Metropolis Importance Sampling for Rugged Dynamical Variables

TL;DR

The paper introduces a funnel transformation for Metropolis sampling that biases probabilities towards the global energy minimum, improving efficiency in rugged energy landscapes, demonstrated on Met-Enkephalin with a twofold computational gain.

Contribution

It proposes a novel recursive funnel transformation to enhance Metropolis sampling in rugged energy landscapes, with a simple approximation tested on a biomolecular system.

Findings

Twofold computational speed-up for Met-Enkephalin at 300 K

RM$_1$ effectively biases sampling towards global minima

Simple implementation suitable for rugged systems

Abstract

A funnel transformation is introduced, which acts recursively from higher towards lower temperatures. It biases the a-priori probabilities of a canonical or generalized ensemble Metropolis simulation, so that they zoom in on the global energy minimum, if a funnel exists indeed. A first, crude approximation to the full transformation, called rugged Metropolis one (RM$_1$), is tested for Met-Enkephalin. At 300$ $K the computational gain is a factor of two and, due to its simplicity, RM$_1$ is well suited to replace the conventional Metropolis updating for these kind of systems.