data/md_simulation/Utils.py

import matplotlib.pyplot as plt
import numpy as np
import torch
from ase import Atoms
from ase.calculators.calculator import Calculator, all_changes
from ase.geometry.analysis import Analysis
from ase.optimize import FIRE
from matsciml.datasets.trajectory_lmdb import data_list_collater
from matsciml.datasets.transforms import (
    FrameAveraging,
    PeriodicPropertiesTransform,
    PointCloudToGraphTransform,
)
from matsciml.preprocessing.atoms_to_graphs import AtomsToGraphs

a2g = AtomsToGraphs(
    max_neigh=200,
    radius=6,
    r_energy=False,
    r_forces=False,
    r_distances=False,
    r_edges=True,
    r_fixed=True,
)
f_avg = FrameAveraging(frame_averaging="3D", fa_method="stochastic")
PBCTransform = PeriodicPropertiesTransform(cutoff_radius=6.0, adaptive_cutoff=True)
GTransform = PointCloudToGraphTransform(
    "dgl",
    node_keys=["pos", "atomic_numbers"],
)


def convAtomstoBatchmace(atoms):
    data_obj = a2g.convert(atoms)
    Reformatted_batch = {
        "cell": data_obj.cell,
        "natoms": torch.Tensor([data_obj.natoms]).unsqueeze(0),
        "edge_index": [data_obj.edge_index.shape],
        "cell_offsets": data_obj.cell_offsets,
        "atomic_numbers": data_obj.atomic_numbers,
        "pos": data_obj.pos,
        "y": None,
        "force": None,
        "fixed": [data_obj.fixed],
        "tags": None,
        "sid": None,
        "fid": None,
        "dataset": "S2EFDataset",
        "graph": data_list_collater([data_obj]),
    }

    Reformatted_batch = PBCTransform(Reformatted_batch)

    return Reformatted_batch


def convAtomstoBatchtensornet(atoms):
    data_obj = a2g.convert(atoms)
    Reformatted_batch = {
        "cell": data_obj.cell,
        "natoms": torch.Tensor([data_obj.natoms]).unsqueeze(0),
        "edge_index": [data_obj.edge_index.shape],
        "cell_offsets": data_obj.cell_offsets,
        "atomic_numbers": data_obj.atomic_numbers,
        "pos": data_obj.pos,
        "y": None,
        "force": None,
        "fixed": [data_obj.fixed],
        "tags": None,
        "sid": None,
        "fid": None,
        "dataset": "S2EFDataset",
        # 'graph' : data_list_collater([data_obj]),
    }
    Reformatted_batch = PBCTransform(Reformatted_batch)
    Reformatted_batch = GTransform(Reformatted_batch)
    return Reformatted_batch


def convAtomstoBatchfaenet(atoms):
    data_obj = a2g.convert(atoms)
    Reformatted_batch = {
        "cell": data_obj.cell,
        "natoms": torch.Tensor([data_obj.natoms]).unsqueeze(0),
        "edge_index": [data_obj.edge_index.shape],
        "cell_offsets": data_obj.cell_offsets,
        "atomic_numbers": data_obj.atomic_numbers,
        "pos": data_obj.pos,
        "y": None,
        "force": None,
        "fixed": [data_obj.fixed],
        "tags": None,
        "sid": None,
        "fid": None,
        "dataset": "S2EFDataset",
        "graph": data_list_collater([data_obj]),
    }
    Reformatted_batch = f_avg(Reformatted_batch)
    Reformatted_batch = PBCTransform(Reformatted_batch)

    return Reformatted_batch


def convBatchtoAtoms(batch):
    # data_obj=a2g.convert(atoms)
    curr_atoms = Atoms(
        positions=batch["graph"].pos,
        cell=batch["cell"][0],
        numbers=batch["graph"].atomic_numbers,
        pbc=True,
    )  # True or false

    return curr_atoms


def minimize_structure(atoms, fmax=0.05, steps=50):
    """
    Perform energy minimization on the given ASE Atoms object using the FIRE optimizer.

    Parameters:
    atoms (ase.Atoms): The Atoms object to be minimized.
    fmax (float): The maximum force tolerance for the optimization (default: 0.01 eV/Å).
    steps (int): The maximum number of optimization steps (default: 1000).

    Returns:
    ase.Atoms: The minimized Atoms object.
    """
    dyn = FIRE(atoms, trajectory=None)
    dyn.run(fmax=fmax, steps=steps)
    return atoms


def to(data, device):
    """Simple utility function to move things to correct device"""
    new_dict = {}
    for key, value in data.items():
        if hasattr(value, "to"):
            new_dict[key] = value.to(device)
        else:
            new_dict[key] = value
    return new_dict


def min_height(cell_matrix):
    """
    Calculate the perpendicular heights in three directions given a 3x3 cell matrix.
    """
    a, b, c = cell_matrix[:, 0], cell_matrix[:, 1], cell_matrix[:, 2]
    volume = abs(np.dot(a, np.cross(b, c)))
    # Calculate the cross products
    a_cross_b, b_cross_c, c_cross_a = (
        np.linalg.norm(np.cross(a, b)),
        np.linalg.norm(np.cross(b, c)),
        np.linalg.norm(np.cross(c, a)),
    )
    # Calculate the perpendicular heights
    height_a, height_b, height_c = (
        abs(volume / a_cross_b),
        abs(volume / b_cross_c),
        abs(volume / c_cross_a),
    )
    return min(height_a, height_b, height_c)


def perturb_config(atoms, displacement_std=0.01):
    # Create a new Atoms object with the perturbed positions
    positions = atoms.get_positions()
    displacements = np.random.normal(scale=displacement_std, size=positions.shape)
    new_positions = positions + displacements
    new_perturbed_atoms = atoms.copy()
    new_perturbed_atoms.set_positions(new_positions)
    return new_perturbed_atoms


def plot_pair_rdfs(Pair_rdfs, shift=0):
    counter = 0
    plt.figure()
    for key in Pair_rdfs.keys():
        plt.plot(Pair_rdfs[key][0], Pair_rdfs[key][1] + shift * counter, label=key)
        counter += 1
    plt.legend(loc=(1.2, 0))
    plt.xlabel("r (Angstrom)")
    plt.ylabel("g(r)")
    plt.show()


def replicate_system(atoms, replicate_factors):
    """
    Replicates the given ASE Atoms object according to the specified replication factors.
    """
    nx, ny, nz = replicate_factors
    original_cell = atoms.get_cell()
    original_positions = atoms.get_positions() #@ original_cell  # Scaled or Unscaled ?
    original_numbers = atoms.get_atomic_numbers()
    x_cell, y_cell, z_cell = original_cell[0], original_cell[1], original_cell[2]
    new_numbers = []
    for i in range(nx):
        for j in range(ny):
            for k in range(nz):
                new_numbers += [original_numbers]
    pos_after_x = np.concatenate([original_positions + i * x_cell for i in range(nx)])
    pos_after_y = np.concatenate([pos_after_x + i * y_cell for i in range(ny)])
    pos_after_z = np.concatenate([pos_after_y + i * z_cell for i in range(nz)])
    new_cell = [nx * original_cell[0], ny * original_cell[1], nz * original_cell[2]]
    new_atoms = Atoms(
        numbers=np.concatenate(new_numbers),
        positions=pos_after_z,
        cell=new_cell,
        pbc=atoms.get_pbc(),
    )
    return new_atoms


def write_xyz(Filepath, atoms):
    """Writes ovito xyz file"""
    R = atoms.get_position()
    species = atoms.get_atomic_numbers()
    cell = atoms.get_cell()
    f = open(Filepath, "w")
    f.write(str(R.shape[0]) + "\n")
    flat_cell = cell.flatten()
    f.write(
        f'Lattice="{flat_cell[0]} {flat_cell[1]} {flat_cell[2]} {flat_cell[3]} {flat_cell[4]} {flat_cell[5]} {flat_cell[6]} {flat_cell[7]} {flat_cell[8]}" Properties=species:S:1:pos:R:3 Time=0.0'
    )
    for i in range(R.shape[0]):
        f.write(
            "\n"
            + str(species[i])
            + "\t"
            + str(R[i, 0])
            + "\t"
            + str(R[i, 1])
            + "\t"
            + str(R[i, 2])
        )


def symmetricize_replicate(curr_atoms, max_atoms, box_lengths):
    replication = [1, 1, 1]
    atom_count = curr_atoms
    lengths = box_lengths
    while atom_count < (max_atoms // 2):
        direction = np.argmin(box_lengths)
        replication[direction] += 1
        lengths[direction] = box_lengths[direction] * replication[direction]
        atom_count = curr_atoms * replication[0] * replication[1] * replication[2]
    return replication, atom_count


def get_pairs(atoms):
    Atom_types = np.unique(atoms.get_chemical_symbols())
    Pairs = []
    for i in range(len(Atom_types)):
        for j in range(i, len(Atom_types)):
            Pairs += [[Atom_types[i], Atom_types[j]]]
    return Pairs


def getfirstpeaklength(r, rdf, r_max=6.0):
    bin_size = (r[-1] - r[0]) / len(r)
    cut_index = int(r_max / bin_size)
    cut_index = min(cut_index, len(r))
    Peak_index = np.argmax(rdf[:cut_index])
    # Returns : Peak index and Bond length
    return Peak_index, r[Peak_index]


def get_partial_rdfs(Traj, r_max=6.0, dr=0.01):
    rmax = min(r_max, min_height(Traj[0].get_cell()) / 2.7)
    analysis = Analysis(Traj)
    dr = dr
    nbins = int(rmax / dr)
    pairs_list = get_pairs(Traj[0])
    Pair_rdfs = dict()
    for pair in pairs_list:
        rdf = analysis.get_rdf(
            rmax=rmax, nbins=nbins, imageIdx=None, elements=pair, return_dists=True
        )
        x = rdf[0][1]
        y = np.array([rdf[k][0] for k in range(len(rdf))]).mean(axis=0)
        Pair_rdfs["-".join(pair)] = [x, y]
    return Pair_rdfs


def get_partial_rdfs_smoothened(
    inp_atoms, perturb=10, noise_std=0.01, max_atoms=300, r_max=6.0, dr=0.01
):
    atoms = inp_atoms.copy()
    replication_factors, _ = symmetricize_replicate(
        len(atoms),
        max_atoms=max_atoms,
        box_lengths=atoms.get_cell_lengths_and_angles()[:3],
    )
    atoms = replicate_system(atoms, replication_factors)
    Traj = [perturb_config(atoms, noise_std) for k in range(perturb)]
    return get_partial_rdfs(Traj, r_max=r_max, dr=dr)


def get_bond_lengths_noise(
    inp_atoms, perturb=10, noise_std=0.01, max_atoms=300, r_max=6.0, dr=0.01
):
    Pair_rdfs = get_partial_rdfs_smoothened(
        inp_atoms,
        perturb=perturb,
        noise_std=noise_std,
        max_atoms=max_atoms,
        r_max=r_max,
        dr=dr,
    )
    Bond_lengths = dict()
    for key in Pair_rdfs:
        r, rdf = Pair_rdfs[key]
        Bond_lengths[key] = getfirstpeaklength(r, rdf)[1]
    return Bond_lengths, Pair_rdfs


def get_bond_lengths_TrajAvg(Traj, r_max=6.0, dr=0.01):
    Pair_rdfs = get_partial_rdfs(Traj, r_max=r_max, dr=dr)
    Bond_lengths = dict()
    for key in Pair_rdfs:
        r, rdf = Pair_rdfs[key]
        Bond_lengths[key] = getfirstpeaklength(r, rdf)[1]
    return Bond_lengths, Pair_rdfs


def get_initial_rdf(
    inp_atoms,
    perturb=10,
    noise_std=0.01,
    max_atoms=300,
    replicate=False,
    Structid=0,
    r_max=6.0,
    dr=0.01,
):
    atoms = inp_atoms.copy()
    # write_xyz(f"StabilityXYZData2/{Structid}.xyz",atoms.get_positions(),atoms.get_chemical_symbols(),atoms.get_cell())
    if replicate:
        replication_factors, size = symmetricize_replicate(
            len(atoms),
            max_atoms=max_atoms,
            box_lengths=atoms.get_cell_lengths_and_angles()[:3],
        )
        atoms = replicate_system(atoms, replication_factors)
    rmax = min(r_max, min_height(atoms.get_cell()) / 2.7)
    # atoms.rattle(0.01)
    analysis = Analysis([perturb_config(atoms, noise_std) for k in range(perturb)])
    # write_xyz(f"StabilityXYZDataReplicated2/{Structid}.xyz",atoms.get_positions(),atoms.get_chemical_symbols(),atoms.get_cell())
    dr = dr
    nbins = int(rmax / dr)
    rdf = analysis.get_rdf(
        rmax=rmax, nbins=nbins, imageIdx=None, elements=None, return_dists=True
    )
    x = rdf[0][1]
    y = np.array([rdf[k][0] for k in range(len(rdf))]).mean(axis=0)
    return x, y


def get_rdf(Traj, r_max=6.0, dr=0.01):
    rmax = min(r_max, min_height(Traj[0].get_cell()) / 2.7)
    analysis = Analysis(Traj)
    dr = dr
    nbins = int(rmax / dr)
    rdf = analysis.get_rdf(
        rmax=rmax, nbins=nbins, imageIdx=None, elements=None, return_dists=True
    )
    x = rdf[0][1]
    y = np.array([rdf[k][0] for k in range(len(rdf))]).mean(axis=0)
    return x, y


## MAce Calculator
class ASEcalculator(Calculator):
    """Simulation ASE Calculator"""

    implemented_properties = ["energy", "forces", "stress"]

    def __init__(self, model, model_name, **kwargs):
        Calculator.__init__(self, **kwargs)
        self.results = {}

        self.model = model
        if model_name == "mace":
            self.convAtomstoBatch = convAtomstoBatchmace

        elif model_name == "faenet":
            self.convAtomstoBatch = convAtomstoBatchfaenet

        elif model_name == "tensornet":
            self.convAtomstoBatch = convAtomstoBatchtensornet

        else:
            print("Wrong Model Name")

    # pylint: disable=dangerous-default-value
    def calculate(self, atoms=None, properties=None, system_changes=all_changes):
        """
        Calculate properties.
        :param atoms: ase.Atoms object
        :param properties: [str], properties to be computed, used by ASE internally
        :param system_changes: [str], system changes since last calculation, used by ASE internally
        :return:
        """
        # call to base-class to set atoms attribute
        Calculator.calculate(self, atoms)

        # prepare data
        batch = self.convAtomstoBatch(atoms)

        # predict + extract data
        out = self.model.predict(to(batch,self.model.device))
        #out = self.model.forward(batch)
        energy = out['energy'].detach().cpu().item()
        forces = out["force"].detach().cpu().numpy()
        stress = out["stress"].squeeze(0).detach().cpu().numpy()
        # store results
        E = energy
        stress = np.array(
            [
                stress[0, 0],
                stress[1, 1],
                stress[2, 2],
                stress[1, 2],
                stress[0, 2],
                stress[0, 1],
            ]
        )
        self.results = {
            "energy": E,
            # force has units eng / len:
            "forces": forces,
            "stress": stress,
        }