I Stephane,
thanks for your answer
Stephane Del Pino wrote:
Hello Pierre,
At first reading I did not noticed any mistake in your ff3d's file ...
This might be a bug in ff3d, but I am not sure about that (this is a quite
basic test case, but one never knows ...). What does happen when your mesh is
more refined along the Z-axis? Is it converging?
The difference with the structured case could be explained by the fact that
structured discretization uses Q1 (hexahedra) finite element, which is exact
for tri-affine functions ((a+x)*(b+y)*(d+z)+e) and P1 is only exact for
affine functions (ax+by+cz+d) functions ...
Please let me know if this converges.
I have tried with a much finer mesh (>200000 tetrahedra), but I obtain
the same wrong value
as with the coarse mesh.
By the way I have a few remarks concerning your ff3d's command file:
- you should use cg instead of bicgstab: your problem is symetric, so it is
more efficient, and you can use the ichol preconditionner if you use the
sparse matrix storage (default is no preconditionner);
- since you are consumming to much memory you could use the
'memory(matrix=none)' option. It is a bit slower but requires few memory.
Actually, in the more complex (and asymetric) problems I am studying, I
am using the fictitious domain method
because my geometry is not a simple box, and I use the "extract" command
to define boundaries.
I reach the memory limit during the initial phase, before the solver
launches.
Pierre
_______________________________________________
ff3d-users mailing list
[email protected]
http://lists.nongnu.org/mailman/listinfo/ff3d-users
