Abstract
Directed Acyclic Graphs (DAGs) are directed graphs in which there is no path from a vertex to itself. DAGs are an omnipresent data structure in computer science and the problem of counting the DAGs of a given number of vertices has been solved in the 70’s by Robinson. In many applications one needs to construct connected DAGs and to control their number of edges, but the adaptation of Robinson’s enumeration to take this into account led to counting formulas based on the inclusion-exclusion principle, inducing a high computational cost for the uniform random sampling of DAGs based on this formula.In the present paper we propose two contributions. First we enumerate a new class of DAGs, enriched with an independent ordering of the children of each vertex, according to their numbers of vertices and edges. We obtain a constructive recursive counting formula for them (i.e. without using the inclusion-exclusion principle) using a new decomposition scheme. Then we show the applicability of our method by proposing a constructive enumeration of Robinson’s labelled DAGs, by vertices and edges, based on the same decomposition. As a consequence we are able to derive efficient uniform random samplers for both models.
Highlights
Directed Acyclic Graphs (DAGs) are directed graphs in which there is no path from a vertex to itself
We propose an alternative model of unlabelled DAGs which we call Directed Ordered Acyclic Graphs (DOAGs) that are similar to regular unlabelled DAGs with additional structure on the edges: the set of outgoing edges of each vertex is totally ordered
We introduce a model of directed acyclic graphs called “Directed Ordered Acyclic Graphs” which is similar to the classical model of unlabelled DAGs but where, in addition, we have a total order on the outgoing edges of each vertex
Summary
Directed Acyclic Graphs (DAGs) are directed graphs in which there is no path (sequence of incident edges) from a vertex to itself. We propose an alternative model of unlabelled DAGs which we call Directed Ordered Acyclic Graphs (DOAGs) that are similar to regular unlabelled DAGs with additional structure on the edges: the set of outgoing edges of each vertex is totally ordered This “local” ordering of the edges captures a structure that is naturally present in other DAG-related mathematical objects such as compressed logical formulas, where the order of the arguments of a function is relevant, or plane tree-like data structures with sharing [3, 12]. We obtain a constructive recursive formula enumerating vertex-labelled DAGs with one sink and one source, counted by number of vertices and edges. To the best of our knowledge, the efficient uniform generation of unlabelled DAGs is still an open problem
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have