AdaptiveBandit: A Multi-armed Bandit Framework for Adaptive Sampling in Molecular Simulations.

Area of Science:

• Computational chemistry
• Molecular dynamics
• Statistical mechanics

Background:

• Sampling equilibrium distributions in molecular simulations is computationally intensive due to high-dimensional conformational spaces.
• Various unbiased sampling methods, including parallel and adaptive sampling, have been developed to address this challenge.

Purpose of the Study:

• To develop a novel adaptive sampling algorithm for molecular simulations.
• To frame adaptive sampling strategies within the context of multi-armed bandit theory.

Main Methods:

• Recasting adaptive sampling schemes using multi-armed bandit principles.
• Developing and implementing the AdaptiveBandit algorithm.
• Testing the algorithm on simplified potentials and a protein folding scenario.

Main Results:

• The AdaptiveBandit algorithm demonstrates comparable or superior performance to existing methods across various test potentials.
• The multi-armed bandit framework provides a robust foundation for developing new sampling algorithms.
• The proposed method shows promise for enhancing sampling efficiency in complex molecular systems.

Conclusions:

• The AdaptiveBandit algorithm offers an effective approach to improve sampling in molecular simulations.
• The multi-armed bandit framework presents a novel perspective for advancing molecular simulation sampling techniques.
• This work contributes to the development of more efficient computational methods for studying molecular systems.