Improved determination of bimodal size distributions from measurements with diffusional size classification

The size distribution of the unattached fraction of the short-lived radon p rogeny is reported in the literature to have a bimodal structure. Due to th e weak size resolution of diffusional size classification, a wide variety o f bimodal size distributions yields similar measurement results, obstructin g the reconstruction of size distribution parameters from measured data. Fo r example, it could be shown that although 2 of the commonly used nonlinear approximation algorithms performed well for the reconstruction of a monomo dal size distribution, the reproduction of parameters of a bimodal size dis tribution was unsatisfactory. In consequence, a "random walk" approach is p resented. The basic idea for this approach consists of probing the complete parameter space as an ideal method for locating the best set of parameters . Additionally, 2 steps are introduced for the reduction of computation tim e to render the ideal approach to an applicable method. The range of geomet ric standard deviations for the calculation was restricted to values betwee n 1 and 5. The range of median diameters was limited according to the penet ration functions of the diffusional samplers. This restricted volume of par ameter space was subdivided into 10(4) cells. During the calculation, cells with exceptionally large values of the minimization functional were elimin ated from further computation. Compared to the results of EM and Simplex al gorithms, this "random walk" method was able to retrieve parameters of both monomodal and bimodal size distributions with improved accuracy.