Lecture 5:

Quantization of Yang-Mills elds

Ludwig Faddeev
12 November 1996

This lecture treats Yang-Mills theory, an analogue of the eld theory for electromag-
netism (Example 2 of Lecture 4) in which the gauge group C 1(R4; U (1)) is replaced by
G = C 1(R4; G) where G is a nonabelian compact Lie group. In this case Feynman discov-
ered the disturbing fact that if one attempts to treat this theory following the approaches
which worked in the case of the electromagnetic eld, one obtains an S-matrix which is not
unitary. Here we shall see that the reason for this is a dierence of the geometry of the
gauge orbits, which are linear in the electromagnetic case but nonlinear for the case of the
nonabelian gauge eld.
We begin much as in our treatment of electromagnetism in Lecture 4. As in electromag-
netism, we have a eld A(~x; t) which is a vector representation of the Lorentz group (in
other words, it is a one-form on V = R4); however each component A0; A1; A2; A3 now takes
values in the Lie algebra Lie(G), so the one-form A = Adx takes values in Lie(G). (More
generally it could be taken to have values in an arbitrary unitary representation of G, but for
simplicity we take the representation to be Lie(G) endowed with the adjoint representation.)
An element g of the gauge group G = C 1(R4; G) acts on A by the transformation
A 7! Ag def
= gAg 1 + (@g)g 1
(1)
The curvature is given by
F = dA + A ^ A; (2)
where if A = Pa Xa
a for a basis Xa of Lie(G) and a collection of one-forms a, we dene
X
A ^ A = [Xa; Xb ]a ^ b:
a;b

The curvature (which is a Lie(G)-valued two-form on R4) transforms under the gauge trans-
formation g as
F 7! F g = gFg 1: (3)
As in the abelian case, the Lagrangian is
L = 1 Tr (F 2);
2 (4)

1
where if F = Pa Xa
a for a collection of two-forms a, we dene
X
Tr (F 2) = < Xa ; Xb > a ^ b: (5)
a;b
(A symmetric quadratic form on the space of two-forms on R4 is dened in the same way, and
is also denoted by Tr .) Here, we have introduced an Ad-invariant inner product <
;
> on
Lie(G), which in the case of matrix groups is given by the trace Tr . We have also introduced
a coupling constant .
This Lagrangian contains terms of the form (dA)2 as well as terms cubic in A (of order
1 in derivatives) and terms quartic in A (with no derivatives). There are thus two types
of vertices, a 3-valent vertex (corresponding to the term cubic in A) and a 4-valent vertex
(corresponding to the term quartic in A). By studying the corresponding perturbation
expansion, Feynman found that the S-matrix was not unitary.
To proceed further, we need to nd the physical variables. We write the Lagrangian
in a form that is rst order in derivatives, by the device introduced in Lecture 4 where we
regarded A and F as independent variables. Thus we see that the Lagrangian is (suppressing
the overall multiplicative constant 1=2 for the moment)
L = Tr (dA + A2; F ) 21 Tr F 2: (6)
As in Lecture 4, we have ten variables
A0; Ak ; F0k = Ek ; Fik = iklHl :
Exactly as in Lecture 4, the variables Ak and Ek are canonically conjugate, while the Fik (or
equivalently the Hl) are excludable variables and A0 is a Lagrange multiplier multiplying a
constraint
G = rk E k = @k E k + [Ak ; Ek ]: (7)
As in Lecture 4 we obtain an equation of motion (Gauss's Law)
G = 0;
the dierence from the case of abelian G (electromagnetism) is the presence in (7) of the
quadratic term [Ak; Ek ].
The canonical commutation relations are
n o
Aak(~x); Ebk (~y) = labl (3)(~x ~y): (8)
Here, a and b index a basis fXag of the Lie algebra Lie(G) and we have decomposed Ak =
P
a Ak
Xa for functions Ak which form a one-form A on R .
a a a 3

We introduce a source term G() = R Tr G(~x)(~x)d~x (where  : R3 ! Lie(G) is an

element of the Lie algebra of the gauge group). We nd that the Poisson brackets of the
source terms are
fG(1); G(2)g = G([1; 2]); (9)
2
in other words the map  7! G() (which may be regarded as the momentum map for
the gauge group action) is a homomorphism of Lie algebras from the Lie algebra of the
gauge group (equipped with the Lie bracket induced by [; ]) to the Lie algebra of functions
(equipped with the Poisson bracket). Equivalently, the constraints are a representation of the
Lie algebra of the gauge group. The constraint G generates gauge transformations through
the Poisson bracket:
fG(); Ak(x)g = @k  + [Ak; ]: (10)
The Hamiltonian is
H = 12 (E 2 + H 2):
As in the case of electromagnetism, the variables Fik are excludable. Likewise as in the case
of electromagnetism, the constraint equations are rst class: we must set G = 0 and also the
Poisson brackets fG; f g (for arbitrary functions f ) must all be set to zero.
The observables the reduced system are thus functionals
(A; E )jG=0 (11)
subject to the condition
fG; g = 0: (12)
Reduction and solving constraints break the manifest Lorentz invariance. So if we wish
to retain this invariance, we must learn how to describe dynamics (the functional integral)
without explicitly solving the constraints. To understand how to treat a simpler case, let us
return to a nite dimensional phase space , and recall the treatment of this situation as at
the beginning of Lecture 4. We assume we have a Lagrangian
N
X M
X
l= pidqi H +  '; (13)
i=1 =1
here we assume that M < N (so that the dimension 2(N M ) of the reduced phase space
is positive). To parametrize the physical phase space, we take
~ = f(p; q) 2 : '(p; q) = 0 for all g=G (14)
where G is the gauge group generated by the constraints ' through the Poisson bracket. (In
the Yang-Mills example the action of the constraints via Poisson bracket actually does expo-
nentiate to a group action, though this is not always true more generally.) To parametrize
the physical degrees of freedom we introduce a slice
 (p; q) = 0;  = 1; : : :; M (15)
which intersects each orbit of G exactly once: the slice (15) is called a subsidiary condition.
We choose the subsidiary conditions  to satisfy
f;  g = 0: (16)
In order to obtain a nondegenerate symplectic structure, we require also that
det(f';  g) 6= 0: (17)
3
Imposing the subsidiary conditions (15) and also the constraint equations
'(p; q) = 0
we nd a smaller phase space . We may take local coordinates on the big phase space
given by
q = (  ; q  )
and
p = (p; p);
here p and q are the physical variables, while p and  must be excluded. Let us write
the Liouville measure as
dp dq = ()(') det(f'; g)dpdq; (18)
if we were able to solve
'(p; q) = 0 () p = p(p; q) (19)
this would correspond to the transformation of the delta function under a change of variables,
(') = (@'=@p)(p p(p ; q)) (20)
which would give rise to the determinant in (18).
We additionally make use of the fact that
Z
(') = ei'd;
so that we can omit the delta function (') if the action already contains a term P '.
Thus the functional integral giving rise to the evolution operator is
Z Z
i
exp ~ (pq_ H ')dt dp(t)dq2(t~)d(t) () det(f; 'g):
Y
(21)
t
To pass to the S -matrix only the physical degrees of freedom are xed at the asymptotic
values of t, while the Lagrange multiplier and the values of the constraints are allowed to
vary freely. Returning to the Yang-Mills case, we get the path integral
Z  Z YY
i 1
exp ~ 4 ((@0A)E 2 (E + H ) A0(r
E ))dx
2 2
(@k Ak )(det MC )dAdEk ;
R t ~x
(22)
where as our subsidiary condition we have taken the Coulomb gauge condition,
@ k Ak = 0 ; (23)
and have introduced the operator MC on Lie(G)-valued functions on R3, dened by
MC  = @k @k  [Ak; ]: (24)
The only objects appearing in (22) which are not Lorentz invariant are det MC and @k Ak.
4
 We can explicitly integrate in (22) over the variables E to get
Z  Z Y
exp ~i 4 (F )2)dx dA(x)
YY
R
(@kAk ) det MC (t): (25)
x t ~x
We interpret this expression as an integral over A=G , where A is a space of connections and
G is a gauge group. Our subsidiary condition (23) is the choice of a point on the orbit of G:
the determinant det MC gives the correct measure.
To check that we have indeed obtained the correct measure on the space of G orbits
reduced from the measure Qx Q dA(x), it suces to check that
Z
dgxj@k Ak =0 = det1M ;
YY
(@k Agk ) (26)
x g 2G C

here dgx is the Haar measure on G at the point gx. We can linearize the action of G so that
innitesimally we obtain Z
(MC ) dx = det1M
Y
x C
where we have dened the operator MC by (24). After this understanding we can proceed
to choose a dierent subsidiary condition, in particular a manifestly Lorentz invariant one
@ A = c; (27)
where c is an arbitrary scalar function. The reduced measure is given by
Y Y
det ML (@A c(x)) dA(x)
x
where det ML is obtained by averaging the -function
1 = Z (@ Ag c(x)) Y dg(x):
 
det ML x
Here, we have dened the operator ML (compare the denition (24) of MC ) by
ML =  + @[A; ]: (28)
The path integral now becomes
Z  Z Y
exp i (F dx)
2
Y
(@A c)(det ML) dA (x): (29)
~ x x

The nal step is to average over c(x). Ultimately the expression (29) becomes
Z  Z Y
i Y Y i R c2 dx
exp ~ 4 (F ) dx (@A(x) c)(det ML) dA(x) e
2
(30)
R x c
Z  Z Y
= exp i (F )2 + (@ A )2)dx (det M ) dA (x):
Y
(31)
~ R4   
x
C 

5
Here,  is an arbitrary constant. The integrand in (30) is not local (because the determi-
nant of the operator MC is not local): to transform it into a local expression we introduce
Grassmann variables c = (c1; : : :; cr ) and c = (c1; : : :; cr ); in other words odd variables in a
Grassmann algebra, satisfying the anticommutation relations
cicj + cj ci = 0;
c2i = ci2 = 0
and recall the identity Z
det A = ecAc dcdc (32)
where A is an endomorphism of a vector space of dimension r and we have introduced the
corresponding endomorphism of a vector space of dimension 2r,
" #
0 A
A= A 0 : (33)

We now replace the Grassmann variables c and c (which took values in a nite dimensional
vector space) by Grassmann variables c and c taking values in the innite dimensional space
of functions on R4. Rewriting the determinant in (30) by formally applying (32), (30) is
transformed into the form
Z  Z   Y
i 
exp ~2 Tr R4 (F ) + (@A ) + (@c; @c + [A; c] dx dAdcdc: (34)
2 2

We have now transformed the path integral into an integral of the form R exp(iAction), where
the action is local and has a nondegenerate quadratic form, and we integrate over all elds
(which are assumed to satisfy appropriate boundary conditions). Here c and c are elds
which should be regarded as carrying charge (because of the terms mixing A, c and c )
and which should be regarded as fermionic (because they are odd elements in a superalgebra
or Grassmann algebra). The expression (34) is the generating functional from which the
S -matrix of Yang-Mills theory may be extracted. There are still the 3-valent and 4-valent
vertices in which all the lines emanating from the vertex correspond to the propagator of A.
Additionally we now have another 3-valent vertex with one A propagator, one c propagator
and one c propagator: the c and c propagators may be distinguished by writing an arrow
on the c propagator pointing toward the vertex and an arrow on the c propagator pointing
away from the vertex.
Near a classical solution of the equations of motion for the Yang-Mills elds, the correction
terms entering the perturbation expansion are proportional to F2 (recall that a term of the
form F2 appeared in our original Lagrangian). The term 12 (F2 ) in the Lagrangian is
renormalized to
( 1 + aln())F 2
2 
for an appropriate cuto parameter  and a suitable constant a. We notice that a must be
negative in order for the renormalization procedure to succeed: this property is referred to
in the literature as \asymptotic freedom".