Development of integro-differential methods of parameter determination for a sum of exponential functions (Q1375678)

The authors consider the model \[ Y(x)=\sum_{i=1}^{M} A_i e^{-l_i x},\;l_i > 0, A_i > 0, l_i\neq l_j, 1\leq i,j\leq M,i\neq j,\tag{1} \] from observations of \(Y\) on a grid \(X=\big\{x_i; a\leq x_i\leq b\big\}\) which is not necessarily uniform. The true values \(Y(x_j)\) are unobservable, and instead the values \[ y_j=Y(x_j)+e_j,\;j=1,\dots,N \] are observed, where the errors \(e_j\) are random and \(E e_j=0\). The quantities \(M, \big\{A_i\big\}, \big\{l_i\big\}\) are assumed unknown and have to be determined from the observations \(\big\{y_j,j=1, \dots, N\big\}\). This problem is ill-conditioned and its numerical solution often involves difficulties. The authors present a modified method of \textit{Foss} [Biometrics 26, 815-821 (1970)], based on the idea that \((1)\) is a representation of a general solution of a homogeneous linear differential equation of \(M\)-th order with constant coefficients. The method proposed in this paper tries to overcome the problems with the numerically ill-posed problems, and make it possible to estimate the errors in the result.