230b1203

More curvature calculations.

January 31, 2012

In todays lecture I will continue to illustrate by examples how to

compute the Riemann curvature of the deSitter universe and show
that it is a solution of the Einstein field equations with
cosmological term. I will use two methods: 1) by relating the
curvature of the metric on a submanifold in the induced metric
from that of the full manifold and 2) more on Cartan
computations, especially to Friedmann-Robertson-Walker metrics
which can be thought of as generalizations of polar coordinate
metrics.
I will begin with some general review, especially about Ricci and
scalar curvature.

Contents.

1 Review of the definition of various curvatures.

2 The curvature of the induced metric of a submanifold.

The case of a hypersurface.

The de Sitter universe and its relatives.
3 More Cartan style computations in semi-Riemannian geometry.

The second Bianchi identity and Schurs theorem.

Friedmann Robertson Walker metrics.

The Levi-Civita theorem and Koszuls formula.

The Levi-Civita theorem says that on any semi-Riemannian

manifold there is a unique connection with with zero torsion which
is isometric. For any vector field V , the Levi-Civita connection V
is determined by the Koszul formula
2hV W , X i =

V hW , X i+W hX , V iX hV , W ihV , [W , X ]i+hW , [X , V ]i+hX , [V , W ]i

where W and X are any two other vector fields.

The hereditary character of the Levi-Civita connection.

Suppose that S M is a submanifold which is itself a

semi-Riemannian manifold under the induced metric from M. That
is, the scalar product of two vectors in Tp S is the same as their
scalar product in Tp M. (Since the restriction of the scalar product
to such a subspace may not be non-degenerate, we have to impose
the condition of non-degeneracy as a hypothesis. In the
Riemannian case this is automatic.)
Let X and Y be vector fields on S and extend them locally to
vector fields on M (which we can do by the implicit function
theorem).

At each p S we have the orthogonal decomposition

Tp M = Tp S + (Tp S) . We will denote the decomposition of any
Tp M according to this direct sum as
= tan + nor .
Now we can consider the covariant derivative X Y where X and
Y are thought of as vector fields on M, or we can consider their
covariant derivative given by the Levi-Civita connection
corresponding to the induced metric on S - let us denote this
covariant derivative by SX Y .
It follows immediately from Koszuls formula that
SX Y = (X Y )tan .

(1)

We will use this fact a bit later to compute the curvature of the
induced metric.

So if C is a non-singular curve on S and Z is a vector field on S

tangent to S, then the derivative of Z along this curve relative to
the Levi-Civita connection of S in its induced metric is obtained by
computing is derivative relative to the Levi-Civita metric on M and
then projecting onto TC (t) S.
In particular, C will be a geodesic on S if and only if, at each t,
C 00 (t) is orthogonal to S where C 00 is computed relative to the
Levi-Civita metric of M.

For example, if M = Rn with its standard metric, then Koszul tells

us that i j = 0 for all i and j where the k are the standard
constant vector fields in the coordinate directions, and hence the
Christoffel symbols vanish and so the derivative of any vector field
d
along any curve is given by Z 0 (t) = dt
Z (t).
Hence a curve on S will be a geodesic if and only if its acceleration
(in the sense of freshman calculus) is orthogonal to S.
For example, the great circles on a sphere in Euclidean space have
this property, hence they are the geodesics on the sphere.

Review: the curvature of a connection.

The curvature R = R() of the connection is defined to be the

map V(M)3 V(M) assigning to three vector fields X , Y , Z the
value
RXY Z := [X , Y ]Z [X ,Y ] Z .
(2)
We know that R is a tensor i.e. that the value of RXY Z at a point
depends only on the values of X , Y , and Z at that point and is
trilinear over R as a function of these values.
In the case that is the Levi-Civita connection, R is called the
Riemann curvature tensor.

The curvature operator.

Consider R(X , Y , W , V ). This is anti-symmetric in X , Y and in

V , W and symmetric under the exchange of the pairs X , Y with
W , V . So it defines a symmetric bilinear form on 2 (TM) by
R(X Y , V W ) = R(X , Y , W , V ).

(3)

Notice the reversal of V and W in this definition.

This bilinear form in turn defines a linear operator
R : 2 TM 2 TM by
hR(X Y ), V W i = R(X Y , V W ).

(4)

The h , i occurring on the left hand side of (4) is the scalar

product induced on 2 (TM): If we are given a a bilinear form
h , i on a vector space V it induces a h , i on each of the exterior
powers k V .
For the case k = 2, for example, this scalar product on
decomposable vectors is given by
hv1 v2 , w1 w2 i = hv1 , w1 ihv2 , w2 i hv1 , w2 ihv2 , w1 i.

Degenerate and non-degenerate planes.

For a pair of vectors in a vector space with a non-degenerate
symmetric bilinear form h , i define
Q(v , w ) := hv , v ihw , w i hv , w i2 .
In the positive definite case this is the square of the area of the
parallelogram spanned by v and w . Assuming that v and w are
independent, if Q(v , w ) = 0, the same is true for any two vectors
in the plane that they span. Such a plane is called degenerate
because the bilinear form restricted to that subspace is degenerate.
Of course, in the positive definite case, every two dimensional
subspace is non-degenerate. In terms of the induced
non-degenerate form on 2 V
Q(v , w ) = hv w , v w i.

We will let = (v , w ) denote the plane spanned by v and w .

In what follows we will assume unless otherwise stated that
(v , w ) is non-degenerate. Notice that a slight perturbation will
carry a degenerate plane into a non-degenerate one.

Sectional curvature.

Define the directional curvature operator Rv by Rv w = R(v , w )v

and the operator R we introduced above. Define
hRv (w ), w i
Q(v , w )
hR(w , v )v , w i
Q(v , w )
hR(v w ), v w i
.
Q(v , w )

sec(v , w ) =
=
=

(5)

From the last expression it is clear that sec(v , w ) depends only on

the plane (v , w ) so we will also write sec(). So 7 sec() is a
real valued function on the space of non-degenerate planes.

The sectional curvature determines the Riemann curvature

tensor.

Over the next few slides I wish to prove that sec (together with the
metric) determines R. This is a purely algebraic result for which
we will need the following definitions and a lemma: We let F be a
four tensor whose values on four tangent vectors we shall write as
F,,, or hF, , i.
We say that F is curvaturelike if it satifies the same algebraic
identities as the Riemann curvature tensor. In more detail:

Curvaturelike tensors.
Definition
A multilinear function F on a vector space is called curvaturelike if
1
2

F = F

hF , i = h, F i

F + F + F = 0

hF , i = hF , i.

If F is curvaturelike tensor we define its analogue of sec by:

Definition
If F is a curvaturelike multilinear function on a vector space with a
non-degenerate symmetric bilinear form we define
KF (v , w ) =

F (w , v , v , w )
,
Q(v , w )

if Q(v , w ) 6= 0

A lemma.

Just as in the definition of the sectional curvature KF (v , w )

depends only on the plane spanned by v and w and so we may
write KF ().
Lemma
If F is a curvaturelike multilinear function and KF = 0 then F = 0.
As a consequence, if F1 and F2 are curvaturelike multilinear
functions with KF1 = KF2 then F1 = F2 .
The consequence follows from the first statement in the lemma
since the evaluation map F 7 F (v , w , v , w ) is linear in F .
To begin the proof observe that the hypothesis is that
F (v , w , w , v ) = 0 for all non-degenerate pairs. But since these are
dense and F is continuous F (v , w , w , v ) = 0 for all v and w .

Proof.
As a second step we will prove that F (w , v , v , x) 0. Indeed we
know from multilinearity that
0 = F (w + x, v , v , w + x)
= F (w , v , v , w ) + F (x, v , v , w ) + F (w , v , v , x) + F (x, v , v , x).
The first and last term on the right vanish, from our assumption.
The two middle terms are equal by one of the curvature-like
identities. So we get 2F (w , v , v , x) = 0 for all v , w , x.
We know that F (w , v , v , x) 0. So F (w , v + y , v + y , x) 0 and
therefore
F (w , v , y , x) = F (w , y , v , x) = F (y , w , v , x).
So F (w , v , y , x) is unchanged by a cyclic permutation of the first
three variables, Then 3. implies that 3F (w , v , y , x) = 0 so
F = 0.


Spaces of constant curvature.

M is said to have constant curvature k if sec() = k (at a point

p) for all non-degenerate two planes at p. This will certainly be
the case if for any three vectors x, y , z at p we have
R(x, y )z = k(hy , zix hx, ziy )

(6)

for then
hR(w , v )v , w i = kQ(v , w ).
A famous result of Riemann says the converse - that if the
sec() = k for all then the curvature has the form given by (6).
Indeed, the function given by the right hand side of (6) is
curvaturelike (as you can check) and the result follows from the
lemma.

All possible values of k can occur as the constant curvature.

Indeed spheres have constant positive curvature, Euclidean spaces
have k = 0 and k < 0 in hyperbolic geometry.
In all of these cases, the sectional curvature is not only constant on
the two planes at a point, but k is the same at all points of M.
Later today we will prove a theorem of F. Schur which says that if
dim M 3, and if the Riemann metric on M has constant
curvature f (p) at each point p of M, then f is a constant. Later
today, I will give very simple proof that I learned from Prof. Chern
(in 1953!) of this fact using the Cartan formalism.

The Ricci curvature.

If we hold TMx and TMx fixed in Rv then the map

v 7 Rv

v TMx

is a linear map of TMx into itself. Its trace (which is bilinear in

and ) is known as the Ricci curvature tensor.
Ric(, ) := tr[v 7 Rv ].

(7)

Ricci curvature plays a key role in general relativity because it is

the Ricci curvature rather than than the full Riemann curvature
which enters into the Einstein field equations.

A useful property of the Ricci curvature.

By polarization and scalar multiplication the Ricci tensor is

determined at a point by its values Ric(u, u) on unit vectors, i.e.
vectors satisfying |hu, ui| = 1. On such vectors, it follows from the
definition that
Ric(u, u) = hu, ui

n
X

sec(u, ei ),

(8)

where e2 , . . . , en is an orthonormal basis of u . In other words,

Ric(u, u) = hu, ui the sum of the sectional curvatures of any
n 1 orthogonal non-degenerate planes through u.

(Riccis definition of) the scalar curvature.

The (Ricci) scalar curvature S is the contraction of the Ricci

curvature. So in terms of an orthonormal basis of TMm
X
S(m) =
sec(ei , ej ).
(9)
i6=j

In particular, if (M, g) is a semi-Riemannian manifold with

constant sectional curvature, k, then it follows from the above that
Ric = (n 1)kg

and S = n(n 1)k.

(10)

We have seen that the Levi-Civita connection of the metric

induced on a submanifold is the tangential component of the
Levi-Civita connection of M.
Let us now consider the normal component (X Y )nor where X
and Y are vector fields on S (extended so as to be vector fields on
M). This assigns to every point s S an element of TMs which is
orthogonal to TSs . It is a vector field of M along S.

Since fX Y = f X Y for f F(M) the value of (X Y )nor at a

point s S depends only on X (s). Let g be a smooth function on
S, extended to be a smooth function on M. Then
X (gY ) = (Xg )Y + g X Y .
Upon taking the normal component, the first term disappears,
since Y is tangential. So (X Y )nor (s) depends only on X (s) and
Y (s). To shorten notation we will set
N (X , Y ) := (X Y )nor
for any X , Y V(S).

Let V , W , X , Y be vector fields on S. We can consider

hRVW X , Y i
where R is the curvature tensor of M, and also
S
hRVW
X,Yi

where R S is the curvature tensor of the induced metric and its

connection. The following formula gives the relation between the
two:
S
hRVW
X,Yi
= hRVW X , Y i hN (V , X ), N (W , Y )i + hN (V , Y ), N (W , X )i.
(11)

Proof.

With no loss of generality we may assume that [V , W ] = 0 so that

RVW = V W W V . We write W X = SW X + N (W , X )
so that
hV W X , Y i = hV SW X , Y i + hV (N (W , X )), Y i.

()

Since Y is tangent to S, the first term in () is the same as would

be obtained by replacing V SW X by its tangential component
which is SV SW X . We massage the second term in () by writing
it as
V hN (W , X ), Y i hN (W , X ), V Y i.

()

hV W X , Y i = hV SW X , Y i + hV (N (W , X )), Y i.

()

The second term in () is

V hN (W , X ), Y i hN (W , X ), V Y i.

()

Since Y is tangent to S the function hN (W , X ), Y i vanishes on S,

and since V is tangent to S the first term in () vanishes. Since
N (W , X ) is normal to S we may replace the V Y in the second
term in () by its normal component which is N (V , Y )i. So ()
becomes
hV W X , Y i = hSV SW X , Y i hN (V , X ), N (W , Y )i.
Interchanging V and W and subtracting yields (11).


The induced sectional curvature.

Let v and w span a non-degenerate plane in the tangent space

TSs at some point s S. Let secM (v , w ) denote the sectional
curvature of this plane considered as a plane in TMs and
secS (v , w ) its sectional curvature considered as a plane in TSs
relative to the induced metric on S. Recalling the definition of the
sectional curvature it follows from (11) that
secS (v , w ) = secM (v , w )+

hN (v , v ), N (w , w ) hN (v , w ), N (v , w )i
.
hv , v ihw , w i hv , w i2
(12)

The case of a hypersurface.

The above formulas simplify somewhat when S is a hypersurface.
The assumption that S is non-degenerate implies that the line
normal to S is non-degenerate. So if y is a non-zero vector normal
to S, we have hy , y i =
6 0. But (in the semi-Riemannian case) this
might be positive or negative.
Let us choose, locally, a normal vector field, with |hU, Ui| = 1 and
let  = hU, Ui 1. For V tangent to S we have
1
0 = V hU, Ui = hV U, Ui.
2
So V U is tangent to S. Let W be another vector field tangent to
S. Then since hW , Ui 0, we have
hV U, W i = hV W , Ui = hN (V , W ), Ui.

So if we define the (1, 1) tensor S on S which assigns to each

vector field V the vector field V U we have
hS(V ), W i = hN (V , W ), Ui.
So
hN (V , W ), N (V 0 , W 0 )i = hS(V ), W ihS(V 0 ), W 0 ihU, Ui
= hS(V ), W ihS(V 0 ), W 0 i

for any four vector fields V , W , V 0 W 0 tangent to S. So (12)

becomes
secS (v , w ) = secM (v , w ) + 

hSv , v ihSw , w i hSv , w i2

hv , v ihw , w i hv , w i2

for any pair of tangent vectors at a point of S which span a

non-degenerate plane.

(13)

The de Sitter universe and its relatives.

In 1917, de Sitter wrote down a solution of the Einstein field
equations (see below in this section) for a universe devoid of matter
provided that these equations contain a cosmological constant.
The de Sitter universe has a vey simple description: it is a quadric
hypersurface ( a pseudosphere) in the five dimensional space
R1,4 - the space R5 equipped with the quadratic form
q(y) = y12 + y22 + y32 + y42 + y52 .
The de Sitter space with parameter r > 0 is just the hypersurface
Sr := q 1 (r 2 ).
We will find, using the method described above, that in the
induced metric, this pseudosphere has constant curvature 1/r 2 .
There is nothing special about the values (1, 4). The argument
works for a pseudosphere in Rr ,s for any r and s.

The tautological vector field on a vector space,

On any vector space we have the identification of the tangent

space at any point with the vector space itself. In particular, with
have the vector field E which assigns to each point p itself,
thought of as an element of the tangent space at p. In terms of a
system x1 , . . . , xn of linear coordinates we have
X
E=
xi i .
If v is a tangent vector at any point of our vector space, we have
X
v E =
vi i = v
where is the covariant derivative of the (flat) metric.

We have
hE, Ei = q.

where q is the quadratic form associated to the metric on R1,4 . So

vq = v hE, Ei = 2hv E, Ei = 2hv , Ei.
In particular, if v is tangent to Sr , the above expressions vanish.
So E is normal to Sr . Therefore
1
U := E
r
is the outward pointing unit normal vector to Sr and the
operator S (from the preceding discussion) defined by
S(v ) = v U,
is given by Sv = v /r .

v tangent to Sr

In other words,

1
S = Id.
r
(
Since the curvature of R 1, 4) vanishes, it follows from
secS (v , w ) = secM (v , w ) + 

hSv , v ihSw , w i hSv , w i2

hv , v ihw , w i hv , w i2

(13)

that
Proposition
The de Sitter space Sr has constant curvature 1/r 2 .
Notice that the the group O(1, 4), the orthogonal group of R1,4
(which is ten dimensional) acts transitively as isometries of Sr (just
as the orthogonal group in three dimensions acts transitively as
isometries of the sphere). So Sr has maximal possible dimension of
its isometry group.

The Einstein field equations with cosmological constant .

In covariant form this is the equation
1
Ric Sg + g = T
2

(14)

as an equation for the Lorentzian metric g in terms of the source

term T which is the stress energy tensor describing the
distribution of matter and energy of the universe.
From
Ric = (n 1)kg and S = n(n 1)k (10)
it follows that if S is a four dimensional manifold with constant
(sectional) curvature k, then
Ric = 3kg

and

S = 12k.

So S is a solution of the Einstein field equations with T = 0 if

= 3k. So

Proposition
The de Sitter universe Sr is a solution of the Einstein field
equations with T = 0 and cosmological constant
=

3
.
r2

I will leave it to the professionals to decide whether to move the

term to the other side of the Einstein equations and consider its
contribution as consisting of dark matter.

The second Bianchi identity.

Take the exterior derivative of Cartans second structural equation

d + = .
We obtain
d d = d.
Substitute d = in the left hand side. The terms
involving cancel, and we are left with
d =
which is known as the second Bianchi identity.

(15)

A theorem of F. Schur.

Theorem
Let M be a connected Riemannian manifold with the property that
at every point p M the sectional curvature sec() is independent
of . Then if the dimension of M is 3 this sectional curvature is
independent of p.

Proof, 1.
Let K (p) denote the sectional curvature at p. So K is a function
on M and, in terms of a coframe field, we are assuming that
ij = K i j .
Take d of both sides. On the left we get
X
(ik jk ki kj )
k

by the second Bianchi identity. On the right we get

dK i j + Kdi j K i dj .
X
= dK i j K
(ki k j i kj k )
k

by the first structural equation.

Proof, 2.
On the right dK i j + Kdi j K i dj .
X
= dK i j K
(ki k j i kj k ).
k

Now K

ki k j =

X
k

ki kj and

i kj k =

X
k

ik kj =

X
k

ik jk

by the antisymmetry of . So we arrive at the conclusion that

dK i j = 0
P
for all i and j. Expand dK = k ak k . If dim M 3 we can, for
each k find i and j both 6= k so ak = 0 and hence dK = 0. 

Friedmann Robertson Walker metrics.

These are metrics on the product M of an interval I R (with a

negative definite metric dt 2 ) and a three dimensional Riemannian
manifold S of constant curvature. The metric has the form
dt 2 + f 2 d 2
where f is a function on I and d 2 is the metric on S.
The reason for the constant curvature assumption has to do with
Schurs theorem. If we believe that space time is of the form I S,
and if we believe that at all points of S all directions are the are
the same, a sort of democracy principle, then if follows from
Schurs theorem that the curvature must be constant.

One of the consequences of the computations we will do here are

that the Einstein field equations reduce to an ordinary differential
equation, from which we can deduce the big bang under
plausible additional assumptions. This result is due to Friedmann.
Over the next few slides I will do Cartan calculations without
assuming that d 2 has constant curvature, and then specialize to
the constant curvature case. I will keep the condition that S be
three dimensional in order that the formulas fit on a page. But
there is no difference between 3 and n 3.

The coframe field.

Let 1 , 2 , 3 be a coframe field on S so that

0 = dt, 1 = f 1 , 2 = f 2 , 3 = f 3
is a coframe field on M and the metric is given by
(0 )2 + (1 )2 + (2 )2 + (3 )2 .
We seek satisfying d + = 0 and the
symmetry-anti-symmetry conditions
i0 = 0i , 00 = 0,

ji = ij ,

i, j = 1, 2, 3.

Since d0 = d(dt) = 0, we know that i0 must be a multiple of

i , i = 1, 2, 3. In fact, we can be more precise: For i = 1, 2, 3 we
have i = f i so
di = df i + fdi = f 0 dt i + fdi =

f0 0
i + fdi .
f

From this we draw the conclusions that

0i = i0 =

f0 i
,
f

i = 1, 2, 3

and that these ji on M coincide with the ji of S (pulled back to

M).

f 1
f 2
0

f
f 0 1 f
0
21

= ff 0 2
1
f 2
0
f0 3
1
3 32
f

f0 3
f

31

32

where the ji , i = 1, 2, 3 are the connection forms on S.

Computing d.
In the lower 3 3 block we will simply write dji (coming from S)
solet us look at the
We have
 top row:
f0 i
f0
f0
i
d f = d f + f di ,
d
and

 0
f 00 f (f 0 )2 0
f
=
,
f
f2
3

X
f0
d = i 0
ji j .
f
i

j=1

So the terms involving (f 0 )2 /f 2 cancel, and the entries in the top

row of d are:
f 00
f0 X i
i0 := 0 i
j j .
f
f
j

So

0
10
20
30
0
0
d21 d31
1

d =
0 d 1
0
d32
2
2
30 d31 d32
0
00
0 P
i
j
where i0 := ff 0 i ff
j j . Lets look at the top entry

0 
in the second column of . It is ff 2 21 + 3 31 . This
cancels the term involving f 0 /f in 10 . So the expression for the
curvature is:
= d + =

 = d + =

f !! 0
f

f !!
0 1
f

!!
f 0
f 2

f !! 0
3
f

0
$ ! %2
f
f

$ ! %2
f
f

f !! 0
f

$ ! %2
f
f

2 1 + 21
3 1 + 31

1 2 + 12
0

$ ! %2
f
f

f !! 0
f

3 2 + 32

where the ij are the curvature forms of S.

$ ! %2
f
f

$ ! %2
f
f

1 3 + 13
2 3 + 23
0

Now assume that S has constant curvature.

Now let us make the assumption that S has constant curvature k

so ij = ki j = fk2 i j . We get =

f 0
f

f 0 1
f

f 0 2
f

f
0
3
f

f
f

f
f

f 0

0
+

k
f2

k
f2

f
2

f
f

2 1
3

f
f

+ fk2 1 2
0
+

k
f2

3 2

f
+ fk2 1 3

.
2
f
k
2
3
+

f
f2

0
f 0

f
2

The expanding universe and the big bang.

From this it is easy to compute the Ricci curvature and the scalar
curvature. Let us look at the special case of our formula for
where k = 0 and f = e t/r . Then the coefficients of all the i j
are equal to 1/r 2 . In other words, the space with this
Robertson-Walker metric has constant curvature 1/r 2 . So it is
locally symmetric. In fact, on general principles, it must be
isometric to an open subset of the deSitter space-time that we
studied above. Thus the Friedmann-Robertson-Walker metric
dt 2 + e 2t/r (dx 2 + dy 2 + dz 2 )

(16)

(so with flat space) describes an open subset of deSitter space.

To see this explicitly, consider the map

(t, x, y , z) 7 (W , V , X , Y , Z ) given by
V +W

= e t/r

V W

= (x 2 + y 2 + z 2 )e t/r + r 2 e t/r

= e t/r x

= e t/r y

= e t/r z.

Then W 2 + V 2 + X 2 + Y 2 + Z 2 = r 2 and the metric

dW 2 + dV 2 + dX 2 + dY 2 + dZ 2 pulls back, under this map to
the metric (16) given above, as can easily be checked.

Notice that the hypersurfaces t = const. go over under this map

to the intersection of the quadric with the hyperplanes V + W =
constant. Since e t/r > 0, we must have V + W > 0 so only half
of deSitter space is covered by this map.
It is this model (and modifications thereof) which are currently
used in cosmology.

In deSitters original model, he chose his cosmic time so the the

quadric was sliced by horizontal hyperplanes, rather than the
diagonal ones as in the figure which was Lemaitres choice of
cosmic time. This led to a stationary universe.

Back to more general Robertson-Walker metrics.

According to Hubble (1929) all distant galaxies are moving away

from us at a rate proportional to their distance. The distance
between p (t) and q (t) in S(t) is f (t)d(p, q) where d is the
Riemannian distance in S. Hubbles constant
H0 =
is estimated as

f 0 (t)
f (t)

1
.
(18 2) 109 yr

In particular f 0 is positive so the universe is expanding.

The distribution of energy and pressure in the universe, together

with the Einstein field equations (which become ordinary
differential equations) show that f 00 < 0. This, together with the
value of Hubbles constant, shows that there must have been a
singularity between ten and twenty billion years ago. In other
words, that the universe as we know it had a definite beginning
somewhere between ten and twenty billion years ago.