82 ⌅ General Relativity and its Applications

This is a tensor equation, and must hold in any coordinate frame; therefore, in a generic
frame the equation of motion of the free particle is

u↵ uµ ;↵ = 0 . (3.121)

If we expand the covariant derivative

µ
u↵ uµ ;↵ = u↵ uµ ,↵ + u↵ ↵ u , (3.122)

and replace uµ = dxµ /d , Eq. 3.121 becomes

d2 xµ µ dx↵ dx
+ ↵ = 0, (3.123)
d 2 d d
which is the geodesic equation, Eq. 3.114. Thus, the geodesic equation can also be written
in the form 3.121, which describes the parallel transport of the tangent vector ~u along the
geodesic. This means that if we parallely transport the tangent vector from a point p to a
point p0 along the geodesic line, the transported vector in p0 will be tangent to the curve.
Consequently, a curve C with tangent vector ~u is a geodesic if

r~u ~u = 0 . (3.124)

Therefore, geodesics are those curves which parallel-transport their own tangent vectors.
The parameter along the geodesic is not unique. Let s be a new parameter; since dd =
d ds
ds d , Eq. 3.123 becomes

 µ ⌫ " ✓ ◆2 # ↵
d2 x↵ ↵ dx dx d2 s . ds dx
2
+ µ⌫ = 2
. (3.125)
ds ds ds d d ds

This equation reduces to Eq. 3.123 only if s is related to by a linear transformation

s = a + b, a, b = const , (3.126)

in which case the right-hand side of Eq. 3.125 vanishes. The parameters for which the
geodesic equation takes the form 3.123 are called affine parameters and are all related by
a linear transformation.
Eq. 3.121 describes timelike, spacelike, and null geodesics. If the geodesic is timelike, i.e.
~u · ~u < 0, it represents the wordline of a massive particle; in this case, by performing the
linear transformation in Eq. 3.126 it is possible to change the affine parameter in such a
way that the new parameter is the particle proper time (see Box 2-F) multiplied by c, i.e.
the parameter ⌧ appearing in Eq. 3.114. With this choice, ~u · ~u = 1. If the geodesic is a
null curve, i.e. ~u · ~u = 0, it represents the wordline of a massless particle; in this case it is
not possible to choose the proper time as affine parameter, because it cannot be defined;
the geodesic can be parametrized with a di↵erent affine parameter, for instance with the
arc length. If the geodesic is spacelike, i.e. ~u · ~u > 0, it does not represent the worldline of
a physical particle.

3.9 FERMI COORDINATES

In this section we shall show how to construct a coordinate frame adapted to an observer.
This frame is especially useful to describe physical experiments.
An observer in General Relativity is characterized by a timelike curve, ⌧ 7! (⌧ ), where
⌧ is the proper time (times c). The tangent vector to the curve is the four-velocity of the
 Affine connection and parallel transport ⌅ 83

!(")

e(0)* u

p(")
0

e(2)* e(1)*
v s
p

Σ(")

Figure 3.5: Fermi coordinates adapted to an observer. For simplicity, we do not show one of the
three space dimensions.

observer ~u, and ~u · ~u = 1. The observer can either be freely-falling (i.e. with vanishing
acceleration aµ = u⌫ uµ;⌫ = 0) or accelerated (aµ 6= 0).
It is possible to define, in a region close to the curve (a “worldtube” surrounding ),
coordinates {x⇤µ } adapted to the observer, called the Fermi coordinates, as follows (see
Fig. 3.5):
• At any point of the curve p0 (⌧ ) 2 , we can define a basis {~e(µ) ⇤
} of the vector space
⇤ ⇤ ⇤
Tp0 such that ~e(0) ⌘ ~u, and ~e(i) (i = 1, . . . , 3) are orthogonal to ~e(0) and to each
⇤
other. In addition, we impose that ~e(i) have norm one. Thus, the basis is orthonormal
⇤ ⇤
(~e(µ) · ~e(⌫) = ⌘µ⌫ ), and the timelike basis vector coincides with the four-velocity of the
observer.
• At each value of ⌧ , the three spacelike vectors ~e(i)
⇤
locally define a three-dimensional
spacelike surface ⌃, to which they are tangent in p0 (⌧ ).
• Given a point p in the worldtube, such that p 2 ⌃(⌧ ), let us consider a spacelike
geodesic from p0 (⌧ ) to p on the surface ⌃(⌧ ). Let ~v be the tangent vector in p0 (⌧ )
to this geodesic with unit length. Since the geodesic belongs to ⌃(⌧ ), ~v is a linear
⇤
combination of the three spacelike vectors ~e(i) (i = 1, . . . , 3), i.e.
⇤
~v = v i~e(i) . (3.127)

• Let s be the affine parameter of the geodesic from p0 (⌧ ) to p. We define the Fermi
coordinates of the point p as follows:
x⇤0 = ⌧ , x⇤i = sv i . (3.128)
It can be shown that s is also the proper length of the path (as defined in Eq. 2.211)
of this spacelike geodesic between p0 and p.
84 ⌅ General Relativity and its Applications

• Repeating the same procedure for all points of the observer worldline, we can define,
in the entire worldtube, the Fermi coordinate system {x⇤µ } adapted to the observer
worldline.
⇤
It can be shown that (with an appropriate choice of the three-dimensional basis {~e(i) } for
each surface ⌃(⌧ )) the spacetime metric in Fermi coordinates is

ds2 = (1 + 2aµ x⇤µ )(dx⇤0 )2 + ij dx

⇤i
dx⇤j + O(x⇤2 ) , (3.129)

where a⌫ are the components of the observer’s four-acceleration. If the observer is freely-
falling (~a = 0), ds2 = ⌘µ⌫ + O(x⇤2 ), and this coordinate system coincides with the LIF
centered in p0 . Indeed, in a LIF the metric coincides with Minkowski’s metric modulo
quadratic terms in the coordinates. If, instead, the observer is accelerated, the deviation
of the metric 3.129 from Minkowski’s metric is linear in the coordinates x⇤µ : g00 ⌘00 =
2aµ x⇤µ . The linear term is precisely due to the acceleration of the observer.
Remarkably, the metric 3.129 does not depend on the spacetime curvature (up to
O(x⇤2 )). Indeed, introducing the Fermi coordinates for an accelerated observer in Special
Relativity, i.e. ⇠ ↵ ! x⇤µ (⇠ ↵ ), leads to the metric ds2 = (1 + 2aµ x⇤µ )(dx⇤0 )2 + ij dx⇤i dx⇤j ,
which di↵ers from Eq. 3.129 by the O(x⇤2 ) terms only. We conclude that due to the Equiv-
alence Principle, according to which the laws of physics in a LIF take, locally, the form
prescribed by Special Relativity for inertial frames, the laws of physics in a Fermi frame
take, locally, the form prescribed by Special Relativity for accelerated frames.

3.10 NON-COORDINATE BASES

0
In Boxes 2-G and 2-P we showed that changing from Minkowskian coordinates {⇠ ↵ } ⌘
(ct, x, y) to polar coordinates {x↵ } ⌘ (ct, r, ✓), the coordinate basis for vectors 3
8
>
<~e(00 ) ! (1, 0, 0)
{~e(↵0 ) } = ~e(10 ) ! (0, 1, 0) (3.130)
>
:
~e(20 ) ! (0, 0, 1)

transforms to {~e(↵) }, where (see Eq. 2.73)

8
>
<~e(0) = ~e(00 )
~e(1) = ~e(r) = cos ✓~e(10 ) + sin ✓~e(20 ) (3.131)
>
:
~e(2) = ~e(✓) = r sin ✓~e(10 ) + r cos ✓~e(20 ) .

These expressions follow from the transformation laws

0
~e(↵) = ⇤µ ↵ ~e(µ0 ) , (3.132)
0 µ0 ↵
where ⇤µ ↵ = @x ↵ 0 @x
@x↵ and ⇤ µ = @xµ0 . We now want to choose a di↵erent basis for vectors.
For example, while the vectors {~e(↵) } are not normalized because
0 1
1 0 0
~e(↵) · ~e( ) = g↵ = @ 0 1 0 A 6= ⌘↵ , (3.133)
0 0 r2
3 Note that, at variance with Boxes 2-G and 2-P, we indicate with a prime the Minkowskian coordinates.