Kilonova Color Evolution

• Kilonova color evolution is the time-dependent change in spectral energy distribution caused by radioactive decay and varying opacities in merger ejecta.
• Studies show a rapid transition from blue/UV to red/infrared emissions over hours to days, linking evolution to electron fraction, lanthanide content, and ejecta velocity.
• Multi-band photometry combined with advanced radiative-transfer modeling helps disentangle ejecta mass, geometry, and remnant power to classify diverse kilonova events.

Kilonova color evolution is the time-dependent change in the spectral energy distribution of the radioactive transient produced by compact-object merger ejecta. In the literature summarized here, the defining phenomenology is a rapid migration of emission from ultraviolet or blue optical wavelengths toward red optical, near-infrared, and in some cases mid-infrared bands, on timescales of hours to days, with further evolution over weeks to months in especially well-observed events. This evolution is not a superficial photometric effect: it encodes the ejecta’s electron fraction $Y_e$, lanthanide abundance, velocity structure, mass, geometry, thermalization history, and radiative-transfer regime, including recombination and light-travel-time effects (Tanaka, 2016, Arcavi et al., 2017).

1. Opacity, composition, and the physical origin of reddening

The standard explanation for kilonova color evolution is the interplay between radioactive $r$-process heating and wavelength-dependent line opacity in homologously expanding ejecta. The key compositional distinction is between lanthanide-poor material and lanthanide-rich material. In disk-wind models, a threshold near $Y_e \gtrsim 0.25$ suppresses lanthanide synthesis; above this threshold the outflow is lower-opacity and can produce a blue optical transient, whereas lower-$Y_e$ ejecta remain lanthanide-rich and radiate preferentially at redder wavelengths (Kasen et al., 2014). Atomic-structure calculations further show why this distinction is so strong: open $f$-shell lanthanides such as Nd and Er have much higher bound-bound opacities over a wide wavelength range than open $s$-, $p$-, or $d$-shell species, so post-merger lanthanide-free ejecta are brighter and bluer in the optical, while lanthanide-rich dynamical ejecta are redder and near-infrared dominated (Tanaka et al., 2017).

Diffusion arguments connect opacity directly to timescale and color. For a homologously expanding ejecta of mass $M$, velocity $v$, and opacity $r$0, the characteristic peak time scales as

$r$1

while the peak luminosity decreases as opacity increases (Tanaka, 2016). In the analytical relations used for GW170817/AT2017gfo, the same dependence appears as

$r$2

with corresponding luminosity and temperature scalings showing that low opacity and relatively fast ejecta favor an earlier and bluer peak, although those simple relations did not reproduce the observed early blue temperature of AT2017gfo (Arcavi et al., 2017).

Opacity itself evolves rapidly. Detailed early-time calculations for lanthanide-free ejecta with $r$3–0.40 find a Planck mean opacity of $r$4–$r$5 at $r$6 day, rising to $r$7–$r$8 at $r$9 day as ionization changes and bound-bound transitions become more effective (Banerjee et al., 2020). At later times, recombination can reverse part of that trend. In GRB 230307A, cooling below roughly $Y_e \gtrsim 0.25$0 K was interpreted as lanthanide and actinide recombination from singly ionized to neutral states, reducing the number of free electrons and lowering the infrared bound-bound line density by roughly an order of magnitude; the result was photospheric recession and a cooler, redder, quasi-thermal spectrum (Yang et al., 2023). A related temperature-dependent prescription,

$Y_e \gtrsim 0.25$1

was adopted in dynamical-ejecta transport calculations to represent recombination-driven opacity decline at late times (Just et al., 2021).

2. The early blue phase and its competing interpretations

The clearest early optical color evolution was measured in AT2017gfo. Dense $Y_e \gtrsim 0.25$2 coverage showed an initial blue excess followed by rapid reddening: the transient peaked at about 1 day after merger, then faded at about 2 magnitudes per day in $Y_e \gtrsim 0.25$3, about 1 magnitude per day in $Y_e \gtrsim 0.25$4, and about 0.8 magnitudes per day in $Y_e \gtrsim 0.25$5. Blackbody fits gave a temperature decline from about $Y_e \gtrsim 0.25$6 K at 1.4 days after peak to about $Y_e \gtrsim 0.25$7 K at 2.5 days after peak, while luminous infrared emission was already present by 2.5–3.5 days ($Y_e \gtrsim 0.25$8, $Y_e \gtrsim 0.25$9, $Y_e$0 mag), showing that the spectral energy distribution had shifted strongly toward the near-infrared within only a few days (Arcavi et al., 2017).

The physical interpretation of that blue phase remains non-unique in the early hours. A compilation of the GW170817 ultraviolet, optical, and infrared data found that the optical bands rose on a $Y_e$1 day timescale, which is more naturally reproduced by radioactive decay in low-opacity ejecta, but the earliest ultraviolet data were obtained about 4 hours after the first optical/IR discovery. Including those delayed UV points lowered the inferred first-epoch temperature by about 1500 K, changing the early bolometric slope enough that shock cooling, cocoon or wind-heating models, boosted-radioactive models, and hybrids could not be excluded. The resulting conclusion was deliberately limited: the $Y_e$2 day optical rise mildly favors low-opacity radioactive ejecta, but the evidence is not decisive (Arcavi, 2018).

Detailed opacity calculations indicate that the first hours are a distinct radiative regime rather than a simple extension of later optical behavior. At $Y_e$3 day, representative blue-kilonova conditions are $Y_e$4 and $Y_e$5, with ions reaching at least the tenth ionization stage for $Y_e$6–56. In that regime, bound-bound transitions of heavy elements dominate the opacity, electron scattering is only $Y_e$7, and a fiducial spherical model with $Y_e$8 reaches a bolometric luminosity of $Y_e$9 at $f$0 day. The same model predicts a UVW1 peak of about $f$1 mag at $f$2 hours, or about 20.5 mag at 200 Mpc, implying that early UV data directly probe the fastest outer ejecta and any extra heating source (Banerjee et al., 2020).

A long-lived remnant offers another route to early blue emission. End-to-end HMNS outflow calculations found that longer-lived remnants inject more high-$f$3, lanthanide-poor material, producing brighter and longer-lasting blue emission. In the models explored, a remnant lifetime of $f$4 ms and a viewing angle around $f$5 from the pole approached the blue component of AT2017gfo at early times, whereas shorter-lived cases were dimmer and more susceptible to lanthanide-curtain suppression (Groenewegen et al., 31 Jul 2025).

3. Transition to red and infrared dominance

Although the existence and origin of an early blue phase are event-dependent, rapid reddening is common to essentially all of the cited kilonova calculations and observations. In multi-dimensional kilonova models applied to GRB 160821B, the optical-to-infrared color evolution was extreme: the predicted $f$6 color peaks between 7 and 16 magnitudes, implying that the source can be orders of magnitude brighter in $f$7 than in $f$8 at the epochs relevant for follow-up of short GRBs or gravitational-wave events. The same analysis noted that $f$9 is relatively small, so either $s$0 or $s$1 can be effective from the ground, with $s$2-band potentially favored because of lower sky brightness (Kasliwal et al., 2017).

Late-time observations of GRB 230307A provided a direct empirical realization of this redward drift. The near-infrared spectral peak shifted from roughly $s$3 \AA\ at $s$4 d to $s$5 \AA\ at $s$6 d. The fitted blackbody temperature evolved from $s$7 K at 1.2 d to $s$8 K at 7.4 d, $s$9 K at 28.9 d, and $p$0 K at 61.4 d, while the photospheric radius first increased and later receded. JWST photometry at 28.9 d was strongly weighted toward the reddest bands ($p$1, $p$2, $p$3) and much fainter at shorter wavelengths ($p$4, $p$5, $p$6); by 61.4 d the transient had reddened further, with non-detections at $p$7 and $p$8. The bolometric luminosity declined as

$p$9

from $d$0 at 29 d to $d$1 at 61 d, supporting lanthanide-rich ejecta undergoing recombination rather than a late-time magnetar-dominated phase (Yang et al., 2023).

Dynamical-ejecta-only calculations generally predict that the red or near-infrared phase is not merely late but intrinsic. In simulations with direct merger ejecta trajectories and neutrino treatment, the kilonova typically peaks after $d$2–$d$3 days in the near-infrared, with luminosities between $d$4 and $d$5 and photospheric temperatures of $d$6–$d$7 K. The peak band is usually $d$8, and the near-infrared peak is explicitly incompatible with the early blue component of AT2017gfo, indicating that dynamical ejecta alone are too red and too faint for that event (Just et al., 2021).

A comparable conclusion holds for black hole–neutron star mergers. In those models, the low-opacity neutrino-driven wind is small, while the large projected area of the tidal dynamical ejecta makes the emission increasingly infrared dominated. After $d$9 days post-merger, the peak emission moves from optical to infrared. Peak luminosities are typically a few times $M$0, always $M$1, corresponding to optical absolute magnitudes fainter than $M$2 and infrared magnitudes fainter than $M$3. Relative to AT2017gfo, these kilonovae are cooler at the same epoch and more infrared weighted, making $M$4-band follow-up especially favorable (Zhu et al., 2020).

4. Multi-component ejecta, geometry, and viewing-angle dependence

Color evolution is inseparable from ejecta structure. Disk-wind calculations already established a two-component picture in which outer, higher-$M$5, lanthanide-free layers produce a brief $M$6 day blue optical transient, while inner, lower-$M$7, lanthanide-rich material produces a longer $M$8 day infrared transient. The remnant controls how much of the outflow reaches $M$9: longer-lived neutron stars and rapidly spinning black holes generate more lanthanide-free wind and thus bluer optical emission. This optical component, however, is fragile. Even $v$0 of overlying neutron-rich dynamical ejecta can suppress the blue optical flux by more than an order of magnitude, and $v$1 can render the wind emission effectively invisible. This “lanthanide curtain” is strongly angle-dependent, with pole-on views less obscured than edge-on views (Kasen et al., 2014).

Viewing-angle-dependent kilonova plus afterglow population models sharpened the observational implications of that geometry. Using a GW170817-like anisotropic kilonova with polar lanthanide-free ejecta and equatorial lanthanide-rich ejecta, the predicted reddening is strongest in mixed optical/infrared colors such as $v$2, $v$3, and $v$4. In the same framework, afterglows retain nearly constant color because they are synchrotron sources, so kilonova optical-to-infrared color evolution becomes a discriminant unavailable from single-band decay rates alone (Zhu et al., 2021).

Not all multidimensional radiative-transfer calculations assign the same weight to orientation. A 3D ARTIS model of dynamical ejecta found polar lines of sight up to a factor of $v$5 brighter than equatorial ones, but no significant angle dependence in the color curves themselves; both polar and equatorial directions exhibited a rapid blue-to-red transition, with the dominant driver identified as global cooling rather than orientation (Collins et al., 2022). This contrasts with models in which angle-dependent composition gradients control the visibility of a blue component, and it shows that “viewing angle” is not a universal shorthand for the same radiative effect across model classes.

Jet interaction adds another geometric modifier. In 3D Monte Carlo calculations comparing a no-jet Wind model to jet-driven models Jet49 and Jet51, the blue bands ($v$6) decline monotonically in all cases, but most slowly without a jet and most rapidly in the strongest-jet model. The red and near-infrared bands ($v$7) are more persistent and can show an initial dip followed by rise or plateau. The jet redistributes density and scattering material toward the polar region, so unlike some earlier expectations, the jet models can be fainter near the axis than the no-jet model and become redder faster in the blue bands because short-wavelength photons are more effectively scattered and reprocessed (Shrestha et al., 2023).

5. Benchmark events, survey populations, and diversity

AT2017gfo remains the benchmark for observed chromatic evolution. In comparative short-GRB studies, its $v$8-band was described as showing no apparent peak within the span of the available data and declining from $v$9 days as a simple exponential, while the redder $r$00, $r$01, and $r$02 bands were progressively broader and slower. This “strongly chromatic behaviour” made AT2017gfo a template for evaluating short-GRB kilonova candidates, but it also showed that some events are substantially different: deep upper limits for GRBs 050509B, 061201, and 080905A rule out an AT2017gfo-like kilonova by several magnitudes, and the overall spread across the sample reaches five magnitudes at a few days after trigger. The cited analysis concluded that observer angle alone is probably insufficient to explain that diversity; merger type, ejecta mass, mass ratio, spins, and opacity differences must also contribute (Gompertz et al., 2017).

GRB 160821B is instructive because it demonstrates how color arguments can separate afterglow and kilonova components. Keck/MOSFIRE obtained three $r$03-band epochs at 4.3, 7.5, and 8.4 days after the burst. The first yielded a marginal $r$04 detection at $r$05 AB, while the later epochs gave non-detections of $r$06 AB and $r$07 AB. The first-epoch flux was interpreted primarily as a reddened relativistic afterglow because the inferred $r$08 color of $r$09 mag AB was much bluer than the kilonova models, which are redder in $r$10. The later limits then constrained the infrared kilonova brightness, implying $r$11 for velocities greater than $r$12 (Kasliwal et al., 2017).

Population studies make these event-level lessons operational. Simulations of binary neutron star kilonovae and afterglows found that, at current survey depths $r$13 mag, serendipitous detections are more likely to be afterglows, but for future depths of $r$14–24 mag the kilonova detection rate could become comparable in order of magnitude. The same work stressed that kilonovae are difficult to identify from single-band fading alone; their distinctive signature is rapid optical-to-infrared reddening over the first $r$15–10 days, particularly in $r$16, $r$17, and $r$18 (Zhu et al., 2021). A complementary grey-opacity population synthesis, focused on LSST-relevant wavelengths, predicted peak $r$19-band absolute magnitudes spanning

$r$20

with a bimodal brightness distribution tied to lanthanide-poor versus lanthanide-rich ejecta and a time within one magnitude of peak clustered around $r$21 days. That study explicitly cautioned that further improvements are needed to reproduce detailed spectral shapes across the full light-curve evolution, but it still located the first 1–3 days as the most informative phase for color-based classification (Setzer et al., 2022).

6. Time dependence, remnant power, and unresolved modeling issues

An important refinement is that observed color evolution is not always the same as intrinsic instantaneous color evolution. In a time-dependent reverberation treatment, photons observed at a single epoch are emitted over a range of photospheric times $r$22 and arrive at a detector time

$r$23

with a geometric spread

$r$24

for a spherical photosphere. In rapidly evolving kilonovae this is not negligible. Time-independent calculations can therefore appear too bright and too blue at early times, while scattered photons can persist and drift redward, helping to explain the delayed emission component of the $r$25 feature in AT2017gfo. The central methodological conclusion is explicit: time dependence must be included when the photosphere evolves on timescales comparable to photon flight times through the ejecta (McNeill et al., 10 Oct 2025).

Remnant power sources introduce a second layer of complexity. In radiative-transfer models with an added central engine, pulsar dipole power tends to prolong a blue component, whereas fallback power declines as $r$26 and allows more rapid reddening. In the parameter space explored, the same model could not simultaneously reproduce both the bolometric luminosity and the $r$27 evolution of AT2017gfo, but the $r$28- and $r$29-band trends were better matched by fallback than by pulsar input. The specific inference drawn was that the time dependence of the remnant source influences the color evolution, not merely its integrated energy budget (Wollaeger et al., 2019).

A more explicitly diagnostic proposal treats color evolution itself as a remnant classifier. In the prompt-black-hole case, where the kilonova is powered only by radioactive decay, the predicted behavior is little or no significant color evolution before peak, followed by a rapid post-peak increase in color index. In a long-lived magnetar or stable-neutron-star case, the emission receives contributions from radioactive heating, magnetar spin-down, and pulsar-wind-nebula synchrotron radiation; then the post-peak color evolution becomes more complex, can decrease and then increase, and finally tends toward a constant at late times when the spectrum becomes non-thermal. The same study notes, however, that a late-time constant color is not unique to pulsar-wind-nebula emission, because off-axis afterglow or recombination could mimic similar flattening (Wang et al., 21 Aug 2025). This suggests that color evolution is a powerful but non-exclusive diagnostic, strongest when combined with spectroscopy, cadence, and multi-band coverage.

Taken together, the cited work supports a layered picture rather than a single universal sequence. Some kilonovae begin with a brief blue phase generated by low-opacity ejecta or remnant-supported outflows; others are near-infrared dominated from the outset if only dynamical ejecta are visible. Nearly all credible models then redden rapidly as expansion, cooling, changing ionization, and opacity gradients redirect the emergent flux toward longer wavelengths. The rate, amplitude, and band dependence of that reddening remain sensitive to ejecta composition, geometry, remnant survival time, and radiative-transfer treatment, which is precisely why color evolution has become a central observable in kilonova physics.