The mystery of kilonova late-time infrared glow

When two neutron stars collide, the resulting explosion, known as a kilonova, ejects neutron-rich material at a significant fraction of the speed of light. Some of that material undergoes rapid neutron capture, the r-process, and decays into the heaviest elements on the periodic table. In 2017, the gravitational-wave event GW170817 provided the first direct confirmation of this process through the electromagnetic counterpart AT2017gfo. Since then, a second kilonova candidate, AT2023vfi, associated with the gamma-ray burst GRB 230307A, has deepened our understanding even further.

However, both events revealed an unexpected puzzle. Late-time observations taken weeks after the mergers showed strong infrared emission at surprisingly low temperatures, below 1000 K, far too cool to be explained by atomic processes alone. In AT2023vfi, JWST captured a spectrum peaking at approximately 5 micrometers, consistent with blackbody radiation at roughly 660 K. Standard models of atomic absorption and emission in kilonova ejecta could not reproduce this glow.

Exotic dust from the r-process

A new study led by Nanae Domoto at the University of Tokyo, together with Kenta Hotokezaka and Daniel Kasen, now offers a compelling explanation: the infrared emission comes from dust grains composed of refractory r-process elements, including zirconium (Zr), tungsten (W), and osmium (Os).

The key insight is that, as the ejected material expands and cools, temperatures drop below the condensation thresholds of these refractory species, roughly 1800 K, at about 10 to 20 days after the merger. Under those conditions, atoms of these heavy elements begin to aggregate into tiny solid dust grains through a kinetic process of molecular association.

This result contradicts a previous 2014 study by Takami and collaborators, which concluded that r-process dust could not form efficiently in kilonovae. The difference lies in methodology. The earlier work relied on classical nucleation theory, which considers only monomer-by-monomer growth of clusters. Domoto and collaborators instead solved the full kinetic rate equations, tracking all possible association and fragmentation pathways among clusters of different sizes. When all reaction routes are included, dust formation proceeds far more efficiently, with dimer formation (the pairing of two atoms) acting as the kinetic bottleneck rather than as a total barrier.

Slow ejecta and the velocity threshold

The team’s calculations reveal a strong dependence on ejecta velocity. Dust condenses efficiently in the slower, denser components of the ejecta, those moving at less than about 0.1 times the speed of light. Faster ejecta expand too rapidly and at too low a density for dust grains to assemble before conditions freeze out. This velocity threshold is consistent with the photospheric velocities inferred from the infrared continuum emission in both AT2017gfo and AT2023vfi, providing independent observational support for the dust formation model.

The resulting dust grains, estimated to be nanometer-scale in size, produce a smooth continuum opacity in the infrared that keeps the inner ejecta optically thick for over two months after the merger. In the Rayleigh limit, the infrared opacity depends only on the total dust mass and density, not on the individual grain size, a simplification that makes the model predictions robust across a range of assumed grain size distributions.

A new diagnostic for heavy-element production

Radiative transfer simulations using the Sedona code confirm that the dusty kilonova model naturally reproduces the observed spectrum of AT2023vfi at 29 days. Without dust, the model predicts an optically thin, line-dominated nebular spectrum by day 20, inconsistent with the observations. With dust, the dense inner regions produce a nearly blackbody continuum, while the outer, dust-free layers contribute a smaller emission component near 2.1 micrometers.

The efficiency of dust formation depends sensitively not only on ejecta dynamics but also on the abundance pattern of heavy elements. Ejecta with a composition enriched in third r-process-peak elements, similar to the abundance pattern observed in the metal-poor star HD 222925, can form substantial amounts of dust. In contrast, ejecta with a lighter r-process signature similar to HD 122563 are expected to remain dust-free. This compositional sensitivity opens a new observational window: infrared emission from r-process dust can serve as a direct probe of the nucleosynthetic yields in neutron star mergers, helping to determine what elements are actually produced in these extreme events.

The team notes that similar dust formation processes could also occur in other proposed r-process sites, including collapsars, magnetorotational supernovae, and accretion-induced collapse events, though those environments also produce lighter elements capable of forming carbonaceous or silicate grains, resulting in a more complex dust emission picture.

The paper was posted to arXiv on July 1, 2026, and is authored by Nanae Domoto, Kenta Hotokezaka, and Daniel Kasen.

N. Domoto, K. Hotokezaka, D. Kasen, «Heavy element dust explains the late-time spectra of kilonovae,» arXiv:2607.00433 (2026). DOI: 10.48550/arXiv.2607.00433

© 2026 SKYCR.ORG | Homer Dávila Gutiérrez, FRAS. All rights reserved. Total or partial reproduction without express authorization is prohibited. Original source: phys.org / arXiv:2607.00433.