EE5110: Probability Foundations for Electrical Engineers July-November 2015

Lecture 24: Probability Generating Functions

Lecturer: Dr. Krishna Jagannathan Scribe: Debayani Ghosh

24.1 Probability Generating Functions (PGF)

Definition 24.1 Let X be an integer valued random variable. The probability generating function (PGF)
of X is defined as :
X
GX (z) , E[z X ] = z i P(X = i).
i

24.1.1 Convergence

For a non-negative valued random variable, there exists R, possibly +∞, such that the PGF converges for
|z| < R and diverges for |z| > R where z ∈ C. GX (z) certainly converges for |z| < 1 and possibly in a larger
region as well. Note that,
X X
|GX (z)| = z i P(X = i) ≤ |z|i .
i i

This implies that GX (z) converges absolutely in the region |z| < 1. Generating functions can be defined
for random variables taking negative as well as positive integer values. Such generating functions generally
converge for values of z satisfying α < |z| < β for some α, β such that α ≤ 1 ≤ β.

Example 1 : Consider the Poisson random variable X with probability mass function
e−λ λi
P(X = i) = , i ≥ 0.
i!
Find the PGF of X.

Solution : The PGF of X is

∞
X z i λi e−λ
GX (z) = = eλ(z−1) , ∀z ∈ C.
i=1
i!
Example 2 : Consider the geometric random variable X with probability mass function
P(X = i) = (1 − p)i−1 p, i ≥ 1.
Find the PGF of X.

Solution : The PGF of X is

∞
i−1
X
GX (z) = (1 − p) pz i ,
i=1
pz 1
= , if |z| < .
1 − z (1 − p) 1−p

24-1
24-2 Lecture 24: Probability Generating Functions

24.1.2 Properties
1. GX (1) = 1.
dGX (z)
2. dz = E[X].
z=1

Proof : From definition

  X i
GX (z) = E z X = z P(X = i).
i

Now,
dGX (z) d X i
= z P(X = i),
dz dz i
(a) X d i
= z P(X = i),
i
dz
X
= iz i−1 P(X = i),
i

where the interchange

P i of differentiation and summation in (a) is a consequence of absolute convergence
of the series z P(X = i). Thus,
i

dGX (z)
= E[X].
dz z=1

dk GX (z)
3. dz k = E [X (X − 1) (X − 2) · · · (X − k + 1)].
z=1

4. If X and Y are independent and Z = X + Y , then GZ (z) = GX (z)GY (z). The ROC for the PGF of z
is the intersection of the ROCs of the PGFs of X and Y .
Proof :

GZ (z) = E[z Z ] = E[z X+Y ] = E[z X .z Y ].

Since X and Y are independent, they are uncorrelated. This implies that

E[z X .z Y ] = E[z X ]E[z Y ] = GX (z)GY (z).

Hence proved.
N
P
5. Random sum of discrete RVs : Let Y = Xi , where Xi ’s are i.i.d discrete positive integer valued
i=1
random variables and N is independent of Xi ’s. The PGF of Y is GY (z) = GN (GX (z)).
Proof :

GY (z) = E[z Y ] = E E z Y |N
  
(By law of iterated expectation).

Now,
 P 
xi
Y
|N = n = E GX (z)N .
   
E z |N = n = E z i

This implies that

GY (z) = GN (GX (z)) .

Lecture 24: Probability Generating Functions 24-3

24.2 Exercise
1. Find the PMF of a random variable X whose probability generating function is given by

( 31 z+ 32 )4
GX (z) = z

2. Suppose there are X0 individuals in initial generation of a population. In the nth generation, the Xn in-
(n) (n) (n) (n) (n) (n)
dividuals independently give rise to numbers of offspring Y1 , Y2 , ..., YXn , where Y1 , Y2 , ..., YXn
st
are i.i.d. random variables. The total number of individuals produced at the (n + 1) generation will
(n) (n) (n)
then be Xn+1 = Y1 + Y2 + ... + YXn . Then, {Xn } is called a branching process. Let Xn be the size
of the nth generation of a branching process with family-size probability generating function G(z), and
let X0 = 1. Show that the probability generating function Gn (z) of Xn satisfies Gn+1 (z) = G(Gn (z))
for n ≥ 0. Also, prove that E[Xn ] = E[Xn−1 ]G′ (1).