Evaluating MS-ANI vs SS-ANI for Surface Hopping Simulations: A Cyclohexadiene Photochemical Ring-Opening Case Study

Area of Science:

• Computational Chemistry
• Photochemistry
• Machine Learning

Background:

• Traditional quantum mechanical (QM) methods are computationally expensive for simulating excited-state dynamics.
• Machine learning potentials (MLPs) offer a promising alternative for cost-effective simulations.
• Accurate modeling of excited-state dynamics is crucial for understanding photochemical reactions.

Purpose of the Study:

• To compare the performance of two MLPs, SS-ANI and MS-ANI, for simulating excited-state dynamics.
• To evaluate the accuracy of MLPs in modeling the photochemical ring-opening of 1,3-cyclohexadiene (CHD).
• To identify factors influencing MLP accuracy, such as data set quality and diversity.

Main Methods:

• Simulated photochemical ring-opening dynamics of 1,3-cyclohexadiene (CHD) using a Landau-Zener-based surface hopping algorithm.
• Trained two MLPs, SS-ANI and MS-ANI, using data generated from QM-based surface hopping simulations (SA-3-CASSCF/cc-pVDZ).
• Compared the energy predictions, energy gaps, and population dynamics obtained from SS-ANI and MS-ANI.

Main Results:

• MS-ANI demonstrated superior performance compared to SS-ANI in modeling multiple electronic states.
• MS-ANI provided more accurate energy predictions, stable energy gaps, and reliable population dynamics.
• The quality and diversity of training data significantly impacted MLP performance, with MS-ANI benefiting most from larger, well-curated datasets.

Conclusions:

• MS-ANI is a robust and scalable approach for accurate excited-state simulations in complex photochemical systems.
• MS-ANI effectively captures interstate correlations and maintains consistent energy profiles essential for surface hopping dynamics.
• The findings highlight the potential of MS-ANI for advancing computational studies in photochemistry.