<div dir='auto'><div>Hi Ansgar,</div><div dir="auto"><br></div><div dir="auto">you might want to look into dune-composites. That's essentially a linear elasticity module, specifically intended for layered composites including all the bells and whistles like serendipity elements, stress calculations etc. When it comes to the modeling code, Anne Reinarz is the expert.</div><div dir="auto"><br></div><div dir="auto">Indeed we found the dune AMG to hit it's limits with those particular systems. We added a (parallel, but sequential should work too) wrapper for hypre, their AMG performs better after some tweaking. You might be able to reuse that wrapper for your code.</div><div dir="auto">As the main solver, we went with a two-level schwarz method with GenEO coarse space. That has high setup cost, but is provably robust for such problems and outperformed the AMGs in our case.</div><div dir="auto">My GenEO implementation is in dune-pdelab since 2.6, almost black-box (you need to provide extra matrices, there are tools to easily generate them though based on your existing local operators).</div><div dir="auto"><br></div><div dir="auto">There's a paper on dune-composites already out, it's mostly focusing on the modeling side but also comparing to AMG. Another one more focused on the math and software is on the way.</div><div dir="auto"><br></div><div dir="auto">https://www.sciencedirect.com/science/article/pii/S0263822317321797<br></div><div dir="auto"><br></div><div dir="auto">Best,</div><div dir="auto"><br></div><div dir="auto">Linus</div><div dir="auto"><div class="gmail_extra" dir="auto"><br><div class="gmail_quote">Am 22.11.2018 19:49 schrieb Ansgar Burchardt <ansgar.burchardt@tu-dresden.de>:<br type="attribution"><blockquote class="quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">Hi,
<br>
<br>
do any DUNE users have experience with using AMG for linear elasticity
<br>
problems?
<br>
<br>
I understand that unsmoothed aggregation AMG as implemented by dune-
<br>
istl should not work well as rigid body movements are not reproduced
<br>
well enough on the algebraic coarse grid levels.
<br>
<br>
Is there any experience which alternative (sequential) AMG
<br>
implementations are easy to use from an existing code using DUNE? I'm
<br>
so far considering PETSc's AMG (or hypre which can be used via PETSc)
<br>
or Trilinos' ML.
<br>
<br>
Regards,
<br>
Ansgar
<br>
<br>
<br>
<br>
_______________________________________________
<br>
Dune mailing list
<br>
Dune@lists.dune-project.org
<br>
https://lists.dune-project.org/mailman/listinfo/dune
<br>
<br>
</p>
</blockquote></div><br></div></div></div>