You are on page 1of 9

Noethers Theorem: Proof + Where it Fails (Diffeomorphisms)

There is a simple way to state and prove the Noether theorem that allows you to see all the important features the
(symmetric) stress tensor, the force law, the conservation laws ... and most importantly: the places where the theorem
does not apply (i.e. where it actually breaks down), but in which the theorem commonly and naively used.

So, this note is just as much about dispelling some folklore as it is about proving a result.

This is a reformatted, reworked version of an article posted in sci.physics.research some time around 2010 or 2011.

Mark Hopkins

This proof is an adaptation of one posed by Hehl in the early 1990s [1]. Start out by assuming a dynamics given by an
action principle where the action S is an integral
U
U
S L =

of a Lagrangian n -form L , over an n -dimensional


space. Ill use ( ) _
U

to denote the integral over a region U . The Lagrangian is assumed to be a function of the form
( ) , , , L L q v x = ; that is, a Lagrangian of the first order. The field consists of a form-valued configuration
( )
: 0, , 1
A
q q A N = = , its exterior differential v dq =
( )
A A
v dq = . In addition, we also allow for possible
dependence on the spatial coordinates
( )
: 0, , 1 x x n

= = and their exterior differentials dx =


( )
dx

= .

The first-order differential of the Lagrangian may be written as:
( ) 1 .
q
L q f v p x K T = + +
This notation compresses the wedge product and contraction over the coordinate and field indices. For simplicity, were
assuming the forms q and v are of homogeneous degrees, which we denote q and 1 q + . Then f and p are,
respectively, of degrees n q and 1 n q , while K is of degree n (the force density) and T is of degree 1 n (the
stress tensor density). In index form this expression is given by:
( ) 1
q
A A
A A m
L q f v p x K T

= + +
Thus, we write
A
A
q f as q f , etc. The summation convention is used here and below to denote contraction on
the indices.

In the action integral as is generally done the fields are subject to what we may call the kinematic constraint. That
is, we dont actually take the general function ( ) , , , L L q v x = as the integrand, but its restriction
( ) ( ) ( ) , , ,
U
U
S L q x dq x x dx =


in which one imposes the kinematic constraints v dq = and dx = . Also following from these constraints are
1. the Bianchi identities, 0 dv = , 0 d =
2. the inherited constraints on the variationals: ( ) v d q = , ( ) d x = .

Using script notation (e.g., , , , K T P F ) to denote densities,
0 1 n
dx dx dx

= to denote the coordinate n -form, and
0 1 1 1 n
dx dx dx dx dx
+

= to denote the ( ) 1 n hypersurface forms, the force density and stress tensor
density can be written respectively as
, . K dx T dx


= = K T

Example: (Electromagnetism)
An important example includes the 1-form fields which may be expanded in 3+1 Minkowski space form using the
following notations
( ) ( ) , , , , , , d dx dy dz d dy dz dz dx dx dy dV dx dy dz = = = r S
as
, , , . q d dt v d d dt p d d dt f dV d dt = = + = = A r B S E r D S H r J S
This generalizes, for curvilinear coordinates and curved backgrounds to the following

1
, , , .
2 2
q A dx v dq F dx dx p dx f dx


1
= = = = = G J

Historically, Noether (in 1918) [2] posed the Noether Theorem in a form that is, in some ways, more general than
whats described below. Though it was restricted to 0-form valued configuration coordinates (as are most treatments of
the theorem), it included an account of Lagrangians of 2nd or higher degree in the derivatives and functional
Lagrangians. Our development is suitable for natural bundles [4] and uses form-valued quantities, which restricts us to
degree-1 derivatives (since
2
0 d = ).

A notable failing in Noethers paper and one which has been carried on almost by habit since then was the failure to
explicitly include the region of integration in the actual notation (though it was mentioned in the narrative). The
reason this becomes important is that the operation ( ) _

is generally not well-defined unless is compact. The


typical sleight of hand, let fields drop to 0 at asymptotic infinity is used as a compromise on this, but is not
particularly meaningful or relevant when dealing with diffeomorphism symmetry meant to include all of space-time;
i.e. there is no asymptotic infinity in any realistic model of the universe and even if we use the more general definition
(which applies for locally compact spaces) of dropping to 0 at infinity, were still placing restrictions on the set of
functions that may be used for variationals, which may enable important non-trivial results to slip through.

This forces us to
(1) confine our attention to compact regions ,
(2) explicitly indicate how the let everything drop off to asymptotic infinity links up to this (in particular, how a
very important loophole is lost if we dont do this analysis right) and
(3) note that the conserved quantities are not constants, but actually functionals that depend on the shape of .
This last feature is especially important for external symmetries, when is moved. The importance becomes
prevalent when trying to do this in a quantum setting, where moving also means changing the state-space cutoff (i.e.
effecting an incomplete Bogoliubov transformation).

So, heres the proof.

1. Force and Power Laws
The first thing to note is that the variational applies for all selections of as natural derivations [3, 4]. This includes,
most importantly,
x
= L the Lie derivative with respect to a vector field x x

= (where

denotes both the


partial differential operator x

and the corresponding vector field). It also applies to ( ) ( ) _ _ x = the


contraction by the vector field x . (The operator ( ) ( ) _ _ is frequently used to denote contraction, one also sees
x

in the literature to denote contraction by the vector field x ).

Note, that the ( ) 1 n forms can be expressed as dx dx

= . We can also define ( ) 2 n -forms dx dx

= ,
similarly for forms of lower degree.

The special cases


= L L and ( ) _

= give us explicit expressions for the force n -form K and stress
tensor ( ) 1 n -form T . This comes by way of the identities:

( ) ( )
, 0,
0, .
x dx d x d
x dx




= = = =
= =
L L L

Thus, we find the following relations:

( ) ( ) ( ) ( ) ( ) ( )
1 , 1 .
q q
L q f v p K L q f v p T

= + + = + L L L




1.1 Stress Tensor
The second of these relations, upon multiplication by x

and summation over , leads to the following expression


for the stress tensor density:
( ) ( ) ( ) ( ) 1 .
q
x T x dx x q f x v p x L


= = + T
For instance, for the 1-form valued electromagnetic field, with the Maxwell-Lorentz Lagrangian

0 0
1 1
,
4 4
ML ML
L dx c g g g F F c g F F


= = = L L
(where indices are raised with the metric, F g g F

= ) we have the following expressions for the Lagrangian


derivatives

0
, 0.
ML ML
c g F
F A



= = = =

L L
G J
Substituting into the expression for x T , we obtain the following

0
1
4
A F c g F F F F


| |
= = +
|
\
T J G L .
The first of these equalities is generic, and not specific to the Maxwell-Lorentz Lagrangian at all. In Minkowski space
for arbitrary pure field Lagrangians ( ) F = L L it yields:

( )
( ) ( )
( )
0 0
0
0
, ,
, 1, 2, 3 .
, ,
j
j
i
i i i i i
j j j j
i j
D E B H
= =
=
= = + +
D E B D
E H B H
T L T
T T L

When restricting attention to the Maxwell-Lorentz Lagrangian
( )
2 2 2 1
0 2 ML
E c B = L , this reduces to the usual
expressions

( )
( )
2 2 2
0 0 0 0
0 0 1 2 3 0
1 1 1
1 2 3
2 2 2
1 2 3 2 2 2 2 2
0 0 0 0 1 2 3 0
3 3 3
1 2 3
, , , ,
2
, , , ,
2
E c B
E c B
c c
+
= =
| |
| | | +
= = +
| |
\
|
\
B E
E B EE BB
T T T T
T T T
T T T T T T
T T T
I

with the spatial stress tensor written in dyad form, with I as the identity dyad.

1.2. Force Law
If we differentiate the expression for the stress tensor and add it to the expression for the force density, we get

( ) ( ) ( ) ( ) ( ) ( )
1 1
q q
L d L q d q f q df v d v p v dp K dT

= + + + L L L ,
since the action of the Lie derivative
X
L on a form-valued quantity is given by
X X X
d d = + L . Additionally,
since L is an n -form, then 0 dL = and
X X
L d L = L . Therefore, this expression can be reduced to

( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
0 1 1
1 .
q q
q
dq f q df dv p v dp K dT
v f dp q df K dT


= + + + +
= + +

Thus,
( ) ( ) ( ) ( ) ( ) 1 .
q
x K dT x v dp f x q df + = +

Ultimately, the application of the action principle would give us the field law dp f , and the continuity equation
0 df . So, denoting on-shell identities by (i.e. identities that follow from application of the field law and all of its
derivatives, including the continuity equation), it follows that
( ) 0. x K dT +
In component form, weve found: 0.


+ K T

2. Reduction of the Variational
The point of finding expressions for the stress tensor and force density is to remove them from the total variational L .
But first, it pays to perform an integration by parts and then to perform the indicated substitutions.
( ) ( ) ( ) ( ) ( ) 1 , .
q
v p d q p d q p q dp T d x T d x T x dT = = = =
Thus
( )
( )
( ) ( ) 1 .
q
L d q p x T q f dp x K dT = + + +

When used to formulate the Euler-Lagrange equations, this shows that f dp , while the equation involving the
variational x , 0 K dT

+ already follows., independent of its appearance as an Euler-Lagrange equation.

Upon substitution for ( ) x K dT + and x T , we obtain:
( ) ( ) ( )
( )
( ) ( ) ( ) ( ) 1 1 .
q q
L d q x v p x q f x L x v q dp f x q df = + + +

3. Noether Theorems as Continuity Equations
This leads directly to the following result.

Noether Theorem #1: (Total Variations)
Assume the Lagrangian L is quasi-invariant under a given variation , with L dL

= . Then, defining the Noether


current
( ) ( ) ( ) 1
q
J x q f x v q p x L L

+ +
it follows that
( ) ( ) ( ) ( ) 1 0.
q
dJ x v q dp f x q df

= +

To arrive at the second form of the theorem, we can separate out the internal variation and external variation
x
L by
the decomposition
x
= + L and write

Noether Theorem #2: (Internal Variations)
Assume the Lagrangian L is quasi-invariant under internal variations , with L dL

= . Then with the corresponding


Noether current
( ) 1
q
J q p L

+
it follows that
( ) 0. dJ q dp f

=
Unlike Theorem #1, this one does not require the continuity equation 0 df .

Noether Theorem #3: (Diffeomorphisms)
Finally, noting that ( )
x
L L L L L d x L

= + = + it follows that L L x L

= + , so that we can relate the two
currents by
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1 1 .
q q
x
J J J d x q p x q f dp d x q p

= +
The residual
x
J

is the actual Noether current associated with the diffeomorphism generated by x . This is in contrast
to what one may find in the literature, where it is asserted that the stress tensor density T x is the Noether current
associated with the diffeomorphisms generated by x .

So, one important folklore item that is dispelled is this:
The stress tensor T

is not the Noether current for diffeomorphism symmetry, but is the variational of the
Lagrangian with respect to the (negative of the) co-frame dx

= .

4. Noether Theorems as Conservation Laws
The second important folklore item to dispel is the idea that the job is done once the continuity equations are written
down. This is the myth based on a casual conflation of continuity equation and conservation law that is so widespread
that the two are practically regarded as synonymous. Thus, the second important folklore item to be dispelled is:
Continuity Equations are not Conservation Laws.

To pose a conservation law, we first assume the region is compact and that it can be sliced into a set of layers
t
as
t ranges over some interval [ ] , t t
+
, such that
t t
can be generated by a ( ) 1 n -diffeomorphism
t
t

. That is,
is foliated.

Notice, by the way, that this foliation need not be timelike, despite the use of t as the index variable. Nor does it need
to be the case the space even have a time-like dimension. For instance, one can apply this just as well to 3 n = , for 3-
dimensional statics, where t indexes the unfolding of a static configuration in space.

The underlying assumption is that the foliation is generated by a vector field d dt that has compact support. So, each
of the
t
s share a common boundary:
,
t
H =
that is: the horizon. Define the region
b
a
b t
t a =
=

, where a b < . Then the boundary is given by


.
a
b b a
=
Then, each continuity equation yields a conservation law. Define
( ) ( ) , .
t t
Q t J Q t J





Then we find that
( ) ( ) 0.
a a
b a b b
Q b Q a J J J dJ


= = =


Similarly, ( ) ( ) 0. Q b Q a

Therefore, each of the charges Q

, Q

are conserved quantities, expressed as functions


of t .

Notice, however, the appearance of the horizon H . The difference between the two is
( ) ( ) ( ) ( ) 1
q
H
Q t Q t x q p


and involves an integral on the horizon H . Similarly, if each of the conserved quantities is derived by an on-shell
equation from a superpotential, e.g. J dj

, then we only have the expression
( ) .
t
H
H
Q t dj j Q

= = =


They are functions of H . Hence, we dispel another folklore item:
Conserved quantities are not universal constants, but are functions of the horizon.

Example: (The Foliations Associated with the Cosmological Horizon)
An observer lies on a world line ( ) t t M in the space-time manifold M , experiencing time as a continuing
progression. Part of this is accounted for by the fact that each instant t , the corresponding point ( ) x t = is
associated with a unique state space that comprises all the field configurations which evolve from the compact region
contained on the initial cosmological hypersurface on its past light cone and in its interior.

In a cosmological setting, the most important horizon is that associated with a given space-time point, obtained by
taking the intersection of its past light cone with the initial hypersurface. Let x M denote the point, and let
x

denote the past light cone of x . Denoting cosmological time of any y M by ( ) t y , with 0 t = the time of the
cosmological singularity, we then have the horizon ( ) { } : 0
x x
H y t y = = .

A foliation may then be defined by first associating each
x
y with its comoving geodesic
( ) ( ) 0, t t y y t M (


. This partitions the interior of the light cone
x
, projecting each point of the interior to a
unique point on the compact region ( ) { }
0
: 0 y M t y = bounded by the horizon
x
H . This foliation is defined by
the hypersurfaces

( )
( )
( ) ( )
: 0, .
s x
t y
y s y s t x
t x
| |

= ( | `

|
\ )


For all ( ) s t x < , this yields a space-like surface diffeomorphic to
0
. At ( ) s t x = , it yields the light cone
( ) x t x
= .

5. The Breakdown of Diffeomorphism Symmetry
The appearance of the explicit dependence on the horizon becomes especially important when dealing with external
symmetries, where 0 x ; in particular the diffeomorphisms.

Classically, this effect is already accounted for, by the version #1 theorem. But at a quantum level, a new effect comes
into play that is absent both from the classical level for general symmetries and from the quantum level for internal
symmetries. The boundary moves:
( )
~ S S S



where ~
x
e

is the diffeomorphism generated by x , with parameter .



Each region has associated with it a state-space K

. When we move ( )

= , we are also changing the


state space, not just the action. This results in an incomplete Bogoliubov transform K K

. Part of the region
within may now lie outside of and becomes averaged over in the cut-off associated with K

. Conversely, part
of the region in lies outside of and is averaged over in the cut-off associated with K

.

This is particularly true if the horizon H moves. So, there are 3 components to the transition :
(1) a motion along the field d dt ,
(2) a motion within the
t
s that preserves the horizon H and
(3) a motion that changes the location of H .
It is (3) that generates the anomalous behavior.

As a consequence, the Noether theorem must break down when applied at the quantum level to diffeomorphisms or
other external symmetries that involve any motion of type (3). This will lead to the appearance of an anomalous
contribution to the expressions for the stress tensor ( ) T L dx = or the current
x
J

.

This anomalous contribution is none other than the Einstein tensor. That is,
gravity is already implied by quantum field theory on a curved background.
Right now thats the detail Ive not yet been able to fully work out and prove, so it stands as a conjecture. But the
closest Ive seen to filling in the missing pieces is the recent effective gravity approach pursued by Padmanabhan, et al.
[5, 6, 7]

6. Addendum: Application to Gauge Fields
The foregoing may be expanded to a form suitable for gauge-natural bundles [4]; in particular to allow application to
form-valued fields acted on by a gauge symmetry. With that in mind, we assume that our original field q now consists
of a subset of field components, which we will continue to label q , plus 1-form field A that is associated with the
gauge symmetry.

In addition, we adopt a covering algebra that includes both the basis ( )
a
Y for the Lie algebra underlying the symmetry
group, and the basis ( )
A
e for the vector space acted on by the symmetry group. We assume that within the covering
algebra, we have the following actions:
( ) ( ) , , .
c C
a b a b ab c a b b a a B a B aB C
Y Y Y Y f Y Y Y Y Y Y e Y e f e = = = = = (


The basic requirements for consistency are
Anti-Commutativity:
0 , ,
0 ,
a b b a
c c
ab ba
Y Y Y Y
f f
= + ( (

= +

Jacobi identity:
0 , , , , , ,
0 ,
b c a c a b a b c
d e d e d e
bc da ca db ab dc
Y Y Y Y Y Y Y Y Y
f f f f f f
( ( ( = + + ( ( (

= + +

Representation consistency:
( ) ( ) ,
.
a b C a b C b a C
d E E D E E
ab dC aD bC bD aC
Y Y e Y Y e Y Y e
f f f f f f
= (

=

Finally, we generalize differential forms to forms that take values within this algebra. Thus, for instance, we may write

( ) ( )
( )
2
1
2
1
.
2
a b
a b
a b
a b
a b
a b b a
c a b
ab c
A A dx Y A dx Y
A A dx dx Y Y
A A dx dx Y Y Y Y
f A A dx dx Y








=
=
=
=


The kinematics are then revised to the following
2
, ; , 0. dq Aq v dA A F dv Av Fq dF AF FA + = + = + = + =
We will also write the ungauged forms of the kinematics with the following notation
, ; 0, 0. dq v dA F dv dF = = = =
Thus, the Lagrangian is now assumed to be a function of the form ( ) , , , , , L L q v A F x =

In order to represent the Lagrangian derivatives, we will also need to extend the covering algebra to include the
coadjoint action on the dual basis
( )
a
Y and the induced actions on the dual basis
( )
A
e :

( ) ( )
, .
c c b c c C C C B
a ab a a a a aB
Y Y f Y Y Y Y Y Y e e Y f e = = = =
Denoting contraction by
a a
b b
Y Y = and
A A
B B
e e = , these actions are characterised by the following identities:

( ) ( ) ( ) ( ) 0, 0.
c c C C
a b a b a B a B
Y Y Y Y Y Y Y e e e Y e + = + =
In particular, the contraction can be represented by a trace operator ( ) Tr _ if the operator satisfies the cyclicity
property, ( ) ( ) Tr Tr ab ba = . Then the second of the contraction identities could be written as

( ) ( ) ( ) ( )
( )
( ) ( ) ( )
( ) ( ) ( )
Tr
Tr 0,
Tr 0.
c c c c c
a b a b a a b a b b c
c c
a b b a
C C C C
a B a B a B a B
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y
Y e e e Y e e Y e e Y e
+ = +
= =
+ = + =

Finally, we may also define an induced action of the dual basis
( )
A
e on the basis ( )
A
e by the condition:

( ) ( ) ( )
Tr Tr .
C C C D C C C a
B a a B aB D aB B aB
e e Y Y e e e f e f e e f Y = = = =
This produces a term for the gauge current associated with the extra field.

Thus, the Lagrangian variational may be denoted
( )
( )
1
1 ,
q
q
L q f v p A J F G x K T
q f v p A J F G x K T
= + + +
= + + +

with the ( ) ( ) _ _ used to denote both contraction and exterior product.

The two sets of differential coefficients are related by the following
( ) ( ) ( ) 1 , 1 1 .
n q q n
f f pA J q p J AG GA

= + + = +
The Euler-Lagrange equations dp f = and dG J = then become
( ) ( ) ( ) 1 , 1 1 .
n q n q
dp pA f dG AG GA J qp

+ = + = +
The continuity equations 0 df = and 0 dJ = become
( ) ( ) ( ) 1 , 1 1 .
n q n q
df fA pF dJ AJ JA qf FG GF vp

= + + + =
The expressions for the stress tensor and force law become
( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
1 ,
1 1 1
1 1
1 1
0
q
n q q n q
n q
n q
x T x q f x A J x v p x F G x L
x K dT x v dp pA f x q df fA pF
x F dG AG GA J qp
x A dJ AJ JA qf FG GF vp

= + +
+ = + +
+ +
+ + + + + +



Finally, for variationals satisfying the condition L dL

= , the expressions for the current densities become


( ) ( ) ( ) ( ) ( )
( )
1 ,
1 ,
q
q
J q x v p A x F G x q f x A J L
J q p A G L






and the version #1 and version #2 Noether theorems respectively become
( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
1 1 1
1 1
1 1
0
n q q n q
n q
n q
dJ q x v dp pA f x q df fA pF
A x F dG AG GA J qp
x A dJ AJ JA qf FG GF vp

= + + +
+ +
+ + + + +


and
( )
( )
( ) ( )
( )
1 1 1 0.
n q n q
dJ q dp pA f A dG AG GA J qp

= + + +
Finally, for the version #3 theorem, we obtain the following:
( ) ( ) ( )
( )
( ) ( )
( )
( ) ( ) ( )
( )
1
1
1 1 .
q
x
n q
n q
J d x q p x A G
x q dp pA f
x A dG AG GA J qp

=
+ +
+ +


Thus ( ) ( ) ( )
( )
1
q
J J d x q p x A G

+ .

Example: (Gauge Symmetries)
The fields are acted on by gauge transformation as follows
, , , , . q q v v A d A A F F F L dL

= = = + = =
This leads to the following specialization of the theorem #2:
( ) ( )
( )
( )
( ) ( )
1 1 0.
n q
d J L dJ AJ JA qf FG GF vp

= + + +

References
[1] Friedrich W. Hehl, Jrgen Lemke, and Echehard W. Mielke (1991), Two Lectures on Fermions and Gravity
(lecture 2), in J. Debrus and A. C. Hirshfeld (Eds.), Geometry and Theoretical Physics, Springer-Verlag.
[2] Emmy Noether (1918), Invariante Variationsprobleme, Nachr. d. Knig. Gesellsch. d. Wiss. zu Gttingen,
Math-phys. Klasse, 235--257. Translated into English by M. A. Tavels in Transport Theory and Statistical
Physics, 1 (3), 183--207 (1971), reproduced by Frank Y. Wang in arXiv:physics/0503066v1 [physics.hist-ph] 8
Mar 2005.
[3] A. A. Kirillov (2004), Lectures on the Orbit Method. Providence, RI: American Mathematical Society. (Appendix
section on natural bundles, natural objects and natural operators).
[4] (Kopczynski et al. reference on Natural bundles, including the chapter on gauge-natural bundles).
[5] T. Padmanabhan, Thermodynamical Aspects of Gravity: New insights, arXiv:0911.5004v2 [gr-qc] 19 Jan 2010.
[6] Ted Jacobson, Thermodynamics of Spacetime: The Einstein Equation of State, arXiv:9504004 v2 [gr-qc] 06
Jun 1995.
[7] Erik Verlinde, On the Origin of Gravity and the Laws of Newton, arXiv:1001.0785v1 [hep-th] 6 Jan 2010.

You might also like