Active-Learning Assisted General Framework for Efficient Parameterization of Force-Fields.

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Sulfone molecules are crucial for energy storage electrolytes.
• Accurate molecular modeling is essential but traditional methods are inefficient.
• Existing force field parametrization is computationally expensive and labor-intensive.

Purpose of the Study:

• To develop an efficient and accurate method for optimizing force field parameters for sulfone molecules.
• To enhance the predictive power of molecular simulations for sulfone-based electrolytes.
• To reduce the computational cost and manual effort in force field development.

Main Methods:

• Integration of genetic algorithms (GA) and Gaussian process regression (GPR) in an active learning framework.
• Optimization of force field parameters using a limited dataset (300 points) over 12 iterations.
• Validation against experimental data (density, viscosity, diffusion, surface tension) and comparison with state-of-the-art techniques.

Main Results:

• Achieved optimized force field parameters with significantly fewer iterations and data points compared to traditional methods.
• Demonstrated excellent agreement between GA-GPR predictions and experimental values, outperforming the OPLS force field.
• Successfully captured bulk and interfacial properties, including molecular mobility and caging effects.

Conclusions:

• The GA-GPR approach provides a robust, transferable, and efficient force field for sulfone molecules.
• This method significantly enhances the accuracy and efficiency of molecular simulations for energy storage applications.
• Establishes a foundation for future machine learning-driven force field development in complex molecular systems.