Conjugate Grids for Unconstrained Optimisation

Authors:
D. Byatt;I. D. Coope;C. J. Price
Affiliations:
Department of Mathematics and Statistics, University of Canterbury, Private Bag 4800, Christchurch, New Zealand. d.byatt@math.canterbury.ac.nz;Department of Mathematics and Statistics, University of Canterbury, Private Bag 4800, Christchurch, New Zealand. i.coope@math.canterbury.ac.nz;Department of Mathematics and Statistics, University of Canterbury, Private Bag 4800, Christchurch, New Zealand. c.price@math.canterbury.ac.nz
Venue:
Computational Optimization and Applications
Year:
2004

Citing 5
Cited 1

`` Direct Search'' Solution of Numerical and Statistical Problems

Journal of the ACM (JACM)
On the Convergence of Grid-Based Methods for Unconstrained Optimization

SIAM Journal on Optimization
On the Convergence of Pattern Search Algorithms

SIAM Journal on Optimization
A convergent variant of the Nelder-Mead algorithm

Journal of Optimization Theory and Applications
Positive Bases in Numerical Optimization

Computational Optimization and Applications

A new quasi-Newton pattern search method based on symmetric rank-one update for unconstrained optimization

Computers & Mathematics with Applications

Quantified Score

Hi-index	0.00

Visualization

Abstract

Several recent papers have proposed the use of grids for solving unconstrained optimisation problems. Grid-based methods typically generate a sequence of grid local minimisers which converges to stationary points under mild conditions.In this paper the location and number of grid local minimisers is calculated for strictly convex quadratic functions in two dimensions with certain types of grids. These calculations show it is possible to construct a grid with an arbitrary number of grid local minimisers. The furthest of these can be an arbitrary distance from the quadratic's minimiser. These results have important implications for the design of practical grid-based algorithms.Grids based on conjugate directions do not suffer from these problems. For such grids only the grid points closest (depending on the choice of metric) to the minimiser are grid local minimisers. Furthermore, conjugate grids are shown to be reasonably stable under mild perturbations so that in practice, only approximately conjugate grids are required.