Scaling strategies for symmetric rank-one method in solving unconstrained optimization problems
Symmetric rank-one update (SR1) is known to have good numerical performance among the quasi-Newton methods for solving unconstrained optimization problems as evident from the recent study of Farzin et al. (2011), However, it is well known that the SR1 update may not preserve positive definiteness...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
2014
|
| Subjects: | |
| Online Access: | http://eprints.unisza.edu.my/5079/1/FH02-FIK-14-00733.jpg http://eprints.unisza.edu.my/5079/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Symmetric rank-one update (SR1) is known to have good numerical performance among the quasi-Newton methods for solving unconstrained optimization problems as evident from the recent study of
Farzin et al. (2011), However, it is well known that the SR1 update may
not preserve positive definiteness even when updated from a positive
definite approximation and can be undefined with zero denominator. In
this paper, we propose some scaling strategies to overcome these well
known shortcomings of the SR1 update. Numerical experiment showed
that the proposed strategies are very competitive, encouraging and have
exhibited a clear improvement in the numerical performance over SR1
algorithms with some existing strategies in avoiding zero denominator
and preserving positive-definiteness. |
|---|