Minimum-coordinate elimination on a simplex: polynomial first-step safety and coNP-complete correctness
math.OC
Submitted: 2026-10-24
Updated: 2026-10-24
Comments: 25 pages
License: http://creativecommons.org/licenses/by/4.0/
The gist: We study a deletion rule for strictly convex quadratic minimization on a simplex: compute the affine minimizer on the current face, permanently remove a most-negative coordinate, and repeat.
Abstract
We study a deletion rule for strictly convex quadratic minimization on a simplex: compute the affine minimizer on the current face, permanently remove a most-negative coordinate, and repeat. Correctness is required for every linear term and every resolution of ties. We characterize first-step safety by a finite maximum principle for the constrained inverse, and correctness from every initial face by principal-inverse row inequalities. Correctness from the full simplex is a different property. The two properties coincide with at most four coordinates, but an explicit rational five-coordinate example separates them. An exact reachability condition explains the separation: boundary equalities in the constrained inverse restrict the tasks that can be entered on a facet. Away from these equalities, full-start and hereditary correctness coincide. We also determine the recognition complexity. First-step safety is decidable in polynomial bit time, by combining classical strict n-step-vector recognition with diagonal regularization. In contrast, full-start and hereditary correctness are coNP-complete, even on a reduction family with strict column diagonal maxima on every face. The proofs include the non-strict boundary and all affine anchors; explicit rational certificates for the finite examples are given in the text.
Related papers
- An Accelerated Proximal Gradient Method for Bures-Wasserstein Space
- FOMTools: Development and AI-Assisted Code Generation for First-Order Methods
- Lions and Muons: Optimization via Stochastic Frank-Wolfe under Heavy-Tailed Noise
- Dynamic Service Recommendation with Congestion-Dependent Joining: Near-Optimal and Constant-Factor Approximation Algorithms
- Integrating Learning and Optimization for Open Transport Capacity Pooling under Disruption
- Zeroth-Order Gradient Estimators under Decision-Dependent Noise