MMA incorrect behavior after 2.6.2

[smartalecH]

The MMA algorithm no longer converges (or even progresses) after the 2.6.2 update. It just idles around the starting cost function evaluation within machine precision. The same optimization problems work well under 2.6.1, however.

I'll try to generate a simple MWE. Unfortunately, all the examples I'm currently testing with are complicated TO problems...

[stevengj]

Maybe this is due to #183 changing the stopping criteria slightly as explained in https://github.com/stevengj/nlopt/pull/183#issuecomment-408465734? Maybe just increase your xreltol slightly? You could also try reverting #183…