Random variable z is deﬁned by ⎧ ⎪ −1. 2/15. 0. (c )Fz(3/2).5 ≤ (1/ γ )− < 1 √ − 1 ≤ (1/ γ ) . 1/15. ⎨ z= x.5 Fx (α) = ⎪ 0. Answers: 0. (b)Fz(1/2).5γ −1/2 . Fz(γ ) = 1 − ⎪ 0. (c ) f z(15). For γ > 0.5 ≤ α < 1 ⎪ ⎪ ⎩ 1.5γ −1/2 . 0. 1/15.52 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. Determine: (a) Fz(−1/2). − √ γ so that 1 ∪ √ . (b)Fz(1).5 + 0. Drill Problem 6. Random variable z = 3x + 2. 1. For γ < 0.
(d )Fz(4). (b)Fz(1/2). Determine: (a) Fz(1/8). The PDF technique. i = 1. (c )Fz(3/2).1. Second.0519. (c )Fz(4). First.5α(u(α) − u(α − 2)). x + 0. 2. . Let αi = αi (γ ) = g −1 (γ ). is only applicable for functions of random variables in which z = g (x) is continuous and does not equal a constant in any interval in which f x is nonzero. ⎪ ⎩ 3. 0.2 UNIVARIATE PDF TECHNIQUE The previous section solved the problem of determining the probability distribution of a function of a random variable using the cumulative distribution function. a method for handling mixed random variables with the PDF technique is introduced.089. we consider computing the conditional PDF of a function of a random variable using the PDF technique. Assume g is continuous and not constant over any interval for which f x = 0 . Now.022.2.5. however.11) .5 0. 6. 6. Answers: 1/4. Determine: (a) Fz(−1). Finally. 1. we will ﬁnd the PDF method most useful in extensions to multivariate functions.2.1.1. we introduce a second method for calculating the probability distribution of a function z = g (x) using the probability density function. we discuss a wide variety of situations using the PDF technique with functions of continuous random variables. In this section. in many situations it is much simpler to use than the CDF technique. Then. (d ) f z(4). .617. 0. Random variable x has PDF f x (α) = e −α−1 u(α + 1). Random variable x has the PDF f x (α) = 0. Drill Problem 6. We introduce the PDF technique for two reasons. Answers: 0. 1/16. . Let x be a continuous RV with PDF f x (α) and let the RV z = g (x) . Random variable z = 1/x 2 .5 < x ≤ 1 x > 1.4. 0. (b)Fz(0). 1/16. . called the PDF technique. Random variable z is deﬁned by z= ⎧ ⎪ ⎨ −1. x ≤ 0.3.1 Continuous Random Variable Theorem 6.TRANSFORMATIONS OF RANDOM VARIABLES 53 Drill Problem 6. (6.
h)) − g (a i (γ . Let h > 0 and deﬁne I (γ . h→0 ∞ i=1 ∞ i=1 ∞ i=1 f x (αi (γ )) . h) = Then Fz(γ ) − Fz(γ − h) = By hypothesis. .12) i = 1. h))) lim a i (γ . h). h) The desired result follows by taking the product of the above limits. h) = {x : γ − h < g (x) ≤ γ }. . whereas g (1) may be positive or negative. Proof. (Fx (b i (γ . Then f z(γ ) = where we interpret f x (αi (γ )) = 0.54 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS denote the distinct solutions to g (αi ) = γ . b i (γ . h)) − Fx (a i (γ . h) = lim b i (γ . h) = αi (γ ). 2. . Ii (γ . Partition I (γ . h) b i (γ . h) . . h)). h)) h g (αi (γ )) lim and that h→0 lim Fx (b i (γ . such that I (γ . The absolute value appears because by construction we have b i > a i and h > 0. . h) − a i (γ . g (1) (αi (γ )) (6. b i (γ . h) = (a i (γ . h) into disjoint intervals of the form Ii (γ . g (1) (αi (γ )) if f x (αi (γ ) = 0. h). h) − a i (γ . h→0 Note that (for all γ with f x (αi (γ )) = 0) 1 b i (γ . = lim = (1) h→0 h→0 g (b i (γ . h)) = f x (αi (γ )). h) − a i (γ . h)) − Fx (a i (γ .
75(1 − α 2 )(u(α + 1) − u(α − 1)). f z(γ ) = or f z(γ ) = 0.75(1 − γ −1 )(u(1 − γ −1/2 ) − 0 + 1 − u(γ −1/2 − 1)) u(γ ). given by α1 = (γ − 1)/2. Since γ = g (α) = α −2 . For γ < 0. Random variable x has PDF f x (α) = 0. 0. so that f z(γ ) = 0 for γ < 0. We easily ﬁnd g (1) (α) = 2. Solution.2. . γ and 1 α2 (γ ) = √ . 2γ 3/2 0. In this case.2. Random variable x is uniformly distributed in the interval 0–4. Find the PDF for random variable z = g (x) = 1/x 2 .2. Example 6.1. γ 1/8. Solution.75(γ − 2 − γ − 2 )u(γ − 1). and f z(γ ) = Substituting. g (1) (αi ) = 2/αi 3 = 2γ 3/2 .2. there are no solutions to g (αi ) = γ . 2γ 3/2 0. Random variable x has PDF 3 5 f x (−γ −1/2 ) + f x (γ −1/2 ) u(γ ).TRANSFORMATIONS OF RANDOM VARIABLES 55 Example 6. For γ > 0 there are two solutions to g (αi ) = γ : 1 α1 (γ ) = − √ . 6 Find the PDF for random variable z = g (x) = x 2 . f z(γ ) = Simplifying. Hence f z(γ ) = f x ((γ − 1)/2)/2 = Example 6. 2γ 3/2 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)).75(1 − γ −1 )(u(γ − 1) − 0 + 1 − u(1 − γ )) u(γ ). 1<γ <9 otherwise. Find the PDF for random variable z = g (x) = 2x + 1. we have g (1) (α) = −2α −3 . there is only one solution to the equation γ = 2α + 1. hence.3.
2. f z(γ ) = Simplifying. ⎨ f z(γ ) = (γ −1/2 + γ 1/2 )/12.2. To illustrate the use of this technique.5) + δ(α − 0. and α2 (γ ) = γ . Here. 8 8 8 Find the PDF for the RV z = g (x) = e −x . so that f z(γ ) = 0 for γ < 0. We have g (1) (α1 ) = −e ln(γ ) = −γ . f z(γ ) = or 1+γ √ (u(1 − γ ) − 0 + 1 − u(γ − 4))u(γ ). For γ > 0 there are two solutions to g (αi ) = γ : √ √ α1 (γ ) = − γ . Example 6. 12 γ ⎧ ⎪ (γ −1/2 + γ 1/2 )/6. 6. f z(γ ) = √ 2 γ Substituting. 1+γ √ √ √ √ √ (u(1 − γ ) − u(−2 − γ ) + u( γ + 1) − u( γ − 2))u(γ ).5 . respectively.5).56 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. g (1) (αi ) = 2αi  = 2 γ . consider the following example. For γ < 0 there are no solutions to g (αi ) = γ .5 are mapped to probability masses of 1/8 for z at e 0. and we wish to ﬁnd the PDF for z = g (x). and √ √ f x (− γ ) + f x ( γ ) u(γ ). hence.2 Mixed Random Variable Consider the problem where random variable x is mixed. There is only one solution to g (α) = γ : α1 (γ ) = − ln(γ ). Random variable x has PDF 1 1 3 f x (α) = (u(α + 1) − u(α − 1)) + δ(α + 0. and then combine the results to yield f z. 12 γ 0<γ <1 1<γ <4 elsewhere.4.5 and 0. ⎪ ⎩ 0.5 and e −0. we have g (1) (α) = 2α. The probability masses of 1/8 for x at −0. we treat the discrete and continuous portions of f x separately. √ Since γ = g (α) = α 2 . The continuous part of the PDF of x is handled by the PDF technique. For all γ > 0 such that . Solution.
6 Find the PDF for random variable z = g (x) = x 2 . we have f zA (γ A) = f z(γ ) . (6. one of the following two methods may be used to determine the conditional PDF of z using the PDF technique. We have P (A) = 1 6 2 0 7 (1 + α 2 )d α = . Random variable x has the PDF 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)). Then.5. Depending on whether the event A is deﬁned on the range or domain of z = g (x). Consider the problem where random variable x has PDF f x . based on f xA . f zA . by the deﬁnition of a conditional PDF. the above techniques apply to ﬁnd the conditional PDF for z = g (x). g (1) (αi (γ )) (6. (ii) If A is an event deﬁned for an interval on x.14) Example 6.2. γ 6. given that event A occurred.2.3 Conditional PDF Technique Since a conditional PDF is also a PDF. is computed by ﬁrst evaluating f z using the technique in this section.5 ) + δ(γ − e −0. and we wish to evaluate the conditional PDF for random variable z = g (x).TRANSFORMATIONS OF RANDOM VARIABLES 57 α1 (γ ) ± 0. we will use the conditional PDF of x to evaluate the conditional PDF for z as f zA (γ A) = ∞ i=1 f zA (αi (γ )A) . First.5 ).5 > 0 we have f z(γ ) = Combining these results. the conditional PDF. we solve for the conditional PDF for x and then ﬁnd the conditional PDF for z. given event A. P (A) γ ∈ A. given A = {x : x > 0}. Solution. 9 . (i) If A is an event deﬁned for an interval on z.13) and f zA (γ A) = 0 for γ ∈A. we ﬁnd f z(γ ) = 3 1 1 (u(γ − e −1 ) − u(γ − e )) + δ(γ − e 0. 8γ 8 8 f x (− ln(γ )) .
9 Random variable z = (x − 1)2 . (c ) f zz≤1 (1/4z ≤ 1). We √ √ have α1 (γ ) = γ and g (1) (α1 (γ )) = 2 γ . 48. Random variable x has a uniform PDF in the interval 0–8. 14 There is only one solution to γ = g (α) = α 2 on the interval 0 < α < 2 where f xA = 0. 0. (c )E(z). otherwise.1. 1/24. (c ) f zA (2A). Random variable x has the PDF 2 f x (α) = α(u(α) − u(α − 3)).(d ) f zA (9A). 5/27. (d )E(zz ≤ 1).2. Answers: 1.2.3. f zA (γ A) = 3 √ 1 γ+√ 28 γ (u(γ ) − u(γ − 4)). (b) f z(1/4).43. Answers: 13. Consequently.58 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS so that f xA (αA) = 3 (1 + α 2 )(u(α) − u(α − 2)). Random variable x has the PDF ⎧ ⎪ 9α 2 . 0 ≤ α < 0.5 ⎨ 2 f x (α) = 3(1 − α ). Drill Problem 6.5 ≤ α ≤ 1 ⎪ ⎩ 0. 2. Random variable z = x 3 . 9 Random variable z = 2x 2 and event A = {x : x ≥ 0}.52. 3. (c ) f zz>1/8 (1/4z > 1/8). 1/2. Drill Problem 6.Use the PDF method to determine: (a) f z(1/27). (b) f z(6). (d ) f z(2). Drill Problem 6. Drill Problem 6. Answers: 4/9.2. 0.Use the PDF method to determine: (a) f z(0).4. 1/3. Random variable 2 z = 3x + 1 .2. (b) f z(9/4). . Use the PDF method to determine: (a) f z(1/4). 1/8. 8/9. Answers: 0. Random variable x has the PDF 2 f x (α) = (α + 1)(u(α + 1) − u(α − 2)). (b) f xA (1A). 0. Determine: (a) P (A). (d )σz .2. 1.
0 < β < 2 otherwise. y) created from jointly distributed random variables x and y.17) This result cannot be continued further until a speciﬁc Fx. Solution. .y and g (x. in the problem solution. y) : g (x. In this section. With z = g (x. Let us consider several examples to illustrate the mechanics of the CDF technique and also to provide further insight. y) ≤ γ ) = P ((x. our efforts are concentrated on evaluating Fz through integrals. The ease in solution for Fz is dependent on transforming A(γ ) into proper limits of integration.16) dFx. Because we are dealing with regions in a plane instead of intervals on a line. β) where α ≤ γ − β: Fz(γ ) = ∞ −∞ γ −β −∞ 1/4. 0.y (the region where f x. y) ≤ γ }. Sketching the support region for f x. (6.y (α.3. Pay careful attention to the limits of integration to determine the range of integration in which the integrand is zero because f x. Remember that in the case of a single random variable.TRANSFORMATIONS OF RANDOM VARIABLES 59 6. Random variables x and y have joint PDF f x. Example 6. y).y is not constant) and the region A(γ ) is often most helpful. these problems are not as straightforward and tractable as before. f x. we have Fz(γ ) = P (z ≤ γ ) = P (g (x.1. β) : α + β ≤ γ }. The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6.y (α.y = 0. even crucial. We have A(γ ) = {(α.y (α. y) are considered. β) = Find the CDF for z = x + y.y = 0. β)d α dβ. Here.15) (6. y) ∈ A(γ )). the probability distribution of z = g (x.y (α. 0 < α < 2. with the ease of solution critically dependent on g (x. our efforts primarily dealt with algebraic manipulations. or Fx. β). y).3 ONE FUNCTION OF TWO RANDOM VARIABLES Consider a random variable z = g (x. y) is computed using a CDF technique similar to the one at the start of this chapter. where A(γ ) = {(x. We require the volume under the surface f x.
Random variables x and y have joint PDF f x. We have A(γ ) = {(α. 0.3. β)d αdβ. For γ < −1 we have Fz(γ ) = 0 and f z(γ ) = 0.3: Plots for Example 6. f x. β) : α − β ≤ γ }.3(b). 0 < α < 1. β) = Find the PDF for z = x − y.1.y (α.y (α. β) where α ≤ γ + β: Fz(γ ) = ∞ −∞ γ +β −∞ 2 γ −2 2 γ −β 1 1 d αdβ = 1 − (4 − γ )2 . Solution. for −1 ≤ γ < 0. 4 8 For 2 ≤ γ < 4.3(a) we obtain Fz(γ ) = 0 2 0 γ −β 1 1 d αdβ = γ 2 . 0 < β < 1 otherwise.3.60 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS γ β γ 2 2 β γ −2 2 (a)0 < γ < 2 0 γ α 0 γ −2 2 (b)2 < γ < 4 γ α FIGURE 6. Fz(γ ) = 1. With the aid of Figure 6. For 0 ≤ γ < 2. Fz(γ ) = 1 −γ 0 γ +β 1 d αdβ = (1 + γ )2 . We require the volume under the surface f x. 4 8 1. referring to Figure 6.y (α. with the aid of Figure 6. Example 6. for 4 ≤ γ .4(a). we consider the complementary region (to save some work): Fz(γ ) = 1 − Finally. 2 . For γ < 0 we have Fz(γ ) = 0.2.
TRANSFORMATIONS OF RANDOM VARIABLES 61 β 1 1 1− γ 1+ γ β −γ 0 1 (a) −1 < γ < 0 α 0 1 (b) 0 < γ < 1 γ α FIGURE 6. Inside the support region for f x.4(b) (to save some work): Fz(γ ) = 1 − 0 1−γ 1 d αdβ = 1 − (1 − γ )2 .5 it is easiest to integrate with respect to β ﬁrst: for 1 ≤ γ . Finally. we consider the complementary region shown in Figure 6. where x and y have the joint PDF f x.3. Example 6. so that f z(γ ) = 1 + γ . Solution.y (α.5: Integration region for Example 6. Find the CDF for z = x/y.3.2. 0. Fz(γ ) = 0 1 α α/γ 1 1 dβd α = 1 − . We have A(γ ) = {(α.y . 2 γ +β 1 so that f z(γ ) = 1 − γ . β) = 1/α. for γ < 1 we have Fz(γ ) = 0. so that f z(γ ) = 0. hence. for 1 ≤ γ . β) : α/β ≤ γ }.3.4: Plots for Example 6. α γ β 1 1 γ 0 1 α FIGURE 6. For 0 ≤ γ < 1. .3. 0<β<α<1 otherwise.3. we have α/β > 1. As shown in Figure 6. Fz(γ ) = 1.
we obtain Fz(γ ) = 0. γ we ﬁnd with the aid of Figure 6. β = r sin(θ). We have A(γ ) = {(α. Transforming to polar coordinates: α = r cos(θ ). Fz(γ ) = 0 2π r 2 ≤γ f x. 2 π/4 θ1 0 √ γ √ α γ −1 0 3r cos(θ)dr d θ + 2 0 1 3αdβ d α. .6: Integration regions for Example 6. Fz(γ ) = 0 π/4 0 √ γ 1 3 3r 2 cos(θ)dr d θ = √ γ 2 . Referring to Figure 6.62 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6. For γ < 0. otherwise. 2 For 1 ≤ γ < 2. Find the CDF for z = x 2 + y 2 .y (α.3. Solution. With √ γ −1 sin(θ1 ) = √ .4. the other using rectangular coordinates. Finally.4. 0 < β < α < 1 0. we split the integral into two parts: one with polar coordinates. for 0 ≤ γ < 1. where x and y have the joint PDF f x. β) = 3α.y (r cos(θ). r sin(θ))r dr d θ. β) : α 2 + β 2 ≤ γ }.6(b) that Fz(γ ) = or 3 1 3 Fz(γ ) = √ γ 2 − (γ − 1) 2 .3. β γ γ −1 1 β γ 1 0 γ 1 α 0 (b) 1 < γ < 2 1 γ α (a) 0 < γ < 1 FIGURE 6. we have Fz(γ ) = 1 for 2 ≤ γ .6(a).
β) = 0. γ ]) = {(x. 3 e −1 .20) It is often difﬁcult to perform the integration indicated in (6. 4 4 4 4 6. y) = x + y and w = h(x. we consider a bivariate CDF technique.18) (6. 6. First. y) ≤ ψ}. the conditional joint PDF is presented. and (d) f z(1). y) and w = h(x. Finally.y (α.e. 0 ≤ α ≤ 2. y) ≤ γ and w = h(x. Then Fz. y) = y. With z = g (x. y). Answers: 1 e −1 . We have g −1 ((−∞.1 Bivariate CDF Technique Let x and y be jointly distributed RVs on the probability space (S..ψ) (6. the joint PDF technique for ﬁnding the joint PDF for random variables z and w formed from x and y is described. Note that A(γ .3. ψ) to be the region of the x − y plane for which z = g (x. A(γ . . γ ]) ∩ h −1 ((−∞.y (α.20).4 BIVARIATE TRANSFORMATIONS In this section. β). y) : x + y ≤ γ } . y) from jointly distributed random variables x and y. y) ≤ γ . we ﬁnd the joint distribution of random variables z = g (x. and let z = g (x. (6. (c )Fz(1). i. ψ) = g −1 ((−∞. (b) f z(−1).1. Random variables x and y have the joint PDF f x. Then. Next. 1 − 3 e −1 .w (γ . Solution. y) ≤ ψ.4.4. 3 e −1 .19) dFx. h(x. 0.y (α. y) : g (x.w . ψ) = {(x. y) and w = h(x.25. Random variable z = x − y .1. Find (a) Fz(−1/3). the case of one function of two random variables is treated by using the joint PDF technique with an auxiliary random variable. ψ) = A(γ . 0 ≤ β ≤ 2 otherwise.TRANSFORMATIONS OF RANDOM VARIABLES 63 Drill Problem 6. Random variables x and y have joint PDF f x. Deﬁne A(γ . ψ]). Example 6. β) = e −α e −β u(α)u(β). ﬁnd the joint PDF f z. P ).
The intersection of these regions is A(γ .w (γ . ψ) = 0. β).w directly.w (γ . ψ) = {(x. 0 < γ − ψ < 2. y) : y ≤ min(γ − x. 6. y) : y ≤ ψ}. β). ψ]) = {(x.64 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS y ψ γ γ −ψ x FIGURE 6.7: Integration region for Example 6.y (α. We ﬁnd 1 ∂ Fz. ψ) = f x. ψ)}.w (γ . we ﬁnd that f z.7. β). ψ).w (γ .y (α.4. it is usually easier to ﬁnd the joint PDF than to carry out the integral indicated in (6.y (α. f z. we differentiate the above integral to obtain the PDF f z.y (γ − ψ. ψ) 1 = lim lim h 2 →0 h 1 →0 h 1 h 2 ∂ψ ∂γ ψ ψ−h 2 γ −β γ −h 1 −β ψ −∞ γ −β γ −h 1 −β d Fx.w When the RVs x and y are jointly continuous. which is illustrated in Figure.25. Instead of carrying out the above integration and then differentiating the result. we obtain f z.7 and (6. . Performing the indicated limits. 0. dFx.w (γ .20) we ﬁnd Fz.20).1. and h −1 ((−∞. With the aid of Figure 6. 0 < ψ < 2 otherwise. ψ) = ψ −∞ γ −β −∞ d Fx. ψ) = lim h 1 →0 h 1 ∂γ and ∂ 2 Fz. substituting.
21) Ii (γ .y (αi . γ ]) ∩ h −1 ((ψ − h 2 . h 2 ) = g −1 ((γ − h 1 .4.22) Let (αi .24) J (α. . .2 Bivariate PDF Technique A very important special case of bivariate transformations is when the RVs x and y are jointly continuous RVs and the mapping determined by g and h is continuous with g and h having continuous partial derivatives. The result is summarized by the theorem below. ψ. Let (αi (γ .h 2 ) d α dβ. Let x and y be jointly continuous RVs with PDF f x. βi ) Ii (γ . y) . h 1 .h 1 . ψ) ≈ i f x.4. (6. 2. βi ) = γ and h(αi . ψ).25) i = 1.w (γ . and let z = g (x. Then for small h 1 and small h 2 we have h 1 h 2 f v. βi ) denote the unique element of h 1 →0 h 2 →0 lim lim Ii (γ .26) (6. h 2 > 0 and deﬁne I (γ . h 1 . h 2 ). β) ∂β ∂h(α. ψ)). y) and w = h(x.y . Partition I into disjoint regions so that I (γ . ψ]). (6. ψ. Theorem 6. be the distinct solutions to the simultaneous equations g (αi . . h 2 ). Let h 1 > 0. . Deﬁne the Jacobian ∂g (α.1. h 2 ) = i (6.27) .ψ. (6. h 1 . the approximation becomes an equality. β) ∂β (6. h 1 . ψ.TRANSFORMATIONS OF RANDOM VARIABLES 65 6. β) ∂α ∂h(α. βi ) = ψ. β) ∂α ∂g (α. β) = . βi (γ . (6.23) Dividing both sides of the above by h 1 h 2 and letting h 1 → 0 and h 2 → 0. ψ.
0.8. J  1 0 1 = 1 · 1 − 1 · 0 = 1.y (α.y (γ − ψ. βi ) and ψ = h(αi . Then f z. β) = 12αβ(1 − α). Random variables x and y have the joint PDF f x. y) = y. 1 The region of support for f z. Find the joint PDF for z = g (x. Example 6. There is only one solution to γ = α + β and ψ = β: α1 = γ − ψ and β1 = ψ. ψ) = 1/4. βi ) for αi and βi .w (γ .w .w (γ .3. Find the joint PDF for z = g (x. 0 < β < 2 otherwise. y) = x + y and w = h(x. ψ). Example 6. Solution. and determining the support region for f z. .66 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where the indicated partial derivatives are assumed to exist. 0. y) = y. we obtain f z. J (αi . Let us consider several examples to illustrate the mechanics of this PDF technique. βi ) (6. β1 ) = f x.w (γ . 0 < ψ < 2 otherwise.y (α1 .y (α.w is illustrated in Figure 6. 0 < α < 1. y) = x 2 y and w = h(x. βi ) . 0.28) The most difﬁcult aspects to ﬁnding f z.w using the joint PDF technique are solving the simultaneous equations γ = g (αi .4. we ﬁnd f z. f x. 0 < γ − ψ < 2. 0 < α < 2. 0 < β < 1 otherwise. Random variables x and y have the joint PDF f x. ψ) = ∞ i=1 f x.2.4.y (αi . ψ) = Substituting. The Jacobian of the transformation is J = Applying Theorem 1. β) = 1/4.
y ( γ /ψ. ψ) = √ 2 γψ √ 6(1 − γ /ψ). the solution (α2 . Example 6. β2 ) is in the third quadrant of the α − β plane. Solution. − γ ψ) f z.y ( γ /ψ.y (α. β1 ) is in the ﬁrst quadrant of the α − β plane. ψ) = + .2. Random variables x and y have the joint PDF f x.TRANSFORMATIONS OF RANDOM VARIABLES 67 ψ 2 1 0 1 2 3 4 γ FIGURE 6. y) = xy and w = h(x. we ﬁnd α1 = γ /ψ and β1 = ψ. ψ) f z. Note that the solution (α1 . √ Solution. Hence β −β/α 2 β α =2 . = 0. The Jacobian is J1 = Applying Theorem 1. y) = y/x. β) = 0 for α < 0. Let√ √ √ √ ting β1 = γ ψ we have α1 = γ /ψ. we ﬁnd 2α1 β1 0 2 α1 = 2α1 β1 = 2 γ ψ.y (− γ /ψ. Solving γ = α 2 β and ψ = β.4.y (α.8: Support region for Example 6. −1 < β < 1 otherwise. The other solution α2 = −α1 is not needed since f x. γ ψ) f x. 1/α α √ √ √ √ f x. we ﬁnd α = γ /β = β/ψ. We ﬁnd J = so that J 1 = 2ψ = J 2 .25(α + 1). 0 < ψ < 1 otherwise. 2ψ 2ψ .w (γ . Then β2 = − γ ψ and α2 = − γ /ψ. so that β 2 = γ ψ.4. 1 √ f x. Find the joint PDF for z = g (x. β) = 0. 0. 0< √ γ /ψ < 1.w (γ . Solving γ = αβ and ψ = β/α.4. −1 < α < 1.
0. γ <1 ψ and 0 < γ ψ < 1. Consequently.29) It is usually advisable to let the auxiliary random variable w equal a quantity which allows a convenient solution of the Jacobian and/or the inverse equations. the support region is γ −1 < ψ < γ < 0. or γ −1 < ψ < γ < 0 otherwise.w have identical regions of support in the γ − ψ plane.w and then ﬁnding the marginal PDF f z using f z(γ ) = ∞ −∞ f z. we ﬁnd f z.y ( γ /ψ. − γ ψ) 2ψ has support region speciﬁed by −1 < − γ /ψ < 0 and or 0< γ <1 ψ and 0 < γ ψ < 1. − 1 < − γ ψ < 0.w (γ . Finally. the two components of the PDF f z. In the ﬁrst quadrant of the γ − ψ plane. This support region is illustrated in Figure 6. 0 < γ < ψ < γ −1 . Similarly. ψ) = 1/(4ψ). (6.y to random variable z = g (x. Auxiliary Random Variables The joint PDF technique can also be used to transform random variables x and y with joint PDF f x. y) to ﬁnd f z. γ ψ) 2ψ has support region speciﬁed by 0< or 0< Similarly. this support region is easily seen to be 0 < γ < ψ < γ −1 . y) by introducing auxiliary random variable w = h(x.9. in the third quadrant.y (− γ /ψ.68 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS In the γ − ψ plane. ψ)d ψ. γ /ψ < 1 and 0 < γ ψ < 1. √ √ f x.w (γ . This method is an alternative to the CDF technique presented earlier. . √ √ f x.
The only solution inside the support region for f x. y) = x 2 . Let auxiliary variable w = h(x. otherwise. y) = y. Random variables x and y have the joint PDF f x. 0 < α < 1. for 0 < γ < 1 and f z(γ ) = 0.4.y ( γ . Solution.TRANSFORMATIONS OF RANDOM VARIABLES 69 ψ 3 2 1 −3 −2 −1 1 2 3 γ −1 −2 −3 FIGURE 6. we ﬁnd α = √ √ ± γ and β = ψ.y is α = γ and β = ψ.5. 2 γ f z. We ﬁnd the marginal PDF for z as f z(γ ) = ∞ −∞ f z. Solving γ = α 2 and ψ = β.9: Support region for Example 6.y (α.4. ⎧ √ 4 γψ √ f x. 0 < γ < 1. . 1 4αβ. ψ) ⎨ √ = 2ψ. ψ)d ψ = 0 1 2ψ d ψ. 0.w (γ . β) = Find the PDF for z = g (x. Example 6.4.w (γ . ψ) = = √ ⎩ 2 γ 0. The Jacobian of the transformation is J = so that 2α 0 0 = 2α. 0 < ψ < 1 otherwise. 0 < β < 1 otherwise.
for 0 < γ < 1. β) = 4αβ. Consider the problem where random variables x and y have joint 2 8 4(γ − ψ)ψ d ψ = − γ 3 + 4γ − .70 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS ψ 1 0 1 2 γ FIGURE 6. Solving γ = α + β and ψ = β. Solution. y) = x + y. Random variables x and y have the joint PDF f x. f z = 0.4. Conditional PDF Technique Since a conditional PDF is also a PDF. y) = y. the above techniques apply to ﬁnd the conditional PDF for z = g (x. 0. 0 < ψ < 1 otherwise. Let auxiliary variable w = h(x. 0. ψ) = 4(γ − ψ)ψ. 0 < β < 1 otherwise. 1 The support region for f z. 3 3 γ −1 1 .y (γ − ψ. Referring to Figure 6. 0 < α < 1. 1 0 1 = 1.w (γ .10. f z(γ ) = Otherwise.10.6.4. given event A. Example 6. Find the PDF for z = g (x. we ﬁnd α = γ − ψ and β = ψ. f z(γ ) = 0 γ 2 4(γ − ψ)ψ d ψ = γ 3 . We ﬁnd J = so that f z.y (α. 3 For 1 < γ < 2. y). ψ) = f x. 0 < γ − ψ < 1.6.w is shown in Figure 6.10: Support region for Example 6.
30) ii.yA (α.5}. ⎧ ⎨ f x.5 0 α 1 3α dβ d α = .. Depending on whether the event A is deﬁned on the range or domain of z = g (x. β) in region A.y (α.yA to evaluate the conditional PDF f zA as follows. by the deﬁnition of a conditional PDF.wA (γ .TRANSFORMATIONS OF RANDOM VARIABLES 71 PDF f x. y). i.y (α. given A. J (αi (γ . . First. ψ)) (6. β) = 3α. β) and ψ = h(α. 2. introduce an auxiliary random variable w = h(x. We begin by ﬁnding the conditional PDF for x and y.y . 8 consequently.11(a) we ﬁnd P (A) = A f x.yA (αi (γ . we have f zA (γ A) = f z(γ ) .wA (γ . βi (γ . ψA)d ψ. If A is an event deﬁned for a region in the x − y plane. and we wish to evaluate the conditional PDF for random variable z = g (x.7. given event A = {max(x. βi ). the conditional PDF. is computed by ﬁrst evaluating f z using the technique in this section. From Figure 6. βi (γ . ψA) = ∞ i=1 f x. 0. βA) = P (A) ⎩ 0. . . and evaluate f z. are the solutions to γ = g (α.y (α. Find the PDF for z = x + y.32) Example 6. 0 < β < α < 0. ψ). f zA . i = 1. β)d α dβ = 0 0. 0<β<α<1 otherwise. (6. P (A) γ ∈ A. If A is an event deﬁned for an interval on z. Then evaluate the marginal conditional PDF f zA (γ A) = ∞ −∞ f z. y). ψ). f x. β) = 24α. Random variables x and y have joint PDF f x. we will use the conditional PDF f x.31) where (αi .5 otherwise. . y) ≤ 0. Solution. y). (6. given that event A occurred. one of the following two methods may be used to determine the conditional PDF of z using the bivariate PDF technique. ψ)A) .4. Then.
11(b). The conditional PDF for z. Letting the auxiliary random variable w = h(x. f zA (γ A) = 0.11(b).5γ γ −0.wA is illustrated in Figure 6.yA (γ − ψ. . ψA) = Substituting.5 < γ < 1 we have f zA (γ A) = 0 0.wA (γ . 0. given A. 0 < ψ < γ − ψ < 0. For 0. we ﬁnd f z.wA (γ . ψA) = 24(γ − ψ).4.5 < γ < 1 we obtain f zA (γ A) = For γ < 0 or γ > 1. β) = α + β and ψ = h(α.5 24(γ − ψ)d ψ = 3(1 − γ 2 ). The integration is easily carried out with the aid of Figure 6.7. For 0. β) = β is α = γ − ψ and β = ψ. The Jacobian of the transformation is J = 1 so that f z. f x. 0. y) = y. 1 The support region for f z. ψA)d ψ.72 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS FIGURE 6. can be found from f zA (γ A) = ∞ −∞ f z. the only solution to γ = g (α.5γ 24(γ − ψ)d ψ = 9γ 2 .wA (γ .11: Support region for Example 6.5 otherwise. ψA) .
13/12. is called the CDF technique.5 SUMMARY This chapter presented a number of different approaches to ﬁnd the probability distribution of functions of random variables. e −2 . Answers: 0. 1/12.1. Answers: 0. β) = 4αβ.2. β) = e −α−β u(α)u(β). Random variable z = x + y and w = y 2 . z = g (x). Random variables x and y have joint PDF f x. 15/16. y) : x + y ≤ 1}. 0 < α < 1. Random variable z = x 2 y.3. 1). 1/4). (b) f zA (−1/2A). 1.4. 1).4.w (−1. 0. (b) f z(1/4). 0 < β < 1 otherwise. 0 < β < 1 otherwise. (d ) f zA (1/2A). (6. Random variables x and y have joint PDF f x. we write Fz(γ ) = P (z ≤ γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )). Determine: (a) f z(−1). 0. (b) f z.w (1.TRANSFORMATIONS OF RANDOM VARIABLES 73 Drill Problem 6.4. Answers: 0. From the deﬁnition of the CDF. (c ) f z(1/2). 0. Random variables x and y have joint PDF f x. 0 < α < 1. 0 < α < 1.4. 6. Two methods are presented to ﬁnd the probability distribution of a function of one random variable.2(α 2 + β). (c )FzA (0A). (d ) f z(3/2). 3/4. (d ) f z(3/2). Find: (a) f z. Random variables x and y have joint PDF f x. (c ) f z(1/2).y (α. β) = 1.3. Answers: 0. Random variable z = x and w = xy.y (α.7172. (b) f z(−1/2). β) = 2β. Random variable z = x − y and event A = {(x. Drill Problem 6. Drill Problem 6. 0. 3/16.y (α. Determine: (a) f z.33) .y (α.4. 1. The ﬁrst. Drill Problem 6. 0 < β < 1 otherwise. and the most general approach. Determine: (a) f zA (−3/ 2A). 0.w (1.
y) from jointly distributed random variables x and y. By partitioning A(γ ) into disjoint intervals. called the PDF technique. y) and w = h(x. y) : g (x. We wish to remind the reader that the special case of a convolution integral is obtained when random variables x and y are independent and z = x + y. Typically.y (α. a bivariate CDF as well as a joint PDF technique were presented. The ﬁrst approach. i = 1. g (1) (αi (γ )) (6. i = 1.w (γ .38) The ease of solution here involves transforming A(γ ) into proper limits of integration. (6. .34) denote the distinct solutions to g (αi ) = γ . involves evaluating f z(γ ) = where αi = αi (γ ) = g −1 (γ ). y) ≤ γ ) = P ((x. The second approach. the PDF technique is much simpler to use than the CDF technique. βi ) . the PDF technique is applicable only when z = g (x) is continuous and does not equal a constant in any interval in which f x is nonzero.37) dFx. y) ∈ A(γ )). βi ) (6. .39) . involves evaluating Fz(γ ) = P (z ≤ γ ) = P (g (x. βi (γ . (6. Next. the CDF Fz may be found using the CDF Fx .35) ∞ i=1 f x (αi (γ )) . . ψ)). The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6.74 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where A(γ ) = {x : g (x) ≤ γ }. . To ﬁnd the joint probability distribution of random variables z = g (x. ψ).40) ∞ i=1 f x. y) ≤ γ }. 2. J (αi . where A(γ ) = {(x. β). The second approach involves introducing an auxiliary random variable and using the PDF technique applied to two functions of two random variables. . y) created from jointly distributed random variables x and y using two approaches. . However.36) (6. ψ) = where (αi (γ . Using the joint PDF technique.y (αi . a CDF technique. we evaluated the probability distribution of a random variable z = g (x. . (6. 2. . the joint PDF for z and w can be found as f z.
43) where the indicated partial derivatives are assumed to exist. Let random variable x be uniform between 0 and 2 with z = exp(x). x + 1. Given f x (α) = and z= Use the CDF technique to ﬁnd Fz. x<0 x > 0. the most difﬁcult aspect of the joint PDF technique is in solving the simultaneous equations. −1 < α < 1 otherwise. α<0 0≤α<1 1≤α and z = exp(−x).6 PROBLEMS 1. βi ) = ψ. The Jacobian ∂g (α.5(1 + α). 0. ∂h(α. ⎨ Fx (α) = α 2 .w . Find Fz using the CDF technique.41) (6. β) ∂β (6. 0. Suppose random variable x has the CDF ⎧ ⎪ 0. Typically.TRANSFORMATIONS OF RANDOM VARIABLES 75 are the distinct solutions to the simultaneous equations g (αi . and then integrating to obtain the marginal PDF f z. 2. βi ) = γ and h(αi . 6. Using the CDF technique. 3. ⎪ ⎩ 1. The joint PDF technique can also be used to ﬁnd the probability distribution for z = g (x. ﬁnd Fz.y by introducing an auxiliary random variable w = h(x. x − 1.42) (6. y) from f x. β) = ∂h(α. β) ∂β . β) ∂α J (α. β) ∂α ∂g (α. y) to ﬁnd f z. .
2 Find Fz using the CDF technique when z = x. Using the CDF method. so that z = x. ﬁnd: (a) E(z) using f x . (d) E(z) using f z (compare with the answer to part a). ⎨ z= x. Random variable x has the PDF f x (α) = e −α u(α). Given that x is uniformly distributed between −2 and 2. ⎧ ⎪ −1. The PDF of random variable x is 1 1 f x (α) = e − 2 (α+2) u(α + 2). (b) Fz using the CDF method. x < −1 −1 ≤ x ≤ 2 x > 2. (b) Fz using the CDF method. ﬁnd: (a) E(z) using f x .76 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 4. (b) Fz. Random variable x represents the input to a halfwave rectiﬁer and z represents the output. Determine: (a) a. 0<α<1 otherwise. 5. so that z = u(x). Random variable x represents the input to a fullwave rectiﬁer and z represents the output. 6. ⎨ z= x. (c) f z. Find Fz for z = x 2 using the CDF method. ⎪ ⎩ 1. Suppose random variable x has the PDF f x (α) = Random variable z is deﬁned by ⎧ ⎪ −1. . Given that x is uniformly distributed between −2 and 2. 0. (d) E(z) using f z (compare with the answer to part a). (c) f z. determine Fz. 7. 8. ⎪ ⎩ 2. 2α. Given f x (α) = and a 1 + α2 x ≤ −1 −1 < x < 1 x ≥ 1. 9.
Random variable x has the CDF ⎧ ⎪ 0. (c) z = (x − 1/4)2 .5 ≤ x < 3. Find the PDFs for the following random variables: (a) z = x 3 . 13. −1 < α < 2 otherwise. Let z = 1/x 2 .5 ≤ x < 2.TRANSFORMATIONS OF RANDOM VARIABLES 77 10.5 x ≥ 3. ⎪ ⎪ 3.5 0. ﬁnd: (a) Fz using the CDF technique. 11.5 −0. Random variable x has the PDF f x (α) = e −α u(α). Given that the input follows a Gaussian distribution with x ∼ G(2. Use the CDF technique to determine Fz. 0. (b) Fz using the PDF technique. Random variable x has the following PDF f x (α) = 4α 3 . ⎪ ⎪ ⎪ 1. ﬁnd f z using the CDF method. 14. ⎨ Fx (α) = (α 2 + 2α + 1)/9. . ⎪ ⎩ 1. x < 0. With z = 100 − 25x.5 ≤ x ≤ 0 0 < x. Determine Fz.5.5 2. ⎪ ⎨ z = 2.5 ≤ x < 1.5 1. 0. ⎨ z = 0. and ⎧ ⎪ x. α < −1 −1 ≤ α < 2 2≤α x < −0. Suppose f x (α) = (1 + α 2 )/6. Random variables x and z are the input and output of a quantizer. 0<α<1 otherwise. 12. ⎪ ⎩ x. ⎪ ⎪ ⎪ ⎩ 4. 0.25.49). (b) z = (x − 1/2)2 . The relationship between them is deﬁned by: ⎧ ⎪ 0.
⎧ ⎪ ⎨ α < −1 −1 ≤ α < 3 3 ≤ α. + 1K Ω + x 3K Ω z − − FIGURE 6.12: Plot for Problem 15.14 is a Gaussian random variable with mean ηx = 0 and standard deviation σx = 3. and g (x) shown in Figure 6. Assume the diode and the operational ampliﬁer are ideal. Find the PDF f z. Assume that g (x) is a second degree polynomial for x ≥ 0.25(α + 1).16. ﬁnd the PDF f z. With z = g (x) shown in Figure 6. 18.12. Assume the diodes are ideal. Find the PDF f z. The voltage x in Figure 6. The voltage x in Figure 6. Random variable x is Gaussian with mean ηx = 0 and standard deviation σx = 1. 15. .13 is a random variable which is uniformly distributed from −1 to 2. with RV z = g (x). Find the CDF Fz. 17. ⎪ ⎩ 1. 16.15 is a Gaussian random variable with mean ηx = 1 and standard deviation σx = 1. Find the PDF f z. Assume the diode is ideal. 19.78 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1 −2 −1 −1 1 2 3 x FIGURE 6. The voltage x in Figure 6. Random variable x has the CDF 0. Fx (α) = 0.13: Circuit for Problem 16.
Random variable x has the following PDF f x (α) = Find Fz if z = x 2 .14: Circuit for Problem 17.5).15: Circuit for Problem 18. Find f z using the PDF technique. 5K Ω 1K Ω x z FIGURE 6. 0. 24. 20. 0<α<1 otherwise.5δ(α − 0. 0. Random variable z = 4x + 2. 21. Random variable x is uniform in the interval 0 to 12. α + 0. Let z = cos(x). 0. Random variable x has the PDF f x (α) = 2(α + 1)/9. Find f z if z = 1/x 2 and x is uniform on −1 to 2. α < π/2 otherwise. Find the PDF of z = 2x 2 using the PDF technique.TRANSFORMATIONS OF RANDOM VARIABLES + R + 79 x 2V 3V z − − FIGURE 6. 22. Find f z if: (a) f x (α) = 1/π. 23. −1 < α < 2 otherwise. .
0. 27. 28.5 1 2 x FIGURE 6.80 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1. 0. Random variable z = (x − 1)2 and event A = {x : x ≥ 1/2}. ﬁnd f z. 26. Find (a) f w . α<0 0≤α<1 1 ≤ α. 0<α<3 otherwise. (b) f wA if A = {v ≥ 0}. given event A. ⎧ ⎪ 0. Given that x has the CDF f x (α) = 8α/π 2 .16: Transformation for Problem 19. 0 < α < π/2 otherwise. ⎪ ⎩ 1. ⎨ Fx (α) = α. x. . (b) 25. x 2. Random variable w = v2 /R represents the power dissipated in a resistor of R with v volts across the resistor. Random variable x is uniform between −1 and 1. Random variable z= Using the PDF technique.4 −2 −1 −0. Find the PDF of random variable z. Find the PDF of z = −2 ln(x) using the PDF technique. x<0 x ≥ 0. A voltage v is a Gaussian random variable with ηv = 0 and σv = 2. Random variable x has the PDF f x (α) = 2α/9.
0<α<1 1<α<2 otherwise. 34. Using the PDF technique. ⎨ f x (α) = 2 − α. −1 < α < 1 otherwise. and z = −2 ln(x). Determine: (a) f z. ⎧ ⎪ ⎨ 0 < α < 0. Determine f z using the PDF technique.TRANSFORMATIONS OF RANDOM VARIABLES −x 81 29. 0<α<1 otherwise. use the PDF technique to determine: (a) f zA (γ A) if A = {x : x > 2}. Given that random variable x has the PDF ⎧ ⎪ α. f x (α) = 3(1 − α 2 ). (d) E(z2 ). Given 9α 2 . 0. (b) f zB (γ B) if B = {z : z < 1/2}. 33. Determine f z using the PDF technique.5α(u(α) − u(α − 2)). 32. 36.75(1 − α 2 ). Random variable x has an exponential distribution with mean one. and z = 8x 3 . (b) Fz. + 1) 31. Let f x (α) = 0. Find f z if z = 1/x and f x (α) = π(α 2 1 . . ﬁnd the PDF of z = x 2 . 35. Suppose RV x has PDF f x (α) = 0. 0. ⎪ ⎩ 0. Suppose f x (α) = 2α. and z = x 2 . (c) E(z). Using the PDF technique. If z = e . Find a transformation g such that z = g (x) has PDF f z(γ ) = c γ 2 (u(γ ) − u(γ − 1)). ﬁnd f z if z = x and x is a standard Gaussian random variable. 30.5 < α < 1 otherwise. ⎪ ⎩ 0.5 0.
0 ≤ α < 2π otherwise. 40.17: Transformation for Problem 40. Find the PDF f z.5 − 0.25α. P (x = −2) = P (x = 2) = 0. Let random variable x have the PDF f x (α) = 1/12. If z = 1/x. Random variable x has the PDF ⎧ ⎪ 0. −2 < α < 1 1 < α < 2. Find f z if z = sin(x) and f x (α) = 1/(2π). 0. 38. (a) Find the transformation z = g (x) so that ⎧ ⎪ 1 − 0. with g shown in Figure 6. ⎨ f x (α) = 0.25γ .5 − 0. ⎪ ⎩ 0. ⎪ ⎩ 0. ﬁnd f z.25. −1 < α < 0 0<α<2 otherwise. in addition.17.82 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 37. 39. 3 z = g(x) 2 1 x −1 0 1 2 FIGURE 6.25γ . 0<γ <1 1<γ <2 otherwise.5. Random variable x has PDF 1 1 f x (α) = δ(α + 1) + δ(α) + (u(α) − u(α − 2)). . 1/4. ⎨ f z(γ ) = 0. Random variable x has PDF f x (α) = α2 (u(α + 1) − u(α − 2)). 4 4 Random variable z = g (x). 41. (b) Determine f z if z = (2x + 2)u(x).
The PDF for random variable x is f x (α) = e −α u(α). 44. With g denoting the equivalent conductance of a parallel connection of R1 and R2 .18. With r denoting the equivalent resistance of a series connection of R1 and R2 . determine: (a) f zA (γ A) where A = {x : x > 2}. 0<α<1 otherwise. If z = x/y. Use the PDF technique to ﬁnd: (a) f z. With z = x + y. The joint PDF of x and y is f x. 0<β<α<1 otherwise. 47. β) = 0. . (b) f zA with A = {x > 0}. (b) f zB (γ B) where B = {z : z < 1/2}. 0.25α. Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 . with g shown in Figure 6.18: Transformation for Problem 41. 42. 0<β<α otherwise.TRANSFORMATIONS OF RANDOM VARIABLES z = g( x ) 2 1 x 83 −1 0 1 2 FIGURE 6. just write the integral(s) necessary to ﬁnd Fz. ﬁnd Fz when x and y have the joint PDF f x.y (α. 0. Suppose the voltage v across a resistor is a random variable with PDF f v (α) = 6α(1 − α). With z = e −x . 45. 43. 0. Random variable z = g (x). 46. write an expression(s) for Fz(γ ) (do not solve. ﬁnd the PDF f g using convolution. Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 .y (α. β) = e −α . ﬁnd the PDF fr using convolution.
0. The joint PDF for random variables x and y is f x. 0. (b) f zA . (b) f z. 0 < β < 1 otherwise. The joint PDF for random variables x and y is f x. 49. suppose that v and r are independent random variables. . 94 < β < 106 otherwise. 0<β<α<1 otherwise. p = v2 /r . Let A = {(x. 50. 0.y (α. β) = With z = y/x. 0.w . Find f z. The joint PDF for random variables x and y is f x. Find: (a) f z. 52.y (α. β) = 3(α 2 + β 2 ). ﬁnd f z. Find the joint PDF for random variables z = x + y and w = x − y.y (α. 0 < β < 1 otherwise. 0 < β < 2 otherwise. β) = Let z = x/y and w = y.w . The joint PDF for random variables x and y is f x. 0 < α < 1. 48. 51. 0 < α 2 + β 2 < 1.y (α.w . Let z = x 2 and w = xy. y) : x > y}. Let z = 2x + y and w = x + 2y. 0 < β otherwise. β) = α + β. 53. 1/α. 0 < α < 1. 0. β) = 4αβ. 0. 0 < α. β) = 8αβ. The joint PDF for random variables x and y is f x. 0. Find the PDF for power.y (α.wA .84 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS and that resistance r is a random variable with PDF fr (β) = 1/12. z = x and w = x 2 + y 2 . 0.y (α. Find f z. Find: (a) f z. 0<β<α<1 otherwise.5βe −α . Moreover. 0 < α. The joint PDF for random variables x and y is f x.
0 < α. Find f z. Find: (a) f z.TRANSFORMATIONS OF RANDOM VARIABLES 85 54. 0 < α < 1.y (α. 0. 0 < α < 1. Let z = xy and w = y/x.y (α. 0. The joint PDF for random variables x and y is f x. 0 < β < π otherwise.y (α. Find f z.y (α. β) = α sin(β). The joint PDF for random variables x and y is f x. β) = Let z = x 2 y. 12αβ(1 − α). 55.25(α + β). 0 < β otherwise. β) = 0. 0 < β < 1 otherwise.w using the joint CDF technique. α < 1. 56. (b) f z.w . 0. β) = e −α−β . Let z = x 2 and w = cos(y).w . Let z = x + y and w = x/(x + y). 0. Find f z. 57. β < 1 otherwise. The joint PDF for random variables x and y is f x. The joint PDF for random variables x and y is f x. .
.
05 0.998617 +1 0.528220 0.000000 0.999997 0.912960 0.973388 0.999970 0.237305 0.077760 0.25 0.835210 0.998971 1.4 0.736099 0.000000 0.968750 .196874 0.999991 0.632813 0.999865 +2 0.998365 1.999540 0.2 0.500000 +3 0.998842 0.348678 0.15 k 0 0 0 0 0 0 0 0 0 0 k 0 5 0 5 0 5 +0 0.000000 +4 0.000000 (Continued ) +2 0.000000 0.977408 0.999936 1.031250 +0 0.428415 0.999924 0.969220 0.868780 0.544300 0.168070 0.989760 0.000000 0.35 0.773781 0.993280 0.5 p 0.991440 0.593127 0.981547 0.764831 0.918540 0.327680 0.1 0.598737 0.737280 0.997773 0.590490 0.820197 0.000000 0.116029 0.682560 0.994748 0.87 APPENDIX A Distribution Tables TABLE A.987205 1.999853 0.999680 0.999023 0.336960 0.187500 n = 10 +1 0.15 0.990126 1.1 0.836920 0.050328 0.45 0.945978 0.3 0.05 0.812500 +4 1.000000 0.443705 0.256218 0.942080 0.929809 1.896484 0.000000 0.913862 1.988496 1.950030 1.999990 0.997570 0.984375 0.999991 +3 0.000000 0.1: Bernoulli CDF for n = 5 and n = 10 n=5 p 0.
815939 0.028248 0.945238 0.000000 0.997750 1.677800 0.987280 1.5 k 0 5 0 5 0 5 0 5 0 5 0 5 0 5 +0 0.000000 0.3 0.000000 +2 0.973976 0.999023 TABLE A.000000 +1 0.05 k 0 5 10 0.999385 1.999922 0.046357 0.775875 0.000000 0.002533 0.829047 0.000000 .999460 0.167290 0.266038 0.944444 0.999970 0.000000 1.504405 0.996494 0.549043 0.999856 0.376953 0.898005 0.999689 1.649611 0.000000 1.244025 0.010742 0.023257 0.205891 0.000000 1.999996 1.006047 0.952651 0.2 0.25 0.000000 0.013463 0.987705 0.000000 0.623047 +1 0.149308 0.054688 0.000000 0.375810 0.921873 0.4 0.828125 +2 0.999996 0.999966 1.879126 0.999947 1.1: Bernoulli CDF for n = 5 and n = 10 (Continued) n = 10 p 0.513827 0.967206 1.999659 0.994533 1.999584 0.35 0.972608 0.999136 0.995498 0.382281 0.000000 1.88 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.107374 0.995179 0.261607 0.000000 +3 0.099560 0.999997 1.382783 0.849732 0.999999 0.989258 +4 0.963800 1.171875 0.751495 0.945313 +3 0.993631 0.000977 0.833761 0.056314 0.905066 0.998322 0.2: Bernoulli CDF for n = 15 n = 15 p 0.738437 0.999994 0.633103 0.999972 0.463291 0.1 0 5 10 +0 0.989408 0.525593 0.000000 +4 0.085954 0.45 0.980272 0.998410 0.999895 0.
013363 0.886769 0.786897 0.564282 0.995807 1.999969 .999999 0.696381 0.984757 1.995760 1.993673 0.014179 0.754842 0.990652 0.010652 0.822655 0.000000 0.996394 1.999521 0.004748 0.126828 0.035268 0.983190 0.2 0 5 10 0.000488 0.835766 0.167126 0.982422 +2 0.999512 +4 0.999908 0.003693 0.999999 0.236088 0.999215 1.999328 0.000470 0.999919 1.059235 0.999999 0.080181 0.35 0 5 10 0.999887 1.035184 0.5 0 5 10 +0 0.000127 0.999885 0.999994 0.648162 0.452160 0.938295 0.515491 0.999721 0.000000 0.999992 1.000000 0.981941 0.000000 0.260760 0.15 k 0 5 10 0.999991 0.904953 0.987557 1.042421 0.997169 0.999996 0.000031 0.303619 0.000000 0.998893 0.999943 0.398023 0.000000 0.001562 0.996347 1.461287 0.296868 0.999390 1.938949 0.849121 0.351943 0.609813 0.087354 0.943380 0.999975 0.721621 0.150879 0.000000 0.923071 0.4 0 5 10 0.45 0 5 10 0.001692 0.000000 0.957806 0.868857 0.999988 0.818240 0.996307 +3 0.998072 0.090502 0.217278 0.686486 0.2: Bernoulli CDF for n = 15 (Continued) n = 15 p 0.027114 0.851632 0.999205 1.005172 0.318586 0.604225 0.949987 0.940765 +1 0.403216 0.966167 0.120399 0.000000 0.974534 0.982700 0.653504 0.25 0 5 10 0.017578 0.DISTRIBUTION TABLES 89 TABLE A.999988 0.999879 0.061734 0.999999 0.000000 0.000000 0.000000 0.500000 0.172696 0.3 0 5 10 0.
000000 0.999996 1.2 0 5 10 15 0.069175 0.000000 1.091260 0.999961 1.994079 0.000000 0.225156 0.000000 +4 0.998671 1.999898 1.414842 0.785782 0.000000 0.804208 0.000000 0.012118 0.676927 0.990018 0.3: Bernoulli CDF for n = 20 n = 20 p 0.999065 1.608010 0.999970 1.999994 0.999671 1.999584 1.000000 0.000000 0.967857 0.997405 1.998479 1.000000 0.000000 0.000000 0.107087 0.000000 1.999950 +1 0.000000 0.999966 1.988747 0.391747 0.416625 0.15 0 5 10 15 0.999940 1.999739 1.000000 0.999752 1.000000 0.35 0 5 10 15 +0 0.999437 1.999995 1.898188 0.90 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.999999 0.002133 0.617173 0.999999 1.629648 0.000000 1.999998 1.000000 0.993985 0.000000 1.000000 1.999993 1.000000 0.175558 0.000181 0.000000 0.946833 0.237508 0.601027 0.735840 0.000000 0.024313 0.878219 0.924516 0.000000 1.05 k 0 5 10 15 0.000000 0.999999 1.984098 1.999994 +2 0.000000 1.404896 0.994862 0.959075 0.999985 1.978065 0.000000 0.411449 0.000000 0.000000 0.038760 0.996058 1.035483 0.000000 1.000798 0.000000 0.999689 1.997614 1.3 0 5 10 15 0.000000 .956825 0.998721 1.999997 1.044376 0.000000 0.829847 0.762378 0.011529 0.000000 0.1 0 5 10 15 0.358486 0.118197 0.000000 1.999816 1.121577 0.416371 0.886669 0.25 0 5 10 15 0.999957 1.007637 0.999999 +3 0.980421 0.913307 0.000000 0.003171 0.000000 0.000000 0.000000 0.952038 0.982855 0.997426 1.986136 0.867047 0.000000 1.647725 0.932692 0.000000 1.245396 0.772272 0.000000 0.206085 0.000000 1.000000 1.000000 1.
909796 0.999991 +2 0.999845 0.985612 1.750711 0.999910 +1 0.755337 0.999605 1.999999 0.740818 0. 1.978971 0.45 0 5 10 15 0.000006 0.125599 0.999939 0.000037 0.595599 0.999997 0.252006 0.5 0 5 10 15 +0 0.999683 0.005909 0.000020 0.3 0.976885 1.995321 0.979305 0.131588 0.1 0.000000 0.2 0.999723 0.999224 0.000000 0.000201 0.415893 0.999980 +4 0.000000 0.998248 1.878099 0. .748278 0.4 k 0 5 10 15 0.4 0.999953 0.001288 0.942341 0.000000 0.2.000000 0.000000 0.998388 1.965858 0.943474 0.818731 0.999799 +3 0.004933 0.000000 +4 1.055334 0.941966 0.999984 0.999999 TABLE A. .015961 0.000001 0.999996 0.904837 0.994091 +1 0. .869235 0.000524 0.251722 0.591361 0.994247 1.999734 0.411901 0. 2.999995 0.050952 0. 4.996400 0.5.963064 0.999828 1.018863 0.6 0.999214 1.993534 1.DISTRIBUTION TABLES 91 TABLE A.000927 0.496585 0.982477 0.548812 0.868412 0.992074 0.7 k 0 0 0 0 0 5 0 5 0 5 +0 0.000111 0.978586 0.000000 0.414306 0.998712 +2 0.020695 0.844195 0.000000 0. .996642 1.998469 0.4: Poisson CDF for λ = 0.057659 0.000000 (Continued ) .670320 0.000000 0.250011 0.129934 0.000000 0.999986 0. .588099 0.999999 +3 0.872479 0.5 0.993566 1.5 λ 0.938448 0.999997 0. .1.999998 0. 0.003611 0.999964 0.999943 0. 1.3: Bernoulli CDF for n = 20 (Continued) n = 20 p 0.999961 0.998852 0. .606531 0. .
985813 0.1. 2.916082 0.959743 0.135888 0.999999 0.978637 0.988095 0.735759 0.8 0.000000 0.999708 0.998589 1.757576 0.999954 0.999748 1.998897 0.999987 0.999957 0.676676 0.999998 0.999999 0.061099 0.082085 0.999763 0.990126 0.999980 +1 0.999074 0.981424 0.5 2 2.981012 0.966491 0.000000 0.999990 0.423190 0.999997 0.934358 0.000000 0.999999 +3 0.536633 0.948866 0.999980 0.9 1 1. 1.947347 0. 1. .998981 0. .998860 1.999979 0.091578 0.986541 1.913414 0.049787 0.934712 0.997656 1.995753 0.995466 0.995544 0.199148 0.999917 0.320847 0.999981 0.5 (Continued) λ 0.990920 1.018316 0.999926 1.342296 0.999972 0. 0. .449329 0.000000 0.999723 1.999995 0.999995 +2 0.725445 0.999406 0.937143 0.998903 0.000000 0.999195 0.000000 0.815263 0. .628837 0.702930 0.543813 0.785130 0.5 0 5 10 4 0 5 10 4.973261 0. 4.831051 0.557825 0.857614 0.952577 0.030197 0.999726 0.287297 0.999656 0.999998 0.5 k 0 5 0 5 0 5 0 5 0 5 0 5 10 +0 0.999830 0.011109 0.919699 0.000000 3 0 5 10 3.5.857123 0.000000 +4 0.999938 0. .996685 0.92 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.406006 0.238103 0.999924 0.433470 0.999996 0.173578 0. .808847 0.999711 0.999816 0.996197 0.991868 0.957979 0.891178 0. .996340 1.647232 0.982907 0.5 0 5 10 15 .983436 0.999984 0.889326 0.532104 0.000000 0.367879 0.135335 0.997596 0.999924 0.999085 0.772482 0.406570 0.997160 0.4: Poisson CDF for λ = 0.993331 0. .999929 0.999996 0.808792 0.2.223130 0.
998315 0.615961 0.357518 0.968172 0.000000 0.866628 0.000553 0.999995 0.999825 0.999638 0.061969 0.378155 0.957334 0.743980 0.946223 0.241436 0.043036 0.986305 0.999937 0.999491 0.997044 0.445680 0.992900 0.001503 0.007295 0.999984 0.989740 0.999943 0.026564 0.916076 0.151204 0.933161 0.059145 0.999956 0.978435 0.5 0 5 10 15 6 0 5 10 15 6.995392 0.999999 0.081765 0.974749 0.598714 0.983973 0.931906 0.979908 0.5 0 5 10 15 7 0 5 10 15 7.995549 0.285057 0.004701 0.946650 0.5.004087 0.999999 +4 0. .672758 0.999961 +1 0.124652 0.989012 0.847238 0.994547 0.528919 0.999995 0.111850 0.999987 +2 0.999982 0.996372 0.029636 0.172992 0.999042 0.999996 +3 0.088376 0.5: Poisson CDF for λ = 5.999980 0. .017351 0.999570 0.040428 0.000000 (Continued ) 5.686036 0.920759 0.998041 0.201699 0.809485 0.369041 0.DISTRIBUTION TABLES 93 TABLE A.002479 0.994283 0. .997981 0.999889 1.440493 0.987189 0.000912 0.606303 0.999800 0.877384 0.223672 0.862238 0.991173 0.999949 0.999302 0. . 5.132062 0.999995 0.5 λ 5 k 0 5 10 15 +0 0.966120 0.973000 0.997593 0.776408 0.999931 0.265026 0.729091 0.999870 0.998840 0.5 0 5 10 15 20 .999210 0.830496 0.901479 0.300708 0.791573 0.661967 0.524639 0.006738 0.957379 0.999999 0.894357 0.762183 0.526524 0.020257 0.999981 0.999849 0. 8.999774 1.999401 0.998600 0.449711 0.011276 0.999697 0.
115691 0.995716 0.936203 0.457930 0.013754 0.001933 0.999944 0.5 0 5 10 15 20 TABLE A.6: Poisson CDF for λ = 9.999999 0.000786 0.5.029253 0.999465 0.074364 0.999561 0.988894 0. . 12.587408 0.999141 0. . 5. 9.000075 0. . 11.001234 0.998703 0.5: Poisson CDF for λ = 5.999975 0.268663 0.898136 0.966527 0.994680 0. .523105 0.926149 0.040263 0.645328 0.977964 0.751990 0.958534 0.130141 0. 8.864464 +4 0.5 (Continued) λ 8 k 0 5 10 15 20 +0 0.009283 0.993387 0.006232 0.000123 0.996282 0.999933 0.5.385597 0.021226 0.5 0 5 10 15 20 10 0 5 10 .999997 8.982273 0.705988 0.256178 0.088528 0.991769 0.206781 0.940008 0.948589 0.763362 0.999921 +2 0.972575 0.149597 0.313374 0.909083 0.803008 0.003019 0.042380 0.067086 0.220221 0.965819 0.014860 0.998038 0.875773 0.999991 0.332820 0.999979 0.996998 0.791556 +3 0.999789 +1 0.030109 0.997574 0.999855 0.998944 0.836430 0.323897 0.916542 9.991072 0.999825 0.94 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.583040 +1 0.455653 0.521826 0.999967 0.999350 0.999639 0.716624 0.999990 +4 0.888076 0.000335 0.010336 0.999971 +3 0.004164 0.054964 0.999996 0.999747 0.848662 0.815886 0.164949 0.696776 +2 0.002769 0. 10.191236 0.000499 0.000045 0.999989 0.982743 0.998406 0.999906 0.391824 0. 13 λ 9 k 0 5 10 15 20 +0 0.986167 0.652974 0.452961 0.099632 0.592547 0.000203 0.
DISTRIBUTION TABLES
95
TABLE A.6: Poisson CDF for λ = 9, 9.5, 10, 11, 12, 13 (Continued)
λ
k 15 20
+0 0.951260 0.998412 0.000017 0.037520 0.459889 0.907396 0.995329 0.999918 0.000006 0.020341 0.347229 0.844416 0.988402 0.999692 0.000002 0.010734 0.251682 0.763607 0.974988 0.999034
+1 0.972958 0.999300 0.000200 0.078614 0.579267 0.944076 0.997748 0.999967 0.000080 0.045822 0.461597 0.898709 0.993935 0.999867 0.000032 0.025887 0.353165 0.835493 0.985919 0.999548
+2 0.985722 0.999704 0.001211 0.143192 0.688697 0.967809 0.998958 0.999987 0.000522 0.089505 0.575965 0.937034 0.996953 0.999944 0.000223 0.054028 0.463105 0.890465 0.992378 0.999796
+3 0.992813 0.999880 0.004916 0.231985 0.781291 0.982313 0.999536 0.999995 0.002292 0.155028 0.681536 0.962584 0.998527 0.999977 0.001050 0.099758 0.573045 0.930167 0.996028 0.999911
+4 0.996546 0.999953 0.015105 0.340511 0.854044 0.990711 0.999801 0.999998 0.007600 0.242392 0.772025 0.978720 0.999314 0.999991 0.003740 0.165812 0.675132 0.957331 0.998006 0.999962
11
0 5 10 15 20 25
12
0 5 10 15 20 25
13
0 5 10 15 20 25
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18
λ 14
k 0 5 10
+0 0.000001 0.005532 0.175681
+1 0.000012 0.014228 0.260040
+2 0.000094 0.031620 0.358458
+3 0.000474 0.062055 0.464448
+4 0.001805 0.109399 0.570437 (Continued )
96
ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ
k 15 20 25 30
+0 0.669360 0.952092 0.997392 0.999940 0.000000 0.002792 0.118464 0.568090 0.917029 0.993815 0.999803 0.000000 0.001384 0.077396 0.466745 0.868168 0.986881 0.999433 0.000000 0.000675 0.049124 0.371454 0.805481 0.974755 0.998552
+1 0.755918 0.971156 0.998691 0.999974 0.000005 0.007632 0.184752 0.664123 0.946894 0.996688 0.999910 0.000002 0.004006 0.126993 0.565962 0.910773 0.992541 0.999724 0.000001 0.002062 0.084669 0.467738 0.861466 0.984826 0.999253
+2 0.827201 0.983288 0.999365 0.999989 0.000039 0.018002 0.267611 0.748859 0.967256 0.998284 0.999960 0.000016 0.010000 0.193122 0.659344 0.941759 0.995895 0.999869 0.000007 0.005433 0.135024 0.564023 0.904728 0.991166 0.999626
+3 0.882643 0.990672 0.999702 0.999996 0.000211 0.037446 0.363218 0.819472 0.980535 0.999139 0.999983 0.000093 0.021987 0.274511 0.742349 0.963314 0.997811 0.999940 0.000041 0.012596 0.200873 0.654958 0.936704 0.995016 0.999817
+4 0.923495 0.994980 0.999864 0.999998 0.000857 0.069854 0.465654 0.875219 0.988835 0.999582 0.999993 0.000400 0.043298 0.367527 0.812249 0.977685 0.998869 0.999973 0.000185 0.026125 0.280833 0.736322 0.959354 0.997273 0.999913
15
0 5 10 15 20 25 30
16
0 5 10 15 20 25 30
17
0 5 10 15 20 25 30
DISTRIBUTION TABLES
97
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ 18
k 0 5 10 15 20 25 30
+0 0.000000 0.000324 0.030366 0.286653 0.730720 0.955392 0.996669
+1 0.000000 0.001043 0.054887 0.375050 0.799124 0.971766 0.998187
+2 0.000003 0.002893 0.091669 0.468648 0.855090 0.982682 0.999040
+3 0.000018 0.007056 0.142598 0.562245 0.898890 0.989700 0.999506
+4 0.000084 0.015381 0.208077 0.650916 0.931740 0.994056 0.999752
TABLE A.8: Marcum’s Q function for α = 0, 0.01, . . . , 1.99
α 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65
+0.00 0.500000 0.480061 0.460172 0.440382 0.420740 0.401294 0.382089 0.363169 0.344578 0.326355 0.308538 0.291160 0.274253 0.257846
+0.01 0.496011 0.476078 0.456205 0.436441 0.416834 0.397432 0.378281 0.359424 0.340903 0.322758 0.305026 0.287740 0.270931 0.254627
+0.02 0.492022 0.472097 0.452242 0.432505 0.412936 0.393580 0.374484 0.355691 0.337243 0.319178 0.301532 0.284339 0.267629 0.251429
+0.03 0.488033 0.468119 0.448283 0.428576 0.409046 0.389739 0.370700 0.351973 0.333598 0.315614 0.298056 0.280957 0.264347 0.248252
+0.04 0.484046 0.464144 0.444330 0.424655 0.405165 0.385908 0.366928 0.348268 0.329969 0.312067 0.294598 0.277595 0.261086 0.245097 (Continued )
032884 0.40 1.063008 0.226627 0.068112 0.037538 0.072145 0.074934 0.100273 0. 1.102042 0.173609 0.048457 0.146859 0.241964 0.095098 0.111233 0.00 1.35 1.20 1.70 1.80 1.049471 0.077804 0.60 1.203269 0.109349 0.040059 0.042716 0.069437 0.070781 0.235762 0.171056 0.137857 0.15 1.035148 0.220650 0.064256 0.040929 0. 0.054799 0.80 0.232695 0.238852 0.024419 +0.55 1.052616 0.066807 0.028067 0.036727 0.055917 0.04 0.208970 0.131357 0.229650 0.026190 0.161087 0.30 1.166023 0.153864 0.091759 0.027429 0.85 1.025588 +0.95 1.085344 0.023852 +0.117023 0.061780 0.107488 0.086915 0.080757 0.176186 0.034379 0.211855 0.041815 0.113140 0.142310 0.059380 0.158655 0.024998 +0.192150 0.133500 0.090123 0.039204 0.163543 0.149170 0.065522 0.103835 0.99 (Continued) α 0.8: Marcum’s Q function for α = 0.181411 0.144572 0.082264 0.45 1.98 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.047460 0.088508 0.044565 0.65 1.168528 0.01.05 1.178786 0.093418 0.75 1.189430 0.75 0. .127143 0.197662 0.076359 0.045514 0.105650 0.129238 0.02 0.057053 0. .023295 .01 0.123024 0.135666 0.033625 0.125072 0.031443 0.10 1.90 0.030742 0.026803 0.096801 0.95 +0.121001 0.051551 0.030054 0.119000 0.029379 0.079270 0.058208 0.206108 0.25 1.186733 0.050503 0.038364 0.060571 0.214764 0.00 0.035930 0.70 0.043633 0.85 0.151505 0.50 1.028716 0. . .083793 0.046479 0.115070 0.032157 0.194894 0.03 0.200454 0.053699 0.217695 0.098525 0.184060 0.223627 0.073529 0.156248 0.90 1.140071 0.
006756 0.014629 0.000641 0.018763 0.DISTRIBUTION TABLES 99 TABLE A.001695 0.00 2.009387 0.001807 0.014262 0.019699 0.015003 0.012874 0.004145 0.013209 0.004799 0.022750 0.45 2.007549 0.015386 0.65 2.35 2.000845 0.003072 0. 3.80 2.002327 0. .000577 0.15 2.020182 0.019226 0.011304 0.011011 0.60 2.30 2. .50 2.000664 0.75 2.001070 0.022216 0. .001489 0.15 3. 3.007760 0.008894 0.008198 0.000904 0.00 3.000519 0.008656 0.001035 0.006037 0.007143 0.001223 0.01 0.002401 0.004940 0.001001 0.000467 +0.006569 0.005868 0.9: Marcum’s Q function for α = 3.20 2.009137 0.01.004025 0.011911 0.005085 0.003467 0.000501 0.009642 0.017429 0.003907 0.017003 0.004397 0.004661 0. .013903 0.001395 0.05 3.000762 0.02 0.000711 0.00 0.002980 0.007344 0.002803 0.004269 0.005703 0.002635 0.002118 0.016177 0.017864 0.008424 0.000816 0.25 2.021178 0.003793 0.005386 0.000450 +0.003681 0.20 3.70 2.05 2.001441 0.000483 +0.003573 0.90 2.003167 0.009903 0.018309 0.000557 0.10 3.000968 0.006387 0.002718 0.000687 0.95 3.003364 0.40 2.002052 0.000619 0.012224 0.015778 0.000419 (Continued ) .013553 0.006210 0.001589 0.99 α 2.006947 0.000736 0.012545 0.003264 0.002256 0.001107 0.005543 0.016586 0.001641 0.010170 0.007976 0.002477 0.03 0.30 +0.001988 0.010444 0.000936 0.55 2.000434 +0.002890 0.001538 0.002186 0.000598 0.004527 0.25 3.021692 0.020675 0.000538 0.85 2.000789 0.000874 0.010724 0.001306 0.005234 0.002555 0.001144 0.001750 0.10 2.001183 0.001866 0.04 0.001264 0.001926 0.011604 0.001350 0.
000070 0.04 0.000325 0.02 0.35 3.65 3.000165 0.70 3.000050 0.000072 0.000034 +0.90 3.000059 0.000260 0.000142 0.000033 .000057 0.000092 0.000104 0.000048 0. 3. .000362 0.000039 +0.000302 0.000251 0.000075 0.000062 0.000121 0.40 3. .00 0.000147 0.000044 0.000078 0.01.000185 0.000404 0.01 0.000067 0.50 3.80 3.000242 0.000041 0.000085 0.100 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.03 0.000108 0.000052 0.000153 0.000096 0.000136 0.000100 0.000179 0.000313 0.75 3. .000131 0.000193 0.95 +0.55 3.000337 0.000376 0.000088 0.000216 0.000082 0.000291 0.99 (Continued) α 3.000117 0.000159 0.9: Marcum’s Q function for α = 3.000064 0.000042 0.000270 0.000126 0.000037 +0.000112 0.000280 0.000046 0.000224 0. 3.000233 0.000350 0.000200 0.60 3.000390 0.45 3.85 3.000054 0. .000208 0.000172 0.000036 +0.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.