Publisher
University of Tennessee at Chattanooga
Place of Publication
Chattanooga (Tenn.)
Abstract
Water is an important component in many scientific and engineering applications, but precisely modeling its thermal, electrical, and physical properties remains difficult in molecular dynamics (MD) simulations. The widely used TIP4P water model, while useful in many circumstances, requires parameterization to better agreement with experimental results under various conditions. In this study, interpretable machine learning techniques were used to systematically tune the TIP4P model parameters where mathematical relationship between features and targets has been developed using deep symbolic optimization (DSO), resulting in a more accurate depiction of water behavior in molecular dynamics simulation. A data-driven approach was created to optimize critical model parameters while maintaining physical interpretability by combining molecular dynamics simulations and neural network-based modeling. A new optimized TIP4P water model has been developed to reproduce the thermal conductivity, diffusion coefficient, density and dielectric constant with accurate dipole calculations.
Document Type
posters
Language
English
Rights
http://rightsstatements.org/vocab/InC/1.0/
License
http://creativecommons.org/licenses/by/4.0/
Recommended Citation
Dey, Khowshik; Barisik, Murat; and Liang, Yu, "Machine Learning for Reparameterization of Molecular Dynamics Water Model to Accurately Capture Electrical and Thermal Behavior". ReSEARCH Dialogues Conference proceedings. https://scholar.utc.edu/research-dialogues/2025/posters/10.
Machine Learning for Reparameterization of Molecular Dynamics Water Model to Accurately Capture Electrical and Thermal Behavior
Water is an important component in many scientific and engineering applications, but precisely modeling its thermal, electrical, and physical properties remains difficult in molecular dynamics (MD) simulations. The widely used TIP4P water model, while useful in many circumstances, requires parameterization to better agreement with experimental results under various conditions. In this study, interpretable machine learning techniques were used to systematically tune the TIP4P model parameters where mathematical relationship between features and targets has been developed using deep symbolic optimization (DSO), resulting in a more accurate depiction of water behavior in molecular dynamics simulation. A data-driven approach was created to optimize critical model parameters while maintaining physical interpretability by combining molecular dynamics simulations and neural network-based modeling. A new optimized TIP4P water model has been developed to reproduce the thermal conductivity, diffusion coefficient, density and dielectric constant with accurate dipole calculations.