Bayesian machine learning for quantum molecular dynamics.

Area of Science:

• Quantum chemistry
• Computational physics
• Machine learning

Background:

• Quantum molecular dynamics simulations are crucial for understanding chemical reactions.
• Traditional methods often face limitations in accuracy and computational cost.
• Bayesian statistics offers a framework to quantify uncertainties in scientific models.

Purpose of the Study:

• To explore the application of Bayesian machine learning for quantum molecular dynamics.
• To develop a machine learning simulator for the Schrödinger equation.
• To incorporate uncertainty quantification into quantum dynamics predictions.

Main Methods:

• Formulating quantum dynamics as a machine learning simulator of the Schrödinger equation.
• Integrating Bayesian statistics to estimate uncertainties in dynamical results.
• Developing non-parametric distributions of potential energy surfaces conditioned by experimental data.

Main Results:

• The developed simulator provides quantum predictions with associated error bars.
• Identified sensitivity of dynamical properties to various input parameters (potential energy surface, collision energy, etc.).
• Demonstrated potential for accurate interpolation and physical extrapolation of Schrödinger equation solutions.

Conclusions:

• Bayesian machine learning provides a powerful tool for quantum molecular dynamics, offering uncertainty quantification.
• This approach can guide the design of efficient quantum dynamics calculations.
• It enables exploration of physical properties beyond current computational and experimental reach, accelerating discovery.