From: John Peterson <peterson@cf...>  20070911 14:21:49

Isn't the Darcy velocity u ~ grad(p)=3F In that case I don't understan= d the rationale for using a higherorder approximation for velocity. Mixed methods are notorious for stability/compatibility restrictions on the pair of approximation spaces used. I'm not familiar enough with= Darcy flow to know whether the one you've chosen is LBBstable. John Roy Stogner writes: > On Tue, 11 Sep 2007, Shengli Xu wrote: >=20 > > Hi, I use the mixed galerkin finite element to sovle darcy equatio= n. > > pressure is linear interpretation, velocity is quadratic interpret= ation=A1=A3 The > > domain is square. The result is correct when I apply noslip b.c o= n top and > > bottom side, zero pressure on rightbottom corner, Ux=3D1.0 on lef= t side. But > > I can not get correct result if I apply noslip b.c on top and bot= tom side, > > zero pressure on the right side, P=3D1 on the left side. Why can I= get correct > > result with this b.c.=3F >=20 > Unless someone has seen a similar Darcyflow problem before (John=3F= ), > which is a long shot, you're not likely to find someone who can answ= er > this. If you've already got the problem boiled down to a simple > benchmark, I'd try running it on a single square element and compari= ng > the numerical linear system to what you can compute by hand. Doing > so for even that bare minimum (22x22 in your case, I suppose) system= > can be tedious, but in the past it's been the only way I've been abl= e > to find one or two subtle bugs in application codes. >  > Roy 