School of Mathematics and Statistics

UNSW Sydney

Introduction to Probability and Stochastic Processes

OPEN LEARNING
 Chapter 9
Branching Processes

2 / 40
Outline:

9.1 Introduction
9.2 What is Probability of Extinction?
9.3 Calculation of Probability of Extinction
9.4 Probability Generating Functions
9.5 Supplementary Material

3 / 40
9.1 Introduction

4 / 40
Branching processes are used in biological, sociological, and engineering
sciences.

Suppose an organism at the end of its lifetime produces a random number Z

of offspring with probability distribution

P(Z = j) = Pj , j = 0, 1, 2, . . . (1)
P
where Pj ≥ 0 and j Pj = 1.

Suppose that all offspring act independently of each other and produce their
offspring according to the probability law given by Equation (1).

Notation: In this chapter and the next, we will use P instead of P to avoid
the confusion of the notation of Pj = P (Z = j). The notation, P, is used in
advanced courses.

5 / 40
Definition
Consider a population consisting of the organisms described above.

The number of organisms initially present, denoted X0 , is the size of the

zeroth generation.

All offspring of the zeroth generation constitute the first generation, and
their number is denoted by X1 .

Let Xn be the population size of the nth generation, and

{Xn , n = 0, 1, 2, . . . } is called a discrete-time branching process.

It turns out, under these conditions, {Xn , n ≥ 0} is a Markov chain with

states {0, 1, 2, . . . , }.

6 / 40
State 0 is an absorbing and thus recurrent state since

P00 = P(Xn = 0 Xn−1 = 0) = 1.

In addition, if P0 > 0, the probability of starting in state i going to state 0 is

Pi0 = P0i for all i ≥ 1.

Therefore, all other states except state 0 are transient.

An important outcome is if P0 > 0, then the population will either die

out or its size will converge to infinity.

7 / 40
9.2 What is the Probability of Extinction?

8 / 40
Population extinctions were first raised in 1889 by Galton1 in connection
with the extinction of family surnames.

Assume that the initial population consists of one individual, i.e. X0 = 1.

Galton-Watson2 definition of the process is, for every n = 1, 2, 3, . . . ,

Xn−1
X
Xn = Zi , (2)
i=1

where Zi are independent random variables with a common distribution

given by Equation (1). That is, Zi is the number of offspring of the ith
individual of the (n − 1)th generation.
1
Galton, F. (1889). Natural Inheritance. Macmillan, London.
2
Galton, F., and Watson, H. W. (1875). ”On the probability of the extinction of families”. Journal of the Royal
Anthropological Institute, 4, 138–144.

9 / 40
 P
Let µ = j jPj be the mean number of offspring of a single individual,
and

let σ 2 = j (j − µ)2 Pj be the variance of the number of offspring

P
produced by a single individual.

Result
From Equation (2), we see that

E(Xn ) = µn ,

and (  n 
−1
σ 2 µn−1 µµ−1 if µ ̸= 1
V ar(Xn ) =
n σ2 if µ = 1.

10 / 40
Proof
By conditioning on Xn−1 , we have


E(Xn ) = E E(Xn Xn−1 )
Xn−1
X

= E E( Zi Xn−1 )
i=1
= E(Xn−1 µ)
= µ E(Xn−1 ).

Note that E(X0 ) = 1. Hence, we have

E(X1 ) = µ,
E(X2 ) = µE(X1 ) = µ2 ,
E(X3 ) = µE(X2 ) = µ3 ,
..
.
E(Xn ) = µE(Xn−1 ) = µn .

11 / 40
Proof - continued
Similarly, we can find V ar(Xn ) by the conditional variance formula

V ar(Xn ) = E(V ar(Xn Xn−1 )) + V ar(E(Xn Xn−1 )).

Given Xn−1 , Xn is the sum of Xn−1 independent random variables each

having the distribution {Pj , j ≥ 0}.

Now V ar(Xn Xn−1 ) = Xn−1 σ 2 . The conditional variance formula yields

V ar(Xn ) = E(Xn−1 σ 2 ) + V ar(Xn−1 µ)

= σ 2 µn−1 + µ2 V ar(Xn−1 ).

12 / 40
Proof.
However, V ar(X0 ) = V ar(1) = 0, by mathematical induction,

V ar(X1 ) = σ 2 µ1−1 + µ2 V ar(X0 ) = σ 2 ,

V ar(X2 ) = σ 2 µ2−1 + µ2 V ar(X1 ) = σ 2 µ + µ2 σ 2 = σ 2 µ(1 + µ),

V ar(X3 ) = σ 2 µ3−1 + µ2 V ar(X2 )

= σ 2 µ2 + µ2 (σ 2 µ(1 + µ))
= σ 2 µ2 (1 + µ + µ2 )
 3 
2 2 µ −1
= σ µ ,
µ−1
..
.

Therefore, the variance of the population size increases geometrically if µ > 1 and
increases linearly if µ = 1 and decreases geometrically if µ < 1.
13 / 40
9.3 Calculation of Probability of Extinction

14 / 40
The population becomes extinct if the population size is reduced to zero.

Let π0 denote the probability that the population will eventually die out under the
assumption that X0 = 1. Then
π0 = lim P(Xn = 0 X0 = 1).
n→∞

First note that π0 = 1 when µ < 1.

This follows since

∞
X ∞
X
µn = E(Xn ) = j P(Xn = j) ≥ 1 · P(Xn = j) = P(Xn ≥ 1).
j=1 j=1

When µ < 1, µn → 0 and hence, P(Xn ≥ 1) → 0. Therefore, P (Xn = 0) → 1.

In fact, it can be shown that π0 = 1 even when µ = 1.

15 / 40
When µ > 1, it turns out that π0 < 1.

We can derive the equation determining π0 by conditioning on the number of

offspring of the initial individual, as follows.

π0 = P(population dies out)

X∞
= P(population dies out X1 = j) Pj .
j=0

Given that X1 = j, the population will eventually die out if and only if each
of the j families started by members of the first generation eventually dies
out.

16 / 40
Since each family is assumed to act independently and since the probability
that any particular family dies out is just π0 , this yields

P(population dies out X1 = j) = π0j .

Thus, π0 satisfies
∞
X
π0 = π0j Pj . (3)
j=0

In fact, when µ > 1, it can be shown that π0 is the smallest positive number
satisfying Equation (3).

17 / 40
Example
If P0 = 1/2, P1 = 1/4 and P2 = 1/4, then determine π0 .

18 / 40
Example
If P0 = 1/2, P1 = 1/4 and P2 = 1/4, then determine π0 .

Solution:
Using Equation (3) , we have

π0 = π00 P0 + π01 P1 + π02 P2

1 1 1
= + π0 + π02
2 4 4
=⇒
1 3 1
0 = − π0 + π02 .
2 4 4
By the quadratic formula , we get π0 = 2 and π0 = 1, so the only viable
solution is π0 = 1 because probabilities are bounded by one.
P2 3
Alternatively, we see that µ = j=0 j Pj = 0 P0 + 1 P1 + 2P2 = 4
< 1, it follows that π0 = 1.

18 / 40
Example
If P0 = 1/4, P1 = 1/4 and P2 = 1/2, then determine π0 .

19 / 40
Example
If P0 = 1/4, P1 = 1/4 and P2 = 1/2, then determine π0 .

Solution:
Using Equation (3) , we have

π0 = π00 P0 + π01 P1 + π02 P2

1 1 1
= + π0 + π02
4 4 2
=⇒
1 3 1
0 = − π0 + π02 , or
4 4 2
0 = 2π02 − 3π0 + 1.
P
By the quadratic formula , we get π0 = 1/2 and π0 = 1. Here µ = j j Pj = 5/4 > 1.
The smallest positive solution of this quadratic equation is π0 = 1/2.

19 / 40
Example
In the previous two examples, what is the probability that the population
will die out if it initially consists of k individuals?

20 / 40
Example
In the previous two examples, what is the probability that the population
will die out if it initially consists of k individuals?

Solution:
Recall that the population will die if and only if the families of each initial
generation die out. So, the desired probability is π0k .

For the example with P0 = 1/2, P1 = 1/4 and P2 = 1/4 , it is π0k = 1k = 1.

For the example with P0 = 1/4, P1 = 1/4 and P2 = 1/2 , it is π0k = (1/2)k .

20 / 40
9.4 Probability Generating Functions

21 / 40
Probability generating functions are very useful tools when studying
branching processes.

Definition
The probability generating function (pgf ) of a non-negative
integer-valued random variable X is defined by
∞
X
X
GX (s) = E(s ) = sj P(X = j),
j=0

for all s ∈ R, for which the sum converges.

22 / 40
Properties

1 GX (0) = P(X = 0),

GX (1) = 1, and

G′X (1) = E(X).

(j)
2 GX (0)/j! = P(X = j) and

(j)
GX (1) = E(X (X − 1) · · · (X − j + 1)).

23 / 40
Properties - continued

➌ Suppose that X = ni=1 Xi , where {Xi , i = 0, 1, 2, . . . } are independent

P
identically distributed random variables. Then,
 n
GX (s) = GXi (s) .

➍ Suppose that X = N
P
i=1 Xi , a compound random variable, or a random
sum, with N a positive integer-valued random variable and
{Xi , i = 0, 1, 2, . . . } are independent identically distributed random
variables and independent of N . Then,

GX (s) = GN (GXi (s)).

24 / 40
The probability-generating functions of the branching process Xn is
∞
X
Xn
Gn (s) ≡ E(s )= sj P (Xn = j), n = 1, 2, 3, . . . , 0 ≤ s ≤ 1.
j=0

Then, it can be shown that

Gn+1 (s) = Gn (G1 (s)),

Gn+1 (s) = G1 (Gn (s)).

This can be easily seen by induction.

25 / 40
Write
π0 = lim P(Xn = 0 X0 = 1)
n→∞

as the probability that the population will eventually die out, or we say it is
the ultimate probability that the population will die out.

Note that the limit exists since Xn = 0 is increasing in n.

Denote π0 (k) = limn→∞ P(Xn = 0 X0 = k), then it can be easily shown that

π0 (k) = π0k .

26 / 40
The following theorem shows the properties of π0 . It also provides a method
on how to find π0 .

The proof of this theorem can be found in Ross, S.M. (1970) Applied Probability Models
with Optimization Applications, Dover Publication, N.Y. USA on page 77.

Theorem
Suppose P0 > 0, and P0 + P1 < 1. Then

1 π0 is the smallest positive number p satisfying

G1 (p) = p.

2 π0 = 1 if and only if µ ≤ 1.

27 / 40
Example
Let G(s) = (1 − p + p s), where 0 < p < 1 (this is the offspring distribution of
a Bernoulli(p)).

The associating branching process is a pure death process.

1 Find the probability generating function of the associating

branching process, Gn (s).
2 Let T be the time of extinction. Calculate P(T = n X0 = k).

28 / 40
Example

Solution:
1 Given G1 (s) = (1 − p + p s), we obtain by induction,

G2 (s) = G1 (G1 (s)) = (1 − p + p (1 − p + p s)) = (1 − p2 ) + p2 s,

..
.
Gn (s) = G1 (Gn−1 (s)) = (1 − pn ) + pn s.

Hence, Xn ∼ Bernoulli(pn ).

29 / 40
Example

Solution - continued:
➋ Let T be the random time of extinction. It is the first time n when
Xn = 0 and obviously, Xn = 0 for all n ≥ T . That is,

T = min{n ≥ 0 : Xn = 0}.

We have
P(Xn = 0) = P(T ≤ n).

30 / 40
Example

Solution - continued:
➋ Let the initial population size X0 = 1. Then

P(T = n X0 = 1) = P(Xn = 0 X0 = 1) − P(Xn−1 = 0 X0 = 1)

= Gn (0) − Gn−1 (0)
= (1 − pn ) − (1 − pn−1 )
= pn−1 (1 − p).

31 / 40
Example

Solution - continued:
➋ Suppose now the initial population size X0 = k. We can determine the
probability distribution of the time to extinction:

P(T = n X0 = k) = P(Xn = 0 X0 = k) − P(Xn−1 = 0 X0 = k)

k
k
= P(Xn = 0 X0 = 1) − P(Xn−1 = 0 X0 = 1)

k
k
= Gn (0) − Gn−1 (0)
= (1 − pn )k − (1 − pn−1 )k

32 / 40
Example
Consider a branching process with Z following the distribution

P(Z = j) = (1 − p)j p, j = 0, 1, 2, . . .

for some 0 < p < 1.

Find the probability of extension of the process.

33 / 40
Example

Solution:
Let q = 1 − p. Firstly, we need to find the generating function of Z,i.e.,

∞
X
Z
G(s) = E(s ) = sj P (Z = j)
j=0
∞
X ∞
X
j j
= s q p=p (sq)j
j=0 j=0
1
= p for all s such that sq < 1
1 − sq
p 1
= for |s| < .
1 − sq q

34 / 40
Example

Solution:
1
We will only consider the case when p = q, so G1 (s) = 2−s
.

1
G2 (s) = G1 (G1 (s)) = 1
2 − 2−s
2−s 2−s
= =
4 − 2s − 1 3 − 2s
1
2− 2−s
G3 = G2 (G1 (s)) = 1
3− 2 2−s
4 − 2s − 1 3 − 2s
= = .
6 − 3s − 2 4 − 3s
Now, we see a pattern.
35 / 40
Example

Solution - continued
We have
n − (n − 1)s
Gn (s) = , for n = 1, 2, . . . .
(n + 1) − ns
We will prove this by proof of induction. Assume Gn (s). Consider
 
1
n − (n − 1) 2−s
Gn+1 (s) = Gn (G(s)) =  
1
(n + 1) − n 2−s
2n − ns − (n − 1)
=
2n + 2 − ns − s − n
(n + 1) − ns
= .
(n + 2) − (n + 1)s
36 / 40
Example

Solution - continued:
n − (n − 1)s
Gn (s) = , for n = 1, 2, . . . .
(n + 1) − ns

The number of Zn of the nth generation satisfies

n
P (Zn = 0) = Gn (0) = , if p = q.
n+1

For n ≥ 1 and p = q,
1
P(T = n) = P(Zn = 0) − P(Zn−1 = 0) = .
n(n + 1)

It follows that E(T ) < ∞ if and only if p = q.

37 / 40
Example

Solution - continued:
For p ̸= q, we have the number Zn of the nth generation satisfies

q (pn − q n )
P(Zn = 0) = Gn (0) = .
pn+1 − q n+1

For n ≥ 1,

pn−1 q n (p − q)2
P (T = n) = P(Zn = 0) − P(Zn−1 = 0) = .
(pn − q n ) (pn+1 − q n+1 )

38 / 40
9.5 Supplementary Material

39 / 40
Supplementary Material - Quadratic Formula

The quadratic formula helps us solve any quadratic equation in the form
ax2 + bx + c = 0, where a, b, and c are coefficients. The formula is

√
−b ± b2 − 4 a c
.
2a

40 / 40