dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Mar 04, 2009 at 01:21:18PM +0100, Johan Hake wrote:
> On Wednesday 04 March 2009 13:13:26 Anders Logg wrote:
> > On Wed, Mar 04, 2009 at 09:08:49AM +0100, Johan Hake wrote:
> > > On Wednesday 04 March 2009 08:21:11 Anders Logg wrote:
> > > > On Tue, Mar 03, 2009 at 08:08:56PM -0500, Hatef Monajemi wrote:
> > > > > Hi
> > > > >
> > > > > I was just wondering if it is possible to implement the
> > > > > Petrov-Galerkin type of weighting functions in dolfin. I am trying to
> > > > > solve a convection dominated problem in which the use of standard
> > > > > Galerkin weighting generates the oscillations which can be removed by
> > > > > upwinding. Is there any demo which shows the implementation of
> > > > > Petrov-Galerkin type of testfunctions.
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Hatef
> > > >
> > > > Yes, look at
> > > >
> > > >   demo/pde/stokes/stabilized
> > > >
> > > > It's very straightforward, just write
> > > >
> > > >   v = TestFunction(V)
> > > >   v = v + delta*A(v)
> > > >
> > > > where delta is a stabilization parameter and A is a suitable operator.
> > >
> > > The stabilizing parameter showed in this demo is quite crude. If it wont
> > > work for you, you can consider:
> > >
> > >   P = FunctionSpace(mesh,"CG",1)
> > >   V = VectorFunctionSpace(mesh,"CG"2)
> > >   v = TestFunction(P)
> > >
> > >   field = Function(V,"Some vector valued function that drives the
> > > advection") stab = Function(V,open("SUPG_stab.h").read()) # V is a Vector
> > >   stab.field = field
> > >
> > >
> > >   v = v + 0.2*dot(stab,grad(v))
> > >
> > > Here 0.2*stab relates to the delta Anders mention and A relates to grad
> > >
> > > We could consider adding the handed function as a SpecialFunction in
> > > DOLFIN. It is a general SUPG stabilization term, which depends not only
> > > on the local meshsize but also on the local field size.
> >
> > Would that work? The added term must be linear in a TestFunction, so
> > it can't be a Function.
> 
> Not sure what you are aiming at here. The stab function does not depend on the 
> testfunction.

ok, I thought you mean the entire thing. Yes, that could be nice to
add.

-- 
Anders

Attachment: signature.asc
Description: Digital signature