<div class="gmail_quote">On Fri, Dec 23, 2011 at 12:52, Dominik Szczerba <span dir="ltr"><<a href="mailto:dominik@itis.ethz.ch">dominik@itis.ethz.ch</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div id=":2b4">I am scaling my systems when dealing with small (microscopic) scales<br>
in SI dimensions. </div></blockquote><div><br></div><div>The best choice is to choose suitable units so that the system is intrinsically well-scaled.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div id=":2b4">This is because otherwise I have very small (near<br>
epsilon) entries in the matrix and the solution fails or takes<br>
significantly longer to converge. I used to do it by hand so far, if<br>
there is a way to do it in Petsc - especially automatically detecting<br>
the optimal scale - I am all ears.</div></blockquote></div><br><div>You are best off doing it by hand, it is better to avoid -ksp_diagonal_scale when it's reasonable to do so. If you have trouble determining a reasonable scale at assembly time, it might make sense to use.</div>