Reconstruction of the event vertex in the PandaX-III experiment with convolution neural network

bstract The PandaX-III experiment uses a high-pressure xenon gaseous time projection chamber (TPC) to search for the neutrinoless double beta decay (0 νββ ) of 136 Xe. The absence of the vertex position in the electron drift direction at which the event takes place in the detector limits the PandaX-III TPC’s performance. The charged particle tracks recorded by the TPC provide a possibility for vertex reconstruction. In this paper, a convolution neural network (CNN) model VGGZ0net is proposed for the reconstruction of vertex position. An 11 cm precision is achieved with the Monte Carlo simulation events uniformly distributed along a maximum drift distance of 120 cm. The electron loss during the drift under the different gas conditions is studied, and after the distance-based correction, the detector energy resolution is significantly improved. The CNN model is also verified successfully using the experimental data of the PandaX-III prototype detector.