[petsc-users] FW: [Fwd: Re: Question on asm]

[Isik Ali Ozcer]

Hi All,

Regarding ILU convergence at subdomain boundaries, I need to add to my colleague's message that we followed a very basic tutorial that shows matrix and rhs assembly, preconditioner and solver setup, and finally a KSPsolve.  We did not prepare any application ordering information or index sets. I am not sure if Petsc in our implementation has all the info it needs to perform proper overlapping operations with ASM. Do we need to worry about setting up AO and IS contexts before attempting KSPsolve with ASM and ILU? 

Thanks a lot and have a great day,

Isik
petsc-newb



-------- Original Message --------
Subject:        Re: [petsc-users] Question on asm
Date:   Tue, 14 May 2013 13:03:38 -0500
From:   Jed Brown <jedbrown at mcs.anl.gov>
To:     Konstantinos Kontzialis <ckontzialis at lycos.com>,
petsc-users at mcs.anl.gov



Konstantinos Kontzialis <ckontzialis at lycos.com> writes:

> Dear all,
>
> I am using asm as a preconditioner for solving the Navier Stokes
> equations. When using ILU as the subdomain preconditioner, I notice
> convergence problems in the partition boundaries.

I.e., it converges slowly there?  That is natural because the
preconditioner is more accurate in the interior of subdomains.

> This does not happen with the Jacobi preconditioner.

Naturally, because Jacobi is identical regardless of the partition.

> Do I need to provide any connectivity, node mapping, and/or overlap
> information to Petsc before calling KSPsolve?

No, the matrix is enough.