[Libmesh-devel] Re: AW: libmesh - slepc

[Benjamin S. K. <be...@cf...>]

Sorry that I have been SO SLOW in getting back to you about this...

I looked at what you sent, and looks good.  I have made a few changes, 
like only looking for SLEPc if PETSc has been found, etc...  but the 
rest looks good.

The only thing I'd like to change is the SlepcInterface class...  Right 
now it inherits from LinearSolverInterface, which seems a bit 
contrived.  For a while now I've been thinking about creating a more 
complicated Solver class structure.  It would encompass LinearSolvers, 
NonlinearSolvers, TransientLinearSolvers, etc...  In this case there 
should simply be another, the EigenSolver.  The SlepcInterface would 
then be derived from an EigenSolver and implement specific functionality.

I'll talk to John a little more about this when he gets back tomorrow 
and see if we can put together some more complete thoughts on this...

-Ben



Steffen Petersen wrote:

>Find attached a tarball including the slepc interface.
>There is some stuff in it that is not really nice and clear,
>but I hope it is good enough to show the basic concept and
>to discuss some things.
>
>To use the inteface the code should be configured with
>--enable-slepc. I tried to #ifdef the changes, so that
>most stuff should only be there when slepc is enabled.
>
>I wrote a small example (examples/ex12), where the eigenapairs
>of a mass matrix are computed
>(usage: ex12 -n <requested number of eigenpairs> --use-slepc).
>
> Steffen
>
>
>  
>
>>that sounds good to me...  I'm in Australia right now, so I 
>>might be a little slow in getting back to you once you send 
>>me something.  A tarball would be fine...  I can play with it 
>>here.  I have my laptop.  
>>
>>-Ben
>>    
>>