Ondrej Certik wrote:

On Fri, Aug 1, 2008 at 12:23 PM, Ondrej Certik <ond...@certik.cz> wrote:

If you plugged the basis into the equation naively (e.g. directly), you'd get unsymmetric matrices, so you first integrate the equation per partes (or the green theorem, as you use in the paper). We do not care so much about symmetry (we use UMFPACK, right?), but

BTW, we do actually care a lot, because in my case we are using an eigensolver and if the matrix is not symmetric, the eigenvalues are in general complex. Not mentioning that pysparse (imho) can only solve symmetric matrices. So using the per partes is actually crutial just to be able to solve it easily.

Yes, in your case we do care.

The argument about linear elements is imho not that important, you can always use higher order elements.

But as you said, it would not be that easy.

So from my point of view, you really want to rewrite the problem as symmetric just to move forward. Everything else is just a sideeffect, that can be fixed somehow.

Symmetry is a win if one can achieve it, I agree with that. But believe me, weak formulations have other merits as well :)

r.