PRQP - Potential Reduction solver for Quadratic Programming

PRQP is a Fortran 90 Interior Point (IP) solver for the solution of large-scale convex Quadratic Programming (QP) problems. It implements an infeasible primal-dual Potential Reduction method (S. Mizuno, M. Kojima, and M.J. Todd, SIAM J. Optim. 5 (1995), 52-67), where the Newton step at each iteration is obtained by solving a KKT (saddle-point) linear system by either the LDL' factorization or the Conjugate Gradient (CG) method with a Constraint Preconditioner (CP). A symmetric QMR solver is also available, for use with inexact CPs. If the QP problem has only bound constraints, the IP method is feasible. Furthermore, the KKT linear systems are reduced to the normal equation form, and the iterative method used is CG with a limited-memory incomplete Cholesky factorization preconditioner. The package uses the MA57 routine from the HSL Library (http://www.hsl.rl.ac.uk/catalogue/ma57.html) for computing the LDL' factorization and solving the associated triangular systems, and the ICFS package (http://www.mcs.anl.gov/~more/icfs/) for computing the limited-memory incomplete Cholesky factorization. PRQP is a "research code", mainly used by the authors to develop and analyze preconditioning techniques for KKT systems arising in IP methods.