hed a mu
h higher plane of
reativity when he bla
ked out everything but a,
an and the. That ere
ted more dynami
intralinear tensions.
, Cat h22
J. Heller
Nonperturbative Methods
in Gauge Theory
A Set of Le tures
by
Y. M. Makeenko
Samizdat, 1995
Yuri Makeenko
Institute of Theoreti
al and Experimental Physi
s, Mos
ow, Russia and
The Niels Bohr Institute, Copenhagen, Denmark
Prefa
e
These le
ture notes are based on
ourses given at:
1) Autonoma University of Madrid, Winter Semester of 1993;
2) Leipzig University, Winter Semester of 1995;
3) Mos
ow Physi
al and Te
hni
al Institute, Spring Semester of 1995.
My intention was to introdu
e graduate and Ph.D. students to the methods of
ontemporary quantum eld theory. The term \nonperturbative" in
the title means literally \beyond the s
ope of perturbation theory". Therefore, it is assumed that the reader is familiar with quantum me
hani
s as well
as with the standard methods of perturbative expansion in quantum eld
theory and, in parti
ular, with the theory of renormalization.
The se
ond purpose was to make the
ourse useful for senior people (in
luding those working in
ondensedmatter theory), as a survey of ideas, terminology and methods, whi
h were developed in quantum eld theory in the
seventies and the beginning of the eighties. For this reason, these notes do not
go very deeply into details, so the presentation is sometimes a bit s
hemati
.
Correspondingly, the subje
ts whi
h are usually
overed by modern
ourses
in string theory, su
h as the twodimensional
onformal eld theories, are not
tou
hed. It is assumed that su
h a
ourse will follow this one.
The main body of the le
ture notes deals with latti
e gauge theories and
largeN methods. These two Chapters are pre
eded by Chapter 1 whi
h
is devoted to the method of path integrals. The pathintegral approa
h is
loosely used in quantum eld theory and statisti
al me
hani
s. In Chapter 1,
I shall pay most attention to aspe
ts of the path integrals, whi
h are then
used in the next two Chapters.
In ea
h Chapter, I was going to be as
losed to the original papers, where
the involved methods were proposed, as possible. The list of these papers
respe
tively in
ludes:
i
ii
1. R.P. Feynman, An operator
al
ulus having appli
ations in quantum
ele
trodynami
s, Phys. Rev. 84 (1951) 108.
2. K.G. Wilson, Connement of quarks, Phys. Rev. D10 (1974) 2445.
3. G.'t Hooft, A planar diagram theory for strong intera
tions, Nu
l. Phys.
B72 (1974) 461.
The le
tures were followed by seminars where some problems for deeper
studies had been solved on a bla
kboard. They are inserted in the text as
the problems, whi
h
an be omitted at rst reading. Some more information
is also added as remarks after the main text. Both of them
ontain some
relevant referen
es.
The referen
es, whi
h are
olle
ted at the end of ea
h Chapter, are usually given only to either a rst paper (or papers) in a series or those
ontaining a pedagogi
al presentation of the material. With the modern ele
troni
database at qspires (SLAC), a list of subsequent papers
an, in most
ases,
be retrieved by downloading
itations of the rst paper.
I would like to thank the students for their attention, patien
e, and questions. I am indebted to Martin Gurtler for his help in preparing these le
ture
notes.
1995{1996
Y. M.
Contents
1 Path Integrals
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2
2
5
10
12
17
20
29
35
35
37
39
40
41
42
46
48
48
49
52
56
61
66
66
69
72
iv
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
76
78
82
87
88
88
91
96
99
102
102
105
111
113
117
121
123
127
127
132
135
138
141
145
149
149
151
156
158
162
165
165
169
171
174
179
182
CONTENTS
3 1=N Expansion
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
187
188
188
192
202
211
213
215
215
220
227
233
237
241
248
251
254
254
260
263
267
271
274
278
280
288
295
295
300
303
304
vi
CONTENTS
List of Figures
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
1.10
1.11
1.12
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8
13
14
39
40
57
59
64
67
70
77
78
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
98
100
103
104
104
106
107
114
119
120
122
125
128
131
133
vii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
viii
LIST OF FIGURES
2.16
2.17
2.18
2.19
2.20
2.21
2.22
2.23
2.24
2.25
2.26
2.27
2.28
2.29
2.30
2.31
2.32
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
134
138
141
146
147
148
154
156
157
161
163
164
172
177
179
180
181
3.1
3.2
3.3
3.4
3.5
3.6
191
192
192
193
196
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
3.17
3.18
3.19
3.20
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
218
220
221
222
223
224
225
225
226
228
229
230
232
233
235
LIST OF FIGURES
ix
238
240
242
244
247
255
258
261
270
271
272
277
282
283
284
298
LIST OF FIGURES
, Cat h22
J. Heller
Chapter 1
Path Integrals
The path integral is a method of quantization whi
h is equivalent to the
operator formalism. It re
overs the operator formalism in quantum me
hani
s
and perturbation theory in quantum eld theory (QFT).
The approa
h based on path integrals has several advantages over the
operator formalism. It provides a useful tool for nonperturbative studies
in
luding:
instantons,
analogy with statisti
al me
hani
s,
numeri
al methods.
A standard way of deriving the path integral is from the operator formalism:
operator formalism
()
path integral .
We shall pro
eed in the opposite dire
tion, following the original paper by
Feynman [Fey51.
CHAPTER 1.
PATH INTEGRALS
1.1.
The deltafun
tion (1) (x0 y0 ) emerges when (=t)2 is applied to the operator of the T{produ
t in (1.1.1).
Derive Eq. (1.1.2) in the operator formalism.
Let us apply the operator on the lefthand side (LHS) of Eq. (1.1.2) to
the T{produ
t whi
h is dened by
Problem 1.1
Solution
' (x) ' (y) = (x0 y0 )' (x) ' (y) + (y0 x0 )' (y) ' (x)
with
(x0 y0 ) =
1 for x0 y0
0 for x0 < y0 :
(1.1.5)
(1.1.6)
Eq. (1.1.3) implies a nonvanishing result to emerge only when (=x0 )2 is applied
to the operator of the T{produ
t. One gets
h0jT '_ (x) ' (y) j0i
x0
y0 ) h0j [' (y ) ; '_ (x) j0i = i (d) (x y ) ;
G (x y ) =
(1.1.1)
2
OPERATOR CALCULUS
m2 G ( x
y) =
i (d) (x
y) ;
(1.1.3)
1) (~
x
~y) ;
(1.1.4)
1 The ordered produ
ts of operators were introdu
ed by Dyson [Dys49. This paper and
other
lassi
al papers on quantum ele
trodynami
s are
olle
ted in the book edited by
S
hwinger [S
h58.
(1.1.7)
The expli
it solution to Eq. (1.1.2) for the free propagator is wellknown
and is most simply given by the Fourier transform:
G (x y) =
(1.1.2)
where d = 4 is the dimension of spa
etime, however the formulas are appli
able at any value of d. In the operator formalism, Eq. (1.1.2) is a
onsequen
e
of the free equations
2 m2 ' (x) 0 = 0 ;
0 2 m2 ' (x) = 0
= (1) (x0
dd p ip(x y)
i
:
e
d
2
p m2 + i"
(2)
(1.1.8)
An extra i" (with " ! +0) in the denominator is due to the T{produ
t
in the denition (1.1.1) and unambiguously determines the integral over p0 .
The propagator (1.1.8) is known as the Feynman propagator whi
h respe
ts
ausality.
Problem 1.2 Perform the Fourier transformation of the free momentumspa
e
propagator in the energy p0 :
+1
Z
p
dp0 e ip0 (t t0 )
i
G! (t t0 ) =
;
! = p~ 2 + m2 : (1.1.9)
2
2
2
p0 ! + i"
1
Solution
p0 =
! i" :
(1.1.10)
CHAPTER 1.
PATH INTEGRALS
For t > t0 (t < t0 ), the
ontour of integration
an be
losed in the upper (lower)
halfplane whi
h gives
i!(t t0 )
i!(t t0 )
G! (t t0 ) = (t t0 ) e
+ (t0 t) e
2!
2!
0
e i!jt t j
=
:
2!
The Green fun
tion (1.1.11) obeys the equation
2
t2
!2
G! t t0 = i (1) t t0
(1.1.11)
(1.1.12)
G (x y) =
2
x :
2
m
(1.1.13)
x jxi = x jxi :
(1.1.14)
dd x jxihxj
= 1;
= g(x) ;
hx j f i
(1.1.15)
= f (x) :
dd x f (x)jxi ;
hgj
dd y g(y)hyj :
(1.1.18)
dd x
(1.1.19)
hy j1j xi
= hyjxi = (d) (x y) ;
(1.1.20)
the formula
hy jOj xi
O (d) (x y)
(1.1.21)
dd x O(y; x) f (x) :
(1.1.22)
hg jOj f i
(1.1.23)
jf i
= O(y; x) :
If spa etime is approximated by a dis rete set of points, then the operator
while the wave fun tions, asso iated with hgj and jf i, are given by
hg j xi
The a
tion of a linear operator O on the bra and ket ve
tors in the Hilbert
spa
e is determined by its matrix element hy jOj xi, whi
h is also known as
the kernel of the operator O and is denoted by
hg jOj f i
OPERATOR CALCULUS
hy jOj xi
and therefore
oin
ides with the
ausal Green fun
tion for a harmoni
os
illator
with frequen
y ! .
y
1.1.
(1.1.17)
t =
i x4 :
(1.1.24)
CHAPTER 1.
PATH INTEGRALS
pM
i M
i ; i
t ~x
Minkowski spa e ;
(1.1.25)
1.1.
We shall use the same notation v for a fourve
tor in Minkowski and
Eu
lidean spa
es:
= (v ; ~v )
vM
0
vE = (~v; v4 )
pE =
i E
; i
~x x4
Eu lidean spa e :
(1.1.26)
p0
i p4
(1.1.27)
for the relation between energy and the fourth
omponent of the Eu
lidean
fourmomentum.
The passage to Eu
lidean spa
e results in
hanging the Minkowski signature of the metri
g to the Eu
lidean one:2
)
(+ + + +)
Minkowski signature
Eu lidean signature
(+
(1.1.28)
As su h, one gets
p2M = p20 p~ 2
p2E = p~ 2 p24 :
(1.1.29)
p x = Et + p~~x
! pExE = p~~x + p4 x4 :
(1.1.30)
f (p) =
f (x) =
Z
Z
OPERATOR CALCULUS
dd p ipx
e f (p) :
(2)d
(1.1.31)
2 An older generation is familiar with Eu
lidean notations whi
h are used throughout
the book by Akhiezer and Berestetskii [AB69. On the
ontrary, the two other
anoni
al
books on quantum eld theory by Bogoliubov and Shirkov [BS76 and by Bjorken and
Drell [BD65 use Minkowskian notations whi
h are due to Feynman.
Eu lidean spa e ;
(1.1.32)
iv4 :
(1.1.33)
with
v0 =
2 + m2 G (x y) = (d) (x y)
(1.1.34)
G (x y ) =
dd x e ipx f (x) ;
Minkowski spa e ;
dd p ip(y x) 1
e
:
p2 + m2
(2)d
(1.1.36)
Note that the RHS of Eq. (1.1.36) is nothing but the Fourier transform of the
free momentumspa
e Eu
lidean propagator, and there is no need to retain an
i" in the denominator sin
e the integration pres
ription is now unambiguous.
It is now
lear why we keep the same notation for the
oordinatespa
e
Green fun
tions: the Feynman propagator in Minkowski spa
e and the Eu
lidean propagator. They are the same analyti
fun
tion of the timevariable.
CHAPTER 1.
t = i
q
z
Minkowski
spa
e
6E =
tm
XXnEu
lidean
spa
e
}
PATH INTEGRALS
H
Eu
lidean
spa
e
q
XX
ip4 Em
q
H
a)
}
{q
Minkowski
spa
e
b)
Fig. 1.1: Dire
tion of the Wi
k rotation from Minkowski to Eu
lidean spa
e (indi
ated by the arrows) for a) time and b) energy. The dots represent singularities of a free propagator in a)
oordinate and b) momentum spa
es.
The
ontours of integration in Minkowski spa
e are asso
iated with
ausal
Green fun
tions. They
an obviously be deformed in the dire
tions of the
arrows.
Problem 1.3
Solution
G! (
0) =
+1
Z
dp4 ip4 ( 0 ) 1
e
:
2
p24 + ! 2
(1.1.37)
The integral on the RHS
an be
al
ulated for > 0 ( < 0 ) by
losing the
ontour
in the lower (upper) halfplane, and taking the residues at p4 = i! (p4 = i! ),
respe
tively. This yields
G! (
!( 0 )
0) e
2!
0
e !j j :
2!
0 ) = (
+ ( 0
) e
!( 0 )
2!
(1.1.38)
The Eu
lidean Green fun
tion (1.1.38)
an obviously be obtained from the
Minkowskian one, Eq. (1.1.11), by the substitution
= it ;
0 = it0
(1.1.39)
1.1.
OPERATOR CALCULUS
2 + ! 2 G 0 = (1) 0
(1.1.40)
!
2
and, therefore, is the Green fun
tion for a Eu
lidean harmoni
os
illator with frequen
y ! .
f M ; M g = 2 g I :
(1.1.41)
Therefore,
0 is Hermitean while the Minkowskian spatial
matri
es are antiHermitean.
Analogously, the Eu
lidean
matri
es satisfy
f
;
g = 2 I ;
(1.1.42)
(ipb + m)
= 0
(1.1.46)
10
CHAPTER 1.
PATH INTEGRALS
(1.1.48)
sin e the order of operators is now essential. In parti ular, one has
6 eA eB ;
e A+B =
(1.1.49)
M
1
= lim
= lim 1 + (A + B )
M !1
M !1
M
1
1
1
1 + (A + B ) 1 + (A + B ) 1 + (A + B ) :
M
M
M
{z
} (1.1.50)

M times
e A+B
The stru
ture of the produ
t on the RHS prompts us to introdu
e an index
i running from 1 to M and repla
e (A + B ) in ea
h multiplier by (Ai + Bi ).
Therefore, one writes
e A+B =
lim
M !1
M
Y
i=1
1+
1
= lim 1 + (AM + BM )
M !1
M
where the index i
ontrols the order of the operators whi
h are all treated
dierent. The ordering is su
h that the larger i is, the later the operator
with the index i a
ts. This order of operators is pres
ribed by quantum me
hani
s, where initial and nal states are represented by ket and bra ve
tors,
respe
tively.
The RHS of Eq. (1.1.51)
an be rewritten as
1 PM
1
1 + (A1 + B1 ) ; (1.1.51)
M
R1
d(A()+B ())
;
(1.1.53)
where A(i=M ) = Ai and B (i=M ) = Bi while the operator A() + B () a
ts
at order .
Eq. (1.1.53) is in fa
t obvious sin
e it only involves the operator A + B
whi
h
ommutes with itself. For
ommuting operators there is no need for
ordering so that A() + B () does not depend on in this
ase. The integral
in the exponent on the RHS of Eq. (1.1.53)
an then be done, and reprodu
es
the LHS.
Eq. (1.1.53)
an however be manipulated as though A() and B () were
just fun
tions rather than operators sin
e the order would be automati
ally
spe
ied by the pathordering operation. This is analogous to the wellknown
fa
t that operators
an be written in an arbitrary order under the T{produ
t.
Therefore, we
an rewrite Eq. (1.1.53) as
e A+B = P e
R1
R
d0 A(0 ) 01 dB ()
:
This is the operator analog of the law of addition of exponents.
Problem 1.4
Solution
11
OPERATOR CALCULUS
in B .
1
(A + Bi )
M i
1.1.
(1.1.54)
Cal ulate expli itly the rst term of the expansion of exp (A + B )
eA +
Z1
0
R
R1
0 0
d0 A(0 )
B ( ) e 0 d A( ) + : : :
d e
(1.1.55)
12
CHAPTER 1.
PATH INTEGRALS
There is no need for a pathordering sign in this formula, sin
e the order of the
operators A and B is written expli
itly. There is also no ambiguity in dening the
exponentials of the operator A as already explained.
Sin
e the order is expli
it, one drops the formal dependen
e of A and B on the
ordering parameter whi
h gives
e A+B = e A +
Z1
d e (1 )A B e A + : : :
1.1.
z (t)
6
z ( )
(1.1.56)
The formulas (1.1.55) and (1.1.56) are known from timedependent perturbation
theory in quantum me
hani
s.
Pe
R
0
z (0)=x
dt z_ (t) (t) (d)
(x y) ;
13
OPERATOR CALCULUS
(1.1.58)
Fig. 1.2: The traje
tory z (t). The operator (t) a
ts at the order t.
where the integration runs over all traje
tories z (t) whi
h begin at the point
x, as depi
ted in Fig. 1.2.
Sin
e the operator (t) a
ts at the order t, these operators are ordered
along the traje
tory z (t) with P, in Eq. (1.1.58), standing for the pathordering operator. Note, that z_ (t) and (t)
ommute sin
e
d
(t)z_ (t) = = 0
(1.1.59)
dt
so that their order is inessential in Eq. (1.1.58). With these rules of manipulations, Eq. (1.1.58)
an be proven by the \translation"
z (t)
z 0 (t) = z (t) +
Zt
dt0 (t0 )
(1.1.60)
14
CHAPTER 1.
1.1.
15
OPERATOR CALCULUS
z1
PATH INTEGRALS
LL
L
LL
L !!z3
L!z2
Pe
((( z6
z5
R
Pe
M
Y
= lim
!0 i=1 [1 + (zi
zi 1 ) D (i) :
(1.1.63)
R
0
dt z_ (t) (t)
= exp (z ( ) x )
;
x
(1.1.64)
whi
h is nothing but the shift operator. Applying the operator on the RHS
of Eq. (1.1.64) to the deltafun
tion, one gets

Pe
Fig. 1.3: Dis
retization of the traje
tory z (t) (depi
ted for M = 6).
The
ontinual path integral
an be approximated by a nite one. To this
aim, let us
hoose M points ti = i, where is the step of dis
retization, and
M = =. We then
onne
t the points
z0 = x ; zi = z (i) i = 1; 2; : : : ; M
(1.1.61)
by straight lines. Su
h a dis
retization of the traje
tory z (t) is depi
ted in
Fig. 1.3. The measure in Eq. (1.1.58)
an then be dis
retized by
Dz (t) : : : =
dt z_ (t)D (t)
z4
M Z
Y
i=1
dd zi
::: :
(2)d=2
(1.1.62)
The expli
it form of the operator in Eq. (1.1.34) was not essential
in deriving Eq. (1.1.58). If in Eq. (1.1.34) is repla
ed by an arbitrary
operator D with non
ommuting
omponents, then Eq. (1.1.58) holds with
(t) substituted by D (t). The dis
retized pathordered exponential of a
R
0
y) = (d) (z ( ) y) :
(1.1.65)
Therefore, z ( ) has to
oin
ide with y due to the deltafun
tion, whi
h
disappears after the integration over z ( ) has been performed. Thus the
nal answer is
Z
Z1
R
1 dt z_ 2 (t)
1 m2
1
: (1.1.66)
2
2
0
d e
Dz (t) e
G (x y ) =
2
0
z (0)=x
z ( )=y
This path integral goes over all traje
tories z (t) whi
h
onne
t the initial
point x and the nal point y .
Problem 1.5 Derive Eqs. (1.1.58) and (1.1.66) by introdu
ing a path integral over
velo
ity v (t) = z_ (t).
Solution The operator on the RHS of Eq. (1.1.57)
an be disentangled by the
following Gaussian path integral
e2
R
0
dtD2 (t)
Dv (t) e
1 dtv2 (t)
2 0
R
0
(1.1.67)
16
CHAPTER 1.
PATH INTEGRALS
Substituting D (t) = (t) and
al
ulating the a
tion of the pathordered exponential on (d) (x y ), we get
Z
Z1
R
2
1
1 2
G(x y ) = 1 d e 2 m Dv (t) e 2 0 dtv (t)
2
0
(d) (x +
Z
0
dt v (t) y ):
the measure
dt v 2 (t)
!
0
Z
Dv (t) : : : =
M Z
Y
i=1
M
X
i=1
vi2 ;
dd vi : : :
(2=)d=2
(1.1.70)
(1.1.71)
obviously re
overs Eq. (1.1.62) after
al
ulating the Ja
obian from the variables vi
to the variables zi . Therefore, Eq. (1.1.68) reprodu
es Eq. (1.1.66) provided
z (t) = x +
Zt
dt0 v (t0 ) :
(1.1.72)
The dis
retized traje
tory in Fig. 1.3
an be analyti
ally written as the expansion
M
X
i=1
are nonvanishing only for the ith and (i+1)th intervals. The measure (1.1.62)
is dened, therefore, via the
oe
ients zi as a multiple produ
t of dzi 's.
While the basis fun
tions fi (t) are not orthogonal:
1
Z
dt fi (t)fj (t) =
1
1
2
ij + i(j+1) + i(j
3
6
6
(1.1.73)
(1.1.74)
1) ;
(1.1.75)
z_ (t) =
where
M
X
i=1
(zi
(1.1.76)
1=
for t 2 [(i 1); i ; :
(1.1.77)
0
otherwise
This shows why the dis
retized velo
ities from Problem 1.5 are natural variables.
One
an
hoose, instead, another set of (orthogonal) basis fun
tions and
expand
i (t) =
z (t) =
M
X
n=1
n n (t)
(1.1.78)
with some oe ients n . Then the measure (1.1.62) takes the form
Dz (t) : : :
z (t) =
17
OPERATOR CALCULUS
(1.1.68)
The integration over Dv (t) in this formula has no restri
tions.
To derive Eq. (1.1.66) from Eq. (1.1.68), let us note that the dis
retized velo
ities
read
z zi 1
vi = i
:
(1.1.69)
Sin
e
Z
1.1.
M
Y
n=1
dd n : : :
(1.1.79)
z (t) =
y x
t + (t) + x:
(1.1.80)
18
CHAPTER 1.
PATH INTEGRALS
The boundary
onditions for the variable (t) are determined by Eq. (1.1.80)
to be
(0) = ( ) = 0:
1.1.
Problem 1.6
Solution
(1.1.81)
dt z_ 2 (t)
(y x)2
(y x)
=
+2
( ( ) (0)) +
Z
The se
ond term on the RHS vanishes due to the boundary
onditions
(1.1.81) so that the propagator be
omes
1
G (x y) =
2
:
Z1
1 m2
2
d e
(y x)2
2
D e
1 dt _2 (t)
2 0
:
(1.1.83)
(0)=( )=0
D e
1
2 0
dt _2 (t)
= F ( ) :
(1.1.84)
(1.1.85)
F ( )
1
:
(2 )d=2
Z M
Y1
i=1
dd zi e
(2)d=2
1
2
PM
2
i=1 (zi zi 1 ) =
(1.1.87)
(0)=( )=0
Z1
2
2
dd p ip(x y) 1
d e 2 (p +m )
e
G (x y) =
d
2
(2)
0
Z1
(x y)2
2
1
1
1
d e 2 m e 2
=
:
2
(2 )d=2
1
=
(2)d=2
1 dt z_ 2 (t)
2 0
Problem 1.7
The path integral over on the RHS of Eq. (1.1.83) is a fun
tion solely of
Z
Dz (t) e
z (0)=x
z ( )=y
Substituting Eq. (1.1.80) into the exponent in Eq. (1.1.66), one gets
Z
19
OPERATOR CALCULUS
(1.1.86)
jzi zi 1 j p
(1.1.89)
are essential as ! 0. Su
h traje
tories are typi
al Brownian traje
tories. They
are
ontinuous as ! 0 but not smooth (jzi zi 1 j for smooth traje
tories).
In mathemati
al language, these fun
tions are said to belong to the Lipshitz
lass
1/2.
e.g., the books [Ka 59, S h81, Wie86 for a des ription of the pathintegral approa h to Brownian motion.
20
CHAPTER 1.
PATH INTEGRALS
d
dt
! (t) ;
0;
(1.1.91)
s =
ds
(1.1.92)
z_ 2()d
(1.1.93)
yx
where
ds =
t =
S [ yx
Z
2
m2 + 12 dt z_ 2 (t) :
(1.1.95)
4 The notation S [ with square bra kets means that S is a fun tional of
while f (x)
21
OPERATOR CALCULUS
The sense of this notation is that the RHS
oin
ides with the
lassi
al a
tion of
a relativisti
free (s
alar) parti
le in the propertime parametrization (1.1.94)
when
Z
(1.1.90)
Noti
e that yx stands for a traje
tory as a geometri
obje
t, while z (t)
is a fun
tion whi
h des
ribes a given traje
tory in some parametrization t.
This fun
tion (but not the geometri
obje
t itself) depends on the
hoi
e of
parametrization and
hanges under the reparametrization transformation
1.1.
dtz_ 2(t) = m
Z
ds = m Length[
(1.1.96)
sin
e
dz (s)
ds
2
= 1
(1.1.97)
G(x y) =
X
yx
e S[
yx :
(1.1.98)
This sum is split in Eq. (1.1.66) into the traje
tories along whi
h the parti
le
propagates during the proper time and the integral over .
Equation (1.1.98) implies that the transition amplitude in quantum me
hani
s is a sum over all paths whi
h
onne
ts x and y. In other words, a
parti
le propagates from x to y along all paths yx in
luding the ones whi
h
are forbidden by the free
lassi
al equation of motion
z (t) = 0:
(1.1.99)
Only the
lassi
al traje
tory (1.1.99) survives the path integral in the
lassi
al limit h ! 0. The reason for this is that if the dependen
e on the
Plan
k's
onstant is restored, it appears in the exponent:
G(x y) =
X
yx
e S[
yx =h :
(1.1.100)
22
CHAPTER 1.
PATH INTEGRALS
Sfree[ = m Length[ ;
(1.1.101)
where the length is given for some parametrization of the traje
tory
Zf
Length[ =
d z_ 2 ():
by
(1.1.102)
i
H = 2m + V (x)
H xE
= Dv (t) e
m
2
R
0
dtv2 (t)
R
0
R
0
y ): (1.1.104)
Here the argument t in V (x; t) is just the ordering parameter, while the same formula
holds when the potential is expli
itly time dependent.
In
ontrast to Eq. (1.1.67), we have put the minus sign in front of the linearinv term in the exponent in Eq. (1.1.104), so that it agrees with Appendix B
of Feynman's paper [Fey51. In fa
t it does not matter what sign is used sin
e
the integral over v (t) is Gaussian so that only even powers of v survive after the
integration.
The pathordered exponential in Eq. (1.1.104) reads expli
itly
P
R
0
dt v (t) (t)
R
0
dtV (x;t)
= lim
!0
M h
Y
i=1
1 vi
x
whi
h
an be rewritten as
P
R
0
R
0
dtV (x;t)
= lim
!0
M h
Y
1 vi
i=1
R
0
R
0
dtV (x;t)
h
Y
t=0
1 dtv (t)
i [1 dt V (x; t) : (1.1.107)
x
There is no need to write down the tdependen
e of (t) in these formulas sin
e
the order of the operators is expli
it.
To disentangle the operator expression (1.1.107), let us note that
[1 dtv (t) = U 1 (t + dt)U (t)
with
U (t) = exp
Z t
0
(1.1.108)
(1.1.109)
(1.1.110)
1 dt V x +
Z t
0
dt0 v (t0 ); t
; (1.1.111)
1 dt v (t)
= U 1 ( )
i [1 dt V (x; t)
x
Y
t=0
1 dt V x +
= U 1 ( ) exp 4
23
OPERATOR CALCULUS
(1.1.103)
as a path integral.
Solution
The
al
ulation is similar to the one whi
h is already done in Subse
t. 1.1.4. It is most
onvenient to use a pathintegral over velo
ity whi
h was
onsidered in Problem 1.5. The proper disentangling formula reads
D
y e
Z
1.1.
Z
0
Z t
dt V x +
dt0 v (t0 ); t
Z t
0
3
dt0 v (t0 ); t 5;
(1.1.112)
24
CHAPTER 1.
PATH INTEGRALS
(d) (x y ) =) (d) x +
Z
0
dt v (t) y :
H xE =
Dz (t) e
R
0
dt L(t)
(1.1.114)
z (0)=x
z ( )=y
(1.1.115)
is the Lagrangian asso
iated with the Hamiltonian H . The unusual plussign in
this formula is due to Eu
lideanspa
e formalism. It is
lear from the derivation
that Eq. (1.1.114) holds for timedependent potentials as well.
Noti
e that the path integral in Eq. (1.1.114) is now over traje
tories whi
h the
parti
le propagates in the xed proper time with no integration over .
A spe
ial
omment about the operator U 1 ( ) in Eq. (1.1.112) is needed. In the
S
hrodinger representation of quantum me
hani
s, one is interested in the matrix
elements on the evolution operator between some ve
tors hg j and jf i in the Hilbert
spa
e. A
ording to Eq. (1.1.23) one has in the
oordinate representation
D
g e
H f E =
dd x g (x) e H f (x) :
(1.1.116)
g (x) =) U ( ) g (x) U 1 ( ) = g x +
Z
0
dt v (t) :
(1.1.117)
H f E =
Problem 1.9
potential V (x):
R! (x; x; V ) =
Dz (t) e
R
0
dt L(t)
E
1
2
2
G + ! + V x
(1.1.119)
z (0)=x
z ( )=x
z (t) = x ;
(1.1.121)
whi
h is asso
iated with a parti
le standing at the point x. Substituting V at this
saddlepoint, i.e. repla
ing V (z (t)) by V (x), and
al
ulating the Gaussian integral
over quantum
u
tuations around the traje
tory (1.1.121) by the use of Eqs. (1.1.84)
and (1.1.86), one gets
R! (x; x; V ) =
1
p
2 ! 2 + V (x)
(1.1.122)
in d = 1.
Equation (1.1.122)
an be alternatively derived by applying the Gelfand{Di
key
te
hnique [GD75 whi
h says that R! (x; x; V ) obeys the third order linear dierential equation
1 1 3
2 2 G
(1.1.123)
Problem 1.10
Solution
There are no restri
tions on the initial and nal points of the traje
tories z (t) in
this formula.
D
x
where
25
OPERATOR CALCULUS
(1.1.113)
1.1.
R! (x; y ; V ) =
D
y
E
1
G 2 + !2 + V x
(1.1.124)
26
CHAPTER 1.
2
G x
2
PATH INTEGRALS
2 + ! 2 + V (y ) R (x; y ; V ) = (1) (x y ) :
G y
!
2
(1.1.125)
+ ! 2 + V (x) f (x) = 0 ;
G x
2
(1.1.126)
1.1.
sure by
One more dierentiation of Eq. (1.1.130) with respe
t to x results in Eq. (1.1.123).
It is worth noting that Eq. (1.1.130) is very
onvenient for
al
ulating the semi
lassi
al expansion of R! (x; x; V ) in G . In parti
ular, the leading order (1.1.122) is
obvious.
G(x y)
Dz () e
z (i )=x
z (f )=y
R
m0 f d z_ 2 ()
i
(1.1.131)
Cal ulate the path integral in Eq. (1.1.131), dis retizing the mea
Dz
being the Wronskian of these solutions. Applying =x to Eq. (1.1.128), it is easy
to show that W! is an xindependent fun
tion of ! .
The simplest way to prove Eq. (1.1.123) is to dierentiate
27
OPERATOR CALCULUS
1 Y
M
X
dd zi
(2)d=2
M =1 i=1
(1.1.132)
(zi 1 ! zi ) = e m0 jzi zi 1 j
(1.1.134)
being an (unnormalized) probability fun
tion and is a parameter of the dimension
of [length2 . The probability interpretation of ea
h term in the sum is standard for
random walk models, and means, as usual, that a parti
le propagates via independent intermediate steps. The dis
retization of the measure given by Eq. (1.1.133)
looks like that in Eq. (1.1.62), but the summation over M is now added.
Sin
e the integral in Eq. (1.1.133) is a
onvolution, the
entral limit theorem
states that
M
1 X
0
G (x y ) =
(2)d=2 M (2m20 )d=2
21 d=2 e m20 (x y)2 =(22 M )+O(M
(2 M )
2)
(1.1.135)
28
CHAPTER 1.
PATH INTEGRALS
at large M , where 0 and 2 are the zeroth and (normalized) se ond moments of :
0 =
2
dd x e jxj = 2 d=2
1
=
0
G (x y )
(1.1.136)
p 0
(1.1.137)
2
m20
G (x y ) =
e
Ea
h term with M
X
m2
2 2 M
m2
d=2
M
dM ln (mo =m
) m20 (x y)2 =(22 M )+O(M 2 ) :
(1.1.139)
(1.1.140)
This justies the using of the
entral limit theorem in this
ase. The typi
al
distan
es between the zi 's, whi
h are essential in the integral on the RHS of
Eq. (1.1.133), are
jzi zi 1 j m10 p
(1.1.141)
like in Eq. (1.1.89). The relation (1.1.138) between the essential values of M and
is also similar to what we had in Subse
t. 1.1.4.
The sum over M in Eq. (1.1.139)
an be repla
ed by a
ontinuous integral over
the variable
2 M
=
;
(1.1.142)
m2
whi
h is O(1) for M
d m2 m 2 > 0 ;
2 0
(1.1.143)
29
OPERATOR CALCULUS
(d)
;
(d=2)
dd x x2 e jxj = d(d + 1) :
m0 > m =
1.1.
m2
2
!0
Z1
d
1
e
(2 )d=2
1 2 (x y)2
2m
2
(1.1.144)
whose RHS is proportional to that in Eq. (1.1.85) for the Eu lidean propagator.
p ! p eA (x) :
(1.1.145)
! r = ieA (x) :
(1.1.146)
Inverting the operator r2 by the use of the disentangling pro
edure, one
gets
G (x; y; A)
Z1
1
=
d e
2
0
y
Z
1 m2
2
1
2
2
r + m x
Dz (t) e
z(0)=x
z( )=y
1
2 0
R
dt z_ 2 (t)+ie 0 dt z_ (t)A (z(t))
: (1.1.147)
30
CHAPTER 1.
PATH INTEGRALS
Noti
e that the exponent is just the
lassi
al (Eu
lidean) a
tion of a parti
le
in an external ele
tromagneti
eld. Therefore, this expression is again of the
type in Eq. (1.1.98).
The pathintegral representation (1.1.147) for the propagator of a s
alar
parti
le in an external ele
tromagneti
eld is due to Feynman [Fey50 (Appendix A).
Derive Eq. (1.1.147) using Eq. (1.1.67) with D = r .
The
al
ulation is analogous to that of Problem 1.8. We have
Problem 1.12
Solution
D (t)
(1.1.148)
e
=
R
0
dt v (t)r (t)
h
Y
t=0
dtv (t)
h
Y
t=0
1 dt v (t)
i [1 + ie dt v (t)A (x; t) :
x
(1.1.149)
ie v (t)A (x; t) :
(1.1.150)
Substituting this potential in Eq. (1.1.114) and remembering the additional integration over , we obtain the pathintegral representation (1.1.147).
G (x; y; A) =
X0 ie
yx
yx
dz A (z)
(1.1.151)
where, pro
eeding as in Subse
t. 1.1.6, we have in
luded the free a
tion in
the denition of the sum over traje
tories:
X0
X
(1.1.152)
e Sfree [ yx ;
=
yx
yx
and rewritten the (parametri invariant) integral over dt as the ontour integral over
dz = dt z_ (t)
along the traje
tory yx.
(1.1.153)
31
OPERATOR CALCULUS
U [ yx = e
ie
yx
dz A (z)
(1.1.154)
A (z )
! A (z ) + 1e (z ) ;
g.t.
(1.1.155)
U [ yx g.t.
! e i(y) U [ yx e i(x) :
(1.1.156)
Noting that a wave fun
tion at the point x is transformed under the gauge
transformation (1.1.155) as
'(x)
V (x; t) =
1.1.
g.t.
e i(x) '(x);
(1.1.157)
we
on
lude that the phase fa
tor is transformed as the produ
t '(y)'y (x):
U [ yx g.t.
\'(y)'y (x)" ;
(1.1.158)
U [ yx '(x)
g.t.
\'(y)" ;
(1.1.159)
'y (y) U [ yx
g.t.
\'y (x)" :
(1.1.160)
and analogously
Equations (1.1.159) and (1.1.160) show that the phase fa
tor plays the
role of a parallel transporter in an ele
tromagneti
eld, and that in order to
ompare phases of a wave fun
tion at points x and y, one should rst make a
parallel transport along some
ontour yx . The result is, generally speaking,
dependent ex
ept when A (z ) is a pure gauge. The su
ient and ne
essary
ondition for the phase fa
tor to be independent is the vanishing of the eld
32
CHAPTER 1.
PATH INTEGRALS
1.1.
Problem 1.14
G(x; y ; A) =
Solution
ln det D = Sp ln D ;
(1.1.162)
whi
h relates the determinant and the tra
e of a Hermitean operator (or a
matrix) D.
Prove Eq. (1.1.162).
Let us rst redu
e D to a diagonal form by a unitary transformation
and denote eigenvalues as Di . Then Eq. (1.1.162) reads
Problem 1.13
Solution
ln
Di =
ln Di
(1.1.163)
The phase fa
tor for a
losed
ontour enters Eq. (1.1.161). It des
ribes
a parallel transportation along a
losed loop, and is gauge invariant as a
onsequen
e of Eq. (1.1.156):
H
g.t.
e ie dz A (z) ! e ie dz A (z) :
(1.1.164)
z(0)=z( )
where the path integral goes over traje
tories whi
h are
losed due to the
boundary
ondition z (0) = z ( ). To derive Eq. (1.1.161), we have used the
formula
33
OPERATOR CALCULUS
yx yx
d
(d) (
where
$
=
z) : : : =
X0
yz
$ X0
::: ;
z
zx
!
+ :
+ :::
y (1.1.165)
(1.1.166)
(1.1.167)
Problem 1.15
ln det r2 =
1
2
(1.1.168)
+ ::: ;
with the
ombinatori
fa
tor 1=2 in the third diagram being asso
iated with a
symmetry fa
tor.
Solution The same as in the previous problem.
Z = Sp e H
(1.1.169)
34
CHAPTER 1.
PATH INTEGRALS
with the partition fun
tion of a
losed elasti
string, whose energy is proportional to its length, and intera
ts with an external ele
tromagneti
eld. This
shows an analogy between Eu
lidean quantum me
hani
s in ddimensions and
statisti
al me
hani
s in d (spatial) and 1 (temporal)dimensions whose timedependen
e disappears, sin
e nothing depends on time at equilibrium. We
shall explain this analogy in more detail in Chapter 2 (Se
t. 2.5) when dis
ussing quantum eld theory at nite temperature.
1.2.
35
SECOND QUANTIZATION
Z =
D' (x) e S
where the a
tion in the exponent is given for the free
ase by
Z
1
Sfree[' =
dd x ( ')2 + m2 '2 :
2
The measure D'(x) is dened analogously to Eq. (1.1.62):
+1
Z
Y Z
D' (x) : : : =
d' (x) : : :
x 1
(1.2.2)
(1.2.3)
(1.2.4)
6 It is also alled the statisti al sum in an analogy with statisti al me hani s (see Sub
se t. 1.1.7).
36
CHAPTER 1.
PATH INTEGRALS
where the produ
t runs over all spa
e points x and the integral over d' is the
Lebesgue one.
The propagator is given by the average
G (x; y) =
h F [' i
= Z
h1i = 1:
(1.2.6)
(1.2.7)
1
x ;
2
2
+m
h 'x 'y i
Z
Solution
(1.2.8)
d'x exp
1
2
x;y
(1.2.9)
'x
1=2
xy 'y :
1
y
i
x :
(1.2.12)
f y ; x g
= 0;
y ; x = 0 ;
y ; x = 0 :
(1.2.13)
= x2 = 0 :
(1.2.14)
'0x =
(1.2.11)
Path integrals over anti
ommuting Grassmann variables are used to des
ribe
fermioni
systems.
The Grassmann variables x and y obey the anti
ommutation relations
d
D' e i d x 'D' ;
whi
h
oin
ides with (1.1.57). Therefore, we have obtained the same propagator (1.1.85) as in the previous Se
tion.
Problem 1.16
D' e iS =
Sin
e the free a
tion (1.2.3) is Gaussian, the average (1.2.5) equals
y
equals
The notation is obvious sin
e we average on the RHS of Eq. (1.2.6) over all
eld
ongurations with the same weight as in the partition fun
tion (1.2.2).
The normalization fa
tor Z 1 provides the ne
essary property of an average
G (x y) =
37
SECOND QUANTIZATION
(1.2.5)
1.2.
(1.2.10)
Note that the integrals over '(x) are
onvergent in Eu
lidean spa
e. If
a dis
retization of spa
e is introdu
ed, the path integrals in Eqs. (1.2.2) or
(1.2.6) are rigorously dened.
D D e
R d
d x
= det D
(1.2.15)
D'y D' e
R d y
d x'
D' = (det D) 1 :
(1.2.16)
38
CHAPTER 1.
PATH INTEGRALS
1.2.
d x=
d x y=
d x y =
Z
Z
d x y = xy ;
d x y = 0 :
(1.2.17)
d x d x e
x x
= 1:
(1.2.18)
The average over the Fermi elds, dened with the same weight as in
Eq. (1.2.15), equals
(x) (y) =
1 x
(1.2.19)
(1.2.20)
= e
Sp ln D :
S [' =
dd x
1
1
( ')2 + m2 '2 + '3 ;
2
2
3!
(1.2.21)
The relative minus sign in the exponents on the RHS's of Eqs. (1.2.20)
and (1.2.21) is a famous minus sign whi
h emerges for
losed fermioni
loops
whi
h
ontribute to the (logarithm of the) partition fun
tion.
(1.2.22)
' (x)
3!
dd x1 '3 (x1 )
3!
free
free
+ : : : (1.2.23)
The term whi
h is linear in (as well as the Gaussian average of any odd
power of ') vanishes due to the re
e
tion symmetry ' ! ' of the Gaussian
a
tion. The last term on the RHS is depi
ted graphi
ally in Fig. 1.4.
Further
al
ulation of the RHS of Eq. (1.2.23) is based on the free average
Fig. 1.4: Diagrammati representation of the last term on the RHS of Eq. (1.2.23).
The formula (1.2.15)
an now easily be derived: representing hy jD j xi in the diagonal form, expanding the exponential up to a term whi
h is linear in all the Grassmann variables and
al
ulating the integrals of this term a
ording to Eq. (1.2.17).
See more details in the book by Berezin [Ber86 (x3 of Part I).
d x = 0 ;
SS
S
x2
S
x1 S
S
Problem 1.17
Solution
39
SECOND QUANTIZATION
= G (xi
xj )
(1.2.24)
40
CHAPTER 1.
PATH INTEGRALS
x1
x2
x2
41
SECOND QUANTIZATION
S [' w.s.
= h
;
' (x)
' (x)
x1
a)
1.2.
b)
Fig. 1.5: Some of the Feynman diagrams whi
h appear from (1.2.23) after the Wi
k
pairing.
(1.2.27)
where we have expli
itly written the dependen
e on the Plan
k's
onstant h.
It appears this way sin
e the a
tion S is divided by h in Eq. (1.2.2) when h
is restored.
Equation (1.2.27) is to be understood in the weak sense, i.e. it is valid
under the sign of averaging when applied to a fun
tional F ['. In other
words the variation of the a
tion on the LHS of Eq. (1.2.27)
an always be
substituted by the variational derivative on the RHS when integrated over
elds with the same weight as in Eq. (1.2.6). Therefore, one arrives at the
following fun
tional equation
fa
tor in the rst term on the RHS of Eq. (1.2.23)). This is a general property
that only
onne
ted diagrams are left in h ' (xi ) ' (xj ) i.
Let us note nally that the
ombinatori
fa
tor 1=2 is
orre
tly reprodu
ed
for the diagram of Fig. 1.5a. For an arbitrary diagram, this pro
edure of
pairing reprodu
es the usual
ombinatori
fa
tor whi
h is equal to the number
of automorphisms of the diagram (i.e. the symmetries of a given graph).
In order to show how Eq. (1.2.28) reprodu es Eq. (1.1.34) for the free propagator, let us hoose
' (x)
(1.2.25)
This invarian
e is obvious sin
e the fun
tional integration goes over all the
elds while the shift (1.2.25) is just a transformation from one eld
onguration to another.
Sin
e the measure is invariant, the path integral in the average (1.2.6)
does not
hange under the shift (1.2.25):
Z
dd x ' (x)
D' e S['
F ['
S ['
F [' +
'(x)
'(x)
= 0:
(1.2.26)
S ['
F ['
' (x)
F ['
:
= h
' (x)
(1.2.28)
(1.2.29)
Substituting into Eq. (1.2.28) and
al
ulating the variational derivative, one
gets
' (y)
' (x)
= h (d) (x y)
(1.2.30)
Sfree
=
' (x)
2 + m2 ' (x)
(1.2.31)
42
CHAPTER 1.
PATH INTEGRALS
1.2.
while the RHS, whi
h results from the variational derivative, emerges in the
operator formalism from the
anoni
al
ommutation relations
(x0 y0 ) [' (x0 ; ~x) ; '_ (y0 ; ~y) = i (d) (x y)
(1.2.32)
as is explained in Subse
t. 1.1.1.
For this reason, the RHS of Eq. (1.2.30) and, more generally, the RHS of
Eq. (1.2.28) is
alled the
ommutator term. The variational derivative on the
RHS of Eq. (1.2.27) plays the role of the
onjugate momentum in the operator
formalism. The
al
ulation of this variational derivative in Eu
lidean spa
e is
equivalent to dierentiating the T{produ
t and using
anoni
al
ommutation
relations in Minkowski spa
e.
When the Plan
k's
onstant vanishes, h ! 0, the RHS of Eq. (1.2.27)
(or Eq. (1.2.28)) vanishes. Therefore it redu
es to the
lassi
al equation of
motion for the eld ':
S ['
= 0:
(1.2.33)
' (x)
This implies that the path integral over elds has a saddle point as h ! 0
whi
h is given by Eq. (1.2.33).
Another lesson we have learned is that the average (1.2.5), whi
h is dened via the Eu
lidean path integral, is asso
iated with the Wi
k rotated
T{produ
t. We have already seen this property in the previous Se
tion in the
language of rst quantization. More generally, the Eu
lidean average (1.2.6) is
asso
iated with the va
uum expe
tation value of h0 j TF [' j 0i in Minkowski
spa
e.
and
2 + m2 h ' (x) ' (y) i + '2 (x) ' (y) = (d) (x y) : (1.2.34)
2
Rederive Eq. (1.2.34) by analyzing Feynman diagrams.
Let us introdu
e the Fouriertransformed 2 and 3point Green fun
tions
G3 (p; q; p q ) =
G(p) =
dd x e ipx h '(x)'(0) i ;
(1.2.35)
(1.2.36)
G3 (p; q; p q ) =
(1.2.37)
(1.2.38)
It is standard to trun
ate three external legs, introdu
ing the vertex fun
tion
(p; q; p q ) = G3 (p; q; p q )G 1 (p)G 1 (q )G 1 (p + q )
(1.2.39)
3
(p; q; p q ) =
T
""bb = ""bb + "" Tbb +
s
:::
(1.2.41)
where the lled
ir
le on the LHS represents the exa
t vertex and the thin lines are
asso
iated with the bare propagator (1.2.37).
An analogous expansion of the propagator is
+
Problem 1.18
Solution
43
SECOND QUANTIZATION
+
+ ::: ;
(1.2.42)
44
CHAPTER 1.
PATH INTEGRALS
where the bold line represents the exa
t propagator. It is
ommonly rewritten as
an equation for the selfenergy G0 1 (p) G 1 (p) whi
h involves only the 1parti
le
irredu
ible (1PI) diagrams. The last diagram on the RHS of Eq. (1.2.42) is not 1PI.
Resumming the diagrams a
ording to the denition (1.2.41) of the exa
t vertex and the exa
t propagator G, the propagator equation
an be represented
graphi
ally as
G0 (p) G0 (p)G
1 (p)G
0 (p)
(1.2.43)
s
where the bold lines represent the exa
t propagator G while the external (thin) ones
are asso
iated with the bare propagator G0 . One vertex on the RHS of Eq. (1.2.43)
is exa
t and the other is bare. It does not matter whi
h one is exa
t and whi
h one
is bare sin
e we
an
olle
t the diagrams of Eq. (1.2.42) into the exa
t vertex either
on the left or on the right. Eq. (1.2.43)
an be written analyti
ally as
Z
dd q
G(q ) ( q; p; q p)G(p q ) : (1.2.44)
G0 1 (p) G 1 (p) =
2 (2 )d
Multiplying Eq. (1.2.44) by G(p) and using the denition (1.2.36), we obtain
exa
tly the Fourier transform of Eq. (1.2.34).
Noti
e that Eq. (1.2.34) is not
losed. It relates the 2point average (propagator) to the 3point average (whi
h is asso
iated with a vertex). The
losed
set of the S
hwinger{Dyson equations
an be obtained for the npoint averages
Gn (x1 ; x2 ; : : : ; xn ) = h ' (x1 ) ' (x2 ) : : : ' (xn ) i :
(1.2.45)
They are also
alled the
orrelators, in analogy with statisti
al me
hani
s, or the npoint Green fun
tions, in analogy with the Green fun
tions in
Minkowski spa
e.
Choosing
F [' = ' (x2 ) : : : ' (xn )
(1.2.46)
and
al
ulating the variational derivative, one gets from Eq. (1.2.28) the following
hain of equations
2 + m2 h ' (x) ' (x2 ) : : : ' (xn ) i
n
X
j =2
(1.2.47)
1.2.
45
SECOND QUANTIZATION
where ' (xj ) means that the
orresponding term ' (xj ) is missing in the
produ
t. Using the notation (1.2.45), Eq. (1.2.47)
an be rewritten as
2 + m2 Gn (x; x2 ; : : : ; xn ) + Gn+1 (x; x; x2 ; : : : ; xn )
2
n
X
j =2
(d) (x xj ) Gn 2 (x2 ; : : : ; xj ; : : : ; xn )
(1.2.48)
Z [J =
d
D' (x) e S+ d xJ (x)'(x);
(1.2.49)
and varying with respe
t to the sour
e. The npoint fun
tions (1.2.45) are
then given by
Gn (x1 ; : : : ; xn ) =
Z [J
Z [J J (x1 ) J (xn )
(1.2.50)
ln Z [J :
J (x1 ) J (xn )
(1.2.51)
W [J = ln Z [J
(1.2.52)
involves only a set of
onne
ted diagrams while dis
onne
ted ones emerge
in Z [J after the exponentiation. We already tou
hed this property in Subse
t. 1.2.3 to order 2 . The fun
tional W [J is
alled, for this reason, the
generating fun
tional for
onne
ted diagrams.
46
CHAPTER 1.
PATH INTEGRALS
q1 + : : : + q n = p 1 + : : : + p k :
(1.2.53)
An!k (q1 ; : : : ; qn ; p1 ; : : : ; pk )
=
n
Y
k
Y
Z
dd pi nY1 dd qj
d
d
j =1 (2 )
i=1 (2 )
(1.2.54)
The unusual sign of the square of the parti
le mass m is due to the Eu
lidean
metri
.
Equation (1.2.54) makes sense for timelike pi , when p2i < 0, while qj2 is
arbitrary. The amplitude for the
ase of onmassshell in
oming parti
les is
given by Eq. (1.2.54) with qj2 ! m2 .
1.2.7 Regularization
The ultraviolet divergen
ies (i.e. those at small distan
es or large momenta)
are an intrinsi
property of QFT whi
h makes it dierent from the quantum
me
hani
s of a nite number of degrees of freedom. The divergen
ies emerge,
roughly speaking, be
ause of the deltafun
tion in the
anoni
al
ommutation
relations.
The idea of regularization is to somehow smooth the ee
t of the deltafun
tion. The usual pro
edures of regularization are to
47
SECOND QUANTIZATION
(d)(x y) Reg.
=) R (d)(x y) = R(x; y) :
(1.2.55)
R = e a2 ( 2
(1.2.56)
Z k
Y
lim p2 + m2
2 ! m2 i
p
i
i=1
j =1
P
P
i ik=1 pi xi +i nj=11 qj xk+j
e
Gn (x1 ; : : : ; xn+k 1 ; 0)
qj2 + m2
1.2.
m2 ) ;
where the parameter a with the dimension of length plays the role of an
ultraviolet
uto. The
uto disappears as a ! 0 when
R ! 1;
R (x; y) ! (d) (x y) :
(1.2.57)
It is easy to
al
ulate how the regularization (1.2.55) modies the propagator. The result reads
1
R (d) (x y)
GR (x y) =
2 + m2
Z1
(x y)2
2
1
1
1
:
(1.2.58)
=
d e 2 m e 2
2
(2 )d=2
2
a
The lower limit in the integral over the proper time is now a2 rather than
0 as in the nonregularized expression (1.1.85). This parti
ular way of point
splitting
oin
ides with the S
hwinger propertime regularization.
A regularization via point splitting
an be done nonperturbatively while
the dimensional regularization, whi
h is listed in item 4, is dened only in the
framework of perturbation theory. The regularization of the measure, whi
h
is listed in item 5, will be
onsidered in the next Se
tion.
When a regularization is introdu
ed, some of the rst prin
iples (
alled
the axioms), on whi
h quantum eld theory is built on, are violated. For
instan
e, the regularization via the point splitting (1.2.55) violates lo
ality.
7 That is in the quadrati
inelds part of the a
tion.
48
CHAPTER 1.
PATH INTEGRALS
Z =
DA
D D e S[A; ;
(1.3.1)
opposite sign (i.e. is substituted by i ) whi
h results in a multipli
ation of the fermioni
part of the a
tion (1.3.4) by an extra fa
tor of i.
1.3.
49
S [A; ; =
1 2
dd x
r + m + F
;
4
(1.3.4)
where
r
=
ieA (x)
(1.3.5)
F = A
A
(1.3.6)
is the eld strength, and
are the Eu
lidean
matri
es whi
h are dis
ussed
in Subse
t. 1.1.2.
e i(x)
5 (x) ;
(x) e i(x)
5 :
(1.3.7)
5 =
1
2
3
4
(1.3.8)
and have the same transformation law sin
e in
5y =
5 ;
0y =
0
(1.3.9)
S =
(1.3.10)
50
CHAPTER 1.
PATH INTEGRALS
J5 = i
5
(1.3.11)
is the Noether
urrent asso
iated with the
hiral transformation.
It follows from Eq. (1.3.10) that the divergen
e of the axial
urrent (1.3.11)
reads
J5 = 2im
5 ;
(1.3.12)
so that it is
onserved in the massless
ase (m = 0) at the
lassi
al level:
J5 m==0
Problem 1.19
0:
(1.3.13)
rb + m (x) = 0 ;
rb = r :
(1.3.14)
1.3.
where sp is the tra
e only over the
matrix indi
es i and j . The RHS vanishes
naively sin
e the tra
e vanishes. A subtlety with the appearan
e of the innite
fa
tor (d) (0) will be dis
ussed in the next Subse
tion.
Note that the innitesimal version the transformation (1.3.7) is a parti
ular
ase of the more general one
(x) ! 0 (x) = (x) + (x) ;
(x) ! 0 (x) = (x) + (x) ;
(1.3.19)
whi
h is an analog of the transformation (1.2.25) and leaves the measure
invariant. The
al
ulation given in Eq. (1.3.18) is an expli
it illustration of
this fa
t.
The general transformation (1.3.19) leads, when applied to the path integral in Eq. (1.3.1), to the S
hwinger{Dyson equations
rb + m
Solution
(x)
rb m
= 0
(1.3.15)
r = +ieA(x) :
(1.3.16)
with
Let us now dis
uss how the measure in the path integral
hanges under
the
hiral transformation (1.3.7). The old and new measures are related by
h
i
D D = D 0 D 0 det e 2i(x)
5 (d) (x y) ;
(1.3.17)
where the determinant is over the spa
e indi
es x and y as well as over the
matrix indi
es i and j . Noti
e that the determinant, whi
h is nothing but the
Ja
obian of the transformation (1.3.7), emerges for the Grassmann variables
to the positive rather than negative power as for
ommuting variables. This is
a known property of the integrals (1.2.17) over Grassmann variables [Ber86
whi
h look more like derivatives.
The logarithm of the Ja
obian in Eq. (1.3.17)
an be
al
ulated as
h
= Sp ln e 2i
5
51
(x)
;
(x)
w.s.
;
=
(x)
(x) w.s.
=
r m
b
(1.3.20)
whi
h hold in the weak sense, i.e. under the sign of averaging over and .
More restri
tive transformations of the type of (1.3.7), whi
h are asso
iated with symmetries of the
lassi
al a
tion and result in
onserved
urrents,
lead to some (less restri
tive) relations between
orrelators whi
h are
alled
Ward identities. This terminology goes ba
k to the fties when a proper relation between the 2 and 3point Green fun
tions was rst derived for the
gauge symmetry in QED.
The simplest Ward identity, whi
h is asso
iated with the
hiral transformation (1.3.7), reads
i (d) (y)
i (x)
5 (0) : (1.3.21)
j
It is
lear from the way Eq. (1.3.21) was derived, that it is always satised as
a
onsequen
e of the quantum equations of motion (1.3.20).
Derive Eq. (1.3.21) in the operator formalism when the averages
are substituted by the va
uum expe
tation values of the T{produ
ts.
Solution Use Eq. (1.3.14) (whi
h a
quires an extra
i in Minkowski spa
e, where
the spatial
matri
es are antiHermitean rather than Hermitean as in Eu
lidean
Problem 1.20
52
CHAPTER 1.
PATH INTEGRALS
1.3.
53
spa
e, and holds in the quantum
ase in the weak sense, i.e. when applied to a
state) and
anoni
al equaltime anti
ommutation relations for and with the
only nonvanishing anti
ommutator being
(x0 y0 ) i (y ); j (x) = ij (d) (x y ) :
(1.3.22)
Remark on 5 in d dimensions
5 = 1 2 d
(1.3.23)
2d=2 2d=2.
D D
(d)(0)
sp 5 =
1 0:
(1.3.24)
in in (x) ;
i (x) =
(1.3.25)
where there is no summation over the spinor index i. Here
in and
in are
Grassmann variables. The measure is then similar to that of Eq. (1.1.79) and
1 Y
Y
n=1 i
(D )R (D )R =
d in
M Y
Y
n=1 i
1 Y
Y
m=1 j
d in
d jm :
M Y
Y
m=1 j
d jm :
(1.3.26)
(1.3.27)
(1.3.29)
and Eq. (1.2.20), we rewrite the determinant on the RHS of Eq. (1.3.28) for
an innitesimal parameter as
det
Z
kj
dd x kn y (x) e 2i(x)
5 jm (x)
= 1 + 2i
M Z
X
n=1
where the spinor indi
es are
ontra
ted in the usual way.
It is easy to see how this formula re
overs Eq. (1.3.18) sin
e
1
X
in (x)jny (y) = (d) (x y) ij
n=1
(1.3.30)
(1.3.31)
54
CHAPTER 1.
PATH INTEGRALS
in the nonregularized
ase due to the
ompleteness of the basis fun
tions.
In the regularized
ase, the sum over n on the LHS of Eq. (1.3.31) is
restri
ted by M from above so that the RHS is no longer equal to the deltafun
tion. We substitute
M
X
n=1
(1.3.32)
with the RHS being the matrix element of some regularizing operator R.
It
an be
hosen in many ways. We shall work with several forms:
R = e a2 rb 2 ;
1
;
R =
b2
1 a2 r
1
R =
;
b
1 + ar
or
(1.3.33)
or
(1.3.34)
::: ;
(1.3.35)
dd x (x) J5 = 2i Sp
5 R
= 2i
dd x (x) sp 5 R (x; x) :
(1.3.36)
Sp OR ;
(1.3.37)
1.3.
55
M
X
(1.3.38)
k:
n=1
k
n
This denition extends the standard mathemati
al one11 to the
ase of spinor
indi
es. The sum over k is in
luded in order for the regularized variational
derivative to feel variations of all the spinor
omponents of
n when the variation is not diagonal in the spinor indi
es.
When applied to
j
R (y )
i
R (x)
it yields
or, equivalently,
i
R (x)
j
R (y )
M
X
n=1
jny (y)
M
X
in in (x) ;
(1.3.39)
(1.3.40)
n=1
ij (d) (x y) Reg.
=) Rij (x; y) :
(1.3.41)
whi
h is a fermioni
analog of Eq. (1.2.55).
Thus, we
on
lude that the regularization of the measure in the path
integral is equivalent to smearing the deltafun
tion in
ommutator terms.
rb + m (x) w.s.
=
dd yR(x; y) ;
(y)
Z
w.s.
(x) r
b
m
=
dd yR(x; y)
:
(1.3.42)
(y)
11 See,
e.g
56
CHAPTER 1.
PATH INTEGRALS
1.3.
These equations are understood again in the weak sense, i.e. under the sign
of averaging over and and obviously reprodu
e Eq. (1.3.20) as a ! 0.
q1
57
k q1
Problem 1.21
J5 w.s.
= i
dd y
A
Aq A
AK q + k
AA
dd yR(x; y )
(y )
(1.3.43)
whi h is equivalent to Eq. (1.3.36) sin e the (x) there is an arbitrary fun tion.
Fig. 1.6: Triangular diagram asso
iated with
hiral anomaly in d = 4. The solid
lines
orrespond to R0 given by Eq. (1.3.51). The wavy lines
orrespond
to the eld strength.
4 :
(1.3.46)
Expanding in e:
R = R0 + R0 (: : :) R0 + : : :
(1.3.47)
with
R0 = 1 1a2 2 ;
(1.3.48)
eF
R0 e2F R0
2
e2
d4 x (x)
F F~ ;
162
Sp 5 R = a4 Sp 5 R0
(1.3.49)
where
(1.3.50)
R0 (p) =
1
:
1 + a2 p2
(1.3.51)
The wavy lines
orrespond to the eld strength. The lower vertex is asso
iated
with
5 .
The integral over the fourmomentum q, whi
h
ir
ulates along the triangular loop,
an be easily
al
ulated by introdu
ing ! = aq and transforming
58
CHAPTER 1.
the integral as
d4 q f (q)
PATH INTEGRALS
Z
! a14 d4 ! f !a :
1.3.
(1.3.52)
Noti
e that the integral involves a 4 whi
h
an
els a4
oming from the expansion in e whose proper term is given by the intermediate expression in
Eq. (1.3.49). Therefore, the result is nonvanishing and a independent as
a ! 0. Higher terms of the expansion in e are proportional to higher powers
in a and vanish as a ! 0.
From Eqs. (1.3.36) and (1.3.49), we get nally
ie2
(1.3.53)
F F~ :
82
The anomaly on the RHS is known as the Adler{Bell{Ja
kiw anomaly. Its
appearan
e is usually related to the fa
t that any regularization
an not be
simultaneously gauge and
hiral invariant.
J5 =
Problem 1.22 Cal
ulate the
oe
ient in Eq. (1.3.49) and show that it is regulator
independent.
Solution The
ontribution of the triangular diagram of Fig. 1.6, whi
h represents
the intermediate expression in Eq. (1.3.49), reads expli
itly
4e2 a4
2 Sp 5 R =
d4 x
d4 y
2e2
16 2
(1.3.54)
d4 q1
F (q )F~ (k q1 )
(2 )4 1
Z
4
(2dq)4 (1 + a2 q2 ) (1 + a2 (q +1q1 )2 ) (1 + a2 (q + k)2 ) (1.3.52)
=
Z1 2 2
Z
Z
d4 k ikx d4 q1
! d!
4
d x(x)
e
F (q )F~ (k q1 )
(1.3.55)
(2 )4
(2 )4 1
(1 + ! 2 )3
2e2 a4
d4 x(x)
d4 k e ikx
(2 )4
r(a2 p2 )
R0 (p) :
q+k
r
5
Fig. 1.7: The diagram asso
iated with
hiral anomaly in d = 2. The solid lines
orrespond to R0 given by Eq. (1.3.51). The wavy line
orresponds to the
eld strength.
Then the only dieren
e with Eq. (1.3.55) is that the last integral over ! 2 is repla
ed
by
Z1
! 2 d! 2 r00 (! 2 ) = r(0) = 1
(1.3.57)
0
for reasonable fun tions r whi h look like those given by Eqs. (1.3.33) { (1.3.35).
d4 z
59
(1.3.56)
Solution
2iea2
d2 x
60
Z
2
2iea
CHAPTER 1.
PATH INTEGRALS
d2 k e ikx F (k)
d2 q
1
2
(2 )
(2 )2 (1 + a2 q 2 ) (1 + a2 (q + k)2 )
Z1
Z
Z
(1.3.59) ie
d! 2
d2 k e ikx F (k)
= 2
d2 x(x)
2
4
(2 )
(1 + ! 2 )2
d2 x(x)
d2 x(x)
ieF (x)
:
2
(1.3.60)
(1.3.61)
in d = 2.
The result is again regulator independent sin
e the integral over ! is repla
ed
for an arbitrary R0 (p) by
Z1
d! 2 r0 (! 2 ) = r(0) = 1
(1.3.62)
0
1.3.
61
(1.3.65)
(1.3.66)
The index l' is
alled the s
ale dimension of the eld '. The value of l' in
a free theory is
alled the
anoni
al dimension, whi
h equals (d 2)=2 for
bosons (s
alar or ve
tor elds) and (d 1)=2 for the spinor Dira
eld, i.e. 1
and 3=2 in d = 4, respe
tively. Sometimes l' is
alled, for histori
al reasons,
the anomalous dimension. More often the term \anomalous dimension" is
used for the dieren
e between l' and the
anoni
al value.
The proper Noether
urrent, whi
h is
alled the dilatation
urrent, is
expressed via the energymomentum tensor as
D = x
(1.3.67)
so that its divergen
e equals the tra
e of the energymomentum tensor over
the spatial indi
es:
D = ;
ig2 X a ~ a
F F :
162 a
=
m
(1.3.69)
S =
(1.3.64)
The
oe
ient in this formula diers by 1=2 from that in Eq. (1.3.53) whi
h
is due to dierent normalizations in the Abelian and nonAbelian
ases. This
will be explained in more detail in Subse
t. 2.1.1 below.
The d = 2 anomaly (1.3.58) exists for the singlet axial
urrent only in the
Abelian
ase.
(1.3.68)
sin
e the energymomentum tensor is
onserved. For the a
tion (1.3.4) one
gets
(1.3.63)
Here f ab
are the stru
ture
onstants of the gauge group and g is the
oupling
onstant. The nonAbelian analog of Eq. (1.3.53) for the axial
urrent, whi
h
is a singlet w.r.t. the gauge group, reads12
J5 =
dd x (x) D (x)
(1.3.70)
or
D (x) =
S
:
(x)
(1.3.71)
In the massless
ase, m = 0, the RHS of Eq. (1.3.69) vanishes and the
dilatation
urrent is
onserved. This is a wellknown property of ele
trodynami
s with massless ele
tron whi
h is s
ale invariant at the
lassi
al level. A
62
CHAPTER 1.
PATH INTEGRALS
generi
s
aleinvariant theory does not depend on parameters of the dimension of mass or length. This usual dimension is to be distinguished from the
s
ale dimension whi
h is dened by Eq. (1.3.66). The dimensional parameters
do not
hange under the s
ale transformation (1.3.65).
In the quantum
ase, the s
ale invarian
e is broken by the (dimensional)
uto a. The energymomentum tensor is no longer tra
eless due to loop
ee
ts. The relation (1.3.68) holds in the quantum
ase in the weak sense,
i.e. for the averages
D F [A; ; = F [A; ; ;
(1.3.72)
where F [A; ; is a gaugeinvariant fun
tional of A, and .
For a renormalizable theory like QED, the RHS of Eq. (1.3.72) is proportional to the GellMannLow fun
tion B(e2) whi
h is dened by
a de2
(1.3.73)
= B e2 :
da
A nontrivial property of a renormalizable theory is that the RHS in this
formula is a fun
tion solely of e2  the bare
harge.
The meaning of the renormalizability is very simple: physi
al quantities do
not depend on the
uto a, provided the bare
harge e is
hosen to be
uto
dependent a
ording to Eq. (1.3.73). This dependen
e of e on a ee
tively
a
ounts for the distan
es smaller than a, whi
h are ex
luded from the theory.
The pre
ise relation between the tra
e of the energymomentum tensor
and the GellMann{Low fun
tion reads
2
w.s. B e
F2 ;
(1.3.74)
=
4e2
where the equality is understood again in the weak sense. This formula was
rst obtained in Refs. [Cre72, CE72 to leading order in e2 and proven in
Ref. [ACD77 to all orders in e2 .
Note that this formula holds in the operator formalism only when applied
to a gauge invariant state. The reason is that otherwise a
ontribution from
a gaugexing term in the a
tion would be essential. It does not
ontribute,
however, to gaugeinvariant averages whi
h
an be formally proven using the
gauge Ward identity.
Prove the relation (1.3.74).
Let us absorb the
oupling e into A introdu
ing
A = ieA ;
F = ieF :
Problem 1.24
Solution
1.3.
63
1 2
F :
4e2
L = b Ab
(1.3.76)
To prove Eq. (1.3.74), let us use the
hain of Eqs. (1.3.68) and (1.3.71). It is
ru
ial that in the absen
e of other dimensional parameters the derivative =
an
be repla
ed by =a sin
e all dimensionless quantities depend in a theory with a
uto only on the ratios of the type x=a.13 Sin
e the dependen
e on the
uto a
2 , Eq. (1.3.74)
an be
enters in Eq. (1.3.76) formally only via e 2 in front of F
heuristi
ally proven by rst dierentiating with respe
t to a and then expressing
the result via F again. We have used herein the fa
t that F is invariant under
the renormalizationgroup transformation and, therefore, does not depend on a.
D (x) =
sp R(x; x) :
(1.3.77)
Problem 1.25
1
2 + 1
$
F
r
4
4
$
+
r
(1.3.78)
(1.3.79)
=
=
m +
m
Z
dd yR(x; y )
(x)
2
(y )
sp R (x; x) ;
(x)
dd yR(x; y )
(y )
(1.3.80)
13 This is the reason why the Callan{Symanzik equations, whi h are nothing but the
64
CHAPTER 1.
q1
q1
k q1
PATH INTEGRALS
k q1
q+k
r
a)
b)
Fig. 1.8: The diagrams whi
h
ontribute to the s
ale anomaly in d = 4. The wavy
line
orresponds to the eld strength.
+
*
1
x
sp x
1 + a2 i b 2
*
1
sp x
1 + a2 (i + eA )2
+
1 a2 e F x
(1.3.81)
e2 a2 A2 ;
1 2
2 ea F :
65
sp R(x; x) =
q + q1
A
Aq A
AK q + k
A
A
1.3.
2 (x)
e2 F
:
482
(1.3.82)
w.s.
B g2
4g2
a F~ a :
F
(1.3.83)
66
CHAPTER 1.
PATH INTEGRALS
1.4.
2 2
2
x
4
1 2 2 1 4 4
=
x + x + :
(1.4.1)
2
4
4
This is nothing but an anharmoni
os
illator with a wrong sign for the
oe
ient of the quadrati
term,14 whi
h usually appears with a positive
oe
ient
!2 =2. We have introdu
ed
V (x) =
2 =
!2
(1.4.2)
in order to work with real numbered values. The
onstant term is added for
later
onvenien
e. The potential (1.4.1) as a fun
tion x is depi
ted in Fig. 1.9.
The (Eu
lidean) a
tion is dened by
Z
1 2
S [x =
d x_ ( ) + V (x( ))
2
(1.4.3)
14 It is often alled the mass term. This terminology omes from quantum eld theory,
where the potential (1.4.1) is
onsidered in the
ontext of a spontaneous breaking of the
re
e
tion symmetry x ! x. In our quantumme
hani
al problem, dened by the Eu
lidean a
tion (1.4.3), the mass of the nonrelativisti
parti
le is absorbed in whi
h has,
therefore, the dimension of [length2 . This is explained already in Subse
t. 1.1.6.
67
6V (x)
+ p
p
x
Fig. 1.9: The doublewell potential (1.4.1). The short verti al lines represent the
position of the minima (1.4.4). The dashed lines
orrespond to the energy
E0 of the lowest state in a single well, i.e. to that in the limit ! 0.
with V (x) given by Eq. (1.4.1). The plussign between the kineti
and potential energies is be
ause we are in Eu
lidean spa
e.
It follows from Eqs. (1.4.1) and (1.4.3) that the parameter has the
dimension of [length 2 or, in other words, the dimensions of x and are
[ 1=2 and [ 1 , respe
tively. Analogously, the dimension of the
onstant
is [3 .
For 3 , the potential (1.4.1) has super
ially two degenerate va
ua
(1.4.4)
x0 = p ;
whose positions
oin
ide with the minima of the potential in Fig. 1.9.
The degenera
y between the two minima is preserved at all orders of perturbation theory, where an expansion near one of the minima of the potential
(either the left or right one) is
arried out:
(1.4.5)
x ( ) = p + ( )
p
with ( ) = . The
orrelator at asymptoti
ally large is
2
hx (0) x ( )i !
+::: :
(1.4.6)
Its nonvanishing value means that a parti
le is lo
alized at one of the two
va
ua.
68
CHAPTER 1.
PATH INTEGRALS
1.4.
The next terms of the perturbative expansion in do not spoil this result
sin
e the potential (1.4.1) be
omes
and
V = 2 2 3 + 4
(1.4.7)
4
after the shift (1.4.5), and has the positive sign of the quadrati
term. Therefore, perturbation theory around the va
uum x
0 is a usual one, and the
parti
le lives perturbatively in one of the two va
ua.
However, we know from quantum me
hani
s that (nonperturbatively)
hx (0) x ( )i
jxn0 j2 e
(En E0 )
(1.4.8)
hx (0) x ( )i
for large , where
r
48 2 2 3
E =
exp
3
E
2 2 3
3
(1.4.9)
!
(1.4.10)
Derive Eq. (1.4.10) modulo a
onstant fa
tor within standard quantum me
hani
s.
Solution Let us use the semi
lassi
al formula [LL74 (Problem 3 in x50)
Problem 1.26
+a
2
a dx 2(V (x) E0 ) ;
e
where a are the
lassi
al turning points, whi
h are determined by
V (a) = E0 ;
E0 =
h =
p 3;
2
2
z =
69
(1.4.13)
x;
(1.4.14)
1 h
Z
1+h
p
dz (1 z 2 )2 4h2 = 2 2 + ln h + O(1) :
3h
(1.4.15)
Substituting into Eq. (1.4.11), one re overs Eq. (1.4.10) modulo a onstant fa tor.
is the energy splitting between the two lowest states (symmetri
and antisymmetri
) for 3 whi
h exponentially vanishes as ! 0.
The appearan
e of imaginary time in Eq. (1.4.8) is be
ause under a barrier
parti
les live in imaginary time. We may say that imaginary time is an
adequate language to des
ribe a tunneling through a barrier.
Sin
e the RHS of Eq. (1.4.9) vanishes as ! 1, the re
e
tion symmetry x ! x, whi
h is broken in perturbation theory, is restored nonperturbatively as ! 1.
E =
(1.4.11)
(1.4.12)
h x (0) x ( ) i
(1.4.16)
2
+::: :
(1.4.17)
Noti
e that x(0) and x( ) in the integrand
an be substituted by the saddlepoint values after whi
h the integral over Gaussian
u
tuations
an
els by
the same one in the denominator. In other words we have reprodu
ed that
ea
h of the trivial minima (1.4.4) results in Eq. (1.4.6).
hx (0) x ( )i
70
CHAPTER 1.
PATH INTEGRALS
1.4.
71
An
p analogous solution whi
h interpolates between = at = 1 and
= at = +1 is
alled an antiinstanton. It diers from Eq. (1.4.19)
by an overall minus sign:
6
xainst (
0
p
0 ) =
tanh
(
p 0 ) ;
2
(1.4.20)
p Find all solutions pof Eq. (1.4.18) with the boundary
onditions
1) = = and x(+1) = = .
Problem 1.27
x(
Equation (1.4.18) looks like the Newton equation for a
lassi
al parti
le,
whose mass is equal to unity, in the upsidedown potential V (x) (its shape
an
be obtained from the one depi
ted in Fig. 1.9 by re
e
ting with respe
t to the
horizontal axis V = 0). The rst integral of motion is the energy
Solution
p
+ p
x
Fig. 1.10: Graphi
representation of the onekink solution (1.4.19) as a fun
tion of
.
Minima of the a tion (1.4.3) an be alternatively obtained from the lassi al equation of motion
x 2 x + x3 = 0 :
(1.4.18)
x2
4
1
= x_ 2
2
2
2
(1.4.21)
x_ =
2(E + V (x));
(1.4.22)
x_ =
2
2
x2
(1.4.23)
whi
h results after integration in Eq. (1.4.19) with 0 being the integration
onstant.
It is evident that the solution is unique.
2 2 3
;
S [xinst =
3
(1.4.24)
whi
h
oin
ides with the (minus) exponent in Eq. (1.4.10) for the energy
splitting E .
72
CHAPTER 1.
PATH INTEGRALS
h x (0) x ( ) i
2 4
2 2 3
=
1 C
exp
3
2 2 3
3
1.4.
The
u
tuations around the instanton are Gaussian ex
ept for one mode, whi
h
is asso
iated with a translation of the instanton
enter, 0 . This zero mode is given
by
0 ( )
d2
d 2
2 + 3x2
h x (0) x ( ) i
+ e S [xinst
D( ) e
e (S [x S [xinst)
d ( 12 _ 2 +2 2 )
(1.4.26)
where the path integral in the numerator is over
u
tuations around the instanton
solution (1.4.19). The normalizing fa
tor in the denominator is asso
iated with
averaging over the Gaussian
u
tuations around the trivial minima (1.4.4) whose
potential energy is des
ribed by the quadrati
term in Eq. (1.4.7). There are two
su
h trivial minima (x+ and x ) and two onekink minima (instanton and antiinstanton) so these fa
tors of 2
an
el.
Keeping the quadrati
term in the expansion around the instanton:
x( ) = xinst (
one gets
1
S [x S [xinst =
2
d
0 ) + (
_ 2
0 ) ;
(1.4.27)
2 2 + 3x2
inst
2
x_ inst = 0
+1
Z
1 =
(1.4.30)
d (u[x )
(1.4.31)
in the path integral in the numerator on the RHS of Eq. (1.4.26). Here u[x is
determined by the equation
+1
Z
1
2
(1.4.29)
Problem 1.28
2
=
R
Dx( ) x(0)x( )
inst
/ x_ inst ( ) :
!3
5 ; (1.4.25)
73
d y (
u[x) x( ) = 0
(1.4.32)
with
y ( ) =
hR
+1
x_ ( )
2
1 dt x_ (t)
i1=2
(1.4.33)
! 0 = 0 ;
(1.4.34)
one gets
x( )
! x( 0) = x( 0 ) :
(1.4.35)
This leaves the measure and the a tion in the path integral (1.4.26) invariant, while
(1.4.28)
u[x
! u[x + 0 :
(1.4.36)
74
CHAPTER 1.
PATH INTEGRALS
2 2 3 D( )
3
Z +1
1
2
d yinst ( )( ) e
d (_ 2 2 2 +3x2inst 2 )
1
2
d (_ 2 2 2 +3x2inst 2 )
:(1.4.42)
1
p
Noti
e the appearan
e of the fa
tor S [xinst .
We have thus obtained Eq. (1.4.25) with
R
R
R
2 2 2
2
2
1
D( ) +11 d yinst ( )( ) e 2 d (_ +3xinst )
R
: (1.4.43)
C = 2
R
1 2 2 2
D( ) e d ( 2 _ + )
1.4.
75
Problem 1.29
Solution
Noting that
x2inst ( ) = 2 1
1 ;
osh2 z
(1.4.44)
(1.4.45)
R! [v
det
D2 + ! 2 + v (z )
det [ D2 + ! 2
(1.4.48)
v (z ) =
let us note that
6
;
osh2 z
(1.4.49)
i
h
1
1
ln R [v = Sp
Sp
!
2
2
2
2
2
!
D + ! + v (z )
D +!
+
1
Z
dz R! (z; z ; v ) 1 ;
=
(1.4.50)
2!
1
where the diagonal resolvent R! (z; z ; v ) is dened by Eq. (1.1.119) with G = 1 and
V v . The term 1=2! on the RHS, whi
h equals the diagonal resolvent in the
free
ase when v = 0 (see Eq. (1.1.38)),
omes from the free determinant in the
denominator on the RHS of Eq. (1.4.48).
76
CHAPTER 1.
PATH INTEGRALS
A
ru
ial observation is that the diagonal resolvent for the potential (1.4.49) is
given by the simple formula
R! (z; z ; v ) =
1
2!
v (z )
v 2 (z )
+
;
2
2
4! (! 1) 8! (! 1)(! 2 4)
(1.4.51)
whi
h
an easily be veried by substituting into the Gelfand{Di
key equation (1.1.123) with G = 1. The reason is that the potential (1.4.49) is integrable
and possesses two bound states (see, e.g., x23 of Ref. [LL74).
Cal
ulating the integral over z on the RHS of Eq. (1.4.50), by the formulas
+1
+1
Z
Z
dz = 4 ;
dz = 2 ;
(1.4.52)
2
3
osh z
osh4 z
1
1
we obtain
ln R [v = 1 1 + 2 ;
!
! 2
! !2 1 !2 4
whi
h is easily integrated over ! to give
det
(1.4.53)
D2 + ! 2 6=
osh2 z
(! 2)(! 1)
=
:
det [ D2 + ! 2
(! + 2)(! + 1)
(1.4.54)
!!1
R! [v = 1 :
(1.4.55)
C = 2B =
48
(1.4.56)
be approximately
onstru
ted from several single kinks and antikinks, whi
h
are separated along the axis by the some distan
e R 1=, sin
e the
intera
tion between kinks would be exp f Rg. Su
h a
onguration is
1.4.
p
+=
p
77
=
Fig. 1.11: The manykink onguration xMkink ( ) whi h is ombined from the solution (1.4.19).
depi
ted in Fig. 1.11 for the
ase when the number of kinks is equal to the
number of antikinks. An analogous
onguration with the number
p of kinks
p
one more than the number of antikinks
onne
ts the = and =
va
ua.
It is not an exa
t solution of Eq. (1.4.18) sin
e the kink and the antikink
attra
t and have a tenden
y to annihilate. However, it is an approximate
solution as ! 0.
Analyti
ally, the M kink
onguration
an be represented as
xMkink ( ) =
p
M
Y
i=1
sign(
i ) ;
(1.4.57)
where i are the
enters of the instantons (or antiinstantons), of whi
h the
M kink
onguration is built out, and
1 2 : : : M :
(1.4.58)
Equation (1.4.57) assumes that the kinks do not intera
t and are innitely
thin as ! 0. The a
tion of the
onguration (1.4.57) is therefore given by
2 23
M;
3
i.e. it equals M times the a
tion for the one kink.
Summing over manykink
ongurations, one gets [Pol77
S [xMkink =
(1.4.59)
2 E
e
;
(1.4.60)
where E is given by Eq. (1.4.10). The x ! x symmetry is now restored
as ! 1. This restoration is produ
ed by instantons =
lassi
al traje
tories
with a nite (Eu
lidean) a
tion.
h x (0) x ( ) i
78
CHAPTER 1.
PATH INTEGRALS
Fig. 1.12: Graphi
representation of a periodi
potential.
Obtain the exponentiation of the onekink
ontribution (1.4.25)
after summing over the M kink
ongurations (1.4.57) in the dilute gas approximation when the intera
tion between kinks is disregarded.
Solution The
al
ulation of the
ontribution of the M kink
onguration (1.4.57)
to the path integral is quite analogous to that for the one kink whi
h is des
ribed
in Problem 1.28. One gets
Z
Z 1
Z M 1
1
2 X
h x(0)x( ) i = ( E )M d1 d2 : : :
dM ; (1.4.61)
0
0
0
M =0
Problem 1.30
whi
h reprodu
es Eq. (1.4.60) by noting that the ordered integral is equal to
Z
0
d1
Z 1
0
d2 : : :
Z M 1
0
dM =
M :
M!
(1.4.62)
This
al
ulation is very similar to the one in statisti
al me
hani
s for the exponentiation of a singleparti
le
ontribution to the partition fun
tion in the
ase of
an ideal gas.
x(1) = x(0)
in perturbation theory ;
x(1) = x(0) + n
(1.4.63)
(1.4.64)
79
where k (M ) is the kth homotopy group whose elements are
lasses of
ontinuous maps of the ksphere S k onto M . Eq. (1.4.64) says that an (integer)
winding number n 2 Z is asso
iated with the mapping S 1 ! S 1 , whi
h
ounts how many times the target is
overed.
We see the dieren
e between the M kink
onguration for the doublewell potential and the multiinstanton solution for the periodi
potential.
The former was not an exa
t solution of the
lassi
al eld equation (1.4.18).
Only a single instanton or antiinstanton was a solution that
onne
ts the
two va
ua. This is why we need a periodi
potential for the multiinstanton
solution to exist due to the topologi
al argument.
The value of n in the boundary
ondition (1.4.63) is
alled the topologi
al
harge of the instantons, while n < 0 is asso
iated with antiinstantons. The
va
uum states are labelled by n: jni. The ninstanton
onguration
onne
ts
the jmi and hm + nj states. Therefore, instantons are asso
iated in Minkowski
spa
e with the pro
ess of tunneling between topologi
ally distin
t va
ua16
rather than with real parti
les. For this reason, they are sometimes
alled
pseudoparti
les in Eu
lidean spa
e.
It is
onvenient to
onsider another representation of va
uum states
1
X
(1.4.65)
ji =
e in j ni ;
n= 1
whi
h are
alled the va
ua. The va
ua are orthogonal
1 X
1
X
0
h j 0 i =
e i(n m) hm j ni = 2 ( 0 ) ; (1.4.66)
m= 1 n= 1
where 2 is a periodi
deltafun
tion with period 2. Here we used the
orthogonality of the nstates:
hm j ni
1 (S 1 ) = Z ;
15 See, e.g., the book [DNF86 (x17.5 of Part II).
1.4.
= mn :
(1.4.67)
16 The Minkowskispa
e interpretation of instantons is attributed to V.N. Gribov (unpublished). It is based on the fa
t that when the parti
le is lo
alized in one of the two wells
its momentum is indenite and
an sometimes be very large so that the proper energy is
above the barrier between the two wells. Su
h a parti
le jumps from the given well to the
other one. The
hara
teristi
time of this pro
ess is small in the typi
al units given by
. In other words this pro
ess is instantaneous that explains the term \instanton" whi
h
was introdu
ed by 't Hooft. The exponential suppression with of the oneinstanton
ontribution (1.4.25) represents quantitatively the fa
t that the probability of having large
momentum is small.
80
CHAPTER 1.
Z =
PATH INTEGRALS
Dx e S[x+i
d x_ ( )
d x_ ( ) = x (1) x (0)
(1.4.68)
(1.4.69)
whi h never shows up in perturbation theory. Therefore, the partition fun tion (1.4.68) an be alternatively represented as
Z =
n x(1)=x(0)+n
81
1.4.
Dx e S[x+in :
(1.4.70)
The se
ond term in the exponent in Eq. (1.4.68) is known as the term.
The parameter plays a role of a new fundamental
onstant whi
h does not
show up in perturbation theory. Amplitudes of physi
al pro
esses generated
by instantons may depend on the .
3 (SU(2)) = Z :
(1.4.72)
n =
g2
322
d4 x
3
X
a=1
a (x)F~ a (x) ;
F
(1.4.73)
whi
h equals one half of the non
onservation of the axial
harge given by the
Minkowskispa
e integral of the
hiral anomaly (1.3.64). This expression is
also known in topology as the Pontryagin index or the se
ond Chern
lass.
See, e.g., Refs. [Col77, VZNS82 for an introdu
tion to instantons in the
Yang{Mills theory.
82
CHAPTER 1.
PATH INTEGRALS
Referen
es to Chapter 1
[AB69
84
REFERENCES TO CHAPTER 1
[DNF86 B.A. Dubrovin, C.P. Novikov and A.T. Fomenko, Modern geometry. Methods and appli
ations, Nauka, M., 1986 (English transl.:
Springer, N.Y., Part I 1984, Part II 1985).
[Dys49 F.J. Dyson, The radiation theories of Tomonaga, S
hwinger, and
Feynman, Phys. Rev. 75 (1949) 486.
[Fey50 R.P. Feynman, Mathemati
al formulation of the quantum theory
of ele
tromagneti
intera
tion, Phys. Rev. 80 (1950) 440.
[Fey51 R.P. Feynman, An operator
al
ulus having appli
ations in quantum ele
trodynami
s, Phys. Rev. 84 (1951) 108.
[Fey72 R.P. Feynman, Statisti
al me
hani
s, Benjamin, Mass., 1972 (Russian transl.: Mir, M., 1975).
[FH65 R.P. Feynman and A.R. Hibbs, Quantum me
hani
s and path integrals, M
GrawHill, N.Y., 1965 (Russian transl.: Mir, M, 1968).
[Fuj79 K. Fujikawa, Path integral measure for gauge invariant fermion
theories, Phys. Rev. Lett. 42 (1979) 1195.
[GD75 I.M. Gelfand and L.A. Di
key, Asymptoti
s of development of the
Sturm{Liouville equations and the Korteveg{de Vries algebra, Usp.
Mat. Nauk 30 (1975) v.5, pp.67100 (English transl.: Russ. Math.
Surv. 30 (1975) v.5, 77).
[IZ80
C. Itzykson and J.B. Zuber, Quantum eld theory, M
GrawHill,
N.Y., 1980 (Russian transl.: Mir, M., 1984).
[Ka
59 M. Ka
, Probability and related topi
s in the physi
al s
ien
e, Inters
ien
e, N.Y., 1959 (Russian transl.: Mir, M., 1972).
[Kle95 H. Kleinert, Path integrals in quantum me
hani
s, statisti
s, and
polymer physi
s, 2nd Ed., World S
i., 1995.
[Lan67 J.S. Langer, Theory of the
ondensation point, Ann. Phys. 41
(1967) 108.
[Lev51 P. Levy, Probleems
on
rets d'analyse fon
tionnelle, Paris, 1951
(Russian transl.: Nauka, M., 1967).
[LL74
L.D. Landau and E.M. Lifshits, Quantum me
hani
s. Nonrelativisti
theory, Nauka, M., 1974 (English transl.: Pergamon, N.Y.,
1977).
[MM81 Yu.M. Makeenko and A.A. Migdal, Quantum
hromodynami
s as
dynami
s of loops, Nu
l. Phys. B188 (1981) 269 (Russian version:
Yad. Fiz. 32 (1980) 838).
[Mor86 A.Yu. Morozov, Anomalies in gauge theories, Usp. Fiz. Nauk 150
(1986) 337 (English transl.: Sov. Phys. Usp. 29 (1986) 993).
REFERENCES TO CHAPTER 1
[Nie77
85
86
REFERENCES TO CHAPTER 1
, Cat h22
J. Heller
Chapter 2
87
88
CHAPTER 2.
g.t.
0 (x) =
(x) (x) :
(2.1.3)
In analogy with QCD, the gauge group G = SU(N
) is usually asso
iated
with
olor while the proper index of is
alled the
olor index.
The gauge transformation (2.1.1) of the matter eld
an be
ompensated
by a transformation of the nonAbelian gauge eld A whi
h belongs to the
adjoint representation of G:
A (x)
g.t.
! A0 (x) =
(x) A (x)
y (x)
(x)
y (x) : (2.1.4)
89
(2.1.5)
Aa (x) =
(2.1.1)
Here
(x) 2 G with G being a semisimple Lie group whi
h is
alled the gauge
group (G = SU(3) for QCD). Equation (2.1.1) means that belongs to the
fundamental representation of G.
We restri
t ourselves to a unitary gauge group when
1 (x) =
y (x) ;
(2.1.2)
y (x) g.t.
!
2.1.
2
tr A (x) ta :
ig
(2.1.8)
Substituting
(x) = e i(x) ;
(2.1.9)
A (x) g.t.
= i radj
(x):
(2.1.10)
radj
[A ;
(2.1.11)
Here
is the
ovariant derivative in the adjoint representation of G while
rfun
A
(2.1.12)
radj
B (x)
= [rfun
; B (x)
(2.1.13)
90
CHAPTER 2.
A; ;
where
d4 x
(
A )
+m
1
2 ; (2.1.14)
tr F
2g2
F
or
91
= A A [A ; A
(2.1.15)
is the (antiHermitean) matrix of the nonAbelian eld strength.
The a
tion (2.1.14) is manifestly invariant under the lo
al gauge transformation (2.1.1), (2.1.4) sin
e
F (x) g.t.
!
(x) F (x)
y (x)
(2.1.16)
Solution
2.1.
In order to
ompare the phases of wave fun
tions at distin
t points, one needs
a nonAbelian extension of the parallel transporter whi
h was
onsidered in
Subse
t. 1.1.7. The proper extension of the Abelian formula (1.1.154) is
written as
R
dz A (z)
U [ yx = P e yx
:
(2.1.22)
While the matri
es A (z ) do not
ommute, the pathordered exponential
on the RHS of Eq. (2.1.22) is unambiguously dened by the general method of
Subse
t. 1.1.3. This is obvious after rewriting the phase fa
tor in an equivalent
form
R
R1
dz A (z)
P e yx
= P e 0 d z_ ()A (z()) :
(2.1.23)
as1
U [ yx =
Y
t=0
[1 + dt z_ (t)A (z (t)) :
(2.1.24)
We already used this notation for the produ
t on the RHS in Problem 1.8.
Using Eq. (1.1.153), Eq. (2.1.24)
an also be written as
U [ yx =
y
Y
x
[1 + dz A (z ) :
(2.1.25)
If the
ontour yx is dis
retized as is shown in Fig. 1.3, then the nonAbelian phase fa
tor is approximated by
M
Y
1 + (zi zi 1 ) A
U [ yx = Mlim
!1 i=1
zi + z i
2
; (2.1.26)
92
CHAPTER 2.
Problem 2.2 Write down an expli
it expansion of the nonAbelian phase fa
tor
(2.1.22) in A .
Solution Let us use the notation
Zy
dz : : :
dz : : :
Ry
x dz A (z)
1 Zy
X
as
k=0 x
dz 1
Zy
Zy
For the Abelian
ase, when A(zi )
ommute, Eq. (2.1.32) results in
Zy
zk 1
z1
dt1
dt1
dt2 : : :
t1
0
Z
Z
Z
0
dt2 : : :
Z
(2.1.28)
tk 1
(2.1.29)
where
1
tk 2 ) (t2 t1 ) (2.1.30)
orders the points along the ontour. We shall also denote this theta in a parametrization independent form as
(k; k 1; k 2; : : : ; 2; 1)
Zy
1
k!
zk 1
1k
0 y
Z
dz A (z )A
(2.1.33)
Problem 2.3 Disentangle the nonAbelian phase fa
tor using a path integral over
Grassmann variables living in a
ontour.
Solution Let us dene the average
(tk ; tk 1 ; tk 2 ; : : : ; t2 ; t1 ) :
(2.1.31)
(k; k 1; k 2; : : : ; 2; 1) + (k 1; k; k 2; : : : ; 2; 1)
+ (other permutations of k; : : : ; 1) = 1:
F [ ;
_
D (t)D (t) e 0 dt (t) (t) (0) (0) F [ ;
R
: (2.1.34)
R
_
D (t)D (t) e 0 dt (t) (t) (0) (0)
The path integral in this formula looks like those of Se
t. 1.2 with i (t) and j (t)
being Grassmann variables whi
h depend on the onedimensional variable t 2 [0;
that parametrizes a
ontour, and i and j are the
olor indi
es.
The simplest average, whi
h des
ribes a propagation of the
olor indi
es along
the
ontour, is
(2.1.35)
= ij (t2 t1 ) ;
0 t 1 ; t2 :
i (t2 ) j (t1 )
dz 2 : : :
Z
Zy
z1
dz 1
1
yx
dz 2 : : :
93
(2.1.27)
2.1.
M
X
n=1
(n) [ (n)
i (n) j (m)
ij
0
for n m ;
for n < m :
(2.1.38)
94
CHAPTER 2.
The term (0) (0) is needed to provide nonvanishing integrals over (0) and (0).
It is also seen from the dis
retized version that the path integral in the denominator
on the RHS of Eq. (2.1.34) is equal to unity.
The fermioni
pathintegral representation for the nonAbelian phase fa
tor
reads (see, e.g., Ref. [GN80) as
R
e 0 dt z_ (t)A (z(t))
ij
R
e 0 dt z_ (t) (t)A (z(t)) (t) i ( ) j (0)
:(2.1.39)
There is no pathordering sign on the RHS sin
e the matrix indi
es of A are
onta
ted by and .
In order to prove Eq. (2.1.39), one expands the exponential in A and
al
ulates
the average using Eq. (2.1.35) and the rules of Wi
k's pairing whi
h yields
1
k!
=
Z
0
Z
i ( )
dt1
Z
0
1k
dt2 : : :
Z
(2.1.40)
fun + m2 x
rfun
r
1 m2
2
Dz (t) e
2
1
2 0 dt z_ (t)
z (0)=x
R z( )
rfun
dz
x : (2.1.41)
y P e x
The integral over z ( )  the nal point of the traje
tory  of the matrix element
on the RHS equals
Z
dd z ( )
G(x; y ; A) =
where
P0
R z( ) fun
y P e x dz r x
Ry
dz A (z)
:
Pe x
X0
R
P
yx
yx
dz A (z)
(2.1.43)
Problem 2.5
Solution
ab gf ab
A
rab
=
ab
and the Green fun
tion G (x; y ; A) whi
h obeys
b
2 ab bd
ad (d)
ra
r + m G (x; y ; A) = (x y ) :
X0
yx
2 tr tb U [ yx ta U y [ yx ;
(2.1.44)
(2.1.45)
(2.1.46)
G(x; y ; A)
Z1
1
d e
=
2
Then we get
95
Gab (x; y ; A) =
y
2.1.
(2.1.42)
Sin
e matri
es are rearranged in an inverse order under Hermitean
onjugation, one has2
U y [ yx = U [ xy :
(2.1.47)
In parti
ular, the phase fa
tors obey the ba
ktra
king
ondition
U [ yx U [ xy = 1 :
(2.1.48)
We have
hosen A in the dis
retized phase fa
tor (2.1.26) at the
enter
on the ith interval in order to satisfy Eq. (2.1.48) at nite .
Problem 2.6 Establish the relation between nonAbelian phase fa
tors and the
group of paths.
Solution The group of paths (or loops) is dened as follows. The elements of the
group are the paths yx . The produ
t of two elements zx and yz is the path yx
whi
h is a
omposition of zx and yz . In other words one passes rst the path zx
and then the path yz . The produ
t is denoted as
yz zx =
yx :
(2.1.49)
2 The notation yx means that the
ontour is oriented from x to y while xy stands
for the opposite orientation from y to x. In the pathordered produ
t (2.1.25), these two
ontours result in the opposite orders of multiplying the matri
es.
96
CHAPTER 2.
The multipli
ation of paths is obviously asso
iative but non
ommutative. The inverse element is dened as
1
yx = xy ;
i.e.
(2.1.52)
1 + dz A0 (z )
(z + dz ) [1 + dz A (z )
y (z )
(2.1.54)
U [ yx (x) g.t.
\ (y)" ;
(2.1.55)
and, analogously,
(y) U [ yx g.t.
\ (x)" :
Therefore, U [ yx is, indeed, a parallel transporter.
(2.1.57)
(2.1.58)
These properties of the nonAbelian phase fa
tor are quite similar to those
of the Abelian one whi
h was
onsidered in Subse
t. 1.1.7.
Cal
ulate U [ yx =x and U [ yx =y .
It is
onvenient to start from Eq. (2.1.26). Then only (z1 x) in the
last element of the produ
t should be dierentiated with respe
t to x or (y zM 1 )
in the rst element of the produ
t should be dierentiated with respe
t to y . As
! 0, we get
Problem 2.7
Solution
rfun (y) U [ yx = 0 ;
U [ yx r fun
(x) = 0 :
(2.1.60)
U [ yx
(2.1.56)
g.t.
tr P e dz A (z) ! tr P e dz A (z) :
(2.1.51)
[1 + dz A (z )
U [ yx xy = 1 :
97
(2.1.50)
U [ yz U [ zx = U [ yz zx :
2.1.
U[
yz Czz zx
U [ yx =
U [ yx F (z ) (z )
(2.1.62)
98
CHAPTER 2.
2.1.
99
The formulas of the type of Eq. (2.1.61) are wellknown in dierential geometry where the parallel transport around a small
losed
ontour determines
the
urvature. Therefore, F in Yang{Mills theory is the proper
urvature
in an internal
olor spa
e while A is the
onne
tion.
6dz
dz
A histori
al remark
x
Fig. 2.1: The re
tangular loop Czz , whi
h is added to the
ontour
intermediate point z in the ; plane.
yx at the
sin
e the P{produ
t will automati
ally put F (z ) at the point z of the
ontour
yx .
A
onvenient way to prove Eq. (2.1.61) is to
hoose Czz to be a re
tangle
whi
h is built on the ve
tors dz and dz , as is depi
ted in Fig. 2.1. Using the
representation (2.1.42), we see that the phase fa
tor a
quires the extra fa
tor
[1 + dz r [1 + dz r [1 dz r [1 dz r = 1 dz dz [r ; r (2.1.63)
at the proper order in the pathordered produ
t. Then Eq. (2.1.20) results in
Eq. (2.1.62). Alternatively, one
an prove Eq. (2.1.62) using the dis
retized formula (2.1.26).
Derive a nonAbelian version of the Stokes theorem.
The ordered
ontour integral
an be represented as the doubleordered
surfa
e integral [Are80, Bra80
Problem 2.9
Solution
exp
Cxx
dz A (z ) =
exp
d \F (x)"
(2.1.64)
An analog of the phase fa
tors was rst introdu
ed by Weyl [Wey19 in his
attempt to des
ribe gravitational and ele
tromagneti
intera
tion of an ele
tron on equal footing. What he did is asso
iated in the modern language
with the s
ale rather than gauge transformation, i.e. the ve
torpotential was
not multiplied by i as in Eq. (1.1.154). This explains the term \gauge invarian
e"  gauging literally means xing a s
ale. The fa
tor of i was inserted
by London [Lon27 after the
reation of quantum me
hani
s and re
ognition
of the fa
t that ele
tromagneti
intera
tion
orresponds to the freedom of a
hoi
e of the phase of a wave fun
tion and not with a s
ale transformation.
However, the terminology was left.
2.1.4 Aharonov{Bohm ee
t
The simplest example of a gauge eld is the ele
tromagneti
eld whose
transversal
omponents des
ribe photons. Otherwise, the longitudinal
omponents of the ve
torpotential, whi
h are
hangeable under the gauge transformation, are related to gauging the phase of a wave fun
tion, i.e. permit to
ompare its values at dierent spa
etime points when an ele
tron is pla
ed
in an external ele
tromagneti
eld.
As is wellknown in quantum me
hani
s, the wavefun
tion phase itself
is unobservable. Only the phase dieren
es are observable, e.g. via interferen
e phenomena. For the ele
tron in an ele
tromagneti
eld, the
urrent
(gauged) value of the phase of the wave fun
tion at the point y is related,
as is dis
ussed in Subse
t. 1.1.7, to its value at some referen
e point x by
the parallel transport whi
h is given by Eq. (1.1.159). Therefore, the phase
dieren
e depends on the value of the phase fa
tor for a given path yx along
whi
h the parallel transport is performed.
It is essential that the phase fa
tors are observable in quantum theory,
ontrary to
lassi
al theory. This is seen in the Aharonov{Bohm ee
t. The
orresponding experiment is depi
ted s
hemati
ally in Fig. 2.2.
100
CHAPTER 2.
2.1.
101
XXXXXz
XXXXX
*
XXX interferen
e
solenoid
plane
H
H 6= 0
sour
e
HH
HHH
*
j HH
HH
ele
tron H beam
s
reen
Bohm ee
t. Ele
trons do not pass inside the solenoid where the magneti
eld is
on
entrated. Nevertheless, a phase dieren
e arises between the
ele
tron beams passing through the two slits. The interferen
e pi
ture
hanges with the value of ele
tri
urrent.
It allows one to measure the phase dieren
e between ele
trons passing through the two slits and, therefore, going a
ross opposite sides of the
solenoid. The ne point is that magneti
eld is nonvanishing only inside the
solenoid where ele
trons do not penetrate. Hen
e the ele
trons pass throughout the region of spa
e where the magneti
eld strength vanishes! Nevertheless, the ve
tor potential A itself does not vanish whi
h results in observable
onsequen
es.
The probability amplitude for an ele
tron to propagate from the sour
e at
the point x to the point y at the interferen
e plane is given by the Minkowskispa
e analog of Eq. (1.1.151):
(x; y) =
yx
+
yx
yx
dz A (z)
yx
dz A (z)
ie
yx
dz A (z)
= e ie dz A (z) ;
(2.1.67)
where the
losed
ontour is
omposed from +yx and xy . This is nothing
but the phase fa
tor asso
iated with a parallel transport along the
losed
ontour .
For the given pro
ess this phase fa
tor does not depend on the shape of
+ and
yx
yx . Applying the Stokes theorem, one gets
H
e ie d A = eie d F = eieHS ;
(2.1.68)
where HS is the magneti ow through the solenoid. Therefore, the interferen e pi ture hanges when H hanges3.
Fig. 2.2: Prin ipal s heme of the experiment whi h demonstrates the Aharonov{
X0 ie
ie
X0 ie
yx
dz A (z)
A moral from the Aharonov{Bohm experiment is that the phase fa
tors are
observable in quantum theory while in
lassi
al theory only the ele
tri
and
magneti
eld strengths are observable. The ve
tor potential plays, in
lassi
al theory, only an auxiliary role to determine the eld strength.
For the nonAbelian gauge group G = SU(N
), a quark
an alter its
olor
under the parallel transport so the nonAbelian phase fa
tor (2.1.22) is a
unitary N
N
matrix. A nonAbelian analogue of the quantity, whi
h is
measurable in the Aharonov{Bohm experiment, is the tra
e of the matrix of
the parallel transport along a
losed path. It is gauge invariant a
ording to
Eq. (2.1.58).
It looks suggestive to reformulate gauge theories entirely in terms of these
observable quantities. How to do that will be explained in Chapter 3.
(2.1.66)
yx
102
CHAPTER 2.
Susskind [KS75.
2.2.
103
2q
3q
4q
5q
6q
a6
? a
Fig. 2.3: Twodimensional latti
e with periodi
boundary
onditions. The sites labelled by the same numbers are identied. The latti
e spa
ing equals a
while the spatial size of the latti
e
orresponds to L1 = 6 and L2 = 4.
x = n a ;
(2.2.1)
n = (n1 ; n2 ; : : : ; nd )
(2.2.2)
are integer numbers. The points (2.2.1) are
alled the latti
e sites.
The dimensional
onstant a, whi
h is equal to the distan
e between the
neighboring sites, is
alled the latti
e spa
ing. Dimensional quantities are
usually measured in the units of a, putting therefore a = 1.
A twodimensional latti
e is depi
ted in Fig. 2.3. A fourdimensional latti
e for whi
h the distan
es between sites are the same in all dire
tions (as
for the latti
e in Fig. 2.3) is
alled the hyper
ubi
latti
e.
The next
on
ept is the link of a latti
e. A link is a line whi
h
onne
ts two
neighboring sites. A link is usually denoted by the letter l and is
hara
terized
by the
oordinate x of its starting point and its dire
tion = 1; : : : ; d:
l =
fx; g :
(2.2.3)
104
CHAPTER 2.
l = + n1 d + n2 dL + : : : + nd dLd 1 ;
x + a^ + a^
e
(2.2.6)
p
e
105
Cal
ulate the numbers of sites, links and plaquettes for a symmetri
hyper
ubi
latti
e with periodi
boundary
onditions.
Solution Let us denote L1 = L2 = : : : = Ld = L. Then
d(d 1) d
Nl = dLd ;
Np =
L :
(2.2.5)
Ns = Ld ;
2
Problem 2.11 Label the latti
e links by a natural number l 2 [1; Nl .
Solution One of the
hoi
es is as follows
Fig. 2.4: A link of a latti
e. The link
onne
ts the sites x and x + a^.
e
Problem 2.10
x + a^
x + a^
2.2.
x + a^
Fig. 2.5: A plaquette of a latti
e. The plaquette boundary is made of four links.
The link l
onne
ts sites with the
oordinates x and x + a^, where ^ is a unit
ve
tor along the dire
tion, as shown in Fig. 2.4. The lengths of all links are
equal to a for a hyper
ubi
latti
e.
The elementary square en
losed by four links is
alled the plaquette. A
plaquette p is spe
ied by the
oordinate x of a site and by the two dire
tions
and along whi
h it is built on:
p = fx; ; g :
(2.2.4)
A plaquette is depi
ted in Fig. 2.5. The set of four links whi
h bound the
plaquette p is denoted as p.
If the spatial size of the latti
e is innite, then the number of dynami
al
degrees of freedom is also innite (but enumerable). In order to limit the
number of degrees of freedom, one deals with a latti
e whi
h has nite size
L1 L2 Ld in all dire
tions (see Fig. 2.3).
Usually, one imposes periodi
boundary
onditions to redu
e nitesize
ee
ts whi
h are due to the nite extent of the latti
e. In other words, one
identies pairs of sites whi
h lie on parallel bounding hyperplanes. Usually
the sites with the
oordinates (0; n2; : : : ; nd ) and (L1 ; n2 ; : : : ; nd) are identied
and similarly along other axes.
The next step is to des
ribe how matter elds and gauge elds are dened on
a latti
e.
A matter eld, say a quark eld, is attributed to the latti
e sites. One
an just think that a
ontinuous eld ' (x) is approximated by its values at
the latti
e sites5 :
A (x)
=) Ux; :
(2.2.8)
It looks natural sin
e a link is
hara
terized by a
oordinate and a dire
tion
(see Eq. (2.2.3))  the same as A (x). Sometimes the notation U (x) is
alternatively used for Ux; .
The link variable Ux;
an be viewed as
Ux; =
R x+a^
P e x dz A (z) ;
where the integral is along the link fx; g. As a ! 0, this yields
Ux; ! e aA (x)
(2.2.9)
(2.2.10)
5 We write arguments of fun tions whi h are dened on a latti e as subs ripts keeping
106
q
CHAPTER 2.
b)
a)
Fig. 2.6: Des
ription of
ontinuum eld
ongurations by latti
es a) \
oarse" and b)
\ne". The latti
e a)
an represent the given
ontinuum eld
onguration
very roughly, while the latti
e b) has the spa
ing whi
h is small enough.
so that Ux; is expressed via the exponential of the th
omponent of the
ve
tor potential, say, at the
enter of the link to agree with Eq. (2.1.26).
Sin
e the pathordered integral in Eq. (2.2.9) depends on the orientation,
the
on
ept of the orientation of a given link arises. The same link, whi
h
onne
ts the points x and x + a^,
an be written either as fx; g or as
fx + a^; g. The orientation is positive for > 0 in the former
ase (i.e.
the same as the dire
tion of the
oordinate axis) and is negative in the latter
ase.
We have assigned the link variable Ux; to links with positive orientations.
The U matri
es whi
h are assigned to links with negative orientations are
given by
(2.2.11)
Ux+a^; = U y :
x;
2.2.
107
x + a^
?
x
x + a^ + a^
x + a^
Ux; g.t.
!
x+a^ Ux;
yx ;
(2.2.13)
where the matrix
x is equal to the value of
(x) at the latti
e site x. This
denes the latti
e gauge transformation.
The plaquette variable transforms under the latti
e gauge transformation
as
g.t.
Up !
x Up
y :
(2.2.14)
x
g.t.
tr Up :
(2.2.15)
The invarian
e of the tra
e under the latti
e gauge transformation is used
in
onstru
ting an a
tion of a latti
e gauge theory. The simplest (Wilson)
a
tion is
X
1
(2.2.16)
Re tr Up :
SLat [U =
1
N
p
The summation is over all the elementary plaquettes of the latti
e (i.e. over
all x, , and ), regardless of their orientations.
Sin
e a reversal of the orientation of the plaquette boundary results, a
ording to Eq. (2.1.47), in
omplex
onjugation:
reorient
! tr U y = (tr U ) ;
tr U
(2.2.17)
p
p
p
108
CHAPTER 2.
SLat [U =
1 X
1
2 oriented p
1
;
tr U
N
p
(2.1.18).
Solution
X
(2.2.18)
(2.2.22)
109
where the sum is also over the two possible orientations of the boundary of a
given plaquette.
In the limit a ! 0, the latti
e a
tion (2.2.16) be
omes in d = 4 the a
tion
of a
ontinuum gauge theory. In order to show this, let us rst note that
Up
2.2.
6=
U fx;; g Uyfx;; g
? 6
x
= 0;
(2.2.24)
1
C
CCCC
? 6 CC
 A
= 0:
(2.2.25)
In the last equation we have depi
ted only plaquettes with the positive orientation,
while those with the negative one are re
overed by the sum over for < 0.
Equation (2.2.24) (or (2.2.25)) is the latti
e analog of the nonAbelian Maxwell
equation.
In order to show how this equation reprodu
es the
ontinuum one (2.1.18) as
a ! 0, let us rewrite the se
ond term on the LHS of Eq. (2.2.25) using (2.2.11):
0
B
B
X B
B
B
B
6= B
? 6
x
?6? 6

x a^
C
C
C
C
C
C
C
A
= 0
(2.2.26)
6=
U fx;; g
Ux a^; U fx a^;; g Uxy a^; = 0 :
(2.2.27)
It is now
lear that the plaquette boundary in the se
ond term on the LHS, whi
h is
the same as the rst one but transported by one latti
e spa
ing in the dire
tion,
is asso
iated with F (x a^). Using Eqs. (2.2.10) and (2.2.19), we re
over the
ontinuum Maxwell equation (2.1.18).
The limit a ! 0, when Eqs. (2.2.10) and (2.2.19) hold reprodu ing the ontinuum a tion (2.2.21), is alled the naive ontinuum limit. It is assumed in the
110
CHAPTER 2.
naive
ontinuum limit that A (x) is weakly
u
tuating at neighboring latti
e links. Flu
tuations of the order of 1=a are not taken into a
ount, sin
e
dis
ontinuities of the ve
tor potential in the
ontinuum theory are usually
asso
iated with an innite a
tion.
Another subtlety with the naive
ontinuum limit is that next order in a
terms of the expansion of the latti
e a
tion (2.2.16), say the term / a2 tr F 3 ,
are asso
iated with nonrenormalizable intera
tions and the smallness of a2
an be
ompensated, in prin
iple, by quadrati
divergen
ies.
The a
tual
ontinuum limit of latti
e gauge theories is in fa
t very similar
to the naive one modulo some nite renormalizations of the gauge
oupling
onstant. The large
u
tuations of A (x) of the order of 1=a be
ome frozen
when passing to the
ontinuum limit. How to pass to the
ontinuum limit of
latti
e gauge theories is explained in Subse
t. 2.2.7 below.
adj (U ) = jtr U j
1;
(2.2.28)
Sadj [U =
X
X
1
jtr Up j2 :
N
2
X
1
1
Re tr Up + A
N
2 p
(2.2.29)
1
jtr Up j2 : (2.2.30)
N
2
111
oin
ide, these two gauge theories
oin
ide in the
ontinuum while they dier
on the latti
e.
One more possibility is to use the phase fa
tor asso
iated, say, with the
boundary of two plaquettes having a
ommon link, or the phase fa
tors for
more
ompli
ated
losed
ontours of nite size on the latti
e to
onstru
t the
a
tion. These a
tions will also reprodu
e, in the naive
ontinuum limit, the
a
tion of the
ontinuum gauge theory.
The independen
e of the
ontinuum limit of latti
e gauge theories on the
hoi
e of latti
e a
tions in
alled the universality. We shall say more about
this in the next Se
tion when dis
ussing renormalization group on the latti
e.
Z ( ) =
Z Y
x;
dUx; e S[U ;
(2.2.31)
The naive
ontinuum limit will be the same as for the Wilson a
tion (2.2.16).
Moreover, one
an dene the latti
e a
tion as a mixture of the fundamental
and adjoint representations [BC81, KM81:
Smixed [U
2.2.
2N
:
g2
(2.2.32)
112
CHAPTER 2.
U = a4 I + i~a~
(2.2.34)
with ~ being the Pauli matri
es. The Haar measure for SU(2) then reads
4
1 Y
dU = 2
da (1) a2
=1
1 ;
(2.2.35)
Problem 2.14
The partition fun
tion (2.2.31)
hara
terizes va
uum ee
ts in the quantum theory. Physi
al quantities are given by the averages of the type of
Eq. (1.2.6):
h F [U i
= Z
1 ( )
Z Y
x;
dUx; e S[U F [U ;
(2.2.39)
2.2.
113
where F [U is a gauge invariant fun
tional of the link variable Ux; . The averages (2.2.39) be
ome the
orresponding expe
tation values in the
ontinuum
theory as a ! 0 and is related to g2 by Eq. (2.2.32).
C = fx; 1 ; : : : ; n g :
(2.2.40)
114
CHAPTER 2.
(R; 0)
?

(0; 0)
2.2.
(R; T )
we then have
(0; T )
we get
= Z
1 ( )
Z Y
1
dUx; e S[U tr UC :
N
x;
(2.2.42)
Problem 2.15
Solution
WRT
(2.2.41)
For the links with a negative dire
tion it is again
onvenient to use
Eq. (2.2.11).
A
losed
ontour has ^1 + : : : + ^n = 0. The tra
e of the phase fa
tor for
a
losed
ontour, whi
h is gauge invariant, is
alled the Wilson loop.
The average of the Wilson loop is determined by the general formula (2.2.39) to be
WC
ij
(2.2.44)
1 tr (0) y (T E
(2.2.45)
ij
ij ) :
N
Inserting in the last equation a sum over a
omplete set of intermediate states
1
tr U
N
C
RR
WRT =
ij (t)
115
jn ih nj = 1 ;
1 tr h (0)j ni
n y (T )
ij
ji
N
n
X 1
=
jh ij (0)j nij2 e En T
N
n
(2.2.46)
(2.2.47)
where En is the energy of the state jni. As T ! 1, only the ground state with the
lowest energy survives in the sum over states and we nally nd
large T
WRT ! e E0 T ;
(2.2.48)
whi
h results in Eq. (2.2.43).
Noti
e that nothing relies in this derivation on the latti
e. Therefore,
Eq. (2.2.43) holds for a re
tangular loop in a
ontinuum theory as well.
dz A (z ) =
dd xJ (x)A (x) ;
(2.2.49)
where
J (x) = e
dz (d) (x z )
(2.2.50)
is a fourve
tor
urrent of a
lassi
al parti
le moving along the traje
tory C
whi
h is des
ribed by the fun
tion z (t).
116
CHAPTER 2.
ln W (C ) =
ln e
R
E
i d4 xJ (x)A (x)
(2.2.51)
determines a
hange of the a
tion of the
lassi
al parti
le due to ele
tromagneti
intera
tion in a
ordan
e with Eq. (2.2.43).
A similar interpretation of Eq. (2.2.43) in the nonAbelian
ase is somewhat more
ompli
ated. For a heavy parti
le moving along some traje
tory
in spa
etime,
olor degrees of freedom are quantum and easily respond to
hanges of the gauge eld A (x), whi
h intera
ts with them. Let us suppose
that a quark and an antiquark are
reated at the same spa
etime point in
some
olor state. Then this state must be singlet with respe
t to
olor (or
olorless) sin
e the average over the gauge eld would vanish otherwise. When
quarks are going away, their
olor
hanges from one point to another simultaneously with
hanging the
olor of the gauge eld, in order for the system of
the quarks plus the gauge eld to remain
olorless. Therefore, the averaging
over the gauge eld leads to an averaging over
u
tuations of quark
olor
degrees of freedom. E0 (R) in Eq. (2.2.43) is asso
iated with the intera
tion
energy averaged over
olor in this way.
Derive a nonAbelian analog of Eq. (2.2.50).
The proper nonAbelian extension of Eq. (2.2.50) is [Won70
Problem 2.16
Solution
Ja (x) =
Z
(2.2.52)
where I a (t), whi
h des
ribes the
olor state of a
lassi
al parti
le moving along the
traje
tory z (t) in an external Yang{Mills eld A (z ), is a solution of the equation
(2.2.53)
(2.2.54)
117
It is
lear that
D
2.2.
(2.2.55)
g2 N
2 1
4a 2N
(2.2.56)
as a ! 0.8
The potential energy of the intera
tion between the stati
quarks is therefore dened as the dieren
e
(2.2.57)
(2.2.58)
where the Haar measure (given for SU(2) by Eq. (2.2.35)) is normalized by
Z
dU = 1 :
(2.2.59)
118
CHAPTER 2.
dU Uji U y l
= A li jk + B ji lk :
(2.2.61)
AN
+ B = 1 :
(2.2.62)
One more relation between A and B
omes from the fa
t that the
hara
ter in
the adjoint representation is given by Eq. (2.2.28). Contra
ting Eq. (2.2.61) by ij
and kl , and using the orthogonality of
hara
ters whi
h says
Z
dU jtr U j2 1
= 1;
(2.2.63)
we get
AN
+ BN
2 = 1 :
Therefore, A = 1=N
and B = 0 whi
h proves Eq. (2.2.60).
(2.2.64)
Wp =
x;
dUx;
R Q
x;
P
1 + N1
Re tr Up0
p0
dUx; 1 +
1
N
tr Up
P 1
N Re tr Up0
p0
?6
0
p

119
?6p
Fig. 2.9: Boundaries of the plaquettes p and p0 with the opposite orientations: p
and p0 , respe
tively.
Problem 2.17
Solution
2.2.
i  j = 1 i j ;
N
l
k
l k
where the semi
ir
les are asso
iated with the Krone
ker deltas:
i = i :
(2.2.69)
l l
This notation is
onvenient sin
e the lines whi
h denote the deltas in the last
equation
an be asso
iated with propagation of the
olor indi
es. Analogously
a
losed line represents the
ontra
ted delta, whi
h is summed up over the
olor indi
es,
j
2
+ O : (2.2.66)
(2.2.68)
= ii = N :
(2.2.70)
9 A
al
ulation of more
ompli
ated group integrals (2.2.58) in the graphi
notations
is dis
ussed in the le
tures by Wilson [Wil75 and in the Chapter 8 of the book by
Creutz [Cre83. An alternative way of
al
ulating the group integrals by the
hara
ter
expansion is des
ribed in the review by Droue and Zuber [DZ83.
120
CHAPTER 2.
  6 ?6 ?6 ?6  ?6 ?6 ?6 ?6 ?6 ?6 ?
      6
6 ?6 ?6?6 ?6 ?6 ?
WC = (Wp )Amin(C ) ;
2N
2
Wp =
4
= 1;
(2.2.71)
(2.2.73)
where the
ontra
ted Krone
ker deltas are asso
iated with the four sites of
the plaquette.
The nal answer for the plaquette average is
Wp =
121
where Wp is given by Eq. (2.2.72) and Amin(C ) is the area (in units of a2 )
of the minimal surfa
e.
For the re
tangular loop, whi
h is depi
ted in Fig. 2.8, the minimal surfa
e
is just a pie
e of the plane bounded by the re
tangle. Therefore, we get
WRT = (Wp )R T
(2.2.74)
Using the graphi
representation (2.2.68) for ea
h of the four links depi
ted
in Fig. 2.9, we get
y 0 = 14
dUx; tr Up tr Up
N
x;
R Q
2.2.
(2.2.72)
The result for SU(2) diers by a fa
tor of 1=2 be
ause tr Up is real for SU(2)
so that the orientation of the plaquettes
an be ignored.
The graphi
representation (2.2.68) is useful for evaluating the leading
order of the strong
oupling expansion for more
ompli
ated loops. A
ording
to Eq. (2.2.67), a nonvanishing result emerges only when plaquettes,
oming
from the expansion of the exponentials of Eq. (2.2.42) in ,
ompletely
over
a surfa
e en
losed by the given loop C as is depi
ted in Fig. 2.10. In this
ase
ea
h link is en
ountered twi
e (or never), on
e in the positive dire
tion and
on
e in the negative dire
tion, so that all the group integrals are nonvanishing.
The leading order in
orresponds to lling a minimal surfa
e, whose area
The exponential dependen
e of the Wilson loop average on the area of the
minimal surfa
e (as in Eq. (2.2.73)) is
alled the area law. It is
ustomarily
assumed that if an area law holds for loops of large area in pure gluodynami
s (i.e. in the pure SU(3) gauge theory) then quarks are
onned. In other
words, there are no physi
al jini or houtj quark states. This is the essen
e of
Wilson's
onnement
riterion. The argument is that physi
al amplitudes,
e.g. the polarization operator, do not have quark singularities when the Wilson
riterion is satised. I refer the reader to the wellwritten original paper
by Wilson [Wil74, where this point is
laried.
Another, somewhat oversimplied, justi
ation for the Wilson
riterion is
based on the relationship (2.2.43) between the Wilson loop average and the
potential energy of intera
tion between stati
quarks. When the area law
W (C ) large!C e KAmin(C )
(2.2.75)
holds for large loops, the potential energy is a linear fun
tion of the distan
e
between the quarks:
E (R) = KR :
(2.2.76)
122
CHAPTER 2.
2.2.
123
the gauge eld between quarks would be distributed over the whole spa
e as
is depi
ted in Fig. 2.11b. The Wilson loop average would have the behavior
W (C )
r
a)
large C
e onstL(C ) ;
(2.2.79)
where L(C ) stands for the length (or perimeter) of the
losed
ontour C .
This behavior of the Wilson loops is
alled the perimeter law. To ea
h
order of perturbation theory, it is the perimeter law (2.2.79), rather than the
area law (2.2.75), that holds for the Wilson loop averages. A perimeter law
orresponds to a potential whi
h
an not
onne quarks.
b)
Fig. 2.11: Lines of for
e between stati
quarks for a) linear and b) Coulomb intera
tion potentials. For the linear potential the lines of for
e are
ontra
ted into a tube, while they are distributed over the whole spa
e for the
Coulomb one.
To distinguish between the area and perimeter law behavior of the Wilson
loop averages, Creutz [Cre80 proposed to
onsider the ratio
(I; J ) =
W W
ln I J (I 1)(J
W(I 1)J WI (J
1)
1)
(2.2.80)
is the energy of the string per unit length. This string is stret
hed with the
distan
e between quarks and prevents them from moving apart to ma
ros
opi
distan
es.
Equation (2.2.74) gives
1 2N
2
ln
a2
1
= 2 ln N
g2
(2.2.77)
a
for the string tension to the leading order of the strong
oupling expansion.
Next orders of the strong
oupling expansion result in
orre
tions in to this
formula.
Therefore,
onnement holds in the latti
e gauge theory to any order of
the strong
oupling expansion.
K =
(I; J )
large I; J
a2 K ;
(2.2.81)
(2.2.82)
i.e. does not depend on I or J and
oin
ides with the string tension. This
property of the Creutz ratio was used for numeri
al
al
ulations of the string
tension.
E (R) =
L(I J ) = 2I + 2J ;
g2 N
2 1
;
4R 2N
(2.2.78)
124
CHAPTER 2.
2.2.
experimental value10
K = (400 MeV)2
1 Gev/fm :
(2.2.83)
Then the renormalizability pres
ribes that variations of a, whi
h plays the role
of a latti
e
uto, and of the bare
harge g2 should be done simultaneously
in order that K does not
hange.
Given Eq. (2.2.77), this pro
edure
alls for a ! 1 as g2 ! 1. In other
words, the latti
e spa
ing is large in the strong
oupling limit,
ompared to
1 fm  the typi
al s
ale of the strong intera
tion. This is a situation of the
type shown in Fig. 2.6a. Su
h a
oarse latti
e
an not des
ribe
orre
tly the
ontinuum limit and, in parti
ular, the rotational symmetry.
In order to pass to the
ontinuum, the latti
e spa
ing a should be de
reased
to have a pi
ture like that in Fig. 2.6b. Equation (2.2.77) shows that a
de
reases with de
reasing g2 . However, this formula
eases to be appli
able
in the intermediate region of g2 1 and, therefore, a 1 fm.
The re
ipe for further de
reasing of a is the same as in the strong
oupling
region: further de
reasing of g2 . While no analyti
formulas are available at
intermediate values of g2, the expe
ted relation between a and g2 for small
g2 is predi
ted by the known twoloop GellMann{Low fun
tion of QCD.
For pure SU(3) gluodynami
s, Eq. (2.2.77) is repla
ed at small g2 by
1 162
K =
onst 2
a 11g2
102
121
162
11g2 :
(2.2.84)
125
strong
oupling
ln a2 K
JJ
J asymptoti
s
aling
JJ
J
1=g2
JJ
JJ
Fig. 2.12: Dependen
e of the string tension on 1=g2 . The strong
oupling formula
(2.2.77) holds for small 1=g2 . The asymptoti
s
aling formula (2.2.84) sets
in for large 1=g2 . Both formulas are not appli
able in the intermediate
region of 1=g2 1 whi
h is depi
ted by the dashed line.
1
exp
a
dg2
B(g2) :
(2.2.85)
Up to now there was no dieren
e between QCD and QED. The dieren
e
stems from the fa
t that the GellMann{Low fun
tion B(g2 ) is positive for
QED and negative for QCD. In QED e2(a) in
reases with de
reasing a while in
QCD g2 (a) de
reases with de
reasing a. The latter behavior of the
oupling
126
CHAPTER 2.
1
QCD
= a exp
Z
dg2
B(g2 ) :
(2.2.86)
The only
han
e for the RHS to diverge is to have a zero of the GellMann{
Low fun
tion B(g2) at some xed point g2 = g2 . Therefore, the bare
oupling
should approa
h the xedpoint value g2 to des
ribe the
ontinuum.
As we have dis
ussed, B(0) = 0 for a nonAbelian gauge theory so that
g2 = 0 is a xedpoint value of the
oupling
onstant. Therefore, the
ontinuum limit is asso
iated with g2 ! 0 as is said above.
2.3.
LATTICE METHODS
127
128
CHAPTER 2.
2.3.
129
LATTICE METHODS
6 Wp
SP = Z ;
Ux;
Z = I e 2in =N ;
(2.3.3)
/ 1=g2
Fig. 2.13: Typi al dependen e of the plaquette average for a rstorder phase transition whi h o urs at = .
the plaquette average (2.2.65) as is depi
ted in Fig. 2.13. The form of Wp ( )
at small is given to the leading order of the strong
oupling expansion by
Eq. (2.2.72), while that at large is pres
ribed by the latti
e perturbation
theory11 to be
dG
Wp ( ) = 1
+ O( 2 ) ;
(2.3.1)
d
where dG is the dimensionality of the gauge group (dG = N
2 1 for SU(N
),
dG = N
2 for U(N
)) and d is the dimensionality of the latti
e as before.
This behavior of the plaquette average is quite analogous to a dependen
e
of the internal energy per unit volume (
alled the spe
i
energy) in statisti
al
systems. In order to see an analogy between the spe
i
energy and (1 Wp ),
let us remember that is an analog of the inverse temperature and rewrite
Eq. (2.2.65) as
1
ln Z ( ) ;
(2.3.2)
Wp ( ) = 1 +
Np
where the partition fun
tion is given by Eq. (2.2.31) and the number of plaquettes Np is an analog of the volume of a statisti
al system.
Derive Eq. (2.3.1) for the SU(N
) gauge group.
The partition fun
tion (2.2.31)
an be
al
ulated at large by the saddlepoint method. The saddlepoint
ongurations are given by solutions of the
lassi
al
Problem 2.18
Solution
n = 1; : : : ; N :
(2.3.4)
It is evident that this is a solution be
ause elements of the
enter
ommute so that
Z and Z
an
el ea
h other in Ux;; .
In order to take into a
ount
u
tuations around the saddlepoint solution (2.3.3), let us expand
SP e i ta ax; ;
Ux; = Ux;
(2.3.5)
where the order of multipli
ation is inessential sin
e Z
ommute with the generators
ta . The expansion of tr Up to the quadrati
order in a reads
1
1 2
tr U
= 1
E ;
(2.3.6)
N
x;;
4N
x;;
where
(2.3.7)
Due to the lo
al gauge invarian
e, we
an always
hoose, say, x;d = 0 so that there
are only Nl Ns independent 's.
Substituting into Eq. (2.2.31) and expanding the Haar measure, we get
+1
N
d X
Y
Y Z
2
(2.3.8)
Z ( ) /
dax; e Ex;; =4N
:
=1 n =1 x;<d;a 1
The sum over n , whi
h is due to the degenerate saddle points, is just an irrelevant
onstant.
We see from Eq. (2.3.8) that only
1
ax; p
(2.3.9)
are essential whi
h justies the expansion in . Res
aling the integration variables
in Eq. (2.3.8), we get, therefore,
Z ( )
(Nl Ns ) dG =2 :
(2.3.10)
130
CHAPTER 2.
2.3.
Problem 2.19 Repeat the derivation of the previous Problem for the adjoint a
tion (2.2.29).
Solution The only dieren
e from the Wilson a
tion (2.2.16) is that the saddlepoint solution (2.3.3) is now modied as
SP = Z ;
Ux;
x;
n =1
=)
N
d X
Y
x; nx; =1
(2.3.12)
whi
h only
hanges an irrelevant overall
onstant. Therefore, Eq. (2.3.1) remains
un
hanged providing the plaquette average is also taken in the adjoint representation. This supports the expe
tation that the
ontinuum limits for the both a
tions
oin
ide.
The rstorder phase transitions of the type in Fig. 2.13 are usually harmless and are not asso
iated with de
onnement. They are related with dynami
s of some latti
e degrees of freedom (say, with large
u
tuations of the
link variable Ux; whi
h o
ur independently at adja
ent links) whi
h do not
ae
t the
ontinuum limit and are
alled latti
e artifa
ts. Moreover, these
latti
e degrees of freedom be
ome frozen for > whi
h is ne
essary for
the
ontinuum limit to exist. A brief des
ription of some latti
e artifa
ts and
of the related phase transitions
an be found in the Se
tion 3 of the review
arti
le [Mak84.
Another possibility for a latti
e system is to undergo a se
ondorder phase
transition in analogy with spin systems. In this
ase Wp is
ontinuous but
the derivative Wp = be
omes innite at the
riti
al point = as is
depi
ted in Fig. 2.14. Given Eq. (2.3.2), this derivative is to be
onsidered
as an analog of the spe
i
heat of statisti
al systems. Its behavior at small
and large is governed by Eqs. (2.2.72) and (2.3.1), respe
tively.
Dierentiating Eq. (2.2.65) with respe
t to , the derivative Wp =
an
be expressed via the sum of the
onne
ted
orrelators:
6 Wp
(2.3.11)
i.e. may take on dierent values at dierent links. It is evident that this is a
minimum of the a
tion (2.2.29).
The only modi
ation of Eq. (2.3.8) is
N
d X
Y
131
LATTICE METHODS
1
1 X
Wp
1
:
tr Up tr Up0
=
2 oriented p0 N
N
onn
(2.3.13)
/ 1=g2
This formula shows also that Wp = is positive denite, sin
e the RHS
an
be rewritten using translational invarian
e as
1 X
1
1
tr U
tr U 0
2 oriented p0 N
p N
p
onn
1
=
4Np
0;
1
tr Up
N
oriented p
X
!2 +
1
4Np
1
tr Up
N
oriented p
X
!+2
(2.3.14)
where the equality is possible only for a Gaussian averaging, i.e. for a free
theory. This repeats the standard proof of the positivity of spe
i
heat in
statisti
al me
hani
s.
Sin
e ea
h term of the sum in Eq. (2.3.13) is nite (remember that a tra
e
of a unitary matrix takes on values between N
and N
), the only possibility
for the RHS to diverge is for the sum over plaquettes p0 to diverge. This is
possible only when longrange (in the units of the latti
e spa
ing)
orrelations
are essential or, in other words, the
orrelation length is innite. Thus, we
have on
e again reprodu
ed the argument that the
ontinuum limit of latti
e
theories is rea
hed at the points of se
ondorder phase transitions.
Su
h a se
ondorder phase transition seems to o
ur in
ompa
t QED (i.e.
the U(1) latti
e gauge theory with fermions) at e2 1. It is asso
iated there
with de
onnement of ele
trons. Ele
trons are
onned for e2 > e2 , similar
to quarks in the latti
e QCD, and are liberated for e2 < e2. The intera
tion
132
CHAPTER 2.
potential looks like that of Fig. 2.11b for e2 < e2 and like that of Fig. 2.11a in
the
onnement region e2 > e2 . 12 In order to rea
h the
ontinuum limit with
de
onned ele
trons, the bare
harge e2 should be
hosen slightly below the
riti
al value. Then the renormalized physi
al
harge
an be made as small as
the experimental value ( = 1=137) a
ording to the renormalization group
arguments whi
h are presented in the Remarks to Subse
t. 2.2.7.
The nature of the phase transition in a fourdimensional
ompa
t U(1)
latti
e gauge theory without fermions was investigated by numeri
al methods.
While the very rst paper [LN80 indi
ated that the phase transition is of
se
ond order, some more advan
ed re
ent investigations say that it may be
weakly rst order (see, e.g., the talks at the Latti
e Conferen
e [Lat94, Part
2.7). Anyway, we need fermions whi
h usually weaken a phase transition that
happens in a pure latti
e gauge theory.
There are no indi
ations that a se
ondorder phase transition o
urs in
nonAbelian pure latti
e gauge theories at intermediate values of . This
very strongly supports a behavior of the string tension of the type depi
ted
in Fig. 2.12. The se
ondorder phase transition o
urs in four dimensions at
= 1 (or g2 = 0) a
ording to the general arguments of Subse
t. 2.2.7,
whi
h is ne
essary for the
ontinuum limit to exist.
2.3.
se t. 2.5.4.
m
m
Fig. 2.15: Graphi
representation of the self
onsisten
y
ondition (2.3.17). The link
variables are repla
ed by m I at all links ex
ept for a given one denoted
by the bold line.
In the simplest version of the meaneld method, the link variable Ux;
is repla
ed by the meaneld value m I everywhere but at a given link (see
Fig. 2.15) at whi
h the self
onsisten
y
ondition
D
[Ux;ij
133
LATTICE METHODS
= m ij
(2.3.15)
is imposed.
The average on the LHS of Eq. (2.3.15) is
al
ulated with the a
tion whi
h
is obtained from (2.2.16) by the substitution of m I for all the link variables
(or their Hermitean
onjugate) ex
ept at the given link. Sin
e the given link
enters 2(d 1) plaquettes, the average on the LHS of Eq. (2.3.15) is to be
al
ulated with the a
tion
S0 [U = 2 (d 1) m3 Re tr Ux; +
onst :
(2.3.16)
Therefore, the self
onsisten
y
ondition (2.3.15)
an be written by
the substitution of the meaneld ansatz into the latti
e partition fun
tion (2.2.31) as
R
with
Re tr U 1
dU eN
N
tr U = m
R
dU e N
Re tr U
(2.3.17)
= 2 (d 1) m3 2 :
N
(2.3.18)
134
CHAPTER 2.
2.3.
135
LATTICE METHODS
or for the SO(3) group), agreement with numeri
ally
al
ulated positions of
the phase transitions is remarkable.
6m
Cal
ulate for the SU(1) latti
e gauge theory, when the group
2 for 1 (a strong
oupling phase)
integral on the LHS of Eq. (2.3.17) equals: =
and 1 1=2 for 1 (a weak
oupling phase).
Solution For the strong
oupling phase, the self
onsisten
y equation
(d 1) m3 2 = m
(2.3.20)
N
has the only solution m = 0. The other solutions are una
eptable due to stability
onsideration.
The nontrivial solutions of the self
onsisten
y equation appear in the weak
oupling phase when dm=d = 1 or d=dm = 0. Dierentiating, we get in d = 4
Problem 2.20
/ 1=g2
Fig. 2.16: Typi al behavior of the solutions of the self onsisten y equation (2.3.17).
The only solution with m = 0 exists for < . Two more solutions
appear for > . The one depi
ted by the dashed line is unstable. The
a
tual value of m versus is depi
ted by the bold lines. A rstorder phase
transition is asso
iated with = .
The meaning of Eq. (2.3.17) is very simple: the average of the tra
e of the
link variable at the given link should
oin
ide with m whi
h is substituted for
all other links of the latti
e.
In order to verify whether the self
onsisten
y
ondition (2.3.17) admits
nontrivial solutions, one should rst
al
ulate the group integral on the LHS
and then solve the self
onsisten
y equation for m versus . A typi
al behavior of the solution is depi
ted in Fig. 2.16. For all values of , there exists a
trivial solution m = 0 whi
h is asso
iated with no mean eld. At some value
, two more solutions of the self
onsisten
y equation appear. The upper of
them is asso
iated with positive spe
i
heat while the lower one
orresponds
to negative one. This
an be seen by noting that
Wp = m4
(2.3.19)
in the meaneld approximation whi
h follows from the substitution of the
linkvariables in the denition (2.2.65) by the meaneld values. This nontrivial solution is preferred for > sin
e the partition fun
tion for it is
larger (or the free energy is smaller) than for the m = 0 solution. The value
of is often asso
iated with the point of a rstorder phase transition.
The meaneld method in su
h a simple form was rst applied to nonAbelian latti
e gauge theories in Ref. [GL81, CGL81. For the
ases when a
rstorder phase transition o
urs (say, for the SU(N
) groups with N
> 3
1 N
2 = 12 3m2
m
4m3 ;
(2.3.21)
whi h yields
= 43 = 0; 79 :
m = 3 ;
(2.3.22)
4
N
2
34
It is still left to verify that the proper is indeed asso
iated with the weak
oupling
phase. From Eq. (2.3.18), we get = 2 and this is the
ase.
F
e hF i0
(2.3.23)
e
0
to Chapter 1.
e.g.
136
CHAPTER 2.
whi
h is due to the
onvexity of the exponential fun
tion, where h: : :i0 means
averaging with respe
t to a trial a
tion.
Let us
hoose the trial partition fun
tion
Z0 =
Z Y
x;
dUx; e
N
x; Re tr Ux;
(2.3.24)
as a produ
t of onelink integrals. Adding and subtra
ting the trial a
tion,
we write down the following bound on the partition fun
tion (2.2.31):
P
P Re tr Ux;
Re tr Up N
N
x;
p
0
Z0 e
X
X
Re tr Up N
Re tr Ux;
N
p
x;
2 Nl m ;
= Np m4 N
Z0 =
Z Y
x;
dUx; e N
P
y
x; Re tr Bx; Ux;
Problem 2.21
(2.3.27).
Solution
(2.3.27)
ij equation
but is arbitrary otherwise. Now Ux;
vanishes
after summing over equiv0
alent maxima whi
h results in integrations over d
x .
Let us denote
ij =
Mx;
RQ
P
y
ij
x; Re tr Bx; Ux; Ux;
P
y Ux;
Re tr Bx;
N
dUx; e
x;
x; dUx; e
RQ
x;
X Re tr U
p
N
p
X Re tr M
=
p
N
p
x;
y Ux;
Re tr Bx;
x;
(2.3.29)
+
0
y Mx;
Re tr Bx;
(2.3.30)
(2.3.26)
137
LATTICE METHODS
(2.3.25)
Here h: : :i0 means averaging with respe
t to the same a
tion as in Eq. (2.3.24).
Sin
e the expression whi
h is averaged in the exponent in Eq. (2.3.25) is
linear in ea
h of the link variables, it
an be
al
ulated via the onematrix
integral given by the LHS of Eq. (2.3.17). Therefore, we get
*
2.3.
P
y
P Re tr M
p N
x; Re tr Bx; Mx; :
p
Z0 e N
(2.3.31)
Bx;
an now be determined by maximizing with respe
t to Bx; and taking into
a
ount Eq. (2.3.29).
It is easy to see that if Bx; = I is a solution as before, then (2.3.28) is also
a solution. Therefore, we get
ij
U
x; 0
= m
d
x+a^ d
x
x+a^
yx = 0 ;
(2.3.32)
138
CHAPTER 2.
r.g.
b)
Fig. 2.17: Latti
e renormalization group transformation (2.3.33). The thin lines of
the old latti
e a) represent links on whi
h integration is performed. The
new latti
e b) has the latti
e spa
ing 2a but the same spatial extent La.
a =r.g.
) 2a ;
(2.3.33)
whi
h is depi
ted in Fig. 2.17b. The spa
e size of the latti
e is L before the
transformation and be
omes L=2 after the transformation,
L
;
(2.3.34)
2
so that the latti
e extent is L a in both
ases whi
h is expe
ted to redu
e an
in
uen
e of nite size ee
ts on the transformation.
The Wilson a
tion on the latti
e of Fig. 2.17a be
omes a more general one
r.g.
L =)
139
LATTICE METHODS
=)
a)
2.3.
(2.3.35)
dx
B
x3=2
6
x
2 ln 2
p :
6
(2.3.37)
Here the ln 2 on the RHS is due to Eq. (2.3.33) and the relation (2.2.32)
between and g2 is used with N
= 3.
For the pure SU(3) gauge theory, we get from Eq. (2.3.37)
= 0; 579 +
1
0; 204
+O 2
(2.3.38)
140
CHAPTER 2.
2.3.
141
LATTICE METHODS
e S[U =
fr dr r (U )
(2.3.39)
where
dr = r (I )
(2.3.40)
a =) a ;
(2.3.41)
by the formula
0 0
e S [U =
"
fr dr r (U 0 )
2
# d 2
(2.3.42)
Fig. 2.18: Monte Carlo data by Akemi et al. [Ake93 for . The errorbars represent
statisti
al errors. The solid line represents the asymptoti
s (2.3.38).
C =
ij
U
x; ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : :
: (2.3.43)
There are as many positions in this row as the multipli
ity of the integral.
Then the sequential integral
an be represented as
Z Y
x;
dUx; : : : =
::: :
(2.3.44)
142
CHAPTER 2.
h F (C ) i
e S(C )F (C )
CP
e S(C )
(2.3.45)
where S (C ) and F (C ) are the values of F and S for the given
onguration
C.
The task of Monte Carlo
al
ulations is not to sum over all possible
ongurations whose number is innite but rather to
onstru
t an ensemble, say,
of N
ongurations
E =
fC1 ; : : : ; CN g
(2.3.46)
(2.3.47)
h F [U i
N
1 X
F (Cn )
N n=1
(2.3.48)
be ause ea h onguration \weighs" already as mu h as is required. In parti ular, the Wilson loop average for a re tangular ontour is given by
WRT =
N
1
1 X
tr URT (Cn ) :
N n=1 N
(2.3.49)
2.3.
143
LATTICE METHODS
P (Cn
! Cn )
= P (Cn 1 ; Cn ) :
(2.3.50)
P 0 (C ) =
C0
P (C; C 0 ) P (C 0 )
(2.3.53)
P (C; C 0 ) =
C0
P (C; C 0 ) = 1
(2.3.54)
144
CHAPTER 2.
It is now easy to prove the statement 1). Summing the detailed balan
e
ondition (2.3.51) over C 0 , we get
X
P (C; C 0 )P (C 0 ) ;
(2.3.55)
P (C ) =
Bol
Bol
C0
0
i.e. the Boltzmann distribution is an eigenve
tor of P (C; C ). Comparing with
Eq. (2.3.53), we see that the new distribution is again the Boltzmann one, whi
h
proves 1).
To prove the statement 2), let us
ompare the distan
es from E and E 0 to some
equilibrium ensemble E Bol asso
iated with the Boltzmann distribution (2.3.47).
We have the inequality
X
P 0 (C ) P
kE 0 E k =
(C )
Bol
Bol
X X
P (C; C 0 ) P (C 0 )
C C0
X
P (C; C 0 ) P (C 0 ) P
X
P (C 0 )
kE E Bolk ;
Bol
CC 0
C0
P Bol (C 0 )
(C 0 )
P Bol (C 0 )
(2.3.56)
where Eqs. (2.3.53), (2.3.55) and (2.3.54) are used. Thus, 2) is proven.
Spe
i
Monte Carlo algorithms dier in the
hoi
e of the transition probability P (C; C 0 ) while the detailed balan
e
ondition (2.3.51) is always satised. The two most popular algorithms, whi
h a
t at one link, are
0 is sele
ted randomly from
Heat bath algorithm: A new link variable Ux;
the group manifold with a probability given by the Boltzmann fa
tor
P U0
/ e S(C 0 ) :
(2.3.57)
x;
Then this pro
edure is repeated for the next link and so on until the
whole latti
e is passed. This
an be imagined as if a reservoir with a
temperature 1= tou
hes ea
h link of the latti
e in su
ession. It is
lear from physi
al intuition that the system will be brought to the
thermodynami
equilibrium sooner or later.
2.3.
LATTICE METHODS
145
146
CHAPTER 2.
2.3.
LATTICE METHODS
147
Fig. 2.19: Monte Carlo data by Creutz [Cre79 for the string tension in the SU(2)
pure latti
e gauge theory.
surfa
es for the intera
tion between stati
quarks. In the strong
oupling
region g2 ! 1, they appear as in Fig. 2.20a sin
e the intera
tion potential
is given by
E (x; y; z ) = K ( jxj + jyj + jz j )
(2.3.60)
be
ause the distan
e between the quarks is measured along the latti
e. This
is asso
iated with a
ubi
symmetry on the latti
e (i.e. rotations through
an angle whi
h is a multiple of =2 around ea
h axis and translations by a
multiple of the latti
e spa
ing along ea
h axis) rather than with the Poin
are
group. The rotational symmetry must be restored in the
ontinuum limit.
The Monte Carlo data by Land and Rebbi [LR82 are shown in
Figs. 2.20b,
. They demonstrate the restoration of rotational symmetry
when passing from = 2 (Fig. 2.20b) to = 2:25 (Fig. 2.20
).
The old Monte Carlo
al
ulations played an important role for developing
the method. A dramati
improvement of the Monte Carlo te
hnology in
latti
e gauge theories happened during the last ten years.
As an example of modern Monte Carlo
al
ulations, the Monte Carlo
data by Bali and S
hilling [BS92 for the intera
tion potential in the SU(3)
Fig. 2.20: Behavior of equipotential lines at dierent values of : a) the strong ou
148
CHAPTER 2.
2.4.
149
FERMIONS ON A LATTICE
Fig. 2.21: Monte Carlo data [BS92 for the
ontinuum intera
tion potential E (R).
The solid
urve is a sum of linear and Coulomb potentials.
pure latti
e gauge theory are shown in Fig. 2.21. The solid
urve represents
how the data are des
ribed by a sum of linear and Coulomb potentials. The
maximal latti
e size was 324 and the spatial extent of the latti
e was up to
3.3 fm.
The Pro
eeding of the Latti
e Conferen
e [Lat94 are very useful, as is
already mentioned, for a today's look at the subje
t.
It turned out to be most di
ult in the latti
e approa
h to QCD to deal with
fermions. Putting fermions on a latti
e is an ambiguous pro
edure sin
e the
ubi
symmetry of a latti
e is less restri
tive than the
ontinuous Lorentz
group.
The simplest
hiralinvariant formulations of latti
e fermions lead to a
doubling of fermioni
degrees of freedom, as was rst noted by Wilson [Wil75,
and des
ribe from 16 to 4 relativisti
ontinuum fermions, depending on the
formulation. One half of them has a positive axial
harge and the other
half has a negative one, so that the
hiral anomaly
an
els. There is a nogo theorem whi
h says that the fermioni
doubling is always present under
natural assumptions about a latti
e gauge theory.
A pra
ti
al way out is to
hoose the fermioni
latti
e a
tion to be expli
itly
noninvariant under the
hiral transformation and to have, by tuning a mass
of the latti
e fermion, 1 relativisti
fermion in the
ontinuum and the masses
of the doublers to be of the order of the inverse latti
e spa
ing. The
hiral
anomaly is re
overed in this way.
We
onsider in this Se
tion various formulations of latti
e fermions and
the doubling problem. We dis
uss brie
y the results on spontaneous breaking
of
hiral symmetry in QCD.
x+a^ Ux; x :
(2.4.2)
14 A standard formula diers from this one by an inter
hange of U and U y due to the
inverse ordering of matri
es in the phase fa
tors (see the footnote on p. 91). It does not
matter how to dene Ux; sin
e the Haar measure is invariant under Hermitean
onjugation.
150
CHAPTER 2.
The rst term on the RHS is the pure gauge latti
e a
tion (2.2.16). The
se
ond term is a quark mass term on a latti
e. The sum in the third term is
over all latti
e links (i.e. over all sites x and positive dire
tions ). This a
tion
is Hermitean and invariant under the latti
e gauge transformation (2.2.13),
(2.4.1) at a nite latti
e spa
ing.
The partition fun
tion of latti
e QCD with fermions is dened by
Z (; M ) =
Z Y
x;
dUx;
d x d x e S[U; ; ;
(2.4.3)
F U; ;
Z
1 (; M )
Z Y
x;
dUx;
d x d x e
S [U; ;
F U; ;
; (2.4.4)
whi
h extends Eq. (2.2.39) to the
ase of fermions. Sin
e both the a
tion
and the measure in Eq. (2.4.4) are gauge invariant at nite
latti
e spa
ing, a
nonvanishing result is only when the integrand, F U; ; , is gauge invariant
as well.
In order to show how the latti
e a
tion (2.4.2) reprodu
es (2.1.14) in the
naive
ontinuum limit a ! 0, let us assume that the latti
e quark eld x
slowly varies from site to site and substitute
x
x+a^
! a3=2 (x) ;
! a3=2 ( (x) + a (x))
(2.4.5)
! a3
5
+ a4 rfun
+ O a ;
(2.4.6)
151
FERMIONS ON A LATTICE
The fermioni
latti
e a
tion (2.4.2) was proposed in Ref. [Wil74. For
M = 0 it is invariant under the global
hiral transformation
x
:t:
e i 5 x ;
t: i
5
x
:!
:
xe
(2.4.7)
For this reason, these latti
e fermions are
alled
hiral fermions. Sin
e the
latti
e a
tion is both gauge and
hiral invariant, there is no Adler{Bell{Ja
kiw
anomaly a
ording to the general arguments of Se
t. 1.3.
Problem 2.23
where the a
tion is given by Eq. (2.4.2). The integration over Ux; is as in
Eq. (2.2.31), and the integral over the quark eld is the Grassmann one. The
averages are dened by
2.4.
! i 4 5 ( 1)t=a x :
(2.4.8)
(2.4.9)
T = i 5 ( 1)x =a :
(2.4.10)
The other generators are given by produ
ts of (2.4.10). Their expli
it form
is [KS81a
= a5=2
xe
ikx :
(2.4.12)
152
CHAPTER 2.
S0 ;
=a
Z
=a
d4 k 1
k G (k ) k
(2)4
(2.4.13)
with
G 1 (k) =
4
1X
i
sin k a
a =1
(2.4.14)
for M = 0.
In the naive
ontinuum limit, the sin in Eq. (2.4.14)
an be expanded in
the power series in a, whi
h results in the free (inverse)
ontinuum propagator
G 1 (k)
! i
4
X
=1
k = i k^ :
(2.4.15)
3
X
=1
sin2 p a :
(2.4.16)
Let us look for solutions of Eq. (2.4.16) with positive energy E > 0
(solutions with negative energy are asso
iated as usual with antiparti
les).
Suppose that a parti
le moves along the z axis so that
omponents of the
fourmomentum
p(1) = (E; 0; 0; pz )
(2.4.17)
sinh Ea = sin pz a ;
(2.4.18)
are related by
2.4.
153
FERMIONS ON A LATTICE
whi
h follows from the substitution of (2.4.17) into the dispersion law.
Sin
e the sin is a periodi
fun
tion, the fourve
tor
p(2) = E; 0; 0;
(2.4.19)
pz
a
is also a solution of Eq. (2.4.18) if (2.4.17) is. Quite analogously, the fourve
tors
p(3) = E; ; 0; pz ;
a
: : : ;
p(8) = E; ; ;
pz ;
(2.4.20)
a a a
whi
h are obtained from p(1) and p(2) by
hanging zeros for =a, also satisfy Eq. (2.4.18). Therefore, a quark state with the energy E is eightfold
degenerate.
The quark states with the fourmomenta p(1); : : : ; p(8) are dierent states.
Their wave fun
tions equal
(j) (t; x; y; z )
(2.4.21)
The wave fun
tion in the state with the momentum p(3) diers, say, from the
wave fun
tion in the state p(1) by an extra fa
tor ( 1)x=a . In other words, it
strongly
hanges as a ! 0 with one step along the latti
e in the xdire
tion.
One more step returns the initial value.
For su
h fun
tions, the naive
ontinuum limit of the latti
e a
tion (2.4.2)
is as good as for the slowly varying fun
tions when Eq. (2.4.5) holds. In order
to see that, let us rewrite the a
tion (2.4.2) as
X
x x
S U; ;
= SLat [U + M
x
1 X h y
+
U
2 x;>0 x x; x+a^
Ux;y x a^
i
(2.4.22)
Even if x has opposite signs at neighboring latti
e sites along the axis, as
is illustrated by Fig. 2.22a, i.e.
(2.4.23)
x+a^ !
x;
then the dieren
e x+a^
x a^ on the RHS of Eq. (2.4.22) is still of the
orre
t order in a:
3=2 (x) ;
2a5=2 (x) ; (2.4.24)
x ! a
x+a^
x a^ !
154
CHAPTER 2.
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
+r
a)
so that the
ontinuum fermioni
a
tion is reprodu
ed ex
ept for the sign of
the
matrix whi
h is opposite to that in Eq. (2.1.14).
This extra minus sign
an be absorbed in the redenition of the
ontinuum fermioni
eld (x) ! i
5 (x), whi
h
hanges its
hirality. Therefore,
the axial
harge of the doublers is opposite. Analogously, four of the eight
doublers have positive axial
harge and four others have negative one dependently on whether the sign of x alters at neighboring sites along even or odd
number of axes (see Fig. 2.22). In Eu
lidean spa
e the doubling o
urs also
along the temporal axis, so the number of doublers is equal to 2d = 16: 8 of
them with positive and 8 with negative axial
harge. This explains why the
hiral anomaly
an
els.
Problem 2.24 Cal
ulate the ve
tor and axial
harges of the doublers deriving the
ve
tor and axial
urrents on a latti
e.
Solution The ve
tor and axial
urrents on a latti
e
an be derived using a latti
e
analog of the Noether theorem. The invarian
e of the latti
e fermioni
a
tion under
i
x ! x e ix
(2.4.25)
x ! e x x;
(2.4.26)
(Vx;
(2.4.27)
>0
Vx a^; ) = 0 :
155
FERMIONS ON A LATTICE
Ux;y x a^ + M x w.s.
= ;
x
(2.4.28)
! e ix 5
x;
x ! x e ix 5
(2.4.29)
b)
2.4.
y x+a^ + x+a^
5 Ux; x ;
Ax; = i x
5 Ux;
2
(2.4.30)
It is worth noting that the mass term in Eq. (2.4.2) is not
5 invariant
but does not remove the fermioni
doubling.
One might think of removing the doubling problem by modifying the
expression for the inverse latti
e propagator G 1 (k) in the free fermioni
latti
e a
tion (2.4.13), for instan
e, by adding nexttoneighbor terms. It is
easy to see that it does not help if the fun
tion G 1 (k) is periodi
as it should
be on a latti
e. A typi
al form of G 1 (k) as a fun
tion of, say, k4 is depi
ted
in Fig. 2.23. The behavior around k4 = 0 is pres
ribed by Eq. (2.4.15) and is
just a straight line with a positive slop. Therefore, G 1 (k) will have inevitably
another zero at k4 = =a due to periodi
ity.
This is the dieren
e between the fermioni
and bosoni
ases. For bosons
G 1 (k) is quadrati
in k4 near k4 = 0 rather than linear as for fermions. A
typi
al behavior of G 1 (k) for bosons is shown in Fig. 2.24. There is no
doubling of states in the bosoni
ase.
156
CHAPTER 2.
6G
2.4.
6G
a
157
FERMIONS ON A LATTICE
+ a
k4
1 for the hiral latti e fermions. The periodi ity leads to an extra zero at k4 = =a.
+ a
a
k4
Let us substitute
(2.4.33)
or expli itly
x; = 1
for = 1 ;
x
x; = ( 1) 1 =a
for = 2 ;
::: ;
x; = ( 1)(x1 +:::+xd 1 )=a
for = d ;
(2.4.34)
does not depend on spinor indi
es.
The idea is to leave only one
omponent of ix in order to redu
e the
degenera
y:
1
0
1
0
x
x
B 0 C
B 0 C
C
C
ix = B
ix = B
(2.4.35)
0 A;
0 A:
0
0
158
CHAPTER 2.
These latti
e fermions are known as the staggered fermions, sin
e x; is
staggering from one latti
e site to another. They are more often
alled the
Kogut{Susskind fermions due to their relation to those of Refs. [KS75, Sus77.
The a
tion of the Kogut{Susskind fermions is
S U; ;
+
= SLat [U + M
1 X
U y
2 x;>0 x; x x; x+a^
2d=2
x x
x+a^ Ux; x :
(2.4.36)
2d=2 1
2.4.
159
FERMIONS ON A LATTICE
4
:
(2.4.38)
a
Therefore, the Wilson latti
e fermions des
ribe a relativisti
fermion of the
mass m in the
ontinuum when
m =
M ! 4 + ma :
(2.4.39)
It des
ribes
= 4 spe
ies for
omplex x or
= 2 spe
ies for Majorana x. Components of a
ontinuum bispinor are distributed in this approa
h
over four latti
e sites.
There is no axial anomaly for the Kogut{Susskind fermions as with the
hiral fermions.
In order to see that there are no other relativisti
fermion states in the
limit (2.4.39), let us
onsider the fermioni
propagator whi
h is given by
S U; ;
= SLat [U + M
x x
1
y x+a^ + x+a^ (1 +
) Ux; x : (2.4.37)
(1
) Ux;
2 x;>0 x
X
G 1 (k ) = M
osh Ea =
3
P
1+ M
=1
2 M
os p a
3
P
=1
!2
=1
sin2 p a
os p a
(2.4.41)
E =
The dieren
e between this a
tion and the a
tion (2.4.2) for
hiral fermions
is due to the proje
tor operators (1
) whi
h pi
k only one fermioni
state.
4
1X
(1
) e ik a + (1 +
) e ik a : (2.4.40)
2 =1
~p2 + m2
(2.4.42)
M 4.
160
CHAPTER 2.
2.4.
161
FERMIONS ON A LATTICE
For M 4, we
an repla
e the LHS of Eq. (2.4.41) by 1 and substitute M = 4 on the RHS. Then Eq. (2.4.41) redu
es to the equation for spatial
omponents of the fourmomentum:
Solution
3
X
=1
!2
os p a
+ 3
3
X
=1
os2 p a
= 0;
(2.4.43)
It is instru
tive to dis
uss what happens with the fermion doublers under
the
hange of
by (1
) in the latti
e fermioni
a
tion. Let us
onsider
one of su
h states, say, that with p1 = =a, p2 = p3 = 0. Its energy is
determined by Eq. (2.4.41) to be 1=a so that this state is inessential as
a ! 0.
The
hiral anomaly is
orre
tly re
overed by the Wilson fermions. The 15
states of the mass 1=a play thereat a role of regulators whi
h result in the
anomaly as a ! 0.
Cal
ulate the masses of all 16 fermioni
states.
Substituting Eqs. (2.4.23), (2.4.24) and so on into the a
tion (2.4.37),
Problem 2.26
we get
m =
where
s = e ip a
M
P4
=1 s
+1 p = 0
1 p = =a :
(2.4.44)
(2.4.45)
P =
whose only solution is p1 = p2 = p3 = 0 sin e both terms on the LHS are nonnegative.
Solution
the negative ones. In parti
ular, there are no ba
ktra
kings in the (latti
e)
sum over paths sin
e
P+ P = 0 :
(2.4.47)
This removes the doubling.
Problem 2.27 Represent the fermion propagator in an external Yang{Mills eld
as a sum over paths on a latti
e, performing an expansion in 1=M .
Solution Let us res
ale the fermion eld, absorbing the parameter M in front of
the mass term. The fermioni
part of the new a
tion reads
X
X
y x+a^ + x+a^ P + Ux; x ; (2.4.48)
x x
x P Ux;
S =
x;>0
where = 1=M is usually
alled the hopping parameter. The large mass expansion
in 1=M is now represented as the hopping parameter expansion in .
It is
onvenient to depi
t ea
h of the two terms in the square bra
kets in
Eq. (2.4.48) by a string bit as in Fig. 2.4 with the quark elds at the ends and
the gauge variable at the link. The rst term
orresponds to the negative dire
tion
of the link, and the se
ond term
orresponds to the positive dire
tion. Substituting Eq. (2.4.48) into the denition (2.4.4) and expanding the exponential in ,
we get a
ombination of terms
onstru
ted from the string bits. A nonvanishing
ontribution to the quark propagator
i
j
Gijmn (x; y ; U ) =
(2.4.49)
m (x) n (y ) ;
where i; j and m; n represent, respe
tively,
olor and spinor indi
es, emerges when
the links form a path yx that
onne
ts x and y on the latti
e as is depi
ted
in Fig. 2.25. Otherwise, the average over and vanishes due to the rules of
integration over Grassmann variables des
ribed in Problem (1.17).
Therefore, we get
Gijmn (x; y ; U ) =
X
yx
1
U ij [ yx 4
M L( )+1
Y
yx
P 5
;
mn
(2.4.50)
162
CHAPTER 2.
where P+ or P are asso
iated with the positive or negative dire
tion of a given
link 2 yx . For the Wilson fermions, they are given by Eq. (2.4.46), while
P =
hiral fermions
(2.4.51)
2
for
hiral fermions. The sum in Eq. (2.4.50) runs over all the paths between x and y
on the latti
e, while L( ) stands for the length of the path yx in the latti
e units.
Eq. (2.4.50) is a latti
e analog of the
ontinuum formula (2.1.43) for the
ase of
fermions.
Represent the integral over fermions in Eq. (2.4.3) as a sum over
losed paths on a latti
e, performing an expansion in 1=M .
Solution The
al
ulation is analogous to that of the previous problem. The result
reads
Problem 2.28
Z Y
d x d x e S = e Sind [U
(2.4.52)
2.4.
am
!!!
!
!
!!

M
Sind [U =
tr U [
sp
L( ) M L( )
P ;
(2.4.53)
where the
ombinatori
fa
tor 1=L( ) is due to identity of L links forming the
losed
ontour , and the minus sign is be
ause of fermions.
Equation (2.4.53) denes an ee
tive a
tion of a pure latti
e gauge theory,
whi
h is nonlo
al sin
e involves arbitrary large loops. However, it
an be made
the singleplaquette latti
e a
tion (2.2.16) by introdu
ing many
avors of latti
e
fermions [Ban83, Ham83.
Fig. 2.26: Dependen e of the meson mass on the latti e quark mass M . At
For the
hiral or Kogut{Susskind fermions with M = 0, the latti
e a
tion is invariant under the global
hiral transformation (2.4.7). The order
parameter for breaking the
hiral symmetry is
with
163
FERMIONS ON A LATTICE
:t:
e 2i 5 ;
(2.4.54)
whi
h is not invariant under the
hiral transformation. Therefore, the average
of must vanish if the symmetry is not broken spontaneously15. Su
h
spontaneous breaking results in
6= 0 :
(2.4.55)
This nonvanishing value of the average of x x does not depend on x due to
translational invarian
e and is
alled the quark
ondensate.
The spontaneous breaking of the
hiral symmetry in QCD was demonstrated by the Monte Carlo
al
ulations of the quark
ondensate. This quantity has a dimension of [mass3 and should depend on g2 at small g2 as is
pres
ribed by the asymptoti
s
aling. The Monte Carlo data for the quark
ondensate from the pioneering paper by Hamber and Parisi [HP81 are shown
in Fig. 2.27. Its agreement with asymptoti
s
aling demonstrates that the
hiral symmetry is spontaneously broken in the
ontinuum QCD.
164
CHAPTER 2.
2.5.
165
FINITE TEMPERATURES
Fig. 2.27: Monte Carlo data by Hamber and Parisi [HP81 for the quark
ondensate
in the quen
hed approximation.
left with the determinant, say for the Kogut{Susskind fermions, of the matrix
D [U = M x;y +
1 X
Uy
2 >0 x; x; x(x+a^)
for a given
onguration of the gluon eld Ux;. This results in a pure gaugeeld problem with the ee
tive a
tion given by
e Seff [U = det D [U e S[U :
(2.4.57)
The matrix that appears in this determinant has at least N
L4 N
L4 elements, and is to be
al
ulated at ea
h Monte Carlo upgrading of Ux; .
Several methods are proposed to manage the quark determinant approximately. The simplest one is not to take it into a
ount at all. This approximation is known as the quen
hed approximation when only valen
e quarks are
onsidered, while the ee
ts of virtual quark loops are disregarded. Re
ently
a progress in the full theory has been a
hieved using some tri
ks for evaluating
the quark determinants (see Ref. [Lat94 for a review of the subje
t).
Z (T; V ) =
e En =T
Sp e H=T
(2.5.1)
166
CHAPTER 2.
hgj ~xi
h~x jf i
= g (~x) ;
= f (~x) ;
(2.5.2)
hgj e
H=T
jf i
D' (~x; t) e
R 1=T
(2.5.3)
'(x~;0)=f (~x)
'(~x;1=T )=g(~x)
L ['
dt L['
d3~x
1
1
( ')2 + m2 '2 + '3
2
2
3!
Sp e H=T =
=
Df (~x) hf j e H=T jf i
Z
D' (~x; t) e
real time for a system at equilibrium. The variable t in Eq. (2.5.5) is just
the proper time of the disentangling pro
edure. This analogy between the
partition fun
tions of statisti
al systems and the Eu
lidean formulation of
quantum eld theory is already mentioned in the Remark to Subse
t. 1.1.7.
Derive the Feynman{Ka
formula for a quantum parti
le with nonrelativisti
Hamiltonian (1.1.103).
Solution The matrix element hx jexp ( H=T )j xi is determined by Eq. (1.1.114)
to be
Z
R 1=T
H=T
x
xe
=
(2.5.7)
Dz (t) e 0 dt L(t) ;
Problem 2.29
z (0)=x
z (1=T )=x
(2.5.4)
dt L['
(2.5.5)
'(~x;1=T )='(~x;0)
Noti
e that the path integral in Eq. (2.5.5) is taken with periodi
boundary
onditions for the eld ':
(2.5.6)
dd x x e H=T x
Dz (t) e
R 1=T
0
dt L(t)
(2.5.8)
z (0)=z (1=T )
This integral is over the traje tories with periodi boundary onditions
z (0) = z (1=T ) :
(2.5.9)
Cal
ulate the partition fun
tion (2.5.8) for the free
ase.
The Gaussian path integral with the boundary
onditions
Problem 2.30
Solution
R 1=T
167
FINITE TEMPERATURES
Sp e H=T =
for the
ubi
selfintera
tion of '. The derivation is quite analogous to that
of Problem 1.8.
In order to
al
ulate the tra
e over states, one should put g(~x) = f (~x) and
perform the additional integration over f (~x). This yields the Feynman{Ka
formula16
Z
2.5.
z (0) = z (1=T ) = x
(2.5.10)
is
al
ulated in Subse
t. 1.1.5 with the result given by Eq. (1.1.86). In order to
al
ulate the partition fun
tion (2.5.8), we need to integrate this expression over x
whi
h yields [Fey53
Z (T; V ) =
dd x F (1=mT ) = V
mT d=2 :
2
(2.5.11)
The formula (2.5.11) is to be
ompared with that given by the Boltzman distribution in
lassi
al statisti
s. Sin
e the energy of a free nonrelativisti
parti
le
is
2
E (p~) = ~p ;
(2.5.12)
2m
168
CHAPTER 2.
the Boltzmann distribution is given by the sum over positions of the parti
le in a
box of volume V and the integration over its momentum p~:
Z
dd p~ e E (p~)=T = V mT d=2 ;
Z (T; V ) = V
(2.5.13)
(2 )d
2
whi
h
oin
ides with Eq. (2.5.11) derived from the path integral.
Cal
ulate the partition fun
tion (2.5.5) for the free
ase.
Sin
e the path integral over '(~x; t) is Gaussian, it
an be represented as
1
ln det 2 + m2
ln Z (T; V ) =
2
1
=
Sp ln 2 + m2
2 Z
dd p~ Sp ln D2 + ! 2 ;
1
V
(2.5.14)
=
2
(2 )d t
Problem 2.31
Solution
where
(2.5.15)
! = ~p2 + m2 :
We have used the fa
t that the ~x variable is not restri
ted while the remaining
tra
e of the onedimensional operator is to be
al
ulated with periodi
boundary
onditions.
We shall perform the
al
ulation by expressing the tra
e via the diagonal resolvent of the same operator as it was already done in the Problem 1.29. The
Green fun
tion G! (t t0 ) is no longer given by Eq. (1.1.38) be
ause of the periodi
boundary
onditions. We get, instead, the sum over even frequen
ies:
+1
0
X
e i2nT (t t)
G! (t t0 ) = T
;
(2.5.16)
(2nT )2 + ! 2
n= 1
whi
h satises G! (1=T ) = G! (0), as it should be for periodi
boundary
onditions,
and reprodu
es Eq. (1.1.38) as T ! 0. The diagonal resolvent is given by
+1
X
1
!
1
(2.5.17)
G! (0) = T
2 + ! 2 = 2!
oth 2T :
(2
nT
)
n= 1
Therefore,
D2 + !
Spt ln
=
Z!
2
Z!
d! 2
1=T
Z
0
e !=T
(2.5.18)
169
FINITE TEMPERATURES
dd p~ h ! + ln 1
(2 )d 2T
e !=T ;
(2.5.19)
whi
h is the standard result for an ideal Bose gas in quantum statisti
s modulo the
rst term on the RHS asso
iated with the zeropoint energy of the va
uum.
Z (T; V ) =
DA D D e
R 1=T
0
dt V d3 ~x L[A ; ;
;
(2.5.20)
whi
h is the proper analog of Eq. (2.5.5). The path integral is taken with the
boundary
onditions
(2.5.21)
(2.5.22)
(2.5.23)
whi
h are periodi
for the gauge eld (gluon) and antiperiodi
for the Fermi
elds (quarks). The antiperiodi
ity of the Fermi elds is related, roughly
speaking, with the famous extra minus sign of fermioni
loops in the va
uum
energy.
Cal
ulate the partition fun
tion for free massive onedimensional
fermions with antiperiodi
boundary
onditions
(1=T ) =
(0) :
(1=T ) =
(0) ;
(2.5.24)
Problem 2.32
Solution
ln Z (T; V ) = ln det (D + m) = Sp ln (D + m) :
(2.5.25)
The fermion Green fun
tion Gm (t t0 ) is given by the sum over odd frequen
ies:
Gm (t t0 ) = T
dt G! (0)
!
!
1
= + 2 ln 1
d!
oth
T
2T
T
2.5.
+1
X
e i(2n+1)T (t t)
;
i(2n + 1)T + m
n= 1
(2.5.26)
170
As T
CHAPTER 2.
! 0, we get
Gm (t t0 ) =
+1
Z
0
d e i(t t)
= t t0
2 i + m
(2.5.27)
sin
e the
ontour of integration over
an be
losed for t > t0 (t < t0 ) in the lower
(upper) halfplane. We have thus reprodu
ed the fermioni
Green fun
tion (2.1.35)
from Problem 2.3.
The diagonal resolvent is given by
+1
X
1
1
m
Gm (0) = T
= tanh ;
(2.5.28)
i
(2
n
+
1)
T
+
m
2
2
T
n= 1
whi
h diers from Eq. (2.5.17) by the
hange of the
oth for the tanh. Therefore,
Zm
(2.5.29)
dm 1 tanh m = m + ln 1 + e m=T
T
2T
2T
modulo an mindependent
onstant. The se
ond term on the RHS involves the
plus sign whi
h
hara
terizes the Fermi statisti
s (remember that ! = m if there is
no spatial dimensions). If we were
hoose periodi
boundary
onditions instead of
antiperiodi
ones, we would get the minus sign as in Eq. (2.5.19) whi
h is wrong for
fermions. The rst term on the RHS is asso
iated again with the zeropoint energy
of the va
uum.
ln Z (T; V ) =
The dis
ussion of the previous Subse
tion about the relation between the
nitetemperature and Eu
lidean formulations explains why the latter allows
to
al
ulate in QCD only stati
quantities, say hadron masses or intera
tion
potentials, whi
h do not depend on the time. It is also worth noting that
we did not add a gauge xing term in Eq. (2.5.20) having in mind a latti
e
quantization as before.
The latti
e formulation of nitetemperature QCD is espe
ially simple.
One should take an asymmetri
latti
e whose size along the temporal axis is
mu
h smaller than that along the spatial ones:
1
Lt =
L:
(2.5.30)
Ta
This guarantees the system to be in a thermodynami
limit. Then the temperature is given by
1
;
(2.5.31)
T =
aLt
2.5.
171
FINITE TEMPERATURES
i.e.
oin
ide with the inverse extent of the latti
e along the temporal axis.
The periodi
boundary
onditions are usually imposed on the latti
e by
onstru
tion.
Sin
e the latti
e spa
ing a and the bare
oupling
onstant g2 are related
by Eq. (2.2.85), the temperature (2.5.31)
an be rewritten as
T =
1
exp
Lt QCD
Z
dg2
B(g2) :
(2.5.32)
Therefore, one
an
hange the temperature on the latti
e varying either the
size along temporal axis Lt or g2 .
L(~x) = tr P e
R 1=T
0
dt Ad (~x;t)
(2.5.33)
It is gauge invariant be
ause of the periodi
boundary
onditions for the gauge
eld and is
alled the Polyakov loop or the Wilson line. One
an imagine that
the timevariable t xd is
ompa
tied so that the Polyakov loop winds
around the torus in the temporal dire
tion as is shown in Fig. 2.28.
The latti
e Polyakov loop
L~x = tr
xd
Ux;d
(2.5.34)
is just the tra
e of the produ
t of the link variables along a line whi
h goes in
the temporal dire
tion through the latti
e with imposed periodi
boundary
onditions.
Using the latti
e gauge transformation (2.2.13), almost all link variables,
asso
iated with links pointing in the temporal dire
tion,
an be put equal 1
ex
ept for one time sli
e sin
e the gauge transformation is periodi
:
(~x; 0) =
(~x; 1=T ) :
(2.5.35)
172
CHAPTER 2.
A Polyakov A
loop
? t 6*~x
2.5.
While the latti
e a
tion is invariant under the transformation (2.5.39), the
Polyakov loop transforms as
L~x
Z =
Fig. 2.28: Polyakov loop whi
h winds around
ompa
tied temporal dire
tion.
The average of the Polyakov loop is related to the free energy F0 (~x) of
a single quark (minus that of the va
uum) nailed at the point ~x of a threedimensional spa
e by
= e F0 =T :
(2.5.36)
hL(~x)i
= 0
onnement :
de onnement
(2.5.37)
(2.5.38)
is asso
iated with de
onnement. This is the Polyakov
riterion of
onnement at nite temperature.
This
riterion establishes a
onne
tion on a latti
e between
onnement
and the Z(3) symmetry  the
enter of the SU(3). The Z(3)transformation
of the link variables
Ux;d
! Zxd Ux;d
( Zxd 2 Z(3) )
( Z 2 Z(3) ) ;
(2.5.40)
(2.5.39)
xd
Zxd :
(2.5.41)
Therefore, Eq. (2.5.37) holds if the symmetry is unbroken, while Eq. (2.5.38)
signalizes about spontaneous breaking of the symmetry. Thus,
onnement
or de
onnement are asso
iated with the unbroken or broken global Z(3)
symmetry, respe
tively.
On a latti
e of a nite volume, the number of degrees of freedom is nite
and spontaneous breaking of the Z(3) symmetry is impossible. It is more
onvenient to use then the
riterion whi
h is based on the
orrelator of two
Polyakov loops separated by a distan
e R along a spatial dire
tion. This
orrelator determines the intera
tion energy E (R) between a quark and an
antiquark by the formula
On the ontrary,
hL(~x)i 6=
! Z L~x
where
hL(~x)i
173
FINITE TEMPERATURES
L(~x)Ly (~y)
onn = e E(R)=T :
(2.5.42)
A nite
orrelation length is now asso
iated with
onnement while an innite
one
orresponds to de
onned quarks.
More about the Z(3) symmetry in nitetemperature latti
e gauge theories
an be found in the review by Svetitsky [Sve86.
Cal
ulate the
orrelator (2.5.42) to the leading order of the strong
oupling expansion.
Solution The
al
ulation is analogous to that of Subse
t. 2.2.5. The group integral
is nonvanishing when the plaquettes
ompletely ll a
ylinder, spanned by two
Polyakov loop, whose area is R=T . This is an analog of the lling shown in Fig. 2.10.
Contra
ting the indi
es, we get
Problem 2.33
L~x Ly~y
onn = (Wp )R=T ;
(2.5.43)
where Wp is given by Eq. (2.2.72). This yields the same intera
tion potential E (R)
as before (see Eqs. (2.2.76) and (2.2.77)).
174
CHAPTER 2.
g32d = g2 T ;
(2.5.44)
E (T ) = V1 (1=T ) ln Z (T; V ) ;
V
(2.5.45)
X
T
gi m3i K1 (mi =T ) + 3m2i K2 (mi =T ) ;
2
2 i=;;!;:::
(2.5.46)
FINITE TEMPERATURES
175
Derive Eq. (2.5.46) starting from the partition fun
tion (2.5.19).
For a dilute gas, the logarithm in Eq. (2.5.19)
an be expanded in
exp ( E=T ). Therefore, we get
Z
p2 2
onst
d3 p~
ln Z (T; V ) =
+V
e ~p +m =T
3
T
(2 )
onst V T m2
=
+
K (m=T ) :
(2.5.47)
T
2 2 2
The se
ond term on the RHS des
ribes the
lassi
al statisti
s of an ideal gas of
relativisti
parti
les. Equation (2.5.46)
an now be derived by dierentiating this
formula with respe
t to 1=T a
ording to Eq. (2.5.45) and taking into a
ount
statisti
al weights of the hadron states. The zeropoint energy term gives a T independent
ontribution to Eh , whi
h only
hanges the referen
e level of energy.
Problem 2.34
At high temperatures T ! 1, the temporal dire
tion shrinks and the partition fun
tion (2.5.20) redu
es to a threedimensional one with the
oupling
onstant
Eh (T )
2.5.
Solution
176
CHAPTER 2.
2.5.
6 E =T 4
(2.5.49)
The energy density E (T ) is des
ribed by a free gas of hadrons for low
temperatures, as is already mentioned, and by a free gas of quarks and gluons
at high temperatures. The latter statement is due to asymptoti
freedom,
whi
h says that the ee
tive
oupling
onstant des
ribing strong intera
tion
at the temperature T is given by
g2 (T )
b ln
QCD
T
(2.5.52)
with
2
(2.5.53)
11 + Nf
3
and Nf being the number of fermion spe
ies (or
avors) whose mass is mu
h
less than T . This formula has the same stru
ture as the running
onstant
g2 (Q) whi
h des
ribes strong intera
tion at the momenta Q. Sin
e Q T for
thermal
u
tuations, these two
oupling
onstants
oin
ide with logarithmi
a
ura
y.18
The energy density E (T ) of the quarkgluon plasma is given by Boltzmann
law
2
(2.5.54)
Ep (T ) = gp 30 T 4 + B
b =
18 The perturbative al ulations in QCD at nite temperature are des ribed in the book
by Kapusta [Kap89.
hadrons
gh
82
plasma
gp
(2.5.50)
1
;
(2.5.51)
T
= K S
l
when the
hanges of the energy and entropy
ompensate ea
h other, so that the free
energy
eases to depend on l. Therefore, the phase transition is asso
iated with
a
ondensation of arbitrary long strings.
177
FINITE TEMPERATURES
Fig. 2.29: Temperature dependen e of the energy density of the hadron matter. E (T )
for the hadron and plasma phases are given by Eqs. (2.5.46) and (2.5.54).
The dieren
e Ep Eh at the temperature T
of the de
onnement phase
transition is equal to the latent heat E .
where
gp = 2 8 +
7
2 2 3 Nf
8
(2.5.55)
178
CHAPTER 2.
Problem 2.36 Cal
ulate T
and the latent heat E approximating Eh by an ideal
gas of massless mesons.
Solution It is reasonable to disregard the mass of the mesons for T > 200 Mev.
Then
Eh (T ) = gh 30 T 4
(2.5.57)
and
2
Pp (T ) = gp 90 T 4 B :
(2.5.58)
The positive
onstant B in the energy density (2.5.54) leads to a negative pressure in the plasma state at low temperatures. Therefore, the hadron phase is preferable at low temperatures. This is in a spirit of the bag model of hadrons. At high
energies the pressure is higher for the plasma phase, sin
e
gp = 37 > gh = 3 ;
(2.5.59)
so that the plasma phase is realized. The pressure against T 4 is shown in Fig. 2.30
for both phases of the hadron matter.
The de
onnement phase transition o
urs when the pressures in both phases
oin
ide. Therefore, we get
T
4 = 2 B
90 (gp gh )
(2.5.60)
and
E Ep (T )
Eh (T ) = 4B :
(2.5.61)
If we were put gh = 0 in Eq. (2.5.60), this would
hange the value of T
by few
per
ents. This justies the approximation of massless mesons.
179
FINITE TEMPERATURES
(2.5.56)
with gh = 3 being due to the three isotopi
states ( + , , and ). Eh (T ) for the
plasma state is given by Eq. (2.5.54).
The pressure for the relativisti
gases with the energy densities (2.5.56) and
(2.5.54) is given, respe
tively, by
Ph (T ) = gh 90 T 4
2.5.
plasma
hadrons
T
4
T4
Fig. 2.30: Pressure against T 4 for the two phases of the hadron matter. The solid
and dashed lines represent Eqs. (2.5.57) and (2.5.58), respe
tively. The
hadron phase is stable for T < T
, while the plasma phase is stable for
T > T
.
is
appli
able at high T . This restoration o
urs
as a phase transition with being the proper order parameter. Therefore,
the quark
ondensate is destroyed at T = T
h. Monte Carlo simulations
indi
ates this phase transition to be of the rst order.
However, there is a subtlety in the above string pi
ture of quark
onnement when virtual quarks are taken into a
ount. The ee
ts of virtual
180
CHAPTER 2.
2.5.
Fig. 2.31: Breaking of the
ux tube by
reating a quarkantiquark pair (depi
ted by
quarks are suppressed when their mass m is innitely large and the pi
ture
of
onnement is the same as in pure gluodynami
s: quarks are permanently
onned by strings made of the
ux tubes of the gluon eld. This is asso
iated
with a linear intera
tion potential.
For light virtual quarks, the
ux tube
an break
reating a quarkantiquark pair out of the va
uum as is shown in Fig. 2.31. This happens
when the energy saved in the
ux tube is large enough to
ompensate the
kineti
energy of the produ
ed parti
les. Hen
e, the linear growth of the
potential will stop at su
h distan
es.
The average of the Polyakov loop (2.5.33) is no longer
riterion for quark
onnement in the presen
e of virtual quarks. The test stati
quark
an
always be s
reened by an antiquark
reated out of the va
uum (a quark
reated at the same time will go to innity). Therefore, the free energy F0 in
Eq. (2.5.36) is always nite so that hL(~x)i 6= 0 in both phases.
The ee
ts of virtual quarks usually weaken a phase transition in a pure
gauge theory. If the de
onning phase transition in the SU(3) pure gauge theory was of the se
ond order rather than of the rst order, it would presumably
disappear for an arbitrary large but nite value of m. Su
h a phenomenon
happens in the Ising model where an arbitrary small external magneti
eld
(whi
h is an analog of the quark mass) destroys the se
ond order phase transition. A dis
ontinuity of hL(~x)i at the rst order de
onning transition
ontinues in the T; mplane as is illustrated by Fig. 2.32. It seems to terminate
at some value m
of the quark mass.
This situation with the order parameter for the de
onnement phase
transition is somewhat similar to that for the
hiral phase transition. vanishes in the unbroken phase only for m = 0. If m 6= 0 but is small, there is a
small expli
it breaking of
hiral symmetry due to the quark mass. Sin
e the
hiral phase transition is
of the rst order for m = 0, it is natural to expe
t
that a dis
ontinuity of
ontinues in the T; mplane up to some value
m
h of the quark mass.
181
FINITE TEMPERATURES
Hadrons
Plasma
m h
0
0
T
h
T
Fig. 2.32: Expe ted phase diagram of the hadron matter in the T; mplane. The
If m
h < m
, the phase diagram in the T; mplane may look like that
shown in Fig. 2.32. In
the
intermediate region m
h < m < m
, the behavior
of neither hL(~x)i nor
an answer the question whether a phase transition
(or two separate transitions) o
urs. A proper parameter, whi
h signals about
a phase transition is this region,
ould be the temperature dependen
e of the
energy density E (T ) that undergoes dis
ontinuities at the points of rst order
phase transitions.
It is worth noting that an alternative behavior of the phase diagram in the
T; mplane, when m
h > m
, is not
onrmed by Monte Carlo simulations.
182
CHAPTER 2.
Referen
es to Chapter 2
[Ake93
184
[CGS86
[CJR79
[CJR83
[Cre79
[Cre80
[Cre83
[Dro81
[DZ83
[Eli75
[Fey53
[FLZ82
[GL81
[GN80
[Gri78
[Ham83
[HJS77
REFERENCES TO CHAPTER 2
REFERENCES TO CHAPTER 2
[HP81
185
186
[NN81
[Pol78
[Sei82
[STW81
[Sus77
[Sus79
[Sve86
[Weg71
[Wey19
[Wil74
[Wil75
[Wil77
[WK74
[Won70
[YM54
REFERENCES TO CHAPTER 2
, Cat h22
J. Heller
Chapter 3
1=N
Expansion
188
CHAPTER 3.
1=N
EXPANSION
dd x ^ + m
G
2
2
(3.1.1)
1 In d = 2 this model was studied in the largeN limit in Ref. [GN74 and is often alled
(N ) VECTOR MODELS
189
Here ^ = and
The simplest models, whi
h be
ome solvable in the limit of a large number of
eld
omponents, deal with a eld whi
h has N
omponents forming an O(N )
ve
tor in an internal symmetry spa
e. A model of this kind was rst
onsidered by Stanley [Sta68 in statisti
al me
hani
s and is known as the spheri
al
model. The extension to quantum eld theory was done by Wilson [Wil73
both for the fourFermi and '4 theories.
In the framework of perturbation theory, the fourFermi intera
tion is
renormalizable only in d = 2 dimensions and is nonrenormalizable for
d > 2. The 1=N expansion resums perturbationtheory diagrams after whi
h
the fourFermi intera
tion be
omes renormalizable to ea
h order in 1=N for
2 d < 4. An analogous expansion exists for the nonlinear O(N ) sigma
model. The '4 theory remains trivial in d = 4 to ea
h order of the 1=N expansion while has a nontrivial infraredstable xed point for 2 < d < 4.
The 1=N expansion of the ve
tor models is asso
iated with a resummation
of Feynman diagrams. A very simple
lass of diagrams  the bubble graphs
 survives to the leading order in 1=N . This is why the largeN limit of the
ve
tor models is solvable. Alternatively, the largeN solution is nothing but
a saddlepoint solution in the pathintegral approa
h. The existen
e of the
saddle point is due to the fa
t that N is large. This is to be distinguished from
a perturbationtheory saddle point whi
h is due to the fa
t that the
oupling
onstant is small. Taking into a
ount
u
tuations around the saddlepoint
results in the 1=N expansion of the ve
tor models.
We begin this Se
tion with a des
ription of the 1=N expansion of the
N 
omponent fourFermi theory analyzing the bubble graphs. Then we introdu
e fun
tional methods and
onstru
t the 1=N expansion of the O(N )symmetri
'4 theory and nonlinear sigma model. At the end we dis
uss the
fa
torization in the O(N ) ve
tor models at large N .
S ;
3.1.
= ( 1; : : : ; N )
(3.1.2)
N
X
i=1
i i :
(3.1.3)
(3.1.4)
For this reason, the perturbation theory for the fourFermi intera
tion is
renormalizable in d = 2 but is nonrenormalizable for d > 2 (and, in parti
ular, in d = 4). This is why the old Fermi theory of weak intera
tions was
repla
ed by the modern ele
troweak theory, where the intera
tion is mediated
by the W and Z bosons.
The a
tion (3.1.1)
an be equivalently rewritten as
S ; ; =
dd x
^ + m
2
+
;
2G
(3.1.5)
where is an auxiliary eld. The two forms of the a
tion, (3.1.1) and (3.1.5),
are equivalent due to the equation of motion whi
h reads in the operator
notation as
(3.1.6)
= G : :;
where : : : : : stands for the normal ordering of operators. Equation (3.1.6)
an
be derived by varying the a
tion (3.1.5) with respe
t to .
In the pathintegral quantization, where the partition fun
tion is dened
by
Z =
DD D e S[ ; ;
(3.1.7)
with S ; ; given by Eq. (3.1.5), the a
tion (3.1.1) appears after performing the Gaussian integral over . Therefore, one alternatively gets
Z =
D D e S[ ;
(3.1.8)
190
CHAPTER 3.
1=N
EXPANSION
with S ; given by Eq. (3.1.1).
The perturbative expansion of the O(N )symmetri
fourFermi theory
an
be
onveniently represented using the formulation (3.1.5) via the auxiliary
eld . Then the diagrams are of the type of those in Yukawa theory, and
resemble the ones for QED with and being an analog of the ele
tronpositron eld and being an analog of the photon eld. However, the auxiliary eld (x) does not propagate, sin
e it follows from the a
tion (3.1.5)
that
D0 (x y) h (x)(y) iGauss =
G (d)(x
y)
(3.1.9)
or
3.1.
(N ) VECTOR MODELS
I
6 6
I
I 
I
I
I

a)
b)
(3.1.10)
in momentum spa
e.
It is
onvenient to represent the fourFermi vertex as the sum of two terms
I
I
I
I
I
I

I
I
where the empty spa
e inside the vertex is asso
iated with the propagator (3.1.9) (or (3.1.10) in momentum spa
e). The relative minus sign makes
the vertex antisymmetri
in both in
oming and outgoing fermions as is pres
ribed by the Fermi statisti
s.
The diagrams that
ontribute to se
ond order in G for the fourFermi vertex are depi
ted, in these notations, in Fig. 3.1. The O(N ) indi
es propagate
through the solid lines so that the
losed line in the diagram in Fig. 3.1b
orresponds to the sum over the O(N ) indi
es whi
h results in a fa
tor of N .
Analogous oneloop diagrams for the propagator of the eld are depi
ted
in Fig. 3.2.
Cal
ulate the oneloop GellMann{Low fun
tion of the fourFermi
theory in d = 2.
Solution Evaluating the diagrams in Fig. 3.1 whi
h are logarithmi
ally divergent
in d = 2, and noting that the diagrams in Fig. 3.2 do not
ontribute to the wavefun
tion renormalization of the eld, whi
h emerges to the next order in G, one
gets
Problem 3.1
B (G) =
(N
1) G2
:
2
d)
(3.1.11)
(3.1.12)
191
I
I

e)
Fig. 3.1: Diagrams of the se
ond order of perturbation theory for the fourFermi
vertex. The diagram b) involves the sum over the O(N ) indi
es.
The fourFermi theory in 2 dimensions is asymptoti
ally free as was rst noted by
Anselm [Ans59 and redis
overed in Ref. [GN74.
The vanishing of the oneloop GellMann{Low fun
tion in the Gross{Neveu
model for N = 1 is related to the same phenomenon in the Thirring
model. The
2
latter model is asso
iated with the ve
torlike intera
tion
of one spe
ies
of fermions with
being the
matri
es in 2 dimensions. Sin
e a bispinor has
in d = 2 only two
omponents 1 and 2 , both the ve
torlike and the s
alarlike
intera
tion (3.1.1) for N = 1 redu
e to 1 1 2 2 sin
e the square of a Grassmann
variable vanishes. Therefore, these two models
oin
ide. For the Thirring model, the
vanishing of the GellMann{Low fun
tion for any G was shown by Johnson [Joh61
to all loops.
192
CHAPTER 3.
1=N
EXPANSION
3.1.
(N ) VECTOR MODELS
?
6 
6 ?
 
a)
b)
I
193
I
6
I
6
I
Fig. 3.2: Oneloop diagrams for the propagator of the eld. The diagram b)
I
I
Fig. 3.3: Bubble diagram whi h survives the largeN limit of the O(N ) ve tor models.
eld".
The perturbationtheory expansion of the O(N )symmetri
fourFermi theory
ontains, in parti
ular, the diagrams of the type depi
ted in Fig. 3.3 whi
h are
alled bubble graphs. Sin
e ea
h bubble has a fa
tor of N , the
ontribution of
the nbubble graph is / Gn+1 N n whi
h is of the order of
Gn+1 N n G
(3.1.13)
as N ! 1 sin
e
1
:
(3.1.14)
G
N
Therefore, all the bubble graphs are essential to the leading order in 1=N .
Let us denote
n loops
= G + : : : + G2
+ Gn+1
+:::
(3.1.15)
6
6
I
Fig. 3.4: Some diagrams of the 1=N expansion for the O(N ) fourFermi theory. The
wavy line represents the (innite) sum of the bubble graphs (3.1.15).
In fa
t the wavy line is nothing but the propagator D of the eld with the
bubble
orre
tions in
luded. The rst term G on the RHS of Eq. (3.1.15) is
nothing but the free propagator (3.1.10).
Summing the geometri
series of the fermionloop
hains on the RHS of
Eq. (3.1.15), one gets analyti
ally2
D 1 (p) =
I
6
6
1
G
dd k
(2)d
sp k^ + im k^ + p^ + im
:
(k2 + m2 ) ((k + p)2 + m2 )
(3.1.16)
2 Re
all that the free Eu
lidean fermioni
propagator is given by S0 (p) = (ip^ + m) 1
due to Eqs. (3.1.5), (3.1.7) and the additional minus sign is asso
iated with the fermion
loop.
194
CHAPTER 3.
1=N
EXPANSION
The main advantage of the expansion in 1=N for the fourFermi intera
tion, over the perturbation theory, is that it is renormalizable in d < 4 while
the perturbationtheory expansion in G is renormalizable only in d = 2.
Moreover, the 1=N expansion of the fourFermi theory in 2 < d < 4 demonstrates [Wil73 an existen
e of an ultravioletstable xed point, i.e. a nontrivial zero of the GellMann{Low fun
tion.
Problem 3.2 Show that the 1=N expansion of the fourFermi theory is renormalizable in 2 d < 4 (but not in d = 4).
Solution In order to demonstrate renormalizability, let us analyze indi
es of the
diagrams of the 1=N expansion.
First of all, we shall get rid of an ultraviolet divergen
e of the integral over the
dmomentum k in Eq. (3.1.16). The divergent part of the integral is proportional
to d 2 (logarithmi
ally divergent in d = 2) with being an ultraviolet
uto. It
an be
an
elled by
hoosing
2
G = g 2 d ;
N
(3.1.17)
3.1.
(N ) VECTOR MODELS
Note that the integral is linearly divergent in d = 3 and is the
uto for the
integration over jkj. This divergen
e
an be
an
elled by
hoosing G a
ording to
Eq. (3.1.17) with g equal to
g = :
Solution
D 1 (p) =
Z
sp k^k^
=
2
k2 k2
d3 k
(2 )3
1
1
= 2
k2
Z
djkj =
: (3.1.19)
2
2N
= 2N
d3 k
k2 + kp
3
2
(2 ) k (k + p)2
d3 k p2 + kp :
(2 )3 k2 (k + p)2
1
k2
(3.1.21)
We have
Z
d3 k p2 + kp =
(2 )3 k2 (k + p)2
Z1
d1
Z1
d2
after whi
h the Gaussian integral over d3 k
an easily be performed. We get then
Z1
Z1
1 2 2
N
1
1
2
1 +2 p :
D (p) = 3=2 p d1 d2
(3.1.24)
e
4
(1 + 2 )5=2
0
The remaining integration over 1 and 2
an easily be done introdu
ing the new
variables 2 [0; 1 and x 2 [0; 1 so that
1 = x ;
2 = (1 x) ;
(1 ; 2 ) = :
(x; )
Solution
d3 k
(2 )3
(3.1.20)
Problem 3.4
Problem 3.3
Z
195
(3.1.25)
D(p) =
8
:
N jpj
(3.1.26)
196
CHAPTER 3.
1=N
EXPANSION
3.1.
(N ) VECTOR MODELS
197
R
R
a)
(p1 ; p2 ) = 1 +
Fig. 3.5: Diagrams for the 1=N 
orre
tion to the eld propagator (a) and the
threevertex (b).
Problem 3.5
8i
N
Z
d3 k k^ + p^
:
(2 )3 jkj(k + p)2
(3.1.27)
The (logarithmi
ally) divergent
ontribution emerges from the domain of integration
jkj jpj so we
an expand the integrand in p. The pindependent term vanishes
after integration over the dire
tions of k so that we get
Z
d3 k 1
(2 )3 jkj3
2
2
= ip^ 1 + 2 ln 2 + nite :
3 N p
S 1 (p) = ip^ 1 +
8
N
Z
(3.1.29)
where p1 and p2 the in
oming and outgoing fermion momenta, respe
tively. The
logarithmi
domain is jkj jpjmax with jpjmax being the largest of jp1 j and jp2 j.
This gives
b)
S 1 (p) = ip^ +
8
N
(3.1.28)
2
2
ln
+ nite :
2 N p2max
(p1 ; p2 ) = 1
(3.1.30)
An analogous
al
ulation of the 1=N
orre
tion for the eld is a bit more
ompli
ated sin
e involves three twoloop diagrams (see Ref. [CMS93). The resulting
expression for D 1 (p) reads
N D(p)
1
= 2+
g
jpj
1
+
+ 2 2
2 8
N
jpj
2 2
ln + nite
3 p2
:(3.1.31)
(3.1.32)
whi h determines g to order 1=N . After this D 1 (p) takes the form
D 1 (p) =
N jpj 1
8
16
2
ln 2
2
3 N p
(3.1.33)
To make all three expressions (3.1.28), (3.1.30), and (3.1.33) nite, we need logarithmi
renormalizations of the wave fun
tions of  and elds and of the vertex
. This
an be a
hieved by multiplying them by the renormalization
onstants
2
Zi () = 1
i ln 2
(3.1.34)
where stands for a referen
e mass s
ale and
i are anomalous dimensions. The
index i stands for , , or v for the  and propagators or the threevertex ,
respe
tively. We have, therefore,
al
ulated
2
=
;
3 2 N
2
v =
;
2 N
16
=
(3.1.35)
3 2 N
198
CHAPTER 3.
1=N
EXPANSION
to order 1=N . Due to Eq. (3.1.6)
oin
ides with the anomalous dimension of the
omposite elds
= :
(3.1.36)
Note, that
Z 2 Zv 2 Z = 1 :
(3.1.37)
This implies that the ee
tive
harge is not renormalized and is given by Eq. (3.1.32).
Thus, the nontrivial zero of the GellMann{Low fun
tion persists to order 1=N (and,
in fa
t, to all orders of the 1=N expansion).
dg2
= exp
(3.1.38)
B(g2) ;
whi
h is the same as Eq. (2.2.86) sin
e the
orrelation length . If B has a
nontrivial xed point g2 near whi
h
B(g2)
= b g2
g2
(3.1.39)
3.1.
(N ) VECTOR MODELS
199
where i stands generi
ally either for verti
es or for inverse propagators,
possess the s
ale invariant solutions
g2
(3.1.44)
i / i( ) :
This
omplements the heuristi
onsideration of Subse
t. 1.3.5 on the relation
between s
ale invarian
e and the vanishing of the GellMann{Low fun
tion.
For the fourFermi theory in d = 3, Eq. (3.1.44) yields
1 p2
;
(3.1.45)
S (p) =
ip^ 2
2
p
8
;
(3.1.46)
D(p) =
N jpj 2
2
v 2
p pp
(p1 ; p2 ) =
f 22 ; 1 2 2 ;
(3.1.47)
2
p1
p1 p1
where f is an arbitrary fun
tion of the dimensionless ratios whi
h is not
determined by s
ale invarian
e. The indi
es here obey the relation
1
(3.1.48)
v =
+
2
whi
h guarantees that Eq. (3.1.37), implied by s
ale invarian
e, is satised.
The indi
es
i are given to order 1=N by Eqs. (3.1.35). When expanded
in 1=N , Eqs. (3.1.45) and (3.1.46) obviously reprodu
es Eqs. (3.1.28) and
(3.1.33). Therefore, one gets the exponentiation of the logarithms whi
h
emerge in the 1=N expansion. The
al
ulation of the next terms of the 1=N expansion for the indi
es
i is
ontained in Ref. [Gra91.
K = 2x x x2 :
(3.1.49)
200
CHAPTER 3.
Group
Transformations
Lorentz
Poin
are
d(d 1)
2
rotations
+ d translations
1=N
EXPANSION
# Parameters
x0 =
x
d(d 1)
2
x0 = x + a
d(d+1)
2
3.1.
(N ) VECTOR MODELS
201
Equation (3.1.51), whi
h results from
onformal invarian
e, unambiguously xes the fun
tion f in Eq. (3.1.47). In
ontrast to innitedimensional
onformal symmetry in d = 2, the
onformal group in d > 2 is less restri
tive.
It xes only the treepoint vertex while, say, the fourpoint vertex remains an
unknown fun
tion of two variables.
Problem 3.6
1=N .
The integral on the RHS of Eq. (3.1.51) looks in d = 3 very mu
h like that
in Eq. (3.1.29) and
an easily be
al
ulated to the leading order in 1=N when only
the region of integration over large momenta with jkj >
jpjmax maxfjp1 j; jp2 jg is
essential to this a
ura
y.
Let us rst note that the
oe
ient in front of the integral is /
v 1=N , so
that one has to peak up the term 1=
v in the integral for the vertex to be of order
1. This term
omes from the region of integration with jkj >
jpjmax . Re
alling that
jp1 p2 j < jpjmax in Eu
lidean spa
e, one gets
Z 3
dk
k^ + p^2
1
k^ + p^1
1+
=
2
2
2 [(k + p1 )
[(k + p2 )2 1+
=2 jkj1+2
=2
Z1
1
dk2 =
=
;
(3.1.52)
2
[k 1+
v
v (p2max )
v
Solution
Weyl
+ 1 dilatation
x0 = x
x0
(x0 )2
d2 +d+2
2
= xx2 +
(d+1)(d+2)
2
Table 3.1: Contents and the number of parameters of groups of spa
etime symmetry.
Dierentiating, we get
K = 2x ;
(3.1.50)
whi
h is analogous to Eqs. (1.3.67) and (1.3.68) for the dilatation
urrent.
Therefore, both the dilatation and
onformal
urrents vanish simultaneously
when is tra
eless whi
h is provided, in turn, by the vanishing of the
GellMann{Low fun
tion.
Conformal invarian
e
ompletely xes threeverti
es as was rst shown by
Polyakov [Pol70 for s
alar theories. The proper formula for the fourFermi
theory reads [Mig71
( d2 ) ( d2
v )
(
v )
dd k
k^ + p^2
k^ + p^1
1+
=
2
d=
2
2
[(k + p1 )
[(k + p2 )2 1+
=2 jkjd
(p1 ; p2 ) = 2 v
2+2
=2
; (3.1.51)
where the
oe
ient in the form of the ratio of the fun
tions is pres
ribed
by the normalization (3.1.45) and (3.1.46) and the indi
es are related by
Eq. (3.1.48) but
an be arbitrary otherwise3.
p2max
2 v :
(3.1.53)
(p1 ; p2 ) =
p2max
While the integral in Eq. (3.1.52) is divergent in the ultraviolet for
v < 0, this
divergen
e disappears after the renormalization.
Equation (3.1.30) is reprodu
ed by Eq. (3.1.52) when expanding in 1=N . This
dependen
e of the threevertex solely on the largest momentum is typi
al for logarithmi
theories in the ultraviolet region where one
an put, say, p1 = 0 without
hanging the integral with logarithmi
a
ura
y. This is valid if the integral is fast
onvergent in infrared regions whi
h is our
ase.
202
CHAPTER 3.
1=N
EXPANSION
S ['a =
dd x
1
1
( 'a )2 + m2 'a 'a + ('a 'a )2
2
2
8
(3.1.54)
where
(3.1.55)
The
oupling in the a
tion (3.1.54) must be positive for the theory to be
well dened. The verti
es of Feynman diagrams are asso
iated with .
Cal
ulate the oneloop GellMann{Low fun
tion of the O(N )symmetri
'4 theory in d = 4.
Solution The
orresponding diagrams are similar to those of Fig. 3.1, though now
the arrows are not essential sin
e the eld is real. The diagrams are logarithmi
ally
divergent in 4 dimensions. Ea
h diagram
ontribute with the positive sign while
the diagram of Fig. 3.1b has now an extra
ombinatori
fa
tor of 1=2. The diagrams of Fig. 3.2 result in a mass renormalization and there is no wavefun
tion
renormalization of the 'eld in one loop so that one gets
Problem 3.7
B () =
(N + 8)2
16 2
(3.1.56)
The positive sign in this formula is the same as for QED and is asso
iated with
\triviality" of the '4 theory in 4 dimensions. It is also worth noting that the
oe
ient (N + 8) is large even for N = 1.
Introdu
ing the auxiliary eld (x) as in Subse
t. 3.1.1, the a
tion (3.1.54)
an be rewritten as
Z
a
2
1 a
2
2
a
d
:
(3.1.57)
S [' ; =
d x ' + m + '
2
2
(N ) VECTOR MODELS
203
Z [J a ; K
D (x)
"R
D'a (x)
R
e
(3.1.59)
whi
h is a fun
tional of the sour
es J a and K for the elds 'a and and
extends Eq. (1.2.49).
To make the path integral over (x) in Eq. (3.1.59)
onvergent, we integrate at ea
h point x over a
ontour whi
h is parallel to imaginary axis. This
is spe
i
to the Eu
lidean formulation. The propagator of the eld in the
Gaussian approximation reads
D0 (p) =
h ( p)(p) iGauss
;
(3.1.60)
Z [J a ; K
=
"
d 2 1
d d a
a
d
D (x) e d x 2 + 2 d xd yJ (x)G(x;y;)J (y)+ d xK (x)(x)
N Sp ln G 1 [
2
:
e
(3.1.62)
G 1 [ =
2 + m2 + :
(3.1.63)
g f = hgjf i
dd x f (x)g(x) :
(3.1.64)
= 2 : 'a 'a : :
3.1.
(3.1.58)
Z [J a ; K
"
D (x) e
+ 1 J a G[J a +K N Sp ln G 1 [
2 2
2
:
(3.1.65)
204
CHAPTER 3.
1=N
EXPANSION
(3.1.66)
N
sp (x)
G (x; x; sp )
2
+ J a G (; x; sp ) G (x; ; sp ) J a + K (x) = 0 : (3.1.67)
2
If K 1=, ea
h term here is O(1) sin
e
1
(3.1.68)
N
in analogy with Eq. (3.1.14).
When the sour
es J a and K vanish so that the last two terms on the LHS
of Eq. (3.1.67) equal zero, this equation redu
es to
N
G (x; x; sp ) = 0 :
(3.1.69)
2
Its solution is x independent due to translational invarian
e and
an be
parametrized as
sp
sp = m2R m2
(3.1.70)
where m and mR are the bare and renormalized mass, respe tively. Equation (3.1.69) then redu es to the standard formula [Wil73
m2
m2R
N
2
1
dd k
d
2
(2) (k + m2R )
(3.1.71)
(x)
N
1=2 :
(3.1.72)
(3.1.73)
3.1.
(N ) VECTOR MODELS
205
(3.1.74)
Then we get
W [J a ; K =
1
N
Sp ln G 1 [sp
2 sp sp 2
1
+ J a G [sp J a + K sp
2
1
Sp ln D 1 [sp + O N 1 ;
2
(3.1.75)
where
1 (d)
N
D 1 (x; y; ) =
(x y )
G (x; y; ) G (y; x; )
2
a
+J G (; x; ) G (x; y; ) G (y; ; ) J a :
(3.1.76)
This operator emerges when integrating over the Gaussian
u
tuations
around the saddle point. The
orresponding (last) term on the RHS of
Eq. (3.1.75) is asso
iated with the preexponential fa
tor and, therefore, is
1.
The next terms of the 1=N expansion
an be
al
ulated in a systemati
way by substituting (3.1.72) in Eq. (3.1.65) and performing the perturbative
expansion in .
If the sour
es J a and K vanish so that the saddlepoint value sp is given
by the
onstant (3.1.70), than the RHS of Eq. (3.1.76) simplies to
N
1 (d)
(x y)
G (x; y; sp ) G (y; x; sp ) :(3.1.77)
2
Remembering the denition (3.1.61) of G and passing to the momentumspa
e
representation, we get
D 1 (x; y; sp ) =
D 1 (p) =
1
N
2
1
dd k
: (3.1.78)
2
d
2
(2) (k + mR ) ((k + p)2 + m2R )
206
CHAPTER 3.
1=N
EXPANSION
The sign of the rst term on the RHS is
onsistent with Eq. (3.1.60).
Equation (3.1.78) is an analog of Eq. (3.1.16) in the fermioni
ase and
an
be alternatively obtained by summing bubble graphs of the type in Fig. 3.3
for
D (p) = h ( p)(p) i :
(3.1.79)
The extra symmetry fa
tor 1=2 in Eq. (3.1.78) is the usual
ombinatori
one
for bosons. Therefore, the largeN saddlepoint
al
ulation of the propagator (3.1.79) results pre
isely in the zerothorder of the 1=N expansion.
We see from Eq. (3.1.75) the dieren
e between perturbation theory and
the 1=N expansion. The perturbation theory in
an be
onstru
ted as an
expansion (3.1.72) around the saddle point sp given again by Eq. (3.1.67),
with the omitted se
ond term on the LHS, whi
h is now justied by the fa
t
that is small (even for N 1). The se
ond term on the RHS of Eq. (3.1.75),
whi
h is asso
iated with a oneloop diagram, appears in perturbation theory
as a result of Gaussian
u
tuations around this saddlepoint.
3.1.
(N ) VECTOR MODELS
207
where
D 1 (x; y; l ) =
1 (d)
N
(x y)
G (x; y;
l ) G (y; x;
l )
2
+'a
l (x)G (x; y;
l ) 'a
l (y)
(3.1.84)
= J a (x) ;
'a
l (x)
= K (x) :
l (x)
(3.1.85)
Therefore, 'a
l (x) and
l (x) are determined in the absen
e of external sour
es
by the equations
['a
l ;
l
= 0
'b
l (x)
(3.1.86)
['a
l ;
l
= 0:
l (x)
(3.1.87)
and
Substituting (3.1.83) into Eqs. (3.1.86) and (3.1.87), we get to the leading
order in 1=N , respe
tively, the equations
(3.1.88)
and
a
N
'
l (x)'a
l (x) +
G (x; x;
l ) :
(3.1.89)
2
2
The rst equation is just a
lassi
al equation of motion in an external eld
l (x) while the se
ond one is just the average of the (quantum) equation (3.1.58). Equation (3.1.89) is often
alled the gap equation.
A solution to Eqs. (3.1.88) and (3.1.89) depends on what initial (or boundary)
onditions are imposed.
l (x) =
208
CHAPTER 3.
1=N
EXPANSION
Problem 3.8
l =
(3.1.90)
by the formula
= Vol: V ('a ; ) :
(3.1.91)
V =
1 2 N
+
2
2
(3.1.92)
whi
h obviously re
overs Eqs. (3.1.88) and (3.1.89) after varying with respe
t to
onstant 'a and .
It is
onvenient to perform renormalization by introdu
ing, in d = 4, the renormalized
oupling R given by
1
1 1
= +
R
2
Z
1
N
2
d4 k
1
ln 2
= +
2
4
2
2
2
(2 ) k (k + mR ) 32
mR
(3.1.93)
and
R
Assuming that R 2 (also mR
the form
R 1
(3.1.94)
1 2 m2R R
N 2
+
+
2R R
R
64 2 R
'a = 0 ;
R = m2R for m2R > 0 ;
(3.1.99)
2
2mR
2
2
' =
(3.1.100)
;
R = 0
for mR < 0 :
R
The rst of them is asso
iated with an unbroken O(N ) symmetry, while the se
ond
one
orresponds to a spontaneous breaking of O(N ) down to O(N 1). Both
formulas look like the proper treelevel ones while the only ee
t of loop
orre
tions
at large N is the renormalization of the
oupling
onstant and mass.
A subtle point is a question of stability of these solutions. For small deviations of '2 from the mean value given by Eqs. (3.1.99) and (3.1.100), the ee
tive
potential VR is a monotoni
ally in
reasing fun
tion of '2 , as
an be shown for
R N < 32 2 eliminating the auxiliary eld R from Eq. (3.1.98) by solving the gap
equation (3.1.97) iteratively in '2 , and the solutions are lo
ally stable. Both solutions are, however, unstable globally with respe
t to large
u
tuations of the elds.
This
an be seen by eliminating '2 from VR by solving the gap equation (3.1.97)
for '2 whi
h yields
VR = 1 2R 1
2
R
N ln R
32 2 m2R
N 2 :
128 2 R
(3.1.96)
(3.1.97)
1
1
+ ln R2 + R '2 ;
2
mR
2
(3.1.98)
(3.1.101)
(3.1.102)
where the theory be
omes unstable. This is related to the usual problem of \triviality" of the '4 theory whi
h makes sense only for small
ouplings R N as an ee
tive
theory and
annot be fundamental at very small distan
es of the order of
r
Problem 3.9
VR =
209
(3.1.95)
R 'a = 0 ;
R N ln R
= m2R + R '2 :
32 2 m2R
2
(N ) VECTOR MODELS
322
= + m2 :
m2R = m2
N 2 ;
R
32 2
we rewrite Eqs. (3.1.88) and (3.1.89) as [S
h74, CJP74
3.1.
de reases as
mR 1 e
(3.1.103)
'a
l (x) = a mR e mR
where
162
R N
for
! 1
(3.1.104)
Solution The dieren
e from the previous problem is that '
l is no longer translationally invariant along the timevariable due to the initial
ondition (3.1.104). Let
us denote
'a
l (x)
l (x)
a ( ) ;
v( ) :
(3.1.105)
210
CHAPTER 3.
1=N
EXPANSION
and
v ( ) = a ( )a ( ) + N
2
2
= 0
(3.1.106)
d3 k G (; ; v ) ;
(2 )3 !
D d ;
(3.1.108)
d
p
k 2 + m2
! =
(3.1.109)
and we have introdu
ed the Fourier image of the Green fun
tion (3.1.61)
=
D
E
1
D2 + ! 2 + v
211
(3.1.107)
where
G! (; ; v )
(N ) VECTOR MODELS
3.1.
(3.1.110)
diers from Eq. (3.1.93) only by an additional nal renormalization and the renormalized mass mR is dened in Eq. (3.1.71). This solution is nontrivial for 2 N
and obviously satises the initial
ondition (3.1.104).
The solution is so simple be
ause the diagonal resolvent (3.1.110) takes on the
very simple form
1
v ( )
G! (; ; v ) =
(3.1.114)
2! 4! (! 2 m2R )
for the potential v ( ) given by Eqs. (3.1.111), (3.1.112). This
an be easily veried
by substituting into the Gelfand{Di
key equation (1.1.123) with G = 1. This is a
feature of an integrable potential whi
h was already dis
ussed in Problem 1.29.
The fun
tion a ( ) given by Eq. (3.1.111) des
ribes largeN amplitudes of multiparti
le produ
tion at a threshold [Mak94.
Z =
D~n
~n2
1
e
g2
1
2
R 2
d x( ~n)2
(3.1.115)
where
~n = (n1 ; : : : ; nN )
(3.1.116)
is an O(N ) ve
tor. While the a
tion in Eq. (3.1.115) is pure Gaussian, the
model is not free due to the
onstraint
1
~n2 (x) = 2 ;
(3.1.117)
g
whi
h is imposed on the ~n eld via the (fun
tional) deltafun
tion.
The sigma model in d = 2 is sometimes
onsidered as a toy model for
QCD sin
e it possesses:
1) asymptoti
freedom [Pol75;
2) instantons for N = 3 [BP75.
"
Du (x)
D~n (x) e
1
2
R 2
d x ( ~n)2 u ~n2 g12
;
(3.1.118)
"
Du (x) e
N Sp ln( 2 +u(x))+ 1
2
2g2
R 2
d x u(x)
(3.1.119)
4 The name
omes from elementary parti
le physi
s where a nonlinear sigma model in
4 dimensions is used as an ee
tive Lagrangian for des
ribing lowenergy s
attering of the
Goldstone mesons.
212
CHAPTER 3.
1=N
EXPANSION
The rst term in the exponent is as before nothing but the sum of oneloop
diagrams in 2 dimensions
N
Sp ln
2
2 + u (x) =
1
n
n
NG (x; x; usp ) = 0
(3.1.121)
is quite analogous to Eq. (3.1.69) for the '4 theory while G is dened by
G (x; y; u) =
y
1
x ;
2 + u
(3.1.122)
NX1
1u
nN = t1 g2
n2a
g
a=1
(N ) VECTOR MODELS
(3.1.123)
213
O(N 1). The parti
les in perturbation theory are massless (like Goldstone
bosons) and it suers from infrared divergen
ies.
On the
ontrary, the solution to Eq. (3.1.121) has the form
usp = m2R 2 e
(3.1.120)
where the auxiliary eld u is denoted again by the wavy line. Equation (3.1.119) looks very mu
h like Eq. (3.1.65) if we put there J a = K = 0.
The dieren
e is that the exponent in (3.1.119) involves the term whi
h is
linear in u, while the analogous term in (3.1.65) is quadrati
in .
Now the path integral over u(x) in Eq. (3.1.119) is a typi
al saddlepoint
one: the a
tion N while the entropy 1 sin
e only one integration over u
is left. The saddlepoint equation for the nonlinear sigma model
1
g2
3.1.
4
Ng2
(3.1.124)
where is an ultraviolet
uto. Therefore, all N parti
les a
quires the same
mass mR in the 1=N expansion so that the O(N ) symmetry is restored. This
appearan
e of mass is due to dimensional transmutation whi
h says in this
ase that the parameter mR rather than the renormalized
oupling
onstant
gR2 is observable. The emergen
e of the mass
ures the infrared problem.
Show that (3.1.124) is a solution to Eq. (3.1.121).
Let us look for a translationally invariant solution
Problem 3.10
Solution
usp (x)
m2R :
(3.1.125)
Z
1
N
d2 k
=
(2 )2 k2 + m2R
4
Z 2
0
dk2
2
k + m2R
N 2
ln
: (3.1.126)
4 m2R
Ng 4 :
(3.1.127)
2
The analogous oneloop perturbationtheory formula for any N reads [Pol75
2
=
B(g2 ) ddg
2)g 4
:
(3.1.128)
2
Thus, the sigmamodel is asymptoti
ally free in 2dimensions for N > 2 whi
h is
the origin of the dimensional transmutation. There is no asymptoti
freedom for
N = 2 sin
e O(2) is Abelian.
B(g2 ) =
(N
214
CHAPTER 3.
1=N
EXPANSION
h u (x1 ) : : : uZ(xk ) i
Z 1 Du e
"
N Sp ln[ 2 +u+ 1
2
2g2
R 2
d xu
(3.1.130)
h u (x1 ) : : : u (xk ) i
1
; (3.1.131)
N
3.2.
215
MULTICOLOR QCD
whi
h is similar to that used by 't Hooft [Hoo74a and diers from (2.1.5) by
a fa
tor of ig:
Aij (x) = igAij (x) :
(3.2.2)
216
CHAPTER 3.
1=N
EXPANSION
ij
1 il kj 1 ij kl
kl
A (x) A (y) Gauss =
D (x y) ; (3.2.3)
2
N
where we have assumed, as usual, a gauge xing to dene the propagator in
perturbation theory. For instan
e, one has
D (x y) =
1
42 (x y)2
(3.2.4)
a
A
(x)
Ab (y)
Gauss
= ab D (x y)
(3.2.5)
a=1
1 il kj
2
1 ij kl
N
where the fa
tor of 1=2 is due to the normalization (2.1.6). Eq. (3.2.3)
an
be derived dire
tly from a path integral over matri
es.
We
on
entrate in this Se
tion only on the stru
ture of diagrams in the
index spa
e, i.e. the spa
e of the indi
es asso
iated with the SU(N
) group.
We shall not
onsider, in most
ases, spa
etime stru
tures of diagrams whi
h
are pres
ribed by Feynman's rules.
Omitting at large N
the se
ond term in parentheses on the RHS of
Eq. (3.2.3), we depi
t the propagator by the double line
i  l
:
(3.2.7)
j k
Ea
h line represents the Krone
ker deltasymbol and has orientation whi
h
is indi
ated by arrows. This notation is obviously
onsistent with the spa
etime stru
ture of the propagator whi
h des
ribes a propagation from x to
y.
The arrows are due to the fa
t that the matrix Aij is Hermitean and its
odiagonal
omponents are
omplex
onjugate. The independent elds are,
Gauss
/ il kj
3.2.
217
MULTICOLOR QCD
say, the
omplex elds Aij for i > j and the diagonal real elds Aii . The
arrow represents the dire
tion of the propagation of the indi
es of
the
omplex
ij
eld Aij for i > j while the
omplex
onjugate one, Aji
= A , propagates
in the opposite dire
tion. For the real elds Aii , the arrows are not essential.
The doubleline notation (3.2.7) looks similar to that of Subse
t. 2.2.5.
The reason for that is deep: double lines appear generi
ally in all models
des
ribing matrix elds in
ontrast to ve
tor (in internal symmetry spa
e)
elds whose propagators are depi
ted by single lines as in the previous Se
tion.
The threegluon vertex, whi
h is generated by the a
tion (2.1.14), is depi
ted in the doubleline notations as
j1
i1
... ....
... ...
i2 ............................................j3
.
........
.
.
.
.
.
...
j2
i3
i1
j1
.... .....
... ...
j2 ...........................................i3
.
.
........
.
.
.
.
...
i2
j3
/ g i1 j3 i2 j1 i3 j2 i1 j2 i3 j1 i2 j3
(3.2.8)
where the subs
ripts 1, 2 or 3 refer to ea
h of the three gluons. The relative minus sign is due to the
ommutator in the
ubi
in A term in the
a
tion (2.1.14). The
olor part of the threevertex is antisymmetri
under
inter
hanging the gluons. The spa
etime stru
ture, whi
h reads in the momentum spa
e as
1 2 3 (p1 ; p2 ; p3 )
= 1 2 (p1 p2 )3 + 2 3 (p2 p3 )1 + 1 3 (p3 p1 )2 ; (3.2.9)
is antisymmetri
as well. We
onsider all three gluons as in
oming so that
their momenta obey p1 + p2 + p3 = 0. The full vertex is symmetri
as is
pres
ribed by Bose statisti
s.
The
olor stru
ture in Eq. (3.2.8)
an alternatively be obtained by multiplying the standard vertex
a1 a2 a3
1 2 3 (p1 ; p2 ; p3 )
= f a1 a2 a3 1 2 3 (p1 ; p2 ; p3 )
(3.2.10)
by (ta1 )i1 j1 (ta2 )i2 j2 (ta3 )i3 j3 , with f ab
being the stru
ture
onstants of the
SU(N
) group, and using the formula
(3.2.11)
218
CHAPTER 3.
1=N
EXPANSION
3.2.
.....................
....... ....................................
................
...... ....
.
.
.... ...
.......
.. ...
..... ...
.. ..
.
..
... ....
..
... ...
.... .....
.
... ....
.
.
.
. ..
.... ......
...... ............ .........................
.......... .... ...........
...........
l .....
..... i
....
.....
k
j
... ...
k j
/ g2
(3.2.12)
and there are no deltas on the RHS sin
e the
olor stru
ture is xed. In other
words, we pi
k up only one
olor stru
ture by equaling indi
es pairwise.
The diagrams of perturbation theory
an now be
ompletely rewritten
in the doubleline notation [Hoo74a. The simplest one whi
h des
ribes the
oneloop
orre
tion to the gluon propagator is depi
ted in Fig. 3.6.6 This
diagram involves two threegluon verti
es and a sum over the N
indi
es whi
h
is asso
iated with the
losed index line analogous to Eq. (2.2.70). Therefore,
the
ontribution of this diagram is g2 N
.
In order for the largeN
limit to be nontrivial, the bare
oupling
onstant
g2 should satisfy
g2
N1 :
(3.2.13)
This dependen
e on N
is similar to Eqs. (3.1.14) and (3.1.68) for the ve
tor
6 Here and in the most gures below the arrows of the index lines are omitted for
simpli ity.
(3.2.14)
The sum over the N
indi
es is asso
iated with the
losed index line. The
ontribution of this diagram is g2 N
1.
... ....
242
11N
ln (=QCD )
g2 =
Fig. 3.6: Doubleline representation of a oneloop diagram for the gluon propagator.
l.
219
MULTICOLOR QCD
Fig. 3.6
g2 N
(3.2.15)
TA =
tr T AT B =
ta ; p2IN ;
1 AB
:
2
(3.2.16)
N
X
A=1
TA
ij
TA
kl
1 il kj
2
for U(N ) :
(3.2.17)
The point is that elements of both the SU(N
) group and the U(N
) group
an be represented in the form
U = eiB ;
(3.2.18)
where B is a general Hermitean matrix for U(N
) and a tra
eless Hermitean
matrix for SU(N
).
Therefore, the doubleline representation of perturbationtheory diagrams
whi
h is des
ribed in this Se
tion holds, stri
tly speaking, only for the U(N
)
gauge group. However, the largeN
limit of both the U(N
) group and the
SU(N
) group is the same.
220
CHAPTER 3.
1=N
EXPANSION
..................
...................................................
...........
............... ...
.
.
.
.... ...
.. ...
... ...
... ...
.. ..
.... ....
.. ..
. ...
..
... ..
.... ...
.. ...
.
... ..
.
. ...
..........
............
..............
.
.
.
.
.
.
......... ...................... ..........
............................
The sum over the N
indi
es is asso
iated with ea
h of the four
losed index
lines whose number is equal to the number of loops. The
ontribution of
this diagram is g8 N
4 1.
1:
(3.2.19)
g2 N
n2
gluon propagator. The diagram has six threegluon verti
es but only
one
losed index line (while three loops!). The order of this diagram is
g6 N
1=N
2 .
g 2 N
221
MULTICOLOR QCD
...............
...................... .........
.................. .......................
.
.
.
.
... .............
....... ...
... ... ................... ................ ..........
.......................
... ....
.. .
...
... ....
... ......................
... ...
.
.
.
.
... ...
.
.
.
.
.
.
..
... .... .................. ...........................
.... . ....
.... ..........
.................. ..................
........... ...... ..........
...........
Fig. 3.7: Doubleline representation of a fourloop diagram for the gluon propagator.
Fig. 3.7
3.2.
1;
(3.2.20)
Fig. 3.8
g6 N
N12 :
(3.2.21)
onstn0 ;
(3.2.23)
222
CHAPTER 3.
1=N
EXPANSION
..
..
..
.. ..
.. ..
.... ...
. . .
.. ...
... ... .... ............
.
..
.
.
.. ....
... ... ........ ......
... .. .... ... ....................................................................
.
............... . .
................................................ .... .......... ... .....
. .... . .
.... ........
......... .............. .... ...
................. ...
................. ......................................... ..................................
........
.
. . ...... ..
... ......... .........................
.
.......
.... ..... ............ ....................................................................... .....................
..............
.
.
.
............................................... ...................
.
.
.
.
.
.
.
.
.
...........
..... ....
.. ..
.. . ..... ..........
....
.. ..
.. ... ........
....
..... ..... ..... ... .......
....
...
.. .. ... ....
..
..
.
..
line. The numbers of verti
es of ea
h part and of the whole graph obey
Eq. (3.2.24).
while the number of all the graphs grows with n0 fa
torially. There is no
dependen
e in Eq. (3.2.23) on the number of external lines of a planar graph
whi
h is assumed to be mu
h less than n0 .
It is instru
tive to see the dieren
e between the planar diagrams and,
for instan
e, the ladder diagrams whi
h des
ribe e+ e elasti
s
attering in
QED. Let the ladder has n rungs. Then there are n! ladder diagrams, but
only one of them is planar. This simple example shows why the number of
planar graphs is mu
h smaller than the number of all graphs, most of whi
h
are nonplanar.
In the rest of these le
ture notes, we shall dis
uss what is known about
solving the problem of summing the planar graphs.
Problem 3.11 Show that Eq. (3.2.23) for the number of planar graphs is
onsistent
with its independen
e of the number of external lines.
Solution Let us split a planar graph in two parts by
utting along some line as is
depi
ted in Fig. 3.9. The numbers of verti
es of ea
h part, n00 and n000 , are obviously
related to that of the original graph, n0 , by
(3.2.24)
223
MULTICOLOR QCD
. . . .... . .. . . . . . .
.. .
..
... .....................
.
......... ...
.. ..........
.
.......
..
.
.
.
.
... ..
.. ....
... ....... . . . .
... ..
. . . . .. ...
.. . . ......... . . .. ..
..
.
...
....
..
...
..
....
.
..
.
.
.. . . ..... . . ..
.
.
.
.
...
.
.
.
.
. . . . ..... ..
... ...... . .... .
......
......
....
.
.
.
.
.
.
.. ......
.. .....
.... .......
.. .
.. .
.
.
.
.
...
..
..
Fig. 3.9: Cutting a planar graph into two graphs. The utting is along the dotted
n00 + n000 = n0 :
3.2.
.......
........
.........
.........
.... ...
.... ...
... ...
... ...
............................................................
.... ..... .... ..... .... ..... .... ....
.... ..... ....
..... ....
.......... ...
.... .....
.
..
.
.... ....
.... .......
.
.
.... ....
.
.
.
.
.
.
.... ....
.... ...
.... ....
.... ....
.... ....
.... .... ....... .......
.... .... .... ...
.... .... ....
....
.... ......
.... ....
....
b)
a)
Fig. 3.10: Cutting a planar graph into trees and ar hes. The line of utting is depi ted
whi
h is obviously satised by the formula (3.2.23) a
ounting for Eq. (3.2.24).
Problem 3.12
(3.2.26)
(3.2.27)
224
CHAPTER 3.
.. .
.. .
1=N
EXPANSION
3.2.
.. .. ..
..
..
.. .. ..
.. ..
... .. ... ....... .. .. ... ... ..... .. .. ... .... ..... ... ....... ..... ... ..
.
.
.
.
.
.
.
.
.
... .... . .. ..
.
.. .. ... .... ...... ......... .. .. ... ......
.
.
.... ..
... .... ......
.. ...
... ..
.... ... ... ...
.... .......
.... ..
.. .... ...
..... ...
.
.
.
.
.
.
.
.
. ..... .... .... ....
.. . .... .... ..... ........
..
....
...
...
...
....
....
.
...
.
....
.
.
.
...
....
....
...
.... ......
....
....
.... ...
.
.... .
.
.
.
.....
..
... ...
.... ..
. .... ...
. ..... ..... .... .... .... ....
.......
.......
........... .............
........... ...........
...... .............. ..........
...... ............. .........
....
.
...
. .
............................................................
.....
. .
. ...
. ....... ......
.... ...... ...................
... ...
........
.... ...
..
.... .... .............. .... .....
....
....... .. ..
.... ...
.
.
.
..........
.... ...
.
.
.... ....
.... ....
.... ...
... .....
.... ...
..
.... ...
.....
........
....
.......
.
.
.
....... ....
..... ...
..
. . . . ......................................
.....
......
.... . . . . . . . . . . . . .......
...
.
...
.
.
.
. .... .
...
... .... . . . .. ... .. ..... ...
..
...
.
.
....
..
...
.
.
.
... ...... . . . . . . . . . . . . . ..
.
.
.
. . . . . . ....
.. ....
....
.... . . . . . . .. ......
.. .... .
...
. . . ...... ......
.. .
.... .....
.
.
....
..
225
MULTICOLOR QCD
Fig. 3.12: A tree graph (the solid lines) and its dual (the dashed ar
hes).
a)
..
..
..
..
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.............
...
..
..........
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..... ...
. .....................
.
......
.
.
. .
.. . ..............
.
......... ..........
... ............ .............
....... ... .. ... ....... .... ........... ...........
.. ... .. ... ... ... .. ... ... ... .. ... .. .. ... .. ... ... ... .. ... ...
..
..
..
..
.
b)
Fig. 3.11: Alternative
utting of the same planar graph as in Fig. 3.10 into trees and
ar
hes.
where #A (n2 ) stands for the number of ar
hes and #T (n0 ; 2n2 ) stands for the
number of trees with n0 verti
es and 2n2 tips. An alternative way of
utting the
same planar graph, whi
h leads to a dierent
ombination of ar
hes and trees, is
depi
ted in Fig. 3.11.
The number of ar
hes is wellknown in mathemati
s and is given by the Catalan
number of order n:
2n!
n!1 n
#A (n) =
! 4 :
(3.2.28)
n!(n + 1)!
The number of trees is not independent sin
e a graph, dual to a tree graph,
onsists
of ar
hes as is illustrated by Fig. 3.12. The number of ar
hes of this dual graph
equals the sum of the number n0 of verti
es and the number 2n2 of tips, i.e. equals
n0 +2n2 . Given the number n2 of loops, the number n0 of verti
es is maximal when
all verti
es are trivalent, so that
n0 2n2
(3.2.29)
(n0 = 2n2 for trivalent and n0 = n2 for fourvalent verti
es when n2 is large).
Therefore, the number of ar
hes of the dual graph is bounded by 4n2 , so that
#T (n0 ; 2n2 )
#A (4n2 ) :
(3.2.30)
Substituting in (3.2.27), we get [KNN77 the inequality (3.2.26). Finally, the inequality (3.2.23) an be obtained by noting that n0 n2 for large n2 .
...
....... .......
... ... ... ...
Fig. 3.13: Re urren e relation for the number of ar hes. The dotted line separates
Problem 3.13
Solution
#A (n) =
n
X
n0 =1
#A n0
1 #A n n 0 ;
(3.2.31)
where the number of ar
hes to the left of the dotted line is des
ribed by #A (n0 1)
be
ause one ar
h en
ir
les n0 1 others. Eq. (3.2.31) expresses #A (n) re
urrently
via #A (0) = 1.
Introdu
ing the generating fun
tion
1
X
fA (g ) =
g n+1 #A (n) ;
(3.2.32)
n=0
226
CHAPTER 3.
1=N
EXPANSION
3.2.
227
MULTICOLOR QCD
. . . . . . . . . ..
. . . . . . . . . . . . . ..
...
.. . .
.. .
...
.
..
.
....
..
.. ...
..
... ...... ..... ...... ..... .. .... ......... ..... ... ..... .... ....
.
.
.
.
.
.
.
.
.
.
. .......
....... .. .... .......
..... ....
..
......
... .. .. ...
..
. ..
..
......
... ..
..
..
...
... ......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...... ...
..
.
.
..
.
.
.
.
.
.
.
...... ... ....
... .....
..
..
...... .. .. .
.
.
.
.
..
.
......... . ..
.
..
..
....... .. ...........
..
..
..
....... ... ......
...
.
.
.
.
.
. .... ...... .
.. . .
... ..... . . .. .... . . . ..
. .. .... . . . ..
...
fT (g ) 1 = gfT2 (g ) :
Its solution
fT (g ) = 1
p1 4g X
1
(2n 2)!
=
gn 1
2g
n
!(n 1)!
n=1
(3.2.38)
(3.2.39)
gives
#T (n) =
Fig. 3.14: Re
urren
e relation for the number of trees. The trees inside the left and
right dotted
ir
les have n0 and n n0 tips, respe
tively.
fA (g ) g =
Its solution
fA (g ) = 1
fA2 (g ) :
(3.2.33)
p1 4g X
1
(2n)!
=
g n+1
2
n!(n + 1)!
n=0
(3.2.34)
Problem 3.14
#T (n)
#T (n 1; n)
(3.2.35)
in the notation of Problem 3.12. Pi
king up the rst vertex in a tree as is depi
ted
in Fig. 3.14, we get the following re
ursion relation
#T (n) =
n 1
X
n0 =1
#T n 0 #T n n 0 ;
(3.2.36)
(3.2.37)
(3.2.40)
(2n 2)!
:
n!(n 1)!
(3.2.41)
The inequality here is due to (3.2.29) and the equality is be
ause of the expli
it
formulas (3.2.28) and (3.2.40). We have improved, thus, the estimate (3.2.30) having
al
ulated the number of tree graphs. The inequality (3.2.26) is now improved by
#p
(64)n2 :
(3.2.42)
The a
tual number of the trivalent planar graphs was evaluated by Tutte [Tut62
and reads
n2
#p 256
(3.2.43)
27
for asymptoti
ally large n2 .
228
CHAPTER 3.
1=N
EXPANSION
...................................
...............
........
........
.......
.
.
.
.
.
......
......
.
.....
.
.
.
.
.
....
.
.
.
.
....
.
.....
...
.
...
....
.....
...
..........................................................
.
.
.
.
...
....
.
. ......
...........
.
.
.
...
.
.
.
.
.
.......................
....
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
. . ............. ........... ... ...
...
......................
...
.
... ....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
.
...
... ...
.
.
.
.
.
.
.
... .
..
.. ..
...
... ...
...... ..........
...
... .... ................ ...............................
.
...
.
.
.... ...........
...
..
...... ............ ..........................
..
...
............ ..... ............
...
...
... ...
.
.
....
...
....
...
.
....
.
.
.....
....
......
.....
.......
.....
.
.
.
.
.........
.
.
......
................
.... ........................
(3.2.44)
G(n1)n (x1 ; : : : ; xn )
(ig)n
h tr [A1 (x1 ) An (xn )i :
N
(3.2.45)
229
MULTICOLOR QCD
....... .................
.........
......
.
......
....
..... ......
............
.....
.
..
........
..
...
....
..
.
.
..
.
.
.
.
.
.
....... .
.
.
.
.
...
.
.
.
..
.
... ................. . .................. .................. ....
............ ........
....... ..............
........
....
....
............
.....
.......
.............. ...........................
........
Fig. 3.15: Generi index diagram with n0 = 10 verti es, n1 = 10 gluon propagators,
G(3)
1 2 3 (x1 ; x2 ; x3 )
3.2.
a)
b)
Fig. 3.16: Planar a) and nonplanar b)
ontributions of the two
olor stru
tures in
Eq. (3.2.8) for threegluon vertex to G(3) in the lowest order of perturbation theory.
The fa tor 1=N , whi h normalizes the tra e, provides the natural normalization
G(0) = 1
(3.2.46)
of the averages.
Though the two terms in the indexspa
e representation (3.2.8) of the
threegluon vertex look very similar, their fate in the topologi
al expansion
is quite dierent. When the
olor indi
es are
ontra
ted anti
lo
kwise, the
rst term leads to the planar
ontributions to G(3) , the simplest of whi
h is
depi
ted in Fig. 3.16a. The anti
lo
kwise
ontra
tion of the
olor indi
es in
the se
ond term leads to a nonplanar graph in Fig. 3.16b whi
h
an be drawn
without
rossing of lines only on a torus. Therefore, the two
olor stru
tures
of the threegluon vertex
ontribute to dierent orders of the topologi
al
expansion. The same is true for the fourgluon vertex.
230
CHAPTER 3.
...
............. ...................
...........
..
.
.
.. ........
.............
.............. .....
... ..................
.
.
.
.
....
.......... ........
..
................
..
....
.
.......................
.
...
.
.
.
...
......
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
......
........... ..
.
.
.
.
......... ..
.... .........
............
...........
......
..
...................................
...
....
...
.
1=N
..........................
............
......
.....
......
.
.
.
.
...
...
...
...
....
....
.......
............ ..................
..............
a)
EXPANSION
..
..
.
...
...
b)
Fig. 3.17: Example of a)
onne
ted and b) dis
onne
ted planar graphs.
of the lines is pres
ribed by the orientation of the external boundary whi
h
we
hoose to be, say, anti
lo
kwise like in Fig. 3.16a. Sin
e the lines are
oriented, the fa
es of the Riemann surfa
e asso
iated with a given graph are
oriented too  all in the same way  anti
lo
kwise. Vi
e versa, su
h an
orientation of the Riemann surfa
es unambiguously xes the orientation of
all the index lines. This is the reason why we omit the arrows asso
iated with
the orientation of the index lines: their dire
tions are obvious.
H
1
tr P e ig dx A (x)
N
xZn 1
Zx1
1 I
X
2
1
=
dx1 dx2 : : : dxnn G(n1)n (x1 ; : : : ; xn ) :
n=0
x1
3.2.
planar while the graph in Fig. 3.17b is dis
onne
ted planar.
The usual relation (1.2.52) between the generating fun
tionals W [J and
Z [J for
onne
ted graphs and all graphs, whi
h is dis
ussed in the Remark
on p. 45, does not hold for the planar graphs. The reason is that an exponentiation of su
h a
onne
ted planar diagram for the
y
li
ordered Green's
fun
tions (3.2.45)
an give dis
onne
ted nonplanar ones.
The generating fun
tionals for all and
onne
ted planar graphs
an be
onstru
ted [Cvi81 by means of introdu
ing non
ommutative sour
es j (x).
\Non
ommutative" means that there is no way to transform j 1 (x1 )j 2 (x2 )
into j 2 (x2 )j 1 (x1 ). This non
ommutativity of the sour
es re
e
ts the
y
li
ordered stru
ture of the Green's fun
tions (3.2.45) whi
h possess only
y
li
symmetry.
Using the shorthand notations (3.1.64) where the symbol in
ludes the
sum over the dve
tor (or whatever available) indi
es ex
ept for the
olor
ones:
j A
(3.2.47)
The reason is be ause the ordering along a losed path implies the y li ordering in the index spa e.
XZ
dd x j (x)A (x) ;
(3.2.48)
we write down the denitions of the generating fun
tionals for all planar and
onne
ted planar graphs, respe
tively, as
1 1
X
n
Z [j
tr (j A)
(3.2.49)
n=0 N
and
W [j
x1
231
MULTICOLOR QCD
1
X
n=0
1
tr (j A)n
N
onn
(3.2.50)
The planar
ontribution to the Green's fun
tions (3.2.45) and their
onne
ted
ounterparts
an be obtained, respe
tively, from the generating fun
tionals Z [j and W [j applying the non
ommutative derivative whi
h is
dened by
j (y) f (j ) = (d) (x y) f (j ) ;
j (x)
(3.2.51)
232
CHAPTER 3.
1=N
EXPANSION
... ...
...
Z [j = W [j Z [j ;
(3.2.52)
W [j Z [j = W [Z [j j :
(3.2.53)
j (x) = J (x)W [j ;
(3.2.55)
More about this approa
h to the generating fun
tionals for planar graphs
an
be found in Ref. [CLS82.
Solve iteratively Eq. (3.2.54) for the Gaussian
ase.
In the Gaussian
ase, only G(2) is nonvanishing whi
h yields
Problem 3.15
Solution
(3.2.56)
W [j = 1
where the propagator D is given by Eq. (3.2.4). Using Eq. (3.2.52), we get expli
itly
Z
dd x dd y D (x y ) j (x) Z [j j (y ) Z [j :
a)
b)
Fig. 3.19: Diagrams for the gluon propagator with a quark loop whi h is represented
by the single lines. The diagram a) involves one quark loop and has no
losed index lines so that its order is g2 1=N
. The diagram b) involves
three loops one of whi
h is a quark loop. Its order is g6 N
2 1=N
.
Z [j =
1
R
;
1 + g 2 dd x dd y D (x y ) j (x) Z [j j (y )
(3.2.57)
(3.2.58)
Z [j =
1 + g2 j
:
D
j
D j
1 + g2 j
D j
1 + g2 j
.
(3.2.59)
..
g2 j D j ;
Z [j = 1 g 2
............................
....................... ...........................
.
.
.
.... ...
... ...
... ...
.. ..
... ... ........ ........
... ..
..
.. ....
..
.
.
...
... ...
.
.
.. ....
..... .....
.
... ...
.
.
.
.
.
.
.
.
.
... ....
. ..
.... ......
... ...
...... ......... .....................
............. .............
............
(3.2.54)
Z [j = W [J :
233
MULTICOLOR QCD
.......................
......
.........
.....
....
.
.
.
...
..
...
.....
...
..
...
...
...
...
...
.
.
....
....
......
.....
..........
......................
....
3.2.
i j
/ ij
(3.2.60)
The arrow indi
ates, as usual, the dire
tion of propagation of a (
omplex)
eld . We shall omit these arrows for simpli
ity.
The diagram for the gluon propagator whi
h involves one quark loop is
depi
ted in Fig. 3.19a. It has two three gluon verti
es and no
losed index
234
CHAPTER 3.
1=N
EXPANSION
g2 N1 :
(3.2.61)
Analogously, the order of a more
ompli
ated treeloop diagram in Fig. 3.19b,
whi
h involves one quark loop and two
losed index lines, is
Fig. 3.19b
g6N
2 N1 :
(3.2.62)
It is evident from this
onsideration that quark loops are not a
ompanied
by
losed index lines. One should add a
losed index line for ea
h quark loop
in order for a given diagram with L quark loops to have the same doubleline representation as for pure gluon diagrams. Therefore, given Eq. (3.2.22),
diagrams with L quark loops are suppressed at large N
by
1 L+2genus
L quark loops
:
(3.2.63)
N
The singleline representation of the quark loops is similar to the one of
the external boundary in Fig. 3.15. Moreover, su
h a diagram emerges when
one
al
ulates perturbative gluon
orre
tions to the va
uum expe
tation value
of the quark operator
1
;
O =
N
(3.2.64)
where the fa
tor of 1=N
is introdu
ed to make it O(1) in the largeN
limit.
Therefore, the external boundary
an be viewed as a single line asso
iated
with valen
e quarks. The dieren
e between virtual quark loops and external
boundaries is that ea
h of the latter
omes along with the fa
tor of 1=N
due
to the denitions (3.2.45) and (3.2.64).
In order to prove Eqs. (3.2.22) and its quark
ounterpart (3.2.63), let
us
onsider a generi
diagram in the index spa
e whi
h has n(3)
0 threepoint
(4)
verti
es (either threegluon or quarkgluon ones), n0 fourgluon verti
es, n1
propagators (either gluon or quark ones), n2
losed index lines, L virtual
quark loops and B external boundaries. A typi
al su
h diagram is depi
ted
in Fig. 3.20. Its order in 1=N
is
(4)
1 n(3)
g 0 +2n0 N
n2
B
N
N n2
(4)
n(3)
0 =2 n0 B
(3.2.65)
3.2.
235
MULTICOLOR QCD
..................................
...............
.........
........
.......
.
.
.
.
.
......
......
.
.....
.
.
.
.
.
....
.
.
.
....
.
.
.....
...
.
...
....
..................
.
.
.
.
...
.
.
.
.
.
.
.
. .................................
.
.
.
...
.
.
....
...........
... .....
..
.
.
.
.
.
.........................
....
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
. . ............. ............ ... ...
......................
.
...
..
... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
............
.
...
... ...
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
...
... ...
.
. . .. ..
..
... .... ................. ..............................
.
...
.
.
.
.... ..........
...
..
...... ........... .........................
...
..
..............................
...
... ..
....
...
...
....
...
.... .......
.....
.
.
.
.
.
.....
.
...
...
......
.....
..... .......
.......
.....
..
.
.
.
.
........
.
.
.....
...............
...... ..........................
Fig. 3.20: Generi diagram in the index spa e whi h has L = 1 quark loop and B = 1
loop asso
iated with the external boundary. Its order in 1=N
is des
ribed
by Eq. (3.2.71).
as is already explained. The extra fa
tor of 1=N
B is due to the extra normalization fa
tor of 1=N
in operators asso
iated with external boundaries.
The number of propagators and verti
es are related by
(4)
2n1 = 3n(3)
0 + 4n0 ;
(3.2.66)
sin
e three and fourpoint verti
es emit three or four propagators, respe
tively, and ea
h propagator
onne
ts two verti
es. Using the relation (3.2.66),
we rewrite the RHS of (3.2.65) as
(3)
(4)
N
n2 n0 =2 n0 B = N
n2 n1 +n0 B ;
(3.2.67)
(3.2.68)
is introdu
ed.
The exponent on the RHS of Eq. (3.2.67)
an be expressed via the Euler
hara
teristi
s of a given graph of genus h. Let us rst mention that a
proper Riemann surfa
e, whi
h is asso
iated with a given graph, is open and
has B + L boundaries (represented by single lines). This surfa
e
an be
losed
by atta
hing a
ap to ea
h boundary. The single lines then be
ome double
236
CHAPTER 3.
1=N
EXPANSION
lines together with the lines of the boundary of ea
h
ap. We have already
onsidered this pro
edure when dedu
ing Eq. (3.2.63) from Eq. (3.2.22).
The number of fa
es for a
losed Riemann surfa
e
onstru
ted in su
h a
manner is n2 + L + B , while the number of edges and verti
es are n1 and n0 ,
respe
tively. Euler's theorem says that
2 2h = n2 + L + B
n 1 + n0 :
(3.2.69)
N n2 n1 +n0 B
N 2 2h L 2B :
(3.2.70)
We have thus proven that the order in 1=N
of a generi
graph does not
depend on its order in the
oupling
onstant and is
ompletely expressed via
the genus h and the number of virtual quark loops L and external boundaries
B by
generi
graph
1
N
2h+L+2(B 1)
(3.2.71)
3.2.
237
MULTICOLOR QCD
The fa
t that the a
tion is O(N
2 ) in the largeN
limit is a generi
property
of the models des
ribing matrix elds. It will be
ru
ial for developing saddlepoint approa
hes at large N
whi
h are
onsidered below.
Rederive the formula (3.2.71) using the
alligraphi
notations.
The propagator of the Aeld is g 2 while both three and fourgluon
verti
es are now g 2 as a
onsequen
e of Eq. (3.2.73). The
ontribution of a
generi
graph is now of the order
Problem 3.16
Solution
g 2 n1 n0 N
n2 B
N n2
n1 +n0 B
(3.2.74)
for g 2 1=N
. This
oin
ides with the RHS of Eq. (3.2.67) whi
h results in
Eq. (3.2.71).
i j
/ ij :
(3.2.75)
= Nf :
(3.2.76)
NNf
(3.2.77)
238
CHAPTER 3.
1=N
EXPANSION
a)
Nf
b)
Fig. 3.21: Diagrams with quark loops in the Veneziano limit. The
olor and
avor
indi
es of a quark loop whi
h are represented by the solid and dashed single
lines, respe
tively. The diagram a) is g2 Nf Nf =N
. The diagram b)
is g6 N
2 Nf Nf =N
.
in the 't Hooft limit. The ratio on the LHS of Eq. (3.2.77) is experimentally
10{15% for the meson. A hope to solve QCD in the 't Hooft limit is a hope
to des
ribe QCD with this a
ura
y.
An alternative largeN
limit of QCD when Nf N
as N
! 1 was
proposed by Veneziano [Ven76. Some diagrams for the gluon propagator,
whi
h involve one quark loop, are depi
ted in Fig. 3.21. The dashed single line
represents propagation of the
avor index. Ea
h
losed loop of the dashed line
is asso
iated with the fa
tor Nf a
ording to Eq. (3.2.76). This is analogous
to the ve
tor models whi
h exa
tly des
ribe the O(Nf )
avor symmetry in
these notations.
The diagrams in Fig. 3.21
ontribute, respe
tively,
N
(3.2.78)
Fig. 3.21a g2 Nf f
N
and
N
Fig. 3.21b g6N
2 Nf f
(3.2.79)
N
in the limit (3.2.13). Likewise, a more general diagram with L quark loops
will
ontribute
L quark loops
Nf L 1 genus
:
N
N
2
239
MULTICOLOR QCD
We see from Eq. (3.2.80) that quark loops are not suppressed at large N
in the Veneziano limit
...................
............... ...........................
................
..... ....
.
.
..
.... ..
... ...
... ...
..... ............
.
.
.
.... ...
.
.
... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ...
..
.
... ..
... ... ............. ...........
.... ....
... ....
... ..
.... .....
..... ....
...............
.............. ...........................
...............
..................
......... .... .... .................
.... ....
......... ...
........ .
... ....
.. ...
........
.. ..
..
... ..
.. ..
... ..
.. ....
... ...
.
.
. ...
.... ...
...... . .... . . .... ............
.......... .... ..............
............
3.2.
(3.2.80)
This formula obviously follows from Eq. (3.2.63) sin
e ea
h quark loop results
in Nf .
N ! 1
(3.2.81)
D D e
R 4
b +m
d x r
= e Sp ln rb +m
(3.2.82)
as is dis
ussed in Subse
t. 1.2.2. The tra
e in the exponent involves summation both over
olor and
avor indi
es, so that
Sp ln
rb + m N Nf :
(3.2.83)
240
CHAPTER 3.
...............
...... .... .....
.. ...... .... ...
..
.
.. ..
..
.. ..
..
... ..
... ..
..
... ...
.... . .. ..
...... .. .... ...
..............
1=N
EXPANSION
b =
241
MULTICOLOR QCD
.........
.. ......................
.... ....
..
..
.. ...
.. ....
..
..
.. ..
.. ....
..
.
.. .... .........
..... .... .....
...............
g6 N Nf2 Nf =N
3.2.
All these hadron properties (ex
ept the last one) approximately agree with
experiment, and were wellknown even before 1974 when multi
olor QCD was
introdu
ed. Glueballs are not yet dete
ted experimentally (maybe be
ause of
their property listed in the item 4).
The va
uum expe
tation values of several
olorless or white operators, whi
h
are singlets with respe
t to the gauge group, fa
torize in the largeN
limit
of QCD (or other matrix models). This property is similar to that already
dis
ussed in Subse
t. 3.1.5 for the ve
tor models.
The simplest gaugeinvariant operator in a pure SU(N
) gauge theory is
the square of the nonAbelian eld strength:
O (x) =
1
2 (x) :
tr F
N
2
(3.2.85)
1
tr F 2 (x)
N
2
1 a
a (x)
F (x) F
2N
2
1:
(3.2.86)
N12 N
2 N12 N
2 1 ;
(3.2.87)
242
CHAPTER 3.
......................
......................................
.........
... ..
.
... ..
.. .
... ...
.. ....
.
... ......
.
.
.
.
.
............................
....................
......................
......................................
.........
... ..
.
... ..
.. .
... ...
.. ....
.
... ......
.
.
.
.
.
............................
....................
......................
..... ................... .....
........................ ...........
... ... ................ .... ..
... .... . ..
... ...
.......... .....................
............................
..............
a)
........
....................... ......................
..............................................................................................
.
.
.
.
.
.
.............
.... .....
..... ....
.........
... ...
.. ..
.. ..
.... ....
.. ..
... ...
... ...
........
.
.
.........
..... ...
............
...........
................
................................ ............................................
.................... ......... ...................
.................
1=N
EXPANSION
......................
.....................................
.... .....
... ...
.
... ..
.
... .
... ..
.......
...........
.
.
.
................................
..............
b)
3.2.
Therefore, it is suppressed by 1=N
2 in the largeN
limit. For this graph, the
gluon line is emitted and absorbed by dierent operators O (x1 ) and O (x2 ).
This lowestorder example illustrates the general property that only (planar) diagrams with gluon lines emitted and absorbed by the same operators
survive as N
! 1. Hen
e,
orrelations between the
olorless operators
O (x1 ) and O (x2 ) are of order 1=N
2, so that the fa
torization property holds
as N
! 1:
........................
...................................
...........
... ...
... ...
..
... ...
... ..
.......
...........
.
.
.
.
.............................
..............
........................
....................................
...........
... ...
.
... .....
... ..
.. ..
... ....
....................................
.......................
d)
Fig. 3.23: Demonstration of the largeN
fa
torization to the lowest orders of perturbation theory. The
losed double line represents the average of the operator (3.2.85) to zeroth order in g. The diagrams a) and b), whi
h are asso
iated with the fa
torized part of the average on the LHS of Eq. (3.2.91),
are O(1). The diagrams
) and d), whi
h would violate the fa
torization,
are suppressed by 1=N
2 .
the normalization fa
tor 1=N
4 , and g2 due to two threegluon verti
es. Its
ontribution is
Fig. 3.23b g2 N
1 ;
(3.2.88)
i.e. also O (1) in the limit (3.2.13).
The graph in Fig. 3.23
is of the same type as the graph in Fig. 3.23a,
but the double lines now
onne
t two dierent operators. It has two
losed
index lines (the fa
tor N
2 ) and the normalization fa
tor 1=N
4 , so that its
ontribution
1
(3.2.89)
Fig. 3.23
2
N
is suppressed by 1=N
2.
The graph in Fig. 3.23d, whi
h is of the same order in the
oupling
onstant as the graph in Fig. 3.23b, involves only three
losed index lines (the
fa
tor N
3 ) and is of order 1=N
2:
1
N12 :
(3.2.90)
Fig. 3.23d g2
N
243
MULTICOLOR QCD
1
1
tr F 2 (x1 ) 2 tr F 2 (x2 )
N
2
N
1
1
1
2
2
:
tr F (x1 )
tr F (x2 ) + O
=
N
2
N
2
N
2
(3.2.91)
h O1 O n i
h O1 i h O n i + O
1
:
N
2
(3.2.92)
O =
Here
1
Nf N
(3.2.93)
1
[
;
; : : : :
(3.2.94)
2i
The lowestorder diagrams of perturbation theory for the average of two
quark operators (3.2.93) are depi
ted in Fig. 3.24. Estimation of their order
in 1=N
is analogous to that for the pure gluon graphs in Fig. 3.23.
The graph in Fig. 3.24a represents the zeroth order of perturbation theory
for the average of two quark operators. It involves two
losed
olor and
= I;
5 ;
; i
5 ; =
244
CHAPTER 3.
... .... .
... ..........................
.... ..
..........
.. .
..
.. ....
... ...
.
.
.
... ....
.
..
.
.
....................... ... .
.... .... ....
... .... .
... ..........................
.... ..
..........
.. .
..
.. ....
... ...
.
.
.
... ....
.
..
.
.
....................... ... .
.... .... ....
1=N
... .... .
... ............................
... ...
.........
.. .
..
..
... ....
.. ..
.
.
.
.
... ....
..
.
.
.
....................... ... .
.... .... ....
... .... .
... ..................... ...
.................... ...........
. . ... ...
..
... .... ................... .... ...
.
.
.
.
.
.
.
.
.
.
... ....
.. ... ..
......................... ... .
.... .... ....
a)
EXPANSION
b)
........
....................... ......................
............. .... .... .... .... .... .... .... ........................
.
.
.
.
.
.
.
.... .......
.... ..
.... ...
.... ...
... ...
.. ...
.... ..
... ...
.
... ...
.. ....
... ...
. ...
.... ....
.
.
.
..... ....
..
....... .... .
.. .. .....
............ .... .... .. .... .... .... ....................
.................... .. .... ...................
.................
........................
..... .... ..... .... ........
.. ...
....... ..
... ..
..
... .
.. ..
. .....
...... ...
.
.
........ .... ...........
..............
........................
..... .... ..... .... .........
.. ..
....... ..
. ..
... ..
.. .....
... ...
..... .... .... .... .........
.......................
d)
Fig. 3.24: Same as in Fig. 3.23 but for quark operators (3.2.93). The solid and dashed
lines are asso
iated with
olor and
avor indi
es, respe
tively. The diagram
a), whi
h
ontributes to the fa
torized part of the average on the LHS of
Eq. (3.2.99), are O(1). The diagrams b) and
), whi
h would violate the
fa
torization, are suppressed by 1=N
2 and 1=(Nf N
), respe
tively.
Nf2 N 2 )
N 21N 2 Nf2 N
2 1 :
f
(3.2.95)
N 21N 2 Nf2 g2 N
3
f
(3.2.96)
3.2.
245
MULTICOLOR QCD
N 1N
f
(3.2.97)
is suppressed by 1=(Nf N
).
Finally, the graph in Fig. 3.24d involves a gluon line whi
h is emitted by
one quark operator and absorbed by the other. It has one
losed
olor and two
losed
avor index lines (fa
tor Nf2 N
), the normalization fa
tor 1=(Nf N
)2 ,
and g2 due to two quarkgluon verti
es. Its
ontribution
Fig. 3.24d
(3.2.98)
246
CHAPTER 3.
1=N
EXPANSION
3.2.
N 2
(3.2.100)
C1
247
MULTICOLOR QCD
..
.......
...........
.
.
.
.
.
......... .
........ ........
........... ..............
......... ...
........ ........
....... ...............
...
....... ........
........... .................
.....
.........
.............
.
.
......
...........
.
.
.
.
.
........
......... .........
........... ...............
...... ..
......... .........
....... ...............
. ...... ....
....... ..........
...... ................
........
..........
......
C2
C1
....... .... ...... .... .... . .... ...... .... .... . .... ......
............. .......... . ......... .. . ......... . . .......... .. . ..........
......... . .................. ................. ................ . .................. ...................
.
.
.
.
.
.
.... .. .. ...... . .. ........ .. ...... . .................
......... ............. ............... .. ........... .. ........... ... ..........
........ ...... ..... ...... .. .. ...... .. .. ...... .. ... ........
..................... ................... ......................................
.... . . . . . ..
. . . . . ..
...... .... .... .... .... .... .... ....
.
............................................................................................................................... .................
...... .... .......... .... ......... ..... .......... .... .......... .... ...........
.
.
....... .... .......... .... ......... .... ......... .... .......... ................
....... . .. ...... . . ....... . ....... . . ...... . ...........
.... .... .... ....... .... ....... .... ....... .... ........ .... .........
C2
b)
a)
Fig. 3.25: Fa torization of the Wilson loop operators in the strong oupling expansion
spanned by ea
h individual loop. Its
ontribution is O(1) as N
! 1. A nonfa
torized part emerges from surfa
es of the type depi
ted in Fig. 3.25b, whi
h
onne
t two dierent Wilson loops. They look like a
ylinder and have two boundaries.
Their
ontribution is suppressed by 1=N
2 a
ording to the general formula (3.2.101).
Thus, we have proven the fa
torization property
D
1 tr U 1 tr U E =
N
C1 N
C2
1 tr U E D 1 tr U E + O 1 (3.2.102)
N
C1 N
C2
N
2
; (3.2.101)
where we have used Euler's theorem (3.2.69). In the limit (3.2.100), the
ontribution
does not depend on the order of the strong
oupling expansion and is
ompletely
determined by the number B of boundaries and the genus h of the surfa
e. This
is analogous to the perturbation theory. For the minimal surfa
e, we reprodu
e
previous results.
We are now in a position to analyze the order in N
of dierent terms in the
strong
oupling expansion of the average of two Wilson loop operators. The fa
torized part results from the surfa
es of the type depi
ted in Fig. 3.25a whi
h are
Find the relation between the Wilson loop averages in the fundamental and adjoint representations for a SU(N
) pure gauge theory at large N
.
Solution The
hara
ters in the fundamental and adjoint representations are related by Eq. (2.2.28). Using the fa
torization formula (3.2.102) with
oin
iding
ontours C1 and C2 , we get
Problem 3.18
Wadj (C ) = [Wfun (C )2 + O
1
:
N
2
(3.2.103)
As is dis
ussed in Chapter 2, the Wilson loop average in the fundamental representation obeys area law (2.2.75). The same is true at N
= 1 for the Wilson loop
average in the adjoint representation due to Eq. (3.2.103). In parti
ular, the string
248
CHAPTER 3.
1=N
EXPANSION
Kadj = 2Kfun :
(3.2.104)
On the other hand, the adjoint test quark
an be s
reened at nite N
by a gluon
produ
ed out of the va
uum. This is similar to the breaking of the
ux tube
in the fundamental representation by a quarkantiquark pair, whi
h is dis
ussed
in Subse
t 2.5.5. Therefore, the perimeter law (2.2.79) must dominate for large
ontours. The point is that the perimeter law appears due to
onne
ted diagrams
whi
h are suppressed as 1=N
2 :
3.2.
249
MULTICOLOR QCD
The idea is to rewrite the path integral over A for the Yang{Mills theory as that over a
olorless
omposite eld [A, likewise it was done in
Subse
t. 3.1.4 for the sigmamodel. The expe
ted new pathintegral representation of the partition fun
tion (3.2.106) would be something like
A
[A
Aa
DAa e
R 4 1 a a
d x 4 F F
(3.2.106)
DAa
e N :
(3.2.107)
2
e N
J [
(3.2.110)
J [
The largeN
fa
torization in QCD assumes that gaugeinvariant obje
ts behave as
numbers, rather than as operators. Likewise the ve
tor models, this
suggests that the path integral is dominated by a saddle point.
We already saw in Subse
t. 3.1.5 that the fa
torization in the ve
tor models does not mean that the fundamental eld itself, for instan
e ~n in the
sigmamodel, be
omes \
lassi
al". It is the
ase, instead, for a singlet
omposite eld.
We are now going to apply a similar idea to the Yang{Mills theory whose
partition fun
tion reads
(3.2.109)
The Ja obian
Z =
2
1
D [A e N
S[ :
a
(3.2.111)
2
2
D e N
J [ N
S[ ;
(3.2.112)
where S [ represents the Yang{Mills a
tion in the new variables. The new
\entropy" fa
tor D is O(1) be
ause the variable [A is a
olor singlet. The
large parameter N
enters Eq. (3.2.112) only in the exponent. Therefore, the
saddlepoint equation
an be immediately written:
S
J
=
:
Remembering that is a fun
tional of A :
[A ;
(3.2.113)
(3.2.114)
S
J
= (r F )a =
:
(3.2.115)
Aa
Aa
It diers from the
lassi
al Yang{Mills equation (3.2.108) by the term on the
RHS
oming from the Ja
obian (3.2.110).
250
CHAPTER 3.
1=N
EXPANSION
Given J [ whi
h depends on the pre
ise from of the variable [A,
Eq. (3.2.115) has a solution
A (x) = A l (x) :
(3.2.116)
Let us rst assume that there exists only one solution to Eq. (3.2.115). Then
the path integral is saturated by a single
onguration (3.2.116), so that the
va
uum expe
tation values of gaugeinvariant operators are given by their
values at this
onguration:
hOi
= O A l (x) :
(3.2.117)
A
l (x) = A
l (0) :
(3.2.118)
In this gauge, rotations must be equivalent to a global gauge transformation,
so that A
l (0) transforms as a Lorentz ve
tor.
In fa
t, the idea about su
h a master eld in multi
olor QCD may be
just wrong as was pointed out by Haan [Haa81. The
onje
ture about an
existen
e of only one solution to the mastereld equation (3.2.115) seems to
be somewhat in
orre
t. If several solutions exists, one needs an additional
averaging over these solutions. This is a very deli
ate matter, sin
e this
additional averaging must still preserve the fa
torization property. One might
better think about this situation as if A
l (0) would be an operator in some
Hilbert spa
e rather than a
valued fun
tion. This is simply be
ause A
l (0) is,
3.2.
251
MULTICOLOR QCD
3.2.8
1=N
A natural
andidate for the
omposite operator [A from the previous Subse
tion is given by the tra
e of the nonAbelian phase fa
tor for
losed
ontours  the Wilson loop. It is labeled by the loop C in the same sense as the
eld A (x) is labeled by the point x, so we shall use the notation
(C )
[A =
H
1
tr P e ig C dx A (x) :
N
(3.2.119)
252
CHAPTER 3.
1=N
EXPANSION
S
hwinger{Dyson equations are a
onvenient way of performing the semi
lassi
al expansion, whi
h is an alternative to the path integral. Let us illustrate an idea how to do this by an example of the '3 theory whose S
hwinger{
Dyson equations are given by Eq. (1.2.47).
The RHS of Eq. (1.2.47) is proportional to the Plan
k's
onstant h as is
explained in Subse
t 1.2.5. In the semi
lassi
al limit h ! 0, we get
12 + m2 h' (x1 ) : : : ' (xn )i + '2 (x1 ) : : : ' (xn ) = 0 ; (3.2.120)
2
whose solution is of the fa
torized form
provided that
(3.2.121)
(3.2.122)
obeys
2 + m2 '
l (x) + '2
l (x) = 0 :
(3.2.123)
2
Equation (3.2.123) is nothing but the
lassi
al equation of motion for
the '3 theory, whi
h spe
ies extrema of the a
tion (1.2.22) entering the
path integral (1.2.2). Thus, we have reprodu
ed, using the S
hwinger{Dyson
equations, the wellknown fa
t that the path integral is dominated by a
lassi
al traje
tory as h ! 0. It is also
lear how to perform the semi
lassi
al
expansion in h in the language of the S
hwinger{Dyson equations: one should
solve Eq. (1.2.47) by iterations.
The reformulation of multi
olor QCD in terms of the loop fun
tionals
(C ) is, in a sense, a realization of the idea of the master eld in the weak
sense, when the master eld a
ts as an operator in the spa
e of loops. The
loop equation of the next Se
tion will be a sort of the mastereld equation
in the loop spa
e.
3.2.
253
MULTICOLOR QCD
Then (C ) reads
N
1 X
e igj (C ) :
(C ) =
N
j=1
(3.2.125)
/ ij ;
(3.2.126)
one gets
[ (C ) ; (C 0 )
g2 N1 N12
(3.2.127)
254
1=N
CHAPTER 3.
EXPANSION
3.3.
x1
This approa
h is espe
ially useful in the largeN
limit where everything
is expressed via the va
uum expe
tation value of (C )  the Wilson loop
average. Observables are given by summing the Wilson loop average over
paths with the same weight as in free theory. The Wilson loop average obeys
itself a
lose fun
tional equation  the loop equation.
We begin this Se
tion with presenting the formulas whi
h relate observables to the Wilson loops. Then we translate quantum equation of motion
of Yang{Mills theory into loop spa
e. We derive the
losed equation for the
Wilson loop average as N
! 1 and dis
uss its various properties, in
luding
a nonperturbative regularization. Finally, we brie
y
omment on what is
known about solutions of the loop equation.
h (C1 ) (Cn ) i ;
(3.3.1)
whi
h are analogous to the npoint Green fun
tions (1.2.45). The appropriate
formulas for the
ontinuum theory
an be found in Ref. [MM81.
Great simpli
ations o
ur in these formulas at N
= 1, when all observables are expressed only via the oneloop average
W (C ) =
h (C ) i N1 tr P e ig
C dx A
...........................
.....
.....
..
...
....
...
.
.
.....
..
.
.
.
.
.
.
.
.
.
.
.
.
.......
.....
.....
...
................................
(3.3.2)
........................
.....
....
.
...
....
.
.
..
.
.
...
.....
.
.
.
.
.
.
.......
......
.......... x
...
...... .......................
3
......
x2 ............
x2
x1
Wn (C1 ; : : : ; Cn ) =
255
b)
a)
Fig. 3.26: Contours in the sum over paths representing observables: a) in Eq. (3.3.3)
and b) in Eq. (3.3.4). The
ontour a) passes two nailed points x1 and x2 .
The
ontour b) passes three nailed points x1 , x2 , and x3 .
This is asso
iated with the quen
hed approximation dis
ussed on p. 164.
For example, the average of the produ
t of two
olorless quark ve
tor
urrents (3.2.93) is given at large N
by
(x1 ) (x2 ) =
C 3x1 ;x2
J (C ) h (C ) i ;
(3.3.3)
where the sum runs over
ontours C passing through the points x1 and x2 as is
depi
ted in Fig. 3.26a. An analogous formula for the (
onne
ted)
orrelators
of three quark s
alar
urrents reads
J (C ) h (C ) i ;
(3.3.4)
where the sum runs over
ontours C passing through the three points x1 , x2 ,
and x3 as is depi
ted in Fig. 3.26b. A general (
onne
ted)
orrelator of n
quark
urrents is given by a similar formula with C passing through n points
x1 , . . . , xn (some of them may
oin
ide).
The weights J (C ) in Eq. (3.3.3) and J (C ) in Eq. (3.3.4) are
ompletely
determined by free theory. If quarks were s
alars rather than spinors, then
we would get
J (C ) = e
1 2
2m
1
2 0
dt z_2 (t)
= e mL(C )
where L(C ) is the length of the ( losed) ontour C , as was shown in Subse t. 1.1.6. Using the notation (1.1.152), we an rewrite Eq. (3.3.4) for s alar
256
CHAPTER 3.
quarks as
onn
X0
1=N
EXPANSION
h (C ) i :
(3.3.6)
Therefore, we get the sum over paths of the Wilson loop, likewise in Subse
t. 1.1.7 and Problem 2.5.
For spinor quarks, an additional disentangling of the gammamatri
es is
needed. This
an be done in terms of a path integral over the momentum
variable, with k (t) (0 t ) being an appropriate traje
tory. The result
reads9
J (C ) =
and
J (C ) =
Dk (t) sp P e
R
0
R
(3.3.7)
;
(3.3.8)
where the values t1 and t2 of the parameter t are asso
iated with the points
x1 and x2 in Eq. (3.3.3), and the symbol of P{ordering puts the matri
es
and
at a proper order.
0
and represent the matrix element of the exponential of the Dira
operator as
y e rb+m x
= e m P e
R
dt r
b (t) (d)
0
(x
y) :
(3.3.10)
z(0)=x
9 See,
, Ref. [BNZ79.
e.g.
Dz (t)
Dp(t) e
R
i 0 dt p (t)z_ (t)
(3.3.11)
3.3.
257
P e
= e m
Dz (t)
i.e.
Dp(t)
z(0)=x
dt
ip
(
t
)
z
_
(
t
)
ip
[
(
(t)+A (t))
(t)+ (t)z_ (t) (d) (x
0
R
y ) ; (3.3.12)
(3.3.13)
and pro
eeding as in Problem 1.12, we represent the RHS of Eq. (3.3.12) by
b +m
y e r
x
= e m
Dz (t)
Dk(t)
z(0)=x
dt
ik
(
t
)
z
_
(
t
)
[ (
(t)) z_ (t)A (t)+ (t)z_ (t) (d) (x
0
R
P e
= e m
Dz (t)
z(0)=x
z( )=y
Pe
Ry
Dk(t) P e x dz A (z)
R
i 0 dt k (t)[z_ (t)
(t)
y)
(3.3.14)
where the rst P{exponential on the RHS depends only on
olor matri
es (it is
nothing but the nonAbelian phase fa
tor), and the se
ond one depends only on
spinor matri
es. In Ref. [BNZ79, Eq. (3.3.14) is derived by dis
retizing paths.
Equation (3.3.14) leads to Eqs. (3.3.7) and (3.3.8).
(3.3.15)
258
CHAPTER 3.
.........................
....
....
...
....
...
....
.
...
.
.. .
....
.
.
...
.
..
...
.....
...............................
1=N
EXPANSION
.........................
..
......
....
...... ... .
....
........
....
... . ..
..
... .. ....
...
.....
.
.
.
....
..
...
.....
....... ....................
.....
a)
b)
Fig. 3.27: Examples of a) smooth
ontour and b)
ontour with a
usp. The tangent
ve
tor to the
ontour jumps through angle
at the
usp.
where a is the
uto, L(C ) is the length of C , and Wren (C ) is nite when
expressed via the renormalized
harge gR . The exponential fa
tor is due to the
renormalization of the mass of a heavy test quark, whi
h was already dis
ussed
in the Remark on p. 117. This fa
tor does not emerge in the dimensional
regularization where d = 4 ". The multipli
ative renormalization of the
smooth Wilson loop was shown in Refs. [GN80, Pol80, DV80.
If the
ontour C has a
usp (or
usps) but no selfinterse
tions as is
illustrated by Fig. 3.27b, then W (C ) is still multipli
atively renormalizable [BNS81:
W (C ) = Z ( ) Wren (C ) ;
(3.3.16)
while the (divergent) fa
tor Z (
) depends on the
usp angle (or angles)
(or
's) and Wren (C ) is nite when expressed via the renormalized
harge gR .
Cal
ulate divergent parts of the Wilson loop average (3.3.2) for
ontours without selfinterse
tions to order g 2 . Consider the
ases of a smooth
ontour C and a
ontour with a
usp.
2
Solution Expanding the Wilson loop average (3.3.2) in g (see Eq. (3.2.47) and
Problem 2.2), we get
Problem 3.20
W (C ) = 1 + W (2) (C ) + O(g 4 )
with
W (2) (C ) =
g 2 (N
1)
4N
dx
dy D (x y ) ;
(3.3.17)
(3.3.18)
3.3.
259
Sin
e the
ontour integral in Eq. (3.3.18) diverges for x = y , we introdu
e the
regularization by
Reg: 1
D (x y ) =
(3.3.19)
) 42 (x y)2 + a2
with a being the ultraviolet
uto. Parametrizing the
ontour C by the fun
tion
z ( ), we rewrite the
ontour integral in Eq. (3.3.18) as
I
I
Z
Z
z_ (s)z_ (s + t)
1
dx dy
=
ds dt
: (3.3.20)
(x y )2 + a2
[z (s + t) z (s)2 + a2
C
260
CHAPTER 3.
1=N
EXPANSION
3.3.
261
The se
ond term in this formula is asso
iated with the bremsstrahlung radiation of
a parti
le
hanging its velo
ity when passing the
usp. The answers in the Abelian
and nonAbelian
ases
oin
ide to this order in g 2 .
= h
Aa (x)
Q [A :
(3.3.27)
H
1
tr P r F (x) e C d A
N
2
H
g
tr
P e C d A ;
=
2N
A (x)
(3.3.28)
Akl
N
(x)
Cyx
6 Cxy
Fig. 3.28: Contours Cyx and Cxy whi
h enter the RHS's of Eqs. (3.3.28) and (3.3.32).
whi
h is a
onsequen
e of
Aa (y)
= (d) (x y) ab :
(3.3.30)
Ab (x)
The se
ond term in the parentheses in Eq. (3.3.29)  same as in Eq. (3.2.6)
 is be
ause A is a matrix from the adjoint representation of SU(N
).
By using Eq. (3.3.29), we get for the variational derivative on RHS of
Eq. (3.3.28):
I
H
d A
C
tr
=
dy (d) (x y)
Pe
A (x)
R
R
d A
1
1
1
C d A : (3.3.31)
tr P e Cyx
tr P e Cxy
tr
P
e
3
N
N
N
The
ontours Cyx and Cxy , whi
h are depi
ted in Fig. 3.28, are the parts of
the loop C : from x to y and from y to x, respe
tively. They are always
losed
due to the presen
e of the deltafun
tion. It implies that x and y should be
the same points of spa
e but not ne
essarily of the
ontour (i.e. they may be
asso
iated with dierent values of the parameter ).
We nally rewrite Eq. (3.3.28) as
H
1
tr P r F (x) e C d A
N
I
1
(
d
)
h (C ) i (3.3.32)
= dy (x y) h (Cyx ) (Cxy ) i
N2
d A
=
g2 N
:
2
(3.3.33)
262
CHAPTER 3.
1=N
EXPANSION
Noti
e that the RHS of Eq. (3.3.32) is
ompletely represented via the (
losed)
Wilson loops.
Prove the
an
ellation of the
ontributions of the gaugexing and
ghost terms in the Lorentz gauge.
Solution The Yang{Mills a
tion, asso
iated with the Lorentz gauge, reads
3.3.
263
We have thus proven that the
ontribution of gaugexing and ghost terms in
Eq. (3.3.36) are mutually
an
elled, when applied to the Wilson loop (and in fa
t
to any gaugeinvariant fun
tional).
Problem 3.21
Sg:f : =
1
g2
h
i
2 + 1 tr ( A )2 :
dd x 1 tr F
2
(3.3.34)
Sin
e (C ) is gauge invariant, the (innite) groupvolume fa
tors, in the numerator
and denominator in the denition of the average,
an
el when xing the gauge (see
the Remark on p. 113), and we get
W (C )
DA e S (C )
R
DA e S
DA det r e Sg:f : (C )
R
;
DA det r e Sg:f :
(3.3.35)
w:s:
b
rab
F (x) = h Aa (x)
+ 1 Aa + f ab
Gb
x0 = x; x; A ;
(3.3.36)
x
where Gb
(x0 ; x; A) is the Green fun
tion of the ghost in an external eld A .
Applying this equation to the Wilson loop and using the gauge Ward identity (the
Slavnov{Taylor identity), we transform the
ontribution from the se
ond term on
the RHS to
E
1D1
tr A U (Cxx )
N
g:f :
I
D
E
1
2
b
( ) G
a (; x; A)
d
= g
tr U (Cx ) ta U (Cx ) tb
r
N
x
g:f :
C
g2
g2
D
1 tr U (C ) ta U (C ) tb Gba (; x; A)E
d
x
x
N
x
g:f :
(3.3.37)
whi h exa tly an els the ontribution from the ghost term in Eq. (3.3.36).
s() =
Z
0
d0 x_ 2 (0 )
(3.3.38)
10 Let us remind that L2 stands for the Hilbert spa
e of fun
tions x () whose square is
R
integrable over the Lebesgue measure: f dx2 () < 1. We already mentioned this in
0
the Remark on p. 19.
264
1=N
CHAPTER 3.
EXPANSION
F (C )
(x)
x
> A
1 4
j j F
F
13
A5
(3.3.39)
where an innitesimal loop C (x) is atta
hed to a given loop at the point x
in the plane and j j stands for the area en
losed by the C (x). For
a re
tangular loop C (x), one gets
(3.3.40)
x F (Cxx )
1 4
jx j F
 A
xq
F
13
A5
(3.3.41)
where x is an innitesimal path along whi
h the point x is shifted from the
loop and jx j stands for the length of the x .
These two dierential operations are welldened for so
alled fun
tionals
of the Stokes type whi
h satisfy the ba
ktra
king
ondition  they do not
hange when an appendix passing ba
k and forth is added to the loop at some
point x:
0
F
1
A
F
(3.3.42)
, Ref. [Tav93 whi h ontains denitions of path and area derivatives in this
e.g.
language.
3.3.
265
L(C ) of the loop C is not a Stokes fun
tional, sin
e the lengths of
ontours
on the LHS and RHS of Eq. (3.3.42) are dierent.
For the Stokes fun
tionals, the variation on the RHS of Eq. (3.3.39) is
proportional to the area en
losed by the innitesimally small loop C (x)
and does not depend on its shape. Analogously, the variation on the RHS of
Eq. (3.3.41) is proportional to the length of the innitesimal path x and
does not depend on its shape.
If x is a regular point (like any point of the
ontour for the fun
tional (3.3.2)), the RHS of Eq. (3.3.41) vanishes due to the ba
ktra
king
ondition (3.3.42). In order for the result to be nonvanishing, the point x
should be a marked (or irregular) point. A simple example of the fun
tional
with a marked point x is
R
tr a
d A ()
a [Cxx
(3.3.43)
t P e Cxx
N
with the SU(N
) generator ta inserted in the pathordered produ
t at the
point x.
The area derivative of the Wilson loop is given by the Mandelstam formula
H
H
1
1
tr P e C d A =
tr P F (x) e C d A : (3.3.44)
(x) N
N
In order to prove it, it is
onvenient to
hoose C (x) to be a re
tangle
in the plane, as was done in Problem 2.8, and straightforwardly use the
denition (3.3.39). The sense of Eq. (3.3.44) is very simple: F is a
urvature
asso
iated with the
onne
tion A , as we dis
ussed in the Remark on p. 99.
The fun
tional on the RHS of Eq. (3.3.44) has a marked point x, and is of
the type in Eq. (3.3.43). When the path derivative a
ts on su
h a fun
tional
a
ording to the denition (3.3.41), the result reads
H
H
1
1
tr P r B (x) e C d A ; (3.3.45)
x tr P B (x) e C d A =
N
N
where
r B = B [A ; B
(3.3.46)
is the
ovariant derivative (2.1.11) in the adjoint representation (see also
Problem 2.7).
Combining Eqs. (3.3.44) and (3.3.45), we nally represent the expression
on the LHS of Eq. (3.3.28) (or Eq. (3.3.32)) as
H
H
1
1
266
CHAPTER 3.
Ordinary spa e
EXPANSION
(x)
ovariant derivative
The standard variational derivative, =x (),
an be expressed via the path
and area derivatives by the formula
area derivative
m
X
= x_ ()
+ xi ( i ) ;
x ()
(x()) i=1
r^F =0
Bian hi identity
r F
S
hwingerDyson
equations
Loop
equations
= =A
loop spa e.
i.e. via the a
tion of the path and area derivatives on the Wilson loop. It is
therefore rewritten in loop spa
e.
A resume of the results of this subse
tion is presented in Table 3.2 as a
vo
abulary for translation of Yang{Mills theory from the language of ordinary
spa
e in the language of loop spa
e.
(x)
(C ) = 0 ;
(3.3.48)
H
1
tr P r F (x) e C d A = 0 :
N
(3.3.50)
where the sum on the RHS is present for the
ase of a fun
tional having m
marked (irregular) points xi x(i ). A simplest example of the fun
tional
with m marked points is just a fun
tion of m variables x1 ; : : : ; xm .
By using Eq. (3.3.50), the path derivative
an be
al
ulated as the limiting
pro
edure
Z+0
x
(
)
d0
=
:
(3.3.51)
x (0 )
0
Table 3.2: Vo abulary for translation of Yang{Mills theory from ordinary spa e in
x
267
3.3.
Therefore, Eq. (3.3.48) represents the Bian hi identity (2.1.19) in loop spa e.
Loop spa e
[A phase fa tor
rx
1=N
(3.3.49)
The result is obviously nonvanishing only when x is applied to a fun
tional
with x() being a marked point.
It is nontrivial that the area derivative
an also be expressed via the
variational derivative [Pol80:
=
(x())
Z+0
d0 (0
0
)
:
x (0 ) x ()
(3.3.52)
The point is that the six
omponent quantity, = (x()), is expressed via
the four
omponent one, =x (), whi
h is possible be
ause the
omponents
of = (x()) are dependent due to the loopspa
e Bian
hi identity (3.3.48).
268
CHAPTER 3.
1=N
EXPANSION
or, using the denitions (3.3.1) and (3.3.2) of the loop averages, as
x
W (C )
(x)
I
1
(
d
)
W (C ) : (3.3.54)
= dy (x y) W2 (Cyx ; Cxy )
N
2
C
x
W (C ) = dy (d) (x y) W (Cyx ) W (Cxy ) (3.3.56)
(x)
C
as N
! 1.
Equation (3.3.56) is a
losed equation for the Wilson loop average in the
largeN
limit. It is referred to as the loop equation or the Makeenko{Migdal
equation.
To nd W (C ), Eq. (3.3.56) should be solved in the
lass of Stokes fun
tionals with the initial
ondition
W (0) = 1
(3.3.57)
for loops whi
h are shrunk to points. This is a
onsequen
e of the obvious
property of the Wilson loop
H
(3.3.58)
e 0 d A = 1
and the normalization
h1i
= 1
(3.3.59)
3.3.
269
of the averages.
The fa
torization (3.3.55)
an itself be derived from the
hain of loop
equations. Pro
eeding as before, we get
1 x
W (C ; : : : ; Cn )
(x) n 1
=
C1
h
1
(d) (x y ) W
dy
n
2
j 2 N
Cj
1
W ( C ; : : : ; Cn )
N
2 n 1
C1 Cj ; : : : ; Cj ; : : : ; Cn
W n ( C1 ; : : : ; C n ) :
(3.3.60)
x
W (C ; : : : ; Cn )
(x) n 1
=
(3.3.61)
C1
1
Wn (C1 ; : : : ; Cn ) = h (C1 )i : : : h (Cn )i + O
N
2
1
W (C1 ) W (Cn ) + O N 2
(3.3.62)
provided W (C ) obeys Eq. (3.3.56) whi
h plays the role of a \
lassi
al" equation in the largeN
limit. Thus, we have given a nonperturbative proof of
the largeN
fa
torization of the Wilson loops.
270
1=N
CHAPTER 3.
. . .
...........
...
.....
. . .
. . .
C + p
a)
3.3.
. . .
...
.....
..... .. .......
.. . .. .......
.......... ... .........
........ ....
.. ..
.....
...........
x1
p
p pp p
ppp p ..
pppp
ppppp ............
ppp
.
.
.
pppp pp
.
ppp
..
.....
.
pp
..
pp
.......
pp
.
.
pp p
pp
...... .......
pp
pp
.
.
.
.
.
......... ........
ppp
pppp
.
.
.
.
p
...
..
.
ppp
(n) .................................................ppppp
p..p................................................
.
p
ppp
...
..
.....
pp
pp
........................
ppp
pp
p
p
ppp
p
..
........
pp
ppp
..
pp
......
ppp
.. ..
ppp pp
......
pppp
p
.
p
ppppp
.
pppp
ppp pp ..........
ppppppp
p
p
pp ppppp ..
p
p
p
p
p
p
p
p
.p.pppppppp ppppppppp pppppppppppppppp pp pppppppppp
ppppppp pp
xi
x2
b)
Fig. 3.29: Contours a) C + p and b) C p on the RHS of the latti e loop equation (3.3.63).
Fig. 3.30: Graphi representation of the terms on the RHS of Eq. (3.3.64).
Problem 3.22
X
X [W
(l) WCyx WCxy :
C +p WC p =
2
2N
p
l2C
271
x............
x ...................
....
..
...........
...
............. ...........
........ ...
.. ..
....
...........
...........
EXPANSION
(3.3.63)
Here the
ontours C + p and C p are obtained from Cxx by adding the boundary
of the plaquette p ( p means that the orientation of the boundary is opposite)
and the sum over p goes over the 2(d 1) plaquettes involving the link at whi
h the
shift of Ux; is performed. This
ontours are depi
ted in Fig. 3.29.
The sum on the RHS goes over the links belonging to the
ontour C . The unit
ve
tor (l) = 0; 1 stands for the proje
tion of the (oriented) link l 2 C on the axis
( (l) = 1, 1 or 0 when the dire
tions are parallel, antiparallel or perpendi
ular,
respe
tively). The point y is dened as the beginning of the link l if it has the
positive dire
tion, or as the end of l if it has the negative one. Su
h an asymmetry
is due to the fa
t that we have performed the right shift (2.2.22) of Ux; . The
Krone
ker symbol xy guarantees that Cyx and Cxy are always
losed.
Equation (3.3.63) is a latti
e regularization of the
ontinuum loop equation
(3.3.56). The loop equation on the latti
e was rst dis
ussed in Refs. [Foe79, Egu79.
The perturbationtheory expansion of the Wilson loop average
an be
al
ulated from Eq. (3.2.47) whi
h we represent in the form
I
I
1 I
X
1
2
W (C ) = 1 +
dx1 dx2 : : : dxnn
n=2 C
C
C
(n)
(1; 2; : : : ; n) G1 2 n (x1 ; x2 ; : : : ; xn ) ;
(3.3.64)
272
CHAPTER 3.
1=N
EXPANSION
3.3.
273
Problem 3.23
Solution
a)
b)
Fig. 3.31: Planar diagrams for W (C ): a) of order with gluon propagator, and of
order 2 b) with two nonintera
ting gluons and
) with the threegluon
vertex. Diagrams of order 2 with oneloop insertions to gluon propagator
with some unknown fun tion D (x y ). Its tensor stru ture reads
The largeN
loop equation (3.3.56) des
ribes the sum of the planar diagrams. Its iterative solution in reprodu
es the set of planar diagrams for
W (C ) provided the initial
ondition (3.3.57) and some boundary
onditions
for asymptoti
ally large
ontours are imposed.
Equation (3.3.64)
an be viewed as an ansatz for W (C ) with some unknown fun
tions G(n1)n (x1 ; : : : ; xn ) to be determined by the substitution
into the loop equation. To preserve symmetry properties of W (C ), the fun
tions G(n) must be symmetri
under a
y
li
permutation of the points 1, : : :,
n and depend only on xi xj (translational invarian
e). A main advantage of
this ansatz is that it automati
ally
orresponds to a Stokes fun
tional, due to
the properties of ve
tor integrals, and the initial
ondition (3.3.57) is satised.
The a
tion of the area and path derivatives on the ansatz (3.3.64) is easily
al
ulable. For instan
e, the area derivative reads
I
1 I
X
W (C )
=
dx1 1 : : : dxnn
(1; 2; : : : ; n)
(z )
n=1
C
C
h
(n+1)
z
z
G1 n (z; x1 ; : : : ; xn )
i
n+2)
+ ( ) G(
1 n (z; z; x1 ; : : : ; xn ) :
D (x y ) = D (x y ) + f (x y ) :
(3.3.68)
The se
ond (longitudinal) term in this formula does not
ontribute to W (C ) sin
e
the
ontour integral of this term vanishes in Eq. (3.3.67). We
an thus write
1
2
W (C ) = 1
dx
dy D (x y ) + O 2 :
(3.3.69)
The area derivative
an be easily
al
ulated by using the Stokes theorem, whi
h
gives
(z )
=
dx
dy D (x y )
0
I
2
dy D (z y )
and
sin
e
z
(z )
dx
dy D (z y )A
(3.3.70)
dy D (x y ) = 2
(3.3.65)
dy 2 D (z y )
(3.3.71)
dy D (x y ) = 0 :
(3.3.72)
dy
2 D (x
y) =
dy (d) (x y )
(3.3.73)
274
CHAPTER 3.
1=N
EXPANSION
whi h is equivalent to
2 D (x
y) =
(d) (x
y)
(3.3.74)
sin
e the
ontour C is arbitrary. The solution to Eq. (3.3.74) is unique, provided
D (x y ) de
reases for large x y , and re
overs the propagator (3.2.4).
dx x
W (C )
(x)
C
=
dx
(3.3.75)
Now both the operator on the LHS and the fun
tional on the RHS are s
alars
without labeled points and are welldened in loop spa
e. The operator on
the LHS of Eq. (3.3.75)
an be interpreted as an innitesimal variation of
elements of loop spa
e.
Equations (3.3.56) and (3.3.75) are
ompletely equivalent. A proof of
equivalen
e of s
alar Eq. (3.3.75) and original dve
tor Eq. (3.3.56) is based
on the important property of Eq. (3.3.56) whose both sides are identi
ally
annihilated by the operator x. It is a
onsequen
e of the identity (see Subse
t. 2.1.1)
(3.3.76)
r r F = 21 [F ; F = 0
3.3.
275
in the ordinary spa
e. Due to this property, the vanishing of the
ontour
integral of some ve
tor is equivalent to vanishing of the ve
tor itself, so that
Eq. (3.3.56)
an in turn be dedu
ed from Eq. (3.3.75).
Equation (3.3.75) is asso
iated with the so
alled se
ondorder S
hwinger{
Dyson equation
Z
a (x)
dd x r F
Aa (x)
Z
w:s:
(3.3.77)
= h dd x dd y (d) (x y) a
a
A (y) A (x)
in the same sense as Eq. (3.3.56) is asso
iated with Eq. (3.3.26). It is
alled
\se
ond order" sin
e the RHS involves two variational derivatives with respe
t
to A .
The operator on the LHS of Eq. (3.3.75) is a welldened obje
t in loop
spa
e. When applied to regular fun
tionals whi
h do not have marked points,
it
an be represented, using Eqs. (3.3.51) and (3.3.52), in an equivalent form
I
dx x
(x)
=
C
Zf
0
d
Z+0
0
d0
:
0
x ( ) x ()
(3.3.78)
As was rst pointed out by Gervais and Neveu [GN79b, this operator is
nothing but a fun
tional extension of the Lapla
e operator, whi
h is known
in mathemati
s as the Levy operator.12 Equation (3.3.75)
an be represented
in turn as an (inhomogeneous) fun
tional Lapla
e equation
W (C ) =
dx
(3.3.79)
Reg:
ab (d) (x y) =
) y Rab x = Rab (d)(x y)
(3.3.80)
12 See the book by Levy [Lev51, ited in the referen es to Chapter 1, and a re ent
review [Fel86.
276
CHAPTER 3.
with
Rab =
1=N
EXPANSION
2 2 ab
e a r =2 ;
3.3.
(3.3.81)
where r is the
ovariant derivative in the adjoint representation. The regularized version of Eq. (3.3.77) is
Z
a (x)
dd x r F
w:s:
h
dd x dd y y Rab x
:
Ab (y) Aa (x)
(3.3.82)
To translate Eq. (3.3.82) in loop spa
e, we use the pathintegral representation (see Problem 2.5)
E
D
y ab x
Dr(t) e
R 2
1 a
2 0
dt r_ 2 (t)
2 tr
a
t U (r
b
yx )t U (rxy )
r(0)=x
r(a2 )=y
(3.3.83)
with
U (ryx ) =
(3.3.84)
Dr(t) e
2
1 a2
2 0 dt r_ (t)
2 tr ta tb
r(0)=x
r(a2 )=y
2 2
= ab e a =2 (d) (x y) = ab
1
e
(2a2 )d=2
(x y)2
2a2
(3.3.85)
Cal
ulating the variational derivatives on the RHS of Eq. (3.3.82), using
Eq. (3.3.83) and the
ompleteness
ondition (3.2.6), we get as N ! 1:
Z
dd x dd y
E
D
y ab x
r(0)=x
r(a2 )=y
Dr(t) e
(C ) =
dx
R 2
1 a dtr_ 2 (t)
2 0
(Cyxrxy ) (Cxy ryx )
dy
(3.3.86)
rxy
ryx
6 Cxy
Fig. 3.32: Contours Cyx rxy and Cxy ryx whi
h enter the RHS's of Eqs. (3.3.86) and
(3.3.87).
where the
ontours Cyxrxy and Cxy ryx are depi
ted in Fig. 3.32. Averaging
over the gauge eld and using the largeN
fa
torization, we arrive at the
regularized loopspa
e Lapla
e equation [HM89
W (C ) =
Ry
P e x dr A (r) ;
where the integration is over regulator paths r (t) from x to y whose typi
al
length is a. The
onventional measure is implied in (3.3.83) so that
Z
Cyx
Aa (x)
277
dx
dy
Dr(t) e
R 2
2
1 a
2 0 dt r_ (t) W (C r )W (C r )
yx xy
xy yx
r(0)=x
r(a2 )=y
(3.3.87)
whi
h manifestly re
overs Eq. (3.3.79) when a ! 0.
The
onstru
ted regularization is nonperturbative while perturbatively
reprodu
es regularized Feynman diagrams. An advantage of this regularization of the loop equation is that the
ontours Cyx rxy and Cxy ryx on the RHS
of Eq. (3.3.87) both are
losed and do not have marked points if C does not
have. Therefore, Eq. (3.3.87) is written entirely in loop spa
e.
2 U [x =
1
2
Z f
0
00 [x(x ())2 + : : : :
d Uxx
(3.3.88)
278
CHAPTER 3.
1=N
EXPANSION
Z f
0
00 [x :
d Uxx
(3.3.89)
Here U [x
an be an arbitrary, not ne
essarily parametri
invariant, fun
tional. To emphasize this obsta
le, we use the notation U [x for generi
fun
tionals whi
h are dened on L2 spa
e in
omparison to U (C ) for the
fun
tionals whi
h are dened on elements of loop spa
e. It is easier to deal
with the whole operator , rather than separately with the area and path
derivatives.
The fun
tional Lapla
ian is parametri
invariant and possesses a number
of remarkable properties. While a nitedimensional Lapla
ian is an operator
of the se
ond order, the fun
tional Lapla
ian is that of the rst order and
satises the Leibnitz rule
(UV ) = (U ) V + U (V ) :
(3.3.90)
G (C; C 0 ) =
SC;C 0
::: ;
(3.3.91)
whi
h is analogous to the representation (1.1.98) of the Green fun
tion of the
ordinary Lapla
ian. The standard perturbation theory
an then be re
overed
by iterating Eq. (3.3.79) (or its regularized version (3.3.87)) in with the
Green fun
tion (3.3.91).
3.3.
279
the
uto, was not investigated. In order to do this, one needs more detailed
information about the behavior of W (C ) for intermediate loops.
The free bosoni
Nambu{Goto string whi
h is dened as a sum over surfa
es spanned by C
W (C ) =
S :S =C
e K A(S) ;
(3.3.93)
with the a
tion being the area A (S ) of the surfa
e S , is not a solution for intermediate loops. Consequently, QCD does not redu
e to this kind of string,
as was originally expe
ted in Refs. [GN79a, Nam79, Pol79. Roughly speaking, the ansatz (3.3.93) is not
onsistent with the fa
torized stru
ture on the
RHS of Eq. (3.3.56).
Nevertheless, it was shown that if a free string satises Eq. (3.3.56), then
the same intera
ting string satises the loop equations for nite N
. Here
\free string" means, as usual in string theory, that only surfa
es of genus
zero are present in the sum over surfa
es, while surfa
es or higher genera are
asso
iatedwith a string intera
tion. The
oupling
onstant of this intera
tion
is O N
2 .
A formal solution of Eq. (3.3.56) for all loops was found by Migdal [Mig81
in the form of a fermioni
string
W (C ) =
S :S =C
D e
R 2
d [ k k + m4 pg
(3.3.94)
where the world sheet of the string is parametrized by the
oordinates 1 and
2 for whi
h the 2dimensional metri
is
onformal, i.e. diagonal. The eld
( ) des
ribes 2dimensional elementary fermions (elves) living in the surfa
e
S , and m stands for their mass. Elves were introdu
ed to provide fa
torization
whi
h now holds due to some remarkable properties of 2dimensional fermions.
For large loops, the internal fermioni
stru
ture be
omes frozen, so that the
empty string behavior (3.3.92) is re
overed. For small loops, the elves are
ne
essary for asymptoti
freedom. However, it is un
lear whether or not the
string solution (3.3.94) is pra
ti
ally useful for a study of multi
olor QCD,
sin
e the methods of dealing with the string theory in four dimensions are
not yet developed.
A very interesting solution of the largeN
loop equation on a latti
e was
found by Egu
hi and Kawai [EK82. They showed that the SU(N
) gauge
theory on an innite latti
e redu
es at N
= 1 to the model on a hyper
ube,
280
CHAPTER 3.
1=N
EXPANSION
Z Y
d
=1
dU e
P
y y
1
> Re N
tr(U U U U ) :
(3.3.95)
A1 = 0 ;
(3.3.96)
C
(1; 2; : : : ; 2k) D1 1 (x1
281
where the points x1 , : : :, x2k are still
y
li
ordered along the
ontour. Similar
to Problem 2.2, we
an exponentiate the RHS of Eq. (3.3.98) to get nally
ZEK ( ) =
3.3.
C
C
C
x2 ) Dk k (x2k 1 x2k ) ;(3.3.98)
W (C ) = e
2 C dx
H dy D (x y)
C
:
(3.3.99)
D (x y) =
D (x y) = D (x y) :
(3.3.101)
D (x y ) =
`2
1
ln
;
4 (x y)2
(3.3.102)
dx
dy D (x y) = A (C )
(3.3.103)
W (C ) = e
A(C )
2
D (x y ) =
1
d
2
4d=2
1
d=2 1 :
(x y)2
1
(3.3.104)
282
CHAPTER 3.
x2 =
1=N
EXPANSION
.....................
........
......
.....
....
.
.
..
...
....
..
...
..
y2 ......
.....
.
....
...
......
................................
Fig. 3.33: Graphi
representation of the
ontour integral on the LHS of Eq. (3.3.103)
in the axial gauge. The bold line represents the gluon propagator (3.3.100)
with x2 = y2 due to the deltafun
tion.
Therefore, area law holds in two dimensions both in the nonAbelian and
Abelian
ases. This is, roughly speaking, be
ause of the form of the twodimensional propagator (3.3.102) whi
h falls down with the distan
e only
logarithmi
ally in the Feynman gauge.
Prove Eq. (3.3.104) in the Feynman gauge.
To prove Eq. (3.3.103) in the Feynman gauge (3.3.101), (3.3.102), we
note that the area element in two dimensions
an be represented by
Problem 3.24
Solution
d (x)
(3.3.105)
where e is the antisymmetri
tensor e12 = e21 = 1. Therefore, the area
an be
represented by the double integral
A (C ) = 1
2
d (x)
S (C )
d (y ) (2) (x y )
(3.3.106)
S (C )
whi
h goes along the surfa
e S (C ) en
losed by the (noninterse
ting) loop C .
Applying the Stokes' theorem, we get
I
dx
dy D (x y ) =
d (x)
Z
d (x)
S (C )
=
1
=
2
1
2
dy D (x y )
........................
.........
.....
....
.
...
.
.
...
.....
...
.
...
A
1
...
...
.
.
..
...
.
....
.
......
.....
......... .............
.
.
.
.. . ...
..... .....
.... A2 ....
.... ....
.............
d (x)
S (C )
S (C )
d (y ) D (x y )
Z
d (y ) 2 D (x y )
S (C )
d (x)
S (C )
d (y ) (2) (x
b)
Fig. 3.34: Contours with one selfinterse
tion: A1 and A2 stand for the areas of
the proper windows. The total area en
losed by the
ontour in Fig. a) is
A1 + A2 . The areas en
losed by the exterior and interior loops in Fig. b)
are A1 + A2 and A2 , respe
tively, while the total area of the surfa
e with
the folding is A1 + 2A2 .
The dieren
e between the Abelian and nonAbelian
ases shows up for
the
ontours with selfinterse
tions.
We rst note that the simple formula (3.3.103) does not hold for
ontours
with arbitrary selfinterse
tions.
The simplest
ontours with one selfinterse
tion are depi
ted in Fig. 3.34.
There is nothing spe
ial about the
ontour in Fig. 3.34a. Equation (3.3.103)
still holds in this
ase with A (C ) being the total area A (C ) = A1 + A2 .
The Wilson loop average for the
ontour in Fig. 3.34a
oin
ides both for
the Abelian and nonAbelian
ases and equals
(A1 +A2 )
2
(3.3.108)
y ) :(3.3.107)
.............
........... ..................
.....
......
.
.
.
.
....
...
...
.....
...
A
1
..
....
...
...
...
...
.
...
.
.
.
.
.
.
.
.
.
.... ......
..
...
...
.
..
....
..... ..... A2 .... ........
....... .... ..... ........
...........................
a)
W (C ) = e
S (C )
283
S (C )
3.3.
dx
(3.3.109)
This is easy to understand in the axial gauge where the ends of the propagator
line
an lie both on the exterior and interior loops, or one end at the exterior
284
CHAPTER 3.
.............................
.........
.....
.....
....
.
.
.
...
...
....
...
...
...
..
.
...
...
...
.
.
.
.
.
.
.
.
.
...
... ......
.
.
.
.
.
... ...
....
..... .....
. .
....... ..... ....................
.
.............. ......
...........................
.........
.....
.....
....
.
.
.
...
...
.....
..
....
..
...
.
...
...
.
...
.
.
.
.
.
.
.
.
. .....
.
.
.
.
.
.
...
..
..
..... ....
..... .... .... ..........
.................................
...
...........................
..........
.....
.....
....
.
.
.
...
...
.....
...
....
...
...
...
...
.
...
.
.
.
.
.
.
.
..
.... .......
...
..
.... ....
.
...... ... .... .........
...................................
...
b)
a)
1=N
EXPANSION
......
.............. ...................
......
.....
.
.
.
...
..
...
...
...
....
..
...
.
...
...
...
.
.
.
.
.
.
.
.
.
.
...
.. ... ......
.
.
....
.. ....
.
. .
..... ....
....... ....... ....................
........... ......
3.3.
line lie both on a) exterior and b) interior loops, or
), d) one end on the
exterior loop and another end on the interior loop.
Problem 3.25
x
d)
Fig. 3.35: Three type of ontribution in Eq. (3.3.109) The ends of the propagator
285
(x)
W (C ) =
dy (d) (x y ) W (C ) :
(3.3.113)
Solution
Obtain Eqs. (3.3.108) and (3.3.111) for the
ontours with one selfinterse
tion by solving the loop equation (3.3.56).
Solution Let us multiply Eq. (3.3.56) in d = 2 by e and integrate over dx along
a small (open) pie
e C 0 of the
ontour C in
luding the point of selfinterse
tion. We
get
Problem 3.26
loop and the other end on the interior loop. These
ases are illustrated by
Fig. 3.35. The
ontributions of the diagrams in Fig. 3.35a,b,
,d are A1 + A2 ,
A2 , A2 , and A2 , respe
tively. The result given by Eq. (3.3.109) is obtained
by summing over all four diagrams.
For the
ontour in Fig. 3.34b, the Wilson loop average is
W (C ) = e
(A1 +4A2 )
(3.3.110)
W (C ) = (1 A2 ) e
(A1 +2A2 )
2
(3.3.111)
W (C ) = P (A1 ; : : : ; An ) e
Area
2
(3.3.112)
e
C0
dx x
(x)
W (C ) = e
C0
dx
(3.3.114)
The RHS of Eq. (3.3.114)
an be
al
ulated analogously to the known representation for the number of selfinterse
tions of a loop in twodimensions. For the
ase
of one selfinterse
tion, we have
e
C0
dx
where C1 and C2 stands, respe
tively, for the upper and lower loops in Fig. 3.34a
or the exterior and interior loops in Fig. 3.34b.
The LHS of Eq. (3.3.114)
an be transformed as
e
C0
dx x
(x)
W (C ) =
C0
dx x W (C ) ;
(x)
(3.3.116)
where = (x) denotes the variational derivative with respe
t to the \s
alar" area
(3.3.117)
(x) = 1 e (x) :
2
The integrand on the RHS of Eq. (3.3.116) is a total derivative and the
ontour
integral redu
es to the dieren
e of the
variations at the end points of the
ontour
286
CHAPTER 3.
1=N
EXPANSION
C 0 , whi
h would vanish if no selfinterse
tions. The RHS of Eq. (3.3.114) also
vanishes if no selfinterse
tions so W (C ) is determined in this
ase by Eq. (3.3.113)
rather than Eq. (3.3.114).
For the
ontour in Fig. 3.34a, this gives
.........
........... ..............
...
.....
...
.....
...
...
A
1
...
...
...
..
.
.
...
.
.....
...
............. ............
.............
.
.
.
.. ..
.... A2 ....
....... ......
......
.............
......... .............
.....
...
....
.
A1 ......
.....
.
...
...
...
..
.
.....
.
.
.
........ ........
...................
..... ...
.... A2 .....
.................
W (C ) :
(3.3.118)
A1 A2
The
variation of the
ontour on the LHS represents the variational derivative.
The minus sign in front of =A1 on the RHS is be
ause adding the
variation
in the rst term on the LHS de
reases the area A1 while that in the se
ond term
in
reases A2 . Then Eq. (3.3.114) takes for the
ontour in Fig. 3.34a the form
=
W (C ) = W (C ) W (C ) :
1
2
A1 A2
For the
ontour in Fig. 3.34b, we get quite similarly
..........................
.....
..........
.....
.....
...
......
...
A
...
1
...
..
...
...
...
..
.
...
.
........
..
...
..... ....
..... .... A2 .... ......
..... ... .. ... .......
.......... .......................
......
(3.3.119)
..........................
.....
..........
.....
.....
...
......
...
...
A1
...
..
...
...
...
..
.
...
.
........
..
...
..... ....
.... .... A2 .... ......
..... .... ... .......
...... ........................
........
A1
A1
W (C ) = e
A2
(A1 +A2 )
2
W (C ) = e 2 (A1 +2A2 ) ;
(3.3.123)
A1 A2
respe
tively. Their solution is uniquely given by Eqs. (3.3.108) and (3.3.111).
It is worth nothing that the linear Abelian loop equation (3.3.113)
an be written
for the
ontours in Fig. 3.34a,b as
ln W (C ) = ;
(3.3.124)
A1 A2
ln W (C ) = :
2
(3.3.125)
A1 A2
The operators on the LHS's are always the same for the nonAbelian and Abelian
loop equations, whi
h is a general property, but the RHS's generi
ally differ: Eqs. (3.3.122) and (3.3.124) for the
ontour in Fig. 3.34a
oin
ide while
Eqs. (3.3.123) and (3.3.125) for the
ontour in Fig. 3.34b dier. The solution to
Eq. (3.3.125) is given by (3.3.110).
Prove Eq. (3.3.115) for the
ontours with one selfinterse
tion.
Let the interse
tion
orresponds to the values s1 and s2 of the parameter
s, i.e. x (s1 ) = x (s2 ). Noting that only the vi
inities of s1 and s2
ontribute to
the integral on the LHS of Eq. (3.3.115), we get
Problem 3.27
Solution
C0
dx
(3.3.121)
The RHS's of Eqs. (3.3.119) and (3.3.121) are known sin
e C1 and C2 have
no selfinterse
tions so that Eq. (3.3.104) holds for W (C1 ) and W (C2 ). Equations (3.3.119) and (3.3.121) take nally the expli
it form [KK80
and
W (C ) = W (C ) W (C ) :
1
2
A2
(3.3.122)
287
e
W (C ) :
(3.3.120)
2
A1 A2
Now adding the
variation in the rst term on the LHS in
reases A1 and de
reases
A2 while that in the se
ond term de
reases A1 . Equation (3.3.114) takes the form
=
3.3.
ds
e x_ (s1 ) x_ (s2 )
W
C
W
C
x
(
s
)
x
(
s
)
x
(
s
)
x
(
s
)
2
1
1
2
x_ 2 (s1 ) x_ 2 (s2 ) (x_ (s1 ) x_ (s2 ))2
= W Cx(s2 )x(s1 ) W Cx(s1 )x(s2 )
(3.3.126)
interse tions.
288
CHAPTER 3.
1=N
EXPANSION
Z2d ( ) =
Z Y Y
x =1;2
dUx; e S[U ;
(3.3.127)
Ux;1 = 1 :
(3.3.129)
where Np stands for the number of plaquettes of the latti
e and Z1p is the
onematrix integral
Z1p ( ) =
1
dU e ( N
Re tr U 1) :
(3.3.130)
289
Let us rst introdu
e the proper `observables' for the onematrix model:
1
n
Wn =
;
(3.3.131)
tr U
N
1p
where the average is taken with the same weight as in Eq. (3.3.130). The
interpretation of Wn 's in the language of the singleplaquette model is that
these are the Wilson loop averages for
ontours whi
h go along the boundary
of n sta
ked plaquettes.
In order to derive the loop equation for the onematrix model, we pro
eed
quite analogous to the derivation of the loop equation in the latti
e gauge
theory (given in Problem 3.22). Let us
onsider the obvious identity
0 =
h tr ta U ni1p ;
(3.3.132)
! U (1 ita a ) ; U y ! (1 + ita a ) U y
(3.3.133)
of the integration variable on the RHS of Eq. (3.3.132). Sin
e the Haar
measure is invariant under the
hange (3.3.133), we nally get
(W
2N
2 n
(3.3.128)
Z2d = (Z1p)Np ;
3.3.
where
Wn+1 ) =
n
X
k=1
Wk Wn k
for n 1 ;
W0 = 1 ;
(3.3.134)
1
=
N
2
(3.3.135)
1;
(3.3.137)
and 1 as N
! 1.
Equation (3.3.134) has the following exa
t solution
1
; Wn = 0 for n 2
(3.3.136)
W1 =
2
whi
h reprodu
es the strong
oupling expansion. The leading order of the
strong
oupling expansion turns out to be exa
t at N
= 1.
However, the solution (3.3.136)
an not be the desired solution at any
values of the
oupling
onstant. Sin
e Wk are (normalized) averages of unitary
matri
es, they must obey
Wn
290
CHAPTER 3.
1=N
EXPANSION
3.3.
291
whi
h is not the
ase for W1 , given by Eq. (3.3.136), at small enough values
of .
In order to nd all solutions to Eq. (3.3.134), let us introdu
e the generating fun
tion
1
X
f (z )
Wn z n
(3.3.138)
where A is the area (in the latti
e units) en
losed by the
ontour C . W1 in
this formula is given by Eq. (3.3.136) in the strong
oupling phase ( 1)
and Eq. (3.3.141) in the weak
oupling phase ( 1).
The
ontinuum formula (3.3.104) is re
overed for small from
Eq. (3.3.142) as follows:
WC = ( W 1 ) A
n=0
1
(f 1) + W1 = 2 f 2
z
A formal solution to Eq. (3.3.139) is
fz
f :
(3.3.139)
(1 + 2z + z 2 )2 + 4z 2 (2W1 1)
1 2z
f (z ) =
+
; (3.3.140)
4z
4z
where the positive sign of the square root is
hosen to satisfy f (0) = 1.
The RHS of Eq. (3.3.140) depends on an unknown fun
tion W1 () whi
h
must guarantee for f (z ) to be a holomor
fun
tion of the
omplex variable z
inside the unit
ir
le jz j < 1. This is a
onsequen
e of the inequality (3.3.137)
whi
h stems from the unitarity of U 's.
There exist two solutions for whi
h f (z ) is holomor
inside the unit
ir
le:
the strong
oupling solution given for 1 by Eq. (3.3.136) and the weak
oupling solution given for 1 by
z2
:
(3.3.141)
2
A
omparison with Eq. (2.3.1) for d = 2 shows that the leading order of
the weak
oupling expansion is now exa
t. Therefore, f (z ) is given by two
dierent analiti
fun
tions for > 1 and < 1.
At the point = 1, a phase transition o
urs as was dis
overed by Gross
and Witten [GW80 who rst solved latti
e QCD2 in the largeN
limit. This
phase transition is of the third order sin
e both the rst and se
ond derivatives
of the partition fun
tion are
ontinuous at = 1. The dis
ontinuity resides
only in the third derivative. This phase transition is pretty unusual from the
point of view of statisti
al me
hani
s where phase transitions usually o
ur
in the limit of an innite volume (otherwise the partition fun
tion is analyti
in temperature). Now the Gross{Witten phase transition o
urs even for the
W1 = 1
(3.3.142)
Problem 3.28
U = diag e i1 ; : : : ; e ij ; : : : ; e iN
:
(3.3.144)
The density of eigenvalues (or the spe
tral density), (), is then dened as a
fra
tion of the eigenvalues whi
h lie in the interval [; + d. In other words,
introdu
ing the
ontinuum variable x = j=N
(0 x 1) in the largeN
limit, we
have
() = dx 0
(3.3.145)
d
whi
h obey the obvious normalization
Z
d () =
Z 1
dx = 1 :
(3.3.146)
d () os n :
(3.3.147)
Given (), we an al ulate Wn by
Wn =
Z
292
CHAPTER 3.
1=N
EXPANSION
f (z ) =
d ()
i
2
1
:
1 z e i
(3.3.148)
Z
r
os + +
1 2W1
os +
Z1h =
Z Y
ij
2
M
dij e N
2 tr :
(3.3.155)
The dieren
e from the previous problem is that ij is now a Hermitean
N
N
matrix whose eigenvalues pi
an take on values along the whole real axis.
Let us dene
Z
D
E
1
n
Wn
tr
=
d ( ) n :
(3.3.156)
N
1h
The S
hwinger{Dyson equations for the Hermitean onematrix model
an be derived [Wad81 using the invarian
e of the measure in (3.3.155) under the
hange
ij ! ij + ij
(3.3.157)
where ij is (innitesimal) Hermitean.
Pro
eeding as before and using the largeN
fa
torization, we get the set of
equations
Solution
d ()
ot ! :
2
(3.3.149)
() = 1
2
293
Problem 3.29
3.3.
1 2W1 : (3.3.150)
For W1 given by Eqs. (3.3.136) and (3.3.141) for the strong and weak
oupling
phases, we get nally
1
1
1 +
os
for 1
(3.3.151)
() =
2
and
1
q sin2
() =
os
for 1
(3.3.152)
2
2
for the strong and weak
oupling solutions, respe
tively. Noti
e, that (3.3.151) is
nonnegative for 1 as it should due to the inequality (3.3.145). For < 1,
the strong
oupling solution (3.3.151) be
omes negative somewhere in the interval
[ ; whi
h
an not happen for a dynami
al system. This is the reason why the
other solution (3.3.152) is realized for < 1. It has the support on the smaller
interval [
;
where 0 <
< is determined by the equation
sin2
=
(3.3.153)
2
whi
h always has a solution for < 1. The weak
oupling spe
tral density (3.3.152)
is nonnegative for p1.
For small ,
= 2 so that
p
() = 1 4 2 :
(3.3.154)
2
As ! 0, () ! () and U freezes, modulo a gauge transformation, near a unit
matrix. This guarantees the existen
e of the
ontinuum limit of QCD2 .
The spe
tral densities (3.3.151) and (3.3.152) were rst
al
ulated in
Ref. [GW80 by a dire
t solution of the saddlepoint equation at large N
.
M Wn+1 =
n 1
X
k=0
for n 0 ;
Wk Wn k
W0 = 1 :
(3.3.158)
W (p)
1
1
tr
N
p
1h
1
X
Wn ;
n+1
p
n=0
(3.3.159)
Mp W (p) M = W 2 (p)
whose solution reads
(3.3.160)
Mp
M 2 p2 M ;
W (p) =
(3.3.161)
2
4
where we have
hosen the minus sign of the root to satisfy the asymptoti
s W (p) !
1=p for large p as is pres
ribed by the denition (3.3.159).
W (pp
) given by Eq. p
(3.3.161) is an analyti
fun
tion of p with the
ut from
p = 2= M to p = +2= M along the real axis. The dis
ontinuity of W (p) at the
ut determines the spe
tral density:
W ( i0) = M i ( ) :
(3.3.162)
2
294
CHAPTER 3.
We have
( ) = M
2
1=N
EXPANSION
3.4.
NC
LARGE
295
REDUCTION
2 :
(3.3.163)
Noti
e
p that the
p spe
tral density is nonnegative and has support on a nite interval
[ 2= M; 2= M in analogy with the unitary onematrix model it the weak
oupling
regime. The spe
tral density (3.3.163) was rst
al
ulated by Wigner [Wig51 and
is
alled Wigner's semi
ir
le law.
We have already semi
ir
le law in the previous Problem for the spe
tral density
of the unitary onematrix model at small (see Eq. (3.3.154)). This is be
ause we
an always substitute U = exp (i) where U is unitary and is Hermitean and
expand for small in up to the quadrati
term. We then obtain the Hermitean
model (3.3.155) with M = 1= from the unitary model (3.3.130).
The largeN
redu
tion was rst dis
overed by Egu
hi and Kawai [EK82
who showed that the Wilson latti
e gauge theory on a ddimensional hyper
ubi
latti
e is equivalent at N
= 1 to the one on a hyper
ube with periodi
boundary
onditions. This
onstru
tion is based on an extra (ZN
)d symmetry whi
h the redu
ed theory possesses to ea
h order of the strong
oupling expansion.
Soon after it was re
ognized that a phase transition o
urs in the redu
ed model with de
reasing the
oupling
onstant, so that this symmetry
is broken in the weak
oupling regime. To
ure the
onstru
tion at weak
oupling, the quen
hing pres
ription was proposed by Bhanot, Heller and
Neuberger [BHN82 and elaborated by many authors. An elegant alternative redu
tion pro
edure based on twisting pres
ription was advo
ated by
GonzalezArroyo and Okawa [GAO83. Ea
h of these pres
riptions results in
the redu
ed model whi
h is fully equivalent to multi
olor QCD, both on the
latti
e and in the
ontinuum.
While the redu
ed models look as a great simpli
ation, sin
e the spa
etime is redu
ed to a point, they still involve an integration over d innite
matri
es whi
h is in fa
t a
ontinual path integral. It is not
lear at the
moment whether or not this is a real simpli
ation of the original theory
whi
h
an make it solvable. Nevertheless, the redu
ed models are useful and
elegant representations of the original theory at large N
.
We shall start this Se
tion by a simplest example of a pure matrix s
alar
theory. The quen
hed redu
ed model for this
ase was proposed by Parisi
[Par82 on the latti
e end elaborated by Gross and Kitazawa [GK82 in the
ontinuum, while the twisted redu
ed model was advo
ated by Egu
hi and
Nakayama [EN83. Then we
on
entrate on the Egu
hi{Kawai redu
tion of
multi
olor QCD both on the latti
e and in the
ontinuum.
Z=
Z YY
x ij
d'ijx e
P
x N
tr
V ['x +
P
'x 'x+a^
(3.4.1)
296
CHAPTER 3.
1=N
EXPANSION
Here 'x is a N
N
Hermitean matrix eld and V [' is some intera
tion
potential, say
M 2 3 3 4 4
V [' =
' + ' + ' :
(3.4.2)
2
3
4
The pres
ription of the largeN
redu
tion is formulated as follows. We
substitute
'x ! Sx S y ;
(3.4.3)
x
where
[Sxkj = e ipk x kj
= diag e ip1 x ; : : : ; e ipN
x
F ['x
Z1
Z Y
d'x e
P
x N
tr
P
'(x)'(x+a^) F ['x ;
V ['x +
(3.4.5)
F ['x
Z Y
N
a d Y
N
d
a
a =1 i=1
F ['x
2
x Sx
Redu ed
ij kl
(3.4.6)
where the average on the RHS is
al
ulated [Par82 for the quen
hed redu
ed
model whose averages are dened by
D
E
1
F [
Redu
ed ZRedu
ed
Z Y
P
2P
dij e N
tr V [+N
ij jij j
os ((pi pj )a) F [ : (3.4.7)
ij
The partition fun
tion of the redu
ed model reads
Z Y
P
2P
dij e N
tr V [+N
ij jij j
os ((pi pj )a) (3.4.8)
ZRedu
ed =
ij
! F [Sx s Sxy ;
(3.4.9)
where the momenta pi are uniformly distributed in the hyper
ube. Therefore,
this saddle point
onguration plays the role of a master eld in the sense of
Subse
t. 3.2.7.
In order to show how Eq. (3.4.6) works, let us demonstrate how the planar diagrams of perturbation theory for the s
alar matrix theory (3.4.1) are
re
overed in the quen
hed redu
ed model.
The quen
hed redu
ed model (3.4.8) is of the general type dis
ussed in
Se
t. 3.2. The propagator is given by
D
dpi D
F [S
297
REDUCTION
whi
h
an be dedu
ed, modulo the volume fa
tor, from the partition fun
tion
(3.4.1) by the substitution (3.4.3).
Noti
e that the integration over the momenta pi on the RHS of Eq. (3.4.6)
is taken after the
al
ulation of averages in the redu
ed model. Su
h variables are usually
alled quen
hed in statisti
al me
hani
s whi
h
laries the
terminology.
Sin
e N
! 1 it is not ne
essary to integrate over the quen
hed momenta
in Eq. (3.4.6). The integral should be re
overed if pi 's would be uniformly
distributed in a ddimensional hyper
ube. Moreover, a similar property holds
for the matrix integral over as well, whi
h
an be substituted by its value
at the saddle point
onguration s :
an be
al
ulated at N
= 1 by
D
NC
LARGE
(3.4.4)
3.4.
Gauss
1
G (pi pj ) il kj
N
(3.4.10)
with
G (pi pj ) =
1
os
(pi
pj )a
(3.4.11)
298
CHAPTER 3.
1=N
EXPANSION
...........
..........
..........
.................. .............
...... ............................. ...........
.
.
.
.
.... ..
... ......
k ......... .......
... ...
.. ..
.... ....
..
.
... .....
... ...
.
.
... ...
. .
... ..
... .....
.... ....... k .......... ......
...... ............................ ......
........
...................................
...........
...........
Fig. 3.36: Simplest planar diagram of the se ond order in 3 for the propagator in
the quen
hed redu
ed model (3.4.8). The momentum pi
ows along the
index line i. The momentum pi pj is asso
iated with the double line ij .
23
2X
G
(
p
p
)
G ( p i pk ) G ( p k
i
j
N
2
k
pj ) ;
(3.4.12)
where the summation over the index k is just a standard one over indi
es
forming a
losed loop.
In order to show that the quen
hedmodel result (3.4.12) reprodu
es the
orre
tion to the propagator in the original theory on an innite latti
e, we
pass to the variables of the total momenta
owing along the double lines:
pi pj = p; pk
pj = q ; p i pk = p q ;
(3.4.13)
Z
a
a
dd q
f (q) :
(2)d
(3.4.14)
a
a
dd q
G (q) G (p q)
(2)d
(3.4.15)
3.4.
NC
LARGE
REDUCTION
299
for the se
ondorder
ontribution of the perturbation theory for the propagator on the latti
e.
It is now
lear how a generi
planar diagram is re
overed by the redu
ed
model. We rst represent the diagram by the double lines and asso
iate
the momentum pi with an index line
arrying the index i. Then we write
down the expression for the diagram in the redu
ed model with the propagator (3.4.11). Passing to the momenta
owing along the double lines, similar
to Eq. (3.4.13), we get an expression whi
h
oin
ides with the integrand of the
Feynman diagram for the theory on the whole latti
e. It is
ru
ial that su
h
a
hange of variables
an always be done for a planar diagram
onsistently
with the momentum
onservation at ea
h vertex. The last step is that the
summation over indi
es of
losed index lines reprodu
es the integration over
momenta asso
iated with ea
h of the loops a
ording to Eq. (3.4.14). It is
assumed that the number of loops is mu
h less than N
whi
h is always true
for a given diagram sin
e N
is innite.
We thus have shown how planar diagrams of the latti
e theory dened by
the partition fun
tion (3.4.1) are re
overed by the redu
ed model (3.4.8). The
latti
e was needed only as a regularization to make all integrals welldened
and was not
ru
ial in the
onsideration. This
onstru
tion
an be formulated
dire
tly for the
ontinuum theory [GK82, DW82 where the propagator turns
into
1
G ( p i pj ) =
(3.4.16)
( pi p j ) 2 + m 2
and a Lorentzinvariant regularization
an be a
hieved by
hoosing p2 < 2 .
Sx = x1 1=a x2 2 =a x3 3 =a x4 4 =a
(3.4.17)
where the
oordinates of the (latti
e) ve
tor x are measured in the latti
e
units. The matri
es are expli
itly
onstru
ted in Ref. [GAO83 and
ommute by
(3.4.18)
= Z
y being elements of ZN
.
with Z = Z
300
CHAPTER 3.
1=N
EXPANSION
For the twisting redu
tion pres
ription, Eq. (3.4.6) is valid providing the
average on the RHS is
al
ulated for the twisted redu
ed model whi
h is dened
by the partition fun
tion [EN83
ZTRM =
d e
N tr V [+N
y
tr :
(3.4.19)
Now the perturbation theory for the unredu
ed model (3.4.1) is re
overed due
to the expli
it form of Sx given by Eq. (3.4.17).
We
an
hange the order of 's in Eq. (3.4.17) dening a more general
pathdependent fa
tor
Y
Sx = P
(3.4.20)
:
l2Cx1
The pathordered produ
t in this formula runs over all links l = (z; ) forming
a path Cx1 from innity to the point x.
Due to Eq. (3.4.18),
hanging the form of the path multiplies Sx by the
Abelian fa
tor
Y
Z (C ) =
Z (2)
(3.4.21)
22S :S =C
where (; ) is the orientation of the plaquette 2. The produ
t runs over any
surfa
e spanned by the
losed loop C whi
h is obtained by passing the original
path forward and the new path ba
kward. Due to the Bian
hi identity
Y
22 ube
Z (2) = 1
(3.4.22)
where the produ
t goes over six plaquettes forming a 3dimensional
ube on
the latti
e, the produ
t on the RHS of Eq. (3.4.21) does not depend on the
form of the surfa
e S and is a fun
tional of the loop C .
It is now easy to see that under this
hange of the path we get
[Sx ij [S y kl ! jZ (C )j2 [Sx ij [S y kl
(3.4.23)
x
x
jZ (C )j2
3.4.
NC
LARGE
redu
ed model whi
h is nothing but its redu
tion to a point. The a
tion of
the redu
ed model is given by
1
SEK =
2g2d
tr [A ; A 2 ;
(3.4.24)
H
1
tr P e i d A ()
N
d dim
H
1
tr P e i d A
N
(3.4.25)
EK
where the LHS is
al
ulated with the a
tion (2.1.14) and the RHS is
al
ulated
with the redu
ed a
tion (3.4.24). Stri
tly speaking, this naive statement is
valid only in d = 2 or supersymmetri
ase for the reason whi
h will be
explained in a moment.
The pre
ise equivalen
e is valid only if the average of open Wilson loops
vanish in the redu
ed model:
R
i C d A
1
yx
= 0;
(3.4.26)
tr P e
N
EK
as it does in the ddimensional theory due to the lo
al gauge invarian
e under
whi
h
d A ()
i
P e Cyx
ij
y (y)P e
Cyx d A ()
(x)
ij
(3.4.27)
The point is that this gauge invarian
e transforms in the redu
ed model into
(global) rotation of the redu
ed eld by
onstant matri
es
:
A !
y A
:
(3.4.28)
whi
h does not guarantee su
h vanishing in the redu
ed model.
There exists, however, a symmetry of the redu
ed a
tion (3.4.24) under
the shift of A by a unit matrix15:
301
REDUCTION
Aij
! Aij + a ij ;
(3.4.29)
15 This symmetry is rigorously dened on a latti e where it is asso iated with a dire tion
dependent ZN transformation.
302
1=N
CHAPTER 3.
EXPANSION
Pe
i C d A
yx
ij
! e i(y
x )a
Pe
i C d A
yx
(3.4.30)
ij
whi
h guarantees, if the symmetry is not broken, the vanishing of the open
Wilson loops
WEK (Cyx )
i
1
tr P e
N
R
Cyx d A
EK
=0
(3.4.31)
x
1
d A
i
tr P[A ; [A ; A e Cxx
WEK (C ) =
N
(x)
H
i Cxx d A
1
d
tr P
e
=
N
A
EK
= d
EK
(3.4.32)
The RHS is pretty mu
h similar to the one in Eq. (3.3.56) while (d)(x y)
is missing.
This delta fun
tion
an be re
overed if the Rd symmetry is not broken
sin
e
(d)(x y)
WEK (Cyx )
(3.4.33)
WEK (Cyx ) (d)
(0)
due to Eq. (3.4.31) for the open loops.
This is not a rigorous argument sin
e a regularization is needed. What
a
tually happens is the following. If we smear the delta fun
tion introdu
ing
(d) (x) =
then
p
2
d
2 2
e x =2 ;
2 2
1 (d) 2
(0) / d e x
(d)
(0)
reprodu
ing the delta fun
tion.
! (d) (x) ;
LARGE
303
REDUCTION
(3.4.36)
(N
)
A
l p = diag p(1)
; : : : ; p
(3.4.37)
A = A l + gAq ;
NC
3.4.
(3.4.34)
(3.4.35)
(3.4.38)
S2 = tr
1
[p ; Aq 2 + [p ; b[p ;
2
(3.4.39)
(3.4.40)
(3.4.41)
dp dAq e S2 : : : =
Z Y
N
k=1
dp(k)
Yh
i<j
(p(i)
p(j) )2
i1 d=2
::: ;
(3.4.42)
304
CHAPTER 3.
1=N
EXPANSION
3.4.
NC
LARGE
REDUCTION
305
this fa
tor does not vanish and is needed to provide the equivalen
e with
ddimensional Yang{Mills perturbation theory, sin
e the
lassi
al extrema
of the twisted redu
ed model are A
l =
and the perturbation theory is
onstru
ted expanding around this
lassi
al solution.
The proof of the equivalen
e
an be done using the loop equation quite
similarly to that of Subse
t. 3.4.2 for the Egu
hi{Kawai model with an unbroken Rd symmetry.
306
CHAPTER 3.
1=N
EXPANSION
Referen
es to Chapter 3
[Ans59
[BGSN82
[BHN82
[BNS81
[BNZ79
[BP75
[Bra80
[BW93
[CJP74
[CMS93
[CLS82
A.A. Anselm, A model of a eld theory with nonvanishing renormalized
harge, Zh. Exp. Teor. Fiz. 36 (1959) 863 (English transl.:
Sov. Phys. JETP 9 (1959) 608).
R.A. Brandt, A. Go
ks
h, M. Sato, and F. Neri, Loop spa
e, Phys.
Rev. D26 (1982) 3611.
G. Bhanot, U. Heller and H. Neuberger, The quen
hed Egu
hi{
Kawai model, Phys. Lett. 113B (1982) 47.
R.A. Brandt, F. Neri and M. Sato, Renormalization of loop fun
tions for all loops, Phys. Rev. D24 (1981) 879.
R.A. Brandt, F. Neri and D. Zwanziger, Lorentz invarian
e from
lassi
al parti
le paths in quantum eld theory of ele
tri
and magneti
harge, Phys. Rev. D19 (1979) 1153.
A.A. Belavin and A.M. Polyakov, Metastable states of twodimensional isotropi
ferromagnet, Pis'ma v ZhETF 22 (1975)
503 (English transl.: JETP Lett. 22 (1975) 245).
N. Brali
, Exa
t
omputation of loop averages in twodimensional
Yang{Mills theory, Phys. Rev. D22 (1980) 3090.
E. Brezin and S. Wadia, The large N expansion in quantum eld
theory and statisti
al physi
s: from spin systems to 2dimensional
gravity, World S
i., 1993.
S. Coleman, R. Ja
kiw and H.D. Politzer, Spontaneous symmetry
breaking in the O(N ) model at large N , Phys. Rev. D10 (1974)
2491.
W. Chen, Y. Makeenko, and G.W. Semeno, Fourfermion theory
and the
onformal bootstrap, Ann. Phys. 228 (1993) 341.
P. Cvitanovi
, P.G. Lauwers, and P.N. S
harba
h, The planar se
tor of eld theories, Nu
l. Phys. B203 (1982) 385.
307
308
[Col79
REFERENCES TO CHAPTER 3
REFERENCES TO CHAPTER 3
[GN74
[GN79a
[GN79b
[GN80
[Gra91
[GW70
[GW80
[Ja
72
[Joh61
[Haa81
[HM89
[Hoo74a
[Hoo74b
[IKKT97
[KK80
309
D.J. Gross and A. Neveu, Dynami
al symmetry breaking in asymptoti
ally free eld theories, Phys. Rev. D10 (1974) 3235.
J.L. Gervais and A. Neveu, The quantum dual string wave fun
tional in Yang{Mills theories, Phys. Lett. 80B (1979) 255.
J.L. Gervais and A. Neveu, Lo
al harmoni
ity of the Wilson loop
integral in
lassi
al Yang{Mills theory, Nu
l. Phys. B153 (1979)
445.
J.L. Gervais and A. Neveu, The slope of the leading Regge traje
tory in quantum
hromodynami
s, Nu
l. Phys. B163 (1980) 189.
J.A. Gra
ey, Cal
ulation of exponent to O(1=N 2 ) in the O(N )
Gross Neveu model, Int. J. Mod. Phys. A6 (1991) 395, 2755(E).
D. Gross and J. Wess, S
ale invarian
e,
onformal invarian
e, and
the highenergy behavior of s
attering amplitudes, Phys. Rev. D2
(1970) 753.
D. Gross and E. Witten, Possible thirdorder phase transition in
the largeN latti
e gauge theory, Phys. Rev. D21 (1980) 446.
R. Ja
kiw, Field theoreti
investigations in
urrent algebra, in Le
tures on Current Algebra and its Appli
ations by S.B. Treiman,
R. Ja
kiw, and D.J. Gross, Prin
eton Univ. Press, 1972, p. 97; also
in Current Algebra and Anomalies by S.B. Treiman, R. Ja
kiw,
B. Zumino, and E. Witten, World S
i., 1985, p. 81.
K. Johnson, Solution of the equations for the Green's fun
tions of
a two dimensional relativisti
eld theory, Nuovo Cim. 20 (1961)
773.
O. Haan, Large N as a thermodynami
limit, Phys. Lett. 106B
(1981) 207.
M.B. Halpern and Yu.M. Makeenko, Continuumregularized loopspa
e equation, Phys. Lett. 218B (1989) 230.
G. 't Hooft, A planar diagram theory for strong intera
tions, Nu
l.
Phys. B72 (1974) 461.
G. 't Hooft, A twodimensional model for mesons, Nu
l. Phys.
B75 (1974) 461.
N. Ishibashi, H. Kawai, Y. Kitazawa and A. Tsu
hiya, A largeN
redu
ed model as superstring, Nu
l. Phys. B498 (1997) 467.
V.A. Kazakov and I.K. Kostov, Nonlinear strings in twodimensional U(1) theory, Nu
l. Phys. B176 (1980) 199.
310
[KK81
REFERENCES TO CHAPTER 3
REFERENCES TO CHAPTER 3
[Par82
311