Hartigan's Method: k-means Clustering without Voronoi

Telgarsky, Vattani2010Vector quantization and Voronoi theory

Hartigan one-point terminal geometry (30 Aug 2026)

Does not transfer

its objective is additive SSE. It gives no nuisance block, profiled criterion, information price, or population funnel, and in particular does not make an exchange-stable partition a Voronoi partition.

Telgarsky & Vattani study greedy one-point relocation for finite ordinary k-means, including strict descent and the geometry separating a terminal partition from its induced Voronoi partition. It is the nearest published precedent for the exchange-stability notion used in the DS16 measurements.

AISTATS 2010, PMLR 9:820-827. Finite one-point relocation geometry and separation from the induced Voronoi partition for ordinary k-means; no profiled-Fisher or population-funnel result.

Cited by

Relevant to

Claims this source bears on without being cited in their proofs.