<div dir="ltr"><div class="gmail_quote">On Tue, Mar 29, 2011 at 09:49, Kontsantinos Kontzialis <span dir="ltr"><<a href="mailto:ckontzialis@lycos.com">ckontzialis@lycos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
I have another question. I use the Discontinoues Galerkin discretization for the equations of gas dynamics. I use a matrix free formulation and that's why I need a very good preconditioner, which I compute it numerically and untill now it works ok, but it is a bit slow The preconditioner is just the Jacobian of the system. I wonder if PETSc has a built in capability to compute its entries.</blockquote>
</div><br><div>You can use coloring, see snes/examples/tutorials/ex5.c for an example. It is more expensive in 3D and if the stencil is large.</div></div>