Iterative charge equilibration for fourth-generation high-dimensional neural network potentials

Area of Science:

• Computational Chemistry
• Materials Science
• Physics

Background:

• Machine learning potentials (MLPs) enable accurate, large-scale molecular dynamics simulations.
• Fourth-generation MLPs are crucial for systems with long-range charge transfer, requiring global electrostatic interactions.
• Direct charge equilibration methods exhibit cubic scaling, limiting their application to large systems.

Purpose of the Study:

• To introduce an efficient iterative charge equilibration (iQEq) method for determining atomic partial charges.
• To implement iQEq within LAMMPS for fourth-generation high-dimensional neural network potentials (4G-HDNNPs).
• To assess the accuracy and efficiency of iQEq for large-scale simulations.

Main Methods:

• Iterative solution of the charge equilibration problem (iQEq).
• Implementation of iQEq in LAMMPS, compatible with the n2p2 library.
• Quadratic scaling analysis of the iQEq method with system size.

Main Results:

• The iQEq method demonstrates quadratic scaling with system size, significantly improving computational efficiency.
• Accurate atomic partial charges were determined for a benchmark system (FeCl3 in water).
• The iQEq method is general and applicable to various fourth-generation MLPs.

Conclusions:

• The iQEq method offers a computationally efficient alternative to direct charge equilibration for large systems.
• This advancement facilitates more accurate and scalable molecular dynamics simulations using advanced MLPs.
• The implemented iQEq method enhances the applicability of 4G-HDNNPs in complex chemical systems.