Parallel algorithms for a class of convex optimization problems
This thesis is principally concerned with a piecewise-linear trust region method for solving a class of structured convex optimization problems, which includes the traffic-assignment problems. Piecewise-linear approximation of nonlinear convex objective functions in linearly constrained optimization produces subproblems that may be solved as linear programs. In order to have additional control of the accuracy of the piecewise-linear approximation, two devices are considered: rectangular trust regions and dynamic scaling. The use of rectangular trust regions in conjuction with the type of piecewise-linear approximation considered actually serves to simplify rather than complicate the approximating problems. The approach to dynamic scaling considered may be applied to problems in which each objective function term is a convex function of a linear function of the variables. Another emphasis is the development of parallel algorithms suited to distributed computing and the comparison of the relative efficiencies of these algorithms on different architectures. Computational experience is cited for some large-scale problems arising from traffic-assignment applications.
- Research Organization:
- Wisconsin Univ., Madison (USA)
- OSTI ID:
- 6930113
- Country of Publication:
- United States
- Language:
- English
Similar Records
An algorithm for linearizing convex extremal problems
A new algorithm for structural optimization problems