{"id":"d41aeb1e-48c1-456b-808b-0252800ac4de","arxiv_id":"2608.06810","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For large enough ground sets, every nontrivial t-intersecting k-uniform family has adjacency-tensor spectral radius bounded by the larger of two explicit extremal families, with equality only for copies of those families.","lead":"This paper proves a spectral version of a classic extremal set theory theorem: among nontrivial families of k-element sets that pairwise overlap in at least t elements, the largest spectral radius is always attained by one of two explicitly described families. The result extends the spectral Erdos-Ko-Rado theorem to the nontrivial case and identifies a sharp asymptotic phase transition between the two extremal candidates.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main spectral bound appears sound; however, the proof that κ_t is not an integer in Appendix A contains an invalid 2-adic valuation step, leaving a stated part of Theorem 1.7 unsupported.","rationale":"The reader identified Proposition 3.1 as the weakest spot. I examined that proposition and its use in the main proof and found the structural dichotomy sound: the covering argument for τ_t ≥ t+2, the lifting property L_n(K) ⊆ G, the proof that K is t-intersecting via n ≥ 2k+2, and the classification into the star or complete (r+1)-graph models all hold. The product-layer formula in Lemma 3.2, the coefficient comparisons, and the error absorption in Lemma 4.1 are also consistent. The only real defect I found is in Appendix A's non-integrality proof, which is a stated part of Theorem 1.7's asymptotic comparison. The gap is narrow and likely fixable, so the correct verdict is conditional acceptance rather than rejection or unverdictable: the main spectral theorem appears correct, but the manuscript needs a corrected appendix before the full statement of Theorem 1.7 is rigorously established.","tokens_in":17570,"tokens_out":42113,"duration_ms":334289,"concrete_test":"Independently solve the final 2-adic equation in Appendix A: for integer y>2, test all possibilities in 24 = (y+5)v_2(y). If v_2(y)=0 the RHS is 0, impossible; if v_2(y)≥1 then y is even, so y+5 is an odd divisor of 24, and since y+5>7 no such divisor exists. Confirm that the set {7,19} does not actually satisfy the equation, and check whether replacing the flawed 'forces y∈{7,19}' step with this parity contradiction restores a complete proof of κ_t ∉ Z.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The main inequality and equality characterization in Theorem 1.7 survive scrutiny: the kernel classification in Proposition 3.1, the product-layer estimates in Section 3.2, and the constant absorption in Lemma 4.1 all check out. The concrete flaw is in Appendix A, which proves κ_t ∉ Z. After substituting t = 7 into (29), the text obtains 24 = (y+5)v_2(y) and states that this forces y even and y+5 | 24, hence 'since y>2, this forces y∈{7,19}'. This inference is wrong: if v_2(y)≥1 then y is even, so y+5 is odd; the only odd divisors of 24 are 1 and 3, and y+5 > 7 rules them out. Moreover, y=7 and y=19 are odd and do not satisfy 24 = (y+5)v_2(y). No integer y>2 solves the 2-adic valuation equation, so the contradiction is even cleaner than stated, but the written proof's intermediate step is invalid. Because Theorem 1.7's phase-transition statement ('A for k ≤ floor(κ_t), H for k ≥ ceil(κ_t), no tie') depends on κ_t ∉ Z, this leaves a stated part of the theorem without a valid proof as written, although the defect appears repairable by a parity argument.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":17872,"tokens_out":24032,"duration_ms":198000,"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":[{"comment":"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.","section":"Appendix A"}],"minor_comments":[{"comment":"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.","section":"Lemma 3.5"},{"comment":"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.","section":"Proposition 3.1, case (i)"},{"comment":"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.","section":"Proof of Theorem 1.7, proper subkernel of T^{(r)}"}],"recommendation":"major_revision","confidential_remarks":"The main theorem and its equality characterization appear correct, and the only mathematical gap I found is the invalid 2-adic step in Appendix A. That gap concerns the non-integrality of κ_t and hence the no-tie clause of the asymptotic comparison, but it is easily repaired by the parity argument described in my report. I also note the typo in Lemma 3.5. I recommend major revision rather than rejection: after the authors supply the corrected appendix argument and the formula fix, I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a real theorem, not a repackaging. It gives the first spectral analogue of the complete nontrivial-intersection theorem for all t in the stated range, with equality characterization and an asymptotic phase transition. The Fang–Gao–Chang result covers only k=3, t=1, so the extension is genuine.\n\nThe proof is mostly self-contained and elementary: kernel classification, product-layer decomposition, and explicit coefficient bounds. The structural dichotomy in Proposition 3.1 is the load-bearing step, and it is proved rather than assumed. The product-layer formula in Lemma 3.2 is exact, and the numerical inequalities in Lemma 4.1 are explicit enough to check. I found no circularity: the Ahlswede–Khachatrian theorem is cited for context, but the needed dichotomy is reproved in the paper. The equality characterization via strict monotonicity of the spectral radius is clean.\n\nThe soft spot is Appendix A. The proof that the phase threshold κ_t is not an integer contains an invalid 2-adic step. After substituting t=7 into (29), the text obtains 24 = (y+5)v_2(y) and says this forces y even and y+5 | 24, hence y ∈ {7,19}. That inference is wrong: if v_2(y) ≥ 1 then y is even, so y+5 is odd; the only odd divisors of 24 are 1 and 3, and y+5 > 7 rules them out. Moreover y=7 and y=19 are odd and do not satisfy the valuation equation. No integer y > 2 solves it, so the contradiction is actually cleaner than written, but the printed intermediate step is invalid. This leaves the stated part of Theorem 1.7 about the phase transition endpoints unsupported as written, though a parity argument repairs it and the numerical table makes the claim plausible.\n\nThis defect does not touch the main inequality, the equality characterization, or the asymptotic comparison formula. It affects only the assertion that no integer k gives a tie. The citations look standard and appropriate; I see no self-citation padding or invented entities. The paper is honest about its constants not being optimal.\n\nWho gets value: extremal set theory and spectral hypergraph people. It deserves a serious referee, not a desk reject. I would send it out, with a note to the authors to fix the appendix rather than treating the flaw as fatal.","headline":"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.","tokens_in":18389,"tokens_out":1609,"would_cite":true,"duration_ms":15020,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C50","05D05","05C65","15A69"],"pacs":[],"model":"deepseek-v4-flash","headline":"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 $\\mathcal A_{n,k,t}$…","keywords":["t-intersecting family","Hilton-Milner-Frankl theorem","spectral radius","adjacency tensor","hypergraph","extremal set theory","phase transition","covering kernel"],"falsifier":"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.","tokens_in":17408,"feed_emoji":"🕸️","tokens_out":16443,"duration_ms":142709,"temperature":0.7,"pith_summary":"The paper proves a spectral analogue of the Hilton--Milner--Frankl theorem for nontrivial $t$-intersecting $k$-uniform families. A family is $t$-intersecting if any two of its edges share at least $t$ vertices, and nontrivial if it is not contained in a full $t$-star. The main theorem states that, for $1 \\le t \\le k-2$ and $n \\ge 100\\cdot 2^k k^7$, the adjacency-tensor spectral radius of any such family is at most the larger of the two explicit constructions $\\mathcal H_{n,k,t}$ and $\\mathcal A_{n,k,t}$, and equality forces the family to be isomorphic to one of them. It also shows that, asymptotically, which of the two wins is decided by a unique non-integer transition point $\\kappa_t$ defined by a simple exponential equation. This is the spectral counterpart of the classical size extremal theorem, replacing edge counts by spectral radius and identifying exactly the same two extremal families.","feed_headline":"Two families cap every nontrivial t-intersecting k-graph","feed_subtitle":"A spectral Hilton–Milner–Frankl theorem with an explicit phase transition between the two extremal families.","key_machinery":"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.","core_discovery":"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 $k \\ge \\lceil\\kappa_t\\rceil$.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"Supplies the spectral Erdős--Ko--Rado theorem and the edge-count bound used as the baseline and transfer method.","marker":"[16]"},{"why":"Supplies the complete nontrivial-intersection theorem that identifies the two extremal candidate families.","marker":"[1]"},{"why":"Provides the complete intersection theorem that frames the full $t$-star extremal problem in the full parameter range.","marker":"[2]"},{"why":"Gives the original Hilton--Milner theorem for $t=1$ whose family construction is one of the two candidates.","marker":"[13]"},{"why":"Introduces the adjacency-tensor spectral radius and the Perron--Frobenius framework for uniform hypergraphs.","marker":"[4]"},{"why":"Supplies the strict monotonicity principle for weakly irreducible tensors used to rule out proper subfamilies at equality.","marker":"[14]"},{"why":"Gives Wilson's exact bound for $t$-intersecting families, used to justify the large-$n$ regime.","marker":"[21]"}],"fun_headline_variants":["Two spectra cap every nontrivial t-intersecting family","Spectral HMF theorem with explicit phase transition between two families","Phase transition determines spectral extremal t-intersecting family","Exact spectral extremal families for nontrivial t-intersecting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Two spectra cap every nontrivial t-intersecting family","Spectral HMF theorem with explicit phase transition between two families","Phase transition determines spectral extremal t-intersecting family","Exact spectral extremal families for nontrivial t-intersecting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000879,"raw_usage":{"total_tokens":3939,"prompt_tokens":1223,"completion_tokens":2716,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":839,"completion_tokens_details":{"reasoning_tokens":2644}},"tokens_in":839,"tokens_out":2716,"duration_ms":17622,"temperature":1.0,"reasoning_tokens":2644,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:15:53.289383+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":"Keevash, J","cited_arxiv_id":null,"evidence_quote":"Supplies the spectral Erdős--Ko--Rado theorem and the edge-count bound used as the baseline and transfer method."},{"cited_title":"Ahlswede and L","cited_arxiv_id":null,"evidence_quote":"Supplies the complete nontrivial-intersection theorem that identifies the two extremal candidate families."},{"cited_title":"Ahlswede and L","cited_arxiv_id":null,"evidence_quote":"Provides the complete intersection theorem that frames the full $t$-star extremal problem in the full parameter range."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the original Hilton--Milner theorem for $t=1$ whose family construction is one of the two candidates."},{"cited_title":"Rajesh Kannan, N","cited_arxiv_id":null,"evidence_quote":"Supplies the strict monotonicity principle for weakly irreducible tensors used to rule out proper subfamilies at equality."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives Wilson's exact bound for $t$-intersecting families, used to justify the large-$n$ regime."}],"review_version":1}