Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Simulation-Based Inference of Supernova Power Sources and their Physical Properties

Edgar P. Vidal, Alexander T. Gagliano, Carolina Cuesta-Lazaro

TL;DR

Problem: inferring multiple concurrent power sources and their physical parameters from irregular, multi-band SN light curves. Approach: a hierarchical SBI framework that learns a shared light-curve summary $x_s=f_\psi(x)$, a source posterior $p(\mathcal{S}|x_s)$, and a parameter posterior $p(\theta|\mathcal{S},x_s)$ via flow-matching posterior estimation with a transformer velocity network conditioned on $(x_s,\mathcal{S})$; training combines $L = L_\theta(\phi) + \lambda L_{\mathrm{source}}(\psi)$ with $\lambda \approx 0.83$. Results: on ~150k MOSFiT light curves, the model achieves about 90% accuracy in identifying energy sources, yields well-calibrated posteriors, and reveals anomaly separation (e.g., TDEs) in the latent space. Impact: enables scalable population studies and rapid discovery of novel transients for Rubin LSST.

Abstract

Time domain surveys such as the Vera C. Rubin Observatory are projected to annually discover millions of astronomical transients. This and complementary programs demand fast, automated methods to constrain the physical properties of the most interesting objects for spectroscopic follow up. Traditional approaches to likelihood-based inference are computationally expensive and ignore the multi-component energy sources powering astrophysical phenomena. In this work, we present a hierarchical simulation-based inference model for multi-band light curves that 1) identifies the energy sources powering an event of interest, 2) infers the physical properties of each subclass, and 3) separates physical anomalies in the learned embedding space. Our architecture consists of a transformer-based light curve summarizer coupled to a flow-matching regression module and a categorical classifier for the physical components. We train and test our model on $\sim$150k synthetic light curves generated with $\texttt{MOSFiT}$. Our network achieves a 90% classification accuracy at identifying energy sources, yields well-calibrated posteriors for all active components, and detects rare anomalies such as tidal disruption events (TDEs) through the learned latent space. This work demonstrates a scalable joint framework for population studies of known transients and the discovery of novel populations in the era of Rubin.

Hierarchical Simulation-Based Inference of Supernova Power Sources and their Physical Properties

TL;DR

Problem: inferring multiple concurrent power sources and their physical parameters from irregular, multi-band SN light curves. Approach: a hierarchical SBI framework that learns a shared light-curve summary , a source posterior , and a parameter posterior via flow-matching posterior estimation with a transformer velocity network conditioned on ; training combines with . Results: on ~150k MOSFiT light curves, the model achieves about 90% accuracy in identifying energy sources, yields well-calibrated posteriors, and reveals anomaly separation (e.g., TDEs) in the latent space. Impact: enables scalable population studies and rapid discovery of novel transients for Rubin LSST.

Abstract

Time domain surveys such as the Vera C. Rubin Observatory are projected to annually discover millions of astronomical transients. This and complementary programs demand fast, automated methods to constrain the physical properties of the most interesting objects for spectroscopic follow up. Traditional approaches to likelihood-based inference are computationally expensive and ignore the multi-component energy sources powering astrophysical phenomena. In this work, we present a hierarchical simulation-based inference model for multi-band light curves that 1) identifies the energy sources powering an event of interest, 2) infers the physical properties of each subclass, and 3) separates physical anomalies in the learned embedding space. Our architecture consists of a transformer-based light curve summarizer coupled to a flow-matching regression module and a categorical classifier for the physical components. We train and test our model on 150k synthetic light curves generated with . Our network achieves a 90% classification accuracy at identifying energy sources, yields well-calibrated posteriors for all active components, and detects rare anomalies such as tidal disruption events (TDEs) through the learned latent space. This work demonstrates a scalable joint framework for population studies of known transients and the discovery of novel populations in the era of Rubin.
Paper Structure (14 sections, 4 figures, 2 tables)

This paper contains 14 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Hierarchical Simulation-Based Inference
  3. The light curve shared summarizer.
  4. Modeling the source posterior.
  5. Modeling the parameter posterior.
  6. Results
  7. Sampling the joint posterior
  8. Detection of physical anomalies
  9. Conclusions
  10. Acknowledgments and Disclosure of Funding
  11. Supernova Multi-Component Priors
  12. Loss Curves
  13. Posterior Coverage Tests
  14. Hyperparameter Tuning

Figures (4)

  • Figure 1: Example joint inference of power sources and their physical properties for a supernova light curve. On the left, we show the posterior samples for the two most likely source combinations powering the light curve, CSM (pink) and CSM+Mag (blue). We show that the presence of the magnetar engine can significantly alter the posteriors for the CSM parameters. In the upper right corner, we show the observed light curve and overlay the light curves drawn from the posteriors of the two power sources.
  • Figure 2: tSNE embedding of the latent features of all simulated SN light curves, colored by their power energy sources. TDE light curves are shown as black stars.
  • Figure 3: Loss curves for the flow network (pink), categorical network (blue), and combined model (orange) on the training (dashed line) and validation set (solid line).
  • Figure 4: The expected coverage probability compared to the credibility level for the primary components (Ni, CSM, and Mag) and all possible combinations. The black dashed diagonal line indicates perfect calibration.