Flexible Framework for Surface Hopping: From Hybrid Schemes for Machine Learning to Benchmarkable Nonadiabatic...

Area of Science:

• Computational Chemistry
• Quantum Dynamics
• Photochemistry

Background:

• Nonadiabatic molecular dynamics (NAMD) is crucial for understanding photochemical and photophysical processes.
• Surface hopping schemes are widely used NAMD methods, integrable with quantum chemistry and machine learning.
• Existing methods require efficient and flexible computational frameworks.

Purpose of the Study:

• To present a flexible framework for NAMD in the MLatom package.
• To implement Tully's fewest-switches surface hopping algorithm and its time-dependent Baeck-An variant.
• To demonstrate the framework's utility through representative examples and analysis tools.

Main Methods:

• Implementation of Tully's fewest-switches surface hopping and its time-dependent Baeck-An variant within MLatom.
• Utilizing user-defined custom models for energy, gradients, and nonadiabatic couplings.
• Comparison of curvature-driven surface hopping schemes, including Landau-Zener and time-dependent Baeck-An.

Main Results:

• Demonstrated computational time savings with flexible user-defined models.
• Showcased the utility of custom models for benchmarking machine learning models.
• The Landau-Zener approach showed superior performance over the time-dependent Baeck-An scheme in comparisons.

Conclusions:

• The MLatom framework facilitates accelerated development of machine learning models for NAMD.
• The framework provides deeper insights into nonadiabatic dynamics.
• The open-source MLatom package offers a comprehensive tool for NAMD research.