CEENs: causality-enforced evolutional networks for solving time-dependent partial differential equations (Q6557798)

We propose CEENs: Causality-enforced evolutional networks, which integrate temporal causality into physics-informed neural networks (PINNs) by dividing the time domain into nonoverlapping subintervals and assigning unique neural networks to each, sequentially training them while enforcing causality through a novel loss function founded on the integral form of PDEs within these intervals.