Suppose I have a sequence of functions $f_n: X \to Y$ between metric spaces that converges uniformly to $f$. Then if I change the metric on $Y$ to a Lipschitz equivalent metric, the sequence of functions will still converge uniformly to $f$.

However, I am told that if I change the metric on $Y$ in such a way that does not disturb the induced topology, then $f_n$ may no longer converge uniformly. My question is: is there a simple example of a sequence of functions from $X$ to two homeomorphic metric spaces $Y$, $Y'$ converging uniformly in one but not the other?

EDIT: question modified slightly in response to comments.

  • 3
    $\begingroup$ The different metric must be on $Y$, not on $X$, because the metric (not even the topology, actually) of $X$ never really comes into play. You could pick a sequence of functions $f_n:A\to (Y,d)$, where $A$ is just a set, and define uniform convergence in the same way. $\endgroup$ – user228113 Jan 13 '17 at 12:05
  • $\begingroup$ No, the domain $X$ need not even be a topological space to speak of uniform convergence of a sequence (filter, net) of functions $X\to Y$. For that, only $$\sup \{ d(f_n(x), f(x)) : x \in X\}$$ is relevant, and that only involves the metric (or more generally the uniform structure) on the codomain. $\endgroup$ – Daniel Fischer Jan 13 '17 at 12:15
  • $\begingroup$ Thanks for the reply --- I just realised I was being silly (indeed $X$ can be any set) and deleted my comment. $\endgroup$ – gj255 Jan 13 '17 at 12:16
  • 1
    $\begingroup$ The topology (more precisely, the metric, or more broadly, the uniformity) on $X$ will matter when you want to talk about uniform continuity of a function. $\endgroup$ – tomasz Jan 13 '17 at 12:42

$Y$ must not be compact for this to be possible (because compact spaces have unique uniform structure).

You can simply construct an artificial example. For instance, let $X={\bf N}$ and $Y={\bf N}\times [0,1]$, while $Y'$ is $Y$ with metric on $\{n\}\times [0,1]$ scaled by $1/n$ (so that the diameter is $1/n$, you can put the distance between components in any way you want, for example put $2$ whenever two points are in different components). Clearly, $Y'$ is homeomorphic to $Y$ (even by a locally bi-Lipschitz function), but if you take sequence of functions $(f_m)$ with $$f_m(n)=\begin{cases}(n,1) &\textrm{if }m>n\\ (n,0) &\textrm{otherwise}\end{cases}$$ you will have $f_m$ converge uniformly as sequence of functions into $Y'$, but not so into $Y$. You can replace $[0,1]$ here with any nontrivial metric space.

For a different example, take for $Y$ the interval $(0,1]$ with the standard metric, and for $Y'$ the same interval with the metric scaled by a singular function, e.g. for $y_1<y_2$ put $d(y_1,y_2)=y_1^{-1}-y_2^{-1}$, and put $$f_m(n)=\begin{cases}\frac{1}{n} &\textrm{if }m>n\\ \frac{1}{n+1} &\textrm{otherwise}\end{cases}$$ This sequence will be uniformly convergent in the standard metric, but not in the scaled metric.

  • $\begingroup$ Thanks a lot for the answer. For the first function you defined, aren't the $f_m$ converging to $f: n \mapsto (n,1)$? $\endgroup$ – gj255 Jan 13 '17 at 13:33
  • $\begingroup$ @gj255: yes, that's right, my mistake. $\endgroup$ – tomasz Jan 13 '17 at 13:34

Consider $f_n(x)=x+\frac1n$, from $X=(0,1)$ to $Y=\Bbb R$, with the usual metrics. This converges uniformly to $f(x)=x$.

Now, change the metric of the codomain to $d'(x,y)=\frac1x-\frac1y$. That is, let $Y'$ be $(\Bbb R,d')$. Then $f_n$, as a function from $X$ to $Y'$ no longer converges uniformly to $f$.


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.