dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, May 04, 2006 at 04:40:40PM +0000, Alexander Jarosch wrote:
> Sorry for so may posts about the subject. I solved it with upgrading to 
> petsc 2.3.1-p12 with umfpack , using this form:
> 
> P1 = FiniteElement("Lagrange", "triangle", 1)
> P2 = FiniteElement("Vector Lagrange", "triangle", 2)
> TH = P2 + P1
> 
> (v, q) = TestFunctions(TH)
> (u, p) = TrialFunctions(TH)
> nu = Function(P1)
> f = Function(P2)
> 
> a = (nu*dot(grad(v), grad(u)) - div(v)*p + nu*q*div(u))*dx
> L = dot(v, f)*dx
> 
> and the direct solver setup as in the 
> src/demo/pde/stokes/taylor-hood/main.cpp file.

Good to see you figured it out. I don't know how you specified the
boundary condition for the pressure before, but it sometimes helps (in
particular for the LU solver) to specify it at one point. (It's
undetermined up to a constant anyway so you can set whatever value you
want.)

> BUT i found that the dolfin-convert script skips sometimes a vertex in 
> 2D in the conversion and in the final mesh it is considered as a 
> boundary. it happens if I use smaller mesh sizes...
> 
> let me know if I can provide more info on that bug...
> 
> Alex

It would be good if you could send an input mesh (as small and simple
as possible) and point out exactly what vertex is missing.

/Anders