Polyhedral Games: Theory, Computation, and Play

• Polyhedral games are game-theoretic models that use polyhedral geometry to structure strategy sets, equilibrium conditions, and dynamics.
• They integrate methods from online learning, equilibrium computation, and combinatorial optimization to solve complex strategy problems.
• Polyhedral approaches enable efficient best-response optimization and dynamic gameplay, bridging theoretical insights with practical applications.

Polyhedral games are games, game-theoretic models, or game-like combinatorial systems in which polyhedra, polytopes, polyhedral cones, or polyhedral complexes are structurally primary. Across current research, the term appears in several distinct but related senses: players’ feasible strategy sets may be compact polytopes; equilibrium, regret, or optimality conditions may be encoded by epigraphs, fans, or systems of linear inequalities; cooperative and population games may be analyzed through facets, extremal rays, tangent cones, and normal cones; and some educational or puzzle-oriented systems make polyhedra and tilings themselves the objects of play (Farina et al., 2024, Garcia-Segador et al., 24 Jan 2025, Joswig et al., 2019). The common denominator is not a single canonical model, but the use of polyhedral geometry as the organizing language for strategy, value, stability, or construction.

1. Polyhedral strategy spaces and formal models

A central formalization treats a game as polyhedral when each player’s feasible set is a convex polytope. In a polyhedral convex game, the game is given by a tuple $\Gamma = (m, \{\Omega_i\}, \{\bar U_i\})$, where $\Omega_i \subset \mathbb{R}^{d_i}$ is a convex polytope representing player $i$’s randomized strategy space and $\bar U_i$ is multilinear in the players’ strategies. A typical linear description is $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$ (Farina et al., 2022). In a closely related formulation, player $p$ has a rational polytope $A_p \subseteq \mathbb{R}^{d_p}$, utilities are multi-linear, and the vertices $V(A_p)$ index pure strategies of the associated “corner” game (Farina et al., 2024).

This representation subsumes classical normal-form and extensive-form settings. In normal-form games, a player’s mixed strategies are the simplex

$$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$

which is a $0/1$-vertex polytope (Chakrabarti et al., 2023). In perfect-recall extensive-form games, the sequence-form strategy space is

$\Omega_i \subset \mathbb{R}^{d_i}$0

and its vertices are deterministic sequence-form strategies (Farina et al., 2022). The same general viewpoint appears in concise combinatorial games with exponentially many pure actions: every player may have an action set $\Omega_i \subset \mathbb{R}^{d_i}$1, with $\Omega_i \subset \mathbb{R}^{d_i}$2 and $\Omega_i \subset \mathbb{R}^{d_i}$3 for all $\Omega_i \subset \mathbb{R}^{d_i}$4, so each action is an extreme point of an underlying combinatorial polytope (Kontogiannis et al., 25 Sep 2025).

In this setting, best-response optimization coincides with a linear minimization oracle. For a compact polytope $\Omega_i \subset \mathbb{R}^{d_i}$5, one writes

$\Omega_i \subset \mathbb{R}^{d_i}$6

and for polytopes the oracle always returns a vertex, hence a pure strategy or deterministic sequence-form policy (Chakrabarti et al., 2023). This equivalence between best-response computation and linear optimization is one of the main reasons polyhedral formulations support both algorithmic game theory and online learning.

A broader misconception is that “polyhedral” refers only to linear utility. The literature uses the term more widely: utilities may be multilinear (Farina et al., 2024), losses may be piecewise linear on a fixed polyhedral partition (Li et al., 13 May 2026), and some models are defined by polyhedral relaxations of nonlinear optimality equations rather than by polyhedral strategy sets alone (Kozachinskiy, 2020).

2. Online learning and no-regret dynamics on polyhedral domains

A major research direction studies online learning in polyhedral games without enumerating exponentially many actions. For $\Omega_i \subset \mathbb{R}^{d_i}$7-polyhedral games, Kernelized OMWU (KOMWU) simulates Vertex OMWU on the normal-form equivalent of an extensive-form game using a kernel trick, and in extensive-form games runs in time linear in the game-tree size per iteration (Farina et al., 2022). The core construction uses the feature map

$\Omega_i \subset \mathbb{R}^{d_i}$8

and the kernel

$\Omega_i \subset \mathbb{R}^{d_i}$9

With $i$0, the Vertex OMWU weights satisfy $i$1, and the primal iterate is recovered by

$i$2

For extensive-form games, the necessary kernel evaluations are computed by dynamic programming on the sequence-form tree (Farina et al., 2022).

Kernelization has also been extended beyond full-information feedback. In polyhedral games with $i$3, the kernel

$i$4

supports efficient computation of first and second moments of multiplicative-weights distributions, enabling semi-bandit and bandit learning with regret bounds

$i$5

respectively, with concrete instantiations for Colonel Blotto, graphic matroid congestion games, and DAG-based network congestion games (Kontogiannis et al., 25 Sep 2025). The same framework gives efficient no-regret-to-CCE conversion, so the empirical distribution of play becomes an $i$6-CCE with $i$7 (Kontogiannis et al., 25 Sep 2025).

A separate oracle model uses exact best-response oracles rather than explicit kernels. AFW-ROMD instantiates Reflected Online Mirror Descent with an approximate proximal oracle solved by away-step Frank–Wolfe, so each inner iteration is a best-response call. With $i$8, the number of best-response queries at iteration $i$9 is $\bar U_i$0; in zero-sum games this yields constant social regret, Nash gap $\bar U_i$1 as a function of total best-response calls $\bar U_i$2, and last-iterate linear convergence under SP-MS, while in general-sum games it gives $\bar U_i$3 (Chakrabarti et al., 2023). The geometry of these rates is governed by facial distance,

$\bar U_i$4

for which lower bounds are proved for standard-form, sequence-form, flow, and matching polytopes (Chakrabarti et al., 2023).

In online convex optimization with convex piecewise-linear losses, regret can itself be controlled by polyhedral structure. Under a fixed finite partition $\bar U_i$5 of $\bar U_i$6, polyhedral instability is measured by the region-switch count

$\bar U_i$7

and the minimax rate is

$\bar U_i$8

where $\bar U_i$9 is the maximum number of vertices per region (Li et al., 13 May 2026). For online submodular–concave games under Lovász convexification, this specializes to the permutation-switch count $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$0 and the rate $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$1 (Li et al., 13 May 2026). This suggests that in polyhedral online problems, local combinatorial complexity and the frequency of active-region changes can matter more than ambient dimension alone.

3. Exact equilibrium computation and polyhedral reductions

Polyhedral methods also support exact equilibrium computation. In linear $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$2-equilibria, each player’s deviations are drawn from a polytope $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$3 of linear transformations preserving $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$4, and equilibrium computation is reduced to a bilinear zero-sum saddle-point problem over the Correlator polytope $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$5 and the Deviator polytope $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$6 (Farina et al., 2024). A linear $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$7-equilibrium is a solution of

$X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$8

The generalized Ellipsoid Against Hope algorithm solves the dual feasibility problem with a separation oracle for $X = \{x \in \mathbb{R}^n : Bx = b, x \ge 0\}$9 and a good-enough-response oracle, records polynomially many responses, and then solves a compressed primal LP. Under the polynomial utility gradient property, linearity of deviations, and a polynomial-time separation oracle for $p$0, this yields an exact $p$1-equilibrium in time polynomial in $p$2, utility encoding length, and facet complexity (Farina et al., 2024). Correlated equilibrium in normal-form games, EFCE, and exact linear-deviation correlated equilibrium in extensive-form games are special cases (Farina et al., 2024).

For constrained bimatrix games, the polyhedral objects are the epigraphs of the optimal value functions

$p$3

and Proposition 3.1 states that $p$4 and $p$5 are convex polyhedra (Feinstein et al., 15 Sep 2025). Nash equilibria satisfy

$p$6

and extremal equilibria are characterized by

$p$7

(Feinstein et al., 15 Sep 2025). Three methods are developed for computing extremal Nash equilibria: vertex enumeration, polyhedral calculus, and vector linear programming (Feinstein et al., 15 Sep 2025).

Low-rank structure enters through $p$8 for fixed $p$9. With $A_p \subseteq \mathbb{R}^{d_p}$0, the reduced game is

$A_p \subseteq \mathbb{R}^{d_p}$1

The reduction is exact precisely when

$A_p \subseteq \mathbb{R}^{d_p}$2

equivalently when

$A_p \subseteq \mathbb{R}^{d_p}$3

(Feinstein et al., 15 Sep 2025). Under this restorability condition, maximal Nash faces of the original and reduced games correspond bijectively (Feinstein et al., 15 Sep 2025). A plausible implication is that low-rank reduction is not merely dimensional compression; it is a face-preserving transport of equilibrium geometry.

4. Cooperative, population, and simple games as polyhedral objects

In cooperative transferable-utility games, balancedness itself defines a polyhedral region. For $A_p \subseteq \mathbb{R}^{d_p}$4, the set of balanced games

$A_p \subseteq \mathbb{R}^{d_p}$5

is a closed convex polyhedral cone, and Theorem 3 shows that it is a $A_p \subseteq \mathbb{R}^{d_p}$6-dimensional polyhedral cone that is not pointed (Garcia-Segador et al., 24 Jan 2025). Its lineality space has dimension $A_p \subseteq \mathbb{R}^{d_p}$7 with basis $A_p \subseteq \mathbb{R}^{d_p}$8, its facets are in bijection with minimal balanced collections, and its extremal rays are explicitly identified (Garcia-Segador et al., 24 Jan 2025). Fixing $A_p \subseteq \mathbb{R}^{d_p}$9 yields an affine nonpointed cone; imposing $V(A_p)$0 and $V(A_p)$1 yields the polytope

$V(A_p)$2

which is a $V(A_p)$3-dimensional bounded polyhedron (Garcia-Segador et al., 24 Jan 2025). Its vertices are exactly the balanced $V(A_p)$4–$V(A_p)$5 games, the number of facets is $V(A_p)$6, and the adjacency graph is Hamilton-connected (Garcia-Segador et al., 24 Jan 2025). The same paper characterizes when the core is a singleton and shows that interior points of $V(A_p)$7 never have singleton core (Garcia-Segador et al., 24 Jan 2025).

In affine population games, the polyhedral language appears through tangent and normal cones of the product simplex. If

$V(A_p)$8

then the tangent cone at $V(A_p)$9 is

$$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$0

and the normal cone is described by supportwise affine equalities and off-support inequalities (Sun et al., 26 May 2026). State-Robust Equilibrium is defined by local best-response invariance under perturbations of the evaluation state, and in affine games exposure can be tested by finitely many linear programs with objective $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$1 over the tangent cone (Sun et al., 26 May 2026). The main implication is sharp: robust mixing requires local payoff identity on the support, and generically, in affine games, SRE reduce to strict pure Nash equilibria, although weak boundary equilibria can survive through feasible-set protection (Sun et al., 26 May 2026).

Simple games generate canonical polyhedral companions. For a simple game $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$2 with losing complex $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$3, the Bier sphere is

$$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$4

and the canonical fan $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$5 is a complete simplicial fan in

$$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$6

with rays generated by $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$7 and $$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$8 (Timotijević et al., 2023). Canonical polytopality is equivalent to weightedness, and canonical pseudo-polytopality is equivalent to rough weightedness (Timotijević et al., 2023). The paper also shows, by experimental and theoretical argument, that all simple games with at most five players are polytopal (Timotijević et al., 2023). This distinguishes two notions that are often conflated: being polytopal as a sphere, and being canonically polytopal through the normal fan associated with the game.

5. Dynamic games, Shapley operators, and lattice-game formulations

For discounted and energy games, polyhedral geometry enters through relaxations of optimality equations. In discounted games, the optimality polyhedron is

$$\mathcal{X}_i=\Delta(\mathcal{A}_i)=\left\{x\in\mathbb{R}_{\ge 0}^{|\mathcal{A}_i|}: \sum_{a\in\mathcal{A}_i}x_a=1\right\},$$9

over all edges $0/1$0, and the value vector is characterized by tightness of at least one outgoing edge at each node (Kozachinskiy, 2020). Polyhedral value iteration starts from a feasible point, computes a feasible shift from an auxiliary discounted normal play game on the tight-edge graph, moves along the boundary until a new inequality becomes tight, and repeats until the optimality equations hold (Kozachinskiy, 2020). The resulting deterministic runtimes are $0/1$1 for discounted games, $0/1$2 for bipartite discounted games, and $0/1$3 for energy games (Kozachinskiy, 2020). Here the “polyhedral game” viewpoint is explicitly algorithmic: optimal strategies correspond to faces determined by tight constraints (Kozachinskiy, 2020).

For undiscounted zero-sum games with finite state space, fixed-point sets of Shapley operators are characterized as hyperconvex subsets of $0/1$4 and as lattices in the induced partial order (Akian et al., 2021). In deterministic games with finite action spaces, these fixed-point sets are supports of polyhedral complexes with cells indexed by stationary strategies, and each cell is an alcoved polyhedron of type $0/1$5 (Akian et al., 2021). A finitely generated Shapley operator has the form

$0/1$6

and $0/1$7 is a finite union of alcoved polyhedra (Akian et al., 2021). The local homogeneous structure is described by polyhedral fans canonically associated to lattices in the Boolean hypercube (Akian et al., 2021).

Impartial finite combinatorial games admit a different polyhedral encoding as lattice games. A rational polyhedron $0/1$8 defines a board $0/1$9, a finite rule set $\Omega_i \subset \mathbb{R}^{d_i}$00 defines legal moves $\Omega_i \subset \mathbb{R}^{d_i}$01, and a rational strategy is a short rational generating function for the set $\Omega_i \subset \mathbb{R}^{d_i}$02 of $\Omega_i \subset \mathbb{R}^{d_i}$03-positions (Guo et al., 2011). For a set $\Omega_i \subset \mathbb{R}^{d_i}$04,

$\Omega_i \subset \mathbb{R}^{d_i}$05

and membership is tested by the Hadamard product with a monomial (Guo et al., 2011). Given a rational strategy, there are polynomial-time algorithms in fixed dimension to decide whether a position is winning, to find a move to a winning position if not, and to decide whether two positions are congruent in the sense of misère quotient theory (Guo et al., 2011). This is a polyhedral framework in the strict sense of integer points in rational polyhedra, rather than continuous strategy simplices.

6. Polyhedra as objects of play: nets, polyforms, and construction systems

A distinct usage of “polyhedral games” concerns games whose subject matter is polyhedral geometry itself. “MatchTheNet” is a single-player, browser-based game in which each round presents $\Omega_i \subset \mathbb{R}^{d_i}$06 three-dimensional polytopes and $\Omega_i \subset \mathbb{R}^{d_i}$07 planar nets, and the task is to match them correctly (Joswig et al., 2019). One game lasts five rounds; the player chooses language, difficulty level, and $\Omega_i \subset \mathbb{R}^{d_i}$08 with $\Omega_i \subset \mathbb{R}^{d_i}$09 and level 7 restricted to $\Omega_i \subset \mathbb{R}^{d_i}$10 (Joswig et al., 2019). The dataset contains 215 precomputed solids—Platonic, Archimedean, Catalan, proper Johnson solids, and their duals—plus 50 random convex polytopes (Joswig et al., 2019). The mathematical backend is the dual-graph unfolding formalism: choosing a spanning tree $\Omega_i \subset \mathbb{R}^{d_i}$11 of the dual graph $\Omega_i \subset \mathbb{R}^{d_i}$12 determines edge cuts $\Omega_i \subset \mathbb{R}^{d_i}$13 in the primal graph $\Omega_i \subset \mathbb{R}^{d_i}$14, and the unfolding map

$\Omega_i \subset \mathbb{R}^{d_i}$15

is a planar net when $\Omega_i \subset \mathbb{R}^{d_i}$16 is injective (Joswig et al., 2019). Nets are precomputed in polymake using a heuristic with backtracking, while visualization is implemented with three.js in standard web browsers (Joswig et al., 2019).

Another construction-oriented direction studies polyforms on periodic tilings and honeycombs. A polyform is a connected collection of cells from a fixed tiling or honeycomb, glued face-to-face, and can be formalized as a connected induced subgraph of the dual graph $\Omega_i \subset \mathbb{R}^{d_i}$17 modulo translations or modulo the full automorphism group $\Omega_i \subset \mathbb{R}^{d_i}$18 (Dobbelaere et al., 26 Feb 2026). The enumeration algorithm grows $\Omega_i \subset \mathbb{R}^{d_i}$19-polyforms by adding one adjacent cell at a time, then canonicalizes by applying orientation coset representatives in $\Omega_i \subset \mathbb{R}^{d_i}$20, translating to nonnegative coordinates, sorting lexicographically, and hashing the canonical name (Dobbelaere et al., 26 Feb 2026). The paper gives detailed counts for several tilings and honeycombs, including snub trihexagonal free polyforms $\Omega_i \subset \mathbb{R}^{d_i}$21 for $\Omega_i \subset \mathbb{R}^{d_i}$22–$\Omega_i \subset \mathbb{R}^{d_i}$23, rectified cubic free polyforms $\Omega_i \subset \mathbb{R}^{d_i}$24 for $\Omega_i \subset \mathbb{R}^{d_i}$25–$\Omega_i \subset \mathbb{R}^{d_i}$26, and tetrahedral–octahedral free polyforms $\Omega_i \subset \mathbb{R}^{d_i}$27 for $\Omega_i \subset \mathbb{R}^{d_i}$28–$\Omega_i \subset \mathbb{R}^{d_i}$29 (Dobbelaere et al., 26 Feb 2026). The same framework is linked directly to puzzle design, complete and balanced piece sets, and Tetris-like dynamics on non-square tilings and 3D honeycombs (Dobbelaere et al., 26 Feb 2026).

These educational and puzzle-oriented systems do not study Nash, core, or value in the usual sense. Instead, they treat polyhedral combinatorics, unfoldings, symmetries, and tiling automorphisms as gameplay mechanics. This suggests that the phrase “polyhedral games” now names both a family of analytic techniques in game theory and a family of games whose playable content is polyhedral structure itself.