arXiv:1209.

2199v3 [hep-th] 25 Dec 2016

Notes On Supermanifolds And Integration

Edward Witten
School of Natural Sciences, Institute for Advanced Study,
1 Einstein Drive, Princeton, NJ 08540 USA

Abstract: These are notes on the theory of supermanifolds and integration on them, aim-
ing to collect results that are useful for a better understanding of superstring perturbation
theory in the RNS formalism.
Contents

1 Introduction 1

2 Supermanifolds 2
2.1 Smooth Supermanifolds 2
2.1.1 Families of Supermanifolds 4
2.1.2 Open Sets And Other Topological Notions 4
2.2 Submanifolds Of A Smooth Supermanifold 5
2.3 Complex Supermanifolds 6
2.3.1 Holomorphic Splittings 7

3 Integration On Supermanifolds 9
3.1 Sections Of The Berezinian 9
3.1.1 Relevance To Superstring Perturbation Theory 12
3.1.2 A Note On Notation 13
3.1.3 Integrating Over The Fibers Of A Fibration 13
3.2 Differential And Integral Forms 13
3.2.1 Clifford Algebras 13
3.2.2 Weyl Algebras 15
3.2.3 Forms On Supermanifolds 17
3.3 Integration Of Integral Forms 18
3.3.1 Integration On ΠT M 18
3.3.2 Some Technical Remarks 19
3.3.3 Equivalence With The Berezin Integral on M 20
3.3.4 Integration Over Submanifolds Of Codimension r|0 21
3.4 The Supermanifold Version Of Stokes’s Theorem 22
3.5 Supermanifold With Boundary 24
3.6 Integration On More General Submanifolds 25
3.6.1 Another Example 28

4 More Operations On Forms 28

4.1 Wedge Products And Contractions 28
4.2 Picture-Changing 31

5 Complex and Smooth Supermanifolds 32

5.1 First Orientation 32
5.2 Complex Supermanifold As A Smooth Supermanifold 33
5.2.1 Critique 34
5.3 Submanifolds Of A Complex Supermanifold 36
5.3.1 Periods On A Complex Supermanifold 36
5.4 Alternative Point Of View 37

–i–
 5.5 Application To String Theory: The Worldsheet 38
5.6 Application To String Theory: Supermoduli Space 39

1 Introduction

Supersymmetric field theories have been studied from many points of view since their
discovery roughly forty years ago. Formulating a supersymmetric field theory in superspace
– that is on a supermanifold – is, when possible, often very helpful. In practice, however,
natural physics questions often require only the most basic facts about supermanifolds.
One topic stands out as a conspicuous exception. This is superstring perturbation
theory in the RNS formalism. This perturbation theory is formulated in terms of integration
on the moduli space of super Riemann surfaces. That moduli space is a rather subtle
supermanifold and simple questions about superstring perturbation theory quickly lead to
relatively subtle issues of supergeometry. Superstring perturbation theory really does seem
like one topic that can be better understood with more input from supergeometry.
The present notes aim to present background material on supermanifolds and integra-
tion. The material is not novel, except possibly for a few details, and the presentation
does not aim for either completeness or full rigor. Rather, the goal has been to collect
in a relatively simple way some background material for a reconsideration of superstring
perturbation theory, which will appear elsewhere [1]. A companion article will contain
background material on super Riemann surfaces [2].
Of course, there is an extensive literature on this topic and it is impossible to give
complete references. Much of the material outlined here can be found in books such as
[3–6] and review articles such as [7–10]. A useful and extremely concise introduction
is [11]. The fundamental structure theorem for smooth supermanifolds was proved in
[12–14] and the theory of integral forms was initiated in [15]. Other useful references
include [16–19]. The superstring literature is likewise too vast to be cited in full. The
classic work [20] introduced some key concepts such as the role of different representations
of the Weyl algebra, the papers [21–24] construct measures on supermoduli space via
superconformal field theory, and the paper [25], which is unfortunately little-known, does
this via algebraic geometry. The papers [26–28], which again are unfortunately little-known,
are valuable both as an exposition of aspects of supergeometry and for insight about its
role in superstring perturbation theory.
In section 2, we describe the basic idea of a supermanifold. In section 3, we sketch
the theory of integration on supermanifolds, and in section 4, we describe some additional
useful facts and constructions. Section 5 is devoted to a close look at some basic ideas
needed in string perturbation theory.

–1–
2 Supermanifolds

2.1 Smooth Supermanifolds

Roughly speaking, a supermanifold M of dimension p|q (that is, bosonic dimension p
and fermionic dimension q) can be described locally by p bosonic coordinates t1 . . . tp and q
fermionic coordinates θ 1 . . . θ q . Sometimes we abbreviate the whole collection of coordinates
as t1 . . . | . . . θ q or simply as x.
We cover M by open sets Uα each of which can be described by coordinates t1α . . . | . . . θαq .
On the intersection Uα ∩ Uβ , the tiα are even functions of t1β . . . | . . . θβq and the θαs are odd
functions of the same variables. We call these functions gluing functions and denote them
as fαβ and ψαβ :

tiα = fαβ
i
(t1β . . . | . . . θβq )
θαs = ψαβ
s
(t1β . . . | . . . θβq ). (2.1)

On the intersection Uα ∩ Uβ , we require that the gluing map defined by fαβ 1 . . . | . . . ψ q is

αβ
inverse to the one defined by fβα 1 . . . | . . . ψ q , and we require an obvious compatibility of
βα
the gluing maps on triple intersections Uα ∩ Uβ ∩ Uγ .
Now we have to be more precise about what sort of supermanifold we want. The most
obvious notion is a real supermanifold of dimension p|q. This would mean that the tiα and
θαs are all real variables, and the gluing functions fαβ i and ψ s are all real. To be more
αβ
pedantic, reality of the gluing functions means that if we expand these functions in powers
of the θ’s, for example
X
i
fαβ i
(t . . . | . . . θ) = gαβ (t1β . . . tpβ ) + θβs gαβs
i
(t1β . . . tpβ ) + . . . , (2.2)
s

then, for real tiβ , the functions arising in this expansion are all real.1 If this condition is
obeyed, we say that M is a real supermanifold and that for each α the local coordinate
system t1α . . . | . . . θαq gives an isomorphism of Uα with an open set in Rp|q .
Real supermanifolds are the right framework for superspace descriptions of super-
symmetric field theories in Lorentz signature and for many other applications in Lorentz
signature. But they are often not convenient for Euclidean signature quantum field the-
ory, largely because spinors in Euclidean signature often do not admit a real structure. A
related fact is that they are not convenient for superstring perturbation theory. The most
important supermanifolds for superstring perturbation theory are super Riemann surfaces
and the moduli spaces thereof; in each case, the fermionic variables have no real structure,
so these are not real supermanifolds.
For superstring perturbation theory and for many other Euclidean signature applica-
tions, one wants a more general notion that is called a cs manifold in [11], p. 94 (where it is
stated that cs stands for complex supersymmetric). Informally, in a cs manifold, although
1
For the moment we consider a single supermanifold M rather than a family of supermanifolds
parametrized by some other space, so we assume that the gluing functions depend only on t1 . . . | . . . θq . See
section 2.1.1.

–2–
the t’s are real at θ = 0, there is no reality condition (for either t’s or θ’s) for θ 6= 0. To be
more precise, in terms of the gluing functions, we require that the bosonic gluing functions
i are real at θ 1 = · · · = θ q = 0, but we impose no reality condition on the θ-dependent
fαβ
terms in fαβi , and no reality condition at all on ψ i . In the case of a cs manifold, we say
αβ
that the coordinate functions t1α . . . | . . . θαq give an isomorphism of the set Uα with an open
set in Rp|∗q , where the asterisk is meant to remind us that there is only a real structure
when the odd variables vanish. In this paper, when not stated otherwise, our “supermani-
folds” are cs manifolds. It is usually clear that the statements can be naturally specialized
to real supermanifolds. On rare occasions, we note differences between the two cases.
On a cs supermanifold, there is no notion of taking the complex conjugate of a function.
This only makes sense once the odd variables are set to zero. In particular, we are never
allowed to talk about θ, a hypothetical complex conjugate of an odd variable θ.
An important point is that to a supermanifold M , one can in a natural way associate a
reduced space Mred that is an ordinary real manifold, naturally embedded in M , and of the
same bosonic dimension. One simply sets the odd variables θα1 . . . θαq to zero in the gluing
law. This is consistent because the odd gluing functions ψαβ s are of odd order2 in θ 1 . . . θ s
β β
and hence vanish when the θ’s do, so the gluing law implies that all θαi vanish if and only if
all θβi do. Moreover, once we set the θ’s to zero, the fαβ i become real, by the definition of a
i 1 p
cs manifold. The functions fαβ (tβ . . . tβ |0 . . . 0) are then the gluing functions of an ordinary
p-dimensional manifold that we call Mred . Moreover, there is a natural embedding

i : Mred → M (2.3)

that takes the point in Mred with coordinates t1α . . . tpα to the point in M with coordinates
t1α . . . |0 . . . 0.
Though we have defined supermanifolds by means of gluing, they can also be defined
by any familiar method for defining ordinary manifolds. For instance, a real supermanifold
M of dimension 2|2 can be defined by a real equation such as

x4 + y 4 + z 4 + θ1 θ2 = 1, (2.4)

with real variables x, y, z, θ1 , θ2 . To present M in the gluing language, one would for
example cover it by open sets Uα in each of which one can solve for one of the bosonic
coordinates x, y, or z in terms of the other variables (for example, in one open set, one might
solve for z by z = (1 − x4 − y 4 − θ1 θ2 )1/4 ). The equation (2.4) defines a real supermanifold
because the parameters in the equation are all real. To get a cs manifold that is not real,
one could add to the equation an additional term that is not real but that vanishes at
θ1 = θ2 = 0. For example, if λ is a complex number that is not real, then a suitable
equation is
x4 + y 4 + z 4 + θ1 θ2 (1 + λx2 ) = 1. (2.5)
2
We still make the assumption of footnote 1, so the only odd variables that can appear in the gluing
functions are θ1 . . . θq .

–3–
2.1.1 Families of Supermanifolds
Often one wishes to consider not a single supermanifold M but a family of supermanifolds
parametrized by some other supermanifold N . For example, M might be a super Riemann
surface, which depends on bosonic and fermionic moduli that parametrize N ; in this ex-
ample, N could be the moduli space of super Riemann surfaces. The best way to think
about this situation is to consider a supermanifold X that is fibered over N with the fibers
being copies of M .
In this situation, X is a supermanifold in the sense that we have already described
and therefore it has a reduced space Xred . In defining Xred , all odd variables are set to
zero, both the odd parameters in N , which we will call η 1 . . . η s , and the odd parameters
θ 1 . . . θ q in M .
Though M depends on η 1 . . . η s , it does not have a reduced space that depends on
those parameters. The reason is that since the gluing functions ψαβ i can depend on the η’s,

we will in general get gluing laws such as θα = θβ + η and we cannot consistently set the
θ’s to zero unless we also set the η’s to zero.
So for example if M is a single super Riemann surface, it has a reduced space Mred
that is an ordinary Riemann surface. But if M depends on some odd parameters η1 , . . . , ηs ,
then we cannot define a reduced space without setting those parameters to zero. That is
why there is no elementary map from the moduli space of super Riemann surfaces to
the moduli space of ordinary Riemann surfaces. This fact led to complications in the
superstring literature of the 1980’s.

2.1.2 Open Sets And Other Topological Notions

The intuition concerning the concept of an “open set” is that if U is an open set in a
topological space Y , and p is a point in U , then any point in Y sufficiently close to p is
also contained in U .
Now suppose that y is an even coordinate on a supermanifold M and that, for example,
ζ1 and ζ2 are odd quantities (either odd functions on M or odd moduli). Then since ζ1 ζ2
is nilpotent, we should think of it as smaller in value than any complex number. So we
should consider, for instance, y + ζ1 ζ2 to be “sufficiently close” to y in any reasonable sense.
The upshot of this is that it is not helpful to introduce a concept of “open set in M ”
that is different from the concept “open set in Mred .” We just consider the two concepts
to be synonyms. So similarly the statement “the Uα form an open cover of M ” means the
same as “the Uα form an open cover of Mred .” For another example, a “neighborhood” in
M of a subset Yred ⊂ Mred is an open set in M whose reduced space is a neighborhood of
Yred in Mred .
A common approach in rigorous mathematical treatments is to define the Uα as open
sets in Mred , but to endow each Uα with a larger ring of functions than the obvious ring of
smooth functions on Uα – namely the functions of the whole set of even and odd coordinates
t1 . . . | . . . θ q , with the t’s restricted to Uα . (In a more fancy language, Mred is endowed
with “a sheaf of Grassmann algebras.”) We will not need this language in these notes.

–4–
However, the reader might find it helpful to develop the intuition that because fermions
are infinitesimal, covering M by open sets is equivalent to covering Mred by open sets.
More generally, for similar reasons, one identifies various topological notions on M
with the same notions for Mred . For example, an orientation or spin structure on M is by
definition an orientation or spin structure on Mred . One says that M is compact if and
only if Mred is compact. The Euler characteristic of M is defined to be that of Mred , and
if Mred is a Riemann surface of genus g, then we also refer to g as the genus of M .

2.2 Submanifolds Of A Smooth Supermanifold

We will now describe another general fact about a smooth supermanifold M . If Nred is any
submanifold of Mred of codimension r, then one can “thicken” it slightly in the fermionic
dimensions to make a submanifold3 N ⊂ M of codimension r|0. N is not unique, but it is
unique up to “homology,” in fact up to an infinitesimal wiggling in the odd directions. The
importance of this is that once we know the appropriate analog of a differential form on
a supermanifold, a closed form on M of the appropriate degree (for example, an integral
form of codimension r in the language of section 3.2) can be integrated over N with a result
that does not depend on the choice of thickening. In addition to being uniquely determined
up to homology, N is also uniquely determined as a supermanifold, up to a (non-unique)
diffeomorphism that acts trivially on Nred .
One should think of existence of N as an intuitively obvious reflection of the fact that
the fermionic directions in a supermanifold are infinitesimal. Roughly, if Nred ⊂ Mred is
defined locally by the vanishing of r real-valued functions h1 , . . . , hr , then upon taking the
h’s to depend upon the odd coordinates θ s in an arbitrary fashion, one gets functions on
M whose vanishing defines the desired submanifold N . A precise proof follows from what
one may call the fundamental structure theorem of smooth supermanifolds. As described
below, this structure theorem lets one construct a (non-unique) projection π : M → Mred ,
and one can define N by N = π −1 (Nred ). For the original proofs of the structure theorem,
see [12–14]; see for example section 4.2 of [4], section 3 of [9], or Theorem 8.2.1 of [6] for
expositions.
To explain the structure theorem, observe first that instead of merely setting the θ’s
to zero in the gluing relations, we could consistently drop all terms of order θ 2 and higher.
This puts the gluing relations in the following form:

tiα = fαβ
i
(t1β . . . tpβ )
X p u
θαs = ψαβs 1
u (tβ . . . tβ )θβ . (2.6)
u

Here we should think of the objects ψαβ s

u as matrix elements of a linear transformation
ψαβ acting on the odd variables. The consistency relations on the gluing data say that
these linear transformations are the transition matrices of a vector bundle V → Mred . This
is a bundle with purely odd fibers, of dimension 0|q.
3
We usually refer to sub-supermanifolds simply as submanifolds, as the term sub-supermanifold is clumsy.

–5–
 Thus, from every supermanifold M , one can extract an ordinary manifold Mred and a
purely fermionic vector bundle4 V → Mred . The total space of this bundle is a superman-
ifold M ′ .
The fundamental structure theorem says that as a smooth supermanifold, M is always
isomorphic to M ′ . The proof is made by expanding the gluing functions in a power series
in the θ’s and showing that, order by order, each term beyond those that we have kept
in (2.6) can be eliminated by a suitable redefinition of the coordinates. (Moreover, the
coordinate change in question shifts the t’s only by terms of order θ 2 or higher, and the θ’s
only by terms of order θ 3 or higher.) Since there are only finitely many θ’s, this process
terminates after finitely many steps.5
A supermanifold that is presented in the form (2.6) is said to be split. The structure
theorem says that every smooth supermanifold can be split, but not in a unique fashion.
Once a splitting is picked, there is a natural projection map π : M → Mred that simply
forgets the θ’s (and thus maps the point in M labeled by t1α . . . | . . . θαq to the point in Mred
labeled by t1α . . . tpα ). The projection π is related to the inclusion i : Mred → M that we
defined earlier by π ◦ i = 1.
The drawback of the structure theorem is that the projection π : M → Mred whose
existence is guaranteed by the theorem is far from unique, and the theorem comes with no
advice about finding a useful or natural choice. In the context of superstring perturbation
theory, for example, the structure theorem says (modulo some issues discussed in section
5) that if we wish we can pick a projection π from the moduli space M of super Riemann
surfaces to the moduli space M of ordinary Riemann surfaces, and reduce a measure on
M to a measure on M by integrating first over the fibers of this projection. However, in
the absence of a natural projection, this procedure may not be illuminating.

2.3 Complex Supermanifolds

A complex supermanifold is defined similarly, except that the gluing functions are holo-
morphic functions. Ca|b is a supermanifold parametrized by complex bosonic coordinates
z 1 , . . . , z a and fermionic coordinates θ 1 . . . θ b . (Recall that in this paper there is never a re-
ality condition on fermionic variables.) A complex supermanifold M of complex dimension
a|b can be covered by open sets Uα each of which is parametrized by even and odd complex
coordinates zα1 . . . zαa and θα1 . . . θαb . These coordinates give a holomorphic identification of
Uα with an open set in Ca|b .
4
The structure group of this bundle is GL(q, C) in general. If it cannot be reduced to GL(q, R) – for
example, the Chern classes of V may present an obstruction – then topologically the θ’s cannot be given a
real structure globally. In this case, M must be viewed as a cs manifold rather than a real supermanifold.
There is never any problem locally in giving the θ’s a real structure, by picking a local basis of θ’s and
declaring them to be real. In applications to string theory, there is typically a global obstruction to giving
the θ’s a real structure and no natural way to do so locally.
5
Concretely, to eliminate the unwanted terms from the gluing functions, one needs to know vanishing of
certain sheaf cohomology classes on Mred that can be extracted from the expansion of the gluing functions
in powers of θ. These classes all vanish because in general sheaf cohomology of any smooth manifold, such
as Mred , with values in the sheaf of sections of any vector bundle always vanishes except in degree zero.
The analogous sheaf cohomology for complex manifolds is in general nonzero, which is why a complex
supermanifold need not be holomorphically split. See section 2.3.1.

–6–
 On intersections Uα ∩ Uβ there are gluing relations analogous to (2.1), with the differ-
ence that the gluing functions are now required to be holomorphic in zβ1 . . . | . . . θβb :

zαi = fαβ
i
(zβ1 . . . | . . . θβb )
θαs = ψαβ
s
(zβ1 . . . | . . . θβb ). (2.7)

To be precise, this holomorphy means that if the functions fαβ i and ψ i are expanded as a
αβ
polynomial in the θ’s, then the coefficient of each term is an ordinary holomorphic function
of z 1 . . . z a .
A complex supermanifold M of dimension a|b has a reduced space Mred obtained by
setting the θ’s to zero in the gluing relation. The gluing relations then reduce to

zαi = fαβ
i
(zβ1 . . . zβa |0 . . . 0). (2.8)

These are gluing functions for an ordinary complex manifold Mred of complex dimension
a. There is an evident holomorphic embedding i : Mred → M , mapping zα1 . . . zαa to
zα1 . . . zαa |0 . . . 0. As in our discussion of (2.6), by keeping in the gluing relations the terms
that are linear in the θ’s, we can define a holomorphic vector bundle V → Mred , with fibers
of dimension 0|q. The total space M ′ of this bundle is a complex supermanifold that is an
approximation to M (but in contrast to the smooth case, M ′ and M are not necessarily
isomorphic as complex supermanifolds; see section 2.3.1).
Examples of complex supermanifolds are easily given. For example, let us define
complex projective superspace CPa|b of dimension a|b. It has homogeneous coordinates
z 1 . . . z a+1 |θ 1 . . . θ b , subject to an overall scaling of all z’s and θ’s by a nonzero complex
parameter λ, and with a requirement that the bosonic coordinates z a are not allowed to all
simultaneously vanish. (In supermanifold theory, to say that a bosonic variable is “non-
zero” means that it is invertible or in other words remains nonzero after setting all odd
variables to zero.) To express CPa|b in the above language, for α = 1, . . . , a + 1, let Uα
be defined by the condition z α 6= 0. The Uα give an open cover of CPa|b . Each Uα can
be parametrized by the ratios z β /z α , β 6= α, as well as θ j /z α , with obvious holomorphic
gluing relations.
We can construct many additional examples by imposing an equation F (z 1 . . . | . . . θ b ) =
0, where F is a homogeneous polynomial in the homogeneous coordinates of CPa|b that is
either even or odd. If F is sufficiently generic, this will give a complex supermanifold of
dimension a − 1|b if F is even, or of dimension a|b − 1 if F is odd.

2.3.1 Holomorphic Splittings

We now consider for a complex supermanifold the questions analogous to those that for a
smooth supermanifold are addressed by the fundamental structure theorem described in
section 2.1.
If M is a complex supermanifold, then the reduced space Mred is an ordinary complex
manifold. We recall that there is always a natural holomorphic embedding i : Mred → M .
M is said to be holomorphically projected if there is a holomorphic map π : M → Mred
with π ◦ i = 1. M is said to be holomorphically split if the structure group of the fibration

–7–
M → Mred given by π reduces to the group GL(q) of linear transformations of the fibers
of π. In other words, M is holomorphically split if it is holomorphically isomorphic to M ′ ,
the total space of a purely fermionic vector bundle V → Mred .
In terms of the gluing data (2.7), M is holomorphically projected if the local coordi-
nates can be chosen so that the bosonic gluing functions fαβ i are functions of the z’s only.

It is holomorphically split if the local coordinates can be chosen so that in addition the
fermionic gluing functions ψαβ s are linear in the θ’s. It is rare to encounter in practice a

complex supermanifold that is holomorphically projected but not holomorphically split,

though it is not difficult to construct an example.
As we will see shortly in an example, a generic complex supermanifold is not holomor-
phically projected. The significance of the question is that when holomorphic projections
exist, they tend to be unique or nearly so and may be natural and useful.
A case in point is the moduli space of super Riemann surfaces, which we will call M.
Explicit computations that have been done to date in superstring theory – including the
very beautiful two-loop computations of [29] – make use of the fact that the moduli space
M admits a holomorphic splitting for low genus. However, M is not holomorphically split
in general [30]. In fact, one goal in [1] will be to make it clear that the existence of a
systematic algorithm for superstring perturbation theory does not depend on the existence
of a holomorphic splitting.
For a simple example of a complex supermanifold that cannot be holomorphically
projected (see section 4.2.10 of [4] for another description of this example), let M be the
hypersurface
z12 + z22 + z32 + θ1 θ2 = 0 (2.9)
in the projective space CP2|2 . The reduced space Mred is a hypersurface in an ordinary
projective space CP2 . We write zb1 . . . zb3 for homogeneous coordinates of this CP2 and
define Mred as the hypersurface zb12 + zb22 + zb32 = 0. A holomorphic projection π : M → Mred
would express the zbi as holomorphic functions of the zi and θj . Since we want π ◦ i = 1
where i : Mred → M is the natural inclusion, zbi must coincide with zi at θ1 = θ2 = 0.
So we must have zbi = zi + θ1 θ2 ui for some ui . The ui cannot exist if they are supposed
to be holomorphic; they would have to be homogeneous of degree −1 under scaling of
z1 . . . | . . . θ2 .
The structure theorem of smooth supermanifolds tells us that if M were a smooth
supermanifold, then the ui would exist. Here we run into a subtlety that will be the subject
of section 5. It is possible to endow M with a smooth structure – or more precisely to define
a smooth supermanifold Mcs that admits a complex structure in which it is isomorphic to
M – but there is no truly canonical way to do this. Postponing this somewhat knotty story
to section 5, here we will just define an appropriate Mcs in our example and show that
with this choice, the ui do exist. We define Mcs as a cs supermanifold by starting with
√ √
coordinates t1 . . . t6 |θ1 θ2 , and defining zk = tk + −1 tk+3 , zek = tk − −1 tk+3 , k = 1, . . . , 3.
Then we impose equation (2.9) along with
3
X
zek zk = 1, (2.10)
k=1

–8–
and we divide by the equivalence

zk → eiα zk , k = 1 . . . 3
θs → eiα θs s = 1, 2
zek → e−iα zek , k = 1 . . . 3, (2.11)

where α is a real parameter (more precisely an even parameter that is real modulo the odd
√
variables) and i = −1. If one prefers, one can express all this in terms of the t’s and θ’s
without introducing the z’s and ze’s. This procedure defines a smooth supermanifold Mcs
of dimension 4|2 that admits a complex structure in which it is isomorphic to M . The
structure theorem of smooth supermanifolds tells us that Mcs must split, and indeed a
splitting is given by
zei
zbi = zi + θ1 θ2 , (2.12)
2
P
since this condition along with (2.9) and (2.10) implies that 3k=1 zbk2 = 0.
If M is a complex supermanifold of odd dimension 1, and with no odd moduli on which
the gluing functions depend, then M is inevitably split. The reason for this is simply that
if a single odd coordinate θ is the only odd variable that appears in the gluing functions
of eqn. (2.7), then inevitably those gluing functions have the split form (f is independent
of θ since there is no way to make a fermion bilinear, and similarly ψ is homogeneous and
linear in θ). An important example of this in superstring theory is the one-loop dilaton
tadpole. This involves a moduli space of odd dimension 1, so it is naturally split and some
of the subtleties of superstring perturbation theory do not arise.

3 Integration On Supermanifolds

We will give two different explanations of what sort of object can be integrated on a
supermanifold. The first explanation is possibly slightly abstract, but is directly related
to the way that a measure on supermoduli space has been extracted in the literature from
superconformal field theory [21–24]. The second explanation is possibly more concrete and
gives a convenient framework for the supermanifold version of Stokes’s theorem [15]. We
also will give several descriptions of how to construct objects than can be integrated over
suitable submanifolds of a supermanifold.

3.1 Sections Of The Berezinian

The basic idea of the Berezin integral is presumably familiar. On Rp|∗q , with bosonic and
fermionic coordinates t1 . . . | . . . θ q , a general function g can be expanded as a polynomial
in the θ’s:
g(t1 . . . | . . . θ q ) = g0 (t1 . . . tp ) + · · · + θ q θ q−1 . . . θ 1 gq (t1 . . . tp ). (3.1)
We have written explicitly the first and last terms in the expansion. If g is compactly
supported, or at least if gq vanishes fast enough at infinity, then the integral of g over Rp|∗q
is defined as
Z Z
 1 q
 1 q
dt . . . | . . . dθ g(t . . . | . . . θ ) = dt1 . . . dtp gq (t1 . . . tp ). (3.2)
Rp|q Rp

–9–
The θ’s are treated in a purely algebraic fashion, so the question of whether they admit a
real structure is immaterial.
To generalize this Berezin integral to a general supermanifold, we want to know what
 
sort of object is the “integration form” dt1 . . . | . . . dθ q . On an ordinary oriented manifold,
this would be a differential form of top degree, but on a supermanifold that is the wrong
interpretation. For example, the formula (3.1) implies that if we rescale one of the θ’s by
 
a constant λ, the symbol dt1 . . . | . . . dθ q is multiplied by λ−1 , rather than by λ as one
would expect for a differential form.
To elucidate the meaning of the integration form, we begin with an approach that is
slightly abstract but actually closely related to formulas in the superstring literature of the
1980’s. First we practice with a vector space.6 Let V be a super vector space of bosonic
and fermionic dimensions p|q. A basis of V therefore consists of p even vectors e1 . . . ep and
q odd vectors ρ1 . . . ρq , with the whole collection being linearly independent. We abbreviate
the basis as (e1 . . . | . . . ρq ).
The Berezinian of V , denoted Ber(V ), is a one-dimensional vector space that one can
think of as the space of densities on V . It is defined as follows. For every basis (e1 . . . | . . . ρq )
 
of V , there is a corresponding vector in Ber(V ) that we denote as e1 . . . | . . . ρq . If
(e′1 , . . . | . . . , ρ′q ) is a second basis, related to the first by a linear transformation
   
e′1 e1
e′  e 
 2  2
. .
 ..   .. 
   
  = W  , (3.3)
− −
   
 ..   .. 
. .
ρ′q ρq
then the corresponding two elements of Ber(V ) are related by
 ′   
e1 . . . | . . . , ρ′q = Ber(W ) e1 , . . . | . . . , ρq . (3.4)

Here Ber(W ), known as the Berezinian of W , is the superanalog of the determinant,

in the sense that it possesses the same multiplicative property: if W = W1 W2 , then
Ber(W ) = Ber(W1 )Ber(W2 ). However, its definition is more subtle than that of the ordi-
nary determinant. Picking a decomposition of V as V = Veven ⊕ Vodd , where the summands
are of dimension p|0 and 0|q, we can write a linear transformation of V in block form as
!
A B
W = , (3.5)
C D
where here A and D are respectively p × p and q × q even blocks, while B and C are odd.
The Berezinian is defined for matrices W such that D is invertible, a condition that is
automatically obeyed for the change of basis matrix in (3.3). An explicit formula is

Ber(W ) = det(A − BD −1 C) det−1 (D). (3.6)

6
The matrix elements of the blocks B and C in eqn. (3.5) below are odd, and to allow them to be
nonzero, we should work not over a field but over a Z2 -graded ring that has odd elements.

– 10 –
If W is upper or lower triangular – so that B or C vanishes – then simply
det A
Ber(W ) = . (3.7)
det D
It is rather tricky to show that the formula (3.6) does not depend on the chosen decomposi-
tion and implies the multiplicative property; for instance, see [8], pp. 15-18; [5], section 1.6;
or [11], pp. 59-60. However, it is straightforward to show that the multiplicative property
together with (3.7) implies (3.6). This simply follows from the factorization
! ! !
A B A − BD −1 C B 1 0
= . (3.8)
C D 0 D D −1 C 1

Similarly, the factorization

! ! !
A B A 0 1 A−1 B
= (3.9)
C D C D − CA−1 B 0 1

implies another formula Ber(W ) = det A · det−1 (D − CA−1 B).

Now let M be a compact supermanifold7 of dimension p|q, as described in section 2.1.
Let T ∗ M be the cotangent bundle of M . We will introduce on M a line bundle known as
its Berezinian line bundle Ber(M ). Ber(M ) is defined by saying that every local coordinate
system T = t1 . . . | . . . θ q on M determines a local trivialization of Ber(M ) that we denote
 1 
dt . . . | . . . dθ q . Moreover, if Te = e t1 . . . | . . . θeq is another coordinate system, then the two
trivializations of Ber(M ) are related by
 
 1  ∂T  e1 
dt . . . | . . . dθ q = Ber dt . . . | . . . dθeq . (3.10)
∂ Te

t1 . . . | . . . θeq ).
Here ∂T /∂ Te is the matrix of derivatives of (t1 . . . | . . . θ q ) with respect to (e
We claim that what can be naturally integrated over M is a section of Ber(M ). To
show this, first let σ be a section of Ber(M ) whose support is contained in a small open
set U on which we are given local coordinates t1 . . . | . . . θ q , establishing an isomorphism of
U with an open set in Rp|∗q . This being so, we can view σ as a section of the Berezinian
of Rp|∗q . This Berezinian is trivialized by the section [dt1 . . . | . . . dθ q ] and σ must be the
product of this times some function g:

σ = [dt1 . . . | . . . dθ q g(t1 . . . | . . . θ q ). (3.11)

So we define the integral of σ to equal the integral of the right hand side of eqn. (3.11):
Z Z
 1 
σ= dt . . . | . . . dθ q g(t1 . . . | . . . θ q ). (3.12)
M Rp|∗q
7
To avoid having to include in the formulas some minus signs which could be confusing on first reading,
we will assume that the reduced space Mred of M is oriented. When we identify Uα ⊂ M with an open
subset of Rp|∗q , the orientation of Mred determines an orientation of the reduced space of Rp|∗q . This lets
us view the quantity dt1 . . . dtp in (3.2) and related formulas as a differential form rather than a density.

– 11 –
The integral on the right is the naive Berezin integral (3.2). For this definition to make
sense, we need to check that the result does not depend on the coordinate system t1 . . . | . . . θ q
on Rp|∗q that was used in the computation. This follows from the rule (3.10) for how the
 
symbol dt1 . . . | . . . dθ q transforms under a change of coordinates. The Berezinian in this
formula is analogous to the usual Jacobian in the transformation law of an ordinary integral
under a change of coordinates. For more detail, see for instance [5], pp. 40-1; [6], Theorem
11.2.3; or [11], p. 80.
So far, we have defined the integral of a section of Ber(M ) whose support is in a
sufficiently small region in M . To reduce the general case to this, we pick a cover of M
by small open sets Uα , each of which is isomorphic to an open set in Rp|∗q , and we use
the existence of a partition of unity. Just as on a bosonic manifold, one can find bosonic
P
functions hα on M such that each hα is supported in the interior of Uα and α hα = 1.
P
Then we write σ = α σα where σα = σhα . Each σα is supported in Uα , so its integral
R P R
can be defined as in (3.12). Then we define M σ = α M σα . That this does not depend
on the choice of the open cover or the partition of unity follows from the same sort of
arguments used in defining the integral of a differential form on an ordinary manifold. For
example, see Theorem 11.3.2 of [6].

3.1.1 Relevance To Superstring Perturbation Theory

All this is perhaps a little abstract and in a sense a tautology: we simply postulated
 
the right transformation law for the symbol dt1 . . . | . . . dθ q so that the integral does not
depend on a choice of coordinates. One reason that the definition is useful is that it matches
the superstring literature, a fact that will be relevant in [1]. In applications to superstring
perturbation theory, it is helpful to know the following characterization of Ber(M ). To
every basis ∂t1 . . . | . . . ∂θq of the tangent space of M , a section σ of Ber(M ) assigns a
number, which we will denote as σ(∂t1 . . . | . . . ∂θq ). We define this mapping by saying that

it is linear in σ and for σ = [dt1 . . . | . . . dθ q , we have σ(∂t1 . . . | . . . ∂θq ) = 1. Under a change
of basis, we require
 
∂T
σ(∂et1 . . . | . . . ∂θeq ) = σ(∂t1 . . . | . . . ∂θ ) Ber
q , (3.13)
∂ Te
which is compatible in a natural way with (3.10). Conversely, a function on bases of the
tangent space that transforms in this way is equivalent to a section of Ber(M ). In pertur-
bative string theory, M will be supermoduli space, the basis ∂t1 . . . | . . . ∂θq will correspond
to a basis of super Beltrami differentials, and a section σ of Ber(M ) will be defined by
saying that σ(∂t1 . . . | . . . ∂θq ) is the value of the worldsheet path integral with superghost
insertions corresponding to those super Beltrami differentials.
The construction in this section shows that a smooth section of Ber(M ) can always be
integrated over a compact supermanifold M . So, in the context of superstring perturbation
theory, given that superconformal field theory can be used to define a section of Ber(M ), the
only possible difficulty comes from the lack of compactness of supermoduli space, arising
from the infrared region.

– 12 –
3.1.2 A Note On Notation
A point to stress is that the section of the Berezinian that we have written as [dt1 . . . | . . . dθ q ]
is an irreducible object; we have not built it by multiplying differential forms dt and dθ.
We have not yet even introduced differential forms. The notation is meant to evoke the
idea of a volume form, but the bracket [ ] surrounding the dt’s and dθ’s is a warning
that the symbols inside the bracket have only an abstract meaning.
In section 3.2, we will introduce differential (and integral) forms on a supermanifold;
when we do so, for an odd variable θ, the one-form dθ will be an even variable. So for
example we will have dθdθ ′ = dθ ′ dθ and (dθ)2 6= 0.
By contrast, the symbol [dt1 . . . | . . . dθ q ] is odd under exchange of any two θ’s. This
assertion is a special case of (3.4) or (3.10). The exchange of two θ’s is a coordinate
transformation with Ber(∂ Te/∂T ) = −1, as one can see from (3.7).

3.1.3 Integrating Over The Fibers Of A Fibration

′ ′ ′ ′
Consider a space Rp+p |∗(q+q ) , which we decompose in some fashion as Rp|∗q × Rp |∗q . The
′ ′
Berezin integral of a smooth, compactly supported function on Rp+p |∗(q+q ) can be per-
′ ′
formed by first integrating over Rp|∗q to get a function on Rp |∗q , which one then integrates
′ ′
over Rp |∗q . In fact, if we integrate first over the odd variables, then the statement just
′ ′
made reduces to the fact that a similar ordinary integral on Rp+p = Rp × Rp can be
′
performed by integrating first over Rp and then over Rp .
Now consider a general fibration of supermanifolds π : X → N , with fiber M . We claim
that by integrating over the fibers of π, one can define a natural linear map σ → π∗ (σ)
from a section σ of Ber(X) to a section π∗ (σ) of Ber(N ), obeying the fundamental relation
Z Z
σ= π∗ (σ). (3.14)
X N

Just as in the original definition of integration on a supermanifold, by linearity of the

integral it suffices to consider the case that the support of σ is in a sufficiently small open
set in X. In particular, we can assume that the support of σ projects in N to an open set
that is isomorphic to an open set in some Rp|∗q . We can also assume that the fibration π
is a product when restricted to this open set, and that along the fibers of π, the support
′ ′
of σ is contained in an open set that is isomorphic to an open set in some Rp |∗q . So
the definition of the operation π∗ and the verification of (3.14) reduce to the special case
mentioned in the last paragraph.

3.2 Differential And Integral Forms

Now we are going to describe a phenomenon that in a sense is at the root of the picture-
changing phenomenon in superstring perturbation theory.

3.2.1 Clifford Algebras

Let V ∼
= R0|p be a purely odd vector space of dimension p. Let ζ 1 , . . . , ζ p be a basis of
V and let η1 , . . . , ηp be a basis of the dual space V ∗ . There is a natural nondegenerate

– 13 –
symmetric bilinear form on the direct sum V ⊕ V ∗ which we can write

(ζ i , ζ j ) = (ηi , ηj ) = 0, (ζ i , ηj ) = (ηj , ζ i ) = δji . (3.15)

This quadratic form does not depend on the chosen basis.

Given this nondegenerate quadratic form on V ⊕ V ∗ , we can “quantize” by introducing
the corresponding Clifford algebra. This simply means that we promote the ζ i and ηj to
operators that are supposed to obey the anticommutation relations

{ζ i , ζ j } = 0 = {ηi , ηj }, {ζ i , ηj } = δji . (3.16)

Anticommutation (rather than commutation) relations are natural since ζ i and ηj are odd
variables. An irreducible module8 S for this Clifford algebra (or any Clifford algebra of
even rank) is unique up to isomorphism. We can construct it by starting with a vector
|↓i annihilated by the ηi ; then the states ζ i1 . . . ζ ik |↓i, i1 < · · · < ik , k = 0, . . . , p furnish
a basis of S. Alternatively, we can start with a state |↑i that is annihilated by the ζ j ,
and build a basis by acting on |↑i with the ηi . The two constructions are equivalent, since
|↑i = ζ 1 ζ 2 . . . ζ p |↓i can be reached from |↓i after finitely many steps, and vice-versa.
We would like to interpret this construction more geometrically, but in doing so we
may as well consider a more general situation involving a purely bosonic manifold M of
dimension p. Roughly speaking, we want to consider functions on the tangent bundle T M
of M . But there is a very important twist: we want to consider the fiber directions of the
tangent bundle to be fermionic rather than bosonic. The tangent bundle with this twist is
frequently denoted as ΠT M , where the symbol Π stands for reversal of statistics in the
fiber directions; in the literature, this is often called reversal of parity. If t1 . . . tp are local
coordinates on M , then to give a local coordinate system on ΠT M , we need to double the
coordinates, adding a second set dt1 . . . dtp . These now are fermionic variables since we
have taken the fiber coordinates of ΠT M to be odd. A general function on ΠT M has an
expansion in powers of the dti , and of course this is a finite expansion since these variables
anticommute. A term of order k
X
ai1 ...ik (t1 . . . tp )dti1 . . . dtik (3.17)
i1 ...ik

corresponds to what usually is called a k-form on M . So taking all values of k from 0 to

p, the functions on ΠT M correspond to the whole space of differential forms on M , often
denoted Ω∗ (M ).
At any point m ∈ M , we can easily find a Clifford algebra acting on the functions on
the fiber of ΠT M at m. We take ζ i to be multiplication by dti ,

ψ → dti ∧ ψ (3.18)

(we usually omit the wedge product symbol) and we take ηj to be the corresponding
derivative operator
∂
ηj = . (3.19)
∂dtj
8
A module for the Clifford algebra is simply a vector space on which the algebra acts, analogous to a
representation of a group.

– 14 –
In differential geometry, this operator is usually called the operator of contraction with ∂tj ,
ω → i∂tj ω, (3.20)
but it is simpler to think of it as the derivative with respect to the bosonic variable dtj .
The ζ i and ηj as just defined obey the fiberwise Clifford algebra {ζ i , ηj } = δji , with
other anticommutators vanishing. Differential forms on M or equivalently functions on
ΠT M give a natural module for this family of Clifford algebras; any other irreducible
module would be equivalent (except that globally one could consider differential forms on
M with values in a line bundle).
The exterior derivative operator corresponds to a simple odd vector field on ΠT M :
X ∂ X ∂
d= ζi i = dti i . (3.21)
∂t ∂t
i i

It obeys d2 = 0. Since it has degree 1 under scaling of the fiber coordinates of ΠT M ,

it maps forms of degree k – that is functions homogeneous of degree k in the odd vari-
ables dti – to forms of degree k + 1. When a differential form is interpreted as a function
ω(t1 . . . tp |dt1 . . . dtp ), the wedge product of forms becomes obvious – it is simply the multi-
plication of functions. That is why we usually omit the wedge product symbol. The wedge
product and exterior derivative are related by
d(ω · ν) = dω · ν + (−1)|ω| ω · dν, (3.22)
where |ω| = 0 or 1 for ω even or odd.

3.2.2 Weyl Algebras

Now let us repeat this exercise for a purely even vector space W ∼ = Rq|0 . Again we pick a
basis α , . . . , α of W and a dual basis β1 , . . . , βq of the dual space W ∗ . Because the α’s and
1 q

β’s are now even, it is more useful to introduce a skew-symmetric rather than symmetric
bilinear form on W ⊕ W ∗ :
hαi , αj i = hβi , βj i = 0, hβj , αi i = −hαi , βj i = δji . (3.23)
Upon quantization, the even variables αi and βj obey a Weyl algebra rather than a Clifford
algebra:
[αi , αj ] = [βi , βj ] = 0, [βj , αi ] = δji . (3.24)
In contrast to the finite-dimensional Clifford algebra, the finite-dimensional Weyl algebra
has many irreducible modules. We can postulate a state |↓i that is annihilated by the β’s.
Then a module V for the Weyl algebra can be constructed by acting repeatedly with α’s.
A basis of this module consists of states of the form
αi1 αi2 . . . αik |↓i, k ≥ 0. (3.25)
Because the α’s are commuting variables, there is no upper bound on k. Alternatively, we
can postulate the existence of a state |↑i annihilated by the α’s, in which case we form a
module V ′ with a basis of states
βj1 βj2 . . . βjk |↑i, k ≥ 0. (3.26)

– 15 –
The two modules are inequivalent, as V contains no state annihilated by the α’s and V ′
contains no state annihilated by the β’s. Of course one can form a mixture of the two cases
(and we will discuss such mixtures in section 3.6). But these two cases are of particular
importance.
It is convenient to construct both modules by representing the αi as multiplication
operators and the βj as derivatives:
∂
βj = . (3.27)
∂αj
If we do this, then the two modules differ by the classes of functions allowed. To obtain
V, we consider polynomial functions of the α’s, with the state |↓i corresponding to the
function f (α1 . . . αq ) = 1, which is annihilated by the β’s. To construct V ′ , we need a state
|↑i that is annihilated by the operation of multiplication by αi , i = 1 . . . q. As a function
of the α’s, |↑i corresponds to a delta function supported at α1 = · · · = αq = 0. Certainly
the state δq (α1 . . . αq ) is annihilated by multiplication by any of the α’s. Acting repeatedly
with the β’s, we see that V ′ is spanned by distributions supported at the origin; a basis of
V ′ consists of states of the form
∂ ∂ ∂ q 1
i i
... δ (α . . . αq ), k ≥ 0. (3.28)
∂α ∂α1 2 ∂αik
To interpret this more geometrically, we introduce a purely fermionic supermanifold
M of dimension 0|q. We may as well take M to be R0|q , with coordinates θ 1 . . . θ q . We
introduce an “exterior derivative” d, which we consider to be odd, just as in the bosonic
case, so that it obeys, for example,
d(θ 1 θ 2 ) = dθ 1 · θ 2 − θ 1 · dθ 2 . (3.29)
So the objects αi = dθ i , which we will call one-forms, are even variables. The βj then
become contraction operators
∂
βj = i∂θj = . (3.30)
∂αj
Just as in the bosonic case, a differential form on R0|∗q can be interpreted as a function
ω(dθ 1 . . . dθ q |θ 1 . . . θ q ) (our convention is to list the even variables first), the only subtlety
being that the class of functions considered depends on whether we want the module V or
V ′ . For V, the functions have polynomial dependence on the dθ i , but for V ′ , they are delta
functions supported at dθ i = 0. Also as in the bosonic case, the wedge product of forms
represented by two functions ω and ν is simply represented by the product ων of the two
functions. Here, however, there is a subtlety: one can multiply two polynomials, and one
can multiply a distribution by a polynomial, but one cannot multiply two distributions.
So the wedge product makes sense as a map V × V → V, and also as a map V × V ′ → V ′ ,
but there is no way to multiply two elements of V ′ .
On either module V or V ′ , we can define an exterior derivative operator
X ∂ X ∂
d= αi i = dθ i i . (3.31)
∂θ ∂θ
i i
Wherever the wedge product is defined, the wedge product and the exterior derivative obey
the relation (3.22).

– 16 –
3.2.3 Forms On Supermanifolds
We have treated separately the purely bosonic and purely fermionic cases, but there is no
problem to consider in the same way a general supermanifold M , say of dimension p|q.
Given local coordinates t1 . . . | . . . θ q , we introduce the correponding one-forms dt1 . . . dtp
and dθ 1 . . . dθ q , which are respectively fermionic and bosonic. Forms will correspond to
functions ω(t1 . . . dθ q |θ 1 . . . dtp ) of all the variables, including the dt’s and dθ’s, or in other
words to functions on ΠT M . The only subtlety is what sort of functions are allowed. We
get what one may call differential forms on M if we require ω to have polynomial dependence
on the even one-forms dθ i . And we get integral forms (a concept that originated in [15]) if
we require that in its dependence on dθ 1 . . . dθ q , ω is a distribution supported at the origin.
We write Ω∗ (M ) for the differential forms and Ω∗int (M ) for the integral forms. We take
diffeomorphisms to act on the variables dti and dθ j by the usual chain rule. For example,
t1 . . . | . . . θeq is another coordinate system, then
if e
X ∂ti X ∂ti
dti = tk +
de dθes
∂ e
t k
∂ es
θ
k s
X ∂θ r X ∂θ r
dθ r = tk +
de dθes . (3.32)
∂te k e s
k s ∂θ

Forms with only polynomial dependence on dθ 1 . . . dθ q are mapped to forms of the same
type by a change of coordinates; the same is true for forms with support only at dθ 1 =
· · · = dθ q = 0. Hence the space of differential forms and the space of integral forms are
each invariant under gluing. So these spaces are globally-defined on any supermanifold M ,
even though our initial definition used a local coordinate system.
A wedge product of two differential forms or of a differential form and an integral form
is defined by multiplying the corresponding functions. The exterior derivative is defined in
the obvious way as a vector field on ΠT M
X ∂ X ∂
d= dti + dθ j j (3.33)
∂ti ∂θ
i j

that obeys d2 = 0 and satisifies the usual relation (3.22).

Differential forms are graded by degree in the usual way – a function f (t1 . . . | . . . θ q )
on M is a zero-form, and multiplying k times by one-forms dti or dθ j gives a k-form. In
the space of differential forms, there are forms of lowest degree – namely degree 0 – but
there are no forms of highest degree. If we introduce a scaling symmetry of ΠT M that
acts trivially on t1 . . . | . . . θ p but scales the fiber coordinates by a common factor λ

dt1 . . . | . . . dθ q → λdt1 . . . | . . . λdθ q , (3.34)

then a k-form scales as λk .

In the case of integral forms, we will say that for any function f on M , the object

f (t1 . . . | . . . θ q )dt1 . . . dtp δq (dθ 1 . . . dθ q ) (3.35)

– 17 –
is a top form. Such a form is annihilated by multiplication by dti or dθ j . A form obtained
by acting k times with operators ∂/∂(dti ) or ∂/∂(dθ j ) will be called a form of codimension
k. Again there is no upper bound on k, so in the space of integral forms, there are top
forms but no bottom forms. Under the scaling symmetry of ΠT M , a top form scales as
λp−q , and a form of codimension k scales as λp−q−k .
Differential forms can be multiplied, and one can do many other things with them,
but they cannot be integrated, roughly because there is no top form. To be more exact,
if ω is a p-form on M , then by essentially the ordinary definitions of calculus,9 ω can be
integrated over a purely bosonic oriented submanifold N ⊂ M of the right dimension, in
fact a submanifold of dimension p|0. But ω cannot be integrated over any submanifold
with positive odd dimension.
Integral forms can be integrated over M , as we will discuss shortly (and more generally
over submanifolds of purely bosonic codimension, as we will see in section 3.3.4), but they
cannot be multiplied.
The relation between differential forms and integral forms is a prototype for the notion
of different “pictures” in superstring theory. The concept of different pictures has roots
[31, 32] in the early days of what developed into superstring theory, and was interpreted
in [20] in terms of the existence of inequivalent modules for the Weyl algebra.

3.3 Integration Of Integral Forms

We will give two related explanations of how integral forms of top degree can be integrated.

3.3.1 Integration On ΠT M
One approach starts with a basic difference between M and ΠT M . On M , there is in
general no natural way to pick a section of the Berezinian, but on ΠT M there is always
a natural choice because of the way the variables come in bose-fermi pairs. For every t,
there is a dt, and for every θ, there is a dθ, in each case with opposite statistics. Think-
ing of the whole collection t1 . . . dθ q |θ 1 . . . dtp as a local coordinate system on ΠT M , the
corresponding object
[dt1 . . . d(dθ q )|dθ 1 . . . d(dtp)] (3.36)
is independent of the underlying choice of coordinates t1 . . . | . . . θ q on M and gives a natu-
ral section of Ber(ΠT M ). For example, if we rescale one of the even coordinates of M by
t → λt, then one of the odd coordinates on ΠT M is similarly rescaled by dt → λdt. The
symbol [dt1 . . . d(dθ q )|dθ 1 . . . d(dtp )] changes by the Berezinian of the change of coordinates,
according to (3.10). The relevant Berezinian is 1 (essentially because det D is in the denom-
inator in (3.7) while det A is in the numerator), so the symbol [dt1 . . . d(dθ q )|dθ 1 . . . d(dtp )]
is invariant under this change of coordinates and indeed it is invariant under any change
of coordinates on M .
9
If N is of dimension p|0, we can parametrize it locally by bosonic variables s1 . . . sp . In terms of these
variables, ω becomes an ordinary p-form on an ordinary p-dimensional manifold and we integrate it in the
usual way. It does not matter that s1 . . . sp can only be defined locally; as usual, we write ω as a sum of
p-forms ωα each of which is supported in a small open set Uα in which suitable coordinates exist.

– 18 –
 To streamline our notation, we will abbreviate the whole set of coordinates t1 . . . | . . . θ q
on M as x, and write just D(x, dx) for [dt1 . . . d(dθ q )|dθ 1 . . . d(dtp )]. Similarly, we regard
an integral form ω on M as a function ω(x, dx) on ΠT M . Now we define the integral of ω
over M as a Berezin integral on ΠT M :
Z Z
ω= D(x, dx) ω(x, dx). (3.37)
M ΠT M
It is crucial here that ω is an integral form rather than a differential form. Because
ω(x) has compact support as a function of even variables dθ 1 . . . dθ q (and in fact is a
distribution with support at the origin), the integral over those variables makes sense.
A similar approach to integrating a differential form on M would not make sense, since if
ω(x) is a differential form, it has polynomial dependence on dθ 1 . . . dθ q and the integral over
those variables does not converge. (This is why differential forms can only be integrated on
purely bosonic submanifolds.) The formula (3.37) for integration of ω makes sense for an
integral form ω of any codimension, but if ω has positive codimension, then this formula
vanishes.
Since we have expressed the integral of an integral form in terms of the Berezin integral
in a space with twice as many variables, the reader may wonder if we are making any
progress. Why not stick with the original Berezin integral on M ? One answer is that
in the framework of integral forms, one can formulate a supermanifold analog of Stokes’s
theorem. A related answer is that the formulation with integral forms turns out to be
useful in superstring perturbation theory.

3.3.2 Some Technical Remarks

Two technical remarks are unfortunately difficult to avoid here. First, we need to clarify
what is meant by integration over the bosonic variable dθ. If we are on a real supermanifold,
then dθ is an ordinary real variable, and integration over dθ requires no special explanation.
However, we are mainly interested in cs manifolds, so that a fermionic variable θ has no
real structure, and we cannot claim that dθ is a real variable. This being so, we will have to
interpret integration over dθ as a formal operation, constrained so that one can integrate
by parts, just like the original Berezin integral over θ. The only “functions” of dθ that we
need to integrate are distributions supported at the origin. So we define the integral over
dθ by Z
∂n
[d(dθ)] δ(dθ) = δn,0 . (3.38)
∂(dθ)n
Second, we want to interpret the symbol [d(dθ)] as a section of the Berezinian, not a
density. So under a change of variables θ → λθ, λ ∈ C, which induces dθ → λdθ, we want
[d(dθ)] → λ[d(dθ)], with no absolute value; this is a special case of (3.4). So consistency
with (3.38) requires
1
δ(λdθ) = δ(dθ), (3.39)
λ
again with no absolute value. Another consequence becomes clear if there are two odd
variables θ and θ ′ . We would like
Z
[d(dθ) d(dθ ′)] δ(dθ)δ(dθ ′ ) = 1. (3.40)

– 19 –
On the other hand, the integration measure [d(dθ) d(dθ ′)] is odd in dθ and dθ ′ . So the dual
delta functions are also anticommuting

δ(dθ)δ(dθ ′ ) = −δ(dθ ′ )δ(dθ). (3.41)

Thus, the calculus of distributional functions of dθ and integrals over them is really a
formal algebraic machinery, like the Berezin integral.
Though this will not be important in the present notes, one might wonder how to
interpret the above formulas if M is a real supermanifold, so that we hope to interpret the
dθ’s as real variables. The unfamiliar signs in eqns. (3.39) and (3.41) mean that the symbol
δ(dθ) differs slightly from its usual meaning. Instead of defining a delta function as a linear
function on smooth functions, as is common, we define it as a linear function on smooth
differential forms (mathematically, an object of this kind is called a current). To explain
the idea, let R be a copy of the real line but with no chosen orientation, and let Ω1 (R)
be the space of smooth one-forms on R. An element of Ω1 (R) is an expression f (x) dx
where f (x) is a smooth function. Such an expression cannot be integrated until we pick
an orientation on R. We interpret δ(x) as a linear function on Ω1 (R) that maps f (x) dx
to f (0). With this interpretation of δ(x), we have δ(−x) = −δ(x), since d(−x) = −d(x).
Roughly speaking, δ(x) in this sense differs from the usual δ(x) by a choice of orientation
of the normal bundle to the submanifold x = 0.
Similarly, with this interpretation, the minus sign in eqn. (3.41) is natural. Let R
be a copy of the real line parametrized by x, and let Z be any manifold. We define δ(x)
as a map from smooth k-forms on R × Z to smooth k − 1-forms on Z as follows: δ(x)
annihilates a form that does not contain dx, and it maps ψ = dx ∧ ω (for any form ω), to
the restriction of ω to {0} × Z, where {0} is the point x = 0 in R. This operation would
be integration of δ(x)ψ over the first factor of R × Z, with a standard interpretation of
δ(x), except that integration would require an orientation of that first factor, which we
have not assumed. If y is another real coordinate, we define δ(y) in the same way. Now
let us consider smooth forms on a product Rx × Ry , where Rx and Ry are factors of R
parametrized respectively by x and by y. We understand the product δ(x)δ(y) to represent
successive action of the operator δ(y), mapping smooth k-forms on Rx × Ry to smooth
k − 1-forms on Rx × {y = 0}, and δ(x), mapping smooth k − 1-forms on Rx to smooth
k − 2-forms at the point x = y = 0. δ(y)δ(x) is understood similarly. With this meaning
of the symbols, we have δ(x)δ(y) = −δ(y)δ(x); indeed, δ(y)δ(x) maps the form dx ∧ dy to
+1 and δ(x)δ(y) maps it to −1.

3.3.3 Equivalence With The Berezin Integral on M

We have defined integration of an integral form ω on M by means of a Berezin integral
over ΠT M of D(x, dx) ω(x, dx), where ω(x, dx) is the function on ΠT M corresponding to
the integral form ω, and D(x, dx) is the natural integration measure of ΠT M .
On the other hand, we also have a fibration π : ΠT M → M . And as explained in
section 3.1.3, a Berezin integral on the total space of a fibration can be performed by first
integrating over the fibers of a fibration. So if ω is an integral form, we can define a section

– 20 –
σ of Ber(M ) by acting with π∗ on the section D(x, dx) ω(x, dx) of Ber(ΠT M ):

σ = π∗ (D(x, dx) ω(x, dx)) . (3.42)

This operation is defined for all integral forms ω, but if ω has positive codimension then
σ = 0. (For ω to have positive codimension, it is a linear combination of terms that are
either missing an odd variable dti for some i and vanish upon integration over dti , or are
proportional to some ∂ n δ(dθ s ), n > 0, and vanish upon integration over dθ s . Such terms
are annihilated by π∗ .)
Now comparing the basic property (3.14) of integration over the fibers of a fibration
with the definition (3.37) of integration of an integral form, we see that
Z Z Z
σ= D(x, dx) ω(x, dx) = ω. (3.43)
M ΠT M M
R R
Here M σ is a Berezin integral and M ω is the integral of an integral form. This is the
equivalence between the two notions of integration on a supermanifold.
The attentive reader may notice a sleight of hand in this explanation. In discussing
the integral over the fibers of a fibration in section 3.1.3, we assumed that we were dealing
with an ordinary integral. But for cs manifolds, the integral over the fibers of ΠT M → M
is a formal algebraic operation, as explained at the end of section 3.3.1. However, this
algebraic operation does have the necessary properties – notably it transforms like an
ordinary integral under a change of variables – for the above derivation. (In fact, locally,
the θ’s can be given a real structure, and the algebraic operation is equivalent to an ordinary
integral. See footnote 4 in section 2.2. As usual, the discussion of the equivalence between
the two types of integral can be reduced to the local case by taking ω to be a sum of forms
each of which is supported in a small open set.)

3.3.4 Integration Over Submanifolds Of Codimension r|0

So far we have understood that a compactly supported integral form on M of codimen-
sion zero can be naturally integrated over M . What can we do with an integral form of
codimension r?
The answer to this question is that if N ⊂ M is a supermanifold of codimension r|0
whose normal bundle is oriented, then an integral form of codimension r can be naturally
integrated over N . All that one really needs to know here is that associated to such an N
there is a Poincaré dual r-form δN , a sort of delta function r-form supported on N . If N
is locally defined by vanishing of even functions f1 . . . fr , which are real-valued when the
odd variables θ s all vanish, then one can define

δN = δ(f1 ) . . . δ(fr )df1 ∧ · · · ∧ dfr , (3.44)

where one orders the factors so as to agree with the orientation of the normal bundle to
N . Just as on a bosonic manifold, this formula for δN does not depend on the choice of
the functions fi , so it makes sense globally.

– 21 –
 Now recall that there is a naturally defined wedge product of a differential form with
an integral form. If µ is an integral form of codimension r, then δN ∧ µ is an integral form
of top dimension. So we can define
Z Z
µ= δN ∧ µ. (3.45)
N M

Suppose that we displace N slightly to a nearby submanifold N ′ (or in general, suppose

we replace N by a homologous submanifold N ′ ). In such a situation, just as for bosonic
manifolds, one has
δN − δN ′ = dτ (3.46)
where τ is a compactly supported r − 1-form. So
Z Z Z
µ− µ= dτ ∧ µ. (3.47)
N N′ M

Now suppose that dµ = 0. Then dτ ∧ µ = d(τ ∧ µ) so

Z Z Z
µ− µ= d(τ ∧ µ). (3.48)
N N′ M

The supermanifold version of Stokes’s theorem, to which we turn presently (eqn. (3.53)),
ensures that the right hand side of (3.48) vanishes. So if dµ = 0 and N is homologous to
N ′ , we have Z Z
µ= µ, (3.49)
N N′
just as for differential forms on an ordinary manifold.

3.4 The Supermanifold Version Of Stokes’s Theorem

If ν is an integral form of codimension 1 on a supermanifold, then dν is an integral form
of codimension 0 and one can try to integrate it. The most basic statement of the super-
manifold version of Stokes’s theorem is simply that if ν is a compactly supported integral
form on Rp|∗q of codimension 1, then
Z
dν = 0. (3.50)
Rp|∗q

If fact, if we write d = d0 + d1 with

p
X X q
∂ ∂
d0 = dti i
, d1 = dθ s s , (3.51)
∂t ∂θ
i=1 s=1

then Z Z
d0 ν = d1 ν = 0. (3.52)
Rp|∗q Rp|∗q
The integral of d1 ν over the odd variables vanishes because d1 ν is a sum of terms none of
which are proportional to the product θ 1 . . . θ q of all odd variables. And the integral of d0 ν
over the even variables vanishes by the ordinary bosonic version of Stokes’s formula.

– 22 –
 We can immediately extend this to a general supermanifold M . If ν is a compactly
supported integral form on M of codimension 1, then
Z
dν = 0. (3.53)
M

To show this, we proceed just as in the definition of the Berezin integral. We write ν as the
sum of codimension 1 integral forms να , each of which is supported in an open set Uα ⊂ M
that is isomorphic to an open set in Rp|∗q . So Stokes’s formula for an arbitrary M follows
from the special case M = Rp|∗q .
Just as in the bosonic case, a more general version of Stokes’s theorem applies to a
supermanifold with boundary. First we have to define a supermanifold with boundary. This
is a little tricky and there are several ways to proceed. The simplest way to find the right
definition is to first consider the case that everything is happening inside a supermanifold
Y without boundary. In Y , one considers a submanifold N of codimension 1 defined by an
equation f = 0. As usual, f is required to be real when (but in general only when) the odd
variables vanish. For the moment, we assume that the function f is globally-defined, in
which case,10 roughly speaking, we can define a compact supermanifold M , with boundary
N , by the condition f ≤ 0. Let Θ(x) be the function of a real variable that is 1 for x ≥ 0
and 0 for x < 0. Then for any integral form σ on Y of codimension 0 such that Θ(−f )σ has
compact support (we do not assume that the support of σ is contained in M ), we define
Z Z
σ= Θ(−f )σ. (3.54)
M Y

Since f is only real modulo nilpotents, the interpretation of this formula involves consider-
ations such as Θ(a + bθ 1 θ 2 ) = Θ(a) + bθ 1 θ 2 δ(a), where a is real but b need not be. Because
even nilpotent expressions like bθ 1 θ 2 are neither positive nor negative (or even real, in gen-
eral), it is actually the integration formula that gives a precise meaning to the statement
that M is defined by the condition f ≤ 0.
Why is this an interesting situation to consider? Superstring perturbation theory
provides a good example. In that context, let Y be the moduli space of super Riemann
surfaces. Y is not compact – it has noncompact ends corresponding to the infrared region
– and one often needs to integrate over Y an integral form whose behavior at infinity is
delicate. One may want to introduce an infrared regulator by restricting the integral over
Y to an integral over a large compact subset M ⊂ Y . The version of Stokes’s theorem that
we are in the process of describing governs the boundary terms that will arise in integration
by parts in this sort of situation.
With this understood, let us take the form σ in (3.54) to be exact, say σ = dν. Then
Z Z Z Z
dν = Θ(−f )dν = d(Θ(−f ) ν) − (d(Θ(−f ))) ν. (3.55)
M Y Y Y
10
The topological fact that we are using is that a codimension 1 submanifold Nred ⊂ Yred can be defined
by a globally-defined real-valued function fred if and only if Nred is the boundary of some Mred ⊂ Yred . In
one direction, if fred exists, we define Mred by the condition fred < 0. In the other direction, one uses the
fact that the cohomology class Poincaré dual to Nred vanishes if Mred exists. This implies that the object
that can always be written locally as δNred = δ(f )df is in fact dΘ(−f ) for a globally-defined f .

– 23 –
On the right hand side of (3.55), we can drop the exact term d(Θ(−f )ν), by using Stokes’s
theorem (3.53) for a supermanifold without boundary. On the other hand, d(Θ(−f )) =
−δ(f )df = −δN , where δN was defined in eqn. (3.44). So, from (3.45), we have
Z Z
dν = ν, (3.56)
M N

where N = ∂M is defined by f = 0. This is Stokes’s theorem in this situation.

This construction did not really depend on the specific choice of the function f . That
is because the function Θ(−f ), which is all we really used, is invariant under

f → eφ f, (3.57)

where φ is real when the odd variables vanish. By contrast, Θ(−f ) is not invariant under
something like f → f + αβ where α and β are odd variables (coordinates or moduli),
since Θ(−f − αβ) = Θ(−f ) − αβδ(−f ). So the definition of M and N really relied on
an equivalence class of functions f modulo the relation (3.57). We do not need f to be
globally-defined (though it actually is always possible to find a globally-defined f ); only
Θ(−f ) has to be globally-defined. It suffices to cover the region near the boundary of M
with open sets Va in each of which one is given a function fa such that fa = exp(φab )fb in
Va ∩ Vb , for some function φab . Then Θ(−fa ) = Θ(−fb ) in Va ∩ Vb . This gives a function
that we can call Θ(−f ) that is defined near the boundary of M (that is, near the vanishing
locus of any of the fa ) and since it is 1 in the interior of M wherever it is defined (that is
in ∪a Va ), we can extend its definition so that it equals 1 throughout the interior of M .

3.5 Supermanifold With Boundary

Having come this far, we can give the appropriate definition of a supermanifold M with
boundary without requiring M to be embedded in a supermanifold Y without boundary.
The only subtle point is that part of the definition of M involves a function f , or more
precisely an equivalence class (3.57) of such functions, whose vanishing defines the boundary
of M .
We suppose that the reduced space Mred of M is an ordinary manifold with boundary
of dimension p. We cover M with small open sets Uα that do not intersect its boundary,
and small open sets Va that do intersect its boundary. Each Uα is endowed as usual with
local coordinates t1α . . . tpα |θα1 . . . θαq providing an isomorphism with an open subset in Rp|∗q .
Each Va is likewise endowed with local coordinates t1a . . . tpa |θa1 . . . θaq , now providing an
isomorphism with an open subset of the half-space tp ≤ 0 in Rp|∗q . Thus, the boundary of
M is always at tpa = 0. As usual, we have gluing functions in intersections Uα ∩ Uβ , Uα ∩ Vb ,
and Va ∩ Vb . The only detail that is in any way special to the case of a supermanifold with
boundary is that in Va ∩ Vb , we put a special condition on one of the gluing functions: we
require that the gluing function defining tpa when expressed in terms of t1b . . . | . . . θbq takes
the form
tpa = exp(φab (t1b . . . | . . . θbq ))tpb . (3.58)

– 24 –
In other words, tpa and tpb are equivalent in the sense of (3.57). (In fact, it is always possible
to pick coordinates such that φab = 0 and tpa = tpb , though there is no particularly natural
way to do this.)
The above construction defines what we mean by a p|q-dimensional cs supermanifold
with boundary. The relation (3.58) ensures that we can consistently set tpa = tpb = 0 and
these conditions define an ordinary cs supermanifold N without boundary of dimension
p − 1|q.
Integration on a supermanifold with boundary is defined in a way that should be
almost obvious. If σ is a compactly supported section of Ber(M ) whose support is in just
one of the Uα or Va , its integral is defined by a naive Berezin integral (3.2); the integral
of a general section of Ber(M ) is defined with the help of a partition of unity. The only
R
subtlety is that if σ has compact support in one of the boundary open sets Va , then Va σ
is invariant under those coordinate transformations that act on tpa by tpa → eφ tpa , but not
under something like tpa → tpa + θa1 θa2 . This is the reason that the condition (3.58) is part of
the definition of a supermanifold with boundary.11
Stokes’s theorem for a supermanifold with boundary says that if ν is an integral form
on M of codimension 1 than Z Z
dν = ν. (3.59)
M N
If the gluing functions of M are real analytic, so that they can continued to positive (but
perhaps small) values of tpa , then this theorem is not more general than (3.56). However, it
holds whether or not there is such a continuation. As usual, one first proves it by reduction
to the ordinary form of Stokes’s theorem for the case that ν is compactly supported in just
one of the Uα or Va ; the general case follows by using a partition of unity.
This construction is useful in superstring perturbation theory in the presence of D-
branes and/or orientifold planes, since the moduli space of open and/or unoriented super
Riemann surfaces is a supermanifold with boundary in the sense just described.

3.6 Integration On More General Submanifolds

We have so far considered two types of representation of the Weyl algebra that was intro-
duced in section 3.2. Differential forms correspond to functions on ΠT M with polynomial
dependence on the even variables dθ i. They can be integrated over submanifolds of M of
dimension p|0, that is submanifolds with zero fermionic dimension. Integral forms corre-
spond to distributional functions on ΠT M with support at dθ i = 0. They can be integrated
over submanifolds with maximal fermionic dimension.
If we want to be able to integrate over submanifolds of M whose odd dimension is
positive, but not maximal, we must introduce more general representations of the Weyl
algebra. We need functions on ΠT M that have polynomial behavior with respect to some
of the dθ i and compact support with respect to others. Let us practice with a toy example of
11
Here is a simple example on a half-space t ≤ 0 in R1|∗2 . The Berezin integral I = [dt|dθ1 dθ2 ]w(t)
R

vanishes if w is a function of t only. But if we are permitted to transform from t to t∗ = t + θ1 θ2 , we get

I ∗ = [dt∗ |dθ1 dθ2 ]w(t∗ − θ1 θ2 ) = [dt∗ |dθ1 dθ2 ] w(t∗ ) − θ1 θ2 w′ (t∗ ) , and now there is a boundary term at
R R

t = 0. If w is compactly supported, there is no boundary term at t∗ = −∞.

∗

– 25 –
R0|∗2 , with odd coordinates θ 1 , θ 2 . We can represent the Weyl algebra starting with a state
|↑↓i that is annihilated by dθ 1 and by ∂/∂dθ 2 . In the language of superstring perturbation
theory, this choice of “picture” is midway between differential forms and integral forms.
For an example of integrating a form of this type, let us take

ω = θ 2 δ(dθ 1 ) (3.60)

and try to integrate over the 0|1-dimensional subspace N ⊂ R0|∗2 defined by the equation

aθ 1 + bθ 2 = 0, a, b ∈ C. (3.61)

Along N , we can eliminate θ 2 by θ 2 = −(a/b)θ 1 , dθ 2 = −(a/b)dθ 1 . So we can parametrize

N by θ 1 , and the integral we have to do is
Z
a
D(θ 1 , dθ 1 ) (−a/b)θ 1 δ(dθ 1 ) = − . (3.62)
R0|∗1 b
This is homogeneous in the parameters a, b, as it must be, since N is unchanged in scaling
those parameters. A noteworthy fact, however, is that the integral has a pole at b = 0.
It arises because if b = 0, the definition of N gives θ 1 = 0, so dθ 1 is zero when restricted
to N . But we cannot restrict δ(dθ 1 ) to dθ 1 = 0. So σ can be integrated over a generic
codimension 0|1 submanifold of R0|∗2 , but not over every one.
More generally, given any supermanifold M , we consider any class of (distributional)
functions on ΠT M with three basic properties:
(1) For any local coordinate system t1 . . . | . . . θ q on M , the class of functions is closed
under the Clifford-Weyl algebra, that is under the action of dx and ∂/∂dx, where x is any
of t1 . . . | . . . θ q .
(2) The class of functions on ΠT M is also closed under multiplication by any function
f (t1 . . . | . . . θ p ) on M , and under addition.
(3) Finally, we require that the given class of forms is invariant under the scaling
symmetry (3.34) of ΠT M that scales all fiber coordinates by a common factor λ, and we
say that a form that transforms as λr under this operation has scaling weight r.
We call such objects pseudoforms (this terminology differs slightly from the literature)
or sometimes simply forms. The conditions imply among other things that the exterior
derivative
X ∂
d= dxI I (3.63)
∂x
I=1...|...q

acts on functions of the given class, increasing the scaling weight by 1, and that it is possible
to multiply a function of the given class by an ordinary differential form (understood as a
function on ΠT M with polynomial dependence on the dθ’s), increasing its scaling weight
by the degree of the differential form.
The requirement of scale-invariance of the given class of functions implies that the
support of the functions in the space parametrized by dθ 1 . . . dθ q is a conical submanifold.
For superstring perturbation theory, it seems sufficient to consider the case that the cone
is just a linear subspace – so that we consider wavefunctions with polynomial dependence

– 26 –
on some of the dθ’s and delta function dependence on the others. If the wavefunctions
are localized in s variables, we say that the class of functions in question correspond to
pseudoforms of picture number −s. The terminology is suggested by the usual terminology
in superstring perturbation theory. The picture number is constant for a whole class of
pseudoforms corresponding to a representation of the Clifford-Weyl algebra. Clearly, there
are many classes of pseudoforms with the same picture number, since there are many linear
subspaces (or nonlinear cones) with the same dimension. Some operations that change the
picture number will be described in section 4.
If a form has scaling weight r and is localized with respect to n dθ’s (so its picture
number is −n), we call it a form of superdegree m|n, with m = r + n. A simple example
of a form of superdegree m|n is

ω = f (t1 . . . | . . . θ q )dt1 . . . dtm δ(dθ 1 ) . . . δ(dθ n ). (3.64)

We have used the first m dt’s and the first n dθ’s in writing this formula. The form ω
contains n delta functions of dθ’s, and has scaling weight m−n, so it is indeed of superdegree
m|n. The exterior derivative increases the scaling weight by 1 without changing the picture
number, so it maps forms of superdegree m|n to forms of superdegree m + 1|n.
Given a form of superdegree m|n, we can try to integrate it on a submanifold N ⊂ M
of dimension m|n. We say “try” because though there is a rather natural operation, it
is not defined for all N ; for some choices of N , one will run into problems, as in the
simple example that we gave above with R0|∗2 . This should not be too discouraging, since
something similar happens in ordinary calculus: if one is given a k-form on an ordinary
noncompact manifold, one can try to integrate it over k-dimensional submanifolds, but
sometimes the integral will turn out to diverge.
The procedure for integration uses the fact that if N is embedded in M , then ΠT N is
embedded in ΠT M . So we can try to restrict a function ω(x, dx) on ΠT M to a function
on ΠT N , which we will call by the same name. The only thing that may go wrong is that
to make the restriction, we may need to evaluate a delta function δ(dθ) (where θ is some
linear combination of the odd coordinates) at dθ = 0; this is what happened at b = 0 in
the practice example. If ω(x, dx) is localized with respect to n dθ’s, and N has fermionic
dimension n, then this will not occur at a generic point on a generic N . If N is such
that ω(x, dx) can be restricted to ΠT N without running into trouble anywhere, then the
restriction is everywhere localized with respect to all dθ’s, and is an integral form on N . If
ω(x, dx) is m|n-form, then its scaling dimension is m − n, which ensures that if ω(x, dx)
can indeed be restricted to give an integral form on N , then the resulting integral form is
a top form. Given all this, the form ω(x, dx) can be integrated over N in the usual way.
Suppose instead that U ⊂ M has dimension m + 1|n, with ω still an m|n-form, and
suppose that ω(x, dx) can be restricted to U . Then its support as a function of dθ is
entirely at the origin, so the restriction of ω(x, dx) to U is again an integral form. But the
scaling dimension of ω(x, dx) is too small by 1 to make a top form on U ; rather, ω(x, dx) is
an integral form of codimension 1. So we cannot integrate ω over U , but we can integrate
dω.

– 27 –
 Now let U have boundary12 N . Then, applying the supermanifold version of Stokes’s
theorem to the integral form ω on U , we have
Z Z
dω = ω. (3.65)
U N
R
It follows from this that if dω = 0, then N ω is invariant under small displacements of
N ⊂ M and more generally under a certain class of allowed homologies.

3.6.1 Another Example

We conclude by describing another entertaining example. In R1|1 , with coordinates t|θ, we
consider the form
ω = θ dt δ′ (dθ), (3.66)
This form has scaling degree −1 and picture number −1. So it is a 0|1-form, and we can
hope to integrate it over a submanifold N of R1|1 of dimension 0|1. We define a suitable
N by the equation t = αθ, with α an odd parameter. Upon restriction to N , we have
dt = −αdθ, so ω = −θαdθδ′ (dθ) = θαδ(dθ). We used the fact that for an even variable y,
yδ′ (y) = −δ(y). So Z Z
ω= D(θ, dθ)θαδ(dθ) = α. (3.67)
N

4 More Operations On Forms

4.1 Wedge Products And Contractions

Here we will describe some interesting operations on pseudoforms, largely following [27, 28].
Some of this material may be useful background for superstring perturbation theory. (On
some points whose relevance is not immediately clear, we provide only references to the
literature.)
One basic operation is the wedge product with a one-form. If α is a one-form on M ,
we define an operator eα that acts on pseudoforms by multiplication by the corresponding
function α(x, dx) on ΠT M :

[eα ω](x, dx) = α(x, dx)ω(x, dx). (4.1)

This operation increases the degree by 1 and does not change the picture number. The
statistics of eα are the same as those of α. So eα is odd if α = dt with t an even variable,
but eα is even if α = dθ where θ is odd.
For a very simple operation that can change the picture number, we define an operator
δ(eα ) as multiplication by δ(α(x, dx)):

[δ(eα )ω](x, dx) = δ(α(x, dx))ω(x, dx). (4.2)

12
Strictly speaking, to match the definition we gave in section 3.4 of integration on a supermanifold with
boundary, we should proceed here in more steps. We should first introduce a dimension m+1|n submanifold
Y ⊂ M without boundary. Then we let N be a codimension one submanifold of Y defined by an equation
f = 0, and we define U by f < 0.

– 28 –
These operators obey
eα δ(eα ) = δ(eα )eα = 0. (4.3)
If α is an odd one-form such as dt, then eα is an odd variable. Since an odd variable is its
own delta function, we have simply

δ(eα ) = eα . (4.4)

In particular, in this case δ(eα ) is fermionic, does not change the class of a form, and maps
an m|n-form to an m + 1|n-form. But if α is even, for instance α = dθ, then multiplication
by δ(α) = δ(dθ) is the most simple example of an operator that changes the class of a
form. It maps an m|n-form to an m|n + 1-form. The operator δ(eα ) is odd regardless of
whether α is even or odd, though for even α this is subtle; see eqn. (3.41). For even α, the
operator δ(eα ) is not defined on all classes of pseudoform, since an object with support at
dθ = 0 cannot be multiplied by δ(dθ).
If M is a supermanifold of dimension p|q, and γ 1 . . . γ q |ζ 1 . . . ζ p is a basis of even and
odd one-forms, then the operation

f → δ(eγ 1 ) . . . δ(eγ q )δ(eζ 1 ) . . . δ(eζ p )f (4.5)

supplies all the missing factors of dt and δ(dθ) and maps a function f to an integral form
of top degree.
The dual operation to multiplication by a one-form is contraction with a vector field
on M . For a vector field
X ∂
V = VI I, (4.6)
∂x
I=1...|...q

we define the contraction operator

X ∂
iV = VI (4.7)
∂dxI
I=1...|...q

This operator has the opposite statistics to V . It maps a form of superdegree m|n to one
of superdegree m − 1|n. It is again useful to introduce delta function operators that will
obey
iV δ(iV ) = δ(iV )iV = 0. (4.8)
Here at first we treat separately the cases of an even vector field v or an odd vector field ν.
If v is an even vector field, so that iv is odd, then again the definition of δ(iv ) is obvious:

δ(iv ) = iv . (4.9)

It is less obvious how to define δ(iν ) for an odd vector field ν. However, the appropriate
definition has been given in [27]. If ν is an odd vector field on M , it can be viewed as a
section of ΠT M . And hence, for u an even scalar, uν is a section of ΠT M and it makes
sense to act on ΠT M by shifting the fiber coordinates dx by dx → dx + uν. This makes
possible the definition
Z
[δ(iν )ω](x, dx) = [du] ω(x, dx + uν), (4.10)

– 29 –
(for ω in a suitable class of pseudoforms) which can be seen to satisfy (4.8). This operation
maps a form of superdegree m|n to a form of superdegree m|n − 1. The operator δ(iν ) is
odd, since the integration form [du] is odd. Thus, for example, if ν and ν ′ are two odd
vector fields, we have
Z
[δ(iν )δ(iν )ω](x, dx) = [du du′ ] ω(x, dx + uν + u′ ν ′ ),
′ (4.11)

and this is odd in ν and ν ′ since [du du′ ] = −[du′ du]. Thus δ(iV ) is odd regardless of
whether V is even or odd.
A little thought shows that actually we can define the delta function operation in the
same way also for an even vector field v. If v is an even vector field and η is an odd constant,
then ηv is an odd vector field and thus again a section of ΠT M . So we can define
Z
[δ(iv )ω](x, dx) = [dη] ω(x, dx + ηv), (4.12)

and this is equivalent to the previous definition.

Now suppose that we are given a collection of even and odd vector fields v1 . . . vm |ν1 . . . νn .
Acting with the whole product of delta functions

δ(iv1 ) . . . δ(ivm )δ(iν1 ) . . . δ(iνn ) (4.13)

we get an operator that integrates over the m|n-dimensional subspace of the fibers of ΠT M
that is generated by v1 . . . | . . . νn . Explicitly, to act with the product of delta function
operators, we introduce n even and m odd integration variables u1 . . . un and η 1 . . . η m and
perform the Berezin integral
Z  X X 
[du1 . . . | . . . dη m ] ω x, dx + ui ν i + η j vj , (4.14)

assuming that ω is a form of an appropriate type so that this integral makes sense. Let us
denote the integral as σ(v1 . . . | . . . νn ). Actually, σ is a function of x and possibly the fiber
variables in ΠT M that we have not integrated over; it depends on the values at x of the
vector fields v1 . . . | . . . νn . To simplify the notation, we denote σ simply as σ(v1 . . . | . . . νn ).
The vector fields v1 . . . | . . . νn span a subbundle Vm|n of the tangent bundle T M . Re-
versing the statistics gives a subbundle ΠVm|n of ΠT M . The integral in (4.14) is an
integral over ΠVm|n , but it is not quite true that the integral σ(v1 . . . | . . . νm ) depends on
the chosen vector fields only via the subspace they generate. If we replace v1 . . . | . . . νn by
another collection of vector fields v1′ . . . | . . . νn′ that span the same subspace, with
   
v1′ v1
 v′  v 
 2  2
. .
 ..   .. 
   
  = W  , (4.15)
− −
   
 ..   .. 
. .
νn′ νn

– 30 –
for some linear automorphism W of ΠVm|n , then we can compensate for this in (4.14) by
redefining the integration variables u1 . . . | . . . η n by a dual linear automorphism, but this
will change the integration measure by a factor Ber(W ). So we get

σ(v1′ . . . | . . . ηn′ ) = Ber(W )σ(v1 . . . | . . . ηn ). (4.16)

A special case of this is that m|n equals the dimension p|q of M . In this case,
v1 . . . | . . . ηn is a basis of the tangent space to M . A function depending on such a ba-
sis and obeying (4.16) is a section of Ber(M ), as explained in section 3.1.1. The integral
in (4.15) is simply an integral over the fibers of ΠT M → M , and what we have arrived
at is an operation already described in section 3.3.3: the map from functions on ΠT M to
sections of Ber(M ), by integrating over the fibers of ΠT M .
If on the other hand m|n does not coincide with p|q, then we have described something
more general. For some choices of ω(x, dx), a function σ(v1 . . . | . . . νn ) obtained by inte-
grating over an m|n-dimensional subbundle of ΠT M may still depend on the other fiber
coordinates. However, if ω is a pseudoform of degree m|n in the sense described in section
3.6, then (for a given choice of vector fields v1 . . . | . . . ηn ) σ will be a function on M , and
not a more general function on ΠT M . In this case, σ(v1 . . . | . . . νn ) is a differential form
on M of degree m|n, in the language of [17, 18], as reviewed in [9]. Such an object is by
definition a function of a point x ∈ M that depends on m even and n odd vectors in the
tangent space to x in M , with the restriction (4.16), and also obeys a certain fundamental
relation that is described on p. 57 of [9]. This relation is automatically satisfied when σ is
defined by an integral (4.14). The fundamental relation might be important for some sort
of string or brane actions on supermanifolds.

4.2 Picture-Changing
For a vector field V , the operator δ(iV ) defined in section 4 is not invariant under multiply-
ing V by a constant, V → λV . Rather this operation rescales δ(iV ) by λ or λ−1 , depending
on the statistics of V .
However, for the case of an odd vector field ν, there is a natural “picture-changing”
operation, defined in [27], which is invariant under ν → λν and only depends on the 0|1-
dimensional group of automorphisms of M generated by ν. Let us call this group F ; it is
isomorphic to R0|1 (or R0|∗1 ). The F action on M corresponds to a map m : F × M → M .
Given a pseudoform ω(x, dx) on M , we can pull it back to a form m∗ (ω) on F × M . Then
we have a projection π : F × M → M that forgets the first factor. Integrating over the
fibers of π, we get again a pseudoform on M . So this gives an operation Γν = π∗ m∗ on
pseudoforms that (because we carry out one odd integration in integrating over the fibers
of π) maps an m|n form to an m|n − 1 form.
To make this explicit, we parametrize R0|1 by an odd variable τ , and write the action
of the group on M as
xI → exp(τ ν J ∂/∂xJ )xI = xI + τ ν I . (4.17)

– 31 –
As usual, this expansion stops quickly since τ 2 = 0. The pseudoform m∗ (ω) on F × M is
simply  
∂ν A
ω(x + τ ν, d(x + τ ν)) = ω x + τ ν, dx + dτ ν − τ A dx . (4.18)
∂x
And integration over the fibers means integrating over τ and dτ with the natural measure
D(τ, dτ ). So the picture-changing operator Γν is defined by
Z
[Γν ω](x, dx) = D(τ, dτ ) ω(x + τ ν, d(x + τ ν)). (4.19)

As explained in [27, 28], this is an abstract version of the picture-changing operator of

superstring perturbation theory. It is shown there that
1
Γν = (δ(iν )Lν + Lν δ(iν )) , (4.20)
2
where Lν is the Lie derivative,
Lν = diν + iν d. (4.21)
The exterior derivative d commutes with m∗ and π∗ , so it commutes with Γν . Additional
interesting facts can be found in [27].

5 Complex and Smooth Supermanifolds

5.1 First Orientation

In this concluding section, we will compare complex supermanifolds to smooth ones. The
relevance for superstring perturbation theory will become clear.
For a first orientation to the problem, suppose that X is an ordinary complex manifold,
of dimension a, with local holomorphic coordinates z 1 . . . z a . X can always be viewed as a
smooth manifold of dimension 2a, local real coordinates being ti = Re z i = (z i + z i )/2 and
√
ta+i = Im z i = (z i − z i )/2 −1, with i = 1 . . . a.
We can do the same thing on a supermanifold X – and this is often done in the literature
– if we are allowed to take the complex conjugates (and therefore the real and imaginary
parts) of both even and odd variables. This would turn a complex manifold of dimension
a|b into a real supermanifold of dimension 2a|2b. But for superstring perturbation theory,
that is not what one wants to do, since we are never allowed to take the complex conjugate
of an odd variable.13 Instead a typical thing that we want to do is to relate X to a cs
supermanifold of dimension 2a|b, with no doubling of the fermionic dimension.
For example, let X be the worldsheet of a heterotic string. From a holomorphic
point of view, X is a complex supermanifold of dimension 1|1 with local coordinates z|θ.
(Holomorphically, a heterotic string worldsheet is not a generic complex supermanifold of
dimension 1|1; it has the additional structure of a super Riemann surface. The details,
which are reviewed in [2], need not concern us here.) But in heterotic string theory, there
13
And consequently, unless we are presented with a splitting, so that we know which even variables are
“purely bosonic,” we cannot take the complex conjugate of an even variable either.

– 32 –
are antiholomorphic as well as holomorphic worldsheet fields, and the antiholomorphic
dimension is 1|0. So we want to be able to view X as a smooth supermanifold of dimension
2|1. In what sense can we do this?
The question is trickier than one might at first think. We will consider primarily
two points of view, which generalize the following considerations in conformal field theory
on an ordinary Riemann surface. One typically considers correlation functions which are
neither holomorphic nor antiholomorphic. The expectation value of a product of operators
Φ1 . . . Φs is often written
hΦ1 (z 1 ; z1 ) . . . Φs (z s ; zs )i. (5.1)
There are two contrasting points of view about this formula:
(1) z is really the complex conjugate of z. Denoting an operator as Φ(z; z) is merely
a way of saying that Φ is an (operator-valued) function on Σ that is neither holomorphic
nor antiholomorphic.
(2) A Riemann surface Σ is a real-analytic two-manifold and as such it can be analyt-
ically continued and viewed as a real slice in a two-dimensional complex manifold. (How
to do this concretely is explained in section 5.4.) When this is done, z and z become
independent complex variables; to emphasize this, we write ze instead of z. The correlation
functions are likewise real analytic (away from singularities when distinct points collide)
so they can be analytically continued to holomorphic functions

hΦ1 (e
z1 ; z1 ) . . . Φs (e
zs ; zs )i (5.2)

of independent complex variables zi and zei . To be more precise, these functions are holo-
morphic when zei is sufficiently close to z i . Setting zei = z i , we get the usual correlation
functions on Σ. The notation (5.1) is a shorthand for all this.
In this section, we will attempt to generalize both points of view to supermanifolds.
The first point of view is perhaps more obvious, but the second point of view seems to be
more robust.

5.2 Complex Supermanifold As A Smooth Supermanifold

Let X be a complex manifold of dimension a|b. We would like to view X as a smooth cs
supermanifold of dimension 2a|b. We will first explain what appears to be the best that
one can do, and then explain why the construction is not completely natural.
First of all, if Y is an ordinary complex manifold of complex dimension a, then the
complex conjugate of Y , which we denote Y , is defined to be the same manifold with
opposite complex structure. So holomorphic functions on Y are antiholomorphic functions
on Y , and vice-versa. Concretely, if one covers Y by open sets Uα with local holomorphic
coordinates zαi , i = 1 . . . a and gluing maps

zαi = fαβ
i
(zβ1 . . . zβa ), (5.3)

then Y is covered by the same open sets Uα with local holomorphic coordinates zeαi = zαi
and gluing maps
i
zβ1 . . . zeβa ).
zeαi = f αβ (e (5.4)

– 33 –
We recall that the definition of the function f is such that f (w) = f (w), so that the above
relations are consistent with zeαi = zαi .
Now let X be a complex supermanifold of dimension a|b. We cover X by open sets Uα
and as usual describe X by local holomorphic coordinates zα1 . . . | . . . θαb with holomorphic
gluing laws:

zαi = fαβ
i
(zβ1 . . . | . . . θβb )
θαs = ψαβ
s
(zβ1 . . . | . . . θβb ). (5.5)

X also has a reduced space Xred , which is an ordinary complex manifold of dimension a.
Its gluing relations are obtained from those in (5.5) by setting all odd variables – both
the θ’s and the possible odd moduli of X – to zero. (Why the odd variables must be
set to zero to define the reduced space was explained in section 2.1.1.) Gluing laws for
X red , the complex conjugate of Xred , are obtained by setting the odd variables to zero and
complex-conjugating:
i
zβ1 . . . |0 . . . 0).
zeαi = f αβ (e (5.6)
This ensures that
feαβ
i
(z 1β . . . z aβ ) = fαβ
i (z 1 . . . z a |0 . . . 0).
β β (5.7)
Now we introduce 2a real coordinates t1 . . . t2a :
√
zαi = tiα + −1 ti+a
α
√
zeαi = tiα − −1 ti+a
α . (5.8)

By virtue of (5.7), the gluing relations (5.5) and (5.6) are compatible with reality of t1 . . . t2a
when all odd variables (including odd moduli) vanish. So when regarded as gluing rela-
tions for the whole collection of variables t1 . . . t2a |θ 1 . . . θ b , these formulas define a smooth
manifold Xcs of dimension 2a|b. In particular, the odd coordinates of Xcs are the same as
those of X.
Starting with a complex supermanifold X, we have defined a smooth supermanifold Xcs
on which it makes sense to discuss both holomorphic functions (functions of zα1 . . . | . . . θαb )
and antiholomorphic functions (functions of zeα1 . . . zeαa ). This corresponds to point of view
(1) of section 5.1.

5.2.1 Critique
Though this construction is valid as far as it goes, there is a flaw: the passage from X to
Xcs is not as natural as one would like. It depends in a subtle way on the specific gluing
construction.
In fact, if we had a completely natural way to transform a complex supermanifold X to
a smooth supermanifold Xcs , it would follow that for any isomorphism ϕ : X ∼ = Y between
complex supermanifolds, we would get a corresponding isomorphism ϕcs : Xcs → Ycs . The
isomorphisms ϕcs would obey the same algebraic relations as the ϕ. So in particular, setting
Y = X, the supergroup G of automorphisms of X would act on Xcs .

– 34 –
 To see that this is a problem, suppose that X = CP1|1 , with homogeneous coordinates
u, v|θ. This example simply corresponds to a genus 0 worldsheet of the heterotic string.
The automorphism supergroup is14 G = PGL(2|1), acting by linear transformations of
the homogeneous coordinates. To promote X to a smooth supermanifold, we would want
to introduce antiholomorphic homogeneous coordinates u e, ve, which roughly speaking are
complex conjugates of u, v, but of course we introduce no corresponding odd variable θ. e
There is no way for PGL(2|1) to act on the pair u e, ve, so the passage from X to Xcs cannot
be completely natural.
Let us see what happens if we study this example with the gluing construction. We
can cover X with an open set U1 in which u 6= 0 and a second set U2 in which v 6= 0. In U1 ,
we take coordinates z1 = v/u, ζ1 = θ/u, and in U2 , we set z1 = u/v, ζ2 = θ/v. Following
the above recipe, the gluing laws of Xcs are
1
z2 =
z1
ζ1
ζ2 =
z1
1
ze2 = . (5.9)
ze1
In the starting point, without changing anything else, we could have replaced z1 by
′
z1 = z1 + αζ, where α is an odd parameter. This is an equally valid starting point, and
the above recipe gives a smooth supermanifold Xcs ′ with

1 αζ1
z2 = + ′ 2
z1′ (z1 )
ζ1
ζ2 = ′
z1
1
ze2 = . (5.10)
ze1
Note that the gluing law for ze1 and ze2 is unchanged, since to define it, we are supposed to
first set the odd variables to zero.
′ is isomorphic to X by an isomorphism that maps holomorphic functions
In fact, Xcs cs
to holomorphic functions and antiholomorphic functions to antiholomorphic functions. The
problem is that there are multiple equally natural isomorphisms that do this. We could
transform the gluing formulas (5.10) back into (5.9) by replacing z1′ by z1′′ = z1′ −αζ1 (which
happens to be the same as z1 ) or by replacing z2 with z2′ = z2 − αζ2 z2 .
So Xcs is unique up to isomorphism, but not up to a unique isomorphism. There is no
good way to pick a particular isomorphism.
The author suspects that it is better to develop a different approach, following point
of view (2) from section 5.1. In fact, in the process, we will get a new understanding of the
smooth supermanifold Xcs that was defined above.
As a preliminary, we will describe smooth submanifolds of a complex supermanifold.
Then we return to our theme in section 5.4.
14
If CP1|1 is viewed as a super Riemann surface, its automorphism group is reduced from PGL(2|1) to
OSp(1|2). This does not really affect the discussion in the text.

– 35 –
5.3 Submanifolds Of A Complex Supermanifold
So far we have found it difficult to give a completely natural notion of a not necessarily
holomorphic function on a complex supermanifold X.
A function on X would be a map from X to C, or perhaps to some ring generated over
C by odd elements.
Maps in the opposite direction behave much better. Thus, instead of a map from X
to C, let us consider a map from some smooth supermanifold M to X.
There is no problem at all in defining what we mean by a smooth map φ : M → X.
Locally, such a map expresses the local coordinates zα1 . . . | . . . θαb of X as smooth functions
of local coordinates t1τ . . . | . . . ητs of M . (To compare the descriptions in different local
coordinate systems, one just asks that the image in X of a given point in M should not
depend on the coordinates used, on either M or X.) Moreover, there is no problem in
deciding whether such a map is an embedding. We require that the map of reduced spaces
φred : Mred → Xred is an embedding, and that the differential of the map φ in the odd
directions is sufficiently generic.15 If φ : M → X is an embedding, we call M a smooth
submanifold (or subsupermanifold) of X.
There are many natural examples of such smooth submanifolds. In fact, we can adapt
something explained in section 2.2. Let Nred be any submanifold of the reduced space
Xred of X. We can view X as a smooth supermanifold by following the construction of
section 5.2 (for the present purpose, it does not matter that this construction is slightly
unnatural). Then as explained in section 2.2, to the submanifold Nred ⊂ Xred , we can
associate a submanifold N ⊂ X not quite uniquely, but in a way that is unique up to
homology (up to infinitesimal wiggling of the fermionic directions).
This gives an abundant source of smooth submanifolds of a complex supermanifold,
and the construction can be further generalized by thickening Nred in only some of the
fermionic directions. As an application, we will generalize to a complex supermanifold the
notion of the periods of a holomorphic differential form of top dimension on an ordinary
complex manifold.

5.3.1 Periods On A Complex Supermanifold

If Y is an ordinary complex manifold of complex dimension a, then a top-dimensional
holomorphic form σ on Y is in particular a closed differential form of degree a. If N is an
oriented middle-dimensional cycle in Y , and thus of real dimension a, we can define the
R
period N σ, which only depends on the homology class of N .
To generalize this to a complex supermanifold X, we need to define the holomorphic
analog of the Berezinian line bundle, which we introduced in the smooth case in section
15
If Xcs is a smooth supermanifold associated to X as in section 5.2 (the following definition does not
b
depend on the precise construction of Xcs ) with local coordinates u1α . . . | . . . θα , then one requires that at
b
1
least one of the maximum rank minors of the matrix of derivatives ∂(uα . . . | . . . θα )/∂(t1τ . . . | . . . ητs ) should
have nonzero Berezinian. This echoes the condition for a smooth map of ordinary manifolds to be a local
embedding.

– 36 –
3.1. The cotangent bundle of X in the holomorphic sense16 is a holomorphic bundle T ∗ X
of rank a|b. We define a holomorphic version of the Berezinian of X, which we will call
Ber (X), by imitating the definition used for smooth supermanifolds. To any local system
T = z 1 . . . | . . . θ b of holomorphic coordinates on M , we associate a local holomorphic section
of Ber (X) that we denote [dz 1 . . . | . . . dθ b ]. If Te = ze1 . . . | . . . θeb is a second local holomorphic
coordinate system, then we relate the two sections of Ber (X) by the formula (3.10).
Now we want to ask in what sense a holomorphic section σ of Ber (X) can be integrated.
For this, one approach is to view X as a smooth supermanifold Xcs of dimension 2a|b.
From that point of view, σ corresponds to an integral form on Xcs of codimension a, which
moreover is closed, dσ = 0. The map from a section of Ber (X) to an integral form takes
[dz 1 . . . | . . . dθ b ] to the integral form dz 1 . . . dz a δ(dθ 1 ) . . . δ(dθ b ). This is of codimension a
in the real sense as dze1 . . . dzea are missing.
If Nred ⊂ Xred is of middle dimension, then the corresponding cycle N ⊂ X is of
real codimension a|0. So (given an appropriate orientation condition), there is a natural
R
integral N σ. This only depends on the homology class of Nred .
So σ ∈ H 0 (X, Ber (X)) defines a linear function on the middle-dimensional homol-
ogy of Xred . By ordinary topology, this means that σ determines a class [σ] in the
middle-dimensional complex-valued cohomology H a (Xred , C). For example, if X is a su-
per Riemann surface, then a holomorphic section of Ber (X) has periods associated to
ordinary A-cycles and B-cycles in the ordinary Riemann surface Xred , and defines a class
in H 1 (Xred , C). This enables one to define the super period matrix of a super Riemann
surface, though it is tricky to show that the super period matrix is symmetric.

5.4 Alternative Point Of View

Having completed the preliminaries of section 5.3, we will now generalize viewpoint (2) of
section 5.1 to supermanifolds.
First let us spell out more precisely what viewpoint (2) means, again taking Riemann
surfaces as an example. For Σ a Riemann surface, let ΣR be a copy of Σ and let ΣL be a
copy of Σ with opposite complex structure. Then ΣL ×ΣR is a complex manifold of complex
dimension 2, and Σ is naturally embedded in ΣL × ΣR as the diagonal. If we view Σ is a
real analytic manifold of dimension 2, then we can regard ΣL × ΣR as a complexification
of Σ.
In the last paragraph, we do not literally need to take Σ to be the diagonal in ΣL × ΣR .
Σ can be any real-analytic submanifold of ΣL × ΣR that is sufficiently close to the diagonal.
ΣL × ΣR can be understood as a complexification of any such Σ. One can study conformal
field theory on any such Σ and its content is essentially independent of Σ since the correla-
tion functions can anyway be analytically continued to holomorphic functions on an open
set in ΣL ×ΣR . Restricted to any such Σ, the correlation functions hΦ1 (ez1 ; z1 ) . . . Φs (e
zs ; zs )i
have only the usual CFT singularities when pairs of points coincide.
By a “holomorphic function on Σ,” we mean the restriction to Σ of a holomorphic
function on ΣR , and similarly an “antiholomorphic function on Σ” is the restriction to Σ
16
It seems that, just as X does not have a completely natural structure of smooth supermanifold, it does
not have a completely natural tangent or cotangent bundle except in the holomorphic sense.

– 37 –
of an antiholomorphic function on ΣR . Unless Σ is actually the diagonal in ΣL × ΣR , it is
not true that a antiholomorphic function on Σ is the complex conjugate of a holomorphic
function. This is analogous to the situation encountered in section 5.2 when we associated a
smooth supermanifold Xcs to a complex supermanifold X: the antiholomorphic coordinate
ze is not quite the complex conjugate of z; it has this interpretation only modulo the odd
variables.
We can now reinterpret the object Xcs defined in section 5.2. Let XR be a copy of X
and let XL be X red , that is, XL is the reduced space Xred with complex structure reversed.
The formula (5.8) enabled us to define a smooth supermanifold Xcs with local coordinates
t1α . . . t2a 1 b i
α |θα . . . θα . However, bearing in mind that the odd coordinates θα of XR are the
same as the odd coordinates of Xcs , we can read the first line of eqn. (5.8) as defining
a continuous map from Xcs to XR . And the second line of eqn. (5.8) similarly defines a
continuous map from Xcs to XL . Altogether what we have is a smooth supermanifold Xcs
with an embedding Xcs ֒→ XL × XR .
The reduced space of Xcs , moreover, is the diagonal in the reduced space of XL × XR .
And we can interpret XL × XR as the complexification of Xcs . (This complexification
is defined by interpreting all local coordinates t1α . . . t2a 1 b
α |θα . . . θα of Xcs as independent
complex variables. From eqn. (5.8), this just means that zαi and zeαi are independent
complex variables, as is appropriate for defining XL × XR .)
In short, the relation between Xcs and XL × XR is very similar to the relation between
Σ and ΣL × ΣR . The main difference is that in the bosonic case, there is a completely
natural choice of Σ (namely the diagonal in ΣL × ΣR ), but for X a complex supermanifold,
it does not seem that there is a completely canonical choice of Xcs . We can simply take
Xcs to be any subsupermanifold of XL × XR of codimension 2a|0 whose reduced space is
sufficiently close to the diagonal in (XL × XR )red .
If we wish, we can take the reduced space of Xcs to be precisely the diagonal in
(XL × XR )red , as we did in section 5.2 in the original definition of Xcs . But even then, it
does not seem that there is a completely canonical choice for Xcs itself.

5.5 Application To String Theory: The Worldsheet

The most prominent supermanifolds in superstring perturbation theory are the string
worldsheet and the supermoduli space over which one must integrate to compute scat-
tering amplitudes. Let us consider both from the viewpoint just proposed.
First we consider the string worldsheet, starting with the heterotic string. We let
ΣR be a super Riemann surface; this is a complex supermanifold of dimension 1|1 that
obeys some additional conditions that are described, for example, in [2]. These additional
conditions need not concern us here. We let ΣL be an ordinary Riemann surface that is
sufficiently close to the complex conjugate of the reduced space of ΣL . So ΣL × ΣR is
a complex supermanifold of dimension 2|1. To get a candidate for the worldsheet of a
heterotic string, we let Σ be a smooth submanifold of ΣL × ΣR of dimension 2|1 whose
reduced space is sufficiently close to the diagonal in (ΣL × ΣR )red . For example, we may
start by picking Σred ⊂ (ΣL × ΣR )red to be any cycle sufficiently close to the diagonal and
then “thicken” it slightly in the fermionic directions (as explained in section 2.2) to get Σ.

– 38 –
 If we wish, we can assume that ΣL is precisely the complex conjugate of ΣR,red and
take Σred to be the diagonal in (ΣL × ΣR )red . If we do so, then Σ can be ΣR,cs , the cs
version of ΣR as defined (with some choice of gluing law) in section 5.2. But there is no
reason to limit ourselves to precisely this case. At any rate, even if one wishes to restrict
to the case that Σ is ΣR,cs , the present approach makes it more clear how natural ΣR,cs is
or is not.
The main thing we need to know to make sure that this approach to heterotic string
theory makes sense is that given any Σ ⊂ ΣL × ΣR as above, the worldsheet action of
the heterotic string can be defined as an integral over Σ, and moreover this integral does
not depend on the choice of Σ. The relevant facts are explained in [2]. The Lagrangian
density of the heterotic string will be a holomorphic section of the holomorphic Berezinian
Ber (ΣL × ΣR ) (defined in a suitable neighborhood of the diagonal in the reduced space)
and as explained in section 5.3.1, this can be integrated over the real cycle Σ, with a result
that only depends on the homology class of Σ. We do not have any way to pick a canonical
Σ, but any choice of Σ will lead to the same integrated worldsheet action.
For Type II superstring theory, the basic idea is the same. The holomorphic and
antiholomorphic dimensions of the string worldsheet will now both be 1|1, but all the con-
siderations that we have described are still relevant since antiholomorphic odd coordinates
are not supposed to be in any sense complex conjugates of holomorphic odd coordinates.
For Type II superstring theory, we let both ΣL and ΣR be super Riemann surfaces, such
that the reduced space of ΣL is sufficiently close to the complex conjugate of the reduced
space of ΣR . (No relationship is assumed between the spin structures of ΣL and ΣR .)
ΣL × ΣR is a complex supermanifold of dimension 2|2. Now we let Σ be any smooth sub-
supermanifold of ΣL × ΣR of dimension 2|2 whose reduced space is sufficiently close to the
diagonal. Just as for the heterotic string, we may start by picking Σred ⊂ (ΣL ×ΣR )red to be
any cycle sufficiently close to the diagonal and then “thicken” it slightly in the fermionic
directions to get Σ. Again, the worldsheet action can be defined by integrating over Σ
a holomorphic section of Ber (ΣL × ΣR ) (defined sufficiently close to the diagonal in the
reduced space) and the integrated action does not depend on the precise choice of Σ.
Whenever we say that two objects are “sufficiently close,” one may interpret this to
mean that the objects in question are equal if the odd variables are set to zero. (For
reduced spaces, this means simply that they are equal.) That will always be sufficiently
close for any purpose. But in any concrete case, the two objects need not be quite so close
as that, and it is sometimes better to allow oneself a little more elbow room.

5.6 Application To String Theory: Supermoduli Space

The other supermanifold that plays a prominent role in superstring perturbation theory is
the moduli space over which one integrates in computing scattering amplitudes.
In each of the above constructions, let ML be the moduli space parametrizing the
possible choices of ΣL , and let MR be the moduli space parametrizing ΣR . Concretely,
for either the heterotic string or Type II, MR is the moduli space M of super Riemann
surfaces and its reduced space is what we will call Mspin , which parametrizes an ordinary
Riemann surface with a choice of spin structure. As for ML , for the heterotic string, it

– 39 –
is what we will call M, the moduli space of ordinary Riemann surfaces; for Type II, it is
again the moduli space M of super Riemann surfaces. The reduced space of ML × MR is
therefore M × Mspin for the heterotic string, or Mspin × Mspin for Type II.
If we simply let ΣL and ΣR vary independently, the moduli space parametrizing ΣL ×
ΣR is the product ML × MR . However, in string theory, we do not let ΣL and ΣR vary
independently. Roughly speaking, the bosonic moduli of ΣL are supposed to be complex
conjugates of the bosonic moduli of ΣR , though we want no relation between the odd moduli
of ΣL and those of ΣR . To implement this, we proceed as follows. In the reduced space
(ML ×MR )red , we define a submanifold Γred by requiring that the complex structures of the
ordinary Riemann surfaces parametrized by ML,red and MR,red are complex conjugates
(one assumes no relationship between the spin structures). Then in the usual way, we
thicken Γred slightly to a subsupermanifold Γ ⊂ ML × MR of the same codimension.17
Γ has the necessary properties to be the integration cycle of superstring perturbation
theory. The worldsheet path integral determines a holomorphic section (defined in a neigh-
borhood of Γred ) of the holomorphic Berezinian Ber (ML × MR ). For the usual reasons,
subject to a caveat that we explain momentarily, the integral of this section over Γ does
not depend on the choice of Γ .
The reason that a caveat is needed is that ML , MR , and Γ are all not compact. The
integrals that one encounters in superstring perturbation theory have a delicate behavior at
infinity. Dealing with these integrals requires, among other things, some care in specifying
how Γ should behave near infinity. The region at infinity that causes the subtleties is the
infrared region, and the subtleties go into showing that superstring perturbation theory
has the expected infrared behavior.
We can summarize all this as follows. There are completely natural moduli spaces
ML and MR for antiholomorphic and holomorphic variables. It seems doubtful that there
is a natural moduli space that combines the two types of variable, but up to homology
there is a natural integration cycle Γ ⊂ ML × MR , which is what one actually needs for
superstring perturbation theory.

Acknowledgments Research supported in part by NSF grant PHY-0969448. I thank

P. Deligne for much valuable advice over the years on matters related to supergeometry.
Helpful comments on the manuscript have been given by G. Moore, T. Voronov, P. Zhao,
and especially D. Robbins and B. Safdi.

References

[1] E. Witten, “Superstring Perturbation Theory Revisited,” arXiv:1209.5461.

[2] E. Witten, “Notes On Super Riemann Surfaces And Their Moduli,” arXiv:1209.2459.
[3] F. A. Berezin, Introduction To Superanalysis, A. A. Kirillov, ed. (D. Reidel, 1987).
17
This definition has been given in [11], p. 95. As usual, we could replace Γred by any cycle sufficiently
close to the one stated.

– 40 –
 [4] Yu. I. Manin, Gauge Field Theory And Complex Geometry (Springer-Verlag, 1988).
[5] B. DeWitt, Supermanifolds, second edition (Cambridge University Press, 1992).
[6] A. Rogers, Supermanifolds: Theory And Applications (World-Scientific, 2007).
[7] V. Kac, “Lie Superalgebras,” Adv. Math. 26 (1977) 8-96.
[8] D. Leites, “Introduction To The Theory Of Supermanifolds,” Russian Math Surveys 35
(1980) 1-64.
[9] T. Voronov, “Geometric Integration Theory On Supermanifolds,” Soviet Scientific Reviews,
Section C: Mathematical Physics, vol. 9, part 1.
[10] P. C. Nelson, “Lectures On Supermanifolds And Strings,” in Particles, Strings, and
Supernovae, eds. A. Jevicki and C. I. Tan (World Scientific, 1989).
[11] P. Deligne and J. W. Morgan, “Notes On Supersymmetry (following Joseph Bernstein),” in
P. Deligne et. al., eds., Quantum Fields And Strings: A Course For Mathematicians, Vol. 1
(American Mathematical Society, 1999).
[12] F. A. Berezin, “Supermanifolds,” ITEP preprint (1979).
[13] M. Bachelor, “The Structure Of Supermanifolds,” Trans. Am. Math. Soc. 253 (1979)
329-338.
[14] K. Gawedzki, “Supersymmetries – Mathematics Of Supergeometry,” Ann. Inst. H. Poincaré,
Sect. A 27 (1977) 335-366.
[15] J. Bernstein and D. A. Leites, “Integral Forms And The Stokes Formula On
Supermanifolds,” Funct. Anal. Appl. 11 (1977) 55-6.
[16] M. A. Baranov and A. S. Schwarz, “Cohomology Of Supermanifolds,” Functional Analysis
and its Applications 18 (1984) 236-238.
[17] T. Voronov and A. V. Zorich, “Integral Transformations Of Pseudodifferential Forms,” Usp.
Mat. Nauk. 41 (1986) 167-8.
[18] T. Voronov and A. V. Zorich, “Cohomologies Of Supermanifolds And Integral Geometry,”
Doklady Akad. Nauk SSR 298 (1988) 528-33.
[19] P. A. Grassi and M. Marescotti, “Integration Of Superforms And Super-Thom Class,”
arXiv:0712.2600.
[20] D. Friedan, E. Martinec, and S. Shenker, “Covariant Quantization Of Superstrings,” Phys.
Lett. B160 (1985) 55, “Conformal Invariance, Supersymmetry, and String Theory,” Nucl.
Phys. B271 (1986) 93.
[21] E. D’Hoker and D. H. Phong, “The Geometry Of String Perturbation Theory,” Rev. Mod.
Phys. 60 (1988) 917-1065.
[22] G. Moore, P. Nelson, and J. Polchinski, “Strings And Supermoduli,” Phys. Lett. 169B
(1986) 47-53.
[23] L. Alvarez-Gaumé, C. Gomez, P. C. Nelson, G. Sierra, and C. Vafa, “Fermionic Strings In
The Operator Formalism,” Nucl. Phys. B311 (1988) 333.
[24] E. Verlinde and H. Verlinde, “Multiloop Calculations In Covariant Superstring Theory,”
Phys. Lett. B192 (1987) 95.
[25] A. A. Rosly, A. S. Schwarz, and A. A. Voronov, “Superconformal Geometry And String

– 41 –
 Theory,” Commun. Math. Phys. 120 (1989) 437-450.
[26] A. Belopolsky, “De Rham Cohomology Of The Supermanifolds And Superstring BRST
Cohomology,” Phys. Lett. B403 (1997), hep-th/9609220.
[27] A. Belopolsky, “New Geometrical Approach To Superstrings,” hep-th/9703183.
[28] A. Belopolsky, “Picture Changing Operators In Supergeometry And Superstring Theory,”
hep-th/9706033.
[29] E. D’Hoker and D. H. Phong, “Lectures On Two-Loop Superstrings,” Adv. Lect. Math. 1
85-123, hep-th/0211111, and references therein.
[30] R. Donagi and E. Witten, “Supermoduli Space Is Not Projected,” Proc. Symp. Pure Math.
90 (2015) 19-72, arXiv:1304.7798.
[31] A. Neveu, J. H. Schwarz, and C. B. Thorn, “Reformulation Of The Dual Pion Model,” Phys.
Lett. 35B (1971) 529.
[32] L. Brink, D. Olive, C. Rebbi, and J. Scherk, “The Missing Gauge Conditions For The Dual
Fermion Emission Vertex And Their Consequences,” Phys. Lett. 45B (1973) 379.

– 42 –