Lecture 19

M. Siddikov

Marat.Siddikov@usm.cl

November 7, 2023
Outline
Revision of Special Relativity
−4-vectors in electrodynamics
−Electrodynamics in tensorial notations.
Electromagnetic tensor and Maxwell’s
equations

Certamenes
Certamen 2:
−14/11 10:55-12:05
Certamen 3 (+Examen final):
−28/11 (?)
Electromagentism before Special relativity

Short summary of EM results (theory and experiment, 1905)

Maxwell’s equations (ME) describe all ME are incompatible with Galileo’s
EM very well transformation of coordinates
− no signs of extra terms, motion of ether
− Predict that speed of light c = const Invariant w.r.t. some (strange, purely
−... confirmed by all experiments
mathematical) Lorentz transformations:

x ′ = cosh η x − sinh η ct, ct ′ = − sinh η x + cosh η ct (1)

v /c
sinh η = p
1 − v 2 /c 2
This transformation resembles rotations, yet with “imaginary angle” η (which eventually
gave use hyperbolic functions)
Lorentz transformations (3+1)
Generalization to 3+1 dimensional case:

x02 + x 2 = (x0′ )2 + (x ′ )2 = const

⃗ = (x0 , x) should be linear,
We expect that transformation of components of vector X
akin to rotations in ordinary 3D space:

X µ → Λµν X ν
where Λµν are some matrices. If we had ordinary 4D Euclidean space, we would conclude
that matrices Λ are orthogonal,
ΛT · Λ = 1
However, now we should be careful because x0 = ict , so some “rotation angles”
are purely imaginary. At the same time, we expect that “pure” rotations in 3D are
described by subset of Λ-matrices which correspod to normal, real-valued angles
rotations (!)

We’ll need some additional mathematical concept to be able to formulate carefully

conditions on Λµν .
Minkowsky space
Some mathematical terminology: Some standard notations & definitions:
Minkowski space We’ll use notation aµ (greek index) for 4-
n-dimensional space in which the vector aµ = (a0 , a), and ai (latin index) for
norm of 4D vector is defined as spatial components
Metrics tensor: introduce matrix ηµν =
|a|2 = a02 − a12 − a22 − a32
diag (1, −1, −1, −1) and rewrite |a| in the
form
(some authors use notation a4 ≡ a0 ) |a|2 ≡
X
ηµν aµ aν
µν
In some books might be defined as We’ll call a vector with components aµ ≡
ηµν aµ covariant, and vector with components
|a|2 = −a02 + a12 + a22 + a32
aµ contravariant. In case of Euclidean space
(set of signs is called signature, + − (ηµν = δµν ) they just coincide, but in
− − or − + + +) Minkowski space they differ by sign in front
Mathematical generalization: of ai and thus transform differently.
Pseudo-Euclidean space Einstein’s implied summation over dummy
n-dimensional vector space in which (repetitive index):
the norm of vector is defined as X3 X
|a|2 = aµ aµ ≡ aµ aµ = ηµν aµ aν
|a|2 = a12 + ... + ak2 − ak+1
2
− ... − an2 µ=0 µ,ν
Minkowsky space

Some standard notations & definitions:

Transformation of vectors and covectors is done with different matrices:

xµ → Λµν xν

x µ → Λµν x ν
with Minkowskian orthogonality

Λαµ Λαν = δµν , η αβ Λα, µ Λβ,ν = ηµν ,

Note that α
Λµα (v ) = Λ−1 (v ) µ ≡ Λαµ (−v )

Control question

Assume that we work in n + 1 dimensional pseudoeuclidean space with metrics

|a|2 = a12 − a22 − ... − an+1

2

−How many parameters are needed in order to parametrize the matrix Λµν ?

−How many parameters are needed in order to parametrize the pure spacial rotations
in n-dimensional subspace?

−The difference between the two previous responses corresponds to a number of

pure boosts. Does it coincide with a number of components of velocity v in this
pseudoeuclidean space?
Electromagentism vs Special relativity
Short summary of EM results (theory Albert Einstein, 1905:
and experiment, 1905)
Maxwell’s equations (ME) describe The principle of relativity
all EM very well The laws of physics apply (are equivalent)
− no signs of extra terms, motion of in all inertial reference frames
ether
− Predict that speed of light c = Generalization of Galileo’s principle from
const mechanics to all possible interactions
−... confirmed by all experiments Eliminates the concept of ether
ME are incompatible with Galileo’s
The universal speed of light
transformation of coordinates
Invariant w.r.t. Lorentz transforma- The speed of light in vacuum is the same
tions: in all reference frames.

x ′ = cosh η x − sinh η ct All special relativity is based on these prin-

′ ciples
ct = − sinh η x + cosh η ct
′ ′ The second postulate effectively replaces
y = y, z =z
Galileo’s transform with Lorentz transfor-
v /c mations
sinh η = p
1 − v 2 /c 2
(though this was considered as math-
ematical transform)
Minor technical comments

The principle of relativity

The laws of physics apply (are equivalent) in all inertial reference frames

This principle is valid only in inertial frames. It is NOT valid if we consider non-
inertial frame (accelerated, rotating, etc).

The universal speed of light

The speed of light in vacuum is the same in all reference frames.

Valid only for the speed of light in vacuum. Inside matter (dielectric), the speed of
light
v = c/n < c,
and is NOT constant (depends on frame, velocity of dielectric, ...). We’ll see a bit
later in detail what happens there (and will see in detail Fizeau’s experiment).
Special relativity
Lorentz transformations: −Time is not absolute, depends on frame
x − vt (recall “synchronization” exercise from
x′ = p (4′ ) General Physics courses)
1 − v 2 /c 2
... time intervals are not constant
t − cvx2 ... simultaneity of events in different
ct ′ = p (5′ )
1 − v 2 /c 2 points is not absolute
y′ = y, z′ = z −However, such events cannot be related
by causality (would imply propagation
η ≡ arctanh(v /c)-rapidity . of signal with velocity v > c)
−Lorentz boost cannot change time or-
dering for events which are related by
causality.

Interval (Lorentz invariant)

−Distance between moving objects de-
pends on frame, contracts in longi- Interval between two events (x1 , t1 ) and
tudinal direction: (x2 , t2 ) is defined as

∆x ′ = p
∆x
̸= ∆x I ≡ c 2 ∆t 2 − ∆x 2 ,
1 − v 2 /c 2 ∆x ≡ x 2 − x 1 ,
∆y ′ = ∆y , ∆z ′ = ∆z ∆t = t2 − t1 .
Transformation of the coordinate grid
Rotations: Lorentz boosts:
 Each grid node (x1 , x2 ) keeps its dis-  Each grid node (x0 , x1 ) = (ct, x1 ) keeps
tance from center, its Minkowskian distance from center (in-
terval),
x12 + x22 = const x02 − x12 = const
in the process of rotation, each grid in the process of Lorentz boost rotation,
node remains on some circle (shown each grid node remains on some hyperbola
by green)
Intervals vs Lorentz transformations
Assume we have 2 events, A and B separated by some interval. We’ll choose reference
frame in which A happens at moment t = 0 at the origin of coordinates (x = 0), and
B at some later time

−For any point above blue dashed line (I > 0) can make boost after which axis ct
will pass through it, i.e. the event will happen in the origin of cooridnates

−For any point below blue dashed line (I < 0) can make boost after which axis x̂ will
pass through it, i.e. the event B will happen at moment t = 0, simultaneuosly with
A. We can even choose frame in which B happened before A.
−Any point on the blue dashed line (I = 0) remains there after any boost, i.e. the
light-like separation is invariant in all frames.
Interval vs Causality
For observer making experiment at (t, x):
1) All points inside the light cone going to the past, are separated by time-like intervals,
might be related by causality/affect his measurents
2) All points outside the cone (but with t < 0) are separated by space-like intervals,
cannot be related by causality or affect his measurents. May choose reference frame
in which they happen after the measurement.
Special relativity

Lorentz transformations (aka Inverse Lorentz transformations may be

“boost”): found changing direction of velocity
x − vt v → −v
x′ = p ,
1 − v 2 /c 2
:
t − cvx2
ct ′ = p , x′ + v t′
1 − v 2 /c 2 x= p ,
y′ = y, z′ = z 1 − v 2 /c 2
′
t ′ + vcx2
ct = p ,
1 − v 2 /c 2
y′ = y, z′ = z

Write the Lorentz transformation for arbitrary velocity v , whose direction is not collinear
to x̂, ŷ , ẑ
Special relativity

Write the Lorentz transformation for arbitrary v not collinear to x̂, ŷ , ẑ

We need to split the vector x into parts collinear and orthogonal to v ,

(x · v ) v
x || = (x · v̂ ) v̂ = ,
v2
(x · v ) v
x⊥ = x − x || = x −
v2
 Apply Lorentz transformation for collinear and ⊥ parts:
 
(x · v ) v
x ′|| = γ x || − v t = γ


− v t
v2
(x · v ) v
x ′⊥ = x ⊥ = x −
v2
hx · v i
x ′ = x ′|| + x ′⊥ = x + v 2
(γ − 1) − γt
v
 x ·v
t′ = γ t − 2
c
Lorentz boosts
Lorentz transformations (aka  Note that for pure boosts in limit c = 1
“boost”) in general: matrix Λµν becomes symmetric, Λµν = Λνµ
x ′µ = Λµν (v )x ν , Λµν Λαν = δαµ
 boosts in the same direction commute
Λαν (v ) = η νµ ηαβ Λβµ (v ) = Λνα (−v ) (similar to rotations)
Explicit form:
 boosts in different directions do not
hx · v i
x′ = x + v (γ − 1) − γt commute (similar to rotations)
v 2
 x ·v
t′ = γ t − 2  In general Λ(v 1 )Λ(v 2 ) = Λ(v ) · R̂ where
c
R̂ is some rotation matrix⇒”pure” boosts
1
γ= p do not form a group
1 − v 2 /c 2
Boosts+rotations do form a Lorentz
vi vj
Λij = δij + 2 (γ − 1) , group
v
Λi
Λ0 = −γvi , Λ0i = 20 ,
i
c
Λ00 = γ
Control problems

Assume that we make two consecutive Lorentz boosts with velocities v1,2 in direction
x̂. Demonstrate that the result will be also a boost in direction x̂ with rapidity
v  v 
1 2
η = η1 + η2 = arctanh + arctanh
c c
(i.e. rapidities sum up, as angles in case of rotation around the same axis)

Assume that we make two consecutive Lorentz boost in directions x̂ and ŷ .

Demonstrate, multiplying Lorentz boost matrices

(Λ1 )µν ≡ Λµ
ν (v x̂), (Λ2 )µ µ
ν ≡ Λν (u ŷ ) ,

that the order of multiplication is important, i.e. boosts are not commutative. Also,
prove that in both cases the resulting matrix is not pure boost.
Control problems
We found earlier that Lorentz transofrmation in direction x̂ is given by

x ′ = cosh η x − sinh η ct
ct ′ = − sinh η x + cosh η ct
y′ = y, z′ = z

Let’s disregard coordinates y , z which transform trivially and rewrite this transformation
as
 ′     
x cosh η − sinh η x
=
ct ′ − sinh η cosh η ct

For superposition of 2 boosts have

 ′     
x cosh η2 − sinh η2 cosh η1 − sinh η1 x
=
ct ′ − sinh η2 cosh η2 − sinh η1 cosh η1 ct
  
cosh η − sinh η x
=
− sinh η cosh η ct

where η = η1 + η2 and we simply multiplied 2 colored matrices, using

cosh (η1 + η2 ) = cosh η1 cosh η2 + sinh η1 sinh η2 ,

sinh (η1 + η2 ) = sinh η1 cosh η2 + cosh η1 sinh η2 ,
Control problems
 Explicit multiplication; use c = 1 for simplicity

   
γ(v ) −v γ(v ) 0 0 γ(u) 0 −u γ(u) 0
µ
 −v γ(v ) γ(v ) 0 0  µ
 0 1 0 0 
(Λ1 )ν = 
 , (Λ2 )ν = 
 ,
0 0 1 0  −u γ(u) 0 γ(u) 0 
0 0 0 1 0 0 0 1
 
γ(u)γ(v ) −v γ(v ) −u γ(u) 0
 −v γ(u)γ(v ) γ(v ) γ(u)γ(v )uv 0 
(Λ1 )µα (Λ2 )α
=
ν

 −u γ(u) 0 γ(u) 0 
0 0 0 1
 
γ(u)γ(v ) −v γ(u)γ(v ) −u γ(u) 0
µ α
 −v γ(v ) γ(v ) 0 0 
(Λ2 )α (Λ1 )ν = 
 −u γ(u)

γ(u)γ(v )uv γ(u) 0 
0 0 0 1
From pairwise comparison of the same-color terms, can see that Λ1 · Λ2 ̸= Λ2 · Λ1
From general stucture of Λµν expect that for pure boost Λµν = Λνµ . In our case this is
NOT so, matrices Λ1 · Λ2 , Λ2 · Λ1 are NOT symmetric
vi vj
Λij = δij + 2 (γ − 1) ,
v
Λi0 = −γvi , Λ0i = Λi0 ,
Special relativity
Lorentz transformations (aka  Multiplication of boost matrices in
“boost”) in arbitrary direction: general is tedious, difficult for understand
hx · v
group structure
i
x′ = x + v 2
(γ − 1) − γt
v
x ·v  Expect: for v = 0 Λµν = δν
µ

t′ = γ t − 2
c  Let’s approximate for v ≪ c:
X 2

Λµ µ
ν (v ) ≈ δν + i va (Ta )µ
ν +O v
a

x ′µ = Λµν x ν ,
where (Ta )µ
ν are some matrices which do
NOT depend on velocity, characterize the
vi vj
Λij = δij + (γ − 1) , Lorentz group.
v2
 In Mathematics such matrices (Ta )µ
ν are
Λi
Λi0 = −γvi t, Λ0i = 20 , called generators, play central role in anal-
c
Λ00 = γ ysis of the group structure

1
γ= p
1 − v 2 /c 2
Generators & their properties Examples:
Definition X
Λ−1
1 ≈ 1−i θa Ta
A derivative of the matrix (operator) a
Λ (θ1 , ..., θn ) with respect to one of the Λ2 ≈ 1 + i
X
ϕa Ta
parameters θa at point g = e (θ = 0) : a

1 ∂Λ (θ1 , ..., θn ) so can linearize the product:

Ta =
i ∂θa θ=0 X
Λ1 ·Λ2 ≈ 1+i (θ + ϕ)a Ta +O(θ2 , ϕ2 , θ·ϕ)
−In mathematical books factor 1/i is usu- a

ally omitted; in physical books it is in- For noncommuting groups Λ1 · Λ2 ̸= Λ2 Λ1

cluded to have Ta† = Ta for several phys- −We can’t simply take Λ1 Λ2 − Λ2 Λ1 to
ically interesting groups study the difference, result might be not
Why generators matter? boost or rotation at all
−Close to unit element we may approxi-
mate all elements as: −Define commutator as Λ1 Λ2 Λ−1 −1
1 Λ2 , get
X
Λ1 ≈ 1 + i θa Ta Λ1 Λ2 Λ−1 −1 2
1 Λ2 = 1+i [Ta , Tb ] θa ϕb +O(3)
a

and simplify group multiplications.

∗
For abelian group Λ1 Λ2 Λ−1 −1
1 Λ2 = 1, so
[Ta , Tb ] = 0
Some facts from the theory of Lie groups
Definition of the generator:
1 ∂Λ Lie Algebra (simplified)
Ta =
i ∂θa θ=0 A set of commutator relations between
the generators
where θ is a set of parameters (veloc-
itiesAlgebra
Lie and rotation angles in our case) [Ta , Tb ] = i fabc Tc
A set of linear operators acting on where fabc are called structure functions of
vector space for which we define a the group
bilinear antisymmetric Lie bracket
(“anticommutator”)
Lie’s theorem
[g1 , g2 ] = −[g2 , g1 ] which satisfies
the Jacobi identity: Each Lie algebra g corresponds to a
[[g1 , g2 ] , g3 ] + [[g3 , g1 ] , g2 ] simple connected Lie group G whose
elements are defined via exponential map.
+ [[g2 , g3 ] , g1 ] = 0 If ∃ isomorphism φ : g → h, then there
exists a unique homomorphism f of two
groups f : G → H
Algebra of Lorentz group
Lorentz boost (use c = 1 in this Generators of Lorentz boosts:
slide):  
0 i 0 0
′µ
x = Λµν x ν , (K x )µ =
 i
 0 0 0 
,
  ν  0 0 0 0 
γ −v γ 0 0 0 0 0 0
µ
 −v γ γ 0 0   
Λν (v x̂) = 
 0
, 0 0 i 0
0 1 0   0 0 0 0 
y µ
0 0 0 1 (K )ν =   ,
i 0 0 0 
1 0 0 0 0
γ= √  
1 − v2 0 0 0 i
 0 0 0 0 
Definition of the generator of the (K z )µ
ν = 
 0
,
0 0 0 
Lie group: i 0 0 0

1 ∂Λ
Ta = µ
⇒ (K a )ν = i (δaµ δν0 + δaν δµ0 )
i ∂αa α=0
Generators of rotations (only spacial in-
where α is a set of parameters (veloc- dices participate!):  
ities and rotation angles in our case) 0 0 0 0
a
0 0 −i 0
(J )bc = −i εabc , e.g . : Jz = 
 ,
0 i 0 0
0 0 0 0
Algebra of Lorentz group
Generators of Lorentz boosts:
µ
⇒ (K z )ν = i (δaµ δν0 + δaν δµ0 )

Generators of rotations (only spacial indices participate!):

Straightforward evaluation: (J a )bc = −i εabc .
[K a , K b ] = −i εabc J c
⇒ boosts without rotations do not form closed algebra

[K a , J b ] = i εabc K c , [J a , J b ] = i εabc J c
⇒rotations form closed subalgebra (so(3))
(±)
−If we introduce the generators Ta = (Ja ± iKa )/2, the above-given relations (Lie
algebra) have a form
h i h i h i
(+) (−) (−)
Ta(+) , Tb = iεabc Tc(+) , Ta(−) , Tb = iεabc Tc(−) , Ta(+) , Tb =0

i.e. this algebra is isomorphic to su(2) ⊕ su(2) ∼ so(3) ⊕ so(3)

Control problem

Earlier we found that general Lorentz transformation has a form

hx · v i
x′ = x + v (γ − 1) − γt
v2
′
 x ·v
t =γ t− 2
c
−Write out the infinitesimal transformation (v ≪ c), keeping only linear O(v ) terms.

−Write out the transformation of coordinates under infinitesimal rotation (ϕ ≪ 1),

keeping only linear O(ϕ) terms.

(±)
−Earlier we defined operators Ta = (Ja ± iKa )/2 which allow to rewrite algebra of
Lorentz group as a direct sum su(2) ⊕ su(2) ∼ so(3) ⊕ so(3). Write out how the
P (±)
corresponding coordinates (including time) transform under action of a va Ta .
Algebra of Lorentz group
h i h i h i
(+) (−) (−)
Ta(+) , Tb = iεabc Tc(+) , Ta(−) , Tb = iεabc Tc(−) , Ta(+) , Tb =0

Casimir operator
Operator Cˆ which commutes with all generators
h i
Ta , Cˆ = 0

Each irreducible representation is fully characterized by eigenvalue of Casimirs.

Number of Casimirs=rank of group

E.g. Casimir for rotation group: L2 = L2x + L2y + L2z

Since Lorentz algebra ∼ so(3) ⊕ so(3), expect that it has two Casimirs:
X
2 J2 − K 2 J ·K
C± = Ta± = ±i
a
4 2

⇒Rank of Lorentz group is 2; each irrep of Lorentz algebra is characterized by

eigenvalues of two Casimirs.
Einstein’s velocity addition
Lorentz transformations (aka  If the particle moves with velocity v ′ in
“boost”): some reference frame R ′ , and the refer-
x − vt ence frame moves in direction x̂ of the
x′ = p (4′ )
1 − v 2 /c 2 lab frame with constant velocity V , then
t − cvx2 the velocity of particle in the lab frame is
t′ = p (5′ )
1 − v 2 /c 2 given by
y′ = y, z′ = z dx dx ′ + V dt ′ vx′ + V
vx = = V dx ′ = v′V
Inverse Lorentz transformations: dt ′
dt + c 2 1 + cx2
v → −v
dy ′ 1 − V 2 /c 2 vy′ 1 − V 2 /c 2
p p
dy
x′ + v t′ vy = = ′ = v′V
x= p (4′′ ) dt dt ′ + Vcdx
2 1 + cx2
1 − v 2 /c 2
′ For c → ∞
t ′ + vcx2 ′′
t= p (5 ) vx ≈ vx′ + V , vy ≈ vy′
1 − v 2 /c 2
y′ = y, z′ = z in agreement with Galileo’s principle
For vx′ = c, vy′ = 0 we get vx = c, vy = 0
We could make Lorentz boosts in in agreement with 2nd postulate
any direction (x, y , z) Check what happens for
vx′ = c cos α, vy′ = c sin α
Control problem
 Assume the particle moves with velocity v ′ in direction of x̂ in some reference
frame R ′ , and the reference frame moves in the lab frame with constant velocity
V in direction of x̂. Demonstrate, Using Einstein’s formula, that the velocity of
the particle in the lab frame might be found summing rapidities of the particle in
frame R ′ and of the reference frame, i.e. rapidities add up
For superposition of 2 boosts have
 ′     
x cosh η2 − sinh η2 cosh η1 − sinh η1 x
′ =
ct − sinh η2 cosh η2 − sinh η1 cosh η1 ct
  
cosh η − sinh η x
=
− sinh η cosh η ct

where η = η1 + η2 and we simply multiplied 2 colored matrices, using

cosh (η1 + η2 ) = cosh η1 cosh η2 + sinh η1 sinh η2 ,

sinh (η1 + η2 ) = sinh η1 cosh η2 + cosh η1 sinh η2 ,
Control problem

vx′ + V tanh ηv ′ + tanh ηV

vx = vx′ V
=c = c tanh (ηv ′ + ηV )
1+ 1 + tanh ηv ′ tanh ηV
c2
This finding agrees with our earlier exercise with superposition of 2 boosts:
 ′     
x cosh η2 − sinh η2 cosh η1 − sinh η1 x
′ =
ct − sinh η2 cosh η2 − sinh η1 cosh η1 ct
  
cosh η − sinh η x
=
− sinh η cosh η ct

where η = η1 + η2 and we simply multiplied 2 colored matrices, using

cosh (η1 + η2 ) = cosh η1 cosh η2 + sinh η1 sinh η2 ,

sinh (η1 + η2 ) = sinh η1 cosh η2 + cosh η1 sinh η2 ,
Control problem 2

 Assume the particle moves with velocity v ′ in some reference frame R ′ , and the
reference frame moves in the lab frame with constant velocity V which does not
coincide with x̂ What will be the velocity of the particle in the lab frame ?
Control problem 2

x′ · V x′ · V
   
1
x = x′ + V (γ − 1) + γt ′
, t = γ t′ + , γ= p
V2 c2 1 − V 2 /c 2
h ′ i
dx v ′ + V vV·V2 (γ − 1) + γ
v= = ′
γ 1 + v c·V


dt 2
Einstein’s velocity addition
dx dx ′ + V dt ′ vx′ + V
vx = = V dx ′ = v′V
dt dt ′ + c 2 1 + cx2
dy ′ 1 − V 2 /c 2 vy′ 1 − V 2 /c 2
p p
dy
vy = = ′ = v′V
dt dt ′ + Vcdx2 1 + cx2
 The velocity transformation looks complicated (nonlinear expressions).
-Difficult to work with.

Most of the complexity (in denominators) due to time interval transformation

dt → dt ′ , dx ′

Assume that now we redefine now all velocities as

u µ = dx µ /dτ, (1)

where dτ is the proper time interval (in the rest frame of the partcle)

Interval dτ is:
-Lorentz scalar: time interval in rest frame of a particle
p
-related to lab frame interval dt as dτ = dt 1 − v 2 /c 2
Proper 4-velocity
Example:
Assume that a particle moves along x̂ di-
4-velocity (=Proper velocity)
rection, and its trajectory is given by
! p
dx µ c ⃗v x(t) = a2 + c 2 t 2 , y = z = 0
uµ ≡ = p , p
dτ 1 − v 2 /c 2 1 − v 2 /c 2 (hyperbolic motion in x − t plane). We’ll
evaluate the instant velocity ⃗v and the 4-
The norm of the 4-velocity vector fulfils velocity u µ for this case
uµ u µ ≡ ηµν u µ u ν = c 2
dx c 2t
The 4-velocity transforms as components of: vx = = √ ,
dt a + c 2t2
2
4-vector (same as for coordinates)
vy = vz = 0
u µ → Λµν u ν r
v2 a
Special case: boost in x̂-direction: 1− 2 ≡ √
c a + c 2t2
2

′
u 1 = γ u 1 − vu 0


!
c ⃗v
vu 1
  µ

′
u0 = γ u0 − 2 u = p , p
c 1 − v 2 /c 2 1 − v 2 /c 2
c p 2 
(u µ )′ = Λµν u ν = a + c 2 t 2 , ct
a
Proper 4-momentum
 The particle at rest has (huge!) energy
Covariant 4-momentum
Erest = mc 2
!
mc m⃗v
pµ ≡ m uµ = p , p , Was strange at the beginning of 20th
1 − v 2 /c 2 1 − v 2 /c 2 century, now fully confirmed:
 Decay P of a muon into fragments,
in v ≪ c limit
mµ ≫ mi . Rest frame energy con-
mc

v2
 verts into kinetic energy of fragments
p ≈ mc 1 + 2
(1)
1 − v 2 /c 2 2c
 

1 mv 2 
≈ mc 2 +
 

c  2 
| {z }
kinetic energy Massive particle (m ̸= 0) cannot be ac-
celerated up to v ≥ c, energy E → ∞
m⃗v The ratio |p|/E = v /c 2 , so is we as-
p ≈ m⃗v (2)
1 − v 2 /c 2 |{z} sume that some particle might move with
momentum
v ≈ c (ultrarelativistic limit), then
⇒Identify components of p µ with energy-
momentum, E = |⃗
p| c
mc 2
 
µ E
p = , p⃗ , E = p
c 1 − v 2 /c 2
Covariance of equations
Example:
 Equations of motion (EOM) in nonco-
The principle of relativity
variant (3D) notations.
The laws of physics apply (are equivalent)
in all inertial reference frames d⃗p ⃗
=F (1)
dt
... in covariant 4D notations:
⇓
An equation is said to be Lorentz covari- dp µ µ
= fexternal (2)
dτ
ant if has the same form in any inertial !
reference frame. To guarantee this, we µ f ·v ⃗f
f = p , p
2
c 1 − v /c 2 1 − v 2 /c 2
should use Lorentz covariant quantities
Principle of covariance: All physical laws
Lorentz force:
should be formulated using only those  
⃗ =q E
F ⃗ + ⃗v × B
⃗
physical quantities the measurements of
which the observers in different frames of
-is 3-vector, includes ×, so clearly is NOT
reference could unambiguously correlate. covariant.
=we should NOT mix covariant 4-vectors We need to learn how E ⃗ and B ⃗ trans-
and 3-vectors
Our goal: learn how to write equations form, and how to write Lorentz force F
covariantly. in covariant notations
3D vs 4D covariant notations
Caution:
Levi-Civita εabc is NOT Lorentz-covariant. Should replace it with 4D εµναβ or omit
altogether/rewrite in terms of covariant structures.
 Such definition is Lorentz-invariant

∂x α ∂x β ∂x γ ∂x δ ∂x δ
′ ′ ′ ′ ϵαβγδ = ϵµ′ ν ′ ρ′ λ′ ′
∂x ′µ ∂x ′ν ∂x ′ρ ∂x ′λ ∂x ′λ

′
and Jacobian ∂x δ /∂x ′λ = 1 as (recall proof of phase volume invariance)
−Instead of rotations around axis ϕ speak about rotation in plane
−Have element of a surface (2-form, bivector)

dAµν = dx µ dy ν − dy µ dx ν
formed by vectors dx µ and dy ν instead of “dA = n |dA| where n is normal to dx, dy ”
−Bivector (“antisym. tensor of rank 2”)

Aµ B ν − Aν B µ

instead of vector A × B (recall that [A × B]3 = A1 B2 − A2 B1 , ...)

Example: Angular momentum
Classical mechanics: Relativistic mechanics: can’t use 3D Levi-
M ⃗ = ⃗r × p⃗ Civita (not covariant)
Ma -generalized momentum correspond- Consider transformations
ing to rotations around axis na
x µ → x µ + Ωµν xν , Ωµν = −Ωνµ (1)
∂L
Ma =  If both µ, ν are spacial in a given frame,
∂ ϕ̇a
reproduce rotations
transformation of coordinates and mo-
menta (ϕ ≪ 1):  If one of indices µ, ν is temporal,
r → r + ϕ × r, reproduce boosts
p → p + ϕ × p,
generalization of angular momentum:
in components: X
Mµν = (xµ pν − xν pµ )
δra = ϵabc ϕb rc = Ωac rc , Ωac = −Ωca

r · δr = 0, r 2 ≈ inv + O ϕ2


  Let’s consider tensor in which we contract
Transformation of tensors
Earlier we introduced co- and con- over pair of indices (co- and contravariant):
travariant components of a vector a, να1 ...αn
Tνβ → Λνγ0 Λαγ11 ...Λαγnn ×
1 ...βk
which are related as
δ
ηµν aν = aµ , aµ = η µν aν Λνδ0 Λβδ11 ...Λβk k Tδγ00δγ11...δ
...γn
k

aµ transforms differently from aµ if the two Λ-matrices just cancel in view of or-
we want aµ aµ = const: thogonality:
aµ → Λµν (v )aν ,
Λνγ0 Λνδ0 = δγδ00
aµ → Λµν (v )aν ,
Λµν (v ) = Λµν (−v ) να1 ...αn
Tνβ → Λαγ11 ...Λαγnn ×
1 ...βk

 If we have some tensor Tβα11...β

...αn δ
Λβδ11 ...Λβk k Tνδ
νγ1 ...γn
k 1 ...δk
(e.g. direct product of co- and con-
travariant vectors), then we must add ⇒After contraction we get tensor of a lower
Λµν (v ) for each contravariant Lorentz rank r − 2. For its Lorentz transformation
index, and Λµν (v ) for each covariant need to add Λ-matrices only for free (uncon-
index in transformation law
tracted) Lorentz indices, e.g.:

Aµν bν
Tβα11...β
...αn
k
→ Λαγ11 ...Λαγnn ×
δ
×Λβδ11 ...Λβk k Tδγ11...δ
...γn will transform as contravariant components of
k
vector aµ .
Lagrangian formulation
Landau: We should start from minimal ac- Write out the equations of motion
tion principle, (Euler-Lagrange equations) and find
r generalized momenta pa = ∂L/∂va .
v2
Z Z
S = −mc 2 dτ = −mc 2 dt 1 − 2
c
Evaluate the energy of a particle using
and postulate that for real trajectory S = Legendre’s transform
min.
E =p·v −L
R
We used proper time dτ because in the
rest frame of particles there is no other quan-
tities Your results should coincide with what we
For v ≪ c guessed earlier for p µ = (E /c, p), and

p 2 ≡ ηµν p µ p ν = m2 c 2
r
v2 1 v2
1− 2 ≈1−
c 2 c2
so Generalize the action for a system of
mv 2
Z
S = −mc 2 ∆t + dt noninteracting particles.
2
in agreement with classical mechanics.
Lagrangian formulation
Minimal action for a system of free particles:  Euler-Langrange equations give con-
X Z servation of momentum
S0 = − mi c 2 dτi = X mi v i
i P= p
r i
1 − vi2 /c 2
vi2
X Z
=− mi c 2 dt 1−
c2 Legendre’s transform: energy
i
X
and postulate that for real trajectory S = E = pi · v i − L =
min. i

mi c 2
R
We used proper time dτ because in the X
=
rest frame of particles there is no other quan-
p
i
1 − vi2 /c 2
tities
For v ≪ c For a system of particles E , P = const
P
r (but mi v i is not conserved at all, it
v2 1 v2
1− 2 ≈1− does not have a meaning of momen-
c 2 c2
so tum)
I skip here detailed discussion of rel-
XZ mi vi2
S0 = −mc 2 ∆t + dt ativistic mechanics (I assume You’ve
2
i seen it in undergraduate courses or in
in agreement with classical mechanics. courses of mechanics).
Lagrangian formulation in EM fields

Minimal action for a free particle:

r
v2
Z Z
S0 = −mc 2 dτ = −mc 2 dt 1−
c2
Now we would like to introduce EM fields into our effective action.
We want to keep things covariant.
Before that we want to understand how EM fields E , B transform under Lorentz
boosts (and if it is possible to introduce them in a covariant way in the action S)
Evidence that E , B mix under Lorentz boosts
Magnetic and electric fields mix under boosts
Charge at rest:
⃗ 0 = q R̂
E
4πε0 R 2
Charge moving with constant velocity ⃗v :

⃗ = q 1 − v 2 /c 2 R̂
E
4πε0 1 − v 2 sin2 θ/c 2 3/2 R 2

where θ is angle between R̂ and ⃗v

⃗ (⃗ q 1 − v 2 /c 2 v × R̂
⃗
B r , t) =
4πε0 1 − v 2 sin2 θ/c 2 3/2 R 2

-magnetic field B is generated after the boost from electric field E 0

This example clearly indicates that electric and magntic fields mix with each other, however it is
difficult to extract transformation formulas since vector R as well as time transform nontrivially
from one frame to another
Transformations of EM field
Analysis of simple systems in various frames suggests that fields E , B should transform
nontrivially, mix under Lorentz boosts

(final result for boost in x-direction):

Ex′ = Ex , Bx′ = Bx , Ey′ = γ (Ey − v Bz ) ,

Ez′ = γ (Ez + v By ) ,
 v   v 
By′ = γ By + 2 Ez , Bz′ = γ Bz − 2 Ey .
c c
We will prove validity a bit later, using more general formulation from Landau.
−Can see that E , B do not transform as 4-vectors
−Results are different for || and ⊥ components
Motion in external EM field

Landau: Minimal action for a particle in external field:

r
v2
Z Z
2 2
S = S0 +Sint = −mc dτ +Sint = −mc dt 1− +Sint
c2
postulate that for real trajectory S = min.

Assume that some field is described by 4-vector potential Aµ (x)

Need to add interaction Sint to action S
Result should be covariant (the same in all frames)
For EM fields ∃ superposition principle, fields from different sources sum up as vectors

E = E1 + E2

⇒Sint MUST be linear function of field Aµ [check that for example that quadratic
dependence in interaction term breaks this property]
Motion in external EM field
Let’s assume
Z Z
Sint = −q dxµ Aµ (x) = q dτ (u · A) ,

dxµ
uµ ≡ ; u · A ≡ ηµν u µ Aν
dτ
If we use notation Aµ = (ϕ, A), then
Z Z
ϕ−A·v
Sint = −q dτ p = −q dt (ϕ − A · v )
1 − v 2 /c 2
r
2 v2
L = −mc 1 − 2 − q (ϕ − A · v )
c

Write out the Equations of Motion for such lagrangian and demonstrate that it agrees
with Lorentz formula.
dp
= q (E + v × B)
dt
Motion in external EM field
Claim:
r
v2   dA(x, t) ∂A(x, t) dx b ∂A(x, t)
L = −mc 2 1− ⃗ · ⃗v
−q ϕ−A ≡ + =
c2 dt  ∂t dt ∂x b
∂A(x, t)
Generalized momentum: = + (v · ∇) A
∂t
∂L mv a
Pa = = +q Aa so dp
∂v a
q
v2
1 − c2 = q (E + v × B)
dt
| {z } ∂A
pa E = −∇ϕ − ,
∂t
p a -momentum of free particle; P a ̸= p a B =∇×A
Equation of motion:
Magnetic field does not change kinetic
dP a dp a dAa (x, t) energy:
≡ +q =
dt dt dt dEkin dp
=v· = qE ·v
∂L dt dt
= −q ∇a ϕ + q ∇a (A · v )
∂r a
Claim: A particle moves in constant EM field
∇ (A · v ) ≡ v × [∇ × A] + (v · ∇) A (E , B = const). Find the trajectory of a
relativistic particle
(a×[b×c] = b (a·c)−c (a·b), so b (a·c) =
a × [b × c] + c (a · b))
Motion in uniform electric field

dp 1
q
= q (E + v × B) x(t) = E02 + (qEt + p0x )2
dt qE
∂A
E = −∇ϕ − , -hyperbolic motion in x-direction
∂t
B =∇×A p0y
vy = c 2 q
Case B = 0, select x̂ in direction of E
2
p0y + m2 c 2 + (qEt + p0x )2
 
ṗx = q E , ṗy = 0 p0y c qEt + p0x
y (t) = arcsinh  q 
px = q E t + p0x , py = p0y qE p 2 + m2 c 2
0y

Can solve algebraically to fix vx , vy Trajectory of particle=catenary curve

2
dx pc
 
v= = E0 qEy
x(y ) = cosh
dt Ekin qE p0y c
qEt + p0x -different from parabola in classical theory
vx = c 2 q
2
p0y + m2 c 2 + (qEt + p0x )2
Motion in uniform magnetic field
Evaluate (1′ ) + i (2′ ):
dp V̇ = −i V Ω
= q (E + v × B)
dt
2
∂A where Ω ≡ q c B/Ekin and V = vx + i vy
E = −∇ϕ − ,
∂t
B =∇×A ⇒ V = V0 e −iΩt
iV0 −iΩt
Case E = 0, select ẑ in direction of B ⇒ x + iy = e (3)
Ω
ṗx = q vy B, (1) vx = ℜe (V) , vy = ℑm (V)
ṗy = −q vx B, (2) Similar to nonrelativistic case, particle’s
ṗz = 0, (3) trajectory is helicoidal line (circumference
if v0z = 0)
Recall that Ekin = const and p = Ekin v /c 2 Angular frequency in transverse plane
v̇x = vy q c 2 B/Ekin , (1′ ) qBp
Ω = q c 2 B/Ekin = 1 − v 2 /c 2
2
v̇y = −vx q c B/Ekin , (2 ) ′ m
v̇z = 0, (3) ⇒ vz = v0z Radius of circumference in transverse
plain from (3):
mv⊥
R = v⊥ /Ω = p
qB 1 − v 2 /c 2
Motion in external EM field
Action for a free particle: You should evaluate change of action δS in
case of variation of the trajectory
S = S0 + Sint
r x µ (τ ) → x µ (τ ) + δx µ (τ )
v2
Z Z
2 2
S0 = −mc dτ = −mc dt 1 − 2
c and after that evaluate variational deriva-
tive
Z Z δS
Sint = −q dxµ Aµ (x) = q dτ (u · A) , .
δx µ
dxµ As usual, You may assume that the varia-
uµ ≡ ; u · A ≡ ηµν u µ Aν tions δx µ vanish at the endpoints of inte-
dτ
gration domain
Using minimal action principle, derive the
equations of motion in covariant notations

(assuming that the trajectory is given by

x µ (τ ) and NOT separating explicitly coor-
dinates and time in lab-frame).
Motion in external EM field Variation of integrand of Sint

Action for a free particle: δ (dxµ Aµ (x)) = (dδxµ ) Aµ (x)

∂Aµ (x) ν
S = S0 + Sint + (dxµ ) δx
r ∂x ν
v2
Z
For the terms dτ fµ (...)dδx µ we use in-
R
2
S0 = −mc dt 1 − 2
c tegration by parts

fµ (x)dδx µ ≡ d (fµ (x)δx µ ) − dfµ (...)δx µ

Z Z
Sint = −q dxµ Aµ (x) = q dτ (u · A) ,
∂fµ (...) ν µ
dxµ = d (fµ (x)δx µ ) − dx δx
uµ ≡ ; u · A ≡ ηµν u A µ ν ∂x ν
dτ
Z τ2
Variation of S0
dτ fµ (...)dδx µ = fµ (x)δx µ |ττ21 −
p τ1
dτ ≡ ηµν dx µ dx ν Z
∂fµ (x) ν µ
µ
ηµν dx dδx ν
dx µ − dx δx =
δ dτ = p = ηµν dδx ν ∂x ν
ηαβ dx α dx β dτ Z
∂fµ (x) ν µ
− dτ u δx
= ηµν u µ dδx ν ≡ uν dδx ν ∂x ν
Motion in external EM field
Action for a free particle: Variaton of action:
du ν
Z  
S = S0 + Sint δS = dτ δx µ mc 2 − q F µα uα = 0,
r dτ
v2
Z
2 Fµν ≡ ∂µ Aν − ∂ν Aµ
S0 = −mc dt 1 − 2
c
du ν dp ν
Z Z m ≡ = q F µν uν (1)
Sint = −q dxµ Aµ (x) = q dτ (u · A) , dτ dτ
Fµν is called electromagnetic field tensor
dxµ µ ν
uµ ≡ ; u · A ≡ ηµν u A
dτ Write out explicitly all the components of
Variation of S0 Fµν and demonstrate that (1) is
equivalent to
δ dτ = ηµν u µ dδx ν ≡ uν dδx ν
dp
= q (E + v × B)
Variation of integrand of Sint dt
∂A
E = −∇ϕ − ,
δ (dxµ Aµ (x)) = (dδxµ ) Aµ (x) ∂t
∂Aµ (x) ν B =∇×A
+ (dxµ ) δx
∂x ν
(assume Aµ = (ϕ/c, A),
Aµ = ηµν Aν = (ϕ/c, −A))
Motion µin external
µ
EM field dτ = dt
p
1 − v 2 /c 2
du dp
m ≡ = q F µν uν (1)
dτ dτ c
Fµν ≡ ∂µ Aν − ∂ν Aµ u 0 = u0 = p
1 − v 2 /c 2
Explicit
 form: vj
0 −Ex /c −Ey /c −Ez /c
 u j = −uj = p ,
1 − v 2 /c 2
µν
Ex /c 0 −Bz By  
F =
−Bx  -factors 1/ 1 − v 2 /c 2 appear in both
p
Ey /c Bz 0
Ez /c −By Bx 0 parts and cancel
 Eq. (1) in components (p0 ≡ Ekin /c):
Fµν = ηµα F αβ ηβν = dp0 q q
= (Ex vx + Ey vy + Ez vz ) = E · v ,
dt c c
 
0 Ex /c Ey /c Ez /c
−Ex /c 0 −Bz By 
= dpx
= q (Ex + vy Bz − vz By ) ,

−Ey /c Bz 0 −Bx 
dt
−Ez /c −By Bx 0 dpy
= q (Ey + vz By − vy Bz ) ,
Use ηµν , η µν to raise/lower indices dt
-signs in front of electric field change when dpz
= q (Ez + vx By − vy Bx ) ,
raising/lowering indices, so be careful dt

1 in agreement with results which we got in

Ei = c F0i , Bi = − ϵijk F jk lab-frame
2