Intercomparison of Approximation Algorithms for the Determination of the Size Distribution of the "Unattached" Fraction of Radon Progeny

Starting in 1996, nonlinear approximation algorithms used for the determination of size distributions of ''unattached'' radon progeny from measurements made with diffusional size classification instruments were compared. Seven participants (one American and six European) took part using various techniques (Simplex, Expectation Maximization, Extreme Value Estimation, Twomey, and Random Walk) which are widely used for this type of work. Simulated input data was supplied to the participants. Their output data indicated that the quality of the results varied according to the algorithm used. This was particularly the case when retrieving parameters from bimodal distributions where large inaccuracies were shown. Variations were found between participants using basically identical algorithms. The influence of measurement uncertainties was investigated and indicated a serious reduction of retrieval accuracy, especially for algorithms with better than average performance for precise data. This work suggests that when high precision results are required, a random walk algorithm should be used and attention should be paid to reducing errors in the input penetration data.