Fig. 1: Validation results for our machine-learning-based emulators.
From: Inferring three-nucleon couplings from multi-messenger neutron-star observations
In both panels, the color of the curves corresponds to the percentage uncertainty in the emulators' prediction. a Shows 70 results for the neutron-matter EOS up to 2nsat predicted by a PMM built using a training set of 30 different samples. At nuclear saturation density, individual validation uncertainties Δsat are below 0.3%, with the average uncertainty being 0.04%. b Shows ~60,000 tidal deformability-mass curves predicted by an ensemble of 50 neural networks trained on the 5-parameter EOS model. The uncertainty at 1.4M⊙, Δ1.4, is at most 2.5% with an average of 0.02%. We note that less than 0.1% of EOSs have emulator uncertainties of around 2%, making these very rare occurrences. Source data for this figure are provided as a Source Data file.
