---
title: Rook Statistics in Combinatorics
url: https://www.emergentmind.com/topics/rook-statistics
type: topic
---

# Rook Statistics in Combinatorics

Searching arXiv for relevant rook-statistics papers to ground the article.
arxiv_search query="rook statistics rook placements q-rook polyominoes" max_results=10

arxiv_search query="site:arxiv.org rook theory q-rook numbers finite general linear group" max_results=10

Rook statistics denotes a family of enumerative, weighted, geometric, and algorithmic invariants built from rook placements and rook-like visibility models on discrete boards. In the classical setting the basic objects are non-attacking placements on Ferrers boards and their rook numbers or rook polynomials; in modern extensions the same language covers fixed-margin multirook configurations, \(q\)- and elliptic weights, domination and guarding on polyominoes, higher-dimensional rook graphs, and random-placement asymptotics for safe cells [1812.00533] [1707.08192] [1810.10961] [2409.04423]. The subject is therefore not a single statistic but a coherent collection of statistics whose common feature is that row-column exclusion, or an analogue of rook attack, organizes the combinatorics.

## 1. Classical board models and rook polynomials

For a Ferrers board \(B=(b_1,\dots,b_n)\), a non-attacking \(k\)-rook placement is a \(k\)-subset of cells with no two rooks in the same row or column, and the \(k\)-th rook number \(r_k(B)\) counts such placements. A central structural result is the Goldman–Joichi–White factorization
\[
p(B,x)=\sum_{k=0}^n r_k(B)\,x^{\underline{n-k}}
      =\prod_{i=1}^n \bigl(x+b_i-(i-1)\bigr),
\]
which packages the rook numbers into a falling-factorial basis and converts them into a root-vector description of the board [1812.00533]. Two Ferrers boards are rook equivalent exactly when their root vectors are rearrangements of one another, and this leads to the rook equivalence graph: vertices are rook-equivalent boards, and edges correspond to moving squares from one column into a single other column while staying inside the same equivalence class. Connectivity of this graph is characterized by a multiplicity condition on the root-vector values: if \(v_i\) denotes the multiplicity of the value \(i\), then the graph is connected exactly when \(v_i>1\) implies \(v_{i+1}>1\) for all \(i<M\), where \(M\) is the maximum value appearing [1812.00533].

A second classical axis generalizes the row condition to levels. In \(m\)-level rook theory, rows are partitioned into blocks of \(m\) consecutive rows, and an \(m\)-level rook placement may contain at most one rook in each column and at most one rook in each level. The associated \(m\)-level rook numbers \(r_{k,m}(B)\) interpolate between ordinary rook theory and several higher-level Ferrers-board models. For singleton boards one has the factorization
\[
\sum_{k=0}^{N} r_{k,m}(B)\,x^{\downarrow (N-k);m}
=\prod_{i=1}^{N}\bigl(x+b_i-(i-1)m\bigr),
\]
and every Ferrers board is \(m\)-level rook equivalent to a unique \(m\)-increasing board [1508.05944]. The same work constructs explicit bijections between \(m\)-rook-equivalent boards and proves preservation of the \(m\)-inversion statistic \(\mathrm{inv}_m\), so the equivalence is not merely numerical.

Rook statistics also appear in partition theory through Ferrers boards attached to integer partitions. For a partition \(\pi\), the max-rook number \(R(\pi)\) is the maximal number of non-intersecting rooks that can be placed in its Ferrers board, and it is identified with the size of the Durfee triangle. Equivalently,
\[
R(\pi)=\max\{k:\lambda_i\ge k+1-i\text{ for }1\le i\le k\}.
\]
This yields a decomposition of the partition function \(P(n)=\sum_{k=1}^n R_k(n)\), where \(R_k(n)\) counts partitions of \(n\) with max-rook number \(k\), together with explicit rational generating functions for \(k=3,4,5\) and polynomial formulas for \(R_3(n)\) and \(R_4(n)\) on residue classes [2507.20260]. In this setting rook statistics measure a geometric threshold inside Ferrers diagrams rather than an entire placement distribution.

## 2. Weighted, \(q\)-deformed, and elliptic rook statistics

The Garsia–Remmel \(q\)-rook numbers refine ordinary rook numbers by weighting a placement \(C\) with \(q^{\mathrm{inv}(C)}\), where \(\mathrm{inv}(C)\) is the number of cells left after cancelling, for each rook, the cells above it in its column and to the left of it in its row. For a Ferrers shape \(\lambda\),
\[
R_k(\lambda;q)=\sum_{C\in\mathcal C_k(\lambda)} q^{\mathrm{inv}(C)},
\qquad
R_k(\lambda;1)=r_k(\lambda),
\]
and these polynomials satisfy a recursion obtained by deleting the first row of \(\lambda\) [2507.00766]. A recent development gives a positive tableaux formula: if \(\pi\) is a Dyck path with \(\lambda(\pi)=\lambda\), then \(R_k(\lambda;q)\) is a sum over \(\pi\)-standard Young tableaux of shapes \(\mu\vdash n\) with \(\mu_1=n-k\). This expresses \(q\)-rook numbers in terms of a tableau weight involving \(n(\mu')\), the Dyck-path area, a tableau statistic \(\gamma(T)\), and local \(\pi\)-arm factors, and it links rook statistics to \(q\)-Whittaker coefficients of unicellular LLT functions [2507.00766].

A different \(q\)-analogue counts matrices over finite fields rather than placements of combinatorial rooks. For a board \(B\subseteq[m]\times[n]\), let \(m_r(B,q)\) be the number of \(m\times n\) matrices over \(\mathbb F_q\) of rank \(r\) whose support is contained in \(B\), and define the normalized counts \(M_r(B,q)=m_r(B,q)/(q-1)^r\). These specialize modulo \(q-1\) to classical rook numbers. On rectangular boards, the associated \(q\)-hit polynomial
\[
P(B,q,t)=\sum_{i=0}^m H_i(B,q)t^i
\]
is defined by an explicit transform from the \(M_i(B,q)\), and it satisfies the reciprocity law
\[
P(\overline B,q,t)=q^{-|B|}t^m\,P(B,q,q^n/t).
\]
For Ferrers boards and, more generally, NE-property boards, \(M_r(B,q)\) coincides up to normalization with Garsia–Remmel \(q\)-rook numbers, while for general boards polynomiality and positivity can fail [1707.08192]. The same framework ties rook statistics to Delsarte’s rank-metric MacWilliams theory and partitions \(|GL_n(\mathbb F_q)|\) by \(q\)-hit numbers.

Further extensions replace the single parameter \(q\) by richer weight systems. In the \((a;q)\)-rook theory on Ferrers boards one introduces small and big weights
\[
w_{a;q}(k)=q\frac{1-aq^{2k+1}}{1-aq^{2k-1}},\qquad
W_{a;q}(k)=q^k\frac{(aq^3;q^2)_k}{(aq;q^2)_k},
\]
and builds weighted rook numbers whose product formulas specialize to the \(q\)-Pfaff–Saalschütz summation and a \( {}_4\phi_3 \) summation of Jain [1606.03877]. At the far end of the deformation hierarchy, augmented rook boards admit an elliptic extension in which \(q\)-numbers are replaced by elliptic numbers \([z]_{a,b;q,p}\) defined by theta functions, and Miceli–Remmel’s general product formula becomes an elliptic factorization theorem on extended augmented boards [1601.07834]. These theories show that rook statistics naturally support hypergeometric, elliptic, and finite-field structures.

## 3. Rook statistics as permutation, set-partition, and tree statistics

One major use of rook statistics is to encode permutation-like distributions. For integers \(c,n\ge 1\), a \(c\)-rook placement on an \(n\times n\) board is a nonnegative integer matrix with every row sum and column sum equal to \(c\), and \(A_c(n,k)\) counts those placements with exactly \(k\) rooks strictly below the main diagonal. When \(c=1\), \(A_1(n,k)\) is the Eulerian number counting permutations with \(k\) descents. In full generality the generalized Eulerian numbers satisfy the symmetry
\[
A_c(n,k)=A_c\bigl(n,c(n-1)-k\bigr),
\]
proved by a cyclic right shift of columns followed by transpose; the corresponding polynomial \(F_c(n;t)=\sum_k A_c(n,k)t^k\) is therefore palindromic [1508.03673]. For fixed small \(k\), the stabilization lemma \(A_c(n,k)=A_k(n,k)\) for \(c\ge k\) reduces enumeration to finitely many cases, and explicit rational ordinary generating functions are known for \((k,c)=(2,2),(3,2),(3,3)\) [1508.03673].

On the triangular board
\[
B_n=\{(i,j):1\le i<j\le n\},
\]
rook coincidences encode random set partitions and permutations. Distinguishable rooks are placed independently and uniformly on \(B_n\), allowing multiple occupancy, and pairs of rooks are classified by five coincidence types: RR, CC, RC, CR, and SS. In the sparse regime \(r=n-k\sim t\sqrt n\), attacks and alignments are rare enough for Chen–Stein theory to yield a process-level Poisson approximation with total variation error \(O((n-k)^3/n^2)\). After conditioning on no attacks, the number of size-3 blocks in a random set partition with \(k\) blocks converges to \(\mathrm{Poisson}\!\left(\frac23 t^2\right)\), while the number of 3-cycles in a random permutation with \(k\) cycles converges to \(\mathrm{Poisson}\!\left(\frac43 t^2\right)\); with high probability all remaining components have size \(1\) or \(2\) [1807.08192]. Here rook statistics control an entire component process rather than a single count.

Rook placements also organize generalized Bell and Stirling identities. In the Goldman–Haglund model on the staircase board \(J_n\), each rook adds \(s-1\) to the pre-weight of cells to its left in the same row and cancels cells above it in its column. The weighted sums
\[
S_{s,q}[n,k]
\]
satisfy a two-term recurrence interpolating between Stirling numbers of the first and second kind, and the associated Bell polynomials \(B_{s,q}[n;x]\) obey Spivey-type convolution formulas. Setting \(s=0\) and \(q=1\) recovers Spivey’s Bell number formula, while modified bottom-row preweights produce Type II generalized \(q\)-Stirling numbers in the sense of Remmel–Wachs [1410.0742].

A further combinatorial avatar appears in Gessel polynomials. Maximal nonattacking rook placements on \((n-1)\times kn\) rectangles carry generalized excedance and subcedance vectors \(\mathrm{exc}(g)\) and \(\mathrm{sub}(g)\), obtained by decomposing columns into \(k\) blocks of width \(n\) and comparing rook positions with the diagonal inside each block. Under Tewari’s bijection these become the descent and ascent vectors of labeled rooted plane \(k\)-ary trees, and the corresponding generating polynomial
\[
G_{n,k}(u,v)=\sum_g u^{\mathrm{exc}(g)}v^{\mathrm{sub}(g)}
\]
coincides with a tree ascent–descent enumerator [1604.06894].

## 4. Guarding, domination, and complexity on polyominoes

A geometric form of rook statistics studies visibility on polyominoes and polycubes. A rook at a tile guards all tiles reachable along coordinate directions while remaining inside the polyomino, and a dominating set is a set of rooks or queens whose attack lines cover every tile. For a \(d\)-polycube \(P\) with \(n\) tiles, the guard numbers satisfy
\[
g_R(P)\le \left\lfloor\frac n2\right\rfloor,\qquad
g_Q(P)\le \left\lfloor\frac n3\right\rfloor,
\]
with both bounds tight for infinite families, obtained respectively by a parity \(2\)-coloring argument and by mod-\(3\) layering in \(\ell^1\)-distance together with the lemma that a queen guards every tile at \(\ell^1\)-distance at most \(2\) [1810.10961]. The extremal constructions are the alternating-domino stack for rooks and the alternating-3-row ladder for queens.

The same paper separates domination from non-attacking placement theory. In two dimensions, rows and columns of a polyomino are defined intrinsically by line segments lying entirely in the shape, and the associated bipartite graph \(G(P)\) has row-vertices, column-vertices, and one edge per tile. A set of \(k\) non-attacking rooks corresponds exactly to a matching of size \(k\) in \(G(P)\), so the maximum number of non-attacking rooks equals the maximum matching size and is computable in polynomial time. Moreover,
\[
\max(P)=\left\lceil\frac n2\right\rceil,
\]
and if \(n=2m-1\) and \(P\) admits \(m\) non-attacking rooks, then that set is both dominating and unique [1810.10961]. By contrast, computing the minimum number of rooks or queens needed to dominate a polyomino is NP-hard via a reduction from planar 3SAT with bounded variable occurrences.

The hardness result is robust in dimension: any polyomino instance embeds in a single layer of a \(d\)-polycube, so the rook- and queen-visibility guard set problems remain NP-hard for all \(d\ge 2\) [1810.10961]. At the same time, domination by rooks in two-dimensional polyominoes can always be realized by a non-attacking dominating set of no larger cardinality. This produces an unusual complexity boundary: maximizing non-attacking rooks is polynomial-time by matching theory, but minimizing dominating rooks or queens is NP-hard.

## 5. Higher-dimensional rook geometries and probabilistic regimes

Higher-dimensional rook models replace the ordinary grid by simplices, cyclic quotients, Latin hypercubes, or full \(n^k\) chessboards. The simplicial rook graph \(\mathcal{SR}(m,n)\) has vertex set
\[
\{\mathbf x\in\mathbb N^m:\sum_i x_i=n\},
\]
with adjacency when two vertices differ in exactly two coordinates. Its independence number \(\alpha(\mathcal{SR}(m,n))\) is the maximum number of non-attacking rooks on an \((m-1)\)-dimensional simplicial chessboard. If \(p\) is prime and \(p\ge \max\{m,n\}\), then
\[
\frac1p\binom{n+m-1}{n}\le \alpha(\mathcal{SR}(m,n))
\le \frac1m\binom{n+m-1}{n},
\]
and in regimes with \(m\ge n\to\infty\),
\[
\alpha(\mathcal{SR}(m,n))=\bigl(1-o(1)\bigr)\frac1m\binom{n+m-1}{n}.
\]
The domination number satisfies \(\gamma(\mathcal{SR}(m,n))=\Theta(n^{m-2})\) for fixed \(m\), and \(\mathcal{SR}(m,n)\) is Hamiltonian except for \(m=1\) and \((m,n)=(2,1)\) [2102.00535]. The cyclic simplicial rook graph \(\mathcal{CSR}(m,n)\) adds a modular constraint, has automorphism group \(\mathcal S_m\times \mathbb Z_n^\times\times \mathbb Z_n^{m-1}\) for \(m,n>3\), diameter \(m-\lfloor (m-1)/n\rfloor -1\), and an NP-hard exact distance problem [2102.00535].

In another higher-dimensional direction, a \(d\)-dimensional Latin super cube is an \(n^d\)-cell board containing \(n^{d-1}\) non-attacking rooks, with exactly one rook on every axis-parallel file. For a real Hamming brick \(T_0\) of size \(e_1\times\cdots\times e_d\), if \(T_k\) is obtained by complementing exactly \(k\) coordinate sets, then the rook counts satisfy the distribution theorem
\[
R(T_k)=\frac{|T_k|}{n}-(-1)^k\!\left(\frac{|T_0|}{n}-R(T_0)\right),
\]
equivalently \(df(T_k)=(-1)^k df(T_0)\) for the deflection \(df(X)=|X|/n-R(X)\). The sum of Hamming distances from \(T_0\) to all rooks is
\[
h_n^d(T_0)=n^{d-2}\bigl[(n-e_1)+\cdots+(n-e_d)\bigr],
\]
independent of the actual rook count in \(T_0\) [2208.04113]. These formulas make rook distribution on subsystem partitions an inclusion–exclusion-like statistic on the Boolean lattice of coordinate complements.

Random-placement asymptotics supply yet another higher-dimensional statistic. On an \(n\times n\) board with \(n\) uniformly chosen distinct rook positions, the probability that a fixed square is safe and empty is
\[
\frac{\binom{(n-1)^2}{n}}{\binom{n^2}{n}}
\longrightarrow e^{-2}.
\]
For \(k\)-dimensional hyper-rooks, which attack all cells sharing at least one coordinate, the analogous probability is
\[
\frac{\binom{(n-1)^k}{n}}{\binom{n^k}{n}}
\longrightarrow e^{-k},
\]
and for \(k\)-dimensional line-rooks, with \(n^{k-1}\) rooks attacking along the \(k\) axis lines through each position, the safe-and-empty probability also tends to \(e^{-k}\) [2409.04423]. The normalized variance of the proportion of safe squares tends to \(0\), so the safe density concentrates around these limits.

## 6. Related structures, path models, and open directions

Rook statistics interact with several additional combinatorial structures. On Ferrers boards with row bounds \(f=(f_1,\dots,f_n)\), full rook placements correspond to restricted permutations \(\pi(i)\le f_i\), and this perspective extends to colored permutations in the wreath product \(C_r\wr S_n\). In that setting the sorting index \(\mathsf{sor}\) is equidistributed with the Coxeter length \(\ell\), and on rook-restricted subsets \(G_{r,n,f}\) there is a joint equidistribution between long tuples of set-valued statistics built from right-to-left minima, left-to-right minima, left-to-right maxima, cycle minima, and their color refinements [1409.8093]. This realizes rook statistics as Euler–Mahonian and Stirling-type distributions on colored Ferrers-board permutations.

A more indirect use of rook language appears in Catalan rook paths. A rook path allows arbitrary proper horizontal and vertical steps, and the Catalan constraint requires the path from \((0,0)\) to stay strictly left of \(y=x-1\). The generating function for Catalan rook paths ending at \((n,n)\) is algebraic:
\[
G_R(t)=\frac{1+3t-\sqrt{1-10t+9t^2}}{8t},
\]
with quadratic equation \(4tG_R(t)^2-(1+3t)G_R(t)+1=0\) and coefficients \(1,1,5,29,185,1257,\dots\) [1109.1806]. Queen paths and spider-step extensions satisfy analogous quadratic equations, indicating that “rook” statistics extend beyond placements to step-set models with rook-like motion.

Rook placements on Young diagrams also enter map enumeration. In a \(q\)-analogue of the Harer–Zagier framework, the statistic
\[
S(\mu,C)=\mathrm{inv}(C)+|\mu|-\binom{n+1}{2}
\]
on full rook placements \(C\) in Young diagrams \(\mu\in Y(n,s)\) yields a conjectured identity, now proved by Stanton, equating a weighted rook sum to an explicit \(q\)-hypergeometric expression involving \([2n-1]_q!!\), \(q\)-binomial coefficients, and \(\prod_{i=1}^k(1+q^{n+i})\) [1412.3182]. The same paper constructs a bijection between certain rook placements and tree-rooted maps, showing that rook statistics can encode topological enumeration.

Several open problems remain persistent. For polyomino guarding, the hardness of rook-visibility domination on simply connected polyominoes, the characterization of bipartite graphs arising as \(G(P)\), and higher-dimensional analogues of matching-based rook tools are explicitly posed as open [1810.10961]. For generalized Eulerian numbers \(A_c(n,k)\), simple recurrences and closed forms beyond small \(k\) remain open [1508.03673]. For \(q\)-hit numbers on general boards, natural combinatorial interpretations and broad positivity statements are unresolved [1707.08192]. For simplicial rook graphs, exact independence numbers outside the currently controlled regimes, and sharper domination constants, remain difficult [2102.00535]. These questions illustrate a general pattern: rook statistics are often governed by exact product formulas or bijections in special geometries, while the transition to unrestricted board classes, higher dimensions, or richer weight systems rapidly exposes hard structural and computational phenomena.

Source: https://www.emergentmind.com/topics/rook-statistics