Toward transferable empirical valence bonds: Making classical force fields reactive.

Area of Science:

• Computational chemistry
• Materials science
• Chemical physics

Background:

• Classical force fields are essential for atomistic simulations but struggle to model reactive processes.
• The empirical valence bond (EVB) technique extends classical force fields to reactive simulations.
• EVB requires extensive parametrization, often involving computationally expensive quantum mechanical calculations or experimental data for each reaction.

Purpose of the Study:

• To develop a novel method for predicting EVB parameters using machine learning.
• To eliminate the need for quantum mechanical calculations in EVB parametrization.
• To facilitate the use of classical force fields for reactive atomistic simulations.

Main Methods:

• A neural network model was trained to predict EVB parameters.
• The model utilized SMILES strings, a chemical notation, to describe chemical reactions.
• The predicted parameters were integrated into classical force field simulations.

Main Results:

• The neural network successfully predicted EVB parameters from reaction SMILES strings.
• This approach obviates the necessity for quantum mechanical calculations during parametrization.
• The method demonstrates the feasibility of using machine learning for EVB parameter prediction.

Conclusions:

• A new procedure for enabling reactive atomistic simulations has been established.
• Researchers can now leverage existing classical force fields for reactive simulations without additional quantum calculations.
• This work paves the way for more accessible and efficient reactive simulations in computational chemistry.