Dynamically Polarizable Water Potential Based on Multipole Moments Trained by Machine Learning

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Physical chemistry

Background:

• Accurate simulation of water properties requires accounting for polarization.
• Existing methods often involve complex iterative calculations or damping terms.

Purpose of the Study:

• To develop a novel water potential for molecular dynamics simulations that explicitly includes polarization.
• To accurately reproduce structural, dynamic, and thermodynamic properties of water clusters.

Main Methods:

• Training artificial neural networks (NNs) to predict atomic multipole moments.
• Utilizing quantum chemical topology (QCT) for atomic partitioning.
• Input for NNs comprises coordinates of surrounding water molecules.

Main Results:

• A novel potential for water dimer, trimer, tetramer, pentamer, and hexamer is proposed.
• The method provides a dynamic multipolar representation of water electron density.
• Eliminates the need for iterative self-consistency calculations and damping terms.

Conclusions:

• The proposed NN-based potential offers an efficient and accurate way to include polarization in water simulations.
• This approach simplifies simulations by avoiding polarization catastrophes and iterative procedures.
• The method is suitable for molecular dynamics simulations requiring precise water property reproduction.