Spectroscopic r-Process Abundance Retrieval for Kilonovae III: Linking Spectral and Light Curve Modeling of the GW170817 Kilonova

TL;DR

The paper presents SPARK, a spectral inference framework that derives epoch-specific $r$-process abundances from GW170817 spectra using TARDIS within a Bayesian approach, then translates these abundances into time- and wavelength-dependent heating rates $\dot{q}(t)$ and opacities $\kappa(\lambda,t,v)$. These quantities feed a 1D, gray-opacity kilonova light-curve model to reproduce the observed bolometric and multi-band photometry and to infer the total ejecta mass, yielding $M_{\textrm{ej}}\approx 0.11\,M_\odot$ with uncertainties. The analysis supports a two-component ejecta (blue, high $Y_e$; red, low $Y_e$) and points to a physical origin involving a long-lived hypermassive neutron star remnant launching blue winds alongside magnetized disk winds for the red component. This integrated spectral–time-domain approach links the chemical composition of the ejecta to its dynamical origin and post-merger physics, offering a robust pathway to interpret future kilonova observations.

Abstract

The observed spectra and light curves of the kilonova produced by the GW170817 binary neutron star merger provide complementary insights, but modeling both the spectral- and time-domain has proven challenging. Here, we model the optical/infrared light curves of the GW170817 kilonova, using the properties and physical conditions of the ejecta as inferred from detailed modeling of its spectra. Using our software tool SPARK, we first infer the r-process abundance pattern of the kilonova ejecta from spectra obtained at 1.4, 2.4, 3.4, and 4.4 days post-merger. From these abundances, we compute time-dependent radioactive heating rates and the wavelength-, time-, and velocity-dependent opacities of the ejecta. We use these inferred heating rates and opacities to inform a kilonova light curve model, to reproduce the observed early-time light curves and to infer a total ejecta mass of $M_{\mathrm{ej}} = {0.11}~M_{\odot}$, towards the higher end of that inferred from previous studies. The combination of a large ejecta mass from our light curve modeling and the presence of both red and blue ejecta from our spectral modeling suggests the existence of a highly magnetized hypermassive neutron star remnant that survives for $\sim$$0.01 - 0.5$ s and launches a blue wind, followed by fast, red neutron-rich winds launched from a magnetized accretion disk. By modeling both spectra and light curves together, we demonstrate how combining information from both the spectral and time domains can more robustly determine the physical origins of the ejected material.

Figures (14)

• Figure 1: Schematic diagram illustrating our spectral and light curve inference on observations of the GW170817 kilonova. We input a grid of abundance patterns from nuclear reaction network calculations and lab-measured atomic line lists into radiative transfer simulations to fit the observed optical/IR spectra of the GW170817 kilonova using our tool SPARK. This spectral modeling enables Bayesian inference of the abundance pattern of $r$-process elements synthesized in the ejecta. From these inferred abundance patterns, we compute the heating rates and opacities of the ejecta, which we then input into a light curve model. We fit the bolometric light curve estimated from the observed multi-band light curves of the kilonova to infer a total ejecta mass. This procedure combines both spectral- and time-domain information, ensuring that our spectral and light curve analyses make minimal assumptions on key parameters such as abundances, heating rates, and opacities.
• Figure 2: Compilation of the best-fit synthetic spectra obtained with SPARK, obtained by comparing radiative transfer simulations to the observed spectra of the GW170817 kilonova at 1.4, 2.4, 3.4, and 4.4 days post-merger. The 1.4 and 2.4 day spectra are best-described by a single-component, bluer ejecta. At 3.4 and 4.4 days, an additional red ejecta component emerges, and thus the spectra are best-described with a two-component model. The corresponding best-fit abundance patterns are shown in Figure \ref{['fig:bestfit-abundances']}.
• Figure 3: Abundance patterns of the ejecta at 1.4, 2.4, 3.4, and 4.4 days, inferred from the spectra with SPARK. The abundances at 1.4 and 2.4 days are produced by a single blue component, while the ejecta at 3.4 and 4.4 days is multi-component, containing an additional red component that is richer in heavier elements. The corresponding best-fit spectra are shown in Figure \ref{['fig:bestfit-spectra']}. We also show the Solar $r$-process abundance pattern for comparison, computed using the abundances from lodders09 with the $s$-process residual subtraction of bisterzo14.
• Figure 4: Inferred posterior distributions of the radioactive heating rates of the ejecta at 1.4, 2.4, 3.4, and 4.4 days. As with the abundances (Figure \ref{['fig:bestfit-abundances']}), the heating rates are extracted from the nuclear network calculations of wanajo18 given the inferred $Y_e,~v_{\mathrm{exp}},~\mathrm{and}~s$ at each epoch. At 3.4 and 4.4 days, our best-fit models are multi-component, and each of the two components has its own composition and thus its own inferred heating rate; we report the mass-weighted sum of both components. Appendix \ref{['app:radioactive-components']} includes posteriors for the heating rates of these distinct components.
• Figure 5: Inferred total radioactive heating rates of the ejecta as a function of time. As with the abundances, the heating rates are extracted from the nuclear network calculations of wanajo18, given the inferred $Y_e,~v_{\mathrm{exp}},~\mathrm{and}~s$ at each epoch. The total heating rates at 3.4 and 4.4 days are the mass-weighted total heating rate of the red and blue components at each respective epoch. Fitting the inferred heating rates, we find a power-law of $\dot{q}(t) = \dot{q_0} (t / {\mathrm{1~day}})^\alpha$ with $\dot{q_0} = 10^{10.12 \pm 0.04}~\mathrm{erg~g^{-1}~s^{-1}}$ and $\alpha = -0.99 \pm 0.14$. This fit is shown as solid line, and the gray shaded band reflects uncertainties in $\dot{q_0}$ and $\alpha$. For comparison, we include the heating rate fit of korobkin12 (K+12), which holds primarily for $Y_e \approx 0.05$, and the more broadly-applicable fit from rosswog24 (R&K 24) for two fiducial sets of $(Y_e,~v_{\mathrm{exp}})$.