Let $X,Y$ be metric space and suppose that $f:X\rightarrow Y$ is a uniform embedding; i.e.: $$ \omega(d_X(x,z))\leq d_Y(f(x),f(z)) \leq \Omega(d_X(x,z)), $$ where $\omega\leq \Omega$ are both strictly increasing continuous functions mapping $[0,\infty)$ to itself and which fix $0$.

Is $f$ a quasisymmetry? I.e.: does there exist a monotone function $\eta:[0,\infty)\rightarrow [0,\infty)$ satisfying $$ \frac{d_Y(f(x),f(y))}{d_{Y}(f(x),f(z))} \leq \eta\left(\frac{d_X(x,y)}{d_X(x,z)}\right) . $$

Note on Edit: I have reduced my previous question down to this more general one; since quasisymmetries preserve the doubling property.

  • $\begingroup$ If it were clear, the proof would also most likely provide a clear estimate. But it's just false. Just playing zigzag, you have a surjective 1-Lipschitz map from the reals (or half-reals) onto the 1-skeleton of a regular trivalent tree, which is not doubling. $\endgroup$
    – YCor
    Mar 28 at 9:12
  • $\begingroup$ For a Lipschitz map the inverse simply doesn't exist. $\endgroup$
    – YCor
    Mar 28 at 15:54
  • $\begingroup$ But the map need not be injective, and injectivity is not a necessary condition. And even for bijective maps between doubling spaces, uniform continuity can fail... $\endgroup$
    – YCor
    Mar 28 at 16:53
  • $\begingroup$ this question has Lipschitz in the title but not in the question itself. It is also strangely stated. are you assuming that $f$ is a bijection? Please restate the question clearly. $\endgroup$ Mar 28 at 18:01
  • $\begingroup$ @YCor and VitaliKpovitch I have reduced the earlier incarnation of my question down to the current formulation: when are uniform embeddings quasi-symmetries (if their moduli are well-behaved). $\endgroup$ Mar 29 at 6:58

1 Answer 1


This is false. For example take $f: [0,\infty)\to [0,\infty)$ given by $$ f(x)=\begin{cases}0& \text{ if } x=0\\ e^{-1/x}& \text{ if } x>0 \end{cases}$$

Then $\lim_{x\to 0}\frac{f(2x)}{f(x)}=\infty$ so $f$ is not quasisymmetric. But $f$ is continuous and monotone hence gives a homeomorphism onto the image when restricted to $[0,1]$. Since $[0,1]$ is compact $f$ is uniformly continuous on it. Same for $f^{-1}$.

Let me now answer the original question (which the OP erased) which asked for a weaker conclusion which is still false. The question was the following.

Let $f: X\to Y$ be a bijection such that both $f$ and $f^{-1}$ are uniformly continuous. Suppose $X$ is doubling i.e. there exists a constant $C>$ such any ball of radius $r$ can be covered by at most $C$ balls of radius $r/2$.

Question: Does this imply that $Y$ is also doubling?

The answer is NO.

The answer is yes if $f$ is bi-Lipschitz in which case the statement is trivial but any weaker conditions on the modulus of continuity should not be sufficient.

Here is an example of a metric of the closed $2$-disk which is not doubling.

Take a 2-disk $\bar D=\{x^2+y^2\le 1, z=0\}$ in the $xy$ plane in $\mathbb R^3$ and attach 10 vertical intervals of length 2 at different points of the interior of the disk. The doubling constant of the resulting space $\tilde X_1$ on scale 1 will be about 10.

The space $\tilde X_1$ is no longer a disk but we can make it one by replacing the vertical intervals by very thin "fingers" (graphs of bump functions on tiny disks) of the same height 2 so that the the new space $X_1$ is Hausdorff close to $\tilde X_1$ and is a graph of a function from $\bar D$ to $\mathbb R$.

Now take a very small $\epsilon$, take small disk $D_\epsilon$ in $D$ if radius $\epsilon$ away from the fingers and repeat the procedure on $D_\epsilon$ on scale $\epsilon$ by attaching 100 vertical intervals of length $2\epsilon$. Replace the smaller intervals by fingers. then the doubling constant on scale $\epsilon$ will be 100. Iterate. The limit space $X_\infty$ will be a graph of a function on $\bar D$ but will not be doubling. The projection map $f: X_\infty\to \bar D$ is a homeomorphism. It's 1-Lipschitz. The inverse is uniformly continuous because the spaces involved are compact.

  • $\begingroup$ This is an extremely nice example. So in your opinion what would you expect that we need of the inverse $f^{-1}$ since your example shows that uniform continuity of $f^{-1}$ isn't enough? My intuition was based on the snowflaking $f:(X,d)\mapsto(X,d^{\alpha})$ ($\alpha\in (0,1)$) in which case $f$ has moduli $\alpha$ and $\cdot^{1/\alpha}$ . $\endgroup$ Mar 29 at 12:54
  • $\begingroup$ @Carl_Petterson As I said I think you need $f$ to be bi-Lipschitz. I see no reason why any weaker modulus of continuity would suffice. For example should be possible to construct counterexamples when $f$ is bi-Hölder but not bi-Lipschitz. $\endgroup$ Mar 29 at 13:00
  • $\begingroup$ Ah okay, so then I guess the "correct condition" is simply $f$ is quasi-symmetric and we cannot hope for much more? $\endgroup$ Mar 29 at 13:01
  • 1
    $\begingroup$ ok, yes, that would work too. $\endgroup$ Mar 29 at 13:01
  • $\begingroup$ Cool or weakly quasi-symmetric (assuming $X$ is connected); ok cool, then that also covers both examples (bi-Lipschitz and the snowflake). Thanks Vitali really nice construction btw, it really helped :) $\endgroup$ Mar 29 at 13:02

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.