A Note on Pushdown Automata Systems

In (Csuhaj-Varju et. al. 2000) Parallel Communicating Systems of Pushdown Automata (PCPA) were introduced and shown to be able to simulate nondeterministic one-way multi-head pushdown automata in returning mode, even if communication is restricted to be one-way having a single target component. A simulation of such centralized PCPA by one-way multi-head pushdown automata (Balan 2009) turned out to be incomplete (Otto 2012). Subsequently it was shown that centralized returning PCPA are universal even if the number of components is two (Petersen 2013) and thus are separated from one-way multi-head pushdown automata. Another line of research modified the definition of PCPA such that communication is asynchronous (Otto 2013). While the simulation of one-way multi-head pushdown automata is still possible, now a converse construction shows this model in returning mode to be equivalent to the one-way multi-head pushdown automaton in a very precise sense. It was left open, whether non-centralized returning PCPA of degree two are universal. In the first part of the paper we show this to be the case. Then we turn our attention to Uniform Distributed Pushdown Automata Systems (UDPAS). These systems of automata work in turn on a single tape. We show that UPDAS accepting with empty stack do not form a hierarchy depending on the number of components and that the membership problem is complete in NP, answering two open problems from (Arroyo et. al. 2012).

💡 Research Summary

The paper consists of two main contributions concerning two families of distributed pushdown automata.

First, it settles an open question about the computational power of non‑centralized returning parallel communicating pushdown automata (PCPA) of degree two. Earlier work had shown that centralized returning PCPA with two components are universal, but it was unknown whether the same holds when the system is not centralized (i.e., there is no single target component). The author proves universality by constructing a deterministic two‑component PCPA that simulates an arbitrary Post machine, which is known to accept exactly the recursively enumerable languages. The simulation works as follows: component 1 reads the input, stores it reversed on its stack, and repeatedly executes a “cycle” that mimics one step of the Post machine. In each cycle component 1 pops the top stack symbol, writes the special response symbol R, and then waits for component 2 to receive the whole stack content (returning mode empties the source stack). Component 2 is kept in an accepting state and simply copies the received stack back to component 1, allowing the next cycle to start. By faithfully reproducing the three types of Post‑machine instructions (halt, test, and assignment) the constructed PCPA accepts exactly the language of the simulated Post machine. Because the construction uses only deterministic pushdown automata, the result shows that even deterministic, non‑centralized returning PCPA of degree two are computationally universal.

The second part of the paper addresses Uniform Distributed Pushdown Automata Systems (UDPAS). A UDPAS consists of several identical pushdown automata that work in turn on a common input tape; when the active automaton blocks, the next one becomes active. The author resolves two open problems left by Arroyo et al. (2012).

1. No hierarchy with respect to the number of components.
   By choosing a finite non‑empty language M over a single‑letter alphabet (e.g., M = {a, aa, aaa}) and assuming that a UDPAS with n components accepts M, Lemma 1 of the cited work would imply that M must be expressible as Lⁿ for some language L. Since M contains words of different lengths, such a representation is impossible, leading to a contradiction. Hence increasing the number of components does not yield strictly larger language families; the class of languages accepted by UDPAS is independent of the component count.

2. NP‑completeness of the membership problem for UDPAS with empty‑stack acceptance.
   The problem is in NP because a nondeterministic Turing machine can guess a distribution of the input symbols among the copies of the underlying pushdown automaton and verify acceptance locally. For NP‑hardness the author reduces the well‑known NP‑complete shuffle‑membership problem for two context‑free languages. Given two pushdown automata A and B, the construction creates extended automata A′, B′ (adding self‑loops on a new symbol #) and then duplicates each state to obtain A′′, B′′. A new automaton C combines the control units of A′′ and B′′ and uses two special symbols # and $ to switch between the two copies. For an input word w the reduction builds a transformed word
     w′ = # $ # w₁ # w₂ … # wₙ,
   and asks whether two copies of C accept w′. The construction guarantees that w belongs to the shuffle of L(A) and L(B) iff w′ is accepted by the UDPAS consisting of the two copies of C. Consequently, the shuffle‑membership problem reduces to UDPAS membership, establishing NP‑hardness.

Together, these results demonstrate that (i) non‑centralized deterministic returning PCPA of degree two are as powerful as any Turing‑equivalent model, and (ii) UDPAS, despite their simple turn‑based operation, have a rich structural property (no hierarchy) and a computationally hard membership problem (NP‑complete). The paper closes with a brief discussion of possible extensions, such as characterising context‑free languages whose shuffle can be accepted by a UDPAS with empty‑stack acceptance.