2
$\begingroup$

I have a simple question. We know that functions where every stationary point is a global minimum are invex functions. Is there a name for functions where every local minimum is a global minimum?

And also can we naturally put convex functions into this scheme? As I see, claiming that it has only one global minimum that is the only stationary point is not enough.

Thank you!

Also, there is no tag for invexity

$\endgroup$
9
  • $\begingroup$ you are thinking of a function which is not convex and yet every local minimum is a global minimum? $\endgroup$ – Carlo Beenakker Apr 4 '17 at 17:23
  • 1
    $\begingroup$ There days people are calling such functions: "functions without bad local minima", though I must admit, that is not much of name :-) $\endgroup$ – Suvrit Apr 4 '17 at 17:35
  • $\begingroup$ @Carlo yes, it can be very crazy $\endgroup$ – Eugene Apr 4 '17 at 19:38
  • $\begingroup$ @Suvrit I like the phrase actually $\endgroup$ – Eugene Apr 4 '17 at 19:38
  • 2
    $\begingroup$ When local≠global minimum, the local minimum has been called as "spurious local minimum". See for instance the arXiv pre-prints 1605.07272 and 1704.00708. $\endgroup$ – Shamisen Apr 6 '17 at 2:14
2
$\begingroup$

I don't know of a specific term that describes this property, but it is different from convexity and invexity:

This function has a single global mimumum, but it has multiple stationary points, and therefore it is not invex:

$$ f(x) = 3 x^4 - 4 x^3 $$

And also can we naturally put convex functions into this scheme?

For a convex function, a local minumum will also be global.

However, the inverse is not true. A function may have only one local minimum, but be non-convex, for example

$$ f(x) = - e^{-x^2} $$

is not convex, but has only one local minumum (and also a single stationary point)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.