Iterative unbiasing of quasi-equilibrium sampling

TL;DR

This paper presents an iterative unbiasing method that efficiently reconstructs unbiased distributions from biased quasi-equilibrium sampling data, applicable to high-dimensional collective variables and complex systems.

Contribution

It introduces a novel iterative unbiasing scheme that leverages all trajectory data without grid evaluation, improving upon existing adaptive sampling re-weighting methods.

Findings

The method accurately recovers unbiased distributions in model systems.

It performs well with high-dimensional bias potentials.

Benchmark results show advantages over existing schemes.

Abstract

Atomistic modelling of phase transitions, chemical reactions, or other rare events that involve overcoming high free energy barriers usually entails prohibitively long simulation times. Introducing a bias potential as a function of an appropriately-chosen set of collective variables can significantly accelerate the exploration of phase space, albeit at the price of distorting the distribution of microstates. Efficient re-weighting to recover the unbiased distribution can be nontrivial when employing adaptive sampling techniques such as Metadynamics, Variationally Enhanced Sampling or Parallel Bias Metadynamics, in which the system evolves in a quasi-equilibrium manner under a time-dependent bias. We introduce an iterative unbiasing scheme that makes efficient use of all the trajectory data, and that does not require the distribution to be evaluated on a grid. The method can thus be used even when the bias has a high dimensionality. We benchmark this approach against some of the existing schemes, on models systems with different complexities and dimensionalities.