Jeffreys Flow: Robust Boltzmann Generators for Rare Event Sampling via Parallel Tempering Distillation
This addresses rare event sampling for computational physics and chemistry, offering a robust solution to mode collapse in generative models, though it builds incrementally on existing Boltzmann generators.
The paper tackles the problem of rare event sampling in physical systems with rough energy landscapes by introducing Jeffreys Flow, a generative framework that uses Jeffreys divergence and parallel tempering distillation to prevent mode collapse in Boltzmann generators. The method demonstrates scalability and accuracy on non-convex benchmarks, accelerating exact importance sampling in Path Integral Monte Carlo by orders of magnitude.
Sampling physical systems with rough energy landscapes is hindered by rare events and metastable trapping. While Boltzmann generators already offer a solution, their reliance on the reverse Kullback--Leibler divergence frequently induces catastrophic mode collapse, missing specific modes in multi-modal distributions. Here, we introduce the Jeffreys Flow, a robust generative framework that mitigates this failure by distilling empirical sampling data from Parallel Tempering trajectories using the symmetric Jeffreys divergence. This formulation effectively balances local target-seeking precision with global modes coverage. We show that minimizing Jeffreys divergence suppresses mode collapse and structurally corrects inherent inaccuracies via distillation of the empirical reference data. We demonstrate the framework's scalability and accuracy on highly non-convex multidimensional benchmarks, including the systematic correction of stochastic gradient biases in Replica Exchange Stochastic Gradient Langevin Dynamics and the massive acceleration of exact importance sampling in Path Integral Monte Carlo for quantum thermal states.