Source code for hiphive.cluster_space

"""
Contains the ClusterSpace object central to hiPhive
"""

import numpy as np
import tarfile

from ase.data import chemical_symbols
from collections import OrderedDict
from copy import deepcopy

from .core.cluster_space_builder import build_cluster_space
from .core.atoms import Atoms
from .core.orbit import Orbit
from .io.logging import logger
from .io.read_write_files import (add_items_to_tarfile_pickle,
                                  add_items_to_tarfile_custom,
                                  add_list_to_tarfile_custom,
                                  read_items_pickle,
                                  read_list_custom)
from .cluster_filter import Cutoffs, CutoffMaximumBody

logger = logger.getChild('ClusterSpace')


[docs]class ClusterSpace: # TODO: Inherit from base class to ease the use of FCP # TODO: Is this type of docstring correct? # TODO: Fix cutoffs parameter when new Cutoffs objects is implemented """Primitive object handling cluster and force constants for a structure. Parameters ---------- prototype_struture : ASE Atoms object prototype structure; spglib will be used to find a suitable cell based on this structure. cutoffs : list or Cutoffs obj cutoff radii for different orders starting with second order sum_rules : bool If True the aucostic sum rules will be enforced by constraining the parameters. symprec : float numerical precision that will be used for analyzing the symmetry (this parameter will be forwarded to `spglib <https://atztogo.github.io/spglib/>`_) length_scale : float This will be used as a normalization constant for the eigentensors Examples -------- To instantiate a ClusterSpace object one has to specify a prototype structure and cutoff radii for each cluster order that should be included. For example the following snippet will set up a ClusterSpace object for a BCC structure including second order terms up to a distance of 5 A and third order terms up to a distance of 4 A. >>> from ase.build import bulk >>> prim = bulk('W') >>> cs = ClusterSpace(prim, [5.0, 4.0]) """ # TODO: This class probably need some more documentation and a reference to # the thesis # TODO: Fix doc for n-body cutoff def __init__(self, prototype_structure, cutoffs, sum_rules=True, symprec=1e-5, length_scale=0.1): if not all(prototype_structure.pbc): raise ValueError('prototype_structure must have pbc.') if isinstance(cutoffs, Cutoffs): self._cutoffs = cutoffs if isinstance(cutoffs, list): self._cutoffs = CutoffMaximumBody(cutoffs, len(cutoffs) + 1) self._symprec = symprec self._length_scale = length_scale self._sum_rules = sum_rules self._atom_list = None self._cluster_list = None self._symmetry_dataset = None self._permutations = None self._prim = None self._orbits = None self._constraint_vectors = None # TODO: How to handle the constraint matrices? Should they even be # stored? self._constraint_matrices = None # Is this the best way or should the prim be instantiated separately? build_cluster_space(self, prototype_structure) # TODO: Should everything here be properties? deepcopy/ref etc.? # TODO: Docstrings for properties @property def number_of_dofs(self): """int : number of free parameters in the model If the sum rules are not enforced the number of DOFs is the same as the total number of eigentensors in all orbits. """ return self._get_number_of_dofs() @property def cutoffs(self): """ Cutoffs object : cutoffs used for constructing the cluster space """ return deepcopy(self._cutoffs) @property def symprec(self): """ float : symprec value used when constructing the cluster space """ return self._symprec @property def sum_rules(self): """ bool : True if sum rules are enforced """ return self._sum_rules @property def length_scale(self): """ float : normalization constant of the force constants """ return self._length_scale @property def primitive_structure(self): """ ASE Atoms object : structure of the lattice """ return self._prim.copy() @property def spacegroup(self): """ str : space group of the lattice structure obtained from spglib """ return self._symmetry_dataset['international'] + ' ({})'.format( self._symmetry_dataset['number']) @property def wyckoff_sites(self): """ list : wyckoff sites in the primitive cell""" return self._symmetry_dataset['equivalent_atoms'] @property def rotation_matrices(self): """ list of 3x3 matrices : symmetry elements representing rotations """ return self._symmetry_dataset['rotations'].copy() @property def translation_vectors(self): """ list of 3-vectors : symmetry elements representin translations """ # TODO: bug incoming! return (self._symmetry_dataset['translations'] % 1).copy() @property def permutations(self): """ list of vectors : lookup for permutation references """ return deepcopy(self._permutations) @property def atom_list(self): """ BiMap object : atoms inside the cutoff relative to the of the center cell """ return self._atom_list @property def cluster_list(self): """ BiMap object : clusters possible within the cutoff """ return self._cluster_list @property def orbits(self): # TODO: add __getitem__ method """ list of Orbit objects : orbits associated with the lattice structure. """ return self._orbits @property def orbit_data(self): """ list : list of dictionaries containing detailed information for each orbit, e.g., cluster radius and atom types. """ data = [] p = 0 for orbit_index, orbit in enumerate(self.orbits): d = {} d['index'] = orbit_index d['order'] = orbit.order d['radius'] = orbit.radius d['maximum_distance'] = orbit.maximum_distance d['number_of_clusters'] = len(orbit.orientation_families) d['eigentensors'] = orbit.eigentensors d['number_of_parameters'] = len(d['eigentensors']) types, wyckoff_sites = [], [] for atom_index in self.cluster_list[orbit.prototype_index]: atom = self.atom_list[atom_index] types.append(self.primitive_structure.numbers[atom.site]) wyckoff_sites.append(self.wyckoff_sites[atom.site]) d['prototype_cluster'] = self.cluster_list[orbit.prototype_index] d['prototype_atom_types'] = tuple(types) d['prototype_wyckoff_sites'] = tuple(wyckoff_sites) d['geometrical_order'] = len(set(d['prototype_cluster'])) d['parameter_indices'] = np.arange(p, p + len(orbit.eigentensors)) p += len(orbit.eigentensors) data.append(d) return data
[docs] def print_orbits(self): """ Prints a list of all orbits. """ orbits = self.orbit_data def str_orbit(index, orbit): elements = ' '.join(chemical_symbols[n] for n in orbit['prototype_atom_types']) fields = OrderedDict([ ('index', '{:^3}'.format(index)), ('order', '{:^3}'.format(orbit['order'])), ('elements', '{:^18}'.format(elements)), ('radius', '{:^8.4f}'.format(orbit['radius'])), ('prototype', '{:^18}' .format(str(orbit['prototype_cluster']))), ('clusters', '{:^4}'.format(orbit['number_of_clusters'])), ('parameters', '{:^3}'.format(len(orbit['eigentensors']))), ]) s = [] for name, value in fields.items(): n = max(len(name), len(value)) if index < 0: s += ['{s:^{n}}'.format(s=name, n=n)] else: s += ['{s:^{n}}'.format(s=value, n=n)] return ' | '.join(s) # table header width = max(len(str_orbit(-1, orbits[-1])), len(str_orbit(0, orbits[-1]))) print(' List of Orbits '.center(width, '=')) print(str_orbit(-1, orbits[0])) print(''.center(width, '-')) # table body for i, orbit in enumerate(orbits): print(str_orbit(i, orbit)) print(''.center(width, '='))
def _get_number_of_dofs(self): """ Returns the number of degrees of freedom. """ return self._cvs.shape[1] def __str__(self): def str_order(order, header=False): formats = {'order': '{:2}', 'num_orbits': '{:5}', 'num_params': '{:5}', 'num_clusters': '{:5}'} s = [] for name, value in order.items(): str_repr = formats[name].format(value) n = max(len(name), len(str_repr)) if header: s += ['{s:^{n}}'.format(s=name, n=n)] else: s += ['{s:^{n}}'.format(s=str_repr, n=n)] return ' | '.join(s) # collect data orbits = self.orbit_data orders = OrderedDict( (order, OrderedDict(order=order)) for order in self.cutoffs.orders) for orbit in orbits: k = orbit['order'] orders[k]['num_orbits'] = orders[k].get('num_orbits', 0) + 1 orders[k]['num_clusters'] = orders[k].get('num_clusters', 0) + \ orbit['number_of_clusters'] # basic information prototype = orders[list(orders.keys())[-1]] n = max(len(str_order(prototype)), 54) s = [] s.append(' Cluster Space '.center(n, '=')) data = [('Spacegroup', self.spacegroup), ('symprec', self.symprec), ('Sum rules', self.sum_rules), ('Length scale', self.length_scale), ('Cutoffs', self.cutoffs), ('Cell', self.primitive_structure.cell), ('Basis', self.primitive_structure.basis), ('Numbers', self.primitive_structure.numbers), ('Total number of orbits', len(orbits)), ('Total number of clusters', sum([order['num_clusters'] for order in orders.values()])), ('Total number of parameters', self._get_number_of_dofs() )] for field, value in data: if str(value).count('\n') > 1: s.append('{:26} :\n{}'.format(field, value)) else: s.append('{:26} : {}'.format(field, value)) # table header s.append(''.center(n, '-')) s.append(str_order(prototype, header=True)) s.append(''.center(n, '-')) for order in orders.values(): s.append(str_order(order).rstrip()) s.append(''.center(n, '=')) return '\n'.join(s) def __repr__(self): s = 'ClusterSpace({!r}, {!r}, {!r}, {!r}, {!r})' return s.format(self.primitive_structure, self.cutoffs, self.sum_rules, self.symprec, self.length_scale) def _map_parameters(self, parameters): """ Maps irreducible parameters to the real parameters associated with the eigentensors. """ txt = ('Please provide {} parameters' .format(self.number_of_dofs)) assert len(parameters) == self.number_of_dofs, txt return self._cvs.dot(parameters)
[docs] def copy(self): return deepcopy(self)
[docs] def write(self, fileobj): """ Saves to file or to a file-like object. The instance is saved into a custom format based on tar-files. The resulting file will be a valid tar file and can be browsed by by a tar reader. The included objects are themself either pickles, npz or other tars. Parameters ---------- fileobj : str or file-like object If the input is a string a tar archive will be created in the current directory. Otherwise the input must be a valid file like object. """ # Create a tar archive if isinstance(fileobj, str): tar_file = tarfile.open(name=fileobj, mode='w') else: tar_file = tarfile.open(fileobj=fileobj, mode='w') # Attributes in pickle format pickle_attributes = ['_symprec', '_length_scale', '_sum_rules', '_symmetry_dataset', '_permutations', '_atom_list', '_cluster_list'] items_pickle = dict() for attribute in pickle_attributes: items_pickle[attribute] = self.__getattribute__(attribute) add_items_to_tarfile_pickle(tar_file, items_pickle, 'attributes') # Constraint matrices and vectors in hdf5 format items = dict(cvs=self._cvs) add_items_to_tarfile_pickle(tar_file, items, 'constraint_vectors') # Cutoffs and prim with their builtin write/read functions items_custom = {'_cutoffs': self._cutoffs, '_prim': self._prim} add_items_to_tarfile_custom(tar_file, items_custom) # Orbits add_list_to_tarfile_custom(tar_file, self._orbits, 'orbits') add_list_to_tarfile_custom(tar_file, self._dropped_orbits, 'dropped_orbits') # Done! tar_file.close()
[docs] def read(f): """ Reads a ClusterSpace instance from file. Parameters ---------- f : string or file object name of input file (string) or stream to load from (file object) """ # Instantiate empty cs obj. cs = ClusterSpace.__new__(ClusterSpace) # Load from file on disk or file-like if type(f) is str: tar_file = tarfile.open(mode='r', name=f) else: tar_file = tarfile.open(mode='r', fileobj=f) # Attributes attributes = read_items_pickle(tar_file, 'attributes') for name, value in attributes.items(): cs.__setattr__(name, value) # Load the constraint matrices into their dict items = read_items_pickle(tar_file, 'constraint_vectors') cs._cvs = items['cvs'] # Cutoffs and prim via custom save funcs fileobj = tar_file.extractfile('_cutoffs') cs._cutoffs = Cutoffs.read(fileobj) fileobj = tar_file.extractfile('_prim') cs._prim = Atoms.read(fileobj) # Orbits are stored in a separate archive cs._orbits = read_list_custom(tar_file, 'orbits', Orbit.read) cs._dropped_orbits = read_list_custom( tar_file, 'dropped_orbits', Orbit.read) return cs