3 Tensor Algebra in a Curved Spacetime

Here we first briefly review technique of tensors and differential geometry. Needless to say that there exist a lot more
than presented here about the subjects. Then we discuss the metric tensor and the Riemann curvature tensor in a curved
spacetime.

3.1 Tensor Analysis & Curved Geometry

3.1.1 Basics in Tensor Calculus
• Manifold.— is a geometry which locally looks like a bit of n-dimensional Euclidean space Rn . For example, a 2d
sphere is a manifold and is locally like R2 . However, they are globally different, i.e., you need many (at least two) patches
of R2 to cover the sphere. This set of coordinate patches is called an atlas or a chart. So, a manifold is in a simple term a
set of open neighborhoods that look like Rn .

• Congruence of curves.— is the family of curves xµ (u) that only one curve goes through each point in the manifold.
For a given vector field X µ , the congruence of curves whose tangent vector is X µ is called orbits or trajectories of X µ :
dxµ
= Xµ . (3.1)
du
• Contravariant and covariant tensors.— Given a coordinate transformation to cover the manifold,

@ x̃µ @xµ 1
x̃µ = x̃µ (x⌫ ) , J := , xµ = xµ (x̃⌫ ) , = , (3.2)
@x⌫ @ x̃ ⌫ J

contravariant tensors transform like a differential dxµ as

@ x̃µ ⌫ @ x̃µ @ x̃⌫ ⇢
X̃ µ (x̃) = X (x) , X̃ µ⌫ (x̃) = X (x) , ··· , (3.3)
@x⌫ @x⇢ @x
while covariant tensors transform like a gradient @µ as

@x⌫ @x⇢ @x
X̃µ (x̃) = X⌫ (x) , X̃µ⌫ (x̃) = X⇢ (x) , ··· . (3.4)
@ x̃µ @ x̃µ @ x̃⌫
Of course a tensor of rank zero, or scalar, is invariant:
˜(x̃) = (x) . (3.5)

General tensors are of rank (p, q), and one can create one Z = XY of (p1 + p2 , q1 + q2 ) by a direct product of X
and Y with each type. Contraction in contrast reduces the rank to (p 1, q 1) by summing over one upper and one
lower indicies together.

• Tangent space and co-tangent space.— Tensors are geometric objects, i.e., independent of coordinate systems. For this
interpretation, we can express a contravariant vector as an operator X := X µ @µ , acting on a function f (for example,
coordinates xµ to obtain X µ ). In this expression, it is made obvious that the vector is independent of coordinate systems.
At a given point, a set of all vectors make up a tangent vector space Tp , and its dual Tp⇤ is called co-tangent vector space.
For example. the contravariant vector X µ lies in the tangenet space (which itself is different from the manifold, with the
only exceptions of Euclidean and Minkowski manifolds), while the coordinate basis @µ lies in the co-tangent space.
For example, a vector can be expressed in terms of basis vector [ea ] (or tetrad)

X = X a [ea ] , [ea ]µ / @µ , (3.6)

and this basis is orthonormal

⌘ab = gµ⌫ [ea ]µ [eb ]⌫ . (3.7)
AST511 General Relativity JAIYUL YOO

A vector product is then

X · Y = X a Y b ([ea ] · [eb ]) = ⌘ab X a Y b , (3.8)
where X a is contra-variant and [ea ] is co-variant. A co-vector (or one-form) turns a vector into a real number:

pa := p([e]a ) , p(V ) = p(V a [ea ]) = V a p([ea ]) = V a pa . (3.9)

One can generalize this concept to general tensor of rank (p, q).

• Lie derivative.— In a curved spacetime, no two quantities at two different points can be meaningfully compared. This
implies that a differentiation in a curved manifold needs to be re-considered, and a simple coordinate differentiation of
any tensor does not transform tensorially. The Lie derivative £X defines a differentiation of any tensor along a tangent
vector X µ . Now consider a coordinate transformation

x̃µ = xµ + " X µ , |"| ⌧ 1 , (3.10)

and interpret it in an active way, i.e., we move the point xµp to another point xµq := x̃µ along X µ by an infinitesimal
amount ", rather than the same physical point in a passive interpretation. A tensor at xq can be approximated as

T µ⌫ (xq ) = T µ⌫ (xp ) + " X ⇢ @⇢ T µ⌫ (xp ) + · · · , (3.11)

in the limit " ! 0. With

@ x̃µ
= ⌫µ + " @⌫ X µ , (3.12)
@x⌫
we shift the tensor at the point p to the point q (active transformation):

µ⌫ @ x̃µ @ x̃⌫ ⇢
T̃shift (x̃) = T (x) = T (x) + " @⇢ X µ T ⇢⌫ (x) + @ X ⌫ T µ (x) + O("2 ) ,
µ⌫
(3.13)
@x⇢ @x
Now we define the Lie derivative as
1 ⇥ µ⌫ µ⌫ ⇤
£X T µ⌫ := lim T (xq ) Tshift = X ⇢ @⇢ T µ⌫ T µ⇢ @⇢ X ⌫ T ⇢⌫ @⇢ X µ , (3.14)
"!0 "

and a few more examples are

£X Aµ = Aµ ,⌫ X ⌫ X µ ,⌫ A⌫ , £X Tµ⌫ = Tµ⌫,⇢ X ⇢ + T⇢⌫ X ⇢ ,µ + Tµ⇢ X ⇢ ,⌫ . (3.15)

The Lie derivative is independent of coordinate systems and preserves the tensor rank. It reduces to ordinary differentia-
tion with respect to x⌫ , in case X µ = ⌫µ . Furthermore, it is linear in the tensor field and satisfies the Leibniz chain rule.
All the partial derivatives in the Lie derivative can be replaced in terms of covariant derivatives defined below.

• Covariant derivatives.— Similarly, a vector field V µ at a point q can be approximated

V µ (xq ) = V µ (xp ) + x⌫ @⌫ V µ (xp ) + · · · , xµq := xµp + xµ , (3.16)

if two points are close to each other. However, this vector cannot be directly compared to the vector at a point p, so we
shift the vector at p to the point q. With infinitesimal displacement, the shifted vector at q can be written as (or again
actively transform)
µ
Vshift (xq ) = V µ (xp ) µ ⇢
⇢⌫ (xp )V (xp ) x⌫ + · · · , xµ ,⇢ V ⇢ =: µ
⇢⌫ V
⇢
x⌫ , (3.17)

where the affine connection is defined in terms of a proportionality constant. Now we define the covariant derivative as
1 ⇥ µ µ ⇤
V µ ;⌫ := r⌫ V µ = lim V (xq ) Vshift = @⌫ V µ + µ
⇢⌫ V
⇢
. (3.18)
x!0 x⌫
Given the definition, we can readily derive
⇢ µ µ µ µ
rµ = @ µ , Xµ;⌫ = Xµ,⌫ µ⌫ X⇢ , T⌫;⇢ = T⌫,⇢ + ⇢ T⌫ ⌫⇢ T . (3.19)

20
AST511 General Relativity JAIYUL YOO

Contraction over two indicies commutes with the covariant derivative:

⇣ ⌘ ⇣ ⌘ ai =c
···ai =c···an a1 ···an
rµ Tba11···b i c···bm
= r µ T b1 ···cm . (3.20)
bi =c

• Geodesic and Affine connection.— Geodesic is in a simple term a straightest possible curve in a curved manifold. For
example, in a sphere, geodesic provides a shorted path (a great circle on a sphere) between two points, and a parallel
direction at a one point is no longer parallel along the geodesic. Note that any parallel direction remains parallel in
Euclidean geometries, when shifted to any points.
If we demand that the covariant derivative of a contravariant tensor transforms as (1, 1) tensor, we derive that the
affine connection transforms as
⌧ @ x̃ @ 2 x⇢
˜ µ⌫ = @ x̃ @x @x ⇢
⌧ + . (3.21)
⇢ µ
@x @ x̃ @ x̃ ⌫ @x⇢ @ x̃µ @ x̃⌫
Any quantity which transforms as above is called an affine connection, and a manifold with affine connection is an affine
manifold. A torsion tensor can be defined as
T⇢µ := µ⇢ µ
⇢ , (3.22)
and indeed it transforms tensorially, as the non-tensorial part in µ⌫ is removed. If the torsion tensor vanishes identically,
the affine connection is symmetric over the lower indicies.

• Parallel transport and geodesic equation.— Given a curve xµ (u) and its tangent vector X µ , we define a total derivative
and a parallel transport along the curve:
D D
T := rX T = X µ rµ T , T ⌘0. (3.23)
du du
Furthermore, if a tangent vector X µ is parallel transported along its curve xµ (u) to be parallel to the tangent vector, this
curve is called an affine geodesic:

d2 xµ dx⇢ dx
rX X µ = X µ , + µ
⇢ = (u)X µ , (3.24)
du2 du du
where is a constant. Alternatively, this can be derived by shifting the tangent vector at p to q and using xµ = X µ u.
When ⌘ 0, or the parallel transported tangent vector is exactly the tangent vector, this parameter is called an affine
parameter s, and the geodesic is invariant under re-parametrization s ! c1 s + c2 with arbitrary constants c1 and c2 .

• Tensor density of weight W.— transforms tensorially, but with extra power of the Jacobian J:

µ··· @ x̃µ @x⇢ ···

T̃⌫··· = JW · · · T . (3.25)
@x @ x̃⌫ ⌫···
p
For instance, we will always deal with the metric tensor gµ⌫ , and its determinant is denoted as g. Then, g is a scalar
density of weight 1 (other literature might use different sign convention, or weight +1):

@x⇢ @x p p
g̃µ⌫ (x̃) = µ ⌫
g⇢ (x) , g̃ = J 1 g. (3.26)
@ x̃ @ x̃
p µ
The volume factor d4 x is of weight +1, so that g d4 x is an ordinary scalar. In the same way, for any vector density V(W )
(or tensors) of weight W can be made an ordinary vector:
p W µ
V µ := g V(W ) , r⇢ V µ = @ ⇢ V µ + µ
⇢⌫ V
⌫
. (3.27)

Under the assumption that the covariant derivative of the metric tensor vanishes (which is true, if the connection is the
µ
Christoffel symbol), we can show that the covariant derivative of a vector density V(W ) of W is

µ µ µ ⌫ ⌫ µ
r⇢ V(W ) = @⇢ V(W ) + ⇢⌫ V(W ) +W ⌫⇢ V(W ) , (3.28)

21
AST511 General Relativity JAIYUL YOO

with one extra term, which is also true for any rank tensor density T⇢µ⌫···
··· (W ) . As a special case, we obtain

rµ V(µ 1) = @µ V(µ 1) , (3.29)

and hence we derive
1 @ p
rµ V µ = p ( g V µ) . (3.30)
g @xµ
This can be directly verified by an explicit calculation of the covariant derivative with
⇢ 1 @ p
⇢µ =p g, (3.31)
g @xµ
from Eq. (2.11).

3.1.2 Levi-Civita Tensor and Matrix Determinant

The Levi-Civita tensor is defined always in any coordinates as
"µ⌫⇢ = +1 for even of txyz , "µ⌫⇢ = 1 for odd , (3.32)
where the coordinates run in sequence.1 Given this definition of the Levi-Civita tensor and any 4-by-4 matrix M = Mµ⌫ ,
the determinant of M can be derived as
1
det M := "µ⌫⇢ M1µ M2⌫ M3⇢ M4 = "µ⌫⇢ "↵ M↵µ M ⌫ M ⇢ M . (3.33)
4!
To demonstrate that the Levi-Civita symbol is in fact a tensor density of weight +1, we consider
@ x̃↵ @ x̃ @ x̃ @ x̃ µ⌫⇢
" = J"↵ . (3.34)
@xµ @x⌫ @x⇢ @x
Given the nature of index structure in LHS, the right-hand side should be proportional to "↵ , and the proportionality
constant can be computed by considering the reference sequence (txyz for ↵ ). Furthermore, we infer another useful
relation
det M · "µ⌫⇢ = "↵ M µ ↵ M ⌫ M ⇢ M , (3.35)
for any matrix representation M = M µ ⌫ . The covariant Levi-Civita, defined by lowering the indicies with gµ⌫ as
"↵ := g↵µ g ⌫ g ⇢ g "µ⌫⇢ = g "↵ , (3.36)
is then a tensor density of weight 1, as g is a density of weight 2 and the weights add up for a product of two tensor
densities. The second equality can be derived by the same inference and using the reference sequence.
Now consider a matrix representation M = Mµ⌫ and its inverse M 1 = M µ⌫ , there exists a useful identity

µ⌫ @ @
M ↵
M⌫µ ⌘ Tr M 1 @↵ M = ln | det M| , (3.37)
@x @x↵
and we can readily show that
1 @ 1 @ p
@⇢ g = g g µ⌫ @⇢ gµ⌫ = 2g µ
µ⇢ , ) µ
µ⇢ = g=p g, (3.38)
2g @x⇢ g @x⇢
⇢
where we used the relation between gµ⌫ and µ⌫ in Eq. (2.11). Since g is a scalar of weight 2, this relation leads to
rµ g = 0 , r⇢ gµ⌫ = 0 . (3.39)
When the metric is Minkowski, g = 1, and the Levi-Civita symbol satisfies
µ ⌫

"µ⌫⇢ = "µ⌫⇢ , "µ⌫⇢ "µ⌫⇢ = 4! , "µ⌫⇢ "⌫⇢ = 3! µ
 , "µ⌫⇢ "✏⇢ = 2 µ⌫
✏ := 2 µ

⌫ .
✏ ✏
(3.40)
where we assumed four spacetime dimension.
1
For example, coordinates can be t, x, y, z or t, r, ✓, . However, mind that "txyz = "xyzt , so the coordinate sequence should be checked.
The normalization convention can also be different, for example, "0123 ⌘ 1, or "1230 ⌘ 1 = "0123 . In some cases, the convention "0123 = 1 is
adopted.

22
AST511 General Relativity JAIYUL YOO

3.1.3 Differential Forms

A differential p-form is a tensor of rank (0, p) with completely anti-symmetric indicies. A scalar is (0, 0), and hence a
0-form. A co-vector (dual vector) is (0, 1), and a 1-form. The E&M field strength tensor Fµ⌫ is (0, 2) and anti-symmetric
over its indicies, and hence a 2-form. There is no p-form in d-dimension, if p > d. In n-dimension, there exist only
✓ ◆
n n!
= , (3.41)
p p!(n p)!

number of independent p-forms. A wedge product of p-form A and q-form B is a simple product, but their indicies should
be anti-symmetrized:
(p + q)!
A^B = A[µ1 ···µp Bµ1 ···µq ] = ( 1)pq B ^ A . (3.42)
p!q!
For instance, consider , Aµ , Bµ , and we obtain

( ^ A)µ = Aµ , (A ^ B)µ⌫ = 2A[µ B⌫] = Aµ B⌫ A⌫ B µ . (3.43)

With a coordinate independent notation, a differential p-form ! can be expressed as

! = !µ1 ···µp (dxµ1 ^ · · · ^ dxµp ) . (3.44)

An exterior derivative d increases the rank of a p-form A by one:

(dA)µ1 ···µp+1 = (p + 1)@[µ1 Aµ2 ···µp+1 ] , (3.45)

and the simplest examples are the gradient and the curl

(d )µ = @µ , (dA)µ⌫ = 2@[µ A⌫] = @µ A⌫ @ ⌫ Aµ . (3.46)

Due to the anti-symmetric nature, a repeat application of the exterior derivative is always zero:

0 = d(dA) . (3.47)

A p-form F is called closed if dF = 0, and is called exact if there exists a (p 1)-form A with F = d(A). All exact
forms are always closed, but the converse is true, provided that the manifold is simply connected. For example, the E&M
field strength tensor Fµ⌫ is a 2-form, which can be expressed in terms of a 1-form Aµ :

Fµ⌫ = @µ A⌫ @µ A⌫ = (dA)µ⌫ , dF = 0 . (3.48)

As noted in E&M, however, Aµ is not uniquely determined, as

F = d(A) = d(Ã) , Ã = A + du , (3.49)

where we introduced a (p 2)-form u. In E&M, du ! @µ , corresponding to the gauge transformation.

• Stokes’ theorem.— While it makes little sense to compare or manipulate quantities at two different points, it is fine if
the quantities are scalar. For example, a summation of a scalar field at two points remains invariant under a coordinate
transformation:
˜(x̃1 ) + ˜(x̃2 ) = (x1 ) + (x1 ) . (3.50)
Drawing on this concept, we can construct an integral over some region in the manifold. As discussed before, however,
the coordinate volume factor d4 x is of weight +1, such that we need to integrate a scalar of weight 1 to compensate for
it. Consider the volume element of m-dimensional subspace in n-dimensional manifold

@x⌫ @x⌫
dV µ1 ···µm := µ1 ···µm
⌫1 ···⌫m ··· du1 · · · dum , (3.51)
@u1 @um

23
AST511 General Relativity JAIYUL YOO

where the curves are parametrized as xµ (u). This volume element is a tensor of rank m (and density 0) and invariant
under re-parametrization of the curves with u. A covariant tensor Xµ1 ···µm of rank m can be used to construct a scalar
and form a volume integral over the m-dimensional subspace ⌦:
Z
Xµ1 ···µm dV µ1 ···µm . (3.52)
⌦

For example, consider a 2D curve = (x, y) parametrized in terms of uµ = (r, ✓), and the volume element is then
xµ
dV xy = dV yx = r drd✓. In 4D, the ordinary volume element and the surface element are related as
1 1
dV = "µ⌫⇢ dV µ⌫⇢ = d4 x , dSµ = "µ⌫⇢ dV ⌫⇢ = (dxdydz, dtdydz, dtdxdz, dtdxdy) . (3.53)
4! 3!
The Stoke’s theorem states that in a m-dimensional space the integral of a covariant tensor of rank (m 1) over the
boundary @⌦ is equal to the integral of a derivative over the volume ⌦:
Z Z
µ1 ···µm 1
Xµ1 ···µm 1 dV = @µm Xµ1 ···µm 1 dV µ1 ···µm . (3.54)
@⌦ ⌦

For example, in 4D the Stoke’s theorem says

Z Z
X µ dSµ = rµ X µ d⌦ , (3.55)
@⌦ ⌦

corresponding to the divergence theorem. d⌦ is a scalar density of weight +1. Put it differently,
Z Z
µp p
X gdSµ = rµ X µ g d4 x . (3.56)
@⌦ ⌦

3.2 Metric and Riemann Curvature Tensors

With the equivalence principle, one can always find a coordinate, in which the metric is locally Minkowski gµ⌫ = ⌘µ⌫
and hence the Christoffel symbol vanishes ⇢µ⌫ = 0. Hence, anything that distinguishes from the Minkowski spacetime
should come from at least the second derivatives of gµ⌫ . This also implies that the metric alone is not enough to prove that
the spacetime is curved, i.e., the metric in a spherical coordinate is not Minkowski, but the spacetime is in fact not curved.

3.2.1 Metric Tensor

Any symmetric covariant tensor gµ⌫ of rank two can be used to define distances of any vectors. In particular, the line
element or infinitesimal distance between two neighboring points is

ds2 = gµ⌫ (x)dxµ dx⌫ , (3.57)

where ds2 can be either positive or negative despite the notation. Our convention is mostly plus, such that the interval
is called time-like for ds2 < 0, space-like for ds2 > 0, and null for ds2 = 0. The metric tensor is often called the first
fundamental form. A manifold with a metric is called a Riemannian manifold. Given the metric tensor, its inverse g µ⌫ and
the determinant g are
g µ⇢ g⇢⌫ = ⌫µ , g := det gµ⌫ . (3.58)

• Metric geodesic.— Consider a time-like curve xµ (u) and the distance between two points is
Z q Z Z r
ds dxµ dx⌫
s= ds = du = du gµ⌫ . (3.59)
p du du du
By using the Euler-Lagrange equation with respect to dxµ /du, we find that the path xµ minimizing the distance between
two points should satisfy
✓ ◆ 1
d2 x⌫ dx⌫ dx⇢ d2 s ds dx⌫
gµ⌫ 2
+ {⌫⇢, µ} = 2
gµ⌫ , (3.60)
du du du du du du

24
AST511 General Relativity JAIYUL YOO

or by removing gµ⌫
⇢ ✓ ◆ 1
d2 xµ µ dx⌫ dx⇢ d2 s ds dxµ
+ = , (3.61)
du2 ⌫⇢ du du du2 du du
where we defined the Christoffel symbols of the first kind and the second kind:
1
{⌫⇢, } := (@⇢ g⌫ + @⌫ g⇢ @ g⌫⇢ ) , (3.62)
⇢ 2
µ 1
:= g µ {⌫⇢, } = g µ (@⇢ g⌫ + @⌫ g⇢ @ g⌫⇢ ) . (3.63)
⌫⇢ 2

If we parametrize the curve in terms of s, the geodesic equation becomes

⇢
d2 xµ µ dx⌫ dx⇢ dxµ dx⌫
+ = 0 , 1 = g µ⌫ . (3.64)
ds2 ⌫⇢ ds ds ds ds

It is now apparent that the affine and the metric geodesics coincide, if we equate the affine connection to the Christoffel
symbol of the second kind: ⇢
µ µ 1
⌫⇢ := = g µ (@⇢ g⌫ + @⌫ g⇢ @ g⌫⇢ ) . (3.65)
⌫⇢ 2
This automatically leads to the vanishing torsion tensor and the metricity condition
µ ✏ ✏
T⌫⇢ =0, 0 = r⇢ gµ⌫ = gµ⌫,⇢ ⇢µ g✏⌫ ⇢⌫ gµ✏ . (3.66)

For the latter, one can go to a free-falling frame to see the triviality. Alternatively, if we assume the metricity condition,
we can show that the affine connection is the metric connection.
In short, if we take the Christoffel symbol as our affine connection, the metricity relation holds, and the converse is
also true (see § 3.1 of Wald (1984) for the uniqueness). The metricity relation can also be derived, if we demand that the
inner product of two vectors remains unchanged, when both vectors are parallel transported.

3.2.2 Riemann Curvature Tensor

Unlike coordinate derivatives, covariant differentiation does not commute. A straightforward computation yields

X µ ;⇢ = @ @⇢ X µ + µ
⌫⇢ X
⌫
+ µ
 @⇢ X  + ⌫⇢ X ⌫  µ µ
⇢ (@ X + ⌫ X )
⌫
, (3.67)
µ µ µ ⌫ µ   ⌫  µ µ ⌫
X ; ⇢ = @⇢ (@ X + ⌫ X ) + ⇢ (@ X + ⌫ X ) ⇢ (@ X + ⌫ X ) , (3.68)

and we obtain
r r⇢ X µ r⇢ r X µ = 2X µ ;[⇢ ] = Rµ ⌫ ⇢ X ⌫ T⇢ r X µ , (3.69)
where T⇢ is the torsion tensor and we defined the Riemann tensor in terms of the Christoffel symbols only:
µ µ µ ✏ µ ✏ µ µ µ ✏ µ ✏ µ
R⌫⇢ := ⌫ ,⇢ ⌫⇢, + ⌫ ⇢✏ ⌫⇢ ✏ = ⌫ ;⇢ ⌫⇢; ⌫ ⇢✏ + ⌫⇢ ✏ . (3.70)

Despite the appearance, the Riemann tensor is indeed a tensor, transforming tensorially. We will consider only the con-
nections with vanishing torsion T⇢ ⌘ 0: The Riemann tensor has all the information of the geometry, such that how any
four vector changes locally is fully determined by the Riemann tensor
1
2uµ;[⌫⇢] = u R µ⌫⇢ , u⇢ ;[⌫µ] = R⇢ µ⌫ u . (3.71)
2
Consider a small closed path (or a round trip) and parallel transport a vector along the path. Only when the Riemann
vanishes identically, the transported vector (or any tensors) will come back to the original configuration, regardless of
paths taken (see my note).

• Symmetries in Riemann tensor.— With the metric tensor, we define the Ricci tensor Rµ⌫ and the Ricci scalar R:

Rµ⌫ := R µ⌫ , R := Rµµ = g µ⌫ R⌫µ , Rµ⌫⇢ = gµ R ⌫⇢ . (3.72)

25
AST511 General Relativity JAIYUL YOO

Given the definition, the Riemann tensor has the symmetry in its indices

Rµ⌫⇢ = R[µ⌫][⇢ ] = R⇢ µ⌫ . (3.73)

Furthermore, due to the symmetry in the connection, the Riemann tensor satisfies

Rµ ⌫⇢ + Rµ ⇢ ⌫ + Rµ ⌫⇢ = Rµ [⌫⇢ ] =0. (3.74)

It can be shown that the Riemann tensor satisfies the Bianchi identities: (use the normal coordinate in Section 3.3.1)

Rµ⌫⇢ ; + Rµ⌫⇢; + Rµ⌫ ;⇢ =0. (3.75)

• 2D Sphere as an example.— Given a metric tensor in a 2D sphere of radius r

ds2 = gµ⌫ dxµ dx⌫ = r2 (d✓2 + sin2 ✓ d 2

), (3.76)

we first compute the Christoffel symbols

✓
= sin ✓ cos ✓ , ✓ = cot ✓ , (3.77)

and the Riemann tensor

2
R✓ ✓ = sin2 ✓ , R = sin2 ✓ , R✓✓ = 1 , R= . (3.78)
r2

3.2.3 Weyl Conformal Tensor

The symmetries in the Riemann tensor reduce the degree of freedom from n4 to

n2 (n2 1)
Rµ ⌫⇢ 3 dof . (3.79)
12
For the case of four-dimensional spacetime, the degree of freedom is reduced to 20 from 44 = 256:
• n = 1: the Riemann tensor vanishes identically Rµ ⌫⇢ = 0, i.e., dof is zero,

• n = 2: only one component of the Riemann tensor is independent, i.e., dof is one, essentially the Ricci scalar R,

• n = 3: six dofs are essentially captured by the Ricci tensor Rµ⌫ ,

• n = 4: twenty dofs exist. Ten of them is represented by the Ricci tensor Rµ⌫ and the remaining ten by the Weyl
tensor C µ ⌫⇢ .
The (traceless) Weyl curvature tensor or the conformal tensor is defined as
1 R
Cµ⌫⇢ := Rµ⌫⇢ (gµ⇢ R⌫ + g⌫ Rµ⇢ g⌫⇢ Rµ gµ R⌫⇢ ) + (gµ⇢ g⌫ gµ g⌫⇢ ) , (3.80)
2 6
where the factors 2 and 6 in the denominator are indeed (n 2) and (n 1)(n 2) in n-dimension.
Like the Riemann curvature tensor, the Weyl tensor expresses the tidal force that a body feels when moving along
a geodesic (see Section 3.3.2). The Weyl tensor differs from the Riemann curvature tensor in that it does not convey
information on how the volume of the body changes, but rather only how the shape of the body is distorted by the tidal
force. The Ricci curvature, or trace component of the Riemann tensor contains precisely the information about how
volumes change in the presence of tidal forces, so the Weyl tensor is the traceless component of the Riemann tensor. In
other words, the Ricci tensor is algebraically set by matter distribution through the Einstein equation, but the Weyl tensor
is determined by differential equations with suitable boundary conditions.
The Weyl tensor has the same symmetries as in the Riemnann tensor, in addition to the traceless condition (metric
contraction on any pair of indices yields zero)
C µ ⌫µ = 0 . (3.81)

26
AST511 General Relativity JAIYUL YOO

Two metric tensors are conformally related if

ĝµ⌫ = ⌦2 (x)gµ⌫ , (3.82)
for a non-vanishing differentiable function ⌦ in the manifold. Two metrics describe two different geometries, but the
causal structure (or null geodesic) is identical, and the Weyl tensors are also identical:

Ĉ µ ⌫⇢ = C µ ⌫⇢ . (3.83)

If the Weyl tensor vanishes identically, the metric is conformally flat:

gµ⌫ = ⌦2 ⌘µ⌫ , (3.84)

and the converse is also true. For instance, the background FRW metric is conformally flat.

3.2.4 Killing Vector and Maximally Symmetric Spacetime

• Killing vectors.— A metric gµ⌫ (x) is called form-invariant, if

g̃µ⌫ (x) = gµ⌫ (x) , (3.85)

under a coordinate transformation for all x, and this transformation is called isometry. Note that this condition is different
from the condition for a scalar ˜(x̃) = (x). Given the transformation law for a metric tensor, the condition for a form-
invariant metric puts a very complicated restriction on the transformation. For an infinitesimal transformation, this can be
expressed as
0 = ⇠µ;⌫ + ⇠⌫;µ = L⇠ g , x̃µ = xµ + ⇠ µ (x) . (3.86)
Any four vector that satisfies this equation is called a Killing vector. The vanishing Lie derivative states that the metric
tensor is unaffected along ⇠ µ . Given the structure of the Killing vector, the full functional form of ⇠ µ (x) in the manifold
is fully determined by ⇠ µ and its derivative ⇠ µ ;⌫ at some point X (see Weinberg (1972) § 13 for more details).
For two Killing vectors ⇠1µ and ⇠2µ , we can readily show that

[⇠1 , ⇠2 ]µ := ⇠2⌫ r⌫ ⇠1µ ⇠1⌫ r⌫ ⇠2µ , (3.87)

is another Killing vector

[L⇠1 , L⇠2 ] gµ⌫ = L[⇠1 ,⇠2 ] gµ⌫ = 0 . (3.88)
For a geodesic xµ and its tangent vector k µ , the inner product ⇠µ k µ with any Killing vector ⇠ µ is conserved along the
geodesic:
D 1
(⇠µ k µ ) := k ⌫ r⌫ (⇠µ k µ ) = k µ k ⌫ r⌫ ⇠µ ⌘ k µ k ⌫ (r⌫ ⇠µ + rµ ⇠⌫ ) = 0 . (3.89)
d 2
In the same way, a current J µ := T µ⌫ ⇠⌫ of an energy-momentum tensor Tµ⌫ and a given Killing vector ⇠⌫ is also
conserved:
1
rµ J µ = T µ⌫ rµ ⇠⌫ = T µ⌫ (r⌫ ⇠µ + rµ ⇠⌫ ) = 0 . (3.90)
2

• Stationary metric.— The spacetime metric is said to be stationary, if there exists a special coordinate, in which the
metric tensor is independent of a time coordinate (or stationary),
@
0= gµ⌫ , (3.91)
@t
which means the metric tensor can depend on x0 in other coordinates. This observation can be generalized by considering
a time-like vector ⇠ µ as
@
⇠ µ := tµ , £⇠ gµ⌫ = gµ⌫ = 0 , (3.92)
@t
i.e., a spacetime metric is stationary, only if a time-like Killing vector exists.

27
AST511 General Relativity JAIYUL YOO

• Static metric.— In the same way, if the metric admits a time-like Killing vector orthogonal to a coordinate hypersurface,
the spacetime is called static. Since the Killing vector is time-like, the spacetime is stationary, and with the orthogonality
to the hypersurface as an extra condition, the space-time cross-term cannot exist. Precisely, one can show that any time-
like Killing vector ⇠ µ is hyper-surface othogonal, only if

⇠[µ r⌫ ⇠⇢] = 0 . (3.93)

For example, a rotating system can have stationary fields, but it is not invariant under time-reversal, while a static space-
time is invariant.

• Spherically symmetric metric.— Furthermore, a spacetime is said to be spherically symmetric, if three linearly inde-
pendent spacelike Killing vectors Xi exist and satisfy

[X1 , X2 ] = X3 , [X2 , X3 ] = X1 , [X3 , X1 ] = X2 . (3.94)

• Maximally Symmetric Space.— is defined as a space that possesses the largest possible number of Killing vectors, and
it is homogeneous and isotropic. On an n-dimensional manifold this number is n(n + 1)/2, and it is easy to understand.
Consider an Euclidean space Rn , where the isometries are translations and rotations — there are n translations, one for
each direction we can move, and n(n 1)/2 rotations, i.e., (n 1) rotation axes for each direction we want to rotate.
This applies to spacetime with non-Euclidean signature. For n 2, there are more Killing vectors than the number of
dimension. Strange at first, because more than n vectors in n-dimension cannot be independent at any point. However,
Killing vectors here mean actually Killing vector fields, and a linear combination of Killing vector fields with constant
coefficients is still a Killing vector field.
Maximally symmetric spacetime is described by a constant curvature (Ricci scalar R) at every space and time. There
exist three cases: Minkowski (R = 0), de Sitter (R > 0), and Anti-de Sitter (R < 0). D-dimensional de Sitter spacetime
is a sphere SD (D 3) embedded in a (D + 1)-dimensional flat spacetime. D-dimensional AdS is a hyperboloid
in a (D + 1)-dimensional flat spacetime. The Riemann tensor and the curvature scalar of a maximally symmetric D-
dimensional spacetime are

Rµ⌫⇢ = K̂ (gµ⇢ g⌫ gµ g⌫⇢ ) , R = D(D 1)K̂ , (3.95)

where K̂ and R are both constant. The normalization convention is consistent with our FLRW notation, i.e., R(3) = 6K̂.

• Birkhoff’s theorem.— The metric outside a spherically symmetric source is static in general relativity (Rµ⌫ = 0). In
Newtonian gravity, time-dependence and spherical symmetry are independent. For example, if a pulsating mass distribu-
tion maintains spherical symmetry, the metric outside the source is static, i.e., no gravitational waves.

3.3 More on the Curvature

3.3.1 Riemann Normal Coordinate
General covariance allows a coordinate system, where the metric is locally Minkowski and the first derivative of the metric
vanishes (so do the Christoffel symbols)
⇢
gµ⌫ = ⌘µ⌫ , gµ⌫,⇢ = 0 , µ⌫ =0. (3.96)

Riemann normal coordinates realize such coordinates by using a set of four geodesics (one time-like and three space-like)
starting from the fixed point P . Fermi normal coordinates are a specific extension of Riemann coordinates such that holds
for every point along a fixed time-like geodesic.
Consider a coordinate transformation between x̃µ and xa , in which xa represents a local normal coordinate. The
metric tensor transforms as
@ x̃µ @ x̃⌫
gab = g̃µ⌫ = g̃µ⌫ [ea ]µ [eb ]⌫ , (3.97)
@xa @xb

28
AST511 General Relativity JAIYUL YOO

and with 16 dof in the transformation at P (tetrad vectors)

@ x̃µ
[ea ]µP := , (3.98)
@xa P

we can make the local metric Minkowski gab = ⌘ab at P (only the symmetric part of 16 dofs participates). We want to
show that the first derivative of the metric in the normal coordinate vanishes at P . Now consider three geodesic paths
emanating from P along the spatial tetrad directions, and we parametrize them in terms of proper distance d = ds.
Using these geodesics and assigning the proper interval to the local coordinate xi , we can construct a coordinate system
in the small neighborhood of a point P
1
X 1 dn x̃µ 1
µ
x̃ = n
= x̃µP + [ei ]µP xi µ
⇢ [ei ]⇢P [ej ]P xi xj + O x3 , xaP ⌘ 0 , (3.99)
n! d n P 2 P
n=0

where we used
dxi d2 x̃µ ⇢
µ dx̃ dx̃ d2 x̃µ
=1, 0= + ⇢ = + µ⇢ [ei ]⇢ [ej ] . (3.100)
d d 2 d d d 2
To compute the metric in the normal coordinate, we need to compute the derivatives of the coordinate transformation:
✓ ◆
@ x̃µ @ x̃µP @
0
= 0
+ 0 [ei ]P xi + O(x2 ) = [et ]µP
µ µ
⇢ [ei ]⇢P [et ]P xi + O(x2 ) , (3.101)
@x @x @x P
@ x̃µ
= 0 + [ek ]µP µ
⇢ [ei ]⇢P [ek ]P xi + O(x2 ) , (3.102)
@xk P

where the tetrad vectors are constructed along a time-like path parametrized by x0

D µ @[ei ]µ µ
0= [e i ] = + ⇢ [ei ]⇢ [et ]⌫ , (3.103)
dx0 @x0
and they are independent of xi .
Therefore, the metric tensor in the normal coordinate around P is to the zeroth order and the linear order in xi :
✓ ◆✓ ◆✓ ◆
@ x̃µ @ x̃⌫ µ µ ⇢ i ⌫ ⌫ ⇢ j P P ⇢ k
gab = g̃µ⌫ = [ea ]P ⇢ [ei ]P [ea ]P x [eb ]P ⇢ [ej ] [eb ]P x g̃µ⌫ + g̃µ⌫,⇢ [ek ]P x
@xa @xb P P
✓ ◆
= [ea ]µP [eb ]⌫P g̃µ⌫
P
+ gµ⌫,⇢ g ⌫ ⇢µ gµ ⇢⌫ [ea ]µP [eb ]⌫P [ei ]⇢P xi + O(x2 )
P
= ⌘ab + 0 + O(x2 ) , (3.104)

where the round bracket at P vanishes, as r⇢ gµ⌫ = 0. Indeed, the local metric tensor at P (xi = 0) is the Minkowski and
the first derivative of the metric at P vanishes.

3.3.2 Geodesic Deviation Equation

Due to the equivalence principle, one cannot measure the gravity in a free fall, as the test particle and the laboratory are
moving together. However, two test particles separated by ⇠ µ will follow different geodesic, and the time evolution of its
separation can be measured. That is described by the geodesic deviation equation in general relativity. In the Newtonian
gravity, it is described by the tidal force (or differential gravity):

d2 ⇠ i ij @2
= ⇠k . (3.105)
dt2 @xj @xk
We will derive the geodesic equation in a curved spacetime. Consider a geodesic path xµ in terms of an affine
parameter , and the geodesic equation is

d2 xµ µ dx⇢ dx dtµ µ dxµ

0= + ⇢ = + ⇢ t ⇢ t = t ⌫ r ⌫ tµ , tµ := , (3.106)
d 2 d d d d

29
AST511 General Relativity JAIYUL YOO

where we defined the tangent direction tµ . Again consider another geodesic path x̃µ parametrized by the same affine
parameter, and the separation between two geodesics at is

⇠ µ := x̃µ xµ . (3.107)

Assuming two geodesics are nearby |⇠ µ | ⌧ 1, we can expand the geodesic equation for x̃µ and derive the equation for ⇠ µ
to the linear order as
d2 ⇠ µ µ
⇢
⌫ dx dx
⇢
µ d⇠ dx
0= + ⇢ ,⌫ ⇠ + 2 ⇢ . (3.108)
d 2 d d d d
Now we compute how the separation vector is transported along the geodesic, or the change in ⇠ µ along the flow:
D µ d⇠ µ
⇠ = t⌫ r ⌫ ⇠ µ = + µ
⇢ ⇠⇢t , (3.109)
d d
and the rate of change in ⇠ µ along the flow (or the acceleration) is
✓ ◆ ✓ ✏ ◆
D2 µ d d⇠ µ µ ⇢ µ d⇠
⇠ = + ⇢ ⇠ t + ✏ + ⇢ ⇠ t t
✏ ⇢
d 2 d d d
d2 ⇠ µ  @ µ ⇢ µ d⇠ 
✏
= + t ⇠ t + t + ✏⇢ µ✏ ⇠ ⇢ t t
d 2 @x ⇢ ✏
d
= 0 + ⇠ ⇢ t t µ
⇢ ,
µ
 ,⇢ + ⇢
✏ µ
✏
✏ µ µ  ⇢
 ⇢✏ = R ⇢ t t ⇠ , (3.110)

called the geodesic deviation equation, where we manipulated in the third line
@ d⇠ ⇢
t µ
⇢ ⇠⇢t = µ ⇢ 
⇢ , ⇠ t t + µ
⇢ t + µ
⇢ ⇠⇢ ( ✏ t
 ✏
t) . (3.111)
@x d
In the Newtonian gravity, the vacuum equation is the Laplace equation:

r2 = 0 , (3.112)

i.e., the tidal tensor is traceless. In the same way, the vacuum equation in a curved space should correspond to

0 = Rµ µ t t , (3.113)

for any given direction tµ , implying that the Laplace equation corresponds to

) Rµ⌫ = 0 . (3.114)

3.3.3 Palatini Equation

⇢
In the Riemann normal coordinate, the Christoffel symbol vanishes µ⌫ = 0, while its derivatives are non-vanishing
⇢
µ⌫, 6= 0. Hence the Riemann tensor in this coordinate is simply

Rµ ⌫⇢ = @⇢ µ
⌫ @ µ
⌫⇢ . (3.115)

Now we change a coordinate, and all quantities will transform accordingly. The variation of the Riemann tensor in the
normal coordinate is then
Rµ ⌫⇢ = @⇢ ( µ⌫ ) @ ( µ⌫⇢ ) , (3.116)
where we used the commutation relation between partial derivatives @µ and variation . Since the variation in the Christof-
fel symbol
⇢
µ⌫ = µ⌫
˜⇢ ⇢
µ⌫ , (3.117)
is the difference in Christoffel symbols, it transforms tensorially (one can verify by an explicit computation), and hence
the expression for the variation of the Riemann in any coordinates should be

Rµ ⌫⇢ = r⇢ ( µ
⌫ ) r ( µ
⌫⇢ ) . (3.118)

This is called the Palatini equation. Contracting µ and ⇢, we then obtain the variation of the Ricci tensor as
⇢ ⇢
Rµ⌫ = r⇢ µ⌫ r⌫ µ⇢ . (3.119)

30