I want some examples (hopefully parametric families!) of non-convex functions which satisfy the following properties simultaneously,

  1. It should be at least twice differentiable.

  2. It should have a unique global minima at the origin.

  3. It should have a few local minima which are not global. (preferably symmetrically positioned around the global-minima/origin) If it has saddle points then thats okay too.

  4. The norm of the gradients of the function be bounded by some constant.

  5. It should be gradient Lipschitz.

Feel free to let me know if this is not an appropriate question for MO! Just that I tried quite a bit but couldn't come up with one such function!

For $\lambda\in\mathbf{R}$, $a,x\in\mathbf{R}^n$, let $$f_{\lambda;a}(x) = -\exp(-\lambda\|x-a\|^2)$$ Then linear combinations of $f_{\lambda;a}$ can satisfy these conditions, where $a$ ranges over your chosen minima. E.g., if $n=2$, $$f_{6;(0,0)}+ f_{1;(1,0)}+ f_{1;(0,1)}$$ will have a global minimum at the origin and two symmetric local minima, as in these plots and computations.

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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