This page provides some performance profiles in which intervenes the SQPpro solver. These profiles have been generated within the Libopt environment. The figures have a better appearance when they are viewed without reduction in a separate window.

1. Large scale unconstrained optimization problems.
   
   Here are performance profiles comparing SQPpro with L-BFGS and M1QN3 on the set of 61 unconstrained problems of the CUTEr collection, having between 1000 and 10000 variables. All the solvers use an inverse l-BFGS approximation of the Hessian of the cost function, with 20 updates. The stopping criterion is a gradient sup-norm less than 1.e-5. The number of gradient evaluations is used as performance criterion on the left and the CPU time on the right.
     

2. Large scale bound constrained optimization problems.
   
   Here are performance profiles comparing SQPpro with L-BFGS-B on the set of 78 bound constrained problems of the CUTEr collection, which have between 1 and 11130 variables. Both solvers use a direct l-BFGS approximation of the Hessian of the cost function, with 20 updates. The stopping criterion is a projected gradient sup-norm less than 1.e-5. The number of gradient evaluations is used as performance criterion on the left and the CPU time on the right.