3

3.1

Parallel transport and geodesics

Dierentiation along a curve

As a prelude to parallel transport we consider another form of dierentiation: dierentiation along a curve. A curve is a parametrized path through spacetime: x(), where
is a parameter that varies smoothly and monotonically along the path. The curve
has a tangent vector V
d
x/d = (dx /d)
e . Here one must be careful about the
interpretation: x are not the components of a vector; they are simply 4 scalar elds.
However, V
= d
x/d is a vector (i.e. a tangent vector in the manifold).
If we wish, we could make V
a unit vector (provided V
is non-null) by setting d =
|d
x d
x |1/2 to measure path length along the curve. However, we will impose no such
restriction in general.

Now, suppose that we have a scalar eld fX dened along the curve. We dene the
derivative along the curve by a simple extension of equations (36) and (38) of the rst
set of lecture notes:
df
V
 = V f ,
V f f,
d

d
x
V
=
.
d

(21)

We have introduced the symbol V for the directional derivative, i.e. the covariant
derivative along V
, the tangent vector to the curve x(). This is a natural generalization
of , the covariant derivative along the basis vector
e .
For the derivative of a scalar eld, V involves just the partial derivatives . Sup
X along the curve. Now the compopose, however, that we dierentiate a vector eld A

nents of the gradient A are not simply the partial derivatives but also involve the
connection. The same is true when we project the gradient onto the tangent vector V

along a curve:


DA
dA


A,

V
 = V ( A )
e =

e V A
d
d

dA
+ A V
e .
d

(22)

We retain the symbol V to indicate the covariant derivative along V but we have
introduced the new notation D/d = V = d/d = V .

3.2

Parallel transport

The derivative of a vector along a curve leads us to an important concept called parallel


dened
transport. Suppose that we have a curve x() with tangent V
and a vector A(0)
at one point on the curve (call it = 0). We dene a procedure called parallel transport


by dening a vector A()
along each point of the curve in such a way that DA /d = 0:

=0
V A

along V
.
parallel transport of A

(23)

Over a small distance interval this procedure is equivalent to transporting the vector A
along the curve in such a way that the vector remains parallel to itself with constant
length: A( + ) = A() + O()2 . In a locally at coordinate system, with the
connection vanishing at x(), the components of the vector do not change as the vector


is transported along the curve. If the space were globally at and we used rectilinear
coordinates (with vanishing connection everywhere), the components would not change
at all no matter how the vector is transported. This is not the case in a curved space or
in a at space with curvilinear coordinates because in these cases the connection does
not vanish everywhere.

3.3

Geodesics

Parallel transport can be used to dene a special class of curves called geodesics. A
geodesic curve is one that parallel-transports its own tangent vector V
= d
x/d, i.e.,
a curve that satises V V
= 0. In other words, not only is V
kept parallel to itself
(with constant magnitude) along the curve, but locally the curve continues to point
in the same direction all along the path. A geodesic is the natural extension of the
denition of a straight line to a curved manifold. Using equations (22) and (23), we
get a second-order dierential equation for the coordinates of a geodesic curve:
dV
DV
=
+ V V = 0 for a geodesic ,
d
d

dx
.
d

(24)

Indeed, in locally at coordinates (such that the connection vanishes at a point), this
is the equation of a straight line. However, in a curved space the connection cannot be
made to vanish everywhere. A well-known example of a geodesic in a curved space is a
great circle on a sphere.

There are several technical points worth noting about geodesic curves. The rst is
that V
V
= g(V
, V
) is constant along a geodesic because dV
/d = 0 (eq. 24) and
V g = 0 (metric compatibility with gradient). Therefore, a geodesic may be classied
by its tangent vector as being either timelike (V
V
< 0), spacelike (V
V
> 0) or null
(V
V
= 0). The second point is that a nonlinear transformation of the parameter will
invalidate equation (24). In other words, if x () solves equation (24), y () x (())
will not solve it unless = a + b for some constants a and b. Only a special class of
parameters, called ane parameters, can parametrize geodesic curves.

The ane parameter has a special interpretation for a non-null geodesic. We deduce
this relation from the constancy along the geodesic of V
V
= (d
xd
x)/(d2 ) a, implying
ds = ad and therefore s = a + b where s is the path length (ds2 = g dx dx ). For
a non-null geodesic (V
V
= 0), all ane parameters are linear functions of path length
(or proper time, if the geodesic is timelike). The linear scaling of path length amounts
simply to the freedom to change units of length and to choose any point as = 0.
Note that originally we imposed no constraints on the parameterization. However, the
solutions of the geodesic equation automatically have being an ane parameter. There
is no fundamental reason to use an ane parameter; one could always take a solution
of the geodesic equation and reparameterize it or eliminate the parameter altogether by
replacing it with one of the coordinates along the geodesic. For example, for a timelike
trajectory, xi (t) is a perfectly valid description and is equivalent to x (). But the spatial
components as functions of t = x0 clearly do not satisfy the geodesic equation for x ().
Another interesting point is that the total path length is stationary for a geodesic:
1/2
B
dx dx

g
ds =
d = 0

d d
A
A
B

(25)

if is an ane parameter. The refers to a variation of the integral arising from

a variation of the curve, x () x () + x (), with xed endpoints. The metric
components are considered here to be functions of the coordinates. The variational
principle is discussed in section 2 of the 8.962 notes Hamiltonian Dynamics of Particle
Motion, where it is shown that stationary path length implies the geodesic equation (24)
if the parameterization is ane. Equation (25) is invariant under reparameterization, so
its stationary solutions are a broader class of functions than the solutions of equation
(24). In general, the tangent vector of the stationary solutions are not normalized:
 constant, implying that is not ane. It is easy to show that any
|V
V
|1/2 = Q() =
stationary solution may be reparameterized, through d /d = Q(), and that
the resulting curve x (( )) obeys the geodesic equation with ane parameter . This
transformation replaces the unnormalized tangent vector V
by V
/Q(). For an ane
parameterization, the tangent vector must always have constant length.
Equation (25) is a curved space generalization of the statement that a straight line
is the shortest path between two points in at space.

3.4

Integrals of motion and Killing vectors

Equation (24) is a set of four second-order nonlinear ordinary dierential equations for
the coordinates of a geodesic curve. One may ask whether the order of this system
can be reduced by nding integrals of the motion. An integral, also called a conserved
quantity, is a function of x and V = dx /d that is constant along any geodesic. At
least one integral always exists: V
V
= g V V . (For an ane parameterization, V
V

is constant along the curve.) Are there others? Sometimes. One may show that equation
(24) may be rewritten as an equation of motion for V g V , yielding
dV
1
= ( g )V V .
d
2

(26)

Consequently, if all of the metric components are independent of some particular coordinate x , the corresponding component of the tangent one-form is constant along the
geodesic. This result is very useful in reducing the amount of integration needed to
construct geodesics for metrics with high symmetry. However, the condition g = 0
is coordinate-dependent. There is an equivalent coordinate-free test for integrals, based

call Killing vectors. Killing vectors are, by
on the existence of special vector elds K
denition, solutions of the dierential equation
K + K = 0 .

(27)

(The Killing vector components are, of course, K = g K .) The Killing equation (27)
usually has no solutions, but for highly symmetric spacetime manifolds there may be
9

one or more solutions. It is a nice exercise to show that each Killing vector leads to the
integral of motion

 = K V = constant along a geodesic .
V , K

(28)

Note that if one of the basis vectors (for some basis) satises the Killing equation, then
the corresponding component of the tangent one-form is an integral of motion. The test
for integrals implied by equation (26) is a special case of the Killing vector test when the
Killing vector is simply a coordinate basis vector.
The discussion here has focused on geodesics as curves. The notes Hamiltonian
Dynamics of Particle Motion interprets them as worldlines for particles because, as
we will see, a fundamental postulate of general relativity is that, in the absence of nongravitational forces, particles move along geodesics. Given this fact, we are free to choose
units of the ane parameter so that dx /d is the 4-momentum P , normalized by
P
P
= m2 for a particle of mass m (instead of dx /d = V , V
V
= 1). Thus,
the tangent vector, denoted V
above, is equivalent to the particle 4-momentum vector.
The ane parameter then measures proper time divided by particle mass. Although
one might fear this makes no sense for a massless particle, in fact it is the only way to
anely parameterize null geodesics because the proper time change d vanishes along a
null geodesic so dx /d is undened. For a massless particle, one takes the limit m 0
starting from the solution for a massive particle, with the result that d = d /m is nite
as m 0.

10