Important Random Variables

Binomial: SX = { 0,1,2, . . . , n}
n
Pk    p k (1  p ) ( n k ) k  0,1,..., n
k 
0  p 1

Geometric: SX = { 0,1,2, . . . }
Pk  p(1  p) k k  0,1,...,
0  p 1

Poisson: SX = { 0,1,2, . . . }
 k 
Pk  e k  0,1,...,
k!
 0
Poisson Distribution:
Used for modeling number of events in an interval or set if they occur
randomly and independently.

If N = Number of events in an interval T,

 k 
P( N  k )  e
k!
Where α = Event rate  T
= Average number of events per size T Interval.

e.g. Failure rate for chips in a system = 2 / year

210  2
P (10 failures in a year)  e
10!

For a large number of Bernoulli trials with small success rate, p,

Number of successes = Poisson
 n
P( k hits in n attempts) =   p k (1  p) ( n  k )
k 

 k 
 e with α = np as n → ∞ and np is constant.
k!
e.g.
 10 6
 10

P (10 jackpots in 10 attempts)  
6
 p (1  p )106 10

10 
(106 p )10 106 p
 e for small p
10!
Examples of Poisson Distribution Applications
1)
P(bit error in a communication line)  10 3
P( 10 errors in 1000 bit block)  ?
n 1000 , p  10 3 , np  1000(10 3 )
1000  3 k
  
10
P(10 errors)  1  P(10 errors)  1     10 1  10 3 1000  k

k 0  k 

1 
10
1000(10 ) 3 k 10
1
k 0 k!
e 1000(10 3 )
 1 e 1
 k!
k 0

2)
P(defective component)  10 3
number of components  1000
P( machine functions)  P (number of defectivecomponents)


1000 (10 ) 3 0

e 1000(10
3
)

0!
 e 1
 Continuous Random Variables
Uniform: S X  [a, b]
1
f X ( x)  a xb
ba
ab

Gaussian: S X  [, ]
2 2
e ( x  m ) 2
f X ( x) 
2 

Exponential: S X  [0, ]
f X ( x )   e  x x0
 0
Mean: 
E( X ) 

 sf X ( S ) ds

also called the expected value of X.

If X is Discrete

E ( X )   xk PX ( xk )
k

E(X) does not exist for all random variables.

It requires that

E X   sf X ( s ) ds 


or x P x 
k
k X k 

for Discrete X
Variance:

 X2  Var ( X )  E [ X  E ( X )]2 
 E ( X 2 )  ( E ( X )) 2

Standard Deviation = √Var

Variance measures the dispersion of X about the mean.

Moments: 
n th Moment (X) = E ( X ) 
 f X ( x) dx
n n
x

n th Central Moment = 
E ( X  E ( X ))n 
n
n th Absolute Moment = E( X )

n th Generalized Moment about a = E((X a) )

n
Markov Inequality:
E( X )
P( X  a)  if X  0
a
[useful only if a  E ( X )] a0
Chebychev Inequality:
2
P( X  m  a)  2 a 0
a
m  E( X )

Beinayme Inequality:
n
E( X  b )
P( X  b  a)  a 0
an
Chebychev inequality is a special case of this with b=m, n=2
Memoryless Random Variables:
A random variable , X , is called memoryless if, for h>0
P ( X  x  h | X  x )  P ( X  h)

i.e. the incremental probability of x+h is independent of x.

Context is meaningless.

Geometric is the only memoryless discrete Random Variable.

Exponential is the only memoryless continuous Random Variable.
P( X  x  h, X  x) P ( X  x  h)
P( X  x  h | X  x)  
P( X  x) P ( X  x)
 e  h  P ( X  h )
 Gaussian Random Variable
1 2
22
f X ( x)  e ( x m )
2 

1
x fX(x)
2  
( s  m ) 2 22
FX ( x)  e ds  

There is no closed form for

FX (x)
m x

The Gaussian distribution is also called the normal distribution

and is often popularly referred to as the bell curve.

It is found to be a good model for random variables in many real-world systems,

and has many useful properties (as we will see later in the class).
Standard Gaussian Random Variable

x
1 Note:
e
 y2 2
Define G ( x)  dy The textbook uses  instead
2  of G, but we will later use 
( xm)  for something else.
 xm 1
e
 y2 2
FX ( x)  G  dy
   2 

X m
 G ( y ) is the cdf of random variable Y  .

 Y is called a standard Gaussian random variable.
x
1
 dy
2
y 2
 G ( x)  e
2 

is called the standard Gaussian cdf.

 x
1
 dy  erf ( x) is called the error function.
2
y 2
e
2 0

G ( x )  erf ( x )  1 x  0
2

G( x ) 1  erf (  x ) x  0
2
Define Q (x)  1  G (x)
x 
1 1 1
e e
y2 2 2
 1  dy  y 2
dy   erf ( x )
2  2 x
2

1
Q( 0 ) 
2
Q ( x)  1  Q ( x)

  1  x2
Q (x)    1 
1
1
 e 2
for 0  x  
  1  x 
   
x2 2 
 2
De Moivre-Laplace Theorem:

if np(1-p) >> 1
n k 1
  p ( 1  p )( n  k )  2
e ( k  np ) / 2 np (1 p )
k  2np(1  p)
 
i.e. Binomial  Gaussian for large n and finite p

where m = np and  = np(1-p)

DL 
if k ~ binomially
k2n k
P(k1  k  k 2 )     p (1  p) n  k
 
k  k1 k
 

 k 2  np   k1  np 
G    G 
 np (1  p)   np(1  p ) 
   
Functions of Random Variables:
If X is a random variable
Y = g(X) is also a random variable.

E ( g ( X ))   g ( x) f

X ( x) dx
Any event {g(X) ≤ a} can be seen as a union of events in SX . This is called the
equivalent event.
e.g.

g(x)

{g ( X )  a}  {i, j , k}
a

i j k x

P( g  X   a)  P( X  i )  P( X  j )  P( X  k )
Example: y+dy
y

x1 x1 + dx1 x2 x2 + dx2
x3 x3+ dx3

P( y  Y  y  dy)  P( x1  X  x1  dx1 )  P( x2  X  x2  dx2 )

 P( x3  X  x3  dx3 )

if t  length of t

fY ( y ) dy  f X ( x1 ) dx1  f X ( x2 ) dx2  f X ( x3 ) dx3

In General, if y=g(X) has n solutions, {xk}

dx
fY ( y)  
k
f X ( x)
dy x  xk