If we use an aggresive coarsening how does this affect the convergence of a MG cycle?Is the convergence slower?Do we need more cycles in this case?
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Yes, quicker coarsening means that your coarse grid correction is not as good and that you need more MG cycles. For a well-tuned multigrid, you reduce the error by about a factor of 5-10 for each CG iteration with MG V-cycle preconditioner. If you were to coarsen too aggressively, the reduction factor would be smaller.
answered Feb 2, 2020 at 8:37
-
$\begingroup$ And the reason for this is that with a more aggressive coarsening we lose information about some nodes?(assume canonical restriction) $\endgroup$– spyrosFeb 4, 2020 at 3:28
-
$\begingroup$ Correct. The solution on the more-coarser level is just not as good an approximation as the one on the less-coarser level. $\endgroup$ Feb 4, 2020 at 17:23