Summary: This paper modifies our earlier linearly convergent algorithm for solving the quadratically constrained quadratic programming problem. It focuses on the judicious use of heuristics for circumventing computationally intensive steps in the original procedure. In particular, feasible points of convex sets that are required by the algorithm are obtained by the use of quadratic solutions and relaxation techniques. We also examine alternative stopping criteria, including a new procedure for generating lower bounds on the optimal objective value based on nearest points to boxes. Computational experience with over thirthy-five test problems indicates that the problem size has little effect on the number of iterations, and the execution times depend more on the number of constraints.
For the entire collection see [Zbl 0806.00046].