Cambridge Part III Maths

Lent 2019

Supersymmetry

based on a course given by written up by

David Skinner Josh Kirklin
Please send errors and suggestions to [email protected].

Contents

1 Introduction 2
1.1 What is SUSY? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Why is SUSY interesting? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 SUSY in zero dimensions 5

2.1 Grassman variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Supersymmetric integrals and localisation . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 d = 0 Landau-Ginzburg theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Supersymmetric Quantum Mechanics 13

3.1 Supersymmetric ground states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Path integrals in quantum mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Path integrals for fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Non-linear sigma models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Supersymmetric NLSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 The Atiyah-Singer index theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Supersymmetric Quantum Field Theory 31

4.1 Superspace in d = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Supersymmetry invariant actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3 Wess-Zumino model in d = 1 + 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Seiberg non-renormalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5 Kähler geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.6 Kähler manifolds and SUSY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1
 1 INTRODUCTION

1 Introduction
Lecture 1
21/01/19
1.1 What is SUSY?

Consider a quantum theory with Hilbert space H, ground state |Ωi ∈ H and Hamiltonian H.
Suppose that this theory contains fermions, i.e. that there are fermionic operators acting on H.
Then H necessarily decomposes as the tensor sum HB ⊕ HF , where HB and HF contain states with
an even and odd number of fermionic excitations respectively. What this means is that |Ωi ∈ HB ,
and all fermionic operators map states in HB to states in HF and vice versa.
Such a theory is supersymmetric if it supports the following object:

Definition. A supercharge Q is a fermionic operator satisfying

n o
Q, Q† = 2H, Q2 = Q† 2 = 0.

Here {X, Y } = XY + Y X is the anticommutator of X and Y , and Q† is the Hermitian conjugate

of Q with respect to the inner product on H. The factor of 2 in front of the Hamiltonian is a useful
convention.
There are a couple of immediate consequences of the existence of a supercharge. First, the
commutator of the Hamiltonian and a supercharge vanishes, since we have
hn o i
2[H, Q] = Q, Q† , Q
= (QQ† + Q† Q)Q − Q(QQ† + Q† Q)
= QQ† Q − QQ† Q = 0,

where the third line is reached by using Q2 = 0. This means that Q is conserved, and that the
transformations it generates are symmetries. The unitary operator implementing such symmetries is
written exp(iη Q), where in order for the exponent to be bosonic, η must be a fermionic parameter.
Whenever an object is parametrised by fermionic parameters, a common naming pattern in
physics is to attach the prefix “super-”. Therefore, we refer to the symmetries generated by Q as
supersymmetries.
Second, for any state |ψi ∈ H, the expectation value of the energy is non-negative:

hHi = hψ|H|ψi
1
= hψ|(QQ† + Q† Q)|ψi
2
1 † 2 1
= Q |ψi + kQ |ψik2 ≥ 0.
2 2
Furthermore, equality hHi = 0 is attained if and only if Q† |ψi = Q |ψi = 0, i.e. if the state ψ is
invariant under the supersymmetry transformations. We call such a state supersymmetric. Since
the ground state must have zero energy, it must be supersymmetric.

2
1.2 Why is SUSY interesting? 1 INTRODUCTION

1.2 Why is SUSY interesting?

Phenomenology

In QFT, the coupling constants α of interactions depend on the energy scale Λ at which we
are making observations. One says that the coupling constants run with the energy scale, and
understanding this running requires studying renormalisation. In the case of the Standard Model,
three interesting coupling constants are the fine structure constants of the strong, weak and
electromagnetic forces. One finds that these three constants come very close to being equal at a
certain energy scale ΛGUT ≈ 1016 GeV.

1/α

EM

weak

g
stron

ΛGUT log Λ

The three constants do not appear to be exactly equal at this scale. But there is a reason that we
might expect them to be. Matter in the standard model transforms in representations of the gauge
group SU(3) × SU(2) × U(1). In fact, all of the matter in the standard model can be considered
as a single representation of the larger group SO(10), which contains SU(3) × SU(2) × U(1) as a
subgroup. It is far from the case that such an extension can always be done, and therefore it seems
natural that perhaps SO(10) is the ‘true’ unified gauge group of reality, which is broken down to
SU(3) × SU(2) × U(1) at some energy scale. But for this to be the case, the three fine structure
constants should exactly agree at ΛGUT .
So we might want to change the standard model such that this exact agreement does happen –
the result of such a change is referred to as a GUT, or ‘Grand Unified Theory’. One way of making
such a change is to implement a supersymmetric extension to the Standard Model. There is a
broad class of such supersymmetric extensions.
Another phenomenological reason to study SUSY is as follows. Various sources of evidence (e.g.
galactic rotation curves) point to the existence of ‘dark matter’ which interacts very weakly with
standard model particles. We don’t know what dark matter is yet, but the particles necessary for a
SUSY extension of the Standard Model provide candidates.
A further phenomenological reason (which has been called into question by observations at the
LHC) is that SUSY may shed light on the Higgs mass. The Higgs boson is a scalar particle, and
we expect such particles to recieve large quantum corrections to their masses, of order the Planck
scale. However, according to observations the mass of the Higgs does not appear to recieve these
corrections, so we are forced to ask what protects it from them. There are certain supersymmetric
extensions of the Standard Model that could do this – but other observations at CERN have ruled

3
1.2 Why is SUSY interesting? 1 INTRODUCTION

these extensions out. (Note that these observations do not rule out supersymmetry entirely. They
just mean that it cannot explain the Higgs mass.)

Theory

The argument could be made that the above phenomological motivations are secondary to the
following theoretical motivations.
QFT is hard, and SUSY makes it easier to understand. When learning quantum mechanics, one
starts with the formalism, and then proceeds to analyse easy basic models such as the harmonic
oscillator and the hydrogen atom. Then one finds out that these models are unrealistic, and uses
perturbation theory to study deformations of them which are more realistic. The reason the basic
models were easy is because they were idealised and contained a lot of symmetry. The perturbation
theory enables us to study approximations to the consequences of breaking this symmetry.
In QFT, the only ‘idealised’ system one studies initially is free theory, with no interactions. Free
theory is fine, but it is not obvious that it is a good approximation to reality. More importantly,
free theory only allows us to study a small set of aspects of QFT.
SUSY allows us to do better. When we have supersymmetry we can compute some quantities
exactly. This is almost always only true when studying toy models, with no pretense of realism, but
this is a good starting poing. These quantities often reveal extremely mathematically interesting deep
connections between QFT, geometry, and topology. Many modern breakthroughs in mathematics
(e.g. mirror symmetry, Seiberg-Witten theory, Donaldson invariants...) are motivated by this point
of view.
In this course, we will focus on these more theoretical and mathematical aspects, and not
emphasise so much the phenomonology of supersymmetry.

Path integrals

Consider the following Euclidean signature path integral:

Z
Z= e−S[X]/~ DX .
C

X is some field, S[X] is the action, and C is the space of all possible field configurations. The
configuration space C is an infinite-dimensional function space, which we need to be able to integrate
over for the above expression to make sense. Such an integral is not often well-defined, and even if
it were it is not always entirely obvious what it would mean.
The exact evaluation of a path integral is a formidable goal. As a very simple toy, suppose the
whole universe M is just a single point. Then a field could be a map from that point to R. The
path integral would then just be a normal integral over the values taken by X:
Z
Z= e−S[X]/~ dX .
R

Suppose the action takes the following very simple form:

m 2 λ 4 g
S[X] = X + X + X 6.
2 4! 6!

4
 2 SUSY IN ZERO DIMENSIONS

m might be interpreted as the mass of X, and λ and g might be interpreted as higher order coupling
constants. Even with such a simple action, the path integral is hard to evaluate explicitly. It is
only easy if λ = g = 0, as then we have a Gaussian integral. But this would just be a free theory,
which we are not particularly interested in.
One thing we can do is obtain an asymptotic series for Z in the semiclassical limit ~ → 0. If
S[X] has an isolated minimum at X0 ∈ R, one can approximate the path integral with a saddlepoint
approximation as
!−1/2
~→0 −S[X0 ]/~ ∂2S
Z ≈ e| {z } × × (1 + O(~)) .
∂X 2 | {z }
tree level | {z } higher loops
one loop

Each of the factors above come in at various loop orders in a Feynman diagram approach to the
path integral. This series is at most asymptotic, as it can’t possibly converge. If it did, it would
have to converge for ~ in an open neighbourhood of 0 in the complex plane. But for negative ~,
the path integral clearly blows up. So this semiclassical expression can be viewed at best as an
approximation.
Supersymmetry allows us to do better than this approximation. This course will not be focussed
only on classical supersymmetry, just adding new particles to a theory to make it supersymmetric,
perhaps employing a superspace formalism. This can quickly become messy and complicated. We
will try to understand why it is worthwhile, and in particular why doing it makes the path integral
easier.

2 SUSY in zero dimensions

2.1 Grassman variables

Definition. A set of n Grassman variables are elements ψ a , a = 1, . . . , n which obey

ψ a ψ b = −ψ b ψ a .

In particular (ψ a )2 = 0. Grassman variables commute with bosonic variables x (e.g. real numbers):

ψ a x = xψ a .
n o
Grassman variables are vaguely reminiscent of the Dirac field ψ α (x), ψ β (y) = 0, and differen-
tial forms dxa ∧ dxb = − dxb ∧ dxa . This will turn out to not be a coincidence.
Any function F (ψ) of Grassman variables has a terminating expansion:

F (ψ) = f + ρa ψ a + φab ψ a ψ b + · · · + gψ 1 ψ 2 . . . ψ n .

Note that we can take the coefficients to be antisymmetric in their indices, e.g. φab = −φba . If F (ψ)
is bosonic (commuting), then the even coefficients f, φab , . . . must also be bosonic, and the odd
coefficients ρa , . . . must be fermionic.

5
 2.1 Grassman variables 2 SUSY IN ZERO DIMENSIONS

Definition. Derivatives with respect to Grassman variables are defined by

∂  b  ∂
a
ψ . . . = δab (. . . ) − ψ b a (. . .).
∂ψ ∂ψ

This is reminiscent of the product rule, but with a minus sign instead of a plus sign, which
Lecture 2 reflects the fact that ∂ψ∂ a is fermionic and so anticommutes with ψ b .
23/01/19
We can also define integration
R
with
R
respect to Grassman variables. Note that since ψ 2 = 0,
it
R
is sufficient toR define 1 dψ and ψ dψ. We want our R
integral to be ‘translation invariant’,
R
i.e.
(ψ + η) dψ = ψ dψ for any η. Then we must have η dψ = 0, which implies that 1 dψ = 0.
Another way we could arrive at this condition is by insisting that the fundamental theorem of
calculus holds, Rand supposing
R ∂
that the range of integration behaves
R
as if it has no boundary. Then
we would have 1 dψ = ∂ψ ψ dψ = 0. It remains to define ψ dψ, and doing so is essentially just
a normalisation convention. We will use the standard choice:

Definition. Berezin integration is defined by

Z Z
1 dψ = 0, ψ dψ = 1.

Note that performing a Berezin integral actually just gives the same result as doing a derivative,
∂ ∂
because ∂ψ 1 = 0 and ∂ψ ψ = 1.

Suppose we have n fermions ψ 1 , . . . , ψ n . We have

Z
ψ 1 ψ 2 . . . ψ n dψ n dψ n−1 . . . dψ 1 = 1.
| {z }
dn ψ

Because the ψ a anticommute, the order of the measures dψ a matters here – they may be taken to
be anticommuting. In fact, we may write
Z
ψ a1 ψ a2 . . . ψ an dn ψ = a1 a2 ...an ,

where a1 a2 ...an is the totally antisymmetric symbol in n indices.

Now consider a change of variables ψ 0a = N ab ψ b , where N ∈ GL(n) is some n × n matrix. We
have
Z Z
0a1 0a2 0an
ψ ψ ...ψ n
d ψ= N a1b1 N a2b2 . . . N anbn ψ b1 ψ b2 . . . ψ bn d n ψ

= N a1b1 N a2b2 . . . N anbn b1 b2 ...bn

= det(N )a1 a2 ...an
Z
= det(N ) ψ 0a1 ψ 0a2 . . . ψ 0an dn ψ 0 .

This implies that under this change of variables the measure changes as dn ψ 0 = det(N
1 n
) d ψ. Note
that this is the opposite way round to what happens in bosonic integration, where we would have
dn x0 = det(N ) dn x if x0a = N ab xb .

6
2.2 Supersymmetric integrals and localisation 2 SUSY IN ZERO DIMENSIONS

Example. Consider a single fermion ψ, and the change of variables χ = aψ. One has
Z Z
χ dχ = 1 = aψ dχ ,

which implies that dχ = dψ /a.

In QFT, we will often need to perform Gaussian integrals. Suppose ψ 1 , ψ 2 are fermions, and let
the ‘action’ be S(ψ a ) = 12 ψ 1 M ψ 2 . One can think of this as a kinetic mass term (we are in zero
dimensions, so there cannot be any derivatives). Then the path integral is

1 1
Z Z  
−S(ψ a ) 1 2
e dψ dψ = 1 − ψ 1 M ψ 2 dψ 1 dψ 2 = M,
2 2

where the series expansion for the exponential truncates at its first non-trivial term because (ψ 1 )2 =
(ψ 2 )2 = 0. More generally, suppose we have 2m fermions ψ a , and the action is S(ψ a ) = 12 ψ a Mab ψ b ,
where Mab is a 2m × 2m which may be taken to be antisymmetric. Then
∞
(−1)k
Z Z X
a
e−S(ψ ) d2n ψ = (ψ a Mab ψ b )k d2m ψ
2k k!
Note that theRonly contributing term is at k = m. This is because for k < m there will be at least
one factor of 1 dψ a = 0, and for k > m there will be at least one factor of (ψ a )2 = 0. So

(−1)m
Z Z
a
e−S(ψ ) d2n ψ = (ψ a Mab ψ b )m d2m ψ
2m m!
(−1)m a1 b1 ...am bm
= m  Ma1 b1 . . . Mam bm
q2 m!
= det(M ) = Pfaff(M ).

So the path integral works out to the square root of the determinant of M , sometimes known as
the Pfaffian. Note that this is again opposite to the bosonic case, where we would get

(2π)m (2π)m
Z
1 a b
e− 2 x M ab x
d2m x = p = .
det(M ) Pfaff(M )

2.2 Supersymmetric integrals and localisation

Consider a d = 0 theory of one bosonic variable x and two fermions ψ 1 , ψ 2 . The action must be
bosonic and so generally takes the form

S(x, ψ a ) = V (x) − ψ 1 ψ 2 U (x),

where V, U are functions of x. If we imagine that V, U are polynomials, then the quadratic term in
V is the kinetic term for x, and higher order terms give interactions. Even in d = 0, for generic
U, V the path integral Z
a
e−S(x,ψ ) dx dψ 1 dψ 2

is difficult.

7
2.2 Supersymmetric integrals and localisation 2 SUSY IN ZERO DIMENSIONS

However, suppose we choose a polynomial W (x), and take

1
S(x, ψ, ψ̄) = (∂W )2 − ψ̄ψ∂ 2 W,
2
∂
where ψ = ψ1 + iψ2 , ψ̄ = ψ1 − iψ2 , and ∂ = ∂x . Then the functions V = 12 (∂W )2 , U = ∂ 2 W are
now related. We can view this action as more symmetric than the generic case, and in fact this
extended symmetry is an example of supersymmetry, as we will now demonstrate. Consider the
following variation:

δx = ψ − ¯ψ̄,
δψ = ¯∂W,
δ ψ̄ = ∂W,

where , ¯ are some fermionic parameters. The action S(x, ψ, ψ̄) is invariant under this transforma-
tion, since
1
δS = δ (∂W )2 − δ ψ̄ψ∂ 2 W



2
1
= δx∂ (∂W )2 − δ ψ̄ψ∂ 2 W − ψ̄δψ∂ 2 W − ψ̄ψδx∂ 3 W


2
= (ψ − ¯ψ̄)∂W ∂ 2 W − ψ∂W ∂ 2 W + ¯ψ̄∂W ∂ 2 W − ψ̄ψ(ψ − ¯ψ̄)∂ 3 W.

The first three terms in the last line cancel each other, while the other vanishes because ψ 2 = ψ̄ 2 = 0.
Hence δS = 0. We will write δ = Q + ¯Q̄, so that

Qx = ψ, Q̄x = −ψ̄,
Qψ = 0, Q̄ψ = ∂W,
Qψ̄ = ∂W, Q̄ψ̄ = 0.

Q, Q̄ are the generators of the above transformation, and may be written as vector fields
∂ ∂ ∂ ∂
Q=ψ + ∂W , Q̄ = −ψ̄ + ∂W .
∂x ∂ ψ̄ ∂x ∂ψ
n o
These generators obey Q, Q̄ = 0. Furthermore, they obey Q2 = Q̄2 = 0 when the equations
ψ∂ 2 W = ψ̄∂ 2 W = 0 hold, which are just the equations of motion for the ψ, ψ̄ fields. We say that
Q2 = Q̄2 = 0 ‘on-shell’. Later on, when we use the superspace formalism, we will see that this
on-shell condition becomes unnecessary. Note that in 0 dimensions the Hamiltonian H must vanish,
because there is no concept of time evolution. Therefore, we have (on-shell)
n o
Q, Q̄ = 2H, Q2 = Q̄2 = 0.

So Q, Q̄ are supercharges, and hence they generate supersymmetries.

The presence of these supercharges makes the path integral much easier. Consider the super-
symmetric path integral Z
e−S(x,ψ,ψ̄) dx dψ dψ̄ ,

and suppose we rescale W → λW , for some λ ∈ R+ . Then the action changes

1
S → Sλ = λ2 (∂W )2 − ψ̄ψλ∂ 2 W,
2

8
 2.2 Supersymmetric integrals and localisation 2 SUSY IN ZERO DIMENSIONS

and so do the supercharges

∂ ∂ ∂ ∂
Q → Qλ = ψ + λ∂W , Q̄ → Q̄λ = −ψ̄ + λ∂W .
∂x ∂ ψ̄ ∂x ∂ψ
R −S
Consider the path integral after this rescaling I(λ) = e λ dx d2 ψ. We have
dI ∂ −Sλ
Z
= e dx d2 ψ
dλ ∂λ
Z
λ(∂W )2 − ψ̄ψ∂ 2 W e−Sλ dx d2 ψ


=−
Z
=− Q̄λ (∂W ψ)e−Sλ dx d2 ψ
Z
Q̄λ ∂W ψe−Sλ dx d2 ψ


=−

∂ ∂
But since Q̄λ = −ψ̄ ∂x + (λ∂W ) ∂ψ , the first term is a total derivative in the x integral, while the
dI
second is a total derivative in the ψ integral, and so the whole expression vanishes. So dλ = 0. The
original path integral I(1) is therefore equivalent to I(λ), no matter the value of λ. In particular
we can take Z
e−S(x,ψ,ψ̄) dx dψ dψ̄ = lim I(λ).
λ→∞
2
− λ2 (∂W )2
As λ → ∞, the e factor suppresses the integrand except near where ∂W = 0. So the path
Lecture 3 integral ‘localises’ to the critical points of W (x).
29/01/19
Let’s consider a different, slightly more general perspective. Suppose there is a Lie group G
which acts freely (which means there are no elements g ∈ G other than the identity for which
g(x) = x for all x – so there are no fixed points) on the space of field configurations. Also, let us
assume that both the action S and the integration measure on the space of fields are invariant
under this G-action. In this case, one may decompose the configuration space C = G × C /G, where
C /G is the space of equivalence classes of configurations under the G-action. Then one may carry
out the integral separately over each factor in this decomposition, and the integral over G just gives
a factor of vol(G), i.e. the volume of the group G.
Now consider the case where G is a fermionic group, which means that the elements of G are
parametrised by fermionic variables θ. Then we have vol(G) = 0, because by definition
Z
vol(G) = 1 ddim(G) θ = 0.
G

So the path integral, being proportional to vol(G), vanishes!

The situation changes when there are fixed points x ∈ C , i.e. configurations such that there is a
non-identity g ∈ G satisfying g(x) = x. This is because the decomposition C = G × C /G is then
no longer valid. However, if we exclude the fixed points, then the above construction still works.
As a consequence, we can only get contributions to the path integral from a neighbourhood of the
fixed points. So again the path integral ‘localises’ to a certain region in configuration space.
In our case, the supersymmetry transformation is δψ = ¯∂W , δ ψ̄ = ∂W , and the only fixed
points of this action are the critical points of W (x). Away from such critical points, let us change
coordinates in configuration space, defining y, χ, χ̄ as follows:

ψ̄ψ √
y =x− , χ = ψ ∂W , χ̄ = ψ̄.
∂W

9
2.2 Supersymmetric integrals and localisation 2 SUSY IN ZERO DIMENSIONS

The
p theory of superdeterminants (see the first example sheet) allows us to deduce dx d2 ψ =
2
∂W (y) dy d χ.
The first point of interest is that
δ ψ̄ ψ + ψ̄ δψ ψ̄ψ
δy = δx − + δx∂ 2 W
∂W (∂W )2
∂W ψ ψ̄¯
∂W
= ψ − ψ̄ − −
∂W ∂W
= 0,

where we used the fact that ¯ψ̄ = −ψ̄¯

. So y is invariant under the supersymmetry transformations.
The second point of interest is that
1
S(y, 0, 0) = (∂W (x))2 − ∂ 2 W (x)ψ̄ψ = S(x, ψ, ψ̄).
2
So in these coordinates, the action is independent of the fermionic variables! Thus, if we let U ⊂ C
be an open neighbourhood of the critical points, and let U c be its complement, then the path
integral over U c vanishes:
Z Z q
e−S(c,ψ,ψ̄) dx d2 ψ = e−S(y,0.0) ∂W (y) dy d2 χ = 0.
Uc Uc
So again, the path integral localises to the critical points.
Let us now evaluate the contributions from the critical points. We shall assume that all such
critical points x∗ are isolated, so that
c∗  
W (x) = W (x∗ ) + (x − x∗ )2 + O (x − x∗ )3 .
2
The action, to quadratic order in the fields, is then given by
c2∗
S (2) (x, ψ, ψ̄) =
(x − x∗ )2 − ψ̄ψc∗ + . . .
2
Because of the scaling symmetry W → λW described above, one can always localise to an
infinitesimal neighbourhood of x∗ , and so we are free to ignore all of the higher order terms. Hence
(including now a conventional factor of √12π for all bosonic fields), the contribution to the path
integral from this critical point is given by
dx
Z
I= e−S(x,ψ,ψ̄) √ d2 ψ
2π
c2 dx
Z
∗
= e− 2 (x−x) (−1 + ψ̄ψc∗ ) √ d2 ψ
2π
c∗ c2
Z
∗
=√ e− 2 (x−x) dx
2π R
c∗
= = sign(c∗ ) = ±1.
|c∗ |
P
If W has several critical points, we thus have I = c∗ sign(c∗ ). So the path integral reflects the
topology of the polynomial W :

0

 if W has odd degree
I= −1 if W has even degree and W → −∞ as |x| → ∞

if W has even degree and W → +∞ as |x| → ∞

1

10
2.3 d = 0 Landau-Ginzburg theory 2 SUSY IN ZERO DIMENSIONS

So there are two ‘miracles’. First, we can actually do the path integral, and second, the answer is
simple.

2.3 d = 0 Landau-Ginzburg theory

This is a complex generalisation of the theory described in the previous subsection. Our fields are
now one complex number z ∈ C and two complex fermions ψ1 , ψ2 . We choose a holomorphic W (z),
and set the action to
S(z, ψi ) = |∂W |2 + ∂ 2 W ψ1 ψ2 − ∂ 2 W ψ̄1 ψ̄2 ,
where here ∂ = ∂z is a holomorphic derivative.
This theory is supersymmetric. The supersymmetry transformations are

δz = 1 ψ1 + 2 ψ2 , δ̄z̄ = ¯1 ψ̄1 + ¯2 ψ̄2 ,

δψ1 = 2 ∂W , δ̄ ψ̄1 = ¯2 ∂W,
δψ2 = −1 ∂W , δ̄ ψ̄2 = −¯
1 ∂W,

with all other variations vanishing, δ̄z = δ̄ψi = δz̄ = δ ψ̄i = 0. If we let Qi , Q̄i be the generators of
the transformations with parameters i , ¯i , then one may verify that the supersymmetry relations
n on o
{Qi , Qj , =} Q̄i , Q̄j , = Qi , Q̄j , = 0

hold (on-shell).
By again rescaling W → λW for λ ∈ R+ , one may see that the path integral localises to the
critical points z∗ of W (z). Writing
α∗  
W (z) = W (z∗ ) + (z − z∗ )2 + O (z − z∗ )3 ,
2
the quadratic order action takes the form

S (2) (z, ψi ) = |α∗ |2 |z − z∗ |2 + α∗ ψ1 ψ2 − alpha

¯ ψ̄1 ψ̄2 + . . .
∗

Therefore one may evaluate the path integral:

1
Z
I= e−S(z,ψi ) d2 z d4 ψ
2π
X 1 Z 2
= e−|α∗ (z−z? )| |α∗ |2 d2 z
z∗ 2π
X |α∗ |2 X
= = 1.
z∗ |α∗ |2 z∗

So this time the path integral counts the number of critical points of W .
More generally, let f (z) be any holomorphic function. Then its expectation value1
1
Z
hf (z)i = e−S(z,ψi ) f (z) d2 z d4 ψ
2π
1 1
Note that we are defining this expectation value without a normalising factor of I
.

11
 2.3 d = 0 Landau-Ginzburg theory 2 SUSY IN ZERO DIMENSIONS

is still invariant under the δ̄ transformations, because δ̄f (z) = δ̄z ∂f (z) = 0. So again this integral
will localise to the critical points of W̄ (z̄), and we get
X
hf (z)i = f (z∗ ).
z∗

This of course only works if δ̄f = 0. The holomorphic functions are however not the only ones
satisfying this condition, and it is useful to think about how in general one might construct such
a function. Since δ̄ = ¯1 Q̄1 + ¯2 Q̄2 , what we are really talking about are Q̄i -invariant functions.
Since Q̄2i = 0, one way to construct such functions is simply to take Q̄i Λ, where Λ = Λ(z, z̄, ψj ψ̄k )
is any smooth function. But these functions are actually not very interesting, because

d2 z 4
D E Z
Q̄Λ = (Q̄Λ)e−S d ψ
2π
d2 z 4
Z
= Q̄(Λe−S ) d ψ = 0,
2π
Lecture 4 which vanishes because it is the integral of a total derivative.
31/01/19 D E
A clear consequence is F + Q̄Λ = hF i. So the ‘interesting’ functions are the ones which are
Q̄-invariant, but only up to addition of functions which are of the form Q̄Λ (which we say are
Q̄-exact). This is an example of cohomology. The interesting functions are elements of

ker Q̄
HQ̄ = ,
im Q̄

which is called the Q̄-cohomology. HQ̄ is clearly closed under addition and subtraction, but note
also that the product of a Q̄-exact function with a Q̄-invariant function is Q̄-exact:

F Q̄Λ = Q̄(F Λ),

which implies that HQ̄ is also closed under multiplication. Thus HQ̄ is a ring. It is often called the
chiral2 ring in the context of supersymmetry.
In the case of d = 0 Landau-Ginzburg theory, for an example of a Q̄-exact operator, we need
only look at the transformation rule δ̄ ψ̄i = ¯i ∂W , which implies that ∂W is Q̄ of something. Hence
if an operator O contains ∂W as a factor, then its correlator hOi is trivial (it vanishes.) As a more
specific example, suppose

z n+1
W (z) = − az, ∂W = z n − a.
n+1

Then the non-trivial Q̄-invariant operators are polynomials subject to the identification z n = a,
and these operators form a ring generated by {1, z, z 2 , . . . , z n−1 }, i.e.

C[z]
HQ̄ = .
zn − a

2
‘Chiral’ because we chose Q̄ out of the two possible choices Q, Q̄.

12
 3 SUPERSYMMETRIC QUANTUM MECHANICS

3 Supersymmetric Quantum Mechanics

We now move onto d = 1. We will try to understand this from two perspectives: first in a canonical
framework, and then in a path integral framework. These two perspectives are fundamentally
equivalent, so the results we get will be the same – but they will look quite different. Comparing
the results in the two forms is a powerful tool.
Consider a worldline theory of a single bosonic field x(t) and a single complex fermion ψ(t) with
action
1 2 i ˙ − 1 (∂h)2 − ψ̄ψ∂ 2 h dt ,
Z  
S[x, ψ, ψ̄] = ẋ + (ψ̄ ψ̇ − ψ̄ψ)
2 2 2
where h = h(x(t)) is a function which plays the role of a potential. This theory is supersymmetric.
We see this in the usual way: S is invariant under the following transformation:
δx = ψ̄ − ¯ψ,
δψ = (iẋ + ∂h),
δ ψ̄ = ¯(−iẋ + ∂h).
If we allow the parameters , ¯ to be time-dependent, we find
Z
δS = ˙ + ¯˙Q̄) dt ,
(Q

so by the Noether procedure Q, Q̄ are the generators of the transformations parametrised by , ¯

respectively. We have Q2 = Q̄2 = 0. Also,
n o
Q, Q̄ x = (QQ̄ + Q̄Q)x
= −Qψ + Q̄ψ̄
= −(iẋ + ∂h) + (−iẋ + ∂h) = −2iẋ,
n o
Q, Q̄ ψ = Q̄(iẋ + ∂h),
= −iψ̇ − ψ∂ 2 h,
' −2iψ̇,

where ' means an equality that holds on shell (this particular equality holds o the ψ̄ equation
n when
of motion is satisfied). Thus, up to the fermionic equations of motion, Q, Q̄ generates time
evolution,
n o with a factor of 2. The Hamiltonian generates time evolution by definition, so we have
Q, Q̄ = 2H. So these are supercharges, and the theory is indeed supersymmetric.
To canonically quantise, first we need to move to a classical Hamiltonian mechanical point of
view. The conjugate momenta are
δL
p= = ẋ,
δ ẋ
δL
π= = iψ̄.
δ ψ̇
The Hamiltonian is the Legendre transform of the Lagrangian:
H = pẋ + π ψ̇ − L
1 1 1
= p2 + (∂h)2 + ∂ 2 h(ψ̄ψ − ψ ψ̄).
2 2 2

13
3.1 Supersymmetric ground states 3 SUPERSYMMETRIC QUANTUM MECHANICS

The reason we have written the final term in this way will become apparent.
Upon quantisation, we have (setting ~ = 1)

ψ̂, ψ̄ˆ = 1,
n o
[x̂, p̂] = i,

where a hat indicates that an object is an operator on a Hilbert space H. For the Hilbert space
factor describing x, we can as usual take L2 (R, dx), the space of square integrable functions on the
∂
real line. Then x̂, p̂ are represented by x, −i ∂x respectively. To deal with the fermions, we define a
ˆ
fermionic ‘number operator’ F̂ = ψ̄ ψ̂. We have

F̂ , ψ̄ˆ = +ψ̄ˆ,
h i h i
F̂ , ψ̂ = −ψ̂,

so ψ̄ˆ adds 1 to the number of fermions, while ψ̂ subtracts 1. We let the vacuum of the fermionic
system be |0i, defined by ψ̂ |0i = 0. The first excited state is |1i = ψ̄ˆ |0i. Actually, since ψ̄ˆ2 = 0,
there are no further excited states. Therefore the full Hilbert space is
H = L2 (R, dx) |0i ⊕ L2 (R, dx) |1i = L2 (R, dx) ⊗ C2 .

Now we need to convert the supercharges and the Hamiltonian to their quantum counterparts.
For the supercharges, we can just write

Q̂ = ψ̄ˆ(ip̂ + ∂h),
ˆ = ψ̂(−ip̂ + ∂h),
Q̄
and there is no ordering ambiguity for the various operators involved. For the Hamiltonian, there
is some ordering ambiguity. However, supersymmetry comes to n the orescue here – if we want it to
remain valid in the quantum theory, then we must have 2Ĥ = Q̂, Q̄ ˆ . Using this relation, we find

1 1 1
Ĥ = p̂2 + (∂h)2 + ∂ 2 h(ψ̄ˆψ̂ − ψ̂ ψ̄ˆ).
2 2 2
So the ordering is the same as the one we chose for the classical expression above.
From now on we will drop all hats on quantum operators.

3.1 Supersymmetric ground states

As before, we have hΨ|H|Ψi ≥ 0, with equality if and only if Q |Ψi = Q̄ |Ψi = 0. Thus a state of
zero energy in SQM must be invariant under supersymmetry, and will be a ground state.
Let us represent the fermionic Hilbert space with |0i → 10 , |1i → 01 . Then the full Hilbert

space consists of rank 2 vectors whose entries are square integrable functions on the real line.
Suppose the state |Ψi = fg(x)
(x)

satisfies Q |Ψi = Q̄ |Ψi = 0. Then we may write this as
! ! ! !
∂
0 0 f (x) 0 − ∂x + ∂h f (x)
∂ = =0
∂x + ∂h 0 g(x) 0 0 g(x)

These equations are simple to solve, and we find that a zero energy ground state must be of the
form !
Ae−h(x)
|Ψi = ,
Be+h(x)

14
 3.1 Supersymmetric ground states 3 SUPERSYMMETRIC QUANTUM MECHANICS

where A, B are some constants. For this to really be a state, it must be normalisable – so one of
A, B must vanish. In fact:
Ae−h(x)

• If h(x) → +∞ as |x| → ∞, then the supersymmetric ground state is 0 .
0

• If h(x) → −∞ as |x| → ∞, then the supersymmetric ground state is Be+h(x) .

• If h(x) → ±∞ as x → +∞ but h(x) → ∓∞ as x → −∞, then neither solution is normalisable,

so there is no zero energy state! This is fine, since there is still a ground state, it just has
non-zero energy, and is not supersymmetric. In this case we say that supersymmetry is
Lecture 5 spontaneously broken.
05/02/19
In a supersymmetric theory, excited states (i.e. states with postive energy E > 0) come in pairs.
To see this, let us decompose the Hilbert space into energy eigenspaces,
M
H= Hn where H |ψn i = En |ψn i ∀ |ψn i Hn .
n

We can further decompose each eigenspace Hn = Hn,B ⊕ Hn,F , where the bosonic part Hn,B
contains states with an even number of fermions, and the fermionic part Hn,F contains states
with an odd number of fermions. In particular, Q maps Hn,F to Hn,B , since [Q, H] = 0. Also, Q
annihilates Hn,B . Thus, given a |bi ∈ Hn,B , we have
2En |bi = (QQ̄ + Q̄Q) |bi = Q(Q̄ |bi).
So, for En > 0, |bi is Q of something in Hn,F , namely 2E1 n Q̄ |bi. Similarly, any state |f i in Hn,F with
En > 0 may be written as Q̄ of a state in Hn,B . Thus the two spaces are isomorphic, Hn,B ' Hn,F ,
if En > 0. We say that each pair of excited states has a bosonic and fermionic partner.
This is not the case for states of zero energy, which makes the following a non-trivial object of
study.

Definition. The Witten index of a supersymmetric theory is defined as the number of inde-
pendent bosonic zero energy states minus the number of independent fermionic zero energy
states:
IW = dim H0,B − dim H0,F .

We may write the Witten index as

IW = TrH (−1)F ,
the rationale being that each pair of excited states contribute a +1 and a −1 to this sum, which
cancel cancel. One might be concerned that in the case of an infinite dimensional Hilbert space
this sum is not well defined. A way to get around this is to instead write
 
IW = TrH (−1)F e−βH ,

where now each pair of excited states of energy E contribute a +e−βE and −e−βE to this sum,
which again cancel. This expression has much more reasonable convergence properties.

Definition. The supertrace of an operator O is defined as

 
STr O = Tr (−1)F O .

15
3.2 Path integrals in quantum mechanics 3 SUPERSYMMETRIC QUANTUM MECHANICS

So we may write
IW = STr(e−βH ).
This expression is completely independent of β. We included the e−βH factor to regularise the
trace, but it also has another use – it allows us to make a connection to the path integral.

3.2 Path integrals in quantum mechanics

Consider a free particle travelling on R. The time evolution operator e−iHt becomes e−Hτ under a
Wick rotation t → −iτ . By definition, if our particle is located at y0 at τ = 0, then the amplitude
for it to be found at y1 at τ = β is given by
!
−βH 1 (y0 − y1 )2
hy1 |e |y0 i = Kβ (y1 , y0 ) = √ exp − .
2πβ 2β
This is sometimes called the heat kernel. We may break this evolution into N steps of length
∆τ = β/N , to obtain
Z
hy1 |e−βH |y0 i = hy1 |e−∆τ H |xN −1 i hxN −1 |e−∆τ H |xN −2 i . . . hx2 |e−∆τ H |x1 i hx1 |e−∆τ H |y0 i dN −1 x

Z
= K∆τ (y1 , xN −1 ) . . . K∆τ (x1 , y0 ) dN −1 x
N −1
NY
" #
∆τ xi+1 − xi 2
Z  
(2π∆τ )−N/2
X
= exp − dxi
i=0
2 ∆τ i=1
where xN = y1 and x0 = y0 . Taking the limit as N → ∞ but keeping β fixed (so ∆τ = β/N → 0),
and defining
−1
! " N # NY
Z β
1 2 X ∆τ  xi+1 − xi 2
exp − ẋ dτ Dx = lim exp − (2π∆τ )−N/2 dxi ,
0 2 N →∞
i=0
2 ∆τ i=1
then we have (at least heursitically)
Z β !
1 2
Z
−βH
hy1 |e |y0 i = exp − ẋ dτ Dx ,
C[y0 ,y1 ] 0 2
where C[y0 , y1 ] is the space of all continuous maps x : [0, β] → R such that x(0) = y0 and x(β) = y1 .
p2
A similar derivation

works when

the particle is in a potential, so H = 2 + V (x). In that case,
the action is S = 0β 12 ẋ2 + V (x) dτ .
R

The partition function Z(β) is closely related to the heat kernel. It is defined as
Z
−βH
Z(β) = TrH (e )= hx|e−βH |xi dx .
R
Using the above expression, this may be written
Z "Z #
S[x]
Z(β) = e Dx dy
C[y,y]
Z
= e−S[x] Dx ,
CS 1

where the range of integration in the last line is now the space of continuous maps from the circle
S 1 to the real line.

16
3.3 Path integrals for fermions 3 SUPERSYMMETRIC QUANTUM MECHANICS

3.3 Path integrals for fermions

Consider a Hilbert space for a fermionic theory. We can obtain a path integral in much the same
way as with bosons. However, there are some subtleties of which one should be aware.

Definition. The fermionic coherent states are defined by

|ηi = eψ̄η |0i ,

where ψ̄ is the fermion creation operator, but η is just a c-number fermion.

These states obey the following:

ψ |ηi = η |ηi ,
hχ̄|ηi = eχ̄η ,
Z
IH = e−η̄η |η̄i hη̄| d2 η ,
Z
Tr(A) = h−η̄|A|ηi e−η̄η d2 η ,
Z
STr(A) = hη̄|A|ηi e−η̄η d2 η .

Note the presence of the minus sign in the bra in the fourth line, and its absence in the fifth line.
Using these, and following the same procedure as for the bosons, we have
Z −1
NY
0 −βH 0 −∆τ H −∆τ H
χ̄ e χ = χ̄ e ηN −1 . . . hη̄1 |e |χi e−η̄k ηk d2 ηk .
i=1

By using the commutation relations, we can order the operators that make up the Hamiltonian so
that all ψs appear to the right hand side of all ψ̄s. Then we have (to leading order in ∆τ )

hη̄k+1 |e−∆τ H(ψ̄,ψ) |ηk i = hη̄k+1 |(1 − ∆τ H(ψ̄, ψ))|ηk i

= hη̄k+1 |(1 − ∆τ H(η̄k+1 , ηk ))|ηk i
= e−∆τ H(η̄k+1 ,ηk ) hη̄k+1 |ηk i
= e−∆τ H(η̄k+1 ,ηk ) eη̄k+1 ηk .

Using this, we have

Z N
! N −1
0 −βH
e−η̄k ηk d2 ηk
X Y
χ̄ e χ = lim exp η̄k ηk−1 − ∆τ H(η̄k , ηk−1 )
N →∞
k=1 k=1
N  NY−1
!
ηk − ηk−1
Z X 
= lim exp − η̄k + H(η̄k , ηk−1 ) ∆τ eη̄N ηN d 2 ηk ,
N →∞
k=1
∆τ k=1

where η0 = χ, ηN = χ0 . It is natural to write this as a path integral in the following way:

Z
χ̄0 e−βH χ = e−S[η̄,η] eη̄(β)η(β) Dη Dη̄ ,

17
 3.3 Path integrals for fermions 3 SUPERSYMMETRIC QUANTUM MECHANICS

where the range of integration is over all fermions satisfying η(0) = χ, η(β) = χ0 , and the action is
Z β
S[η̄, η] = (η̄ η̇ − H(η̄, η)) dτ .
0

The partition function is

Z
Z(β) = TrH (e−βH ) = h−χ̄|e−βH |χi e−χ̄χ d2 χ
Z  
= exp −S[ψ̄, ψ] Dψ Dψ̄ .

Because of the minus sign, this integral must be done over all fermions satisfying ψ(τ + β) = −ψ(τ ).
This is known as antiperiodic boundary conditions.
On the other hand, we have
Z
STr(e−βH ) = hχ̄|e−βH |χi e−χ̄χ d2 χ
Z  
= exp −S[ψ̄, ψ] Dψ Dψ̄ ,

where this time there is no minus sign, and so this integral is done over all configurations with
Lecture 6 periodic boundary conditions, i.e. ψ(τ + β) = ψ(τ ).
07/02/19
So the Witten index has the path integral representation
  Z
IW = STr(e−βH ) = Tr (−1)F e−βH = e−SE [x,ψ,ψ̄] Dx Dψ Dψ̄ ,
periodic

where the Euclidean action

1 2 1
I  
SE = ẋ + ψ̄ ψ̇ + (∂h)2 + ∂ 2 hψ̄ψ dτ
2 2
is invariant under the supersymmetry transformations

δx = ψ̄ − ¯ψ,
δψ = (−ẋ + ∂h),
δ ψ̄ = ¯(ẋ + ∂h).

Note that these transformations make sense globally on S 1 because of the perioricity of x, ψ, ψ̄.
Let’s now compute the Witten index with this path integral. We expect to get the same answer
as before. As in d = 0, we consider a rescaling h → λh for λ ∈ R+ , and consider how the Witten
index changes under this rescaling. We have
d
Z I 
IW (λ) = − λ(∂h) + ∂ hψ̄ψ e−SE [x,ψ,ψ̄] Dx Dψ Dψ̄ .
2 2
dλ
But
dx
I  I  
2 2
Qλ ∂hψ dτ = ∂ hψ̄ψ + λ(∂h) − ∂h dτ .
dτ
Note that ∂h dx dh
dτ = dτ is a total derivative, and h(x(τ )) is periodic in τ , so the last term in the
integrand vanishes. Thus
d
IW (λ) = hQλ (. . . )i = 0.
dλ

18
3.4 Non-linear sigma models 3 SUPERSYMMETRIC QUANTUM MECHANICS

 2

So Iλ is independent of the choice of λ. In particular as λ → ∞, the term exp − λ2 (∂h)2 dτ
H

suppresses all maps x : S 1 → R, except in a neighbourhood of constant maps to critical points of h.

Near such critical points we can write x(τ ) = x∗ + δx(τ ), where δx is small. To quadratic order in
the fields, we then have
I " !  #
1 d2 d

(2)
SE = δx − 2 + h00 (x∗ )2 δx + ψ̄ + h00 (x∗ ) ψ dτ .
2 dτ dτ

Since δx and the fermions must be periodic, they permit Fourier expansions
2πinτ
X  
δx(τ ) = δxn exp ,
n∈Z
β
2πinτ
X  
ψ(τ ) = ψn exp ,
n∈Z
β

where δx−n = δxn , since δx is real. We now find, near a critical point

det(∂τ + h00 (x∗ ))

Z
(2)
e−SE Dδx Dψ Dψ̄ = p
det(−∂τ2 + h00 (x∗ )2 )
 
2πin
+ h00 (x∗ )
Q
n∈Z β
=s  
2
2πn
h00 (x )2
Q
n∈Z β + ∗

h00 (x∗ )
= .
|h00 (x∗ )|
In the first line we used our formulae for Gaussian integrals
 we derived previously. In the second
2πinτ
line, we used the fact that the Fourier modes exp β comprise an eigenbasis of each of the
operators whose determinants we are computing. The third line resulted from noting that all factors
with n 6= 0 cancel each other in the numerator and denominator – in particular, the combined
contributions from n = ±k for k ∈ Z+ cancel to 1.
Therefore,
sign(h00 (x∗ )),
X
IW =
x∗

which is in fact the same answer as the canonical approach.

It is nice that we can study the same object from these two different points of view, but with
this reasonably simple supersymmetric theory, the amount we can learn is limited. By investigating
more involved theories, we can obtain more mathematically interesting results. We start to do this
in the next section.

3.4 Non-linear sigma models

A non-linear sigma model is a theory involving a map x from a space M to a target space N . To
start with we will study the case where M is one-dimensional, and in particular given by [0, β]
or S 1 , but one can also consider higher-dimensional versions – for example, string theory is a
two-dimensional non-linear sigma model. In the one-dimensional case we call M the worldline, and
we will assume that N is a compact Riemannian manifold with metric g.

19
3.4 Non-linear sigma models 3 SUPERSYMMETRIC QUANTUM MECHANICS

x(τ )

M
N

Often we will let xa be coordinates on some patch U ⊂ N , and xa (τ ) the corresponding

fields, where indices a, b, . . . take values in 1, . . . , n = dim N . We can choose the following (non-
supersymmetric) action:
1
Z
S[x] = gab (x)ẋa ẋb dτ .
2
This action gives the length of the image of x. Note that gab (x) generically depends on x, so there
are cubic and higher order in x terms appearing in the action – so this is an interacting worldline
QFT.
If one varies x, the action changes by
d b 1
Z  
δS = gab (x)ẋa
δx + ∂c gab ẋa ẋb δxc dτ
M dτ 2
d 1
Z   h i
= − (gac ẋa ) + ∂c gab ẋa ẋb δxc dτ + gab (x)ẋa δxb .
M dτ 2 ∂M

Hence the classical equations of motion are just the geodesic equations

d 2 xa
+ Γabc ẋb ẋc = 0,
dτ 2
where Γ is the Levi-Civita connection on (N , g). So, classically, the particle travels on a geodesic.
For the quantum version, we have
δL
pq = = gab ẋb ,
δ ẋa
and the canonical quantisation relations are

[x̂a , p̂b ] = iδba .

√
We can choose the Hilbert space to be H = L2 (N , g dn x) (the space of square integrable functions
on N with respect to the volume element induced from the metric gab ). The canonical operators
are then represented as usual by x̂a = xa and p̂a = −i ∂x∂ a .
However, there is no preferred choice of quantum Hamiltonian. Classically, as usual, the
Hamiltonian is given by
1
H = pa ẋa − L = g ab (x)pa pb ,
2
√
but there’s an ordering ambiguity in turning this into a quantum operator on L2 (N , g dn x).
We need some general principles which might fix this ordering ambiguity, and the following are
reasonable choices:

20
 3.5 Supersymmetric NLSM 3 SUPERSYMMETRIC QUANTUM MECHANICS

• Ĥ should be generally covariant.

∂ 2
• Ĥ should reduce to − 21 ∂x n
2 in the case of flat space, (N , g) = (R , η).

• Ĥ should contain no more than 2 derivatives acting on either the wavefunction or the metric.
Although we will not do so here, one can show that there is a 1-parameter family of such Ĥs, given
by !
1 1 ∂ ab √ ∂
 
Ĥ = √ g g b + αR[g] .
2 g ∂xa ∂x
The term on the left is just the Laplacian of the metric g, and R[g] is its Ricci scalar, but α is an
unfixed real number. There is no preferred choice of α, and different choices correspond to different
ways to regularise the path integral.
It is possible to completely eliminate this ambiguity by adding supersymmetry.

3.5 Supersymmetric NLSM

Lecture 7
12/02/19
Let us now define a supersymmetric version of the NLSM introduced in the previous section. As
before, the bosonic fields consist of a map x : M → N . To these we add a set of fermionic fields
ψ a , a = 1, . . . , n. Together these fields can be considered as a single fermionic function on M that
takes values in the tangent space Tx N . We choose the action
1 1
Z  
S[x, ψ] = gab ẋa ẋb + igab ψ̄ a (∇t ψ)b − Rabcd ψ a ψ̄ b ψ c ψ̄ d dt ,
M 2 2
where ∇ is the (pullback of) the Levi-Civita connection on N , so that
dψ a dxb c
∇t ψ a = + Γabc ψ ,
dt dt
and Rabcd is its Riemann curvature.
This action is invariant under the following supersymmetry transformations (again, the origin of
these transformations will become clear when we have learnt about superfields and superspace):
δxa = ψ̄ a − ¯ψ a ,
δψ a = (iẋa − Γabc ψ̄ b ψ c ),
δ ψ̄ a = ¯(−iẋa − Γabc ψ̄ b ψ c ).
Using the Noether procedure, one may deduce that the transformations parametrised by , ¯ are
generated by the supercharges Q, Q̄ respectively, where
Q = iψ̄ a (gab ẋb + igbc ψ̄ b Γcad ψ d ),
Q̄ = −iψ a (gab ẋb + igbc ψ b Γcad ψ̄ d ).

There is another more obvious symmetry of the action given by ψ a → eiα ψ a , ψ̄ a → e−iα ψ̄ a . It
is generated by F = gab ψ a ψ̄ b , and the conservation of F in the quantum theory means that no
fermionic excitations are created or destroyed by time evolution.
Now let’s quantise. The conjugate momenta are
δL δL
pa = a
= gab ẋb + igbc ψ̄ b Γcad ψ d , πa = = igab ψ̄ b ,
δ ẋ δ ψ̇ a

21
3.5 Supersymmetric NLSM 3 SUPERSYMMETRIC QUANTUM MECHANICS

which leads the canonical commuation and anticommutation relations

[x̂a , p̂b ] = iδba ,

ψ̂ a , ψ̄ˆb = g ab (x),
n o

with all others vanishing.

√
For the bosonic part of the Hilbert space, we can use L2 (N , g dn x), as before. For the fermions,
we proceed as previously, choosing ψ̄ˆa to be a raising operator, and ψ̂ a to be a lowering operator.
The vacuum is the unique state such that ψ̂ a |0i = 0 for all a. All other fermion states are generated
by acting with ψ̄ˆs on |0i.
There is a remarkable geometric interpretation to the fermionic states. We interpret them as
forms on N . In particular, we make the following identifications (dropping now the hats):

|0i ↔ 1,
ψ̄ |0i ↔ dxa ,
a

ψ̄ a ψ̄ b |0i ↔ dxa ∧ dxb ,

and so on. One can then identify the overall Hilbert space of this supersymmetric NLSM as
n
H = Ω• (N ) =
M
Ωp (N ),
p=0

i.e. the space of all forms (technically with square-integrable coefficients) on N . Normally, when
considering forms, we do not consider addition of a p-form to a q-form if p = 6 q. We will allow such
an operation here, and the result is just a form of mixed degree. For example, a state Ψ ∈ H may
be written
Ψ(x, ψ̄) = f (x) + αa (x)ψ̄ a + βab (x)ψ¯a ψ¯b + . . . ,
where f is a function on N , α = αa dxa is a 1-form on N , β = βab dxa ∧ dxb is a 2-form on N , and
so on.
Acting on this space, the canonical operators x̂a , p̂a , ψ̄ˆa , ψ̂ a are represented in the following way:

x̂a → xa ×,
∂
p̂a → −i a ,
∂x
ψ̄ˆa → dxa ∧,
ψ̂ a → ιgab ∂ .
∂xb

Here, ιV ω is the form obtained by contracting the vector V into the form ω.
The inner product on H is just the Hodge inner product. In particular, if α, β are two p-forms,
then Z
hα|βi = ᾱ ∧ ∗β,
N
where ∗ is the Hodge star associated to the metric gab . In coordinates, if β = βa1 ...ap dxa1 ∧ · · · ∧ dxap
then we have
1 √ a1 ...ap bp+1
∗β = g bp+1 ...bn βa1 ...ap dx ∧ · · · ∧ dxbn .
(n − p)!

22
 3.5 Supersymmetric NLSM 3 SUPERSYMMETRIC QUANTUM MECHANICS

The Hodge inner product may be written

√
Z
hα|βi = ᾱa1 ...ap βa1 ...ap g dn x .

If α, β have differing degree, then we define hα|βi = 0.

Furthermore, in the quantum theory, the supercharges are represented by the exterior derivative
and its adjoint:
∂
Q = iψ̄ a pa → dxa = d,
∂xa
Q̄ = −iψ a pa → d† .
D E
Lecture 8 The adjoint of d is defined through α d† β = hdα|βi.
14/02/19
Recall that in the non-supsersymmetric theory there was an ordering ambiguity in the choice
of the quantum Hamiltonian. Now that we have supersymmetry, we can fix this ambiguity by
demanding that the quantum theory hasn theo same supersymmetry as the classical theory. In
ˆ = Ĥ should still hold. In terms of the geometric
particular, the supersymmetry algebra Q̂, Q̄
interpretation, this fixes the quantum Hamiltonian to be Ĥ = 12 ∆, where

∆ = d† d + dd† .

The operator ∆ is known as the Laplace-de Rham operator. Since d raises the degree of a form
by 1, and d† lowers the degree of a form by 1, ∆ preserves the degree of the form it is acting on.
Acting on functions, it may be shown that
1  √ 
∆f = √ ∂b g ab g∂a f ,
g

which is just the Laplacian of f . ∆ should be viewed as that natural generalisation of the usual
Laplacian acting on functions (i.e. 0-forms), to an operator that acts on all forms.
Consider the expectation value of the energy of a state |ωi, where ω is a form. We have
1D E 1D E
hω|Ĥ|ωi = ω d d† ω + ω d† dω
2 2
1D † E 1
= d ω d† ω + hdω|dωi
2 2
1 † 2 1
= d ω + kdωk2 ≥ 0.
2 2

So energy is non-negative. Also, iff Ĥω = 0, i.e. iff |ωi is a supersymmetric ground state, then
dω = d† ω = 0.
A form obeying ∆ω = 0 is said to be harmonic. Therefore, supersymmetric ground states are in
one-to-one correspondence with harmonic forms of any degree on the target space N . In notation,
this space is denoted
n
Harm• (N ) =
M
Harmp (N ),
p=0
p
where Harm (N ) is the space of harmonic p-forms on N .

23
3.5 Supersymmetric NLSM 3 SUPERSYMMETRIC QUANTUM MECHANICS

We are assuming that the metric gab on N is of Riemannian signature, i.e. that it is positive
definite. In this case, the space of Harmonic forms is very closely related to certain topological
properties of N . The study of this connection is known as Hodge theory, and the main result is the
following:
Theorem 1 (Hodge’s theorem). The space of Harmonic p-forms is in one-to-one correspondence
with the pth de Rham cohomology:
p
Harmp (N ) ' HdR (N ).

As a reminder:

Definition. The pth de Rham cohomology is the space of exact p-forms modulo closed p-forms:

p {ω ∈ Ωp (N ) s. t. dω = 0}
HdR (N ) = .
{dα s. t. α ∈ Ωp−1 (N )}

We may concisely write this as

p ker dp
HdR (N ) = ,
im dp−1
where dq : Ωq (N ) → Ωq+1 (N ) is the restriction of the exterior derivative to only acting on q-forms.
A harmonic form ω must be closed. In de Rham cohomology, two closed forms are considered
equivalent if they differ by an exact form, so ω ∼ ω + dα. Hodge’s theorem works because the
co-closure condition d† ω works to select a unique representative of the resulting equivalence class.
In particular, if dω = d† ω = 0, then we can only rewrite ω ∼ ω + dα if d† dα = 0. But one may
show that the only solution to this equation is α = 0.
So the space of supersymmetric ground states is equivalent to
n
• p
M
HdR (N ) = HdR (N ).
p=0

But how does de Rham cohomology tell us about the topology of N ? Consider the space of
submanifolds of N , and let the boundary operator δ map a submanifold to its boundary. ∂ maps
p-dimensional submanifolds to (p − 1)-submanifolds. Moreover, the boundary of any submanifold
has no boundary, so ∂ 2 = 0. So we can define:

Definition. The pth cohomology group of N is

ker ∂p
Hp (N ) = ,
im ∂p+1

where ∂q is the restriction of the boundary operator to acting on q-dimensional submanifolds.

In this cohomology, two submanifolds without boundary are considered equivalent if they differ
by the boundary of a submanifold.
Recall:

24
3.5 Supersymmetric NLSM 3 SUPERSYMMETRIC QUANTUM MECHANICS

Theorem 2 (Stokes’ theorem). If U is a (p + 1)-dimensional submanifold, and ω is a p-form, then

Z Z
dω = ω.
U ∂U

Stokes’ theorem indicates that there is a connection between the exterior derivative d and the
boundary operator ∂. It is possible to exploit this connection to prove the following:
Theorem 3 (de Rham’s theorem). The pth de Rham cohomology and the pth cohomology group
are in one-to-one correspondence:
p
HdR (N ) ' Hp (N ).

So, in fact, the space of supersymmetric ground states is equivalent to

n
M
H• (N ) = Hp (N ).
p=0

The cohomology groups of N tells us many things about its topology. For example:

Definition. The kth Betti number bk of N is the rank of its kth homology group.

One may think of bk as the number of independent ways one may cut N along a k-dimensional
surface, without increasing the number of connected components of N .

Example. The Betti numbers of a Riemann surface Σg of genus g are

b0 = 1, b1 = 2g, b2 = 1.

Definition. The Euler characteristic χ(N ) of N is given by the alternating sum of its Betti
numbers: n X
χ(N ) = (−1)k bk .
k=0

Example. The Euler characteristic of Σg is 2 − 2g.

In our theory of a supersymmetric non-linear sigma model, the Betti numbers have a particular
interpretation. Hp (N ) corresponds to the space of ground state p-forms, i.e. ground states with p
fermionic excitations. The rank of Hp (N ) tells us how many independent such states there are.
Thus, the kth Betti number tells us how many ground states there are with k fermions.
Let us now compute the Witten index. Using the above, we have
 
IW = Tr (−1)F e−βH
= # bosonic ground states − # fermionic ground states
n
X
= (−1)k bk = χ(N ).
k=0

25
 3.6 The Atiyah-Singer index theorem 3 SUPERSYMMETRIC QUANTUM MECHANICS

So the Witten index is none other than the Euler characteristic of N !

As we have seen previously, one can compute the Witten index another way, using the periodic
path integral. We thus may write
Z
χ(N ) = e−S[x,ψ,ψ̄] Dx Dψ Dψ̄ ,

where all fields are taken to be periodic with period β.

It is instructive to see how IW = χ is obtained from the path integral perspective. Note that
the action itself is supersymmetrically trivial, in that it is Q̄-exact:
1 1 1
I  
S= gab ẋa ẋb + gab ψ̄ a ∇t ψ b + Rabcd ψ̄ a ψ b ψ̄ c ψ d dt
2 2 4
1 ab a b
I 
= Q̄ g ψ̄ (ix + Γbcd ψ̄ c ψ d ) dt .
2
Lecture 9
19/02/19
An infinitesimal change in β amounts to an insertion of the Lagrangian in the path integral.
Consequently, the path integral is independent of β (of course, we already knew this had to be true
by construction). In particular, if we expand
xa (τ ) = xa0 + δxa (τ ),
ψ(τ ) = ψ0a + δψ a (τ ),
where I I
a
δx dτ = δψ a dτ = 0,
then as β → 0, all the contributions from the non-zero modes δxa , δψ a are highly suppressed. In
fact, one can show that the contributions from δxa , δψ a precisely cancel each other, and one is
left with just an integral over the zero-modes xa0 , ψ0a . In other words, the path integral localises to
constant maps. Since there is no preferred point in N , we still need to integrate over N , as well as
the zero-modes of the fermions. Thus, the path integral becomes
dn x0 n
Z
χ(N ) = e−S[x0 ,ψ0 ] n/2
d ψ0 dn ψ̄0
(2π)
1 dn x0 n
Z  
= exp − Rabcd (x0 )ψ̄ a ψ b ψ̄ c ψ d d ψ0 dn ψ̄0
2 (2π)n/2

0 if n odd,
= 1 R

(2π)n/2 N tr(R ∧ R ∧ · · · ∧ R) if n even,

where Rab = Rcdab dxc ∧ dxd is the curvature 2-form, and there are n/2 of these appearing in
the last line. The last line is a well-known expression – it is the n-dimensional generalisation of
the Gauss-Bonnet formula, and gives exactly the Euler characteristic of N . So the path integral
approach agrees with the canonical approach.

3.6 The Atiyah-Singer index theorem

Let us consider the case where N has even dimension n = 2m, and consider again the supersymmetric
non-linear sigma model of the previous section, but now with the restriction that ψ a is real. Then
ψ a = ψ̄ a , and the action simplifies to
1 i
I  
S[x, ψ] = gab ẋa ẋb + gab ψ a ∇τ ψ b dτ ,
2 2

26
3.6 The Atiyah-Singer index theorem 3 SUPERSYMMETRIC QUANTUM MECHANICS

where the term involving the Riemann tensor now vanishes due to the Bianchi identity Ra[bcd] = 0.
This action is still invariant under supersymmetry transformations which respect the restriction
to real ψ a . One can obtain such transformations by setting  = −¯
. They take the form

δxa = ψ a , δψ a = −ẋa .

The momenta conjugate to xa , ψ a are

δL i
pa = = gab ẋb + ψc Γcab ψ b ,
δ ẋa 2
δL
πa = = ψa ,
δ ψ̇ a

where ψa = gab ψ b . The canonical commutators therefore take the form

[x̂a , p̂b ] = iδba ,

n o
ψ̂ a , ψ̂ b = 2g ab .

1
This theory is sometimes referred to as a N = 2 theory, since there is half as much supersymmetry.
Before continuing with our analysis of this theory, it is useful to recall some basic facts about
spinors in n = 2m dimensions. We assume the existence of the following set of matrices:

Definition. The Dirac γ matrices are a set of n linear operators γ i , i ∈ 1, . . . , n on the space
of spinors obeying the anticommutation relations
n o
γ i , γ j = 2δ ij .

In a curved space, one should choose an orthonormal frame, and the indices i, j, . . . should be
understood as frame indices. We can construct m raising and lowering operators (over C) from the
γ matrices:
I 1
γ± = (γ 2I ± iγ 2I−1 ), I = 1, . . . , m,
2
which obey n o n o
I J
γ+ , γ− = δ IJ , I
γ± J
, γ± = 0.

Starting from a spinor χ obeying γ− I χ = 0 for all I, one can construct a basis of the space S of

spinors by acting with any combination of the raising operators γ+ I . Since γ I 2 = 0, each γ I can
+ +
act at most once, so dim(S) = 2n/2 .
The group Spin(n), which is defined as the double cover of SO(n), acts on these spinors via the
generators
1h i
Σij = − γ i , γ j ,
4
which obey the SO(n) algebra
h i  
Σij , Σkl = i δ ik Σjl + δ jl Σik − δ jk Σil − δ il Σjk .

This is not an irreducible representation, and it is useful to decompose it.

27
3.6 The Atiyah-Singer index theorem 3 SUPERSYMMETRIC QUANTUM MECHANICS

Definition. The n + 1-th γ matrix is

γ n+1 = in/2 γ 1 γ 2 . . . γ n .

We have n o
(γ n+1 )2 = 1, γ n+1 , γ i = 0.

Moreover, γ n+1 transforms trivially under the above action of Spin(n), since
h i
γ n+1 , Σij = 0.

In other words, it is a Lorentz scalar. We may decompose S = S + ⊕ S − , where S ± are the ±1

eigenspaces of γ n+1 , and correspond to states constructed from an even/odd number of raising
operators γ+ I respectively acting on χ. A spinor in one of S ± is called chiral. Spinors in S + , S −

are called left-handed, right-handed respectively.

Definition. The Dirac operator is

i∂/ = iγ i ∂i .

The Dirac operator anticommutes with γ n+1 , and thus decomposes as

!
0 ∂+
i∂/ = ,
∂− 0

where with this matrix notation the first row/column corresponds to S − , while the second
row/column corresponds to S + . The operators ∂ ± map S ± → S ∓ respectively, and ∂ ± anni-
hilates S ∓ , so (∂ ± )2 = 0.

Definition. The index of the Dirac operator is defined as

/ = dim(ker(∂ + )) − dim(ker(∂ i )).

index(i∂)

Let us now return to the N = 12 theory. The bosonic part of the wavefunction may again be
√
taken to be in L2 (N , g dn x). For the fermionic part, let eia , i = 1, . . . , n be a basis of c-number
1-forms on N obeying eia ejb g ab = δ ij (such a basis is called an orthonormal frame). Then we have
n o
eia ψ̂ a , ejb ψ̂ b = 2δ ij ,

so the operators ψ̂ i = eia ψ̂ a obey the same anticommutation relations as the Dirac γ matrices, and
all of the above constructions with the γ matrices also apply to these operators. In particular, we
may construct raising and lowering operators ψ̂± I = ψ̂ 2I ± iψ̂ 2I−1 , and choose a fermionic vacuum

state |0i which is annihilated by all lowering operators ψ̂− I . The rest of the fermionic Hilbert space

is generated by application of the raising operators ψ̂+ I to the vacuum state.

By identifying the operators ψ̂ i with the γ matrices, and the vacuum state with the spinor χ,
we can give fermionic states a geometric interpretation – they are just spinors on N . Therefore, a

28
 3.6 The Atiyah-Singer index theorem 3 SUPERSYMMETRIC QUANTUM MECHANICS

state in the full Hilbert space, including bosons and fermions, assigns an amplitude to each possible
spinor at each possible x ∈ M. In other words, the Hilbert space is
√
H = L2 (S(N ), g dn x),

the space of square integrable functions on the spin bundle over M, which is just the bundle
whose sections give a choice of a spinor at each point in M. In this geometric interpretation, the
supercharge
Q = ψ a (igab ẋb + ψc Γcab ψ b )
/ = iγ i ∇i , where ∇i is the Levi-Civita
can be shown to correspond to the covariant Dirac operator i∇
i / decomposes into two operators
covariant derivative associated with the frame ea . As above, i∇
± ± ∓ + −
i∇ : S (N ) → S (N ), where S(N ) = S (N ) ⊕ S (N ) is the chiral decomposition of the spin
bundle. The Witten index of the theory is then just the index of the covariant Dirac operator:
 
Tr (−1)F e−βH = dim(ker(∇+ )) − dim(ker(∇− )).

We can obtain another expression for this index by using a path integral approach. The path
integral is again independence of β. As before, let us split separate the zero modes from the rest of
the fields, writing

xa (τ ) = xa0 + δxa (τ ),
ψ a (τ ) = ψ0a + δψ a (τ ).

We will use Riemann normal coordinates near x0 to write the metric as

1  
gab (x) = δab − Racbd (x0 )δxc δxd + O δx3 .
3
In these coordinates, the Levi-Civita symbol may be written
1  
Γabc (x) = ∂d Γabc (x0 )δxd = − (Rabcd (x0 ) + Racbd (x0 ))δxd + O δx2 .
3
Using this, one may write the action to second order in the fluctuations as
I " #
1 d2 i d i
S (2)
[x0 , ψ0 , δx, δψ] = − δxb 2 δxa + δab δψ b δψ a − Rabcd ψ0a ψ0b δxc δ ẋd dτ .
2 dτ 2 dτ 4
Lecture
10
21/02/19 For any fixed values of the zero modes xa0 , ψ0a , this is a free action, so the path integral over
fluctuations gives
q
Z det0 (∂τ δab )
e−S[x0 ,ψ0 ,δx,δψ] Dδx Dδψ = q
det0 (−∂τ2 δba − iRab (x0 , ψ0 )∂τ )
q
det0 (∂τ δab )
=q ,
det0 (∂τ δab ) det0 (−∂τ δab − iRab (x0 , ψ0 ))

a a c d 0
q R b = R bcd (x0 )ψ0 ψ0 , and det denotes a determinant without zero modes. The factors
where
of det0 (∂τ δab ) in the numerator and denominator clearly cancel. To deal with the remaining

29
3.6 The Atiyah-Singer index theorem 3 SUPERSYMMETRIC QUANTUM MECHANICS

expression, note that Rab is a real antisymmetric 2m-dimensional matrix. This means we can
decompose the tangent space Tx0 M into m 2-dimensional subspaces on which Rab , acts as
!
0 ωi
Rab = ,
−ωi 0

where i = 1, . . . , m, and ±iωi are the eigenvalues of Rab with eigenvectors vi± . Let −Di be the
restriction of the operator −δba ∂τ − iRab to the ith subspace. On the basis vi± e2πikτ , k 6= 0, −Di
acts as
−Di (vi± e2πikτ ) = −(2πik ∓ ωi )vi± e2πikτ
Thus we have

det0 (−Di ) =
Y
(2πik + ωi )(2πik − ωi)
k6=0
Y
= (−4π 2 k 2 − ωi2 )
k6=0
∞ ∞
!2
Y
4
Y ωi2
= (2πk) 1+ .
k=1 k=1
(2πk)2

This product is clearly divergent, and so must be regularised in some way. One way to do this is
with zeta-function regularisation, the result being the replacement ∞ 4
k=1 (2πk) → 1. What is left is
Q

the second factor in the last line, which, unlike the first factor, is dependent on ωi . Recalling that
∞
!
Y z2
sinh(z) = z 1+ 2 2 ,
k=1
π k

we may therefore write

q
sinh(ωi /2)
det0 (−Di ) = .
ωi /2
Therefore,
m
1 Y ωi /2
q = .
det 0
(−∂τ δab − iRab (x0 , ψ0 )) i=1
sinh(ω i /2)

To compute the Witten index, we still need to integrate over the zero modes. We have
m
ωi /2 d n x0 n
Z Y
/ =
IW = index(i∇) d ψ0
i=1
sinh(ωi /2) (2π)n/2
!
Rab (x0 , ψ0 )/2 d n x0 n
Z
= det d ψ0
sinh Rab (x0 , ψ0 )/2 (2π)n/2

1 R/2
Z  
= det ,
(2π)n/2 N sinh(R/2)

where Rab = Rabcd dxc ∧ dxd is the curvature 2-form on N .

This result, and the Gauss-Bonnet theorem in the previous section, are special cases of the
Atiyah-Singer index theorem. This theorem allows us in a very general way to write the index of
an elliptic operator (the Laplacian on forms, the Dirac operator, etc.) acting on sections of some
bundle over N as the integral of certain quantities on N .

30
 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

4 Supersymmetric Quantum Field Theory

n a d-dimensional
If we have o theory that is Lorentz invariant, we must compute the supersymmetry
algebra Q, Q† = 2H. But the Hamiltonian is only one part of the d-momentum multiplet Pµ ,
which transforms non-trivially under Lorentz symmetry. Therefore it is necessary to introduce
further supercharges which generate the rest of the d-momentum. Since Pµ is a vector, it is spin 1.
The momentum is made from combining two supercharges with equal spin, and so the supercharges
must have spin 21 – in other words, they are spinors.
More specifically, the supersymmetry algebra in d dimensions is
n o
Qα , Q†β = 2γαβ
µ
Pµ ,

where α, β, . . . are spinor indices, and γ µ is a Dirac γ matrix.

We’ll mostly be concerned with d = 2, whose Dirac spinors have 2d/2 = 2 complex components.
ψ−

We will write ψ = ψ+ . Choosing coordinates (t, s) ∈ R2 in which the Minkowski metric takes the
form η = diag(+1, −1), we can represent the Dirac matrices with
! !
t 0 1 s 0 −1
γ = , γ = ,
1 0 1 0

which obey {γ µ , γ ν } = 2η µν .
The action for a free, massless Dirac spinor in d = 2 is
1
Z
S[ψ] = / dt ds ,
iψ̄ ∂ψ
2π R

where ∂/ = γ µ ∂µ , and ψ̄ = ψ † γ t . We may expand this out to

1
Z h i
S[ψ] = iψ̄− (∂t + ∂s )ψ− + iψ̄+ (∂t − ∂s )ψ+ dt ds
2π R

Classically, one finds that the equations of motion (∂t + ∂s )ψ− = (∂t − ∂s )ψ+ = 0 may be solved by
a right-moving ψ− (t, s) = f (t − s), and a left-moving ψ+ (t, s) = g(t + s).
Under a SO(1, 1) transformation
! ! !
t cosh γ sinh γ t
7→ ,
s sinh γ cosh γ s

the spinors transform as

ψ± 7→ e±γ/2 ψ± , ψ̄± 7→ e±γ/2 ψ̄± .

4.1 Superspace in d = 2
Lecture
11
26/02/19 Definition. A superspace is a space with both bosonic and fermionic directions.

Let R2|4 denote the superspace with bosonic coordinates x0 , x1 = t, s and fermionic coordinates
θ+ , θ− , θ̄+ , θ̄− . These coordinates are acted on by an SO(1, 1) transformation in the way just

31
4.1 Superspace in d = 2 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

described, with θ± , θ̄± transforming like spinors of chirality ±1 respectively. We can introduce the
following fermionic derivatives on R2|4 :
∂ ∂
Q± = + iθ̄± ± ,
∂θ± ∂x
∂ ∂
Q̄± = − ± − iθ± ± ,
∂ θ̄ ∂x
 
∂ 1 ∂ ∂
where ∂± = ∂x± = 2 ∂x0
± ∂x1
. These derivatives obey the anticommutation relations
n o
Q± , Q̄± = −2i∂± ,

with others vanishing. Thus they represent the supersymmetry algebra on R2|4 . SUSY transforma-
tions act geometrically on R2|4 , being generated by

δ = + Q− − − Q+ − ¯+ Q̄− + ¯− Q̄+ ,

where ± , ¯± are some spinor parameters.

Definition. A superfield is a function on a superspace.

Since the θ± , θ̄± coordinates are anticommuting, a generic superfield F on R2|4 permits an ex-
pansion in the fermionic coordinates, with coefficients given by functions of the bosonic coordinates:

F(x± , θ± , θ̄± ) = f0 (x± )+θ+ f+ (x± )a+θ− f− (x± )+θ̄+ g+ (x± )+θ̄− g− (x± )+· · ·+θ+ θ̄+ θ− θ̄− D(x± ).

There are 24 = 16 such coefficient functions, sometimes collectively known as a supermultiplet.

Note that under a supersymmetry transformation F → F + δF , the highest component field D(x± )
can only change at most by total bosonic derivatives.
It’s often useful to have ‘smaller’ superfields that are constrained in some way. For this purpose,
we introduce
∂ ∂
D± = ± − iθ̄± ∂± , D¯± = − ± + iθ± ∂± ,
∂θ ∂ θ̄
which differ from Q± , Q̄± by the signs of the second terms. One can think of D± and D̄± as
holomorphic antiholomorphic derivatives respectively. The anticommutators of these derivatives
obey n o
D± , D̄± = +2i∂± ,
with others vanishing. Crucially, we also have
n o
{D, Q} = D̄, Q = · · · = 0,

i.e. all anticommutators between these derivatives and the supercharges vanish. This makes the
following definition a useful one.

Definition. A chiral superfield is a superfield Φ which obeys D̄± Φ = 0.

32
4.2 Supersymmetry invariant actions 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

Chiral superfields can depend on x± , θ± , θ̄± only through y ± , θ± , where y ± = x± − iθ± θ̄± , since
D̄± y ± = D̄∓ y ± = 0. We can therefore expand a chiral superfield as

Φ(y ± , θ± ) = φ(y ± ) + θ+ ψ+ (y ± ) + θ− ψ− (y ± ) + θ+ θ− F (y ± ).

The product of any two chiral superfields is again chiral. A holomorphic function of any chiral
superfield is chiral (holomorphic meaning no dependence on Φ̄). The conjugate of a chiral superfield
obeys
D± Φ̄ = 0,
and is called antichiral. Under a supersymmetry transformation Φ → Φ + δΦ, the chirality of Φ is
preserved (because the supercharges anticommute with the Ds).
To work out how the supersymmetry transformations act on the component fields of a chiral
superfield, it is useful to first write the supercharges in terms of y ± instead of x± . Using the chain
rule, we have
∂ ∂
Q± = + iθ̄±
∂θ± x,θ̄ ∂x± θ,θ̄
∂ ∂y ± ∂ ∂
= + + iθ̄±
∂θ± y,θ̄ ∂θ± x,θ̄
∂y ± θ,θ̄ ∂y ± θ,θ̄
| {z }
−iθ̄±
∂
= .
∂θ± y,θ̄

One may similarly compute that

∂ ∂
Q̄± = − − 2iθ± .
∂ θ̄± y,θ ∂y ± θ,θ̄

Using these, one finds that the components transform in the following way:

δφ = + ψ− − − ψ+ ,
δψ± = ± F ± ¯∓ ∂± φ,
δF = −2i¯
+ ∂− ψ+ − 2i¯
− ∂+ ψ− .

Again, note that the supersymmetry transformation of the highest component field F (y ± ) is at
most a bosonic total derivative.

4.2 Supersymmetry invariant actions

The fact that the D term of a generic superfield and the F term of a chiral superfield vary only by
total derivatives allows us to readily construct supersymmetry invariants.
In particular, let K(Fi , Φa , Φ̄a ) be any real smooth function of real superfields Fi and chiral
superfields Φa and their complex conjugates Φ̄a (K is sometimes called a Kähler potential). Then
Z
K(Fi , Φa , Φ̄a ) d2 x d2 θ d2 θ̄
R2|4

is a supersymmetry invariant, provided the component fields behave appropriately as |x± | → ∞.

This is because after the fermionic integration only the θ2 θ̄2 coefficient in the expansion of K

33
 4.3 Wess-Zumino model in d = 1 + 1 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

remains, and this changes by a total derivative under a supersymmetry transformation, which
vanishes upon doing the bosonic integration.
Similarly, suppose W (Φa ) is a holomorphic function of Φa (in this context sometimes called a
superpotential). Then W is chiral, and so
Z
W (Φa ) d2 y d2 θ
R2|2

is again a supersymmetry invariant, provided the coefficient fields are sufficiently well behaved as
|y ± | → ∞.

4.3 Wess-Zumino model in d = 1 + 1

Let’s consider the simplest case of a single chiral superfield Φ and its conjugate Φ̄. We will take
K(Φ, Φ̄) = |Φ|2 + Φ̄Φ, but leave the superpotential generic. Then we can use the real functional
Z Z Z
2 4 2 2
S[Φ, Φ̄] = Φ̄Φ d x d θ + W (Φ) d y d θ + W̄ (Φ̄) d2 y d2 θ
R2|4 R2|2 R2|2
| {z } | {z }
kinetic terms interactions

as the action for a field theory. For the reasons given in the previous subsection, this action is
guaranteed to be supersymmetric, assuming the fields have the appropriate asymptotic behaviour.
Lecture This theory is known as the Wess-Zumino model.
12
28/02/19 Let’s do the integration over the fermionic variables in the Wess-Zumino action. To do so, we
need to find the θ2 term in the superpotential, and the θ4 term in the kinetic terms.
A Taylor series expansion gives that the coefficient of θ+ θ− in the superpotential is given by

∂2W
W (Φ)|θ+θ− = F ∂W (φ) − ψ+ ψ− .
∂φ2

For the kinetic terms, we need to write the chiral superfield Φ(y ± , θ± ) as a function of x± , θ± , θ̄± .
Using y ± = x± − iθ± θ̄± , we can expand the coefficients of the chiral superfield in a series in powers
of θ± θ̄± . Doing so, one finds

Φ(x± , θ± , θ̄± ) = φ(x± ) − iθ+ θ̄+ ∂+ φ(x± ) − iθ− θ̄− ∂− φ(x± ) − θ+ θ̄+ θ− θ̄− ∂+ ∂− φ(x± )
+ θ+ ψ+ (x± ) + θ− ψ− (x± ) − iθ+ θ− θ̄− ∂− ψ+ (x± ) − iθ− θ+ θ̄+ ∂+ ψ− (x± ) + θ+ θ− F (x± ).

Similarly, for the conjugate Φ̄ we have

Φ̄(x± , θ± , θ̄± ) = φ̄(x± ) + iθ+ θ̄+ ∂+ φ̄(x± ) + iθ− θ̄− ∂− φ̄(x± ) − θ+ θ̄+ θ− θ̄− ∂+ ∂− φ̄(x± )
− θ̄+ ψ̄+ (x± ) − θ̄− ψ̄− (x± ) − iθ̄+ θ− θ̄− ∂− ψ̄+ (x± ) − iθ̄− θ+ θ̄+ ∂+ ψ̄− (x± ) + θ̄+ θ̄− F̄ (x± ).

Using these, one can extract the coefficient of θ2 θ̄2 in Φ̄Φ. One finds

Φ̄Φ|θ4 = −φ̄∂+ ∂− φ + ∂+ φ̄∂− φ + ∂− φ̄∂+ φ − ∂+ ∂− φ̄φ

+ iψ̄+ ∂− ψ+ − i∂− ψ̄+ ψ+ + iψ̄− ∂+ ψ− − i∂+ ψ̄− ψ− + |F |2 .

34
4.3 Wess-Zumino model in d = 1 + 1 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

Combining all the pieces, we may now evaluate all of the fermionic integrals to get the Wess-
Zumino action in terms of the component fields:
Z h
S[φ, ψ, ψ̄, F, F̄ ] = ∂ µ φ̄∂µ φ + iψ̄− ∂+ ψ− + iψ̄+ ∂− ψ+ + |F |2
R2
i
+ F W 0 (φ) + F̄ W̄ 0 (φ̄) − ψ+ ψ− W 00 (φ) − ψ̄− ψ̄+ W̄ 00 (φ̄) d2 x .

Note that the field F is auxiliary in this action, meaning that no derivatives of F appear, so the
equations of motion for F are purely algebraic. This means that we can eliminate it using its
equations of motion F + W̄ 0 (φ̄) = 0. Doing so yields
Z h i
S[φ, ψ, ψ̄] = ∂ µ φ̄∂µ φ + iψ̄− ∂+ ψ− + iψ̄+ ∂− ψ+ − |W 0 (φ)|2 − ψ+ ψ− W 00 (φ) − ψ̄− ψ̄+ W̄ 00 (φ̄) d2 x .
R2

The first three terms are kinetic terms for the boson φ and fermions ψ± . The other terms are
interactions, the first one being a potential V (φ) = |W 0 (φ)|2 for the scalar field.

Symmetries of the WZ model

By construction, the Wess-Zumino model is invariant under supersymmetry transformations. The

Noether currents for these transformations may be found to be

G0± = 2∂± φ̄ψ± ± iψ̄± F,

G1± = ∓2∂± φ̄ψ± + iψ̄∓ F.

Ḡµ± are just the conjugates of Ḡµ± . As usual, we may obtain the Noether charges by integrating the
Noether currents over a spatial surface, so
Z
Q± = G0± dx1 .
R

Notice that these currents Gµ±

are spin 32 . Integration reduces the spin by one, so the Noether
1
charges are each spin 2 , as expected for supercharges.
The Wess-Zumino model has some more important symmetries. For example ‘axial’ transforma-
tions act on the chiral superfield as

Φ(x± , θ± , θ̄±) → Φ(x± , e∓iα θ± , e±iα θ̄± ).

Under this transformation, θ+ θ− is invariant, and so W (Φ)|θ2 , Φ̄Φ|θ4 are also invariant. This
means that the action does not change, and so this is a symmetry. The group generated by axial
symmetries is often denoted U (1)A . In terms of the component fields, we can equivalently think of
an axial transformation as
φ → φ, ψ± → e∓iα ψ± , F → F.
The corresponding Noether charge is
Z
FA = (ψ̄+ ψ+ − ψ̄− ψ− ) dx1 .
R

Now consider the ‘vector’ transformation

β
Φ(x± , θ± , θ̄± ) → eiq Φ(x± , e−iβ θ± , eiβ θ̄± ).

35
 4.3 Wess-Zumino model in d = 1 + 1 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

The group of these transformations is often denoted U (1)V . Under this transformation, we have
θ+ θ− → e−2iβ θ+ θ− , while θ2 θ̄2 is invariant. This means that the Kähler contribution Φ̄Φ|θ4 will be
invariant in the action. However, the superpotential term will only be invariant if W has charge 2,
meaning it transforms like W → e2iβ W . In particular, for a monomial superpotential

W (Φ) = cΦk ,

vector transformations are only symmetries if kq = 2. If this is the case, then vector transformations
can be viewed as acting on the component fields in the following way:
2i 2 2
φ → e k β φ, ψ± → e( k −1)iβ ψ± , F → e( k −2)iβ F.

Vacuum moduli space

Lecture
13
05/03/19 So far we have only been considering the Wess-Zumino model classically. Similar classical field
theories exist in higher dimensions. But we want to do quantum field theory. Generically, when
quantising a classical field theory, quantum corrections from the microscopic theory become
important. But a theory with supersymmetry is to some degree protected from these corrections.
In particular, we will see that in a 2-dimensional Wess-Zumino model, the superpotential is not
quantum corrected.
As noted above, the scalar potential of the WZ model is V (φ, φ̄) = |W 0 (φ)|2 . We only had one
chiral superfield, but more generally we could have some number of them Φa , indexed by a and
set the Kähler potential to K(Φa , Φ̄a ) = Φ̄a Φa (summing over a). The superpotential is allowed
to depend on all of these fields. Carrying out the above analysis, we would find that the scalar
potential is given by
∂W 2
V (φa , φ̄a ) = .
∂φa

Recall that ground states |Ωi are supersymmetric if H |Ωi = 0. In particular, since V ≥ 0, a
field configuration can only be a ground state if it sits in a minimum of V (φ) over all space. This
ground state is supersymmetric if and only if V = 0, which is equivalent to requiring
∂W
= 0 for all φa
∂φa
when evaluated on the ground state. Let φa0 be the configuration of the scalar field on the ground
state. In the quantum theory, this configuration is determined by an expectation value

φa0 = hΩ|φa |Ωi .

Typically, the holomorphic function W (φa ) is a polynomial over C, so the vacuum conditions
∂φa = 0 are a system of polynomials. The space of vacua, denoted M , is then the zero set of these
∂W

polynomials, i.e. the set satisfying

∂W ∂W
1
= = · · · = 0.
∂φ ∂φ2
This is an affine algebraic variety, which is a central object of study in algebraic geometry. M is
sometimes called the vacuum moduli space.

36
4.3 Wess-Zumino model in d = 1 + 1 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

mφ2 λφ3 ∂W
Example. Suppose W (φ) = 2 + 3 . Then ∂φ = mφ + λφ2 = 0 at φ = 0, − m
λ.

V (φ) = |W 0 |2

Thus there are two isolated supersymmetric minima, and these make up the vacuum space
M = {− m
λ , 0}.

Example. Suppose W (ρ, h) = λ2 ρh2 . Then we have

∂W λ ∂W
= h2 , = λρh.
∂ρ 2 ∂h
The first of these implies that a supersymmetric vacuum must have h = 0. If this is true, then
the second is automatically satified. In particular ρ can take any value in the complex plane.
Thus the vacuum moduli space M is the complex ρ-plane.
For hρi =
6 0, the field h is massive, with mass |λ hρi|. The field ρ is always massless in the
vacuum, because hhi = 0, and so there is no term proportional to ρ2 in the action.

Example. Suppose we have three fields X, Y, Z, and the superpotential takes the form
W (X, Y, Z) = XY Z. Then

∂X W = Y Z, ∂Y W = ZX, ∂Z W = XY,

so
M = {X = Y = 0} ∪ {Y = Z = 0} ∪ {Z = X = 0}.
This has the structure of three complex planes meeting at the origin:

hXi

hY i hZi

37
4.4 Seiberg non-renormalisation 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

In general the vacuum moduli space is the affine variety

C[φ1 , . . . ]/∂a W.

If M is not just a set of isolated points, one says that the potential has flat directions. In this case,
we can change some combination of the hφa is continuously without leaving V (hφa i) = 0.
In a generic QFT, the structure of the classical potential is changed by quantum corrections.
Couplings run with scale, and new couplings are generated (at least in an effective theory). In
general, these corrections tend to lift flat directions, but supersymmetric theories are special.

4.4 Seiberg non-renormalisation

The claim: in a supersymmetric theory, the effective superpotential Weff (Φ) in the Wilsonian action
is identical to the classical superpotential W (Φ).
We won’t prove this statement, but will instead try to understand the ideas behind it in a
particular example. We set the superpotential to
m 2 λ 3
W (Φ) = Φ + Φ .
2 3
This theory has interactions, and in particular has the following Feynman vertices:

φ̄ φ φ̄ φ

φ φ̄

φ̄ φ φ̄ φ

ψ+ ψ̄+

φ φ̄

ψ− ψ̄−

As a consequence there are many non-trivial Feynman diagrams. For example, the 1-loop
corrections to the mass m recieve contributions from:

φ φ + φ φ

+ φ φ

38
 4.4 Seiberg non-renormalisation 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

However, it’s possible to show that these diagrams cancel each other exactly. Moreover, the
same cancellation holds to all orders in λ! Why has this happened?
Usually when a theory is protected from quantum corrections, it is due to a symmetry of that
theory. Here, W (Φ) breaks the vector symmetry U (1)V , so that can’t possible help. Also, the axial
symmetry U (1)A acts trivially on the scalars φ, so it can’t constrain the form of the effective mass.
The right idea, due to Seiberg, is to promote the couplings m, λ themselves to chiral superfields
M, Λ, such that m, λ are the vacuum expectation values of M, Λ. Note that M, Λ must be chiral,
because they appear in the superpotential W (Φ, M, Λ). In promoting these couplings to fields, we
give them kinetic terms, modifying the Kähler potential by
1
K(Φ, Φ̄) → K(Φ, Φ̄) + (M̄ M + Λ̄Λ).

1
The factor of  means that fluctuations in M, Λ will be highly suppressed as  → 0.
The point is that if we give M, Λ the right charges, the superpotential W (Φ, M, Λ) now does
preserve the U (1)V vector symmetry. There is also an additional symmetry U (1)W Z which acts
trivially on θ± , θ̄± . The correct charges are given by

Φ M Λ
U (1)V 1 0 -1
U (1)W Z 1 -2 -3

Provided we choose a regularisation which is supersymmetric, and preserves these U (1)s, then
the effective superpotential Weff (Φ, M, Λ) in the Wilsonian action is constrained to:
• Be holomorphic in (Φ, M, Λ) (for supersymmetry),
• Have U (1)V charge +2 and U (1)W Z charge 0,

Lecture • Reduce to the classical W (Φ) in the limit M, Λ (i.e. weak coupling).
14
07/03/19 The first two of these conditions fix the effective superpotnetial to be of the form
ΦΛ
 
Weff (Φ, M, Λ) = M Φ2 f ,
M

where f (t) must be holomorphic in t. In particular, f (t) must be regular as t → 0, and f (t)
t must
be regular as t → ∞, which means it must be of the form f (t) = a + bt. The final condition fixes
a = 12 and b = 31 . Hence
M 2 Λ 3
Weff (Φ, M, Λ) = Φ + Φ = W (Φ, M, Λ).
2 3

Finally, we freeze the superfields M, Λ to their vacuum expectation values m, λ by sending  → 0

in the kinetic term 1 (M̄ M + Λ̄Λ). The value of  also can’t effect Weff , because we can always
promote 1 to a real superfield, and the superpotential can’t depend on real superfields. (More
generally, the quantum superpotential is always independent of couplings that appear only in the
Kähler potential.)
Therefore:
m 2 λ 3
Weff (Φ) = Φ + Φ = W (Φ).
2 3

39
4.5 Kähler geometry 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

In other words, there are no quantum corrections to W , and we have verified the claim at the
beginning of this subsection (for this example).
However, the Kähler potential does generically get quantum corrections, and we will look at
some examples of this.

4.5 Kähler geometry

Definition. A Kähler manifold M is a manifold with three compatible (in the sense described
below) structures: a Riemannian metric g, a positive symplectic form ω, and a complex
structure J.

Definition. A 2-form ω is symplectic if it is closed (dω = 0) and non-degenerate (for any

vector field X, ω(X, Y ) = 0 for all Y if and only if X = 0).

Definition. A rank 11 tensor J : TM → TM is an almost complex structure if J 2 = −id,

where id is the identity.

n o
∂ ∂
Example. On M = R2 , we have the tangent space basis ∂x , ∂y , and we could choose J
such that
∂ ∂ ∂ ∂
   
J = , J =− .
∂x ∂y ∂y ∂x
In particular, such a J rotates the plane by π2 in the anticlockwise direction. This should be
compared with multiplication by i in the complex plane.

Note that since J is real, its eigenvalues must be ±i. Complex vector fields are sections of TM ⊗ C.
Thus we can define:

Definition. The holomorphic and antiholomorphic tangent bundles over M are

1
T(1,0) M = {X ∈ TM ⊗ C s. t. (1 − iJ)X = X},
2
1
T(0,1) M = {X ∈ TM ⊗ C s. t. (1 + iJ)X = X},
2
i.e. the ±i eigenspaces of J respectively.

An almost complex structure J is said to be integrable if for all X, Y ∈ TM ⊗ C, we have

1 + iJ 1 − iJ 1 − iJ
     
X, Y = 0,
2 2 2
which is equivalent to the condition that the Lie bracket preserves (anti-)holomorphicity. Taking
the real and imaginary parts of the above equation, it can be shown that it is equivalent to

N (X, Y ) = −J 2 ([X, Y ]) + J [JX, Y ] + [X, JY ] − [JX, JY ] = 0.

N is known as the Nijenhuis tensor.

40
 4.5 Kähler geometry 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

Definition. A complex structure J is an integrable almost complex structure.

Theorem 4 (Newlander-Nirenberg). If J is an almost complex structure on a real manifold M

whose Nijenhuis tensor vanishes, then there exist complex coordinates xi → (z a , z̄ a ) on any patch
U ⊂ M, and the transition functions on overlaps of patches are purely holomorphic.
If J is a complex structure, we can split

TM ⊗ C = T(1,0) M ⊕ T(0,1) M

globally. Similarly we may split the complexified cotangent bundle

T∗ M ⊗ C = T∗(1,0) M ⊕ T∗(0,1) M,
∗(1,0) (1,0)
where Tp M is the dual vector space to Tp M for all p ∈ M. Likewise, we can split
M
Ωk (M, C) = Ω(p,q) (M),
k=p+q

where Ωk (M, C) is the space of complex valued k-forms, and if η ∈ Ω(p,q) (M) then we may write
a1
η(z, z̄) = ηa1 ...ap b̄1 ...b̄q dz
| · · ∧ dz ap} ∧ |dz̄ b̄1 ∧ ·{z
∧ ·{z · · ∧ dz̄ b̄q} .
p q

The exterior derivative d maps k-forms to (k + 1)-forms. On a complex manifold (i.e. a manifold
¯ where
with a complex structure), the exterior derivative itself splits as d = ∂ + ∂,

∂ : Ω(p,q) → Ω(p+1,q) ,
∂¯ : Ω(p,q) → Ω(p,q+1) .

For example, on R2 we may write

∂ ∂ ∂ ∂
d = dx + dy = dz + dz̄ ,
∂x ∂y | {z∂z} | {z∂ z̄}
∂ ∂¯

where z = x + iy. These derivatives ∂, ∂¯ are known as the Dolbeault operators. Since

0 = d2 = ∂ 2 + (∂ ∂¯ + ∂∂)
¯ + ∂¯2 ,

we may write
∂ 2 = 0, ∂¯2 = 0, ∂ ∂¯ + ∂∂
¯ = 0,

Lecture because each of these operators take values in different vector spaces.
15
12/03/19 A Kähler manifold has a symplectic form that is compatible with the complex structure in the
sense that
ω(JX, JY ) = ω(X, Y ) for all vectors X, Y .
This implies that ω ∈ Ω(1,1) (M), i.e. it has one holomorphic and one antiholomorphic component,
so we may write
ω = ωab̄ (z, z̄) dz a ∧ dz̄ b̄ .

41
4.5 Kähler geometry 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

Given any such ω, we get a Hermitian metric for free, by writing g(X, Y ) = ω(X, JY ). We can
check that this obeys all the properties we expect of a Hermitian metric. We have

g(Y, X) = ω(Y, JX) = −ω(J 2 Y, JX) = ω(JX, J 2 Y ) = ω(X, JY ) = g(X, Y ),

so it is symmetric. Also,

g(JX, JY ) = ω(JX, J 2 Y ) = −ω(JX, Y ) = ω(Y, JX) = g(Y, X) = g(X, Y ),

so it is Hermitian. Additionally g is positive if and only if ω is positive (i.e. ω(x, JX) > 0 for
non-zero X). g is non-degenerate because ω, J are non-degenerate.
Since dω = 0, on a complex manifold we have ∂ω + ∂ω ¯ = 0, and since these two forms live in
¯
different vector spaces, we must have ∂ω = 0, ∂ω = 0. The complex form of the Poincaré lemma
¯ In this case we can apply
works just like the real form, but there is one result for each of ∂ and ∂.
it to locally write
¯
ω = i∂ ∂K,
where K is a real function on M. Thus the metric may be written

gab̄ = ∂a ∂b̄ K.
∂
(In this notation, ∂b̄ = ∂ z̄ b̄
is an antiholomorphic derivative.) K is called the Kähler potential, and
is defined up to
K(z, z̄) → K(z, z̄) + f (z) + f¯(z̄),
which leaves ω, g unchanged. This is called a Kähler transformation.

Example. Cn is Kähler, with

n
|z a |2 ,
X
K=
a=1
Xn
g= dz a dz̄ ā ,
a=1
Xn
ω=i dz a ∧ dz̄ ā .
a=1

Example. CPn is Kähler, with

n
!
a 2
X
K = ln 1 + |z | ,
a=1

where z a , z̄ ā are defined on a Cn coordinate patch. The associated metric is called the Fubini-
Study metric.

42
4.6 Kähler manifolds and SUSY 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

On a Kähler manifold, the only non-vanishing Christoffel symbols are those for which the indices
are either all holomorphic or all antiholomorphic. In particular we have
1
Γabc = g ai (∂b gic + ∂c gib − ∂i gbc )
2
1 ¯
= g ad (∂b gdc ¯ + ∂c gdb
¯)
2
1 ¯
= g ad (∂b ∂d¯∂c K + ∂c ∂d¯∂b K)
2
¯
= g ad ∂b gcd¯,

where i is a real index. Similarly

Γāb̄c̄ = g ād ∂b̄ gc̄d .
All others vanish.

4.6 Kähler manifolds and SUSY

The general kinetic term in a supersymmetric nonlinear sigma model on R2|4 is

Z
K(Φa , Φ̄a ) d2 x d4 θ .
R2|4

ThisR is defined up to K → K + F (Φ) + F̄ (Φ̄) as, up to total derivatives, the F terms to not survive
the d4 θ integration.
Performing the fermionic integrals gives
Z h
Skin [Φa , Φ̄b ] = − gab̄ ∂ µ φa ∂µ φ̄b̄ + igab̄ ψ̄+
b̄ a
∇− ψ+ b̄
+ igab̄ ψ̄− a
∇+ ψ−
R2
a b̄ c d ¯ ¯
i 2
ē f
ψ̄+ ψ− ψ̄− + gab̄ F a − Γacd ψ+
c d
ψ− F̄ b̄ − Γb̄ēf¯ψ̄−


+ Rab̄cd¯ψ+ ψ̄+ d x,

where gab̄ (φ, φ̄) = ∂a ∂b̄ K(φ, φ̄) is the Kähler metric, and
a a
∇µ ψ+ = ∂µ ψ+ + Γabc ∂µ φb ψ+
c
, etc.

It is not trivial that this is true, but it is clearly plausible.

Eliminating the auxiliary fields by their equations of motion F a = Γabc ψ+
b ψ c , the remaining
−
action is invariant under:
• Complex coordinate transformations of the target space.
• Kähler transformations.
• Supersymmetry transformations on the worldsheet R2 .
• U (1)V transformations ψ± → eiβ ψ± , ψ̄± → e−iβ ψ̄± .
• U (1)A transformations ψ± → e±iα ψ± , ψ̄± → e∓iα ψ̄± .
• Dilatations (scale transformations) on the worldsheet.

43
 4.6 Kähler manifolds and SUSY 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

In fact it is also invariant under conformal transformations, which are positition dependent
generalisations of dilatations. One can see this by noting that the action has no dimensionful
couplings. The metric g contains a very complicated hidden dependence on φ, which generates
many interactions – but all of the couplings are dimensionless.
Thus at the classical level, this defines a supersymmetric conformal field theory. Turning on a
superpotential
1
Z Z  
2 2 a a b
W (Φ) d θ d x = F ∂a W − ∂a ∂b W ψ+ ψ− d2 x
R2|2 R2 2
typically breaks conformal invariance, because the couplings in W (Φ) can be dimensionful.
Lecture These symmetries may or may not survive at the quantum level.
16
14/02/19 To better understand the dilatation invariance, let us explicitly include the worldsheet metric
hµν and gamma matrices γ µ in the action. We’ll also allow the worldsheet Σ to be more general
than R2 . Then, after eliminating F a , F̄ ā , the action may be written
Z h
¯
i√
S[φ, ψ] = gab̄ hµν ∂µ φa ∂ν φ̄b̄ + igab̄ ψ̄ b̄ γ µ ∇µ ψ a + Rab̄cd¯ψ+
a c b̄ d
ψ− ψ̄− ψ̄+ h d2 x .
Σ

Classically, this is invariant under the scale transformation

1
hµν → λ2 hµν , γ µ → λ−1 γ µ , ψ → λ− 2 ψ, λ ∈ R>0 .

The quantum mechanical violation of scale invariance is encoded in the β-function, and in this case
there can be a non-zero β-function for the target space metric gab̄ (φ). To understand this, let us
first consider a purely bosonic NLSM with Riemannian target. The action is
1
Z
S[φ] = gij (φ)∂ µ φi ∂µ φj d2 x .
2 Σ

Let us write φi = φi0 + ξ i , where φi0 is constant, and ξ i is assumed to be some small vector. Then
we have in Riemannian normal coordinates
1  
gij (φ) = δij − Rikjl ξ k ξ l + O ξ 3 ,
3
and the action becomes
1 1
Z    
S[ξ] = µ i j
δij ∂ ξ ∂µ ξ − Rikjl (φ0 )ξ k ξ l ∂µ ξ i ∂ µ ξ j + O ξ 5 d2 x .
2 Σ 3
The first term gives a propagator and the second term gives a four-point vertex for the ξ field. We
may use these to compute a 1-loop correction to the propagator. We have
D E1-loop
ξ i (x)ξ j (y) = ξ i (x) ξ j (y) + ξ i (x) ξ j (y)
k k k
p
" #
d2 k eik·(x−y) ij 1 d2 p 1 ij
Z Z
= δ + R (φ0 ) .
(2π)2 k2 3 (2π)2 p2

To do the p integral, we need to introduce a UV cutoff Λ, and an IR cutoff µ. Then we have

d2 p 1
Z Λ
1 dp 1 Λ
Z  
2 2
= = ln .
µ≤|p|≤Λ (2π) p 2π µ p 2π µ

44
4.6 Kähler manifolds and SUSY 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

By comparing with the classical propagator

d2 k eik·(x−y) ij
D E0-loop Z
i j
ξ (x)ξ (y) = δ ,
(2π)2 k2
we may conclude that the metric at φ0 recieves the following correction:
1 Λ
 
gij (µ) = δij + Rij ln .
6π µ
This is a renormalisation with β-function
1
βij = Rij
6π
which is proportional to the Ricci curvature. There are then three possibilities:
• If Rij > 0, then βij > 0, so the model is asymptotically free. This means that the curvature of
the target becomes less important at short distances on the worldsheet, so the theory makes
sense in the UV.
• If Rij < 0, then the theory only makes sense as an effective theory, but becomes trivial in the
IR.
• If Rij = 0 (so the target is Ricci-flat, i.e. it solves the vacuum Einstein equations), then the
theory is conformally invariant to (at least) 1-loop order accuracy. Higher order corrections
may still change this.
Exactly the same calculations can be done in supersymmetric models. In this case there is a
non-zero running of the Kähler metric, which changes by

gab̄ → gab̄ + cRab̄ ,

where c is some constant. Again, conformal invariance is respected to 1-loop if the target is Ricci-flat.
Ricci-flat Kähler spaces are called Calabi-Yau. It used to be the belief that this continues to hold to
all loop orders in the supersymmetric theory. Indeed, the 2-loop and 3-loop corrections also vanish
on a Calabi-Yau target space. However, there actually are further corrections, which start being
non-trivial at 4 loops. This is apparently a problem for string theory, which we want to remain
conformally invariant.
However, there are ways to compute exact constraints for conformal invariance. In particular,
this may be done without going to higher and higher loops in a Feynman diagram expansion. As
an example, consider the following correlation function:
D E
f (h, g) = (ψ− )k (ψ̄+ )k ,

where k is some positive integer. f = f (h, g) depends on the worldsheet metric hµν and target
metric gab̄ . This correlation function usually vanishes, because there are no Feynman diagrams
that one can draw which will connect ψ− and ψ̄+ . However, the Feynman expansion only works
perturbatively, and there are non-perturbative effects which can contribute to the correlator. In
particular, if there are exactly k zero-modes of ψ− and ψ̄+ , f (h, g) does not have to vanish. A zero
mode of ψ− is one that satisfies ∇+ ψ− = 0. This means it does not contribute to the quadratic
part of the action, and is said to be non-propagating. A zero mode ψ− is a member of

H 0 (Σ, φ∗ T(1,0) N ⊗ S− ).

45
4.6 Kähler manifolds and SUSY 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

In other words, it is holomorphic on the worldsheet Σ, carries a holomorphic target space index,
and is a spinor.
Let us focus on the case where the worldsheet is a 2-torus Σ = T 2 . Then it is possible to apply
the index theorem (and a vanishing theorem) to show that the number of zero modes is equal to

i
Z
Rab̄ dφa ∧ dφ̄b̄ ,
2π T2

which is sometimes called the ‘first Chern class’ of the image of the worldsheet in the target space,
and is a positive integer. Since ψ̄+ is related to ψ− by complex conjugation, it has the same number
of zero modes.
Let’s set k to be equal to the number of zero modes. We have

f (h, g) = f (λ2 h, g)λk .

This is because a rescaling h → λ2 h of the worldsheet metric can be compensated by a rescaling

1
ψ → λ− 2 ψ of the fermions. Also, since the fermions ψ− , ψ̄+ are invariant under Q̄+ , Q− respectively,
we expect some kind of localisation to apply. In fact, one can use this localisation to show that
2 ,φ∗ g)
f (h, g) = nh e− Area(T ,

where nh is some number that depends on h, and

Z √
Area(Σ, φ∗ g) = hµν ∂µ φa ∂ν φ̄b̄ gab̄ h d2 x
Σ

is the area of the worldsheet according to the pullback of the target metric.
Combining the above, we see that

f (h, g) = nλ2 h e−(Area−k ln λ) = g(λ2 h, g 0 ),

where g 0 is a target space metric which satisfies

Area(T 2 , φ∗ g 0 ) = Area(T 2 , φ∗ g) − k ln λ.

However
Z
Area(T 2 , φ∗ g) = i gab̄ (∂z φa ∂z φ̄b̄ + ∂z̄ φa ∂z φ̄b̄ ) d2 z
T2
Z Z
a b̄
= 2i gab̄ ∂z̄ φ ∂z φ̄ d z + i 2
(∂z φa ∂z φ̄b̄ − ∂z̄ φa ∂z φ̄b̄ ) d2 z .
T2 T2

In the course of the localisation procedure referred to above, one finds that the correlator localises
to holomophic φ, so the first term in the last line vanishes. Also, the second term is just equal to
the integral of φ∗ ω (the pullback of the target symplectic form to the worldsheet) over T 2 . Thus
Z
∗
Area(T , φ g) =2
φ∗ ω.
T2

Thus, the effect of rescaling h → λ2 h means that

i ln λ
Z Z Z
∗ ∗
φ ω→ φ ω− Rab̄ dφa ∧ dφ̄b̄ ,
2π

46
4.6 Kähler manifolds and SUSY 4 SUPERSYMMETRIC QUANTUM FIELD THEORY

where we used the index theorem result for k. In other words

ln λ
[ω] → [ω] − [R],
2π
where the notation [α] refers to the equivalence class of the form α up to addition of exact forms.
Thus, the class of the symplectic form changes by the first Chern class. For the theory to be
scale-invariant, we therefore require that the first Chern class of the target space vanishes. This
result is exact in a supersymmetric theory.

Fin

47