Free Energy Surface Sampling via Reduced Flow Matching

2026-05-01 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Physical Sciences & Chemistry, Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, medium

Summary

Sampling free energy surfaces (FES) is critical in statistical physics for understanding chemical reactions and conformational transitions, but traditional methods are computationally intensive. This paper introduces FES-FM, a novel reduced flow matching (FM) method designed for direct and efficient free energy sampling by training a dynamical transport map within the collective variable (CV) space. For many-particle systems, FES-FM constructs a prior distribution using the Hessian at a local potential minimum, ensuring physically meaningful configurations and rotation-translation invariance. Comparative experiments demonstrate that FES-FM significantly reduces computational costs while achieving superior accuracy per unit sampling time across various potential functions and collective variables. This approach offers a substantial advancement in the efficiency and precision of FES sampling.

Key takeaway

FES-FM introduces a novel reduced flow matching (FM) method for directly sampling free energy surfaces (FES) in collective variable (CV) space, addressing the high computational cost of traditional methods. It trains a dynamical transport map in CV space, utilizing a Hessian-based prior for many-particle systems to ensure rotation-translation invariance and physically meaningful configurations. This approach drastically reduces computational costs while delivering superior accuracy per unit sampling time, accelerating research in chemical reactions and conformational transitions.

Topics

• Free Energy Surface Sampling
• Reduced Flow Matching
• Collective Variables
• Statistical Physics
• Dynamical Transport Maps

Code references

• kwanseokk/SourceFM

Best for: AI Scientist, Research Scientist

Related on AIssential

• Reactive Flux Matching: Mechanism Discovery and Adaptive Sampling of Rare Events — Flux Matching learns reaction pathways and coordinates directly from reactive trajectory data without prior dynamic knowledge.
• Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage — Fun-DDPS combines diffusion models and neural operators for robust, efficient, and physics-consistent CCS modeling.
• Operator learning for the 2D incompressible Navier-Stokes equations: a conformal prediction approach in the data-scarce regime — The perturbation-based conformal prediction framework quantifies uncertainty in neural operators efficiently, especially in data-scarce regimes, by comparing noisy and original label predictions.
• The Score Hamiltonian: Mapping Diffusion Models to Adiabatic Transport — Score Hamiltonians map diffusion models to adiabatic transport, revealing sampling limits and guiding annealing schedules.
• Force-free molecular dynamics through autoregressive equivariant networks — TrajCast uses equivariant MPNNs to directly predict molecular states, enabling faster, data-efficient MD simulations with strong generalization.
• Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data — Exascale multi-task graph foundation models accelerate materials discovery by enabling billion-scale screening and data-efficient fine-tuning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.