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
Next. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. Engineering Statistics.iv ABSTRACT This is the third in a series of short books on probability theory and random processes for biomedical engineers. Pertinent biomedical engineering examples are throughout the text. The primary subjects of the ﬁnal chapter are methods for determining the probability distribution of a function of a random variable. The audience includes students. The aim of all three books is as an introduction to probability theory. Then. as well as important approximations to the Bernoulli PMF and Gaussian CDF. Poisson and Gaussian distributions are introduced. Poisson distributions. The exponential. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. follow most sections. Many important properties of jointly Gaussian random variables are presented. The theory material is presented in a logical manner—developing special mathematical skills as needed. Random Processes. Drill problems. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. 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. Gaussian distributions Bernoulli PMF and Gaussian CDF. KEYWORDS Probability Theory. The mathematical background required of the reader is basic knowledge of differential calculus. Probability and Statistics for Biomedical Engineers. This book focuses on standard probability distributions commonly encountered in biomedical engineering. Exponential distributions. Gaussian random variables . straightforward exercises designed to reinforce concepts and develop problem solution skills.
. . 26 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Univariate PDF Technique . . . . 4 5. . 20 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . 56 6. . . . . . . .1 Bivariate CDF Technique . . 57 6. . .1 Constant Contours. . . . . . . . . . . . . . . . . . . . . . . . 46 6. . . . .5 Univariate Gaussian Distribution . . . . . . .4 Poisson Distribution .1 Interarrival Times. . . . . . . .5 Summary . . . . . . . . . . . . . . . . . . . . . . . .2 Mixed Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Uniform Distributions . . . . 36 5. . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . 1 5. . . . . . . . . . . . . . . . . . . . .3 One Function of Two Random Variables . 11 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Summary . .3 Conditional PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . 59 6. . .1 CDF Technique with Monotonic Functions . 45 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Continuous Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 5. . . . . . . . . . . . 45 6. . . . . . . . .4. . . . . . . . . . . . . 36 Transformations of Random Variables . . . .2 Exponential Distributions . . . .4 Bivariate Transformations . . . . . . . . . . . .3 Bernoulli Trials . . . . 25 5. . . . . . . . . .8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Standard Probability Distributions . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . 1 5. . . . . . . . . . . . . . . . . . . 73 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .32 5. . . . . . . . . . . . . . . . .2 CDF Technique with Arbitrary Functions . . . . . . .2 Bivariate PDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Univariate CDF Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Bivariate Gaussian Random Variables . . .4. . . . . . . .1 Marcum’s Q Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. .1 Poisson Approximation to Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6. . 75 6. . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6. . . . . .18 5. . . . . 45 6. . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . 63 6. . . . . . . . .2. . 63 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Problems . . .2 Gaussian Approximation to Bernoulli . . . . . . . . . . . . . . . . 12 5. .v Contents 5. . . . 65 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
There are only a few key concepts involved in the basic theory of probability theory. conditional expectation. 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. joint moments. Our approach is to present a uniﬁed treatment of the subject.developing special mathematical skills as needed. Later chapters simply expand upon these key ideas and extend the range of application. standard deviation. Here in Chapter 5. Next. and the characteristic function.as well as pursue these topics at a more advanced level. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. The primary subjects of Chapter 6 are methods for determining the probability distribution of a function of a random variable. A considerable effort has been made to develop the theory in a logical manner . and the joint characteristic function. junior or senior level for the application of this theory to a wide variety of problems . Many important properties of jointly distributed Gaussian random variables are presented. This short book focuses on standard probability distributions commonly encountered in biomedical engineering. We have found it best to introduce this material using simple examples such as dice and cards. as well as important approximations to the Bernoulli PMF and Gaussian CDF. The fourth chapter introduces jointly distributed random variables. In addition. Then. The second chapter introduces the topic of random variables. The goal was to prepare students at the sophomore. the probability distribution for a single random variable is determined from a function of two random variables using the CDF.vii Preface This is the third in a series of short books on probability theory and random processes for biomedical engineers. Poisson and Gaussian distributions are introduced. The mathematical background required of the reader is basic knowledge of differential calculus. The third chapter focuses on expectation. along with joint expectation. The applications and examples given reﬂect the authors’ background in teaching probability theory and random processes for many years. moments. the exponential. rather than more complex biological . These key concepts are all presented in the ﬁrst chapter. Convolution is also developed. This text is written as an introduction to probability theory. conditional moments and the conditional characteristic function are also discussed.
. At the end of each chapter is a wide selection of problems. follow most sections. we do introduce some pertinent biomedical engineering examples throughout the text. Students in other ﬁelds should also ﬁnd the approach useful. straightforward exercises designed to reinforce concepts and develop problem solution skills. Many of the examples and end of chapter problems are based on examples from the textbook by Drake [9]. We acknowledge and thank William Pruehsner for the technical illustrations. ranging from simple to difﬁcult. The answers to the drill problems follow the problem statement in random order.viii PREFACE and biomedical phenomena. However. presented in the same general order as covered in the textbook. Drill problems.
The discrete RV x has a uniform distribution over n points (n > 1) on the interval [a. the PMF or PDF is derived according to the characteristics of the experiment. special properties are pointed out along with relationships among other probability distributions. .1 UNIFORM DISTRIBUTIONS Deﬁnition 5.3) Simplifying. n 2 nn−1 2 2 (5. Each section introduces a new PMF or PDF. and characteristic function are found. α = kh + a. the mean. n − 1 otherwise.1 CHAPTER 5 Standard Probability Distributions A surprisingly small number of probability distributions describe many natural probabilistic phenomena.3. 1.1. 12 n − 1 (5. Because of the vast number of probability distributions. . This chapter presents some of these discrete and continuous probability distributions that occur often enough in a variety of problems to deserve special mention.4) . . .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 = . In some instances. 2 σx = (b − a)2 n + 1 . Following this.1) The mean and variance of a discrete uniform RV are easily computed with the aid of Lemma 2. variance. We will see that many random variables and their corresponding experiments have similar properties and can be described by the same probability distribution.1. 2 (n − 1) n−1 4 (5. k = 0. (5. we cannot possibly discuss them all here in this chapter.2) 1 n n−1 kh − k=0 b −a 2 2 = (b − a)2 n n−1 k=0 k 1 k2 − + . 5. 0. b] if x is a lattice RV with span h = (b − a)/(n − 1) and PMF p x (α) = 1/n. Additionally.
1: (a) PMF and (b) characteristic function magnitude for discrete RV with uniform distribution over 20 points on [0. Deﬁnition 5.1 illustrates onehalf period of φx (·). a ≤α≤b otherwise.2. The continuous RV x has a uniform distribution on the interval [a. 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 .5 (a) 1 α 10 30 (b) 50 t FIGURE 5. π /h].8) .2 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS px ( α ) . 1]. where the span h = 1/19.05 1 φ x (t ) 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 = . Recall from Section 3.1 illustrates the PMF and 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. (5. n 1 − e jht (5. 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 . 2(b − a) 2 (5. Thus. (5. Figure 5. 0.3 that φx (−t) = φx (t) and that φx (t) is periodic in t with period 2π /h.6) Figure 5.5) Simplifying with the aid of Euler’s identity. b] if x has PDF f x (α) = 1/(b − a).
2.8.4) is tossed once.2: (a) PDF and (b) characteristic function magnitude for continuous RV with uniform distribution on [0.1. 1]. (d) σx . 1]. Let x be a random variable equaling ten times the number tossed. Note that the characteristic function in this case is not ∗ periodic but φx (−t) = φx (t).3. 12 (5. Random variable x is uniformly distributed on the interval [−1.STANDARD PROBABILITY DISTRIBUTIONS f x (α) 1 1 3 φx (t ) 0 .5 (a) 1 10 30 (b) 50 t FIGURE 5. (d) σx .1. A pentahedral die (with faces labeled 0. (b) P(10 ≤ x ≤ 50).2 illustrates the PDF and the magnitude of the characteristic function for a continuous RV uniformly distributed on [0. Answers: 1. Answers: 20.1. (b) Fx (5). 2 (c) E(x). and 2 σx 1 = b −a b α− a b +a 2 2 dα = (b − a)2 .2.1.2. 5]. φx (t) = exp a +b j t 2 sin b−a t 2 b−a t 2 . 0. Drill Problem 5. (5.10) Figure 5. 200. 0.9) 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. Determine: (a) p x (20). (c) ηx . 3. 1/6. −(b−a)/2 Simplifying with the aid of Euler’s identity. 2. Drill Problem 5. . Deter2 mine: (a) Fx (0).
13) Figure 5.1.2 EXPONENTIAL DISTRIBUTIONS Deﬁnition 5. . α = 1.18127]. px (α) .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 .2. 1 − q e jt (5.15) φx (t ) . otherwise. (5.18 1 (5. p2 (5. 0. (5. It can be shown that a discrete exponentially distributed RV has a memoryless property: p xx> (αx > ) = p x (α − ). The results are ηx = 1 .3: (a) PMF and (b) characteristic function magnitude for discrete RV with geometric distribution [ 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. ≥ 0. where u(·) is the unit step function. 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 .4 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 5.14) Deﬁnition 5. The continuous RV x has an exponential distribution with parameter λ(λ > 0) if x has PDF f x (α) = λe −λα u(α). . .18127. 2. .2. p and 2 σx = q .3 illustrates the PMF and the characteristic function magnitude for a discrete RV with geometric distribution and parameter p = 0.2.09 0 10 (a) 20 α 0 1 (b) 2 3 t FIGURE 5.
Suppose a system contains a component that has an exponential failure rate. Determine the number of hours reliable at 99%.1 0 5 10 (a) 15 α 10 30 (b) 50 t FIGURE 5.17) A continuous exponentially distributed RV.2 1 5 φx (t ) .2. τ ≥ 0. . Reliability theory.STANDARD PROBABILITY DISTRIBUTIONS f x (α) . grew primarily out of military applications and experiences with multicomponent systems. The mean and variance of a continuous exponentially distributed RV can be obtained using the moment generating property of the characteristic function. λ 2 σx = 1 . The exponential probability distribution is also a very important probability density function in biomedical engineering applications.4: (a) PDF and (b) characteristic function magnitude for continuous RV with exponential distribution and parameter λ = 0. arising in situations involving reliability theory and queuing problems. Reliability engineers determined its reliability at 5000 hours to be 95%.18) Example 5.1. λ2 (5. (5.4 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 . satisﬁes a memoryless property: f xx>τ (αx > τ ) = f x (α − τ ). which describes the time to failure for a system or component. like its discrete counterpart. The characteristic function can be found as ∞ φx (t) = λ 0 e α( j t−λ) d α = λ . λ − jt (5.2.16) Figure 5. Queuing theory describes the waiting times between events.2.
Random variable x. 5000 Then.99) = 980 hours.2. Answer: 80.95 = P (x > 5000) = 5000 λe −λα d α = e −5000λ . ⎨ g (x) = 75 + 39. Answers: 0.95) = 1.02.44(x − 4/3).5. Determine the average examination grade. 5. First. (c) 99%. Thus λ= − ln(0.3. λ Drill Problem 5.1. to determine the number of hours reliable at 99%. ⎪ ⎩ 0. 2.03 × 10−5 . Drill Problem 5.3 BERNOULLI TRIALS A Bernoulli experiment consists of a number of repeated (independent) trials with only two possible events for each trial.6 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. Determine the number of years reliable at: (a) 90%.2.6. Suppose a system has an exponential failure rate in years to failure with λ = 0. 3 The examination results are given by ⎧ ⎪ 75. such as a head and a tail in a coin toss. 0 < x < 4/3 x ≥ 4/3 otherwise. (b) 95%. representing the length of time in hours to complete an examination in Introduction to Random Processes. The events for each trial can be thought of as any two events which partition the sample space. we solve for α from P (x > α) = e −λα = 0. a zero or one in a computer . the parameter λ is determined from ∞ 0. 5. has PDF 4 4 f x (α) = e − 3 α u(α).99 or α= − ln(0.2.
STANDARD PROBABILITY DISTRIBUTIONS 7 bit. Deﬁnition 5.1. n p x (k) = k ⎪ ⎩ 0. . the Bernoulli distribution is used in infectious disease problems and other applications. (5. Let us call one of the events a success.3. The Bernoulli PMF describes the probability of k successes in n trials of a Bernoulli experiment. . The Bernoulli distribution is also known as a Binomial distribution. Using the moment generating property of characteristic functions. .2. 1.09 0 10 (a) 20 α 0 1 (b) 2 t FIGURE 5. .5 illustrates the PMF and the characteristic function magnitude for a discrete RV with Bernoulli distribution. k (5. px (α) . . For biomedical engineers. The characteristic function can be found using the binomial theorem: n (5. where p = probability of success and q = 1 − p. p = 0. A discrete RV x is Bernoulli distributed if the PMF for x is ⎧ ⎪ n ⎨ p k q n−k . or an even and odd number in a die toss.5: (a) PMF and (b) characteristic function magnitude for discrete RV with Bernoulli distribution. the mean and variance of a Bernoulli RV can be shown to be ηx = np.20) Figure 5. p = 0. and n = 30.19) φx (t) = k=0 n ( pe j t )k q n−k = (q + pe j t )n . k = 0.2 and n = 30. otherwise. 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. the other a failure.21) φx (t ) .18 1 2 σx = npq .
(b) from 9 to 12 games. G(n. k.7. c) p x (11) = Fx (11) − Fx (10) = 0. 0. Now.0 − 0. p) = 1 − G(n. b) P (9 ≤ x ≤ 12) = Fx (12) − Fx (8) = 0. (5. 15. Making the change of variable m = n − yields n G(n. and p = 0. .2784 = 0. p) = =0 n p (1 − p)n− . In a 15 game season. Tables A. 0. . Using the Binomial Theorem.7421. . a closed form expression for the Bernoulli CDF is not easily obtained.22) This result is easily applied to obtain values of the Bernoulli CDF for values of p > 0.1–A.1–A. Let k ∈ {0. k = 9.5. . p) = m=0 n m n−k−1 p n−m (1 − p) − m m=0 n m p n−m (1 − p)m .7216. n − k − 1. 2 d) ηx = np = 10.22) with n = 15.15.5 and n = 5.7. 0. since n n−m = n! = m! (n − m)! n m . and 20. Solution. 10.15. k.1311 = 0. (c) exactly 2 11 games? (d) With x denoting the number of games won.4845 = 0.8 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Unlike the preceding distributions. With x a Bernoulli random variable. 1. k. we consult Table A. Example 5.2.1. . n − 1} and deﬁne k G(n. what is the probability that they win: (a) at least 10 games. .5 from Tables A. p) = m=n−k n n−m p n−m (1 − p)m .3.1. . .8732 − 0. The probability that Fargo Polytechnic Institute wins a game is 0. we ﬁnd a) P (x ≥ 10) = 1 − Fx (9) = 1. n G(n. 1 − p).7031 − 0. ﬁnd ηx and σx .3. using (5. .05.3 in the Appendix list values of the Bernoulli CDF for p = 0.2186. σx = np(1 − p) = 3. k.
. Note that P (success in th trial r − 1 successes in previous − 1 trials) = p. r + 1. The moment generating function for nr can be expressed as Mnr (λ) = Letting m = − r . . n = k ⎪ ⎩ 0. otherwise. . pr q −r (5.25) . . e λr p r ∞ (m + r − 1)(m + r − 2) · · · (m + 1)(q e λ )m . . . otherwise. (5. − 1) p. 1−x x < 1. (5. for = r.STANDARD PROBABILITY DISTRIBUTIONS 9 We now consider the number of trials needed for k successes in a sequence of Bernoulli trials. we have P (nr = ) = p(r − 1. hence. n) = ⎧(k successes in n trials) P ⎪ n ⎨ p k q n−k . .23) where p = p(1. r + 1. also known as the P´ lya and the Pascal distribution. . . Let RV nr represent the number of trials to obtain exactly r successes (r ≥ 1). Note that with o r = 1 the negative binomial PMF is the geometric PMF. Let p(k. . . k = 0. . = r. 1. 1) and q = 1 − p. The PMF for the RV nr is called the negative binomial distribution. (r − 1)! m=0 xk = 1 . we obtain Mnr (λ) = With s (x) = ∞ k=0 ∞ =r ( − 1)( − 2) · · · ( − r + 1) r p q (r − 1)! −r λ e .24) (5. Discrete RV nr thus has PMF p nr ( ) = ⎧ ⎪ ⎨ ⎪ ⎩ −1 r −1 0.26) where the parameter r is a positive integer.
q e λ < 1.27) The mean and variance for nr are found to be ηn r = r . Mnr (λ) = pe λ 1 − qeλ r . (5.6 illustrates the PMF and the characteristic function magnitude for a discrete RV with negative binomial distribution. p2 (5.29) where RV x has a discrete geometric distribution. and p = 0.18127. Figure 5. (5. φx (t ) 1 px (α) .18127. r = 3.10 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) . r = 3. and p = 0. .6: (a) PMF and (b) magnitude characteristic function for discrete RV with negative binomial distribution.03 0 20 (a) 40 α 0 1 (b) 2 t FIGURE 5.06 . p and 2 σn r = rq .28) We note that the characteristic function is simply r φnr (t) = Mnr ( j t) = φx (t).
k = 0. The Poisson PMF is treated in the following section. . Lemma 5. n→∞. .3. .3 are useless for ﬁnding values for the Bernoulli CDF. so that p(k) = lim Now. We have p(k) = lim n k p k q n−k ⎧ k −λ ⎨λ e .1–A.1 Poisson Approximation to Bernoulli When n becomes large in the Bernoulli PMF in such a way that np = λ = constant. Substituting p = and q = 1 − λ . = lim ln 1 − 1 n λ n n→∞ = −λ lim 1− λ n n = e −λ . from which the desired result follows. the Bernoulli PMF approaches another important PMF known as the Poisson PMF.STANDARD PROBABILITY DISTRIBUTIONS 11 5. The motivation for using this approximation is that when n is large.1. = k! ⎩ 0. Tables A.3. 1. i=0 n→∞ lim n−k 1 − λ n −k k−1 i=0 (n − i) = 1. While there are no prescribed rules regarding the values of n and p for this approximation. n λ k k 1 p(k) = lim n→∞ k! Note that 1− λ n n−k k−1 (n − i). . the better the approximation. the larger the value of n and the smaller the value of p.30) Proof. Satisfactory results are obtained with np < 10. 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.np=λ λ n (5. λ lim ln 1 − n→∞ n so that n→∞ n λk n→∞ k! 1− λ n n . otherwise.
2 Gaussian Approximation to Bernoulli Previously. This approximation is quite useful since the Bernoulli table lists only CDF values for n up to 20.2.3. Suppose x is a Bernoulli random variable with n = 5000 and p = 0. npq where x is a Bernoulli RV. Incidentally. We have np t φx φ y (t) = exp − j √ npq Substituting for φx (t). . if p is close to one.3.5) is also used to approximate a Bernoulli PMF under certain conditions. Our solution involves approximating the Bernoulli PMF with the Poisson PMF since n is quite large (and the Bernoulli CDF table is useless).22). The accuracy of this approximation is best when n is large. p is small and np < 10. Solution. and npq > 3.6160. Lemma 5. Then the characteristic function for y satisﬁes φ(t) = lim φ y (t) = e −t n→∞ 2 (5. which results in a value of p close to zero—see (5. Since λ = np = 5. and p is very close to zero. (5. when n is large. φ y (t) = exp − j Simplifying.5 (the Poisson CDF table is covered in Section 4) that P (x ≤ 5) = 0. .001. Find P (x ≤ 5).2. Then either the Poisson or the Gaussian approximation will yield good results. we ﬁnd from Table A.12 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 5. the Poisson PMF was used to approximate a Bernoulli PMF under certain conditions. p is close to 1/2. Notice that in some circumstances np < 10 and npq > 3. np t q q + p exp t j√ npq n . 5. we can still use this approximation by reversing our deﬁnition of success and failure in the Bernoulli experiment. The Gaussian PDF (see Section 5.32) Proof. φ y (t) = q exp − j t p qn + p exp jt q np n t √ npq . that is.3.31) /2 . Let x − np y= √ .
33) e −α −∞ 2 /2 d α = 1 − Q(γ ) (5.39) = 0.34) and Q(·) is Marcum’s Q function which is tabulated in Tables A.91774. . npq b − np b = √ .5. Hence. a − np a = √ .STANDARD PROBABILITY DISTRIBUTIONS 13 Letting β= we obtain n→∞ q . With np = 2000. α→0 α2 Applying L’Hˆ spital’s Rule twice. npq = 1200 and b = (2048 − 2000)/34.39) = 1 − Q(1.3. npq (5.8 that P (x ≤ 2048) ≈ F(1. Suppose x is a Bernoulli random variable with n = 5000 and p = 0. Find P (x ≤ 2048). we ﬁnd from Table A. α→0 2α 2 2 lim φ y (t) = exp lim ln φ y (t) = e −t n→∞ /2 Consequently. where 1 F(γ ) = √ 2π is the standard Gaussian CDF. 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 = =− . Evaluation of the above integral as well as the Gaussian PDF are treated in Section 5. p and α= 1 .39. 2 n→∞ .3. The limiting φ(t) in the above lemma is the characteristic function for a Gaussian RV with zero mean and unit variance.641 = 1. The solution involves approximating the Bernoulli CDF with the Gaussian CDF since npq = 1200 > 3. Solution.35) γ (5. Example 5.4.9 of the Appendix.8 and A. n lim ln φ y (t) = lim ln( pβ 2 e − j tα/β + pe j tβα ) . for large n and a < b P (a < x < b) = P (a < y < b ) ≈ F(b ) − F(a ).
2.5) ≈ F(−0. Prof. (c) the number of times she must ﬁre so that the probability of hitting the target at least once is greater than 1/2. Suppose x is a Bernoulli random variable with n = 20 and p = 0. or within a given region. 4.7748.3. Solution. The probability she will hit the target is only 0. a = (7. which is very close to the above approximation.1201. Rensselaer is an avid albeit inaccurate marksman. In the previous section.67. 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. (c) from two to ﬁve of the next twenty cars purchased will be white. typically a time interval. we found that the limit (as n → ∞ and constant mean np) of a Bernoulli PMF is a Poisson PMF.4088. Example 5. or the number of errors per page in this textbook.5 − 10)/ √ 5 = −0. Determine: (a) the expected number of hits scored in 15 shots. Answers: 2. For example.5 < x < 8. p x (8) ≈ 0. (b) the standard deviation for 15 shots.67) − F(−1. . With np = 10. 5. Operationally.5. Find P (x = 8). 0. A survey of residents in Fargo. hence.12) = 0.25143 − 0.4 POISSON DISTRIBUTION A Poisson PMF describes the number of successes occurring on a continuous line. Since npq = 5 > 3.12. (b) at least ﬁve of the next twenty cars purchased will be white. North Dakota revealed that 30% preferred a white automobile over all other colors. the Gaussian approximation is used to evaluate the Bernoulli √ PMF. p x (8) = 0. to compute the probability that a Bernoulli RV takes an integer value using the Gaussian approximation we must round off to the nearest integer. 0.5.3. npq = 5. a continuous distribution is used to calculate probabilities for a discrete RV. we derive the Poisson probability distribution from two fundamental assumptions about the phenomenon based on physical characteristics. a Poisson random variable might represent the number of telephone calls per hour. we have p x (8) = P (7. p x (8). Drill Problem 5.4. 1.3.1.7625.12007. In this section.14 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS When approximating the Bernoulli CDF with the Gaussian CDF.13136.5 − 10)/ 5 = −1. From the Bernoulli table.3. Drill Problem 5. and b = (8.1789. Determine the probability that: (a) exactly ﬁve of the next 20 cars purchased will be white. Answers: 0.
We use the notation p(k. h) = λh + o (h). τ ] ∩ (τ. if I1 ∩ I2 = . h→0 h lim For example. we have p(0. τ + h] = [0. τ ] ∪ (τ. The probability of more than one success occurring during an interval vanishes as the length of the interval approaches zero. we have P (A ∩ B) = P (A)P (B). τ + h] and [0.40) (5. and B = { successes in interval I2 }. g (h) = 15h 2 + 7h 3 = o (h). Thus. τ ) = P (k successes in interval [0.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. (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. h) = 1 − λh + o (h). Since [0. τ )(1 − λh + o (h)). and p(0. The longer the interval.41) This second property indicates that for a series of very small intervals. τ + h) = p(0. 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.42) (5. τ ) p(0. τ + h] = . τ ]). (5. the greater the probability of success. each with a probability of success p = λh + o (h). h) = p(0.39) (5. the Poisson process is composed of a series of Bernoulli trials.37) (5. (5. .38) As we will see. Hence p(1. with A = {k successes in interval I1 }.
h) + p(k. . τ ) = e −λτ u(τ ). or p(k.16 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Noting that p(0. τ + h) = p(k − 1. τ + h) − p(0. τ ). Applying the above properties. τ ) = −λτ 0 e λt p(k − 1. τ ) + o (h) = h h and taking the limit as h → 0. (5. For ease in subsequent development. It can be shown ([7. 0) = 0. it is readily seen that p(k. t)d t (5.45) p(k. τ )(1 − λh) + o (h). d p(0.47) The RV x = number of successes thus has a Poisson distribution with parameter λτ and PMF p x (k) = p(k. (5. τ ) + λp(k. . p x (k) = p(k. τ )λh + p(k. τ ) p(0. τ ) p(1. τ ) −λh p(0. . τ ) = λp(k − 1. .48) . τ ). τ ) + . τ ).43) (5. 1)) φx (t) = e −λ ∞ k=0 (λe j t )k = e −λ exp(λe j t ) = exp(λ(e j t − 1)). we replace the parameter λτ with λ. 8]) that τ p(0. so that o (h) p(k. dτ with p(k.44) k = 1. k! k = 0. τ + h) − p(k.46) (λτ )k e −λτ u(τ ). k! (5. dτ This differential equation has solution p(0. τ + h) = p(k − 1. 0) = 1. τ ) = −λp(0. τ ) + λp(k. τ ) = λp(k − 1. (5. h h Taking the limit as h → 0 d p(k. . . h) + o (h). The characteristic function for a Poisson RV x with parameter λ is found as (with parameter λ. The rate of the Poisson process is λ and the interval length is τ . τ ) = λe and hence that p(k. . 1. . 2.
then ψx (γ ) = e λ(γ −1) . (5. The moments of a Poisson RV are tedious to compute using techniques we have seen so far.7: (a) PMF and (b) magnitude characteristic function for Poisson distributed RV with parameter λ = 10.12 1 φx (t ) 17 . Figure 5. 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. then φx1 +x2 (t) = exp((λ1 + λ2 )(e j t − 1)). so that k−1 E i=0 (k) (x − i) = ψx (1). (5. (5. It is of interest to note that if x1 and x2 are independent Poisson RVs with parameters λ1 and λ2 .50) (x − i) .52) .51) If x is Poisson distributed with parameter λ. x1 + x2 is also a Poisson with parameter λ1 + λ2 . 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.e.STANDARD PROBABILITY DISTRIBUTIONS px (α) ..49) i. respectively.
and ﬁnd a) p x (8) = Fx (8) − Fx (7) = 0. Example 5. h) = p(r − 1. known as an interarrival time. Additionally. E(x(x − 1)) = λ2 = E(x 2 ) − λ. and (c) at least 15 problems are graded? Solution.4A.4. τ − h)λh + o (h) . we consult the Poisson CDF table with λ = 10. With x a Poisson random variable. c) P (x ≥ 15) = 1 − Fx (14) = 1 − 0. hand computation of the CDF is quite tedious. τ − h) p(1. From the Poisson CDF table. Here.3628.2202 = 0.4.5830 − 0. we will brieﬂy examine the subject of interarrival times from the basis of the Poisson PMF. however. While it is quite easy to calculate the value of the Poisson PMF for a particular number of successes. 5.9165 = 0. 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. the Poisson CDF is tabulated in Tables A.53) 2 In particular. and then decreases in value as k increases from the mean.1126. Let RV tr denote the length of the time interval from zero to the r th success.7 of the Appendix for selected values of λ ranging from 0. For example. (b) 8–10 problems are graded. E(x) = λ. Professor Rensselaer grades 10 problems per day. of a Poisson random variable is more important than the actual number of successes.2202 = 0. we note that the value of the Poisson PMF increases as the number of successes k increases from zero to the mean.3328 − 0. What is the probability that on a given day (a) 8 problems are graded. b) P (8 ≤ x ≤ 10) = Fx (10) − Fx (7) = 0. On the average. the length of time between successes. so that σx = λ2 + λ − λ2 = λ.1.18 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS hence. the subject of reliability theory is so important that entire textbooks are devoted to the topic. k−1 E i=0 (x − i) = λk . Indeed. even though an inﬁnite number of Poisson successes are theoretically possible. Therefore.1 to 18. Then p(τ − h < tr ≤ τ ) = p(r − 1. in evaluating the reliability of a medical device. (5.0835. the time to failure is far more signiﬁcant to the biomedical engineer than the fact that the device failed.1 Interarrival Times In many instances.
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 ),
We will have occasion to use the shorthand notation x ∼ G(η.08 .9. Unfortunately.67) The Gaussian PDF. illustrated with η = 75 and σ 2 = 25. Additionally.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 . is a bellshaped curve completely determined by its mean and variance. a normalization is performed to obtain a standardized Gaussian RV (with zero mean f x (α) . the Gaussian PDF curve must be symmetrical as previously described. 25. the maximum value of the Gaussian PDF. the more spread in the distribution and the smaller the maximum value of the PDF. σ 2 ) to denote that the RV has a Gaussian PDF with mean η and variance σ 2 . If. (2πσ 2 )−1/2 . σ 2 ) then (x = σ z + η) φx (t) = e j ηt e − 2 σ t . which necessitates numerical integration.12 . with σ > 0 and x = −σ z + η we ﬁnd Fx (α) = P (−σ z + η ≤ α) = 1 − Fz((η − α)/σ ). the larger the value of the variance. so that f x is as above. a closed form expression does not exist for the Gaussian CDF.66) Similarly. The PDF approaches zero asymptotically as α approaches ±∞. . η = 25.9: Gaussian probability density function for η = 75 and σ 2 = 9. in fact. 2σ 2 (5. As can be seen. Rather than attempting to tabulate the general Gaussian CDF. and the area under the curve must equal one. Naturally.04 65 75 85 α FIGURE 5. as well as with η = 75 and σ 2 = 9 in Figure 5. Note that if x ∼ G(η. 1 2 2 (5. identically shaped curves could be drawn. For any combination of the mean and variance. centered now at 25 instead of 75. the Gaussian PDF is symmetrical about the vertical axis through the expected value. occurs at α = η.
1 2 (5. The probability P (α1 < x ≤ α2 ) can be obtained as P (α1 < x ≤ α2 ) = Fx (α2 ) − Fx (α1 ).73) 0 e −t d t = 1 − erf(α) 2 (5.9 for 0 ≤ α < 4 using the approximation presented in Section 5. (5.72) e −t d t 2 (5.68) (5.STANDARD PROBABILITY DISTRIBUTIONS 23 and unit variance).5. The error and complementary error functions. A simple change of variable reveals that √ erfc(α) = 2Q(α/ 2). Note that 1 Fz(α) = √ 2π where Q(·) is Marcum’s Q function: 1 Q(α) = √ 2π ∞ α (5.5. where z ∼ G(0.69) e − 2 τ d τ = 1 − Q(α). .1.1. If x ∼ G(η. It is easy to show that Q(−α) = 1 − Q(α) = Fz(α). using the fact that Fx (α) = Fz((α − η)/σ ).8 and A.74) α are also often used to evaluate the standard normal integral. 1). deﬁned by 2 erf(α) = π and 2 erfc(α) = π ∞ α (5.71) α Marcum’s Q function is tabulated in Tables A. 1).70) −∞ e − 2 τ d τ. the RV z = (x − η)/σ is a standardized Gaussian RV: z ∼ G(0. σ 2 ).75) Example 5. 1 2 (5. Compute Fz(−1. This transformation is always applied when using standard tables for computing probabilities for Gaussian RVs.74).
01 0 30 60 α FIGURE 5.6) − Fz(0. x − 35 10 f x (α) .2.73% of the area under the curve is contained between −3.2).24 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. 65]. .9. To compute Fz(−1. The PDF is essentially zero outside the interval [η − 3σ.8.04 . we note that 99. From the transformation z = (x − η)/σ .0027). With z= we have P (37 ≤ x ≤ 51) = P (0.0.02 .74) = 0.2.2 ≤ z ≤ 1. but the probability of it occurring is really very small (2Q(3) = 0.6) = 0. η + 3σ ] = [5. Solution.36594 from Table A.72) and Table A.10: PDF for Example 5. Suppose x is a Gaussian random variable with η = 35 and σ = 10. While the value a Gaussian random variable takes on is any real number between negative inﬁnity and positive inﬁnity.03 .04093.74) = 1 − Q(−1. Sketch the PDF and then ﬁnd P (37 ≤ x ≤ 51). Example 5. Hence P (37 ≤ x ≤ 51) = Q(0.5.6) = Fz(1. the range of values random variable x takes on is then approximately η ± 3σ . This notion does not imply that random variable x cannot take on a value outside this interval.74) = Q(1.0 and 3. we ﬁnd Fz(−1. the realistic range of values is much smaller.9. From Table A.5. Indicate this probability on the sketch. using (5.2) − Q(1. The sketch of this PDF is shown in Figure 5.74).10 along with the indicated probability.
1) then Q(γ ) = 1 − Fz(γ ). (5.1 Marcum’s Q Function ∞ Marcum’s Q function.. If the RV z ∼ G(0.77) i.5 × 10−8 . 932]: Q(γ ) ≈ e − 2 γ h(t).79) The error in using this approximation is less than 7.80) 1 . 1 2 γ > 0. p.5. Since we are measuring to the nearest μF.5. where z = (x − 25)/6 ∼ G(0. A very accurate approximation to Q(γ ) is presented in [1.083) = 0.09. Note that Q(0) = 0.31938153 −0. Let the RV x denote the value of a capacitor. 2π The constants are i 1 2 3 4 5 ai 0. deﬁned by 1 Q(γ ) = √ 2π e−2α dα 1 2 (5.78) where t= and 1 h(t) = √ t(a 1 + t(a 2 + t(a 3 + t(a 4 + a 5 t)))). Solution. Assuming that capacitance follows a Gaussian distribution.2316419γ (5.5 ≤ x) = P (1. 1). A machine makes capacitors with a mean value of 25 μF and a standard deviation of 6 μF.3. (5.STANDARD PROBABILITY DISTRIBUTIONS 25 Example 5. .781477937 −1.76) γ has been extensively studied. ﬁnd the probability that the value of capacitance exceeds 31 μF if capacitance is measured to the nearest μF.083 ≤ z) = Q(1. 1 + 0.08 and 1. the probability that the measured value exceeds 31 μF is P (31.13941. and that Fz(−γ ) = Q(γ ). 5.356563782 1.330274429 (5. This result is determined by linear interpolation of the CDF between equal 1.821255978 1. Q(∞) = 0. Q(γ ) is the complement of the standard Gaussian CDF.5.e.
946 when α = 3 and 0.5. Determine: (a) P (1 < x < 3).5.1. it is desired to ﬁnd the value of γ for which Q(γ ) = q . 639. The bound improves as α increases. Answers: 0.92535. 0. Suppose grade point averages are computed to the nearest tenth.5. 0. (5. i = 0. 298] e−2α 2 < Q(α) ≤ √ π α + α2 + 4 1 2 e−2α 2 . Determine the number of freshmen you would expect to score: (a) from 2.0.967 when α = 4. Then compute ti = and γi+1 h(ti ) = 2 ln q 1/2 1 1 + 0. (b)P (x > 4).2. (d )Fx (4.83) The procedure is terminated when γi+1 ≈ γi to the desired degree of accuracy. Professor Rensselaer loves the game of golf. we discuss the joint Gaussian PDF and its characteristics by drawing on our univariate Gaussian PDF experiences. Drill Problem 5.6 BIVARIATE GAUSSIAN RANDOM VARIABLES The previous section introduced the univariate Gaussian PDF along with some general characteristics. 150 ± d . Drill Problem 5. 322. 1179. Helstrom [14] offers an iterative procedure which begins with an initial guess γ0 > 0.5π 1 2 (5. (c )P (x = 4). Students attend Fargo Polytechnic Institute for an average of four years with a standard deviation of onehalf year.32. Answer: 33.82) .721). 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 0.2316419γi (5. Let the random variable x denote the length of attendance and assume that x is Gaussian. (b) less than 2.02275.5. .6 to 3. Determine the value of d so that the range.26 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS A very useful bound for Q(α) is [1. . √ π α + α 2 + 0.7. 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 17.0 and 3. . Answers: 167. 0. 5.5. . Now. (d) greater than 3.3. .5.5. covers 95% of the shots. and signiﬁcantly expanding the scope of applications. 1. (c) between 3. p.81) The ratio of the upper bound to the lower bound is 0. Sometimes. Drill Problem 5.
and let x and y be jointly distributed Gaussian RVs. the results for z1 and z2 are applied to obtain corresponding quantities φx.e. −∞ (5.y .z2 . i. which we now proceed to do.86) . (5. σx = 0.1. x ∼ G(ηx . The following development is signiﬁcantly simpliﬁed by considering the standardized versions of x and y.84) so that z1 ∼ G(0. β) = 2π ∞ I (α. the special cases ρx. 2 2 2 t1 + 2t1 t2 ρ + t2 = (t1 + ρt2 )2 + (1 − ρ 2 )t2 . Let z1 = x − ηx σx and z2 = y − ηy . t1 + 2t1 t2 ρ + t2 . like the univariate case. so that φz1 . Finally.y = ±1. Next. Since z1 and z2 are jointly Gaussian. Completing the square. The above deﬁnition of a bivariate Gaussian RV is sufﬁcient for determining the bivariate PDF.6. σ y ). we assume that ρx. In this case we also say that the RVs x and y are jointly distributed Gaussian RVs. Then 2 w is a univariate Gaussian RV for all real constants a and b. Let the RV w = ax + by. σy (5. 1).85) From (6) we have 1 f z1 .STANDARD PROBABILITY DISTRIBUTIONS 27 Numerous applications of this joint PDF are found throughout the ﬁeld of biomedical engineering and. we ﬁrst ﬁnd the joint characteristic function for the standardized RVs z1 and z2 .z2 (t1 . and σ y = 0 are discussed.y  < 1. σx = 0. y) is a bivariate Gaussian RV if every linear combination of x and y has a univariate Gaussian distribution. the marginal PDFs for a joint Gaussian PDF are univariate Gaussian. then the conditional PDF f z2 z1 and the joint PDF f z1 . the joint Gaussian PDF is considered the most important joint distribution for biomedical engineers. σx ) and 2 y ∼ G(η y . t2 ) = E(e j t1 z1 + j t2 z2 ) = e − 2 (1−ρ 1 2 2 )t2 − 1 (t1 +ρt2 )2 2 e . t2 )e − jβt2 d t2 . and σ y = 0. Deﬁnition 5.z2 (α. In particular. Below. the RV t1 z1 + t2 z2 is univariate Gaussian: 2 2 t1 z1 + t2 z2 ∼ G 0. Also. 1) and z2 ∼ G(0.y . f yx and f x.. The bivariate RV z = (x.
From (5. β) = 1 α 2 − 2ραβ + β 2 exp − . −∞ Substituting (5. 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 .90) We now turn our attention to using the above results for z1 and z2 to obtain similar results for x and y. 2(1 − ρ 2 ) (5.z2 (α. t2 ) = e or I (α. 1 2π (1 − ρ 2 ) exp − (β − ρα)2 . t2 ) = φ(t2 ) f z1 (α). we obtain I (α. t2 ) = 2π ∞ φz1 .89) After some algebra.85) and letting τ = t1 + t2 ρ. t2 )e − j αt1 d t1 .z2 (t1 .28 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where 1 I (α. .88) (5. 2π (1 − ρ 2 )1/2 2(1 − ρ 2 ) (5.z2 (α. Substituting into (5. β) = f z2 z1 (βα) = f z1 (α) so that E(z2  z1 ) = ρz1 and 2 σz2 z1 = 1 − ρ 2 . 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.z2 (α. Thus f z1 . we ﬁnd f z1 .86) we ﬁnd 1 f z1 . 1 2 −∞ .87) (5.84) we ﬁnd that x = σx z1 + ηx and y = σ y z2 + η y .
(5. σ y t2 )e − j (α−ηx )t1 e − j (β−η y )t2 d t1 d t2 . 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. (5. τ2 = σ y t2 .96) .95) 2 2πσ y (1 − ρ 2 ) .93) and (5.y (α.87) into (5. the joint characteristic function for x and y can be found from the joint characteristic function of z1 and z2 as φx. β) = (2π)2 ∞ ∞ φz1 .94) 1 fz σx 1 α − ηx σx .93) Substituting (5.z2 (σx t1 .y (t1 . Consequently.y (α. σx σy .66).90) and (5. we obtain f x.y (α. 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 .z2 (σx t1 .y (α.92) Making the change of variables τ1 = σx t1 .93) to obtain f yx (βα) = 1 f z z σy 2 1 β − η y α − ηx σy σx .91) Using (4. β) = f yx (βα) f x (α) and f x (α) = we may apply (5. t2 ) = φz1 . the joint characteristic function φx. σ y t2 )e j ηx t1 e j η y t2 . (5. β) as 1 f x.STANDARD PROBABILITY DISTRIBUTIONS 29 so that the joint characteristic function for x and y is φx. −∞ −∞ (5. (5.y (α.94) we ﬁnd f x. 1 f z . β) = Since f x.z σx σ y 1 2 α − ηx β − η y .y can be transformed to obtain the joint PDF f x.y (t1 .
Since x and y are jointly Gaussian. 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 . . β ) . (5.99) By interchanging x with y and α with β. ηx = η y = 0. ⎛ 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 .97) (5. (5. σy (5.11. β) with σx = σ y = 1.100) 2 2 σxy = σx (1 − ρ 2 ). fx .y (α.98) ⎞ ⎠ . 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 ). and ρ = −0. t1 σx + 2t1 t2 σx.8. x − ηx σx (5.y (α.11: Bivariate Gaussian PDF f x.265 β =3 α = −3 β = −3 α =3 FIGURE 5.101) A threedimensional plot of a bivariate Gaussian PDF is shown in Figure 5.y + t2 σ y .
103) x − ηx σx y → ηy ± σy in probability1 .y = 0. A very special property of jointly Gaussian RVs is presented in the following theorem.105) when ρ = ±1. the joint characteristic function for x and y is φx. Interchanging the roles of x and y we ﬁnd that the joint PDF for x and y may also be written as f x. These results can also be obtained “directly” from the joint characteristic function for x and y.102) which is valid for all σx. then ρx.y (α.98) we ﬁnd E(yx) → η y ± σ y 2 and σ yx → 0. Proof. Then f yx (βα) = f y (β). We showed previously that if x and y are independent. This is an application of the Chebyshev Inequality. t2 ) = e − 2 (t1 σx +2t1 t2 σx.y (α. the probability that the RV deviates from its mean by more than an arbitrarily small ﬁxed amount approaches zero.97) and (5. The jointly Gaussian RVs x and y are independent iff ρx. 1) and such that v and x are independent. 2 2 Example 5. If ρ = 0 then (from (5. σx and σ y . Suppose that ρ = ρx.1. Let x and y be jointly Gaussian with zero means.1.6..104) for ρ = ±1. .y +t2 σ y ) e j t1 ηx + j t2 η y . β) = f x (α)δ β − η y ± σ y (5. and ρ = ±1. Hence.95)) f x.6.STANDARD PROBABILITY DISTRIBUTIONS 31 Consequently. β) = f y (β)δ α − ηx ± σx β − ηy σy (5. We conclude that x − ηx σx α − ηx σx f x. 1 2 2 2 2 (5.y = 0. from (5. Theorem 5. As ρ → ±1.y (α. Find constants a and b such that v = ax + by ∼ G(0.y = 0.y . (5.y (t1 . i.e. We now consider some special cases of the bivariate Gaussian PDF. RVs x and y are independent. β) = f x (α) f y (β). 1 As the variance of a RV decreases to zero. σx = σ y = 1.
1 − 2ρq + q 2 (5. We require 2 2 σv = a 2 + b 2 + 2abρx. β) is constant for α and β satisfying α 2 − 2ραβ + β 2 = c 2 . β) on the contour to the origin is given by d 2 (q ) = α 2 + β 2 = c 2 (1 + q 2 ) .z2 (α. we now investigate the shape of the joint PDF f z1 .110) . β) by ﬁnding the locus of points where the PDF is constant. By inspection of (5.90).106) where c is a positive constant.z2 (α.y x 2 1 − ρx. We assume that ρ < 1.107) The square of the distance from a point (α.32 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution.6. (5. (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. we ﬁnd that f z1 . 2 Hence a = −bρx.y and b 2 = 1/(1 − ρx. so that v= 2 is independent of x and σv = 1.y = 0.1 Constant Contours Returning to the normalized jointly Gaussian RVs z1 and z2 .y ).109) ±c 1 − 2ρq + q 2 . (5.y = 1 and E(vx) = a + bρx.108) (5.y 5. y − ρx. 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 .
the line β = α intersects the constant contour at ±c ±β = α = .118) 2 − 2ρ α − ηx σx β − ηy σy + β − ηy σy 2 = c 2. 1 β =√ 2 β − ηy α − ηx − σy σx (5. we ﬁnd that d 2 (q ) attains its extremal values at q = ±1. Thus α 2 − 2ραβ + β 2 = c 2 is transformed into β2 c2 α2 + = .117) transforms the constant contour to (5. (5.116).113) √ The above equation represents an ellipse with major axis length 2c / 1 − ρ and minor axis √ length 2c / 1 + ρ.93). β) are solutions to α − ηx σx Using the transformation 1 α =√ 2 β − ηy α − ηx + σx σy .114) (5. the constant contours for f x. Thus.112) √ √ Consider the rotated coordinates α = (α + β)/ 2 and β = (β − α)/ 2. √ 2 The rotated coordinate system is a rotation by π/4 counterclockwise. so that α +β =β √ 2 and α −β = α. In the α − β plane.y (α. (5.116) (5. (5. With α = 0 we ﬁnd that one axis is along β − ηy α − ηx =− σx σy . intersects the constant contour at ±β = −α = ±c 2(1 − ρ) . the line β = −α. From (5. 1+ρ 1−ρ 1 − ρ2 (5.111) 2(1 − ρ) Similarly.115) (5.STANDARD PROBABILITY DISTRIBUTIONS 33 Differentiating. the major and minor axes of the ellipse are along the lines β = ±α.
117). 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 . 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 . With β = 0 we ﬁnd that the other axis is along β − ηy α − ηx = σx σy with endpoints at ±c σx α=√ √ + ηx . 2 1−ρ 1−ρ . the value of the joint PDF f x. 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. α =√ 1+ρ β =√ β 1−ρ exp − c2 .34 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS with endpoints at ±c σx α=√ √ + ηx .y on this curve is 1 2πσx σ y 1 − ρ 2 A further transformation α . 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) .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 2 ) (5. Points on this ellipse in the α − β plane satisfy (5.117).
we have 1 √ √ σx 2 1 + ρ J (α. 956–958].y (α. 5. 1 − ρ 2 J (α.y (α. β) = −1 √ √ σx 2 1 − ρ so that J (α. E(xy) = 7/6 + y/6. pp. Additional techniques for treating these computations are given in [1. Substituting and transforming to polar coordinates. Drill Problem 5.y . and σxy = 0. (b) η y . Given that x and y are jointly distributed Gaussian random variables with 2 2 E(yx) = 2 + 1. . we ﬁnd f x. β) d α dβ = A 1 2πσx σ y e − 2 (α +β ) d α dβ . 1. β) = 1 σx σ y 1 − ρ 2 .5x. β) 1 2 2 where the Jacobian of the transformation is ∂α ∂α J (α.STANDARD PROBABILITY DISTRIBUTIONS 35 in the α − β plane. Answers: 0.6. Determine: (a) ηx . (d) 2 σ y .1. 1 √ √ σy 2 1 − ρ ∂α ∂β ∂β ∂β .75. 1 √ √ σy 2 1 + ρ . (c) σx . This bivariate Gaussian probability computation is one of the few which is “easily” accomplished. and (e) ρx.5. β) = ∂β ∂α Computing the indicated derivatives. 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 ) . 9. 2. we have f x.
36 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Drill Problem 5. Suppose you are playing a game with a friend in which you roll a die 10 times. p = 0.6. 5. The length of time William Smith plays a video game is given by random variable x distributed exponentially with a mean of four minutes. (h) n = 10. Poisson and Gaussian distributions.6. Drill Problem 5. 2 2 σx = 21. 5. σ y = 31. p = 0.3394.8 PROBLEMS 1. p = 0.2212 in (154).3. η y = 3. (b)P (1 < y ≤ 10  x = 0). Determine P (x ≤ 3) using the Bernoulli CDF table if: (a) n = 5. 2.7 SUMMARY This chapter introduces certain probability distributions commonly encountered in biomedical engineering.1.6. (f ) n = 20.1. Assume x is a Bernoulli random variable.3. Answers: 0.1. For the Poisson approximation use η = np. (i) n = 20. Special emphasis is placed on the exponential.6. p = 0. Answers: exp(−1). Determine: (a) the probability that William is still playing after four minutes. (b) the length of the minor axis.6. Random variables x and y are jointly Gaussian with ηx = −2. Drill Problem 5. Unfortunately. (c) n = 20. (g) n = 5. he has played at least one game for more than four minutes. (c) the length of the major axis.3.3889. Bernoulli event probabilities may be approximated by the Poisson PMF when np < 10 or by the Gaussian PDF when npq > 3.899. 2. and ρ = −0. (c )P (−1 < x < 7).1. η y = 3. Many important properties of jointly Gaussian random variables are presented.3394. and ρ = −0.5212. 2 2 σx = 21. With c 2 = 0. p = 0. 0.3. 2. 30◦ . your friend gives you a dollar and if the die comes up with an odd number you give your friend a dollar. (b) n = 10. (e) n = 10. His play during each game is independent from all other games.7. For the Gaussian approximation use η = np and σ 2 = npq . p = 0. the die is loaded . out of ﬁve games.2. p = 0. (d) n = 5. σ y = 31. If the die comes up with an even number. Random variables x and y are jointly Gaussian with ηx = −2. p = 0. Important approximations to the Bernoulli PMF and Gaussian CDF are developed. (b) the probability that. p = 0. ﬁnd: (a) the smallest angle that either the minor or major axis makes with the positive α axis in the α − β plane.1753. Find: (a) E(y  x = 0). 0. Answers: 3.
7. If there are 20. One lab class has 5 students and they will perform this experiment next week. (a) If we take a sample of 20 cars. The probability that Professor Rensselaer bowls a strike is 0. Determine: (a) how many dollars your friend can expect to win in this game. (d) the number starting is less than 12. (b) exactly 12. (g) standard deviation for the number of strikes in 20 rolls. one experiment introduces students to the transistor.6. (b) 4 baskets. 8. 4.STANDARD PROBABILITY DISTRIBUTIONS 37 so that a 1 or a 3 are three times as likely to occur as a 2. 5. (a) If he tries 15 times to hit the target. Determine the: (e) expected number of strikes in 20 rolls. The probability of a man hitting a target is 0.2. (c) the probability that fewer than 2 transistors are destroyed. Let random variable x show the possible numbers of students who destroy transistors. (b) the probability of your friend winning more than 4 dollars. 3. North Dakota. Each student is given only one transistor to use. (c) the number starting is between 11. a 5 or a 6. (d) less than 2 baskets. (b) at least 4 of the next 20 rolls are strikes.500.000 will start. On a frosty January morning in Fargo. Consider Problem 7. How many times must he throw the dart so that the probability of hitting the dart board is at least 0. 9. (d) She is to keep rolling the ball until she gets a strike.3.100 will start. (f ) variance for the number of strikes in 20 rolls. the probability that a car parked outside will start is 0.4. A dart player has found that the probability of hitting the dart board in any one throw is 0. In Junior Bioinstrumentation Lab. a 4. Determine the probability that: (a) 3 of the next 20 rolls are strikes. 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. (a) Sketch the PMF for x.150.2. ﬁnd the probability that: (a) at least 12. If he makes 10 attempts. (e) the expected number of baskets. Determine the probability it will take more than 5 rolls. 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 15.000 cars to be started. (c) from 3 to 7 of the next 20 rolls are strikes. Determine: (b) the expected number of destroyed transistors. what is the probability he will make: (a) at least 4 baskets.900 and 12. (c) from 7 to 9 baskets. The probability that a basketball player makes a basket is 0. The probability of a student destroying a transistor is 0.6? .7.
80% of the students take the course at most one time. (a) If you give him a one page letter to type. Determine: (a) q. 17. Let random variable x be Bernoulli with n = 15 and p = 0. 14. what is the probability of him making fewer than 50 mistakes in a 25 page report? 18. Suppose x is a Bernoulli random variable with η = 10 and σ 2 = 10/3. Determine the probability that the student makes: (a) at least 1 mistake. a sample of radioactive material emits 20 alpha particles per minute. he averages 5 mistakes per page. 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. On the average.5. 12. (b) n. You give him another one page letter to type. Suppose a typist makes an average of 30 mistakes per page. 11. (c) exactly 2 mistakes. a student makes 6 mistakes per test. what is the probability that he makes exactly 30 mistakes? (b) The typist decides to take typing lessons. (d) at most 9 quizzes are given during the quarter. Professor Rensselaer gives 11 quizzes per quarter in Introduction to Random Processes. the course instructor. What is the probability that more than 2 accidents will occur during any given week? 15. (b) The academic vicepresident of FPI wants to ensure that on the average. when appropriate. (b) exactly 11 quizzes are given during the quarter. (c) p. What is the probability of at least 6 alpha particles being emitted in 1 min? 19. An electronics manufacturer is evaluating its quality control program. Repeat Problem 1. the material averages 6 alpha particles emitted per minute. What proportion of 20% defective components will be shipped? 13. thinks so many students repeat the course because they enjoy it so much. (b) 10 min? (c) Many years later. 16. What is the probability of him making fewer than 5 mistakes.38 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 10. To what value should the mean be adjusted to ensure this? .) (a) Determine the probability that a student takes the course more than once. What is the probability of 10 alpha particles being emitted in: (a) 1 min.4. (Professor Rensselaer. (b) from 3 to 5 mistakes. Suppose that on the average. after the lessons. using the Poisson approximation to the Bernoulli PMF. (c) at least 10 quizzes are given during the quarter. (d) more than the expected number of mistakes. On the average. (c) With the 5 mistakes per page average. Suppose the average number of times a student takes Introduction to Random Processes is 1. A certain intersection averages 3 trafﬁc accidents per week. and. a student may take a course as many times as desired. At Fargo Polytechnic Institute (FPI). Determine E(x 2 ). Determine the probability that: (a) from 8 to 12 quizzes are given during the quarter.
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. FPI admits only 1000 freshmen per year. The caller will ask the person answering the phone if it’s his or her birthday. A perfect car is assembled with a probability of 2 × 10−5 . then that lucky person will be mailed a brand new programmable calculator. a bran cereal with milk expanding (exploding) marshmallow lizards. 95% of the spoonfuls will each have at least one lizard. 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. k! k = 0. If it is. The probability that a student will major in Bioengineering is 0. (a) Every time a carpenter pounds in a nail. . (b) more than 15 defective transistors in a sample of 1000 transistors. Otherwise.002.STANDARD PROBABILITY DISTRIBUTIONS 39 20. that person will get nothing. Determine the probability that fewer than 9 students major in Bioengineering. wants to ensure that on the average. . 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. how many extra minutes can he expect to take off in building a house with 3000 nails? 24. The manufacturer of Leaping Lizards. 21. determine: (a) E(z).01. If in building a house he pounds 1250 nails. Assuming that . Determine the probability that there are: (a) 3 defective transistors in a sample of 200 transistors. Suppose 1% of the transistors in a box are defective. the probability that he hits his thumb is 0. (c) 0 defective transistors in a sample of 20 transistors. to what value should the mean of the lizards per spoonful be set at to ensure this? 25. 23. This year.000 cars are produced in a month. If 15. 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 . 1. (b) σz. Assuming that the lizards are randomly distributed in the cereal box. on its anniversary day. what is the probability that none are perfect? 22. . 26.
A battery manufacturer has found that 8. Find: (a) P (−1 < x < 0).1 < x ≤ −0. ﬁnd the probability that: (a) the computer store mails out exactly 16 calculators. C: 70–80. (f ) P (x is an integer). Using Tables A.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. 28. (g) FxB (αB). Determine: (a) the probability it takes the runner exactly 4 min to run a mile. (b) P (x < 0). that is Gaussian distributed with a mean of 77 and a variance of 11.9 and A. 33. (b) P (73 < h ≤ 80). 27. ﬁnd: (a) the mean. using the Gaussian approximation to the Bernoulli PMF.2 min. (e) FxA (αA). Grades are assigned as follows: A: 90–100. (d) P (−1. (c) σx .15 min. Event A = {0 < x ≤ 2} and B = {−1 < x ≤ 0} ∪ {1 < x ≤ 2}.98 years. what is the probability that grades will be equally distributed? 36. Find the probability that a bulb burns between 1000 and 1300 h. Assume x is a standard Gaussian random variable. 34. (b) the computer store mails out from 20 to 40 calculators. Assuming the battery lives are Gaussian distributed. (b) P (49 < r < 54). (b) Find the probability that a particular resistor is within 2 ohms of the mean.10. 29.3 years and 2.0 to 4. The time it takes a runner to run a mile is equally likely to lie in an interval from 4. 37. 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: (a) the probability that a particular resistor is within 2 ohms of the mean.2). (c) P (x < 0. (b) the probability it takes the runner from 4.08% of their batteries last less than 2. determine: (a) P (x = 0).471). In a class of 25 students.5). 32. (d) f xA (αA). B: 80–90. (a) Write the PDF for the resistance value. (f ) f xB (αB). (e) P (−2.1 to 4. (c) Find P (49 < r < 54). (c) P (h − ηh  < 3σh ). Consider Problem 32.5% of their batteries last more than 3. A certain transistor has a current gain.583 ≤ x < 1. (b) ηx . Assume that the scores on an examination are Gaussian distributed with mean 75 and standard deviation 10. Find: (a) P (h > 74). Consider Problem 36. Repeat Problem 1. Random variable x is uniform between −2 and 3. and F: below 60. . when appropriate. 31. (b) variance. A certain type of resistor has resistance values that are Gaussian distributed with a mean of 50 ohms and a variance of 3. 35. If resistances are measured to the nearest ohm. D: 60–70. 30. h. Find the value of d so that the range 77 + d covers 95% of the current gains.
8). 44. Suppose a system contains a component whose length is normally distributed with a mean of 2. A 250 question multiple choice ﬁnal exam is given. and ρx.STANDARD PROBABILITY DISTRIBUTIONS 41 38. Determine: (a) E(yx = 0). A large box contains 10. (d) P (3 < x < 10). how many resistors have resistances that are within 10% of the average? 42. (e) ρx. (c) P (70 < x ≤ 80). (c) Find σx using direct integration. and ρx. (b) P (x ≤ 78).8 + 0. .1? 41. 1). 40. and that x and y are jointly Gaussian with ρx. Determine the probability that a student guesses the correct answers for 20–25 of 85 questions about which the student has no knowledge. Draw a sketch of the joint Gaussian contour equation showing the original and the translatedrotated sets of axes.5. 43. and σ yx = 3. determine: (a) P (x = 78). If the average resistance is 1000 ohms with a standard deviation of 200 ohms. Find: (a) P (1 < y < 7  x = 0). Consider Problem 45. (c) σx .0 and a standard deviation of 0.y (∞.y = −0. ∞).8835. Assume x and y are jointly distributed Gaussian random variables with x ∼ G(−2. 1). what is the probability that at least 2 have a length greater than 2. (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 . 4). (c) P (−1 < x < 1. (b) η y . 46. 1 < y < 7). y ∼ G(3. The RV x has PDF f x (α) = a exp − 1 (α − η)2 . Assume x and y are jointly Gaussian with x ∼ G(2.y . Draw a sketch of the joint Gaussian constant contour for x and y. The average golf score for Professor Rensselaer is 78 with a standard deviation of 3. (d) the probability that x is less than 75 if the score is measured to the nearest unit. If 5 of these components are removed from different systems. 45.2. ﬁnd E(x). Assume x ∼ G(3. 39. (c) P (0 < y < 4x = 0). 13). 47. (b) P (1 < y < 7).y . E(xy) = −1 + 0.) (b) Using direct integration. 2σ 2 (a) Find the constant a. Assuming a Gaussian distribution for random variable x describing her golf game.5y. (d) σ y . 9).y = 0.000 resistors with resistances that are Gaussian distributed.32x. Suppose x and y are jointly distributed Gaussian random variables with E(yx) = 2. Determine: (a) ηx . (b) f yx (β0).y = −5. (Hint: assume RV y is independent of x and has PDF f y (β) = 2 f x (β) and evaluate Fx. (b) Using the results of (a). y ∼ G(−2.67. y ∼ G(1. Each question has 5 possible answers and only one correct answer. (f ) σx.
5. (b) 95%. If 20 of these systems are installed. Consider Problem 49. where T is a constant. (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 . Determine: (a) E(y  x = 0). In the circuit shown in Figure 5.6. Assume x and y are jointly Gaussian with x ∼ G(−1. (b) f y  x (β  0). 4).42 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 48. (b) f z(α) if z = t − T.378. . A component with an exponential failure rate is 90% reliable at 10. if he does not pin his opponent with his trick move. For successful operation of the circuit. and ρx. William’s 1 A 3 2 B 4 FIGURE 5. 53. (c) P (0 < y < 4x = 0). 50. Determine the number of hours reliable at 95%. at least two of these components must connect A with B.0433 years.12: Circuit for Problem 54. Determine the probability that the circuit is operating successfully at 10. (b) f yx (β  0). (c) 99%. (d) P (−3 < x < 0). Determine the number of years reliable at: (a) 90%.000 h. (c) P (0 < y < 4x = 0). Consider Problem 52. (b) Using the results of (a). 56. Without exception. Determine the probability that at most three will be alive at the end of 2. The survival rate of individuals with a certain type of cancer is assumed to be exponential with η = 4 years.12. 51. 55. Draw a sketch of the joint Gaussian contour for x and y. y ∼ G(1. (d) P (3 < x < 10). 49. 54.3 years. 57.000 h. Determine: (a) E(y  x = 0). Consider Problem 47. determine the probability that 10 are operating at the end of 2. Suppose a system has an exponential failure rate in years to failure with η = 2. he loses the match on points. each of the four components operate independently of one another and have an exponential failure rate (in hours) with η = 105 . William Smith is a varsity wrestler on his high school team.y = 0. 7). Five individuals have this cancer. 52. Random variable t is exponential with η = 2. Determine: (a) P (t > η).
prints brochures describing their vacation packages as “Unique. Determine: (a) the probability that there will be at least one power failure during the coming year.1149599α)u(α). the queue is 6. Find the probability that William Smith wins: (a) at least 4 of his ﬁrst 5 matches. they are so unexpected that the length of the vacations is random. 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. 62. (c) the probability that you will have more than ﬁve people in the queue ahead of you. Each customer takes 4 min to check out. (b) the probability that there will be at least two power failures during the coming year. the emergency generator has an expected time between failures of once every 200 h. The queue for the cash register at a popular package store near Fargo Polytechnic Institute becomes quite long on Saturdays following football games. Inc. 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. What is the probability that the hospital will be without power during the next 24 h? 61. Outer Space Adventures.3 people long. (b) the probability that you will have less than four people in the queue ahead of you. following an exponential distribution. 59. 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. period 2 : f xA (αA) = 0.STANDARD PROBABILITY DISTRIBUTIONS 43 trick move also prevents him from ever getting pinned. Consider Problem 57.1149599 exp(−0. Suppose Fargo Community Hospital has a backup emergency generator to provide auxilary power during a power failure. In fact.2299197α)u(α). where A = {Smith did not pin his opponent during the previous periods}. The average time between power failures in a Coop utility is once every 1.4598394 exp(−0. What is the probability that you will be back home in time for next fall quarter (that is. Consider Problem 59 statement. period 3 : f xA (αA) = 0. Assume each period is 2 min and the match is 3 periods. Moreover. Assume that a power failure will last at least 24 h. (b) more matches than he is expected to during a 10 match season.” The vacations certainly live up to their advertisement. Unexpected and Unexplained.4 years. Determine: (a) your expected waiting time to make a purchase. On the average. 9 months later)? .2299197 exp(−0. 60. 58.
If a bulb is selected at random from the box. A certain brand of light bulbs has an average lifeexpectancy of 750 h. Sevenhundred and ﬁfty of these bulbs were put in light ﬁxtures in four rooms. the bulbs were taken from the ﬁxtures and placed in a box. HOURS 1000 750 500 1500 NUMBER OF BULBS 125 250 150 225 After the speciﬁed length of time.44 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 63. The failure rate of these light bulbs follows an exponential PDF. 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. what is the probability that it is burnt out? .
the difﬁculties associated with computing the probability distribution of z = g (x) are investigated when the restrictions on g (x) are relaxed. is a random variable on the probability space (S. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. Of course. For example. a continuous monotonic function g . a full wave rectiﬁer circuit produces an output that is the absolute value of the input. and mixed random variables. The input/output characteristics of many physical devices can be represented by a nonlinear memoryless transformation of the input. 6. Proceeding from the deﬁnition of the CDF for z. The CDF technique is applicable for all functions z = g (x). and we wish to ﬁnd the CDF for random variable z = g (x).45 CHAPTER 6 Transformations of Random Variables Functions of random variables occur frequently in many applications of probability theory. consequently. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. ) and P (z(ζ ) ∈ {−∞. P ). +∞}) = 0.1. Then. discrete. we require that the function z : S → R∗ . We will refer to this method as the CDF technique. (Recall that if g is a monotonic function then a onetoone correspondence between g (x) and x exists.1 CDF Technique with Monotonic Functions Consider the problem where the CDF Fx is known for the RV x.1 UNIVARIATE CDF TECHNIQUE This section introduces a method of computing the probability distribution of a function of a random variable using the CDF. The primary subjects of this chapter are methods for determining the probability distribution of a function of a random variable. we start the discussion of computing the probability distribution of z = g (x) with the simplest case. we have for a . To make the CDF technique easier to understand. 6.) Then. and for all types of continuous. . with z(ζ ) = g (x(ζ )). we require z to be a measurable function on the measurable space (S. The ease of use of the CDF technique depends critically on the functional form of g (x). the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. Next.
the CDF for z is given by ⎧ ⎪ 0. Additionally. (γ − 1)/2 < 0 ⎨ γ −1 Fz(γ ) = Fx = (γ − 1)/8.2) 0. The following example illustrates this technique. the only requirement on g is that z = g (x) be a random variable. the CDF of x is ⎧ ⎪ 0. 0 ≤ α < 4 ⎪ ⎩ 1. this case is conceptually no more difﬁcult than .1. ⎧ ⎪ ⎨ (6. (γ − 1)/8. Applying (6. Example 6.1) where g −1 (γ ) is the value of α for which g (α) = γ . we see that g is monotone increasing and that g −1 (γ ) = (γ − 1)/2. ⎪ ⎩ 1. more than one solution of z = g (x) can exist—resulting in a manytoone mapping from x to z. As (6. we obtain Fz(γ ) = which is also a uniform distribution.1. Since random variable x is uniformly distributed. However. In this general case. Solution. 4 ≤ (γ − 1)/2. 4 ≤ α. Fz(γ ) is no longer found by simple substitution. the relationship between x and z can take on any form. with the argument α replaced by g −1 (γ ). Letting g (x) = 2x + 1. the CDF of random variable z is written in terms of Fx (α). In fact. (6. under these conditions it is impossible to write a general expression for Fz(γ ) using the CDF technique.1. 6. Similarly.2 CDF Technique with Arbitrary Functions In general.1).1) indicates. γ <1 1≤γ <9 9 ≤ γ. if z = g (x) and g is monotone decreasing. Random variable x is uniformly distributed in the interval 0 to 4. α<0 ⎨ Fx (α) = α/4.46 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 )). Find the CDF for random variable z = 2x + 1. including discontinuities. In general. Simplifying. the function does not have to be monotonic. then Fz(γ ) = P (z = g (x) ≤ γ ) = P (x ≥ g −1 (γ )) = 1 − Fx (g −1 (γ )− ). 0 ≤ (γ − 1)/2 < 4 ⎪ 2 ⎩ 1.
6) The above probabilities are easily found from the CDF Fx . Similarly. if interval Ai (γ ) is of the form Ai (γ ) = [a i (γ ). α < −1 −1 ≤ α < 1 1 ≤ α.7) The success of this method clearly depends on our ability to partition A(γ ) into disjoint intervals. then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )). i=1 (6. . . . then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )− ). ⎪ ⎩ 1.} so that A(γ ) = ∞ (6. (6. any function g and CDF Fx is amenable to a solution for Fz(γ ). . ⎨ Fx (α) = (3α − α 3 + 2)/4. Then Fz(γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )).4) Ai (γ ). Fz(γ ) = ∞ i=1 P (x ∈ Ai (γ )). The following examples illustrate various aspects of this technique.TRANSFORMATIONS OF RANDOM VARIABLES 47 the previous case. Find the CDF for the RV z = x 2 . (6. b i (γ )]. Using this method.8) (6. Random variable x has the CDF ⎧ ⎪ 0. Since the Ai ’s are disjoint.10) (6. Let A(γ ) = {x : g (x) ≤ γ }. Note that A(γ ) = g −1 ((−∞. and involves only careful book keeping. γ ]).9) (6.5) Note that the intervals as well as the number of nonempty intervals depends on γ .1.3) (6. If interval Ai (γ ) is of the form Ai (γ ) = (a i (γ ). 2. Example 6. b i (γ )]. Partition A(γ ) into disjoint intervals {Ai (γ ) : i = 1.2.
√ √ [− γ . 3 ≤ α. γ <0 ⎨ √ √ Fz(γ ) = (3 γ − ( γ )3 )/2. Example 6. γ < −1 . −1 ≤ γ < 2 2 ≤ γ. −3 ≤ α < 3 ⎪ ⎩ 1. Find Fz(γ ). ⎪ 0≤x<1 ⎪ ⎪ ⎩ 2.1. γ <0 γ ≥ 0. Random variable z is deﬁned by ⎧ x < −2 ⎪ −1.1. ⎪ −2 ≤ x < −1 ⎨ z = g (x) = −3x − 1. ⎪ ⎪ ⎨ γ −5 −γ − 1 γ + 1 A(γ ) = −∞. Noting that Fx (α) is continuous and has the same functional form for −1 < α < 1. 0≤γ <1 ⎪ ⎩ 1 1 ≤ γ. 1 ≤ x. γ ].48 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. ⎪ 3 3 3 ⎪ ⎩ R Consequently. we obtain ⎧ ⎪ 0. ∞)). Letting g (x) = x 2 . ⎪ ⎪ ⎪ 3x + 5. Referring to Figure 6.1 we ﬁnd ⎧ ∅. α < −3 ⎨ Fx (α) = (α + 3)/6. γ +1 − Fx 3 1 −γ − 1 3 − γ < −1 −1 ≤ γ < 2 2 ≤ γ. √ √ Fz(γ ) = Fx ( γ ) − Fx ((− γ )− ). −1 ≤ x < 0 ⎪ ⎪ 3x − 1. ⎧ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎩ Fx γ −5 3 0. Solution. ∪ . Let A(γ ) = g −1 ((−∞. we ﬁnd A(γ ) = g −1 ((−∞. Fz(γ ) = + Fx . γ ]) = so that ∅. Plots for this example are given in Figure 6.3. The CDF for random variable x is ⎧ ⎪ 0. . Random variable x is uniformly distributed in the interval −3 to 3.
Moreover. As presented in the previous example.1: Plots for Example 6. discrete or mixed.3.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. Fz(γ ) = Note that Fz(γ ) has jumps at γ = −1 and at γ = 2. 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. z is a mixed random variable whenever g (α) is constant in an interval where f x (α) = 0. random variable z is mixed due to the constant value of g (x) in the intervals −3 ≤ x < −2 and 1 ≤ x < 3. Whenever random variable x is continuous and g (x) is constant over some interval or intervals. γ −5 γ + 1 −γ − 1 +3+ − 3 3 3 1 γ +2 = . .1. 6 γ < −1 −1 ≤ γ < 2 2 ≤ γ. Substituting. 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. In fact. dependent on the CDF for random variable x.
50 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6. Consequently. ⎪ ⎪ ⎩ 1. we ﬁnd ⎧ ⎪ (−∞. . α < −1 ⎨ 3 Fx (α) = (α + 3α + 4)/18. 0. −1 −1 −2 1 2 FIGURE 6. g( x ) 1 x γ ≤ −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ. −1 < α < 2 otherwise.5). γ + 1].4.1. then differentiate this result. Find the PDF of random variable z deﬁned by ⎧ ⎪ x − 1. γ ≤ −1 ⎪ ⎪ ⎨ (−∞. With the aid of Figure 6. 1 ≤ γ. x≤0 0 < x ≤ 0. 0]. The CDF for random variable x is ⎧ ⎪ 0.5 < x.2: Transformation for Example 6. Random variable x has the PDF f x (α) = (1 + α 2 )/6. ⎨ z = g (x) = 0. Figure 6. To ﬁnd f z.2 shows a plot of g (x).4.1.5]. −1 ≤ α < 2 ⎪ ⎩ 1. Solution. 0. ⎪ ⎪ ⎨ F (0). −1 ≤ γ < 0 A(γ ) = {x : g (x) ≤ γ } = ⎪ (−∞.2.5 0. we evaluate Fz ﬁrst. 0≤γ <1 ⎪ ⎪ ⎩ (−∞. ∞). ⎪ ⎩ 1. x Fz(γ ) = ⎪ Fx (0. 2 ≤ α. ⎧ ⎪ Fx (γ + 1).
continuous and discontinuous. The CDF for x is ⎧ ⎪ 0. . Additionally. Differentiating Fz.1. For mixed random variables. ⎪ ⎪ ⎪ ⎪ ((γ + 1)3 + 3γ + 7)/18. ⎪ ⎪ ⎪ 5/16. the CDF technique is applicable if random variable x is mixed or discrete. Find the CDF for random variable z = g (x) = − ln(x). 18 144 16 Example 6. ⎧ 0. Note that P (x ≤ 0) = 0.5. Example 6. Random variable x is uniformly distributed in the interval from 0 to 10.TRANSFORMATIONS OF RANDOM VARIABLES 51 Substituting. as required since g (x) = − ln(x) is not deﬁned (or at least not real–valued) for x ≤ 0. For γ > 0. ∞). ⎪ ⎪ ⎩ 1.5). Random variable x has PDF f x (α) = 0.1γ . The previous examples illustrated evaluating the probability distribution of a function of a continuous random variable using the CDF technique.1) ≤ γ . We ﬁnd Fz(γ ) = 0. ⎪ ⎩ 1. the CDF technique is used without any changes or modiﬁcations as shown in the next example.1) ln(0. so that Fz(γ ) = 1 − Fx ((e −γ )− ). γ < ln(0. α<0 0 ≤ α < 10 10 ≤ α. f z(γ ) = 13 11 3γ 2 + 6γ + 6 (u(γ + 2) − u(γ + 1)) + δ(γ ) + δ(γ − 1). This technique is applicable for all functions z = g (x). Solution.6.5δ(α − 0.1. respectively. we ﬁnd A(γ ) = {x : − ln(x) ≤ γ } = (e −γ . Find the CDF for z = g (x) = 1/x 2 . Note that Fz(γ ) has jumps at γ = 0 and γ = 1 of heights 13/144 and 11/16. 1 − e −0. ⎨ Fx (α) = α/10.5(u(α) − u(α − 1)) + 0. γ < −2 −2 ≤ γ < −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ. ⎪ ⎨ Fz(γ ) = 2/9.
(b)Fz(1). 0. 1 A(γ ) = {x : x −2 ≤ γ } = −∞. (c ) f z(15). For γ < 0. (b)Fz(1/2). 1/16. ⎪ ⎪ ⎨ 0. Drill Problem 6. ⎪ ⎪ ⎩ 1.5 √ 0. γ √ Since Fx (−1/ γ ) = 0 for all real γ . (d ) f z(1/2). ⎨ z= x. x<1 −1 ≤ x ≤ 1 x > 1. After some algebra. √ (1/ γ )− < 0 √ 0 ≤ (1/ γ )− < 0. 1≤γ <4 ⎪ ⎩ 1 − 0. Random variable x has the PDF f x (α) = 0.1. 4 ≤ γ. Random variable z = 3x + 2. Random variable z is deﬁned by ⎧ ⎪ −1. (c )Fz(3/2).2.5γ −1/2 . 1/15. 0 ≤ α < 0.5γ −1/2 . 0.1. Fz(γ ) = 1 − ⎪ 0.5 ≤ α < 1 ⎪ ⎪ ⎩ 1.∞ . 1/4. . 1 ≤ α.5α.5α(u(α) − u(α − 2)).5 + 0. The mixed random variable x has CDF ⎧ ⎪ 0.5α. − √ γ so that 1 ∪ √ . For γ > 0. we have ⎧ ⎪ 0.5 + 0.5 ≤ (1/ γ )− < 1 √ − 1 ≤ (1/ γ ) . Drill Problem 6.1.5γ . 1. 2/15. Determine: (a) Fz(−1/2). ⎪ ⎩ 1.5γ −1/2 . γ <1 −1/2 Fz(γ ) = 0. Fz(γ ) = 0. Answers: 0. ⎧ ⎪ ⎨ √ √ Fz(γ ) = Fx (−1/ γ ) + 1 − Fx ((1/ γ )− ). 1/15. α<0 ⎪ ⎪ ⎨ 0.5 Fx (α) = ⎪ 0. Determine: (a) Fz(0). (c ) f z(0).5 − 0. Answers: 0.52 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. Random variable x is uniformly distributed in the interval −1 to 4.
Determine: (a) Fz(1/8).5. 1/16. (b)Fz(1/2). Answers: 0. Random variable x has PDF f x (α) = e −α−1 u(α + 1). (b)Fz(0). (c )Fz(4). 1. (d )Fz(4).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.1. Then.1. i = 1. Let αi = αi (γ ) = g −1 (γ ). in many situations it is much simpler to use than the CDF technique. Assume g is continuous and not constant over any interval for which f x = 0 . Finally.617.5 < x ≤ 1 x > 1. Drill Problem 6.11) .TRANSFORMATIONS OF RANDOM VARIABLES 53 Drill Problem 6.5α(u(α) − u(α − 2)). however. First. Let x be a continuous RV with PDF f x (α) and let the RV z = g (x) . Now.1. Random variable x has the PDF f x (α) = 0. . 0. ⎪ ⎩ 3. a method for handling mixed random variables with the PDF technique is introduced. 0. x ≤ 0.022. Determine: (a) Fz(−1). called the PDF technique.2. 2. we will ﬁnd the PDF method most useful in extensions to multivariate functions. (6.4. .1 Continuous Random Variable Theorem 6.5 0. Second. Random variable z = 1/x 2 . 6.0519. Random variable z is deﬁned by z= ⎧ ⎪ ⎨ −1. The PDF technique. . 1/16. Answers: 1/4. . In this section. x + 0. we introduce a second method for calculating the probability distribution of a function z = g (x) using the probability density function. (c )Fz(3/2). we consider computing the conditional PDF of a function of a random variable using the PDF technique. we discuss a wide variety of situations using the PDF technique with functions of continuous random variables.3. (d ) f z(4). We introduce the PDF technique for two reasons.2. 0.089. 6. 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.
h) − a i (γ . b i (γ . h). = lim = (1) h→0 h→0 g (b i (γ . Partition I (γ . h)). Then f z(γ ) = where we interpret f x (αi (γ )) = 0. . h))) lim a i (γ . . h)) − g (a i (γ . . such that I (γ . The absolute value appears because by construction we have b i > a i and h > 0. h→0 Note that (for all γ with f x (αi (γ )) = 0) 1 b i (γ . h)) h g (αi (γ )) lim and that h→0 lim Fx (b i (γ . 2. h→0 ∞ i=1 ∞ i=1 ∞ i=1 f x (αi (γ )) . h). . . h) The desired result follows by taking the product of the above limits. h)) − Fx (a i (γ . g (1) (αi (γ )) (6. g (1) (αi (γ )) if f x (αi (γ ) = 0. h) − a i (γ . whereas g (1) may be positive or negative. Let h > 0 and deﬁne I (γ . Ii (γ . h) . h)) − Fx (a i (γ . h) = {x : γ − h < g (x) ≤ γ }. Proof. b i (γ . h) = Then Fz(γ ) − Fz(γ − h) = By hypothesis. (Fx (b i (γ . h) = αi (γ ). h) into disjoint intervals of the form Ii (γ . h) − a i (γ . h)) = f x (αi (γ )).12) i = 1. h) = lim b i (γ . h) = (a i (γ .54 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS denote the distinct solutions to g (αi ) = γ . h) b i (γ .
and f z(γ ) = Substituting. We easily ﬁnd g (1) (α) = 2. there is only one solution to the equation γ = 2α + 1.TRANSFORMATIONS OF RANDOM VARIABLES 55 Example 6. f z(γ ) = or f z(γ ) = 0. For γ < 0.75(1 − α 2 )(u(α + 1) − u(α − 1)). 6 Find the PDF for random variable z = g (x) = x 2 . In this case. Solution. g (1) (αi ) = 2/αi 3 = 2γ 3/2 . 1<γ <9 otherwise.75(γ − 2 − γ − 2 )u(γ − 1).2. . Find the PDF for random variable z = g (x) = 1/x 2 . Random variable x has PDF f x (α) = 0. 2γ 3/2 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)). Hence f z(γ ) = f x ((γ − 1)/2)/2 = Example 6.2. Find the PDF for random variable z = g (x) = 2x + 1. we have g (1) (α) = −2α −3 .75(1 − γ −1 )(u(1 − γ −1/2 ) − 0 + 1 − u(γ −1/2 − 1)) u(γ ). Solution. 2γ 3/2 0. Since γ = g (α) = α −2 .2. 2γ 3/2 0. Example 6. f z(γ ) = Simplifying. For γ > 0 there are two solutions to g (αi ) = γ : 1 α1 (γ ) = − √ . 0. γ 1/8. given by α1 = (γ − 1)/2.1.2. Random variable x has PDF 3 5 f x (−γ −1/2 ) + f x (γ −1/2 ) u(γ ). there are no solutions to g (αi ) = γ . so that f z(γ ) = 0 for γ < 0.3. hence. Random variable x is uniformly distributed in the interval 0–4. γ and 1 α2 (γ ) = √ .75(1 − γ −1 )(u(γ − 1) − 0 + 1 − u(1 − γ )) u(γ ).
√ Since γ = g (α) = α 2 .5 are mapped to probability masses of 1/8 for z at e 0. ⎨ f z(γ ) = (γ −1/2 + γ 1/2 )/12. Solution. The continuous part of the PDF of x is handled by the PDF technique. ⎪ ⎩ 0. We have g (1) (α1 ) = −e ln(γ ) = −γ . 6.2 Mixed Random Variable Consider the problem where random variable x is mixed. 12 γ ⎧ ⎪ (γ −1/2 + γ 1/2 )/6. g (1) (αi ) = 2αi  = 2 γ .2. 1+γ √ √ √ √ √ (u(1 − γ ) − u(−2 − γ ) + u( γ + 1) − u( γ − 2))u(γ ).5 and 0. Here. Random variable x has PDF 1 1 3 f x (α) = (u(α + 1) − u(α − 1)) + δ(α + 0. There is only one solution to g (α) = γ : α1 (γ ) = − ln(γ ). Example 6. we treat the discrete and continuous portions of f x separately. For all γ > 0 such that . To illustrate the use of this technique. and then combine the results to yield f z. and √ √ f x (− γ ) + f x ( γ ) u(γ ). 8 8 8 Find the PDF for the RV z = g (x) = e −x .5 and e −0. For γ > 0 there are two solutions to g (αi ) = γ : √ √ α1 (γ ) = − γ . For γ < 0 there are no solutions to g (αi ) = γ .2.5) + δ(α − 0.56 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. respectively. f z(γ ) = Simplifying. so that f z(γ ) = 0 for γ < 0. f z(γ ) = √ 2 γ Substituting. 12 γ 0<γ <1 1<γ <4 elsewhere. hence. consider the following example. f z(γ ) = or 1+γ √ (u(1 − γ ) − 0 + 1 − u(γ − 4))u(γ ). The probability masses of 1/8 for x at −0.4.5).5 . and we wish to ﬁnd the PDF for z = g (x). and α2 (γ ) = γ . we have g (1) (α) = 2α.
8γ 8 8 f x (− ln(γ )) .14) Example 6. given A = {x : x > 0}.13) and f zA (γ A) = 0 for γ ∈A.TRANSFORMATIONS OF RANDOM VARIABLES 57 α1 (γ ) ± 0. 9 .3 Conditional PDF Technique Since a conditional PDF is also a PDF. one of the following two methods may be used to determine the conditional PDF of z using the PDF technique. the above techniques apply to ﬁnd the conditional PDF for z = g (x).2. given that event A occurred. P (A) γ ∈ A.5 > 0 we have f z(γ ) = Combining these results. we have f zA (γ A) = f z(γ ) . based on f xA . given event A. 6 Find the PDF for random variable z = g (x) = x 2 . and we wish to evaluate the conditional PDF for random variable z = g (x). (ii) If A is an event deﬁned for an interval on x. Then. γ 6.5. Consider the problem where random variable x has PDF f x . the conditional PDF.2. We have P (A) = 1 6 2 0 7 (1 + α 2 )d α = . 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) . Depending on whether the event A is deﬁned on the range or domain of z = g (x). First.5 ) + δ(γ − e −0. Solution. f zA . by the deﬁnition of a conditional PDF.5 ). Random variable x has the PDF 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)). is computed by ﬁrst evaluating f z using the technique in this section. (6. (i) If A is an event deﬁned for an interval on z. we solve for the conditional PDF for x and then ﬁnd the conditional PDF for z. we ﬁnd f z(γ ) = 3 1 1 (u(γ − e −1 ) − u(γ − e )) + δ(γ − e 0. g (1) (αi (γ )) (6.
1/3.2. Random variable x has the PDF 2 f x (α) = (α + 1)(u(α + 1) − u(α − 2)). (d )E(zz ≤ 1).5 ≤ α ≤ 1 ⎪ ⎩ 0. Drill Problem 6. (c ) f zA (2A). 1/24. .4. (c ) f zz≤1 (1/4z ≤ 1). We √ √ have α1 (γ ) = γ and g (1) (α1 (γ )) = 2 γ . 9 Random variable z = (x − 1)2 . (b) f xA (1A). 9 Random variable z = 2x 2 and event A = {x : x ≥ 0}. 5/27. 14 There is only one solution to γ = g (α) = α 2 on the interval 0 < α < 2 where f xA = 0. 0 ≤ α < 0. 2.2. f zA (γ A) = 3 √ 1 γ+√ 28 γ (u(γ ) − u(γ − 4)).5 ⎨ 2 f x (α) = 3(1 − α ).58 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS so that f xA (αA) = 3 (1 + α 2 )(u(α) − u(α − 2)).(d ) f zA (9A). Answers: 0. Drill Problem 6. 3. 1/2. (d ) f z(2). Random variable x has the PDF 2 f x (α) = α(u(α) − u(α − 3)). 1/8. 0. 48.2. (c )E(z). Determine: (a) P (A).1. Random variable x has a uniform PDF in the interval 0–8.2.43. otherwise. Random variable z = x 3 . Random variable 2 z = 3x + 1 . Consequently. Answers: 13.52. 1.3. (b) f z(6). 8/9.2. 0. (c ) f zz>1/8 (1/4z > 1/8). (b) f z(9/4). Random variable x has the PDF ⎧ ⎪ 9α 2 .Use the PDF method to determine: (a) f z(0).Use the PDF method to determine: (a) f z(1/27). Drill Problem 6. (d )σz . Answers: 1. 0. Drill Problem 6. Answers: 4/9. (b) f z(1/4). Use the PDF method to determine: (a) f z(1/4).
y) is computed using a CDF technique similar to the one at the start of this chapter.y = 0.y (α. Pay careful attention to the limits of integration to determine the range of integration in which the integrand is zero because f x. Here. .3 ONE FUNCTION OF TWO RANDOM VARIABLES Consider a random variable z = g (x. We require the volume under the surface f x. our efforts are concentrated on evaluating Fz through integrals. β) = Find the CDF for z = x + y.y (the region where f x.3. We have A(γ ) = {(α. Let us consider several examples to illustrate the mechanics of the CDF technique and also to provide further insight. y) created from jointly distributed random variables x and y. The ease in solution for Fz is dependent on transforming A(γ ) into proper limits of integration. 0 < α < 2. β)d α dβ. Solution. the probability distribution of z = g (x. even crucial. our efforts primarily dealt with algebraic manipulations.16) dFx. y) ≤ γ }.y is not constant) and the region A(γ ) is often most helpful.y (α. Random variables x and y have joint PDF f x.TRANSFORMATIONS OF RANDOM VARIABLES 59 6. with the ease of solution critically dependent on g (x.y and g (x. Remember that in the case of a single random variable. or Fx. y) : g (x. f x. β).15) (6. we have Fz(γ ) = P (z ≤ γ ) = P (g (x. 0 < β < 2 otherwise.1. With z = g (x.y (α.17) This result cannot be continued further until a speciﬁc Fx. Sketching the support region for f x.y = 0. The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6. β) where α ≤ γ − β: Fz(γ ) = ∞ −∞ γ −β −∞ 1/4. (6. y) are considered. y) ∈ A(γ )). these problems are not as straightforward and tractable as before. y). In this section. y) ≤ γ ) = P ((x. 0. where A(γ ) = {(x. β) : α + β ≤ γ }. Example 6. Because we are dealing with regions in a plane instead of intervals on a line.y (α. in the problem solution. y).
Fz(γ ) = 1. Solution. 0 < β < 1 otherwise.1. β)d αdβ. For γ < −1 we have Fz(γ ) = 0 and f z(γ ) = 0. we consider the complementary region (to save some work): Fz(γ ) = 1 − Finally. Fz(γ ) = 1 −γ 0 γ +β 1 d αdβ = (1 + γ )2 . 0.60 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.y (α. Example 6. 4 8 For 2 ≤ γ < 4.3(b).y (α. f x. For 0 ≤ γ < 2. With the aid of Figure 6.4(a). 2 . We have A(γ ) = {(α. with the aid of Figure 6.y (α. For γ < 0 we have Fz(γ ) = 0.3: Plots for Example 6. 0 < α < 1. β) where α ≤ γ + β: Fz(γ ) = ∞ −∞ γ +β −∞ 2 γ −2 2 γ −β 1 1 d αdβ = 1 − (4 − γ )2 . 4 8 1. for 4 ≤ γ . We require the volume under the surface f x. referring to Figure 6.3. β) : α − β ≤ γ }.3(a) we obtain Fz(γ ) = 0 2 0 γ −β 1 1 d αdβ = γ 2 .2.3. β) = Find the PDF for z = x − y. Random variables x and y have joint PDF f x. for −1 ≤ γ < 0.
3. so that f z(γ ) = 0.3. Find the CDF for z = x/y. hence. . Solution.4: Plots for Example 6. α γ β 1 1 γ 0 1 α FIGURE 6. Inside the support region for f x.3.5 it is easiest to integrate with respect to β ﬁrst: for 1 ≤ γ . we have α/β > 1. 2 γ +β 1 so that f z(γ ) = 1 − γ .3. where x and y have the joint PDF f x.2. for γ < 1 we have Fz(γ ) = 0. β) = 1/α.3. As shown in Figure 6. We have A(γ ) = {(α.4(b) (to save some work): Fz(γ ) = 1 − 0 1−γ 1 d αdβ = 1 − (1 − γ )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. 0<β<α<1 otherwise. Example 6.y . for 1 ≤ γ . Fz(γ ) = 1. β) : α/β ≤ γ }. Finally. 0. For 0 ≤ γ < 1.y (α. Fz(γ ) = 0 1 α α/γ 1 1 dβd α = 1 − . we consider the complementary region shown in Figure 6. so that f z(γ ) = 1 + γ .5: Integration region for Example 6.
β = r sin(θ). Referring to Figure 6.6(a). γ we ﬁnd with the aid of Figure 6. β) : α 2 + β 2 ≤ γ }. Solution. for 0 ≤ γ < 1. where x and y have the joint PDF f x.y (r cos(θ).y (α. we split the integral into two parts: one with polar coordinates. β) = 3α. we have Fz(γ ) = 1 for 2 ≤ γ . the other using rectangular coordinates. Fz(γ ) = 0 2π r 2 ≤γ f x. For γ < 0.6: Integration regions for Example 6.3. .4. r sin(θ))r dr d θ.62 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6. We have A(γ ) = {(α. Fz(γ ) = 0 π/4 0 √ γ 1 3 3r 2 cos(θ)dr d θ = √ γ 2 .6(b) that Fz(γ ) = or 3 1 3 Fz(γ ) = √ γ 2 − (γ − 1) 2 . Transforming to polar coordinates: α = r cos(θ ). 2 For 1 ≤ γ < 2. 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. Finally.4. 0 < β < α < 1 0. Find the CDF for z = x 2 + y 2 . With √ γ −1 sin(θ1 ) = √ . 2 π/4 θ1 0 √ γ √ α γ −1 0 3r cos(θ)dr d θ + 2 0 1 3αdβ d α.3. otherwise. we obtain Fz(γ ) = 0.
Deﬁne A(γ . Example 6. and let z = g (x.3.w (γ . 1 − 3 e −1 .ψ) (6.y (α.4 BIVARIATE TRANSFORMATIONS In this section. y) and w = h(x. Then Fz. Random variables x and y have joint PDF f x. y) : x + y ≤ γ } . and (d) f z(1). A(γ . γ ]) ∩ h −1 ((−∞. β). we consider a bivariate CDF technique.20) It is often difﬁcult to perform the integration indicated in (6.25.. Next. the conditional joint PDF is presented. y) from jointly distributed random variables x and y.4.TRANSFORMATIONS OF RANDOM VARIABLES 63 Drill Problem 6.19) dFx. Note that A(γ . Answers: 1 e −1 . 0 ≤ β ≤ 2 otherwise. 4 4 4 4 6. γ ]) = {(x.w .4. (6. y) and w = h(x. Random variables x and y have the joint PDF f x. 3 e −1 .1. the joint PDF technique for ﬁnding the joint PDF for random variables z and w formed from x and y is described. ψ) = {(x. (b) f z(−1). We have g −1 ((−∞. Then.1.20).y (α. ψ) = A(γ . y) ≤ γ . Random variable z = x − y . y) = x + y and w = h(x. y) ≤ γ and w = h(x. . i. Finally. y) = y.y (α. Find (a) Fz(−1/3). y). ψ]). (c )Fz(1). 0 ≤ α ≤ 2. 0. y) : g (x. β) = e −α e −β u(α)u(β). 3 e −1 . ψ) to be the region of the x − y plane for which z = g (x. β) = 0. y) ≤ ψ.1 Bivariate CDF Technique Let x and y be jointly distributed RVs on the probability space (S. h(x. First.18) (6.e. the case of one function of two random variables is treated by using the joint PDF technique with an auxiliary random variable. 6. ψ) = g −1 ((−∞. With z = g (x. P ). Solution. ﬁnd the joint PDF f z. y) ≤ ψ}. we ﬁnd the joint distribution of random variables z = g (x.
7. and h −1 ((−∞. which is illustrated in Figure. ψ) = ψ −∞ γ −β −∞ d Fx. The intersection of these regions is A(γ . ψ) = {(x.25. β). y) : y ≤ ψ}. β). Performing the indicated limits. β).y (α. 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 (γ . .7: Integration region for Example 6. ψ)}. it is usually easier to ﬁnd the joint PDF than to carry out the integral indicated in (6.y (α. we obtain f z.64 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS y ψ γ γ −ψ x FIGURE 6. y) : y ≤ min(γ − x.w (γ .4. 6.20). ψ).7 and (6. ψ]) = {(x. ψ) = 0. dFx.y (γ − ψ. We ﬁnd 1 ∂ Fz.w When the RVs x and y are jointly continuous. 0 < γ − ψ < 2. we ﬁnd that f z.w (γ .w directly. Instead of carrying out the above integration and then differentiating the result.y (α.20) we ﬁnd Fz. ψ) = lim h 1 →0 h 1 ∂γ and ∂ 2 Fz.1.w (γ . With the aid of Figure 6. ψ) = f x. substituting. 0. we differentiate the above integral to obtain the PDF f z. 0 < ψ < 2 otherwise. f z.w (γ .
23) Dividing both sides of the above by h 1 h 2 and letting h 1 → 0 and h 2 → 0. .24) J (α. βi ) = γ and h(αi . (6. h 2 ) = g −1 ((γ − h 1 . β) = . h 2 ).26) (6. ψ]).h 1 . β) ∂β ∂h(α.4.ψ. βi ) denote the unique element of h 1 →0 h 2 →0 lim lim Ii (γ . h 1 .y (αi .1. . ψ. h 2 ) = i (6. β) ∂α ∂g (α. ψ. y) . h 1 . . βi ) Ii (γ .y .w (γ . h 1 . ψ) ≈ i f x.25) i = 1. Then for small h 1 and small h 2 we have h 1 h 2 f v. βi ) = ψ. 2. Let x and y be jointly continuous RVs with PDF f x. and let z = g (x.21) Ii (γ . the approximation becomes an equality. . (6. β) ∂α ∂h(α. Theorem 6. Let h 1 > 0. (6. ψ). h 2 > 0 and deﬁne I (γ . h 1 .h 2 ) d α dβ. be the distinct solutions to the simultaneous equations g (αi .22) Let (αi . ψ)).TRANSFORMATIONS OF RANDOM VARIABLES 65 6. ψ. Let (αi (γ . (6. ψ. Partition I into disjoint regions so that I (γ . γ ]) ∩ h −1 ((ψ − h 2 . βi (γ . h 2 ).4. y) and w = h(x.27) .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. β) ∂β (6. Deﬁne the Jacobian ∂g (α.
Then f z. Random variables x and y have the joint PDF f x. we obtain f z.y (α1 . 1 The region of support for f z. Random variables x and y have the joint PDF f x. 0 < α < 1.y (γ − ψ.66 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where the indicated partial derivatives are assumed to exist. βi ) .8. There is only one solution to γ = α + β and ψ = β: α1 = γ − ψ and β1 = ψ. . Example 6.y (αi . ψ) = 1/4. J (αi .w (γ .w using the joint PDF technique are solving the simultaneous equations γ = g (αi .28) The most difﬁcult aspects to ﬁnding f z.y (α.4. 0. 0. Example 6. βi ) (6. f x. ψ) = ∞ i=1 f x. β) = 1/4. Let us consider several examples to illustrate the mechanics of this PDF technique. y) = y.w .2. y) = y. Solution.w (γ . The Jacobian of the transformation is J = Applying Theorem 1. y) = x 2 y and w = h(x. βi ) and ψ = h(αi .w is illustrated in Figure 6. βi ) for αi and βi . and determining the support region for f z. J  1 0 1 = 1 · 1 − 1 · 0 = 1. Find the joint PDF for z = g (x. ψ). β) = 12αβ(1 − α). Find the joint PDF for z = g (x. 0 < γ − ψ < 2. 0.4. 0 < β < 2 otherwise. y) = x + y and w = h(x. β1 ) = f x.w (γ . 0 < β < 1 otherwise. 0 < α < 2.3. 0 < ψ < 2 otherwise. ψ) = Substituting. we ﬁnd f z.y (α.
Let√ √ √ √ ting β1 = γ ψ we have α1 = γ /ψ. Example 6. −1 < β < 1 otherwise.y ( γ /ψ. 0 < ψ < 1 otherwise.TRANSFORMATIONS OF RANDOM VARIABLES 67 ψ 2 1 0 1 2 3 4 γ FIGURE 6. Find the joint PDF for z = g (x. the solution (α2 . 1 √ f x. β) = 0.w (γ . β) = 0 for α < 0. we ﬁnd α = γ /β = β/ψ. so that β 2 = γ ψ. we ﬁnd 2α1 β1 0 2 α1 = 2α1 β1 = 2 γ ψ. 0.w (γ .25(α + 1).8: Support region for Example 6. 2ψ 2ψ . − γ ψ) f z. Solving γ = αβ and ψ = β/α. Solution. Hence β −β/α 2 β α =2 . Solving γ = α 2 β and ψ = β. Then β2 = − γ ψ and α2 = − γ /ψ. y) = xy and w = h(x. y) = y/x. β1 ) is in the ﬁrst quadrant of the α − β plane. The other solution α2 = −α1 is not needed since f x.4. ψ) = √ 2 γψ √ 6(1 − γ /ψ). √ Solution. we ﬁnd α1 = γ /ψ and β1 = ψ. Random variables x and y have the joint PDF f x. Note that the solution (α1 . = 0. ψ) f z. β2 ) is in the third quadrant of the α − β plane. γ ψ) f x.y (− γ /ψ. The Jacobian is J1 = Applying Theorem 1.2. ψ) = + .y (α.4.y ( γ /ψ. −1 < α < 1. 1/α α √ √ √ √ f x.y (α.4. 0< √ γ /ψ < 1. We ﬁnd J = so that J 1 = 2ψ = J 2 .
we ﬁnd f z. This method is an alternative to the CDF technique presented earlier. γ ψ) 2ψ has support region speciﬁed by 0< or 0< Similarly.w have identical regions of support in the γ − ψ plane. γ /ψ < 1 and 0 < γ ψ < 1.w (γ . ψ) = 1/(4ψ). . the support region is γ −1 < ψ < γ < 0. √ √ f x.68 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS In the γ − ψ plane. − 1 < − γ ψ < 0. the two components of the PDF f z.y (− γ /ψ. y) by introducing auxiliary random variable w = h(x.y to random variable z = g (x. or γ −1 < ψ < γ < 0 otherwise. y) to ﬁnd f z. 0 < γ < ψ < γ −1 . 0. Auxiliary Random Variables The joint PDF technique can also be used to transform random variables x and y with joint PDF f x. (6. − γ ψ) 2ψ has support region speciﬁed by −1 < − γ /ψ < 0 and or 0< γ <1 ψ and 0 < γ ψ < 1. this support region is easily seen to be 0 < γ < ψ < γ −1 . γ <1 ψ and 0 < γ ψ < 1.w and then ﬁnding the marginal PDF f z using f z(γ ) = ∞ −∞ f z. in the third quadrant. This support region is illustrated in Figure 6. √ √ f x. Similarly.w (γ .y ( γ /ψ. Consequently. In the ﬁrst quadrant of the γ − ψ plane. ψ)d ψ.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.9. Finally.
Let auxiliary variable w = h(x.4. Solution. The Jacobian of the transformation is J = so that 2α 0 0 = 2α. y) = x 2 .9: Support region for Example 6. ⎧ √ 4 γψ √ f x. .4.TRANSFORMATIONS OF RANDOM VARIABLES 69 ψ 3 2 1 −3 −2 −1 1 2 3 γ −1 −2 −3 FIGURE 6.y is α = γ and β = ψ. 0 < β < 1 otherwise. y) = y. ψ)d ψ = 0 1 2ψ d ψ. 0. Example 6.w (γ .y ( γ . 0 < γ < 1. we ﬁnd α = √ √ ± γ and β = ψ. otherwise. 1 4αβ. 0 < ψ < 1 otherwise. 0 < α < 1.w (γ . We ﬁnd the marginal PDF for z as f z(γ ) = ∞ −∞ f z.5. 2 γ f z. ψ) ⎨ √ = 2ψ.y (α.4. for 0 < γ < 1 and f z(γ ) = 0. Solving γ = α 2 and ψ = β. Random variables x and y have the joint PDF f x. ψ) = = √ ⎩ 2 γ 0. β) = Find the PDF for z = g (x. The only solution inside the support region for f x.
w (γ .6. Find the PDF for z = g (x. f z(γ ) = 0 γ 2 4(γ − ψ)ψ d ψ = γ 3 . f z = 0.70 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS ψ 1 0 1 2 γ FIGURE 6. f z(γ ) = Otherwise. Solving γ = α + β and ψ = β. for 0 < γ < 1. ψ) = 4(γ − ψ)ψ. Let auxiliary variable w = h(x.y (γ − ψ.4. 0. 0 < α < 1. 0 < ψ < 1 otherwise. 0 < γ − ψ < 1. y).10.10. 1 0 1 = 1.10: Support region for Example 6. the above techniques apply to ﬁnd the conditional PDF for z = g (x. given event A.y (α. ψ) = f x. 0. Consider the problem where random variables x and y have joint 2 8 4(γ − ψ)ψ d ψ = − γ 3 + 4γ − . Referring to Figure 6. 1 The support region for f z. Solution. 3 3 γ −1 1 . 3 For 1 < γ < 2. y) = y. We ﬁnd J = so that f z. Example 6.w is shown in Figure 6. Conditional PDF Technique Since a conditional PDF is also a PDF. y) = x + y. we ﬁnd α = γ − ψ and β = ψ.4.6. β) = 4αβ. 0 < β < 1 otherwise. Random variables x and y have the joint PDF f x.
we have f zA (γ A) = f z(γ ) .5 0 α 1 3α dβ d α = .32) Example 6. (6.5 otherwise. βi (γ . y) ≤ 0. one of the following two methods may be used to determine the conditional PDF of z using the bivariate PDF technique. Random variables x and y have joint PDF f x. ψ). we will use the conditional PDF f x. βA) = P (A) ⎩ 0. Find the PDF for z = x + y. by the deﬁnition of a conditional PDF. J (αi (γ .y (α.. From Figure 6. and evaluate f z.7. i = 1. β) = 3α. β) in region A. βi (γ . . 0 < β < α < 0.11(a) we ﬁnd P (A) = A f x. Depending on whether the event A is deﬁned on the range or domain of z = g (x. is computed by ﬁrst evaluating f z using the technique in this section.30) ii. .TRANSFORMATIONS OF RANDOM VARIABLES 71 PDF f x. ψA)d ψ. . ψ)) (6. f x. β) and ψ = h(α. 2. given A. Then evaluate the marginal conditional PDF f zA (γ A) = ∞ −∞ f z. 8 consequently. Solution. i. If A is an event deﬁned for an interval on z. 0<β<α<1 otherwise. β) = 24α. .wA (γ . 0. given event A = {max(x. Then. introduce an auxiliary random variable w = h(x. are the solutions to γ = g (α. ψ).wA (γ . First. and we wish to evaluate the conditional PDF for random variable z = g (x. given that event A occurred. y). P (A) γ ∈ A. y). We begin by ﬁnding the conditional PDF for x and y. (6.31) where (αi .yA (αi (γ .y (α.4. β)d α dβ = 0 0. ψA) = ∞ i=1 f x. the conditional PDF.yA to evaluate the conditional PDF f zA as follows.5}. βi ). ⎧ ⎨ f x.y (α. If A is an event deﬁned for a region in the x − y plane. ψ)A) .y .yA (α. f zA . y).
Letting the auxiliary random variable w = h(x.5γ γ −0. β) = β is α = γ − ψ and β = ψ. f zA (γ A) = 0.5 < γ < 1 we obtain f zA (γ A) = For γ < 0 or γ > 1. the only solution to γ = g (α. 0 < ψ < γ − ψ < 0.wA is illustrated in Figure 6. For 0. ψA) = 24(γ − ψ).11: Support region for Example 6.wA (γ . The Jacobian of the transformation is J = 1 so that f z. 0.5γ 24(γ − ψ)d ψ = 9γ 2 . we ﬁnd f z. For 0. 0. β) = α + β and ψ = h(α. . The integration is easily carried out with the aid of Figure 6.5 24(γ − ψ)d ψ = 3(1 − γ 2 ).5 < γ < 1 we have f zA (γ A) = 0 0.wA (γ . can be found from f zA (γ A) = ∞ −∞ f z. ψA) = Substituting. f x.5 otherwise.11(b).7.72 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS FIGURE 6.yA (γ − ψ.11(b). ψA)d ψ. given A.4.wA (γ . 1 The support region for f z. ψA) . y) = y. The conditional PDF for z.
0. (c ) f z(1/2). Answers: 0. 1). Random variables x and y have joint PDF f x. (6. and the most general approach.2. (d ) f zA (1/2A).4. (b) f zA (−1/2A).7172.3. Find: (a) f z. Random variables x and y have joint PDF f x.4.y (α. 3/4.TRANSFORMATIONS OF RANDOM VARIABLES 73 Drill Problem 6. Random variable z = x 2 y. Answers: 0. (d ) f z(3/2). 0 < α < 1. β) = e −α−β u(α)u(β).4. 13/12. Determine: (a) f z.33) . is called the CDF technique.y (α. Random variables x and y have joint PDF f x. 0. (c )FzA (0A). Determine: (a) f z(−1). (d ) f z(3/2).w (−1. 0.5 SUMMARY This chapter presented a number of different approaches to ﬁnd the probability distribution of functions of random variables.4. Drill Problem 6. Random variable z = x + y and w = y 2 .y (α. 1/12. 0 < α < 1. 0. e −2 . Two methods are presented to ﬁnd the probability distribution of a function of one random variable. β) = 1. Drill Problem 6. y) : x + y ≤ 1}.2(α 2 + β). β) = 4αβ. Random variable z = x and w = xy. 0 < β < 1 otherwise. Random variable z = x − y and event A = {(x. 15/16. 6.3. 0 < β < 1 otherwise.y (α.w (1.1. (b) f z(−1/2). The ﬁrst.w (1. 1). 1/4). 1. 3/16. β) = 2β. Answers: 0. (b) f z(1/4). we write Fz(γ ) = P (z ≤ γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )). Answers: 0. 0 < α < 1. Drill Problem 6.4. (b) f z. z = g (x). 0 < β < 1 otherwise. Determine: (a) f zA (−3/ 2A). From the deﬁnition of the CDF. (c ) f z(1/2). 1. 0. Random variables x and y have joint PDF f x.
βi ) . involves evaluating f z(γ ) = where αi = αi (γ ) = g −1 (γ ). .38) The ease of solution here involves transforming A(γ ) into proper limits of integration. βi ) (6. However. ψ).y (α. βi (γ . The second approach. . y) ≤ γ }. . the CDF Fz may be found using the CDF Fx . (6. y) ≤ γ ) = P ((x. 2. Typically. . y) created from jointly distributed random variables x and y using two approaches. 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. 2. 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 random variables z = g (x. ψ)). i = 1.34) denote the distinct solutions to g (αi ) = γ . β). a bivariate CDF as well as a joint PDF technique were presented. By partitioning A(γ ) into disjoint intervals.40) ∞ i=1 f x. .36) (6. a CDF technique. (6. the PDF technique is much simpler to use than the CDF technique. where A(γ ) = {(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. .39) . the joint PDF for z and w can be found as f z. we evaluated the probability distribution of a random variable z = g (x. (6.w (γ .74 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where A(γ ) = {x : g (x) ≤ γ }. ψ) = where (αi (γ .37) dFx. y) and w = h(x. . Next. involves evaluating Fz(γ ) = P (z ≤ γ ) = P (g (x. y) from jointly distributed random variables x and y. called the PDF technique. . The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6. g (1) (αi (γ )) (6. y) : g (x. The ﬁrst approach. i = 1. Using the joint PDF technique. y) ∈ A(γ )).35) ∞ i=1 f x (αi (γ )) .y (αi . J (αi .
α<0 0≤α<1 1≤α and z = exp(−x). ﬁnd Fz. 6. y) to ﬁnd f z. .w . β) ∂α ∂g (α.43) where the indicated partial derivatives are assumed to exist. x<0 x > 0. βi ) = γ and h(αi . Find Fz using the CDF technique. Suppose random variable x has the CDF ⎧ ⎪ 0. The joint PDF technique can also be used to ﬁnd the probability distribution for z = g (x. y) from f x. −1 < α < 1 otherwise.41) (6. 2. The Jacobian ∂g (α. β) ∂β (6. 3. x + 1. β) ∂α J (α.42) (6. Using the CDF technique. 0.TRANSFORMATIONS OF RANDOM VARIABLES 75 are the distinct solutions to the simultaneous equations g (αi .6 PROBLEMS 1.y by introducing an auxiliary random variable w = h(x. ⎨ Fx (α) = α 2 . ∂h(α. ⎪ ⎩ 1. Given f x (α) = and z= Use the CDF technique to ﬁnd Fz. β) = ∂h(α. Typically.5(1 + α). Let random variable x be uniform between 0 and 2 with z = exp(x). the most difﬁcult aspect of the joint PDF technique is in solving the simultaneous equations. 0. βi ) = ψ. x − 1. and then integrating to obtain the marginal PDF f z. β) ∂β .
(c) f z. (b) Fz. Using the CDF method. 2α. . ⎪ ⎩ 2. ⎨ z= x. ﬁnd: (a) E(z) using f x . (b) Fz using the CDF method. (d) E(z) using f z (compare with the answer to part a). so that z = u(x). ﬁnd: (a) E(z) using f x . Determine: (a) a. (d) E(z) using f z (compare with the answer to part a). ⎨ z= x. The PDF of random variable x is 1 1 f x (α) = e − 2 (α+2) u(α + 2). Random variable x represents the input to a fullwave rectiﬁer and z represents the output. (b) Fz using the CDF method. ⎧ ⎪ −1. Find Fz for z = x 2 using the CDF method. 5. determine Fz. so that z = x. 9. ⎪ ⎩ 1. Given that x is uniformly distributed between −2 and 2. 0<α<1 otherwise. Given f x (α) = and a 1 + α2 x ≤ −1 −1 < x < 1 x ≥ 1. x < −1 −1 ≤ x ≤ 2 x > 2. 6.76 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 4. Random variable x has the PDF f x (α) = e −α u(α). 7. Random variable x represents the input to a halfwave rectiﬁer and z represents the output. Suppose random variable x has the PDF f x (α) = Random variable z is deﬁned by ⎧ ⎪ −1. 2 Find Fz using the CDF technique when z = x. 0. (c) f z. 8. Given that x is uniformly distributed between −2 and 2.
0. 0. 0. (b) Fz using the PDF technique. ⎪ ⎨ z = 2.5 ≤ x < 1. Let z = 1/x 2 . Random variables x and z are the input and output of a quantizer. 14. (b) z = (x − 1/2)2 .TRANSFORMATIONS OF RANDOM VARIABLES 77 10. ⎪ ⎪ ⎪ 1.5. 0<α<1 otherwise.5 ≤ x < 3. ⎪ ⎩ 1. Random variable x has the following PDF f x (α) = 4α 3 .25. 13. Use the CDF technique to determine Fz.5 −0. 12. Random variable x has the CDF ⎧ ⎪ 0.5 ≤ x ≤ 0 0 < x. ⎨ z = 0.5 2. With z = 100 − 25x. ⎪ ⎪ ⎪ ⎩ 4. (c) z = (x − 1/4)2 .49).5 0. and ⎧ ⎪ x. Given that the input follows a Gaussian distribution with x ∼ G(2. 11. −1 < α < 2 otherwise. Suppose f x (α) = (1 + α 2 )/6. Find the PDFs for the following random variables: (a) z = x 3 . ﬁnd: (a) Fz using the CDF technique. Random variable x has the PDF f x (α) = e −α u(α). ﬁnd f z using the CDF method. ⎪ ⎪ 3.5 1. The relationship between them is deﬁned by: ⎧ ⎪ 0.5 x ≥ 3. ⎨ Fx (α) = (α 2 + 2α + 1)/9. α < −1 −1 ≤ α < 2 2≤α x < −0.5 ≤ x < 2. ⎪ ⎩ x. . x < 0. Determine Fz.
Assume the diode is ideal. Random variable x is Gaussian with mean ηx = 0 and standard deviation σx = 1. 19. 15.12: Plot for Problem 15. Assume the diodes are ideal. The voltage x in Figure 6. Find the PDF f z.14 is a Gaussian random variable with mean ηx = 0 and standard deviation σx = 3. + 1K Ω + x 3K Ω z − − FIGURE 6.13: Circuit for Problem 16. Assume that g (x) is a second degree polynomial for x ≥ 0. Random variable x has the CDF 0.78 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1 −2 −1 −1 1 2 3 x FIGURE 6. ﬁnd the PDF f z. Find the PDF f z. 16.25(α + 1). Fx (α) = 0.12.16.15 is a Gaussian random variable with mean ηx = 1 and standard deviation σx = 1. The voltage x in Figure 6. .13 is a random variable which is uniformly distributed from −1 to 2. Assume the diode and the operational ampliﬁer are ideal. ⎧ ⎪ ⎨ α < −1 −1 ≤ α < 3 3 ≤ α. Find the CDF Fz. with RV z = g (x). With z = g (x) shown in Figure 6. 17. and g (x) shown in Figure 6. ⎪ ⎩ 1. 18. The voltage x in Figure 6. Find the PDF f z.
0. Find f z using the PDF technique. . 0<α<1 otherwise.14: Circuit for Problem 17. Find the PDF of z = 2x 2 using the PDF technique. 21. 22. 23. 0. Let z = cos(x). 20. 0. Find f z if: (a) f x (α) = 1/π.5). α < π/2 otherwise.15: Circuit for Problem 18. 24. −1 < α < 2 otherwise. Random variable x has the PDF f x (α) = 2(α + 1)/9. Random variable x has the following PDF f x (α) = Find Fz if z = x 2 . Random variable x is uniform in the interval 0 to 12. α + 0.5δ(α − 0. 5K Ω 1K Ω x z FIGURE 6. Find f z if z = 1/x 2 and x is uniform on −1 to 2.TRANSFORMATIONS OF RANDOM VARIABLES + R + 79 x 2V 3V z − − FIGURE 6. Random variable z = 4x + 2.
0<α<3 otherwise. Random variable x is uniform between −1 and 1. ⎧ ⎪ 0. . Random variable z = (x − 1)2 and event A = {x : x ≥ 1/2}. x 2. given event A.5 1 2 x FIGURE 6. A voltage v is a Gaussian random variable with ηv = 0 and σv = 2. (b) 25. Find the PDF of z = −2 ln(x) using the PDF technique.16: Transformation for Problem 19. Given that x has the CDF f x (α) = 8α/π 2 . 27. Find the PDF of random variable z. 28. ⎪ ⎩ 1. Random variable w = v2 /R represents the power dissipated in a resistor of R with v volts across the resistor. α<0 0≤α<1 1 ≤ α. 0. 0 < α < π/2 otherwise. (b) f wA if A = {v ≥ 0}. Find (a) f w . x<0 x ≥ 0. 26. Random variable x has the PDF f x (α) = 2α/9. ⎨ Fx (α) = α.80 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1.4 −2 −1 −0. 0. Random variable z= Using the PDF technique. x. ﬁnd f z.
5 < α < 1 otherwise. Given that random variable x has the PDF ⎧ ⎪ α. ⎪ ⎩ 0. Find f z if z = 1/x and f x (α) = π(α 2 1 .5α(u(α) − u(α − 2)). 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)). 33. (c) E(z).75(1 − α 2 ). 0<α<1 1<α<2 otherwise. ⎨ f x (α) = 2 − α. 35. 36. Random variable x has an exponential distribution with mean one. 0. Determine f z using the PDF technique. Determine f z using the PDF technique. Determine: (a) f z. and z = x 2 . use the PDF technique to determine: (a) f zA (γ A) if A = {x : x > 2}.TRANSFORMATIONS OF RANDOM VARIABLES −x 81 29. 30.5 0. −1 < α < 1 otherwise. (d) E(z2 ). + 1) 31. Suppose f x (α) = 2α. ﬁnd f z if z = x and x is a standard Gaussian random variable. Suppose RV x has PDF f x (α) = 0. and z = −2 ln(x). Using the PDF technique. f x (α) = 3(1 − α 2 ). 32. Let f x (α) = 0. ﬁnd the PDF of z = x 2 . 0. (b) f zB (γ B) if B = {z : z < 1/2}. 0<α<1 otherwise. ⎧ ⎪ ⎨ 0 < α < 0. and z = 8x 3 . (b) Fz. 34. Given 9α 2 . Using the PDF technique. ⎪ ⎩ 0. .
40. Let random variable x have the PDF f x (α) = 1/12. (b) Determine f z if z = (2x + 2)u(x). ⎪ ⎩ 0.17: Transformation for Problem 40. ﬁnd f z. 38. −2 < α < 1 1 < α < 2. 1/4.25γ . in addition. 0 ≤ α < 2π otherwise.5. 3 z = g(x) 2 1 x −1 0 1 2 FIGURE 6. Find the PDF f z. ⎪ ⎩ 0. P (x = −2) = P (x = 2) = 0. 41. 0<γ <1 1<γ <2 otherwise. −1 < α < 0 0<α<2 otherwise. Find f z if z = sin(x) and f x (α) = 1/(2π). 4 4 Random variable z = g (x).5 − 0. Random variable x has PDF 1 1 f x (α) = δ(α + 1) + δ(α) + (u(α) − u(α − 2)).5 − 0. ⎨ f x (α) = 0.17. ⎨ f z(γ ) = 0. If z = 1/x. Random variable x has PDF f x (α) = α2 (u(α + 1) − u(α − 2)).82 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 37.25. 39. (a) Find the transformation z = g (x) so that ⎧ ⎪ 1 − 0. .25α. Random variable x has the PDF ⎧ ⎪ 0.25γ . 0. with g shown in Figure 6.
y (α. 47. ﬁnd the PDF f g using convolution. With g denoting the equivalent conductance of a parallel connection of R1 and R2 . β) = e −α . 0. With r denoting the equivalent resistance of a series connection of R1 and R2 . The PDF for random variable x is f x (α) = e −α u(α). 46. Random variable z = g (x). 0<β<α otherwise. 43. 45. just write the integral(s) necessary to ﬁnd Fz. (b) f zA with A = {x > 0}. The joint PDF of x and y is f x. ﬁnd Fz when x and y have the joint PDF f x. 0. 0<β<α<1 otherwise. 0.TRANSFORMATIONS OF RANDOM VARIABLES z = g( x ) 2 1 x 83 −1 0 1 2 FIGURE 6. ﬁnd the PDF fr using convolution. With z = e −x .25α.18: Transformation for Problem 41. . determine: (a) f zA (γ A) where A = {x : x > 2}. Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 . Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 .y (α. β) = 0. 44. If z = x/y. Suppose the voltage v across a resistor is a random variable with PDF f v (α) = 6α(1 − α). 0<α<1 otherwise. with g shown in Figure 6.18. (b) f zB (γ B) where B = {z : z < 1/2}. Use the PDF technique to ﬁnd: (a) f z. With z = x + y. write an expression(s) for Fz(γ ) (do not solve. 42.
y (α. 0. 0 < α < 1. 0 < β otherwise. Moreover. ﬁnd f z. β) = 4αβ. 0 < α < 1. The joint PDF for random variables x and y is f x.y (α. 52. The joint PDF for random variables x and y is f x. 50.y (α. Find f z. 0. The joint PDF for random variables x and y is f x.84 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS and that resistance r is a random variable with PDF fr (β) = 1/12. 51. 0.wA . 0.w . 94 < β < 106 otherwise. 0 < β < 1 otherwise. 0. suppose that v and r are independent random variables. The joint PDF for random variables x and y is f x. Let z = 2x + y and w = x + 2y. 53. β) = With z = y/x. 49. β) = 8αβ. 0<β<α<1 otherwise. β) = α + β.y (α.w . Find: (a) f z. 0. . 1/α. Find the joint PDF for random variables z = x + y and w = x − y.y (α. β) = 3(α 2 + β 2 ). (b) f zA . Find the PDF for power. Find: (a) f z. Let z = x 2 and w = xy. p = v2 /r .w . 0 < α 2 + β 2 < 1. Let A = {(x. 48. 0. z = x and w = x 2 + y 2 . 0 < α. The joint PDF for random variables x and y is f x. (b) f z. Find f z. 0 < β < 1 otherwise. 0 < α. The joint PDF for random variables x and y is f x. y) : x > y}.5βe −α . 0 < β < 2 otherwise.y (α. 0<β<α<1 otherwise. 0. β) = Let z = x/y and w = y.
0. Find f z. α < 1. The joint PDF for random variables x and y is f x. 0 < α. Find f z.y (α. 12αβ(1 − α). 55.w .25(α + β). The joint PDF for random variables x and y is f x.y (α. The joint PDF for random variables x and y is f x. 0 < β < 1 otherwise. Let z = x + y and w = x/(x + y).TRANSFORMATIONS OF RANDOM VARIABLES 85 54. 0 < β < π otherwise. The joint PDF for random variables x and y is f x.y (α. β) = 0. β < 1 otherwise. β) = e −α−β . 0 < α < 1. Find f z. 0.y (α. . 0. 0 < β otherwise. β) = Let z = x 2 y.w using the joint CDF technique. Let z = xy and w = y/x.w . 0. 0 < α < 1. 56. β) = α sin(β). 57. Find: (a) f z. (b) f z. Let z = x 2 and w = cos(y).
.
991440 0.593127 0.977408 0.4 0.598737 0.999936 1.077760 0.820197 0.1: Bernoulli CDF for n = 5 and n = 10 n=5 p 0.428415 0.984375 0.987205 1.000000 0.999540 0.15 0.35 0.981547 0.031250 +0 0.168070 0.632813 0.256218 0.773781 0.999997 0.998365 1.05 0.998617 +1 0.000000 0.994748 0.999990 0.973388 0.737280 0.999680 0.993280 0.764831 0.812500 +4 1.87 APPENDIX A Distribution Tables TABLE A.116029 0.912960 0.990126 1.000000 0.918540 0.050328 0.999970 0.336960 0.682560 0.989760 0.000000 0.2 0.999991 0.443705 0.3 0.000000 0.000000 0.15 k 0 0 0 0 0 0 0 0 0 0 k 0 5 0 5 0 5 +0 0.999991 +3 0.1 0.05 0.25 0.836920 0.000000 0.988496 1.237305 0.835210 0.590490 0.997570 0.929809 1.868780 0.999853 0.942080 0.1 0.997773 0.544300 0.736099 0.998971 1.45 0.5 p 0.945978 0.913862 1.348678 0.000000 (Continued ) +2 0.969220 0.500000 +3 0.968750 .528220 0.000000 +4 0.998842 0.896484 0.999865 +2 0.327680 0.196874 0.999924 0.000000 0.999023 0.950030 1.187500 n = 10 +1 0.
999996 0.999584 0.244025 0.998322 0.266038 0.999895 0.996494 0.028248 0.879126 0.751495 0.463291 0.999997 1.149308 0.999023 TABLE A.999970 0.085954 0.1: Bernoulli CDF for n = 5 and n = 10 (Continued) n = 10 p 0.945238 0.45 0.5 k 0 5 0 5 0 5 0 5 0 5 0 5 0 5 +0 0.999996 1.375810 0.000000 0.921873 0.023257 0.000000 1.046357 0.000000 +4 0.623047 +1 0.549043 0.815939 0.989258 +4 0.633103 0.980272 0.998410 0.905066 0.056314 0.002533 0.013463 0.973976 0.000000 .525593 0.999966 1.677800 0.000000 0.000000 0.738437 0.649611 0.000000 1.999460 0.994533 1.898005 0.999659 0.261607 0.1 0 5 10 +0 0.999856 0.382783 0.945313 +3 0.828125 +2 0.000000 0.987280 1.999689 1.952651 0.4 0.05 k 0 5 10 0.995179 0.967206 1.000000 0.963800 1.999999 0.989408 0.944444 0.000000 +2 0.107374 0.829047 0.993631 0.999994 0.010742 0.999922 0.000000 1.995498 0.376953 0.849732 0.999385 1.513827 0.000977 0.775875 0.000000 1.987705 0.000000 +3 0.999136 0.205891 0.3 0.2 0.054688 0.382281 0.504405 0.006047 0.997750 1.972608 0.25 0.000000 +1 0.999947 1.88 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.999972 0.000000 0.171875 0.833761 0.099560 0.167290 0.2: Bernoulli CDF for n = 15 n = 15 p 0.35 0.
042421 0.564282 0.999328 0.999215 1.2 0 5 10 0.3 0 5 10 0.4 0 5 10 0.236088 0.167126 0.000000 0.090502 0.721621 0.000000 0.990652 0.995760 1.984757 1.943380 0.000127 0.000000 0.452160 0.938949 0.004748 0.035268 0.403216 0.25 0 5 10 0.013363 0.217278 0.2: Bernoulli CDF for n = 15 (Continued) n = 15 p 0.604225 0.172696 0.515491 0.999999 0.996394 1.999988 0.998072 0.754842 0.998893 0.996307 +3 0.000000 0.45 0 5 10 0.461287 0.080181 0.126828 0.868857 0.904953 0.999996 0.000470 0.014179 0.938295 0.851632 0.648162 0.DISTRIBUTION TABLES 89 TABLE A.000000 0.000031 0.981941 0.999919 1.010652 0.000000 0.001692 0.005172 0.296868 0.003693 0.000488 0.987557 1.849121 0.974534 0.822655 0.999991 0.609813 0.996347 1.318586 0.027114 0.999887 1.000000 0.351943 0.995807 1.398023 0.000000 0.999390 1.999908 0.000000 0.059235 0.949987 0.999988 0.835766 0.000000 0.886769 0.035184 0.983190 0.999969 .000000 0.966167 0.15 k 0 5 10 0.001562 0.999999 0.997169 0.940765 +1 0.5 0 5 10 +0 0.696381 0.957806 0.087354 0.150879 0.999879 0.017578 0.999999 0.999521 0.999975 0.120399 0.999512 +4 0.686486 0.818240 0.999205 1.35 0 5 10 0.303619 0.982422 +2 0.999885 0.786897 0.999999 0.999721 0.982700 0.999992 1.061734 0.500000 0.923071 0.260760 0.999994 0.999943 0.000000 0.993673 0.653504 0.
772272 0.000000 0.000000 0.000000 0.000000 0.999995 1.978065 0.999999 +3 0.000000 0.000000 1.999739 1.000000 0.000000 0.035483 0.000000 0.999816 1.000000 0.997405 1.999065 1.999970 1.924516 0.038760 0.000000 1.998479 1.000000 1.416371 0.000000 0.999966 1.416625 0.000000 0.25 0 5 10 15 0.999999 1.990018 0.997614 1.988747 0.118197 0.984098 1.000000 .000000 0.932692 0.946833 0.003171 0.002133 0.000000 1.107087 0.000000 1.3: Bernoulli CDF for n = 20 n = 20 p 0.000000 1.000000 1.000000 1.000000 1.629648 0.414842 0.886669 0.000000 0.000000 0.225156 0.999584 1.999752 1.000000 1.999898 1.999437 1.999993 1.000000 0.411449 0.998721 1.000000 0.000000 0.986136 0.829847 0.3 0 5 10 15 0.000000 +4 0.245396 0.237508 0.867047 0.999999 1.000798 0.996058 1.999957 1.05 k 0 5 10 15 0.000000 1.999950 +1 0.404896 0.000000 0.358486 0.804208 0.000000 0.999996 1.000000 1.676927 0.391747 0.999689 1.2 0 5 10 15 0.999940 1.000000 1.608010 0.000000 0.091260 0.999998 1.617173 0.175558 0.000000 0.980421 0.999961 1.121577 0.785782 0.999994 +2 0.000000 0.000000 0.000000 0.994079 0.15 0 5 10 15 0.952038 0.999994 0.044376 0.069175 0.999671 1.007637 0.735840 0.206085 0.998671 1.762378 0.999997 1.994862 0.35 0 5 10 15 +0 0.999985 1.90 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.000181 0.011529 0.000000 0.997426 1.956825 0.999999 0.024313 0.1 0 5 10 15 0.878219 0.000000 0.959075 0.000000 0.898188 0.000000 0.913307 0.647725 0.967857 0.993985 0.000000 1.982855 0.012118 0.601027 0.
000111 0.5 0 5 10 15 +0 0.755337 0.1 0.909796 0.000000 0.999605 1.963064 0.878099 0.999999 0.670320 0.057659 0.000000 (Continued ) .606531 0.872479 0.000000 0. 0.996400 0.411901 0.999910 +1 0.999996 0. 1. .999845 0. .3 0.5 λ 0.999984 0.000006 0. .993566 1.938448 0.999953 0.588099 0.000927 0.4 0.591361 0.2.750711 0.982477 0.050952 0.994091 +1 0.999991 +2 0.000020 0.999943 0.055334 0.415893 0.942341 0.999986 0.000524 0.998248 1.999998 0.2 0.251722 0.131588 0.965858 0.999939 0.250011 0.998388 1.904837 0.999224 0.5 0.999997 0.000037 0.985612 1.999997 0.999214 1.000201 0.976885 1.000000 0.999995 0.943474 0.998469 0.252006 0.844195 0.999961 0.004933 0.740818 0.595599 0.000000 0.998712 +2 0.005909 0.999999 +3 0.6 0.998852 0.748278 0.125599 0.018863 0.999723 0.000000 0.993534 1. .999734 0.45 0 5 10 15 0. 4.999999 TABLE A.1. .003611 0.979305 0.4: Poisson CDF for λ = 0.496585 0.996642 1.994247 1.7 k 0 0 0 0 0 5 0 5 0 5 +0 0. 2.5.001288 0.414306 0.000000 0.000001 0.868412 0.978586 0.000000 +4 1.978971 0.999964 0. .999799 +3 0.999683 0. .999980 +4 0.818731 0.DISTRIBUTION TABLES 91 TABLE A.869235 0.548812 0.129934 0.4 k 0 5 10 15 0.999828 1.941966 0.000000 0.3: Bernoulli CDF for n = 20 (Continued) n = 20 p 0.992074 0.000000 0.000000 0. 1. .000000 0.015961 0.020695 0.995321 0.
999917 0.999995 +2 0.999074 0.815263 0.995753 0.735759 0.831051 0.986541 1.4: Poisson CDF for λ = 0.92 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.999748 1.808847 0.000000 3 0 5 10 3. .952577 0. 0.973261 0.9 1 1.000000 0.173578 0.997656 1.999995 0.532104 0.725445 0.999984 0.919699 0.000000 0.999924 0.999926 1.999999 +3 0.990126 0.991868 0.5 0 5 10 4 0 5 10 4.808792 0.557825 0.983436 0.999830 0.857614 0.536633 0.999954 0.978637 0.947347 0.998860 1.647232 0.998903 0.676676 0.998981 0.091578 0.997596 0.000000 0.367879 0. . .1.999711 0.990920 1.543813 0.287297 0.999929 0. .999979 0.999980 0.982907 0.934358 0.937143 0.406570 0.999998 0.999726 0.988095 0.981424 0.999972 0.981012 0.999195 0.996197 0.5 0 5 10 15 . 2.997160 0.999406 0.030197 0.999997 0.199148 0.000000 0.5 (Continued) λ 0.966491 0.423190 0.998897 0.2.135888 0.857123 0. 1.996685 0. .785130 0.223130 0.018316 0.999763 0. 4.998589 1.999938 0.449329 0.702930 0. 1.999656 0.628837 0.5.891178 0.999996 0.948866 0.999816 0.999924 0.985813 0. .995466 0.5 2 2.999987 0. .999957 0.999999 0.061099 0.000000 0.959743 0.406006 0.996340 1.082085 0.000000 0.5 k 0 5 0 5 0 5 0 5 0 5 0 5 10 +0 0.433470 0.320847 0.000000 +4 0.999980 +1 0.999998 0.913414 0.934712 0.889326 0.999999 0.000000 0.772482 0.999708 0.238103 0.8 0.011109 0.995544 0.342296 0.957979 0.999996 0.999981 0.999990 0.049787 0. .757576 0.999085 0.916082 0.999723 1.993331 0.135335 0.
996372 0. .151204 0.999302 0.987189 0.006738 0. .999800 0.999931 0.862238 0.672758 0. .043036 0.986305 0.999982 0.124652 0.999995 0.973000 0.999995 0.776408 0.992900 0.999570 0.686036 0.524639 0.999210 0.997593 0.998840 0.011276 0.661967 0.999949 0.999980 0.994547 0.916076 0.933161 0.995549 0.999889 1.300708 0.998315 0.017351 0.957334 0.743980 0.059145 0.791573 0.999984 0.983973 0.762183 0.946650 0.002479 0.026564 0.999999 0.989012 0.997981 0. .999961 +1 0.877384 0.DISTRIBUTION TABLES 93 TABLE A.901479 0.088376 0.979908 0.847238 0.931906 0.999870 0.999638 0.000000 (Continued ) 5.029636 0.241436 0.000912 0.369041 0.007295 0.378155 0.994283 0.440493 0.201699 0.223672 0.111850 0.526524 0.000000 0.999996 +3 0.991173 0.999981 0.809485 0.920759 0.997044 0.999042 0.020257 0.5 0 5 10 15 20 .999937 0.081765 0.729091 0.040428 0.974749 0.598714 0.5 0 5 10 15 6 0 5 10 15 6.001503 0.968172 0.528919 0.132062 0.830496 0.357518 0.966120 0.172992 0.615961 0.004701 0.999999 0.995392 0.5 λ 5 k 0 5 10 15 +0 0.946223 0.265026 0.957379 0.999956 0.999849 0.999401 0.978435 0.004087 0.999697 0.5 0 5 10 15 7 0 5 10 15 7.999999 +4 0.999995 0.866628 0.999987 +2 0.061969 0.989740 0.999943 0.606303 0.449711 0.894357 0.999774 1.998600 0.998041 0.000553 0.5: Poisson CDF for λ = 5. 8.285057 0.445680 0.5.999825 0. 5.999491 0.
982743 0.220221 0.803008 0.323897 0.999855 0.009283 0.898136 0.001933 0.999944 0.998406 0.592547 0.130141 0.5: Poisson CDF for λ = 5.6: Poisson CDF for λ = 9.645328 0.999997 8.958534 0.848662 0.000499 0.000075 0.999991 0.999971 +3 0.999747 0.5 (Continued) λ 8 k 0 5 10 15 20 +0 0.994680 0.999990 +4 0.067086 0.001234 0.5.000203 0.164949 0.999825 0. 8.115691 0.391824 0.999967 0.998038 0.191236 0.948589 0.94 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.998703 0.999933 0.000335 0. 10.004164 0.791556 +3 0. 12.014860 0.991769 0. .909083 0.999979 0.332820 0.455653 0.982273 0.003019 0.006232 0.457930 0.149597 0.999789 +1 0.940008 0.010336 0. 9.042380 0. .074364 0.991072 0.965819 0.206781 0.521826 0.000786 0.716624 0.999639 0.040263 0.836430 0.998944 0.997574 0.988894 0.587408 0.763362 0.705988 0.815886 0.256178 0.916542 9. 11.751990 0.313374 0.999975 0.000045 0.996282 0.996998 0.999996 0.013754 0.966527 0.583040 +1 0.523105 0.926149 0.999465 0.029253 0.999921 +2 0.977964 0.999141 0.999989 0.999999 0. .021226 0.268663 0.864464 +4 0.888076 0. 5.972575 0.995716 0.099632 0.999350 0. .030109 0.993387 0.000123 0.986167 0.054964 0.652974 0.088528 0. 13 λ 9 k 0 5 10 15 20 +0 0.385597 0.002769 0.452961 0.5.999561 0.999906 0.875773 0.936203 0.5 0 5 10 15 20 10 0 5 10 .696776 +2 0.5 0 5 10 15 20 TABLE A.
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 )
039204 0.025588 +0.80 1.158655 0.133500 0.028716 0.082264 0.047460 0.15 1.043633 0.146859 0.074934 0.211855 0.85 1.166023 0.088508 0. .208970 0.75 0.90 1.65 1.123024 0.119000 0.055917 0.186733 0.176186 0.023295 .151505 0.032884 0.026190 0.04 0.149170 0. .102042 0.107488 0.140071 0.090123 0.093418 0.99 (Continued) α 0.135666 0.050503 0.098525 0.144572 0.142310 0.98 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.178786 0.03 0.80 0.037538 0.038364 0.8: Marcum’s Q function for α = 0.030742 0.083793 0.034379 0.117023 0.55 1.90 0.091759 0.40 1.032157 0.235762 0.70 0.040059 0.192150 0.01.045514 0.040929 0.026803 0. .041815 0.35 1.033625 0.060571 0.066807 0.206108 0.184060 0.95 1.049471 0.080757 0.051551 0.042716 0.217695 0.214764 0.70 1.229650 0.156248 0.220650 0.072145 0.50 1.073529 0.024419 +0.173609 0.241964 0.129238 0.127143 0.058208 0.200454 0.063008 0.10 1.029379 0.194894 0.121001 0.095098 0.45 1.85 0.077804 0.232695 0.111233 0.054799 0.238852 0.115070 0.086915 0.113140 0.226627 0.163543 0.125072 0.052616 0.027429 0.070781 0.25 1.028067 0. .60 1.137857 0.171056 0.057053 0.30 1.030054 0.046479 0.161087 0.223627 0.096801 0.103835 0.024998 +0.105650 0.181411 0.20 1.069437 0.061780 0.035148 0.75 1.02 0.059380 0.076359 0.044565 0.189430 0.01 0.168528 0.023852 +0.109349 0.053699 0.068112 0.197662 0.131357 0.203269 0.153864 0.035930 0.036727 0. 1.05 1.95 +0.048457 0. 0.00 1.065522 0.100273 0.031443 0.085344 0.079270 0.00 0.064256 0.
018309 0.001306 0.007549 0.016586 0.15 2.004145 0.005085 0. 3.004940 0.006947 0.004799 0.003364 0.001350 0.011604 0.003264 0.002118 0.002186 0.006387 0.70 2.25 3.001395 0.002477 0.013209 0.001107 0.001866 0.001223 0.000664 0.001144 0.000538 0.005543 0.03 0.000450 +0.DISTRIBUTION TABLES 99 TABLE A.002890 0.85 2.9: Marcum’s Q function for α = 3.005386 0.002052 0.30 +0.003467 0.002635 0.55 2.000874 0. 3.007344 0.008424 0.011304 0.002555 0.002980 0.000519 0.40 2.017003 0. .010724 0.003793 0.000687 0.012224 0.000968 0.00 2.012874 0.001441 0.000789 0.10 2.000598 0.013903 0.90 2.001183 0.022750 0.000577 0.000619 0.008894 0.017864 0.017429 0.000816 0.009642 0.35 2.50 2.10 3.001589 0.008656 0.002401 0.001489 0.02 0.003167 0.006756 0.002256 0.007143 0.001807 0.000557 0.009903 0.001264 0.005868 0.05 3.002327 0.000419 (Continued ) .20 3.01 0.00 3.015778 0.15 3.001035 0.60 2.05 2.018763 0.001988 0.001070 0.019699 0.000501 0.000845 0.001926 0.001641 0.007760 0.001538 0.004025 0.021692 0.003907 0.003681 0.008198 0.65 2.011011 0.006569 0.004661 0.25 2.000711 0.000467 +0.003072 0.75 2.003573 0.011911 0. .99 α 2.006037 0.015386 0. .004269 0.004397 0.005234 0.000483 +0.005703 0.001001 0.04 0.001750 0.80 2.012545 0.002803 0.000641 0.014629 0.007976 0.009387 0.000762 0.000904 0.021178 0.000434 +0.010170 0.013553 0.016177 0.010444 0.009137 0.01.000736 0.022216 0.014262 0.30 2.20 2.001695 0.004527 0.019226 0.95 3.45 2.020182 0.015003 0.020675 0.00 0.000936 0. .002718 0.006210 0.
000108 0.80 3.000121 0.55 3. .99 (Continued) α 3.01 0.000153 0.60 3.000172 0.000075 0.000200 0. .100 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.000390 0.000302 0.50 3.000054 0.04 0.000082 0.000136 0.000112 0. .000362 0.000404 0.000034 +0.000044 0.000052 0.85 3.01.000260 0. 3.000251 0.000142 0.000185 0.03 0.000092 0.000280 0.000350 0.000165 0.9: Marcum’s Q function for α = 3.000096 0.000242 0.000033 .000233 0.000072 0.70 3.000104 0. 3.000313 0.00 0.000070 0.000037 +0.000041 0.000126 0.000147 0.000048 0.000224 0.40 3.95 +0.000064 0.000270 0.000208 0.000117 0.000067 0.000376 0. .000131 0.000050 0.000057 0.000325 0.000085 0.000159 0.000046 0.02 0.000193 0.000078 0.000062 0.000216 0.000179 0.000100 0.35 3.65 3.000039 +0.000042 0.000291 0.000036 +0.000059 0.000337 0.90 3.75 3.45 3.000088 0.