amg preconditioner

This is an old bug for linearalgebra backends, which do not include amg
as a preconditioner. Only PETSc, Trilinos do that. The bug has been
fixed in the stable and development branches of DOLFIN, but
unfortunately not released.

Not sure if there are any nightly builds for Windows binaries, otherwise
you need to get the code your self and compile.

Johan

On 08/24/2012 10:01 AM, Sven Fanghänel wrote:
> New question #206671 on FEniCS Project:
> https://answers.launchpad.net/fenics/+question/206671
>
> Hi
> i would like to set boundary condition in a rectangular mesh, with 2 subdoimain that are circles. I want to two Dirichlet boundary condition on the border of each circle.
> i got an error mesage: unknown preconditioner `amg`
> the code is:
> I use windows 7.
> Has everyone a tip for this problem.
>
> Thanx Sven
>
> from dolfin import *
> import numpy as np
> not_working_in_parallel("mesh smoothing demo")
>
> # Create rectangular mesh
> mesh = Rectangle(0.0, 0.0, 1.4, 0.8, 60, 20)
>
> # Define a circular hole
> center = Point(0.1, 0.1)
> center2 = Point(0.6, 0.6)
> radius = 0.05
>
> class Hole(SubDomain):
>
> def inside(self, x, on_boundary):
> r = sqrt((x[0] - center[0])**2 + (x[1] - center[1])**2)
> return r < 1.5*radius # slightly larger
>
> def snap(self, x):
> r = sqrt((x[0] - center[0])**2 + (x[1] - center[1])**2)
> if r < 1.5*radius:
> x[0] = center[0] + (radius / r)*(x[0] - center[0])
> x[1] = center[1] + (radius / r)*(x[1] - center[1])
>
> class Hole2(SubDomain):
>
> def inside(self, x, on_boundary):
> r = sqrt((x[0] - center2[0])**2 + (x[1] - center2[0])**2)
> return r < 1.5*radius # slightly larger
>
> def snap(self, x):
> r = sqrt((x[0] - center2[0])**2 + (x[1] - center2[1])**2)
> if r < 1.5*radius:
> x[0] = center2[0] + (radius / r)*(x[0] - center2[0])
> x[1] = center2[1] + (radius / r)*(x[1] - center2[1])
>
> # Mark hole and extract submesh
> hole = Hole()
> hole2 = Hole2()
> sub_domains = MeshFunction("uint", mesh, mesh.topology().dim())
> sub_domains.set_all(0)
> hole.mark(sub_domains, 1)
> hole2.mark(sub_domains, 2)
> mesh = SubMesh(mesh, sub_domains, 0)
> mesh.snap_boundary(hole)
> mesh.snap_boundary(hole2)
>
> # Refine and snap mesh
> plot(mesh, title="Mesh 0")
> num_refinements = 3
> for i in range(num_refinements):
>
> # Mark cells for refinement
> markers = MeshFunction("bool", mesh, mesh.topology().dim())
> markers.set_all(False)
> for cell in cells(mesh):
> if cell.midpoint().distance(center2) < 2*radius:
> markers[cell.index()] = True
>
> # Refine mesh
> mesh = refine(mesh, markers)
>
> # Snap boundary
> mesh.snap_boundary(hole)
> mesh.snap_boundary(hole2)
> # Plot mesh
> #plot(mesh, title=("Mesh %d" % (i + 1)))
> #interactive()
>
> # Initialize mesh function for interior domains
> boundaries = CellFunction("uint", mesh)
> boundaries.set_all(0)
> hole.mark(boundaries, 1)
> hole2.mark(boundaries, 2)
> h = CellSize(mesh)
> # Create mesh and define function space
>
> V = FunctionSpace(mesh, "CG", 1)
> R = FunctionSpace(mesh, "R", 0)
> W = V * R
> # Define input data
> a0 = Constant(1.0)
> a1 = Constant(0.01)
> g_L = Expression("- 10*exp(- pow(x[1] - 0.5, 2))")
> g_R = Constant("1.0")
> f = Constant(1.0)
>
> # Define function space and basis functions
> V = FunctionSpace(mesh, "CG", 2)
> u = TrialFunction(V)
> v = TestFunction(V)
>
> # Define Dirichlet boundary conditions at top and bottom boundaries
> bcs = [DirichletBC(V, 1.0, boundaries, 1),
> DirichletBC(V, 0.0, boundaries, 2)]
>

Hi Johan thanx for your quick answer.
is it runnuing if i installed fenics on ubuntu 12.04

At least then you have possibilities to run a nightly build of the
development version, which should incorporate the bug fix.

Johan

On 08/24/2012 01:15 PM, Sven Fanghänel wrote:
> Question #206671 on FEniCS Project changed:
> https://answers.launchpad.net/fenics/+question/206671
>
> Status: Answered => Open
>
> Sven Fanghänel is still having a problem:
> Hi Johan thanx for your quick answer.
> is it runnuing if i installed fenics on ubuntu 12.04
>

There is also a nightly build for Windows, but this is not working reliably at the moment. And the Windows binaries does not come with PETSc or Trilinos if that matters.