For a metric space $E$, let $\mathcal{H}(E)$ be the metric space consisting of the set of nonempty compact subsets of $E$ and the Hausdorff metric. Consider the following two statements.

  1. Let $X$ and $Z$ be topological spaces and let $Y$ be a compact metric space. Let $f : X \times Y \rightarrow Z$ be a continuous map. Let the map $g : Y \times Z \rightarrow \mathcal{H}(X)$ be given by $g(y,z) = (f(y,\cdot))^{-1}(z) = \{ x \in X : f(x,y) = z \}$. The map $g$ is continuous.

  2. The maps $\min : \mathcal{H}(\mathbb{R}) \rightarrow \mathbb{R}$ and $\max : \mathcal{H}(\mathbb{R}) \rightarrow \mathbb{R}$ are continuous.

I am looking for references to literature where these statements (which do sound true) can be found. Much thanks in advance.


Addendum 1


I made question 1 more general than I needed it to be, and in the process introduced a few mistakes, which Alex Becker and Robert Israel helped me see. I am reformulating statement 1 as follows.

Let $X$ be a compact metric space and $Y$ and $Z$ be topological spaces. Let $f : X \times Y \rightarrow Z$ be a continuous map. Let the map $g : Y \times Z \rightarrow \mathcal{P}(X)$ be given by $g(y,z) = (f(\cdot,y))^{-1}(z) = \{ x \in X : f(x,y) = z \}$. Let $z \in Z$ such that $g(y,z) \neq \varnothing$ for all $y \in Y$. The map $g(\cdot,z) : Y \rightarrow \mathcal{H}(X)$ is (well-defined and) continuous.

I probably need the spaces $X$, $Y$, $Z$ not to have pathologies, e.g. multiple connected components; maybe just convex in Euclidean space to get started.

An intuition for this problem is as follow. Consider that $f(x,y) = z$ is a solvable equation where $z$ is fixed, $y$ is a parameter ranging over $Y$, and $x$ is the unknown. I would like to say that the solution set, i.e. $g(y,z)$, depends continuously on the parameter $y$.

up vote 2 down vote accepted
  1. Is false. Consider $Y = Z = [0,1]$, $X = [0,1] \cup \{2\}$ and $$f(x,y) = \begin{cases} y & \mbox{if }x = 2 \\ x & \mbox{otherwise}\end{cases}$$ Then $g(y,z) = \{x: f(x,y) = z\}$ is either $\{2,z\}$ if $y=z$ or $\{z\}$ if not, and this is not continuous.

  2. is quite easy, because $|\min(A) - \min(B)| \le \mbox{dist}(A,B)$, but I don't have a reference to give you.

EDIT: if you want a counterexample where $X$, $Y$, $Z$ are convex sets in Euclidean space, take all three to be $[-1,1]$ with $f(x,y) = x + y - 3 x^3$. Then, for example, $g(2/9,0) = \{-1/3,2/3\}$, but $g(y,0) \subset [0,1]$ for $y > 2/9$.

  • Somehow I convinced myself that this should be true; some kind of Topological Implicit Function Theorem. But as you showed, one need not look too far to find counterexamples. Thank you! – Gilles Gnacadja Aug 27 '13 at 3:40
  1. Seems to be false. For example, let $X=Z=\mathbb R, Y=[0,1]$ and let $f$ be the zero map. Then for any $y\in [0,1]$ we have that $g(y,0)\notin \mathcal H(\mathbb R)$ so $g$ is not even defined. If you object to this example, then letting $f(x,y)=xy$ and $X=[0,1]$ we get that $g(0,0)=[0,1]$ but $g(y,z)=\{z/y\}$ for all other $y\in [0,1]$, so this is clearly not continuous.

  2. I do not think you need a reference for this, since $|min(A)-min(B)|$ and $|max(A)-max(B)|$ are clearly lower bounds for the Hausdorff distance between $A$ and $B$.

  • lower bounds, you mean... – Robert Israel Aug 26 '13 at 5:20
  • @RobertIsrael Yes, thank you. – Alex Becker Aug 26 '13 at 19:17

Your Answer

 
discard

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.