{"id":"e4b31cff-4edd-444d-afb9-c3f49c7efe04","arxiv_id":"2607.08677","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"Large-N melonic tensor integrals are universal: they depend only on vertex counts, not on tensor rank D≥3 or the fine-grained combinatorics of the melonic invariants.","lead":"This paper proves that large-N integrals of melonic tensor invariants depend only on the number of vertices of the invariant, not on the tensor rank D or the combinatorial structure of the invariant. A generalist might read it because it identifies a universal simplification in a class of models used in random geometry and quantum gravity.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"Prop. 3.13's M_n⊂G_n proof assumes undoing a cycle-edge swap gains exactly D faces (Eq. 3.20), but Eq. 3.17 only justifies this for swaps between different connected components; for same-component swaps the gain could be zero, breaking the contradiction.","rationale":"The reader correctly identified the load-bearing concern: the M_n ⊂ G_n direction of Proposition 3.13, specifically Eq. (3.20). I agree this is the weakest link in the proof chain. The face-count change of D per swap-undoing is only justified for inter-component swaps (Eq. 3.17), not for intra-component swaps arising from thin-graph cycles. The concern is genuine and in the critical path to Theorem 4.2. However, the paper provides substantial computational evidence (thousands of integrals via feyntensor, Figure 1) supporting the result, and the paper is transparent about which parts are proven vs. conjectured (Theorem-Conjecture 4.4 is correctly flagged as unused in the main results). The gap is a proof-completeness issue in a combinatorial argument, not a sign of incorrectness. The CONDITIONAL verdict is appropriate: the result is likely correct but the proof of Proposition 3.13 needs the face-count argument for same-component swaps to be completed, either by showing the gain is strictly positive (even if not exactly D) or by a different argument entirely. The concrete test (n=3, p_i=1, D=3) is the minimal case where the concern could manifest and is computable by hand or with feyntensor.","tokens_in":21585,"tokens_out":9951,"duration_ms":652229,"concrete_test":"For n=3, p₁=p₂=p₃=1, D=3 (three 2-vertex melons), explicitly enumerate all Wick contractions of ⊔ᵢBᵢ that use 3 swaps forming a cycle in the thin graph. Compute #F for each. Compare with the tree configuration (2 swaps, #F = (D-1)(3-3+1)+1 = 3 by Lemma 3.12). If any cycle configuration has #F ≥ 3, the characterization M_n = G_n is incomplete and Proposition 3.13 needs revision. If all cycle configurations have #F < 3, the argument holds for this case and the concern is localized to larger n.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The entire proof chain culminating in Theorem 4.2 depends on Proposition 3.13 (M_n = G_n). In the M_n ⊂ G_n direction, the proof argues: if the thin graph is not a tree, undoing l swaps increases faces by l×D (Eq. 3.20), yielding a graph Γ with strictly more faces, contradicting maximality. This invokes the reverse of Eq. (3.17), which states that a swap between two *different connected components* decreases faces by exactly D. The -D arises because, for each color c, the two faces are necessarily in different graphs and thus distinct. However, for a swap corresponding to a cycle edge in the thin graph, the two vertex pairs are already in the *same* connected component (connected through other swaps). For each color c, the face containing (x,y') and the face containing (x',y) may already be the same face — in which case undoing the swap splits it (+1) rather than merging two distinct faces. If all D colors have coincident faces, the net gain is 0, not D, and no contradiction follows. If some colors coincide, the gain is between 0 and 2D, and the argument requires it to be strictly positive — which is not proven. If any cycle configuration achieves the same face count as the tree, M_n ⊋ G_n, Corollary 3.14's independence claim fails, and Theorem 4.2's universality is unproven. The gap is in the critical path: Prop 3.13 → Cor 3.14 → Thm 4.2 → Thm 5.1.","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"The paper proves a ","tokens_in":21956,"tokens_out":356,"duration_ms":484153,"significance":"The paper introduces a universal melonic measure and melonic polynomials, proving that large-N melonic tensor integrals are independent of both the tensor rank D≥3 and the fine-grained combinatorics of the melonic traces. The core proof chain (Lemmas 3.1–3.12, Proposition 3.13, Theorem 4.2) is structurally coherent. The uniqueness of maximal Wick contractions for single melonic graphs (Proposition 3.6) is convincingly argued by induction using dipole reduction. The application to quartic models (Corollary 5.2) leverages the Dartois-Eynard-Nguyen solution [NDE15] to derive a D-independent critical locus and two-point function, which is a concrete, falsifiable prediction. The computational verification of melonic polynomials via feyntensor [Pér26b] for thousands of integrals (Figure 1) is a notable strength.","major_comments":[],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":"The stress-test concern regarding Eq. (3.20) is, in my assessment, the central issue. The author should be given the opportunity to address it, as the fix may be straightforward (e.g., arguing that the tree structure of the thin graph prevents face coincidence, or that the swap topology forces distinct faces). If the author can close this gap, the paper makes a strong contribution. I recommend major revision rather than reject because the issue is local to the M_n ⊂ G_n direction and potentially fixable without restructuring the paper."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The headline: this paper proves that large-N melonic tensor integrals are independent of both the tensor rank D (for D≥3) and the fine-grained combinatorics of the melonic invariants—only vertex counts matter. That's a new and clean observation, and the single-trace case is solidly proved. The multi-trace case has a gap in a load-bearing position, and I think the stress-test concern about it is correct, or at least not obviously dismissible. Let me explain both sides briefly. The good: the proof chain for single melonic graphs is tight. Lemma 3.1 (dipole vertices must be Wick-contracted in maximal pairings) → Proposition 3.6 (unique maximal contraction) → Lemma 3.7 (face count) → Corollary 3.8 (integral equals 1) is clean and convincing. The construction of the universal melonic measure (Definition 4.1) and the reduction of arbitrary-rank integrals to rank-3 is a useful framework. The paper is also honest that Theorem-Conjecture 4.4 is conjectural and correctly notes it's not used downstream. The gap: Proposition 3.13 claims M_n = G_n, characterizing maximal Wick contractions of multiple melonic graphs. The M_n ⊂ G_n direction argues by contradiction: if the 'thin graph' isn't a tree, undoing l cycle-edge swaps increases faces by l×D (Eq. 3.20), contradicting maximality. But Eq. 3.17, which gives the -D face change per swap, is derived only for swaps between *different connected components*. For a cycle-edge swap, the two vertex pairs are already in the same component (connected through other swaps along the cycle). For each color c, the face containing (x,y') and the face containing (x',y) may already coincide, in which case undoing the swap splits one face (+1) rather than merging two distinct ones. The net face gain per color is in {-1,0,+1}, not necessarily +1. If enough colors have coincident faces, the total gain could be zero or negative, and the contradiction fails. This is in the critical path: Prop 3.13 → Cor 3.14 → Thm 4.2 → Thm 5.1. The result might still be true—the extensive numerical checks reported in Figure 1 support it—but the proof as written doesn't close the loop. The fix likely requires showing that cycle-edge swaps always gain a strictly positive number of faces (even if not exactly D), or a different characterization of maximality for the multi-trace case. This deserves a serious referee who can either patch the argument or identify a counterexample. I'd recommend conditional acceptance pending resolution of this gap.","headline":"Genuine universality result with a real gap in the multi-trace maximality argument","tokens_in":22424,"tokens_out":6059,"would_cite":false,"duration_ms":410321,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C30","82B20","60B99"],"pacs":[],"model":"glm-5.2","headline":"Melonic tensors forget their rank at large N","keywords":["melonic graphs","tensor models","large N limit","universality","Wick contraction","coloured graphs","tensor integrals","1/N expansion"],"falsifier":"Find a melonic graph B (or a collection B₁,...,Bₙ) and a Wick contraction Π that maximizes the number of faces but whose associated thin graph is not a tree, which would contradict Proposition 3.13 and break the chain leading to Theorem 4.2. Alternatively, find a case where the leading order of a melonic integral depends on D≥3 or on the combinatorial structure of the melonic graphs beyond vertex counts, which would directly contradict Theorem 4.2.","tokens_in":21857,"feed_emoji":"","tokens_out":1334,"duration_ms":266695,"temperature":0.7,"pith_summary":"The paper proves that integrals of products of melonic tensor invariants—a class of graph-encoded contractions of tensor indices built by recursive dipole insertion—have a leading large-N behavior that depends neither on the tensor rank D (for D≥3) nor on the combinatorial details of the specific melonic graphs involved, but only on how many vertices each graph has. The author constructs a universal measure on rank-3 tensors that reproduces, at large N, the cumulants of any melonic tensor model of any rank D≥3, provided the coupling constants and vertex counts match. The mechanism rests on showing that melonic graphs admit a unique maximal Wick contraction, and that the number of ways to combine these contractions across multiple graphs into a connected structure is itself a combinatorial invariant depending only on vertex counts. A family of melonic polynomials is introduced that conjecturally gives the exact leading-order values of these integrals.","feed_headline":"Melonic tensors forget their rank at large N","feed_subtitle":"Integrals of melonic tensor invariants depend only on vertex counts, not on rank D≥3 or graph combinatorics, collapsing all melonic models","key_machinery":"The argument turns on three load-bearing pieces: (1) Lemma 3.1, which shows that any maximal Wick contraction must pair the two vertices of any dipole; (2) Proposition 3.6, which uses this to prove uniqueness of the maximal Wick contraction π_max for any connected melonic graph; and (3) Proposition 3.13, which characterizes the set of maximal connected Wick contractions of multiple melonic graphs as the set G_n defined by two conditions—that the contraction is obtained by swaps from the product of individual maximal contractions (condition i), and that a contracted 'thin graph' is a tree (condition ii). The face-count formula in Lemma 3.12, which gives the maximum number of faces as (D−1)(P−","core_discovery":"The central result is that the rescaled connected integral of a product of melonic tensor invariants B₁(T)···Bₙ(T) of rank D, when multiplied by the appropriate power of N, converges as N→∞ to a limit that is identical for all D≥3 and depends on the graphs Bᵢ only through their vertex counts 2pᵢ. This is proved by showing that (a) each connected melonic graph has a unique Wick contraction maximizing the number of faces, (b) the number of maximal connected Wick contractions of a disjoint union of melonic graphs is a function only of the vertex counts (Proposition 3.13 and Corollary 3.14), and (c) a canonical rank-3 melonic graph u_p with 2p vertices can therefore replace any rank-D melonic Bᵢ","pith_inferences":["If the universal measure is rank-independent, then numerical simulations or bootstrap methods at finite N could extract rank-dependent corrections whose structure might reveal a systematic 1/N expansion organized by D, providing a finite-N fingerprint of the tensor rank.","The tree condition (ii) in the definition of G_n suggests a connection to combinatorial species or exponential generating functions: the count #G_n depending only on vertex counts hints that melonic multi-trace integrals might be expressible as coefficients of a single generating function indexed by n, analogous to the Harer-Zagier formula mentioned in the introduction.","The replacement of arbitrary melonic graphs by canonical ones u_p depending only on vertex count p suggests a decoupling between the 'topology' of the melonic graph (irrelevant at large N) and its 'size' (the only relevant parameter), which could extend to non-melonic graphs if a suitable analogue of unique maximal Wick contraction exists."],"forward_implications":["Any two melonic tensor models with the same coupling structure and vertex counts are indistinguishable at large N, regardless of their rank D≥3—this means the large-N phase diagram of a melonic tensor model is a universal object.","The critical locus of the quartic melonic tensor model is given by Σgᵢ + 1/4 = 0 for any D≥3 and any number of quartic interactions M≤D, extending the previously known equal-coupling case.","The melonic polynomials Mel_n(p₁,...,pₙ) = p₁···pₙ × (p₁+...+pₙ−1)_{n−2} conjecturally give closed-form leading-order values for all melonic multi-trace integrals, reducing a combinatorial enumeration to a Pochhammer-symbol expression.","Finite-N methods become essential for distinguishing tensor ranks, since all rank information is lost in the large-N limit for melonic observables."],"fun_headline_variants":["All melonic tensor models share one large-N limit regardless of rank","Melonic tensor integrals lose dependence on rank and trace combinatorics","Large-N melonic integrals reduce to functions of vertex counts only","Melonic tensor models are indistinguishable at large N","Rank and interaction detail drop out of melonic tensor integrals at large N"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The proof that every maximal Wick contraction of multiple melonic graphs satisfies the tree condition (condition ii in the definition of G_n) proceeds by contradiction: if the thin graph is not a tree, one undoes swaps until it becomes one, claiming each undo strictly increases the face count. The intermediate graphs produced by undoing individual swaps are assumed to remain valid Wick contractions with well-defined face structure, but this is stated rather than fully argued.","fun_headline_variants_meta":{"raw":{"variants":["All melonic tensor models share one large-N limit regardless of rank","Melonic tensor integrals lose dependence on rank and trace combinatorics","Large-N melonic integrals reduce to functions of vertex counts only","Melonic tensor models are indistinguishable at large N","Rank and interaction detail drop out of melonic tensor integrals at large N","Melonic polynomials unify large-N tensor integrals across all D≥3","A universal measure simplifies melonic tensor integration at large N","Two universality principles unify melonic tensor models at large N"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":1717,"prompt_tokens":828,"completion_tokens":889,"prompt_tokens_details":null},"tokens_in":828,"tokens_out":889,"duration_ms":46042,"temperature":1.0,"reasoning_tokens":761,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T03:11:06.626742+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"Find a melonic graph B (or a collection B₁,...,Bₙ) and a Wick contraction Π that maximizes the number of faces but whose associated thin graph is not a tree, which would contradict Proposition 3.13 and break the chain leading to Theorem 4.2. Alternatively, find a case where the leading order of a melonic integral depends on D≥3 or on the combinatorial structure of the melonic graphs beyond vertex counts, which would directly contradict Theorem 4.2.","supporting_citations":[],"review_version":1}