The effect of nonrandom errors on the results from regularized inversions of dynamic light scattering data

Dynamic light scattering data from measurements of polystyrene latex beads and lipoprotein particles were analyzed with the regularization algorithms CONTIN and ORT. In addition to the methods for the selection of appropriate ly regularized solutions of these programs, we applied the method of L-curv es. We have studied the effect of systematic errors in the data due to an e rror in the experimental baseline and of partly correlated errors of the in tensity fluctuation noise on the results obtained with these selection meth ods. The solutions determined with the F-test were most sensitive to the pr esence of nonrandom errors. Then, the P-test yielded too weakly regularized solutions associated with complex and highly variable size distributions. In this situation, the other two methods (the stability plot and the L-curv e method) provided too strongly regularized solutions but with less variabl e and more reliable size distributions. When data contained only randomly d istributed errors, all three selection methods yielded practically the same result. This study confirmed the importance of the accuracy of the baselin e. Normalization with the correct value improved the reliability of the siz e distribution and the quality of the fit by up to 100-fold. Baseline error s were determined with the baseline variation option of the program ORT, wh ere the optimal value is found from the minimum of the mean deviation curve with highest sensitivity for variations of the baseline, and a similar var iation method used with CONTIN. The sensitivity parameter, with which the o ptimal regularization strength is determined, proved to be useful when data contained only random noise but failed in the presence of nonrandom statis tical errors. The new method used with CONTIN, where the optimal regulariza tion strength and the solution associated with the optimal baseline were de termined with L-curves, was successful in all cases.