Tuning Monte Carlo generator parameters to measured data by using genetic algorithms
Siegfried Hahn · 2004
Monte Carlo Generators are important tools for analyzing the data measured in High Energy Physics experiments. They describe complete physics events on the basis of various underlying physical models. All these event generators include several free parameters which can not be predicted by theory but have to be determined by comparing simulated events with measured data. Adjusting these parameters is a difficult task due to the complicated nonlinear correlations between different parameters, the multimodal structure of the search space, and the statistical fluctuations of the quality function. Using conventional fitting strategies for this optimization problem requires the knowledge, experience and to some extent the intuition of a human expert in order to reduce the huge amount of calculation time to a reasonable limit. In contrast, genetic algorithms offer an automated procedure which does not require any previous knowledge about the search space. The global character of the search pr...