Molecular dynamics using ASE#

This example is modified from the official home page and Constant temperature MD to use the ASE interface of TorchANI as energy calculator.

As always, we start by importing the modules we need

import ase
from ase.lattice.cubic import Diamond
from ase.md.langevin import Langevin
from ase.optimize import LBFGS

from torchani.models import ANI2x

First we set up our system (in this case a diamond crystal, with PBC enabled)

atoms = Diamond(symbol="C", pbc=True)
len(atoms)  # The number of atoms in the system
8

After, we create a calculator from an ANI model and attach it to our atoms

atoms.calc = ANI2x().ase()

Then we minimize our system using the L-BFGS optimizer, which is included in ASE, under ase.optimize.LBFGS.

opt = LBFGS(atoms)
opt.run(fmax=0.0002)
       Step     Time          Energy          fmax
LBFGS:    0 10:16:22    -8162.894043        0.000044

np.True_

We want to run constant temperature MD, and print some quantities throughout the MD. For this we need to create a callback (function) that prints the quantities we are interested in. For example, to print the energy of the system we can use:

def print_energy(atoms: ase.Atoms):
    pot_energy = atoms.get_potential_energy() / len(atoms)
    kin_energy = atoms.get_kinetic_energy() / len(atoms)
    temperature = kin_energy / (1.5 * ase.units.kB)
    tot_energy = pot_energy + kin_energy
    print(
        "Energy per atom: \n"
        f"    E_pot = {pot_energy:.3f} eV\n"
        f"    E_kin = {kin_energy:.3f} eV (T = {temperature:.1f} K)\n"
        f"    E_tot = {tot_energy:.3f} eV\n"
    )

We will use the Langevin thermostat to control the temperature. To do this we need to first construct an ase.md.langevin.Langevin object. Here we use with a time step of 1 fs, a temperature of 300 K and a friction coefficient of 0.2

dyn = Langevin(
    atoms,
    timestep=1 * ase.units.fs,
    temperature_K=300,
    friction=0.2,
)
dyn.attach(print_energy, interval=5, atoms=atoms)
/opt/hostedtoolcache/Python/3.11.18/x64/lib/python3.11/site-packages/ase/md/langevin.py:102: FutureWarning: The implementation of `fixcm=True` in `Langevin` does not strictly sample the correct NVT distributions. The deviations are typically small for large systems but can be more pronounced for small systems. Use `fixcm=False` together with `ase.constraints.FixCom`. `fixcm` is deprecated since ASE 3.28.0 and will be removed in a future release.
  warnings.warn(msg, FutureWarning)

Finally we run the dynamics using dyn.run(steps)

print_energy(atoms)  # Print the initial energy
dyn.run(50)
Energy per atom:
    E_pot = -1020.362 eV
    E_kin = 0.000 eV (T = 0.0 K)
    E_tot = -1020.362 eV

Energy per atom:
    E_pot = -1020.362 eV
    E_kin = 0.000 eV (T = 0.0 K)
    E_tot = -1020.362 eV

Energy per atom:
    E_pot = -1020.391 eV
    E_kin = 0.039 eV (T = 303.1 K)
    E_tot = -1020.351 eV

Energy per atom:
    E_pot = -1023.577 eV
    E_kin = 3.203 eV (T = 24776.6 K)
    E_tot = -1020.375 eV

Energy per atom:
    E_pot = -1028.395 eV
    E_kin = 6.954 eV (T = 53796.4 K)
    E_tot = -1021.442 eV

Energy per atom:
    E_pot = -1030.734 eV
    E_kin = 8.457 eV (T = 65424.6 K)
    E_tot = -1022.277 eV

Energy per atom:
    E_pot = -1029.290 eV
    E_kin = 5.653 eV (T = 43735.1 K)
    E_tot = -1023.637 eV

Energy per atom:
    E_pot = -1029.974 eV
    E_kin = 5.151 eV (T = 39851.8 K)
    E_tot = -1024.823 eV

Energy per atom:
    E_pot = -1029.821 eV
    E_kin = 4.286 eV (T = 33160.1 K)
    E_tot = -1025.535 eV

Energy per atom:
    E_pot = -1031.193 eV
    E_kin = 5.011 eV (T = 38763.3 K)
    E_tot = -1026.183 eV

Energy per atom:
    E_pot = -1032.903 eV
    E_kin = 5.640 eV (T = 43634.0 K)
    E_tot = -1027.263 eV

Energy per atom:
    E_pot = -1032.539 eV
    E_kin = 4.322 eV (T = 33434.8 K)
    E_tot = -1028.218 eV


True