What kind of tests are sufficient: Some personal thoughts

I think you misunderstand my point: in general, optimization problems do not have algorithms that guarantee a (global) optimum. Some special cases exist (eg convex problems), but generally no matter how excellent and bug-free an implementation is, for every general algorithm you can devise a problem where the global optimum is not found.

No, I didn’t. I know your point very well. Indeed, even for the algorithms that guarantee a global solution in theory, and even if such an algorithm is applied to a convex problem, you most likely only obtain an approximate solution within finite time — even this is usually based on the assumption that your code is doing real-number calculation, whereas it really performs only floating-point arithmetic. However, this is irrelevant to the topic discussed here.

We all like to believe that our work matters, but perhaps this is taking it too far :wink:

Thank you for pointing this trivial fact out, although it is again a topic unrelated to the discussion.

Thanks.