--- title: 'Tournament Minimum Rank: Extremal Problems' url: https://www.emergentmind.com/topics/tournament-minimum-rank type: topic --- # Tournament Minimum Rank: Extremal Problems Across the cited literature, “Tournament Minimum Rank” does not denote a single standardized object. In different research traditions it refers to at least five distinct extremal questions: the smallest numerical finishing position in a competitive format, the probability that tournament selection chooses the best-ranked individual, a linear order minimizing backward arcs in a tournament digraph, the smallest matrix rank or representation dimension needed to realize a tournament, and the minimum order of a tournament satisfying prescribed domination or realization properties. These uses share the language of extremality, but the underlying mathematical objects are different [0803.2925][2606.31565][2110.05188]. ## 1. Finishing positions and ranking quality in competitive tournaments In Swiss-system and tournament-design research, the closest interpretation of minimum rank is usually a player’s tendency to finish with a small numerical placing rather than a literal per-player worst-case rank theorem. A Swiss-system study models a tournament on a fixed player set \(P=\{p_1,\dots,p_n\}\), tracks pre-tournament Elo, current score, current rank, and color difference, and evaluates pairing systems by how well the final ranking matches the latent true-strength order. Its central ranking concept is **ranking quality**, measured primarily by normalized Kendall \(\tau\), with supplementary validation by normalized Spearman \(\rho\) and NDCG. That study states explicitly that it does **not** provide formal per-player minimum-rank guarantees; instead it compares how pairing mechanisms reduce rank distortion in aggregate. Under its simulations, the pairing systems were ordered by expected ranking quality as \[ \text{Burstein} > \text{Random2} > \text{Dutch} = \text{Dutch BBP} > \text{Random} > \text{Monrad}, \] with mean normalized Kendall \(\tau\) values \(0.671\), \(0.654\), \(0.634\), \(0.633\), \(0.605365\), and \(0.481577\), respectively, in the default \(32\)-player, \(7\)-round setting [2112.10522]. A broader tournament-design study makes the same interpretive move more explicitly: if minimum rank is understood as letting strong players end up at small numerical positions such as rank \(1\), \(2\), or \(3\), then the relevant metrics are the average rank of Top 1, the average rank of Top 8, \[ \frac{\sum_{\ell=1}^k r_\ell}{k(k+1)/2}, \] the number of inversions, and a weighted inversion measure that penalizes upward displacement near the top by \[ \sum_{k=i+1}^{j} \frac{1}{\ln(k)}. \] Within a \(32\)-player, fixed-match-budget simulation program, the Swiss-system is reported as the most accurate format for ranking all participants, while draw and process can be superior for champion selection in some regimes. For example, under the theoretical \(skill=5\) model and an \(80\)-match budget, a \(5\)-round Swiss produced \(70.553\) inversions versus \(96.793\) for knockout, weighted inversions \(22.203\) versus \(29.490\), and Top 8 average rank \(1.302\) versus \(1.651\) [2103.06023]. Taken together, these works treat minimum rank not as a guaranteed attainable placement but as a question of **rank fidelity**: how much a tournament design compresses or amplifies the gap between latent strength and observed final standing. ## 2. Probabilistic tournaments and the probability of selecting rank \(1\) In evolutionary computation, minimum rank has a literal meaning: **rank \(1\)** is the best individual, rank \(2\) the next best, and so on up to rank \(n\). The relevant problem is the rank distribution induced by probabilistic tournament selection. A probabilistic tournament of size \(t\) samples \(t\) individuals uniformly with replacement, sorts them by rank, and then chooses tournament seed \(s\) with probability \(\alpha_s\), where seed \(1\) is the best within the sample. If \(I\) denotes the selected population rank and \(I_s\) the population rank of seed \(s\), then \[ P(I=k)=\sum_{s=1}^t \alpha_s P(I_s=k), \] with \(\alpha_s\ge 0\) and \(\sum_{s=1}^t \alpha_s=1\). The exact seed-rank distribution is \[ P(I_s=k)=\sum_{r=s}^t \binom{t}{r} \left[ \left(\frac{k}{n}\right)^r\left(1-\frac{k}{n}\right)^{t-r} - \left(\frac{k-1}{n}\right)^r\left(1-\frac{k-1}{n}\right)^{t-r} \right]. \] This gives a closed-form description of the entire selection probability over ranks [0803.2925]. For minimum rank in the strict sense, the key quantity is the probability of selecting the globally best individual, \(k=1\). Since rank \(1\) can only appear as seed \(1\), the paper derives \[ P(I=1)=\alpha_1\left[1-\left(1-\frac1n\right)^t\right]. \] The decomposition is exact: the best individual must appear at least once among the \(t\) draws, and the tournament rule must then select the best sampled seed. In the deterministic special case \(\alpha_1=1\), \[ P(I=1)=1-\left(1-\frac1n\right)^t, \] which increases monotonically with tournament size \(t\). More generally, deterministic tournament selection yields \[ P(I=k)=\left(1-\frac{k-1}{n}\right)^t-\left(1-\frac{k}{n}\right)^t. \] The same work proves that every probabilistic tournament of size \(t\) is equivalent to a unique polynomial rank scheme of degree at most \(t-1\). Writing \[ P(I=k)=\sum_{l=1}^{t} a_l k^{l-1}, \] the coefficients are obtained by a linear map \(a=T\alpha\), and the minimum-rank probability is simply the polynomial evaluated at \(k=1\): \[ P(I=1)=\sum_{l=1}^{t} a_l. \] In this literature, tournament minimum rank is therefore a fully explicit order-statistic quantity rather than a qualitative property [0803.2925]. ## 3. Minimum-backward-arc rankings and feedback arc sets In combinatorial optimization on tournaments, the central “minimum rank” problem is not a player’s best placement but a **minimum-upset ranking**. A tournament is a digraph \(T=(V,A)\) in which for each distinct \(u,v\in V\), exactly one of \(uv\) or \(vu\) is present. A ranking is a linear ordering \(v_1,\dots,v_n\). An arc \(v_jv_i\) with \(j>i\) is backward relative to that ordering, and the optimization problem is to minimize the number or total weight of such backward arcs. This is exactly the minimum feedback arc set problem on tournaments: a set \(F\subseteq A\) is a feedback arc set if \(T-F\) is acyclic, and in a tournament the backward arcs of an ordering form a feedback arc set, while any feedback arc set yields an acyclic remainder whose topological orders are optimal rankings. In the weighted setting, the objective is \[ \min \sum_{a\in F} w(a) \] over feedback arc sets \(F\), equivalently \[ \min \sum_{(u,v)\in A:\ v \prec u} w(u,v) \] over linear orders \(\prec\) [2606.31565]. The corresponding polyhedral quantities are \[ \tau_w(G) := \min\{ w^T x: x \in {\mathbb Z}_+^A,\ Mx \ge \mathbf 1\}, \qquad \nu_w(G) := \max\{ y^T \mathbf 1: y\in {\mathbb Z}_+^{\mathcal C},\ y^TM\le w^T\}, \] together with their LP relaxations \(\tau_w^*(G)\) and \(\nu_w^*(G)\), where \(\mathcal C\) is the family of directed cycles and \(M\) is the cycle–arc incidence matrix. The basic inequality chain is \[ \nu_w(G) \le \nu_w^*(G)=\tau_w^*(G)\le \tau_w(G). \] For **cycle Mengerian** tournaments, equivalently **Möbius-free** tournaments, the exact min–max relation \[ \tau_w(T)=\nu_w(T) \] holds for every nonnegative integral arc-weight function. On this class, exact combinatorial polynomial-time algorithms are available: minimum feedback arc set in \(O(n^9)\) time and maximum cycle packing in \(O(n^7)\) time. The same work emphasizes that minimum feedback arc set on general tournaments is \(NP\)-hard and admits a PTAS [2606.31565]. This formulation makes tournament ranking an optimization problem over linear orders. A minimum feedback arc set does not merely encode disagreement; after deleting it, any topological ordering of the resulting acyclic digraph is an optimal ranking. In that sense, “minimum rank” here means a ranking that minimizes contradiction with the tournament orientation. ## 4. Axiomatic ranking, worst-case guarantees, and order-theoretic limits A complementary tradition treats tournament ranking as an axiomatic object rather than a purely combinatorial optimization problem. A ranking \(r\) of a tournament \(T=(V,E)\) induces the set of backward arcs \[ \overleftarrow{\mathsf E}(r,T)=\{(x,y)\in E(T): r(x)s\) and all feasible profiles. This relation is neither complete nor transitive in general. If it is complete, however, it is transitive, \(\succeq\) becomes a complete preorder, and its asymmetric part is a finite union of transitive tournaments. In that complete case, the lowest rank is naturally identified with the bottom equivalence class rather than an isolated numerical position [1507.01327]. ## 5. Matrix-theoretic minimum rank and representation dimension In representation theory, tournament minimum rank becomes a linear-algebraic invariant. A tournament \(T\) on \(n\) nodes is represented by a real skew-symmetric matrix \(M\in\mathbb R^{n\times n}\) such that \[ i \succ_T j \iff M_{ij} > 0. \] The minimum rank of \(T\) is the smallest even \(d\) for which such a matrix of rank \(d\) exists. Equivalently, \(T\) can be represented in \(d\) dimensions by vectors \(h_1,\dots,h_n\in\mathbb R^d\) through the bilinear form \[ i \succ_{T[H]} j \iff h_i^T A^{rot} h_j > 0, \] where \(A^{rot}\) is the canonical block-rotation matrix and \[ M = H^T A^{rot} H. \] Within this framework, forbidden configurations for fixed-rank tournament classes must occur as unions of **flip classes**, since negating a subset of representing vectors flips precisely the cut edges and preserves representability rank. The exact rank-\(2\) characterization is especially sharp: \[ \operatorname{minrank}(T)=2 \iff T \text{ is locally transitive}. \] The same work establishes a worst-case lower bound \[ d\ge \sqrt n \] for the minimum dimension needed to represent all tournaments on \(n\) nodes, and an upper bound for any individual tournament, \[ \operatorname{minrank}(T)\le 2(\mu(T)+1), \] where \(\mu(T)\) is the minimum, over the flip class and over permutations, of the number of unique nodes participating in a feedback arc set [2110.05188]. A different matrix-theoretic notion is developed for symmetric matrices \(M_T(f,\mathbf a)\) built from tournament orientations and a function \(f\). In the two-value regime \(\mathbf a\in\{\alpha,\beta\}^N\), every such matrix has rank at least \(N/2\). The key theorem ties low rank to the multiplicity of a particular eigenvalue \(\mu(f;\alpha,\beta)\) in an associated bipartite graph, giving \[ m+n-2-\nu \le \operatorname{rank}(M)\le m+n+2-\nu, \] where \(\nu\) is the multiplicity of \(\mu(f;\alpha,\beta)\). For linear \(f\), there are real examples with \[ \operatorname{rank}(M)\le \frac{N}{2}+O(1), \] and rational examples with \[ \operatorname{rank}(M)\le \left(\frac12+\varepsilon\right)N+O(1) \] for every \(\varepsilon>0\). This is not the skew-symmetric representation rank of a tournament, but it is another well-developed minimum-rank problem attached to tournament structure [2401.14015]. ## 6. Other minimum notions: minimum order, minimum standards, and robust max–min design Several adjacent literatures use “minimum” in ways that are closely related to, but not identical with, minimum rank. One such notion is the **minimum order** of a tournament having a realization property. For the strong \(S_k\)-property, a tournament must realize every prescribed orientation pattern on every \(k\)-set by some outside vertex. The minimum order is denoted \(F(k)\), and the paper establishes \[ 2^{k-1}(k+2)-1 \le F(k)\le \min\left\{n:2^k\binom{n}{k}(1-2^{-k})^{\,n-k}<1\right\}\le (1+o(1))\ln(2)\,k^2 2^{2k}. \] It also proves a constructive quadratic-residue bound: for a prime \(p\equiv 3\pmod 4\), if \[ p > k^2 2^{2k-2}, \] then the quadratic-residue tournament of order \(p\) has the strong \(S_k\)-property [2205.08357]. Another neighboring concept is a **minimum qualifying standard** in a rank-order tournament. In that model, only agents with realized performance \(Y_i\ge \rho\) are eligible for prizes, and among qualifiers the prizes are assigned by rank. For any fixed prize schedule, the effort-maximizing standard is set at a mode of performance that is weakly higher than the global mode. When the prize scheme is also optimized, winner-take-all is optimal under increasing failure rate, while equal prize sharing among all qualifying agents is optimal under decreasing failure rate. Under log-concavity and log-convexity, respectively, these tournament-with-standard schemes remain optimal within a larger class of anonymous monotone contracts that may depend on cardinal performance [2412.01139]. Robust tournament design introduces a further max–min interpretation. In a rank-order tournament with unknown noise distribution, known only through an upper bound \(\bar H\) on Shannon entropy, the designer chooses prizes to maximize the **lowest** equilibrium effort across all admissible noise environments. The robust prize scheme awards positive prizes to all ranks except the last and gives a distinct top prize. Asymptotically, the prize differentials follow \[ d_r^\infty=\frac{1}{(n-1)r}, \] and the prize vector becomes \[ v_r^\infty=\frac{H_{n-1}-H_{r-1}}{n-1}. \] The least favorable noise distribution induced by this asymptotic design is exponential with rate \(e^{-\bar H}\) [2507.16348]. Finally, some tournament-theoretic work uses minimum in a purely structural sense, such as minimum semidegree thresholds guaranteeing strong subtournaments. For balanced \(c\)-partite tournaments, sufficient regularity conditions force a maximal subtournament \(T\) of order \(c\) with \[ \delta(T)\ge \left\lfloor \frac{c-2}{4}\right\rfloor+1, \] hence strong connectivity. This is not a rank notion, but it illustrates how “minimum” in tournament theory often refers to an extremal constraint on structure rather than on numerical ranking [2010.09835]. The phrase “Tournament Minimum Rank” therefore names a family of non-equivalent ideas rather than a single invariant. In competitive-format studies it concerns rank fidelity and placement distortion; in evolutionary computation it is the explicit probability of selecting rank \(1\); in graph optimization it is minimum-upset linear ordering; in representation theory it is minimum skew-symmetric or related matrix rank; and in adjacent extremal theories it shifts toward minimum order, minimum standards, or worst-case guaranteed effort. The technical content of the term is determined entirely by which tournament model is being used. Source: https://www.emergentmind.com/topics/tournament-minimum-rank