{"id":"f47ab5c6-305d-4415-be68-98e87220b15d","arxiv_id":"2511.02517","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Random Brooks–Makover surfaces have first Laplacian eigenvalue > 1/4 − ε with probability → 1 for every ε > 0.","lead":"Random closed hyperbolic surfaces built from random 3-regular graphs with orientations—the Brooks–Makover model—have Laplace spectral gap approaching the absolute bound 1/4 with probability tending to 1. This settles the last open case of the nearly-optimal spectral-gap conjecture among the three standard models of random hyperbolic surfaces.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm completeness in §3.2.2 is asserted but unproven; if it misses equivalence classes, the trace expansion fails.","rationale":"The reader’s weakest assumption is precisely the exhaustiveness of the §3.2.2 algorithm, and I agree this is the most load-bearing unproven step. The main structure of the proof — trace expansion, norm bound, strong convergence, then transfer via large cusps — is credible and consistent with the cited literature. However, Proposition 3.7 is the first point where an infinite sum over equivalence classes must be controlled, and all subsequent estimates flow from it. The algorithm is a plausible enumeration, but without a formal proof of surjectivity (and of the invariant relating a to η), the bound |F_h|≤k^{2h} could be too small, or the summation in (10) could be missing classes. The proposed test—constructing the inverse map—is the natural way to settle the issue. I do not see a more serious concern: the Λ1(H/PSL(2,Z))>1/4 statement is likely a minor misstatement (the needed fact is that there are no eigenvalues below 1/4−ε), and the other cited black-box steps are standard. Hence the reader’s conditional verdict remains appropriate; no verdict change is needed.","tokens_in":31348,"tokens_out":20585,"duration_ms":194401,"concrete_test":"Prove the missing bijection directly: for each word ω = x1 x2^{i1} x1 ··· x1 x2^{ik}, define a map from algorithm outputs (A, with a=h+1) to equivalence classes in I(ω) with η(Γ)=h, and construct the inverse as follows. Given (g,Γ)∈I(ω), record the ordered images of the x2-cycles Δ_1,...,Δ_k of Γ_X(ω); show that the algorithm’s step-by-step choices reproduce exactly this sequence, that every intermediate graph G_t is the image of the subgraph Γ_t, and that the number of Operation II steps equals η(Γ)+1. If this inverse cannot be constructed for some ω with k≤4, the equality used in Prop. 3.7 is false and the theorem is in jeopardy; if it succeeds, the completeness claim is verified in the only way that matters.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Proposition 3.7 relies on the equality #{ (g,Γ)∈I(ω) : η(Γ)=h } = |F_{h+1}(ω)| (used immediately before equation (10)) and on the bound |F_h(ω)| ≤ k^{2h} from Lemma 3.5. The algorithm on pages 24–26 is stated to “output all possible elements in I(ω)”, but no formal bijection or induction is supplied. The algorithm’s nondeterministic choices (Operation I vs. II, choice of w_l, final step) must generate every equivalence class exactly once up to the defined equivalence, and the bookkeeping parameter a must satisfy a = η(G)+1. If any class is omitted, or if the correspondence between a and η is off by even one, the sum over h in Proposition 3.7 is not controlled; the stated asymptotic expansion for En[fix_ω] is unsupported. Consequently Proposition 4.1 (the trace expansion), Proposition 4.3 (the norm bound), and ultimately Theorem 4.6 (strong convergence) collapse, since they all depend on the expectation En[fix_ω] being polynomial in 1/n with the stated error. No alternative proof of this enumeration is given, so this is a load-bearing gap in the central argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves that a random hyperbolic surface in the Brooks–Makover model has first Laplacian eigenvalue larger than 1/4 - ε with probability tending to 1, for every ε > 0. The proof gives a new description of the Brooks–Makover ensemble as degree-6n covers of the modular surface with constrained generators (σ of order 2 with no fixed points, τ of order 3 with no fixed points), preserving the uniform measure up to a constant fiber size (Prop. 2.4). The main technical work is a fixed-point count for words in F_2: fixed points of the random permutation are encoded as X-labeled graph homomorphisms into the associated graph Γ(σ,τ) (Lemma 3.1), and the homomorphism set is decomposed into surjective and injective parts. Proposition 3.7 gives an asymptotic expansion for E_n[fix_ω], Proposition 4.1 converts this into a trace expansion for the permutation representation of PSL(2,Z), and Proposition 4.3 gives the needed norm bound u1(α^p). Following the template of Magee–Puder–van Handel, these imply strong convergence (Thm. 4.6), spectrum coincidence for the cusped surfaces (Cor. 4.7), and finally Theorem 1.1 via the large-cusps comparison of Brooks–Makover.","tokens_in":31646,"tokens_out":6512,"duration_ms":66689,"significance":"If correct, this is a breakthrough: it completes the nearly optimal spectral gap conjecture for all three principal models of random closed hyperbolic surfaces. The paper contains genuinely novel ingredients: a measure-preserving reformulation of the Brooks–Makover model as a constrained random PSL(2,Z)-cover, and a graph-theoretic enumeration scheme for the fixed-point counts with explicit, non-fitted coefficients u_i(γ). The announced theorem is an unambiguous, falsifiable asymptotic statement, and the main line of argument — trace expansion plus norm bound implying strong convergence — is the now-standard and credible polynomial-method template. However, the paper's central combinatorial enumeration step contains an asserted-but-unproven completeness statement, and the norm-bound proof uses an unexplained positivity reduction; these are load-bearing and must be repaired before the result can be considered established.","major_comments":[{"comment":"The algorithm on pages 24–26 is introduced as one that will 'output all possible elements in I(ω)', but no formal proof of completeness is given. Proposition 3.7 uses the equality #{(g,Γ)∈I(ω): η(Γ)=h} = |F_{h+1}(ω)| and the bound |F_h(ω)| ≤ k^{2h} to control the infinite sum over I(ω). If any equivalence class of surjective homomorphisms is not generated by the recursive choices, or if the parameter a is not equal to η(G)+1 for every generated class, then the asymptotic expansion of E_n[fix_ω] is unsupported, and Propositions 4.1, 4.3 and Theorem 4.6 collapse. The manuscript needs an explicit bijection or induction showing that every element of I(ω) is realized by exactly one run of the algorithm up to the defined equivalence, with the bookkeeping invariant a=η+1 verified for all classes, not only for outputs of the algorithm.","section":"§3.2.2, Lemma 3.5 and Proposition 3.7"},{"comment":"Even if the algorithm is complete, Proposition 3.8's identity v_0(ω)=|F_1(ω)|=d(µ) requires more than Lemma 3.6 supplies. Lemma 3.6 identifies the target graph G as Γ_X(ν) with ω=ν^q, but it does not show that for a fixed divisor q there is exactly one equivalence class of surjective homomorphisms f: Γ_X(ω) → Γ_X(ν). The lower bound constructs one natural homomorphism per divisor, but the upper bound |F_1(ω)|≤d(µ) needs a uniqueness argument. Without it, v_0(ω), and hence u_1(γ) in Lemma 4.2, could be larger than d(µ)-1, affecting the norm bound in Proposition 4.3.","section":"§3.2.2, Lemma 3.6 and Proposition 3.8"},{"comment":"The proof states: 'From [MdlS24, Proposition 6.3], we can assume that α has positive coefficients' and then normalizes Σ a_γ=1. This reduction is not immediate, because u_1 is linear on C[PSL(2,Z)] while the operator norm ∥λ(α)∥ is not linear in α. For a general self-adjoint α with signed coefficients, replacing α by a positive-coefficient version changes the value of ∥λ(α)∥, so the inequality for the modified polynomial does not imply the required inequality for the original α. The authors must either state the precise form of [MdlS24, Prop. 6.3] that applies to this u_1 and to the signed expansion u_1(α^p), or prove Proposition 4.3 directly for signed α (e.g. by decomposing α into positive and negative parts and using the rapid decay property invoked earlier in the same paragraph). As written, this is a load-bearing gap in the proof of strong convergence.","section":"§4.3, proof of Proposition 4.3"}],"minor_comments":[{"comment":"Typo: 'there there exists' should be 'there exists'.","section":"§2.2, Lemma 2.2"},{"comment":"The name Bollobás is spelled 'Bollabás' in the introduction; also the notation F⋆_n is used inconsistently in a few places (e.g. the proof of Theorem 4.8 writes (σ,τ)∈F⋆_n where (Γ,O)∈F⋆_n is meant).","section":"§1"},{"comment":"The proof refers to 'Lemma 3.2' while the statement is Proposition 3.2; the numbering should be corrected.","section":"§3.2.1, proof of Proposition 3.2"},{"comment":"The proof of part (2) says 'the proofs for the cases of c and c^2 are similar, that we leave to interested readers.' Since Lemma 4.4 is used essentially in Proposition 4.3, the omitted cases should be written out or at least summarized in enough detail for verification.","section":"§4.4, Lemma 4.4"}],"recommendation":"major_revision","confidential_remarks":"This is a strong and potentially important paper, and the main architecture is credible. The decision hinges on whether the authors can supply a complete proof of the enumeration completeness in §3.2.2 and clarify the positivity reduction in §4.3. These are not cosmetic issues; they are exactly the points on which the trace expansion and the norm bound rest. I would encourage the editor to seek a revised version with these proofs added."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is the real thing — it closes the last of the three standard models. If the proof holds, random Brooks–Makover surfaces have λ1 > 1/4 − ε with probability tending to 1, exactly the conjecture. The paper deserves serious refereeing.\n\nWhat is genuinely new: the identification of each Brooks–Makover surface with a degree-6n cover of the modular surface (Thm 2.5) is clean and useful. The trace expansion (Prop 4.1) and norm bound (Prop 4.3) fit the MPvH25 template, but the word-counting section is a real adaptation to PSL(2,Z), not a copy. The counting of injective homomorphisms is standard and correctly done. I see no circularity: u1 comes from fixed-point counts, not from the target gap, and the external dependencies are all legitimate.\n\nSoft spots, in proportion:\n\n1. The load-bearing issue is exactly the one flagged in the stress test. The §3.2.2 algorithm is asserted to output all equivalence classes in I(ω), but no formal bijection or induction is given. Lemma 3.5 bounds the number of outputs, and Prop 3.7 needs the equality #{(g,Γ)∈I(ω): η(Γ)=h} = |F_{h+1}(ω)|. That equality is not proved. If some classes are missed, the trace expansion collapses. The algorithm is plausible — Operation II is the only way to close an x1-cycle — and I suspect a referee can fill the gap, but it is not a cosmetic omission.\n\n2. The abstract promises a c/log n rate; Theorem 1.1 only states the ε version. The proof as written gives ε. Either the abstract is overclaiming or there is a missing argument.\n\n3. “λ1(H/PSL(2,Z)) > 1/4” is an abuse of notation: the modular surface has eigenvalue 0. The needed fact is that there are no eigenvalues in (0,1/4−ε), which is the standard result. That should be stated correctly.\n\nThe citation pattern is fine; the reliance on MPvH25 and HM23 is explicit and appropriate.\n\nWho is this for: anyone working on random hyperbolic surfaces or strong convergence. I would bring it to a reading group and would cite it once the enumeration proof is written. It deserves a real referee, not a desk reject. Recommendation: send to a good journal, with the referee explicitly asked to verify the Lemma 3.5 / Prop 3.7 enumeration claim.","headline":"Credible proof of the last open random-surface spectral-gap conjecture, with one load-bearing enumeration gap that should be fixed before the result is regarded as settled.","tokens_in":32194,"tokens_out":5682,"would_cite":true,"duration_ms":65293,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["58J50","30F10","05C80"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that a random hyperbolic surface in the Brooks–Makover model has first Laplacian eigenvalue λ1 > 1/4 − ε with probability converging to 1, confirming the nearly optimal spectral gap conjecture for this model.","keywords":["random Belyi surfaces","Brooks-Makover model","spectral gap","first Laplacian eigenvalue","hyperbolic surfaces","strong convergence","random permutations","PSL(2,Z)"],"falsifier":"Compute, for a small word $\\omega$ (such as $\\omega = x_1 x_2 x_1 x_2^2 x_1 x_2 x_1 x_2 x_1 x_2^2$ used as an example in the paper), all surjective $X$-labeled graph homomorphisms from $\\Gamma_X(\\omega)$ to subgraphs of $\\Gamma(\\sigma, \\tau)$ for small $n$, and verify whether the recursive Operations I/II produce every equivalence class. Exhibiting one missed class, or showing $|F_h(\\omega)| > k^{2h}$ for some $h$, would invalidate Lemma 3.5 and the expected-trace expansion.","tokens_in":31200,"feed_emoji":"📐","tokens_out":7124,"duration_ms":71880,"temperature":0.7,"texified_at":"2026-08-05T20:36:29.687339+00:00","pith_summary":"Random closed hyperbolic surfaces of large genus cannot have first Laplacian eigenvalue above $1/4$, and $1/4$ is conjecturally achievable. This paper proves that in the Brooks–Makover model—random surfaces glued from $2n$ ideal triangles according to a random 3-regular graph with orientation—the first eigenvalue exceeds $1/4 - \\epsilon$ for every $\\epsilon > 0$, with probability tending to 1. It is the last of the three standard models of random closed hyperbolic surfaces for which this nearly optimal spectral gap conjecture was open, so the result completes a common picture. The proof works by reinterpreting each Brooks–Makover surface as a degree $6n$ cover of the modular surface $H/PSL(2,\\mathbb{Z})$ and proving strong convergence of the associated random permutation representations of $PSL(2,\\mathbb{Z})$; since the modular surface itself has $\\lambda_1 > 1/4$, the gap transfers to the random covers.","texify_model":"deepseek-v4-flash","texify_usage":{"total_tokens":11623,"prompt_tokens":886,"completion_tokens":10737,"prompt_tokens_details":{"cached_tokens":0},"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":886,"completion_tokens_details":{"reasoning_tokens":9833}},"feed_headline":"Random Belyi surfaces attain near-optimal spectral gaps","feed_subtitle":"The first Laplacian eigenvalue exceeds any value below 1/4 with probability tending to 1, proving the last open random-surface case.","key_machinery":"The central device is the identification of a Brooks–Makover surface with the covering surface $PSL(2,\\mathbb{Z}) \\backslash (\\Phi H \\times [6n])$ determined by the homomorphism $\\Phi: PSL(2,\\mathbb{Z}) \\to S_{6n}$ sending the standard generators to $\\sigma, \\tau$. The proof then follows the polynomial method for strong convergence: expected traces of random permutation matrices are expanded in powers of $1/(6n)$, with coefficients controlled by counting fixed points of random permutations. Fixed points are counted by $X$-labeled graph homomorphisms from the word graph $\\Gamma_X(\\omega)$ to the permutation graph $\\Gamma(\\sigma, \\tau)$: surjective homomorphisms are enumerated by a recursive algorithm (Operations I/II) and injective ones by falling-factorial expectations; the lea","core_discovery":"For any $\\epsilon > 0$, with probability tending to 1 as $n \\to \\infty$, the closed conformal compactification $S_C(\\Gamma, O)$ of a random Brooks–Makover cusped surface $S_O(\\Gamma, O)$ satisfies $\\lambda_1(S_C) > 1/4 - \\epsilon$. Equivalently, the random cusped surface $S_O$ has no eigenvalues in $(0, 1/4 - \\epsilon)$. The paper establishes this by giving a new geometric description of the Brooks–Makover ensemble: each surface is isometric to a degree $6n$ covering of $H/PSL(2,\\mathbb{Z})$ coming from a pair $(\\sigma,\\tau)$ of permutations, $\\sigma$ of order 2 and $\\tau$ of order 3, and then proving that the permutation representation of $PSL(2,\\mathbb{Z})$ defined by $(\\sigma,\\tau)$ is strongly convergent to the regular representation as $n$ grows. The large-cusps lemma transfers the spectral gap from the","pith_inferences":["Extension: If the explicit constants in the trace expansion are tracked, the method may yield effective finite-n lower bounds for λ1, not just an asymptotic statement.","Extension: The description of the Brooks–Makover ensemble as an asymptotically measure-zero subset of all PSL(2,Z)-homomorphisms suggests that other geometric statistics—diameter, systole, Cheeger constant—could be attacked with the same permutation-model machinery.","Extension: The counting algorithm might generalize from PSL(2,Z) ≅ Z2 ⋆ Z3 to other free products of finite cyclic groups, giving spectral-gap results for random surfaces built from ideal polygons with more sides.","Extension: The divisor-counting function d(µ) appearing in the leading coefficient is the same arithmetic input seen in random-cover eigenvalue statistics, hinting that the method could be pushed toward the question of a positive proportion of Ramanujan-like surfaces with λ1 ≥ 1/4."],"forward_implications":["For every ε > 0, with probability tending to 1, the closed Brooks–Makover surface S_C(Γ,O) has λ1 > 1/4 − ε; the same holds for the cusped surface S_O.","The Brooks–Makover model joins the other two standard models in which the nearly optimal spectral gap conjecture is now known to hold.","The strong-convergence result for permutation representations of PSL(2,Z) applies to every finite-support element of the group algebra, so traces and norms of arbitrary words converge to their regular-representation limits.","Because the base modular surface has λ1 > 1/4, the proof transfers a spectral gap from a fixed base to random covers, the same mechanism used in covering models.","The abstract's quantitative version (1/4 − c/log n for a universal c > 0) is stronger than the ε statement, giving an explicit rate of convergence in addition to the asymptotic result."],"fun_headline_variants":["Random Belyi surfaces hit spectral gaps within ε of 1/4","Near-optimal spectral gap proved for random Belyi surfaces","Belyi surfaces approach the optimal spectral gap 1/4","With high probability, Belyi surfaces beat 1/4 − ε","Random Belyi surfaces achieve spectral gap > 1/4 − ε"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing assumption is the exhaustiveness of the recursive algorithm in Section 3.2.2—that it generates every equivalence class of surjective $X$-labeled graph homomorphisms, so the bounds on $|F_h(\\omega)|$ and the fixed-point expectation are complete—which the paper asserts but does not prove by an explicit bijection.","fun_headline_variants_meta":{"raw":{"variants":["Random Belyi surfaces hit spectral gaps within ε of 1/4","Near-optimal spectral gap proved for random Belyi surfaces","Belyi surfaces approach the optimal spectral gap 1/4","With high probability, Belyi surfaces beat 1/4 − ε","Random Belyi surfaces achieve spectral gap > 1/4 − ε"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000276,"raw_usage":{"total_tokens":1421,"prompt_tokens":617,"completion_tokens":804,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":361,"completion_tokens_details":{"reasoning_tokens":708}},"tokens_in":361,"tokens_out":804,"duration_ms":7784,"temperature":1.0,"reasoning_tokens":708,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T00:09:12.553285+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a small word $\\omega$ (such as $\\omega = x_1 x_2 x_1 x_2^2 x_1 x_2 x_1 x_2 x_1 x_2^2$ used as an example in the paper), all surjective $X$-labeled graph homomorphisms from $\\Gamma_X(\\omega)$ to subgraphs of $\\Gamma(\\sigma, \\tau)$ for small $n$, and verify whether the recursive Operations I/II produce every equivalence class. Exhibiting one missed class, or showing $|F_h(\\omega)| > k^{2h}$ for some $h$, would invalidate Lemma 3.5 and the expected-trace expansion.","supporting_citations":[],"review_version":1}