{"id":"7859d1df-e726-4cca-a86f-932d6f31f23f","arxiv_id":"2608.07342","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For every positive integer m there is a matroid whose rank-by-rank counts of flats have exactly m peaks, disproving Rota's unimodality conjecture and Mason's log-concavity conjecture.","lead":"The authors disprove Rota's 1971 conjecture that the counts of flats at each rank in any matroid always rise and then fall like a single hill. They construct matroids whose flat counts instead have any chosen number of peaks, settling a long-open question in combinatorics.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The truncation step in Theorem 1.2 is asserted rather than proved: the paper shows at least m+1 peaks, then claims truncation can reduce to exactly m without verifying the peak count of the truncated sequence.","rationale":"The central construction is sound: Lemma 2.1 and Lemma 2.4 give the necessary asymptotics, and the q-lift calculation correctly converts log-concavity failures into strict local maxima. The only substantive gap is the truncation step, which is asserted in a single sentence. It is elementary to fill—repeated truncations delete a prefix of the corank sequence, and for the constructed sequence the k=m+1-fold truncation leaves exactly the m peaks at old corank positions m+3,...,3m+1—but the paper should state this. The external Lemma 3.1 is published and correctly applied. No ad hominem, no circularity, no parameter-fitting. The theorem's central claim survives the scrutiny.","tokens_in":9105,"tokens_out":53780,"duration_ms":426623,"concrete_test":"For a fixed m (e.g., m=2), take t large, p a prime with p=Θ(t^{3/2}), and inspect the k-fold truncation for k=m+1. Analytically verify the inequalities: (i) W_0=1 is smaller than the first surviving element W_{m+2}; (ii) W_{m+2} is smaller than the following old peak W_{m+3}; (iii) each surviving peak W_{m+2s+3} for s=0..m-1 satisfies W_{m+2s+3} > W_{m+2s+2} and W_{m+2s+3} > W_{m+2s+4} using the ratios from Lemma 2.4 with p=Θ(t^{3/2}). If all three hold, the truncated matroid has exactly m peaks and Theorem 1.2 is fully justified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 1.2 establishes that the p-lift of G_t^{⊕m} has peaks at corank η=m+2s+1 for s=0..m, giving at least m+1 peaks. To reach exactly m peaks, the proof invokes truncation: omitting the W_1 term reduces the peak count by at most one, 'and that we can thereby achieve any desired number of peaks.' This is the only step that converts 'arbitrarily many peaks' into the stated 'exactly m' claim, yet it is not demonstrated. In the constructed sequence, truncation deletes a prefix W_1,...,W_k of the corank sequence, and one must check that the surviving old peaks remain strict local maxima at the new boundary and that no new peak appears at W_0 or at the first surviving valley. For the specific sequence, the check works: after k=m+1 truncations the surviving sequence is (W_0, W_{m+2},...,W_{r+1}), which contains exactly the m old peaks at m+3,...,3m+1. But this verification is absent; without it, the theorem as written only proves existence of matroids with arbitrarily many peaks, not with a prescribed exact number.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper disproves Rota's conjecture that the Whitney numbers of the second kind of a matroid form a unimodal sequence. The main theorem, Theorem 1.2, asserts that for every positive integer m there is a matroid whose flat counts by rank have exactly m peaks. The proof combines asymptotic estimates for the flat counts of a family of generalized theta graphs G_t (Lemma 2.1), a convolution analysis for direct sums (Lemma 2.4), and Whittle's q-lift together with Bonin and Qin's formula for the flat counts of the lift (Lemma 3.1). The authors also exhibit concrete small matroids for which log-concavity and unimodality fail.","tokens_in":77,"tokens_out":29901,"duration_ms":721067,"significance":"If the construction is correct, it refutes a fifty-year-old conjecture of Rota and, because log-concave sequences with no internal zeroes are unimodal, it also gives counterexamples to Mason's log-concavity conjecture for Whitney numbers of the second kind. The paper is largely self-contained and rests on published external results, namely Whittle's q-lift and Bonin-Qin's flat-count formula; the asymptotic arguments in Section 2 are carefully presented. The main weakness is the truncation step at the end of the proof of Theorem 1.2, which is asserted rather than proved; this step is load-bearing for the 'exactly m peaks' formulation, but the missing verification is elementary and local.","major_comments":[{"comment":"The truncation step is load-bearing and is not proved. The proof establishes that the p-lift has at least m+1 peaks, and then asserts that truncation reduces the number of peaks by at most one, 'and that we can thereby achieve any desired number of peaks.' This assertion needs a proof: one must show that deleting a single entry from a sequence never increases the number of peaks (under the plateau convention of Definition 1), that the peak count eventually drops to 0 under repeated truncation, and that for the specific sequence constructed here the surviving old peaks remain strict local maxima and no new peaks appear at the boundary. Concretely, after k=m+1 truncations the corank sequence becomes (W_0, W_{m+2}, W_{m+3}, ..., W_r), and the old peaks at original coranks m+3, m+5, ..., 3m+1 survive; this verification is absent. Without it, the theorem as written only proves the existence of matroids with arbitrarily many peaks, not with a prescribed exact number.","section":"Section 3, proof of Theorem 1.2, final paragraph"}],"minor_comments":[{"comment":"The observation that a sequence is log-concave with no internal zeroes 'if and only if' every c-weighted sequence is unimodal is false as stated. For example, (1,2,5) is not log-concave, but for every c>0 the sequence (1, 2c, 5c^2) is increasing and hence unimodal. The proof of the main theorem does not use this claim, so the authors should remove it or replace it with a correct statement.","section":"Section 1, paragraph after Theorem 1.2"},{"comment":"In the estimate for p^{-ℓ} \\tilde W_{m+2s+2}/W_{m+2s}, the term p^{-ℓ} O(t^3) is exponentially small because ℓ is linear in t, so writing the result as Θ(t^{-1/2}) is misleading. This should be stated as o(1) or O(t^{-1/2}) with an explicit note.","section":"Section 3, proof of Theorem 1.2, third display"},{"comment":"The exact values of the Whitney numbers in these examples are stated without derivation or a reproducibility statement. Since the examples are used to support the 'smallest' claims, a computation outline or accompanying code would improve verifiability.","section":"Examples 2.2 and 3.2"},{"comment":"The sentence about omitting the W_1 term is only correct in corank indexing, which the paper introduced in Section 2 but does not repeat here. This should be clarified to avoid confusion with the rank-indexed W_i used in the introduction.","section":"Section 3, truncation paragraph"},{"comment":"The reference '[DL W26]' contains an unintended space in the author initials; it should read '[DLW26]'.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The main construction is convincing and the counterexample to Rota's conjecture appears to be correct. My concern is the truncation step in the proof of Theorem 1.2, which is essential for the exact-m formulation and is currently asserted rather than demonstrated. I expect the gap can be closed by adding a short lemma about deleting one entry from a sequence and a direct verification for the constructed sequence. If the authors supply that, along with the small clarifications above, I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is the real thing: a disproof of Rota's 1970 conjecture that matroid flat counts are unimodal, and it goes further by showing the failure can be arbitrarily large. The two-step architecture is clean. Lemma 2.1 gives sharp asymptotic counts for the generalized theta graph matroid G_t; Lemma 2.4 computes the direct-sum convolution; and the q-lift argument converts the asymptotic log-concavity violations into strict local maxima. The Bonin–Qin formula is cited properly, and the use of Whittle's q-lift is careful about the representation-independence caveat. Example 3.2 even supplies an explicit 4957-element matroid with non-unimodal flat counts, which alone refutes the conjecture. The core asymptotic proof for 'arbitrarily many peaks' is solid and checkable.\n\nThe soft spot is the final truncation step in the proof of Theorem 1.2. After the q-lift construction, the paper has shown at least m+1 peaks. To get exactly m peaks, it invokes truncation and says 'it is clear that the number of peaks goes down by at most one' when the W_1 term is omitted. That claim is not true for arbitrary sequences: e.g., (1,5,5,4,6,1) has one peak (at the 6), while omitting the W_1 term gives (1,5,4,6,1), which has two peaks. So the blanket assertion fails. The paper doesn't supply the verification for its specific sequence. In this particular case the check does go through: apply m+1 truncations, the first peak at corank m+1 disappears, the remaining m peaks at m+3,...,3m+1 survive, and no new peak appears because the new boundary entry is a valley. But none of that appears in the text. As written, Theorem 1.2 is only proved for 'at least m+1 peaks,' not 'exactly m peaks,' unless the reader fills in the missing argument.\n\nThis is a genuine but small gap. It does not threaten the main counterexample to Rota's conjecture, which is established by Example 3.2 and by the arbitrarily-many-peaks construction. A referee should ask the authors to replace the hand-wave with a concrete truncation count. I'd send this to peer review and expect acceptance after that fix.\n\nWho for: matroid theorists and combinatorial geometers; anyone tracking Mason's conjectures. I would cite it once the truncation step is repaired.\n\nRegards.","headline":"Solid disproof of Rota's conjecture with a clean construction, but the proof of the 'exactly m peaks' theorem leans on an unproven and in general false truncation claim that needs a small fix.","tokens_in":9876,"tokens_out":15095,"would_cite":true,"duration_ms":115144,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B35"],"pacs":[],"model":"deepseek-v4-flash","headline":"Flat counts of matroids can have arbitrarily many peaks","keywords":["matroid","Whitney numbers of the second kind","flat counts","unimodality","log-concavity","q-lift","generalized theta graph","counterexample"],"falsifier":"Enumerate the flats of the rank-82 matroid from Example 3.2 and compare the exact counts at ranks 78, 79, and 80; if the middle count is not strictly larger than both neighbors, the claimed failure of unimodality for that example is wrong.","tokens_in":8868,"feed_emoji":"📈","tokens_out":10249,"duration_ms":74836,"temperature":0.7,"pith_summary":"Matroids are combinatorial structures that capture independence, and their flats—closed sets—can be counted by rank. This paper disproves the long-standing conjecture, posed at the 1970 International Congress of Mathematicians, that those rank-by-rank counts always rise to a single maximum and then fall. It proves the opposite: for every positive integer $m$, there is a matroid whose flat counts have exactly $m$ peaks. The construction finds generalized theta graphs where the counts fail log-concavity badly, takes direct sums to spread the failure across many ranks, and then applies the $q$-lift operation to turn those failures into strict peaks. The result settles the unimodality conjecture negatively and, as a direct consequence, also refutes the weaker log-concavity conjecture for these numbers.","feed_headline":"Flat counts of matroids can have arbitrarily many peaks","feed_subtitle":"A theta-graph and q-lift construction produces matroids with exactly m peaks in their rank-by-rank flat counts.","key_machinery":"The $q$-lift is the central construction: given a simple matroid represented by vectors in $\\mathbb{F}_q^r$, the lift adds one new vector and places scaled copies $(z,v)$ for all $z\\in\\mathbb{F}_q$, producing a rank-$(r+1)$ matroid over $\\mathbb{F}_q$. Its flat counts obey the identity $\\widetilde W_i=q^i W_i+W_{i-1}$, which is the mechanism that converts a severe failure of log-concavity in the original sequence into a strict peak after scaling by powers of $q$. The other half of the machinery is the asymptotic analysis of $G_t^{\\oplus m}$, whose corank-$k$ counts have exponents $\\lfloor 3(k+m)/2\\rfloor$ in the middle range; these exponents are what locate the peaks.","core_discovery":"The central claim is that the Whitney numbers of the second kind—the counts $W_i$ of flats of rank $i$—need not form a unimodal sequence. The proof fixes $m$, takes the direct sum of $m$ copies of the graphic matroid of a generalized $\\theta$ graph with four internally disjoint paths (three of length $t$ and one single edge), and forms a $p$-lift for a prime $p$ chosen so that $p=\\Theta(t^{3/2})$. A quoted identity for $q$-lifts gives $\\widetilde W_i=p^i W_i+W_{i-1}$ for the lifted flat counts, and asymptotic estimates for the direct sum show that the indices $m+2s+1$ for $0\\le s\\le m$ are strict peaks. Truncation then removes peaks one at a time, yielding exactly $m$ peaks for any prescribed $m$.","pith_inferences":["The same $q$-lift recipe should turn other families with known asymptotic exponents, such as the cothickened cocircuit examples mentioned in the paper, into further non-unimodal matroids that are not necessarily graphic.","The gap between the 78-element log-concavity failure and the 4957-element unimodality failure suggests the $q$-lift is the size bottleneck; using larger fields or theta graphs with more strands may shrink the examples substantially.","The method leaves the points-lines-planes inequality, which concerns only the first three entries of the flat-count sequence, untouched; a counterexample to that inequality, if one exists, would have to come from a different construction."],"forward_implications":["The unimodality conjecture for matroid flat counts is false, and in fact the number of peaks is unbounded.","The log-concavity conjecture for Whitney numbers of the second kind is also false, because any log-concave sequence with no internal zeros is unimodal.","For every positive integer $m$, there is an explicit matroid with exactly $m$ peaks; the method also gives a non-unimodal matroid on 4957 elements and a non-log-concave one on 78 elements.","Direct sums break log-concavity at many places simultaneously, so the failure is not confined to one rank in a specially chosen family.","The construction is representable over every field because the theta graphs are regular, so any future proof that flat counts are unimodal for representable matroids would have to leave these examples behind."],"supporting_citations":[{"why":"supplies the identity $\\widetilde W_i=q^iW_i+W_{i-1}$ for flat counts of any $q$-lift, the key step that turns log-concavity failures into peaks.","marker":"[BQ01]"},{"why":"introduces the $q$-lift operation that the construction applies to the direct sums of theta graphs.","marker":"[Whi89]"},{"why":"states the unimodality conjecture for matroid flat counts that the paper refutes.","marker":"[Rot71]"},{"why":"states the stronger log-concavity conjecture that is refuted as a direct consequence.","marker":"[Mas72]"},{"why":"provides the standard matroid conventions, the dual rank formula used in the asymptotics, and the truncation operation used to tune the number of peaks.","marker":"[Oxl11]"}],"fun_headline_variants":["Matroid flat counts break unimodality, can peak many times","Many peaks in matroid flat counts: Rota's conjecture false","Arbitrarily many peaks in rank-wise flat counts of matroids","Theta graph and q-lift give matroids with many flat-count peaks","Matroids with exactly m peaks in flat counts for any m"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction rests on the quoted formula for how many flats a $q$-lift has in each rank, and on the elementary, unproved fact that omitting the $W_1$ term from a sequence can destroy at most one peak.","fun_headline_variants_meta":{"raw":{"variants":["Matroid flat counts break unimodality, can peak many times","Many peaks in matroid flat counts: Rota's conjecture false","Arbitrarily many peaks in rank-wise flat counts of matroids","Theta graph and q-lift give matroids with many flat-count peaks","Matroids with exactly m peaks in flat counts for any m"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000576,"raw_usage":{"total_tokens":2643,"prompt_tokens":793,"completion_tokens":1850,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":409,"completion_tokens_details":{"reasoning_tokens":1758}},"tokens_in":409,"tokens_out":1850,"duration_ms":11974,"temperature":1.0,"reasoning_tokens":1758,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T05:57:45.578283+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate the flats of the rank-82 matroid from Example 3.2 and compare the exact counts at ranks 78, 79, and 80; if the middle count is not strictly larger than both neighbors, the claimed failure of unimodality for that example is wrong.","supporting_citations":[],"review_version":1}