Topics in Model Building. Part II. On Nonlinear Least Squares.

Abstract

Gauss suggested that, when the model is a nonlinear function of parameters, least square parameter estimates might be obtained by iterative linearization. To prevent difficulties in convergence, Levenberg, and later Marquardt, proposed a constrained minimization procedure. On critically examining this method with a linearly invariant metric for the parameters, the authors find this to be equivalent to a simple modification of the Gauss method which had been proposed earlier. Procedures to decide how far one should go along the Gauss solution vector are introduced which use only quantities already computed. (Author)

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 1972
Accession Number
AD0770966

Entities

People

  • George E. P. Box
  • H. Kansmasu

Organizations

  • University of Wisconsin–Madison

Tags

Fields of Study

  • Mathematics

Readers

  • Operations Research
  • Regression Analysis.