I like his min-max analysis. He could change his function to return mc(nom,tola) instead of just nom to get an unbanded graph, if desired. One problem with this approach is there are so many min-max combinations; if there are even 30 components in a circuit, each with only 2 min-max bounds (some like transistors will have many), then there are 2^30 combinations. With a gaussian distribution this means 2^30 runs (if I remember my ancient statistics classes right) will on average miss 5% of coverage. And, the pathological combination might just be among the 5%... So either some additional heuristics or problem partitioning is needed. Or try to figure out mentally what the worst case is (which is actually hard for many closed feedback circuits, active filters, etc). But for something like a many-stage filter each stage could be analyzed to determine *its* min-max, then those are stepped globally. It's clearly NP-complete - meaning there is no polynomial-time solution and that verifying the solution is complete, i.e. the worst case has indeed been found, is also NP.