We examine the effect of diagonal scaling on the projected gradient method of Moré and Toraldo [SIAM J. Optimization 1, (1991), pp. 93-114] when the constraints are bound constraints and the quadratic form is positive definite. It is shown that scalings causes the method to visit fewer faces and to minimize the function more quickly. Numerical results are given with the journal bearing problem and some dense matrix bound constrained problems.
All Science Journal Classification (ASJC) codes
- Control and Optimization
- Applied Mathematics