You are on page 1of 5

Classical Hamiltonian Field Theory

Alex Nelson
August 26, 2009
Abstract
This is a set of notes introducing Hamiltonian eld theory, with focus on the scalar
eld.
Contents
1 Lagrangian Field Theory 1
2 Hamiltonian Field Theory 2
1 Lagrangian Field Theory
As with Hamiltonian mechanics, wherein one begins by taking the Legendre transform of the
Lagrangian, in Hamiltonian eld theory we transform the Lagrangian eld treatment. So
lets review the calculations in Lagrangian eld theory.
Consider the classical elds
a
(t, x). We use the index a to indicate which eld we are
talking about. We should think of x as another index, except it is continuous. We will use the
confusing short hand notation for the column vector
1
, . . . ,
n
. Consider the Lagrangian
L() =
_
all space
L(,

)d
3
x (1.1)
where L is the Lagrangian density. Hamiltons principle of stationary action is still used to
determine the equations of motion from the action
S[] =
_
L(,

)dt (1.2)
where we nd the Euler-Lagrange equations of motion for the eld
d
dx

L
(

a
)
=
L

a
(1.3)
where we note these are evil second order partial dierential equations. We also note that we
are using Einstein summation convention, so there is an implicit sum over but not over a.
So that means there are n independent second order partial dierential equations we need to
solve.
But how do we really know these are the correct equations? How do we really know
these are the Euler-Lagrange equations for classical elds? We can obtain it directly from the
action S by functional dierentiation with respect to the eld. Taking to be a single scalar
1
2 Hamiltonian Field Theory 2
eld (for simplicitys sake, it doesnt change anything if we work with n elds), functional
dierentiation can be dened by
S
(x)
def
= lim
0
1

_
S
_
(y) +
(4)
(x y)
_
S [(y)]
_
(1.4)
where
(4)
(yx) is the 4-dimensional densitized Dirac delta function. Note that we will often
use the shorthand notation
(4)
x
=
(4)
(y x). Applying this to the action yields
S
(x)
= lim
0
1

_
_
L
_
+
(4)
x
,

(4)
x
_
L(,

)
_
d
4
y (1.5a)
= lim
0
1

_ _
L(,

) +
L

(4)
x
+
L
(

(4)
x
+ O(
2
) L(,

)
_
d
4
y (1.5b)
= lim
0
_ _
L

(4)
x
+
L
(

(4)
x
+ O()
_
d
4
y (1.5c)
=
_ _
L

(4)
x
+
L
(

(4)
x
_
d
4
y (1.5d)
=
_ _
L

L
(

)
_

(4)
x
d
4
y (1.5e)
Where we justify the second line by Taylor expanding to rst order, then in the third line we
factor through by the (1/) factor, in the fourth line we take the limit, and integrate by parts
to yield the last line. Note also that we factored out the delta function to make the last line
prettier. Now the last line is zero if and only if
L

a

d
dx

L
(

a
)
= 0 (1.6)
which is precisely the Euler-Lagrange equations of motion!
Why are we working with these delta functions? Well, we are working with something a little more
than just time. We are working with points in space. Locality means, mathematically, we work
with vectors sharing the same base point. Or in the jargon of dierential geometry, we are working in the
tangent space TpM where M is our manifold, and p M is our base point. If we work with multiple base
points at a time, not only is it mathematically not well dened, but it is nonlocal which results in a loss of
causality! Needless to say this is bad, so we try to work with the evolution of the eld at a specied (but
arbitrary) tangent space. If time permits, we will revisit this notion of spatial coordinates as an index in
the appendix.
More precisely, we have the base space be the manifold M representing spacetime. We have our elds
assign to each point p M some physical information (x). The question presents itself Where does this
information live? It lives in a generalization of the tangent space, called the ber. In the Lagrangian
setting, we work with ordered pairs (x

, (x

)) which is actually something called the section of the ber


bundle. That is, the eld is a mapping
: MMF (1.7)
where F is where the elds live, i.e. its the ber.
2 Hamiltonian Field Theory
If we want to begin the canonical treatment of elds, we need to start by nding the canoni-
cally conjugate momenta. We need to rst foliate spacetime into space and time. Why do
we do this? Well, we are concerned about the time derivatives of the eld (how it changes in
time), and we need to foliate spacetime into constant-time surfaces to have this be meaningful
in the obvious way.
2 Hamiltonian Field Theory 3
Let be a (column) vector of scalar elds. So to nd the canonically conjugate momenta
to
a
, we follow our nose to nd

a
=
L
(
0

a
)
. (2.1)
We will write for the row vector = (
1
,
2
, . . . ,
n
). If the mapping
(,
0
,
1
,
2
,
3
) (, ,
1
,
2
,
3
) (2.2)
is invertible (bijective, but when wouldnt it be?
1
), then we dene the Hamiltonian density
function by
H(, ,
1
,
2
,
3
) =
0
L (2.3)
which is a generalization of the Legendre transformation.
By direct computation we can nd Hamiltons equations to be

a
=
H

a
, and

a
=
H

a
+
d
dx
m
H
(
m

m
)
(2.4)
where the equation on the right hand side has an implicit sum over m = 1, 2, 3. These
equations naturally folow from Hamiltons principle applied to
S =
_
(
0
H)d
4
x (2.5)
the action in canonical form.
Example 1. (Working in the East coast convention + ++) Consider the Lagrangian density
L =
1
2

+

2
2

2
(2.6)
where is some mass scalar. We nd the equations of motion being
S

= 0
L

L
()
= 0. (2.7)
We nd by direct computation
L

= +
2
(2.8a)

L
()
= (

) =
2
(2.8b)
which yields the equations of motion to be
`

+
2

= 0. (2.9)
This is precisely the Klein-Gordon equation!
We nd that the canonically conjugate momenta is
=
L
(
0
)
=
0
=
0
. (2.10)
Thus we can nd the Hamiltonian density to be
H =
0
L (2.11a)
=
0

1
2

+

2
2

(2.11b)
=
1
2

0

1
2





2
2

2
(2.11c)
=
1
2

+



+
2

(2.11d)
1
The condition of this being invertible is equivalent to Henneaux and Teiteilboims criteria of regularity
it seems...
2 Hamiltonian Field Theory 4
Note our convention makes the sign of the
2
term dierent than what is conventionally used in most particle
physics texts. Because of this choice of signature convention, our energy is negative.
We can nd Hamiltons equations for the Klein Gordon scalar eld quite easily. One we already found
accidentally, it is
=
L
(
0
)
=
0
=
0
. (2.12)
The other is more interesting, it is

=
H

+
d
dx
m
H
(m)
(2.13a)
=
2
m(
m
) (2.13b)
= (
2
+
2
) (2.13c)
Note that the sign of our results diers from what we derived due to our choice of signature convention!
Now the Hamiltonian approach to elds, as previously mentioned, is slightly diferent
because we work with a one-parameter family of spatial hypersurfaces instead of spacetime.
To consider the scalar eld (x) in this setting, we let it become ( x; t) where we use the
semicolon to remind ourselves that we are working with some parameter t. We obtain the
Hamiltonian simply by integrating the density over all space
H(, ) =
_
all space
H(, ,
1
,
2
,
3
)d
3
x. (2.14)
We need to consider how the functional derivative behaves under this change from spacetime
to space plus time. The functional derivative with respect to ( x; t) is
A[( y; t)]
( x; t)
= lim
0
1

_
A[
t
+
(3)
x
] A[
t
]
_
(2.15)
where
t
( y) = ( y; t) =
t
, and
(3)
x
=
(3)
( x y).
We obtain the equations of motion from the principle of stationary action applied to
the canonical form of the action. The Euler-Lagrange equations encoded in the canonical
formalism are

a
=
H

a
, and

a
=
H

a
+
d
dx
m
H
(
m

m
)
(2.16)
as we previously mentioned.
We can directly compute that
H
( x; t)
=
_
H
( x; t)
d
4
y (2.17a)
= lim
0
1

_
_
H(,
t
+
(3)
x
,
k

t
+
k

(3)
x
) H(,
t
,
m

t
)
_
d
3
y (2.17b)
= lim
0
1

_ _
H(,
t
,
m

t
) +
H


(3)
x
+
H
(
m
)

(3)
x
(2.17c)
+ O(
2
) H(,
t
,
m

t
)
_
d
3
y
=
_ _
H


(3)
x
+
H
(
m
)

(3)
x
_
d
3
y (2.17d)
=
_ _
H


m
H
(
m
)
_

(3)
x
d
3
y (2.17e)
=
_ _

+
m
H
(
m
)
_

(3)
x
d
3
y (2.17f)
References 5
=
_

(3)
x
d
3
y (2.17g)
=

(x). (2.17h)
This is precisely one of Hamiltons equations, and by similar reasoning we nd

=
H

t
( x)
, and

t
( x) =
H

t
( x)
(2.18)
are both generalizations of Hamiltons equations to eld theoretic setting. This motivates us
to dene an analogous Poisson bracket:
{A, B}
def
=
_ _
A

t
( x)
B

t
( x)

B

t
( x)
A

t
( x)
_
d
3
x. (2.19)
This allows us to recover the equations of motion in a familiar way
{A, H} =
_ _
A

t
( x)

t
( x) +

t
( x)
A

t
( x)
_
d
3
x. (2.20)
Similarly, the canonical commutation relations hold classically as
{
a
( x; t),
b
( y; t)} =
a
b

(3)
( x y). (2.21)
Note that this only makes sense if the two variables are considered on the same time slice. If
we were to consider two dierent timeslices, wed necessarily have the commutation relations
vanish due to locality constraints.
References
[1] R. Ticciati, Quantum eld theory for mathematicians,. Cambridge, UK: Univ. Pr.
(1999) 699 p.