Using gradient directions to get global convergence of Newton-type methods
作者：
Highlights：
• Newton and quasi-Newton methods are globalized by using scaled gradient directions.
• Scaling gradient directions by Barzilai-Borwein step lengths is a key issue.
• Search directions invariant to objective function scaling are obtained.
• Improvement over widely used globalization techniques is achieved.
• Extensive numerical experiments show the effectiveness of the proposed approach.
摘要
•Newton and quasi-Newton methods are globalized by using scaled gradient directions.•Scaling gradient directions by Barzilai-Borwein step lengths is a key issue.•Search directions invariant to objective function scaling are obtained.•Improvement over widely used globalization techniques is achieved.•Extensive numerical experiments show the effectiveness of the proposed approach.
论文关键词：Newton-type methods,Globalization strategies,Steepest descent step
论文评审过程：Received 31 March 2020, Revised 22 June 2020, Accepted 9 August 2020, Available online 26 August 2020, Version of Record 11 July 2021.
论文官网地址：https://doi.org/10.1016/j.amc.2020.125612
