A Time-Dependent Parametric Matrix Model for Nuclear Reaction Networks

Nuclear reaction networks are high-dimensional and stiff differential equations that describe the evolution of nuclear species over time. Parametric Matrix Models (PMMs) are a new class of machine learning techniques useful for model emulation and dimensionality reduction. We present a PMM emulator which provides a low-dimensional and accurate representation of the reaction dynamics incorporating one-, two-, and three-body reactions. The model rigorously respects physical constraints and utilizes a high-order adaptive solver for stable time integration.