Skip to content
Tutorials Center
My learning

Topics

Lesson 33 of 36 · Applications and Simulation

Branching Processes and Extinction

Model a growing population with the Galton-Watson process, find the mean generation size, and compute the extinction probability as a fixed point.

Will a family name die out? Will a chain reaction fizzle or explode? Will a new virus spread or vanish? In all of these, individuals independently produce a random number of offspring, and the population is made of generations. The Galton-Watson branching process is the classic model. It was introduced in the 19th century by Francis Galton and Henry Watson to study the extinction of surnames.

The model

Start with one individual: Z₀ = 1. Every individual in every generation has a random number of children with the same distribution, independently of all others. Let ξ be the offspring count with P(ξ = k) = pₖ, mean m = E[ξ]. Then

Text
Z_{n+1} = ξ_1 + ξ_2 + ... + ξ_{Z_n}

a random sum of Zₙ independent offspring counts. If Zₙ = 0 the population is extinct and stays extinct.

Mean generation size

Taking expectations, E[Z_{n+1}] = m E[Zₙ], so

Text
E[Z_n] = m^n
  • m < 1 (subcritical): the mean population shrinks geometrically; extinction is certain.
  • m = 1 (critical): the mean stays 1, but extinction is still certain (unless every individual has exactly one child).
  • m > 1 (supercritical): the mean explodes, and there is a positive probability of survival forever.

Extinction probability

Let q be the probability of eventual extinction starting from one individual. Condition on the first generation: if the root has k children, each of their lineages dies out independently with probability q, so all die with probability qᵏ. Hence

Text
q = Σ p_k q^k = G(q)

where G(s) = E[s^ξ] is the probability generating function of the offspring distribution. The extinction probability is the smallest non-negative root of s = G(s). For a supercritical process this root is strictly less than 1; otherwise q = 1.

A practical way to find it: start at q₀ = 0 and iterate q_{k+1} = G(qₖ). The value qₖ is the probability of extinction by generation k, and it increases to q.

Example: offspring is Poisson with mean m. Then G(s) = e^(m(s - 1)) and q solves q = e^(m(q-1)).

Python Extinction probability by fixed-point iteration

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

For m ≤ 1 the iteration climbs to 1 (certain extinction, slowly at m = 1); for m > 1 it stops below 1. Compare the last line with the algebra: the smallest root of the quadratic is 0.5, which the iteration finds.

Simulating the process

For Poisson offspring, the sum of Zₙ independent Poisson(m) counts is Poisson(m Zₙ), so each generation is a single random draw. A run that reaches a very large size is practically certain to survive, so we stop it at a cap.

Python Simulate extinction frequency

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The simulated extinction frequencies match the fixed-point values, and the generation means follow mⁿ. Runs that die out do so mostly in the first few generations; if a lineage gets large, it almost never returns to 0.

Where it applies

  • Epidemics: early in an outbreak, one infected person infects a random number of others. The mean is the basic reproduction number R₀. If R₀ > 1, an outbreak can take off, yet may still die out by chance. If the offspring distribution is Poisson with R₀ = 1.5, the outbreak fizzles from a single case with probability about 0.42.
  • Nuclear and chemical chain reactions: neutrons produce further neutrons.
  • Genetics: survival of a new mutation (a gene copy with a small advantage has survival probability roughly 2 × advantage in a population of large size).
  • Social media: will a post spread, or die after a few shares?

Real epidemics have more variable offspring counts than Poisson (a few super-spreaders and many people who infect nobody). More variability around the same mean makes extinction more likely, so the Poisson model overstates the chance of a large outbreak.

flowchart TD
  Z0["Z0 = 1"] --> Z1["Z1 children"]
  Z1 --> Z2["Z2 grandchildren"]
  Z2 --> Z3["Z3 ..."]

Generations of a branching process; each individual reproduces independently.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. A branching process has mean offspring m = 0.7. What happens to the population in the long run?

    Subcritical (m < 1): E[Z_n] = 0.7^n goes to 0 and extinction is certain.
  2. 2. What is E[Z_n] when Z_0 = 1 and the mean offspring count is m?

    Each generation multiplies the mean by m.
  3. 3. The extinction probability q is which root of s = G(s)?

    q is the smallest non-negative solution; s = 1 is always a root.
  4. 4. For offspring law P(0) = 0.2, P(2) = 0.8, what are G(s) and the extinction probability?

    0.8 s^2 - s + 0.2 = 0 has roots 1 and 0.25; the smaller is q = 0.25.

Summary

  • In a Galton-Watson process each individual has an independent random number of children; Zₙ₊₁ is a random sum.
  • E[Zₙ] = mⁿ; subcritical and critical processes die out, supercritical ones may survive.
  • The extinction probability is the smallest non-negative root of s = G(s), found by iterating G from 0.
  • Simulation and fixed-point computation agree.

Next: Practice Set: Poisson, CTMC and Random Walks