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
KEYWORDS Probability Theory. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. The primary subjects of the ﬁnal chapter are methods for determining the probability distribution of a function of a random variable. The aim of all three books is as an introduction to probability theory. The audience includes students. Drill problems. The exponential. The mathematical background required of the reader is basic knowledge of differential calculus. follow most sections. straightforward exercises designed to reinforce concepts and develop problem solution skills. Next. Gaussian random variables . Probability and Statistics for Biomedical Engineers. Poisson distributions.iv ABSTRACT This is the third in a series of short books on probability theory and random processes for biomedical engineers. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. Poisson and Gaussian distributions are introduced. 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. This book focuses on standard probability distributions commonly encountered in biomedical engineering. as well as important approximations to the Bernoulli PMF and Gaussian CDF. Gaussian distributions Bernoulli PMF and Gaussian CDF. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. Random Processes. Many important properties of jointly Gaussian random variables are presented. Exponential distributions. Pertinent biomedical engineering examples are throughout the text. Then. Engineering Statistics.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Univariate CDF Technique . . . . . . . . . . . . . . 57 6. . . . . . . .1 Constant Contours. . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Poisson Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Continuous Random Variable . . . . . . . . . .4. . . . . . . .2. . . . . . . 45 6. . . 73 6. .1 Uniform Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Mixed Random Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 5. . . 56 6. . . . . . . . . . . . . . . . 65 6. 45 6. . . . . . . . .3 One Function of Two Random Variables . . . . . . . . . . . . . .v Contents 5. . . . . . . . . . . . . . . . . . .4. . . . . . .4. . . . . . .5. . . . 59 6. . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . .1 Poisson Approximation to Bernoulli . . . . . . . . . . . . . . . . . . . . 75 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6. . . . .2 CDF Technique with Arbitrary Functions .3. . . . . .2 Gaussian Approximation to Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 5. . . . . . . . . . . . . . . . . . . . . . . . . . .6 Bivariate Gaussian Random Variables . . . . . . . . . . . . . . . .1. . . . . . . . . . . 26 5. . . . .5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6. . . . . . . . . . . . . . . . 36 Transformations of Random Variables . . . . . . . . . . . . . .1 Bivariate CDF Technique . . . . . . . . . . . . . 20 5. . . . . . . . . . . 63 6.6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Exponential Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Standard Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Conditional PDF Technique . . . . . . . .18 5. . . . . . . . . . . . 12 5. . . . . . . 1 5. . . . . . . . . . . . . . . . . . .32 5. . . . .1 Interarrival Times. . . . . . . . . . . . . . .4 Bivariate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Problems . . . . . . . . . . . . 1 5. . . . . . . . . . . . . . . 46 6. . . 36 5. . . . . . . . . . . . . . . . . . . . . . . . .1 CDF Technique with Monotonic Functions . . . . . . . . . . . 25 5. . . . . . . . .7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Problems . . . 45 6. . . . . . .2 Bivariate PDF Technique . . . . . 11 5. . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Univariate Gaussian Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Univariate PDF Technique . . . . . . . . . . . . . . . . . .1 Marcum’s Q Function . . . . . . . . . . . . . .2. . . . . . . . . . . . .3 Bernoulli Trials . . .
.
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. The fourth chapter introduces jointly distributed random variables. conditional moments and the conditional characteristic function are also discussed.developing special mathematical skills as needed. as well as important approximations to the Bernoulli PMF and Gaussian CDF. The second chapter introduces the topic of random variables. The primary subjects of Chapter 6 are methods for determining the probability distribution of a function of a random variable. along with joint expectation. standard deviation. rather than more complex biological . the probability distribution for a single random variable is determined from a function of two random variables using the CDF. The mathematical background required of the reader is basic knowledge of differential calculus. The goal was to prepare students at the sophomore. Later chapters simply expand upon these key ideas and extend the range of application. These key concepts are all presented in the ﬁrst chapter. A considerable effort has been made to develop the theory in a logical manner .as well as pursue these topics at a more advanced level. joint moments. and the joint characteristic function. moments. In addition. Next. Our approach is to present a uniﬁed treatment of the subject. Poisson and Gaussian distributions are introduced. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF. The third chapter focuses on expectation. This short book focuses on standard probability distributions commonly encountered in biomedical engineering. This text is written as an introduction to probability theory. the exponential. and the characteristic function. conditional expectation. Then. 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. Here in Chapter 5. We have found it best to introduce this material using simple examples such as dice and cards. There are only a few key concepts involved in the basic theory of probability theory. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. Convolution is also developed.vii Preface This is the third in a series of short books on probability theory and random processes for biomedical engineers. The applications and examples given reﬂect the authors’ background in teaching probability theory and random processes for many years.
ranging from simple to difﬁcult. we do introduce some pertinent biomedical engineering examples throughout the text. straightforward exercises designed to reinforce concepts and develop problem solution skills. 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. Students in other ﬁelds should also ﬁnd the approach useful.viii PREFACE and biomedical phenomena. follow most sections. . We acknowledge and thank William Pruehsner for the technical illustrations. presented in the same general order as covered in the textbook. Many of the examples and end of chapter problems are based on examples from the textbook by Drake [9]. However. Drill problems.
.4) . 1.1) The mean and variance of a discrete uniform RV are easily computed with the aid of Lemma 2. we cannot possibly discuss them all here in this chapter. This chapter presents some of these discrete and continuous probability distributions that occur often enough in a variety of problems to deserve special mention.3) Simplifying. 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. the PMF or PDF is derived according to the characteristics of the experiment.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 = . the mean. In some instances. n 2 nn−1 2 2 (5. Additionally. special properties are pointed out along with relationships among other probability distributions. and characteristic function are found. Each section introduces a new PMF or PDF.1. (5.1 CHAPTER 5 Standard Probability Distributions A surprisingly small number of probability distributions describe many natural probabilistic phenomena. Because of the vast number of probability distributions. Following this. 5.3. 12 n − 1 (5. We will see that many random variables and their corresponding experiments have similar properties and can be described by the same probability distribution. . k = 0. α = kh + a. . 2 σx = (b − a)2 n + 1 . n − 1 otherwise. 2 (n − 1) n−1 4 (5.1 UNIFORM DISTRIBUTIONS Deﬁnition 5. . variance. 0.1.2) 1 n n−1 kh − k=0 b −a 2 2 = (b − a)2 n n−1 k=0 k 1 k2 − + .
1].6) Figure 5.1 illustrates onehalf period of φx (·). The characteristic function is plotted ∗ over [0.7) 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 = . 2(b − a) 2 (5. (5. a ≤α≤b otherwise. Deﬁnition 5. where the span h = 1/19. n 1 − e jht (5. Thus.1 illustrates the PMF and the magnitude of the characteristic function for a discrete RV which is uniformly distributed over 20 points on [0. Recall from Section 3.5 (a) 1 α 10 30 (b) 50 t FIGURE 5. b] if x has PDF f x (α) = 1/(b − a).1.5) Simplifying with the aid of Euler’s identity.8) .1: (a) PMF and (b) characteristic function magnitude for discrete RV with uniform distribution over 20 points on [0. 1].2.05 1 φ x (t ) 0 .3 that φx (−t) = φx (t) and that φx (t) is periodic in t with period 2π /h.2 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS px ( α ) . (5. The continuous RV x has a uniform distribution on the interval [a. Figure 5. 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 . π /h]. 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 . 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.1. 0.10) Figure 5.1. and 2 σx 1 = b −a b α− a b +a 2 2 dα = (b − a)2 . Answers: 20. 2 (c) E(x). 200. (b) P(10 ≤ x ≤ 50). −(b−a)/2 Simplifying with the aid of Euler’s identity.5 (a) 1 10 30 (b) 50 t FIGURE 5. Let x be a random variable equaling ten times the number tossed. 3.2 illustrates the PDF and the magnitude of the characteristic function for a continuous RV uniformly distributed on [0. (5.STANDARD PROBABILITY DISTRIBUTIONS f x (α) 1 1 3 φx (t ) 0 . 1/6. Deter2 mine: (a) Fx (0). Answers: 1. (c) ηx . (d) σx . (d) σx . 1]. Drill Problem 5. Determine: (a) p x (20). Note that the characteristic function in this case is not ∗ periodic but φx (−t) = φx (t).2.2: (a) PDF and (b) characteristic function magnitude for continuous RV with uniform distribution on [0. 1]. φx (t) = exp a +b j t 2 sin b−a t 2 b−a t 2 . A pentahedral die (with faces labeled 0. 2. Drill Problem 5.8. (b) Fx (5). .1.3. 12 (5. Random variable x is uniformly distributed on the interval [−1. 0. 5].1.2.2.4) is tossed once.
3 illustrates the PMF and the characteristic function magnitude for a discrete RV with geometric distribution and parameter p = 0. (5. p and 2 σx = q .18127.1.18 1 (5. .2.3: (a) PMF and (b) characteristic function magnitude for discrete RV with geometric distribution [ p = 0. p2 (5.13) Figure 5. .2. px (α) .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.4 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 5.2 EXPONENTIAL DISTRIBUTIONS Deﬁnition 5. . It can be shown that a discrete exponentially distributed RV has a memoryless property: p xx> (αx > ) = p x (α − ).14) Deﬁnition 5. 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 . α = 1.15) φx (t ) .2. 0. 2.11) 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 . where u(·) is the unit step function.18127]. ≥ 0. . otherwise. The continuous RV x has an exponential distribution with parameter λ(λ > 0) if x has PDF f x (α) = λe −λα u(α). The results are ηx = 1 .09 0 10 (a) 20 α 0 1 (b) 2 3 t FIGURE 5. 1 − q e jt (5. (5.
λ2 (5.2. arising in situations involving reliability theory and queuing problems. Reliability engineers determined its reliability at 5000 hours to be 95%. Determine the number of hours reliable at 99%. . Queuing theory describes the waiting times between events.1 0 5 10 (a) 15 α 10 30 (b) 50 t FIGURE 5. τ ≥ 0. which describes the time to failure for a system or component. Reliability theory.18) Example 5.17) A continuous exponentially distributed RV. satisﬁes a memoryless property: f xx>τ (αx > τ ) = f x (α − τ ). Suppose a system contains a component that has an exponential failure rate.16) Figure 5.4: (a) PDF and (b) characteristic function magnitude for continuous RV with exponential distribution and parameter λ = 0. (5. The exponential probability distribution is also a very important probability density function in biomedical engineering applications.1. λ − jt (5.2 1 5 φx (t ) .4 illustrates the PDF and the magnitude of the characteristic function for a continuous RV with exponential distribution and parameter λ = 0. like its discrete counterpart. grew primarily out of military applications and experiences with multicomponent systems. λ 2 σx = 1 . The characteristic function can be found as ∞ φx (t) = λ 0 e α( j t−λ) d α = λ .2. The results are ηx = 1 .2.STANDARD PROBABILITY DISTRIBUTIONS f x (α) . The mean and variance of a continuous exponentially distributed RV can be obtained using the moment generating property of the characteristic function.
Determine the average examination grade. 5. (c) 99%. First. 2. Determine the number of years reliable at: (a) 90%.95) = 1.02. Random variable x.99) = 980 hours.3 BERNOULLI TRIALS A Bernoulli experiment consists of a number of repeated (independent) trials with only two possible events for each trial. Answers: 0. Suppose a system has an exponential failure rate in years to failure with λ = 0. Answer: 80. 5000 Then. 5. ⎨ g (x) = 75 + 39. we solve for α from P (x > α) = e −λα = 0.3. such as a head and a tail in a coin toss. has PDF 4 4 f x (α) = e − 3 α u(α). λ Drill Problem 5.6. (b) 95%.1. The events for each trial can be thought of as any two events which partition the sample space.95 = P (x > 5000) = 5000 λe −λα d α = e −5000λ . 3 The examination results are given by ⎧ ⎪ 75.44(x − 4/3). a zero or one in a computer .99 or α= − ln(0.03 × 10−5 .2.5.2. to determine the number of hours reliable at 99%. the parameter λ is determined from ∞ 0. ⎪ ⎩ 0. 0 < x < 4/3 x ≥ 4/3 otherwise. Drill Problem 5. representing the length of time in hours to complete an examination in Introduction to Random Processes. Thus λ= − ln(0.6 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution.2.
.STANDARD PROBABILITY DISTRIBUTIONS 7 bit. . px (α) .19) φx (t) = k=0 n ( pe j t )k q n−k = (q + pe j t )n . p = 0. or an even and odd number in a die toss. The Bernoulli distribution is also known as a Binomial distribution.3.5: (a) PMF and (b) characteristic function magnitude for discrete RV with Bernoulli distribution.18 1 2 σx = npq . and n = 30.09 0 10 (a) 20 α 0 1 (b) 2 t FIGURE 5. The characteristic function can be found using the binomial theorem: n (5. The Bernoulli PMF describes the probability of k successes in n trials of a Bernoulli experiment. (5. k (5. . the other a failure. the Bernoulli distribution is used in infectious disease problems and other applications. Using the moment generating property of characteristic functions. p = 0. 1.2 and n = 30. Deﬁnition 5. . the mean and variance of a Bernoulli RV can be shown to be ηx = np. k = 0. n p x (k) = k ⎪ ⎩ 0.2. .20) Figure 5.5 illustrates the PMF and the characteristic function magnitude for a discrete RV with Bernoulli distribution. where p = probability of success and q = 1 − p. For biomedical engineers. A discrete RV x is Bernoulli distributed if the PMF for x is ⎧ ⎪ n ⎨ p k q n−k . 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.1. otherwise. Let us call one of the events a success.21) φx (t ) .
.1311 = 0.5. σx = np(1 − p) = 3.1. 0. (b) from 9 to 12 games. we consult Table A. 10.7421. 2 d) ηx = np = 10.1. k = 9.7031 − 0. k. n G(n.2784 = 0. Tables A. and 20. . using (5.7216. Example 5. n − k − 1. k. k.3 in the Appendix list values of the Bernoulli CDF for p = 0.22) with n = 15. b) P (9 ≤ x ≤ 12) = Fx (12) − Fx (8) = 0. since n n−m = n! = m! (n − m)! n m . .1–A. ﬁnd ηx and σx .8732 − 0. what is the probability that they win: (a) at least 10 games.0 − 0. (c) exactly 2 11 games? (d) With x denoting the number of games won. p) = m=n−k n n−m p n−m (1 − p)m . c) p x (11) = Fx (11) − Fx (10) = 0.22) This result is easily applied to obtain values of the Bernoulli CDF for values of p > 0. .4845 = 0. p) = m=0 n m n−k−1 p n−m (1 − p) − m m=0 n m p n−m (1 − p)m .05. 0. In a 15 game season. 0. .7.1–A. p) = =0 n p (1 − p)n− . Making the change of variable m = n − yields n G(n. Now. Solution. and p = 0. The probability that Fargo Polytechnic Institute wins a game is 0.15.3. (5.2186. Using the Binomial Theorem. 15. Let k ∈ {0. 1 − p). G(n. k.5 from Tables A.8 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Unlike the preceding distributions.2.5 and n = 5. .15. p) = 1 − G(n.7. we ﬁnd a) P (x ≥ 10) = 1 − Fx (9) = 1. n − 1} and deﬁne k G(n. a closed form expression for the Bernoulli CDF is not easily obtained. . With x a Bernoulli random variable.3. . . 1.
r + 1. . . we obtain Mnr (λ) = With s (x) = ∞ k=0 ∞ =r ( − 1)( − 2) · · · ( − r + 1) r p q (r − 1)! −r λ e . otherwise. Let RV nr represent the number of trials to obtain exactly r successes (r ≥ 1). k = 0. for = r. Note that P (success in th trial r − 1 successes in previous − 1 trials) = p. Note that with o r = 1 the negative binomial PMF is the geometric PMF. (r − 1)! m=0 xk = 1 . The moment generating function for nr can be expressed as Mnr (λ) = Letting m = − r .23) where p = p(1.26) where the parameter r is a positive integer. Discrete RV nr thus has PMF p nr ( ) = ⎧ ⎪ ⎨ ⎪ ⎩ −1 r −1 0. . 1. . − 1) p. r + 1. . . e λr p r ∞ (m + r − 1)(m + r − 2) · · · (m + 1)(q e λ )m . . also known as the P´ lya and the Pascal distribution. otherwise. Let p(k. we have P (nr = ) = p(r − 1. . n) = ⎧(k successes in n trials) P ⎪ n ⎨ p k q n−k . (5. pr q −r (5. hence. .25) . . . = r. 1) and q = 1 − p.24) (5. The PMF for the RV nr is called the negative binomial distribution. (5. 1−x x < 1. .STANDARD PROBABILITY DISTRIBUTIONS 9 We now consider the number of trials needed for k successes in a sequence of Bernoulli trials. n = k ⎪ ⎩ 0.
29) where RV x has a discrete geometric distribution. and p = 0. p2 (5. (5. φx (t ) 1 px (α) .03 0 20 (a) 40 α 0 1 (b) 2 t FIGURE 5.27) The mean and variance for nr are found to be ηn r = r . and p = 0. q e λ < 1.28) We note that the characteristic function is simply r φnr (t) = Mnr ( j t) = φx (t).18127. r = 3. Figure 5.6: (a) PMF and (b) magnitude characteristic function for discrete RV with negative binomial distribution. . p and 2 σn r = rq .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) .06 . r = 3.6 illustrates the PMF and the characteristic function magnitude for a discrete RV with negative binomial distribution. Mnr (λ) = pe λ 1 − qeλ r .18127. (5.
30) Proof. While there are no prescribed rules regarding the values of n and p for this approximation.1 Poisson Approximation to Bernoulli When n becomes large in the Bernoulli PMF in such a way that np = λ = constant. from which the desired result follows. k = 0. Tables A.1–A. the better the approximation. = lim ln 1 − 1 n λ n n→∞ = −λ lim 1− λ n n = e −λ .3.3. The motivation for using this approximation is that when n is large. so that p(k) = lim Now. n λ k k 1 p(k) = lim n→∞ k! Note that 1− λ n n−k k−1 (n − i). Substituting p = and q = 1 − λ . = k! ⎩ 0. . n→∞. λ lim ln 1 − n→∞ n so that n→∞ n λk n→∞ k! 1− λ n n .np=λ λ n (5. the Bernoulli PMF approaches another important PMF known as the Poisson PMF. Satisfactory results are obtained with np < 10. i=0 n→∞ lim n−k 1 − λ n −k k−1 i=0 (n − i) = 1. . 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. . . We have p(k) = lim n k p k q n−k ⎧ k −λ ⎨λ e .STANDARD PROBABILITY DISTRIBUTIONS 11 5. Lemma 5. 1. the larger the value of n and the smaller the value of p. The Poisson PMF is treated in the following section.1.3 are useless for ﬁnding values for the Bernoulli CDF. otherwise.
5 (the Poisson CDF table is covered in Section 4) that P (x ≤ 5) = 0. that is. Notice that in some circumstances np < 10 and npq > 3.12 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 5.3. Suppose x is a Bernoulli random variable with n = 5000 and p = 0. (5.2 Gaussian Approximation to Bernoulli Previously.2. we can still use this approximation by reversing our deﬁnition of success and failure in the Bernoulli experiment.2. We have np t φx φ y (t) = exp − j √ npq Substituting for φx (t). Our solution involves approximating the Bernoulli PMF with the Poisson PMF since n is quite large (and the Bernoulli CDF table is useless). . p is small and np < 10. .3. This approximation is quite useful since the Bernoulli table lists only CDF values for n up to 20. The accuracy of this approximation is best when n is large. Lemma 5. Incidentally. Then either the Poisson or the Gaussian approximation will yield good results. Solution.6160. φ y (t) = exp − j Simplifying. Since λ = np = 5.31) /2 . np t q q + p exp t j√ npq n .5) is also used to approximate a Bernoulli PMF under certain conditions.22). when n is large.001. npq where x is a Bernoulli RV. and p is very close to zero. the Poisson PMF was used to approximate a Bernoulli PMF under certain conditions. φ y (t) = q exp − j t p qn + p exp jt q np n t √ npq . Find P (x ≤ 5). The Gaussian PDF (see Section 5. if p is close to one. 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. 5. p is close to 1/2.3.32) Proof. we ﬁnd from Table A. and npq > 3. Let x − np y= √ .
Suppose x is a Bernoulli random variable with n = 5000 and p = 0.91774. 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. n lim ln φ y (t) = lim ln( pβ 2 e − j tα/β + pe j tβα ) .39. Find P (x ≤ 2048).35) γ (5. 2 n→∞ . With np = 2000.3. . Hence. where 1 F(γ ) = √ 2π is the standard Gaussian CDF. for large n and a < b P (a < x < b) = P (a < y < b ) ≈ F(b ) − F(a ). p and α= 1 . The limiting φ(t) in the above lemma is the characteristic function for a Gaussian RV with zero mean and unit variance.STANDARD PROBABILITY DISTRIBUTIONS 13 Letting β= we obtain n→∞ q . npq b − np b = √ . npq (5. Example 5. a − np a = √ .5. 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 = =− . npq = 1200 and b = (2048 − 2000)/34.33) e −α −∞ 2 /2 d α = 1 − Q(γ ) (5.4. α→0 α2 Applying L’Hˆ spital’s Rule twice. Evaluation of the above integral as well as the Gaussian PDF are treated in Section 5.9 of the Appendix.34) and Q(·) is Marcum’s Q function which is tabulated in Tables A. Solution.641 = 1.39) = 1 − Q(1. The solution involves approximating the Bernoulli CDF with the Gaussian CDF since npq = 1200 > 3.8 and A.3.8 that P (x ≤ 2048) ≈ F(1.39) = 0.
5 − 10)/ 5 = −1.3.3.7625. 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.67) − F(−1. Answers: 2. p x (8) ≈ 0. p x (8) = 0. Example 5. Operationally. (c) from two to ﬁve of the next twenty cars purchased will be white. typically a time interval.1789. North Dakota revealed that 30% preferred a white automobile over all other colors. A survey of residents in Fargo. p x (8).12007. a = (7. or the number of errors per page in this textbook.5. or within a given region. Since npq = 5 > 3.12) = 0.3. Solution. (b) at least ﬁve of the next twenty cars purchased will be white. which is very close to the above approximation. 0. Suppose x is a Bernoulli random variable with n = 20 and p = 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.13136. we have p x (8) = P (7.5 < x < 8. In this section. a continuous distribution is used to calculate probabilities for a discrete RV. (c) the number of times she must ﬁre so that the probability of hitting the target at least once is greater than 1/2. 5. 4. .2. Drill Problem 5.5) ≈ F(−0. we derive the Poisson probability distribution from two fundamental assumptions about the phenomenon based on physical characteristics. 1.14 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS When approximating the Bernoulli CDF with the Gaussian CDF.25143 − 0. Drill Problem 5.3. The probability she will hit the target is only 0.7748. Rensselaer is an avid albeit inaccurate marksman. Determine the probability that: (a) exactly ﬁve of the next 20 cars purchased will be white.1201.67.5 − 10)/ √ 5 = −0. 0. Determine: (a) the expected number of hits scored in 15 shots.4088. For example. npq = 5. Answers: 0. In the previous section. (b) the standard deviation for 15 shots.12. the Gaussian approximation is used to evaluate the Bernoulli √ PMF. hence. Prof.4. From the Bernoulli table.1.5. a Poisson random variable might represent the number of telephone calls per hour. and b = (8. Find P (x = 8).4 POISSON DISTRIBUTION A Poisson PMF describes the number of successes occurring on a continuous line. With np = 10. we found that the limit (as n → ∞ and constant mean np) of a Bernoulli PMF is a Poisson PMF.
42) (5. (5. g (h) = 15h 2 + 7h 3 = o (h). Since [0. Thus.41) This second property indicates that for a series of very small intervals. if I1 ∩ I2 = . τ ] ∩ (τ. τ ) = P (k successes in interval [0. and B = { successes in interval I2 }. We use the notation p(k. h→0 h lim For example.39) (5. τ + h] = .37) (5. τ ] ∪ (τ. τ + h) = p(0. The probability of more than one success occurring during an interval vanishes as the length of the interval approaches zero. Hence p(1. and p(0. with A = {k successes in interval I1 }. we have p(0. the greater the probability of success. τ ]).40) (5. h) = 1 − λh + o (h). 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. τ + h] and [0. τ ) p(0. τ )(1 − λh + o (h)). (2) The probability of a single success during a very small time interval is proportional to the length of the interval. τ + h] = [0. the Poisson process is composed of a series of Bernoulli trials. each with a probability of success p = λh + o (h). we have P (A ∩ B) = P (A)P (B). h) = λh + o (h). (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.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. The longer the interval. . (5.38) As we will see.
τ ) −λh p(0. 2. d p(0. (5. . h h Taking the limit as h → 0 d p(k.44) k = 1.48) . .16 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Noting that p(0.47) The RV x = number of successes thus has a Poisson distribution with parameter λτ and PMF p x (k) = p(k. τ ) = λp(k − 1. t)d t (5. τ ) p(0. . τ ) + λp(k. τ + h) = p(k − 1. τ ) + o (h) = h h and taking the limit as h → 0. τ ). τ ) = λe and hence that p(k. τ ) = −λp(0. τ ) = −λτ 0 e λt p(k − 1. . τ + h) − p(0. h) + p(k. k! (5. τ + h) − p(k. τ ) p(1. τ ) = λp(k − 1. (5. τ ) + . τ ) = e −λτ u(τ ). k! k = 0. . dτ This differential equation has solution p(0.45) p(k. For ease in subsequent development. The characteristic function for a Poisson RV x with parameter λ is found as (with parameter λ. (5. 1)) φx (t) = e −λ ∞ k=0 (λe j t )k = e −λ exp(λe j t ) = exp(λ(e j t − 1)). so that o (h) p(k. 0) = 0. . . h) + o (h). τ ) + λp(k.43) (5. 0) = 1. 1. τ + h) = p(k − 1.46) (λτ )k e −λτ u(τ ). p x (k) = p(k. τ )(1 − λh) + o (h). dτ with p(k. . τ ). τ ). The rate of the Poisson process is λ and the interval length is τ . τ )λh + p(k. or p(k. Applying the above properties. It can be shown ([7. we replace the parameter λτ with λ. 8]) that τ p(0. it is readily seen that p(k.
(5.7: (a) PMF and (b) magnitude characteristic function for Poisson distributed RV with parameter λ = 10. (5. Consider the function ψx (γ ) = E(γ x ) and note that k−1 (k) ψx (γ ) = E γ x−k i=0 (5.STANDARD PROBABILITY DISTRIBUTIONS px (α) .7 illustrates the PMF and characteristic function magnitude for a discrete RV with Poisson distribution and parameter λ = 10.12 1 φx (t ) 17 .52) ..e. then φx1 +x2 (t) = exp((λ1 + λ2 )(e j t − 1)).49) i. The moments of a Poisson RV are tedious to compute using techniques we have seen so far. It is of interest to note that if x1 and x2 are independent Poisson RVs with parameters λ1 and λ2 . Figure 5. so that k−1 E i=0 (k) (x − i) = ψx (1).51) If x is Poisson distributed with parameter λ. then ψx (γ ) = e λ(γ −1) . respectively.06 0 10 (a) 20 α 0 1 (b) 2 t FIGURE 5. x1 + x2 is also a Poisson with parameter λ1 + λ2 . (5.50) (x − i) . so that (k) ψx (γ ) = λk e λ(γ −1) .
Let RV tr denote the length of the time interval from zero to the r th success. of a Poisson random variable is more important than the actual number of successes. and (c) at least 15 problems are graded? Solution. k−1 E i=0 (x − i) = λk . hand computation of the CDF is quite tedious.5830 − 0.1. h) = p(r − 1. the length of time between successes. From the Poisson CDF table. Therefore. Professor Rensselaer grades 10 problems per day. Additionally. Example 5. known as an interarrival time.2202 = 0.3328 − 0.1126. b) P (8 ≤ x ≤ 10) = Fx (10) − Fx (7) = 0. 5. (5. and ﬁnd a) p x (8) = Fx (8) − Fx (7) = 0. Then p(τ − h < tr ≤ τ ) = p(r − 1. Indeed. the subject of reliability theory is so important that entire textbooks are devoted to the topic. so that σx = λ2 + λ − λ2 = λ.7 of the Appendix for selected values of λ ranging from 0. in evaluating the reliability of a medical device.4. While it is quite easy to calculate the value of the Poisson PMF for a particular number of successes. however. τ − h) p(1. With x a Poisson random variable. E(x(x − 1)) = λ2 = E(x 2 ) − λ. c) P (x ≥ 15) = 1 − Fx (14) = 1 − 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. (b) 8–10 problems are graded. we note that the value of the Poisson PMF increases as the number of successes k increases from zero to the mean. τ − h)λh + o (h) . even though an inﬁnite number of Poisson successes are theoretically possible. we consult the Poisson CDF table with λ = 10.9165 = 0.4.1 Interarrival Times In many instances. the Poisson CDF is tabulated in Tables A.2202 = 0. What is the probability that on a given day (a) 8 problems are graded. On the average.1 to 18. E(x) = λ. the time to failure is far more signiﬁcant to the biomedical engineer than the fact that the device failed.4A.53) 2 In particular.3628. we will brieﬂy examine the subject of interarrival times from the basis of the Poisson PMF. Here.18 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS hence. and then decreases in value as k increases from the mean. For example.0835.
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 ),
The PDF approaches zero asymptotically as α approaches ±∞. centered now at 25 instead of 75. which necessitates numerical integration.08 . the more spread in the distribution and the smaller the maximum value of the PDF. If. illustrated with η = 75 and σ 2 = 25. Rather than attempting to tabulate the general Gaussian CDF.04 65 75 85 α FIGURE 5. σ 2 ) then (x = σ z + η) φx (t) = e j ηt e − 2 σ t .66) Similarly. Additionally. We will have occasion to use the shorthand notation x ∼ G(η. a closed form expression does not exist for the Gaussian CDF. is a bellshaped curve completely determined by its mean and variance. with σ > 0 and x = −σ z + η we ﬁnd Fx (α) = P (−σ z + η ≤ α) = 1 − Fz((η − α)/σ ). Naturally. so that f x is as above.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 . as well as with η = 75 and σ 2 = 9 in Figure 5. the Gaussian PDF curve must be symmetrical as previously described. occurs at α = η.9. the maximum value of the Gaussian PDF.12 . η = 25. 1 2 2 (5. . a normalization is performed to obtain a standardized Gaussian RV (with zero mean f x (α) . the Gaussian PDF is symmetrical about the vertical axis through the expected value. 2σ 2 (5. the larger the value of the variance.67) The Gaussian PDF. σ 2 ) to denote that the RV has a Gaussian PDF with mean η and variance σ 2 . (2πσ 2 )−1/2 . As can be seen. For any combination of the mean and variance. in fact. Unfortunately. 25. identically shaped curves could be drawn. and the area under the curve must equal one.9: Gaussian probability density function for η = 75 and σ 2 = 9. Note that if x ∼ G(η.
5.72) e −t d t 2 (5. 1).8 and A. where z ∼ G(0. 1 2 (5.9 for 0 ≤ α < 4 using the approximation presented in Section 5.70) −∞ e − 2 τ d τ. Compute Fz(−1.1. σ 2 ).71) α Marcum’s Q function is tabulated in Tables A.75) Example 5.74). It is easy to show that Q(−α) = 1 − Q(α) = Fz(α). This transformation is always applied when using standard tables for computing probabilities for Gaussian RVs. A simple change of variable reveals that √ erfc(α) = 2Q(α/ 2).73) 0 e −t d t = 1 − erf(α) 2 (5. The probability P (α1 < x ≤ α2 ) can be obtained as P (α1 < x ≤ α2 ) = Fx (α2 ) − Fx (α1 ).68) (5. The error and complementary error functions.74) α are also often used to evaluate the standard normal integral.1. Note that 1 Fz(α) = √ 2π where Q(·) is Marcum’s Q function: 1 Q(α) = √ 2π ∞ α (5. using the fact that Fx (α) = Fz((α − η)/σ ).5. 1). . (5.STANDARD PROBABILITY DISTRIBUTIONS 23 and unit variance). 1 2 (5. deﬁned by 2 erf(α) = π and 2 erfc(α) = π ∞ α (5. If x ∼ G(η.69) e − 2 τ d τ = 1 − Q(α). the RV z = (x − η)/σ is a standardized Gaussian RV: z ∼ G(0.
This notion does not imply that random variable x cannot take on a value outside this interval.03 .2. we ﬁnd Fz(−1.6) = Fz(1.0027). Sketch the PDF and then ﬁnd P (37 ≤ x ≤ 51).2.24 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution.04093. Example 5.74).10 along with the indicated probability.6) − Fz(0. using (5. Solution. To compute Fz(−1. but the probability of it occurring is really very small (2Q(3) = 0. Suppose x is a Gaussian random variable with η = 35 and σ = 10.5. Hence P (37 ≤ x ≤ 51) = Q(0. From the transformation z = (x − η)/σ . η + 3σ ] = [5.01 0 30 60 α FIGURE 5. x − 35 10 f x (α) . The sketch of this PDF is shown in Figure 5.74) = Q(1. From Table A. The PDF is essentially zero outside the interval [η − 3σ. .10: PDF for Example 5. the range of values random variable x takes on is then approximately η ± 3σ .36594 from Table A.6) = 0. 65]. the realistic range of values is much smaller.5. With z= we have P (37 ≤ x ≤ 51) = P (0.72) and Table A.9.0.74) = 0. we note that 99.02 . Indicate this probability on the sketch.8.2). While the value a Gaussian random variable takes on is any real number between negative inﬁnity and positive inﬁnity.73% of the area under the curve is contained between −3.04 .0 and 3.9.2 ≤ z ≤ 1.2) − Q(1.74) = 1 − Q(−1.
31938153 −0. 932]: Q(γ ) ≈ e − 2 γ h(t).e. 1).STANDARD PROBABILITY DISTRIBUTIONS 25 Example 5. Q(γ ) is the complement of the standard Gaussian CDF.77) i. This result is determined by linear interpolation of the CDF between equal 1.80) 1 .08 and 1.2316419γ (5.5 ≤ x) = P (1. (5. p.5 × 10−8 . Q(∞) = 0. 5.356563782 1. Note that Q(0) = 0. Since we are measuring to the nearest μF.09. 2π The constants are i 1 2 3 4 5 ai 0. Assuming that capacitance follows a Gaussian distribution. Solution. If the RV z ∼ G(0.5.1 Marcum’s Q Function ∞ Marcum’s Q function.5.330274429 (5.5.13941. A very accurate approximation to Q(γ ) is presented in [1.79) The error in using this approximation is less than 7.3.76) γ has been extensively studied. where z = (x − 25)/6 ∼ G(0. deﬁned by 1 Q(γ ) = √ 2π e−2α dα 1 2 (5. . 1 2 γ > 0. 1) then Q(γ ) = 1 − Fz(γ ).78) where t= and 1 h(t) = √ t(a 1 + t(a 2 + t(a 3 + t(a 4 + a 5 t)))).083 ≤ z) = Q(1. ﬁnd the probability that the value of capacitance exceeds 31 μF if capacitance is measured to the nearest μF. the probability that the measured value exceeds 31 μF is P (31.781477937 −1. and that Fz(−γ ) = Q(γ ). Let the RV x denote the value of a capacitor.. A machine makes capacitors with a mean value of 25 μF and a standard deviation of 6 μF.821255978 1. 1 + 0. (5.083) = 0.
The bound improves as α increases.02275.721). and signiﬁcantly expanding the scope of applications.5. Drill Problem 5. 1. The quality point averages of 2500 freshmen at Fargo Polytechnic Institute follow a Gaussian distribution with a mean of 2. (d )Fx (4.2316419γi (5.5 and a standard deviation of 0. 0.7. 1179.0. 0.3. Students attend Fargo Polytechnic Institute for an average of four years with a standard deviation of onehalf year.6 BIVARIATE GAUSSIAN RANDOM VARIABLES The previous section introduced the univariate Gaussian PDF along with some general characteristics. Answers: 167. Then compute ti = and γi+1 h(ti ) = 2 ln q 1/2 1 1 + 0. (c) between 3. . 0. √ π α + α 2 + 0.5.0 and 3. Answers: 0. (b) less than 2. covers 95% of the shots. Determine: (a) P (1 < x < 3).82) .5. 150 ± d . Helstrom [14] offers an iterative procedure which begins with an initial guess γ0 > 0.5. (c )P (x = 4). Determine the number of freshmen you would expect to score: (a) from 2. it is desired to ﬁnd the value of γ for which Q(γ ) = q .83) The procedure is terminated when γi+1 ≈ γi to the desired degree of accuracy. Suppose grade point averages are computed to the nearest tenth. 298] e−2α 2 < Q(α) ≤ √ π α + α2 + 4 1 2 e−2α 2 . Answer: 33.2.6 to 3.81) The ratio of the upper bound to the lower bound is 0. 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. (b)P (x > 4).5. . p.5π 1 2 (5. .92535. Drill Problem 5. (5. we discuss the joint Gaussian PDF and its characteristics by drawing on our univariate Gaussian PDF experiences. Drill Problem 5. . i = 0. Determine the value of d so that the range.5. (d) greater than 3. 5.967 when α = 4. Professor Rensselaer loves the game of golf.26 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS A very useful bound for Q(α) is [1. Now. .1. 639.946 when α = 3 and 0. Sometimes. Let the random variable x denote the length of attendance and assume that x is Gaussian.32.5. 322.
z2 (α. t2 ) = E(e j t1 z1 + j t2 z2 ) = e − 2 (1−ρ 1 2 2 )t2 − 1 (t1 +ρt2 )2 2 e . −∞ (5. the RV t1 z1 + t2 z2 is univariate Gaussian: 2 2 t1 z1 + t2 z2 ∼ G 0.84) so that z1 ∼ G(0. Also.. y) is a bivariate Gaussian RV if every linear combination of x and y has a univariate Gaussian distribution. β) = 2π ∞ I (α. f yx and f x.1. and σ y = 0. t2 )e − jβt2 d t2 . Below. The following development is signiﬁcantly simpliﬁed by considering the standardized versions of x and y.z2 . so that φz1 .y  < 1. the results for z1 and z2 are applied to obtain corresponding quantities φx.y = ±1.z2 (t1 . Completing the square. (5. the joint Gaussian PDF is considered the most important joint distribution for biomedical engineers. 2 2 2 t1 + 2t1 t2 ρ + t2 = (t1 + ρt2 )2 + (1 − ρ 2 )t2 .85) From (6) we have 1 f z1 . the marginal PDFs for a joint Gaussian PDF are univariate Gaussian.y . In this case we also say that the RVs x and y are jointly distributed Gaussian RVs. which we now proceed to do. t1 + 2t1 t2 ρ + t2 . σx ) and 2 y ∼ G(η y . we assume that ρx. 1) and z2 ∼ G(0. In particular. like the univariate case.e. x ∼ G(ηx . then the conditional PDF f z2 z1 and the joint PDF f z1 . σx = 0. 1). Since z1 and z2 are jointly Gaussian. Let z1 = x − ηx σx and z2 = y − ηy . σx = 0.86) .STANDARD PROBABILITY DISTRIBUTIONS 27 Numerous applications of this joint PDF are found throughout the ﬁeld of biomedical engineering and. Finally. we ﬁrst ﬁnd the joint characteristic function for the standardized RVs z1 and z2 . σy (5. Next.6. Then 2 w is a univariate Gaussian RV for all real constants a and b.y . σ y ). i. and let x and y be jointly distributed Gaussian RVs. Let the RV w = ax + by. the special cases ρx. Deﬁnition 5. and σ y = 0 are discussed. The above deﬁnition of a bivariate Gaussian RV is sufﬁcient for determining the bivariate PDF. The bivariate RV z = (x.
t2 )e − j αt1 d t1 .z2 (t1 .87) (5. t2 ) = 2π ∞ φz1 . 2(1 − ρ 2 ) (5. 1 2 −∞ .z2 (α. Substituting into (5. −∞ Substituting (5.z2 (α.86) we ﬁnd 1 f z1 . where φ(t2 ) = e j αρt2 e − 2 (1−ρ 1 2 2 )t2 2 − 1 (1−ρ 2 )t2 2 1 2π ∞ e − 2 τ e − j α(τ −ρt2 ) d τ. 1 2π (1 − ρ 2 ) exp − (β − ρα)2 .90) We now turn our attention to using the above results for z1 and z2 to obtain similar results for x and y.85) and letting τ = t1 + t2 ρ. t2 ) = e or I (α. β) = f z2 z1 (βα) = f z1 (α) so that E(z2  z1 ) = ρz1 and 2 σz2 z1 = 1 − ρ 2 .z2 (α. t2 ) = φ(t2 ) f z1 (α). .84) we ﬁnd that x = σx z1 + ηx and y = σ y z2 + η y .88) (5. we obtain I (α.89) After some algebra. 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 . 2π (1 − ρ 2 )1/2 2(1 − ρ 2 ) (5. β) = 1 α 2 − 2ραβ + β 2 exp − .28 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where 1 I (α. From (5. Thus f z1 . we ﬁnd f z1 .
y (t1 .93) Substituting (5. t2 ) = E(e j t1 x+ j t2 y ) = E(e j t1 σx z1 + j t2 σ y z2 )e j t1 ηx + j t2 η y .93) to obtain f yx (βα) = 1 f z z σy 2 1 β − η y α − ηx σy σx .95) 2 2πσ y (1 − ρ 2 ) . (5.z2 (σx t1 . the joint characteristic function for x and y can be found from the joint characteristic function of z1 and z2 as φx. (5.y (t1 .92) Making the change of variables τ1 = σx t1 .91) Using (4.90) and (5. the joint characteristic function φx. t2 ) = φz1 . Consequently. β) = f yx (βα) f x (α) and f x (α) = we may apply (5. 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 (5.94) we ﬁnd f x. σ y t2 )e − j (α−ηx )t1 e − j (β−η y )t2 d t1 d t2 . (5. we obtain f x.z2 (σx t1 .94) 1 fz σx 1 α − ηx σx .y (α.96) .y (α.y (α.y (α. τ2 = σ y t2 .z σx σ y 1 2 α − ηx β − η y .93) and (5. (5. β) as 1 f x.y can be transformed to obtain the joint PDF f x.87) into (5.y (α.STANDARD PROBABILITY DISTRIBUTIONS 29 so that the joint characteristic function for x and y is φx.66). β) = Since f x. σx σy . 1 f z . −∞ −∞ (5. σ y t2 )e j ηx t1 e j η y t2 . β) = (2π)2 ∞ ∞ φz1 .
11.265 β =3 α = −3 β = −3 α =3 FIGURE 5. x − ηx σx (5.8.99) By interchanging x with y and α with β.98) ⎞ ⎠ .100) 2 2 σxy = σx (1 − ρ 2 ). (5. β) with σx = σ y = 1.y + t2 σ y . (5. and ρ = −0. Since x and y are jointly Gaussian. ⎛ 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 .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 ). σy (5.11: Bivariate Gaussian PDF f x. t1 σx + 2t1 t2 σx.97) (5.101) A threedimensional plot of a bivariate Gaussian PDF is shown in Figure 5. The bivariate characteristic function for x and y is easily obtained as follows. the RV t1 x + t2 y is a univariate Gaussian RV: 2 2 2 2 t1 x + t2 y ∼ G t1 ηx + t2 η y .y (α. . β ) .y (α. fx . ηx = η y = 0.
104) for ρ = ±1.y = 0. 2 2 Example 5.y (α. This is an application of the Chebyshev Inequality. σx = σ y = 1.105) when ρ = ±1.6. Then f yx (βα) = f y (β). We showed previously that if x and y are independent. β) = f y (β)δ α − ηx ± σx β − ηy σy (5. and ρ = ±1.y . then ρx. Proof.95)) f x. If ρ = 0 then (from (5. Suppose that ρ = ρx. A very special property of jointly Gaussian RVs is presented in the following theorem.y (t1 . The jointly Gaussian RVs x and y are independent iff ρx.. (5.1. β) = f x (α) f y (β).y = 0.102) which is valid for all σx.y (α. the joint characteristic function for x and y is φx. Let x and y be jointly Gaussian with zero means. 1) and such that v and x are independent. the probability that the RV deviates from its mean by more than an arbitrarily small ﬁxed amount approaches zero.y (α. Hence.STANDARD PROBABILITY DISTRIBUTIONS 31 Consequently. from (5.97) and (5. Find constants a and b such that v = ax + by ∼ G(0.103) x − ηx σx y → ηy ± σy in probability1 . These results can also be obtained “directly” from the joint characteristic function for x and y. t2 ) = e − 2 (t1 σx +2t1 t2 σx.1. 1 As the variance of a RV decreases to zero. Interchanging the roles of x and y we ﬁnd that the joint PDF for x and y may also be written as f x. As ρ → ±1. 1 2 2 2 2 (5.y +t2 σ y ) e j t1 ηx + j t2 η y .e. RVs x and y are independent. We conclude that x − ηx σx α − ηx σx f x. Theorem 5. β) = f x (α)δ β − η y ± σ y (5. We now consider some special cases of the bivariate Gaussian PDF. σx and σ y . i.98) we ﬁnd E(yx) → η y ± σ y 2 and σ yx → 0.y = 0. .6.
110) .6. (5. 1 − 2ρq + q 2 (5.y and b 2 = 1/(1 − ρx. y − ρx.y = 1 and E(vx) = a + bρx. β) by ﬁnding the locus of points where the PDF is constant. We require 2 2 σv = a 2 + b 2 + 2abρx. β) on the contour to the origin is given by d 2 (q ) = α 2 + β 2 = c 2 (1 + q 2 ) . We assume that ρ < 1. 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 .y ).106) where c is a positive constant.32 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. we now investigate the shape of the joint PDF f z1 . so that v= 2 is independent of x and σv = 1. We have E(v) = 0.90).109) ±c 1 − 2ρq + q 2 . 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 .y = 0. (5. β) is constant for α and β satisfying α 2 − 2ραβ + β 2 = c 2 .y x 2 1 − ρx. 2 Hence a = −bρx.z2 (α.107) The square of the distance from a point (α. By inspection of (5.1 Constant Contours Returning to the normalized jointly Gaussian RVs z1 and z2 .z2 (α.108) (5.y 5. (5.
118) 2 − 2ρ α − ηx σx β − ηy σy + β − ηy σy 2 = c 2. the constant contours for f x. the line β = α intersects the constant contour at ±c ±β = α = . Thus. From (5. (5. so that α +β =β √ 2 and α −β = α.112) √ √ Consider the rotated coordinates α = (α + β)/ 2 and β = (β − α)/ 2.111) 2(1 − ρ) Similarly. (5. the line β = −α. √ 2 The rotated coordinate system is a rotation by π/4 counterclockwise. With α = 0 we ﬁnd that one axis is along β − ηy α − ηx =− σx σy . the major and minor axes of the ellipse are along the lines β = ±α.STANDARD PROBABILITY DISTRIBUTIONS 33 Differentiating. (5. β) are solutions to α − ηx σx Using the transformation 1 α =√ 2 β − ηy α − ηx + σx σy .y (α. Thus α 2 − 2ραβ + β 2 = c 2 is transformed into β2 c2 α2 + = .93). intersects the constant contour at ±β = −α = ±c 2(1 − ρ) .114) (5. 1 β =√ 2 β − ηy α − ηx − σy σx (5.116).116) (5. In the α − β plane.117) transforms the constant contour to (5.113) √ The above equation represents an ellipse with major axis length 2c / 1 − ρ and minor axis √ length 2c / 1 + ρ.115) (5. we ﬁnd that d 2 (q ) attains its extremal values at q = ±1. 1+ρ 1−ρ 1 − ρ2 (5.
117). 2(1 − ρ 2 ) (5. the value of the joint PDF f x. α =√ 1+ρ β =√ β 1−ρ exp − c2 . 2 1−ρ the length of this axis in the α − β plane is √ 2c 2 2 σx + σ y ±c σ y β=√ √ + ηy .34 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS with endpoints at ±c σx α=√ √ + ηx . Points on this ellipse in the α − β plane satisfy (5.y on this curve is 1 2πσx σ y 1 − ρ 2 A further transformation α .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 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 1−ρ .117). 2 1+ρ 1+ρ . With β = 0 we ﬁnd that the other axis is along β − ηy α − ηx = σx σy with endpoints at ±c σx α=√ √ + ηx . Letting A denote the region bounded by the ellipse in the α − β plane and A denote the image (a circle) . 1 − ρ2 This transformation provides a straightforward way to compute the probability that the joint Gaussian RVs x and y lie within the region bounded by the ellipse speciﬁed by (5.
Substituting and transforming to polar coordinates. 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 ) . β) 1 2 2 where the Jacobian of the transformation is ∂α ∂α J (α. 1. 1 √ √ σy 2 1 + ρ . we have 1 √ √ σx 2 1 + ρ J (α.1. (b) η y .y (α. 1 √ √ σy 2 1 − ρ ∂α ∂β ∂β ∂β .STANDARD PROBABILITY DISTRIBUTIONS 35 in the α − β plane.y . .75.5x. and (e) ρx. β) = −1 √ √ σx 2 1 − ρ so that J (α. This bivariate Gaussian probability computation is one of the few which is “easily” accomplished. 1 − ρ 2 J (α. (c) σx . E(xy) = 7/6 + y/6. β) = ∂β ∂α Computing the indicated derivatives. Additional techniques for treating these computations are given in [1. β) = 1 σx σ y 1 − ρ 2 . 956–958]. 5. Drill Problem 5. and σxy = 0. Given that x and y are jointly distributed Gaussian random variables with 2 2 E(yx) = 2 + 1.y (α. (d) 2 σ y . β) d α dβ = A 1 2πσx σ y e − 2 (α +β ) d α dβ . 9. we have f x. Determine: (a) ηx . pp.5.6. we ﬁnd f x. 2. Answers: 0.
3394. and ρ = −0.8 PROBLEMS 1. For the Gaussian approximation use η = np and σ 2 = npq .3.6. Unfortunately.3.6. 2 2 σx = 21. (c) n = 20. Poisson and Gaussian distributions. 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.7.3889. (c )P (−1 < x < 7). Random variables x and y are jointly Gaussian with ηx = −2. (b) the probability that. and ρ = −0. p = 0. Assume x is a Bernoulli random variable. σ y = 31. p = 0.5212. (h) n = 10. η y = 3.3394. 5. p = 0. 2.7 SUMMARY This chapter introduces certain probability distributions commonly encountered in biomedical engineering. Drill Problem 5. For the Poisson approximation use η = np. Determine P (x ≤ 3) using the Bernoulli CDF table if: (a) n = 5.2212 in (154). Many important properties of jointly Gaussian random variables are presented. p = 0. p = 0. (d) n = 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.6. he has played at least one game for more than four minutes. (f ) n = 20. (e) n = 10.36 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Drill Problem 5. p = 0.3. 2.1753. Drill Problem 5. (g) n = 5.3.1. Special emphasis is placed on the exponential. (i) n = 20. your friend gives you a dollar and if the die comes up with an odd number you give your friend a dollar. out of ﬁve games. Determine: (a) the probability that William is still playing after four minutes.6. 30◦ . Find: (a) E(y  x = 0).899. the die is loaded . Important approximations to the Bernoulli PMF and Gaussian CDF are developed. Answers: exp(−1). 5. (b) n = 10. p = 0. 2.1. (c) the length of the major axis. Answers: 0. Answers: 3. Suppose you are playing a game with a friend in which you roll a die 10 times. Bernoulli event probabilities may be approximated by the Poisson PMF when np < 10 or by the Gaussian PDF when npq > 3. σ y = 31. 2 2 σx = 21. 0. p = 0.6.1. If the die comes up with an even number. 0. (b) the length of the minor axis. η y = 3. Random variables x and y are jointly Gaussian with ηx = −2. His play during each game is independent from all other games.1. (b)P (1 < y ≤ 10  x = 0).2. With c 2 = 0.
2. (c) from 3 to 7 of the next 20 rolls are strikes. (c) from 7 to 9 baskets. Each student is given only one transistor to use. (b) the probability of your friend winning more than 4 dollars. The probability of a man hitting a target is 0.6? . A dart player has found that the probability of hitting the dart board in any one throw is 0. (a) If we take a sample of 20 cars. Determine: (a) how many dollars your friend can expect to win in this game. If there are 20. a 4. North Dakota. Determine the probability that: (a) 3 of the next 20 rolls are strikes. In Junior Bioinstrumentation Lab.150. (c) the number starting is between 11.3.2. the probability that a car parked outside will start is 0. what is the probability that exactly 12 cars will start and 8 will not? (b) What is the probability that the number of cars starting out of 20 is between 9 and 15. one experiment introduces students to the transistor. (a) Sketch the PMF for x. (e) the expected number of baskets. a 5 or a 6. (b) 4 baskets. what is the probability he will make: (a) at least 4 baskets. 9. Consider Problem 7. Determine: (b) the expected number of destroyed transistors.000 cars to be started.500. 3.4. The probability of a student destroying a transistor is 0. (d) less than 2 baskets. How many times must he throw the dart so that the probability of hitting the dart board is at least 0. On a frosty January morning in Fargo.7. Determine the probability it will take more than 5 rolls. 7. ﬁnd the probability that: (a) at least 12.100 will start. (a) If he tries 15 times to hit the target. 8. (d) the number starting is less than 12. Let random variable x show the possible numbers of students who destroy transistors. One lab class has 5 students and they will perform this experiment next week. Determine the: (e) expected number of strikes in 20 rolls. 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. (b) exactly 12.6.000 will start. (g) standard deviation for the number of strikes in 20 rolls. 5. (c) the probability that fewer than 2 transistors are destroyed. (b) at least 4 of the next 20 rolls are strikes. (f ) variance for the number of strikes in 20 rolls. The probability that a basketball player makes a basket is 0. 4. If he makes 10 attempts. The probability that Professor Rensselaer bowls a strike is 0.STANDARD PROBABILITY DISTRIBUTIONS 37 so that a 1 or a 3 are three times as likely to occur as a 2.900 and 12. (d) She is to keep rolling the ball until she gets a strike.
An electronics manufacturer is evaluating its quality control program. You give him another one page letter to type. To what value should the mean be adjusted to ensure this? . (c) With the 5 mistakes per page average. (b) exactly 11 quizzes are given during the quarter. Suppose the average number of times a student takes Introduction to Random Processes is 1. Suppose x is a Bernoulli random variable with η = 10 and σ 2 = 10/3. (a) If you give him a one page letter to type. What is the probability of at least 6 alpha particles being emitted in 1 min? 19. 14. and. a sample of radioactive material emits 20 alpha particles per minute. a student may take a course as many times as desired. the material averages 6 alpha particles emitted per minute. (d) at most 9 quizzes are given during the quarter. (b) 10 min? (c) Many years later. Determine the probability that the student makes: (a) at least 1 mistake. (b) from 3 to 5 mistakes. Repeat Problem 1. using the Poisson approximation to the Bernoulli PMF. a student makes 6 mistakes per test.4. (c) p. thinks so many students repeat the course because they enjoy it so much. What is the probability of him making fewer than 5 mistakes. What proportion of 20% defective components will be shipped? 13. after the lessons.5. Professor Rensselaer gives 11 quizzes per quarter in Introduction to Random Processes. 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 is the probability that more than 2 accidents will occur during any given week? 15. (c) exactly 2 mistakes. he averages 5 mistakes per page. 12. 17. (b) The academic vicepresident of FPI wants to ensure that on the average. 11. 80% of the students take the course at most one time. (d) more than the expected number of mistakes.) (a) Determine the probability that a student takes the course more than once. 16. Determine the probability that: (a) from 8 to 12 quizzes are given during the quarter. Determine: (a) q. Let random variable x be Bernoulli with n = 15 and p = 0. (b) n. when appropriate. Suppose that on the average. (Professor Rensselaer. what is the probability that he makes exactly 30 mistakes? (b) The typist decides to take typing lessons. the course instructor. what is the probability of him making fewer than 50 mistakes in a 25 page report? 18. A certain intersection averages 3 trafﬁc accidents per week. Suppose a typist makes an average of 30 mistakes per page. At Fargo Polytechnic Institute (FPI). On the average. On the average.38 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 10. Determine E(x 2 ). (c) at least 10 quizzes are given during the quarter. What is the probability of 10 alpha particles being emitted in: (a) 1 min.
on its anniversary day. 23. FPI admits only 1000 freshmen per year. (c) 0 defective transistors in a sample of 20 transistors. a bran cereal with milk expanding (exploding) marshmallow lizards. 21. 26. Otherwise. the probability that he hits his thumb is 0. Determine the probability that there are: (a) 3 defective transistors in a sample of 200 transistors. This year. then that lucky person will be mailed a brand new programmable calculator. to what value should the mean of the lizards per spoonful be set at to ensure this? 25. how many extra minutes can he expect to take off in building a house with 3000 nails? 24. Suppose 1% of the transistors in a box are defective. . The probability that a student will major in Bioengineering is 0. (b) σz. (b) more than 15 defective transistors in a sample of 1000 transistors.01. 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 ). Assuming that the lizards are randomly distributed in the cereal box. .000 cars are produced in a month. .002. k! k = 0. Assuming that . If it is. Determine the probability that fewer than 9 students major in Bioengineering. that person will get nothing. 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. a computer store is going to run an advertising campaign in which the employees will telephone 5840 people selected at random from the population of North America. If in building a house he pounds 1250 nails.STANDARD PROBABILITY DISTRIBUTIONS 39 20. determine: (a) E(z). (a) Every time a carpenter pounds in a nail. If 15. 1. what is the probability that none are perfect? 22. If random variable z equals the total number of students Professor Rensselaer helps each day. 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 . 95% of the spoonfuls will each have at least one lizard. wants to ensure that on the average. A perfect car is assembled with a probability of 2 × 10−5 . The caller will ask the person answering the phone if it’s his or her birthday. The manufacturer of Leaping Lizards. .
A battery manufacturer has found that 8. (d) f xA (αA). (a) Write the PDF for the resistance value. Find: (a) P (−1 < x < 0). (b) Find the probability that a particular resistor is within 2 ohms of the mean. 27. (f ) P (x is an integer). Event A = {0 < x ≤ 2} and B = {−1 < x ≤ 0} ∪ {1 < x ≤ 2}. (f ) f xB (αB). 33. Consider Problem 36. 31. (d) P (−1. (e) P (−2. (b) P (x < 0). 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. 35.1 to 4.3 years and 2. Determine: (a) the probability it takes the runner exactly 4 min to run a mile. determine: (a) P (x = 0). A certain transistor has a current gain. Assuming the battery lives are Gaussian distributed. (b) P (49 < r < 54). The time it takes a runner to run a mile is equally likely to lie in an interval from 4. A certain type of resistor has resistance values that are Gaussian distributed with a mean of 50 ohms and a variance of 3. and F: below 60. Find: (a) P (h > 74). what is the probability that grades will be equally distributed? 36.15 min.471). (c) Find P (49 < r < 54). (b) P (73 < h ≤ 80). Random variable x is uniform between −2 and 3. Assume that the scores on an examination are Gaussian distributed with mean 75 and standard deviation 10. 28. 32. . 29. (c) σx . Using Tables A. (b) ηx . (c) P (h − ηh  < 3σh ). Assume x is a standard Gaussian random variable. Find the probability that a bulb burns between 1000 and 1300 h.5% of their batteries last more than 3. 34. Consider Problem 32.9 and A. C: 70–80.2). (b) variance. D: 60–70. 37. Repeat Problem 1. (b) the computer store mails out from 20 to 40 calculators. In a class of 25 students.1 < x ≤ −0. (b) the probability it takes the runner from 4. ﬁnd: (a) the mean.0 to 4. that is Gaussian distributed with a mean of 77 and a variance of 11. ﬁnd the probability that: (a) the computer store mails out exactly 16 calculators. (g) FxB (αB). B: 80–90.10. If resistances are measured to the nearest ohm. Find the value of d so that the range 77 + d covers 95% of the current gains.583 ≤ x < 1.98 years. (e) FxA (αA). using the Gaussian approximation to the Bernoulli PMF. h. when appropriate.5). Grades are assigned as follows: A: 90–100. 30.2 min.08% of their batteries last less than 2.40 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS the person answering the phone won’t lie and that there is no such thing as leap year. (c) P (x < 0.
(b) P (x ≤ 78). (Hint: assume RV y is independent of x and has PDF f y (β) = 2 f x (β) and evaluate Fx. (b) P (1 < y < 7). (e) ρx. how many resistors have resistances that are within 10% of the average? 42.67.y = 0. 4). 1). 2σ 2 (a) Find the constant a. ∞). y ∼ G(−2. Assume x ∼ G(3. Find: (a) P (1 < y < 7  x = 0). (d) the probability that x is less than 75 if the score is measured to the nearest unit. Determine: (a) E(yx = 0).0 and a standard deviation of 0.STANDARD PROBABILITY DISTRIBUTIONS 41 38. Draw a sketch of the joint Gaussian contour equation showing the original and the translatedrotated sets of axes. what is the probability that at least 2 have a length greater than 2. A 250 question multiple choice ﬁnal exam is given. and that x and y are jointly Gaussian with ρx. 44. 47.000 resistors with resistances that are Gaussian distributed. Suppose x and y are jointly distributed Gaussian random variables with E(yx) = 2. 45. Assuming a Gaussian distribution for random variable x describing her golf game.1? 41. 1). A large box contains 10. Assume x and y are jointly Gaussian with x ∼ G(2. (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 . Each question has 5 possible answers and only one correct answer. If the average resistance is 1000 ohms with a standard deviation of 200 ohms. (b) f yx (β0). y ∼ G(3. E(xy) = −1 + 0.32x. Determine the probability that a student guesses the correct answers for 20–25 of 85 questions about which the student has no knowledge. 46.5y. (f ) σx. Assume x and y are jointly distributed Gaussian random variables with x ∼ G(−2.y = −5. (c) P (0 < y < 4x = 0). ﬁnd E(x). (c) Find σx using direct integration. (b) η y . 39.y = −0. Draw a sketch of the joint Gaussian constant contour for x and y. If 5 of these components are removed from different systems. and σ yx = 3. The average golf score for Professor Rensselaer is 78 with a standard deviation of 3. (c) P (−1 < x < 1. and ρx. Consider Problem 45. and ρx.y (∞.8 + 0. (d) P (3 < x < 10). Determine: (a) ηx . The RV x has PDF f x (α) = a exp − 1 (α − η)2 . . 43.y . determine: (a) P (x = 78). 1 < y < 7). (c) P (70 < x ≤ 80). Suppose a system contains a component whose length is normally distributed with a mean of 2. (d) σ y . 13). y ∼ G(1. 9). (c) σx .y . (b) Using the results of (a). 40.2.5.8835.) (b) Using direct integration. 8).
if he does not pin his opponent with his trick move.12: Circuit for Problem 54. . (a) Draw a sketch of the constant contour equation for the standardized RVs z1 and z2 . (b) Using the results of (a). at least two of these components must connect A with B. 51. (d) P (3 < x < 10). A component with an exponential failure rate is 90% reliable at 10. (c) P (0 < y < 4x = 0). Without exception. where T is a constant. 49. he loses the match on points. 50. 4).42 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 48.y = 0.12. Five individuals have this cancer. Consider Problem 52. 55. In the circuit shown in Figure 5.000 h. Draw a sketch of the joint Gaussian contour for x and y. 57. (c) 99%. Determine the number of years reliable at: (a) 90%.5. Assume x and y are jointly Gaussian with x ∼ G(−1. Suppose a system has an exponential failure rate in years to failure with η = 2. Determine the probability that at most three will be alive at the end of 2. 7). The survival rate of individuals with a certain type of cancer is assumed to be exponential with η = 4 years. (b) 95%. If 20 of these systems are installed. determine the probability that 10 are operating at the end of 2. Consider Problem 47. (c) P (0 < y < 4x = 0). y ∼ G(1.000 h.3 years. (b) f yx (β  0). William Smith is a varsity wrestler on his high school team.6. 56. William’s 1 A 3 2 B 4 FIGURE 5. (b) f z(α) if z = t − T. For successful operation of the circuit. Determine the probability that the circuit is operating successfully at 10. 54. Determine the number of hours reliable at 95%.378. Determine: (a) E(y  x = 0). Determine: (a) P (t > η).0433 years. Consider Problem 49. and ρx. 53. Random variable t is exponential with η = 2. 52. (d) P (−3 < x < 0). Determine: (a) E(y  x = 0). each of the four components operate independently of one another and have an exponential failure rate (in hours) with η = 105 . (b) f y  x (β  0).
Each customer takes 4 min to check out. In fact.3 people long. 60.1149599 exp(−0. The queue for the cash register at a popular package store near Fargo Polytechnic Institute becomes quite long on Saturdays following football games. Assume each period is 2 min and the match is 3 periods. Consider Problem 59 statement. What is the probability that the hospital will be without power during the next 24 h? 61. prints brochures describing their vacation packages as “Unique. the queue is 6. Suppose Fargo Community Hospital has a backup emergency generator to provide auxilary power during a power failure. On the average. 9 months later)? . 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.” The vacations certainly live up to their advertisement. (b) the probability that there will be at least two power failures during the coming year. period 3 : f xA (αA) = 0.4598394α)u(α). Outer Space Adventures. 58. Moreover. where A = {Smith did not pin his opponent during the previous periods}. Find the probability that William Smith wins: (a) at least 4 of his ﬁrst 5 matches. Unexpected and Unexplained. (b) more matches than he is expected to during a 10 match season. period 2 : f xA (αA) = 0. Assume that a power failure will last at least 24 h.4598394 exp(−0. Suppose that you have signed up for a vacation trip that starts at the end of this quarter. (b) the probability that you will have less than four people in the queue ahead of you. 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. Determine: (a) your expected waiting time to make a purchase.4 years. Determine: (a) the probability that there will be at least one power failure during the coming year.STANDARD PROBABILITY DISTRIBUTIONS 43 trick move also prevents him from ever getting pinned. (c) the probability that you will have more than ﬁve people in the queue ahead of you.2299197 exp(−0. following an exponential distribution. they are so unexpected that the length of the vacations is random. What is the probability that you will be back home in time for next fall quarter (that is. Consider Problem 57. Inc. 59. The average time between power failures in a Coop utility is once every 1. the emergency generator has an expected time between failures of once every 200 h. having an average length of 6 months.2299197α)u(α).1149599α)u(α). 62.
A certain brand of light bulbs has an average lifeexpectancy of 750 h.44 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 63. 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 1 2 3 4 TIME BULBS LEFT ON. The failure rate of these light bulbs follows an exponential PDF. what is the probability that it is burnt out? . HOURS 1000 750 500 1500 NUMBER OF BULBS 125 250 150 225 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.
Next. We will refer to this method as the CDF technique. the joint probability distribution is found from a function of two random variables using the joint PDF and the CDF.) Then. The ease of use of the CDF technique depends critically on the functional form of g (x).1. and mixed random variables. We ﬁrst evaluate the probability distribution of a function of one random variable using the CDF and then the PDF. and we wish to ﬁnd the CDF for random variable z = g (x). To make the CDF technique easier to understand.1 CDF Technique with Monotonic Functions Consider the problem where the CDF Fx is known for the RV x. the probability distribution for a single random variable is determined from a function of two random variables using the CDF. Of course. is a random variable on the probability space (S. with z(ζ ) = g (x(ζ )). (Recall that if g is a monotonic function then a onetoone correspondence between g (x) and x exists. and for all types of continuous.1 UNIVARIATE CDF TECHNIQUE This section introduces a method of computing the probability distribution of a function of a random variable using the CDF. we require that the function z : S → R∗ . consequently.45 CHAPTER 6 Transformations of Random Variables Functions of random variables occur frequently in many applications of probability theory. 6. For example. a full wave rectiﬁer circuit produces an output that is the absolute value of the input. a continuous monotonic function g . discrete. The CDF technique is applicable for all functions z = g (x). P ). we start the discussion of computing the probability distribution of z = g (x) with the simplest case. 6. ) and P (z(ζ ) ∈ {−∞. The input/output characteristics of many physical devices can be represented by a nonlinear memoryless transformation of the input. Proceeding from the deﬁnition of the CDF for z. Then. . we require z to be a measurable function on the measurable space (S. +∞}) = 0. The primary subjects of this chapter are methods for determining the probability distribution of a function of a random variable. the difﬁculties associated with computing the probability distribution of z = g (x) are investigated when the restrictions on g (x) are relaxed. we have for a .
the CDF of x is ⎧ ⎪ 0. the function does not have to be monotonic. the relationship between x and z can take on any form. The following example illustrates this technique. Letting g (x) = 2x + 1.46 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 )). the CDF for z is given by ⎧ ⎪ 0. 6. (6. (γ − 1)/8.1) indicates.2) 0. ⎪ ⎩ 1. Applying (6. However. Additionally. γ <1 1≤γ <9 9 ≤ γ. the only requirement on g is that z = g (x) be a random variable. ⎧ ⎪ ⎨ (6. Since random variable x is uniformly distributed. we obtain Fz(γ ) = which is also a uniform distribution. if z = g (x) and g is monotone decreasing. In fact.1). 4 ≤ (γ − 1)/2. As (6. Solution. In general. more than one solution of z = g (x) can exist—resulting in a manytoone mapping from x to z. we see that g is monotone increasing and that g −1 (γ ) = (γ − 1)/2. Example 6. Simplifying.2 CDF Technique with Arbitrary Functions In general. with the argument α replaced by g −1 (γ ). (γ − 1)/2 < 0 ⎨ γ −1 Fz(γ ) = Fx = (γ − 1)/8. the CDF of random variable z is written in terms of Fx (α). Find the CDF for random variable z = 2x + 1.1. In this general case.1. 0 ≤ α < 4 ⎪ ⎩ 1. this case is conceptually no more difﬁcult than . under these conditions it is impossible to write a general expression for Fz(γ ) using the CDF technique. α<0 ⎨ Fx (α) = α/4. 0 ≤ (γ − 1)/2 < 4 ⎪ 2 ⎩ 1. including discontinuities. Similarly. Fz(γ ) is no longer found by simple substitution. 4 ≤ α. then Fz(γ ) = P (z = g (x) ≤ γ ) = P (x ≥ g −1 (γ )) = 1 − Fx (g −1 (γ )− ).1) where g −1 (γ ) is the value of α for which g (α) = γ .1. Random variable x is uniformly distributed in the interval 0 to 4.
Since the Ai ’s are disjoint. If interval Ai (γ ) is of the form Ai (γ ) = (a i (γ ). Then Fz(γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )). . Let A(γ ) = {x : g (x) ≤ γ }.4) Ai (γ ). Partition A(γ ) into disjoint intervals {Ai (γ ) : i = 1. γ ]). . α < −1 −1 ≤ α < 1 1 ≤ α. Random variable x has the CDF ⎧ ⎪ 0.10) (6. Similarly. Find the CDF for the RV z = x 2 . .2.8) (6. Using this method. b i (γ )].9) (6. ⎪ ⎩ 1. 2. (6. then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )). and involves only careful book keeping. then P (x ∈ Ai (γ )) = Fx (b i (γ )) − Fx (a i (γ )− ). . Example 6. ⎨ Fx (α) = (3α − α 3 + 2)/4.} so that A(γ ) = ∞ (6. if interval Ai (γ ) is of the form Ai (γ ) = [a i (γ ). b i (γ )].1.TRANSFORMATIONS OF RANDOM VARIABLES 47 the previous case.6) The above probabilities are easily found from the CDF Fx . any function g and CDF Fx is amenable to a solution for Fz(γ ).5) Note that the intervals as well as the number of nonempty intervals depends on γ . (6.7) The success of this method clearly depends on our ability to partition A(γ ) into disjoint intervals. The following examples illustrate various aspects of this technique. i=1 (6.3) (6. Fz(γ ) = ∞ i=1 P (x ∈ Ai (γ )). Note that A(γ ) = g −1 ((−∞.
Random variable z is deﬁned by ⎧ x < −2 ⎪ −1. Let A(γ ) = g −1 ((−∞. ∞)).1. −1 ≤ γ < 2 2 ≤ γ. √ √ [− γ . ⎪ ⎪ ⎨ γ −5 −γ − 1 γ + 1 A(γ ) = −∞.3. −1 ≤ x < 0 ⎪ ⎪ 3x − 1.48 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. γ < −1 . ⎪ ⎪ ⎪ 3x + 5. 0≤γ <1 ⎪ ⎩ 1 1 ≤ γ. Plots for this example are given in Figure 6. 3 ≤ α. Fz(γ ) = + Fx . ⎪ 3 3 3 ⎪ ⎩ R Consequently. 1 ≤ x. γ <0 γ ≥ 0. Random variable x is uniformly distributed in the interval −3 to 3. . ⎪ −2 ≤ x < −1 ⎨ z = g (x) = −3x − 1. γ ]. ⎪ 0≤x<1 ⎪ ⎪ ⎩ 2. The CDF for random variable x is ⎧ ⎪ 0. we obtain ⎧ ⎪ 0. ⎧ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎩ Fx γ −5 3 0. √ √ Fz(γ ) = Fx ( γ ) − Fx ((− γ )− ). Noting that Fx (α) is continuous and has the same functional form for −1 < α < 1. −3 ≤ α < 3 ⎪ ⎩ 1. Solution.1. Find Fz(γ ). Letting g (x) = x 2 .1 we ﬁnd ⎧ ∅. Example 6. γ +1 − Fx 3 1 −γ − 1 3 − γ < −1 −1 ≤ γ < 2 2 ≤ γ. γ ]) = so that ∅. Referring to Figure 6. we ﬁnd A(γ ) = g −1 ((−∞. ∪ . γ <0 ⎨ √ √ Fz(γ ) = (3 γ − ( γ )3 )/2. α < −3 ⎨ Fx (α) = (α + 3)/6.
z = g (x) is a discrete random variable if g (α) changes values only on intervals where f x (α) = 0. This results in a jump in Fz and an impulse function in f z.1. Moreover.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 6.1: Plots for Example 6.3. Substituting. Whenever random variable x is continuous and g (x) is constant over some interval or intervals. 6 γ < −1 −1 ≤ γ < 2 2 ≤ γ. dependent on the CDF for random variable x. γ −5 γ + 1 −γ − 1 +3+ − 3 3 3 1 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. In fact. discrete or mixed. random variable z is mixed due to the constant value of g (x) in the intervals −3 ≤ x < −2 and 1 ≤ x < 3. then random variable z can be continuous. Fz(γ ) = Note that Fz(γ ) has jumps at γ = −1 and at γ = 2. . we obtain ⎧ ⎪ ⎪ ⎨1 ⎪6 ⎪ ⎩ 0. As presented in the previous example.
⎪ ⎪ ⎩ 1. ⎨ z = g (x) = 0. x Fz(γ ) = ⎪ Fx (0.4. ⎧ ⎪ Fx (γ + 1).1. g( x ) 1 x γ ≤ −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ. 0. Consequently.50 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6. 0]. ⎪ ⎩ 1.5 < x. α < −1 ⎨ 3 Fx (α) = (α + 3α + 4)/18. γ ≤ −1 ⎪ ⎪ ⎨ (−∞. −1 ≤ γ < 0 A(γ ) = {x : g (x) ≤ γ } = ⎪ (−∞. With the aid of Figure 6.4.2: Transformation for Example 6. x≤0 0 < x ≤ 0. To ﬁnd f z.5). 2 ≤ α. 0≤γ <1 ⎪ ⎪ ⎩ (−∞.2. −1 < α < 2 otherwise. Solution. The CDF for random variable x is ⎧ ⎪ 0. Random variable x has the PDF f x (α) = (1 + α 2 )/6. then differentiate this result. Figure 6. 1 ≤ γ. 0. −1 ≤ α < 2 ⎪ ⎩ 1.1. .2 shows a plot of g (x). ∞). we evaluate Fz ﬁrst. γ + 1]. ⎪ ⎪ ⎨ F (0).5 0. Find the PDF of random variable z deﬁned by ⎧ ⎪ x − 1. we ﬁnd ⎧ ⎪ (−∞. −1 −1 −2 1 2 FIGURE 6.5].
the CDF technique is applicable if random variable x is mixed or discrete.TRANSFORMATIONS OF RANDOM VARIABLES 51 Substituting. The previous examples illustrated evaluating the probability distribution of a function of a continuous random variable using the CDF technique. the CDF technique is used without any changes or modiﬁcations as shown in the next example. The CDF for x is ⎧ ⎪ 0. ⎨ Fx (α) = α/10.1. ⎪ ⎨ Fz(γ ) = 2/9. Differentiating Fz. Note that P (x ≤ 0) = 0. continuous and discontinuous. γ < −2 −2 ≤ γ < −1 −1 ≤ γ < 0 0≤γ <1 1 ≤ γ. Example 6. ⎪ ⎪ ⎪ 5/16.1. ⎪ ⎩ 1. . This technique is applicable for all functions z = g (x). ∞). We ﬁnd Fz(γ ) = 0. Find the CDF for random variable z = g (x) = − ln(x).1) ln(0. For γ > 0. Random variable x has PDF f x (α) = 0. Find the CDF for z = g (x) = 1/x 2 . 18 144 16 Example 6. respectively.1γ . we ﬁnd A(γ ) = {x : − ln(x) ≤ γ } = (e −γ . Random variable x is uniformly distributed in the interval from 0 to 10. ⎪ ⎪ ⎩ 1.5). Note that Fz(γ ) has jumps at γ = 0 and γ = 1 of heights 13/144 and 11/16. 1 − e −0. Solution. ⎧ 0.1) ≤ γ . For mixed random variables. so that Fz(γ ) = 1 − Fx ((e −γ )− ). γ < ln(0. ⎪ ⎪ ⎪ ⎪ ((γ + 1)3 + 3γ + 7)/18.6. f z(γ ) = 13 11 3γ 2 + 6γ + 6 (u(γ + 2) − u(γ + 1)) + δ(γ ) + δ(γ − 1). as required since g (x) = − ln(x) is not deﬁned (or at least not real–valued) for x ≤ 0. α<0 0 ≤ α < 10 10 ≤ α.5δ(α − 0.5. Additionally.5(u(α) − u(α − 1)) + 0.
Random variable z is deﬁned by ⎧ ⎪ −1.5 Fx (α) = ⎪ 0.5 + 0. Fz(γ ) = 1 − ⎪ 0. (b)Fz(1/2). (c ) f z(15). ⎪ ⎩ 1. α<0 ⎪ ⎪ ⎨ 0. 1/16.5γ −1/2 .52 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. 0. Drill Problem 6. (d ) f z(1/2). 4 ≤ γ.5α(u(α) − u(α − 2)). Answers: 0.5γ . Random variable x has the PDF f x (α) = 0. 1/4. x<1 −1 ≤ x ≤ 1 x > 1.1. (c )Fz(3/2).5 ≤ (1/ γ )− < 1 √ − 1 ≤ (1/ γ ) . 2/15.5 + 0.1. 1/15.∞ .5α. (b)Fz(1).5 − 0. 1 ≤ α. γ √ Since Fx (−1/ γ ) = 0 for all real γ .5 ≤ α < 1 ⎪ ⎪ ⎩ 1. 1≤γ <4 ⎪ ⎩ 1 − 0. 1 A(γ ) = {x : x −2 ≤ γ } = −∞. √ (1/ γ )− < 0 √ 0 ≤ (1/ γ )− < 0. ⎪ ⎪ ⎩ 1.5 √ 0. Random variable x is uniformly distributed in the interval −1 to 4.1. Fz(γ ) = 0. we have ⎧ ⎪ 0. After some algebra. Answers: 0. Determine: (a) Fz(−1/2). For γ < 0. ⎪ ⎪ ⎨ 0. − √ γ so that 1 ∪ √ . ⎧ ⎪ ⎨ √ √ Fz(γ ) = Fx (−1/ γ ) + 1 − Fx ((1/ γ )− ). For γ > 0. The mixed random variable x has CDF ⎧ ⎪ 0. ⎨ z= x.5γ −1/2 . Drill Problem 6. Determine: (a) Fz(0).5α. 0. 1. 1/15. Random variable z = 3x + 2. (c ) f z(0). 0 ≤ α < 0.2.5γ −1/2 . . γ <1 −1/2 Fz(γ ) = 0.
5 0. 0. i = 1. a method for handling mixed random variables with the PDF technique is introduced. in many situations it is much simpler to use than the CDF technique. x ≤ 0.11) . (c )Fz(4).5.TRANSFORMATIONS OF RANDOM VARIABLES 53 Drill Problem 6. Drill Problem 6. Determine: (a) Fz(1/8). . 6. (c )Fz(3/2). ⎪ ⎩ 3. 2. .617.1. called the PDF technique. Let x be a continuous RV with PDF f x (α) and let the RV z = g (x) . 1/16. Let αi = αi (γ ) = g −1 (γ ). . however. We introduce the PDF technique for two reasons.089. Random variable x has the PDF f x (α) = 0. Then. we will ﬁnd the PDF method most useful in extensions to multivariate functions.1. (d )Fz(4). First.2. (6. (d ) f z(4).3. In this section.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. 6. The PDF technique. Random variable z = 1/x 2 .4. Second. (b)Fz(1/2). 1/16. 0. Random variable z is deﬁned by z= ⎧ ⎪ ⎨ −1. Answers: 1/4. Finally. 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. .5 < x ≤ 1 x > 1.5α(u(α) − u(α − 2)).2. (b)Fz(0). x + 0. Random variable x has PDF f x (α) = e −α−1 u(α + 1).1 Continuous Random Variable Theorem 6.022. we discuss a wide variety of situations using the PDF technique with functions of continuous random variables. we consider computing the conditional PDF of a function of a random variable using the PDF technique. Assume g is continuous and not constant over any interval for which f x = 0 . Now.0519. 1. 0. we introduce a second method for calculating the probability distribution of a function z = g (x) using the probability density function.1. Determine: (a) Fz(−1). Answers: 0.
h)) − Fx (a i (γ . h). h→0 Note that (for all γ with f x (αi (γ )) = 0) 1 b i (γ . (Fx (b i (γ . . 2. h) = Then Fz(γ ) − Fz(γ − h) = By hypothesis. h→0 ∞ i=1 ∞ i=1 ∞ i=1 f x (αi (γ )) . such that I (γ . h) . Partition I (γ .54 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS denote the distinct solutions to g (αi ) = γ . . h)). g (1) (αi (γ )) if f x (αi (γ ) = 0. h) = αi (γ ). h) into disjoint intervals of the form Ii (γ . . h) = {x : γ − h < g (x) ≤ γ }. b i (γ . .12) i = 1. Let h > 0 and deﬁne I (γ . h). h) = lim b i (γ . h) − a i (γ . h) − a i (γ . b i (γ . whereas g (1) may be positive or negative. h) The desired result follows by taking the product of the above limits. h) b i (γ . Then f z(γ ) = where we interpret f x (αi (γ )) = 0. h))) lim a i (γ . Ii (γ . h)) h g (αi (γ )) lim and that h→0 lim Fx (b i (γ . Proof. h)) − Fx (a i (γ . h)) = f x (αi (γ )). g (1) (αi (γ )) (6. h) − a i (γ . = lim = (1) h→0 h→0 g (b i (γ . h) = (a i (γ . . h)) − g (a i (γ . The absolute value appears because by construction we have b i > a i and h > 0.
g (1) (αi ) = 2/αi 3 = 2γ 3/2 . 0.2. f z(γ ) = or f z(γ ) = 0. Random variable x has PDF 3 5 f x (−γ −1/2 ) + f x (γ −1/2 ) u(γ ). Find the PDF for random variable z = g (x) = 1/x 2 . For γ > 0 there are two solutions to g (αi ) = γ : 1 α1 (γ ) = − √ . we have g (1) (α) = −2α −3 . so that f z(γ ) = 0 for γ < 0. and f z(γ ) = Substituting.75(1 − γ −1 )(u(1 − γ −1/2 ) − 0 + 1 − u(γ −1/2 − 1)) u(γ ). Solution.2. . 1<γ <9 otherwise. there are no solutions to g (αi ) = γ . Random variable x is uniformly distributed in the interval 0–4.75(1 − γ −1 )(u(γ − 1) − 0 + 1 − u(1 − γ )) u(γ ).3. For γ < 0.1. γ and 1 α2 (γ ) = √ . hence. Solution. 2γ 3/2 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)).2.TRANSFORMATIONS OF RANDOM VARIABLES 55 Example 6.75(γ − 2 − γ − 2 )u(γ − 1). In this case. Hence f z(γ ) = f x ((γ − 1)/2)/2 = Example 6. Since γ = g (α) = α −2 . 2γ 3/2 0. 6 Find the PDF for random variable z = g (x) = x 2 .2. Find the PDF for random variable z = g (x) = 2x + 1. Random variable x has PDF f x (α) = 0. Example 6. f z(γ ) = Simplifying. 2γ 3/2 0. We easily ﬁnd g (1) (α) = 2. there is only one solution to the equation γ = 2α + 1.75(1 − α 2 )(u(α + 1) − u(α − 1)). given by α1 = (γ − 1)/2. γ 1/8.
so that f z(γ ) = 0 for γ < 0. Here. 6. Solution. and then combine the results to yield f z. Example 6. The continuous part of the PDF of x is handled by the PDF technique. we have g (1) (α) = 2α.2 Mixed Random Variable Consider the problem where random variable x is mixed. 12 γ 0<γ <1 1<γ <4 elsewhere. f z(γ ) = or 1+γ √ (u(1 − γ ) − 0 + 1 − u(γ − 4))u(γ ).2.5 are mapped to probability masses of 1/8 for z at e 0. hence.2. There is only one solution to g (α) = γ : α1 (γ ) = − ln(γ ). and √ √ f x (− γ ) + f x ( γ ) u(γ ). The probability masses of 1/8 for x at −0.5) + δ(α − 0. For γ < 0 there are no solutions to g (αi ) = γ .4.56 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Solution. and we wish to ﬁnd the PDF for z = g (x).5 and 0. For all γ > 0 such that . g (1) (αi ) = 2αi  = 2 γ .5 and e −0. consider the following example. We have g (1) (α1 ) = −e ln(γ ) = −γ . 12 γ ⎧ ⎪ (γ −1/2 + γ 1/2 )/6. 8 8 8 Find the PDF for the RV z = g (x) = e −x .5 . f z(γ ) = √ 2 γ Substituting. respectively. ⎪ ⎩ 0. f z(γ ) = Simplifying. Random variable x has PDF 1 1 3 f x (α) = (u(α + 1) − u(α − 1)) + δ(α + 0. ⎨ f z(γ ) = (γ −1/2 + γ 1/2 )/12.5). √ Since γ = g (α) = α 2 . For γ > 0 there are two solutions to g (αi ) = γ : √ √ α1 (γ ) = − γ . and α2 (γ ) = γ . 1+γ √ √ √ √ √ (u(1 − γ ) − u(−2 − γ ) + u( γ + 1) − u( γ − 2))u(γ ). we treat the discrete and continuous portions of f x separately. To illustrate the use of this technique.
2. (i) If A is an event deﬁned for an interval on z. 9 . (ii) If A is an event deﬁned for an interval on x.TRANSFORMATIONS OF RANDOM VARIABLES 57 α1 (γ ) ± 0.5 > 0 we have f z(γ ) = Combining these results. by the deﬁnition of a conditional PDF. Then. P (A) γ ∈ A. Depending on whether the event A is deﬁned on the range or domain of z = g (x). First. we ﬁnd f z(γ ) = 3 1 1 (u(γ − e −1 ) − u(γ − e )) + δ(γ − e 0. the above techniques apply to ﬁnd the conditional PDF for z = g (x). we have f zA (γ A) = f z(γ ) .14) Example 6. and we wish to evaluate the conditional PDF for random variable z = g (x). g (1) (αi (γ )) (6. We have P (A) = 1 6 2 0 7 (1 + α 2 )d α = .5 ) + δ(γ − e −0. given event A. f zA . given that event A occurred. one of the following two methods may be used to determine the conditional PDF of z using the PDF technique. 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) . Solution. the conditional PDF. we solve for the conditional PDF for x and then ﬁnd the conditional PDF for z. Consider the problem where random variable x has PDF f x . Random variable x has the PDF 1 f x (α) = (1 + α 2 )(u(α + 1) − u(α − 2)). based on f xA .3 Conditional PDF Technique Since a conditional PDF is also a PDF. γ 6.2.13) and f zA (γ A) = 0 for γ ∈A. 6 Find the PDF for random variable z = g (x) = x 2 . 8γ 8 8 f x (− ln(γ )) .5. is computed by ﬁrst evaluating f z using the technique in this section. (6.5 ). given A = {x : x > 0}.
2. (b) f z(6). (c )E(z). We √ √ have α1 (γ ) = γ and g (1) (α1 (γ )) = 2 γ .58 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS so that f xA (αA) = 3 (1 + α 2 )(u(α) − u(α − 2)). 1. 48. (c ) f zA (2A). (b) f z(9/4). (d ) f z(2). (c ) f zz≤1 (1/4z ≤ 1).(d ) f zA (9A). 0. Drill Problem 6. (d )σz . 8/9.2. 3. Random variable x has a uniform PDF in the interval 0–8. 0. . 0 ≤ α < 0. Drill Problem 6. 1/24.5 ≤ α ≤ 1 ⎪ ⎩ 0.2. f zA (γ A) = 3 √ 1 γ+√ 28 γ (u(γ ) − u(γ − 4)). (b) f xA (1A).2. 1/3. Consequently. Random variable x has the PDF 2 f x (α) = (α + 1)(u(α + 1) − u(α − 2)). Answers: 1. (c ) f zz>1/8 (1/4z > 1/8).43. Determine: (a) P (A). 14 There is only one solution to γ = g (α) = α 2 on the interval 0 < α < 2 where f xA = 0. Random variable x has the PDF 2 f x (α) = α(u(α) − u(α − 3)). Answers: 13.4. 1/8.2.5 ⎨ 2 f x (α) = 3(1 − α ). Random variable 2 z = 3x + 1 . Drill Problem 6. 9 Random variable z = 2x 2 and event A = {x : x ≥ 0}. 9 Random variable z = (x − 1)2 . Answers: 4/9. (b) f z(1/4). Random variable x has the PDF ⎧ ⎪ 9α 2 . (d )E(zz ≤ 1).1. Drill Problem 6. 5/27.Use the PDF method to determine: (a) f z(0).2. otherwise. 0.3. Answers: 0. Use the PDF method to determine: (a) f z(1/4).Use the PDF method to determine: (a) f z(1/27). Random variable z = x 3 .52. 1/2.
Random variables x and y have joint PDF f x. Pay careful attention to the limits of integration to determine the range of integration in which the integrand is zero because f x. where A(γ ) = {(x. we have Fz(γ ) = P (z ≤ γ ) = P (g (x. the probability distribution of z = g (x.17) This result cannot be continued further until a speciﬁc Fx.1. y) : g (x. Sketching the support region for f x. We have A(γ ) = {(α.y = 0. Here.y and g (x.y (α. . y) ≤ γ ) = P ((x. We require the volume under the surface f x.y (the region where f x. y) ∈ A(γ )).TRANSFORMATIONS OF RANDOM VARIABLES 59 6. β) where α ≤ γ − β: Fz(γ ) = ∞ −∞ γ −β −∞ 1/4. y) are considered. Let us consider several examples to illustrate the mechanics of the CDF technique and also to provide further insight. in the problem solution. β)d α dβ. y) is computed using a CDF technique similar to the one at the start of this chapter. even crucial. The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6. (6.3 ONE FUNCTION OF TWO RANDOM VARIABLES Consider a random variable z = g (x.3.y (α. Because we are dealing with regions in a plane instead of intervals on a line. our efforts are concentrated on evaluating Fz through integrals. 0 < α < 2. y) ≤ γ }. y).y (α. Solution. with the ease of solution critically dependent on g (x. In this section. f x. The ease in solution for Fz is dependent on transforming A(γ ) into proper limits of integration. With z = g (x. β).y (α.y is not constant) and the region A(γ ) is often most helpful. 0 < β < 2 otherwise. 0. Example 6. these problems are not as straightforward and tractable as before. or Fx. β) : α + β ≤ γ }. y).y = 0.16) dFx. our efforts primarily dealt with algebraic manipulations. y) created from jointly distributed random variables x and y. Remember that in the case of a single random variable.15) (6. β) = Find the CDF for z = x + y.
y (α. 0 < α < 1.1.2. For γ < −1 we have Fz(γ ) = 0 and f z(γ ) = 0.3(a) we obtain Fz(γ ) = 0 2 0 γ −β 1 1 d αdβ = γ 2 . Random variables x and y have joint PDF f x. For γ < 0 we have Fz(γ ) = 0. for −1 ≤ γ < 0. β) where α ≤ γ + β: Fz(γ ) = ∞ −∞ γ +β −∞ 2 γ −2 2 γ −β 1 1 d αdβ = 1 − (4 − γ )2 . Fz(γ ) = 1 −γ 0 γ +β 1 d αdβ = (1 + γ )2 .y (α.3. We require the volume under the surface f x. 0 < β < 1 otherwise. Example 6. for 4 ≤ γ . with the aid of Figure 6. 2 .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. β)d αdβ.y (α. 4 8 1.4(a). β) : α − β ≤ γ }. With the aid of Figure 6.3. We have A(γ ) = {(α. Fz(γ ) = 1. For 0 ≤ γ < 2. β) = Find the PDF for z = x − y. 4 8 For 2 ≤ γ < 4. Solution. f x. we consider the complementary region (to save some work): Fz(γ ) = 1 − Finally. referring to Figure 6. 0.3(b).3: Plots for Example 6.
α γ β 1 1 γ 0 1 α FIGURE 6.2. we have α/β > 1.5 it is easiest to integrate with respect to β ﬁrst: for 1 ≤ γ . Example 6. 0<β<α<1 otherwise. We have A(γ ) = {(α. for 1 ≤ γ . where x and y have the joint PDF f x. Inside the support region for f x. we consider the complementary region shown in Figure 6. Solution. Fz(γ ) = 1. Fz(γ ) = 0 1 α α/γ 1 1 dβd α = 1 − . As shown in Figure 6. .3. hence. Finally. 2 γ +β 1 so that f z(γ ) = 1 − γ .y (α. Find the CDF for z = x/y.5: Integration region for Example 6.3. β) = 1/α. so that f z(γ ) = 1 + γ .4: Plots for Example 6.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. β) : α/β ≤ γ }.3.y . so that f z(γ ) = 0. 0. For 0 ≤ γ < 1.3. for γ < 1 we have Fz(γ ) = 0.3.4(b) (to save some work): Fz(γ ) = 1 − 0 1−γ 1 d αdβ = 1 − (1 − γ )2 .
2 For 1 ≤ 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 ). β = r sin(θ). γ we ﬁnd with the aid of Figure 6. Referring to Figure 6.3. Find the CDF for z = x 2 + y 2 . β) : α 2 + β 2 ≤ γ }. β) = 3α. 0 < β < α < 1 0. . where x and y have the joint PDF f x.y (r cos(θ). otherwise. we obtain Fz(γ ) = 0. Solution. we have Fz(γ ) = 1 for 2 ≤ γ .6(a).6: Integration regions for Example 6.62 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS Example 6.4. we split the integral into two parts: one with polar coordinates.4. the other using rectangular coordinates.3. Fz(γ ) = 0 2π r 2 ≤γ f x. Fz(γ ) = 0 π/4 0 √ γ 1 3 3r 2 cos(θ)dr d θ = √ γ 2 . We have A(γ ) = {(α. 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. For γ < 0. for 0 ≤ γ < 1. r sin(θ))r dr d θ.y (α. With √ γ −1 sin(θ1 ) = √ . Finally. 2 π/4 θ1 0 √ γ √ α γ −1 0 3r cos(θ)dr d θ + 2 0 1 3αdβ d α.
y (α. h(x. A(γ . β).y (α. y) and w = h(x. y) ≤ γ .18) (6. Finally. y) and w = h(x. y) ≤ ψ. Random variables x and y have the joint PDF f x. we ﬁnd the joint distribution of random variables z = g (x. ψ]). (c )Fz(1). y) ≤ ψ}. y) ≤ γ and w = h(x.25.y (α.w (γ . With z = g (x. Then. 6.ψ) (6. 3 e −1 .20). Example 6.1. 0. the case of one function of two random variables is treated by using the joint PDF technique with an auxiliary random variable. β) = e −α e −β u(α)u(β). β) = 0. y). Deﬁne A(γ .w .. we consider a bivariate CDF technique. y) from jointly distributed random variables x and y. 1 − 3 e −1 .3. γ ]) = {(x. 0 ≤ β ≤ 2 otherwise. Random variables x and y have joint PDF f x. ﬁnd the joint PDF f z. (b) f z(−1). Find (a) Fz(−1/3). 3 e −1 . and let z = g (x.4. We have g −1 ((−∞. 0 ≤ α ≤ 2.e. ψ) = g −1 ((−∞. Random variable z = x − y . y) = x + y and w = h(x. Note that A(γ . the joint PDF technique for ﬁnding the joint PDF for random variables z and w formed from x and y is described. y) = y. Then Fz. the conditional joint PDF is presented. (6. First. ψ) to be the region of the x − y plane for which z = g (x. y) : g (x. γ ]) ∩ h −1 ((−∞. y) : x + y ≤ γ } . 4 4 4 4 6. ψ) = {(x. . and (d) f z(1).TRANSFORMATIONS OF RANDOM VARIABLES 63 Drill Problem 6.1 Bivariate CDF Technique Let x and y be jointly distributed RVs on the probability space (S. Solution.1. i.4 BIVARIATE TRANSFORMATIONS In this section. Answers: 1 e −1 .20) It is often difﬁcult to perform the integration indicated in (6. ψ) = A(γ .4.19) dFx. Next. P ).
With the aid of Figure 6.y (γ − ψ.w (γ . Instead of carrying out the above integration and then differentiating the result.7 and (6. The intersection of these regions is A(γ .7: Integration region for Example 6. y) : y ≤ min(γ − x.w (γ . ψ). Performing the indicated limits. ψ) = lim h 1 →0 h 1 ∂γ and ∂ 2 Fz.w directly.y (α. which is illustrated in Figure. it is usually easier to ﬁnd the joint PDF than to carry out the integral indicated in (6. and h −1 ((−∞.w (γ . 0. 6.25. ψ) = {(x. β).y (α.7. ψ)}. we ﬁnd that f z. f z.64 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS y ψ γ γ −ψ x FIGURE 6.1.w (γ . ψ) = 0.w (γ .w When the RVs x and y are jointly continuous. we obtain f z. β). We ﬁnd 1 ∂ Fz. ψ) = ψ −∞ γ −β −∞ d Fx. ψ]) = {(x. substituting. 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. . dFx.4. β). 0 < ψ < 2 otherwise. y) : y ≤ ψ}. 0 < γ − ψ < 2. we differentiate the above integral to obtain the PDF f z. ψ) = f x.20).y (α.20) we ﬁnd Fz.
h 2 ). h 2 > 0 and deﬁne I (γ . . h 1 .1. βi ) = γ and h(αi . Let x and y be jointly continuous RVs with PDF f x.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. ψ]).4. ψ. βi ) denote the unique element of h 1 →0 h 2 →0 lim lim Ii (γ .22) Let (αi . h 1 . Deﬁne the Jacobian ∂g (α. h 1 .24) J (α.25) i = 1. ψ.26) (6. β) ∂β ∂h(α. (6. y) . h 2 ) = g −1 ((γ − h 1 . ψ). h 2 ). β) ∂β (6. β) ∂α ∂g (α. ψ) ≈ i f x.21) Ii (γ . β) ∂α ∂h(α. ψ. The result is summarized by the theorem below. βi ) = ψ. Theorem 6. .w (γ . Let (αi (γ . β) = . . h 1 . 2.27) . and let z = g (x.h 2 ) d α dβ.4. . (6. Partition I into disjoint regions so that I (γ . (6.y .h 1 .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. Then for small h 1 and small h 2 we have h 1 h 2 f v. ψ. Let h 1 > 0. γ ]) ∩ h −1 ((ψ − h 2 . βi (γ .y (αi . y) and w = h(x. βi ) Ii (γ . h 2 ) = i (6.ψ.TRANSFORMATIONS OF RANDOM VARIABLES 65 6. be the distinct solutions to the simultaneous equations g (αi . ψ)). (6.
4. βi ) for αi and βi . Example 6. f x. ψ). ψ) = 1/4. we obtain f z. Example 6. J  1 0 1 = 1 · 1 − 1 · 0 = 1. 0 < γ − ψ < 2.y (γ − ψ.w is illustrated in Figure 6.2. Find the joint PDF for z = g (x. 0. β) = 12αβ(1 − α). we ﬁnd f z.w (γ .28) The most difﬁcult aspects to ﬁnding f z.w (γ .66 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where the indicated partial derivatives are assumed to exist. Random variables x and y have the joint PDF f x. 0 < α < 2. β) = 1/4. and determining the support region for f z. y) = y.y (αi . Random variables x and y have the joint PDF f x. 0 < β < 1 otherwise.y (α. βi ) .4. . βi ) and ψ = h(αi .3. 0.w . βi ) (6. β1 ) = f x. There is only one solution to γ = α + β and ψ = β: α1 = γ − ψ and β1 = ψ. 0 < β < 2 otherwise. ψ) = ∞ i=1 f x. 1 The region of support for f z. ψ) = Substituting. y) = y. The Jacobian of the transformation is J = Applying Theorem 1. Let us consider several examples to illustrate the mechanics of this PDF technique. J (αi . 0 < ψ < 2 otherwise.y (α.w (γ . Then f z.y (α1 . 0. Find the joint PDF for z = g (x. y) = x + y and w = h(x.8. Solution. y) = x 2 y and w = h(x. 0 < α < 1.w using the joint PDF technique are solving the simultaneous equations γ = g (αi .
2. = 0. β1 ) is in the ﬁrst quadrant of the α − β plane. Then β2 = − γ ψ and α2 = − γ /ψ. Let√ √ √ √ ting β1 = γ ψ we have α1 = γ /ψ.y (α. ψ) = √ 2 γψ √ 6(1 − γ /ψ).w (γ . ψ) = + . Random variables x and y have the joint PDF f x. The Jacobian is J1 = Applying Theorem 1. Solution.4. we ﬁnd α1 = γ /ψ and β1 = ψ. 1 √ f x. we ﬁnd 2α1 β1 0 2 α1 = 2α1 β1 = 2 γ ψ.y ( γ /ψ. Note that the solution (α1 . y) = xy and w = h(x. 2ψ 2ψ . we ﬁnd α = γ /β = β/ψ. − γ ψ) f z. Hence β −β/α 2 β α =2 . Find the joint PDF for z = g (x. −1 < β < 1 otherwise.4. −1 < α < 1.8: Support region for Example 6. so that β 2 = γ ψ. Solving γ = αβ and ψ = β/α.4.y (α. y) = y/x. γ ψ) f x. 0 < ψ < 1 otherwise. Solving γ = α 2 β and ψ = β. β) = 0 for α < 0. 0.w (γ .TRANSFORMATIONS OF RANDOM VARIABLES 67 ψ 2 1 0 1 2 3 4 γ FIGURE 6.y (− γ /ψ. 0< √ γ /ψ < 1.y ( γ /ψ. β2 ) is in the third quadrant of the α − β plane. ψ) f z. The other solution α2 = −α1 is not needed since f x. Example 6. We ﬁnd J = so that J 1 = 2ψ = J 2 . 1/α α √ √ √ √ f x.25(α + 1). β) = 0. √ Solution. the solution (α2 .
This support region is illustrated in Figure 6. 0 < γ < ψ < γ −1 . √ √ f x. Similarly. we ﬁnd f z.29) It is usually advisable to let the auxiliary random variable w equal a quantity which allows a convenient solution of the Jacobian and/or the inverse equations. the support region is γ −1 < ψ < γ < 0. ψ)d ψ. y) by introducing auxiliary random variable w = h(x. . Consequently.y to random variable z = g (x.w and then ﬁnding the marginal PDF f z using f z(γ ) = ∞ −∞ f z.y ( γ /ψ.9. γ <1 ψ and 0 < γ ψ < 1. the two components of the PDF f z. this support region is easily seen to be 0 < γ < ψ < γ −1 . Auxiliary Random Variables The joint PDF technique can also be used to transform random variables x and y with joint PDF f x. This method is an alternative to the CDF technique presented earlier. in the third quadrant. γ /ψ < 1 and 0 < γ ψ < 1. √ √ f x.y (− γ /ψ.w have identical regions of support in the γ − ψ plane. 0. (6. Finally. y) to ﬁnd f z. − 1 < − γ ψ < 0. or γ −1 < ψ < γ < 0 otherwise. ψ) = 1/(4ψ). In the ﬁrst quadrant of the γ − ψ plane.w (γ .68 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS In the γ − ψ plane. γ ψ) 2ψ has support region speciﬁed by 0< or 0< Similarly.w (γ . − γ ψ) 2ψ has support region speciﬁed by −1 < − γ /ψ < 0 and or 0< γ <1 ψ and 0 < γ ψ < 1.
Solution. y) = x 2 . Let auxiliary variable w = h(x. we ﬁnd α = √ √ ± γ and β = ψ. ⎧ √ 4 γψ √ f x. The only solution inside the support region for f x. Example 6.4. 0 < ψ < 1 otherwise. otherwise.TRANSFORMATIONS OF RANDOM VARIABLES 69 ψ 3 2 1 −3 −2 −1 1 2 3 γ −1 −2 −3 FIGURE 6. ψ) ⎨ √ = 2ψ.4. 1 4αβ. y) = y. 0. . β) = Find the PDF for z = g (x. 2 γ f z.y is α = γ and β = ψ.9: Support region for Example 6. We ﬁnd the marginal PDF for z as f z(γ ) = ∞ −∞ f z.y (α. for 0 < γ < 1 and f z(γ ) = 0. 0 < β < 1 otherwise. 0 < α < 1.y ( γ .w (γ . The Jacobian of the transformation is J = so that 2α 0 0 = 2α. Solving γ = α 2 and ψ = β. 0 < γ < 1. ψ)d ψ = 0 1 2ψ d ψ. Random variables x and y have the joint PDF f x.4.w (γ . ψ) = = √ ⎩ 2 γ 0.5.
Referring to Figure 6. ψ) = 4(γ − ψ)ψ. y) = x + y. f z = 0.70 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS ψ 1 0 1 2 γ FIGURE 6. y). Consider the problem where random variables x and y have joint 2 8 4(γ − ψ)ψ d ψ = − γ 3 + 4γ − .6. 0 < ψ < 1 otherwise. Solution.y (α. Find the PDF for z = g (x. we ﬁnd α = γ − ψ and β = ψ.4. Solving γ = α + β and ψ = β.4.10: Support region for Example 6. 3 3 γ −1 1 .6. 1 The support region for f z. 0 < β < 1 otherwise.10. β) = 4αβ. Random variables x and y have the joint PDF f x. 0. f z(γ ) = 0 γ 2 4(γ − ψ)ψ d ψ = γ 3 .w (γ . Let auxiliary variable w = h(x. 0 < α < 1. Conditional PDF Technique Since a conditional PDF is also a PDF.w is shown in Figure 6. ψ) = f x. 0. for 0 < γ < 1. 0 < γ − ψ < 1. y) = y. f z(γ ) = Otherwise. We ﬁnd J = so that f z. 1 0 1 = 1.10. Example 6. the above techniques apply to ﬁnd the conditional PDF for z = g (x. 3 For 1 < γ < 2. given event A.y (γ − ψ.
Depending on whether the event A is deﬁned on the range or domain of z = g (x.y (α. 8 consequently. i = 1. Find the PDF for z = x + y. If A is an event deﬁned for a region in the x − y plane.5 0 α 1 3α dβ d α = . y).7. ψ)) (6. β) = 24α. β)d α dβ = 0 0. given event A = {max(x. β) in region A. i. one of the following two methods may be used to determine the conditional PDF of z using the bivariate PDF technique. βi (γ .11(a) we ﬁnd P (A) = A f x. y) ≤ 0. given A. From Figure 6.TRANSFORMATIONS OF RANDOM VARIABLES 71 PDF f x.y (α.y (α. ψ)A) . y).4. and evaluate f z. we have f zA (γ A) = f z(γ ) . f x. Then evaluate the marginal conditional PDF f zA (γ A) = ∞ −∞ f z. the conditional PDF. J (αi (γ . βi ).31) where (αi .5}. ⎧ ⎨ f x. 0. . 0 < β < α < 0. (6. If A is an event deﬁned for an interval on z. First. and we wish to evaluate the conditional PDF for random variable z = g (x. introduce an auxiliary random variable w = h(x. Random variables x and y have joint PDF f x.30) ii.yA (α. β) and ψ = h(α. Then. ψA) = ∞ i=1 f x. βi (γ . .wA (γ .yA to evaluate the conditional PDF f zA as follows. 2. y).yA (αi (γ . .5 otherwise. ψA)d ψ. 0<β<α<1 otherwise. Solution. are the solutions to γ = g (α. . P (A) γ ∈ A.y . is computed by ﬁrst evaluating f z using the technique in this section. by the deﬁnition of a conditional PDF. f zA . (6. We begin by ﬁnding the conditional PDF for x and y. βA) = P (A) ⎩ 0. given that event A occurred. we will use the conditional PDF f x. ψ). ψ). β) = 3α..32) Example 6.wA (γ .
wA (γ . ψA) = Substituting.11(b).72 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS FIGURE 6.11(b). β) = α + β and ψ = h(α.4.5 24(γ − ψ)d ψ = 3(1 − γ 2 ). given A. we ﬁnd f z.5 < γ < 1 we have f zA (γ A) = 0 0.wA (γ . f x. The conditional PDF for z.7. 0. f zA (γ A) = 0. the only solution to γ = g (α.5γ γ −0. . Letting the auxiliary random variable w = h(x. For 0. The integration is easily carried out with the aid of Figure 6.wA (γ . ψA) . 0. 1 The support region for f z. ψA)d ψ.5 otherwise. The Jacobian of the transformation is J = 1 so that f z. y) = y. 0 < ψ < γ − ψ < 0.11: Support region for Example 6.wA is illustrated in Figure 6.5γ 24(γ − ψ)d ψ = 9γ 2 . β) = β is α = γ − ψ and β = ψ.yA (γ − ψ. can be found from f zA (γ A) = ∞ −∞ f z. ψA) = 24(γ − ψ).5 < γ < 1 we obtain f zA (γ A) = For γ < 0 or γ > 1. For 0.
(c ) f z(1/2). Answers: 0.3. 1/12.4. (d ) f z(3/2). 0. 3/4.y (α. Determine: (a) f z(−1). β) = e −α−β u(α)u(β).w (−1. (b) f z. (d ) f z(3/2). z = g (x). (b) f z(1/4).y (α.4. Random variables x and y have joint PDF f x. 6.1. (6.4. Random variables x and y have joint PDF f x. β) = 1. Determine: (a) f z. The ﬁrst. y) : x + y ≤ 1}.3. 1). 0 < β < 1 otherwise. Answers: 0. Random variables x and y have joint PDF f x. (c )FzA (0A). Answers: 0.33) . (d ) f zA (1/2A). Determine: (a) f zA (−3/ 2A). 1/4). Answers: 0.w (1. 1. 15/16. Random variable z = x and w = xy. 0.4.y (α. β) = 4αβ. 0. β) = 2β.2(α 2 + β).2. we write Fz(γ ) = P (z ≤ γ ) = P (g (x) ≤ γ ) = P (x ∈ A(γ )). 0 < β < 1 otherwise. 13/12. Drill Problem 6. Drill Problem 6. Drill Problem 6. is called the CDF technique.5 SUMMARY This chapter presented a number of different approaches to ﬁnd the probability distribution of functions of random variables. Random variables x and y have joint PDF f x.y (α. 0. 1). 1. (b) f z(−1/2). 0. 0 < α < 1. Two methods are presented to ﬁnd the probability distribution of a function of one random variable. Random variable z = x 2 y.TRANSFORMATIONS OF RANDOM VARIABLES 73 Drill Problem 6. Find: (a) f z. and the most general approach. Random variable z = x + y and w = y 2 . 0 < α < 1. 3/16. e −2 . 0 < β < 1 otherwise. (c ) f z(1/2).w (1. From the deﬁnition of the CDF. (b) f zA (−1/2A).4. Random variable z = x − y and event A = {(x. 0 < α < 1.7172.
a bivariate CDF as well as a joint PDF technique were presented. 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. The second approach. Using the joint PDF technique.w (γ .34) denote the distinct solutions to g (αi ) = γ .39) . i = 1. . called the PDF technique. J (αi .38) The ease of solution here involves transforming A(γ ) into proper limits of integration. ψ). (6. . where A(γ ) = {(x. y) : g (x. (6. y) and w = h(x. The ﬁrst approach. . . involves evaluating Fz(γ ) = P (z ≤ γ ) = P (g (x. a CDF technique.37) dFx.y (α. involves evaluating f z(γ ) = where αi = αi (γ ) = g −1 (γ ). The CDF for the RV z can then be found by evaluating the integral Fz(γ ) = A(γ ) (6. y) from jointly distributed random variables x and y. βi ) (6.74 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS where A(γ ) = {x : g (x) ≤ γ }. y) ∈ A(γ )). Typically. . .35) ∞ i=1 f x (αi (γ )) . To ﬁnd the joint probability distribution of random variables z = g (x. y) ≤ γ }. . . βi ) . y) created from jointly distributed random variables x and y using two approaches. (6.40) ∞ i=1 f x.36) (6. By partitioning A(γ ) into disjoint intervals. ψ) = where (αi (γ . Next. the PDF technique is much simpler to use than the CDF technique. 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. 2.y (αi . y) ≤ γ ) = P ((x. βi (γ . we evaluated the probability distribution of a random variable z = g (x. i = 1. β). the CDF Fz may be found using the CDF Fx . The second approach involves introducing an auxiliary random variable and using the PDF technique applied to two functions of two random variables. 2. g (1) (αi (γ )) (6. However. the joint PDF for z and w can be found as f z. ψ)).
.w . ⎨ Fx (α) = α 2 . 0. β) ∂β (6. β) = ∂h(α. the most difﬁcult aspect of the joint PDF technique is in solving the simultaneous equations. x − 1. βi ) = ψ. Suppose random variable x has the CDF ⎧ ⎪ 0. Typically. ﬁnd Fz.41) (6.43) where the indicated partial derivatives are assumed to exist.y by introducing an auxiliary random variable w = h(x. β) ∂β .42) (6. Given f x (α) = and z= Use the CDF technique to ﬁnd Fz. Let random variable x be uniform between 0 and 2 with z = exp(x). The Jacobian ∂g (α. y) to ﬁnd f z.5(1 + α). −1 < α < 1 otherwise. Find Fz using the CDF technique. β) ∂α ∂g (α. and then integrating to obtain the marginal PDF f z. 2. α<0 0≤α<1 1≤α and z = exp(−x). 3.TRANSFORMATIONS OF RANDOM VARIABLES 75 are the distinct solutions to the simultaneous equations g (αi . ⎪ ⎩ 1. y) from f x. Using the CDF technique. βi ) = γ and h(αi . 6. x<0 x > 0. 0. ∂h(α. x + 1. β) ∂α J (α. The joint PDF technique can also be used to ﬁnd the probability distribution for z = g (x.6 PROBLEMS 1.
⎧ ⎪ −1. Find Fz for z = x 2 using the CDF method. Given that x is uniformly distributed between −2 and 2. (d) E(z) using f z (compare with the answer to part a). 0<α<1 otherwise. ﬁnd: (a) E(z) using f x . so that z = u(x). (d) E(z) using f z (compare with the answer to part a). Random variable x represents the input to a fullwave rectiﬁer and z represents the output. Determine: (a) a. 0. The PDF of random variable x is 1 1 f x (α) = e − 2 (α+2) u(α + 2). so that z = x.76 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 4. ⎨ z= x. 9. Using the CDF method. 5. (b) Fz using the CDF method. x < −1 −1 ≤ x ≤ 2 x > 2. Random variable x has the PDF f x (α) = e −α u(α). Random variable x represents the input to a halfwave rectiﬁer and z represents the output. 8. ﬁnd: (a) E(z) using f x . 2α. 7. ⎪ ⎩ 2. ⎨ z= x. Suppose random variable x has the PDF f x (α) = Random variable z is deﬁned by ⎧ ⎪ −1. (b) Fz using the CDF method. 6. (b) Fz. determine Fz. (c) f z. Given f x (α) = and a 1 + α2 x ≤ −1 −1 < x < 1 x ≥ 1. . (c) f z. ⎪ ⎩ 1. 2 Find Fz using the CDF technique when z = x. Given that x is uniformly distributed between −2 and 2.
14. Find the PDFs for the following random variables: (a) z = x 3 . Suppose f x (α) = (1 + α 2 )/6. ⎪ ⎩ 1. ⎪ ⎩ x. and ⎧ ⎪ x. ⎪ ⎨ z = 2. ⎪ ⎪ ⎪ ⎩ 4. 0.5 ≤ x ≤ 0 0 < x.5 0.5 x ≥ 3.49). ⎨ Fx (α) = (α 2 + 2α + 1)/9. Let z = 1/x 2 . Random variable x has the CDF ⎧ ⎪ 0.5 ≤ x < 2. 13.25. ⎨ z = 0. x < 0. Random variable x has the PDF f x (α) = e −α u(α). 0<α<1 otherwise. −1 < α < 2 otherwise. Determine Fz. . ﬁnd f z using the CDF method. 0. Use the CDF technique to determine Fz. 0. Random variable x has the following PDF f x (α) = 4α 3 . ⎪ ⎪ 3. With z = 100 − 25x. 12.TRANSFORMATIONS OF RANDOM VARIABLES 77 10.5 1. (b) Fz using the PDF technique. The relationship between them is deﬁned by: ⎧ ⎪ 0. (b) z = (x − 1/2)2 .5.5 ≤ x < 1. α < −1 −1 ≤ α < 2 2≤α x < −0.5 ≤ x < 3.5 −0. ⎪ ⎪ ⎪ 1. Given that the input follows a Gaussian distribution with x ∼ G(2. 11. ﬁnd: (a) Fz using the CDF technique. Random variables x and z are the input and output of a quantizer. (c) z = (x − 1/4)2 .5 2.
Random variable x has the CDF 0. Find the PDF f z. 19. ⎧ ⎪ ⎨ α < −1 −1 ≤ α < 3 3 ≤ α.12. Find the CDF Fz. 15.78 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1 −2 −1 −1 1 2 3 x FIGURE 6. Assume the diodes are ideal. + 1K Ω + x 3K Ω z − − FIGURE 6. The voltage x in Figure 6. ⎪ ⎩ 1.12: Plot for Problem 15. Assume that g (x) is a second degree polynomial for x ≥ 0. 18. The voltage x in Figure 6. .13: Circuit for Problem 16.25(α + 1). 17. Find the PDF f z. ﬁnd the PDF f z. Fx (α) = 0. 16.13 is a random variable which is uniformly distributed from −1 to 2. Random variable x is Gaussian with mean ηx = 0 and standard deviation σx = 1. Assume the diode and the operational ampliﬁer are ideal. Assume the diode is ideal. Find the PDF f z.15 is a Gaussian random variable with mean ηx = 1 and standard deviation σx = 1. The voltage x in Figure 6. with RV z = g (x). With z = g (x) shown in Figure 6. and g (x) shown in Figure 6.16.14 is a Gaussian random variable with mean ηx = 0 and standard deviation σx = 3.
Random variable x has the following PDF f x (α) = Find Fz if z = x 2 .14: Circuit for Problem 17.15: Circuit for Problem 18. 0. −1 < α < 2 otherwise. 24. 0. 20.5). Random variable z = 4x + 2. Random variable x is uniform in the interval 0 to 12. 22. 0. Let z = cos(x). 23.TRANSFORMATIONS OF RANDOM VARIABLES + R + 79 x 2V 3V z − − FIGURE 6. Random variable x has the PDF f x (α) = 2(α + 1)/9. Find f z if: (a) f x (α) = 1/π.5δ(α − 0. 0<α<1 otherwise. 5K Ω 1K Ω x z FIGURE 6. 21. Find the PDF of z = 2x 2 using the PDF technique. α < π/2 otherwise. Find f z using the PDF technique. Find f z if z = 1/x 2 and x is uniform on −1 to 2. α + 0. .
Random variable w = v2 /R represents the power dissipated in a resistor of R with v volts across the resistor. ⎧ ⎪ 0. 28. 0<α<3 otherwise. Given that x has the CDF f x (α) = 8α/π 2 . 26.80 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS z = g( x ) 1.16: Transformation for Problem 19. given event A.4 −2 −1 −0. Random variable x has the PDF f x (α) = 2α/9. x. . Find the PDF of z = −2 ln(x) using the PDF technique. (b) 25. α<0 0≤α<1 1 ≤ α. Random variable z = (x − 1)2 and event A = {x : x ≥ 1/2}. 0. x 2. ⎪ ⎩ 1. Find (a) f w . Find the PDF of random variable z.5 1 2 x FIGURE 6. Random variable z= Using the PDF technique. (b) f wA if A = {v ≥ 0}. 27. Random variable x is uniform between −1 and 1. ﬁnd f z. 0. 0 < α < π/2 otherwise. ⎨ Fx (α) = α. A voltage v is a Gaussian random variable with ηv = 0 and σv = 2. x<0 x ≥ 0.
Given 9α 2 . ⎧ ⎪ ⎨ 0 < α < 0.5 < α < 1 otherwise. 34. (d) E(z2 ). 35. . ⎪ ⎩ 0. and z = x 2 . Determine f z using the PDF technique.5α(u(α) − u(α − 2)). If z = e .75(1 − α 2 ). Suppose RV x has PDF f x (α) = 0. Find a transformation g such that z = g (x) has PDF f z(γ ) = c γ 2 (u(γ ) − u(γ − 1)). 30. 32. Let f x (α) = 0. 0<α<1 otherwise. Using the PDF technique. Random variable x has an exponential distribution with mean one. 36. ﬁnd the PDF of z = x 2 . Suppose f x (α) = 2α. (c) E(z).5 0. ﬁnd f z if z = x and x is a standard Gaussian random variable. f x (α) = 3(1 − α 2 ).TRANSFORMATIONS OF RANDOM VARIABLES −x 81 29. (b) Fz. and z = −2 ln(x). Find f z if z = 1/x and f x (α) = π(α 2 1 . Determine f z using the PDF technique. and z = 8x 3 . ⎪ ⎩ 0. 0. 0. use the PDF technique to determine: (a) f zA (γ A) if A = {x : x > 2}. Determine: (a) f z. Given that random variable x has the PDF ⎧ ⎪ α. ⎨ f x (α) = 2 − α. −1 < α < 1 otherwise. 0<α<1 1<α<2 otherwise. Using the PDF technique. (b) f zB (γ B) if B = {z : z < 1/2}. + 1) 31. 33.
5.5 − 0. ⎨ f z(γ ) = 0.25γ . 4 4 Random variable z = g (x). −2 < α < 1 1 < α < 2.5 − 0.25α. 3 z = g(x) 2 1 x −1 0 1 2 FIGURE 6. (a) Find the transformation z = g (x) so that ⎧ ⎪ 1 − 0.82 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS 37.17: Transformation for Problem 40. 0. Random variable x has the PDF ⎧ ⎪ 0. 39. −1 < α < 0 0<α<2 otherwise. 40. Random variable x has PDF 1 1 f x (α) = δ(α + 1) + δ(α) + (u(α) − u(α − 2)). ﬁnd f z. 38.17. 41. 1/4. ⎪ ⎩ 0. P (x = −2) = P (x = 2) = 0.25. (b) Determine f z if z = (2x + 2)u(x). . in addition. ⎨ f x (α) = 0. 0 ≤ α < 2π otherwise. ⎪ ⎩ 0.25γ . Find the PDF f z. Random variable x has PDF f x (α) = α2 (u(α + 1) − u(α − 2)). with g shown in Figure 6. 0<γ <1 1<γ <2 otherwise. If z = 1/x. Find f z if z = sin(x) and f x (α) = 1/(2π). Let random variable x have the PDF f x (α) = 1/12.
Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 . 0<β<α<1 otherwise. 46. The PDF for random variable x is f x (α) = e −α u(α). If z = x/y. 42. 0. 44. 0. 0.18: Transformation for Problem 41. With z = x + y. With z = e −x . determine: (a) f zA (γ A) where A = {x : x > 2}.18. with g shown in Figure 6. With g denoting the equivalent conductance of a parallel connection of R1 and R2 . β) = 0. 43. ﬁnd the PDF fr using convolution. 45. (b) f zA with A = {x > 0}.y (α. write an expression(s) for Fz(γ ) (do not solve. just write the integral(s) necessary to ﬁnd Fz.y (α. Resistors R1 and R2 have values r 1 and r 2 which are independent RVs uniformly distributed between 1 and 2 . ﬁnd the PDF f g using convolution. β) = e −α . (b) f zB (γ B) where B = {z : z < 1/2}.TRANSFORMATIONS OF RANDOM VARIABLES z = g( x ) 2 1 x 83 −1 0 1 2 FIGURE 6. ﬁnd Fz when x and y have the joint PDF f x. Use the PDF technique to ﬁnd: (a) f z. Suppose the voltage v across a resistor is a random variable with PDF f v (α) = 6α(1 − α). With r denoting the equivalent resistance of a series connection of R1 and R2 . . 0<β<α otherwise. 0<α<1 otherwise.25α. 47. The joint PDF of x and y is f x. Random variable z = g (x).
0. 0<β<α<1 otherwise. suppose that v and r are independent random variables. Let z = x 2 and w = xy. The joint PDF for random variables x and y is f x. 0<β<α<1 otherwise. The joint PDF for random variables x and y is f x. 0 < β otherwise. z = x and w = x 2 + y 2 . ﬁnd f z. 1/α. (b) f zA . The joint PDF for random variables x and y is f x.y (α. Let A = {(x. 0. (b) f z. 0 < α.y (α. β) = With z = y/x. Find: (a) f z. 0.w .y (α. 0 < α. 0 < α < 1. Let z = 2x + y and w = x + 2y.wA .w .y (α. 51. 0.w . The joint PDF for random variables x and y is f x. 50. 0 < β < 1 otherwise. β) = 8αβ. β) = α + β. 0 < β < 2 otherwise.5βe −α . y) : x > y}. 0. 0 < β < 1 otherwise. Find the joint PDF for random variables z = x + y and w = x − y. Moreover.y (α. β) = 3(α 2 + β 2 ). 0 < α 2 + β 2 < 1. p = v2 /r . 0. Find f z. 49.84 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS and that resistance r is a random variable with PDF fr (β) = 1/12. The joint PDF for random variables x and y is f x. Find f z. 0. The joint PDF for random variables x and y is f x. β) = Let z = x/y and w = y. 53. 94 < β < 106 otherwise. 48. β) = 4αβ. 52. 0 < α < 1. Find: (a) f z. Find the PDF for power.y (α. 0. .
57. . 0 < β otherwise. β) = Let z = x 2 y. The joint PDF for random variables x and y is f x. 0 < β < π otherwise.y (α. β) = 0. β) = α sin(β). Find f z.25(α + β). 0 < α < 1. (b) f z. Let z = x + y and w = x/(x + y). Find f z. 0. α < 1. 0. β < 1 otherwise. The joint PDF for random variables x and y is f x.w .TRANSFORMATIONS OF RANDOM VARIABLES 85 54. Let z = x 2 and w = cos(y). The joint PDF for random variables x and y is f x. Let z = xy and w = y/x. 0 < α < 1.y (α.w . 0 < α. 0.w using the joint CDF technique. 12αβ(1 − α). Find f z.y (α. 0. 0 < β < 1 otherwise. 55.y (α. 56. Find: (a) f z. The joint PDF for random variables x and y is f x. β) = e −α−β .
.
000000 (Continued ) +2 0.000000 0.4 0.912960 0.443705 0.590490 0.773781 0.988496 1.913862 1.077760 0.991440 0.000000 0.256218 0.999680 0.45 0.050328 0.15 0.929809 1.196874 0.998971 1.999853 0.05 0.998365 1.1 0.2 0.25 0.997773 0.000000 0.812500 +4 1.950030 1.987205 1.998617 +1 0.999936 1.835210 0.031250 +0 0.999991 0.116029 0.945978 0.981547 0.544300 0.999990 0.327680 0.5 p 0.35 0.187500 n = 10 +1 0.973388 0.998842 0.999924 0.632813 0.764831 0.999970 0.918540 0.736099 0.000000 0.000000 0.168070 0.87 APPENDIX A Distribution Tables TABLE A.737280 0.336960 0.3 0.999023 0.999991 +3 0.000000 0.05 0.000000 0.000000 0.593127 0.15 k 0 0 0 0 0 0 0 0 0 0 k 0 5 0 5 0 5 +0 0.868780 0.1 0.836920 0.997570 0.994748 0.999865 +2 0.500000 +3 0.348678 0.968750 .000000 +4 0.989760 0.942080 0.993280 0.990126 1.1: Bernoulli CDF for n = 5 and n = 10 n=5 p 0.820197 0.598737 0.977408 0.999540 0.969220 0.428415 0.984375 0.682560 0.896484 0.999997 0.237305 0.528220 0.
000000 1.649611 0.463291 0.549043 0.999659 0.05 k 0 5 10 0.833761 0.000000 .513827 0.989258 +4 0.244025 0.525593 0.000000 1.905066 0.751495 0.945313 +3 0.952651 0.000000 0.999895 0.999460 0.2 0.054688 0.999996 1.999136 0.382783 0.623047 +1 0.167290 0.261607 0.085954 0.2: Bernoulli CDF for n = 15 n = 15 p 0.023257 0.000000 1.000000 0.999385 1.006047 0.375810 0.945238 0.993631 0.382281 0.972608 0.999584 0.999856 0.989408 0.999947 1.000000 +4 0.677800 0.999922 0.829047 0.000000 1.000000 0.999994 0.997750 1.994533 1.1 0 5 10 +0 0.849732 0.999972 0.987705 0.995498 0.000000 +3 0.998322 0.980272 0.000000 0.998410 0.000977 0.376953 0.775875 0.999970 0.046357 0.000000 0.1: Bernoulli CDF for n = 5 and n = 10 (Continued) n = 10 p 0.000000 +1 0.149308 0.973976 0.999023 TABLE A.999996 0.4 0.999966 1.963800 1.633103 0.000000 0.995179 0.002533 0.879126 0.056314 0.171875 0.010742 0.504405 0.25 0.815939 0.999689 1.967206 1.828125 +2 0.099560 0.987280 1.88 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.266038 0.45 0.000000 +2 0.205891 0.996494 0.898005 0.999999 0.3 0.738437 0.35 0.5 k 0 5 0 5 0 5 0 5 0 5 0 5 0 5 +0 0.921873 0.107374 0.013463 0.028248 0.999997 1.944444 0.
999988 0.15 k 0 5 10 0.999991 0.005172 0.886769 0.000488 0.999721 0.3 0 5 10 0.2 0 5 10 0.080181 0.999521 0.999999 0.983190 0.982422 +2 0.035184 0.000000 0.990652 0.461287 0.999919 1.059235 0.172696 0.999885 0.648162 0.000000 0.995807 1.982700 0.398023 0.943380 0.904953 0.996347 1.000127 0.25 0 5 10 0.010652 0.923071 0.999328 0.090502 0.000031 0.35 0 5 10 0.999996 0.000000 0.087354 0.4 0 5 10 0.999999 0.2: Bernoulli CDF for n = 15 (Continued) n = 15 p 0.000000 0.686486 0.999943 0.167126 0.957806 0.403216 0.318586 0.851632 0.260760 0.999999 0.042421 0.998893 0.515491 0.999992 1.004748 0.150879 0.940765 +1 0.126828 0.014179 0.027114 0.500000 0.997169 0.721621 0.035268 0.754842 0.938295 0.996394 1.999205 1.236088 0.966167 0.984757 1.061734 0.000000 0.001562 0.999994 0.45 0 5 10 0.696381 0.786897 0.003693 0.303619 0.949987 0.000470 0.000000 0.974534 0.999887 1.996307 +3 0.564282 0.999999 0.000000 0.604225 0.013363 0.987557 1.995760 1.000000 0.000000 0.998072 0.999975 0.999390 1.981941 0.999969 .653504 0.999215 1.993673 0.120399 0.999988 0.000000 0.DISTRIBUTION TABLES 89 TABLE A.017578 0.452160 0.296868 0.835766 0.999908 0.001692 0.818240 0.938949 0.609813 0.000000 0.849121 0.868857 0.351943 0.217278 0.000000 0.5 0 5 10 +0 0.999879 0.999512 +4 0.822655 0.
999752 1.998721 1.999999 +3 0.804208 0.000000 1.1 0 5 10 15 0.012118 0.967857 0.993985 0.982855 0.35 0 5 10 15 +0 0.994079 0.002133 0.024313 0.000000 0.617173 0.035483 0.000000 0.091260 0.000000 1.000000 0.90 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.762378 0.000000 1.000000 1.959075 0.000000 1.000000 1.3: Bernoulli CDF for n = 20 n = 20 p 0.999999 1.999671 1.175558 0.999957 1.946833 0.000000 1.999940 1.2 0 5 10 15 0.867047 0.003171 0.000000 1.676927 0.999994 +2 0.999689 1.999961 1.999985 1.000000 0.999950 +1 0.990018 0.000000 0.011529 0.735840 0.000000 0.044376 0.999998 1.956825 0.000000 0.829847 0.245396 0.000000 1.000000 1.999437 1.225156 0.997426 1.878219 0.601027 0.980421 0.000000 0.608010 0.416371 0.986136 0.000000 0.118197 0.996058 1.107087 0.999970 1.000000 0.999739 1.237508 0.999995 1.414842 0.999966 1.000000 0.000000 0.000000 +4 0.000000 0.999997 1.391747 0.997614 1.000000 0.404896 0.999816 1.000000 0.038760 0.000798 0.886669 0.000000 0.952038 0.358486 0.007637 0.000000 0.932692 0.629648 0.000000 0.785782 0.000000 0.000000 1.25 0 5 10 15 0.994862 0.984098 1.000000 0.913307 0.999065 1.000000 0.000000 1.997405 1.999999 0.3 0 5 10 15 0.411449 0.000000 0.924516 0.999994 0.898188 0.998671 1.988747 0.999996 1.999898 1.999993 1.772272 0.000000 0.000000 0.999999 1.05 k 0 5 10 15 0.121577 0.647725 0.000000 0.000000 1.978065 0.000181 0.000000 0.416625 0.999584 1.15 0 5 10 15 0.998479 1.000000 .000000 0.069175 0.206085 0.000000 0.000000 1.
000037 0.818731 0.000201 0.000000 0.000020 0.4: Poisson CDF for λ = 0.999214 1.878099 0.999943 0.050952 0.000524 0.999980 +4 0. .844195 0.7 k 0 0 0 0 0 5 0 5 0 5 +0 0.976885 1.998388 1.999997 0.4 0.125599 0.999828 1.996642 1.999964 0.1 0.000000 0.872479 0.000927 0.904837 0. 1.999910 +1 0.000000 0.999723 0.055334 0.496585 0.909796 0.000000 (Continued ) .978586 0.748278 0.999984 0.999224 0.999999 TABLE A. 4.996400 0.998712 +2 0.999999 0.998469 0.994247 1. 1.595599 0.869235 0.999799 +3 0.588099 0.999734 0.5 0.1. .985612 1.414306 0.999991 +2 0. .2.670320 0.415893 0. .994091 +1 0.129934 0.45 0 5 10 15 0.942341 0.750711 0.943474 0.251722 0.250011 0.998248 1.606531 0.755337 0.000000 0.005909 0.004933 0.999997 0. .938448 0.941966 0.5 λ 0. 2.999998 0.3 0.999999 +3 0.992074 0.000000 0.000000 0.252006 0.993534 1.5.000000 +4 1.057659 0.3: Bernoulli CDF for n = 20 (Continued) n = 20 p 0.999683 0. .999986 0.998852 0. 0.591361 0.999953 0.963064 0.999845 0.978971 0.020695 0.DISTRIBUTION TABLES 91 TABLE A.001288 0.868412 0.000000 0.000006 0. .000001 0.5 0 5 10 15 +0 0.979305 0.015961 0.000000 0.548812 0.003611 0.995321 0.131588 0.740818 0.6 0.000000 0.965858 0. .993566 1.000111 0.999939 0.4 k 0 5 10 15 0.999961 0.000000 0.411901 0.999996 0.999995 0.999605 1.018863 0.2 0.982477 0.
999763 0.999972 0.948866 0.891178 0.831051 0.996340 1.238103 0.5 0 5 10 15 .735759 0.999656 0.999074 0.991868 0.725445 0.998981 0.5.999708 0.952577 0.406006 0.937143 0.628837 0. .049787 0.889326 0.406570 0.985813 0.676676 0.999997 0. .999748 1.000000 0.018316 0. 4.543813 0.135335 0.000000 0.082085 0.287297 0.557825 0.000000 0.091578 0.423190 0.934358 0.999929 0.702930 0.999723 1.342296 0.532104 0.000000 0.995466 0.957979 0.92 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.536633 0.996685 0.966491 0.981424 0.999999 0.999924 0. 0.998860 1.999711 0.000000 0.772482 0.000000 0. . 1.999995 0.999957 0. .857614 0.999926 1.000000 0.199148 0.999085 0.8 0.857123 0.808792 0.999999 0.9 1 1.999987 0.999981 0.996197 0.959743 0.223130 0. .997160 0. 2.173578 0.947347 0.999195 0.135888 0.757576 0.919699 0.999938 0.990126 0. .999406 0. 1.916082 0.999954 0.997596 0.808847 0.998897 0.999984 0.999996 0.999980 +1 0.2.995544 0.999830 0.999980 0.990920 1.647232 0.999917 0.981012 0.998589 1.5 (Continued) λ 0.367879 0.5 0 5 10 4 0 5 10 4.999990 0.934712 0.997656 1.5 k 0 5 0 5 0 5 0 5 0 5 0 5 10 +0 0.999999 +3 0.1.999996 0.433470 0.986541 1. .998903 0.999998 0.993331 0.030197 0.011109 0. .983436 0.4: Poisson CDF for λ = 0.5 2 2.982907 0.973261 0.000000 +4 0.999979 0.815263 0.449329 0.999726 0.999998 0.999816 0.320847 0.913414 0.000000 3 0 5 10 3.061099 0.978637 0.988095 0.999924 0.785130 0.995753 0.999995 +2 0.
088376 0.986305 0.791573 0.920759 0.000000 (Continued ) 5.957334 0.999302 0.809485 0.931906 0.999956 0.916076 0.598714 0.999999 0.528919 0.000553 0.997044 0.997593 0.830496 0.989740 0.847238 0.999949 0. .223672 0.999995 0.999995 0. 8.081765 0.661967 0.999984 0.357518 0.999042 0.743980 0.999937 0.998600 0.132062 0.998041 0.378155 0.992900 0.999996 +3 0.5 0 5 10 15 6 0 5 10 15 6.449711 0.776408 0.011276 0.999870 0.017351 0.999987 +2 0.999210 0.994547 0.007295 0.445680 0.004701 0.994283 0.999638 0.974749 0.999825 0.862238 0.006738 0.983973 0.5 0 5 10 15 7 0 5 10 15 7.999981 0.999931 0. .729091 0.999800 0.061969 0.004087 0.999570 0.995392 0.894357 0.5.020257 0.059145 0.987189 0.526524 0.DISTRIBUTION TABLES 93 TABLE A. .999995 0.026564 0.991173 0.866628 0.999491 0.995549 0.998840 0.5 0 5 10 15 20 .606303 0.978435 0.996372 0.933161 0.265026 0.989012 0.998315 0.901479 0.999943 0.043036 0.999961 +1 0.968172 0.686036 0.5: Poisson CDF for λ = 5.001503 0.000912 0.172992 0.877384 0.300708 0.999999 0.5 λ 5 k 0 5 10 15 +0 0.999401 0.999999 +4 0.151204 0.002479 0.000000 0.973000 0.999774 1.999697 0.999980 0.672758 0.440493 0.946223 0.999889 1.369041 0.241436 0.979908 0. .999849 0.762183 0.524639 0.285057 0.997981 0.111850 0.957379 0.966120 0.946650 0.029636 0.124652 0. 5.201699 0.040428 0.999982 0.615961 0.
521826 0.000499 0.993387 0. 13 λ 9 k 0 5 10 15 20 +0 0.5 0 5 10 15 20 TABLE A.523105 0.999921 +2 0.040263 0.986167 0.014860 0.583040 +1 0.977964 0.587408 0.021226 0.5.998038 0.001933 0. 9.115691 0.054964 0.991769 0.998406 0. .999971 +3 0.006232 0.815886 0.220221 0.999975 0.000075 0.936203 0.848662 0.909083 0.999465 0.996282 0. 5.999933 0.999639 0.898136 0.999855 0.999979 0.009283 0.099632 0.791556 +3 0.067086 0.001234 0.864464 +4 0.074364 0.013754 0.999747 0.926149 0.452961 0.029253 0.982273 0.716624 0.256178 0.995716 0.149597 0.958534 0.088528 0.5: Poisson CDF for λ = 5.999906 0.948589 0.965819 0. .6: Poisson CDF for λ = 9. 10.999997 8.982743 0.5 0 5 10 15 20 10 0 5 10 .999789 +1 0.875773 0.999991 0.332820 0.999996 0.991072 0.030109 0.998703 0.164949 0.323897 0.999999 0.705988 0.999350 0.592547 0.000123 0.042380 0.94 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.763362 0.751990 0.988894 0.130141 0.191236 0.999990 +4 0.972575 0.455653 0.003019 0.010336 0.391824 0.268663 0.652974 0.000335 0.385597 0. 11.004164 0.5.836430 0.5 (Continued) λ 8 k 0 5 10 15 20 +0 0.000045 0.999141 0.888076 0. .999967 0. 12.999944 0.000203 0.966527 0.999825 0.002769 0. .206781 0. 8.645328 0.000786 0.916542 9.803008 0.998944 0.999561 0.940008 0.313374 0.999989 0.457930 0.994680 0.996998 0.997574 0.696776 +2 0.
DISTRIBUTION TABLES
95
TABLE A.6: Poisson CDF for λ = 9, 9.5, 10, 11, 12, 13 (Continued)
λ
k 15 20
+0 0.951260 0.998412 0.000017 0.037520 0.459889 0.907396 0.995329 0.999918 0.000006 0.020341 0.347229 0.844416 0.988402 0.999692 0.000002 0.010734 0.251682 0.763607 0.974988 0.999034
+1 0.972958 0.999300 0.000200 0.078614 0.579267 0.944076 0.997748 0.999967 0.000080 0.045822 0.461597 0.898709 0.993935 0.999867 0.000032 0.025887 0.353165 0.835493 0.985919 0.999548
+2 0.985722 0.999704 0.001211 0.143192 0.688697 0.967809 0.998958 0.999987 0.000522 0.089505 0.575965 0.937034 0.996953 0.999944 0.000223 0.054028 0.463105 0.890465 0.992378 0.999796
+3 0.992813 0.999880 0.004916 0.231985 0.781291 0.982313 0.999536 0.999995 0.002292 0.155028 0.681536 0.962584 0.998527 0.999977 0.001050 0.099758 0.573045 0.930167 0.996028 0.999911
+4 0.996546 0.999953 0.015105 0.340511 0.854044 0.990711 0.999801 0.999998 0.007600 0.242392 0.772025 0.978720 0.999314 0.999991 0.003740 0.165812 0.675132 0.957331 0.998006 0.999962
11
0 5 10 15 20 25
12
0 5 10 15 20 25
13
0 5 10 15 20 25
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18
λ 14
k 0 5 10
+0 0.000001 0.005532 0.175681
+1 0.000012 0.014228 0.260040
+2 0.000094 0.031620 0.358458
+3 0.000474 0.062055 0.464448
+4 0.001805 0.109399 0.570437 (Continued )
96
ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ
k 15 20 25 30
+0 0.669360 0.952092 0.997392 0.999940 0.000000 0.002792 0.118464 0.568090 0.917029 0.993815 0.999803 0.000000 0.001384 0.077396 0.466745 0.868168 0.986881 0.999433 0.000000 0.000675 0.049124 0.371454 0.805481 0.974755 0.998552
+1 0.755918 0.971156 0.998691 0.999974 0.000005 0.007632 0.184752 0.664123 0.946894 0.996688 0.999910 0.000002 0.004006 0.126993 0.565962 0.910773 0.992541 0.999724 0.000001 0.002062 0.084669 0.467738 0.861466 0.984826 0.999253
+2 0.827201 0.983288 0.999365 0.999989 0.000039 0.018002 0.267611 0.748859 0.967256 0.998284 0.999960 0.000016 0.010000 0.193122 0.659344 0.941759 0.995895 0.999869 0.000007 0.005433 0.135024 0.564023 0.904728 0.991166 0.999626
+3 0.882643 0.990672 0.999702 0.999996 0.000211 0.037446 0.363218 0.819472 0.980535 0.999139 0.999983 0.000093 0.021987 0.274511 0.742349 0.963314 0.997811 0.999940 0.000041 0.012596 0.200873 0.654958 0.936704 0.995016 0.999817
+4 0.923495 0.994980 0.999864 0.999998 0.000857 0.069854 0.465654 0.875219 0.988835 0.999582 0.999993 0.000400 0.043298 0.367527 0.812249 0.977685 0.998869 0.999973 0.000185 0.026125 0.280833 0.736322 0.959354 0.997273 0.999913
15
0 5 10 15 20 25 30
16
0 5 10 15 20 25 30
17
0 5 10 15 20 25 30
DISTRIBUTION TABLES
97
TABLE A.7: Poisson CDF for λ = 14, 15, 16, 17, 18 (Continued)
λ 18
k 0 5 10 15 20 25 30
+0 0.000000 0.000324 0.030366 0.286653 0.730720 0.955392 0.996669
+1 0.000000 0.001043 0.054887 0.375050 0.799124 0.971766 0.998187
+2 0.000003 0.002893 0.091669 0.468648 0.855090 0.982682 0.999040
+3 0.000018 0.007056 0.142598 0.562245 0.898890 0.989700 0.999506
+4 0.000084 0.015381 0.208077 0.650916 0.931740 0.994056 0.999752
TABLE A.8: Marcum’s Q function for α = 0, 0.01, . . . , 1.99
α 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65
+0.00 0.500000 0.480061 0.460172 0.440382 0.420740 0.401294 0.382089 0.363169 0.344578 0.326355 0.308538 0.291160 0.274253 0.257846
+0.01 0.496011 0.476078 0.456205 0.436441 0.416834 0.397432 0.378281 0.359424 0.340903 0.322758 0.305026 0.287740 0.270931 0.254627
+0.02 0.492022 0.472097 0.452242 0.432505 0.412936 0.393580 0.374484 0.355691 0.337243 0.319178 0.301532 0.284339 0.267629 0.251429
+0.03 0.488033 0.468119 0.448283 0.428576 0.409046 0.389739 0.370700 0.351973 0.333598 0.315614 0.298056 0.280957 0.264347 0.248252
+0.04 0.484046 0.464144 0.444330 0.424655 0.405165 0.385908 0.366928 0.348268 0.329969 0.312067 0.294598 0.277595 0.261086 0.245097 (Continued )
149170 0.153864 0.063008 0.091759 0.200454 0.35 1.211855 0.035148 0.040929 0.111233 0.095098 0.206108 0.060571 0.046479 0.235762 0.051551 0.131357 0. 0.214764 0.146859 0.171056 0.030054 0.70 1.30 1.161087 0.02 0.030742 0.192150 0.176186 0.142310 0.70 0.061780 0.025588 +0.052616 0.038364 0.082264 0.03 0.115070 0.50 1.098525 0.088508 0.033625 0.023295 . .133500 0.096801 0.105650 0.184060 0.00 0. .039204 0.04 0.137857 0.093418 0. 1.073529 0.121001 0.068112 0.00 1.173609 0.40 1.048457 0.238852 0.083793 0.032884 0.10 1.119000 0.059380 0.232695 0.113140 0.98 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.25 1.151505 0.047460 0.050503 0.041815 0.65 1.065522 0.125072 0.99 (Continued) α 0.85 1.85 0.074934 0.027429 0.144572 0.109349 0.035930 0.194894 0.077804 0.069437 0.178786 0.217695 0.024998 +0.023852 +0.163543 0.140071 0.15 1.95 1.186733 0.226627 0.042716 0.055917 0.156248 0.05 1.072145 0.058208 0.90 0.95 +0.079270 0.80 1.103835 0.01 0.203269 0.024419 +0.040059 0.75 1.085344 0.129238 0.135666 0.049471 0.036727 0.158655 0.034379 0.064256 0. .166023 0.127143 0.90 1.208970 0.8: Marcum’s Q function for α = 0.037538 0.053699 0.070781 0.123024 0.028067 0.75 0.066807 0.029379 0.100273 0.031443 0.054799 0.241964 0.026803 0.189430 0. .045514 0.117023 0.220650 0.032157 0.107488 0.028716 0.102042 0.20 1.181411 0.057053 0.090123 0.080757 0.168528 0.229650 0.043633 0.223627 0.55 1.197662 0.80 0.026190 0.01.076359 0.45 1.60 1.086915 0.044565 0.
001183 0.015386 0.000816 0.DISTRIBUTION TABLES 99 TABLE A.011011 0.003573 0.017864 0.005868 0.004661 0.021692 0.004269 0.005543 0.019226 0.30 +0.011911 0.000419 (Continued ) .015003 0. .022750 0.002118 0.001223 0.008424 0.000434 +0.000450 +0.012545 0.012224 0.010444 0.000845 0.007976 0.001538 0.000483 +0.001035 0.004397 0.006947 0.20 3.003072 0.02 0.002186 0.003793 0.003681 0.009642 0.007549 0.45 2.001926 0.020182 0.00 0.001695 0.000874 0.008198 0.30 2.002635 0.001807 0.70 2.014262 0.004940 0.000736 0.002256 0.25 2.005703 0. .001489 0.016586 0.005234 0.002890 0.001144 0.25 3.002980 0.15 2.03 0.021178 0. 3.000641 0.002327 0.002718 0.014629 0.003467 0.000789 0.001306 0.020675 0.004025 0.007344 0.05 2.004799 0.006210 0.001988 0.000664 0.000519 0.001264 0.018763 0.04 0.013553 0.016177 0.002052 0.000936 0.000467 +0.40 2.50 2.000538 0.000598 0.006037 0.00 2.003907 0.011604 0.009903 0.000619 0.017429 0.000968 0.000577 0.001001 0. 3.15 3.018309 0.004145 0.001441 0.001866 0.55 2.022216 0.009387 0.015778 0.005386 0.007760 0.90 2.001641 0.008656 0.002803 0.000762 0.000904 0.003167 0.006569 0.003364 0.001589 0.017003 0.10 3.001107 0.000687 0.95 3.003264 0.008894 0.01 0.001070 0.01.007143 0. .012874 0.9: Marcum’s Q function for α = 3.80 2.001350 0.004527 0.20 2. .00 3.002555 0.011304 0.000501 0.000557 0.010170 0.006756 0.85 2.006387 0.75 2.99 α 2.10 2.001750 0.65 2.001395 0.002401 0.35 2.05 3.013903 0.009137 0.010724 0.60 2.019699 0.000711 0.005085 0.002477 0.013209 0.
000142 0.000200 0.000104 0.000131 0.000280 0. .000165 0.80 3.55 3.000216 0. .000270 0.000057 0.000224 0.9: Marcum’s Q function for α = 3.000048 0.000291 0.000208 0.000036 +0.02 0.000193 0.95 +0.000337 0.000059 0.000041 0.000126 0.000260 0.000062 0. 3.03 0.000242 0.000136 0. .000233 0.01 0.70 3.000070 0.000078 0.000033 .000302 0.000054 0.000064 0.000075 0.000052 0.000100 0.00 0.000117 0.000325 0.000179 0.000404 0.000044 0.000185 0.000390 0.000092 0.35 3.60 3.000067 0.40 3.50 3.000147 0.99 (Continued) α 3.000042 0.000088 0.000376 0. .000362 0.65 3. 3.000034 +0.000251 0.000159 0.000050 0.90 3.000085 0.85 3.000082 0.100 ADVANCED PROBABILITY THEORY FOR BIOMEDICAL ENGINEERS TABLE A.000313 0.45 3.000037 +0.000046 0.000121 0.000153 0.000096 0.000350 0.000072 0.75 3.000108 0.000039 +0.01.000172 0.04 0.000112 0.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.