Ultrametric spaces and clouds

I. N. Mikhailov

REVIEW 1 major objections 5 minor 15 references

Ultrametric spaces and clouds

T0 review · 1 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proves that ultrametrization is 1-Lipschitz for every metric space, and uses this to show that unbounded clouds cannot mix ultrametric and dotted connected spaces.

desk verdict A small, honest generalization of the Carlsson–Memoli 1-Lipschitz bound to unbounded metric spaces, with a few new structural results; one easily fixable gap in Theorem 3.4(ii). read the letter →

arxiv 2501.19346 v1 pith:6Y73RLJ3 submitted 2025-01-31 math.MG

I. N. Mikhailov This is my paper · ORCID

classification math.MG MSC 54E3553C23

keywords ultrametricspaceGromov–Hausdorffdistancecloudsdottedconnectedmetric1-Lipschitzmapultrametrizationunboundedspaces

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper shows that the construction sending a metric space to the ultrametric space obtained by closing distances under finite chains is 1-Lipschitz for all metric spaces, not only bounded ones: the Gromov–Hausdorff distance between two ultrametrizations never exceeds the distance between the original spaces. This makes the ultrametrization map available as a lower-bound tool throughout the whole class of metric spaces, including unbounded ones. The main structural consequence is a separation law for clouds, the equivalence classes of metric spaces lying at finite Gromov–Hausdorff distance from one another: an unbounded cloud cannot contain an ultrametric space and a dotted connected space simultaneously. A reader interested in the geometry of non-compact metric spaces gets a simple criterion that separates two natural families inside each cloud.

What carries the argument

The load-bearing object is the ultrametrization map $U$: for a metric space $X$, define $u_X(x,x')$ as the infimum over finite chains from $x$ to $x'$ of the maximum step length, then quotient $X$ by the points at $u_X$-distance zero; this is the single-linkage chain-closure construction. A dotted connected space is one in which any two points can be joined by finite chains with arbitrarily small maximum step length; every path-connected space is an example. The proof mechanism is the correspondence formula $2 d_{GH}(X,Y) = \inf\{\operatorname{dis} R : R \in \mathcal{R}(X,Y)\}$, which turns a low-distortion multivalued matching between $X$ and $Y$ into one between $(X,u_X)$ and $(Y,u_Y)$ with no larger distortion. The cloud-level results are carried by two further facts: a dotted connected space $A$ satisfies $U(A)=\Delta_1$, so any space in the same cloud has bounded $U$-image, and the standard isometric embedding of a metric space into bounded continuous functions allows construction of path-connected spaces at controlled Gromov–Hausdorff distance from any space with bounded $U$. Finally, the identity $U(X\times_\rho Y)=U(X)\times_{\ell^\infty}U(Y)$ for fair metrics bounded below by the $\ell^\infty$ metric makes the product map $\Psi$ behave as an isometric embedding of bounded ultrametric spaces, with $U$ as a left inverse.

What would settle it

Look for a single cloud of unbounded metric spaces containing both an ultrametric space and a dotted connected space; the paper's Corollary 3.3(ii)–(iii) asserts that no such cloud exists. Failing that, testing Claim 2.1 on a concrete unbounded pair, for example a geometric progression inside $\mathbb{R}$ and a path-connected unbounded space, would locate the exact point where the extension could break.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is Theorem 3.1: for arbitrary metric spaces $X$ and $Y$, $$d_{GH}(U(X),U(Y)) \leq d_{GH}(X,Y),$$ where $U(X)$ is the quotient of $X$ by the pseudometric $u_X(x,x') = \inf\{\max_{0\le i\le n-1} d_X(x_i,x_{i+1}) : x=x_0,\ldots,x_n=x'\}$. The proof takes a correspondence between $X$ and $Y$ of distortion $c$ and shows the same correspondence has distortion at most $c$ between the chain-closed pseudometrics, so the bounded-space stability theorem carries over unchanged. With this inequality, the paper derives Corollary 3.3: a cloud of unbounded metric spaces cannot contain both an ultrametric space and a dotted connected space, and dually an unbounded ultrametric cloud contains no dotted connected space. It also proves that bounded ultrametric spaces form a closed subclass of the bounded cloud, and that for a dotted connected space $A$ the map $\Psi(X)=X\times A$ preserves Gromov–Hausdorff distance on that subclass, with $U$ inverting it.

Load-bearing premise

The load-bearing premise is Claim 2.1, the formula $2 d_{GH}(X,Y) = \inf\{\operatorname{dis} R : R \in \mathcal{R}(X,Y)\}$, which the paper applies to arbitrary metric spaces while citing [5] and [15] rather than proving it beyond the bounded case; if an unbounded pair violated this formula, the 1-Lipschitz inequality and the cloud separation theorem would lose their basis.

Editorial extensions

If this is right

The inequality $d_{GH}(U(X),U(Y)) \leq d_{GH}(X,Y)$ holds for arbitrary metric spaces, so the ultrametrization map is a universal lower-bound tool for Gromov–Hausdorff distances even when both spaces are unbounded.
Every cloud is mapped by $U$ into a single cloud, and any cloud that contains an ultrametric space is invariant under $U$.
A cloud of unbounded metric spaces that contains a dotted connected space contains no ultrametric space; an unbounded ultrametric cloud contains no dotted connected space.
The bounded ultrametric spaces form a closed subclass of the cloud of bounded metric spaces, so a bounded space sufficiently close to an ultrametric space is itself ultrametric.
For a dotted connected space $A$, the map $\Psi(X)=X\times A$ preserves Gromov–Hausdorff distance on bounded ultrametric spaces, and $U$ inverts it.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper leaves implicit that the diameter of $U(X)$ is a 1-Lipschitz numerical invariant of every cloud, so detecting whether this diameter is finite gives a cheap way to recognize clouds that contain path-connected members.
One testable extension is to ask whether the product identity $U(X\times_\rho Y)=U(X)\times_{\ell^\infty}U(Y)$ yields exact Gromov–Hausdorff computations for unbounded spaces, where the correspondence formula is less settled.
A natural next step, not addressed in the paper, is whether the cloud obstruction persists under approximate ultrametricity, for instance for spaces whose $U$-image is a bounded perturbation of an ultrametric space.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a carefully written, honest paper. Its advertised main result, Theorem 3.1, really is a simple observation: the proof of the bounded-case 1-Lipschitz property of the ultrametrization map U goes through verbatim once you work with the correspondence formulation of the Gromov–Hausdorff distance. That is the paper's main value, and the author says so himself. It makes the stability result available for all clouds, not just the bounded ones, which matters for the recent unbounded-cloud program of Bogatyi–Tuzhilin and collaborators.

What else is actually new: Theorem 3.2(ii), the product formula U(X ×_ρ Y) = U(X) ×_ℓ∞ U(Y) for fair metrics bounded below by the ℓ∞-metric, the closedness of the ultrametric subclass in the bounded cloud, and the cloud dichotomy in Corollary 3.3: no cloud can contain both an unbounded ultrametric space and a dotted connected space. These all follow from standard techniques, but I do not see them in the cited literature, and the proofs are compact and correct.

The soft spots are minor. In Theorem 3.4(ii) the proof sets c = diam U(A) and invokes Lemma 2.1(i), which requires diam U(A) < c. When diam U(A) = 0 the strict inequality fails. The fix is trivial: any c > 0 works, since the assumption is diam U(A) < c. The conclusion still holds. There are a couple of typos (spacing, an index in Definition 2.13), nothing substantive.

The one thing that might raise an eyebrow is that Claim 2.1, the equality 2 dGH(X,Y) = inf dis R for arbitrary (not necessarily bounded) metric spaces, is only cited to [5] and [15], not proven. I checked the standard argument: for finite distortion c, the adjunction metric on X ⊔ Y realizes the pair with Hausdorff distance ≤ c/2, no boundedness needed. Tuzhilin's lectures cover exactly this. So the citation is fine, and the reader's worry about this point does not land.

Who gets value: metric geometers working with the Gromov–Hausdorff class and clouds, and anyone using ultrametric lower bounds in clustering. It is a small, solid contribution, not a landmark. For peer review: yes, it deserves refereeing; the referee's job is a small fix to Theorem 3.4(ii) and a check of typos.

Referee Report

1 major / 5 minor

Summary. The paper studies the Carlsson-Memoli ultrametrization map U, which sends a metric space to the quotient ultrametric space obtained from the maximal chain-step pseudo-ultrametric. It extends the known 1-Lipschitz stability of U from bounded metric spaces to arbitrary metric spaces (Theorem 3.1), proves a product formula for fair metrics bounded below by the ℓ∞ metric (Theorem 3.2), shows that the class Ult of bounded ultrametric spaces is closed in the bounded cloud (Theorem 3.3(i)), and proves that multiplication by a dotted connected metric space is an isometry on Ult (Theorem 3.3(ii)). The paper then derives a cloud dichotomy: an unbounded cloud cannot contain both an ultrametric space and a dotted connected space (Corollary 3.3(ii)-(iii)). The proofs use the correspondence formula for the Gromov-Hausdorff distance, the Kuratowski embedding, and the Carlsson-Memoli construction.

Significance. If the results stand, the paper gives a clean extension of a standard stability estimate to unbounded metric spaces and a sharp structural dichotomy for clouds in the Gromov-Hausdorff class. The arguments are concise and mostly self-contained, and they are grounded in standard external benchmarks rather than in fitting or ad hoc assumptions. The main consequence, Corollary 3.3(ii)-(iii), is a crisp falsifiable statement about the geometry of unbounded clouds. The paper does not provide machine-checked proofs or reproducible code, but the traditional proofs are short and verifiable. The main caveat is a local but genuine gap in the proof of Theorem 3.4(ii), which is easily repaired and does not affect the truth of the stated results.

major comments (1)

[Theorem 3.4(ii)] The proof sets c = diam U(A) and then applies Lemma 2.1(i), whose hypothesis is diam U(A) < c. This strict inequality fails whenever diam U(A) is finite, including the case diam U(A) = 0. The result is still true: choose any c with diam U(A) < c < ∞; by Lemma 2.1(i) the space D_c(A) is path-connected, and by Lemma 2.1(ii) one has dGH(A, D_c(A)) ≤ c/2 < ∞, hence D_c(A) lies in [X]. The proof should be corrected accordingly.

minor comments (5)

[Definition 2.3] The word "infinum" should be "infimum".
[Example 2.1] The sentence "Since γ is continuous, [0,1] is compact, γ is also uniformly continuous" is grammatically awkward; it should say that γ is uniformly continuous because [0,1] is compact.
[Theorem 3.3(ii)] The proof cites Theorem 3.2 for the identities U(U) = U(A × U) and U(U′) = U(A × U′); the precise source is Corollary 3.2, which follows from Theorem 3.2(ii) and Remark 2.1. Please adjust the reference.
[Claim 2.1 / Theorem 3.1] The proof of Theorem 3.1 invokes Claim 2.1 for arbitrary, not necessarily bounded, metric spaces. This is standard and is covered by the cited reference [15], but adding one sentence making this explicit would help readers who are used to the compact-case formulation.
[Notation] In statements such as Lemma 3.1 and Theorem 3.3(ii), the notation X ∈ [∆1] conflates a metric space with its Gromov-Hausdorff class. Since Lemma 3.1 shows the map is 1-Lipschitz, this is harmless, but writing representatives and classes explicitly would remove ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and relies on external benchmarks, not on fitted inputs or load-bearing self-citation.

full rationale

The central result, Theorem 3.1, extends the Carlsson–Memoli stability theorem to unbounded metric spaces. Its proof uses only the definition of the ultrametric distance u_X, the external correspondence formula Claim 2.1 (cited to [5] and [15]), and elementary triangle-inequality arguments; it does not assume the conclusion or fit any parameter. Claim 2.1 is an independent standard result, not derived from the paper's own claims, so citing it for arbitrary metric spaces is not circular even though the proof sketch is not repeated. Theorem 3.3(ii) is proved by a chain of inequalities, dGH(U,U') >= dGH(A×U,A×U') >= dGH(U(A×U),U(A×U')) = dGH(U,U'), where the equality uses the independently proved Theorem 3.2 and Corollary 3.2; this is not circular because the target equality is obtained by combining two 1-Lipschitz inequalities rather than assumed. Corollary 3.3 follows from Theorem 3.4 and Theorem 3.2 without circularity. The author's own previous works [11] and [12] appear only in the introductory list of recent literature and are not load-bearing. The only notable defect is a boundary-case gap in Theorem 3.4(ii) when diam U(A) = 0, since Lemma 2.1(i) requires diam U(X) < c with c > 0; this is a correctness issue, not a circularity. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors, and no ansatz is smuggled in via self-citation. The paper is self-contained against external benchmarks, so the circularity score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters or invented entities. The paper introduces the term 'dotted connected' for a known metric-space property and otherwise relies on standard external theorems.

assumptions (5)

domain assumption Claim 2.1: 2 dGH(X,Y) = inf{dis R : R ∈ R(X,Y)} for all metric spaces
Used in proofs of Theorems 3.1 and 3.3; cited to [5] and [15], applied here to unbounded spaces.
domain assumption The ultrametrization construction U from [6] is well-defined and satisfies Theorem 2.2 for bounded spaces
The whole paper builds on this construction; Theorem 2.2 is quoted from [6] and [13].
standard math Kuratowski embedding (Theorem 2.3) is isometric for arbitrary metric spaces
Used in Lemma 2.1 to build path-connected approximations; cited to [10].
standard math Gromov-Hausdorff distance is a generalized pseudometric on all metric spaces
Basis for defining clouds; Theorem 2.1 cited to [5].
domain assumption Boundedness is preserved under finite GH distance: a cloud of unbounded spaces contains only unbounded spaces
Used implicitly in Corollary 3.3(ii)-(iii); follows from the definition of GH distance but is not stated as a lemma.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Ultrametric spaces and clouds." pith.science (2026). https://pith.science/paper/6Y73RLJ3

@misc{pith2026250119346,
  author       = {Pith},
  title        = {Pith review of: Ultrametric spaces and clouds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6Y73RLJ3}},
  note         = {Machine review of arXiv:2501.19346}
}

abstract

In ``Characterization, stability and convergence of hierarchical clustering methods'' by G. E. Carlsson, F. Memoli, the natural way to construct an ultrametric space from a given metric space was presented. It was shown that the corresponding map $\textbf{U}$ is $1$-Lipschitz for every pair of bounded metric spaces, with respect to the Gromov-Hausdorff distance. We make a simple observation that $\textbf{U}$ is $1$-Lipschitz for pairs of all, not necessarily bounded, metric spaces. We then study the properties of the mapping $\textbf{U}$. We show that, for a given dotted connected metric space $A$, the mapping $\Psi\colon X\mapsto X\times A$ from the proper class of all bounded ultrametric spaces ($X\times A$ is endowed with the Manhattan metric) preserves the Gromov-Hausdorff distance. Moreover, the mapping $\textbf{U}$ is inverse to $\Psi$. By a dotted connected metric space, we mean a metric space in which for an arbitrary $\varepsilon > 0$ and every two points $p,\,q$, there exist points $x_0 = p,\,x_1,\,\ldots,\,x_n = q$ such that $\max_{0\le j \le n-1}|x_jx_{j+1}|\le \varepsilon$. At the end of the paper, we prove that each class (proper or not) consisting of unbounded metric spaces on finite Gromov-Hausdorff distances from each other cannot contain an ultrametric space and a dotted connected space simultaneously.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 12 canonical work pages

[5]

Burago, Yu

D. Burago, Yu. Burago, S. Ivanov, A Course in Metric Geometry , Graduate Studies in Mathe- matics 33, AMS, 2001

work page 2001
[15]

A. A. Tuzhilin, Lectures on Hausdorﬀ and Gromov–Hausdorﬀ distance geometr y, arXiv:2012.00756, 2019. 10

arXiv 2012
[1]

S. A. Bogaty, A. A. Tuzhilin, Gromov–Hausdorﬀ class: its completeness and cloud geometr y, ArXiv e-prints, arXiv:2110.06101, 2021

arXiv 2021
[2]

S. I. Bogataya, S. A. Bogatyy, Isometric stabilizers of clouds , Topology and its applications, 329:108381 (2023)

work page 2023
[3]

S. I. Bogataya, S. A. Bogatyy, V. V. Redkozubov, A. A. Tuzh ilin, Clouds in Gromov–Hausdorﬀ Class: their completeness and centers , Topology and its applications, 329:108374 (2023)

work page 2023
[4]

S. I. Borzov, A. O. Ivanov, A. A. Tuzhilin, Extendability of Metric Segments in Gromov-Hausdorﬀ Distance, ArXiv e-prints, arXiv:2009.00458. (2020)

arXiv 2020
[6]

G. E. Carlsson, F. Memoli, Characterization, stability and convergence of hierarchi cal clustering methods, J. Mach. Learn., vol. 11, № 47, 1425-1470, 2010

work page 2010
[7]

Gromov, Structures m´ etriques pour les vari´ et´ es riemanniennes, Textes Math

M. Gromov, Structures m´ etriques pour les vari´ et´ es riemanniennes, Textes Math. 1 (1981)

work page 1981

Show all 15 references

[8]

Gromov, Metric structures for Riemannian and non-Riemannian space s, Birkh¨ user (1999)

M. Gromov, Metric structures for Riemannian and non-Riemannian space s, Birkh¨ user (1999)

1999
[9]

D. A. Herron, Gromov–Hausdorﬀ Distance for Pointed Metric Spaces , J. Anal., 2016, v. 24, N 1, pp. 1–38

2016
[10]

Kuratowski, Quelques probl` emes concernant les espaces m´ etriques non -s´ eparables, Funda- menta Mathematicae 25: pp

C. Kuratowski, Quelques probl` emes concernant les espaces m´ etriques non -s´ eparables, Funda- menta Mathematicae 25: pp. 534–545. (1935)

1935
[11]

I. N. Mikhailov, Gromov–Hausdorﬀ distances between normed spaces , arXiv:2407.01388, 2024 (in print)

2024 arXiv
[12]

I. N. Mikhailov, A. A. Tuzhilin, When the Gromov–Hausdorﬀ distance between ﬁnite-dimensio nal space and its subset is ﬁnite? , Commun. Math. Res., 2024 (accepted to appear in Vol. 41)

2024
[13]

Sunhyuk Lim, Facundo Memoli, Zane Smith, The Gromov–Hausdorﬀ distance between spheres , 2022, ArXiv e-prints, arXiv:2105.00611v5

2022 arXiv
[14]

Talipov, The Gromov–Hausdorﬀ Distance between Vertex Sets of Regula r Polygons Inscribed in a Single Circle , Moscow Univ

T. Talipov, The Gromov–Hausdorﬀ Distance between Vertex Sets of Regula r Polygons Inscribed in a Single Circle , Moscow Univ. Math. Bull., 78, 130–135 (2023)

2023

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [5]

Burago, Yu

D. Burago, Yu. Burago, S. Ivanov, A Course in Metric Geometry , Graduate Studies in Mathe- matics 33, AMS, 2001

work page 2001

[2] [15]

A. A. Tuzhilin, Lectures on Hausdorﬀ and Gromov–Hausdorﬀ distance geometr y, arXiv:2012.00756, 2019. 10

arXiv 2012

[3] [1]

S. A. Bogaty, A. A. Tuzhilin, Gromov–Hausdorﬀ class: its completeness and cloud geometr y, ArXiv e-prints, arXiv:2110.06101, 2021

arXiv 2021

[4] [2]

S. I. Bogataya, S. A. Bogatyy, Isometric stabilizers of clouds , Topology and its applications, 329:108381 (2023)

work page 2023

[5] [3]

S. I. Bogataya, S. A. Bogatyy, V. V. Redkozubov, A. A. Tuzh ilin, Clouds in Gromov–Hausdorﬀ Class: their completeness and centers , Topology and its applications, 329:108374 (2023)

work page 2023

[6] [4]

S. I. Borzov, A. O. Ivanov, A. A. Tuzhilin, Extendability of Metric Segments in Gromov-Hausdorﬀ Distance, ArXiv e-prints, arXiv:2009.00458. (2020)

arXiv 2020

[7] [6]

G. E. Carlsson, F. Memoli, Characterization, stability and convergence of hierarchi cal clustering methods, J. Mach. Learn., vol. 11, № 47, 1425-1470, 2010

work page 2010

[8] [7]

Gromov, Structures m´ etriques pour les vari´ et´ es riemanniennes, Textes Math

M. Gromov, Structures m´ etriques pour les vari´ et´ es riemanniennes, Textes Math. 1 (1981)

work page 1981