Direct Grid-Based Nonadiabatic Dynamics on Machine-Learned Potential Energy Surfaces: Application to Spin-Forbidden...

Area of Science:

• Quantum chemistry
• Computational chemistry
• Chemical dynamics

Background:

• Accurate simulation of molecular dynamics is crucial for understanding chemical reactions.
• Existing methods often require pre-fitting of potential energy surfaces (PESs), which can be computationally expensive and time-consuming.
• Machine learning (ML) has shown promise in accelerating computational chemistry calculations.

Purpose of the Study:

• To adapt and demonstrate a machine learning-based quantum dynamics scheme for modeling spin-forbidden nonadiabatic processes, specifically intersystem crossing (ISC).
• To enable direct dynamics calculations without the need for pre-fitted potential energy surfaces.
• To provide a powerful computational tool for studying important photochemical systems.

Main Methods:

• Integration of high-accuracy wave function grid-based propagation schemes (e.g., multiconfiguration time-dependent Hartree - MCTDH) with machine learning (ML) descriptions of potential energy surfaces (PESs).
• Development of modified diabatization schemes to accurately treat electronic states with different spin multiplicities.
• Application of the ML-based quantum dynamics scheme to model ISC in SO2 and thioformaldehyde.

Main Results:

• Successful adaptation of the ML-based quantum dynamics scheme to model spin-forbidden nonadiabatic dynamics, including intersystem crossing (ISC).
• Demonstration of accurate and robust treatment of electronic states with different spin multiplicities through modified diabatization schemes.
• Benchmarking of results against previous computational studies for SO2 and thioformaldehyde ISC, showing good agreement.

Conclusions:

• The developed ML-based quantum dynamics scheme offers an efficient approach for modeling spin-forbidden nonadiabatic dynamics without PES pre-fitting.
• This methodology presents a powerful tool for investigating complex photochemical phenomena, including those relevant to photoactivated pro-drugs and organometallic catalysts.
• The study expands the applicability of ML in quantum dynamics to spin-dependent processes, opening new avenues in computational chemistry.