[petsc-users] AMG preconditioners

[Manav Bhatia]

> On Mar 9, 2016, at 10:19 AM, Jed Brown <jed at jedbrown.org> wrote:
> 
> Manav Bhatia <bhatiamanav at gmail.com> writes:
> 
>> Thanks! 
>> 
>> I am intrigued by this. 
>> 
>> I am solving a system of equations where the natural block size is
>> 2x2. This is obtained by rewriting a complex linear system of
>> equations in real numbers.
> 
> What equations are you solving?  Is this system a leading cost in your
> application?  (Using complex scalars is another option.  Yes, it's lame
> that PETSc can't currently use both in the same library.)
> 

The application is that of small-disturbance frequency domain analysis about a steady-state. 

This needs a mix of linear solves for system of equations in real numbers and complex numbers.

I was able to setup the complex solves using real numbers with a relatively small piece of code. It is working very well for the most part, except that I got some errors while trying to use the AMG preconditioners with it. 

Making the code generic for BAIJ or AIJ, I think, will be a very small fix in comparison with recompiling PETSC with complex scalars, since a lot of my application code assumes that the underlying libraries (libMesh, PETSC, SLEPC) are compiled for real scalars. 

-Manav


>> I use MatSetValuesBlocked to set the values of each 2x2 block in my
>> matrix. I am assuming that this method works only for blocked matrix. 
> 
> No, just call MatSetBlockSize and MatSetValuesBlocked (and variants)
> will work for any matrix type.