Existence of optimal solution for exponential model by least squares (Q676157)

The authors consider the exponential model \[ f\big(t;b,c\big)=be^{ct},\qquad b,c\in\mathbb{R},\tag{1} \] parameters of which are to be found from the experimental data \(\big(p_i,t_i,f_i\big), i=1,\dots,m\), \(m\geq2\), where \(t_i\)'s denote the values of independent variable, \(f_i\)'s are the measured values of the corresponding dependent variable and \(p_i\)'s are the data weights. Concerning the errors, the authors assume that they occur only in the measured values of the dependent variable \(f_i\). An existence theorem for the best least squares approximation of the parameters \(b,c\) in model (1) is proved. In addition, the problem of choosing a good initial approximation is considered as well. Finally, results of several examples and numerical experiments, in which some possibilities for the choice of the initial approximation are tested, are given.