<div class="gmail_quote">On Fri, Dec 23, 2011 at 07:43, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com">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 id=":221">Even if you lose orthogonality, it will just stagnate since you always have the solution<div>from the first few vectors. Likewise, I do not see how you can get an diverging sequence</div><div>from restarting since you always have the last residual, you just stagnate.</div>
</div></blockquote></div><br><div>The true residual can grow even though the estimate of the preconditioned residual decreases monotonically. When you restart, you re-evaluate the actual residual, which might be much larger.</div>