Improvement in the electron/positron proton separation based on the Electromagnetic Calorimeter of the Alpha Magnetic Spectrometer

BackgroundElectron/positron proton separation based on Electromagnetic Calorimeter (ECAL) is crucial for the search for dark matter through precision measurement of cosmic ray positrons for the Alpha Magnetic Spectrometer (AMS-02) experiment. Proton rejection power with Boosted Decision Trees (BDT) technique in existing AMS-02 software decreases in high energy range (beyond 500 GeV), because there are fewer pure proton samples in data as background for BDT training, and proton Monte Carlo (MC) simulation shows disagreement in ECAL energy distribution compared to the data.
PurposeImprove the proton rejection power based on ECAL in high energy range.
MethodTuning the distribution of the variables of proton MC used in BDT to agree with the proton data. Using proton MC as background training sample with a two-step BDT training approach.
ResultsThe proton rejection power above 1.0 TeV is increased to 1.2×10 ⁴, representing an improvement by a factor of 5 compared to the BDT in existing AMS-02 software.