1 Special Theory of Relativity

Here we use the convention ⌘µ⌫ = ( 1, 1, 1, 1), but the other convention ⌘µ⌫ = (1, 1, 1, 1) is also popular in litera-
ture.

Maxwell’s equations says that the E&M phenomena are described by the speed c of light, and all the known waves al-
ways propagate through a medium. Aether was proposed to be such medium for light propagation, and the light propagates
at c in Aether. However, according to the Galilean transformation, the speed of light will be different, if the laboratory
or the source is moving. The Michelson-Morley experiment in 1895 was devised to detect the Earth motion relative to
Aether by measuring the change in the speed of light.
In 1895, Lorentz, Fitzgerald, Poincaré explained the null results of the Michelson-Morley experiment by adopting
two ad hoc hypotheses: contraction of rigid bodies and time dilation relative to Aether, and Lorentz provided a concise
formula (or the Lorentz transformation) that captures it in a mathematical form. In fact, all the essence of special relativity
is contained in the Lorentz transformation. Einstein in 1905 re-evaluated the idea of spacetime and re-derived the Lorentz
transformation based on two principles, providing its correct physical interpretations and highlighting the limitation of
classical mechanics in terms of simultaneity.
Einstein’s special relativity is based on two principles:
1 laws of physics are the same in any inertial frames (same from old days),
2 the speed of light is the same to all observers in any inertial frames.

1.1 Lorentz Transformation and Physical Interpretations

A general coordinate transformation from xµ to x̃µ (x⌫ ) can be expanded for an infinitesimal change:
x̃µ = xµ + ⇠ µ , (1.1)
and one can integrate to obtain a finite coordinate transformation, where µ, ⌫, · · · represent the spacetime indices, while
i, j, k, · · · represent the spatial indices. In classical Newtonian mechanics, Newton’s law is invariant (or valid) only in the
inertial frames, and the inertial frames are related in terms of Galilean transformation:
x̃i = Ri j xj + v i t + ci , t̃ = t + c0 , (1.2)
where R is a rotation matrix, and v i , ci , c0 are constant. In short, the inertial frames are related to each other by a simple
spatial and temporal translation ci , c0 , a rotation R, or a relative motion with constant speed v i . Note that in Galilean
transformation there is no upper limit to the relative speed.
However, it is clear that the Galilean transformation does not preserve the speed of light, and hence the Maxwell’s
equations appear different in different inertial frames. The Lorentz transformation preserves the speed of light and the
metric ⌘µ⌫ in all inertial frames:
x̃µ = ⇤µ ⌫ x⌫ + cµ , (1.3)
where the constant shift cµ is a space-time translation. The Lorentz transformation is made of rotation and (Lorentz)
boost, and the Poincaré transformation includes translation in addition to the Lorentz transformation. Given the full gen-
erality in Eq. (1.1), we consider only very specific transformations (or Lorentz transformations) in Minkowski spacetime.
Furthermore, since the spacetime is intertwined, it is clear that we have to consider four (spacetime) vectors for physical
quantities, rather than ordinary three (spatial) vectors.

1.1.1 Transformation Laws

The transformation law is determined by the (symmetric) Lorentz matrix ⇤ (its components are denoted as ⇤µ ⌫ ). The
matrix ⇤ describes the transformation of a four vector in the rest frame to a four vector in the frame moving with the
velocity i relative to the rest frame:
2 i
j
⇤0 0 = , ⇤ 0 i = ⇤i 0 = i
, ⇤i j = i
j + i
j = i
j +( 1) 2
, (1.4)
1+
AST511 General Relativity JAIYUL YOO

where we used the relation

i 1/2 2 2 2
:= 1 i , = 1. (1.5)
Please see the lecture note for the derivation of the Lorentz transformation. The components of four vectors represent
those measured in each rest frame, and the Lorentz transformation of the four vectors provides the relation between two
frames. It is symmetric, in the sense that when the velocity vector i is replaced with i , the same formula can be used

for the inverse transformation.

For example, consider the relative velocity to be aligned along the z-direction, and the matrix becomes
0 1
0 0
B 0 1 0 0 C
⇤µ ⌫ = B
@ 0
C , i
:= i
z , (1.6)
0 1 0 A
0 0

and two coordinates of an event are related as1

t̃ = (t z) , z̃ = (z t) . (1.8)

While in Newtonian mechanics two simultaneous events can be specified only in terms of its spatial separation, the
simultaneity is now a relative concept, and the events are now characterized in terms of its spacetime interval:

s2 := t̃2 + z̃ 2 = t2 + z 2 , (1.9)

invariant under Lorentz transformations. The Lorentz transformation can also be derived by starting with the invariance
of the spacetime interval.
Now consider three inertial frames with two relative velocities 12 and 23 . The relation between one and three frames
should be described by the Lorentz transformation, which leads to

12 + 23
13 = . (1.10)
1+ 12 23

Note that this relation to the lowest order is just an addition of two velocities. Furthermore, the Lorentz transformation
allows only three possibilities, strictly separated by the speed of light = 1: sub-luminal, luminal or super-luminal in all
inertial frames.

• Manipulations of the Lorentz transformation.—: The Minkowski metric is invariant under the Lorentz transformation
(by construction):

⌘µ⌫ = ⌘↵ ⇤↵ µ ⇤ ⌫ , ⇤ ⌘ ⇤↵ , ⌘ ⌘ ⌘µ⌫ 7! ⌘ = ⇤t ⌘⇤ , (1.11)

where the Lorentz matrix is symmetric. Using the transformation of the inverse Minkowski metric and the identity matrix
⇣ ⌘
µ
⌘ ↵ = ⇤↵ µ ⇤ ⌫ ⌘ µ⌫ , ↵
⌘ ⌘ ⌘ ↵ = ⇤↵ µ ⌘ ⇤ ⌫ ⌘ ⌫µ ⌘ ⇤↵ µ ⇤ 1 , (1.12)

we derive the expression for the inverse transformation

1 1 µ 1
⇤ ⌘ (⇤ ) ⌫ = ⌘⇤⌘ . (1.13)

In Minkowski space, the indices are raised and lowered by ⌘:

1
⌘ := ⌘ µ⌫ = ⌘µ⌫ = ⌘ . (1.14)
1
Since the un-tilde coordinate is moving with in the opposite direction relative to the tilde coordinate, we can readily derive the inverse
transformation relation:
t = (t̃ + z̃) , z = (z̃ + t̃) . (1.7)

3
AST511 General Relativity JAIYUL YOO

Furthermore, we can also define two variants of the Lorentz matrix, satisfying

⇤ µ := ⌘ ⌫ ⇤⌫ µ , ⇤ ⇢
:= ⇤ µ ⌘ µ⇢ , ⇤ ⇢⇤ ⌫ ⌘ ⇢
⌫ , (1.15)

to find that
) (⇤ 1 µ
) ⌫ ⌘ ⇤⌫ µ . (1.16)
Note that ⇤µ⌫ is not symmetric, while ⇤µ ⌫ is symmetric; the Lorentz matrix is not orthogonal or unitary:

⇤ = ⇤t , orthogonal : ⇤ 1
6= ⇤ , unitary : ⇤ 1
6= ⇤† . (1.17)

Unitary matrices preserve the amplitude of complex vectors, and real unitary matrices are orthogonal.
Physical phenomena are described by tensors (vectors and scalars), and the relation between two frames with relative
velocity is described by the Lorentz transformation ⇤:

p̃µ = ⇤µ ⌫ p⌫ . (1.18)

A co-vector pµ can be obtained by lowering the index with ⌘µ⌫ , and hence it transforms with the inverse Lorentz matrix:

pµ := ⌘µ⌫ p⌫ , p̃µ = ⌘µ⌫ p̃⌫ = ⌘µ⌫ ⇤⌫ ⇢ (⌘ ⇢ p ) = ⇤ 1

µp = ⇤µ p . (1.19)

The notation convention is constructed to facilitate the manipulation. Note that only the four vectors and tensors transform
with the Lorentz matrix; one cannot apply the Lorentz transformation to three spatial vectors.

1.1.2 Length Contraction and Time-Dilation

• Length contraction.— Consider an observer at rest in a tilde coordinate, but moving relative to an un-tilde coordinate,
and consider a small stick with length Lp , described by the observer as located in coordinates z̃1 and z̃2 with Lp := z̃2 z̃1 .
The proper length Lp is the length measured by the observer in the rest frame with the stick. Another observer in a un-
tilde coordinate would then see the stick as located in z1 and z2 , with these coordinate points transforming according to
Eq. (1.8) as
z̃1 = (z1 t1 ) , z̃2 = (z2 t2 ) . (1.20)
At some time t, this observer measures the length L of the stick by measuring two end points z1 and z2 at the same time
(t = t1 = t2 ):
1
L̃p = z̃2 z̃1 = (z2 z1 ) = L , L = Lp  Lp . (1.21)

The length of the stick is smaller than that in the rest frame. This is not an illusory effect, but a real effect due to different
simultaneity.

• Time dilation.— Consider instead a clock at rest in a tilde coordinate, and emits signals at t̃1 and t̃2 to indicate that the
clock is ticking:
t1 = (t̃1 + z̃0 ) , t2 = (t̃2 + z̃0 ) . (1.22)
The proper time interval dtp := t̃2 t̃1 is the time interval measured by the observer in the rest frame with the clock. The
observer in a un-tilde coordinate would then see the clock tick slower as

dt = t2 t1 = dtp dtp . (1.23)

Again, the time dilation is not an illusory effect. For example, muons decay with half-life ⌧ ⇡ 2.2 µ sec. When accelerated
to 1 in particle accelerators, the change in their half-life can be measured. Also, muons are produced at the
atmosphere from a collision with high-energy cosmic rays. With 2.2 µ sec, muons can only travel 0.6 km, even at
the speed of light, but those muons created in the atmosphere hit the Earth surface all the time, due to time dilation.

4
AST511 General Relativity JAIYUL YOO

1.1.3 Transformation of Velocities and Accelerations

Consider a particle moving with some velocity v i = dxi /dt, and consider two observers with relative velocity , measur-
ing the particle velocity in their own rest frames. Using Eq. (1.8), we can readily derive
f
dz vz f
dx vx
ṽz = = , ṽx = = , (1.24)
e
dt 1 vz e
dt (1 vz )
where the relation for vy is similar. The velocity vz along the direction of the moving observer is to the leading order just
a Galilean transformation. The perpendicular direction is also affected due to the change in dt.
Instead of three velocity v i = dxi /dt, we consider a four velocity in terms of proper time
dxµ
uµ := = , vi , 1 = ⌘µ⌫ uµ u⌫ = 2
+ 2 i i
vv , (1.25)
d⌧
where for uµ is one for the moving object with v i . We can show that the four velocity transforms as a four-vector under
the Lorentz transformation.
ũµ = ⇤µ ⌫ u⌫ , (1.26)
which implies
˜ = ⇤0 , ṽ i = ⇤0 1 ⇤i 0 + ⇤i j v j , (1.27)
where we defined
⇤0 := ⇤0 0 + ⇤0i v i = (1 · v) . (1.28)
To put it in a vector form, we obtain the general expression
!
v· ˆ ˆ + v (v · ) ,
ˆ ˆ (1 v 2 )2
ṽ = 1 ṽ 2 = . (1.29)
1 ·v (1 · v) (1 · v)2

In the same way, we consider an acceleration ai = dv i /dt of a particle, measured by two observers with relative
velocity . In the same way we derived the transformation of the velocity, we first compute

fz = 1
dv
dvz
, fx =
dv
dvx
+
vx dvz
, (1.30)
2 (1 vz )2 (1 vz ) (1 vz )2
and then obtain the transformation of three-acceleration
az ax v x az
ãz = 3 (1
, ãx = + . (1.31)
vz )3 2 (1 vz )2 2 (1 vz )3
Similarly, the relative velocity between two observers affects not only the parallel component az , but also the perpendicular
component ax , ay . One can also compute the four acceleration: aµ := duµ /d⌧ .
In the Newtonian dynamics (or Galilean transformation), the position and the velocity are relative, while the time and
the acceleration are absolute (or identical) in all inertial frames. In special relativity, the time (or simultaneity) is now
relative, but the acceleration is still absolute in all inertial frames, while its value is not invariant for all inertial observers,
i.e., non-vanishing accelerations are observed by all inertial observers, which is not true in general relativity.

1.1.4 Uniform Acceleration

Consider two observers with relative velocity (in fact infinitely many observers) and a moving object that is uniformly
accelerating. Mind that the amount of acceleration depends on observers. As a special case, consider an observer moving
with , which is identical to the velocity of the moving object at a given moment (note of the observer is constant,
while v of the object is changing). At this instance, the object is at rest in the observer rest frame, and the observer
measures the acceleration:
=v, ) ã = 3 a , (1.32)
and this acceleration ã is constant. This uniform acceleration would correspond to a situation, in which a rocket is
constantly accelerating in its rest frame by ejecting the same amount of rocket propellants. Note that v? = a? = 0 in this
special case.

5
AST511 General Relativity JAIYUL YOO

At a moment later, the rocket moves ahead of the observer, so we have to consider another observer, who is initially
moving faster and ahead of the object. The object will soon catch up this observer, when their velocities are identical and
this observer will again measure the acceleration. Note that with ã constant, the acceleration a in an un-tilde coordinate
continuously decreases as the speed of the object increases, and the object will never reach the speed of light. Note that
the acceleration is not bounded, while the speed is. The acceleration is a dimension of inverse length, such that when
the acceleration is stronger (or the length of interest is larger than the scale of acceleration), (general) relativistic effects
become more important (like curvature). For an earth gravity, the length scale of the acceleration is a light year. In
comparison, a typical acceleration in nucleus is a ⇠ v 2 /r ⇠ 1028 earth gravity. Gravity is very weak on Earth!
Now we derive the trajectory x(t) of the uniformly accelerating object. First, consider the identities
!
d d 1 d
= p = 3 , v= , (1.33)
dt dt 1 2 (t) dt

hence we derive
d( ) d d d 2 2 d
= + = +1 = 3 = ã . (1.34)
dt dt dt dt dt
Given the constant acceleration ã along the trajectory, we can integrate the above equation to obtain the solution for the
velocity (t) as
ã t ã t
= , ) (t) = p 1, (1.35)
( ) 1 + ã2 t2
and one more integration of the solution yields the trajectory of the uniformly accelerating object with boundary condition
= 0 at t = 0:
1
x2 t2 = 2 := L2 . (1.36)
ã
The uniformly accelerating object follows a hyperbolic trajectory i.e., it starts at x = 1 in the past t = 1, moving
almost at the speed of light toward x = 0, but with constant acceleration ã toward positive direction (or slowing down);
it reaches the closest point x = L at t = 0 and the velocity at the moment is zero; it then turns around and moves to
x = 1 again with increasing velocity. The trajectories are bounded by two light cones that pass through x = t = 0, such
that anything in the upper quadrant cannot be seen for those objects, acting as a horizon. This trajectory of a uniformly
accelerating object is used to describe Rindler coordinates.

1.1.5 Null Vectors: Photon Wavevectors

The photon propagation is described by a null vector k µ , and it is fully characterized in terms of its frequency ⌫, direction,
and polarization:
k µ = ! 1, ni , 0 = kµ k µ , ! := 2⇡⌫ , (1.37)
where ni is a unit observed direction, opposite to the propagation direction. Consider a light source with relative velocity
with respect to an observer, and the light emitted from the source is measured by the observer in the rest frame. The
Lorentz transformation k̃ µ = ⇤µ ⌫ k ⌫ yields the physical relation of what the observer measures to the physical quantities
in the source rest frame:
!˜ = ⇤! ! , ñi = ⇤! 1 ⇤i 0 ⇤i j nj , (1.38)
where we defined (which is similar to ⇤0 )

⇤! := ⇤0 0 ⇤ 0 i ni = (1 + · n) . (1.39)

The ratio of angular frequencies is the relativistic Doppler effect:

!
˜ ⌫˜
= = = ⇤! . (1.40)
! ⌫ ˜
For the observed direction, we can express the transformation relation in a vector form:
!
n· ˆ ˆ + n (n · ) ,
ˆ ˆ
ñ = (1.41)
1 n· (1 n · )

6
AST511 General Relativity JAIYUL YOO

where two frames are moving with relative velocity . The observed angle n in one frame is then related to the observed
angle ñ in the other frame. Note that some formulas are expressed in terms of the propagation direction n, instead of
the observed direction.
As an example, consider a source moving approaching the observer along z-direction with photons emitted with an
angle ✓. For convenience, the emission direction is n = (sin ✓, 0, cos ✓) confined in x-z plane ( ⌘ 0). In the tilde
coordinate, the observer frame is moving with = ẑ against the source and measures
cos ✓ + sin ✓
⇤! = (1 + cos ✓) , cos ✓˜ = , sin ✓˜ = , (1.42)
1 cos ✓ (1 cos ✓)
where in our notation ✓˜ is the angle the observer measures (not the observed propagation direction). For ✓ = 0,
s
1+
⇤! = ⇡ 1 + + O( 2 ) , ✓˜ = ⇡ , (1.43)
1

which reduces to the classical Doppler effect in the non-relativistic limit. The frequency is blueshift (✓˜ is the observed
direction). For ✓ = ⇡/2, we obtain
1
⇤! = , cos ✓˜ = , sin ✓˜ = . (1.44)

If the source is moving relativistically 1, the observed angle becomes narrow, called relativistic beaming:

˜ ' 1
|✓| , (1.45)

beyond which the object is hardly visible, while highly boosted along the direction. If the source is moving away from the
observer, the only difference is that we have to flip the emission direction. Now consider a source moving perpendicular
to the light propagation direction:
1
⇤! = = p ⇡ 1 + O( 2 ) , (1.46)
1 2

and this purely relativistic effect is called the transverse Doppler effect.
Further details in astrophysical applications of the relativistic effects can be found in Rybicki and Lightman (1979).

• Perpendicular components.— Now we define a projection tensor P ij = ij ni nj with ni and consider how the
perpendicular components of a vector Aµ orthogonal to a null vector k µ transform. Given the constraint 0 = Aµ k µ , we
can write
Aµ =: Ak , Ai? + ni Ak , Ai? = P ij Aj , Ai? ni = 0 , (1.47)
and in the tilde coordinate the expression takes the same form. From the transformation law
⌘µ⌫ = ⌘⇢ ⇤⇢ µ ⇤ ⌫ = ⇤0 µ ⇤0 ⌫ + ⇤ i µ ⇤i ⌫ , (1.48)
we can derive useful relations
2
⇤i 0 ⇤i 0 = 1 + ⇤0 0 = 1+ 2 2
, ⇤ j 0 ⇤ j i = ⇤0 0 ⇤ 0 i = 2
i , (1.49)
k k 0 0 2
⇤ i⇤ j = ij + ⇤ i⇤ j = ij + i j . (1.50)
Under the transformation, the perpendicular components transform as
Ãi? = (⇤? )i j Aj? , (1.51)
with
⇤i 0 ⇤ i k nk 0 ⇣ ⌘
(⇤? )i j := ⇤i j ⇤ j = ⇤i j ⇤k 1 ⇤i 0 ⇤ i k nk ⇤ 0 j , (1.52)
⇤0 0 ⇤ 0 l nl
and we can show that
i
X̃? Ỹ?i = (⇤? )i k X?
k
(⇤? )i l Y?l = X?
i
Y?i , (1.53)
where we need to replace all the summed ⇤’s, e.g., µ ⌫ by using the above relations and use the orthogonality 0 =
⇤i ⇤i
i n = Y i n . The derivation is relatively straightforward. The inner product of X is invariant under the transformation.
X? i ? i ?

7
AST511 General Relativity JAIYUL YOO

1.2 Some Paradoxes in Relativity

1.2.1 Twin Paradox
A twin sets a journey to a remote star by accelerating the rocket and maintaining a constant speed for most of the journey
and decelerating just before arrival. Once arrives at the star, the twin reverses the engine and comes back to the other twin
at Earth, finding that the person at Earth has already died old.
It is called the paradox, because one can switch the argument by saying things are relative, such that one at Earth is
moving relative to the one on the rocket. However, the outcome is not symmetric. This is the case due to the acceleration,
no matter how short the period of acceleration can be made.
Check the diagram for the simultaneity in § 3.9 D’Inverno (1992) or p.88 in C. Will 2005.

1.2.2 Causality & Tachyons

In case one can move faster than the speed of light, how is the causality violated? Even though one can move faster,
nothing strange happens in its rest frame. However, to other time-like observers, the sequence of events is not preserved.
For example, at a point A, Alice and Bob are chatting, sending light signal of this situation to Charlie at a point B. As
soon as the signal is sent, Alice and Bob move to B faster than the speed of light, and all three can meet together at B,
before the signal from A arrives, at which point Alice and Bob exist both A and B.
Particles moving along a spacelike geodesic are called Tachyons. Their existence violates the causality. For example,
an observer A sends a signal via a Tachyon to B, who is at another position in a surface of simultaneity of A. Once B
receives the signal, B waits a bit and sends it back to A via Tachyon. In A’s world line, A receives a signal from B first,
before A sends a signal to B. So this is a no-go.
According to special relativity, tachyons with v > 1 have imaginary energy, which is not defined, so that one consider
instead an imaginary mass with real energy. As it gains more energy, the speed is reduced asymptotically to the speed
of light, while at its low energy the speed approaches to infinity. It was shown that tachyons are spin zero, but follow
Fermi-Dirac statistics (hence many issues). Arnold Sommerfeld was the first to conceive tachyons, but its name was given
by someone else later.

1.2.3 Barn-Pole Paradox & Bell’s Spaceship Paradox

• Barn-Pole paradox.— Consider a barn (or a garage), whose size is 10m long on side, and a 20m-long pole. The pole
cannot fit in the barn. However, a fast runner runs with the pole at = 2, such that from the barn’s rest frame, the pole
is length contracted to 10m to be fit in the barn, while from the runner’s rest frame, the barn is length contracted to 5m
too small to be fit. This paradox can be made more interesting, by closing the doors in the barn, when the pole fits the barn.

• Bell’s spaceship paradox.— Originally by Dewan & Beran, and later modified by Bell: In the inertial frame S, there
are two spaceships with separation d at rest, but they are connected by a string. They both accelerate equally at the same
time to some speed in S, hence they maintain the same speed and the same separation at each moment with respect to
the frame S. Therefore, the string should remain unaffected, which is the ordinary point of view. However, due to length
contraction in special relativity, all lengths contract equally, i.e., two spaceships are smaller, and the string is smaller. That
is the paradox. In this lab frame, the string is torn apart due to length contraction. In the rest frame of first spaceship with
the final speed, two spaceships did not accelerate at the same time. The first one accelerates first, and then the second one
later. Hence the distance between two spaceships increased, breaking apart the string. In this frame, the string is broken
by external forces.

1.3 Relativistic Mechanics

1.3.1 Relativistic Mass and Energy
As the length and time measurements depend on observers, the mass measurements should also depend on observers.
Consider two identical particles of mass m0 at rest. One particle is moving with and of relativistic mass m( ), and the
other particle is at rest. After collision, two particles are stuck, moving at u with mass M (u). Assuming the conservation

8
AST511 General Relativity JAIYUL YOO

of the relativistic mass and the momentum, we demand the relativistic mass terms satisfy

m( ) + m0 = M (u) , m( ) + 0 = M (u)u . (1.54)

Removing M (u), we obtain ✓ ◆

u
m( ) = m0 . (1.55)
u
Now consider a center of mass frame, where the combined particles are at rest after collision. This frame has the relative
velocity u, compared to the lab frame. The first particle in the center of mass frame should also move with velocity u,
because the second particle that is at rest in the lab frame moves with velocity u in the opposite direction. Using Eq. (1.10),
we derive
2u 1⇣ p ⌘
= , ) u = 1 1 2 , (1.56)
1 + u2
and the relativistic mass is
m( ) = m0 . (1.57)
Observing that
1
E := m( )c2 ⇡ m0 c2 + m0 ( c)2 + O( 4 ) , (1.58)
2
the conservation of the relativistic mass includes the classical energy conservation, and the total energy of the particle
is E = m0 c2 . The photon is always moving at the speed of light. For any energy E for photons, the rest mass is zero:
E
m0 = =0. (1.59)
c2

1.3.2 Particle Dynamics

We generalize the Newtonian dynamics in special relativity. The energy-momentum four vector pµ and four velocity uµ
are defined as
dxµ dxµ
pµ := m = muµ , uµ := , (1.60)
d⌧ d⌧
where m hereafter represents the rest mass (without subscript 0) and d⌧ is the proper time interval for the particle:
p 1 dxi
i
d⌧ = dt2 dx2 = dt , = . (1.61)
dt
The particle moves on a time-like path, fixing the normalization:

d⌧ 2 = ⌘µ⌫ dxµ dx⌫ , 1 ⌘ ⌘µ⌫ uµ u⌫ , u0 ⌘ , (1.62)

which yields
pi
= i, p0 = m=E, m2 = ⌘µ⌫ pµ p⌫ . (1.63)
E
The Newton’s force-law should then be
d2 xµ
m = fµ , (1.64)
d⌧ 2
which becomes in the rest frame
d2 ⇠ µ µ 0 i
m = frest , frest =0, frest , (1.65)
dt2

with f~rest being the ordinary forces (three vector) in Newtonian dynamics, where we use ⇠ µ to represent a coordinate
where the particle is at rest. Given a force in the rest frame such as E&M force, the four force f µ can then be obtained by
Lorentz transformation:
· frest
f µ = ⇤µ ⌫ frest
⌫
, f i = frest
i
+( 1) i
2
, f0 = · frest = i i
f . (1.66)

9
AST511 General Relativity JAIYUL YOO

1.3.3 Energy-Momentum Tensor

Consider a system of point particles, and we define the energy-momentum tensor as the usual macroscopic way as the
energy density T 00 / ⇢, the energy flux T 0i / pi (or momentum density), and the pressure T ij / pi pj . For a macroscopic
fluid or perfect fluid, the energy-momentum tensor in the rest frame is

T 00 = ⇢ , T 0i = 0 , T ij = p ij
, (1.67)

and Lorentz transform to a general coordinate

1
⇤0 0 = , ⇤↵ 0 = v ↵ , ⇤↵ = ↵
+ v↵v , (1.68)
v2
which yields the energy-momentum tensor
0 2 1
(⇢ + pv 2 ) 2 (⇢ + p)v i
T =@
µ⌫ A = ⇢uµ u⌫ + pHµ⌫ , (1.69)
2 (⇢ + p)v i p + (⇢ + p) 2vivj
ij

where the projection tensor is

Hµ⌫ := ⌘µ⌫ + uµ u⌫ . (1.70)
Given T µ⌫ , we can readily derive

Tµ⌫ = ⌘µ⇢ ⌘⌫ T ⇢ = ⌘µµ ⌘⌫⌫ T µ⌫ , T00 = T 00 , T0i = T 0i , Tij = T ij , (1.71)

µ µ⇢ µ⌫ 0 00 0 0i i i ij
T ⌫ =T ⌘⇢⌫ = T ⌘⌫⌫ , T 0 = T , T i =T = T 0 , T j =T ,

where no summation in the second equality over µ, ⌫. Mind that T µ ⌫ is not symmetric over indices.

Using the energy-momentum tensor, we can define the angular momentum tensor J µ⌫ and the spin four-vector Sµ :
Z
⇢µ⌫ µ ⌫⇢ ⌫ µ⇢ 1
M := x T x T , J := d3 x M 0µ⌫ = J ⌫µ ,
µ⌫
Sµ := "µ⌫⇢ J ⌫⇢ u , (1.72)
2
and the conservation equation guarantees that the angular momentum tensor is constant in time:

@⇢ M ⇢µ⌫ = @⌫ T µ⌫ = 0 . (1.73)

The spin vector is the intrinsic angular momentum that does not vanish in the rest frame:

rest frame : S1 = J 23 , S2 = J 31 , S3 = J 12 , S0 = 0 . (1.74)

1.4 Electromagnetism
Here we provide a concise review of E&M in view of special relativity.

1.4.1 E&M Basics

The physical laws governing the electromagnetism are

qq 0 q
Coulomb : F1 =: k1 , E := k1 2 , (1.75)
r2 r
dF2 II 0 dq k1
Ampere : =: 2k2 , I := , = c2 , (1.76)
dl r dt k2
@B 2k2 I
Faraday : r ⇥ E + k3 =0, B := , (1.77)
@t k3 r

10
AST511 General Relativity JAIYUL YOO

where three dimensionful proportionality constants k1 , k2 , k3 are left unspecified, q, q 0 are electric charges, and I, I 0 are
electric currents (I = dq/dt). The standard units are
1 1 10
cgs units : k1 = 1 , k2 = , k3 = , q = 4.803 ⇥ 10 esu , (1.78)
c2 c
1 7 2 µ0 7
MKS units : k1 = = 10 c , k2 = = 10 , k3 = 1 , q = 1 coulomb ,
4⇡✏0 4⇡
where cgs units are sometimes called Gaussian cgs units. The Maxwell’s equations are

1 @E 4⇡k2 @B
r · E = 4⇡k1 ⇢ , r·B = 0, r⇥B = J, r ⇥ E + k3 = 0 . (1.79)
k3 c2 @t k3 @t
In this note, we use the cgs units, and the Maxwell’s equations in the cgs units become dynamical equations with
source
1 @E 4⇡
r · E = 4⇡⇢ , r⇥B = J, (1.80)
c @t c
and the source-free equations
1 @B
r·B=0, r⇥E+ =0, (1.81)
c @t
in conjunction to the continuity equation and the Lorentz force

@⇢ ⇣ v ⌘
+r·J=0, F=q E+ ⇥B . (1.82)
@t c
The macroscopic field variables are

D = E + 4⇡P ⌘ ✏E , H=B 4⇡M . (1.83)

1.4.2 Field Strength Tensor

Here we work with ( , +, +, +), but compare to (+, , , ) convention with tilde. We consider a four current and a
four vector:

J µ := (c⇢, J) ⌘ J˜µ , Aµ := ( , A) ⌘ õ , Aµ = ( , A) ⌘ õ , (1.84)

and they are related as

Z
@⇢
µ 1 @A
0 = @µ J = +r·J, E =: r , B =: r ⇥ A , Q := d3 x J 0 . (1.85)
@t c @t
The field strength tensor is then defined as

Fµ⌫ := 2A[⌫,µ] = @µ A⌫ @ ⌫ Aµ , F̃µ⌫ := 2Ã[⌫,µ] ⌘ Fµ⌫ , F µ⌫ = F̃ µ⌫ . (1.86)

With ⌘µ⌫ , mind that

F 0i = F0i = F̃0i = F̃ 0i , F ij = Fij = F̃ij = F̃ ij . (1.87)

Using the above relations of Aµ to the electric and the magnetic fields, we derive
1 1
E= F0i = F̃0i , B = ✏ijk F jk = ✏ijk F̃ jk , (1.88)
2 2
where for E and B we do not distinguish their components with upper index or lower index, as it is defined in the
Euclidean space, and note that the Levi-Civita symbols in 3D satisfy

✏↵ = ✏˜↵ , ✏↵ = ⌘ ↵↵ ⌘ ⌘ ✏↵ = ✏˜↵ , 1 = ✏123 = ✏123 = ✏˜123 , ✏˜123 = 1.

(1.89)

11
AST511 General Relativity JAIYUL YOO

In the matrix notation, we derive

0 1 0 1
0 Ex /c Ey /c Ez /c 0 Ex /c Ey /c Ez /c
B Ex /c 0 Bz By C B Ex /c 0 Bz By C
F µ⌫ = B@ Ey /c
C= F̃ µ⌫ , Fµ⌫ =B C = F̃µ⌫ .
Bz 0 Bx A @ Ey /c Bz 0 Bx A
Ez /c By Bx 0 Ez /c By Bx 0
(1.90)
It is convenient to define the dual tensor
0 1
0 Bx /c By /c Bz /c
1 1 B Bx /c 0 Ez Ey C 1
⇤
Fµ⌫ := ✏µ⌫⇢ F ⇢ = ✏⇢ µ⌫ F
⇢
=B
@ By /c
C= ⇤
F̃µ⌫ , ⇤
F̃µ⌫ := ✏˜µ⌫⇢ F̃ ⇢ ,
2 2 Ez 0 Ex A 2
Bz /c Ey Ex 0
(1.91)
where the totally anti-symmetric Levi-Civita symbol in 4D is
✏0123 := 1 , ✏0123 = 1, ✏µ⌫⇢ = ✏˜µ⌫⇢ , ✏µ⌫⇢ = ✏µ⌫⇢ , ✏˜µ⌫⇢ = ✏˜µ⌫⇢ .
(1.92)
The field strength tensor satisfies (Lorentz invariant)
✓ ◆
E2 1 4
µ⌫
Fµ⌫ F = Fµ⌫ F ⇤ ⇤µ⌫
=2 B 2
2
, det F = 2 (E · B)2 , ⇤
Fµ⌫ F µ⌫ = (E · B) . (1.93)
c c c2

1.4.3 Maxwell’s Equations and E&M Action

The Maxwell’s equations with the source terms in Eq. (1.80) become a field equation
@⌫ F µ⌫ = 4⇡J µ , or ⇤Aµ @ µ @ ⌫ A⌫ = 4⇡Jµ , (1.94)
and the remaining source-free equations (1.81) become the Bianchi identity
0 = ✏⇢ µ⌫
@ Fµ⌫ , (1.95)
where 0- and i-components become
1 ijk
0 = @x Fyz + @y Fzx + @z Fxy , 0 = ✏ijk @j Fk0 ✏ @0 Fjk . (1.96)
2
Note that the Bianchi identity can be re-arranged to be the source-free Maxwell’s equation in terms of the dual tensor F̃ µ⌫ :
0 = ✏⇢ µ⌫
@ Fµ⌫ = @[µ F⌫ ] = @µ F̃ µ⌫ , (1.97)
where
1
[abc] := (abc acb + bca bac + cab cba) . (1.98)
3!

• Action.— The Lagrangian for the E&M is and the Noether current (under spacetime translation invariance) is
✓ ◆
1 @L 1
L= Fµ⌫ F µ⌫ + J µ Aµ , tµ⌫ = ⌘µ @⌫ A⇢ + ⌘µ⌫ L = Fµ⇢ @⌫ A⇢ ⌘µ⌫ F⇢ F ⇢ . (1.99)
4 @(@ A⇢ ) 4
However, the Noether current is anti-symmetric, nor gauge-invariant, such that the correct energy-momentum tensor is2
0 1 2 2
1
2 E +B (E ⇥ B)i
i 1
Tµ⌫ = tµ⌫ ( 2iFµ⇢ @ ⇢ A⌫ ) = Fµ⇢ F⌫ ⇢ ⌘µ⌫ F⇢ F ⇢ = @ A , (1.101)
2 4
(E ⇥ B)i ij
2
The correct energy-momentum tensor is obtained by the Belinfante-Rosenfeld procedure:
i
Tµ⌫ = tµ⌫ @ (Sµ⌫ + S µ⌫ S⌫ µ ) , (1.100)
2
where Sµ⌫ is the contribution of the intrinsic angular momentum (see EnM.2016.pdf or Maggiore).

12
AST511 General Relativity JAIYUL YOO

where T00 is the energy density, T0i is the Poynting vector, the spatial components ij are the momentum density, and we
used
1 1 1
⌘µ⌫ F⇢ F ⇢ = ⌘µ⌫ B 2 E 2 , ij ⌘
2
ij E + B
2
Ei Ej B i B j . (1.102)
4 2 2
Note that the energy-momentum tensor of E&M is not conserved
µ⌫
@⌫ TEM = F µ⌫ J ⌫ , (1.103)

but in the presence of E&M (or external force other than gravity) the energy-momentum tensor of (charged) particles is
subject to
@⌫ T µ⌫ = F µ ⌫ J ⌫ , (1.104)
so that the total energy-momentum tensor is conserved.

• Gauge choice.— Finally, in terms of the field strength tensor, the Lorentz force becomes

F = q (E + v ⇥ B) 7 ! F µ = qF µ⌫ u⌫ . (1.105)

The electric and the magnetic fields are invariant under the gauge transformation of the four vector

@
7! , A 7! A + r . (1.106)
@t
Popular choices of gauge condition are as follows:

@
Lorenz : 0 = @µ Aµ = +r·A ! r2 @t2 = @ µ @µ = ⇤ = 4⇡⇢ , (1.107)
@t
Coulomb : 0 = r · A ! r2 = 4⇡⇢ , (1.108)

where Lorenz and Lorentz are different.

• Particle in E&M Field.— The Lagrangian for a charged particle is

1 1
L = mṙ2 e( ṙ · A) , p = mṙ + eA , H= (p eA)2 + e . (1.109)
2 2m

13