Active Learning: Ru/H Part 3, Gas Snapshots¶
Continue the Ru/H workflow by generating gas-phase and interface snapshots, replaying them with the reference method, and adding the results to the training data. Run parts 1 and 2 first.
Downloads: Notebook | Script ?
Additional files required by this example:
common_ru_h.py
Requires: AMS2026 or later
Required AMS Packages: m3gnet
Related examples
Related tutorials
Related documentation
Initial imports¶
import scm.plams as plams
from scm.params import ResultsImporter
from scm.plams import Settings, AMSJob, log, Molecule, packmol_on_slab
from pathlib import Path
import matplotlib.pyplot as plt
# common_ru_h.py must exist in the current working directory
from common_ru_h import (
rotation,
dft_settings,
QEKPointsConfig,
m3gnet_up_settings,
replay_settings,
slice_slab,
check_installation,
)
# register dependencies for AMSjobs, to support submitting this notebook directly to a cluster in AMS2025+
# dependency: {} common_ru_h.py
# dependency: {} reference_data_2
Initialize PLAMS working directory¶
old_ref_dir = "reference_data_2"
check_installation(old_ref_dir)
new_ref_dir = "reference_data_3"
ri = ResultsImporter.from_yaml(old_ref_dir)
plams.init()
Current AMS version: 2024.102
05-31 14:23:05 m3gnet is installed: M3GNet ML Backend v[0.2.4] - build:0 [06668e0a45ce742d8f66ff23484b8a1e]
05-31 14:23:06 qe is installed: Quantum ESPRESSO (AMSPIPE) v[7.1] - build:115 [777d72eb480fe4d632a003cc62e9c1cb]
PLAMS working folder: /path/to/plams_workdir.003
Construct the Ru(10-10)/H2(gas) interface¶
For details about the construction of the slab, see the previous notebook.
optimized_bulk = ri.job_collection["hcp_lattopt_Ru_dft"].molecule
slab_100 = slice_slab(
optimized_bulk, miller=(1, 0, 0), thickness=7.0, cell_z=15, ref_atom=0
)
slab_100 = slab_100.supercell(3, 2, 1)
for at in slab_100:
at.properties = Settings() # remove details about supercell generation
plams.plot_molecule(slab_100, rotation=rotation)
plt.title("Ru(10-10)");
Now use the packmol_slab function to add hydrogen molecules:
from scm.plams import packmol_on_slab
h2_mol = plams.from_smiles("[HH]")
density = 0.3 # approximate density of the gas phase in g/cm^3
slab_100_H2_gas_raw = packmol_on_slab(slab_100, h2_mol, density=density)
slab_100_H2_gas = plams.preoptimize(
slab_100_H2_gas_raw, settings=m3gnet_up_settings(), maxiterations=100
)
plams.plot_molecule(slab_100_H2_gas, rotation=rotation, radii=0.8)
Now let’s run some MD:
mdjob = plams.AMSNVTJob(
settings=m3gnet_up_settings(),
name="md_ru10-10_h2_m3gnet",
temperature=501,
nsteps=10000,
molecule=slab_100_H2_gas,
samplingfreq=100,
)
mdjob.run();
[31.05|14:23:35] JOB md_ru10-10_h2_m3gnet STARTED
[31.05|14:23:35] JOB md_ru10-10_h2_m3gnet RUNNING
[31.05|15:08:27] JOB md_ru10-10_h2_m3gnet FINISHED
[31.05|15:08:28] JOB md_ru10-10_h2_m3gnet SUCCESSFUL
from scm.params import ResultsImporter
ri = ResultsImporter.from_yaml(old_ref_dir)
settings = dft_settings(QEKPointsConfig(3, 3, 1))
settings += replay_settings(mdjob.results.rkfpath(), frames=[10, 30, 50, 70, 90])
dft_replay_job = plams.AMSJob(
settings=settings,
name="snapshots_from_md_ru10-10_h2_dft",
)
dft_replay_job.run(watch=True)
[31.05|15:08:28] JOB snapshots_from_md_ru10-10_h2_dft STARTED
[31.05|15:08:28] JOB snapshots_from_md_ru10-10_h2_dft RUNNING
[31.05|15:08:29] snapshots_from_md_ru10-10_h2_dft: AMS 2024.102 RunTime: May31-2024 15:08:29 ShM Nodes: 1 Procs: 1
[31.05|15:08:29] snapshots_from_md_ru10-10_h2_dft: Starting trajectory replay ...
[31.05|15:08:29] snapshots_from_md_ru10-10_h2_dft: Replaying frame #1/5 (#10 in original trajectory)
... output trimmed ....
[31.05|15:45:36] snapshots_from_md_ru10-10_h2_dft: Trajectory replay complete.
[31.05|15:46:36] snapshots_from_md_ru10-10_h2_dft: Number of QE evaluations: 5
[31.05|15:46:36] snapshots_from_md_ru10-10_h2_dft: NORMAL TERMINATION
[31.05|15:46:36] JOB snapshots_from_md_ru10-10_h2_dft FINISHED
[31.05|15:46:36] JOB snapshots_from_md_ru10-10_h2_dft SUCCESSFUL
<scm.plams.interfaces.adfsuite.ams.AMSResults at 0x14c005cbf7c0>
ri.add_trajectory_singlepoints(dft_replay_job, properties=["energy", "forces"])
ri.store(new_ref_dir)
['reference_data_3/job_collection.yaml',
'reference_data_3/results_importer_settings.yaml',
'reference_data_3/training_set.yaml']
See also¶
Python Script¶
#!/usr/bin/env python
# coding: utf-8
# ## Initial imports
import scm.plams as plams
from scm.params import ResultsImporter
from scm.plams import Settings, AMSJob, log, Molecule, packmol_on_slab
from pathlib import Path
import matplotlib.pyplot as plt
# common_ru_h.py must exist in the current working directory
from common_ru_h import (
rotation,
dft_settings,
QEKPointsConfig,
m3gnet_up_settings,
replay_settings,
slice_slab,
check_installation,
)
# register dependencies for AMSjobs, to support submitting this notebook directly to a cluster in AMS2025+
# dependency: {} common_ru_h.py
# dependency: {} reference_data_2
# ## Initialize PLAMS working directory
old_ref_dir = "reference_data_2"
check_installation(old_ref_dir)
new_ref_dir = "reference_data_3"
ri = ResultsImporter.from_yaml(old_ref_dir)
plams.init()
# ## Construct the Ru(10-10)/H2(gas) interface
#
# For details about the construction of the slab, see the previous notebook.
optimized_bulk = ri.job_collection["hcp_lattopt_Ru_dft"].molecule
slab_100 = slice_slab(optimized_bulk, miller=(1, 0, 0), thickness=7.0, cell_z=15, ref_atom=0)
slab_100 = slab_100.supercell(3, 2, 1)
for at in slab_100:
at.properties = Settings() # remove details about supercell generation
plams.plot_molecule(slab_100, rotation=rotation)
plt.title("Ru(10-10)")
plt.gcf().savefig("picture1.png")
plt.clf()
# Now use the ``packmol_slab`` function to add hydrogen molecules:
from scm.plams import packmol_on_slab
h2_mol = plams.from_smiles("[HH]")
density = 0.3 # approximate density of the gas phase in g/cm^3
slab_100_H2_gas_raw = packmol_on_slab(slab_100, h2_mol, density=density)
slab_100_H2_gas = plams.preoptimize(slab_100_H2_gas_raw, settings=m3gnet_up_settings(), maxiterations=100)
plams.plot_molecule(slab_100_H2_gas, rotation=rotation, radii=0.8)
# Now let's run some MD:
mdjob = plams.AMSNVTJob(
settings=m3gnet_up_settings(),
name="md_ru10-10_h2_m3gnet",
temperature=501,
nsteps=10000,
molecule=slab_100_H2_gas,
samplingfreq=100,
)
mdjob.run()
from scm.params import ResultsImporter
ri = ResultsImporter.from_yaml(old_ref_dir)
settings = dft_settings(QEKPointsConfig(3, 3, 1))
settings += replay_settings(mdjob.results.rkfpath(), frames=[10, 30, 50, 70, 90])
dft_replay_job = plams.AMSJob(
settings=settings,
name="snapshots_from_md_ru10-10_h2_dft",
)
dft_replay_job.run(watch=True)
ri.add_trajectory_singlepoints(dft_replay_job, properties=["energy", "forces"])
ri.store(new_ref_dir)
Required additional files¶
common_ru_h.py
"""
This file is part of the Ru/H active learning example in AMS.
The most important things to modify/inspect are:
1) the TESTING_MODE (see below), and
2) the dft_settings() function.
"""
import matplotlib.pyplot as plt
import numpy as np
import os
import scm.plams as plams
from dataclasses import dataclass
from pathlib import Path
from scm.base import Units, Lattice
from scm.utils.conversions import plams_molecule_to_chemsys, chemsys_to_plams_molecule
from typing import List, Optional, Literal, Tuple, ClassVar, Sequence, Union
from scm.version import release
import subprocess
ang2bohr = Units.conversion_factor("angstrom", "bohr") # angstrom to bohr
rotation = "-85x,-5y,0z" # plotting view in Jupyter notebook
TESTING_MODE = False # set to True to use a custom M3GNet instead of DFT calculations for quick testing run-through
@dataclass
class QEKPointsConfig:
"""Class representing the K_Points input for Quantum ESPRESSO"""
x: int = 1
y: int = 1
z: int = 1
shift_x: bool = False
shift_y: bool = False
shift_z: bool = False
def to_settings(self) -> plams.Settings:
"""Converts config to PLAMS Settings at the top level"""
s = plams.Settings()
if self.x == 1 and self.y == 1 and self.z == 1:
s.input.QuantumEspresso.K_Points._h = "gamma"
return s
s.input.QuantumEspresso.K_Points._h = "automatic"
s.input.QuantumEspresso.K_Points._1 = (
f"{self.x} {self.y} {self.z} " f"{int(self.shift_x)} {int(self.shift_y)} {int(self.shift_z)}"
)
return s
###########################################
#### Calculation settings functions
###########################################
def dft_settings(kpoints: QEKPointsConfig = QEKPointsConfig(), conv_thr: float = 1e-6) -> plams.Settings:
"""Returns PLAMS Settings for Quantum ESPRESSO PBE-D3(BJ) with Gaussian Smearing
kpoints will default to the Gamma point if not specified.
"""
if TESTING_MODE:
d = Path(os.path.expandvars("$AMSRESOURCES/MLPotential/M3GNet/RuH/engine.txt"))
if not d.exists():
raise FileNotFoundError(f"Couldn't find {d}. This file/directory was added in AMS2024.102.")
plams.log(
f"Warning: Using custom M3GNet potential from {d} and not DFT! "
"Set TESTING_MODE = False in common_ru_h.py for running proper calculations."
)
s = plams.AMSJob.from_inputfile(str(d)).settings
s.runscript.nproc = 1
return s
s = plams.Settings()
s += kpoints.to_settings()
s.input.QuantumEspresso.System.occupations = "smearing"
s.input.QuantumEspresso.System.degauss = 0.015
s.input.QuantumEspresso.Pseudopotentials.Family = "SSSP-Efficiency"
s.input.QuantumEspresso.Pseudopotentials.Functional = "PBE"
s.input.QuantumEspresso.System.dftd3_version = 4
s.input.QuantumEspresso.System.vdw_corr = "Grimme-D3"
s.input.QuantumEspresso.Electrons.conv_thr = conv_thr
# delete the *.save, worker.*, and *.xml files since they are no longer needed
# for parametrization jobs
s.runscript.postamble_lines = ["rm -rf quantumespresso.save PESPoint*.save worker.* *.xml"]
return s
def replay_settings(rkf: Union[str, os.PathLike], frames: Optional[List[int]] = None) -> plams.Settings:
"""Settings for AMS Replay jobs. Always sets Properties%Gradients = "Yes".
:param rkf: Path to ams.rkf file from which to extract frames
:type rkf: os.PathLike
:param frames: Which frames (indexing starts with 1) to replay. For details, see the AMS Replay documentation.
:type frames: Optional[List[int]], optional
:raises FileNotFoundError: If the ams.rkf file does not exist.
:return: Returns PLAMS Settings object at the top level.
:rtype: plams.Settings
"""
s = plams.Settings()
s.input.ams.Task = "Replay"
s.input.ams.Properties.Gradients = "Yes"
rkf = Path(rkf).resolve()
if not rkf.exists():
raise FileNotFoundError(f"{rkf} does not exist")
s.input.ams.Replay.File = str(rkf)
if frames:
s.input.ams.Replay.Frames = " ".join(str(x) for x in frames)
return s
def m3gnet_up_settings() -> plams.Settings:
"""Returns PLAMS Settings for the M3GNet Universal Potential"""
s = plams.Settings()
s.input.MLPotential.Model = "M3GNet-UP-2022"
s.runscript.nproc = 1 # always run AMS Driver in serial for MLPotential
return s
def lattice_optimization_settings() -> plams.Settings:
"""Returns PLAMS settings for lattice optimization"""
s = plams.Settings()
s.input.ams.Task = "GeometryOptimization"
s.input.ams.GeometryOptimization.OptimizeLattice = "Yes"
return s
def constraints_settings(constrained_atoms: Optional[Union[int, Sequence[int]]] = None) -> plams.Settings:
"""Sets settings.input.ams.Constraints.Atom. Returns Settings at top level"""
s = plams.Settings()
if constrained_atoms is None:
return s
if isinstance(constrained_atoms, int):
s.input.ams.Constraints.Atom = [constrained_atoms]
else:
s.input.ams.Constraints.Atom = list(constrained_atoms)
return s
def pesscan_settings(
scan_coordinates: Sequence["SingleScanCoordinate"],
n_points=10,
) -> plams.Settings:
s = plams.Settings()
s.input.ams.Task = "PESScan"
# fix at least 1 Ru atom in place to prevent the entire slab from diffusing
s.input.ams.PESScan.ScanCoordinate = [plams.Settings()]
s.input.ams.PESScan.ScanCoordinate[0].nPoints = n_points
s.input.ams.PESScan.ScanCoordinate[0] += scan_coordinate_list_to_settings(scan_coordinates)
s.input.ams.PESScan.CalcPropertiesAtPESPoints = "Yes"
return s
###########################################
##### General structure functions #########
###########################################
def slice_slab(
bulk: plams.Molecule,
miller: Tuple[int, int, int],
thickness: float = 7.0,
cell_z: float = 15.0,
ref_atom: int = 0,
) -> plams.Molecule:
"""Returns a slab cut out from the bulk crystal.
:param bulk: The bulk crystal
:type bulk: plams.Molecule
:param miller: Miller indices. For hexagonal crystals with 4 indices, do not specify the third index.
:type miller: Tuple[int, int, int]
:param thickness: Approximate thickness of slab (angstrom), defaults to 7.0
:type thickness: float, optional
:param cell_z: Lattice parameter in the surface normal direction (angstrom), defaults to 15.0
:type cell_z: float, optional
:param ref_atom: Reference atom index (0-based), defaults to 0
:type ref_atom: int, optional
:return: Returns a slab (3D periodicity with a vacuum gap)
:rtype: plams.Molecule
"""
def with_vacuum_lattice(system, vacuum_height: float):
"""Return the chemical system with a fresh 3D lattice including vacuum."""
vectors = np.asarray(system.lattice.vectors)
new_vectors = np.concatenate((vectors, [[0.0, 0.0, vacuum_height]]), axis=0)
system.lattice = Lattice(new_vectors)
return system
if get_ams_version() < "2024.2":
# Backwards-compatible version for AMS2024 where Chemical System was in Bohr, had transposed coords and different slice API
slab = plams_molecule_to_chemsys(bulk)
slab.slice_thickness(
ref_atom=ref_atom,
top=(cell_z / 2 + thickness / 2) * ang2bohr, # bohr
bottom=(cell_z / 2 - thickness / 2) * ang2bohr, # bohr
miller=miller, # 1, 0, -1, 0 (third index ignored for hexagonal cell)
translate=0.0,
)
slab = with_vacuum_lattice(slab, cell_z * ang2bohr)
else:
bulk_cs = plams_molecule_to_chemsys(bulk)
slab = bulk_cs.make_slab_thickness(
miller=miller, # 1, 0, -1, 0 (third index ignored for hexagonal cell)
top=(cell_z / 2 + thickness / 2),
bottom=(cell_z / 2 - thickness / 2),
translate=0.0,
ref_atom=ref_atom,
)
slab = with_vacuum_lattice(slab, cell_z)
# slab.supercell([3, 2]) # create a 3x2 supercell
slab.map_atoms([0, 0, 0]) # maps atoms into the 0..1 0..1 unit cell
return chemsys_to_plams_molecule(slab)
def add_adsorbate(
slab: plams.Molecule,
symbol: str = "H",
frac_x: float = 0.0,
frac_y: float = 0.0,
delta_z: float = 2.0,
) -> plams.Molecule:
"""
Adds an atom with given fractional xy coordinates on top of the slab, with a distance of delta_z (angstrom).
Returns a new plams.Molecule
"""
surface_lattice = np.array(slab.lattice)[:2, :2]
x, y = frac_x * surface_lattice[0] + frac_y * surface_lattice[1]
max_z = np.max(slab.as_array()[:, 2])
z = max_z + delta_z
ret = slab.copy()
ret.add_atom(plams.Atom(symbol=symbol, coords=(x, y, z)))
return ret
def get_bottom_atom_index(mol: plams.Molecule) -> int:
"""Returns 1-based index of the atom with the smallest z coordinate"""
return int(np.argmin(mol.as_array()[:, 2])) + 1
##########################################################
##### Functions and classes for PES Scan coordinates #####
##########################################################
@dataclass
class SingleScanCoordinate:
key: ClassVar[str] = "MustBeOverridden"
@dataclass
class CartesianScanCoordinate(SingleScanCoordinate):
"""
Small helper class for data inside a cartesian coordinate in the PESScan task.
Multiple cartesian scan coordinates can be combined into a single PESScan scan coordinate.
"""
key: ClassVar[str] = "Coordinate"
atom: int # first atom has index 1
coordinate: Literal["x", "y", "z"]
start: float # starting coordinate in angstrom
end: float # final coordinate in angstrom
def __str__(self) -> str:
"""Returns a string in the AMS input format"""
return f"{self.atom} {self.coordinate} {self.start:.4f} {self.end:.4f}"
@dataclass
class CellVolumeScalingRangeScanCoordinate(SingleScanCoordinate):
"""
Small helper class for data inside a cell volume scaling range coordinate in the PESScan task.
"""
key: ClassVar[str] = "CellVolumeScalingRange"
start: float # starting coordinate in angstrom
end: float # final coordinate in angstrom
def __str__(self) -> str:
"""Returns a string in the AMS input format"""
return f"{self.start} {self.end}"
@dataclass
class DistanceScanCoordinate(SingleScanCoordinate):
"""
Small helper class for data inside a distance coordinate in the PESScan task.
"""
key: ClassVar[str] = "Distance"
atom1: int # first atom has index 1
atom2: int # first atom has index 1
start: float # starting coordinate in angstrom
end: float # final coordinate in angstrom
def __str__(self) -> str:
"""Returns a string in the AMS input format"""
return f"{self.atom1} {self.atom2} {self.start:.4f} {self.end:.4f}"
def scan_coordinate_list_to_settings(scan_coordinates: Sequence[SingleScanCoordinate]) -> plams.Settings:
"""
Return value is not at the top level.
Example: s.input.ams.PESScan.ScanCoordinate[0] = scan_coordinate_list_to_settings([sc1, sc2])
"""
s = plams.Settings()
for x in scan_coordinates:
if x.key not in s:
s[x.key] = []
s[x.key].append(str(x))
return s
def get_surface_diffusion_scan_coordinates(
slab: plams.Molecule,
atom_index: int,
) -> List[CartesianScanCoordinate]:
"""Cause the atom with index ``atom_index`` (1-based) to diffuse to the top right corner of the unit cell.
:param slab: Slab including adsorbate
:type slab: plams.Molecule
:param atom_index: One-based atom index, defaults to None
:type atom_index: Optional[int], optional
:return: List of CartesianScanCoordinate that can be used for PES Scans
:rtype: List[CartesianScanCoordinate]
"""
orig_x, orig_y = slab[atom_index].coords[0], slab[atom_index].coords[1]
lattice = np.array(slab.lattice)
target_xy = np.round(lattice[0, :2] + lattice[1, :2], decimals=4)
target_x, target_y = target_xy
print(
f"Atom {atom_index} will diffuse from "
f"(x, y) = {orig_x}, {orig_y} to {target_x}, {target_y} "
"with the z coordinate optimized"
)
return [
CartesianScanCoordinate(atom=atom_index, coordinate="x", start=orig_x, end=target_x),
CartesianScanCoordinate(atom=atom_index, coordinate="y", start=orig_y, end=target_y),
]
def get_bulk_diffusion_scan_coordinates(
slab: plams.Molecule, atom_index: int, delta_z: float = 2.0
) -> List[CartesianScanCoordinate]:
"""Cause the atom with index ``atom_index`` (1-based) to diffuse through the slab.
:param slab: _description_
:type slab: plams.Molecule
:param atom_index: _description_
:type atom_index: int
:return: _description_
:rtype: List[CartesianScanCoordinate]
"""
target_z = np.min(slab.as_array()[:, 2]) - delta_z
orig_z = slab[atom_index].coords[2]
print(
f"Atom {atom_index} will diffuse from "
f"z = {orig_z:.4f} to {target_z:.4f} "
"with the x and y coordinates optimized"
)
return [CartesianScanCoordinate(atom=atom_index, coordinate="z", start=orig_z, end=target_z)]
def get_surface_bond_scan_coordinates(
slab: plams.Molecule, atom_index: int, start: Optional[float] = None, end: float = 1.2
) -> List[DistanceScanCoordinate]:
"""Cause the atom with index ``atom_index`` (1-based)
to move towards its nearest neighbor until the distance is ``target_length.
start: float
If None will use the current distance
end: float
Target distance (angstrom)
"""
ase_atoms = plams.toASE(slab)
distance_matrix = ase_atoms.get_all_distances(mic=True)
row = distance_matrix[atom_index - 1, :]
row[atom_index - 1] = 1e10
min_d = np.min(row)
min_i = int(np.argmin(row) + 1)
start = start or min_d
return [DistanceScanCoordinate(atom1=atom_index, atom2=min_i, start=start, end=end)]
########
## Plotting functions
########
def plot_pesscan(job: plams.AMSJob, title: Optional[str] = None, ax=None) -> plt.Axes:
"""Plots a PESScan finished PESScan job. If multiple scan coordinates only uses the first one on the x axis."""
if ax is None:
fig, ax = plt.subplots()
r = job.results.get_pesscan_results()
ax.plot(r["RaveledPESCoords"][0], r["PES"])
ax.set_xlabel(f"{r['RaveledScanCoords'][0]} ({r['RaveledUnits'][0]})")
ax.set_ylabel("Energy (Ha)")
ax.ticklabel_format(useOffset=False)
ax.set_title(title or str(job.name))
return ax
#######
## Check that correct versions of packages are installed
#######
def run_amspackages_check(package: str) -> subprocess.CompletedProcess:
x = subprocess.run(
["sh", os.path.expandvars("$AMSBIN/amspackages"), "-v", "check", package], stdout=subprocess.PIPE, text=True
)
return x
def get_qe_version_and_build() -> Tuple[Union[str, None], Union[int, None]]:
x = run_amspackages_check("qe")
print(x.stdout.strip())
if x.returncode != 0:
return None, None
try:
version_number = x.stdout.split("v[")[1].split("]")[0]
build_number = int(x.stdout.split("build:")[1].split()[0])
except (KeyError, IndexError, TypeError):
return None, None
return version_number, build_number
def check_qe_installation():
if TESTING_MODE:
return True
min_build = 115
msg = f"QE must be at least version 7.1, build {min_build}. Install or update through the AMS package manager."
version_number, build_number = get_qe_version_and_build()
assert version_number is not None, msg
assert build_number is not None, msg
assert version_number >= "7.1", msg
if version_number == "7.1":
assert build_number >= min_build, msg
def check_m3gnet_installation():
x = run_amspackages_check("m3gnet")
print(x.stdout.strip())
assert x.returncode == 0, f"m3gnet is not installed. Install it through the AMS package manager."
def get_ams_version() -> str:
return release
def check_ams_installation(min_version="2024.102"):
ams_version = get_ams_version()
print(f"Current AMS version: {ams_version}")
assert ams_version >= min_version, f"AMS version must be at least {min_version}"
def check_installation(ref_dir: Optional[Union[Path, str]] = None):
check_ams_installation()
check_m3gnet_installation()
check_qe_installation()
if ref_dir:
assert Path(ref_dir).exists(), f"{ref_dir} must exist in the current working directory."