[Trilinos-Users] Incomplete Cholesky Conditioning

[Chris Siefert]

Kimberly,

The IC preconditioners are a "structural" incomplete factorization.  
This means that powers of the matrix (it's graph, strictly speaking) are 
used to determine the sparsity pattern (e.g. IC(0) has the sparsity 
pattern of A, IC(1) has the sparsity pattern of A^2, etc. ).

The ICT preconditioners use a dropping threshold to pick the sparsity 
pattern (e.g. it keeps all entries bigger than your threshold).

Which one you should choose is problem dependent.

The Ifpack docs at trilinos.sandia.gov detail the differences in a bit 
more precision:

http://trilinos.sandia.gov/packages/docs/dev/packages/ifpack/doc/html/ifp_ilu.html

As for the STAND_ALONE option, this applies to what to do in parallel.  
The stand-alone versions drop all off processor entries.  The 
non-stand-alone versions wrap the incomplete factorization inside an 
overlapping Schwarz method.  This usually gives a better (albeit more 
expensive) preconditioner.

Again, the "correct" choice will depend on your problem.

I hope this helps.

Sincerely,
Chris Siefert


On 01/09/2012 02:18 PM, Kennedy, Kimberly R wrote:
> Hello,
> I am trying figure out which Preconditioner (IC,IC_STAND_ALONE,ICT, or ICT_STAND_ALONE)to use from IFPACK I should use for applying Incomplete Cholesky Conditioning and what the differences are between each. Any help would be appreciated.
> --------------------------------------------------------------
> Kimberly Kennedy
> Graduate Part Time Instructor
> Texas Tech University
> Department of Mathematics and Statistics
> Email: kimberly.r.kennedy at ttu.edu<mailto:kimberly.r.kennedy at ttu.edu>
>
>
>
>
>
>
>
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users