Skip to content

api_other_calc.md

Latest commit

 

History

History
64 lines (50 loc) · 2.63 KB

api_other_calc.md

File metadata and controls

64 lines (50 loc) · 2.63 KB
Raw

How to Use Polynomial MLPs with Other Calculator Tools

LAMMPS

To use polynomial MLPs in LAMMPS, please use the polymlp-lammps-package. Molecular dynamics and other calculations available in LAMMPS, including those using user packages, can be performed by following the standard LAMMPS documentation, except for the specification of pair_style and pair_coeff.

The following LAMMPS input commands specify a machine learning potential (e.g., for a binary Ti–Al system):

pair_style  polymlp
pair_coeff * * polymlp.lammps Ti Al

If you are using a polynomial MLP generated by pypolymlp (version ≥ 0.9.0), the potential file can be specified as follows:

pair_style  polymlp
pair_coeff * * polymlp.yaml Ti Al

Hybrid polynomial MLPs in LAMMPS

In hybrid polynomial MLPs, the potential energy is represented as a sum of multiple polynomial MLPs, where the regression coefficients of all models are determined simultaneously. When performing simulations with hybrid polynomial MLPs in LAMMPS, pair_style hybrid command can be used:

pair_style hybrid polymlp polymlp polymlp
pair_coeff * * polymlp.yaml.1 Ti Al
pair_coeff * * polymlp.yaml.2 Ti Al
pair_coeff * * polymlp.yaml.3 Ti Al

ASE

To use polynomial MLPs in ASE, generate a calculator instance as shown below. This calculator can be used for various calculations provided by ASE, in accordance with the ASE documentation.

from pypolymlp.calculator.utils.ase_calculator import PolymlpASECalculator

calculator = PolymlpASECalculator(pot="polymlp.yaml")

OpenKIM

A model driver for polynomial MLPs PolyMLP model driver is available in OpenKIM. Many polynomial MLPs from the Polynomial MLP repository are also available in OpenKIM.

To convert developed MLPs into portable models for OpenKIM, see OpenKIM Portable Model for Polynomial MLP.

Phonopy

from phonopy import Phonopy
from pypolymlp.utils.phonopy_utils import phonopy_cell_to_structure
from pypolymlp.api.pypolymlp_calc import PypolymlpCalc

supercell_matrix = np.diag([2, 2, 2])
ph = Phonopy(unitcell, supercell_matrix) # unitcell: PhonopyAtoms
ph.generate_displacements(distance=0.01)
structures = [phonopy_cell_to_structure(cell) for cell in ph.supercells_with_displacements]

polymlp = PypolymlpCalc(pot="polymlp.yaml")
_, forces, _ = polymlp.eval(structures) # Forces. shape=(n_disp, 3, natom), unit: eV/angstrom.
forces = np.array(forces).transpose((0, 2, 1))

ph.forces = forces
ph.produce_force_constants()