Iterative DFT vs MLIP benchmark workflow for phonons#
We will not perform VASP calculations in realtime for this tutorial, but rather mock vasp runs. Thus, it is necessary to set folders with pre-computed VASP output files for execution in the notebook.
# Please note that I am reusing the same relaxations here in several steps.
# This is only to save storage on our repo. It has influence on the result.
ref_paths = {
"tight relax 1_mp-117_0.94_pre1": "tutorial_data/tight_relax_1_mp-117_0.94_0_42",
"tight relax 1_mp-117_0.94_0": "tutorial_data/tight_relax_1_mp-117_0.94_0_42",
"tight relax 1_mp-117_0.94_1": "tutorial_data/tight_relax_1_mp-117_0.94_0_42",
"dft tight relax 1_mp-117_0.94_0": "tutorial_data/tight_relax_1_mp-117_0.94_0_42",
"tight relax 2_mp-117_0.94_pre1": "tutorial_data/tight_relax_2_mp-117_0.94_0_43",
"tight relax 2_mp-117_0.94_0": "tutorial_data/tight_relax_2_mp-117_0.94_0_43",
"tight relax 2_mp-117_0.94_1": "tutorial_data/tight_relax_2_mp-117_0.94_0_43",
"dft tight relax 2_mp-117_0.94_0": "tutorial_data/tight_relax_2_mp-117_0.94_0_43",
"dft static_mp-117_0.94_0": "tutorial_data/tight_relax_2_mp-117_0.94_0_43",
"tight relax 1_mp-117_1.0_pre1": "tutorial_data/tight_relax_1_mp-117_1.0_0_47",
"tight relax 1_mp-117_1.0_0": "tutorial_data/tight_relax_1_mp-117_1.0_0_47",
"tight relax 1_mp-117_1.0_1": "tutorial_data/tight_relax_1_mp-117_1.0_0_47",
"dft tight relax 1_mp-117_1.0_0": "tutorial_data/tight_relax_1_mp-117_1.0_0_47",
"dft tight relax 1_mp-117_1.0_1": "tutorial_data/tight_relax_1_mp-117_1.0_0_47",
"tight relax 2_mp-117_1.0_pre1": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"tight relax 2_mp-117_1.0_0": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"tight relax 2_mp-117_1.0_1": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"dft tight relax 2_mp-117_1.0_0": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"dft tight relax 2_mp-117_1.0_1": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"dft static_mp-117_1.0_0": "tutorial_data/tight_relax_2_mp-117_1.0_0_48",
"tight relax 1_mp-117_1.06_pre1": "tutorial_data/tight_relax_1_mp-117_1.06_0_52",
"tight relax 1_mp-117_1.06_0": "tutorial_data/tight_relax_1_mp-117_1.06_0_52",
"tight relax 1_mp-117_1.06_1": "tutorial_data/tight_relax_1_mp-117_1.06_0_52",
"dft tight relax 1_mp-117_1.06_0": "tutorial_data/tight_relax_1_mp-117_1.06_0_52",
"dft tight relax 1_mp-117_1.06_1": "tutorial_data/tight_relax_1_mp-117_1.06_0_52",
"tight relax 2_mp-117_1.06_pre1": "tutorial_data/tight_relax_2_mp-117_1.06_0_53",
"tight relax 2_mp-117_1.06_0": "tutorial_data/tight_relax_2_mp-117_1.06_0_53",
"tight relax 2_mp-117_1.06_1": "tutorial_data/tight_relax_2_mp-117_1.06_0_53",
"dft tight relax 2_mp-117_1.06_0": "tutorial_data/tight_relax_2_mp-117_1.06_0_53",
"dft static_mp-117_1.06_0": "tutorial_data/tight_relax_2_mp-117_1.06_0_53",
"Sn-stat_iso_atom_0": "tutorial_data/Sn-stat_iso_atom_0_25",
"Sn-stat_iso_atom_1": "tutorial_data/Sn-stat_iso_atom_0_25",
"dft rattle static 1/1_mp-117_0.94_0": "tutorial_data/dft_rattle_static_1_1_mp-117_0.94_0_63",
"dft rattle static 1/1_mp-117_0.94_1": "tutorial_data/dft_rattle_static_1_1_mp-117_0.94_1_257",
"dft rattle static 1/1_mp-117_1.0_0": "tutorial_data/dft_rattle_static_1_1_mp-117_1.0_0_65",
"dft rattle static 1/1_mp-117_1.0_1": "tutorial_data/dft_rattle_static_1_1_mp-117_1.0_1_255",
"dft rattle static 1/1_mp-117_1.06_0": "tutorial_data/dft_rattle_static_1_1_mp-117_1.06_0_67",
"dft rattle static 1/1_mp-117_1.06_1": "tutorial_data/dft_rattle_static_1_1_mp-117_1.06_1_259",
"dft phonon static 1/1_mp-117_0.94_0": "tutorial_data/dft_phonon_static_1_1_mp-117_0.94_0_193",
"dft phonon static 1/1_mp-117_1.0_0": "tutorial_data/dft_phonon_static_1_1_mp-117_1.0_0_189",
"dft phonon static 1/1_mp-117_1.06_0": "tutorial_data/dft_phonon_static_1_1_mp-117_1.06_0_191",
}
import warnings
from atomate2.vasp.flows.core import DoubleRelaxMaker
from atomate2.vasp.jobs.core import StaticMaker, TightRelaxMaker
from atomate2.vasp.jobs.phonons import PhononDisplacementMaker
from atomate2.vasp.sets.core import StaticSetGenerator, TightRelaxSetGenerator
from jobflow import Flow, run_locally
from mock_vasp import TEST_DIR, mock_vasp
from pymatgen.core.structure import Structure
from autoplex.auto.phonons.flows import (
CompleteDFTvsMLBenchmarkWorkflow,
IterativeCompleteDFTvsMLBenchmarkWorkflow,
)
warnings.filterwarnings("ignore")
First, we define all relevant Makers for the workflow, used to train and finetune ML potentials for phonons. We will pre-relax the structures before starting the workflow. We will now define the relax maker, a displacement maker (static maker), static energy maker and a static energy maker for isolated atoms.
phonon_bulk_relax_maker = DoubleRelaxMaker.from_relax_maker(
TightRelaxMaker(
run_vasp_kwargs={"handlers": ()},
input_set_generator=TightRelaxSetGenerator(
user_incar_settings={
"GGA": "PE",
"ISPIN": 1,
"KSPACING": 0.1,
"ALGO": "Normal",
"LAECHG": False,
"ISMEAR": 1,
"ENCUT": 700,
"IBRION": 1,
"ISYM": 0,
"SIGMA": 0.05,
"LCHARG": False,
"LWAVE": False,
"LVTOT": False,
"LORBIT": None,
"LOPTICS": False,
"LREAL": False,
"ISIF": 4,
"NPAR": 4,
}
),
)
)
phonon_displacement_maker = PhononDisplacementMaker(
name="dft phonon static",
run_vasp_kwargs={"handlers": ()},
input_set_generator=StaticSetGenerator(
user_incar_settings={
"GGA": "PE",
"IBRION": -1,
"ISPIN": 1,
"ISMEAR": 1,
"ISIF": 3,
"ENCUT": 700,
"EDIFF": 1e-7,
"LAECHG": False,
"LREAL": False,
"ALGO": "Normal",
"NSW": 0,
"LCHARG": False,
"LWAVE": False,
"LVTOT": False,
"LORBIT": None,
"LOPTICS": False,
"SIGMA": 0.05,
"ISYM": 0,
"KSPACING": 0.1,
"NPAR": 4,
},
auto_ispin=False,
),
)
phonon_static_energy_maker = phonon_displacement_maker
static_isolated_atom_maker = StaticMaker(
run_vasp_kwargs={"handlers": ()},
input_set_generator=StaticSetGenerator(
user_kpoints_settings={"reciprocal_density": 1},
user_incar_settings={
"GGA": "PE",
"ALGO": "Normal",
"ISPIN": 1,
"LAECHG": False,
"ISMEAR": 0,
"LCHARG": False,
"LWAVE": False,
"LVTOT": False,
"LORBIT": None,
"LOPTICS": False,
"NPAR": 4,
},
),
)
First, collect a number of structures and then optimize them in advance of the workflow. One can also perform subsequent optimizations with different k-point settings, for example.
job_list = []
structure_list = []
benchmark_structure_list = []
start_mpids = ["mp-117"]
start_poscars = [TEST_DIR / "tutorial_data/POSCAR-mp-117"]
mpids = []
for mpid, start_poscar in zip(start_mpids, start_poscars):
for scale in [0.94, 1.0, 1.06]:
structure = Structure.from_file(start_poscar)
volume = structure.copy().volume
structure = structure.scale_lattice((scale**3) * volume) # added the cube
job_opt = phonon_bulk_relax_maker.make(structure)
job_opt.append_name("_" + mpid + "_" + str(scale) + "_pre1")
job_list.append(job_opt)
structure_list.append(job_opt.output.structure)
mpids.append(mpid + "_" + str(scale))
mpbenchmark = mpids
benchmark_structure_list = structure_list
iteration_flow = IterativeCompleteDFTvsMLBenchmarkWorkflow(
max_iterations=2, # with the current test data, you can switch between 1 and 2
rms_max=0.2,
complete_dft_vs_ml_benchmark_workflow_0=CompleteDFTvsMLBenchmarkWorkflow(
symprec=1e-3,
apply_data_preprocessing=True,
add_dft_rattled_struct=True,
add_dft_phonon_struct=True,
volume_custom_scale_factors=[1.0],
rattle_type=0,
distort_type=0,
rattle_std=0.1, #
benchmark_kwargs={"relax_maker_kwargs": {"relax_cell": False}},
supercell_settings={
"min_length": 10,
"max_length": 15,
"min_atoms": 10,
"max_atoms": 300,
"fallback_min_length": 9,
},
# settings that worked with a GAP
split_ratio=0.33,
regularization=False,
separated=False,
num_processes_fit=48,
displacement_maker=phonon_displacement_maker,
phonon_bulk_relax_maker=phonon_bulk_relax_maker,
phonon_static_energy_maker=phonon_static_energy_maker,
rattled_bulk_relax_maker=phonon_bulk_relax_maker,
isolated_atom_maker=static_isolated_atom_maker,
),
complete_dft_vs_ml_benchmark_workflow_1=CompleteDFTvsMLBenchmarkWorkflow(
symprec=1e-3,
apply_data_preprocessing=True,
add_dft_phonon_struct=False,
add_dft_rattled_struct=True,
volume_custom_scale_factors=[1.0],
rattle_type=0,
distort_type=0,
rattle_std=0.1, # maybe 0.1
benchmark_kwargs={"relax_maker_kwargs": {"relax_cell": False}},
supercell_settings={
"min_length": 10,
"max_length": 15,
"min_atoms": 10,
"max_atoms": 300,
"fallback_min_length": 9,
},
# settings that worked with a GAP
split_ratio=0.33,
regularization=False,
separated=False,
num_processes_fit=48,
displacement_maker=phonon_displacement_maker,
phonon_bulk_relax_maker=phonon_bulk_relax_maker,
phonon_static_energy_maker=phonon_static_energy_maker,
rattled_bulk_relax_maker=phonon_bulk_relax_maker,
isolated_atom_maker=static_isolated_atom_maker,
),
).make(
structure_list=structure_list,
mp_ids=mpids,
benchmark_structures=benchmark_structure_list,
benchmark_mp_ids=mpbenchmark,
rattle_seed=0,
fit_kwargs_list=[
{
"soap": {
"delta": 1.0,
"l_max": 12,
"n_max": 10,
"atom_sigma": 0.5,
"zeta": 4,
"cutoff": 5.0,
"cutoff_transition_width": 1.0,
"central_weight": 1.0,
"n_sparse": 6000,
"f0": 0.0,
"covariance_type": "dot_product",
"sparse_method": "cur_points",
},
"general": {
"two_body": True,
"three_body": False,
"soap": True,
"default_sigma": "{0.001 0.05 0.05 0.0}",
"sparse_jitter": 1.0e-8,
},
}
],
)
job_list.append(iteration_flow)
autoplex_flow = Flow(jobs=job_list, output=iteration_flow.output)
Now, we are mocking the VASP execution. If you would like leave all folders from the run, set clean_folders to False!
with mock_vasp(ref_paths=ref_paths, clean_folders=False) as mf:
run_locally(
autoplex_flow,
create_folders=True,
ensure_success=True,
raise_immediately=True,
)