<div class="gmail_quote">On Tue, Apr 3, 2012 at 14:05, 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">
2) "I turn off norms for this -- its in the eigen estimate -- because I don't usually want them. " But doesn't GMRES compute the norms always? So why not have the GMRES version still kick out early if the norm reduces by same 10^10 before it ran out of its? (Since the norm is "free" for GMRES there is no harm done, and good things occasionly like your case below.</blockquote>
</div><br><div>Yeah, though this is an ideal scenario for TSQR-based GMRES in which case we would have done several iterations with one reduction. You will still have the norms, but I guess we will need to check them and decide which estimates to use. So I guess no problem at this point, but something to keep in mind with spectral estimates using TSQR-based or pipelined (latency-tolerant) GMRES.</div>
<div><br></div><div>Mark, because CG runs in a norm that is stronger with respect to low frequencies/weaker with respect to high frequencies, but we only want an accurate estimate of the highest eigenvalue, I would expect GMRES to give this estimate sooner. Do you not observe this?</div>