<div class="gmail_quote">On Sat, Nov 19, 2011 at 22:02, Gong Ding <span dir="ltr"><<a href="mailto:gdiso@ustc.edu">gdiso@ustc.edu</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div>I have the same request as explictly building schur complement matrix when K_PiB inv(K_BB) K_BPi is small.<br>The dynamic insert of matrix entry can be really useful here.<br></div></blockquote><div><br></div><div>I don't believe it is necessary or desirable to use dynamic preallocation in this case. Additionally, the FETI-DP or BDDC Schur complements have quite special structure that can be exploited to assemble them. I don't know whether it makes sense to have a public API for exploiting that structure. For one thing, building the MatSchurComplement produces a more synchronous algorithm than necessary. My preference would be to exploit the structure as part of the PCFETIDP implementation instead of by adding special-purpose code to MatSchurComplement.</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"><div>Several months ago I had committed a patch for this purpos.<br>Will anyone try to merge the patch? </div>
</blockquote></div><br><div>We didn't want a whole new matrix type for this. I think we gave a suggestion for making the patch work in a less intrusive way with existing formats. I still have it on my list to modify your patch for this purpose, but sadly, haven't had time to complete it. So it's not forgotten and I'm pretty confident that it will be done before the next release (probably in the spring).</div>