{"id":"37229340-e7a4-4c4d-b7b5-051751a64f16","arxiv_id":"2412.12466","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For all even n at least 10 there exist latin squares with at least floor(n/6) entries common to every transversal and more than half of all entries in no transversal; for orders 3m, every transversal hits all nine equal subsquares.","lead":"This paper constructs latin squares of every even size at least 10 where all transversals share at least one sixth of the entries, even though transversals exist. It also builds odd-order squares divided into nine equal blocks where every transversal must touch every block.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the inclusion–exclusion bookkeeping in Lemmas 2.6–2.8 is intricate but survives symbolic and small-case spot-checks.","rationale":"The Reader's weakest-assumption diagnosis is correct: the inclusion–exclusion sections are the most likely place for an unnoticed off-by-one or a missed wrap-around exception, and they are not machine-checked. My stress-test targeted that spot. I re-derived the main counting identities for representative k in all three residue classes and found the stated bounds reproduced exactly. The small-k anomaly in Lemma 2.8 is a real defect in the exposition—negative cardinalities are asserted as equalities—but it is used only as a lower bound and therefore does not threaten the theorem. The pinned-entry argument via the Delta-sum identity is elementary and sound, and the transversals exhibited in Lemmas 2.3–2.5 check out as transversal via the column/symbol set decompositions. The odd-order construction in Theorem 1.3, after resolving the intended modular interpretation of a+b in the block definition, is internally consistent; the automorphisms and the symbol-availability argument in Lemmas 3.1–3.2 are coherent. Overall, no load-bearing concern lands, and the ACCEPT verdict with moderate confidence remains appropriate.","tokens_in":18778,"tokens_out":39910,"duration_ms":338905,"concrete_test":"Implement T_n, U_n, V_n exactly as in (2), (4), (6), together with the sets M,N,O,P,Q,R defined in Lemmas 2.6–2.8, and for all even n from 10 to 1000 compute |M∪N∪O| directly; verify that it reaches the claimed bound for each residue class modulo 6 and that every stated pairwise and triple intersection count in Lemmas 2.6–2.8 matches the direct count.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I found no load-bearing flaw. The fragile point of the paper is exactly the one the Reader flags: the lower bounds in Theorem 1.2 depend on the exhaustive exceptional-entry lists (9)–(15) and on the exact sizes of the triple intersections P∩N∩O, Q∩N∩O, R∩N∩O in Lemmas 2.6–2.8. I checked the arithmetic for representative parameter values in each congruence class (n=12,14,18,20,24,26,28,32) and the counts reproduce the stated lower bounds; Table 1 provides additional support. The only genuine blemish I found is purely presentational: in Lemma 2.8, for small k the displayed formulas for |Q∩N∩O| and |R∩N∩O| evaluate to negative integers (for k=2 they are −1 and −2), and the text writes these as equalities. Since these quantities are subsequently used only through the lower bounds `floor((k−3)/2) ≥ (k−4)/2` and `k−4`, the inequality |M∪N∪O| ≥ (19n²−73n−182)/36 remains valid; the equalities are false for k<4 but harmless. No error affecting the 19/36 coefficient was found.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs, for every even n ≥ 10, a latin square with at least floor(n/6) pinned entries, and at least (19/36)n² + O(n) entries that lie in no transversal. For every odd m ≥ 3 it constructs an order 3m latin square divided into nine m × m subsquares such that every transversal hits each subsquare. The even-order proofs split into the congruence classes n ≡ 0, 2, 4 mod 6, using the Δ-function sum invariant (Lemma 2.1) to force the maximum-Δ entries of selected rows into every suitable diagonal, and then exhibiting explicit transversals. The transversal-free count is obtained by inclusion-exclusion over the rows missing their maximum Δ, together with carefully chosen column and symbol sets N and O. Section 3 introduces a block construction and uses a mod m analogue of the Δ-sum plus automorphisms to prove that every transversal meets every block.","tokens_in":74,"tokens_out":8779,"duration_ms":540434,"significance":"If the constructions are valid, the results are a clear improvement over the earlier one-pinned-entry theorem of Egan and Wanless and over previous bounds on transversal-free entries in squares that have transversals. The proofs are constructive and elementary, and the claimed numbers are concrete and falsifiable, with Table 1 providing actual counts for small orders that confirm the stated lower bounds. The main theorems are built on the simple Δ-sum lemma, which makes the paper accessible; the bookkeeping in Lemmas 2.6–2.8 is intricate but, as far as I could verify by spot checks, consistent. The odd-order block result is novel and its use of automorphisms to propagate the hitting property is elegant.","major_comments":[{"comment":"The arrays T_n, U_n, V_n, and L_n are called latin squares, but the paper never proves that each row and column contains every symbol exactly once. This is load-bearing: Lemma 2.6 explicitly invokes the \"latin property\" to compute |N ∩ O|, and Theorem 1.2 is about latin squares. Please add a verification (or a precise reference) that each of these arrays is a latin square. If the verification is intended to be routine, it should at least be stated explicitly and sketched, because the piecewise definitions involve many exceptional cases.","section":"§2, definitions (2), (4), (6); §3, definition of L_n"},{"comment":"The lower bound in Theorem 1.2 rests on the assertion that the exceptional-entry lists (9)–(15) are exhaustive and on the stated sizes of the triple intersections P ∩ N ∩ O, Q ∩ N ∩ O, and R ∩ N ∩ O. The text often says \"the structure reveals\" or \"one can see\" for these counts. I did not find an actual error in representative cases, but given how intricate the definitions are, the proof should either expand these derivations into a verifiable case analysis or supply an independent check (for example, a short computer verification for each congruence class) so that completeness is not left to the reader.","section":"§2, Lemmas 2.6–2.8"}],"minor_comments":[{"comment":"The displayed equalities |Q ∩ N ∩ O| = floor((k−3)/2) and |R ∩ N ∩ O| = k−4 are false for small k: for k = 2 they give −1 and −2, which are impossible for cardinalities. Since the proof only uses the subsequent lower bounds (k−4)/2 and k−4, the final bound is unaffected, but the equalities should be stated as lower bounds, or restricted to k ≥ 4 with the small cases handled separately.","section":"§2, Lemma 2.8, Subcases 3.b and 3.c"},{"comment":"In the displayed 8 × 8 square, the first row is broken across two lines in the typeset version, making it look like a row of length 4. Please format the full matrix on a single line or in a way that each row is visibly one of the eight rows.","section":"§2, Lemma 2.8"},{"comment":"The phrase \"by a similar argument to Lemma 2.1\" is a little terse for the mod m function Δ_m, because Lemma 2.1 concerns Δ modulo n and gives n/2 for even n, whereas here the right-hand side is 0 modulo m. A one-sentence derivation of ∑(r+c) = n²(n−1) ≡ 0 (mod m) would make the step fully explicit.","section":"§3, equation (16)"},{"comment":"The condition n² > τ(L) is equivalent to the existence of at least one transversal; since this is not stated explicitly, a brief parenthetical remark would help readers who might otherwise wonder why the strict inequality appears alongside the lower bound.","section":"§1, Theorem 1.2"}],"recommendation":"major_revision","confidential_remarks":"The core results appear to be correct and the paper is likely to be accepted after revision. My main request is the missing proof that the constructed arrays are latin squares; this should be easy to supply. The small-k issue in Lemma 2.8 is purely presentational and does not threaten the theorem. I would support acceptance once the latin property and the inclusion-exclusion bookkeeping are made fully verifiable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is a genuine step forward on an old question: how much can transversals of a latin square be forced to overlap? The authors improve the known pinned-entry guarantee from one entry to floor(n/6) for even n ≥ 10, and show the same squares have at least 19n^2/36 + O(n) entries that lie in no transversal. That is the first even-order family with more than half of the entries transversal-free while still having a transversal. The odd-order result—every transversal hits each of the nine m×m subsquares in a constructed 3m×3m square—is new and neatly proved.\n\nWhat I like best is that the proofs are constructive and self-contained. The Delta-lemma from Evans and Wanless-Webb is elementary, and the paper applies it cleanly: in each congruence class mod 6, the maximum-Delta argument forces the pinned entries, then an explicit transversal is exhibited. The transversal-free counting uses inclusion-exclusion over sets P, Q, R, N, O. That part is intricate—dozens of congruence subcases, exceptional entry lists (9)–(15), and triple intersection counts. The stress-test note is right: the displayed formulas for |Q∩N∩O| and |R∩N∩O| in Lemma 2.8 evaluate to −1 and −2 for k=2, so the equalities are not literally true for small k. But the proof only uses the weaker inequalities, so the bound survives. That is a presentational bug, not a mathematical one, and the small-order table (Table 1) independently supports the bounds.\n\nThe only genuinely soft spot is the burden that carries: the 19/36 coefficient depends on the exceptional lists being exhaustive and the triple intersection counts being exact. The paper gives enough detail to check, and the spot-checks pass, but this is exactly the kind of counting that could hide a small error. A serious referee should verify a few of the exceptional lists from the square definitions, and ask the authors to fix the k<4 formulas in Lemma 2.8. The open problems at the end are honest and relevant. Self-citation is present but not circular; Lemma 2.1 is standard and cited independently.\n\nWho is this for? Researchers in combinatorics working on transversals, latin squares, or SDRs. It is a solid advance, not a blockbuster, but the constructions are reusable and the barrier it breaks (more than half transversal-free with a transversal present) is real. I would send it to a serious referee. If I were refereeing, I'd recommend accept after minor revisions.","headline":"Solid constructive advance: floor(n/6) pinned entries and more than half transversal-free entries for infinitely many even orders; the inclusion-exclusion is intricate but checks out aside from a harmless small-k blemish in Lemma 2.8.","tokens_in":19562,"tokens_out":2090,"would_cite":true,"duration_ms":17933,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs latin squares whose transversals all share n/6 entries.","keywords":["latin squares","transversals","pinned entries","transversal-free entries","suitable diagonals","Delta function","inclusion-exclusion counting","block subsquares"],"falsifier":"Take the smallest constructed representatives of the three residue classes, $T_{12}$, $V_{16}$, and $U_{20}$, and exhaustively enumerate all suitable diagonals or all transversals. The table reports exact transversal-free counts 67, 107, and 190; if any claimed pinned entry is missing from a suitable diagonal, or if the true count is lower than reported, the inclusion-exclusion lemmas behind Theorem 1.2 would be wrong.","tokens_in":18584,"feed_emoji":"🧩","tokens_out":12344,"duration_ms":101186,"temperature":0.7,"pith_summary":"This paper builds explicit latin squares for every even order $n\\ge 10$ in which all transversals pass through the same $\\lfloor n/6 \\rfloor$ entries, while at least $19n^2/36+O(n)$ entries lie in no transversal at all. If the construction is right, it shows that a latin square can have transversals and yet force all of them to agree on a linearly growing set of cells. The same ideas produce squares of odd order $3m$, split into nine $m\\times m$ blocks, where every transversal must touch each block. These are the best current examples of how much a transversal spectrum can be forced to share while still containing at least one transversal.","feed_headline":"New latin squares force all transversals through the same n/6 entries","feed_subtitle":"For every even order n≥10, explicit squares leave at least 19/36 of cells outside all transversals.","key_machinery":"The central machinery is the integer-valued function $\\Delta(r,c,s) \\equiv s-r-c \\pmod n$, together with the $\\Delta$ Lemma: the sum of $\\Delta$ over any transversal is $0$ modulo $n$ when $n$ is odd and $n/2$ modulo $n$ when $n$ is even. For even $n$, a suitable diagonal is any set of $n$ entries, one from each row and column, whose $\\Delta$-sum is $n/2$; since every transversal is a suitable diagonal, forcing an entry into every suitable diagonal forces it into every transversal. The construction makes the row-wise maxima of $\\Delta$ strict and makes their total exactly $n/2$, so every suitable diagonal must take the maximum from every row, hitting the listed pinned entries. The transversal-free count then comes from inclusion-exclusion over the columns and symbols that the proof forces to avoid all suitable diagonals.","core_discovery":"For orders $n\\equiv 0,2,4 \\pmod 6$, the paper writes down explicit cyclic latin squares $T_n$, $U_n$, and $V_n$. On each square it defines $\\Delta(r,c,s)$ and observes that every transversal must be a suitable diagonal. The squares are engineered so that the sum of the maximum $\\Delta$-values over all rows is exactly $n/2$; a suitable diagonal must therefore take the maximum from every row, and the entries listed in (3), (5), and (7) are exactly those maxima. Those $\\lfloor n/6 \\rfloor$ entries are pinned. Explicit transversals are listed, so the squares are not transversal-free. Inclusion-exclusion over carefully chosen columns and symbols then certifies at least $19n^2/36+O(n)$ entries that no transversal can use. For odd $n=3m$, a separate block construction with $\\Delta_m(r,c,s)\\equiv r+c \\pmod m$ forces every transversal to hit all nine $m\\times m$ subsquares.","pith_inferences":["The authors report that their bounds are within $O(n)$ of the true transversal-free counts; if that belief holds, the three constructed families have transversal-free density tending to exactly $19/36$, an asymptotic statement not asserted in the theorems.","The row-maximum forcing scheme is modular: any even-order square whose row-wise $\\Delta$-maxima are strict and sum to $n/2$ produces pinned entries, so perturbing the construction might push the pinned count above $\\lfloor n/6 \\rfloor$ while staying below the $n-3$ ceiling.","The odd-order block result is one step short of pinned entries: refining the block-hitting invariant to force a specific cell in each block would yield the first pinned entry for a large odd order, a case where none is currently known."],"forward_implications":["For every even $n\\ge 10$, there is a latin square with at least one transversal but with at least $\\lfloor n/6 \\rfloor$ entries common to all transversals.","Those squares have at least $19n^2/36+O(n)$ transversal-free entries, so for $n\\ge 88$ more than half the cells are outside every transversal.","For every odd $m\\ge 3$, the order-$3m$ block square has every transversal meeting each of the nine $m\\times m$ subsquares.","The construction improves the previously known one-pinned-entry result for even orders to a pinned set whose size grows linearly with $n$.","Combined with the parity of the number of transversals, the examples show that the pinned-entry number can be forced to grow while remaining well below the $n-3$ ceiling."],"supporting_citations":[{"why":"Supplies the earlier weaker pinned-entry theorem and the previous family with more than a constant fraction of transversal-free entries, which this paper extends.","marker":"[4]"},{"why":"One of the two independent sources of the Delta Lemma that every transversal has a fixed sum of Delta values.","marker":"[6]"},{"why":"The other independent source of the Delta Lemma, jointly providing the key tool behind suitable diagonals.","marker":"[7]"},{"why":"Gives the fact that even-order latin squares have an even number of transversals, used in the discussion of pinned-entry limits.","marker":"[1]"},{"why":"Used to state the $n-3$ upper bound on pinned entries in even-order squares.","marker":"[2]"}],"fun_headline_variants":["All transversals share n/6 entries for every even order n","New squares force transversals to coincide on n/6 cells","Even n: all transversals agree on n/6 entries exactly","Explicit squares pin down n/6 common transversal entries","Odd order: every transversal hits all nine subsquares"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof depends on a long count of which cells are forced to be transversal-free, and that count must be complete; if overlooked exceptions change the totals by more than a constant, the claim that at least $19n^2/36$ cells are transversal-free fails.","fun_headline_variants_meta":{"raw":{"variants":["All transversals share n/6 entries for every even order n","New squares force transversals to coincide on n/6 cells","Even n: all transversals agree on n/6 entries exactly","Explicit squares pin down n/6 common transversal entries","Odd order: every transversal hits all nine subsquares"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000438,"raw_usage":{"total_tokens":2179,"prompt_tokens":854,"completion_tokens":1325,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":1240}},"tokens_in":470,"tokens_out":1325,"duration_ms":11667,"temperature":1.0,"reasoning_tokens":1240,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T14:03:19.818598+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the smallest constructed representatives of the three residue classes, $T_{12}$, $V_{16}$, and $U_{20}$, and exhaustively enumerate all suitable diagonals or all transversals. The table reports exact transversal-free counts 67, 107, and 190; if any claimed pinned entry is missing from a suitable diagonal, or if the true count is lower than reported, the inclusion-exclusion lemmas behind Theorem 1.2 would be wrong.","supporting_citations":[{"cited_title":"Egan and I","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier weaker pinned-entry theorem and the previous family with more than a constant fraction of transversal-free entries, which this paper extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"One of the two independent sources of the Delta Lemma that every transversal has a fixed sum of Delta values."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The other independent source of the Delta Lemma, jointly providing the key tool behind suitable diagonals."},{"cited_title":"Balasubramanian, On transversals in latin squares, Linear Algebra Appl","cited_arxiv_id":null,"evidence_quote":"Gives the fact that even-order latin squares have an even number of transversals, used in the discussion of pinned-entry limits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Used to state the $n-3$ upper bound on pinned entries in even-order squares."}],"review_version":1}