Font Size:
New Hybrid Conjugate Gradient Method for Unconstrained Optimization
Last modified: 2014-02-05
Abstract
A new hybrid conjugate gradient algorithm is considered. Werequire a specific determination of parameters. The parameter$\beta_k$ is computed as a convex combination of $\beta_k^{LS}$(Liu and Storey) and $\beta_k^{CD}$ (Conjugate Descent). Wecompute the parameter $\theta_k$ in such a way that the conjugacycondition is satisfied. The standard Wolfe line search conditionsare used. Numerical comparisons show that the present hybridconjugate gradient algorithm often behaves better than someprevious algorithms.