Interacting Particle Systems — Mathematics Knowledge System

Overview

Interacting particle systems are continuous-time Markov processes on state spaces of the form {0,1}^S or {1,...,k}^S where S is a countable set (often Z^d). Each site hosts a particle with a state, and particles evolve according to random rules that depend on the states of nearby sites. Classical examples include the contact process (spread of infection), the voter model (opinion dynamics), and the exclusion process (particles on a lattice). These models bridge probability theory and statistical physics, exhibiting phase transitions, ergodicity, and hydrodynamic limits.

Intuition

Imagine a grid where each cell is either occupied (1) or empty (0). Particles jump to neighbouring sites, infect neighbours, or vote like their majority neighbours. The resulting system is an IPS. Key questions: Does it reach a unique stationary distribution (ergodicity)? Does the density of particles (1s) have a deterministic limit as the system size grows (hydrodynamic limit)? Does it exhibit a phase transition -- a sharp threshold where the long-run behaviour changes qualitatively (e.g., the contact process dies out for small infection rate but survives for large rate)?

Formal Definition

Definition

An interacting particle system is a Markov process on {0,1}^S (or {1,...,k}^S for k states) with generator L f(eta) = sum_{x in S} c(x, eta) * [f(eta^x) - f(eta)] where eta is a configuration, eta^x is the configuration with state at x flipped, and c(x, eta) is the flip rate at x depending on the local configuration. The contact process has c(x, eta = 1) = 1 (recovery rate) and c(x, eta = 0) = lambda * (number of infected neighbours) (infection rate).

Lf(\eta) = \sum_{x \in S} c(x, \eta)[f(\eta^x) - f(\eta)]

Generator of an IPS

c_\mathrm{CP}(x, \eta) = \begin{cases} 1 & \text{if } \eta(x) = 1 \\ \lambda \sum_{y \sim x} \eta(y) & \text{if } \eta(x) = 0 \end{cases}

Contact process flip rates

c_\mathrm{voter}(x, \eta) = |\{y \sim x : \eta(y) \ne \eta(x)\}| / |\partial x|

Voter model flip rates

\lambda_c = \sup\{\lambda : \text{contact process dies out for all initial conditions}\}

Critical infection rate

Notation

Notation	Meaning
$\eta \in \{0,1\}^S$	Configuration of the particle system
$\eta^x$	Configuration with state at x flipped
$c(x, \eta)$	Flip rate at site x in configuration eta
$\lambda_c$	Critical infection rate of the contact process

Theorems

Theorem 1: Phase Transition for Contact Process

The contact process on Z^d has a critical infection rate lambda_c > 0 such that: for lambda < lambda_c the process dies out (goes to all-0 state) from any initial condition; for lambda > lambda_c there exists a non-trivial stationary distribution under which the process survives with positive probability. On Z, lambda_c = 1.64...

Theorem 2: Voter Model Duality

The voter model is dual to coalescing random walks: the probability that sites x_1, ..., x_k all agree at time t in the voter model equals the probability that k coalescing random walks started from x_1, ..., x_k have all coalesced into one walk by time t. This duality implies that in d >= 3, the voter model clusters (any two sites eventually agree with probability 1).

Theorem 3: Hydrodynamic Limit

For the exclusion process (SSEP) on Z/NZ with N sites, the empirical density rho_N(t, x) = (1/N)*sum_{k=1}^N eta_{[Nx]}(t) converges as N -> inf to the solution of the heat equation: d/dt rho = (1/2)*Delta rho. This is the hydrodynamic limit: the microscopic particle dynamics gives the macroscopic diffusion equation.

Worked Examples

1
In the voter model, each site copies a randomly chosen neighbour's opinion at rate 1.
2
The voter model on Z^2 is recurrent (2D random walk is recurrent). By the duality with coalescing random walks, two walks from any two sites eventually meet.
$P(\text{two coalescing walks from } x, y \text{ meet}) = 1$
3
As t -> inf, any two sites eventually agree with probability 1. The system clusters into large monochromatic regions.
4
The stationary distributions are the pointmasses on all-0 and all-1 (consensus states). From any initial condition, the system converges weakly to a mixture of these.
$\eta_t \Rightarrow \rho \delta_{\mathbf{1}} + (1-\rho)\delta_{\mathbf{0}}$

✓ Answer

On Z^2, the voter model clusters: any two sites eventually agree with probability 1. The system converges to a mixture of all-0 and all-1 states with mixture probability equal to the initial density.

Practice Problems

Mediumfree response

Describe the exclusion process and explain why it is particle-conserving. What is the hydrodynamic limit?

Common Mistakes

Common Mistake

Thinking all IPS have a unique stationary distribution.

Many IPS have multiple stationary distributions: the voter model has an uncountable family of stationary measures on the infinite lattice. The ergodicity question (whether there is a unique stationary measure) is non-trivial and depends on the specific system and parameters.

Quiz

The contact process on Z undergoes a phase transition at lambda_c ≈ 1.64, meaning:

For any lambda, the infection survives with positive probabilityFor lambda < lambda_c the infection dies out; for lambda > lambda_c it survivesThe infection always dies out in finite timeThe critical value depends on the initial configuration

Historical Background

Interacting particle systems were introduced by Frank Spitzer in 1970 and systematically developed by Thomas Liggett in his landmark 1985 monograph. Harris (1972) proved the first major results on the contact process. Spohn (1991) connected the theory to hydrodynamics and scaling limits. The exclusion process (ASEP in particular) became central to the KPZ universality class and random matrix theory.

1970
Spitzer introduces interacting particle systems
Frank Spitzer
1972
Harris proves existence and ergodicity for contact process
Theodore Harris
1985
Liggett's monograph establishes the mathematical theory
Thomas Liggett
1986
Kardar, Parisi, Zhang introduce the KPZ equation, connected to ASEP
Mehran Kardar, Giorgio Parisi, Yi-Cheng Zhang

Summary

IPS are continuous-time Markov processes on configuration spaces, modelling locally interacting particles.
Contact process: infection spreading on a lattice with phase transition at lambda_c > 0.
Voter model: opinion dynamics dual to coalescing random walks; clusters in d <= 2.
Exclusion process: conserved particles diffusing on a lattice; hydrodynamic limit is the heat equation.

References

BookLiggett, T. Interacting Particle Systems. Springer, 1985.
BookLiggett, T. Stochastic Interacting Systems: Contact, Voter and Exclusion Processes. Springer, 1999.
WebsiteWikipedia -- Interacting particle system