Enhanced sampling of multidimensional free-energy landscapes using adaptive biasing forces

TL;DR

This paper introduces an adaptive biasing algorithm that improves sampling efficiency of complex free-energy landscapes in molecular dynamics by generalizing existing methods to multidimensional reaction coordinates.

Contribution

It presents a novel multidimensional adaptive biasing force method with theoretical convergence analysis and practical validation on realistic test cases.

Findings

Enhanced sampling of multimodal measures achieved.

The method converges reliably over long times.

Effective for weakly coupled reaction coordinates.

Abstract

We propose an adaptive biasing algorithm aimed at enhancing the sampling of multimodal measures by Langevin dynamics. The underlying idea consists in generalizing the standard adaptive biasing force method commonly used in conjunction with molecular dynamics to handle in a more effective fashion multidimensional reaction coordinates. The proposed approach is anticipated to be particularly useful for reaction coordinates, the components of which are weakly coupled, as illuminated in a mathematical analysis of the long-time convergence of the algorithm. The strength as well as the intrinsic limitation of the method are discussed and illustrated in two realistic test cases.