VENUSpy: A Chemical Dynamics Simulation Program in the Era of Machine Learning and Exascale Computing

Area of Science:

• Computational Chemistry
• Chemical Physics
• Materials Science

Background:

• Traditional ab initio molecular dynamics (AIMD) provides high accuracy but is computationally expensive, limiting simulations to small systems and short timescales.
• Advances in machine learning (ML) potentials and exascale computing offer new avenues for developing potential energy surfaces (PESs).
• These advancements enable simulations of chemical reaction dynamics for larger systems and longer durations.

Purpose of the Study:

• Introduce VENUSpy, a Python-based reimplementation and extension of the VENUS code.
• Facilitate the integration of ML potentials and exascale-ready quantum chemistry packages like NWChemEx.
• Enable advanced ML/ab initio hybrid dynamics simulations for complex reactive systems.

Main Methods:

• Developed VENUSpy as a Python framework, extending the classical VENUS code.
• Demonstrated interfacing capabilities with NWChemEx's top-level classes and objects for reaction dynamics.
• Integrated modern Python tools to enhance the original code's versatility, including initial sampling and trajectory propagation.
• Enabled ML/ab initio hybrid dynamics simulations.

Main Results:

• VENUSpy successfully interfaces with ML potentials and the developing NWChemEx package.
• The framework preserves the core functionalities of the original VENUS code while adding modern Python integration.
• Demonstrated the capability for ML/ab initio hybrid dynamics simulations, offering advantages over standalone AIMD and MLMD.
• Facilitated rapid development of new methodologies for studying complex reactive systems.

Conclusions:

• VENUSpy provides a versatile, modular framework for advanced chemical reaction dynamics simulations.
• It bridges the gap between high-accuracy AIMD and efficient ML-based methods.
• The tool accelerates research in complex reactive systems by enabling hybrid simulation approaches.