PySurf: A Framework for Database Accelerated Direct Dynamics.

Area of Science:

• Computational Chemistry
• Theoretical Chemistry
• Machine Learning Applications

Background:

• Computational expense of electronic structure calculations limits theoretical studies of photoinduced processes.
• Machine learning (ML) offers a promising approach to significantly reduce computational demands.
• Existing methods often require extensive computational resources for accurate dynamics simulations.

Purpose of the Study:

• Introduce PySurf, an innovative code framework for data science in computational chemistry.
• Develop a rapid prototyping tool with customizable Plugin and Workflow engines.
• Enable efficient and accurate simulations of nonadiabatic molecular dynamics.

Main Methods:

• PySurf framework with Plugin and Workflow engines for task customization.
• Database framework for automatic data storage and property interpolation.
• Implementation of energy-only nonadiabatic surface hopping simulations using Landau-Zener algorithm and interpolated potential energy surfaces.

Main Results:

• PySurf enables full-dimensional nonadiabatic surface hopping simulations using only adiabatic energies.
• Simulations of pyrazine and SO2 molecules demonstrate accurate prediction of nonadiabatic dynamics.
• Database-accelerated, energy-only simulations show competitiveness with traditional semiclassical methods.

Conclusions:

• PySurf significantly reduces computational cost for studying photoinduced processes.
• The energy-only surface hopping approach provides a computationally efficient yet accurate method.
• PySurf is a powerful tool for advancing data science applications in computational chemistry.