1 Distribution of the Simple (Linear) Birth–Death

Process
Consider a simple birth–death process {X(t), t ≥ 0} with
• per capita birth rate λ > 0,
• per capita death rate µ > 0,
• initial population X(0) = i.

1.1 1. Probability Generating Function

The probability generating function (PGF) is defined as
G(s, t) = E[sX(t) | X(0) = i].
For λ ̸= µ, let
r = λ − µ.
Then the PGF is given by
i
(µ − λs)e−rt

G(s, t) = .
µ − λse−rt
This expression completely determines the distribution of X(t).

1.2 2. Probability of Extinction by Time t

The probability that the population is extinct at time t is obtained by setting
s = 0:
i
µ(1 − e−rt )

P {X(t) = 0 | X(0) = i} = , λ ̸= µ.
λ − µe−rt

1.3 3. Distribution for n ≥ 1

For n ≥ 1, the transition probabilities can be obtained by expanding the gener-
ating function in powers of s.
Define
λ(1 − e−rt )
α(t) = .
λ − µe−rt
Then
min(i,n)
X i−k k n−i+k
P {X(t) = n | X(0) = i} = ik (1 − p0 (t)) (p0 (t)) n − 1i − k − 1α(t)i−k (1 − α(t)) ,
k=0

where
µ(1 − e−rt )
p0 (t) = .
λ − µe−rt

1
1.4 4. Critical Case: λ = µ
When λ = µ, the process is called critical.
In this case,
λt
P {X(t) = 0 | X(0) = 1} = .
1 + λt
For n ≥ 1,

(λt)n−1
P {X(t) = n | X(0) = 1} = .
(1 + λt)n+1

For general i, the distribution is the i-fold convolution of the above distri-
bution.

1.5 5. Limiting Behaviour as t → ∞

• If λ < µ (subcritical case):

P (X(t) = 0) → 1.

• If λ > µ (supercritical case):

 µ i
P (X(t) = 0) → .
λ

• If λ = µ (critical case):
P (X(t) = 0) → 1.

1.6 6. Remarks
• The distribution belongs to the branching process family.
• When µ = 0, the model reduces to the pure birth (Yule) process and the
distribution becomes negative binomial.
• When λ = 0, the model reduces to the pure death process.

• The generating function provides the most compact representation of the

distribution.

2 Feller’s Method for the Mean and Variance of

the Simple Birth–Death Process
2.1 1. Model Setup
Let {X(t), t ≥ 0} be a simple (linear) birth–death process with:

2
 • Per capita birth rate λ > 0
• Per capita death rate µ > 0
• Initial state X(0) = i
Transition rates:
λn = λn, µn = µn.
Let
Pn (t) = P (X(t) = n | X(0) = i).
For n ≥ 1, the Kolmogorov forward equations are
dPn (t)
= λ(n − 1)Pn−1 (t) + µ(n + 1)Pn+1 (t) − (λ + µ)nPn (t).
dt
For n = 0:
dP0 (t)
= µP1 (t).
dt

2.2 2. Feller’s Method for the Mean

Define the first moment:
∞
X
m(t) = E[X(t)] = nPn (t).
n=0

Differentiate:
∞
dm(t) X dPn (t)
= n .
dt n=0
dt

Substitute the forward equations:

∞
dm(t) X h i
= n λ(n − 1)Pn−1 + µ(n + 1)Pn+1 − (λ + µ)nPn .
dt n=1

Now evaluate each term separately.

Birth term
∞
X
nλ(n − 1)Pn−1 .
n=1

Let k = n − 1. Then n = k + 1:
∞
X ∞
X
=λ (k + 1)kPk = λ (k 2 + k)Pk .
k=0 k=0

3
Death term

∞
X
nµ(n + 1)Pn+1 .
n=1

Let k = n + 1. Then n = k − 1:
∞
X ∞
X
=µ (k − 1)kPk = µ (k 2 − k)Pk .
k=1 k=0

Loss term

∞
X
−(λ + µ) n 2 Pn .
n=0

Combine terms
Collecting:

dm(t) X X X
=λ (k 2 + k)Pk + µ (k 2 − k)Pk − (λ + µ) k 2 Pk .
dt
Expanding:
X X X
= (λ + µ) k 2 Pk + (λ − µ) kPk − (λ + µ) k 2 Pk .

The k 2 terms cancel.

Thus:
dm(t) X
= (λ − µ) kPk .
dt
Therefore,

dm(t)
= (λ − µ)m(t)
dt

2.3 Solution
This is a first-order linear ODE:

m(t) = ie(λ−µ)t .

4
2.4 3. Feller’s Method for the Second Moment
Define:
∞
X
M2 (t) = E[X 2 (t)] = n2 Pn (t).
n=0

Differentiate:
dM2 (t) X 2 dPn (t)
= n .
dt dt
Substitute forward equations:
dM2 (t) X 2 h i
= n λ(n − 1)Pn−1 + µ(n + 1)Pn+1 − (λ + µ)nPn .
dt
After similar index shifting and simplification (details omitted for brevity
but following identical algebraic steps), we obtain:
d
E[X 2 (t)] = 2(λ − µ)E[X 2 (t)] + (λ + µ)E[X(t)]
dt

2.5 4. Variance
Recall:
Var(X(t)) = E[X 2 (t)] − (E[X(t)])2 .
Solving the coupled system gives:
(λ + µ)i (λ−µ)t  (λ−µ)t 
Var(X(t)) = e e −1 , λ ̸= µ.
λ−µ

2.6 5. Critical Case (λ = µ)

If λ = µ, then:

E[X(t)] = i,
and solving the simplified second-moment equation yields:

Var(X(t)] = 2λit.

2.7 6. Interpretation
• If λ > µ: exponential growth in mean and variance.
• If λ < µ: exponential decay in mean.
• If λ = µ: mean constant but variance increases linearly.
Feller’s method shows that moment evolution depends only on drift (λ − µ)
and fluctuation intensity (λ + µ).

5
2.8 Mean and Variance: Introduction to Feller’s Method
In the study of pure birth processes, the probability generating function (PGF)
was used to determine the mean and variance. While the PGF approach is
powerful and elegant, it becomes algebraically cumbersome for more general
birth–death processes.
An alternative and more direct approach for determining moments is due to
Feller. This method avoids the use of generating functions and instead works
directly with the Kolmogorov forward equations.
Idea of Feller’s Method.
Let {X(t), t ≥ 0} be a birth–death process with transition probabilities
Pn (t) = P (X(t) = n) satisfying the forward equations.
The k-th moment of the process is defined by
∞
X
k
mk (t) = E[X (t)] = nk Pn (t).
n=0

To determine the evolution of mk (t):

1. Differentiate mk (t) with respect to t,

∞
d X dPn (t)
mk (t) = nk .
dt n=0
dt

dPn (t)
2. Substitute the Kolmogorov forward equations for dt .

3. Interchange summation and differentiation.

4. Simplify using index shifts.

This procedure typically yields a differential equation involving lower-order

moments. Solving the resulting system gives explicit expressions for the mean
and variance.
Advantages of Feller’s Method

• Avoids solving partial differential equations for the generating function.

• Provides direct differential equations for the moments.
• Clearly separates the roles of drift (λ−µ) and fluctuation intensity (λ+µ).
• Extends naturally to more general birth–death processes.

In what follows, we apply Feller’s method to the simple (linear) birth–death

process to determine its mean and variance.