<div class="gmail_quote">On Thu, Jan 13, 2011 at 18:16, 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"> If I use PCFIELDSPLIT with PCFieldSplitSetType(pc,PC_COMPOSITE_ADDITIVE), is it the same as using PCASM?<br>
<br>
</div> Same in what sense? It solves a bunch of subproblems independently and adds together all the solutions. There can be overlapping in the fields or not depending how what you choose. The decomposition in ASM is by "geometry" while the decomposition in the PCFIELDSPLIT is between different "fields" or "types of variables". So yes they have many similarities.</blockquote>
</div><br><div>To put it differently, PCASM exposes as much concurrency as possible, making it efficient to use with spatially "local" subdomains. In contrast, PCFIELDSPLIT decomposes a multi-physics problem into sub-problems that are hopefully "better understood" so that efficient solvers are available. While you can make PCFIELDSPLIT use the same decomposition as PCASM uses, so that the algorithm is functionally equivalent, it does not expose the same concurrency, so would not scale in parallel.</div>