Surrogate models for lightcurves and photosphere properties of Type II supernovae

TL;DR

This paper introduces surrogate emulators trained on a large grid of STELLA-type simulations to efficiently infer Type II supernova progenitor and explosion properties from lightcurves and spectra, addressing the computational bottleneck for large-scale LSST analyses. The authors develop independent surrogates for bolometric luminosity, photosphere temperature and radius, and a spectrum, achieving ~30 ms per prediction and enabling robust inference with quantified uncertainties. Validation on simulated data and application to real events (SN 2004et, SN 2012aw, SN 2017gmr) demonstrate accurate recovery of input parameters and competitive progenitor-mass constraints compared with pre-explosion imaging, albeit with some tensions in late-time nebular epochs and hydrodynamical-model comparisons. The approach, implemented in Redback, offers a scalable framework for Type II SN population studies and can be extended to other transients, with planned improvements including non-LTE treatment, varied nickel mixing, and inclusion of spectral features.

Abstract

Inferences on the properties Type II supernovae (SNe) can provide significant insights into the lives and deaths of the astrophysical population of massive stars and potentially provide measurements of luminosity distance, independent of the distance ladder. Here, we introduce surrogate models for the photospheric properties and lightcurves of Type II SNe trained on a large grid of simulations from the radiation hydrodynamics code, {\sc stella}. The trained model can accurately and efficiently ($\sim 30$ms) predict the lightcurves and properties of Type II SNe within a large parameter space of progenitor ($10-18 M_{\odot}$ at ZAMS) and nickel masses ($0.001-0.3M_{\odot}$), progenitor mass-loss rate ($10^{-5}-10^{-1}~M_{\odot}$yr$^{-1}$), CSM radius ($1-10\times10^{14}$cm), and SN explosion energies ($0.5-5 \times 10^{51}$erg). We validate this model through inference on lightcurves and photosphere properties drawn directly from the original {\sc stella} simulations not included in training. In particular, for a synthetic Type II SNe observed within the 10-year LSST survey, we find we can measure the progenitor and nickel masses with $\approx 9\%$ and $\approx 25\%$ precision, respectively, when fitting the photometric data while accounting for the uncertainty in the surrogate model itself. Meanwhile, from real observations of SN~2004et, SN~2012aw, and SN~2017gmr we infer a progenitor ZAMS mass of $12.15_{-1.06}^{+1.03} M_{\odot}$, $10.61_{-0.32}^{+0.37} M_{\odot}$, $10.4 \pm 0.3 M_{\odot}$, respectively. We discuss systematic uncertainties from our surrogate modelling approach and likelihood approaches to account for these uncertainties. We further discuss future extensions to the model to enable stronger constraints on properties of Type II SNe and their progenitors, and applications of our surrogate modelling approach to other transients.

Figures (10)

• Figure 1: Bolometric luminosity and photosphere temperature and radius for random samples from our training data (solid curves) alongside the surrogate model predictions (dashed curves), while the bottom panels show the normalised residuals on each quantity.
• Figure 2: $1\sigma$ credible interval of the normalised residuals of our surrogate model predictions compared to the testing data.
• Figure 3: $90\%$ credible interval (shown by blue shaded region) of the bolometric luminosity, photosphere temperature and radius generated from $50000$ random samples our prior. We also show the predicted properties for a Type II SN with $M_{\rm ZAMS}=13~M_{\odot}$, ${}^{56}\rm{Ni}=0.02~M_{\odot}$, $\dot{M}=10^{-3.1}$$M_{\odot}$yr$^{-1}$, $\beta=1.2$, $R_{\rm csm} = 5.5\times10^{14}$cm, and $E_{\rm sn}=2.1\times10^{51}$erg (red curves), a sample
• Figure 4: The top panel shows the true absolute magnitudes (solid curves) for a range of ZTF filters with our surrogate model prediction shown with dashed curves and the prediction assuming a blackbody SED (in squares). The bottom panel shows a $90\%$ credible interval (shown by blue shaded region) absolute magnitudes against time (in observer frame) for the lsst/u (blue) and lsst/y (green) bands drawn from $5000$ random samples from our prior.
• Figure 5: Example of a bolometric luminosity (top panel) and spectrum predictions (bottom panel) from our model alongside an estimate of the uncertainty estimated through Monte Carlo sampling ($3\sigma$ uncertainty shown by shaded bands).