A Method for Solving Certain Quadratic Programming Problems Arising in Nonsmooth Optimization

We present a finite algorithm for minimizing a piecewise linearconvex function augmented with a simple quadratic term. To solvethe dual problem, which is of least-squares form with an additionallinear term, we include in a standard active-set quadratic programmingalgorithm a new column-exchange strategy for treating positivesemidefinite problems. Numerical results are given for an implementationusing the Cholesky factorization.