Abstract

Based on the concept of Big Data Modeling, the errors in determining peak parameters of noisy Gaussian triplets using nonlinear least squares curve fitting have been evaluated. Calculations were performed using Gauss-Newton (with Levenberg-Marquardt modifications) and genetic algorithms The probability that the relative error in estimating each model parameter is not greater than a priory given limit for a given fitting error has been calculated. It was shown that using the derivative mode for curve fitting has no advantage over the normal mode. It was demonstrated that the mean errors for the genetic algorithm are significantly greater than for the Gauss-Newton algorithm. It was found that the mean probability for triplets is higher than that for quartets. Obtained results showed that small fitting error does not guarantee that a fitting algorithm does converge to the correct peak parameters.

Authors and Affiliations