Automated Selection of Nuclear Coordinates for Reduced Dimensionality Nonadiabatic Dynamics

Area of Science:

• Computational Chemistry
• Photochemistry
• Molecular Dynamics

Background:

• Simulating photochemical processes is computationally expensive due to high dimensionality.
• Current methods struggle to scale efficiently with the number of atoms.
• Accurate simulations are crucial for understanding molecular behavior during light-induced reactions.

Purpose of the Study:

• To investigate dimensionality reduction techniques for enhancing computational efficiency in dynamics simulations.
• To compare the effectiveness of Principal Component Analysis (PCA) and Normal Mode Variance (NMV) in reducing simulation dimensionality.
• To maintain accuracy in simulations of photoreactive molecules using reduced dimensionality.

Main Methods:

• Mixed quantum-classical Trajectory Surface Hopping (TSH) simulations were employed.
• Two dimensionality reduction techniques, PCA and NMV, were applied to three photoreactive molecules: trans-azomethane (tAZM), butyrolactone (Bulac), and furanone (Fur).
• Simulations were performed in both full and reduced dimensionality to evaluate accuracy.

Main Results:

• Both PCA and NMV successfully identified lower-dimensional spaces that reproduced full-dimensionality dynamics.
• PCA demonstrated superior performance over NMV, enabling greater dimensionality reduction without sacrificing accuracy.
• PCA's accuracy advantage was evident in electronic properties for tAZM and in the ring-opening reaction for Bulac and Fur.

Conclusions:

• Automated dimensionality reduction, particularly using PCA, offers a viable path to simulating larger photochemical systems.
• This approach enhances computational efficiency and avoids human bias in selecting simulation parameters.
• The findings facilitate more accurate and scalable simulations of complex photochemical events.