On 10/22/21 2:53 AM, richard....@gmail.com wrote:

I think I know what it is: the ML guide on p 16 says in footnote 11:
"Note that the hierarchy produced by ReComputePreconditioner() can differ from that produced by ComputePreconditioner() for non-zero threshold values." So when the matrix changes too much, the const double aggregation_threshold in PreconditionAMG::AdditionalData would lead to different matrix entries being considered. That is why it was also working for (constant) Laplace and (almost constant Hyper-) Elasticity, but not so well for Navier-Stokes, since I initialized the AMG in the first timestep, where convective terms are completely zero starting from the quiescent state. Hope someone finds this in the future and updates their AMG every few timesteps as well!

Richard -- I cannot recall who implemented the AMG preconditioner, but suspect that that parameter was simply never used. I don't think we use it in ASPECT, for example.

Since you figured out how to use it, is there a place where one could add a paragraph that describes what you found out and how one would use this feature? You might know this one piece of the library better than anyone else at this time :-) It would be nice to write your experience down in the place where others might look for it.

Best
 Wolfgang

--
------------------------------------------------------------------------
Wolfgang Bangerth          email:                 bange...@colostate.edu
                           www: http://www.math.colostate.edu/~bangerth/

--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see 
https://groups.google.com/d/forum/dealii?hl=en
--- You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to dealii+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/dealii/2ae2c842-9d59-0d27-cf45-8b034d4114f3%40colostate.edu.

Reply via email to