Queens: Combinatorics and Algorithmics

• Queens are a family of chessboard placement problems defined by the non-attacking queen relation across rows, columns, and diagonals.
• They encompass classical and toroidal variants, utilizing permutation encodings and hypergraph formulations to analyze feasibility and count solutions.
• Applications extend to asymptotic enumeration, optimization algorithms, and spectral graph theory, providing insights into combinatorial and probabilistic methods.

Queens, in combinatorics and algorithmics, denotes a family of chessboard placement problems defined by the queen attack relation—common row, common column, or common diagonal. The central object is the $n$-queens problem: place $n$ mutually non-attacking queens on an $n\times n$ board, or count how many such placements exist. The same attack primitive also underlies toroidal, reflecting, Toeplitz, infinite-board, higher-dimensional, completion, coverage, and peaceability variants, together with graph-theoretic, probabilistic, spectral, and optimization formulations (Bowtell et al., 2021, Clinch et al., 2024, Dai et al., 2024).

1. Classical and toroidal formulations

On the classical $n\times n$ board, a queen at $(x,y)$ attacks any square $(x',y')$ with $x'=x$, $y'=y$, $x'+y'=x+y$, or $x'-y'=x-y$. A partial configuration is a set $n$0 with at most one queen in each row, column, and non-toroidal diagonal; an $n$1-queens configuration is a partial configuration of size $n$2 (Luria et al., 2021).

A standard encoding uses a permutation $n$3, where row $n$4 contains a queen in column $n$5. The row and column constraints are then automatic, and the remaining obstruction is diagonal collision. Formally, a classical solution is a $n$6 for which all values $n$7 and $n$8 are distinct. This permutation viewpoint is central both to asymptotic counting and to probabilistic algorithms (Bowtell et al., 2021).

The toroidal variant replaces the board by a torus: rows and columns are indexed modulo $n$9, and both diagonal families wrap around modulo $n\times n$0. There are exactly $n\times n$1 toroidal diagonals of each type, all of size $n\times n$2. A toroidal configuration is therefore a permutation $n\times n$3 such that the residues $n\times n$4 are all distinct and the residues $n\times n$5 are all distinct (Luria et al., 2021).

This modular structure admits a hypergraph formulation. A toroidal configuration is exactly a perfect matching in a natural $n\times n$6-partite, $n\times n$7-uniform hypergraph $n\times n$8 with parts $n\times n$9, each of size $n\times n$0, whose edges are quadruples $n\times n$1. The classical problem similarly corresponds to matchings in a related non-regular hypergraph $n\times n$2 with non-wrapping diagonal parts of size $n\times n$3 (Bowtell et al., 2021).

2. Existence theorems and early milestones

The classical existence theory is elementary only at a coarse level. The values $n\times n$4 are trivial, and classical solutions exist for all $n\times n$5. Historically, $n\times n$6 was found by Nauck in 1850 and proved by Glaisher and Pauls in 1874; values of $n\times n$7 are tabulated up to $n\times n$8 in OEIS A000170 (Bowtell et al., 2021).

The toroidal existence criterion is much sharper. Pólya proved in 1918 that

$n\times n$9

Thus toroidal feasibility is governed by a complete congruence obstruction, unlike the classical board (Bowtell et al., 2021).

Several other queen models exhibit distinct arithmetic thresholds. On a symmetric $(x,y)$0 Toeplitz matrix $(x,y)$1, where queens attack by row, column, or equal Toeplitz value $(x,y)$2, $(x,y)$3 nonattacking queens can be placed if and only if $(x,y)$4. In the same model, $(x,y)$5 nonattacking queens can be placed for every $(x,y)$6 (Szaniszlo et al., 2010).

Reflecting queens provide another existence theory. On an $(x,y)$7 board with a $(x,y)$8 reflecting strip attached along one side, diagonal paths may reflect via the strip. Klarner’s reflecting $(x,y)$9-queens problem is equivalent to Slater’s number-theoretic pairing problem, and reflecting $(x',y')$0-queens configurations exist for all sufficiently large $(x',y')$1. Small nonexistence is known for $(x',y')$2; existence was constructed for $(x',y')$3, and computationally for $(x',y')$4 (Dai et al., 2024).

3. Asymptotic enumeration

The modern counting theory separates the classical count $(x',y')$5 from the toroidal count $(x',y')$6. Rivin, Vardi, and Zimmerman conjectured that

$(x',y')$7

on the Pólya residue classes for the toroidal problem. Luria proved the first general asymptotic upper bounds: $(x',y')$8 with $(x',y')$9, and conjectured that the toroidal upper bound is tight when $x'=x$0 (Luria, 2017).

Bowtell and Keevash proved that conjecture. For the toroidal problem,

$x'=x$1

and $x'=x$2 otherwise. They also proved

$x'=x$3

for all sufficiently large $x'=x$4. Their proof combines a random greedy algorithm with randomized algebraic construction and iterative absorption, phrased in the language of hypergraph matchings (Bowtell et al., 2021).

For the classical problem, the constant $x'=x$5 is not the final answer. Simkin proved that there exists a constant $x'=x$6 with $x'=x$7 and $x'=x$8 such that

$x'=x$9

The proof introduces limit objects called queenons, defines a strictly concave entropy functional $y'=y$0 on the compact convex set of queenons, and identifies $y'=y$1 by the variational principle $y'=y$2 (Simkin, 2021).

The toroidal constant $y'=y$3 admits a heuristic interpretation. Starting from $y'=y$4 permutations, the two diagonal families are approximately collision-free with probability about $y'=y$5 each, and a further joint constraint contributes a third $y'=y$6 factor, suggesting $y'=y$7. Bowtell–Keevash formalize the lower bound, while Luria’s entropy argument formalizes the matching upper bound (Bowtell et al., 2021).

4. Completion, construction, and computation

The $y'=y$8-queens completion problem asks whether a given partial non-attacking configuration can be extended to a full $y'=y$9-queens configuration. Let $x'+y'=x+y$0 be the largest $x'+y'=x+y$1 such that every placement of $x'+y'=x+y$2 mutually non-attacking queens is completable. Glock, Munhá Correia, and Sudakov proved that, for all sufficiently large $x'+y'=x+y$3,

$x'+y'=x+y$4

The lower bound uses rainbow matchings in bipartite graphs, probabilistic sparsification, and a generalized rainbow matching lemma; the upper bound comes from explicit non-completable partial configurations and a linear-programming dual line-weighting certificate (Glock et al., 2021).

That upper bound was improved in 2026. For all sufficiently large $x'+y'=x+y$5,

$x'+y'=x+y$6

The improvement preserves the line-weighting method but refines the covering construction for the unattacked squares of a carefully chosen partial configuration (Nielsen, 23 Jun 2026).

For direct construction of full configurations, a simple two-stage randomized algorithm gives a strong lower bound for the classical counting problem. Stage 1 runs a toroidal random greedy process up to

$x'+y'=x+y$7

and with high probability places $x'+y'=x+y$8 queens while maintaining dynamic concentration

$x'+y'=x+y$9

for $x'-y'=x-y$0. Stage 2 uses absorbers to convert the partial toroidal configuration into a full classical one, yielding

$x'-y'=x-y$1

(Luria et al., 2021).

Optimization-oriented computation has produced different objective variants. Integer programming models with row, column, and diagonal constraints were used to compute many new lexicographically optimal solutions for $x'-y'=x-y$2, and a lexicographic bottleneck “most beautiful” variant was solved up to $x'-y'=x-y$3 (Fischetti et al., 2019). A separate Monte Carlo line of work reformulates counting as a rare-event problem, uses quantile re-ordering based on the Lorenz curve, and combines vertical-likelihood Monte Carlo, importance sampling, simulated annealing, nested sampling, and a Swendsen–Wang style algorithm for sampling binary matrices (Polson et al., 2024).

5. Structural abstractions

Several structural formalisms organize the queens literature beyond direct board combinatorics. For fixed $x'-y'=x-y$4 identical queens on a dilated rational convex polygon $x'-y'=x-y$5, inside-out Ehrhart theory implies that the number $x'-y'=x-y$6 of nonattacking placements is a quasipolynomial in $x'-y'=x-y$7 of degree $x'-y'=x-y$8. Its leading coefficient is $x'-y'=x-y$9, and reciprocity identifies $n$00 with the number of unlabelled combinatorial configuration types. The exact formula is indexed by the intersection semilattice of attack hyperplanes and a slope matroid of weighted graphs (Chaiken et al., 2013).

The asymptotic theory of the classical problem introduces queenons, which are probability measures on $n$01 with uniform marginals and sub-uniform diagonal marginals. They extend permutons by encoding diagonal sparsity. The count of configurations approximating a fixed queenon is controlled by an entropy functional involving Kullback–Leibler divergence of the measure itself and of its diagonal pushforwards, and a large deviations principle shows that typical large $n$02-queens configurations concentrate near the unique entropy maximizer (Simkin, 2021).

A distinct structural notation appears in graph theory: the $n$03-Queens’ graph $n$04 has the $n$05 board’s squares as vertices, with two vertices adjacent when the corresponding squares share a row, a column, or a diagonal. This graph decomposes as the sum of block-structured matrices coming from two triangular queen graphs $n$06 and $n$07, a disjoint union of cliques along anti-diagonals, and two disjoint unions of complete bipartite graphs $n$08. The triangular graphs are regular and integral, with explicitly determined spectra, and Weyl’s inequalities then give eigenvalue bounds for the full attack graph (Cardoso et al., 1 Aug 2025).

These abstractions are complementary. Hypergraphs encode feasibility and counting on toroidal and classical boards; inside-out polytopes control exact quasipolynomiality for fixed $n$09; queenons govern the classical asymptotic constant; and attack graphs expose spectral structure on the square-board geometry.

6. Variants, extensions, and asymptotic regimes

The peaceable queens problem replaces mutual non-attack by bipartite non-attack: maximize $n$10 such that one can place $n$11 white queens and $n$12 black queens on an $n$13 board with no opposite-color capture. For the regular board, $n$14 for all sufficiently large $n$15, improving the previous $n$16 bound. On the torus, parity causes a sharp odd–even split: $n$17 with asymptotic lower bounds $n$18 for even $n$19 and $n$20 for odd $n$21 (Clinch et al., 2024).

Higher-dimensional random-placement models replace exact non-attack by the expected proportion of safe squares. For $n$22 random queens on an $n$23 board, that proportion converges to

$n$24

The same framework defines line-queens and hyper-queens on $n$25-dimensional boards, with rook-like contributions $n$26 and additional bishop-like factors; in dimension $n$27, the line-queen safe-square proportion converges to

$n$28

(Cashman et al., 2024).

Infinite-board greedy placement leads to another asymptotic geometry. On the doubly infinite board $n$29 numbered along a square spiral, the greedy non-attacking queens are described by the Tribonacci word; the placements lie on four lines of slopes $n$30 and $n$31, where $n$32 is the Tribonacci constant. On the single-quadrant board $n$33 numbered along antidiagonals, rows and columns of the associated Sprague–Grundy table are permutations of $n$34, and the queen positions are conjectured to lie on two lines of slopes $n$35 and $n$36 (Dekking et al., 2019).

A fixed-$n$37 coverage variant asks, not for non-attack, but for the maximum number of covered squares. For each fixed $n$38, there is a non-attacking threshold beyond which all optimal configurations are pairwise non-attacking, and a stabilizing threshold beyond which the set of optimal configurations becomes constant. Using the resulting loss-function analysis, all optimal large-board configurations were determined for $n$39 (Adhikari et al., 4 Aug 2025).

Taken together, these variants show that queens problems are not a single counting puzzle but a broad research domain. Their shared attack geometry supports arithmetic existence theorems, entropy maximization, quasipolynomial counting, rainbow matching methods, spectral decomposition, probabilistic asymptotics, and a growing catalogue of board-dependent threshold phenomena.