Machine learning phase space quantum dynamics approaches

Area of Science:

• Quantum dynamics
• Computational chemistry
• Statistical mechanics

Background:

• Equilibrium continuity dynamics (ECD) offers trajectory-based methods for quantum systems.
• ECD methods conserve quantum Boltzmann distributions and are exact in classical/harmonic limits.
• Key ECD elements include effective force and effective mass matrix.

Purpose of the Study:

• To introduce a machine learning (ML) approach for fitting ECD elements.
• To improve the efficiency of integrating ECD equations of motion.
• To assess the accuracy and feasibility of ML-driven quantum phase space dynamics.

Main Methods:

• Developed an ML model to fit effective force and mass matrix in quantum phase space.
• Utilized the zeroth term of the phase space propagator expansion.
• Integrated ML-fitted elements into ECD equations of motion.

Main Results:

• Demonstrated the feasibility of ML-driven quantum phase space dynamics.
• Achieved reasonably accurate results for realistic molecular systems.
• Showcased a significant increase in computational efficiency.

Conclusions:

• ML approaches are competent for quantum phase space dynamics.
• This method offers a computationally efficient pathway for accurate molecular dynamics simulations.
• The ML-ECD framework holds promise for advancing quantum dynamics studies.