Measurement of Gamma(Z going to bottom quark antibottom quark)/Gamma(Z going to hadrons) partial decay width using multiple independent tags.

Abstract

The decay of Z&rarr;bb offers many insights into the Standard Model, and provides a window to test for new physics if the ratio of partial decay widths Rb=G Z&rarr;bb</fen><de>G Z&rarr;hadrons</fen></de> can be measured with sufficient precision. This measurement is based on data collected in 1993 by the L3 detector at the LEP collider for an integrated luminosity of 32.8 pb<super>-1</super> of e<super>+</super>e<super>-</super> with a centre of mass energy s&sime;mz. The data sample consists of 157085 selected hadronic events. The branching ratio is measured using a multiple tagging scheme that includes information from a neural network and from semi-leptonic decays. This analysis involved a study of the effectiveness of neural networks for identifying certain high energy physics processes. The neural network that was chosen for this task consisted of three layers, with 6 input nodes, 2 hidden nodes, and one output node trained on 150,000 <italic>Z</italic> normalised Monte Carlo events using a Langevin updating procedure with a variable learning rate. Independent validation and testing samples consisted of 50,000 and 150,000 events respectively. Various other architectures and training mechanisms were evaluated, and it was found that any increase in complexity over this network produced only marginal improvements in performance. The network was found to be considerably more effective at tagging hadronic <italic> b</italic> decays than equivalent linear discriminant analyses, and contributed significantly to the precision of the measurement. However, inaccuracies in the Monte Carlo caused a degradation in performance as optimal inputs could not be used, and resulted in increased systematic errors. The measured ratio of partial decay width was found to be: <display-math> <fd> Rb=0.214+/-.002 stat.</fen>+/-.010 syst.</fen> </fd> </display-math> This measurement is in agreement with the Standard Model predicted value, as well as recent results from other experiments.