Open AccessA Globally Convergent Hybrid FR-PRP Conjugate Gradient Method for Unconstrained Optimization Problems
Author(s) -
O. J. Adeleke,
Idowu Ademola Osinuga,
Rafiq Raji
Publication year - 2022
Publication title -
wseas transactions on mathematics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.211
H-Index - 21
eISSN - 2224-2880
pISSN - 1109-2769
DOI - 10.37394/23206.2021.20.78
Subject(s) - conjugate gradient method , line search , mathematics , benchmark (surveying) , convergence (economics) , conjugacy class , nonlinear conjugate gradient method , gradient method , mathematical optimization , gradient descent , conjugate residual method , descent (aeronautics) , derivation of the conjugate gradient method , algorithm , computer science , combinatorics , artificial intelligence , computer security , geodesy , aerospace engineering , economic growth , economics , artificial neural network , engineering , radius , geography
In this paper, a new conjugate gradient (CG) parameter is proposed through the convex combination of the Fletcher-Reeves (FR) and Polak-Ribiére-Polyak (PRP) CG update parameters such that the conjugacy condition of Dai-Liao is satisfied. The computational efficiency of the PRP method and the convergence profile of the FR method motivated the choice of these two CG methods. The corresponding CG algorithm satisfies the sufficient descent property and was shown to be globally convergent under the strong Wolfe line search procedure. Numerical tests on selected benchmark test functions show that the algorithm is efficient and very competitive in comparison with some existing classical methods.