A tool to help fix infeasible optimization models

What is it about?

Optimization problems are routinely formulated and solved to answer challenging questions in decision making, science and engineering. When these optimization models are infeasible, detailed domain knowledge and expertise are often required to determine the cause of infeasibility. More often than not, infeasibility in optimization formulations is due to incorrect modelling or incorrect use of parameters. Our work implements algorithms for automated isolation of small infeasible parts of an optimization model, that can aid the expert in resolving causes of infeasibilities in a model. We further propose a novel deletion presolve algorithm as a preprocessing step that augments the capabilities of these infeasibility isolation algorithms. Compilation of an extensive test library, detailed computational tests, and results demonstrate the value of this work.

Featured Image

Why is it important?

As the size of optimization models solved in practice has increased in recent years, so has the difficulty associated with diagnosing and fixing infeasible models. An automated method for isolating infeasibilities in optimization can save tremendous amounts of time and effort toward resolving infeasibilities. Our work makes our infeasibility diagnosis algorithms available to a broad array of users through the global solver BARON. BARON is accessible through commercial vendors as well as freely through the NEOS server for optimization.

Perspectives