Reproduce “Multifield Cosmology with Artificial Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Francisco Villaescusa-Navarro, Daniel Anglés-Alcázar, Shy Genel, David N. Spergel, Yin Li, Benjamin Wandelt, Andrina Nicola, Leander Thiele and 5 more
Villaescusa-Navarro et al. demonstrate that convolutional neural networks can infer cosmological parameters and astrophysical feedback efficiencies from 2D projected maps of multifield hydrodynamic simulations with few-percent precision. A multifield input combining 11 baryonic tracers achieved approximately 2.5% average relative accuracy for Ωm, surpassing individual fields and gravity-only simulations by leveraging complementary information and cross-correlations.
Key insights
- Neural networks can effectively extract cosmological information from complex hydrodynamic simulations, implicitly marginalizing over the confounding effects of baryonic physics.
- Combining information from multiple baryonic fields (multifield analysis) significantly enhances the precision of cosmological parameter inference, even outperforming gravity-only simulations for some parameters.
- Different physical fields exhibit varying sensitivities to cosmological versus astrophysical parameters, highlighting their unique information content and the benefit of a multi-tracer approach.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.