On Sun, Mar 11, 2012 at 8:20 AM, Timo Heister <span dir="ltr"><<a href="mailto:heister@math.tamu.edu">heister@math.tamu.edu</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">
<div class="im">> [Timo: Question -- do we want this anyway? If so, please apply to the<br>
> repository since I will be offline for the next couple of days.]<br>
<br>
</div>I think so, yes. This was on my "10% time to be done on weekends" list.<br>
<div class="im"><br>
> The second part you talked about, namely not building the matrix as<br>
> individual blocks but instead as one matrix from which you let PETSc pick<br>
> and choose is likely going to be much more difficult. I'll try to explain<br>
> how this currently works when we meet in 20 minutes down at the AGU<br>
> building. For now I've gotta run :-)<br>
<br>
</div>Matt, you want to use PETSc solvers for the whole linear system? It<br>
shouldn't be too hard to make this change. To explain: right now a<br>
PETScWrappers::BlockSparseMatrix is a two-dimensional array of PETSc<br>
matrices (same with vectors). Of course, we are using this block<br>
structure in many places.</blockquote><div><br></div><div>Yes, do no blocking. Just give the whole matrix, do not hardcode the blocking</div><div>or the matrix type, or the solver type or the preconditioner. That is design from</div>
<div>the 1980s.</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="HOEnZb"><font color="#888888"><br>
--<br>
Timo Heister<br>
<a href="http://www.math.tamu.edu/~heister/" target="_blank">http://www.math.tamu.edu/~heister/</a><br>
</font></span></blockquote></div><br><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<br>