On the performance of algorithms for large-scale bound constrained problems
We discuss issues that affect the performance of algorithms for the solution of large-scale bound constrained problems on parallel computers. The discussion centers on the solution of the elastic-plastic torsion problem and the journal bearing problem. These two problems are model large-scale quadratic programming problems that arise as finite element approximations to elliptic variational inequalities. Performance issues are illustrated with the GPCG algorithm of More and Toraldo. This algorithm uses the gradient projection method to select an active set and the conjugate gradient method to explore the active set defined by the current iterate. We show that significant improvements in the performance of the GPCG algorithm can be obtained by using partitioning techniques in a parallel environment. We also show that these partitioning techniques lead to almost linear speedups on function-gradient evaluations and Hessian-vector products for partially separable functions. 12 refs., 2 figs., 8 tabs.
- Research Organization:
- Argonne National Lab. (ANL), Argonne, IL (United States)
- Sponsoring Organization:
- DOE/ER
- DOE Contract Number:
- W-31109-ENG-38
- OSTI ID:
- 6937883
- Report Number(s):
- CONF-8910400-1; ON: DE90010084
- Resource Relation:
- Conference: Workshop on large-scale numerical optimization, Ithaca, NY (USA), 19-20 Oct 1989
- Country of Publication:
- United States
- Language:
- English
Similar Records
Large-scale sequential quadratic programming algorithms
A trust region algorithm for PDE-constrained optimization with bound constraints using reduced order modeling
Related Subjects
TORSION
BOUNDARY-VALUE PROBLEMS
ALGORITHMS
ARRAY PROCESSORS
ELASTICITY
FINITE ELEMENT METHOD
ITERATIVE METHODS
PARALLEL PROCESSING
PERFORMANCE
MATHEMATICAL LOGIC
MECHANICAL PROPERTIES
NUMERICAL SOLUTION
PROGRAMMING
TENSILE PROPERTIES
990200* - Mathematics & Computers