Settling SETH vs. Approximate Sparse Directed Unweighted Diameter (up to (NU)NSETH)

We prove several tight results on the fine-grained complexity of approximating the diameter of a graph. First, we prove that, for any $\varepsilon>0$, assuming the Strong Exponential Time Hypothesis (SETH), there are no near-linear time $2-\varepsilon$-approximation algorithms for the Diameter of a sparse directed graph, even in unweighted graphs. This result shows that a simple near-linear time 2-approximation algorithm for Diameter is optimal under SETH, answering a question from a survey of Rubinstein and Vassilevska-Williams (SIGACT '19) for the case of directed graphs. In the same survey, Rubinstein and Vassilevska-Williams also asked if it is possible to show that there are no $2-\varepsilon$ approximation algorithms for Diameter in a directed graph in $O(n^{1.499})$ time. We show that, assuming a hypothesis called NSETH, one cannot use a deterministic SETH-based reduction to rule out the existence of such algorithms. Extending the techniques in these two results, we characterize whether a $2-\varepsilon$ approximation algorithm running in time $O(n^{1+\delta})$ for the Diameter of a sparse directed unweighted graph can be ruled out by a deterministic SETH-based reduction for every $\delta\in(0,1)$ and essentially every $\varepsilon\in(0,1)$, assuming NSETH. This settles the SETH-hardness of approximating the diameter of sparse directed unweighted graphs for deterministic reductions, up to NSETH. We make the same characterization for randomized SETH-based reductions, assuming another hypothesis called NUNSETH. We prove additional hardness and non-reducibility results for undirected graphs.