<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Tue, Oct 28, 2014 at 9:03 AM, Jed Brown <span dir="ltr"><<a href="mailto:jed@jedbrown.org" target="_blank">jed@jedbrown.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">Matthew Knepley <<a href="mailto:knepley@gmail.com">knepley@gmail.com</a>> writes:<br>
<br>
> On Tue, Oct 28, 2014 at 8:00 AM, Tabrez Ali <<a href="mailto:stali@geology.wisc.edu">stali@geology.wisc.edu</a>> wrote:<br>
><br>
>> Matt<br>
>><br>
>> With MUMPS it is indeed the same but not faster. E.g., see 1.txt and 2.txt<br>
>><br>
><br>
> There is a big problem here. This system is never actually being solved.<br>
> Look at 1.txt:<br>
><br>
> 0 KSP preconditioned resid norm 9.320055451716e+05 true resid norm<br>
> 1.755998647494e+02 ||r(i)||/||b|| 1.000000000000e+00<br>
> 380 KSP preconditioned resid norm 9.034871458425e-05 true resid norm<br>
> 3.118451181896e+09 ||r(i)||/||b|| 1.775884728811e+07<br>
><br>
> The preconditioned residual did drop 10 order of magnitude, but the true<br>
> residual (b - Ax) is still very very large.<br>
> This looks like the system is singular and the preconditioner is masking<br>
> this. This needs to be figured out first.<br>
<br>
</span>It could also be very poor scaling between the fields. It's important<br>
to nondimensionalize so that you don't ask to solve for velocity 1e15<br>
more accurately than pressure, or vice-versa.<br>
</blockquote></div><br>Yes, this was a big deal in PyLith. There is a large gap between displacements and</div><div class="gmail_extra">the Lagrange multipliers. Without non-dimensionalization, we could not solve the system.</div><div class="gmail_extra"><br></div><div class="gmail_extra"> Matt<br clear="all"><div><br></div>-- <br>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>-- Norbert Wiener
</div></div>