A Radiation-Hydrodynamic Light Curve Grid and Interpolation Framework for Stripped-Envelope Supernovae

We present a grid of 8,148 stripped-envelope supernovae (SESNe) light curves based on radiation-hydrodynamic simulations of exploding helium-star progenitors. The grid spans an ejecta-mass range representative of typical SESNe, together with broad ranges of explosion energies, radioactive nickel masses, and degrees of material mixing. We systematically investigate how these physical parameters shape the light curves and develop an interpolation-based parameter inference framework that enables the application of the model grid to observational data. Using the simulated light curves as mock observations, we assess the reliability of the widely used Arnett model. Although it reproduces the overall light-curve morphology, the inferred ejecta masses show no significant correlation with the true values. Moreover, the analytical model produces a spurious correlation between explosion energy and nickel mass, despite these parameters being independent in the underlying model grid. These results demonstrate that good light-curve fits do not necessarily imply reliable physical parameter estimates and highlight the need for physically motivated radiation-hydrodynamic models in population studies of SESNe.

Improvements to AI systems:

1. Physics-constrained surrogate modeling for supernova parameter inference: Build a neural network (e.g., a normalizing flow or mixture density network) trained on the 8,148 simulated light curves to map observed multi-band photometry directly to posterior distributions of ejecta mass, explosion energy, nickel mass, and mixing degree. This replaces the flawed Arnett analytical model with a radiation-hydrodynamic-informed emulator, enabling fast, accurate, and uncertainty-aware parameter estimation for large transient surveys (e.g., LSST, ZTF).

2. Bias-aware model selection and validation framework: Implement an AI system that automatically detects when a simplified analytical model (like Arnett) is being used and flags cases where inferred parameters are unreliable. Specifically, train a classifier or regression model to predict the discrepancy between analytical and simulation-based inferences, then use this to reject or re-weight fits that show spurious correlations (e.g., energy–nickel correlation) before population-level analyses.

3. Generative data augmentation for rare SESNe subtypes: Use the grid to train a conditional generative model (e.g., a diffusion model or GAN) that can produce synthetic light curves for arbitrary combinations of physical parameters not fully covered by the grid. This improves the robustness of downstream AI classifiers (e.g., for typing SESNe or estimating redshift) by expanding the training distribution and reducing overfitting to observed sparse samples.

4. Interpretable feature extraction for physical parameter recovery: Develop an AI system that learns a latent representation of light curves (via an autoencoder or contrastive learning) that is explicitly regularized to correlate with the known physical parameters from the grid. This yields a low-dimensional, physically meaningful embedding that can be used for clustering, anomaly detection, or rapid parameter regression, with built-in uncertainty quantification.

5. Automated model calibration and prior refinement: Use the grid as a training set to calibrate a Bayesian neural network or Gaussian process that predicts the likelihood of a given light curve under the radiation-hydrodynamic model. This system can then be used to iteratively refine priors on explosion parameters for individual events, improving the accuracy of population studies by avoiding the biases introduced by analytical approximations.

What the improved AI system can do:

• Reliable population-level inference: It can estimate ejecta masses, explosion energies, and nickel masses from real SESNe light curves with known accuracy and without the spurious correlations that plague the Arnett model, enabling robust studies of explosion mechanisms and progenitor properties.

• Real-time classification and parameter estimation: It can process incoming transient alerts within seconds, providing posterior distributions for physical parameters and flagging events where the model grid is insufficient (e.g., due to missing physics like circumstellar interaction).

• Self-aware uncertainty reporting: It can explicitly state when a light curve is poorly fit by the grid or when the analytical model would give misleading results, preventing erroneous astrophysical conclusions.

• Simulation-to-observation transfer: It can generate synthetic surveys with realistic noise and cadence to test observational strategies, optimize follow-up scheduling, and quantify selection biases in supernova samples.

We present a grid of 8,148 stripped-envelope supernovae (SESNe) light curves based on radiation hydrodynamic simulations of exploding helium-star progenitors. The grid spans an ejecta-mass range representative of typical SESNe, together with broad ranges of explosion energies, radioactive nickel masses, and degrees of material mixing. We systematically investigate how these physical parameters shape the light curves and develop an interpolation-based parameter inference framework that enables the application of the model grid to observational data. Using the simulated light curves as mock observations, we assess the reliability of the widely used Arnett model. Although it reproduces the overall light-curve morphology, the inferred ejecta masses show no significant correlation with the true values. Moreover, the analytical model produces a spurious correlation between explosion energy and nickel mass, despite these parameters being independent in the underlying model grid. These results demonstrate that good light-curve fits do not necessarily imply reliable physical parameter estimates and highlight the need for physically motivated radiation hydrodynamic models in population studies of SESNe.