Preprint typeset in JHEP style - HYPER VERSION SLAC-PUB-13144

NSF-KITP-08-16

Center-stabilized Yang-Mills theory: confinement and

large N volume independence
arXiv:0803.0344v2 [hep-th] 22 Mar 2008

Mithat Unsal1 and Laurence G. Yaffe2

1
SLAC and Physics Department, Stanford University, Stanford, CA 94305
2
Department of Physics, University of Washington, Seattle, Washington 981951560

Abstract: We examine a double trace deformation of SU (N ) Yang-Mills theory which, for

large N and large volume, is equivalent to unmodified Yang-Mills theory up to O(1/N 2 )
corrections. In contrast to the unmodified theory, large N volume independence is valid
in the deformed theory down to arbitrarily small volumes. The double trace deformation
prevents the spontaneous breaking of center symmetry which would otherwise disrupt large
N volume independence in small volumes. For small values of N , if the theory is formulated
on R3 S 1 with a sufficiently small compactification size L, then an analytic treatment of the
non-perturbative dynamics of the deformed theory is possible. In this regime, we show that
the deformed Yang-Mills theory has a mass gap and exhibits linear confinement. Increasing
the circumference L or number of colors N decreases the separation of scales on which the
analytic treatment relies. However, there are no order parameters which distinguish the small
and large radius regimes. Consequently, for small N the deformed theory provides a novel
example of a locally four-dimensional pure gauge theory in which one has analytic control
over confinement, while for large N it provides a simple fully reduced model for Yang-Mills
theory. The construction is easily generalized to QCD and other QCD-like theories.

Keywords: nonperturbative QCD.

[email protected]

[email protected]
Contents

1. Large N volume independence 1

2. Deformed Yang-Mills theory 3

2.1 Stabilization of center symmetry 6
2.2 Multiple compactified dimensions 7
2.3 Large N equivalence between ordinary and deformed YM 9
2.4 Large N volume independence of deformed YM theory 12
2.5 Addition of matter fields 12

3. Confinement at small radius and small N 13

3.1 Mass gap 19
3.2 String tensions 20
3.3 Larger size or larger N 22
3.4 Connection to integrable Toda theory 23

A. Monopole measure 25

1. Large N volume independence

The large N limit of SU (N ) Yang-Mills theories, when formulated on toroidal compactifi-

cations of Rd , are independent of volume provided the (ZN )d center symmetry is not spon-
taneously broken [1, 2, 3].1 However, above two dimensions, center symmetry does break
spontaneously when the (smallest) compactification circumference L is less than a critical
size Lc [4]. (If just one dimension is compactified, then this center-symmetry breaking transi-
tion is the usual thermally induced deconfinement transition.) In four dimensions, the critical
size Lc is approximately 1 where is the M S strong scale of the theory [4, 5].
Notwithstanding the limitation to L Lc , the volume independence of large N Yang-
Mills theory (partial reduction) has practical utility for lattice studies, because simulations
on lattices of size (Lc )d are sufficient to extract infinite volume properties of large N Yang-
Mills theory [4, 6, 7, 8]. But it would be even more helpful to have a formulation of the theory
in which volume independence holds for arbitrarily small volumes since this allows one to
reduce the lattice all the way down to a single site.

1
Center symmetry transformations are gauge transformations which are periodic only up to an element of
the center of the gauge group. Volume independence applies to the leading large N behavior of expectation
values and connected correlators of topologically trivial Wilson loops.

1
 Several schemes for preserving volume independence in arbitrarily small volumes have
been proposed. In so-called quenched reduced models, one constrains the eigenvalues of link
variables (or Wilson lines) in a manner which prevents Wilson lines from acquiring expectation
values [3].2 In the N limit, quenched reduced models correctly reproduce properties of
infinite volume Yang-Mills theory. However, corrections to the N = limit scale as 1/N , not
1/N 2 , in quenched reduced models. This makes extracting large N properties from numerical
simulations of quenched reduced models quite challenging. An alternative proposal, known
as twisted reduced models, involves modifying the Wilson action of a single-site model so
that the action explicitly disfavors configurations in which Wilson lines in different directions
mutually commute [10, 11]. Unfortunately, this clever scheme fails to work sufficiently close
to the continuum limit [12, 13]. In essence, the penalty imposed by the twisting of the action
is insufficient to overcome entropic effects which favor breaking of the center symmetry.
If light adjoint representation fermions are added to an SU (N ) Yang-Mills theory, and
periodic (not anti-periodic) boundary conditions imposed on the fermions, then the fermion
contribution to the Wilson line effective potential stabilizes the unbroken center symmetry
phase. Hence these QCD-like theories satisfy large-N volume independence for arbitrarily
small volumes [14]. (In addition, in the large-N limit C-even observables coincide between
these theories and corresponding theories, in sufficiently large volume, with fermions in the
rank-two symmetric or antisymmetric tensor representations [15, 16, 17].)
Motivated by this fermion-induced stabilization of center symmetry, in this paper we
introduce a simple scheme for preserving volume independence in pure Yang-Mills theory.
We add double trace terms to the action which prevent spontaneous breaking of center sym-
metry, while simultaneously perturbing the dynamics of the unbroken symmetry phase only
by O(1/N 2 ) corrections. This leads to a stabilized reduced model which reproduces the
dynamics of infinite volume Yang-Mills theory up to corrections which scale as 1/N 2 . The
construction may be easily generalized to other QCD-like theories with matter fields in rank-
one or rank-two representations.
In addition to providing a simple large N reduced model, the deformed Yang-Mills theory
is interesting in its own right when N is not large.3 When formulated on R3 S 1 , we show
that the large distance dynamics of the theory is analytically tractable provided N L  1.
In this regime, a semiclassical analysis (closely related to Polyakovs classic treatment of 3d
SU (2) adjoint Higgs theory [22]) reveals the existence of a mass gap and area law behavior of
spatial Wilson loops. It is noteworthy that our compactified, deformed Yang-Mills theory is
an analytically tractable confining theory with no fundamental scalar fields or supersymmetry,
in contrast to other instructive models of confinement [22, 25, 26, 27]. The confinement mech-
anism involves the formation of a dilute plasma of magnetic monopoles (and antimonopoles)
carrying topological charge 1/N .4

2
See Ref. [9] for an extended discussion of quenched and twisted reduced models.
3
See also related recent work in Refs. [18, 19, 20, 21].
4
Confinement due to such topological objects has been previously discussed in, for example, Refs. [23, 24]
and references therein. What is novel about our deformed theory at small N L is that this confinement
mechanism operates in a regime in which one has analytic control over the long distance dynamics.

2
2. Deformed Yang-Mills theory

We consider pure Yang-Mills (YM) theory with gauge group SU (N ) defined on the four
manifold R3 S 1 , with the S 1 having circumference L. The extension to multiple compactified
dimensions will be discussed below, but we begin with a single compactified dimension to
simplify the exposition. We start with the usual continuum action,
Z
YM 1 2
S = 2
tr F (x) , (2.1)
3
R S 1 2g
or a lattice formulation with the Wilson action,
X
S YM = tr U [p] + U [p] ,


(2.2)
2
p4

where the sum is over all oriented plaquettes, 4 is the four dimensional spacetime lattice, and
U [p] denotes the usual product of link matrices around the boundary of plaquette p. The
lattice coupling 2/g 2 . In our discussion, we will use both continuum and lattice formu-
lations, and benefit from both perspectives. As usual, a key virtue of the lattice formulation
is that it provides an Rexplicit non-perturbative definition of the theory.
Let (x) P (ei dx4 A4 (x,x4 ) ) denote the Wilson line (or Polyakov loop) operator the
holonomy of the gauge field around a circle wrapping the S 1 and sitting at the point x R3 .
We will construct a deformation of the Yang-Mills action on our compactified geometry by
adding terms, respecting all symmetries of the unmodified theory, built from the Wilson line
operator. The deformed action is given by

S deformed = S YM + S , (2.3)

with
Z
1
S P [(x)] (2.4a)
R3 L3
in the continuum, or
1 X
S P [(x)] (2.4b)
Nt3
x3

on the lattice. In the lattice form, Nt L/a denotes the size of the lattice in the compact-
ified direction and 3 4 is a three dimensional sublattice of the four dimensional lattice
corresponding to a fixed Euclidean time-slice. We want the deformation potential P [] to
guarantee the stability of the phase with unbroken center symmetry (at small volume). It
will be chosen to have the form
bN/2c
X
P [] an |tr (n )|2 , (2.5)
n=1

with positive coefficients {an } (and bN/2c denoting the integer part of N/2). In other words,
P [] is a sum of the double trace operators tr(n ) tr(n ) . When considering the large N
limit, the coefficients {an } will be held fixed as N .

3
 Under a center symmetry transformation by some element z ZN , the Wilson loop
tr(p ) is multiplied by z p . The value of p mod N , which determines the ZN representation,
is referred to as the N -ality. It will be important that the deformation potential (2.5) only
contains, by construction, absolute squares of Wilson loops with non-zero N -ality.
If P [] were simply proportional to |tr |2 , with a sufficiently large positive coefficient,
then this would prevent breaking of the center symmetry with htr()i as an order parameter.
But if N > 3 then this single term is not sufficient to prevent any spontaneous breaking
of center symmetry, as this term alone does nothing to prevent tr(2 ) from developing an
expectation value. In other words, a stabilizing term proportional to |tr()|2 cannot prevent
ZN breaking to Z2 (assuming N is even) with tr(2 ) as an order parameter. Adding an
additional stabilizing term proportional to |tr(2 )|2 could prevent such a breaking to Z2 , but
does not prevent breaking to Z3 (if N mod 3 = 0), or to any larger discrete subgroup of ZN .
This is why we have allowed P [] to include terms up to |tr bN/2c |2 .
We will argue that the deformed theory satisfies the following:

i) For suitable choices of the deformation parameters an (i.e., each coefficient sufficiently
large and positive) the stabilizing potential (2.5) will prevent the ZN center symmetry
from breaking to any subgroup.

ii) In the N limit, pure Yang-Mills theory on R4 is equivalent to the deformed theory
formulated on any R3 S 1 (for choices of the {an } satisfying point i). This equivalence
applies to expectation values of Wilson loops (on R4 ), or the leading large N behavior
of their connected correlators. In the lattice formulation, the number of sites in the
compactified direction may be reduced to one.

iii) When N L  1 limit, the deformed Yang-Mills theory is solvable in the same sense
as the Polyakov model. The existence of a mass gap and linear confinement can be
shown analytically. One can regard this regime as having spontaneous breaking of the
SU (N ) gauge symmetry down to U (1)N 1 , but this is a perturbative gauge dependent
description with no well-defined invariant content.

iv) There exist no order parameters which can distinguish the N L  1 Higgs regime
from the N L  1 regime in which gauge symmetry is restored.

As noted earlier, pure Yang-Mills theory on R3 S 1 satisfies volume independence (in

the large N limit) so long as the ZN center symmetry remains unbroken [2, 3, 4]. We will
show that this is equally true for the deformed theory; the additional terms in the action
do not affect the proof that volume independence, at N = , is an automatic consequence
of unbroken center (and translation) symmetry. In the undeformed theory, the unbroken
center symmetry phase is the low temperature confined phase, L > Lc with Lc 1 the
inverse deconfinement temperature. But in the deformed Yang-Mills theory, the stability of
the unbroken center symmetry phase is enforced by hand for all values of L.
Large N volume independence, and the large N equivalence between ordinary Yang-
Mills theory on R4 and the deformed theory on R3 S 1 , may be demonstrated by comparing
Dyson-Schwinger equations (i.e., Migdal-Makeenko loop equations) for expectation values

4
 ordinary YangMills deformed YangMills

deformation equivalence

orbifold
equivalence
L
Lc

combined
deformationorbifold
equivalence

0 0

Figure 1: Large N equivalences relating ordinary and deformed SU (N ) Yang-Mills theories, as a

function of the size L of the periodic volume in which the theories are defined. In deformed YM
volume independence holds for all L, while in ordinary YM volume independence fails below a critical
size, L < Lc , (shaded region) due to spontaneous breaking of center symmetry. This prevents reduction
all the way down to a single-site matrix model for ordinary YM. Large N equivalence holds between
ordinary and deformed YM theories as long as center symmetry is unbroken in ordinary YM. Large N
volume independence is a type of orbifold equivalence, as discussed in Ref. [14]. The combination of
volume changing orbifold projections in the deformed theory, along with the deformation equivalence
in sufficiently large volume, provides a useful equivalence between deformed YM in small volume
and ordinary YM in large volumes. In particular, a single-site matrix model of the deformed theory
will reproduce properties of ordinary Yang-Mills theory in infinite volume. The construction can be
generalized to QCD, with the deformed theory providing a fully reduced matrix model for QCD.

and correlators of Wilson loops, or alternatively by comparing the large N classical dynamics
that may be derived by using appropriate large N coherent states [2, 30, 31, 29]. Figure 1
summarizes the relation between the large N limits of ordinary and deformed Yang-Mills
theories. As long as the ZN center symmetry is not spontaneously broken, the dynamics of
the theories defined by S YM and S deformed are indistinguishable at leading order in the 1/N
expansion. In particular, glueball spectra of the two theories can differ only by order 1/N 2
effects. Agreement up to O(1/N 2 ) terms also applies to the string tension which characterizes
the area law behavior of large Wilson loops.
The large N equivalence between ordinary Yang-Mills theory and the deformed theory
in large volume, combined with the large N volume independence of the deformed theory,
circumvents the problems with previous formulations of reduced models for pure Yang-Mills
theory. Unlike the original Eguchi-Kawai model [1], its twisted variant [10, 11], and the partial

5
reduction of Refs. [4, 5], the equivalence to deformed YM theory remains valid in the limit
of zero compactification radius, irrespective of the value of the (bare) gauge coupling. And
for finite N , corrections to the large N limit scale as 1/N 2 , not 1/N as in quenched reduced
models [3]. Consequently, it should be possible to study the deformed theory, for relatively
modest values of N and vanishingly small volume, and obtain accurate results for properties
of ordinary Yang-Mills theory, in the large N limit, on R4 . As we discuss below, it is also
instructive to study the deformed theory for small values of N . In this regime, it will be seen
to provide a novel example of a confining theory, only involving an SU (N ) gauge field, which
is analytically soluble.

2.1 Stabilization of center symmetry

The possibility of preventing spontaneous breaking of center symmetry through the addition
of a deformation potential of the form (2.5) is largely self-evident. A positive coefficient an
suppresses configurations in which tr(n ) is non-zero. Although the pure-gauge dynamics
of the undeformed theory, in small volume, leads to an effective potential for the Wilson
line which is minimized when is an element of ZN , adding the deformation potential P []
changes the shape of the Wilson line effective potential. For sufficiently large values of the co-
efficients {an }, the effective potential will be minimized by configurations in which tr(n ) = 0
for 1 n bN/2c. This implies that tr(n ) = 0 for any integer n which is non-zero mod-
ulo N because, for SU (N )-valued matrices, tr(n ) is not independent of lower order traces
when n > bN/2c. Vanishing of these traces implies that the eigenvalues of are uniformly
spaced around the unit circle, so that the set of eigenvalues is invariant under ZN transfor-
mations (which multiply every eigenvalue by e2i/N ). This shows that the center symmetry
is not spontaneously broken. Henceforth, we assume that the coefficients {an } of the defor-
mation potential P [] are suitably chosen so as to enforce unbroken center symmetry for all
compactification radii.
This argument may be made much more explicit if one considers small compactifica-
tions, L  1 , so that (due to asymptotic freedom) the gauge coupling at the scale of the
compactification is small and the theory is amenable to a perturbative treatment. Quantum
fluctuations generate a nontrivial potential for the Wilson line [32]. In ordinary Yang-Mills
theory, integrating out the gauge field (and Faddeev-Popov ghosts) produces the functional
determinants
2
[det+ (Dadj )]1/2 [det+ (Dadj
2 2
)] = [det+ (Dadj )]1 , (2.6)
where det+ denotes a determinant in the space of periodic functions with period L. Therefore,
the effective potential for the Wilson line is

V [] = L1 ln det+ (Dadj
2
). (2.7)

For constant (or slowly varying) configurations, the evaluation of the functional determinant
is straightforward and yields [32]
Z
1
V [] = 4
V[(x)] , (2.8)
R3 L
with

6

2 X 1
V[] 2 |tr n |2 . (2.9)
n4
n=1

For sufficiently small L, corrections to this one-loop result are negligible. The effective po-
tential (2.9) is minimized when the Wilson line is an element of the center, = e2ik/N ,
k = 0, , N 1, demonstrating the spontaneous breaking of ZN symmetry, in ordinary
Yang-Mills theory, for sufficiently small compactifications.
To force unbroken center symmetry in the deformed theory, the deformation potential
P [] must overcome the effect of the one-loop potential (2.9). A simple specific choice for
the deformation coefficients which, in the continuum limit and for sufficiently R large N , ac-
2 4 4
complishes this is an = 4/( n ). For this choice, the deformation potential R3 L P [] is
minus twice the one-loop Wilson line effective potential (2.9) of the undeformed theory, so the
net effect of the deformation is to flip the sign of the effective potential for the Wilson line.
The resulting combined potential is minimized when tr n = 0 for all n which are non-zero
modulo N , indicating unbroken center symmetry.

2.2 Multiple compactified dimensions

Instead of compactifying just one dimension, one may consider the gauge theory on R4d T d ,
where T d is a d > 1 dimensional torus. For simplicity, we will discuss the case of a symmetric
torus having size L in each dimension. Two particularly interesting cases are d = 3, where
the L 0 limit will reduce to SU (N ) matrix quantum mechanics, and d = 4, where the zero
size limit will reduce to a matrix model.
With multiple compactified dimensions, the center symmetry is (ZN )d . The preceding
discussion generalizes in a straightforward fashion, but there is one significant change: with
multiple compactified dimensions there is a richer set of possible symmetry realizations in
which the center symmetry is partially broken. One needs to craft the deformation potential
in such a fashion that all unwanted symmetry realizations are suppressed.
With multiple compactified directions, one may define distinct Wilson line operators
1 , 2 , ..., d each of which wraps one elementary cycle of the torus. The classical Yang-
Mills action suppresses configurations in which Wilson lines in different directions are non-
commuting. (In other words, configurations with vanishing field strength must have Wilson
lines which commute and are covariantly constant.) A deformation potential with the form
(2.5), involving just one Wilson line i , can prevent symmetry realizations in which the
ZN center symmetry associated with direction i spontaneously breaks in a manner which
is independent of the ZN symmetries associated with other directions. However, it is also
possible for the center symmetry to break in a manner which leaves a unbroken subgroup
involving correlated symmetry transformations in different directions. With, for example,
d = 2, suppose that the (ZN )2 symmetry breaks to the diagonal ZN subgroup. Such a
symmetry realization does not allow expectations of tr k1 or tr k2 to be non-vanishing (for
any k which is non-zero modulo N ). But this realization does allow tr 1 2 to have a non-
zero expectation value. To prevent such a symmetry realization, it may be necessary to add
a |tr 1 2 |2 term to the deformation potential.

7
 To see that this concern is not moot, it is instructive to examine the one-loop effective
potential for Wilson lines when there are multiple compactified directions. Once again, if the
physical size of the d-torus is much smaller than the inverse strong scale 1 , then Kaluza-
Klein modes with momenta at and above the compactification scale can be integrated out
perturbatively. This produces a one loop effective potential for mutually commuting Wilson
lines given by [33, 34]

tr n1 nd 2


1 X 1 d
V[1 , , d ] 2
2 . (2.10)
n21 + + n2d
(n1 , ,nd )(Zd 0)

This effective potential has N d degenerate global minima at which each Wilson line is an
independent element of the center. This demonstrates the spontaneous breaking of (ZN )d
symmetry (down to the identity) in ordinary Yang-Mills theory on R4d (S 1 )d for sufficiently
small compactifications.5 Adding a deformation to the Yang-Mills action which is the direct
generalization of the previously discussed deformation (2.4), namely

XZ 1
S = P [i ] , (2.11)
R4d L4d
i

with P [] having the form (2.5), will deform the effective potential and can prevent this
symmetry realization in which the (ZN )d symmetry breaks down to nothing.
However, the fact that all terms in the effective potential (2.10) have negative coefficients
demonstrates that the center symmetry will partially break spontaneously even in the pres-
ence of stabilizing terms of the form (2.11). In particular, even when all traces wrapping a
single compactified direction vanish, the potential (2.10) favors configurations in which traces
wrapping multiple cycles (such as tr 1 2 ) are non-zero relative to the center-symmetry pre-
serving configuration we are trying to stabilize (for which tr n1 1 nd d = 0 for all ni that
are non-zero modulo N ).
Therefore, to prevent any spontaneous breaking of the center symmetry when there are
multiple compactified directions one must allow the deformation of the action to include
absolute squares of order parameters which wrap multiple cycles of the torus. In other words
we need to have Z
1
S = 4d
P [1 , , d ] , (2.12)
R4d L

5
More precisely, this demonstrates the spontaneous breaking of (ZN )d symmetry down to the identity
when d = 1 or 2. If d > 2 then there are fewer than two non-compactified directions. In this case, the center
symmetry cannot break spontaneously for any finite value of N , due to the non-zero probability of fluctuations
(or tunneling events) which effectively average over all degenerate minima of the Wilson line effective potential.
However, the probability of such fluctuations vanish exponentially with increasing N . Consequently, in the
N limit one can have spontaneous breaking of the (ZN )d center symmetry even when all directions are
compactified. In order to have a large N equivalence between the compactified theory and Yang-Mills theory
on R4 , one must prevent spontaneous breaking of the center symmetry in the large N limit. Therefore, the
more complicated form of the stabilizing potential discussed below is necessary for all d > 1.

8
where the deformation potential has the form
bN/2c
X
2
an1 nd tr n1 1 nd d  ,

P [1 , , d ] = (2.13)
n1 , ,nd =bN/2c

with coefficients an1 nd which are sufficiently large and positive.

A simple specific choice for the coefficients which, in the continuum
2 limit and for suf-
ficiently large N , accomplishes this is an1 nd = 22 n21 + + n2d . For this choice, the
deformation potential is minus twice the one-loop effective potential (2.10) of the undeformed
theory, so once again the net effect of the deformation is to flip the sign of the effective po-
tential for the Wilson line. The resulting combined potential has a unique center-symmetry
preserving minimum.6
In the following sections we will, for simplicity, largely focus on the case of just one
compactified direction.

2.3 Large N equivalence between ordinary and deformed YM

The most direct way to demonstrate equivalence between ordinary Yang-Mills theory and our
deformed theory is to compare the Schwinger-Dyson (or loop) equations for gauge invariant
observables. As usual, for a rigorous treatment it is appropriate (and convenient) to work with
lattice regulated formulations of both theories. It is also convenient to consider U (N ) gauge
theories instead of SU (N ); the difference in gauge groups only affects subleading O(1/N 2 )
relative corrections to Wilson loop expectation values or connected correlators.
Let `a denote an operator which varies individual link fields according to `a (U [`0 ])
``0 ta U [`], where {ta } is a set of U (N ) Lie algebra basis matrices satisfying tr ta tb = 21 ab .
Invariance of the Haar measure implies that the integral of any variation vanishes,
Z
d0 `a (anything) = 0 , (2.14)

where d0 `0 dU [`0 ]. Choose (anything) to be eS `a W [C], where W [C] N1 tr U [C] is the

Q

Wilson loop around some closed contour C. Summing over the Lie algebra index a and the
link ` yields Z
d0 eS S W [C] + 2 W [C] = 0 ,


(2.15)

6
In the case of R T 3 , the undeformed theory was studied in detail by Luscher, van Baal and others in
the mid 80s. (See the reviews [35, 36] and references therein.) The goal of this QCD in a box program
was to use asymptotic freedom combined with the absence of the phase transitions in finite volumes to extract
lessons about QCD on R4 . This approach confines the theory to its short distance perturbative regime where
it is perturbatively solvable. However, features of this small universe are very different from QCD on R4 . In
particular, when the physical box size becomes comparable to the inverse strong scale 1 there is a cross-over
from a world of hadrons to a world of quarks and gluons. As noted in footnote 5, this cross-over becomes a
sharp phase transition in the large N limit.
In our deformed Yang-Mills theory, as will be discussed below, there are no phase transitions as a function
of box size even in the N = limit. For the deformed theory we will find that it is not 1/L which acts
as an infrared cutoff, but rather 1/LN . If LN  1 then the deformed theory reproduces the dynamics of
Yang-Mills theory on R4 .

9
where the dotR product is shorthand for the sum over a and `. Dividing by the partition
function Z d0 eS yields relations among expectation values,

S W [C] + 2 W [C] = 0 .



(2.16)

These are Schwinger-Dyson equations for Wilson loop expectation values.

After working out the action of the variations, the result may be expressed in a purely
geometric form. For lattice gauge theory with the Wilson action one finds [37],

1

 X X h

i
2 |C| W [C] =
4N
W [(p)C] W [(p)C]
`C p| `p
X
W [C 0 ]W [C 00 ] .


+ (2.17)
selfintersections

Here |C| is the length of the loop C (i.e., the number of links in the loop), W [(p)C] denotes a
Wilson loop which goes around the boundary of plaquette p (which contains a link contained
in the contour C) and then around the contour C, and p denotes the oppositely oriented
plaquette boundary. The sum over self-intersections runs over all ways of decomposing a
loop C which multiply traverses some link ` into two separate loops, C = C 0 C 00 , with the
associated sign determined by whether C 0 and C 00 traverse the link ` in the same or opposite
directions. See Ref. [28] for more detailed discussion.

In the large N limit, with e N = 2 held fixed (where g 2 N is the t Hooft coupling),
all N dependence disappears. Fluctuations in the values of Wilson loops vanish in this limit
(their distributions become arbitrarily sharply peaked). This is a reflection of the classical
nature of the large N limit [2], and implies that the expectation value of a product of loops
factorizes, up to 1/N 2 corrections,

W [C 0 ]W [C 00 ] = W [C 0 ] W [C 00 ] + O(1/N 2 ) .




(2.18)

(The O(1/N 2 ) remainder is the connected correlator.) Consequently, in the large N limit
Wilson loop expectation values satisfy a closed set of nonlinear algebraic equations,

1

 X X e h

i
2 |C| W [C] =
4
W [(p)C] W [(p)C]
`C p|lp
X
W [C 0 ] W [C 00 ] .



+ (2.19)
selfintersections

The loop equations in the concise form (2.16) are equally valid for the deformed theory.
The only difference is that S now includes the double trace deformation S, and this generates
new terms in the loop equations given by h(S) W [C]i. Just as the usual Wilson action
leads to terms in which a plaquette is inserted into the loop C, the piece of S proportional
to |tr k |2 generates terms in the loop equation in which the topologically non-trivial loop
k (or its inverse) is sewn into the loop C (if C contains links pointing in the compactified

10
direction). But because S contains absolute squares of traces, each such term is multiplied
by the complex conjugate of the trace of the inserted loop. Hence,
 X
bk [C] W [k C] W [k ] ,



(S) W [C] = (2.20)
k6=0

where k C denotes a loop obtained by concatenating k and C at their intersection links,

and the coefficients bk [C] are proportional to a|k| but also depend on the number of links in
C which point in the compactified direction. The essential point is that the variation acts
on one of the two traces comprising the double trace deformation, leaving the other trace
unchanged. In the large N limit, due to factorization,

W [k C]W [k ] = W [k C] W [k ] + O(1/N 2 ) .




(2.21)

But for any k which is non-zero modulo N , W [k ] transforms non-trivially (acquiring a

phase e2ik/N ) under a ZN center symmetry transformation. Hence its expectation value is
an order parameter for the center symmetry and hW [k ]i must vanish in any phase with
unbroken center symmetry. As discussed above, the deformed theory is constructed so as to
ensure unbroken center symmetry for all compactification radii. Consequently, all additional
terms in the loop equations generated by the deformation of the action vanish in the large N
limit,
(S) W [C] = O(1/N 2 ) ,


(2.22)
implying that Wilson loops in the original and deformed Yang-Mills theory satisfy identical
large-N Schwinger-Dyson equations.
Ordinary Yang-Mills theory has unbroken center symmetry only for sufficiently large
compactifications, L > Lc . The coinciding large N loop equations in ordinary and deformed
Yang-Mills theories imply that Wilson loop expectation values in these two theories have
identical large N limits when L > Lc .7
The same approach may be used to compare the Schwinger-Dyson equations satisfied by
connected correlators of two or more Wilson loops, with exactly the same conclusion: the
leading large N behavior of connected correlators coincide between ordinary and deformed
Yang-Mills theories, provided L > Lc . Thus, in sufficiently large volume the net effect of
the double trace deformation on the dynamics of the theory is O(1/N 2 ), and vanishes in the
large N limit. In another words, the physics of the deformed Yang-Mills theory depends on
the deformation parameters {ai } only in the combination ai /N 2 which vanishes at N = .
This demonstrates the nonperturbative equivalence of ordinary Yang-Mills theory and the
deformed YM theory, formulated on R3 S 1 (or more generally, on any toroidal compacti-
fication of flat space), provided the compactification size is above the critical size for center
symmetry breaking in the undeformed theory.
7
This argument, that coinciding loop equations imply coinciding expectation values, glosses over the possi-
bility that the infinite set of loop equations may have multiple solutions which respect center symmetry, with
different theories potentially corresponding to different solutions of the same set of equations. The alternative
approach of comparing the N = classical dynamics generated by appropriate large N coherent states, dis-
cussed in Ref. [29], eliminates this loophole and demonstrates equivalence in any phase of the theories which
satisfy the necessary and sufficient symmetry realization conditions.

11
2.4 Large N volume independence of deformed YM theory
Unbroken center symmetry is necessary and sufficient for the validity of the large N volume
independence of Yang-Mills theory (or more general gauge theories containing adjoint repre-
sentation matter fields). This may be demonstrated by comparing large N loop equations,
or the N = classical dynamics generated by suitable coherent states [14]. Corrections to
this equivalence for finite N scale as 1/N 2 . The loop equation analysis is very similar to that
sketched above. In the large N loop equations for topologically trivial Wilson loops, one finds
that the only volume-dependent terms (arising from self-intersections) automatically vanish
as long as the center symmetry is not spontaneously broken.
The analysis of large N volume independence in Ref. [14] applies equally well to the
deformed theory which, by construction, has unbroken center symmetry for any compactifi-
cation radius. Because the double trace operators in P [] are squares of loops with non-zero
N -ality, the presence of the deformation potential P [] has no effect on the large N classical
dynamics of center-symmetry symmetric states. Consequently, deformed Yang-Mills theory,
in the large N limit, is completely volume independent.
In the lattice formulation, if one compactifies all directions then one may reduce the
lattice size all the way down to a single site, in which case the theory becomes a simple
matrix model of Wilson lines {i } running in each lattice direction with action,
d
X
deformed
tr i j i j + j i j i + P [1 , , d ] .


Ssinglesite = (2.23)
2
i>j=1

The large N limit of this matrix model will reproduce the leading large N behavior of expec-
tation values and connected correlators of Wilson loops in uncompactified Yang-Mills theory.
As discussed in the Introduction, the single-site deformed Yang-Mills theory (2.23) provides
a simple generalization of Eguchi-Kawai reduction which is valid for any value of the lattice
coupling .

2.5 Addition of matter fields

Consider adding Nf species of matter fields (either fermions or scalars) in the fundamental
representation to SU (N ) Yang-Mills theory, either ordinary or deformed, with one dimension
compactified. The addition of fundamental representation matter explicitly breaks the ZN
center symmetry. However, if Nf is held fixed as N , then the fundamental representation
matter fields have only a subleading O(Nf /N ) effect on the gauge field dynamics. As a result,
everything discussed above remains valid. That is, the leading large N behavior of expectation
values or connected correlators of Wilson loops in the undeformed theory, in sufficiently
large volume, coincide with the corresponding observables in the theory, in arbitrary volume,
deformed by the addition of the stabilizing potential P []. In addition, one may also show
that the same equivalence applies to the leading large N behavior of mesonic expectation
values and connected correlators.8 Note, however, that these large N equivalences cease to
apply if Nf /N is held fixed as N [14].
8
This is easiest to understand by considering the equivalent gluonic observables produced by integrating
out the matter fields. For the case of large N volume independence in the undeformed theory, see Ref. [14] for
details. The presence of the deformation potential does not affect this analysis.

12
 If adjoint representation matter fields are added to the theory (ordinary or deformed),
then the large N equivalences discussed above also remain valid. Adding adjoint representa-
tion fields enlarges the natural set of gauge invariant observables from simple Wilson loops to
Wilson loops decorated by arbitrary numbers of insertions of adjoint matter fields. But the
presence of adjoint matter fields preserves the center symmetry of the underlying Yang-Mills
theory. As a result, the above-described comparison of large N loop equations (or large N
classical dynamics) between the ordinary and deformed theories immediately generalizes to
the case of Yang-Mills theories with adjoint matter, with exactly the same conclusions.9
Finally, one may also consider the addition of matter fields in rank-two antisymmetric or
symmetric tensor representations (yielding theories we will refer to as QCD(AS) or QCD(S),
respectively). The presence of fields in these representations reduces the U (1) center sym-
metry of U (N ) Yang-Mills theory down to Z2 . Given the central role the center symmetry
played in the above large N equivalences, one might think this reduction in center symmetry
would destroy these large N equivalences. This is not the case. One way to see this is to
note, as discussed in Ref. [14], that volume-dependent terms in the N = loop equations
only appear if loops with non-zero winding number around the compactified direction acquire
non-zero expectation values. The addition of the deformation potential P [] prevents topo-
logically non-trivial Wilson loops from acquiring non-zero large N expectation values, even in
small volumes. Consequently, the situation is analogous to the pure gauge case: the leading
large N behavior of expectation values or connected correlators of single trace observables in
QCD(AS/S) in sufficiently large volume coincide with the corresponding observables in the
theory modified by the addition of the deformation potential in arbitrary volume, as depicted
in Fig.1.10

3. Confinement at small radius and small N

When compactified on a small circle, L  1 , the gauge coupling of the deformed theory is
small at the compactification scale, g 2 (1/L)  1. As discussed earlier, the combined potential

9
For a detailed discussion of loop equations in theories with adjoint matter, see Ref. [28].
10
Another way to understand this is to note the existence of a large N equivalence (so-called orientifold
equivalence) between theories with rank-two symmetric or antisymmetric representation matter and corre-
sponding theories with adjoint representation matter [QCD(adj)] [15, 16, 17]. This large N equivalence
applies to the charge-conjugation even sectors of the two theories, and only holds if charge conjugation sym-
metry is not spontaneously broken. When, for example, the matter fields are fermions with periodic boundary
conditions, examination of the Wilson line effective potential shows that QCD(AS/S) does spontaneously
break both charge conjugation and center symmetry when compactified with sufficiently small size [15]. But
the addition of a deformation potential of the form (2.5) (with sufficiently positive coefficients) will prevent this
spontaneous symmetry breaking, just as it does in the pure Yang-Mills case. Since QCD(adj) satisfies large N
volume independence (as long as its center symmetry is not spontaneously broken), the same large N volume
independence must also apply to QCD(AS/S) (as long as charge conjugation is not broken). In sufficiently large
volumes, there is no reason to believe that charge conjugation symmetry breaks spontaneously in QCD-like
theories with rank-two tensor representation matter. Therefore, large N orientifold equivalence combines with
large N volume independence of QCD(adj) to imply volume independence in QCD(AS/S) as long as center
and charge conjugation symmetries are not spontaneously broken which is what the deformation potential
ensures.

13
V[] + P [] is minimized when
 
= Diag 1, e2i/N , e4i/N , , e2i(N 1)/N , (3.1)

up to conjugation by an arbitrary SU (N ) matrix. Working in a gauge in which is diagonal,

and using (temporarily) gauge-dependent language, this configuration may be regarded as
breaking the gauge symmetry down to the maximal Abelian subgroup,

SU (N ) U (1)(N 1) . (3.2)

Modes of the diagonal components of the SU (N ) gauge field with no momentum along the
compactified x4 direction describe photons associated with the Cartan subgroup of SU (N ).
Modes of the diagonal components of the gauge field with non-zero momentum in the com-
pactified direction form a Kaluza-Klein tower and receive masses which are integer multiples
of 2/L. The off-diagonal components of the SU (N ) gauge field describe Kaluza-Klein tow-
ers of W -bosons which are charged under the unbroken U (1)N 1 gauge group. The non-zero
value of A4 (i/L) ln shifts the masses of these off-diagonal components by multiples of
2/(N L). The net effect is that there are charged W -bosons with masses

2 k
mWk = , k = 1, 2, , . (3.3)
NL
For later convenience, we define mW to be the mass of the lightest W bosons,
2
mW . (3.4)
NL
This is the mass scale below which the dynamics is effectively Abelian.11 Note that, at fixed
L, the lightest W bosons have masses which become small when N . This will be
important in the discussion of the large N behavior of the deformed YM theory. But first, in
this section, we consider the dynamics of the deformed theory when N is fixed and small.
The N 1 photons of the Cartan subgroup do not couple (directly) to the Wilson line and
remain massless to all orders in perturbation theory. Thus, a strictly perturbative analysis
would lead one to expect that the deformed theory, for sufficiently small L, would have a
non-confining Coulomb phase. We will see that this is incorrect nonperturbative effects
lead to the generation of a mass gap and produce confining long distance physics.
The analysis of non-perturbative properties in our compactified deformed Yang-Mills
theory is very similar to Polyakovs treatment of the 3d Georgi-Glashow model [22]. But
instead of a three-dimensional theory with a non-compact Higgs field, we have a compactified
four-dimensional theory with the group-valued Wilson line serving as a compact Higgs
field. For theories involving massless complex fermions, the difference between compact and

11
Fluctuations in the eigenvalues of away from the minimum (3.1) correspond to neutral Higgs bosons.
The masses ofthese fluctuations
depend on the coefficients {an } of the deformation potential but parametrically
are of order /L N mW . This scale will be large compared to the mass scale of the non-perturbative
3d dynamics, and these fluctuations will play no role in the following discussion.

14
noncompact Higgs systems can be major [38]. However, in our case, the differences relative
to Polyakovs classic discussion are rather minimal.
Due to the SU (N ) U (1)N 1 gauge symmetry breaking, there exist topologically
stable, semiclassical field configurations, namely monopoles [32]. At the center of a monopole,
one pair of eigenvalues of the Wilson line become degenerate. For fundamental (i.e., minimal
action) monopoles, this will be a pair of eigenvalues which are nearest-neighbors at infinity.
If the adjoint Higgs field was noncompact, then there would be N 1 species of fundamental
monopoles. This follows from the topological considerations: the second homotopy group
2 [SU (N )/U (1)N 1 ] = 1 [U (1)N 1 ] = ZN 1 , implying that fundamental monopoles come in
N 1 varieties. However, with a compact Higgs field there is an extra fundamental (Kaluza-
Klein) monopole which arises due to the fact that the underlying theory is formulated on a
cylinder, R3 S 1 , or equivalently that the configuration space of is compact.
The monopoles may be characterized by their magnetic charges, topological charge, and
their action. The magnetic charges of the N different types of fundamental monopoles are
proportional to the simple roots and affine root of the Lie algebra of the unbroken U (1)N
gauge group.12 The simple roots are given by13

1 = (1, 1, 0, . . . , 0) = e1 e2 , (3.5a)
2 = (0, 1, 1, , . . . , 0) = e2 e3 , (3.5b)
..
.
N 1 = (0, . . . , 0, 1, 1) = eN 1 eN , (3.5c)

and the affine root is

N
X 1
N j = (1, 0, 0, . . . , 1) = eN e1 . (3.6)
j=1

For later convenience, let 0aff denote the affine (extended) root system of the the associated
Lie algebra,
0aff {1 , 2 , . . . , N 1 , N } . (3.7)
It is the affine root system which is relevant for compact Yang-Mills Higgs systems. The roots
i 0aff obey
i j = 2i,j i,j+1 i,j1 , i, j = 1, . . . N . (3.8)
The form (3.8) of these inner products will translate into self and nearest neighbor interactions
between monopoles in the Dynkin space. The above choice of basis is natural due its visual
12
Even though the gauge symmetry breaking is SU (N ) U (1)N 1 , for ease of presentation it is convenient
to add an extra photon to the original theory and discuss U (N ) U (1)N . This simplifies the discussion of
charge assignments of monopoles, and the affine roots of the associated Lie algebra. In the continuum limit
of the theory this extra photon completely decouples from the other degrees of freedom and may simply be
ignored. It should not be confused with the N 1 photons which have non-trivial nonperturbative dynamics.
13
This set of simple roots corresponds to choosing Lie algebra generators normalized to satisfy tr ta tb = ab ,
instead of 12 ab as in the previous section.

15
simplicity, but the inner products (3.8) of the roots of the associated Lie algebra are basis
independent.
1 4
Let F 2g  F denote the U (1)N valued 3d magnetic field, with conventional
perturbative normalization. (In a gauge where is diagonal, F is just the list of diagonal
elements of the original non-Abelian field strength, multiplied by 1/g.) The magnetic charges
of a monopole of type i = 1, , N are given by the root i (up to a factor of 2/g),
Z
2
d F = i [type (i) monopole] . (3.9)
S2 g

(The S 2 is an arbitrarily large sphere in R3 . The flux is independent of the value of x4 at

which the integral is performed, as the long distance monopole fields are independent of x4 .)
The topological charge is correlated with the magnetic charge of the monopole. For
fundamental monopoles with magnetic charges i 0aff , the topological charge is
Z
1 1
2
tr F Fe = . (3.10)
R3 S 1 16 N

For antimonopoles with magnetic charges i , the topological charge = 1/N .

The electric charges of W -bosons may also be simply expressed in terms of the affine
roots 0aff . The lightest W bosons, with mass mW , may be labeled by a single root which
gives their electric charges (up to a factor of g),

QWi = g i . (3.11)

W -bosons in the next heavier multiplet are labeled by a pair of neighboring roots, and have
charges
QWi +i+1 = g (i + i+1 ) , (3.12)
etc. Dot products of the W -boson charges and monopole charges obey the Dirac quantization
condition,

2 4 , for i = j ;
QWi QMj = g i j = 2(2ij i,j+1 i,j1 ) = 2 , for i = j 1 ; (3.13)
g
0, otherwise.

Conjugation by a ZN shift matrix, which is part of the global gauge symmetry, cyclically
permutes the Wilson line eigenvalues and hence cyclically permutes the N different species of
fundamental monopoles. The presence of this symmetry (which is one of the features which
distinguishes compact and non-compact Higgs systems) guarantees that the N different types
of fundamental monopoles have identical values of the action. Monopole solutions are self-
dual,
F = Fe , (3.14)
and hence the Yang-Mills action of a fundamental monopole (or antimonopole) is
 8 2 8 2
Z Z 
1 2
 1
SYM = 2
tr F =  2
tr F Fe  = 2 || = 2 . (3.15)
R3 S 1 2g R3 S 1 2g g g N

16
After adding the contributions of the deformation potential P [] and the induced one-loop
effective potential V[], the complete monopole action will differ from this value. But the
deviation is perturbative in g 2 , so the monopole action

8 2
S0 SYM + S + S1loop = + O(1) . (3.16)
g2N

The correct infrared description of the deformed Yang-Mills theory on R3 S 1 at small

radius is generated by a dilute gas of monopoles (and antimonopoles) of N different types,
interacting via the species-dependent long range Coulomb potential,
 2  2
2 (i ) (j ) 2 2ij i,j+1 i,j1
Vi,j (r) = L = L . (3.17)
g 4|r| g 4|r|

(The overall sign is plus for monopole-monopole, and minus for monopole-antimonopole.)
Hence, we are dealing with a multi-component classical plasma, with nearest-neighbor inter-
actions in the Dynkin space. As with any classical plasma, this system will exhibit Debye
screening. The field due to a static external magnetic charge will fall exponentially with
distance, |F| emD r /r, with m1 D the characteristic Debye screening length. This implies
that external fields cannot propagate coherently over distances large compared to the Debye
length, which will be the longest correlation length in the system. The Debye mass mD will
appear as a dynamically generated photon mass. This will be shown explicitly.
For momentum scales small compared to the lightest W mass, the equilibrium dynamics
is correctly represented by a grand canonical ensemble of all types of monopoles and anti-
monopoles. Consider a configuration in which there are n(i) monopoles and n(i) antimonopoles
(i) (i)
of types i = 1, , N , located at positions rk , k = 1, , n(i) and rl , l = 1, , n(i) , re-
spectively. The magnetic field generated by this ensemble of magnetic charges is

N n(i) (i) n(i) (i)
X 2 X x rk X x rl
B(x) = i (i) 3
(i) 3
. (3.18)
g 4|x r | 4|x r |
i=1 k=1 k l=1 l

The action of such a monopole configuration is the sum of the monopole self-energies plus
their potential energy due to Coulomb interactions,
N 
X 
Smonopolegas = S0 n(i) + n(i) + Sint , (3.19)
i=1

with

N n(i) X
n(j) n(i) X
n(j) n(i) X
n(j)
2 2 L X X (i) (j)
X (i) (j)
X (i) (j)
Sint = i j G(rk rl ) + G(rk rl ) 2 G(rk rl ) ,
g2
i,j=1 k=1 l=1 k=1 l=1 k=1 l=1
(3.20)
and
1
G(r) . (3.21)
4|r|

17
 The grand canonical partition function of this multi-component Coulomb gas is

N X n(i) (i)
Y n
(i)
X n
(i) Z
Y (i) Z nY (i)
Z= drk drl eSint , (3.22)
(i) n(i) ! (i) n(i) ! R3 3
R l=1
i=1 n =0 nk=1=0

where
2 /N g 2 (m
C eS0 = A m3W (g 2 N )2 eS e8 W)
(3.23)
is the monopole fugacity. The prefactor C represents the one-loop functional determinant in
the monopole background. Extracting the zero-modes of the small fluctuation operator via
the usual collective coordinate procedure leads to factors of (g 2 N )2 m3W . (See the appendix
for details.) If the coupling is evaluated at the scale mW , which is natural for this problem,
then the non-zero mode part of the one-loop determinant merely gives rise to an overall
dimensionless (and N independent) coefficient A. In the final form of (3.23), S denotes the
deformation term in the action (2.4) evaluated in the background of a fundamental monopole.
This is an O(1) number, independent of the coupling g 2 , whose explicit value depends, of
course, on the deformation parameters {an }. [For large N , S scales as O(1/N ).]
Using the fact that G(r) is the Greens function for the 3d Laplacian, this partition
function can be exactly transformed into a 3d scalar field theory with an N -component real
scalar field,
Z YN
Z= Di eSdual [] , (3.24)
i=1

where
Z h 1  g 2 N
X i
dual 2
S = () cos(i ) . (3.25)
R3 2L 2
i=1

To verify this, it is easiest to start with the functional integral (3.24), rewrite the cosines in
terms of exponentials of , expand the exponential of each of the resulting interaction terms
in a power-series in eS0 eii , and then perform the functional integral over . The scalar
fields i appearing in this representation are dual fields for the 3d Abelian gauge fields Ai .14
The fields {i } should be regarded as compact scalar fields defined modulo 2. In addi-
tion to invariance under 2 shifts in any component of , note that the monopole induced

14
In three dimensions, Abelian duality relates a photon to a compact scalar. With j (x) the compact scalar
(j)
dual to the photon A (x) of the jth U (1) subgroup, the Abelian duality relations are

4i g2
d j = 21 L Im( ) F (j) , (L1 ) = + , (j)
F =  j .
g2 2 2L
(j) (j)
The 3d Maxwell action becomes 4gL2 (F )2 = LIm( 16
)
(F )2 = 2LIm(
1
)
( j )2 . The path integral of the
Abelian gauge theory in the presence of a monopole with charge j located at position x is equivalent to
the insertion of eij (x) into the path integral over the dual scalar fields [22, 27]. The complete partition
function of the long distance effective theory is a sum over all topological sectors, each of which may contain
an arbitrary number of monopoles and antimonopoles (whose charges sum to give the appropriate topolog-
ical class). Summing over all numbers and locations of monopoles (and antimonopoles), weighted with the
appropriate fugacity, directly yields the result (3.25).

18
interaction vertex has the additional shift symmetry

+ 2i , i = 1, . . . , N 1 (3.26)

where {i } are the N 1 fundamental weights of the SU (N ) algebra. These are defined by
the reciprocity relation with the simple roots,

i j = 21 ij j2 = ij (3.27)

for i = 1, , N 1, which implies that the fundamental weights {i } form a basis which is
dual to the fundamental roots {j }. The presence of the symmetry (3.26) is related to the
fact that the vacuum of the original theory can be probed by N 1 different types of external
charges, distinguished by their (non-zero) values of N -ality. This will be discussed below.
Including a non-zero theta parameter in the original Yang-Mills action,
Z
YM YM 1
S S + i 2
tr F Fe , (3.28)
3
R S 1 16
has the effect, in the grand canonical partition function (3.22), of multiplying the monopole
fugacity by ei/N and antimonopole fugacity by ei/N . In the dual representation (3.25),
this amounts to shifting the argument of the cosine by /N , so that the interaction becomes
N
X
cos(i + i/N ) . (3.29)
i=1

In this form, 2 periodicity of the theory with respect to is not manifest. However, a shift of
the dual scalar fields, + (/N ) with (0, 1, 2, , N 1), converts the interaction
term to the manifestly 2 periodic form
"N 1 #
X
cos(i ) + cos(N + ) , (3.30)
i=1

in which theta dependence only appears in the term involving the affine root. For simplicity,
in the following subsections we will focus on the case of = 0.

3.1 Mass gap

The cosine potential in the dual action (3.25) generates a mass term for the photons. Rescaling
to put the kinetic term into canonical form and expanding the potential to quadratic order
around the minimum at = 0 gives
N
X
1 2
V (i ) = (const.) + 2 m (i+1 i )2 , (3.31)
i=1

with N +1 1 and
3
(2)2

2 2 /(N g 2 (m ))
m2 L = A m2W eS e8 W
. (3.32)
g2 g2N

19
A ZN Fourier transform,
N
1 X 2ipj/N
p e j , p = 0, , N 1 , (3.33)
N j=1

diagonalizes this mass term and yields

N
X 1
V (i ) = (const.) + 1
2 m2p |p |2 , (3.34)
p=0

with
p
mp m sin. (3.35)
N
Expressing m in terms of the renormalization group invariant scale , defined by

b1 /b0 2 /( N g 2 ())
b0 = b0 N g 2 () e8 (3.36)

(with b0 = 11/3 and b1 = 17/3), yields

e (N L)5/6 |ln N L|9/11 ,
m = A (3.37)

where A e is an O(1) coefficient. Relative corrections suppressed by powers of g 2 (mW )

1/| ln N L| have, of course, been neglected. The result (3.35), for p = 1, , N 1, shows
that the N 1 photons of the unbroken U (1)N 1 gauge group receive non-zero masses due
to nonperturbative effects.15

3.2 String tensions

Let us first examine the vacuum structure of the dual theory in more detail. The dual scalars
are defined to be periodic with period 2. This implies that shifting by 2 times any root
vector is an identity, + 2i for all i 0aff . As noted earlier, the dual action (3.25)
is also invariant under + 2i , where {i } are the fundamental weights of the SU (N )
gauge group, defined by the reciprocity relations (3.27). The simple roots {i } generate the
root lattice r . Its dual, the weight lattice w is generated by the fundamental weights {i }.
The root lattice is a sublattice of the weight lattice and their quotient is

w /r = ZN . (3.38)
P
This implies that the dual theory potential, V () i cos(i ), has N isolated minima
lying within the unit cell of r . These minima are located at = 0 and

= 2j , j = 1, , N 1 . (3.39)

(Equivalently, one may describe the minima as lying at 2 j 1 , for j = 0, 1, , N , since

j = j 1 + for some r .)
15
The vanishing p = 0 mass corresponds to the extra decoupled photon which was added to simplify the
duality transformation but is not present in the original theory. It should be ignored.

20
 Let R be some chosen irreducible representation of SU (N ). The expectation value of
the Wilson loop WR (C) characterizes the response of the system to external test charges in
the representation R. In a confining phase with a non-zero mass gap, if external charges in
representation R cannot be screened by gluons, then expectation values of large Wilson loops
in this representation are expected to decrease exponentially with the area of the minimal
spanning surface,
WR (C) eT (R) Area() .


(3.40)
Here denotes the minimal surface with boundary C, and T (R) is the string tension for
representation R. Such area law behavior implies the presence of an asymptotically linear
confining potential between static charges in representation R and anti-charges in represen-
tation R, VR (x) T (R) |x| as |x| .
The irreducible representation R may be associated with its highest weight vector w
w . Identifying weight vectors which differ by elements of the root lattice produces a ZN
grading of representations which corresponds to their N -ality (the charge of the representation
under the ZN center). In particular, if this equivalence associates the highest weight vector
w with k times the fundamental weight 1 ,

w = k 1 + , for some r , (3.41)

then the representation R has N -ality k.16

As discussed in Refs. [27, 22], the insertion of a Wilson loop WR (C) in a representation
R with non-zero N -ality k corresponds, in the low-energy dual theory, to the requirement
that the dual scalar fields have non-trivial monodromy,
Z
d = 2k , (3.42)
C0

where C 0 is any closed curve whose linking number with C is one. In other words, in the
presence of the Wilson loop WR (C) the dual scalar fields must have a discontinuity of 2k
across some surface which spans the loop C. One way to see this is to go back to the
duality relation. For simplicity, consider the case of a large planar loop lying in the xy-
plane. As the size of the loop grows, the spanning surface approaches an infinite flat plane.
In the presence of the Wilson loop, the Abelian duality relation F d is replaced by
F d + k (z) dx dy. Therefore the dual scalars must be discontinuous across in
order for the field strength F to be continuous.
The fact that dual low energy theory depends on the representation R of the Wilson loop
only through its N -ality k shows that there are only N 1 distinct string tensions, referred
to as k-string tensions, {Tk }. (Charge conjugation symmetry implies that Tk = TN k .) The
dual theory representation of Wilson loops also shows that external charges in representations
with zero N -ality will not be confined. These are precisely the representations which can be
screened by adjoint representation gluons.

16
Representations contained in the product of m powers of the fundamental representation with n powers
of the antifundamental have N -ality mn.

21
 To evaluate a Wilson loop expectation value, one must minimize the dual action in the
space of field configurations satisfying the monodromy condition (3.42). To extract the string
tension,
ln hWR (C)i
Tk lim , (3.43)
area() area()
it is sufficient to consider the limit where fills the xy-plane. In this case, the field (x)
will only depend on z. It must approach some minimum of the dual potential at infinity,
limz (z) = 2l , and must be discontinuous across z = 0 with a jump given by the
prescribed fundamental weight, limz0+ (z) limz0 (z) = 2k (mod 2). Because
shifts by 2k are an invariance of the dual potential V (), one may equally well minimize
the action for field configurations (z) which are continuous but whose asymptotic values
differ, 
S() 
Tk = min , (3.44)
(z) area(R2 ) =2 (mod 2)
k

where ()(), and S() is the dual action minus its vacuum value. Explicitly,

Z n 1  g 2  2 X o
Tk = min dz + [1 cos(i i+1 )]  . (3.45)

(z) 2L 2 z 
i =2k (mod 2)

In other words, the k-string tension Tk equals the action of a kink solution with topological
charge k in this one dimensional theory.
The width of the kink solution must be of order of the inverse photon mass m1 . Con-
sequently, the k-string tension will have the form Tk = fk T , where

T /m 2 (LN )1/6 |log(LN )|3/11 , (3.46)

and fk is an O(1) coefficient. Even without finding the minimizing kink solutions explicitly,
it is apparent that the resulting k-string tension Tk will be non-zero (for k = 1, , N 1),
and must satisfy the convexity relation Tk+l Tk + Tl .
We were unable to solve the kink equations of motions analytically for general N , but
when N = 2 the equations of motion reduce to Sine-Gordon model. In this case, one finds

T T1 = 4 2 /m . (3.47)

3.3 Larger size or larger N

The above semiclassical analysis of the deformed Yang-Mills theory is reliable provided there
is a parametrically large separation of scales between the lightest W -boson mass, mW =
2/(N L), and the nonperturbatively induced dual photon mass m . Their ratio scales as
mW
(LN )11/6 | log(LN )|9/11 , (3.48)
m

and hence there is a large separation of mass scales provided LN  1. In this regime, the
monopole gas is highly dilute and a semiclassical analysis is justified. Increasing LN , by

22
increasing N , L, or both, decreases the separation of scales; the heaviest photon mass, m ,
grows while the lightest W mass, mW , drops. When LN 1, the scale separation is entirely
lost, the effective t Hooft coupling g 2 N at the scale of mW ceases to be small, and the
long distance dynamics can no longer be described by a weakly coupled U (1)N 1 effective
theory.
One can consider sending N to infinity while staying within the analytically tractable
regime. This is a double scaling limit in which g 2 N and LN are both are held fixed (and
both are much less than unity) as N . Taking a large N limit in this fashion allows
monopole effects to survive and to continue dictating the nonperturbative physics of the
deformed Yang-Mills theory. However, this region shrinks to a vanishingly small window in
the large N limit, since the double scaling implies that 0 < L  Lmax with Lmax 1/N .
For any fixed compactification size L, if one sends N the deformed YM theory ceases
to possess a monopole dominated, Abelian long distance regime.17

3.4 Connection to integrable Toda theory

The ZN symmetric model (3.25) is a deformation of a complex affine Toda theory with action
Z h 1  g 2 N
X i
S affine Toda = ()2 ei(i i+1 ) . (3.49)
R3 2L 2
i=1

This complex (CPT-noninvariant) action describes a plasma which is composed solely of

monopoles with no antimonopoles. (Due to the existence of the affine root, one can have
a neutral plasma composed solely of monopoles!) Interestingly, the soliton spectrum of the
affine Toda theory is exactly computable. When reduced to one dimension, this theory is an
integrable system as shown by Hollowood [40], using techniques due to Hirota [41, 42].
As discussed above, the k-string tension Tk is equal to the action of the kink solution
with topological charge k. Borrowing the exact soliton spectrum from Ref. [40], one finds
that the k-string tensions in the affine Toda theory are given by
k
Tkaffine Toda = T N sin , k = 1, N 1 , (3.50)
N
with T given above in Eq. (3.46).
The long distance effective theory (3.25) for our deformed Yang-Mills theory (when
LN  1) is a deformation of the affine Toda system by complex conjugation. Unfortu-
nately, unlike the integrable affine Toda system, when N > 2 the resulting CPT invariant
17
An analog of this double scaled limit was previously discussed by Douglas and Shenker in mass deformed
SU (N ) N = 2 supersymmetric Yang-Mills theory on R4 [26]. This deformed N = 1 supersymmetric theory,
just like our deformed Yang-Mills theory, possess a regime in which the long distance gauge dynamics reduces
to the Abelian subgroup U (1)N 1 . Ref. [26] shows that in the N limit of the mass deformed theory,
the Abelian long distance regime is preserved only if the mass deformation m is sent to zero in a correlated
fashion, m/ 1/N 4 . In particular, at any fixed non-zero m, if one takes N first then there is no
regime of the supersymmetric gauge theory in which the long distance dynamics remains Abelian. Although
this phenomena only appeared previously in the context of supersymmetric gauge theories, it is generic in
deformed Yang-Mills and other deformed QCD-like theories.

23
system is no longer exactly integrable according to Hirotas criteria.18 Consequently, we do
not expect k-string tensions in the deformed Yang-Mills theory to have the sine-law form
(3.50).
Recently, there have been attempts [43], to model the strongly coupled confined regime
of Yang-Mills theory assuming the Wilson line has the center-symmetric form (3.1). (See also
earlier related work in Refs. [44, 45, 46, 39].) A few remarks concerning the connection with
Ref. [43] may be in order. First, our results for the k-string tensions do not support the claim
of Ref. [43], which asserts that k-string tensions will have the sine law form (3.50). As just
noted, sine-law string tensions are a property of the affine Toda subsystem, whereas the center-
stabilized Yang-Mills in a weak coupling regime is dual to a real deformation of the affine
Toda theory. We see no reason to believe that the k-dependence of the string tensions will be
unaffected by the deformation. Secondly, it should be emphasized that the deformation (2.5)
stabilizes the center symmetric vacuum in the weakly coupled regime, and thereby provides
a window in which a semiclassical analysis is reliable. Many earlier discussions of center
symmetric backgrounds do not clearly distinguish the weakly coupled Higgs regime, in
which fluctuations of the Wilson line eigenvalues are small, from the strong coupling regime
in which the eigenvalues have large fluctuations and are essentially randomized over the unit
circle. In our deformed Yang-Mills theory, both regimes exist. As the compactification size L
increases, the theory moves from the weakly coupled regime to the strongly coupled regime.
These two regimes are expected to be smoothly connected no physical order parameter
sharply distinguishes the two regimes. Nevertheless, the long distance physics of the weak
coupling Higgs regime is effectively Abelian, while in the strong coupling regime there is no
length scale beyond which the dynamics can be described accurately in terms of Abelian
degrees of freedom.

Acknowledgments

We thank Barak Bringoltz and Steve Sharpe for calling our attention to the issue of sponta-
neous breaking to diagonal subgroups when there are multiple compactified directions. This
work was supported in part by the U.S. Department of Energy under Grant Nos. DE-AC02-
76SF00515 and DE-FG02-96ER40956, and by the U.S. National Science Foundation under
Grant No. PHY05-51164. L.Y. thanks the Kavli Institute for Theoretical Physics for its
hospitality during the completion of this paper.

18
In the absence of the complex conjugate term in the potential, there is a change of variables which converts
the soliton equation of motion into Hirota bilinear type, which is synonymous with solvability [40]. The
presence of the complex conjugate term spoils the bi-linearity.

24
A. Monopole measure

The appropriate one-loop measure for integrating over configurations of a single monopole
(of any type) may be expressed as19

d3 a d 1
dmonopole = 4 e(SYM +S) J det0 (Dadj
2

3/2
Ja 1/2
) , (A.1)
(2) (2)

where a R3 is the monopole position, [, ] is the internal U (1) angle of the monopole,
and is the (Pauli-Villars) renormalization scale. Global U (1) gauge transformations (in the
U (1) subgroup associated with the given type of monopole) shift the angle . Fluctuations in
the position and U (1) angle of the monopole represent the four zero modes in the monopole
small fluctuation operator;20 the factor of 4 can be viewed as the contributions of the Pauli-
Villars regulator fields associated with these bosonic zero modes. The exponential factor
is, of course, the exponential of minus the classical action of the monopole. The collective
coordinate Jacobians are given by [39]
1/2
3/2 2 SYM 1/2
Ja = SYM , J = = N L SYM , (A.2)
j

where = {0, 2, 4, , 2(N 1)}/(N L) are the eigenvalues of i ln . The primed deter-
minant represents the result of Gaussian integrals over all fluctuations other than zero modes;
the prime on the determinant denotes omission of the zero modes. The contributions from
gauge bosons and ghosts,
 0 1/2
det (D2 2F )adj det(D2 )adj , (A.3)
| {z } | {z }
gauge bosons ghosts

combine to give this simple form because

0
[det (D2 2F )adj ]1/2 = [det(D2 )adj ]2 (A.4)

in any self-dual background. These functional determinants may be regularized using the
Pauli-Villars scheme.
The fields of fundamental monopoles reside entirely within an SU (2) subgroup of SU (N ),
and the characteristic size of these monopoles is given by the inverse of the lightest W -
boson mass, m1 W N L. (This is the only scale which appears in the classical equations

19
The following summary is an adaptation of the appendix of Ref. [39], which treats the monopole measure
in supersymmetric Yang-Mills theory.
20
For comparison, recall that an instanton in SU (N ) Yang-Mills theory has 4N zero-modes. (See, for
example, Ref. [47].) For SU (2) gauge theory, these are the four zero modes corresponding to changes in the
instanton position, one for its size, and three global gauge rotations. For SU (N ) with N > 2, there are
in addition 4N 8 zero modes associated with changes in the embedding of SU (2) within SU (N ). When
compactified in one dimension, with non-trivial holonomy , one may regard an instanton as being composed
of N independent monopole constituents [45, 44], each of which carries four zero-modes.

25
for the monopole.) The regularized scalar determinant depends on the cube root of the
renormalization scale, det(D2 ) 1/3 . Since the determinant is dimensionless, it must
have the form
[det(D2 )]1 = 2C (N L)1/3 , (A.5)
where C is a pure number (N -independent). Consequently, the one-loop monopole measure
equals
2
dmonopole = C11/3 (N L) 3 (SYM )2 eSYM +S d3 a d . (A.6)
Performing the trivial integral over the angle , the result is the d3 a, with the monopole
fugacity. Choosing to use mW as the value of the renormalization point yields the expression
(3.23) for the fugacity.

References
[1] T. Eguchi and H. Kawai, Reduction of dynamical degrees of freedom in the large N gauge
theory, Phys. Rev. Lett. 48 (1982) 1063.
[2] L. G. Yaffe, Large N limits as classical mechanics, Rev. Mod. Phys. 54 (1982) 407.
[3] G. Bhanot, U. M. Heller and H. Neuberger, The quenched Eguchi-Kawai model, Phys. Lett. B
113 (1982) 47.
[4] R. Narayanan and H. Neuberger, Large N reduction in continuum, Phys. Rev. Lett. 91 (2003)
081601, hep-lat/0303023.
[5] J. Kiskis, R. Narayanan and H. Neuberger, Does the crossover from perturbative to
nonperturbative physics in QCD become a phase transition at infinite N ?, Phys. Lett. B 574
(2003) 65, hep-lat/0308033.
[6] B. Lucini, M. Teper and U. Wenger, Properties of the deconfining phase transition in SU(N)
gauge theories, J. High Energy Phys. 0502 (2005) 033, hep-lat/0502003.
[7] B. Bringoltz and M. Teper, In search of a Hagedorn transition in SU (N ) lattice gauge theories
at large-N , Phys. Rev. D 73 (2006) 014517, hep-lat/0508021.
[8] T. D. Cohen, Large Nc continuum reduction and the thermodynamics of QCD, Phys. Rev. Lett.
93 (2004) 201601, hep-ph/0407306.
[9] Y. Makeenko, Methods of contemporary gauge theory, Cambridge, 2002.
[10] A. Gonzalez-Arroyo and M. Okawa, The twisted Eguchi-Kawai model: A reduced model for large
N lattice gauge theory, Phys. Rev. D 27 (1983) 2397.
[11] A. Gonzalez-Arroyo and M. Okawa, A twisted model for large N lattice gauge theory, Phys. Lett.
B 120 (1983) 174.
[12] M. Teper and H. Vairinhos, Symmetry breaking In twisted Eguchi-Kawai models, Phys. Lett. B
652 (2007) 359 , hep-th/0612097.
[13] T. Azeyanagi, M. Hanada, T. Hirata and T. Ishikawa, Phase structure of twisted Eguchi-Kawai
model arXiv:0711.1925 [hep-lat].
[14] P. Kovtun, M. Unsal, and L. G. Yaffe, Volume independence in large Nc QCD-like gauge
theories, J. High Energy Phys. 0706 (2007) 019, hep-th/0702021.

26
[15] M. Unsal and L. G. Yaffe, (In)validity of large N orientifold equivalence, Phys. Rev. D 74
(2006) 105019, hep-th/0608180.
[16] A. Armoni, M. Shifman and G. Veneziano, SUSY relics in one-flavor QCD from a new 1/N
expansion, Phys. Rev. Lett. 91 (2003) 191601, hep-th/0307097.
[17] A. Armoni, M. Shifman and G. Veneziano, Refining the proof of planar equivalence, Phys. Rev.
D 71 (2005) 045015, hep-th/0412203.
[18] M. Schaden, A center-symmetric 1/N expansion, Phys. Rev. D 71 (2005) 105012,
hep-th/0410254.
[19] R. D. Pisarski, Effective theory of Wilson lines and deconfinement, Phys. Rev. D 74 (2006)
121703, hep-ph/0608242.
[20] J. C. Myers and M. C. Ogilvie, New phases of SU (3) and SU (4) at finite temperature,
arXiv:0707.1869 [hep-lat].
[21] M. Shifman and M. Unsal, QCD-like theories on R3 S1 : a smooth journey from small to large
r(S1 ) with double-trace deformations, arXiv:0802.1232 [hep-th].
[22] A. M. Polyakov, Quark confinement and topology of gauge groups, Nucl. Phys. B 120 (1977)
429458.
[23] A. R. Zhitnitsky, Confinement-deconfinement phase transition and fractional instanton quarks
in dense matter, hep-ph/0601057.
[24] D. Toublan and A. R. Zhitnitsky, Confinement-deconfinement phase transition at nonzero
chemical potential, Phys. Rev. D 73 (2006) 034009, hep-ph/0503256.
[25] N. Seiberg and E. Witten, Electric-magnetic duality, monopole condensation, and confinement
in N = 2 supersymmetric Yang-Mills theory, Nucl. Phys. B 426 (1994) 19 [Erratum-ibid. B
430 (1994) 485 ], hep-th/9407087.
[26] M. R. Douglas and S. H. Shenker, Dynamics of SU (N ) supersymmetric gauge theory, Nucl.
Phys. B 447 (1995) 271, hep-th/9503163.
[27] P. Deligne, P. Etingof, D. Freed, L. Jeffrey, D. Kazhdan, J. Morgan, D. Morrison, E. Witten,
eds., Quantum fields and strings: A course for mathematicians. vol. 1, 2, Providence, USA:
AMS (1999) 11501.
[28] P. Kovtun, M. Unsal and L. G. Yaffe, Non-perturbative equivalences among large N gauge
theories with adjoint and bifundamental matter fields, J. High Energy Phys. 0312 (2003) 034,
hep-th/0311098.
[29] P. Kovtun, M. Unsal and L. G. Yaffe, Necessary and sufficient conditions for non-perturbative
equivalences of large N orbifold gauge theories, J. High Energy Phys. 0507 (2005) 008,
hep-th/0411177.
[30] F. R. Brown and L. G. Yaffe, The Coherent State Variational Algorithm: A numerical method
for solving large N gauge theories, Nucl. Phys. B 271 (1986) 267.
[31] T. A. Dickens, U. J. Lindqwister, W. R. Somsky and L. G. Yaffe, The Coherent State
Variational Algorithm. 2. Implementation and testing, Nucl. Phys. B 309 (1988) 1.
[32] D. J. Gross, R. D. Pisarski, and L. G. Yaffe, QCD and instantons at finite temperature, Rev.
Mod. Phys. 53 (1981) 43.

27
[33] M. Luscher, Some analytic results concerning the mass spectrum of Yang-Mills gauge theories
on a torus, Nucl. Phys. B 219 (1983) 233.
[34] P. van Baal, The small volume expansion of gauge theories coupled to massless fermions, Nucl.
Phys. B 307 (1988) 274 [Erratum-ibid. B 312, 752 (1989)].
[35] M. Luscher, Advanced lattice QCD, hep-lat/9802029.
[36] P. van Baal, QCD in a finite volume, hep-ph/0008206.
[37] Y. M. Makeenko and A. A. Migdal, Exact equation for the loop average in multicolor QCD,
Phys. Lett. B 88 (1979) 135 [Erratum-ibid. B 89 (1980) 437].
[38] M. Unsal, in preparation, 2008.
[39] N. M. Davies, T. J. Hollowood, and V. V. Khoze, Monopoles, affine algebras and the gluino
condensate, J. Math. Phys. 44 (2003) 36403656, hep-th/0006011.
[40] T. J. Hollowood, Solitons in affine Toda field theories, Nucl. Phys. B 384 (1992) 523.
[41] R. Hirota and J. Satsuma, A variety of nonlinear network equations generated from the
Backlund transformation for the Toda lattice, Prog. Theor. Phys. Suppl. 59 (1976) 64100.
[42] M. Toda, Nonlinear waves and solitons, Springer, 1989.
[43] D. Diakonov and V. Petrov, Confining ensemble of dyons, Phys. Rev. D 76 (2007) 056001,
arXiv:0704.3181 [hep-th].
[44] K. M. Lee and P. Yi, Monopoles and instantons on partially compactified D-branes, Phys. Rev.
D 56 (1997) 3711, hep-th/9702107.
[45] T. C. Kraan and P. van Baal, Periodic instantons with non-trivial holonomy, Nucl. Phys. B
533 (1998) 627, hep-th/9805168.
[46] F. Bruckmann, D. Nogradi and P. van Baal, Higher charge calorons with non-trivial holonomy,
Nucl. Phys. B 698 (2004) 233, hep-th/0404210.
[47] E. J. Weinberg and P. Yi, Magnetic monopole dynamics, supersymmetry, and duality, Phys.
Rept. 438 (2007) 65, hep-th/0609055.

28