A Constructive Arboricity Approximation Scheme

The arboricity $Γ$ of a graph is the minimum number of forests its edge set can be partitioned into. Previous approximation schemes were nonconstructive, i.e., they only approximated the arboricity as a value without computing a corresponding forest partition. This is because they operate on the related pseudoforest partitions or the dual problem of finding dense subgraphs. We propose an algorithm for converting a partition of $k$ pseudoforests into a partition of $k+1$ forests in $O(mk\log k + m \log n)$ time with a data structure by Brodal and Fagerberg that stores graphs of arboricity $k$. A slightly better bound can be given when perfect hashing is used. When applied to a pseudoforest partition obtained from Kowalik's approximation scheme, our conversion implies a constructive $(1+ε)$-approximation algorithm with runtime $O(m \log n \log Γ\, ε^{-1})$ for every $ε> 0$. For fixed $ε$, the runtime can be reduced to $O(m \log n)$. Our conversion also implies a near-exact algorithm that computes a partition into at most $Γ+2$ forests in $O(m\log n \,Γ\log^* Γ)$ time. It might also pave the way to faster exact arboricity algorithms. We also make several remarks on approximation algorithms for the pseudoarboricity and the equivalent graph orientations with smallest maximum indegree, and correct some mistakes made in the literature.