Definition
A superprocess is a class of measure‑valued stochastic processes that model the random evolution of a continuous mass distribution subject to both spatial motion and branching (reproduction or death). It generalizes classical branching particle systems by allowing an infinite number of infinitesimally small particles, resulting in a continuous‑state, often diffusion‑type, process.
Overview
Superprocesses arise as scaling limits of discrete branching particle systems, such as branching Brownian motion or branching random walks, when the number of particles tends to infinity while individual particle mass tends to zero. The most studied examples include the Dawson–Watanabe superprocess (also called the super‑Brownian motion) and the Fleming–Viot process. They are employed in probability theory, population genetics, statistical physics, and various applications requiring stochastic modeling of spatially distributed mass or population densities.
Mathematically, a superprocess $(X_t)_{t\ge 0}$ takes values in the space of finite measures on a Polish space $E$ (often $E=\mathbb{R}^d$). Its dynamics are characterized by a spatial motion component—typically a diffusion process generated by a second‑order differential operator $L$—and a branching mechanism described by a function $\Phi$ governing the reproduction rate. The probabilistic law of $X_t$ is frequently defined via a martingale problem, a stochastic partial differential equation (SPDE), or a log‑Laplace functional equation.
Etymology/Origin
The term “superprocess” reflects its nature as a “super‑” (i.e., extended or generalized) version of classical branching processes. The concept was first rigorously introduced in the early 1970s by Donald Dawson, Edwin Perkins, and others, who investigated scaling limits of critical branching particle systems. The name combines “super” (indicating the passage to a continuous‑mass setting) with “process,” a standard term for a stochastic evolution.
Characteristics
- State Space: Finite (or locally finite) measures on a spatial domain $E$; often equipped with the weak‑* topology.
- Spatial Motion: Determined by a Markov process on $E$ (e.g., Brownian motion, symmetric stable process).
- Branching Mechanism: Given by a branching (or Lévy) exponent $\Phi(\lambda) = -\beta\lambda + \alpha\lambda^2 + \int_{(0,\infty)} (e^{-\lambda r} - 1 + \lambda r) , \pi(dr)$, where $\beta$, $\alpha$, and $\pi$ encode death, binary branching, and more general offspring distributions.
- Log‑Laplace Functional: For a non‑negative test function $f$, the functional $u(t,x)= -\log \mathbb{E}_{\delta_x}[e^{-\langle X_t, f\rangle}]$ solves a nonlinear partial differential equation of the form $\partial_t u = L u - \Phi(u)$.
- Martingale Problem: The process can be characterized as the unique solution to a martingale problem involving operators that combine the spatial generator $L$ and the branching term.
- Scaling Limits: Obtained as limits of critical branching particle systems under diffusive scaling (space scaled by $\varepsilon^{-1}$, time by $\varepsilon^{-2}$, particle mass by $\varepsilon$).
- Path Properties: Sample paths are generally continuous in the weak‑* topology, though the total mass process may possess jumps if the branching mechanism includes a Lévy component.
Related Topics
- Branching Processes – Discrete‑state stochastic models of reproduction (e.g., Galton–Watson processes).
- Measure‑Valued Diffusions – General class of processes that include superprocesses as particular cases.
- Dawson–Watanabe Superprocess – The prototypical superprocess with quadratic branching (binary splitting).
- Fleming–Viot Process – Measure‑valued process modeling gene frequencies with resampling.
- Stochastic Partial Differential Equations (SPDEs) – Many superprocesses satisfy SPDEs such as the stochastic heat equation with multiplicative noise.
- Historical Processes – Extensions of superprocesses that record ancestral lineages.
- Interacting Particle Systems – Discrete models whose scaling limits can yield superprocesses.
Superprocesses constitute a central object of study in modern probability theory, connecting stochastic analysis, partial differential equations, and various applied fields that require modeling of randomly evolving spatial mass distributions.