def _make_sparse_precon(self, atoms, initial_assembly=False,
force_stab=False):
"""Create a sparse preconditioner matrix based on the passed atoms.
Args:
atoms: the Atoms object used to create the preconditioner.
Returns:
A scipy.sparse.csr_matrix object, representing a d*N by d*N matrix
(where N is the number of atoms, and d is the value of self.dim).
BE AWARE that using numpy.dot() with this object will result in
errors/incorrect results - use the .dot method directly on the
sparse matrix instead.
"""
#print('creating sparse precon: initial_assembly=%r, '
# 'force_stab=%r, apply_positions=%r, apply_cell=%r' %
# (initial_assembly, force_stab, self.apply_positions,
# self.apply_cell))
N = len(atoms)
#start_time = time.time()
if self.apply_positions:
# compute neighbour list
i_list, j_list, rij_list, fixed_atoms = get_neighbours(
atoms, self.r_cut)
#print('--- neighbour list created in %s s ---' %
# (time.time() - start_time))
row = []
col = []
data = []
# precon is mu_c*identity for cell DoF
if isinstance(atoms, Filter):
i = N - 3
j = N - 2
k = N - 1
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(x)
col.extend(x)
if self.apply_cell:
data.extend(np.repeat(self.mu_c, 9))
else:
data.extend(np.repeat(self.mu_c, 9))
#print('--- computed triplet format in %s s ---' %
# (time.time() - start_time))
conn = sparse.lil_matrix((N, N), dtype=bool)
if self.apply_positions and not initial_assembly:
if self.morses is not None:
for n in range(len(self.morses)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_morse_potential_reduced_hessian(
atoms, self.morses[n])
elif self.hessian == 'spectral':
i, j, Hx = ff.get_morse_potential_hessian(
atoms, self.morses[n], spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
conn[i, j] = True
conn[j, i] = True
if self.bonds is not None:
for n in range(len(self.bonds)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_bond_potential_reduced_hessian(
atoms, self.bonds[n], self.morses)
elif self.hessian == 'spectral':
i, j, Hx = ff.get_bond_potential_hessian(
atoms, self.bonds[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
conn[i, j] = True
conn[j, i] = True
if self.angles is not None:
for n in range(len(self.angles)):
if self.hessian == 'reduced':
i, j, k, Hx = ff.get_angle_potential_reduced_hessian(
atoms, self.angles[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, Hx = ff.get_angle_potential_hessian(
atoms, self.angles[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(np.repeat(x, 9))
col.extend(np.tile(x, 9))
data.extend(Hx.flatten())
conn[i, j] = conn[i, k] = conn[j, k] = True
conn[j, i] = conn[k, i] = conn[k, j] = True
if self.dihedrals is not None:
for n in range(len(self.dihedrals)):
if self.hessian == 'reduced':
i, j, k, l, Hx = \
ff.get_dihedral_potential_reduced_hessian(
atoms, self.dihedrals[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, l, Hx = ff.get_dihedral_potential_hessian(
atoms, self.dihedrals[n], self.morses,
spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2,
3 * k, 3 * k + 1, 3 * k + 2,
3 * l, 3 * l + 1, 3 * l + 2]
row.extend(np.repeat(x, 12))
col.extend(np.tile(x, 12))
data.extend(Hx.flatten())
conn[i, j] = conn[i, k] = conn[i, l] = conn[
j, k] = conn[j, l] = conn[k, l] = True
conn[j, i] = conn[k, i] = conn[l, i] = conn[
k, j] = conn[l, j] = conn[l, k] = True
if self.apply_positions:
for i, j, rij in zip(i_list, j_list, rij_list):
if not conn[i, j]:
coeff = self.get_coeff(rij)
x = [3 * i, 3 * i + 1, 3 * i + 2]
y = [3 * j, 3 * j + 1, 3 * j + 2]
row.extend(x + x)
col.extend(x + y)
data.extend(3 * [-coeff] + 3 * [coeff])
row.extend(range(self.dim * N))
col.extend(range(self.dim * N))
if initial_assembly:
data.extend([self.mu * self.c_stab] * self.dim * N)
else:
data.extend([self.c_stab] * self.dim * N)
# create the matrix
#start_time = time.time()
self.P = sparse.csc_matrix(
(data, (row, col)), shape=(self.dim * N, self.dim * N))
#print('--- created CSC matrix in %s s ---' %
# (time.time() - start_time))
if not initial_assembly:
if len(fixed_atoms) != 0:
self.P.tolil()
for i in fixed_atoms:
self.P[i, :] = 0.0
self.P[:, i] = 0.0
self.P[i, i] = 1.0
self.P = self.P.tocsr()
# Create solver
if self.use_pyamg:
#start_time = time.time()
self.ml = smoothed_aggregation_solver(
self.P, B=None,
strength=('symmetric', {'theta': 0.0}),
smooth=(
'jacobi', {'filter': True, 'weighting': 'local'}),
improve_candidates=[('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 4}),
None, None, None, None, None, None, None,
None, None, None, None, None, None, None],
aggregate='standard',
presmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
postsmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
max_levels=15,
max_coarse=300,
coarse_solver='pinv')
#print('--- multi grid solver created in %s s ---' %
# (time.time() - start_time))
return self.P
def _make_sparse_precon(self, atoms, initial_assembly=False,
force_stab=False):
"""Create a sparse preconditioner matrix based on the passed atoms.
Args:
atoms: the Atoms object used to create the preconditioner.
Returns:
A scipy.sparse.csr_matrix object, representing a d*N by d*N matrix
(where N is the number of atoms, and d is the value of self.dim).
BE AWARE that using numpy.dot() with this object will result in
errors/incorrect results - use the .dot method directly on the
sparse matrix instead.
"""
logger.info('creating sparse precon: initial_assembly=%r, '
'force_stab=%r, apply_positions=%r, apply_cell=%r',
initial_assembly, force_stab, self.apply_positions,
self.apply_cell)
N = len(atoms)
start_time = time.time()
if self.apply_positions:
# compute neighbour list
i_list, j_list, rij_list, fixed_atoms = get_neighbours(
atoms, self.r_cut)
logger.info('--- neighbour list created in %s s ---' %
(time.time() - start_time))
row = []
col = []
data = []
# precon is mu_c*identity for cell DoF
if isinstance(atoms, Filter):
i = N - 3
j = N - 2
k = N - 1
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(x)
col.extend(x)
if self.apply_cell:
data.extend(np.repeat(self.mu_c, 9))
else:
data.extend(np.repeat(self.mu_c, 9))
logger.info('--- computed triplet format in %s s ---' %
(time.time() - start_time))
conn = sparse.lil_matrix((N, N), dtype=bool)
if self.apply_positions and not initial_assembly:
if self.morses is not None:
for n in range(len(self.morses)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_morse_potential_reduced_hessian(
atoms, self.morses[n])
elif self.hessian == 'spectral':
i, j, Hx = ff.get_morse_potential_hessian(
atoms, self.morses[n], spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
conn[i, j] = True
conn[j, i] = True
if self.bonds is not None:
for n in range(len(self.bonds)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_bond_potential_reduced_hessian(
atoms, self.bonds[n], self.morses)
elif self.hessian == 'spectral':
i, j, Hx = ff.get_bond_potential_hessian(
atoms, self.bonds[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
conn[i, j] = True
conn[j, i] = True
if self.angles is not None:
for n in range(len(self.angles)):
if self.hessian == 'reduced':
i, j, k, Hx = ff.get_angle_potential_reduced_hessian(
atoms, self.angles[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, Hx = ff.get_angle_potential_hessian(
atoms, self.angles[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(np.repeat(x, 9))
col.extend(np.tile(x, 9))
data.extend(Hx.flatten())
conn[i, j] = conn[i, k] = conn[j, k] = True
conn[j, i] = conn[k, i] = conn[k, j] = True
if self.dihedrals is not None:
for n in range(len(self.dihedrals)):
if self.hessian == 'reduced':
i, j, k, l, Hx = \
ff.get_dihedral_potential_reduced_hessian(
atoms, self.dihedrals[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, l, Hx = ff.get_dihedral_potential_hessian(
atoms, self.dihedrals[n], self.morses,
spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2,
3 * j, 3 * j + 1, 3 * j + 2,
3 * k, 3 * k + 1, 3 * k + 2,
3 * l, 3 * l + 1, 3 * l + 2]
row.extend(np.repeat(x, 12))
col.extend(np.tile(x, 12))
data.extend(Hx.flatten())
conn[i, j] = conn[i, k] = conn[i, l] = conn[
j, k] = conn[j, l] = conn[k, l] = True
conn[j, i] = conn[k, i] = conn[l, i] = conn[
k, j] = conn[l, j] = conn[l, k] = True
if self.apply_positions:
for i, j, rij in zip(i_list, j_list, rij_list):
if not conn[i, j]:
coeff = self.get_coeff(rij)
x = [3 * i, 3 * i + 1, 3 * i + 2]
y = [3 * j, 3 * j + 1, 3 * j + 2]
row.extend(x + x)
col.extend(x + y)
data.extend(3 * [-coeff] + 3 * [coeff])
row.extend(range(self.dim * N))
col.extend(range(self.dim * N))
if initial_assembly:
data.extend([self.mu * self.c_stab] * self.dim * N)
else:
data.extend([self.c_stab] * self.dim * N)
# create the matrix
start_time = time.time()
self.P = sparse.csc_matrix(
(data, (row, col)), shape=(self.dim * N, self.dim * N))
logger.info('--- created CSC matrix in %s s ---' %
(time.time() - start_time))
if not initial_assembly:
if len(fixed_atoms) != 0:
self.P.tolil()
for i in fixed_atoms:
self.P[i, :] = 0.0
self.P[:, i] = 0.0
self.P[i, i] = 1.0
self.P = self.P.tocsr()
# Create solver
if self.use_pyamg and have_pyamg:
start_time = time.time()
self.ml = smoothed_aggregation_solver(
self.P, B=None,
strength=('symmetric', {'theta': 0.0}),
smooth=(
'jacobi', {'filter': True, 'weighting': 'local'}),
improve_candidates=[('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 4}),
None, None, None, None, None, None, None,
None, None, None, None, None, None, None],
aggregate='standard',
presmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
postsmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
max_levels=15,
max_coarse=300,
coarse_solver='pinv')
logger.info('--- multi grid solver created in %s s ---' %
(time.time() - start_time))
return self.P
def _make_sparse_precon(self, atoms, initial_assembly=False,
force_stab=False):
""" """
#start_time = time.time()
N = len(atoms)
row = []
col = []
data = []
if self.morses is not None:
for n in range(len(self.morses)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_morse_potential_reduced_hessian(
atoms, self.morses[n])
elif self.hessian == 'spectral':
i, j, Hx = ff.get_morse_potential_hessian(
atoms, self.morses[n], spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
if self.bonds is not None:
for n in range(len(self.bonds)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_bond_potential_reduced_hessian(
atoms, self.bonds[n], self.morses)
elif self.hessian == 'spectral':
i, j, Hx = ff.get_bond_potential_hessian(
atoms, self.bonds[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
if self.angles is not None:
for n in range(len(self.angles)):
if self.hessian == 'reduced':
i, j, k, Hx = ff.get_angle_potential_reduced_hessian(
atoms, self.angles[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, Hx = ff.get_angle_potential_hessian(
atoms, self.angles[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(np.repeat(x, 9))
col.extend(np.tile(x, 9))
data.extend(Hx.flatten())
if self.dihedrals is not None:
for n in range(len(self.dihedrals)):
if self.hessian == 'reduced':
i, j, k, l, Hx = \
ff.get_dihedral_potential_reduced_hessian(
atoms, self.dihedrals[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, l, Hx = ff.get_dihedral_potential_hessian(
atoms, self.dihedrals[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j +
2, 3 * k, 3 * k + 1, 3 * k + 2, 3 * l, 3 * l + 1,
3 * l + 2]
row.extend(np.repeat(x, 12))
col.extend(np.tile(x, 12))
data.extend(Hx.flatten())
row.extend(range(self.dim * N))
col.extend(range(self.dim * N))
data.extend([self.c_stab] * self.dim * N)
# create the matrix
#start_time = time.time()
self.P = sparse.csc_matrix(
(data, (row, col)), shape=(self.dim * N, self.dim * N))
#print('--- created CSC matrix in %s s ---' %
# (time.time() - start_time))
fixed_atoms = []
for constraint in atoms.constraints:
if isinstance(constraint, FixAtoms):
fixed_atoms.extend(list(constraint.index))
else:
raise TypeError(
'only FixAtoms constraints are supported by Precon class')
if len(fixed_atoms) != 0:
self.P.tolil()
for i in fixed_atoms:
self.P[i, :] = 0.0
self.P[:, i] = 0.0
self.P[i, i] = 1.0
self.P = self.P.tocsr()
#print('--- N-dim precon created in %s s ---' %
# (time.time() - start_time))
# Create solver
if self.use_pyamg:
#start_time = time.time()
self.ml = smoothed_aggregation_solver(
self.P, B=None,
strength=('symmetric', {'theta': 0.0}),
smooth=(
'jacobi', {'filter': True, 'weighting': 'local'}),
improve_candidates=[('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 4}),
None, None, None, None, None, None, None,
None, None, None, None, None, None, None],
aggregate='standard',
presmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
postsmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
max_levels=15,
max_coarse=300,
coarse_solver='pinv')
#print('--- multi grid solver created in %s s ---' %
# (time.time() - start_time))
return self.P
def _make_sparse_precon(self, atoms, initial_assembly=False,
force_stab=False):
""" """
start_time = time.time()
N = len(atoms)
row = []
col = []
data = []
if self.morses is not None:
for n in range(len(self.morses)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_morse_potential_reduced_hessian(
atoms, self.morses[n])
elif self.hessian == 'spectral':
i, j, Hx = ff.get_morse_potential_hessian(
atoms, self.morses[n], spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
if self.bonds is not None:
for n in range(len(self.bonds)):
if self.hessian == 'reduced':
i, j, Hx = ff.get_bond_potential_reduced_hessian(
atoms, self.bonds[n], self.morses)
elif self.hessian == 'spectral':
i, j, Hx = ff.get_bond_potential_hessian(
atoms, self.bonds[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j + 2]
row.extend(np.repeat(x, 6))
col.extend(np.tile(x, 6))
data.extend(Hx.flatten())
if self.angles is not None:
for n in range(len(self.angles)):
if self.hessian == 'reduced':
i, j, k, Hx = ff.get_angle_potential_reduced_hessian(
atoms, self.angles[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, Hx = ff.get_angle_potential_hessian(
atoms, self.angles[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 *
j + 1, 3 * j + 2, 3 * k, 3 * k + 1, 3 * k + 2]
row.extend(np.repeat(x, 9))
col.extend(np.tile(x, 9))
data.extend(Hx.flatten())
if self.dihedrals is not None:
for n in range(len(self.dihedrals)):
if self.hessian == 'reduced':
i, j, k, l, Hx = \
ff.get_dihedral_potential_reduced_hessian(
atoms, self.dihedrals[n], self.morses)
elif self.hessian == 'spectral':
i, j, k, l, Hx = ff.get_dihedral_potential_hessian(
atoms, self.dihedrals[n], self.morses, spectral=True)
else:
raise NotImplementedError('Not implemented hessian')
x = [3 * i, 3 * i + 1, 3 * i + 2, 3 * j, 3 * j + 1, 3 * j +
2, 3 * k, 3 * k + 1, 3 * k + 2, 3 * l, 3 * l + 1,
3 * l + 2]
row.extend(np.repeat(x, 12))
col.extend(np.tile(x, 12))
data.extend(Hx.flatten())
row.extend(range(self.dim * N))
col.extend(range(self.dim * N))
data.extend([self.c_stab] * self.dim * N)
# create the matrix
start_time = time.time()
self.P = sparse.csc_matrix(
(data, (row, col)), shape=(self.dim * N, self.dim * N))
logger.info('--- created CSC matrix in %s s ---' %
(time.time() - start_time))
fixed_atoms = []
for constraint in atoms.constraints:
if isinstance(constraint, FixAtoms):
fixed_atoms.extend(list(constraint.index))
else:
raise TypeError(
'only FixAtoms constraints are supported by Precon class')
if len(fixed_atoms) != 0:
self.P.tolil()
for i in fixed_atoms:
self.P[i, :] = 0.0
self.P[:, i] = 0.0
self.P[i, i] = 1.0
self.P = self.P.tocsr()
logger.info('--- N-dim precon created in %s s ---' %
(time.time() - start_time))
# Create solver
if self.use_pyamg and have_pyamg:
start_time = time.time()
self.ml = smoothed_aggregation_solver(
self.P, B=None,
strength=('symmetric', {'theta': 0.0}),
smooth=(
'jacobi', {'filter': True, 'weighting': 'local'}),
improve_candidates=[('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 4}),
None, None, None, None, None, None, None,
None, None, None, None, None, None, None],
aggregate='standard',
presmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
postsmoother=('block_gauss_seidel',
{'sweep': 'symmetric', 'iterations': 1}),
max_levels=15,
max_coarse=300,
coarse_solver='pinv')
logger.info('--- multi grid solver created in %s s ---' %
(time.time() - start_time))
return self.P