Nuclear Neural Networks

About 

Emulator designed to predict the results of nucleosynthesis calculations, based on Grichener et al. 2025.  It receives as input the temperature (on log scale), density (on log scale) and composition of a burning zone, and predicts the composition, nuclear energy generation (per unit mass) and energy lost by neutrinos (per unit mass per unit time) after a given timestep. 

We currently produced nuclear neural network (NNN) trained models for temperatures and densities corresponding to silicon core burning, but our method could be extended to cover nucleosynthesis regimes for many more physical scenarios.  All the datasets and scripts needed to reproduce our results, or use and/or create NNNs  for your own science can be found here, and detailed documentation on how to create trainings sets for NNNs, how to modify the NNN architecture, how to train NNNs and how to test NNN predictions are available in the guide you are currently reading.

Whether you are trying to reproduce our results or design you own NNNs, don't hesitate to contact me in: agrichener@arizona.edu

Training Sets➡️