Event-Chain Monte Carlo: The global-balance breakthrough

TL;DR

Event-Chain Monte Carlo (ECMC) replaces traditional Metropolis-style random moves with rejection-free, deterministic chains that propagate activity through collisions, yielding global-balance-based sampling of the Boltzmann distribution $\pi(x) \propto e^{-\beta U(x)}$. The framework generalizes to continuous potentials via Event-Driven Monte Carlo (EDMC) and the factorized Metropolis filter, enabling localized, collision-driven dynamics that preserve the correct equilibrium while enhancing mixing in dense systems. Key contributions include lifted Metropolis schemes, the factorized acceptance concept, and efficient event-driven implementations with techniques like Cell-Veto for long-range forces, together with rigorous connections to PDMPs and the Bouncy Particle Sampler. The work has substantially impacted simulations of dense hard-sphere systems and extended to polymers, spin systems, and all-atom sampling, while highlighting practical challenges in parallelization and the potential for hybrid MD-EDMC architectures to exploit these gains in modern hardware.

Abstract

The seminal 2009 paper by Bernard, Krauth, and Wilson marked a paradigm shift in Monte Carlo sampling. By abandoning the restrictive condition of detailed balance in favor of the more fundamental principle of global balance, they introduced the Event-Chain Monte Carlo (ECMC) algorithm, which achieves rejection-free, deterministic sampling for hard spheres. This breakthrough demonstrated that persistent, directional dynamics could dramatically accelerate equilibration in dense particle systems. In this commentary, we review this foundational work and elucidate its underlying mechanism using the broader Event-Driven Monte Carlo (EDMC) framework developed in subsequent years. We show how the original hard-sphere concept naturally generalizes to continuous potentials and modern lifted Markov chain formalisms, transforming a surprising specific result into a powerful general class of sampling algorithms.

Figures (4)

• Figure 1: An event-chain move. The particle $1$ is picked to move to the right. When it collides with particle $2$, particle $1$ halts and $2$ takes over etc., forming an event chain. The motion stops of the sum of displacements add up to a preset value, $l$, which was one fifth of the box size in this case. Plot (a) shows the initial positions of particles involved in the chain, and (b) the final positions.
• Figure 2: The Metropolis scheme in 1D. Incoming and outgoing equilibrium probability flows add up to the stationary probability $\pi_i$. Detailed balance holds because the probability flows between any two nodes are symmetric.
• Figure 3: A lifted Markov-chain scheme in 1D. Each physical state $x_i$ is split into two lifted states, typically interpreted as "moving left" or "moving right". Detailed balance is violated, but global balance is maintained: for every lifted state, the sums of incoming and outgoing equilibrium flows both equal the stationary probability. Unlike in the Metropolis scheme (Fig. \ref{['fig:metropolis']}), there are no self-transitions and thus no rejections.
• Figure 4: Visualization of a purely deterministic ECMC trajectory in a 2D hard-disk system. The simulation is initialized on a lattice (box size $L=8$, disk radius $R=0.9$). The algorithm iterates through particles in sequence, moving them with a fixed event-chain length of $l=0.7\sqrt{2}$. The direction of motion alternates strictly between $+x$ and $+y$. The resulting "woven" pattern illustrates how the algorithm explores configuration space via cooperative, multi-particle chains rather than independent random walks, but still mixing the system.