ePubs

The open archive for STFC research publications

You may experience service outages on ePubs over the coming days due to work being carried out to enhance our network infrastructure. The service should be considered at risk from 23/11 - 03/12.

Full Record Details

Persistent URL http://purl.org/net/epubs/work/51570
Record Status Checked
Record Id 51570
Title On the complexity of steepest descent, Newton’s method and regularized Newton methods for nonconvex unconstrained optimization
Contributors
Abstract It is shown that the steepest descent and Newton's method for unconstrained nonconvex optimization under standard asumptions may require numbers of iterations and function evaluations arbitrarily close to O(e-2) to drive the norm of the gradient below e. This shows that the upper bound of O(e-2) evaluations known for the steepest descent method is tight, and that Newton's method may be as slow as steepest descent in the worst case. the improved evaluation complexity bound of O(e-3/2) evaluations known for cubically-regularised Newton methods is also shown to be tight.
Organisation CSE , CSE-NAG , STFC
Keywords
Funding Information
Related Research Object(s): 49215710
Licence Information:
Language English (EN)
Type Details URI(s) Local file(s) Year
Journal Article SIAM J Optimiz 20, no. 6 (2010): 2833-2852. 2010