def aveF2CC(self):
"Construct the averaging operator on cell faces to cell centers."
if getattr(self, '_aveF2CC', None) is None:
n = self.vnC
if self.isSymmetric:
avR = Utils.av(n[0])[:, 1:]
avR[0, 0] = 1.
self._aveF2CC = ((0.5)*sp.hstack((sp.kron(Utils.speye(n[2]),
avR),
sp.kron(Utils.av(n[2]),
Utils.speye(n[0]))),
format="csr"))
else:
raise NotImplementedError('wrapping in the averaging is not '
'yet implemented')
# self._aveF2CC = (1./3.)*sp.hstack((Utils.kron3(Utils.speye(n[2]),
# Utils.speye(n[1]),
# Utils.av(n[0])),
# Utils.kron3(Utils.speye(n[2]),
# Utils.av(n[1]),
# Utils.speye(n[0])),
# Utils.kron3(Utils.av(n[2]),
# Utils.speye(n[1]),
# Utils.speye(n[0]))),
# format="csr")
return self._aveF2CC
def W(self):
"""Regularization matrix W"""
if getattr(self, '_W', None) is None:
ntx = self.xyz_line.shape[0]
meshline = Mesh.TensorMesh([ntx])
Gx = meshline.cellGradx
Wx = np.sqrt(self.alpha_x) * sp.kron(Gx, Utils.speye(self.ntau))
Ws = np.sqrt(self.alpha_s) * Utils.speye(self.ntau * ntx)
wlist = (Wx, Ws)
self._W = sp.vstack(wlist)
return self._W
def faceDivz(self):
"""
Construct divergence operator in the z component
(face-stg to cell-centres).
"""
if getattr(self, "_faceDivz", None) is None:
D3 = Utils.kron3(Utils.ddx(self.nCz), Utils.speye(self.nCy), Utils.speye(self.nCx))
S = self.r(self.area, "F", "Fz", "V")
V = self.vol
self._faceDivz = Utils.sdiag(1 / V) * D3 * Utils.sdiag(S)
return self._faceDivz
def faceDivz(self):
"""
Construct divergence operator in the z component
(face-stg to cell-centres).
"""
if getattr(self, '_faceDivz', None) is None:
D3 = Utils.kron3(Utils.ddx(self.nCz), Utils.speye(self.nCy),
Utils.speye(self.nCx))
S = self.r(self.area, 'F', 'Fz', 'V')
V = self.vol
self._faceDivz = Utils.sdiag(1/V)*D3*Utils.sdiag(S)
return self._faceDivz
def aveF2CCV(self):
"Construct the averaging operator on cell faces to cell centers."
if getattr(self, "_aveF2CCV", None) is None:
n = self.vnC
if self.isSymmetric:
avR = Utils.av(n[0])[:, 1:]
avR[0, 0] = 1.0
self._aveF2CCV = sp.block_diag(
(sp.kron(Utils.speye(n[2]), avR), sp.kron(Utils.av(n[2]), Utils.speye(n[0]))), format="csr"
)
else:
raise NotImplementedError("wrapping in the averaging is not " "yet implemented")
return self._aveF2CCV
def faceDivy(self):
"""
Construct divergence operator in the y component
(face-stg to cell-centres).
"""
raise NotImplementedError("Wrapping the Utils.ddx is not yet " "implemented.")
if getattr(self, "_faceDivy", None) is None:
# TODO: this needs to wrap to join up faces which are
# connected in the cylinder
D2 = Utils.kron3(Utils.speye(self.nCz), Utils.ddx(self.nCy), Utils.speye(self.nCx))
S = self.r(self.area, "F", "Fy", "V")
V = self.vol
self._faceDivy = Utils.sdiag(1 / V) * D2 * Utils.sdiag(S)
return self._faceDivy
def get_grad_horizontal(self, xy, hz):
"""
Compute Gradient in horizontal direction using Delaunay
"""
tri = sp.spatial.Delaunay(xy)
# Split the triangulation into connections
edges = np.r_[tri.simplices[:, :2], tri.simplices[:, 1:],
tri.simplices[:, [0, 2]]]
# Sort and keep uniques
edges = np.sort(edges, axis=1)
edges = np.unique(edges[np.argsort(edges[:, 0]), :], axis=0)
# Create 2D operator, dimensionless for now
nN = edges.shape[0]
nStn = xy.shape[0]
stn, count = np.unique(edges[:, 0], return_counts=True)
col = []
row = []
dm = []
avg = []
for ii in range(nN):
row += [ii] * 2
col += [edges[ii, 0], edges[ii, 1]]
scale = count[stn == edges[ii, 0]][0]
dm += [-1., 1.]
avg += [0.5, 0.5]
D = sp.sparse.csr_matrix((dm, (row, col)), shape=(nN, nStn))
A = sp.sparse.csr_matrix((avg, (row, col)), shape=(nN, nStn))
# Kron vertically for nCz
Grad = sp.sparse.kron(D, Utils.speye(hz.size))
Avg = sp.sparse.kron(A, Utils.speye(hz.size))
# Override the gradient operator in y-drection
# This is because of ordering ... See def get_2d_mesh
# y first then x
self.regmesh._cellDiffyStencil = self.regmesh.cellDiffxStencil.copy()
# Override the gradient operator in x-drection
self.regmesh._cellDiffxStencil = Grad
# Do the same for the averaging operator
self.regmesh._aveCC2Fy = self.regmesh.aveCC2Fx.copy()
self.regmesh._aveCC2Fx = Avg
self.regmesh._aveFy2CC = self.regmesh.aveFx2CC.copy()
self.regmesh._aveFx2CC = Avg.T
return tri
def aveF2CCV(self):
"Construct the averaging operator on cell faces to cell centers."
if getattr(self, '_aveF2CCV', None) is None:
n = self.vnC
if self.isSymmetric:
avR = Utils.av(n[0])[:, 1:]
avR[0, 0] = 1.
self._aveF2CCV = sp.block_diag(
(sp.kron(Utils.speye(n[2]),
avR), sp.kron(Utils.av(n[2]), Utils.speye(n[0]))),
format="csr")
else:
raise NotImplementedError('wrapping in the averaging is not '
'yet implemented')
return self._aveF2CCV
def faceDivy(self):
"""
Construct divergence operator in the y component
(face-stg to cell-centres).
"""
raise NotImplementedError('Wrapping the Utils.ddx is not yet '
'implemented.')
if getattr(self, '_faceDivy', None) is None:
# TODO: this needs to wrap to join up faces which are
# connected in the cylinder
D2 = Utils.kron3(Utils.speye(self.nCz), Utils.ddx(self.nCy),
Utils.speye(self.nCx))
S = self.r(self.area, 'F', 'Fy', 'V')
V = self.vol
self._faceDivy = Utils.sdiag(1/V)*D2*Utils.sdiag(S)
return self._faceDivy
def Pac(self):
"""
diagonal matrix that nulls out inactive cells
:rtype: scipy.sparse.csr_matrix
:return: active cell diagonal matrix
"""
if getattr(self, "_Pac", None) is None:
if self.indActive is None:
self._Pac = Utils.speye(self.mesh.nC)
else:
e = np.zeros(self.mesh.nC)
e[self.indActive] = 1.0
# self._Pac = Utils.speye(self.mesh.nC)[:, self.indActive]
self._Pac = Utils.sdiag(e)
return self._Pac
def Pafy(self):
"""
diagonal matrix that nulls out inactive y-faces
to full modelling space (ie. nFy x nindActive_Fy )
:rtype: scipy.sparse.csr_matrix
:return: active face-y diagonal matrix
"""
if getattr(self, '_Pafy', None) is None:
if self.indActive is None:
self._Pafy = Utils.speye(self.mesh.nFy)
else:
indActive_Fy = (self.mesh.aveFy2CC.T * self.indActive) >= 1
e = np.zeros(self.mesh.nFy)
e[indActive_Fy] = 1.
self._Pafy = Utils.sdiag(e)
return self._Pafy
def Pafz(self):
"""
diagonal matrix that nulls out inactive z-faces
to full modelling space (ie. nFz x nindActive_Fz )
:rtype: scipy.sparse.csr_matrix
:return: active face-z diagonal matrix
"""
if getattr(self, "_Pafz", None) is None:
if self.indActive is None:
self._Pafz = Utils.speye(self.mesh.nFz)
else:
indActive_Fz = (self.mesh.aveFz2CC.T * self.indActive) >= 1
e = np.zeros(self.mesh.nFz)
e[indActive_Fz] = 1.0
self._Pafz = Utils.sdiag(e)
return self._Pafz
def Pafx(self):
"""
diagonal matrix that nulls out inactive x-faces
to full modelling space (ie. nFx x nindActive_Fx )
:rtype: scipy.sparse.csr_matrix
:return: active face-x diagonal matrix
"""
if getattr(self, "_Pafx", None) is None:
if self.indActive is None:
self._Pafx = Utils.speye(self.mesh.nFx)
else:
indActive_Fx = self.mesh.aveFx2CC.T * self.indActive >= 1
e = np.zeros(self.mesh.nFx)
e[indActive_Fx] = 1.0
self._Pafx = Utils.sdiag(e)
return self._Pafx
def getJtJdiag(self, m, W=None, threshold=1e-8):
"""
Compute diagonal component of JtJ or
trace of sensitivity matrix (J)
"""
J_sigma = self.getJ_sigma(m)
J_matrix = J_sigma*(Utils.sdiag(1./self.sigma)*(self.sigmaDeriv))
if self.hMap is not None:
J_height = self.getJ_height(m)
J_matrix += J_height*self.hDeriv
if W is None:
W = Utils.speye(J_matrix.shape[0])
J_matrix = W*J_matrix
JtJ_diag = (J_matrix.T*J_matrix).diagonal()
JtJ_diag /= JtJ_diag.max()
JtJ_diag += threshold
return JtJ_diag
def getAdiag(self, tInd):
"""
System matrix at a given time index
.. math::
(\mathbf{I} + \mathbf{dt} \mathbf{C} \mathbf{M_{\sigma}^e}^{-1}
\mathbf{C}^{\\top} \mathbf{M_{\mu^{-1}}^f})
"""
assert tInd >= 0 and tInd < self.nT
dt = self.timeSteps[tInd]
C = self.mesh.edgeCurl
MeSigmaI = self.MeSigmaI
MfMui = self.MfMui
I = Utils.speye(self.mesh.nF)
A = 1./dt * I + ( C * ( MeSigmaI * (C.T * MfMui ) ) )
if self._makeASymmetric is True:
return MfMui.T * A
return A
def getAdiag(self, tInd):
"""
System matrix at a given time index
.. math::
(\mathbf{I} + \mathbf{dt} \mathbf{C} \mathbf{M_{\sigma}^e}^{-1}
\mathbf{C}^{\\top} \mathbf{M_{\mu^{-1}}^f})
"""
assert tInd >= 0 and tInd < self.nT
dt = self.timeSteps[tInd]
C = self.mesh.edgeCurl
MeSigmaI = self.MeSigmaI
MfMui = self.MfMui
I = Utils.speye(self.mesh.nF)
A = 1. / dt * I + (C * (MeSigmaI * (C.T * MfMui)))
if self._makeASymmetric is True:
return MfMui.T * A
return A
def run(plotIt=True, n=60):
"""
Mesh: Operators: Cahn Hilliard
==============================
This example is based on the example in the FiPy_ library.
Please see their documentation for more information about the
Cahn-Hilliard equation.
The "Cahn-Hilliard" equation separates a field \\\\( \\\\phi \\\\)
into 0 and 1 with smooth transitions.
.. math::
\\frac{\partial \phi}{\partial t} = \\nabla \cdot D \\nabla \left( \\frac{\partial f}{\partial \phi} - \epsilon^2 \\nabla^2 \phi \\right)
Where \\\\( f \\\\) is the energy function \\\\( f = ( a^2 / 2 )\\\\phi^2(1 - \\\\phi)^2 \\\\)
which drives \\\\( \\\\phi \\\\) towards either 0 or 1, this competes with the term
\\\\(\\\\epsilon^2 \\\\nabla^2 \\\\phi \\\\) which is a diffusion term that creates smooth changes in \\\\( \\\\phi \\\\).
The equation can be factored:
.. math::
\\frac{\partial \phi}{\partial t} = \\nabla \cdot D \\nabla \psi \\\\
\psi = \\frac{\partial^2 f}{\partial \phi^2} (\phi - \phi^{\\text{old}}) + \\frac{\partial f}{\partial \phi} - \epsilon^2 \\nabla^2 \phi
Here we will need the derivatives of \\\\( f \\\\):
.. math::
\\frac{\partial f}{\partial \phi} = (a^2/2)2\phi(1-\phi)(1-2\phi)
\\frac{\partial^2 f}{\partial \phi^2} = (a^2/2)2[1-6\phi(1-\phi)]
The implementation below uses backwards Euler in time with an
exponentially increasing time step. The initial \\\\( \\\\phi \\\\)
is a normally distributed field with a standard deviation of 0.1 and
mean of 0.5. The grid is 60x60 and takes a few seconds to solve ~130
times. The results are seen below, and you can see the field separating
as the time increases.
.. _FiPy: http://www.ctcms.nist.gov/fipy/examples/cahnHilliard/generated/examples.cahnHilliard.mesh2DCoupled.html
"""
np.random.seed(5)
# Here we are going to rearrange the equations:
# (phi_ - phi)/dt = A*(d2fdphi2*(phi_ - phi) + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*(d2fdphi2*phi_ - d2fdphi2*phi + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*d2fdphi2*phi_ + A*( - d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - phi = dt*A*d2fdphi2*phi_ + dt*A*(- d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - dt*A*d2fdphi2 * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*dfdphi - dt*A*d2fdphi2*phi - dt*A*L*phi_ + phi
# (dt*A*d2fdphi2 - I) * phi_ = dt*A*d2fdphi2*phi + dt*A*L*phi_ - phi - dt*A*dfdphi
# (dt*A*d2fdphi2 - I - dt*A*L) * phi_ = (dt*A*d2fdphi2 - I)*phi - dt*A*dfdphi
h = [(0.25, n)]
M = Mesh.TensorMesh([h, h])
# Constants
D = a = epsilon = 1.
I = Utils.speye(M.nC)
# Operators
A = D * M.faceDiv * M.cellGrad
L = epsilon**2 * M.faceDiv * M.cellGrad
duration = 75
elapsed = 0.
dexp = -5
phi = np.random.normal(loc=0.5, scale=0.01, size=M.nC)
ii, jj = 0, 0
PHIS = []
capture = np.logspace(-1, np.log10(duration), 8)
while elapsed < duration:
dt = min(100, np.exp(dexp))
elapsed += dt
dexp += 0.05
dfdphi = a**2 * 2 * phi * (1 - phi) * (1 - 2 * phi)
d2fdphi2 = Utils.sdiag(a**2 * 2 * (1 - 6 * phi * (1 - phi)))
MAT = (dt * A * d2fdphi2 - I - dt * A * L)
rhs = (dt * A * d2fdphi2 - I) * phi - dt * A * dfdphi
phi = Solver(MAT) * rhs
if elapsed > capture[jj]:
PHIS += [(elapsed, phi.copy())]
jj += 1
if ii % 10 == 0:
print(ii, elapsed)
ii += 1
if plotIt:
fig, axes = plt.subplots(2, 4, figsize=(14, 6))
axes = np.array(axes).flatten().tolist()
for ii, ax in zip(np.linspace(0, len(PHIS) - 1, len(axes)), axes):
ii = int(ii)
M.plotImage(PHIS[ii][1], ax=ax)
ax.axis('off')
ax.set_title('Elapsed Time: {0:4.1f}'.format(PHIS[ii][0]))
def run(plotIt=True, n=60):
np.random.seed(5)
# Here we are going to rearrange the equations:
# (phi_ - phi)/dt = A*(d2fdphi2*(phi_ - phi) + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*(d2fdphi2*phi_ - d2fdphi2*phi + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*d2fdphi2*phi_ + A*( - d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - phi = dt*A*d2fdphi2*phi_ + dt*A*(- d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - dt*A*d2fdphi2 * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*dfdphi - dt*A*d2fdphi2*phi - dt*A*L*phi_ + phi
# (dt*A*d2fdphi2 - I) * phi_ = dt*A*d2fdphi2*phi + dt*A*L*phi_ - phi - dt*A*dfdphi
# (dt*A*d2fdphi2 - I - dt*A*L) * phi_ = (dt*A*d2fdphi2 - I)*phi - dt*A*dfdphi
h = [(0.25, n)]
M = Mesh.TensorMesh([h, h])
# Constants
D = a = epsilon = 1.
I = Utils.speye(M.nC)
# Operators
A = D * M.faceDiv * M.cellGrad
L = epsilon**2 * M.faceDiv * M.cellGrad
duration = 75
elapsed = 0.
dexp = -5
phi = np.random.normal(loc=0.5, scale=0.01, size=M.nC)
ii, jj = 0, 0
PHIS = []
capture = np.logspace(-1, np.log10(duration), 8)
while elapsed < duration:
dt = min(100, np.exp(dexp))
elapsed += dt
dexp += 0.05
dfdphi = a**2 * 2 * phi * (1 - phi) * (1 - 2 * phi)
d2fdphi2 = Utils.sdiag(a**2 * 2 * (1 - 6 * phi * (1 - phi)))
MAT = (dt*A*d2fdphi2 - I - dt*A*L)
rhs = (dt*A*d2fdphi2 - I)*phi - dt*A*dfdphi
phi = Solver(MAT)*rhs
if elapsed > capture[jj]:
PHIS += [(elapsed, phi.copy())]
jj += 1
if ii % 10 == 0:
print(ii, elapsed)
ii += 1
if plotIt:
fig, axes = plt.subplots(2, 4, figsize=(14, 6))
axes = np.array(axes).flatten().tolist()
for ii, ax in zip(np.linspace(0, len(PHIS)-1, len(axes)), axes):
ii = int(ii)
M.plotImage(PHIS[ii][1], ax=ax)
ax.axis('off')
ax.set_title('Elapsed Time: {0:4.1f}'.format(PHIS[ii][0]))
def run(plotIt=True, n=60):
np.random.seed(5)
# Here we are going to rearrange the equations:
# (phi_ - phi)/dt = A*(d2fdphi2*(phi_ - phi) + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*(d2fdphi2*phi_ - d2fdphi2*phi + dfdphi - L*phi_)
# (phi_ - phi)/dt = A*d2fdphi2*phi_ + A*( - d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - phi = dt*A*d2fdphi2*phi_ + dt*A*(- d2fdphi2*phi + dfdphi - L*phi_)
# phi_ - dt*A*d2fdphi2 * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*(- d2fdphi2*phi + dfdphi - L*phi_) + phi
# (I - dt*A*d2fdphi2) * phi_ = dt*A*dfdphi - dt*A*d2fdphi2*phi - dt*A*L*phi_ + phi
# (dt*A*d2fdphi2 - I) * phi_ = dt*A*d2fdphi2*phi + dt*A*L*phi_ - phi - dt*A*dfdphi
# (dt*A*d2fdphi2 - I - dt*A*L) * phi_ = (dt*A*d2fdphi2 - I)*phi - dt*A*dfdphi
h = [(0.25, n)]
M = Mesh.TensorMesh([h, h])
# Constants
D = a = epsilon = 1.
I = Utils.speye(M.nC)
# Operators
A = D * M.faceDiv * M.cellGrad
L = epsilon**2 * M.faceDiv * M.cellGrad
duration = 75
elapsed = 0.
dexp = -5
phi = np.random.normal(loc=0.5, scale=0.01, size=M.nC)
ii, jj = 0, 0
PHIS = []
capture = np.logspace(-1, np.log10(duration), 8)
while elapsed < duration:
dt = min(100, np.exp(dexp))
elapsed += dt
dexp += 0.05
dfdphi = a**2 * 2 * phi * (1 - phi) * (1 - 2 * phi)
d2fdphi2 = Utils.sdiag(a**2 * 2 * (1 - 6 * phi * (1 - phi)))
MAT = (dt * A * d2fdphi2 - I - dt * A * L)
rhs = (dt * A * d2fdphi2 - I) * phi - dt * A * dfdphi
phi = Solver(MAT) * rhs
if elapsed > capture[jj]:
PHIS += [(elapsed, phi.copy())]
jj += 1
if ii % 10 == 0:
print(ii, elapsed)
ii += 1
if plotIt:
fig, axes = plt.subplots(2, 4, figsize=(14, 6))
axes = np.array(axes).flatten().tolist()
for ii, ax in zip(np.linspace(0, len(PHIS) - 1, len(axes)), axes):
ii = int(ii)
M.plotImage(PHIS[ii][1], ax=ax)
ax.axis('off')
ax.set_title('Elapsed Time: {0:4.1f}'.format(PHIS[ii][0]))
def get_grad_horizontal(
self, xy, hz, dim=3,
use_cell_weights=True,
minimum_distance=None
):
"""
Compute Gradient in horizontal direction using Delaunay
"""
if use_cell_weights:
self.cell_weights = np.tile(hz, (xy.shape[0], 1)).flatten()
if dim == 3:
tri = sp.spatial.Delaunay(xy)
# Split the triangulation into connections
edges = np.r_[
tri.simplices[:, :2],
tri.simplices[:, 1:],
tri.simplices[:, [0, 2]]
]
# Sort and keep uniques
edges = np.sort(edges, axis=1)
edges = np.unique(
edges[np.argsort(edges[:, 0]), :], axis=0
)
# Compute distance
if minimum_distance is not None:
dx = xy[edges[:, 0], 0]-xy[edges[:, 1], 0]
dy = xy[edges[:, 0], 1]-xy[edges[:, 1], 1]
distance = np.sqrt(dx**2+dy**2)
inds = distance < minimum_distance
edges = edges[inds, :]
# Create 2D operator, dimensionless for now
nN = edges.shape[0]
nStn = xy.shape[0]
stn, count = np.unique(edges[:, 0], return_counts=True)
col = []
row = []
dm = []
avg = []
for ii in range(nN):
row += [ii]*2
col += [edges[ii, 0], edges[ii, 1]]
scale = count[stn == edges[ii, 0]][0]
dm += [-1., 1.]
avg += [0.5, 0.5]
D = sp.sparse.csr_matrix((dm, (row, col)), shape=(nN, nStn))
A = sp.sparse.csr_matrix((avg, (row, col)), shape=(nN, nStn))
# Kron vertically for nCz
Grad = sp.sparse.kron(D, Utils.speye(hz.size))
Avg = sp.sparse.kron(A, Utils.speye(hz.size))
# Override the gradient operator in y-drection
# This is because of ordering ... See def get_2d_mesh
# y first then x
self.regmesh._cellDiffyStencil = self.regmesh.cellDiffxStencil.copy()
# Override the gradient operator in x-drection
self.regmesh._cellDiffxStencil = Grad
# Do the same for the averaging operator
self.regmesh._aveCC2Fy = self.regmesh.aveCC2Fx.copy()
self.regmesh._aveCC2Fx = Avg
self.regmesh._aveFy2CC = self.regmesh.aveFx2CC.copy()
self.regmesh._aveFx2CC = Avg.T
return tri
elif dim == 2:
# Override the gradient operator in y-drection
# This is because of ordering ... See def get_2d_mesh
# y first then x
temp_x = self.regmesh.cellDiffxStencil.copy()
temp_y = self.regmesh.cellDiffyStencil.copy()
self.regmesh._cellDiffyStencil = temp_x
# Override the gradient operator in x-drection
self.regmesh._cellDiffxStencil = temp_y
# Do the same for the averaging operator
temp_x = self.regmesh.aveCC2Fx.copy()
temp_y = self.regmesh.aveCC2Fy.copy()
self.regmesh._aveCC2Fy = temp_x
self.regmesh._aveCC2Fx = temp_y
temp_x = self.regmesh.aveCC2Fx.copy()
temp_y = self.regmesh.aveCC2Fy.copy()
self.regmesh._aveFy2CC = temp_x
self.regmesh._aveFx2CC = temp_y
return True
vec = np.c_[mx, my, mz]
nC = mesh.nC
atp = Utils.matutils.xyz2atp(vec)
theta = atp[nC:2 * nC]
phi = atp[2 * nC:]
# theta = np.ones(mesh.nC) * 45
# phi = np.ones(mesh.nC) * 0
indActive = np.zeros(mesh.nC, dtype=bool)
indActive[actv] = True
print("Building operators")
Pac = Utils.speye(mesh.nC)[:, indActive]
Dx1 = Regularization.getDiffOpRot(mesh, np.deg2rad(0.), theta, phi, 'X')
Dy1 = Regularization.getDiffOpRot(mesh, np.deg2rad(0.), theta, phi, 'Y')
Dz1 = Regularization.getDiffOpRot(mesh, np.deg2rad(0.), theta, phi, 'Z')
Dx1 = Pac.T * Dx1 * Pac
Dy1 = Pac.T * Dy1 * Pac
Dz1 = Pac.T * Dz1 * Pac
Dx2 = Regularization.getDiffOpRot(mesh,
np.deg2rad(0.),
theta,
phi,
'X',
forward=False)
def get_grad_horizontal(
self, xy, hz, dim=3,
use_cell_weights=True,
minimum_distance=None
):
"""
Compute Gradient in horizontal direction using Delaunay
"""
self.cell_weights = np.tile(hz, (xy.shape[0], 1)).flatten()
if dim == 3:
tri = sp.spatial.Delaunay(xy)
# Split the triangulation into connections
edges = np.r_[
tri.simplices[:, :2],
tri.simplices[:, 1:],
tri.simplices[:, [0, 2]]
]
# Sort and keep uniques
edges = np.sort(edges, axis=1)
edges = np.unique(
edges[np.argsort(edges[:, 0]), :], axis=0
)
# Compute distance
if minimum_distance is not None:
dx = xy[edges[:, 0], 0]-xy[edges[:, 1], 0]
dy = xy[edges[:, 0], 1]-xy[edges[:, 1], 1]
distance = np.sqrt(dx**2+dy**2)
inds = distance < minimum_distance
edges = edges[inds, :]
# Create 2D operator, dimensionless for now
nN = edges.shape[0]
nStn = xy.shape[0]
stn, count = np.unique(edges[:, 0], return_counts=True)
col = []
row = []
dm = []
avg = []
for ii in range(nN):
row += [ii]*2
col += [edges[ii, 0], edges[ii, 1]]
scale = count[stn == edges[ii, 0]][0]
dm += [-1., 1.]
avg += [0.5, 0.5]
D = sp.sparse.csr_matrix((dm, (row, col)), shape=(nN, nStn))
A = sp.sparse.csr_matrix((avg, (row, col)), shape=(nN, nStn))
# Kron vertically for nCz
Grad = sp.sparse.kron(D, Utils.speye(hz.size))
Avg = sp.sparse.kron(A, Utils.speye(hz.size))
# Override the gradient operator in y-drection
# This is because of ordering ... See def get_2d_mesh
# y first then x
self.regmesh._cellDiffyStencil = self.regmesh.cellDiffxStencil.copy()
# Override the gradient operator in x-drection
self.regmesh._cellDiffxStencil = Grad
# Do the same for the averaging operator
self.regmesh._aveCC2Fy = self.regmesh.aveCC2Fx.copy()
self.regmesh._aveCC2Fx = Avg
self.regmesh._aveFy2CC = self.regmesh.aveFx2CC.copy()
self.regmesh._aveFx2CC = Avg.T
return tri
elif dim == 2:
# Override the gradient operator in y-drection
# This is because of ordering ... See def get_2d_mesh
# y first then x
temp_x = self.regmesh.cellDiffxStencil.copy()
temp_y = self.regmesh.cellDiffyStencil.copy()
self.regmesh._cellDiffyStencil = temp_x
# Override the gradient operator in x-drection
self.regmesh._cellDiffxStencil = temp_y
# Do the same for the averaging operator
temp_x = self.regmesh.aveCC2Fx.copy()
temp_y = self.regmesh.aveCC2Fy.copy()
self.regmesh._aveCC2Fy = temp_x
self.regmesh._aveCC2Fx = temp_y
temp_x = self.regmesh.aveCC2Fx.copy()
temp_y = self.regmesh.aveCC2Fy.copy()
self.regmesh._aveFy2CC = temp_x
self.regmesh._aveFx2CC = temp_y
return True
