Implicit extra-update multi-step quasi-Newton methods (Q2627817)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Implicit extra-update multi-step quasi-Newton methods |
scientific article |
Statements
Implicit extra-update multi-step quasi-Newton methods (English)
0 references
31 May 2017
0 references
Summary: A new approach to the implicit update multi-step methods, introduced earlier by \textit{J. A. Ford} and \textit{S. Tharmlikit} [J. Comput. Appl. Math. 152, No. 1--2, 133--146 (2003; Zbl 1025.65035)] and \textit{I. A. R. Moghrabi} [Sib. Zh. Vychisl. Mat. 12, No. 2, 189--200 (2009; Zbl 1212.65229)], is addressed in this paper. Two algorithms are developed. One of the algorithms proposes a method that uses exact quantities in doing the update that are computed cheaply rather than relying on crude approximations, as is the case in the work of \textit{J. A. Ford} [Comput. Math. Appl. 42, No. 8--9, 1083--1091 (2001; Zbl 0989.65063)], Ford and Tharmlikit [Zbl 1025.65035] and Moghrabi [Zbl 1212.65229]. The other method introduced here exploits \textit{M. Al-Baali}'s [Optim. Methods Softw. 13, No. 3, 159--179 (2000; Zbl 0957.90115)] extra update technique and is motivated by the improvements achieved by the application of such methods to the minimisation of nonlinear unconstrained problems. The implicit extra updates idea performs, on iterative basis, implicit updates to the Hessian approximation multiple times at each iteration. The updates are intended to be performed cheaply prior to updating the actual Hessian approximation at each iteration. The implicitly updated matrix serves as a metric for measuring distances among recent iterates, as required in the multistep Methods developed by \textit{J. A. Ford} and \textit{I. A. Moghrabi} [J. Comput. Appl. Math. 50, No. 1--3, 305--323 (1994; Zbl 0807.65062)]. The method displays numerical merits and motivates promising further investigation.
0 references
unconstrained optimisation: quasi-Newton methods
0 references
multi-step updates
0 references
extra updates
0 references
Hessian approximation
0 references
implicit update
0 references