5

Let $M$ be a metric space and let $S\subset M$. We define the $\varepsilon$-neighbouhood of $S$ as the set $U_\varepsilon = \{x\in M\ :\ d(x, S) < \varepsilon\}$, where the distance between a point and a set is defined as $d(x, S) = \inf\{d(x, s)\ :\ s\in S\}$.

If $B_r(x)\subset M$ is the open ball of radius $r$ centered around $x$, we have $U_\varepsilon(B_r(x))\subset B_{r+\varepsilon}(x)$. Is there a name or anything to be read on metric spaces that have the property $B_{r+\varepsilon}(x) = U_\varepsilon(B_r(x))\ \forall r > 0, \varepsilon > 0, x\in M$?

Edit: Not all spaces satisfy this condition. Consider the two point subset of the line $\{0,1\}$. We have

$$U_{\frac{1}{2}}(B_{1}(0)) = U_{\frac{1}{2}}(\{0\}) = \{0\} \neq B_{\frac{3}{2}}(0) = \{0,1\}$$

Motivation: I'm reading a Mikhael Gromov's paper on Groups of Polynomial Growth (http://www.numdam.org/item/?id=PMIHES_1981__53__53_0). On section 3 he defined a space with this property, which he names connectivity. I'm interested in learning more about this kind of space.

Carlyle
  • 2,807
  • 2
  • 22
Eric Vaz
  • 397
  • 1
    Your notation could be improved by calling the set $S_\varepsilon$ (rather than $U_{\varepsilon}$ ($S_\epsilon$ is an $\varepsilon$-neighborhood of $S$). Do you have any examples of spaces which do not satisfy this criteria? spaces which do? In what context does this come up? – Xander Henderson Dec 23 '23 at 23:09
  • Not that it matters much, but the notion introduced by the author requires $r$ to be an integer – Carlyle Dec 24 '23 at 09:58

1 Answers1

3

This is closely related to a class of metric spaces which I created, and termed "interpolation spaces" (unfortunately the name is taken by something in functional analysis already) but you won't find anything about this online yet, so I'll give the definition briefly:

A metric space $(M,d)$ is called an interpolation space if given any $x,y\in M$ and any $r\in\mathbb{R}$ such that $r> -d(x,y)$, we can find a $z\in M$ such that \begin{align} d(y,z) &= |r| \\ d(x,z) &= d(x,y)+ r \end{align}

Those last two conditions require that the triangle inequality is an equality for the 3 points, so these are spaces where you can find a point between and beyond two given points at a given distance, so that all $3$ points are "colinear".

These are related to your spaces (connectivity is a bad name in my opinion, seeing as these spaces do not have to be connected, but I will stick with it for now)

Every interpolation space has the connectivity property

The proof is rather straightforward, we need to show the inclusion (we can assume $\varepsilon < r$)

$$B_{r+\varepsilon}\subseteq U_{\varepsilon}(B_r(x)) $$

Thus let $$y\in B_{r+\varepsilon}(x)$$ but $$y\notin B_{r}(x)$$ then we can define $$ r_0:= - \frac{d(y,B_r(x))+\varepsilon}{2}> -\varepsilon$$ (This last inequality is not trivial, you should prove it from the interpolation property)

And then by the interpolation property there is some $z\in M$ such that

\begin{align} d(y,z) &= |r_0| < \varepsilon \\ d(x,z) &= d(x,y) - \frac{d(y,B_r(x))+\varepsilon}{2} < r \end{align}

(This last inequality follows since $d(x,y) = r + d(y,B_r(x))$, this too is a non-trivial consequence of the interpolation property)

Then $z\in B_r(x) $ and $y\in B_{\varepsilon}(z) $ and hence by the triangle inequality,

$$ y\in B_{r+\varepsilon}(x) $$

Which concludes the proof.

However, the two notions are not equivalent, the rationals, $\mathbb{Q}$ have connectivity, but they do not have the interpolation property. There are other nice properties possessed by interpolation spaces that these spaces do not possess as well, for example a closed ball is compact if and only if the boundary of every closed ball contained in the original closed ball, is compact.

Carlyle
  • 2,807
  • 2
  • 22
  • This is very interesting! It seems to me like there is a connection between interpolation spaces and length spaces, which is what Gromov seems to be using on the second page of section 7. As far as you are aware are there interpolation spaces that are not length spaces? – Eric Vaz Dec 24 '23 at 18:15
  • @EricVaz This is a good question, and indeed there is a connection, most natural examples turn out to be length spaces, but there are quite pathological examples which I suspect (but have not proven) are not locally path connected (which I believe is necessary for a length space). – Carlyle Dec 24 '23 at 18:23