REVIEW 1 major objections 5 minor 17 references
Satisfying sequences for rainbow partite matchings
T0 review · 1 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash
Pith's one-line read The paper establishes which asymmetric sequences of size thresholds are 'satisfying' for rainbow matchings in $k$-partite hypergraphs, proving that thresholds can be spread from $(i-1)n^{k-1}$ upward while still forcing a cross-matching.
desk verdict A mostly solid paper that completes a line of work on asymmetric satisfying sequences, but Theorem 2 has a real constant bug (C≥24, not C≥20) that should be fixed. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The proof runs on three interlocking tools. Spread approximation (the method of [15]) decomposes each $\mathcal F_i$ into a cross-dependent family $\mathcal S_i$ of sets of size at most 2 and a leftover family of size at most $r^3 n^{k-3}$, where $r=2^5s\log_2(sk)$; a family is $r$-spread if every subfamily induced by fixing a set $X$ has size at most $r^{-|X|}|\mathcal F|$, and the spread lemma then says a randomly colored ground set contains a monochromatic member with high probability. For Theorem 2, a uniformly random perfect matching $M$ is used: the degrees $\zeta_i=|\mathcal F_i\cap M|$ concentrate near their expectations by a theorem from [9], the usual marriage condition forces a subset $B$ where all degrees are small, and an anticoncentration theorem (Theorem 11) shows that if $|\mathcal F\cap M|=s-1$ almost always, then $\mathcal F$ has $s-1$ fat parallel hyperplanes. This structural consequence lets the authors reduce to a single matching lying inside $s-1$ common hyperplanes, which gives the contradiction. Theorem 3 instead maps $[p]^2$ bijectively into a field $F=\mathbb Z_p(\alpha)$ with $\alpha^2$ a non-residue, so that $(x_i-x_j)^2\in\mathbb Z_p$ exactly when the two tuples share a coordinate; the Combinatorial Nullstellensatz turns a nonzero coefficient into a point where the product avoids $\mathbb Z_p$, i.e., into a rainbow matching.
What would settle it
Check the constant bookkeeping in Section 5: for $C\in[20,24)$, the condition $k<Cn/(3\sqrt{s/\log s})$ does not imply the displayed lower bound $|\mathcal F_i'|\ge (i+\tfrac C2\sqrt{s\log s})n^{k-1}$, because the proof's reduction requires $(C/2-4)\ge C/3$, i.e. $C\ge24$. If the inequalities indeed fail for $C=20$, then Theorem 2 as stated needs correction. A second concrete test is to search for cross-dependent families with $|\mathcal F_i|=f_i+1$ in the stated $n,k$ range using the Section 2 constructions, which already give counterexamples for larger $k$.
Extended reading notes
Core claim
On its own terms, the paper claims three theorems. Theorem 9: for $n>2^5 s\log_2(sk)$, the sequence $f_i=(i-1)n^{k-1}+4(s-1)^2n^{k-2}+2^{15}s^3\log_2^3(sk)n^{k-3}$ is satisfying; Theorem 1 follows as a corollary for $n\ge\max\{2^8s^{3/2}\log_2^{3/2}(sk),8s^2\}$ with $f_i=i n^{k-1}$. Theorem 2: for $C\ge20$, $s>s_0(C)$, and $k< Cn/(3\sqrt{s/\log s})$, the truncated sequence $f_i=\min(s-1,i+C\sqrt{s\log s})n^{k-1}$ is satisfying, while the constructions in Section 2 show the restriction on $k$ is tight up to a constant factor. Theorem 3: if $n=p$ is prime, $k=2$, and the coefficient of $x_1^{f_1}\cdots x_s^{f_s}$ in $\prod_{1\le i<j\le s}(x_j-x_i)^2$ is nonzero modulo $p$, then $(pf_1,\ldots,pf_s)$ is satisfying. Together these map out the landscape of asymmetric thresholds for rainbow matchings in $k$-partite hypergraphs.
Load-bearing premise
The load-bearing premise is a previously proved concentration bound, quoted from [9], that for a family of size just above $(s-1)n^{k-1}$, the expected excess of $|\mathcal F\cap M|$ over $s-1$, conditional on being positive, is at most $3.7\sqrt{s\log s}$; the contradiction chain in Theorem 2 collapses without it, and the Section 5 constants appear to require $C\ge24$ rather than the stated $C\ge20$.
Editorial extensions
If this is right
- For $n>2^5s\log_2(sk)$, ordered families with $|\mathcal F_i|>(i-1)n^{k-1}+4(s-1)^2n^{k-2}+2^{15}s^3\log_2^3(sk)n^{k-3}$ are guaranteed a rainbow matching, so the full uniform threshold is only needed for the largest family.
- For $n\ge\max\{2^8s^{3/2}\log_2^{3/2}(sk),8s^2\}$, the arithmetic sequence $i n^{k-1}$ is satisfying, settling the first conjectured sequence in this range.
- If Theorem 2 is correct, then for $k<Cn/(3\sqrt{s/\log s})$ one can take $M=C\sqrt{s\log s}$ families at the full $(s-1)n^{k-1}$ threshold while every other family needs only $(i+C\sqrt{s\log s})n^{k-1}$; the Section 2 constructions show this $k$-range cannot be improved by more than a constant factor.
- For prime $n=p$ and $k=2$, every assignment $f_1,\dots,f_s$ whose monomial occurs in $\prod_{i<j}(x_j-x_i)^2$ with a nonzero coefficient modulo $p$ gives a satisfying sequence $pf_1,\dots,pf_s$, yielding many nonuniform thresholds beyond the uniform one.
- The negative examples show $(i-\tfrac12)n^{k-1}-1$ can fail, so the additive slack $m$ in the linear spread $m+(i-1)n^{k-1}$ lies below $\tfrac12 n^{k-1}$ while the positive results achieve lower-order $m$, narrowing the remaining open problem.
Reading between the lines
- The coefficient criterion in Theorem 3 suggests that, for $k=2$ and prime $n$, the set of satisfying sequences may be exactly the integer points in the Newton polytope of $\prod_{i<j}(x_j-x_i)^2$; if true, this would give a complete algebraic description for the prime case.
- The three-case split in Theorem 2 (concentration, $n\ge s^5$, and anticoncentration) suggests the true boundary between satisfying and non-satisfying truncated sequences is smoother than the proof's case division, and the $n\approx s^5$ threshold is likely an artifact of the method.
- The spread-approximation decomposition used here could be adapted to non-partite families $\binom{[n]}{k}$, where the analogous uniform threshold is still open; asymmetric bounds of the same shape might give new constraints on cross-dependent families there.
- For $k>2$, a Nullstellensatz analogue of Theorem 3 would need a polynomial whose vanishing detects whether two chosen tuples share one of $k$ coordinates; the squared-difference trick does not extend directly, so prime-power generalizations would require a different algebraic encoding.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies sequences f_1,...,f_s such that |F_i| > f_i for families F_i ⊂ [n]^k forces a rainbow matching. It proves Theorem 1: for n ≥ max{2^8 s^{3/2} log_2^{3/2}(sk), 8s^2}, the sequence {i n^{k-1}} is satisfying. Theorem 2: for C ≥ 20, s > s_0(C), and k < Cn/(3√(s/log s)), the truncated sequence {min(s−1, i+C√s log s)n^{k-1}} is satisfying. Theorem 3: for prime n=p and k=2, sequences (p f_1,...,p f_s) are satisfying when the coefficient of x_1^{f_1}...x_s^{f_s} in ∏_{i<j}(x_j−x_i)^2 is nonzero mod p. It also provides examples showing optimality up to constants and proves a spread-approximation result (Theorem 9) and a large-n result (Theorem 17).
Significance. If the results hold, they answer questions of Kiselev and Kupavskii, establish the linear sequence in a large-n regime, and delineate the valid range for the truncated sequence up to constant factors. The methods—spread approximations, anticoncentration for matching intersections, and Combinatorial Nullstellensatz—are appropriate, and the proofs of Theorems 1, 3, 9, and 17 are largely self-contained. The main caveat is Theorem 2's dependence on a quantitative concentration bound from [9] and the constant mismatch discussed below; both are local and fixable.
major comments (1)
- [Section 5, proof of Theorem 2, inequality after (8)] The proof requires C ≥ 24 rather than C ≥ 20. The displayed lower bound on |F'_i| uses k ≤ n(C/2−4)/√(s/log s), and the text states that this is implied by the theorem's hypothesis k ≤ Cn/(3√(s/log s)) for C ≥ 20. Algebraically, the implication needs C/3 ≤ C/2 − 4, i.e., C ≥ 24. For 20 ≤ C < 24 the stated k-range does not imply the needed bound, so the subsequent Hall-contradiction argument is not established for the full stated parameter range. The theorem is repairable by taking C ≥ 24 (or by replacing the k-range with k ≤ n(C/2−4)/√(s/log s)), but as written the proof does not prove Theorem 2 for C ∈ [20,24).
minor comments (5)
- [Section 1 (Theorem 1) and Section 3 (Theorem 9)] The exponents in the lower bounds on n are typographically ambiguous: '28' should be '2^8' and '25' should be '2^5', matching the proof's choice r = 2^5 s log_2(sk). Please clarify throughout.
- [Section 2, Claim 5] The bound 'k−1 ≥ 3Cn√(log s)/s' is inconsistent with the proof, which yields α > 3C√(log s)/√s (i.e., k−1 ≥ 3Cn√(log s)/√s). Please correct the displayed formula.
- [Abstract] 'Nullstellenzats' is a typo for 'Nullstellensatz'.
- [Section 5, Claim 14 proof] The condition '10 ≤ C ≤ √(s/log s)' should be phrased as 'for s sufficiently large relative to C', since C is fixed while s varies.
- [Section 5, Case 3] The expression 's^{-10,5}' should use a period as the decimal separator.
Circularity Check
No circularity: the main theorems are derived from independent concentration, spread-approximation, and Nullstellensatz lemmas; self-citations are upstream and not load-bearing.
full rationale
I walked the derivation chain of Theorems 1, 2 and 3. Theorem 1 is deduced from Theorem 9 by a direct parameter inequality, and Theorem 9 is proved via spread approximation, with the spread lemma itself cited to external work [4, 16, 17]. Theorem 2 uses Hall's theorem together with concentration tools: Theorem 13 and the bound E[zeta_i | zeta_i > 0] <= 3.7 sqrt(s log s) are quoted from [9], and Theorem 11 is proved in this paper. These are upstream, parameter-free external lemmas with stated assumptions; they are not restatements of the target sequences, and the paper does not fit any parameter to the conclusion. Theorem 3 is a direct application of Alon's Combinatorial Nullstellensatz, with the nonzero-coefficient condition as an explicit hypothesis and the coefficient translation established by the displayed polynomial identities. I found no step where a definition is given in terms of the conclusion, no fitted input renamed as a prediction, and no known result merely renamed. The skeptical note about the proof of Theorem 2 requiring C >= 24 rather than C >= 20 is a possible correctness gap in the algebra of Section 5, not a circularity: it does not make the theorem's conclusion equivalent to its inputs. Accordingly, no specific circular reduction can be exhibited, and the paper is self-contained relative to its cited external lemmas.
Assumptions & free parameters
assumptions (6)
- domain assumption Concentration inequality for |F cap M| (Theorem 13 of [9]).
- domain assumption Conditional expectation bound E[|F cap M|-s+1 | |F cap M| >= s] <= 3.7 sqrt(s log s) from [9].
- domain assumption Spread lemma (Theorem 8) due to Alweiss-Lovett-Wu-Zhang, Tao, and Stoeckl.
- standard math Alon's Combinatorial Nullstellensatz.
- standard math Expander mixing lemma for powers of complete graphs.
- standard math Identity prod_{q in Z_p}(Y-q) = Y^p - Y over F_p.
Cite this review
Pith. "Pith review of Satisfying sequences for rainbow partite matchings." pith.science (2026). https://pith.science/paper/6RX2LO3B
@misc{pith2026250203105,
author = {Pith},
title = {Pith review of: Satisfying sequences for rainbow partite matchings},
year = {2026},
howpublished = {\url{https://pith.science/paper/6RX2LO3B}},
note = {Machine review of arXiv:2502.03105}
}
abstract
Let $\mathcal F_1,\ldots, \mathcal F_s\subset [n]^k$ be a collection of $s$ families. In this paper, we address the following question: for which sequences $f_1,\ldots, f_s$ the conditions $|\ff_i|>f_i$ imply that the families contain a rainbow matching, that is, there are pairwise disjoint $F_1\in \ff_1,\ldots F_s\in \ff_s$? We call such sequences {\em satisfying}. Kiselev and the first author verified the conjecture of Aharoni and Howard and showed that $f_1 = \ldots = f_s=(s-1)n^{k-1}$ is satisfying for $s>470$. This is the best possible if the restriction is uniform over all families. However, it turns out that much more can be said about asymmetric restrictions. In this paper, we investigate this question in several regimes and in particular answer the questions asked by Kiselev and Kupavskii. We use a variety of methods, including concentration and anticoncentration results, spread approximations, and Combinatorial Nullstellenzats.
Reference graph
Works this paper leans on
-
[9]
S. Kiselev, A. Kupavskii, Rainbow matchings in k-partite hypergraphs, Bulletin of the London Math. Society 53 (2021), N2, 360–369
work page 2021
-
[1]
R. Aharoni and D. Howard, A rainbow k-partite version of the Erd˝ os–Ko–Rado Theorem, Comb. Probab. Comput. 26 (2017), N3, 321–337
work page 2017
-
[2]
N, Alon, Combinatorial nullstellensatz , Comb. Probab. Comput. 8 (1999), NN1-2, 7–29
work page 1999
-
[3]
N. Alon and J. H. Spencer, The Probabilistic Method , Fourth Edition, 2015
work page 2015
-
[4]
R. Alweiss, S. Lovett, K. Wu, and J. Zhang, Improved bounds for the sunflower lemma , arXiv:1908.08483 (2019)
arXiv 2019
-
[5]
H. Huang, P.-S. Loh and B. Sudakov, The Size of a Hypergraph and its Matching Num- ber, Comb. Probab. Comput. 21 (2012), N3, 442–450
work page 2012
-
[6]
Erd˝ os,A problem on independent r-tuples , Ann
P. Erd˝ os,A problem on independent r-tuples , Ann. Univ. Sci. Budapest. 8 (1965) 93–95
work page 1965
-
[7]
Frankl, Improved bounds for Erd˝ os’ Matching Conjecture , J
P. Frankl, Improved bounds for Erd˝ os’ Matching Conjecture , J. Comb. Theory Ser. A 120 (2013), 1068–1072. 18
work page 2013
Show all 17 references
-
[8]
Frankl, A
P. Frankl, A. Kupavskii, The Erd˝ os Matching Conjecture and Concentration Inequalities, Journal of Comb. Theory Ser B. 157 (2022), 366–400
2022
-
[10]
Kolupaev, A
D. Kolupaev, A. Kupavskii, Erd˝ os Matching Conjecture for almost perfect matchings , Discrete Math. 346 (2023), N4
2023
-
[11]
Kupavskii, Erd˝ os–Ko–Rado type results for partitions via spread appr oximations (2023), arXiv.2309.00097
A. Kupavskii, Erd˝ os–Ko–Rado type results for partitions via spread appr oximations (2023), arXiv.2309.00097
2023
-
[12]
Kupavskii, Intersection theorems for uniform subfamilies of heredita ry families (2023), arXiv.2311.02246
A. Kupavskii, Intersection theorems for uniform subfamilies of heredita ry families (2023), arXiv.2311.02246
2023 arXiv
-
[13]
Kupavskii, An almost complete t-intersection theorem for permutation s (2024), arXiv:2405.07843
A. Kupavskii, An almost complete t-intersection theorem for permutation s (2024), arXiv:2405.07843
2024 arXiv
-
[14]
Kupavskii, F
A. Kupavskii, F. Noskov, Linear dependencies, polynomial factors in the Duke–Erd˝ o s forbidden sunflower problem (2024), arXiv:2410.06156
2024 arXiv
-
[15]
Kupavskii and D
A. Kupavskii and D. Zakharov, Spread approximations for forbidden intersections prob- lems, to appear in Advances in Mathematics, available at arXiv:2 203.13379
-
[16]
Stoeckl, Lecture notes on recent improvements for the sunflower lemma https://mstoeckl.com/notes/research/sunflower_notes.html
M. Stoeckl, Lecture notes on recent improvements for the sunflower lemma https://mstoeckl.com/notes/research/sunflower_notes.html
-
[17]
Tao, The sunflower lemma via shannon entropy , https://terrytao.wordpress.com/2020/07/20/the-sunflower-lemma-via-shannon-entropy/ 19
T. Tao, The sunflower lemma via shannon entropy , https://terrytao.wordpress.com/2020/07/20/the-sunflower-lemma-via-shannon-entropy/ 19
2020
Reviewed August 9, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.