New article in MNRAS 📃

Cold Dark Matter accounts for most of the matter content of the Universe, and is hence a key ingredient in cosmological simulations. Traditional N-Body simulations discretize the CDM fluid into particles (Diracs in the phase space), leading to inacurracies at small scales. My new article, Solving the Cosmological Vlasov-Poisson Equations with Physics-Informed Kolmogorov-Arnold Networks, to be published in MNRAS, provides an alternative path to evolve the CDM fluid sheet using neural networks.

In the image above, the PIKAN output is compared to the N-body simulation in the phase space, for a 1D single halo collapse. The CDM sheet (dark line) folds in a spiral pattern. The PIKAN achieves remarkable results, without discretizing the CDM sheet.

The code behind the paper is publicly available in the cdm-pikan github repo.