Source code for ase.calculators.dftb

""" This module defines a FileIOCalculator for DFTB+

http://www.dftbplus.org/
http://www.dftb.org/

Initial development: markus.kaukonen@iki.fi
"""

import os
import numpy as np
from ase.calculators.calculator import (FileIOCalculator, kpts2ndarray,
                                        kpts2sizeandoffsets)
from ase.units import Hartree, Bohr


[docs]class Dftb(FileIOCalculator): if 'DFTB_COMMAND' in os.environ: command = os.environ['DFTB_COMMAND'] + ' > PREFIX.out' else: command = 'dftb+ > PREFIX.out' implemented_properties = ['energy', 'forces', 'charges', 'stress', 'dipole'] discard_results_on_any_change = True def __init__(self, restart=None, ignore_bad_restart_file=FileIOCalculator._deprecated, label='dftb', atoms=None, kpts=None, slako_dir=None, **kwargs): """ All keywords for the dftb_in.hsd input file (see the DFTB+ manual) can be set by ASE. Consider the following input file block: >>> Hamiltonian = DFTB { >>> SCC = Yes >>> SCCTolerance = 1e-8 >>> MaxAngularMomentum = { >>> H = s >>> O = p >>> } >>> } This can be generated by the DFTB+ calculator by using the following settings: >>> calc = Dftb(Hamiltonian_='DFTB', # line is included by default >>> Hamiltonian_SCC='Yes', >>> Hamiltonian_SCCTolerance=1e-8, >>> Hamiltonian_MaxAngularMomentum_='', >>> Hamiltonian_MaxAngularMomentum_H='s', >>> Hamiltonian_MaxAngularMomentum_O='p') In addition to keywords specific to DFTB+, also the following keywords arguments can be used: restart: str Prefix for restart file. May contain a directory. Default is None: don't restart. ignore_bad_restart_file: bool Ignore broken or missing restart file. By default, it is an error if the restart file is missing or broken. label: str (default 'dftb') Prefix used for the main output file (<label>.out). atoms: Atoms object (default None) Optional Atoms object to which the calculator will be attached. When restarting, atoms will get its positions and unit-cell updated from file. kpts: (default None) Brillouin zone sampling: * ``(1,1,1)`` or ``None``: Gamma-point only * ``(n1,n2,n3)``: Monkhorst-Pack grid * ``dict``: Interpreted as a path in the Brillouin zone if it contains the 'path_' keyword. Otherwise it is converted into a Monkhorst-Pack grid using ``ase.calculators.calculator.kpts2sizeandoffsets`` * ``[(k11,k12,k13),(k21,k22,k23),...]``: Explicit (Nkpts x 3) array of k-points in units of the reciprocal lattice vectors (each with equal weight) Additional attribute to be set by the embed() method: pcpot: PointCharge object An external point charge potential (for QM/MM calculations) """ if slako_dir is None: slako_dir = os.environ.get('DFTB_PREFIX', './') if not slako_dir.endswith('/'): slako_dir += '/' self.slako_dir = slako_dir self.default_parameters = dict( Hamiltonian_='DFTB', Hamiltonian_SlaterKosterFiles_='Type2FileNames', Hamiltonian_SlaterKosterFiles_Prefix=self.slako_dir, Hamiltonian_SlaterKosterFiles_Separator='"-"', Hamiltonian_SlaterKosterFiles_Suffix='".skf"', Hamiltonian_MaxAngularMomentum_='', Options_='', Options_WriteResultsTag='Yes') self.pcpot = None self.lines = None self.atoms = None self.atoms_input = None self.do_forces = False self.outfilename = 'dftb.out' FileIOCalculator.__init__(self, restart, ignore_bad_restart_file, label, atoms, **kwargs) # Determine number of spin channels try: entry = kwargs['Hamiltonian_SpinPolarisation'] spinpol = 'colinear' in entry.lower() except KeyError: spinpol = False self.nspin = 2 if spinpol else 1 # kpoint stuff by ase self.kpts = kpts self.kpts_coord = None if self.kpts is not None: initkey = 'Hamiltonian_KPointsAndWeights' mp_mesh = None offsets = None if isinstance(self.kpts, dict): if 'path' in self.kpts: # kpts is path in Brillouin zone self.parameters[initkey + '_'] = 'Klines ' self.kpts_coord = kpts2ndarray(self.kpts, atoms=atoms) else: # kpts is (implicit) definition of # Monkhorst-Pack grid self.parameters[initkey + '_'] = 'SupercellFolding ' mp_mesh, offsets = kpts2sizeandoffsets(atoms=atoms, **self.kpts) elif np.array(self.kpts).ndim == 1: # kpts is Monkhorst-Pack grid self.parameters[initkey + '_'] = 'SupercellFolding ' mp_mesh = self.kpts offsets = [0.] * 3 elif np.array(self.kpts).ndim == 2: # kpts is (N x 3) list/array of k-point coordinates # each will be given equal weight self.parameters[initkey + '_'] = '' self.kpts_coord = np.array(self.kpts) else: raise ValueError('Illegal kpts definition:' + str(self.kpts)) if mp_mesh is not None: eps = 1e-10 for i in range(3): key = initkey + '_empty%03d' % i val = [mp_mesh[i] if j == i else 0 for j in range(3)] self.parameters[key] = ' '.join(map(str, val)) offsets[i] *= mp_mesh[i] assert abs(offsets[i]) < eps or abs(offsets[i] - 0.5) < eps # DFTB+ uses a different offset convention, where # the k-point mesh is already Gamma-centered prior # to the addition of any offsets if mp_mesh[i] % 2 == 0: offsets[i] += 0.5 key = initkey + '_empty%03d' % 3 self.parameters[key] = ' '.join(map(str, offsets)) elif self.kpts_coord is not None: for i, c in enumerate(self.kpts_coord): key = initkey + '_empty%09d' % i c_str = ' '.join(map(str, c)) if 'Klines' in self.parameters[initkey + '_']: c_str = '1 ' + c_str else: c_str += ' 1.0' self.parameters[key] = c_str def write_dftb_in(self, outfile): """ Write the innput file for the dftb+ calculation. Geometry is taken always from the file 'geo_end.gen'. """ outfile.write('Geometry = GenFormat { \n') outfile.write(' <<< "geo_end.gen" \n') outfile.write('} \n') outfile.write(' \n') params = self.parameters.copy() s = 'Hamiltonian_MaxAngularMomentum_' for key in params: if key.startswith(s) and len(key) > len(s): break else: # User didn't specify max angular mometa. Get them from # the .skf files: symbols = set(self.atoms.get_chemical_symbols()) for symbol in symbols: path = os.path.join(self.slako_dir, '{0}-{0}.skf'.format(symbol)) l = read_max_angular_momentum(path) params[s + symbol] = '"{}"'.format('spdf'[l]) # --------MAIN KEYWORDS------- previous_key = 'dummy_' myspace = ' ' for key, value in sorted(params.items()): current_depth = key.rstrip('_').count('_') previous_depth = previous_key.rstrip('_').count('_') for my_backsclash in reversed( range(previous_depth - current_depth)): outfile.write(3 * (1 + my_backsclash) * myspace + '} \n') outfile.write(3 * current_depth * myspace) if key.endswith('_') and len(value) > 0: outfile.write(key.rstrip('_').rsplit('_')[-1] + ' = ' + str(value) + '{ \n') elif (key.endswith('_') and (len(value) == 0) and current_depth == 0): # E.g. 'Options {' outfile.write(key.rstrip('_').rsplit('_')[-1] + ' ' + str(value) + '{ \n') elif (key.endswith('_') and (len(value) == 0) and current_depth > 0): # E.g. 'Hamiltonian_Max... = {' outfile.write(key.rstrip('_').rsplit('_')[-1] + ' = ' + str(value) + '{ \n') elif key.count('_empty') == 1: outfile.write(str(value) + ' \n') elif ((key == 'Hamiltonian_ReadInitialCharges') and (str(value).upper() == 'YES')): f1 = os.path.isfile(self.directory + os.sep + 'charges.dat') f2 = os.path.isfile(self.directory + os.sep + 'charges.bin') if not (f1 or f2): print('charges.dat or .bin not found, switching off guess') value = 'No' outfile.write(key.rsplit('_')[-1] + ' = ' + str(value) + ' \n') else: outfile.write(key.rsplit('_')[-1] + ' = ' + str(value) + ' \n') if self.pcpot is not None and ('DFTB' in str(value)): outfile.write(' ElectricField = { \n') outfile.write(' PointCharges = { \n') outfile.write( ' CoordsAndCharges [Angstrom] = DirectRead { \n') outfile.write(' Records = ' + str(len(self.pcpot.mmcharges)) + ' \n') outfile.write( ' File = "dftb_external_charges.dat" \n') outfile.write(' } \n') outfile.write(' } \n') outfile.write(' } \n') previous_key = key current_depth = key.rstrip('_').count('_') for my_backsclash in reversed(range(current_depth)): outfile.write(3 * my_backsclash * myspace + '} \n') outfile.write('ParserOptions { \n') outfile.write(' IgnoreUnprocessedNodes = Yes \n') outfile.write('} \n') if self.do_forces: outfile.write('Analysis { \n') outfile.write(' CalculateForces = Yes \n') outfile.write('} \n') def check_state(self, atoms): system_changes = FileIOCalculator.check_state(self, atoms) # Ignore unit cell for molecules: if not atoms.pbc.any() and 'cell' in system_changes: system_changes.remove('cell') if self.pcpot and self.pcpot.mmpositions is not None: system_changes.append('positions') return system_changes def write_input(self, atoms, properties=None, system_changes=None): from ase.io import write if properties is not None: if 'forces' in properties or 'stress' in properties: self.do_forces = True FileIOCalculator.write_input( self, atoms, properties, system_changes) with open(os.path.join(self.directory, 'dftb_in.hsd'), 'w') as fd: self.write_dftb_in(fd) write(os.path.join(self.directory, 'geo_end.gen'), atoms, parallel=False) # self.atoms is none until results are read out, # then it is set to the ones at writing input self.atoms_input = atoms self.atoms = None if self.pcpot: self.pcpot.write_mmcharges('dftb_external_charges.dat') def read_results(self): """ all results are read from results.tag file It will be destroyed after it is read to avoid reading it once again after some runtime error """ with open(os.path.join(self.directory, 'results.tag'), 'r') as fd: self.lines = fd.readlines() self.atoms = self.atoms_input charges, energy, dipole = self.read_charges_energy_dipole() if charges is not None: self.results['charges'] = charges self.results['energy'] = energy if dipole is not None: self.results['dipole'] = dipole if self.do_forces: forces = self.read_forces() self.results['forces'] = forces self.mmpositions = None # stress stuff begins sstring = 'stress' have_stress = False stress = list() for iline, line in enumerate(self.lines): if sstring in line: have_stress = True start = iline + 1 end = start + 3 for i in range(start, end): cell = [float(x) for x in self.lines[i].split()] stress.append(cell) if have_stress: stress = -np.array(stress) * Hartree / Bohr**3 self.results['stress'] = stress.flat[[0, 4, 8, 5, 2, 1]] # stress stuff ends # eigenvalues and fermi levels fermi_levels = self.read_fermi_levels() if fermi_levels is not None: self.results['fermi_levels'] = fermi_levels eigenvalues = self.read_eigenvalues() if eigenvalues is not None: self.results['eigenvalues'] = eigenvalues # calculation was carried out with atoms written in write_input os.remove(os.path.join(self.directory, 'results.tag')) def read_forces(self): """Read Forces from dftb output file (results.tag).""" from ase.units import Hartree, Bohr # Initialise the indices so their scope # reaches outside of the for loop index_force_begin = -1 index_force_end = -1 # Force line indexes for iline, line in enumerate(self.lines): fstring = 'forces ' if line.find(fstring) >= 0: index_force_begin = iline + 1 line1 = line.replace(':', ',') index_force_end = iline + 1 + \ int(line1.split(',')[-1]) break gradients = [] for j in range(index_force_begin, index_force_end): word = self.lines[j].split() gradients.append([float(word[k]) for k in range(0, 3)]) return np.array(gradients) * Hartree / Bohr def read_charges_energy_dipole(self): """Get partial charges on atoms in case we cannot find charges they are set to None """ with open(os.path.join(self.directory, 'detailed.out'), 'r') as fd: lines = fd.readlines() for line in lines: if line.strip().startswith('Total energy:'): energy = float(line.split()[2]) * Hartree break qm_charges = [] for n, line in enumerate(lines): if ('Atom' and 'Charge' in line): chargestart = n + 1 break else: # print('Warning: did not find DFTB-charges') # print('This is ok if flag SCC=No') return None, energy, None lines1 = lines[chargestart:(chargestart + len(self.atoms))] for line in lines1: qm_charges.append(float(line.split()[-1])) dipole = None for line in lines: if 'Dipole moment:' in line and 'au' in line: words = line.split() dipole = np.array( [float(w) for w in words[-4:-1]]) * Bohr return np.array(qm_charges), energy, dipole def get_charges(self, atoms): """ Get the calculated charges this is inhereted to atoms object """ if 'charges' in self.results: return self.results['charges'] else: return None def read_eigenvalues(self): """ Read Eigenvalues from dftb output file (results.tag). Unfortunately, the order seems to be scrambled. """ # Eigenvalue line indexes index_eig_begin = None for iline, line in enumerate(self.lines): fstring = 'eigenvalues ' if line.find(fstring) >= 0: index_eig_begin = iline + 1 line1 = line.replace(':', ',') ncol, nband, nkpt, nspin = map(int, line1.split(',')[-4:]) break else: return None # Take into account that the last row may lack # columns if nkpt * nspin * nband % ncol != 0 nrow = int(np.ceil(nkpt * nspin * nband * 1. / ncol)) index_eig_end = index_eig_begin + nrow ncol_last = len(self.lines[index_eig_end - 1].split()) self.lines[index_eig_end - 1] += ' 0.0 ' * (ncol - ncol_last) eig = np.loadtxt(self.lines[index_eig_begin:index_eig_end]).flatten() eig *= Hartree N = nkpt * nband eigenvalues = [eig[i * N:(i + 1) * N].reshape((nkpt, nband)) for i in range(nspin)] return eigenvalues def read_fermi_levels(self): """ Read Fermi level(s) from dftb output file (results.tag). """ # Fermi level line indexes for iline, line in enumerate(self.lines): fstring = 'fermi_level ' if line.find(fstring) >= 0: index_fermi = iline + 1 break else: return None fermi_levels = [] words = self.lines[index_fermi].split() assert len(words) in [1, 2], 'Expected either 1 or 2 Fermi levels' for word in words: e = float(word) # In non-spin-polarized calculations with DFTB+ v17.1, # two Fermi levels are given, with the second one being 0, # but we don't want to add that one to the list if abs(e) > 1e-8: fermi_levels.append(e) return np.array(fermi_levels) * Hartree def get_ibz_k_points(self): return self.kpts_coord.copy() def get_number_of_spins(self): return self.nspin def get_eigenvalues(self, kpt=0, spin=0): return self.results['eigenvalues'][spin][kpt].copy() def get_fermi_levels(self): return self.results['fermi_levels'].copy() def get_fermi_level(self): return max(self.get_fermi_levels()) def embed(self, mmcharges=None, directory='./'): """Embed atoms in point-charges (mmcharges) """ self.pcpot = PointChargePotential(mmcharges, self.directory) return self.pcpot
class PointChargePotential: def __init__(self, mmcharges, directory='./'): """Point-charge potential for DFTB+. """ self.mmcharges = mmcharges self.directory = directory self.mmpositions = None self.mmforces = None def set_positions(self, mmpositions): self.mmpositions = mmpositions def set_charges(self, mmcharges): self.mmcharges = mmcharges def write_mmcharges(self, filename): """ mok all write external charges as monopoles for dftb+. """ if self.mmcharges is None: print("DFTB: Warning: not writing exernal charges ") return with open(os.path.join(self.directory, filename), 'w') as charge_file: for [pos, charge] in zip(self.mmpositions, self.mmcharges): [x, y, z] = pos charge_file.write('%12.6f %12.6f %12.6f %12.6f \n' % (x, y, z, charge)) def get_forces(self, calc, get_forces=True): """ returns forces on point charges if the flag get_forces=True """ if get_forces: return self.read_forces_on_pointcharges() else: return np.zeros_like(self.mmpositions) def read_forces_on_pointcharges(self): """Read Forces from dftb output file (results.tag).""" from ase.units import Hartree, Bohr with open(os.path.join(self.directory, 'detailed.out'), 'r') as fd: lines = fd.readlines() external_forces = [] for n, line in enumerate(lines): if ('Forces on external charges' in line): chargestart = n + 1 break else: raise RuntimeError( 'Problem in reading forces on MM external-charges') lines1 = lines[chargestart:(chargestart + len(self.mmcharges))] for line in lines1: external_forces.append( [float(i) for i in line.split()]) return np.array(external_forces) * Hartree / Bohr def read_max_angular_momentum(path): """Read maximum angular momentum from .skf file. See dftb.org for A detailed description of the Slater-Koster file format. """ with open(path, 'r') as fd: line = fd.readline() if line[0] == '@': # Extended format fd.readline() l = 3 pos = 9 else: # Simple format: l = 2 pos = 7 # Sometimes there ar commas, sometimes not: line = fd.readline().replace(',', ' ') occs = [float(f) for f in line.split()[pos:pos + l + 1]] for f in occs: if f > 0.0: return l l -= 1