On Wed, Oct 10, 2012 at 4:33 PM, <span dir="ltr"><<a href="mailto:panourg@mech.upatras.gr" target="_blank">panourg@mech.upatras.gr</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
What do you mean when you say that i do not use a matrix free method at all?<br>
<br>
With matrix-free method only the operator is "free". Not the<br>
preconditioning matrix.<br></blockquote><div><br></div><div>Yes, but if you are assembling the Jacobian using coloring, but then only using it as a preconditioner and using a separate matrix-free finite differencing step to apply the operator, then you are just goofing off. Of course you can do it, but it doesn't make sense.</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
For this i always set -snes_mf_operator option.<br>
<br>
Either way is any way to have preconditioner without forming the its prec<br>
matrix??<br>
<br></blockquote><div><br></div><div>Sure, but you have to understand the problem, e.g., multilevel structure, plane waves, an embedded discretization. If you want to use PETSc's automatic preconditioners instead of rolling your own, then you should assemble the operator.</div>
</div>