REVIEW 1 major objections 3 minor 22 references
A Spectral Hilton--Milner--Frankl Theorem for $t$-Intersecting Families
T0 review · 1 major / 3 minor · reviewed 2026-08-11 · deepseek-v4-flash
Pith's one-line read For $1 \le t \le k-2$ and $n \ge 100\cdot 2^k k^7$, every nontrivial $t$-intersecting $k$-uniform family has adjacency-tensor spectral radius at most the larger of the two explicit families $\mathcal H_{n,k,t}$ and $\mathcal A_{n,k,t}$…
desk verdict Solid spectral Hilton–Milner–Frankl theorem with a broken appendix step; the main proof checks out, but the non-integrality of the phase threshold needs a parity fix. 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 load-bearing object is the $t$-covering kernel $\mathcal K$ of a maximal nontrivial family $\mathcal G$: the set of all $(t+1)$-sets that meet every edge of $\mathcal G$ in at least $t$ vertices. Proposition 3.1 classifies any maximal kernel as a subgraph of one of two finite models --- the $(t+1)$-uniform star $S^{(t+1)}_q$ with $q = k-t+1$ leaves, or the complete $(t+1)$-graph $T^{(t+1)}$ on $t+2$ vertices --- and shows that all edges outside the lifted kernel are covered by few complete $(t+2)$-stars. The spectral engine is the pure product-layer decomposition $M_{n,k}(\mathcal K) = \{T \cup D : T \in \mathcal K,\ D \in \binom{[n]\setminus U}{d}\}$ together with the exact product formula of Lemma 3.2, which expresses $\lambda(M_{n,k}(\mathcal K))$ as a constant $C(\mathcal K)$ times $\beta_d(n-|U|) \sim n^{d\theta}$. The gaps $\Delta_S$ and $\Delta_T$ between full and proper kernels absorb every error term, and strict monotonicity of the tensor spectral radius rules out equality for proper subfamilies.
What would settle it
Using formulas (26) and (27) with $t=1$, $k=3$, and $n = 100\cdot 2^3\cdot 3^7$, compute the exact spectral radii of $\mathcal A_{n,3,1}$ and $\mathcal H_{n,3,1}$; since $\kappa_1 \approx 3.385$, the asymptotic comparison predicts $\rho(\mathcal A) > \rho(\mathcal H)$, but the theorem itself only requires the maximum to dominate all nontrivial families. A more decisive test targets the subkernel gap: build the maximal family whose kernel is $T^{(3)}$ with one edge deleted and all lift edges, and compute its spectral radius by Lemma 3.2; the proof requires it to lie below $\rho(\mathcal A_{n,4,2})$ by at least $(\Delta_T/2)n^{d\theta}$ for $t=2$, $k=4$, $n = 100\cdot 2^4\cdot 4^7$. If that gap fails numerically, the rigidity statement would collapse.
Extended reading notes
Core claim
The central discovery is that, once full $t$-stars are excluded, the spectral extremal problem has exactly the same two candidates as the cardinality problem in the Wilson range: $\mathcal H_{n,k,t}$, whose edges contain $[t]$ and meet $[t+1,k+1]$ plus $t$ exceptional edges, and $\mathcal A_{n,k,t}$, whose edges meet $[t+2]$ in at least $t+1$ vertices. The main theorem asserts that for every $k \ge t+2$ and $n \ge 100\cdot 2^k k^7$, every nontrivial $t$-intersecting $k$-uniform family $\mathcal F$ satisfies $\rho(\mathcal F) \le \max\{\rho(\mathcal A_{n,k,t}), \rho(\mathcal H_{n,k,t})\}$, with equality exactly when $\mathcal F$ is isomorphic to the candidate, or either candidate when the two radii coincide, attaining the maximum. The proof compares not the full families but their pure product layers over a finite kernel, showing that every proper subkernel loses a fixed fraction of the leading coefficient $n^{(k-t-1)(1-1/k)}$. Asymptotically the comparison is governed by the function $\Phi_t(k) = (k-t-1)\log(t+2) + (t+1)\log(t+1) - (k-1)\log(k-t+1)$; its unique zero $\kappa_t$ is never an integer, so the Frankl family $\mathcal A_{n,k,t}$ has larger spectral radius for $k \le \lfloor\kappa_t\rfloor$ and the Hilton--Milner family $\mathcal H_{n,k,t}$ for $k \ge \lceil\kappa_t\rceil$.
Load-bearing premise
The load-bearing premise is the kernel dichotomy: every maximal nontrivial $t$-intersecting family has its minimal $t$-cover sets arranged in one of two simple patterns, a star of $(t+1)$-sets sharing a fixed $t$-set or the complete $(t+1)$-graph on $t+2$ vertices. A maximal family with any other minimal-cover pattern would break the argument.
Editorial extensions
If this is right
- The spectral Hilton--Milner--Frankl bound holds with an explicit ground-set threshold: no nontrivial $t$-intersecting $k$-uniform family can have adjacency-tensor spectral radius above the larger of $\rho(\mathcal A_{n,k,t})$ and $\rho(\mathcal H_{n,k,t})$.
- Equality is rigid: if one candidate is strictly larger, any extremal family is isomorphic to it; if the two radii are equal, exactly both candidates are extremal.
- The asymptotic phase boundary is quantized by the non-integer $\kappa_t$: the Frankl family wins for $k \le \lfloor\kappa_t\rfloor$, the Hilton--Milner family wins for $k \ge \lceil\kappa_t\rceil$, and no integer $k$ gives an asymptotic tie.
- The leading spectral coefficient of each candidate is explicit, so for fixed $n,k,t$ the exact comparison reduces to the two-variable optimization in formulas (26) and (27), and Proposition 4.3 gives a sufficient condition for either side to win.
Reading between the lines
- As a transferable extension, the kernel dichotomy and product-layer estimates should yield analogous spectral Hilton--Milner--Frankl theorems for other monomial spectral parameters, such as $\alpha$-spectral or $p$-spectral radii, not only the $k$-uniform adjacency tensor.
- The explicit threshold $100\cdot 2^k k^7$ sits far above the structural assumption $n \ge 2k+2$; a Perron-vector-aware refinement of the error absorption should bring the range down toward the Wilson range $n > (t+1)(k-t+1)$ without changing the two candidates.
- Because $\kappa_t = 2t + \tfrac12\log t + \tfrac12 + O((\log t)^2/t)$, the Frankl family's asymptotic spectral dominance extends a logarithmic window beyond $k = 2t$, a phenomenon absent from the cardinality comparison; testing whether this persists at finite $n$ through formulas (26) and (27) would answer the paper's open question about the finite phase rule.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper proves a spectral analogue of the Hilton–Milner–Frankl theorem for nontrivial t-intersecting k-uniform hypergraphs. For 1≤t≤k−2 and n≥100·2^k k^7, Theorem 1.7 asserts that ρ(F) ≤ max{ρ(A_{n,k,t}), ρ(H_{n,k,t})}, with equality only for the candidate attaining the maximum. The proof extends F to a maximal family, classifies its t-covering kernel via Proposition 3.1 into two model families, decomposes the lifted family into product layers, and uses explicit spectral formulas (Lemma 3.2, Proposition 3.4, Lemma 3.5) to absorb all error terms via the numerical inequalities of Lemma 4.1. The paper also compares the two candidates asymptotically by evaluating the sign of Φ_t(k), proving that the transition occurs at a non-integer κ_t determined by (5), and gives a finite-n comparison in Proposition 4.3. The main inequality and equality characterization appear sound and are proved elementarily without invoking asymptotic stability theorems.
Significance. The result is significant: it resolves the spectral nontrivial-intersection problem in a broad explicit parameter range and gives a sharp equality characterization matching the two classical extremal constructions. The approach is transparent and largely self-contained: the product-layer formula in Lemma 3.2 is exact, the kernel dichotomy in Proposition 3.1 is proved directly rather than imported from Ahlswede–Khachatrian, and the numerical estimates in Lemma 4.1 are fully explicit. The asymptotic phase transition with threshold κ_t is a new phenomenon for the spectral problem and is backed by a concrete formula and approximation. The equality case is handled by strict monotonicity. These are genuine strengths. However, the proof of the non-integrality of κ_t in Appendix A contains an invalid step, so one clause of Theorem 1.7 is not proved as written; the defect is localized and easily repairable.
major comments (1)
- [Appendix A] The proof of κ_t ∉ Z contains an invalid valuation step. After substituting t=7 into (29), the text derives 24=(y+5)v_2(y) and then states that this forces y even and y+5 | 24, 'hence since y>2, this forces y∈{7,19}'. This inference is wrong: if y is even, then y+5 is an odd integer greater than 7, and the only odd divisors of 24 are 1 and 3, so no such y exists; if y is odd, then v_2(y)=0 and the valuation equation reads 24=0, which is impossible. Thus the contradiction is actually immediate, but the written argument does not prove it. Since Theorem 1.7's phase-transition statement ('A for k≤⌊κ_t⌋, H for k≥⌈κ_t⌉, no tie') depends on κ_t ∉ Z, this is a load-bearing gap in a stated part of the theorem, even though the main inequality and equality characterization are unaffected. The repair is a simple parity argument, and I expect it to be routine.
minor comments (3)
- [Lemma 3.5] The displayed formula appears to have a typographical error: it reads λ(S^k_{n,c}) = k! binom(n−c,k−c)^{k−1}(k−c)^{(k−c)/k}(n−c)^{−(k−c)/k}, but the derivation in the proof and the endpoint checks give λ(S^k_{n,c}) = (k−1)! binom(n−c,k−c)(k−c)^{(k−c)/k}(n−c)^{−(k−c)/k}. Please correct the statement; the later use in (19) is consistent with the corrected formula.
- [Proposition 3.1, case (i)] In the covering argument for τ_t(G)≥t+2, the phrase 'at most k choices for x' relies on choosing B_T once for each T and C_{T,x} once for each pair (T,x), rather than separately for each F. A brief clarification would prevent a misreading of the counting step.
- [Proof of Theorem 1.7, proper subkernel of T^{(r)}] The step 'the loss estimate gives λ(M_{n,k}(K)) ≤ λ(M_{n,k}(T^{(r)})) − (Δ_T/2)n^{dθ}' suppresses a use of Lemma 3.3 to lower-bound β_d(N) by (1/2)n^{dθ}; adding this sentence would make the displayed estimate fully transparent.
Circularity Check
No significant circularity: the spectral bound is derived from an internal kernel classification and explicit product-layer spectral estimates, with no fitted parameters or load-bearing self-citations.
full rationale
The paper's central claim, Theorem 1.7, is established by reducing an arbitrary maximal nontrivial t-intersecting family to a finite kernel K (Proposition 3.1), decomposing the family into product layers, and then bounding each layer with spectral estimates proven in Section 3.2. The dichotomy for K is proved directly by elementary case analysis rather than imported from the Ahlswede–Khachatrian theorem; the classical theorem appears only as background and is not used as a proof ingredient. The two extremal families H_{n,k,t} and A_{n,k,t} are explicitly defined constructions, and the equality characterization is obtained from the kernel classification together with strict monotonicity of the spectral radius, not from assuming the desired maximum. The asymptotic comparison of the two candidates follows from the exact product-layer formulas for lambda(H) and lambda(A), leading to the sign analysis of Phi_t(k); no fitted parameter is renamed as a prediction and no self-citation carries a load-bearing assumption. The Appendix A argument that kappa_t is non-integral contains a flawed 2-adic valuation step, but that is a correctness defect, not circularity; it does not make any derivation equivalent to its inputs. Overall, the derivation chain is self-contained against the stated hypotheses and the score is 0.
Assumptions & free parameters
assumptions (5)
- standard math Weak irreducibility implies strict spectral monotonicity for coefficientwise ordered nonnegative tensors, used in Lemma 2.2.
- standard math Subadditivity and monotonicity of the k-norm spectral radius under unions and inclusions of hypergraphs, Lemma 2.1.
- standard math The edge-count spectral bound lambda(G)≤(k!|E(G)|)^theta, Lemma 2.3.
- standard math Maclaurin's inequality and the power-mean inequality for bounding products over d-subsets, Lemma 3.2.
- standard math Mihailescu's theorem (Catalan's conjecture): 8 and 9 are the only consecutive perfect powers, used in Appendix A.
Cite this review
Pith. "Pith review of A Spectral Hilton--Milner--Frankl Theorem for $t$-Intersecting Families." pith.science (2026). https://pith.science/paper/DUZR6PMR
@misc{pith2026260806810,
author = {Pith},
title = {Pith review of: A Spectral Hilton--Milner--Frankl Theorem for $t$-Intersecting Families},
year = {2026},
howpublished = {\url{https://pith.science/paper/DUZR6PMR}},
note = {Machine review of arXiv:2608.06810}
}
abstract
Keevash, Lenz, and Mubayi proved a spectral Erd\H{o}s--Ko--Rado theorem, showing that, for sufficiently large $n$, the complete $t$-star uniquely maximizes the adjacency-tensor spectral radius among all $t$-intersecting $k$-uniform families. In this paper, we establish a spectral Hilton--Milner--Frankl theorem for nontrivial $t$-intersecting families in the explicit range $1\le t\le k-2$ and $n\ge 100\cdot 2^k k^7$. More precisely, we prove that, for every nontrivial $t$-intersecting $k$-uniform family $\mathcal F$, the spectral radius satisfies \[ \rho(\mathcal F)\le \max\{\rho(\mathcal H_{n,k,t}),\rho(\mathcal A_{n,k,t})\}, \] where $\mathcal H_{n,k,t}$ and $\mathcal A_{n,k,t}$ are the two extremal families appearing in the classical Hilton--Milner--Frankl theorem. Moreover, equality holds only for the extremal candidates attaining the maximum, up to isomorphism. We further compare the two candidates asymptotically. For each fixed $t$, the unique real solution $x=x_t$ of \[ (t+2)^{x-t-1}(t+1)^{t+1}=(x-t+1)^{x-1} \] determines, as $k$ varies, which of $\mathcal H_{n,k,t}$ and $\mathcal A_{n,k,t}$ has the larger asymptotic spectral radius.
Reference graph
Works this paper leans on
-
[1]
R. Ahlswede and L. H. Khachatrian, The complete nontrivial-intersection theorem for sys- tems of finite sets,J. Combin. Theory Ser. A76(1996), 121–138
work page 1996
-
[2]
R. Ahlswede and L. H. Khachatrian, The complete intersection theorem for systems of finite sets,European J. Combin.18(1997), 125–136. 20
work page 1997
-
[3]
Borg, Intersecting and cross-intersecting families of labeled sets,Electron
P. Borg, Intersecting and cross-intersecting families of labeled sets,Electron. J. Combin. 15(2008), N9
work page 2008
-
[4]
Cooper and A
J. Cooper and A. Dutle, Spectra of uniform hypergraphs,Linear Algebra Appl.436(2012), 3268–3292
2012
-
[5]
Erd˝ os, C
P. Erd˝ os, C. Ko, and R. Rado, Intersection theorems for systems of finite sets,Quart. J. Math. Oxford Ser. (2)12(1961), 313–320
1961
-
[6]
L. Fang, G. Gao, and A. Chang, Spectral Radius Conditions for 3-Uniform Intersecting Families,arXiv:2607.08468v1
-
[7]
Frankl, On intersecting families of finite sets,J
P. Frankl, On intersecting families of finite sets,J. Combin. Theory Ser. A24(1978), 146–161
work page 1978
-
[8]
Frankl, On cross-intersecting families,Discrete Math.108(1992), 291–295
P. Frankl, On cross-intersecting families,Discrete Math.108(1992), 291–295
work page 1992
Show all 22 references
-
[9]
Frankl and N
P. Frankl and N. Tokushige,Extremal Problems for Finite Sets, Vol. 86, AMS, Providence, RI, 2018
2018
-
[10]
Frankl and N
P. Frankl and N. Tokushige, Some best possible inequalities concerning cross-intersecting families,J. Combin. Theory Ser. A61(1992), 87–97
1992
-
[11]
Frankl and J
P. Frankl and J. Wang, A product version of the Hilton–Milner–Frankl theorem,Sci. China Math.67(2024), 455–474
2024
-
[12]
Friedland, S
S. Friedland, S. Gaubert, and L. Han, Perron–Frobenius theorem for nonnegative multilin- ear forms and extensions,Linear Algebra Appl.438(2013), 738–749
2013
-
[13]
A. J. W. Hilton and E. C. Milner, Some intersection theorems for systems of finite sets, Quart. J. Math. Oxford Ser. (2)18(1967), 369–384
1967
-
[14]
Rajesh Kannan, N
M. Rajesh Kannan, N. Shaked-Monderer, and A. Berman, On weakly irreducible nonneg- ative tensors and interval hull of some classes of tensors,Linear Multilinear Algebra64 (2016), 667–679
2016
-
[15]
L. Kang, Y. Lu, X. Yuan, and J. Zhou, A spectral confirmation of the Erd˝ os matching conjecture,arXiv:2607.07392v1, 2026
2026 arXiv
-
[16]
Keevash, J
P. Keevash, J. Lenz, and D. Mubayi, Spectral extremal problems for hypergraphs,SIAM J. Discrete Math.28(2014), 1838–1854
2014
-
[17]
L. Liu, Z. Ni, J. Wang, and L. Kang, Hypergraph extensions of spectral Tur´ an theorem, arXiv:2408.03122v1, 2024
2024 arXiv
-
[18]
Nikiforov, Analytic methods for uniform hypergraphs,Linear Algebra Appl.457(2014), 455–535
V. Nikiforov, Analytic methods for uniform hypergraphs,Linear Algebra Appl.457(2014), 455–535
2014
-
[19]
Mih˘ ailescu, Primary cyclotomic units and a proof of Catalan’s conjecture,J
P. Mih˘ ailescu, Primary cyclotomic units and a proof of Catalan’s conjecture,J. Reine Angew. Math.572(2004), 167–195
2004
-
[20]
Qi, Eigenvalues of a real supersymmetric tensor,J
L. Qi, Eigenvalues of a real supersymmetric tensor,J. Symbolic Comput.40(2005), 1302– 1324
2005
-
[21]
R. M. Wilson, The exact bound in the Erd˝ os–Ko–Rado theorem,Combinatorica4(1984), 247–257. 21
1984
-
[22]
Zhang and X.-D
P.-L. Zhang and X.-D. Zhang, The spectral radii of intersecting uniform hypergraphs,Com- mun. Appl. Math. Comput.3(2021), 243–256. 22
2021
Reviewed August 11, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.