Lattice field theory simulations frequently encounter significant computational bottlenecks, particularly when researchers attempt to analyze systems near phase transitions or the continuum limit. These challenges often manifest as critical slowing down, where Markov chain Monte Carlo (MCMC) simulations become increasingly inefficient.
To address this, researchers have developed a novel approach called the Stochastic Path Sampler (SPS), which leverages principles from nonequilibrium thermodynamics to generate configurations for unnormalized target distributions without requiring prior training data. The method operates by establishing a trajectory-level balance between forward and backward stochastic dynamics, connecting the system’s prior and target states.
The central innovation of the SPS involves minimizing path-space variational free energy, a metric linked to the entropy-production upper bound. By defining this bound through the log-ratio of forward and auxiliary backward trajectory measures, the researchers enhance the reversibility of the processes involved. This variational framework allows the forward process to provide high-quality, independent proposals, which are then refined through an extended-space Independence Metropolis-Hastings correction step.
Empirical tests of the sampler were conducted using two-dimensional φ4 theory, where the authors compared the performance of their neural sampler against Hybrid Monte Carlo (HMC). The findings indicate that the SPS achieves comparable sampling quality to HMC while demonstrating a significantly shorter autocorrelation time within the pseudocritical region. This reduction in autocorrelation suggests a substantial improvement in sampling efficiency for the specified theoretical model.
This study introduces a stochastic-quantization-inspired method for data-free proposal construction within lattice field theory. By utilizing the variational free-energy principle derived from path-space irreversibility, the researchers offer a new path forward for computational physics.
Subject of Research: Lattice Field Theory Computational Physics
Article Title: Stochastic path sampler for lattice field theory
Article References: Chen, S., Qian, M., Aarts, G., Lucini, B., & Zhou, K. (2026). Stochastic path sampler for lattice field theory. Journal of High Energy Physics, 2026(11), Article 164. https://doi.org/10.1007/jhep11(2026)164
Image Credits: AI Generated
Keywords: Lattice Field Theory, Stochastic Path Sampler, Nonequilibrium Thermodynamics, Markov Chain Monte Carlo, Computational Physics, Variational Free Energy, Independence Metropolis-Hastings, Theoretical Physics, High Energy Physics, Autocorrelation Time, Neural Sampler, Path-space Irreversibility, Sampling Efficiency, Pseudocritical Region, Data-free Proposal
News Source: Denise Maddox. (November 17, 2026). New Stochastic Path Sampler Improves Efficiency in Lattice Field Theory. Scienmag.



