Yeah, Roy is right, better to use quadrature.
The approach I suggested initially would require you to project the exact
solution into the FE space first, but you probably want to avoid that since
it introduces a projection error into your calculation.
David
On Tue, Jan 19, 2016 at 6:20 PM, Roy
On Tue, 19 Jan 2016, Roy Stogner wrote:
> In the short term the workaround ought to be to use
> WRITE_SERIAL_FILES; I'm not sure that's working but if it's not I'll
> try to fix it ASAP.
WRITE_SERIAL_FILES | WRITE_DATA works.
Parallel xda mesh support ought to be relatively easy to add, but I'm
On Tue, 19 Jan 2016, Salazar De Troya, Miguel wrote:
Related to this issue, I would like to pass the rigid body modes to
a multigrid preconditioner. I’ve seen that there is an interface in
libMesh through PetscNonlinearSolver::nearnullspace, but my system
is a FEMSystem with a NewtonSolver. It
Hello
I am trying to speed up a linear elasticity problem with isotropic and
heterogeneous properties. It is a topology optimization problem, therefore some
regions have an almost zero stiffness whereas other regions have a higher
value, making the matrix ill-conditioned. So far, from having se
This may not be what you want to do, IMHO.
In the extreme case: If you compute the Uexact vector and then do the
vec*mat*vec, then you'll get a result of 0 for problems where your
method gives you an interpolant of the exact solution, even if that
interpolant doesn't equal the exact solution.
If
On Thu, 14 Jan 2016, Roy Stogner wrote:
> You know what? It doesn't for me either, my previous reports to the
> contrary. But that use case *should* work, and does work on lots of
> other meshes I've tried. I'll keep looking into it.
And now I think I understand the problem. Our solution res
Thanks, that is helpful.
--Junchao Zhang
On Tue, Jan 19, 2016 at 4:10 PM, David Knezevic
wrote:
> Assemble the Uexact vector, then compute the difference e = U - Uexact.
>
> You can also refer to error_estimation/exact_solution.h, but I believe
> that's used for computing L2 or H1 error, not an
Assemble the Uexact vector, then compute the difference e = U - Uexact.
You can also refer to error_estimation/exact_solution.h, but I believe
that's used for computing L2 or H1 error, not an arbitrary energy norm.
David
On Tue, Jan 19, 2016 at 5:04 PM, Junchao Zhang
wrote:
> How to get e =
How to get e = U - Uex? Is there a libmesh interface for that?
--Junchao Zhang
On Tue, Jan 19, 2016 at 3:57 PM, David Knezevic
wrote:
> Assuming you've already assembled K, so you can just do a matvec
> (SparseMatrix::vector_mult) followed by a dot product (NumericVector::dot).
>
> David
>
>
>
Assuming you've already assembled K, so you can just do a matvec
(SparseMatrix::vector_mult) followed by a dot product (NumericVector::dot).
David
On Tue, Jan 19, 2016 at 4:53 PM, Junchao Zhang
wrote:
> Hello,
> I want to compute e^TKe as a measure of the error of a solution. Here e =
> U -
Hello,
I want to compute e^TKe as a measure of the error of a solution. Here e =
U - Uex, supposing I know the analytic answer to the PDE.
How can I do it in libmesh? Is there an example?
Thank you.
--Junchao Zhang
-
11 matches
Mail list logo
