Ill conditioning in LP and MILP models remains a challenge for optimization practitioners. The condition number of a square matrix provides a measure of the level of ill conditioning, but it offers limited insight into causes or remedies. Several LP and MILP solvers offer functionality that provides explanations of infeasibility, but until now, none have offered similar functionality to provide concise explanations of ill conditioning.

In this hands-on webinar, Dr. Ed Klotz describes gurobi-modelanalyzer, an open-source repository containing an ill conditioning explainer that filters out rows or columns of an ill conditioned basis matrix. The resulting submatrix often facilitates a diagnosis of the source of ill conditioning that would be significantly more difficult if examining the entire basis matrix. While the underlying computational methodology will be discussed, the tutorial will focus on how to interpret the explanations, using examples based on publicly available models.