Elements of Probability Theory: 2.1 Probability, Random Variables and Random Matrices

Chapter 2
Elements of Probability Theory
In this chapter, we formally review some basic concepts of probability theory.

Most of this material is standard and available in classical references, such as
[108, 189, 319]; more advanced material on multivariate statistical analysis can
be found in [22]. The definitions introduced here are instrumental to the study of
randomized algorithms presented in subsequent chapters.
2.1 Probability, Random Variables and Random Matrices

2.1.1 Probability Space
Given a sample space and a -algebra S of subsets S of (the events), a probability P R {S} is a real-valued function on S satisfying:
1. P R {S} [0, 1];
2. P R {} = 1;
3. If the events Si are mutually exclusive (i.e., Si Sk = for i = k), then

Si =
P R {Si }
PR
iI
iI
where I is a countable1 set of positive integers.

The triple (, S, P R {S}) is called a probability space.
A discrete probability space is a probability space where is countable. In this
case, S is given by subsets of and the probability P R : [0, 1] is such that

P R {w} = 1.
1 By
countable we mean finite (possibly empty) or countably infinite.
R. Tempo et al., Randomized Algorithms for Analysis and Control of Uncertain Systems,
Communications and Control Engineering, DOI 10.1007/978-1-4471-4610-0_2,
Springer-Verlag London 2013
2 Elements of Probability Theory
2.1.2 Real and Complex Random Variables

We denote with R and C the real and complex field respectively. The symbol F is
also used to indicate either R or C. A function f : R is said to be measurable
with respect to a -algebra S of subsets of if f 1 (A) S for every Borel set
A R.
A real random variable x defined on a probability space (, S, P R {S}) is a
measurable function mapping into Y R, and this is indicated with the shorthand
notation x Y. The set Y is called the range or support of the random variable x.
A complex random variable x C is
a sum x = xR + j xI , where xR R and xI R
.
are real random variables, and j = 1. If the random variable x maps the sample
space into a subset [a, b] R, we write x [a, b]. If is a discrete probability
space, then x is a discrete random variable mapping into a countable set.
Distribution and Density Functions
of a random variable x is defined as
The (cumulative) distribution function (cdf)
.
Fx (x) = P R {x x}.
The function Fx (x) is nondecreasing, right continuous (i.e., Fx (x) = limzx+ Fx (z)),
and Fx (x) 0 for x , Fx (x) 1 for x . Associated with the concept
of distribution function, we define the percentile of a random variable

x = inf x : Fx (x) .
For random variables of continuous type, if there exists a Lebesgue measurable
function fx (x) 0 such that
x
fx (x) dx
Fx (x) =
then the cdf Fx (x) is said to be absolutely continuous, and

fx (x) =
dFx (x)
dx
holds except possibly for a set of measure zero. The function fx (x) is called the
probability density function (pdf) of the random variable x.
For discrete random variables, the cdf is a staircase function, i.e. Fx (x) is constant
except at a countable number of points x1 , x2 , . . . having no finite limit point. The
total probability is hence distributed among the mass points x1 , x2 , . . . at which
the jumps of size
.
fx (xi ) = lim Fx (xi + ) Fx (xi ) = P R {x = xi }
0
occur. The function fx (xi ) is called the mass density of the discrete random variable x. The definition of random variables is extended to real and complex random
matrices in the next section.
2.1 Probability, Random Variables and Random Matrices
2.1.3 Real and Complex Random Matrices

Given n random variables x1 , . . . , xn , their joint distribution is defined as
.
Fx1 ,...,xn (x1 , . . . , xn ) = P R {x1 x1 , . . . , xn xn }.
When the above distribution is absolutely continuous, we can define the joint density
function fx1 ,...,xn (x1 , . . . , xn )
. n Fx1 ,...,xn (x1 , . . . , xn )
.
fx1 ,...,xn (x1 , . . . , xn ) =
x1 xn
The random variables x1 , . . . , xn are said to be independent if
Fx1 ,...,xn (x1 , . . . , xn ) =
Fxi (xi )
i=1
where Fxi (xi ) = P R {xi xi }.

A real random matrix X Rn,m is a measurable function X : Y Rn,m .
That is, the entries of X are real random variables [X]i,k for i = 1, . . . , n and k =
1, . . . , m. A complex random matrix X Cn,m is defined as the sum X = XR + j XI ,
where XR and XI are real random matrices. A random matrix is discrete if its entries
are discrete random variables.
The distribution function FX (X) of a real random matrix X is the joint cdf of the
entries of X. If X is a complex random matrix, then its cdf is the joint cdf of XR
and XI . The pdf fX (X) of a real or complex random matrix is analogously defined as
the joint pdf of the real and imaginary parts of its entries. The notation X fX (X)
means that X is a random matrix with probability density function fX (X).
Let X Fn,m be a real or complex random matrix (of continuous type) with pdf
fX (X) and support Y Fn,m . Then, if Y Y, we have

P R {X Y } = fX (X) dX.
Y
Clearly, P R {X Y} = Y fX (X) dX = 1. When needed, to further emphasize that

the probability is relative to the random matrix X, we explicitly write P RX {X Y }.
2.1.4 Expected Value and Covariance

Let X Y Fn,m be a random matrix and let J : Fn,m Rp,q be a Lebesgue
measurable function. The expected value of the random matrix J (X) is defined as

.
J (X)fX (X) dX
EX J (X) =
Y
where Y is the support of X. We make use of the symbol EX (J (X)) to emphasize

the fact that the expected value is taken with respect to X. The suffix is omitted when
clear from the context.
10
If X Fn,m is a discrete random matrix with countable support Y = {X1 , X2 , . . .},

Xi Fn,m and Y Y, then

P R {X Y } =
fX (Xi ) =
P R {X = Xi }.
Xi Y
Xi Y
The expected value of J (X) is defined as

.
J (Xi )fX (Xi ).
E J (X) =
Xi Y
The expected value of

of x Rn is defined as
X Rn,m
is usually called the mean. The covariance matrix

T

.
Cov (x) = Ex x Ex (x) x Ex (x)
where X T denotes the transpose of X. The covariance of x R is called the variance

and is given by

2
.
Var (x) = Ex x Ex (x) .
The square root of the variance (Var (x))1/2 is called the standard deviation.
2.2 Marginal and Conditional Densities

Consider a random vector x = [x1 xn ]T Rn with joint density function
fx (x) = fx1 ,...,xn (x1 , . . . , xn ).
The marginal density of the first i components of the random vector x = [x1 xn ]T
is defined as

.
(2.1)
fx1 ,...,xi (x1 , . . . , xi ) = fx (x1 , . . . , xn ) dxi+1 dxn .
The conditional density fxi |x1 ,...,xi1 (xi |x1 , . . . , xi1 ) of the random variable xi conditioned to the event x1 = x1 , . . . , xi1 = xi1 is given by the ratio of marginal
densities
. fx1 ,...,xi (x1 , . . . , xi )
.
(2.2)
fxi |x1 ,...,xi1 (xi |x1 , . . . , xi1 ) =
fx1 ,...,xi1 (x1 , . . . , xi1 )
2.3 Univariate and Multivariate Density Functions

We next present a list of classical univariate and multivariate density functions. The
reader is referred to Chap. 14 for numerical methods for generating random variables with the mentioned densities.
2.3 Univariate and Multivariate Density Functions
11
Binomial Density
The binomial density with parameters n, p is defined as

. n x
bn,p (x) =
p (1 p)nx , x {0, 1, . . . , n}
(2.3)
x

n!
. The binomial distribuwhere xn indicates the binomial coefficient xn = x!(nx)!
tion is denoted as
x
. n k
Bn,p (x) =
(2.4)
p (1 p)nk , x {0, 1, . . . , n}.
k
k=0
Normal Density The normal (Gaussian) density with mean x R and variance
2 R is defined as
1
1
.
2 / 2
e 2 (xx)
, x R.
(2.5)
Nx,
2 (x) =
2
Multivariate Normal Density
The multivariate normal density with mean
x Rn and symmetric positive definite covariance matrix W Sn , W 0, is defined as
1
.
n/2
T W 1 (xx)
Nx,W
|W |1/2 e 2 (xx)
, x Rn .
(2.6)
(x) = (2)
The uniform density on the interval [a, b] is defined as
1
.
if x [a, b];
U[a,b] (x) = ba
0
otherwise.
Uniform Density
(2.7)
Uniform Density over a Set Let S be a Lebesgue measurable set of nonzero

volume (see Sect. 3.1.3 for a precise definition of volume). The uniform density
over S is defined as
1
. Vol(S)
if X S;
US (X) =
(2.8)
0
otherwise.
If instead S is a finite discrete set, i.e. it consists of a finite number of elements
S = {X1 , X2 , . . . , XN }, then the uniform density over S is defined as
1
.
if X S;
US (X) = Card(S)
0
otherwise
where Card (S) is the cardinality of S.
Chi-Square Density
dom is defined as
The unilateral chi-square density with n > 0 degrees of free-
.
n2 (x) =
1
x n/21 ex/2 ,
(n/2)2n/2
where () is the Gamma function

. x1
e d,
(x) =
0
x R+
x > 0.
(2.9)
12
Weibull Density
The Weibull density with parameter a > 0 is defined as

a
.
Wa (x) = ax a1 ex , x R.
(2.10)
Laplace Density The unilateral Laplace (or exponential) density with parameter
> 0 is defined as
.
(2.11)
L (x) = ex , x R+ .
Gamma Density
defined as
The unilateral Gamma density with parameters a > 0, b > 0 is

.
Ga,b (x) =
1
x a1 ex/b ,
(a)ba
x R+ .
(2.12)
Generalized Gamma Density The unilateral generalized Gamma density with

parameters a > 0, c > 0 is defined as
c
. c
Ga,c (x) =
(2.13)
x ca1 ex , x R+ .
(a)
2.4 Convergence of Random Variables

We now recall the formal definitions of convergence almost everywhere (or almost
sure convergence), convergence in the mean square sense and convergence in probability. Other convergence concepts not discussed here include vague convergence,
convergence of moments and convergence in distribution, see e.g. [108].
Definition 2.1 (Convergence almost everywhere) A sequence of random variables
x(1) , x(2) , . . . converges almost everywhere (a.e.) (or with probability one) to the
random variable x if

P R lim x(N ) = x = 1.
N
Definition 2.2 (Convergence in the mean square sense) A sequence of random variables x(1) , x(2) , . . . converges in the mean square sense to the random variable x if
2

lim E x x(N ) = 0.
N
Definition 2.3 (Convergence in probability) A sequence of random variables x(1) ,

x(2) , . . . converges in probability to the random variable x if, for any > 0, we have

lim P R x x(N ) > = 0.
N
Convergence a.e. and convergence in the mean square sense both imply convergence in probability, while there is no implicative relationship between convergence
a.e. and convergence in the mean square sense.
http://www.springer.com/978-1-4471-4609-4

Elements of Probability Theory: 2.1 Probability, Random Variables and Random Matrices

Uploaded by

Copyright:

Available Formats

Elements of Probability Theory: 2.1 Probability, Random Variables and Random Matrices

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elements of Probability Theory: 2.1 Probability, Random Variables and Random Matrices

Uploaded by

Copyright:

Available Formats

Chapter 2

Elements of Probability Theory

In this chapter, we formally review some basic concepts of probability theory.

2.1 Probability, Random Variables and Random Matrices

where I is a countable1 set of positive integers.

countable we mean finite (possibly empty) or countably infinite.

2 Elements of Probability Theory

2.1.2 Real and Complex Random Variables

The (cumulative) distribution function (cdf)

then the cdf Fx (x) is said to be absolutely continuous, and

2.1 Probability, Random Variables and Random Matrices

2.1.3 Real and Complex Random Matrices

where Fxi (xi ) = P R {xi xi }.

Clearly, P R {X Y} = Y fX (X) dX = 1. When needed, to further emphasize that

2.1.4 Expected Value and Covariance

where Y is the support of X. We make use of the symbol EX (J (X)) to emphasize

2 Elements of Probability Theory

If X Fn,m is a discrete random matrix with countable support Y = {X1 , X2 , . . .},

The expected value of J (X) is defined as

The expected value of

is usually called the mean. The covariance matrix

where X T denotes the transpose of X. The covariance of x R is called the variance

2.2 Marginal and Conditional Densities

2.3 Univariate and Multivariate Density Functions

2.3 Univariate and Multivariate Density Functions

The binomial density with parameters n, p is defined as

Uniform Density over a Set Let S be a Lebesgue measurable set of nonzero

The unilateral chi-square density with n > 0 degrees of free-

where () is the Gamma function

2 Elements of Probability Theory

The Weibull density with parameter a > 0 is defined as

The unilateral Gamma density with parameters a > 0, b > 0 is

Generalized Gamma Density The unilateral generalized Gamma density with

2.4 Convergence of Random Variables

Definition 2.3 (Convergence in probability) A sequence of random variables x(1) ,

You might also like