Centroid Molecular Dynamics Can Be Greatly Accelerated through Neural Network Learned Centroid Forces Derived from...

Area of Science:

• Computational chemistry
• Quantum dynamics
• Machine learning applications

Background:

• Centroid molecular dynamics (CMD) is a classical-like phase space method for quantum dynamics.
• Calculating the centroid effective force in CMD is computationally expensive, limiting its efficiency for condensed phase systems.

Purpose of the Study:

• To develop a faster and more cost-effective method for calculating quantum dynamical properties using CMD.
• Introduce machine-learned centroid molecular dynamics (ML-CMD) to address the computational bottleneck in standard CMD.

Main Methods:

• A neural network is trained to learn the centroid effective force from path integral molecular dynamics data.
• The learned force field is used to evolve centroids directly within the CMD algorithm.
• The DeepMD software kit facilitates the training process for ML-CMD.

Main Results:

• ML-CMD demonstrates significantly reduced computational cost compared to "on the fly" CMD and ring polymer molecular dynamics (RPMD).
• The method achieves accuracy comparable to CMD and RPMD in estimating quantum dynamical properties.
• Applied successfully to model systems like liquid para-hydrogen and water.

Conclusions:

• ML-CMD offers a computationally efficient and accurate approach for studying condensed phase quantum dynamics.
• This machine learning-based strategy overcomes the limitations of traditional CMD, enabling broader applications.