Initial data generation from VASP

[Asif-Iqbal-Bhatti]

Hello,

Thank you for the code. I would like to use this technique for my case. Have you provided how the initial data should be organized? I don't see any scripts that generated the AIMD data from VASP, although MACE provides scripts to standardize the data before training.

Further, I see there are no scripts for validation of the model.

[recisic]

Hello @Asif-Iqbal-Bhatti, here is the data preprocessing code example for LGPS which takes VASP XDATCAR as inputs.

import ase
import ase.io
import numpy as np

def unwrap_trajectory(
    positions: np.ndarray,  # [n_frames, n_atoms, 3]
    lattice: np.ndarray,  # [3, 3]
) -> np.ndarray:
    """Unwraps a trajectory of atomic positions. This function assumes that the
    positions do not jump more than half the box length between frames."""
    frac_diff = np.diff(positions, axis=0) @ np.linalg.inv(lattice)
    frac_diff_unwrap = frac_diff - np.floor(frac_diff + 0.5)
    diff_unwrap = frac_diff_unwrap @ lattice
    return np.cumsum(np.concatenate((positions[0, None], diff_unwrap), axis=0), axis=0)

positions_data = dict()

for temp in [650, 900, 1150, 1400]:
    # Load the wrapped trajectory
    traj = ase.io.read(
        f"data/LGPS/Li10GeS2P12_AIMD_stoichiometric_{temp}K_XDATCAR.sid",
        index=":",
        format="vasp-xdatcar",
    )
    # Extract and unwrap the trajectory
    positions = np.array([atoms.positions for atoms in traj])
    lattice = traj[0].cell.array
    atomic_numbers = traj[0].get_atomic_numbers()
    positions = unwrap_trajectory(positions, lattice)
    # Store the unwrapped trajectory
    positions_data[temp] = positions

np.save("data/LGPS/atomic_numbers.npy", {"LGPS": atomic_numbers})
np.save("data/LGPS/lattice.npy", {"LGPS": lattice})
for temp in positions_data:
    np.savez_compressed(
        f"data/LGPS/positions_{temp}K.npz", **{"LGPS": positions_data[temp]}
    )

The model validation loop is integrated into the module source code. You can find this in https://github.com/learningmatter-mit/liflow/blob/main/liflow/model/modules.py#L84