Fast prediction of the hydrodynamic QGP evolution in ultra-relativistic heavy-ion collisions using Fourier Neural Operators

Supporting data and models for FNO models in QGP flow (https://doi.org/10.1103/jsh8-6tm7) Three PyTorch models here for QGP flow in 200 GeV Au_Au collisions, using nucleon width of 1.12 fm unless otherwise marked: model_10K_central.pt : trained on 10,000 0-10% central events model_10K_peripheral.pt : trained on 10,000 40-60% central events model_500_of_four_types.py : trained on 500 0-10% central events, 500 40-60% central events, and 500 more central and periphral events with nucleon widths of 0.8 fm. (For 2,000 total events). These three models have been traced for used with JETSCAPE. The traced models are also attached as files: traced_JS3.7_10k_3feat_fno_model_cpu_0_10_59bins.pt traced_JS3.7_10k_3feat_fno_model_cpu_40_60_59bins.pt traced_JS3.7_2k_3feat_fno_model_cpu_4types_59bins.pt To use a PyTorch model with python, use one of the three numbered models above: There are also seven sets of event data. These are formatted and TTrees containing a single branch array floats of length 648,000 per event. To make sense, they can be resized in as: np.reshape(data_array, (n_events, 3, 60, 60, 60)), where the indices are [i_event, i_param, i_x, i_y, i_tau[, and i_param indeces (as shown to the left) 0..3 (for energy density, vx, and vy). The data sets are: cent_0_10_10K_events_flat_xy60_t60.root : 10,000 central events (used to train the model_10K_central.pt) cent_40_60_10K_events_flat_xy60_t60.root : 10,000 peripheral events (used to train the model _10K_peripheral.pt) sample101_40_60_nw9p6_flat_xy60_t60.root : 101 peripheral events with nucleon width = 9.6 fm sample2K_0_10_flat_xy60_t60.root : 2,000 central events sample2K_0_10_spikey_flat_xy60_t60.root : 2,000 central events, nucleon width = 0.8 fm sample2K_40_60_flat_xy60_t60.root : 2,000 peripheral events sample2K_40_60_spikey_flat_xy60_t60.root : 2,000 peripheral events, nucleon width = 0.8 fm