Normalizing Flows for High-Dimensional Detector Simulations

Whenever invertible generative networks are needed for LHC physics, normalizing flows show excellent performance. A challenge is their scaling to high-dimensional phase spaces. We investigate their performance for fast calorimeter shower simulations with increasing phase space dimension. In addition to the standard architecture we also employ a VAE to compress the dimensionality. Our study provides benchmarks for invertible networks applied to the CaloChallenge.