def sp_reshape(A, shape, format='csr'):
"""
Reshapes a sparse matrix.
Parameters
----------
A : sparse_matrix
Input matrix in any format
shape : list/tuple
Desired shape of new matrix
format : string {'csr','coo','csc','lil'}
Optional string indicating desired output format
Returns
-------
B : csr_matrix
Reshaped sparse matrix
References
----------
http://stackoverflow.com/questions/16511879/reshape-sparse-matrix-efficiently-python-scipy-0-12
"""
if not hasattr(shape, '__len__') or len(shape) != 2:
raise ValueError('Shape must be a list of two integers')
C = A.tocoo()
nrows, ncols = C.shape
size = nrows * ncols
new_size = shape[0] * shape[1]
if new_size != size:
raise ValueError('Total size of new array must be unchanged.')
flat_indices = ncols * C.row + C.col
new_row, new_col = divmod(flat_indices, shape[1])
if format == 'csr':
return arr_coo2fast(C.data, new_row, new_col, shape[0], shape[1])
else:
B = sp.coo_matrix((C.data, (new_row, new_col)), shape=shape)
if format == 'coo':
return B
elif format == 'csc':
return B.tocsc()
elif format == 'lil':
return B.tolil()
else:
raise ValueError('Return format not valid.')
def sp_reshape(A, shape, format='csr'):
"""
Reshapes a sparse matrix.
Parameters
----------
A : sparse_matrix
Input matrix in any format
shape : list/tuple
Desired shape of new matrix
format : string {'csr','coo','csc','lil'}
Optional string indicating desired output format
Returns
-------
B : csr_matrix
Reshaped sparse matrix
References
----------
http://stackoverflow.com/questions/16511879/reshape-sparse-matrix-efficiently-python-scipy-0-12
"""
if not hasattr(shape, '__len__') or len(shape) != 2:
raise ValueError('Shape must be a list of two integers')
C = A.tocoo()
nrows, ncols = C.shape
size = nrows * ncols
new_size = shape[0] * shape[1]
if new_size != size:
raise ValueError('Total size of new array must be unchanged.')
flat_indices = ncols * C.row + C.col
new_row, new_col = divmod(flat_indices, shape[1])
if format == 'csr':
return arr_coo2fast(C.data, new_row, new_col, shape[0], shape[1])
else:
B = sp.coo_matrix((C.data, (new_row, new_col)), shape=shape)
if format == 'coo':
return B
elif format == 'csc':
return B.tocsc()
elif format == 'lil':
return B.tolil()
else:
raise ValueError('Return format not valid.')
def bloch_redfield_tensor(H, a_ops, spectra_cb=None, c_ops=[], use_secular=True, sec_cutoff=0.1):
"""
Calculate the Bloch-Redfield tensor for a system given a set of operators
and corresponding spectral functions that describes the system's coupling
to its environment.
.. note::
This tensor generation requires a time-independent Hamiltonian.
Parameters
----------
H : :class:`qutip.qobj`
System Hamiltonian.
a_ops : list of :class:`qutip.qobj`
List of system operators that couple to the environment.
spectra_cb : list of callback functions
List of callback functions that evaluate the noise power spectrum
at a given frequency.
c_ops : list of :class:`qutip.qobj`
List of system collapse operators.
use_secular : bool
Flag (True of False) that indicates if the secular approximation should
be used.
sec_cutoff : float {0.1}
Threshold for secular approximation.
Returns
-------
R, kets: :class:`qutip.Qobj`, list of :class:`qutip.Qobj`
R is the Bloch-Redfield tensor and kets is a list eigenstates of the
Hamiltonian.
"""
if not (spectra_cb is None):
warnings.warn("The use of spectra_cb is depreciated.", DeprecationWarning)
_a_ops = []
for kk, a in enumerate(a_ops):
_a_ops.append([a,spectra_cb[kk]])
a_ops = _a_ops
# Sanity checks for input parameters
if not isinstance(H, Qobj):
raise TypeError("H must be an instance of Qobj")
for a in a_ops:
if not isinstance(a[0], Qobj) or not a[0].isherm:
raise TypeError("Operators in a_ops must be Hermitian Qobj.")
if c_ops is None:
c_ops = []
# use the eigenbasis
evals, ekets = H.eigenstates()
N = len(evals)
K = len(a_ops)
#only Lindblad collapse terms
if K==0:
Heb = qdiags(evals,0,dims=H.dims)
L = liouvillian(Heb, c_ops=[c_op.transform(ekets) for c_op in c_ops])
return L, ekets
A = np.array([a_ops[k][0].transform(ekets).full() for k in range(K)])
Jw = np.zeros((K, N, N), dtype=complex)
# pre-calculate matrix elements and spectral densities
# W[m,n] = real(evals[m] - evals[n])
W = np.real(evals[:,np.newaxis] - evals[np.newaxis,:])
for k in range(K):
# do explicit loops here in case spectra_cb[k] can not deal with array arguments
for n in range(N):
for m in range(N):
Jw[k, n, m] = a_ops[k][1](W[n, m])
dw_min = np.abs(W[W.nonzero()]).min()
# pre-calculate mapping between global index I and system indices a,b
Iabs = np.empty((N*N,3),dtype=int)
for I, Iab in enumerate(Iabs):
# important: use [:] to change array values, instead of creating new variable Iab
Iab[0] = I
Iab[1:] = vec2mat_index(N, I)
# unitary part + dissipation from c_ops (if given):
Heb = qdiags(evals,0,dims=H.dims)
L = liouvillian(Heb, c_ops=[c_op.transform(ekets) for c_op in c_ops])
# dissipative part:
rows = []
cols = []
data = []
for I, a, b in Iabs:
# only check use_secular once per I
if use_secular:
# only loop over those indices J which actually contribute
Jcds = Iabs[np.where(np.abs(W[a, b] - W[Iabs[:,1], Iabs[:,2]]) < dw_min * sec_cutoff)]
else:
Jcds = Iabs
for J, c, d in Jcds:
elem = 0+0j
# summed over k, i.e., each operator coupling the system to the environment
elem += 0.5 * np.sum(A[:, a, c] * A[:, d, b] * (Jw[:, c, a] + Jw[:, d, b]))
if b==d:
# sum_{k,n} A[k, a, n] * A[k, n, c] * Jw[k, c, n])
elem -= 0.5 * np.sum(A[:, a, :] * A[:, :, c] * Jw[:, c, :])
if a==c:
# sum_{k,n} A[k, d, n] * A[k, n, b] * Jw[k, d, n])
elem -= 0.5 * np.sum(A[:, d, :] * A[:, :, b] * Jw[:, d, :])
if elem != 0:
rows.append(I)
cols.append(J)
data.append(elem)
R = arr_coo2fast(np.array(data, dtype=complex),
np.array(rows, dtype=np.int32),
np.array(cols, dtype=np.int32), N**2, N**2)
L.data = L.data + R
return L, ekets
def _permute(Q, order):
Qcoo = Q.data.tocoo()
if Q.isket:
cy_index_permute(Qcoo.row, np.array(Q.dims[0], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [[Q.dims[0][i] for i in order], Q.dims[1]]
elif Q.isbra:
cy_index_permute(Qcoo.col, np.array(Q.dims[1], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [Q.dims[0], [Q.dims[1][i] for i in order]]
elif Q.isoper:
cy_index_permute(Qcoo.row, np.array(Q.dims[0], dtype=np.int32),
np.array(order, dtype=np.int32))
cy_index_permute(Qcoo.col, np.array(Q.dims[1], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [[Q.dims[0][i] for i in order],
[Q.dims[1][i] for i in order]]
elif Q.isoperket:
# For superoperators, we expect order to be something like
# [[0, 2], [1, 3]], which tells us to permute according to
# [0, 2, 1 ,3], and then group indices according to the length
# of each sublist.
# As another example,
# permuting [[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]] by
# [[0, 3], [1, 4], [2, 5]] should give
# [[[1, 1], [2, 2], [3, 3]], [[1, 1], [2, 2], [3, 3]]].
#
# Get the breakout of the left index into dims.
# Since this is a super, the left index itself breaks into left
# and right indices, each of which breaks down further.
# The best way to deal with that here is to flatten dims.
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[0], []), dtype=np.int32)
cy_index_permute(Qcoo.row, q_dims, flat_order)
# Finally, we need to restructure the now-decomposed left index
# into left and right subindices, so that the overall dims we return
# are of the form specified by order.
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [new_dims, [1]]
elif Q.isoperbra:
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[1], []), dtype=np.int32)
cy_index_permute(Qcoo.col, q_dims, flat_order)
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [[1], new_dims]
elif Q.issuper:
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[0], []), dtype=np.int32)
cy_index_permute(Qcoo.row, q_dims, flat_order)
cy_index_permute(Qcoo.col, q_dims, flat_order)
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [new_dims, new_dims]
else:
raise TypeError('Invalid quantum object for permutation.')
return arr_coo2fast(Qcoo.data, Qcoo.row, Qcoo.col, Qcoo.shape[0],
Qcoo.shape[1]), new_dims
def _permute(Q, order):
if Q.isket:
dims, perm = _perm_inds(Q.dims[0], order)
nzs = Q.data.nonzero()[0]
wh = np.where(perm == nzs)[0]
data = np.ones(len(wh), dtype=complex)
cols = perm[wh].T[0]
perm_matrix = arr_coo2fast(data, wh.astype(np.int32),
cols.astype(np.int32), Q.shape[0],
Q.shape[0])
return perm_matrix * Q.data, Q.dims
elif Q.isbra:
dims, perm = _perm_inds(Q.dims[1], order)
nzs = Q.data.nonzero()[1]
wh = np.where(perm == nzs)[0]
data = np.ones(len(wh), dtype=complex)
rows = perm[wh].T[0]
perm_matrix = arr_coo2fast(data, rows.astype(np.int32),
wh.astype(np.int32), Q.shape[1], Q.shape[1])
return Q.data * perm_matrix, Q.dims
elif Q.isoper:
dims, perm = _perm_inds(Q.dims[0], order)
data = np.ones(Q.shape[0], dtype=complex)
rows = np.arange(Q.shape[0], dtype=np.int32)
perm_matrix = arr_coo2fast(data, rows, (perm.T[0]).astype(np.int32),
Q.shape[1], Q.shape[1])
dims_part = list(dims[order])
dims = [dims_part, dims_part]
return (perm_matrix * Q.data) * perm_matrix.T, dims
elif Q.issuper or Q.isoperket:
# For superoperators, we expect order to be something like
# [[0, 2], [1, 3]], which tells us to permute according to
# [0, 2, 1 ,3], and then group indices according to the length
# of each sublist.
# As another example,
# permuting [[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]] by
# [[0, 3], [1, 4], [2, 5]] should give
# [[[1, 1], [2, 2], [3, 3]], [[1, 1], [2, 2], [3, 3]]].
#
# Get the breakout of the left index into dims.
# Since this is a super, the left index itself breaks into left
# and right indices, each of which breaks down further.
# The best way to deal with that here is to flatten dims.
flat_order = sum(order, [])
q_dims_left = sum(Q.dims[0], [])
dims, perm = _perm_inds(q_dims_left, flat_order)
dims = dims.flatten()
data = np.ones(Q.shape[0], dtype=complex)
rows = np.arange(Q.shape[0], dtype=np.int32)
perm_matrix = arr_coo2fast(data, rows, (perm.T[0]).astype(np.int32),
Q.shape[0], Q.shape[0])
dims_part = list(dims[flat_order])
# Finally, we need to restructure the now-decomposed left index
# into left and right subindices, so that the overall dims we return
# are of the form specified by order.
dims_part = list(_chunk_dims(dims_part, order))
perm_left = (perm_matrix * Q.data)
if Q.type == 'operator-ket':
return perm_left, [dims_part, [1]]
elif Q.type == 'super':
return perm_left * perm_matrix.T, [dims_part, dims_part]
else:
raise TypeError('Invalid quantum object for permutation.')
def _permute(Q, order):
Qcoo = Q.data.tocoo()
if Q.isket:
cy_index_permute(Qcoo.row,
np.array(Q.dims[0], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [[Q.dims[0][i] for i in order], Q.dims[1]]
elif Q.isbra:
cy_index_permute(Qcoo.col,
np.array(Q.dims[1], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [Q.dims[0], [Q.dims[1][i] for i in order]]
elif Q.isoper:
cy_index_permute(Qcoo.row,
np.array(Q.dims[0], dtype=np.int32),
np.array(order, dtype=np.int32))
cy_index_permute(Qcoo.col,
np.array(Q.dims[1], dtype=np.int32),
np.array(order, dtype=np.int32))
new_dims = [[Q.dims[0][i] for i in order], [Q.dims[1][i] for i in order]]
elif Q.isoperket:
# For superoperators, we expect order to be something like
# [[0, 2], [1, 3]], which tells us to permute according to
# [0, 2, 1 ,3], and then group indices according to the length
# of each sublist.
# As another example,
# permuting [[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]] by
# [[0, 3], [1, 4], [2, 5]] should give
# [[[1, 1], [2, 2], [3, 3]], [[1, 1], [2, 2], [3, 3]]].
#
# Get the breakout of the left index into dims.
# Since this is a super, the left index itself breaks into left
# and right indices, each of which breaks down further.
# The best way to deal with that here is to flatten dims.
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[0], []), dtype=np.int32)
cy_index_permute(Qcoo.row, q_dims, flat_order)
# Finally, we need to restructure the now-decomposed left index
# into left and right subindices, so that the overall dims we return
# are of the form specified by order.
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [new_dims, [1]]
elif Q.isoperbra:
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[1], []), dtype=np.int32)
cy_index_permute(Qcoo.col, q_dims, flat_order)
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [[1], new_dims]
elif Q.issuper:
flat_order = np.array(sum(order, []), dtype=np.int32)
q_dims = np.array(sum(Q.dims[0], []), dtype=np.int32)
cy_index_permute(Qcoo.row, q_dims, flat_order)
cy_index_permute(Qcoo.col, q_dims, flat_order)
new_dims = [q_dims[i] for i in flat_order]
new_dims = list(_chunk_dims(new_dims, order))
new_dims = [new_dims, new_dims]
else:
raise TypeError('Invalid quantum object for permutation.')
return arr_coo2fast(Qcoo.data, Qcoo.row, Qcoo.col, Qcoo.shape[0], Qcoo.shape[1]), new_dims
def _permute(Q, order):
if Q.isket:
dims, perm = _perm_inds(Q.dims[0], order)
nzs = Q.data.nonzero()[0]
wh = np.where(perm == nzs)[0]
data = np.ones(len(wh), dtype=complex)
cols = perm[wh].T[0]
perm_matrix = arr_coo2fast(data, wh.astype(np.int32), cols.astype(np.int32),
Q.shape[0], Q.shape[0])
return perm_matrix * Q.data, Q.dims
elif Q.isbra:
dims, perm = _perm_inds(Q.dims[1], order)
nzs = Q.data.nonzero()[1]
wh = np.where(perm == nzs)[0]
data = np.ones(len(wh), dtype=complex)
rows = perm[wh].T[0]
perm_matrix = arr_coo2fast(data, rows.astype(np.int32), wh.astype(np.int32),
Q.shape[1], Q.shape[1])
return Q.data * perm_matrix, Q.dims
elif Q.isoper:
dims, perm = _perm_inds(Q.dims[0], order)
data = np.ones(Q.shape[0], dtype=complex)
rows = np.arange(Q.shape[0], dtype=np.int32)
perm_matrix = arr_coo2fast(data, rows, (perm.T[0]).astype(np.int32),
Q.shape[1], Q.shape[1])
dims_part = list(dims[order])
dims = [dims_part, dims_part]
return (perm_matrix * Q.data) * perm_matrix.T, dims
elif Q.issuper or Q.isoperket:
# For superoperators, we expect order to be something like
# [[0, 2], [1, 3]], which tells us to permute according to
# [0, 2, 1 ,3], and then group indices according to the length
# of each sublist.
# As another example,
# permuting [[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]] by
# [[0, 3], [1, 4], [2, 5]] should give
# [[[1, 1], [2, 2], [3, 3]], [[1, 1], [2, 2], [3, 3]]].
#
# Get the breakout of the left index into dims.
# Since this is a super, the left index itself breaks into left
# and right indices, each of which breaks down further.
# The best way to deal with that here is to flatten dims.
flat_order = sum(order, [])
q_dims_left = sum(Q.dims[0], [])
dims, perm = _perm_inds(q_dims_left, flat_order)
dims = dims.flatten()
data = np.ones(Q.shape[0], dtype=complex)
rows = np.arange(Q.shape[0], dtype=np.int32)
perm_matrix = arr_coo2fast(data, rows, (perm.T[0]).astype(np.int32),
Q.shape[0], Q.shape[0])
dims_part = list(dims[flat_order])
# Finally, we need to restructure the now-decomposed left index
# into left and right subindices, so that the overall dims we return
# are of the form specified by order.
dims_part = list(_chunk_dims(dims_part, order))
perm_left = (perm_matrix * Q.data)
if Q.type == 'operator-ket':
return perm_left, [dims_part, [1]]
elif Q.type == 'super':
return perm_left * perm_matrix.T, [dims_part, dims_part]
else:
raise TypeError('Invalid quantum object for permutation.')
