def test_rhs_shape_is_maintained(self):
A = scipy.sparse.csr_matrix(
np.array([
[1, 0, 2],
[0, 0, 3],
[-4, 5, 6],
], dtype=np.complex128))
b = np.array([0, 2, 0], dtype=np.complex128)
out = mkl_spsolve(A, b, verbose=True)
assert b.shape == out.shape
b = np.array([0, 2, 0], dtype=np.complex128).reshape((3, 1))
out = mkl_spsolve(A, b, verbose=True)
assert b.shape == out.shape
def test_mkl_spsolve7():
"""
MKL spsolve : Solution shape same as input RHS vec
"""
row = np.array([0,0,1,2,2,2])
col = np.array([0,2,2,0,1,2])
data = np.array([1,2,3,-4,5,6], dtype=complex)
A = sp.csr_matrix((data,(row,col)), shape=(3,3), dtype=complex)
b = np.array([0,2,0],dtype=complex)
out = mkl_spsolve(A,b)
assert_(b.shape==out.shape)
b = np.array([0,2,0],dtype=complex).reshape((3,1))
out = mkl_spsolve(A,b)
assert_(b.shape==out.shape)
def test_mkl_spsolve7():
"""
MKL spsolve : Solution shape same as input RHS vec
"""
row = np.array([0, 0, 1, 2, 2, 2])
col = np.array([0, 2, 2, 0, 1, 2])
data = np.array([1, 2, 3, -4, 5, 6], dtype=complex)
A = sp.csr_matrix((data, (row, col)), shape=(3, 3), dtype=complex)
b = np.array([0, 2, 0], dtype=complex)
out = mkl_spsolve(A, b)
assert_(b.shape == out.shape)
b = np.array([0, 2, 0], dtype=complex).reshape((3, 1))
out = mkl_spsolve(A, b)
assert_(b.shape == out.shape)
def test_single_rhs_vector_real(self):
Adense = np.array([[0, 1, 1], [1, 0, 1], [0, 0, 1]])
As = scipy.sparse.csr_matrix(Adense)
np.random.seed(1234)
x = np.random.randn(3)
b = As * x
x2 = mkl_spsolve(As, b, verbose=True)
np.testing.assert_allclose(x, x2)
def test_symmetric_solver(self, dtype):
A = qutip.rand_herm(np.arange(1, 11)).data
if dtype == np.float64:
A = A.real
x = np.ones(10, dtype=dtype)
b = A.dot(x)
y = mkl_spsolve(A, b, hermitian=1, verbose=True)
np.testing.assert_allclose(x, y)
def test_mklspsolve2():
"""
MKL spsolve : Single RHS vector (Complex)
"""
A = rand_herm(10)
x = rand_ket(10).full()
b = A * x
y = mkl_spsolve(A.data,b)
assert_array_almost_equal(x, y)
def test_mkl_spsolve9():
"""
MKL spsolve : Hermitian (complex) solver
"""
A = rand_herm(np.arange(1,11)).data
x = np.ones(10,dtype=complex)
b = A.dot(x)
y = mkl_spsolve(A, b, hermitian=1)
assert_array_almost_equal(x, y)
def test_mklspsolve2():
"""
MKL spsolve : Single RHS vector (Complex)
"""
A = rand_herm(10)
x = rand_ket(10).full()
b = A * x
y = mkl_spsolve(A.data, b)
assert_array_almost_equal(x, y)
def test_mkl_spsolve9():
"""
MKL spsolve : Hermitian (complex) solver
"""
A = rand_herm(np.arange(1, 11)).data
x = np.ones(10, dtype=complex)
b = A.dot(x)
y = mkl_spsolve(A, b, hermitian=1)
assert_array_almost_equal(x, y)
def test_mkl_spsolve10():
"""
MKL spsolve : Hermitian (real) solver
"""
A = rand_herm(np.arange(1, 11)).data
A = sp.csr_matrix((np.real(A.data), A.indices, A.indptr), dtype=float)
x = np.ones(10, dtype=float)
b = A.dot(x)
y = mkl_spsolve(A, b, hermitian=1)
assert_array_almost_equal(x, y)
def test_mkl_spsolve10():
"""
MKL spsolve : Hermitian (real) solver
"""
A = rand_herm(np.arange(1,11)).data
A = sp.csr_matrix((np.real(A.data), A.indices, A.indptr), dtype=float)
x = np.ones(10, dtype=float)
b = A.dot(x)
y = mkl_spsolve(A, b, hermitian=1)
assert_array_almost_equal(x, y)
def test_mkl_spsolve8():
"""
MKL spsolve : Sparse RHS matrix
"""
A = sp.csr_matrix([[1, 2, 0], [0, 3, 0], [0, 0, 5]])
b = sp.csr_matrix([[0, 1], [1, 0], [0, 0]])
x = mkl_spsolve(A, b)
ans = np.array([[-0.66666667, 1], [0.33333333, 0], [0, 0]])
assert_array_almost_equal(x.toarray(), ans)
def test_mkl_spsolve1():
"""
MKL spsolve : Single RHS vector (Real)
"""
Adense = np.array([[0., 1., 1.], [1., 0., 1.], [0., 0., 1.]])
As = sp.csr_matrix(Adense)
np.random.seed(1234)
x = np.random.randn(3)
b = As * x
x2 = mkl_spsolve(As, b)
assert_array_almost_equal(x, x2)
def test_mkl_spsolve1():
"""
MKL spsolve : Single RHS vector (Real)
"""
Adense = np.array([[0., 1., 1.],
[1., 0., 1.],
[0., 0., 1.]])
As = sp.csr_matrix(Adense)
np.random.seed(1234)
x = np.random.randn(3)
b = As * x
x2 = mkl_spsolve(As, b)
assert_array_almost_equal(x, x2)
def _pseudo_inverse_sparse(L, rhoss, method='splu', **pseudo_args):
"""
Internal function for computing the pseudo inverse of an Liouvillian using
sparse matrix methods. See pseudo_inverse for details.
"""
N = np.prod(L.dims[0][0])
rhoss_vec = operator_to_vector(rhoss)
tr_op = tensor([identity(n) for n in L.dims[0][0]])
tr_op_vec = operator_to_vector(tr_op)
P = sp.kron(rhoss_vec.data, tr_op_vec.data.T, format='csr')
I = sp.eye(N*N, N*N, format='csr')
Q = I - P
if pseudo_args['use_rcm']:
perm = reverse_cuthill_mckee(L.data)
A = sp_permute(L.data, perm, perm, 'csr')
Q = sp_permute(Q, perm, perm, 'csr')
else:
if not settings.has_mkl:
A = L.data.tocsc()
A.sort_indices()
if method == 'splu':
if settings.has_mkl:
LIQ = mkl_spsolve(A,Q.toarray())
else:
lu = sp.linalg.splu(A, permc_spec=pseudo_args['permc_spec'],
diag_pivot_thresh=pseudo_args['diag_pivot_thresh'],
options=dict(ILU_MILU=pseudo_args['ILU_MILU']))
LIQ = lu.solve(Q.toarray())
elif method == 'spilu':
lu = sp.linalg.spilu(A, permc_spec=pseudo_args['permc_spec'],
fill_factor=pseudo_args['fill_factor'],
drop_tol=pseudo_args['drop_tol'])
LIQ = lu.solve(Q.toarray())
else:
raise ValueError("unsupported method '%s'" % method)
R = sp.csr_matrix(Q * LIQ)
if pseudo_args['use_rcm']:
rev_perm = np.argsort(perm)
R = sp_permute(R, rev_perm, rev_perm, 'csr')
return Qobj(R, dims=L.dims)
def test_sparse_rhs(self):
A = scipy.sparse.csr_matrix([
[1, 2, 0],
[0, 3, 0],
[0, 0, 5],
])
b = scipy.sparse.csr_matrix([
[0, 1],
[1, 0],
[0, 0],
])
x = mkl_spsolve(A, b, verbose=True)
ans = np.array([[-0.66666667, 1], [0.33333333, 0], [0, 0]])
np.testing.assert_allclose(x.toarray(), ans)
def test_multi_rhs_vector(self, dtype):
M = np.array([
[1, 0, 2],
[0, 0, 3],
[-4, 5, 6],
], dtype=dtype)
sM = scipy.sparse.csr_matrix(M)
N = np.array([
[3, 0, 1],
[0, 2, 0],
[0, 0, 0],
], dtype=dtype)
sX = mkl_spsolve(sM, N, verbose=True)
X = scipy.linalg.solve(M, N)
np.testing.assert_allclose(X, sX)
def test_mkl_spsolve4():
"""
MKL spsolve : Multi RHS vector (Complex)
"""
row = np.array([0, 0, 1, 2, 2, 2])
col = np.array([0, 2, 2, 0, 1, 2])
data = np.array([1, 2, 3, -4, 5, 6], dtype=complex)
sM = sp.csr_matrix((data, (row, col)), shape=(3, 3), dtype=complex)
M = sM.toarray()
row = np.array([0, 0, 1, 1, 0, 0])
col = np.array([0, 2, 1, 1, 0, 0])
data = np.array([1, 1, 1, 1, 1, 1], dtype=complex)
sN = sp.csr_matrix((data, (row, col)), shape=(3, 3), dtype=complex)
N = sN.toarray()
sX = mkl_spsolve(sM, N)
X = la.solve(M, N)
assert_array_almost_equal(X, sX)
def test_mkl_spsolve4():
"""
MKL spsolve : Multi RHS vector (Complex)
"""
row = np.array([0,0,1,2,2,2])
col = np.array([0,2,2,0,1,2])
data = np.array([1,2,3,-4,5,6],dtype=complex)
sM = sp.csr_matrix((data,(row,col)), shape=(3,3), dtype=complex)
M = sM.toarray()
row = np.array([0,0,1,1,0,0])
col = np.array([0,2,1,1,0,0])
data = np.array([1,1,1,1,1,1],dtype=complex)
sN = sp.csr_matrix((data, (row,col)), shape=(3,3), dtype=complex)
N = sN.toarray()
sX = mkl_spsolve(sM, N)
X = la.solve(M, N)
assert_array_almost_equal(X, sX)
def test_mkl_spsolve8():
"""
MKL spsolve : Sparse RHS matrix
"""
A = sp.csr_matrix([
[1, 2, 0],
[0, 3, 0],
[0, 0, 5]])
b = sp.csr_matrix([
[0, 1],
[1, 0],
[0, 0]])
x = mkl_spsolve(A, b)
ans = np.array([[-0.66666667, 1],
[0.33333333, 0],
[0, 0]])
assert_array_almost_equal(x.toarray(), ans)
def _steadystate_direct_sparse(L, ss_args):
"""
Direct solver that uses scipy sparse matrices
"""
if settings.debug:
logger.debug('Starting direct LU solver.')
dims = L.dims[0]
n = int(np.sqrt(L.shape[0]))
b = np.zeros(n**2, dtype=complex)
b[0] = ss_args['weight']
if ss_args['solver'] == 'mkl':
has_mkl = 1
else:
has_mkl = 0
ss_lu_liouv_list = _steadystate_LU_liouvillian(L, ss_args, has_mkl)
L, perm, perm2, rev_perm, ss_args = ss_lu_liouv_list
if np.any(perm):
b = b[np.ix_(perm, )]
if np.any(perm2):
b = b[np.ix_(perm2, )]
if ss_args['solver'] == 'scipy':
ss_args['info']['permc_spec'] = ss_args['permc_spec']
ss_args['info']['drop_tol'] = ss_args['drop_tol']
ss_args['info']['diag_pivot_thresh'] = ss_args['diag_pivot_thresh']
ss_args['info']['fill_factor'] = ss_args['fill_factor']
ss_args['info']['ILU_MILU'] = ss_args['ILU_MILU']
if not ss_args['solver'] == 'mkl':
# Use superLU solver
orig_nnz = L.nnz
_direct_start = time.time()
lu = splu(L,
permc_spec=ss_args['permc_spec'],
diag_pivot_thresh=ss_args['diag_pivot_thresh'],
options=dict(ILU_MILU=ss_args['ILU_MILU']))
v = lu.solve(b)
_direct_end = time.time()
ss_args['info']['solution_time'] = _direct_end - _direct_start
if (settings.debug or ss_args['return_info']) and _scipy_check:
L_nnz = lu.L.nnz
U_nnz = lu.U.nnz
ss_args['info']['l_nnz'] = L_nnz
ss_args['info']['u_nnz'] = U_nnz
ss_args['info']['lu_fill_factor'] = (L_nnz + U_nnz) / L.nnz
if settings.debug:
logger.debug('L NNZ: %i ; U NNZ: %i' % (L_nnz, U_nnz))
logger.debug('Fill factor: %f' % ((L_nnz + U_nnz) / orig_nnz))
else: # Use MKL solver
if len(ss_args['info']['perm']) != 0:
in_perm = np.arange(n**2, dtype=np.int32)
else:
in_perm = None
_direct_start = time.time()
v = mkl_spsolve(L,
b,
perm=in_perm,
verbose=ss_args['verbose'],
max_iter_refine=ss_args['max_iter_refine'],
scaling_vectors=ss_args['scaling_vectors'],
weighted_matching=ss_args['weighted_matching'])
_direct_end = time.time()
ss_args['info']['solution_time'] = _direct_end - _direct_start
if ss_args['return_info']:
ss_args['info']['residual_norm'] = la.norm(b - L * v, np.inf)
ss_args['info']['max_iter_refine'] = ss_args['max_iter_refine']
ss_args['info']['scaling_vectors'] = ss_args['scaling_vectors']
ss_args['info']['weighted_matching'] = ss_args['weighted_matching']
if ss_args['use_rcm']:
v = v[np.ix_(rev_perm, )]
data = dense2D_to_fastcsr_fmode(vec2mat(v), n, n)
data = 0.5 * (data + data.H)
if ss_args['return_info']:
return Qobj(data, dims=dims, isherm=True), ss_args['info']
else:
return Qobj(data, dims=dims, isherm=True)
def steady_state(self,
use_mkl=True,
mkl_max_iter_refine=100,
mkl_weighted_matching=False):
"""
Compute the steady state of the system.
Parameters
----------
use_mkl : bool, default=False
Whether to use mkl or not. If mkl is not installed or if
this is false, use the scipy splu solver instead.
mkl_max_iter_refine : int
Specifies the the maximum number of iterative refinement steps that
the MKL PARDISO solver performs.
For a complete description, see iparm(8) in
http://cali2.unilim.fr/intel-xe/mkl/mklman/GUID-264E311E-ACED-4D56-AC31-E9D3B11D1CBF.htm.
mkl_weighted_matching : bool
MKL PARDISO can use a maximum weighted matching algorithm to
permute large elements close the diagonal. This strategy adds an
additional level of reliability to the factorization methods.
For a complete description, see iparm(13) in
http://cali2.unilim.fr/intel-xe/mkl/mklman/GUID-264E311E-ACED-4D56-AC31-E9D3B11D1CBF.htm.
Returns
-------
steady_state : Qobj
The steady state density matrix of the system.
steady_ados : :class:`HierarchyADOsState`
The steady state of the full ADO hierarchy. A particular ADO may be
extracted from the full state by calling :meth:`.extract`.
"""
n = self._sys_shape
b_mat = np.zeros(n**2 * self._n_ados, dtype=complex)
b_mat[0] = 1.0
L = deepcopy(self.RHSmat)
L = L.tolil()
L[0, 0:n**2 * self._n_ados] = 0.0
L = L.tocsr()
L += sp.csr_matrix(
(np.ones(n), (np.zeros(n), [num * (n + 1) for num in range(n)])),
shape=(n**2 * self._n_ados, n**2 * self._n_ados))
if mkl_spsolve is not None and use_mkl:
L.sort_indices()
solution = mkl_spsolve(
L,
b_mat,
perm=None,
verbose=False,
max_iter_refine=mkl_max_iter_refine,
scaling_vectors=True,
weighted_matching=mkl_weighted_matching,
)
else:
L = L.tocsc()
solution = spsolve(L, b_mat)
data = dense2D_to_fastcsr_fmode(vec2mat(solution[:n**2]), n, n)
data = 0.5 * (data + data.H)
steady_state = Qobj(data, dims=self._sys_dims)
solution = solution.reshape((self._n_ados, n, n))
steady_ados = HierarchyADOsState(steady_state, self.ados, solution)
return steady_state, steady_ados
def _pseudo_inverse_sparse(L, rhoss, w=None, **pseudo_args):
"""
Internal function for computing the pseudo inverse of an Liouvillian using
sparse matrix methods. See pseudo_inverse for details.
"""
N = np.prod(L.dims[0][0])
rhoss_vec = operator_to_vector(rhoss)
tr_op = tensor([identity(n) for n in L.dims[0][0]])
tr_op_vec = operator_to_vector(tr_op)
P = zcsr_kron(rhoss_vec.data, tr_op_vec.data.T)
I = sp.eye(N * N, N * N, format='csr')
Q = I - P
if w is None:
L = 1.0j * (1e-15) * spre(tr_op) + L
else:
if w != 0.0:
L = 1.0j * w * spre(tr_op) + L
else:
L = 1.0j * (1e-15) * spre(tr_op) + L
if pseudo_args['use_rcm']:
perm = sp.csgraph.reverse_cuthill_mckee(L.data)
A = sp_permute(L.data, perm, perm)
Q = sp_permute(Q, perm, perm)
else:
if ss_args['solver'] == 'scipy':
A = L.data.tocsc()
A.sort_indices()
if pseudo_args['method'] == 'splu':
if settings.has_mkl:
A = L.data.tocsr()
A.sort_indices()
LIQ = mkl_spsolve(A, Q.toarray())
else:
pspec = pseudo_args['permc_spec']
diag_p_thresh = pseudo_args['diag_pivot_thresh']
pseudo_args = pseudo_args['ILU_MILU']
lu = sp.linalg.splu(A,
permc_spec=pspec,
diag_pivot_thresh=diag_p_thresh,
options=dict(ILU_MILU=pseudo_args))
LIQ = lu.solve(Q.toarray())
elif pseudo_args['method'] == 'spilu':
lu = sp.linalg.spilu(A,
permc_spec=pseudo_args['permc_spec'],
fill_factor=pseudo_args['fill_factor'],
drop_tol=pseudo_args['drop_tol'])
LIQ = lu.solve(Q.toarray())
else:
raise ValueError("unsupported method '%s'" % method)
R = sp.csr_matrix(Q * LIQ)
if pseudo_args['use_rcm']:
rev_perm = np.argsort(perm)
R = sp_permute(R, rev_perm, rev_perm, 'csr')
return Qobj(R, dims=L.dims)
def _steadystate_direct_sparse(L, ss_args):
"""
Direct solver that uses scipy sparse matrices
"""
if settings.debug:
logger.debug('Starting direct LU solver.')
dims = L.dims[0]
n = int(np.sqrt(L.shape[0]))
b = np.zeros(n ** 2, dtype=complex)
b[0] = ss_args['weight']
if ss_args['solver'] == 'mkl':
has_mkl = 1
else:
has_mkl = 0
ss_lu_liouv_list = _steadystate_LU_liouvillian(L, ss_args, has_mkl)
L, perm, perm2, rev_perm, ss_args = ss_lu_liouv_list
if np.any(perm):
b = b[np.ix_(perm,)]
if np.any(perm2):
b = b[np.ix_(perm2,)]
if ss_args['solver'] == 'scipy':
ss_args['info']['permc_spec'] = ss_args['permc_spec']
ss_args['info']['drop_tol'] = ss_args['drop_tol']
ss_args['info']['diag_pivot_thresh'] = ss_args['diag_pivot_thresh']
ss_args['info']['fill_factor'] = ss_args['fill_factor']
ss_args['info']['ILU_MILU'] = ss_args['ILU_MILU']
if not ss_args['solver'] == 'mkl':
# Use superLU solver
orig_nnz = L.nnz
_direct_start = time.time()
lu = splu(L, permc_spec=ss_args['permc_spec'],
diag_pivot_thresh=ss_args['diag_pivot_thresh'],
options=dict(ILU_MILU=ss_args['ILU_MILU']))
v = lu.solve(b)
_direct_end = time.time()
ss_args['info']['solution_time'] = _direct_end - _direct_start
if (settings.debug or ss_args['return_info']) and _scipy_check:
L_nnz = lu.L.nnz
U_nnz = lu.U.nnz
ss_args['info']['l_nnz'] = L_nnz
ss_args['info']['u_nnz'] = U_nnz
ss_args['info']['lu_fill_factor'] = (L_nnz + U_nnz)/L.nnz
if settings.debug:
logger.debug('L NNZ: %i ; U NNZ: %i' % (L_nnz, U_nnz))
logger.debug('Fill factor: %f' % ((L_nnz + U_nnz)/orig_nnz))
else: # Use MKL solver
if len(ss_args['info']['perm']) != 0:
in_perm = np.arange(n**2, dtype=np.int32)
else:
in_perm = None
_direct_start = time.time()
v = mkl_spsolve(L, b, perm=in_perm, verbose=ss_args['verbose'],
max_iter_refine=ss_args['max_iter_refine'],
scaling_vectors=ss_args['scaling_vectors'],
weighted_matching=ss_args['weighted_matching'])
_direct_end = time.time()
ss_args['info']['solution_time'] = _direct_end-_direct_start
if ss_args['return_info']:
ss_args['info']['residual_norm'] = la.norm(b - L*v, np.inf)
ss_args['info']['max_iter_refine'] = ss_args['max_iter_refine']
ss_args['info']['scaling_vectors'] = ss_args['scaling_vectors']
ss_args['info']['weighted_matching'] = ss_args['weighted_matching']
if ss_args['use_rcm']:
v = v[np.ix_(rev_perm,)]
data = dense2D_to_fastcsr_fmode(vec2mat(v), n, n)
data = 0.5 * (data + data.H)
if ss_args['return_info']:
return Qobj(data, dims=dims, isherm=True), ss_args['info']
else:
return Qobj(data, dims=dims, isherm=True)
def test_single_rhs_vector_complex(self):
A = qutip.rand_herm(10)
x = qutip.rand_ket(10).full()
b = A.full() @ x
y = mkl_spsolve(A.data, b, verbose=True)
np.testing.assert_allclose(x, y)
def countstat_current_noise(L,
c_ops,
wlist=None,
rhoss=None,
J_ops=None,
sparse=True,
method='direct'):
"""
Compute the cross-current noise spectrum for a list of collapse operators
`c_ops` corresponding to monitored currents, given the system
Liouvillian `L`. The current collapse operators `c_ops` should be part
of the dissipative processes in `L`, but the `c_ops` given here does not
necessarily need to be all collapse operators contributing to dissipation
in the Liouvillian. Optionally, the steadystate density matrix `rhoss`
and the current operators `J_ops` correpsonding to the current collapse
operators `c_ops` can also be specified. If either of
`rhoss` and `J_ops` are omitted, they will be computed internally.
'wlist' is an optional list of frequencies at which to evaluate the noise
spectrum.
Note:
The default method is a direct solution using dense matrices, as sparse
matrix methods fail for some examples of small systems.
For larger systems it is reccomended to use the sparse solver
with the direct method, as it avoids explicit calculation of the
pseudo-inverse, as described in page 67 of "Electrons in nanostructures"
C. Flindt, PhD Thesis, available online:
http://orbit.dtu.dk/fedora/objects/orbit:82314/datastreams/file_4732600/content
Parameters
----------
L : :class:`qutip.Qobj`
Qobj representing the system Liouvillian.
c_ops : array / list
List of current collapse operators.
rhoss : :class:`qutip.Qobj` (optional)
The steadystate density matrix corresponding the system Liouvillian
`L`.
wlist : array / list (optional)
List of frequencies at which to evaluate (if none are given, evaluates
at zero frequency)
J_ops : array / list (optional)
List of current superoperators.
sparse : bool
Flag that indicates whether to use sparse or dense matrix methods when
computing the pseudo inverse. Default is false, as sparse solvers
can fail for small systems. For larger systems the sparse solvers
are reccomended.
Returns
--------
I, S : tuple of arrays
The currents `I` corresponding to each current collapse operator
`c_ops` (or, equivalently, each current superopeator `J_ops`) and the
zero-frequency cross-current correlation `S`.
"""
if rhoss is None:
rhoss = steadystate(L, c_ops)
if J_ops is None:
J_ops = [sprepost(c, c.dag()) for c in c_ops]
N = len(J_ops)
I = np.zeros(N)
if wlist is None:
S = np.zeros((N, N, 1))
wlist = [0.]
else:
S = np.zeros((N, N, len(wlist)))
if sparse == False:
rhoss_vec = mat2vec(rhoss.full()).ravel()
for k, w in enumerate(wlist):
R = pseudo_inverse(L,
rhoss=rhoss,
w=w,
sparse=sparse,
method=method)
for i, Ji in enumerate(J_ops):
for j, Jj in enumerate(J_ops):
if i == j:
I[i] = expect_rho_vec(Ji.data, rhoss_vec, 1)
S[i, j, k] = I[i]
S[i, j, k] -= expect_rho_vec(
(Ji * R * Jj + Jj * R * Ji).data, rhoss_vec, 1)
else:
if method == "direct":
N = np.prod(L.dims[0][0])
rhoss_vec = operator_to_vector(rhoss)
tr_op = tensor([identity(n) for n in L.dims[0][0]])
tr_op_vec = operator_to_vector(tr_op)
Pop = sp.kron(rhoss_vec.data, tr_op_vec.data.T, format='csr')
Iop = sp.eye(N * N, N * N, format='csr')
Q = Iop - Pop
for k, w in enumerate(wlist):
if w != 0.0:
L_temp = 1.0j * w * spre(tr_op) + L
else: #At zero frequency some solvers fail for small systems.
#Adding a small finite frequency of order 1e-15
#helps prevent the solvers from throwing an exception.
L_temp = 1.0j * (1e-15) * spre(tr_op) + L
if not settings.has_mkl:
A = L_temp.data.tocsc()
else:
A = L_temp.data.tocsr()
A.sort_indices()
rhoss_vec = mat2vec(rhoss.full()).ravel()
for j, Jj in enumerate(J_ops):
Qj = Q.dot(Jj.data.dot(rhoss_vec))
try:
if settings.has_mkl:
X_rho_vec_j = mkl_spsolve(A, Qj)
else:
X_rho_vec_j = sp.linalg.splu(
A, permc_spec='COLAMD').solve(Qj)
except:
X_rho_vec_j = sp.linalg.lsqr(A, Qj)[0]
for i, Ji in enumerate(J_ops):
Qi = Q.dot(Ji.data.dot(rhoss_vec))
try:
if settings.has_mkl:
X_rho_vec_i = mkl_spsolve(A, Qi)
else:
X_rho_vec_i = sp.linalg.splu(
A, permc_spec='COLAMD').solve(Qi)
except:
X_rho_vec_i = sp.linalg.lsqr(A, Qi)[0]
if i == j:
I[i] = expect_rho_vec(Ji.data, rhoss_vec, 1)
S[j, i, k] = I[i]
S[j, i,
k] -= (expect_rho_vec(Jj.data * Q, X_rho_vec_i, 1) +
expect_rho_vec(Ji.data * Q, X_rho_vec_j, 1))
else:
rhoss_vec = mat2vec(rhoss.full()).ravel()
for k, w in enumerate(wlist):
R = pseudo_inverse(L,
rhoss=rhoss,
w=w,
sparse=sparse,
method=method)
for i, Ji in enumerate(J_ops):
for j, Jj in enumerate(J_ops):
if i == j:
I[i] = expect_rho_vec(Ji.data, rhoss_vec, 1)
S[i, j, k] = I[i]
S[i, j, k] -= expect_rho_vec(
(Ji * R * Jj + Jj * R * Ji).data, rhoss_vec, 1)
return I, S
def countstat_current_noise(L, c_ops, wlist=None, rhoss=None, J_ops=None,
sparse=True, method='direct'):
"""
Compute the cross-current noise spectrum for a list of collapse operators
`c_ops` corresponding to monitored currents, given the system
Liouvillian `L`. The current collapse operators `c_ops` should be part
of the dissipative processes in `L`, but the `c_ops` given here does not
necessarily need to be all collapse operators contributing to dissipation
in the Liouvillian. Optionally, the steadystate density matrix `rhoss`
and the current operators `J_ops` correpsonding to the current collapse
operators `c_ops` can also be specified. If either of
`rhoss` and `J_ops` are omitted, they will be computed internally.
'wlist' is an optional list of frequencies at which to evaluate the noise
spectrum.
Note:
The default method is a direct solution using dense matrices, as sparse
matrix methods fail for some examples of small systems.
For larger systems it is reccomended to use the sparse solver
with the direct method, as it avoids explicit calculation of the
pseudo-inverse, as described in page 67 of "Electrons in nanostructures"
C. Flindt, PhD Thesis, available online:
http://orbit.dtu.dk/fedora/objects/orbit:82314/datastreams/file_4732600/content
Parameters
----------
L : :class:`qutip.Qobj`
Qobj representing the system Liouvillian.
c_ops : array / list
List of current collapse operators.
rhoss : :class:`qutip.Qobj` (optional)
The steadystate density matrix corresponding the system Liouvillian
`L`.
wlist : array / list (optional)
List of frequencies at which to evaluate (if none are given, evaluates
at zero frequency)
J_ops : array / list (optional)
List of current superoperators.
sparse : bool
Flag that indicates whether to use sparse or dense matrix methods when
computing the pseudo inverse. Default is false, as sparse solvers
can fail for small systems. For larger systems the sparse solvers
are reccomended.
Returns
--------
I, S : tuple of arrays
The currents `I` corresponding to each current collapse operator
`c_ops` (or, equivalently, each current superopeator `J_ops`) and the
zero-frequency cross-current correlation `S`.
"""
if rhoss is None:
rhoss = steadystate(L, c_ops)
if J_ops is None:
J_ops = [sprepost(c, c.dag()) for c in c_ops]
N = len(J_ops)
I = np.zeros(N)
if wlist is None:
S = np.zeros((N, N,1))
wlist=[0.]
else:
S = np.zeros((N, N,len(wlist)))
if sparse == False:
rhoss_vec = mat2vec(rhoss.full()).ravel()
for k,w in enumerate(wlist):
R = pseudo_inverse(L, rhoss=rhoss, w= w, sparse = sparse, method=method)
for i, Ji in enumerate(J_ops):
for j, Jj in enumerate(J_ops):
if i == j:
I[i] = expect_rho_vec(Ji.data, rhoss_vec, 1)
S[i, j,k] = I[i]
S[i, j,k] -= expect_rho_vec((Ji * R * Jj
+ Jj * R * Ji).data,
rhoss_vec, 1)
else:
if method == "direct":
N = np.prod(L.dims[0][0])
rhoss_vec = operator_to_vector(rhoss)
tr_op = tensor([identity(n) for n in L.dims[0][0]])
tr_op_vec = operator_to_vector(tr_op)
Pop = sp.kron(rhoss_vec.data, tr_op_vec.data.T, format='csr')
Iop = sp.eye(N*N, N*N, format='csr')
Q = Iop - Pop
for k,w in enumerate(wlist):
if w != 0.0:
L_temp = 1.0j*w*spre(tr_op) + L
else: #At zero frequency some solvers fail for small systems.
#Adding a small finite frequency of order 1e-15
#helps prevent the solvers from throwing an exception.
L_temp = 1.0j*(1e-15)*spre(tr_op) + L
if not settings.has_mkl:
A = L_temp.data.tocsc()
else:
A = L_temp.data.tocsr()
A.sort_indices()
rhoss_vec = mat2vec(rhoss.full()).ravel()
for j, Jj in enumerate(J_ops):
Qj = Q.dot( Jj.data.dot( rhoss_vec))
try:
if settings.has_mkl:
X_rho_vec_j = mkl_spsolve(A,Qj)
else:
X_rho_vec_j = sp.linalg.splu(A, permc_spec
='COLAMD').solve(Qj)
except:
X_rho_vec_j = sp.linalg.lsqr(A,Qj)[0]
for i, Ji in enumerate(J_ops):
Qi = Q.dot( Ji.data.dot(rhoss_vec))
try:
if settings.has_mkl:
X_rho_vec_i = mkl_spsolve(A,Qi)
else:
X_rho_vec_i = sp.linalg.splu(A, permc_spec
='COLAMD').solve(Qi)
except:
X_rho_vec_i = sp.linalg.lsqr(A,Qi)[0]
if i == j:
I[i] = expect_rho_vec(Ji.data,
rhoss_vec, 1)
S[j, i, k] = I[i]
S[j, i, k] -= (expect_rho_vec(Jj.data * Q,
X_rho_vec_i, 1)
+ expect_rho_vec(Ji.data * Q,
X_rho_vec_j, 1))
else:
rhoss_vec = mat2vec(rhoss.full()).ravel()
for k,w in enumerate(wlist):
R = pseudo_inverse(L,rhoss=rhoss, w= w, sparse = sparse,
method=method)
for i, Ji in enumerate(J_ops):
for j, Jj in enumerate(J_ops):
if i == j:
I[i] = expect_rho_vec(Ji.data, rhoss_vec, 1)
S[i, j, k] = I[i]
S[i, j, k] -= expect_rho_vec((Ji * R * Jj
+ Jj * R * Ji).data,
rhoss_vec, 1)
return I, S
def steady_state(self, max_iter_refine = 100, use_mkl = False, weighted_matching = False, series_method = False):
"""
Computes steady state dynamics
max_iter_refine : Int
Parameter for the mkl LU solver. If pardiso errors are returned this should be increased.
use_mkl : Boolean
Optional override default use of mkl if mkl is installed.
weighted_matching : Boolean
Setting this true may increase run time, but reduce stability (pardisio may not converge).
"""
nstates = self._N_he
sup_dim = self._sup_dim
n = int(np.sqrt(sup_dim))
unit_h_elems = sp.identity(nstates, format='csr')
L = deepcopy(self.RHSmat)# + sp.kron(unit_h_elems,
#liouvillian(H).data)
b_mat = np.zeros(sup_dim*nstates, dtype=complex)
b_mat[0] = 1.
L = L.tolil()
L[0, 0 : n**2*nstates] = 0.
L = L.tocsr()
if settings.has_mkl & use_mkl == True:
print("Using Intel mkl solver")
from qutip._mkl.spsolve import (mkl_splu, mkl_spsolve)
L = L.tocsr() + \
sp.csr_matrix((np.ones(n), (np.zeros(n),
[num*(n+1)for num in range(n)])),
shape=(n**2*nstates, n**2*nstates))
L.sort_indices()
solution = mkl_spsolve(L, b_mat, perm = None, verbose = True, \
max_iter_refine = max_iter_refine, \
scaling_vectors = True, \
weighted_matching = weighted_matching)
else:
if series_method == False:
L = L.tocsc() + \
sp.csc_matrix((np.ones(n), (np.zeros(n),
[num*(n+1)for num in range(n)])),
shape=(n**2*nstates, n**2*nstates))
# Use superLU solver
LU = splu(L)
solution = LU.solve(b_mat)
else:
L = L.tocsc() + \
sp.csc_matrix((np.ones(n), (np.zeros(n),
[num*(n+1)for num in range(n)])),
shape=(n**2*nstates, n**2*nstates))
# Use series method
L.sort_indices()
solution,fidelity = lgmres(L, b_mat)
dims = self.H_sys.dims
data = dense2D_to_fastcsr_fmode(vec2mat(solution[:sup_dim]), n, n)
data = 0.5*(data + data.H)
solution = solution.reshape((nstates, self.H_sys.shape[0]**2))
return Qobj(data, dims=dims), solution