hep-th/9407087, RU-94-52, IAS-94-43

Electric-Magnetic Duality,
Monopole Condensation, And Confinement
arXiv:hep-th/9407087v1 15 Jul 1994

In N = 2 Supersymmetric Yang-Mills Theory

N. Seiberg
Department of Physics and Astronomy
Rutgers University, Piscataway, NJ 08855-0849, USA
and
Institute for Advanced Study
Princeton, NJ 08540, USA
and
E. Witten
Institute for Advanced Study
Princeton, NJ 08540, USA

We study the vacuum structure and dyon spectrum of N = 2 supersymmetric gauge theory
in four dimensions, with gauge group SU (2). The theory turns out to have remarkably rich
and physical properties which can nonetheless be described precisely; exact formulas can
be obtained, for instance, for electron and dyon masses and the metric on the moduli space
of vacua. The description involves a version of Olive-Montonen electric-magnetic duality.
The strongly coupled vacuum turns out to be a weakly coupled theory of monopoles,
and with a suitable perturbation confinement is described by monopole condensation.

6/94
1. Introduction

The dynamics of N = 1 supersymmetric gauge theories in four dimensions have been

much explored, in part because of the possible phenomenological interest. Recently results
have emerged about their strong coupling behavior [1-4]. Such results are largely conse-
quences of the fact that the low energy effective superpotential is holomorphic both in the
chiral superfields and in the parameters of the fundamental Lagrangian [1] (other aspects
of such holomorphy were discussed in [5,6]). When the behavior of the superpotential
near its singularities is combined with the symmetries and its holomorphy, the superpo-
tential can often be determined exactly. The resulting superpotentials sometimes exhibit
new physical phenomena [2-4] which might have applications also to non-supersymmetric
gauge theories.
Our goal in the present paper is, among other things, to obtain similar strong coupling
information about the corresponding N = 2 theories, at least in the basic case of SU (2)
gauge theory without matter multiplets. (Such multiplets will be incorporated in a sub-
sequent paper [7].) In N = 2 theories, the Kahler potential and the masses of the stable
particles are controlled by a holomorphic object, the prepotential. Therefore, they can
be studied in a way somewhat similar to the determination of the superpotential in N = 1
theories and ultimately determined.
Although our results are derived in the special case of N = 2 supersymmetry, they
exhibit physical phenomena of general interest, including asymptotic freedom, chiral sym-
metry breaking, generation of a mass scale from strong coupling, confinement of electric
charge via condensation of magnetic monopoles, and a version of electric-magnetic duality.
These results are also likely to shed some light on the phenomena in N = 4 and perhaps
in string theory, and are likely to help in understanding the topological field theories that
can be obtained from N = 2 theories by twisting [8,9].
As electric-magnetic duality will play an important role in solving the model, we will
review some of the background here. The existence of such duality in supersymmetric
Yang-Mills theories was first suspected from properties of the dyons particles carrying
electric and magnetic charge that exist at the classical level in the N = 2 and N = 4
gauge theories. Semiclassically one finds by an argument that originated with the work

1
of Prasad and Sommerfeld [10] and Bogomolnyi [11] that the mass of a dyon of magnetic
and electric quantum numbers (nm , ne ) is

M 2|Z| (1.1)

with  
1
Z = v ne + i nm . (1.2)

Here v is the Higgs expectation value, and = g 2 /4 with g the gauge coupling constant.1
States for which the inequality in (1.1) is actually an equality are said to be BPS-saturated.
To see that a similar inequality must hold quantum mechanically, one interprets the in-
equality as a consequence of a central extension of the supersymmetry algebra [12]: we will
see later that the parameters in the central extension (and thus the parameters appearing
in the definition of Z) have an interesting renormalization for N = 2 but not for N = 4.
The mass formula (1.2) has a symmetry under ne nm , 1/, v v/.
Olive and Montonen [13] pointed out this symmetry and conjectured that it was an exact
property of a suitable quantum theory. According to this dramatic conjecture, the strong
coupling limit of the theory is equivalent to the weak coupling limit with ordinary particles
and solitons exchanged. The N = 2 theory appears not to possess Olive-Montonen duality
because the electrons and monopoles have different Lorentz quantum numbers (electrons
are in a supersymmetric multiplet with spins 1, while the monopoles have spins 1/2).
On the other hand [14], for N = 4 the electrons and monopoles have the same quantum
numbers, making Olive-Montonen duality more plausible. Actually, in this paper we will
find that a version of Olive-Montonen duality holds for N = 2.
The version in question is necessarily quite different from the one that seems to hold
for N = 4. The N = 2 theory is asymptotically free, so the coupling constant is equivalent
to a choice of scale. And an anomaly in the U (1)R R symmetry prevents the existence of
a physical microscopic theta angle. (There is still an effective theta angle at low energies.)
For N = 4, the anomalies cancel in both conformal invariance and the U (1) symmetry, so
the theory possesses both a natural dimensionless gauge coupling g and a theta angle ;

1 1 1
We normalize the Higgs field such that its kinetic term is multiplied by 4
= g2
.

2

the Olive-Montonen duality should be extended to an action of SL(2, Z) on = 2 + i 4
g2 ,

as was first recognized in lattice models [15,16] and in string theory [17]. (Dramatic new
evidence for this has appeared recently in considerations of multi-monopole bound states
[18,19] and the partition function on various manifolds [20,21].) For N = 2, these natural
dimensionless parameters are absent, and duality as we interpret it has to do with the
behavior of the theory as a function of the expectation value of the Higgs field.
The organization of the paper is as follows. Section 2 is a review of some known
facts about N = 2 supersymmetric gauge theories. The classical theory has a continuous
manifold of inequivalent ground states the classical moduli space. Quantum correc-
tions do not lift the vacuum degeneracy, so also the quantum theory has a manifold of
inequivalent ground states a quantum moduli space. We construct a low energy effec-
tive Lagrangian for the light degrees of freedom. As described in [22], the metric on the
quantum moduli space is a Kahler metric that is written locally in terms of a holomorphic
function. In certain N = 1 models [2], the quantum corrections change the topology of the
moduli space. For N = 2 we will argue that there is no change in topology but a marked
change in geometry, singularity structure, and physical interpretation.
In section 3, we analyze more fully the geometry of the low energy Lagrangian. The
local structure is unique only up to transformations of a certain kind (flat space limits
of the special geometry transformations [23-26] that appear in certain supergravity and
string theories). Physically these correspond to duality transformations on the low energy
fields. The low energy Lagrangian is mapped to another Lagrangian of the dual fields.
In section 4 we use the coupling of the light fields to the massive dyons to find an ex-
pression for the dyon masses. It includes the quantum corrections to the classical result and
is manifestly dual. We also (following a previous two-dimensional analysis [27]) describe
conditions under which the spectrum of BPS-saturated states (that is, states obeying the
BPS mass formula (1.2)) does not vary continuously. The possibility of this phenomenon
turns out to be an essential difference between N = 2 and N = 4.
In sections 5 and 6 we make our proposal for solving the model. We begin section 5
by explaining that to get a sensible Kahler metric on the quantum moduli space, at least
two singularities are needed in strong coupling. We propose that these singularities arise

3
when a magnetic monopole or dyon goes to zero mass. As a check we show that, under
a further perturbation, condensation of monopoles occurs precisely when confinement of
electric charge is expected. This for the first time gives a real relativistic field theory model
in which confinement of charge is explained in this long-suspected fashion. We also show
that the monodromies resulting from massless monopoles and dyons fit together in just the
right way. Then in section 6 we show that, with the assumption that the singularities come
from massless monopoles and dyons, it is possible to get a unique metric on the moduli
space (and unique formulas for particle masses) obeying all the necessary conditions.

2. Review of N=2 SUSY

2.1. Representations

All N = 2 theories have a global SU (2)R symmetry which acts on the two supercharges
of given chirality. Scale invariant N = 2 theories have also a U (1)R symmetry under which
the supercharges of positive chirality have charge minus one.
We will be studying two types of N = 2 multiplet:

1. The first is the N = 2 chiral multiplet (sometimes called a vector multiplet), con-
taining gauge fields A , two Weyl fermions , , and a scalar , all in the adjoint
representation. We arrange the fields as

(2.1)

to exhibit the SU (2)R symmetry which acts on the rows; A and are singlets and
, are a doublet. In terms of N = 1 supersymmetry, these fields can be organized
into a vector multiplet W (containing (A , )) and a chiral multiplet (containing
(, )). In this formalism, only one generator of SU (2)R , which we will call U (1)J , is
manifest. U (1)J and U (1)R are both N = 1 R symmetries, acting as

U (1)J : (ei )
(2.2)
2i i
U (1)R : e (e ) .

4
 2. The second type of multiplet is the hypermultiplet (sometimes called the scalar mul-
tiplet), consisting of two Weyl fermions q and and complex bosons q and qe ;
e
q
SU (2)R again acts on the rows of the diamond:

q qe . (2.3)

e
q

In terms of N = 1 supersymmetry, these fields make up two chiral multiplets Q and

e The symmetries in (2.2) act on them as
Q.

U (1)J : Q ei Q(ei )
e ei Q(e
Q e i )
(2.4)
U (1)R : Q Q(ei )
e Q(e
Q e i )

e are dual to those of Q.

The gauge quantum numbers of Q

2.2. Renormalizable Lagrangians and Classical Flat Directions

We will consider an N = 2 gauge theory, described by chiral superfields in the adjoint

representation of a gauge group G, possibly coupled to additional matter hypermultiplets.
N = 2 supersymmetry relates the gauge couplings to certain interactions which in the
N = 1 language are described by the superpotential. In fact the requisite term in the
superpotential is (in the notation of [28])

W = e
2QQ. (2.5)

e and Q transform in dual representations and is in the adjoint

This makes sense since Q
representation.
Classically, the Lagrangian is invariant under the SU (2)R U (1)R symmetry (2.2),
(2.4) as well as, possibly, some flavor symmetries depending on the gauge representations
of the hypermultiplets. In the quantum theory, the U (1)R symmetry is generally broken
by an anomaly. For SU (Nc ) gauge symmetry with Nf massless hypermultiplets (quarks)

5
in the fundamental representation, U (1)R is broken to Z4Nc 2Nf . In this paper, we will
mainly consider the SU (2) gauge theory without quarks. In this case the global symmetry
is (SU (2)R Z8 )/Z2 since 4Nc 2Nf = 8 (the division by Z2 arises because the center of
SU (2)R is contained in Z8 ).
The classical potential of the pure N = 2 theory (without hypermultiplets) is

1
V () = 2
Tr[, ]2 . (2.6)
g

For this to vanish, it is not necessary that should vanish; it is enough that and
commute. The classical theory therefore has a family of vacuum states. For instance,
1
if the gauge group is SU (2), then up to gauge transformation we can take = 2
a 3 ,
with 3 = diag(1, 1) and a a complex parameter labeling the vacua. The Weyl group of
SU (2) acts by a a, so the gauge invariant quantity parametrizing the space of vacua is
u = 12 a2 = Tr 2 . For non-zero a the gauge symmetry is broken to U (1) and the global Z8
symmetry is broken to Z4 . The residual Z4 acts trivially on the u plane since the U (1)R
charge of u is 4. The global symmetry group acts on the u plane as a spontaneously broken
Z2 , acting by u u.
Classically, there is a singularity at u = 0, where the full SU (2) gauge symmetry is
restored and more fields become massless.

2.3. Low Energy Effective Action

We now study the low energy effective action of the light fields on the moduli space.
For generic hi the low energy effective Lagrangian contains a single N = 2 vector multiplet,
A. The terms with at most two derivatives and not more than four fermions are constrained
by N = 2 supersymmetry. They are expressed in terms of a single holomorphic function
F (A), as explained in [22]. In N = 1 superspace, the Lagrangian is
Z Z 2

1 4 F (A) 2 1 F (A)
Im d A+ d W W (2.7)
4 A 2 A2

where A is the N = 1 chiral multiplet in the N = 2 vector multiplet A whose scalar

component is a.
We would like to make a few comments:

6
1. For large a, asymptotic freedom takes over and the theory is weakly coupled. More-
over, since it is impossible to add an N = 2 invariant superpotential to (2.7), the
vacuum degeneracy cannot be removed quantum mechanically. Therefore, the quan-
tum theory has a non-trivial moduli space which is in fact a one complex dimensional
Kahler manifold. The Kahler potential can be written in terms of the effective low
energy F function as  
F (A)
K = Im A . (2.8)
A
The metric is thus concretely

2 F (a)
(ds)2 = Im da da. (2.9)
a2

In the classical theory, F can be read off from the tree level Lagrangian of the SU (2)
gauge theory and is F (A) = 21 cl A2 with cl =
2
+ i 4
g2
. Asymptotic freedom means
that this formula is valid for large a if g 2 is replaced by a suitable effective coupling.
The small a behavior will however turn out to be completely different. Classically, the
parameter has no consequences. Quantum mechanically, the physics is dependent,
but since there is an anomalous symmetry, it can be absorbed in a redefinition of the
fields. Therefore, we will set = 0.

2. The formula for the Kahler potential does not look covariant the Kahler potential
can be written in this way only in a distinguished class of coordinate systems, which
we will analyze later. In fact, A is related by N = 2 supersymmetry to the photon
A , which has a natural linear structure; this gives a natural coordinate system (or
what will turn out to be a natural class of coordinate systems) for A.

3. The low energy values of the gauge coupling constant and theta parameter can be

read off from the Lagrangian. If we combine them in the form = 2 + i 4
g 2 , and

denote the effective couplings in the vacuum parametrized by a as (a), then

2F
(a) = . (2.10)
a2

4. The generalization to an arbitrary compact gauge group G of rank r is as follows.

The potential is always given by (2.6), so the classical vacua are labeled by a complex

7
 adjoint-valued matrix with [, ] = 0. The unbroken gauge symmetry at the
generic point on the moduli space is the Cartan subalgebra and therefore the complex
dimension of the moduli space is r. The low energy theory is described in terms of r
abelian chiral multiplets Ai , and the generalization of (2.7) is [22]
Z Z 
1 F (A) i
4 2 1 2 F (A) i j
Im d A + d W W . (2.11)
4 Ai 2 Ai Aj

Here i labels the generators in the Cartan subalgebra and locally F is an arbitrary
holomorphic function of r complex variables.

5. The SU (2) theory, studied on the flat direction with u 6= 0, has in addition to the
massless chiral or vector multiplet A, additional charged massive vector multiplets.
One can easily write a gauge invariant effective action for the triplet of chiral mul-
tiplets Aa , a = 1 . . . 3, which reduces at low energies to (2.7) for the massless fields
and incorporates the massive ones. Using the same function F as above, we set

F ( A A) = H(A A) and write
Z Z 
1 4 a V

b 1

Im
d H A e ab
A + d H ab + 2H Aa Ab Wa Wb
2
(2.12)
2 2

where we used the SU (2)-invariant metric ab to raise and lower indices. (2.12) has
N = 2 supersymmetry and manifest gauge invariance, and reduces at low energies to
(2.7).

6. The Lagrangian (2.7) is unchanged if we add to F terms linear in A. This has the
effect of shifting F /A by a constant. We will later assign physical meaning to

F
h(A) = . (2.13)
A

The additive constant will always be fixed by comparing with the high energy theory
as in (2.12).
As we have already mentioned, classically the F function is

1
F0 = cl A2 . (2.14)
2
8
The quantum corrections were analyzed in [22]. The tree level and one loop contributions
add up to
1 2 A2
Fone
loop =i A ln 2 (2.15)
2
where is the dynamically generated scale. This logarithm is related to the one loop beta
function and also ensures the anomalous transformation laws under U (1)R . Higher order
perturbative corrections are absent. Instantons lead to new terms. The anomaly and the
instanton action suggest that
 4k
1 A2 X
F = i A2 ln 2 + Fk A2 (2.16)
2 A
k=1
where the kth term arises as a contribution of k instantons. A detailed calculation of
the k = 1 term [22] indicates that F1 6= 0. We will soon see that infinitely many Fk are
nonzero.
Corrections to the classical formula (2.14) are related to the beta function, and for
N = 4 supersymmetric Yang-Mills theory, whose beta function vanishes, the formula (2.14)
is exact.

3. Duality

We have noted above that locally, by virtue of N = 2 supersymmetry, the metric on

the moduli space is of the form

(ds)2 = Im (a) da da, (3.1)

with (a) the holomorphic function = 2 F /a2 . The one loop formula (2.15) shows that


for large |a|, (a) i ln(a2 /2 ) + 3 / is a multivalued function whose imaginary part
is single-valued and positive. However, if Im (a) is globally defined it cannot be positive
definite as the harmonic function Im cannot have a minimum. This indicates that the
above description of the metric must be valid only locally.
To what extent is it possible to change variables from a to some other local parameter,
while leaving the metric in the form (3.1)? The answer to this question is at the heart of
the physics. We define aD = F /a. The metric can then be written
i
(ds)2 = Im daD da = (daD da dadaD ) . (3.2)
2
9
This formula is completely symmetric in a and aD , so if we use aD as the local parameter,
the metric will be in the same general form as (3.1), with a different harmonic function
replacing Im . As we will see presently, this transformation corresponds to electric-
magnetic duality. Before entering into that, let us identify the complete class of local
parameters in which the metric can be written as in (3.2).

3.1. Mathematical Description

To treat the formalism in a way that is completely symmetric between a and aD , we

introduce an arbitrary local holomorphic coordinate u, and treat a and aD as functions of
u. u is a local coordinate on a complex manifold M the moduli space of vacua of the
theory. Eventually we will pick u to be the expectation value of Tr 2 a good physical
parameter but for now u is arbitrary.
Introduce a two dimensional complex space X
= C2 with coordinates (aD , a). Endow
X with the symplectic form = Im daD da. The functions (aD (u), a(u)) give a map f
from M to X. The metric on M is
 
2 daD da i daD da da daD
(ds) = Im dudu = dudu. (3.3)
du du 2 du du du du

This formula is valid for an arbitrary local parameter u on M. If one picks u = a, one
gets back the original formula (2.9) for the metric. (The above formula can be described
in a coordinate-free way by saying that the Kahler form of the induced metric on M is
f ().) Notice that had no particular positivity property and thus, if a(u) and aD (u)
are completely arbitrary local holomorphic functions, the metric (3.3) is not positive. We
will eventually construct a(u) and aD (u) in a particular way that will ensure positivity.
It is easy to see what sort of transformations preserve the general structure of the
metric. If we set a = (aD , a), = 1, 2, and let be the antisymmetric tensor with
12 = 1, then
i da da
(ds)2 = dudu. (3.4)
2 du du
This is manifestly invariant under linear transformations that preserve and commute
with complex conjugation (the latter condition ensures that a and a transform the same
way). These transformations make the group SL(2, R) (or equivalently Sp(2, R)). Also,

10
(3.4) is obviously invariant under adding a constant to aD or a. So if we arrange (aD , a)
as a column vector v, the symmetries that preserve the general structure are

v M v + c, (3.5)

where M is a 2 2 matrix in SL(2, R), and c is a constant vector. Later we will find (from
considerations involving the gauge fields and the electric and magnetic charges) that M
must be in SL(2, Z) and that in the pure N = 2 gauge theory, c must vanish. (In coupling
to matter, c will play an important role [7].) In general, the group of transformations (3.5)
can be thought of as the group of 3 3 matrices of the form
 
1 0
, (3.6)
c M

acting on the three objects (1, aD , a).

Generalization To Dimension Greater Than One
Though it will not be exploited in the present paper, and can thus be omitted by the
reader, let us briefly discuss the generalization to other gauge groups. If the gauge group
G has rank r, then M has complex dimension r. Locally from (2.11) the metric is

2F
(ds)2 = Im dai daj , (3.7)
ai aj

with distinguished local coordinates ai and a holomorphic function F . We again reformu-

late this by introducing
F
aD,j = . (3.8)
aj
Then we can write
X
(ds)2 = Im daD,i dai . (3.9)
i

To formulate this invariantly, we introduce a complex space X = C2r with coordinates

P

ai , aD,j . We endow X with the symplectic form = 2i i dai daD,i daD,i dai of
P
type (1, 1) and also with the holomorphic two-form h = i dai daD,i . Then we introduce
arbitrary local coordinates us , s = 1, . . . , r, on the moduli space M, and describe a map
f : M X by functions ai (u), aD,j (u). We require f to be such that f (h ) = 0; this

11
precisely ensures that locally, if we pick ui = ai , then aD,j must be of the form in (3.8)
with some holomorphic function F . Then we take the metric on M to be the one whose
Kahler form is f (); in formulas the metric is
X aD,i ai
(ds)2 = Im s ut
dus dut . (3.10)
s,t,i
u

If again we arrange a, aD as a 2r-component column vector v, then the formalism is in-

variant under transformations v M v + c with M a matrix in Sp(2r, R) and c a constant
vector. Again, considerations involving the charges will eventually require that M be in
Sp(2r, Z) and impose restrictions on c.

3.2. Physical Interpretation Via Duality

So far we have seen that the spin zero component of the N = 2 multiplet has a Kahler
metric of a very special sort, constructed using a distinguished set of coordinate systems.
This rigid structure is related by N = 2 supersymmetry to the natural linear structure
of the gauge field. We have found that, for the spin zero component, the distinguished
parametrization is not completely unique; there is a natural family of parametrizations
related by SL(2, R). How does this SL(2, R) (which will actually be reduced to SL(2, Z))
act on the gauge fields?
SL(2, R) is generated by the transformations
   
1 b 0 1
Tb = , and S = (3.11)
0 1 1 0
with real b. The former acts as aD aD +ba, a a; this acts trivially on the distinguished
coordinate a, and can be taken to act trivially on the gauge field. By inspection of (2.12),
the effect of aD aD + ba on the gauge kinetic energy is just to shift the angle by 2b;
in the abelian theory, this has no effect until magnetic monopoles (or at least non-trivial
U (1) bundles) are considered. Once that is done, the allowed shifts in the angle are
by integer multiples of 2; that is why b must be integral and gives essentially our first
derivation of the reduction to SL(2, Z).
The remaining challenge is to understand what S means in terms of the gauge fields.
We will see that it corresponds to electric-magnetic duality. To see this, let us see how
duality works in Lagrangians of the sort introduced above.

12
 We work in Minkowski space and consider first the purely bosonic terms involving
2
only the gauge fields. We use conventions such that F = ( F )2 and ( F ) = F
where F denotes the dual of F . The relevant terms are
Z Z
1 2 1
Im
(a) (F + i F ) = Im (a) (F 2 + i F F ) . (3.12)
32 16

Duality is carried out as follows. The constraint dF = 0 (which in the original description
follows from F = dA) is implemented by adding a Lagrange multiplier vector field VD .
Then F is treated as an independent field and integrated over. The normalization is set as
follows. The U (1) SU (2) is normalized such that all SO(3) fields have integer charges
(matter multiplets in the fundamental representation of SU (2) therefore have half integer
charges). Then, a magnetic monopole corresponds to 0 F = 8 (3) (x). For VD to
couple to it with charge one, we add to (3.12)
Z Z Z
1 1 1
VD F = FD F = Re ( FD iFD )(F + i F ) (3.13)
8 8 16

where FD = VD VD is the the field strength of VD . We can now perform the

Gaussian functional integral over F and find an equivalent Lagrangian for VD ,

1 1 1 1 2
Im (FD + i FD )2 = Im (FD + i FD FD ) . (3.14)
32 16
1
R
We now repeat these steps in N = 1 superspace. We treat W in 8 Im d2 (A)W 2
as an independent chiral field. The superspace version of the Bianchi identity dF = 0 is
Im DW = 0 (D is the supercovariant derivative). It can be implemented by a real vector
superfield VD Lagrange multiplier. We add to the action
Z Z Z
1 4 4 1 4 4 1
Im d xd VD DW = Re d xd iDVD W = Im d4 xd2 WD W. (3.15)
4 4 4

Performing the Gaussian integral over W we find an equivalent Lagrangian

Z
1 1
Im d2 2
WD . (3.16)
8 (A)

To proceed further, we need to transform the N = 1 chiral multiplet A to AD . The

kinetic term Z
Im d4 h(A)A (3.17)

13
is transformed by
AD = h(A) (3.18)

to Z
Im d4 hD (AD )AD (3.19)

where hD (h(A)) = A is minus the inverse function. Then using h (A) = (A) the
coefficient of the gauge kinetic term (3.16) becomes
1 1
= = hD (AD ) = D (AD ). (3.20)
(A) h (A)
Note that a shift of h by a constant does not affect the Lagrangian. Therefore, the duality
transformation has a freedom to shift AD by a constant.
The relations AD = h(A) and hD = A mean that the duality transformation pre-
cisely implements the missing SL(2, Z) generator S. The function = h is mapped by
1
D (AD ) = (3.21)
(A)
(a) 4
Remembering that (a) = 2 + i g(a)2 , we see that the duality transformation inverts

rather than the low energy gauge coupling g(a). (A similar phenomenon is known in
1
R R
duality in string theory [29], where Bij + iGij , which is analogous to our , is
inverted rather than Gij .)
It is important to stress that unlike + 1, the duality transformation is not a
symmetry of the theory. It maps one description of the theory to another description of
the same theory.
For other gauge groups G the low energy Lagrangian has several abelian fields, Ai , in
the Cartan subalgebra. Then

(AD )i = hi (Ai ) = i F (Ai ) (3.22)

which leads to
hiD (hj (Ak )) = Ai (3.23)

and the metrics

ij (A) =i j F (A) = j hi (A)
(3.24)
ij i j
D (AD ) = FD (AD ) = j hiD (AD )

14
satisfy
jk
fij fD = ik . (3.25)

The above transformation together with the more obvious shifts ADi ADi + Mij Aj
generate Sp(2r, Z).

3.3. Coupling To Gravity

Before concluding this section, we would like to compare the structure we have found to
the special geometry that appears if the chiral multiplet is coupled to N = 2 supergravity
[23]. (This will not be used in the present paper.) In N = 2 supergravity, the general
Kahler metric for a system of r chiral superfields is described locally by a holomorphic
function G0 (a1 , . . . , ar ) of r complex variables ai . The Kahler potential is
 !
i X i G0 G 0
Kgrav = ln 2i(G0 G 0 ) + a ai i . (3.26)
2 i ai a

In global supersymmetry we had a local holomorphic function F with

 
i X i F i F
K= a a . (3.27)
2 i ai ai

One would expect that there is some limit in which gravitational effects are small and
(3.26) would reduce to (3.27). How does this occur?
It suffices to set
MP l 2
G0 = i + F, (3.28)
4
with MP l the Planck mass. Then if MP l is much larger than all relevant parameters, we
get
K
Kgrav = ln MP l 2 + + O(MP l 4 ). (3.29)
MP l 2
The constant term ln MP l 2 does not contribute to the Kahler metric, so up to a nor-
malization factor of 1/MP2 l , the Kahler metric with supergravity reduces to that of global
N = 2 supersymmetry as MP l keeping everything else fixed.
More fundamentally, we would like to compare the allowed monodromy groups. In
supergravity, the global structure is exhibited as follows. One introduces an additional

15
variable a0 and sets G = (a0 )2 G0 . One also introduces aD,j = G/aj for j = 0, . . . , r. Then
one finds that the special Kahler structure of (3.26) allows Sp(2r + 2, R) transformations
acting on (aDi , aj ).2 Now, in decoupling gravity, we consider G to be of the special form
in (3.28). In that case,
MP l
aD,0 = i . (3.30)
2

The other ai , aD,j are independent of MP l . To preserve this situation in which MP l

appears only in aD,0 , we must consider only those Sp(2r + 2, R) transformations in which
the transformations of all fields are independent of aD,0 . These transformations all leave
a0 invariant. There is no essential loss then in scaling the as so that a0 = 1. Arrange
the aD,i , aj with i, j = 1 . . . r as a column vector v. The Sp(2r + 2, R) transformations
that leave invariant a0 = 1 act on v by v M v + c where M Sp(2r, R) and c is a
constant. (The transformations with c 6= 0 do not leave aD,0 invariant, but its variation
is independent of MP l and so is negligible in the limit in which gravity is weak.) This is
precisely the duality group that we found in the global N = 2 theory.

4. Dyon masses

The SU (2) gauge theory under discussion has electrically and magnetically charged
particles whose masses satisfy
M 2 = 2|Z|2 (4.1)

with
Zcl = a(ne + cl nm ) (4.2)

(cl = 2
+ i 4
g2
) where ne and nm are the electric and magnetic charges; they are integers
as long as all elementary fields are in representations of SO(3) (fields that are in SU (2)
representations of half-integral spin have half-integral ne ). The origin of this formula [12]
is that Z arises as a central extension in the N = 2 supersymmetry algebra and equation

2
Once once considers the gauge fields, this is reduced to Sp(2r + 2, Z). The symplectic form
P
preserved by Sp(2r + 2, R) is the usual one i
dai daD,i .

16
(4.1) follows from the representation theory of N = 2 for small representations. One
uses the algebra to show that for any state of given (nm , ne ),

M 2|Z| (4.3)

with equality precisely for the small representations of N = 2 (four helicity states instead
of sixteen). We will not review the argument here. States saturating the inequality are
called BPS-saturated states.
As stressed in [12], the same interpretation should apply quantum mechanically, but
the coefficients of ne and nm in Z might be modified. One way to find the modification is to
calculate the central extension of the algebra from the the low energy effective Lagrangian
(2.7). This leaves an ambiguity in shifting A and AD by a constant. As remarked above,
such an ambiguity can be resolved by considering the full high energy theory in its effective
form as in (2.12).
Alternatively, we can couple the theory based on (2.7) to a hypermultiplet two N = 1
f with electric charge ne and a canonical kinetic term. As in
chiral multiplets M and M
(2.5), the coupling to the gauge field is N = 2 supersymmetric only with a superpotential

f .
2ne AM M (4.4)

There is also a possibility of adding a mass term; this has the effect of shifting ne A by a
constant corresponding to the ambiguity mentioned above. By embedding the theory in a
higher energy theory such as (2.12), this ambiguity is removed, and one learns in the pure

f in the superpotential is precisely 2ne A, with no
gauge theory that the coefficient of M M
additional additive constant. With additional matter multiplets included [7], such a term
will arise and play an important role.
f are massive. As the corresponding states are in
For non-zero a, the fields M and M
a small representation, their mass is determined by the central extension in the algebra
[12]. Comparing (4.1) and (4.4), we conclude that Z = ane . Using the duality transfor-
mation, it is clear that for magnetic monopoles with magnetic charge nm , Z = aD nm and
for dyons
Z = ane + aD nm . (4.5)

17
These formulas can be verified directly; for instance, to study the BPS bound for
monopoles, we consider the full high energy theory in its effective form (2.12) and ex-
amine the bosonic terms in the Hamiltonian of a magnetic monopole for real a :
Z  
1 3 a b 1 a b
E = Im d x ab (Di ) (Di ) + ab Bi Bi
4 2
Z  ab !
1 1 1
= Im d3 x (Di h)a (Di h)b + ab Bia Bib
4 2
(Z  ab   )
1 1 1 1 (4.6)
= Im d3 x (Di h)a ac Bic (Di h)b bd Bid
4 2 2

2i (Bia ha )
 I 
 2 
 2 a 
 d sBi ha  = 2|nm aD |.
 4 

This inequality confirms that for monopoles Z = nm aD .

We would like to make a few comments:

1. The expression (4.5) has the correct semiclassical limit and manifest duality.

2. Unlike (4.2), it is renormalization group invariant. Note that it does not differ from
(4.2) merely by replacing i 4
g2
by (a); i.e. by the running coupling. The distinction
between them appears already at one loop.

3. From this expression, it follows that the renormalization of the BPS formula vanishes
for N = 4 supersymmetric Yang-Mills theory. For N = 4, the classical expression
(2.14) is exact (note the comment at the end of section 2) so there are no corrections
to the classical formula aD = cl a.

4. The generalization of (4.5) to an arbitrary gauge group is

Z = ai ne,i + hi (a)nm i = a ne + aD nm (4.7)

where ai are local coordinates on the quantum moduli space and ne,i , nm i are the
electric and magnetic charges.
Now let us discuss the restrictions that the dyon mass formula puts on the duality
discussed in section 3. First of all, it is clear that Z, since it determines particle masses

18
(or appears in the supersymmetry algebra) must be invariant under the monodromies. In
section 3, we arranged (aD , a) as a column vector v, and found that the analysis of the
Kahler metric permitted monodromies v M v + c. Since ne a + nm aD is not invariant
under addition of a constant to a or aD , and there is no way to compensate for this by
3
any transformation of ne or nm , we must set c = 0. Moreover, under v M v, we need
w wM 1 where w is the row vector w = (nm , ne ). But since (nm , ne ) are integers,
M 1 must be integer-valued. As the determinant of M is 1, it follows that also M is
integer-valued; hence the monodromy group is at most SL(2, Z).
Stability of BPS-Saturated States
Now we will discuss the stability of BPS-saturated states. Many of the following
remarks are well known and none are new.
A BPS-saturated state of given (nm , ne ) determines a vector Z = nm aD + ne a in the

complex plane. Its mass is 2 times the length of that vector. According to (4.3), all
other states with the same (nm , ne ) are heavier. Assuming that the ratio aD /a is not real,
the complex numbers a and aD generate a lattice in the complex plane, and Z is a point
in that lattice.
Let us analyze a possible decay process of a BPS-saturated state S with Z = ane +

aD nm and mass M = 2|Z| to states Si with Zi = nm,i aD +ne,i a and masses Mi 2|Zi |.
P
Since the charges (nm , ne ) must be conserved, Z = i Zi . It is clear from the triangle
inequality that
X
|Z| |Zi | (4.8)
i

and hence
X
M Mi . (4.9)
i
P
Of course if M < i Mi the decay is impossible. Equality is achieved in (4.8) when and
only when all the states are BPS-saturated and all the vectors Z and Zi are aligned, that is
P
ti = Zi /Z is real and positive and i ti = 1. Assuming that the ratio of aD /a is not real,
this is possible only if the charge vectors (nm , ne ) of the initial particles and (nm,i , ne,i)

3
When matter is included, additional terms appear in Z and one no longer gets c = 0 [7].

19
of the final particles are proportional. This in turn is possible only if nm and ne are not
relatively prime; i.e. (nm , ne ) = (qm, qn) with integers q, m and n.
Conversely, states with (nm , ne ) = (qm, qn) are in fact only neutrally stable against
decay to q states with charges (m, n).
Now what happens to these stable particles (BPS saturated with (nm , ne ) relatively
prime) as we vary some of the parameters that determine the vacuum? Such small changes
in the vacuum can be described by emission of zero momentum particles the neutral u
quanta. Possible emission of such particles does not affect the argument for stability of
the stable BPS saturated states so those states must persist as the parameters are varied.
So far we have assumed that aD /a is not real. As explained in [27], the above argument
fails if, upon varying the parameters, aD /a passes through the real axis.4 When that
happens, the two dimensional lattice generated by aD and a collapses to a one dimensional
configuration, and it becomes much easier for the triangle inequality to collapse to an
equality. For instance, if aD /a is real and irrational, there are infinitely many points
nm,1 aD + ne,1 a on the segment between 0 and Z = nm aD + ne a; letting Z1 be one such
point, and Z2 = Z Z1 , a BPS-saturated state of given Z is only neutrally stable against
decaying to possible BPS-saturated particles of given Z1 and Z2 .
Moreover and this is the main point it was shown in [27] that at least in two
dimensions, a BPS-saturated particle S of given (nm , ne ) can disappear (or appear) when
one passes through the point in parameter space at which aD /a is real. What happens
is that, for, say, Im(aD /a) > 0, the S particle, if it exists, is stable against decay to,
say, S1 + S2 , but for Im(aD /a) 0, the mass of the S particle goes up to the S1 + S2
threshold. Varying the parameters still further, to Im(aD /a) < 0, the S particle no longer
exists it has decayed to S1 + S2 . A BPS-saturated state of the given (nm , ne ) would
again be stable when Im(aD /a) < 0, but such a particle may not exist. In [27], the precise
number of BPS-saturated states that appear or disappear in this way was computed in
two dimensions. An analogous computation in four dimensions would be desirable.
In the present paper, discontinuity of the BPS-saturated spectrum will be found (in

4
Their formulation is slightly more general; they do not assume that the allowed values of Z
are integer linear combinations of two basic numbers a and aD .

20
section 6) only for strong coupling where it is difficult to explicitly check what is going on.
In a subsequent paper [7], we will see such jumping also for weak coupling.
Such jumping does not occur for the N = 4 theory since then one has the exact formula
aD = cl a ensuring that aD /a is not real. Consequently, the spectrum of BPS-saturated
states of given (nm , ne ) is independent of the coupling and so can be computed semiclassi-
cally for weak coupling. This has, in fact, been assumed in tests of Olive-Montonen duality
for N = 4.

5. Structure Of The Moduli Space

In section 2, we developed the general local framework for the low energy effective
action of the N = 2 theory. At the outset of section 3, we noted that this framework
could not be satisfactory globally because the metric on the moduli space of vacua could
not be positive definite. Instead, we found that the global structure could involve certain
monodromies; as we have explained, the group generated by the monodromies is a subgroup
of SL(2, Z).

5.1. The Singularity at Infinity

It is actually quite easy to see explicitly the appearance of non-trivial monodromies. In

fact, asymptotic freedom implies a non-trivial monodromy at infinity. The renormalization
group corrected classical formula Fone loop = iA2 ln(A2 /2 )/2 gives for large a

F 2ia ia
aD = ln(a/) + . (5.1)
a

It follows that aD is not a single-valued function of a for large a. If we recall that the
physical parameter is really u = 12 a2 (at least for large u and a), then the monodromy can
be determined as follows. Under a circuit of the u plane at large u, one has ln u ln u+2i,
and hence ln a ln a + i. So the transformation are

aD aD + 2a
(5.2)
a a.

21
Thus, there is a non-trivial monodromy at infinity in the u plane,
 
2 1 2
M = P T = (5.3)
0 1

where P is the element 1 of SL(2, Z) and as usual

 
1 1
T = . (5.4)
0 1

The factor of P in the monodromy exists already at the classical level. As we said
above, a and a are related by a gauge transformation (the Weyl subgroup of the SU (2)
gauge group) and therefore we work on the u plane rather than its double cover, the a
plane. In the anomaly free Z8 subgroup of the R symmetry group U (1)R , there is an
operation that acts on a by a a; when combined with a Weyl transformation, this
is the unbroken symmetry that we call P . Up to a gauge transformation it acts on the
bosons by , so it reverses the sign of the low energy electromagnetic field which
in terms of SU (2) variables is proportional to Tr (F ). Hence it reverses the signs of all
electric and magnetic charges and acts as 1 SL(2, Z). The P monodromy could be
removed by (perhaps artificially) working on the a plane instead of the u plane.
The main new point here is the factor of T 2 which arises at the quantum level. This
factor of T 2 has a simple physical explanation in terms of the electric charge of a magnetic
monopole. As explained in [30], magnetic monopoles labeled by (nm , ne ) have anomalous
ef f
electric charge ne + 2 nm . The appropriate effective theta parameter is the low energy
one
daD daD /du
ef f = 2 Re (a) = 2 Re = 2 Re . (5.5)
da da/du
For large |a|, we have ef f 4arg(a) which can be understood from the anomaly in
the U (1)R symmetry. The monodromy at infinity transforms the row vector (nm , ne ) to
(nm , ne 2nm ), which implies that (aD , a) transforms to (aD + 2a, a). The electric
charge of the magnetic monopole can in fact be seen in the formula for Z, which if we take
aD from (5.1) and set a = a0 eief f /4 (with a0 > 0) is
   
ief f /4 ef f nm 2 ln a0 / + 1
Z a0 e ne + + inm . (5.6)
2

22
The monodromy under ef f ef f + 4 is easily seen from this formula to transform
(nm , ne ) in the expected fashion. Of course, this simple formula depended on the semi-
classical expression (5.1) for aD ; with the exact expressions we will presently propose, the
results are much more complicated, in part because the effective theta angle is no longer
simply the argument of a.

5.2. Singularities at Strong Coupling

The monodromy at infinity means that there must be an additional singularity (or
topological complication) somewhere in the u plane. If M is the moduli space of vacua
with all singularities deleted, then the monodromies must give a representation of the
fundamental group of M in SL(2, Z). Can this representation be abelian? If the mon-
odromies all commute with P T 2 , then a2 is a good global complex coordinate, and the
metric is globally of the form (3.1) with a global harmonic function Im (a). As we have
already noted, such a metric could not be positive.
The alternative is to assume a nonabelian representation of the fundamental group.
This requires at least two more punctures of the u plane (in addition to infinity). Since
there is a symmetry u u acting on the u plane, the minimal assumption is that there
are precisely two more punctures exchanged by this symmetry. In this paper, we will
find that this assumption leads to a unique and elegant solution that passes many tests.
(In a following paper [7], we will in some sense derive this assumption from more general
properties of N = 2 systems with matter.)
The most natural physical interpretation of singularities in the u plane is that some
additional massless particles are appearing at a particular value of u. Such a phenomenon
of singularities in moduli space associated with the occurrence of extra massless particles
has already been observed in N = 1 theories [2] and we will argue that it also happens in
our problem.
For instance, in the classical theory, at u = a = 0, the SU (2) gauge symmetry is
restored; all the gluons become massless. In fact classically aD = 4ia/g 2 also vanishes
at this point, and the monopoles and dyons become massless as well. One might be
tempted to believe that the missing singularity comes from an analogous point in the

23
quantum theory at which the gauge boson masses vanish. Though this behavior might
seem unusual in asymptotically free theories in general, there are good indications that
some N = 1 theories have an infrared fixed point with massless nonabelian gluons [2,31].
However, there are good reasons to doubt that the N = 2 theory has this behavior.
First of all, to make sense of the monodromies, one needs (as we saw above) not a single
singularity but (at least) a pair of singularities at non-zero hTr 2 i. One might be willing
to believe that hTr 2 i =
6 0 in the theory with massless nonabelian gauge bosons because of
spontaneous breaking of the discrete chiral symmetry. This assumption, however, clashes
with the asymptotic conformal invariance that one would expect in the infrared if the
gauge bosons are massless. In fact, a non-zero expectation value of Tr 2 contradicts
conformal invariance unless Tr 2 is of dimension zero. In a unitary quantum field theory,
the only operator of zero dimension is the identity; Tr 2 cannot mix with the identity
under renormalization because it is odd under a global symmetry.
Moreover, a conformally invariant point for the N = 2 theory is far-fetched because
conformal invariance together with N = 2 supersymmetry implies invariance under the
full N = 2 superconformal algebra including the U (1)R symmetry. Thus, the instanton
anomaly in the U (1)R symmetry would have to somehow disappear. Moreover, for a field
such as O = Tr 2 which is in a chiral multiplet, superconformal invariance implies that
the dimension D(O) and U (1)R charge R(O) are related by D(O) = R(O)/2. Thus at a
conformal point the dimension of Tr 2 would have the canonical value 2, not the value 0
that it should have in order to have an expectation value.

5.3. Interpretation Of The Singularities

Since the above discussion of massless gauge bosons does not appear promising, we
will assume that the singularities come from massive particles of spin 1/2 that become
massless at particular points in the moduli space. Since there are no such elementary
multiplets, these must be bound states or collective excitations. It might sound counter
intuitive that such objects can become massless. However, a similar phenomenon of mass-
less bound states (which are not Goldstone bosons) has been observed in N = 1 theories
[2] and we will argue that it also happens in N = 2.

24
 The possibilities are severely restricted by the structure of N = 2 supersymmetry: a
massive multiplet of particles of spins 1/2 must be a hypermultiplet that saturates the
BPS bound.
In the semiclassical approximation the only such hypermultiplets in the N = 2 gauge
theory are the monopoles and dyons whose mass renormalization was the subject of sec-
tion 4. We will interpret the needed singularities as arising when these particles become
massless. For instance, from the discussion of masses in section 4, the monopole becomes
massless, while the gluons remain massive, at a point where aD = 0 while a 6= 0. Similarly
a (1, 1) dyon becomes massless if a + aD = 0 while a, aD 6= 0.
Proposing that magnetic monopoles become massless and dominate the low energy
landscape (at certain points in the moduli space of vacua) may seem bold. In the rest of
this paper we will give evidence for this hypothesis as follows:

1. We will use the renormalization group to compute the SL(2, Z) monodromy that arises
near a point at which a hypermultiplet becomes massless. We will then show that if
the hypermultiplets that are relevant are the monopoles and dyons that are visible
semiclassically, then the monodromies work out consistently.

2. The underlying N = 2 theory can be perturbed to an N = 1 theory by adding

Tr 2 as a superpotential. It is believed that this causes confinement of quarks. We
will see that the same perturbation, added near the point at which monopoles are
becoming massless, causes the monopoles to condense (develop a vacuum expectation
value). This naturally leads to confinement of charges, giving for the first time
an example in which confinement in a nonabelian theory is naturally understood in
terms of monopole condensation.

3. Finally, we will show that if it is assumed that there is a minimal set of singularities
coming from massless monopoles and dyons, then one can determine uniquely and
exactly the full structure of the low energy theory, including the Kahler metric on the
moduli space of vacua and the particle masses as a function of the parameters. In
particular the puzzle with positivity of the metric mentioned at the outset of section
3 is naturally overcome.

25
 Explaining these points will occupy the rest of this paper.

5.4. Effects Of A Massless Monopole

Our first task is to analyze the behavior of the effective Lagrangian near a point u0
on the moduli space where magnetic monopoles become massless, that is, where

aD (u0 ) = 0. (5.7)

Since monopoles couple in a non-local way to the original photon, we cannot use that
photon in our effective Lagrangian. Instead, we should perform a duality transformation
and write the effective Lagrangian in terms of the dual vector multiplet AD . The low energy
theory is therefore an abelian gauge theory with matter (an N = 2 version of QED). The
unusual fact that the light matter fields are magnetically charged rather than electrically
charged does not make any difference to the low energy physics. The only reason we call
these particles monopoles rather than electrons is that this language is appropriate at large
|u| where the theory is semiclassical.
The dominant effect on the low energy gauge coupling constant is due to loops of
light fields. In our case, these are the light monopoles. The low energy theory is not
asymptotically free and therefore its gauge coupling constant becomes smaller as the mass
of the monopoles becomes smaller. Since the mass is proportional to aD , the low energy
coupling goes to zero as u u0 . The electric coupling constant which is the inverse of the
magnetic one diverges at that point.
More quantitatively, using the one loop beta function, near the point where aD = 0,
the magnetic coupling is
i
D ln aD . (5.8)

Since aD is a good coordinate near that point,

aD c0 (u u0 ) (5.9)

with some constant c0 . Using D = dhD /daD , we learn that

i i
a(u) = hD (u) a0 + aD ln aD a0 + c0 (u u0 ) ln(u u0 ) (5.10)

26
for some constant a0 = a(u = u0 ). This constant a0 cannot be zero because if it had been
zero, all the electrically charged particles would have been massless at u = u0 and the
computation using light monopoles only would not be valid. (In fact, there would be no
local effective field theory incorporating both the light electrons and light monopoles.)
Now we can read off the monodromy. When u circles around u0 , so ln(u u0 )
ln(u u0 ) + 2i, one has
aD aD
(5.11)
a a 2aD .
This effect is a sort of dual of the monodromy at infinity. Near infinity, the monopole gains
electric charge, and near u = u0 , the electron gains magnetic charge. (It does not come
back as a dyon but as a pair of particles for reasons explained at the end of section 6.)
(5.11) can be represented by the 2 2 monodromy matrix
 
2 1 1 0
M1 = ST S = . (5.12)
2 1

5.5. The Third Singularity

With our assumption that there are only three singularities (counting u = ) and
with two of the three monodromies determined in (5.3) and (5.12), we can now determine
the third monodromy, which we will call M1 . (The motivation for the notation is that
we will eventually introduce parameters in which the singularities at finite u are at u = 1
and u = 1.) With all of the monodromies taken in the counter clockwise direction as in
figure 1, the monodromies must obey M1 M1 = M , and from this we get
 
2 1 1 2
M1 = (T S)T (T S) = . (5.13)
2 3

The matrix M1 is conjugate to M1 . In fact, if

 
1 1
A = T M1 = , (5.14)
2 1

then
M1 = AM1 A1 . (5.15)

27
 P

-1 1

M M1
-1

Fig. 1: The u plane with monodromies around 1, 1, and . Note the

choice of base point in the definition of the monodromies.

Hence, M1 can arise from a massless particle, just like M1 . (5.15) would also hold if A is
replaced by AM1 r for any integer r.
What kind of particle should become massless to generate this singularity? If one
arranges the charges as a row vector q = (nm , ne ), then the massless particle that produces
a monodromy M has qM = q. For instance, monodromy M1 arises from a massless
monopole of charge vector q1 = (1, 0), and using the known form of M1 , one has q1 M1 = q1 .
Duality symmetry implies that this must be so not just for the particular monodromy M1
but for any monodromy coming from a massless particle. Upon setting q1 = (1, 1), we
get q1 M1 = q1 , and hence the monodromy M1 arises from vanishing mass of a dyon
of charges (1, 1).
It seems that we are seeing massless particles of charges (1, 0) or (1, 1). However,
there is in fact a complete democracy among dyons. The BPS-saturated dyons that exist
semiclassically have charges (1, n) (or (1, n)) for arbitrary integer n. The monodromy
at infinity brings about a shift (1, n) (1, n2). If one carries out this shift n times before
proceeding to the singularity at u = 1 or u = 1, the massless particles producing those
singularities would have charges (1, 2n) and (1, 1 2n), respectively. This amounts to

28
conjugating the representation of the fundamental group by M n .
The particular matrix A in (5.14) obeys A2 = 1, which is equivalent to the iden-
tity as an automorphism of SL(2, Z). Conjugation by A implements the underlying Z2
symmetry of the quantum moduli space which according to our assumptions exchanges
the two singularities. The Z2 maps M1 M1 = M1 , M1 M1

= M1 and

M M = M1 M1

= M1 M1 . Note that M is not just obtained from M by
conjugation, but the relation M = M1 M1 is preserved. The reason for that is that (as
in any situation in which one is considering a representation of the fundamental group of
a manifold in a nonabelian group), the definition of the monodromies requires a choice of
base point, as shown in figure 1. The operation u u acts on the base point, and this
has to be taken into account in determining how M transforms under Z2 .
One can go farther and show that if one assumes the existence of a Z2 symmetry
between M1 and M1 , then they must be conjugate to T 2 , and not some other power of T .
In our derivation of the monodromy (5.12), the 2 came from something entirely independent
of the assumption of a Z2 symmetry, namely, from the charges and multiplicities of the
monopoles that exist semiclassically.

5.6. Monopole Condensation And Confinement

We will now explain a quite satisfying physical phenomenon which was in fact at the
heart of how some of these things were originally discovered.
We recall that the underlying N = 2 chiral multiplet A decomposes under N = 1
supersymmetry as a vector multiplet W and a chiral multiplet . Breaking N = 2 down
to N = 1, one can add a superpotential W = mTr 2 for the chiral multiplet. This gives
a bare mass to , reducing the theory at low energies to a pure N = 1 gauge theory. The
low energy theory has a Z4 chiral symmetry. This theory is strongly believed to generate a
mass gap, with confinement of charge and spontaneous breaking of Z4 to Z2 . Furthermore,
there is no vacuum degeneracy except what is produced by this symmetry breaking, so
that there are precisely two vacuum states [32].
How can this be mimicked in the low energy effective N = 2 theory? That theory
has a moduli space M of quantum vacua. The massless spectrum at least semiclassically

29
consists solely of the abelian chiral multiplet A of the unbroken U (1) subgroup of SU (2). If
those are indeed the only massless particles, the effect in the low energy theory of turning
on m can be analyzed as follows. The operator Tr 2 is represented in the low energy
theory by a chiral superfield U . Its first component is the scalar field u whose expectation
value is
hui = hTr 2 i (5.16)

( is the = 0 component of the superfield ). This is a holomorphic function on the

moduli space. At least for small m we should add to our low energy Lagrangian an effective
superpotential Wef f = mU (soon we will show that this is the exact expression also for
large m).
Turning on the superpotential mU would perhaps eliminate almost all of the vacua
and in the surviving vacua give a mass to the scalar components of A. But if there are no
extra degrees of freedom in the discussion, the gauge field in A would remain massless. To
get a mass for the gauge field, as is needed since the microscopic theory has a mass gap
for m 6= 0, one needs either (i) extra light gauge fields, giving a non-abelian gauge theory
and possible strong coupling effects, or (ii) light charged fields, making possible a Higgs
mechanism.
Thus we learn, as we did in discussing the monodromies, that somewhere on M extra
massless states must appear. The option (i) does not seem attractive, for reasons that we
have already discussed. Instead we will consider option (ii), with the further proviso, from
our earlier discussion, that the light charged fields in question are monopoles and dyons.
Near the point at which there are massless monopoles, the monopoles can be repre-
f, as long as
sented in an N = 1 language by ordinary (local) chiral superfields M and M
we describe the gauge field by the dual to the original photon, AD . The superpotential is

c=
W f + mU (AD ),
2AD M M (5.17)

where the first term is required by N = 2 invariance of the m = 0 theory, and the
second term is the effective contribution to the superpotential induced by the microscopic
perturbation m Tr 2 .

30
 The fact that the superpotential (5.17) is exact can be established by using the non-
renormalization theorem of [1] as follows. For m = 0 the theory is invariant under SU (2)R .
It will suffice to consider its U (1)J subgroup (2.2). This is an N = 1 R symmetry under
f are in an N = 2
which has charge zero. The two N = 1 chiral fields M and M
hypermultiplet. Therefore, according to (2.4), they both have charge one. The presence
of a term mTr 2 in the microscopic superpotential shows that the parameter m carries
c (m, M M
charge two. The low energy superpotential is holomorphic in its variables W f, AD )
f = 0,
and should have charge two under U (1)J . Imposing that it is regular at m = M M
c = mf1 (AD ) + M M
we find that it is of the form W ff2 (AD ). The functions f1 and f2 are

independent of m and can be determined by examining the limit of small m, leading to

(5.17).
The low energy vacuum structure is easy to analyze. Vacuum states correspond to
solutions, up to gauge transformation, of

c=0
dW (5.18)

that obey the additional condition

f|
|M | = |M (5.19)

f both the superfields and their first components). The latter

(we denote by M and M
condition comes from vanishing of the D terms. Implementing these conditions, one finds
f = 0 with arbitrary aD ; this is simply
if m = 0 that vacuum states correspond to M = M
the familiar moduli space M. If m 6= 0 the result is quite different. We get
du
f+m
2M M =0
daD (5.20)
f = 0.
aD M = aD M

f 6= 0, whence the second equation

Assuming that du 6= 0, the first equation requires M, M
requires aD = 0. Imposing also (5.19), we get a unique solution up to gauge transformation,
with
 1/2
f = mu (0)/ 2
M =M . (5.21)

Expanding around this vacuum, it is easy to see that there is a mass gap. For instance,
f 6= 0. The Higgs mechanism
the gauge field gets a mass by the Higgs mechanism, since M, M

31
in question is a magnetic Higgs mechanism, since the fields with expectation values are
monopoles! Condensation of monopoles will induce confinement of electric charge. Thus,
we get an explanation in terms of the low energy effective action of why the microscopic
theory becomes confining when the m Tr 2 superpotential is added.
We have also noted that in the presence of the perturbing superpotential, the micro-
scopic theory has a Z4 global symmetry, spontaneously broken down to Z2 . This symmetry
breaking is manifest in the effective theory, since the broken symmetry exchanges the point
where aD = 0 and there is a massless monopole with a point where a aD = 0 and there
is a massless dyon. Thus, the effective theory has two vacuum states from the two points
of extra massless particles (corresponding to the two singularities in the u plane that
were discussed above), related by a broken symmetry, in parallel with what is expected
microscopically.
In the microscopic description, one attributes the properties of the two vacuum states
(for m 6= 0) to a difficult-to-understand strong gauge coupling. In the low energy theory, we
have found a perfectly peaceful description involving a weakly coupled theory of monopoles
and photons; the coupling constant flows to zero in the infrared if m = 0, and in general
flows to a value of order 1/ ln m, since this is the behavior of weakly coupled QED. The
original, electric, gauge coupling is the inverse of the magnetic coupling, so flows in the
infrared to a value of order ln m; in this sense the two vacua that survive when m is
small and non-zero are strongly coupled.
The effective superpotential (5.17) gives a good description of the low energy physics
near the point u0 where the monopoles are light. What is its meaning far from u0 ? Then,
except for the expectation values of the monopoles and their masses, few of their properties
could be correctly inferred from the effective Lagrangian. The effective Lagrangian has
even greater difficulties if one tries to continue it to u = u0 where massless dyons should
appear. There cannot be an effective field theory containing both the monopoles and dyons
as elementary fields, as they are not relatively local. The theory we are discussing is an
interesting example of a theory in which, while in the neighborhood of any one vacuum
there is a good description by a low energy effective theory, there is no low energy effective
theory that is reasonable everywhere.

32
5.7. The Effective Parameter

Finally, we will tie up some remaining loose ends and at the same time prepare for
the solution of the model in the next section.

In the above, we found two vacuum states for m 6= 0, while assuming that du 6= 0.
f = 0, for any point at which du = 0. We do
We would get additional states, with M = M
not want any additional vacuum states, since two is the correct number of vacua in the
microscopic theory, so we assume that du 6= 0 everywhere.

The fact that du 6= 0 means that u is everywhere a good local coordinate on the space
of vacua. Is it also a good coordinate globally? The question is whether there is precisely
one vacuum for given u. According to (5.16), u is simply the expectation value of Tr 2 ,
regarded as a function on the space of vacua. At least when u = Tr2 is large, perturbation
theory is reliable and there is precisely one vacuum for given u. This combination of facts
strongly suggests that u is a good global coordinate on the space M of vacua.

In proceeding in the next section to propose a solution of the model, we will assume
that M is just the u plane (with u equal to the expectation value of Tr 2 ) with precisely
the two singularities that we proposed above. We will pick a renormalization convention
such that the singularities in the u plane are at 1 and 1. Using the solution of the model
that we will propose in the next section, it would be possible to compare to perturbation
theory and determine how this convention compares to other conventions such as M S.

6. The Solution Of The Model

In this section, we will, finally, put the pieces together, with a couple of new ingredi-
ents, and make our proposal for the solution of the model.

The moduli space M of quantum vacua is to be the u plane with singularities at 1, 1,

and and a Z2 symmetry acting by u u. Duality means that over the punctured u
plane there is a flat SL(2, Z) bundle V with the following monodromies around , 1, and

33
1:  
1 2
M =
0 1
 
1 0
M1 = (6.1)
2 1
 
1 2
M1 = .
2 3
The quantities (aD (u), a(u)) are a holomorphic section of the bundle V (or of V C to be
precise). This section is to be determined by its asymptotic behavior. Near u = ,

a 2u
(6.2)
2u
aD i ln u.

Near u = 1,
aD c0 (u 1)
i (6.3)
a a0 + aD ln aD ,

with constants a0 , c0 . The behavior near u = 1 is similar, with a aD replacing aD .
There is also one more important constraint. The metric on the u plane can be written

(ds)2 = Im( ) |da|2 (6.4)

with
daD /du
= (6.5)
da/du.
To ensure positivity of the metric, Im( ) must be positive definite. We need not specify
the asymptotic behavior of near 1, 1, and , since this is determined by the asymptotic
behavior of a and aD .
The key point in making it practical to solve these conditions is that the flat SL(2, Z)
bundle has a nice interpretation. To begin with, note that the monodromy matrices in
equation (6.1) are all congruent to 1 modulo 2. So these matrices do not generate SL(2, Z)
at most they could generate the subgroup of SL(2, Z) consisting of matrices congruent
to 1 modulo 2. This subgroup is usually called (2). In fact, M and M1 do generate
(2).5 Moreover, the u-plane punctured at 1, 1, and has a very special interpretation.
5
For a quick proof of this and other assertions made presently, see pp. 92-3 of [33].

34
It can be thought of as the quotient of the upper half plane H by (2). Indeed, as (2)
is of index six in SL(2, Z), the quotient H/(2) is a six-fold cover of the usual modular
domain. This quotient has three cusps, which we can take to be at 1, 1, and , with
precisely the monodromies in (6.1).
The family of curves parametrized by H/(2) can be described very explicitly, by the
equation

y 2 = (x 1)(x + 1)(x u). (6.6)

The idea here is that for every u, there is a genus one Riemann surface Eu , determined
by equation (6.6). This equation describes a double cover of the x plane branched over
1, 1, , and u. The curve Eu becomes singular when (and only when) two branch points
coincide. This occurs precisely for u = 1, 1, or .
Note among other things that (6.6) has a manifest symmetry w that maps u u,
x x, y iy. This generates a Z4 symmetry, but only a Z2 quotient acts on the u
plane. Indeed, P = w2 acts trivially on u and x while mapping y y. It will turn out
that a and aD are odd under y y and so odd under P . The symmetry structure just
described is precisely that of the field theory that we are aiming to solve.
The Riemann surface Eu has a two dimensional first homology group Vu = H 1 (Eu , C).
The Vu are fibers of a flat bundle V over the punctured u plane, and this is the bundle
of which the pair (aD , a) is a section. This is a convenient description of the bundle, as
we will see. The bundle V can be trivialized locally by picking a pair of independent
and continuously varying one-cycles 1 , 2 on Eu ; these can be normalized so that the
intersection number is

1 2 = 1. (6.7)

The space H 1 (Eu , C) can be thought of as the space of meromorphic (1, 0)-forms on
Eu of vanishing residues, modulo exact forms (or total derivatives). The heuristic idea is
that if is such a one-form, then it represents an element of H 1 (Eu , C) because it can be
paired with one-cycles by
I
. (6.8)

35
The condition that the residues of vanish ensures that this pairing is invariant under
deformation of even across a pole of . One identifies +dw (with w a meromorphic
function) since the exact differential dw would not contribute to the contour integral (6.8).
As H 1 (Eu , C) is two-dimensional, a basis is provided by any two linearly independent
elements, for instance
dx x dx
1 = and 2 = . (6.9)
y y
Here 1 is actually a holomorphic differential, having no poles even at infinity; it is up
to a scalar multiple the unique holomorphic differential on Eu . (In terms of the Hodge
decomposition of H 1 (Eu , C), it represents an element of H 1,0 .) If one picks on Eu a basis
of one-cycles normalized as in (6.7), then the periods
I
bi = 1 , i = 1, 2 (6.10)
i

obey
b1
= u , (6.11)
b2
with u the usual parameter of the elliptic curve Eu , which has the fundamental property

Im(u ) > 0. (6.12)

Under a change in the basis of s, u would be transformed by the standard action of

SL(2, Z) on the upper half plane. As for 2 , its only pole is a double pole at infinity
(where the residue vanishes, since in any case the sum of the residues of a meromorphic
differential is zero).
As (6.9) gives a basis of H 1 (Eu , C), an arbitrary section of the flat bundle Vu can be
represented in the form
= a1 (u)1 + a2 (u)2 . (6.13)

This representation is not necessarily convenient, as it may be convenient to use the freedom
of adding dw for some function w. In any event, once a section is chosen, to extract the
components aD and a of this section, one picks the basis 1 , 2 of one-cycles and defines
I
aD =
1
I (6.14)
a= .
2

36
If a different choice is made for the i , the pair (aD , a) will be transformed by an element
of SL(2, Z).
Now we would like to impose the condition that as defined in (6.5) has Im( ) > 0.
There is an obvious way to satisfy this condition. One has
I
daD d
=
du du
I 1 (6.15)
da d
= .
du 2 du

Suppose that
d dx
= f (u)1 = f (u) (6.16)
du y
with f (u) a function of u only. Then we get
daD
=f (u)b1
du (6.17)
da
=f (u)b2 ,
du
with bi defined in (6.10). Consequently,
daD /du b1
= = = u , (6.18)
da/du b2
and therefore Im( ) > 0. Conversely, if Im( ) > 0 everywhere, then = u and is as
above. To see this, note first if Im( ) > 0 everywhere, then for every u, (u) is the -
parameter of an elliptic curve. Also, the pair (daD /du, da/du) transforms under SL(2, Z)
the same way as (aD , a), and so gives a section of the same flat bundle. The family of curves
determined by u (u) consequently has the same monodromies (and singularities) as
the family Eu determined by u u . It follows on general grounds that these families
coincide.
It remains to select f and verify the desired asymptotic behavior of aD and a near

u = 1, 1, . We will in fact show that everything works out correctly with f = 2/4.
It will then be clear that this is the unique choice that works; in particular, a non-constant
f would somewhere introduce unwanted poles or zeroes.
First of all, given (6.16), there is no problem in finding . We can take

2 dx x u 2 dx y
= = . (6.19)
2 x2 1 2 x2 1

37

Obviously, the u derivative of this expression reproduces (6.16) with f = 2/4. The
residues of vanish since

2(2 u1 )
= (6.20)
2

with i as above. is odd under P = w2 , as the second formula in (6.19) makes clear,
and that is why P reverses the sign of a and aD as defined presently.

-1 1

Fig. 2: A non-trivial one-cycle on Eu comes from a contour in the x plane

that loops around two of the branch points. The cuts in the x plane are
indicated with dotted lines.

Now we need an explicit basis of one-cycles on Eu . These can be constructed as

follows. A circle C in the x plane that loops around two of the branch points lifts to
a non-trivial cycle on Eu . (Apart from any topological arguments, the non-triviality
of this cycle follows from the non-vanishing of the integral that we are about to write.)
Suppose, for instance, that as in figure 2 we take C to loop once around the branch points
at x = 1, 1. The differential form that we wish to integrate, viewed by the first formula
p
in (6.19) as a form on the x plane, has a factor (x u)/(x2 1) that requires branch
cuts. We can take the cuts to run from 1 to 1 and from u to , as shown in figure 2.

38
Then we can deform C so it runs just around the cut between 1 and 1. So
I I Z 1
2 dx x u 2 dx x u
= = . (6.21)
2 C x2 1 1 x2 1
A factor of two in the last step comes because the integral over C contains an integral
from 1 to 1 then from 1 back to 1 on the other side of the cut; the two segments make
equal contributions.
Now we can implement the definition of a and aD in (6.14). Choosing the contour
just described to be 2 , we have
Z 1
2 dx x u
a= . (6.22)
1 x2 1
Similarly, defining another cycle 1 by using a circle that loops around the branch points
at 1 and u, we can take Z u
2 dx x u
aD = . (6.23)
1 x2 1
Of course, this particular basis of one-cycles was picked with a view to getting the desired
behavior near the singularities in the u plane. The occurrence of a square root means that
the overall signs of a and aD are ill-defined. That is in keeping with the fact that, from
the outset, the classical relation a2 = 2u means that a can be recovered from the gauge
invariant quantity u only up to sign.
It will suffice to study the behavior of a and aD near u = and u = 1, since the
behavior near u = 1 is determined by the Z2 symmetry of the u plane that was described
earlier.
Near u = , we get Z 1
2u dx
a = 2u (6.24)
1 1 x2
and (after a change of variables x = uz)
Z 1
2u dz z 1
aD = . (6.25)
1/u z 2 u2
For u the integral develops a logarithmic divergence near z = 0; extracting the
divergent term, we get
2u ln u
aD i . (6.26)

39
Of course, this logarithm is the signal of asymptotic freedom in the semiclassical region of
large u.
The behavior near u = 1 is equally easy to determine. From (6.25), we get
Z 1 Z  
2u dz z 1 1 1 dz z 1 i 1 i(u 1)
aD = = 1 .
1/u z + u1 z u1 1/u z u1 2 u 2
(6.27)
What about a? At u = 1 the integral for a is:
Z 1
2 dx 4
a(u = 1) = = . (6.28)
1 x + 1

However, the derivative of a with respect to u is given by an integral

Z 1
da 2 dx
= p (6.29)
du 2 1 (x + 1)(x 1)(x u)

that for u 1 becomes logarithmically divergent near x = 1. Extracting the coefficient

of the logarithm, we find that the expansion of a is

4 (u 1) ln(u 1)
a= + .... (6.30)
2

Comparing (6.30) and (6.27), we get the desired monodromy a a 2aD near u = 1.
This completes our verification of the expected properties. But there is still one point
to discuss.
The Spectrum
The remaining point concerns, in a sense, the physical meaning of the duality that we
have used to solve the theory.
In looping around u = 1 or u = 1, the pair (aD , a) are transformed by monodromies
M1 and M1 . The charges (nm , ne ) are transformed similarly. Naively, one would think
that the spectrum of BPS-saturated states would be transformed by the monodromy ma-
trices. In that case, since the monodromies generate (2), the spectrum of BPS-saturated
states would be (2) invariant.
In fact, that is not true. In the semiclassical region of large u, the BPS-saturated states
are the electrons and W bosons of (nm , ne ) = (0, 1) and the dyons of (nm , ne ) = (1, n);
moreover the W bosons are chiral multiplets (spin 1) and the dyons are hypermultiplets

40
(spin 1/2). The fact that this spectrum is not duality invariant is precisely the reason
that it was concluded many years ago that Olive-Montonen duality did not hold for N = 2
super Yang-Mills theory.

From the discussion at the end of section 4, there is a possible mechanism for curing
the problem. The spectrum of BPS-saturated states can jump on crossing a curve in the
u plane on which the ratio aD /a is real. If a curve on which aD /a is real passes through
u = 1, then in looping around u = 1, one would have to cross that curve and the resulting
jumping of the spectrum would invalidate conclusions based on the monodromies.

It is easy to see that this situation does arise. Near u = 1, a is nearly real - in fact
a(u = 1) = 4/. But aD can have an arbitrary phase near u = 1 since aD i(u 1)/2.
Thus, jumping can occur on a curve that near u = 1 looks like u = 1 + it, t real, where
aD /a is real. A similar curve on which aD /a is real passes through u = 1.

It is consistent with everything we know, and will resolve all the puzzles about lack
of duality in the spectrum, if the curve on which aD /a is real looks something like |u| = 1.
Then one could avoid the jumping phenomenon only if one stays in the region u > 1; the
only monodromy that can be seen in that region is M , under which the spectrum of
BPS-saturated states is indeed invariant. However, we do not know a practical way to
determine the curve on which aD /a is real.

There is, however, one important region in which one can easily prove that aD /a is not
real. This is on the real u axis for |u| > 1, as one can easily see from (6.22) and (6.23). For u
real and |u| > 1, the quantity (x u)/(x2 1) is positive in the integration region for a and
negative in the integration region for aD , so a is real and aD is imaginary. In particular, by
sticking to the real u axis, one can come in from the semiclassical region of large u to the
singularities at u = 1 without crossing any jumping curves. Therefore, whatever particles
become massless at u = 1 must evolve continuously from the BPS-saturated states that
can be seen semiclassically near infinity. So for our picture of the strong coupling region
to make sense, the monopoles and dyons that we need for the singularities must exist in
the semiclassical region. Happily, they do.

41
 Acknowledgements
We would like to thank T. Banks for a useful discussion on confinement and conden-
sation of magnetic monopoles, P. Deligne for some conceptual assistance, G. Moore for
suggesting a possible role of hypergeometric functions, and C. Vafa for explaining certain
two-dimensional results. We have also benefitted from discussions with K. Intriligator,
R. Leigh, and S. Shenker. This work was supported in part by DOE grant #DE-FG05-
90ER40559 and in part by NSF grant PHY92-45317.

42
 References

[1] N. Seiberg, Phys. Lett. 318B (1993) 469.

[2] N. Seiberg, hep-th/9402044, Phys. Rev. D49 (1994) 6857.
[3] K. Intriligator, R. Leigh and N. Seiberg, hep-th/9403198, RU-94-26, Phys. Rev. in
press.
[4] V. Kaplunovsky and J. Louis, hep-th/9402005.
[5] M.A. Shifman and A.I. Vainshtein, Nucl. Phys. B277 (1986) 456; Nucl. Phys. B359
(1991) 571.
[6] D. Amati, K. Konishi, Y. Meurice, G.C. Rossi and G. Veneziano, Phys. Rep. 162
(1988) 169 and references therein.
[7] N. Seiberg and E. Witten, to appear.
[8] E. Witten, Comm. Math. Phys. 117 (1988) 353.
[9] E. Witten, IASSNS-HEP-94-5, hep-th/9403195.
[10] M. K. Prasad and C. M. Sommerfield, Phys. Rev. Letters 35 (1975) 760.
[11] E. B. Bogomolnyi, Sov. J. Nucl. Phys. 24 (1976) 449.
[12] E. Witten and D. Olive, Phys. Lett. 78B (1978) 97.
[13] C. Montonen and D. Olive, Phys. Lett. 72B (1977) 117; P. Goddard, J. Nuyts, and
D. Olive, Nucl. Phys. B125 (1977) 1.
[14] H. Osborn, Phys. Lett. B83 (1979) 321.
[15] J. Cardy and E. Rabinovici, Nucl. Phys. B205 (1982) 1; J. Cardy, Nucl. Phys. B205
(1982) 17.
[16] A. Shapere and F. Wilczek, Nucl. Phys. B320 (1989) 669.
[17] A. Font, L. Ibanez, D. Lust, and F. Quevedo, Phys. Lett. B249 (1990) 35.
[18] A. Sen, TIFR-TH-94-08, hep-th/9402002.
[19] G. Segal, to appear.
[20] L. Girardello, A. Giveon, M. Porrati and A. Zaffaroni, NYU-TH-94/06/02, IFUM/472/FT,
SISSA 80/94/EP, RI-6-94, hep-th/9406128.
[21] C. Vafa and E. Witten, to appear.
[22] N. Seiberg, Phys. Lett. 206B (1988) 75.
[23] B. De Wit and A. Van Proeyen, Nucl. Phys. B245 (1984) 89.
[24] S. Ferrara, Mod. Phys. Lett. A6 (1991) 2175.
[25] A. Strominger, Commun. Math. Phys. 133 (1990) 163.
[26] P. Candelas and X. de la Ossa, Nucl. Phys. B355 (1991) 455.
[27] S. Cecotti, P. Fendley, K. Intriligator, and C. Vafa, Nucl. Phys. B386 (1992) 405; S.
Cecotti and C. Vafa, Commun. Math. Phys. 158 (1993) 569.
[28] J. Wess and J. Bagger, Supersymmetry and Supergravity (Princeton University Press,
1982).
[29] For a review, see A. Giveon, M. Porrati and E. Rabinovici, RI-1-94, hep-th/9401139.

43
[30] E. Witten, Phys. Lett. 86B (1979) 283.
[31] N. Seiberg, to appear.
[32] E. Witten, Nucl. Phys. B202 (1982) 253.
[33] C. H. Clemens, A Scrapbook Of Complex Curve Theory (Plenum Press, 1980).

44