Thanks, this is now fixed by Handle sparsity patterns which are specified as missing some diagonal by ChrisRackauckas · Pull Request #1820 · SciML/OrdinaryDiffEq.jl · GitHub. The issue is that, as documented, the sparsity pattern should be the sparsity of W = M - gamma*J for arbitrary gamma. But that’s a bit hard to always remember, so that PR adds any of the missing mass matrix parts (usually just I, so the diagonal) automatically, which fixes the issue.
1 Like