Abstract
Classical liveness in $\omega$-regular games guarantees that a desired event will eventually occur, but it fails to bound the delay between a request and its response. This paper investigates "finitary winning," a stronger condition requiring a fixed, though initially unknown, bound on response times. We explore finitary parity and Streett games, establishing their determinacy, complexity, and the memory requirements for winning strategies. Notably, while the system player can win finitary parity games with memoryless strategies, the environment may require infinite memory.
Introduction
In the design of reactive systems—such as elevator controllers or communication protocols—requirements are often modeled as infinite games played on graphs between a "System" player and an "Environment" player. A core component of these specifications is liveness: the promise that "something good eventually happens." However, the classical mathematical definition of liveness is weak; it allows the time between a request and a response to grow without bound, potentially rendering a system practically useless.
The authors address this by studying finitary liveness, which requires that there exists some bound $b$ such that every request is satisfied within $b$ steps. While $b$ does not need to be known upfront, it must remain fixed throughout the infinite play.
Methods
The authors focus on two standard types of winning conditions: Parity and Streett objectives. Parity games assign integer priorities to states, where the goal is to ensure the minimum priority seen infinitely often is even. Finitary parity games refine this by requiring a bounded distance between an odd priority and a subsequent lower even priority.
To solve these games, the authors introduce "distance sequences" to track response times and develop a recursive algorithm, FinitaryParity. For finitary Streett games, which involve multiple request-response pairs, they use an Index Appearance Record (IAR) construction to reduce the problem to a finitary parity game on a larger graph.
Key Insights and Results
A striking finding is the memory asymmetry between players. In finitary parity games, Player 1 (the system) can always win using a memoryless strategy. Conversely, Player 2 (the environment) may require infinite memory to win, essentially needing to "stretch" the response time longer in each successive round to defeat any fixed bound.
The paper proves that both finitary parity and Streett games are determined. Furthermore, the "Boundedness Theorem" shows that if Player 1 can win a finitary parity game on a graph with $n$ states, they can guarantee a response within $n$ steps. This provides a concrete upper bound derived from the system's size.
Complexity analysis places the problem of deciding the winner in finitary parity games in NP $\cap$ coNP, the same class as classical parity games. The proposed algorithm solves these games in $O(n^{2d-3} \cdot d \cdot m)$ time, where $n$ is the number of states, $m$ is the number of edges, and $d$ is the number of priorities. Finitary Streett games are shown to be decidable in EXPTIME.
Conclusion
Finitary winning provides a more robust framework for reactive system specification by imposing practical constraints on liveness. While challenges remain—such as the state space explosion in Streett-to-Parity reductions and the search for a polynomial-time algorithm for parity games—the theoretical results regarding memoryless strategies and concrete boundedness offer significant steps toward more reliable automated synthesis.
Summary updated