dolfin team mailing list archive

[Neilen Marais <nmarais@xxxxxxxxx>]

Hi Marie,

On Thu, May 5, 2011 at 7:34 PM, Marie E. Rognes <meg@xxxxxxxxx> wrote:
> On 05/05/2011 07:16 PM, Neilen Marais wrote:
>>
>>  In part this is
>> simply because the "standard" dolfin doesn't cater quite that well for
>> stuff I need for electromagnetic simulation (yet?)
>
> Just out of curiousity, what stuff to you need for electromagnetic
> simulation that DOLFIN doesn't cater well for?

At the moment I'm missing a point source that can be applied to vector
valued functions in some way, see
https://answers.launchpad.net/dolfin/+question/152565

Anders went ahead and implemented a point source class, but as far as
I can tell there is no way of making it work with a Nedelec type
element. See the last comment on launchpad answers.

The other biggie (that I don't expect to be fully resolved soon) is
the lack of complex support. While it is possible to formulate a
complex system entirely in dolfin by decomposing it into two coupled
real systems, the resulting real matrix is very ill conditioned. See
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.23.2071&rep=rep1&type=pdf

The paper above proposes a method for using real solvers to solve
complex problems fairly efficiently, but it seems that for the most
part it only approaches the performance of a "real" complex solver.
That method ("Komplex") is apparently implemented in Trilinos, so it
may be possible to use it through dolfin? In the mean time I'm just
constructing real and imaginary part matrices seperately in uBLASS
format and then combining the csr data into a scipy.sparse complex
matrix.

Best Regards
Neilen

>
> --
> Marie
>
> _______________________________________________
> Mailing list: https://launchpad.net/~dolfin
> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~dolfin
> More help   : https://help.launchpad.net/ListHelp
>