On Mon, 2008-07-14 at 12:18 +0200, Robert Cimrman wrote:
On Sat, 2008-07-12 at 16:03 -0500, Ryan Krauss wrote:
This is interesting - how you would use a test variable for a posteriori estimates? I know almost nothing about this, but am interested to have some error estimates in SfePy, preferably such that do not depend on the equations solved (i.e. discretization-error based). Have you some link?
hmm.. I guess I was wrong, wrote that at 3am local time! Just replied without reading it closely. When I saw "dual" I thought you were talking about "adjoint" that is in one of your presentations (closed channel shape optimization). I was also looking at one of the Fenics documents, and they use dual/adjoint for estimating the error in some integrated quantity (in this case they were looking at lift/drag coeffs). And use it for auto refinements. I would like to do the same using sfepy.
I see. Yes, we use the adjoint problem to a weak formulation of Navier-Stokes equations to perform a shape sensitivity analysis.
It would be great to have error estimators in sfepy. If you have some specific wishes, submit an issue, please, at 'http://code.google.com/p/sfepy/issues/list'.