We introduce a type of surgery on metric spaces. This surgery, in some sense, seeks to replace a subspace S of a metric space X with another metric space T via a function f : S \to T. When T is a discrete space, this amounts to collapsing the subspace according to the function. This surgery results in a new metric space we denote \widehat{X}_f and there is a natural function F : X \to \widehat{X}_f induced from f. Our primary interest is investigating if properties of the original function f are inherited by the induced function F. We show that if f is a pseudo-isometry then so is F. However, for a quasi-isometry, a very natural generalization of a pseudo-isometry that is prevalent in geometric group theory, such a result does not hold.
Suppose (X,d_X) and (T,d_T) are metric spaces and consider a subset S \subset X as a metric space with metric induced from X. If f \colon\thinspace S -.6 T is a pseudo-isometry, then the natural map F \colon\thinspace X -.6 \widehat X_f is a pseudo-isometry as well.
Let (S,d_S) and (T,d_T) be metric spaces. A map f \colon\thinspace S -.6 T is a pseudo-isometry if there exist contants K \geq 1 and C \geq 0 such that the following hold.
For all x_0,x_1 \in S, we have:
For all y \in T, there is an x \in S with d_T(f(x),y) \leq C.
If we allow the upper bound to also have an additive constant, i.e.,
then the map is called a quasi-isometry.
An admissible sequence is a sequence of pairs of the form:
where:
x_0,y_{k+1} \in X,
x_i,y_i \in S for i = 1,\ldots,k,
u_i,v_i \in T for i = 1,\ldots,k, and
u_i = f(y_i) and v_i = f(x_i) for i = 1,\ldots,k.
We allow for the possibility that x_i = y_{i+1} or u_i = v_i for each i = 0,\ldots,k. Moreover, we allow for the possibility that the pair (x_0,y_1) is omitted. In this case, the only restriction on u_1 is that it lies in T. Likewise, we allow for the possibility that the pair (x_k,y_{k+1}) is omitted. In this case, the only restriction on v_k is that it lies in T. We say the sequence is from x_0 to y_{k+1}, modifying to use u_1 or v_k accordingly if the pair (x_0,y_1) or (x_k,y_{k+1}) respectively is omitted. A schematic for an admissible sequence appears in Figure ?.
The length of an admissible sequence \gamma as defined in (2.3) is:
Let \gamma be an admissible sequence from x to y, where x,y \in X. Then
Let the admissible sequence \gamma be given by:
where x_0 = x and y_{k+1} = y.
By the definition of an admissible sequence, we have f(x_i) = v_i and f(y_i) = u_i for i = 1,\ldots,k. The assumption that f is a pseudo-isomtery implies
Combining the triangle inequality with the first inequality gives (Proof), (Proof) and (Proof). Regrouping gives (Proof) and the second of these inequalities is used in eqrefeq:5. As K \geq 1, further rearranging gives (Proof). Finally, (Proof) follows from the definition of the length of \gamma.
This completes the proof of the lemma.
Note that in the proof above it is necessary that f is a pseudo-isometry and not merely a quasi-isometry. In passing from (Proof) to (Proof), the d_X summation has no additive term, effectively allowing us to bound the lengths with a multiplicative constant. Had f been just a quasi-isometry this summation would induce k-1 additive constants. The number of such constants reflects the number of steps in the admissible sequence which is not bounded by the distance. This makes it impossible to bound the distance between x and y in terms of the length of an admissible sequence between them.
If x,y \in X and F(x) = F(y), then d_X(x,y) \leq 3KC.
Fix points x,y \in X and suppose that F(x) = F(y) = \hat{x}. Let x' = j(x) and y' = j(y). By the construction of \widehat X_f, we have that for any two points in q^{-1}(\hat{x}) the pseudo-distance is equal to 0. Hence p_{X'}(x',y') = 0. Therefore, for any \epsilon > 0, there must be points x_0,y_0 \in X \cup T with j(x_0) = x', j(y_0) = y', and an admissible sequence \gamma from x_0 to y_0 of length less than \epsilon.
By the definition of j \colon\thinspace X \cup T -.6 X' we can assume that x_0 and y_0 lie in X. Indeed, if x_0 \in T then we must have x_0 = f(s_0) for some s_0 \in S as j is injective on T - f(S). We could then prepend the sequence \gamma by the ordered pair (s_0,s_0) to get a new sequence with the same properties that starts in X. Likewise if y_0 \in T.
Since j(x) = j(x_0), we must have that f(x) = f(x_0) and hence
since f is a pseudo-isometry. Likewise we have that d_X(y,y_0) \leq KC. Using Lemma 2.5, we have that d_X(x_0,y_0) \leq K^2\epsilon + KC. Combining these with the triangle inequality, we find:
As this holds for all \epsilon > 0, we have d_X(x,y) \leq 3KC as claimed.
Let X, S, T be as in the statement of the theorem and f \colon\thinspace S -.6 T a pseudo-isometry with constants K \geq 1 and C > 0 as in Definition 2.1. Let X' and \widehat X_f be as defined in Section ? and F \colon\thinspace X -.6 \widehat X_f the map induced by the composition:
First we show F is coarsely surjective. By construction, for any \hat{y} \in \widehat X_f - F(X) there is some t \in T such that \hat{y} = q(j(t)). As f \colon\thinspace S -.6 T is coarsely surjective, there is an s \in S where d_T(f(s),t) \leq C. The admissible sequence (s,s), (f(s),t) from s to t has length at most C. Therefore p_{X'}(j(s),j(t)) \leq C and hence
as well.
Next we demonstrate an upper bound on d_{\widehat X_f}(F(x),F(y)). Given points x,y \in X, for the admissible sequence \gamma : (x,y) we find that p_{X'}(j(x),j(y)) \leq \ell(\gamma) = d_X(x,y). Therefore
Finally, we demonstrate a lower bound on d_{\widehat X_f}(F(x),F(y)). Let \epsilon > 0. Given points x,y \in X, we fix points x_0, y_0 \in X where F(x) = F(x_0), F(y) = F(y_0), and for which there exists an admissible sequence \gamma from x_0 to y_0 of length less than d_{\widehat X_f}(F(x),F(y)) + \epsilon. We note that such points and admissible sequence exist by an argument similar to the one presented in Lemma 2.7.
By Lemmas 2.5 and 2.7 we now have that:
As this holds for all \epsilon > 0, rearranging we find that:
Let X be the regular tree of degree 4 where every edge is isometric to the unit interval [0,1]. We consider the subset S \subset X consisting of every other horizontal edge, including its incident vertices, as indicated in Figure 4.1. Specifically, we can identify X as the Cayley graph of the free group of rank 2, F_2, corresponding to a basis \{g_1,g_2\} (see Chapter 2 of [Clay] for details). Then there are two orbits of edges corresponding to the set of horizontal edges and the set of vertical edges respectively in Figure 4.1. Let e_1 be the edge whose originating vertex is the identity and whose terminal vertex is g_1. Then the set S corresponds to the set of edges of the form we_1 where the terminal syllable of w is an even power of g_1. In other words, w \in F_2 is of the form:
where w' \in F_2 is either trivial or ends in g_2^{\pm 1} and p \in \mathbb{Z}.
Let T \subset X be the set of midpoints of the edges in S, considered as a metric space using the metric from X. The map f \colon\thinspace S -.6 T that sends each edge in S to its midpoint is a pseudo-isometry where K = 2 and C = 1. As shown in Figure 4.1, the surgered space \widehat X_f is the regular tree of degree 6.
Let X be a regular tree of degree n where every edge is isometric to the unit interval [0,1]. As we show, a partial fold of two edges of X is a pseudo-isometry. This produces a new tree whose vertices have degree either three or n-1 and applying this map iteratively produces a tree of degree three. Because each stage of the map is a pseudo-isometry, we get a pseudo-isometry from any regular tree to a tree of degree three. Moreover, it is quite clear that regularity of X is not required in this construction.
At each vertex choose two edges u and v. Since each edge is isometric to [0,1], assume the vertex shared by u and v corresponds to 0. Let u_0 and v_0 denote the half-edges of u and v, respectively, that are identified with [0,\frac{1}{2}]. The union of each pair of half-edges is a half-corner. Let S \subset X be the set of all the half-corners and let T \subset X be the set of all the u_0. i The map f \colon\thinspace S -.6 T that sends each half-corner in S to its corresponding u_0 is a pseudo-isometry where K = 3 and C = 1. As suggested in Figure ?, the surgered space \widehat X_f is a tree where every vertex is either degree three or degree n-1. By iterating this process on vertices of degree not equal to three, we arrive at a regular tree of degree three, see Figure ?.
Let X be the closed positive ray in \mathbb{R}, i.e., X = [0,\infty). Define two sequences (a_i), (b_i) \subset X as follows starting with i=0:
These sequences are related by the equations a_i = b_{i-1} + \left(\sfrac{1}{2}\right)^i and b_i = a_i + 1. We consider the subset S \subset X given by:
See Figure 4.3.
Let T \subset X be the set of all a_i, considered as a metric space using the metric from X. The map f \colon\thinspace S -.6 T that sends each interval [a_i,b_i] to a_i is a (continuous) quasi-isometry with K = 1, C = 2, but it is not a pseudo-isometry for any choice of constants K and C. Indeed, d_S(b_{i-1},a_i) = \left(\sfrac{1}{2}\right)^i -.6 0 as i -.6 \infty while d_T(f(b_{i-1}),f(a_i)) = d_T(a_{i-1},a_i) \geq 1 for all i. The map collapses all but intervals of the form [b_n,a_{n+1}], each having length (\frac{1}{2})^{n+1}. Since \sum_0^\infty (\frac{1}{2})^{n+1} = 1, the surgered space \widehat X_f is isometric to the half-open interval [0,1) \subset \mathbb{R}. In particular, X is not pseudo-isometric (nor quasi-isometric) to \widehat X_f.
The failure of this example is directly tied to the failure for Lemma 2.5 in this setting. There are admissible sequences with bounded length connecting points arbitrarily far apart in X. See Remark 2.6.
The text on this page is from arXiv:2507.23666v1 by Matt Clay and Josh Thompson, under the licence CC0 1.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2507/2507.23666.jsonl.