Research

Phys. Rev. Research · 2026 · 6 citations

Joint electromagnetic and gravitational wave inference of binary neutron star merger GW170817 using forward-modeling ejecta predictions

Can the kilonova from GW170817 tell us how big neutron stars are? We combined the gravitational-wave signal with our simulation-trained kilonova surrogate and three published fits to numerical relativity simulations that predict how much mass a merger ejects, and how fast. Even with a generous allowance for systematic uncertainty, the kilonova pins down a narrow range of ejecta properties, but each fit implies a different binary: the inferred radius of a 1.4 solar-mass neutron star spans roughly 8–15, 10–20, or 10–40 km depending on the fit. The data also favor ejecting only about 4–16% of the post-merger disk, well below the up to 40% often assumed. Using kilonovae to constrain the dense-matter equation of state therefore needs better first-principles models of how mergers eject mass.

Read paper arXiv

ApJL · 2026 · 5 citations

Kilonovae and Long-duration Gamma-Ray Bursts

Kilonova-like glows following the long gamma-ray bursts GRB 211211A and GRB 230307A have been read as evidence that neutron star mergers can power long bursts, with their red, infrared emission taken as the signature of freshly made lanthanides. We tested an alternative: a collapsar, the core collapse of a massive, rapidly rotating star, producing only weak r-process elements (those up to a mass number of about 130). In our radiative transfer models, a single lanthanide-free ejecta component reproduces both the early blue and the late red light of each event, which merger kilonovae need two components to do, and the large ejecta masses required are hard for a merger to supply. A red kilonova therefore does not by itself prove heavy-element production; gravitational-wave detections of future events will help tell the scenarios apart.

Read paper arXiv

In the news LANL NewsPhys.org

Phys. Rev. Research · 2023 · 8 citations

Interpolated kilonova spectra models: Examining the effects of a phenomenological, blue component in the fitting of AT2017gfo spectra

Spectra hold more information about a kilonova's ejecta than broadband light curves, but radiative transfer spectra are far too costly to compute on demand. We showed that random-forest interpolation trained on our existing simulation library produces accurate spectra across ejecta masses and velocities, viewing angles, and times after merger. Fitting the spectra of AT2017gfo recovers ejecta parameters similar to our earlier light-curve analysis, but the models consistently fall short of the observed blue light. A commonly invoked fix, a third light, slow, lanthanide-free ejecta component, makes the deficit worse once it is included self-consistently in the radiative transfer, because it reprocesses blue photons to redder wavelengths; the shortfall is clear by a week after merger. Explaining the blue emission will need more sophisticated physics rather than another component.

Read paper arXiv

ApJ · 2023 · 8 citations

Constraining Inputs to Realistic Kilonova Simulations through Comparison to Observed r-process Abundances

If neutron star mergers are the main source of the universe's heaviest elements, the material a kilonova ejects should reproduce the r-process abundance pattern seen in the Sun. We turned that requirement into a prior: comparing the compositions assumed in our kilonova simulations with solar abundances constrains the ratio of wind to dynamical ejecta, which we combine with fits to the light curves of AT2017gfo. The best joint match comes from very neutron-rich dynamical ejecta (electron fraction 0.035, with FRDM2012 nuclear masses and FRLDM fission) plus moderately neutron-rich winds, giving a wind-to-dynamical mass ratio of about 0.47, roughly double our earlier estimate. That ratio shifts by a factor of about four with the assumed nuclear physics, a reminder of how much these inferences depend on nuclear inputs.

Read paper arXiv

Phys. Rev. Research · 2022 · 22 citations

Interpolating detailed simulations of kilonovae: Adaptive learning and parameter inference applications

Radiative transfer simulations capture the detailed physics of kilonovae, but each one is too expensive to run inside a parameter-inference loop, while the fast semi-analytic models used instead leave out that physics and can bias the results. This paper bridges the two. Starting from a grid of two-dimensional, anisotropic simulations, we used adaptive learning to pick which new simulations to run, then trained Gaussian-process and random-forest surrogates that predict multiband light curves continuously across the masses and velocities of two ejecta components. Fitting AT2017gfo, the kilonova from GW170817, with these surrogates recovers different ejecta properties than simplified analytic models do, showing how strongly modeling choices shape what we infer. The light curves, interpolation code, and inference engine are publicly released.

Read paper arXiv