<div class="gmail_quote">On Tue, Mar 1, 2011 at 14:24, Umut Tabak <span dir="ltr"><<a href="mailto:u.tabak@tudelft.nl">u.tabak@tudelft.nl</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Yep, then converges, so that you iterate for 146 times(for 155 restart). But it seems to me that with a low dimensional subspace it can not catch the information. And needs a lot of vectors and orthogonalization which makes it costly, are there any further comments?</blockquote>
</div><br><div>1. Make a better preconditioner</div><div><br></div><div>2. Use a different Krylov method that does not require storing a fully subspace. Such methods cannot have a monotone residual and are less robust to inexact Jacobians, but can be better if you need a large number of Krylov iterations.</div>