Id: arxiv/2611.15492 Url: https://arxiv.org/abs/2611.15492 Tags: math.OC Score: 0 Title: Probabilistically Exact Convexification of Mixed-Integer Quadratic Optimization with Indicators Domain: math.OC Source: arxiv Authors: Salar Fattahi, Andres Gomez Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-11-11 Page: https://delphi.science/r/arxiv/2611.15492 We study the convexification of mixed-integer quadratic optimization with indicator variables. The problem remains NP-hard even when the quadratic matrix has bandwidth two, posing a fundamental obstacle to tractable exact convexification. To overcome this challenge in realistic settings, we introduce probabilistically exact (inner) convexification, a framework that relaxes the requirement of exactness for every objective while preserving exact optimality for statistically relevant ones. Our approach is motivated by inference applications, where the objective is determined by a data vector…