Let $X$ be a compact, geodesic metric space which is Gromov hyperbolic with a constant $\delta>0$. To fix scaling, let us also assume that $X$ has diameter $1$.

To fix a definition of Gromov hyperbolicity, I will use the first definition here under ``Definitions using triangles''.

I understand that mostly one does not care about the precise value of $\delta$ because we think of this as an asymptotic notion, and that all compact metric spaces are Gromov hyperbolic for $\delta$ larger than their diameters. Nonetheless, in this question I would like to care about the size of $\delta$.

Gromov hyperbolicity is often motivated as saying that a metric space looks ``tree-like'' (at least at large scales), since trees are Gromov hyperbolic with $\delta=0$. So:

Question 1: Given $\epsilon>0$, is there $\delta>0$ such that if $X$ is $\delta$-hyperbolic, then $X$ is within Gromov-Hausdorff distance $\epsilon$ from a compact metric tree?

Here a metric tree means a geodesic metric space in which each pair of points is joined by a unique arc.

I am nervous that someone might answer the above question, if the answer is indeed yes, by an appeal to ultralimits. Although this would be fine for question 1, I would be happier with a more constructive understanding. So I also ask:

Question 2: If the answer to question 1 is yes, is there an explicit estimate for the rate of decay for $\delta$ given $\epsilon$? Equivalently, is there an upper bound, tending to zero with $\delta$, for the maximal Gromov-Hausdorff distance to a tree among all metric spaces $X$ with the above properties?

I apologize if these questions are simple or well-known. If the latter, a reference would be great.

  • 5
    $\begingroup$ Hint: Round disks in hyperbolic plane will provide counter examples to Q1. $\endgroup$ Nov 20, 2020 at 0:11

2 Answers 2


Just to remove this question from the un-answered list. First of all, if $d_{GH}(X,Y)\le \epsilon$ then there is a $(1,2\epsilon)$-quasi-isometry $X\to Y$. See for instance

Burago, D.; Burago, Yu.; Ivanov, S., A course in metric geometry, Graduate Studies in Mathematics. 33. Providence, RI: American Mathematical Society. 2001.

If a geodesic metric space $X$ is $(L,A)$-quasi-isometric to a tree then it satisfies the following bottleneck property:

For every pair of points $x, y\in X$ and every mid-point $m$ of this pair, every path in $X$ connecting $x$ and $y$ passes through the open ball $B(m, L(A/2 +D) +A)$ centered at $m$.

Here $D=D(L,A)$ is some function which I can write for you if you really want. Incidentally, the bottleneck property also characterizes geodesic metric spaces quasi-isometric to trees. For a proof, see Theorem 4.6 in

J. Manning, Geometry of pseudocharacters. Geom. Topol. 9 (2005), 1147–1185.

On the other hand, if $X_R$ is the closed ball $\bar{B}(m, R)$ of radius $R$ in the hyperbolic plane and $x, y$ are boundary points of $\bar{B}(m,R)$ at the distance $2R$ from each other (with the mid-point $m$), then either arc on the boundary circle of $\bar{B}(m,R)$ connecting the points $x, y$ avoids $B(m,R)$. Hence, assuming that $R>(A +D) +2A$, the GH-distance from $X_R$ to any tree is $>A$. Here $D=D(1,2A)$. Of course, all the spaces $X_R$ are $\delta$-hyperbolic, where $\delta$ is the hyperbolicity constant of the hyperbolic plane.

Thus, your question 1 has negative answer. However, the following might be true, I am not sure:

Conjecture. There exists a separable geodesic Gromov-hyperbolic spaces $U$ (the "universal hyperbolic space") and a function $\epsilon(\delta)$ such that every compact $\delta$-hyperbolic hyperbolic space $X$ is within Gromov-Hausdorff distance $\epsilon(\delta)$ from a subset of $U$.

See also

M. Bonk, O. Schramm, Embeddings of Gromov hyperbolic spaces. Geom. Funct. Anal. 10 (2000), no. 2, 266–306.


A straightforward estimates show that any $n$-point $\delta$-hyperbolic space lies on distance at most $n{\cdot}\delta$ from a $0$-hyperbolic space. The following statement improves this bound to $\log n\cdot \delta$. I learned it from Rostislav Matveyev, who attributed it to Étienne Ghys.

Let $d$ be a $\delta$-hyperbolic metric on an $n$-points set $F$. Then there is a $0$-hyperbolic metric $d'$ on $F$ such that $$d \leqslant d' \leqslant d+\mathrm{const}\cdot\log n\cdot \delta.$$

To prove it, you choose a base point and define new Gromov's product $(\ |\ )'$ using the old one $(\ |\ )$ $$(x|y)'\mathrel{:=}\max \{\,\min \{\,(x_0|x_1),\dots,(x_{m-1}|x_m)\,\}\,\},$$ where the maximum is taken for all chains of points $x=x_0,\dots, x_m=y$. Plus you need to apply the following lemma:

For any sequence of points $x_0,\dots,x_m$ in a $\delta$-hyperbolic space, we have $$(x_0|x_m)\geqslant \min \{\,(x_0|x_1),\dots,(x_{m-1}|x_m)\,\}-\mathrm{const}\cdot\log m\cdot \delta.$$

The proof is recursive application of the inequality for Gromov's product for triples $$(x_i|x_k)\geqslant \min\{\,(x_{i}|x_j),(x_{j}|x_k)\,\}- \delta.$$ You start with $(x_0|x_m)$ and choose roughly the middle index each time. This way you get $\lceil\log_2 n\rceil$ iterations; each iteration gives you extra $\delta$.

Postscript. Actually, it is 6.1.B in Gromov's "Hyperbolic groups".


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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