scientific article; zbMATH DE number 4133416
The authors consider the solution of the unconstrained optimization problem min\(\{\) f(x)\(|\) \(x\in R^ n\}\), f: \(R^ n\to R\), by the trust region method. At a current point \(x^ i\) the step \(s^ i\) is calculated as a solution of the problem (*) \(\min \{F_ i(s)|\) \(\| s\| \leq \delta_ i\}\), and then set \(x^{i+1}=x^ i+s^ i\) and repeat. \(F_ i(s)\) is a local quadratic model: \(F_ i(s)=f(x^ i)+g(x^ i)^ Ts+s^ TB_ is\), where \(B_ i\) is a symmetric, nonsingular matrix. In the paper the general case is presented in which B is an indefinite matrix. The main idea is that, in a ``hard case, the authors compute the eigenvector v corresponding to the least eigenvalue \(\lambda_ 1\) of B, and then get a nearly optimal solution of (*) along v. Hence, only one iteration is needed for solving (*) instead of the large number of iterations in the algorithm of \textit{D. M. Gay} [SIAM J. Sci. Stat. Comput. 2, 186-197 (1981; Zbl 0467.65027)]. Numerical results are presented. Some nonlinear least squares functions of the form: \(f(x)=\sum^{m}_{i=1}f^ 2_ i(x),\) \(x\in R^ n\), are used as test functions.
- scientific article; zbMATH DE number 2069282
- Trust Region Methods
- A trust region algorithm with conjugate gradient technique for optimization problems
- An adaptive method for constrained optimization least-squares problems
- An ODE-based trust region method for unconstrained optimization problems
- A Family of Trust-Region-Based Algorithms for Unconstrained Minimization with Strong Global Convergence Properties
- scientific article; zbMATH DE number 3871040
- A Trust Region Method for Norm Constrained Problems
- scientific article; zbMATH DE number 1932416
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3034748)