Advanced Search

Query:

Fill in the form and click »Search«...

Format:

Display: entries per page entries

Zbl 0957.65061
Dai, Y.H.; Yuan, Y.
A nonlinear conjugate gradient method with a strong global convergence property. (English)
[J] SIAM J. Optim. 10, No.1, 177-182 (1999). ISSN 1052-6234; ISSN 1095-7189/e

The paper presents a new version of the conjugate gradient method $x_{k+1}=x_k +\alpha_k d_k, d_{k+1}=-f''(x_{k+1})+ \beta_k d_k, \beta_k=\|f''(x_{k+1})\|^2/ d_k^T (f''(x_{k+1})-f''(x_k))$ for solving unconstrained optimization problem $\min_{x\in \Bbb R^n} f(x)$. Unlike the classical stepsize rule $\alpha_k =\text{argmin}\{ f(x_k+\alpha d_k): \alpha \geq 0 \}$ the authors analyse a scheme in which $\alpha_k$ is chosen arbitrarily subject to the conditions $f(x_k)-f(x_k+\alpha_k d_k) \geq -\delta \alpha_k f^{\prime T}(x_k) d_k$ and $f^{\prime T}(x_k+\alpha_k d_k) d_k > \sigma f^{\prime T}(x_k) d_k; 0<\delta <\sigma <1$. The convergence $f''(x_k) \to 0$ is established provided that $f(x)$ is bounded below on $N=\{x \in \Bbb R^n: f(x) \leq f(x_1)\}$ and $f''(x)$ is Lipschitz continuous on $N$. The boundedness of $N$ is not assumed.
[Mikhail Yu.Kokurin (Yoshkar-Ola)]

MSC 2000:: *65K05 Mathematical programming (numerical methods)
90C30 Nonlinear programming
90C53 Methods of quasi-Newton type

Keywords: unconstrained optimization; conjugate gradient method; Wolfe conditions; global convergence

Cited in: Zbl 1192.65074 Zbl 1154.90586 Zbl 1181.90221 Zbl 1158.90422 Zbl 1137.90718 Zbl 1149.90425 Zbl 1217.90159 Zbl 1006.65063

Comment on this Item PDF MathML XML ASCII DVI PS BibTeX Online Ordering Article Journal

	Scientific prize winners of the ICM 2010
	Overhang
	Lie groups, physics and geometry. An introduction for physicists, engineers and chemists.

Advanced Search

Zentralblatt MATH Berlin [Germany]