Institut für Organische Chemie, Freie Universität Berlin, D-1000 Berlin 33
Whereas spectrum simulation requires an a priori knowledge of all spectral parameters, automated spectrum fitting allows the determination or at least the refinement of spectral parameters. Besides modules for data and parameter input and output, a fit program has at least three components. First, there must be a module for the calculation of simulated spectra which has to be adequate for the problem under study. Second, a goodness-of-fit criterion is required for measuring the match between experimental and simulated spectra. Third, there must be a procedure for improving the original set of parameters, usually in an iterative way. Finally, the program should provide error estimates for the adjustable parameters.
The programs EPRFT and HFFIT use a non-algorithmic iterative method in the fitting procedure [1]. In this Monte Carlo method, the parameters are varied by small but random amounts, and the new set is retained if a better fit is obtained, following the principle of evolution. This method is particularly suited for microcomputers, because it avoids the storage demands and numerical problems of deterministic algorithms. However, the procedure is fairly slow because typically a few hundred iterations are required, and error estimates of adjustable parameters are not provided. The method will be compared with alternative procedures such as the Marquardt procedure or multidimensional Gauss-Newton method and the simplex method.
Finally, some questions concerning data formats and programming languages will be addressed.
[1] B. Kirste, J. Magn. Reson. 73, 213 (1987).