REVIEW 3 major objections 4 minor 18 references
Hausdorff dimension of some subsets of the Lagrange and Markov spectra near $3$
T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash
Pith's one-line read This paper proves that arbitrarily close to 3, the Lagrange spectrum contains full-dimensional subsets $B_r$ where $D(t)=HD(\ell^{-1}(t))$ and $D$ is strictly increasing, and that the dimension of the Markov-minus-Lagrange spectrum below…
desk verdict New local structure for Lagrange/Markov spectra near 3 with a genuine new bound for M\L, but Theorem 1.1's level-set equality is delegated to a same-group preprint and needs referee scrutiny. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the horseshoe $\Lambda(2)$ of bi-infinite sequences over $\{1,2\}$ carrying the continued-fraction function $\lambda$, together with its hyperbolic sets of finite type called subhorseshoes. The central mechanism is a connection dichotomy: a subhorseshoe whose maximum is just above 3 either connects with the fixed orbit $\psi_b$ (meaning there is a larger subhorseshoe with supremum of $f$ below a prescribed level containing both), or it fails to connect and is forced, by a long list of connection schemes built from Markov-tree alphabets and bad cuts, to have word factors that force at least $n/5$ of each continuation, leading to dimension at most $C_0/n$. The Farey-sequence coding of ordered alphabets $(\alpha,\beta)$ and the renormalization algorithm supply the word-level constraints, while bounded-distortion estimates convert the forced letters into the dimension bound.
What would settle it
Take any $r$ and any $t$ in the constructed set $B_r$. Numerically estimate the Hausdorff dimension of the set of irrational continued fractions whose Lagrange value is exactly $t$ and compare it with $D(t)=HD(\mathcal{L}\cap(-\infty,t))$; a value strictly below $D(t)$ would disprove the level-set identity. The decisive structural check is whether the homeomorphism from the unstable Cantor set of $\Lambda(s_0,\epsilon_0)$ to $\ell^{-1}(t)$ constructed in Section 3.3 of [12] really has a H\"older inverse for the whole nested family of subhorseshoes used in the proof of Theorem 1.1; if any member of that family admits only a non-H\"older inverse, the strict monotonicity argument also fails.
Extended reading notes
Core claim
On the horseshoe model $\Lambda(2)$ for the classical spectra below $\sqrt{12}$, the paper proves Proposition 5.1: if a subhorseshoe has maximum $f$-value below $3+6^{-3n}$ and does not connect, before that maximum plus a small slack, with the periodic orbit $\psi_b$, then its Hausdorff dimension is at most $C_0/n$. From this, the proof decomposes each interval $(a_{r+1},a_r)$ into subhorseshoes that do or do not connect with $\psi_b$, shows the non-connecting ones are dimension-negligible, and defines $B_r$ as the points on which the connecting subhorseshoes accumulate. The identity $D(t)=HD(\ell^{-1}(t))$ on $B_r$ is obtained by following the construction of a homeomorphism from an unstable Cantor set to the exact level set, and strict increase follows from comparing nested subhorseshoes via the spectral decomposition theorem. For $\mathcal{M}\setminus\mathcal{L}$, every point below $3+\rho$ lies on a transient component joining two subhorseshoes, at least one of which does not connect with $\psi_b$; averaging the two dimension bounds gives the stated upper bound.
Load-bearing premise
The proof borrows from a companion preprint the claim that a correspondence between the unstable Cantor set of a subhorseshoe and the set of numbers with Lagrange value exactly $t$ is dimension-preserving; if that borrowed construction does not extend to the subhorseshoes used here, the exact-level-set identity and the strict increase of $D$ on $B_r$ collapse.
Editorial extensions
If this is right
- For each $r$, the set of $t$ in $(a_{r+1},a_r)$ where the exact level-set dimension $D(t)=HD(\ell^{-1}(t))$ is known to hold has the same Hausdorff dimension as the full interval, so the exceptions are dimension-negligible in that interval.
- The strict increase of $D$ on $B_r$ gives intervals arbitrarily close to 3 on which a full-dimensional subset of the spectrum has strictly increasing exact-level dimension, ruling out local constancy of $D$ along that subset.
- Because $B_r\subset\mathcal{L}'$ and $HD(B_r)=HD((a_{r+1},a_r)\cap\mathcal{L})$, the accumulation points of the Lagrange spectrum near 3 carry the full local dimension of the spectrum.
- The bound on $HD((\mathcal{M}\setminus\mathcal{L})\cap(-\infty,3+\rho))$ tends to 0 as $\rho\to0$ at the rate $(\log|\log\rho|)/|\log\rho|$, so the difference between the Markov and Lagrange spectra is dimension-sparse just to the right of 3.
- Since $d(t)=2D(t)$ for $t<t_1$, the theorem implies the cumulative dimension $d$ is realized, on $B_r$, by exact level sets, so $d(t)=2HD(\ell^{-1}(t))$ holds on a full-dimensional set of levels arbitrarily close to 3.
Reading between the lines
- The same connection/non-connection dichotomy may transfer to other endpoints of gaps in the Markov tree, not only near 3, to locate full-dimensional subsets where level-set dimensions are known exactly.
- The upper bound for $\mathcal{M}\setminus\mathcal{L}$ near 3 suggests the natural conjecture that the true asymptotic is exactly $(\log|\log\rho|-\log\log|\log\rho|+O(1))/|\log\rho|$, with the constant $C$ identified; the paper's proof gives $C=C_2+3\log 6\cdot C_0$ but does not claim sharpness.
- A testable numerical consequence is that for $t\in B_r$, algorithms estimating the dimension of continued-fraction level sets should return values matching $d(t)/2$; any systematic gap would indicate that the homeomorphism construction imported from the companion preprint does not extend to the full nested family of subhorseshoes used here.
- If the imported construction extends as claimed, the same method should yield $D(t)=HD(\ell^{-1}(t))$ on a set of full dimension in $(3,a_1)$, not just on the specific intervals selected by the sequence $a_r$.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the Hausdorff dimension of the Lagrange spectrum L and the difference set M \ L near the value 3. Theorem 1.1 constructs a decreasing sequence a_r → 3 such that, for each r, there is a set B_r ⊂ (a_{r+1}, a_r) ∩ L that carries the full Hausdorff dimension of the spectrum in that interval, satisfies D(t) = HD(ℓ^{-1}(t)) for every t ∈ B_r, makes D strictly increasing on B_r, and is contained in the accumulation set L′. Theorem 1.2 gives a quantitative upper bound HD((M \ L) ∩ (-∞, 3+ρ)) ≤ (log|log ρ| - log(log|log ρ|) + C)/|log ρ| for small ρ > 0. The central technical result is Proposition 5.1, which states that a subhorseshoe of Λ(2) with max f < 3 + 6^{-3n} that does not connect with the orbit ψ_b before max f + ε has Hausdorff dimension at most C_0/n. The proof of Proposition 5.1 occupies most of the paper and is built from a case analysis of connection schemes, forced-letter arguments, and an imported lemma, Lemma 5.7, from the same-group preprint [13].
Significance. If the main results hold, the paper makes a genuine contribution to the fine structure of the classical Lagrange and Markov spectra near 3. Proposition 5.1 is a substantial and plausibly useful estimate: it turns non-connection with ψ_b into a uniform dimension bound, and Theorem 1.2 derives from it a nontrivial upper bound for M \ L near 3. Theorem 1.1 identifies a large-dimensional subset of L where the level sets ℓ^{-1}(t) have the same dimension as the initial segments ℓ^{-1}(-∞,t), a property previously known only for the interior of the spectra. The paper contains no free parameters and the main new estimates are not used as inputs to the derivation. However, the proof of Theorem 1.1 depends crucially on an imported construction from [12, §3.3] that is not reproduced or verified for the specific subhorseshoe sequences used here, and the strict monotonicity step relies on a one-sentence citation to [12, Corollary 3.9]. These dependencies are load-bearing and make the current version conditional. The paper is not fully self-contained in exactly the places where its main theorem is most novel.
major comments (3)
- [§6.1, proof of Theorem 1.1] The proof of the second and fourth bullets of Theorem 1.1 is not carried out in this paper. After constructing the subhorseshoes Λ(s_n, ε_n), the text states that 'These conclusions are similar to the hypothesis of proposition 3.3 of [12]' and asserts that an increasing sequence of subhorseshoes yields a homeomorphism θ: K^u(Λ(s_0, ε_0)) → ℓ^{-1}(t) with Hölder inverse of exponent arbitrarily close to 1, with 'the details in Section 3.3 of [12]'. The present text verifies none of the conditions needed for that construction: nestedness of the sequence, convergence of max f|Λ(s_n, ε_n) to t, convergence of HD(Λ(s_n, ε_n)) to HD((Λ(2))_t), and the connecting-orbit condition required to produce a bijective map with Hölder inverse. The asserted limits do not by themselves give the homeomorphism, and the strict monotonicity statement D|_{B_r} is afterwards justified by Proposition 6.1, which is itself asserted in one sentence. Because these are load-bearing for Theorem 1.1, the authors should either reproduce the full argument of [12, §3.3] in the present setting or state and prove a precise theorem whose hypotheses are checked for the particular sequence Λ(s_n, ε_n).
- [§5.1.2, Lemma 5.7] Lemma 5.7 is stated without proof and attributed to [13], yet it is used in all of the connection schemes (Lemmas 5.8 through 5.13) that form the backbone of Proposition 5.1. The lemma compares λ(σ^j(... ; ... β_2)) with the maximum of two Markov values plus 1/2^{T-1}, and its exact hypotheses on the tails β_1, β_2, β_3 and on the finite common block are essential for the contradiction arguments. Since the proof of the main theorems rests on this lemma, the authors should include a proof or a self-contained statement with all hypotheses made explicit and verified in the context where it is applied.
- [§6.1, Proposition 6.1] Proposition 6.1 claims that for two subhorseshoes Λ̃_1 ⊂ Λ̃_2 with proper inclusion one has HD(Λ̃_1) < HD(Λ̃_2), and it is justified only by 'the spectral decomposition theorem and Corollary 3.9 of [12]'. This is a strong rigidity statement and it is used directly to prove that D|_{B_r} is strictly increasing. The cited corollary is not stated or proved in this paper, and the spectral decomposition theorem alone does not imply strict dimension increase for arbitrary proper inclusions of transitive hyperbolic sets. The authors should either prove Proposition 6.1 or state the precise result from [12] and verify that its hypotheses hold for the specific subhorseshoes Λ(t, ε) appearing in the proof.
minor comments (4)
- [§2.1, Proposition 2.4] The statement begins 'Suppose Λ_1 and Λ_1 are subhorseshoes'; the second symbol should be Λ_2.
- [§5.1.2, Lemma 5.7] The notation is not fully defined: the Markov value m of a finite or one-sided sequence is used without specifying how the one-sided tails are completed to bi-infinite sequences, and the role of the parameter T in the inequality is not discussed.
- [References] Reference [2] has a broken title, 'S. Roma˜naFractal dimensions' with a missing space and comma; reference [3] lacks publication data; reference [4] is listed as arXiv only despite being cited for specific numerical constants.
- [Abstract and Theorem 1.1] The abstract states B_r ⊂ (a_{r+1}, a_r) ∩ L′ while Theorem 1.1 states B_r ⊂ (a_{r+1}, a_r) ∩ L and adds B_r ⊂ L′ as the last bullet; the notation L′ is used before it is defined, and the two formulations should be reconciled.
Circularity Check
Theorem 1.1's key level-set equality and strict monotonicity are delegated to the same authors' preprint [12]; the proof asserts the hypotheses are 'similar' rather than verifying them, leaving the second and fourth bullets conditional on that self-citation.
-
self citation load bearing
[Section 6.1, proof of Theorem 1.1, paragraph before Proposition 6.1]
"These conclusions are similar to the hypothesis of proposition 3.3 of [12]. ... one is able to construct a homeomorphism θ :K u(Λ(s0,ǫ 0)) → ℓ−1(t) whose inverse is H¨ older with exponent arbitrarily close to one. ... one gets easily that D(t) = HD (ℓ−1(t)). The details are in Section 3.3 of [12]."
The proof of the second bullet of Theorem 1.1, namely D(t)=HD(ℓ^{-1}(t)) for t∈B_r, is not carried out here. Instead, the text asserts that the constructed subhorseshoes satisfy hypotheses 'similar' to Proposition 3.3 of [12] and refers to Section 3.3 of that same-author preprint for the homeomorphism construction. The present paper does not verify the specific hypotheses needed for Λ(s_n,ε_n), such as nestedness, convergence of max f to t, dimension convergence to HD((Λ(2))_t), and the connecting-orbit condition required by [12]. Thus the central equality is imported from a companion paper rather than derived in this paper, and the theorem's second bullet is load-bearing on that self-citation.
-
self citation load bearing
[Section 6.1, Proposition 6.1 and the sentence introducing it]
"Now, the spectral decomposition theorem and Corollary 3.9 of [12] let us conclude the following proposition Proposition 6.1. Given two subhorseshoes ˜Λ 1 and ˜Λ 2 of Λ such that ˜Λ 1 ⊈ ˜Λ 2, we have HD (˜Λ 1)<HD (˜Λ 2)."
The strict monotonicity of D on B_r, the fourth bullet of Theorem 1.1, is justified through Proposition 6.1, which is itself introduced as a consequence of 'Corollary 3.9 of [12]'. Corollary 3.9 belongs to the same authors' earlier preprint and is neither proved nor stated in this paper. Therefore the strict monotonicity assertion also rests on an unverified, load-bearing self-citation rather than on a self-contained proof.
full rationale
The paper contains substantial independent work: Proposition 5.1 is proved in Section 5 via the connection schemes, and Theorem 1.2 is derived from Proposition 5.1 together with the published estimates from [2]; those parts are not circular. The circularity concern is concentrated in Theorem 1.1, whose second and fourth bullets are not proved in the paper but are delegated to the same authors' preprint [12], with only a claim that the hypotheses are 'similar'. This is a load-bearing self-citation chain: if the imported construction does not extend to the particular subhorseshoes Λ(s_n,ε_n) used here, the asserted properties of B_r fail. There are no fitted parameters, no quantity is renamed as a prediction, and no equation is equal to itself by construction, so the issue is not a definitional circle. However, the central claim is partially conditional on an internal companion paper, so the appropriate score is 4 rather than a higher score reserved for full reduction by construction or by definition.
Assumptions & free parameters
assumptions (7)
- domain assumption Symbolic characterization of the Lagrange and Markov spectra via bi-infinite sequences and lambda(omega), and the realization of the initial segment of the classical spectra up to sqrt(12) as dynamical spectra of the horseshoe Lambda(2).
- domain assumption Theorem 2.8: for n at least 68, the sets of length-n subwords of Sigma(3+6^{-3n}) and Sigma(3) coincide.
- standard math Conservative horseshoe dimension formulas: HD(Lambda~)=2 HD(K^u(Lambda~)) and the transient set formula HD(T_{\Lambda_1,\Lambda_2})=(HD(Lambda_1)+HD(Lambda_2))/2.
- domain assumption Proposition 3.7: for every epsilon greater than 0 there is a subhorseshoe inside (Lambda(2))_{3+epsilon} containing all periodic orbits psi_p for periods from the Farey tree.
- domain assumption Lemma 5.7 from [13]: comparing sequences with a common prefix gives a bound on Lagrange values by the maximum of two continuations plus a small error term.
- domain assumption Section 3.3 and Corollary 3.9 of [12]: existence of a homeomorphism from K^u(Lambda(s_0, eps_0)) to ell^{-1}(t) with Holder inverse, and strict monotonicity of Hausdorff dimension of nested subhorseshoes.
- domain assumption Inequalities (1.1) and (1.2) for d(3+rho) from [2].
Cite this review
Pith. "Pith review of Hausdorff dimension of some subsets of the Lagrange and Markov spectra near $3$." pith.science (2026). https://pith.science/paper/RKM7M5P7
@misc{pith2026250420300,
author = {Pith},
title = {Pith review of: Hausdorff dimension of some subsets of the Lagrange and Markov spectra near $3$},
year = {2026},
howpublished = {\url{https://pith.science/paper/RKM7M5P7}},
note = {Machine review of arXiv:2504.20300}
}
abstract
We study the sets $\mathcal{L}$ and $\mathcal{M}\setminus\mathcal{L}$ near $3$, where $\mathcal{L}$ and $\mathcal{M}$ are the classical Lagrange and Markov spectra. More specifically, we construct a strictly decreasing sequence $\{a_r\}_{r\in \mathbb{N}}$ converging to $3$, such that for any $r$ one can find a subset $\mathcal{B}_r\subset (a_{r+1},a_r)\cap \mathcal{L}^{'}$ with the property that the Hausdorff dimension of $((a_{r+1},a_r)\cap \mathcal{L})\setminus \mathcal{B}_r$ is less than the Hausdorff dimension of $\mathcal{B}_r$ and for $t\in \mathcal{B}_r$ the sets of irrational numbers with Lagrange value bounded by $t$ and exactly $t$ respectively, have the same Hausdorff dimension. We also show that, as $t$ varies in $\mathcal{B}_r$, this Hausdorff dimension is a strictly increasing function. Finally, in relation to $\mathcal{M}\setminus \mathcal{L}$, we find $C>0$ such that we can bound from above the Hausdorff dimension of $(\mathcal{M}\setminus \mathcal{L})\cap (-\infty,3+\rho)$ by $\frac{\log (\abs{\log \rho})-\log (\log(\abs{\log \rho}))+C}{\abs{\log \rho}}$ if $\rho>0$ is small.
Reference graph
Works this paper leans on
- [13]
- [2]
-
[12]
C. G. Moreira and C. Villamil. Concentration of dimension in extrema l points of left-half lines in the Lagrange spectrum, Preprint (2023) available at arXiv:2309.1 4646
2023
-
[1]
Bombieri, Continued fractions and the Markoff tree , Expo
E. Bombieri, Continued fractions and the Markoff tree , Expo. Math. 25, 187–213 (2007) Zbl 1153.11030 MR 2345177
-
[3]
Lima and C
D. Lima and C. G. Moreira, Dynamical characterization of initials segments of the Mar kov and Lagrange spectra
-
[4]
C. Matheus, C. G. Moreira, M. Pollicott and P. Vytnova, Hausdorff dimension of Gauss–Cantor sets and two applications to classical Lagr ange and Markov spectra . https://arxiv.org/abs/2106.06572
-
[5]
Cerqueira, C
A. Cerqueira, C. Matheus, C. G. Moreira. Continuity of Hausdor ff dimension across generic dynamical Lagrange and Markov spectra. Journal of Modern Dynamics , 12:151–174, 2018
2018
-
[6]
D. Lima, C. Matheus, C. G. Moreira and S. Roma˜ na Classical and Dynamical Markov and Lagrange spectra, World Scientific , 2020
2020
Show all 18 references
-
[7]
Freiman, Non-coincidence of the spectra of Markov and of Lagrange , Mat
G.A. Freiman, Non-coincidence of the spectra of Markov and of Lagrange , Mat. Zametki 3, 195–200 (1968) (in Russian) Zbl 0155.09602 MR 0227110
1968
-
[8]
Freiman, Diophantine approximation and geometry of numbers (The Mar koff spectrum) , Kalininskii Gosudarstvennyi Universitet, Moscow, 1975
G.A. Freiman, Diophantine approximation and geometry of numbers (The Mar koff spectrum) , Kalininskii Gosudarstvennyi Universitet, Moscow, 1975
1975
-
[9]
Lima and C
D. Lima and C. G. Moreira. Phase transtitions on the Markov and L agrange dynamical spectra. Annales de L’Institute Henri Poincar´ e (C), Analyse non-lineaire, 1 –31,2020
2020
-
[10]
A. Markoff. Sur les formes quadratiques binaires ind´ efinies. Math.Ann., 15:381–406, 1879
-
[11]
C. G. Moreira, C. Villamil and D. Lima. Continuity of fractal dimens ions in conservative generic Markov and Lagrange dynamical spectra, Preprint (2023) availab le at arXiv:2305.07819
2023
-
[14]
Villamil
C. Villamil. Concentration of dimension in the Lagrange spectrum, Preprint (2024) available at arXiv:2411.16939
2024 arXiv
-
[15]
C. G. Moreira. Geometric properties of Markov and Lagrange s pectra. Annals of Math. , 188: 145–170, 2018
2018
-
[16]
Perron, ¨Uber die Approximation irrationaler Zahlen durch rational e II , S.-B
O. Perron, ¨Uber die Approximation irrationaler Zahlen durch rational e II , S.-B. Heidelberg Akad. Wiss., Abh. 8, 1921, 12 pp
1921
-
[17]
Cusick and M
T. Cusick and M. Flahive, The Markoff and Lagrange spectra, Mathematical Surveys and Mono- graphs, 30. American Mathematical Society, Providence, RI, 1989. x+97 pp
1989
-
[18]
Matheus, C., Moreira, C.G.: Fractal geometry of the complemen t of Lagrange spectrum in Markov spectrum. Comment. Math. Helv. , 95, no. 3, 593–633 (2020) SUSTech International Center for Mathematics, Shenzhen, G uangdong, People’s Republic of China and IMPA, Estrada Dona Casto...
2020
Reviewed August 16, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.