Skip to content

Commit

Permalink
Update fixes
Browse files Browse the repository at this point in the history
  • Loading branch information
@Zhuoying
Zhuoying committed Mar 6, 2024
1 parent b8d0246 commit 743666e
Showing 1 changed file with 6 additions and 167 deletions.
173 changes: 6 additions & 167 deletions atomate/vasp/analysis/lattice_dynamics.py
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
from itertools import product
from typing import Dict, List, Tuple
from joblib import Parallel, delayed
from typing import Dict, List, Tuple, Optional

import numpy as np
import scipy as sp
Expand All @@ -8,20 +9,11 @@
from copy import copy

from atomate.utils.utils import get_logger
<<<<<<< HEAD
<<<<<<< HEAD
from pymatgen import Structure
=======
=======

from hiphive import (ForceConstants, ForceConstantPotential,
enforce_rotational_sum_rules, ClusterSpace,
StructureContainer)
<<<<<<< HEAD
>>>>>>> 1e75a82a (imports)
=======
from hiphive.force_constant_model import ForceConstantModel
>>>>>>> 0d613579 (revert to structure container for fitting)
from hiphive.cutoffs import is_cutoff_allowed, estimate_maximum_cutoff
from hiphive.fitting import Optimizer
from hiphive.renormalization import Renormalization
Expand All @@ -30,7 +22,6 @@

from pymatgen.core.structure import Structure
from pymatgen.io.ase import AseAtomsAdaptor
>>>>>>> 8f4f7e37 (gruneisen and CTE)
from pymatgen.io.phonopy import get_phonopy_structure

from ase.atoms import Atoms
Expand All @@ -40,8 +31,8 @@
from phono3py.phonon3.gruneisen import Gruneisen


__author__ = "Alex Ganose, Rees Chang, Junsoo Park"
__email__ = "[email protected], [email protected], [email protected]"
__author__ = "Alex Ganose, Rees Chang, Junsoo Park, Zhuoying Zhu"
__email__ = "[email protected], [email protected], [email protected], [email protected]"

logger = get_logger(__name__)

Expand Down Expand Up @@ -69,11 +60,7 @@

def get_cutoffs(structure: Structure):
"""
<<<<<<< HEAD
Get a list of trial cutoffs based on a structure.
=======
Get a list of trial cutoffs based on a supercell structure for grid search.
>>>>>>> b637d9a8 (update)
An initial guess for the lower bound of the cutoffs is made based on the
average period (row) of the elements in the structure, according to:
Expand All @@ -99,15 +86,9 @@ def get_cutoffs(structure: Structure):
====== ==== =====
Cutoff Max Step
====== ==== =====
<<<<<<< HEAD
2ND +3.0 0.5
3RD +1.5 0.25
4TH +1.0 0.25
=======
2ND +2.0 1.0
3RD +1.5 0.75
4TH +0.6 0.6
>>>>>>> b637d9a8 (update)
====== ==== =====
Args:
Expand All @@ -123,28 +104,11 @@ def get_cutoffs(structure: Structure):
3: {1: 3.0, 2: 3.5, 3: 4.5, 4: 5.5, 5: 6.0, 6: 6.5, 7: 7.0},
4: {1: 2.5, 2: 3.0, 3: 3.5, 4: 4.0, 5: 4.5, 6: 5.0, 7: 5.5},
}
<<<<<<< HEAD
<<<<<<< HEAD
<<<<<<< HEAD
inc = {2: 3, 3: 1.5, 4: 1}
steps = {2: 0.5, 3: 0.25, 4: 0.25}

row = min([s.row for s in structure.species])
=======
inc = {2: 2, 3: 1, 4: 0.5}
steps = {2: 1, 3: 0.5, 4: 0.5}
=======
inc = {2: 2, 3: 1.5, 4: 0.5}
steps = {2: 1, 3: 0.75, 4: 0.5}
>>>>>>> c69d0316 (fixed cutoff)
=======
inc = {2: 2, 3: 1.5, 4: 0.6}
steps = {2: 1, 3: 0.75, 4: 0.6}
>>>>>>> b637d9a8 (update)

row = int(np.around(np.array([s.row for s in supercell_structure.species]).mean(),0))
factor = row/4
>>>>>>> cc7520c5 (separate fitting)
mins = {
2: min_cutoffs[2][row], 3: min_cutoffs[3][row], 4: min_cutoffs[4][row]
}
Expand All @@ -153,9 +117,6 @@ def get_cutoffs(structure: Structure):
range_three = np.arange(mins[3], mins[3] + factor*(inc[3]+steps[3]), factor*steps[3])
range_four = np.arange(mins[4], mins[4] + factor*(inc[4]+steps[4]), factor*steps[4])

<<<<<<< HEAD
return list(map(list, product(range_two, range_three, range_four)))
=======
cutoffs = np.array(list(map(list, product(range_two, range_three, range_four))))
max_cutoff = estimate_maximum_cutoff(AseAtomsAdaptor.get_atoms(supercell_structure))
cutoffs[cutoffs>max_cutoff] = max_cutoff-0.2
Expand All @@ -164,7 +125,6 @@ def get_cutoffs(structure: Structure):
good_cutoffs = np.all(cutoffs < max_cutoff-0.1, axis=1)
logger.info('GOOD CUTOFFS \n{}'.format(good_cutoffs))
return cutoffs[good_cutoffs].tolist()
>>>>>>> c094a175 (cutoff vs cell_size)


def fit_force_constants(
Expand All @@ -175,13 +135,9 @@ def fit_force_constants(
disp_cut: float = 0.055,
imaginary_tol: float = IMAGINARY_TOL,
fit_method: str = FIT_METHOD,
<<<<<<< HEAD
) -> Tuple["SortedForceConstants", Dict]:
=======
n_jobs: int = -1,
fit_kwargs: Optional[Dict] = None
) -> Tuple["SortedForceConstants", np.ndarray, ClusterSpace, Dict]:
>>>>>>> fddd9681 (get_cutoffs)
"""
Fit force constants using hiphive and select the optimum cutoff values.
Expand Down Expand Up @@ -220,14 +176,7 @@ def fit_force_constants(
``SortedForceConstants`` object, array of parameters, cluster space,
and a dictionary of information on the fitting results.
"""
<<<<<<< HEAD
from hiphive.fitting import Optimizer
from hiphive import ForceConstantPotential, enforce_rotational_sum_rules

logger.info("Starting fitting force constants.")
=======
logger.info("Starting force constant fitting.")
>>>>>>> 8f4f7e37 (gruneisen and CTE)

fitting_data = {
"cutoffs": [],
Expand All @@ -242,20 +191,13 @@ def fit_force_constants(

best_fit = {
"n_imaginary": np.inf,
<<<<<<< HEAD
"free_energy": np.inf,
=======
"rmse_test": np.inf,
"cluster_space": None,
>>>>>>> cd00c732 (cluster_space to best_fit)
"force_constants": None,
"parameters":None,
"cutoffs": None,
}
n_cutoffs = len(all_cutoffs)
<<<<<<< HEAD
for i, cutoffs in enumerate(all_cutoffs):
=======

fit_kwargs = fit_kwargs if fit_kwargs else {}
if fit_method == "rfe" and n_jobs == -1:
Expand Down Expand Up @@ -374,48 +316,15 @@ def _run_cutoffs(
supercell_atoms = structures[0]

if not is_cutoff_allowed(supercell_atoms, max(cutoffs)):
>>>>>>> 8f4f7e37 (gruneisen and CTE)
logger.info(
"Testing cutoffs {} out of {}: {}".format(i, n_cutoffs, cutoffs)
"Skipping cutoff due as it is not commensurate with supercell size"
)
<<<<<<< HEAD
sc = get_structure_container(cutoffs, structures)
<<<<<<< HEAD
opt = Optimizer(sc.get_fit_data(), fit_method)
=======
logger.info('{}, {}: FINE UNTIL SC'.format(i,cutoffs))
opt = Optimizer(sc.get_fit_data(), fit_method, **fit_kwargs)
logger.info('{}, {}: FINE UNTIL OPT'.format(i,cutoffs))
>>>>>>> fddd9681 (get_cutoffs)
=======
return None

cs = ClusterSpace(supercell_atoms, cutoffs, symprec=1e-3, acoustic_sum_rules=True)
logger.debug(cs.__repr__())
n2nd = cs.get_n_dofs_by_order(2)
nall = cs.n_dofs
<<<<<<< HEAD

if separate_fit:
logger.info('Fitting harmonic force constants separately')
# fit_data = get_fit_data(cs, supercell_atoms, structures, separate_fit,
# disp_cut, ncut=n2nd, param2=None)
sc = get_structure_container(cs, structures, separate_fit, disp_cut,
ncut=n2nd, param2=None)
opt = Optimizer(sc.get_fit_data(),
fit_method,
[0,n2nd],
**fit_kwargs)
<<<<<<< HEAD
logger.info('Optimizer set up for cutoff: {}, {}'.format(i,cutoffs))
>>>>>>> 79348b89 (bulk_mod input)
=======
>>>>>>> cc7520c5 (separate fitting)
opt.train()
param_harmonic = opt.parameters # harmonic force constant parameters

logger.info('Fitting anharmonic force constants separately')
=======

logger.info('Fitting harmonic force constants separately')
separate_fit = True
Expand Down Expand Up @@ -446,7 +355,6 @@ def _run_cutoffs(

if imaginary:
logger.info('Imaginary modes found! Fitting anharmonic force constants separately')
>>>>>>> 2e30eb4a (remove separate_fit)
# fit_data = get_fit_data(cs, supercell_atoms, structures, separate_fit,
# disp_cut, ncut=n2nd, param2=param_harmonic)
sc = get_structure_container(cs, structures, separate_fit, disp_cut,
Expand All @@ -457,75 +365,7 @@ def _run_cutoffs(
**fit_kwargs)
opt.train()
param_anharmonic = opt.parameters # anharmonic force constant parameters

<<<<<<< HEAD
parameters = enforce_rotational_sum_rules(
sc.cluster_space, opt.parameters, ["Huang", "Born-Huang"]
)
fcp = ForceConstantPotential(sc.cluster_space, parameters)
fcs = fcp.get_force_constants(supercell_atoms)
logger.info('FCS generated for cutoff {}, {}'.format(i,cutoffs))
n_imaginary, min_freq, free_energy, entropy, Cv, grun, cte, dLfrac = evaluate_force_constants(
parent_structure, supercell_matrix, fcs, bulk_modulus, T_QHA, imaginary_tol
)
<<<<<<< HEAD
<<<<<<< HEAD
<<<<<<< HEAD

fitting_data["cutoffs"].append(cutoffs)
fitting_data["rmse_test"].append(opt.rmse_test)
fitting_data["n_imaginary"].append(n_imaginary)
fitting_data["min_frequency"].append(min_freq)
fitting_data["300K_free_energy"].append(free_energy)

if (
min_freq > -np.abs(max_imaginary_freq)
and n_imaginary <= max_n_imaginary
and n_imaginary < best_fit["n_imaginary"]
and free_energy < best_fit["free_energy"]
):
best_fit.update(
{
"n_imaginary": n_imaginary,
"free_energy": free_energy,
"force_constants": fcs,
"cutoffs": cutoffs,
}
)
fitting_data["best"] = cutoffs

logger.info("Finished fitting force constants.")

return best_fit["force_constants"], fitting_data
=======
=======
logger.info('evaluate_force_constants success!')
>>>>>>> fddd9681 (get_cutoffs)
=======
logger.info('evaluate_force_constants success: {}, {}'.format(i,cutoffs))
>>>>>>> 79348b89 (bulk_mod input)
return {
"cutoffs": cutoffs,
"rmse_test": opt.rmse_test,
"n_imaginary": n_imaginary,
"min_frequency": min_freq,
"temperature": T_QHA,
"free_energy": free_energy,
"entropy": entropy,
"heat_capacity": Cv,
"thermal_expansion": cte,
"cluster_space": sc.cluster_space,
"parameters": parameters,
"force_constants": fcs
}
except Exception:
return None
>>>>>>> 8f4f7e37 (gruneisen and CTE)


def get_structure_container(
cutoffs: List[float], structures: List["Atoms"]
=======

parameters = np.concatenate((param_harmonic,param_anharmonic)) # combine
assert(nall==len(parameters))
logger.info('Training complete for cutoff: {}, {}'.format(i,cutoffs))
Expand Down Expand Up @@ -572,7 +412,6 @@ def get_structure_container(
disp_cut: float,
ncut: int,
param2: np.ndarray
>>>>>>> cc7520c5 (separate fitting)
) -> "StructureContainer":
"""
Get a hiPhive StructureContainer from cutoffs and a list of atoms objects.
Expand Down

0 comments on commit 743666e

Please sign in to comment.