Given $x_i$'s in $\mathbb{R}^2$ and $\mathcal{S}$ defined arbitrarily as a set of two $x_i$'s, i.e., $$\mathcal{S}=\{(x_1,x_4), (x_2, x_3), (x_3, x_6), (x_6,x_4), (x_4, x_7)\},$$ the function is defined by $$f(x_1, x_2, \ldots, x_n)=\max_{(x,y) \in \mathcal{S}} \lVert x-y \rVert.$$

The physical meaning of $f$ is the maximum edge length in a tree, where the tree has $n$ vertices and $\lvert\mathcal{S}\rvert$ edges.

Is the function $f$ convex?


In general, if $f_\alpha$ are convex, then $f=\sup_\alpha f_\alpha$ is convex.

In the above case, the functions (indexed by $i,j$ rather than $\alpha$) are $(x_1,...,x_n) \to \|x_i-x_j\|$ which are convex.

That is, the function $f: (\mathbb{R}^2)^n \to \mathbb{R}$ is convex.

  • 1
    $\begingroup$ OMG, I also know the fact, as you said, that the supremum of a set of convex functions is again convex. However, I couldn't think of it rigorously and mathematically. Thanks to you, I wholly understood. God bless you! $\endgroup$ – Danny_Kim Feb 6 '18 at 7:08

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.