Branching-time model checking of one-counter processes

One-counter processes (OCPs) are pushdown processes which operate only on a unary stack alphabet. We study the computational complexity of model checking computation tree logic (CTL) over OCPs. A PSPACE upper bound is inherited from the modal mu-calculus for this problem. First, we analyze the periodic behaviour of CTL over OCPs and derive a model checking algorithm whose running time is exponential only in the number of control locations and a syntactic notion of the formula that we call leftward until depth. Thus, model checking fixed OCPs against CTL formulas with a fixed leftward until depth is in P. This generalizes a result of the first author, Mayr, and To for the expression complexity of CTL’s fragment EF. Second, we prove that already over some fixed OCP, CTL model checking is PSPACE-hard. Third, we show that there already exists a fixed CTL formula for which model checking of OCPs is PSPACE-hard. To obtain the latter result, we employ two results from complexity theory: (i) Converting a natural number in Chinese remainder presentation into binary presentation is in logspace-uniform NC^1 and (ii) PSPACE is AC^0-serializable. We demonstrate that our approach can be used to obtain further results. We show that model-checking CTL’s fragment EF over OCPs is hard for P^NP, thus establishing a matching lower bound and answering an open question of the first author, Mayr, and To. We moreover show that the following problem is hard for PSPACE: Given a one-counter Markov decision process, a set of target states with counter value zero each, and an initial state, to decide whether the probability that the initial state will eventually reach one of the target states is arbitrarily close to 1. This improves a previously known lower bound for every level of the Boolean hierarchy by Brazdil et al.

💡 Research Summary

The paper investigates the computational complexity of model checking Computation Tree Logic (CTL) over one‑counter processes (OCPs), a subclass of push‑down systems whose stack alphabet consists of a single symbol. While a PSPACE upper bound follows from known results for the modal μ‑calculus, the authors go far beyond this baseline by analysing the structural behaviour of CTL formulas on OCPs and by establishing tight lower bounds.

A central contribution is the introduction of the leftward‑until depth (LUD), a syntactic measure that captures how deeply “until” operators are nested towards the left (i.e., towards the past) within a CTL formula. By exploiting the periodic nature of OCP configurations, the authors devise an algorithm whose running time is exponential only in the number of control locations and the LUD of the formula. Consequently, when the LUD is fixed, model checking any OCP against such a CTL formula can be performed in polynomial time. This result generalizes earlier work on the EF fragment, showing that the whole of CTL enjoys a similar fixed‑parameter tractability with respect to LUD.

On the hardness side, the paper presents three distinct PSPACE‑hardness proofs. First, it shows that for a fixed OCP, CTL model checking is PSPACE‑hard, by encoding quantified Boolean formulas into the counter dynamics of that process. Second, it proves that for a fixed CTL formula, the problem remains PSPACE‑hard when the OCP varies. This proof cleverly combines two complexity‑theoretic facts: (i) converting a number given in Chinese‑remainder representation to binary lies in log‑space uniform NC¹, and (ii) PSPACE is AC⁰‑serializable. By using (i) to encode counter values efficiently and (ii) to decompose PSPACE computations into a sequence of AC⁰ evaluations, the authors simulate any PSPACE computation within the model‑checking task.

The paper also settles the exact difficulty of the EF fragment: model checking EF over OCPs is Pⁿᴾ‑hard, matching the known upper bound and answering an open question left by earlier work of the first author, Mayr, and To.

Finally, the authors extend their techniques to probabilistic systems. They consider one‑counter Markov decision processes (MDPs) with a set of target states that must be reached with counter value zero. They prove that deciding whether the probability of eventually hitting a target can be made arbitrarily close to 1 is PSPACE‑hard. This improves upon previously known lower bounds that only placed the problem at various levels of the Boolean hierarchy.

Overall, the work provides a comprehensive picture of CTL model checking over OCPs: (1) an upper bound that becomes polynomial for formulas of bounded leftward‑until depth, (2) several PSPACE‑hardness results that hold even when either the system or the formula is fixed, (3) a precise Pⁿᴾ‑hardness classification for the EF fragment, and (4) a new PSPACE‑hardness result for probabilistic one‑counter MDPs. The methodological blend of automata‑theoretic periodicity arguments with fine‑grained complexity tools (NC¹ conversions, AC⁰ serializability) is likely to inspire further investigations into the verification of other restricted infinite‑state models.