def _ode_solve(dt, t, b, y):
initializer.assert_initialized()
if _diffusion_matrix is None: _setup_matrices()
lo = _rxd_offset
hi = lo + len(_nonzero_volume_indices)
n = len(_node_get_states())
# TODO: this will need changed when can have both 1D and 3D
if species._has_3d:
if species._has_1d:
raise Exception(
'development issue: cvode currently does not support hybrid simulations (fix by shifting for zero volume indices)'
)
# NOTE: only working on the rxd part
rxd_b = b[lo:hi]
# TODO: make sure can handle both 1D and 3D
m = eye_minus_dt_J(n, dt)
# removed diagonal preconditioner since tests showed no improvement in convergence
result, info = _scipy_sparse_linalg_bicgstab(m, dt * rxd_b)
assert (info == 0)
b[lo:hi] = _react_matrix_solver(result)
else:
# 1D only; use Hines solver
full_b = numpy.zeros(n)
full_b[_nonzero_volume_indices] = b[lo:hi]
b[lo:hi] = _react_matrix_solver(_diffusion_matrix_solve(
dt, full_b))[_nonzero_volume_indices]
def _ode_fun(t, y, ydot):
initializer.assert_initialized()
lo = _rxd_offset
hi = lo + len(_nonzero_volume_indices)
if lo == hi: return
states = _node_get_states()
states[_nonzero_volume_indices] = y[lo : hi]
# need to fill in the zero volume states with the correct concentration
# this assumes that states at the zero volume indices is zero (although that
# assumption could be easily removed)
#matrix = _scipy_sparse_dok_matrix((len(_zero_volume_indices), len(states)))
"""
for i, row in enumerate(_zero_volume_indices):
d = _diffusion_matrix[row, row]
if d:
nzj = _diffusion_matrix[row].nonzero()[1]
print 'nzj:', nzj
for j in nzj:
matrix[i, j] = -_diffusion_matrix[row, j] / d
states[_zero_volume_indices] = matrix * states
"""
if len(_zero_volume_indices):
states[_zero_volume_indices] = _mat_for_zero_volume_nodes * states
"""
for i in _zero_volume_indices:
v = _diffusion_matrix[i] * states
d = _diffusion_matrix[i, i]
if d:
states[i] = -v / d
"""
# TODO: make this so that the section1d parts use cptrs (can't do this directly for 3D because sum, but could maybe move that into the C)
# the old way: _section1d_transfer_to_legacy()
for sr in _species_get_all_species().values():
s = sr()
if s is not None: s._transfer_to_legacy()
if ydot is not None:
# diffusion_matrix = - jacobian
ydot[lo : hi] = (_rxd_reaction(states) - _diffusion_matrix * states)[_nonzero_volume_indices]
states[_zero_volume_indices] = 0
def _ode_fun(t, y, ydot):
initializer.assert_initialized()
lo = _rxd_offset
hi = lo + len(_nonzero_volume_indices)
if lo == hi: return
states = _node_get_states()
states[_nonzero_volume_indices] = y[lo:hi]
# need to fill in the zero volume states with the correct concentration
# this assumes that states at the zero volume indices is zero (although that
# assumption could be easily removed)
#matrix = _scipy_sparse_dok_matrix((len(_zero_volume_indices), len(states)))
"""
for i, row in enumerate(_zero_volume_indices):
d = _diffusion_matrix[row, row]
if d:
nzj = _diffusion_matrix[row].nonzero()[1]
print 'nzj:', nzj
for j in nzj:
matrix[i, j] = -_diffusion_matrix[row, j] / d
states[_zero_volume_indices] = matrix * states
"""
if len(_zero_volume_indices):
states[_zero_volume_indices] = _mat_for_zero_volume_nodes * states
"""
for i in _zero_volume_indices:
v = _diffusion_matrix[i] * states
d = _diffusion_matrix[i, i]
if d:
states[i] = -v / d
"""
# TODO: make this so that the section1d parts use cptrs (can't do this directly for 3D because sum, but could maybe move that into the C)
# the old way: _section1d_transfer_to_legacy()
for sr in _species_get_all_species().values():
s = sr()
if s is not None: s._transfer_to_legacy()
if ydot is not None:
# diffusion_matrix = - jacobian
ydot[lo:hi] = (_rxd_reaction(states) -
_diffusion_matrix * states)[_nonzero_volume_indices]
states[_zero_volume_indices] = 0
def _ode_solve(dt, t, b, y):
initializer.assert_initialized()
if _diffusion_matrix is None: _setup_matrices()
lo = _rxd_offset
hi = lo + len(_nonzero_volume_indices)
n = len(_node_get_states())
# TODO: this will need changed when can have both 1D and 3D
if species._has_3d:
if species._has_1d:
raise Exception('development issue: cvode currently does not support hybrid simulations (fix by shifting for zero volume indices)')
# NOTE: only working on the rxd part
rxd_b = b[lo : hi]
# TODO: make sure can handle both 1D and 3D
m = _scipy_sparse_eye(n, n) - dt * _euler_matrix
# removed diagonal preconditioner since tests showed no improvement in convergence
result, info = _scipy_sparse_linalg_bicgstab(m, dt * rxd_b)
assert(info == 0)
b[lo : hi] = _react_matrix_solver(result)
else:
# 1D only; use Hines solver
full_b = numpy.zeros(n)
full_b[_nonzero_volume_indices] = b[lo : hi]
b[lo : hi] = _react_matrix_solver(_diffusion_matrix_solve(dt, full_b))[_nonzero_volume_indices]
