Pith. sign in

REVIEW 1 major objections 5 minor 33 references

Distance preservers for Lobachevsky space

T0 review · 1 major / 5 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read The paper classifies every entrywise transform that maps Lorentz–Gram matrices to Lorentz–Gram matrices, completing the negative-curvature counterpart of the classical zero- and positive-curvature results.

desk verdict Completes Schoenberg's classification for negative curvature; the only real soft spot is a load-bearing external theorem (Krein's screw-line trichotomy) imported without proof. read the letter →

arxiv 2608.22568 v1 pith:OMXDKAE7 submitted 2026-08-23 math.CA

Alexander Belton , Dominique Guillot , Apoorva Khare , Mihai Putinar This is my paper · ORCID
classification math.CA MSC 15A4551M1046E2242A82
keywords LobachevskyspaceLorentz–GrammatricesentrywisepreservershyperbolicdistanceBernsteinfunctionsscrewlineskernelsofrealtypeinfinitelydivisible
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

This paper asks which functions, applied entrywise to a Lorentz–Gram matrix, always produce another Lorentz–Gram matrix. For Lobachevsky space, the answer is exactly the parameterized family in display (1.1): a baseline power term, a possible jump at the identity, and an integral term, subject to one inequality coupling the drift to the jump and measure. If correct, this closes the last curvature case left open by the classical Euclidean and spherical classifications and gives a complete description of distance preservers of hyperbolic space.

What carries the argument

The load-bearing tool is the screw-line trichotomy: every metric function of a continuous helical path in Lobachevsky space is elliptic, parabolic, or hyperbolic, meaning that $\Lambda=\cosh\circ\lambda$ has one of three integral representations involving cosine transforms. Combined with the Hankel-kernel theorem and the fact that the Hadamard inverse of a Lorentz–Gram matrix is positive semidefinite and infinitely divisible, this trichotomy forces any preserver's transform $f\circ\cosh$ into the three integral forms from which the additive representation follows.

What would settle it

Take the parameter inequality seriously: choose $\alpha=1$, $\gamma=0$, $\nu=0$, and $\beta<1$, so $f(t)=1+\beta(t-1)$ is increasing but violates (1.2); if the $3\times3$ Lorentz–Gram matrix $\begin{bmatrix}1&a&b\\a&1&a\\b&a&1\end{bmatrix}$ with $b=2a^2-1$ is mapped by $f$ to a matrix with two positive eigenvalues for some $a>1$, the necessity direction is refuted. Conversely, any preserver not representable in form (1.1), for example a flat bounded function with $f(1)=1$ and $f(t)-1$ decaying faster than $t^{-s}$ for all $s>0$, would refute sufficiency if it preserves the class $M_{1+}$.

Watch

Extended reading notes

Core claim

A function $f:[1,\infty)\to[1,\infty)$ with $f(t)=1$ only at $t=1$ preserves the class of Lorentz–Gram matrices under entrywise application if and only if $f$ has the form $$f(t)=1+\gamma\mathbf{1}_{(1,\infty)}(t)+\$\beta$\frac{t^\$\alpha$-1}{\$\alpha$}+\int_{(0,\infty)}(1-$t^{{-s}}$)\,d\nu(s),$$ with $\alpha\in[0,1]$, $\beta,\gamma\ge 0$, and $\nu$ a positive Radon measure satisfying $\beta\ge\alpha(1+\gamma+\nu((0,\infty)))$ when $\alpha>0$ and $\int\min\{s,1\}\,d\nu(s)<\infty$ when $\alpha=0$, with not all of $\beta,\gamma,\nu$ zero. The same class is exactly the set of functions for which the inner transform $\varphi=\operatorname{arcosh}\circ f\circ\cosh$ sends the hyperbolic metric into another metric on Lobachevsky space. Continuous preservers also admit a multiplicative exponential-Bernstein representation, which the paper relates to subordinators.

Load-bearing premise

The characterization inherits its structure from a previously established trichotomy for the distance functions of helical paths in Lobachevsky space; if that trichotomy has a missing boundary case, the whole characterization would inherit the gap.

Editorial extensions

If this is right

  • Every power map $t\mapsto t^\alpha$ with $\alpha\in[0,1]$ is a preserver, so the known result that such powers preserve kernels of real hyperbolic type appears as a special case.
  • The class is closed under composition and pointwise limits, so the family (1.1) forms a complete semigroup of nonlinear transforms on Lorentz–Gram matrices.
  • Each continuous preserver admits a multiplicative representation $f(t)=t^\alpha\exp\!\left(\int_{(0,\infty)}(1-t^{-s})\,d\mu(s)\right)$, connecting the classification to Bernstein-function theory and to subordination of stochastic processes.
  • The fractional Ornstein–Uhlenbeck covariance $\exp(-a\,d(x,y)^\alpha)$ on hyperbolic space is not positive semidefinite for $\alpha>1$ in all sufficiently large dimensions, settling an open question about such fields.

Reading between the lines

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

  • Inference: The explicit family gives a direct route to data-driven hyperbolic embeddings: one can fit the parameters $(\alpha,\beta,\gamma,\nu)$ rather than fix a single kernel transform, because every admissible parameter choice provably preserves the hyperbolic geometry constraints.
  • Inference: The theorem suggests a precise analogy between Euclidean conditionally negative definite kernels and Lorentz–Gram kernels; the condition $\beta\ge\alpha(1+\gamma+\nu((0,\infty)))$ is the hyperbolic analogue of Bernstein-function monotonicity, so algorithms developed for Euclidean distance transforms may have direct hyperbolic counterparts.
  • Inference: A natural testable extension is to Gram matrices with more than one negative eigenvalue; the screw-line trichotomy likely resolves into a finite number of regimes indexed by the signature, and the same additive-integral method could apply.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 5 minor

Summary. The paper studies entrywise maps f:[1,∞)→[1,∞) that send the class of Lorentz–Gram matrices (Gram matrices of points in infinite-dimensional Lobachevsky space) to itself. The main result, stated in display (1.1), characterizes such distance preservers as functions of the form f(t)=1+γ 1_{(1,∞)}(t)+β(t^α−1)/α+∫_{(0,∞)}(1−t^{−s}) dν(s), where α∈[0,1], β,γ≥0, and ν is a positive Radon measure subject to β≥α(1+γ+ν((0,∞))) for α>0 and ∫ min{s,1} dν(s)<∞ for α=0. The necessity proof combines Krein's embedding theorem with Krein's classification of screw lines; the sufficiency is shown through explicit matrix decompositions, Bernstein-function theory, and closure properties. The paper also establishes closure under composition and pointwise limits, uniqueness of the representation, a multiplicative representation, and a negative answer to a question of Cohen and Lifshits on fractional Ornstein–Uhlenbeck fields.

Significance. If the main theorem is correct, it completes Schoenberg's classification of entrywise preservers in the constant-curvature trichotomy by filling the negative-curvature case. The representation is explicit and yields concrete consequences: it unifies and extends the Monod–Py power preservation result, resolves the Cohen–Lifshits question, and connects to complete Nevanlinna–Pick kernels and Bochner subordination. The proof is detailed and largely self-contained, with careful handling of boundary cases, pointwise limits, and the regularisation of discontinuous preservers. The main residual risk is the unproved importation of Krein's screw-line classification, which is load-bearing for the necessity direction.

major comments (1)
  1. [Section 4 (Screw lines) and Section 5 (Characterisation of distance preservers)] The necessity direction of the main theorem (display (1.1)) depends crucially on the trichotomy for metric functions of screw lines stated in Section 4. This theorem is imported from Krein [17] and Iokhvidov–Krein [15, Sections 26 and 27] without proof. The trichotomy is the bridge that forces f∘cosh to be elliptic, hyperbolic, or parabolic, and any mismatch in hypotheses or boundary conventions (e.g., the integrability conditions in (4.1), the support condition on σ, or the no-mass-at-0 statement) would propagate into the characterization. Since the paper claims to be self-contained, I request that the authors either prove the classification or provide a precise statement with all hypotheses and a verification that the classical sources establish exactly this version. At a minimum, a remark detailing the boundary cases would substantially reduce the residual risk.
minor comments (5)
  1. [Section 1, display (1.1)] The main characterization is stated in the abstract and in the introduction but is not labelled as a theorem. Numbering this result, e.g., as Theorem 1.1, and restating it in Section 5 would improve citability and readability.
  2. [Section 3.8] The uniqueness of the representing measure ν is attributed to 'standard Laplace-transform arguments'. Given that the rest of the paper is very careful, a brief proof or a precise reference for this uniqueness would be welcome.
  3. [Section 4] The screw-line trichotomy would benefit from a uniqueness statement for the representing measures in each of the three cases, since the later proofs identify the parameters a and b from asymptotic limits.
  4. [Introduction, Section 1] The sentence 'To ensure our presentation is self contained ... and full proofs of auxiliary results in Appendix A' is too strong: the proofs of Krein's embedding theorem (Theorem 3.1) and the screw-line classification (Section 4) are not provided in the appendix. Please tone down this claim or add the missing material.
  5. [Section 5.2] In the display involving the second difference, the notation uses forward increments defined just after; this is fine, but the definition of Δ_δ should be placed before its first use for readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central characterization is independently grounded in Krein's screw-line theorem and the paper's own matrix-analysis arguments.

full rationale

The main claim is the representation (1.1) for entrywise preservers of Lorentz–Gram matrices. The necessity direction proceeds from preservation of M_{1+} to an isometric embedding via Krein's certificate (Theorem 3.1), then to screw lines, then invokes Krein's classical trichotomy for metric functions of screw lines in Lobachevsky space. That trichotomy is an external, parameter-free classification theorem with stated hypotheses that do not include the target representation; it is cited but not proved, which is a correctness/verification consideration, not a circularity. The sufficiency direction constructs preservers directly from the additive Bernstein-type representation and proves preservation using elementary matrix inertia arguments, the Bapat–Micchelli Hadamard-inverse fact, and complete monotonicity. The only self-citation is [5, Theorem 2.7] in Section 3.6, used to assert that a 3-PMP matrix is positive semidefinite when proving that the jump-regularized version of a candidate preserver acts correctly. This is an auxiliary published matrix-positivity fact, independent of the classification being proved, and it does not reduce the main representation to its own input. No equation in the paper is defined in terms of the target result, and no fitted parameter is renamed as a prediction. The derivation chain is therefore not circular, and the score is set to 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim is a pure classification theorem: no experimental data, no fitted constants, and no invented entities. The list above records the main published theorems imported without proof; all are standard tools, with the exception of [5], a self-cited 3-PMP theorem used in Section 3.6. The parameters α, β, γ and ν in the final representation are classification data, not ad hoc fitted constants.

assumptions (6)
  • standard math Krein's screw-line trichotomy: every metric function of a screw line in Lobachevsky space has one of the elliptic, hyperbolic, or parabolic cosine-integral representations.
    Invoked in Section 4 to classify Λ=f∘cosh; quoted from [17] and [15, Sections 26 and 27], not proved in the present paper.
  • standard math Widder's Theorem VI.21: an analytic positive-semidefinite Hankel kernel h(x+y) has a representation as the Laplace transform of a finite positive Radon measure.
    Used in Sections 5.1 and 5.2 to pass from positive-semidefinite kernels to the additive representation of f.
  • standard math Bapat-Micchelli theorem: the Hadamard inverse of a matrix in M_{1+} is positive semidefinite and infinitely divisible.
    Used in Sections 3.3 and 3.6 and in Theorem 3.6 to show that terms t^{-s} preserve M_{1+}.
  • standard math Faraut-Harzallah kernel representation: positive-definite radial kernels on Lobachevsky space have the form h(t)=∫ t^{-s} dμ(s), and Corollaire 8.2 gives the multiplicative exponential-Bernstein representation.
    Used in Section 3.7 for positive-definite kernels and in Section 6 for the multiplicative representation f(t)=t^α exp(∫(1-t^{-s}) dμ(s)).
  • standard math Belton-Guillot-Khare-Putinar 3-PMP theorem: a symmetric nonnegative matrix whose 3-by-3 principal minors satisfy the 3-positive-minor condition is positive semidefinite.
    Cited as [5, Theorem 2.7] in Section 3.6 to prove the direct preserver property for the pointwise closure with a jump γ. This is a self-cited published theorem, not derived in the present paper.
  • standard math Quiggin-McCullough criterion: a positive kernel is a complete Nevanlinna-Pick kernel if and only if its reciprocal kernel matrices have exactly one positive eigenvalue.
    Used in Section 3.7 for Proposition 3.7 on complete Nevanlinna-Pick kernels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distance preservers for Lobachevsky space." pith.science (2026). https://pith.science/paper/OMXDKAE7

@misc{pith2026260822568,
  author       = {Pith},
  title        = {Pith review of: Distance preservers for Lobachevsky space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OMXDKAE7}},
  note         = {Machine review of arXiv:2608.22568}
}
read the original abstract

We obtain a complete description of the class of entrywise preservers of Lorentz-Gram matrices. This resolves, for the case of constant negative curvature, the classification of entrywise preservers obtained by Schoenberg in the zero-curvature (Euclidean) and constant-positive-curvature (spherical) settings.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [17]

    M.G. Krein. Helical curves in an infinite-dimensional Lobachevskii space and Lorentz transforma- tions. (Russian). InMeetings of the Moscow Mathematical Society,Uspekhi Mat. Nauk3:3(25):152– 165, 1948

    work page 1948

  2. [1]

    Agler and J.E

    J. Agler and J.E. McCarthy, Complete Nevanlinna–Pick kernels.J. Funct. Anal.175(1):111–124, 2000

    work page 2000

  3. [2]

    Bapat, Multinomial probabilities, permanents and a conjecture of Karlin and Rinott.Proc

    R.B. Bapat, Multinomial probabilities, permanents and a conjecture of Karlin and Rinott.Proc. Amer. Math. Soc.102(3):467–472, 1988

    work page 1988

  4. [3]

    Bapat and T.E.S

    R.B. Bapat and T.E.S. Raghavan,Nonnegative Matrices and Applications. Encyclopedia Math. Appl. 64. Cambridge University Press, Cambridge, 1997

    work page 1997

  5. [4]

    Bavaud, On the Schoenberg transformations in data analysis: theory and illlustrations.J

    F. Bavaud, On the Schoenberg transformations in data analysis: theory and illlustrations.J. Classification28(3):297–314, 2011

    work page 2011

  6. [5]

    Belton, D

    A. Belton, D. Guillot, A. Khare and M. Putinar, Simultaneous kernels of matrix Hadamard powers. Linear Algebra Appl.576:142–157, 2019

    work page 2019

  7. [6]

    Bertoin, Subordinators: examples and applications

    J. Bertoin, Subordinators: examples and applications. InLectures on Probability and Statistics, Ecole d’Et´ e de Probabilit´ es de Saint-Flour XXVII, Springer, Lect. Notes Math. 1717, Berlin, 1999, pp.1–91

    work page 1999

  8. [7]

    Cohen and M

    S. Cohen and M. A. Lifshits. Stationary Gaussian random fields on hyperbolic spaces and on Euclidean spheres.ESAIM Probab. Stat.16:165–221, 2012

    work page 2012

Show all 33 references
  1. [8]

    Faraut, Fonction brownienne sur une vari´ et´ e riemannienne

    J. Faraut, Fonction brownienne sur une vari´ et´ e riemannienne. InS´ eminaire de Probabilit´ es VII, edited by C. Dellacherie, P. A. Meyer and M. Weil, Lect. Notes Math. 321, Springer, Berlin, 1973, pp.61–76

    1973

  2. [9]

    Faraut and K

    J. Faraut and K. Harzallah. Distances hilbertiennes invariantes sur un espace homog` ene.Ann. Inst. Fourier (Grenoble)24(3):171–217, 1974

    1974

  3. [10]

    Ganea, G

    O. Ganea, G. B´ ecigneul and T. Hofmann, Hyperbolic neural networks. InAdvances in Neural Information Processing Systems (NeurIPS)31, edited by S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi and R. Garnett, Curran Associates, Inc., 2018, pp.7341–7352

    2018

  4. [11]

    Gangolli

    R. Gangolli. Positive definite kernels on homogeneous spaces and certain stochastic processes related to L´ evy’s Brownian motion of several parameters.Ann. Inst. H. Poincar´ e Sect. B3(2):121– 226, 1967

    1967

  5. [12]

    Haynsworth

    E.V. Haynsworth. Determination of the inertia of a partitioned Hermitian matrix.Linear Algebra Appl.1(1):73–81, 1968

    1968

  6. [13]

    R.A. Horn. The theory of infinitely divisible matrices and kernels.Trans. Amer. Math. Soc. 136:269–286, 1969

    1969

  7. [14]

    Horn and C.R

    R.A. Horn and C.R. Johnson.Matrix Analysis(second edition). Cambridge University Press, 2013

    2013

  8. [15]

    Iokhvidov and M.G

    I.S. Iokhvidov and M.G. Krein. Spectral theory of operators in spaces with indefinite metric. II. (Russian)Trudy Moskov. Mat. Obsc.8:413–496, 1959

    1959

  9. [16]

    J. Istas. Spherical and hyperbolic fractional Brownian motion.Electron. Commun. Probab.10:254– 262, 2005

    2005

  10. [18]

    M.G. Krein. Hermitian-positive kernels on homogeneous spaces. I. (Russian).Ukrain. Mat. ˇZurnal 1(4):64–98, 1949

    1949

  11. [19]

    Krioukov, F

    D. Krioukov, F. Papadopoulos, M. Kitsak, A. Vahdat and M. Bogu˜ n´ a, Hyperbolic geometry of complex networks.Phys. Rev. E82:036106, 2010

    2010

  12. [20]

    C. Loewner. Determination of the critical exponent of the Green’s function. InContemporary Problems in the Theory of Analytic Functions (Erevan, 1965), edited by M.A. Lavrentiev, Izdat. “Nauka”, Moscow, 1966, pp.184–187

    1965

  13. [21]

    C. Loewner. On schlicht-monotonic functions of higher order.J. Math. Anal. Appl.14(2):320–325, 1966

    1966

  14. [22]

    M. Merkle. Completely monotone functions: a digest. InAnalytic Number Theory, Approximation Theory, and Special Functions, edited by G.V. Milovanovi´ c and M.T. Rassias, Springer, New York, 2014, pp.347–364

    2014

  15. [23]

    Micchelli, Interpolation of scattered data: distance matrices and conditionally positive def- inite functions.Constr

    C.A. Micchelli, Interpolation of scattered data: distance matrices and conditionally positive def- inite functions.Constr. Approx.2(1):11-22, 1986. LOBACHEVSKY DISTANCE PRESERVERS 47

    1986

  16. [24]

    Monod and P

    N. Monod and P. Py, Self-representations of the M¨ obius group.Annales Henri Lebesgue2:259–280, 2019

    2019

  17. [25]

    Monod, Notes on functions of hyperbolic type.Bull

    N. Monod, Notes on functions of hyperbolic type.Bull. Belg. Math. Soc. Simon Stevin27(2):167– 202, 2020

    2020

  18. [26]

    Nickel and D

    M. Nickel and D. Kiela, Learning continuous hierarchies in the Lorentz model of hyperbolic geo- metry. InProc. 35th Int. Conf. on Machine Learning, edited by J. Dy and A. Krause, Proceedings of Machine Learning Research 80, 2018, pp.3779–3788

    2018

  19. [27]

    Roberts and D.E

    A.W. Roberts and D.E. Varberg.Convex Functions. Pure and Applied Mathematics 57, Academic Press, New York, 1973

    1973

  20. [28]

    F. Sala, C. De Sa, A. Gu and C. R´ e, Representation tradeoffs for hyperbolic embeddings. InProc. 35th Int. Conf. on Machine Learning, edited by J. Dy and A. Krause, Proceedings of Machine Learning Research 80, 2018, pp.4460–4469

    2018

  21. [29]

    Schilling, R

    R. Schilling, R. Song and Z. Vondraˇ cek.Bernstein Functions(second edition). De Gruyter Stud. Math. 37, Walter de Gruyter & Co., Berlin, 2012

    2012

  22. [30]

    Schoenberg

    I.J. Schoenberg. Remarks to Maurice Fr´ echet’s article ”Sur la d´ efinition axiomatique d’une classe d’espace distanci´ es vectoriellement applicable sur l’espace de Hilbert”.Annals of Mathematics 36(3):724–732, 1935

    1935

  23. [31]

    Tabaghi and I

    P. Tabaghi and I. Dokmani´ c, Hyperbolic distance matrices. InProc. 26th ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining, 2020, pp.1728–1738

    2020

  24. [32]

    von Neumann and I.J

    J. von Neumann and I.J. Schoenberg. Fourier integrals and metric geometry.Trans. Amer. Math. Soc.50(2):226–251, 1941

    1941

  25. [33]

    Widder.The Laplace Transform

    D.V. Widder.The Laplace Transform. Princeton University Press, 1946. (A. Belton)School of Engineering, Computing and Mathematics, University of Ply- mouth, Plymouth, UK Email address:alexander.belton@plymouth.ac.uk (D. Guillot)University of Delaware, Newark, DE, USA Email addr...

    1946

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.