Dresden 2013 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

HK: Fachverband Physik der Hadronen und Kerne

HK 2: Hadronenstruktur und -spektroskopie

HK 2.6: Talk

Monday, March 4, 2013, 12:30–12:45, HSZ-105

An Evolutionary Algorithm for Model Selection — •Karl Bicker², Suh-Urk Chung¹, Jan Friedrich¹, Boris Grube¹, Florian Haas¹, Bernhard Ketzer¹, Sebastian Neubert¹, Stephan Paul¹, and Dimitry Ryabchikov¹ — ¹Technische Universität München — ²CERN, Geneva, Switzerland

When performing partial-wave analyses of multi-body final states, the choice of the fit model, i.e. the set of waves to be used in the fit, can significantly alter the results of the partial wave fit. Traditionally, the models were chosen based on physical arguments and by observing the changes in log-likelihood of the fits. To reduce possible bias in the model selection process, an evolutionary algorithm was developed based on a Bayesian goodness-of-fit criterion which takes into account the model complexity. Starting from systematically constructed pools of waves which contain significantly more waves than the typical fit model, the algorithm yields a model with an optimal log-likelihood and with a number of partial waves which is appropriate for the number of events in the data. Partial waves with small contributions to the total intensity are penalized and likely to be dropped during the selection process, as are models were excessive correlations between single waves occur. Due to the automated nature of the model selection, a much larger part of the model space can be explored than would be possible in a manual selection. In addition the method allows to assess the dependence of the fit result on the fit model which is an important contribution to the systematic uncertainty. This work is supported by BMBF, MLL München and the DFG Cluster of Excellence Exc153.