<div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">On Sun, Mar 3, 2013 at 8:14 AM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div>You can see our example of a parallel unstructured FVM solver in TS ex11 in petsc-dev.</div><div>
<br></div><div>Specifically, I can see several ways to do this. You could partition faces, and have ghost cells, which we can handles. Or</div>
<div>you could partition cells, and assemble these kind of function over faces additively if they are linear, which I think yours is,</div><div>which is what we do in ex11. It depends on your particular application, but we have the tools to do this.</div>
<div></div></blockquote></div><br></div><div class="gmail_extra">This isn't an accurate description of what ex11 does. We build a vector Grad on the owned cells, then communicate like this<br><br> ierr = DMGlobalToLocalBegin(dmGrad,Grad,INSERT_VALUES,locGrad);CHKERRQ(ierr);<br>
ierr = DMGlobalToLocalEnd(dmGrad,Grad,INSERT_VALUES,locGrad);CHKERRQ(ierr);<br><br></div><div class="gmail_extra">and evaluate fluxes using locGrad.<br></div></div>