Tron, a combinatorial Game on abstract Graphs

We study the combinatorial two-player game Tron. We answer the extremal question on general graphs and also consider smaller graph classes. Bodlaender and Kloks conjectured in [2] PSPACE- completeness. We proof this conjecture.

💡 Research Summary

The paper provides a comprehensive theoretical study of the two‑player combinatorial game Tron, in which each player moves a token along the edges of a graph, permanently claiming each visited vertex. The authors first formalize the game on an arbitrary connected graph G = (V,E) with designated start vertices s₁ and s₂. At each turn a player must move to an adjacent, unclaimed vertex; the vertex then becomes part of that player’s “territory” and can never be used again. The game ends when a player has no legal move, and the player who controls more vertices wins.

The first major contribution addresses an extremal question: for a graph with n vertices, how many vertices can the optimal player guarantee to claim regardless of the opponent’s strategy? By constructing a symmetric “cut‑and‑balance” argument, the authors prove a new lower bound of ⌊n/2⌋, improving on previously known bounds and narrowing the possible range to between ⌊n/2⌋ and ⌈2n/3⌉. This result holds for all connected graphs and is tight for certain families (e.g., complete bipartite graphs).

The central technical achievement is a proof that Tron is PSPACE‑complete, confirming the conjecture of Bodlaender and Kloks. The reduction starts from the canonical PSPACE‑hard problem Quantified Boolean Formula (QBF). The authors design a collection of graph gadgets—variable widgets, clause widgets, and logical gate widgets (AND, OR, NOT)—that simulate the quantifier alternation and logical evaluation of a QBF instance. Each variable widget offers two mutually exclusive paths representing true/false assignments; clause widgets are traversable only if at least one of their incident literals is satisfied; gate widgets enforce the correct propagation of truth values through conditional connections.

To keep the construction within realistic graph classes, the authors introduce a “flower” subgraph that acts as a high‑degree buffer, allowing the overall graph to have maximum degree three while preserving planarity. By embedding the gadgets on a planar layout and carefully routing edges, they show that the reduction yields a planar, bounded‑degree graph. Consequently, Tron remains PSPACE‑complete even when restricted to planar graphs of degree ≤ 3, a significant strengthening of the hardness result.

Beyond the general case, the paper investigates several restricted graph families. For trees (acyclic graphs), the game becomes tractable: the authors present a linear‑time algorithm that computes Grundy numbers for each vertex in a bottom‑up fashion, thereby determining the optimal move from any position. This algorithm extends to forests and yields a complete characterization of winning and losing positions on trees.

For 2‑connected planar graphs, the authors prove that an “outer‑face strategy” is optimal. By fixing a planar embedding, a player who consistently moves along the outer face can block the opponent’s progress and force a win in many instances. They formalize this intuition with a structural theorem and provide an O(n²) algorithm that decides the winner on such graphs.

The experimental section validates the theoretical findings on synthetic and real‑world networks. Random Erdős–Rényi graphs, scale‑free Barabási–Albert graphs, and real road‑network data are used to benchmark the PSPACE‑hardness reduction, the tree algorithm, and the planar‑graph heuristic. Results show that while the general game remains intractable on large dense graphs, the specialized algorithms run efficiently on the targeted classes, and the planar heuristic achieves high win rates on low‑degree planar instances.

In conclusion, the paper settles the long‑standing open problem of Tron’s computational complexity, establishes tight extremal bounds for vertex capture, and delineates the boundary between tractable and intractable graph classes. It opens several avenues for future work, including multi‑player extensions, dynamic graphs where vertices or edges may appear or disappear during play, and probabilistic analyses of random game play.