Continuous-Time Multitype Branching Process

• Continuous-time multitype branching processes are Markov models that describe populations evolving through stochastic reproduction, death, and type changes.
• They employ branching mechanisms, infinitesimal generators, and Lamperti-type representations to model complex inter-type interactions and scaling limits.
• Analyses classify these processes into subcritical, critical, and supercritical regimes, while also addressing extinction criteria and backbone decompositions for genealogical insights.

A continuous-time multitype branching process is a Markov process modeling the joint evolution of populations with multiple discrete or continuous "types" (often interpreted as species, lineages, or genetic classes), where individuals reproduce and die stochastically, possibly changing type according to prescribed probabilistic rules. These processes are fundamental in stochastic population theory, mathematical genetics, and the analysis of random trees and genealogies.

1. Formal Definition and Basic Structure

Let $d \ge 1$ denote the number of types, labeled $1,\ldots,d$. The process is described by a vector-valued Markov process $$Z_t = (Z^{(1)}_t, \ldots, Z^{(d)}_t) \in \mathbb{Z}_+^d$$ (discrete populations) or $$X_t = (X_t^{(1)}, ..., X_t^{(d)}) \in \mathbb{R}_+^d$$ (continuous mass or measure). The process evolves under the following rules:

• Reproduction (multitype offspring):
  • Each type-$i$ individual lives for an independent $\text{Exp}(\lambda_i)$ amount of time, then dies, producing a random vector $(n_1, \ldots, n_d)$ offspring in types $1,\ldots,d$ with law $\nu_i(n_1, ..., n_d)$.
• Branching property:
  • Conditioned on the present, the future evolution of each individual's progeny is independent.

For continuous-state analogues (MCSBP/MCBI), the transition semigroup admits an affine Laplace transform

$\E_x\left[e^{-\langle f, X_t \rangle}\right] = \exp\left\{ -\langle x, v(t, f) \rangle \right\},$

where $v(t, f)$ solves a vector-valued system of ODEs encoding the (possibly nonlocal) branching mechanisms, jump measures, and cross-type interactions (Kyprianou et al., 2017, Casanova et al., 2024, Barczy et al., 2014, Kyprianou et al., 2016).

2. Generating Mechanisms and Infinitesimal Generators

The process is fully characterized by its vector-valued branching mechanism $\Psi = (\Psi_1, ..., \Psi_d)$, each typically of Lévy–Khintchine type:

$$\Psi_i(u) = -a_i u_i + \frac{1}{2}q_i u_i^2 + \sum_{j\ne i} -a_{i,j} u_j + \int_{(0,\infty)^d}\left(e^{-\langle u,x\rangle} - 1 + \langle u, x \rangle\mathbf{1}_{\{\|x\|\leq 1\}}\right)\mu_i(dx),$$

where $a_i$, $a_{i,j}$, $q_i$, and $\mu_i$ encode the per-type drift, cross-type mutation, variance, and jump intensities, respectively (Chaumont et al., 2021, Kyprianou et al., 2017). The case with immigration incorporates additional drift and jump terms (Barczy et al., 2014, Barczy et al., 2014).

The generator $\A$ of a continuous-state process on $C^2_c(\mathbb{R}_+^d)$ takes the integro-differential form

$\A f(x) = \sum_{i=1}^d c_i x_i \partial_{ii}^2 f(x) + \langle \beta + Bx, \nabla f(x) \rangle + \int_{\mathbb{R}_+^d\setminus\{0\}}[f(x+z)-f(x)]\,\nu(dz) + \sum_{i=1}^d x_i\!\int_{\mathbb{R}_+^d\setminus\{0\}}[f(x+z)-f(x)-z_i \partial_{i}f(x)]\,\mu_i(dz),$

for admissible parameters $(c, \beta, B, \nu, \mu)$ (Barczy et al., 2014, Barczy et al., 2014).

3. Lamperti Representation and Pathwise Construction

Multitype continuous-time processes admit an explicit Lamperti-type representation: letting $X^{(j)}$ be independent $\mathbb{R}^d$-valued Lévy processes (with $X^{(j),i}$ spectrally positive if $i=j$, subordinator otherwise), the process can be constructed as the unique nonnegative strong solution to

$$Z_t^{(i)} = r_i + \sum_{j=1}^d X^{(j),i}\left( \int_0^t Z_s^{(j)}\,ds \right), \quad i=1, ..., d.$$

This representation is equivalent to a system of time-changed Lévy fields whose Laplace exponents match the vector branching mechanism $\Psi$. These couplings underpin many scaling limits and genealogical encodings (Chaumont et al., 2021, Fittipaldi et al., 2022, Jr, 2021, Hernández et al., 7 Feb 2025).

4. Asymptotic Behavior and Classification

The process exhibits trichotomy (subcritical, critical, supercritical) governed by the spectral properties of the mean matrix associated to the first-moment semigroup or Laplace exponents:

• Spectral classification:
  • The spectral radius or principal eigenvalue $\rho$ (discrete) or $\lambda_1$ (continuous) of the mean matrix (or its exponentially-evolved counterpart) determines growth/extinction regimes (Chaumont et al., 2015, Kyprianou et al., 2017, Barczy et al., 2014).
• Growth theorems:
  • In the supercritical regime ($\lambda_1>0$), $e^{-\lambda_1 t} Z_t \to W \pi$ almost surely, with $W$ a nondegenerate random variable and $\pi$ the left Perron (eigen/vector) (Kyprianou et al., 2017, Chaumont et al., 2015). Precise analogues extend to integer and measure-valued processes (Kyprianou et al., 2016).
• Critical case:
  • For critical irreducible processes with immigration, scaling limits converge to squared Bessel (Feller) diffusions on rays determined by the Perron vector $u$ of the mean matrix (Barczy et al., 2014).

5. Extinction Criteria and Times

Extinction analysis hinges on the branching mechanism. For continuous-state branching processes (MCSBP), the probability of extinction at infinity is

$$\P_r\left( \lim_{t\to\infty} Z_t = 0 \right) = \exp\left( -\langle r, \Phi(0) \rangle \right),$$

where $\Phi(0)$ solves $\Psi(\mathbf{u}) = 0$ for $\mathbf{u}>0$ (Chaumont et al., 2021). The extension of Grey's criterion establishes that extinction occurs in finite time if and only if

$$\int^\infty \frac{ds}{\Psi_i(s\,e_i)} < \infty, \quad \forall\, i,$$

where $\Psi_i(s\,e_i)$ is the one-dimensional diagonal branching mechanism (Chaumont et al., 2021). This aligns with classical one-type results but reveals nuanced dependence on cross-type interactions.

6. Genealogies, Coalescence, and Backbone Decompositions

Genealogies of multitype continuous-time branching processes are encoded via forests (in the discrete skeleton) or via Poissonian sampling and duality with multitype $\Lambda$-coalescents (Hernández et al., 7 Feb 2025, Johnston et al., 2019, Casanova et al., 2024):

• Genealogical scaling limits: Under suitable limits and time-changes, genealogies of uniform samples from critical multitype processes converge to universal structures, such as Kingman's coalescent, with type structure introduced via size-bias and inter-type dependencies (Hernández et al., 7 Feb 2025, Johnston et al., 2019).
• Backbone decomposition: In the supercritical regime, the process admits a prolific-line (backbone) decomposition: a multitype Galton–Watson process (the backbone) along which conditionally independent subcritical clusters immigrate at rates determined by excursion (Dynkin–Kuznetsov) measures, generalizing the classical backbone decomposition (Fekete et al., 2018).

7. Special Cases: Search Tree Processes, Interacting Branching, and Ergodicity

• Continuous-time $m$-ary search tree processes: Model node-composition with nontrivial phase transitions (Gaussian for $m \le 26$, non-Gaussian for $m\ge 27$). The limiting distribution of second-order terms solves a complex-valued smoothing equation with explicit contraction, Fourier, and Mandelbrot cascade arguments (Chauvin et al., 2011).
• Interacting multitype branching: Recent models include explicit interaction terms motivated by stochastic Lotka–Volterra dynamics, leading to strong solutions of SDEs with cross-type quadratic drifts and generalized Lamperti representations (Fittipaldi et al., 2022).
• Ergodicity and long-term stability: Exponential ergodicity, Wasserstein convergence, and stationary distributions are established in mixed-state models and CBI processes under spectral and moment conditions, with explicit generator characterizations (Chen et al., 2021, Barczy et al., 2014).

---

References:

• "On mutations in the branching model for multitype populations" (Chaumont et al., 2015)
• "Almost sure growth of supercritical multi-type continuous state branching process" (Kyprianou et al., 2017)
• "The coalescent structure of multitype continuous-time Galton-Watson trees" (Hernández et al., 7 Feb 2025)
• "Limit distributions for multitype branching processes of m-ary search trees" (Chauvin et al., 2011)
• "Continuous Time Mixed State Branching Processes and Stochastic Equations" (Chen et al., 2021)
• "On multitype Branching Processes with Interaction" (Fittipaldi et al., 2022)
• "Extinction times of multitype, continuous-state branching processes" (Chaumont et al., 2021)
• "Extinction properties of multi-type continuous-state branching processes" (Kyprianou et al., 2016)
• "Multitype $Λ$-coalescents and continuous state branching processes" (Casanova et al., 2024)
• "Backbone decomposition of multitype superprocesses" (Fekete et al., 2018)
• "Encoding multitype Galton-Watson forests and a multitype Ray-Knight theorem" (Jr, 2021)
• "Asymptotic behavior of critical irreducible multi-type continuous state and continuous time branching processes with immigration" (Barczy et al., 2014)
• "Multitype branching process with nonhomogeneous Poisson and generalized Polya immigration" (Rabehasaina et al., 2019)
• "Stochastic differential equation with jumps for multi-type continuous state and continuous time branching processes with immigration" (Barczy et al., 2014)
• "The coalescent structure of uniform and Poisson samples from multitype branching processes" (Johnston et al., 2019)