DATA-ANALYSIS FOR ROTATIONALLY RESOLVED SPECTRA - A SIMULATED ANNEALING APPROACH

A new method for the analysis of rotationally resolved spectra is presented. The method employs a simulated annealing algorithm with two modifications. First, the standard simulated annealing process is extended to take advantage of parallel computing. Second, rather than using a continuous random search of the parameter space, at distinct intervals new optimization processes are started at points in parameter space that have promising χ2 values. Together these modifications make more efficient use of computer time than a standard simulated annealing approach. The technique is applied to the analysis of simulated data as well as real high-resolution experimental spectra to demonstrate the effectiveness of the parallel simulated annealing algorithm. © 1992.