A Appendix In the following sections, we provide additional details about the network architecture, training, and

We decided to use a fixed axis, which in our case corresponds to the spatial y-axis. Additional Modifications In addition to the properties of our algorithm as discussed in Section 2.3 The output is denormalized again before continuing with the time integration steps. Moreover, to satisfy the condition of the constraint 10 from above, i.e. GPU (48GB) for the 3D data set. Preprocessing As discussed in Section 2.4, we use custom preprocessing steps to improve longtime stability.