Normalizing Flows for Cherenkov Detector Simulation

One of the important tasks of elementary particle physics is modeling the passage of charged elementary particles through matter. With an increase in the collision energy of beams, the multiplicity of events increases, and hence the number of particles whose motion must be modeled. Classical modeling methods based on the description of the process are not CPU effective enough to have enough events generated. To solve this problem, it is necessary to develop new methods for simulating the response of the detectors to the charged particles. In this paper, it is proposed to use normalizing flows to sample the result of the particle reconstruction algorithm. Experiments were carried out in the work, which showed that this approach allows us to surpass in quality other generating models (primarily generative adversarial networks) at a similar sampling rate. Also, a method for further processing the probability distribution was proposed, which allows us to further improve the quality.