ARRIVAL: Recursive Framework & 𝓁₁-Contraction

Abstract

ARRIVAL is the problem of deciding which out of two possible destinations will be reached first by a token that moves deterministically along the edges of a directed graph, according to so-called switching rules. It is known to lie in NP ∩ CoNP, but not known to lie in 𝖯. The state-of-the-art algorithm due to Gärtner et al. (ICALP `21) runs in time 2^{𝒪(√n log n)} on an n-vertex graph.
We prove that ARRIVAL can be solved in time 2^{𝒪(k log² n)} on n-vertex graphs of treewidth k. Our algorithm is derived by adapting a simple recursive algorithm for a generalization of ARRIVAL called G-ARRIVAL. This simple recursive algorithm acts as a framework from which we can also rederive the subexponential upper bound of Gärtner et al.
Our second result is a reduction from G-ARRIVAL to the problem of finding an approximate fixed point of an 𝓁₁-contracting function f : [0, 1]ⁿ → [0, 1]ⁿ. Finding such fixed points is a well-studied problem in the case of the 𝓁₂-metric and the 𝓁_∞-metric, but little is known about the 𝓁₁-case. 
Both of our results highlight parallels between ARRIVAL and the Simple Stochastic Games (SSG) problem. Concretely, Chatterjee et al. (SODA `23) gave an algorithm for SSG parameterized by treewidth that achieves a similar bound as we do for ARRIVAL, and SSG is known to reduce to 𝓁_∞-contraction.