Physics 411 Lecture 18

Continuity Equations and the Energy-Momentum

Tensor
Lecture 18
Physics 411
Classical Mechanics II
October 8th, 2007
We have nished the denition of Lagrange density for a generic space-time
described by a metric g

. Now we must carry out the actual change-of-

variables and see how the action changes formally. The fact that the action
cannot change under a change-of-variables will lead to some constraints on
the eld solutions (and their derivatives). This is an example of Noethers
theorem, and we will recover a number of familiar results.
Keep in mind that all we have discussed so far are free (scalar) elds, that is,
functions that are self-consistent but source-less. We have not yet described
the connection between sources and elds, nor the physical eect of elds
on test particles. Both eld sources and eld forces can be put into our
current Lagrange framework, but before we do that, we must nd the explicit
manifestation of coordinate invariance for the elds. For this, we do not need
more than the free eld action.
While the following refers to general eld actions (scalar, vector, and higher),
it is useful to continue our connection to scalar elds. In addition to devel-
oping the energy-momentum tensor and its set of four immediate continuity
equations, we will also look at a Hamiltonian-ization of the Lagrange pro-
gram, an equivalent formulation of the eld equations and one which will
help with some interpretative issues that show up in the energy-momentum
tensor itself.
18.1 Coordinates and Conservation
Returning to our discussion of the coordinate invariance of the action, the
idea is to make a formal, innitesimal coordinate transformation, see what
1 of 7
18.1. COORDINATES AND CONSERVATION Lecture 18
changes (to rst order) that induces in the action S S + S and require
that S = 0, enforcing the overall isometric nature of S. We begin with the
action
S =
_
L(,
,
, g

) d (18.1)
where L =

g

L(,
,
) is the Lagrange density, and we indicate the metric
dependence in L. Now we introduce an innitesimal coordinate transforma-
tion:
x

= x

(18.2)
with

a small, but general perturbation

(x). The coordinate

transformation generates a perturbation to both the eld (x) and the
metric g

(x) lets write these as

(x

) = (x) + (x) and g

(x

) =
g

(x) +g

(x). We can connect these explicitly to the perturbation

(x)
by Taylor expansion take (x

):

(x

) = (x

) = (x + ) = (x) +

x

+ O(
2
) (18.3)
so we would call =

x

. Well return to the analagous procedure for

g

(x

) in a moment.
Viewing the Lagrange density as a function of ,
,
and g

we have
L(+,
,
+
,
, g

+g

) L(,
,
, g

)+
L

+
L

,
+
L
g

,
(18.4)
and it is pretty clear what will happen to the action when we put this in
and integrate by parts well get:
S

[ +] = S[] +
_ _
L

_
L

,
__
d
. .
=0
+
_
L
g

d
. .
S
. (18.5)
Notice that the rst perturbative term above vanishes by the eld equations
indeed, this rst term is just the result of an induced variation of (the
eld equation for comes from arbitrary variation, we are just picking a
particular variation associated with coordinate transformations, so the
eld equations still hold). The second term, labelled S must be zero for S
to remain unchanged. Now we need to connect the change in the metric to
the change in coordinates. It is tempting to simply set
L
g
= 0, but this
is overly restrictive (g

here refers to a specic type of variation for the

metric, not a generic one).
2 of 7
18.1. COORDINATES AND CONSERVATION Lecture 18
Consider the general transformation of a second rank covariant tensor
g

(x

) =
x

(x). (18.6)
If we re-write the transformation in terms of the original variables (so that
we can recover S in its original form), then
g

(x) = g

+
_

,
g

,
g

g
,

_
(18.7)
with both sides functions of x now.
A moral point: we rarely (if ever) leave partial derivatives alone a comma
(as in g
,
) is, as we have learned, not a tensor operation. The whole point
of introducing covariant dierentiation was to generate tensor character, so
it is always a good idea to use it. We are in luck, the metrics covariant
derivative vanishes (by denition, eectively, but we saw this a while ago),
so that:
g
;
= g
,

= 0. (18.8)
The partial derivative on

,
can also be replaced via

;
=

,
+

, (18.9)
and similarly for the

,
. Until, nally, we have
g

(x) = g

;

;
. (18.10)
Evidently, the perturbation is g

= (
;
+
;
) and we are, nally,
ready to return to S.
Referring to (18.5), we can dene a new density built from a tensor T

1
2

g T

L
g

(18.11)
which is symmetric, so that
S =
1
2
_
d

g T

= +
1
2
_
d

g T

(
;
+
;
)
=
_
d

g T

;
.
(18.12)
We have used the symmetric property of T

to simplify (notice how the

strange factor of
1
2
and the minus sign from (18.11) have played a role). Now
3 of 7
18.2. THE STRESS-ENERGY TENSOR Lecture 18
we can break the above into a total derivative, which will turn into a surface
integral that vanishes (by assumption, on the boundary of the integration,

goes away) and an additional term:

_
d

g T

;
=
_
d
_
g T

_
;
. .
=0

_
d

g T

. (18.13)
For arbitrary

(x), then, we have

T

;
= 0. (18.14)
18.2 The Stress-Energy Tensor
Think of a four-divergence in Minkowski space an expression of the form
A

;
= 0 reduces to (in at space, ; , and we take Minkowski in its usual
Cartesian form)
1
v
A
0
t
+ A = 0
A
0
t
= v A (18.15)
with A the spatial components of the four-vector A

. This is a continu-
ity statement in generic language, we say that the zero component is
the conserved charge, and the vector portion the current, terms com-
ing from one of the most common examples: charge conservation where
J

=(c , J
x
, J
y
, J
z
) = (c , J), and then

t
= J (18.16)
(with v = c for E&M). The Lorentz gauge condition can also be expressed
this way, with A

=(/c, A
x
, A
y
, A
z
)
T
, leading to
1
c
2

t
= A. (18.17)
For a second rank tensor in this setting, the statement (18.14) becomes four
equations
0 =
1
v
T
00
t
+
T
0j
x
j
j = 1,2,3
0 =
1
v
T
j0
t
+
T
jk
x
k
.
(18.18)
4 of 7
18.2. THE STRESS-ENERGY TENSOR Lecture 18
If we integrate over an arbitrary spatial volume, and use the usual form of
Gausss law, we can interpret these four equations as continuity equations
as well

t
_
1
v
T
00
d =
_
T
0j
da
j

t
_
1
v
T
0j
d =
_
T
jk
da
k
,
(18.19)
with the obvious identication of a scalar and three-vector in at space.
In order to understand the actual physics of this T

tensor, we will nd
the explicit form in terms of the Lagrange density and see what this implies
about our scalar eld.
18.2.1 The Tensor T

in General
From its denition:
T

=
2

g
L
g

, (18.20)
we need the derivative of the density L in terms of g

. A typical density,
like our scalar eld, will depend on g

through g, the determinant, and

potentially, a metric that is used to contract terms like
,
g

,
. Lets
assume the form is
L(,
,
, g

) =

g

L(,
,
, g

) (18.21)
and note that

g
g
=
1
2

g g

, so
L
g

=
1
2

g g

L +

g


L
g

. (18.22)
The tensor of interest is
T

=
_
g

L + 2


L
g

_
. (18.23)
5 of 7
18.2. THE STRESS-ENERGY TENSOR Lecture 18
18.2.2 Free Scalar Field (Minkowski)
For the free scalar eld, the Lagrange scalar is

L =
1
2

,
g

,
. If we input
this Lagrange scalar in (18.23), we need to evaluate the term
1

L
g

=
1
2

,

,
g

, (18.24)
and the T

tensor is
T

=
,

,
g

1
2
g

,
g

,
. (18.25)
Again, we have the question of interpretation here referring to our original
discrete Lagrangian, from whence all of this came, we can transform to a
Hamiltonian, and this will give us an expression (upon taking the continuum
limit) for what might reasonably be called the energy density of the eld.
The procedure is motivated by its classical analogue where the temporal
derivative is privileged this seems strange in our current homogeneous
treatment, but lends itself to an obvious energetic interpretation.
In the particle case, the canonical momentum associated with the motion of
an individual portion of the string (for example) would be
L

( x
j
,t)
=

=
suggesting that we take, as the canonical momentum for the eld :


L

. Starting from

L =
1
2

,
g

,
=
1
2
_

1
v
2

2
+
2
_
(18.26)
we identify =
1
v
2

, and then

H =



L =
1
v
2

1
2 v
2

2
+
1
2

2
_
=
1
2
_
1
v
2

2
+
2
_
E,
(18.27)
where we dene the energy density E to be the negative of

H (this is just
a matter of the metric signature, nothing deep). Compare this with T
00
from (18.25) that component is
T
00
=
,0

,0
+
1
2
_

2
,0
+
2
,1
_
=
1
2
_
1
v
2

2
+
2
_
= E,
(18.28)
1
Using
g

g
= g

6 of 7
18.2. THE STRESS-ENERGY TENSOR Lecture 18
so that the zero-zero (pure temporal) component of T

is naturally identi-
ed with the energy density of the system.
If we introduce our overall factor (multiply by v
2
in

L) to make contact
with real longitudinal oscillations, then the energy density is (with units)
E =
1
2

2
+
1
2
v
2

2
(18.29)
which is pretty clearly the kinetic and potential energy per unit length (think
of what we would have gotten out of the Hamiltonian for balls and springs).
Now we want to nd the natural momentum density here, and we can work
directly from the energy for a short segment of string E =
_
b
a
E dx as time
goes on, energy will ow into and out of this segment, and we can calculate
the temporal dependence of that ow via:
E(t + dt) = E(t) +
dE
dt
dt =
_ _
1
2
(

(t + dt))
2
+
1
2
v
2
(

(t + dt))
2
_
dx
=
_ _
1
2


(t)
2
+
1
2
v
2

(t)
2
+ dt
_

+ v
2

_
_
dx,
(18.30)
and using the eld equation:

= v
2

, we can write the derivative as:

dE
dt
=
_

x
_
v
2

_
dx (18.31)
which, as a total derivative, can be evaluated at the endpoints as usual. In
terms of the local statement, we have:
E
t
=

x
_
v
2

_
. (18.32)
This is supposed to be related to the T
0x
element of the stress tensor we
expect, from (18.18), that: T
0x
= v

but that is precisely what we

get from (18.25). In addition, we can interpret this component in terms of
momentum density for a patch of eld between x and x + dx, we have
mass displaced from x into the interval: (x), and mass displaced out of the
interval on the right: (x + dx) then the momentum in this interval is
pdx =

((x) (x + dx)) dx

dx (18.33)
and this is related to T
0x
by one factor of v: p =
1
v
T
0x
.
7 of 7