Solving a Tropical Optimization Problem via Matrix Sparsification

An optimization problem, which arises in various applications as that of minimizing the span seminorm, is considered in the framework of tropical mathematics. The problem is to minimize a nonlinear function defined on vectors over an idempotent semifield, and calculated by means of multiplicative conjugate transposition. We find the minimum of the function, and give a partial solution which explicitly represents a subset of solution vectors. We characterize all solutions by a system of simultaneous equation and inequality, and exploit this characterization to investigate properties of the solutions. A matrix sparsification technique is developed to extend the partial solution to a wider solution subset, and then to a complete solution described as a family of subsets. We offer a backtracking procedure that generates all members of the family, and derive an explicit representation for the complete solution. Numerical examples and graphical illustrations of the results are presented.