Abstract:Decades of advances in mixed-integer linear programming (MILP) and recent development in mixed-integer second-order-cone programming (MISOCP) have translated very mildly to progresses in global solving nonconvex mixed-integer quadratically constrained programs (MIQCP). In this paper we propose a new approach, namely Compact Disjunctive Approximation (CDA), to approximate nonconvex MIQCP to arbitrary precision by \textit{convex} MIQCPs, which can be solved by MISOCP solvers. For nonconvex MIQCP with $n$ variables and $m$ general quadratic constraints, our method yields relaxations with at most $O(n\log(1/\varepsilon))$ number of continuous/binary variables and linear constraints, together with $m$ \textit{convex} quadratic constraints, where $\varepsilon$ is the approximation accuracy. The main novelty of our method lies in a very compact lifted mixed-integer formulation for approximating the (scalar) square function. This is derived by first embedding the square function into the boundary of a three-dimensional second-order cone, and then exploiting rotational symmetry in a similar way as in the construction of BenTal-Nemirovski approximation. We further show that this lifted formulation characterize the union of finite number of simple convex sets, which naturally relax the square function in a piecewise manner with properly placed knots. We implement (with JuMP) a simple adaptive refinement algorithm. Numerical experiments on synthetic instances used in the literature show that our prototypical implementation (with hundreds of lines of Julia code) can already close a significant portion of gap left by various state-of-the-art global solvers on more difficult instances, indicating strong promises of our proposed approach.

Error analysis of approximation algorithm for standard bi-quadratic programming

Conic approximation to nonconvex quadratic programming with convex quadratic constraints

Approximating Quadratic 0-1 Programming via SOCP

Nonlinearly Constrained Best Approximation in Hilbert Spaces: the Strong CHIP and the Basic Constraint Qualification.

Adaptive computable approximation to cones of nonnegative quadratic functions

Convex Relaxations for Nonconvex Quadratically Constrained Quadratic Programming: Matrix Cone Decomposition and Polyhedral Approximation

Sparse Conic Reformulation of Structured QCQPs based on Copositive Optimization with Applications in Stochastic Optimization

Quadratic Optimization over A Polyhedral Cone

Tighter bound estimation for efficient biquadratic optimization over unit spheres

Conic Approximation to Quadratic Optimization with Linear Complementarity Constraints.

An Improved Lower Bound and Approximation Algorithm for Binary Constrained Quadratic Programming Problem

An Improved Analysis of Semidefinite Approximation Bound for Nonconvex Nonhomogeneous Quadratic Optimization with Ellipsoid Constraints

Exact Computable Representation of Some Second-Order Cone Constrained Quadratic Programming Problems

Constraint Qualification, the Strong CHIP, and Best Approximation with Convex Constraints in Banach Spaces.

Tightening a Copositive Relaxation for Standard Quadratic Optimization Problems

Computable Representation of the Cone of Nonnegative Quadratic Forms over A General Second-Order Cone and Its Application to Completely Positive Programming

Compact Disjunctive Approximations to Nonconvex Quadratically Constrained Programs

Simplex QP-based methods for minimizing a conic quadratic objective over polyhedra

Globally Solving Quadratic Programs with Convex Objective and Complementarity Constraints Via Completely Positive Programming

Approximating optimal solutions to biconvex parametric programs

A new inexact SQP algorithm for nonlinear systems of mixed equalities and inequalities