Path following in the exact penalty method of convex programming (Q493687): Difference between revisions

The authors propose a following path method for constrained convex programming problems in the context of an exact penalty method. The path is parameterized by the penalty constant and consists of the optimal solutions of the corresponding penalized problems; the parameter increases from its initial value \(0\). The path is built by numerically solving a differential equation satisfied by the Lagrange multipliers of the penalized problems. The method is illustrated by several examples, including quadratic programming, geometric programming and semidefinite programming. The authors have developed a \textsc{Matlab} code, which they offer to share with interested researchers.