Dynamics of N = 1 Supersymmetric Field

Theories in Four Dimensions

Nathan Seiberg

Introduction
The exact solutions of supersymmetric eld theories have four distinct appli-
cations:
1. They teach us about the dynamics of strongly coupled eld theories. Here
the main lesson is the role of electric-magnetic duality in the dynamics.
2. The ideas from eld theory have extensions to string theory whose dynamics
is richer. This has already led to a revolution in our understanding of string
theory.
3. Supersymmetry might be discovered experimentally in the next generation
of experiments. In this case, the dynamics of supersymmetric eld theories
is of direct phenomenological interest, especially for the question of how
supersymmetry is spontaneously broken leading to a non-supersymmetric
spectrum.
4. Witten's topological eld theory is based on supersymmetry. Understanding
these theories has applications to mathematics.
In these lectures we will focus on the rst application. Other applications were
described by other lecturers.
Exact solutions play a crucial role in physics. It is often the case that a simple
model exhibits the same phenomena which are also present in more complicated
examples. The exact solution of the simple model then teaches us about more
generic situations. The main point of these lectures is to show how four dimensional
supersymmetric quantum eld theories can play a similar role as laboratories and
testing grounds for ideas in more generic quantum eld theories. This follows from
the fact that these theories are more tractable than ordinary, non-supersymmetric
theories, and many of their observables can be computed exactly. Yet, it turns out
that these theories exhibit explicit examples of various phenomena in quantum eld
theory. Some of them had been suggested before without an explicit realization and
others are completely new.
The generic eld theory in four dimensions has no supersymmetry at all. With
current techniques its dynamics is dicult to control. With more supersymmetry
Notes by Siye Wu

1
2 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
the theories are less generic. The most constrained theories have N = 4 super-
symmetry. A more general class of theories has N = 2. An even more general
class has the minimal amount of supersymmetry: N = 1 supersymmetry. These
lectures are devoted to the analysis of this largest class. (The theories with N = 1
supersymmetry are also the relevant ones for the third application above|particle
physics phenomenology.) Since this class includes more generic theories than the
more special N = 2 or N = 4 theories, the dynamics which is being exhibited
is richer; it includes phenomena which are impossible with more supersymmetry.
Therefore these theories are more useful for our rst application. They are the most
generic, yet still amenable to exact analysis.
In Lecture 1 we discuss the simplest theories (Wess-Zumino models) and prove
their non-renormalization theorem. We will then turn to a discussion of gauge
theories. The key feature is the existence of at directions of the classical potential
leading to a moduli space of classical ground states. In Lecture 2 we analyze
some of the quantum eects in these gauge theories. We show how the moduli
space of classical vacua can disappear (by generating a potential) in the quantum
theory. Alternatively, the moduli space is deformed. We will also see an example
where the space is not deformed, but the massless particles at the singularities
are dierent in the quantum theory than in the short distance classical theory.
Finally, in Lecture 3 we complete the analysis of these theories by uncovering a new
phenomenon|electric-magnetic duality in N = 1 theories. It generalizes previously
noticed dualities in N = 4 and N = 2 theories and explains all the phenomena which
we discuss in the earlier lectures.
 LECTURE 1
Basic Aspects of N = 1 QCD
We will study various supersymmetric theories in 4 dimensions, i.e., the un-
derlying bosonic spacetime is the Minkowski space R1;3. Supersymmetry put con-
straints on the geometry setting of the eld theory. In fact, N = 0; 1; 2; 4 supersym-
metry corresponds to having a target space of Riemannian, Kahler, hyperKahler
(or special Kahler) and at geometry, respectively.
x1.1. Wess-Zumino model
Wess-Zumino model is a simple example to demonstrate the principle of holomorphy
and the non-renormalization theorem. It is an N = 1 supersymmetric theory on
R1;3j4 with a multiplet of chiral superelds  valued in a complex vector space R.
In components, we have
(1.1)  =  +  + 2 F;
where , and F are complex boson, Weyl fermion and the auxiliary eld, respec-
tively, all valued in R. The most general Lagrangian (density) with at most two
derivatives
Z is Z 
L = d4  K(; )  + d2 W() + c:c:
(1.2) = g(@; @) + g(6 @ ; ) + g(F; F)
 + (Hessian(W) ( ; ) + @W (F) + c:c:):
Here K is a Kahler potential on R, g, the corresponding metric, and W, a holo-
morphic function on R called the superpotential. After integrating out the aux-
iliary eld F by its (algebraic) equation of motion, we get a bosonic potential
V = g(@W; @W ). Supersymmetry is unbroken if and only if there exists a solution
to the equation @W = 0.
We consider the following examples. In all the cases considered, let K be a
Kahler potential on R such that the metric is at.
Example 1.  is a single chiral supereld valued in C and W = 21 m2 (m 2 C ).
This is a free theory in which the boson  and the fermion are of equal mass
jmj. The free action contains terms m + jmj2jj2. The phase of m 2 C , and
more generally that of W, can be rotated away by a U(1) rotation of the fermionic
coordinate  (called the R-symmetry). We denote this U(1) group by U(1)R . By
(1.1), the U(1)R rotation produces a redenition of in component language.
3
4 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
Example 2.  is valued in C and W = 12 m2 + 13 3 (m;  2 C ,  =6 0). This is
the original Wess-Zumino model with the Yukawa coupling  and the quartic
interaction jj2jj4. The equation @W = m + 2 = 0 has two solutions:  = 0
and  = m , corresponding to two classical ground states. The bosonic potential of
 is shown in Figure 1.1. Under the transformation  7! m  which interchanges
the two ground states, W 7! W up to an additive constant. The sign of W can
be rotated away by a U(1)R symmetry which maps to i . Therefore we have a
Z4 discrete symmetry spontaneously broken to Z2 : the generator of Z4 relates the
two ground states 0 and m ; its square acts as a 2 rotation in Spin (1; 3) and is
therefore unbroken.
V( φ )

m 0 φ
λ

Figure 1.1. The bosonic potential of Example 2, with two minima

Example 3.  consists of two complex valued elds L and H (which stand for
light and heavy, respectively). The superpotential is W = 2 LH 2 . The minimum of
the bosonic potential is achieved at H = 0 and arbitrary L. The eld L is massless
(light) whereas H has mass 2 hLi; H is heavy except when hLi = 0. The moduli
space of vacua is shown in Figure 1.2 These (classical) vacua are inequivalent -
there are no symmetries that relate them. The metric on the L-space is given
by Ke = L L, which is the restriction of the original Kahler potential K to the
L-space.

L-plane

L=0

Figure 1.2. The moduli space of vacua of Example 3 with H being massless
at L = 0
Example 4.  consists of three chiral superelds X, Y and Z, and the superpo-
tential is W = XY Z. The minimum is achieved at X = Y = 0, Y = Z = 0 and
 LECTURE 1. BASIC ASPECTS OF N = 1 QCD 5

Z = X = 0. The moduli space of vacua has three branches, as shown in the Figure
1.3. Even in the classical theory, there is a singularity X = Y = Z = 0, where new
massless elds appear.

Z O X

Figure 1.3. The singularity of the moduli space in Example 4

At the quantum level, all theories considered above are renormalizable (but
not asymptotically free) and make sense perturbatively in 4 dimensions. The non-
renormalization theorem says that the superpotential is not changed by the quan-
tum eects. Although the Wess-Zumino model in 4 dimensions is not an interesting
theory non-perturbatively and the standard way to prove the theorem in pertur-
bation theory is to use Feynman diagrams, we are looking for a conceptual and
non-perturbative proof which will shed light on other theories.
We need a digression on two notions of eective actions. First, the 1PI (one-
particle irreducible) eective action is obtained by adding sources to the (classical)
Lagrangian, performing the entire path integral, and doing a Legendre transfor-
mation. This procedure fails if the integral diverges in the infrared. On the other
hand, in dening the Wilsonian eective action, we separate a eld, say , into
the
R Dhigh Renergy modes H and the low energy
R modes L in the path integral, i.e.,
= DH DL , do the integration DH only, and obtain a Lagrangian for
L. Here integrating over H is putting a cut-o. The infrared diculty is absent
in the Wilsonian eective action. When there are no interacting massless elds, the
two notions are identical.
The eective action is subject to constraints of symmetries present in the classi-
cal theory. In addition, supersymmetry requires that the superpotential is holomor-
phic in the eective elds. This is the principle of holomorphy. Finally, the eective
action is constrained at various limits where the theory is in the perturbative or
classical regimes.
A key technique in our proof of non-renormalization theorem is to think of
all coupling constants, such as m,  in Example 2, as expectation values of some
background superelds. (We assume that the extended theory make sense and
is still supersymmetric at the quantum level.) As a result of holomorphy, the
superpotential is holomorphic not only in the eective elds, but also in the coupling
constants.
We consider Example 2, in which the superpotential is W = 21 m2 + 31 3
classically. To nd the (Wilsonian) eective superpotential, we regard m,  as
background superelds and assign the following U(1)  U(1)R charges to the elds:
6 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES

U(1)  U(1)R
(1.3)  1 1
m 2 0
 3 1
U(1)R is the rotation on the fermionic coordinates in superspace. W is invariant
under the rst U(1) factor and has U(1)R charge 2, as required by supersymmetry.
By holomorphy, We is holomorphic not only in , but also in m, . (If super-
symmetry was unbroken for W() but broken for W(; m; ), this argument would
fail.) So
We (; m; ) = f(m2; 3) (from the rst U(1) symmetry)
(1.4) = m2 h m (from R-charge of W is 2)
for some yet unknown holomorphic functions f and h. We can now forget that m,
 are background elds. Next, we consider the limits. If jj << 1, perturbation
theory should be valid. So
  1
2 + 1 3 + X a  
1 n n+2
(1.5) m2 h m = 2 m 3 n mn 1 :
n=2
The n-th term must come from the diagram in Figure 1.4 with n vertices and n+2
external legs, with each internal line contributing a factor of m 1 . No loop could
appear because each loop would increase the power of  by 1 but not that of .
However for n  2, the diagram is not 1PI and therefore can not contribute. So
We = W, i.e., there is no quantum correction to the superpotential. This is the
non-renormalization theorem. Notice however there is no claim about the quantum
correction to the metric K.

φ φ φ

φ m -1 φ

Figure 1.4. The diagram for the n-th term in formula (1.5)

Theories with more elds behave similarly. Consider a variant of Example 3, a

theory with two chiral superelds L, H and a superpotential W = 12 mH 2 + L2 H.
Classically, the minimumis reached at @W 2 @W
@H = mH +L = 0, @L = 2LH = 0, i.e.,
 LECTURE 1. BASIC ASPECTS OF N = 1 QCD 7

L = H = 0. So L is a light eld. Fixing the value of L, the minimumof W is reached

at H = Lm , and we get We = 2m L4 . This procedure is called integrating out
2 2

the heavy eld H. The eective superpotential We corresponds to the tree diagram
in Figure 1.5. Non-renormalization theorem in this case says that W is renormalized
by the tree diagram only and not by loop diagrams. In fact, we again regard m, 
as background elds and assign the following U(1)H  U(1)L  U(1)R charges to
the elds

U(1)H  U(1)L  U(1)R

H 1 0 1
(1.6) L 0 1 1
m 2 0 0
 1 2 1
Invariance under U(1)H  U(1)L and holomorphy imply that
 2 4
(1.7) We (L; m; ) = h  mL
for some holomorphic function h. Using the fact that We has R-charge 2 or mass
dimension 3, we conclude that We  m L4 . 2

L L

L L

Figure 1.5. The tree diagram that produces the interaction L4

Several remarks are in order.
1. Supersymmetry is crucial in the above arguments; without holomorphy, we
can not rule out terms such as e 1= .
2. Strictly speaking, we need to show the existence of a regularization scheme
that respects all the symmetries. Here we simply assume its existence with-
out discussing the details.
3. These arguments are valid both in perturbation theory and non-perturba-
tively.
4. If supersymmetry is unbroken classically, it is unbroken quantum mechani-
cally because We = W.
8 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
5. In usual quantum eld theories, we have to include in the Lagrangian all
terms compatible to renormalizability and the symmetries. For supersym-
metric theories, this is not necessary because renormalization does not gen-
erate new terms in the superpotential.
6. Supersymmetry is the only known way to have widely separated scales in a
theory which are not ruined by quantum corrections. Therefore the gauge
hierarchy problem could be explained if supersymmetry holds in Nature.
7. The kinetic term is usually very complicated. However, in N = 2 theories
(for example, N = 2 super Yang-Mills), it is possible to determine the kinetic
term by an extra supersymmetry.
8. We will see examples of supersymmetric gauge theories in which the non-
renormalization theorem can be avoided: new terms are generated non-
perturbatively. Even so, the constraints of symmetries and holomorphy are
strong enough to determine the superpotential completely.
x1.2. Pure supersymmetric gauge theory
Recall that an N = 1 gauge eld is a connection (with some constraints on the
curvature) on a principal G-bundle P over the superspace R1;3j4. In components, a
constrained connection on the superspace consists of a gluon eld A (a connection
on the ordinary spacetime R1;3) and a gluino eld  (a section of the spinor bundle
over R1;3 twisted by the adjoint bundle). Let W be the supereld strength and
F, the usual eld strength (curvature of A). The Lagrangian density is
(1.8) L = 4i Z d2 tr W W  + c:c: = 1 (tr F ^ F +  i D 6 ) i tr F ^ F;
 g 2 82
where g is the coupling constant,  is the -angle, and  = 2 + 4gi .
2

Classically, there is a U(1)R symmetry. Its action on the gluino eld  is the
chiral rotation  7! ei  (ei 2 U(1)R ). Here 5 is 1 on the positive (negative)
5

chiral spinors, respectively, and can be regarded as an operator acting on the space
of sections of the twisted spinor bundle. The associated current j5 has an anomaly
(1.9) @ j5 = c32
2(G)  tr F ^ F;
2
which perturbatively comes from the Feynman diagram in Figure 1.6. Here c2(G)
is the second Casimir of the gauge group G. Numerically, it is equal to 2h, h being
the dual Coxeter number of G. Consequently, U(1)R is not a symmetry at the
quantum level. This is called the axial anomaly. In the path-integral language, a
classical symmetry survives at the quantum level if it preserves the path-integral
measure in addition to the classical action. Under ei 2 U(1)R , the measure D
transforms as
(1.10) D 7! D (det ei ) 1 = D e i tr = D e i ind 6D :
5 5 adj

Since the index of the Dirac operator twisted by the adjoint bundle is always an
integral multiple of c2 (G) = 2h, the measure D is invariant only under the Z2h
subgroup of U(1)R . Thus the symmetry that remains at the quantum level is Z2h .
For G = SU(Nc ) (Nc is called the number of colors), h = Nc and c2 = 2Nc . Hence
the symmetry is Z2Nc .
 LECTURE 1. BASIC ASPECTS OF N = 1 QCD 9

λ
g
λ
A

Figure 1.6. A Feynman diagram that causes the axial anomaly

The standard lores are that the theory has a mass gap, that it has connement,
i.e., hW(C)i  e Area for a Wilson loop in the fundamental representation, and that
hi =6 0. The non-zero value of hi spontaneously breaks the Z2h symmetry to
Z2 . The non-trivial element of Z2 can be identied as the 2 rotation in Spin (1; 3)
because its action is equal to ( 1)F . Therefore this Z2 subgroup is unbroken.
Consequently the theory has h vacua (Nc if G = SU(Nc )). In the k-th vacuum,
(1.11) hik  e hi k 3;
2

where  is the mass scale generated by renormalization, which we now explain.

Due to the regularization procedure, a quantum eld theory depends not only
on the classical Lagrangian, but also on a cuto . If we change  but keep the
theory xed, the coupling constant g then depends on . The change of g is governed
by
(1.12) dg = (g);
 d
where the (g) is the -function. Up to 1-loop,
(1.13) (g) = (4)b0 g3
2
for some number b0 . If b0 6= 0, a renormalization group invariant mass scale  given
by the equation
  b0 8 2
(1.14)  = e g ( )2

is generated. This is called dimensional transmutation. Notice that  is intrinsic

to the theory; it is not a cuto. Because  has mass dimension 1, the quantum
theory is not scale invariant even if the classical theory is.
In gauge theories, it is convenient to dene  as a complex number satisfying
  b
g  +i = e2i () :
0 
8 2
(1.15)  = e ( )2

The exponent in (1.15) is simply the value of the classical action for one-instanton.
Moreover, any quantity holomorphic in , which appears in the Lagrangian density
(1.8), is also holomorphic in . Of course  dened by (1.15) is not single valued.
If a physical quantity expressed in term of  is not single valued, it is often because
there are many vacua in the theory. For N = 1 pure gauge theory, b0 = 32 c2 (G) =
10 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
3h. 3 is h-valued and there are h vacua in the theory. Fixing a choice of 3 , hi
takes values according to (1.11) in the h vacua.
We need
H a digression on connement. Recall that the Wilson loop WR(C) =
trR P exp C A is the trace in the representation R of G of the holonomy around
a closed curve C. For the adjoint representation, we always have hWadj (C)i 
e Perimeter(C ) . This is because it is possible to screen the charges by the gluons. So
the potential is independent of the distance. Likewise for a theory with fundamental
quarks that transform in a representation where the center of G acts faithfully, we
have hWR (C)i  e Perimeter(C ) for any representation R because the quarks can
screen any charges. In both cases, the Wilson loop obeys the perimeter law and
the charges are not conned. However in a theory without fundamental quarks, the
Wilson loop in fundamental representation may exhibit connement, in which case
we have the area law hWR (C)i  e Area(D), where D is the disk bounded by C.

x1.3. Supersymmetric QCD

Supersymmetric QCD is a theory of N = 1 matter coupled to N = 1 gauge theory.
It is labeled by a compact Lie group G, a complex representation R of G (where
the quarks take value), and a classical superpotential W , which is a G-invariant
function on the representation space R. Recall that we have a principal G-bundle
P ! R1;3j4 in N = 1 gauge theory. Let R be an Hermitian vector space with a
unitary representation of G. An N = 1 matter (or quark) eld Q is a chiral section
of the associated bundle P G R ! R1;3j4. The superpotential W (Q) may contain
a mass term for Q. It is at most cubic in Q if the theory is to be renormalizable.
Our goal is to nd the physics of the theory and its dependence on the bare
mass. Does the theory have connement? Which phase is the theory in? If the
theory ow to a non-trivial infrared xed point (a conformal eld theory in four
dimensions), what are the anomalous dimensions of the operators?

x1.3.1. Classical theory

We start with the case W = 0. Since Q couples to the N = 1 gauge theory,
the potential for the bosonic part of Q (also denoted by Q) is V (Q) = jD(Q)j2 ,
where D is the moment map of the Hamiltonian G-action on R. The minimum
is reached at Q 2 D 1 (0) (which is non-empty if G is semi-simple). Modulo the
gauge transformations (the G-action), the moduli space of classical vacua M is the
symplectic quotient D 1 (0)=G, which is an open dense set in the complex quotient
R=GC.
In the lectures, we will consider theories with the gauge group G = SU(Nc ). Let
C be the fundamental representation (or the dening representation) of G. C is an
Hermitian vector space of dimension Nc with a complex volume form vc preserved
by G = SU(C) (and GC = SL(C) as well). The matter eld consists of Nf copies
of quarks in the fundamental representation C and Nf copies of anti-quarks in the
conjugate representation C  = C  . Nf is called the number of avors. There are
altogether 2Nf Nc chiral superelds. It is convenient to introduce two Hermitian
vector spaces F, F 0 of complex dimension Nf and let R = Hom(F; C)  Hom(C; F 0).
The values of the quark, anti-quark elds are Q 2 Hom(F; C), Q0 2 Hom(C; F 0),
respectively, or (Q; Q0) 2 R.
 LECTURE 1. BASIC ASPECTS OF N = 1 QCD 11

The action of the group G = SU(C) on R is Hamiltonian and the moment map
D: R ! su(C) is
(1.16) hD(Q; Q0 ); X i = trF Qy XQ trF 0 Q0XQ0y
for any X 2 su(C). Under the isomorphism su(C)  = su(C), D takes values in
su(C) and is given by
(1.17) D(Q; Q0) = traceless part of i(QQy Q0yQ0 ):
So (Q; Q0) 2 D 1 (0) if and only if QQy Q0y Q0 = rIC for some r 2 R. For
Nf < Nc , we have r = 0. There exist unitary bases of C, F, F 0 such that
0a 1
1
(1.18) QB @ ... CA  Q0y
aNf
for some ak  0 (1  k  Nc ). Generically, these ak > 0 and the gauge group
SU(Nc ) is broken to SU(Nc Nf ), the isotropy group of (Q; Q0). For Nf  Nc , r
can be non-zero. There exist unitary bases such that
0a 1 0 a0 1
1 1
B ... CC 0y BB . . . CC
(1.19) QB B@ CA ; Q  B@ C
aN c a0Nc A

for some ak  0, a0k  0 (1  k  Nc ) satisfying a2k a0k 2 = r (independent of k).

Generically, the gauge group is completely broken.
We now construct gauge invariants from Q and Q0. First, we have the meson
eld M = Q0Q, which is clearly invariant under SL(C). If Nf  Nc , then we have
additional invariants B = Q vc 2 ^Nc F  and B 0 = Q0 vc 2 ^Nc F 0, called the
baryon elds. They are subject to the constraints
(1.20) B; B 0 are decomposable
and
(1.21) ^Nc M = B
B 0 :
If Nf > Nc , then further constraints
(1.22) M ^ B = 0; B 0 ^ M = 0
shall be satised. In particular, if Nf = Nc , then (1.20) is vacuous and (1.21)
reduces to
(1.23) det M = B
B 0 :
If Nf = Nc +1, then B, B 0 can be identied as elements of F
^Nf F , ^Nf F 0
F 0 ,
respectively. In this case, (1.20) is again vacuous and (1.22) reduces to
(1.24) MB = 0; B 0 M = 0:
The gauge invariant quantities M, B and B 0 subject to the constraints (1.20),
(1.21) and (1.22) provide a good coordinate system on an open dense subset of
R=SL(C) and hence of M. This is because, rst, there exist Q, Q0 such that M =
12 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
Q0Q, B = Q vc and B 0 = Q0 vc if M, B, B 0 satisfy these constraints,1 and secondly,
generically, the gauge invariants M and B, B 0 (if Nf  Nc ) determine Q, Q0 up to
a (complexied) gauge transformation.2 As a consistency check, when Nf < Nc ,
the complex dimension of a generic GC-orbit in R is dimC (SL(Nc )=SL(Nc Nf )) =
2Nf Nc Nf2. Therefore the quotient M has complex dimension dimC R (2Nf Nc
Nf2) = Nf2 and is labeled by Nf2 gauge invariants M.
x1.3.2. Quantum theory
With an equal number of quarks and anti-quarks in the representation
R = Hom(F; C)  Hom(C; F 0), the quantum theory is free from gauge and gravita-
tional anomalies. (We shall come to this point later.) But the quantum theory is
dierent from the classical theory in several aspects.
First, classically the global symmetry is U(F)  U(F 0 )  U(1)X , where U(1)X
is the U(1) rotation on the fermionic coordinate  in the superspace R1;3j4. (The
notation U(1)R is reserved for an anomaly-free group.) The group U(1)X and
the U(1) subgroups of U(F) and U(F 0) are anomalous at the quantum level. Let
U(1)A , U(1)B be the anti-diagonally and diagonally embedded U(1) subgroups in
U(F)  U(F 0), respectively. The weights of these U(1) groups on the fermionic elds
and on the path-integral measure (in k-instanton background), calculated similarly
as in (1.10), are given by

U(1)A  U(1)B  U(1)X

Q 1 1 1
(1.25) Q0 1 1 1
 0 0 1
DD Q D Q0 2Nf k 0 2(Nc Nf )k

So the group U(1)B is anomaly free. Let U(1)R be a linear combination of U(1)A
and U(1)X under which the fermionic measure in (1.25) is invariant. The generator
is chosen so that it projects to that of U(1)X . Thus a weight of U(1)R is fractional
in general and is related to those of U(1)A and U(1)X by
(1.26) R = X NfN Nc A:
f
1
This is obvious if Nf < Nc , when B , B 0 do not exist and if Nf = Nc , when the constraint is
(1.23). If Nf > Nc , then B (if non-zero) determines a proper subspace F0  F of dimension Nc .
On can choose an isomorphism Q0 : F0 ! C such that ^Nc Q0 = B , let Q0 = MQ0 1 and extend
Q0 to Q on F by setting it to be 0 on F0? .
2
When Nf < Nc , then generically ker Q = 0, imQ is a proper subspace in C , imQ0 = F 0
and C = imQ  ker Q0 . If Q01 Q1 = Q0 Q for generic variables, then there is an invertible map
U : imQ ! imQ1 such that Q1 = UQ. Therefore Q01 U = Q0 on imQ. Extend U by adding an
invertible map ker Q0 ! ker Q01 such that U 2 SL(C ). Then Q01 U = Q0 on C . When Nf  Nc ,
then generically imQ = C and ker Q0 = 0. If Q01 Q1 = Q0 Q, Q1 vc = Q vc , Q01 vc = Q0 vc for
generic variables, then there exists U 2 GL(C ) dened as follows: for any x 2 C , choose y 2 F such
that Qy = x, and set Ux = Q1 y. U is well-dened because ker Q = ker Q1 (from Q1 vc = Qvc ).
Since Q1 vc = det U Q vc , we conclude that U 2 SL(C ). It is easy to see Q01 = Q0 U 1.
 LECTURE 1. BASIC ASPECTS OF N = 1 QCD 13

The global symmetry of the quantum theory is SU(F)  SU(F 0 )  U(1)B  U(1)R .
The matter elds Q, Q0 (together with their composites M, B, B 0 ) transform
according to

SU(F)  SU(F 0 )  U(1)B  U(1)R

Q F 1 1 Nf Nc
Nf
Q0 1 F0 1 Nf Nc
(1.27) Nf
M F F0 0 2 NfNfNc
B ^Nc F  1 Nc Nc NNfc
2

B0 1 ^Nc F 0 Nc Nc NNcf
2

Here 1 is the trivial representation. The representations of the U(1) groups are
labeled by their weights.
Secondly, whereas the classical Lagrangian depends only on a dimensionless
coupling constant g, we need to introduce a mass scale  to dene the quantum
theory; this is due to dimensional transmutation discussed earlier. The running of
the coupling constant is governed by the -function as in (1.12) and (1.13). For
N = 1 supersymmetric QCD with gauge group G and matter representation R,
(1.28) b0 = 23 c2 (G) 12 c2 (R);
where c2 (R) is the value of the Casimir operator in R. In the theories we are
considering, G = SU(C) and R = Hom(F; C)  Hom(C; F 0). So b0 = 3Nc Nf .
The theory is asymptotically free if Nf < 3Nc . b is a scalar under U(1)B .
0

However, since the U(1) subgroups of U(F) and U(F 0) shift the -angle by units
of Nf and Nf , respectively, it is more appropriate to regard b in (1.15) as an
0

element in a suitable complex line, i.e.,

(1.29) b 2 ^Nf F 
^Nf F 0:
0

Thirdly, we have a moduli space of classical vacua M and want to know whether
the vacuum degeneracy will be lifted after quantization. There are several possi-
bilities. The space of vacua could disappear or be modied. If the moduli space
remains the same, the metric on it could be changed. We will see examples of each
case.
Consider the region in M where all ak >> . The gauge group SU(Nc ) is
broken to SU(Nc Nf ) if Nf < Nc 1 and is completely broken if otherwise. The
light elds are the massless elds M and B, B 0 (if Nf  Nc ) and the N = 1 pure
SU(Nc Nf ) gauge elds (if Nf < Nc 1), which are uncoupled to the former.
(Gauge bosons in the directions complement to SU(Nc Nf ) have acquired masses
of the order of ak through spontaneous symmetry breaking.) Since the N = 1
SU(Nc Nf ) gauge theory has Nc Nf vacua, we obtain an Nc Nf cover of
the space M. The N = 1 pure gauge theory has a mass gap. Going to a lower
energy, we have only M, and B, B 0 (if Nf  Nc ), which are combinations of Q, Q0 .
We wish to nd out whether an eective superpotential We could be generated
for these light elds. The only holomorphic combination of Q, Q0 that is invariant
under both the gauge and the global SU(F)  SU(F 0 )  U(1)B symmetry is det Q0 Q.
Since We has U(1)R charge 2, we get We  (det Q0Q) 1=(Nc Nf ) . Since We has
14 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
mass dimension 3 and the only parameter in our theory which has a mass dimension
is , we get
 3Nc Nf  Nc Nf
We (Q; Q0) = bNc ;Nf det Q0Q
1

(1.30)
for some coecient bNc ;Nf . (Both 3Nc Nf = b and det Q0Q are in ^Nf F 

^Nf F 0. Therefore their ratio is a number.) By holomorphy, (1.30) should also

be valid in M away from innity. We emphasize here that (1.30) is derived using
constraints of global symmetry, holomorphy and the limit far out in the moduli
space M. Note that the form of an eective potential need not be constrained by
the renormalizability condition.
For Nf > Nc , det Q0 Q = 0. For Nf = Nc , the exponent in (1.30) is singular.
So We = 0 in both cases. Thus the vacuum degeneracy is not lifted, and we have a
moduli space of vacua in the quantum theory. However, the quantum moduli space
may be modied away from innity. For Nf < Nc , we have a superpotential (1.30)
6
compatible with all symmetries; this is unlike the Wess-Zumino model. If bNc ;Nf =
0, then there is no non-renormalization theorem which is valid non-perturbatively
for supersymmetric theories with non-Abelian gauge elds.
In N = 2 gauge theory, the N = 1 supersymmetry already forbids a superpo-
tential. In fact, the pure N = 2 gauge theory is an N = 1 supersymmetric QCD
with matter eld Q in the adjoint representation. Since the fermions  and Q are
both in adjoint representation, and have opposite charges under the U(1)R rotation
in N = 1 superspace, this U(1)R is anomaly-free. So Q has R-charge 0 and it is
not possible to have a superpotential holomorphic in Q that has an R-charge 2.
 LECTURE 2
Quantum Behavior of Super QCD: Nf Small
We review some aspects of N = 1 supersymmetric QCD with Nc number of
colors and Nf number of avors covered in the last lecture. The gauge group is
G = SU(C) where C is an Hermitian vector space of dimension Nc equipped with
a complex volume form vc . The matter (quark) elds Q and Q0 take values in
Hom(F; C) and Hom(C; F 0), respectively. Here F, F 0 are Hermitian vector spaces
of dimension Nf . Under SU(C), the quarks Q, Q0 transform as C and C   
= C,
respectively. The (anomaly-free) global symmetry is SU(F)  SU(F 0 )  U(1)B 
U(1)R , where U(1)B is the diagonally embedded U(1) subgroup in U(F)  U(F 0 ),
and U(1)R is a linear combination of the anti-diagonal U(1) subgroup in U(F) 
U(F 0) and the U(1) rotation in the superspace. Under this global symmetry group,
Q, Q0 transform as (1.27).
The moduli space Mc of classical vacua can be parametrized by gauge invariant
quantities M = Q0 Q and (if Nf  Nc ) B = Q vc , B 0 = Q0 vc constructed from
Q, Q0 . In fact, an open dense subset of Mc is equivalent to fM g if Nf < Nc ,
fM; B; B 0 j det M = B
B 0 g if Nf = Nc , and fM; B; B 0 j ^Nc M = B
B 0 ; MB =
B 0 M = 0g if Nf = Nc + 1. This classical moduli space may be modied in the
quantum theory. We showed using constraints of global symmetry, holomorphy
that for Nf  Nc , no superpotential can be dynamically generated, whereas for
Nf < Nc , the only possible form is (1.30). The central question question is whether
the coecient bNc ;Nf is non-zero when Nf < Nc . We shall give an outline of the
computation of bNc ;Nf .

x2.1. Nf = Nc 1
Consider the case Nf = Nc 1. The gauge group SU(C) is completely broken
at a generic point of Mc. Having no massless non-Abelian gluons, the theory is
not strongly interacting at low energies. Moreover, there is no branch cut in the
superpotential (1.30). This suggests that the theory does not have multiple vacua.
In this case, the superpotential is generated by instanton for the following three
reasons.
15
16 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
Suppose in a eld theory with Euclidean action S() we do perturbation
 around
an instanton 0, which a solution of the equation of motion  
S = 0. Then
2 
0

(2.1) S() = S(0 ) + 12 S2  +


:


0

The result of any perturbative path integral computation in the instanton back-
ground 0 is of the form
(2.2) 0
detF


;
e S ( ) det
B

where detB , detF are the determinants of the operators in S  acting on the
2
2

bosonic, fermionic elds, respectively. In gauge theories, the instanton contribution

0

is non-zero and is proportional to


e S ( ) = e g  +i = e2i () :
8 2
(2.3) 0 ( )2

So (2.2) is proportional to b = 3Nc Nf by (1.15); the appropriate power of 

0

comes from regularizing the determinants and other factors. The superpotential
(1.30) has the same power of  when Nf = Nc 1. Thus for Nf = Nc 1,
the superpotential is instanton generated, and the coecient bNc ;Nc 1 is non-zero.
We can choose a renormalization subtraction scheme such that bNc ;Nc 1 = 1. In
other words, the non-zero coecient is absorbed in the mass scale . Here we see
how in a supersymmetric theory, the instanton action, -function and anomaly are
interrelated.
Secondly, in instanton perturbation theory outlined above, because of the zero
modes of the Dirac operator, an instanton absorbs 2Nc of the  and Nf each
of the Q and Q0 . This can be represented graphically by an instanton vertex
(Figure 2.1(a)). On the other hand, the Lagrangian contains interactions Q Q
and  Q0 Q0 which are represented by the usual vertices (Figure 2.1(b)). In the
presence of expectation values hQi, hQ0i, the external Q, Q0 lines can be absorbed.
When Nf = Nc 1, it is possible to form diagrams with one external line of Q
and Q0 . (An example is shown in Figure 2.1(c).) The existence of such diagrams
is the hallmark of having a superpotential.
Finally, if Nf = Nc 1, instanton computations are reliable and yields nite
results. In theories with strong interaction, small instantons correspond to a small
coupling constant g (by asymptotic freedom), whereas a large instanton size means
strong coupling. When the instanton size becomes large, instantons overlap and
have long range interactions. The integration over instanton size diverges. In
our theory, at the energy much higher than hQi, hQ0i, the gauge group SU(C) is
unbroken, so g is small by asymptotic freedom. It turns out that our theory does
not reach strong coupling in the infrared as well. This is because when the energy is
smaller than hQi, hQ0 i, the gauge group is completely broken. So the instantons can
not grow and our theory is free from the usual troubles in the instanton calculations.
Higher order corrections beyond one instanton contribution would have a de-
pendence on  not allowed by symmetry and therefore are not present. So when
Nf = Nc 1, the one instanton contribution gives the exact answer of the eective
superpotential. In this case, we see that there is no non-renormalization theorem
because a non-zero superpotential is generated non-perturbatively.
 LECTURE 2. QUANTUM BEHAVIOR OF SUPER QCD: Nf SMALL 17

Nf
ψ’
ψ’ Q or Q ’

2 Nc ψ or ψ ’
ψ Nf
λ

(a) (b)

Q Q

ψ
λ

λ ψ

λ λ

ψ’

ψ’

Q’ Q’

(c)

Figure 2.1. (a) The instanton vertex; (b) Yukawa interaction vertex; (c) A
diagram with one external line of Q and Q0

x2.2. Nf < Nc 1
We start with the theory with Nc 1 number of avors, i.e., dimF = dimF 0 = Nc 1
and Q 2 Hom(F; C), Q0 2 Hom(C; F 0). We then add to it a mass term. At short
distances (ultraviolet), the mass term is the tree level superpotential
(2.4) Wtree = tr mQ0 Q;
18 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
where m 2 Hom(F 0; F) is of rank Nc Nf 1, Nf being the number of light avors.
At long distances (infrared), we shall have an eective superpotential We (M; ; m)
which reduces to (1.30) for Nc 1 avors when m = 0. (Recall that we have chosen
a subtraction scheme such that bNc ;Nc 1 = 1.) Since We is holomorphic in m, is
invariant under SU(F)  SU(F 0 ), and has R-charge 2, mass dimension 3, it must
be of the form
2Nc +1
(2.5) We = det M + tr mM:
When the energy drops below m, the eective theory has Nf number of avors.
Take F = F^  F^ ? and F 0 = F^ 0  F^ 0? with dim F^ = dim F^ 0 = Nf such that
 
(2.6) m = 00 m0? ;
where m? 2 Hom(F^ ?; F^ 0? ) is non-degenerate.
The mass scales ^ of the infrared theory and  of the ultraviolet theory are
related by the renormalization group ow. At high energy (large ), b0 = 2Nc + 1.
So the coupling constant g() runs according to
8 2
  2Nc+1
(2.7) e g()2 +i
=  :
At low energy (small ), b0 = 3Nc Nf . So g() ows as
^
!3Nc Nf
(2.8)
 +i
e g
8 2
= 
( )2
 :

Matching g() at a typical mass scale  = (det m? )1=(Nc Nf ) (see Figure 2.2), we
get in a specic subtraction scheme (called DR scheme)
(2.9) ^3Nc Nf = 2Nc +1 det m? :

b0 = 3 N c - N f

b0 = 2 N c + 1

0 µ

Figure 2.2. Matching of renormalization group ows at high and low energies
 LECTURE 2. QUANTUM BEHAVIOR OF SUPER QCD: Nf SMALL 19

Since (2.5) is an eective potential, we can integrate out the heavy quarks
classically. The light components of
 ^ 
(2.10) M = MM M 12
21 M ?
are in M^ 2 Hom(F; ^ F^ 0). We integrate out the heavy components M12, M21 and
?
M of M. Fixing M, ^ the minimum of We in (2.5) is reached at M12 = M21 = 0
and
 2Nc+1 det m?  Nc Nf
1

(2.11) M = ? (m? ) 1:
det M^
Substituting these values into (2.5) and using (2.9), we get
!
^3Nc Nf Nc Nf
1

(2.12) ^
We (M) = (Nc Nf )  :
det M^
This is of the same form as (1.30). Moreover we conclude that in our renormalization
scheme, bNc ;Nf = Nc Nf 6= 0 when Nf < Nc .
We also want to nd out the physical mechanism that generates the superpo-
tential (1.30) when Nf < Nc 1 without embedding the theory into a larger one.
At a generic point in M, expectation values hQi, hQ0i break the SU(Nc ) gauge
symmetry to SU(Nc Nf ). The gauge bosons not in the unbroken subgroup ac-
quire a mass classically. At an energy below this mass scale, we have an eective
theory whose Lagrangian density is
Z Z (M) 
(2.13) 4 y
Le = d  K(M; M ) + d  4i tr W W + c:c: ;
2 

where K is a Kahler potential on M which determines the kinetic term of M, W is

the supereld strength of the SU(Nc Nf ) gauge theory, and (M) = 2 + g(4Mi) 2

is determined by the coupling constant g(M) at scale M. Using the 1-loop -

function, (M)  logdet M. The pure SU(Nc Nf ) gauge theory has a mass
gap and depends on a scale SU (Nc Nf ) . Below this mass scale, we have M only.
Again, SU (Nc Nf ) is related to  by matching two renormalization group ows at
 = (det M)1=2Nf . In fact,
3(Nc Nf )
(2.14) 3Nc Nf = SU (Nc Nf ) det M;
where the same subtraction scheme is used. We claim that the eective superpo-
tential of M is generated by the pure gauge theory. Indeed, the leading order term
of W W  is . According to (1.11),
(2.15) hi  3SU (Nc Nf ) :
Using (2.14), we get
 3Nc Nf  Nc Nf
hi  det M
1

(2.16) :
Therefore the bosonic potentials from (1.30) and (2.13) agree:
(2.17) @We (M)  @(M) hi:
@M @M
That bNc ;Nf 6= 0 also conrms that hi = 6 0 in the pure gauge theory.
20 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
The bosonic potential of Q, Q0 is qualitatively illustrated in Figure 2.3(a).
We see that this potential pushes the vacuum to innity. Therefore theories with
Nf < Nc do not have a vacuum. We can add a bare mass to the theory and recover
the original theory by taking the limit as the mass goes to zero. If Nf < Nc , the
eective superpotential is
 3Nc Nf  Nc Nf
We = (Nc Nf ) det M
1

(2.18) + tr mM:
The bosonic potential now looks like Figure 2.3(b). The minimum is reached at
(2.19) hM i = (3Nc Nf det m) Nc m 1 : 1

The Nc solutions (in the choices of the Nc -th root) correspond to the Nc vacua of
the pure gauge theory. Indeed as m ! 0, all the eigenvalues of M go to innity.

V( M ) V( M )

0 M 0 M

(a) (b)

Figure 2.3. The bosonic eective potential in supersymmetric QCD: (a) m =

0; (b) m 6= 0
The result (2.19) is consistent with all the symmetries. We claim that it is
also valid in the range Nf  Nc , where no superpotential is generated. To show
this, we imagine adding a huge mass to some of the quarks, integrating them out,
and reducing to a problem of Nf < Nc . Since hM i is holomorphic in m, (2.19) for
Nf  Nc follows from the case Nf < Nc . For Nf  Nc , the limit of hM i as m ! 0
depends on the way the limit is taken. This is not a surprise because there is a
moduli space of vacua when m = 0. With dierent ways of taking the limit, the
massive theory lands on dierent points on the moduli space.
x2.3. Nf = Nc
As before, we rst add a mass term tr mM, then take the limit m ! 0. If m 2
Hom(F 0 ; F) is invertible then hM i is given by (2.19). The invariance under the
U(1)B global symmetry implies that hB i = hB 0 i = 0. These expectation values lie
on the surface
(2.20) det M B
B 0 = 2Nc :
Because (2.20) is independent of m, the quantum ground states are not on the
classical moduli space Mc = fM; B; B 0 j det M B
B 0 = 0g even in the limit
m ! 0. We call Mq = fM; B; B 0 j det M B
B 0 = 2Nc g the quantum moduli
space.
 LECTURE 2. QUANTUM BEHAVIOR OF SUPER QCD: Nf SMALL 21

The constraint (2.20) is consistent with all the symmetries. We postulate that
the low energy eective theory consists of elds M, B, B 0 constrained on Mq .
Unlike the classical moduli space Mc, the quantum moduli space Mq has no sin-
gularities near the origin because of the correction 2Nc , as shown in Figure 2.4.
This suggests that there are no extra massless particles (other than the modes in
the at directions) { a point we shall check later by 't Hooft's anomaly matching
condition. If so, the original gluons are not light, even near the origin, the region
of strong coupling. Thus the gluons are conned. Far out along the at directions
(when M is big), the quantum correction is small. There the classical moduli space
Mc is a good approximation to Mq , as expected. Again there are no massless
gluons. But in that region, it is more natural to interpret it as Higgs phenomena,
in which the gauge group is completely broken by hM i and all the gauge bosons ac-
quire a mass. Therefore there is a continuous interpolation between the connement
and the Higgs descriptions.

(a) (b)

Figure 2.4. (a) The classical moduli space has a singularity at the origin; (b)
The quantum moduli space is smooth.

We also remark on chiral symmetry breaking. The global symmetry SU(F) 

SU(F 0 )  U(1)B  U(1)R is unbroken at the origin of the classical moduli space
Mc. But at any point on the quantum moduli space Mq , the symmetry is always
broken to a proper subgroup H, the isotropy subgroup, which varies according
to the point. The massless particles live in the tangent spacepof Mq and form
a representation of H. For example, at M = 0, B = B 0 = 2Nc , we have
H = SU(F)  SU(F 0 )  U(1)R . So we have connement with chiral symmetry
breaking.
Since the values of M B, B 0 in the low energy eective theory are constrained
on Mq , the theory can be described by
(2.21) We = 2N1c 1 A (B
B 0 det M + 2Nc );
where A is a supereld (of mass dimension 2) that imposes the constraint (2.20)
Now M, B, B 0 can take arbitrary values. We check that our theory reduces to
supersymmetric QCD with Nf < Nc after integrating out the quarks to which a
22 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
bare mass is added. The eective potential with a mass term is
(2.22) We = 2N1c 1 A (B
B 0 det M + 2Nc ) + tr mM;
where m 2 Hom(F 0; F) is of rank Nc Nf . (Nf is the number of light avors.)
Write F = F^  F^ ? and F 0 = F^ 0  F^ 0? with dim F^ = dim F^ 0 = Nf such that m
has the form (2.6). Again the light components in M of the form (2.10) are in
M^ 2 Hom(F; ^ F^ 0). Integrating out the rest of M and B, B 0 in (2.22), we get
^
! 1
3Nc Nf Nc Nf
(2.23) W^ e = (Nc Nf )  :
det M^
This is precisely the superpotential (1.30) with the correct coecient. Here ^3Nc Nf
= 2Nc det m? , which is again consistent with the renormalization group ow.
x2.4. Nf = Nc + 1
Again, we perturb the theory by adding a mass term tr mM. Then we have hB i =
hB 0 i = 0 and using (2.19), ^Nc hM i = 2Nc 2 det m (^Nc m 1 ) = 2Nc 2 m. In the
limit m ! 0, the expectation values satisfy the constraints (1.21) and (1.24) of the
classical congurations. However, for m = 6 0, they are not on the classical moduli
space Mc . In fact, all the classical values of M with B = B 0 = 0 can be obtained
by taking an appropriate limit m ! 0. Unlike the case Nf = Nc , there is not a
constraint which hB i, hB 0 i and hM i satisfy for all values of m. We propose that
the long distance eective theory of the original theory without a bare mass is a
Wess-Zumino model with the unconstrained elds M, B, B 0 and a superpotential
(2.24) We (M; B; B 0 ) = 2N1c 1 (B 0 MB det M):
The stationary points of We are at @W @M = B
B 0 ^Nc M = 0, @W 0
@B = B M = 0,
@W0 = B 0 M = 0; these are precisely the constraints that every classical congura-
@B
tion satises. However, the moduli space is interpreted dierently: it is embedded
into a larger space. As a result, more particles become light in the eective theory.
For example, at M = 0, B, B 0 are the new massless elds.
Similar to the theory with Nf = Nc , we see connement at M = 0 since
there are no massless gluons. There is again a continuum between the connement
description near the origin and the Higgs description in the semiclassical region.
However, at M = B = B 0 = 0, the full global symmetry is unbroken. So we can
have connement without chiral symmetry breaking.
We check that by adding a mass term to the superpotential (2.24), i.e,
(2.25) We (M; B; B 0 ) = 2N1c 1 (B 0 MB det M) + tr mM;
the theory with Nf = Nc can be recovered. Here m is of rank 1. Again we write
m in the form (2.6). The light components in
 ^     
(2.26) M = M A ; B = B^ ; B 0 =  ; B^ 0
^ A, B,
are M, ^ B^ 0 . Integrating out the rest, we get
(2.27) We (M; ^ B^ 0) = 2N1 1 A (B^
B^ 0 det M^ + ^2Nc );
^ A; B;
 c
 LECTURE 2. QUANTUM BEHAVIOR OF SUPER QCD: Nf SMALL 23

where ^2Nc = m? 2Nc 1. This agrees with the superpotential for Nf = Nc after
a rescaling of the eld A.
We shall further check the postulate on the eld content of the low energy
theory by 't Hooft's anomaly matching condition.
x2.5. 't Hooft anomaly matching condition
A low energy theory may have massless fermions very dierent from the fundamen-
tal theory. The anomaly matching condition of 't Hooft is a highly non-trivial test
on the compatibility of the two theories. We rst introduce the condition, then
apply it to check some of the claims we made about supersymmetric QCD with
Nf = Nc and Nf = Nc + 1.

λ or ψ

Figure 2.5. The triangular Feynman that causes the gauge anomaly
Any gauge theory with Weyl fermions may suer from anomalies. If so, the
quantum theory is ill-dened. Suppose the gauge group is G and the fermions
are in the representation R of G. Perturbatively, the gauge anomaly comes from
the triangular Feynman diagram in Figure 2.5. This is proportional to a totally
symmetric cubic form dR on the Lie algebra g of G given by
(2.28) dR (x; y; z) = trR xfy; z g (x; y; z 2 g):
dR = 0 if R is a real or quaternionic (pseudo-real) representation. For simple Lie
groups, dR 6= 0 only if g = su(Nc ) (Nc  3). In the path-integral language, the
fermionic path integral is the determinant of the Dirac operator D6 R+ (twisted by R)
on the spacetime 4-manifold X. This determinant is a section of the determinant
line bundle Det 6DR+ over A=G , the space A of connections modulo the group G of
gauge transformations. By local family index theorem, the curvature FR of Det D 6 R+
0
evaluated on two tangent vectorsZ B, B at A 2 A is given by
(2.29) FR (B; B 0 ) = 24i 3 dR (F; F;
A1  [B; B 0 ]) + dR (F; B; B 0 );
X
where F is the curvature of A 2 A and
A , the Laplacian acting on TA A. The
quantum theory makes sense only when the bundle Det 6DR+ is canonically trivial.
This requires FR = 0, hence dR = 0, in agreement with the perturbative argument.
^ = 1. In general, FR contains terms
Strictly speaking, (2.29) is true only when A(X)
24 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
proportional to the linear form trR on g (which is non-zero only if G has a U(1)
factor). This is the mixed gauge and gravitational anomaly. So the quantum theory
is free from gauge and gravitational anomalies if and only if the representation R
is arranged so that dR = trR = 0. For example, if G = SU(C), it is easy to check
that dC = dC and dadj = 0. So N = 1 supersymmetric QCD with equal number
of quarks and anti-quarks is free from anomalies.
Recall that the global symmetry of our theory is SU(F)  SU(F 0 )  U(1)B 
U(1)R . At a particular point in the moduli space, this symmetry may be broken
to a subgroup H. The massless fermions will form a representation R of H. H
is not gauged; if we gauge it (adding gauge elds interacting with the fermions),
there will be anomalies proportional to dR and trR . But imagine that in the gauged
theory we add some massless fermions in another representation R0 that couple only
to the gauge elds of H. The representation R0 is chosen so that the combined
theory is anomaly free, i.e., dR + dR = 0, trR + trR = 0. Suppose our original
0 0
theory with representation R has a low energy description, in which the fermions
are in representation R0 . Then the low energy description of the combined theory
is anomaly free because it comes from an anomaly free ultraviolet theory. Since the
additional massless fermions stay the same, we have dR0 +dR = 0, trR0 +trR = 0.
0 0
Therefore
(2.30) dR = dR0 ; trR = trR0 :
This is the 't Hooft anomaly matching condition. It is a stringent check on whether
two theories with fermions in representations R and R0 describe the same low energy
physics.
Now we check the claimed solutions of the low energy theories with Nf = Nc
and Nf = Nc + 1 against 't Hooft's criterion. Using (1.27), the fermions in the
ultraviolet theory transform under the global symmetry according to

SU(F)  SU(F 0 )  U(1)B  U(1)R

Q F 1 1 Nc
Nf
(2.31)
Q0 1 F0 1 Nc
Nf
 1 1 0 1
In the infrared theory, the fermionic components M , B , B0 of the composite
elds M, B, B 0 transform as

SU(F)  SU(F 0 )  U(1)B  U(1)R

M F  F 0 0 1 2 NNfc
(2.32)
^Nc F  1 Nc Nc 1 NNfc
2
B
1 ^Nc F 0 Nc Nc 1 NNfc
2
B0

We can calculate the cubic and the trace forms on the Lie algebra su(F)  su(F 0) 
u(1)B  u(1)R in both the ultraviolet and the infrared theories. The non-zero
combinations are given by
 LECTURE 2. QUANTUM BEHAVIOR OF SUPER QCD: Nf SMALL 25

(2.33)
ultraviolet infrared
su(F )3 Nc ( dF ) Nf ( dF ) + ( d^Nc (F ) )
su(F 0 )3 Nc d F 0 Nf dF 0 + d^Nc F 0
su(F )2 u(1)B Nc c2 (F )
( 1) c2 (^Nc F )
( Nc )
su(F 0 )2 u(1)B N c c 2 (F 0 )
1 c2 (^Nc F 0 )
Nc
su(F )2 u(1)R N c c 2 (F )
( N Nc ) Nf c2 (F )
(1 2NNfc ) + c2 (^Nc F )
(Nc Nc2 )
f 1 Nf
su(F 0 )2 u(1)R N c c 2 (F 0 )
( N Nf c2 (F 0 )
(1 N 0
Nc ) Nc N2
Nf ) + c2 (^ c F )
(Nc Nfc )
2
1
f N 

N Nc2
Nc [( Nc ) + Nc ]
(Nc 1 Nf )
f
u(1)2B u(1)R Nc Nf [( 1)2 + 12 ] ( Nfc ) 2 2 2

N 3 Nf2 (1 2NNfc )3 + 2 N Nc2 3

u(1)3R 2Nc Nf ( Nfc ) + (Nc f
Nc (Nc 1 Nf )
2 1)13
 Nc2
u(1)R 2Nc Nf ( Nc
Nf ) + (Nc
2 1)1 Nf2 (1 2NNfc ) + 2 N f
Nc (Nc 1 Nf )

When Nf = Nc , the quantum modulipspace Mq is dened by the constraint

(2.20). At the point M = 0, B = B 0 = 2Nc , the global symmetry group is
broken to H = SU(F)  SU(F 0 )  U(1)R . Using the condition Nf = Nc , it is
straightforward to check that for su(F)3, su(F 0 )3, su(F)2u(1)R and su(F 0)2 u(1)R ,
the entries in the left and right columns match. However for u(1)3R and u(1)R ,
the entry in left column is (Nc2 + 1) whereas as that in the right is (Nc2 + 2).
The discrepancy is because M, B, B 0 are constrained on Mq . Alternatively, we
can introduce a Lagrange multiplier A as in (2.22). A is invariant under SU(F) 
SU(F 0 )  U(1)B . Since the R-charges of M, B, B 0 are 0 and that of We should
be 2, the R-charge of A is 2. So A has R-charge 1 and contributes 13 and 1 to
u(1)3R and u(1)R , respectively, in the right column. Thus we have the matching
(Nc2 + 1) = (Nc2 + 2) + 1. This computation conrms that for Nf = Nc , the
elds M, B, B 0 are constrained on Mq and that there are no additional light elds.
For Nf = Nc + 1, consider the point M = B = B 0 = 0 on the moduli space,
where the full global symmetry group is unbroken. Using Nf = Nc + 1, it is
straightforward to check that in (2.33), all the corresponding entries in the two
columns match. This shows that in the low energy theory, M, B, B 0 are indeed
unconstrained and that they are the only light elds.
 LECTURE 3
Quantum Behavior of Super QCD: Nf Large
In the previous lectures, we studied N = 1 supersymmetric QCD with Nf 
Nc + 1. More concretely, we have an N = 1 gauge theory with gauge group G =
SU(C) coupled to N = 1 matter which consists of the quarks Q 2 Hom(F; C) and
the anti-quarks Q0 2 Hom(C; F 0). Nc = dimC is called the number of colors and
Nf = dimF = dimF 0, the number of avors. The anomaly-free global symmetry
of the theory is SU(F)  SU(F 0 )  U(1)B  U(1)R , under which Q, Q0 transform as
(1.27). The moduli space Mc of classical ground states are parametrized by M, B,
B 0 , which are also the elds in the low energy eective theory. Their transformations
under the global symmetry are also given by (1.27).
If Nf < Nc , a non-zero superpotential (1.30) is dynamically generated. Con-
sequently, the quantum theory has no vacuum. If Nf = Nc , the classical moduli
space Mc is modied in the quantum theory. The light elds M, B, B 0 satisfy the
constraint (2.20). If Nf = Nc + 1, the quantum moduli space is the same as Mc .
But all the components of M, B, B 0 become physical in the low energy theory. We
checked these statements on the low energy eld theory using 't Hooft's anomaly
matching condition.
In this lecture, we will study the theory for all ranges of Nf . If Nf  3Nc ,
then the -function (g) > 0, so the theory is infrared free. (If Nf = 3Nc , the
1-loop -function vanishes but the 2-loop contribution is positive. See below.) At
long distances, we see elementary degrees of freedom: the massless quarks and
gluons. We will show that when 3N2 c < Nf < 3Nc , the theory has a non-trivial
infrared xed point, which is a superconformal eld theory in four dimensions.
When Nc + 2 < Nf < 3N2 c , the theory is strongly coupled at the infrared. But
there is a dual description, the magnetic theory, which is infrared free.

x3.1. N < Nf < 3Nc : non-trivial infrared xed points

3
2
c

Up to 2-loop, the -function is

g 3 g 5  N f

(3.1) (g) = 162 (3Nc Nf ) + 1284 2Nc Nf 3Nc N + o(g7 ): 2
c
Note that the 2-loop term is positive when Nf = 3Nc . Consider the limit Nc ; Nf !
1 while NNfc = 3  is xed for some  > 0. The -function is shown in Figure 3.1.
If  is suciently small, the higher order terms in (g) are negligible. So there is a
27
28 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
solution g of (g) = 0 with
2
(3.2) g2 Nc = 83  + o(2 ):
The coupling constant runs and stops at this point in the infrared.

β( g)

0 g* g

Figure 3.1. The behavior of  -function to two loops

At g , the we have a scale invariant and supersymmetric eld theory, i.e.,

a superconformal eld theory in four dimensions. It is moreover an non-trivial
interacting theory; this will be conrmed by computing the mass dimension of
certain elds later. At such a xed point, the notion of particle is ill-dened.
Because of scale invariance, there is no mass gap. The energy spectrum contains
a vacuum and a continuum above it. So massless particles can decay arbitrarily
since they are not free at long distances. Consequently, there are no well-dened
one-particle states in the Hilbert space and there is no S-matrix.
Unlike in two dimensions, the conformal algebra in four dimensions is nite
dimensional. Yet it is still possible to extract information from it. We consider
radial quantization, i.e., regard the Euclidean spacetime R4 f0g as an expanding
continuum of S 3 . There is a one-to-one correspondence between operators and
states in radial quantization. Let j0i be the vacuum. Then an operator O denes
a state jOi = Oj0i in the Hilbert space. Since the theory is unitary, the operators
form a unitary representation of the superconformal algebra.
Recall that the Lie algebra of the conformal group in spacetime of dimension
d  3 is generated by translations P = @ , Lorentz generators M = x@ x @ ,
dilation D = x @ and special inversions K = 2x x @ x2@ . P , M form
the standard Poincare Lie algebra. The commutation relations of other generators
are
(3.3) [D; P ] = P ; [D; K ] = K ; [K ; P ] = 2( D M ):
We look for representations in which P are the creation operators and K are the
annihilation operators, and Py = K  . There is a vacuum or highest weight vector
jOi such that it is an eigenvector of D (whose eigenvalue is also denoted by D) and
K jOi = 0. Assume that O is a scalar operator, i.e., M jOi = 0. Using (3.3), the
 LECTURE 3. QUANTUM BEHAVIOR OF SUPER QCD: Nf LARGE 29

square of the norm of P jOi is

(3.4) hOjK  P jOi = 2dD:
Therefore unitarity implies that D  0 and D = 0 if and only if P jOi = 0, i.e., O is
a constant operator. Furthermore, using (3.3), the square of the norm of P P jOi
is
(3.5) hOjK  K P  P jOi = 8dD(D d 2 2 ):
For d = 4, unitarity implies that D  1 and D = 1 if and only if P  PjOi = 0, i.e.,
O corresponds to a free eld.
When d = 4, the superconformal algebra has extra generators Q, Q _ of con-
formal dimension 12 and S  , S_ of conformal dimension 21 . The latter are the
supersymmetric partners of Km u, i.e., [Q; K  ] = _ S_ and [Q _ ; K  ] = 
 S .
_
Moreover, we have 1

(3.6) fQ; S  g = ( ) M + 2 (D 32 R);

where R is the generator of the U(1)R transformation (whose eigenvalue is also
denoted by R). Again, we look for highest weight representations such that Py =
K  and Qy = S  . An operator O is called chiral if Q _ jOi = 0. Again we assume
that O is a scalar operator. If jOi is also a highest weight vector, i.e., K jOi = 0
and S  jOi = 0, then the square of the norm of QjOi is
(3.7) hOjS  Q jOi = 4(D 32 R):
So D  32 R and the equality holds for chiral operators only.
A consequence of the above study is that the chiral operators form a ring.
Consider the product of two chiral operators 1(x1) and 2(x2 ). When x1 is close
to x2, we have an expansion
X
(3.8) 1 (x1)2 (x2)  (x1 x2)i i (x2):
i
All the operators i in the expansion have R(i) = R(1) + R(2). So
(3.9) D(i )  23 (R(1) + R(2)) = D(1 ) + D(2 ):
Therefore all the exponents i  0. In the limit x1 ! x2, the surviving terms terms
have i = 0 and saturate the inequality (3.9). These are the chiral operators. So
chiral operators form a ring.
We return to the study of the non-trivial xed point g with (g ) = 0. The
dimension of the chiral elds can be computed by their R-charges. For example,
using R-charges in (1.27), we get, in the infrared limit,
(3.10) D(M) = 32 R(M) = 3 Nf N Nc
c
and
3  N2 
(3.11) 0
D(B) = D(B ) = 2 Nc Nc :
f

1
See for example Phys. Rep. 128 (1985) 39.
30 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
The meson M has dimension 2 in the ultraviolet. The dierence with (3.10) is
called the anomalous dimension. Perturbatively, the anomalous dimension of mass
is
(3.12) (g) = 8 g2 Nc2 1 + o(g4 ):
2 Nc

At the infrared xed point g given by (3.2), (g ) = 3 +o(2). In the perturbative
expansion, (3.10) is D(M) = 2 3 + o(2 ). This matches D(M) = 2 + (g ).
The requirement D(M)  1 implies that Nf  32 Nc . For Nf  3Nc , the
theory is free at the infrared. We claim that for 23 Nc < Nf < 3Nc , the theory is
an interacting superconformal eld theory in the infrared limit. The xed point
is non-trivial because D(M) > 1 in this range. If Nf = 32 Nc (for Nc even), then
D(M) = 1. So M (and perhaps the whole low energy theory) is free. If Nf < 23 Nc ,
(3.10) would imply D(M) < 1. Therefore the picture breaks down and new physics
should emerge.

x3.2. Nc + 2  Nf  N : infrared free magnetic theory

3
2
c

Associated with the original supersymmetric QCD which we now call the electric
theory, we introduce a dual theory when Nf  Nc + 2, called the magnetic theory.
The dual theory has a gauge group G~ = SU(C), ~ where C~ is an Hermitian vector
space of dimension N~c = Nf Nc and is equipped with a complex volume form
v~c 2 ^N~c C~  invariant under G~ (and G~ C = SL(C)).
~ The quarks and anti-quarks
~ ~ ~ ~ 0 ~ ~ 0
are Q 2 Hom(F; C) and Q 2 Hom(C; F ), respectively, where
(3.13) F~ = F 
(^Nf F)1=N~c ; F~ 0 = F 0
(^Nf F 0)1=N~c
are the spaces of dual avors (dened up to a root of unity). In addition, the dual
theory has a meson M 2 Hom(F; F 0), which comes from the electric theory but is
regarded as an elementary eld here. In the electric theory, M has mass dimension
2 in the ultraviolet and acquires an anomalous dimension in the infrared. Therefore
in the magnetic theory, it is more natural to regard the eld  1 M 2 Hom(F~ 0; F)
~
of mass dimension 1 as the elementary meson, where
(3.14)  2 (^Nf F 
^Nf F 0)1=N~c
~ gauge elds but
is a parameter of mass dimension 1. M is uncoupled to the SU(C)
interact with the dual quarks through a superpotential
(3.15) W = 1 trF Q~ 0 QM:
~

~ F~ 0) 
Here Q~ 0 Q~ 2 Hom(F; = Hom(F 0; F)
(^Nf F 
^Nf F 0 )1=N~c is the meson of
the magnetic theory. W has mass dimension 3.
Under the global symmetry SU(F)  SU(F 0 )  U(1)B  U(1)R of the electric
~ Q~ 0 , M transform as, according to (3.13) and (1.26),
theory, the elds Q,
 LECTURE 3. QUANTUM BEHAVIOR OF SUPER QCD: Nf LARGE 31

SU(F)  SU(F 0 )  U(1)B  U(1)R

Q~ F 1 Nc
N~c
Nc
Nf
(3.16)
Q~ 0 1 F 0 Nc
N~c
Nc
Nf
M F F0 0 2 NfNfNc

So the superpotential (3.15) has R-charge 2 NfNfNc + 2 NfNfN~c = 2, as required.

Their fermionic components Q~ , Q~0 , M and the gluino eld ~ (of the dual theory)
transform as

SU(F)  SU(F 0 )  U(1)B  U(1)R

Q~ F 1 Nc
N~c
Nc
Nf 1
(3.17) Q~0 1 F0 Nc
N~c
Nc
Nf 1
M F F~ 0 1 2 NNfc
~ 1 1 0 1

It is easy to check that the measure D Q~ D Q~0 D ~ of the fermions that are coupled
to the SU(C) ~ gauge elds is invariant under the global symmetry. Therefore the
anomaly-free global symmetry of the dual theory is SU(F)  SU(F 0 )  U(1)B 
U(1)R as well.
We argue that the electric and magnetic theories describe the same physics at
least at long distances. First, we already see that both have the same anomaly-free
global symmetry SU(F)  SU(F 0 )  U(1)B  U(1)R . That they do not have the
same gauge group is not a problem. Strictly speaking, gauge symmetry is not a
symmetry, but a redundancy in the description of the theory. So gauge symmetry
is not fundamental. In fact, the gauge elds in the magnetic theory are composites
of the electric degrees of freedom, and vice versa.
Secondly, the electric and magnetic theories have the same gauge invariant
operators, though their eld contents seem rather dierent. We construct the map-
ping of gauge invariant chiral operators. The composite meson Q0Q of the electric
theory corresponds to the elementary meson M of the magnetic theory, whereas the
composite meson Q~ 0 Q~ of the magnetic theory vanishes by the equation of motion
of M. The baryons of the magnetic theory are B~ = Q~  v~c 2 ^N~c F~   = ^Nc F  and
B~ 0 = Q~ 0v~c 2 ^Nc F~ 0 
~
= ^Nc F 0. Thus they have the same quantum numbers as
0
B, B of the electric theory, which we shall identify. In fact, taking into account
(1.29), (3.14) as well as their mass dimensions, the only possible relations are
(3.18) 0 ~ B 0 = iN~c +1 b =2 N~c =2 B~ 0 ;
B = iN~c +1 b =2 N~c =2 B; 0

where the numerical factors are chosen to t future calculations. However, Q, ~ Q~ 0

0
are not polynomials of Q, Q and vice versa because the U(1)B charges of Q, ~ Q~ 0
are fractional. Rather, the dual quarks Q, ~ Q~ 0 should be interpreted as solitons of
Q, Q0 in the electric theory, and vice versa.
32 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
Thirdly, we check the 't Hooft anomaly matching condition. Under the global
symmetry, the fermions in the electric theory transform as (2.31). Their contribu-
tion to anomaly is given by the left column of (2.33). In the magnetic theory, the
fermions transform as (3.17). Their contribution to anomaly is given by

magnetic
su(F)3 ~
Nc dF + Nf ( dF )
su(F 0)3 N~c ( dF 0 ) + Nf dF 0
su(F)2u(1)B N~c c2 (F)
( NN~cc )
su(F 0)2 u(1)B N~c c2(F 0)
NN~cc
(3.19)
su(F)2u(1)R N~c c2 (F)
( NNfc 1) + Nf c2(F)
(1 2 NNfc )
su(F 0)2 u(1)R N~c c2(F 0)
( NNfc 1) + Nf c2(F 0 )
(1 2 NNfc )
u(1)2B u(1)R N~c Nf [( NN~cc )2 + ( NN~cc )2 ]
u(1)3
R 2N~c Nf ( NNfc 1)3 + Nf2(1 2 NNfc )3 + (N~c2 1)13
u(1)R 2N~c Nf ( NNfc 1) + Nf2 (1 2 NNfc ) + (N~c2 1)1

It is straightforward to check that these two columns match.

In the range Nc + 2  Nf  3N2 c (not yet studied except for Nf = 23 Nc ), we
have N~c  2 and Nf  3N~c , therefore the magnetic theory is infrared free. So
the magnetic theory is a better description in this range, where the electric theory
is hard to study because of the strong coupling in the infrared. In particular, at
Nf = 3N2 c (provided that Nc is even), we calculated D(M) = 1 in the electric
theory. This conrms that M is a free eld in the infrared, a consequence of of the
magnetic description. Generally, as Nf decreases, the coupling in the electric theory
gets stronger until the theory connes when Nf  Nc + 1, whereas the coupling
of the magnetic theory gets weaker and the theory becomes infrared free when
Nf  3N2 c . The infrared behavior of the theory in various ranges is summarized in
the following table.

electric theory magnetic theory

Nc  2, Nf  3Nc infrared free strongly coupled
(N~c + 2  Nf  3N2~c )
3Nc < N < 3N non-trivial non-trivial
2 f c
3 ~c
N ~
( 2 < Nf < 3Nc ) infrared xed point infrared xed point
Nc + 2  Nf  3N2 c strongly coupled infrared free
~ ~
(Nc  2, Nf  3Nc )
Nf  Nc + 1 analysed in ||{
lectures 1, 2
 LECTURE 3. QUANTUM BEHAVIOR OF SUPER QCD: Nf LARGE 33

x3.3. Further tests of duality

Dynamically, the electric theory has a scale  as in (1.29), where b0 = 3Nc Nf
determines the -function. The scale ~ of the magnetic theory must be related to
~  all have mass dimension 1 and because of
that of the electric theory. Because , ,
(1.29) (and its magnetic counterpart) and (3.14), the only possible relation among
them is, up to a numerical factor,
(3.20) 3Nc Nf ~3N~c Nf = ( 1)N~c Nf :
The phase ( 1)N~c will be needed in a number of places below. This, together
with (1.15), shows that if the electric theory is strongly coupled, then the magnetic
theory is weakly coupled, and vice versa.
x3.3.1. Duality is an involution
We claim that the dual of the magnetic theory is the electric theory itself. The
number of colors in the double dual theory is Nf N~c = Nc . Therefore, its color
space can be identied with C. Next, the avor spaces are F~ 
(^Nf F) ~ 1=N~c 
=F
~ 0
and F
(^ F ) N f ~ 0 1 =N~ c  0
= F , the same as the electric theory. The mass scale of
the double dual theory is again ; this follows from (3.20), provided that
(3.21) ~ =  2 (^NF F~ 
^Nf F~ 0 )1=Nc  = (^Nf F 
^Nf F 0)1=N~c :
The double dual theory has a new term 1~ trF~ Q0 QM~ in the superpotential where
M~ 2 Hom(F; ~ F~ 0) is a new meson eld. This, combined with (3.15), gives the total
superpotential
(3.22) W = 1 trF M(M
~ Q0Q):
The equation of motion of M~ implies that M = Q0Q. Therefore we recover the
electric theory.
x3.3.2. Deformation along the at directions
So far, we have only studied the theory with Nf  Nc + 2 at the origin of the
moduli space M. Recall that in the electric theory, M is the space of M, B, B 0
subject to the classical constraints (1.20), (1.21), (1.22).
We show that the procedure of deforming M in M commutes with taking the
dual. Suppose rank M = r. We write
(3.23) F = F^  F^ ? ; F 0 = F^ 0  F^ 0? (dim F^ ? = dim F^ 0? = r)
such that M has the form  
(3.24) M = 00 M0? :
At low energies, the gauge group is broken to SU(Nc r), and there are Nf r
light avors. The mass scale ^ of the low energy eective theory is given by
3Nc Nf
(3.25) ^3(Nc r) (Nf r) = det M ? :
In the dual description, the fundamental theory has N~c colors and Nf avors, and
has a scale ~ related to  by (3.20). Now  1 M 2 Hom(F~ 0; F)
~ gives mass to some
34 N. SEIBERG, DYNAMICS OF N = 1 SUSY THEORIES
components of the dual quarks Q, ~ Q~ 0 . The low energy magnetic theory has N~c
colors and Nf r avors. Its scale ^~ is given by
~3N~c Nf det M ?
(3.26) ^~3N~c (Nf r) =  r :
It follows from (3.20), (3.25), (3.26) that
(3.27) ^3(Nc r) (Nf r) ^~3N~c (Nf r) = ( 1)N~c Nf r :
Thus the low energy magnetic theory is dual to the low energy electric theory.
We further check that the magnetic theory has the same space of ground states
M. In the electric theory, the restriction r = rank M  Nc is a classical constraint.
In the magnetic theory, the same constraints are enforced by non-perturbative
quantum eects. In fact, if r > Nc , then the number of avors in the low energy
magnetic theory is Nf r < N~c . In this case, a superpotential of the form (1.30) is
generated, and consequently there is no vacuum. So generically, r = Nc from both
the electric and the magnetic points of view. In this case, the number of avors
of the low energy magnetic theory is Nf r = N~c , equal to the number of colors.
According to (2.20), the light components M, ^~ B,
^~ B~^0 of the dual meson and baryons
satisfy
(3.28) det M^~ B^~
B~^0 = ^~2N~c :
Moreover, M^~ = 0 here because of the equation of motion of M. Using (3.26) and
(3.20) when r = Nc , we obtain
~3Nc Nf M ? ( 1)N~c det M ?
(3.29) ~^2N~c =  Ndet = 3Nc Nf N~c :
c  
By the mapping relation (3.18), we get
(3.30) B^
B^0 = det M ?:
This is precisely the constraint (1.23) in the electric theory. Other constraints are
trivially satised.

x3.3.3. Mass deformation

We now check that mass deformation commutes with taking dual. Suppose a mass
term tr mM with rank m = r is added to the electric theory. Again, we write
(3.23) such that m has the form (2.6). The low energy theory has light quarks
Q^ 2 Hom(F;^ C), Q^0 2 Hom(C; F^ 0) and has a scale ^ given by
(3.31) ^3Nc Nf +r = 3Nc Nf det m? :
M has the form (2.10), where M^ contains the light degrees of freedom. Let F^~ =
F^ 
(^Nf F)1=N~c and F~^0 = F^ 0
(^Nf F 0)1=N~c . Then the equations of motion of
the heavy components of M imply that
^
!
(3.32) Q~ Q~ =
0 M~ 0
0 m?
 LECTURE 3. QUANTUM BEHAVIOR OF SUPER QCD: Nf LARGE 35

for some light dual meson M^~ 2 Hom(F; ^~ F~^0). This breaks the SU(N~c ) gauge sym-
metry to SU(N~c r). We assume that N~c r  2, i.e., r  Nf (Nc + 2).
Integrating out the heavy components in M, Q, ~ Q~ 0 , we obtain a low energy eec-
tive superpotential
(3.33) W^ = 1 tr M^~ M:
^
This is exactly a magnetic theory of N~c r avors and Nf r quarks. Its mass
scale ^~ is given by
~3N~c Nf
(3.34) ^~3(N~c r) (Nf r) = ( )r det m? :
From (3.31), (3.34) and (3.20), we obtain
(3.35) ^3Nc (Nf r) ^~3(N~c r) (Nf r) = ( 1)N~c r Nf r :
So the low energy theory of the magnetic theory after mass deformation is dual to
that of the electric theory.
If r = Nf Nc 1, then in the low energy electric theory, the number of avors
is Nf r = Nc + 1 and in the low energy magnetic theory, the gauge symmetry
is completely broken since N~c r = 1. The light baryons in the magnetic theory
B^~ = Q,
^~ B~^0 = Q~^0 are related to B^ and B^0 by (3.18). The superpotential (3.15) can
be written as
(3.36) 1 ~^0 ^ ^~ 1 ^0 ^ ^
 Q M Q = ^2Nc 1 B M B:
In addition, because the gauge symmetry is completely broken, instanton contri-
bution should be included just as in the case Nf = Nc 1. Now there are more
than N~c 1 zero modes of Q~ , Q~ 0 ; these Nc + 1 extra zero modes are absorbed by
the interaction (3.15), generating a term proportional to the (Nc + 1)-th power of
( 1 M)light in the eective superpotential. So the instanton contribution is
~3N~c Nf det( 1 M)light = ~3N~c Nf det( 1M) ^
= ~3N~c Nf det M^
(3.37)
~ heavy
det(Q~ 0Q) det( m? ) ( 1)N~c +1 Nf det m?
= 3Nc det M^ det M^
Nf det m? = ^2Nc 1 ;
where (3.32), (3.20), (3.31) have been used. Combining (3.36) and (3.37), we get
the total eective superpotential
(3.38) We = ^2N1c 1 (B^0 M^ B^ det M); ^

which is exactly (2.24).
There are many other checks of duality. We remark here that a theory with
less avors can be regarded as the deformation of a larger theory by a mass term.
This duality is related to various phenomena in N = 2 theories and to Langlands
duality in N = 4 theories.