import numpy as np from abc import ABC, abstractmethod from ase.data import covalent_radii from ase.geometry import get_distances from ase import Atoms from ase.visualize import view from utils import check_valid_bondlengths class CandidateGenerator(ABC): """Baseclass for mutation and crossover operations as well as the startgenerator. Parameters: blmin: The minimum allowed distance between atoms in units of the covalent distance between atoms, where d_cov=r_cov_i+r_cov_j. blmax: The maximum allowed distance, in units of the covalent distance, from a single isolated atom to the closest atom. If blmax=None, no constraint is enforced on isolated atoms. force_all_bonds_valid: If True all bondlengths are forced to be valid according to blmin/blmax. If False, only bondlengths of atoms specified in bondlength checks during operations are tested. The specified atoms are typically the ones changed during operations. Default is False, as True might cause problems with GOFEE, as GPR-relaxations and dual-steps might result in structures that does not obey blmin/blmax. """ def __init__(self, blmin=0.7, blmax=1.4, force_all_bonds_valid=False): self.blmin = blmin self.blmax = blmax self.force_all_bonds_valid = force_all_bonds_valid self.description = 'Unspecified' def check_valid_bondlengths(self, a, indices=None, check_too_close=True, check_isolated=True): if self.force_all_bonds_valid: # Check all bonds (mainly for testing) return check_valid_bondlengths(a, self.blmin, self.blmax+0.1, check_too_close=check_too_close, check_isolated=check_isolated) else: # Check only specified ones # (typically only for the atoms changed during operation) return check_valid_bondlengths(a, self.blmin, self.blmax+0.1, indices=indices, check_too_close=check_too_close, check_isolated=check_isolated) @abstractmethod def get_new_candidate(self): pass def finalize(self, a, a0=None, successfull=True): if successfull: description = self.description else: description = 'failed ' + self.description try: a.info['key_value_pairs']['origin'] = description except: a.info['key_value_pairs'] = {'origin': description} if self.force_all_bonds_valid: # Check all bonds valid_bondlengths = self.check_valid_bondlengths(a) assert valid_bondlengths, 'bondlengths are not valid' return a class OperationSelector(): """Class to produce new candidates by applying one of the candidate generation operations which is supplied in the 'operations'-list. The operations are drawn randomly according to the 'probabilities'-list. operations : "list" or "list of lists" of mutations/crossovers. probabilities : probability for each of the mutations/crossovers in operations. Must have the same dimensions as operations. """ def __init__(self, probabilities, operations): cond1 = isinstance(operations[0], list) cond2 = isinstance(probabilities[0], list) if not cond1 and not cond2: operations = [operations] probabilities = [probabilities] element_count_operations = [len(op_list) for op_list in operations] element_count_probabilities = [len(prob_list) for prob_list in probabilities] assert element_count_operations == element_count_probabilities, 'the two lists must have the same shape' self.operations = operations self.rho = [np.cumsum(prob_list) for prob_list in probabilities] def __get_index__(self, rho): """Draw from the cumulative probalility distribution, rho, to return the index of which operation to use""" v = np.random.random() * rho[-1] for i in range(len(rho)): if rho[i] > v: return i def get_new_candidate(self, parents): """Generate new candidate by applying a randomly drawn operation on the structures. This is done successively for each list of operations, if multiple are present. """ for op_list, rho_list in zip(self.operations, self.rho): to_use = self.__get_index__(rho_list) anew = op_list[to_use].get_new_candidate(parents) parents[0] = anew return anew def random_pos(box): """ Returns a random position within the box described by the input box. """ p0 = box[0].astype(float) vspan = box[1] r = np.random.random((1, len(vspan))) pos = p0.copy() for i in range(len(vspan)): pos += vspan[i] * r[0, i] return pos class StartGenerator(CandidateGenerator): """ Class used to generate random initial candidates. The candidates are generated by iteratively adding in one atom at a time within the box described. Parameters: slab: The atoms object describing the super cell to optimize within. Can be an empty cell or a cell containing the atoms of a slab. stoichiometry: A list of atomic numbers for the atoms that are placed on top of the slab (if one is present). box_to_place_in: The box within which atoms are placed. The box should be on the form [p0, vspan] where 'p0' is the position of the box corner and 'vspan' is a matrix containing the three spanning vectors. blmin: The minimum allowed distance between atoms in units of the covalent distance between atoms, where d_cov=r_cov_i+r_cov_j. blmax: The maximum allowed distance, in units of the covalent distance, from a single isolated atom to the closest atom. If blmax=None, no constraint is enforced on isolated atoms. cluster: If True atoms are required to be placed within blmin*d_cov of one of the other atoms to be placed. If False the atoms in the slab are also included. """ def __init__(self, slab, stoichiometry, box_to_place_in, blmin=0.7, blmax=1.4, cluster=False, description='StartGenerator'): CandidateGenerator.__init__(self, blmin=blmin, blmax=blmax) self.slab = slab self.stoichiometry = stoichiometry self.box = box_to_place_in self.cluster = cluster self.description = description def get_new_candidate(self, parents=None): a = self.make_structure() a = self.finalize(a) return a def make_structure(self): """ Generates a new random structure """ Nslab = len(self.slab) Ntop = len(self.stoichiometry) num = np.random.permutation(self.stoichiometry) for i_trials in range(1000): a = self.slab.copy() for i in range(Ntop): pos_found = False for _ in range(300): # Place new atom posi = random_pos(self.box) a += Atoms([num[i]], posi.reshape(1,3)) # Check if position of new atom is valid not_too_close = self.check_valid_bondlengths(a, indices=[Nslab+i], check_too_close=True, check_isolated=False) if len(a) == 1: # The first atom not_isolated = True else: if self.cluster: # Check isolation excluding slab atoms. not_isolated = self.check_valid_bondlengths(a[Nslab:], indices=[Nslab+i], check_too_close=False, check_isolated=True) else: # All atoms. not_isolated = self.check_valid_bondlengths(a, indices=[Nslab+i], check_too_close=False, check_isolated=True) valid_bondlengths = not_too_close and not_isolated if not valid_bondlengths: del a[-1] else: pos_found = True break if not pos_found: break if pos_found: break if i_trials == 999 and not pos_found: raise RuntimeError('StartGenerator: No valid structure was produced in 1000 trials.') else: return a if __name__ == '__main__': from ase.io import read from ase.visualize import view from candidate_operations.basic_mutations import RattleMutation, RattleMutation2, PermutationMutation print(0.7*2*covalent_radii[1], 1.3*2*covalent_radii[1]) np.random.seed(7) #a = read('/home/mkb/DFT/gpLEA/Si3x3/ref/gm_unrelaxed_done.traj', index='0') #a = read('si3x3.traj', index='0') #a = read('c6h6.traj', index='0') traj = read('c6h6_init.traj', index=':') #a = read('sn2o3.traj', index='0') #slab = read('slab_sn2o3.traj', index='0') """ stoichiometry = 6*[50] + 10*[8] c = slab.get_cell() c[2,2] = 3.3 p0 = np.array([0,0,14]) box = [p0, c] sg = StartGenerator(slab, stoichiometry, box) """ a = traj[0] rattle = RattleMutation(n_top=len(a), Nrattle=3, rattle_range=2) rattle2 = RattleMutation2(n_top=16, Nrattle=0.1) permut = PermutationMutation(n_top=16, Npermute=2) candidategenerator = OperationSelector([1], [rattle]) #candidategenerator = CandidateGenerator([0., 1., 0.], [rattle, rattle2, permut]) #candidategenerator = CandidateGenerator([[1],[1]], [[rattle2], [permut]]) """ for a in traj: vb = rattle.check_valid_bondlengths(a) print(vb) """ traj_rattle = [] for i in range(100): for j, a in enumerate(traj[13:14]): print('i =', i, 'j =', j) a0 = a.copy() anew = candidategenerator.get_new_candidate([a0,a0]) traj_rattle += [a0, anew] view(traj_rattle) """ a_mut = rattle.get_new_candidate([a]) view([a,a_mut]) """