dolfin team mailing list archive

[Garth Wells <question123790@xxxxxxxxxxxxxxxxxxxxx>]

Question #123790 on DOLFIN changed:
https://answers.launchpad.net/dolfin/+question/123790

    Status: Open => Answered

Garth Wells proposed the following answer:
There is a subtly here related to the Neumann boundary condition. DOLFIN
interpolates the BCs in finite element spaces, which can lead to some
'creep' around corners if the domain is not properly specified. What you
need to do is mark the boundary on which you're applying the force bc,
e.g

In the UFL file, do something like:

  L = f*vdx + h*v*ds(3)

This means the h*v should only be integrated over the boundary subdomain
'3'.

In the C++ code:

  // Right boundary
  class RightBoundary : public SubDomain
  {
    bool inside(const Array<double>& x, bool on_boundary) const
    {
      if (1.0 - x[0] < DOLFIN_EPS && on_boundary)
        return true;
      else
        return false;
    }
  };

  // Create subdomain
  RightBoundary right_boundary;

  // Create mesh function over facets (FacetFunction is new in 0.9.9, otherwise use 
  // MeshFunction<unsigned int> right_boundary_function(mesh, 1) for a 2D problem);
  FacetFunction<unsigned int> right_boundary_function(mesh);

  // Mark the boundary with '3'
  right_boundary.mark(right_boundary_function, 3);

If you're using the VariationalProblem class, then do

  // Create variational problem
  VariationalProblem pde(a_form, L, bc, 0, &right_boundary_function, 0);

This will pass the subdomains to the assembler. If you're calling the assembler directly, take a look in 
  
  dolfin/fem/Assembler.h 

for how to pass subdomains.

The elastodynamics demo also shows how to use boundary subdomains when
applying force bcs (both in C++ and Python).

-- 
You received this question notification because you are a member of
DOLFIN Team, which is an answer contact for DOLFIN.