scientific article; zbMATH DE number 3915534
[For the entire collection see Zbl 0553.00001.] Various results on the rate of convergence of variable metric algorithms for unconstrained optimization are discussed. Usually line searches are assumed to be perfect and for the strong results one additionally assumes that the matrices formed by n successively occuring normed search directions are bounded away from singularity. Both assumptions fail for many practical problems. Numerical and theoretical results for problems satisfying resp. not satisfying these assumptions are given. In particular, a weaker convergence result without the above assumption on the search directions is derived and an example shows how the use of step-length one instead of perfect line searches severely deteriorates the rate of convergence under the BFGS-update.
- scientific article; zbMATH DE number 4180681 (Why is no real title available?)
- On the Least Q-order of Convergence of Variable Metric Algorithms
- The convergence of variable metric matrices in unconstrained optimization
- scientific article; zbMATH DE number 3913572 (Why is no real title available?)
- Analogues of Dixon’s and Powell’s Theorems for Unconstrained Minimization with Inexact Line Searches
- Gobally convergent variable metric method for nonconvex nondifferentiable unconstrained minimization
- An example of numerical nonconvergence of a variable-metric method
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3690977)