MACRIB High efficiency - high purity hadron identification for DELPHI

Abstract

Analysis of the data shows that hadron tags of the two standard DELPHI particle identification packages RIBMEAN and HADSIGN are weakly correlated. This led to the idea of constructing a neural network for both kaon and proton identification using as input the existing tags from RIBMEAN and HADSIGN, as well as preproccessed TPC and RICH detector measurements together with additional dE/dx information from the DELPHI vertex detector. It will be shown in this note that the net output is much more efficient at the same purity than the HADSIGN or RIBMEAN tags alone. We present an easy-to-use routine performing the necessary calculations.