English

Sparse Navigable Graphs for Nearest Neighbor Search: Algorithms and Hardness

Data Structures and Algorithms 2025-07-21 v1

Abstract

We initiate the study of approximation algorithms and computational barriers for constructing sparse α\alpha-navigable graphs [IX23, DGM+24], a core primitive underlying recent advances in graph-based nearest neighbor search. Given an nn-point dataset PP with an associated metric d\mathsf{d} and a parameter α1\alpha \geq 1, the goal is to efficiently build the sparsest graph G=(P,E)G=(P, E) that is α\alpha-navigable: for every distinct s,tPs, t \in P, there exists an edge (s,u)E(s, u) \in E with d(u,t)<d(s,t)/α\mathsf{d}(u, t) < \mathsf{d}(s, t)/\alpha. 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 Ω~(n)\widetilde{\Omega}(n) 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 O(n3)O(n^3)-time (lnn+1)(\ln n + 1)-approximation algorithm, as well as establishing NP-hardness of achieving an o(lnn)o(\ln n)-approximation. Building on this equivalence, we develop faster O(lnn)O(\ln n)-approximation algorithms. The first runs in O~(nOPT)\widetilde{O}(n \cdot \mathrm{OPT}) 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 O(lnn)O(\ln n)-approximation to the sparsest 2α2\alpha-navigable graph, running in O~(nω)\widetilde{O}(n^{\omega}) time. Finally, we complement our upper bounds with a query complexity lower bound, showing that any o(n)o(n)-approximation requires examining Ω(n2)\Omega(n^2) distances. This result shows that in the regime where OPT=O~(n)\mathrm{OPT} = \widetilde{O}(n), our O~(nOPT)\widetilde{O}(n \cdot \mathrm{OPT})-time algorithm is essentially best possible.

Keywords

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}
}
R2 v1 2026-07-01T04:08:10.130Z