On Sat, Oct 6, 2012 at 4:44 PM, Barry Smith <bsmith at mcs.anl.gov> wrote: > On Oct 5, 2012, at 7:38 PM, John Fettig <john.fettig at gmail.com> wrote: >>What is wrong with the diagonal >> block when inode.size==0? >> >> Thanks, >> John > > John, > > When innode.size is zero the "diagonal blocks" are all 1 by 1. ... > Are you using MatMultDiagonalBlock() for some other case?
Barry, Thanks for the explanation, it makes sense now. I was trying to use it in the context of a multigrid, and didn't think about the coarse level. It must have been failing when the diagonal block was 1. In the end, though, I ended up not using this function. I was just curious why there should be any restriction. John
