High-performance solver for binary quadratic problems

Hrga, Timotej (author), Povh, Janez (author), Wiegele, Angelika (author)

Many problems in combinatorial optimization can be formulated as constrained binary quadratic problems(BQPs),which are in genera lNP hard. We present a method for finding exact solutions of large-scale linearly constrained binary quadratic programming problems. Our exact solution method combines parallelization techniques and exact penalty method approach to obtain optimal solutions of problems of sizes that are due to their size unsolvable by existing method sand tools. We use exact penalty method to first efficiently transform constrained BQP into an instance of max-cut. The obtained problem is then solved by parallel branch-and-bound algorithm that uses SDP based relaxations and is implemented using MPI (distributed memory parallel model) on supercomputer HPC located at Faculty of Mechanical Engineering of the University of Ljubljana. We will present numerical results and demonstrate how to submit a graph instance to new opensource parallel solver BiqBin for (BQPs) which is available as an online service.

• Language: English
• Keywords: high-performance computing, combinatorial optimization, max-cut problem
• Typology: 1.12 - Published Scientific Conference Contribution Abstract
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2018
• Number of pages: Str. 309
• PID: 20.500.12556/RUL-101999
• UDC: 519.8(045)
• COBISS.SI-ID: 16150811
• Publication date in RUL: 19.07.2018

Record is a part of a monograph

• Title: ISMP 2018
• Place of publishing: Bordeaux
• Publisher: University of Bordeaux
• Year: 2018
• COBISS.SI-ID: 16150299

Secondary language

• Language: Slovenian
• Keywords: visoko zmogljivo računalništvo, kombinatorična optimizacija, problem maksimalnega prereza