You are on page 1of 59

Progress of Theoretical Physics Supplement 1

Post-Newtonian Approximation
Its Foundation and Applications
Hideki Asada and Toshifumi Futamase
 Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-01
Astronomical Institute, Graduate School of Science, Tohoku University
Sendai 980-77, Japan
We discuss various aspects of the post-Newtonian approximation in general relativity.
After presenting the foundation based on the Newtonian limit, we use the (3+1) formalism to
formulate the post-Newtonian approximation for the perfect uid. As an application we show
the method for constructing the equilibrium con guration of nonaxisymmetric uniformly
rotating uid. We also discuss the gravitational waves including tail from post-Newtonian
systems.

x1. Introduction
The motion and associated emission of gravitational waves (GW) of self gravi-
tating systems have been a main research interest in general relativity. The problem
is complicated conceptually as well as mathematically because of the nonlinearity
of Einstein's equations. There is no hope in any foreseeable future to have exact
solutions describing motions of arbitrary shaped, massive bodies so that we have to
adopt some sort of approximation scheme for solving Einstein's equations to study
such problems. In the past years many types of approximation schemes have been
developed depending on the nature of the system under consideration. Here we shall
focus on a particular scheme called the post-Newtonian (PN) approximation. There
are many systems in astrophysics where Newtonian gravity is dominant, but general
relativistic gravity plays also important roles in their evolution. For such systems it
would be nice to have an approximation scheme which gives Newtonian description
in the lowest order and general relativistic e ects as higher order perturbations. The
post-Newtonian approximation is perfectly suited for this purpose. Historically Ein-
stein computed rst the post-Newtonian e ects, the precession of the perihelion 36),
but a systematic study of the post-Newtonian approximation was not made until the
series of papers by Chandrasekhar and associates 21) - 25) . Now it is widely known that
the post-Newtonian approximation is important in analyzing a number of relativistic
problems, such as the equation of motion of binary pulsar 19); 40); 53); 31) , solar-system
tests of general relativity 97); 98) , and gravitational radiation reaction 25); 20) .
Any approximation scheme nesseciates a small parameter(s) characterizing the
nature of the system under consideration. Typical parameter which most of schemes
adopt is the magnitude of the metric deviation away from a certain background
metric. In particular if the background is Minkowski spacetime and there is no other
parameter, the scheme is sometimes called as the post-Minkowskian approximation
in the sense that the constructed spacetime reduces to Minkowski spacetime in the

typeset using PT P TEX.sty <ver.0.8 >


2 H. Asada and T. Futamase
limit as the parameter tends to zero. This limit is called as the weak eld limit.
In the case of the post-Newtonian approximation the background spacetime is also
Minkowski spacetime, but there is another small parameter, that is, the typical
velocity of the system divided by the speed of light which we call  henceforth.
These two parameters (the deviation away from the at metric and the velocity)
have to have a certain relation in the following sense. As the gravitational eld
gets weaker, all velocities and forces characteristic of the material systems become
smaller, in order to permit the weakening gravity to remain an important e ect on
the system's dynamics. For example in case of a binary system, the typical velocity
would be the orbital velocity v=c   and the deviation from the at metric would
be the Newtonian potential, say . Then these are related by =c2  v2 =c2  2
which guarantees that the system is bounded by its own gravity.
In the post-Newtonian approximation, the equations in general relativity take
the form of Newton's equations in an appropriate limit as  ! 0. Such a limit is called
as the Newtonian limit and it will be the bases of constructing the post-Newtonian
approximation. However the limit is not in any sense trivial since the limit may be
thought of two limits tied together as just described. It is also worth noting that
the Newtonian limit cannot be uniform everywhere for all time. For example any
compact binary systems, no matter how weak the gravity between components and
slow its orbital motion, will eventually spiral together due to backreaction of the
emission of gravitational waves. As the result the e ects of its Newtonian gravity
will be swamped by those of its gravitational waves. This will mean that higher
order e ect of the post-Newtonian approximation eventually dominates the lowest
order Newtonian dynamics and thus if the post-Newtonian approximation is not
carefully constructed, this e ect can lead to many formal problems, such as divergent
integrals 35) . It has been shown that such divergences may be avoided by carefully
de ning Newtonian limit 44). Moreover, the use of such limit provides us a strong
indication that the post-Newtonian hierarchy is an asymptotic approximation to
general relativity 47) . Therefore we shall rst discuss in this paper the Newtonian
limit and how to construct the post-Newtonian hierarchy before attacking practical
problems in later sections.
Before going into the details, we mention the reason for the growth of interest
in the post-Newtonian approximation in recent years. Certainly the discovery of
the binary neutron star system PSR 1913+16 was a strong reason to have renewed
interest in the post-Newtonian approximation since it is the rst system in which
general relativitistic gravity plays fundamental roles in its evolution 53) . Particularly
indirect discovery of gravitational wave by the observation of the period shortening
led to many fruitful studies of the equation of motion with gravitational radiation
emission in 1980's 27); 82); 93) . The e ect of radiation reaction appears as the form of
the potential force at the order of 5 higher than Newtonian force in the equation
of motion. There have been various attempts to show the validity of the so-called
quadrupole formula for the radiation reaction 2); 44); 54) - 58); 81); 94); 95) .
At that time, however, no serious attempts had been made for the study of
higher order e ects in the equation of motion. The situation changed gradually in
1990's because of the increasing expectation of direct detection of gravitational waves
Post-Newtonian Approximation 3
by Kilometer-size interferometric gravitational wave detectors, such as LIGO 1) ,
VIRGO 18) and TAMA 61) now under construction. Coalescing binary neutron stars
are the most promising candidate of sources of gravitational waves for such detec-
tors. The reasons are that (1) we expect to detect the signal of coalescence of binary
neutron stars about several times per year 76) , and (2) the waveform from coalescing
binaries can be predicted with a high accuracy compared with other sources 1); 91); 98) .
Informations carried by gravitational waves tell us not only various physical param-
eters of neutron star 90); 91) , but also the cosmological parameters 83); 66); 42); 96) if and
only if we can make a detailed comparison between the observed signal with theoret-
ical prediction during the epoch of the so-called inspiraling phase where the orbital
separation is much larger than the radius of component stars 26) . This is the place
where the post-Newtonian approximation may be applied to make a theoretical tem-
plate for gravitational waves. The problem is that in order to make any meaningful
comparison between theory and observation we need to know the detailed waveforms
generated by the motion up to 4PN order which is of order 8 higher than the Newto-
nian order 88) . This request from gravitational wave astronomy forces us to construct
higher order post-Newtonian approximation.
It should be mentioned that Blanchet and Damour have developed a system-
atic scheme to calculate the waveform at higher order where the post-Minkowskian
approximation is used to construct the external eld and the post-Newtonian ap-
proximation is used to construct the eld near the material source 9) - 13) . Blanchet
has obtained the waveform up to the 2.5 PN order which is of order 5 higher than
the lowest qudraupole wave 14); 7); 8) by using the equation of motion up to that or-
der 28) - 30); 50); 60) . We shall also study the waveform from di erent point of view
because of its importance. We obtain essentially the same results including the tail
term with them.
There is another phase in the evolution of the binary neutron star which we can
withdraw useful informations from the observed signals of gravitational waves. That
is the intermediate phase between the inspiralling phase and the event of merging.
In the inspiralling phase the components are usually treated as point particles, while
the extendedness becomes important in the intermediate phase and the merging.
Full general relativistic hydrodynamic simulation is needed for the understanding of
the merging. The material initial data for the simulation will be the density and the
velocity distribution of the uid. The post-Newtonian approximation will be used to
construct the data. Furthermore the post-Newtonian approximation is able to take
into account of the tidal e ect which is important for the orbital instability. Recently,
Lai, Rasio and Shapiro 62); 63) have pointed out that such a tidal coupling of binary
neutron stars is very important for their evolution in the nal merging phase because
the tidal e ect causes the instability of the circular motion even in the Newtonian
gravity. Also important is the general relativistic gravity because in the intermediate
phase, the orbital separation is about ten times as small as the Schwarzschild radius
of the system. Thus, we need not only a hydrodynamic treatment, but also general
relativistic one in order to study the nal stage of the evolution of binary neutron
stars. Since the usual formulation of the post-Newtonian approximation is based on
the particular gauge condition which is not so convenient for numerical purposes,
4 H. Asada and T. Futamase
we shall reformulate it by the (3+1) formalism often used in numerical approach of
general relativity.
This paper is organized as follows. In section 2 we introduce the Newtonian
limit in general relativity and present how to construct the post-Newtonian hierar-
chy. There we mention how to avoid divergent integrals which appears at higher
order in the previous treatments, and we also discuss how to incorporate strong in-
ternal gravity in the post-Newtonian approximation. Next we reformulate the post-
Newtonian approximation appropriate for numerical treatment in section 3. There
we shall adopt the (3+1) formalism which is frequently used in numerical relativity.
Based on the formalism developed in section 3, we present a formulation for con-
structing numerically equilibrium solutions of uniformly rotating uid in the 2PN
approximation in section 4. We shall also discuss the propagation of gravitational
waves from slow motion systems in section 5.
x2. Foundation of the post-Newtonian approximation
Since the Newtonian limit is the basis of the post-Newtonian approximation, we
shall rst formulate the Newtonian limit. We shall follow the formulation by Futa-
mase and Schutz 47) . We will not mention other formulation of Newtonian limit by
Ehlers 38); 39) , because it has not yet used to construct the post-Newtonian approxi-
mation.
2.1. Newtonian limit along a regular asymptotic Newtonian sequence
The formulation is based on the observation that any asymptotic approximation
of any theory need a sequence of solutions of the basic equations of the theory 87); 92) .
Namely, if we write the equations in abstract form as
E (g) = 0; (2.1)
for an unknown function g, one would like to have a one-parameter (or possibly
multiparameter) family of solutions,
E (g()) = 0; (2.2)
where  is some parameter. Asymptotic approximation then says that function f ()
approximates g() to order p if jf () g()j=p ! 0 as  ! 0. We choose the
sequence of solutions with appropriate properties in such a way that the properties
re ect the character of the system under consideration.
We shall formulate the post-Newtonian approximation according to the general
idea just described. As stated in the introduction, we would like to have an approx-
imation which applies the systems whose motion is described almost by Newtonian
theory. Thus we need a sequence of solutions of Einstein's equations parameterized
by  (the typical velocity of the system divided by the speed of light) which has the
Newtonian character as  ! 0.
The Newtonian character is most conveniently described by the following scal-
ing law. The Newtonian equations involve six variables, density (), pressure (P ),
Post-Newtonian Approximation 5
gravitational potential (), and velocity (vi ; i = 1; 2; 3):
r  4 = 0;
2
(2.3)
@t  + ri(vi ) = 0; (2.4)
@t vi + vj rj vi + riP + ri = 0; (2.5)
supplemented by an equation of state. For simplicity we have considered perfect
uid. Here we have set G = 1.
It can be seen that the variables f(xi ; t); P (xi ; t); (xi ; t); vi (xj ; t)g obeying the
above equations satisfy the following scaling law.
(xi; t) !  (xi; t);
2

P (xi; t) !  P (xi; t);


4

vi (xk ; t) ! vi(xk ; t);


(xi; t) !  (xi; t):
2
(2.6)
One can easily understand the meaning of this scaling by noticing that  is the
magnitude of typical velocity (divided by the speed of light). Then the magnitude of
the gravitational potential will be of order 2 because of the balance between gravity
and centrifugal force. The scaling of the time variable expresses the fact that the
weaker the gravity ( ! 0) the longer the time scale.
Thus we wish to have a sequence of solutions of Einstein's equations which has
the above scaling as  ! 0. We shall also take the point of view that the sequence
of solutions is determined by the appropriate sequence of initial data. This has a
practical advantage because there will be no solution of Einstein's equations which
satisfy the above scaling (2.6) even as  ! 0. This is because Einstein's equations
are nonlinear in the eld variables so it will not be possible to enforce the scaling
everywhere in spacetime. We shall therefore impose it only on the initial data for
the solution of the sequence.
Here we shall rst make a general discussion on the formulation of the post-
Newtonian approximation independent of any initial value formalism and then present
the concrete treatment in the harmonic coordinate. The condition is used because of
its popularity and some advantages in the generalization to the systems with strong
internal gravity.
As the initial data for the matter we shall take the same data set in Newtonian
case, namely, the density , the pressure P , and the coordinate velocity vi . In most
of the application, we usually assume a simple equation of state which relates the
pressure to the density. The initial data for the gravitational eld are g ; @g =@t.
Since general relativity is overdetermined system, there will be constraint equations
among the initial data for the eld. We shall write the free data for the eld as
(Qij ; Pij ) whose explicit forms depend on the coordinate condition one assumes. In
any coordinate we shall assume these data for the eld vanish since we are interested
in the evolution of an isolated system by its own gravitational interaction. It is
expected that these choice corresponds to the absence of radiation far away from the
6 H. Asada and T. Futamase
source. Thus we choose the following initial data which is indicated by Newtonian
scaling:
(t = 0; xi ; ) = 2a(xi);
P (t = 0; xi ; ) = 4b(xi);
vi(t = 0; xk ; ) = ci(xk );
Qij (t = 0; xi ; ) = 0;
Pij (t = 0; xi ; ) = 0; (2.7)
where the functions a; b , and ci are C 1 functions that have compact support con-
tained entirely within a sphere of a nite radius.
Corresponding to the above data, we have a one-parameter set of spacetime
parameterized by . It may be helpful to visualize the set as a ber bundle, with
base space R the real line (coordinate ) and ber R4 the spacetime (coordinates
t; xi). The ber  = 0 is Minkowski spacetime since it is de ned by zero data. In
the following we shall assume that the solutions are suciently smooth functions
of  for small  6= 0. We wish to take the limit  ! 0 along the sequence. The
limit is, however, not unique and is de ned by giving a smooth nowhere vanishing
vector eld on the ber bundle which is nowhere tangent to each ber 51); 87) . The
integral curves of the vector eld give a correspondence between points in di erent
bers, namely events in di erent spacetime with di erent values of . Remembering
the Newtonian scaling of the time variable in the limit, we introduce the Newtonian
dynamical time:
 = t; (2.8)
and de ne the integral curve as the curve on which  and xi stay constant. In fact
if we take the limit  ! 0 along this curve, the orbital period of the binary system
with  = 0:01 is 10 times of that of the system with  = 0:1 as expected from the
Newtonian scaling. This is what we de ne as the Newtonian limit. Notice that this
map never reaches the ber  = 0 (Minkowski spacetime). There is no pure vacuum
Newtonian limit as expected.
In the following we assume the existence of such a sequence of solutions con-
structed by the initial data satisfying the above scaling with respect to . We shall
call such a sequence is a regular asymptotically Newtonian sequence. We have to
make further mathematical assumptions about the sequence to make explicit calcu-
lations. We will not go into details of them partly because to prove the assumptions
we need a deep understanding of the existence and uniqueness properties of Cauchy
problem of Einstein's equations with perfect uids of compact support which are not
available at present.
2.2. Post-Newtonian Hierarchy
We shall now de ne the Newtonian, post-Newtonian and higher approximations
of various quantities as the appropriate higher tangents of the corresponding quanti-
ties to the above integral curve at  = 0. For example the hierarchy of approximations
for the spacetime metrics can be expressed as follows:
g (; ; xi ) = g (0; ; xi ) + (LV g )(0; ; xi )
Post-Newtonian Approximation 7
n
+ 21 2 (L2V g )(0; ; xi ) + ::::: + n! (LnV g )(0; ; xi ) + Rn+1 ; (2.9)
where LV is the Lie derivative with respect to the tangent vector of the curves de ned
above, and the remainder term Rn+1 is
Rn+1 = n Z d`(1 `)n (Ln g )(`; ; xi ):
+1 1
+1 +1
(2.10)
(n + 1)! 0
V 
Taylor's theorem guarantees that the series is asymptotic expansion about  = 0
under certain assumptions mentioned above. It might be useful to point out that
the above de nition of the approximation scheme may be formulated purely geomet-
rically in terms of jet bundle.
The above de nition of the post-Newtonian hierarchy gives us asymptotic series
in which each term in the series is manifestly nite. This is based on the  depen-
dence of the domain of dependence of the eld point (; xk ). The region is nite with
nite values of , and the diameter of the region increases as  1 as  ! 0. With-
out this linkage of the region with the expansion parameter , the post-Newtonian
approximation leads to divergences in the higher orders. This is closely related with
the retarded expansion. Namely, it is assumed that the slow motion assumption
enabled one to Taylor expand the retarded integrals in retarded time such as
Z Z Z
drf ( r; :::) = drf (; :::)  drrf (; :::); + :: (2.11)
and assign the second term to a higher order because of its explicit  in front. This
is incorrect because r !  1 as  ! 0 and thus r is not uniformly small in the
Newtonian limit. Only if the integrand falls o suciently fast, can retardation be
ignored. This happens in the lower-order PN terms. But at some higher order there
appear many terms which do not fall o suciently fast because of the nonlinearity of
Einstein's equations. This is the reason that the formal PN approximation produces
the divergent integrals. It turns out that such divergence appears at 3PN order
indicating the breakdown of the PN approximation in the harmonic coordinate.
This sort of divergence may be eliminated if we remember that the upper bound
of integration does depend on  as  1 . Thus we would get something like n ln 
instead of n ln 1 in the usual approach. This shows that the asymptotic Newtonian
sequence is not di erentiable in  at  = 0, but there are no divergence in the
expansion and it has still an asymptotic approximation in  that involves logarithms.
2.3. Explicit calculation in Harmonic coordinate
Here we shall use the above formalism to make explicit calculation in the har-
monic coordinate. The reduced Einstein's equation in the harmonic condition is
written as
g~ g~; = 16 g~ ; g~ ; ; (2.12)
@ [~g @ x ] = 0; (2.13)
where
g~ = ( g)1=2 g ; (2.14)
8 H. Asada and T. Futamase
 = ( g)(T + t
LL ); (2.15)
where t
LL is the Landau-Lifshitz pseudotensor 64) . In the following we shall choose

an isentropic perfect uid for T which is enough for most of applications.
T = ( +  + P )u u + Pg ; (2.16)
where  is the rest mass density,  the internal energy, P the pressure, u the four
velocity of the uid with normalization;
g u u = 1: (2.17)
The conservation of the energy and momentum is expressed as
r T = 0: (2.18)
De ning the gravitational eld variable as
h =  ( g)1=2 g ; (2.19)
where  is the Minkowski metric, the reduced Einstein's equation (2.12) and the
gauge condition (2.13) take the following form;
( h )h ; = 16 + h  ; h  ; ; (2.20)
h; = 0: (2.21)
Thus the characteristics is determined by the operator ( h )@ @ , and thus
the light cone deviates from that in the at spacetime. We shall use this form of
the reduced Einstein's equations in the calculation of wave form far away from the
source because the deviation plays a fundamental role there. However in the study
of the gravitational eld near the source it is not neccesary to consider the deviation
of the light cone away from the at one and thus it is convenient to use the following
form of the reduced Einstein's equations 2) .
 h ; = 16 ; (2.22)
where
 =  +  ; ; (2.23)
  = (16) (h  h  h h  ):
1
(2.24)
Equations (2.21) and (2.22) together imply the conservation law
 ; = 0: (2.25)
We shall take as our variables the set f; P; vi ; h g, with the de nition
vi = ui=u :0
(2.26)
Post-Newtonian Approximation 9
The time component of 4-velocity u0 is determined from Eq.(2.17). To make a well-
de ned system of equations we must add the conservation law for number density n
which is some function of the density  and pressure P :
r (nu ) = 0: (2.27)
Equations (2.25) and (2.27) imply that the ow is adiabatic. The role of the equation
of state is played by the arbitrary function n(; p).
Initial data for the above set of equations are h ; h ;0 ; ; P , and vi , but not
all these data are independent because of the existence of constraint equations.
Equations.(2.21) and (2.22) imply the four constraint equations among the initial
data for the eld.
h 0 + 16 0 ij h i;j ;0 = 0; (2.28)
where  is the Laplacian in the at space. We shall choose h ij and h ij;0 as free
data and solve Eq.(2.28) for h 0 ( = 0; ::; 3) and Eq.(2.21) for h 0 ;0 . Of course
these constraints cannot be solved explicitly, since  0 contains h 0 , but they can
be solved iteratively as explained below. As discussed above, we shall assume that
the free data h ij and h ij;0 for the eld vanish. One can show that such initial data
satisfy the O'Murchadha and York criterion for the absence of radiation far away
from the source 75) .
In the actual calculation, it is convenient to use an expression with explicit
dependence of . The harmonic condition allows us to have such expression in terms
of the retarded integral.
Z
h (; ; xi ) = 4 d3 y ( r; yi ; )=r + h i
H (; ; x ); (2.29)
C (;;xi)
where r = jyi xi j and C (; ; xi ) is the past at light cone of event (; xi ) in the
spacetime given by , truncated where it intersects with the initial hypersurface
 = 0. h 
H is the unique solution of the homogeneous equation
 h ; = 0: (2.30)
We shall henceforce ignore the homogeneous solution because they play no important
roles. Because of the  dependence of the integral region, the domains of integration
are nite as long as  6= 0 and their diameter increases as  1 as  ! 0.
Given the formal expression (2.29) in terms of initial data (2.7), we can take
the Lie derivative and evaluate these derivatives at  = 0. The Lie derivative is
nothing but partial derivative with respect to  in the coordinate system for the
ber bundle given by (; ; xi ). Accordingly one should convert all the time indices
to  indices. For example, T  = 2 T tt which is of order 4 since T tt   is of order
2. Similarly the other components of stress energy tensor T i = T ti and T ij are of
order 4 as well. Thus we expect the rst non-vanishing derivative in (2.29) will be
the forth-derivatives. In fact we nd
Z (; yi )

(4) h
 i
(; x ) = 4 3 (2) r d3 y; (2.31)
R
10 H. Asada and T. Futamase
Z (; yk ) vi (; xk )
(4) h i (; xi) = 4 (2) (1)

r
3
d y; (2.32)
R 3
Z (; yk ) vi (; yk ) vj (; yk ) + tijLL(; yk )
(4) h ij (; xk ) = 4 (2) (1) (1)

r
(4) 3
d y;
R
3
(2.33)
where we have adopted the notation
f (; xi) = n1! lim @ n f (; ; xi); (2.34)
( ) n ! @n
0

and
(4) tijLL = 641 ( h ;i h ;j 12 ij h ;k h ;k ):
(4) (4) (4) (4) (2.35)
In the above calculation we have taken the point of view that h  is a tensor
eld, de ned by giving its componentsp in the assumed harmonic coordinate as the
di erence between the tensor density gg and  .
The conservation law (2.25) also has its rst nonvanishing derivative at this
order, which are
k k i k
(2) (; x ); + ((2) (; x )(1) v (; x ));i = 0; (2.36)
((2) (1) vi ); + ((2) (1) vi (1) vj );j + (4) P ;i 41 (2) (4) h ;i = 0: (2.37)
Equations (2.31), (2.36), and (2.37) consist of Newtonian theory of gravity. Thus the
lowest non-vanishing derivative with respect to  is indeed Newtonian theory, and
the 1PN and 2PN equations emerge from sixth and eighth derivatives, respectively,
in the conservation law (2.25). At the next derivative, the quadrupole radiation
reaction term comes out.
2.4. Strong point particle limit
If we wish to apply the post-Newtonian approximation to the inspiraling phase
of binary neutron stars, the strong internal gravity must be taken into account. The
usual post-Newtonian approximation assumes explicitly the weakness of gravitational
eld everywhere including inside the material source. It is argued by appealing
the strong equivalence principle that the external gravitational eld which governs
the orbital motion of the binary system is independent of internal structure of the
components up to tidal interaction. Thus it is expected that the results obtained
under the assumption of weak gravity also apply for the case of neutron star binary.
Experimental evidence for the strong equivalence principle is obtained only for the
system with weak gravity 97); 98) , but no experiment is available in case with strong
internal gravity at present.
In theoretical aspect, the theory of extended object in general relativity 34) is still
in preliminary stage for the application to realistic systems. Matched asymptotic
expansion technique has been used to treat the system with strong gravity in certain
situations 32); 33); 55); 56); 27) . Another way to handle with strong internal gravity is by
Post-Newtonian Approximation 11
the use of Dirac's delta function type source with a xed mass 37) . However, this
makes Einstein's equations mathematically meaningless because of their nonlinearity.
Physically, there is no such source in general relativity because of the existence of
black holes. Before a body shrinks to a point, it forms a black hole whose size is
xed by its mass. For this reason, it has been claimed that no point particle exists
in general relativity.
This conclusion is not correct, however. We can shrink the body keeping the
compactness (M=R), i.e., the strength of the internal eld xed. Namely we should
scale the mass M just like the radius R. This can be tted nicely into the concept of
regular asymptotic Newtonian sequence de ned above because there the mass also
scales along the sequence of solutions. In fact, if we take the masses of two stars as M,
and the separation between two stars as L, then   M=L. Thus the mass M scales
as 2 if we x the separation. In the above we have assumed that the density scales
as 2 to guarantee this scaling for the mass understanding the size of the body xed.
Now we shrink the size as 2 to keep the compactness of each component. Then
the density should scale as  4 . We shall call such a scheme as strong point particle
limit since the limit keeps the strength of internal gravity. The above consideration
suggests the following initial data to de ne a regular asymptotic Newtonian sequence
which describes nearly Newtonian system with strong internal gravity 45) . The initial
data are two uniformly rotating uids with compact spatial support whose stress-
energy tensor and size scales as  4 and 2 , respectively. We also assume that each
of these uid con gurations would be a stationary equilibrium solution of Einstein's
equation if other were absent. This is neccesary for the suppression of irrelevant
internal motions of each star. Any remaining motions are the tidal e ects caused
by the other body, which will be of order 6 smaller than the internal self-force.
This data allows us to use the Newtonian time  = t as a natural time coordinate
everywhere including inside the stars.
This choice of the data leads naturally to the introduction of the body zones BA
and the body zone coordinates xkA de ned by
BA = fxk ; jxk Ak j < Rg; (2.38)
k k
xA =  (x A);
2 k (2.39)
where R is some constant, and Ak ( ); A = I; II are the coordinates of the origin
of the two stars, where we have used letters with bar to distinguish the body-zone
coordinates from their counterparts. The scaling by  2 means that as the star
shrinks with respect to the coordinate xk , it remains of xed size in the body-zone
coordinate xk . The boundary of the body zone shrinks to a point with respect to xk ,
and expands to in nity with respect to xk as  ! 0. This makes a clean separation
of the body zone from the exterior geometry generated by other star.
g = (gB ) + (gC B ) ; (2.40)
where (gB ) is the contribution from the body zones, (gC B ) from elsewhere.
Actual calculations are most easily performed in the harmonic coordinate. Take
two stationary solutions of Einstein's equations for the perfect uid fTA (xi ); gA (xi )g
12 H. Asada and T. Futamase
as our initial data in the body zone. As explained above every component of the
stress-energy tensor TA has the same  4 scaling in the inertia coordinates (t; xk )
for a rapidly rotating star. In the body zone coordinates (; xk ), these have the
following scalings:
TA = 2 TAtt   2; (2.41)
TAi =  1TAti   5; (2.42)
TA =  TA   :
ij 4 ij 8
(2.43)
Transformed to the near zone coordinates (; xk ), xk = Ai + 2xk , they take the form
TN = TA ; (2.44)
TNi = 2 TAi + vAi TA ; (2.45)
TNij = 4 TAij + 22 vA(i TAj) + vAi vAj TA ; (2.46)
where vAi = dAi =d is the velocity of the origin of the body A. If there were only one
body, these data would produce a stationary solution in the body zone, which moves
with uniform velocity vAi in the near-zone coordinates. Now we know the ordering of
the source, we can solve Eq.(2.29),iteratively as in the weak eld case. The di erence
is that we transform the integration variables to the body zone coordinates in the
expression of the elds whose contributions come from the body zone.
XZ
h  i
B (; ; x ) = 4
6
d yAjxA  yAj  ( jxA  yj; yi ; ); (2.47)
3 2 1 2

A BA
where xiA = xi Ai . Expanding these expressions in terms of , we obtain
X MA
h  i
B (; ; x ) = 4
4
jx j + O( );
6
(2.48)
A A
i
h iB(; ; xi ) = 43 jxJA j
X X MA vA
+ 44 j x j + O ( ); 5
(2.49)
A A A A
X ij X vAi JAj X MA vAi vAj
h ijB (; ; xk ) = 42 jZxAj
( )

+ 83
j x j + 4  j x
4
j + O( ); (2.50)
5

A A A A A A
where
Z
MA = lim
!
 2
d y
3
A (; ; y); (2.51)
ZBA
0

JAi = lim
!
 3
d yiA (; ; y);
3
(2.52)
ZBA
0

ZAij = lim
!
 4
d yijA (; ; y):
3
(2.53)
0 BA
The MA de ned above is the conserved ADM mass the body A would have if it
were isolated. Without loss of generality one can set the linear momentum JAi of
the overall internal motion of each body to zero by choosing appropriate origin of
Post-Newtonian Approximation 13
the coordinates. If we did not assume the internal stationarity in the initial data,
then ZAij would be nite and no approximation would be possible. However under
the condition of internal stationarity one can show that ZAij vanishes 45) . Above
expressions for the body zone contribution are used to calculate the pseudotensor in
the near zone and then to evaluate the contribution h 
C B 6outside the body zone.
As the result we shall obtain the metric variables up to O( ).
h  (; xi ) = 44 MA + O( );
X 6
(2.54)
A jxA j
i X k
h i (; xi ) = 4 MjxAvjA + +2 jxxAj MAki + O( );
X
4 5 6
(2.55)
A A A A
i j
h ij (; xk ) = 4 MAjxvAjvA 2 Iorbij + O( );
X
4 5 (3) 6
(2.56)
A A
where
Z
MAij = lim
!
2 4
d yy ijA (; ; y);
3 [ ]
(2.57)
B
X i j A i
0
ij
Iorb = AAMA JA: (2.58)
A
These are the angular momentum of the body A and the quadrupole moment for
the orbital motion, respectively. This kind of calculation is also performed at 2.5PN
order to get the standard quadrupole formula except that the mass in the de nition
of the quadrupole moment is replaced with the ADM mass 45) . The above method
has been also applied for the calculation of spin precession and the same form of the
equation in the case of weak gravity is obtained. Thus we can extend the strong
equivalence principle for bodies with strong internal gravity to the generation of
gravitational waves and the spin precession. It is an open question if the strong
point particle limit may be taken to lead a well de ned equation of motion at higher
post-Newtonian orders.
x3. Post-Newtonian Approximation in the (3+1) Formalism
In the evolution of binary neutron stars, the orbital separation eventually be-
comes comparable to the radius of the neutron star due to radiation reaction. Then,
the point particle picture does not apply and each star of the binary begins to behave
as a hydrodynamic object because of tidal e ect. As described in the introduction we
need not only a hydrodynamic treatment, but also general relativistic one in order to
study the nal stage of the evolution of binary neutron stars. Full general relativistic
simulation will be a direct way to answer such a question, but it is one of the hardest
problem in astrophysics. Although there is much progress in this direction 69), it will
take a long time until numerical relativistic calculations become reliable. One of the
main reasons for this would be that we do not know the behavior of geometric vari-
ables in the strong gravitational eld around coalescing binary neutron stars. Owing
14 H. Asada and T. Futamase
to this, we do not know what sort of gauge condition is useful and how to give an
appropriate general relativistic initial condition for coalescing binary neutron stars.
The other reason is a technical one: In the case of coalescing binary neutron
stars, the wavelength is of order of   R3=2 M 1=2 , where R and M are the orbital
radius and the total mass of binary, respectively. Thus, in numerical simulations, we
need to cover a region L >  / R3=2 with numerical grids in order to attain good
accuracy. This is in contrast with the case of Newtonian and/or PN simulations,
in which we only need to cover a region  > L > R. Since the circular orbit of
binary neutron stars becomes unstable at R  10M owing to the tidal e ects 62); 63)
and/or the strong general relativistic gravity 59) , we must set an initial condition of
binary at R  10M . For such a case the grid must cover a region L >   100M in
numerical simulations to perform an accurate simulation. When we assume to cover
each neutron star of its radius  5M with  30 homogeneous grids 72) - 74); 85); 86) , we
need to take grids of at least  5003 , but it seems impossible to take such a large
amount of mesh points even for the present power of supercomputer. At present,
we had better search other methods to prepare a precise initial condition for binary
neutron stars.
In the case of PN simulations, the situation is completely di erent because we
do not have to treat gravitational waves explicitly in numerical simulations, and
as a result, only need to cover a region at most L  20 30M . In this case, it
seems that  2003 grid numbers are enough. Furthermore, we can take into account
general relativistic e ects with a good accuracy: In the case of coalescing binary
neutron stars, the error will be at most  M=R  a few  10% for the rst PN
approximation, and  (M=R)2  several % for 2PN approximation. Hence, if we
can take into account up through 2PN terms, we will be able to give a highly accurate
initial condition (the error  several %).
For these reasons, we present in this section the 2.5PN hydrodynamic equations
including the 2.5PN radiation reaction potential in such a way that one can apply
the formulation directly in numerical simulations. As for the PN hydrodynamic
equation, Blanchet, Damour and Schafer 15) have already obtained the (1+2.5) PN
equations. In their formulation, the source terms of all Poisson equations take nonva-
nishing values only on the matter, like in the Newtonian hydrodynamics. Although
their formulation is very useful for PN hydrodynamic simulations including the radi-
ation reaction 72) - 74); 85); 86); 78) , they did not take into account 2PN terms. In their
formulation, they also xed the gauge conditions to the ADM gauge, but in numer-
ical relativity, it has not been known yet what sort of gauge condition is suitable
for simulation of the coalescing binary neutron stars and estimation of gravitational
waves from them. First, we develop the formalism for the hydrodynamics using the
PN approximation. In particular, we use the (3+1) formalism of general relativity
so that we can adopt more general class of slice conditions 46); 4) . Next, we present
methods to obtain numerically terms at the 2PN order 4) .
3.1. (3+1) Formalism for the post-Newtonian Approximation
According to the above discussion, we shall apply the (3+1) formalism to formu-
late the PN approximation. In the (3+1) formalism 3); 92); 68) , spacetime is foliated
Post-Newtonian Approximation 15
by a family of spacelike 3D hypersurfaces whose normal one-form is taken as
n^  = ( ; 0): (3.1)
Then the line element takes the following form
ds2 = ( 2 i i)dt2 + 2 i dtdxi + ij dxidxj ; (3.2)
where ; i and ij are the lapse function, shift vector and metric on the 3D hyper-
surface, respectively. Using the (3+1) formalism, the Einstein's equation
G = 8T ; (3.3)
is also split into the constraint equations and the evolution equations. The formers
are the so-called Hamiltonian and momentum constraints which respectively become
trR Kij K ij + K 2 = 16H ; (3.4)
Di K ij Dj K = 8Jj ; (3.5)
where Kij , K , trR and Di are the extrinsic curvature, the trace part of Kij , the
scalar curvature of 3D hypersurface and the covariant derivative with respect of ij .
H and Jj are de ned as
H = T n^ n^  ; (3.6)
Jj = T n^  j : (3.7)
The Evolution equations for the spatial metric and the extrinsic curvature are re-
spectively
@ = 2 K + D + D ; (3.8)
@t ij ij i j j i
@ K = (R + KK 2K K l ) D D + (D m )K + (D m )K
@t ij ij ij il j i j

j mi

i mj
+ m D 1 l (3.9)
m Kij 8 Sij + 2 ij (H S l ) ;
@ = 2 ( K + D i); (3.10)
@t i
@ i j l (3.11)
@t K = (trR + K ) D Di + Dj K + 4 (S l 3H );
2

where Rij , and Sij are, respectively, the Ricci tensor with respect of ij , the
determinant of ij and
Sij = Tkl ki l j : (3.12)
Hereafter we use the conformal factor = = instead of for simplicity.
1 12

To distinguish the wave part from the non-wave part (for example, Newtonian
potential) in the metric, we use ~ij = 4 ij instead of ij . Then det(~ ij ) = 1 is
satis ed. We also de ne A~ij as
 
~
Aij  Aij 
4 4 1
Kij 3 ij K : (3.13)
16 H. Asada and T. Futamase
We should note that in our notation, indices of A~ij are raised and lowered by ~ij ,
so that the relations, A~i j = Ai j and A~ij = 4 Aij , hold. Using these variables, the
evolution equations (3.8)-(3.11) can be rewritten as follows;
@ ~ = 2 A~ + ~ @ l + ~ @ l 2 ~ @ l ; (3.14)
@n ij  
ij il @xj jl @xi 3 ij @xl
  
@ A~ = 1 R 1 trR D~ i D~ j 1 ~ij  ~
@n ij 4 ij 3 ij 3
2  + 2 ~ ~ kl 
;i ;j ;j ;i 3 ij ;k ;l
m @ m A~ 2 @ m A~
+ (K A~ij 2A~il A~l j ) + @ A~ +
i mj @xj mi 3 @xm ij
  @x
8 4 Sij 13 ij S l l ; (3.15)
@ =  K + @ i ; (3.16)
@n 6 @xi 
@ K = A~ A~ij + 1 K 2 1 
~ 2 ~ kl + 4 (S i +  ); (3.17)
@n ij 3 4 5 ;k ;l i H

where D~ i and ~ are the covariant derivative and Laplacian with respect to ~ij and
@ = @ i @ : (3.18)
@n @t @xi
The Hamiltonian constraint equation then takes the following form.
5 
~ 1 ~
 = 8 trR 2H ~ ~ij 2 (3.19)
8 Aij A 3 K ;
5 2

where trR~ is the scalar curvature with respect to ~ij . The momentum constraint is
also written as
D~ j ( 6 A~ji) 23 6 D~ iK = 8 6 Ji: (3.20)
Now let us consider Rij in Eq.(3.15), which is one of the main source terms of
the evolution equation for A~ij . First we split Rij into two parts as
Rij = R~ij + Rij ; (3.21)
where R~ ij is the Ricci tensor with respect to ~ij and Rij is de ned as
Rij = 2 D~ i D~ j 2 ~ij D~ k D~ k + 62 (D~ i )(D~ j ) 22 ~ij (D~ k )(D~ k ): (3.22)
Using det(~ ij ) = 1, R~ ij is written as

~ 1
Rij = 2 hij;kk + hjl;li + hil;lj + f kl;k (hlj;i + hli;j hij;l)

+ f (hkj;il + hki;jl hij;kl ) ~kjl ~lik ;
kl (3.23)
Post-Newtonian Approximation 17
where ;i denotes @=@xi , ~ijk is the Christo el symbol, and we split ~ij and ~ ij as
ij + hij and ij + f ij , by writing the at metric as ij .
We shall consider only the linear order in hij and fij assuming jhij j; jfij j  1.
(As a result, hij = f ij + O(h2 ).) Such a treatment is justi ed because hij turns
out to be 2PN quantity in our choice of gauge condition (see below). Here, to clarify
the wave property of ~ij , we impose a kind of transverse gauge to hij as
hij;j = 0; (3.24)
which was rst proposed by Nakamura in relation to numerical relativity 69) . Here-
after, we call this condition merely the transverse gauge. The equation (3.14) shows
that this condition is guaranteed if i satis es
 
;k j ~ij;k = 2 A~ij + ~il l ;j + ~jl l ;i 32 ~ij l ;l : (3.25)
;j
Using the transverse gauge, Eq.(3.23) becomes
R~ij = 12 hij + O(h2 ); (3.26)
where  is the Laplacian with respect to ij . Note that trR~ = O(h2 ) is guaranteed in
the transverse gauge because the traceless property of hij holds in the linear order.
We show the equations for the isentropic perfect uid (2.16). The conservation
law of mass density, (2.27), may be written as
@ + @ ( vi ) = 0; (3.27)
@t @xi
where  is the conserved density de ned as
 = 6 u0: (3.28)
The equation of motion and the energy equation are obtained from the conservation
law (2.18) which takes the following forms.
@Si + @ (Si vj ) = 6 P S 0 + S j 1 jk (3.29)
@t @xj ;i ;i j ;i 2S 0 Sj Sk ;i;
and
@H + @ (Hvj ) = P  @ ( 6 u0 ) + @ ( 6 u0vj ) ; (3.30)
@t @xj @t @xj
where  
6 0 P
Si = ( +  + P )u ui =  1 +  +  ui(= 6 Ji );
 6
S 0 = 6 ( +  + P )(u0 )2 = (H + P ) ;

H = u =  ;
6 0

i ij
vi  uu = i + S Sj :
0 0
(3.31)
Finally, we note that in the above equations, only i appears, and i does not, so
that, in the subsequent section, we only consider the PN expansion of i , not of i .
18 H. Asada and T. Futamase
3.2. Post-Newtonian approximation in the (3+1) formalism
Next, we consider the PN approximation of the above set of equations. First
of all, we review the PN expansion of the variables. Each metric variable may be
expanded up to the relevant order as
= 1 + (2) + (4) + (6) + (7) + : : : ;
= 1 + (2) + (4) + (6) + (7) + : : : ;
 
U 2
= 1 U + 2 + X + + +:::; (6) (7)

i = i + i + i + i + i + : : : ;
(3) (5) (6) (7) (8)

hij = hij + hij + : : : ;


(4) (5)

A~ij = A~ij + A~ij + A~ij + : : : ;


(3) (5) (6)

K = K + K + K + :::;
(3) (5) (6) (3.32)
where subscripts denote the PN order(n) and U is the Newtonian potential satisfying
U = 4: (3.33)
X depends on the slice condition, and in the standard PN gauge , we usually use 21)

 = X=2, which satis es


 
1
 = 4 v + U + 2  + 2  :
2 3 P (3.34)

Note that the terms relevant to the radiation reaction appear in , , i and (7) (7) (8)
hij , and the quadrupole formula is derived from and hij .
(5) (7) (5)
The four velocity is expanded as
     
u =  1 + 2 12 v2 + U + 4 38 v4 + 52 v2U + 12 U 2 + (3) ivi X + O(6 ) ;
     
u =  1 1 + 2 12 v2 U + 4 38 v4 + 32 v2 U + 12 U 2 + X + O(6 ) ;
    
ui = vi 1 + 2 12 v2 + U + 4 38 v4 + 52 v2 U + 12 U 2 + (3) i vi X + O(7);
     
ui = vi + 3 (3) i + vi 12 v2 + 3U + 5 (5) i + (3) i 12 v2 + 3U + (4) hij vj
   
+ vi 83 v4 + 72 v2 U + 4U 2 X + 4(4) + (3) j vj + 6 (6) i + (5) hij vj
+O(7 ); (3.35)
where vi = O() and v2 = vi vi . From u u = 1, we obtain the useful relation
( u )2 = 1 + ij ui uj  
= 1 + 2 v2 + 4 v4 + 4v2 U + 2(3) i vi + O(6 ): (3.36)
Post-Newtonian Approximation 19
Thus H , Ji and Sij are respectively expanded as
      
H =   1 +  v +  +  v + v 4U +  + P + 2 ivi + O( );
2 2 2 4 4 2
(3)
8

    
Ji =   vi 1 +  v + 3U +  + P +  i + O( );
3 2 2 3
(3)
7

   
4 j P 2 P
Sij =   v v +  ij +  v + 6U +  +  vivj + vi j + vj i
i 2
(3) (3)

UP
+ 2  ij + O( ); 8

   
l 3P i i
Sl =   v +  +  2 v + v v + 4U +  + 
4 2 2 P 2
+ O( ): (3.37)
2 8
(3)

The conformal factor (and in the conformal slice) is determined by the


Hamiltonian constraint. In the PN approximation, the Laplacian with respect to ~ij
for the scalar is expanded as
~ =  (4(4) hij + 5(5) hij )@i @j + O(6 ): (3.38)
At the lowest order, the Hamiltonian constraint becomes
(2) = 2: (3.39)
Thus, (2) = 2(2) = U is satis ed in this paper. At the 2PN and 3PN orders,
the Hamiltonian constraint equation becomes, respectively,
 
(4) = 2 v +  + 2 U ;
2 5 (3.40)
and
   
(6) = 2 v4 + v2  + P + 132 U + 2(3) ivi + 25 U + 52 U 2 + 5(4)
 
+ 21 (4) hij U;ij 18 (3) A~ij (3) A~ij 32 (3) K 2 : (3.41)
The term relevant to the radiation reaction rst appears in (7) whose equation
becomes
(7) = 21 (5) hij U;ij : (3.42)
Hence, (7) may be also relevant to the radiation reaction, depending on the slice
condition.
Let us now derive equations for i . From Eq.(3.25), the relation between (3) A~ij
and (3) i becomes
2(3) A~ij + (3) ;ji + (3) j ; i 32 ij (3) ;ll = 0: (3.43)
where we used the boundary condition that (3) A~ij and (3) i vanishes at the spatial
in nity. (3) A~ij must also satisfy the momentum constraint. Since (3) A~ij does not
20 H. Asada and T. Futamase
contain the transverse-traceless (TT) part and only contains the longitudinal part,
it can be written as
~ 2 W ; (3.44)
(3) Aij = (3) Wi;j + (3) Wj;i
3 ij (3) k;k
where (3) Wi is a vector on the 3D hypersurface and satis es the momentum constraint
at the rst PN order as follows;
(3) Wi + 13 (3) Wj;ji 23 (3) K;i = 8vi : (3.45)
From Eq.(3.43), the relation,
i
(3) = 2(3) Wi ; (3.46)
holds and at the rst PN order, Eq.(3.16) becomes
3U_ = (3) K + (3) l ; l ; (3.47)
where U_ denotes the derivative of U with respect to time. Thus Eq.(3.45) is rewritten
as  
i i
(3) = 16v + (3) K;i U;i : _ (3.48)
This is the equation for the vector potential at the rst PN order.
At the higher order, (n) i is also determined by the gauge condition, hij;j = 0.
The equation for (5) i is obtained by using the momentum constraint and the 2PN
order of Eq.(3.16) as follows,
   
(5) i = 16 vi v2 + 2U +  + P + (3) i 8U;j (3) A~ij

+(5) K;i U (3) K;i + 13 U;i (3) K 2(4) _ ;i + 12 (U U_ );i + ((3) l U;l );i :
(3.49)
The equation for (6) i takes the purely geometrical form since the material contri-
bution Ji at the 1.5PN order vanishes.
(6) i = (6) K;i : (3.50)
Then, let us consider the wave equation for hij . From Eqs.(3.14), (3.15), (3.21)
and (3.26), the wave equation for hij is written as
  
2hij = 1 2
hij + @n @ 2
@ h2

 
4 2 @t ij 
2

2 2 D~ D~ 1 ~ 6 D~ D~ 1 ~ D~ D~ k 
+ 4 i j 3 ij ~ + i j 2 3 ij k
   
D~ iD~ j 1 ~ ~ 2 D~ D~ + D~ ~ 2 ~ k
Di 3 ~ij D Dk ~

3 ij 
i j

j

+2 2 K A~ij 2A~il A~l j + 2 ;mi A~mj + m;j A~mi 23 m;m A~ij


16 Sij
2
1 S l  @  m ~ + m ~ 2 m ~  + 2 @ A~
4 3 ij l @n ;i mj ;j mi 3 ;m ij @n ij
 ij ; (3.51)
Post-Newtonian Approximation 21
where 2 is the at spacetime wave operator de ned as
@ + :
2 = @t
2
(3.52)
2

We should note that (4) ij has the TT property, i.e., (4) ij;j = 0 and (4) ii = 0. This
is a natural consequence of the transverse gauge, hij;j = 0 and hii = O(h2 ). Thus
(4) hij is determined from
(4) hij = (4) ij : (3.53)
Since O(h2 ) turns out to be O(8 ), it is enough to consider only the linear order of
hij in the case when we perform the PN approximation up to the 3.5PN order. We
can obtain (5) hij by evaluating
h (  ) = 1 @ Z  (; y)d3 y; (3.54)
(5) ij
4 @ (4) ij
and the quadrupole mode of gravitational waves in the wave zone is written as
Z jx yj; y) 3
1 (4) ij (
hij (; x) = 4 jxlim
rad
j!1 jx yj d y: (3.55)
Later, we shall derive the quadrupole radiation-reaction metric in the near zone using
Eq.(3.54).
Finally, we show the evolution equation for K . Since we adopt slice conditions
which do not satisfy K = 0 (i.e. the maximal slice condition), the evolution equation
for K is necessary. The evolution equations appear at the 1PN, 2PN and 2.5PN
orders which become respectively
@ K = 42v +  + 2U + 3 P  X;
2
(3.56)
@ (3)

@ K = 42v + v 6U + 2 + 2 P   + 3P U 4U + 4 + X
4 2 2
@ (5)

  (4)

+ 4 i vi + A~ij A~ij + 13 K
(3) hij U;ij + i K;i
(3) (3) (3)
2
(4) (3) (3)

3 UU U U X + 2U (3.57)
2 ;k ;k ;k ;k ;k (4) ;k (6) + 2UX;
@ K=  hij U;ij : (3.58)
@ (6) (7) (5)

We note that for the PN equations of motion up to the 2.5PN order, we need
i i i
(2) , (4) , (6) , (7) , (2) , (4) , (3) , (5) , (6) , (4) hij , (5) hij , (3) K , (5) K and (6) K .
Therefore, if we solve the above set of the equations, we can obtain the 2.5 PN
equations of motion. Up to the 2.5PN order, the hydrodynamic equations become
@Si + @ (Sivj ) = 1 + 2U + 5 U 2 + 6 + X P
@ @xj 4 (4) ;i
 
+ U;i 1 +  + P + 32 v2 U + 58 v4 + 4v2 U
22 H. Asada and T. Futamase
   
3 P
+ 2 v U  +  + 3(3) v
2 j j
 2
X;i 1 +  + P + v2 + 2v2 (4) ;i (6) ;i (7) ;i
   
;i 1 +  + P
v 2
+ vj j j j
(3)
 + 2 + 3U + ; i + ;i (5) (6)

+ j ;ij + 21 vj vk ( hjk;i + hjk;i) + O( ); (3.59)


(3) (3) (4) (5)
8

@H + @ (Hvj ) = P vj + @  1 v + 3U  + @  1 v + 3U vj  + O( );


2 2 5
@ @xj ;j @t 2 @xj 2
(3.60)
where we have omitted n coecients for the sake of simplicity, and used relations
 2  
S 0 =  1 +  + P + v2 + v2  + P + 38 v4 + 2v2 U + (3) j vj ) + O(6 );
2

   
i P v 2
i
Si =  v 1 +  +  + 2 + 3U + (3) + O(5 ): (3.61)
3.3. Strategy to obtain 2PN tensor potential
Although the 2PN tensor potential is formally solved as
h ( ; x ) = 1 Z (4) ij (; y) d3 y; (3.62)
(4) ij 4 jx yj
it is dicult to estimate this integral numerically since (4) ij ! O(r 3 ) for r ! 1
and the integral is taken all over the space. Thus it is desirable to replace this
equation by some tractable forms in numerical evaluation. We shall present two
methods to do so. One is to change Eq.(3.62) into the form in which the integration
is performed only over the matter distribution like as in the Newtonian potential.
The other is to solve Eq.(3.62) as the boundary value problem 4).
Strategy1: Direct integration method
The explicit form of (4) ij is
 
^ ^ i j 1
(4) ij = 2@ij (X + 2(4) ) + U @ij U 3U;i U;j + ij U;k U;k 16 v v 3 ij v 2

 
_ i _ j 2 _ k (3.63)
(3) ;j + (3) ;i
3 ij (3) ;k ;
where
@^ij  @x@i@xj 31 ij :
2
(3.64)
Although (4) ij looks as if it depends on the slice condition, the independence is
shown as follows. Eq.(3.48) is solved formally as
Z 
i = pi 1 (3) K
(3) 4 jx yj d 3
y ; (3.65)
;i
Post-Newtonian Approximation 23
where iZ Z 
v 1
pi = 4 jx yj d y 2 _jx yjd y : 3 3
(3.66)
;i
From Eqs.(3.40) and (3.57), we obtain

_ (X + 2(4) ) + 4 v2 + 3 P U : (3.67)
(3) K =
2
Combining Eq.(3.65) with Eq.(3.67), the equation for (3) _ i is written as
 
Z v + 3P U=2
2 
_ i _i (X + 2(4) );i
(3) = p dy : 3
(3.68)
jx yj ;i
Using this relation, the source term, (4) ij , is split into ve parts
(4) ij = ijS + ijU + ijC + ij + ijV ;
(4)
( )
(4)
( )
(4)
( )
(4)
( )
(4)
( )
(3.69)
where we introduced the following notations.
 
 (S )
= 16  v i vj 1 ij v2 ;
(4) ij
3
(U )
(4) ij = UU;ij 13 ij UU 3U;i U;j + ij U;k U;k ;
 (C )
= 4 @ Z (vi) d3y + 4 @ Z (vj ) d3 y 8  @ Z (vk ) d3 y;
(4) ij
@xZj jx yj @xi jx yj 3 ij @xk jx yj
() ^ jx yjd3 y;
(4) ij = @ij
 
V
v + 3P U=2
Z 2
( ^
ij = 2@ij
)
d y: (3.70) 3
(4)
jx yj
Thus it becomes clear that hij and hij as well as ij are expressed in terms
(4) (5) (4)
of matter variables only and thus their forms do not depend on slicing conditions,
though values of matter variables depend on gauge conditions.
Then, we de ne (4) h(ijS ) =  1 (4) ij(S ) , (4) h(ijU ) =  1 (4) ij(U ) , (4) h(ijC ) =  1 (4) ij(C ) ,
() ()
(4) hij = 
1
(4) ij and (4) h(ijV ) =  1 (4) ij(V ) , and consider each term separately.
First, since (4) ij(S ) is a compact source, we immediately obtain
 
Z vi vj ij v 1 2

hijS = 4 (3.71)
3
(4)
( )

jx yj d y: 3

Second, we consider the following equation


G(x; y1; y2) = jx y jj1x y j : (3.72)
1 2
24 H. Asada and T. Futamase
It is possible to write (4) h(ijU ) using integrals over the matter if this function, G, is
used. Eq.(3.72) has solutions 43); 71) ,
G(x; y1 ; y2) = ln(r1 + r2  r12 ); (3.73)
where
r = jx y1j; r = jx y2j; r = jy1 y2j:
1 (3.74) 2 12

Note that ln(r + r r ) is not regular on the interval between y1 and y2 , while
1 2 12
ln(r + r + r ) is regular on the matter. Thus we use ln(r + r + r ) as a kernel.
1 2 12 1 2 12
Using this kernel, UU;ij and U;i U;j are rewritten as
 Z Z
UU;ij = @x@i@xj jx(y1y) j d y
2
(y2) d y  3 3

1 jx y2j 1 2

Z  
@
= d y d y (y1 )(y2 ) i j jx y jjx y j
3 3 1 2

@y @y
1
1
2
2 1 1
Z
=  d y d y (y1 )(y2 ) @i j ln(r + r + r );
2
3 3

@y @y 1 2
1 1
1 2 12

 Z  Z 
@  ( y 1) @
U;iU;j = @xi jx y j d y @xj jx y j d y ( y2) 3 3
1 2
1 2
Z  
@
= d y d y (y1 )(y2 ) i j jx y jjx y j
3 3 1 2

@y @y
1
1
2
2 1 2
Z
=  d y d y (y1 )(y2 ) @i j ln(r + r + r ):
2
3 3
(3.75)
@y @y 1 2
1 2
1 2 12

Thus we can express (4) h(ijU ) using the integral over the matter as

Z
hijU = d y d y (y1)(y2 )
( ) 3 @ 1  4  3 @
3
2 2
1  4 
(4)
@yi @yj 3 ij
1 2
@yi @yj 1 1
1
1 2
3 ij 12
 ln(r + r + r ); 1 2 12 (3.76)
where we introduced
4 = @y@k @yk ; 4 = @y@k @yk :
2 2

1 12 (3.77)
1 1 1 2

Using relations jx yj = 2=jx yj and jx yj = 12jx yj, hijC , 3


(4)
( )
(4) hij and
( )

hijV are solved as


(4)
( )

Z Z Z
C ( @ ) j  @ i  4
hij = 2 @xi (v ) jx yjd y + 2 @xj (v ) jx yjd y + 3 ij jx 3 3
yjd y; 3
(4)

(3.78)
Z Z
 1 @ 1
hij = 12 @xi@xj jx yj d y 3 ij jx yjd y;
( )
2
3 3 3
(3.79)
(4)
Post-Newtonian Approximation 25
 
Z 
2  Z v + 3P U=2 d y:
2

hijV = @x@i@xj v + 3P U
2

(4)
( ) 2
2 j x y j d y 3 ij
3
jx yj
3

(3.80)
Thus we nd that the 2PN tensor potentials can be expressed as the integrals only
over the matter.
Strategy2: Partial use of boundary value approach
Although the above expression for (4) hij is quite interesting and might play
an important role in some theoretical applications, it will take a very long time
to evaluate double integration numerically. Therefore, we propose another strategy
where Eq.(3.53) is solved as the boundary value problem. Here, we would like to
emphasize that the boundary condition should be imposed at r(= jxj)  jy1 j; jy2 j,
but r does not have to be greater than , where  is a typical wave length of
gravitational waves. We only need to impose r > R (a typical size of matter). This
means that we do not need a large amount of grid numbers compared with the case
of fully general relativistic simulations, in which we require r >   R.
First of all, we consider the equation
 
 (4) hij + (4) hij = (4) ij(S) + (4) ij(U ) :
(S ) (U )
(3.81)
Since the source (4) ij(U ) behaves as O(r 6 ) at r ! 1, this equation can be accurately
solved under the boundary condition at r > R as
h(S )
+ h (U )
= 2 I 1  I 
(4) ij (4) ij
r ij 3 ij kk 
+ 32r2 nk Iijk 13 ij nk Illk + 2nk (S_ ikj + S_ jki) 43 ij nk S_ lkl
+O(r 3 ); (3.82)
where Z
Iijk = xixj xk d3 x;
Z
Sijk = (vi xj vj xi)xk d x: 3
(3.83)
Next, we consider the equations for (4) h(ijC ) , (4) h(ij) and (4) h(ijV ) . Using the iden-
tity,
 = (vi);i = (vivj );ij + P (U;i);i ; (3.84)
we nd the following relations;
Z Z i i
jx yjd y =
3
d y jxx yy j (vi );
3

Z Z " i i j j
jx yj d y = 3 d y vivj (x jyx )(xyj y )
3 3 3

  #
+ 4P + v 2
U;i (xi yi ) jx yj : (3.85)
26 H. Asada and T. Futamase
Using Eqs.(3.85), (4) h(ijC ) , (4) h(ij) and (4) h(ijV ) in Eqs.(3.78)-(3.80) can be written as
Z
C) = 2 (vj ) x
i yi d y + 2 Z (vi ) xj yj d y 4  Z (vk ) xk yk d y;
(4) hij
( 3 3 3
jx yj jx yj 3 ij jx yj
(3.86)
Z
hij = 14 @x@i @xj vk vl (x
( )
2 k yk )(xl yl) d y + 1  Z (vk ) xk yk d y
3 3
(4)
(
jx yj 3 ij j)x yj
1
+ 2 @x @ Z P 0 (xj 3
Z
yj ) d y + @ P 0 (xi yi) d y 3
i jx yj @xj jx yj
( Z
1 2  U;j (xi yi) + U;i(xj yj ) d y 3
8 jx yj
@ Z
U ;k ( x j yj ) @ Z
U ;k ( x i yi ) )
+ xk @x i  jx yj d y + x @xj  jx yj d y ; (3.87)
k 3 3

and
V) = 1  @ Z v2 + 3P U  xj yj d y
(4) hij
( 3
2 @xi 2 jx yj
Z  i i 
@ U x
+ @xj v + 3P 2 jx yj d y y2 3

 
2  Z v + 3P U=2 d y;
2

3 ij jx yj
3
(3.88)

where P 0 = P + v2 =4 + U;l yl =4. Furthermore they are rewritten in terms of the
potentials as follows.
 
(C ) i _ j j _ i 4 Q kk k _ k
(4) hij = 2(x (3) P + x (3) P Qij ) + 3 ij 2 x (3) P ;

   
hij = 14 @x@i@xj Vklv xk xl 2Vk v xk + V v + 31 ij xk P_k Q2kk
2
( ) ( ) ( ) ( )
(4) (3)
    
1 @ P j
+ 2 @xi V x Vj + @xj V x ViP( ) @ P i (P ) ( ) ( )

1 2xi V U + xi V U V U V U 
( ) ( ) ( ) ( )

8 j i ij ji
   
k @
+ x @xi x Vk j U U k @
Vkj + x @xj x Vk
( ) i U U
Vki (
; ) ( ) ( )

V) = 1 Q(I ) xi + Q(I ) xj Q(I ) Q(I )  + 1 Q(I )  ; (3.89)


(4) hij
(

2 ;j ;i i;j j;i 3 ij
Post-Newtonian Approximation 27
where the potentials are de ned as
 Pi = 4vi ;
(3)
 
Qij = 4 xj (vi ) + xi(vj ) ;
 
Q I = 4 v + 3P 21 U ;
( ) 2

 
QiI = 4 v + 3P 12 U xi;
( ) 2

Vijv =
( )
4vi vj ;
Vi v =
( )
4vi vj xj ;
V v =
( )
4(vj xj )2 ;
V P =
( )
4P 0 ;
Vi P =
( )
4P 0 xi ;
Vi U =
( )
4U;i ;
VijU =
( )
4U;i xj : (3.90)
It should be noted that these Poisson equations have compact sources.
In this strategy (4) h(ijS ) and (4) h(ijU ) are solved as the boundary value problem,
while other parts are obtained by the same method as in Newtonian gravity.
Strategy 3: Boundary Value approach
Instead of the above procedure, we may solve the Poisson equation for (4) hij as
a whole carefully imposing the boundary condition for r  R as
  
1 1 3 k i j
(4) hij =
r 4 Iij + 4 n n Ikj + n Iki
(2) (2) (2)

5 ni nj I (2) + 3 ni nj nk nl I (2) + 1  I (2) 5  nk nl I (2) 



8 kk 8 kl 8 ij kk 8 ij kl
+ r12 5 k (2) 1 i (2) j (2) 5 k l i (2) j (2)
12 n Iijk 24 (n Ijkk + n Iikk ) + 8 n n (n Ijkl + n Iikl )
7 ni nj nk I (2) + 5 ni nj nk nl nm I (2) + 11  nk I (2) 5  nk nl nm I (2) 
8 kll 8 klm 24 ij kll 8 ij klm

+ 23 nk (S_ ikj + S_ jki) 34 (ni S_ jkk + nj S_ ikk )

+ 2nk nl (ni S_ jkl + nj S_ ikl ) + 2ni nj nk S_ kll + 23 ij nk S_ kll + O(r 3 ):
(3.91)
It is veri ed that O(r 1 ) and O(r 2 ) parts satisfy the traceless and divergence-free
conditions respectively. It should be noted that (4) hij obtained in this way becomes
meaningless at the far zone because Eq.(3.53), from which (4) hij is derived, is valid
only in the near zone.
28 H. Asada and T. Futamase
3.4. The Radiation Reaction due to Quadrupole Radiation
This topic has been already investigated by using some gauge conditions in
previous papers 25); 79); 15) . However, if we use the combination of the conformal slice
and the transverse gauge, calculations are simpli ed.
3.4.1. conformal slice
In combination of the conformal slice 84) and the transverse gauge, Eq.(3.54)
becomes 4)
h (  ) = 1 @ Z  16vi vj 1  v2 
(5) ij
4 @  3 ij 
+ UU;ij 13 ij UU 3U;i U;j + ij U;k U;k d3 y
@ Z  _ i; j + _ j 2  _ k d3 y:
+ 41 @ (3.92)
(3) (3) ;i
3 ij (3) ;k
From a straightforward calculation, we nd 4) that the sum of the rst and second
lines becomes 2 I (3) ij and the third line becomes 6 I ij =5, where I ij = d I ij =dt
(3) (3) 3 3

and
I ij = Iij 31 ij Ikk : (3.93)
Thus, (5) hij in the near zone becomes
4 I (3) : (3.94)
(5) hij =
5 ij
Since hij has the transverse and traceless property, it is likely that (5) hij remains
the same for other slices. However it is not clear whether the TT property of hij
is satis ed even after the PN expansion is taken in the near zone and, as a result,
whether (5) hij is independent of slicing conditions or not. The fact that slicing
conditions never a ect (5) hij is understood on the ground that (4) ij does not depend
on slices, as already shown in Eq.(3.69).
Then the Hamiltonian constraint at the 2.5PN order, Eq.(3.42), turns out to be
(7) = 52 I (3) 1 (3)
ij U;ij = 5 I ij ;ij ; (3.95)
where  is the superpotential 21) de ned as
Z
= jx yjd y;
3
(3.96)
which satis es the relation  = 2U . From this, we nd (7) takes the following
form,
Z j j
(7) = 5 I ij ;i (xjx yy j ) d3 y
1 (3)

Z j !
1 j  ;i
= 5 I ij x U;i + jx yj d y :
(3) y 3
(3.97)
Post-Newtonian Approximation 29
Therefore, the lapse function at the 2.5PN order, (7) = 2(7) , is derived from U
and Ur , where Ur satis es 15)
Ur = 4 I (3) j
ij ;ix : (3.98)
Since the right-hand side of Eq.(3.58) cancels out, (6) K disappears if the (6) K does
not exist on the initial hypersurface, which seems reasonable under the condition that
there are no initial gravitational waves. Also, (6) i vanishes according to Eq.(3.50).
Hence, the quadrupole radiation reaction metric has the same form as that derived
in the case of the maximal slice 79); 15) .
From Eq.(3.29), the PN equation of motion becomes
v_ i + vj v;i j = P;i + U;i + Fi PN + Fi PN + Fi : PN + O( );
1 2 25
(3.99)
8

where Fi PN and Fi PN are, respectively, the 1PN and 2PN forces. Since the radi-
1 2

ation reaction potentials, hij and , are the same as those by Schafer (1985)
(5) (7)
and Blanchet, Damour and Schafer (1990) in which they use the ADM gauge, the
radiation reaction force per unit mass, Fi2:5PN  Fir , is the same as their force and
 
Fir = ((5) hij vj ) + vk v;kj (5) hij + (7) ;i
 Z k yk )(xl yl ) 
4 (3) j  4 (3) k j 2 (3) @ ( x
= 5 ( I ij v ) + 5 I ij v v;k + 5 I kl @xi (t; y) jx yj3 d y :
3

(3.100)
Since the work done by the force (3.100) is given by
Z
W  Fir vid x 3

 Z 
4 d
= 5 d I ij v v d x 15 I ij I
i j
(3) 3 (3) (3)ij ; (3.101)
we obtain the so-called quadrupole formula of the energy loss by averaging Eq.(3.101)
with respect to time as

dEN  = 1 D I (3) I (3)ij E + O(6): (3.102)
d 5 ij
3.4.2. Radiation reaction in other slice conditions
In this subsection, we do not specify the slice condition. In this case, the reaction
force takes the following form
Fir = 54 ( I (3) j  4 (3) k j
ij v ) + 5 I ij v v;k (7) ;i (6) _ + v (6) ;i v (6) ;j : (3.103)
i j j j i

Here, (7) corresponds to the slice condition. From Eq.(3.103), we obtain the work
done by the reaction force as
Z
W  Fir vid x 3
30 H. Asada and T. Futamase
 
Z
= d 3
xvi 4 I (3) vj  + 4 I (3) vj vl
5 ij 5 ij ;l 
(7) ;i
_ i j j
(6) + v (6) ;i
j i
v (6) ;j : (3.104)
Explicit calculations are done separately: For the rst and second terms of
Eq.(3.104), we obtain
4 Z d3 x I (3) vj vi = 4 d  I (3) Z d3 xvi vj  1 I (3) I (3)
5 ij 5 dt ij 5 ij ij
Z Z
2 I (3) d3 x(x)vk @ d3 y (y)(xi yi )(xj yj )
5 ij Z @xk jx yj3
2 I (3) d3 xv_ i vj ; (3.105)
5 ij
and
4 Z d3 x I (3) vj vl vi = 2 I (3) Z d3 xv
_ i vj : (3.106)
5 ij ;l 5 ij
As for the fourth term in the integral of Eq.(3.104), we nd the relation
_
i= 2 I (3) @ Z d3 y (xk yk )(xl yl ) ; (3.107)
(7) ;i
(6)
5 kl @xi jx yj3
which is given by Eqs.(3.50), (3.58) and (3.94).
Using Eqs.(3.104), (3.105), (3.106) and (3.107), we obtain
 Z 
d
W = 5 d I ij v v d x 15 I (3)
4 i j (3)ij
(3.108)
ij I :
(3) 3

This expression for W does not depend on the slice condition. However, this never
means that the value of W is invariant under the change of the slice condition, since
the meaning of the time derivative depends on the slice condition.
3.5. Conserved quantities
The conserved quantities are gauge-invariant so that, in general relativity, they
play important roles to characterize various systems described in di erent gauge
conditions. From the practical point of view, these are also useful for checking the
numerical accuracy in simulations. Thus, we show several conserved quantities in
the 2PN approximation.
Conserved Mass And Energy
In general relativity, we have the following conserved mass;
Z
M = d x:3
(3.109)
In the PN approximation,  de ned by Eq.(3.28) is expanded as 4)
  
1
 =  1 + 2 v + 3U
2

  
3 4 7 2 15 2 i i
+ 8 v + 2 v U + 4 U + 6(4) + (3) v + (6)  ) + O(7 ); (3.110)
Post-Newtonian Approximation 31
where (6)  denotes the 3PN contribution to  . This term (6)  will be calculated
later.
Next, we consider the ADM mass which is conserved. Since the asymptotic
behavior of the conformal factor becomes
 
M
= 1 + 2r + O r12 ;
ADM (3.111)
the ADM mass in the PN approximation becomes
Z
1
MADM = 2  d3 x
Z    
= d3 x 1 + v2 +  + 52 U + v4 + 13 2 v2U + v2 + P v2 + 52 U
  
+ 52 U 2 + 5(4) + 2(3) i vi + 161 (3) A~ij (3) A~ij 23 (3) K 2

+ (6) ADM + O( ) ;
7
(3.112)
where (6) ADM denotes the 3PN contribution. This term (6) ADM will be calculated
later.
Using these two conserved quantities, we can de ne the conserved energy as
follows;
EM M
Z ADM   
= d x 21 v +  12 U
3 2

 
+ 58 v4 + 3v2 U + v2  + P v2 + 52 U 45 U 2 (4) + (3) i vi
    
+ 161 (3) A~ij (3) A~ij 23 (3) K 2 + (6) ADM (6)  + O(c ) 7

 EN + E1PN + E2PN +    : (3.113)


We should note that the following equation holds
Z
Z
~
A ~
A d 3
x = 8 
Z
v i i d 3
x + 2 K 2 + 2U_ K d3 x; (3.114)
(3) ij (3) ij (3)
3 (3) (3)

where we used Eqs.(3.47) and (3.48). Then, we obtain the Newtonian and the rst
PN energies as Z  
EN =  12 v2 +  12 U d3 x; (3.115)
and
 
Z
E1PN = d x  58 v4 + 25 v2 U + v2  + P v2 + 2U
3 5 U 2 + 1 i vi 

2 2 (3)
1 _
+ 8 U (3) K : (3.116)
32 H. Asada and T. Futamase
E PN can be rewritten immediately in the following form used by Chandrasekhar ;
1
23)

Z    
E PN = d x 85 v + 52 v U + v  + P + 2U 52 U 12 vi qi ; (3.117)
1
3 4 2 2 2

where qi is the rst PN shift vector in the standard PN gauge , which turns out 23)

to be K = 0 in the (3+1) formalism and it satis es


(3)

qi = 16vi + U_ ;i: (3.118)


The total energy at the 2PN order E PN is calculated from the 3PN quantities
2

(6)  and ADM . In a straightforward manner, we obtain


(6)
Z
(6) M =   d x;(6)
3
(3.119)
where
 
(6)  = 165 v + 33
6
8 v U + v
4
5
2
+ 93 U + 3 i vi X + 6
8(4)
2
2
(3) (6) + 15U (4)
5 1 1
+ 2 U + 7 i vi U + 2 hij vi vj + 2 i i + i vi :
3
(3.120)
(3) (4) (3) (3) (5)

Next, we consider (6) ADM . The Hamiltonian constraint at O(8 ) becomes


(8) (4) hij (4) ;ij 21 (6) hij U;ij
1 (2 h
= 32 (4) kl;m (4) hkm;l + (4) hkl;m (4) hkl;m ) 2(6) 
 
1 A~ A~ 2 K K 1 U A~ A~ 2 K 2 ; (3.121)

4 (3) ij (5) ij 3 (3) (5) 16 (3) ij (3) ij 3 (3)
where we de ne (6)  as
 
 =  v 6
+ v 4
 + P + 21 U  + v2  13 U  + P  + 9 2 X + 20 U 2

(6)

 2 2  (4)

+  5(4) + 52 U 2 + 5(6) + 10U (4) + 54 U 3 + (4) hij vi vj


  
+ 2(3) i vi 2v2 +  + P + 13 2 U + 2 (5) i vi + i i : (3.122)
(3) (3)

Making use of the relations (4) hij;j = 0 and (6) hij;j = 0, we obtain
Z Z  
1 ~ ~ 2
(6) MADM = d x(6)  + 8 d y (3) Aij (5) Aij 3 (3) K (5) K
3 3

Z  
+ 321 d3 yU (3) A~ij (3) A~ij 23 (3) K 2 ; (3.123)
where we assumed (6) hij ! O(r 1 ) as r ! 1. Although this assumption must
be veri ed by performing the 3PN expansions which have not been done here, it
Post-Newtonian Approximation 33
seems reasonable in the asymptotically at spacetime. From (6) MADM and (6) M ,
we obtain the conserved energy at the 2PN order
E PN = MADM
2 (6) M
(6)
Z   
11
= d x 16 v + v  +  + 8 U
3 6 P4 47
   
+v 4 2 P 41
X + 6U  +  + 8 U + 2 v 5 i i 2
(4) (3)
 
+  5 + 4U 5 15 U 2 5U 3
(4)
2 2
 
(4)

1 i j i i P
+ 2 hij v v + 2 v  +  + 5U + v + 2 i i 1 i i
(4) (3) (5) (3) (3)
Z  
1 ~
+ 8 d y hij UU;ij + Aij Aij 3 K K :
3 ~ 2 (3.124)
(4) (3) (5) (3) (5)

R R
Here we used Eq.(3.41) and the relation, d3 x(6) = 4
1
 d xU
3
(6) , in order
to eliminate (6) .
Conserved linear momentum
When we use the center of mass system as usual, the linear momentum of the
system should vanish. However, it may arise from numerical errors in numerical
calculations. Since it is useful to check the numerical accuracy, we mention the
linear momentum derived from
I 
1
Pi = 8 rlim j i
!1 Kij n Kn dS
I 
1
= 8 rlim 4 ~ j 2 i
Aij n 3 Kn dS; (3.125)
!1
where the surface integrals are taken over a sphere of constant r. Since the asymp-
totic behavior of A~ij is determined by
 
~ 1 i j 2 l (3.126)
(3) Aij =
2 (3) ;j + (3) ;i 3 ij (3) ;l + O(r );
3

and
 
~ 1 i j 2 l ~TT (3.127)
(5) Aij =
2 (5) ;j + (5) ;i 3 ij (5) ;l + (5) Aij + O(r );
3

the leading term of the shift vector is necessary. Using the asymptotic behavior
derived from Eq.(3.48)

i = 27 lri 1 ni nj lj + O(r 2 ); (3.128)


(3)
2 r
we obtain Z
i j 2  l nj dS = 16l ; (3.129)
(3) ;j + (3) ;i
3 ij (3) ;l i
34 H. Asada and T. Futamase
R
where we de ned li = vi d3 x. Therefore the Newtonian linear momentum is
Z
PN i = d3 xvi : (3.130)
Similarly the rst PN linear momentum is obtained as
Z    
P1PN i = d 3
x vi v2 +  + 6U P i
+ + : (3.131)
(3)

We obtain P2PN i by the similar procedure as


Z    
P PN i= d3
xvi 2(3) ivi + 10 + 6U + v 2 P
+ 
2 (4)

67
+ 4 U + 10Uv + v X :
2 2 4
(3.132)

x4. Formulation for Nonaxisymmetric Uniformly Rotating Equilibrium


Con gurations
We now consider the construction of the spacetime involving a close binary neu-
tron stars as an application of PN approximation. It is assumed that the binary
stars are regarded as uniformly rotating equilibrium con gurations. Here, we men-
tion the importance of this investigation. To interpret the implication of the signal of
gravitational waves, we need to understand the theoretical mechanism of merging in
detail. When the orbital separation of binary neutron stars is  < 10GM=c2 , where M
is the total mass of binary neutron stars, they move approximately in circular orbits
because the timescale of the energy loss due to gravitational radiation is much longer
than the orbital period. However, when the orbital separation becomes 6 10GM=c2 ,
the circular orbit cannot be maintained because of instabilities due to the GR grav-
ity 59) or the tidal eld 62); 63) . As a result of such instabilities, the circular orbit of
binary neutron stars changes into the plunging orbit to merge. Gravitational waves
emitted at this transition may present us important information about the structure
of NS's since the location where the instability occurs will depend sensitively on the
equation of state (EOS) of NS 62); 63); 102) . Thus, it is very important to investigate
the location of the innermost stable circular orbit (ISCO) of binary neutron stars.
In order to search the ISCO, we can take the following procedure: First, ne-
glecting the evolution due to gravitational radiation, we construct equilibrium con-
gurations. Next, we take into account the radiation reaction as a correction to the
equilibrium con gurations. The ISCO is the orbit, where the dynamical instability
for the equilibrium con gurations occurs. Hence we shall present a formalism to
obtain equilibrium con gurations of uniformly rotating uid in 2PN order as a rst
step 5) , though in reality due to the conservation of the circulation for gravitational
radiation reaction, tidally locked con guration such as uniformly rotating uid is not
a good approximation.
4.1. Formulation
We shall use in this section the maximal slice condition Ki i = 0. As the spatial
gauge condition, we adopt the transverse gauge ~ij;j = 0 in order to remove the gauge
Post-Newtonian Approximation 35
modes from ~ij . In this case, up to the 2PN approximation, each metric variable is
expanded as 4)
= 1 + c12 U2 + c14 (4) + O(c 6 ); (4.1)
 2 
= 1 c12 U + c14 U2 + X + c16 (6) + O(c 7); (4.2)
i = 1 i + 1 i + O(c 7 );
c 3 (3) c 5 (5)
(4.3)
1
~ij = ij + c hij + O(c ): 5
(4.4)
4

In this section, we use 1=c instead of  as the expansion parameter because of its
convenience in numerical applications.
For simplicity, we assume that the matter obeys the polytropic equation of state
(EOS);
P = ( 1) = K ; (4.5)
where and K are the polytropic exponent and polytropic constant, respectively.
Up to the 2PN order, the four velocity is given by Eq.(3.35) 24); 4); 5) . Since we need
u0 up to 3PN order to obtain the 2PN equations of motion, we derive it here. Using
Eq.(3.35), we can calculate ( u0 )2 up to 3PN order as
( u0 )2 = 1 + 4 ~ ij ui uj
  
v 2
1 j j
= 1 + + 2 v + 4Uv + v + 2 4 1 j j + 8 j vj U
c c 2 4

(3)
c  
6 (3) (3) (3)

i j i i j j 15
+hij v v + 2 v + 4 v + 4 + 2 U 2X v + 8Uv + v 2 2 4 6
(5) (3) (4)

+O(c ); 7
(4.6)
where we used ~ ij = ij c hij + O(c ). Thus, we obtain u up to the 3PN order
4 5 0

as
   
u = 1 + c1 12 v + U + c1 38 v + 52 v U + 12 U + i vi X
0
2
2
4
4 2 2
(3)
  
+ c1 6
+ 12 j j + hij vivj + j vj + 5 j vj U 2UX
(6) (3) (3) (5) (3)
 
+ 32 (3) j vj + 2(4) + 6U 2 32 X v2 + 27 Uv 4
+ 5 v6 
8 16
+O(c ):7
(4.7)
Substituting PN expansions of metric and matter variables into the Einstein
equation, and using the polytropic EOS, we nd that the metric variables obey the
following Poisson equations 4) ;
U = 4;  
(4.8)
X = 4 2v2 + 2U + (3 2) ; (4.9)
36 H. Asada and T. Futamase
 
(4) = 2 v +  + 2 U ; 52
(4.10)
(3) i = 16vi U_ ;i; (4.11)
"   # 
 = 16 vi v2 + 2U
i +  + (3) i 4U;j (3) ;ji + (3) ;ij 2  k 
(5)
3 ij (3) ;k
2(4) _ ;i + 12 (U U_ );i + ((3) l U;l );i ; (4.12)
   
hij = UU;ij 13 ij UU 3U;iU;j + ij U;k U;k 16 vivj 13 ij v2
  
_ i + _ j 2  _ k 2 ( X + 2 ) 1  (X + 2 );
(3) ;j (3) ;i
3 ij (3) ;k (4) ;ij 3 ij (4)

"
(4#.13)
 
 = 4 2v + 2v 5U +  + (3
(6)
4 2
2)U + 4(4) + X + 4(3) i vi

hij U;ij 32 UU;l U;l + U;l (2 (4) X );l


 
1 i i j 2 k
+ 2 (3) ;j (3) ;j + (3) ;i 3 ij (3) ;k : (4.14)
Here, we consider the uniformly rotating uid around z -axis with the angular
velocity
, i.e.,
vi = ijk
j xk = ( y
; x
; 0); (4.15)
where we choose
j = (0; 0;
) and ijk is the completely anti-symmetric unit tensor.
In this case, the following relations hold;

@ +
@ Q =  @ +
@ Q =  @ +
@ Q = 0; (4.16)
@t @' @t @' i @t @' ij
where Q, Qi and Qij are arbitrary scalars, vectors, and tensors, respectively. Then,
the conservation law (2.18) can be integrated as 65)
Z
dP (4.17)
c +  + P = ln u + C;
0
2

where C is a constant. For the polytropic EOS, Eq.(4.17) becomes


" #
ln 1 + c ( K 1)  2
1
= ln u0 + C: (4.18)

Using Eq.(4.7), the 2PN approximation of Eq.(4.18) is written as


!
H 2Hc + 3Hc = v2 + U + c1 2Uv + v4
2 3 2 4

2 4 2
2
X+ (3) i vi

+ c14 + 12 i i + 4 i viU U6 + ivi v + 2


3

(6) (3) (3) (3) (3)


2
(4) v 2
Post-Newtonian Approximation 37
!
+ 15 1 6
4 U v + 2Uv + 6 v UX v X + (5) v + 2 hij v v
2 2 4 2 i i 1 i j

+C; (4.19)
where H = K 1=( 1), v2 = R2
2 and R2 = x2 + y2 . Note that Eq.(4.19)
can be also obtained from the 2PN Euler equation like in the rst PN case 22) . If
we solve the coupled equations (4.8)-(4.14) and (4.19), we can obtain equilibrium
con gurations of the non-axisymmetric uniformly rotating body.
In the above Poisson equations for metric variables, the source terms in the
Poisson equations for (3) i , (5) i , and hij fall o slowly as r ! 1 because these
terms behave as O(r 3 ) at r ! 1. These Poisson equations do not take convenient
forms when we try to solve them numerically as the boundary value problem. Hence
in the following, we rewrite them into convenient forms.
As for hij , rst of all, we split the equation into three parts as 4)
 
h(ijU ) = U U;ij 31 ij U 3U;iU;j + ij U;k U;k  4Sij(U ) ; (4.20)
 
h(ijS) = 16 vivj 13 ij v2 ; (4.21)
 
h(ijG) = (3) _;ji + (3) _;ij 23 ij (3) _;kk
 
2 (X + 2(4) );ij 13 ij (X + 2(4) ) : (4.22)

The equation for h(ijS ) has a compact source, and also the source term of h(ijU ) behaves
as O(r 6 ) at r ! 1, so that Poisson equations for them are solved easily as the
boundary value problem. On the other hand, the source term of h(ijG) behaves as
O(r 3) at r ! 1, so that it seems troublesome to solve the equation as the boundary
value problem. In order to solve the equation for h(ijG) as the boundary value problem,
we had better rewrite the equation into useful forms. As shown by Asada, Shibata
and Futamase 4) , Eq.(4.22) is integrated to give
Z Z Z
hijG = 2 @x@ i (vj ) jx yjd y + 2 @x@ j (vi )jx yjd y + ij jx yjd y
( ) 3 3 3

Z Z 
1 @ 2
3 3 @ 2
+ 12 @xi @xj jx yj d y + @xi @xj v + 3P 2 jx yjd y
2 U 3

 
2 Z v + 3P U=2
2

3 ij jx yj d3 y: (4.23)
Using the relations
 = (vj );j + O(c 2 );
v_ i = 0;
vi xi = 0; (4.24)
38 H. Asada and T. Futamase
Eq.(4.23) is rewritten as
 
(G) 7 i _ j _
hij = 4 x (3) Pj + x (3) Pi Qij Qji _ (T ) _ (T )
ij xk (3) P_k
" #
1 xk @ xj P_ Q_ (T )  + @ xi P_ Q_ (T ) 
8 @xi (3) k kj @xj (3) k ki
"    #
1 @
+ 2 @xi x Qj (I ) (I )
Qj + @xj x Q@ i (I )
Qi (I ) 2  Q(I ) ; (4.25)
3 ij
where
(3) Pi = 4vi; (4.26)
Q(ijT ) = 4vixj ; (4.27)
 
(I )
Q = 4 v + 3P 2 U ; 2 1 (4.28)
 
(I )
Qi = 4 v + 3P 2 U xi:
2 1 (4.29)
Therefore, h(ijG) can be deduced from variables which satisfy the Poisson equations
with compact sources.
The source terms in the Poisson equations (4.11) and (4.12) for (3) i and (5) i
also fall o slowly. However, we can rewrite them as 4)
 
i = 4 Pi 1 xi U_ q_i ; (4.30)
(3) (3)
2 
i = 4 P 1 2xi _ _ ; (4.31)
(5) i i
(5)
2 (4)

where
qi = 4x"i ; #
(4.32)
   
i i i j
(5) Pi = 4 v v + 2U +  + (3) + U;j (3) ;j + (3) ;i 3 ij (3) ;k
2 2 k

1 (UU
_ 1 l (4.33)
8  );i 4 ((3) U;l );i ;
i = 4 v2 +  + 52 U xi: (4.34)
Thus (3) i and (5) i can be obtained by solving the Poisson equations (4.32)-(4.34)
whose source terms fall o fast enough, O(r 5 ), for numerical calculation.
For later convenience, using the relation (3) Pi = izk qk
and Eq.(4.16), (3) i and
i
(5) may be written in the form with explicit
dependence as
  
i 1
4izk qk + 2 x U;' qi;' 
(3) ^i ;
i (4.35)
(3) =

  
i ^ 1 i (4.36)
(5) =
4(5) Pi + 2 2x (4) ;' i;' ;
Post-Newtonian Approximation 39
where
"   #  
(5) P^i = 4 izk xk v2 + 2U +  + (3) ^i + U;j (3) ^;ji + (3) ^;ij 32 ij (3) ^;kk
+ 18 (UU;' );i 14 ((3) ^k U;k );i : (4.37)
4.2. Basic equations appropriate for numerical approach
Although equilibrium con gurations can be formally obtained by solving Eq.(4.19)
as well as metric potentials, U , X , (4) , (6) , (3) i , (5) i and hij , they do not take
convenient forms for numerical calculation. Thus, we here change Eq.(4.19) into
forms appropriate to obtain numerically equilibrium con gurations.
In numerical calculation, the standard method to obtain equilibrium con gura-
tions is as follows 52); 72) ;
(1) We give a trial density con guration for .
(2) We solve the Poisson equations.
(3) Using Eq.(4.19), we give a new density con guration.
These procedures are repeated until a sucient convergence is achieved. Here, at (3),
we need to specify unknown constants,
and C . In standard numerical methods
52); 72)
, these are calculated during iteration by xing densities at two points; i.e.,
if we put 1 at x1 and 2 at x2 into Eq.(4.19), we obtain two equations for
and
C . Solving these two equations give us
and C . However, the procedure is not
so simple in the PN case:
is included in the source of the Poisson equations for
the variables such as X , (4) , (6) , i , (5) P^i , h(ijS ) , Q(ijT ) , Q(I ) and Q(iI ) . Thus, if we
use Eq.(4.19) as it is, equations for
and C become implicit equations for
. In
such a situation, the convergence to a solution is very slow. Therefore, we transform
those equations into other forms in which the potentials as well as Eq.(4.19) become
explicit polynomial equations in
.
First of all, we de ne q2 , q2i , q4 , qu , qe and qij which satisfy
q2 = 4R2; (4.38)
q2i = 4R2xi ; (4.39)
q4 = 4R4; (4.40)
qu = 4U; (4.41)
qe = 4; (4.42)
qij = 4x x : i j (4.43)
Then, X , (4) , Q(I ) , Q(iI ) , i , (5) P^i , Q(ijT ) , and h(ijS ) are written as
X = 2q
2qu (3  2)qe;
2
2
(4.44)
(4) = 21 q
+ qe + 52 qu ;
2
2
(4.45)
Q I = q
+ 3( 1)qe 12 qu  q
+ Q I ;
( )
2
2
2
2 ( )
0 (4.46)
QiI = q i
+ Q Ii ;
( )
2
2 ( )
0 (4.47)
40 H. Asada and T. Futamase
i = q i
+  i ;
2
2
0 (4.48)
P^i = izk q k
+ P i ;
(5) 2
2
(5) 0 (4.49)
QijT = izl qlj
;
( )
(4.50)
 
hijS = 4
izk jzlqkl 31 ij q ;
( ) 2
2 (4.51)
where Q(0Ii ) , 0i and (5) P0i satisfy
  
(I ) 1
Q0i = 4 3P 2 U x = 4 3( i 1) 1 U xi ; (4.52)
 
2
0i = 4  + 2 U xi; 5 (4.53)
"    # 
(5) P0i = 4 izk 2U +  + (3) ^i + U;j (3) ^;ji + (3) ^;ij 23 ij (3) ^;kk
xk
+ 18 (UU;' );i 41 ((3) ^k U;k );i  4Si(P ) : (4.54)
Note that (5) i and h(ijG) are the cubic and quadratic equations in
, respectively, as
i
(5) = (5)
i (A)
+ i (B)
3 ;
(5) (4.55)
h(ijG) = h(ijA) + h(ijB)
2 ; (4.56)
where
   
i (A) 1 i 5 (4.57)
(5) = 4(5) P0i + 2 x qe + 2 qu 0i;' ;
 ;'

i (B ) 1 i (4.58)
(5) = 4izk q2k + 2 x q2;' q2i;' ;
"     #
hijA
( ) 1 @ i (I ) (I )
= 2 @xj x Q0 Q0i + @xi x Q0 Q0j @ j (I ) (I ) 4  Q(I ) ;
3 ij 0
" (     )
(B ) 1 @ i @
hij = 2 @xj x q2 q2i + @xi x q2 q2j 3 ij q2 j 4

7 xi  q + xj  q 
k
4 ( jzk k;' izk k;' izk qkj;' jzk qki;' + ij x kzl ql
   )#
1 @ @
+ 8 x @xi x kzl ql;' kzl qlj;' + @xj x kzl ql;' kzl qli;' : (4.59)
k j i

Finally, we write (6) as


(6) = (6) 0 + (6) 2

2
2q4
4 ; (4.60)
where (6) 0 and (6) 2 satisfy
"    #
 = 4 3
(6) 0 2 U 3 4 qe + 3qu
Post-Newtonian Approximation 41
 
hijU ) + h(ijA)
(
U;ij 3 UU U + U @  9 q + (3 + 1)q 
2 ;l ;l ;l @xl 2 u e
 4S ( 0 ) ;   
(4.61)
 = 8R 5U +  + 2 ^'
(6) 2
2
4izk jzl qkl 43 ij q2 + h(ijB) U;ij
(3)
 
1 ^i ^i ^j 2 ^
+3q2;l U;l + 2 (3) ;j (3) ;j + (3) ;i 3 ij (3) ;kk

 4S ( 2 ) : (4.62)
Using the above quantities, Eq.(4.19) is rewritten as

H 2Hc + 3Hc = A + B
+ D
+ 6Rc
+ C;
2 3 6

2 4
2 4
4
6
(4.63)
where
    
A = U + c1 2qu + (3 2)qe + c1 U6 + U 2qu + (3 2)qe ;
3

2 4 (6) 0

  
R 2
1 ^
B = 2 + c 2R U + 2q + ' + c 2 1 + 12 ^i ^i + 4 ^'U
2 2 (3) 4 (6) 2 (3) (3) (3)
 
9 15 2 A 1 2 U
+(3 1)qe R + 2 qu R + 4 U R + 2q U + ' + 2 h'' + h'' ; A 2 2 ( ) ( ) ( )
2 (5)

  
R 4
1 ^ 7 B 1 B
D = 4c + c 2q + ' R + 3 q R + 2UR + ' + 2 h'' + 4R qRR :
2 2 4 ( ) ( ) 2
2 4 4 (3) 2 (5)

(4.64)
Note that in the above, we use the following relations which hold for arbitrary vector
Qi and symmetric tensor Qij ,
Q' = yQx + xQy ;
Q'' = y Qxx 2xyQxy + x Qyy ;
2 2

R QRR = x Qxx + 2xyQxy + y Qyy :


2 2 2
(4.65)
We also note that source terms of Poisson equations for variables which appear in
A, B and D do not depend on
explicitly. Thus, Eq.(4.63) takes the desired form
for numerical calculation.
In this formalism, we need to solve 29 Poisson equations for U , qx , qy , qz , (5) P0x ,
(I ) (I ) (I ) (U ) (U ) (U ) (U ) (U )
(5) P0y , 0x , 0y , Q0x , Q0y , Q0z , q2 , q2x , q2y , q2z , qu , qe , hxx , hxy , hxz , hyy , hyz ,
qxx, qxy , qxz , qyz , (6) 0, (6) 2 and q4. In Table 1, we show the list of the Poisson
equations to be solved. In Table 2, we also summarize what variables are needed
to calculate the metric variables U , X , (4) , (6) , (3) i , (5) i , h(ijU ) , h(ijS ) , h(ijA) and
h(ijB) . Note that we do not need (5) P0z , 0z , and qzz because they do not appear in
(U )
any equation. Also, we do not have to solve the Poisson equations for hzz and qyy
(U ) (U ) (U )
because they can be calculated from hzz = hxx hyy and qyy = q2 qxx.
42 H. Asada and T. Futamase
In order to derive U , qi, q2 , q2i , q4 , qe and qij , we do not need any other potential
because only matter variables appear in the source terms of their Poisson equations.
On the other hand, for qu , Q(0Ii ) , 0i and h(ijU ) , we need the Newtonian potential U ,
and for (5) P0i , (6) 0 and (6) 2 , we need the Newtonian as well as PN potentials.
Thus, U , qi, q2 , q2i , q4 , qe and qij must be solved rst, and then qu, Q(0Ii ) , 0i , h(ijU ) ,
(5) P0i and (6) 2 should be solved. (6) 0 must be solved after we obtain qu because
its Poisson equation involves qu in the source term. In Table 1 we also list potentials
which are included in the source terms of the Poisson equations for other potentials.
The con guration which we are most interested in and would like to obtain
is the equilibrium state for BNS's of equal mass. Hence, we show the boundary
condition at r ! 1 for this problem. When we consider equilibrium con gurations
for BNS's where the center of mass for each NS is on the x-axis, boundary conditions
for potentials at r ! 1 become
Z xZ
1
U = d x + O(r );
3 3
q = n x2d3 x + O(r 4);
rZ xr Z 2
y
q = 1r R d x + O(r );
2
2 3 3
qy = nr y d x + O(r );
2
2 3 4

Z zZ
qe = 1r d x + O(r );3 3
qz = nr z d x + O(r );
2
2 3 4

Z Z
qu = 1r Ud x + O(r );3 3
q = 1r R d x + O(r ); (4.66)
4
4 3 3

xZ yZ
(5) P x = nr SxP xd x + nr SyP yd x + O(r );
0 2
( ) 3
2
( ) 3 3

xZ yZ
(5) P y = nr SyP xd x + nr SyP yd x + O(r );
0 2
( ) 3
2
( ) 3 3
(4.67)

xZ  
 x = nr x  + 52 U d x + O(r );
0 2
2 3 4

yZ  
 y = nr y  + 25 U d x + O(r );
0 2
2 3 4
(4.68)

xZ   xZ
Q Ix
( ) n
= r x 3( 2 1 n
1) 2 U d x + O(r ); q2x = r2 R2 x2 d3 x + O(r 4 );
3 4
0 2
yZ   yZ
I
( ) n
Q y = r y 3( 2 1 n
1) 2 U d x + O(r ); q2y = r2 R2 y2 d3 x + O(r 4 );
3 4
0 2
yZ   zZ
I
( ) n
Q z = r z 3( 2 1 n
1) 2 U d x + O(r ); q2z = r2 R2 z 2 d3 x + O(r 4 );
3 4
0 2

(4.69)

hxxU ) =
( 1 Z S (U ) d3 x + O(r 3 ); hxyU ) =
( 3nx ny Z S (U ) xyd3 x + O(r 5 );
r xx r3 xy
Post-Newtonian Approximation 43
Z
hyyU )
( 1
= r Syy (U ) 3
d x + O(r 3); hxzU ) =
( 3nx nz Z S (U ) xzd3 x + O(r 5 );
r3 xz
3 n y nz Z
(U )
hyz = r3 Syz(U ) yzd3 x + O(r 5); (4.70)

Z x yZ
qxx = 1r x d x + O(r );
2 3
qxy = 3nr n x y d x + O(r );
3
3
2 2 3 5

x zZ y zZ
qxz = 3nr n x z d x + O(r ); qyz = 3nr n y z d x + O(r ); (4.71)
3
2 2 3 5
3
2 2 3 5

Z Z
1 ( 0 ) 3 1 ( 2 ) 3 .
(6) 0 =
r S d x + O(r ); (6) 2 =
r S d x + O(r ); (4 72)
3 3

where ni = xi =r. We note that Si(P ) ! O(r 5 ), Sij(U ) ! O(r 6 ), S ( 0 ) ! O(r 4 )


and S ( 2 ) ! O(r 4 ) as r ! 1, so that all the above integrals are well de ned.
We would like to emphasize that from the 2PN order, the tensor part of the
3-metric, ~ij , cannot be neglected even if we ignore gravitational waves. Recently,
Wilson and Mathews 100) , Wilson, Mathews and Marronetti 101) presented numerical
equilibrium con gurations of binary neutron stars using a semi-relativistic approx-
imation, in which they assume the spatially conformal at metric as the spatial
3-metric, i.e., ~ij = ij . Thus, in their method, a 2PN term, hij , was completely
neglected. However, it should be noted that this tensor potential plays an important
role at the 2PN order: This is because they appear in the equations to determine
equilibrium con gurations as shown in previous sections and they also contribute to
the total energy and angular momentum of systems. This means that if we performed
the stability analysis ignoring the tensor potentials, we might reach an incorrect con-
clusion. If we hope to obtain a general relativistic equilibrium con guration of binary
neutron stars with a better accuracy (say less than 1%), we should take into account
the tensor part of the 3-metric.
In our formalism, we extract terms depending on the angular velocity
from
the integrated Euler equation and Poisson equations for potentials, and rewrite the
integrated Euler equation as an explicit equation in
. This reduction will improve
the convergence in numerical iteration procedure. As a result, the number of Poisson
equations we need to solve in each step of iteration reaches 29. However, source terms
of the Poisson equations decrease rapidly enough, at worst O(r 4 ), in the region far
from the source, so that we can solve accurately these equations as the boundary
value problem like in the case of the rst PN calculations.
The formalism presented here will be useful to obtain equilibrium con gurations
for synchronized BNS's or the Jacobi ellipsoid. In this context, Bonazzola, Frieben
and Gourgoulhon (1996) obtained an approximate nonaxisymmetric neutron star
by perturbing a stationary axisymmetric con guration. However, they do not solve
the exact 3D Einstein's equation. Thus, it is interesting to examine their conclu-
sion on the transition between equilibrium con gurations, which are approximately
ellipsoids, by our PN methods.
44 H. Asada and T. Futamase
x5. Gravitational Waves from post-Newtonian sources
In the post-Minkowskian approximation, the background geometry is the Minkowski
spacetime where linearized gravitational waves propagate 89); 11); 12) . The corrections
to propagation of gravitational waves can be taken into account if one performs the
post-Minkowskian approximation up to higher orders. In fact, Blanchet and Damour
obtained the tail term of gravitational waves as the integral over the past history of
the source 12); 13) . They introduced an unphysical complex parameter B , and used
the analytic continuation as a mathematical device in order to evaluate the so-called
log term in the tail contribution. The method is powerful, but it is not easy to see
the origin of the tail term. Will and Wiseman (1996) have also obtained the tail
term from the mass quadrupole moment, by improving the Epstein-Wagoner for-
malism 41). Nakamura 70) and Schafer 80) have obtained the quadrupole energy loss
formula including the contribution from the tail term by studying the wave propa-
gation in the Coulomb type potential. This indicates clearly that the origin of the
tail term is due to the Coulomb type potential generated by the mass of the source.
In the following treatment we calculate the waveform from the slow motion source
and show explicitly how the tail term originates from the di erence between the at
light cone and the true one which is due to the mass of the source GM=c2 as the
lowest order correction.
5.1. Gravitational waves in Coulomb type potential
We wish to clarify that the main part of the tail term is due to the propagation of
gravitational waves on the light cone which deviates slightly from the at light cone
owing to the mass of the source. For this purpose, we shall work in the harmonic
coordinate since the deviation may be easily seen in the reduced Einstein's equation
in this coordinates.
( h )h ; = 16 + h  ; h  ; : (5.1)
Since we look for 1=r part of the solution, the spatial derivative is not relevant in
the di erential operator so that it is more convenient for our purpose to transform
Eq.(5.1) into the following form
(2 h 00 @0 @0 )h  = 16S  ; (5.2)
where we de ned S  as
S  =  161 (h  ; h  ; 2h 0i h ;0i h ij h ;ij ): (5.3)
Substituting the lowest order expression for h 00 derived in section 2, we nally obtain
our basic equation to be solved.
   2 
1+ 4 M @ +  h  = ~ ; (5.4)
r @t 2

where the e ective source ~ is de ned as


 

~ = S  1 
h 00 4 M  ;00 :
h (5.5)
16 r
Post-Newtonian Approximation 45
At the lowest order, we obtain (4) ~ = (4)  .
The solution for (5.4) may be written down by using the retarded Green's func-
tion in the following form.
Z
h  (x) = dx0G(+) 0  0
M (x; x )~ (x ): (5.6)
The retarded Green's function is de ned as satisfying the equation;
2M G(+)
M (x; y) =  (x y);
4
(5.7)
with an appropriate boundary condition. We also de ned the following symbol for
the di erential operator appearing in our basic equation.
   2 
2 = 1+ 4 M @ + ; (5.8)
M r @t 2

where  is the Laplacian in the at space.


The Green's function G0 for M = 0 i.e. in the Minkowski spacetime is well
known. We shall present the detailed derivation of the Green function in appendix
B, and present here the result
X i Z 
GM (x; x0 ) =
(+)
e l d!sgn(!) + ! l m (x) S ! l m  (x0 )(r r0 )
lm 
+ S ! l m (x) + ! l m  (x0 )(r0 r) ; (5.9)
where  denotes the complex conjugate, sgn(!) denotes a sign of ! and we de ned
+ ! l m (x) and S ! l m (x) as
s
+! l m (x) = j!j e i!t  1
ul (; )Yl m ;
(+)

s
2
S ! l m (x) = j2!j e i!t  1
Fl (; )Yl m; (5.10)
where  = !r and u(+) l is a spherical Coulomb function satisfying the outgoing wave
condition at null in nity, and F` is also a spherical Coulomb function which is regular
at the origin.
Since we study an asymptotic form of the waves, we use Eqs.(B .7) and (B .8) in
appendix B to obtain the asymptotic form of the Green function as
lZ
G(+) ( x; x 0 ) = 1 X ( i) d!cl eil e i!(t r 2M ln 2M! t ) (!r0 )l Ylm (
)Y  (
0 )
0

M r lm 2 lm
+O(r 2 ) for r ! 1: (5.11)
In addition, for slow motion sources, we evaluate the asymptotic form of the Green
function up to O(M!) as
l Z   l 1 
0 1 X ( i )
GM (x; x ) = r 2(2l + 1)!! d! 1 + M! + 2iM! ln 2M!
(+)
X
+C
lm s=1 s
46 H. Asada and T. Futamase

+O(M ! ) e i!(t r t ) (!r0 )l Ylm (
)Ylm
2 2 0
 (
0 ) + O(r 2 )

=1
X ( i)l Z d!1 + 2M!i X l 1
+ ln 2 M

+  sgn(!)
r lm 2(2l + 1)!! s=1 s 2
 
+i(ln j!j + C ) + O(M 2 !2 ) e i!(t r t ) (!r0 )l Ylm (
)Ylm (
0 ) 0

+O(r 2 ); (5.12)
where C is Euler's number. In deriving the above expression we have used the
following expansion for cl and l in M!.
cl = 1 + M! + O(M 2 !2 ) ;
(2l + 1)!!
 Xl 1
 = 2M! C l + O(M 2 !2 ): (5.13)
s=1 s
Now we apply the formula 49); 16)
Z Z1
dve ln v + i dv ei!v = 2 sgn(!) i(ln j!j + C );
1
! i!v
v (5.14)
0 1

to Eq.(5.12), then we obtain


l
G(+) ( x; x 0 ) = 1 X ( i)
M r lm 2(2l + 1)!!
  X 
Z
 d! 1 2M
l 1
+ ln 2 M d + 2M Z 1 dvei!v ln v d2
s dt dt2
Z 1 s=1
0
 
ei!v d + O(M ! )
+2M dv 2 2
v dt 1

e i! t r t (!r ) Ylm (
)Ylm (
0 ) + O(r ):
( 0 l 0
) (5.15) 2

It is worthwhile to point out that ln 2M in Eq.(5.15) can be removed by using the


freedom to time translation.
Here, we assume the no incoming radiation condition on the initial hypersurface
so that we may take
!1 e
vlim
i!(t r v t ) ln v ! 0: 0
(5.16)
Thus we can make the following replacement
Z1 Z1 Z 1 
dv ei!v ! i! dvei!v ln v = dvei!v ln v d: (5.17)
1 v 1 1 dt
Inserting Eq.(5.17) into Eq.(5.15), we nally obtain the desired expression for the
retarded Green's function
l Z  X l 1 
0 1 X ( i )
GM (x; x ) = r 2(2l + 1)!! d! 1 + 2M
(+)
ln 2M dtd
lm s=1 s
Post-Newtonian Approximation 47
Z 1  
dvei!v ln v dtd + O(M ! )
2
+2M 2
2 2

e i!(t r t )0
(!r0 )l Ylm (
)Ylm
 (
0 ) + O(r ) 2

= 1r part of
  
G (x; x0 ) + 2M dtd
2 XZ v +X
dv ln 2M
l 1
Gl m (t v; x; x0 )
0 2
lm s s =1
0


+O(M 2 ) + O(r 2 ); (5.18)
where we de ned the spherical harmonic expansion coecient of the at Green's
function as follows.
Z
G0 (x; x ) = Yl m(
) d
0 G0(x; x0 )Yl m (
0):
l m 0 (5.19)
As a result, we obtain the waveform generated by the linear part of the e ective
source which corresponds to (C1) by Blanchet 7) .
1 1  Z    X
l 1
= 4P M~ pqL 2(t r) + 2M dtd 2 v
X 2 2
hTT
ij r ijpq l! nL 2 dv ln 2M + s
l=2 

s =1

~ 2 l
MpqL 2(t r v) l + 1 naL 2 ab(p S~q)bL 2(t r)
2 Z    X l 1 1 
d v ~
+2M dt2 dv ln 2M + s ab(p Sq)bL 2 (t r v)
 s=1
+O(M 2 ) + O(r 2 ); (5.20)
where Pijpq is the transverse and traceless projection tensor and parentheses denote
symmetrization. M~ pqL 2 and S~pqL 2 are the mass and current multipole moments
generated by the full nonlinear e ective source ~ . They take the same forms as in
Thorne (1980).
5.2. Comparison with the previous work
By using the post-Minkowskian approximation, Blanchet obtained the radiative
mass moment and radiative current moment as 7)
Z1    
ULl
( )
(u) = MLl ( )
(u) + 2GM dvMLl ( +2) v
(u v) ln P + l + O(G M ); 2 2

0
(5.21)
and
Z1    
VLl
( )
(u) = SLl ( )
(u) + 2GM dvSLl ( +2) v 0
(u v) ln P + l + O(G M ); 2 2

0
(5.22)
48 H. Asada and T. Futamase
where their moments ML and SL do not contain the nonlinear contribution outside
the matter, P is a constant with temporal dimension, and l and 0l are de ned as
l 2
l = 1s + l(2ll++1)(5ll ++ 42) ;
X 2
(5.23)
s=1
and l 1
0l = 1s + l(ll + 11) :
X
(5.24)
s=1
In black hole perturbation, the same tail corrections with multipole moments induced
by linear perturbations have beenPobtained 77) . Compared with Eqs.(5.21)-(5.24),
Eq.(5.20) shows that log term and 1=s in the radiative moments (5.21) and (5.22)
originate from propagation on the slightly curved light cone determined by Eq.(5.8).
That is to say, log term is produced by propagation of gravitational waves in the
Coulomb type potential in the external region. It is worthwhile to point out the
following fact: Only the log term has a hereditary property expressed as the integral
over the past history of the source, since the constants l and 0l represent merely
instantaneous parts after performing the integral under the assumption that the
source approaches static as the past in nity.
However, Eq.(5.20) does not apparently agree with Eqs.(5.21) and (5.22), be-
cause the de nition of the moments fM~ L ; S~L g and fML ; SL g are di erent. We shall
show the equivalence between our expression and that of Blanchet (1995), by calcu-
lating the contribution from nonlinear terms like M  ML or M  SL . It is noteworthy
that the luminosity of gravitational waves obtained by the present approach agrees
at the tail term i.e. O(c 3 ) with that by the post-Minkowskian approximation.
1
L = (l +(l1)(1) l + 2) 1
X (l+1) (l+1)

l l !(2 l + 1)!! < UL UL >


l=2
1
+ 4(ll(l +1)2) (l + 1)!(2
(l+1) (l+1)
X 1 (5.25)
l=2 l + 1)!! VL VL > :
<
5.3. Contributions from the nonlinear sources
In order to evaluate the waveform produced by the nonlinear sources in ~ , it
is enough to use the at Green's function
XZ 
G0(x; x0 ) = i !
d!  e i!t h (!r)Y (
)ei!t j (!r0 )Y  (
0 )(r
0
r0 )
l lm l lm
lm 
+e i!t j (!r)Y (
)ei!t h (!r0 )Y  (
0 )(r0
0
r) :
l lm l lm
(5.26)
Since the nonlinear sources have the form of either M  ML or M  SL, we have
to treat the following type of retarded integral
   q
2 1 @i( 1r )@^Q F (tr r) = ( )q+1 2j(jq!(+q j )!j )!
X
j =0
Post-Newtonian Approximation 49
 
2 1 (q j()t r):
n^ iQ + 2q q+ 1 i<aq n^ Q
rj+3 F
1
1>

(5.27)
The evaluation can be made by using the formula which will be proved in appendix
C.
  X
1 (l + n + 1)! ( k + l + 3 + 2n ) 
2 1 n^rkL F (t r) = 2l+1 lim
!0 n=0 n!(2l + 2n + 2)!
 n^ L (t r) + O(r 2):
(k 3)

r F (5.28)
Although some terms of nonlinear sources may produce disastrous divergence in
Eq.(5.28), we expect these divergent parts cancel out in total, which is explicitly
demonstrated in the appendix.
Using the above formula we nd
  
2 1 @ ( 1 )@^ F (t r)
i r Q r
 
)q
= 2((q + 1) n^ iQ q+1 1 (q)(t r) + O(r 2 ); (5.29)
2q + 1 i<a1 n^ Q 1> rF
which is same with the expression (C2) obtained by Blanchet 7) .
Applying Eqs.(5.29) to the nonlinear source ~ , we can evaluate its contribution
to the waveform. Together with Eq.(5.20), we obtain the total waveform as
4P X 1 1 
d2 Z dvln v  +  
hTT
ij = n M
r ijpq l=2 l! L 2 pqL 2 ( t r ) + 2M dt2 2M l
 
MpqL 2(t r v) l +2l 1 naL 2 ab(p Sq)bL 2(t r)
2 Z     
d v 0
+2M dt2 dv ln 2M + l ab(p Sq)bL 2 (t r v)

+O(M ) + O(r 2 ):
2
(5.30)
In this section, we have derived the formula (5.30) for gravitational waves in-
cluding tail by using the formula (5.20) and (5.29). We would like to emphasize two
points on the derivation: First, in deriving Eq.(5.20), spherical Coulomb functions
are used, since we use the wave operator (5.8) which take account of the Coulomb-
type potential M=r. As a consequence, ln(v=2M ) appears naturally in Eq.(5.30).
This is in contrast with Blanchet and Damour's method, where an arbitrary con-
stant with temporal dimension P appears in the form of ln(v=P ). Our derivation
shows that the main part of the tail, which needs the past history of the source
only through ln v, is produced by propagation in the Coulomb-type potential. Our
method might make it easy to clarify conditions for wave formula including tail. This
remains as a future work.
50 H. Asada and T. Futamase
The second point relates with physical application: At the starting point of
our derivation, the Fourier representation in frequency space has been used. Such
a representation seems to simplify the calculation of gravitational waveforms from
compact binaries in the quasi-circular orbit, since such a system can be described by
a characteristic frequency. Applications to physical systems will be also done in the
future.
x 6. Conclusion
We have discussed various aspects on the post-Newtonian approximation mainly
based on our own work. After presenting the basic structure of the PN approximation
in the framework of Newtonian limit along a regular asymptotic Newtonian sequence,
we reformulated it in the appropriate form for numerical approach. For this purpose
we have adopted the (3+1) formalism in general relativity. Although we restricted
ourselves within the transverse gauge in this paper, we can use any gauge condition
and investigate its property relatively easily in the (3+1) formalism, compared with
in the standard PN approximation performed so far 21) - 25) . Using the developed
formalism, we have written down the hydrodynamic equation up to 2.5PN order. For
the sake of an actual numerical simulation, we consider carefully methods to solve
the various metric quantities, especially, the 2PN tensor potential (4) hij . We found it
possible to solve them by using the numerical methods familiar in Newtonian gravity.
Thus, the formalism discussed in this paper will be useful in numerical applications.
As an example of the application, we have presented the formalism for constructing
the equilibrium con guration of nonaxisymmetric uniformly rotating uid. We have
also discussed the propagation of gravitational waves explicitly taking into account of
the deviation of the light cone from that of the at spacetime and obtained essentially
the same result with that of Blanchet and Damour for the waveform from post-
Newtonian systems.
Acknowledgements We would like to thank Y. Kojima, T. Nakamura, K.
Nakao, K. Oohaka, M. Sasaki, B. F. Schutz, M. Shibata and T. Tanaka for fruitful
discussion. Part of this work is based on the collaboration with M. Shibata. This
work is supported in part by the Japanese Grant-in-Aid for Science Research fund
of Ministry of Education, Science and Culture No.09640332 (T.F.) and in part by
Soryushi Shogakukai (H.A.).

References
1) A. Abramovici et al. Science, 256 (1992), 325.
2) J. L. Anderson and T. C. Decanio, Gen. Rel. Grav. 6 (1975), 197.
3) R. Arnowitt, S. Deser and C. W. Misner, In An Introduction to Current Research, ed. L.
Witten, (Wiley, New York, 1962).
4) H. Asada, M. Shibata and T. Futamase, Prog. Theor. Phys. 96 (1996), 81.
5) H. Asada and M. Shibata, Phys. Rev. D54 (1996), 4944.
6) L. Blanchet, Phys. Rev. D47 (1993), 4392.
7) L. Blanchet, Phys. Rev. D51 (1995), 2559.
8) L. Blanchet, Phys. Rev. D54 (1996), 1417.
Post-Newtonian Approximation 51
9) L. Blanchet and T. Damour, C. R. Acad. Sci. Paris, 298 (1984a), 431.
10) L. Blanchet and T. Damour, Phys. Lett, 104A (1984), 82.
11) L. Blanchet and T. Damour, Phil. Trans. R. Soc. London. A320 (1986), 379.
12) L. Blanchet and T. Damour, Phys. Rev. D37 (1988), 1410.
13) L. Blanchet and T. Damour, Phys. Rev. D46 (1992), 4304.
14) L. Blanchet, T. Damour, B. R. Iyer, C. M. Will and A. G. Wiseman, Phys. Rev. Lett.
74 (1995), 3515.
15) L. Blanchet, T. Damour and G. Schafer, Mon. Not. R. Astr. Soc. 242 (1990), 289.
16) L. Blanchet and G. Schafer, Class. Quant. Grav. 10 (1993), 2699.
17) S. Bonazzola, J. Frieben and E. Gourgoulhon, Astrophys. J. 460 (1996), 379.
18) C. Bradaschia et al., Nucl. Instrum. Method Phys. Res. Sect. A289 (1990), 518.
19) R. Blandford and S. A. Teukolsky, Astrophys. J. 205 (1969), 55.
20) W. L. Burke, J. Math. Phys. 12 (1971), 401.
21) S. Chandrasekhar, Astrophys. J. 142 (1965), 1488.
22) S. Chandrasekhar, Astrophys. J. 148 (1967), 621.
23) S. Chandrasekhar, Astrophys. J. 158 (1969a), 45.
24) S. Chandrasekhar and Y. Nutku, Astrophys. J. 158 (1969), 55.
25) S. Chandrasekhar and F. P. Esposito, Astrophys. J. 160 (1970), 153.
26) C. Cutler, T. A. Apostolatos, L. Bildsten, L. S. Finn, E. E. Flanagan, D. Kenne ck,
D. M. Markovic, A. Ori, E. Poisson, G. J. Sussman, K. S. Thorne, Phys. Rev. Lett. 70
(1993), 2984.
27) T. Damour, In Gravitaional Radiation, ed. N. Deruelle and T. Piran (Les Heuches; North-
Holland; 1982) p. 59.
28) T. Damour and N. Deruelle, Phys. Lett. 87A (1981a), 81.
29) T. Damour and N. Deruelle, C. R. Acad. Sci. Paris 293 (1981b), 537.
30) T. Damour and N. Deruelle, C. R. Acad. Sci. Paris 293 (1981c), 877.
31) T. Damour and J. H. Taylor, Astrophys. J. 366 (1991), 501.
32) P. D. D'Eath, Phys. Rev. D11 (1975a), 1387.
33) P. D. D'Eath, Phys. Rev. D11 (1975b), 2183.
34) W. G. Dixon, In Isolated gravitating systems in general relativity, ed. J. Ehlers (North-
Hollan, Amsterdam, 1976), p. 156.
35) J. Ehlers, A. Rosenblum, J. N. Golberg and P. Havas, Astrophys. J. 208 (1976), L77.
36) A. Einstein, Preuss. Akad. Wiss. Berlin, 47 (1915), 831.
37) A. Einstein, L. Infeld and B. Ho man, Ann. Math. 39 (1938), 65.
38) J. Ehlers, In Grun agenprobleme der modernen Physik, ed J. Nitsch et. al. (Bibliographis-
ches, Mannheim, 1981).
39) J. Ehlers, Class. Quant. Grav. 14 (1997), A119.
40) R. Epstein, Astrophys. J. 216 (1977), 92.
41) R. Epstein and R. V. Wagoner, Astrophys. J. 197 (1975), 717.
42) L. S. Finn, Phys. Rev. D53, 2878 (1996).
43) V. Fock, The Theory of Space Time and Gravitation (Pergamon Press, Oxford, 1959).
44) T. Futamase, Phys. Rev. D28 (1983), 2373.
45) T. Futamase, Phys. Rev. D36 (1987), 321.
46) T. Futamase, In proceedings of Gtavitational Astronomy (in Japanese), ed. T. Nakamura,
(1992), p. 340.
47) T. Futamase and B. F. Schutz, Phys. Rev. D28, (1983) 2363.
48) T. Futamase and B. F. Schutz, Phys. Rev. D32 (1985), 2557.
49) I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and Products (New York;
Academic 1980).
50) L. P. Grischuk and S. M. Kopejkin, Sov. Astron. Lett. 9 (1983), 230.
51) R. Geroch, Commun. Math. Phys. 13 (1969), 180.
52) I. Hachisu, Astrophys. J. Suppl. 62 (1986), 461.
53) R. A. Hulse and J. H. Taylor, Astrophys. J. Lett. 195 (1975), L51.
54) R. A. Isaacson, J. S. Welling and J. winicour, Phys. Rev. Lett. 53, (1984), 1870.
55) R. E. Kates, Phys. Rev. D22 (1980a), 1853.
56) R. E. Kates, Phys. Rev. D22 (1980b), 1871.
57) D. D. Kerlick, Gen. Rel. Grav. 12 (1980a), 467.
58) D. D. Kerlick, Gen. Rel. Grav. 12 (1980b), 521.
52 H. Asada and T. Futamase
59) L. E. Kidder, C. M. Will and A. G. Wiseman, Phys. Rev. D47 (1993a), 3281.
60) S. M. Kopejkin, Sov. Astron. 29 (1985), 516.
61) K. Kuroda et. al., In proceedings of the international conference on gravitational waves:
Sources and Detections, ed. I. Ciufolini and F. Fidecard (World Scienti c, 1997), p. 100.
62) D. Lai, F. Rasio and S. L. Shapiro, Astrophys. J. Suppl. 88 (1993), 205.
63) D. Lai, F. Rasio and S. L. Shapiro, Astrophys. J. 420 (1994), 811.
64) L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Oxford: Pergamon
1962).
65) A. P. Lightman, W. H. Press, R. H. Price and S. A. Teukolsky, Problem Book in Relativity
and Gravitation (Princeton University Press, Princeton, NJ, 1975).
66) D. Markovic, Phys. Rev. D48 (1993), 4738.
67) A. Messiah, Mecanique Quantique, (Dunod, Paris; 1959).
68) T. Nakamura, K. Oohara and Y. Kojima, Prog. Theor. Phys. Suppl. 90 (1987).
69) T. Nakamura, In proceedings of the eighth Nishinomiya-Yukawa memorial symposium on
Relativistic Cosmology, ed. M. Sasaki (Universal Academy Press, Tokyo, 1994), p. 155.
70) T. Nakamura, talk at the fourth JGRG workshop held at Kyoto, (1994).
71) T. Ohta, H. Okamura, H. Kimura and K. Hiida, Prog. Theor. Phys. 51 (1974b), 1598.
72) K. Oohara and T. Nakamura, Prog. Theor. Phys. 83 (1990), 906.
73) K. Oohara and T. Nakamura, Prog. Theor. Phys. 86 (1991), 73.
74) K. Oohara and T. Nakamura, Prog. Theor. Phys. 88 (1992), 307.
75) N. O'Murchadha and J. W. York, Phys. Rev. D10 (1974), 2345.
76) E. S. Phinney, Astrophys. J. Lett. 380 (1991), 17.
77) E. Poisson and M. Sasaki, Phys. Rev. D51 (1995), 5753.
78) M. Ru ert, H-T. Janka and G. Schafer, Astron. and Astrophys. 311 (1996), 532.
79) G. Schafer, Ann. Phys. 161 (1985), 81.
80) G. Schafer, Astron. Nachr. 311 (1990), 213.
81) B. F. Schutz, Phys. Rev, D22 (1980), 249.
82) B. F. Schutz, T. In Relativistic Astrophysics and Cosmology, ed. X. Fustero and E. Verda-
guer (World Scienti c; 1984) p. 35.
83) B. F. Schutz, Nature, 323 (1986), 210.
84) M. Shibata and T. Nakamura, Prog. Theor. Phys. 88 (1992), 317.
85) M. Shibata, T. Nakamura and K. Oohara, Prog. Theor. Phys. 88 (1992), 1079.
86) M. Shibata, T. Nakamura and K. Oohara, Prog. Theor. Phys. 89 (1993), 809.
87) J. M. Stewart and M. Walker, Proc. Roy. Soc. London A341 (1979), 49.
88) H. Tagoshi and T. Nakamura, Phys. Rev. D49 (1994a), 4016.
89) K. S. Thorne, Rev. Mod. Phys. 52 (1980), 285.
90) K. S. Thorne, In 300 Years of Gravitation, ed. S. Hawking and W. Israel (Cambridge;
1987) p. 330.
91) K. S. Thorne, In proceedings of the eighth Nishinomiya-Yukawa memorial symposium on
Relativistic Cosmology, ed. M. Sasaki (Universal Academy Press, Tokyo, 1994).
92) R. M. Wald, General relativity (The University of Chicago press, 1984).
93) M. Walker, In Relativistic Astrophysics and Cosmology, ed. X. Fustero and E. Verdaguer
(World Scienti c; 1984) p. 99.
94) M. Walker and C. M. Will, Astrophys. J., Lett 242 (1980a), L129.
95) M. Walker and C. M. Will, Phys. Rev. Lett. 45 (1980b), 1741.
96) Y. Wang, A. Stebbins and E. L. Turner, Phys. Rev. Lett. 77 (1996), 2875.
97) C. M. Will, In 300 Years of Gravitation, ed. S. Hawking and W. Israel (Cambridge; 1987)
p. 80.
98) C. M. Will, In proceedings of the eighth Nishinomiya-Yukawa memorial symposium on
Relativistic Cosmology, ed. M. Sasaki (Universal Academy Press, Tokyo, 1994).
99) C. M. Will and A. G. Wiseman, Phys. Rev. D54 (1996), 4813.
100) J. R. Wilson and G. J. Mathews, Phys. Rev. Lett. 75 (1995), 4161.
101) J. R. Wilson, G. J. Mathews and P. Marronetti, Phys. Rev. D54 (1996), 1317.
102) X. Zhung, J. M. Centrella, and S. L. W. McMillan, Phys. Rev. D50 (1994), 6247.
Post-Newtonian Approximation 53
Appendix A
Conserved quantities for 2PN approximation of uniformly rotating uid
The conserved quantities in the 2PN approximation will be useful to investigate
the stability property of equilibrium solutions obtained in numerical calculations.
Since the de nitions of the conserved quantities are given in section 3, we shall
present only the results in this appendix.
(1)Conserved mass 4) ; Z
M  d3 x; (A.1)
where
"     #
 =  1 + c12 12 v2 + 3U + c14 83 v4 + 132 v2 U + 454 U 2 + 3U + (3) i vi + O(c 6 ) :
(A.2)
3); 92); 4)
(2)ADM mass ;
Z Z
1
MADM = 2  d x  ADM d3 x;
3
(A.3)
where
"  
1
ADM =  1 + c2 v +  + 2 U
2 5
  #
1 4 2 2 35 2 3
+ c4 v + 9v U + v + 5U + 4 U + 2 (3) v + O(c ) :(A.4)
i i 6

(3)Total energy, which is calculated from MADM M in the third PN order 4) ;


Z
E  E d x; 3
(A.5)
where
" ! !
E =  12 v2 +  21 U + c12 85 v4 + 52 v2U + v2  + 2U 25 U 2 + 12 (3) ivi
( 
1
+ c4 11 v 6
+ v 4
 + 47 U  + v2 4 + 6 U + 41 U 2 + 5 i vi X 
16 8 (4)
8 2 (3)
5 U 3 + 2 i vi  + 5 + 5U i vi 15 U + 5 U 2 
2 (3) (4) (3)
2 (4) 4
+ 12 hij vi vj + 12 (3) i (3) i
 !) #
U j 2
+ 16 2hij U;ij + (3) ;ji (3) ;ji + (3) ;i 3 ij (3) ;kk + O(c 6 ) : (A.6)

It is noteworthy that terms including (5) i cancel out in total.


54 H. Asada and T. Futamase
(4)Total linear and angular momenta: In the case Ki i = 0, these are calculated from
(Wald 1984)
I
1
Pi = 8 rlim j
!1 Kij n dS
Z
= 81 ( 6 Ki j );j d3 x
Z 
= Ji + 161 4 ~jk;iK jk 6 d3 x; (A.7)
where Ji = (c2 +  + P ) u0 ui . Up to the 2PN order, the second term in the last
line of Eq.(A.7) becomes
1 Z h j 3
jk;i (3) ;k d x
16 " #
Z  
1
= 16 hjk;i(3) j j
hjk;ik (3) d3 x
;k
I
1 j k (A.8)
!1 hjk;i(3) n dS = 0;
= 16 rlim
where we use hjk ! O(r 1 ) and (3) j ! O(r 2 ) at r ! 1, and the gauge condition
hjk;k = 0. Thus, in the 2PN approximation, Pi becomes
Z
Pi  pid x; 3
(A.9)
where
" (
  ) (
i 1
pi =  v + c2 v v +  + 6U + (3) i + c14 hij vj + (5) i
i 2

  
+(3) i v2 + 6U +  + vi 2(3) i vi + 10(4) + 6 U
 ) #
+ 67
4 U 2
+ v 2
+ 10 Uv 2
+ v 4
X + O (c 5 ) : (A.10)
The total angular momentum J becomes
Z
J = p'd x; 3
(A.11)
where p' = ypx + xpy .
Appendix B
Construction of Green's function
In this appendix we shall present a detailed derivation of Green's function for
2M . The following procedure is similar to that by Thorne (1980) for the Minkowski
background spacetime 89) .
Post-Newtonian Approximation 55
The Green's function satisfying Eq.(5.7) can be constructed by using the homo-
geneous solutions for the equation
2M = 0: (B.1)
The homogeneous solution for Eq.(B .1) takes a form of
e i!t f ()Y (; );
l lm (B.2)
where we de ned
 = !r: (B.3)
Then the radial function f~l ()  fl () satis es
2
d + 1 + 4M! l(l + 1) f~ () = 0; (B.4)
2
d   l 2

so that Eq.(B .2) is a solution for Eq.(B .1). Thus we can obtain homogeneous solu-
tions for Eq.(B .1) by choosing f~l () as one of spherical Coulomb functions; ul  (; ),
( )

Fl (; ) and Gl (; ) with = 2M!. Here, we adopted the following de nition of
the spherical Coulomb function 67) as
Fl (; ) = cl ei l F (l + 1 + i j2l + 2j 2i);
+1

ul  (; ) = 2ieil cl eil W (l + 1  i j2l + 2j  2i);


( ) +1
1

Gl (; ) = 12 (ul eil + ul e il );


(+) ( )
(B.5)
where cl and l are de ned as
cl = 2l e  = j ((2
2l + 1 + i )j ;
l + 1)!
l = arg (l + 1 + i ): (B.6)
Here, F and W1 are the con uent hypergeometric function and the Whittaker's
function respectively. These spherical Coulomb functions have asymptotic behavior
as
 
1
Fl  sin  ln 2 2 l + l ;
 
1
Gl  cos  ln 2 2 l + l ;
  
() 1
u  exp i  ln 2 2 l for r ! 1; (B.7)
and
Fl  cl l ; +1

Gl  (2l +1 1)c  l for r ! 0: (B.8)


l
56 H. Asada and T. Futamase
In terms of spherical Coulomb functions introduced above, we can construct the
Green's function in the standard way.
X Z 
GM (x; x0 ) = eil
( )
d!sgn(!)  ! l m (x) S ! l m  (x0)(r r0)
lm 
+ S ! l m (x)  ! l m  (x0 )(r0 r) ; (B.9)
where  denotes the complex conjugate, sgn(!) denotes a sign of ! and we de ned
 ! l m(x) and S ! l m(x) as
s
! l m (x) = j!j e i!t  1
ul (; )Yl m ;
(+)

s
2
S ! l m (x) = j2!j e i!t  1
Fl (; )Yl m: (B.10)
The retarded Green's function is obtained for  = +.
Appendix C
Derivation of the formula (5:28)
We evaluate the asymptotic form of the retarded integral as
  Z  0 
2 1 n^ L F (t r) = d4 x0G (x; x0 ) n^ L F (t0 r0)
rk r 0k0

Z Zr i!r
! i d!!e i!t hl (!r)F! n^ L r0 dr0jl (!r0) er0k
0

0
for large r; (C.1)
where we de ned F! as Z
F! = 21 dt ei!t F (t): (C.2)
Since the Bessel function is de ned as
1
X ( )n (z=2) +2n ;
J (z) = (C.3)
n=0 n! ( + n + 1)
we obtain formally
Z1 1
( )n (b=2) +2n Z 1 dye ay y+ +2n 1
X
dyJ (by)e ay y = 1

0 n=0 n! ( + n + 1) 0
1 n 2) +2n ( +  + 2n)
= ( )n!(ab=
X
+ +2n ( + n + 1) : (C.4)
n=0
Putting a = i, b = 1,  = k + 5=2  and  = l + 1=2 in Eq.(C .4), we obtain
where we introduced small  2 C in order to avoid the pole of ( k + l + 3 + 2n)
Post-Newtonian Approximation 57
for k  l + 3. Thus we obtain
r
Zr i!r
r02dr0 jl (!r0) e0k+ = !k
0

3 i k+l+3 X1 ( k + l + 3 + 2n ) + O(r 1 ):


0 r 2 2l+ 12 n=0 22n n! (l + n + 32 )
(C.5)
From Eqs.(C .1), (C .2) and (C .5), we obtain
  p  X 1 ( k + l + 3 + 2n )  n^ (k 3)
2 1 n^rkL F (t r) = lim!0 2l+1 n=0 22n n! (l + n + 32 )
L
rk F (t r)
+O(r 2 ): (C.6)
This equation is valid for k  3. For 3  k  l + 2, we obtain
 
2 1 n^rkL F (t r) = 2k 3 (k (l3)!( l k + 2)! n^ L (k 3)(t r) + O(r 2)
+ k 2)! r F
for 3  k  l + 2; (C.7)
since
p  X 1 ( k + l + 3 + 2n) 
= 2 k 3 (k 3)!(l k + 2)! for 3  k  l + 2:
2 n=0 2 n! (l + n + 2 )
l +1 2 n 3
(l + k 2)!
(C.8)
This formula for 3  k  l + 2 has been derived by Blanchet and Damour (1988).
Appendix D
Derivation of (5:29)
From Eq.(5.27), we shall derive Eq.(5.29) in this appendix. First, for the rst
term in the right hand side of Eq.(5.27), we use Eq.(C.7) to obtain
q (q + j )!  
( ) q +1
X 1 (q j )
2 n^ iQ rj+3 F (t r)
1

j =0 2 j !(q j )!
j
q
= 2((q +) 1) n^riQ F (t r) + O(r 2 ):
(q )
(D.1)
Next, the second term in the right hand side of Eq.(5.27) is calculated as
qX2  
( )q+1 2q q+ 1 2j(jq!(+q j )!j )! 2 1 i<aq n^ Q 1> rj1+3 F (t r)
(q j )

j =0
 n^
= ( )q 2qq +11 i<aq r Q 1> F (t r) + O(r 2 );
(q )
(D.2)
and

( )q+1 2q q+ 1
Xq (q + j )! 2 i<aq n^ Q 1 (q j()t r)
1> j +3 F
1

j =q 1 2 j !(q j )! r
j
58 H. Asada and T. Futamase
)! i<aq n^ Q 1> (q)(t r)
= ( )q (2qq(2+q1)q! r F
X1  (q + n)!  
 lim
!0 n=1 n!(2q + 2n)!
(2n ) + (2n 1 )
 
q !
+ (2q)! ( ) + ( 1 ) + O(r 2 )

= ( )q+1 2(2 2q 1 i<aq n^ Q 1> (q)(t r) + O(r 2 ): (D.3)


q + 1) r F
Here we used Eq.(5.28),
1
X (q + n)!  (2n) + (2n 1) = q! ; (D.4)
n=1 n!(2q + 2n)! 2q(2q)!
and  
lim
!0
( ) + ( 1 ) = lim !0
( ) ( 1 ) = 1; (D.5)
which, in fact, means the cancellation of the poles.
From Eqs.(5.27) and (D.1)-(D.3), we obtain
  
2 1 @ ( 1 )@^ F (t r)
i r Q r
 
)q
= 2((q + 1) n^ iQ q+1 1 (q)(t r) + O(r 2 ); (D.6)
2q + 1 i<a1 n^ Q 1> rF
which equals to (C2) obtained by Blanchet 7) . As for Eq.(D.6), we used unphysical
but small parameter  in order to avoid the pole of the integrals. However, in the
limit  ! 0, we showed explicitly the cancellation of poles and obtained the nite
values in total.
Post-Newtonian Approximation 59
Table 1
List of potentials to be solved (column 1), Poisson equations for them (column 2),
and other potential variables which appear in the source term of the Poisson equation
(column 3). Note that i and j run x; y; z . Also, note that we do not have to solve
0z , (5) P0z , qyy , qzz and h(zzU ) .

Pot. Eq. Needed pots. Pot. Eq. Needed pots.


U (2.11) None qij (4.6) None
qi (3.14) None Q Ii ( )
0 (4.15) U
q2 (4.1) None i 0 (4.16) U
qi
2 (4.2) None Pi
(5) 0 (4.17) U; qi
q4 (4.3) None (6) 0 (4.21) U; qe; qu; hijU ; Q Ii
( ) ( )
0

qu (4.4) U (6) 2 (4.22) U; q ; qi; q i; qij


2 2

qe (4.5) None hijU( )


(3.1) U
Table 2
Variables to be solved in order to obtain the original metric variables.
Metric Variables to be solved see Eq.
U U (2.11)
(3) i qi, U (3.17)
X q , qu, qe
2 (4.7)
(4) q , qu, qe
2 (4.8)
i (A)
(5) P i ,  i, qu, qe
(5) 0 0 (4.18)
(5)
i (B) q i, q2 2 (4.18)
(6) (6) , , q
0 (6) 2 4 (4.20)
hijU )
(
hijU ( )
(3.1)
h(ijS) qij ; q 2 (4.14)
h(ijA) Q Ii , qu, qe
( )
0 (4.19)
h(ijB) qij , q , q i, qi
2 2 (4.19)