How black holes get their kicks:

Radiation recoil in binary black hole mergers

Scott A. Hughes1 , Marc Favata2 , and Daniel E. Holz3
1

Department of Physics and Center for Space Research, Massachussets Institute of

Technology, 77 Massachusetts Avenue, Cambridge, MA 02139
Department of Astronomy, Cornell University, 611 Space Sciences Building, Ithaca,
NY 14853
Center for Cosmological Physics, University of Chicago, Chicago, IL 60637

Abstract. Gravitational waves from the coalescence of binary black holes carry linear momentum, causing center of mass recoil. This radiation rocket has important
implications for systems with escape speeds of order the recoil velocity. We describe
new recoil calculations using high precision black hole perturbation theory to estimate
the magnitude of the recoil for the slow inspiral coalescence phase; coupled with a
cruder calculation for the final plunge, we estimate the total recoil imparted to a
merged black hole. We find that velocities of many tens to a few hundred km/sec can
be achieved fairly easily. The recoil probably never exceeds about 500 km/sec.

It is very well known that gravitational waves (GWs) carry energy and angular momentum from a binary system, causing decay of the binarys orbit and
eventually driving the system to merge into a single object. Although it has been
understood for quite some time (e.g., [1]), it is somewhat less well-appreciated
that these waves can carry linear momentum from the system as well. The center
of mass in this case must recoil in order to enforce global conservation of momentum. If the recoil velocity is comparable to or greater than the escape velocity of
the binarys host structure, there could be important dynamical consequences,
such as ejection of the merged black hole remnant.
The recoil arises because the radiation field generated by a binary is typically
asymmetric. As a helpful cartoon, consider the following argument due to Alan
Wiseman. In an unequal mass binary (Fig. 1), the smaller member, m1 , moves
with a higher speed than the larger member, m2 . It is thus more effective at
forward beaming its wave pattern. This means that there is an instantaneous
net flux of momentum ejected from the system parallel to the velocity of the
smaller body, and a concomitant recoil opposing this.
Over an orbit, the recoil direction continually changes. If the orbit were
perfectly circular, this means that there would be no net interesting effect
the binarys center of mass would run around in a circle, and the net recoil would
sum to zero. However, when GW emission is strong, the orbit is not perfectly
circular: Because of the secular, dissipative evolution of the binarys energy and
angular momentum, the black holes slowly spiral towards one another. Since the
orbit does not close, the recoil does not sum to zero. The recoil accumulates until
the holes merge and settle down to a quiescent state, shutting off the momentum
flux and yielding a net, non-zero kick.

Scott A. Hughes et al.

Pejected
v
1
m
2

X
v
2

center of mass
m1
Precoil

Fig. 1. GW emission from an unequal mass binary. Momentum is ejected parallel to

the smaller bodys velocity (v 1 ). Conservation of momentum requires that the system
recoil in the opposite direction.

This recoil is not a wierd property of GWs it holds for any form of radiation1 . This can be brought out by considering a multipolar decomposition.
Suppose we build a distribution of charges that has a non-zero electric dipole and
quadrupole moment, as in Fig. 2. Suppose further that we spin this charge arrangement about its center point, driving the system to radiate electromagnetic
waves. What does this radiation distribution look like from far away?

2Q

2Q

Fig. 2. Charge distribution with non-zero dipole and quadrupole moment. Spinning
this distribution about its center point produces radiation carrying non-zero linear
momentum due to beating between the dipolar and quadrupolar radiation fields.

The radiations amplitude has two pieces, dipole and quadrupole:

E
= E dip + E quad ,
where
dip
quad
i(t)
E
e
,
E
e2i(t) .

(1)
(2)

Since the intensity I |E|2 , it will contain three pieces:

I = I dip + I quad + I dipquad ,

(3)

where
I dip
1

|E dip |2 constant ;

I quad |E quad |2 constant

Indeed, electromagnetic or neutrino recoil may impact neutron star kicks [2,3].

(4)

Radiation recoil in binary black hole mergers

h
i
quad cos( t) .
I dipquad Re E dip E

(5)

The intensity has a preferred direction, which rotates as the charge distribution
rotates. The energy from the system is instantaneously beamed in a preferred
direction, and so there is a net flux of momentum in that direction as well.
Since the lowest order GWs are quadrupolar, recoil from GW emission must
come (at lowest order) from a beating of the mass quadrupole with mass octupole
and current quadrupole moments. The mass octupole and current quadrupole
vanish for an equal mass binary, demonstrating in accord with our forward
beaming intuition that unequal masses are needed for there to be any recoil.
This also demonstrates that GW recoil must be a very small effect, except perhaps in the very late stages of coalescence the octupole radiation amplitude is
smaller than the quadrupole by a factor of order v/c (where v is orbital speed).
The first careful analysis of recoil in binary systems due to GW emission is
that of Michael Fitchett [5]. Fitchetts analysis described the orbital dynamics
of the binary using Newtonian gravity and only included the lowest radiative
multipoles which contribute to the recoil. His analysis predicted that the recoil
of the merged remnant took the form
f (q)
vF ' 1480 km/sec
fmax

2G(m1 + m2 )/c2
Rterm

(6)

where Rterm is the orbital separation at which GW emission terminates, q =

m1 /m2 is the binarys mass ratio, and f (q) = q 2 (1 q)/(1 + q)5 is a function
whose maximum is at q ' 0.38, and has the limit f (q) ' q 2 for q 1.
Three features of this formula are particularly noteworthy. First, this result
does not depend on total mass only on the mass ratio (bearing in mind
that Rterm scales with total mass M ). Thus, this scaling holds for any binary
black hole merger stellar mass mergers through supermassive mergers. Second,
the overall scale is quite high. Although there is an important dependence on
mass ratio and the termination radius Rterm is somewhat uncertain, Eq. (6)
indicates that kicks of hundreds of km/sec are not difficult to achieve; kicks
>
1000 km/sec are plausible. This is high enough that we might expect black
hole ejection following a merger to be common.
Notice, however, that the recoil becomes very strong when the separation of
the bodies is small. This is a strong hint that we cannot take Eq. (6) at face value
the strong gravity physics neglected by Ref. [5] is likely to be very important.
A few efforts have improved on Fitchetts analysis over the years. Fitchett
and Detweiler [6] first made a strong field analysis, treating the binary as a
Schwarzschild black hole M orbited by a point mass . The orbiting bodys influence can then be studied using black hole perturbation theory. Their results
suggested that Eq. (6) describes the recoil fairly well. Wiseman [4] analysed
the
p
recoil with post-Newtonian theory (roughly, an expansion in v GM/rc2 ).
He found that Fitchetts formula tended to systematically overestimate the recoil by >
10%. Unfortunately, his results behave somewhat pathologically for
2
9GM/c
due to ill behavior of the expansion in the very strong field.
r<

Scott A. Hughes et al.

This motivates our analysis of this problem. Our formal setup is quite similar
to that of Fitchett and Detweiler: We model the binary as a pointlike body
moving on the exact, geodesic orbits of a Kerr black hole with mass M and spin
parameter a. We compute the GWs emitted from such orbits very accurately
using black hole perturbation theory [79], and extract the recoil from the wave
pattern [10]. The perturbative approach allows us to study the dynamics of the
binarys spacetime with high accuracy. Schematically, we treat this spacetime as
that of the Kerr black hole plus a small perturbation:
Kerr
g = g
(M, a) + h () .

(7)

By requiring that this spacetime satisfy the Einstein field equations, it can be
shown [7] that h is governed by a wave-like equation. The wave operator
automatically captures the most important properties of the strong field physics.
This approach is strictly accurate only when M the smaller body
cannot significantly distort the spacetime if we want the exact Kerr orbits to describe our binary. We believe, though, that we can extrapolate out of this regime
with accuracy good enough for most astrophysical purposes our extrapolation
errors are estimated to be at most several tens of percent, at least up to a mass
ratio q ' 0.3, and to scale with the squared mass ratio.
We focus upon circular, equatorial orbits of Kerr black holes. Circularity is
surely a good approximation, since eccentricity is rapidly reduced during coalescence. The equatorial assumption is not so good; since the binaries of interest
form through captures, we expect no particular alignment between the spin and
orbit. We are working to lift this approximation, which requires moderately substantial modifications to our code; early indications are that the inclination does
not have a very large effect on the recoil, other than to change the radius at
which a transition in the orbital dynamics occurs. Results from the prograde
and retrograde equatorial orbits appear to bound the recoil at any spin.
One technical detail of our code is important enough that it requires some
explanation. Our calculation expands the perturbation as
4 =

1X
Zlm Slm ()eim(t) .
r

(8)

lm

4 represents a curvature perturbation, and serves as a surrogate for h . Slm ()

is a spheroidal harmonic (very similar to a spherical harmonic), and Zlm is a
complex number found by solving a certain differential equation [7,8]. The most
important aspect of this equation for our purposes is the presence of the orbital
frequency : We assume that the orbit is well-described by a Fourier expansion.
This is only true if the orbit is periodic (or nearly so). This in turn is only true
when the separation r of the black holes is greater than the radius of the last
stable orbit (LSO), rLSO (a). (For Schwarzschild holes, a = 0, rLSO = 6GM/c2 .)
In the regime r > rLSO (a), dynamically stable orbits exist. GWs slowly evolve
the system through a sequence of stable orbits of ever decreasing radius. At any
instant in this regime, the orbit is well modeled as periodic with a well-defined

Radiation recoil in binary black hole mergers

Fig. 3. Left: Fitchetts accumulated recoil and our results, versus binary separation
r. This illustrates the importance of a proper treatment of strong-field physics. Right:
Total recoil versus black hole spin a. The upper curve is our overestimate, the lower
our underestimate. The span between them shows the importance of the final plunge.

frequency . The recoil is then simply related to the coefficients Zlm and to an
overlap integral involving the harmonics Slm (); see Ref. [10] for details.
As we approach and cross rLSO (a), the orbit becomes a rapid plunge in
which the small body quickly falls into the event horizon. This final plunge cannot be treated using this Fourier expansion, so our calculation breaks down.
We are currently developing techniques to model this regime accurately. For the
present analysis, we compute a probable overestimate (based on an extrapolation
of the inspiral momentum flux beyond its range of validity) and a probable underestimate (based on a low-order flux formula coupled to the plunging motion)
in order to provide a reasonable range for the likely recoil.
Our analysis shows that Fitchetts calculation consistently overestimates the
recoil velocity, especially as the LSO is approached; see the left-hand panel of Fig.
3. This is due to strong-field physics: A wave packet released near the horizon
redshifts as it propagates to large radius, reducing the energy and momentum
that it carries. This effect is not present in calculations which neglect curved
spacetime physics, as in Ref. [5]. Also, when radiation propagates through a
curved spacetime, the anisotropy of the radiation pattern tends to be somewhat
reduced due to the phenomenon of tails essentially, the backscatter of the radiation from spacetime curvature itself. Both the redshifting and the anisotropy
reduction reduce the recoil relative to Fitchetts original analysis.
The right-hand panel of Fig. 3 summarizes our results for the total recoil
that can be expected in a merger; the curves shown are for q = 0.12, but can be
rescaled using f (q) [Eq. (6) and subsequent text]. The range shown here shows
the uncertainty that results from our inability to model the plunge very well. It
is largest for retrograde orbits (a < 0) of rapidly rotating black holes because the
transition to plunge occurs at relatively large radius there the smaller body
plunges quite a distance before passing through the event horizon. By contrast,

Scott A. Hughes et al.

for prograde orbits (a > 0), the transition occurs at small radius, so the plunge
does not matter quite so much. Our uncertainty is much smaller for those cases.
A very notable feature of this plot is that the recoil, though substantial, never
exceeds about 500 km/sec, even when the mass ratio is tuned to maximize the
recoil. On the other hand, it is not difficult for the recoil to reach several tens
of km/sec, even for our underestimate. Indeed, when we convolve our over- and
underestimates with a distribution of likely mass ratios and spin values (cf. Ref.
[11]) we find that recoils of several tens of km/sec are quite easy to achieve;
recoils 100 km/sec are likely for an interesting fraction of recoils; and recoils
of several hundred km/sec, though possible, are probably rather rare.
In the near future we hope to reduce our error bars, but for now we understand the recoil with sufficient accuracy that these results can be used for
many astrophysical applications The most likely recoil range several tens to
a few hundred km/sec is particularly interesting: While not large enough to
eject black holes from massive galaxies, kicks in this range can lead to ejection
in dwarf galaxies and dark matter halos, affect the nuclear density profiles of
galaxies with SMBHS, and influence the hierarchical growth of supermassive
and intermediate mass black holes [1117].
It is a pleasure to thank Saul Teukolsky and Jerry Ostriker for bringing this

problem to our attention, and Eanna

Flanagan, Avi Loeb, David Merritt, Milos
Milosavljevic, Martin Rees, Joseph Silk, Alan Wiseman, and Yanqin Wu for
many useful discussions. SAH would especially like to thank the organizers of
this conference for a fantastic meeting. MF is supported by NSF grant PHY0140209; SAH by NSF grant PHY-0244424 and NASA grant NAG5-12906; and
DEH by NSF grant PHY-0114422.

References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.

J. D. Bekenstein: Astrophys. J. 183, 657 (1973).

E. R. Harrison and E. Tademaru: Astrophys. J. 201, 447 (1975).
D. Lai, D. F. Chernoff and J. M. Cordes: Astrophys. J. 549, 1111 (2001).
A. G. Wiseman: Phys. Ref. D 46, 1517 (1992).
M. J. Fitchett: Mon. Not. R. Astron. Soc. 203, 1049 (1983).
M. J. Fitchett and S. D. Detweiler: Mon. Not. R. Astron. Soc. 211, 933 (1984).
S. A. Teukolsky: Astrophys. J. 185, 635 (1973).
S. A. Hughes: Phys. Rev. D 61, 084004 (2000).
S. A. Hughes: Phys. Rev. D 64, 064004 (2001).
M. Favata, S. A. Hughes, and D. E. Holz: Astrophys. J. Lett. 607, L5 (2004).
D. Merritt, M. Milosavljevic, M. Favata, S. A. Hughes, and D. E. Holz: Astrophys.
J. Lett. 607, L9 (2004).
P. Madau and E. Quataert: Astrophys. J. Lett. 606, L20 (2004).
M. Volonteri, F. Haardt, P. Madau: Astrophys. J. 582, 559 (2003).
K. Gultekin, M. C. Miller, and D. P. Hamilton, Astrophys. J., submitted; astroph/0402532.
Z. Haiman: Astrophys. J., submitted; astro-ph/0404196.
J. Yoo and J. Miralda-Escude: Astrophys. J. Lett., submitted; astro-ph/0406217.
M. Boylan-Kolchin, C.-P. Ma, and E. Quataert: Astrophys. J. Lett., submitted;
astro-ph/0407488.