Kilonovae and Neutron Star Mergers

When two neutron stars spiral together and merge, tidal forces and the sheer violence of contact fling out a small fraction of their mass — a few thousandths to a tenth of a solar mass — at a tenth of light speed. This ejecta is extraordinarily neutron-rich, with electron fraction $Y_e$ (protons per baryon) often below $0.25$ . As it expands and cools out of nuclear statistical equilibrium, free neutrons are captured onto seed nuclei far faster than those nuclei can beta-decay: the rapid neutron-capture process, or $r$ -process, building elements all the way to the actinides. Mergers are now known to be a — perhaps the — major forge of roughly half the elements heavier than iron, gold and platinum and the lanthanides among them. The freshly minted nuclei are radioactive and decay back toward stability through beta decays, alpha decays, and fission, releasing heat. Because an enormous ensemble of species with a broad spread of half-lives decays together, the heating rate is a nearly universal power law, $\dot q\propto t^{-\alpha}$ with $\alpha\approx1.3$ , almost independent of the exact abundances. That heat, thermalized in the expanding cloud, powers a supernova-like but far fainter transient — the kilonova, named because its peak luminosity is roughly a thousand times a classical nova ( $\sim10^3\,L_\mathrm{nova}$ ), though only $\sim10^{-3}$ of a supernova. Everything observable follows from a race between photon diffusion and expansion — the Arnett picture from supernovae — with one decisive twist: opacity. Lanthanides and actinides, with their open $f$ -shells and millions of bound–bound transitions, raise the effective opacity a hundredfold above iron-group matter, splitting the ejecta into a fast, lanthanide-poor "blue" kilonova peaking within a day and a slow, lanthanide-rich "red" kilonova peaking after a week. GW170817 confirmed the whole picture in 2017: a gravitational-wave chirp, a short gamma-ray burst, and the kilonova AT2017gfo, all from one binary neutron-star merger at $\sim40\,\mathrm{Mpc}$ .

Key takeaways

A kilonova is a thermal, radioactively powered transient from neutron-star (or neutron-star–black-hole) mergers that eject $\sim10^{-3}$ – $10^{-1}\,M_\odot$ of neutron-rich matter undergoing $r$ -process nucleosynthesis.
Decay of unstable $r$ -process nuclei heats the ejecta at $\dot q\propto t^{-1.3}$ — a nearly universal law set by the statistics of many isotopes — powering $L\sim10^{40}$ – $10^{42}\,\mathrm{erg\,s^{-1}}$ .
Opacity is the master switch: lanthanide-rich ejecta have $\kappa\sim10\,\mathrm{cm^2\,g^{-1}}$ (red, slow, near-IR); lanthanide-poor ejecta have $\kappa\sim0.1$ – $1$ (blue, fast, optical/UV).
Peak time and luminosity follow from the Arnett balance: $t_\mathrm{peak}\sim\sqrt{\kappa M_\mathrm{ej}/(v_\mathrm{ej}c)}$ (day-to-week), and at maximum $L_\mathrm{peak}\approx f_\mathrm{th}\,\dot q(t_\mathrm{peak})\,M_\mathrm{ej}$ .
GW170817 (17 August 2017) and its kilonova AT2017gfo, found $\sim11$ hours later at $\sim40\,\mathrm{Mpc}$ , proved mergers are a major $r$ -process site and required $\sim0.05\,M_\odot$ of ejecta.
Kilonovae are electromagnetic counterparts to gravitational-wave sources, probes of the dense-matter equation of state, and standard sirens for an independent Hubble constant.
Key idea: A kilonova is an Arnett-type radioactively powered transient in which the power source is $r$ -process decay heat ( $\dot q\propto t^{-1.3}$ ) rather than $^{56}$ Ni, and the light-curve colour and timescale are set by whether lanthanides are present (high opacity, red, slow) or absent (low opacity, blue, fast).

Where this lives today

Kilonovae now sit at the centre of multi-messenger astronomy. When LIGO, Virgo, and KAGRA report a neutron-star merger, robotic optical and infrared telescopes tile the sky-localization region within hours, hunting the fading thermal glow — the strategy that found AT2017gfo in NGC 4993 eleven hours after GW170817. The physics has sharpened since: spectra of AT2017gfo showed a broad feature near $800\,\mathrm{nm}$ that Watson and collaborators (2019) identified as singly ionized strontium — the first direct spectroscopic detection of a freshly synthesized $r$ -process element. Looking ahead, the Vera Rubin Observatory's LSST and dedicated infrared facilities aim to catch dozens of kilonovae per year, including events with no gamma-ray burst, while next-generation detectors (Einstein Telescope, Cosmic Explorer) will localize mergers well enough to make kilonovae routine standard sirens. That may be the most consequential use: the gravitational-wave inspiral yields an absolute distance while the host gives the redshift, so each merger is a calibration-free ruler for $H_0$ . GW170817 alone gave $H_0\approx70\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$ ; a few dozen more could arbitrate the Hubble tension from a direction the distance ladder cannot reach.

The physical picture

Picture the merger flinging out a warm ( $\sim10^9$ – $10^{10}\,\mathrm{K}$ ), extremely neutron-rich cloud expanding at a tenth of light speed. As it thins, neutrons rain onto seed nuclei, building heavy radioactive isotopes that then decay like a fading nuclear campfire whose heat falls off as a clean power law $t^{-1.3}$ ; the light we see is that campfire glowing through an expanding, cooling fog. But the ejecta is not one cloud — it is at least two, launched by different mechanisms in different directions, and the first figure shows why that matters.

Figure

Two ejecta channels from one merger: neutron-rich tidal ejecta in the equatorial plane with Y_e ∼ 0.1 and κ ≈ 10 cm²g⁻¹ makes the slow red kilonova, while shock- and neutrino-processed polar ejecta with Y_e ≳ 0.3 and κ ∼ 0.5 cm²g⁻¹ makes the fast blue kilonova — both timed by t_peak ∝ √(κ M_ej / v c)

The single most important knob is opacity. If the ejecta carries lanthanides, its fog is optically thick — photons random-walk out slowly, so the transient is dim, slow, and red. If neutrinos from a hot, long-lived remnant have raised $Y_e$ above $\sim0.25$ , lanthanide production is choked off, the fog thins to iron-group values, and the transient is fast, brighter, and blue. Everything else — when the light peaks, how bright, what colour — falls out of a race between two clocks: the diffusion clock $t_\mathrm{diff}\sim\kappa M_\mathrm{ej}/(cR)$ , which shrinks as the cloud dilates, and the age of the explosion. The luminosity peaks when the two read the same, and the second figure shows the two-humped light curve AT2017gfo actually traced.

Figure

The two-component kilonova light curve: a fast blue peak near L ≈ 10⁴² erg s⁻¹ at t ≈ 1 day from low-opacity lanthanide-poor ejecta, and a slow red peak near L ≈ 10⁴¹ erg s⁻¹ at t ≈ 1 week from high-opacity lanthanide-rich ejecta, each fading along the r-process heating law q̇ ∝ t⁻¹·³ — the shape AT2017gfo traced out

Because the heating law and thermalization are close to universal, the peak luminosity and time encode the ejecta mass, velocity, and composition almost directly — making a kilonova reverse-engineerable: measure the light curve, infer how much gold-forming material was thrown out, and tie that back to the neutron-star radii and the fate of the remnant.

The core equations

The engine is radioactive decay heat. Rather than the single dominant chain $^{56}\mathrm{Ni}\to{}^{56}\mathrm{Co}\to{}^{56}\mathrm{Fe}$ of a Type Ia supernova, a kilonova is heated by thousands of $r$ -process species decaying at once, whose superposition is a robust power law normalized to about $10^{10}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ at one day.

\dot q(t)\;\approx\;\dot q_0\left(\frac{t}{t_0}\right)^{-\alpha},\qquad \alpha\approx1.3

where: $\dot q$ — specific heating rate (energy per unit mass per unit time) from radioactive decay; $\dot q_0$ — normalization, $\sim10^{10}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ at $t_0\sim1$ day; $\alpha$ — power-law index, $\approx1.3$ , robust across abundance patterns; $t$ — time since merger. The engine dims slowly, so the light curve is shaped less by the heating than by when radiation can escape. Near and after maximum the observed luminosity tracks the instantaneous heating rate — Arnett's rule — modulated only by the thermalization efficiency.

L(t)\;\approx\;f_\mathrm{th}(t)\,\dot q(t)\,M_\mathrm{ej}\qquad(\text{near and after peak})

where: $L$ — bolometric luminosity; $f_\mathrm{th}$ — thermalization efficiency, $\lesssim1$ , falling at late times as gamma rays and $\beta$ -electrons leak out; $M_\mathrm{ej}$ — total ejecta mass, $\sim10^{-3}$ – $10^{-1}\,M_\odot$ . At peak, luminosity is heating rate times mass — invert it and you weigh the ejecta. The peak occurs when the diffusion time out of the expanding sphere falls to equal the elapsed time; setting $t_\mathrm{diff}=t$ gives a compact expression for it.

t_\mathrm{peak}\;\approx\;\left(\frac{\kappa\,M_\mathrm{ej}}{4\pi\,\beta\,v_\mathrm{ej}\,c}\right)^{1/2}

where: $t_\mathrm{peak}$ — time of light-curve maximum; $\kappa$ — grey opacity ( $\sim0.1$ – $1$ blue, $\sim10$ red, in $\mathrm{cm^2\,g^{-1}}$ ); $v_\mathrm{ej}$ — characteristic ejecta velocity, $\sim0.1$ – $0.3\,c$ ; $\beta$ — geometry factor of order unity ( $\approx3$ for centrally concentrated profiles); $c$ — speed of light. Opacity and mass push the peak later; faster ejecta pulls it earlier. Evaluating the constants for fiducial merger ejecta collapses the scaling into a memorable form.

t_\mathrm{peak}\;\sim\;1.6\ \mathrm{d}\left(\frac{M_\mathrm{ej}}{0.01\,M_\odot}\right)^{1/2}\!\left(\frac{\kappa}{1\,\mathrm{cm^2\,g^{-1}}}\right)^{1/2}\!\left(\frac{v_\mathrm{ej}}{0.1c}\right)^{-1/2}

where: the units are chosen so the prefactor is $1.6$ days; red ejecta ( $\kappa\sim10$ ) push $t_\mathrm{peak}$ to $\sim5$ – $7$ days, blue ejecta ( $\kappa\sim0.1$ – $1$ ) keep it under a day. A factor $100$ in opacity delays and reddens the peak by $\sim10$ — the whole blue/red dichotomy in one line. The photosphere at peak radiates roughly as a blackbody, tying luminosity to radius and temperature and hence to observed colour.

L_\mathrm{peak}\;=\;4\pi R_\mathrm{peak}^2\,\sigma\,T_\mathrm{eff}^4,\qquad R_\mathrm{peak}=v_\mathrm{ej}\,t_\mathrm{peak}

where: $R_\mathrm{peak}$ — photospheric radius at peak; $\sigma$ — Stefan–Boltzmann constant, $5.67\times10^{-5}\,\mathrm{erg\,cm^{-2}\,s^{-1}\,K^{-4}}$ ; $T_\mathrm{eff}$ — effective temperature, $\sim10^4\,\mathrm{K}$ (blue) down to a few $\times10^3\,\mathrm{K}$ (red). Because $R_\mathrm{peak}$ grows while $L$ is roughly fixed, $T_\mathrm{eff}$ falls — the source reddens as it ages. The bulk kinetic energy of the outflow dwarfs everything it radiates — the light curve is a tiny leak from a large reservoir.

E_\mathrm{kin}\;=\;\tfrac12 M_\mathrm{ej}\,v_\mathrm{ej}^2\;\sim\;10^{50}\text{–}10^{51}\,\mathrm{erg}

where: $E_\mathrm{kin}$ — bulk kinetic energy; $M_\mathrm{ej}\sim10^{-2}\,M_\odot$ ; $v_\mathrm{ej}\sim0.1$ – $0.3c$ . The radiated energy ( $\sim10^{47}$ – $10^{48}\,\mathrm{erg}$ ) is a thousandth of this — the kilonova is a faint by-product of a violent event. Finally, the nucleosynthesis itself requires neutron capture to outrun beta decay, fixing the neutron density the ejecta must reach.

\lambda_{n\gamma}\,n_n\;\gg\;\lambda_{\beta}\;\;\Longleftrightarrow\;\;\tau_n\ll\tau_\beta

where: $\lambda_{n\gamma}$ — neutron-capture rate per nucleus per unit neutron density; $n_n$ — free-neutron number density, $\gtrsim10^{24}\,\mathrm{cm^{-3}}$ required; $\lambda_\beta$ — beta-decay rate; $\tau_n,\tau_\beta$ — capture and beta-decay timescales. Only when captures win do nuclei march up to the actinides, seeding the fuel the kilonova later burns.

Derivation AWhy the r-process heating rate is a near-universal power law

The goal is to explain the single most surprising fact about kilonova power — that thousands of different nuclei with wildly different half-lives conspire to produce a heating rate close to $t^{-1}$ , almost independent of which isotopes actually form. This near-universality lets us predict a kilonova's brightness without a full reaction network; it is a statement about the statistics of the decaying ensemble, not about any one nucleus. We build it from a flat distribution of decay timescales, integrate over the ensemble, and explain why the true index is $1.3$ rather than exactly $1$ .

Step 1. Suppose the number of radioactive species with decay timescale near $\tau$ is roughly flat per logarithmic interval, $\mathrm{d}N/\mathrm{d}\ln\tau\approx$ const, over many decades — a fair description of the sprawl of $r$ -process nuclei across the chart of nuclides. Each species stores energy $Q$ and releases it over its lifetime $\tau$ , contributing power $\sim Q/\tau$ while it is still alive and switching off once $t\gg\tau$ . Why: the $r$ -process populates so many isotopes that no single half-life dominates, so the ensemble, not the individual, sets the heating.

Step 2. Sum the contributions with an exponential cutoff for species that have already decayed away.

\dot q(t)\;\sim\;\int \frac{Q}{\tau}\,e^{-t/\tau}\,\frac{\mathrm{d}N}{\mathrm{d}\ln\tau}\,\mathrm{d}\ln\tau

Why: the factor $e^{-t/\tau}$ removes short-lived species ( $\tau\ll t$ ) that have burned out, leaving the integral dominated by those with $\tau\sim t$ — the population that is decaying right now.

Step 3. With $\mathrm{d}N/\mathrm{d}\ln\tau$ constant, the integral is controlled by $\tau\sim t$ , where the integrand scales as $Q/\tau\sim Q/t$ , so the collective heating is

\dot q(t)\;\propto\;t^{-1}.

Why: at each moment the "active" isotopes are those with $\tau\approx t$ ; as $t$ grows the baton passes to ever longer-lived species, and the flat log-distribution makes the hand-off scale-free — a power law.

Step 4. The real index is $\alpha\approx1.3$ , slightly steeper than $1$ , because the density of states declines gently toward long half-lives and the thermalization efficiency $f_\mathrm{th}$ itself falls as gamma rays and electrons begin to escape. Network calculations (Metzger 2020) confirm $\dot q_0\sim10^{10}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ at one day with $\alpha=1.1$ – $1.4$ . Why: the deviation from exactly $t^{-1}$ is the residual imprint of real nuclear physics — a small correction; the power-law form is the robust, model-independent backbone.

Go deeper · Derivation B · 2 minKilonova peak time and luminosity from the Arnett balance

Derivation BKilonova peak time and luminosity from the Arnett balance

The goal is to combine radioactive heating, homologous expansion, and photon diffusion into a single prediction for when and how bright a kilonova peaks, recovering the observed day-to-week timescales and $\sim10^{40}$ – $10^{42}\,\mathrm{erg\,s^{-1}}$ luminosities from a handful of inputs. We treat the ejecta as a uniform freely-expanding sphere heated by $\dot q(t)\propto t^{-\alpha}$ , write energy conservation for its trapped radiation, and find the maximum occurs exactly when radiative losses catch up to the heating — Arnett's rule — with a peak time fixed by the diffusion condition.

Step 1. Track the internal radiation energy $E$ . It gains heat from decay and loses energy two ways — adiabatically as the sphere expands (rate $E/t$ ) and radiatively as photons diffuse out (rate $E/t_\mathrm{diff}$ ).

\frac{\mathrm{d}E}{\mathrm{d}t}=-\frac{E}{t}-\frac{E}{t_\mathrm{diff}}+\dot q(t)\,M_\mathrm{ej}

Why: this is the first law for a radiation-dominated, expanding gas — $PdV$ work drains energy as $E/t$ under homologous expansion, and diffusion carries off the observed luminosity.

Step 2. The escaping luminosity is $L=E/t_\mathrm{diff}$ , and with homologous expansion $R=v_\mathrm{ej}t$ the diffusion time falls inversely with time.

L=\frac{E}{t_\mathrm{diff}},\qquad t_\mathrm{diff}=\frac{\kappa M_\mathrm{ej}}{4\pi\beta c\,v_\mathrm{ej}t}

Why: optical depth $\tau_\mathrm{opt}=\kappa\rho R\propto1/t^2$ and the random-walk escape time $t_\mathrm{diff}\sim(R/c)\tau_\mathrm{opt}\propto1/t$ — the fog thins fast as the cloud grows.

Step 3. At maximum, $\mathrm{d}L/\mathrm{d}t=0$ , and the differential equation reduces to the statement that radiative losses instantaneously balance heating.

\left.\frac{\mathrm{d}L}{\mathrm{d}t}\right|_{t_\mathrm{peak}}=0\;\Longrightarrow\;L(t_\mathrm{peak})=f_\mathrm{th}\,\dot q(t_\mathrm{peak})\,M_\mathrm{ej}

Why: this is Arnett's rule — the defining property of a radioactively powered transient. It holds regardless of the detailed density profile, which is why it is so useful for inverting light curves.

Step 4. The peak coincides with the moment the diffusion time equals the age, $t_\mathrm{diff}=t$ , giving the square-root scaling.

t_\mathrm{peak}=\left(\frac{\kappa M_\mathrm{ej}}{4\pi\beta v_\mathrm{ej}c}\right)^{1/2}

Why: before this instant photons are trapped and the luminosity rises; after it they escape faster than they are produced and the light fades — the crossover is the maximum.

Step 5. Insert fiducial numbers: $M_\mathrm{ej}=0.01\,M_\odot=2\times10^{31}\,\mathrm{g}$ , $\kappa=1\,\mathrm{cm^2\,g^{-1}}$ , $v_\mathrm{ej}=0.1c=3\times10^9\,\mathrm{cm\,s^{-1}}$ , $\beta\approx1$ . Then $t_\mathrm{peak}\approx1.5\times10^5\,\mathrm{s}\approx1.6\,\mathrm{d}$ , and with $\dot q(1.6\,\mathrm{d})\approx5\times10^9\,\mathrm{erg\,g^{-1}\,s^{-1}}$ and $f_\mathrm{th}\approx0.5$ ,

L_\mathrm{peak}\approx0.5\times5\times10^9\times2\times10^{31}\approx5\times10^{40}\,\mathrm{erg\,s^{-1}}.

Why: this reproduces the observed kilonova scale; lanthanide-rich ejecta ( $\kappa\sim10$ ) shift the peak to $\sim5$ – $7$ days and into the near-IR, exactly the AT2017gfo two-component behaviour.

Go deeper · Derivation C · 1 minEjecta mass from the observed peak and the Galactic budget check

Derivation CEjecta mass from the observed peak and the Galactic budget check

The goal is to run the Arnett machine backward — from the measured peak luminosity and time of AT2017gfo to the mass of $r$ -process material actually ejected — and then check whether mergers, at their inferred rate, can supply the heavy elements in the Milky Way. This is the argument that turned kilonovae into the leading explanation for the cosmic origin of gold and platinum, so it is worth doing explicitly and confronting with the Galactic inventory.

Step 1. Read off the observables: AT2017gfo's blue emission peaked near $L_\mathrm{peak}\sim10^{42}\,\mathrm{erg\,s^{-1}}$ at $t_\mathrm{peak}\sim1$ day, its later red emission near $10^{41}\,\mathrm{erg\,s^{-1}}$ at about a week. Why: the two humps are two ejecta components; inverting each separately yields the total mass and its blue/red split.

Step 2. Evaluate the engine at that epoch: $\dot q(1\,\mathrm{d})\sim10^{10}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ with $f_\mathrm{th}\sim0.5$ . Why: at one day roughly half the decay energy is still trapped and thermalized; the rest escapes as gamma rays and neutrinos.

Step 3. Invert Arnett's rule, $M_\mathrm{ej}\approx L_\mathrm{peak}/[f_\mathrm{th}\,\dot q(t_\mathrm{peak})]$ .

M_\mathrm{ej}\;\sim\;\frac{10^{42}}{0.5\times10^{10}}\,\mathrm{g}\;\sim\;2\times10^{32}\,\mathrm{g}\;\sim\;0.1\,M_\odot

Why: this order-of-magnitude estimate brackets the full radiative-transfer result, $M_\mathrm{ej}\approx0.05\,M_\odot$ total, split into a lighter fast blue component and a heavier slow red component.

Step 4. Confront the Galactic budget. With a binary neutron-star merger rate $\mathcal R_\mathrm{BNS}\sim10^{-4}\,\mathrm{galaxy^{-1}\,yr^{-1}}$ , $r$ -process yield $M_\mathrm{ej}^{r}\sim0.05\,M_\odot$ per event, over a Hubble time $t_\mathrm{H}\sim10^{10}\,\mathrm{yr}$ :

\mathcal R_\mathrm{BNS}\,M_\mathrm{ej}^{r}\,t_\mathrm{H}\;\sim\;10^{-4}\times0.05\times10^{10}\;\sim\;5\times10^{4}\,M_\odot.

Why: the Milky Way holds of order $10^{4}\,M_\odot$ of $r$ -process elements, so mergers can supply the bulk — the arithmetic closes, and mergers graduate from "a site" to "a dominant site".

Limiting cases

Kilonova phenomenology partitions by ejecta composition (through $Y_e$ and hence opacity) and by the fate of the remnant, which sets how much neutrino irradiation the ejecta receives.

Case	Result	Meaning
Red (lanthanide-rich) kilonova	$\kappa\sim10\,\mathrm{cm^2\,g^{-1}},\ Y_e\lesssim0.25$	Dynamical tidal ejecta, very neutron-rich; peaks in the near-IR after $\sim1$ week at $L\sim10^{40}$ – $10^{41}\,\mathrm{erg\,s^{-1}}$ . Dominated AT2017gfo at late times.
Blue (lanthanide-poor) kilonova	$\kappa\sim0.1$ – $1\,\mathrm{cm^2\,g^{-1}},\ Y_e\gtrsim0.30$	Neutrino-reprocessed or shock-heated ejecta; peaks in optical/UV within $\sim1$ day, brighter and faster. Seen early in AT2017gfo.
Prompt collapse to black hole	$M_\mathrm{tot}\gtrsim1.2\,M_\mathrm{max}^\mathrm{TOV}$	Little disk wind, no long neutrino bath; ejecta stays very neutron-rich and lanthanide-dominated — a faint, predominantly red kilonova.
Long-lived (supramassive/stable) remnant	$t_\mathrm{collapse}\gg t_\mathrm{dyn}$	Copious neutrinos raise $Y_e$ ; strong blue component plus possible extra energy from remnant rotation or magnetic fields.

Common trapThe kilonova is the same thing as the short gamma-ray burst

It is not. A merger produces (at least) three distinct signals, and confusing them is the classic error. The short gamma-ray burst and its afterglow are non-thermal emission from a collimated, ultra-relativistic jet ( $\Gamma\gtrsim100$ ), prompt in gamma rays and beamed into a narrow cone. The kilonova is thermal (quasi-blackbody), quasi-isotropic emission from mildly relativistic ejecta ( $v\sim0.1$ – $0.3c$ ) that peaks over days in the optical and infrared. They come from different matter moving at different speeds in different directions. A second, subtler trap: do not equate "kilonova" with " $r$ -process". The $r$ -process is the nucleosynthesis and would happen even if we saw no light; the kilonova is merely how the decay heat of that freshly made matter becomes visible. Finally, the $t^{-1.3}$ heating law is a statistical property of many nuclei — it breaks down at late times ( $\gtrsim1$ week) when a few long-lived isotopes and spontaneous fission take over and the decline steepens or develops bumps.

Worked examples

1 · Peak time and luminosity of the fiducial blue component (fully worked)

Take blue ejecta with $M_\mathrm{ej}=0.02\,M_\odot$ , $\kappa=0.5\,\mathrm{cm^2\,g^{-1}}$ , $v_\mathrm{ej}=0.2c$ . Estimate the peak time, the peak luminosity, and the effective temperature at peak.

Step 1. Peak time from the scaled relation, using the prefactor $1.6$ days: $t_\mathrm{peak}=1.6\,(0.02/0.01)^{1/2}(0.5/1)^{1/2}(0.2/0.1)^{-1/2}$ days $=1.6\times1.414\times0.707\times0.707=\mathbf{1.13\ days}$ . The larger mass pushes the peak later, but the low opacity and fast expansion pull it back under $\sim1$ day — the signature of a blue kilonova.

Step 2. Heating rate there: $\dot q(1.13\,\mathrm{d})=10^{10}(1.13)^{-1.3}\approx8.6\times10^{9}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ . With $M_\mathrm{ej}=0.02\,M_\odot=4\times10^{31}\,\mathrm{g}$ and $f_\mathrm{th}\approx0.5$ , Arnett's rule gives $L_\mathrm{peak}=0.5\times8.6\times10^{9}\times4\times10^{31}\approx\mathbf{1.7\times10^{41}\ erg\,s^{-1}}$ .

Step 3. Effective temperature from the blackbody photosphere. The peak radius is $R_\mathrm{peak}=v_\mathrm{ej}t_\mathrm{peak}=(6\times10^9)(1.13\times86400)=5.9\times10^{14}\,\mathrm{cm}$ , so $T_\mathrm{eff}=[L_\mathrm{peak}/(4\pi R_\mathrm{peak}^2\sigma)]^{1/4}=[1.7\times10^{41}/2.5\times10^{26}]^{1/4}\approx\mathbf{5100\ K}$ . A blue-to-optical transient peaking near a day at $\sim2\times10^{41}\,\mathrm{erg\,s^{-1}}$ with $T_\mathrm{eff}\sim5000$ K — consistent with AT2017gfo's early emission, which was hotter still ( $\sim10^4$ K) at first, when its photosphere was smaller and faster.

2 · Peak time of the red component

Lanthanide-rich ejecta with $M_\mathrm{ej}=0.04\,M_\odot$ , $\kappa=10\,\mathrm{cm^2\,g^{-1}}$ , $v_\mathrm{ej}=0.1c$ . When does it peak?

Step 1. Apply the scaled relation: $t_\mathrm{peak}=1.6\,(0.04/0.01)^{1/2}(10/1)^{1/2}(0.1/0.1)^{-1/2}$ days. Evaluate the factors: $(0.04/0.01)^{1/2}=\sqrt4=2$ , $(10)^{1/2}=3.16$ , velocity factor $=1$ .

Step 2. Multiply: $t_\mathrm{peak}=1.6\times2\times3.16\approx\mathbf{10\ days}$ . The high opacity alone stretches the peak from a day to a week and a half — precisely the slow near-infrared red component seen in AT2017gfo, and the reason infrared (rather than optical) follow-up is essential for catching the late kilonova. Taking $X_\mathrm{Au}\sim10^{-2}$ of this $0.04\,M_\odot$ as gold implies $\sim4\times10^{-4}\,M_\odot\sim10^{30}\,\mathrm{g}$ , of order a hundred Earth-masses of gold from one merger.

Physics of the portrait

Kilonovae are the payoff of a prediction made two decades before the confirming observation — and no one shaped it more than the astrophysicist who first tied gamma-ray bursts to merging neutron stars and then, with a student, wrote down the radioactively powered transient itself.

Origins · 1 minBohdan Paczyński and how this was discovered

Bohdan Paczyński

1940–2007 · gamma-ray bursts and the radioactively powered merger transient

Bohdan Paczyński was a Polish-American astrophysicist whose instinct for the decisive idea shaped several fields at once. Working in Warsaw and from 1982 at Princeton, he made foundational contributions to stellar evolution, accretion discs (the "Paczyński–Wiita" potential still used to mimic black-hole gravity), and gravitational microlensing — proposing to monitor millions of stars for dark matter and planets, which launched the OGLE survey. For kilonovae his two crucial moves came a decade apart. In 1986 he argued that cosmological gamma-ray bursts could be powered by merging neutron stars at vast distances, well before their extragalactic origin was established. Then in 1998, with his student Li-Xin Li, he modelled the transient such a merger should produce: an expanding cloud of neutron-rich ejecta heated from within by radioactive decay, glowing for days as the heat diffuses out. That Li–Paczyński model is the direct ancestor of every kilonova light-curve calculation, including those fit to AT2017gfo. Paczyński died in 2007 and did not live to see GW170817 — but the 2017 detection was, in essence, confirmation of a transient he had sketched nineteen years earlier, powered by exactly the radioactive engine he and Li proposed.

History

1974

— Lattimer and Schramm propose r-process ejecta from compact-object tidal disruption

1986

— Paczyński ties cosmological gamma-ray bursts to neutron-star mergers

1998

— Li and Paczyński predict a radioactively powered merger transient

2010

— Metzger and collaborators coin the name kilonova and fix its ∼10⁴¹ erg s⁻¹ scale

2017

— GW170817 and its kilonova AT2017gfo confirm mergers as an r-process site

Going further

Texts. Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 15–16 (compact objects, supernovae, nucleosynthesis) · Ryden, Introduction to Cosmology, 2nd ed. (standard sirens and distance measurement) · Dodelson, Modern Cosmology (gravitational-wave sources) · Mo, van den Bosch & White, Galaxy Formation and Evolution (chemical enrichment context) · Binney & Tremaine, Galactic Dynamics (binary evolution and merger rates).

Reviews and key papers. Metzger, "Kilonovae", Living Reviews in Relativity 23, 1 (2020) · Li & Paczyński, ApJL 507, L59 (1998) · Metzger et al., MNRAS 406, 2650 (2010) · Kasen et al., Nature 551, 80 (2017) · Abbott et al. (LIGO/Virgo + partners), "Multi-messenger Observations of a Binary Neutron Star Merger", ApJL 848, L12 (2017) · Watson et al., Nature 574, 497 (2019). The engine and light-curve scalings follow Arnett-type transient theory adapted to kilonovae. The $\dot q_0\sim10^{10}\,\mathrm{erg\,g^{-1}\,s^{-1}}$ normalization and $\alpha\approx1.3$ index come from detailed $r$ -process network calculations. The peak-time formula $t_\mathrm{peak}=\sqrt{\kappa M_\mathrm{ej}/(4\pi\beta v_\mathrm{ej}c)}$ follows directly from setting $t_\mathrm{diff}=t$ ; substituting cgs constants ( $c=3\times10^{10}\,\mathrm{cm\,s^{-1}}$ , $M_\odot=2\times10^{33}\,\mathrm{g}$ ) into the fiducial parameters recovers $t_\mathrm{peak}\approx1.6\,\mathrm{d}$ and $L_\mathrm{peak}\sim10^{41}\,\mathrm{erg\,s^{-1}}$ . The AT2017gfo benchmarks used throughout — distance $\sim40\,\mathrm{Mpc}$ , discovery delay $\sim11\,\mathrm{h}$ , $M_\mathrm{ej}\approx0.05\,M_\odot$ , $H_0\approx70\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$ — are order-of-magnitude estimates consistent with the published observations. Order-of-magnitude estimates are labelled as such; no non-standard numbers are asserted.

Prerequisites. Neutron Stars · Equation of State of Dense Matter · Tolman–Oppenheimer–Volkoff Equation · Neutron Star Structure and Interior. Leads to. Magnetars · Millisecond and Recycled Pulsars.

Ask the tutor: Ask me to walk through inverting a kilonova light curve to weigh its ejecta, or to show how the electron fraction decides whether a merger glows blue or red.

Why is it called a "kilonova" if it is fainter than a supernova?

The name compares it to a classical nova, not a supernova. At peak a kilonova reaches roughly a thousand times a nova's luminosity — hence "kilo" — but only about $10^{-3}$ of a typical supernova. Metzger and collaborators coined the term in 2010; earlier papers called it a "mini-supernova" or the "Li–Paczyński" transient. It is a genuine radioactively powered explosion, but a small, fast one, because the ejecta mass ( $\sim0.01$ – $0.05\,M_\odot$ ) is a fraction of a supernova's and the fuel is $r$ -process nuclei rather than the much larger nickel mass of a Type Ia.

How can the r-process heating rate be predictable when the nuclear physics is uncertain?

Because the prediction rests on statistics, not on any single isotope. A kilonova is heated by thousands of radioactive species with half-lives spread over many decades. When you sum their decays, the individual quirks — this isotope's exact half-life, that one's decay energy — average out, leaving the collective law $\dot q\propto t^{-1.3}$ , much as a large insurance pool has predictable payouts even though no single claim is. Reaction-network calculations refine the normalization, and the late-time behaviour (when a few long-lived nuclei dominate) is genuinely uncertain, but the power-law backbone near peak is robust — which is what makes kilonovae usable as mass-measuring tools.

What sets whether a given merger produces a blue or a red kilonova?

The electron fraction $Y_e$ , which controls how far the $r$ -process runs. If the ejecta stays very neutron-rich ( $Y_e\lesssim0.25$ ), the $r$ -process reaches the lanthanides, whose enormous opacity makes the transient red and slow. If neutrinos from a hot remnant, or shocks at the collision interface, raise $Y_e$ above $\sim0.25$ – $0.3$ , lanthanide production is suppressed, the opacity drops to iron-group values, and the transient is blue and fast. In practice a single merger launches both: cold, neutron-rich tidal tails in the equatorial plane (red) and hotter, neutrino-irradiated polar winds (blue). AT2017gfo showed blue first, then red — hence its colour evolved from blue to infrared over a week.

Why does the kilonova reddening happen over days rather than all at once?

Two effects combine. First, the two ejecta components peak at different times: the low-opacity blue component peaks within a day, the high-opacity red one takes a week, so the colour drifts red as blue fades and red rises. Second, even a single component reddens on its own, because the photosphere expands while the heating declines slowly. From $L=4\pi R^2\sigma T_\mathrm{eff}^4$ with $R=v t$ growing and $L$ roughly fixed, the temperature falls as $T_\mathrm{eff}\propto R^{-1/2}\propto t^{-1/2}$ . A cooler photosphere emits at longer wavelengths, so any kilonova drifts from optical toward infrared as it ages.

How well do we actually know the ejecta mass from the light curve?

The order-of-magnitude mass follows directly from Arnett's rule, $M_\mathrm{ej}\approx L_\mathrm{peak}/[f_\mathrm{th}\,\dot q(t_\mathrm{peak})]$ , and is robust — which is why AT2017gfo's $\sim0.05\,M_\odot$ was agreed quickly. The uncertainties live in the details: $f_\mathrm{th}$ is known only to a factor of a few, the opacity $\kappa$ depends on the uncertain lanthanide fraction, and $\beta$ absorbs the density profile. Multi-component radiative-transfer models tighten this, but mass, velocity, and composition stay degenerate at the tens-of-percent level from photometry alone. Spectroscopy — like the strontium feature in AT2017gfo — breaks some of that degeneracy by pinning composition.

Do neutron-star–black-hole mergers make kilonovae too?

They can, but only under specific conditions. For the neutron star to be tidally disrupted outside the black hole's horizon — and so leave mass behind to shine — the black hole must be low-mass and/or rapidly spinning and the neutron star relatively large (a stiffer equation of state helps). If the neutron star plunges in whole, it is swallowed with essentially no ejecta and no kilonova, only the gravitational-wave signal. When disruption does occur, the ejecta is cold and extremely neutron-rich, so the kilonova tends to be red and lanthanide-dominated, often with more mass than a binary-neutron-star event. The gravitational-wave-measured masses are the cleaner discriminant between channels.