Professional Documents
Culture Documents
General informa-
tion
Advanced General Relativity and Cosmology
Instructor: Dr. Sergei Winitzki
Oce: Philosophenweg 19, Room 014
Oce hours: Thursdays 14:00-16:00
Home page: http://www.tphys.uni-heidelberg.de/~winitzki/AGRC/
Lectures: Wednesday and Friday, 11:15-13:00,
Philosophenweg 19, Seminarraum
Material shown in green was not presented due
to lack of time or for other reasons
1
Course outline
Problem of the year: ination and dark energy
Tensor calculus explained without indices
Cosmological evolution in f(R) gravity
Birrell-Davies demystied: Basics of quantum
eld theory in curved spacetime
Classical and quantum theory of cosmological
perturbations
Geometry of null surfaces, causal structure,
chronology protection, time machines in GR
Thermodynamics of horizons, entropy of black
holes, "holographic principle"
Pseudotensors demystied: Energy in asymp-
totically at spaces; EMT of gravitational waves
Hawking-Ellis demystied: Classical singularity
theorems
2
Problem of the year
Cosmological observations yield:
- primordial ination
- late-time ination
Early work on ination:
A. Starobinsky, Phys. Lett. B91, p.99, 1980
de Sitter stage in cosmological evolution
Mechanism: modication of gravity due to
quantum backreaction, for example:
S[g] =
_
d
4
x
g
_
R
16G
+R
2
_
Inationary spacetime: de Sitter with at spa-
tial sections, Hubble rate H:
g
dx
dx
= dt
2
e
2Ht
_
dx
2
+dy
2
+dz
2
_
3
Problem of the year
To explain dark energy and primordial ination
within a single model of modied gravity, and
to remain in agreement with other tests of GR.
A model of modied gravity is f(R) gravity:
S =
1
16G
_
d
4
x
g (R +f(R)) +S
matter
Recent proposals: f(R) = c
1
_
1 +c
2
R
n
_
1
How to describe consequences of f(R) gravity?
- Need equations of motion for g
- Solutions for Solar system tests
- Homogeneous cosmological solutions: have
ination and/or dark energy?
- Primordial cosmological perturbations
There are > 50 papers on aspects of f(R)
gravity in 2006-2007
4
Deriving the EOM for f(R) gravity
Note: f(R) is nonlinear EOM are 4th order.
Equations for g
are complicated
Use the method of Lagrange multipliers! Then
we can reduce the Lagrangian to one that is
linear in R and then EOM will be 2nd order
Analogy with mechanics:
S =
_
dt
_
1
2
x
2
+f(x, x)
_
nonlinear in x
EOM would be 2nd order if S were linear in x
Trick: Introduce new variable y and Lagrange
multiplier that enforces y = x:
S =
_
dt
_
1
2
x
2
+f(x, y) +( x y)
_
EOM for y is f
,y
(x, y) = 0.
Since f
,yy
,= 0, we can solve for y through x, :
y = Y (x, ). Also replace x
x. Finally:
S =
_
dt
_
1
2
x
2
x +f(x, Y (x, )) Y (x, )
_
Obtained a rst-order Lagrangian with extra
d.o.f. and potential energy V (x, ) = f f
5
Whether one can/cannot substitute into
Lagrangians
Suppose we have L(x, x, y, y, ...); we can change
variables x, y A(x, y), B(x, y), ...
We cannot change variables to new variables
involving derivatives, x A(y, y, ...)
Example: L = x
2
, substitute x = y + y, get
wrong EOM: y
(4)
y = 0
Suppose we have L(x, x, y, y, ...) and one of the
EOM is x = f(y, y, ...) We can eliminate x.
Suppose we have constraints: L+F. We can
substitute F into L as long as L+F remains.
This works even if F involves derivatives!
But: we cannot substitute EOM into L if
EOM is x = f(x, x, ...) or x = f( y, z) (change
the number of relevant initial conditions). Also
cannot substitute arbitrary relationships, e.g. spe-
cic solutions x = f(y, y, const), unless they
are enforced by constraints.
6
Practice problems:
1. Show that L = xf(x, x) can be reduced to
1st order by adding a total derivative
d
dt
M(x, x).
2. Reduce L = f(x, x, x,
...
x ) to 1st order by
introducing appropriate new variables.
3. Reduce L =
...
x f(x, x, x) to 1st order (rst
reduce to 2nd order, i.e. remove
...
x ).
Incorrect substitution of EOM into Lagrangian
Example: L =
1
2
x
2
+
1
2
x y
2
Equations of motion:
x =
1
2
y
2
d
dt
(x y) = 0
Solve y = C
1
x
1
, substitute into EOM for x:
x =
C
2
1
2x
But: what if we substitute y = C
1
x
1
into L?
L =
1
2
x
2
+
C
2
1
2
1
x
Equation of motion for x now is wrong:
x =
C
2
1
2x
2
7
Want to derive EOM for f(R) gravity!
Apply this technique to f(R) gravity:
S[g] =
1
16G
_
d
4
x
g
_
f rf
+R
_
Now apply a trick: conformal transformation:
g g e
2
g
Choose such that R
g becomes
R
g
plus some terms need a formula for
R!
Result: GR with scalar eld and metric g
Einstein frame vs. Jordan frame these are
dierent variables, not reference frames
Note: coupling to matter is through g, not
through g matter feels and g
8
Tensor calculus without indices
Geometric view: manifolds, topology
Geometric operations on vector and tensor elds
Lie derivative, metric, curvature
Index-free notation: instead of
u
;
= 0 u
_
g
_
= 0
write
u
u = 0
u
g(u, u) = 2g(
u
u, u) = 0
Practical tools for calculations without indices
Vielbein formalism; curvature and Einstein eqs.
9
Quantum eld theory in curved spacetime
Key applications in cosmology:
* Prediction of the primordial uctuations
* Dynamics of chaotic ination
Key applications in black hole physics:
* Hawking radiation
* Black hole entropy
* Thermodynamics of horizons
Many calculations do not require hard QFT!
10
Causality and horizons
In GR, matter determines the causal structure
Time machines? Chronology protection?
Horizons are boundaries of causal regions
Horizons have thermodynamical properties!
Black holes have temperature and entropy
Holographic principle: a bound on entropy
due to quantum eects and nonlinear coupling
to gravity (not due to quantum gravity)
Can study BHs without quantum gravity!
11
Energy and conservation laws in GR
Ill-dened local energy density; pseudotensors
Energy in asymptotically at spaces
Energy of matter + gravity is conserved!
Energy-momentum tensor of gravitational waves
12
Classical singularity theorems
GR predicts singularity in many cases
Limit of applicability of classical GR!
Proofs of singularity theorems have 3 parts:
* Geodesic focusing theorem
* Energy conditions guarantee focusing
* Topological argument ad absurdum
Applications:
* Gravitational collapse of stars
* Past cosmological singularity (Big Bang)
* Future cosmological singularity (Big Crunch)
13
Lecture 2, 2007-10-19. Manifolds
Gravity in GR curved geometry, not a force
Observations: Small scales R
4
, large scales ?
Manifold / is a generalized curved space
Visualize manifolds as embedded surfaces
Examples: sphere x
2
+y
2
+z
2
= R
2
.
Torus:
__
x
2
+y
2
A
_
2
+z
2
= R
2
, A > R
0
z
A
R
14
Local coordinates
Locally like R
n
means local coordinate system
which is a map | R
n
, where | / is a
patch; all of / must be covered by patches
Local coordinates on torus: 0 < ( , ) < 2,
x = cos , y = sin, z = Rcos ,
= A +Rsin
Another choice of coordinates: =
a
bsin
,
z =
c cos
bsin
, for appropriately chosen a, b, c. The
coordinate change map R
n
R
n
is smooth
Intrinsic description: no embedding needed!
Coordinates are scalar functions on /
Functions on manifold = functions on patches
= functions on subsets of R
n
Intrinsic denition of smoothness: smooth func-
tions on subsets of R
n
+ smooth coordinate
change maps
15
Tangent spaces
Vectors in physics mean velocities / directions
Visualize a vector as velocity of a point moving
within /. Path (t) given in local coords
In embedding picture: vector is in tangent
plane
E.g.: hypersurface f(x) = 0, normal vector
n
j
=
x
j
f(x
0
), tangent plane is n (x x
0
) = 0
Tangent planetangent space T
p
/ at point p
Intrinsic picture: velocity means change; tan-
gent vector v = derivative operator on scalar
functions f : / R; this derivative is with
respect to the time of the moving point
D
f =
d
dt
f((t)) =
j
f
x
j
x=(t
0
)
d
j
dt
t=t
0
The derivative operator is v
j
d
j
dt
x
j
. Write
v f instead of D
f.
Tangent vectors v form a vector space; basis
tangent vectors are
x
j
j
.
16
Calculations with tangent vectors
Chain rule: v [f(g)] = f
(g)v g
Consequence: components of v in a coordi-
nate system x
are computed as v x
, and
then v =
(v x
)
x
.
Practice problems: 1. Find components of
x
x
+y
y
in coordinates a = x +y, b = 3xy
2. In 2D, nd tangent vector to the curve
(t) = x = cos t, y = 2sint in coords x, y
3. Given u =
x
+x
y
, nd coords X, Y such
that u =
Y
.
Other properties: v (fg) = fv g + gv f,
linearity v (f + g) = v f + v g as a 1st
derivative. All 1st derivatives are equivalent to
tangent vectors.
Tangent vectors as short curve segments:
f
f((t
0
+) (t
0
))
, f(p
)f(p) ( )f
Local vector eld = choice of tangent vector
v[
p
T
p
/ at each point p | /
Flow of vector eld; show that v t = 1
17
Commutator of vector elds
Consider [u, v] f = u (v f) v (u f)
This is a 1st derivative operator, thus [u, v] is
a new tangent vector
Example: [x
x
,
x
+
y
] has no 2nd derivatives
Proof: show that [u, v] (fg) = ... [Derivation]
Practice problem: compute commutators [a, b],
[a, c], [b, c] for
a = x
x
+y
y
+z
z
, b = x
y
y
x
, c =
z
Geometric view: [a, b] is the arrow p
p here:
a
b
p
p
0
p
p
1
p
2
Approximately f(p) f(p
) (t)
2
[a, b] f[
p
0
18
Lecture 3, 2007-10-24. Connecting vector
elds
Vector eld c is connecting for v if [c, v] = 0
(these vector elds commute)
Connecting vector c ,= v points across v and
along lines of equal time, for some choice of
time across the ow
s=0
= 3
= 2
= 1
v
c
Note: coordinate vector elds
x
always com-
mute (no discrepancy when following coord. lines)
Question: Given a vector eld v, can we nd
coords x
s.t. v =
x
1
? (Yes. Need a basis
of connecting vector elds for v. But [c, v] = 0
is an ODE for c along ow of v.)
19
Practice problems: 1. A vector v is tangent
to a surface f(p) = const i v f = 0. If u, v
are tangent to the surface, show that [u, v] is
also tangent to the same surface.
2. A ow is given explicitly by the equations
x, y, z x +y, y, z x
Compute the vector eld v that generates this
ow. Find two examples of connecting vector
elds for v.
Dierential forms
1-form is a linear function chosen in each tan-
gent space (element of cotangent space T
p
/)
Example: v v f for xed f; this 1-form is
denoted df, so (df) v v f. The 1-form df
is called the gradient of f
Interpretation: (df) v is the change of f over
a short curve segment represented by v
Coordinate 1-forms: dx, dy, dz are gradients
of coordinate functions x, y, z
(dx)
x
= 1, (dx)
y
= 0, etc.
Practice problems: 1. For = d(x
2
y
3
) and
v = y
2
x
, compute v
2. Let a y (a x) y
2
2a y; express
through dx, dy
3. Find all 1-forms = adx + bdy such that
(
x
+x
y
) = 0
4. Find all vectors v = a
x
+ b
y
such that
(xdy +ydx) v = 0
Tensor elds = tensor product of vector elds
and 1-forms
20
Calculus of n-forms
Antisymmetric tensor product of 1-forms:
( ) (a, b) = (a)(b) (b)(a)
Exterior dierential:
d() = (d) + d
d( ) = (d) +()
[[
d
d(d) = 0
Examples: d(xdx) = 0; d(xdyydx) = 2dxdy
General formulas: ( is 1-form, is n-form)
(d) (x, y) = x (y) y (x) ([x, y])
(d) (v
1
, ..., v
n+1
) =
n+1
s=1
()
s1
v
s
(v
1
, ... v
s
..., v
n+1
)
+
1r<sn+1
()
r+s1
_
[v
r
, v
s
] , v
1
, ... v
r
v
s
..., v
n+1
_
Insertion operator: (
x
)(a, b, ...) = (x, a, b, ...)
Integration of over a domain |:
_
|
Stokes theorem:
_
|
d =
_
|
p
/ we nd = df (else
v : v f ,= 0, (v) = 0) d = 0.
2. From d = 0, need to show that
ows of v : (v) = 0 lie within a single sur-
face. Commutator of u, v should remain within
the surface! Need to show u, v : (u) = 0,
(v) = 0 ([u, v]) = 0. Choose a vector
t such that (t) = 1 (transverse). Com-
pute: 0 = ( d)(t, u, v) = (t)d(u, v) =
d(u, v) = u (v) v (u) ([u, v]), thus
([u, v]) = 0.
Practice problem*: If F = 0, where F is
n-form and is 1-form, show that : F =
where is a suitable (n 1)-form.
22
Lie derivative
Need derivative of vectors in direction v but...
Cannot add tangent vectors at dierent points!
Solution: transport vectors by the ow of v
v
a
p
b
p
/
v
a = lim
t0
T
1
v
_
a(p
a(p)
t
/
v
X measures how much X diers from T
v
X
(i.e. X transported by the ow of v)
Properties that follow from this picture:
* /
v
f = v f on scalar functions f
* /
v
a = [v, a] on vectors a
(Note: connecting vector is exactly transported.)
* /
v
(X +Y ) = /
v
X +/
v
Y +(/
v
) Y
* /
v
( a) = (/
v
) a + /
v
a on 1-forms
Note: /
v
X depends on the ow of v, not only
on the value of v at a point!
Lie derivative of tensors: /
v
(XY ) = (/
v
X)
Y +X /
v
Y ; same for wedge product
23
Lecture 4, 2007-10-26. Calculations with
Lie derivatives
/
v
X depends on derivatives of v as well as on
derivatives of X:
/
v
a = /
v
a (a ) v ,= /
v
a
Lie derivative of 1-form:
(/
v
) u = v ( u) [v, u]
Compute /
x
(dy):
_
/
x
dy
_
v = /
x
(v y)
dy [
x
, v] = v (dy
x
) = 0.
Lie derivative of a bilinear form: (/
v
A)(x, y) =
v [A(x, y)] A([v, x] , y) A(x, [v, y])
Cartan homotopy formula: /
x
=
x
d +d
x
x
dx; /
x
x
(x
y
);
/
x
y
(xydx); /
x
y
+xy
x
_
2z
2
dz
_
; /
x
x
(xdx dy).
2. Let A(a, b) (a x) (b y) z(a y) (b z) x;
compute /
x
y
A. (A is a 2nd rank tensor)
3. Let
T(a)b = (b x) ya(a y) xb; compute
/
T. (
1
T = 0 for a tensor T =
,,...
T
...
dx
dx
1
dx
= 0 and
/
= 0
Interpretation: if /
v
T = 0 then components of
T are constant along the ow of v
Which components? In a basis of connecting
vectors for v
The property /
v
T = 0 does not depend on
other basis vectors, only on v!
Note: x-component of a tensor depends on the
function x(p) and on the vector eld
x
. But
x
depends not only on the function x(p), but
also on the choice of every other coordinate.
25
Metric
Need to compute distances along paths
Visualize using the embedding picture:
z
a
y
x
b
Distance between a and b along / is
D(a, b)
_
(a
x
b
x
)
2
+(a
y
b
y
)
2
+(a
z
b
z
)
2
For innitesimal lengths along x(), y(), z():
L =
_
dx
d
_
2
+
_
dy
d
_
2
+
_
dz
d
_
2
Use intrinsic coordinates (u, v): L =
=
_
(x
,u
u +x
,v
v)
2
+... =
_
A u
2
+2B u v +C v
2
Need to know only A, B, C to compute length!
The length in larger space gives the induced
metric. Intrinsically, metric is a quadratic form
describing the length of innitesimal curve seg-
ments : L
_
g( , )
Note: in GR the metric has signature +
26
Applications of metric structure
Example: if g( , ) = A u
2
+ 2B u v + C v
2
in
coords. (u, v), then
g = Adudu+B(du dv +dv du)+Cdvdv
Shorthand notation: g = Adu
2
+ 2Bdudv +
C dv
2
Practice problem: compute the metric for
the surface z = x
2
+y
2
in coords. x, y
* Metric maps vectors into 1-forms and back
v gv : (gv) x g(v, x)
g
1
: g(
_
g
1
_
, x) x
* Orthonormal frame: e
a
; g(e
a
, e
b
) =
ab
* Dual basis of 1-forms
a
;
a
e
b
=
a
b
Levi-Civita tensor
0
1
2
3
independent of basis, up to orientation
(a, b, c, d) = the 4-volume of a parallelepiped
Components in a coordinate basis: g
g(
)
All 4-forms are proportional to . Formula:
= [det g
[
1/2
dx
0
dx
1
dx
2
dx
3
Proof: volume spanned by
equals (
0
, ...)
and also equals [det g(
)[
1/2
(introduce transition matrix
= M
, then
the volume spanned by
equals det
M)
27
Ane connection (covariant derivative)
Want to have directional derivative of vectors:
v
x that depends only on the value v[
p
(Lie
derivative /
v
x contains derivatives of v at p)
Problem: need to subtract vectors in dierent
tangent spaces; need a transport operator
T:
v
x = lim
0
T
p
p
x[
p
x[
p
to p
Tx = x +n
T
p
p
n
x[
p
p
p
x[
p
v
x
(orthogonal projection w.r.t. g onto T
p
/)
Proj
(n)
x x n
g(n, x)
g(n, n)
Need a formula for
v
x without embedding
Note properties:
u
v
v
u = [u, v];
u
g = 0
Dene Tors (a, b)
a
b
b
a [a, b]; ver-
ify that Tors (a, b) is dependent of derivatives;
then Tors (a, b) = Proj
(n)
Tors (a, b) = 0
Verify
v
g = 0 for tangent vectors v:
(
v
g) (a, b) = v g(a, b) g(
v
a, b) g(a,
v
b)
= v g(a, b) g(
v
a, b) g(a,
v
b) = 0
v
x has the properties of 1st order derivative.
Verify:
v
x is independent of derivatives of v:
v
x = Proj
(n)
v
x = Proj
(n)
v
x =
v
x
Express
v
x through the intrinsic metric g!
29
Levi-Civit (LC) connection
Properties: the connection is torsion-free:
[x, y] =
x
y
y
x
Compatible with the metric:
x
g(a, b) = g(
x
a, b) +g(a,
x
b)
Will now derive an explicit formula for
y
x.
Dene derivative tensor B
(x)
(a, b) g(
a
x, b).
Consider rst the antisymmetric part of B
(x)
:
B
(x)
(a, b)B
(x)
(b, a) = g(
a
x, b)g(
b
x, a) =
a
g(x, b)
b
g(x, b) g(x, [a, b]) = (dgx)
(a, b).
Symmetric part: B
(x)
(a, b)+B
(x)
(b, a) = g(
x
a
/
x
a, b)+g(a,
x
b/
x
b) =
x
g(a, b)g(/
x
a, b)
g(a, /
x
b) = (/
x
g) (a, b).
Hence
g(
a
x, b) =
1
2
(d(gx) +/
x
g) (a, b)
Fully explicit formula (Koszul): 2g(
a
x, b) =
ag(x, b) bg(x, a) g(x, [a, b]) +xg(a, b)
g([x, a] , b) g(a, [x, b])
30
Geodesic vector elds and curves
A vector eld v is geodesic if
v
v = 0.
A curve () is geodesic with ane parameter
if
v
v = 0, where v = is the velocity.
Can change A +B if A, B = const
Normalization property:
v
g(v, v) = 0, so we
can choose such that g(v, v) = 0, 1
Null geodesic: g(n, n) = 0. If observed in a ref-
erence frame with 4-velocity u, the frequency
of the photon is = g(u, n).
31
Killing vectors
Motivation: geometry does not change in the
direction of a symmetry
Example: g = dt
2
e
2N(t)
_
dx
2
+dy
2
+dz
2
_
does not change in directions x, y, z
Generalize: components of g are constant along
the ow of v
A vector eld v is a Killing vector if /
v
g = 0.
Example: g = f(r)dt
2
g(r)dr
2
r
2
d
2
has
Killing vectors
t
,
(and others!)
Formula: (/
x
g) (a, b) = g(
a
x, b) +g(a,
b
x)
Killing equation:
k : g(
a
k, b) +g(a,
b
k) = 0 for a, b
Note: for a Killing vector k, the symmetric part
of B
(k)
vanishes B
(k)
=
1
2
dgk is a 2-form
Conformal Killing vector: /
k
g = 2g for some
scalar . Geometry is conformally rescaled
under the ow of k.
Example: FRW universe, g = dt
2
a
2
(t). Ex-
pect conformal Killing vector k = f(t)
t
with
some f(t). Calculation: assume a = a
t
t
+a
x
x
such that
t
a
t,x
= 0, then (/
k
g) (a, b) =
f
t
_
a
2
_
(a, b) +2f
a
t
b
t
. Thus we must have
2f
= 2
a
a
f, hence f(t) = a(t) const.
32
Redshift factor and gravitational potential
Metric g is stationary if there exists a timelike
Killing vector k : g(k, k) > 0, /
k
g = 0.
Note: metric g is static if g(k, k) > 0, /
k
g = 0,
and k is everywhere orthogonal to a surface.
Stationary observers move with 4-velocity
k
g(k,k)
Consider a photon along null geodesic (); lo-
cally observed frequency (energy) is =
g(k, )
g(k,k)
.
Statement: g(k, ) = const along geodesic .
Proof: Let v = , then
v
g(k, v) = g(
v
k, v) =
1
2
(dgk +/
k
g) (v, v) = 0.
Hence =
0
[g(k, k)]
1/2
. Dene redshift z =
_
g(k, k).
Photons are redshifted by factor z
33
Lecture 5, 2007-10-31. Geodesics extrem-
ize proper length
Statement: Geodesic curves () extremize
proper length
_
_
g( , )d
Proof (for timelike curves): Deform () by
transverse vector eld t; obtain connecting eld
v; compute: /
t
_
_
g(v, v)d =
_
g(
t
v,v)
g(v,v)
d =
_
g(
v
t,v)
g(v,v)
d =
_
v
(...) d
_
g(t,
v
v
g(v,v)
)d.
Hence,
v
v
g(v,v)
= 0. Reparameterize
_
_
g(v, v)d to achieve g( , ) = 1.
Note: for spacelike curves need
_
g(v, v); for
null curves need
_
Ng(v, v)d.
Universal variational principle:
_
_
g( , )
N
+KN
_
d = 0 where N N()
EOM: g( , ) = KN
2
;
v
_
N
1
v
_
= 0
For null geodesics: choose K = 0 (but N ,= 0!)
34
Killing vectors II
Compute:
(/
k
g)(a, a) = 2g(
a
k, a) = 2(a ) g(k, a)+
(/
k
g)(a, a), so /
k
g = /
k
g+dgk+gkd
or (/
k
g)
= (/
k
g)
+
,
k
,
+
,
k
,
/
k
(g) = (k ) g +/
k
g
Example: FRW metric g = dt
2
a
2
dx
2
, let us
determine conformal Killing vector k = f
t
:
/
f
t
g = f/
t
_
dt
2
a
2
dx
2
_
+dfg
t
+g
t
df =
2fa a dx
2
+2
fdt dt
If f = a then /
f
t
g = 2 ag.
For timelike conformal Killing vectors: redshift
z =
_
g(k, k) still meaningful since
g( , k) =
1
2
(dgk +/
k
g) ( , ) = 0 for null geodesics
Note: (/
k
g) (k, k) = 2g(
k
k, k) = 2g(k, k),
so =
g(
k
k,k)
g(k,k)
=
k
lnz
Also compute g(
k
k, x) = 2g(k, x)g(
x
k, k) =
2g(k, x)
1
2
x
g(k, k), thus
k
k = zg
1
dz +2k
Practice problem: Can one always choose a
vector eld v such that all covariant derivatives
vanish at a point? (a :
a
v = 0)
35
Redshift and gravitational potential
Consider a stationary spacetime with a Killing
vector k; redshift z =
_
g(k, k)
Redshift is time-independent:
k
z = 0 (since
k
lnz = = 0 for a Killing vector)
Is k geodesic? No:
k
k = zg
1
dz ,= 0 unless
z = const.
Stationary observers move along u =
1
z
k.
Acceleration of stationary observers, a =
u
u =
g
1
d where is the gravitational potential.
Proof: with u = z
1
k we have
u
u =
1
z
k
k
z
=
1
z
2
k
k =
1
z
g
1
dz = g
1
d where lnz.
Example 1: Schwarzschild spacetime, k =
t
,
g =
_
1
2M
r
_
dt
2
_
1
2M
r
_
1
dr
2
r
2
d
2
z =
_
1
2M
r
_
1/2
; =
1
2
ln
_
1
2M
r
_
M
r
Example 2: FRW spacetime, k = a(t)
t
,
g = dt
2
a
2
dx
2
, /
k
g = 2g
z = a(t), = a(t); u =
t
; but
u
u = 0:
2g(
u
u, x) = (dgu +/
u
g) (u, x) = 0
36
Force at a distance. Surface gravity
Consider stationary spacetime; /
k
g = 0
Energy transfer and conservation: zE = zg(u, )
is conserved in local processes
Stationary observers feel acceleration in trans-
verse direction, a = g
1
dlnz
Transferring rope tension: 1) If rope moves by
r and transfers energy E, then Fr = E.
2) If energy E is sent back by a particle, then
zE = const. Hence zF = const
Force needed locally to held a unit mass sta-
tionary: F
0
= g
1
dlnz
Force at innity (z = 1): F
= zF
0
= g
1
dz;
[F
[ =
_
g
1
(dz, dz)
Example: Schwarzschild spacetime; k =
t
;
z =
_
1
2M
r
; dz =
M
r
2
dr
_
1
2M
r
; g
1
dz =
M
r
2
_
1
2M
r
r
;
[F
[ =
M
r
2
. This is still nite at horizon, r =
2M, [F
[ =
1
4M
- surface gravity
Note: at horizon k is null so the formula for
k
k = zg
1
dz does not apply.
37
Change of LC connection under conformal
transformation
Consider g = e
2
g. What is the relationship
between
and ?
Compute: 2g(
a, ) = d
ga +/
a
g = de
2
ga +
/
a
e
2
g = e
2
(dga+/
a
g)+2(d)
ga+2(a ) g,
hence g(
x
a, y) = g(
x
a, y) +(x ) g(a, y)
(y ) g(a, x) +(a ) g(x, y) or, without y,
x
a =
x
a+(x ) a+(a ) xg(a, x)g
1
(d)
Practice exercises: 1. If k is a Killing vector
for g, show that k is a conformal Killing vector
for e
2
g, for any .
2. If () is a null geodesic for metric g, show
that a reparameterization is possible
such that ( ) is a null geodesic for g = e
2
g.
3. Let k be a Killing vector for g and () be
a geodesic. Show that is transported by the
ow of k to another geodesic. (If [v, k] = 0
then /
k
v
v = 0.)
4. In a stationary spacetime, n photons are
sent each second from point a to point b. How
many photons per second arrive at b? (Express
the answer through the redshift function z.)
38
Lecture 6, 2007-11-02. Energy extraction
from horizons
Consider a stationary spacetime, /
k
g = 0, with
a Killing horizon g(k, k) = 0
Lower a mass m very slowly (adiabatically) on
a strong rope from an initial point p
0
to point
p
1
near horizon. How much energy can be
extracted at point p
0
?
Force needed to hold m at p
1
is F
1
= mg
1
dlnz[
p
1
Due to redshift, force at p
0
is F
0
=
z
1
z
0
F
1
=
m
z
0
g
1
dz[
p
1
To describe movement, introduce vector eld
s =
g
1
dz
_
g
1
(dz,dz)
. After a shift L along s,
energy gained at p
0
is E = g(F
0
, sL) =
_
m
z
0
sL
_
z. Shifting slowly from p
0
to p
1
yields E =
_
p
1
p
0
m
z
0
dz =
z
0
z
1
z
0
m
Move to horizon, z
1
0 so E m, extracted
the entire energy (mc
2
)!
39
Energy extraction: nite rope strength
If rope has nite max. tension F
max
, then move
only up to z
min
,= 0: z
min
F
max
z
0
m, hence
E
max
_
1
m
F
max
_
m < m.
Optimum choice of mass: m
opt
=
1
2
F
max
, then
extract only 50% of mc
2
Note: the formula for E works only for m
m
opt
because of the near-horizon approxima-
tion g
1
(dz, dz)
2
Note: E is independent of z
0
, but energy is
redshifted, so more absolute energy is gained
if p
0
is further from the horizon
Practice problem: For the metric g = f(r)dt
2
dr
2
f(r)
r
2
_
d
2
+sin
2
d
2
_
, assume f(r
0
) = 0,
f
(r
0
) ,= 0, so that r = r
0
is a horizon. Com-
pute the surface gravity for that horizon.
40
Expansion (divergence) of a vector eld
The rate of change of volume under ow of t:
Volume spanned by t, a, b, c is (t, a, b, c)
Choose a, b, c connecting for t:
t(t, a, b, c) = /
t
(t, a, b, c) = (/
t
)(t, a, b, c)
Due to antisymmetry, must have /
t
= f
where f div t is the divergence or expan-
sion of the vector eld t
A better formula for computing div t:
Choose orthonormal frame e
a
, g(e
a
, e
b
) =
ab
; then
div t =
aa
g(e
a
,
e
a
t)
=
aa
B
(t)
(e
a
, e
a
) Tr
x
B
(t)
(x, x)
Proof: Consider the dual basis
a
and com-
pute /
t
/
t
_
0
1
...
_
=
_
/
t
0
_
1
... +
... Now /
t
0
=
a
e
a
_
/
t
0
_
, but only the
term with
0
survives. So /
t
= (
e
0
/
t
0
+
...). Simplify one term:
_
/
t
0
_
e
0
=
0
[t, e
0
] =
00
g(e
0
, [t, e
0
]) =
00
g(e
0
,
e
0
t); we
used g(e
0
,
t
e
0
) =
1
2
t
g(e
0
, e
0
) = 0.
41
Calculations without indices
From index to index-free notation:
Introduce appropriate vector & tensor objects
Contract each free index with arbitrary vector
Subtract terms coming from nested derivatives
Traces: lower both indices, introduce g
, then
replace g
A
...
x
... by Tr
a
A(a, a, x, ...)
Substitute a a g
1
into A(a a, x, ...)
Tricks to simplify calculations:
assume zero 1st derivatives of dummy vectors
use properties of d, /, , Tr where it helps
Example: u
w
;
= X
Rewrite X
= g
; introduce tensor
X(a, x, y, z) := g(a,
X(x, y)z)
Result: g(u,
u
u
w
u
u
w) = Tr
a
g(a,
X(v, a)v)
Properties of the trace operation:
Linear: Tr
a
x
F =
x
Tr
a
F; same for /
x
,
x
, d
Tr
a
g(a, a) = 4; Tr
a
g(a, x)F(a, y, ...) = F(x, y, ...);
Tr
a
g(a,
a
b) = div b; Tr
a
a
X = X.
Example:
Px = x 2ng(n, x); Tr
P =?
P
= g
; P
= g(x,
Py); substitute
x
b
c
b
a
c
[a,b]
c
Verify that
R(a, b)c is independent of deriva-
tives of a, b, c
Thus:
R is a transformation-valued 2-form.
Covariant tensor: R(a, b, c, d) = g(
R(a, b)c, d)
Properties: exchange a b, c d, ab cd,
Bianchi identities (derive from Jacobi identity)
Example: Show that R(a, b, x, x) = 0, then set
x = c +d and derive c d antisymmetry.
Start with
a
b
b
a [a, b] g(x, x) = 0;
assume
a,b
x = [a, b] = 0; get 0 = [
a
,
b
] g(x, x)
= 2g([
a
,
b
] x, x) = 2R(a, b, x, x).
Ricci tensor: Ric (a, b) = Tr
x
R(a, x, b, x) =
Ric (b, a)
Ricci scalar: R = Tr
x
Ric(x, x)
Einstein equation: Ric
1
2
Rg = 8GT, where
T is energy-momentum tensor of matter,
T =
2
g
1
S
matter
43
Lecture 7, 2007-11-07. FRW spacetime
Metric g = dt
2
a
2
(t)
_
dx
2
+dy
2
+dz
2
_
= dt dt a
2
(t)h; h is the at 3-dim. metric
Need the Einstein tensor, Ric
1
2
Rg
Compute covariant derivatives:
g(
u
t
, v) =
1
2
_
/
t
g
_
(u, v) = aah(u, v), so
t
= 0 and
t
=
a
a
x
=
x
.
g
x
= a
2
dx; so g(
u
x
, v) =
1
2
(dg
x
)(u, v) =
( aadt dx) (u, v); so g(
x
, ) = 0 and
g(
x
,
t
) = aa, so
x
= aa
t
. (
y
,
z
...)
Need
Ricv
R(
t
, v)
t
1
a
2
s=
x
,
y
,
z
R(s, v)s.
For v =
t
need
R(s,
t
)s =
s
t
s
s
s =
a
2
t
( aa)
t
= aa
t
, so
Ric
t
= 3
a
a
t
For v =
x
we have
Ric
x
=
R(
t
,
x
)
t
1
a
2
R(
y
,
x
)
y
1
a
2
R(
z
,
x
)
z
=
1
a
2
(aa+2 a
2
)
x
.
Therefore
Ricv =
_
a
a
+
2 a
2
a
2
_
v+2
_
a
a
a
2
a
2
_
g(v,
t
)
t
Ric(u, v) =
_
a
a
+
2 a
2
a
2
_
g(u, v)+2
_
a
a
a
2
a
2
_
g(u,
t
)g(v,
t
Ricci scalar: R = Tr
v
g(
Ricv, v) = 6
_
a
a
+
a
2
a
2
_
Einstein tensor:
Ric
1
2
Rg =
_
2a
a
+
a
2
a
2
_
g +2
_
a
a
a
2
a
2
_
dt dt
44
Scalar eld cosmology
Lagrangian for scalar eld:
S[] =
_
d
4
x
g
_
1
2
g
1
(d, d) V ()
_
Energy-momentum tensor:
T
2
g
S
g
1
= dd
_
1
2
g
1
(d, d) V ()
_
g
Cosmological solution: = (t); d =
dt;
T =
2
dt dt
_
2
2
V
_
g
Einstein equations for FRW spacetime:
_
2a
a
+
a
2
a
2
_
= 8G
_
2
2
V
_
2
_
a
a
a
2
a
2
_
= 8G
2
Hence: Friedmann equation
_
a
a
_
2
=
8G
3
_
2
2
+V
_
8G
3
(,
)
EOM for :
+3H
+V
() = 0, H
a
a
Expanding universe: a > 0, then H =
_
8G
3
(,
),
then assume
v > 0, so v = v(), get
dv()
d
= f(, v) autonomous dynamics
45
Attractors and the slow-roll approximation
Consider the auxiliary variable u() dened by
d
dt
u()
d
V
d
u()F(). Then we have
=
_
u
F +uF
_
uF; the EOM for becomes
2 +uu
V
+u
2
F
V
= u
_
24G
_
1 +
1
2
u
2
F
2
V
_
If V () satises the slow-roll conditions,
Fu
V
1,
F
u
2
V
1,
V
24G
V,
V
24G
V,
the EOM becomes u u
0
1
6G
= const
the slow-roll approximation is
_
V
_
6G
Note: need to verify u
2
0
F
2
V , u
2
0
F
V .
Generically, there exists a single solution with
u
0
() const at . This u
0
() is ap-
proximated by the slow-roll formula. This is
an attractor under the slow-roll assumptions:
u() = u
0
() +u,
d
d
u = C(u, )u, C > 0
Further approximations: use perturbative ex-
pansion, u = u
0
()+u(); or substitute u
0
()
into the l.h.s. of the equation, u() =
F
(
u,u
,
)
24G
46
Lecture 8, 2007-11-09. Conformal trans-
formation of curvature
How do Ric and R change under g g = e
2
g?
LC connection:
x
y =
x
y +(x, y) where
(x, y) g(w, y)x +g(w, x)y g(x, y)w,
where w g
1
d, g(w, x) x
Ric(a, c) =
Tr
b
g(
R(a, b)c, b) = Tr
b
g(
R(a, b)c, b)
For calculation of
R(a, b)c =
b
c
a
c =
a
(
b
c +(b, c))+(a,
b
c+(b, c)[...b a...]
=
R(a, b)c+
a
(b, c)+(a, (b, c))[...b a...]
Need to express
x
w; note that g(
w, ) =
1
2
(dgw +/
w
g) but dgw = dd = 0; dene a
symmetric bilinear form H
(x, y) g(
x
w, y)
x
d. (In the index notation, this is
;
)
a
(b, c) =
a
(g(w, b)c +g(w, c)b g(b, c)w)
= H
(a, b)c+H
(a, c)bg(b, c)
a
w; antisym-
metrize [b a] and take g(..., b): H
(a, c)g(b, b)
H
(b, b).
Now take trace:
Tr
b
(1st term) = (N 2) H
(a, c) +g(a, c)
47
Conformal transformation of Ricci tensor
2nd term: lets write xy) instead of g(x, y),
then (a, (b, c)) = wa) (b, c) +w(b, c)) a
a(b, c)) w = wa) (wc) b +wb) c bc) w)+
(wb) wc)+wc) wb)ww) bc))a(ab) wc)+
ac) wb)aw) bc))w. Antisymmetrizing [b a],
only terms 5,6,8 remain: (a, (b, c))(b, (a, c))
= wc) (wb) awa) b)+ww) (bc) a+ac) b)+
(ac) wb)+bc) aw))w. Compute g(b, ...) =
wc) (wb) ab)wa) bb))+ww) (bc) ab)+
ac) bb))+(ac) wb)+bc) aw)) wb). Trace:
Tr
b
(2nd term) = (N 2) (ac) ww) aw) cw))
Finally,
Ric = Ric + (N 2) (H
+ ww) g
dd) +() g
Ricci scalar:
R =
Tr
x
Ric(x, x) = e
2
Tr
x
Ric(x, x)
R = e
2
[R +2(N 1) +(N 1) (N 2) ww)]
Note that ww) g(w, w) = g
1
(d, d).
Einstein tensor:
Ric
1
2
Rg = Ric
1
2
Rg+(N 2) [H
dd
_
+
1
2
(N 3) ww)
_
g ]
Einstein tensor does not change in 2D!
A conformal transformation can make
R = 0
in 2D - any 2D space is conformally at.
48
Equations of motion for f(R) gravity
After introducing Lagrange multiplier :
S[g] +S
m
=
_
d
4
x
_
g
_
R +U()
16G
+L
m
_
,
U() f(r) rf
(r)
r=r()
, = 1 +f
(r)
Now conformal transformation: g e
2
g; note:
g = e
4
g and
R = e
2
_
R +6+6[d[
2
_
Note: is with respect to g now!
Choose such that
g equals R
g plus
some terms:
g
_
R+U()
16G
+L
m
_
= e
4
g
_
1
16G
_
e
2
_
R +6+6[d[
2
_
+U()
_
+L
m
_
.
Set = e
2
and omit (total divergence):
S =
_
d
4
x
g
_
R +6[d[
2
+e
4
U(e
2
)
16G
+e
4
L
m
_
Standard action for Einstein gravity (but with
weird matter coupling) Einstein frame, i.e. the
variables g, with 2nd order EOM
The variables g are Jordan frame (have 4th
order EOM)
49
Lecture 9, 2007-11-14. Cosmology with
f(R) gravity
Dene canonical scalar eld:
_
3
4G
s
Assume f(R) =
1
2
R
2
; neglect L
m
; then =
1+R, U() =
(1)
2
2
, so (in Einstein frame)
S =
_
d
4
x
g
_
R
16G
+
1
2
[d[
2
V ()
_
,
V () =
1
32G
(exp[2s] 1)
2
=
_
e
2s
1
_
2
24s
2
.
Trick: Write EOM for through x()
E
kin
E
pot
,
so
d
dt
_
2x()V () (assumed
< 0, > 0!)
dx
d
+
V
V
(1 +x) 6s
_
x(1 +x) = 0
Set V () V
0
e
2ns
(we will have n = 2), then
x
= 2s
x +1
_
3
x n
x +1
_
. Assume x
x
0
at : x
0
=
n
2
9n
2
;
_
2x
0
V (); so
(t)
1
ns
ln
(tt
0
)n
2
_
12
_
9n
2
_
Stable attractor : x = x
0
+x, then x
Cx,
C =
s
n
_
9 n
2
_
> 0 as long as n < 3.
50
Do we have ination?
Note: slow-roll conditions are not valid,
V
V
1
48G
=
2
3
, 1!
Compute
H(t) s
_
2(1 +x
0
) V () = 3n
2
(t t
0
)
1
a(t) = exp
_
H(t)dt = (t t
0
)
3n
2
Conformal transformation:
g = e
2
g =
const
(t t
0
)
2/n
_
dt
2
a
2
(t)dx
2
_
Dene physical time: want g = d
t
2
a
2
(
t)dx
2
;
so dene
t = (t t
0
)
11/n
, then
a(
t) = (t t
0
)
3n
2
n
1
=
t
3n
n(n1)
With n = 2 we have radiation-dominated era,
a t
1/2
Note: ination will be with < 0 and
> 0!
(Slow roll approximation is then valid)
Practice problem: Determine (approximately)
the potential V () for f(R) gravity with f(R) =
R
2
R
2
+R
n
, n > 0, in the regimes of large
R and small R
51
Conditions for viable f(R) theory
If f
f
f
2
+1=exp(2s)
If f
M
2
Pl
16G
d
2
V ()
d
2
=
1
3f
+
Rf
4(R +f)
(1 +f
)
2
1
3f
< 0,
which gives tachyonic instability (small-scale)
At high R we must have f(R) R at observed
scales (M
L 1); f
0
Hence f(R) must be negative, monotonically
growing in R
Practice exercise: Derive the formula for M
2
.
52
Lecture 10, 2007-11-16. Matter-dominated
era
The full action in Einstein frame is
S =
_
d
4
x
g
_
R
16G
+
[d[
2
2
V () +L
m
e
4s
_
where s
_
4G
3
, V =
1
24s
2
_
e
2s
1
_
2
Mock-up Lagrangian for matter:
L
m
(,
) = p
m
,
m
=
L
m,
L
m
,
and we assume p
m
= w
m
m
(nondynamical
matter with w
m
= const) in Jordan frame
Also assume homogeneous = (t), = (t)
Evolution equations for , :
d
dt
_
g
_
=
gV
,
+4s
ge
4s
p
m
d
dt
_
e
4s
L
m,
g
_
=
ge
4s
L
m,
With
g = a
3
(t) we get
L
m,
+
L
m,
+3HL
m,
+4s
L
m,
L
m,
= 0;
d
dt
m
=
_
L
m,
+
L
m,
L
m,
_
53
Finally, using p
m
= w
m
m
, we get
d
dt
m
=
_
3H +4s
m
(1 +w
m
) ;
= 3H
V
,
+4se
4s
w
m
m
Friedmann equation (Einstein frame):
a
2
a
2
H
2
(t) =
8G
3
_
+e
4s
m
_
= 2s
2
_
2
2
+V () +e
4s
m
_
Obtained a closed system for ,
m
, H
How to analyze attractor behavior?
1. Use energy conservation equation
d
dt
= 3H
_
+p
_
+4se
4s
w
m
p
m
2. Use the energy quotient
q(t)
e
4s
m
e
4s
m
+
, 0 q 1
3. Assume that q(t) const and w
const
on the attractor
4. Derive closed equations for q(t), (t)
dq
dt
= 3Hq
_
_
w
m
w
_
(1 q) +
4s
3
w
m
H
_
(Complete analysis of dynamics is in Amendola
et al., gr-qc/0612180 using Jordan frame)
54
Attractor for dust-dominated era (w
m
= 0)
Integrate the equations for
m
,
m
=
(0)
m
a
3(1+w
m
)
e
4s(1+w
m
)
=
(0)
a
3
(
1+w
)
where we assumed p
= w
with w
const
Now assume w
m
= 0, derive equation for q(t):
q = 3Hq (1 q) w
Fixed point if w
0 or q 1 (not if q 0)
If q 1 then need w
(t)
or at most
m
(t) C
0
(t), hence
H = s
2
_
+
m
e
4s
C
1
a
3/2
and then a(t) t
2/3
, H(t)
2
3
t
1
in any case.
So
=
(0)
t
2
(
1+w
)
and
= C
2
t
1w
.
If w
= 0)
Consider w
0, q q
0
,= 1 (the value q
0
is
determined by the EOM for ) then
H
2
=
_
2
3
t
1
_
2
= 2s
2
_
+e
4s
m
_
,
(t) =
_
2(1 q
0
)
3s
lnt, e
s
= t
2
(
1q
0
)
3
t
, > 0
Transforming back to Jordan frame:
g = t
2
_
dt
2
a
2
dx
2
_
d
2
a
2
()dx
2
= t
1+
, a() =
_
2
3
+
_
/(1+)
,=
2/3
To have a dust-dominated era with a()
2/3
we need 1
(This is a constraint on V () after q
0
is deter-
mined)
(See Amendola et al. for precise conditions)
56
Lecture 11, 2007-11-21. Introduction to
perturbation theory for gravity
What is the metric resulting from a small per-
turbation in the matter source?
Compute
Ric(x, y) for g = g +h to 1st order
in h (treating h as a small perturbation)
Change in the LC connection is a tensor,
(x, y)
x
y
x
y = (y, x)
g((x, y), z) = g(
x
y, z)g(
x
y, z)h(
x
y, z) =
1
2
_
d
hy +/
y
h
_
(x, z)h(
x
y, z) =
1
2
(
x
h) (y, z)
1
2
(
z
h) (x, y) +
1
2
_
y
h
_
(x, z).
Since we are computing to 1st order, we may
set g(, ) g(, ) and disregard .
Assuming
x,y
x, y, z = 0, write the Riemann
tensor
R for g through
R for g:
R(x, y)z =
y
z
x
z
R(x, y)z
+
x
(y, z)
y
(x, z)
57
Compute the Ricci tensor: to 1st order,
Ric(x, z) =
Tr
y
R(x, y, z, y) Tr
y
g(
R(x, y)z, y)
Covariant Riemann tensor: to rst order,
x
Tr
y
g((y, z), y)Tr
y
y
g((x, z), y)= Ric(x, z)+
1
2
x
Tr
y
(
y
h(y, z) +
z
h(y, y)
y
h(y, z))
1
2
Tr
y
y
(
x
h(y, z) +
z
h(x, y)
y
h(x, z))=
Ric(x, z)+
1
2
z
Trh+
1
2
h(x, z)
1
2
x
(divh) (z)
1
2
z
(divh) (x), where (divh) (x) Tr
y
(
y
h)(y, x).
Practice problem: Show that div (g) = d
By taking trace of the Einstein equation, nd
R = 8GTrT, hence Ric = 8G
_
T
1
2
g TrT
_
Assume that a background spacetime (Ric, T)
is given and we have a small perturbation T
of the EMT. We need to solve the equation
1
2
z
Trh +
1
2
h(x, z)
1
2
x
(divh) (z)
1
2
z
(divh) (x) = 8G(T
1
2
g TrT)
to rst order in T. Complicated!
58
The 3+1/SVT decomposition
Questions: 1. What is the metric g
gener-
ated by a small T
in at spacetime?
2. What is the metric g generated by a small
T
in FRW spacetime (g = dt
2
a
2
dx
2
)?
To solve the dicult tensor equation, use tricks:
1. Compute time components separately from
spatial components; h =
_
h
00
, h
0j
, h
ij
_
2. Use scalar-vector-tensor (SVT) decompo-
sition for h and T
Helmholz decomposition: = d+, div = 0
(we dene div div (g
1
) for 1-forms)
Determine : div = div d = , then =
1
div v; C = v g
1
dB.
Note:
_
1
f
_
(x) =
_
f( x, y, z)d
3
x
4[xx[
, where x R
3
Remarks on the SVT decomposition:
1. The decomposition is nonlocal in space
2. Only dened when div v 0 suciently
quickly at spatial innity (at least [x[
2
)
3. Dened also without Fourier transform!
4. Useful only because perturbation equations
are algebraic in Fourier space
60
3+1/SVT decomposition of (0,2) tensors
Consider a symmetric tensor T
ij
in 3-dimensional
Euclidean space with metric
ij
Algebraic decomposition: trace + traceless part:
T
ij
= A
ij
+T
(1)
ij
, T
(1)
aa
= 0, A :=
1
3
T
aa
Fourier decomposition: project along tensors
made out of k
Component parallel to k
i
k
j
1
3
k
2
ij
:
T
(1)
ij
=
_
k
i
k
j
1
3
k
2
ij
_
B + T
(2)
ij
, T
(2)
ij
k
i
k
j
= 0,
T
(2)
aa
= 0, B :=
3
2
k
i
k
j
T
(1)
ij
k
4
Component parallel to k
i
& orthogonal to k
j
:
T
(2)
ij
= i
_
k
i
C
j
+k
j
C
i
_
+T
(3)
ij
, T
(3)
ij
k
j
= 0, T
(3)
aa
=
0, C
a
k
a
= 0, C
j
:=
1
i
k
a
T
(2)
aj
k
2
; relabel D
ij
T
(3)
ij
Final decomposition in real space: T
ij
= A
ij
j
1
3
ij
_
B +C
i,j
+C
j,i
+D
ij
Remarks:
1. T
ij
must decay at least as [x[
2
at innity
2. Terms proportional to
ij
can be gathered
because we cannot have B
,ij
= A
ij
(check!)
61
Properties of SVT decomposition
SVT decomposition can be applied to arbitrary
(Euclidean) tensors. It is useful because:
1. Tensor equations split into simpler eqs.
2. Eqs. for S/V/T components decouple
3. Often get algebraic eqs. in Fourier space
Theorem 1: If A
ij
+B
,ij
+C
i,j
+C
j,i
+D
ij
= 0
and A, B, C
i
, D
ij
0 at least as [x[
2
at innity,
then each of A, B, C
i
, D
ij
= 0 everywhere.
(Not true in spaces with Lorentzian signature!)
Theorem 2: EOM for perturbations of a ten-
sor X in the presence of (several) background
tensors will decouple after using a complete
decomposition of X with respect to projections
on the background tensors and SVT. (In our
case, the background tensors are k and
ij
)
Lemma: If a linear function of k is invariant
under 3D rotations, it is equal to zero.
62
Lecture 12, 2007-11-23. Why 4D/SVT
does not work; gauge issues
Try avoiding 3+1 decomposition (only SVT):
h
= Ag
+B
;
+C
;
+C
;
+D
= u
;
+u
;
; decompose
u
= b
,
+c
with div c c
;
= 0; then h
=
2b
;
+c
;
+c
;
. This removes B, C terms!
Compute: Trh = 4A; div h = dA; so Ric
=
A
;
+
1
2
g
A+
1
2
D
= ... Note: D
has 5 d.o.f.
It appears that we have 5 d.o.f. in D
, but
this is wrong: the gauge has not been xed
(have 3 free solutions of u = 0, div u = 0).
But these gauge solutions are not explicit.
3+1/SVT will make the d.o.f. explicit. Also
we will avoid having to solve u = 0 in FRW
background!
63
Perturbations in at spacetime
Use ordinary derivative instead of
Use 3+1/SVT decomposition for h (notation
from V. F. Mukhanov, Physical Foundations
of Cosmology, chapter 7):
_
_
_
h
00
= 2; h
0j
= B
,j
+S
j
; S
j
,j
= 0
h
ij
= 2
ij
+2E
,ij
+F
i,j
+F
j,i
+t
ij
Gauge transformations: h
+h
where
h
= u
,
+ u
,
. Apply 3+1/SVT to u
:
u
_
a; b
,j
+c
j
_
, c
j
,j
= 0. Now compute:
h
00
= 2a
,0
; h
0j
= a
0,j
+b
,0j
+c
j,0
;
h
ij
= 2b
,ij
+c
i,j
+c
j,i
By the SVT theorem 1, we have
= a, B = a +
b, S
j
= c
j
,
= 0, E = b, F
j
= c
j
, t
ij
= 0
Hence we can set E = B = 0, F
j
= 0 (this is
called longitudinal or Newtonian gauge)
Remarks: 1. Scalar parts of u contribute to
scalar parts of h, vector parts of u to vector
parts of h (cf. SVT theorem 2)
2. The source T
+
/
u
T
, but in at space T
= 0. Will have to
take care with FRW background!
64
Einstein equations in almost at space
Compute Ric
=
(Trh)
,
+h
(divh)
(divh)
2
Tr h = 26
(divh)
0
= g
h
0,
= h
00,0
h
0j,j
= 2
(divh)
j
= h
j0,0
h
ji,i
=
S
j
2
,j
Hence (in the SVT decomposition):
Ric
00
= 3
, Ric
0j
= 2
,j
1
2
S
j
Ric
ij
= ()
,ij
+
1
2
t
ij
+
ij
1
2
_
S
j,i
+
S
i,j
_
R = Ric
00
Ric
jj
= 6
2+4
Need to decompose the EMT in 3+1/SVT:
T
0j
=
,j
+
j
;
j
,j
= 0
T
ij
=
ij
+
,ij
+
i,j
+
j,i
+q
ij
j
,j
= 0, q
jj
= 0, q
ij
,j
= 0
65
Explicit solutions
Explicit expressions for 3+1/SVT components:
=
1
T
0i,i
=
1
_
3
2
1
T
ik,ik
1
2
T
ii
_
=
1
2
_
T
ii
T
ik,ik
_
j
=
1
_
T
ji,i
_
1
T
ik,ik
_
,j
_
Einstein equations in 3+1 decomposition are
2 = 8GT
00
2
,j
1
2
S
j
= 8G
_
,j
+
j
_
()
,ij
+
1
2
t
ij
1
2
_
S
j,i
+
S
i,j
_
_
2
+()
_
ij
= 8G(
ij
+
,ij
+
i,j
+
j,i
+q
ij
)
Explicit solution using SVT theorem 1:
=
1
4GT
00
, = 8G,
S
j
=
1
16G
j
, t
ij
=
1
16Gq
ij
Remarks: 1. Other equations are consequences
of T
ij
,j
= 0, do not need to solve them!
2. , , S
j
are constrained and not free d.o.f.
3. A free wave solution can be added to t
ij
(this gives 2 vacuum d.o.f.)
4. For T
= ( +p) u
pg
we have = 0,
= 0, = p,
j
=
j
= 0, q
ij
= 0, T
00
= +p
66
Lecture 13, 2007-11-28. Perturbations of
gravity and matter in FRW spacetime
(see Mukhanov 2005, chapters 7 and 8)
Perturbations due to a single, ideal comoving
uid (if one matter component dominates and
if vector perturbations are absent, e.g. scalar
eld/dust/radiation but no cosmic strings):
T
= ( +p) u
pg
=
_
1 +2 0
0
ij
+2
ij
+t
ij
_
, u
=
_
_
_
_
_
1
0
0
0
_
_
_
_
_
To pass to FRW spacetime, note that any
FRW spacetime is conformally at:
g = d
t
2
a
2
(
t)dx
2
= a
2
(t)
_
dt
2
dx
2
_
Note: t is the conformal time,
t is the proper
time, t =
_
d
t
a(
t)
Example: de Sitter, a(
t) = e
H
t
, t = H
1
e
H
t
,
a(t) = [Ht[
1
Note that < t < 0 for de Sitter; t 0 is
late times
In matter-dominated FRW, a(
t)
t
p
where
usually 0 < p < 1 and so 0 < t <
67
Ricci tensor for perturbed metric in FRW
Ansatz for metric perturbations: g = a
2
g, so
g
= a
2
(t)
_
1 +2 0
0
ij
(1 2) +t
ij
_
Now need to compute the Einstein tensor of g.
Use linear approximation in , t
ij
. Note: a(t)
is xed and has no perturbations!
Use the conformal transformation formula with
g = e
2
g, lna(t):
2dd(2+[w[
2
)g
where w = g
1
d. Need the Hessian: H
(x, y) =
(
.
.
) (x, y) = (
x
d) (y) = g(
x
w, y).
Compute: d = (
t
lna) dt, denote b(t)
t
lna =
t
a; then d = bdt; w (1 2) b
t
;
[w[
2
(1 2) b
2
to rst order in .
Covariant derivative is computed w.r.t. g
and contains perturbations:
H
(x, y) = g(
x
w, y) = b (1 2) g(
x
t
, y)+
g(
t
, y)
x
b (1 2)
g(
.
t
, ) =
1
2
_
dg
t
+/
t
g
_
= d dt +
1
2
g
So H
b; and = Tr
x
H
(x, x) =
4b
+(1 2)
b because Tr
x
h(x, x) = 4
68
Use 3+1/SVT decomposition to compute
H
ij
+
1
2
b
t
ij
; w
0
= (1 2) b; [w[
2
=
(1 2) b
2
use
Ein = Ein+2H
2dd(2+[w[
2
)g
Ein
00
Ein(
t
,
t
) = 23b
2
+6b
Ein
0j
Ein
_
t
,
x
j
_
= 2
_
+b
_
,j
Ein
ij
=
1
2
t
ij
+b
t
ij
_
b
2
+2
b
_
t
ij
+[
_
b
2
+2
b
_
4
_
b
2
+2
b
_
6b
]
ij
Einstein equations:
Ein
(0)
+Ein
(1)
= 8G
_
T
(0)
+T
(1)
_
Background Friedmann equ.
b
2
=
8G
3
T
(0)
00
; b
2
+2
b = 8G
(0)
Perturbations in 00, 0j, ij:
3b
= 4GT
(1)
00
+b = 4G
(1)
t
ij
+2b
t
ij
2
_
b
2
+2
b
_
t
ij
= 4Gq
(1)
ij
In usual scenarios q
(1)
ij
= 0, so only scalar per-
turbations are sourced
69
Lecture 14, 2007-11-30. Perturbations of
gravity and scalar eld in FRW spacetime
Cosmology with scalar eld:
S[g, f] =
_
d
4
x
g
_
R
16G
+
1
2
[d[
2
V ()
_
Perturbations in scalar eld: = f
0
(t)+f(t, x)
Metric perturbations (only scalar):
g = a
2
(t)
_
1 +2 0
0 (1 2)
ij
_
Energy-momentum tensor of :
T
=
,
,
g
p, p
1
2
[d[
2
V ()
p =
1 2
a
2
_
1
2
f
2
0
+
f
0
f
_
V (f
0
) fV
(f
0
)
T
00
=
f
2
0
+2
f
0
f a
2
(1 +2) p
=
_
1
2
f
2
0
+a
2
V
0
_
+
_
f
0
f +a
2
V
0
f +2a
2
V
0
_
T
0j
=
f
0
f
,j
T
ij
= (1 2) a
2
p
ij
=
_
1
2
f
2
0
a
2
V
0
_
ij
+...
Einstein equations in 3+1/SVT:
(background) 3b
2
= 8G
_
1
2
f
2
0
+a
2
V
0
_
3b
= 4G
_
f
0
f +a
2
V
0
f +2a
2
V
0
+b = 4G
f
0
f
70
Equations of motion for the eld = f
0
(t)+f:
gg
,
_
+
gV
() = 0
Compute:
g = a
4
(1 2);
gg
0
,
= a
4
(1 2) a
2
(1 2)
_
f
0
+
f
_
=
a
2
_
f
0
+
f 4
f
0
_
;
gg
j
,
= a
2
(1 2) (1 2) f
,j
= a
2
f
,j
;
V
() = V
0
+fV
0
; so EOM for is a
2
[a
2
(
f
0
+
f 4
f
0
)]
f +a
2
_
(1 2) V
0
+V
0
f
_
= 0.
Background EOM: a
2
_
a
2
f
0
_
+a
2
V
0
= 0
Perturbation EOM:
a
2
_
a
2
f 4a
2
f
0
_
f 2a
2
V
0
+a
2
V
0
f = 0
Collect terms with using background EOM:
a
2
_
a
2
f
_
f
0
f +2a
2
V
0
+a
2
V
0
f = 0
Change variables = a
1
h, f = a
1
x in the
00 and 0j Einstein equations; use background
EOM (3b
2
= ...; a
2
V
0
f = ...) and simplify:
h 3b
h = 4G
_
f
2
0
h +
f
0
x 3b
f
0
x
f
0
x
_
h = 4G
f
0
x
Make Fourier transform in (at) 3-space:
_
f
2
0
k
2
4G
_
h =
f
0
x
f
0
x;
h
4G
=
f
0
x
EOM for perturbations (
f = ...) is a conse-
quence of this 1st order system
The Mukhanov-Sasaki variable
Would like to have a closed 2nd order equation
For quantization, expect x to be fundamental
(occurs as
1
2
x
2
+... in canonical eld action)
Look for a combination v = x+B(t)h such that
v satises an equation of the form v+k
2
v = Cv
Substitute x = v Bh into 2nd Einstein equ.:
h
4G
=
f
0
(v Bh)
Expect relationships of the form
v = A
1
h +A
2
h, k
2
h = A
3
v +A
4
v +0 h
because then we will have
k
2
v = A
1
(A
3
v +A
4
v) +A
2
(A
3
v +A
4
v)
Compute k
2
h from 1st Einstein equ.:
_
f
2
0
k
2
4G
_
h =
f
0
_
v
Bh B
h
_
f
0
(v Bh)
then substitute
h = 4G
f
0
(v Bh) and nd
k
2
4G
h =
f
0
v
_
4GB
f
2
0
+
f
0
_
v
+h
_
f
0
B +4G
f
2
0
B
2
+
f
0
B
f
2
0
_
71
Now we need to cancel the last term:
f
0
B =
f
2
0
_
4GB
2
1
_
+
f
0
B
This is a Riccati equation; convert to linear:
B =
f
0
Q
Q
;
Q = 4G
f
2
0
Q;
A suitable solution is Q = a
1
due to the back-
ground equation 4G
f
2
0
= b
2
b (found by sub-
tracting b
2
+2
b = 8Gp from 3b
2
= ...); so
B =
f
0
b
, v = x +
f
0
b
h; v =
f
0
b
h +
1
4G
f
0
k
2
4G
h =
f
0
v
_
4G
b
f
3
0
+
f
0
_
v
Equation for v: dene z a
f
0
b
1
, then
v =
z
1
4G
_
zh
f
0
_
,
k
2
h
4G
=
f
0
z
_
v
z
_
k
2
v =
z
1
4G
_
k
2
h
z
f
0
_
= z
1
_
f
0
z
2
f
0
_
v
z
_
= v
z
z
v
Expressing x through v:
x = v
f
0
b
h; h = 4G
f
0
z
1
_
v
z
_
t
f
0
const but
actually slightly growing towards the end of
ination. (Figure from Mukhanov 2005)
/H
f
t t
c
t
f
f
0
1
a
Note: describing the evolution of perturba-
tions in non- dominated epochs requires a
dierent calculation! (We used the fact that
the background is -dominated.)
72
Lecture 15, 2007-12-07. Quantization of
free elds in FRW spacetime
(See also my lecture notes of 2006)
Consider a scalar eld in a xed FRW back-
ground spacetime:
S =
_
gd
4
x
_
1
2
g
)
_
1
2
m
2
2
_
Use conformal time: g = a
2
(t)
_
dt
2
dx
2
_
S =
1
2
_
d
3
xdt a
4
_
a
2
_
2
(
j
)
2
_
m
2
2
_
Change variables: a(t), then
a
2
2
=
2
2
a
a
+
2
_
a
a
_
2
=
2
+
2
a
a
2
a
a
_
.
So the action is
S =
1
2
_
d
3
xdt
_
2
(
j
)
2
m
2
e
(t)
2
_
m
2
e
(t) m
2
a
2
(t)
a(t)
a(t)
Scalar eld (t, x) in FRW spacetime is equiva-
lent to scalar eld (t, x) with a time-dependent
mass m
e
(t) in at Minkowski spacetime
Note: the action for Mukhanov-Sasaki variable
v(t, x) is of this form with m = 0 and using z(t)
instead of a(t)
73
Quantization of free eld with m(t)
Mode expansion: Fourier transform in space
(not in time)
(x, t) =
_
d
3
k
(2)
3/2
k
(t)e
ikx
, (
k
)
=
k
Then a term such as
_
d
3
x
_
_
2
becomes
_
d
3
x
_
_
2
=
_
d
3
x
d
3
kd
3
k
(2)
3
_
e
i
(
k+k
)
x
i
2
k k
_
=
_
d
3
kk
2
k
=
_
d
3
k[
k
[
2
k
2
So the action for
k
(t) becomes
S[
k
(t)] =
_
d
3
kdt
_
1
2
[
k
[
2
1
2
[
k
[
2
_
k
2
+m
2
e
(t)
_
_
Every Re
k
, Im
k
is a separate harmonic os-
cillator with time-dependent frequency
(t)
_
k
2
+m
2
e
(t)
It is more convenient to consider complex-valued
variables
k
Equation of motion:
k
+
2
k
(t)
k
= 0
74
Quantization of time-dependent oscillators
Action for a unit-mass oscillator:
S =
_
dt
_
1
2
q
2
1
2
2
(t)q
2
_
Equation of motion: q +
2
(t)q = 0
General solution (complex-valued) is in 2-dim.
space of solutions spanned by v(t), v
(t):
q(t) = a
+
v(t) +a
(t); v +
2
(t)v = 0
Quantization replaces q(t) q(t) and a
, a
+
v +a
]
= [a
, a
+
](v
v v v
) = i
A mode function is a solution v(t) such that
1 =
v
v v v
i
2Im
_
v
v
_
Note: v(t) must be complex-valued; have one-
parametric family of admissible mode functions
(can set v(t
0
)/v(t
0
) = arbitrarily if Im > 0)
Note: v(t) ,= 0 for all t because Im(v
v) is a
Wronskian and is constant in time
For a given v(t) we have a
=
1
i
[ v(t) q(t) v(t) p(t)],
so the denition of a
depends on v(t)
75
Fock space: a
[0) = 0; [n) =
1
n!
_
a
+
_
n
[0),
depends on the choice of v(t)
Hamiltonian
H =
1
2
p
2
+
1
2
2
q
2
is expressed as
H(t) =
[ v[
2
+
2
[v[
2
2
_
2a
+
a
+1
_
+
v
2
+
2
v
2
2
a
+
a
+
+
v
2
+
2
v
2
2
a
[exercise]
Note: in time-independent cases, (t) =
0
,
one chooses v(t) =
1
2
e
i
0
t
, so
H is diagonal in
the Fock basis:
H[n) = E
n
[n).
In general, [n) are not eigenstates of
H(t), ex-
cept possibly at one t = t
0
if we choose v(t)
such that v(t
0
) = i(t
0
)v(t
0
). The state [0) is
then the instantaneous vacuum at t
0
.
Statement 1: It is impossible to chose v(t)
such that
H(t) is diagonal at all times, i.e. v
2
+
2
v
2
= 0 for all t.
Proof: v = e
i
_
dt
does not satisfy EOM.
Statement 2: The instantaneous vacuum [
t
0)
minimizes 0[
H(t) [0) over all possible [0).
Proof: Consider [
0[
H(t
0
) [
0) =
[ v
0
[
2
+
2
0
[v
0
[
2
2
=
[
2
[+
2
0
2(
)
. If = A +iB, then the
minimum is at A = 0, B =
0
. So v
0
= i
0
v
0
.
76
Lecture 16, 2007-12-12. Quantized free
scalar eld in FRW
Decompose the Fourier modes as
k
() = a
k
v
k
(t) +a
+
k
v
k
(t), a
+
k
=
_
a
k
_
k
v
k
(t) +a
+
k
v
k
(t)
_
e
ikx
=
_
d
3
k
(2)
3/2
_
a
k
v
k
(t)e
ikx
+a
+
k
v
k
(t)e
ikx
_
The canonical commutation relation is
[ (x
1
, t),
(x
2
, t)] = i(x
1
x
2
)
Choose v
k
(t) for every k (same for equal [k[)
a
k
=
v
k
k
v
k
k
i
; [a
k
, a
+
k
] =
_
k k
_
Once v
k
(t) are chosen, dene vacuum: a
k
[0) =
0, k. Excited states are interpreted as parti-
cles because
k
n[
H[
k
n) = 0[
H[0) +n
k
Note: no instantaneous minimum energy pos-
sible if
2
k
< 0. If so, can have e.g.
k
n[
H[
k
n) <
0[
H[0). No particle interpretation possible!
77
Bogoliubov transformations
Consider an oscillator with in-out transition:
PSfrag replacements
(t)
1
t
1
t
2
t
Natural denitions of minimum-energy vacuum:
v
1
(t)[
t<t
1
=
1
2
e
i
1
t
, v
2
(t)[
t>t
2
=
1
2
e
i
2
t
Since
_
v
1
, v
1
_
is a basis, we must have
v
2
(t) = v
1
(t) +v
1
(t)
(Bogoliubov transformation/coecients)
Condition Im
_
v
2
v
2
_
=
1
2
gives [[
2
[[
2
= 1
Express a
2
through a
1
, get a
2
= a
1
a
+
1
The state
t
1
0
_
is not vacuum at t = t
2
but
has particles. Compute mean particle number:
_
t
1
0
a
+
2
a
t
1
0
_
= [[
2
Numerical computation of requires knowing
v
1
(t), v
1
(t), v
2
(t), v
2
(t) at any one moment t
0
:
=
1
i
[ v
1
v
2
v
2
v
1
]
t=t
0
78
Number density for quantum particles
Bogolyubov transformation: u
k
(t) =
k
v
k
(t) +
k
v
k
(t), and [
k
[
2
[
k
[
2
= 1
Fourier expansions must agree:
e
ikx
_
v
k
(t)a
+
k
+v
k
(t)a
k
_
= e
ikx
_
u
k
(t)
b
+
k
+u
k
(t)
k
_
Hence
b
k
=
k
a
k
k
a
+
k
,
b
+
k
=
k
a
+
k
k
a
k
Mean k-particle number in the a
k
-vacuum state:
a
0[
b
+
k
k
[
a
0) =
a
0[ (
k
a
+
k
k
a
k
)(
k
a
k
a
+
k
) [
a
0)
= [
k
[
2
a
0[ a
k
a
+
k
[
a
0) = [
k
[
2
(3D)
(0). The
divergent factor
(3D)
(0) represents 3-volume
of the space. Therefore the number density of
k-particles is n
k
= [
k
[
2
.
Need to calculate the coecients
k
:
k
=
u
k
v
k
u
k
v
k
i
Need to know the mode functions u
k
(t
0
), v
k
(t
0
)
at any one time t
0
79
Particle production by gravity in FRW space-
time
Toy model: a(t) = const for t < t
1
and t > t
2
,
nonstationary regime for t [t
1
, t
2
]
Mode functions are solutions of
v
k
+
_
k
2
+m
2
a
2
a
a
_
v
k
v
k
+
2
k
(t)v
k
= 0
Natural denitions of vacuum for t < t
1
and
t > t
2
are the positive-frequency solutions:
v
k
=
1
_
2
(1)
k
e
i
(1)
k
t
for t < t
1
u
k
=
1
_
2
(2)
k
e
i
(2)
k
t
for t > t
2
(1,2)
k
k
(t
1,2
) =
_
k
2
+m
2
a
2
(t
1,2
) = const
Initial state is vacuum choose v
k
as mode
functions nal state has particle density
[
k
[
2
= [u
k
v
k
u
k
v
k
[
2
=
1
2
(2)
k
v
k
(t
2
) v
k
(t
2
)i
(2)
k
2
Need to compute v
k
(t
2
). If we use WKB,
v
k
(t
2
)
1
_
2
k
(t
2
)
exp
_
i
_
t
2
t
1
k
(t)dt
_
then [
k
[
2
0, so this precision is insucient!
80
Improved WKB approximation: instanta-
neous squeezed state
Consider Bogolyubov coecients
k
(t),
k
(t)
relating instantaneous vacua at t and at t
1
:
k
(t) =
v
k
i
k
v
k
i
k
,
k
(t) =
v
k
i
k
v
k
i
k
Dene instantaneous squeezing parameter
Z(t)
k
(t)
k
(t)
=
v
k
i
k
v
k
v
k
+i
k
v
k
Since [
k
[
2
=
[Z[
2
1[Z[
2
1, we expect Z(t) 1
Equation for Z(t) is
Z +2i
k
Z =
k
2
k
_
1 Z
2
_
, Z(t
1
) = 0
First approximation (neglect Z
2
):
Z(t)
_
t
t
1
k
(t
)
2
k
(t
)
exp
_
2i
_
t
t
k
(t
)dt
_
dt
k
(t
)dt
_
k
4i
2
k
1
2i
k
d
dt
_
k
4i
2
k
_
+...
_
81
This is an adiabatic series (the small param-
eter is the slowness of
k
, i.e.
k
2
k
etc.)
Note: this series is divergent! At t > t
2
this
series gives Z(t
2
) 0 because Z(t
2
) is smaller
than the precision of this series.
Asymptotic estimate of [
k
[
2
[Z(t
2
)[
2
Note: for high-energy particles (k
2
m
2
a
2
,
k
a
a
k
(t)dt, then
Z() e
2i
_
k
2
2
k
e
2i
If t = t
is a zero of
k
(t) of n-th order, then
k
(t)
(0)
k
(t t
)
n
, so we have near t t
that
+
(0)
k
n+1
(t t
)
n+1
. The function
k
2
2
k
n
2
(0)
k
(tt
)
n+1
=
n
2(n+1)
1
.For Im
> 0,
we get a contribution
n
2(n+1)
e
2Im
k
t kt, so [
k
[
2
e
4kt
is exponentially
small at high energy (ultrarelativistic particles)
Lecture 17, 2007-12-14. De Sitter space,
Bunch-Davies vacuum
De Sitter spacetime: a(t) = (Ht)
1
, so the
eective frequency is
k
(t) = k
2
+
_
m
2
H
2
2
_
t
2
Consider m H (relevant in cosmology); then
v
k
+
_
k
2
2
t
2
_
v
k
= 0
has the general solution
v
k
= A
k
_
1 +
i
kt
_
e
ikt
+B
k
_
1
i
kt
_
e
ikt
Normalization yields [A
k
[
2
[B
k
[
2
=
1
2k
How to choose A
k
, B
k
?
Consider instantaneous vacuum at early time
[t[ k
1
, then v
k
(t)
1
2k
e
ikt
is a natural
choice. Interpretation: short-wave modes do
not feel the curvature, behave as in at space.
Bunch-Davies vacuum state: for every k, choose
v
k
/v
k
i
k
at t (Minkowski vacuum at
early times). This yields A
k
=
1
2k
, B
k
= 0
82
Amplitude of quantum uctuations
Fluctuations averaged on spatial scales L:
L
(t)
_
d
3
x (t, x)W
L
(x),
where W
L
(x) is the window function such that
W
L
(x) 1 for [x[ L and
_
d
3
xW
L
(x) = 1,
for example W
L
(x) =
1
(2)
3/2
L
3
exp
_
[x[
2
2L
2
_
The quantum expectation value of
L
L
is
0[
L
(t)
L
(t) [0) =
_
d
3
k
(2)
3
[v
k
(t)[
2
e
[k[
2
L
2
_
L
1
0
k
2
dk
2
2
[v
k
(t)[
2
1
2
2
[v
k
(t)[
2
k
3
k=L
1
This is called the power spectrum
2
L
(k) of
uctuations in on scales L k
1
Power spectrum of free eld in at space:
v
k
=
e
i
k
2
+m
2
t
2(k
2
+m
2
)
1/4
,
2
L
(k) =
1
2
2
k
3
_
k
2
+m
2
Power spectrum in de Sitter space (BD):
L
(t) =
k
3/2
[v
k
(t)[
a(t)
2
=
H
2
_
t
2
k
2
+1
At late times (t 0),
L
H
2
= const!
83
Primordial uctuations from ination
Plan: 1. Quantize the MS variable v. 2. De-
termine vacuum state [0). 3. Determine power
spectrum of at the end of ination.
v
k
+
_
k
2
z
z
_
v
k
= 0, z(t)
a
f
0
b
BD vacuum state for v: mode functions
v
k
(t) =
1
2k
e
ikt
, t
To determine time t
c
(k) of horizon crossing,
need to have approximate expression for z(t)
Use slow roll approximation, conformal time:
b
2
=
a
2
a
=
8G
3
_
1
2
f
2
0
+a
2
V (f
0
)
_
f
0
+2b
f
0
+a
2
V
(f
0
) = 0
Slow-roll parameters: (of order 10
2
)
=
1
16G
V
2
V
2
, =
1
8G
_
V
V
1
2
V
2
V
2
_
Slow-roll solution: (note conformal time)
f
0
_
V
_
6G
a = a
2
3
V (f
0
)
z(t)
a
4G
;
z
z
2b
2
16G
3
a
2
V (f
0
)
84
Approximately treat as a constant for now,
and estimate a(t) t
1
, z(t) t
1
, z/z 2t
2
Horizon crossing: kt
c
(k)
2
Approximate v
k
(t) after horizon crossing:
v
k
(t) A
k
z(t)
_
1 +k
2
C(t)
_
, A
k
unknown
Find leading correction of order k
2
:
1
C +2
z
z
C = z
2
_
Cz
2
_
.
C(t) =
_
t
z
2
(t
)
_
t
z
2
(t
)dt
1
2
t
2
+O(t
3
)
v
k
(t) A
k
z(t)
_
1
1
2
k
2
t
2
_
, kt 1
To determine A
k
, match at horizon crossing:
A
k
z(t
c
(k))
1
2k
e
ikt
c
(k)
, A
k
1
2kz(t
c
(k))
Substitute v
k
into the formula for :
=
1
a
4G
f
0
1
z
_
v
z
_
.
k
(t) A
k
a(t)t(t)
8G
3
V (f
0
(t))
4G(t)
3
_
k(t
c
(k))
a(t)t
a(t
c
(k))
_
V (f
0
(t))
4G(t)
3
_
k(t
c
(k))
2
k
_
V (f
0
(t))
85
Power spectrum
k
(t) =
k
3/2
2
[
k
(t)[ =
4G(t)
3
_
(t
c
(k))
_
V (f
0
(t))
Note: at the end of ination (t) 1, so
k
1/2
(t
c
(k))
Towards the end of ination grows, so the
spectrum is red tilted
86
Lecture 18, 2007-12-19. Null surfaces,
horizons
A smooth hypersurface (a submanifold of
codimension 1) S /is a set of points p : f(p) = 0
where f is a smooth function such that df ,= 0
on S.
Normal vector n g
1
df ; can redene f
f with ,= 0 on S, then n n.
Tangent vectors: t f = 0, g(n, t) = 0
Note: 1. A null vector cannot be orthogonal to
a timelike vector. 2. If g(n, n) = g(l, l) = 0 and
g(n, l) = 0 then n = l. 3. A timelike vector
cannot be orthogonal to a timelike vector.
Hypersurface is timelike / spacelike / null
i the normal vector n is everywhere space-
like / timelike / null. A spacelike hypersurface
has only spacelike tangent vectors. A timelike
hypersurface has timelike, null, and spacelike
tangent vectors. A null hypersurface has 1
null and 2 spacelike basis tangent vectors (no
timelike!). For a null hypersurface, the normal
vector lies within the hypersurface!
Let us call hypersurfaces just simply surfaces.
87
Examples of null surfaces
Example 1. In Minkowski spacetime, consider
f(p) = 0, f(p) t
_
x
2
+y
2
+z
2
t r
then the normal vector is null,
n g
1
df =
t
+
1
r
(x
x
+y
y
+z
z
)
This surface is a lightcone focusing through
the origin.
Example 2. In Schwarzschild spacetime,
g =
_
1
2m
r
_
dt
2
_
1
2m
r
_
1
dr
2
r
2
d
2
,
the surface r = 2m (the horizon 1) is a null
surface.
Proof: Tangent space to r = 2m is spanned
by
,
t
. The vector
t
is null at r = 2m
(but not at r ,= 2m). Also,
t
is normal to
and
. Hence,
t
is normal to T1 everywhere
on 1. Hence, 1 is a null surface.
Denition: a null function f(p) is such that
n g
1
df is null everywhere.
88
Geodesic generators of null surfaces
Null surfaces are made of geodesic lines called
generators. These are the ow lines of the
normal vector n g
1
df. Note: n is the
ane normal to the surface. (Other normals,
n
.)
90
Causal boundaries are null surfaces
Statement: The boundary 7
+
(T) of the fu-
ture inuence domain of any set T is a null
surface (piecewise, and away from T).
Proof: Consider a point p 7
+
(T) where
the normal vector n exists and is not null, and
also p , T. Go to a local Lorentz frame where
7
+
(T) is locally a plane orthogonal to n.
Condition B holds: Points innitesimally
close to 7
+
(T) on one side are in 7
+
(T) but
point innitesimally close to 7
+
(T) on the
other side are not in 7
+
(T).
If n is timelike, then a curve going along n is
causal and violates condition B.
If n is spacelike, then there exists a timelike
vector t within 7
+
(T). Then t + n is time-
like for very small ; a causal curve along t+n
crosses 7
+
(T) and violates condition B.
Note: statement is not true if p T or if n is
undened (a corner of a surface)
Denition: Event horizon 1 is the boundary
of the past inuence domain of the asymptotic
far future (can be timelike or null future)
91
Stationary (Killing) horizons
A null surface 1 is a stationary (Killing) hori-
zon for a Killing vector eld k if k is normal to
1 : g(k, x) = 0 for every tangent vector x to
the surface. (Note: k must be hypersurface-
orthogonal to admit a horizon)
Properties:
1. Killing vector k is null on 1;
k
k = k
2. Surface gravity is constant along ow
lines of k
3. (without proof) The set of points where
k = 0 is a 2-sphere B (bifurcation 2-surface)
4. Surface gravity is constant on B
5. Surface gravity changes sign across B
Derivations:
1. Note: k is proportional to the ane nor-
mal n to 1, so k is itself null on 1. If t is
tangent to 1, we must have 0 =
1
2
t
g(k, k) =
g(
t
k, k) = g(
k
k, t) for every t such that
g(k, t) = 0, hence
k
k = k on 1. Recall
that z
_
g(k, k), g
1
dz =
k
k
z
. Assume
g(k, k) > 0 in some domain bounded by 1.
Surface gravity (dened as force at innity) is
[[ =
g
1
dz
1
= lim
p1
[
k
k[
z
= [[.
Let us dene (not only [[!) by the formula
k
k k on 1. (Signed surface gravity.)
92
Lecture 19, 2007-12-21. Zero-th law of
black hole thermodynamics
2. Need formula:
x
y
k
x
y
k =
R(x, k)y
(derivation postponed), then
k
k
k =
k
k
and so 0 =
k
(
k
k k) = k
k
. Hence
= const along k (unless k = 0)
Similarly,
k
_
2
_
= lim
p1
k
g(
k
k,
k
k)
g(k,k)
= 0
3. Assume ,= 0, introduce an ane null
normal n to 1, write k = n. By denition,
n
n = 0, hence
k
k = n
n
= n, so
n
= = const along a ow line (orbit) of
n, as long as ,= 0. Hence there is a point on
the orbit where changes sign. At that point,
k = 0, changes direction (bifurcation point).
Dene B as the set of points where = 0.
4. Derive another formula for . Since (by
the Frobenius theorem) gk dgk = 0, a useful
identity is found from a double trace with dgk:
0 = Tr
x,y
y
(gk dgk)
x
y
(dgk) = ...
But
x
k
dgk = 2g(
k
k, x), so
k
dgk = 2gk.
Also
1
4
Tr
x,y
dgk(x, y) dgk(x, y) = Tr
x
g(
x
k,
x
k) =
Tr
x
(
x
g(k,
x
k)g(k,
x
x
k)) =
1
2
z
2
Ric(k, k).
93
Now
y
x
(gk dgk) = g(k, x)
y
dgkg(k, y)
x
dgk+
gkdgk(x, y), then we compute 0 =
1
4
Tr
x,y
x
(gk dgk)
y
x
dgk = gk([
k[
2
+ 2
2
), where
we denoted [
k[
2
Tr
x
g(
x
k,
x
k). Hence
2
=
1
2
Tr
x
g(
x
k,
x
k) =
1
4
z
2
1
2
Ric(k, k)
5. For any vector t tangent to B, we have
x
k =
t
x
k +
R(t, k)x, hence
t
2
=
Tr
x
g(
t
x
k,
x
k) = Tr
x
R(t, k, x,
x
k) = 0 since
k = 0 on B.
Derivation of the formula for
a
b
k
g(
a
b
k, c) = g(
a
b
k, c) +R(b, c, a, k)
Assume that all covariant derivatives of vec-
tors a, b, c vanish. We are trying to reverse the
order of vectors a, b by using the Killing prop-
erty g(
a
k, b) = g(
b
k, a) and the denition
of R(). Write g(
a
b
k, c) =
a
g(
b
k, c) =
a
g(
c
k, b) = g(
a
c
k, b) since
a
b, c =
0. Replace g(
a
c
k, b) = g(
c
a
k, b)+R(a, c, k, b)
and so g(
a
b
k, c) = R(c, a, k, b)g(
c
a
k, b).
Now repeat this for the permutations (abkc)
(cakb) (bcka) and nd g(
a
b
k, c) = R(c, a, k, b)
R(b, c, k, a) + R(a, b, k, c) g(
a
b
k, c). Now
use the rst Bianchi identity to express R(b, c, a, k) =
R(a, b, c, k) R(c, a, b, k). Hence we get
2g(
a
b
k, c) = 2R(b, c, a, k)
Vector elds orthogonal to a single 3-surface
Statement: If a vector eld v is either geodesic
or Killing, and is orthogonal to a single 3-
surface
(either spacelike or timelike but not
null) then it is everywhere hypersurface-orthogonal.
Proof: We have
v
g(v, v) = /
v
g(v, v) = 0 in
either case. Hence, g(v, v) is constant along
orbits of v. We may rescale v u = v
such that g(u, u) = const everywhere. (If v
is Killing, we will not need to rescale; see be-
low.) Now it is sucient to show that u is
hypersurface-orthogonal everywhere.
Construct a family of surfaces () by trans-
porting along u, where is the ane pa-
rameter, u = 1. Now prove that tangent
vectors to () remain orthogonal to u.
Choose a connecting vector c T
p
such that
[c, u] = 0 everywhere; by construction g(c, u)
=
0. It remains to prove that /
u
g(c, u) = 0:
/
u
g(c, u) = (/
u
g) (c, u) +g(/
u
c, u) +g(c, /
u
u)
= (/
u
g) (c, u) = g(
c
u, u) +g(
u
u, c).
This vanishes either when /
u
g = 0 (Killing
case, set 1) or when g(u, u) = const and
u
u = 0 (geodesic case).
94
Lecture 20, 2008-01-09. Near-horizon ge-
ometry for Killing horizons
Assumptions: a metric with hypersurface-orthogonal
Killing vector, and a Killing horizon 1 with
surface gravity =
g
1
dz
,= 0, where z
_
g(k, k) is the redshift function (dened only
at that side of the horizon where k is timelike)
Want to investigate near-horizon properties.
Use t as time (k =
t
), due to hypersurface-
orthogonality have g = z
2
dt
2
(spatial metric).
Use z as a spatial coordinate; introduce spatial
coordinates
_
y
1
, y
2
_
within surfaces z = const
(at least locally b/c dz ,= 0 on 1)
Metric is written as
g = z
2
dt
2
g(
z
,
z
)dz
2
A,B
y
A
y
B
AB
(z, y
1
, y
2
)
To compute g(
z
,
z
), consider the inverse met-
ric; the coecient of g
1
at
z
z
is g
1
(dz, dz) =
g
1
dz
2
2
f(z, y
A
) where f = 1 on 1. Hence
g = z
2
dt
2
f
1
2
dz
2
(...)
Note: g is degenerate on 1 (det g = 0 on 1).
95
Local Rindler frame
Coordinate z is undened beyond the horizon.
Trick: Dene z
2
and consider < 0
g = dt
2
d
2
4
2
f()
(...)
Near horizon, f 1, introduce coordinates
T, X via T
_
[[ sinht, X
_
[[ cosht
Then g dT
2
dX
2
(...). (2D at spacetime2D
space)
Rindler spacetime:
T
X
z = 0
z =const
> 0
> 0
< 0
< 0
z < 0
96
Removing coordinate singularity
Metric g = dt
2
_
4
2
f()
_
1
d
2
, Killing
vector k =
t
Singularity det g = 0 at = 0 is not physical
(det g is a coordinate-dependent quantity!)
Introduce (Painlev) infalling coordinates:
t + A(), where A() is chosen so that the
metric is nonsingular on 1 in coords.
_
, , y
A
_
:
g =
_
d A
d
_
2
= d
2
2A
dd
_
1
4
2
f
A
2
_
d
2
(...)
Infalling means that
points inwards: g(
) >
0; this holds if A
() > 0.
Can choose A() such that A
,= 0 and that
1
4
2
f
A
2
is nite as 0. For example, can
have
1
4
2
f
A
2
1 on 1. (Need A
()
1
2
near 1 in any case.)
Examples: spherically symmetric spacetimes
g = Q(r)dt
2
dr
2
Q(r)
r
2
_
d
2
+sin
2
d
2
_
Zeros of Q(r) are horizons. Reissner-Nordstrm:
Q(r) = 1
2m
r
+
q
2
r
2
; Schwarzschild-de Sitter:
Q(r) = 1
2m
r
H
2
r
2
97
Computing surface gravity
Exercise: verify
k
k = k directly in Painlev
coordinates
_
, , y
B
_
using the formula g(
k
k, x) =
1
2
(dgk) (k, x).
Note: surface gravity depends on the normal-
ization of the Killing vector k! (This needs to
be xed ad hoc, say g(k, k) 1 at innity)
For metric of the form
g = P(x)dt
2
dx
2
Q(x)
(x, y
B
)
where P(x
0
) = Q(x
0
) = 0, while P
(x
0
) > 0,
Q
(x
0
) > 0. (Assuming k =
t
in the given
coordinate system!)
Need to transform the metric to g = dt
2
_
4
2
f()
_
1
d
2
where = 0 at x = x
0
and
f(0) = 1: change coordinates to = P(x),
then
Q(x)
Q
(x
0
)
P
(x
0
)
, g = dt
2
(x
0
)
Q
(x
0
)
d
2
P
2
(x
0
)
and hence =
1
2
_
P
(x
0
)Q
(x
0
).
98
Unruh eect
Rindler spacetime (2D): g = (1 +ax)
2
dt
2
dx
2
; Killing vector k =
t
; Killing horizon x =
a
1
; redshift z = 1 +ax (for x > a
1
only!)
Proper acceleration of stationary observers:
a = g
1
dlnz =
a
1 +ax
x
Observer at x = 0 has constant acceleration a
Interpretation: horizon plane x = a
1
due to
constant acceleration. Surface gravity = a
To be shown: quantum eld in Minkowski vac-
uum appears to have thermal spectrum of par-
ticles, as measured by accelerated detectors
Unruh temperature: T =
a
2
in Planck units
(Analytic continuation arguments: 1. Euclidean
metric a
2
x
2
dt
2
E
+ dx
2
is nonsingular at origin
only if t
E
is periodic with period 2/a. 2.
Fields are functions of sinat
E
and must be pe-
riodic in t
E
. Inconclusive!)
99
Lecture 21, 2008-01-11. Hawking radia-
tion. First law for Killing horizons
Metric g = dt
2
_
4
2
f()
_
1
d
2
2
f()
t
2
=
2
, t =
_
d
2
_
f()
Statement: Any null curve in 2D spacetime
is a null geodesic (lightray).
Proof: g(
n
n, n) = 0 but there is only one
vector orthogornal to n, so
n
n = n and
hence
n
(n) = 0 for some .
Dene the tortoise coordinate r
_
d
2
f()
,
then lightrays are t r = const; 2D metric is
conformally at, g =
_
dt
2
dr
2
_
.
Consider scalar eld (any coupling/potential):
S [g, ] =
_
dt d d
2
y
f
_
1
2
2
2
2
f
_
,
_
2
1
2
[
[
2
V (, R)
_
=
_
d
2
y
_
dt dr
_
1
2
1
2
2
,r
+(...)
_
Coupling/potential is unimportant at 0!
Essentially, free eld in 2D Rindler spacetime.
100
Unruh eect
Quantization of free eld in Rindler spacetime:
use two coordinate systems, t, r and T, X
Tortoise coordinate: r = a
1
ln(1 +ax)
Minkowski frame: aT = e
ar
sinhat, aX = e
ar
coshat
g = dT
2
dX
2
= e
2ar
_
dt
2
dr
2
_
Mode functions: u
k
=
1
2
k
e
i
k
t
,
k
= [k[
Mode expansions for quantization:
(T, X) =
_
dk
2
1
_
2[k[
_
e
i[k[T+ikX
a
k
+e
i[k[TikX
a
+
k
_
(t, r) =
_
dk
2
1
_
2[k[
_
e
i[k[t+ikr
k
+e
i[k[tikr
b
+
k
_
Use lightcone coordinates: u = tr, v = t+r,
U = T X, V = T + X and rewrite mode
expansions using
k
as integration variable:
(U, V ) =
_
0
d
2
1
2
_
e
iU
a
+e
iU
a
+
+e
iV
a
+e
iV
a
+
(u, v) =
_
0
d
2
1
2
_
e
iu
+e
iu
b
+
+e
iv
+e
iv
b
+
_
Note: both expansions are for the same oper-
ator
, so we can express
b
through a
101
Bogoliubov transformation
Relationship between u, v and U, V :
aU = e
au
, aV = e
av
Hence the u-dependent part of
is equal to
the U-dependent part: (same for v and V )
_
0
d
2
1
2
_
e
iu
+e
iu
b
+
_
=
_
0
d
2
1
2
_
e
iU
a
+e
iU
a
+
_
Compute Fourier transform in u of both sides:
_
+
du
2
e
iu
_
0
d
2
1
_
e
i
+e
i
b
+
_
=
1
_
2[[
_
_
_
, > 0
b
+
[[
, < 0
_
+
du
2
e
iu
_
0
d
2
1
2
_
e
iU
a
+e
iU
a
+
_
=
_
+
du
2
_
0
d
2
_
e
iuiU
a
+e
iu+iU
a
+
_
0
d
2
_
F(, )a
+F(, )a
+
_
,
where we dened
F(, )
_
+
du
2
exp
_
iu +i
a
e
au
_
102
Then the Bogoliubov transformation is
=
_
+
0
d
_
a
+
_
,
where (for > 0, > 0)
=
_
F(, ),
=
_
F(, )
Need to compute
0[
b
+
0) =
_
+
0
d[
[
2
where [
a
, F
(, ) = F(, )
Derivation: Shift the integration contour into
the line u = ia
1
+t with t R, then e
au
=
e
at
and a factor e
/a
appears.
Trick 2: Canonical commutation condition,
_
b
+
_
= (
), entails
(
) =
_
+
d
_
_
=
_
+
F(, )F
(,
) exp
a
_
+
_
The mean particle density is computed by set-
ting =
) =
_
exp
_
2
a
_
+1
_
1
=
_
exp
_
E(k)
T
_
+1
_
1
First law for Killing horizons: E = TS
where E is the energy ow through 1, T =
2
is the temperature, S =
1
4G
A is the entropy of
the horizon
Consider an almost Killing horizon (
k
k k
on 1) and a small amount of matter (T
)
that falls into the horizon very slowly.
Area of the horizon is the 2-area of the space-
like cross-section of 1 at t = const
Consider spacelike connecting vectors s
1
, s
2
for
k; they remain approximately orthogonal to k
since /
k
g(k, s) = (/
k
g) (k, s) 0
Introduce a null vector l such that g(k, l) = 1,
then
k
l l near 1
The rate of change of 2-area under ow of k:
2-area A = 4-volume spanned by k, l, s
1
, s
2
,
so A = (k, l, s
1
, s
2
) and
/
k
(k, l, s
1
, s
2
) = (div k) A +(k, /
k
l, s
1
, s
2
)
The last term approximately vanishes since
/
k
l has almost no component parallel to l:
g(k,
k
l
l
k) g(k, l +l) = 0. Hence
div k
n
lnA
103
Assume that the horizon is everywhere smooth
(this excludes e.g. collision of black holes!)
Then 1 is (locally) entirely covered by the ow
lines of k. Let t be the ow parameter for k.
Denote (t) div k along ow lines of k. Then
the total change in A is A =
_
+
dt
_
d
2
A
n
lnA =
_
+
dt
_
d
2
A(t)
Compute
t
(t) =
k
Tr
x
g(
x
k, x) = Ric(k, k)+
div
k
k Ric(k, k) +(t) since
k
is small.
Then (t) =
_
t
Ric(k, k)e
(
tt
)
dt
and so
_
+
(t)dt =
1
_
+
Ric(k, k)dt
=
8G
_
+
T(k, k)dt
The quantity E
_
d
2
A
_
+
dt
T(k, k) is the
total ux of energy through the horizon, so
E =
2
A
4G
= TS
Remarks: 1. The null energy condition guar-
antees T(k, k) 0 and thus S 0. But this
is only a weak 2nd law b/c we assumed very
small variations of smooth Killing horizon.
2. Introducing a rotational Killing vector
such that k =
t
+
we get
EJ = TS, J
_
d
2
A
_
+
dt
T(
)
Lecture 22, 2008-01-16. Raychaudhuri equa-
tion for timelike geodesics
Motivation: Want to describe the distortion of
a spacelike 3-volume by the Lie ow of a time-
like geodesic eld, compared with the parallel
transport (comoving ow)
v
c
a
=0
b
a()
b()
c()
Let v be a timelike geodesic eld. Let () be
one ow line. Let c be a connecting vector for
v; values of c on are c(). Let c
P
() be the
result of parallel transport of c(0) along .
104
Distortion of transverse connecting basis
Some useful properties:
Statement 1: A connecting vector c stays
orthogonal to v if v is normalized and geodesic
(or Killing)
Proof:
v
g(v, c) = g(v,
c
v) = 0;/
k
g(k, c) = 0
Statement 2: (Distortion of connecting ba-
sis) a) The connecting vector c satises
v
c =
B
v
(c) where
B
v
is the derivative tensor of v;
g(
B
v
(c), x) g(
c
v, x). b) The transforma-
tion operator
Z() dened by c() =
Z()c
p
()
satises
Z(0) =
1 and
d
d
Z() =
B
v
()
Z()
Proof: g(
v
c, x) = g(
c
v, x) B
(v)
(c, x);
_
d
d
Z
_
c
P
=
v
Zc
P
=
c
v =
B
v
c =
B
v
Zc
P
for
any (parallelly transported) basis vector c
P
.
Statement 3: Transverse 3-volume satises
d
d
V = /
v
(v, a, b, c) = (div v) V = V Tr
B
v
Statement 4: Derivative tensor is transverse,
B
(v)
(v, ) = B
(v)
(, v) = 0
Proof: g(
x
v, v) = 0, g(
v
v, x) = 0
So the tensor B
(v)
is eectively 3-dimensional.
105
Orthogonal decomposition of B
v
for time-
like v
Consider B
(v)
as a bilinear form in 3-dim. space
with reduced metric h
Note: divv = Tr
x
g(
x
v, x) = Tr
x
h(
x
v, x);
Tr
x
h(x, x) = 3; h(v, ) = 0
Decompose B
(v)
into pure trace, traceless sym-
metric, and antisymmetric parts (w.r.t. h):
B
(v)
=
1
3
hdiv v +
(v)
+r
(v)
where
(v)
is shear and r
(v)
=
1
2
dgv is rotation
of v. Note:
(v)
and r
(v)
are transverse to v.
Statement 5: Rotation vanishes for hypersurface-
orthogonal, non-null geodesic v.
Proof: Let gv; by assumption d = 0
and (v) ,= 0. Compute 0 =
v
( d) =
(v)d
v
d = (v)d
v
2r
(v)
=
(v)d, hence d = 0.
Counterexample for null v: set v = y (
t
+
x
)
in Minkowski spacetime; then v is null, hypersurface-
orthogonal, and geodesic, but r
(v)
,= 0.
Practice exercise: check the counterexample
106
Raychaudhuri equation
Compute
d
d
div v. (Note: all 1st derivatives of
x vanish under trace, but [v, x] ,= 0.)
v
Tr
x
g(
x
v, x) = Tr
x
R(v, x, v, x)
+Tr
x
g(
x
v
v, x) +Tr
x
g(
[v,x]
v, x)
= Ric(v, v) Tr
x
g(
B
v
B
v
x, x)
Now simplify for timelike geodesics v using
the 3-dim. decomposition. Since Tr() is a
scalar product in the space of matrices, we
have Tr(
B) = 0 if
A
T
=
A and
B
T
=
B.
Traceless matrices are orthogonal to h. Hence
d
d
div v = Ric(v, v)
1
3
(div v)
2
Tr
v
v
Trr
v
r
v
Comments: shear changes shape but not vol-
ume; rotation does not change shape; Tr can
be computed with h instead of g here; h is
negative-denite.
Focusing conditions
Statement 6: Tr
v
v
0 and Trr
v
r
v
0 for
any timelike v.
Proof: If a matrix
A satises
A
T
=
A then
Tr
Aa = Tr
x
h(
Ax, x) = Tr
x
h(
Ax,
Ax) 0
Strong energy condition: Ric(v, v) 0 for all
timelike v (attractive nature of gravity)
107
Focusing theorem
Statement 7: a) For a hypersurface-orthogonal
timelike geodesic eld v, assuming strong en-
ergy condition, we have
d
d
div v
1
3
(div v)
2
b) If div v =
0
< 0 at = 0 then div v =
at nite time =
0
, where
0
3
1
0
.
Proof: a) Follows from Statements 5 and 6.
b) Consider auxiliary variable x() (div v)
1
,
then
dx
d
1
3
and x(0) =
1
0
. It follows that
x(
0
) = 0 at
0
< 3
1
0
.
Interpretation: The volume of connecting ba-
sis shrinks to zero. So geodesics that are in-
nitesimally nearby focus onto () at =
0
.
This is congruence-dependent, not just a prop-
erty of spacetime.
Statement 8: Volume vanishes as V (
0
)
n
near =
0
, where we may have n = 1, 2, 3
Proof: 3-volume element V () is proportional
to det
Z() while
Z satises
d
d
Z =
B
v
Z; hence
d
d
det
Z = (div v) det
Z. Since
d
d
det
Z is al-
ways nite, we must have det
Z = 0 at =
0
where div v = . Since all components of
Z
are regular at =
0
, the determinant vanishes
as polynomial of
0
of degree at most 3.
108
Lecture 23, 2008-01-18. Rauchaudhuri
equation for null geodesics
Orthogonal decomposition of B
n
requires pro-
jecting onto the 2-dim. spacelike subspace or-
thogonal to n. Choose a null connecting vec-
tor l such that g(n, l) = 1,[n, l] = 0 everywhere.
Properties:
n
g(n, l) = 0; g(
l
n, l) = 0.
Projection onto n
is the operator
P
n
x = x ng(l, x) lg(n, x)
2-dim. reduced metric h
(2)
= g
P
n
x,
P
n
y)+B
n
(
P
n
x, l)g(n, y)+
B
n
(l,
P
n
y)g(n, x)+B
n
(l, l)g(n, x)g(n, y). Denote
for brevity B
n
(
P
n
x,
P
n
y) B
n
(x, y). Compute
trace: Tr
x
g(
B
n
B
n
x, x) = Tr
x,y
B
n
(x, y)B
n
(y, x) =
Tr
x,y
B
n
(x, y)B
n
(y, x) because all other com-
ponents generate
P
n
n = 0 or g(n, n) = 0.
109
Now decompose B
n
(x, y) according to sym-
metry and trace properties:
B
n
=
1
2
h
(2)
div n +
(n)
+r
(n)
n
1
2
[B
n
(x, y) +B
n
(y, x)]
1
2
h
(2)
(x, y)div n
r
n
1
2
[B
n
(x, y) B
n
(y, x)]
Raychaudhuri equation:
d
d
div n = Ric(n, n)
(div n)
2
2
Tr
_
n
_
Tr
_
r
n
r
n
_
Statement 1: r
n
= 0 if n is hypersurface-
orthogonal. (Even if r
n
does not vanish.)
Proof: The tensor r
n
is equivalently dened
by r
n
(x, y) =
1
2
dgn(
P
n
x,
P
n
y). This vanishes
(by construction) if either x or y is not in n
.
It remains to consider x and y both in n
,
so that
P
n
x = x and
P
n
y = y. Then
x
gn =
y
gn = 0 and (by the Frobenius condition) 0 =
y
(gn dgn) = gn
x
y
dgn = 2gnr
(n)
(x, y). So
r
(n)
(x, y) = 0.
Null energy condition (NEC): Ric(n, n) 0 for
any null n. [Equivalently, T(n, n) 0]
Focusing theorem: If NEC holds and div n <
0 then focusing occurs within nite interval of
ane parameter
Proof: Dene x (div n)
1
; then
dx
d
1
2
and
x(0) =
1
0
, so x(
0
) = 0 where
0
< 2
1
0
Geodesic deviation equation
If c is a connecting vector to a geodesic eld
v then c satises the 2nd order equation:
v
c =
R(v, c)v
Choose v as the eld of all the geodesics emit-
ted from a point p. Solve geodesic deviation
equation along a single curve . Obtain c()
such that c(0) = 0,
v
c(0) c ,= 0.
The vector c() is a Jacobi eld for ().
Point q is conjugate to p if c() not
identically zero, solving GDE, c = 0 at p and q.
Interpretation: an innitesimally nearby geodesic
curve emitted at p will intersect again at q.
Dene the operator
A() as the map from
c( = 0) to c() in a parallelly propagated
basis. Then
A =
G
A, where
Gx
R(v, x)v.
Initial conditions:
A(0) = 1
P
v
,
A(0) =
P
v
Conjugate points and focusing
1. A vector eld emitted at p is always hypersurface-
orthogonal (proof below)
2. If div v < 0 at some point p then focus-
ing is guaranteed (assuming energy conditions)
within a nite distance from p
3. How to compute div v for this eld: div v =
d
d
lndet
A. So det
A 0 when div v
110
Lecture 24, 2008-01-23. Extremal prop-
erties of geodesics
To understand the extremal properties it is im-
portant to consider a particular kind of vector
eld: namely geodesics emitted from a point.
This eld is hypersurface-orthogonal (see be-
low) and hence we may use the bounds on
focusing from the Raychaudhuri equation. Fo-
cusing will control the extremal properties.
Statement 2: A geodesic vector eld v emit-
ted at a single point p is hypersurface-orthogonal.
Proof: Let gv; we will show that d =
0. a) Suppose that v is timelike. Since d =
2r
v
while r
v
is transverse to v, the only nonzero
possibility is d(v, a, b) = (v)2r
v
(a, b) where
a, b v
. However, 2r
v
(a, b) = g(
a
v, b)
g(
b
v, a). Let us choose a basis of vector
elds a, b, c connecting to v; by construc-
tion these elds vanish at p. So r
v
(a, b)[
p
=
0. Now compute the derivative of 2r
v
(a, b)
along v. We nd 2
v
r
v
(a, b) =
v
g(
v
a, b)
v
g(
v
b, a) = R(v, a, v, b) R(v, b, v, a) = 0.
Hence r
v
(a, b) = 0 everywhere.
111
b) Suppose that v is null. Choose a null con-
necting vector l such that g(v, l) = 1. The only
nonzero possibility is d(l, a, b) = 2r
v
(a, b)
where a, b v
G is symmetric,
G
T
= G, with respect to the
metric g
Proof: Compute g((
G
G
T
)a, b) = g(
Ga, b)
g(a,
Gb) = R(v, a, v, b) R(v, b, v, a) = 0, so
G
G
T
= 0
Conjugate points (0) and (
0
) correspond to
focusing of innitesimally nearby geodesics:
= 0
()
=
0
c()
Second variation of action in mechanics
Consider mechanical system with action
S[x] =
_
T
0
_
1
2
x
2
V (x)
_
dt
Is the action minimized by a trajectory x()
that solves x +V
(x) = 0?
Compute rst and second variation of S:
S =
_
T
0
_
x V
(x)
_
x(t)dt
2
S =
_
T
0
_
x +V
(x)x
_
x(t)dt
If there exists a solution for geodesic devi-
ation such that x(0) = 0, x(0) ,= 0, and
x(t
0
) = 0 for some t
0
> 0, call t
0
a conju-
gate point to t = 0. Suppose no conjugate
points exist until t = T. Then the solution
A(t) of
A + V
(x)A = 0, A(0) = 0,
A(0) = 1
is nonzero for 0 < t T. For any x(t) dene
s(t) x(t)/A(t) and substitute into
2
S:
2
S =
_
T
0
_
As +2
A s
_
Asdt
=
_
T
0
_
_
A
2
s s
_
.
+A
2
s
2
_
dt
=
_
T
0
A
2
s
2
dt > 0 if s(t) ,= 0 somewhere
Then S has a minimum on x(t).
112
No minimum of action with conjugate points
Statement: If a conjugate point exists at t
0
<
T, then S[x] has no minimum at x(t).
Proof: We will present explicitly a x(t) such
that
2
S < 0. By assumption, a solution c(t)
of c + V
(x)x c(t
0
)(t t
0
) +
O() near t t
0
, so
2
S =
_
T
0
_
x +V
(x)x
_
x(t)dt
=
_
T
0
_
c
2
(t
0
)(t)(t t
0
) +O(
2
)
_
dt
= c
2
(t
0
)(t
0
) +O(
2
) < 0
for suciently small . (Note: sign of is
chosen to cut the corner.)
T
c(t)
0
t
x(t)
t
0
113
Second variation of geodesic length
Consider a timelike curve perturbed by a vec-
tor eld t. Dene Lie-propagated vector eld
v such that /
t
v = 0. Assume that t = 0 at
endpoints
1
,
2
. Proper length of the curve
(; ) displaced by parameter is
L[(; )] =
_
2
1
_
g(v, v)d
Compute rst and second derivatives w.r.t. :
d
d
L[; ] =
_
2
1
g
_
t,
v
v
N
_
d, N
_
g(v, v)
d
2
d
2
L[; ] =
_
2
1
_
g
_
t
t,
v
v
N
_
+g
_
t,
t
v
v
N
__
d
For a geodesic curve,
v
v
N
= 0, and we can
set N = 1 to simplify calculations. Moreover:
g(t,
t
v
v) = R(t, v, v, t) +g(t,
v
v
t)
Hence the 2nd variation evaluated on is
d
2
d
2
L[; ] =
_
2
1
g
_
t,
Gt
v
v
t
_
d
If this is negative for any eld t() then L[]
is locally maximized by the curve
Note: if has conjugate points
1
,
2
with a
Jacobi eld c(), then choosing t = c yields
the absence of maximum! (See next theorem)
114
Theorem of geodesic extremum:
a) If a timelike geodesic curve () has no
conjugate point to = 0 within an interval
[0, T], then is a local maximum of L[] for
curves between (0) and (T).
b) If =
0
< T is a conjugate point to =
0 on () then there exists a longer timelike
curve between (0) and (T).
Proof: a) Consider the operator
A() for .
If there are no conjugate points for (0) within
[0, T], then
A() is nonsingular, so
A
1
()
exists for [0, T]. We will now show that
d
2
d
2
L[; ] =
_
T
0
g
_
t,
Gt
v
v
t
_
d < 0
for all nonzero t() v
that vanish at = 0
and = T. Dene s()
A
1
()t(), denote
v
under the integral by the overdot, and ex-
press the integral through s():
_
T
0
g
_
As,
G
As (
As)
_
d =
_
T
0
g
_
As, 2
A s +
As
_
d
Now we add a total derivative
v
g(
As,
A s) =
g(
As +
A s,
A s) +g(
As,
A s +
As) and obtain
d
2
d
2
L[; ] =
_
T
0
g
_
A s,
A s
_
d
+
_
T
0
_
g
_
As,
A s
_
g
_
As,
A s
__
d
115
Rewrite g(
As,
A s)g(
As,
A s) = g
_
s, (
A
T
A
A
T
A) s
_
and examine this Wronskian-like combination:
A
T
A
A
T
A = 0 at = 0, while
A
T
=
A
T
G
T
=
A
T
G. Hence
_
A
T
A
A
T
A
_
.
= 0 and nally
d
2
d
2
L[; ] =
_
T
0
g
_
A s,
A s
_
d < 0
since
A is nondegenerate (has no zero eigen-
vectors) and
A s is always spacelike, while s ,= 0
at least for some .
b) If there exists a conjugate point at =
0
< T, then Jacobi eld c() ,= 0 such that
c(0) = c(
0
) = 0. Using c() we will explicitly
construct a eld t() such that
d
2
d
2
L[(; )] >
0 when evaluated on the geodesic . The
idea is to cut the corner: set t() c()
for 0 < <
0
and t 0 for >
0
. Explicitly,
t = c()(
0
) q()b, where b = const,
g(b, c(
0
)) = 1, and > 0, q() > 0. Then
Gt
t() = c(
0
)(
0
) +O()
and nally we obtain
d
2
d
2
L[; ] =
_
T
0
g(t,
Gt
t)d
=
_
T
0
g(q()b, c(
0
)(
0
))d
= q(
0
) +O(
2
) > 0
Lecture 25, 2008-01-25. Extremal prop-
erties of null geodesics
Since null curves cannot be obtained by ex-
tremizing
_
_
g(v, v)d, we will use an alterna-
tive variational principle:
L[] =
_
T
0
g( , )
N
d,
where N() is a Lagrange multiplier. Dene
the vector eld v as the Lie transport of by
a perturbation eld t, then /
t
N = 0, /
t
v = 0.
First variation w.r.t. parameter :
d
d
L[; ] = /
t
_
g(v, v)
N
d =
_
2g(
t
v, v)
N
d
=
_
2g(
v
t,
v
N
)d =
_
d
d
(...)d
_
2g(t,
v
v
N
)d
= 2
_
g(t,
v
v
N
)d
Second variation (using
v
v
N
= 0 on shell):
d
2
d
2
L[; ] = 2
_
g(t,
t
v
v
N
)d
= 2
_ _
R(t, v,
v
N
, t) +g(t,
v/N
v/N
t)
_
d
= 2
_
g(t,
G()t
t)Nd,
where
G()t
R(
v
N
, t)
v
N
; overdot means
d
d(N)
.
116
Restrict N = 1 everywhere, so have
v
g(v, t) =
g(v,
v
t) = g(v,
t
v) =
1
2
t
g(v, v) = 0 and
since t = 0 at endpoints, must have g(v, t) = 0
everywhere.
So it suces to consider t v
.
Also, t = v gives no variation since
Gv = 0
and g(t,
up to multiples of v.
Introduce a null vector u such that g(u, v) = 1
and x the projector
P
v
x x vg(x, u) =
ug(x, v). Then it suces to consider t (u, v)
.
Dene
A() by the equation
G
A = 0,
A(0) =
1
P
v
,
A(0) =
P
v
. Then focus point
=
0
exists i a solution
A() exists with
det
A(
0
) = 0.
Now the same argument involving a solution
A s,
A s)d < 0
since
A s is in (u, v)
, c(
0
) = 0,
c(
0
) ,= 0. Consider the curve that follows
v +c until =
0
and then v. The vector is
piecewise null.
2. Consider the tangent vector w at =
0
. The vectors v and w are both future-
directed and null, so g(v, w) > 0, g(v, v) =
g(w, w) = 0. Hence b v w is spacelike:
g(b, b) = g(v w, v w) = 2g(w, v) < 0.
Also g(b, w) > 0, g(b, v) < 0.
3. The deviation vector t() can be chosen as
t = q()b where > 0, q() > 0, sign q() =
sign(
0
) near =
0
, for example q() =
1 [
0
[. Then q()g( , b) > 0 for all .
4. The tangent vector to the deformed curve
is
= +
t = + q()b.
5. We verify that
is everywhere timelike:
g( +
t, +
t) = 2 q()g( , b) +O(
2
) 0
117
t
v
v
w
w
118
Extremal properties for geodesics emitted
from a surface
Consider a spacelike surface and a point p ,
. What is the shortest distance from p to ?
It is the geodesic emitted normally from
towards p unless there exists a conjugate point
on before p.
Example: let be the half-hyperboloid given
by t
2
x
2
y
2
z
2
= 1, t < 0 in Minkowski
spacetime. The normal vector eld is n =
t
t
+x
x
+y
y
+z
z
. A straight line (geodesic)
at point t, x, y, z towards the origin is x =
t, x, y, z and so it is orthogonal to
because x | n.
Hence the origin is the focus of geodesics emit-
ted normally from .
Consequence: There is no (timelike) curve of
maximal proper length connecting 1, 0, 0, 0
and
119
The easiest singularity theorem
If is a compact Cauchy surface with (time-
like) normal vector n, such that div n < 0 ev-
erywhere on , and if the strong energy condi-
tion holds to the future of , then no timelike
geodesic to the future from is complete.
Proof: 1. Consider the timelike geodesic eld
n emitted normally from . Since is com-
pact, div n has an upper bound such that div n <
0
everywhere on with some
0
> 0. Since
n is hypersurface-orthogonal, by focusing the-
orem every geodesic has a conjugate point to
at <
0
3
1
0
.
2. By extremal length theorem, no geodesic
maximizes proper length longer than
0
.
3. Since is compact and Cauchy, for any
point p to the future of there is a nonempty
set of timelike or null curves intersecting and
p. These curves intersect at a compact set,
hence there is a point p
could
be chosen as shifted a little bit towards the
past. Suppose p is a point where the null gen-
erator leaves 1. By assumption p is not a
singularity, so we may continue past p; then
enters B\1. However, arbitrarily nearby curves
2
, generating a portion of 1
2
. The area
of that portion cannot be smaller than the area
of 1
1
since div n 0. Other disconnected
pieces of 1
2
might have appeared, making
the area of 1
2
even larger.
Remark: Since div n 0 everywhere, the area
of each connected piece of 1 is non-decreasing.
122
Singularity in collapsing universe
The existence of a singularity = the existence
of a geodesic that cannot be continued beyond
a nite parameter range .
Note: If a piecewise timelike or null curve be-
tween points p, q has proper length L, then
there exists a timelike geodesic between p, q
of length at least L. Same for curves between
a point p and a surface .
Statement 3: If is a spacelike Cauchy su-
face with normal vector eld n such that div n
0
< 0 everywhere, and if the strong energy
condition holds, then every timelike geodesic
from has proper time 3
1
0
.
Proof: Select any point p to the future of
and consider the set of all causal (timelike or
null) geodesic curves leading to p. All these
curves intersect only once (Cauchy prop-
erty of ). The set of tangent vectors at p is
compact (it includes the past timelike and null
directions), hence the set of intersection points
with is a compact subset of . Hence, there
123
exists the longest (by proper length) geodesic
curve
(p)
max
between and p. Hence,
(p)
max
is also the longest piecewise causal curve be-
tween and p.
The curve
(p)
max
must be timelike (since it has
proper length ,= 0). By the focusing theorem,
every timelike geodesic orthogonal to will
have a focus within 3
1
0
. Hence, no
geodesic longer than 3
1
0
maximizes proper
length. Hence,
(p)
max
is not longer than 3
1
0
,
for any point p to the future of .
Singularity in closed universe
Closed universe = containing a compact space-
like 3-surface without boundary (but not
necessarily a Cauchy surface!)
Cauchy horizon of is 7
+
() and is locally
a null surface such that 7
+
() is on the future
side of 7
+
(). Null generators of 7
+
()
cannot leave it (same reason as before).
Statement 4: If a compact spacelike 3-surface
without boundary, with the normal vector
n such that div n < 0 on , and if the SEC
holds, and if the spacetime has no closed time-
like curves (strongly causal), then there is at
least one incomplete timelike geodesic emitted
from .
Proof: Since is compact, div n <
0
< 0
for some
0
everywhere. By the focusing theo-
rem and extremal theorem, no geodesic emit-
ted from that is longer than 3
1
0
maxi-
mizes length. Let us show that, for every point
p 7
+
(), the set of causal geodesics be-
tween and p is compact. The intersection
124
7
(p) corresponds
to a sequence of tangent vectors in the past
lightcone at p and thus has an accumulation
point, a tangent vectorv. If the geodesic
emitted from p in the direction v does not in-
tersect , then a neighborhood around also
does not intersect ; this is impossible.
Hence for each p there exists a timelike geodesic
(p)
max
of maximum length, among all geodesics.
But no other causal curve between and p
can be longer than 3
1
0
; this is so because
any causal curve can be deformed to a smooth
causal curve of no smaller proper length, and
then, if it is not a geodesic, deformed further
to a curve of longer proper length. Hence,
a nongeodesic curve cannot maximize proper
length. The only remaining possibility is that
there are (nongeodesic) causal curves of ar-
bitrarily large proper length. However, this is
impossible because the set of all causal curves
between a compact and p is compact (in the
C
0
topology in the space of curves). Proper
length is upper semicontinuous as a functional
of piecewise dierentiable curves, and hence
there exist a causal curve of nite maximum
length between and p. This curve must be
a geodesic and thus is shorter than 3
1
0
.
Hence, no p 7
+
() can be further than 3
1
0
from . , so 7
+
()
_
0, 3
1
0
_
. Hence
7
+
() is compact.
Now consider the Cauchy horizon 7
+
(): it
is either compact or noncompact. If it is com-
pact, then it contains null generators that wind
around innitely many times. Then closed
timelike curve. Hence 7
+
() is noncompact.
But 7
+
() is compact; contradiction.
Remark: Hawking & Ellis give a longer proof
without requiring the no-CTC condition.
Lecture 27, 2008-02-01. Singular collapse
There are two families of null geodesics emit-
ted normally from a spacelike 2-surface; in-
wards and outwards.
Trapped 2-surface = both inwards and out-
wards future null geodesics emitted normally
to the 2-surface have negative divergence.
Trapped surface theorem (Penrose): If a
time-orientable spacetime / has a noncom-
pact Cauchy surface ( and a compact, space-
like trapped 2-surface S, then there exists an
incomplete null geodesic to the future of S.
Proof: Consider 7
+
(S): it is generated by
null geodesics n emitted normally fromS. Since
S is compact, div n <
0
< 0 on S. Suppose
that every null geodesic has length 2
1
0
from S. By the focusing theorem and the ex-
tremal theorem for null geodesics, every point
p on a null geodesic after = 2
1
0
will be
inside 7
+
(S). Hence, 7
+
(S) S
_
0, 2
1
0
_
and so 7
+
(S) is compact.
Since / is time-orientable, v : g(v, v) = 1.
Each orbit of v intersects 7
+
(S) and also (
exactly once. Hence orbits of v map 1-into-
1 7
+
(S) (. So the image of 7
+
(S) is
a compact subset of (, hence its boundary is
nonempty, but 7
+
(S) = . Contradiction.
125
Topology change closed timelike curves
Theorem (Geroch): If the spacetime domain
/ between two compact, spacelike, bound-
less surfaces
1
,
2
is compact and admits
no closed timelike curves, then
1
=
2
(have
the same topology) and /
=
1
[0, 1].
Proof: Choose a timelike, nonvanishing vec-
tor eld v in /. Orbits of v from
1
either
end on
2
or never reach
2
. If an orbit never
reaches
2
, it winds around itself and can be
deformed to a CTC. Hence, all orbits reach
2
in nite parameter time and provide a 1-
to-1 map
1
2
. Parameter time can be
rescaled to [0, 1].
Theorem (Hawking): If there is a tube-like
domain T connecting spacelike 3-surfaces
1
and
2
of dierent topology, and if T con-
sists of a timelike surface and of
1,2
, then
there is a closed timelike curve in T .
Proof: One can choose a nonvanishing time-
like vector eld v in T such that v is tangent to
the timelike side boundary of T . If each orbit
of v from
1
ends on
2
, then one would have
a 1-to-1 map
1
2
, but they have dierent
topology. So there are some orbits that wind
around in T and never reach
2
. These orbits
can be deformed into CTCs.
126
Holographic principle
Generalized 2
nd
law of thermodynamics:S 0
where S =combined entropy of matter +
1
4
of
combined area of black hole horizons.
Susskind process: A system can be made into
a black hole with entropy R
2
by adding
more energy to it.
Spacelike entropy bound: If a system is bounded
by a sphere of radius R then its entropy is not
larger than R
2
.
GSL = A black hole is the highest-entropy ob-
ject possible in a given spherical volume.
This suggests that the fundamental degrees of
freedom scale as area rather than as volume;
hence the name holography.
127
Spacelike bound is not generally covariant and/or
not generally valid.
Diculties with generalizing the spacelike bound:
1. Not clear what is inside and what is outside
a sphere if the universe is closed.
2. In an expanding universe, obviously one
can nd an arbitrarily large volume and the en-
tropy density is constant on large scales, hence
S A
3/2
3. In a collapsing star, we may enclose it in
arbitrarily small (infalling) sphere and violate
the bound.
4. Choose a spacelike 3-surface (surface of
constant time) such that the boundary of the
system is a 2-surface of vanishing 2-area, then
area is arbitrarily small while entropy enclosed
in it remains nite. (The entropy bound, as
formulated, is noncovariant!)
These pitfalls are in some sense kinematical
since they do not involve a violation of the
Susskind process. Need a covariant formula-
tion of the entropy bound that avoids these
pitfalls.
128
Covariant entropy bound (Bousso)
A lightsheet / for a 2-surface is a null sur-
face made by emitting null geodesics normally
to such that div n 0 everywhere. (/ needs
to be cut if the nonconvergence condition is
violated somewhere.)
Entropy on a lightsheet is the integral over
/ of the 3-form dual to the entropy current
s. (Note: integration in the null direction is
independent of the choice of the ane param-
eter!)
Covariant entropy bound: The total entropy
on a lightsheet is smaller than
1
4
of the area of
the initial surface .
Interpretation: This is the entropy visible from
. This avoids the counterexamples above!
Spacelike projection theorem (Bousso): If
a system is enclosed in and the lightsheet
/ emitted from has as its only boundary,
then the entire entropy of the system is not
larger than the entropy on the worldsheet.
Proof: All matter within will pass through
/; the entropy will not decrease until that time.
129
Lecture 28, 2008-02-06. Ination is not
eternal to the past
Ination = accelerated expansion of spacetime.
Example: de Sitter spacetime, g = dt
2
e
2Ht
dx
2
Expansion of spacetime = timelike geodesic
vector eld v such that div v > 0 everywhere
throughout a spacetime domain between two
Cauchy surfaces. For de Sitter spacetime: v =
t
. Compute div v = Tr
x
g(
x
v, x) using or-
thonormal basis
_
v, e
Ht
j
_
: since g(
j
,
j
) =
e
2Ht
, we get div v = g(
v
v, v)
3
j=1
e
2Ht
g(
j
v,
j
) = 3e
2Ht
1
2
v g(
j
,
j
) = 3H.
Hence, in a general spacetime we estimate the
local Hubble rate as H
1
3
div v g(n,
n
v)
for any spacelike, normalized n (g(n, n) = 1)
Theorem (Borde & Vilenkin): If a spacetime
is everywhere expanding, i.e. there exists a
timelike geodesic eld v such that g(n,
n
v)
H
0
> 0 for any spacelike vector n (assuming
g(v, v) = 1, g(n, n) = 1, g(v, n) = 0) every-
where to the past of a Cauchy surface , then
there exists a timelike geodesic that is incom-
plete to the past of .
130
Proof: Visually, assume that spacetime is lled
with particles move along worldlines that are
orbits of v. The incomplete timelike geodesic
will be constructed explicitly. It will be the
worldline of an observer not at rest with re-
spect to the comoving particles. One can choose
a geodesic vector eld u such that g(u, u) = 1
and g(v, u) > 1 everywhere. Denote g(v, u)
(this is the gamma factor of special relativity,
= (1 v
rel
)
1/2
, where v
rel
is the relative ve-
locity of observer w.r.t. comoving particles).
Then u = v+n for some constants , and
spacelike vector n orthogonal to v. Explicitly,
projecting u onto v
, we get
n =
u vg(u, v)
[u vg(u, v)[
=
u v
_
2
1
u +v.
The observer can measure
u
by observing
relative velocities of comoving particles. Then
the observer estimates the local expansion fac-
tor as
H = g(
n
v, n) = g(
u+v
v, u +v)
=
2
g(
u
v, u) =
2
u
g(v, u) =
u
2
1
.
Replace
u
=
1
Hd =
_
2
2
1
dt =
1
2
ln
+1
1
1
<
1
2
ln
(
2
) +1
(
1
) 1
since > 1 and ln(...) > 0. On the other hand,
_
1
Hd H
2
(
2
1
). Hence, there is a lower
bound on the value of
1
:
1
>
2
1
2H
0
ln
(
2
) +1
(
2
) 1
1min
.
Hence, the worldline u cannot be extended
to the past beyond
1min
and is incomplete.
(Note: orbits of v are complete, but all other
geodesics are incomplete.)
Conclusion: Either there exists a singularity
to the past, or ination is not past-eternal.
Remarks: 1. The theorem is purely kinemat-
ical - no Einstein equations or energy condi-
tions are used.
2. In de Sitter space, either one uses global
coordinates where the universe is initially con-
tracting, or the worldline u reaches the edge
where the geodesic eld v is not expanding any
more.
Superluminal travel
Warp drive (Alcubierre 1994):
g = dt
2
(dx v
s
f(r
s
)dt)
2
dy
2
dz
2
,
v
s
(t)
dx
s
dt
, r
s
_
(x x
s
(t))
2
+y
2
+z
2
,
f(r)
_
_
_
1, [r[ < R
0, [r[ > R
The worldline , x
s
(), 0, 0 is timelike be-
cause d = dt on (no time dilation!).
Properties: 1. is always a geodesic. Proof:
Let u =
t
+ v
s
(t)
x
. Compute gu = dt,
hence g(
u
u, x) =
1
2
(/
u
g)(u, x) = g(
x
u, u) =
0 since g(u, u) 1.
2. The function x
s
(t) is completely arbitrary,
which allows superluminal travel.
3. All energy conditions are violated. [Omitted
calculation: T(
t
,
t
) v
2
s
f
2
]
Negative energy is concentrated in the skin
of the bubble.
131
Superluminal travel needs negative energy
K. D. Olum, Phys.Rev.Lett.81, p.3567, 1998; gr-qc/9805003
Generic superluminal travel: Assume that there
exists a null geodesic that travels further (in a
xed, externally dened time) than any neigh-
bor null geodesic.
Formalize this construction. Let lightrays be
emitted orthogonally from an extrinsically at
(made of spacelike geodesics), spacelike 2-surface
A
. The externally dened time is a space-
like 3-surface ( intersected by the lightrays.
Let the furthest-reaching lightray be between
A
A
and B (. Hence, there exists an
extrinsically at 2-surface
B
( such that
B
B
but no other point p
B
can be
reached by any lightrays from
A
. Then call
the path between A and B superluminal.
Theorem (K. Olum): The NEC is violated on
a superluminal path if Ric( , ) < 0 at some
point on .
Proof: Consider a congruence n of geodesics
emitted normally from
A
. There are no con-
jugate points to
A
on (or else B is timelike-
reachable from A). Hence, there is no focus-
ing, and div n remains nite everywhere on .
132
Initially, div n = 0 on
A
because there exists a
basis of (spacelike) geodesic vector elds c
i
T
A
and we have g(n, c
i
) = 0 and g(
c
i
n, c
i
) =
c
i
g(n, c
i
) g(n,
c
i
c
i
) = 0.
Raychaudhuris equation gives
n
div n = Ric(n, n) Tr(
n
)
1
2
(div n)
2
Since Ric(n, n) < 0 somewhere on , we must
have div n < 0 at B.
Now we show that div n 0 at B. Dene a
basis of connecting elds c
i
for n, and consider
c
i
at point B. The elds c
i
are not geodesic
at B; the orbits of c
i
must leave
B
at B
towards the future of B, or else points near
B are in 7
+
(
A
). Hence, the deviation
c
i
c
i
must point to the future, so g(n,
c
i
c
i
) 0.
Also, is normal to
B
at B (or else it can be
deformed to reach other points near B).
Also, g(c
i
, n) = 0 everywhere because
n
g(c
i
, n) =
0 while g(c
i
, n) = 0 everywhere on
A
.
To compute div n at B, consider g(
c
i
n, c
i
) =
g(
c
i
c
i
, n) at B. Since g(
c
i
c
i
, n) 0, we
nd div n =
i
g(
c
i
n,c
i
)
g(c
i
,c
i
)
0. Contradiction.
Remarks: 1. The NEC is violated directly on
the path of the signal.
2. Singularities do not help prevent this.
133
Lecture 29, 2008-02-08. Conservation laws.
Komars global energy integral
Consider a stationary spacetime with a Killing
vector k. The derivative tensor B
k
=
1
2
dgk is
a 2-form, hence its divergence is a divergence-
free 1-form (conserved current):
div B
k
= Tr
x
g(
x
x
k, ) = g(k, )
[Why is it conserved? div B
k
= dB
k
, hence
(d)(d)B
k
= 0.]
More generally: any rank 2 tensor satises
;[]
= R
+R
, hence a 2-form
satises
;
= 2Tr
x,y
Ric(x, y)(x, y) = 0.
The conserved charge Q is obtained by inte-
grating the ux of the conserved current q
g
1
div B
k
= k through a 3-surface (:
Q
_
(
g(q, n)d
3
V =
_
(
g(k, n)d
3
V
Since k is a total divergence, rewrite this as
Q =
_
=(
_
_
n
d
2
B
k
(v, n)d
2
, where is a sphere
of large radius R. Vectors normal to the sphere
are v z
1
k, n = z
r
. Need to compute
B
k
(v, n) = g(
v
k, n) = g(
z
1
k
k, z
r
). We
know that the proper acceleration of station-
ary observers is
k
k =
1
2
g
1
dz
2
. Hence
g(
k
k,
r
) =
1
2
r
z
2
=
Gm
r
2
Integrating over the 2-sphere of radius R:
Q =
_
m
r
2
_
d
2
= 4Gm.
This is the mass of the black hole, times (4G).
Now we show that this is not a coincidence:
the total energy (including gravity) of a sta-
tionary spacetime is E =
1
4G
Q
135
Interpretation of Q as the energy
Statement: In the Newtonian limit, Q = 4E,
where E is the total mass of the matter sources.
Proof: The property
x
y
k =
R(x, k)y yields
g(k, x) = Tr
y
g(
y
y
k, x) = Tr
y
R(y, k, y, x) =
Ric(k, x) and hence Q =
_
(
Ric(k, n)d
3
V .
Newtonian limit = particles move along or-
bits of k, normal to the equal-time surface (;
the energy-momentum tensor of matter sources
is T
= k
+8GT
]g
+q
;
)
gd
4
x
where Ein = Ric
1
2
Rg is the Einstein ten-
sor, and q is the vector eld such that q
;
=
g
. (Explicitly, Ric = /
v
Ric and q =
g
1
divdgv.) Since the above is an identity,
while g
= v
;
v
;
, so we have (using
Bianchi identity and conservation of T
) that
[Ein
+8GT
]g
= 2([Ein
+8GT
]v
)
;
and so
(q
2[Ein
+8GT
]v
)
;
= 0.
On-shell (Ein +8GT = 0) we have q
;
= 0,
i.e. a conserved current. Up to a total deriva-
tive, one can write
q
= 2[Ein
+8GT
]v
+
_
v
U
[]
_
,
137
One can add an exact 1-form to q
such that
the result does not contain 2nd derivatives of
g (but contains derivatives of v
):
q
= q
+
_
v
[]
_
,
, U
[]
= ...
Explicitly, one will have
q
= 8GT
v
;
for some (non-tensor) quantities and .
A conserved density is obtained as
gq
such that
_
gq
_
,
=
gq
;
= 0 on-shell.
Choose 4 dierent vectors v
0
, v
1
, v
2
, v
3
and
obtain a pseudotensor. For example, in a co-
ordinate system x
choose v
0
= 1, 0, 0, 0,
v
1
= 0, 1, 0, 0 etc., i.e. v
(noncovariant
denition!). Then one obtains four conserved
currents q
0
, ..., q
3
. Since the numerically cho-
sen components of v
= 8GT
0
+
0
and so on; hence
(8GT
)
,
= 0
so
1
8G
and
dierent coordinate systems x
, various energy-
momentum pseudotensors can be constructed.
These pseudotensors are not tensors for the
following reasons:
1. They are dened through vectors v
given
by numerical components in a xed coordinate
system.
2. They are collections of four conserved vec-
tors, rather than single conserved rank 2 ob-
jects.