<html><head><meta http-equiv="Content-Type" content="text/html charset=us-ascii"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><br><div><div>On Sep 14, 2012, at 7:29 PM, Jed Brown <<a href="mailto:jedbrown@mcs.anl.gov">jedbrown@mcs.anl.gov</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div class="gmail_quote">On Fri, Sep 14, 2012 at 6:17 PM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>Unless I am reading this wrong, you have a bizarre coarse grid solver. It should be redundant LU but instead</div>
<div>you have GMRES/BJacobi?</div></blockquote></div><br><div>That is Mark's default. The proper coarse grid solve ends up being pretty expensive. Mark set the GAMG default to just do bjacobi because it was scaling better in practice. I don't know where this GMRES is coming from because it doesn't show up in my tests with 3.3.</div>
</blockquote><br></div><div>FYI, The proper coarse grid solve should not be expensive. Redundant coarse grid solves do not make any sense to me unless you do it right and PETSc does not do it right.</div><div><br></div><div>Mark</div><br></body></html>