Detecting and Handling Reflection Symmetries in Mixed-Integer (Nonlinear) Programming

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autore principale: Hojny, Christopher
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909201596416000
author Hojny, Christopher
author_facet Hojny, Christopher
contents Symmetries in mixed-integer (nonlinear) programs (MINLP), if not handled appropriately, are known to negatively impact the performance of (spatial) branch-and-bound algorithms. Usually one thus tries to remove symmetries from the problem formulation or is relying on a solver that automatically detects and handles symmetries. While modelers of a problem can handle various kinds of symmetries, automatic symmetry detection and handling is mostly restricted to permutation symmetries. This article therefore develops techniques such that also black-box solvers can automatically detect and handle a broader class of symmetries. Inspired from geometric packing problems such as the kissing number problem, we focus on reflection symmetries of MINLPs. We develop a generic and easily applicable framework that allows to automatically detect reflection symmetries for MINLPs. To handle this broader class of symmetries, we discuss generalizations of state-of-the-art methods for permutation symmetries, and develop dedicated symmetry handling methods for special reflection symmetry groups. Our symmetry detection framework has been implemented in the open-source solver SCIP and we provide a comprehensive discussion of the implementation. The article concludes with a detailed numerical evaluation of our symmetry handling methods when solving MINLPs.
format Preprint
id arxiv_https___arxiv_org_abs_2405_08379
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Detecting and Handling Reflection Symmetries in Mixed-Integer (Nonlinear) Programming
Hojny, Christopher
Optimization and Control
Symmetries in mixed-integer (nonlinear) programs (MINLP), if not handled appropriately, are known to negatively impact the performance of (spatial) branch-and-bound algorithms. Usually one thus tries to remove symmetries from the problem formulation or is relying on a solver that automatically detects and handles symmetries. While modelers of a problem can handle various kinds of symmetries, automatic symmetry detection and handling is mostly restricted to permutation symmetries. This article therefore develops techniques such that also black-box solvers can automatically detect and handle a broader class of symmetries. Inspired from geometric packing problems such as the kissing number problem, we focus on reflection symmetries of MINLPs. We develop a generic and easily applicable framework that allows to automatically detect reflection symmetries for MINLPs. To handle this broader class of symmetries, we discuss generalizations of state-of-the-art methods for permutation symmetries, and develop dedicated symmetry handling methods for special reflection symmetry groups. Our symmetry detection framework has been implemented in the open-source solver SCIP and we provide a comprehensive discussion of the implementation. The article concludes with a detailed numerical evaluation of our symmetry handling methods when solving MINLPs.
title Detecting and Handling Reflection Symmetries in Mixed-Integer (Nonlinear) Programming
topic Optimization and Control
url https://arxiv.org/abs/2405.08379