Deep learning for flow observables in ultrarelativistic heavy-ion collisions

We train a deep convolutional neural network to predict hydrodynamic results for flow coefficients, average transverse momenta and charged particle multiplicities in ultrarelativistic heavy-ion collisions from the initial energy density profiles. We show that the neural network can be trained accurately enough so that it can reliably predict the hydrodynamic results for the flow coefficients and, remarkably, also their correlations like normalized symmetric cumulants, mixed harmonic cumulants and flow-transverse-momentum correlations. At the same time the required computational time decreases by several orders of magnitude. To demonstrate the advantage of the significantly reduced computation time, we generate 10M initial energy density profiles from which we predict the flow observables using the neural network, which is trained using 5k, and validated using 90k events per collision energy. We then show that increasing the number of collision events from 90k to 10M can have significant effects on certain statistics-expensive flow correlations, which should be taken into account when using these correlators as constraints in the determination of the QCD matter properties.