OpenMM

OpenMM has no callback that can call an arbitrary C++ potential. The comparison that the meson test openmm_lj runs (engines/openmm/run_openmm.py) is a CustomNonbondedForce whose expression is the shifted 12-6 potential rgpot evaluates, written in OpenMM’s nanometres and kJ/mol.

The energy is one expression with no named subexpressions. Sigma and epsilon are combined geometrically, which is exact when both atoms carry the same parameters, as they do in the test.

Positions are 0 and 0.15 nm in a 2 nm cube, cutoff 0.8 nm. The context energy, converted with rgpot-lj --kJ-per-eV, matches the direct evaluation to 1e-6 eV, and the force on the second atom matches to 1e-5 eV/Angstrom. The test runs when the openmm Python module imports. -Dwith_openmm=enabled makes a missing openmm module a configure error.

Public constructor

rgpot_openmm_potential stores the same nanometre and kJ/mol factors as rgpot_gromacs_potential; rgpot_md_kj_per_ev() and rgpot_md_nm_per_angstrom() return them. It takes u0 in eV and cutoff and psi in Angstrom. An empty engine path (NULL or "" in C) evaluates the potential in process, in Angstrom and eV. A non-empty path is a library that exports rgpot_engine_create, rgpot_engine_destroy and rgpot_md_native_force in nanometres and kJ/mol, such as libopenmm_engine.so; a library that lacks one of them is refused. That library evaluates the shifted 12-6 expression with an OpenMM CustomNonbondedForce on the CPU platform. The constructor converts u0, cutoff and psi to the library’s units once and passes them to rgpot_engine_create. Every call then converts the positions and the box to nanometres and the energy and forces back to eV and eV/Angstrom. rgpot_external_potential_energy_forces returns 0 and that result, or 1 with a message in err, and rgpot_external_potential_free releases the potential. The errors are the same as for rgpot_gromacs_potential: the C constructor returns NULL and writes err.

Meson builds libopenmm_engine.so when the openmm Python module reports its library path and libOpenMM and openmm/Platform.h are found there. With -Dwith_openmm=enabled, a missing libOpenMM there is a configure error. The meson test external_potentials then loads it through RGPOT_OPENMM_ENGINE and checks the dimer at 1.5 Angstrom: -0.3203213355 eV from rgpot-lj, matched to 1e-6 eV, and the force on the second atom to 1e-5 eV/Angstrom. When the library was not built, the test prints a SKIP openmm engine line with that reason and still passes.

import numpy as np
import rgpot

pot = rgpot.rgpot_openmm_potential(1.0, 8.0, 1.0)
energy, forces, variance = pot(
    np.array([[0.0, 0.0, 0.0], [1.5, 0.0, 0.0]]),
    [1, 1],
    np.diag([20.0, 20.0, 20.0]),
)
assert f"{energy:.10f}" == "-0.3203213355"
#include "rgpot/ExternalPot/external_potentials.h"

#include <stdio.h>

int main(void) {
  char err[256] = "";
  const double xyz[6] = {0, 0, 0, 1.5, 0, 0};
  const int z[2] = {1, 1};
  const double box[9] = {20, 0, 0, 0, 20, 0, 0, 0, 20};
  double forces[6];
  double energy = 0.0;
  rgpot_external_potential *pot =
      rgpot_openmm_potential(NULL, 1.0, 8.0, 1.0, err, sizeof err);
  if (pot == NULL) {
    fprintf(stderr, "%s\n", err);
    return 1;
  }
  if (rgpot_external_potential_energy_forces(pot, xyz, z, 2, box, forces,
                                             &energy, err, sizeof err) != 0) {
    fprintf(stderr, "%s\n", err);
    rgpot_external_potential_free(pot);
    return 1;
  }
  printf("%.10f eV\n", energy);
  rgpot_external_potential_free(pot);
  return 0;
}