Polyhedral-based Methods for Mixed-Integer SOCP in Tree Breeding

Sena Safarina,Tim J. Mullin,Makoto Yamashita

DOI: https://doi.org/10.48550/arXiv.1805.03809

2018-05-10

Abstract:Optimal contribution selection (OCS) is a mathematical optimization problem that aims to maximize the total benefit from selecting a group of individuals under a constraint on genetic diversity. We are specifically focused on OCS as applied to forest tree breeding, when selected individuals will contribute equally to the gene pool. Since the diversity constraint in OCS can be described with a second-order cone, equal deployment in OCS can be mathematically modeled as mixed-integer second-order cone programming (MI-SOCP). If we apply a general solver for MI-SOCP, non-linearity embedded in OCS requires a heavy computation cost. To address this problem, we propose an implementation of lifted polyhedral programming (LPP) relaxation and a cone-decomposition method (CDM) to generate effective linear approximations for OCS. In particular, CDM successively solves OCS problems much faster than generic approaches for MI-SOCP. The approach of CDM is not limited to OCS, so that we can also apply the approach to other MI-SOCP problems.

Optimization and Control

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to select the optimal contributors from candidate individuals while maintaining genetic diversity in forest tree breeding. Specifically, the paper focuses on the Equal Deployment Problem (EDP), that is, when a certain number of individuals are selected, these individuals are required to make equal contributions to the gene pool.

### Background and Problem Description
During the tree breeding process, in order to ensure the health of trees and the potential for future genetic gain, genetic diversity needs to be considered during the selection stage. The goal of Optimal Contribution Selection (OCS) is to maximize the total economic benefit under the condition of satisfying genetic diversity constraints. According to the different types of contributions, the OCS problem can be divided into the Unequal Deployment Problem (UDP) and the Equal Deployment Problem (EDP). Among them, EDP requires that the contributions of the selected individuals to the gene pool must be equal.

### Mathematical Model
For EDP, the mathematical optimization model can be expressed as:
\[
\begin{aligned}
& \text{maximize} & & g^T y / N \\
& \text{subject to} & & e^T y = N, \\
& & & y_i \in \{0, 1\} \quad \text{for } i = 1, \ldots, m, \\
& & & y^T A y \leq 2\theta,
\end{aligned}
\]
where:
- \( y \in \mathbb{R}^m \) is the decision variable, representing the contribution of each candidate individual.
- \( g \in \mathbb{R}^m \) is the Estimated Breeding Values (EBVs), representing the genetic value of the candidate individuals.
- \( e \in \mathbb{R}^m \) is a vector of all 1s, indicating that the total contribution of all candidate individuals is 1.
- \( A \in \mathbb{R}^{m \times m} \) is a numerical relationship matrix, representing the genetic relationships between individuals.
- \( \theta \in \mathbb{R}^+ \) is the threshold of genetic diversity.

### Deficiencies of Existing Methods
Existing methods such as GENCONT and dsOpt have some problems when solving EDP:
- **GENCONT**: Although it generates solutions relatively fast, it is often not the optimal solution, and sometimes it cannot meet the number of selected individuals.
- **dsOpt**: Although it can obtain accurate optimal solutions, it generates a large number of sub - problems in the branch - and - bound framework, resulting in excessive computation time.

### Solutions Proposed in the Paper
The paper proposes a new method based on Lifted Polyhedral Programming Relaxation (LPP) and Cone Decomposition Method (CDM) to solve the EDP problem.

#### Lifted Polyhedral Programming Relaxation (LPP)
The LPP method approximates the second - order cone through polyhedral relaxation, thereby removing non - linear constraints. The specific steps include:
1. Convert the original problem into a Mixed - Integer Second - Order Cone Programming (MI - SOCP) problem.
2. Use the polyhedral relaxation method to generate linear constraints to approximate the second - order cone.
3. Reduce the number of linear constraints by adding the Active Constraint Selection Method (ACSM).

#### Cone Decomposition Method (CDM)
The CDM method generates effective linear approximations based on the cutting - plane method and the Lagrange multiplier method. The specific steps include:
1. Initialize a Mixed - Integer Linear Programming (MI - LP) problem without quadratic constraints.
2. Calculate the orthogonal projection through the Lagrange multiplier method to generate cutting planes.
3. Iteratively add cutting planes until the stop condition is met.

### Experimental Results