Chapter 2 of 6
←  3C 66B: the story and the chapters
Chapter 2 · General relativity

Gravitational waves: how loudly a black hole binary radiates and how long it can live

A billion-solar-mass binary on a one-year orbit radiates gravitational waves strongly enough to destroy its own orbit. The quadrupole formula tells us how fast, and turns a measured frequency drift into a mass.

The two models of 3C 66B differ by a factor of sixteen in chirp mass. Would either binary live long enough for us to watch it? The answer comes from one physical law, the emission of gravitational waves (GWs) by a time-varying mass quadrupole. This chapter derives that law from Einstein’s equations and applies it to the two models.

01Two models, two lifetimes

Gravitational radiation sets a clock on every compact binary. Sudou et al. (2003) estimated that their binary, with total mass of order $6\times10^{10}\,M_\odot$ on a 1.05-year orbit, would merge by gravitational radiation in about 5 years. Iguchi et al. (2010) estimated about 500 years for their lighter binary. A merger within five years of the discovery would be a spectacular prediction, and a dangerous one for the model.

Think first

If the 2003 model were correct, what would astronomers have seen in the years after 2003, before any gravitational-wave search?

Hint

A binary that merges within $\tau$ years must change its orbital period on a timescale comparable to $\tau$.

Answer

The orbital period itself would shrink by several per cent per year (Section 05 gives about 8% per year), so the core ellipse would visibly speed up, and the binary would disappear by around 2006, probably with a powerful electromagnetic and gravitational signal. A model can be tested by its own consequences long before a dedicated experiment is built.

02Gravity in the weak-field limit

Far from the binary, spacetime is nearly flat, and Einstein’s equations reduce to a wave equation. We write the metric as the flat Minkowski metric plus a small perturbation, $g_{\mu\nu} = \eta_{\mu\nu}+h_{\mu\nu}$ with $|h_{\mu\nu}|\ll1$, and keep only terms linear in $h$. It is convenient to use the trace-reversed perturbation $\bar h_{\mu\nu} = h_{\mu\nu}-\tfrac12\eta_{\mu\nu}h$, where $h = \eta^{\mu\nu}h_{\mu\nu}$.

Derivation 2.1 · The quadrupole formula for the wave
Einstein’s equations, linearised$$\Box\bar h_{\mu\nu} = -\frac{16\pi G}{c^4}\,T_{\mu\nu}$$in the Lorenz gauge $\partial^\mu\bar h_{\mu\nu}=0$, with $\Box = -c^{-2}\partial_t^2+\nabla^2$Maggiore Vol. 1, Ch. 1; LL Vol. 2, §107
a wave equation with a source: solve it with the retarded Green’s function of $\Box$
The retarded solution$$\bar h_{\mu\nu}(t,\vect x) = \frac{4G}{c^4}\!\int\!\frac{T_{\mu\nu}(t-|\vect x-\vect x'|/c,\ \vect x')}{|\vect x-\vect x'|}\,\dd^3x'$$the same solution as for the potentials of electrodynamicsLL Vol. 2, §62; Jackson, §6.4
far zone $r\gg\lambda$; slow source, size $\ll\lambda$: replace $|\vect x-\vect x'|$ by $r$; then impose the Lorenz gauge of the first box
Only the spatial components radiate$$\bar h_{ij}(t,\vect x)\simeq\frac{4G}{c^4r}\int T_{ij}\!\left(t-\tfrac rc,\ \vect x'\right)\dd^3x'$$For an outgoing wave, the gauge condition fixes $\bar h_{00}$ and $\bar h_{0i}$ from $\bar h_{ij}$, and both vanish in the TT gauge; they carry no independent radiation.
use energy–momentum conservation $\partial_\mu T^{\mu\nu}=0$ twice, $T^{00}\simeq\rho c^2$
The virial identity$$\int T^{ij}\,\dd^3x = \frac12\,\frac{\dd^2}{\dd t^2}\int\rho\,x^ix^j\,\dd^3x \equiv \frac12\,\ddot I^{ij}$$Maggiore Vol. 1, Ch. 3
project onto the transverse–traceless (TT) part, the only part that is physical far away; the projection removes the trace, so $I_{ij}$ may be replaced by its traceless part $Q_{ij}$
The quadrupole formula$$h_{ij}^{\rm TT}(t,\vect x) = \frac{2G}{c^4r}\,\ddot Q_{ij}^{\rm TT}\!\left(t-\frac rc\right),\qquad Q_{ij} = \int\rho\left(x_ix_j-\tfrac13\delta_{ij}|\vect x|^2\right)\dd^3x$$LL Vol. 2, §110; Maggiore Vol. 1, Ch. 3

Two features of this result deserve a pause. The wave amplitude falls as $1/r$, like any radiation field, which is why gravitational waves from 94 Mpc away can be measured at all. And the source is the second time derivative of the mass quadrupole moment: a static mass, or a spherically pulsating one, does not radiate.

Think first

Electromagnetic radiation is dominated by the electric dipole. Why is there no dipole gravitational radiation?

Hint

The gravitational analogue of the dipole moment is the mass dipole $\vect d = \int\rho\,\vect x\,\dd^3x$. What are $\dot{\vect d}$ and $\ddot{\vect d}$?

Answer

The monopole, the total mass, is conserved, so it cannot radiate. Mass has only one sign, so the mass dipole is the total mass times the position of the centre of mass, $\vect d = M\vect R$, and $\ddot{\vect d} = M\ddot{\vect R} = 0$ by momentum conservation (Chapter 1). In electromagnetism, charges of both signs let the dipole oscillate while the total charge stays fixed. The lowest radiating gravitational multipole is therefore the quadrupole (LL Vol. 2, §110), whose power is smaller by a factor of order $(v/c)^2$ than that of a dipole of the same size and speed.

03How much power the waves carry

A gravitational wave carries energy, and the quadrupole formula tells us how much. Averaged over several wavelengths, the energy density of a gravitational wave is $t^{00} = \frac{c^2}{32\pi G}\langle\dot h^{\rm TT}_{ij}\dot h^{\rm TT}_{ij}\rangle$ (Isaacson 1968). Integrating the flux over a sphere gives the total power.

Derivation 2.2 · The quadrupole formula for the power
Energy of the wave$$\frac{\dd E}{\dd t\,\dd A} = \frac{c^3}{32\pi G}\,\big\langle\dot h^{\rm TT}_{ij}\dot h^{\rm TT}_{ij}\big\rangle$$Isaacson 1968; Maggiore Vol. 1, Ch. 1
The wave$$h_{ij}^{\rm TT} = \frac{2G}{c^4r}\,\Lambda_{ij,kl}(\uv n)\,\ddot Q_{kl}$$$\Lambda$ projects onto the plane transverse to $\uv n$Derivation 2.1
integrate over the sphere, $\dd A = r^2\dd\Omega$: the $r^2$ cancels the $1/r^2$
An angular average of the projector$$\oint\Lambda_{ij,kl}\,\dddot Q_{ij}\dddot Q_{kl}\,\dd\Omega = \frac{8\pi}{5}\,\dddot Q_{ij}\dddot Q_{ij}\quad\text{for traceless } Q$$Maggiore Vol. 1, Ch. 3
collect the constants
Luminosity in gravitational waves$$L_{\rm GW} = \frac{G}{5c^5}\,\big\langle\dddot Q_{ij}\,\dddot Q_{ij}\big\rangle$$LL Vol. 2, §110

The prefactor $G/c^5 = 2.8\times10^{-60}\ {\rm s\,erg^{-1}}$ is tiny, which is why no laboratory source radiates measurably. For astrophysical binaries, the third time derivative of $Q$ is enormous instead: of order the mass times the square of the orbital velocity times the orbital frequency.

04A circular binary

A circular binary is a rotating dumbbell, and its quadrupole moment repeats every half orbit. By Chapter 1, the two-body problem reduces to one body of reduced mass $\mu = m_1m_2/M$ on the relative orbit. Put the relative orbit in the $xy$ plane, $x = a\cos\Phi$, $y = a\sin\Phi$, with orbital phase $\Phi = \omega t$ and angular frequency $\omega^2 = GM/a^3$.

Derivation 2.3 · The waveform of a circular binary
Quadrupole moment of the orbit$$I_{xx} = \tfrac12\mu a^2(1+\cos2\Phi),\quad I_{yy} = \tfrac12\mu a^2(1-\cos2\Phi),\quad I_{xy} = \tfrac12\mu a^2\sin2\Phi$$Chapter 1: $\mu$ on the relative orbit
only $2\Phi$ appears: the GW frequency is twice the orbital frequency, $f = \omega/\pi$
Projection onto the observer’s skyAn observer at inclination $\iota$ to the orbital axis (the inclination $i$ of Chapter 1) sees the wave along $\uv n = \uv y\sin\iota+\uv z\cos\iota$, with polarisation axes $\uv e_1 = \uv x$ and $\uv e_2 = \uv y\cos\iota-\uv z\sin\iota$. Derivation 2.1 gives $$h_+ = \frac{G}{c^4r}\big(\ddot I_{11}-\ddot I_{22}\big),\qquad h_\times = \frac{2G}{c^4r}\,\ddot I_{12},$$ with $\ddot I_{11} = \ddot I_{xx}$, $\ddot I_{22} = \ddot I_{yy}\cos^2\iota$ and $\ddot I_{12} = \ddot I_{xy}\cos\iota$. Each $\ddot I$ has amplitude $2\omega^2\mu a^2$.Maggiore Vol. 1, Ch. 3
eliminate $a$ with Kepler’s third law, $a = (GM/\omega^2)^{1/3}$
The two polarisations$$\begin{aligned}h_+ &= -\frac{4}{r}\left(\frac{G\mathcal M}{c^2}\right)^{5/3}\!\left(\frac{\pi f}{c}\right)^{2/3}\frac{1+\cos^2\iota}{2}\cos2\Phi,\\ h_\times &= -\frac{4}{r}\left(\frac{G\mathcal M}{c^2}\right)^{5/3}\!\left(\frac{\pi f}{c}\right)^{2/3}\cos\iota\,\sin2\Phi\end{aligned}$$with the chirp mass $\mathcal M = \mu^{3/5}M^{2/5} = (m_1m_2)^{3/5}/(m_1+m_2)^{1/5}$. The minus signs only fix the origin of $\Phi$: at $\Phi = 0$ the black holes lie along $\uv e_1$, and test masses along that line are squeezed. Shifting $\Phi\to\Phi+\pi/2$, as Maggiore does, makes both signs positive.Maggiore Vol. 1, Ch. 3 and §4.1

The masses enter the amplitude only through $\mu M^{2/3} = \mathcal M^{5/3}$. This combination, the chirp mass, is what gravitational waves measure first, and it is the quantity that every 3C 66B result below is quoted in.

Freely falling masses register the wave through their separations. By the equation of geodesic deviation, written in the TT gauge, two free masses separated by a distance $L$ along the unit vector $\uv n$ change their separation by $\delta L/L = \tfrac12h_{ij}n^in^j$ (Maggiore Vol. 1, Ch. 1). A ring of free masses facing the wave is therefore squeezed into an ellipse, which rotates or oscillates depending on the polarisation.

Interactive 2.1

The rotating quadrupole and the wave it makes

(A) The binary from above. The shaded ellipse shows the second moment of its mass distribution; after half an orbit it returns to itself. (B) A ring of freely falling test masses facing the wave, displaced by $\delta x = \tfrac12(h_+x+h_\times y)$, $\delta y = \tfrac12(h_\times x-h_+y)$; the amplitude is exaggerated by about $10^{14}$. (C) The two polarisations at the observer, compared with the position of one black hole along $x$: the wave completes two cycles per orbit. Change the inclination and watch both polarisations.
Think first

Set the inclination to 0° and then to 90°. Describe the motion of the ring in each case. In which case are the two polarisations equal?

Answer

Face-on ($\iota = 0$), $h_+$ and $h_\times$ have equal amplitudes and are a quarter cycle apart: the ring’s deformation rotates without changing its shape, which is circular polarisation. Edge-on ($\iota = 90°$), $h_\times = 0$ and $h_+$ is halved: the ring is squeezed and stretched along fixed axes, which is linear polarisation. The orientation of the orbit is written in the polarisation of the wave, as it is for light from a rotating dipole.

05Losing energy: the inspiral and the chirp mass

The power radiated by the binary comes out of its orbital energy, so the orbit shrinks and the frequency rises. Substituting the circular orbit into Derivation 2.2 gives the luminosity of LL Vol. 2, §110, Problem 1, which Peters & Mathews (1963) extended to eccentric orbits. Energy balance then gives the rate at which the orbit decays.

Derivation 2.4 · The inspiral
Power radiated$$L_{\rm GW} = \frac{32}{5}\frac{G^4}{c^5}\frac{\mu^2M^3}{a^5}$$Derivation 2.2 with $\dddot Q_{ij}\dddot Q_{ij} = 32\mu^2a^4\omega^6$ (Problem 2); LL Vol. 2, §110, problem
Orbital energy$$E = -\frac{G\mu M}{2a}$$Chapter 1, Problem 2
energy balance, $\dd E/\dd t = -L_{\rm GW}$
The orbit shrinks$$\dot a = -\frac{64}{5}\frac{G^3\mu M^2}{c^5a^3}\quad\Longrightarrow\quad a^4(t) = a_0^4-\frac{256}{5}\frac{G^3\mu M^2}{c^5}\,t$$
use $f = \omega/\pi\propto a^{-3/2}$, so $\dot f/f = -\tfrac32\,\dot a/a$
The chirp$$\dot f = \frac{96}{5}\,\pi^{8/3}\left(\frac{G\mathcal M}{c^3}\right)^{5/3}f^{11/3},\qquad \tau = \frac38\frac{f}{\dot f} = \frac{5}{256}\left(\frac{G\mathcal M}{c^3}\right)^{-5/3}(\pi f)^{-8/3}$$Only the chirp mass appears: a measured frequency drift weighs the binary.Peters 1964; Maggiore Vol. 1, §4.1

For 3C 66B, with $f = 2/P = 60.4$ nHz:

2003 model, $\mathcal M\approx1.3\times10^{10}\,M_\odot$τ ≈ 5 yr; $\dot f\approx$ 4.6 nHz/yr; $h_0\approx$ 7×10⁻¹³Sudou et al. 2003; Jenet et al. 2004
2010 model, $\mathcal M = 7.9\times10^{8}\,M_\odot$τ ≈ 500 yr; $\dot f\approx$ 0.04 nHz/yr; $h_0\approx$ 7×10⁻¹⁵Iguchi et al. 2010; Cardinal Tremblay et al. 2026

Here $h_0 = 2(G\mathcal M)^{5/3}(\pi f)^{2/3}/(c^4D_L)$ is the strain amplitude in the convention of pulsar-timing papers. In this convention $h_+ = h_0(1+\cos^2\iota)\cos2\Phi$ and $h_\times = 2h_0\cos\iota\sin2\Phi$, identical to Derivation 2.3 with $\Phi$ shifted by $\pi/2$. The amplitude uses the redshifted chirp mass $(1+z)\mathcal M$ and the observed frequency $f$, as do $\dot f$ and $\tau$ when they are measured by clocks at the Earth.

The lifetimes are computed in the rest frame of the binary, where the GW frequency is $f(1+z)$. Iguchi et al. (2010) quoted 510 years from their separation and masses; the formula with the chirp mass gives the same number, because the two are related by Kepler’s third law. The 2010 model drifts by less than 1 nHz in twenty years, less than the frequency resolution $1/T$ of a twenty-year data set, so a search can treat its frequency as constant at the Earth (Cardinal Tremblay et al. 2026).

The power of Derivation 2.4 can be written as $L_{\rm GW} = \tfrac{32}{5}\,(c^5/G)\,\eta^2(v/c)^{10}$, with $\eta = \mu/M$ and $v$ the relative orbital speed. The prefactor $c^5/G = 3.6\times10^{59}$ erg s⁻¹ is enormous; what makes most binaries faint is the tenth power of $v/c$. For the 2010 model, $v/c\approx0.12$ and $\eta = 0.23$, so $L_{\rm GW}\approx10^{50}$ erg s⁻¹, about a million times the starlight of the Milky Way.

Interactive 2.2

The merger clock: how long a binary can live at each period and chirp mass

Colour and contours show the time left before merger by gravitational-wave emission alone, $\tau\propto\mathcal M^{-5/3}P^{8/3}$, so the contours are straight lines of slope 8/5. Amber: the 2003 model; blue: the 2010 model with its uncertainty; red: published upper limits on the chirp mass of 3C 66B, at its period (Chapter 3 and the story page). Click or drag to place a binary. The look-back control gives the GW frequency of the same binary some time ago, which a pulsar far from the Earth records (Chapter 3).
Think first · use Interactive 2.2

For the 2003 model, by how much would the orbital period have changed between the first and the last VLBI epochs, 15 months apart? Could the 2003 data have tested this?

Hint

$\dot P/P = -\dot f/f$. Read $\dot f$ and $f$ from the readouts.

Answer

$\dot f/f\approx4.6/60.4\approx8\%$ per year, so the period would have shortened by about 10% (0.1 yr) over the campaign, three times the quoted period uncertainty of 0.03 yr. The 2003 fits assumed a constant period, and six epochs could not measure a drift. A few more epochs would have tested the prediction directly: a model can be tested by its own consequences.

Think first

The 2010 binary would merge in about 500 years. How lucky would we have to be to find a binary so close to merger in a nearby galaxy?

Hint

The probability of catching a binary in a given stage is the time it spends there divided by the total time, if binaries form at a steady rate.

Answer

If a massive galaxy hosts at most about one such merger in the age of the Universe, $\sim10^{10}$ yr, the chance that it is within 500 years of the merger today is $\sim5\times10^{-8}$. Among the roughly $10^4$ massive galaxies within 100 Mpc, we expect $\sim5\times10^{-4}$ such systems. A candidate that close to merger therefore needs very strong evidence: this is a quantitative form of the principle that extraordinary claims require extraordinary evidence. The same counting, done carefully over the whole population of binaries, sets the gravitational-wave background that pulsar timing arrays measure.

DeeperEccentric orbits radiate more, at many harmonics

For an eccentric orbit the quadrupole moment is no longer a pure $2\Phi$ rotation, and the waves are emitted at all integer multiples $nF_{\rm orb}$ of the orbital frequency. Averaged over an orbit, the luminosity is enhanced by the factor (Peters & Mathews 1963)

$$F(e) = \frac{1+\tfrac{73}{24}e^2+\tfrac{37}{96}e^4}{(1-e^2)^{7/2}},$$

which is 1.8 for $e = 0.3$, 4.9 for $e = 0.5$ and 27 for $e = 0.7$. Eccentric binaries therefore merge faster and emit more power at higher harmonics; this is how the radiation of the Hulse–Taylor pulsar, with $e = 0.62$, was predicted and confirmed (Problem 1). Jenet et al. (2004) derived the timing signature of eccentric binaries for their 3C 66B search.

06What would make this wrong?

  • The environment. At separations of order 0.01 pc and wider, the orbits of massive binaries can also shrink by scattering stars and by torques from gas. At 3C 66B’s separation of $6\times10^{-3}$ pc in the 2010 model, gravitational radiation is expected to dominate, but the expectation is model dependent.
  • Higher post-Newtonian orders. The orbital speed in the 2010 model is $v/c\approx0.12$, so corrections of order $(v/c)^2\approx1.5\%$ are small at this stage, though not in the final orbits.
  • Conventions. Papers differ by factors of two in the definition of the strain amplitude, and in whether masses and frequencies are quoted in the rest frame of the source ($\mathcal M_{\rm obs} = (1+z)\mathcal M$, $f_{\rm obs} = f/(1+z)$). Always check the definition before comparing numbers.

07Problems

Estimate

The Hulse–Taylor binary pulsar has orbital period $P_b = 7.752$ h, eccentricity $e = 0.617$ and masses $1.44$ and $1.39\,M_\odot$. Using $\dot P_b = -\frac{192\pi}{5}\left(\frac{2\pi G\mathcal M}{c^3P_b}\right)^{5/3}F(e)$, predict the decay of its period. Compare with the measured $-2.40\times10^{-12}$ (Weisberg & Huang 2016).

Hint

$\mathcal M = 1.23\,M_\odot$ and $F(0.617) = 11.86$.

Answer

$\dot P_b\approx-2.40\times10^{-12}$: the quadrupole formula, derived above for a circular orbit and extended to eccentric ones, agrees with the measurement to better than one per cent.

Derive

Starting from Derivation 2.2 and the moments in Derivation 2.3, show that $\dddot Q_{ij}\dddot Q_{ij} = 32\mu^2a^4\omega^6$ for a circular orbit, and hence derive $L_{\rm GW} = \frac{32}{5}G^4\mu^2M^3/(c^5a^5)$.

Hint

Each component oscillates at $2\omega$ with amplitude $\tfrac12\mu a^2$; the third derivative multiplies it by $(2\omega)^3$. The trace part does not oscillate.

Derive

Show that the time to coalescence can be written in terms of the separation as $\tau = \frac{5}{256}\frac{c^5a^4}{G^3\mu M^2}$, and evaluate it for the 2010 masses, $1.2\times10^9$ and $7\times10^8\,M_\odot$, on an orbit with a period of 1.028 yr in the rest frame of the galaxy (1.05 yr observed).

Answer

$a = 6.1\times10^{-3}$ pc and $\tau\approx510$ yr, the value quoted by Iguchi et al. (2010).

Compute

Integrate the orbit-averaged Peters equations for $\dot a$ and $\dot e$ (Peters 1964) for a binary with the 2010 masses, a 1.05-yr period and initial eccentricity $e_0 = 0.5$. By what factor does the time to coalescence differ from the circular case?

Hint

$\dot a = -\frac{64}{5}\frac{G^3\mu M^2}{c^5a^3}F(e)$, $\dot e = -\frac{304}{15}\frac{G^3\mu M^2}{c^5a^4}\frac{e\,(1+\frac{121}{304}e^2)}{(1-e^2)^{5/2}}$. Integrate in $\log a$ with an adaptive step.

Answer

It merges in about 0.36 of the circular time, nearly three times faster, because $F(0.5)\approx4.9$ at the start, but the orbit circularises as it shrinks ($e\approx0.40$ halfway, where $F\approx2.8$), so the average enhancement is only about 2.8.

Explore

A pulsar records the gravitational wave that passed it earlier than the wave now passing the Earth, by $L(1-\cos\theta)/c$, where $L$ is its distance and $\theta$ its angle from the source on the sky (Chapter 3). For PSR B1855+09, $L\approx1$ kpc and $\theta = 98.5°$, a lag of about 3700 years. What were the frequencies of the 2003 and 2010 models that long ago? Use the look-back control of Interactive 2.2.

Answer

For a look-back of 3700 years, the 2003 binary had a frequency of about 5 nHz, a twelfth of today’s; the 2010 binary about 28 nHz, half of today’s. A single pulsar thus records the binary at two epochs and two frequencies, the “two-frequency response” of Chapter 3.

08References

  1. M. Maggiore, Gravitational Waves, Vol. 1: Theory and Experiments (Oxford University Press, 2008): Ch. 1 (linearised theory, the TT gauge, the energy of gravitational waves), Ch. 3 (generation of gravitational waves in linearised theory, quadrupole radiation), §4.1 (inspiral of compact binaries).
  2. L. D. Landau & E. M. Lifshitz, The Classical Theory of Fields, Course of Theoretical Physics Vol. 2, 4th ed. (Pergamon, 1975): §62 (retarded potentials), §107 (weak gravitational waves), §110 (radiation of gravitational waves, with the binary problem).
  3. B. F. Schutz, A First Course in General Relativity, 2nd ed. (Cambridge University Press, 2009): Ch. 9 (gravitational radiation).
  4. J. D. Jackson, Classical Electrodynamics, 3rd ed. (Wiley, 1999): §6.4 (Green’s functions for the wave equation).
  5. R. A. Isaacson, Gravitational radiation in the limit of high frequency. II. Nonlinear terms and the effective stress tensor, Phys. Rev. 166, 1272 (1968).
  6. P. C. Peters & J. Mathews, Gravitational radiation from point masses in a Keplerian orbit, Phys. Rev. 131, 435 (1963).
  7. P. C. Peters, Gravitational radiation and the motion of two point masses, Phys. Rev. 136, B1224 (1964).
  8. J. M. Weisberg & Y. Huang, Relativistic measurements from timing the binary pulsar PSR B1913+16, ApJ 829, 55 (2016); arXiv:1606.02744.
  9. H. Sudou et al., Science 300, 1263 (2003); S. Iguchi et al., ApJL 724, L166 (2010); F. A. Jenet et al., ApJ 606, 799 (2004); J. Cardinal Tremblay et al., ApJL 998, L42 (2026). Full entries on the story page.