Speaker
Description
Dark Matter (DM) is a cornerstone of the standard cosmological model, yet its fundamental nature remains elusive. Accurate numerical simulations are essential to test competing DM models against observational data. In this work, we propose a novel approach to DM simulations by replacing traditional N-body methods with Physics-Informed Kolmogorov-Arnold Networks (PIKANs). Specifically, we apply PIKANs to solve the one-dimensional cosmological Vlasov-Poisson equation for Cold Dark Matter. Our model well captures core collapse, the emergence of singularities at shell-crossing times, and the slope of the density profile. We further show that our method achieves lower residuals compared to N-body simulations. These results demonstrate the potential of physics-informed neural networks as a precise tool for DM simulations. This approach is particularly relevant in the context of upcoming Epoch of Reionization experiments, such as with the Square Kilometre Array, which will require high-resolution simulations to distinguish between competing dark matter scenarios.
