It's a bug in tensor representation. As a temporary fix, try setting: parameters["form_compiler"]["representation"] = "quadrature" at the top to disable tensor representation.
Martin On 2 September 2013 10:26, Cotter, Colin J <[email protected]>wrote: > Hi Dave, > This term can't be integrated exactly because it is not polynomial, so > my first guess would be that different quadrature rules are used for those > two different expressions (if you don't suggest a quadrature rule then FFC > tries to guess one from the expression). Try fixing the quadrature rule in > that term and see if the results are still different. > > all the best > --Colin > ------------------------------ > *From:* [email protected] [[email protected]] > on behalf of David Bernstein [[email protected]] > *Sent:* 01 September 2013 20:35 > *To:* [email protected] > *Subject:* [FEniCS] Question about nonlinear problem > > Hi Everyone, Recently I wrote a fenics python code for solving a > nonlinear advection diffusion equation which worked very well. However, I > came across one issue which I couldn't resolve and which is illustrated by > the attached code (simpler than original project but still seems to contain > the problem). > > The example problem is identical to the one described here > > http://fenicsproject.org/documentation/tutorial/nonlinear.html > > except that the nonlinear term q(u) is (1-u)^(-2), i.e., the PDE is > > grad(grad(u)/(1-u)^2) = -f > > The attached code gives different results if q is written as "(1-u)**(-2)" > and "(1/(1-u))**2". Is there a reason for this? I would have thought they > should be equivalent. > > Thanks, > Dave > > _______________________________________________ > fenics mailing list > [email protected] > http://fenicsproject.org/mailman/listinfo/fenics > > _______________________________________________ > fenics mailing list > [email protected] > http://fenicsproject.org/mailman/listinfo/fenics > >
_______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
