[petsc-users] -pc_asm_type none

[Barry Smith]

> On Nov 5, 2016, at 11:47 PM, Fande Kong <fdkong.jd at gmail.com> wrote:
> 
> 
> 
> On Sat, Nov 5, 2016 at 10:10 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> > On Nov 5, 2016, at 9:35 PM, Fande Kong <fdkong.jd at gmail.com> wrote:
> >
> > Hi All,
> >
> > I know what "basic", "restrict" and "interpolate" represent, but have no idea about "none". Anybody knows what does mean when using "-pc_asm_type none"?
> 
>  PC_ASM_NONE         - Residuals from ghost points are not used, computed ghost values are
> $                        discarded.
> $                        Not very good.
> 
>   In the P (A_subdomain)^-1 R  both the R and the P skip values in the overlap region. Sort of the worst of restrict and interpolate.
> 
> Thanks, Barry.  
> 
> Did some experiments, and I observed that "none" works better than "basic" for an incompressible Navier-Stokes defined on a complex domain. Is it normal?

   It is certainly possible, I don't know if it is normal. Once you get away from simply uniformly elliptic model problems all kinds of strange things can happen.

> 
> I think for PC_ASM_BASIC, we could average the values on the overlap region. Right now, we are just simply adding all values from different processors together for the overlap region.

   Yes, there are all kinds of variants. One can weight the values based on how far away they are from the subdomain boundary, one can use different boundary conditions for the subdomains ....    Even doing something simple like averaging requires more information, you need to know how many subdomains contribute to each grid point.

   Barry


> 
> There are some literatures that shows the averaged version works similarly as the restrict/interpolate version.  
> 
> Fande,
> 
> 
>  
> 
> 
> >
> > Regards,
> >
> > Fande Kong,