Comparative analysis of three selected classical conjugate gradient coefficients under the Armijo inexact line search for unconstrained optimization

Hasbullah, Siti Nur’Ain and Abdul Ghani, Nur Hamizah and Wan Sulaiman, Wan Khadijah (2026) Comparative analysis of three selected classical conjugate gradient coefficients under the Armijo inexact line search for unconstrained optimization. Mathematics Letters, 5 (2): 8. pp. 97-109. ISSN eISSN: 2948-3735
Abstract

Optimization plays an important role in solving unconstrained optimization problems that arise in various scientific and engineering applications. Among the available optimization techniques, the conjugate gradient (CG) method is widely used due to its low memory requirement and computational efficiency. The performance of the CG method is highly influenced by the selection of the conjugate gradient coefficient and the line search strategy. This study aims to implement selected classical CG methods, namely Hestenes–Stiefel (HS), Polak–Ribière–Polyak (PRP), and Belloufi–Rahali–Benzine (BRB). These methods use the Armijo inexact line search for solving unconstrained optimization problems. The methods were implemented and tested on a collection of 20 benchmark optimization problems with different dimensions and initial points. Their performance was evaluated based on the number of iteration and CPU time. The numerical results were analysed using performance profiles to compare the efficiency of the three methods. The findings indicate that the HS method consistently outperformed the PRP and BRB methods by solving a larger proportion of test problems with fewer iterations and shorter times. Hence, the HS method is considered the most efficient and reliable among the selected methods under the Armijo inexact line search. This study demonstrates the effectiveness of classical CG methods in unconstrained optimization and provides a useful reference for future research on optimization algorithms.

Item Details
Edit Item
Edit Item
Downloads & Files
[thumbnail of 149389.pdf]
Text
149389.pdf
Download (446kB)
Indexing & Metrics