dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, May 27, 2010 at 04:22:33PM +0100, Garth N. Wells wrote:
>
>
> On 27/05/10 15:58, Anders Logg wrote:
> >Wouldn't it work to just to this:
> >
> >v = v + (h/2.0*vnorm)*dot(velocity, grad(v))
> >U = 0.5*(u0 + u)
> >F = v*(u - u0)*dx + 0.5*k*(v*dot(velocity, grad(U))*dx + c*dot(grad(v), grad(U))*dx) - k*v*f*dx
> >a = lhs(F)
> >L = rhs(F)
> >
> >That's my understanding of SUPG, but perhaps I've missed something?
> >
>
> Looks like the same thing (I'd have to check). I prefer to
> expressing the method as weighting the residual, e.g. Galerkin +
> weighted residual.

I also think of it as weighting the residual, but with a modified test
function v. The above looks like a weighted (weak) residual to me.

The difference from your implementation is that you multiply the
stabilization part of the test function with the strong residual for
which the Laplacian (for P1 elements) is zero. In the above case, the
stabilization also hits the nonzero integrated by parts grad-grad term.

I don't know whether or not that is important.

> The code I added can be made more compact using lhs/rhs, but I was
> having some trouble and initially wanted to make a bare bones
> implementation.

ok.

I might try to make some changes. I'd like to get a feeling for how
SUPG works.

--
Anders

Attachment: signature.asc
Description: Digital signature