Using a QMModel: Performing Geometry Optimization

[1]:

import numpy as np
import ase
from ase.build import molecule
from ase.optimize import LBFGS

[2]:

from qmlearn.api.api4ase import QMLCalculator
from qmlearn.drivers.mol import QMMol
from qmlearn.io.model import db2qmmodel

Geting Energies from an external model using Delta learning by altering the initial geometry

  1. Open a database that contained the Linear Regression model using db2qmmodel

[3]:

qmmodel = db2qmmodel('h2o_vib_QML_set.hdf5')

Guess DB names : {'qmmol': 'rks/qmmol', 'atoms': 'rks/train_atoms_36', 'properties': 'rks/train_props_36'}
Guess mmodels: {'gamma': KernelRidge(alpha=0.1), 'd_gamma': LinearRegression(), 'd_energy': LinearRegression(), 'd_forces': LinearRegression()}

  1. Define second learning features energy and forces

  2. Define the Calculator, based on second_learn models.

[4]:

second_learn = {
        'energy' : 'd_energy',
        'forces' : 'd_forces',
        }
calc = QMLCalculator(qmmodel = qmmodel, second_learn = second_learn, method = 'gamma')

  1. Alter the atom_test positions by 0.2 and perform an optimization using LBFGS method.

[5]:

atoms_test=molecule('H2O')
atoms_test.positions[0,0] += 0.2

[6]:

atoms = atoms_test.copy()
atoms.calc = calc

[7]:

LBFGS(atoms, trajectory = 'opt.traj').run(fmax=0.01)
atoms.positions

       Step     Time          Energy         fmax
LBFGS:    0 16:14:33    -2070.589575        1.2729
LBFGS:    1 16:14:33    -2070.613727        0.5394
LBFGS:    2 16:14:33    -2070.613889        0.5234
LBFGS:    3 16:14:33    -2070.618351        0.0467
LBFGS:    4 16:14:33    -2070.618367        0.0336
LBFGS:    5 16:14:34    -2070.618414        0.0136
LBFGS:    6 16:14:34    -2070.618422        0.0038

[7]:

array([[ 1.99147485e-01, -4.01329618e-05,  9.49149270e-02],
       [ 4.28776896e-04,  7.65808159e-01, -4.64809616e-01],
       [ 4.23770571e-04, -7.65768012e-01, -4.64937199e-01]])

Geting Energies using PySCF engine

  1. Define the basis set, level of theory, exchange correlation functional and the total charge of your system

[8]:

basis = 'cc-pvTZ'
xc = 'lda,vwn_rpa'
method = 'rks'
charge = 0

  1. Create atoms atoms copy, and the refqmmol object to initialize the engine.

  2. Initialize the calculator using QMLCalculator class.

[9]:

atoms = atoms_test.copy()
refqmmol = QMMol(atoms = atoms, method = method, basis=basis, xc = xc, charge=charge)
atoms.calc = QMLCalculator(qmmodel = refqmmol, method = 'engine')

  1. Run the geometry optimization

[10]:

LBFGS(atoms, trajectory = 'opt.traj').run(fmax=0.01)
atoms.positions

       Step     Time          Energy         fmax
LBFGS:    0 16:14:37    -2070.596525        1.3259
LBFGS:    1 16:14:40    -2070.616301        0.2563
LBFGS:    2 16:14:43    -2070.617393        0.1126
LBFGS:    3 16:14:46    -2070.618015        0.1030
LBFGS:    4 16:14:50    -2070.618396        0.0495
LBFGS:    5 16:14:53    -2070.618425        0.0106
LBFGS:    6 16:14:57    -2070.618427        0.0005

[10]:

array([[ 1.92577545e-01, -9.34071419e-15,  9.72371798e-02],
       [ 3.71122733e-03,  7.65804799e-01, -4.66034590e-01],
       [ 3.71122733e-03, -7.65804799e-01, -4.66034590e-01]])

Geting Energies from an external model using Delta learning

  1. Open a model in a database and keep it in qmmodel.

  2. Create atoms atoms copy.

  3. Initialize the calculator using QMLCalculator class, based on the define second_learn models.

[11]:

qmmodel = db2qmmodel('./h2o_md_300_QML_set.hdf5')
atoms = atoms_test.copy()
atoms.calc = QMLCalculator(qmmodel = qmmodel, second_learn = second_learn, method = 'gamma')

Guess DB names : {'qmmol': 'rks/qmmol', 'atoms': 'rks/train_atoms_30', 'properties': 'rks/train_props_30'}
Guess mmodels: {'gamma': KernelRidge(alpha=0.1), 'd_gamma': LinearRegression(), 'd_energy': LinearRegression(), 'd_forces': LinearRegression()}

  1. Run the geometry optimization

[12]:

LBFGS(atoms, trajectory = 'opt.traj').run(fmax=0.01)
atoms.positions

       Step     Time          Energy         fmax
LBFGS:    0 16:15:00    -2070.596523        1.3260
LBFGS:    1 16:15:00    -2070.616322        0.2564
LBFGS:    2 16:15:01    -2070.617412        0.1126
LBFGS:    3 16:15:01    -2070.618032        0.1029
LBFGS:    4 16:15:01    -2070.618411        0.0495
LBFGS:    5 16:15:01    -2070.618441        0.0106
LBFGS:    6 16:15:01    -2070.618441        0.0014

[12]:

array([[ 1.92610265e-01,  1.14244173e-05,  9.72291730e-02],
       [ 3.69633877e-03,  7.65794139e-01, -4.66026200e-01],
       [ 3.69338994e-03, -7.65805574e-01, -4.66034992e-01]])