The Study Of Energy Reconstruction For A Hadronic Calorimeter

CERN announced the discovery of Higgs Boson on LHC in 2012.Shortly after the discovery,the international High Energy Physics community proposed several lepton colliders as Higgs factory,including CEPC by China,FCC-ee by CERN and ILC by Japan.Because the high accuracy of energy resolution is required by the measurements of the Higgs hadronic decay final states,traditional hadronic calorimeter could not achieve the required performance of the future collider.In order to develop new calorimeter techniques,the internatial calorimeter Collaboration CALICE has developed several prototypes of high granularity electromagnetic and hadronic calorimeters based on Particle Flow Algorithm(PFA).Futhermore,some of the calorimeter prototypes have been built to do the beam test and the performance study.This paper focuses on the hadron energy reconstruction performance based on the Semi-Digital Hadronic Calorimeter(SDHCAL)proposed by the CALICE collaboration,with Monte Carlo?~-samples produced by the Geant4 simulation.The energy of a hadron shower is reconstructed by a classical method minimizing?~2 based on number of hits in SDHCAL.The energy linearity of this method is about 3%?4%in the energy ranges from20 GeV to 80 GeV.In order to improve the hadron energy reconstruction,two Machine Learning methods,Multi-Layer Perceptron(MLP)and Boosted Decision Trees with Gradient Boost(BDTG),are employed.In order to fully utilize hadron shower information and further improve the performance,more related variables have been fed into the machine learning methods one by one,like the begin layer of hadronic shower,the number of clusters,number of tracks,the mean radius between hits,etc.The results of the test samples demonstrate significant improvements in energy linearity compared to classical method,from 3%?4% to 1%?2%.The energy resolution with machine learning methods only has marginal improvement since it's dominated by the sampling rate of active layers of SDHCAL.Based on the results above,the two machine learning methods could be considered as energy reconstruction methods of hadronic calorimeters.