8
$\begingroup$

In the euclidean plane $\mathbb{R}^2$ it's possible to state the following theorem:

Theorem: Consider any finite set of $n$ points $\{x_1,\dots,x_n\}\subset \mathbb{R}^2$ and any n-ple of positive real numbers $(r_1,\dots,r_n)$ such that $$\cap_{i=1}^nB_{r_i}(x_i)\neq\emptyset$$ where $B_{r_i}(x_i)$ is the closed ball of center $x_i$ and radius $r_i$. Then for any other set of $n$ points $\{x_1',\dots,x_n'\}\subset \mathbb{R}^2$ such that $$|x'_i-x'_j|\le|x_i-x_j|$$ for every $i,j=1,\dots,n$ it also happens $$\cap_{i=1}^nB_{r_i}(x'_i)\neq\emptyset$$

Question: does this result still hold true if we consider $$X:=\{x\in\mathbb{R}^2 \text{ such that } 0\le\text{arg}(x)\le 7\pi/4 \}$$ instead of $\mathbb{R}^2$? By this I mean that both $\{x_1,\dots,x_n\}$ and $\{x_1',\dots,x_n'\}$ will be in $X$ and the condition $|x'_i-x'_j|\le|x_i-x_j|$ is replaced by $d(x'_i,x_j')\le d(x_i,x_j)$ where $d$ is the induced patch metric by the euclidean metric of $\mathbb{R}^2$ on $X$. Also, the intersections are considered in $X$ and the balls are defined using $d$.

This doesn't seem easy to me because the only proof I know for the theorem in $\mathbb{R}^2$ isn't really easy (it's contained in "A Lipschitz Condition Preserving Extension for a Vector Function" by Valentine and it's for $\mathbb{R}^n$) and doesn't seem to adapt to the case of $X$. As user Moishe Cohen pointed out, I should sketch it.


Lemma 1: (by Alexandrov and Hopf): Let $A_1,\dots,A_n$ be closed sets in $\mathbb{R}^m$ which cover the simplex $T$. Suppose that for each side $a_{i_1},\dots,a_{i_\tau}$ of $T$ it's true $a_{i_1},\dots,a_{i_\tau}\subset A_{i_1}\cup\dots\cup A_{i_\tau}$. Then $A_1\cap \dots\cap A_n\neq \emptyset$.

Lemma 2: Let $\Delta(x_1',\dots,x_n')$ be the euclidean simplex of vertex $x_1',\dots,x_n'$. Under the hypothesis of the theorem $\Delta(x_1',\dots,x_n')$ is covered by the sets $B_{r_i}(x'_i)$

proof of the lemma: suppose $\Delta(x_1',\dots,x_n')$ is not covered by the sets $B_{r_i}(x'_i)$, then we can choose $$x\in \cap_{i=1}^nB_{r_i}(x_i)\neq\emptyset,\qquad x'\in \Delta(x_1',\dots,x_n')\setminus \cup_{i=1}^nB_{r_i}(x'_i)$$ call $R_i$ the vector from $x$ to $x_i$ and $R_i'$ the vector from $x'$ to $x_i'$. Since $|x'_i-x'_j|\le|x_i-x_j|$ it follows $$ |R_i'|>|R_i|, \qquad |R_i'-R_j'|\le|R_i-R_j|$$ for every $i,j=1,\dots,n$. This implies $R_i'\cdot R_j'>R_i\cdot R_j$. Let $(a_1',\dots,a_n')\in \mathbb{R}_+^n$ such that $\sum_{i=1}^na_i'R_i'=0$, then from the previous inequality it follows $\sum_{i=1}^n|a_i'R_i'|^2<0$, which is impossible.

proof of the theorem: since $\cap_{i=1}^nB_{r_i}(x_i)\neq\emptyset$ it follows $B_{r_i}(x_i)\cap B_{r_j}(x_j)\cap \overline{x_ix_j}\neq \emptyset$ and since $|x'_i-x'_j|\le|x_i-x_j|$ it follows $B_{r_i}(x'_i)\cap B_{r_j}(x'_j)\cap \overline{x'_ix'_j}\neq \emptyset$. The result then follows from the two lemmas.


As I said this proof is for the general case of $\mathbb{R}^n$, $n>1$ and is quite complicated. I don't know if for $n=2$ there's a simpler one which would adapt to the case of $X$. I can't understand how to adapt this one because it doesn't seem possible to define $\Delta(x_1',\dots,x_n')$ in $X$ in the first place.

  • 0
    Do you mean $d$ is the induced path-metric? (The notion of "induced distance" is ambiguous.) Also, I assume that you are considering the intersection only in $X$ and the balls are also defined using $d$. Is this correct? Also, I suggest you include a proof or a reference in the case of points in $R^2$.2017-01-26
  • 0
    yes, all your assumptions are correct. This theorem is proven for $\mathbb{R}^n$ in the article "A Lipschitz Condition Preserving Extension for a Vector Function" by Valentine. I don't know a simpler proof for $\mathbb{R}^2$, but I'd really like to have one2017-01-26
  • 0
    Then you should at least sketch the proof to relate it to the extension theorem. This sounds like Kirszbraun's extension which has a generalization to spaces (of curvature bounded above) like your nonconvex set: http://www.math.psu.edu/petrunin/papers/alexandrov/lang-schroeder_kirszbraun.pdf2017-01-26
  • 0
    ok, I'll sketch the proof of Valentine. The generalization of Kirszbraun in the article you linked only works for functions from $CBB(k)$ spaces to $Cat(k)$ spaces. It seems to me that this convex subset is $Cat(0)$ but not $CBB(0)$, so I can't use the result you linked2017-01-26

1 Answers 1

3

The answer is negative already for $n=3$:

There are points $x_1, x_2, x_3, x_1', x_2', x_3'\in C$ and positive numbers $r_1, r_2, r_3$ for every nonconvex cone $C\subset E^2$ such that:

  1. $d(x_i, x_j)= d(x_i', x_j')$ for all $i, j=1, 2, 3$.

  2. $$ \{ x\}= \bigcap_{i} \bar{B}(x_i, r_i)\ne \emptyset. $$

  3. $$ \bigcap_{i} \bar{B}(x'_i, r_i)= \emptyset. $$

In my examples $r_1+r_2=d(x_1, x_2)$, $r_i=d(x,x_i), i=1, 2, 3$; thus $x\in x_1x_2$. The point is that for every nonconvex cone $C$, there exists a geodesic triangle $x_1x_2x_3$ and a point $x\in x_1x_2$ such that in the Euclidean comparison triangle $x_1'x_2'x_3'$ and the comparison point $x'\in x'_1x'_2$, we have $$ d(x'_3, x')> d(x_3, x). $$ [To find such examples place $x_1, x_2$ on the different boundary rays of the cone $C$ and place $x_3$ (in the cone interior) so that the geodesic $x_1x_3$ passes through $x=$ the apex of the cone.]

enter image description here

It follows that $$ \bigcap_{i} \bar{B}(x'_i, r_i)= \emptyset, $$ since $$ \bar{B}(x_1',r_1)\cap \bar{B}(x_2',r_2)=x', $$ but $x'\notin \bar{B}(x_3',r_3)$. Now, the cone $C$ contains arbitrarily large flat disks, so we can place the points $x_i'$ as in your question in such disks, so that they necessarily span flat comparison triangles.

  • 0
    @user294185: No, at the moment I do not have examples satisfying this additional assumption. For n=3 (with this extra assumption) the answer to your question is positive. My suggestion is to read the proof of Lang and Schroeder and check if some parts of it can be used to handle this case.2017-01-31
  • 0
    @user294185: I am saying that I do not know which way it would go (with the extra assumption and $n\ge 4$). The proof for $n=3$ is quite straightforward.2017-01-31
  • 0
    ok, thank you for your answer2017-01-31