Abstract:Mixed-integer convex quadratic programs with indicator variables (MIQP) encompass a wide range of applications, from statistical learning to energy, finance, and logistics. The outer approximation (OA) algorithm has been proven efficient in solving MIQP, and the key to the success of an OA algorithm is the strength of the cutting planes employed. In this paper, we propose a new technique for deriving cutting planes for MIQP from various convex relaxations, and, as a result, we develop new OA algorithms for solving MIQP at scale. The contributions of our work are two-fold: (1) we bridge the work on the convexification of MIQP and the algorithm design to solve large-scale problems, and (2) we demonstrate through a computational study on the sparse portfolio selection problem that our algorithms give rise to significant speedups compared with the state-of-the-art methods in the literature.
What problem does this paper attempt to address?
This paper attempts to solve the mixed - integer convex quadratic programming problem (MIQP), especially those with indicator variables. MIQP has wide applications in fields such as statistical learning, energy, finance and logistics. However, even when variables \(x\) and \(y\) are unconstrained, MIQP is still an NP - hard problem. The main contribution of the paper is to propose a new technique to derive cutting planes from various convex relaxations, and based on this, develop a new outer approximation (OA) algorithm to solve MIQP problems on a large scale. Specifically, the contributions of the paper can be summarized in two aspects:
1. **Bridge between theory and algorithm design**: The paper establishes the connection between MIQP convexification theory and algorithm design, especially in solving large - scale problems.
2. **Algorithm performance improvement**: Through computational research on the sparse portfolio selection problem, the paper shows that the proposed algorithm significantly improves the solution speed compared with the state - of - the - art methods in the existing literature.
### Background of the paper and related work
#### 1.1 Background and literature review
The paper first introduces the general form of MIQP:
\[
\begin{aligned}
& \min_{x, y} \quad y^\top Q y+g^\top y + h^\top x \\
& \text{s.t.} \quad A y \leq b, \\
& \quad C y \leq D x, \\
& \quad y_i(1 - x_i)=0, \quad \forall i = 1,\ldots,n, \\
& \quad x \in X\subseteq \{0, 1\}^n,
\end{aligned}
\]
where \(Q\in \mathbb{R}^{n\times n}\) is a positive definite Hessian matrix, and \(X\) is an arbitrary set of the binary variable vector \(x\). The complementary constraint \(y_i(1 - x_i) = 0\) indicates that the indicator variable \(x_i\) controls the activation / deactivation state of the continuous variable \(y_i\).
#### 1.2 Related research
- **Convexification methods**: In recent years, much research has focused on how to convexify MIQP through decomposition methods. For example, the Hessian matrix \(Q\) can be decomposed into the sum of several matrices with "simple" structures. Perspective Reformulation is a commonly used method, which realizes convexification by decomposing the quadratic function into one - dimensional quadratic terms and the remaining parts, and replacing some terms with perspective functions.
- **Exact solution methods**: Existing exact solution methods are mainly divided into two categories: Branch - and - Bound method and Outer Approximation method. The Branch - and - Bound method solves MIQP by solving a series of sub - problems, while the Outer Approximation method generates cutting planes by recursively solving polyhedral approximations.
### Main contributions of the paper
#### 3. Cutting planes based on perspective reformulation
The paper proposes a new framework for deriving projected cutting planes from different MIQP reformulations. In particular, for perspective reformulation, the paper gives an efficient formula to calculate cutting planes, which is more efficient than existing methods and can handle MIQP problems with linear linking constraints.
#### 4. Cutting planes based on rank - one inequalities
The paper further explores the reformulation method using the epigraphical ideal representation of rank - one quadratic functions. By introducing rank - one inequalities, the paper obtains a class of stronger cutting planes, which are stronger than the cutting planes based on perspective reformulation.
#### 5. Algorithm design and experimental verification
The paper designs an outer approximation program based on the above cutting planes and applies it to the sparse portfolio selection problem. The experimental results show that the proposed algorithm is significantly superior to other state - of - the - art methods in terms of solution speed.
### Conclusion
The paper effectively solves large - scale MIQP problems by proposing new cutting - plane techniques and outer approximation algorithms, and shows its superior performance in practical applications. This provides a solution for solving complex optimization problems.