dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On 01/02/11 20:09, Anders Logg wrote:
> Question #143195 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/143195
> 
>     Status: Open => Answered
> 
> Anders Logg proposed the following answer:
> On Tue, Feb 01, 2011 at 05:48:10PM -0000, David Ham wrote:
>> Question #143195 on DOLFIN changed:
>> https://answers.launchpad.net/dolfin/+question/143195
>>
>>     Status: Answered => Open
>>
>> David Ham is still having a problem:
>> Thanks Kent, this is a useful insight into part of the problem.
>>
>> In the example you gave, the variable which is eliminated is in a
>> continuous space and, if I understand the code correctly, you use mass
>> lumping to invert the mass matrix in the Schur compliment.
>>
>> However this doesn't indicate how one would do this in the DG case for
>> which the mass can be directly inverted. So I guess the direct question
>> is, is there a way in Dolfin assembling the inverse of the local tensor?
>> Ie something like:
>>
>> assemble(inverse(u*v*dx))
> 
> No, not currently but it would be relatively easy to add if we
> conclude that it's important enough to complicate the assembler with
> the extra code (which would be to check a flag and then call something
> like Armadillo to do the inversion in-place).
> 

There is a blueprint on this somewhere.

Garth

> --
> Anders
>