Back to search

Georgia Institute of Technology

Topics in Mixed Integer Nonlinear Optimization

Abstract

dc:description.abstract

Mixed integer nonlinear optimization has many applications ranging from machine learning to power systems. However, these problems are very challenging to solve to global optimality due to the inherent non-convexity. This typically leads the problem to be NP-hard. Moreover, in many applications, there are time and resource limitations for solving real-world problems, and the sheer size of real instances can make solving them challenging. In this thesis, we focus on important elements of nonconvex optimization - including mixed integer linear programming and nonlinear programming, where both theoretical analyses and computational experiments are presented. In the first chapter we look at Mixed Integer Quadratic Programming (MIQP), the problem of minimizing a convex quadratic function over mixed integer points in a rational polyhedron. We utilize the augmented Lagrangian dual (ALD), which augments the usual Lagrangian dual with a weighted nonlinear penalty on the dualized constraints. We first prove that ALD will reach a zero duality gap asymptotically as the weight on the penalty goes to infinity under some mild conditions on the penalty function. We next show that a finite penalty weight is enough for a zero gap when we use any norm as the penalty function. Finally, we prove a polynomial bound on the weight on the penalty term to obtain a zero gap. In the second chapter we apply the technique of lifting to bilinear programming, a special case of quadratic constrained quadratic programming. We first show that, for sets described by one bilinear constraint together with bounds, it is always possible to sequentially lift a seed inequality. To reduce computational burden, we develop a framework based on subadditive approximations of lifting functions that permits sequence-independent lifting of seed inequalities for separable bilinear sets. We then study a separable bilinear set where the coefficients form a minimal cover with respect to the right-hand-side. For this set, we introduce a bilinear cover inequality, which is second-order cone representable. We study the lifting function of the bilinear cover inequality and lift fixed variable pairs in closed-form, thus deriving a lifted bilinear cover inequality that is valid for general separable bilinear sets with box constraints. In the third chapter we continue our research around separable bilinear programming. We first prove that the semidefinite programming relaxation provides no benefit over the McCormick relaxation for separable bilinear optimization problems. We then design a simple randomized separation heuristic for lifted bilinear cover inequalities. In our computational experiments, we separate many rounds of these inequalities starting from the McCormick relaxation of bilinear instances where each constraint is a separable bilinear constraint set. Our main result is to demonstrate that there is a significant improvement in the performance of a state-of-the-art global solver in terms of the gap closed, when these inequalities are added at the root node compared to when these inequalities are not added. In the fourth chapter we look at Mixed Integer Linear Programming (MILP) that arises in operational applications. Many routinely-solved MILPs are extremely challenging not only from a worst-case complexity perspective, but also because of the necessity to obtain good solutions within limited time. An example is the Security-Constrained Unit Commitment (SCUC) problem, solved daily to clear the day-ahead electricity markets. We develop ML-based methods for improving branch-and-bound variable selection rules that exploit key features of such operational problems: similar decisions are generated within the same day and across different days, based on the same power network. Utilizing similarity between instances and within an instance, we build one separate ML model per variable or per group of similar variables for learning to predict the strong branching score. The approach is able to produce branch-and-bound trees which gap closed only slightly worse than that of trees obtained by strong branching, while it outperforms previous machine learning schemes.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Industrial and Systems Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Gu, Xiaoyi
Advisor dc:contributor.advisor
  • Dey, Santanu S.
Committee members dc:contributor.committeemember
  • Richard, Jean-Philippe P.
  • Toriello, Alejandro
  • Cifuentes Pardo, Diego F
  • Sahinidis, Nikolaos

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1853/67235

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Gu, Xiaoyi. Topics in Mixed Integer Nonlinear Optimization. Doctoral thesis, Georgia Institute of Technology, 2022. http://hdl.handle.net/1853/67235