<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><div>(switching back to petsc-users with the previous attachment striped)</div>I’m assuming your problem is kind of ill-posed?<div>PCICC and PCILU have different default parameters, and thus can yield different convergence curves, especially for difficult problems.</div><div>When switching from AIJ to SBAIJ, ILU will switch to ICC.</div><div>If I switch to something more “robust” (PCCHOLESKY from MUMPS), then I get the same convergence (up to machine precision).</div><div>I think this highlight there is no issue with the new PCFIELDSPLIT/MatCreateSubMatri[x,ces]() code, but merely that you need to be a bit careful with the inner solvers.</div><div><br></div><div>Thanks,</div><div>Pierre</div><div><br></div><div></div></body></html>