Energy-Weighted Flow Matching: Unlocking Continuous Normalizing Flows for Efficient and Scalable Boltzmann Sampling

2026-04-22 · Source: stat.ML updates on arXiv.org · Field: Science & Research — Mathematics & Computational Sciences, Physical Sciences & Chemistry, Life Sciences & Biology · Depth: Expert, extended

Summary

Researchers from the Technical University of Munich introduced Energy-Weighted Flow Matching (EWFM), a novel training objective that enables continuous normalizing flows (CNFs) to model Boltzmann distributions using only energy function evaluations, bypassing the need for target samples. This objective reformulates conditional flow matching via importance sampling, allowing training with samples from arbitrary proposal distributions. Based on EWFM, two algorithms were developed: iterative EWFM (iEWFM), which refines proposals progressively, and annealed EWFM (aEWFM), which incorporates temperature annealing for complex energy landscapes. On benchmark systems, including challenging 55-particle Lennard-Jones clusters, these algorithms achieved sample quality competitive with state-of-the-art energy-only methods like Flow Annealed Importance Sampling Bootstrap (FAB) and Iterated Denoising Energy Matching (iDEM), while requiring up to three orders of magnitude fewer energy evaluations than iDEM.

Key takeaway

For AI Scientists and Research Scientists working on molecular dynamics or computational chemistry, EWFM offers a significant advancement in Boltzmann sampling. You can now train highly expressive continuous normalizing flows using only energy function evaluations, drastically reducing the computational cost associated with generating target samples. Consider implementing iEWFM or aEWFM to achieve competitive sample quality with substantially fewer energy evaluations, especially for complex, high-dimensional systems like Lennard-Jones clusters.

Key insights

Energy-Weighted Flow Matching enables continuous normalizing flows to model Boltzmann distributions using only energy evaluations, not target samples.

Principles

• Importance sampling can reformulate generative model objectives.
• Iterative refinement improves proposal distributions for stable training.
• Temperature annealing aids learning on complex energy landscapes.

Method

EWFM reformulates conditional flow matching with importance sampling, reweighting loss by Boltzmann weight and proposal density. iEWFM iteratively refines proposals, while aEWFM adds temperature annealing.

In practice

• Use iEWFM for high-dimensional molecular sampling.
• Employ aEWFM for systems with challenging energy landscapes.
• Buffer samples to amortize CNF density calculation costs.

Topics

• Energy-Weighted Flow Matching
• Continuous Normalizing Flows
• Boltzmann Sampling
• Importance Sampling
• Temperature Annealing

Best for: AI Scientist, Research Scientist

Related on AIssential

• Decision-Weighted Flow Matching for Contextual Stochastic Optimization — Decision-Weighted Flow Matching (DW-FM) aligns generative model training with downstream decision regret by reweighting objectives based on decision sensitivity.
• Hybridizing Equilibrium Propagation with Ising Machines for Efficient Energy-Based Learning — Hybridizing EP with Ising dynamics improves energy-based learning by overcoming local minima and accelerating convergence.
• Neural Galerkin Normalizing Flows for Bayesian Inference of Diffusions with Inaccessible Boundaries — Neural Galerkin Normalizing Flows efficiently learn diffusion transition densities for Bayesian inference, bypassing complex analytical challenges.
• Flow Matching for Efficient and Scalable Data Assimilation — EnFF uses flow matching to create a training-free, scalable data assimilation framework with improved cost-accuracy.
• Strong Stochastic Flow Maps — Strong Stochastic Flow Maps (SSFMs) enable few-step, strong convergence sampling for additive-noise SDEs by learning direct solution maps.
• Theoretical guidelines for annealed Langevin dynamics in compositional simulation-based inference — Theoretical guidelines for annealed Langevin dynamics hyperparameters ensure sampling accuracy in compositional SBI, favoring Linhart's method.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.