{"id":"7e95d0fd-2bb4-465f-b421-17a62d1f27c5","arxiv_id":"2512.07753","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The cumulants of the β-ensemble admit an expansion in 1/N and 2/β whose coefficients are sums over suitably labelled maps, with the subleading order identified with maps on the projective plane via a new 2^{c(θ)-1}-to-1 bijection.","lead":"This paper derives a new large-N expansion of the cumulants of the β-ensemble — the eigenvalue distribution behind many random matrix models — using the Dumitriu-Edelman tridiagonal model, and expresses the coefficients in terms of labelled maps. It also constructs a new bijection connecting the first two orders of this expansion to maps on the projective plane, linking two previously separate combinatorial frameworks.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5.30's 'tent' claim for the leftmost geodesic is underproved and is the load-bearing step for the RP² bijection; a geodesic can have label valleys without creating new local minima.","rationale":"The reader's weakest_assumption identifies Lemma 5.30 as the load-bearing point, and I concur. The paper's central all-orders expansion (Theorem 1.2) is derived in Sections 2–4 using the bijection with suitably labelled maps; that part appears structurally sound and is checkable in small cases, with only typographical normalization issues. However, the advertised subleading-order interpretation in terms of maps on RP² (Corollary 1.3) rests entirely on the slit-opening/mirror-gluing construction of Section 5, whose geometric validity hinges on Lemma 5.30. The one-line proof of that lemma is not convincing: geodesics in 1-Lipschitz labelings can have non-monotone label sequences, and the local-minimum condition alone does not rule out valleys in the interior of the subpath. This is not a mere gap in exposition; if the tent claim is false, the uniqueness of the equilibrium loop and the 2^{c(θ)−1}-to-1 count fail, so the RP² interpretation is unsupported. The proposed computational test would either find a concrete counterexample or provide evidence that the lemma holds in the relevant regimes. Given that the main enumerative theorem does not rely on Section 5, the paper's core contribution is likely salvageable, and the correct verdict remains CONDITIONAL on resolving this lemma. I therefore keep the reader's verdict unchanged.","tokens_in":37863,"tokens_out":14106,"duration_ms":111342,"concrete_test":"Enumerate all maps in S2(θ) for small n (e.g., n/2 ≤ 4) and representative θ using the bijection of Proposition 3.20: generate all (γ,σ) with the appropriate number of vertices, construct the suitably labelled map, then compute the leftmost geodesic from the root minimum to the second minimum as defined in §5.5.2. Check whether the label sequence on the subpath g̃ equals min(ℓ(v°)+i, ℓ(v°)+#g̃−i). If any counterexample is found, Lemma 5.30 is false and Theorem 5.43 is unsupported. If no counterexample appears in an exhaustive search, it provides strong inductive evidence, though not a proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 5.30 asserts that along the subpath g̃ of the leftmost geodesic between v• and v°, the label sequence is exactly min(ℓ(v°)+i, ℓ(v°)+#g̃−i) — a symmetric tent with no interior dips below ℓ(v°). The proof is a single sentence: “as g̃ is a geodesic.” This is not sufficient. In a suitably labelled map, labels are 1-Lipschitz but a geodesic between two equal-label vertices can have valleys: an interior vertex with label below ℓ(v°) need not be a local minimum if it has an off-path neighbor of even lower label, so the map can still have exactly two local minima. Such a valley would break the tent symmetry. The tent property is what guarantees that the slit-opening construction in §5.4 produces a good loop and that the equilibrium loop in §5.5.3 is unique; these are used in Theorem 5.43 to obtain the 2^{c(θ)−1}-to-1 count. If Lemma 5.30 fails, the interpretation of the subleading coefficient as a count of maps on RP² (Corollary 1.3) collapses, even though the main all-orders expansion (Theorem 1.2) may remain correct because it does not depend on Section 5. The proof also implicitly assumes that labels along the prefix from v* (label 0) to v• increase monotonically, which is not obvious and is part of the same gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper derives an all-orders expansion in 1/N and β for the joint cumulants of power sums of the β-ensemble, using the Dumitriu–Edelman tridiagonal model. The coefficients are expressed as sums of elementary symmetric polynomials in distance labels of suitably labelled maps, via the Bouttier–Fusy–Guitter bijection. The first two orders of the expansion are identified with counts of planar maps and of maps on the projective plane, through a new many-to-one construction relating suitably labelled maps with two minima to maps on RP². The main expansion is obtained in Sections 2–4 from a Motzkin-path computation, a conjugation argument, and Faulhaber summation; the RP² interpretation is developed in Section 5.","tokens_in":38232,"tokens_out":28427,"duration_ms":259659,"significance":"If correct, the paper gives a new, parameter-free bridge between β-ensemble cumulants and labelled-map statistics, and a novel interpretation of the first subleading order in terms of maps on RP². The derivation is largely self-contained, uses no fitted quantities, and the leading-order recovery of planar map counts for all β>0 is a convincing check. The core expansion of Theorem 1.2 / Proposition 2.12 does not depend on the RP² section, so the main result is robust modulo the statement errors noted below. The RP² bijection is more fragile and needs a rigorous proof of its key lemmas.","major_comments":[{"comment":"The denominator in (2) is printed as N^{n/2+l-2}. Comparing with Proposition 2.12, where κ_l(n) = Σ ... N^{s+1} ..., and with eq. (15) and the leading-order computation in §4.2 (κ_l(n) ~ N^{n/2-l+2}), the denominator must be N^{n/2-l+2}. With the printed exponent, the left side is of order N^{2-2l} relative to the right side, so the statement is false as written. Please correct the exponent and check all displays for the same sign error.","section":"Theorem 1.2, Eq. (2)"},{"comment":"The subleading term inside the parenthesis is printed as (1/2^{1-l}) N (2/β - 1) #M_{1/2}(θ). This is of order N, while the leading term is O(1), so as written it would dominate the expansion. The proof at the end of §5.6 and the preceding computation give (1/(2^{l-1}N))(2/β - 1)#M_{1/2}(θ) = 2^{1-l}/N (2/β - 1)#M_{1/2}(θ). Please correct the statement: the factor N belongs in the denominator, not the numerator.","section":"Corollary 1.3"},{"comment":"The proof that the label sequence of g̃ equals min(ℓ(v°)+i, ℓ(v°)+#g̃-i) is a single sentence, 'as g̃ is a geodesic'. This is not a consequence of geodesy alone: a shortest path between two equal-label vertices can have label valleys or plateaus. Ruling these out requires using that every non-minimal vertex has a lower neighbour, that the map has exactly two local minima, and that v• is the unique other vertex of label ℓ(v°) on h̃; even then a length-comparison argument is needed. The same gap appears in Lemma 5.34, where the assertion that the maximum is attained only once or at two consecutive vertices is stated without proof. These lemmas feed into the slit-opening construction (§5.4) and the uniqueness of the equilibrium loop (§5.5.3), and hence into the 2^{c(θ)-1}-to-1 count of Theorem 5.43 and the RP² interpretation of Corollary 1.3. This is a load-bearing gap; please give a comple","section":"Lemma 5.30"}],"minor_comments":[{"comment":"The paper uses n/2 in summation limits and map edge counts but never states that n must be even. Please state this explicitly in Theorem 1.2 and Proposition 2.12.","section":"Theorem 1.2 / Section 2"},{"comment":"The last sentence says 'As g is a good loop', but g is a good path, not a loop. Please correct.","section":"Lemma 5.17"},{"comment":"The displayed identity 'Note that #g_{u'}^{-} + #g_{u'}^{-} = #g_1 = #g_2' should read #g_{u'}^{-} + #g_{u'}^{+} = #g_1 = #g_2.","section":"Proof of Lemma 5.37"},{"comment":"Typo: 'in is study' should be 'in his study'.","section":"Abstract"},{"comment":"The symbol \\tildeφ is used both for the new face cycle and as a factor in φ'; define it explicitly as the single cycle (g_{2l-1} ... g_{2l}) and avoid the ambiguous notation.","section":"Construction 5.16"},{"comment":"The recurrence 'det(z - T^1_∞)' appears to have a typo in the index; please make the notation for T^N_∞ consistent.","section":"Appendix A, Eq. (24)"}],"recommendation":"major_revision","confidential_remarks":"The main expansion appears sound, but the two displayed central statements contain sign/order typos that must be fixed, and the RP² bijection relies on an underproved lemma. I recommend major revision; the RP² section needs a complete proof of Lemma 5.30 (and the related uniqueness argument) before the subleading-order interpretation can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth reading for Theorem 1.2. The all-orders 1/N expansion of beta-ensemble cumulants in terms of distance statistics on suitably labelled maps is new, as far as I can tell, and the derivation through the Dumitriu-Edelman tridiagonal model is clean. The moment-to-Motzkin-path computation, the conjugacy step, and the BFG bijection all line up; the leading order recovering AAM14 is a good sanity check. I think that part of the paper is in good shape.\n\nThe soft spots are in the presentation and in Section 5. First, the headline statements have normalization/index typos: Theorem 1.2 writes N^{n/2+l-2} where the text and later usage need N^{n/2-l+2}, and Proposition 4.2 has an exponent that does not match the line after it. These are easy to fix, but right now the main theorem is not stated cleanly.\n\nThe bigger issue is Lemma 5.30. The lemma claims that along the leftmost geodesic between the two minima, the label sequence is exactly a symmetric tent min(ℓ(v°)+i, ℓ(v°)+#g̃−i). The proof is one sentence: \"as g̃ is a geodesic.\" That is not enough. A geodesic in a 1-Lipschitz labelling can dip below the tent without creating a new local minimum, because an interior vertex with a low label may have an even lower neighbor off the path. Such a valley would break the tent symmetry that the slit-opening construction and the uniqueness of the equilibrium loop depend on. So Theorem 5.43 and the RP² interpretation in Corollary 1.3 are not established as written. They may be true, but the geometric argument needs to be made carefully.\n\nI want to be clear: this is a load-bearing gap in Section 5, not a failure of the main expansion. The expansion in Theorem 1.2 does not rely on the RP² bijection. If a referee can get Lemma 5.30 (or a replacement) to work, the paper is very nice. As it stands, I would not trust the subleading-order interpretation until that lemma is fixed.\n\nWho should read it: anyone working on beta-ensembles and map enumeration, and people interested in bijections between orientable and non-orientable maps. It deserves a serious referee, not a desk reject. I'd send it to someone who knows both the BFG machinery and planar-map geodesics.","headline":"Strong new expansion of β-ensemble cumulants in labelled maps; the RP² bijection in Section 5 rests on a geometric lemma that is asserted, not proved.","tokens_in":38721,"tokens_out":2410,"would_cite":true,"duration_ms":21022,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C10","05A15","60B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes a 1/N and β expansion of the β-ensemble cumulants whose coefficients are counts of vertex-labelled maps, with planar maps at leading order and maps on the projective plane at the next order.","keywords":["β-ensemble","cumulants","1/N expansion","map enumeration","suitably labelled maps","projective plane","tridiagonal matrix model"],"falsifier":"Exhaustively enumerate all suitably labelled planar maps of small size (e.g. one face of length 4) and check whether every leftmost geodesic between the two minima satisfies Lemma 5.30 without interior label dips; a single counterexample would invalidate Theorem 5.43. Equivalently, compare #S2(θ) counted directly with (1+n/2−l) 2^{l−1} #M_{1/2}(θ) from the theorem for small θ and n.","tokens_in":37748,"feed_emoji":"🗺️","tokens_out":7370,"duration_ms":63817,"temperature":0.7,"pith_summary":"This paper tries to prove that the joint cumulants of power sums of β-ensemble eigenvalues admit a large-N expansion whose coefficients are naturally counted by vertex-labelled maps. The β-ensemble is the one-parameter family of eigenvalue distributions that includes the GOE, GUE, and GSE at β=1, 2, and 4. The expansion is derived from the tridiagonal matrix model for the β-ensemble, and at every order the coefficient is an elementary symmetric polynomial in the distances from the local-minimum vertices of a suitably labelled map. The leading order recovers the number of planar maps with a given face profile, and the next order is shown, through a new many-to-one mapping, to equal the number of maps on the projective plane. If the expansion is right, random matrix cumulants become a way to read off distance statistics of random planar maps, and map counts on orientable and non-orientable surfaces are unified at the level of the 1/N expansion.","feed_headline":"Beta-ensemble cumulants count vertex-labelled maps","feed_subtitle":"Leading order is the number of planar maps; the next order is the number of maps on the projective plane.","key_machinery":"The expansion's coefficients are carried by suitably labelled maps: maps whose vertices carry nonnegative labels with minimum 0 and |ℓ(v)−ℓ(w)|≤1 across every edge. Each non-local-minimum vertex contributes its distance from the local minima to an elementary symmetric polynomial; the sum over maps of this polynomial is the coefficient. The derivation uses the tridiagonal matrix model for the β-ensemble, whose moments become counts of Motzkin bridges with compatible permutations; cumulants impose transitivity of ⟨θ,σ⟩, selecting connected labelled hypermaps. A bijection from well-labelled hypermaps to suitably labelled maps turns the permutation data into vertex distances. For the subleading","core_discovery":"The paper's central claim is Theorem 1.2: for any partition n=(n_1,...,n_l) of n and any face profile θ, the joint cumulant κ_l(n) of the β-ensemble obeys (2/β)^{1-l} κ_l(n)/N^{n/2-l+2} = Σ_{v=0}^{n/2-l+1} N^{-v} Σ_{u+q+r=v} (2/β)^u (-1)^q B_r/(n/2-l+2-v) binom(r+n/2-l+1-v, r) ⟨e_q⟩_{θ,u+l-1}, where ⟨e_q⟩_{θ,p} is a sum over suitably labelled maps of the q-th elementary symmetric polynomial of the distances from the map's local minima. The proof passes through the tridiagonal model: moments of the matrix entries are read as Motzkin bridges decorated by permutations, cumulants select the connected pieces, and a known bijection between labelled hypermaps and suitably labelled maps converts the","pith_inferences":["Left implicit: the same slit-open/mirror-glue construction may apply at every order of the expansion, with higher-order coefficients counting maps on connected sums of projective planes; the paper only confirms the first two orders, so this is a conjecture.","If the expansion can be computed analytically from known asymptotic analyses of tridiagonal matrices, the distance-statistics interpretation would yield new results on the distribution of distances in random planar maps with fixed face degrees — a direction the paper mentions but does not develop.","The weight (2/β−1) in the subleading term suggests a natural interpolation: replacing β by a continuous parameter turns the expansion into a generating polynomial in non-orientability, analogous to b-deformations of map series studied elsewhere; this could give a new proof route for positivity of that polynomial."],"forward_implications":["For every β>0, the leading 1/N order of κ_l(n) is exactly the number of planar maps with face profile θ, recovering that planar map counts are universal in β.","The first subleading order is the number of maps on the projective plane weighted by (2/β−1), so the GOE/GUE/GSE distinction (β=1,2,4) emerges from the β-dependence of the orientable-versus-non-orientable weight.","At every order the coefficients are labelled-map distance statistics; an analytic handle on the cumulant expansion would give distance statistics of random planar maps with prescribed face profile.","The 2^{c(θ)−1}-to-1 mapping of Theorem 5.43 gives a bijective bridge between orientable labelled maps with two minima and pointed maps on RP², so counts of the latter can be computed from the former.","In the β→∞ limit, the expansion matches the known asymptotics of power sums of Hermite roots, recovering Catalan counts of planar trees and the Brownian-excursion area constant."],"fun_headline_variants":["β-ensemble cumulants count maps via new expansion","Map counting from β-ensemble cumulants","Cumulants of β-ensemble enumerate labelled maps","β-ensemble expansion reveals map counts","Dumitriu-Edelman model links β-ensemble to maps"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire projective-plane interpretation rests on Lemma 5.30: along the chosen segment of the leftmost geodesic between the two minima, the labels must form the exact symmetric tent min(ℓ(v°)+i, ℓ(v°)+#g̃−i), with no interior dip; the paper's proof is a single sentence and a dip would make the equilibrium loop fail to be good and break the 2^{c(θ)−1}-to-1 count.","fun_headline_variants_meta":{"raw":{"variants":["β-ensemble cumulants count maps via new expansion","Map counting from β-ensemble cumulants","Cumulants of β-ensemble enumerate labelled maps","β-ensemble expansion reveals map counts","Dumitriu-Edelman model links β-ensemble to maps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000391,"raw_usage":{"total_tokens":1888,"prompt_tokens":733,"completion_tokens":1155,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":1095}},"tokens_in":477,"tokens_out":1155,"duration_ms":10585,"temperature":1.0,"reasoning_tokens":1095,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T17:56:05.895303+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhaustively enumerate all suitably labelled planar maps of small size (e.g. one face of length 4) and check whether every leftmost geodesic between the two minima satisfies Lemma 5.30 without interior label dips; a single counterexample would invalidate Theorem 5.43. Equivalently, compare #S2(θ) counted directly with (1+n/2−l) 2^{l−1} #M_{1/2}(θ) from the theorem for small θ and n.","supporting_citations":[],"review_version":1}