Class of methods for solving large convex quadratic programs subject to box constraints
Large sparse convex quadratic programs subject only to box constraints (i.e., lower and upper bounds on the variables), called BQP, play an important role in large scale nonlinear optimization. The dual of a large strictly convex quadratic program (subject to general constraints) is a special case of BQP, and such generally-constrained quadratic programs need to be solved efficiently in using the popular Sequential Quadratic Programming methods for solving large nonlinear programs. A conjugate gradient-type heuristic algorithm and a new class of finite algorithms based on such a heuristic were studied. The numerical results with Dembo and Tulowitzki's CRGP algorithm in general and perform better than CRGP for problems that have a low percentage of free variables at optimality. An SSOR preconditioning was also used to improve the efficiency of the heuristic algorithm.
- Research Organization:
- North Carolina Univ., Chapel Hill (USA)
- OSTI ID:
- 5315635
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
Non-convex quadratic programming
Convex Relaxations for Quadratic On/Off Constraints and Applications to Optimal Transmission Switching
Related Subjects
71 CLASSICAL AND QUANTUM MECHANICS
GENERAL PHYSICS
NONLINEAR PROBLEMS
NUMERICAL SOLUTION
OPTIMIZATION
ALGORITHMS
COMPUTER CALCULATIONS
EFFICIENCY
PROGRAMMING
MATHEMATICAL LOGIC
990200* - Mathematics & Computers
658000 - Mathematical Physics- (-1987)