On Tue, Apr 01, 2008 at 09:52:02PM +0200, Murtazo Nazarov wrote:
Quoting Murtazo Nazarov <murtazo@xxxxxxxxxxx>:
Anders Logg wrote:
On Tue, Apr 01, 2008 at 02:32:17PM +0200, Murtazo Nazarov wrote:
Anders Logg wrote:
On Mon, Mar 31, 2008 at 07:15:23PM +0200, Murtazo Nazarov wrote:
Anders Logg wrote:
On Mon, Mar 31, 2008 at 03:52:57PM +0200, Murtazo Nazarov
wrote:
Hi,
I want to apply a boundary condition to the vertices on a
boundary.
For that I define:
// Sub domain for MyBC
class MyBC_Boundary2D : public SubDomain
{
public:
bool inside(const real* p, bool on_boundary) const
{
return on_boundary && (p[0] < xmax - bmarg) && (p[0] >
xmin
+
bmarg);
}
};
Then I initialize:
//-----------------------------------------------------------------------------
void MyBC::init(SubDomain& sub_domain)
{ ...
mesh.init(0);
sub_domains = new MeshFunction<uint>(mesh, 0);
...}
Do you remember to set everything to 1 (number of subdomains)
here?
Yes I do:
// Mark everything as sub domain
1
(*sub_domains) = 1;
// Mark the sub domain as sub domain
0
sub_domain.mark(*sub_domains, 0);
/murtazo
Then I have no idea. You just need to dig into the code and see
what
goes wrong.
I think I found a mistake in Subdomain.mark(). The thing is bool
on_boundary should be defined
inside the loop for computing subdomain marker:
// Always false when not marking
facets
//bool on_boundary =
false;
// Compute sub domain
markers
for (MeshEntityIterator entity(mesh, dim); !entity.end();
++entity)
{
bool on_boundary = false;
...
}
Otherwise, if you want to apply a boundary condition on vertices,
it
becomes always
true after once meeting a vertex on the boundary.
It solved problem for me.
I'm not sure. If you are marking vertices, then they will never be
on
the boundary, unless you are doing this for a 1-dimensional mesh
(which typically has just 2 boundary vertices). Only *facets* will
ever be marked as being on the boundary, so for a 2-dimensional
mesh,
edges on the boundary will be marked as being on the boundary and
for
a 3-dimensional mesh, faces on the boundary will be marked as being
on
the boundary.
With the change I wrote I was able to mark vertices. Because,
if I mark facets or faces, I miss some nodes in the boundary
of the subdomain where I am applying my bc. That destroys
the solution. Now, I apply my boundary only for vertices
I need, and I do not miss any nodes. It work perfectly with
the slip BC.
Sounds like a similar issue that we have for Discontinuous elements.
Have
you
tried to apply BCs using a geometrical (or pointwise) approach?
Yes, I have. But as I said I had a problem with some side nodes. The
boundary condition did not apply to those nodes. Therefore I got
incorrect
result. But, with this approach everything went well.
/murtazo
I still don't understand what the problem is. What kind of mesh do you
have? Triangles or tetrahedra? What are you marking? Vertices?
If you are marking vertices, then on_boundary should always be
false. It is set to false on line 53 and it will never change since
line 60 will not be reached.
Ok, I see. It seems we are looking to some different version of DOLFIN. My
Subdomain.init looks like the following:
// Always false when not marking facets
//bool on_boundary = false;
// Compute sub domain markers
for (MeshEntityIterator entity(mesh, dim); !entity.end(); ++entity)
{
// on_boundary should be initialized here:
bool on_boundary = false;
// Check if entity is on the boundary if entity is a facet
if (dim == D - 1)
on_boundary = entity->numEntities(D) == 1;
if (dim == 0)
{
for (FacetIterator fi(*entity); !fi.end(); ++fi)
{
if(fi->numEntities(D) == 1)
on_boundary = true;
}
}
...
}
Now, you see then in the loop FacetIterator on_boundary is true, even if
we iterate over the vertices. Here dim = 0.
/murtazo
