potential

This keyword is used to specify the interatomic potential model for the system.

Available potential models

• Tersoff-1989 potential

• Tersoff-1988 potential

• Tersoff mini potential

• Embedded atom method (EAM) potential

• Force constant potential (FCP)

• Lennard-Jones (LJ) potential

• Neuroevolution potential (NEP)

• Hybrid NEP+ILP potential

• Hybrid SW+ILP potential

• Hybrid Tersoff+ILP potential

• Deep Potential (DP)

Syntax

The first parameter, potential_filename, is the filename (including relative or absolute path) of the potential file to be used. Some potential types require additional parameters, as described in their corresponding sections.

For standard NEP, temperature-dependent NEP, and nep_spin models, a multi-GPU run can optionally specify the spatial partition direction:

potential <potential_filename> [partition_direction]

Here, partition_direction can be x, y, or z. These correspond to the a, b, and c directions, respectively, for a triclinic box. If the direction is omitted, GPUMD selects it automatically, normally along the thickest direction of the box.

The optional partition direction applies only to the multi-GPU implementation of standard NEP, temperature-dependent NEP, and nep_spin models. Charge NEP models currently use the single-GPU implementation in gpumd.

Example

potential Si_NEP.txt

For a multi-GPU NEP run, one can force the decomposition along a specific direction, for example:

potential Si_NEP.txt x

NEP-spin

A NEP-spin model uses all GPUs visible to the process to compute the same atomic system. The optional partition direction also applies to NEP-spin. Small systems may not benefit from multiple GPUs; global atom arrays remain on GPU 0.

Provide spin:R:3 in model.xyz. Fully periodic orthogonal and triclinic cells are supported. Spins remain fixed under nve and nvt_nhc; other ensembles and multiple-potential runs are not supported. See the Spin runtime instructions for output, restart and external-field usage.