DAG-width is PSPACE-complete

TL;DR

This paper proves that DAG-width, a graph complexity measure, can require super-polynomial size decompositions and that deciding if the DAG-width is below a constant is PSPACE-complete, highlighting computational hardness.

Contribution

It demonstrates the super-polynomial size of optimal DAG decompositions and establishes PSPACE-completeness of DAG-width decision problem.

Findings

Existence of graphs with super-polynomial DAG decomposition size

No polynomial approximation for optimal DAG decompositions

Deciding DAG-width ≤ constant is PSPACE-complete

Abstract

Berwanger et al. show that for every graph $G$ of size $n$ and DAG-width $k$ there is a DAG decomposition of width $k$ and size $n^{O(k)}$. This gives a polynomial time algorithm for determining the DAG-width of a graph for any fixed $k$. However, if the DAG-width of the graphs from a class is not bounded, such algorithms become exponential. This raises the question whether we can always find a DAG decomposition of size polynomial in $n$ as it is the case for tree width and all generalisations of tree width similar to DAG-width. We show that there is an infinite class of graphs such that every DAG decomposition of optimal width has size super-polynomial in $n$ and, moreover, there is no polynomial size DAG decomposition which would approximate an optimal decomposition up to an additive constant. In the second part we use our construction to prove that deciding whether the DAG-width of a given graph is at most a given constant is PSPACE-complete.