Abstract. An algorithm is constructed for recognizing the circulant graphs and finding a canonical labeling for them in polynomial time. This algorithm also yields a cycle base of an arbitrary solvable permutation group. The consistency of the algorithm is based on a new result on the structure of Schur rings over a finite cyclic group. §1. Introduction
A finite graph1 is said to be circulant if its automorphism group contains a full cycle, i.e., a permutation the cycle decomposition of which consists of a unique cycle of full length. This means that the graph admits a regular cyclic automorphism group, and, consequently, is isomorphic to a Cayley graph over a cyclic group. In particular, any circulant graph can be specified in a compact form by a full cycle automorphism and a neighborhood of some vertex. One of the main computational problems concerning circulant graphs is that of finding an efficient algorithm to recognize them. (This problem is a special case of the following NP-complete problem: test whether or not a given graph has an automorphism without fixed points [15].) The first attempt to solve this problem was undertaken in [24], where a polynomial-time algorithm for recognizing circulant tournaments was described. In the subsequent papers [21,22,5] several results on recognizing some special classes of circulant graphs were presented, but the general problem remained open up to now. In the present paper we solve this problem completely. Another problem about circulant graphs is to find an efficient isomorphism test for them. In fact, this problem is polynomial-time reducible to the recognition problem, because two circulant graphs with the same number of vertices are isomorphic if and only if their disjoint union is a circulant graph. In this paper we present a solution to a more difficult problem of finding a canonical labeling for circulant graphs.3 It should be mentioned that the isomorphism problem for Cayley graphs over a cyclic group (which is a special case of the isomorphism problem for circulant graphs) has been extensively studied through the last forty years (see [20]