Joint Electromagnetic and Gravitational Wave Inference of Binary Neutron Star Merger GW170817 Using Forward-Modeling Ejecta Predictions

Abstract

We reassess the capacity for multimessenger inference of AT2017gfo/GW170817 using both kilonova and gravitational wave emission within the context of a recent simulation-based surrogate model for kilonova emission. Independent of the inclusion of gravitational wave observations, comparisons between observations that incorporate our kilonova model favor a narrow range of ejecta properties, even when allowing for a wide range of systematic uncertainties in our modeling approach. Conversely, we find that astrophysical conclusions about the neutron star itself, including its mass and radius, depend strongly on assumptions about how much material is ejected from the neutron star. Looking forward, our analysis highlights the importance of systematic uncertainty in general, the need for better modeling of neutron star merger mass ejection from first principles, and warns against uncontextualized applications of ejecta predictions using fits to numerical relativity simulations.

Figures (8)

• Figure 1: Solid, dashed, and dotted line contours corresponding to the KruFo20 dynamical ejecta mass $m_{\rm{ej}}$, disk mass $m_{\rm{disk}}$, and ejecta velocity $v_{\rm{ej}}$, respectively. Contours were calculated as a function of mass asymmetry $\delta = (m_1 -m_2)/(m_1+m_2)$ and neutron star radius $R_{\rm{1.4}}$. Mass and velocity units are in $M_\odot$ and $c$, respectively.
• Figure 2: In this figure, the colored dots represent the input marginal likelihood data assumed in our analysis. In black, red, and blue, we overplot posteriors corresponding to the $90\%$ credible interval for the KruFo20, DiCo20, and Nedora21 GW+EM predictions, respectively. Here, $\mathcal{M}_{\rm{c}}$ is in the detector frame.
• Figure 3: Binary parameter posterior distributions using EM (dashed) and GW+EM (solid) likelihoods using $\log f_{\rm{disk}}$ priors of [-0.9, -0.75], [-1.4, -1.2], and [-1.2, -1] for DiCo20, KruFo20, and Nedora21, respectively. The inclusion of the GW likelihood most noticeably affects the chirp mass $\mathcal{M}_c$ and mass asymmetry $\delta$, while simultaneously providing tighter constraints on the neutron star radius $R_{\rm{1.4}}$ and ejected disk fraction $f_{\rm{disk}}$. Gray shaded regions represent $\delta > 0.3$ for which the secondary star's mass is sub-solar ($< 1 M_\odot$).
• Figure 4: Ejecta posterior distributions using EM (dashed) and GW+EM (solid) likelihoods with the disk mass ejection fraction left as a free parameter. The wind velocity is fixed to $v_w = 0.10c$ per the posteriors in Ref. Peng24. There is little difference in the ejecta posteriors between the two approaches, modulo artifacts from varying sample counts, due to our kilonova model requiring parameters in a very narrow region to recreate the AT2017gfo light curves. Figure \ref{['fig:corner_binary_narrow_mc_fix_vw']} highlights the differences in the binary parameters when including the GW likelihood.
• Figure 5: Light curves predicted by our kilonova surrogate model for the Nedora21 GW+EM ejecta posteriors presented in Figure \ref{['fig:corner_ejecta_narrow_mc_fix_vw']}, with solid lines corresponding to the median posterior values and shaded bands representing 1$\sigma$ uncertainty. This posterior includes the effect of marginalizing over unknown model systematic uncertainty $\sigma_{\rm sys}$, as described in the text.