Computing sensitivity coefficients in Brownian dynamics simulations by Malliavin weight sampling

TL;DR

This paper introduces a Malliavin weight sampling method for efficiently computing parameter sensitivities in Brownian dynamics simulations, applicable to various system states and outperforming traditional finite difference approaches.

Contribution

The paper presents a novel, easy-to-implement technique using Malliavin weights to compute sensitivities in Brownian dynamics, applicable to both equilibrium and nonequilibrium systems.

Findings

Method accurately computes sensitivities in Brownian dynamics.

Scales more efficiently than finite difference methods.

Applicable to steady state and time-dependent systems.

Abstract

We present a method for computing parameter sensitivities and response coefficients in Brownian dynamics simulations. The method involves tracking auxiliary variables (Malliavin weights) in addition to the usual particle positions, in an unperturbed simulation. The Malliavin weights sample the derivatives of the probability density with respect to the parameters of interest and are also interesting dynamical objects in themselves. Malliavin weight sampling is simple to implement, applies to equilibrium or nonequilibrium, steady state or time-dependent systems, and scales more efficiently than standard finite difference methods