Pith. sign in

REVIEW 2 major objections 3 minor 11 references

Latin Squares whose transversals intersect in unusual ways

T0 review · 2 major / 3 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read For every even order n from 28 to 10,000 (except 30), there is a latin square in which any two transversals share an entry, yet no single entry belongs to all transversals.

desk verdict Clever delta-forcing proofs make the no-disjoint-transversals half rigorous; the no-pinned-entry half of Theorem 6 waits on unreported search certificates. read the letter →

arxiv 2607.17547 v1 pith:CNN6AM27 submitted 2026-07-20 math.CO

Afsane Ghafari , Ian M. Wanless This is my paper · ORCID
classification math.CO MSC 05B1505D15
keywords latinsquarestransversalspinnedentriespairwise-intersectingdominanttransversaldelta-sumrulerandomizedhill-climbingrainbowmatchings
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish a new extreme in how constrained the transversals of an even-order latin square can be. It constructs squares, for n=28 and every even n between 32 and 10,000, where every pair of transversals intersects—so no two transversals can be taken together—while no single entry is 'pinned' in all of them. The construction is explicit: two families G_n and H_n, plus special squares for orders 28, 32 and 34, and the proof that no two transversals are disjoint uses a simple modular sum rule. The no-pinned-entry part, however, is checked only by a randomized hill-climbing search up to n=10,000. In a separate line, the paper proves that for every n>4 with n not congruent to 3 mod 4 and n≠6, some latin square of order n has a 'dominant' transversal that meets every other transversal.

What carries the argument

The Δ-function Δ(r,c,s) ≡ s−r−c (mod n), lifted to integers between −n/2 and n/2, and the 'delta-sum rule' that a transversal must have Δ-sum 0 (n odd) or n/2 (n even) modulo n. The authors design squares whose entries have Δ-values in a narrow band, with only a few entries at the extreme +3. The sum rule then forces any transversal to pick at least two of these extreme entries, so any two transversals collide at one of them. A hill-climbing procedure that swaps two columns while tracking unused symbols and excess rows is used to find the specific transversals witnessing that no entry is universal.

What would settle it

Run the hill-climbing procedure on G_100 with required/forbidden sets as specified in Section 3 and try to find four transversals: one missing each of (1,2,6), (6,6,15), (11,10,24) while containing the other two, plus one avoiding (16,12,29) but containing all three; verify the four have empty common intersection. If for any even n in [36,10000] no such set exists, Theorem 6 fails. For n=28,32,34, check the printed transversals directly.

Watch

Extended reading notes

Core claim

The central discovery is that pairwise intersection of all transversals does not force a common entry. For even n≥28 (except 30), the authors define latin squares—G_n for n≡2 mod 4 and H_n for n≡0 mod 4—together with X_28, X_32, Y_34, in which every transversal must include at least two out of three specially placed 'large-delta' entries, forcing any two transversals to overlap. A fourth or third transversal found by a randomized search shows that no single entry is contained in every transversal, up to n=10,000. The same construction idea yields squares with dominant transversals for a broad range of orders, and the proof isolates why group tables and diagonally cyclic squares can never hav

Load-bearing premise

For every even n in [36,10000], the randomized search actually produced transversals with the specified required/forbidden entries and empty common intersection; if even a single run failed or returned an invalid transversal, Theorem 6 fails for that order.

Editorial extensions

If this is right

  • Conjecture 3 is confirmed for n=28 and all even n from 32 to 10,000; for every even n≥38 the no-disjoint-transversals half holds unconditionally, and for n≥36 in the H_n family, so the only gap is the no-pinned-entry check beyond 10,000.
  • The delta-sum rule gives a reusable sufficient condition: a latin square whose entries all lie in a narrow Δ-band, with few extreme entries, cannot have two disjoint transversals.
  • For every n>4 with n≢3 mod 4 and n≠6, there is a latin square of order n with a dominant transversal; thus most orders admit a transversal that intersects every other one.
  • Group multiplication tables and diagonally cyclic squares never admit a dominant transversal, so the constructed squares are structurally different from these natural families.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Publishing the actual transversals from the search for each n up to 10,000 would turn Theorem 6 into a checkable certificate rather than a search-based assertion.
  • The same Δ-band design might be pushed to orders n≡3 mod 4 or to odd orders; if a square with no pinned entry and no disjoint transversals can be built there, it would sharpen the boundary between the parity cases.
  • For n>10,000, a deterministic construction of the missing transversals—perhaps guided by the Δ-sum constraints—would complete Conjecture 3 without relying on search.
  • Rephrased in graph language, these are edge-coloured complete bipartite graphs whose rainbow perfect matchings are pairwise intersecting but have no common edge; this is an extremal configuration worth testing in broader rainbow-matching settings.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The paper studies latin squares whose transversals are pairwise intersecting but share no common entry. It defines families G_n (n=4k+2), H_n (n=4k), and special examples X_28, X_32, Y_34, and uses a Delta-Lemma to prove that these squares have no two disjoint transversals. For n=28 and all even n in [32,10000] it claims that a randomized hill-climbing algorithm (Procedure 2) finds transversals whose intersection is empty, establishing that no entry is pinned. The paper also introduces dominant transversals, proving that Cayley tables and diagonally cyclic squares have none, and constructing latin squares with a dominant transversal for n>4, n not congruent 3 mod 4, n≠6.

Significance. The rigorous portions of the paper are elegant and, as far as I checked, correct. Lemmas 7–10 give parameter-free Delta-sum arguments that are machine-checkable; the inequalities and the required conclusions all work. The dominant-transversal construction in Theorem 13 is a nice application of a mod-2 Delta argument, and the negative results for group tables and diagonally cyclic squares are clean. If the computational claim in Theorem 6 can be made reproducible, the paper would be a substantial contribution to the understanding of transversal structures in latin squares, directly addressing a question posed by Wormald.

major comments (2)
  1. [§3, Theorem 6 (G_n and H_n cases)] The 'no pinned entry' half of Theorem 6 is not established by the written text. For each n in the stated range the paper asserts that Procedure 2 found transversals with certain inclusion/exclusion constraints, but no code, run logs, found transversals, or certificates are provided. The claim is load-bearing: the universal statement about all transversals depends on it. An independent reader cannot check any single order, let alone all orders up to 10,000. The paper should ship the actual transversals (or a machine-readable certificate) and/or the code used to generate and verify them.
  2. [§3, G_n description after Eq. (4)] Even if the searches were successful, the described four transversals do not obviously prove the absence of a pinned entry. The text states that the first three transversals ensure no entry from (4) is common to all, and that the fourth avoids (16,12,29). It does not state that the four transversals have empty total intersection, nor does it argue that no other entry besides (4) and (16,12,29) lies in all transversals. Thus, as written, the argument does not formally imply the 'no single entry' conclusion for G_n; an explicit statement or certificate of the empty intersection is needed.
minor comments (3)
  1. [§5, Theorem 13] The construction of T_2 in block Z is very terse. A sentence explaining that the count of occurrences of symbols from Z_n\Gamma in block X forces X (and analogously Y) to contain only symbols outside Gamma would make it clear why T_2 is automatically a transversal of the complementary block.
  2. [§2, Procedure 1] In Line 6 of Procedure 1, the condition involving R is correct, but a parenthetical explanation of why the current entries must be checked against R would help readers unfamiliar with Shor #-moves.
  3. [General] The paper jumps quickly between the cases n=4k+2 and n=4k. A short table summarizing which construction covers which order, and which lemmas apply, would improve readability.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the only concern is a reproducibility gap in unreported randomized-search witnesses, not a circular derivation.

full rationale

No significant circularity. The central Theorem 6 is built from parameter-free delta-sum inequalities (Lemma 5 and Lemmas 7-10), and the randomized Procedure 2 is used to produce witness transversals, not to fit a parameter or to predict a value from a fitted input. The R/F constraints placed on Procedure 2 are necessary consequences of Lemma 5 (e.g., fixing allowed delta-values per row); restricting to a necessary condition is not circular. The 'no pinned entry' half relies on unreported runs, so it is a reproducibility gap, but the existence of the witnesses is an input claim, not a reduction of the theorem to itself. The self-citations to [5] and [6] are to prior results: [6]'s pinned-entry construction is a direct external input for even n in Theorem 13, and [5] supplies enumeration data. These are not the conclusions being proved and do not smuggle in the target statement. Thus no step reduces the derivation to its inputs.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central results are parameter-free existence constructions. The load-bearing inputs are: the standard Delta-Lemma; the unproven (but checkable) assertion that the piecewise tables are Latin squares; Ryser's embedding theorem plus the unstated fact that a transversable even-order subsquare exists; and external enumeration data from [5] for small orders. No numbers are fitted to data. The algorithm's heuristic constants (the 100-move loop cap in Procedure 2, restart behavior) are efficiency knobs, not parameters of any claim: any returned transversal is valid by the invariant that Shor #-moves never introduce forbidden entries or remove required ones. 'Dominant transversal' is a new definition, not an invented entity in the rubric's sense.

assumptions (4)
  • standard math Delta-Lemma (Lemma 5): for any transversal of an order-n square, the sum of delta = s - r - c over its entries is 0 mod n (n odd) or n/2 mod n (n even).
    Stated without proof; standard tool in the transversal literature (Wanless survey [11]); every no-disjoint-transversals argument (Lemmas 7-10) rests on it.
  • ad hoc to paper The piecewise tables G_n, H_n, X_n, Y_34 are Latin squares for the stated k ranges.
    Asserted without proof at (3), (7), (11), (12); the delta-counts in Lemmas 7-10 presuppose the Latin-square property, and a flaw here would invalidate Theorem 6.
  • domain assumption Ryser's embedding theorem supplies an order-(4k+1) square M containing an order-2k subsquare S with a transversal T_1.
    Invoked in Theorem 13's n = 1 mod 4 proof; the 'by replacing S if necessary' step is true (e.g., a Klein-four subsquare works when 2k = 4) but the existence of a transversable even-order subsquare is not cited or shown.
  • standard math The mod-2 identity (15): for any transversal T' of M, the sum of theta(s) - theta(r) - theta(c) is odd.
    Proved in the text by a short argument; it makes the constructed T dominant by forcing every transversal through T_2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Latin Squares whose transversals intersect in unusual ways." pith.science (2026). https://pith.science/paper/CNN6AM27

@misc{pith2026260717547,
  author       = {Pith},
  title        = {Pith review of: Latin Squares whose transversals intersect in unusual ways},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CNN6AM27}},
  note         = {Machine review of arXiv:2607.17547}
}
abstract

A latin square of order $n$ is an $n\times n$ array in which each of $n$ symbols occurs exactly once in each row and column. A transversal in such a square is a selection of $n$ entries that includes one representative of each row and column, and one of each symbol. For all even orders $n\ge 28$ except $n=30$, we construct a latin square of order $n$ in which every pair of transversals share at least one entry. We conjecture that in our squares there is no single entry that is common to all transversals. We prove this conjecture for $n\le10\,000$ by finding transversals using an algorithm that is likely to be of independent interest. We say that a transversal is dominant if it intersects every other transversal of the same latin square. We show that there exist latin squares of order $n$ that have a dominant transversal for $n\in\{5,7\}$ and also for all $n\ge8$ such that $n\not\equiv3\bmod4$.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 2 linked inside Pith

  1. [1]

    Bowtell and R

    C. Bowtell and R. Montgomery, Almost every latin square has a decomposition into transversals, arXiv:2501.05438

    arXiv

  2. [2]

    N. J. Cavenagh and I. M. Wanless, On the number of transversals in Cayley tables of cyclic groups,Disc. Appl. Math.158(2010), 136–146

    2010

  3. [3]

    N. J. Cavenagh and I. M. Wanless. Latin squares with no transversals.The Electronic Journal of Combinatorics, 24(2):#P2.45, 2017

    2017

  4. [4]

    H. B. Mann. On orthogonal latin squares.Bull. Amer. Math. Soc., 50:249–257, 1944. 20 AFSANE GHAF ARI AND IAN M. W ANLESS

    1944

  5. [5]

    Egan and I

    J. Egan and I. M. Wanless. Latin squares with restricted transversals.Journal of Combinatorial Designs, 20:344–361, 2012

    2012

  6. [6]

    Ghafari and I

    A. Ghafari and I. M. Wanless, Latin squares whose transversals share many entries, arXiv:2412.12466

    arXiv

  7. [7]

    Montgomery, Transversals in latin squares,London Math

    R. Montgomery, Transversals in latin squares,London Math. Soc. Lecture Note Ser. 493, Cambridge University Press, 2024, 131–158

    2024

  8. [8]

    H. J. Ryser, A combinatorial theorem with an application to latin rectangles,Proc. Amer. Math. Soc.2, (1951) 550–552

    1951

Show all 11 references
  1. [9]

    P. W. Shor, A lower bound for the length of a partial transversal in a latin square,J. Combin. Theory Ser. A33(1982), 1–8

    1982

  2. [10]

    I. M. Wanless, Diagonally cyclic latin squares,European J. Combin.25(2004), 393– 413

    2004

  3. [11]

    I. M. Wanless, Transversals in latin squares: A survey,London Math. Soc. Lecture Note Ser.392, Cambridge University Press, 2011, 403–437

    2011

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.