Let $\mathbb{R}_+$ denote the strictly positive real numbers, let $\mathcal{X} \subset \mathbb{R}^n$ and $\mathcal{P} \subset \mathbb{R}^m$ be compact and convex subsets, let \begin{equation} f: \mathcal{X} \rightarrow \mathbb{R}, \quad g: \mathcal{X} \rightarrow \mathbb{R}^{1\times m},\quad h: \mathcal{X} \rightarrow \mathbb{R}_+, \quad S: \mathcal{X} \rightarrow \mathbb{R}^{m\times m} \end{equation} be continuously differentiable functions and assume $S(x)$ corresponds to a strictly positive definite matrix for every $x\in\mathcal{X}$.

Consider the correspondence $\Pi :\mathcal{X}\rightarrow 2^{\mathcal{P}}$, given by \begin{equation} \Pi(x) = \left\{p \in \mathcal{P} \ \Big\vert \ f(x) + g(x)p \geq \sqrt{h(x) + p^\top S(x) p \ \ } \ \right\}, \end{equation} and assume that $\Pi(x)$ is non-empty for every $x\in\mathcal{X}$.

$\textbf{Question: Is the correspondence }$ $\Pi$ $\textbf{Hausdorff-Lipschitz in}$ $\mathcal{X}$, i.e., does there exist a positive scalar $L$, such that \begin{equation} d_H( \Pi(x), \Pi(x')) \leq L\Vert x-x'\Vert_2 \end{equation} hold for every $x, x'\in \mathcal{X}$, where $d_H$ denotes the Hausdorff distance?


1 Answer 1


$\newcommand\P{\mathcal P}\newcommand\X{\mathcal X}$A counterexample is as follows: $n=m=1$, $\X=[-1,1]$, $\P=[0,2]$, $f(x)=\sqrt2-1/\sqrt2+|x|^{3/2}$, $g(x)=1/\sqrt2$, and $h(x)=1=S(x)$ (for all $x\in\X$).

Indeed, then $$\Pi(x)=\{p\in[0,2]\colon d(p)\le|x|^{3/2}\},$$ where $$d(p):=\sqrt{1+p^2}-(\sqrt2-1/\sqrt2+p/\sqrt2),$$ so that $d$ is a convex function with $d(1)=0=d'(1)$ and $d''(1)=2^{-3/2}$. So, $d(p)\sim2^{-5/2}(p-1)^2$ for $p\to1$ and hence $$\Pi(x)=[p_-(x),p_+(x)]$$ for some functions $p_\pm$ such that $p_\pm(x)-1\sim2^{-5/4}|x|^{3/4}$ as $x\to0$. Since $3/4<1$, the function $\Pi$ is not Lipschitz in any neighborhood of $0$.

  • $\begingroup$ Thanks. Does it help if all functions are smooth? $\endgroup$
    – Heinrich A
    Apr 15 at 19:33
  • 1
    $\begingroup$ @HeinrichA : I don't think so. I think you can modify this example to have $(p-1)^4$ in place of $(p-1)^2$ and then use something like $f(x)=c+x^2$. $\endgroup$ Apr 15 at 19:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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