# Branching process with immigration

2010 Mathematics Subject Classification: Primary: 60J80 [MSN][ZBL]

A model of a branching process (discrete-time or continuous-time, with one or several types of particles, etc.) in which new particles may appear not only as the products of division of the original particles, but also as a result of immigration from some "external source" . For instance, let

$$X _ {t,i } , Y _ {t} ,\ t = 0, 1 ,\dots ; \ i = 1, 2 \dots$$

be independent random variables with generating functions

$$F (s) = \ \sum _ {k = 0 } ^ \infty {\mathsf P} \{ X _ {t,i } = k \} s ^ {k} ,$$

$$G (s) = \sum _ {k = 0 } ^ \infty {\mathsf P} \{ Y _ {t} = k \} s ^ {k} ,$$

respectively; the branching Galton–Watson process with immigration may be defined by the relations $\mu (0) = 0$, where $\mu (t)$ is the number of particles and

$$\mu (t + 1) = \ X _ {t,1 } + \dots + X _ {t, \mu (t) } + Y _ {t} ,\ \ t=0, 1 ,\dots .$$

Here, the variable $X _ {t,i}$ is interpreted as the number of daughter particles of the $i$- th particle of the $t$- th generation, while the variable $Y _ {t}$ is interpreted as the number of the particles which have immigrated into the $t$- th generation. The generating functions

$$H _ {t} (s) = \ {\mathsf E} \{ s ^ {\mu (t) } \mid \mu (0) = 0 \}$$

are given by the recurrence relations

$$H _ {0} (s) = 1,\ \ H _ {t + 1 } (s) = \ G (s) H _ {t} (F (s)).$$

The Markov chain $\mu (t)$ corresponding to the Galton–Watson branching process with immigration is recurrent if ${\mathsf E} X _ {t,i} < 1$ and ${\mathsf E} \mathop{\rm ln} (1 + Y _ {t} ) < \infty$ or ${\mathsf E} X _ {t,i} = 1$ and $B = {\mathsf D} X _ {t,i} > 2C = 2 {\mathsf E} Y _ {t}$, and is transient if ${\mathsf E} X _ {t,i} = 1$ and $B < 2C$ or ${\mathsf E} X _ {t,i} > 1$. For the Markov chain $\mu (t)$ to be ergodic, i.e. for the limits

$$\lim\limits _ {t \rightarrow \infty } \ {\mathsf P} \{ \mu (t) = k \} = p _ {k}$$

to exist and to satisfy

$$\sum _ {k = 0 } ^ \infty p _ {k} = 1,$$

it is necessary and sufficient [FW] that

$$\int\limits _ { 0 } ^ { 1 } \frac{1 - G (s) }{F (s) - s } \ ds < \infty .$$

This condition is met, in particular, if ${\mathsf E} X _ {t,i} < 1$ and ${\mathsf E} \mathop{\rm ln} (1 + Y _ {t} ) < \infty$. If ${\mathsf E} X _ {t,i} = 1$, $B > 0$, $C < \infty$, then [S]

$$\lim\limits _ {t \rightarrow \infty } {\mathsf P} \left \{ \frac{2 \mu (t) }{Bt } \leq x \right \} = \ { \frac{1}{\Gamma (2CB ^ {-1} ) } } \int\limits _ { 0 } ^ { x } y ^ {2CB ^ {-1 } -1 } e ^ {-y} dy,\ x \geq 0.$$

If $A = {\mathsf E} X _ {t,i} > 1$ and ${\mathsf E} \mathop{\rm ln} (1 + Y _ {t} ) < \infty$, then there exists [S2] a sequence of numbers $c _ {t} \downarrow 0$, $c _ {t} / c _ {t+1} \rightarrow A$, such that

$${\mathsf P} \left \{ \lim\limits _ {t \rightarrow \infty } \ c _ {t} \mu (t) \ \textrm{ exists } \textrm{ and } \ \textrm{ is } \textrm{ positive } \right \} = 1.$$

In branching processes with immigration in which the immigration takes place at $\mu (t) = 0$ only, i.e.

$$\mu (t+1) = X _ {t,1} + \dots + X _ {t, \mu (t) } + \delta _ {0, \mu (t) } Y _ {t} ,\ t=0, 1 \dots$$

where $\delta _ {ij}$ is the Kronecker symbol, the following relation is valid if ${\mathsf E} X _ {t,i} = 1$, $1 < {\mathsf E} X _ {t,i} ^ { 2 } < \infty$ and $0 < {\mathsf E} Y _ {t} < \infty$:

$$\lim\limits _ {t \rightarrow \infty } \ {\mathsf P} \left \{ \frac{ \mathop{\rm ln} (1 + \mu (t)) }{ \mathop{\rm ln} t } \leq x \right \} = x,\ 0 \leq x \leq 1.$$

How to Cite This Entry:
Branching process with immigration. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Branching_process_with_immigration&oldid=46155
This article was adapted from an original article by A.M. Zubkov (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article