Hamiltonian em Bruno - Murino
Hamiltonian em Bruno - Murino
Hamiltonian em Bruno - Murino
Hamiltonian formulation of EM
Bruno Murino
Last edited
June 19, 2017
i
Abstract
The study of constraints will be of great importance in developing a quantum theory for elec-
trodynamics since the canonical prescription for quantization requires the correct Hamiltonian
formalism of the classical theory. We start by reviewing the Lagrangian formalism and then we
proceed carefully through the Hamiltonian formalism with constraints, as constructed by Dirac.
Finally, we use such formalism to perform the canonical quantization of the electromagnetic free
field on Coulomb’s gauge.
ii
Content
Abstract i
Content ii
References 17
1
1
Constrained Hamiltonian systems
Hamilton’s principle states that true field Ψ is the one for which δS (Ψ) = 0, which
implies
Z
D ∂L ∂L
δS (Ψ) = d x δΨ + δ (∂µ Ψ)
∂Ψ ∂ (∂µ Ψ)
V
Z Z
D ∂L ∂L D ∂L
= d x − ∂µ δΨ + d x ∂µ δΨ (1.1)
∂Ψ ∂ (∂µ Ψ) ∂ (∂µ Ψ)
V V
Z
D ∂L ∂L ∂L
= d x − ∂µ δΨ + δΨ =0
∂Ψ ∂ (∂µ Ψ) ∂ (∂µ Ψ) ∂V
V
2
If we now make the further assumption that δΨ = 0 at ∂V, then for (1.1) to be
satisfied for all allowed δΨ it suffices that
∂L ∂L
− ∂µ =0
∂Ψ ∂ (∂µ Ψ)
and must always be expressed in terms of the field Ψ with its spatial derivatives ∇Ψ, and
its conjugate momentum π with its spatial derivative ∇π. The full Hamiltonian H, defined
as the spatial integral of H, is then
Z
H = dD−1 x H
V
Its important to notice that now the action is a functional of both Ψ and π, not only of
Ψ as before, so we must vary Ψ and π independently. Treating Ψ and π independently,
the equations of motion are obtained through the following: lets start computing δS by
computing its parts separately:
δ (π · ∂0 Ψ) = π · δ (∂0 Ψ) + δπ · ∂0 Ψ
= π · ∂0 (δΨ) + δπ · ∂0 Ψ
= ∂0 (π · δΨ) − ∂0 π · δΨ + δπ · ∂0 Ψ
3
∂H ∂H ∂H ∂H
δH = δΨ + δ (∇Ψ) + δπ + δ (∇π)
∂Ψ ∂ (∇Ψ) ∂π ∂ (∇π)
∂H ∂H ∂H ∂H
= δΨ + ∇ (δΨ) + δπ + ∇ (δπ)
∂Ψ ∂ (∇Ψ) ∂π ∂ (∇π)
∂H ∂H ∂H
∇ (δΨ) = ∇ · δΨ − ∇ · δΨ
∂ (∇Ψ) ∂ (∇Ψ) ∂ (∇Ψ)
∂H ∂H ∂H
∇ (δπ) = ∇ · δπ − ∇ · δπ
∂ (∇π) ∂ (∇π) ∂ (∇π)
Now we can write δS (after collecting the δΨ and δπ terms) as
Z
D ∂H ∂H
δS = d x −∂0 π − +∇· δΨ +
∂Ψ ∂ (∇Ψ)
V
∂H ∂H
∂0 Ψ − +∇· δπ +
∂π ∂ (∇Ψ)
∂H ∂H
− δΨ − δπ = 0 (1.2)
∂ (∇Ψ) ∂V ∂ (∇π) ∂V
Assuming that δΨ = 0 and δπ = 0 at ∂V, then for (1.2) to be satisfied for all allowed δΨ
and δπ is sufficies that
∂H ∂H δH
∂0 Ψ = −∇· = (1.3)
∂π ∂ (∇π) δπ
and
∂H ∂H δH
∂0 π = − +∇· =− (1.4)
∂Ψ ∂ (∇Ψ) δΨ
{F, {G, H}PB }PB + {G, {H, F }PB }PB + {H, {F, G}PB }PB = 0
With the above definition we can now take {F, H}PB to find
Z
3 δF δH δF δH
{F, H}PB = d x −
δΨ δπ δπ δΨ
Z
3 δF δF
= dx ∂0 Ψ − (−∂0 π)
δΨ δπ
Z
3 δF δF
= dx ∂0 Ψ + ∂0 π
δΨ δπ
also, lets determine the variation of a functional F with respect to time, meaning δΨ =
∂0 Ψ δt and δπ = ∂0 π δt:
Z
∂F 3 δF δF
δF = δt + d x ∂0 Ψ δt + ∂0 π δt
∂t δΨ δπ
and since time is a scalar parameter it follows that variations with respect to it are the same
as derivatives with respect to it, which implies that
δF dF ∂F
≡ = {F, H}PB +
δt dt ∂t
and now, noticing that
δfi (x)
= δij δ(x − y)
δfj (y)
it follows that Hamilton’s equations can be stated as
statement is very important since it shows that, indeed, the Lagrangian must hold every
relevant information about the physical system, including the contraints. There are different
kinds of constraints but all of them are represented as a function φ of the field Ψ and its
conjugate momentum π such that
φ(Ψ, π) = 0
The constraints originated exclusively from the form of the Lagrangian are called
primary constraints, and, of course, they must not vary with time. In the presence of
primary constraints we don’t have a uniquely determined Hamiltonian since we take the
ordinary Hamiltonian H and add to it any combination of the M primary constraints,
which means
H ∗ = H + cm φm , m = 1, ..., M
φm ≈ 0
H∗ ≈ H
constraints, which is
δ (cm φm ) = δ (cm ) φm + cm δ (φm )
∂cm ∂cm ∂φm ∂φm
= δΨ + δπ φm + cm δΨ + δπ
∂Ψ ∂π ∂Ψ ∂π
∂cm ∂φm ∂cm ∂φm
= φm + cm δΨ + φm + cm δπ
∂Ψ ∂Ψ ∂π ∂π
Now comes another interesting thing: irrespective of φm being 0, its partial derivative is
not. For instance, if we take the previous example φ = y − x = 0, we find that ∂φ
∂x
= 1. So
we can impose the constraints φm = 0 now, and obtain the weak equality
∂φm ∂φm
δ (cm φm ) ≈ cm δΨ + cm δπ
∂Ψ ∂π
and such variation results in the following Hamilton’s equations
δH ∂φm
∂0 Ψ = + cm
δπ ∂π
and
δH ∂φm
∂0 π = − − cm
δΨ ∂Ψ
which enable us to write
∂0 F ≈ {F, H ∗ }PB
and notice that {F, φm }PB didn’t vanish even with the constraints imposed since the
Poisson bracket consists of some partial derivatives (that don’t necessarly vanish, as we
saw on the example).
7
and since the constraints must not vary with time, we find the following
Several things can now happen, the first being an inconsistency: we could find 1 ≈ 0, and
this will only happen if the Lagrangian itself has inconsistent equations of motion, so lets
assume this is not the case. We could find an identity like 0 ≈ 0, which is fine but doesn’t
add anything new and the modified hamiltonian is enough to account for every possible
constraint. Now, if we notice that {φn , φm }PB can be the components ξnm of a matrix ξ,
the cm the components of a vector c, and {φn , H}PB the components of a vector F, (1.6)
becomes the equation
ξ · c = −F
c ≈ −ξ −1 F
−1
cm ≈ −ξmn {φn , H}PB
and (1.5) becomes
−1
∂0 F (Ψ, π) = {F, H}PB − {F, φm }PB ξmn {φn , H}PB
However, as we can see this only takes the M primary constraints into account, but
there may be more. The (1.6) may lead to more constraints, which we’ll call secondary
constraints denoted by
χs (Ψ, π) = 0, s = 1, ..., S
8
Again, such constraints must not vary with time, so they must satisfy
We can readily notice that we’ll again have analyse the possibilities and may end up with
some more secondary constraints and have to repeat the whole process. This process is
called the Dirac-Bergmann algorithm. What matter is that if we don’t find an inconsistency
or a triviality, then we’ll end up with M primary constraints and K secondary constraints
with the final requirement that the K secondary constraints do not vary with time, and
having ended the Dirac-Bergmann algorithm, we are sure to finally find the cm . Denoting
the secondary constraints as φk with k = M + 1, ..., M + K, and letting M + K = J be
the total number of constraints, then the requirement that they all do not vary with time
is
M
X
{φj , H}PB + cm {φj , φm }PB ≈ 0 with j = 1, ..., J
m=1
Notice that we have J equations and M ≤ J unknowns, and, of course, this must have
a solution otherwise we have an inconsistency, which means the original Lagrangian is
inconsistent, but since we are not considering inconsistent Lagrangians, we’re fine. The
general solution is
A
X
cm = U m + va Vma
a=1
where Um is the particular solution and Vma are the A linearly independet solutions to the
homogenous equation. This means
M
X
{φj , φm }PB Vma ≈ 0
m=1
where
M
X
0
H =H+ Um φm
m=1
9
and
M
X
Φa = Vma φm
m=1
and notice that the v’s are completely arbitrary functions of time. The equations of motion
are, for some arbitraty function F of Ψ and π, simply
dF ∂F ∂F
= {F, HT }PB + = {F, H}D +
dt ∂t ∂t
2
Source free electromagnetic fields
Aµ = (φ, A)
Let
Fµν = ∂µ Aν − ∂ν Aµ
be the so called field strength. We’ll be using Minkowski metric with signature (− + ++),
which means that we can raise and lower latin lettered indices as we wish, but must always
change sign for each zero-th index raised or lowered. Also, of course, we’ll be using natural
units (c = ~ = 1). Then the Lagrangian which gives rise to the source free Maxwell’s
equations is
1
LEM = − Fµν F µν
4
The field E is, then
Ej = −∂0 Aj + ∂j A0 or E j = ∂ 0 Aj − ∂ j A0
Bj = −jk` ∂k A`
11
∂F αβ
∂LEM 1 ∂Fαβ αβ
=− F + Fαβ
∂ (∂ν Aµ ) 4 ∂ (∂ν Aµ ) ∂ (∂ν Aµ )
αβ
1 ∂Fαβ αβ ab ∂F
=− F + ηaα ηbβ F
4 ∂ (∂ν Aµ ) ∂ (∂ν Aµ )
αβ
!
1 ∂Fαβ ∂ η aα ηbβ F
=− F αβ + F ab
4 ∂ (∂ν Aµ ) ∂ (∂ν Aµ )
1 ∂Fαβ αβ ab ∂Fab
=− F +F
4 ∂ (∂ν Aµ ) ∂ (∂ν Aµ )
1 ∂Fαβ
=− F αβ
2 ∂ (∂ν Aµ )
∂Fαβ
= δαν δβµ − δβν δαµ
∂ (∂ν Aµ )
∂LEM
= F µν (2.1)
∂ (∂ν Aµ )
π 0 = 0 := φ1
12
which is a primary constrain, since it came directly from the form of the Lagrangian. The
other momenta are
π i = F i0 = ∂ i A0 − ∂ 0 Ai = E i
where {φ1 , φ1 }PB ≡ 0 since the Poisson bracket is antisymmetric. Notice that this fact
allows us to simply use the ordinary equations of motion for the momenta
µ ∂H i ∂H
∂0 π = − µ + ∂
∂A ∂ (∂ i Aµ )
H = π µ ∂0 Aµ − LEM
for which it follows that ∂H/∂Aµ ≡ 0, meaning that we have to compute only
i ∂H 0 i i i ∂LEM 0 i i ν ∂LEM 0 ∂LEM
∂ = δµ ∂ π − ∂ = δµ ∂ π − ∂ +∂
∂ (∂ i Aµ ) ∂ (∂ i Aµ ) ∂ (∂ ν Aµ ) ∂ (∂ 0 Aµ )
and since the second term of the right hand side is proportional to equation of motion (2.2)
we find
µ 0 i i 0 ∂LEM
∂0 π = δµ ∂ π + ∂ =0
∂ (∂ 0 Aµ )
meaning
2∂0 π 0 = ∂i π i ≈ 0 = ∇ · E
13
∇ · E ≡ ∂i E i = ∂i ∂ 0 Ai − ∂ i A0 = ∂ 0 ∂i Ai − ∂i ∂ i A0 = 0 := φ2
Using
∂i π i , F = ∂i π i , F
PB PB
we find that
∂i π i , H = ∂i π i , H
PB PB
≡0
∂i π i , π 0 = ∂i π i , π 0
PB PB
≡0
Allowing us to conclude that there are no more constraints and thus c1 is arbitrary.
However, we can do something interesting: we can use the primary constraint to
impose a gauge, we just have to change the Lagrangian:
1 1
− Fµν F µν −→ − Fµν F µν − ∂0 A0 (G)
4 4
where G is some function of Aµ such that G ≈ 0. It’s easy to see that now we find
π 0 = −G ≈ 0
so we retained our previous primary constraint, but now we can express it simply as G ≈ 0.
Recall that the secondary constraint that arises from π 0 ≈ 0 is ∇ · E ≈ 0, which we can
write in terms of Aµ as
∇ · E = ∂0 (∇ · A) − ∇2 A0 = 0
{φ1 , φ2 }PB
∂i Ai , H = ∂i Ai , H
PB PB
So lets find everything about the constraints, recalling that they are
φ1 = ∂i Ai and φ2 = ∂i πi
This means that the Dirac bracket between Ai and πj is, finally
−1
{Ai , πj }D = {Ai , πj }PB − {Ai , φ2 }PB {φ1 , πj }PB ξ21
−1
= δij δ(x − y) − (∂i δ(x − y)) (∂j δ(x − z)) ξ12
−1
= δ(x − y) δij − ∂i ∂j ξ12
−1 1
ξ12 =
∇2
as the Laplacian green’s function, so that the final Dirac bracket is
∂i ∂j
{Ai , πj }D = δij − 2 δ(x − y)
∇
or in momentum space
d3 p
Z
pi pj
{Ai , πj }D = δij − eip·(x−y)
(2π)3 |p| 2
{·,·}D −→ −i[·, ·]
∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ ∂ µ Aν − ∂µ ∂ ν Aµ = 0
∂µ ∂ µ Ai − ∂ν ∂i Ai ≈ ∂µ ∂ µ Ai = 0
∂µ ∂ µ A = 0
16
which is just the wave equation for A, meaning that A is a linear combination of the well
known plane waves of the form
= Aei(p )
µx 0x
0 +p·x
A(x) = Aeip µ
A·p=0
which means that given some p, our field has only 2 degrees of freedom r (p), r = 1, 2,
such that
this are known as the polarization vectors. As we now, the general solution to the wave
equation can be writen as a linear combination of plane waves, and in our case, we have 2
plane waves for each polarization vector, and this is true for every p, which means we can
write A as
2
d3 p
Z X
r† −ip·x
r ip·x
A(x) = F (p) r (p) a p e + ap e
(2π)3 r=1
which shows that these coefficients are the creation and annihilation operators.
17
References