I would like to ask your thoughts on the following problem:

“Characterise the metric spaces in which every function is uniformly continuous”

The first thing I observed is that every subset of such metric spaces (let us call them X for the sake of semplicity) must be open for otherwise there would be some metric spaces Y and some function $f:X \rightarrow Y$ which may not be continuous (hence, also not uniformly continuous). Then, I conjectured that X must be compact, in fact, by the Heine-Cantor theorem this is a sufficient condition; however I can not manage to find a counter-example for when X is not compact in order to conclude that it is also necessary.

Any help or hint is much appreciated!


1 Answer 1


As you rightly point out, since every function from $X$ to any other metric space $Y$ is continuous, every subset of $X$ is an open set.

(By the way, are you sure that the question doesn't mean "every function from $X$ to $X$ is continuous"?)

In the study of general topology, when every subset of $X$ is an open set, we say that $X$ has the discrete topology.

Let's now apply the condition about uniform continuity. My first instinct is that the infimum of $d(x,y)$ for $x\neq y$ has a role to play. And indeed it does. Let's call the infimum $r$.

If $r>0$, then every function is of course uniformly continuous. The image of each $s$-ball is a single point, for $s<r$.

What if $r=0$?

Then, with the help of the axiom of choice, choose sequences $x_i,x'_i$ such that $d(x_i,x'_i)$ decreases strictly to $0$, and such that all these points are distinct, i.e. no point is used more than once here.

Now, you can construct a function from $X$ to $X$, which sends $x_i$ to $x_1$, and $x'_i$ to $x'_1$. It wouldn't matter where the function sends any other point. The function is not uniformly continuous.


In conclusion, for metric spaces $X$,

"every function on $X$ is uniformly continuous" is equivalent to

"there is some $c>0$ such that all pairs of points have distance at least $c$".


EDIT: I forgot to explain a subtle technical point.

We know that a subset containing only one point $p$ is open. This means that for each $p$, there is some radius $s_p$ such that the open ball $B(p,s_p)$ contains only $p$.

This point is relevant to the following question. When we were dealing with the case $r=0$, how can we be sure that we can choose $x_i,x'_i$, so that we do not use any point more than once? (We want to ensure this, because we want all the $x_i$ to be mapped to $x_1$, and all the $x'_i$ to be mapped to a point $x'_1$ which is different from $x_1$. If $x'_6=x_4$, for example, then $x'_6$ will map to both $x'_1$ and $x_1$, which we don't want.)

We can do this, as shown by induction. Suppose you have already chosen $x_1,\ldots,x_n$ and $x'_1,\ldots,x'_n$. We want to choose $x_{n+1}$ and $x'_{n+1}$ so that neither equals any of those. Just draw an open ball around each of those such that each open ball contains only that one point.

Take the minimum $m$ of those radii. And declare that $x_{n+1}$ and $x'_{n+1}$ will have distance less than $m$. Such a choice is always possible, because the infimum of $d(x,y)$ for $x\neq y$ is assumed to be $0$.

It follows that $x_{n+1}$ and $x'_{n+1}$ do not repeat the previous points, because the open ball of radius $m$ around $x_{n+1}$ contains $x'_{n+1}$, and vice versa. Yet an open ball of radius $m$ around each of the previous points contains only one point.

  • 1
    $\begingroup$ Some call this condition for a metric space "uniformly discrete." $\endgroup$ Feb 25 at 16:53
  • $\begingroup$ Man, thank you very much! It is one of those answers that really gives you insight! $\endgroup$ Feb 25 at 17:20
  • $\begingroup$ No problem. I've included an addendum, a technical part of the proof which I didn't explain, if you're interested $\endgroup$ Feb 25 at 17:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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