The Last Three Minutes: Issues in Gravitational Wave Measurements

of Coalescing Compact Binaries

Curt Cutler,(1) Theocharis A. Apostolatos,(1) Lars Bildsten,(1) Lee Samuel

Finn,(2) Eanna E. Flanagan,(1) Daniel Kennefick,(1) Dragoljubov M. Markovic,(1)
Amos Ori,(1) Eric Poisson,(1) Gerald Jay Sussman,(1),∗ and Kip S. Thorne(1)
(1)
Theoretical Astrophysics, California Institute of Technology, Pasadena, CA 91125
(2)
Department of Physics and Astronomy, Northwestern University, Evanston, IL 60208

Gravitational-wave interferometers are expected to monitor the last three minutes of inspiral
arXiv:astro-ph/9208005v1 25 Aug 1992

and final coalescence of neutron star and black hole binaries at distances approaching cosmological,
where the event rate may be many per year. Because the binary’s accumulated orbital phase can be
measured to a fractional accuracy ≪ 10−3 and relativistic effects are large, the waveforms will be
far more complex, carry more information, and be far harder to model theoretically than has been
expected. Theorists must begin now to lay a foundation for extracting the waves’ information.
PACS numbers: 04.30.+x, 04.80.+z, 97.60.Jd, 97.60.Lf

A network of gravitational-wave interferometers (the

American LIGO [1], the French/Italian VIRGO [2] and
possibly others) is expected to be operating by the end
of the 1990s. The most promising waves for this network
are from the inspiral and coalescence of neutron star and
black hole binaries [3,4], with an estimated event rate of
∼ (3/year)(distance/200 Mpc)3 [5].
A binary’s inspiral and coalescence will produce two
gravitational waveforms, one for each polarization. By
cross-correlating the outputs of 3 or more interferome-
ters, both waveforms can be monitored and the source’s
direction can be determined to within ∼ 1 degree [4,6].
We shall divide each waveform into two parts: the in-
spiral waveform, emitted before tidal distortions become
noticeable (3 or fewer orbital cycles before complete dis-
ruption or merger [7]), and the coalescence waveform,
emitted during distortion, disruption, and/or merger.
As the binary, driven by gravitational radiation reac-
tion, spirals together, its inspiral waveform sweeps up-
ward in frequency f . The interferometers will observe
the last several thousand cycles of inspiral (from f ∼ 10
Hz to ∼ 1000 Hz), followed by coalescence.
Theoretical calculations of the waveforms are generally
made using the Post-Newtonian (“PN”) approximation
to general relativity. Previous calculations have focused
on the Newtonian-order waveforms [4,3,8,1] and on PN
modulations of their amplitude and frequency [9].
Two of us [11] recently realized that the PN modula-
tions are far less important than PN contributions
R to the
secular growth of the waves’ phase Φ = 2π f dt (which
arise, largely, from PN corrections to radiation reaction.)
The binary’s parameters are determined by integrating
the observed (noisy) signal against theoretical templates
[10], and if the signal and template lose phase with each
other by one cycle (out of thousands) as the signal sweeps
through the interferometers’ band, their overlap integral

1
will be strongly reduced. Correspondingly, one can infer
each of the system’s parameters λi to an accuracy that is
roughly the change ∆λi which alters by unity the number
of cycles Ncyc spent in the interferometers’ band.
The only parameters λi that can influence the inspiral
template significantly are (i) the initial orbital elements,
(ii) the bodies’ masses and spin angular momenta, and
(iii) if the spins are very large, the bodies’ spin-induced
quadrupole moments. (The quadrupole moments pro-
duce orbital phase shifts a little larger than unity at
f ∼ 10 Hz, but negligible phase shifts later [7]. Below we
shall ignore this tiny effect.)
If (as almost always is the case) the binary’s orbit has
been circularized by radiation reaction, then the num-
ber of cycles spent in a logarithmic interval of frequency,
dNcyc /d ln f = (1/2π)(dΦ/d ln f ), is:
  
dNcyc 5 1 743 11 µ
= 1 + + x
d ln f 96π µM 2/3 (πf )5/3 336 4 M
− [4π + S.O.]x1.5 + [S.S.]x2 + O(x2.5 ) . (0.1)

Here M is the binary’s total mass, µ its reduced mass,

and x ≡ (πM f )2/3 ≃ M/D the PN expansion parameter
(with D the bodies’ separation and c = G = 1). The PN
correction [O(x) term] is from [12]. In the P1.5 N correc-
tion, the 4π is created by the waves’ interaction with the
binary’s monopolar gravitational field as they propagate
from the near zone to the radiation zone [13], and the
“S.O.” denotes contributions due to spin-orbit coupling
[14]. In the P2 N correction the “S.S.” includes spin-spin
coupling effects [14] plus an expression quadratic in µ/M .
Since the leading-order, Newtonian contribution to
Eq. (0.1) gives several thousand cycles of oscillation in
the interferometers’ band, to be measureable the higher-
order corrections need only be as large as a part in several
thousand, even when the amplitude signal to noise ratio
has only its typical value of 8. It is this that gives the
waves’ phasing its high potential accuracy.
The determination of the binary’s parameters (masses,
spins, orbital elements) is made possible by the various
frequency dependences in Eq. (0.1). Analytic [15,16] and
Monte Carlo [17] calculations show that (i) the “chirp
mass” Mc ≡ µ3/5 M 2/5 [which governs the Newtonian
part of (0.1)] will typically be measured to 0.1 per cent,
and (ii) if we somehow knew that the spins were small,
then the reduced mass µ would be measured to ∼ 1 per
cent for NS/NS and NS/BH binaries, and ∼ 3 per cent for
BH/BH binaries. (Here and below NS means a ∼ 1.4M⊙
neutron star and BH means a ∼ 10M⊙ black hole.) Un-
fortunately, the frequency (x) dependences of the various
terms in Eq. (0.1) are not sufficiently different to give a
clean separation between µ and the spins. Preliminary
estimates [17], in which the S.O. term in Eq. (0.1) was
taken into account but not the S.S. term, suggest that
the spin/µ correlation will worsen the typical accuracy
of µ by a factor ∼ 20, to ∼ 20 per cent for NS/NS,
∼ 25 per cent for NS/BH, and ∼ 50 per cent for BH/BH.

2
These worsened accuracies might be improved signifi-
cantly, however, by the waveform modulations [9]—most
especially, modulations caused by spin-induced preces-
sion of the orbit [14]. Much additional theoretical work
is needed to pin down the measurement accuracies for µ
and the spins.
The highest accuracy parameter information will come
from low frequencies where most of the waves’ cycles
are located [15–17]. (It thus is important that the in-
terferometers achieve good low-frequency performance).
When the waves have swept to higher frequencies, the
binary’s parameters may be fairly well known, and the
subsequent, highly relativistic waveforms might be used
to map out the binary’s innermost spacetime geometry
near the orbital plane. It would be useful to develop
a parametrized, “PPN-like” multi-theory formalism for
translating the measured waveforms into spacetime maps
and for comparing those maps with the predictions of
various gravitation theories.
To make optimal use of the interferometers’ data
will require general-relativity-based waveform templates
whose phasing is correct to within a half cycle or so dur-
ing the entire frequency sweep from ∼ 10 Hz to ∼ 1000
Hz. By examining an idealized limit (µ ≪ M and vanish-
ing spins), several of us [13,19] have discovered that, to
compute the templates with the desired accuracy via PN
methods will be very difficult. We calculated the waves
from such a binary to high accuracy using the Teukolsky
[18] black-hole perturbation formalism and then fit a PN
expansion to the results, to obtain [19]

dNcyc f 2 dE/df 5 M/µ

= = ×
d ln f dE/dt 96π x2.5
1 − 23 x − 81 2 675 3
8 x − 16 x + ...
× 1247 . (0.2)
1 − 336 x + 4πx1.5 − 4.9x2 − 38x2.5 + 170x3 + ...

Here dE/df is half the ratio of the changes of orbital en-

ergy dE and orbital frequency d(f /2) as the orbit shrinks,
and dE/dt is the power carried off by gravitational waves.
The coefficients in the denominator after the 4π were ob-
tained from the fit; all other coefficients are known ana-
lytically. The coefficient 4.9 has accuracy ∼ 2% ; the 38,
∼ 10% ; and the 170, ∼ 25%.
We can use Eq. (0.2) to estimate how the phase ac-
curacy depends on the order to which the PN expansion
is carried. If the template were computed through P2 N
order [all x2.5 and x3 terms in Eq. (0.2) omitted], then
the phase error typically would reach a half cycle when
the frequency had swept from 10 Hz to only ∼ 15 Hz,
and the total phase error from 10 Hz to 1000 Hz would
typically be ∼ 6 cycles. For interferometers whose peak
sensitivity is at 70 Hz [1], one could use such templates
for a maximum range of ∼ 50 Hz to ∼ 90 Hz without
accumulating phase errors larger than a half cycle. If
the template were computed through P2.5 N order, there
would be no substantial improvement. It is not at all
clear how far beyond P2.5 N the template must be carried

3
to keep the total phase error below a half cycle over the
entire range from ∼ 10 Hz to ∼ 1000 Hz.
This slow convergence of the PN expansion might be
improved by cleverness in the way one expresses the ex-
pansion (e.g., by the use of Padé approximates). How-
ever, even with great cleverness and fortitude, PN tem-
plates might never cover the entire inspiral range, from
∼ 10 Hz to ∼ 1000 Hz. Evidently, new techniques are
needed for computing templates to higher accuracy. Two
such techniques look promising: A “post-Teukolsky ex-
pansion,” and a “weak-reaction expansion.”
The post-Teukolsky expansion would expand in powers
of µ/M and thus would be useable only for a light body
orbiting a much heavier black hole. The expansion’s first
step would be the unperturbed hole’s Kerr metric; the
second, the Teukolsky formalism [18] for the light body’s
first-order perturbations; and the third, incorporation of
radiation reaction [20]. Each of these first three steps is
now in hand, though studies of the consequences of the
radiation reaction are only beginning [21]. That the ex-
pansion must be carried well beyond these three steps is
evident from the magnitude of the µ/M term in Eq. (0.1),
and from recent studies [22] of the influence of a finite
µ/M on the last stable circular orbit.
The weak-reaction expansion would be a variant of
numerical relativity in which one expands in powers of
1/Q ≡ (dE/dt)(πf E)−1 (a measure of the effects of ra-
diation reaction during one orbit). Because 1/Q ∼ x2.5
(with x the PN parameter) is always ≪ 1, this expan-
sion might produce adequate inspiral templates for all
binaries. The expansion’s first step might be a numerical
solution of the Einstein equations for the binary’s met-
ric, with standing-wave boundary conditions at infinity
and at the BH horizon(s) (if any) [23]. In co-rotating
coordinates, the equations would be elliptic and might
be solvable to high accuracy by relaxation techniques.
The second step might be to switch from standing-wave
to outgoing- and downgoing-wave boundary conditions,
and evaluate the resulting linearized metric perturba-
tions, and with them the leading contributions to the
waveforms. The third step might be the leading effects
of radiation reaction, including the orbital inspiral. To
obtain templates with the desired phase accuracy, one
would have to carry the expansion beyond this third step.
Although an optimally-precise measurement of the bi-
nary’s parameters requires templates faithful to general
relativity’s predictions, a near optimal search for binary-
inspiral waves can use a family of templates that merely
span the range of expected waveform behaviors, without
being closely related to the true general relativistic wave-
forms [24]. Theorists need to develop a set of such search
templates and optimize their efficiency for wave searches.
Turn, now, from a binary’s inspiral waveforms to its
coalescence waveforms. By the beginning of coalescence,
the binary’s parameters (masses, spins, geometry) will
be known with fair accuracy, and from its masses the
nature of its bodies (BH vs NS) should be fairly clear.
The coalescence waveforms can then be used to probe

4
issues in the physics of gravity and atomic nuclei:
If the bodies are BH’s of comparable mass (and espe-
cially if their spins are large and not perpendicular to the
orbital plane), then the coalescence should produce large-
amplitude, highly nonlinear vibrations of spacetime cur-
vature that may reveal aspects of gravity we have never
seen. Although supercomputer simulations of such coa-
lescences are being attempted, they are so difficult, es-
pecially for spinning holes, that they may well not give
definitive results without observational guidance. Con-
versely, the observed waveforms will be very hard to in-
terpret without the guidance of simulations.
The interferometer measurements discussed above may
well be achieved in the early years of LIGO/VIRGO.
Most of the measurements discussed below are more dif-
ficult, and may require more mature interferometers.
For a NS/BH binary with the BH spinning moderately
fast, the NS should disrupt tidally before plunging into
the BH. The NS disruption should be quick, as should the
final coalescence of a NS/NS binary: within about one
orbit, the NS/BH and NS/NS binaries may get smeared
into roughly axially symmetric configurations, thereby
shutting off their waves [25,7]. The rapid wave shut-
off and contamination by laser shot noise may prevent
the coalescence waveforms from revealing more than two
numbers: the maximum frequency fmax reached, and the
total wave energy EGW in the last ∼ 0.02 seconds (when
the energy emission is maximal). Either number, how-
ever, would be valuable. It could tell us the NS radius
RNS (or a combination of the two NS radii) [27]. Since the
NS masses will already be known from the inspiral wave-
forms, such measurements on a number of NS’s could re-
veal the NS radius-mass relation RNS (MNS ), from which
one can infer the equation of state of matter at densities
from ∼ 1 to 10 times nuclear [26].
In preparation for the interferometers’ measurements
of fmax and/or EGW , theorists need to model tidal dis-
ruption in NS/BH binaries and coalescence of NS/NS
binaries, to determine the dependence of fmax and EGW
on RNS and the binary’s other parameters. (Such model-
ing is also driven by the possibility that these events are
the sources of observed gamma-ray bursts [28].) Sim-
ple arguments [7,27] suggest that fmax will be roughly
3 1/2
(1/2π)(MNS /RNS ) , so for a 1.4M⊙ NS, a 15 km ra-
dius will lead to fmax ∼ 1000 Hz, while a 10 km radius
will lead to fmax ∼ 2000 Hz. Similarly, EGW will be pro-
portional to 1/RNS so a 10 km NS will emit ∼ 50 per
cent more energy just before disruption than will a 15
km NS.
To measure the shutoff frequency fmax ∼ 1000 Hz in
the midst of the interferometers’ shot noise will probably
require specially configured (“dual recycled” [29]) inter-
ferometers that operate over adjustable, narrow bands
∆f around adjustable central frequencies fo . Such inter-
ferometers, operated in concert with broad-band ones,
could give simple “yes/no” answers as to whether the
waves swept through their frequency bands. By col-
lecting such data on a number of binaries, with various

5
choices for the frequencies fo , one might zero in on fmax
for various types of binaries, and thence on the NS radius-
mass relation. The dual-recycled sensitivities required for
such a program, however, will be difficult to achieve [27].
Even more difficult, but not totally hopeless, will be
the measurement of EGW via the waves’ “Christodoulou
memory” [30,27] (a slowly changing mean of the wave-
forms, produced by the 1/r gravitational field of the emit-
ted gravitons). When the waves’ mean is monitored by
an optimally filtered, broad-band interferometer, it gets
averaged over a time ∼ 0.02 seconds, so the maximum
level it reaches is proportional to EGW . Unfortunately,
to measure the memory’s level and thence EGW with
even 50 per cent accuracy for binaries at 200 Mpc dis-
tance is likely to require interferometers modestly better
than LIGO’s so-called “advanced detectors” [1].
Turn, next, from a binary’s coalescence to the use of its
waveforms as cosmological probes. Schutz [4] has pointed
out that a binary’s distance from Earth (more precisely,
its “luminosity distance” rL ) can be inferred from its in-
spiral waveforms, and that such gravitationally measured
distances, when combined with electromagnetically mea-
sured redshifts z, might determine the Hubble constant
Ho to better precision than heretofore. However, this
will require identifying electromagnetically the clusters
of galaxies in which the binaries lie, at least in some sta-
tistical sense—a difficult task [4].
Remarkably, for NS/BH binaries at z > ∼ 1, it may be
possible to determine z as well as rL from the gravita-
tional waveforms, and then from rL (z) one might infer
the Universe’s Hubble constant Ho , mean density ρ̄, and
cosmological constant Λ [31]. (A variant of the method
could also work for NS/NS binaries, but NS/BH bina-
ries emit more strongly and thus can be seen to greater
distances, making them more promising than NS/NS.)
The key to measuring z is the expectation (based on
radio pulsar observations) that NS masses will cluster
around 1.4M⊙. (If this fails, but there is a sharp cut-
off in NS masses at some upper limit, then that limiting
mass can be used as the key to the measurements.) For a
binary at z >∼ 1, the waveforms will be measured as func-
tions of redshifted time, and correspondingly they will
reveal redshifted masses (1 + z)MNS and (1 + z)MBH .
From the measured luminosity distance and redshifted
masses, it should be fairly clear whether the binary is
NS/BH or not; and if so, then one can infer a “candidate
redshift” for the binary, 1 + zcand ≡ (1 + z)MNS /1.4M⊙.
If the NS masses cluster around 1.4M⊙ , then a large sam-
ple of measured distances and candidate redshifts should
cluster around the true distance-redshift relation.
This gravitational method of determining cosmological
parameters might be much more immune to evolution-
ary and ill-understood systematic effects than are elec-
tromagnetic methods, because gravitational waves are
immune to absorption and scattering, and neutron star
masses might not be sensitive to the Universe’s evolution.
A detailed analysis by one of us [31] suggests that
this gravitational approach to cosmology might begin to

6
bring useful information when broad-band interferome-
ters reach the LIGO “advanced detector” sensitivities
[1]: with one year of observational data, the one-sigma
accuracies might be δHo ∼ 0.01Ho , δ ρ̄ ∼ 0.1ρcrit, and
δΛ ∼ 0.3Ho2 , where ρcrit is the critical density to close
the Universe.
We conclude by noting that some of the issues dis-
cussed in this Letter have implications for a possible
future space-based interferometer called LAGOS, which
would operate in the band 0.0001–0.03 Hz [32]. We shall
discuss those implications elsewhere.
For helpful discussions we thank David Chernoff, Larry
Kidder, Andrzej Krolak, E. Sterl Phinney, Bernard F.
Schutz, Clifford M. Will, and Alan Wiseman. This re-
search was supported in part by National Science Foun-
dation Grant PHY-9213508, and, in view of its ap-
plications to LAGOS, by NASA grants NAGW-2897,
2920, and 2936, and by a Weingart Foundation, Lee
A. Dubridge Fellowship to L. Bildsten and an Alfred P.
Sloan Foundation Fellowship to L. S. Finn.

∗
Permanent address: Department of Electrical Engineer-
ing, Massachusetts Institute of Technology, Cambridge,
MA.
[1] A. Abramovici et al., Science 256, 325 (1992).
[2] C. Bradaschia et. al., Nucl. Instrum. & Methods A289,
518 (1990).
[3] K. S. Thorne, in 300 Years of Gravitation, eds. S. W.
Hawking and W. Israel (Cambridge University Press,
Cambridge, 1987), p. 330.
[4] B. F. Schutz, Nature 323, 310 (1986); B. F. Schutz,
Class. Quantum Gravity 6, 1761 (1989).
[5] R. Narayan, T. Piran, and A. Shemi, Astrophys. J. 379,
L17 (1991); E. S. Phinney Astrophys. J. 380, L17 (1991).
[6] Y. Gursel and M. Tinto, Phys. Rev. D 40, 3884 (1990);
P. Jaranowski and A. Krolak (unpublished).
[7] L. Bildsten and C. Cutler, Astrophys. J., in press. See
also C. Kochanek, Astrophys. J., in press.
[8] H. D. Wahlquist, Gen. Rel. Gravit. 19, 1101 (1987).
[9] C. W. Lincoln and C. M. Will, Phys. Rev. D 42, 1123
(1990).
[10] See, e.g., L. S. Finn, Phys. Rev. D, submitted.
[11] C. Cutler and L. S. Finn, independent unpublished work.
[12] R. V. Wagoner and C. M. Will, Astrophys. J 210, 764
(1976), and 215, 984 (1977).
[13] E. Poisson, Phys. Rev. D, submitted.
[14] G. J. Sussman, K. S. Thorne, and T. A. Apostolatos (un-
published); L. Kidder and A. G. Wiseman (unpublished).
[15] C. Cutler and E. E. Flanagan (unpublished).
[16] D. Chernoff and L. S. Finn (unpublished).
[17] C. Cutler (unpublished).
[18] S. A. Teukolsky, Astrophys. J. 185, 635 (1973).
[19] C. Cutler, L. S. Finn, E. Poisson and G. J. Sussman,
Phys. Rev. D, submitted.

7
[20] D. V. Gal’tsov, J. Phys. A 15, 3737 (1982).
[21] Two of us (A. Ori and D. Kennefick) have recently proved
that “quasi-circular” orbits (orbits of constant Boyer-
Lindquist radial coordinate r) remain quasicircular as
they shrink under the influence of radiation reaction.
[22] L. Kidder, C. M. Will, and A. G. Wiseman, Phys. Rev.
D, submitted.
[23] S. L. Detweiler and K. Blackburn (unpublished) have de-
veloped an action principle for this first step for an equal-
mass black-hole binary.
[24] C. Cutler, E. E. Flanagan, and G. J. Sussman (unpub-
lished).
[25] F. A. Rasio and S. L. Shapiro, Astrophys. J., in press.
[26] L. Lindblom, Astrophys. J., in press.
[27] E. E. Flanagan, D. Kennefick, D. M. Markovic and K. S.
Thorne (unpublished).
[28] B. Paczynski, Astrophys. J. Lett. 308, L43 (1986), and
Acta Astronomica 41, 257 (1991). C. A. Meegan et. al.,
Nature 355, 143 (1992).
[29] B. J. Meers, Phys. Rev. D 38, 2317 (1988); K. A. Strain
and B. J. Meers, Phys. Rev. Lett. 66, 1391 (1991); A.
Krolak, J. A. Lobo, and B. J. Meers Phys. Rev. D 43,
2470 (1991).
[30] D. Christodoulou, Phys. Rev. Lett. 67, 1486 (1991); K. S.
Thorne, Phys. Rev. D 45, 520 (1992); C. M. Will and A.
Wiseman, Phys. Rev. D 44, 2945 (1991).
[31] D. M. Markovic, paper in preparation. That cosmologi-
cal measurements might be carried out by gravitational-
wave detectors, without the aid of electromagnetic obser-
vations, has been suggested previously by Schutz [4].
[32] D. Hils, P. L. Bender, and R. F. Webbink, Astrophys. J.
360, 75 (1990).