We review the basic algorithms using the gradient direction for unconstrained QP problems, discuss their convergence in Euclidean and energy norm, and examine the effect of the steplength on their performance. We pay special attention to the possibility of enhancing the information from the previous step via the steplength (the Barizilai–Borwein method) or momentum (Polyak’s heavy ball method).

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Gradient Methods

  • Zdeněk Dostál

摘要

We review the basic algorithms using the gradient direction for unconstrained QP problems, discuss their convergence in Euclidean and energy norm, and examine the effect of the steplength on their performance. We pay special attention to the possibility of enhancing the information from the previous step via the steplength (the Barizilai–Borwein method) or momentum (Polyak’s heavy ball method).