S matrix not positive definite

Thread: https://lists.quantum-espresso.org/pipermail/developers/2019-June/002066.html I am quite sure that this occurs when generated correction vectors are too small and contain only noise. Still, it shouldn't happen, never, with NC PPs. To be tested: does the problem persist

• if the set of correction vectors is not allowed to expand (diago_david_ndim=2)?
• if correction vectors are discarded if they are too small?
• with some fiddling of convergence thresholds, number of iterations, etc. ?