Short Paths in the Planar Graph Product Structure Theorem

TL;DR

This paper improves the planar graph product structure theorem by providing tighter bounds on the path length, showing that planar graphs can be embedded into products with nearly optimal path lengths related to their size and treewidth.

Contribution

The paper proves new upper bounds on the length of the path in the product structure theorem for planar graphs, nearly matching known lower bounds and extending the theorem's applicability.

Findings

Path length bound of O((1/ε) * n^{(1+ε)/2}) for planar graphs

Almost tight bounds matching lower bounds of Ω(n^{1/2})

Product structure with treewidth 3 and logarithmic path length for large graphs

Abstract

The Planar Graph Product Structure Theorem of Dujmovi\'c et al. [J. ACM '20] says that every planar graph $G$ is contained in $H\boxtimes P\boxtimes K_3$ for some planar graph $H$ with treewidth at most 3 and some path $P$. This result has been the key to solving several old open problems. Several people have asked whether the Planar Graph Product Structure Theorem can be proved with good upper bounds on the length of $P$. No $o(n)$ upper bound was previously known for $n$-vertex planar graphs. We answer this question in the affirmative, by proving that for any $\epsilon\in (0,1)$ every $n$-vertex planar graph is contained in $H\boxtimes P\boxtimes K_{O(1/\epsilon)}$, for some planar graph $H$ with treewidth 3 and for some path $P$ of length $O(\frac{1}{\epsilon}n^{(1+\epsilon)/2})$. This bound is almost tight since there is a lower bound of $\Omega(n^{1/2})$ for certain $n$-vertex planar graphs. In fact, we prove a stronger result with $P$ of length $O(\frac{1}{\epsilon}\,\textrm{tw}(G)\,n^{\epsilon})$, which is tight up to the $O(\frac{1}{\epsilon}\,n^{\epsilon})$ factor for every $n$-vertex planar graph $G$. Finally, taking $\epsilon=\frac{1}{\log n}$, we show that every $n$-vertex planar graph $G$ is contained in $H\boxtimes P\boxtimes K_{O(\log n)}$ for some planar graph $H$ with treewidth at most 3 and some path $P$ of length $O(\textrm{tw}(G)\,\log n)$. This result is particularly attractive since the treewidth of the product $H\boxtimes P\boxtimes K_{O(\log n)}$ is within a $O(\log^2n)$ factor of the treewidth of $G$.