Deligne P., Freed D. Classical Field Theory (AMS 1999) (90s) .Ps

Classical Field Theory
Pierre Deligne and Daniel S. Freed
Introduction
A physical system is usually described in terms of states and observables. (See
the discussion in [I-Faddeev, x1.1].) In the Hamiltonian framework of classical
mechanics, the states form a symplectic manifold (M;
) and the observables are
functions on M. The dynamics of a (time-invariant) system is described by a
one-parameter group of symplectic dieomorphisms; the generating function is the
energy or hamiltonian . The system is said to be free if (M;
) is an ane symplectic
space and the motion is by a one-parameter group of ane symplectic transforma-
tions. This general description applies to systems which include classical particles,
strings, elds, and other types of objects. Often the dynamics of the theory is
embedded in a larger symmetry group. For example, in relativistic eld theories
one assumes that (M;
) carries a representation of the Poincare group.
Many classical systems admit a lagrangian description in which (M;
) is de-
rived from a relatively simple expression, called the lagrangian density. One of the
main features of a lagrangian description is that the conserved quantity|called
the Noether charge |corresponding to a symmetry is computed directly from the
lagrangian. Furthermore, in eld theories there is a local Noether current which
integrates to the global charge and which gives rise to local conservation laws. This
re ects the physical fact that, for example, we can measure energy in any region
of space, not just the total energy over all of space. In this text we develop the
basic ideas of classical lagrangian eld theory. The examples we have in mind are
the ones which arise in relativistic quantum eld theory as treated in other parts
of the book. One should be careful in trying to apply the formalism developed
here to constrained systems (e.g., nonholonomic constraints in classical mechanical
systems, or constraints imposed on superelds in the superspace descriptions of su-
persymmetric elds theories). In most of the exposition, we will assume that elds
are arbitrary sections of some xed ber bundle E over spacetime. Also, we develop
the formalism in a purely local way from the lagrangian density, hardly mentioning
its integral, the action . If one is interested in calculus of variations problems, then
the emphasis is dierent: the action is of primary interest and boundary conditions
play a crucial role. A last warning is that not all classical eld theories admit a la-
grangian description. There is a free example which is important in two-dimensional
conformal eld theory: a free chiral scalar eld in two dimensions.
137
138 P. DELIGNE, D. FREED, CLASSICAL FIELD THEORY
In Chapter 1 we review some classical mechanics. In a few standard examples
we describe the classical equations of motion and the construction of a symplectic
structure on the space of classical solutions. Noether's theorem|the construction of
a conserved quantity from a one-parameter group of symmetries|is also discussed.
We treat both nonrelativistic and relativistic examples, and in x1.4 we show how
to obtain nonrelativistic Galilean spacetime as a limit of relativistic Minkowski
spacetime.
The general theory of classical elds is laid out in Chapter 2. A classical
lagrangian eld theory consists of a spacetime M , a space of elds F , and a la-
grangian density L. We are mostly interested in Minkowski spacetime, but the
theory is quite general and applies to spacetimes which are curved Lorentzian or
Riemannian manifolds as well. The elds are some sort of functions on M , more
precisely sections of a ber bundle E over M . The lagrangian density is a density
on M for each point of F . It is assumed1 to be of a local nature on M . More
precisely, for some k the value of L() at a point m of M should depend only on
the k-jet of at m. Usually k = 1. The lagrangian density L determines Euler-
Lagrange equations DL = 0, also called equations of motion, which cut out the
space of extremals M F . If is a eld, the tangent space of F at is the space
of sections of the vector bundle T (E=M ) over M . The Euler-Lagrange equation
DL is a morphism of vector bundles from T (E=M ) to the bundle of densities on
M . Its characteristic property is that for a deformation with compact support [u]
of , one has
Z
d L()[u] = Z DL d [u] :
du
u=0 du u=0
The formalism naturally takes place in the double complex of dierential forms
on F M . This is qualied in two ways: (i) as we want L() to be a density, rather
than a dierential form of maximal degree, this double complex should be twisted
by the orientation bundle of M ; and (ii) we want to consider only (p; q)-forms
which are local on M : to a eld and tangent vectors 1 ; : : : ; p 2 T (E=M ),
the form attaches a q-form (; 1 ; : : : ; p ) on M , and for some k the value of
(; 1 ; : : : p ) at a point m of M should depend only on the k-jet at m of and
the i . The cohomology of the double complex of local forms has been investigated
by F. Takens2. Write the exterior dierential as D = + d, with of degree (1; 0)
and d of degree (0; 1). Takens' main result is Theorem 2.15: For p > 0 the complex
(
p;loc ; d) of local dierential forms is exact except in top degree. One can view
the
p;q loc as an inductive limit, in k , of spaces of global sections of soft sheaves
on E , and this makes the exactness of (
p;loc ; d) (p > 0; 6= top) a local question
on E . Let J k (E ) be the bundle over M of k-jets of sections of E . Takens also
observes that the associated simple complex
loc is the inductive limit over k of the
de Rham complexes of the J k (E ). As the projections J k (E ) ! E are brations,
with bers ane spaces, they induce isomorphisms in cohomology. It follows that
the cohomology of E maps isomorphically to that of
loc . We include a proof of
Takens' results in an appendix to Chapter 2.
1 Although we restrict to lagrangians which depend locally on elds, in quantum eld theory one
meets eective lagrangians which are not local. Some of the formal aspects carry over to nonlocal
lagrangians, but we have not taken the trouble to distinguish them.
2 See References at the end of the manuscript.
INTRODUCTION 139
If the lagrangian density L() depends only on the rst jet of , it denes
a (1; n 1)-form (where n = dim M ), which we call the variational 1-form ,
characterized by the following properties: (i) the value of the (n 1)-form (; )
at m 2 M depends only on the 1-jet of at m, and on the value of at m; and (ii)
DL = L + d :
The variational (1; n 1)-form encodes the usual integration by parts argu-
ment which occurs in computing Euler-Lagrange equations. For more general la-
grangians, the choice of a local (1; n 1)-form with DL = L + d should be
considered part of the denition of the theory. Such a form always exists and, by
Takens' theorem, it is unique up to the addition of d , for a local (1; n 2)-form.
Classical mechanics corresponds to M = R (time). In classical mechanics, one
is used to the following package: (i) the space of extremals M is a symplectic man-
ifold; (ii) symmetries give symplectic automorphisms of (M; !), and innitesimal
symmetries are given by generating functions Q, with dQ = ( )!. A particular
case is the one-parameter group of time translations, whose generating function is
minus the hamiltonian. In (ii), Q is ambiguous up to an additive constant (and
its existence can be obstructed by H 1 (M; R)). This ambiguity can often be re-
moved by rening (i) to: (i0 ) the space of extremals carries a canonical principal
R-bundle T with connection r whose curvature is ! . Innitesimal automorphisms
of (M; T; r) can then be identied with pairs (; Q) consisting of a vector eld
on M and a function Q such that dQ = ( )!.
In eld theory one has a similar Hamiltonian interpretation if M is given as time
space, or if one has a suitable notion of space-like hypersurface. A new problem is
to obtain counterparts local on M of those constructions. These local counterparts
do not require a notion of space-like hypersurface, or a splitting into time space,
but rather exist for any spacetime M . In particular, they make sense for Euclidean
analogs, obtained by a Wick rotation t 7! it. For the 2-form on M, which we now
denote `
', this is done in a paper of G. Zuckerman:3 if ! is the local (2; n 1)-
form on F M , then the 2-form
on M is deduced from ! by integration
on any space-like hypersurface. Similarly, one can express the generating functions
of innitesimal symmetries as integrals over space-like hypersurfaces of conserved
currents, called Noether currents . Let be a vector eld on F which is local: at
m 2 M depends only on some jet of at m. Suppose it is a generalized symmetry
of L, in the sense that for some given (0; n 1)-local form , one has
Lie( )L = d :
Then the Noether current j := ( ) is conserved : dj () = 0 for all
in M,. The vector eld is tangent to M, and the integral Q of j on a space-like
hypersurface is a corresponding generating function.
It can be more natural to consider symmetries which act on F and M simulta-
neously. We will say that a symmetry is manifest if it preserves L and exactly. For
example, time translation is usually a manifest symmetry when it acts on both F
and M . If we let it act only on elds, not on spacetime, it is only a generalized
symmetry (with = (@t )L).
3 See References.
Theories whose eld content includes a connection or metric possess an innite
dimensional group of local \gauge" symmetries. (For metrics the gauge symmetry
group is the group of dieomorphisms.) We can freeze the metric or connection at
xed values g0 ; A0 ; then local symmetries which x g0 or A0 act as global symme-
tries in the theory of the remaining elds. The Noether current for these global
symmetries may be computed by dierentiating the total lagrangian with respect
to the metric or connection, as explained in x2.8 and x2.9. The derivative with
respect to the metric is called the energy-momentum tensor.
In x2.10 we discuss time-invariant eld congurations of nite energy on space-
times which are time space. Among these we nd classical vacua and solitons .
We also explain brie y perturbations around a classical vacuum and the Higgs
mechanism.
Chapter 3 summarizes the basic free lagrangian eld theories on Minkowski
spacetime. We treat scalar elds, spinor elds, and abelian gauge elds (connec-
tions). Our goal is to illustrate the general theory in the simplest case and to record
useful formulas.
In classical physics one of the main applications of eld theory is to electro-
magnetism. Chapter 4, which is a discussion of gauge theory in general, begins
with a brief treatment of Maxwell's equations in the lagrangian framework. Some
familiarity with this material is necessary to understand the intuition behind more
complicated models with gauge elds, which are encountered in many lectures in
these volumes. In x4.2 we review the basic geometry of connections in principal
bundles, paying special attention to the universal connection (in terms of which
we write gauge theory lagrangians). Then in x4.3 we write the lagrangian for
Yang-Mills theory and describe some additional \-terms" which may appear in
low dimensions. Finally, we dene electric and magnetic charge in x4.4 and discuss
the relationship to global gauge transformations.
The general bosonic lagrangians without gravity usually include only scalar
elds and gauge elds. We discuss a general lagrangian for these elds in Chapter 5;
it includes many important bosonic theories as special cases.
There are special topological terms in lagrangians which are invariant under
(orientation-preserving) dieomorphisms. Some, like the -terms mentioned above,
are related to primary topological invariants. More subtle are the ones associated
to secondary invariants, like the Wess-Zumino-Witten term in a -model or the
Chern-Simons term in three-dimensional gauge theory. In Chapter 6 we brie y
introduce the main examples and explain how the action acquires a more subtle
geometric meaning. The geometric home for the lagrangian is a \ -calculus" which
extends the usual calculus of dierential forms, as we indicate in x6.3.
Finally, in Chapter 7 we discuss the \Wick rotation" of a lagrangian from
Minkowski spacetime
p to Euclidean spacetime. For reference we collect the signs
and factors of 1 which occur in this analytic continuation.
To a large extent this text presents a preliminary version of this material;
we are not satised at all with our understanding in many places. Still, we felt
it important to include some mathematical framework for the computations in
lagrangian eld theory. Our treatment is guided by what is needed to follow the
lectures and problems recorded in these volumes, and we hope at least to have
provided sucient background for that. Other mathematical accounts appear in
the references.
INTRODUCTION 141
The denition of a classical lagrangian that we use (Denition 2.39) is adopted
from a lecture of Joseph Bernstein. The basic formalism appears in the aforemen-
tioned paper of Gregg Zuckerman. In preparing this text we also beneted from
many discussions with David Kazhdan, John Morgan, Nati Seiberg, and Ed Witten,
among others.
CHAPTER 1
Classical Mechanics
The equation of motion of classical mechanics is the Euler-Lagrange equation
for extremizing the action integral. From this extremal description we construct
a canonical closed 2-form ! on the space M of classical evolutions. It turns out
that ! is nondegenerate, and so M is a symplectic manifold. More precisely, we
construct on M a canonical R-torsor4 with connection whose curvature is !. As
usual in symplectic geometry, to an innitesimal symmetry we associate a function
which is called the Noether charge . (It is often called the momentum or momentum
map .)
In quantum theory we apply exp( ~i ) to turn this R-torsor into a unitary line
bundle with connection whose curvature is ~i !.
x 1.1. The nonrelativistic particle
We treat three cases: the free particle, a system of particles with potential, and the
electromagnetic eld.
Free Particle
Let X = Rn be Euclidean space with its standard inner product h; i. The evolution
of a classical free particle of mass m is described by a map x from R (time) to X .
The lagrangian density 5 is the density on R
(1.1) L = m2 jx_ j2 dt;
where jx_ j2 is the inner product hdx=dt; dx=dti. The integral of the lagrangian
density
Z t1
(1.2) S= L
t0
from time t0 to time t1 is the action integral , or simply action . If we deform x we
have
L = mhx;_ x_ i dt
(1.3) n o
= mhx; xi dt d mhx;_ xi :
4 In this context `R-torsor' means `principal R-bundle'.
5 We should use the density jdtj in place of the form dt in (1.1), but in this section we simply
orient R and so identify 1-forms and densities.
143
Here is the dierential on the space F of trajectories x of the particle, d is the
dierential on R, and the second minus sign6 in the second line of (1.3) arises since
and d anticommute on F R. Integrating we nd
Z t1 Z t1 h it1
(1.6) S = L= dt mhx(t); x(t)i + mhx_ (t); x(t)i t :
t0 t0 0
The extremality condition is that the 1-form S on F vanish on deformations
of x with compact support in (t0 ; t1 ). This leads to the classical equation of motion
(1.7) x = 0
whose solutions are uniform motion. The boundary term leads one to consider, for
each time t, the 1-form
(1.8) (t) = mhx_ (t); x(t)i
on the space F of all paths x. On the subspace M of extremals (solutions to (1.7))
the action S is a function whose dierential is
(1.9) S = (t1 ) (t0 ) on M:
It follows that the 2-form on M dened by
!(t) : = (t)
(1.10)
= mhx_ (t) ^ x(t)i
is independent of t. It is even independent of t for a specic reason, namely (1.9),
with a compatibility among the reasons if three times t0 < t1 < t2 are considered:
Z t1 Z t2 Z t2
(1.11) L+ L= L:
t0 t1 t0
This can be rephrased as dening an R-torsor with connection (T; r) on M whose
curvature is !: for each xed t0 it is the trivial R-torsor R t1
T (t0) with the con-
nection r(t0 ) given by (t0 ). By (1.9)
addition of t0 L gives an isomorphism
from T (t0); r(t0 ) to T (t1 ); r(t1 ) , and by (1.11) these isomorphisms form a
compatible system of isomorphisms. The desired (T; r) is the \common value"
(projective limit) of the T (t); r(t) .
If we x t0 , the map x 7! x(t0 ); x_ (t0 ) from the space M of extremals to the
tangent bundle TX of X is an isomorphism. If we use (1.8) to map the tangent
bundle to the cotangent bundle, then (resp. !) is the pullback of the canonical
1-form (resp. canonical 2-form) on the cotangent bundle.
6 It is more usual to let x denote a tangent vector to F , in which case there is no minus sign. We
emphasize that our computations take place on F R. For example, `x' in (1.1) is the evaluation
map F R ! R and `x_ ' is its time derivative. To convert (1.3) into the more usual formula, let ^
be a vector (eld) on F , and set
(1.4) ^x = (^)x = X:
Apply the contraction (^) to (1.3). Note that in commuting (^) past d in the last term we pick
up a minus sign. Thus we nd
n o
(1.5) (^)L = ^L = mhx; X i dt + d mhx;_ X i :
CHAPTER 1. CLASSICAL MECHANICS 145
System of Classical Particles with Potential
More generally, if we consider a system of classical particles with rigid constraints,
the conguration space is a Riemannian manifold X with Riemannian structure
given by twice the kinetic energy. Evolution is described by a map from R (time)
to X . The lagrangian density is
(1.12) L = 12 jx_ j2 dt;
here the masses are included in the metric. If in addition we have an external eld
of forces depending on a potential, or interaction between the particles described
by potentials, then the potentials are encoded by a real-valued function V on X ,
and the lagrangian density is
L = 12 jx_ j2 V x dt:
n o
(1.13)
The free story can be repeated with the following changes. The Euler-Lagrange
equation (1.7) is now Newton's law
(1.14) rt x_ + grad V = 0:
The 1-form on the space of x is
(1.15) (t) = hx_ (t); x(t)i
and the symplectic 2-form is
(1.16) !(t) = hx_ (t) ^ x(t)i:
The construction of the R-torsor is as before. The identication of the space of
solutions with the initial data at a xed time depends on suitable completeness
assumptions.
Electromagnetic Field
An electromagnetic eld can be described as an R-torsor with connection (P; r) on
spacetime R X . For now choose a trivialization of P (a \gauge"), and so write r
as a 1-form . We separate the time and space components by
(1.17) = V dt + A:
Then V is the scalar potential and A is the vector potential . The evolution of a
single charged particle is described as before by a map x : R ! X , and for a particle
of mass m and charge q the lagrangian density is
(1.18) L = m jx_ j2 dt qx ():
2
R t1
The action S = t0 L is the sum of a kinetic energy term and of q times the parallel
transport along x from x(t0 ) to x(t1 ), computed in the chosen gauge. In L the
boundary term is now

(1.19) (t) = mhx_ (t); x(t)i qA x(t) :

Fix t0 ; then under suitable completeness assumptions the map x 7! x(t0 ); x_ (t0 )
identies the space M of extremals with the tangent bundle TX . Notice that the
map from the tangent bundle to the cotangent bundle given by (1.18) is the previous
map shifted by qA.
If we change the gauge, theR space of extremals does not change. This becomes
clear if we do not interpret tt01 L as a number, but rather as an isomorphism
from qP(t0 ;x(t0 )) to qP(t1 ;x(t1 )) , the sum of a kinetic energy term and of parallel
transport. This isomorphism is manifestly gauge independent. On the space of
extremals M we continue to have an R-torsor with connection whose curvature
is !, independent of the gauge. For t 0 xed it is naturally t 0 ; x(t 0 ) (qP ) with
connection given by t0 ; x(t0 ) (qr)+ mhx_ (t0 ); x(t0 )i. Isomorphisms between this

R-torsor with connection for dierent choices of t0 are given by the action|parallel
transport plus a kinetic energy term.
In this example we see that the action need not be a number, but rather can be
an element of an R-torsor. We discuss such topological terms further in Chapter 6.
x1.2. The relativistic particle
Let X be n-dimensional Minkowski spacetime. This is standard n-dimensional
ane space with a Lorentz metric. Fix ane coordinates t; x1 ; x2 ; : : : ; xn 1 so that
the metric takes the form
(1.20) c2 (dt)2 (dx1 )2 (dxn 1 )2 :
The corresponding basis of the underlying vector space of translations is called an
inertial frame . Here c is the speed of light. It is often convenient to set x0 = ct.
The worldline of a relativistic particle is represented by a map
x: R ! X
(1.21)
x( ) = t( ); x1 ( ); : : : ; xn 1 ( )
with hdx=d; dx=d i 0 and dt=d > 0. The lagrangian density of a free particle
of rest mass m0 is
1=2
(1.22) dx
L = m0c d ; d dx d:
For a physical particle dx=d lies in the positive light cone; in particular, we have
h ddx ; dx
d i > 0. This lagrangian is+ invariant by the Poincare group of symmetries of
X , as well as by the group Di (R) of reparametrizations of x. An action integral
is attached to a region R of spacetime bounded by two space-like hypersurfaces H0
and H1 . One can, for instance, take for R the region t0 t t1 . For a free particle
of rest mass m0 , the action integral is m0c times the arc length of the portion of
the path x contained in R.
The extremals for this action are straight lines. If S is the action integral for
the region R bounded by hypersurfaces H0 ; H1 , and if we deform x, then S is the
sum of two terms: (i) an integral 01 , where x(i ) 2 Hi ; and (ii) boundary terms.
R
The boundary term for H1 is

(1.23) [H1 ] = m0 c hx_ (j1x_)(;x)(j1 )i :
1
Note that this expression is Di(R)-invariant. As previously, ! := [H1] is inde-
pendent of H1 on the space of extremals and turns it into a symplectic manifold.
More precisely, the space of extremals carries an R-torsor T with connection whose
curvature is !. For each choice of space-like hypersurface H it can be identied
with the trivial R-torsor, with connection given by [H ]. Action integrals give
a transitive system of isomorphisms between the descriptions of (T; r) given by
dierent H .
To go to the nonrelativistic limit we proceed as follows. We work in our chosen
coordinate system and consider velocities which are small relative to the speed of
light c. If we take for coordinate the time t, then the path is

(1.24) x(t) = t; x1 (t); : : : ; xn 1 (t) :
The lagrangian is then
p
L = m0 c2 1 v2 =c2 dt

(1.25) 4
= m0 c2 + 21 m0 v2 + O( vc2 ) dt;
where
X 2
(1.26) v2 = dxi
i dt
is the velocity squared. The second line of (1.25) shows that the nonrelativistic
limit v=c ! 0 of L is the kinetic energy 21 m0 v2 minus the potential energy m0 c2 of
the particle at rest. The reader can also check that the nonrelativistic limit of (1.23)
for the hypersurface H1 : ft = constantg is (1.8) (for m = m0 ).
In the relativistic setting it does not make sense to introduce rigid constraints on
a system of particles, nor to introduce potentials. On the other hand, a background
electromagnetic eld can be introduced on Minkowski spacetime analogously to the
nonrelativistic case.
x 1.3. Noether's theorem
We have seen a number of examples of how, because equations of motion are Euler-
Lagrange equations, the space of classical evolutions M carries a symplectic struc-
ture !. More precisely, it carries a canonical R-torsor T with connection r whose
curvature is !. Any automorphism of the data used to dene the extremality con-
dition induces an automorphism of (M; T; r). The same holds for innitesimal
automorphisms. Innitesimal automorphisms of (M; T; r) can be identied with
functions Q on M. Namely, if a vector eld ~ on T is such an innitesimal au-
tomorphism, let denote the projection to M and ^ the horizontal lift of to T .
Then
(1.27) ~ = ^ + Q;
where is the innitesimal action of R which denes the torsor structure. The
innitesimal automorphism ~ of T respects the connection if and only if
(1.28) dQ = ( )!:
If the curvature ! is nondegenerate, i.e., turns M into a symplectic manifold, then
(1.28) shows that vector eld is determined by the generating function Q. For
functions Q1 ; Q2 corresponding to vector elds Q1 ; Q2 , we compute the vector
eld which corresponds to the Poisson bracket fQ1 ; Q2g to be
(1.29) fQ1 ;Q2 g = [Q1 ; Q2 ]:
(Note: Brackets of innitesimal automorphisms are given by the opposite of Poisson
brackets, in the same way that the Lie algebra of dieomorphisms is the set of vector
elds with bracket opposite to the usual brackets of vector elds.)
If we have a trivialization of (T; r) which is preserved by an innitesimal au-
tomorphism ~, then
(1.30) Q = ( ) ;
where is the connection form on M induced from the trivialization. If the triv-
ialization changes, then there is an additional term in (1.30). In x2 we develop a
more systematic formalism in which to make computations, so in this section when
the trivialization changes we simply report the result.
We apply these ideas to the free particle in both the nonrelativistic and rel-
ativistic settings. Consider rst a free nonrelativistic particle of mass m0 moving
in Rn . Let x1 ; : : : ; xn be standard coordinates on Rn and ij the standard metric.
The isometries of Rn induce automorphisms of (M; T; r) which x the trivializa-
tion at a xed time t0 . Thus we can use (1.30) in conjunction with (1.8) to compute
the charges. The charge corresponding to the innitesimal translation = @=@xi
is the linear momentum
(1.31) pi = m0 ij x_ j :
The charge corresponding to an innitesimal rotation = xi @=@xj xj @=@xi is the
angular momentum
(1.32) Mji = m0 jk (xi x_ k xk x_ i ):
Now the hamiltonian or energy E is the charge associated to minus time translation
in the domain R, and this does not preserve the trivialization of (T; r) induced at
a xed time. Rather, the derivative of the trivialization at t0 is L0(t0 ), where we
write the lagrangian density as L0 dt. (The change in trivialization comes because
innitesimal time translation only preserves the lagrangian up to a total derivative.)
Accounting for this we compute the nonrelativistic energy (see Example 2.105).
(1.33) E = 21 m0 jx_ j2 :
There is an additional symmetry called a Galilean boost . We write the innitesimal
version as a time-varying vector eld = t@=@xi . Again there is an additional term
in the formula for the charge Ni since the lagrangian is only preserved up to a total
derivative:
(1.34) Ni = m0 ij (tx_ j xj ):
This is simply minus the initial position times the mass.
Consider now a relativistic particle. The lagrangian (1.22) is invariant under
the isometries of X , the Poincare group, and we would like to compute the corre-
sponding charges. Fix a splitting of Minkowski spacetime into time space and
use the hypersurface t = t0 to trivialize (T; r). We parametrize the worldline as
in (1.24). The subgroup of Poincare which xes this hypersurface also xes this
trivialization and we use (1.30) and (1.23) to compute the associated charges. Thus
the momentum pi is the charge corresponding to = @=@xi , which we compute to
be
j
(1.35) pi = pm0 ij 2x_ 2 = m(v)ij x_ j :
1 v =c
Here we have introduced the relativistic mass
(1.36) m(v) = p m0 2 2 :
1 v =c
Note that the nonrelativistic limit v=c ! 0 of the relativistic mass m(v) is the rest
mass m0 . Similarly, an innitesimal spatial rotation = xi @=@xj xj @=@xi has
charge the angular momentum
(1.37) jk (xi x_ k xk x_ i ) = m(v ) (xi x_ k xk x_ i ):

Mji = m0 p jk
1 v2 =c2
Now the charge corresponding to = @=@t is the relativistic energy E . As in the
nonrelativistic case the trivialization changes according to the lagrangian (1.25)
(divided by the standard density dt), and so we compute Einstein's famous formula
2
(1.38) E = p m0 c 2 2 = m(v)c2 :
1 v =c
Note that (E; pi ) transforms in the dual to the standard representation of the
Lorentz group. In n = 4 dimensions this quantity is called the 4-momentum .
Finally, the innitesimal Lorentz boost
i @
(1.39) = xc2 @t @
+ t @x i
has corresponding charge
j j
(1.40) Ni = m0pij (tx_ 2 x2 ) = m(v)ij (tx_ j xj ):
1 v =c
Here again the symmetry does not preserve the trivialization, so there is an extra
term to compute. The charge Ni is simply minus the initial position of the particle
times the relativistic mass.
The nonrelativistic limits of (1.35), (1.37), (1.38), (1.40) give (1.31), (1.32),
(1.33), (1.34). (From E in (1.38) we must subtract the rest energy m0 c2 to obtain
the nonrelativistic energy (1.33).)
x1.4. Synthesis
Let Mc denote n-dimensional Minkowski spacetime with metric gc given in (1.20).
Let M1 denote the limiting space as c ! 1. Of course, the ane spaces which
underlie Mc and M1 are identical. Let W be the vector space of translations
of M1 . The limit clim!1 gc 1 = g11 of the inverse metric is a degenerate quadratic
form on W . It has a one-dimensional kernel S W whose annihilator S W
determines a codimension one foliation of M1 by ane subspaces. Then g11
on W =S is inverse to a negative denite metric on S . We think of this foliation
(with the metric on the leaves) as dening the simultaneous spatial events in M1 .
We also x a scale dt 2 S for time; then there is a one-dimensional space of
distinguished ane time functions which dier by a constant. The group G1 of
ane transformations of M1 which preserve g11 and dt is the Galilean group . Let
Gc denote the group of ane transformations of Mc which preserve gc ; the double
cover of its identity component is the Poincare group. Note that Gc ; G1 act on the
same ane space, so there is a well-dened sense in which Gc ! G1 as c ! 1.
The translations sit inside Gc and G1 . In G1 the spatial translations lie in
a subgroup H of dimension 2(n 1) generated by vector elds t@ + @ 0 , where t is
any ane time function and @; @ 0 are spatial translations. Vector elds t@ generate
ane transformations called Galilean boosts ; they are the c ! 1 limit of Lorentz
boosts (1.39).
There is a nontrivial central extension of G1 which restricts nontrivially on H .
For a geometric picture we begin with a xed (n +1)-dimensional Minkowski space-
time N with vector space of translations V . For each spacelike vector v 2 V we
consider the subgroup Gv of the Poincare group which xes v. It acts on the quo-
tient ane space Mv = N=R v. This quotient space Mv inherits a Minkowski
metric we describe in two equivalent ways: it is the metric induced from the or-
thogonal space (R v)? V , or equivalently its inverse is the subspace metric on
the annihilator (V=R v) = (R v) V . The subgroup Gv is the trivial central
extension of the Poincare group of Mv by translations along v. Now consider ` 2 V
a lightlike vector. Set M` = N=R `, and let G` be the subgroup of the Poincare
group of N which xed `. Now the inverse metric on V restricts to a degenerate
form on (V=R `) = (R `) V . Also, we take the functional h`; i to be the dt
above. The group of ane transformations of M` which preserves the degenerate
form and ` is the Galilean group. Thus G` is the central extension of this group
by translations along v. Note that the subspace of (V=R `) = (R `) previously
?
called S is here (` ) , the annihilator of the orthogonal subspace to `. (Since ` is
lightlike, we have R ` `? , which is why it makes sense to dene dt = h`; i.) Also,
S is simply `?=R `.
For an explicit description introduce coordinates t+ ; t ; x1 ; : : : xn 1 on N so
that the metric is7
(1.41) dt+ dt (dx1 )2 (dxn 1 )2 :
7 Here the speed of light is inessential, so we set it to one.

Take ` = @+ = @=@t+. Then (`? ) is the span of dt , and the possible time
functions are fat + b : a; b 2 Rg. The Lie algebra of G` is spanned by
@+ (central element)
@ = @t@ ; @i = @x@ i (translations)
(1.42)
xi @j xj @i (spatial rotations)
Bi = t @i + xi @+ (Galilean boosts)
The nontrivial bracket
(1.43) [@i ; Bj ] = ij @+
re ects the nontrivial central extension. Note from (1.31) and (1.34) that for the free
nonrelativistic particle the Poisson bracket of the Noether charges corresponding
to @i and Bj is
(1.44) fpi ; Nj g = ij m0 :
Comparing (1.43) and (1.44) we see that the conserved quantity corresponding
to @+ in a theory should be identied with the mass. Thus in the Galilean theory
there is an additional conserved quantity over the relativistic theory|the mass.
(The bracket of elements in the centrally extended Galilean algebra and the Pois-
son bracket of the corresponding Noether charges are opposite. This is the usual
situation for left group actions on symplectic manifolds; see x5 of [I-Signs].)
Quite generally, suppose we have a classical theory whose state space (M; !)
carries a symplectic action of either the Galilean group or the Poincare group.
Innitesimally we have an antihomomorphism from the Lie algebra g of that group
into the Lie algebra of vector elds on M. We assume that each vector eld in
the image satises ( )! is exact, i.e., is the symplectic gradient of some function.
For a general group G, the existence and uniqueness of a lift g ! C 1 (M) which
is an antihomomorphism of Lie algebras is measured by H 1 (g) and H 2 (g). For the
Poincare group there is a unique lift; for the Galilean group there is a lift of the
central extension we constructed above, unique up to a shift of the total energy.
(If spacetime has dimension 3, then we can also shift angular momentum.) The
central element then maps to a locally constant function which is the total mass of
the system. In the Poincare case the vector space V of translations maps to a vector
space of functions which|after choosing an inertial frame|is ( E; p1 ; : : : ; pn 1),
where E is energy and pi are linear momenta. Under Lorentz transformations this
transforms as the coecients of an element in8 V , and out of it we construct an
invariant using the metric:
(1.45) E 2 =c2 jpj2 = m20 c2 :
This denes the \rest mass" m0 of the system. It is a Poincare invariant, though in
general not locally constant, function on M. In the Galilean case there is a Galilean-
invariant codimension one subspace S V of translations in spatial directions, and
8 Note that the vector dual to (E; p) is (E=c2 ; p).
S carries a metric. If the Galilean algebra, and not a central extension, were to
lift to functions, then we would conclude that the norm square jpj2 of the total
momentum is Galilean-invariant. But clearly jpj2 typically changes under boosts,
and this shows why boosts and spatial translations have a nonzero commutator in
the central extension.
We can formulate the theory of the free nonrelativistic particle by considering
its worldline in the space M1 . The dierence from the relativistic situation is
that the time functions provide distinguished parametrizations of the worldline
up to translation. This allows us to \couple" the free nonrelativistic particle to
an arbitrary potential function V , as in (1.13); the integrated lagrangian (action)
does not change if we translate time and so is Galilean invariant. By contrast, as
stated earlier we cannot introduce a Poincare invariant coupling of the relativistic
particle to a potential function. Rather, we can couple it to elds , specically to
an electromagnetic eld (abelian connection) and to a gravitational eld (variable
Lorentz metric). In the next chapter we take up the general theory of elds.
CHAPTER 2
Lagrangian Theory of Classical Fields
x 2.1. Dimensional analysis
It is often useful to follow the advice we give to beginning students: Check units!
Every physical quantity has units attached to it. The basic units are units of
mass, length, and time. The number measuring a physical quantity Q depends
on the choice of units, and Q is said to be of dimension M a LbT c if the number q
measuring Q is multiplied by a b c when the units are multiplied by ; ; .
Notation: [Q] = M aLb T c. For example, the action integral S of (1.2), where the
integral is taken between prescribed instants, has the dimension of an action:
(2.1) [S ] = ML2T 1:
The dimension of a p-form is dened to be the dimension of its value on a xed
p-vector. The lagrangian density (1.1) (a 1-form on the time line), the variational
1-form of (1.8) (a 1-form on the space of trajectories), and the closed 2-form !
all have the dimension of an action.
The dimension of the conserved quantity Q corresponding to an innitesimal
Galilean transformation is given by
(2.2) [Q] = [action][]:
In the relativistic setting we will usually impose c = 1, so that a unit of length gives
one of time. Then [action] becomes ML, and the conserved quantities corresponding
to innitesimal generators of the Poincare group once again have
(2.3) [Q] = [action][]:
We can consider electric charge to have units C and take the constant q in (1.18)
to have [q] = C ; then the 1-form in (1.17) which represents the electromagnetic
eld has units of action divided by charge:
(2.4) [] = ML2T 1C 1 :
Alternatively, we can work in a system of units in which charge is expressed in
terms of mass, length, and time by
(2.5) C 2 = ML3 T 2:
153
This comes from declaring the constant k in Coulomb's law \F = kq1 q2 =r2 " to be
dimensionless.
Universal physical constants allow us to convert units. In relativistic theories
the speed of light
(2.6) [c] = LT 1
identies time with length. In quantum theories Planck's constant
(2.7) [~] = ML2T 1
has units of action, which eliminates one of M; L; T . In theories of gravity the
Newton constant G in the Newton force law \F = Gm1 m2 =r2 " has units
(2.8) [G] = L3 M 1 T 2;
which again allows us to eliminate one of M; L; T . In a relativistic quantum eld
theory we can use c; ~ to express all units in terms of mass, and so each physical
quantity has a mass dimension .9 If the theory includes gravity (e.g. string theory),
then we can use c; ~; G to express everything in terms of dimensionless quantities.
x2.2. Densities and twisted dierential forms
In this section M is an ordinary (not super) manifold.
If M is oriented of dimension n, then n-forms with compact support can be
integrated on M . Changing the orientation multiplies the integral by 1. This
leads to the consideration of densities , dened to be sections of
Dens M := n T M
oM ;
V
(2.9)
where oM is the orientation line bundle. On any M , oriented (and orientable) or
not, the integral of a density ! with compact support is unambiguously dened:
one writes ! as a sum of densities !i with support in orientable local charts Ui , one
P R Ui , and doing so identies !i with an n-form on Ui , and one denes
orients each
R
M ! := Ui !i .
More generally, we will have to consider the complex of twisted forms , the tensor
product of
M with the local system oM . We treat oM as being in cohomological
degree n, and dene
(2.10)
jM pj :=
M n p
o :
M
j p j
A compactly supported element of
M can be integrated on a normally oriented
submanifold of M of codimension p. Elements of
jM pj can be viewed as sections
of
Vp
(2.11) TM
Dens M ;
the isomorphism with the description (2.10) comes by contracting a p-vector eld
with a density.
In the sequel we often say simply `form' for `twisted form'.
In the case of supermanifolds, densities are related not to an exterior power of
the cotangent bundle, but to its Berezinian, and (2.11) denes the components of
the complex of integral densities. See [I-Supersymmetry, xx3.9{12] for details.
On a Riemannian manifold M the Hodge -operator is an isomorphism
(2.12) :
qM !
Mj qj :
9 Then for a physical quantity Q instead of writing [Q] = M n we usually write [Q] = n.
CHAPTER 2. LAGRANGIAN THEORY OF CLASSICAL FIELDS 155
x2.3. Fields and lagrangians
Let M be a smooth manifold of dimension n. We formulate eld theory on M . In
the standard physical setup M = M n is ane Minkowski spacetime. For n = 1
this is M 1 , which is (ane) time; it is the appropriate \spacetime" M for classical
mechanics. In usual examples of eld theory we can analytically continue to imagi-
nary time and so obtain a eld theory on Euclidean space. Often eld theories can
also be formulated for curved metrics|of Lorentz or Euclidean signature. We also
allow the spacetime M to be a supermanifold. As we cautioned in the introduction,
our framework is not adequate for many types of constrained mechanical systems.
Fields on M are (smooth) sections of a given ber bundle E ! M . Let
F denote the space of all sections. For example, the basic eld in a -model is
a map : M ! X for some auxiliary manifold X . In this case we simply have
E = M X . We can also study a twisted version in which E is not a product. An
important case is when E is a vector bundle. The basic eld in a gauge theory is
a connection A with gauge group some specied Lie group G. If we x a principal
G-bundle P ! M , then A is a section of a certain associated bundle of ane
spaces. For many purposes it is best not to x P and rather to view the collection
of all connections as a category. However, in this section we simply view P as xed.
There is an evaluation map
(2.13) e : F M ! E:
Again we allow E |and so also F |to be a supermanifold, even if M is an ordinary
manifold.
Of course, a eld theory typically contains several elds i and correspondingly
E = Ei is a ber product. Roughly speaking, each Ei decomposes into an intrinsic
part times an extrinsic part. The intrinsic part is associated to the principal frame
bundle of M via a representation of GL (or Spin), and the representation determines
the type of eld. Thus a scalar eld is associated to the trivial representation. The
basic eld : M ! X in a -model is a typical example. A scalar eld may take
values in a nonlinear space, but the extrinsic values of other types of elds are linear.
A p-form eld is a p-form on M ; it may take values in some vector bundle over M .
Physicists often use the word `vector eld' to refer to a 1-form eld. They also use
`vector eld' or `gauge eld' to refer to a connection , which is a type of eld which
was discussed above. There are also \connectionlike" versions of p-forms (see x6.3),
but only for abelian structure groups. If M is a spin manifold then we can also
consider spinor elds , which are sections of a spin bundle possibly tensored with
another extrinsic vector bundle. The precise choice of spinor bundle varies with
the example. Theories of gravity also include a metric on X , sometimes called a
`gravitational eld'. In theories of supergravity there is also a Rarita-Schwinger
eld , which is a section of an irreducible subbundle of S
T M , where S is a spin
bundle.
The spin of a eld depends on the representation of Spin (or GL) which denes
it. Namely, we decompose the complexied representation under the Spin(2) sub-
group which double covers the group of rotations in some 2-plane. We obtain a sum
of one-dimensional representations which we label by half-integers 0; 1=2; 1; : : : .
Replace these numbers by their absolute values; then the largest number which
occurs is the spin of the eld. Thus a scalar eld has spin 0, a spinor eld spin 1=2,
a p-form eld or connection spin 1, a Rarita-Schwinger eld spin 3=2, and a metric
spin 2.
There are physical reasons to restrict to these values of the spin in quantum
eld theories (see [II-Dynamics of QFT, x2.4]).
In unitary quantum eld theories there is a connection between spin and sta-
tistics , the statistics being whether the eld is even or odd (in the sense of super-
geometry): elds of integral spin are even and elds of half-integral spin are odd.
The usual physical terminology is that even elds are bosons and odd elds are
fermions . This spin-statistics connection is violated in some nonunitary topologi-
cal eld theories.
In some theories there are local gauge symmetries (see Denition 2.93) which
act on the elds. This occurs for elds which are p-forms (p 1) and for connections
in gauge theory. (A p-form is gauge equivalent to + d for any (p 1)-form.)
The gauge symmetries of a connection are as usual. Local symmetries are sections
of a bundle of groups over M . Let F denote the quotient of the space of elds F by
the action of local symmetries. In our notation we treat F as a manifold, though
this may not be true and we may need to work equivariantly on some space which
projects onto F . (For example, in gauge theory we often x a basepoint on each
component of M . Then the group of gauge transformations which equal the identity
at the basepoints acts freely on the space of connections, and the nite dimensional
group of automorphisms at the basepoints acts on the quotient.)
Our formulation is not meant to include theories of gravity, where one of the
elds is a metric and the group of dieomorphisms of M acts as a local symmetry.
We will, however, use the metric as a background eld when we discuss the energy-
momentum tensor. Then in some cases (e.g., with spinor elds) the ber bundle
E ! M varies with the metric. The appropriate modication of this setup is
described later when we discuss the energy-momentum tensor.
We will use a complex which is, basically, the de Rham complex of F M .
Re ecting the product structure of F M , it is a double complex. Suppose rst
that M is oriented, of dimension n. The complex
;jj we will use is then the de
Rham complex shifted by n: in other words,
p;j qj (F M ) is the space of p-forms
on F with values in the space of (n q)-forms on M . We let be the exterior
derivative of F , d the exterior derivative10 of forms on M , and D = + d the total
exterior derivative. For general M we use twisted (n q)-forms: instead of the de
Rham complex of M , we use
jj
M =
M
oM .
More relevant is the subcomplex
loc ;jj of local elements of
;jj , where 2
p;j qj is said to be local if for some k, at any 2 F the value of the twisted (n q)-
form (; ^1 ; : : : ; ^p ) at m 2 M depends only on the k-jet at m of ; ^1 ; : : : ; ^p . Here
î are tangents to F at . If F is the space of sections of a bundle E over M , and
if (k) : J k E ! M is the bundle of k-jets of sections of : E ! M , such an is a
global section over J k E of
pJ k E=M
(k)
M j qj . Note that k cannot be kept xed,
as d maps
pJ k E=M
(k)
M j qj to
p
J k+1 E=M

(k+1)
j q+1j .
M
As mentioned in the Introduction, the double complex
loc ;jj has been investi-
gated by Takens, who proved the following.
10 Our sign convention is that for 2
p (F ) and 2
j qj (M ),
(2.14) d( ^ ) = ( 1)p ^ d:
Theorem 2.15 (Takens). For p > 0 the complex
p;locjj(F M ); d of local
dierential forms is exact except in the top degree jj = 0.
This theorem plays a crucial role in the discussion of generalized symmetries (x2.6),
and also plays a part in the general discussion of this section. We also make use of
the following generalization, proved by the same method.
Theorem 2.16 (Takens). Let Vi ! E , i = 1; : : : ; p, be vector bundles, where
we require p 1. Let V = E Vi be the ber product. For 2 F a section
of E , let V be the space of sections of V ! M . Consider in
loc 0;jj (V M )

the subcomplex of forms (; 1 ; : : : ; p ) which are R-multilinear in i . Then this
subcomplex is exact except in the top degree jj = 0.
In the appendix to Chapter 2 we give a proof of these theorems.
We will use the following picture to depict elements in the double complex:
0 1 F
j0j
d"
j 1j !

..
.
M
We remark that certain topological (terms in) lagrangians do not t into this
formalism. We discuss the necessary modications in Chapter 6.
The basic ingredient in a classical eld theory is a lagrangian (density)
(2.17) 0;j0j (F M );
L 2
loc
it has units of action:
(2.18) [L] = ML2 T 1:
The notation in (2.15) indicates that for each eld in F we have a density L()
on M , while (2.16) indicates that the integral of L() on a xed compact region
of M has the dimension of an action. Typically, L() is a local functional of and,
for most fundamental lagrangians, the value of L() at a point m depends only on
the 1-jet of at m (gravityR is an exception).
Typically, the integral M L() is divergent. Even if, as often, one can restrict
one's attention to elds decaying rapidly at spatial innity, so that the integral be-
tween two space-like hypersurfaces converges, the integration over time will diverge.
Computations given below show that otherwise the expected symplectic structure
on the space of extremals would vanish. However, for a deformation with compact
support of , and if we assume L local, then L() is a density with
R compact sup-
port, which can be integrated. One says that 0 is extremal if L = 0 for any
such deformation of 0 . We now make this more precise. One considers families
of elds [u], with [0] = 0 , and with [u] independent of u in the complement
of a compact region of M . The density dud L([u]) u=0 then has compact support.
The extremality condition is that its integral should
R vanish, for any [u]. It is
suggestive, if abusive, to write the condition dud M L([u]) = 0.
Let us suppose that the elds are sections of a bundle E over M . For a
deformation [u] of 0 , the density dud L([u]) at u = 0 will depend only on the eld
of vertical vectors dud on E , along 0 , which is a section ^ of 0 T (E=M ). The
R-linear form
Z
(2.19) ^ 7 ! d L()
M du
on the sections with compact support of 0 T (E=M ) can be uniquely written as
Z
(2.20) DL(^)
M
for DL a (0 T (E=M )) -valued density. The unicity of DL is clear, and its exis-
tence is a local question. It is proved by the usual integration by parts argument
leading to Euler-Lagrange equations. Extremality of 0 means that DL = 0 at 0
(Euler-Lagrange equations). We note that DL is in
1;j0j, with a value at m 2 M
depending only on some jet of at m, and on the value at m of ^.
This formalism does not hold if, as is typical of superspace formulations of
supersymmetric theories, the elds are sections of a bundle E subjected to con-
straints. The problem is that ^ is no longer arbitrary in 0 T (E=M ). Let M be
the space of extremals. We now explain that if spacetime M is time space, or at
least if there is a suitable notion of space-like hypersurface, then the construction
on M of a closed 2-form|on the model of Chapter 1|uses only conditions on L
which hold for such constrained superspace formulations. The crucial condition is
that the constraints allow to be deformed independently in disjoint regions of M .
(Formally: the sheaf of elds, a subsheaf of the sheaf of C 1 -sections of E=M , is
soft). In practice, for superspace formulations, this means that superelds can be
described in components.
Let H1 and H2 be space-like hypersurfaces with, to simplify the picture, H1
before H2 . In superspace formulations, H1 and H2 should be codimension 1j0
submanifolds of M . We consider
Z H2
(2.21) L() ;
H1
which we at rst assume to converge. If 0 is extremal, and if [u] deforms 0
([0] = 0 ), then
(2.22) d Z H2 L([u])
du H1
will in general not vanish at u = 0, except if the support of the deformation is be-
tween H1 and H2 . For deformations with compact support, it will be the dierence
of boundary terms attached to H1 and H2 :
d Z H2 L([u]) = (^)
(2.23) du H1 2 1 (^) :
In this formula, i is a 1-form on F , dened only on M, whose value at a tangent
vector ^ at an extremal 0 depends only on some jet of ^ along Hi . We now restrict
those 1-forms to (the tangent bundle of) M. Equation (2.23) means that, with
the de Rham dierential on M,
Z H2
(2.24) L= 2 1 ;
H1
so that the exterior derivative i is independent of i. One denes the canonical
closed 2-form on M by
(2.25)
:= i :
Remark 2.26. If the integral of L() on the whole of M did make sense, this
construction would collapse: 2 () would be of the integration of L() from the
innite past to H2 , giving
= 2 = 0.
Remark 2.27. Instead of assuming RHH12 to exist, one may consider only elds
which coincide with a xed extremal at spatial innity. The given construction
then denes i for tangent vectors to M whose support intersects the region of
spacetime in between any two space-like hypersurfaces in a compact set.
Remark 2.28. As in Chapter 1, the construction gives an R-torsor T with con-
nection r on M, whose curvature is
. The choice of a space-like Rhypersurface H
trivializes T , and the trivializations given by H1 and H2 dier by HH12 L().
Remark 2.29. Adding to the lagrangian density an exact term d, for in
loc 0;j 1j ,
does not change the space of extremals. For H a space-like hypersurface, it changes
the corresponding 1-form on M by
Z
(2.30) ^ 7 ! d ([u]) at u = 0 :
H du
The torsor with connection (T; r) for L, and the one (T 0 ; r0 ) for L + d, can
be identied, by sending the trivializationR 0H of T corresponding to a space-like
hypersurface H to the trivialization 00H H of T 0 . In Chapter 1 (particle in an
electromagnetic eld), we saw one example where a change of gauge changes L()
by an exact term, and where the torsor T of Remark 2.28 is dened, but where a
trivialization of T is given by the choice of H and of a gauge along H .
Remark 2.31. We now give a variant of the denition of
, to make clearer on
what it depends. Let [u; v] be a 2-parameter family of extremals. At 0 = [0; 0]
in M, we have the two tangent vectors ^1 = @u @ [u; v ] at (0; 0) and ^2 = @ [u; v ]
@u
at (0; 0). What is
(^1 ; ^2 )? Let H be a space-like hypersurface. It has a past and
a future side, and we identify a neighborhood of H with H ( 1; 1), with h 2 H
corresponding to (h; 0) and the future side of H corresponding to H [0; 1). Let
1 [u; v] be a family of elds, deforming [u; 0], agreeing with [u; v] in the future,
and with [u; 0] in the past: for some a, b with 1 < a < b < 1,
1 [u; 0] = [u; 0];
(2.32) 1 [u; v] = [u; v]; at (h; t) for t > b;
1 [u; v] = [u; 0]; at (h; t) for t < a:
Let 2 [u; v] be a similar family, with the roles of u and v interchanged. Then
Z
(2.33)
(^1 ; ^2 ) = @u @v L(1 ) @u @v L(2 ) at (u; v) = (0; 0) :
The integrand vanishes for t > b (where L(1 ) = L(2 )) and for t < a (where
@v L(1 ) = @u L(2 ) = 0). If [u; v] = 0 at (h; t) for h outside a compact of H ,
one can take = 1 = 2 outside of such a compact, and the integrand has then
compact support.
We now relate the denition of
to the previous one. We rst consider a
one-parameter family [u] of extremals, with [0] = 0 . Let 1 [u] agree with [u]
for t > b, and with 0 for t < a ( 1 < a < b < 0). We have
Z t=0
d L( [u]) at u = 0;
(2.34) H (^) = du 1
t= 1
with H the boundary term attached to H . For commuting vector elds, we have
d(X; Y ) = X(Y ) Y (X ). Applying this to the (u; v) plane, and to @u , @v to
compute d H , we get the formula given for
(^1 ; ^2 ).
The construction of the closed 2-form
on the space M of extremals is reas-
suring, but not particularly useful. What is more interesting are analogs local on
M of
and for
nondegenerate of functions on M corresponding to innitesimal
symmetries. Such local analogs will continue to make sense after a Wick rotation,
in Euclidean eld theory.
We will see in x2.4 that if the space of elds is the space of sections of a bundle
E over M , and if the value of L() at m 2 M depends only on the rst jet of
at m, there is a unique
(2.35) 2
loc1;j 1j (F M )
such that
(2.36) DL = L + d
and that is \linear over functions": at , for ^ a tangent vector to F at ,
identied with a section of T (E=M ), the value of the form (^) at m 2 M should
depend only on ^ at m. More generally, we have the following
Denition 2.37. A form 2
loc 1;jj (F M ) is linear over functions at (; m) if
for every ^ 2 T F and every function f on M ,
(2.38) (;m)(f ^) = f (m)(;m) (^):
This is a local counterpart to the 1-form H on M attached to a space-like
hypersurface H . Indeed, H is the integral of on H . Taking an exterior derivative
in the F direction, we obtain a local analog of the 2-form
: the 2-form
is the
integral on H of ! := in
loc2;j 1j (F M ). This integral is a 2-form on the whole
of F , but it is independent of H only as a 2-form on M.
In the case of gauge theories, even if L and are invariant by gauge transfor-
mations, one should not expect to vanish in the direction of the gauge orbits.
For more general Lagrangians, to localize the construction of
, one should
choose in
loc1;j 1j (F M ) such that DL = L + d . We call the variational
1-form . The pair (L; ) denes a eld theory.
Denition 2.39. L = L + denes a classical (lagrangian) eld theory if
(2.40) (DL)1;j0j = L + d
is linear over functions.
Equation (2.40) is a simple rewriting of (2.36).
Since DL is uniquely determined by L, and is local, it follows from Theorem 2.15
that the dierence between any two choices for the variational 1-form is d-exact. A
d-exact change in leads to a d-exact change in the local symplectic form !, dened
below in (2.44). In the Hamiltonian situation, where spacetime is time space,
the symplectic form
on the space of classical solutions and the R-torsor with
connection whose curvature is
do not depend on the choice of .
For K 2
loc 0;j 1j (F M ) we can form a new lagrangian L + dK . Then +
K is a valid choice of variational 1-form, and the new total lagrangian is L +
DK . In this case the local symplectic form ! is unchanged, as are the equations
of motion (DL)1;j0j . In the Hamiltonian situation the global symplectic form
is unchanged, and there is an isomorphism of the R-torsors for L and L + dK

constructed from K .
If is linear over functions, in the sense of Denition 2.37, we write
(2.41) = ^ :
In this formula is a section of E over F M , is the corresponding local
(1; 0)-form with values in T (E=M ), and is a local (0; j 1j)-form with values
in T (E=M ). If is a real scalar eld : M ! R, then 2
loc0;j 1j (F M ) is
called the conjugate momentum (density) to . Roughly speaking, at least locally
we can choose a coordinate system (on the bers of E ) to write any set of elds as
a collection of real scalar elds and so obtain conjugate momenta.
As explained above, we have the
Denition 2.42. The space M F of classical solutions is the space of such
that the restriction of (DL)1;j0j to fg M vanishes:
(2.43) (DL)1;j0j = L + d = DL = 0 on M M:
Physicists refer to the submanifold M (or M M ) as on-shell ; its complement
in F (or F M ) is o-shell , though that term is usually used to describe all of F
(or F M ). We let M denote the quotient of M by gauge symmetries; as with F
we treat M as a manifold. In some contexts M is called the moduli space.
Dene
(2.44) ! := ; 2;j 1j (F M ):
in
loc
Restricted to the space of classical solutions we have
(2.45) ! = DL on M M;
and so
(2.46) D! = 0 on M M:
We call ! the local symplectic form , since in the Hamiltonian situation its integral
over a spacelike hypersurface is the global symplectic form on the space of classical
solutions. ! also has units of action:
(2.47) [!] = ML2 T 1:
The picture of our on-shell data is:
0 1 2
M
j0j L ! 0
"
j 1j ! !
M
0;j 1j (F M ) are called currents . A current j is conserved if
Elements of
loc
(2.48) dj = 0 on M M:
Below we discuss symmetries and show how they lead to conserved (Noether) cur-
rents. Such currents are local counterparts of generating functions, on the sym-
plectic manifold M of extremal elds, of innitesimal symplectic transformations.
More precisely, their integral on a spacelike hypersurface, to the extent that it
makes sense, is such a generating function. The current being conserved, its inte-
gral is invariant under suitable deformations of the spacelike hypersurface. Another
source of conserved currents is topology. Namely, in some theories the space of elds
has nontrivial topology and topological invariants are constructed by integrating
topological currents. Note that the units of a current are not xed; see (2.3).
x 2.4. First order lagrangians
Let : E ! M be a ber bundle. For s1 and s2 two local sections of E=M dened
in a neighborhood of m 2 M , the relation \in local coordinates, the derivatives of
order k at m of s1 and s2 coincide" is an equivalence relation. The equivalence
classes are called the k-jets of sections of E=M at m. They form the ber at m
of a new ber bundle J k (E ) ! M . Local coordinate systems on J k (E ) can be
obtained as follows. One xes near m 2 M a local coordinate system fx g on M ,
a trivialization E = M F of E , and local coordinates fyag on F . Sections of
E ! M near m are given by functions a (x ) from M to F . The x and @ n a
with jnj k form a local coordinate system.
A local section of E ! M denes a section j k () of J k (E ) ! M , with value
at m 2 M the k-jet of at m.
Denition 2.49. A lagrangian L 2
loc 0;j0j (F M ) depends only on the k -jet of
the elds if there exists a morphism ` : J k (E ) ! Dens(M ) of bundles over M such
that

(2.50) L() = ` j k () :
Example 2.51. Most \fundamental" lagrangians L depend only on the 1-jet of
the elds. (A notable exception occurs in theories of gravity, since the curvature
of a metric depends on the 2-jet of the metric.) A simple example to keep in mind
is a theory of a scalar eld : M ! X with values in a Riemannian manifold X .
Then E = M X and the ber of J 1 E ! E at (m; x) is Hom(Tm M; TxX ). Let
V : X ! R be a (potential energy) function. The theory of the scalar eld in the
potential V has
(2.52) `(m; x; T ) = 12 jT j2 V (x);
where T 2 Hom(TmM; Tx X ).
We now assume that L is a lagrangian density depending only on the rst
jet of the eld . The condition DL = L + d means, in integrated form,
that for U a compact integration domain with smooth boundary, one has, for
2 M; T (E=M ) a tangent vector of F at ,
Z Z Z
(2.53) L() = ( )DL + ( ) :
U U @U
The last sign comes from the fact that ( )d = d( ) . If is assumed to satisfy
(f ) = f( ) for f a function on M , i.e., if (; ) at m depends only on the
value of atR m, this formula makes it clear that is unique. Indeed, the formula
determines @U f( ) for any U and f , hence ( ) for any . As any can be
decomposed into sections of M; T (E=M ) with small support, existence is a
local question on E and M . Locally on M , one can trivialize E as M F . Let
us choose local coordinate systems fx g on M and fyag on F . This gives a local
coordinate system fx ; ya ; ya g on J 1 (E ). By assumption, L = `(j 1 ()) with
(2.54) ` = `(x ; ya ; ya )jdn xj:
The j 1j-form ( ) on M is then given by the standard integration by parts:
(2.55) ( ) =
X @` a (@ )jdn xj:
@ya
x 2.5. Hamiltonian theory

Suppose that spacetime is M = M 1 N for some manifold N , where M 1 is ane
one-dimensional Minkowski space, i.e., the ane real line with its standard metric.
We view M 1 as time and N as space. We also assume that the ber bundle
E ! M 1 N is pulled back from a xed bundle on N , or equivalently that time
translation has been lifted to E . In this case we integrate the local symplectic
form ! to obtain a closed 2-form
on the space of classical solutions M:
Z
(2.56)
= ! 2
2 (M):
ftgN
Typically N is noncompact and so to ensure convergence we only evaluate
on
tangent vectors to M with compact support in spatial directions, or at least with
sucient decay at spatial innity. The hyperbolicity of the classical equations
of motion implies nite propagation speed of the classical solutions, and so the
decay conditions are uniform in time. By (2.46) the right hand side of (2.56) is
independent of t 2 M 1 , and also
is a closed 2-form on M.
A generalized innitesimal local symmetry is a construction as follows (see
Denition 2.93). One gives: (i) a vector bundle V over M ; (ii) for each section
of V , a vector eld X on the space F of elds and a (0; j 1j)-form such that
Lie(X )L = d . One requires that X and be local: the value at m 2 M of
X () 2 (M; T (E=M )) should depend only on the k-jet of and at m, and
similarly for . One also requires X and to be additive in . The basic example
is the algebra of innitesimal gauge symmetries, for which V is the adjoint bundle
of a principal bundle.
In the presence of such symmetries, one cannot hope for the closed 2-form
on M to be nondegenerate. Indeed, the X are tangent to M, because X is a

generalized symmetry (see Denition 2.71). We claim that the X preserve
and
are in the kernel of
, at least for with compact support so that everything is
well-dened. Indeed, decomposing one can assume it has a small support. Let
us choose t such that ftg N does not meet the support of . If we compute
using ftg N , then

does not see and the claim follows. Therefore,
is the
pullback of a 2-form
on the quotient M of M by the local symmetries (or on
the quotient by any subgroup). The best one can hope for, then, is that
is a
symplectic structure.
0;j 1j (F M ) is a conserved current then the associated charge Q is
If j 2
loc j
Z
(2.57) Qj = j:
ftgN
If is in M, then since dj = 0 the right hand side is independent of t. This is
a global conservation law. Local conservation laws are obtained by considering a
domain U N . For simplicity assume the closure of U is compact with smooth
boundary @U . Let
Z
(2.58) qt = j
ftgU
be the total charge contained in U at time t. Write
(2.59) j = dt ^ j1 + j2 ;
where j1 ; j2 do not involve dt. Stokes' theorem applied to integration over the bers
of the projection M 1 U ! M 1 implies
dqt + Z
(2.60) dt ftg@U j1 = 0:
This says that the rate of change of the total charge in U is minus the ux through
the boundary. The units of a charge are the same as that of the current.
In theories with local symmetry the global charge is gauge invariant since we
require the local current to be gauge invariant up to an exact form.
A eld on M = M 1 N is static if its time derivative Lie(@=@t) vanishes.
(Recall that is a section of E ! M 1 N and we have xed a lift of @=@t to E .)
When there are gauge symmetries we say that is static if locally on N there is
a gauge in which Lie(@=@t) = 0. For a p-form eld such a local gauge exists if
and only if the gauge-invariant eld d is invariant under t. The same assertion
holds for abelian connections , where d is replaced by the curvature. The proof
of this assertion is straightforward.11 The local gauge in which is static is unique
up to static local gauge transformations. There does not seem to be an analogous
criterion for static nonabelian connections in terms of the curvature.
Let FN denote the space of static elds and F N the quotient by static gauge
symmetries. In x2.10 we dene the energy of a static eld and so a subspace FEN
of static elds of nite energy.
x 2.6. Symmetries and Noether's theorem
We assume that the space of elds F is the space of sections of a ber bundle E
over M , and that L in
0loc;j0j (F M ) is a lagrangian density. An automorphism
g: F ! F of F is local if for some k the value of g() at m 2 M depends only on
the k-jet of at m, and if the same condition holds for the inverse g 1 of g. A
0;j 1j
generalized symmetry of L is a local automorphism g of F , given with in
loc
such that
(2.66) L(g()) L() = d() :
By the locality assumption, deformations with compact support of and g cor-
respond to each other, and by (2.66), the integral of the variation of L() is pre-
served. It follows that g preserves the space M of extremals. As in x2.3, the
lagrangian density L gives rise to an R-torsor with connection (T; r) on M. A
generalized symmetry acts on (T; r) as follows. Let H be a space-like hypersur-
face. It denes a trivialization 0H of T . We map 0H at to
Z
(2.67) g(0H at ) = 0H () at g() :
H
11 Write
= dt ^ A(t) + B(t); A(t) 2
pN 1 ; B(t) 2
pN ;
(2.61)
= dt ^ P (t) + Q(t); P (t) 2
pU 2 ; Q(t) 2
Up 1
on some small open set U N . Then + d is time independent if and only if
(2.62) A_ dP_ + Q = 0
(2.63) B_ + dQ_ = 0;
whereas d is time independent if and only if
(2.64) B dA_ = 0
(2.65) dB_ = 0:
The integrability condition for (2.63) is (2.65), so we can solve for Q_ . Then we solve (2.62) for P_ ;
the integrability condition is (2.64). (A slightly dierent argument is required for p = 1 since in
that case P 0.)
The resulting map g: T ! T is independent of the choice of H . Indeed, if H1 and
H2 are two space-like hypersurfaces, with H1 before H2 , we have at
Z H2
(2.68) 0H2 0H1 = L() :
H1
As HH12 L(g) = HH12 L() + [ ()]HH21 , this is compatible with 0H1 and 0H2 at
R R R
to be mapped, respectively, to
Z Z
(2.69) 0 H1 () and 0H2 ()
H1 H2
at g(). Indeed, at g(),
(2.70)
Z Z Z H2 Z H2
0H2 () 0H1 () = L(g()) ()
H2 H1 H1 H1
= 0H2 0H1 at :
We now consider the innitesimal analog of this construction.
A vector eld ^
^
on F is local if for some k the value of 2 M; T (E=M ) at m 2 M depends
only on the k-jet of at m. In other words, if q is the projection from J k (E ) to E ,
then ^ is given by a section of q T (E=M ) on J k (E ).
Denition 2.71. A generalized innitesimal symmetry of L is a local vector eld
0;j 1j such that
^ on F , given with ^ in
loc
(2.72) Lie(^)L = d^ on F M:
Here, Lie(^)L is simply (^)L. Such a symmetry should not be expected to inte-
grate to a generalized (local) symmetry, as dened above. For instance, the vector
eld on F corresponding to an innitesimal time translation is local. A nite time
translation on F is not. To make it local, one would need to act both on F and M ,
as we will do later.
One can, however, repeat the previous construction, if one views a generalized
innitesimal symmetry as a generalized symmetry (g; ) depending on an innites-
imal parameter ", with (g(0); (0)) = (Identity; 0). Formally, this means working
over Spec(R["]=("2 )). Doing so, we get a lifting of g(") to T , i.e., a lifting of the
vector eld ^ on M to T , respecting the connection. Such a lifting corresponds to
a function Q^ on M, such that
(2.73) dQ^ = (^)
;
for
the curvature form of T . The function Q^ is the dierence between the lifting
of ^ and its horizontal lifting. For any space-like hypersurface H
Z Z
(2.74) Q^ = H (^) ^ = (^) ^ :
H H
The Noether current of (^; ^) is the local (0; j 1j)-form
(2.75) j^ := (^) ^ :
By construction, its integral on a space-like hypersurface H is the charge Q^, giving
R
the lifting of the vector eld ^ on M to (T; r). For in M, H j^ is hence
independent of H . It follows that j^ is a conserved current.
To prove that the Noether current j^ is conserved, i.e. that j^() is closed when
is extremal, we have used a global argument, relying on a notion of \space-like
hypersurface" H . This leads to extraneous convergence problems. They could be
avoided by using instead hypersurfaces homologous to zero, in a small neighborhood
of a point. The corresponding line bundle on M has trivial curvature, but this
does not spoil the argument. Here is another way out, which gives as well a local
counterpart of (2.73).
Proposition 2.76. Suppose (^; ^) is a generalized innitesimal symmetry of L.
1;j 2j (F M ), the identity
Then, for some ^ in
loc
(2.77) Lie(^) = ^ + d^ on M M
holds on shell. The Noether current j^ := (^) ^ is conserved,

(2.78) dj^ = 0 on M M;
and
(2.79) j^ = (^)! + d^ on M M .
Remark 2.80. Both (2.77) and (2.79) hold as equalities, on M M , of (1; j 1j)-
forms on F M : at in M, they give rise to an equality of j 1j-forms on M
when evaluated against any tangent vector of F at , whether or not it is tangent
to M.
The formula (2.79) is the promised local counterpart of (2.73). We summarize
(2.72) and (2.77) in the diagrams:
0 0 1
F M
j0j Lie(^)L j0j Lie(^)L
" "
j 1j ^ j 1j ^ ! Lie(^)
"
M j 2j ^
M
Proof. The vector eld ^ is tangent to M. Since on M M we have D L =
L + d = 0, we also have on M M
(2.81) Lie(^)D L = Lie(^)L + d Lie(^) = 0;
in other words,
(2.82) d( ^ + Lie(^) ) = 0:
If is in M, then this identity holds when evaluated on any tangent vector 2
(M; T (E=M )) of F at . As ()( ^ + Lie(^) ) is local in , it follows from
Takens that for some ^(; ), local in , one has
(2.83) ^() + Lie(^) () = d^ :
Suppose we have chosen locally on M coordinate systems Ui ,! Rn , trivializations
of E as F Ui and local coordinate systems on F . In terms of those, and of a
partition of unity attached to the resulting covering of E , ^ is given by Takens by
explicit local formulas. Those formulas dene ^ on F M , with (2.77) holding on
M M . A fortiori , (2.77) holds as an identity of (1; j 1j)-forms on M M , i.e.,
when applied to vectors tangent to M M only. Applying to it, we obtain that
^ preserves the (2; j 1j)-form ! on M M up to an exact derivative:
(2.84) Lie(^)! = d( ^) :
As ^ is a vector eld on F , Cartan's formula takes the form
(2.85) Lie(^) = (^) + (^) = (^) + (^)! :
Plugging this in (2.77), we obtain (2.79) on M M .
A vector eld on F M is said to be decomposable and local if it is the sum
of a local vector eld ^ on F and of a vector eld on M . For such a vector eld,
the Lie derivative Lie( ) preserves the bigrading of
loc;jj(F M ). We say that
0;j 1j (F M ) is such that
(; ) is a generalized innitesimal symmetry if in
loc
(2.86) Lie( )L = d :
The Lie derivative Lie( ) is the sum of Lie(^) and of Lie(). As Lie()L = d()L,
it hence follows from (2.86) that
(2.87) Lie(^)L = d( ()L) :
the vector eld ^ on F , and ^ := ()L, form a generalized innitesimal
symmetry. The corresponding Noether current is
j = (^) + ()L = ( )L 0;j 1j :

(2.88)
In this formula, as before, L = L + . We say that is a manifest symmetry of L
if Lie( )L = 0. For a manifest symmetry, one can take = 0, hence ^ = ()L,
and the formula (2.88) for the Noether current then simplies to
j = ( )L 0;j 1j

(2.89) (for a manifest symmetry of L).
We say that is a manifest symmetry of L = L + if Lie( )L = 0, i.e. if
preserves L and . As Lie( ) = Lie(^) + Lie(), this gives
(2.90) Lie(^) = (d() + ()d ) :
On shell, ()d = ()L = ()L = ^, hence Lie(^) = ^ + d( () ):
the formula (2.77) holds for the generalized symmetry (^; ^) and ^ = () . For
this choice of ^, (2.79) becomes
(2.91) j = (^)! d() :
Proposition 2.92. Suppose that L depends only on the rst jet of the elds, that
is the canonical variational form, that = ^ + is a manifest symmetry of L,
and that it is induced by an innitesimal automorphism of the bundle E ! M , that
is, by a vector eld X on E projecting to on M . Then, is a manifest symmetry
of L.
Integrating , this reduces to the statement that if an automorphism of the
bundle E ! M preserves L, it also preserves . As is canonically deduced
from L, this is an application of transport of structures.
This argument seems to assume that X can be integrated, i.e. that exp(tX )e
does not go to innity in nite time. However, as the question is local, integrability
is immaterial. One only needs to know that for t t0 and for suitable E 0 ! M 0 with
E 0 open in E and M 0 in M , exp(tX ) and exp(t) map E 0 and M 0 isomorphically
to Et0 E and to Mt0 M .
Following Zuckerman, we give the formal denition of a local, or gauge, sym-
metry.
Denition 2.93. A generalized innitesimal local symmetry of a theory L = L +
is specied by a vector bundle V ! M and linear maps
7 ! X
(2.94) 7 ! ;
where is a section of V and X is a generalized innitesimal symmetry of L:
(2.95) Lie(X )L = d on F M:
We require that the dependence of X and on be local.
Two examples: In gauge theory V is the adjoint bundle of a principal bundle and
X acts by innitesimal gauge transformations. In theories of gravity V = TM and
X acts by the Lie derivative of .
Proposition 2.96. The Noether current attached to a generalized innitesimal
local symmetry is d-exact on M M .
Proof. This follows directly from Theorem 2.16.
Assuming suitable decay at innity, we see that the Noether charge associated to
a local innitesimal symmetry vanishes.
Finally, we introduce a bracket operation on Noether currents. There is a
standard Poisson bracket of functions on M in the Hamiltonian situation. If
j = j (x) @
jdn xj is a current, then we can view j (x) for x xed as den-
ing a (singular) function on M and so compute Poisson brackets of components of
currents. This is a standard procedure in physics. Such computations go under the
name current algebra . The bracket we introduce on Noether currents is dened in
arbitrary eld theories: we need not be in a Hamiltonian situation and there is no
nondegeneracy assumption. The bracket (2.101) below is often simpler to compute
in practice than the standard Poisson bracket of distributions on M.
Consider a theory L = L + on F M for some manifold M .
0;j 1j (F M ) is a current,
Denition 2.97. We call (j; ) a Noether pair if j 2
loc
is a generalized innitesimal symmetry of L, and if
(2.98) Lie( )L = d on F M;
where
= ( )L 0;j 1j j:

(2.99)
Let (ji ; i ); i = 1; 2 be Noether pairs. We dene the bracket

(2.100) (j1 ; 1 ) ; (j2 ; 2 ) = (Lie(1 )j2 + C ; [1 ; 2 ]);
where
h i0;j 1j
(2.101) C = C (1 ; j1 ; 2 ; j2 ) = (2 ) Lie(1 )L + Lie(2 ) (1 )L j1 :
A straightforward computation shows that the right hand side of (2.100) is a
Noether pair.
Remark 2.102. The formula for C may be changed by an exact term, though
this particular choice is nice: for this choice the quantity dened in (2.99) for the
bracket is
(2.103) Lie(1 )2 Lie(2 )1 ;
where i are the corresponding quantities for i . The precise formula for C is not
valuable; the important term in the bracket of currents is Lie(1 )j2 .
For manifest symmetries C = 0, and in all cases C is exact on-shell (after impos-
ing the equations of motion). The bracket makes the space of Noether pairs into a
Lie algebra (o-shell). For nonmanifest symmetries, this Lie algebra is typically in-
nite dimensional and becomes nite dimensional only after we impose the equations
of motion. Then it is nice to work modulo exact forms. Hence dene an e-Noether
pair [j; ] to be an equivalence class of Noether pairs (j; ) where (j1 ; ) (j2 ; )
if j2 j1 is d-exact. Thus j in [j; ] is determined on-shell by up to an element
of the local cohomology group Hloc 0;j 1j (M M ). Under the bracket (2.100) the set
of e-Noether pairs is a Lie algebra on-shell. If we restrict the symmetries to lie
in some Lie algebra h, then the corresponding set of e-Noether pairs is a central
extension of h by Hloc0;j 1j (M M ).
Example 2.110 in x2.7 illustrates the dierence between on-shell and o-shell
symmetries. Note that the algebra of symmetries generated o-shell is innite
dimensional.
Suppose now that we are in the Hamiltonian situation M = M 1 N and that
(M;
) is symplectic. We work with Noether pairs (j; ) such that projects to a
vector eld on M and j is gauge invariant. Then by (2.73) is the symplectic
gradient of the global charge Q 2
0 (M). The charge Q only depends on the
equivalence class of (j; ) in the set of e-Noether pairs. Let [ji ; i ] be e-Noether pairs
with global charges Qi . Set [j; ] = [j1 ; 1 ] ; [j2 ; 2 ] . Then the global charge Q
associated to j is the Poisson bracket
(2.104) Q = fQ1 ; Q2 gM :
Equation (2.104) relates the bracket on Noether pairs to the Poisson bracket on M;
it follows directly from (2.100), the denition of Poisson bracket, and Stokes' the-
orem.
Recall that if h is a Lie algebra of Hamiltonian vector elds on a symplectic
manifold, then the sub-Poisson algebra of functions whose symplectic gradients lie
in h is a central extension of h by the space of locally constant functions. For a Lie
algebra h of symmetries of L as above, the set of e-Noether pairs [j; ] with 2 h
is a local version of the global central extension dened by the charges.
x 2.7. More on symmetries

This section is not strictly speaking needed for the theoretical development. Rather,
we present examples of symmetries and Noether currents. These examples also
illustrate the calculus we use to compute in function spaces.
Example 2.105. For a system of nonrelativistic point particles (1.13) time trans-
lation is a symmetry. As in the discussion of Example 2.110 it is manifest12 for the
vector eld in (2.120). Explicitly,13
( )x = x;_
(2.107)
( )dt = 1:
12 In Example 2.131 below we discuss time translation as a nonmanifest symmetry.
13 Beware that usually the vector eld ^ is omitted and one simply writes
(2.106) x = x_
instead of the rst equation in (2.107). In (2.106) `' does not denote the dierential on F , but
rather the vector eld ^. Then signs are dierent when commuting d and .
Then we compute directly from (1.13) and (1.15) that Lie( )L = 0. The associated
Noether current
j = ( ) + ( )L
= jx_ j2 + 12 jx_ j2 V (x)

(2.108)
= 1 jx_ j2 + V (x)
2
is minus the hamiltonian . The Hamiltonian formulation of classical mechanics
involves the hamiltonian function (2.108) and the symplectic form (1.16) on the
symplectic manifold M. Noether currents coincide with charges.
In this example the eld is a map M 1 ! X , where M 1 is ane time and X is
a Riemannian manifold. One should view `x' in these formulas as the evaluation
map (2.13)
(2.109) x : F M 1 ! X;
though we often also denote a specic eld x : M 1 ! X with the same letter. The
coordinate t on M 1 is a real-valued function on F M 1 which is constant in the
F direction.
Example 2.110. Let X be a real inner product space of dimension r with inner
product h; i, and for x : M 1 ! X let
L = 12 jx_ j2 21 jxj2 jdtj:

n o
(2.111)
This lagrangian describes a system of r identical harmonic oscillators. The equation
of motion (1.14) is
(2.112) x = x:
(We omit the argument `t' from equations for readability.) Consider the vector
eld ^ on F = Map(M 1 ; X ) given by
(2.113) (^)x = Ax + B x_
for some A; B 2 End(X ). Then
(2.114) (^)x_ = Ax_ + B x:
Assume that A is skew-symmetric and B is symmetric. We compute
(2.115) Lie(^)L = (^)L = d^ on F M 1 ;
for
(2.116) ^ = 12 hB x;_ x_ i 21 hBx; xi:
Thus ^ is a generalized innitesimal symmetry for A skew-symmetric and B sym-
metric. Also,
(2.117) Lie(^) = (^) + (^) = hB x;_ x_ i + hB x; xi:
From (2.116) we nd
(2.118) ^ = hB x;_ x_ i hBx; xi;
so that on-shell|that is, after imposing the equation of motion (2.112)|we have
(2.119) Lie(^) = ^ on M M 1 ; :
as implied by Proposition 2.76.
Only when B = 0 is the symmetry manifest. In that case it is the internal
symmetry generated by an innitesimal rotation in X . For A = 0 and B = id we
see that ^ is the vector eld on F induced from the vector eld @=@t on M 1 . We
make this symmetry manifest by forming the vector eld
(2.120) = ^ + @=@t;
then Lie( )L = 0. In other cases the symmetry is nonmanifest and is not induced by
any symmetry of M 1 . Also, note that on-shell the group of symmetries generated
by (2.113), (2.114) is isomorphic to the unitary group U (r), whereas o-shell it
is innite dimensional. We meet this same phenomenon in the more complicated
situation of supersymmetric theories with no o-shell formulation, i.e., no auxiliary
elds. Then the supersymmetry algebra only closes on-shell and the supersymmetry
is not manifest.
The conserved quantity associated to a diagonal element of the Lie algebra u(r)
with one nonzero entry is the energy of one of the r oscillators.
Next, we give an example in eld theory where in Proposition 2.76 is nonzero.
Example 2.121. Let M = M 2 denote two-dimensional Minkowski spacetime with
coordinates x0 ; x1 . Here x0 is the speed of light times a standard time coordinate.
For convenience x the orientation fx0 ; x1 g and use it to identify twisted forms
with forms. Note that dx0 = dx1 and dx1 = dx0 . Let F = f : M 1;1 ! Rg be
the set of real scalar elds. The free (massless) lagrangian is
L = 12 d ^ d
(2.122) = 21 jdj2 dx0 ^ dx1
= 12 (@0 )2 (@1 )2 dx0 ^ dx1 :

From this we derive

(2.123) = @0 ^ dx1 + @1 ^ dx0
(2.124) ! = @0 ^ ^ dx1 + @1 ^ ^ dx0
and the equation of motion
(2.125) @02 @12 = 0 on M M:
Consider innitesimal translation in the x0 (time) direction. It denes a vector
eld on F M by
( ) = @0
(2.126) ( )dx0 = 1
( )dx1 = 0:
Then a routine computation shows that is a manifest symmetry: Lie( )L =
Lie( ) = 0 on F M . (By Proposition 2.92 we need only check Lie( )L = 0.) But
we can also regard innitesimal time translation as a nonmanifest symmetry ^ by
letting it operate only along F :
(^) = @0
(2.127) (^)dx0 = 0
(^)dx1 = 0:
Then we compute
(2.128) Lie(^)L = d on F M;
(2.129) Lie(^) = + d on M M;
where
= 21 (@0 )2 (@1 )2 dx1 2
loc
0 ;j 1j ;

(2.130)
= @1 1 ;j
2
loc 2j :
A simple example from mechanics illustrates the importance of locality in Def-

inition 2.71.
Example 2.131. Consider a free particle x : M 1 ! R with lagrangian L = m2 x_ 2 dt
and the canonical = mx_ x. Let ^ be innitesimal time translation, considered14
as a motion only along F :
(^)x = x;_
(2.132)
(^)dt = 0:
Then
(2.133) Lie(^)L = mx_ x dt = da^
14 In Example 2.105 we discuss time translation as a manifest symmetry.
for
(2.134) a^ = 21 mx_ 2 ;
so that ^ is a generalized innitesimal symmetry. Imposing the equation of mo-
tion x = 0 we compute
(2.135) Lie(^) a^ = 0 on M M 1 ;
as it must by Proposition 2.76. Now x a time t0 and a function F : M 1 ! R and
consider instead
a~^ = 12 mx_ 2 + F x(t0 ) :

(2.136)
(We could put any function

on F here; we take one of this explicit form for ease of
writing.) Since F x(t0 ) is independent of t we have Lie(^)L = da~^, but now

(2.137) Lie(^) a~^ = F 0 x(t0 ) x(t0 ) on M M 1
is nonzero unless F is constant.

Note that F x(t0 ) is a nonlocal function of x|its value at t 2 M 1 does not
just depend on a nite jet of x at t. This example shows that without locality in
Denition 2.71, the energy would be completely ill-dened.
x 2.8. Computing Noether's current by gauging symmetries
Let us make the following assumptions: (a) the space F of elds is the space of maps
from M to a manifold X , that is, E is a product M X ; (b) the lagrangian density
L depends only on the rst order jet of elds; (c) is the canonical variational
form; (d) a Lie group G acts on X ; it preserves L, hence .
Let P be a principal G-bundle on M , and let E P be the twisted form of E
dened by P : a section p of P over an open subset U of M denes an isomorphism
(p): EU ! EUP , and for g a map from U to G, (pg) = (p)g. Let F1 be the space
of pairs (A; ), where A is a connection on P and where is a section of E P . On
F1 M , let L1(A; ) be the lagrangian density whose value at m in M is computed
as follows. Choose a local trivialization p of P which is horizontal (for A) at m.
The section (p) 1 of E can be identied with a map from M to X , and
(2.138) L1 (A; ) at m := L((p) 1 ) at m :
The assumptions (b) and (d) ensure that the second member does not depend on
the choice of p. The value of L1 (A; ) at m depends only on the value of A at m
and on the rst jet of at m. We let 1 be the canonical variational 1-form.
In local coordinates, L is computed as follows. Over a local coordinate system
x: U ,! Rn of M , one chooses a trivialization of P . This identies E P with M X .
The connection A is a 1-form on M with values in Lie(G), which the action of G
on X turns into a 1-form on M with values in vector elds on X . The original
lagrangian L is of the form
(2.139) L() = `(; i )dx
with i = @=@xi . The new one is
(2.140) L1 (A; ) = `(; i + Ai ())dx :
P
It follows that if the variational 1-form () is cI (; i )dxI , then
X
(2.141) 1 (A; ) = cI (; i + Ai ())dxI :
The gauge group GP of automorphisms of P acts on F1 by transport of struc-
tures. As L1 and 1 have been dened (in terms of structures preserved by GP ),
they are preserved. If P is the trivial principal bundle G, so that E P = E , and if
A is the trivial connection A0 , the group G acting on P (on the left) by constant
gauge transformations respects A0 ; its action on is the action of G on F , and
(2.142) L(A0 ; ) = L() :
The construction of F1 , L1 from F , L and the action of G is called \gauging
the symmetry". Our aim in this section is to explain how the Noether current
attached to the manifest innitesimal symmetry of L given by the action of Lie(G)
can be computed by dierentiating L1 (A; ) in A. We will work in a more general
framework than the one above, which covers as well the case where the elds are a
map : M ! X and a TX -valued spinor.
Let P be a xed principal G-bundle on M , and let con(P ) be the bundle over
M whose sections are the connections on P . Let us rst assume only that E is a
ber space over con(P ), so that the space of elds F bers over the space A of
connections, that the action of the gauge group GP of automorphisms of P on A is
lifted to an action of F , that this lifted action is local|g at m 2 M depends only
on some jet of g and at m|and that it preserves a lagrangian density L. For A
in A, let FA be the ber of F over A at A. It is the space of sections of EA , the
inverse image in E of the section A of con(P ). The lagrangian density L induces a
lagrangian density LA on FA M . Let in FA be an extremal for LA. This means
that DL at , a morphism of vector bundles from T (E=M ) to densities on M ,
factors through A T (con(M )=M ). As con(M )=M is an ane space bundle over M ,
the latter is independent of A. It is the vector bundle of 1-forms with values in the
adjoint bundle gP .
The bundle Hom(
1 (gP );
j0j ) is identied with
j 1j(gP_ ): to an element
0
in
j 1j (gP_ ) corresponds the morphism
7! ^: h; 0 i. To in FA , extremal
for LA, we15 have attached a section J1 of
j 1j
gP_ . The gauge invariance of L
gives:
Proposition 2.144. One has
dA J1 = 0:
In particular, if the innitesimal gauge transformation is horizontal for A, then
J1 ( ) = h; J1 i is a conserved current.
15 Physicists typically summarize the denition of J1 by the equation
Z
(2.143) S () = J1 () A ;
M
where S is the action.
Proof. Let be an innitesimal gauge transformation, a section of gP . Then,
induces a vector eld on A whose value at A is dA . If has compact support,
one has
Z
(2.145) h dA ^ J1 i = 0 :
Suppose indeed that g(u) is a family of sections with compact support of the gauge
group, with dud g(u) = at u = 0. By gauge invariance of L and extremality of ,
we have
Z Z D E Z d
(2.146) h dA ^ J1 i = d
DL; du g(u) = du L(g(u)) = 0 :
By integration by parts, i.e., using that

(2.147) d h; J1 i = hdA ^ J1 i + h ^ dA J1 i ;
equation (2.145) gives that for any with compact support,
Z
(2.148) h ^ dA J1 i = 0 ;
so that dA J1 = 0. The second statement follows from (2.147).
We now give conditions under which the current J1 ( ) can be interpreted as a
Noether current. We suppose that:
(a) the ber bundle E over M is a ber product con(P ) E 0 , so that the space F
of elds is a product A F 0 ;
(b) the action of GP on E is deduced from its action on con(P ), and from an action
of GP on E 0 : the value at m of g0 , for g in GP and 0 a section of E 0 , depends
only on the values of g and 0 at m;
(c) the value of L(A; 0 ) at m depends only on the value of A at m, and of the 1-jet
of at m.
The assumptions (a) (b) (c) hold for a lagrangian obtained by gauging a symmetry.
The assumption (c) ensures that the canonical variational 1-form depends only
on 0 , not on A. It agrees with the A for the LA on F 0 = FA .
As GP acts on F 0 , an innitesimal gauge transformation induces a vector eld
on F 0 . If is horizontal for A, i.e. xes A, then the vector eld ^ on F 0 is a
^
manifest symmetry of LA.
Proposition 2.149. If the lagrangian density L is invariant by gauge transforma-
tions, if the assumptions (a) (b) (c) above hold, and if is an innitesimal gauge
transformation which xes the connection A on P , i.e. if dA = 0, then the Noether
current for the manifest symmetry ^ of the lagrangian LA is the conserved current
J1 ( ) of Proposition 2.144.
Proof. Let be an innitesimal gauge transformation. By gauge invariance of L,
(2.150) Lie( )L = ( )L = ( dA )L + ( ^ )LA = 0 :
As L depends only on the 0th -jet of A, we have
(2.151) ( dA )L = ( dA )DL = dA ^ J1 :
We also have
(2.152) ( ^ )LA = ( ^ )DLA ( ^ )d A = DLA ( ^ ) + d( ^ ) A ;
and so
(2.153) d( ^ ) A = dA ^ J1 DLA( ^ ) :
For any function f on M , we have (f )^ = f ^ . Taking (2.153) for f , and
subtracting f times (2.153) for , we get for any f
(2.154) df ^ ( ^ ) A = df ^ J1 ( ) ;
hence
(2.155) ( ^ ) A = J1 ( ) :
For dA = 0, the left side is the Noether current corresponding to the manifest
symmetry ^ .
Elements in the center of g determine innitesimal \global" gauge transforma-
tions which act trivially on all of A, and so by Proposition 2.144 lead to currents
dened on the entire space of classical solutions of a theory L. In x4.4 we relate
these currents to electric charge .
x 2.9. The energy-momentum tensor
First approach
Let M denote (ane) n-dimensional Minkowski spacetime. Fix a coordinate sys-
tem x0 ; x1 ; : : : ; xn 1 with respect to which the metric is
(2.156) g = (dx0 )2 (dx1 )2 (dxn 1 )2 ;
and the natural density is
(2.157) jdn xj = j dx0 ^ dx1 ^ ^ dxn 1 j:
Here x0 = ct is the speed of light times the standard time coordinate. Of course,
the metric and density have their geometric units, a power of length:
(2.158) [g] = L2 ; [ jdn xj ] = Ln :
Suppose L = L + describes a eld theory on M . We assume that the Poincare
group is a manifest symmetry group of the theory. Now the constant vector eld
(2.159) @ = @x@
on M induces a vector eld on F M , which we are assuming is a symmetry, and
we denote minus the 0; j 1j -component of the associated Noether current by
0
(2.160) dx = g (@ 0 )jdn xj
for some functions
(2.161) : F M ! R:
The tensor = ( ) is called the energy-momentum tensor. The conservation
law (2.78) is
X
(2.162) @ = 0:

The units of the functions are those of a spatial momentum density:
(2.163) [ ] = LnM2 T = ML T Ln 1 :

1
Physically correct units for the components involving x0 = ct are obtained by

re-expressing in the coordinate system ft; x1 ; : : : ; xn g.
A vector eld = @ on M denes a current
(2.164) = dx :
Lemma 2.165. is symmetric if and only if is conserved for every inni-
tesimal Lorentz transformation .
Proof. Let g denote the Minkowski metric. Then a skew form B = (B )
corresponds to an innitesimal Lorentz transformation
(2.166) = B g x @ :
Using (2.162) we nd
d( ) = d( B g x dx )
(2.167) = B g g jdn xj
= B jdn xj:
The conclusion follows.
In practice is symmetric only for theories of scalar elds. For other types
of elds there is an \improved" symmetric energy-momentum tensor, as we discuss
below.
The various components of the energy-momentum tensor have a physical in-
terpretation, as the name suggests. (See the general discussion at the end of x1.4.)
In this paragraph we work on-shell. We also insert factors of c to obtain correct
physical units. Let
(2.168) jdn 1 xj = j dx1 ^ ^ dxn 1 j
be the canonical density on a time slice fx0 = constantg. Then c00 jdn 1 xj is
the energy density and i0 jdn 1 xj (i 1) the momentum density . Assuming
suitable decay at innity, we integrate these densities over space to obtain minus
the charges associated to the innitesimal translations @ :
Z
E= c00 jdn 1 xj;
fx0 =constantg
(2.169) Z
Pi = i0 jdn 1 xj
fx0 =constantg
(The minus sign comes since i is minus the Noether current associated to @i ,
whereas the momentum Pi is the Noether charge associated to @i . Recall that E is
the Noether charge associated to c@0 .) The conservation law asserts that on-shell
E and Pi are independent of time. E is the total energy or hamiltonian and Pi the
momentum in the ith spatial direction. The square of the (rest) mass of a eld
conguration is
X
(2.170) M 2 = E 2 =c4 Pi2 =c2 ;
i
as in (1.45). The remaining components of comprise the stress tensor (ij )i;j1 .
Second approach
For eld theories which can be formulated in an arbitrary background metric, an-
other approach to energy-momentum tensors is available. The relation with Noether
currents is to be obtained by arguments parallel to those of x2.8, but with the group
of dieomorphisms replacing the group of gauge transformations.
Let met(M ) ! M denote the ber bundle of Lorentzian metrics on the tan-
gent bundle. Taking the inverse metric, we will view it as an open subbundle of
the bundle of contravariant symmetric 2-tensors g , and identify its relative tan-
gent bundle T met(M )=M with the pull-back from M of the bundle Sym2 (T ) of
contravariant symmetric 2-tensors. In a Euclidean context, one would rather take
for met(M ) the ber bundle whose sections are the Riemannian structures on M .
In x2.8, we assumed the ber bundle E over M to be a product con(P ) M E 0 .
To similarly assume here that E = met(M ) M E 0 is possible only at the cost of
excluding spinor (or Rarita-Schwinger) elds. Indeed, for V a vector space given
with a symmetric bilinear form g, the corresponding space of spinors depends on g
(and on a lifting of the structural group from an orthogonal to a spin group). When
g varies, the spaces of spinors form a vector bundle on the space of g's. Because of
this, we will only assume that E is a ber bundle E ! met(M ) over met(M ).
In x2.8, we assumed that the gauge group acts on F . Here, if among the
elds there is a connection over some principal G-bundle P , it is not convenient
to assume an action on F of the group of dieomorphisms. What will act is an
extension of the group of dieomorphisms by the gauge group: the group of pairs
(f; ') of a dieomorphism f and of a lifting ': P ! f P of f to P . If P is trivial,
a trivialization of P denes a splitting of this extension. Often, it is more natural to
use that a connection on P provides a lifting to P of innitesimal dieomorphisms
(= vector elds on M ), the horizontal lifting. One should however keep in mind
that this horizontal lifting is not compatible with brackets of vector elds.
Because of this, we will at rst only assume given a lagrangian density L on
F M which obeys the weak version of dieomorphism invariance explained below.
A dieomorphism of M acts on Met(M ). An innitesimal dieomorphism of M ,
that is, a vector eld , induces on Met(M ) the vector eld whose value at a metric
g is Lie( )(g_ ), for g the inverse metric. One has
(2.171) Lie( )(g_ ) = (r ) + (r ) ;
where the tensor indices are moved up and down by the metric tensor and its inverse.
At m, this is easily checked in a coordinate system in which the metric g is constant
+O(x2 ), so that the Christoel symbols vanish at m. The dieomorphism
invariance assumption is the following: for any vector eld on M , there is a local
vector eld ~ on F , projecting to Lie( )(g_ ) on Met(M ), which is a generalized
symmetry of the lagrangian density L.
For g in Met(M ), let Fg be the ber of F ! Met(M ) at g. If Eg is the inverse
image of g in E , then Fg is the space of sections of the ber bundle Eg over M .
The lagrangian density L induces a lagrangian density Lg on Fg M .
Suppose that in Fg is an extremal for Lg . The Euler-Lagrange equation
(2.172) DL: T (E=M ) ! densities on M

then vanishes on T E= met(M ) , hence factors through

(2.173) Dg L: g T met(M )=M ! densities on M :
This expresses that if [u] is a deformation with compact supports of , inducing
a deformation g[u] of the metric, then
Z
d
(2.174) du L([u]) at u = 0
depends only on dud (g[u]) at u = 0, and not on the concomitant variation of the
other elds. This allows to unambiguously attach to (assumed to be extremal
for Lg ) a symmetric 2-tensor T with values in densities on M such that (2.173)
is16
(2.176) a 7 ! 12 a T
for a in g T (E=M ) a symmetric contravariant 2-tensor. In local coordinates fx g
on M we write the energy-momentum tensor as
(2.177) Tg = T (g; ) dx
dx g (x);
where g is the canonical density given by the Riemannian metric g.
16 Physicists write (2.176) as
Z
(2.175) S () = 21 T () g ;
M
where S is the action. In our denition we restrict to which are extremal for Lg .
Remark 2.178. We have dened T only on-shell, that is, for an extremal of
Lg . This does not provide a unique dierential expression in giving T . Suppose,
however that we obtained a dierential expression in which (even locally on M )
gives T for any extremal. Suppose also that depends only on the k-jet of the
section of Eg : is a section on J k (Eg =M ) of the pull-back of Sym2 (T )
densities
on M . Then, if any k-jet is the k-jet of an extremal (the validity of this depends on
which Cauchy data the Euler Lagrange equations for Lg require), then is unique
and denes T o-shell.
Proposition 2.179. Under the assumptions above, if in Fg is extremal for Lg ,
the corresponding T obeys the following conservation law: its covariant derivative
T ; is such that its contraction T vanishes.
Proof. Let be a vector eld with compact support on M . It induces the vector
eld Lie( )(g_ ) on Met(M ) and, by assumption, this vector eld can be lifted to
^ on F in such a way that Lie(^)L is an exact dierential. It follows that
Z
(2.180) hT; Lie( )(g_ )i = 0 :
M
Our computations will be local. Let us choose an orientation of M to identify
densities with forms of maximal degree n. Instead of viewing T as a section of
1

1

n , one can use g to identify
1 with T 1, and view T as a section of
1
T 1

n , mapped isomorphically by contraction to
1

n 1 . Raising and
lowering indices by g, Lie( )(g_ ) similarly corresponds to a symmetrization of r
in
1
T 1, and in (2.180), the symbol h i denotes contraction of
1 with T 1 while
simultaneously wedging
n 1 and
1 . One has, for h i denoting only contraction
of
1 and T 1,
(2.181) r hT; i = hrT; i + hT; r i :
After wedging, r hT; i becomes d hT; i and (2.181) gives
Z
(2.182) hrT wedged ; i = 0 :
M
As this holds for all , it follows that rT in
1

1

n 1 , projected to
1

n
by

7!
(^ ), vanishes. This is equivalent to T = 0.
Corollary 2.183. Let us view T as a 1-form with values in (n 1)-forms (or
better, with values in
j 1j). If in Fg is extremal for Lg , and if is a Killing
vector eld, i.e., an innitesimal isometry, then T ( ) is closed.
Proof. As r = 0, we have
(2.184) (rT )( ) = r(T ( )) ;
which, applying ^:
1

n 1 !
n , gives
(2.185) 0 = dT ( ) :
We managed to relate the conserved currents in Corollary 2.183 to Noether's
currents only under the following restrictive assumptions: (a) E is a ber product
met(M ) M E 0 , making the space F of elds a product Met(M ) F 0 ; (b) the
Lagrangian density L(g; 0 ) at m 2 M depends only on the value of g at m and on
the 1-jet of the section 0 of E 0 at m; (c) we take for the canonical choice; (d)
to each vector eld on M is attached, by a local rule, a vector eld on F 0 , so
that (; Lie( )(g_ ); ) is a manifest symmetry of L; (e) the map 7! is linear
over functions.
Remarks 2.186. (i) (a) is needed to make sense of (b); it fails for spinor elds.
(ii) Condition (d) fails for tensor elds, if 7! is given by a Lie derivative.
However, it holds for connections on a xed principal G-bundle P , if is dened
as follows. Given a connection rA , use it to lift to P . The ow exp(t ) generated
by is then lifted to P , and one takes
(2.187) = dtd (exp( t ) (rA )) at t = 0 :
The vector eld is obtained by contracting the curvature 2-form FA with , and
is hence linear over the functions.
(iii) Condition (a) allows us to dene T o-shell as the restriction to the tangent
space to Met(M ) of DL (or L; this amounts to the same by (b)).
Assuming (a) to (e), we now repeat the arguments of x2.8. The manifest
symmetry of L means the vanishing of a Lie derivative of L: at (g; 0 )
(2.188) d( )L + DL( Lie( )g_ ) + ( )L = 0 :
Writing I ( ) for this identity, we now express that I (f ) fI ( ) vanishes.
The rst term contributes df ^( )L. As Lie( )(g_ ) is r , changed to a twice
contravariant tensor and symmetrized and that r(f ) f r = df
, the second
term contributes hT; grad f
i. As L = DL d , with DL linear over functions,
and that ( )d d( ) , the third term contributes df ^( ) . With the notations
of Corollary 2.183, the contribution of the second term can be written df ^T ( ),
giving
(2.189) df ^ (( )L + T ( ) + ( ) ) = 0 :
As this holds for all f , we can suppress the df ^. If is an innitesimal isometry of
g, then (; ) is a manifest innitesimal symmetry of Lg , and we get
Proposition 2.190. Under the assumptions made, if is a Killing vector eld,
the conserved current T ( ) of (3.5) is the opposite of the Noether current of the
corresponding symmetry of Lg .
x 2.10. Finite energy congurations, classical vacua, and solitons
In this section we work in the Hamiltonian framework M = M 1 N . Let t be an
ane coordinate on M 1 and @t the vector eld which generates unit time transla-
tion. Recall that a eld is static if
(2.191) @ = (^ ) = 0;
@t t
where ^t is the vector eld on F induced by the action of @t . Let FN denote the
space of static elds and F N the quotient by gauge symmetries. Equation (2.191)
asserts that the vector eld ^t vanishes on FN .
Consider a theory L = L + which we assume is manifestly invariant under
time translation.17 The energy density is minus the canonical Noether current
for @t :
(t )L 0;j 1j;

(2.192) =
where t = @t ^t . The energy at time t of a eld is (see (2.169))
Z
(2.193) E (t) = ():
ftgN
For a static eld the energy is constant in time. Let FEN denote the space of
nite energy static elds and FE N the quotient by gauge symmetries. These are
subspaces of FN and F N , respectively. Dene MN FEN (and MN FE N ) to
be the space of static classical solutions of nite energy.
The energy density of a static eld is simply related to the lagrangian density.
Proposition 2.194. (i) We have
(2.195) = (@t )L on FN ;
(ii) If is a critical point of energy on FEN , then is a solution to the classical
equations, i.e., 2 MN . Conversely, every element of MN is a critical point of
energy.
Proof. By (2.191) we have (^t ) = 0 on FN . Hence on FN
(2.196) = (t )L = (@t )L + (^t ) = (@t )L;
which is (i). For (ii) we rst note that on FN we have by (2.191) that
0 = Lie(t ) = d(@t ) + (@t )d (^t ) (^t )
(2.197)
= d(@t ) + (@t )d :
Using (2.195) we compute that on FN
= (@t )L
(2.198) = (@t )(DL)1;j0j (@t )d
= (@t )(DL)1;j0j + d(@t ) :
This is the integration by parts equation for the functional E on FN |analogous
to (2.40) for the functional L|and so the Euler-Lagrange equation for E is
(2.199) (@t )(DL)1;j0j = 0;
17 This excludes, for example, the theory of a nonrelativistic particle moving in a time-dependent
potential.
which is equivalent to the equation of motion (DL)1;j0j = 0.
The global minima of energy are called vacuum solutions . For eld theories on
Minkowski spacetime the set of vacuum solutions is referred to as the classical mod-
uli space of vacua . We denote it Mvac . A vacuum solution in Minkowski spacetime
is usually assumed to be Poincare invariant (if the theory is Poincare invariant).
This means that scalar elds are constant, gauge elds are gauge equivalent to a
trivial connection, spinor elds vanish, and p-form elds (p 1) are exact. The
classical moduli space is then the space of constant values of the scalar elds. Now
the energy density (0 ) of a constant scalar eld 0 : M ! X satises
(2.200) dt ^ (0 ) = V (0 ) jdtj jdn 1 xj
for a potential energy function V : X ! R. Assume18 that V 0. The only nite
energy constant scalar elds have V (0 ) = 0, and so
(2.202) Mvac = V 1 (0)
is the classical moduli space. If there is also a connection eld in the theory with
gauge group G, then G acts on X and V is an invariant function. In that case the
eect of dividing by global gauge transformations is that the classical moduli space
is the quotient
(2.203) Mvac = V 1 (0)=G:
Returning for a moment to a general eld theory L = L + on a manifold M , x
a eld conguration 0 2 F . Then there is a perturbation theory for the uctuations
around 0 in which the space of elds is T0 F and the N th order perturbative
lagrangian is the N th order jet of L at 0 .
On Minkowski spacetime M we often perturb around a vacuum solution.19 Re-
call that at a vacuum all gauge elds A0 are trivial, and we can use them to trivialize
all bundles. Then any scalar 0 is a constant in a manifold X . In the perturbation
theory the uctuations of a trivial connection A0 on a principal G-bundle P lie
in
1M (ad P )
=
1M (g), where we use the trivialization. The uctuations ~ of the
constant scalar 0 lie in T0 X . Spinor elds, Rarita-Schwinger elds, and p-form
elds all vanish at a vacuum conguration and we consider the elds in the original
lagrangian to be uctuations about zero. Now since we are at a vacuum solution
the perturbative lagrangian starts out with quadratic terms. In the quadratic ap-
proximation ~ is typically a free scalar eld in the Euclidean space T0 X with
mass20 Hess0 V ; there may be higher derivative terms as well. There is a mass
matrix for the spinor elds which also depends on 0 . The massless uctuations
of the gauge eld lie in
1M (g0 ), where g0 is the Lie algebra of the stabilizer
subgroup G0 at 0 of the G action on X . Often G0 is called the unbroken gauge
group, though this terminology is confusing. Other components of the gauge uctu-
ations are massive due to the Higgs mechanism , which we illustrate in the following
example.
18 In many problems the energy density is nonnegative for all elds: t () 0 for all , where
(2.201) dt ^ () = t () jdtj jdn 1 xj:
19 This is the rst step in a perturbative construction of a quantum theory around the chosen
vacuum.
20 Free elds and this use of `mass' are discussed in x3.
Example 2.204. We work on n-dimensional Minkowski spacetime M n. The elds
are a complex scalar eld , a real spinor eld21 , and a T T connection A.
As in any gauge theory, we work on the quotient of the space of elds by gauge
transformations. Let T T act on C by (1 ; 2 ) z = 1 z2 1 , and suppose that is
a section of the associated hermitian line bundle. The spinor eld is not coupled
to A. Let
(2.205) L = k dA k2 21 jFA j2 + 21 h D= i kk2 (1 kk2)2

jdn xj:
The potential energy is
(2.206) V (0 ) = k0 k2 (1 k0 k2 )2
and V 1 (0) consists of the origin and the unit circle. The classical moduli space
is the quotient of V 1 (0) by T T and so consists of two isolated points. They
are the two classical vacua. At the origin the perturbative lagrangian is the same
as (2.205), except that A should now be viewed as a perturbation A = A0 + of a
trivial connection; is an (iR iR)-valued 1-form on M . We use A0 to construct
a trivializing section of the hermitian line bundle, and so view as a map M ! C .
Then the perturbative lagrangian is
(2.207) L0 = k d + k2 21 jdj2 + 12 h D= i kk2 (1 kk2)2

jdn xj:
From the quadratic part of L0 we read o that is massive (with mass 1), is
massless, and is massless. The entire gauge group T T is unbroken.
Now we expand around 0 = 1, which is most easily accomplished by substi-
tuting = 1 + ~ into L0:
L1 = k d~ + (1 + ~) k2 21 jdj2 + 21 h D= i

k1 + ~k2 (1 k1 + ~k2 )2 (1 + ~) jdn xj
(2.208)
= k d~ k2 12 jdj2 + 12 h D= i 4j(Re ~)j2

+ k1 2 k2 jdn xj + higher order terms:
In the last line we wrote only the quadratic part; it contains the information about
masses. Also, we wrote = (1 ; 2 ); the last term is the norm square of
1 = 1 2 . The imaginary part of the complex scalar eld is gauged away by
constant gauge transformations (which preserve the trivial connection A0 ). Thus
the only scalar eld is the real scalar eld Re ~ with mass square
p 4. The spinor
isp now massive with mass 2 (and so mass square 4). Set22 2 = 1 2 and
2 = 1 + 2 . Then the quadratic part of the lagrangian involving is
(2.209) 1 jd j2 1 jd j2 + 2k k2:
2 2
21 The precise spinor representation S of Spin(1; n 1) is not crucial, but we do assume that there
is an invariant skew form V2 S ! R so that we can write a mass term.
22 Here the massless eld is canonical, but the massive eld is somewhat arbitrary.
So we see that has mass square 4 and is massless. Note that lies in the Lie
algebra of the stabilizer of 1 2 C , which is the diagonal T T T. This is the
unbroken gauge group. The appearance of a mass for is an example of the Higgs
mechanism.
In general, if H is a group of global symmetries of L, then H acts on the space
of static classical solutions MN . For 2 MN we have the subgroup Stab H
which xes . We say that spontaneously breaks H down to Stab , and that
Stab is the group of unbroken symmetries. (For example, we asserted above that in
Minkowski spacetime Poincare symmetry is unbroken at a vacuum solution.) If 0 is
a vacuum solution, the homogeneous space H= Stab 0 is embedded in V 1 (0) by
the H action. Then the perturbative scalar elds with values in h=stab 0 T0 X
are massless. These elds are called (classical) Goldstone bosons . They are massless
scalar elds guaranteed by the symmetry. Of course, there may be other massless
scalar elds which are not related to symmetry.
Example 2.210. Consider a theory of a complex scalar : M ! C with poten-
tial (2.206). Now there is a global T symmetry and the moduli space consists of
the origin and the unit circle. We do not divide by the global symmetry; rather, at
a classical vacuum on the unit circle there is a single (real) Goldstone boson eld
due to the symmetry. At the origin T acts trivially and the entire complex scalar
is massive.
A soliton is a static classical solution whose energy is not a global minimum.
In many examples the space FE N of nite energy static elds is not connected,
and a soliton is a minimum energy conguration in a component where the global
minimum is not achieved. For example, consider a scalar eld : M 2 ! R on
two-dimensional Minkowski spacetime with lagrangian density
L = 21 jdj2 V () jd2 xj:

(2.211)
Suppose V 0 and V 1 (0) = fa; bg. Then the space of nite energy static elds
has 4 components: A static eld (t; x) = (x) depends only on the spatial variable,
1
and the nite energy condition means xlim !1 (x) and x!lim1 (x) lie in V (0) =
fa; bg. There are 2 vacuum solutions (x) a and (x) b. There are solitons
with x!lim1 (x) = a; xlim !1 (x) = b and also (anti-)solitons with x!lim1 (x) =
b; xlim
!1 (x) = a.
x 2.11. Dimensional reduction
Suppose L = L + is a Poincare invariant eld theory on n-dimensional Minkowski
spacetime M n . We obtain a theory on M n 1 as follows. As usual let x0 ; : : : ; xn 1
denote coordinates on M n , and x0 ; : : : ; xn 2 coordinates on M n 1 , viewed as the
quotient of M n by translations in the xn 1 direction. Let ^n 1 denote the vector
eld on F induced by the action of @n 1 , and dene
(2.212) Fn 1 = f 2 F : Lie(^n 1 ) = 0g:
These are the elds which are constant in the xn 1 direction. We identify Fn 1
with a space of elds on M n 1. Then the dimensionally reduced theory is
(2.213) Ln 1 = L
@n 1 2
0loc (Fn 1 M n 1 ):
A scalar eld on M n reduces to a scalar eld on M n 1 , but for elds of higher
spin the identication of Fn 1 with elds on M n 1 is more complicated. For
example, a 1-form on M n reduces to a 1-form plus a scalar eld on M n 1.
More generally, if M ! M 0 is a ber bundle with a section of Det T (M=M 0),
then we can reduce a eld theory on M to a eld theory on M 0 . Instead of restricting
to elds constant along the bers (dimensional reduction) we can also include elds
which uctuate along the bers (compactication); from the point of view of M 0
there is an innite number of such elds.
Appendix: Takens' acyclicity theorem
This appendix gives the proof of Theorem 2.16. Our setting is slightly dierent
from that of Takens, who works on the innite jet bundle JE=M 1 of E=M . To go
from local to global results, he uses that if 0 ! F0 ! ! Fn ! Fn+1 is an
exact sequence of sheaves, with Fi soft for i n, the sequence of groups of global
sections is again exact. In order that the space of \local" lagrangian densities, or
forms, be a space of global sections of some sheaf, he is led to dene \local" to
mean \depending only on k-jets of elds and their variations, where k is bounded
1 ". For us, k is globally bounded, and the sheaf theoretic argument
locally on JE=M
has to be unraveled, replaced by a direct use of partitions of unity.
We x a submersion E ! M . For xed p > 0, we are to prove that the
corresponding complex (
p;loc ; d) is acyclic, except in top degree. It will be more
convenient to prove a more general statement. Fix vector bundles V1 ; : : : ; Vp on E .
Let V be their product. The complex (
0loc; ; d) for V ! M is the complex of forms
on M depending (locally) on a section of E=M and on sections 1 ; : : : ; p of
Vi . For some k, the value of (; 1 ; : : : ; p ) at m 2 M should depend only on
the k-jet of and the k-jet of 1 ; : : : ; p at m. We denote
loc;multi the subcomplex
consisting of the depending R-linearly on each of 1 ; : : : ; p . The following is a
restatement of Theorem 2.16.
Theorem 2.214. If p > 0, the complex
loc;multi is acyclic except in top degree.
If we take all Vi to be the relative tangent bundle TE=M , then (
p;loc ; d) is a direct
factor of
loc;multi: it is the antisymmetric part for the action of the symmetric
group Sp . The acyclicity of (
p;loc ; d) for p > 0 (except in top degree) hence follows
from the theorem.
Proof. For e0 in E , with image m0 in M , one can choose (a) a neighborhood U
of m0 , and a local coordinate system x: U ,! Rd , (b) a neighborhood W of e0 ,
a decomposition W = F U and a local coordinate system F ,! Rf , and (c)
trivializations of the Vi on W . In such a local coordinate system, in
loc;multi can
uniquely be written as a nite sum
X
(2.215) (; 1 ; : : : ; p ) = n1 ;:::;np ()@ n1 1 : : : @ np p :
In the formula, each ni is a multi-index (ni1 ; : : : ; nid ) and @ ni := @1ni1 : : : @dnid is the
corresponding iterated
P derivative in the coordinate system (x1 ; : : : ; xd ). We dene
as usual jni j = nij . For some k each n1 ;:::;np depends only on the k-jet of and
takes values in
Vi_

M : it is a section on J k (E=M ) of the tensor product of the
inverse images of the Vi_ and of
M . We write F for the increasing ltration by
P
jni j. It does not depend on the local coordinate systems used. The dierential
d maps FN to FN +1 .
Replacing in (2.215) each @ ni by its symbol in Symjni j (TM ), we obtain the
following description of GrF (
qloc;multi). It is the inductive limit in k of the space
k of the vector bundle
of sections over JE=M
(2.216)
k Sym (TM )
V i_

qM
with each factor to be replaced by its pull-back to JE=Mk . The degree N is the sum
of the symmetric algebra degrees.
The dierential d induces GrFN (d): GrFN (
qloc;multi) ! GrFN +1 (
qloc+1;multi), which
is linear over the functions. It is deduced from a morphism of vector bundles on M

p p
(2.217)
Sym (TM )
1

qM !
Sym (TM )
1

qM+1
by pull-back and tensorization by
V i_ . Locally on M , for fej g a basis of TM ,
(2.217) is
p
X !
X
(2.218) s
7 ! 1

(ej at ith place)

1 s
ej ^ :
j i=1
The crux of the matter is now the following
Lemma 2.219. On M , for each integer N , the complex of vector bundles with
components
p
(part of degree (N + q) of
Sym (TM ))

qM
1
and dierential the morphisms (2.218) of vector bundles is acyclic, except in top
degree.
Proof. This is to be checked point by point. One then recognizes in the complex
a variant of the Koszul complex, and one can write an explicit homotopy. In more
detail: for T the tangent space at a point of M , one has to consider a complex with
components

p
(2.220)
1 Sym (T )
^q (T _ ) :

p p p
One has
Sym (T ) = Sym T . If we identify T with its image T T by
1 1 1
the diagonal embedding, and if S is any supplement, we have

p
(2.221)
Sym T = Sym (S )
Sym (T ) ;
1
and the complex (2.220) becomes the tensor product of Sym (S ) (of cohomological
degree zero) by the complex whose component of degree q is
(2.222) Sym (T )
^q T _ ;
and whose dierential is
X
(2.223) d(s
) = (ej s)
(ej ^ )
j
for ej a basis of T and ej the dual basis of T _ .
The complex (2.222) is a version of the Koszul complex. It is multiplicative in
T : for T = T 0 T 00 , it is the tensor product of the similar complexes of T 0 and
T 00 . To check its acyclicity except in top degree, it hence suces to check it for
dim(T ) = 1, in which case the complex reduces to
(2.224) R[x]
x! R[x]
in degree 0 and 1. One can also write an explicit homotopy H : for P and
homogeneous, dene H (P) to be 0 if P is of degree 0 and of maximal degree
dim(T ). Otherwise
X
(2.225) H (P) := @ej (P )
(ej ) =(deg P + dim T deg )
One checks that dH + Hd is the identity minus the projection to the part of bidegree
0 in P and dim(T ) in .
Lemma 2.219 implies Theorem 2.214, as the acyclicity (except at the top) of
an associated graded to
loc;multi implies the same acyclicity for
loc;multi itself.
From the proof of Corollary 2.183 one can in addition deduce a local formula to
attach to such that d = 0 a for which = d . Indeed, let us cover E by
open sets W` for which one has local coordinates as in (a), (b), (c) before (2.215).
On each W` , a choice of local coordinates splits the ltration F , and the proof of
Lemma 2.219 gives a homotopy operator H` of ltration 1 such that outside of
top degree Id (dH` + H`d) is of ltration 1, i.e., maps FN to FN 1 . P Let ` be a
partition of unity subordinate to the covering W` and dene H () = H` (` ).
We again have Id (dH + Hd) of ltration 1. If is in FN , and closed,
(2.226) 0 := (dH + Hd) = d(H)
is in FN 1 . One has = dH + 0 , 0 is closed, and repeating the argument for
0 , one eventually obtains with = d .
CHAPTER 3
Free Field Theories
x 3.1. Coordinates on Minkowski spacetime
Let M denote ane n-dimensional Minkowski spacetime. Fix a coordinate sys-
tem x0 ; x1 ; : : : ; xn 1 with respect to which the metric is
(3.1) g = (dx0 )2 (dx1 )2 (dxn 1 )2 :
Note that x0 = ct for t the standard time coordinate. The wave operator on
functions is
= ( 1)n 1 d d
= d d
(3.2)
= @02 @12 @n2 1
= g @ @ ;
where we use the notation
(3.3) @ = @x@ :
The canonical density is
(3.4) jdn xj = j dx0 ^ dx1 ^ ^ dxn 1 j:
For a function f we have
(3.5) d df = f jdn xj:
The symbol `g' usually denotes the Minkowski metric, though occasionally it
denotes a general metric as is clear from the context.
Lagrangians in Minkowski spacetime are real (see Chapter 7). Thus equations
involving complex quantities have complex conjugate equations which also hold,
and which we usually omit.
We dene global charges by integrating over the spacelike submanifold fx0 = 0g
with canonical density
(3.6) jdn 1 xj = j dx1 ^ ^ dxn 1 j:
Indices ; ; : : : run from 0 to n 1; indices i; j; : : : run from 1 to n 1.
191
x3.2. Real scalar elds
A real scalar eld is a map : M ! R. The free lagrangian for a eld of mass m 0
is
2 o
L = 21 jdj2 m2 2 jdn xj
n
2
(3.7) = 12 d ^ d m2 2 jdn xj
2 o
= 12 g @ @ m2 2 jdn xj:
n
It is instructive to check the units in the lagrangian. Comparing the two terms we
see that m must have units L 1 (to match the units of d). In a relativistic quantum
theory we can replace m by mc=~; then m has units of mass. Of course, in that
context we usually work in a natural system of units with c = ~ = 1. In that case
has mass dimension (n 2)=2. The dierential of L along the space of elds is
L = d ^ d m2 jdn xj
(3.8)
= ^ d d + m2 jdn xj d ^ d :
So from (3.5) the classical eld equation is
(3.9) ( + m2 ) = 0
and the variational 1-form is
(3.10) = ^ d:
The local symplectic form ! = is
(3.11) ! = d ^ :
Equation (3.9) is most easily analyzed through the Fourier transform
Z p
^(k) = (21)n=2 e 1hk;xi (x) jdn xj;
(3.12) ZV p
1
(x) = (2)n=2 e+ 1hk;xi ^(k) jdn kj:
V
Then equation (3.9) transforms to

(3.13) jkj2 + m2 ^(k) = 0:
Thus the Fourier transform ^ is supported on the mass shell
(3.14) Om = fk 2 M : jkj2 = m2 g:
Note that since is real we have
(3.15) ^( k) = ^(k):
CHAPTER 3. FREE FIELD THEORIES 193
Write k = (k0 ; : : : ; kn 1 ); then ki has units L 1 and is called the wave number .
The frequency ! = k0 c has units T 1 .
We can formulate the theory for an arbitrary metric g on M as
2
(3.16) Lg = 21 g @ @ m2 2 g (x);
where g is the canonical density associated to the metric g. Now a computation
shows
(3.17) g = 21 g 1 g g = 21 g 1 g g ;
where
(3.18) g 1 g = g g :
So the energy-momentum tensor, as dened in (2.176), is
(3.19) Tg = d d g g Lg :
In local coordinates we write (see (2.177))
2
(3.20) T = @ @ + 12 jdj2 + m2 2 g :
Specialize to Minkowski spacetime. The energy density T00 jdn 1 xj is given by
nX1 2
(3.21) T00 = 2 (@0 ) + 2 (@i )2 + m2 2 :
1 2 1
i=1
Note that this is nonnegative and only vanishes for 0. Minus the momentum
density Ti0 jdn 1 xj is given by
(3.22) Ti0 = @i @0 ; i = 1; 2; : : : ; n 1:
By Proposition 2.190 this agrees with the energy-momentum tensor (2.160) given
by the canonical Noether currents of translations, as we now verify directly. The
vector eld induced by @ is
( )dx =
(3.23)
( ) = @ :
Then minus the associated Noether current is
( )(L + ) = L dx + @ d
(3.24)
= ( g L + @ @ ) dx ;
where g is the Lorentz metric.
The \trace" of T is
2
(3.25) g T = (1 n2 )g @ @ + m2 n 2 ;
which vanishes if n = 2 and m = 0. Thus the theory of a massless scalar eld is
conformally invariant in 2 dimensions.
x3.3. Complex scalar elds
Our convention is that if `h; i' denotes a real bilinear form, then it also denotes
the extension to a bilinear form over the complex numbers. If h; i is a real inner
product on a real vector space W , then the associated hermitian norm on the
complexication WC is
(3.26) w 7 ! hw;
wi:
Over the reals or complexes we always use the notation
(3.27) jwj2 = hw; wi:
A complex scalar eld is a map : M ! C . The free lagrangian for a eld of
mass m 0 is
n o
(3.28) L = hd; di m2 h; i jdn xj:
This theory is equivalent to a theory of two uncoupled free real scalars 1 ; 2 ; simply
set
+
p 1
(3.29) = 1 p 2:
2
Then (3.28) reduces to the sum of two copies of (3.7). The variation of (3.28) is
(3.30)
L = hd ^ di + hd ^ di m2 m2 jdn xj

= d d + m2 jdn xj d d + m2 jdn xj d
for the variational 1-form
(3.31) = ^ d + ^ d:
So the equation of motion is
(3.32) ( + m2 ) = 0:
The analysis of (3.32) proceeds as in the real case. Note the signicant dierence
that there is no reality condition (3.15) in the Fourier transform.
The circle group T C operates on by scalar multiplication, and this sym-
metry manifestly preserves the lagrangian. (In the language of x2 it is an \internal"
symmetry.) The corresponding innitesimal symmetry is
p
(3.33) ( ) = 1:
(Of course, we also have the conjugate equation ( ) =
p 1 .) So the
associated Noether current (see (2.88)) is

j = ( ) L +
(3.34) p
= 1 d d :
The corresponding global charge is
p Z
(3.35) Q = 1 (@0 @0 ) jdn 1 xj:
fx0 =0g
The energy-momentum tensor is
(3.36) Tg = d d g g Lg :
or (see (2.177))
(3.37) T = @ @ +1 hd; di + m2 h; i g :

2 2
We evaluate the energy-momentum tensor for a plane wave

(3.38) (x) = aeik(x) ;
where k = (k0 ; : : : ; kn 1 ) is in the dual of V , the vector space underlying M . We
restore the constants necessary to treat m as a mass, so replace m by mc=~. Compo-
nents of the wave number (k1 ; : : : ; kn 1 ) have unitspL 1 and the frequency ! = k0 c
has units T 1. The complex constant a has units M=TLn 4. By the equation of
motion (3.32), we have
(3.39) jkj2 = m2 c2 =~2:
Then (3.37) reduces to
(3.40) T = jaj2 k k :
x 3.4. Spinor elds
Let V be the vector space underlying Minkowski spacetime M , and let Spin(V )
denote the Lorentz group. Suppose S is a real spin representation of Spin(V ). This
means that there are symmetric pairings
: S
S ! V
(3.41) ~: S
S ! V
which satisfy a Cliord relation. Let fe g be a basis of V and ff ag a basis of S .
We use the dual bases feg; ffag for V ; S . Write
(fa ; fb ) = ab e ;
(3.42) ~ (f a ; f b ) = ~ ab e ;
then the Cliord relation is
(3.43) ~ ab bc + ~ ab bc = 2g ca :
Let S denote the odd vector space which is S with odd parity. A spinor eld
(spin 1/2 fermion) is a map : M ! S . With respect to the basis we write
(3.44) (x) = a (x)f a :
A dual spinor eld is a map : M ! S . The analysis of spinor elds below
adapts easily to dual spinor elds.
A mass pairing for spinor elds is a skew-symmetric pairing
V2
(3.45) M: S ! R:
Nonzero pairings need not exist.23 If there is a nonzero mass pairing M , then there
exists a normalized skew-symmetric pairing
(3.46) : S
S ! R;
the normalization condition is
(3.47) ~ ab = a0 b0 aa0 bb0 :
For simplicity assume that is irreducible. Then we can write
(3.48) M = m; m 2 R:
We write
(3.49) M = M ab a b ;
where the name of a bilinear form is written between the arguments. The kinetic
term in the free lagrangian is the Dirac form
(3.50) D= = ~ ( ; @ ) = ~ ab a @ b :
Equation (3.50) denes a symmetric (in the graded sense) bilinear form up to
an exact term, so it is exactly symmetric after integrating over M , assuming no
contribution at innity. The free lagrangian is
L = 21 D= 12 M jdn xj:
n o
(3.51)
The variation of L is
L = 21 D= + 21 D= M jdn xj:
n o
(3.52)
Now
D= = ~ ab a @ b
(3.53)
= ~ ab @ a b + ~ ab @ ( a b );
23 For example, for n = 2 take S to be a half-spinor representation. It is one-dimensional, so
V2 S = 0.
and so

(3.54) L = D= M jdn xj d
for the variational 1-form
= 21 ~ ( ; ) (@ )jdn xj
(3.55)
= 12 ( ~ ab a b ) (@ )jdn xj:
Now M determines a map S ! S and ~ a map S ! V
S , so that both
M and D= dene operators from spinor elds to dual spinor elds. We use the same
letters M; D= to denote these operators. Then the equation of motion is
(3.56) D= = M :
We analyze (3.56) via the Fourier transform (3.12):
p ~ ab ^
(3.57) 1 k b (k) = M ab ^b (k) for all a:
p
Multiply both sides of (3.57) by 1 ~ cak . After some simplication using (3.43),
(3.47), and (3.57) we nd
(3.58) jkj2 ^c(k) = m2 ^c (k) for all c:
Thus the Fourier transform of a solution is supported on the mass shell Om . Fur-
thermore, equation (3.57) denes a subbundle S 0 of the trivial bundle Om S ,
and ^ is a section of S 0 . The rank of S 0 is24 dim S=2. Again we have a reality
condition (3.15). For m > 0 the subgroup of Spin(V ) which stabilizes k 2 Om is
isomorphic to Spin(n 1); the ber of S 0 at k is a spin representation of Spin(n 1).
For m = 0 the stabilizer has reductive part isomorphic to Spin(n 2), and the ber
of S 0 at k is a spin representation of Spin(n 2).
In some cases the minimal real spin representation S admits an action of C
or H commuting with Spin(V ). Then if ~ is sesquilinear we have an action of T
or Sp1 which preserves the kinetic term D= . (The mass pairing (3.45) may break
this symmetry.) If there are multiple copies of the spin representation, then there
is a larger compact group of manifest symmetries which rotate the various copies.
Each such symmetry has a Noether current which may be computed from (3.55).
As for the energy-momentum tensor we rst compute minus the canonical
Noether current associated to translations, as in (2.160). Translation @ induces a
vector eld with
( )dx =
(3.59)
( ) = @ :
Using (3.51) and (3.55) we compute ( )(L + ) = dx , where
(3.60) = 12 ~ ab a @ b g ( 21 D= 12 M )g :
This is not symmetric in ; (as expected).
24 There is an exceptional case: For a chiral spinor eld in 2 dimensions equation (3.57) says that
the Fourier transform is supported on half of the forward lightcone. In other words, a solution to
the Dirac equation is either left- or right-moving, depending on the chirality. See [I-Supersolu-
tions, x2.6] for the precise formulas.
x3.5. Abelian gauge elds
We restrict ourselves to the massless case. In 3 dimensions the Chern-Simons
functional serves as a mass term (see Problem FP4 of [I-Homework]). In any
dimension the Higgs mechanism can be used to introduce a mass. For another
discussion of this material, see the solution to Problem FP3 of [I-Homework].
The standard physical interpretation (Maxwell's equations) is discussed in x4.1.
The basics of principal bundles and connections are reviewed in x4.2.
We choose as gauge group p the circle group T of unit norm complex numbers.
On the Lie algebra Lie(T) = 1 R we have the positive denite inner product
p
(3.61) ha; bi = ab; a; b 2 1 R:
The eld in a gauge theory is a connection A on M with structure group T, and
we work with elds up to gauge equivalence. The space of equivalence classes is a
real ane space. The lagrangian is
L = 21 hFA ^ FA i
(3.62) = 12 jFA j2 jdn xj
= 14 hF ; F0 0 i g g jdn xj;
0 0
where the curvature is

(3.63) FA = 21 F dx ^ dx :
Note that since
(3.64) (dx ^ dx ) = g0 g 0 (@ 0 )(@0 )jdn xj;
we have
(3.65) FA = 12 F g0 g 0 (@ 0 )(@0 )jdn xj:
Now A is an imaginary 1-form on FA M , and
(3.66) FA = dA;
the sign comes from commuting past d. Thus the variation of the lagrangian is
L = hdA ^ FA i
(3.67)
= dhA ^ FA i hA ^ d FA i:
Hence the canonical 1-form is
= hA ^ FAi
(3.68)
= hA ; F ig g (@ )jdn xj
and the classical equation of motion is
(3.69) d FA = 0;
or
(3.70) g @ F = 0 for all :
This is an ane equation for A, and since gauge transformations act by an ane
action, the moduli space M of classical solutions modulo gauge equivalence is an
ane subspace of the space of elds modulo equivalence. The local symplectic form
is
! = hA ^ dAi
(3.71)
= h(@ A @ A ) ^ A i g g (@ )jdn xj:
Of course, the detailed analysis involves choosing an origin, so we may as well
from the beginning work with the vector space of translations associated to con-
nections. Hence in this paragraph only we take the eld to be a real 1-form up
to addition of an exact real function. The lagrangian is
(3.72) L = 12 d ^ d = 12 jdj2 jdn xj;
and the equation of motion is
(3.73) d d = 0:
As usual, we consider the Fourier transform ^ = ^ dx on V . The vector space
of classical solutions is the rst cohomology space of the complex
(3.74) d
1 (M ) d!

0 (M ) ! d
1 (M ):
On the Fourier transforms we have

p
df^(k) = 1 f^(k)k
(3.75)
d d^(k) = jkj2 ^(k) h^ (k); kik:
An easy argument shows that the rst cohomology is isomorphic to the collec-
tion of Real functions f^ : lightcone ! V
C g on the lightcone which satisfy
h^ (k); ki = 0 modulo the set of functions f^(k) = f^(k)kg, where f^ ranges over the
complex functions on the lightcone. (The Reality condition is that of the Fourier
transform.) The subgroup of Spin(V ) which stabilizes k has reductive part isomor-
phic to Spin(n 2), and ^ is a section of a bundle whose ber at k is the vector
representation of Spin(n 2).
Return now to the abelian gauge eld and lagrangian (3.62) To compute the
energy-momentum tensor, we rst compute the Noether current j of minus trans-
lation by @ (see (2.160)). Note that @ induces the vector eld ^ on the space FA
of elds dened by (see (4.40))
(3.76) ^ A = (@ )FA :
Let = @ + ^ . Then
j = (@ + ^ )(L + )
(3.77)
= 12 (@ )FA ^ FA FA ^ (@ ) FA :

We nd after some computation that
(3.78) = hF ; F i g + 21 jFA j2 g ;

where is dened in (2.160). Notice that it is symmetric.
We can compute the energy-momentum tensor more directly using the alter-
native denition (2.176). Couple the lagrangian (3.62) to an arbitrary metric g:
(3.79) Lg = 21 hF ; F i g g g (x):

Using (3.17) and dierentiating with respect to g 1 we nd the energy-momentum
tensor to be (see (2.177))
(3.80) T = hF ; F ig + 21 jFA j2 g ;

which agrees with (3.78), as it must by Proposition 2.190. In (4.8) we rewrite (3.80)
in terms of the electric and magnetic components.
For a 1-form eld the energy-momentum tensor computed by dierentiating
the metric is again (3.80) (where we interpret F = @ @ ). However, the
canonical Noether current j diers from (3.77) since the induced vector eld ^
acts with an additional term over (3.76):
(3.81) ^ = Lie(@ ) = (@ )d d(@ ):
Then j picks up an extra term

(3.82) d(@ ) ^ d = d (@ ) d + (@ ) d d:
This is written as the sum of an exact term and a term which vanishes on-shell.
The \trace" of T ,
(3.83) g T = ( n2 2)jFA j2 ;
vanishes when n = 4. This says that the lagrangian (3.62) is conformally invariant
in n = 4 dimensions.
CHAPTER 4
Gauge Theory
x 4.1. Classical electromagnetism
As in x3.5 we consider an abelian gauge eld on Minkowski space M . To make
contact with the usual formulas of electromagnetism we take the gauge group to be
the multiplicative group R>0 rather than the circle group T. (In classical physics
the electric charge is not quantized, so this makes more physical sense.) Thus the
gauge eld A is a connection on a principal R>0 -bundle. The curvature FA is a real
2-form. The formulas of x3.5 hold, only now (3.61) is replaced by
(4.1) ha; bi = ab; a; b 2 R:
We work in any dimension n, though of course the electromagnetism of our
world is n = 4. Choose a splitting M = M 1 N of Minkowski spacetime into
time space. For now N can be any Riemannian manifold. Let x0 = ct be the
standard coordinate on M 1 . Dene the electric and magnetic elds
EA 2
1 (N )
(4.2)
BA 2
2 (N )
by
(4.3) FA = BA dt ^ EA :
EA and BA depend on time t, but neither contains dt:
(4.4) (@0 )EA = (@0 )BA = 0:
Let N be the Hodge star operator for the positive denite metric on N . Then if
is the Hodge operator on M 1 N (with its usual metric of signature + ),
we have
(4.5) FA = 1c N EA + c dt ^ NBA :
So the lagrangian (3.62) is
L = 21 jEcA2 j jBA j2 jdn xj;

2
(4.6)
201
where the norms of the forms EA ; BA are computed in the positive denite met-
ric on N . A multiplicative constant is usually inserted in L, depending on the
units used; we omit it.25 The equation of motion (3.69) together with the Bianchi
identity dFA = 0 comprise Maxwell's law in empty space:
dBA = 0 dEA = @B @t
A
(4.7)
d N EA = 0 c2 d N BA = N @E
@t
A
The initial components of the energy-momentum tensor (3.80) reduce to
T00 = 21 jEcA2 j + jBA j2

2
(4.8) nX1
Ti0 = (BA )ij (EA )j ;
j =1
the classical expressions for the energy density and the Poynting vector, while
the space components Tij comprise the Maxwell stress tensor. (Recall that Ti0
integrates to the eld momentum.)
We can couple the electromagnetic eld to a current J 2
j 1j(M ). The
lagrangian density is
(4.9) L = 12 FA ^ FA + J ^ A:
The second term is not well-dened in general, since A is a connection, not a 1-form.
In fact, if J is a conserved current
(4.10) dJ = 0;
then we can write J = dK and so by Stokes' theorem
Z Z
(4.11) J Â= K ^ FA
M M
is well-dened, assuming suitable decay at innity. (We discuss such \topological
terms" in Chapter 6.) A typical conserved current is the Poincare dual of a closed
curve in M , in which case (4.11) is the holonomy of A around the curve. In
coordinates we write
(4.12) J = J (@ )jdn xj;
where J 0 = c is c times the charge density and minus the spatial components J i
comprise the current density j :
(4.13) J = c jdn 1 xj c dt ^ j:
25 In the \mks" system of units, the constant is written 0 c2 = 107 =4.
CHAPTER 4. GAUGE THEORY 203
We can treat J as an external eld, which is xed, or as a dynamical eld which
varies. For example, we can take J to be Poincare dual to the path of a dynamical
moving point charge of rest mass m0 and charge q. In that case we add the kinetic
term (1.22) for the point particle to the lagrangian. Then the elds are the particle
x : R1 ! M and the connection A on M . The classical action is
Z T2 Z T2 Z
(4.14) S=
dx
m0 c d jd j q
x A 1 F ^ F :
A A
T1
T1 M2
In these formulas we compute the action for the proper time interval [T1 ; T2 ] R.
The second term should be interpreted as a multiple of the logarithm of the parallel
transport of A along a piece of the path x; see x6.1. It is not a real number, but
rather lives in a real torsor (depending on the endpoints). Note that the path is
oriented.
The equation of motion for the lagrangian (4.9) is
(4.15) d FA = J:
Rewriting in terms of EA ; BA ; ; j we have
d N EA = c2 jdn 1 xj
(4.16)
c2 d N BA = N @E A
@t + c j;
2
which are two of Maxwell's equations. (The top two equations in (4.7) remain
unchanged.)
The action (4.14) is invariant under reparametrizations of the path x. We
compute the equation of motion ignoring boundary terms. The equation for A is a
special case of (4.16) where the current J is Poincare dual to q times the path x.
The variation of the rst term with respect to the path x|after integration by
parts|leads to a contribution to the equation of motion of (see (1.25))
(4.17) @ p m0 v ;
@t 1 v2 =c2
where v is the velocity measured in our xed system of coordinates. To vary the
second term with respect to x, we use the formula that the variation of the holonomy
is the integral of the curvature evaluated on the variation of the path. Thus the
contribution of this term to the equation of motion is the vector associated via the
metric to the 1-form

(4.18) q @=@t + v FA = qEA q(v)BA :
(We omit the dt component of the 1-form.) So the equation for x is
(4.19) @ p m0 v = q(E (v)B ) ;

@t 1 v2 =c2 A A
`' denoting the dual vector. In four dimensions this is the Lorentz force law.
x4.2. Principal bundles and connections
In this section we establish some notation and review standard notions. Let M be
an arbitrary (super)manifold. All vector elds are even.
Let P ! M be a principal bundle with structure group a Lie group G, often
called the gauge group . If M is a supermanifold, then so is P . We take G to act
on the right . Elements 2 g in the Lie algebra of G induce vertical vector elds ^
1 ; 2 ] = [^1 ; ^2 ].
on P , and [\
Suppose V is a space with a left G action. In many interesting cases V is a
linear space and the action is linear, but it need not be. There is an associated
bundle V P ! M whose sections are equivariant maps f : P ! V . So f satises
(4.20) f (pg) = g 1 f (p); p 2 P; g 2 G;
where `' denotes the action of G on V . Write g = exp(t ) and dierentiate with
respect to t to nd
(4.21) ^f = f; 2 g;
where now `' denotes the innitesimal action of g on V .
The bundle associated to the adjoint action of G on g is the adjoint bundle
gP = ad P = P G g. A section of the adjoint bundle is then a map : P ! g with
(pg) = Adg 1 (p). Using the innitesimal g action on P we can identify with a
G-invariant vertical vector eld ^ on P . Namely, dene ^p to be the vertical vector
corresponding to the innitesimal action by (p). Then by (4.21)
(4.22) ^1 (2 ) = [1 ; 2 ]
and the corresponding vertical vector eld is [^1 ; ^2 ]. The ber of ad P at m 2 M
is the Lie algebra of innitesimal automorphisms of the ber of P at m. It acts on
the left , which explains the minus in (4.22). Equation (4.22) is a special case of the
following: If f : P ! V is a section of an associated bundle, then
(4.23) ^f = f:
(Compare (4.21).)
There is also an adjoint bundle of groups P G G ! M . Sections of this bundle
act as automorphisms of P , often called gauge transformations .
A connection on P is a G-invariant distribution on P which projects isomor-
phically onto TM . Thus a vector eld on M has a horizontal lift ~ which is a
G-invariant vector eld on P . Equivalently, a connection is encoded in a g-valued
1-form A 2
1P (g) which satises
(^)A = ; 2 g;
(4.24)
Rg A = Adg 1 (A); g 2 G:
Here Rg : P ! P is the (right) action of g on P . The curvature FA 2
2P (g) is
(4.25) FA = dA + 21 [A ^ A]:
It follows from (4.24) that FA 2
2M (ad P ) is a 2-form on the base M with values
in the adjoint bundle. If 1 ; 2 are vector elds on M , then

(4.26) [~1 ; ~2 ] [^
1 ; 2 ]
is a G-invariant vertical vector eld on P , and by (4.25) it corresponds to the section

(4.27) (2 )(1 )FA = ([~1 ; ~2 ])A
of the adjoint bundle.
Let : P ! g be a section of the adjoint bundle and ^ the associated G-invariant
vertical vector eld. Then for any vector eld on M ,
(4.28) [~; ^] = ~b:
A connection A on P induces a covariant derivative r on any associated bun-
dle V P = P G V . For a vector eld on M , and a section f : P ! V of V P ,
(4.29) r f = ~f:
The covariant derivative may be viewed as an operator dA :
0 (V P ) !
1 (V P ),
dened by
(4.30) dA f = (d + A)f:
If G is a linear algebraic group, then we have the Tannakian statement: A connec-
tion on P is equivalent to a system of connections on all associated vector bundles
which is compatible with the tensor product of representations.
Automorphisms of P |gauge transformations|act on connections by pullback.
An automorphism ' : P ! P may be represented by an equivariant map g : P ! G,
dened by
(4.31) '(p) = p g(p):
Then if A 2
1P (g) is a connection, we have
(4.32) ' A = Adg 1 A + g 1 dg;
where g 1 dg is the pullback by g of the Maurer-Cartan form on G.
The Bianchi identity is simply the Jacobi identity for horizontal vector elds
on P :
(4.33) [[r1 ; r2 ]; r3 ] + [[r2 ; r3 ]; r1 ] + [[r3 ; r1 ]; r2 ] = 0:
Equivalently,
(4.34) dA FA = 0;
where dA :
2M (ad P ) !
3M (ad P ) is the extension of the dierential d using the
connection A.
The set of all connections on P is an ane space AP . One sees this easily
from (4.24), which are ane equations. The associated vector space of translations
is
1M (ad P ). The group GP of gauge transformations acts on AP on the left using
pullback by the inverse (or pushforward of the associated horizontal distribution).
Let Aut P denote the group of all dieomorphisms ' : P ! P which commute
with the right G action. Such a ' induces a dieomorphism of M , and there is an
exact sequence
(4.35) 1 ! GP ! Aut P ! Di M:
The last map may not be onto; the image is the subgroup of Di M which preserves
the topological type of P ! M . Also, the sequence (4.35) is not usually split.
Innitesimally|at the level of Lie algebras|we do have a short exact sequence
(4.36) 0 ! Lie(GP ) ! Lie(Aut P ) ! X (M ) ! 0:
Furthermore, a connection A gives a splitting as vector spaces (but not as Lie
algebras): to a vector eld 2 X (M ) we attach its horizontal lift ~.
It is often more geometric (and more physical) not to x a particular bundle P .
Then instead of a single ane space of connections, we study the category CM (G)
of all connections on all principal G-bundles over M . A morphism in CM (G) is an
isomorphism of principal bundles which preserves the given connections. The set of
equivalence classes CM (G) may be identied as a disjoint union of spaces AP =GP ,
where P runs over a set of representatives of topological types of G-bundles on M .
In a lagrangian eld theory including connections, the category CM (G) is part of
the \space" of elds.
Fix a bundle P . On the product bundle
(4.37) P = AP P ! AP M
there is a universal connection A. Its restriction to fAg P is A, and its restriction
to AP fpg is zero. It is straightforward to compute the curvature of A at (A; m):
evaluated on 1 ; 2 2 TmM and 1 ; 2 2
1M (ad P ) we obtain
(2 )(1 )FA = (2 )(1 )FA
(4.38) ()()FA = ()
(2 )(1 )FA = 0:
These are equations for elements of the ber (ad P )m .
The group Aut P acts on P (on the left) by the product of its actions on AP
and P . The universal connection A is invariant under this action. Lagrangians in
gauge theory are gauge invariant functions (partial densities) on AP M which are
computed from A, and the (Aut P )-invariance of A implies that such lagrangians
are invariant under subgroups of Di M .
From (4.35) we see that Di M acts on the space CM (G) of equivalence classes
of connections. Also, the discussion after (4.36) shows that there is a canonical lift
of X (M ) to vector elds on the universal bundle P. First, if A 2 AP is a connection
and a vector eld on M , then acts on A using the horizontal lift ~:
(4.39) Lie(~)A = (~ )dA = ()FA :
Since dieomorphisms of M act on elds by pullback, the induced vector eld ^
on AP is
(4.40) ^ = ()FA at A 2 AP :
Thus
(4.41) = (~ )FA +
is the vector eld on AP M induced by . Its horizontal lift ~ to P (using
the universal connection A) is the desired induced vector eld on P. For vector
elds 1 ; 2 on M , we compute

(4.42) [^1 ; ^2 ] = ^[1 ;2 ] dA FA (1 ; 2 )
(4.43) [~1 ; ~2 ] = [^
1 ; 2 ] FA (1 ; 2 )
Equation (4.42) asserts that the map 7! ^ is a homomorphism26 up to innitesi-

mal gauge transformations, as is evident from the sequence (4.36). Equation (4.43)
implies that if 1 and 2 commute, then the symmetry 2 commutes with 1 up to
a gauge transformation. This is used in superspace formulations of supersymmetric
gauge theories, where the symmetry generated by the vector elds Qa on super-
space commute up to a gauge transformation with the action of vector elds Da
used to build lagrangians.
x4.3. Pure Yang-Mills theory

The data which dene a pure gauge theory are:
G Lie group with Lie algebra g
(4.44) h; i bi-invariant inner product on g
We work on Minkowski n-space M . The eld in the theory is a connection A
on a principal G-bundle over M . As explained in x4.2, the collection of elds is
best regarded as a category. We write lagrangians which are invariant under gauge
transformations.
Recall (4.25) that for a nonabelian group the curvature FA is a nonlinear func-
tion of the connection A. So nonabelian gauge theories are not free|the lagrangian
is not quadratic in the eld A. For perturbation theory one introduces a coupling
constant (or several if G is a product of simpler groups). We simply absorb these
constants in the inner product h; i.
The pure Yang-Mills lagrangian is given by the same formulas as (3.62):
(4.45) L = 21 hFA ^ FA i:

26 When a Lie algebra acts on a manifold on the left we expect an anti homomorphism. But if
we view the space of vector elds as the Lie algebra of the group of dieomorphisms, then the
induced bracket is minus the usual Lie bracket of vector elds.
The manipulations following (3.62) are valid for arbitrary gauge groups, except that
we must replace d by dA in the nonabelian case. Thus the equation of motion is
(4.46) dA FA = 0;
and the variational 1-form and local symplectic form are
= hA ^ FA i;
(4.47)
! = hA ^ dA Ai:
Equation (4.46) is called the Yang-Mills equation . Of course, the connection A
always satises the Bianchi identity (4.34).
The discussion of the energy-momentum tensor carries over without change,
and formula (3.80) holds:
(4.48) T = hF ; F ig + 12 jF j2 g :

Pure Yang-Mills theory is conformally invariant in 4 dimensions (see (3.83)).
The energy density is
T00 = 12 jF0i j2 + 21
X X
(4.49) jFij j2 ;
i i<j
where i; j = 1; : : : ; n 1 run over spatial indices. Therefore, the eld congurations
of minimal energy are at, and the moduli space of vacua on Minkowski space is a
point|the equivalence class of the trivial connection.
Fundamental lagrangians in physical theories are constrained by renormaliz-
ability. With that criterion there are a few terms one can add to the Yang-Mills
lagrangian (4.45) in dimensions 2, 3, and 4. These terms are topological in nature.
We often refer to them as \-terms".
In n = 2 dimensions suppose
(4.50) hhii : g ! R
is a trace on g, i.e., a linear map for which hh [a; b] ii = 0 for all a; b 2 g. Then an
additional possible term in the lagrangian is
(4.51) L2 = 2 hhFA ii:

We include the constant 2 R since this is the form in which this -term is usually
written. There is a constant for each independent trace on g. On a compact
2-manifold M (or on the space of elds with nite action on Minkowski space) the
integral of (4.51) is locally constant on the space of elds. Thus L2 does not aect
the equations of motion. It does, however, enter into the formula for the variational
1-form. Namely, the variational 1-form for pure Yang-Mills plus L2 is
(4.52) = hA ^ FA i + 2 hhAii:

The local symplectic form and energy-momentum tensor are unchanged.
In n = 3 dimensions there is a Chern-Simons term; we discuss it in x6.2. It is
dierent than the -terms (4.51) and (4.54): the Chern-Simons term is not locally
constant on the space of elds, and so does have a dynamical eect.
In n = 4 dimensions there is a -term associated to an invariant bilinear form
(4.53) hhii : g
g ! R:
(The form need not be an inner product; in particular, it need not be identical
to (4.44).) The -term is
(4.54) L4 = 162 hhFA ^ FA ii:

As with L2 , the integral of L4 is locally constant on the space of elds. The modied
variational 1-form is
(4.55) = hA ^ FA i + 8 2 hhA ^ FA ii:
x 4.4. Electric and magnetic charge
To begin, let spacetime be a product M = M 1 N . For a classical electromag-
netic eld A, as in x4.1, the electric charge \enclosed" by a closed codimension 1
submanifold N is
Z Z
(4.56) E () = F = 1 E ;
A A c N A

and the magnetic charge enclosed by a closed 2-dimensional submanifold N is
Z Z
(4.57) BA() = FA = BA :

In each case need not be a boundary, hence the quotes around ènclosed'. We use
the same terminology for a general spacetime M , which may also be Euclidean|
then only the rst equations in (4.56) and (4.57) make sense. For a gauge eld with
general gauge group G, electric and magnetic charge are associated to a trace
(4.58) : g !R
on the Lie algebra g:
Z
EA () = (FA );
(4.59)
Z
BA () = (FA ):

Magnetic charge is an example of a topological charge; BA () is a topological in-
variant which depends only on the topological type of the principal bundle under-
lying A. Hence it is a locally constant function on the space of elds|in particular,
the space of classical solutions. It is central in the sense that it Poisson commutes
with any other charge.
Electric charge is not generally conserved in a nite region. For example, in
classical electromagnetism it follows from (4.16) that the time derivative of the
electric charge enclosed by is proportional to the integral of the spatial current j
through . Also, if = @
then the electric charge enclosed by is proportional
to the integral of the charge density over
. This is the usual statement of
Gauss' law. We generalize to eld theory, where the current J is computed from
matter elds, as follows. We work with a gauge eld A for a general gauge group G.
Elements in the center of g give innitesimal global gauge transformations which
act trivially on the space of connections: dA = 0 for all A. From x2.8 it follows that
if L0A is a gauge invariant \matter lagrangian" in which A is a xed background
eld and which only depends on the 1-jet of the matter elds , then the Noether
current j for in the theory L0A is computed by dierentiating L0A with respect
to A (see Proposition 2.149). We now consider A as a dynamical variable by adding
the Yang-Mills lagrangian to L0A . We write the signs for a Lorentz manifold.
Proposition 4.60. Let L0A() be the matter lagrangian described above, and set
(4.61) L(A; ) = 21 hFA ^ FA i + L0A ():
Suppose
is a spacelike (n 1)-dimensional region with smooth boundary . Then
if is the global gauge transformation corresponding to a central element 2 g,
on-shell we have Z
(4.62) EA () () = j ;

where () 2 g is the trace 7! h; i on g and j is the canonical Noether current
for in the theory L0A .
Remark 4.63. In Minkowski space we often take
to be a spacelike hyperplane|
the limit of large balls|in which case is the sphere at \spatial innity"|the limit
of large spheres. Then the limit of EA ()() is the total electric charge in the system.
Proof. There is a canonical variational 1-form since the lagrangian only depends
on the 1-jet of the elds. Let L = L + = LY M + L0A be the total lagrangian. The
equation of motion is
(4.64) (DLY M )1;j0j + (DL0A )1;j0j = 0:
Evaluate this on , where 2
1M is an arbitrary 1-form and is viewed as a
tangent to the space of connections. By manipulations as in x3.5 we have
(4.65) ()(DLY M )1;j0j = ^ h; dA FA i:
By Proposition 2.149 the second term is j (). Since is arbitrary,
(4.66) j = h; dA FA i = dh; FA i:
Equation (4.62) is the integral version of (4.66).
Formula (4.62) is not modied if we add a -term (4.51) or (4.54) to (4.61).
(This has the eect of adding a multiple of FA to FA in (4.65), but that extra
piece vanishes by Bianchi.) However, formula (4.62) is modied in the theory of
monopoles; see [II-Dynamics of QFT, x9.6].
CHAPTER 5
-Models and Coupled Gauge Theories
x 5.1. Nonlinear -models

We begin with some preliminary remarks. First, if M is a manifold and E ! M
a real or complex vector bundle with connection r, then we form the twisted de
Rham complex
(5.1)
0M (E ) dr=r!
1M (E ) d!
r
2 (E ) d!
M
r :
It is not a complex in general, but rather

(5.2) d2r = Rr ;
where Rr is the curvature of r. Next, if : M ! X is a map between manifolds,
then d 2
1M ( TX ). If now TX has a connection r, then
(5.3) dr d = Tr ;
where Tr is the torsion of the connection r. We will apply this to the Levi-Civita
connection of a Riemannian manifold; its torsion vanishes.
The data which dene a nonlinear -model are:
X Riemannian manifold
(5.4) V : X ! R potential energy function
The eld in the theory is a map : M ! X ; the space of elds is F = Map(M; X ).
Dieomorphisms of X act on F by composition, and dieomorphisms of M act
on F using composition by the inverse . In each case the subgroup of isometries
preserves the lagrangian (5.6) below. There is a canonical evaluation map
(5.5) e: F M ! X
which is invariant under the product action of Di M on F M . Innitesimal
symmetries act on derivatives of the eld via the Levi-Civita covariant deriva-
tive r. This introduces curvature terms into the bracket of innitesimal symmetries
in general. Since e is invariant under Di M , and since the lagrangian (5.6) below
211
can be written in terms of e, it is easy to check that isometries of M are manifest
symmetries (as are isometries of X ).
The -model lagrangian with potential is
L = 21 jdj2 V jdn xj
n
(5.6)
= 12 hd ^ di V jdn xj:
Note that jdj is computed using both the metric on M and the metric on X . We
carry out the familiar analysis:
L = hr d ^ di h; grad V i jdn xj
(5.7)
= d h ^ di h ^ dr di h; grad V i jdn xj:
Hence the equation of motion is
(5.8) r = grad V;
where r is the covariant extension of the wave operator (3.2). Despite this no-
tation one shouldn't lose sight of Newton's law, which is the special case n = 1 of
equation (5.8). The variational 1-form and local symplectic form are
= h ^ di;
(5.9)
! = hdr ^ i:
It is easy to couple this theory to an arbitrary metric g on M :
Lg = 12 g @ @ V g (x):

(5.10)
The computation leading to (3.19) is essentially unchanged, and so the energy-
momentum tensor is
(5.11) Tg = hd dig g L;
or more explicitly (see (2.177))
T = @ @ + 12 jdj2 + V g :

(5.12)
On Minkowski space the moduli space of vacua|eld congurations which minimize
the energy density T00 |is
(5.13) Mvac = V 1 (0)
assuming that 0 is the minimum value of the potential energy V .
If is a Killing vector eld (innitesimal isometry) on X , then the associated
Noether current is
(5.14) j = h; di:
There are also topological terms one can add to a pure -model, usually called
Wess-Zumino terms . We describe them in Chapter 6.
CHAPTER 5. -MODELS AND COUPLED GAUGE THEORIES 213
x5.2. Gauge theory with bosonic matter
The theory in this section is the most general bosonic theory without gravity27
(though we do not include all possible terms|e.g. topological terms|in the la-
grangian). We can also describe it as a gauged nonlinear -model, or as a gauge
theory with bosonic matter. The data which denes the theory are:
G Lie group with Lie algebra g
h; i bi-invariant scalar product on g
(5.15) X Riemannian manifold on which G acts by isometries
V:X !R potential function invariant under G
The elds are
A connection on some principal G-bundle P ! M
(5.16) section of the associated bundle P G X ! M
It is often convenient to view as an equivariant map : P ! X . The lagrangian
combines (4.45) and (5.6):
(5.17) L=
1 jF j2 + 1 jd j2 V jdn xj:
2 A 2 A
There is a new term
(5.18) hA ; dA i
in L from the coupling of A and , so a new term in the equations of motion. The
variational 1-form is
(5.19) = h ^ dA i hA ^ FA i:
The energy-momentum tensor is
(5.20) T = (@A ) (@A ) hF ; F i g + 12 jFA j2 12 jdA j2 + V g

The moduli space of vacua is

(5.21) Mvac = V 1 (0) G
assuming that 0 is the minimum value of the potential energy V .
In x2.8 we described a simple example of \gauging a symmetry".
27 By this we mean a theory of scalar and gauge elds only. There are also models with p-form elds
for p 2, for example. The lagrangian here covers most fundamental (vs. eective ) lagrangians
without gravity.
CHAPTER 6
Topological Terms
We have already introduced \-terms" in gauge theory in x4.3. These are
a sort of topological term related to primary topological invariants, in this case
characteristic classes of principal bundles. Corresponding terms occur in a -model
as well. Namely, if : M ! X is a eld in a -model dened on an oriented
spacetime M n , and ! 2
n (X ) is a closed dierential form, then we can insert a
term
(6.1) c !
into a lagrangian. (The minus sign indicates that (6.1) is a contribution to the
potential energy.) Here c 2 R is a constant; if ! has periods which are 2~ times
integers, then only c (mod Z) enters into the quantum theory. In this chapter we
consider a dierent type of topological term which is related to secondary topo-
logical invariants. For example, in a gauge theory the holonomy of a connection
is a secondary invariant associated to a rst Chern class; in three dimensions the
Chern-Simons invariant is associated to a four-dimensional characteristic class. In
-models we meet Wess-Zumino-Witten terms which are secondary invariants as-
sociated to cohomology classes in the target space X .
To t such terms into the general theory of Chapter 2, we need local dierential-
geometric objects which integrate to these secondary invariants. We indicate an
extension of ordinary calculus to include these objects brie y, though a systematic
development of foundations for this extension is lacking. In the rst few sections
we focus instead on examples.
For both -terms and topological terms we need to work over an oriented space-
time. The orientation allows us to pass from dierential forms to densities. Note
that neither -terms nor topological terms depend on a metric on spacetime.
x 6.1. Gauge theory

The simplest example of a topological term was already discussed at the end of x1.1
and in x4.1. Namely, consider a theory on a spacetime M n which includes a con-
nection A for the group R>0 . Suppose x : R ! M is a parametrized path which
may be xed or variable. Then we can introduce a term
(6.2) q x A
215
into the lagrangian of the theory, where q 2 R is a constant. Note that (6.2) is meant
to be integrated over R, not over spacetime M . It is unchanged by orientation-
preserving dieomorphisms of R; in other words, it only depends on the image of x
viewed as an oriented submanifold of M .
Now q x A is dierent from other terms in lagrangians we have seen so far.
A connection A on an R>0 -bundle P ! M is an element of
1 (P ), and x A a
1-form on the pullback x P . It is not a 1-form on the base R. To compute the
action we can x a trivialization of x P . Taking into account the dependence on
this trivialization, the action over an interval [T1; T2 ] R is a homomorphism
of R-torsors, the R-torsors being q times the bers of x P at T1 and T2. In the
p to the quantum theory it is not the action S , but rather the
lagrangian approach
exponential exp( 1 S=~) which enters. Then this term in the action is parallel
transport in the circle bundle associated to P via the homomorphism
R>0 !T
(6.3) p
x 7 ! ( 1)qx=~
For usual kinetic and potential terms, the exponentiated action is an element of T.
This parallel transport is instead a homomorphism of circle torsors.
The equation of motion (Lorentz force law) arising from (6.2) in classical elec-
tromagnetism is discussed at the end of x4.1.
In the quantum theory of electromagnetism the electric charge q is quantized
in suitable units. It makes sense from the beginning to regard the gauge eld A
as a connection in a T-bundle (as opposed to an R>0 -bundle). More generally, in
quantum Yang-Mills theories the gauge eld is a connection for a compact gauge
group G. Then it makesp sense to write a term x (A) in the lagrangian, where
the trace : g ! R is 1= 1 times the dierential of an abelian character G ! T.
This term in the exponentiated action is again interpreted as a homomorphism of
circle torsors.
In three-dimensional gauge theory there is a topological term due to Chern and
Simons. Let M 3 be a 3-dimensional oriented spacetime and consider a theory with
gauge eld A. If G is the gauge group, and h; i an invariant inner product on the
Lie algebra g, then the Chern-Simons term is
(6.4) LCS = hA ^ FA i 61 hA ^ [A ^ A]i:
If A is a connection on P , then this term is a 3-form on the total space of P .
Assume for convenience that M 3 = M 1 2 is the product of time M 1 and a
space 2 which is a closed oriented surface. Then once again the exponentiated
action on [T1 ; T2 ] is best interpreted as a homomorphism between T-torsors T1
and T2 dened from (6.4) at fT1 g and fT2g . (For this we need h; i to
lie in a distinguished lattice of inner products.) For trivializable bundles P the
torsors are trivialized by a choice of trivialization of P , and may be constructed by
such trivializations. We omit the details. (For nontrivializable bundles we need a
renement of the inner product h; i to an element of H 4 (BG; Z) in order to dene
the torsors and the exponentiated action.)
Locally we can break gauge invariance and x a trivialization of the bundle
which carries the connection A. Then (6.4) pulls down to a local 3-form on M . We
compute
(6.5) LCS + d = hA ^ 2FA i;
CHAPTER 6. TOPOLOGICAL TERMS 217
where
(6.6) = hA ^ Ai:
So for pure Chern-Simons theory the equation of motion is
(6.7) FA = 0:
For the particular spacetime M 3 = M 1 2, the space of solutions M up to gauge
equivalence is the moduli space of at connections on . The local symplectic form
(6.8) ! = hA ^ Ai
integrates on to (twice) the usual symplectic form on M.
The equations of motion (6.7) and the local symplectic form (6.8) are gauge
invariant, whereas the variational 1-form (6.6) is not. This suggests that a Noether
charge is no longer a well-dened real number. Indeed, if is a central element
of g, interpreted as a constant gauge transformation, then the associated Noether
current from (6.6) is formally
(6.9) dh ^ Ai:
The integral over a region
2 is minus the logarithm of the holonomy around @
of the bundle associated to the exponential of the trace h; i on g. This logarithm
is well-dened only up to integer shifts. (If the Chern-Simons term is added to
the lagrangian of a standard 3-dimensional theory, as in (4.61), then this term
contributes to the electric charge (4.62).)
x 6.2. Wess-Zumino-Witten terms
Consider a -model with eld : M ! X , where the spacetime M is an oriented n-
manifold. Under the simplifying hypothesis Hn 1 (X ) = Hn (X ) = 0, we construct
a term in an action associated to a closed form
2
n+1 (X ) with periods in 2~Z.
To remove the simplifying topological assumption we need a more precise form
of this data, as we explain in x6.3. For simplicity, consider a product spacetime
M = M 1 N n 1 , where we assume space N to be a compact oriented n-manifold.
We rst construct a T-torsor Tf = Tf (
) for each map f : N ! X as follows.
Since Hn 1 (X ) = 0 there exist extensions F : Y n ! X to oriented n-chains Y n
with @F = f . Each such F trivializes Tf . If F1 ; F2 are two such extensions, then
since Hn (X ) = 0 we can nd an oriented (n + 1)-chain H : Z n+1 ! X with @H =
F2 F1 . Then the isomorphism from the F1 -trivialization to the F2 -trivialization
is
h p 1Z i
(6.10) exp H
:
~ Z
Because of the assumption on the periods of
, this is independent of the choice
of H .
We can allow Hn 1 (X ) to be torsion in this construction if we specify a coho-
mology class c 2 H n (X ; 2~Z) such that the de Rham class of
is the image cR
of c in real cohomology. Then a multiple of f bounds, and we essentially use the
choice of c to divide by that multiple in the abelian group T.
For another perspective, consider the usual diagram
Map(N; X ) N e! X
?
(6.11) ?
y
Map(N; X )
Then e
is a 2-form on Map(
N; X ) with periods in 2 ~Z, and e c is a rene-
2
ment in H Map(N; X ); 2~Z . So there exist p principal T-bundles with connec-
tion over Map(N; X ) whose curvature is ( 1=~) e
and whose Chern class
is e c=2~. The construction above gives a particular such T-bundle; it can be
extended to produce a connection as well. p
Now if : [T1 ; T2] N ! X is a eld, the exponentiated action exp( 1 S=~)
of is naturally an element of the circle torsor
(6.12) T@ = TjfT2 gN Tjf1T1 gN :
Namely, if Fi is a trivialization of TjfTi gN , then F2 + F1 bounds an (n + 1)-
chain H : Z n+1 ! X , and we use formula (6.10) to compute the action in this
trivialization.
We remark that the gauge theory constructions of x6.1 may be understood|at
least heuristically|as special cases in which X = BG is the classifying space of the
gauge group G.
x 6.3. Smooth Deligne cohomology
Fix a smooth manifold X and an integer28 n. Let us consider on X the constant
sheaf Z, the constant sheaf R, and the following complex of sheaves, denoted F n (
)
(or simply F (
)): the subcomplex of the de Rham complex given by

0; if p < n;
(6.13) F (
)p =

p ; if p n:
Both the cohomology of X with coecients in Z and the hypercohomology of X
with coecients in F (
) map to the cohomology of X with coecients in R. The
rst because Z maps to R. The second is just the cohomology of the complex
n (X ) !
n+1 (X ) ! starting in degree n, which is a subcomplex of the
de Rham complex computing the real cohomology. Suppose we have somehow
lifted those maps at the cochain level. In other words, suppose we have found
natural complexes (CZ ; dZ), (CF n ; dF n ) (or simply
(CF ; dF )), and (CR ; dR) with
H (CZ) = H (X; Z), H (CF ) = H X; F (
) , and H (CR ) = H (X; R), and

found morphisms of complexes
(6.14) 'Z; 'F : CZ ; CF n ! CR
28 Our choice of indexing makes the description of products below more natural. However, there is
a shift in the application to lagrangian eld theory: it is cocycles for the cohomology group Dn+1
which enter lagrangians for n-dimensional spacetimes.
CHAPTER 6. TOPOLOGICAL TERMS 219
inducing the maps we described in cohomology. This can be done in many ways|
one way will be described later|but the philosophy of cohomological algebra tells
that they are essentially equivalent.
One can then form a mapping cone (K; d) with
K p := CZp CFp CRp 1
(6.15)
d := dZ + dF dR + 'Z 'F :
A p-cycle c = (cZ; cF ; cR) of K is the data of p-cycles cZ and cF of CZ and CF ,
and of a homology between their images in CR. The mapping cone K behaves as if
one had a short exact sequence of complexes 0 ! K ! CZ CF ! CR ! 0. For
instance, one has a long exact sequence of cohomology groups. The smooth Deligne
cohomology group Dp;n is H p (K ). For p < n it is H p 1 (X; R=Z). For p > n it is
H p (X; Z). We will be mainly interested in Dn := Dn;n . It sits in an exact sequence
(6.16)
H n 1 (X; Z) ! H n 1 (X; R) ! Dn ! H n (X; Z) (closed n-forms) ! H n (X; R)
and it is an extension of the group of closed n-forms with integral periods by
H n 1 (X; R=Z).
This description of Dn is not the most economical, but it suggests the func-
torial properties to be expected. Products : One has product maps on Z and R,
as well as product maps F n
F m ! F n+m . If the corresponding cup-product in
cohomology is expressed at the cochain level, giving products in CZ, CR, and CF
compatible with 'Z and 'R, one obtains products Dp;n
Dq;m ! Dp+q;n+m . In-
deed, homologies c and c0 between the images of cZ and cF (resp. c0Z and c0F ) give
a homology between the images of cZc0Z and cF c0F . In fact, there are two naturally
cohomologous homologies: '(cZ)c0 + c'(c0F ) and c'(c0Z) + '(cF )c0 . Integration : If
a proper submersion f : X ! Y of relative dimension d has oriented bers, inte-
gration along the bers f has meaning in integral and real cohomology, as well as
from F nRon X to F n d on Y . Expressed compatibly at the cochain level, it should
provide X=Y : Dp;n (X ) ! Dp d;n d (Y ), and in particular Dn (X ) ! Dn d (Y ).
One way to nd complexes CZ , CF , and CR is to use the C ech method for
computing cohomology. If fUi gi2I is an open covering of X such that the nonempty
intersections Ui0 :::ip := Ui0 \ \ Uip are contractible, one can use for CZ the C ech
complex
Y
(6.17) CZp = (Ui0 :::ip ; Z);
and for CF and CR the simple complex associated to the double complex
Y
CFp;q = Ui0 :::ip ; F (
)q
(6.18) Y
CRp;q = Ui0 :::ip ;
q :
For n > 0 the map Z F !

of complexes of sheaves on X is injective. If
so is 'Z 'F : CZ CF ! CR , the mapping cone K has the same cohomology, up
to a shift of index by 1, as the cokernel of 'Z 'F , a complex whose cohomology
is the hypercohomology of the complex of sheaves
(6.19) C 1 (values in R=Z) !
1 ! : : : !
n 1 ! 0:
If we denote by
~ this complex, we have in the C echist computation
Y
(6.20) coker('Z 'F ) = (Ui0 :::ip ;
~ p );
and Dp;n is the cohomology group H p 1 of the complex.
For n = 1 we nd that D1 is the group of C 1 maps from X to the circle R=Z.
For n = 2 elements of D2 are represented by cocycles
i ! 0
x
?
(6.21) ?
gij !0
where the diagram indicates the equations
gij + gjk = gik
(6.22) i j = dgij :
p
The exp(2 1 gij )p are the transition functions for a T-bundle, trivialized on
the Ui , and the 2 1 i , viewed as local connection 1-forms, provide this T-
bundle with a connection. The group D2 is the group of isomorphism classes of
T-bundles with connection. More precisely, a cocycle denes a bundle with connec-
tion and a homology denes an isomorphism between bundles.
In x6.2, what we needed to avoid assumptions on the homology of X is not just
a closed (n + 1)-form with integral periods, but rather a class in Dn+1 , or rather a
cocycle giving such a class. If H n (X; R=Z) = 0 (equivalently: Hn (X; Z) = 0), one
has
(6.23) Dn+1 ! (closed (n + 1)-forms with integral periods):
If H n 1 (X; R=Z) = 0 (equivalently: Hn 1 (X; Z) = 0), the ambiguity in the choice
of a cocycle becomes irrelevant: if c1 and c2 are two cocycles representing the same
class in Dn+1 , not only are they homologous, c1 c2 = dc, but any two choices c0
and c00 for c are homologous: c0 c00 = dc~.
If a cocycle c is chosen and if is a eld, i.e., a map from M to X , then
c is a cocycle giving a class in Dn+1 (M ). For N in M a compact subvariety of
dimension n 1, integration of c on N should produce a T-principal homogeneous
space|a one-dimensional complex vector space L(N; c) with metric attached
to N and c. Indeed, integration on N maps Dn+1 (M ) to D2 of a point. If
N is the boundary of a singular chain S , integration of c on S should produce
a unit vector in L(N; c). For instance, in a Hamiltonian picture, a space-like
hypersurface N , supposed here compact, would give L(N; c) and the slice S
between two such hypersurfaces N1 and N2 would provide an isomorphism (the
action integral) from L(N1 ; c) to L(N2 ; c).
A systematic treatment of those expectations has yet to be given.
CHAPTER 7
Wick Rotation: From Minkowski Space to Euclidean Space
A basic constraint on a Minkowski space action is that it be real. An action
SM is the integral of a lagrangian densityZ LM over Minkowski space M :
(7.1) SM = LM :
M
Choose a time t on M . Then we (Wick) rotate to Euclidean space E by introducing
imaginary time
p
(7.2) = 1 t:
p
By convention the Euclidean action is 1= 1 times the rotated Minkowski action:
Z
(7.3) 1
p 1 SM = SE = LE :
E
p
Note that e 1 SM = e SE . Also, SE is not real in general.
We describe the continuation to Euclidean space more precisely for a -model.
The eld is a map : M ! X into some Riemannian manifold. The complexi-
cation of the space of maps M ! X is the space of holomorphic maps MC ! XC
between the complexied spaces. The lagrangian extends to a holomorphic func-
tion on this space, and the Euclidean action is the restriction of this continuation
to maps E ! X . (Note that EC = MC so E MC .) There is a similar picture
for other types of elds.
We consider four types of terms which typically occur in an action: kinetic
terms for bosons, potential terms, topological terms (also -terms), and kinetic
terms for fermions.
In this chapter set the speed of light c = 1. We use the conventions in x3.1 for
Minkowski space M . So the metric is
(7.4) gM = dt2 (dx1 )2 (dxn 1 )2 :
On Euclidean space E we use the positive denite metric
(7.5) gE = d 2 + (dx1 )2 + + (dxn 1 )2 :
So as to avoid confusion, we x the standard orientations t; x1 ; : : : ; xn 1 on M
and ; x1 ; : : : ; xn 1 on E , and we write lagrangians as forms rather than densities,
though we often omit the `^' sign. Let
(7.6) dn 1 x = dx1 ^ ^ dxn 1 :
221
x7.1. Kinetic terms for bosons
Consider a particle of mass m moving in some Riemannian manifold X . It is
described by a map x : M 1 ! X . Then the kinetic energy density is (see (1.13))

2
(7.7) LM = m2 dx
dt dt:

The continuation to imaginary time|after dividing by

p 1|is

2
(7.8) LE = m2 dx
d d:

In higher dimensions, we consider a real scalar eld on Minkowski space, which is

a real function : M ! R. The kinetic lagrangian is
(7.9) LM = 12 jdj2M dt dn 1 x;
where j jM is the norm (7.4) on M . The continuation to E is
(7.10) LE = 12 jdj2E d dn 1 x;
where j jE is the Euclidean norm (7.5).
For a gauge eld29 A the kinetic term is the Yang-Mills lagrangian
(7.11) LM = 21 jFA j2M dt dn 1 x:

The continuation to Euclidean space is
(7.12) LE = 21 jFA j2E d dn 1 x:

x 7.2. Potential terms
For the particle x : R ! X , the potential energy is described by a function V :
X ! R. The corresponding term in the lagrangian is
(7.13) LM = V (x(t)) dt:
The continuation to imaginary time is
(7.14) LE = V (x( )) d:
The extension to higher dimensions is the same: Potential terms appear with a
sign in Minkowski actions and with a + sign in Euclidean actions.
29 For a concrete example of rotating a gauge eld to Euclidean space, see the solution to Prob-
lem 11 of [II-Dynamics of QFT, Exercises].
CHAPTER 7. WICK ROTATION TO EUCLIDEAN SPACE 223
x 7.3. Topological terms and -terms
Let be a real 1-form on X and consider the lagrangian (for x : R ! X )
(7.15) LM = x :
The continuation
p to imaginary time is innocuous except for the conventional divi-
sion by 1:
p
(7.16) LE = 1 x :
Hence in the Euclidean (imaginary time) lagrangian the topological term is imag-
inary. A similar factor is introduced into the continuation of any topological term
(in the sense of Chapter 6).
The -terms in gauge theory (x4.3) behave similarly. For example, the contin-
uation of
(7.17) LM = 2 hhFA ii
to Euclidean space is
p
(7.18) LE = 1 2 hhFA ii:
x 7.4. Kinetic terms for fermions
We begin with the particle x : R ! X and add an odd eld which is a section of
x TX , the parity-reversed pullback of the tangent bundle. Then is real and in
real time its kinetic term in the lagrangian is
(7.19) LM = m2 h ; ddt i dt:
Rotating to imaginary time and dividing by
p 1, we obtain
p m d
(7.20) LE = 1 2 h ; d i d:
Consider now a real spinor eld in n-dimensional Minkowski spacetime. We
retain the notation of x3.4, but we set the mass M = 0. (The mass term is a
potential energy term, so is covered by the discussion of x7.2.) The lagrangian in
Minkowski space is
LM = 12 D=M dt dn 1 x
(7.21)
= 21 ( ~ M )ab a @ b dt dn 1 x:
The rst observation when rotating to Euclidean space is that a real spin represen-
tation S of Spin(1; n 1) extends to a complex representation SC of the complex
spin group SpinC (n), but the restriction to Spin(n) SpinC (n) is not necessarily
real. Thus in the Euclidean theory we take the eld
(7.22) : E ! SC
to be a complex spinor eld. Again: There are no reality conditions on spinor elds
in Euclidean space . Dene pairings
E : SC
SC ! VC
(7.23) ~ E : SC
SC ! VC
by
p p ~ 0ab
( E )0ab = 1 ( M )0ab ; ( ~ E )0ab = 1( M) ;
(7.24) ~ ~
( E )iab = ( M )iab ; ( E ) = ( M )iab ;
iab
where i = 1; : : : ; n 1 runs over the spatial indices. Then E ; ~ E satisfy a Cliord
relation
(7.25) ( ~ E )ab ( E )bc + ( ~ E )ab ( E )bc = 2gE ca :
The Euclidean Dirac form is
(7.26) D=E = ( ~ E )ab a @ b ;
and the factors in (7.24) are chosen so that the lagrangian (7.21) rotates to
(7.27) LE = 21 D=E d dn 1 x
in Euclidean space.
REFERENCES
There are many mathematical accounts of classical mechanics, for example

V. I. Arnold, Mathematical methods of classical mechanics, Translated from
the 1974 Russian original by K. Vogtmann and A. Weinstein, Springer, New
York, 1989.
Many elements of the formalism we adopt here for classical lagrangian eld theory
may be found in
G. J. Zuckerman, Action principles and global geometry, Mathematical As-
pects of String Theory, ed. S. T. Yau, World Scientic Publishing, 1987,
pp. 259{284.
The cohomology of local forms was investigated in
F. Takens, A global version of the inverse problem of the calculus of varia-
tions, J. Di. Geom. 14 (1979), 543{562.
The original reference for Noether's theorem is
E. Noether, Invariante Variationsprobleme, Nachr. Konig. Gesell. Gottingen,
Math.-Phys. Kl. (1918), 235{257 ; English translation in: Transport
Theory and Stat. Phys. 1 (1971), 186{207.
225

Deligne P., Freed D. Classical Field Theory (AMS 1999) (90s) .Ps

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Deligne P., Freed D. Classical Field Theory (AMS 1999) (90s) .Ps

Transféré par

Droits d'auteur :

Formats disponibles

Classical Field Theory

Pierre Deligne and Daniel S. Freed

The boundary term for H1 is

(1.37) jk (xi x_ k xk x_ i ) = m(v ) (xi x_ k xk x_ i ):

(1.41) dt+ dt (dx1 )2 (dxn 1 )2 :

7 Here the speed of light is inessential, so we set it to one.

is unchanged, and there is an isomorphism of the R-torsors for L and L + dK

x 2.5. Hamiltonian theory

on M to be nondegenerate. Indeed, the X are tangent to M, because X is a

using ftg N , then

holds on shell. The Noether current j^ := (^) ^ is conserved,

Let (ji ; i ); i = 1; 2 be Noether pairs. We de ne the bracket

x 2.7. More on symmetries

L = 12 jx_ j2 21 jxj2 jdtj:

From this we derive

A simple example from mechanics illustrates the importance of locality in Def-

By integration by parts, i.e., using that

(2.163) [ ] = LnM2 T = ML T Ln 1 :

Physically correct units for the components involving x0 = ct are obtained by

(3.37) T = @  @  +1 hd; di + m2 h; i g :

We evaluate the energy-momentum tensor for a plane wave

where the curvature is

On the Fourier transforms we have

We nd after some computation that

(3.78)  = hF ; F i g + 21 jFA j2 g ;

(3.79) Lg = 21 hF ; F i g g g (x):

(3.80) T = hF ; F ig + 21 jFA j2 g ;

L = 21 jEcA2 j jBA j2 jdn xj;

The initial components of the energy-momentum tensor (3.80) reduce to

T00 = 21 jEcA2 j + jBA j2

(4.19) @  p m0 v  = q(E (v)B ) ;

is a G-invariant vertical vector eld on P , and by (4.25) it corresponds to the section

Equation (4.42) asserts that the map  7! ^ is a homomorphism26 up to in nitesi-

x4.3. Pure Yang-Mills theory

(4.45) L = 21 hFA ^ FA i:

(4.48) T = hF ; F ig + 12 jF j2 g :

(4.51) L2 = 2 hhFA ii:

(4.52) = hA ^ FA i + 2 hhAii:

(4.54) L4 = 162 hhFA ^ FA ii:

x 5.1. Nonlinear -models

It is not a complex in general, but rather

The moduli space of vacua is

x 6.1. Gauge theory

For n > 0 the map Z  F !

The continuation to imaginary time|after dividing by

In higher dimensions, we consider a real scalar eld on Minkowski space, which is

(7.11) LM = 21 jFA j2M dt dn 1 x:

(7.12) LE = 21 jFA j2E d dn 1 x:

There are many mathematical accounts of classical mechanics, for example

Vous aimerez peut-être aussi

(1.37) jk (xi x_ k xk x_ i ) = m(v ) (xi x_ k xk x_ i ):

on M to be nondegenerate. Indeed, the X are tangent to M, because X is a

holds on shell. The Noether current j^ := (^) ^ is conserved,

Let (ji ; i ); i = 1; 2 be Noether pairs. We dene the bracket

(2.163) [ ] = LnM2 T = ML T Ln 1 :

(3.37) T = @ @ +1 hd; di + m2 h; i g :

(3.78) = hF ; F i g + 21 jFA j2 g ;

(3.79) Lg = 21 hF ; F i g g g (x):

(3.80) T = hF ; F ig + 21 jFA j2 g ;

(4.19) @ p m0 v = q(E (v)B ) ;

Equation (4.42) asserts that the map 7! ^ is a homomorphism26 up to innitesi-

(4.48) T = hF ; F ig + 12 jF j2 g :

(4.51) L2 = 2 hhFA ii:

(4.52) = hA ^ FA i + 2 hhAii:

(4.54) L4 = 162 hhFA ^ FA ii:

x 5.1. Nonlinear -models

For n > 0 the map Z F !

(7.12) LE = 21 jFA j2E d dn 1 x: