The Production of Electron-Capture Elements in Thermonuclear Supernovae: Theory vs. Observations

Type Ia supernovae (SNe Ia) explosively destroy C/O white dwarfs (WDs) in multiple stellar systems. They produce ≈ 50% of the iron-group elements in the Universe, synthesize electron-capture (EC) elements, drive nuclear physics experiments, and underpin high-precision cosmology. To first order, the outcome is governed by nuclear physics, a property often described as “stellar amnesia.” Recently, this stellar amnesia has begun to be broken by the nearly universal detection of EC elements with JWST. These elements trace high-density burning, largely ruling out the currently popular He-triggered, sub-MCh detonation models as the dominant channel. Instead, the ubiquitous presence of EC is shifting back the focus to dynamical and secular mergers, and near-MCh explosions similar to the deflagration model W7, but in which the nuclear flame undergoes a deflagration-to-detonation transition. The early deflagration phase is especially important because spherical simulations identify the central WD density, and thus the WD mass, as a key parameter governing the explosion. Here, we present detailed magneto-hydrodynamical simulations. We find that small-scale, pre-existing turbulence expected from the pre-explosion smoldering phase is essential for overcoming the fundamental challenges imposed by the intrinsic 3D physics. This turbulence systematically reduces the production of EC elements by about a factor of two, implying the need for WD central densities closer to those associated with accretion-induced collapse to a neutron star. We also demonstrate the effect of magnetic fields near the saturation field strength and highlight the need for higher-precision EC rates at low Ye.

Improvements to AI systems based on this paper:

1. Physics-constrained generative models for supernova simulations – Train a neural network to predict 3D flame propagation and deflagration-to-detonation transitions in white dwarfs, using the paper’s finding that small-scale pre-existing turbulence reduces electron-capture (EC) element yields by 2×. The improved AI can generate fast, approximate magneto-hydrodynamical (MHD) simulations that correctly reproduce this turbulence-EC coupling, enabling rapid parameter sweeps over central density and magnetic field strength.

2. Uncertainty-aware nuclear reaction rate emulator – Build an AI surrogate that maps uncertain electron-capture rates (especially at low electron fraction Ye) to final nucleosynthetic abundances (e.g., iron-group and EC elements). The improved system can quantify how rate uncertainties propagate to explosion outcomes, and identify which specific nuclear reactions most strongly influence the predicted WD central density threshold for accretion-induced collapse vs. thermonuclear explosion.

3. Multi-scale turbulence-aware supernova classifier – Develop a deep learning model that ingests JWST observational spectra of SNe Ia (EC element abundances) and outputs the most probable explosion channel (e.g., He-triggered sub-MCh, dynamical merger, near-MCh deflagration-to-detonation). The AI is improved by incorporating the paper’s finding that ubiquitous EC elements rule out He-triggered models, and that turbulence reduces EC yields—so the classifier can infer the central WD density and magnetic field strength from observed EC ratios, even when turbulence is unresolved.

4. Physics-informed reinforcement learning for flame front dynamics – Use reinforcement learning to discover optimal sub-grid models for turbulent flame speed and magnetic field amplification in the early deflagration phase. The improved AI agent can learn closure parameters that reproduce the paper’s simulated factor-of-two reduction in EC production, then generalize to different WD masses and initial turbulence spectra, providing a validated sub-grid model for larger cosmological simulations.

5. Bayesian inverse modeling for WD progenitor properties – Create an AI system that inverts the paper’s simulation results to infer the pre-explosion WD central density (and thus mass) from observed EC element ratios in nearby SNe Ia. The improved system uses a Bayesian neural network trained on the paper’s MHD outputs, explicitly accounting for the turbulence-induced reduction in EC yields, and can output posterior distributions over WD mass and magnetic field strength, directly informing stellar evolution models and the accretion-induced collapse boundary.

What the improved AI system can do specifically:

• Predict EC element abundances (e.g., Ti, Cr, Fe-group) from given initial WD density, turbulence intensity, and magnetic field strength, with accuracy matching the paper’s MHD simulations but at 10,000× lower computational cost.

• Automatically flag which SN Ia observations are inconsistent with He-triggered models, and estimate the required central density to match JWST spectra, including a confidence interval that accounts for nuclear rate uncertainties.

• Provide real-time guidance to nuclear experimentalists by ranking which EC reaction rates (at specific Ye values) most urgently need higher precision to reduce the dominant uncertainty in explosion modeling.

• Enable large-scale population synthesis of SNe Ia by replacing expensive 3D MHD runs with the trained surrogate, allowing for statistical studies of how turbulence and magnetic fields shape the observed diversity of EC element yields.