Sparse Navigable Graphs for Nearest Neighbor Search: Algorithms and Hardness
Abstract
We initiate the study of approximation algorithms and computational barriers for constructing sparse -navigable graphs [IX23, DGM+24], a core primitive underlying recent advances in graph-based nearest neighbor search. Given an -point dataset with an associated metric and a parameter , the goal is to efficiently build the sparsest graph that is -navigable: for every distinct , there exists an edge with . We consider two natural sparsity objectives: minimizing the maximum out-degree and minimizing the total size. We first show a strong negative result: the slow-preprocessing version of DiskANN (analyzed in [IX23] for low-doubling metrics) can yield solutions whose sparsity is times larger than optimal, even on Euclidean instances. We then show a tight approximation-preserving equivalence between the Sparsest Navigable Graph problem and the classic Set Cover problem, obtaining an -time -approximation algorithm, as well as establishing NP-hardness of achieving an -approximation. Building on this equivalence, we develop faster -approximation algorithms. The first runs in time and is thus much faster when the optimal solution is sparse. The second, based on fast matrix multiplication, is a bicriteria algorithm that computes an -approximation to the sparsest -navigable graph, running in time. Finally, we complement our upper bounds with a query complexity lower bound, showing that any -approximation requires examining distances. This result shows that in the regime where , our -time algorithm is essentially best possible.
Cite
@article{arxiv.2507.14060,
title = {Sparse Navigable Graphs for Nearest Neighbor Search: Algorithms and Hardness},
author = {Sanjeev Khanna and Ashwin Padaki and Erik Waingarten},
journal= {arXiv preprint arXiv:2507.14060},
year = {2025}
}