Monte Carlo simulation was used to study the fit parameter uncertainties obtained from the Marquardt Compromise for a noisy data set, data that deviates from the parent function. A known Gaussian was the parent function in this work. Each data point was also weighted by either statistical or uniform weights. The three uncertainties were the Monte Carlo estimate, equal to the standard deviation of a fit parameter, the error matrix estimate, and the propagation of errors estimate. All simulations produced equality of the error matrix and the propagation of errors estimates for the same fit parameter. Each Monte Carlo estimate was linearly dependent on a parameter that determined the amount of noise and was independent of the weights. However, the error matrix uncertainties depended linearly on the weights and were independent of the noise for statistical weights, but depended linearly on the noise and were independent of the weights for fixed weights. Because the error matrix estimates depend on chi square, it was also studied and varied quadratically with the noise and as the inverse square of the weights. Each of these results was also derived from the theory of the Marquardt Compromise.
Skip Nav Destination
Article navigation
Research Article|
September 01 1993
Comparison of three methods for determining fit parameter uncertainties for the Marquardt Compromise
Robert W. Cunningham
Robert W. Cunningham
Department of Physics, Kent State University, Tuscarawas Campus, New Philadelphia, Ohio 44663
Search for other works by this author on:
Robert W. Cunningham
Department of Physics, Kent State University, Tuscarawas Campus, New Philadelphia, Ohio 44663
Comput. Phys. 7, 570–576 (1993)
Article history
Received:
November 04 1991
Accepted:
April 21 1993
Citation
Robert W. Cunningham; Comparison of three methods for determining fit parameter uncertainties for the Marquardt Compromise. Comput. Phys. 1 September 1993; 7 (5): 570–576. https://doi.org/10.1063/1.168470
Download citation file:
Citing articles via
Related Content
Damping–undamping strategies for the Levenberg–Marquardt nonlinear least-squares method
Comput. Phys. (January 1997)
Least‐squares fits when there are errors in X
Comput. Phys. (March 1993)
IMD—Software for modeling the optical properties of multilayer films
Comput. Phys. (July 1998)