Pith.

Combinatorics · arXiv:2611.17197 v1 · submitted 12 Nov 2026 · 72 references

Matching statistics in regular hypergraphs and applications

Hyunwoo Lee

The paper says

“For fixed k ≥ 3, we determine the first sublinear correction to the logarithm of the matching partition function for a class of n-vertex d-regular k-uniform hypergraphs, uniformly for activities in each fixed compact interval of (0,∞).Our hypotheses include every regular linear …”

more of the abstract

… hypergraph of polynomial degree and a class of nonlinear hypergraphs described by higher codegrees and intersection patterns, which we introduce here. The weighted counting estimate implies an averaged version of Kahn's conjecture: under the uniform measure, the expected number of uncovered vertices is (1 + o(1))nd−1/k, with concentration and large deviations on this scale. The proof combines a novel variant of random greedy counting, based on short stages and regular completions, with a new entropy method that uses intersection patterns to control reconstruction after deleting matching edges. We highlight the following applications. (i) Partial transversals in Latin squares. Every Latin square of order n has (n/e2)nexp((3 + o(1))n2/3) partial transversals. Also, almost all partial transversals have (1 + o(1))n2/3 missing cells. (ii) Linear 3-uniform hypergraphs. There are (n/e2)n2/6exp((1/2 + o(1))n5/3) linear 3-uniform hypergraphs on [n], and almost all have n2/6−(1/6 + o(1))n5/3 edges. (iii) Matroids of fixed rank. For each fixed r ≥ 4, the number m(n,r) of rank-r matroids on [n] satisfies logm(n,r) = (nr−1)/(r!)(logn−r + 1) + (1/((r−1)!) + o(1))nr−1−1/r, refining the estimates of van der Hofstad, Pendavingh and van der Pol.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

The passes disagreed. Reader: holds up. Skeptic: holds up only with conditions. Referee: major revision.

A serious, technically deep paper that resolves the averaged form of Kahn's conjecture for a broad class of regular hypergraphs and gives the first sublinear correction to several important enumeration formulas. This line is wrong

4 major objections · 10 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 72 references say.

The paper says, at §4, Lemma 4.2, Theorem 4.3, Claim 4.5

The review did not quote this passage.

Pith says: Objection 1 of 4 · major

The proof of the lower bound (Proposition 6.2) depends on Theorem 4.1, which iterates the short-greedy-stage estimate (Theorem 4.3) through T stages. At each stage the residual hypergraph is completed to an exactly regular hypergraph K_P via Lemma 4.2. The Bennett–Bohman estimate (Theorem 4.3) requires that the maximum pair codegree L_2 of the hypergraph to which it is applied satisfy L_2 = o(D/(log N)^5).

more of objection 1

The paper controls the order of the completed hypergraph (Claim 4.5: log N_i = O(log^2 n)) and the degree D_i, but it does not bound the pair codegree of the completed hypergraph K_P. Lemma 4.2 only asserts Δ_j(K) ≤ max{Δ_j(F), 1} for j ≥ 2, but the completion adds new vertices and edges; a pair consisting of one old vertex and one new vertex can have codegree as large as D_{i+1} (the new vertex is adjacent to every copy of the old vertex in the auxiliary construction). If D_{i+1} is large (it is at least d p_i^{k-1} and can be as large as 2d), then L_2 for K_P could be Θ(D), violating the hypothesis L_2 = o(D/(log N)^5). The paper does not address this, and the iteration may therefore apply Theorem 4.3 outside its stated hypotheses. If the estimate fails for some completion, the geometric accumulation of errors in Claim 4.6 collapses and the lower bound R_λ ≥ na_λ – o(na_λ) is unsupported.

This review says the paper's mathematics is wrong. Pith did not check that against the paper's source.

The paper says, at §4.3, Claim 4.5

The review did not quote this passage.

Pith says: Objection 2 of 4 · major

Claim 4.5 bounds log N_i = O(log^2 n) for i ≤ T, but this bound is derived assuming the completion uses a_i = O(D_{i+1}^2) copies and w_i/a_i ≤ (k-1)N_i/2. However, Lemma 4.2 states that a = O_k(D^2) copies are used, but this constant may depend on k and the choice of prime Q.

more of objection 2

The bound a_i = O_k(D_{i+1}^2) is plausible, but the paper does not verify that the number of copies a_i is polynomial in D_{i+1} with a constant that does not depend on the specific residual hypergraph. Moreover, the bound w_i/a_i ≤ (k-1)S/D is given, but S is the total deficit, which could be as large as O(N_i D_i). The paper then asserts w_i/a_i = O_k(N_i ℓ sqrt(L_2^{(i)}/D_i)) (equation (4.7)), but this relies on the deficit being O(N_i sqrt(D_i L_2^{(i)} ℓ)). This deficit estimate itself depends on the degree estimate from Theorem 4.3, which is only guaranteed to hold with high probability for the original hypergraph, not for the completed one. The argument appears circular: to bound the deficit we need the degree estimate, but to apply the degree estimate we need to bound the codegree of the completed hypergraph, which is not done.

The paper says, at §5, Lemma 5.4

The review did not quote this passage.

Pith says: Objection 3 of 4 · major

Lemma 5.4 (random-order exposure) bounds H(P) for a random perfect matching P of H[U]. The proof uses the Radhakrishnan entropy argument and assumes that for each vertex v, the number of available edges through v given the history is at most 1 + g_v(P) y^{k-1} + sum b_{v,π}(P) y^{m(π)-1}.

more of objection 3

However, this bound counts all edges through v that are disjoint from previously revealed matching edges, but does not account for the fact that some of these edges may not be extendable to a perfect matching of the remaining vertices. The entropy bound (3.2) uses the support size, not the number of edges that can be completed to a perfect matching. The paper implicitly assumes that every edge through v that is disjoint from the revealed matching can be extended to a perfect matching of the remaining vertices, which is not justified. This is a known issue with the Radhakrishnan method for hypergraphs (see Luria's paper, which the paper cites, but Luria's application is for perfect matchings in complete hypergraphs where the condition holds by symmetry). Here the hypergraph is arbitrary, and the bound may overcount the support, potentially invalidating the entropy upper bound.

The paper says, at §6.2, Claim 6.2

The review did not quote this passage.

Pith says: Objection 4 of 4 · major

Claim 6.2 bounds H(W) - E[S_W]/(kd) using Shearer's inequality and a weighting argument. The application of Shearer's inequality (Lemma 3.3) requires that every vertex belongs to at least r members of the family A. Here the family consists of the edges of H plus a_v copies of each singleton {v}, where a_v = d - d_H(v).

more of objection 4

The paper states 'Add a_v copies of {v} to the edge family, so every coordinate belongs to exactly d members.' However, the a_v copies are not edges of the hypergraph; they are artificial. Shearer's inequality as stated applies to a family of subsets of the random vector's coordinates. The paper does not verify that the inequality holds when the family includes repeated singletons. Moreover, the entropy H(W∩e) for a singleton {v} is just H(1_{v∈W}), which is fine, but the inequality's factor 1/r requires that each coordinate appears in at least r members of the family. Here each vertex v appears in d_H(v) edges and a_v singleton copies, total d. So r = d. But the inequality then gives H(W) ≤ (1/d) [ sum_e H(W∩e) + sum_v a_v H(1_{v∈W}) ]. This is valid if Shearer's inequality holds for families with repetitions, which is true (the standard proof via subadditivity of entropy works). However, the paper then uses this to bound H(W) - E[S_W]/(kd). The derivation of (6.9) from (6.8) and Shearer's inequality needs to be checked carefully: S_W = k sum_e 1_{e∩W≠∅} - d|W| + sum_{v∉W} a_v. The expectation of the last term is sum_v a_v (1-p_v). The paper's expression (6.9) seems to omit this term or absorb it into the np/k term. The derivation is not fully explicit, and the cancellation in Claim 6.2 may rely on an algebraic manipulation that is not clearly justified.

10 minor objections

The paper says, at §1, Theorem 1.6

Pith says: minor objection 1 of 10

The statement of Theorem 1.6 includes the condition b_π(H) ≤ C d^{m(π)/k} n^{-ε} for π∈Π_k. The definition of Π_k excludes the partition (k) (which corresponds to the edge itself) and (1,1,...,1) (which corresponds to an edge meeting k different matching edges).

more of objection 1

The paper states that these are excluded, but it is not entirely clear why (1,1,...,1) is excluded; later in the proof, the term g_v(P) y^{k-1} accounts for edges meeting k blocks. The definition of b_π(H) should be clarified to explicitly state which partitions are included.

The paper says, at §4.1, Lemma 4.5

Pith says: minor objection 2 of 10

The proof of Lemma 4.5 uses Freedman's inequality to control the martingale M_t. The bound on the conditional variance is given as O_k(L_j L_{j+1}/N). However, the derivation of this bound is sketchy: it uses E[(ΔY_S(i))^2 | F_i] ≤ k L_{j+1} E[-ΔY_S(i) | F_i].

more of objection 2

This inequality relies on the fact that each removal of a vertex destroys at most L_{j+1} edges counted by Y_S, and each step removes at most k vertices. The factor k L_{j+1} is an upper bound on the maximum possible squared change, but the conditional variance bound should be more carefully justified. Also, the paper uses the bound E[-ΔY_S(i) | F_i] = O_k(L_j/N), which comes from the drift calculation. The product L_j L_{j+1}/N is then used. This seems plausible but the argument is compressed.

show 8 more

The paper says, at §4.2, Lemma 4.6

Pith says: minor objection 3 of 10

The statement of Lemma 4.6 includes the term O_{k,C_0}(ℓ^2 sqrt(L_2/D)). The proof uses the bound w_P/(a_P N) = O_{k,C_0}(ℓ sqrt(L_2/D)) from (4.7). However, the derivation of (4.7) relies on the deficit estimate S = O_{k,C_0}(N sqrt(D L_2) ℓ). This deficit estimate itself depends on the degree estimate from Theorem 4.3, which is only guaranteed for good trajectories.

more of objection 3

The lemma assumes that good trajectories have probability at least 1/2, but the bound on w_P/(a_P N) should be uniform over all P in P. The paper should argue that for every good prefix, the deficit is bounded as claimed, which it does implicitly, but the dependence on the constant C_1 in the definition of D' should be made explicit.

The paper says, at §5.1, Lemma 5.2

Pith says: minor objection 4 of 10

The integral estimate in Lemma 5.2 uses the bound (5.3) for ∫_0^1 log(1 + x y^{-q}) dy. For q=1, the bound is O(x log(e/x)). The paper then applies this to x = b_π(H)/d and q = k - m(π). For q=1, this requires m(π) = k-1.

more of objection 4

The paper states that b_π(H)/d = O(d^{-1/k} n^{-ε}) under the hypotheses, so the contribution is O(d^{-1/k} n^{-ε} log n). This is o(d^{-1/k}) because n^{-ε} log n → 0. However, the paper also uses the bound for q≥2, where the contribution is O((b_π(H)/d)^{1/q}). For q=2, this is O((d^{-1/k} n^{-ε})^{1/2}) = O(d^{-1/(2k)} n^{-ε/2}), which is o(d^{-1/k}) only if d^{-1/(2k)} = o(d^{-1/k}), i.e., d^{1/(2k)} → ∞, which holds. But the paper should check that the exponent 1/q is indeed ≤ 1/2 for q≥2, which is true. The argument is sound but the presentation could be clearer.

The paper says, at §6.1, Lemma 6.1

Pith says: minor objection 5 of 10

The proof of Lemma 6.1 uses Claim 6.1 to bound f(p, θ(1-p)) in terms of u = (log T)/k - g. The claim requires u > 0, which is ensured by the hypothesis g < (log T)/k. The bound then gives (1-p)g + (p/k) log T - f(p, θ(1-p)) ≥ g + θ(1-p) log(1 - T^{-1/k} e^g).

more of objection 5

The paper then drops the factor (1-p) in the logarithm term, using that the logarithm is negative and 1-p ≤ 1. This is valid but the inequality becomes weaker; the final bound is still sufficient. The step is correct but could be explained more clearly.

The paper says, at §7.1, Proposition 7.1

Pith says: minor objection 6 of 10

The proof of Proposition 7.1 uses the exponential moment estimate (7.1) to derive tail bounds via Markov's inequality. The choice of s depends on x, and the paper uses a common o(n(λd)^{-1/k}) error bound for all |s| ≤ s_0. This is valid because the error in (7.1) is uniform over the interval.

more of objection 6

However, the paper then uses this error bound in the tail estimate for both upper and lower tails, and then integrates to get the second moment bound. The integration step uses the tail bound (7.2) for x ≥ η n(λd)^{-1/k}. The derivation of (7.2) from Claim 7.1 uses the fact that o(n(λd)^{-1/k}) ≤ (cη/2) x for x ≥ η n(λd)^{-1/k} and sufficiently large n. This is correct because o(n(λd)^{-1/k}) is o(x) for such x. The argument is sound.

The paper says, at §8.2, Proof of Theorem 2.5

Pith says: minor objection 7 of 10

The proof for k = t+2 and k = t+3 relies on Claims 8.1 and 8.2, which verify the hypotheses of Theorem 1.6. The bounds on b_π(H) are derived by counting the number of candidate blocks B that intersect a given matching P in a specified pattern.

more of objection 7

The arguments are combinatorial and appear correct, but they are somewhat intricate and the paper does not provide full details for all cases. For example, for k = t+3 with h ≥ 2 parts of size t+1, the bound is O(n) (not O(n^2)). The justification is that after fixing the first two matching blocks and their intersections, the candidate B is determined. This relies on the fact that two (t+1)-subsets of B intersect in exactly t-1 points, and that the span of these two subsets is B. The paper should verify that this holds for all configurations, which it does briefly. The argument is plausible but the reader may need to fill in some steps.

The paper says, at §8.3, Proposition 8.1

Pith says: minor objection 8 of 10

The proof of Proposition 8.1 (maximum-degree upper bound) is given in Appendix C. The proof follows the same entropy argument as in Section 5, but with the simplification that the hypergraph is linear, so all b_π vanish. The paper uses the tangent bound and integral estimates to obtain the result.

more of objection 8

The proof seems correct, but it relies on Lemma 5.4, which itself has the issue mentioned in major comment 3. The validity of Proposition 8.1 is therefore contingent on fixing that issue.

The paper says, at §9, Concluding remarks

Pith says: minor objection 9 of 10

The concluding remarks mention several open questions and limitations. The paper acknowledges that the pair-codegree range may be extended, that the variance asymptotic is not determined, and that the pointwise vacancy question remains open for linear hypergraphs. This is appropriate and shows the authors are aware of the scope.

The paper says, at Statement of AI use

Pith says: minor objection 10 of 10

The paper states that ChatGPT carried out the technical calculations in Section 4 and contributed to the development of the entropy method. This is unusual for a serious journal submission. While the paper states that the author has independently verified all AI-generated results, the reliance on AI for core technical calculations raises concerns about reproducibility and the depth of verification.

more of objection 10

The referee cannot independently verify that the AI-generated calculations are correct without redoing them from scratch. This is a meta-concern that may affect the journal's policy.

What would settle it

For a concrete test, construct a small example: take a 3‑uniform linear hypergraph H on n vertices with degree d = n^{1/2}. Apply the completion Lemma 4.2 to a residual hypergraph after one greedy stage (N/2 vertices, degree ≈ 2^{-(k-1)}d). Compute the maximum pair codegree of the completed hypergraph K_P. If it exceeds D_{i+1} / (log N_i)^5, then the Bennett–Bohman hypothesis is violated and the iteration cannot be applied as claimed. The test should be repeated for several random prefixes to see whether the violation is typical.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up confidence moderate
  2. Skeptic partly agrees: says it holds up only with conditions 3 min 14 s · 55,658 tokens read, 848 written The proof that the greedy counting error is o(nd^{-1/k}) relies on the Bennett–Bohman estimate holding uniformly for every regular completion, but the completions can have super-polynomial order and the estimate's hypothesis requires L_2 = o(D/(log N)^5); the paper controls log N_i = O(log^2 n)…
  3. Editor wrote the verdict above 12 s · 56,627 tokens read, 740 written
  4. Referee major revision: 4 major objections, 10 minor 34 s · 56,123 tokens read, 3,844 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

Hyunwoo Lee has written a substantial paper. The main result—determining the first sublinear correction n(λd)−1/k to the matching partition function for a broad class of regular k-uniform hypergraphs—is a genuine advance. The paper also gives the first sublinear correction to several important enumeration formulas (Latin squares, Steiner systems, matroids) and establishes an averaged version of Kahn's conjecture. The proof combines a short-stage regularized greedy iteration with an intersection-pattern entropy method. The paper is well-structured, the hypotheses are explicit, and the author is honest about limitations (e.g., the pair-codegree range is not optimal, the variance asymptotic is not resolved). The applications are nontrivial and the verification of the structural conditions for each application is carried out in detail.

The soft spots are real but not fatal. The stress-test concern about the Bennett-Bohman estimate holding uniformly for all regular completions is the most serious. The paper controls log Ni = O(log2 n) and the degree Di, but does not bound the pair codegree of the completed hypergraph KP. Lemma 4.2 only asserts Δj(K) ≤ max{Δj(F), 1} for j ≥ 2, but the completion adds new vertices and edges; a pair consisting of one old vertex and one new vertex can have codegree as large as Di+1. If Di+1 is large (it can be as large as 2d), then L2 for KP could be Θ(D), violating the hypothesis L2 = o(D/(log N)5) of Theorem 4.3. The paper does not address this, and the iteration may therefore apply Theorem 4.3 outside its stated hypotheses. If the estimate fails for some completion, the geometric accumulation of errors in Claim 4.6 collapses and the lower bound Rλ ≥ na_λ – o(na_λ) is unsupported. This is a genuine gap that a referee would need to check carefully. The paper also states that ChatGPT carried out the technical calculations in Section 4; without independent verification, this is a concern for soundness, though the author states they have verified all AI-generated results.

Who is this paper for? Specialists in extremal combinatorics, probabilistic combinatorics, and design theory. It deserves a serious referee—the paper is important enough and the proof is detailed enough to warrant careful review. The stress-test concern is a genuine soft spot, but it is not obviously fatal; it may be fixable with additional argument. I would recommend sending it to peer review with a request to address the pair-codegree issue in the completion step.

Recommendation: Accept for peer review, with a note that the referee should carefully check Claims 4.4 and 4.5 and the uniform validity of the Bennett-Bohman estimate through all regular completions.

The strongest objection

The central lower bound (Proposition 6.2) depends on Theorem 4.1, which iterates the short‑greedy‑stage estimate (Theorem 4.3) through T stages. At each stage the residual hypergraph is completed to an exactly regular hypergraph K_P via Lemma 4.2. The Bennett–Bohman estimate (Theorem 4.3) requires that the maximum pair codegree L_2 of the hypergraph to which it is applied satisfy L_2 = o(D/(log N)^5). The paper controls the order of the completed hypergraph (Claim 4.5: log N_i = O(log^2 n)) and the degree D_i, but it does not bound the pair codegree of the completed hypergraph K_P. Lemma 4.2 only asserts Δ_j(K) ≤ max{Δ_j(F), 1} for j ≥ 2, but the completion adds new vertices and edges; a pair consisting of one old vertex and one new vertex can have codegree as large as D_{i+1} (the new vertex is adjacent to every copy of the old vertex in the auxiliary construction). If D_{i+1} is large (it is at least d p_i^{k-1} and can be as large as 2d), then L_2 for K_P could be Θ(D), violating the hypothesis L_2 = o(D/(log N)^5). The paper does not address this, and the iteration may therefore apply Theorem 4.3 outside its stated hypotheses. If the estimate fails for some completion, the geometric accumulation of errors in Claim 4.6 collapses and the lower bound R_λ ≥ na_λ – o(na_λ) is unsupported.

The reader identified the weakest assumption as the uniform validity of the Bennett‑Bohman estimate through all regular completions, but did not pinpoint the specific gap: the pair codegree of the completed hypergraph is not bounded. The paper controls the order and degree of the completions but does not bound the pair codegree of the new vertices, which can be as large as the degree itself. This is a concrete technical gap that could invalidate the iteration. The test is feasible and would settle whether the gap is real. The verdict should be CONDITIONAL because the central claim is plausible but depends on a missing verification; if the test shows the hypothesis is satisfied, the paper would be correct. The reader's verdict was ACCEPT with MODERATE confidence; this concern raises the correctness risk to HIGH unless resolved.

The referee's report

If correct, the result provides the first determination of the sublinear correction to the matching partition function for regular hypergraphs, resolving an averaged version of Kahn's conjecture and giving refined counts for several important combinatorial objects. The techniques introduced (short greedy stages with regular completions, intersection pattern coefficients) are likely to be influential.

Pith's reading of the paper (a machine's summary, not the authors')

This paper determines the first sublinear correction to the logarithm of the matching partition function for a broad class of regular k-uniform hypergraphs, uniformly for activities in each fixed compact interval of (0,∞). The result gives the sharp mean, concentration, and large deviations for the number of uncovered vertices, and refines counts for partial Latin squares, Steiner systems, subspace designs, and matroids of fixed rank.

What carries the argument. The proof combines a novel variant of random greedy counting, based on short stages and regular completions, with a new entropy method that uses intersection patterns to control reconstruction after deleting matching edges.

Core claim. For fixed k≥3 and ε>0, every simple n-vertex d-regular k-uniform hypergraph with d ≥ n^ε, Γ(H) ≤ C $d^{{-1/k}}$ $n^{{-ε}}$, and b_π(H) ≤ C $d^{{m(π)/k}}$ $n^{{-ε}}$ for all π satisfies log Z_H(λ) = (n/k)(log(λ d) - k + 1) + (1+o(1)) n (λ d)^{-1/k}, uniformly for λ in each compact subinterval of (0,∞). This determines the first sublinear correction to the matching partition function and, via activity shifts, gives the mean, concentration, and large deviations of the number of uncovered vertices.

The premise it rests on. The proof depends on the Bennett-Bohman fixed-fraction degree estimate remaining valid with high probability for every regular completion in the iterative greedy counting, not just for the original hypergraph.

If the paper is right (Pith's inferences, not the authors')

The counting estimate implies an averaged version of Kahn's conjecture: under the uniform measure, the expected number of uncovered vertices is (1+o(1)) n d^{-1/k}, with concentration and large deviations on this scale.

Every Latin square of order n has (n/e^2)^n exp((3+o(1)) n^{2/3}) partial transversals, and almost all partial transversals have (1+o(1)) n^{2/3} missing cells.

There are (n/e^2)^{n^2/6} exp((1/2+o(1)) n^{5/3}) linear 3-uniform hypergraphs on [n], and almost all have n^2/6 - (1/6+o(1)) n^{5/3} edges.

For each fixed r ≥ 4, the number m(n,r) of rank-r matroids on [n] satisfies log m(n,r) = (n^{r-1}/r!)(log n - r + 1) + (1/((r-1)!)+o(1)) n^{r-1-1/r}, refining prior estimates.

The method suggests that the first sublinear correction to the partition function is universal for regular hypergraphs with sufficiently small higher codegrees, not just those satisfying the specific intersection pattern bounds.

The technique of short greedy stages with regular completions could be adapted to other counting problems where the natural process loses regularity too quickly.

The entropy reconstruction bound using intersection patterns provides a general tool for converting deletion randomness into counting gains, potentially applicable to other hard-core models on hypergraphs.

The refined matroid count opens the possibility of determining the next correction for rank-3 matroids, where four-point hyperplanes contribute at order n^2.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

For fixed k ≥ 3, we determine the first sublinear correction to the logarithm of the matching partition function for a class of n-vertex d-regular k-uniform hypergraphs, uniformly for activities in each fixed compact interval of (0,∞).Our hypotheses include every regular linear hypergraph of polynomial degree and a class of nonlinear hypergraphs described by higher codegrees and intersection patterns, which we introduce here. The weighted counting estimate implies an averaged version of Kahn's conjecture: under the uniform measure, the expected number of uncovered vertices is (1 + o(1))nd−1/k, with concentration and large deviations on this scale. The proof combines a novel variant of random greedy counting, based on short stages and regular completions, with a new entropy method that uses intersection patterns to control reconstruction after deleting matching edges. We highlight the following applications. (i) Partial transversals in Latin squares. Every Latin square of order n has (n/e2)nexp((3 + o(1))n2/3) partial transversals. Also, almost all partial transversals have (1 + o(1))n2/3 missing cells. (ii) Linear 3-uniform hypergraphs. There are (n/e2)n2/6exp((1/2 + o(1))n5/3) linear 3-uniform hypergraphs on [n], and almost all have n2/6−(1/6 + o(1))n5/3 edges. (iii) Matroids of fixed rank. For each fixed r ≥ 4, the number m(n,r) of rank-r matroids on [n] satisfies logm(n,r) = (nr−1)/(r!)(logn−r + 1) + (1/((r−1)!) + o(1))nr−1−1/r, refining the estimates of van der Hofstad, Pendavingh and van der Pol.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-14

The authors' main claim, as the classifier picked it out 0.98

“For fixed $k\ge3$, we determine the first sublinear correction to the logarithm of the matching partition function for a class of $n$-vertex $d$-regular $k$-uniform hypergraphs, uniformly for activities in each fixed compact interval of $(0,\infty)$.Our hypotheses include every regular linear hypergraph of polynomial degree and a class of nonlinear hypergraphs described by higher codegrees and intersection patterns, which we introduce here.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.17 · states limitation 0.30 · human participants 0.01 · controlled comparison 0.02 · randomized assignment 0.03 · states quantitative result 0.98 · compares with prior results 0.77 · mentions code or data availability 0.01

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
For fixed $k\ge3$, we determine the first sublinear correction to the logarithm of the matching partition function for a class of $n$-vertex $d$-regular $k$-uniform hypergraphs, uniformly for activities in each fixed compact interval of $(0,\infty)$.Our hypotheses include every regular linear hypergraph of polynomial degree and a class of nonlinear hypergraphs described by higher codegrees and intersection patterns, which we introduce here. contribution 0.70 0.96 0.74 asserted 1.00
The weighted counting estimate implies an averaged version of Kahn's conjecture: under the uniform measure, the expected number of uncovered vertices is $(1+o(1))nd^{-1/k}$, with concentration and large deviations on this scale. result 0.53 0.85 0.31 asserted 0.98
The proof combines a novel variant of random greedy counting, based on short stages and regular completions, with a new entropy method that uses intersection patterns to control reconstruction after deleting matching edges. method 0.99 0.94 0.04 asserted 1.00
We highlight the following applications. contribution 0.79 0.43 0.01 unclear 0.75
(i) Partial transversals in Latin squares. contribution 0.58 0.49 0.02 unclear 0.86
Every Latin square of order $n$ has $(n/e^2)^n\exp((3+o(1))n^{2/3})$ partial transversals. result 0.99 0.85 0.10 asserted 1.00
Also, almost all partial transversals have $(1+o(1))n^{2/3}$ missing cells. result 1.00 0.85 0.05 asserted 1.00
(ii) Linear $3$-uniform hypergraphs. contribution 0.55 0.67 0.03 asserted 0.72
There are $(n/e^2)^{n^2/6}\exp((1/2+o(1))n^{5/3})$ linear $3$-uniform hypergraphs on $[n]$, and almost all have $n^2/6-(1/6+o(1))n^{5/3}$ edges. result 0.99 0.88 0.05 asserted 1.00
(iii) Matroids of fixed rank. contribution 0.69 0.43 0.02 unclear 0.87
For each fixed $r\ge4$, the number $m(n,r)$ of rank-$r$ matroids on $[n]$ satisfies $\log m(n,r)=\frac{n^{r-1}}{r!}(\log n-r+1)+\left(\frac1{(r-1)!}+o(1)\right)n^{r-1-1/r}$, refining the estimates of van der Hofstad, Pendavingh and van der Pol. result 0.97 0.89 0.07 asserted 1.00

What kind of paper the classifier reports this is: theoretical result 1.00 · evidence from proof or derivation 1.00

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-14: 1 request, 8,574 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (5)
Figure 1 of the paper
Figure 1. Regular completion for one degree deficit at v, illustrated with k = 3 and D = Q = 2. The four edges ei of the auxiliary K2,2 are extended to ei∪ {vi}, where vi lies in the ith copy of F. Each vi gains one incident edge, and every new vertex retains degree two. A fresh auxiliary graph is used for each remaining deficit. Lemma 4.2 (Regular completion). Let F be a simple k-uniform hypergraph on s vertices with maximum degree at most the positive integer D, and let S = P v∈V (F) (D − dF (v)). There is a simple D-regular k-uniform hypergraph K containing a = Ok(D2 ) disjoint copies of F…
Figure 2 of the paper
Figure 2. The two reconstruction problems. The matching J does not dis￾tinguish the original vacancy W = {7} from the six vertices in S made vacant by deletion. Once W is specified, reconstruction amounts to choosing a perfect matching of H[S]; two possible choices are shown. The retained edge on a, b, c is unchanged. matching edges of P alternative edge e (a) π = (1, 1, 1, 1, 1, 1) P1 P2 P3 P4 P5 P6 m = 6 Pr(e ⊆ S) = θ 6 (b) π = (2, 2, 2) P1 P2 P3 m = 3 Pr(e ⊆ S) = θ 3 (c) π = (4, 1, 1) P1 P2 P3 m = 3 Pr(e ⊆ S) = θ 3 Each source edge is deleted independently with probability θ; S is the…
Figure 3 of the paper
Figure 3. Three intersection patterns in a 6-uniform hypergraph. Each blue block is an edge of the source matching P; the red vertices form an alternative edge e. If S is the union of the independently deleted source edges, then P(e ⊆ S) = θ m(π) . The last two patterns meet the same number of source edges but have different intersection sizes
Figure 4 of the paper
Figure 4. The main contribution in the perfect-source calculation, normalized by naλ, as a function of θ/aλ. Its maximum occurs at θ = aλ and gives the correction naλ. The error terms are omitted. Lemma 6.1 (Thinning). Under the hypotheses of Theorem 1.6, let λ ∈ I, T > 1 and 0 < θ < 1, and define g = h(θ) − θ log(λdθk−1 ) + (k − 1)θ k . If g < (log T)/k, then Rλ ≥ RλT + n h g + θ log(1 − T −1/ke g ) i − o(naλ). The error is uniform over T and θ satisfying these conditions. Proof of Lemma 6.1. Let M have the Gibbs measure at λT, and obtain J by deleting each edge independently with…
Figure 5 of the paper
Figure 5. Deleting the shadow of a large hyperplane for r = 4 and n = 7. The hyperplane L = {1, 2, 3, 4, 5} removes its ten triples from H. The two displayed edges of F are the shadows of {1, 2, 6, 7} and {3, 4, 6, 7}; triple labels abbreviate sets. The residual degrees dF (126) = 1 and dF (167) = 4 show why a maximum￾degree bound is needed. For the residual hypergraph F on N − w vertices, with w as in (8.6), Proposition 8.1 gives log |M(F)| ≤ N − w r (log D − r + 1) + ND−1/r + Or

Figures are the authors', reproduced from the paper.

References (72)
  1. [1] N. Alon, J. H. Kim, and J. Spencer,Nearly perfect matchings in regular simple hypergraphs, Israel J. Math.100(1997), 171–187 on Pith
  2. [2] A. S. Asratian and N. N. Kuzjurin,On the number of nearly perfect matchings in almost regular uniform hypergraphs, Discrete Math.207(1999), 1–8 on Pith
  3. [3] A. S. Asratian and N. N. Kuzjurin,On the number of partial Steiner systems, J. Combin. Des.8(2000), no. 5, 347–352 on Pith
  4. [4] N. Bansal, R. Pendavingh, and J. van der Pol,On the number of matroids, Combinatorica 35(2015), no. 3, 253–277 on Pith
  5. [5] P. Bennett and T. Bohman,A natural barrier in random greedy hypergraph matching, Combin. Probab. Comput.28(2019), no. 6, 816–825 on Pith
  6. [6] P. Bennett and A. Frieze,Binomial random matroids, Preprint, arXiv:2603.10293, 2026 on PithPith's reviewarXiv
  7. [7] T. Bohman, A. Frieze, and E. Lubetzky,Random triangle removal, Adv. Math.280(2015), 379–438 on Pith
  8. [8] S. Boucheron, G. Lugosi, and P. Massart,Concentration inequalities: A nonasymptotic theory of independence, Oxford University Press, 2013 on Pith
  9. [9] C. Bowtell and P. Keevash,Then-queens problem, Preprint, arXiv:2109.08083, 2021 on PithPith's reviewarXiv
  10. [10] M. Braun, T. Etzion, P. R. J. ¨Osterg ˚ ard, A. Vardy, and A. Wassermann,Existence of q-analogs of Steiner systems, Forum Math. Pi4(2016), e7 on Pith
  11. [11] T. Carroll, D. Galvin, and P. Tetali,Matchings and independent sets of a fixed size in regular graphs, J. Combin. Theory Ser. A116(2009), no. 7, 1219–1227 on Pith
  12. [12] F. R. K. Chung, R. L. Graham, P. Frankl, and J. B. Shearer,Some intersection theorems for ordered sets and graphs, J. Combin. Theory Ser. A43(1986), 23–37 on Pith
  13. [13] E. Cohen, W. Perkins, M. Sarantis, and P. Tetali,On the number of independent sets in uniform, regular, linear hypergraphs, European J. Combin.99(2022), 103401 on Pith
  14. [14] T. M. Cover and J. A. Thomas,Elements of information theory, 2nd ed., John Wiley & Sons, 2006 on Pith
  15. [15] P. Csikv´ ari,Lower matching conjecture, and a new proof of Schrijver’s and Gurvits’s the- orems, J. Eur. Math. Soc.19(2017), no. 6, 1811–1844 on Pith
  16. [16] T. Dai, A. Divoux, and T. Kelly,Entropy bounds for perfect matchings in bipartite hyper- graphs, Electron. J. Combin.33(2026), no. 2, Paper No. P2.20 on Pith
  17. [17] E. Davies, M. Jenssen, W. Perkins, and B. Roberts,Independent sets, matchings, and occupancy fractions, J. Lond. Math. Soc.96(2017), no. 1, 47–66 on Pith
  18. [18] E. Davies, M. Jenssen, W. Perkins, and B. Roberts,Tight bounds on the coefficients of partition functions via stability, J. Combin. Theory Ser. A160(2018), 1–30 on Pith
  19. [19] M. Delcourt and L. Postle,Finding an almost perfect matching in a hypergraph avoiding forbidden submatchings, J. Lond. Math. Soc.114(2026), no. 2, e70660 on Pith
  20. [20] P. Erd˝ os and H. Hanani,On a limit theorem in combinatorical analysis, Publ. Math. De- brecen10(1963), 10–13 on Pith
show 52 more
  1. [21] T. Etzion, S. Kurz, K. Otal, and F. ¨Ozbudak,Subspace packings: constructions and bounds, Des. Codes Cryptogr.88(2020), no. 9, 1781–1810 on Pith
  2. [22] R. M. Falc´ on and R. J. Stones,Enumerating partial Latin rectangles, Electron. J. Combin. 27(2020), no. 2, Paper No. P2.47. 60 HYUNWOO LEE on Pith
  3. [23] P. Frankl and V. R¨ odl,Near perfect coverings in graphs and hypergraphs, European J. Combin.6(1985), 317–326 on Pith
  4. [24] D. A. Freedman,On tail probabilities for martingales, Ann. Probab.3(1975), no. 1, 100– 118 on Pith
  5. [25] M. Ghandehari, H. Hatami, and E. S. Mahmoodian,On the size of the minimum critical set of a Latin square, Discrete Math.293(2005), no. 1–3, 121–127 on Pith
  6. [26] R. Glebov and Z. Luria,On the maximum number of Latin transversals, J. Combin. Theory Ser. A141(2016), 136–146 on Pith
  7. [27] S. Glock, F. Joos, J. Kim, M. K¨ uhn, and L. Lichev,Conflict-free hypergraph matchings, J. Lond. Math. Soc.109(2024), no. 5, e12899 on Pith
  8. [28] S. Glock, D. K¨ uhn, A. Lo, and D. Osthus,The existence of designs via iterative absorption: hypergraphF-designs for arbitraryF, Mem. Amer. Math. Soc.284(2023), no. 1406, v+131 on Pith
  9. [29] S. Gould and T. Kelly,Advancing the R¨ odl nibble: New bounds on matchings and the list chromatic index of hypergraphs, Preprint, arXiv:2511.11375, 2025 on PitharXiv
  10. [30] A. Gupta,A rate for the average vacancy of uniformly random matchings in linear hyper- graphs, Preprint, arXiv:2609.26133, 2026 on PithPith's reviewarXiv
  11. [31] O. J. Heilmann and E. H. Lieb,Theory of monomer-dimer systems, Comm. Math. Phys. 25(1972), 190–232 on Pith
  12. [32] L. Ilinca and J. Kahn,Asymptotics of the upper matching conjecture, J. Combin. Theory Ser. A120(2013), no. 5, 976–983 on Pith
  13. [33] B. Jany, R. Jurrius, and R. Pendavingh,Countingq-matroids, Preprint, arXiv:2606.20348, 2026 on PithPith's reviewarXiv
  14. [34] J. Kahn,Asymptotics of hypergraph matching, covering and coloring problems, Proceed- ings of the International Congress of Mathematicians, Vol. 1, 2 (Z¨ urich, 1994) (Basel), Birkh¨ auser, 1995, pp. 1353–1362 on Pith
  15. [35] J. Kahn,On some hypergraph problems of Paul Erd˝ os and the asymptotics of matchings, covers and colorings, The Mathematics of Paul Erd˝ os, I, Algorithms Combin., vol. 13, Springer, Berlin, 1997, pp. 345–371 on Pith
  16. [36] J. Kahn and J. H. Kim,Random matchings in regular graphs, Combinatorica18(1998), no. 2, 201–226 on Pith
  17. [37] D. Y. Kang, D. K¨ uhn, A. Methuku, and D. Osthus,New bounds on the size of nearly perfect matchings in almost regular hypergraphs, J. Lond. Math. Soc.108(2023), no. 5, 1701–1746 on Pith
  18. [38] P. Keevash,The existence of designs, Preprint, revised 2024, arXiv:1401.3665, 2014 on PithPith's reviewarXiv
  19. [39] P. Keevash,Counting designs, J. Eur. Math. Soc.20(2018), no. 4, 903–927 on Pith
  20. [40] P. Keevash,The existence of designs II, Preprint, arXiv:1802.05900, 2018 on PithPith's reviewarXiv
  21. [41] P. Keevash, A. Pokrovskiy, B. Sudakov, and L. Yepremyan,New bounds for Ryser’s con- jecture and related problems, Trans. Amer. Math. Soc. Ser. B9(2022), 288–321 on Pith
  22. [42] P. Keevash, A. Sah, and M. Sawhney,The existence of subspace designs, Proc. Lond. Math. Soc.131(2025), no. 1, e70071 on Pith
  23. [43] D. E. Knuth,The asymptotic number of geometries, J. Combin. Theory Ser. A16(1974), no. 3, 398–400 on Pith
  24. [44] M. Kwan,Almost all Steiner triple systems have perfect matchings, Proc. Lond. Math. Soc. 121(2020), no. 6, 1468–1495. MATCHING STATISTICS IN REGULAR HYPERGRAPHS AND APPLICATIONS 61 on Pith
  25. [45] M. Kwan, R. Safavi, and Y. Wang,Counting perfect matchings in Dirac hypergraphs, Com- binatorica46(2026), Paper No. 5, 32 on Pith
  26. [46] M. Kwan, A. Sah, and M. Sawhney,Enumerating matroids and linear spaces, C. R. Math. Acad. Sci. Paris361(2023), 565–575 on Pith
  27. [47] H. Lee,Random matchings in linear hypergraphs, Forum Math. Sigma14(2026), e117 on Pith
  28. [48] M. Lelarge,Counting matchings in irregular bipartite graphs and random lifts, Proceedings of the Twenty-Eighth Annual ACM–SIAM Symposium on Discrete Algorithms, SIAM, 2017, pp. 2230–2237 on Pith
  29. [49] N. Linial and Z. Luria,An upper bound on the number of Steiner triple systems, Random Structures Algorithms43(2013), no. 4, 399–406 on Pith
  30. [50] N. Linial and Z. Luria,An upper bound on the number of high-dimensional permutations, Combinatorica34(2014), no. 4, 471–486 on Pith
  31. [51] Z. Luria,New bounds on the number ofn-queens configurations, Preprint, arXiv:1705.05225, 2017 on PithPith's reviewarXiv
  32. [52] M. Madiman and P. Tetali,Information inequalities for joint distributions, with interpre- tations and applications, IEEE Trans. Inform. Theory56(2010), no. 6, 2699–2713 on Pith
  33. [53] D. Mayhew, M. Newman, D. Welsh, and G. Whittle,On the asymptotic proportion of connected matroids, European J. Combin.32(2011), no. 6, 882–890 on Pith
  34. [54] B. D. McKay and F. Tian,Asymptotic enumeration of linear hypergraphs with given number of vertices and edges, Adv. Appl. Math.115(2020), 102000 on Pith
  35. [55] B. D. McKay and I. M. Wanless,On the number of Latin squares, Ann. Comb.9(2005), no. 3, 335–344 on Pith
  36. [56] R. Montgomery,A proof of the Ryser–Brualdi–Stein conjecture for large evenn, Preprint, arXiv:2310.19779, 2023 on PithPith's reviewarXiv
  37. [57] J. Oxley,Matroid theory, second ed., Oxford Graduate Texts in Mathematics, vol. 21, Oxford University Press, Oxford, 2011 on Pith
  38. [58] R. Pendavingh and J. van der Pol,On the number of matroids compared to the number of sparse paving matroids, Electron. J. Combin.22(2015), no. 2, Paper No. P2.51 on Pith
  39. [59] R. Pendavingh and J. van der Pol,Enumerating matroids of fixed rank, Electron. J. Combin. 24(2017), no. 1, Paper No. P1.8, 28 on Pith
  40. [60] N. Pippenger and J. Spencer,Asymptotic behavior of the chromatic index for hypergraphs, J. Combin. Theory Ser. A51(1989), 24–42 on Pith
  41. [61] J. Radhakrishnan,An entropy proof of Bregman’s theorem, J. Combin. Theory Ser. A77 (1997), 161–164 on Pith
  42. [62] V. R¨ odl,On a packing and covering problem, European J. Combin.6(1985), 69–78 on Pith
  43. [63] V. R¨ odl and L. Thoma,Asymptotic packing and the random greedy algorithm, Random Structures Algorithms8(1996), no. 3, 161–177 on Pith
  44. [64] H. J. Ryser,Neuere Probleme der Kombinatorik, Vortr¨ age ¨ uber Kombinatorik (Oberwol- fach, 1967), Mathematisches Forschungsinstitut Oberwolfach, 1967, pp. 69–91 on Pith
  45. [65] A. Schrijver,Counting 1-factors in regular bipartite graphs, J. Combin. Theory Ser. B72 (1998), no. 1, 122–135 on Pith
  46. [66] M. Simkin,The number ofn-queens configurations, Adv. Math.427(2023), 109127 on Pith
  47. [67] J. Spencer,Asymptotic packing via a branching process, Random Structures Algorithms7 (1995), no. 2, 167–172. 62 HYUNWOO LEE on Pith
  48. [68] S. K. Stein,Transversals of Latin squares and their generalizations, Pacific J. Math.59 (1975), 567–575 on Pith
  49. [69] Y. Tang and Y. Zhao,Counting near-spanning matchings in Latin squares and Steiner triple systems, Preprint, arXiv:2609.11006, 2026 on PithPith's reviewarXiv
  50. [70] R. van der Hofstad, R. Pendavingh, and J. van der Pol,The number of partial Steiner systems andd-partitions, Adv. Comb. (2022), Paper No. 2, 23 on Pith
  51. [71] J. H. van Lint and R. M. Wilson,A course in combinatorics, 2 ed., Cambridge University Press, 2001 on Pith
  52. [72] M. J. Wainwright and M. I. Jordan,Graphical models, exponential families, and variational inference, Found. Trends Mach. Learn.1(2008), no. 1–2, 1–305. Department of Mathematical Sciences, KAIST, South Korea and Extremal Combinatorics and Probability Group (ECOPRO), Institute ... on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment