← Back to all news
arXiv Machine Learning By Yunbum Kook

Beyond the $d^{2.5}$-mixing bound for Dikin walks on polytopes

Read the original on arXiv Machine Learning →

arXiv:2607. 13943v1 Announce Type: cross Abstract: Inspired by interior-point methods (IPM) for structured convex optimization, Kannan and Narayanan introduced the Dikin walk for sampling uniformly from polytopes in 2009.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Aug 31

On two proofs of $d^2$ mixing of weighted Dikin walks

The paper investigates the mixing time of weighted Dikin walks used for sampling from exponential distributions on polytopes and truncated positive-semidefinite cones. It presents a general total-variation mixing bound under conditions of strong self-concordance, ν-symmetry, and mixed-trace regularity, achieving an “~O(d^2)" bound for polytopes and “~O(d^4)" for truncated PSD cones. A second result introduces a fourth-order bootstrap condition that yields stronger χ^2-divergence guarantees and an improved “~O(d^2)" mixing bound for a scaled Lee–Sidford metric.

By Yuansi Chen, Yunbum Kook
More like this →