An Interior Point Algorithm for the General Nonlinear Programming Problem with Trust Region Globalization.

Abstract

This paper presents an SQP-based interior point technique for solving the general nonlinear programming problem using trust region globalization and the Coleman-Li scaling. The SQP subproblem is decomposed into a normal and a reduced tangential subproblem in the tradition of numerous works on equality constrained optimization, and strict feasibility is maintained with respect to the bounds. This is intended to be an extension of previous work by Coleman & Li and Vicente. Though no theoretical proofs of convergence are provided, some computational results are presented which indicate that this algorithm holds promise. The computational experiments have been geared towards improving the semilocal convergence of the algorithm; in particular high sensitivity of the speed of convergence with respect to the fraction of the trust region radius allowed for the normal step and with respect to the initial trust region radius are observed. The chief advantages of this algorithm over primal dual interior point algorithms are better handling of the 'sticking problem' and a reduction in the number of variables by elimination of the multipliers of bound constraints.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 01, 1996
Accession Number
ADA319048

Entities

People

  • Indraneel Das

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Applied Mathematics
  • Computations
  • Computer Programming
  • Convergence
  • Elimination
  • Engineering
  • Equations
  • Evolutionary Algorithms
  • Globalization
  • Inequalities
  • Mathematics
  • Nonlinear Programming
  • Numbers
  • Operations Research
  • Optimization
  • Tensile Strength

Readers

  • Operations Research