<div class="gmail_quote">On Mon, Dec 20, 2010 at 01:31, Barry Smith <span dir="ltr"><<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im">> 2. Is it as easy as PCSetDM? Or provide a coarse DM and get a hierarchy? Is there an example? I agree about having FieldSplit forward the pieces. I recall starting on that. Who is responsible for assembling rediscretized coarse operators?<br>
<br>
</div> src/ksp/ksp/examples/tutorials/ex45.c</blockquote></div><br><div>Neat, this does more than I thought. We currently pick up DMComposite and forward index sets and field names into the splits. Two issues remain</div>
<div><br></div><div>1. Forward DMs into splits. This is superficial, it pretty much just requires a non-vararg DMComposite accessor. It would be a massive simplification to create a DMRedundant instead of all the Array specialization in DMComposite.</div>
<div><br></div><div>2. What if the user wants to run PCMG in only one split? I think this one is actually hard. Suppose we have</div><div><br></div><div>FormFunction1(dm1,X1,X2,F1,user1);</div><div>FormFunction2(dm2,X1,X2,F2,user2);</div>
<div><br></div><div>If we do monolithic multigrid, then we call both of these functions with (dm1,dm2) on each level and there is no conceptual difficulty relative to the single-level method. But if we only do multigrid on physics 2, then we still need a way to get X1 (interpolated into whatever form FormFunction2 needs) on coarse levels. If we want to call FormFunction2 from PCSetUp_FieldSplit:PCSetUp_MG, then we have lost all reference to X1. But handling all the MG stuff from PCSetUp_FieldSplit is breaking the abstraction. I'm not seeing an elegant way to handle this, perhaps we should have a way to cache coupling information on the DM. Any ideas?</div>