Graph Operations on Parity Games and Polynomial-Time Algorithms

Parity games are games that are played on directed graphs whose vertices are labeled by natural numbers, called priorities. The players push a token along the edges of the digraph. The winner is determined by the parity of the greatest priority occurring infinitely often in this infinite play. A motivation for studying parity games comes from the area of formal verification of systems by model checking. Deciding the winner in a parity game is polynomial time equivalent to the model checking problem of the modal mu-calculus. Another strong motivation lies in the fact that the exact complexity of solving parity games is a long-standing open problem, the currently best known algorithm being subexponential. It is known that the problem is in the complexity classes UP and coUP. In this paper we identify restricted classes of digraphs where the problem is solvable in polynomial time, following an approach from structural graph theory. We consider three standard graph operations: the join of two graphs, repeated pasting along vertices, and the addition of a vertex. Given a class C of digraphs on which we can solve parity games in polynomial time, we show that the same holds for the class obtained from C by applying once any of these three operations to its elements. These results provide, in particular, polynomial time algorithms for parity games whose underlying graph is an orientation of a complete graph, a complete bipartite graph, a block graph, or a block-cactus graph. These are classes where the problem was not known to be efficiently solvable. Previous results concerning restricted classes of parity games which are solvable in polynomial time include classes of bounded tree-width, bounded DAG-width, and bounded clique-width. We also prove that recognising the winning regions of a parity game is not easier than computing them from scratch.

💡 Research Summary

Parity games are two‑player infinite‑duration games played on directed graphs whose vertices carry natural‑number priorities. The winner of a play is determined by the parity (even or odd) of the greatest priority that appears infinitely often. Because the decision problem for parity games is polynomial‑time equivalent to model checking the modal μ‑calculus, it occupies a central place in formal verification. Despite intense research, the exact computational complexity of solving parity games remains open: the best known algorithms run in sub‑exponential time, and the problem is known to lie in both UP and co‑UP.

The present paper adopts a structural‑graph‑theoretic viewpoint and asks: for which graph families can parity games be solved in polynomial time? Rather than focusing on width parameters (tree‑width, DAG‑width, clique‑width) that have been extensively studied, the authors investigate three elementary graph operations—join, pasting at a vertex, and addition of a new vertex—and prove that these operations preserve polynomial‑time solvability. Formally, let C be any class of directed graphs on which parity games can be solved in time O(poly(n, m, p)) (n vertices, m edges, p distinct priorities). The paper shows that the class obtained by applying any one of the following operations to members of C also admits a polynomial‑time solution:

1. Join: Given two digraphs G₁ and G₂, create a new digraph whose vertex set is V(G₁) ∪ V(G₂) and whose edge set consists of all edges of G₁, all edges of G₂, plus every possible directed edge between a vertex of G₁ and a vertex of G₂ (both directions).
2. Pasting: Identify a vertex v₁ of G₁ with a vertex v₂ of G₂ (renaming them to a single vertex v) and merge the two graphs at this vertex. The operation may be repeated, producing a tree‑like composition of blocks sharing single vertices.
3. Vertex addition: Insert a fresh vertex u into a graph G and connect u to any subset of existing vertices with arbitrarily oriented edges; the priority and owner of u are chosen arbitrarily.

The core technical contribution is a suite of invariants that allow the winner regions of the original subgames to be “lifted” to the combined game with only a limited amount of extra work. For the join operation, the algorithm first solves the two subgames independently, obtaining their respective winning regions W₁ and W₂. Because all cross‑edges are added, the only way the joint game’s outcome can differ from the disjoint solution is if a cross‑edge creates a new cycle whose maximal priority changes the parity condition. The authors show that checking this possibility reduces to a linear‑time scan of the cross‑edges, preserving overall polynomial complexity. For pasting, the shared vertex v acts as a “gateway” between the components; the algorithm solves each component, then resolves the status of v by examining the two local solutions, which can be done in constant additional time per pasting step. For vertex addition, the impact of the new vertex is confined to its incident edges; the algorithm evaluates whether the new vertex belongs to the winning region of the player who owns it by a simple local test, again incurring only O(deg(u)) overhead.

By iterating these operations, the authors obtain polynomial‑time algorithms for several previously uncharted families:

• Orientations of complete graphs – every pair of vertices is connected by both directed arcs. Such graphs can be built by repeatedly joining single‑vertex graphs, so the join preservation result applies directly.
• Orientations of complete bipartite graphs – the two partite sets are fully connected in both directions; this is also a special case of the join operation.
• Block graphs – each biconnected component (block) is a clique, and blocks intersect in at most one vertex. Block graphs arise by repeatedly pasting cliques at single vertices.
• Block‑cactus graphs – these are graphs whose blocks are either cliques or simple cycles, arranged in a cactus‑like fashion (any two cycles share at most one vertex). They can be generated by a combination of joins (to create cliques) and pastings (to attach cycles), so the preservation theorems cover them as well.

For each of these families the paper presents explicit algorithms whose running time is bounded by a low‑degree polynomial in the size of the input game (typically O(n·m·p) or better). The results significantly broaden the landscape of tractable parity‑game instances beyond the width‑bounded regimes studied earlier.

In addition to the algorithmic contributions, the paper addresses a conceptual question: is recognizing whether a given vertex lies in a player’s winning region easier than computing the entire winning region? The authors prove that the two tasks are polynomial‑time equivalent. Their reduction shows that, given an oracle that decides membership for a single vertex, one can reconstruct the full winning region by a polynomial number of queries, and conversely, a full solution trivially yields membership answers. Consequently, any hardness result for the full problem automatically transfers to the recognition problem.

Overall, the paper introduces a novel, operation‑centric methodology for extending polynomial‑time solvability of parity games. By demonstrating that the join, pasting, and vertex‑addition operations preserve tractability, the authors provide a systematic way to construct large, dense graph families (including complete and complete‑bipartite orientations) on which parity games can be solved efficiently. This complements existing width‑based approaches and offers fresh avenues for both theoretical exploration and practical algorithm design in verification and automata theory.