Semi-Empirical Shadow Molecular Dynamics: A PyTorch Implementation.

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Molecular dynamics simulations are crucial for understanding chemical systems.
• Semi-empirical quantum-mechanical methods offer a balance between accuracy and computational cost.
• Extended Lagrangian Born-Oppenheimer molecular dynamics (XL-BOMD) has advanced capabilities for complex systems.

Purpose of the Study:

• To implement the latest shadow potential energy version of XL-BOMD in the PyTorch-based software PySeQM.
• To incorporate finite electronic temperatures, canonical density matrix perturbation theory, and Krylov subspace approximation (KSA-XL-BOMD).
• To leverage GPU and machine learning hardware accelerators for enhanced simulation performance.

Main Methods:

• Implementation of KSA-XL-BOMD within the PySeQM software using PyTorch.
• Utilizing GPU acceleration for large-scale molecular dynamics simulations.
• Integration of finite electronic temperatures and perturbation theory for electronic equations of motion.

Main Results:

• Successful implementation of KSA-XL-BOMD in PySeQM, enabling simulations of challenging chemical systems.
• Demonstrated efficiency with a simulation of 840 carbon atoms completing one time step in 4 seconds on a single Nvidia RTX A6000 GPU.
• The new formulation effectively handles systems with charge instabilities and low electronic energy gaps.

Conclusions:

• The PySeQM software with the new XL-BOMD formulation provides a powerful tool for large-scale simulations.
• GPU acceleration significantly enhances the speed and feasibility of complex quantum mechanical simulations.
• This development opens avenues for studying previously intractable chemical systems in molecular dynamics.