Strong formulations for conic quadratic optimization with indicator variables

• 成果类型: Article
• 署名作者: Gomez, Andres
• 署名单位: University of Southern California
• 刊物名称: MATHEMATICAL PROGRAMMING
• ISSN/ISSBN: 0025-5610
• DOI: 10.1007/s10107-020-01508-y
• 发表日期: 2021
• 页码: 193-226
• 关键词: perspective cuts; robust solutions; RISK; inequalities; PROGRAMS; reformulations; SURFACES

摘要

We study the convex hull of the mixed-integer set given by a conic quadratic inequality and indicator variables. Conic quadratic terms are often used to encode uncertainties, while the indicator variables are used to model fixed costs or enforce sparsity in the solutions. We provide the convex hull description of the set under consideration when the continuous variables are unbounded. We propose valid nonlinear inequalities for the bounded case, and show that they describe the convex hull for the two-variable case. All the proposed inequalities are described in the original space of variables, but extended SOCP-representable formulations are also given. We present computational experiments demonstrating the strength of the proposed formulations.