The open archive for STFC research publications

Full Record Details

Persistent URL http://purl.org/net/epubs/work/36041926
Record Status Checked
Record Id 36041926
Title A higher order method for solving nonlinear least-squares problems
Abstract We consider the solution of nonlinear least-squares problems. Such problems have traditionally been solved using a Gauss-Newton or Newton approximation, which is in turn globalized by either solving within a trust region, or by solving a regularized problem. We describe a hybrid method that combines these two approaches. While Newton's method uses second derivative information, we argue that, due to the sum-of-squares nature of the problem, a method we call the tensor-Newton approximation makes better use of this information. We show how the subproblem here can be solved using standard nonlinear least-squares techniques, e.g., our hybrid Gauss-Newton/Newton solver. We describe each of these methods, and give numerical results from our nonlinear least-squares solver RALFit, comparing the proposed solver with the commonly used implementation in the GNU Scientific Library.
Organisation STFC , SCI-COMP
Funding Information
Related Research Object(s):
Licence Information:
Language English (EN)
Type Details URI(s) Local file(s) Year
Preprint RAL Preprints RAL-P-2017-010, SIAM J Sci Comput STFC, 2017. RAL-P-2017-010.pdf 2017