ITS Performance Study And Application Of Deep Learning At NICA/MPD Experiment

The Multi-Purpose Detector（MPD）is one of the heavy-ion collision experiments of the Nuclotron-based Ion Collision fAcility（NICA）.Assembly of the MPD began in 2020 and it is scheduled to be operational in 2023.Its main physical purpose is to search for novel phenomena in the baryon-rich region of the QCD phase diagram and to find the critical point of QCD phase transition by means of colliding heavy nuclei in the energy range of (S_NN)^1/2= 4-11GeV.In the future,the Inner Tracking System（ITS）,manufactured based on Monolithic Active Pixel Sensors（MAPS）technology,will be assembled into the MPD as vertex detectors.The ITS consists of five layers of MAPS,which are placed between the beam tube and the Time Projection Chamber（TPC）.With high position resolution,it can more accurately measure the momentum and the topological variables of short-lived particles,such as DCA and decay length.Thus,the background of short-lived particle reconstruction can be effectively depressed and the reconstruction ability of short-lived particles can be improved.The first work of this paper is to study the physical performance of the ITS.We use Ur QMD to simulate Au+Au collisions at (S_NN)^1/2= 11GeV before and after installing the ITS.The performance of the ITS in terms of transverse momentum resolution and neutral strange hadron reconstruction is studied.The results show that the physical performance of the ITS,the TPC and the TOF are consistent with the MPD design.Installation of the ITS can improve track reconstruction capability,transverse momentum resolution and neutral strange hadron reconstruction capability of the MPD.Therefore,it is of great significance to install the ITS in the MPD.The second work of this paper is to apply deep learning to charged particle identification and neutral strange hadron reconstruction.Simulations of the MPD experiments can provide a large amount of labeled data,such as the momentum and the species of particles,which will be used in supervised learning.According to different needs,we have adopted different processing methods for training data,and obtained different deep learning models with different performance.The results show that deep learning is superior to traditional method in particle identification and neutral strange hadron reconstruction.Therefore,the application of deep learning in analysis of experimental data can optimize the performance of the MPD in particle identification and neutral strange hadron reconstruction.