Direct Learning Hidden Excited State Interaction Patterns from ab initio Dynamics and Its Implication as Alternative...

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Quantum chemistry

Background:

• Excited states of polyatomic systems are complex and exhibit meta-stable behaviors.
• Static analysis of reaction pathways often fails to capture non-equilibrium excited state motions.

Purpose of the Study:

• To develop a method for characterizing meta-stable dynamical behaviors in excited states.
• To accurately predict ground and excited state properties from ab initio dynamic trajectories.

Main Methods:

• A time series guided clustering algorithm was proposed to identify meta-stable patterns from dynamic trajectories.
• An interpolation scheme, Prediction with Ensemble Models (PEM), was developed using these identified patterns.

Main Results:

• The PEM method accurately predicts ground and excited state properties with similar accuracy.
• The method requires no training data beyond the initial clustering algorithm.
• Demonstrated effectiveness using sinapic acids as an example.

Conclusions:

• The developed method provides insights into constructing molecular mechanism models.
• It enables accurate prediction of molecular properties for both ground and excited states.
• Offers a way to create compatible energy terms for molecular models.