Convex functions -> quasi-convex functions -> ... can we weaken the assumptions?

193 Views Asked by At

First of all let me say that I'm new to optimization. I realized that quasi-convex functions share with convex functions some nice properties, so I wonder if we can push the weakening a little further. We all know that quasi-convex functions are functions whose sublevel sets are convex. What if we try to build an optimization theory for objective functions whose sublevel sets are...

  • bounded
  • simply connected
  • star-shaped
  • combinations of above... or anything else.

I'm curious if is there a direction that lead to satisfactory results. Ok, nothing can compare to convex optimization, but some results for more general functions may be interesting. Thanks in advance!

1

There are 1 best solutions below

2
On

Minimizing continuous functions over compact level-sets is a nice problem.

Boundedness of level-sets alone is not enough. The function $$ f(x) = \begin{cases} 1-x^2 & |x|<1\\ x^2 & |x|\ge 1 \end{cases} $$ has bounded level sets, but there exists no global minimizer.

The function $f(x)=e^x$ has star-shaped, connected level-sets, but no minimizer exists.