<div class="gmail_quote">On Sun, Feb 27, 2011 at 18:15, 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 id=":zv">This exact text could go into the manual page. The problem now is that this information is in only one location (your head) and no one else knows if<br>
1) BFBt is orthogonal to to lsc<br>
2) BFBt is the replacement to lsc<br>
3) BFBt is to be replaced with lsc.<br>
without consulting you. </div></blockquote><div><br></div><div>Dave knows. :-)</div><div><br></div><div>Indeed, he just did the real work here and put in the MatNest support so that the BFBt methods from his paper now run with LSC and produces identical results. This has been merged into petsc-dev.</div>
<div><br></div><div>About this example, it's quite a hard problem and pretty small (32x32 Q1-P0 elements), but the matrices come from Underworld which we can't easily include in a test. I'm inclined to put the matrix into datafiles and add the test to the nightlies since it seems to be a discerning preconditioner benchmark.</div>
<div><br></div><div>Now about whether to delete PCBFBT now or leave it around. I don't see a problem with adding a man page that puts it in the "developer" level and has a note that it is deprecated and LSC should be used instead, with a link to this example for details on converting code.</div>
</div>