🤖 AI Summary
This work addresses the limitation of Lagrangian methods in large-scale cosmic structure evolution forecasting—namely, their reliance on particle-level information, which is often unavailable. We propose an Eulerian field-based modeling framework to circumvent this constraint. Specifically, we introduce a Generative Adversarial Network-enhanced Autoencoder (GAN-AE) for joint spatiotemporal prediction of dark matter density and velocity fields, trained on 2D/3D N-body simulation data. A key innovation is the explicit incorporation of the velocity field as a physical constraint, which substantially improves robustness and cross-step generalization in 3D prediction: while 2D forecasting achieves high accuracy using density fields alone, 3D prediction with velocity-field integration reduces forecast error significantly and demonstrates markedly superior long-term stability compared to density-only models. This approach establishes a novel paradigm for cosmological fluid modeling in particle-agnostic scenarios.
📝 Abstract
Cosmological simulations play a key role in the prediction and understanding of large scale structure formation from initial conditions. We make use of GAN-based Autoencoders (AEs) in an attempt to predict structure evolution within simulations. The AEs are trained on images and cubes issued from respectively 2D and 3D N-body simulations describing the evolution of the dark matter (DM) field. We find that while the AEs can predict structure evolution for 2D simulations of DM fields well, using only the density fields as input, they perform significantly more poorly in similar conditions for 3D simulations. However, additionally providing velocity fields as inputs greatly improves results, with similar predictions regardless of time-difference between input and target.