Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Is it possible to have a function with the following properties?

(i) The function maps a bounded $n$-dimensional space $\mathcal{X}$ (say $\left[0,1\right]^n$) onto a bounded interval $\mathcal{Y}$ (say $\left[0,1\right]$);

(ii) The function has exactly $k$ minima in $X$, whose locations are distributed randomly, and whose vaues are distributed randomly over $\mathcal{Y}$. Ideally, the value of the global minimum (in $\mathcal{X}$) should correspond to the lowerbound of $\mathcal{Y}$.

Additional notes:

(i) It does not matter if the function has additional minima outside of $\mathcal{X}$;

(ii) The function does not need to be differentiable over $\mathcal{X}$: this means that the local minima could also be, for example, kinks;

(iii) The number of maxima in $\mathcal{X}$ does not matter;

(iv) It is also OK if the minima and maxima are interchanged (i.e. if the function has $k$ maxima and any number of minima).

share|improve this question

1 Answer 1

up vote 3 down vote accepted

On $\mathbb{R}^{n}$ define $f(x)=e^{-\frac{1}{1-|x|^{2}}}$ for $|x|<1$ and $f(x)=0$ for $|x|\geq 1$. This is a smooth function with a maximum at the origin that vanishes outside of the unit ball. Given a point $a\in\mathbb{R}^{n}$ and $\epsilon>0$ define $g_{a,\epsilon}(x)=f((x-a)/\epsilon)$ this has a single maximum at $a$ and vanishes outside of $B_{\epsilon}(a)$. Given arbitrary distinct points $\{a_{1},...,a_{k}\}$ in $[0,1]^{n}$ and arbirary $r_1,...,r_k>0$ define $F(x)=\sum_{i=1}^{k}r_i\cdot g_{a_{i},\epsilon}(x)$ where $\epsilon=\min_{i\neq j}(\{\frac{|a_i-a_j|}{2}\})$. Then the $B_\epsilon(a_i)$ are all pairwise disjoint so $F$ has exactly $k$ maxima.

share|improve this answer
    
Hi Brian, I may be using this in a paper that I should be publishing. I will gladly acknowledge your help, if you don't mind it. I would require your full name and institution. –  user25207 Jul 19 '12 at 1:02

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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