def fit(self):
"""
Fit atom energies using the provided species energies and
corresponding atomization energies from the reference data.
"""
self._load_refdata()
mols = [
Molecule().from_adjacency_list(self.ref_data[lbl].adjacency_list,
raise_atomtype_exception=False,
raise_charge_exception=False)
for lbl in self.species_energies
]
atom_counts = [
Counter(atom.element.symbol for atom in mol.atoms) for mol in mols
]
elements = sorted({element
for ac in atom_counts for element in ac},
key=lambda s: get_element(s).number)
x = np.array([[ac[element] for element in elements]
for ac in atom_counts]) # Nmols x Nelements
atomization_energies = np.array([
self.ref_data[lbl].reference_data[
self.ref_data_src].atomization_energy.value_si /
constants.E_h / constants.Na for lbl in self.species_energies
])
zpes = np.array([
self.ref_data[lbl].reference_data[self.ref_data_src].zpe.value_si /
constants.E_h / constants.Na for lbl in self.species_energies
])
elec_energies = np.array(list(
self.species_energies.values())) # Should already be in Hartree
y = atomization_energies + elec_energies + zpes
w = np.linalg.solve(x.T @ x, x.T @ y)
self.atom_energies = dict(zip(elements, w))
# Get confidence intervals
n = len(y) # Ndata
k = len(w) # Nparam
ypred = x @ w
sigma2 = np.sum((y - ypred)**2) / (n - k - 1) # MSE
cov = sigma2 * np.linalg.inv(x.T @ x) # covariance matrix
se = np.sqrt(np.diag(cov)) # standard error
alpha = 0.05 # 95% confidence level
tdist = distributions.t.ppf(1 - alpha / 2, n - k - 1) # student-t
ci = tdist * se # confidence interval half-width
self.confidence_intervals = dict(zip(
elements, ci)) # Parameter estimates are w +/- ci
def geo_to_mol(coords: np.ndarray, symbols: Iterable[str] = None, nums: Iterable[int] = None) -> Molecule:
"""
Convert molecular geometry specified by atomic coordinates and
atomic symbols/numbers to RMG molecule.
Use Open Babel because it's better at recognizing long bonds.
"""
if nums is None and symbols is None:
raise ValueError('Must specify nums or symbols')
symbols = [get_element(int(n)).symbol for n in nums] if symbols is None else list(symbols)
xyz = f'{len(symbols)}\n\n'
xyz += '\n'.join(f'{s} {c[0]: .10f} {c[1]: .10f} {c[2]: .10f}' for s, c in zip(symbols, coords))
mol = pybel.readstring('xyz', xyz)
return _pybel_to_rmg(mol)
def _pybel_to_rmg(pybel_mol: pybel.Molecule) -> Molecule:
"""
Convert Pybel molecule to RMG molecule but ignore charge,
multiplicity, and bond orders.
"""
mol = Molecule()
for pybel_atom in pybel_mol:
element = get_element(pybel_atom.atomicnum)
atom = Atom(element=element, coords=np.array(pybel_atom.coords))
mol.vertices.append(atom)
for obbond in pybel.ob.OBMolBondIter(pybel_mol.OBMol):
begin_idx = obbond.GetBeginAtomIdx() - 1 # Open Babel indexes atoms starting at 1
end_idx = obbond.GetEndAtomIdx() - 1
bond = Bond(mol.vertices[begin_idx], mol.vertices[end_idx])
mol.add_bond(bond)
return mol