1 Linear theory in a weak field approximation

(More detailed technical and explanatory remarks are marked in red.)

Creating a mathematical formalism describing gravitational waves is extremely difficult, among

other things, because of its nonlinearity. In this case, we consider spacetime ripples very weak
because we are far from the source. Following this assumption, we can linearize the gravitational
field as a slightly deformed Minkowski flat spacetime ηµν = diag(+1, −1, −1, −1) with a small
perturbation hµν (we use the convention + − −− here, other authors may use the opposite, while
the physics stays unaffected),

gµν ≃ ηµν + hµν + O(h2µν ), |hµν | ≪ 1. (1)

We raise and lower the indices of this terms by ηµν : following the principle,

hµν = η µα η νβ hαβ , h = η µν hµν , (2)

we obtain the “upper index” linearization

g µν ≃ η µν − hµν . (3)

It’s because g µν is obtained from gµν as the inverse matrix, i.e., via

g µν gνσ = δ µσ . (4)

Now we formally distinguish “lower” and “upper” indexed perturbations as h and h̃,

gνσ = ηνσ + hνσ , g µν = η µν + h̃µν , (5)

noting again that ηµν = η µν = diag(+1, −1, −1, −1). Following this, we get

g µν gνσ (= δ µσ ) = (η µν + h̃µν )(ηνσ + hνσ ) = δ µσ + hµσ + h̃µσ + O(h̃µν hνσ ), (6)

which means (or, after raising the lower index again with η νσ ),

h̃µσ = −hµσ , h̃µν = −hµν . (7)

So, this results from the consistently applied formalism of lowering and raising indices with ηµν
and η µν , respectively.
Now we want to find the equation of motion of the perturbations hµν by examining Einstein’s
equations in the first order. We adopt the Christoffel symbols to the linearized gravity (since the
derivatives of ηµν are zero) as
1 1
Γρµν = η ρσ (∂µ hνσ + ∂ν hµσ − ∂σ hµν ) = ∂µ hρν + ∂ν hρµ − ∂ ρ hµν .


(8)
2 2
ρ 1 ρσ 1
∂µ hρρ + ∂ρ hρµ − ∂σ hσµ .


Γµρ = η (∂µ hρσ + ∂ρ hµσ − ∂σ hµρ ) = (9)
2 2
The Ricci tensor (since the Christoffel symbols are the first order quantities; we employ only the
derivatives of Γ and neglect the Γ2 terms) will then take the form
1
Rµν ≃ ∂ρ Γρµν − ∂ν Γρµρ ≃ ∂ρ ∂µ hρν + ∂ρ ∂ν hρµ − ∂µ ∂ν hρρ − ∂ρ ∂ ρ hµν .


(10)
2

1
The term hρρ = η ρσ hσρ = h, and the operator ∂ρ ∂ ρ represents the scalar product of the four-
derivatives in Minkowski space with the raised index ∂ ρ = η ρσ ∂σ , that is,

∂ρ ∂ ρ = c−2 ∂t2 − ∂x2 − ∂y2 − ∂z2 = c−2 ∂t2 − ∇2 = □, (11)

where the "box" shortly symbolizes the d’Alembert operator (also called the d’Alembertian).
Applying the η µν Rµν contraction of Ricci tensor, we obtain the Ricci scalar R as
1
∂ρ ∂µ hµρ + ∂ρ ∂ν hνρ − ∂µ ∂ µ hρρ − ∂ρ ∂ ρ hµµ = ∂ρ ∂µ hµρ − ∂ρ ∂ ρ hµµ .


R= (12)
2
Using the contraction of hρρ = h and the d’Alembertian symbol, we can write the Ricci tensor
and Ricci scalar of the linearized perturbed spacetime in the more convenient form as
1
∂ρ ∂µ hρν + ∂ρ ∂ν hρµ − ∂µ ∂ν h − □hµν , R = ∂ρ ∂µ hµρ − □h.


Rµν = (13)
2
We directly obtain the Einstein tensor Gµν of this linearized flat spacetime perturbations by
summing the Ricci tensor and Ricci scalar,
1 1
∂ρ ∂µ hρν + ∂ρ ∂ν hρµ − ∂µ ∂ν h − □hµν − ηµν ∂ρ ∂µ hµρ + ηµν □h .


Gµν = Rµν − ηµν R = (14)
2 2
We can thus simply write the equation of the linearized gravity,
16πG
∂ρ ∂µ hρν + ∂ρ ∂ν hρµ − ∂µ ∂ν h − □hµν − ηµν ∂ρ ∂µ hµρ + ηµν □h = Tµν . (15)
c4
As a next step, it is convenient to introduce a "trace-reversed" (in 4-dimensional spacetime)
tensor ϕµν defined as
1
ϕµν = hµν − ηµν h, (16)
2
which manifestly gives ϕ = −h and so hµν = ϕµν − 12 ηµν ϕ (for the same reason, we may regard the
Einstein tensor as simply the trace-reversed Ricci tensor). The metric perturbation hµν and the
trace-reversed perturbation ϕµν thus contain the same information. In n-dimensional spacetime
it would be

h = hµµ = η µν hµν = h00 − h11 − h22 − h33 − ... − hnn , (17)

1 1
ϕ = ϕµµ = η µν hµν − η µν ηµν h = h − nh, (18)
2 2
noting that η µν ηµν = n (summation convention). If n = 4, then ϕ = h − 2h = −h. We can now
insert the tensor ϕµν into Eq. (15), getting
       
ρµ 1 ρν 1 1
∂ρ ∂µ η ϕµν − ηµν ϕ + ∂ρ ∂ν η ϕµν − ηµν ϕ + ∂µ ∂ν ϕ − □ ϕµν − ηµν ϕ −
2 2 2
  
1 16πG
−ηµν ∂ρ ∂µ η µν η µρ ϕµν − ηµν ϕ − ηµν □ϕ = Tµν ⇒ (19)
2 c4
1 1 1
∂ρ ∂µ ϕρν − ∂µ ∂ν ϕ + ∂ρ ∂ν ϕρµ − ∂µ ∂ν ϕ + ∂µ ∂ν ϕ − □ϕµν + ηµν □ϕ−
2 2 2
µρ 1 16πG
−ηµν ∂ρ ∂µ ϕ + ηµν □ϕ − ηµν □ϕ = Tµν ⇒ (20)
2 c4

2
 16πG
∂ρ ∂µ ϕρν + ∂ρ ∂ν ϕρµ − □ϕµν − ηµν ∂ρ ∂µ ϕµρ = Tµν , (21)
c4
we thus reduced the left-hand side of the equation of the linearized gravity to four terms; by
lowering the upper index ρ in the term ϕµν , we can even reduce the left-hand side to two terms,
16πG
∂ρ ∂ν ϕρµ − □ϕµν = Tµν , (22)
c4
Another step in the solution of this problem is to “choose a gauge.” This is similar to the
“electromagnetic” Lorentz gauge condition on the vector potential Aµ (by setting ϕ′ = ϕ − ψt ,
A⃗′ = A⃗ + ∇ψ,
⃗ where ψ is an arbitrary function), in which ∇µ Aµ = 0. To illustrate the gauge from
basic principles, let’s introduce two coordinate systems xµ and x′µ , deviating from each other by
a very small displacement ξ µ (four functions of the order hµν in a 4-dimensional spacetime),

xµ′ = xµ + ξ µ . (23)

Then, obviously,

∂xµ ∂xµ′
= δ µν − ∂ν ξ µ , = δ µν + ∂ν ξ µ , (24)
∂xν ′ ∂xν
and the first-order displacement is

′ ∂xρ ∂xσ
(x′ ) = ρ ρ

σ
gµν ′
gρσ (x) = δ µ − ∂ µ ξ (δ ν − ∂ν ξ σ ) (ηρσ + hρσ )
∂xµ′ ∂xν
≃ ηµν + hµν − ∂µ ξν − ∂ν ξµ = ηµν + h′µν . (25)

Thus, the metric transforms as

h′µν = hµν − ∂µ ξν − ∂ν ξµ (26)

and, following the contraction h′ = η µν h′µν = η µν (hµν − ∂µ ξν − ∂ν ξµ ) = hµµ − ∂µ ξ µ − ∂ν ξ ν =

h − 2∂α ξ α , we can immediately generalize the gauge condition in the trace-reversed perturbation
as

ϕ′µν = ϕµν − ∂µ ξν − ∂ν ξµ + ηµν ∂λ ξ λ . (27)

In general relativity, solving Einstein’s equations, it is common practice to choose a harmonic

gauge condition (also known as Lorentz, Hilbert, de Donder, or Fock gauge) on the coordinate
system, where (absolutely analogously to the Lorentz gauge in EM)

□xµ = 0. (28)

Since □ = ∇α ∇α (recalling that the covariant derivative ∇α xµ = ∂α xµ + Γµαν xν , ∇α ωµ = ∂α ωµ −

Γναµ ων where xµ is the four-vector and ωµ is the one-form why? because the covariant derivative of
a scalar is simply the partial derivative, so let’s assume an ad-hoc symbolized expression ∇α ωµ =
∂α ωµ + Γe ν ων and let’s impose the covariant derivative on a scalar ∇α (ωµ xµ ) = xµ (∇α ωµ ) +
αµ
ωµ (∇α x ), giving xµ (∂α ωµ + Γ
µ e σαµ ωσ ) + ωµ (∂α xµ + Γµαν xν ), while, simultaneously, ∇α (ωµ xµ ) =
∂α (ωµ xµ ) = xµ (∂α ωµ ) + ωµ (∂α xµ ); this cancels the connection coefficient (Christoffel symbols)
terms, Γe σ ωσ xµ = −Γµαν xν ωµ , so, since Γ e σ ωσ xµ ≡ Γ e µαν ωµ xν , this finally proves that Γ
e µαν =
αµ αµ

3
−Γµαν ), we may this condition (noting that it is imposed on a scalar coordinate function xµ within
the orthogonal four-coordinate system where obviously ∂α xµ = δαµ ) explicitly expand as

0 = □xµ = η ρσ (∂ρ ∂σ xµ − Γλρσ ∂λ xµ ) = −η ρσ Γλρσ . (29)

At this point, it is necessary to strictly keep in mind that the four functions xµ are just functions,
not components of a vector; since the covariant derivative of a scalar function is just the partial
derivative, ∇α xµ = ∂α xµ , we simply arrive at the expression (29). Expanding the Eq. (29)
explicitly in the weak field limit,
1 1
0 = η ρσ η λα (∂ρ hσα + ∂σ hρα − ∂α hρσ ) = ∂ρ hρα − ∂α h. (30)
2 2
The first term on the left-hand side of Eq. (15) then transforms to 12 ∂ν ∂µ h = 12 ηµν □h, while the
fifth term transforms to −∂ρ ∂µ hρν = − 12 ηµν □h; that is, Eq. (15) by imposing the Lorentz gauge
simplifies to
1 16πG
□hµν − ηµν □h = − 4 Tµν , (31)
2 c
or
16πG
□ϕµν = − Tµν , (32)
c4
while the vacuum form of Einstein equation Rµν = 0 (together with the gauge invariance expressed
by Eq. (23) and the Lorentz gauge condition, Eq. (28)) super-reduces Eq. (31) to

□hµν = 0, □ϕµν = 0, □ξ µ = 0. (33)

Moreover, by raising indices and substituting the trace-reversed tensor ϕµν into Eq. (30), we
obtain the alternative important form of the Lorentz gauge condition,
1 1 1 1
0 = (∂ρ hρα − ∂α h)η αν = ∂ρ hρν − ∂ ν h = ∂ρ (ϕρν − η ρν ϕ) + ∂ ν ϕ ⇒
2 2 2 2
∂ρ ϕρν = 0. (34)

Now, we apply the linearized gravity to a gravitational field of an isolated mass in the New-
tonian limit. In this case, we assume that the energy-momentum tensor (see explanations in
Example 2.5 in the textbook "Prakticke_pocetni_metody_pro_fyziky") is dominated by the en-
ergy density ρc2 (Tµν = ρc2 δµ0 δν0 ), the matter is practically static, or it moves slowly enough so
we may neglect the time derivatives (then □ = −∇2 in the convention + − −− of the flat space-
time), and the spacetime is "asymptotically flat," that is, it behaves as the Minkowski spacetime
at large distances. Then Eq. (32) says
16πG
∇2 ϕ00 = ρ (35)
c2
and, after implementing the gravitational Poisson equation ∇2 Φ = 4πGρ (we hereafter consis-
tently distinguish Φ as a gravitational potential from a trace-reversed perturbation ϕ),
4Φ
ϕ00 = . (36)
c2

4
Since the other components of ϕµν are negligible, then ϕ = ϕ00 , and hi0 = ϕi0 − 12 ηi0 ϕ = 0 (where
i = 1, 2, 3 are the spatial components of the metric). The "inverse" Eq. (16) thus immediately
gives
2Φ
hµν = δµν . (37)
c2
Finally, following Eq. (1), the metric form of a perturbed spacetime in this weak-field limit is
 
2 2

2 2Φ
dx2 + dy 2 + dz 2 .


ds = c + 2Φ dt − 1 + 2 (38)
c

We now describe the formalism of the weak-field limit application to gravitational radiation.
Let’s suppose the perturbation within the vacuum solution □ϕµν = 0 radiates plane waves in the
form
α i
ϕµν = Aµν eikα x = Aµν e−i(ki x −ωt) , (39)

where Aµν is a constant and symmetric spacetime tensor of second order (consisting thus of ten
independent components, called polarization tensor including information of the amplitude and
the polarization properties of the gravitational waves), and kα is the wavevector; k0 = k 0 = ω/c,
ki = −k i . Then, the flat-space d’Alembertian imposed on a scalar complex 4-exponential yields

0 = □ϕµν = η αβ ∂α ∂β ϕµν = η αβ ∂α (ikβ ϕµν ) = −η αβ kα kβ ϕµν = −kα k α ϕµν . (40)

Because at least some of the ϕµν components must be nonzero (otherwise, we do not have any
subject to deal with), we have a solution of a null wavevector,

kα k α = 0, (41)

which immediately shows that the gravitational waves propagate with the speed of light. Moreover,
since ω 2 = c2 δij k i k j = c2 k 2 (kα k α = ω 2 − c2 |⃗k|2 = 0) (where k 2 = |⃗k|2 ), this relation explicitly
suggests
ω ∂ω
vphase = = c, vgroup = = c, (42)
k ∂k
that is, the group as well as the phase velocity of the gravitational waves are equal to the speed
of light c.
This simple plane wave, of course does not describe the complete or general solution; (possibly)
an infinite number of distinct plane waves can be superposed and solve the linear wave equations
(33). Imposing the Lorentz gauge condition (34) on Eq. (39), we see that
α α
0 = ∂µ (Aµν eikα x ) = ikµ Aµν eikα x , (43)

which is fulfilled if and only if

kµ Aµν = 0. (44)

Thus, we may regard the wavevector kµ as orthogonal to Aµν ; this condition reduces the number
of independent components of Aµν from ten to six. Explicitly, recalling that due to the symmetry

5
of Aµν (Aµν = Aνµ ),

c kµ Aµ0 = ωA00 − c k1 A10 + k2 A20 + k3 A30 = 0,

(45)
c kµ Aµ1 = ω A01 ≡ A10 − c k1 A11 + k2 A21 + k3 A31 = 0,




(46)
c kµ Aµ2 = ω A02 ≡ A20 − c k1 A12 ≡ A21 + k2 A22 + k3 A32 = 0,




(47)
c kµ Aµ3 = ω A03 ≡ A30 − c k1 A13 ≡ A31 + k2 A23 ≡ A32 + k3 A33 = 0,





(48)

showing that in this case, the ten originally independent components A00 , A10 , A20 , A30 , A11 ,
A21 , A31 , A22 , A32 , A33 , reduce to only six, A11 , A21 , A31 , A22 , A32 , and A33 (the same applies for
symmetric counterparts of the nondiagonal ones). From the last equation (33) and analogously
to Eq. (39), we claim the solution for the displacement
σ
ξµ = Bµ eikσ x , (49)

where Bµ are constant coefficients and kα is the wave four-vector. Then, following (27) with
Eq. (39) plugged in and with Eq. (49) for the transformation of displacement, we obtain the
gravitational wave amplitude changes as

A′ µν = Aµν − ikµ Bν − ikν Bµ + iηµν kλ B λ , (50)

This, after raising indices by η µν changes to

µ
A′ µ = Aµµ − ik ν Bν − ik µ Bµ + 4ikλ B λ = Aµµ + 2ikλ B λ . (51)

which can also be simply modified to

µ
A′ µ = Aµµ + 2ikλ B λ . (52)

This can be rearranged as

µ
A′ µ = Aµµ + 2ikλ Bσ η σλ , (53)

which we may explicitly write as (employing only the relevant terms)

µ
A′ = Aµµ + 2i k0 B0 η 00 + k1 B1 η 11 + k2 B2 η 22 + k3 B3 η 33 =


µ (54)
= Aµµ + 2i (k0 B0 − k1 B1 − k2 B2 − k3 B3 ) , (55)

Following Eqs. (50) and (52), we can construct the matrix transformation for selected amplitudes,
1 ′ µ  1 µ   ω  
2A µ 2A µ c −k1 −k2 −k3 B0
 A′   A01  −k1 − ω 0 0  B1 
 01  =  c
 A′   A02  + i −k2
   , (56)
02 0 − ωc 0  B2 
A′03 A03 −k3 0 0 − ωc B3

where, because we may choose the coordinate shift amplitudes Bµ however we need, we can also
adjust the coordinate system so that
µ
A′ µ = 0, A′ 0ν = 0, (57)

6
to which we convert the "old" Aµν coefficients. We can also extend the solution (44) via

k µ A′ µν = k µ Aµν − ik µ kµ Bν − ik µ kν Bµ + iηµν k µ kλ B λ =
 
= 0 − 0 − ikν k µ Bµ − kλ B λ = 0, that is k µ A′ µν = 0. (58)

For clarity, we readjust the relation (56) as

1 µ   ω  
2A µ −c k1 k2 k3 B0
 A01   k1 ω
c 0 0  B1 
 
 A02  = i  k2
   ω
. (59)
0 c 0  B2 
A03 k3 0 0 ωc B3

In any case, the first equation (57) brings another constraint to the already announced six in-
dependent components Aµν , reducing them to five. The second equation (57) should bring four
more constraints; however, since one of these is already dependent via Eq. (44), we have only
three additional constraints at this point. The total number of independent coefficients Aµν is
thus reduced to two. The following suggestions will yet more clearly explain this.

2 The transverse - traceless (TT) gauge

We can now choose the coordinate system oriented so that the plane gravitational wave is propa-
gating in the direction of the third spatial axis xµ = x3 . In this case, according to the null vector
solution (41), we have
ωω ω
kµ k µ = k0 k 0 + k3 k 3 = + x3 (−x3 ) = 0 ⇒ x3 = (60)
ω c c  c
µ 3 ω ω
k = , 0, 0, x = , 0, 0, . (61)
c c c
In this case, the condition (58) now gives k 0 A′ 0ν + k 3 A′ 3ν = 0, which, together with the second
equation (57), means

A′ 3ν = 0, (62)

reducing the number of independent terms in the A′µν to four, A′11 , A′12 , A′21 , and A′22 . Moreover,
A′21 = A′12 due to the symmetry of A′µν and the first equation (57) implies that A′µν is traceless,
A′ µµ = η µν A′µν = A′00 − A′11 − A′22 − A′33 = 0, A′22 = −A′11 , and
 
0 0 0 0
0 A′11 A′12 0
A′µν =
0 A′12 −A′11
. (63)
0
0 0 0 0

The plane is within this configuration completely described by frequency ω and the two compo-
nents A′11 and A′12 . We call this particular gauge the Transverse Traceless (TT) gauge (or radiation
gauge). Now, since the trace-reversed perturbation ϕµν is in this gauge traceless (because Aµν is),
and it is equal to the trace-reverse of hµν , we have

ϕTµνT = hTµνT . (64)

7
We can, therefore, use both quantities equally in this gauge.
Let’s rename the two numbers to forms that will better illustrate the principles (will be ex-
plained soon), A+ = A′11 and A× = A′12 , then
 
0 0 0 0
0 A+ A× 0 σ
A= 0 A× −A+ 0 ,
 hµν = A eikσ x (65)
0 0 0 0

The quantities A+ and A× are invariant under gauge transformations according to Eq. (50), so
we may use them to compute the amplitudes of an arbitrary gravitational wave. The metric
associated with a gravitational wave within this gauge is

ds2 = gµν dxµ dxν =

σ σ σ
= c2 dt2 − (1 + A+ eikσ x )(dx1 )2 − 2A× eikσ x dx1 dx2 − (1 − A+ eikσ x )(dx2 )2 − (dx3 )2 . (66)

Let’s consider a linearly polarized wave with polarization +, and an ensemble of test particles at
coordinates x1 , x2 , x3 . As the wave passes, the particles separated in the 1st direction oscillate
in this direction. The particles separated in the 2nd direction do the same in the counter-phase.
The distance element in the 3rd direction (the direction of propagation) remains unaffected by
the passage of the gravitational wave. The distance elements in the 1st and 2nd direction are thus
given by
   
1 σ 1 σ
dx1 ≈ 1 + A+ eikσ x dx10 dx2 ≈ 1 − A+ eikσ x dx20 . (67)
2 2

A × polarized wave oscillates along the axes rotated by 45 degrees. The 3rd direction remains
again unaffected. We designate + and × polarizations rather than "horizontal" or "vertical," as
we do for light. Here should be added pictures of polarization-dependent oscillations.
These polarization modes are thus invariant under rotation of 180 degrees; since the spin S
is generally given by the rotational angle θ invariance as S = 360◦ /θ, we may suppose that the
"gravitons" predicted in the quantum gravitation theory will have the spin 2. Of course, such
particles have not yet been detected (and maybe they will never be in the future). However,
relevant quantum field theories suggest their existence as massless particles (since they propagate
with the speed of light) with the corresponding spin.
Now, let’s revoke Eq. (32) as the starting point for describing the coupling of the gravitational
radiation with its source (supposed to be of a material nature). The solution to such an equation
can be obtained using Green’s function in precisely the same way as the analogous problem in
electromagnetism. However, before we come to this, we join the short explanatory notes on "re-
tarded” solutions as they are defined in, e.g., the Liénard-Wiechert potentials in electrodynamics
(in this sense, there can be replaced a mass for a charge, or vice versa). The term "retarded" is
used in this context in the sense of "propagation delays." Consider a particle of charge q moving
along a trajectory r − r ′ (t′ ) whose velocity is u(t′ ) = ṙ ′ (t′ ). Its charge and current is
ˆ ˆ
q = q δ(r − r ′ (t′ )) d3 r , qu = qu δ(r − r ′ (t′ )) d3 r , (68)

where the charge and current densities are ρ(r , t) = qδ(r − r ′ (t′ )) and j (r , t) = quδ(r − r ′ (t′ )),
respectively, and where the general property of the Dirac δ-function is a localization of an integral

8
given by
ˆ
f (x) δ(x − x0 ) dx = f (x0 ). (69)

We now define the retarded potentials (in the system of SI units) as

ˆ ˆ
1 [ρ] d3 r ′ µ0 [j ] d3 r ′
ϕ(r , t) = , A(r , t) = . (70)
4πϵ0 |r − r ′ | 4π |r − r ′ |

where the quantities in the square brackets mean that they are evaluated at the retarded time
t − |r − r ′ |/c, which refers to conditions at the point r ′ that existed at a time earlier by |r − r ′ |/c
than t where |r − r ′ |/c is the time required for light to travel between r ′ and r .
We now evaluate these retarded potentials from Eq. (70) via the charge and current densities
given by integrands of Eq. (68). The scalar potential is then
ˆ ˆ
ρ(r ′ , t′ ) |r − r ′ |
  
1 3 ′ ′
ϕ(r , t) = d r δ t − t− dt′ =
4πϵ0 |r − r ′ | c
ˆ
R(t′ )
 
q −1 ′ ′
= R (t ) δ t − t + dt′ , (71)
4πϵ0 c
ˆ
R(t′ )
 
µ0 q
A(r , t) = u(t′ )R−1 (t′ ) δ t′ − t + dt′ , (72)
4π c

where R(t′ ) = r − r ′ (t′ ) and R(t′ ) = |R(t′ )|.

The argument of the δ-function vanishes for a value of t′ = tret (retarded time) given by

R(tret )
tret = t − . (73)
c
We substitute a new variable t′′ = t′ − t + R(t′ )/c whose differential
 
h i 1 ′
dt = 1 + Ṙ(t ) dt = 1 − n(t ) · u(t ) dt′ ,
′′ ′ ′ ′
(74)
c

where we obtain the latter by differentiating the identity R2 (t′ ) = 2R(t′ )Ṙ(t′ ) = −2R(t′ ) · u(t′ ),
where Ṙ(t′ ) = −u(t′ ) and the unit vector n = R/R. Equations (71) and (72) take the form
ˆ  −1
q −1 ′ 1 ′ ′
ϕ(r , t) = R (t ) 1 − n(t ) · u(t ) δt′′ dt′′ , (75)
4πϵ0 c
ˆ  −1
µ0 q ′ −1 ′ 1 ′ ′
A(r , t) = u(t )R (t ) 1 − n(t ) · u(t ) δt′′ dt′′ . (76)
4π c

Setting t′′ = 0 or equivalently t′ = tret simplifies the notation to

1 h q i µ0 h qu i
ϕ(r , t) = , A(r , t) = , (77)
4πϵ0 κR 4π κR
where we keep the "bracket" notation for retarded potentials and where
1
κ(tret ) = 1 − n(tret ) · u(tret ) = 1 − β(tret ) · n(tret ). (78)
c

9
Equations (77) are the Liénard-Wiechert Potentials. They differ from static electromagnetic
potentials by the factor κ(tret ) that becomes very important at velocities close to the speed of light
c, where it concentrates the potentials into a narrow cone about the particle’s velocity (relativistic
beaming effect). Another difference is that the quantities are evaluated at the retarded time tret ,
which enables a particle to radiate. The potentials fall off as ∝ 1/r so that the fields would
decrease ∝ 1/r2 if the differentiation of potentials acted exclusively on the factor ∝ 1/r. The
retardation involves an implicit dependence on position via the definition of retarded time, and
the differentiation for this dependence transforms the 1/r behavior of the potentials into the fields
themselves. This allows radiation energy to flow to an infinite distance.
The alternative (and maybe better) explanation of this stuff is the following, using the prin-
ciples of special relativity: In the primed frame, the charge is at rest, so (see electro- and magne-
tostatics),
1 q ϕ′ µ0 qc
ϕ′ = , that is, since µ0 ϵ0 = c−2 , = , A′ = 0, (79)
4πϵ0 R′ c 4π R′
where the latter two terms are the components of the four-potential Aα = (ϕ/c, A), while the
other relevant four-quantities are uα = γ(c, u), and Rα is the position four-vector of the inter-
event distance, formed analogously to the four-position definition rα = (ct, r ) as

Rα = c t − t′ , r − r ′ ≡ c t − t′ , R ′ .

 

(80)

We can, however, generalize the four-potential (using the four-notation, in the non-primed quan-
tities) consistently as
µ0 cuα
Aα = q , (81)
4π uβ Rβ
where, reminding that c(t − t′ ) = |r − r ′ | ≡ R,

uβ Rβ = γ(c, −u) · c t − t′ , R = γc (R − β · R) = γcR (1 − β · n) ,



(82)

where n ≡ R/R is a unit vector in the observer’s direction, β ≡ u/c is the normalized velocity
of the particle, and γ ≡ 1/(1 − u2 /c2 )1/2 is the Lorentz factor. In the reference frame of the
charge (in which β = 0 and γ = 1), the expression (82) is reduced to cR so we see that Eq. (81)
is correctly reduced to Eq. (79). We now express the components of Eq. (81) in terms of the lab
frame (in which t′ = tret and R = Rret ),
   
1 1 µ0 β
ϕ(r , t) = q , A(r , t) = qc . (83)
4πϵ0 R (1 − β · n) ret 4π R (1 − β · n) ret
These formulas for the Liénard-Wiechert potentials are identical to those in Eq. (77).
Let’s now do one more formal calculation whose result we will need anyway. We differentiate
the geometric relation

c(t − tret ) = Rret ≡ |r (t) − r ′ (tret )| (84)

over tret and then, independently, over t, assuming that r is fixed. Differentiating both sides of
2 =R
the identity Rret ret · R ret over tret , we have

∂Rret ∂R ret
2Rret = 2R ret · . (85)
∂tret ∂tret

10
Regarding the fixed r , then ∂R ret /∂tret ≡ ∂(r − r ′ )/∂tret = −∂r ′ /∂tret ≡ −u ret , Eq. (85) yields
∂Rret R ret ∂R ret
= · = − (n · u)ret . (86)
∂tret Rret ∂tret
Differentiating in Eq. (84) the same Rret over t gives
∂Rret ∂tret
=c−c , (87)
∂t ∂t
while using the chain rule, we have
∂Rret ∂Rret ∂tret ∂tret
= = − (n · u)ret . (88)
∂t ∂tret ∂t ∂t
Combining Eqs. (87) and (88) gives
 
∂tret c 1
= = . (89)
∂t c − (n · u) ret 1 − β · n ret

The Green’s function G(xα − y α ) for the D’Alembertian operator □ is the solution of the wave
equation in the presence of a delta-function source: □x G(xα − y α ) = δ (4) (xα − y α ), where □x
denotes the D’Alembertian for the coordinates xα . The general solution of Eq. (32) can be then
written as the convolution see the general principles of Fourier transformation
ˆ
16πG
ϕµν (xα ) = − 4 G(xα − y α ) Tµν (y α ) d4 y. (90)
c
We express the retarded Green’s function, which represents the accumulated effects of signals to
the past of the point under consideration as
 
α α 1 0 0

|x − y |
G(x − y ) = − δ x −y − θ(x0 − y 0 ). (91)
4π |x − y | c
where t = x0 and the boldface denote the spatial vectors x = (x1 , x2 , x3 ) and y = (y 1 , y 2 , y 3 ),
1/2
with the norm |x − y | = δij (xi − y i )(xj − y j ) while the function θ(x0 − y 0 ) equals 1 when


x0 > y 0 , and zero otherwise, to protect the time causality.

Inserting (91) into (90), we can perform the integral over y ′ using the delta function,
ˆ  
4G 1 |x − y |
ϕµν (t, x) = 4 Tµν t − , y d3 y, (92)
c |x − y | c
the retarded time now refers to tret = |x − y | /c. We may consider Eq. (92) in the following way:
the perturbation of the gravitational field at the "spacetime point" (t, x) is a sum of the energy
and momentum impulses generated at points on the past light cone. Illustrative picture?
Let’s now consider the case of the gravitational radiation emitted by a sufficiently distant
isolated source comprised of nonrelativistic matter; these approximations will be consistently
explained further on. We employ the Fourier transform solutions, which always make life easier
in case of waves or oscillations. Reminding the general principles: given a whatever function of
spacetime f (t, x), we can construct its Fourier transform (and inverse) for time variable (using
the symmetric scaling convention) as,
ˆ ∞ ˆ ∞
1 −iωt 1
f (ω, x) = √
b f (t, x) e dt, f (t, x) = √ fb(ω, x) eiωt dω, (93)
2π −∞ 2π −∞

11
When applying the same to the function ϕµν including the Green’s function solution, we have

ˆ ˆ ˆ Tµν t − |x−y | , y
 
4G c
ϕbµν (ω, x) = ϕµν (t, x) e−iωt dt = √ dt e−iωt d3 y =
c4 2π |x − y |
ˆ ˆ

Tµν et, y −iω et+ |x−y
 
4G |
= √ det e c
d3 y =
c4 2π |x − y |
ˆ ˆ −iω |x−y | ˆ b
4G
−iωet e c
3 4G Tµν (ω, y ) −iω |x−y | 3
= √ Tµν t, y e
e dt
e d y= 4 e c d y, (94)
4
c 2π |x − y | c |x − y |

where e t = t − |x − y |/c. The first row of equation (94) defines the Fourier transform, the second
reflects the equation (92), the third transforms the variable t to e t, and defines again the Fourier
transform. We now approximate the source as isolated, far away, and slowly moving. We consider
the center of the source at a spatial distance r, with the opposite parts at distances r + δr where
δr ≪ r. Since it is slowly moving, most of the radiation emitted will be at frequencies ω sufficiently
low that δr ≪ ω −1 . (Essentially, light traverses the source much faster than the components of
the source itself do.)
|x−y | r
Under these approximations, the term e−iω c /|x −y | can be replaced by e−iω c /r and brought
outside the integral. This leaves us with
r ˆ
4G e−iω c
ϕbµν (ω, x) = 4 Tbµν (ω, y ) d3 y. (95)
c r

We do not need to calculate all the components of ϕbµν (ω, x), since the harmonic gauge condition
∂µ ϕµν (t, x) = 0 in Fourier space implies

i
ϕb 0ν = ∂i ϕb iν . (96)
ω

We, therefore, only need to concern ourselves with the spacelike components of ϕbµν (ω, x). From
(95), we want to take the integral of the spacelike components of Tbµν (ω, y ). We integrate by parts,
ˆ ˆ   ˆ  
ij 3 i b kj
T (ω, y ) d y = ∂k y T
b d y − y i ∂k Tb kj d3 y,
3
(97)

where the first term is a surface integral which will vanish since the source is isolated, while the
second can be related to Tb 0j by the Fourier-space version of ∂µ T µν = 0,

∂k Tb kµ = −iω Tb 0µ . (98)

Thus,
ˆ ˆ ˆ
iω  i b 0j 
Tb ij (ω, y ) d3 y = iω i b 0j
yT d y= 3
y T + y j Tb 0i d3 y =
2
ˆ h  ˆ
iω i j b 0k

i j

0k
i
3 ω2
= ∂k y y T − y y ∂k T
b d y=− y i y j Tb 00 d3 y. (99)
2 2

The third integral of Eq. (99) is justified since we know that the left-hand side is symmetric in i
and j, while the fourth and fifth integrals are simply repetitions of reverse integration by parts

12
and conservation of T µν . It is conventional to define the quadrupole moment tensor of the energy
density of the source,
ˆ
qij (t) = y i y j T 00 (t, y ) d3 y, (100)

a constant tensor on each surface of constant time. In terms of the Fourier transform of the
quadrupole moment, our solution takes on the compact form
r
2Gω 2 e−iω c
ϕbij (ω, x) = − 4 qbij (ω), (101)
c r
or, transforming back to t by the inverse Fourier transform, we can absorb the factor −ω 2 into
a second time derivative, as well as e−iωr by transforming to the retarded time, following the
previously used steps in reverse. Thus, we finally arrive at the quadrupole formula
ˆ ˆ
1 2G 2 iω(t− rc ) 1 2G d2
ϕij (t, x) = − √ ω e qbij (ω) dω = √ eiωtret qbij (ω) dω =
2π c4 r 2π c4 r dt2
2G d2
= 4 [qij (tret )] . (102)
c R dt2
The gravitational wave produced by an isolated nonrelativistic object is, therefore, proportional
to the second derivative of the quadrupole moment of the energy density at the point where the
observer’s past light cone intersects the source.
In contrast, the leading contribution to electromagnetic radiation comes from the changing
dipole moment of the charge density. The difference can be traced back to the universal nature of
gravitation. A changing dipole moment corresponds to the motion of the center of density - charge
density in the case of electromagnetism and energy density in the case of gravitation. While there
is nothing to stop the center of charge of an object from oscillating, the oscillation of the center
of mass of an isolated system violates the conservation of momentum. (You can shake a body up
and down, but you and the Earth shake slightly in the opposite direction to compensate.) The
quadrupole moment, which measures the shape of the system, is generally smaller than the dipole
moment, and for this reason (as well as the weak coupling of matter to gravity) gravitational
radiation is typically much weaker than electromagnetic radiation.
It is always illustrative to take a general solution and apply it to a specific case of interest.
One case of interest is the gravitational radiation emitted by a binary star (two stars in orbit
around each other). For simplicity, let us consider two stars of mass M in a circular orbit in
the x1 -x2 plane, at a distance R from their common center of mass. Circular orbits in the
Newtonian approximation (where the orbital period of both stars is T = 2πR/V ) are characterized
by equating the gravitational and centrifugal forces,
r r
GM 2 MV 2 GM GM
2
= , giving V = and Ω = , (103)
(2R) R 4R 4R3

where Ω is the angular frequency (hereafter, we distinguish Ω as the angular frequency of the
orbiting system and ω as the angular frequency of the gravitational wave, where ω = 2Ω due to
two "tidal maxima" of the wave during one orbital period). Then, we can express the trajectories
of stars A and B as

x1A = R cos Ωt, x2A = R sin Ωt, x1B = −R cos Ωt, x2B = −R sin Ωt, (104)

13
while the corresponding energy density is

T 00 (t, x) = M δ(x3 ) δ(x1 − R cos Ωt) δ(x2 − R sin Ωt) + δ(x1 + R cos Ωt) δ(x2 + R sin Ωt) .
 

(105)

Following this, we can integrate Eq. (100) (using the principle given in Eq. (69)),

q11 = 2M R2 cos2 Ωt, q22 = 2M R2 sin2 Ωt, q12 = q21 = 2M R2 cos Ωt sin Ωt, qi3 = 0. (106)

We get the components of the metric perturbation from (101) as

 
cos 2Ωtret sin 2Ωtret 0
8GM
ϕij (t, x) = − 4 Ω2 R2  sin 2Ωtret − cos 2Ωtret 0 . (107)
c r
0 0 0

The remaining components of ϕµν could be derived from the harmonic gauge condition.
Another approach to construct the quadrupole moment tensor qij (t) is based on the evaluation
of the "second moment of mass" (moment of inertia) of the nonrelativistic isolated two-body
system (of two masses m1 and m2 where M = m1 + m2 ) in the center-of-mass frame where
x 0 = x 1 − x 2 is the relative coordinate, the center-of-mass coordinate x CM and the reduced mass
µ are (usually) given as
m1 x 1 + m2 x 2 m1 m2
x CM = , µ= . (108)
M M
Then, the second moment of mass can be written as

qij = m1 xi1 xj1 + m2 xi2 xj2 = M xiCM xjCM + µxi0 xj0 . (109)

Following the last expression, we can expand this as

m1 xi1 + m2 xi2 m1 xj1 + m2 xj2 m1 m2 i

M + (x1 − xi2 )(xj1 − xj2 ) =
M M M
m2 xi xj + m1 m2 xi1 xj2 + m1 m2 xj1 xi2 + m22 xi2 xj2 m1 m2 i j
= 1 1 1 + (x1 x1 − xj1 xi2 − xi1 xj2 + xi2 xj2 )
M M
1 h i
= m1 (m1 + m2 ) xi1 xj1 + m2 (m1 + m2 ) xi2 xj2 = m1 xi1 xj1 + m2 xi2 xj2 . (110)
M
Therefore, if we choose the origin of the coordinate system at x CM = 0, the quadrupole moment
becomes the same as that of a particle of mass µ described by the coordinate x 0 (t). In this CM
frame, the mass density is then

ρ(t, x) = µδ (3) (x − x 0 (t)) , (111)

and the quadrupole moment is

qij (t) = µxi0 (t)xj0 (t). (112)

We can thus relate the gravitational radiation emitted by such a two-body system to the mass
quadrupole moment of this system whose relative coordinate performs a given periodic motion,
say simple harmonic oscillations.

14
3 Energy and momentum of gravitational waves emitted from a
two-body system
To derive the energy emitted via gravitational radiation, let us consider vacuum Einstein’s equa-
tions to second order and see how the result can be interpreted as an energy-momentum tensor
for the gravitational field. Once again, if we consider the metric as gµν = ηµν + hµν , then at
(1) (1)
first order we have Gµν [η + h] = 0, where Gµν is the Einstein’s tensor expanded to first order in
hµν . These equations determine hµν up to (unavoidable) gauge transformations, so to satisfy the
(2)
equations to second order, we have to add a higher-order perturbation, gµν = ηµν + hµν + hµν .
The second-order version of Einstein’s equations consists of all terms either quadratic in hµν or
(2)
linear in hµν . Since any cross terms would be of at least third order, we have

G(1) (2) (2)

µν [η + h ] + Gµν [η + h] = 0. (113)
(2)
Here, Gµν is part of the Einstein tensor, which is of second order in the metric perturbation. It
can be computed from the second-order Ricci tensor (involving the nonlinear terms of the Ricci
tensor (not only the linear ones as we did before in the case of the vacuum solution used to find
the solution of perturbations), which is given by
(2)
Rµν = ∂ρ Γρµν − ∂ν Γρµρ + Γρρλ Γλµν − Γρµλ Γλνρ . (114)

The solution of the first nonlinear term within the TT gauge is

 
ρ 1 ρσ 1
λ λ
η (∂ρ hλσ + ∂λ hρσ − ∂σ hρλ ) = Γλµν ∂ρ hρλ + ∂λ hρρ − ∂σ hσλ = 0,


Γρλ Γµν = Γµν (115)
2 2

because the first and third terms cancel (they are the same in the transverse gauge, using the
symmetry in the metric η αβ = η βα and following, therefore, the identities η ρσ ∂ρ hλσ = η σρ ∂σ hλρ ),
and the middle term is zero due to the traceless choice of the gauge. The second nonlinear term
can be expanded as
1
−Γρµλ Γλνρ = − η ρσ (∂µ hλσ + ∂λ hµσ − ∂σ hµλ ) η λσ (∂ν hρσ + ∂ρ hνσ − ∂σ hνρ ) =
4
1
= − ∂µ hρλ + ∂λ hρµ − ∂ ρ hµλ η λσ (∂ν hρσ + ∂ρ hνσ − ∂σ hνρ ) =


4
1
= − ∂µ hρσ + ∂ σ hρµ − ∂ ρ hσµ (∂ν hρσ + ∂ρ hνσ − ∂σ hνρ ) =


4
1
= − (∂µ hρσ )(∂ν hρσ ) + (∂µ hρσ )(∂ρ hνσ ) − (∂µ hρσ )(∂σ hνρ ) + (∂ σ hρµ )(∂ν hρσ )+
4
+(∂ σ hρµ )(∂ρ hνσ ) − (∂ σ hρµ )(∂σ hνρ ) − (∂ ρ hσµ )(∂ν hρσ ) − (∂ ρ hσµ )(∂ρ hνσ )+
+(∂ ρ hσµ )(∂σ hνρ ) ,

(116)

the second term cancels with the third term, and the fourth term cancels with the seventh term
due to the obvious symmetry of the metric, as was already described above. Equation (116) thus
simplifies to
1
−Γρµλ Γλνρ = − (∂µ hρσ )(∂ν hρσ ) + (∂ σ hρµ )(∂ρ hνσ ) − (∂ σ hρµ )(∂σ hνρ )−
4
−(∂ ρ hσµ )(∂ρ hνσ ) + (∂ ρ hσµ )(∂σ hνρ ) ,

(117)

15
where, however, the second and the fifth terms are for the same reason identical due to the
symmetry of indices, ∂ σ hρµ = ∂ ρ hσµ as well as ∂ρ hνσ = ∂σ hνρ , while for the third and the fourth
terms we can also apply this argument. We may thus yet simplify this identity to
1 1 1
−Γρµλ Γλνρ = − (∂µ hρσ )(∂ν hρσ ) + (∂ σ hρµ )(∂ρ hνσ ) − (∂ σ hρµ )(∂σ hνρ ). (118)
4 2 2
We can rewrite the second and the third term on the right-hand side of Eq. (118), using the rule
for the derivative of the product of two functions as
1 σ ρ 1 1
(∂ hµ )(∂ρ hνσ ) = ∂ σ (hρµ ∂ρ hνσ ) − hρµ ∂ρ ∂ σ hνσ , (119)
2 2 2
1 σ ρ 1 σ ρ 1
− (∂ hµ )(∂σ hνρ ) = − ∂ (hµ ∂σ hνρ ) + hρµ ∂ σ ∂σ hνρ , (120)
2 2 2
where, due to the gauge condition (34), the last term in (119) vanishes. Now, let’s label the two
linear terms of the Ricci tensor from Eq. (10) as being the "connections" of the second order (due
to the derivatives of the Christoffel’s symbols); we can then rewrite the complete second-order
Ricci tensor (see Eq. (13) for its linear part only) in this sense as

(2) 1
Rµν = ∂ρ ∂µ hρν + ∂ρ ∂ν hρµ − ∂µ ∂ν h − □hµν −
2 
1 ρσ σ ρ σ ρ ρ
− (∂µ h )(∂ν hρσ ) + ∂ (hµ ∂ρ hνσ ) − ∂ (hµ ∂σ hνρ ) + hµ □hνρ , (121)
2

where the d’Alembertian in the last term comes from ∂ σ ∂σ = ∂σ ∂ σ = □. We now evaluate the
second-order Ricci scalar by contracting the second row of Eq. (121) by η µν (since the first-order
expression, that is, the first row in Eq. (121), was already calculated in Eq. (13)),
 
(2) µρ 1 1 ν ρσ σ νρ σ νρ νρ
R =∂ρ ∂µ h − □h − (∂ h )(∂ν hρσ ) + ∂ (h ∂ρ hνσ ) − ∂ (h ∂σ hνρ ) + h □hνρ . (122)
2 2
(2) (2) (1)
Now, since Gµν = Rµν − 21 ηµν R(2) and, following Eq. (113) where we can put Gµν (η + h(2) ) =
8πG
t , we simply define
c4 µν

8πG
G(2)
µν (η + h) = − tµν , (123)
c4
we formally distinguish tµν as the energy-momentum tensor specifically for the gravitational field
in the weak field regime.
We will not describe here the Bianchi identity and other mathematical tools which may lead
us too far aside (see Caroll and others for the detailed stuff), stating only that it implies ∇µ Rρµ =
1 µ µ
2 ∇ρ R which is equivalent to energy conservation ∇ Gµν = 0 and, therefore, ∇ Tµν = 0. For this
reason, we have in the flat space (weak field regime) the modified energy-momentum conservation
condition

∂µ tµν = 0. (124)
´b
By averaging over a macroscopic region, [ a f (x) dx]/(b − a), we can yet simplify the Einstein
tensor. At this point, it is important to note that we are working with functions constructed from
small perturbations. The difference in the values a and b of the function at any point will also

16
be small. If the wavelength of the gravitational wave is much larger than the region b − a we are
considering, f (a) and f (b) will also be practically the same. So let’s go through the right-hand
sides of equations (121) and (122) term by term and see which ones can be neglected (noting
again that we "are" in the TT gauge): The first four terms and the last term in (121) vanish due
to the conditions ∂µ hµα = 0, h = hαα = 0, and □hµν = 0 (see the explanations above). In contrast,
the second and the third terms from the end vanish due to neglecting the average of the total
derivative, as described closely above. The same principles apply in Eq. (122) where also the third
term (first term in square bracket) vanishes due to the same macroscopic condition within the
averaged value (remembering that u′ v = (uv)′ − uv ′ where the middle averaged term vanishes, so
u′ v ≃ −uv ′ ), that is,
1 1
− ⟨(∂ ν hρσ )(∂ν hρσ )⟩ = ⟨hρσ □hρσ ⟩ = 0. (125)
4 4
Finally, we are only left with a single term (the fifth term on the right-hand side of Eq. (121)) for
the effective stress-energy of gravitational waves,

(2) 1 8πG
Rµν (η + h) = − ⟨(∂µ hρσ )(∂ν hρσ )⟩ = − 4 tµν , (126)
4 c
that is,
c4
tµν = ⟨(∂µ hρσ )(∂ν hρσ )⟩ , (127)
32πG
where we only readjusted Eq. (126) and transferred the "Einstein gravity coefficient" to the
opposite side of the equation.
Regarding the derived energy-momentum tensor, we proceed with the T00 component that
describes the energy density. Therefore, we can find the total energy of the gravitational radiation
contained within the surface Σ of constant time. Using the symmetries of the system, we put the
source of the radiation into the origin, which is especially important when we work with the
retarded time,
ˆ
E= t00 d3 x. (128)
Σ

We can also calculate the energy loss ∆E due to radiation through a sphere S with a radius R
per second because the T0µ components describe the energy flux in the direction µ as
ˆ
∆E = t0µ nµ d2 x dt, (129)
S

where the integral is taken over a spacelike two-sphere in infinity and some interval in time, and
nµ is a unit spacelike vector normal to S. After a little rearrangement and a transcription of the
isotropic situation to spherical coordinates where the only component t0µ nµ is t0r , we can write
the same equation (not in the infinity but at least in a considerable distance r) as
ˆ
dE
= t0r r2 dΩ. (130)
dt S

Then, according to Eq. (127),

c4 D E
t0r = (∂0 hαβ
TT )(∂ hTT
r αβ ) , (131)
32πG

17
where we further employ the derivation of the trace-reversed tensor (101) based on the second-
order time derivative of the quadrupole momentum, noting that ϕµν = hµν in the TT gauge. We
also advancingly use the fact that the amount of radiated energy (129) can be written in terms
of the radiated power P as
ˆ
∆E = P dt, (132)

where the power is then given by the integrand of Eq. (130) in spherical coordinates.
Now, we need to impose the TT gauge to the trace-reversed perturbation so it changes to the
weak field perturbation, and we can use it in equation (127). We have to find a TT tensor qTijT
constructed from q ij . We can use this to find T0µ , after which we can change qTijT back to q ij
without information loss. First, we begin by projecting q ij on its traceless component Qij (whose
Tr Qij = Qii = 0),
1
Qij = q ij − δ ij δkl q kl , (133)
3
second, to make Qij transversal, we want to project its components on a transversal image.
Therefore, we will use the projection operator
xb xa
Pab (x) = δab − which is equivalent to Pab (n) = δab − nb na ,
, (134)
r2
 1 2 3
1 2 3 i 2 x x x
so ni ni = 1, that acts as


where x = x , x , x so x xi = r , and n = , ,
r r r
1
Qij i j ij
T T = Pa Qab Pb − Pab Qab P . (135)
2
After projecting Qij on its transversal image, you can use the projection again, but it stays the
same transversal image. Since a general property of a projection operator is
P 2 = P ∧ Pab Pca = Pcb , (136)
xb xa xa xc xa xc xb xa xb xa xa xc
   
2 b a
P = δa − 2 δc − 2 = δab δca − δab 2 − δca 2 + 2 =
r r r r r r2
xb xc xb xc xb r 2 xc xb xc
= δcb − 2 − 2 + = δc
b
− = P, in other words (137)
 r  r r4 r2
P 2 = δab − nb na (δca − na nc ) = δab δca − δab na nc − δca nb na + nb na na nc =
= δcb − nb nc − nb nc + nb nc = δcb − nb nc = P (138)
using this, we can check that qTijT = Qij ij
T T is traceless and transversal: Let’s first check that QT T
is transversal, that is, x · Qij
T T = 0 in the following way,
1
xi Qij i j
T T = (xi Pa )Qab Pb − Pab Qab (xi P ),
ij
checking in particular that
2
xi xa xi xi
 
xi Pai = xi δai − 2 = xa − 2 xa = xa − xa = 0, in other words (139)
r r
i i i i


ni Pa = ni δa − n na = xa − n ni na = na − na = 0, (140)
xi xj xi xi
 
ij ij
xi P = xi δ − 2 = xj − 2 xj = xj − xj = 0. This is also (141)
r r
ni P = ni δ − n n = n − ni ni nj = nj − nj = 0.
ij ij i j j


(142)

18
Second, we check that Qij i
T T is traceless, that is, Qi = 0, by

1
Qii = gij Qij i i
T T = Pa Qab Pib − Pab Qab Pi , where
2
xi xa 
 
xi xb  xa xb
Pai Pib = δai − 2 δib − 2 = δab − 2 = Pab ,
r r r
i i i


Pa Pib = δa − n na (δib − ni nb ) = δab − na nb = Pab ,
1 i 1 i xi xi
 
1
Pi = δi − 2 = (3 − 1) = 1, that is,
2 2 r 2
1 i 1 i 1
δ − ni ni = (3 − 1) = 1,


P =
2 i 2 i 2
i
Qi = Pab Qab − Pab Qab = 0. (143)

Now, inserting Eq. (101) into Eq. (131), we continue as

2G ∂ h αβ i 2G ∂t
ret ∂
h i
∂0 hαβ
TT = Q̈TT (tret ) = Q̈αβ
TT (tret ) , (144)
c4 r ∂t c4 r ∂t ∂tret
 
2G ∂ 1 αβ 2G ∂t ∂tret ∂ h αβ i 2G h i
∂r hTαβT = 4 Q̈T T (tret ) = 4 Q̈T T (tret ) − 4 2 Q̈TαβT (tret ) , (145)
c ∂r r c r ∂r ∂t ∂tret c r

where we use Eq. (89) for the derivative ∂tret /∂t which however, due to negligible β term, we
regard as 1. Due to the spherical symmetry (one radial spatial coordinate r), we set the radial
derivative ∂r of the quadrupole equal to c times ∂t (the time derivative), except the direct spatial
derivative of r in the denominator. We thus obtain
2G h ...αβ i
∂0 hαβ
TT = Q T T (tret ) (146)
c4 r
2G ...T T
h i 2G h i
∂r hTαβT = 5 Q αβ (tret ) − 4 2 Q̈TαβT (tret ) . (147)
c r c r
Because we are far from the source, we may neglect the last term in Eq. (147). We can thus write
an expression for the power P = dE/dt radiated by a gravitational source,
ˆ  ˆ
c4 4G2 2 ...αβ ...T T G ...αβ ...T T

dE
= 9 2
r Q TT Q αβ dΩ = 5
Q TT Q αβ dΩ . (148)
dt S 32πG c r S 8πc

Since we want to achieve a general solution of the integral, we transform this back to a non-
transversal form. We use again the property of the projection operator P 2 = P ,
...T T ...ij ... ...cd
  
a b 1 ab i j 1 ij
Q ij Q T T = Pi Pj − P Pij Pc Pd − P cd P Q ab Q . (149)
2 2

We can split the two brackets with the projection operator into four terms
1 1 1 1
Pia Pjb Pci Pdj = Pca Pdb , − Pia Pjb P cd P ij = − Pcd P ab , − P ab Pij Pci Pdj = − Pcd P ab ,
2 2 2 2
1 ab 1 1
P Pij P cd P ij = Pcd P ab Pii = Pcd P ab , (150)
4 4 2

19
(see Eq. (143) for justification of the last equation’s solution) we need to solve the two operators,

xa xc xb xd xb xd xa xc xa xc xb xd
  
a b a b
Pc Pd = δ c − 2 δd − 2 = δca δdb − δca 2 − δdb 2 + , (151)
r r r r r4
xc xd  ab xa xb xa xb xa xb xc xd
 
ab
 xc xd
Pcd P = δcd − 2 δ − 2 = δ ab δcd − δ ab 2 − δcd 2 + . (152)
r r r r r4
Substituting this into Eq. (149) gives
... ...cd ... ...ab xb xd ... ...a xa xc ... ...b xa xc xb xd ... ...
Pca Pdb Q ab Q = Q ab Q − 2 Q ab Q d − 2 Q ab Q c + Q ab Q cd , (153)
R R R4
1 ... ...cd 1 ... ... xc xd ... ...cd xa xb ... ... xa xb xc xd ... ...
− Pcd P ab Q ab Q = − Q Q + QQ + Q Q− Q ab Q cd . (154)
2 2 2R2 2R2 ab 2R4
Remembering that the quadrupole moment of mass distribution is still traceless (Q = 0), we can
rewrite equation (149) as
...ij ...T T ... ...ab xb xd ... ...a xa xc ... ...b xa xc xb xd ... ...
Q T T Q ij = Q ab Q − 2 Q ab Q d − 2 Q ab Q c + Q ab Q cd . (155)
R R 2R4
We must integrate this over dΩ to obtain the total power. To perform the integration, we yet
introduce the solutions for a surface S with radius R (cf., e.g., Michele Maggiore, Gravitational
waves, 2008, pp. 104-105)
ˆ ˆ
dΩ a b 1 2 ab dΩ a b c d 1  
x x = R η , x x x x = R4 η ab η cd + η ac η bd + η ad η bc . (156)
4π 3 4π 15
These identities and their generalization to an arbitrary number of n’s can be found as follows.
For an odd number of ni the integral vanishes because the integrand is odd under parity. For
an even number of ni , we use the fact that the tensor ni1 ni2 ... ni2k is symmetric, and therefore
its integral can only depend on the symmetrized product of Kronecker deltas. Once the tensor
structure is fixed, the overall constant is obtained by contracting all indices. This all gives
ˆ
dΩ 1

ni1 ni2 ... ni2k = δi1 i2 δi3 i4 ... δi2k−1 i2k + ... , (157)
4π (2k + 1)!!
where !! denote the "double factorial" n(n − 2)(n − 4) ... and the final dots denote all possible
pairing of indices. In particular, we thus have
ˆ ˆ
dΩ a b 1 ab dΩ a b c d 1  ab cd 
n n = η , n nnn = η η + η ac η bd + η ad η bc . (158)
4π 3 4π 15
which is analogous to Eq. (156).
We can now use these particular solutions to calculate the integral of the total energy release
(power of the gravitational radiation)
ˆ
dΩ ...ij ...T T
Q Q =
4π T T ij
... ...ab η bd ... ...a η ac ... ...b η ab η cd + η ac η bd + η ad η bc ... ...
= Q ab Q − Q ab Q d − Q ab Q c + Q ab Q cd =
3 3 30
... ...ab 2 ... ...ab 1 ... ... ... ...ab ... ...ba
  2 ... ...ab
= Q ab Q − Q ab Q + Q Q + Q ab Q + Q ab Q = Q ab Q , (159)
3 30 5

20
where we use again Q = 0 and Qab = Qba . Inserting this into Eq. (148), we finally find the famous
quadrupole radiation formula, first derived by A. Einstein,
dEgw G D ... ...ab E
= 5 Q ab Q , (160)
dt 5c
...
where, again, Q ab must be evaluated at the retarded time tret . Sometimes, in explicit computa-
dE ... ... ...
tions, it is more practical to use qij instead of Qij , then we have dtgw = 5cG5 q ab q ab − 13 q cc q cc
which is not traceless. Some authors, e.g., Landau and Lifshitz, Vol. II ´ (1979), or Carroll (1997),
3
define the quadrupole moment with a different normalization, QLL ij = d x ρ(t, x)(3xi xj − r2 δ ij ),
where the superscript "LL" stands for Landau & Lifshitz. This is larger byD the factor 3 than
dEgw G
... ...ab E
the definition in Eq. (160); the quadrupole formula, therefore, is dt = 45c5 Q ab Q , and all
other equations involving Qij must be rescaled similarly.
Let’s now consider the evolution of a binary star on a circular orbit in the xy plane. The stars
have masses m1 and m2 and separation a; they thus have the reduced mass µ and they orbit each
other with an angular frequency ω and the total orbital energy Eorb of the binary system,
r
m1 m2 GM 1 Gm1 m2 GµM
m1 a21 + m2 a22 Ω2 −


µ= , Ω= 3
, Eorb = =− , (161)
M a 2 a 2a
where M = m1 + m2 is the total mass and where the term in the bracket in the last equation is
equal µa2 in the center-of-mass coordinate system, as is already described in Eq. (110). We now
follow the principles manifested by Eq. (112) of the quadrupole moment (qij (t) = µxi0 (t)xj0 (t) for
reminder); we consider for the moment an isolated binary system with masses m1 and m2 , and
we assume that the relative coordinate x0 is performing a circular orbit with constant radius R.
We choose the system’s trajectory in xy plane, so in the center-of-mass frame, the xi0 components
are given as

x0 (tret ) = R cos Ωtret , y0 (tret ) = R sin Ωtret , z0 (tret ) = 0, (162)

so, the components of the mass quadrupole moment are

q11 (tret ) = µR2 cos2 Ωtret , q22 (tret ) = µR2 sin2 Ωtret , q12 (tret ) = µR2 sin Ωtret cos Ωtret , (163)

(where q21 (tret ) = q12 (tret )) while the other components vanish. In this sense, we have the time
second derivatives

q̈11 (tret ) = −2µR2 Ω2 cos 2Ωtret , q̈22 (tret ) = 2µR2 Ω2 cos 2Ωtret ,
q̈12 (tret ) = −2µR2 Ω2 sin 2Ωtret , (164)

and the time third derivatives

... ...
q 11 (tret ) = 4µR2 Ω3 sin 2Ωtret , q 22 (tret ) = −4µR2 Ω3 sin 2Ωtret ,
...
q 12 (tret ) = −4µR2 Ω3 cos 2Ωtret . (165)

The consistent matrix notation of the previous thus will be

 
sin 2Ωtret − cos 2Ωtret 0
...
q ij (tret ) = 4µR2 Ω3 − cos 2Ωtret − sin 2Ωtret 0 . (166)
0 0 0 ij

21
 ...
Since this matrix is traceless, according to Eq. (133), it also equals Q ij . Plugging it into the
quadrupole radiation formula (160) where we may the dEgw /dt regard as the "luminosity" of the
gravitational waves radiation Lgw , we have

G D ... ...ab E 32Gµ2 R4 Ω6 32G4 µ2 M 3

Lgw = Q ab Q = = , (167)
5c5 5c5 5c5 R5
where the additional factor 2 comes from summing over all products of the corresponding elements
⟨sin2 2Ωtret + cos2 2Ωtret + cos2 2Ωtret + sin2 2Ωtret ⟩ of the two identical matrices from Eq. (166).
As the gravitating system loses energy by emitting radiation, the distance between the two
bodies shrinks at a rate
dEorb GµM dR dR 64G3 µM 2
Lgw = − =− , that is, =− . (168)
dt 2R2 dt dt 5c5 R3
The orbital frequency increases according to the decrease in the orbital distance as

f˙ Ω̇ 3Ṙ
= =− , (169)
f Ω 2R
and, therefore, if the present separation of the two stars is Rinit , then from Eq. (168), the binary
system will merge in the gravitational wave inspiral time
ˆ 0
5c5 3 5c5 Rinit
4 5c5 M 1/3
tGW = − R dR = = . (170)
64G3 µM 2 Rinit 256G3 µM 2 256G5/3 m1 m2 Ω
8/3
init

As long as Ω̇ ≪ Ω2 , we are in the quasi-circular motion regime. From Eq. (169), we see that
Ṙ = − 23 R Ω̇ 2 Ω̇ 2 Ω̇ 2
Ω = − 3 RΩ Ω2 = − 3 Vϕ Ω2 , then, as long as the condition Ω̇ ≪ Ω is fulfilled, |Ṙ| is
much smaller than the tangential orbital velocity Vϕ = RΩ. The approximation of a circular orbit
with a slowly varying radius is then applicable.
To evaluate the amplitude of the gravitational wave, we first apply the P operator from
Eq. (134) to the traceless form of Eq. (102),
2G
hTijT = Q̈ij (tret ), (171)
c4 r
remembering that in the TT gauge hij = ϕij and Q̈ij = q̈ij . First, we realize that if the propagation
direction of a gravitational wave is z, then the Pij operator becomes (because both nx and ny are
zero, and ni nj = 0 if i ̸= j, so Pxx = δxx = 1, Pyy = δyy = 1, and Pzz = δzz − nz nz = 1 − 1 = 0)
 
1 0 0
1 1
Pij = Pij = P ij = 0 1 0 , so Pia q̈ab Pbj − Pab q̈ab P ij = q̈ij − q̈aa P ij =
2 2
0 0 0 ij
     
q̈11 q̈12 0 1 0 0 (q̈11 − q̈22 )/2 q̈12 0
1
= q̈21 q̈22 0 − Tr(q̈) 0 1 0 =  q̈21 −(q̈11 − q̈22 )/2 0 . (172)
2
0 0 0 0 0 0 0 0 0

We directly see the two polarization amplitudes of the gravitational waves propagating in the z
direction
G 2G
h+ (t, z) = (q̈11 − q̈22 ), h× (t, z) = q̈12 , (173)
c4 r c4 r

22
evaluated at tret . To calculate the amplitude of a wave that in the coordinate frame with axes
(x, y, z) propagates to an arbitrary direction n, we introduce another orthogonal frame (x′ , y ′ , z ′ )
that is identical with (x, y, z) at the beginning and performs two rotations, first around the
axis z ≡ z ′ and subsequently around the axis x′ , after which both frames are inclined in both
angular directions by angles ϕ and θ. If the wave now propagates in the primed frame along
the axis z ′ , then the vector n has coordinates n′i = (0, 0, 1) while in the unprimed frame ni =
(sin θ cos ϕ, sin θ sin ϕ, cos θ). Equation (173) will in the primed frame differ by the argument
(t, n) of the amplitudes (instead of (t, z)) and the q̈ ’s now will be primed. We can construct the
rotational matrix R of two rotations around axes z ′ and x′ , respectively, simply as
    
cos ϕ sin ϕ 0 1 0 0 cos ϕ cos θ sin ϕ sin θ sin ϕ
R = − sin ϕ cos ϕ 0 0 cos θ sin θ  = − sin ϕ cos θ cos ϕ sin θ cos ϕ , (174)
0 0 1 0 − sin θ cos θ 0 − sin θ cos θ
and solve the transformation for q̈ ′ as q̈ij′ = (RT q̈ R)ij , where RT is the transposed matrix R.
Applying this to the relevant terms of the primed version of Eq. (173) and merging symmetric
terms into single ones (with doubled values), we obtain
′
q̈11 = q̈11 cos2 ϕ − q̈12 sin 2ϕ + q̈22 sin2 ϕ, (175)
′ cos θ sin 2ϕ cos θ sin 2ϕ
q̈12 = q̈11 + q̈12 cos θ cos 2ϕ − q̈13 sin θ cos ϕ − q̈22 + q̈23 sin θ sin ϕ (176)
2 2
′
q̈22 = q̈11 cos2 θ sin2 ϕ + q̈12 cos2 θ sin 2ϕ − q̈13 sin 2θ sin ϕ + q̈22 cos2 θ cos2 ϕ −
− q̈23 sin 2θ cos ϕ + q̈33 sin2 θ, (177)
and, therefore,
G
q̈11 cos2 ϕ − cos2 θ sin2 ϕ − q̈12 sin 2ϕ 1 + cos2 θ + q̈13 sin 2θ sin ϕ +



h+ (t, n) = 4
c r
+ q̈22 sin2 ϕ − cos2 θ cos2 ϕ + q̈23 sin 2θ cos ϕ − q̈33 sin2 θ ,

G
h× (t, n) = 4 |(q̈11 − q̈22 ) cos θ sin 2ϕ + 2 q̈12 cos θ cos 2ϕ − 2 q̈13 sin θ cos ϕ +
c r
+ 2 q̈23 sin θ sin ϕ| . (178)
This equation allows us to calculate the angular distribution of the quadrupole radiation once qij
is given.
Let’s now again remind that we consider the evolution of a binary star on a circular orbit in
the xy plane, so if we plug Eq. (3) into (178), after some arithmetic, we have
4GµR2 Ω2 1 + cos2 θ
 
h+ (t, n) = cos (2Ωtret + 2ϕ) , (179)
c4 r 2
4GµR2 Ω2
h× (t, n) = cos θ sin (2Ωtret + 2ϕ) , (180)
c4 r
and, substituting the frequency f of gravitational wave (that is twice the orbital frequency) and
realizing that in this circular orbit approximation where Ω∆tret = ∆ϕ, we can shift the origin of
time so that Ωtret + ϕ → Ωtret , we get a modification of the above,
4GµR2 (πf )2 1 + cos2 θ
 
h+ (t, n) = cos (2πf tret ) , (181)
c4 r 2
4GµR2 (πf )2
h× (t, n) = cos θ sin (2πf tret ) . (182)
c4 r

23
We see that observing the gravitational waves binary system source pole-on (θ = 0, cos θ = 1),
the amplitudes of h+ and h× are identical while the phase is shifted by π/2; the wave is thus
circularly polarized. On the other hand, if we observe the system equator-on (θ = π/2, cos θ = 0),
the amplitude h× vanishes, and the amplitude h+ is of half magnitude, the wave is linearly
polarized.
We make another consideration; following the angular acceleration and substituting Eq. (168),
we find that

96G5/3 µM 2/3 Ω11/3

3Ω 96G7/2 µM 5/2
Ω̇ = − Ṙ = = . (183)
2R 5c5 R11/2 5c5
Now, we rearrange this equation to express the explicit mass-containing term (highlighted by the
bracket) in the dimension of mass [g]; this means that we have to raise the expression µM 2/3 by
the power 3/5, transfer it to the left-hand side of the equation, and leave on the right-hand side
the quantities Ω and Ω̇ observable due to the propagation of gravitational waves. We get
 3/5 3
5 c
µ 3/5
M 2/5
= Ω̇3/5 Ω−11/5 , (184)
3 8G

where the quantity on the left-hand side is the "chirp mass,"

(m1 m2 )3/5
Mc = (185)
(m1 + m2 )1/5

that can thus be measured from observations.

Implementing the chirp mass as a canonical quantity, we define the chirp as a rapid frequency
increase f˙. We follow Eq. (183) keeping in mind that the orbiting angular frequency is twice lower
than the frequency of the wave (which has two peaks during one orbit), so Ω = πf (not 2πf );
after little arithmetic we write its usual form
3 f
 8/3
96 c G
f˙ = πf Mc . (186)
5 G Mc c3
´ df c3 1

8/3 ´ tcoal
Integrating Eq. (186) as f 11/3 = 96 G
5 G Mc c3 πMc t dt, we see that f formally diverges
at a finite value of time (time of coalescence) that we denote tcoal ; in terms of the time remaining
to tcoal , the frequency of a gravitational wave is
5/8  3/8
c3

1 5
f (t) = . (187)
π GMc 256 (tcoal − t)

Inserting Eqs. (186) and (187) into Eq. (169), after a little algebra, we evaluate the expression

Ṙ 2f˙ 1
=− =− , (188)
R 3f 4 (tcoal − t)

which, after integration of the radius according to time, gives

ˆ ˆ 1/4
R t
dt′

dR tcoal − t
=− , that is R(t) = R0 , (189)
R0 R t0 4 (tcoal − t′ ) tcoal − t0

24
where R0 is the value of R at the initial time t0 . This means that after a long phase of relatively
smoothly decreasing R, there is a plunge phase where the approximation of a quasi-circular orbit
is no longer valid. By inversion of (186), we can also express the chirp mass as

c3 5 −8/3 −11/3 ˙ 3/5

 
Mc = π f f . (190)
G 96

Substituting the chirp mass Mc into the angle-independent part of Eq. (181) or (182), we obtain
(after some algebra and renaming the general distance r to the luminosity distance D) the scaling
amplitude
 5/3  2/3
4 GMc πf
h0 = , (191)
D c2 c

In this lowest order Newtonian approximation, the amplitudes depend on masses m1 and m2 only
through the chirp mass Mc . By simple inversion, we can thus calculate the luminosity distance
 5/3  2/3
4 GMc πf
D= , (192)
h0 c2 c

which is a method of measuring the luminosity distance using only gravitational wave observables.
This is extremely useful as an independent distance indicator in astronomy.
The amplitude of the emitted gravitational waves depends on the angle between the line of sight
and the axis of angular momentum; the formula for the amplitude contains angular factors of order
1. The relative strength of the two polarizations depends on that angle as well. If three (or more)
detectors observe the same signal, it is possible to reconstruct the full waveform and deduce many
details about the orbit of the binary system. As the oldest canonical example, the well-studied
pulsar PSR 1913+16 (the Hulse-Taylor pulsar) served a long time. It is expected to merge after
∼ 3.5 × 108 years. The binary system is roughly 5 kpc away from the Earth; the masses of the two
neutron stars are estimated to be ∼ 1.4 M⊙ each, and the present period of the system is ∼ 7 h and
45 min. The predicted rate of period change is Ṫ = −2.4 × 10−12 sec/sec, while the corresponding
observed value is in excellent agreement with the predictions, i.e., Ṫ = (−2.3±0.22)×10−12 sec/sec;
finally, the present amplitude of gravitational waves is of the order of h ∼ 10−23 at a frequency of
∼ 7 × 10−5 Hz.

25