Coresets for Continuous -Center in Hyperbolic Space
arXiv:2606.16061
Abstract
We study the continuous -center problem in fixed-dimensional hyperbolic space. The input is a finite point set , while the centers may be placed anywhere in the ambient space. A nonempty subset is an -coreset if every exact optimal continuous -center solution for is a -approximate solution for . Our construction uses Gonzalez's farthest-first traversal, whose radius gives a constant-factor estimate of the optimum. When is bounded, we map each cluster to the origin of the Poincaré ball model and apply a local Euclidean grid. For larger , we partition each cluster into radial shells and project them onto representative hyperbolic spheres, avoiding a discretization of the ambient ball. By the hyperbolic law of cosines, the intersection of a hyperbolic ball with each sphere corresponds to a spherical cap. A recursive -cap coverage certificate then replaces each shell with a radius-independent set of input witnesses that preserves its farthest-point behavior for every set of at most centers. The resulting coreset has size and satisfies for every optimal continuous -center solution of . It can be constructed in time. Consequently, for fixed , , and , the coreset has constant size and can be computed in linear time. We complement this upper bound by showing that, for every fixed , every , and all sufficiently small , there are instances for which every -coreset has size .
arXiv commentA counterexample to the methodology in the previous version was identified, and the argument has been replaced with a new spherical-cap certificate approach. A coreset-size lower bound has also been added