Faster shortest-path algorithms using the acyclic-connected tree

This paper gives a fixed-parameter linear algorithm for the single-source shortest path problem (SSSP) on directed graphs. The parameter in question is the nesting width, a measure of the extent to which a graph can be represented as a nested collection of graphs. We present a novel directed graph decomposition called the acyclic-connected tree (A-C tree), which breaks the graph into a recursively nested sequence of strongly connected components in topological order. We prove that the A-C tree is optimal in the sense that its width, the size of the largest nested graph, is equal to the nesting width of the graph. We then provide a linear-time algorithm for constructing the A-C tree of any graph. Finally, we show how the A-C tree allows us to construct a simple variant of Dijkstra's algorithm which achieves a time complexity of $O(e+n\log w)$, where $n$ ($e$) is the number of nodes (arcs) in the graph and $w$ is the nesting width. The idea is to apply the shortest path algorithm separately to each component in the order dictated by the A-C tree. We obtain an asymptotic improvement over Dijkstra's algorithm: when $w=n$, our algorithm reduces to Dijkstra's algorithm, but it is faster when $w \in o(n)$, and linear-time for classes of graphs with bounded width, such as directed acyclic graphs.