Quark and gluon jet discrimination by neutral networks

Abstract

As the energy scales of high energy physics experiments increase, the amount of data which is available becomes difficult to manage. A method that can increase the signal to background ratio would be a clear advantage. The focus of the study reported here is on increasing the light quark jet signal to gluon jet background. We begin by demonstrating that there are characteristics common to quark jets and to gluon jets regardless of the interaction that produced them. The classification technique we use depends on the mass of the jet as well as center-of-mass energy of the hard subprocess that produces the jet. In addition, we present the quark-gluon jet separability results of an artificial neural network trained on three-jet e+ e- events at the Z0 mass, using a backpropagation algorithm. The inputs to the network are the longitudinal momenta of the leading hadrons in the jet. We tested the network with quark and gluon jets from both e+e--+ 3jets and pp-+ 2jets. Finally, we compare the performance of the artificial neural network with the results of making well chosen physical cuts.