Stephane,
a nonsymmetric Hessian for our finite elements would indeed be a severe
bug. Is your test program small enough such that I can have a look at
it? In that case, please send it to my email address.
On the other hand, you refer to rotations and I am not sure I understand
how they are related to the discussion of the Hessian. Are you referring
to the Jacobian of the transformation? Since general rotations are not
in Q1, that might indeed cause the trace to be nonzero.
Best,
Guido
On 7/22/2010 3:04 AM, veys wrote:
Hi all,
I'm trying to implement an error estimator based on residuals. And for a
laplacian problem I need to have the laplacian of my unknown. So I have
used "fe_val.get_function_laplacians".
When I use Q1 FE on a square domain, my laplacian is 0 so that's
great :)
But, when I use a quarter of disk, my laplacian is no more 0.
It seems to me that DEAL compute hessian on the reference cell and then
apply transformations to be on real cell.
So I made a test : on my square domain I printed the hessian matrix,
denoted by H, on each quadrature points and then I saw that it's not
symetric, although diagonals terms are 0. A consequence of that is when
there is a rotation (in the case of a quarter of disk) then trace(H)!=0
even if we use Q1 FE.
Are you agree with me ?
Best,
Stephane Veys
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
