Power of Nondetreministic JAGs on Cayley graphs

The Immerman-Szelepcsenyi Theorem uses an algorithm for co-stconnectivity based on inductive counting to prove that NLOGSPACE is closed under complementation. We want to investigate whether counting is necessary for this theorem to hold. Concretely, we show that Nondeterministic Jumping Graph Autmata (ND-JAGs) (pebble automata on graphs), on several families of Cayley graphs, are equal in power to nondeterministic logspace Turing machines that are given such graphs as a linear encoding. In particular, it follows that ND-JAGs can solve co-st-connectivity on those graphs. This came as a surprise since Cook and Rackoff showed that deterministic JAGs cannot solve st-connectivity on many Cayley graphs due to their high self-similarity (every neighbourhood looks the same). Thus, our results show that on these graphs, nondeterminism provably adds computational power. The families of Cayley graphs we consider include Cayley graphs of abelian groups and of all finite simple groups irrespective of how they are presented and graphs corresponding to groups generated by various product constructions, including iterated ones. We remark that assessing the precise power of nondeterministic JAGs and in particular whether they can solve co-st-connectivity on arbitrary graphs is left as an open problem by Edmonds, Poon and Achlioptas. Our results suggest a positive answer to this question and in particular considerably limit the search space for a potential counterexample.