Simulated Kilonova Spectra Interpolation with Comparison to AT2017gfo Observations

Neutron star mergers could be the primary channel for heavy-element nucleosynthesis in the Universe. Kilonovae provide, via observations of their spectra, a direct avenue to assess how much this channel contributes to r-process abundances. In this work, we present a simple kilonova spectral interpolation technique trained against our existing library of kilonova radiative transfer simulations which use state-of-the-art, physically-motivated atomic opacities and lines. Our model accurately reflects spectra for anisotropic (axisymmetric) time-dependent kilonova models over our presently four-dimensional kilonova model space. We compare our interpolated spectra to the AT2017gfo spectral data and find parameters consistent with our previous inferences deduced from long-term multiwavelength light curve observations. However, the spectral observations have significant systematic short-wavelength residuals relative to our models, which we cannot explain within our existing framework. Similar to previous studies, we argue an additional blue component is required. We find that a light, slow-moving lanthanide-free component could supplement early-time short-wavelength model deficits.