QFTI Homework 1 Solution

1 4-vector notation and Maxwell equations

1.1 Problem
Recall that the electric and magnetic fields can be derived in terms of two quantities, the
~ It is also convenient to replace the magnetic
scalar potential Φ and the vector-potential A.
field with the antisymmetric 2-tensor Fij defined as

Fij ≡ ∂i Aj − ∂j Ai (non-covariant index notation). (1)

In terms of this, F12 = B3 , F23 = B1 , and F31 = B2 , e.g.,

ijk
Fij = ijk Bk and Bi = Fjk . (2)
2
Here as usual ijk is the totally antisymmetric symbol (Levi-Civita tensor) with 123 = 1.
In terms of ∂t , ∂i , and these two potentials, write the standard (non-covariant) expressions
for the electric and magnetic fields Ei and Fij in terms of Φ and A. ~
Now we move to 4-vector notation. Define Aµ = (Φ, A) ~ (where µ = 0, 1, 2, 3 and the
notation means that for µ = 0 you choose the first object in the parenthesis and for µ = 1, 2, 3
you choose the component of the second, e.g., A0 = Φ and A1,2,3 = A ~ 1,2,3 . Careful; the 4-
vector objects A1,2,3 are minus the non-covariant components of A.) ~ Also define ∂µ = (∂t , ∂i ).
Introduce
F µν ≡ ∂ µ Aν − ∂ ν Aµ , (3)
and show how F 0i is related to the electric field and how F ij is related to the magnetic
tensor. Here F 0i means F µν for the case where µ = 0 but ν 6= 0–we will use Roman letters
to mean that a Lorentz index µ is not zero.
Also introduce j µ = (ρ, ~j) the 4-current. Show that the covariant equation

∂µ F νµ = [±]j ν (4)

is equivalent to both Gauß’ law and Ampere’s law. Figure out which is the correct sign on
the current; is my ± a + or a −?
Now define µναβ the 4D antisymmetric symbol which generalizes the 3-D Levi-Civita
tensor: νµαβ = −µναβ and similarly for any other permutation of the indices, and 0123 = −1.
[The minus sign is so that 0123 = +1; the sign flips because an odd number of the g µν ’s you
need to raise the indices are negative.]

1
 Show that
µναβ ∂ ν F αβ = 0 (5)
is an identity (is true regardless of what values Aµ take provided they are twice differentiable)
and that this identity is equivalent both to Gauß’ Law for magnetism and to Faraday’s law.

1.2 Solution
Non-covariantly,
Fij = ∂i Aj − ∂j Ai , Ei = −∂i A0 − ∂0 Ai . (6)
Covariantly, A indices should be upper, derivative indices lower; to move a space index from
one to the other introduces a minus sign. Therefore
ij
Fcov = −Fij noncov , F 0i = ∂ 0 Ai − ∂ i A0 = −Ei noncov . (7)

Normally, Maxwell’s equations are

∂i Ei = ρ , ∂j Fij − ∂t Ei = ji . (8)

Covariantly this is
−∂i F 0i = j 0 , −∂j F ij + ∂0 F 0i = j i (9)
or
∂µ F νµ = −j ν . (10)
Now use
µναβ ∂ ν F αβ = µναβ ∂ ν (∂ α Aβ − ∂ β Aα ) . (11)
Rename α ↔ β for the second term and use µναβ = −µνβα to rewrite this as

= 2µναβ ∂ ν ∂ α Aβ = µναβ (∂ ν ∂ α + ∂ α ∂ ν )Aβ (12)

where I used the fact that derivatives commute. Switch ν, α in the second term there, picking
up a − sign and showing the two terms cancel so you get zero.
The time component of this is
ijk ∂ i F jk = 0 (13)
but ijk F jk = ±2Bi , so this is ∇ · B = 0, Gauß’ Law for magnetism. The space component
is
i0jk ∂ 0 F jk + 2ij0k ∂ j F 0k = 0 (14)
which is, noncovariantly,

−2∂t Bi − 2ijk ∂j Ek = 0 = −2(∂t Bi + [∇ × E]i ) . (15)

which is indeed Faraday’s law.

2
2 Condition to be a Lorentz transformation
2.1 Problem
Here we clear up two simple pieces of the derivation of what is and is not a Lorentz trans-
formation.
In class we saw that Λµ ν is a Lorentz transformation if and only if

xα Λµ α gµν Λν β xβ = xα gαβ xβ (16)

for any choice of 4-coordinate xα . Show that this really does require that

Λµ α gµν Λν β = gαβ (17)

should hold. Hint: show that if (17) is NOT true, then there is some xµ such that (16) is
also NOT true. Then argue by contrapositive.
Next, consider
Λµ ν = exp ω µ ν (18)
where multiplication is defined by thinking of the first upper index as a column index and
the second lower index as a row index, and using matrix multiplication, eg,
1 1 1
exp ω µ ν = δνµ + ω µ ν + ω µ α ω α ν + ω µ α ω α β ω β ν + ... (19)
2 6 24
Show that, provided ωµν = −ωνµ , that Λµ ν really is a Lorentz transform, that is, that it
satisfies Eq. (17).

2.2 Solution
Consider the matrix

Aαβ = Λµ α gµν Λν β − gαβ or A = Λ> gΛ − g . (20)

It is a real symmetric matrix. Therefore its eigenvalues are real and its left and right
eigenvectors are the same. Choose xα to be one of its eigenvectors:

xα Aαβ xβ = λ|x|2 . (21)

In order for Eq. (16) to hold for this particular xβ , this has to be zero; since the eigenvector
is nontrivial this requires λ = 0. Applying this reasoning to each eigenvector, we find that all
the eigenvalues are zero, which proves that A is the zero matrix. Therefore Eq. (17) holds.

3
 For the second part of the problem there are many ways to proceed, I will do so by
writing ω µ ν = ω µα gαν so all the indices on ω’s are upper. Then I write it in matrix notation
as (ωg). Using this notation,
∞
1
(ωg)n
X
Λ= (22)
n=0 n!
and we want to show that
∞ ∞
1 1
g = Λ> gΛ = [(ωg)n ]> g (ωg)m .
X X
(23)
n=0 n! m=0 m!

Note that (ωg)> = gω > = −gω since g is a symmetric matrix and ω is an antisymmetric
matrix (this is why I wanted both indices up or down). So the RHS of Eq. (23) is
∞
(−1)n
(gω)n g(ωg)m .
X
(24)
m,n=0 n!m!

The term with m = 0 = n is just g and matches the left hand side. So all we need to
do is to show that, for each r 6= 0, that all the terms with r powers of ω sum to zero. The
terms with r powers of ω are
r r
(−1)n (−1)n
(gω)n g(ωg)r−n = g(ωg)r
X X
. (25)
n=0 n!(r − n)! n=0 n!(r − n)!

The binomial theorem states that

r r
r! xn (−1)n (1 − 1)r
= (1 + x)r
X X
and hence = = 0, (26)
n=0 n!(r − n)! n=0 n!(r − n)! r!

and so all terms with r ≥ 1 powers of ω cancel. Since this is true for all r 6= 0, all terms
except the r = 0 term cancel, which is what we need to show.
A student’s homework presents the simpler approach: The condition ωµν = −ωνµ can be
written in terms of the matrix-notation ω ≡ ω µ ν as ω > g = −gω. Now g 2 = 1 so g = g −1 .
Therefore, multiplying on the right by g, we get ω > = −gωg. Using this,

exp(ω > )g exp(ω) = exp(g(−ω)g)g exp(ω) (27)

Now g(−ω)gg(−ω)g = g(−ω)2 g since g 2 = 1; applying to the definition of the exponent,

exp(g(−ω)g) = g exp(−ω)g ⇒ exp(g(−ω)g)g exp(ω) = g exp(−ω)g 2 exp(ω)

= g exp(−ω) exp(ω) = g (28)

and we are done.

4
3 Why vector fields are Maxwell fields
3.1 Problem
Consider the classical field theory for a 4-vector field Aµ . Assume that the terms in the
Lagrange density have at most two powers of the field and at most two powers of derivatives
(in the spirit of a gradient expansion). Further, L must be a Lorentz scalar.
Recall that any term in a Lagrangian density of form

L ⊃ C∂µ [stuffµ ] (29)

with C a constant can be dropped from L, since its contribution to the action S integrates to
a boundary term and therefore has no physical effect. Such terms are called total derivatives.1
Using the freedom to shift L by such total derivatives, argue that the most general form
for L is
L[Aµ , ∂ν Aµ ] = C1 (∂ν Aµ )(∂ ν Aµ ) + C2 (∂ν Aµ )(∂ µ Aν ) + C3 Aµ Aµ , (30)
with C1 , C2 , C3 some constants. (The linearly independent operators are the ones where
each derivative acts only on one field variable, like Aµ ∂µ ∂ν Aν . You should find 5 independent
2-derivative terms. However you should find 3 independent total derivatives, which allow
you to eliminate three of the 5 2-derivative terms.)
Now use the usual (canonical) procedure to determine Πν , the canonical momentum for
the Aν field:
∂L
Πν ≡ (31)
∂∂0 Aν
(which is the zero component of Πµ ν with µ the “usual” index and ν the index indicating
which field) and then determine the Hamiltonian density H. Write this out–it may be
convenient to do so in non-covariant notation. [You may find it easier to write out H in
terms of derivatives of Aµ , rather than in terms of Πµ . This is correct but is not the form H
must be expressed in if you wanted to get Hamilton’s equations from it.]
Now the punchline: show that C3 6= 0 leads to an H which is unbounded above and
below, meaning that in this case there is some value for the components of Aµ which will
make H arbitrarily large and some different value which will make it arbitrarily negative.
Therefore C3 = 0 is required for the theory to make sense. In addition, show that C2 6= −C1
also leads to an unbounded H, so C2 = −C1 is also required.
Now define
Fµν ≡ ∂µ Aν − ∂ν Aµ . (32)
1
Not to be confused with total differentials d[stuff]

5
Show that, for the special case C2 = −C1 and C3 = 0 we have just found, the Lagrangian
density can be written
C1
L= Fµν F µν . (33)
2
Derive the Euler-Lagrange equations from this action. Show that it is the same as one of
the equations from the last problem, with j µ = 0.

3.2 Solution
The Lagrangian density L must be built out of fields Aµ and derivatives ∂µ . These each
have 1 Lorentz index, and Lorentz scalars involve “tying off” indices in pairs. Therefore
Lagrangian terms must involve 2,4,6,. . . terms (with Aµ and ∂µ each a “term”), at least one
of which must be an Aµ . The two-term objects are

Aµ Aµ , ∂µ Aµ (34)

but the latter is a total derivative. The four-term objects have 1,2,3, or 4 A fields; but of
these the 1A term is a total derivative and we have chosen to drop the 3A and 4A terms since
we consider only Lagrangians at most 2’nd order in the fields (for now). (These terms are
important in Yang-Mills theory which we will see in Phys 673.) I can distribute derivatives
so each derivative acts on an expression with at most one A, for instance writing ∂µ (Aµ Aν )
as (∂µ Aµ )Aν + Aµ ∂µ Aν , in which case the possible terms are

Aµ (∂µ ∂ν Aν ) Aµ (∂ν ∂ ν Aµ ) (∂µ Aµ )(∂ν Aν ) (35)

and the terms written in the Lagrangian. There are however 3 distinct total derivatives:

∂µ (Aµ ∂ν Aν ) ∂µ (Aν ∂ µ Aν ) ∂µ (Aν ∂ν Aµ ) (36)

which I can add to the Lagrangian, so of the 5 terms I wrote only two are linearly independent,
which I can take to be the two in the problem statement.
Now let’s do the problem. The canonical momentum is
∂L
Πµ ≡ = 2C1 ∂ 0 Aµ + 2C2 ∂ µ A0 (37)
∂[∂0 Aµ ]

and the Hamiltonian density is

H = Πµ ∂0 Aµ − L (38)
   
= C1 2∂ 0 Aµ ∂0 Aµ − ∂ ν Aµ ∂ν Aµ + C2 2∂ µ A0 ∂0 Aµ − ∂ µ Aν ∂ν Aµ − C3 Aµ Aµ , (39)

6
which can be written (still including the standard gµν metric but using Roman letters for
indices which are only summed on 1,2,3)
   
H = C1 ∂ 0 Aµ ∂0 Aµ − ∂ i Aµ ∂i Aµ + C2 ∂ µ A0 ∂0 Aµ − ∂ µ Ai ∂i Aµ − C3 Aµ Aµ . (40)

As in the first problem, we go from covariant to 3-dimensional notation by replacing

µ
A = (Φ, A ~ i ) and ∂µ = (∂t , ∂i ), and we will use all lower indices to indicate that we are in
the explicitly 3-d treatment.
First, consider uniform fields, ∂µ Aν = 0. Only the C3 term is then nonzero; it is −C3 (A20 −
A2i ). So long as C3 6= 0, this is unbounded on either side; if A0 6= 0 but Ai = 0, it is of one
sign, if A0 = 0 but Ai 6= 0 it is of the opposite sign. Since A can be as large as we like, the
energy density an be arbitrarily positive or negative if C3 6= 0. Therefore, to make sure the
energy density is bounded from below, we must have C3 = 0.
Now write the remaining terms in noncovariant notation:

H = C1 (∂0 A0 ∂0 A0 − ∂0 Ai ∂0 Ai + ∂i A0 ∂i A0 − ∂i Aj ∂i Aj )
+C2 (∂0 A0 ∂0 A0 − ∂j Ai ∂i Aj ) (41)

To show that this is unbounded, we will consider a few different choices that the field
derivatives could take, and check whether they always return H of the same sign, or if H
can take either sign.
When ∂0 A0 6= 0, you get an energy equal to its square times C1 + C2 .
When ∂1 A1 6= 0, you get an energy equal to its square times −C1 − C2 .
Therefore, unless C1 + C2 = 0, these two field configurations give opposite energies!
By considering arbitrarily large ∂0 A0 or ∂1 A1 , we see that the energy is unbounded unless
C1 + C2 = 0.
However, for this special case, we can rewrite the Lagrangian density as follows:

C1 (∂µ Aν ∂ µ Aν − ∂µ Aν ∂ ν Aµ )
C1
= (∂µ Aν ∂ µ Aν + ∂ν Aµ ∂ ν Aµ − ∂µ Aν ∂ ν Aµ − ∂ν Aµ ∂ µ Aν )
2
C1
= (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ )
2
C1
= Fµν F µν . (42)
2
Writing noncovariantly, we have
C1 h i
L= (−∂i A0 − ∂0 Ai )2 − Fij Fij . (43)
2

7
The Euler-Lagrange equations are
dL dL
∂µ =
d∂µ Aν dAν
µν
2C1 ∂µ F = 0 (44)

which, sure enough, are the Maxwell equations (specifically Gauß’ Law and the Ampere-
Maxwell equation (with vanishing current).

3.3 except that. . .

3.3.1 Problem

Show however that for this case,

1. The field A0 does not have a canonical momentum. Therefore the relation between
derivatives and canonical momenta cannot be inverted and there is no clear procedure
for finding H.

2. The Euler-Lagrange equations do not uniquely determine the evolution of the field.
In particular, assume that Aµ (x, t) is a solution to the Euler-Lagrange equations you
found. Show that Aµ (x, t) + ∂µ Λ(x, t) is also a solution, for any choice of Λ(x, t).
The latter problem will rear its ugly head again some day when we want to do perturbation
theory for electrodynamics.

3.3.2 Solution

Look at Equation (43) above, and note that while ∂0 Ai appears, ∂0 A0 does not (because
F00 = 0 as Fµν is antisymmetric). Therefore ∂L/∂(∂0 A0 ) = 0 and A0 has no canonical
momentum.
What does that mean? A field which appears in the Lagrangian without its time deriva-
tive appearing is actually a Lagrange multiplier; its Euler-Lagrange equation is a constraint
on what initial conditions are permitted, rather than providing dynamical information about
the evolution of the fields. In our case the constraint is Gauß’ Law, which constrains the
initial values of the electric fields (essentially, a constraint on what the first time derivatives
of the remaining fields can be). Gauß’ Law is first order in time derivatives, unlike what we
usually find in equations of motion.
Now assume that Aµ (x) is a spacetime dependent choice of A which gives an F µν which
satisfies the (Maxwell) Euler-Lagrange equations. Define Aµnew (x) = Aµ (x) + ∂ µ Λ(x). The
field strength associated with this is
µν
Fnew = ∂ µ Aν − ∂ ν Aµ + ∂ µ ∂ ν Λ − ∂ ν ∂ µ Λ = F µν (45)

8
that is, it is EXACTLY THE SAME F µν as the choice before adding ∂ µ Λ (since derivatives
commute). But it is F µν which appears in the Euler-Lagrange equations. If one F satisfies
the equations, the other will as well, since they are identical. But Λ was arbitrary (and
in particular could vanish with its derivatives at the current time but be nonzero at some
future time, so the two solutions could be identical for a while and then suddenly start to
differ). Therefore the time evolution of Aµ is not uniquely determined by the Euler-Lagrange
equations, but is unique only up to a transformation Aµ → Aµ + ∂ µ Λ, which is called a gauge
transformation (Eichtransformation).