7

Let's say that $A\subset\mathbb{R}^2$ is a similar universal cover for $n$ points if:

  1. $A$ is closed.
  2. The interior of $A$ is empty.
  3. For every finite set $B\subset\mathbb{R}^2$ containing exactly $n$ points there is a set $A'\subset\mathbb{R}^2$ which is similar to $A$ such that $B\subset A'$.

Let $S_n=\{A\subset\mathbb{R}^2|A$ is a similar universal cover for $n$ points$\}$.

Examples

  1. Any line segment is an element of $S_2$.
  2. A letter "T" is an element of $S_3$.
  3. A circunference with a radius is an element of $S_4$: if the four points are colinear, you can put them on the radius. If they're not, then you can form a circunference with three of them so that the fourth is inside the circle. Then you can rotate the radius until you cover that last point.

Can you find an example of an element of $S_5$?

  • This sounds like a homework problem. What have you tried so far? – mephistolotl Jun 15 '17 at 03:26
  • It is not a homework problem. The only result I got is that if $A$ is made of $n$ intersecting lines not three of them concurrent, then $A\notin S_{n+2}$. My goal is to find a set in $\bigcap_{n=1}^\infty S_n$ or to prove that set is empty. – Sergio Enrique Yarza Acuña Jun 15 '17 at 03:45
  • Intuitively I would expect $ S_5$ is empty. You have four parameters (location, rotation and size) to match your shape and if you have $ n $ points you have $ n $ parameters to move the points away (orthogonal) from the shape. Am I understanding something wrong? – Bananach Jun 23 '17 at 18:01
  • @Bananach Yes, that was my intuition too. I think your missing one parameter, that for me it would be orientation, or reflection. However, apparently I found an example. I encourage you to check it, since it contradicts an already posted answer. – Sergio Enrique Yarza Acuña Jun 23 '17 at 18:07
  • Right. Now that I read both answers I saw that the other answer did exactly what I thought. I'll ask about your answer there – Bananach Jun 23 '17 at 19:26

2 Answers2

3

I am not sure what happens with your definition, but here is an answer when we modify (2) to

(2') $A$ has Hausdorff dimension $\le 1$ (e.g. $A$ is a countable union of smooth curves).

Such sets $A$ have empty interior but the converse is false.

I will use the notation $S_n'$ for the collection of corresponding $n$-universal subsets of $R^2$ (by the way, you should not use the name "universal cover" for this, it is already taken).

Here is a proof that $S_n'=\emptyset$ for $n\ge 5$.

Consider the $n$-fold product $A^n$ of $A\subset R^2$ satisfying (2'); $A^n\subset R^{2n}$. The group $G$ of similarities of the Euclidean plane is a Lie group of dimension $4$. The group $G$ acts smoothly on $R^2$ and we obtain the corresponding diagonal action of $G$ on $R^{2n}$: $g(z_1,...,z_n)=(gz_1,...,gz_n)$, where $z_k\in R^2$ for each $k$. In other words, we have the smooth map $\mu: G\times R^{2n}\to R^{2n}$, $$ \mu(g, z_1,...,z_n)= (gz_1,...,gz_n). $$ The statement that $A$ satisfies your universality condition is equivalent to the property that $\mu(G\times A^n)=R^{2n}$. However, $\mu$ is locally Lipschitz (since it is smooth) $G$ has Hausdorff dimension $4$, $A^n$ has Hausdorff dimension $\le n$, hence, $\mu(G\times A^n)$ has Hausdorff dimension $\le 4+n$. For $n\ge 5$, $4+n< 2n$ and $2n$ is the Hausdorff dimension of $R^{2n}$. (For $n=4$ we have the equality.) Hence, $\mu(G\times A^n)$ cannot equal $R^{2n}$. Thus, $A$ cannot belong to $S'_n$.

The same argument works if (2') is replaced by

(2'') $A$ has Hausdorff dimension $<2$.

I let $S_n''$ denote the corresponding collection of universal subsets of $R^2$.

Then the result is that $$ \bigcap_{n\ge 1} S_n'' =\emptyset. $$

I do not know what happens for your original definition of $S_n$. One can attempt to replace Hausdorff dimension with topological dimension. Then $int(A)=\emptyset$ means that $dim(A)\le 1$, hence $dim(A^n)\le n$. However, there are examples of smooth maps which raise topological dimension. For instance, one can take a topological curve $P\subset {\mathbb R}^3$ (the graph of a Peano curve), such that for the action $\mu$ of ${\mathbb R}$ on ${\mathbb R}^3$ by translations along the $x$-axis, $\mu({\mathbb R}\times P)$ is 3-dimensional, equal to the product of the unit square and the real line.

Edit. The fact that $$ \bigcap_{n\in {\mathbb N}} S_n \ne \emptyset $$ is proven in Theorem 1.12 in

C. G. Wastun, Universal covers of finite sets. J. Geom. 32 (1988), no. 1-2, 192–201.

In fact, he proves even more: There exists a closed subset $A$ with empty interior in $E^2$ such that for each finite subset $F\subset E^2$ there exists a horizontal translation $T$ of $E^2$ such that $T(F)\subset A$. His sets $A$ are products of certain closed, perfect, totally disconnected subsets of the x-axis with the y-axis.

Moishe Kohan
  • 97,719
  • I like your answer a lot, thanks. A few comments are in order. – Sergio Enrique Yarza Acuña Jun 18 '17 at 19:25
  • I like your answer a lot, thanks. A few comments are in order: First, the definition is based in a work by Wastun in the 80's. He defined a universal cover of $n$ points as a set $A\subset\mathbb{R}^2$ which is closed, doesn't contain arbitrarily large discs, and for every set $Q\subset\mathbb{R}^2$ containing exactly $n$ points, there is an isometry $f$ such that $Q\subset f(A)$. So, the above definition is just a similarity version of the original and the name "universal cover" is not my invention. – Sergio Enrique Yarza Acuña Jun 18 '17 at 19:44
  • Second, the second property in the definition, has exactly the same purpose as the second property in the definition in the past comment: if the interior of $A$ is not empty, then you can cover any finite set just by making the ball inside $A$ sufficiently large. – Sergio Enrique Yarza Acuña Jun 18 '17 at 19:48
  • @SergioEnriqueYarzaAcuña: Yes, that (in your last comment) was quite clear. As for the first comment, the blame then belongs to Wastun. – Moishe Kohan Jun 18 '17 at 19:51
0

I think I found an example, but I'm curious because I think it has Hausdorf dimension <2.

Let $C\subset[0,1]$ as the Cantor set obtained by removing open intervals formed by numbers whose base 16 expansion does not have the "digit" 15. Ends of this intervals, like $0.(15)000\dots=0.(14)(15)(15)(15)\dots$ stay on $C$. Now define $B=\bigcup_{i=0}^4(i+C)$, so that $B\times B\subset[0,5]\times[0,5]$. $B\times B$ is closed and has empty interior.

I'll show that $B\times B$ is in $S_5$: Let $A=\{z_1,z_2,z_3,z_4,z_5\}\subset\mathbb{R}^2$. Let $Q$ be a similarity transformation such that $Q(A)\subset[0,5]\times[0,5]$. Now I show that there is a translation $L$ such that $L(Q(A))\subset B\times B$.

Let $p_1(x,y)=x$ and $x_j=p_1(Qz_j)$ and assume, without loss of generality, that $x_1<x_2<x_3<x<4<x_5$. Now, define $d_i=x_{i+1}-x_i$ for $i=1,\dots,4$. I show now that we can realize this four distances in $B$:

Let $d_i-[d_i]=0.d_{i1}d_{i2}\dots$ in base 16, where $[\cdot]$ stands for the floor function. To show that we can realize the four distances in $B$, we only need to prove that we can built a number $p=0.p_1p_2\dots$ (in base 16) so that for every $j$ none of the numbers

$p_j$,

$p_j+d_{1j},p_j+d_{1j}+1$,

$p_j+d_{1j}+d_{2j},p_j+d_{1j}+d_{2j}+1,p_j+d_{1j}+d_{2j}+2,$

$\dots$,

$p_j+d_{1j}+\dots+d_{4j},p_j+d_{1j}+\dots+d_{4j}+1,\dots,p_j+d_{1j}+\dots+d_{4j}+4$,

is congruent with 15 mod 16. Noticing that this list of numbers has at most 15 distinct numbers, we can choose $p_j$ so that none of them is congruent with 15 mod 16. In this way, we can realize the four distances in $B$, so that we can translate $Q(A)$ so that all the projections into the $x$-axis of its elements lie in $B$. Analogously, we can do the same for the projections to the $y$-axis, so that there is a translation $L$ such that $L(Q(A))\subset B\times B$. So, $B\times B$ is in $S_5$.

It is not so difficult to see that this process can be generalized for $n\leq6$, so that $S_n\neq\emptyset$ $\forall n\in\mathbb{N}$.

However, $B\times B$ has Hausdorff dimension <1, so that $S'_5\neq\emptyset$ and therefore $S''_5\neq\emptyset$. Please chech this, because I can't find the mistake on the previous proof nor in mine.

  • How do you know the Hausdorff dimension of $ C\times C $ is smaller than $1$? Can you expand on how you actually chose the translation given a $ p $ as in your answer? – Bananach Jun 23 '17 at 19:42
  • The Hausdorff dimension of $C$ es $log(15)/log(16), because $C$ is made of 15 smaller copies of itself and the scaling factor from $C$ to this copies is $1/16$. Since the Hausdorff dimension of $C$ is less than 1, so is the Hausdorff dimension of $C\times C$. – Sergio Enrique Yarza Acuña Jun 23 '17 at 22:38
  • Hm, you definitely are more trained in this than I am, but isn't $\dim C\times C=2 dim C $? – Bananach Jun 24 '17 at 07:21
  • LOL, you're right. The Hausdorff dimension is $2$log$(15)/$log$(16)$. So, the Hausdorff dimension is <2. – Sergio Enrique Yarza Acuña Jun 24 '17 at 17:31