An efficient algorithm for steepest descent method for unconstrained optimization
Exact line searches along each steepest descent direction converge very slowly. Barzilai and Borwein suggested two stepsizes that ensures superlinear convergence and performs quite well. Barzilai-Borwein method is not monotone, thus it is not easy to be generalized for general nonlinear functions. A...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Published: |
Penerbit UTHM
2009
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/543/ http://eprints.uthm.edu.my/543/1/JST_Vol12_F2.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Exact line searches along each steepest descent direction converge very slowly. Barzilai and Borwein suggested two stepsizes that ensures superlinear convergence and performs quite well. Barzilai-Borwein method is not monotone, thus it is not easy to be generalized for general nonlinear functions. A new stepsize enables fast convergence and possesses monotone property is proposed by Yuan. The new stepsize is modified to obtain modified new steepest descent method, which is for convex quadratic problems only is proposed by Yuan. The new steepest descent method uses the new stepsize after every m exact line search iterations. An algorithm for m=2 is proposed in this paper. |
|---|