Different Methods for Quark/Gluon Jet Classification on Real Data from the DELPHI detector

University dissertation from Department of Elementary Particle Physics, Lund University, Professorsgatan 1, Box 118, 221 00 Lund, Sweden

Abstract: Different methods to separate quark jets from gluon jets have been investigated and tested on data from the DELPHI experiment. A test sample of gluon jets was selected from $bar{b}g$ threejet events where the two b-jets had been identified using a lifetime tag and a quark jet sample was obtained from $qar{q}gamma$ events where the photon was required to have a high energy and to be well separated from the two jets. Three types of tests were made. Firstly, the jetenergy, which is the variable most frequently used for quark/gluon jet separation, was compared with methods based on the differences in the fragmentation of quark and gluon jets. It was found that the fragmentation based classification provides significantly better identification than the jet energy only in events where the jets all have approximately the same energy. In Monte Carlo generated symmetric $e^{+}e^{-} ightarrow qar{q}g$ threejet events, where the jet energy does not provide any identification at all, the gluon jet was correctly assigned in 58\% of the events. More importantly, however, is that the identification has been divided into two independent parts, the energy part and the fragmentation part. Secondly, two different sets of fragmentation sensitive variables were tested. It was found that a slightly better identification could be achieved using information from all the particles of the jet rather than using only the leading ones. Thirdly, three types of statistical discrimination methods were compared: a cut on a single fragmentation variable; a cut on the Fisher statistical discriminant calculated from one set of variables; a cut on the output from an {em Artificial Neural Networks} (ANN) trained on different sets of variable. The three types of classifiers gave about the same performance and one conclusion from this study was that the use of ANNs or Fisher statistical discrimination do not seem to improve the results significantly in quark/gluon jet separation on a jet to jet basis.

  This dissertation MIGHT be available in PDF-format. Check this page to see if it is available for download.