def _assemble(self):
coordinates = np.array(np.zeros(self.mesh.num_vertices()),
dtype=[('x', float), ('y', float)])
for i, vertex in enumerate(df.vertices(self.mesh)):
coordinates['x'][i] = vertex.x(0)
coordinates['y'][i] = vertex.x(1)
self._Ny = len(np.unique(coordinates['y']))
self._Nx = len(np.unique(coordinates['x']))
coordinates = np.sort(coordinates, order=['y', 'x'])
X = coordinates['x'].reshape(self._Ny, self._Nx)
Y = coordinates['y'].reshape(self._Ny, self._Nx)
self._X = np.flipud(X)
self._Y = np.flipud(Y)
# check if uniform mesh
T = np.diff(X, axis=1)
self._dx = T[0, 0]
assert (np.all(np.abs(T - self._dx) < 1e-12))
T = np.diff(Y, axis=0)
self._dy = T[0, 0]
assert (np.all(np.abs(T - self._dy) < 1e-12))
Interpolator = np.zeros(
((self._Ny - 1) * (self._Nx - 1), self.V.dim()))
for i, cell in enumerate(df.cells(self.mesh)):
x = cell.midpoint().x()
y = cell.midpoint().y()
cx = int(x // self._dx)
cy = int(y // self._dy)
cy = (self._Ny - 2) - cy
pixel_id = cy * (self._Ny - 1) + cx
Interpolator[pixel_id, i] = 0.5
ReverseInterpolator = np.zeros(
(self.V.dim(), (self._Ny - 1) * (self._Nx - 1)))
for i, row in enumerate(Interpolator):
ind = np.where(row)[0]
ReverseInterpolator[ind[0], i] = 1
ReverseInterpolator[ind[1], i] = 1
self.Interpolator = Interpolator
self.ReverseInterpolator = ReverseInterpolator
#
a, b = np.where(self.Interpolator != 0)
self._DofToPixelPermutator = torch.tensor(b, dtype=torch.long)
a, b = self.ReverseInterpolator.nonzero()
self._PixelToDofPermutator = torch.tensor(b, dtype=torch.long)
def refine_near_left_boundary(mesh, cycles):
""" Refine mesh near the left boundary.
The usual approach of using SubDomain and EdgeFunction isn't appearing to work
in 1D, so I'm going to just loop through the cells of the mesh and set markers manually.
"""
for i in range(cycles):
cell_markers = fenics.CellFunction("bool", mesh)
cell_markers.set_all(False)
for cell in fenics.cells(mesh):
found_left_boundary = False
for vertex in fenics.vertices(cell):
if fenics.near(vertex.x(0), 0.):
found_left_boundary = True
if found_left_boundary:
cell_markers[cell] = True
break # There should only be one such point.
mesh = fenics.refine(mesh, cell_markers)
return mesh
def refine_initial_mesh(self):
""" Locally refine near the hot boundary """
for i in range(self.initial_hot_boundary_refinement_cycles):
cell_markers = fenics.MeshFunction("bool", self.mesh,
self.mesh.topology().dim(),
False)
cell_markers.set_all(False)
for cell in fenics.cells(self.mesh):
found_left_boundary = False
for vertex in fenics.vertices(cell):
if fenics.near(vertex.x(0), 0.):
found_left_boundary = True
break
if found_left_boundary:
cell_markers[cell] = True
break # There should only be one such point in 1D.
self.mesh = fenics.refine(
self.mesh, cell_markers) # Does this break references?
def initial_mesh(self):
self.initial_hot_wall_refinement_cycles = 8
mesh = self.coarse_mesh()
for i in range(self.initial_hot_wall_refinement_cycles):
cell_markers = fenics.MeshFunction("bool", mesh,
mesh.topology().dim(), False)
cell_markers.set_all(False)
for cell in fenics.cells(mesh):
found_left_boundary = False
for vertex in fenics.vertices(cell):
if fenics.near(vertex.x(0), 0.):
found_left_boundary = True
break
if found_left_boundary:
cell_markers[cell] = True
break # There should only be one such point in 1D.
mesh = fenics.refine(mesh, cell_markers)
return mesh
def refine_initial_mesh(self):
""" Replace 2D refinement method with 3D method. Perhaps one could make an n-dimensional method. """
for i in range(self.initial_hot_wall_refinement_cycles):
cell_markers = fenics.MeshFunction("bool", self.mesh,
self.mesh.topology().dim(),
False)
for cell in fenics.cells(self.mesh):
found_left_boundary = False
for vertex in fenics.vertices(cell):
if fenics.near(vertex.x(0), 0.):
found_left_boundary = True
break
if found_left_boundary:
cell_markers[cell] = True
self.mesh = fenics.refine(self.mesh, cell_markers)
def integrate_field(self, fn_spec, specific, fn_main, r=20, val=0.0):
"""
Assimilate a field with filename <fn_spec> from DataInput object
<specific> into this DataInput's field with filename <fn_main>. The
parameter <val> should be set to the specific dataset's value for
undefined regions, default is 0.0. <r> is a parameter used to eliminate
border artifacts from interpolation; increase this value to eliminate edge
noise.
"""
print "::: integrating %s field from %s :::" % (fn_spec, specific.name)
# get the dofmap to map from mesh vertex indices to function indicies :
df = self.func_space.dofmap()
dfmap = df.vertex_to_dof_map(self.mesh)
unew = self.get_projection(fn_main) # existing dataset projection
uocom = unew.compute_vertex_values() # mesh indexed main vertex values
uspec = specific.get_projection(fn_spec) # specific dataset projection
uscom = uspec.compute_vertex_values() # mesh indexed spec vertex values
d = float64(specific.data[fn_spec]) # original matlab spec dataset
# get arrays of x-values for specific domain
xs = specific.x
ys = specific.y
nx = specific.nx
ny = specific.ny
for v in vertices(self.mesh):
# mesh vertex x,y coordinate :
i = v.index()
p = v.point()
x = p.x()
y = p.y()
# indexes of closest datapoint to specific dataset's x and y domains :
idx = abs(xs - x).argmin()
idy = abs(ys - y).argmin()
# data value for closest value and square around the value in question :
dv = d[idy, idx]
db = d[max(0,idy-r) : min(ny, idy+r), max(0, idx-r) : min(nx, idx+r)]
# if the vertex is in the domain of the specific dataset, and the value
# of the dataset at this point is not abov <val>, set the array value
# of the main file to this new specific region's value.
if dv > val:
#print "found:", x, y, idx, idy, v.index()
# if the values is not near an edge, make the value equal to the
# nearest specific region's dataset value, otherwise, use the
# specific region's projected value :
if all(db > val):
uocom[i] = uscom[i]
else :
uocom[i] = dv
# set the values of the projected original dataset equal to the assimilated
# dataset :
unew.vector().set_local(uocom[dfmap])
return unew
def meshReader(dictMesh):
''' converts json mesh to fenics mesh and apply the boundary conditions.
dict:: dictMesh, the mesh from database in the form of a python dictionary
output:
dolfin.ccp.mesh:: feMesh, mesh defined in fenics object
'''
# conver dict to object
mesh = namedtuple("mesh", dictMesh.keys())(*dictMesh.values())
nodes = np.array(mesh.nodes)
cells = np.array(mesh.connectivity, dtype=np.uintp)
feMesh = fn.Mesh()
editor = fn.MeshEditor()
# cell type, topological, and geometrical dimensions. i.e. 2, 3 for 3d surface mesh
editor.open(feMesh, "triangle", 2, 3)
# cell types available: point, interval, triangle, quadrilateral, tetrahedron, hexahedron
editor.init_vertices(len(mesh.nodes))
editor.init_cells(len(mesh.connectivity))
[editor.add_vertex(i, n) for i, n in enumerate(nodes)]
[editor.add_cell(i, n - 1) for i, n in enumerate(cells)]
editor.close()
# construct faces
faceSets = fn.MeshFunction('size_t', feMesh, 2)
for face, faceCells in mesh.faces.items():
for index in faceCells:
faceSets.set_value(index, int(face), feMesh)
# construct edges
edgeSets = fn.MeshFunction('size_t', feMesh, 1)
meshEdges = {}
for edge in fn.edges(feMesh):
meshEdges[edge.index()] = []
for vert in fn.vertices(edge):
meshEdges[edge.index()].append(vert.index())
for edge, nodes in meshEdges.items():
for edgeName, edgeList in mesh.edges.items():
if nodes in edgeList:
edgeSets.set_value(edge, int(edgeName), feMesh)
# construct points
pointSets = fn.MeshFunction('size_t', feMesh, 0)
for pointName, point in mesh.points.items():
pointSets.set_value(point, int(pointName), feMesh)
return dict(feMesh=feMesh,
faceSets=faceSets,
edgeSets=edgeSets,
pointSets=pointSets)
def solve_poisson_with_fem(lightweight=False):
# Create mesh and define function space
mesh = fs.UnitSquareMesh(8, 8)
V = fs.FunctionSpace(mesh, 'P', 1)
# Define boundary condition
u_code = 'x[0] + 2*x[1] + 1'
u_D = fs.Expression(u_code, degree=2)
def boundary(x, on_boundary):
return on_boundary
bc = fs.DirichletBC(V, u_D, boundary)
# Define variational problem
u = fs.Function(V) # Note: not TrialFunction!
v = fs.TestFunction(V)
# f = fs.Expression(f_code, degree=2)
f_code = '-10*x[0] - 20*x[1] - 10'
f = fs.Expression(f_code, degree=2)
F = q(u) * fs.dot(fs.grad(u), fs.grad(v)) * fs.dx - f * v * fs.dx
# Compute solution
fs.solve(F == 0, u, bc)
# Plot solution
fs.plot(u)
# Compute maximum error at vertices. This computation illustrates
# an alternative to using compute_vertex_values as in poisson.py.
u_e = fs.interpolate(u_D, V)
# Restore numpy object
image1d = np.empty((81, ), dtype=np.float)
for v in fs.vertices(mesh):
image1d[v.index()] = u(*mesh.coordinates()[v.index()])
if not lightweight:
error_max = np.abs(u_e.vector().get_local() -
u.vector().get_local()).max()
print('error_max = ', error_max)
fs.plot(u)
plt.show()
save_contour(image1d, 1.0, 1.0, 'poisson')
return image1d
def get_nearest(self, fn):
"""
returns a dolfin Function object with values given by interpolated
nearest-neighbor data <fn>.
"""
#FIXME: get to work with a change of projection.
# get the dofmap to map from mesh vertex indices to function indicies :
df = self.func_space.dofmap()
dfmap = df.vertex_to_dof_map(self.mesh)
unew = Function(self.func_space) # existing dataset projection
uocom = unew.vector().array() # mesh indexed main vertex values
d = float64(self.data[fn]) # original matlab spec dataset
# get arrays of x-values for specific domain
xs = self.x
ys = self.y
for v in vertices(self.mesh):
# mesh vertex x,y coordinate :
i = v.index()
p = v.point()
x = p.x()
y = p.y()
# indexes of closest datapoint to specific dataset's x and y domains :
idx = abs(xs - x).argmin()
idy = abs(ys - y).argmin()
# data value for closest value :
dv = d[idy, idx]
if dv > 0:
dv = 1.0
uocom[i] = dv
# set the values of the empty function's vertices to the data values :
unew.vector().set_local(uocom[dfmap])
return unew
def get_nearest(self, fn):
"""
returns a dolfin Function object with values given by interpolated
nearest-neighbor data <fn>.
"""
#FIXME: get to work with a change of projection.
# get the dofmap to map from mesh vertex indices to function indicies :
df = self.func_space.dofmap()
dfmap = df.vertex_to_dof_map(self.mesh)
unew = Function(self.func_space) # existing dataset projection
uocom = unew.vector().array() # mesh indexed main vertex values
d = float64(self.data[fn]) # original matlab spec dataset
# get arrays of x-values for specific domain
xs = self.x
ys = self.y
for v in vertices(self.mesh):
# mesh vertex x,y coordinate :
i = v.index()
p = v.point()
x = p.x()
y = p.y()
# indexes of closest datapoint to specific dataset's x and y domains :
idx = abs(xs - x).argmin()
idy = abs(ys - y).argmin()
# data value for closest value :
dv = d[idy, idx]
if dv > 0:
dv = 1.0
uocom[i] = dv
# set the values of the empty function's vertices to the data values :
unew.vector().set_local(uocom[dfmap])
return unew
def sd_nodenode(mesh, V, u_n, De, nexp):
"""
SD node-to-node
Flow routing from node-to-node based on the steepest route of descent
:param mesh: mesh object generated using mshr (fenics)
:param V: finite element function space
:param u_n: solution (trial function) for water flux
:param De: dimensionless diffusion coefficient
:param nexp: water flux exponent
:return:
"""
# get the global coordinates
gdim = mesh.geometry().dim()
if dolfin.dolfin_version() == '1.6.0':
dofmap = V.dofmap()
gc = dofmap.tabulate_all_coordinates(mesh).reshape((-1, gdim))
else:
gc = V.tabulate_dof_coordinates().reshape((-1, gdim))
vtd = vertex_to_dof_map(V)
# first get the elevation of each vertex
elevation = np.zeros(len(gc))
elevation = u_n.compute_vertex_values(mesh)
# loop to get the local flux
mesh.init(0, 1)
flux = np.zeros(len(gc))
neighbors = []
for v in vertices(mesh):
idx = v.index()
# get the local neighbourhood
neighborhood = [Edge(mesh, i).entities(0) for i in v.entities(1)]
neighborhood = np.array(neighborhood).flatten()
# Remove own index from neighborhood
neighborhood = neighborhood[np.where(neighborhood != idx)[0]]
neighbors.append(neighborhood)
# get location
xh = v.x(0)
yh = v.x(1)
# get distance to neighboring vertices
length = np.zeros(len(neighborhood))
weight = np.zeros(len(neighborhood))
i = 0
for vert in neighborhood:
nidx = vtd[vert]
xn = gc[nidx, 0]
yn = gc[nidx, 1]
length[i] = np.sqrt((xh - xn) * (xh - xn) + (yh - yn) * (yh - yn))
flux[vert] = length[i]
# weight[i] = elevation[idx] - elevation[vert]
# # downhill only
# if weight[i] < 0:
# weight[i] = 0
# i += 1
#
# # find steepest slope
# steepest = len(neighborhood)+2
# if max(weight) > 0:
# steepest = np.argmax(weight)
# else:
# weight[:] = 0
# i = 0
# for vert in neighborhood:
# if i == steepest:
# weight[i] = 1
# else:
# weight[i] = 0
# flux[vert] = flux[vert] + length[i]*weight[i]
# i += 1
# sort from top to botton
sortedidx = np.argsort(-elevation)
# accumulate fluxes from top to bottom
for idx in sortedidx:
neighborhood = neighbors[idx]
weight = np.zeros(len(neighborhood))
i = 0
for vert in neighborhood:
weight[i] = elevation[idx] - elevation[vert]
# downhill only
if weight[i] < 0:
weight[i] = 0
i += 1
# find steepest slope
steepest = len(neighborhood) + 2
if max(weight) > 0:
steepest = np.argmax(weight)
else:
weight[:] = 0
i = 0
for vert in neighborhood:
if i == steepest:
weight[i] = 1
else:
weight[i] = 0
flux[vert] = flux[vert] + flux[idx] * weight[i]
i += 1
# calculate the diffusion coefficient
q0 = 1 + De * pow(flux, nexp)
q = Function(V)
q.vector()[:] = q0[dof_to_vertex_map(V)]
return (q)
def solve_heat_with_fem(lightweight=False):
T = 2.0 # final time
num_steps = 100 # number of time steps
dt = T / num_steps # time step size
alpha = 3 # parameter alpha
beta = 1.2 # parameter beta
# Create mesh and define function space
nx = ny = 8
mesh = fs.UnitSquareMesh(nx, ny)
V = fs.FunctionSpace(mesh, 'P', 1)
# Define boundary condition
u_D = fs.Expression('1 + x[0]*x[0] + alpha*x[1]*x[1] + beta*t',
degree=2,
alpha=alpha,
beta=beta,
t=0)
bc = fs.DirichletBC(V, u_D, boundary)
# Define initial value
u_n = fs.interpolate(u_D, V)
# u_n = project(u_D, V)
# Define variational problem
u = fs.TrialFunction(V)
v = fs.TestFunction(V)
f = fs.Constant(beta - 2 - 2 * alpha)
F = u * v * fs.dx + dt * fs.dot(
fs.grad(u), fs.grad(v)) * fs.dx - (u_n + dt * f) * v * fs.dx
a, L = fs.lhs(F), fs.rhs(F)
# Time-stepping
u = fs.Function(V)
t = 0
images1d = []
for n in range(num_steps):
# Update current time
t += dt
u_D.t = t
# Compute solution
fs.solve(a == L, u, bc)
# Restore numpy object
image1d = np.empty((81, ), dtype=np.float)
for v in fs.vertices(mesh):
image1d[v.index()] = u(*mesh.coordinates()[v.index()])
images1d.append(image1d)
if not lightweight:
# Compute error at vertices
u_e = fs.interpolate(u_D, V)
error = np.abs(u_e.vector().get_local() -
u.vector().get_local()).max()
print('t = %.2f: error = %.3g' % (t, error))
# Update previous solution
u_n.assign(u)
# Plotting
if not lightweight:
fs.plot(u)
plt.show()
save_dynamic_contours(images1d, 1.0, 1.0, 'heat2d')
return images1d
