Interval Digraphs: An Analogue of Interval Graphs- M. Sen S. Das DEPARTMENT OF MATHEMATlCS NORTH BENGAL UNlVERSlTY DARJEELING, INDlA A.B. Roy DEPARTMENT OF MATHEMATICS JADA VPUR UN I VERSl TY CALCUTTA, lNDIA D.B. West DEPARTMENT OF MATHEMATlCS UNlVERSITY OF ILLINOIS, URBANA, ILLlNOlS /DA/CRD PRINCETON, NEW JERSEY ABSTRACT Intersection digraphs analogous to undirected intersection graphs are in- troduced. Each vertex is assigned an ordered pair of sets, with a directed edge uu in the intersection digraph when the "source set" of u inter- sects the "terminal set" of u. Every n-vertex digraph is an intersection digraph of ordered pairs of subsets of an n-set, but not every digraph is an intersection digraph of convex sets in the plane. Interval digraphs are those having representations where all sets are intervals on the real line. Interval digraphs are characterized in terms of the consecutive ones property of certain matrices, in terms of the adjacency matrix and in terms of Ferrers digraphs. In particular, they are intersections of pairs of' Ferrers digraphs whose union is a complete digraph. 1. INTRODUCTION The concepts of intersection graph and interval graph have been well studied for undirected graphs. Given a family of sets, each is assigned to a vertex, and the intersection graph of the family of sets has an edge between two of these Journal of Graph Theory, Vol. 13, No. 2, 189-202 (1989) 0 1989 by John Wiley & Sons, Inc. CCC 0364-90241891020189-14$04.00 190 JOURNAL OF GRAPH THEORY vertices if and only if the corresponding sets intersect. A graph is an interval gruph if it is the intersection graph of a family of intervals on the real line. In- terval graphs have a long and rich history. A recent issue of Discrete Mathe- matics 151 was dedicated to papers on this subject, and additional references can be found therein. In this paper, we introduce and study a natural analogue of these concepts for directed graphs D (henceforth “digraphs”) with vertex set V and edge set E. We consider a family V of ordered pairs of sets, and to each ordered pair we assign a vertex u. The first set assigned to u is called its .source set S,, and the second is its terminal set (or sink set) T,. The intersection digraph of a family of or- dered pairs of sets is the digraph such that uu € E if and only if S,, n T, # @. Note that loops are allowed, but there are no multiple edges. By analogy with interval graphs, we define an intervul digraph to be an intersection digraph of a family of ordered pairs of intervals on the real line; i.e., each source set and terminal set is an interval. Several characterizations are known for interval graphs 18,9,12]; our aim in this paper is to give analogous characterizations for interval digraphs. Fulkerson and Gross [S] defined a 0,l -matrix to have the consecutive ones property for rows if its columns can be permuted so that the ones in each row appear con- secutively. They proved that a graph is an interval graph if and only if the inci- dence matrix between its vertices and maximal complete subgraphs has the consecutive ones property for rows. In Section 3, we obtain a similar character- ization for interval digraphs using a simultaneous consecutive ones property for the rows of two incidence matrices. A class of intersection digraphs was introduced and studied by Maehara [ 13 J. He defined a pointed set to be a set S with a distinguished “base point” b E S. Phrased in our terminology, he defined the catch digraph of a family of pointed sets {(S,),b,,)}to be the intersection digraph in which the source set for u is S, and the sink set is b,,;i.e., uu E E if and only if b, E S,,. When the source sets are required to be intervals, this is a class of interval digraphs. Note that b, E S, forces a loop at each vertex. Dropping the requirement b,, E S, yields an intcr- mediate family between the catch digraphs of intervals and the general interval digraphs; we call these intervul-point digraphs. Recall that the adjacency matrix A@) of a digraph D with vertices numbered u,,. ,u, is the 0,I-matrix with a 1 in position i,j.if and only if u,uJis an edge. Maehara characterized the catch digraphs of families of pointed intervals as those whose adjacency matrix has the consecutive ones property for rows, withour al- lowing column permutations. (Actually, his definition of catch digraph discards loops, and his characterization then adds a loop at each vertex before testing for consecutive ones.) In Section 3, we characterize the interval-point digraphs; dropping the requirement b, E S, corresponds to allowing column permutations in testing for the consecutive ones property of the adjacency matrix. In Section 4 we obtain more difficult characterizations of interval digraphs. We show that D is an interval digraph if and only if A(D) has (independent) row and column permutations so that every 0 can be replaced by one of {C,R} INTERVAL DIGRAPHS 191 in such a way that every R has all R’s to its right and every C has all C’s below it. At the same time, we characterize interval digraphs in terms of a special class of digraphs. Introduced independently by Guttman [ 111 and Riguet [ 171, Ferrers digraphs are those whose successor sets are linearly ordered by inclusion, where the successor ser of u is its set of out-neighbors {u E V:uu E E}. (See [16] for an early survey on Ferrers digraphs and related classes and [2] for additional char- acterizations.) It is easy to see that the successor sets are linearly ordered by in- clusion if and only if the analogously defined predecessor sets are ordered by inclusion, and that both are equivalent to the transformability of the adjacency matrix by independent row and column permutations to a 0,I-matrix in which the 1’s are clustered in the lower left in the shape of a Ferrers diagram (hence the term “Ferrers digraph’). (All three conditions are equivalent to forbidding any 2 by 2 submatrix of the adjacency matrix to be a permutation matrix.) We prove that D is an interval digraph if and only if it is the intersection of two Ferrers digraphs whose union is a complete digraph. Intersections of Ferrers digraphs have been studied previously, but without the requirement that the union be complete. The Ferrers dimension of D is de- fined to be the minimum number of Ferrers digraphs whose intersection is D. By our characterization, the Ferrers dimension of an interval digraph is at most 2. The digraphs with Ferrers dimension 2 have been characterized, indepen- dently by Cogis [3] and Doignon, Ducamp, and Falmagne [4] in different con- texts. This characterization yields a polynomial algorithm for testing whether a digraph has Ferrers dimension at most 2; it is possible that our characterization will allow this to be converted into a recognition algorithm for interval digraphs. In Section 5, we translate Cogis’s condition to an adjacency matrix condition for Ferrers dimension 2 that is analogous to our adjacency matrix condition for interval digraphs. We then construct an example to show that not every digraph of Ferrers dimension 2 is an interval digraph. We begin in Section 2 with remarks about general intersection digraphs. 2. INTERSECTION DIGRAPHS It is well known that every finite undirected graph is an intersection graph of finite sets. The simplest construction is to use a set whose elements correspond to the edges of G and assign to each vertex the elements corresponding to its incident edges. Since vertices are adjacent if and only if they share an incident edge, G is the intersection graph of these finite sets. The analogous construc- tion works for directed graphs. If S, consists of the edges with u as source and 7‘” consists of the edges with u as terminus, then S, fl T, # 0 if and only if uu E E. For undirected graphs, it is easy to obtain a more “efficient” representation by using cliques larger than single edges to cover the edges. Indeed, the inrer- section number i#(G) of an undirected graph G is defined to be the minimum 192 JOURNAL OF GRAPH THEORY size of a set U such that G is the intersection graph of subsets of U, and Erdos, Goodman, and Posa [6] showed that the intersection number of G equals the minimum number of complete subgraphs needed to cover its edges. They also proved that i#(G) 5 Ln2/4] for n-vertex graphs, achieved by G = Kl l To develop arialogous results for digraphs, we define a generulized complete bipartite subdigruph (abbreviated GBS) to be a subdigraph generated by vertex sets X, Y, whose edges are all xy such that x E X, y E Y. We say “generalized” because X, Y need not be disjoint, which means that loops may arise. Let the intersection number i#(D) of a digraph be the minimum size of U such that C is the intersection digraph of ordered pairs of subsets of U. The analogue of the Erdos-Goodman-Posa results is as follows: Theorem 1. The intersection number of a digraph equals the minimum num- ber of GBSs required to cover its edges, and the best possible upper bound on this is n for n-vertex graphs. Proof. Suppose {(XI,Y,)} with k members is a minimum collection of GBSs whose union is D.