REVIEW 2 major objections 5 minor 8 references
Further remarks on fractional vs. expectation thresholds
T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read The paper establishes a new case of Talagrand's conjecture for constant functions supported on edge sets of ℓ-uniform hypergraph cliques, with an absolute constant L = 2^{12} e^{16} for sufficiently large ground sets.
desk verdict Genuine extension of DeMarco-Kahn to hypergraph cliques; results are conditional on the authors' earlier reduction, and the constant-weight assumption is stated, not derived. 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 carrying machinery is a probabilistic union construction. One samples s independent uniform k-edges e_{1,1},...,e_{1,s} from supp(g), forms their union e_1, and repeats t times, with t = $p^{{-sk}}$ n; the random family G is the collection of these t unions. The size identity |e_1| = sk - \sum_{j\le s} Y_j, where Y_j = |e_{1,j} \cap (e_{1,1}\cup\cdots\cup e_{1,j-1})|, converts the expected weight \mathbb{E} w(G,p/L) into a generating-function of the overlap sum \sum_j Y_j; choosing sk = \ln n cancels the factor n. Theorem 4 gives a sufficient tail bound on that overlap sum, Theorem 5 gives a sufficient bound using only the two-set overlap distribution, and both imply \mathbb{E}[w(G,p/L)\mid \langle g\rangle\subseteq\langle G\rangle]\le 1, so a suitable G exists. The clique application verifies these conditions by bounding the overlap of two random \tilde k-subsets of [\tilde n] with hypergeometric tail estimates.
What would settle it
The direct falsifier is a counterexample: fix $L=2^{12}e^{16}$, choose $\tilde n$ arbitrarily large and parameters satisfying Definition 8 with $w(g,p)=1$, and compute the minimum of $w(G,p/L)$ over all $G$ with $\langle g\rangle\subseteq\langle G\rangle$; a value above 1 disproves Corollary 11. Since the proof is asymptotic, the first place to look is the boundary $\ell=4\ln\tilde n$, where the two range-splitting estimates in the proof meet and their constants may leave the claimed chain of inequalities invalid.
Extended reading notes
Core claim
Under the clique-hypergraph assumptions of Definition 8, let X be the set of ℓ-subsets of [\tilde n] and let E consist of the edge sets of all cliques of order \tilde k. For g equal to 1/r on E and zero elsewhere, with r chosen between L k and |E|, and with p = (r/|E|)^{1/k} so that w(g,p)=1, Corollary 11 asserts that for L ≥ $2^{{12}}$ $e^{{16}}$ and \tilde n sufficiently large there is a family G of nonempty subsets with \langle g\rangle ⊆ \langle G\rangle and w(G,p/L) ≤ 1. In words, the constant function supported on hypergraph-clique edge sets satisfies the fractional-vs-expectation threshold conjecture, with an explicit absolute constant, in the uniform-size setting k = \binom{\tilde k}{\ell} ≤ \ln|X| identified by the reduction.
Load-bearing premise
The argument depends on the authors' earlier reduction saying that solving the k-uniform case with k ≤ ln|X| is enough for the full conjecture; if that reduction has a gap, the new results do not extend the original conjecture.
Editorial extensions
If this is right
- Within the stated size assumptions, every constant function on $\ell$-uniform hypergraph cliques has its fractional and expectation thresholds within the absolute factor $2^{12} e^{16}$.
- Conjecture 1 follows for this class of functions, provided the earlier reduction to the $k\le \ln|X|$ regime is sound.
- The $\ell=2$ case reproduces the known graph-clique theorem with an explicit constant.
- The witnessing family $G$ is constructive: it is a random collection of unions of $s$ sampled cliques with $s k = \ln\binom{\tilde n}{\ell}$, so the proof gives a concrete way to build the approximating family.
Reading between the lines
- The two overlap conditions in Theorems 4 and 5 are stated abstractly, so the same argument should apply to any $k$-uniform support family whose two-set overlap distribution is tame; paths or cycles are plausible next test cases.
- The constant $2^{12} e^{16}$ comes from crude logarithmic estimates; sharper handling of inequality (5.4) would likely lower it, and the proof's two-range split suggests the value is not intrinsic to the problem.
- The 'large enough' qualifier is a byproduct of the proof's asymptotic comparisons; extracting an explicit threshold on $\tilde n$ would make the result checkable by exhaustive search for small parameter values.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies Talagrand's conjecture comparing the expectation threshold with the fractional expectation threshold. Building on the authors' earlier reduction in [3], it considers k-uniform functions g on the k-subsets of an n-element set with k ≤ ln n. Under the additional standing assumption that g is constant on its support E, taking value 1/r with |E| = d, and with p determined by w(g,p) = 1, the paper states two sufficient conditions (Theorems 4 and 5) for the existence of a set family G with ⟨g⟩ ⊆ ⟨G⟩ and w(G, p/L) ≤ 1. These conditions are then applied to hypergraph cliques: Definition 8 sets up the clique hypergraph, Lemmas 9 and 10 verify the hypotheses of Theorems 4 and 5, and Corollary 11 asserts that for L ≥ 2^12 e^16 and all sufficiently large ñ, the clique-supported constant-weight case satisfies the conjecture.
Significance. If the claims are correct, the paper extends the DeMarco–Kahn method from graph cliques to ℓ-uniform hypergraph cliques of order k̃, which is a genuinely new special case of Talagrand's conjecture. The abstract framework of Theorems 4 and 5 is potentially reusable, and the proofs contain explicit moment and hypergeometric tail estimates that are largely checkable. However, the significance currently depends on two points that need repair: the constant-support assumption in Section 2 is not derived or cited, and the implication (1.3) ⇒ (1.2) in Theorem 4 contains an invalid conditional-probability inequality. The first point sharpens the scope of the result, and the second directly affects the application in Corollary 11.
major comments (2)
- [§2 (after Conjecture 2)] The passage 'we assume that there exists an r > 0 such that g(S) = 1/r · 1_{S∈E}' imposes a constant-weight restriction that is not derived from Conjecture 2 or from [3, Theorems 8 and 10]. For a general g, the upset ⟨g⟩ is defined by weighted sums, and it is not true that any such g can be replaced by a uniform weight 1/r on the same support without changing ⟨g⟩; for example, two edges of weight 0.6 each produce a different threshold family from a uniform weight of 1/r with r = 1/0.6. Consequently, Theorems 4 and 5 and Corollary 11 are proved only for constant-weight functions, and the paper should either prove a reduction to this case, cite one, or explicitly state that the 'special cases' settled are the constant-weight clique hypergraphs. As written, the step from Conjecture 2 to Definition 3 is an unproved assumption.
- [§3, proof of Theorem 4, implication (1.3) ⇒ (1.2)] The inequality ℙ(Y_j = y_j | Y_2 = y_2, …, Y_{j-1} = y_{j-1}) ≤ ℙ(Y_s ≥ y_j | Y_2 = y_2, …, Y_{s-1} = y_{s-1}) is false in general. For k = 2 and s = 3, take the conditioning event Y_2 = 2 (so e_{1,2} = e_{1,1}) and Y_3 = 1 (so e_{1,3} shares exactly one vertex with e_{1,1}). For j = 2, the left side equals 1/|E| > 0, while the right side, ℙ(Y_3 ≥ 2 | Y_2 = 2, Y_3 = 1), equals 0 because the conditioning includes Y_3 = 1. This invalidates the derivation of (1.2) from (1.3) as written. Since Lemma 9 verifies only condition (1.3), the application of Theorem 4 in Corollary 11 is not justified. A repair is likely available: the hypergeometric estimate in Lemma 9 appears to prove the stronger bound for every j, not only for Y_s; the authors should state and prove that stronger condition and use it in the proof of Theorem 4.
minor comments (5)
- [§4, proof of Theorem 5] The phrase 'we may assume that r ∈ ℕ' is not justified. Since Definition 3 allows real r, the construction of G1 should use ⌈r⌉ disjoint edges, and the estimate for w(G1, p/L) needs a short adjustment for the rounding.
- [§5, Lemma 9] The step combining the two factors after equation (5.3) is hard to follow; the authors should display the exponents involving ilde k, k, and ilde y more explicitly, including all cancellations, so that the final expression is verifiable.
- [Corollary 11] The statement says 'for ilde n large enough' without a quantitative threshold. This is acceptable for an asymptotic existence statement for an absolute L, but the authors should explain why the finitely many small cases can be absorbed by increasing L.
- [§1 and references] There is a typo 'Defintion' in Section 1, and the reference [3] is listed as 'submitted'. Since Theorems 8 and 10 of [3] are load-bearing for the reduction to k-uniform functions with k ≤ ln|X|, the authors should state their exact statements or confirm that the preprint is publicly available.
- [§2, Lemma 7] The notation inom{d}{r} is used with a real r before any integrality convention is introduced; a sentence clarifying that r can be replaced by ⌈r⌉ in combinatorial bounds would remove ambiguity.
Circularity Check
No construction-level circularity; the proof is self-contained under its stated restrictions, with the link to Talagrand's conjecture inherited from the authors' previous reduction [3].
full rationale
Theorems 4 and 5 are sufficient-condition results proved directly from Definition 3: Lemma 6 bounds the probability that the random family G covers ⟨g⟩, and the expected weight computation for w(G,p/L) is reduced to the distribution of the overlaps Y_j. Inequality (1.2) (resp. (1.4)) is a sufficient condition, not the desired conclusion. In the clique application, Lemmas 9 and 10 verify these inequalities by explicit hypergeometric estimates, and Corollary 11 merely compares the resulting lower and upper bounds on r and chooses L. No parameter is fitted to the statement being proved, and no conclusion is used as an input. The constant-on-support restriction in Section 2 and the reduction to k-uniform functions with k≤ln|X| from [3, Theorems 8 and 10] are external dependencies: Corollary 11 is a special case of Conjecture 2 conditional on those cited reductions, not an equivalence that defines the conclusion in terms of the assumptions. This is a scope limitation and a non-circular reliance on the authors' prior work, not a circular derivation.
Assumptions & free parameters
assumptions (3)
- standard math Sperner's theorem: the maximum antichain in 2^X has size at most C(n, floor(n/2)).
- standard math Standard hypergeometric tail bounds, e.g., P(Hypergeom(N,K,n) ≥ x) ≤ C(n,x)(K/N)^x or similar forms.
- domain assumption The reduction from Talagrand's Conjecture 1 to Conjecture 2 for k-uniform functions with k ≤ log_L |X|, and specifically to k ≤ ln |X|, is valid.
Cite this review
Pith. "Pith review of Further remarks on fractional vs. expectation thresholds." pith.science (2026). https://pith.science/paper/GO5N2R7C
@misc{pith2026250521782,
author = {Pith},
title = {Pith review of: Further remarks on fractional vs. expectation thresholds},
year = {2026},
howpublished = {\url{https://pith.science/paper/GO5N2R7C}},
note = {Machine review of arXiv:2505.21782}
}
read the original abstract
A conjecture of Talagrand (2010) states that the so-called expectation and fractional expectation thresholds are always within at most some constant factor from each other. In this note we generalize a method of DeMarco and Kahn and settle a few more special cases.
Reference graph
Works this paper leans on
-
[3]
Some results on fractional vs. expectation thresholds
T. Fischer and Y. Person, Some results on fractional vs. expectation thresholds (2023). Submitted, arXiv:2311.08163
work page Pith review arXiv 2023
-
[1]
B. DeMarco and J. Kahn, Note on a problem of M. Talagrand. Random Struct. Algorithms 47.4 (2015), 663–668, doi: 10.1002/rsa.20559
-
[2]
Note on a conjecture of Talagrand: expectation thresholds vs. fractional expectation thresholds
Q. Dubroff, J. Kahn and J. Park, Note on a conjecture of Talagrand: expectation thresholds vs. fractional expectation thresholds (2024). ArXiv:2412.00917
work page Pith review arXiv 2024
-
[4]
K. Frankston, J. Kahn and J. Park, On a problem of M. Talagrand. Random Struct. Algorithms 61.4 (2022), 710–723, doi: https://doi.org/10.1002/rsa.21077
-
[5]
J. Kahn and G. Kalai, Thresholds and expectation thresholds. Combin. Probab. Comput. 16.3 (2007), 495–502
work page 2007
-
[6]
J. Park and H. T. Pham, A proof of the Kahn-Kalai conjecture. Journal of the American Mathematical Society 37.1 (2024), 235–243
work page 2024
-
[7]
H. T. Pham, A sharp version of Talagrand’s selector process conjecture and an application to rounding fractional covers (2024). ArXiv:2412.03540
work page Pith review arXiv 2024
-
[8]
M. Talagrand, Are many small sets explicitly small? Proceedings of the 42nd annual ACM symposium on theory of computing, STOC ’10. Cambridge, MA, USA, June 5–8, 2010. , 13–36, New York, NY: Association for Computing Machinery (ACM), ISBN 978-1-60558-817-9 (2010), doi: 10.1145/1806689.1806693. Institut für Mathematik, Technische Universität Ilmenau, 98684 ...
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.