File size: 9,467 Bytes
3e02ab8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
"""Relax an ASE structure with a compiled NequIP model.

This follows the official NequIP ASE relaxation example: it supports atomic
and cell relaxation, tracks forces at every ionic step, and aborts exploding
relaxations before they can hang indefinitely.
"""

from __future__ import annotations

import argparse
import json
import os
from pathlib import Path
from typing import Any

import numpy as np
import torch
from ase import Atoms
from ase.build import bulk
from ase.filters import ExpCellFilter, FrechetCellFilter
from ase.io import read, write
import ase.optimize as opt

from onescience.utils.nequip.integrations.ase import NequIPCalculator


OPTIMIZERS = {
    "BFGS": opt.BFGS,
    "BFGSLineSearch": opt.BFGSLineSearch,
    "FIRE": opt.FIRE,
    "FIRE2": opt.FIRE2,
    "GOQN": opt.GoodOldQuasiNewton,
    "GPMin": opt.GPMin,
    "LBFGS": opt.LBFGS,
    "LBFGSLineSearch": opt.LBFGSLineSearch,
}
CELL_FILTERS = {
    "exp": ExpCellFilter,
    "frechet": FrechetCellFilter,
}


def default_compiled_model() -> str | None:
    models_dir = os.environ.get("ONESCIENCE_MODELS_DIR")
    if not models_dir:
        return None
    return str(Path(models_dir) / "NequIP" / "NequIP-OAM-L-0.1.nequip.pth")


def load_structure(
    input_path: str | None,
    index: int,
    element: str,
    crystal_structure: str,
    lattice_constant: float,
    displacement: float,
) -> tuple[Atoms, str]:
    if input_path:
        path = Path(input_path).expanduser().resolve()
        if not path.is_file():
            raise FileNotFoundError(f"input structure not found: {path}")
        return read(path, index=index), f"{path}[{index}]"

    atoms = bulk(
        element,
        crystalstructure=crystal_structure,
        a=lattice_constant,
        cubic=True,
    )
    if displacement:
        atoms.positions[0, 0] += displacement
    return atoms, (
        f"ASE bulk {element} {crystal_structure}, a={lattice_constant} Angstrom, "
        f"atom-0 displacement={displacement} Angstrom"
    )


def _max_vector_norm(values: np.ndarray) -> float:
    array = np.asarray(values)
    if array.size == 0:
        return 0.0
    return float(np.linalg.norm(array.reshape(-1, 3), axis=1).max())


def relaxation_snapshot(atoms: Atoms, target: Any, step: int) -> dict[str, Any]:
    forces = atoms.get_forces()
    optimizer_forces = target.get_forces()
    stress = atoms.get_stress()
    return {
        "step": step,
        "energy_ev": float(atoms.get_potential_energy()),
        "energy_ev_per_atom": float(atoms.get_potential_energy() / len(atoms)),
        "volume_angstrom3": float(atoms.get_volume()),
        "max_atomic_force_ev_per_angstrom": _max_vector_norm(forces),
        "max_optimizer_force": _max_vector_norm(optimizer_forces),
        "stress_ev_per_angstrom3_voigt": np.asarray(stress).tolist(),
        "max_abs_stress_ev_per_angstrom3": float(np.abs(stress).max()),
    }


def relax_structure(
    atoms: Atoms,
    *,
    optimizer_name: str,
    cell_filter_name: str,
    fixed_cell: bool,
    fmax: float,
    steps: int,
    force_limit: float,
    logfile: Path,
    trajectory: Path,
) -> tuple[bool, int, list[dict[str, Any]]]:
    if not fixed_cell and not atoms.pbc.all():
        raise ValueError("cell relaxation requires periodic boundaries; use --fixed-cell")

    target = atoms if fixed_cell else CELL_FILTERS[cell_filter_name](atoms)
    optimizer_cls = OPTIMIZERS[optimizer_name]
    history: list[dict[str, Any]] = []
    converged = False

    with optimizer_cls(
        target,
        logfile=str(logfile),
        trajectory=str(trajectory),
    ) as optimizer:
        for converged in optimizer.irun(fmax=fmax, steps=steps):
            snapshot = relaxation_snapshot(atoms, target, optimizer.nsteps)
            history.append(snapshot)
            if max(
                snapshot["max_atomic_force_ev_per_angstrom"],
                snapshot["max_optimizer_force"],
            ) > force_limit:
                raise RuntimeError(
                    f"relaxation force exceeded safety limit {force_limit:g}"
                )

    return bool(converged), int(optimizer.nsteps), history


def main() -> None:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--compiled-model", default=default_compiled_model())
    parser.add_argument(
        "--input",
        help="CIF, POSCAR, XYZ, trajectory, or another ASE-readable structure",
    )
    parser.add_argument("--index", type=int, default=0)
    parser.add_argument("--device", default="cuda")
    parser.add_argument("--optimizer", choices=sorted(OPTIMIZERS), default="GOQN")
    parser.add_argument(
        "--cell-filter", choices=sorted(CELL_FILTERS), default="frechet"
    )
    parser.add_argument(
        "--fixed-cell",
        action="store_true",
        help="relax atomic positions only; the default also relaxes the cell",
    )
    parser.add_argument("--fmax", type=float, default=0.05)
    parser.add_argument("--steps", type=int, default=500)
    parser.add_argument("--force-limit", type=float, default=1.0e6)
    parser.add_argument("--element", default="Si")
    parser.add_argument("--crystal-structure", default="diamond")
    parser.add_argument("--lattice-constant", type=float, default=5.65)
    parser.add_argument("--displacement", type=float, default=0.08)
    parser.add_argument("--output-dir", default="outputs/structure_relaxation")
    parser.add_argument("--output-structure", default="relaxed.cif")
    parser.add_argument("--result", default="result.json")
    parser.add_argument("--trajectory", default="relax.traj")
    parser.add_argument("--log", default="relax.log")
    args = parser.parse_args()

    if not args.compiled_model:
        parser.error("--compiled-model is required when ONESCIENCE_MODELS_DIR is unset")
    compiled_model = Path(args.compiled_model).expanduser().resolve()
    if not compiled_model.is_file():
        parser.error(f"compiled model not found: {compiled_model}")
    if args.fmax <= 0:
        parser.error("--fmax must be positive")
    if args.steps < 1:
        parser.error("--steps must be positive")
    if args.force_limit <= 0:
        parser.error("--force-limit must be positive")

    output_dir = Path(args.output_dir).expanduser().resolve()
    output_dir.mkdir(parents=True, exist_ok=True)
    output_structure = output_dir / args.output_structure
    result_path = output_dir / args.result
    trajectory_path = output_dir / args.trajectory
    log_path = output_dir / args.log

    try:
        atoms, input_source = load_structure(
            args.input,
            args.index,
            args.element,
            args.crystal_structure,
            args.lattice_constant,
            args.displacement,
        )
    except (FileNotFoundError, IndexError, ValueError) as error:
        parser.error(str(error))
    if len(atoms) == 0:
        parser.error("input structure has no atoms")

    species = sorted(set(atoms.get_chemical_symbols()))
    atoms.calc = NequIPCalculator.from_compiled_model(
        compile_path=str(compiled_model),
        chemical_species_to_atom_type_map={symbol: symbol for symbol in species},
        device=args.device,
    )

    try:
        converged, nsteps, history = relax_structure(
            atoms,
            optimizer_name=args.optimizer,
            cell_filter_name=args.cell_filter,
            fixed_cell=args.fixed_cell,
            fmax=args.fmax,
            steps=args.steps,
            force_limit=args.force_limit,
            logfile=log_path,
            trajectory=trajectory_path,
        )
    except ValueError as error:
        parser.error(str(error))

    write(output_structure, atoms)
    result = {
        "compiled_model": str(compiled_model),
        "device": args.device,
        "device_name": torch.cuda.get_device_name(0)
        if args.device.startswith("cuda") and torch.cuda.is_available()
        else "cpu",
        "input_source": input_source,
        "formula": atoms.get_chemical_formula(),
        "num_atoms": len(atoms),
        "chemical_species_to_atom_type_map": {
            symbol: symbol for symbol in species
        },
        "optimizer": args.optimizer,
        "cell_filter": None if args.fixed_cell else args.cell_filter,
        "fixed_cell": args.fixed_cell,
        "fmax_ev_per_angstrom": args.fmax,
        "max_steps": args.steps,
        "force_safety_limit": args.force_limit,
        "converged": converged,
        "steps": nsteps,
        "initial": history[0],
        "final": history[-1],
        "energy_change_ev": history[-1]["energy_ev"] - history[0]["energy_ev"],
        "volume_change_angstrom3": (
            history[-1]["volume_angstrom3"] - history[0]["volume_angstrom3"]
        ),
        "history": history,
        "relaxed_structure": str(output_structure),
        "trajectory": str(trajectory_path),
        "log": str(log_path),
    }
    result_path.write_text(json.dumps(result, indent=2) + "\n", encoding="utf-8")

    print("formula:", result["formula"])
    print("atoms:", result["num_atoms"])
    print("converged:", converged)
    print("steps:", nsteps)
    print("initial energy (eV):", result["initial"]["energy_ev"])
    print("final energy (eV):", result["final"]["energy_ev"])
    print(
        "final max force (eV/Angstrom):",
        result["final"]["max_atomic_force_ev_per_angstrom"],
    )
    print("result:", result_path)


if __name__ == "__main__":
    main()