{"id":"cb6b4466-2b78-49f0-beeb-dc901f573dcd","arxiv_id":"2507.16794","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Random and planted graphs with degree-3 interior and degree-1 boundary vertices yield expander families, and their pants decompositions give hyperbolic surfaces with n comparable to g cusps and a uniform spectral gap.","lead":"This paper introduces a random graph model with degree-3 and degree-1 vertices and proves that, when the degree-1 vertices are not too numerous, a typical graph is connected and has a uniform spectral gap. It then turns such graphs into noncompact hyperbolic surfaces whose cusp count grows linearly with the genus while the spectral gap stays uniformly positive.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 6.4's length decomposition omits closed geodesics contained in a single pair of pants, the loop case used in Theorem 1.3, so the surface transfer may fail without an additional argument.","rationale":"I agree with the reader that Lemma 6.4 is the weakest load-bearing step. The concern can be sharpened: the proof's classification of the realizing multicurve into β_i and γ_j is exhaustive only when no component of α lies entirely inside one pair of pants. That missing type appears exactly when the graph has a loop, and loops are introduced deliberately in Theorem 1.3 to adjust the cusp count while preserving the Cheeger constant. If such a component has length with no positive lower bound in terms of a, the lower bound on h(X(G)) collapses; if it does have such a bound, the proof needs a third term and the constants C(a) must be recomputed. The rest of the construction is plausible: the tree-planting lemma has a repairable isolation argument, the genus arithmetic is correct when read with the intended parentheses, and the random-graph sections are not needed for Theorem 1.4. Since the gap is localized and likely repairable, conditional acceptance remains the right verdict.","tokens_in":25279,"tokens_out":32870,"duration_ms":342843,"concrete_test":"Re-derive Lemma 6.4 with loops allowed, classifying α components as arcs, common boundaries, and interior closed geodesics of a single pants, and verify the length lower bound and the inequality 3(|V3|+q) ≥ |∂V'_2| for a loop at a vertex in V3. As a minimal case, take χ=1, n=1 (one degree-3 vertex with a loop and one boundary vertex), compute H(X(G)) explicitly for the length-a pants with zero twist, and check whether h(X(G)) ≥ C min{h(G),1} holds with an explicit constant; if the minimal case violates it, Theorem 1.4's transfer is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 6.2's Lemma 6.4 is the only bridge from the graph expansion in Theorem 1.3 to the surface spectral gap in Theorem 1.4. Its proof fixes a multicurve α realizing H(X(G)) and decomposes α into β_i, geodesic segments inside pants with endpoints on boundary geodesics, and γ_j, common boundary geodesics of two pants on different sides V1 and V2. The length bound (6.11) and the counting inequality 3(|V3|+q) ≥ |∂V'_2| drive the Cheeger transfer. However, the graphs G(g) in Theorem 1.3 are obtained by adding loops to boundary vertices, so the surfaces X(G(g)) contain pairs of pants in which two boundary components are glued to each other. Such a pants can support a simple closed geodesic component of α that lies entirely in its interior and is neither a β_i segment nor a γ_j common boundary. The proof does not account for this third type, so neither (6.11) nor the graph-boundary comparison is justified for the looped graphs actually used. This is a proof gap in the central claim, not a disagreement with consensus: the transfer may be repairable by adding a loop term with a positive lower bound, but the constant and the inequality chain would change.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a configuration model F_{χ,n} in which χ interior vertices of degree 3 and n boundary vertices of degree 1 are paired by a good partition. It proves that when n=o(χ^{2/3}), a random graph is connected with high probability and has a Laplacian spectral gap bounded below by a uniform constant (Theorem 1.1). It also establishes an upper bound λ1(G) ≤ σ1(G) ≤ 16(g+1)/(3n) for connected graphs (Theorem 1.2). In the critical regime n ≍ g, the authors give an explicit combinatorial construction, the tree-planting method, of connected expander graphs with n(g)/g → θ and λ1 uniformly bounded below (Theorem 1.3). They then replace degree-3 vertices by pairs of pants and boundary vertices by cusps to obtain complete noncompact finite-area hyperbolic surfaces S_{g,n(g)} whose cusp count is linear in genus and whose spectrum has a uniform gap (Theorem 1.4). The main theorems are quantitative and the paper situates itself against recent work on random hyperbolic surfaces and spectral gaps.","tokens_in":25498,"tokens_out":22133,"duration_ms":241381,"significance":"If the proofs are completed, Theorem 1.4 is a striking result: it provides noncompact finite-area hyperbolic surfaces with genus g, cusp count n(g) ≍ g, and a uniform positive spectral gap, improving the arithmetic construction with n ≍ g^{2/3} to a linear cusp count. The paper also gives a probabilistic expander theorem for a boundary-vertex configuration model, with explicit constants, and a negative result when n grows faster than χ^{2/3}. The combinatorial tree-planting construction is explicit and quantitative, and the application to hyperbolic surfaces is natural. The paper does not rely on fitting data or circular derivations; it benchmarks against Bollobás cubic expanders and Shi–Yu’s Steklov comparison. However, the key bridge from graph expansion to surface expansion, Lemma 6.4, is not fully proved as written, and the pruning argument in Lemma 5.3 is also under-justified.","major_comments":[{"comment":"The decomposition of the realizing multicurve α into β_i segments and γ_j common boundary geodesics is not exhaustive. A component of α may be the simple closed geodesic obtained by gluing two boundary components of the same pair of pants to each other; this is exactly the loop case used in Theorem 1.3, since adding a loop at a degree-1 vertex turns it into a degree-3 vertex whose surface replacement has two boundary components identified. Such a component is neither a β_i segment with endpoints on boundary geodesics nor a γ_j common boundary of two distinct pants with v1∈V1 and v2∈V2. Therefore the length bound (6.11) and the subsequent Cheeger comparison are not justified for the surfaces X(G(g)) used in Theorem 1.4. The gap appears repairable by adding a separate term for self-glued boundary components, whose lengths are at least a, and by reworking the inequality chain; but as written the proof does not establish the transfer.","section":"Section 6.2, Lemma 6.4"},{"comment":"The pruning step in Lemma 5.3 asserts that after removing two edges inside T1, the chosen component T2 satisfies |∂V(T2)| ≤ |∂V(T1)|. This is not automatic: the two cut edges become new boundary edges, and the decrease in edges from T2 to T1' may not compensate. For example, if T1 is a path with a single external edge to T1', removing two edges to take a large subcomponent can produce a boundary of size 2, exceeding |∂V(T1)| = 1. Since this lemma underpins the upper bound in Theorem 1.2, the proof needs a more careful accounting of the boundary edges, for instance by choosing T2 to contain all vertices of T1 incident to the external edges, or by a different global argument.","section":"Section 5, Lemma 5.3"},{"comment":"The proof of Sublemma 4.1 asserts a numerical inequality of the form e^{μ/2} μ^{-μ} (1-μ)^{-(1-μ)/2} (1/1.9)^{(1-3μ)/6} < 0.999 for μ<0.02, but the verification is not shown. The argument depends on this constant being strictly less than 1, so a short explicit verification or an analytic monotonicity argument would make the proof complete. This is not a claim of an error, but the delicate constant regime deserves a written justification.","section":"Section 4, Sublemma 4.1"}],"minor_comments":[{"comment":"The title contains a spacing error ('exp anding') and the text has several typographical issues, including 'probabilty' in Section 3 and 'Mirazkhani' in Remark 6.3 (should be Mirzakhani).","section":"Title and Abstract"},{"comment":"In the final summation, the notation 'χ1≥χ2/3' is ambiguous; it should read χ1 ≥ χ^{2/3} to match Case-I. The current typesetting could confuse the reader about which case is being summed.","section":"Section 3, proof of Proposition 3.3"},{"comment":"The universal constant δ in the statement of Theorem 1.4 is never identified in the proof. The proof should explicitly set δ (for example, in terms of the constants C and θ from Lemma 6.4 and (6.12)) to make the claimed δ²/(1+θ)² gap concrete.","section":"Section 6.2, proof of Theorem 1.4"},{"comment":"The display defining a partition has a typo: 'P = (i1j1)(i1j2)...' should be 'P = (i1j1)(i2j2)...'.","section":"Section 2.1, Definition 2.1"}],"recommendation":"major_revision","confidential_remarks":"The main theorems are plausible and the paper is well positioned in the literature, but the surface transfer lemma (Lemma 6.4) has a genuine missing case that is directly used in Theorem 1.4. The pruning lemma for Theorem 1.2 also needs a more careful proof. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read this carefully, and the stress-test note is on target. The paper's core contribution is real: a configuration model with boundary vertices, a clean threshold n = o(χ^{2/3}) for connectivity and expansion, a Steklov upper bound that rules out expanders when g/n → 0, and a deterministic tree-planting construction achieving n ≍ g with a uniform spectral gap. The transfer to hyperbolic surfaces with cusp count linear in genus and uniformly positive spectral gaps is a natural capstone, improving the arithmetic n ≍ g^{2/3} relation. The combinatorial estimates in Sections 3–5 are mostly careful; the good-partition definition is a nice way to force boundary vertices to attach to interior vertices, and the counting in Proposition 4.3 is genuinely involved. The tree-planting argument is explicit and the tuning of n(g)/g → θ is done well. Using Bollobás's cubic expanders as an external ingredient is fine.\n\nThe soft spot is exactly in Lemma 6.4. The decomposition of the realizing multicurve α into arcs β_i inside pants and common boundary geodesics γ_j misses the case where two boundary components of the same pants are glued to each other—i.e., when the graph has loops, as in the construction for Theorem 1.3. Then there is a simple closed geodesic lying entirely in the interior of that pants, contributing neither a β_i nor a γ_j. The length bound (6.11) and the counting inequality 3(|V3|+q) ≥ |∂V2'| are therefore not justified for the looped graphs actually used. This is a genuine proof gap, not a cosmetic issue. But it is probably repairable: add the core geodesics of self-gluings as a third class, give them a lower length bound depending on a, and charge them to the V3 side or redefine the graph boundary. The constants and the inequality chain would change, but the overall strategy should survive. The other issues—terse pruning in Lemma 5.3, the isolation step in Lemma 6.1, and the unaddressed half-edge-level handling of loops in the configuration model—are minor in comparison. The authors should also state plainly how loops enter the random pairing model, since the construction deliberately creates them.\n\nWho is this for? Spectral geometers and graph theorists. It answers a natural question and the main theorems are plausible. I would send it to a serious referee with a conditional accept and a request to close the Lemma 6.4 gap. If that closes, the paper is a strong addition to the expander-surface literature.","headline":"Strong construction paper with a genuine but likely repairable gap in the surface transfer for looped graphs; worth refereeing.","tokens_in":26077,"tokens_out":4778,"would_cite":true,"duration_ms":52364,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C80","58J50"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs non-compact hyperbolic surfaces with linearly many cusps and a uniform positive spectral gap.","keywords":["Cheeger constant","configuration model","expander graphs","spectral gap","hyperbolic surfaces","Laplacian eigenvalues","Steklov eigenvalue","tree-planting method"],"falsifier":"Test Lemma 6.4 on the explicit surfaces $X(G(g))$ by computing the geometric Cheeger constant directly: since only finitely many simple closed multi-geodesics lie below any length bound, enumerate the separating multicurves of bounded length and compute $\\ell(\\alpha)/\\min\\{\\operatorname{Area}(A),\\operatorname{Area}(B)\\}$. If for some $\\theta>0$ these ratios are not bounded below by a positive universal constant (equivalently $h(X(G(g)))/\\min\\{h(G(g)),1\\}\\to0$), the transfer lemma fails and the spectral gap does not follow. The local target is the pair-of-pants arc $\\beta_i$ in Lemma 6.4 with both endpoints on the same boundary geodesic: its length must be bounded below by a constant depending only on the fixed boundary length $a$, and any sequence of such arcs with length tending to zero would break the proof.","tokens_in":25031,"feed_emoji":"📐","tokens_out":12375,"duration_ms":121139,"temperature":0.7,"pith_summary":"This paper proves that expansion can be engineered in a random configuration model with many degree-1 vertices, and then transferred to hyperbolic geometry. Its main theorem (Theorem 1.4) states that for every $\\theta>0$ there is a sequence of complete, non-compact, finite-area hyperbolic surfaces $S_{g,n(g)}$ whose cusp count satisfies $n(g)/g\\to\\theta$ and whose Laplace spectrum has no eigenvalues below a uniform positive constant. Such surfaces are called $\\delta$-expander surfaces, and the paper's construction makes $\\delta$ a universal constant divided by $(1+\\theta)$. If the theorem is right, linear cusp growth does not by itself force small eigenvalues, and the known cusp-count growth rate for uniform spectral gaps improves from the arithmetic $n\\asymp g^{2/3}$ to the natural $n\\asymp g$. The proof runs through graphs: first expanders in the configuration model $\\mathcal{F}_{\\chi,n}$, then a pants-decomposition lemma that exports the graph Cheeger constant to a surface Cheeger constant.","feed_headline":"Cusps can grow linearly without killing the spectral gap","feed_subtitle":"New construction yields, for any cusp-to-genus ratio, surfaces with a uniform Laplacian spectral gap.","key_machinery":"The carrying object is the configuration model $\\mathcal{F}_{\\chi,n}$, random good partitions of $3\\chi+n$ half-edges in which every paired edge contains at least one half-edge from one of the $\\chi$ interior vertices of degree 3, so each of the $n$ boundary vertices of degree 1 is attached to an interior vertex. Four mechanisms run on this object. The first is a counting estimate for disconnected subgraphs that proves connectivity with high probability when $n=o(\\chi^{2/3})$ and failure when $n\\gg\\chi^{2/3}$. The second is the $\\mu$-pair subgraph count: for any $\\mu<0.02$, the expected number of small connected subgraphs whose edge boundary is at most $\\mu$ times their size tends to zero, forcing $\\lambda_1(G)\\ge \\mu^2/18$ by Cheeger's inequality. The third is the Steklov comparison $\\lambda_1(G)\\le\\sigma_1(G)\\le 16(g+1)/(3n)$, obtained by removing $g+1$ edges to split the graph into two trees and using a test function supported on a balanced subsurface. The fourth is the tree-planting method: replace each edge of a cubic expander with the tree $T_k$, adding many degree-1 vertices without destroying expansion; choosing $k$ from $\\theta$ gives $n(g)/g\\to\\theta$. Lemma 6.4 then transfers $h(G)$ to $h(X(G))$ through the pants decomposition, and Proposition 6.2 plus Cheeger's inequality finish Theorem 1.4.","core_discovery":"The central claim is the simultaneous construction of expanding graph families and expanding surfaces in the critical regime where the number of degree-1 'boundary' vertices is proportional to the genus. On the graph side, Theorem 1.3 asserts that for any $\\theta>0$ there are connected graphs $G_g\\in\\mathcal{F}_{2g-2+n(g),n(g)}$ with $n(g)/g\\to\\theta$ and $\\liminf_{g\\to\\infty}\\lambda_1(G_g)\\ge 1/(648(\\theta+4)^2)$. On the surface side, Theorem 1.4 converts each such graph into a complete finite-area hyperbolic surface $X(G_g)$ of genus $g$ with $n(g)$ cusps by replacing every degree-3 vertex with a pair of pants of fixed boundary length and every degree-1 vertex with a cusp. Lemma 6.4 is the load-bearing transfer: the Cheeger constant of the surface is at least a universal constant times the minimum of the graph Cheeger constant and $1$, so Cheeger's inequality and the standard criterion that a Rayleigh quotient below $1/4$ yields a non-zero eigenvalue turn the graph spectral gap into a surface spectral gap $\\delta^2/(1+\\theta)^2$. The construction is explicit enough to answer the paper's own question: expanders can survive at $n(g)\\sim\\theta g$.","pith_inferences":["Beyond the paper: the pants-replacement transfer should apply to any uniformly expanding family of graphs with degree-1 vertices attached to degree-3 vertices, so other configuration-model or combinatorial expander constructions could generate many noncompact expander surfaces with different geometric data; the paper's explicit route is one template.","Beyond the paper: since the surfaces are built from explicit pants decompositions, one could compute Fenchel–Nielsen coordinates, systole, or diameter for the constructed sequence; the paper does not pursue these invariants.","Beyond the paper: the probabilistic model has a sharp connectivity threshold at $n\\asymp\\chi^{2/3}$. A natural next test is whether random graphs in that exact critical window already expand with positive probability, or whether planted constructions like the tree-planting method are genuinely necessary there; the paper proves only the generic failure for faster growth."],"forward_implications":["For every $\\theta>0$, the paper's Question has a negative answer: $n(g)\\sim\\theta g$ does not force $\\lambda_1\\to0$, and explicit connected graphs in $\\mathcal{F}_{2g-2+n(g),n(g)}$ satisfy $\\liminf\\lambda_1\\ge 1/(648(\\theta+4)^2)$.","There exist infinitely many finite-area noncompact hyperbolic surfaces with $n(g)/g\\to\\theta$ and no Laplace eigenvalues below a uniform positive constant, extending the regime of known uniform spectral gaps from arithmetic surfaces with $n\\asymp g^{2/3}$ to the linear regime $n\\asymp g$.","The condition $\\lim n(g)/g=\\infty$ is necessary for the known vanishing result (1.1): at exactly linear cusp growth, positive uniform gaps can still occur, so the rate $\\theta$ alone does not determine spectral collapse.","In the opposite direction, if $n(\\chi)$ grows faster than $\\chi^{2/3}$, almost every graph in $\\mathcal{F}_{\\chi,n}$ is disconnected, and if $g/n\\to0$, Theorem 1.2 gives $\\lambda_1(G)\\le\\sigma_1(G)\\le 16(g+1)/(3n)\\to0$, so those regimes cannot host expanders."],"supporting_citations":[{"why":"Supplies the 3-regular expander seed with Cheeger constant at least 2/11 that the tree-planting construction starts from, and the n=0 case of Theorem 1.1.","marker":"[6]"},{"why":"The tree case g=0 of the Steklov upper bound in Theorem 1.2 reduces to its result, and it supplies the Steklov test-function framework for graphs.","marker":"[24]"},{"why":"Proposition 4.7 compares the geometric Cheeger constant with the surface Cheeger constant; Lemma 6.4 uses it to turn multicurve cuts into h(X(G)).","marker":"[39]"},{"why":"Theorem XIII.1 guarantees that a Rayleigh quotient below 1/4 produces a non-zero first eigenvalue, which is the last step in Theorem 1.4.","marker":"[43]"},{"why":"Theorem 2.5 compares Steklov and Laplacian eigenvalues and is used in the proof of Theorem 1.2.","marker":"[48]"},{"why":"Provides the vanishing result (1.1) whose converse context justifies the necessity claim in Theorem 1.4's remark and the paper's discrete counterpart of its Theorem 3.","marker":"[51]"}],"fun_headline_variants":["Cusps grow linearly, spectral gap survives","Expanding surfaces with many cusps","Critical cusp count keeps Laplacian gap","Uniform spectral gap despite linear cusps","Graph expanders become cusped surfaces"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that replacing each interior vertex by a fixed pair of pants and each boundary vertex by a cusp preserves expansion: the smallest cut ratio of the resulting surface is at least a universal constant times the graph's cut ratio, with constants that do not degrade as the genus and cusp count grow.","fun_headline_variants_meta":{"raw":{"variants":["Cusps grow linearly, spectral gap survives","Expanding surfaces with many cusps","Critical cusp count keeps Laplacian gap","Uniform spectral gap despite linear cusps","Graph expanders become cusped surfaces"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000124,"raw_usage":{"total_tokens":1110,"prompt_tokens":959,"completion_tokens":151,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":575,"completion_tokens_details":{"reasoning_tokens":83}},"tokens_in":575,"tokens_out":151,"duration_ms":2322,"temperature":1.0,"reasoning_tokens":83,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:04:53.555216+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test Lemma 6.4 on the explicit surfaces $X(G(g))$ by computing the geometric Cheeger constant directly: since only finitely many simple closed multi-geodesics lie below any length bound, enumerate the separating multicurves of bounded length and compute $\\ell(\\alpha)/\\min\\{\\operatorname{Area}(A),\\operatorname{Area}(B)\\}$. If for some $\\theta>0$ these ratios are not bounded below by a positive universal constant (equivalently $h(X(G(g)))/\\min\\{h(G(g)),1\\}\\to0$), the transfer lemma fails and the spectral gap does not follow. The local target is the pair-of-pants arc $\\beta_i$ in Lemma 6.4 with both endpoints on the same boundary geodesic: its length must be bounded below by a constant depending only on the fixed boundary length $a$, and any sequence of such arcs with length tending to zero would break the proof.","supporting_citations":[{"cited_title":"The isoperimetric number of random regular graphs.European J","cited_arxiv_id":null,"evidence_quote":"Supplies the 3-regular expander seed with Cheeger constant at least 2/11 that the tree-planting construction starts from, and the n=0 case of Theorem 1.1."},{"cited_title":"Upper bounds for the Steklov eigenvalues on trees","cited_arxiv_id":null,"evidence_quote":"The tree case g=0 of the Steklov upper bound in Theorem 1.2 reduces to its result, and it supplies the Steklov test-function framework for graphs."},{"cited_title":"Growth of Weil-Petersson volumes and random hyperbolic surfaces of large genus","cited_arxiv_id":null,"evidence_quote":"Proposition 4.7 compares the geometric Cheeger constant with the surface Cheeger constant; Lemma 6.4 uses it to turn multicurve cuts into h(X(G))."},{"cited_title":"Methods of modern mathematical physics","cited_arxiv_id":null,"evidence_quote":"Theorem XIII.1 guarantees that a Rayleigh quotient below 1/4 produces a non-zero first eigenvalue, which is the last step in Theorem 1.4."},{"cited_title":"Comparison of Steklov eigenvalues and Laplacian eigenvalues on graphs","cited_arxiv_id":null,"evidence_quote":"Theorem 2.5 compares Steklov and Laplacian eigenvalues and is used in the proof of Theorem 1.2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the vanishing result (1.1) whose converse context justifies the necessity claim in Theorem 1.4's remark and the paper's discrete counterpart of its Theorem 3."}],"review_version":1}