Large-Scale Calculations by Integrating the Fragmentation Approach With Neural Network Potentials

Area of Science:

• Computational chemistry
• Materials science
• Molecular dynamics

Background:

• Neural network potentials (NNPs) offer efficient molecular simulations.
• Scaling limitations of conventional NNPs hinder large-scale system modeling.

Purpose of the Study:

• To introduce a fragmentation method for large-scale molecular simulations using NNPs.
• To overcome the atomistic limitations of current NNP simulations.

Main Methods:

• System partitioning into cube-shaped fragments.
• Many-body expansion formalism for energy reconstruction.
• Distance-based cut-off approximation.

Main Results:

• Accurate energy reconstruction with three-body interactions and 26 neighbors.
• Per-atom energy error reduced to within 0.04 eV for various crystals.
• Enabled simulations of systems up to 1 million atoms.
• Scaling exponent below 1.64 for three-body calculations.

Conclusions:

• The fragmentation method significantly enhances the scale of NNP simulations.
• The approach is computationally feasible for extremely large systems.
• Opens possibilities for simulating complex materials and phenomena at unprecedented scales.