Numerical results for SQIC: Software for large-scale quadratic programming∗

Supplementary material for the article “Methods for Convex and General Quadratic Pro-gramming ” is presented in this document. Table 1 lists the values of various tolerances used to obtain the numerical results. Tables 2–13 provide detailed information of the numerical results from the large-scale quadratic programming software package SQIC. A total of 253 QPs were identified from the CUTEst [6] test set. No linear programs were tested because all of the codes under consideration revert to the simplex method when the objective is linear. The QP problems are grouped into two sets based on the final number of superbasic variables obtained by the default solver SQIC-LUSOL. The final number of superbasics can be slightly different when SQIC is used with other linear solvers. A test problem is included in the “large ” set if the final number of superbasics is greater than 1000 or 12 (m + n). The remaining test problems form the “small ” set. The CUTEst set contains 173 small and 80 large problems. A time limit of 5000 seconds was imposed in each case. (In practice, the 5000 second limit is not exact since the time limit is checked every twenty iterations.) Results are presented for SQIC with its default settings using the three linear solver