def get_extension_bie_symbolic_operator(loc_sign=1):
"""
loc_sign:
-1 for interior Dirichlet
+1 for exterior Dirichlet
"""
logger = logging.getLogger("SETUP")
logger.info(locals())
dim = 2
cse = sym.cse
sigma_sym = sym.make_sym_vector("sigma", dim)
int_sigma = sym.Ones() * sym.integral(2, 1, sigma_sym)
nvec_sym = sym.make_sym_vector("normal", dim)
mu_sym = sym.var("mu")
stresslet_obj = StressletWrapper(dim=dim)
stokeslet_obj = StokesletWrapper(dim=dim)
bdry_op_sym = (
loc_sign * 0.5 * sigma_sym - stresslet_obj.apply(
sigma_sym, nvec_sym, mu_sym, qbx_forced_limit='avg') -
stokeslet_obj.apply(sigma_sym, mu_sym, qbx_forced_limit='avg') +
int_sigma)
return bdry_op_sym
def _operator(self, sigma, normal, mu, qbx_forced_limit):
slp_qbx_forced_limit = qbx_forced_limit
if slp_qbx_forced_limit == "avg":
slp_qbx_forced_limit = +1
# NOTE: we set a dofdesc here to force the evaluation of this integral
# on the source instead of the target when using automatic tagging
# see :meth:`pytential.symbolic.mappers.LocationTagger._default_dofdesc`
dd = sym.DOFDescriptor(None, discr_stage=sym.QBX_SOURCE_STAGE1)
int_sigma = sym.integral(self.ambient_dim, self.dim, sigma, dofdesc=dd)
meanless_sigma = sym.cse(sigma - sym.mean(self.ambient_dim, self.dim, sigma))
op_k = self.stresslet.apply(sigma, normal, mu,
qbx_forced_limit=qbx_forced_limit)
op_s = (
self.alpha / (2.0 * np.pi) * int_sigma
- self.stokeslet.apply(meanless_sigma, mu,
qbx_forced_limit=slp_qbx_forced_limit)
)
return op_k + self.eta * op_s
def _farfield(self, mu, qbx_forced_limit):
length = sym.integral(self.ambient_dim, self.dim, 1)
return self.stokeslet.apply(
-self.omega / length,
mu,
qbx_forced_limit=qbx_forced_limit)
def run_exterior_stokes_2d(ctx_factory,
nelements,
mesh_order=4,
target_order=4,
qbx_order=4,
fmm_order=10,
mu=1,
circle_rad=1.5,
do_plot=False):
# This program tests an exterior Stokes flow in 2D using the
# compound representation given in Hsiao & Kress,
# ``On an integral equation for the two-dimensional exterior Stokes problem,''
# Applied Numerical Mathematics 1 (1985).
logging.basicConfig(level=logging.INFO)
cl_ctx = cl.create_some_context()
queue = cl.CommandQueue(cl_ctx)
ovsmp_target_order = 4 * target_order
from meshmode.mesh.generation import ( # noqa
make_curve_mesh, starfish, ellipse, drop)
mesh = make_curve_mesh(lambda t: circle_rad * ellipse(1, t),
np.linspace(0, 1, nelements + 1), target_order)
coarse_density_discr = Discretization(
cl_ctx, mesh, InterpolatoryQuadratureSimplexGroupFactory(target_order))
from pytential.qbx import QBXLayerPotentialSource
target_association_tolerance = 0.05
qbx, _ = QBXLayerPotentialSource(
coarse_density_discr,
fine_order=ovsmp_target_order,
qbx_order=qbx_order,
fmm_order=fmm_order,
target_association_tolerance=target_association_tolerance,
_expansions_in_tree_have_extent=True,
).with_refinement()
density_discr = qbx.density_discr
normal = bind(density_discr, sym.normal(2).as_vector())(queue)
path_length = bind(density_discr, sym.integral(2, 1, 1))(queue)
# {{{ describe bvp
from pytential.symbolic.stokes import StressletWrapper, StokesletWrapper
dim = 2
cse = sym.cse
sigma_sym = sym.make_sym_vector("sigma", dim)
meanless_sigma_sym = cse(sigma_sym - sym.mean(2, 1, sigma_sym))
int_sigma = sym.Ones() * sym.integral(2, 1, sigma_sym)
nvec_sym = sym.make_sym_vector("normal", dim)
mu_sym = sym.var("mu")
# -1 for interior Dirichlet
# +1 for exterior Dirichlet
loc_sign = 1
stresslet_obj = StressletWrapper(dim=2)
stokeslet_obj = StokesletWrapper(dim=2)
bdry_op_sym = (-loc_sign * 0.5 * sigma_sym - stresslet_obj.apply(
sigma_sym, nvec_sym, mu_sym, qbx_forced_limit='avg') +
stokeslet_obj.apply(
meanless_sigma_sym, mu_sym, qbx_forced_limit='avg') -
(0.5 / np.pi) * int_sigma)
# }}}
bound_op = bind(qbx, bdry_op_sym)
# {{{ fix rhs and solve
def fund_soln(x, y, loc, strength):
#with direction (1,0) for point source
r = cl.clmath.sqrt((x - loc[0])**2 + (y - loc[1])**2)
scaling = strength / (4 * np.pi * mu)
xcomp = (-cl.clmath.log(r) + (x - loc[0])**2 / r**2) * scaling
ycomp = ((x - loc[0]) * (y - loc[1]) / r**2) * scaling
return [xcomp, ycomp]
def rotlet_soln(x, y, loc):
r = cl.clmath.sqrt((x - loc[0])**2 + (y - loc[1])**2)
xcomp = -(y - loc[1]) / r**2
ycomp = (x - loc[0]) / r**2
return [xcomp, ycomp]
def fund_and_rot_soln(x, y, loc, strength):
#with direction (1,0) for point source
r = cl.clmath.sqrt((x - loc[0])**2 + (y - loc[1])**2)
scaling = strength / (4 * np.pi * mu)
xcomp = ((-cl.clmath.log(r) + (x - loc[0])**2 / r**2) * scaling -
(y - loc[1]) * strength * 0.125 / r**2 + 3.3)
ycomp = (((x - loc[0]) * (y - loc[1]) / r**2) * scaling +
(x - loc[0]) * strength * 0.125 / r**2 + 1.5)
return [xcomp, ycomp]
nodes = density_discr.nodes().with_queue(queue)
fund_soln_loc = np.array([0.5, -0.2])
strength = 100.
bc = fund_and_rot_soln(nodes[0], nodes[1], fund_soln_loc, strength)
omega_sym = sym.make_sym_vector("omega", dim)
u_A_sym_bdry = stokeslet_obj.apply( # noqa: N806
omega_sym, mu_sym, qbx_forced_limit=1)
omega = [
cl.array.to_device(queue,
(strength / path_length) * np.ones(len(nodes[0]))),
cl.array.to_device(queue, np.zeros(len(nodes[0])))
]
bvp_rhs = bind(qbx,
sym.make_sym_vector("bc", dim) + u_A_sym_bdry)(queue,
bc=bc,
mu=mu,
omega=omega)
gmres_result = gmres(bound_op.scipy_op(queue,
"sigma",
np.float64,
mu=mu,
normal=normal),
bvp_rhs,
x0=bvp_rhs,
tol=1e-9,
progress=True,
stall_iterations=0,
hard_failure=True)
# }}}
# {{{ postprocess/visualize
sigma = gmres_result.solution
sigma_int_val_sym = sym.make_sym_vector("sigma_int_val", 2)
int_val = bind(qbx, sym.integral(2, 1, sigma_sym))(queue, sigma=sigma)
int_val = -int_val / (2 * np.pi)
print("int_val = ", int_val)
u_A_sym_vol = stokeslet_obj.apply( # noqa: N806
omega_sym, mu_sym, qbx_forced_limit=2)
representation_sym = (
-stresslet_obj.apply(sigma_sym, nvec_sym, mu_sym, qbx_forced_limit=2) +
stokeslet_obj.apply(meanless_sigma_sym, mu_sym, qbx_forced_limit=2) -
u_A_sym_vol + sigma_int_val_sym)
nsamp = 30
eval_points_1d = np.linspace(-3., 3., nsamp)
eval_points = np.zeros((2, len(eval_points_1d)**2))
eval_points[0, :] = np.tile(eval_points_1d, len(eval_points_1d))
eval_points[1, :] = np.repeat(eval_points_1d, len(eval_points_1d))
def circle_mask(test_points, radius):
return (test_points[0, :]**2 + test_points[1, :]**2 > radius**2)
def outside_circle(test_points, radius):
mask = circle_mask(test_points, radius)
return np.array([row[mask] for row in test_points])
eval_points = outside_circle(eval_points, radius=circle_rad)
from pytential.target import PointsTarget
vel = bind((qbx, PointsTarget(eval_points)),
representation_sym)(queue,
sigma=sigma,
mu=mu,
normal=normal,
sigma_int_val=int_val,
omega=omega)
print("@@@@@@@@")
fplot = FieldPlotter(np.zeros(2), extent=6, npoints=100)
plot_pts = outside_circle(fplot.points, radius=circle_rad)
plot_vel = bind((qbx, PointsTarget(plot_pts)),
representation_sym)(queue,
sigma=sigma,
mu=mu,
normal=normal,
sigma_int_val=int_val,
omega=omega)
def get_obj_array(obj_array):
return make_obj_array([ary.get() for ary in obj_array])
exact_soln = fund_and_rot_soln(cl.array.to_device(queue, eval_points[0]),
cl.array.to_device(queue, eval_points[1]),
fund_soln_loc, strength)
vel = get_obj_array(vel)
err = vel - get_obj_array(exact_soln)
# FIXME: Pointwise relative errors don't make sense!
rel_err = err / (get_obj_array(exact_soln))
if 0:
print("@@@@@@@@")
print("vel[0], err[0], rel_err[0] ***** vel[1], err[1], rel_err[1]: ")
for i in range(len(vel[0])):
print("%15.8e, %15.8e, %15.8e ***** %15.8e, %15.8e, %15.8e\n" %
(vel[0][i], err[0][i], rel_err[0][i], vel[1][i], err[1][i],
rel_err[1][i]))
print("@@@@@@@@")
l2_err = np.sqrt((6. / (nsamp - 1))**2 * np.sum(err[0] * err[0]) +
(6. / (nsamp - 1))**2 * np.sum(err[1] * err[1]))
l2_rel_err = np.sqrt((6. /
(nsamp - 1))**2 * np.sum(rel_err[0] * rel_err[0]) +
(6. /
(nsamp - 1))**2 * np.sum(rel_err[1] * rel_err[1]))
print("L2 error estimate: ", l2_err)
print("L2 rel error estimate: ", l2_rel_err)
print("max error at sampled points: ", max(abs(err[0])), max(abs(err[1])))
print("max rel error at sampled points: ", max(abs(rel_err[0])),
max(abs(rel_err[1])))
if do_plot:
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
full_pot = np.zeros_like(fplot.points) * float("nan")
mask = circle_mask(fplot.points, radius=circle_rad)
for i, vel in enumerate(plot_vel):
full_pot[i, mask] = vel.get()
plt.quiver(fplot.points[0],
fplot.points[1],
full_pot[0],
full_pot[1],
linewidth=0.1)
plt.savefig("exterior-2d-field.pdf")
# }}}
h_max = bind(qbx, sym.h_max(qbx.ambient_dim))(queue)
return h_max, l2_err
def Wr(operand: sym.Expression) -> sym.Expression: # noqa: N802
return sym.Ones() * sym.integral(self.ambient_dim, self.dim,
operand)
def build_kernel_exterior_normalizer_table(self,
cl_ctx,
queue,
pool=None,
ncpus=None,
mesh_order=5,
quad_order=10,
mesh_size=0.03,
remove_tmp_files=True,
**kwargs):
r"""Build the kernel exterior normalizer table for fractional Laplacians.
An exterior normalizer for kernel :math:`G(r)` and target
:math:`x` is defined as
.. math::
\int_{B^c} G(\lVert x - y \rVert) dy
where :math:`B` is the source box :math:`[0, source_box_extent]^dim`.
"""
logger.warn("this method is currently under construction.")
if not self.inverse_droste:
raise ValueError()
if ncpus is None:
import multiprocessing
ncpus = multiprocessing.cpu_count()
if pool is None:
from multiprocessing import Pool
pool = Pool(ncpus)
def fl_scaling(k, s):
# scaling constant
from scipy.special import gamma
return (2**(2 * s) * s * gamma(s + k / 2)) / (np.pi**(k / 2) *
gamma(1 - s))
# Directly compute and return in 1D
if self.dim == 1:
s = self.integral_knl.s
targets = np.array(self.q_points).reshape(-1)
r1 = targets
r2 = self.source_box_extent - targets
self.kernel_exterior_normalizers = 1 / (2 * s) * (
1 / r1**(2 * s) + 1 / r2**(2 * s)) * fl_scaling(k=self.dim,
s=s)
return
from meshmode.array_context import PyOpenCLArrayContext
from meshmode.dof_array import thaw, flatten
from meshmode.mesh.io import read_gmsh
from meshmode.discretization import Discretization
from meshmode.discretization.poly_element import \
PolynomialWarpAndBlendGroupFactory
# {{{ gmsh processing
import gmsh
gmsh.initialize()
gmsh.option.setNumber("General.Terminal", 1)
# meshmode does not support other versions
gmsh.option.setNumber("Mesh.MshFileVersion", 2)
gmsh.option.setNumber("Mesh.CharacteristicLengthMax", mesh_size)
gmsh.option.setNumber("Mesh.ElementOrder", mesh_order)
if mesh_order > 1:
gmsh.option.setNumber("Mesh.CharacteristicLengthFromCurvature", 1)
# radius of source box
hs = self.source_box_extent / 2
# radius of bouding sphere
r = hs * np.sqrt(self.dim)
logger.debug("r_inner = %f, r_outer = %f" % (hs, r))
if self.dim == 2:
tag_box = gmsh.model.occ.addRectangle(x=0,
y=0,
z=0,
dx=2 * hs,
dy=2 * hs,
tag=-1)
elif self.dim == 3:
tag_box = gmsh.model.occ.addBox(x=0,
y=0,
z=0,
dx=2 * hs,
dy=2 * hs,
dz=2 * hs,
tag=-1)
else:
raise NotImplementedError()
if self.dim == 2:
tag_ball = gmsh.model.occ.addDisk(xc=hs,
yc=hs,
zc=0,
rx=r,
ry=r,
tag=-1)
elif self.dim == 3:
tag_sphere = gmsh.model.occ.addSphere(xc=hs,
yc=hs,
zc=hs,
radius=r,
tag=-1)
tag_ball = gmsh.model.occ.addVolume([tag_sphere], tag=-1)
else:
raise NotImplementedError()
dimtags_ints, dimtags_map_ints = gmsh.model.occ.cut(
objectDimTags=[(self.dim, tag_ball)],
toolDimTags=[(self.dim, tag_box)],
tag=-1,
removeObject=True,
removeTool=True)
gmsh.model.occ.synchronize()
gmsh.model.mesh.generate(self.dim)
from tempfile import mkdtemp
from os.path import join
temp_dir = mkdtemp(prefix="tmp_volumential_nft")
msh_filename = join(temp_dir, 'chinese_lucky_coin.msh')
gmsh.write(msh_filename)
gmsh.finalize()
mesh = read_gmsh(msh_filename)
if remove_tmp_files:
import shutil
shutil.rmtree(temp_dir)
# }}} End gmsh processing
arr_ctx = PyOpenCLArrayContext(queue)
discr = Discretization(
arr_ctx, mesh,
PolynomialWarpAndBlendGroupFactory(order=quad_order))
from pytential import bind, sym
# {{{ optional checks
if 1:
if self.dim == 2:
arerr = np.abs((np.pi * r**2 - (2 * hs)**2) -
bind(discr, sym.integral(self.dim, self.dim, 1))
(queue)) / (np.pi * r**2 - (2 * hs)**2)
if arerr > 1e-12:
log_to = logger.warn
else:
log_to = logger.debug
log_to("the numerical error when computing the measure of a "
"unit ball is %e" % arerr)
elif self.dim == 3:
arerr = np.abs((4 / 3 * np.pi * r**3 - (2 * hs)**3) -
bind(discr, sym.integral(self.dim, self.dim, 1))
(queue)) / (4 / 3 * np.pi * r**3 - (2 * hs)**3)
if arerr > 1e-12:
log_to = logger.warn
else:
log_to = logger.debug
logger.warn(
"The numerical error when computing the measure of a "
"unit ball is %e" % arerr)
# }}} End optional checks
# {{{ kernel evaluation
# TODO: take advantage of symmetry if this is too slow
from volumential.droste import InverseDrosteReduced
# only for getting kernel evaluation related stuff
drf = InverseDrosteReduced(self.integral_knl,
self.quad_order,
self.interaction_case_vecs,
n_brick_quad_points=0,
knl_symmetry_tags=[],
auto_windowing=False)
# uses "dist[dim]", assigned to "knl_val"
knl_insns = drf.get_sumpy_kernel_insns()
eval_kernel_insns = [
insn.copy(within_inames=insn.within_inames | frozenset(["iqpt"]))
for insn in knl_insns
]
from sumpy.symbolic import SympyToPymbolicMapper
sympy_conv = SympyToPymbolicMapper()
scaling_assignment = lp.Assignment(
id=None,
assignee="knl_scaling",
expression=sympy_conv(
self.integral_knl.get_global_scaling_const()),
temp_var_type=lp.Optional(),
)
extra_kernel_kwarg_types = ()
if "extra_kernel_kwarg_types" in kwargs:
extra_kernel_kwarg_types = kwargs["extra_kernel_kwarg_types"]
lpknl = lp.make_kernel( # NOQA
"{ [iqpt, iaxis]: 0<=iqpt<n_q_points and 0<=iaxis<dim }",
[
"""
for iqpt
for iaxis
<> dist[iaxis] = (quad_points[iaxis, iqpt]
- target_point[iaxis])
end
end
"""
] + eval_kernel_insns + [scaling_assignment] + [
"""
for iqpt
result[iqpt] = knl_val * knl_scaling
end
"""
],
[
lp.ValueArg("dim, n_q_points", np.int32),
lp.GlobalArg("quad_points", np.float64, "dim, n_q_points"),
lp.GlobalArg("target_point", np.float64, "dim")
] + list(extra_kernel_kwarg_types) + [
"...",
],
name="eval_kernel_lucky_coin",
lang_version=(2018, 2),
)
lpknl = lp.fix_parameters(lpknl, dim=self.dim)
lpknl = lp.set_options(lpknl, write_cl=False)
lpknl = lp.set_options(lpknl, return_dict=True)
# }}} End kernel evaluation
node_coords = flatten(thaw(arr_ctx, discr.nodes()))
nodes = cl.array.to_device(
queue, np.vstack([crd.get() for crd in node_coords]))
int_vals = []
for target in self.q_points:
evt, res = lpknl(queue, quad_points=nodes, target_point=target)
knl_vals = res['result']
integ = bind(
discr, sym.integral(self.dim, self.dim,
sym.var("integrand")))(queue,
integrand=knl_vals)
queue.finish()
int_vals.append(integ)
int_vals_coins = np.array(int_vals)
int_vals_inf = np.zeros(self.n_q_points)
# {{{ integrate over the exterior of the ball
if self.dim == 2:
def rho_0(theta, target, radius):
rho_x = np.linalg.norm(target, ord=2)
return (-1 * rho_x * np.cos(theta) +
np.sqrt(radius**2 - rho_x**2 * (np.sin(theta)**2)))
def ext_inf_integrand(theta, s, target, radius):
_rho_0 = rho_0(theta, target=target, radius=radius)
return _rho_0**(-2 * s)
def compute_ext_inf_integral(target, s, radius):
# target: target point
# s: fractional order
# radius: radius of the circle
import scipy.integrate as sint
val, _ = sint.quadrature(partial(ext_inf_integrand,
s=s,
target=target,
radius=radius),
a=0,
b=2 * np.pi)
return val * (1 / (2 * s)) * fl_scaling(k=self.dim, s=s)
if 1:
# optional test
target = [0, 0]
s = 0.5
radius = 1
scaling = fl_scaling(k=self.dim, s=s)
val = compute_ext_inf_integral(target, s, radius)
test_err = np.abs(val - radius**(-2 * s) * 2 * np.pi *
(1 /
(2 * s)) * scaling) / (radius**(-2 * s) *
2 * np.pi *
(1 /
(2 * s)) * scaling)
if test_err > 1e-12:
logger.warn("Error evaluating at origin = %f" % test_err)
for tid, target in enumerate(self.q_points):
# The formula assumes that the source box is centered at origin
int_vals_inf[tid] = compute_ext_inf_integral(
target=target - hs, s=self.integral_knl.s, radius=r)
elif self.dim == 3:
# FIXME
raise NotImplementedError("3D not yet implemented.")
else:
raise NotImplementedError("Unsupported dimension")
# }}} End integrate over the exterior of the ball
self.kernel_exterior_normalizers = int_vals_coins + int_vals_inf
return
def get_path_length(queue, density_discr):
return bind(density_discr, sym.integral(2, 1, 1))(queue)
def compute_biharmonic_extension(queue,
target_discr,
qbx,
density_discr,
f,
fx,
fy,
target_association_tolerance=0.05):
"""Biharmoc extension. Currently only support
interior domains in 2D (i.e., extension is on the exterior).
"""
# pylint: disable=invalid-unary-operand-type
dim = 2
queue = setup_command_queue(queue=queue)
qbx_forced_limit = 1
normal = get_normal_vectors(queue, density_discr, loc_sign=1)
bdry_op_sym = get_extension_bie_symbolic_operator(loc_sign=1)
bound_op = bind(qbx, bdry_op_sym)
bc = [fy, -fx]
bvp_rhs = bind(qbx, sym.make_sym_vector("bc", dim))(queue, bc=bc)
gmres_result = gmres(bound_op.scipy_op(queue,
"sigma",
np.float64,
mu=1.,
normal=normal),
bvp_rhs,
tol=1e-9,
progress=True,
stall_iterations=0,
hard_failure=True)
mu = gmres_result.solution
arclength_parametrization_derivatives_sym = sym.make_sym_vector(
"arclength_parametrization_derivatives", dim)
density_mu_sym = sym.make_sym_vector("mu", dim)
dxids_sym = arclength_parametrization_derivatives_sym[0] + \
1j * arclength_parametrization_derivatives_sym[1]
dxids_conj_sym = arclength_parametrization_derivatives_sym[0] - \
1j * arclength_parametrization_derivatives_sym[1]
density_rho_sym = density_mu_sym[1] - 1j * density_mu_sym[0]
density_conj_rho_sym = density_mu_sym[1] + 1j * density_mu_sym[0]
# convolutions
GS1 = sym.IntG( # noqa: N806
ComplexLinearLogKernel(dim),
density_rho_sym,
qbx_forced_limit=None)
GS2 = sym.IntG( # noqa: N806
ComplexLinearKernel(dim),
density_conj_rho_sym,
qbx_forced_limit=None)
GD1 = sym.IntG( # noqa: N806
ComplexFractionalKernel(dim),
density_rho_sym * dxids_sym,
qbx_forced_limit=None)
GD2 = [
sym.IntG( # noqa: N806
AxisTargetDerivative(iaxis, ComplexLogKernel(dim)),
density_conj_rho_sym * dxids_sym +
density_rho_sym * dxids_conj_sym,
qbx_forced_limit=qbx_forced_limit) for iaxis in range(dim)
]
GS1_bdry = sym.IntG( # noqa: N806
ComplexLinearLogKernel(dim),
density_rho_sym,
qbx_forced_limit=qbx_forced_limit)
GS2_bdry = sym.IntG( # noqa: N806
ComplexLinearKernel(dim),
density_conj_rho_sym,
qbx_forced_limit=qbx_forced_limit)
GD1_bdry = sym.IntG( # noqa: N806
ComplexFractionalKernel(dim),
density_rho_sym * dxids_sym,
qbx_forced_limit=qbx_forced_limit)
xp, yp = get_arclength_parametrization_derivative(queue, density_discr)
xp = -xp
yp = -yp
tangent = get_tangent_vectors(queue,
density_discr,
loc_sign=qbx_forced_limit)
# check and fix the direction of parametrization
# logger.info("Fix all negative signs in:" +
# str(xp * tangent[0] + yp * tangent[1]))
grad_v2 = [
bind(qbx,
GD2[iaxis])(queue,
mu=mu,
arclength_parametrization_derivatives=make_obj_array(
[xp, yp])).real for iaxis in range(dim)
]
v2_tangent_der = sum(tangent[iaxis] * grad_v2[iaxis]
for iaxis in range(dim))
from pytential.symbolic.pde.scalar import NeumannOperator
from sumpy.kernel import LaplaceKernel
operator_v1 = NeumannOperator(LaplaceKernel(dim),
loc_sign=qbx_forced_limit)
bound_op_v1 = bind(qbx, operator_v1.operator(var("sigma")))
# FIXME: the positive sign works here
rhs_v1 = operator_v1.prepare_rhs(1 * v2_tangent_der)
gmres_result = gmres(bound_op_v1.scipy_op(queue, "sigma",
dtype=np.float64),
rhs_v1,
tol=1e-9,
progress=True,
stall_iterations=0,
hard_failure=True)
sigma = gmres_result.solution
qbx_stick_out = qbx.copy(
target_association_tolerance=target_association_tolerance)
v1 = bind((qbx_stick_out, target_discr),
operator_v1.representation(var("sigma"),
qbx_forced_limit=None))(queue,
sigma=sigma)
grad_v1 = bind(
(qbx_stick_out, target_discr),
operator_v1.representation(
var("sigma"),
qbx_forced_limit=None,
map_potentials=lambda pot: sym.grad(dim, pot)))(queue, sigma=sigma)
v1_bdry = bind(
qbx,
operator_v1.representation(var("sigma"),
qbx_forced_limit=qbx_forced_limit))(
queue, sigma=sigma)
z_conj = target_discr.nodes()[0] - 1j * target_discr.nodes()[1]
z_conj_bdry = density_discr.nodes().with_queue(queue)[0] \
- 1j * density_discr.nodes().with_queue(queue)[1]
int_rho = 1 / (8 * np.pi) * bind(
qbx, sym.integral(dim, dim - 1, density_rho_sym))(queue, mu=mu)
omega_S1 = bind( # noqa: N806
(qbx_stick_out, target_discr), GS1)(queue, mu=mu).real
omega_S2 = -1 * bind( # noqa: N806
(qbx_stick_out, target_discr), GS2)(queue, mu=mu).real
omega_S3 = (z_conj * int_rho).real # noqa: N806
omega_S = -(omega_S1 + omega_S2 + omega_S3) # noqa: N806
grad_omega_S1 = bind( # noqa: N806
(qbx_stick_out, target_discr), sym.grad(dim, GS1))(queue, mu=mu).real
grad_omega_S2 = -1 * bind( # noqa: N806
(qbx_stick_out, target_discr), sym.grad(dim, GS2))(queue, mu=mu).real
grad_omega_S3 = (int_rho * make_obj_array([1., -1.])).real # noqa: N806
grad_omega_S = -(grad_omega_S1 + grad_omega_S2 + grad_omega_S3
) # noqa: N806
omega_S1_bdry = bind(qbx, GS1_bdry)(queue, mu=mu).real # noqa: N806
omega_S2_bdry = -1 * bind(qbx, GS2_bdry)(queue, mu=mu).real # noqa: N806
omega_S3_bdry = (z_conj_bdry * int_rho).real # noqa: N806
omega_S_bdry = -(omega_S1_bdry + omega_S2_bdry + omega_S3_bdry
) # noqa: N806
omega_D1 = bind( # noqa: N806
(qbx_stick_out, target_discr),
GD1)(queue,
mu=mu,
arclength_parametrization_derivatives=make_obj_array([xp,
yp])).real
omega_D = (omega_D1 + v1) # noqa: N806
grad_omega_D1 = bind( # noqa: N806
(qbx_stick_out, target_discr), sym.grad(dim, GD1))(
queue,
mu=mu,
arclength_parametrization_derivatives=make_obj_array([xp,
yp])).real
grad_omega_D = grad_omega_D1 + grad_v1 # noqa: N806
omega_D1_bdry = bind( # noqa: N806
qbx, GD1_bdry)(queue,
mu=mu,
arclength_parametrization_derivatives=make_obj_array(
[xp, yp])).real
omega_D_bdry = (omega_D1_bdry + v1_bdry) # noqa: N806
int_bdry_mu = bind(
qbx, sym.integral(dim, dim - 1, sym.make_sym_vector("mu", dim)))(queue,
mu=mu)
omega_W = ( # noqa: N806
int_bdry_mu[0] * target_discr.nodes()[1] -
int_bdry_mu[1] * target_discr.nodes()[0])
grad_omega_W = make_obj_array( # noqa: N806
[-int_bdry_mu[1], int_bdry_mu[0]])
omega_W_bdry = ( # noqa: N806
int_bdry_mu[0] * density_discr.nodes().with_queue(queue)[1] -
int_bdry_mu[1] * density_discr.nodes().with_queue(queue)[0])
int_bdry = bind(qbx, sym.integral(dim, dim - 1, var("integrand")))(
queue, integrand=omega_S_bdry + omega_D_bdry + omega_W_bdry)
debugging_info = {}
debugging_info['omega_S'] = omega_S
debugging_info['omega_D'] = omega_D
debugging_info['omega_W'] = omega_W
debugging_info['omega_v1'] = v1
debugging_info['omega_D1'] = omega_D1
int_interior_func_bdry = bind(qbx,
sym.integral(2, 1,
var("integrand")))(queue,
integrand=f)
path_length = get_path_length(queue, density_discr)
ext_f = omega_S + omega_D + omega_W + (int_interior_func_bdry -
int_bdry) / path_length
grad_f = grad_omega_S + grad_omega_D + grad_omega_W
return ext_f, grad_f[0], grad_f[1], debugging_info
def compute_harmonic_extension(queue,
target_discr,
qbx,
density_discr,
f,
loc_sign=1,
target_association_tolerance=0.05,
gmres_tolerance=1e-14):
"""Harmonic extension.
:param: loc_sign indicates the domain for extension,
which equals to the negation of the loc_sign
for the original problem.
"""
dim = qbx.ambient_dim
queue = setup_command_queue(queue=queue)
# {{{ describe bvp
from sumpy.kernel import LaplaceKernel
kernel = LaplaceKernel(dim)
cse = sym.cse
sigma_sym = sym.var("sigma")
#sqrt_w = sym.sqrt_jac_q_weight(3)
sqrt_w = 1
inv_sqrt_w_sigma = cse(sigma_sym / sqrt_w)
bdry_op_sym = (
loc_sign * 0.5 * sigma_sym + sqrt_w *
(sym.S(kernel, inv_sqrt_w_sigma) + sym.D(kernel, inv_sqrt_w_sigma)))
# }}}
bound_op = bind(qbx, bdry_op_sym)
# {{{ fix rhs and solve
bvp_rhs = bind(qbx, sqrt_w * sym.var("bc"))(queue, bc=f)
from pytential.solve import gmres
gmres_result = gmres(bound_op.scipy_op(queue, "sigma", dtype=np.float64),
bvp_rhs,
tol=gmres_tolerance,
progress=True,
stall_iterations=0,
hard_failure=True)
sigma = bind(qbx,
sym.var("sigma") / sqrt_w)(queue, sigma=gmres_result.solution)
# }}}
debugging_info = {}
debugging_info['gmres_result'] = gmres_result
# {{{ postprocess
repr_kwargs = dict(qbx_forced_limit=None)
representation_sym = (sym.S(kernel, inv_sqrt_w_sigma, **repr_kwargs) +
sym.D(kernel, inv_sqrt_w_sigma, **repr_kwargs))
qbx_stick_out = qbx.copy(
target_association_tolerance=target_association_tolerance)
debugging_info['qbx'] = qbx_stick_out
debugging_info['representation'] = representation_sym
debugging_info['density'] = sigma
ext_f = bind((qbx_stick_out, target_discr),
representation_sym)(queue, sigma=sigma).real
# }}}
# NOTE: matching is needed if using
# pytential.symbolic.pde.scalar.DirichletOperator
# but here we are using a representation that does not have null
# space for exterior Dirichlet problem
if loc_sign == 1 and False:
bdry_measure = bind(density_discr, sym.integral(dim, dim - 1,
1))(queue)
int_func_bdry = bind(qbx, sym.integral(dim, dim - 1,
var("integrand")))(queue,
integrand=f)
solu_bdry = bind((qbx, density_discr),
representation_sym)(queue, sigma=sigma).real
int_solu_bdry = bind(qbx, sym.integral(
dim, dim - 1, var("integrand")))(queue, integrand=solu_bdry)
matching_const = (int_func_bdry - int_solu_bdry) / bdry_measure
else:
matching_const = 0.
ext_f = ext_f + matching_const
def eval_ext_f(target_discr):
return bind((qbx_stick_out, target_discr), representation_sym)(
queue, sigma=sigma).real + matching_const
debugging_info['eval_ext_f'] = eval_ext_f
return ext_f, debugging_info
