Determining Lennard-Jones Parameters Using Multiscale Target Data through Presampling-Enhanced, Surrogate-Assisted...

Area of Science:

• Computational chemistry and molecular modeling.
• Development of physics-based simulation methods.

Background:

• Force field models, Newtonian approximations, are inherently noisy and difficult to scale across molecular domains.
• Parameter transferability between molecular scales (e.g., single molecules to condensed phases) remains a significant challenge.

Purpose of the Study:

• To introduce a novel surrogate-assisted algorithm for optimizing Lennard-Jones parameters.
• To address the difficulties in coupling molecular models across different scales and ensure parameter transferability.

Main Methods:

• A surrogate-assisted global evolutionary optimization method was combined with a presampling phase.
• The presampling phase leverages computationally cheaper molecular scale domains.
• Individual algorithm components were assessed for their specific contributions.

Main Results:

• The presampling method reduces the requirement for accurate initial parameter guesses.
• The surrogate model significantly enhances the efficiency of the parameter optimization process.
• The combined approach demonstrates a significant benefit for force field parametrization.

Conclusions:

• The developed surrogate-assisted algorithm effectively optimizes Lennard-Jones parameters for multi-scale molecular modeling.
• The method improves efficiency and alleviates common challenges in force field development.
• This approach offers a promising strategy for enhancing the accuracy and applicability of molecular simulations.