\end{figure}
-As part of this effort, we ran benchmarks on the TACC Frontera system, where we were able to apply the matrix-free geometric multigrid framework to a variable viscosity Stokes system and achieved weak and strong scaling up to 114k MPI ranks with up to
-\num{2.1e11} unknowns. This is likely the currently largest block system solved with \dealii{} and required various optimizations
+As part of this effort, we ran benchmarks on the TACC Frontera system, where we were
+able to apply the matrix-free geometric multigrid framework to a variable viscosity
+Stokes system and achieved weak and strong scaling up to 114K MPI ranks with up to
+\num{2.1e11} unknowns. This is likely the largest block system currently solved with
+\dealii{} and required various optimizations
and fixes on top of the ones mentioned above:
(i) Bug fixes to concurrent point to point communications.
(ii) Fixes multigrid transfer with adaptive refinement and more than \num{4e9} unknowns.
in \dealii{} as described in \cite{ClevengerHeisterKanschatKronbichler2019}.
The problem considered is a variable viscosity Laplace equation and it is solved
with three different approaches:
- (i) With a matrix-based geometric multigrid based on Trilinos or
- PETSc; (ii) with a matrix-free
- geometric multigrid; (iii) using algebraic multigrid (Trilinos ML).
+ (i) using a matrix-based geometric multigrid based on Trilinos or
+ PETSc; (ii) using a matrix-free
+ geometric multigrid; (iii) using algebraic multigrid (Trilinos ML). The tutorial
+ demonstrates the superiority of the matrix-free method for the problem under consideration,
+ and shows that, for matrix-based formulations, the performance of algebraic and
+ geometric multigrid methods are roughly comparable.
- \todo[inline]{Timo/Conrad/... to write}
\item \texttt{step-58} is a program that solves the nonlinear
Schr{\"o}dinger equation, which in non-dimensional form reads