Parallel transport and geodesics

February 24, 2013

1 Parallel transport
Before dening a general notion of curvature for an arbitrary space, we need to know how to compare vectors
at dierent positions on a manifold. Parallel transport provides a way to compare a vector in one tangent
plane to a vector in another, by moving the vector along a curve without changing it. Suppose we have a
curve with unit tangent t in at space and Cartesian coordinates. Then to move a vector v along this curve
without changing it simply means holding the components constant. At any point along the curve, we
may nd the transported components by solving
(t ) v = 0
or, in components,
t
i

i
v
j
= 0
Since the connection vanishes in Cartesian coordinates, this is the same as writing
t
i
D
i
v
j
= 0
but this expression now holds in any coordinates. The same argument holds in a curved space because close
enough to any point we may nd Cartesian coordinates, transport innitesimally, then change coordinates
to Cartesian again. At each point, the Cartesian expression may be written covariantly, but the covariant
expression is the same at every point of the curve regardless of coordinates. We therefore dene parallel
transport of a vector v

along a curve with tangent u

() =
dx

d
to be the solution v

() to the equation
u

= 0
A curve is called autoparallel if it is transported along its own direction,
v

= 0
Notice that parallel transport preserves the length of the vector because
u

_
|v|
2
_
= u

(g

)
= u

(D

+ g

+ g

)
= g

(u

) v

+ g

(u

)
= 0
2 Example: parallel transport on the 2-sphere
Consider the parallel transport of a vector around a =
0
curve on the 2-sphere. The curve itself may
parameterized using as x
i
= (
0
, ), with tangent, t
i
= (0, 1). The length of this tangent vector is given
1
by
l
2
= g
ij
t
i
t
j
= R
2
sin
2

0
so the unit tangent is
u
i
=
1
Rsin
0
(0, 1)
At = 0, let v
i
=
_
v

0
, v

0
_
, and solve the parallel transport equation for v
i
(),
0 = u
i
D
i
v
j
=
1
Rsin
0
D

v
j
0 =

v
j
+ v
k

j
k
2.1 The metric
Since the metric is given by the line element, ds
2
= R
2
d
2
+ sin
2
d
2
, we have, in matrix components,
g
ij
=
_
R
2
0
0 R
2
sin
2

_
with inverse
g
ij
=
_
1
R
2
0
0
1
R
2
sin
2

_
There are intrinsic ways to get this metric. One approach is to specify the symmetries we require three
independent rotations. There are techniques for nding the most general metric with given symmetry, so we
can derive this form directly. Alternatively, we could ask for 2-dim spaces of constant curvature. Computing
the metric for a general 2-geometry, then imposing constant curvature gives a set of dierential equations
that will lead to this form.
2.2 The connection
Since the only non-constant component of the metric tensor is g

, there are only three nonvanishing

connection components,

=
1
2
g

(g
,
+ g
,
g
,
)
=
1
2
1
R
2
_

_
R
2
sin
2

_
,
_
= sin cos

=
1
2
g

(g
,
+ g
,
g
,
)
=
1
2
1
R
2
sin
2

_
R
2
sin
2

_
,
=
cos
sin
2.3 Parallel transport
The parallel transport equation becomes
0 =

v
j
+ v
k

j
k
=

v
j
+ v

+ v

2
There are two components to check. For j = we have
0 =

+ v

=
v

sin
0
cos
0
For j = ,
0 =

+ v

=
v

+ v

cos
0
sin
0
Therefore, we need to solve the coupled equations,
0 =
v

sin
0
cos
0
0 =
v

+ v

cos
0
sin
0
Taking a second derivative of the rst equation and substituting the second,
0 =

2
v

2

v

sin
0
cos
0
=

2
v

2
+ v

cos
0
sin
0
sin
0
cos
0
=

2
v

2
+ v

cos
2

0
Similarly, dierentiating the second equation and substituting the rst we have
0 =

2
v

2
+
v

cos
0
sin
0
=

2
v

2
+ v

sin
0
cos
0
cos
0
sin
0
=

2
v

2
+ v

cos
2

0
Each of these is just the equation for sinusoidal oscillation, so we may immediately write the solution,
v

() = Acos + B sin
v

() = C cos + Dsin
with the frequency given by
= cos
0
Starting the curve at = 0, it will close at = 2. Then for v

we have the initial condition v

(0) =
_
v

0
, v

0
_
,
and from the original dierential equations we must have
v

=0
= v

0
sin
0
cos
0
v

=0
= v

0
cos
0
sin
0
3
These conditions determine the constants A, B, C, D to be
v

() = v

0
cos +
v

0
sin
0
cos
0

sin
= v

0
cos + v

0
sin
0
sin
v

() = v

0
cos
v

0
sin
0
sin
This gives the form of the transported vector at any point around the circle,
v

() = v

0
cos (cos
0
) + v

0
sin
0
sin(cos
0
)
v

() = v

0
cos (cos
0
)
v

0
sin
0
sin(cos
0
)
Look at the inner product of v with the tangent vector at the same point, u =
1
Rsin 0
(0, 1),
u v = g

_
v

0
cos (cos
0
)
v

0
sin
0
sin(cos
0
)
_
= Rsin
0
_
v

0
cos (cos
0
)
v

0
sin
0
sin(cos
0
)
_
= v

0
Rsin
0
cos (cos
0
) v

0
Rsin(cos
0
)
If the circle is at the equator,
0
=

2
, then
u v = v

0
R
is constant. On the other hand, near the pole,
0
1,
u v = v

0
R
0
cos v

0
Rsin
and the transported vector rotates almost completely around the tangent.
3 Geodesics
Consider a curve, x

() in an arbitrary (possibly curved) spacetime, with the proper interval given by

d
2
= g

dx

dx

Then the 4-velocity along the curve is given by

u

=
dx

d
and in an arbitrary parameterization, the tangent is t

=
dx

d
. Then proper time (or length) along the curve
is given by integrating
=

0
_
g

dx

dx

0
_
g

dx

d
dx

d
d
4
A curve of extremal proper length is called a geodesic. We may nd an equation for geodesics by nding the
equation for the extrema of ,
0 =
=

0
_
g

dx

d
dx

d
d
=

0
1
2
_
g

dx

d
dx

_
g

dx

d
dx

d
_
d
=

0
1
2
_
g

dx

d
dx

d
_
g

dx

d
dx

d
+ g

dx

d
dx

d
+ g

dx

d
dx

d
_
d
=

0
1
2
_
g

dx

d
dx

d
_
g
,
dx

d
dx

d
_
x

d
+

0
d
d
_
_
1
2
_
g

dx

d
dx

d
g

dx

d
_
_
x

d
+

0
d
d
_
_
1
2
_
g

dx

d
dx

d
g

dx

d
_
_
x

d
Now choose the parameter to be proper time (length) so that
g

dx

d
dx

d
= g

dx

d
dx

d
= c
2
= 1
Then we have
0 =
1
2

0
__
g
,
dx

d
dx

d
_
x

+
d
d
_
g

dx

d
_
x

+
d
d
_
g

dx

d
_
x

_
d
=
1
2

0
_
_
g
,
u

_
x

+
d
d
_
g

_
x

+
d
d
(g

) x

_
d
=
1
2

0
_
_
g
,
u

_
x

+
_
g
,
dx

d
u

+ g

du

d
_
x

+
_
g
,
dx

d
u

+ g

du

d
_
x

_
d
=
1
2

0
_
_
g
,
u

_
+
_
g
,
u

+ g

du

d
_
+ g
,
u

+ g

du

d
_
x

d
=
1
2

0
_
g
,
u

+ g
,
u

g
,
u

+ g

du

d
+ g

du

d
_
x

d
The equation for the geodesic is therefore
0 =
1
2
_
g
,
u

+ g
,
u

g
,
u

+ g

du

d
+ g

du

d
_
5
=
1
2
(g
,
+ g
,
g
,
) u

+ g

du

d
0 = g

du

d
+
1
2
g

(g
,
+ g
,
g
,
) u

0 =
du

d
+

But this is just

du

d
+

= u

_
du

dx

+ u

_
= u

and we have the equation for an autoparallel,

u

=
du

d
+

= 0
4 Example: Geodesics on the 2-sphere
Once again consider the 2-sphere, but not look for autoparallels. We have the connection components,

= sin cos

=
cos
sin
and the equations to solve are now,
du

d
+

= 0
du

d
+

= 0
Expanding the rst, there is only one nonvanishing connection term,
du

d
+

= 0
du

d
(u

)
2
sin cos = 0
For the second,
du

d
+

= 0
du

d
+

= 0
du

d
+ 2u

cos
sin
= 0
Let the initial conditions be

0
=

2

0
= 0
u

0
= 1
u

0
= 0
6
Since every point and direction on the sphere are equivalent, there is no loss of generality in this choice.
Then we initially have
_
du

d
_
0
= (u

0
)
2
sin
0
cos
0
= 0
and u

does not change. For the equation, it follows that

du

d
= 2u

cos
sin
= 0
so u

is also constant and the velocity vector is

u
i
= (0, 1)
Integrating to nd the curve,
d
d
= 0
d
d
= 1
so =
0
=

2
and =
0
+ = . The curve is therefore the equator,
(, ) = (0, )
We may characterize the equator as the intersection of the unique plane normal to the surface, containing
the initial velocity vector. Such a plane always passes through the center of the sphere, so all geodesics are
given by great circles.
7