The New Spectral Conjugate Gradient Method for Minimization

• Published in: 2022 International Conference on Data Science and Intelligent Computing (ICDSIC) pp. 273 - 277
• Main Authors: Mohammad T, Saja O., Hamed, Eman T.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-11-2022
• Subjects: conjugate gradient method; Data science; Dolan-Mor'e performance; globally convergence; Gradient methods; Graphics; Minimization; Optimization; Search problems; unconstrained optimization; Wolfe conditions
• DOI: 10.1109/ICDSIC56987.2022.10075933

In this paper, we proposed a spectral conjugate gradient formula based on the Fletcher-Reeves (FR) formula to solve the unconstrained optimization problems, which when some hypotheses are fulfilled, the sufficient descent condition for the search direction of this proposed formula is proven. Also, when using Wolfe's step length finding conditions, our proposed method showed under some conditions that the new search direction is globally convergent. When applying the proposed method to a number of standard optimization functions, we obtained excellent numerical results compared with the basic conjugate gradient methods. To clarify these comparisons, the Dolan-More method was used to calculate the efficiency of the proposed method with intent to the tools used (iter, fgcnt, time).