Streaming beyond sketching for Maximum Directed Cut

We give an $\widetilde{O}(\sqrt{n})$-space single-pass $0.483$-approximation streaming algorithm for estimating the maximum directed cut size ($\textsf{Max-DICUT}$) in a directed graph on $n$ vertices. This improves over an $O(\log n)$-space $4/9 < 0.45$ approximation algorithm due to Chou, Golovnev, Velusamy (FOCS 2020), which was known to be optimal for $o(\sqrt{n})$-space algorithms. $\textsf{Max-DICUT}$ is a special case of a constraint satisfaction problem (CSP). In this broader context, our work gives the first CSP for which algorithms with $\widetilde{O}(\sqrt{n})$ space can provably outperform $o(\sqrt{n})$-space algorithms on general instances. Previously, this was shown in the restricted case of bounded-degree graphs in a previous work of the authors (SODA 2023). Prior to that work, the only algorithms for any CSP were based on generalizations of the $O(\log n)$-space algorithm for $\textsf{Max-DICUT}$, and were in particular so-called "sketching" algorithms. In this work, we demonstrate that more sophisticated streaming algorithms can outperform these algorithms even on general instances. Our algorithm constructs a "snapshot" of the graph and then applies a result of Feige and Jozeph (Algorithmica, 2015) to approximately estimate the $\textsf{Max-DICUT}$ value from this snapshot. Constructing this snapshot is easy for bounded-degree graphs and the main contribution of our work is to construct this snapshot in the general setting. This involves some delicate sampling methods as well as a host of "continuity" results on the $\textsf{Max-DICUT}$ behaviour in graphs.