Explicit Multielement Extension of the Spectral Neighbor Analysis Potential for Chemically Complex Systems.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• The Spectral Neighbor Analysis Potential (SNAP) method uses basis function expansion of weighted neighbor density for atomic interactions.
• Existing SNAP formulations struggle with chemically complex systems due to limitations in handling multiple elements.

Purpose of the Study:

• To develop an extension of the SNAP method for accurate modeling of atomic interactions in chemically complex systems.
• To introduce explicit multielement descriptors for improved accuracy in interatomic potentials.

Main Methods:

• Derived a natural extension of SNAP descriptors by partitioning neighbor density into partial densities for each chemical element.
• Developed an explicit multielement SNAP (EM-SNAP) formulation.
• Implemented the EM-SNAP method in LAMMPS for force calculations.
• Generated an interatomic potential for indium phosphide.

Main Results:

• The new formulation leads to explicit multielement descriptors.
• Computational cost scales favorably with a linear relationship to the number of atoms.
• The EM-SNAP method accurately captures high-energy defects in indium phosphide resulting from radiation damage.
• EM-SNAP reproduces relaxed defect formation energies with substantially greater accuracy than weighted-density SNAP.

Conclusions:

• The explicit multielement SNAP method provides a significant advancement for modeling complex chemical systems.
• This method offers improved accuracy for defect energies in materials like indium phosphide.
• The approach retains favorable computational scaling, making it suitable for large-scale simulations.