The Mathematics of M-Theory

Robbert Dijkgraaf

Abstract. String theory, ot its modern incarnation M-theory, gives a huge

generalization of classical geometry. I indicate how it can be considered as a
two-parameter deformation, where one parameter controls the generalization
from points to loops, and the other parameter controls the sum over topologies
of Riemann surfaces. The final mathematical formulation of M-theory will
have to make contact with the theory of vector bundles, K-theory and non-
commutative geometry.

1. Introduction
Over the years there have been many fruitful interactions between string the-
ory [14] and various fields of mathematics. Subjects like algebraic geometry and
representation theory have been stimulated by new concepts such as mirror sym-
metry [3], quantum cohomology [12] and conformal field theory [4]. But most of
these developments have been based on the perturbative formulation of string the-
ory, either in the Lagrangian formalism in terms of maps of Riemann surfaces
into manifolds and the quantization of loop spaces. This perturbative approach
is however only an approximate description that appears for small values of the
quantization parameter.
Recently there has been much progress in understanding a more fundamental
description of the theory that has become known as M-theory. M-theory seem to
be the most complex and richest mathematical object so far in physics. It seems
to unify three great ideas of twentieth century theoretical physics:
(1) General relativity – the idea that gravity can be described by the Rie-
mannian geometry of space-time.
(2) Gauge theory – the description of forces between elementary particles us-
ing connections on vector bundles. In mathematics this involves K-theory
and index theorems.
(3) Strings, or more generally extended objects, as a natural generalization of
point particles. Mathematically this means that we study spaces primarily
through their (quantized) loop spaces.
At present it seems that these three independent ideas are closely related,
and perhaps essentially equivalent. To some extend physics is trying to build a
dictionary between geometry, gauge theory and strings.
2 R. Dijkgraaf

It must be said that in all developments there have been two further ingredi-
ents that are absolutely crucial. The first is quantum mechanics —the description
of physical reality in terms of operator algebras acting on Hilbert spaces. In most
attempts to understand string theory quantum mechanics has been the foundation,
and there is little indication that this is going to change.
The second ingredient is supersymmetry —the unification of matter and
forces. In mathematical terms supersymmetry is closely related to de Rham com-
plexes and algebraic topology. In some way much of the miraculous interconnec-
tions in string theory only work if supersymmetry is present. Since we are essen-
tially working with a complex, it should not come to a surprise to mathematicians
that there are various ‘topological’ indices that are stable under perturbation and
can be computed exactly in an appropriate limit. From a physical perspective
supersymmetry is perhaps the most robust prediction of string theory.

1.1. A two-parameter deformation of classical geometry

For pedagogical purposes in this lecture M-theory will be considered as a two para-
meter family of deformations of “classical” Riemannian geometry. Let us introduce
these two parameters heuristically. (We will give a more precise definition later.)
First, in perturbative string theory we study the loops in a space-time man-
ifold. These loops can be thought to have an intrinsic length
s , the string length.
At least at an heuristic level it is clear that in the limit
s → 0 the string degen-
erates to a point, a constant loop. The parameter
s controls the “stringyness” of
the model. We will see how the quantity
2s = α
plays the role of Planck’s con-
stant on the worldsheet of the string. That is, it controls the quantum correction
of the two-dimensional field theory on the world-sheet of the string. An important
example of a stringy deformation is quantum cohomology [12].
Secondly, strings can split and join, sweeping out a surface Σ of general
topology in space-time. According to the general rules of quantum mechanics we
have to include a sum over all topologies. Such a sum over topologies can be
regulated if We can introduce a formal parameter λ ∈ R+ , the string coupling,
such that a surface of genus g gets weighted by a factor λ2g−2 . Higher genus
topologies can be interpreted as virtual processes wherein strings split and join
—a typical quantum phenomenon. Therefore the parameter λ controls the quan-
tum corrections. In fact we can equate λ2 with Planck’s constant in space-time.
Only for small values of λ can string theory be described in terms of loop spaces
and sums over surfaces.
In fact, in the case of particles we know that for large values of λ it is better
to think in terms of waves, or more precisely quantum fields. So we expect that
for large λ and α
the right framework is string field theory [21]. This is partly
true, but it is in general difficult to analyze this string field theory directly. In
particular the occurrence of branes, higher-dimensional extended objects that will
play an important role in the subsequent, is often obscure. (See however the recent
work [17].)
 The Mathematics of M-Theory 3

Summarizing we can distinguish two kinds of deformations: stringy effects

parametrized by α
, and quantum effects parametrized by λ. This situation can be
described with the following table

α
> 0 conformal field theory M-theory

strings string fields, branes
α
= 0 quantum mechanics quantum field theory
particles fields

λ=0 λ>0
It is perhaps worthwhile to put some related mathematical fields in a similar
table
α
> 0 quantum cohomology non-commutative geometry
(Gromov, Witten) (Connes)
α
= 0 combinatorical knot invariants 4-manifold, 3-manifolds, knots
(Vassiliev, Kontsevich) (Donaldson, Witten, Jones)

λ=0 λ>0
We will now briefly review these various generalizations. More background
material can be found in [5].

2. Quantum Mechanics and Particles

In classical mechanics we describe point particles on a Riemannian manifold X
that we think of as a (Euclidean) space-time. Pedantically speaking we look at X
through maps
x : pt → X
of an abstract point into X. Quantum mechanics associates to the classical config-
uration space X the Hilbert space H = L2 (X) of square-integrable wavefunctions.
We want to think of this Hilbert space as associated to a point
H = Hpt .
For a supersymmetric point particle we instead work with the space of de Rham
differential forms H = Ω∗ (X).
Classically a particle can go in a time t from point x to point y along some
preferred path, typically a geodesic. Quantum mechanically we instead have a
linear evolution operator
Φt : H → H ,
4 R. Dijkgraaf

that describes the time evolution. Through the Feynman path-integral this oper-
ator is associated to maps of the line interval of length t into X. More precisely,
the kernel Φt (x, y) of the operator Φt is given by the path-integral
 t
[dx]e− 0 dτ |ẋ|
2
Φt (x, y) =
x(τ )

over all paths x(τ ) with x(0) = x and x(t) = y. Φt is the kernel of the heat equation
d
Φt = ∆Φt , Φ0 = δ(x − y) .
dt
These path-integrals have a natural gluing property: if we first evolve over a
time t1 and then over a time t2 this should be equivalent to evolving over time t1 +
t2 .
Φt1 ◦ Φt2 = Φt1 +t2 . (1)
This allows us to write
Φt = e−tH
with H the Hamiltonian. In the case of a particle on X the Hamiltonian is of
course simply given by the Laplacian H = −∆.
The composition property (1) is a general property of quantum field theo-
ries. It leads us to Segal’s functorial view of quantum field theory, as a functor
between the categories of manifolds (with bordisms) to vector spaces (with linear
maps) [15].
The Hamiltonian can be written as
H = −∆ = −(dd∗ + d∗ d) .
Here the differentials d, d∗ play the role of the supercharges. Ground states satisfy
Hψ = 0 and are therefore harmonic forms and in 1-to-1 correspondence with the
de Rham cohomology group
ψ ∈ Harm∗ (X) ∼ = H ∗ (X) .
We want to make two additional remarks. First we can consider also a closed
1-manifold, namely a circle S 1 of length t. Since a circle is obtained by identifying
two ends of an interval we can write
ΦS 1 = TrH Φt .
Here the partition function ΦS 1 is a number associated to the circle S 1 that encodes
the spectrum of ∆. We can also compute the supersymmetric partition function by
using the fermion number F (defined as the degree of the corresponding differential
form). It computes the Euler number
TrH (−1)F Φt = χ(X) .
Secondly, to make the step from the quantum mechanics to the propagation
of a particle in quantum field theory we have to integrate over the metric on the
 The Mathematics of M-Theory 5

1-manifold. In case of an interval we so obtain the usual propagator, the Greens’

function of the Laplacian
 ∞
1
dt et∆ = .
0 ∆

3. Conformal Field Theory and Strings

We will now introduce our first deformation parameter α
and generalize from
point particles and quantum mechanics to strings and conformal field theory.

3.1. Sigma models

A string can be considered as a parametrized loop. So, in this case we study the
manifold X through maps
x : S1 → X
that is, through the free loop space LX.
Quantization will associate a Hilbert space to this loop space. Roughly one
can think of this Hilbert space as L2 (LX), but it is better to think of it as a
quantization of an infinitesimal thickening of the locus of constant loops X ⊂ LX.
These constant loops are the fixed points under the obvious S 1 action on the loop
space. The normal bundle to X in LX decomposes into eigenspaces under this S 1
action, and this gives a description (valid for large volume of X) of the Hilbert
space HS 1 associated to the circle as the normalizable sections of an infinite Fock
space bundle over X.
HS 1 = L2 (X, F+ ⊗ F− )
where the Fock bundle is defined as

F= Sqn (T X) = C ⊕ qT X ⊕ · · ·
n≥1

Here we use the formal variable q to indicate the Z-grading of F and we use the
standard notation

Sq V = qN S N V
N ≥0

for the generating function of symmetric products of a vector space V .

When a string moves in time it sweeps out a surface Σ. For a free string Σ
has the topology of S 1 × I, but we can also consider at no extra cost interacting
strings that join and split. In that case Σ will be a oriented surface of arbitrary
topology. So in the Lagrangian formalism one is let to consider maps
x: Σ → X .
6 R. Dijkgraaf

There is a natural action for such a sigma model if we pick a Hogde star or
conformal structure on Σ (together with of course a Riemannian metric g on X)

S(x) = gµν dxµ ∧ ∗dxν .
Σ

The critical points of S(x) are the harmonic maps.

In the Lagrangian quantization formalism one considers the formal path-
integral over all maps x : Σ → X



ΦΣ = e−S/α .
x : Σ→X

Here the constant α
plays the role of Planck’s constant on the string worldsheet Σ.
It can be absorbed in the volume of the target X by rescaling the metric as g → α
·
g. The semi-classical limit α
→ 0 is therefore equivalent to the limit vol(X) → ∞.

3.2. Functorial description

In the functorial description of conformal field theory the maps ΦΣ are abstracted
away from the sigma model definition.
Starting point is now an arbitrary (closed, oriented) Riemann surface Σ
with boundary. This boundary consists of a collections of oriented circles. One
declares these circles in-coming or out-going depending on whether there orien-
tation matches that of the Σ. To a surface Σ with m in-coming and n out-going
boundaries one associates a linear map
ΦΣ : HS⊗n ⊗n
1 → HS 1 .

These maps are not independent but satisfy gluing axioms that generalize the
simple composition law (1)
ΦΣ1 ◦ ΦΣ2 = ΦΣ
where Σ is obtained by gluing Σ1 and Σ2 on their out-going and incoming bound-
aries respectively.
In this way we obtain what is known as a modular functor. It has a rich
algebraic structure. For instance, the sphere with three holes gives rise to a product
Φ : HS 1 ⊗ HS 1 → HS 1 .
Using the fact that a sphere with four holes can be glued together from two copies
of the three-holed sphere one shows that this product is essentially commutative
and associative. For more details see e.g. [12, 5].

3.3. B-fields and gerbes

There is a straightforward generalization of the sigma model action (3.1) that
includes a 2-form B on X. This so-called B-field adds an extra phase

exp i x∗ B
Σ
 The Mathematics of M-Theory 7

in the path-integral. The two-form B should be considered as a two-form analogue

of a connection. Its curvature H which is locally given by H = dB actually rep-
resents a class [H/2π] ∈ H 3 (X, Z). So more generally, if Σ is considered as the
boundary of a 3-manifold W and the map x is extended over W , then the phase
factor is defined as

exp i x∗ H .
W

By pulling back B to the loop space we see that it acts as a connection on a line
bundle over LX.
Such a B-field can be considered as a connection on a gerbe [9] over the man-
ifold X. Gerbes are generalizations of line bundles that naturally support p-form
connections. Extended objects such as strings and branes are closely connected to
these p-form theories.

3.4. Strings and gravity

The way in which general relativity emerges from string theory is deep and I want
to use some time to explain this here. We have already seen that at the classical
level every two-dimensional sigma model is conformal invariant. In order to write
down the classical action and the resulting equation of motion (the harmonic maps
condition) for maps Σ → X we only need the Hodge star or a conformal structure
on Σ. This is no longer true at the quantum level. Heuristically, we need a full
metric γ on Σ to define the measure in the path-integral over all maps Σ → X,
and this metric dependence of the path-integral distroys the conformal invariance
of the quantum theory.
So the partition function is actually a function of both the metric g on X
and γ on Σ. We can now consider a (constant) rescaling
γ →µ·γ
of the worldsheet metric and compute the µ-dependence of the partition func-
tion Φ(g, γ). In general one finds a non-zero result.
The amazing result of standard renormalization theory in quantum field the-
ory is that a rescaling in γ can be absorbed into a redefinition of the metric g on
X. This leads to the famous renormalization group flow equation [14]
  
∂ ∂
µ + βµν Z = 0.
∂µ X ∂gµν
Here the beta-function βµν is a vector field on the space M ET (X) of Riemannian
metrics on X.
This effect can be observed most easily in a semi-classical expansion of the
sigma model in α
. That is, one consider maps that are approximately constant
x = x0 + ξ
8 R. Dijkgraaf

with ξ : Σ → Tx0 X. In this approximation the action can be written to subleading

order as (schematically)

gµν (x0 )dξ µ ∧ ∗dξ ν + Rµνλρ (x0 )ξ µ ξ ν dξ λ ∧ ∗dξ ρ .

Here the Riemann curvature Rµνλρ can be considered as a small perturbation in

the large volume limit. To leading order one then finds that
βµν = α
Rµν + O(α
) .
2

So to this level of approximations the Ricci flat metrics with

Rµν = 0
correspond to the quantum CFTs.
An important property of the beta-function β is that it is a gradient vector
field on M ET (X)

√
β = ∇S, S= gR(g)

with S the Einstein-Hilbert action for the metric g on X. Conformal field theories
are given by the zeroes of β and thus correspond to the critical points of S. In this
sense a quantum CFT corresponds to a semi-classical solution of gravity.

3.5. ‘Stringy’ geometry and T-duality

Two-dimensional sigma models give a natural one-parameter deformation of classi-
cal geometry. The deformation parameter is Planck’s constant α
. In the limit α
→
0 we localize on constant loops and recover quantum mechanics or point particle
theory. For non-zero α
the non-constant loops contribute.
This structure is most familiar now in the form of quantum cohomology
for Kähler manifolds. Under suitable circumstances the path-integral localizes to
holomorphic maps that get weighted by a phase factor q d where d is the degree
and q = et with t the pull-back of the complexified Kähler form αω
+ iB. So the
CFT maps ΦΣ have a typical expansion [3]

ΦΣ = q d NΣ (d)
d

where NΣ (d) counts the number of holomorphic maps Σ → X. Since essentially

q ∼ e−1/α we see that these corrections are non-perturbative from a world-sheet
point of view. There are invisible in a Taylor expansion in α
.
In fact we can picture the moduli space of CFT’s roughly as follows. It will
have components that can be described in terms of a target spaces X. For these
models the moduli parametrize Ricci-flat metrics plus a choice of B-field. These
components have a boundary ‘at infinity’ which describe the large volume man-
ifolds. We can use the parameter α
as local transverse coordinate on the collar
around this boundary. If we move away from this boundary stringy corrections set
in.
 The Mathematics of M-Theory 9

In the middle of the moduli space exotic phenomena can take place. For ex-
ample, the automorphism group of the CFT can jump, which gives rise to orbifold
singularities at enhanced symmetry points.
The most striking phenomena that the moduli space can have another bound-
ary that allows again for a semi-classical interpretation in terms of a second clas-
sical geometry X̂. These points look like quantum or small volume in terms of the
original variables on X but can also be interpreted as large volume in terms of a
set of dual variables on a dual or mirror manifold X̂. In this case we speak of a
T-duality. In this way two manifold X and X̂ are related since they give rise to
the same CFT.
The most simple example of such a T-duality occurs for toroidal compactifi-
cation. If X = T is an torus, the CFT’s on T and its dual T ∗ are isomorphic. We
will explain this in more detail in §5.

4. M-Theory and Branes

We have seen how CFT gives rise to a rich structure in terms of the modular
geometry as formulated in terms of the maps ΦΣ . To go from CFT to string
theory we have to make two more steps.

4.1. Summing over string topologies

First, we want to generalize to the situation where the maps ΦΣ are not just func-
tions on the moduli space Mg,n of Riemann surfaces but more general differential
forms. In fact, we are particular interested in the case where they are volume forms
since then we can define the so-called string amplitudes as

Ag = ΦΣ .
Mg

This is also the general definition of Gromov-Witten invariants [12].

Although we suppress the depends on the CFT moduli, we should realize
that the amplitudes Ag (now associated to a topological surface of genus g) still
have (among others) α
dependence.
Secondly, it is not enough to consider a string amplitude of a given topology.
Just as in field theory one sums over all possible Feynman graphs, in string theory
we have to sum over all topologies of the string world-sheet. In fact, we have to
ensemble these amplitudes into a generating function.

A(λ) ≈ λ2g−2 Ag .
g≥0

Here we introduce the so-called string coupling constant λ. Unfortunately this

generating function can be at best an asymptotic series expansion of an analytical
function A(λ). A rough estimate of the volume of Mg shows that typically
Ag ∼ 2g!
10 R. Dijkgraaf

Indeed, general physics arguments tell us that the non-perturbative amplitu-

des A(λ) have corrections of the form

A(λ) = λ2g−2 Ag + O(e−1/λ ) .
g≥0

Clearly to approach the proper definition of the string amplitudes these non-
perturbative corrections have to be understood.

4.2. M-theory
The last five years have seen remarkable progress in this direction. It involves two
remarkable new ideas.
1. String theory is not a theory of strings. It is not enough to consider loop
spaces. We should also include other extended objects, collectively known
as branes. One can try to think of these objects as associated to more
general maps Y → X where Y is a higher-dimensional space. But the
problem is that there is not a consistent quantization starting from ‘small’
branes along the lines of string theory, that is, an expansion where we
control the size of Y (through α
) and the topology (through λ). However,
through the formalism of D-branes [13] these can be analyzed exactly in
string perturbation theory.
2. As we stressed, the amplitudes A depend on many parameters or mod-
uli. Apart from the string coupling λ all other moduli have a geometric
interpretation, in terms of the metric and B-field on X. The second new
ingredient is the insight that string theory on X with string coupling λ
can be given a fully geometric realization in terms of a new theory (called
M-theory) on the manifold X × S 1 , where the length of the circle S 1 is
λ [18].

5. Torus Compactifications
So we see that there is a natural hierarchy of generalized geometries, roughly
associated to particles, strings and branes. According this point of view a ‘classical’
manifold can be considered as an element of three different categories. Viewed as
an object in such a category it can inherit different symmetries.
Although we are not yet in a position to give a completely rigorous definition
of what M-theory is, we do know what kind of data we want to associate to a
space X. These data contain at least the following
1. A moduli space M of geometric structures on X. This can be a Ricci-flat
metric, but also B-fields or their generalizations.
2. A charge lattice Γ that labels the various sectors of the theory.
3. A discrete symmetry group G (the duality group) that acts on the lattice Γ,
which will typically form an irreducible representation.
 The Mathematics of M-Theory 11

4. A Hilbert space bundle


H= H(γ)
γ∈Γ

over the moduli space M that is graded by the charge lattice Γ.

This hierarchy of structures can be nicely illustrated with a very simple class
of manifolds: n-tori, written as quotients
T = Rn /L
with L a rank n lattice.

5.1. Particles on a torus

States of a quantum mechanical point particle on T are conveniently labeled by
their momentum
p ∈ L∗ .
The wavefunctions eipx form a basis of H = L2 (T ) that diagonalizes the Hamil-
tonian H = −∆ = p2 . So we can write

H= H(p) ,
p∈L∗

where the graded pieces H(p) are one-dimensional. There is a natural action of
the symmetry group
G = SL(n, Z) = Aut L
on the lattice Γ = L and the Hilbert space H. (These transformations will in
general not leave the metric invariant, but instead give by pull-back another flat
metric on T .)

5.2. Strings on a torus

In the case of a string moving on the torus T states are also labeled by their
winding number
w ∈ L.
The winding number simply distinguishes the various connected components of
the loop space LT , since
π0 LT = π1 T ∼
= L.
We therefore see a natural occurrence of the so-called Narain lattice
Γn,n = L ⊕ L∗ .
This is an even self-dual lattice of signature (n, n) with inner product
p = (w, k), q 2 = 2w · k .
12 R. Dijkgraaf

It turns out that the symmetries of the lattice Γn,n lift to symmetries of the full
conformal field theory. The symmetry group
SO(n, n, Z) = Aut Γn,n
are called T -dualities. A particular example is the interchange T ↔ T ∗ . From
this perspective the string can be considered to be moving on the space T × T ∗ .
If we consider chiral or BPS states in the superstring, then the graded Hilbert
space H(q) is given by
H(p) = F( 12 p2 ) ,
in terms of the Fock space
∞
 
Fq = Sqn (R8 ) ⊗ q n (R
8
)= q N F(N ) .
n=1 N ≥0

Note that the Dynkin diagram of the corresponding lie algebra Dn ∼

= so(n, n)
is obtained from An−1 ∼ = sl(n) by adding an extra root. Reflections in this root
represent the T-duality that maps T to T ∗ .

5.3. Branes on a torus

If we move to the full M-theory the charge lattice becomes more complicated. For
small values of n (n ≥ 4) it can be written as
Γn,n ⊕ H even/odd (T ) .
Here we note that the lattice of branes (which are even or odd depending on the
type of string theory that we consider)
H even/odd (T ) ∼ L∗
even/odd
=
transform as half-spinor representations under the T-duality group SO(n, n, Z).
The full duality group turns out to be the exceptional group over the integers
En+1 (Z) .
So we see that our hierarchy is reflected in the symmetry groups
SL(n, Z) ⊂ SO(n, nZ) ⊂ En+1 (Z)
of rank n − 1, n and n + 1 respectively. It is already a very deep (and generally
unanswered) question what the ‘right’ mathematical structure is associated to a
n-torus that gives rise to the group En+1 (Z).
In this case we also have indirect evidence how the graded Hilbert space H
should behave. If one considers so-called BPS states the graded pieces H(γ) should
be finite dimensional, and for large γ we can estimate their growth
dim H(γ) ∼ exp S(γ)
with S(γ) the entropy. Arguments from black hole physics tell us that
S(γ) = Q(γ) ,
 The Mathematics of M-Theory 13

with Q(γ) an algebraic invariant of the duality group G. For example, for n = 5
with G = E6 the lattice Γ has rank 26 and Q is the famous cubic invariant. Simi-
larly for n = 6 and G = E7 we obtain the quartic invariant of the 56-dimensional
representation.

6. D-Branes
The crucial ingredient to extend string theory beyond perturbation theory are
D-branes [13]. From a mathematical point of view D-branes can be considered as
a relative version of Gromov-Witten theory. The starting point is now a pair of
relative manifolds (X, Y ) with X a d-dimensional manifold and Y ⊂ X closed.
The string worldsheets are defined to be Riemann surfaces Σ with boundary ∂Σ,
and the class of maps x : Σ → X should satisfy
x(∂Σ) ⊂ Y .
That is, the boundary of the Riemann surfaces should be mapped to the sub-
space Y .
Note that in a functorial description there are now two kinds of boundaries to
the surface. First there are the time-like boundaries that we just described. Here we
choose a definite boundary condition, namely that the string lies on the D-brane Y .
Second there are the space-like boundaries that we considered before. These are
an essential ingredient in any Hamiltonian description. On these boundaries we
choose initial value conditions that than propagate in time. In closed string theory
these boundaries are closed and therefore a sums of circles. With D-branes there
is a second kind of boundary: the open string with interval I = [0, 1].
The occurrence of two kinds of space-like boundaries can be understood be-
cause there are various ways to choose a ‘time’ coordinate on a Riemann surface
with boundary. Locally such a surface always looks like S 1 × R or I × R. This
ambiguity how to slice up the surface is a powerful new ingredient in open string
theory.
To the CFT described by the pair (X, Y ) we will associate an extended
modular category. It has two kinds of objects or 1-manifolds: the circle S 1 (the
closed string) and the interval I = [0, 1] (the open string). The morphisms between
two 1-manifolds are again bordisms or Riemann surfaces Σ now with a possible
boundaries. We now have to kinds of Hilbert spaces: closed strings HS 1 and open
strings HI .
Semi-classically, the open string Hilbert space is given by
HI = L2 (Y, F)
with Fock space bundle

F= Sqn (T X) .
n≥1
14 R. Dijkgraaf

Note that we have only a single copy of the Fock space F, the boundary conditions
at the end of the interval relate the left-movers and the right-movers. Also the
fields are sections of the Fock space bundle over the D-brane Y , not over the full
space-time manifold X. In this sense the open string states are localized on the
D-brane.
We have seen that an important new ingredient in the step from classical
geometry to ‘stringy’ geometry was the 2-formB field, technically a connection on
a gerbe. It coupled to the string worldsheet via

B.
Σ

The phase factor


exp i B
Σ

that appears in the path-integral should be considered as the generalization of the

holonomy of a connection on a line bundle associated to a loop. It satisfies a gauge
invariance

B → B + dΛ .

If we now work in the category of Riemann surfaces with boundary, we see that
such maps are not gauge invariant but pick up a phase factor

exp i Λ.
∂Σ

However, on surfaces with boundary we can weight the path-integral with an extra
phase factor. Let A be a connection (on a trivial line bundle) on Y . Then we can
add the holonomy phase factor

exp i A.
∂Σ

Now we see that the system has a generalized gauge invariance where apart from
the transformation of B we have

A → A − Λ.

This leads to a generalized notion of gauge invariant curvature

F = dA + B .

This equation immediately implies that when restricted to the D-brane Y the cur-
vature H = dB should vanish.
 The Mathematics of M-Theory 15

6.1. Branes and matrices

One of the most remarkable facts is that D-branes can be given a multiplicity N
which naturally leads to a non-abelian structure [19].
Given a modular category as described above there is a simple way in which
this can be tensored over the N × N hermitean matrices. We simply replace the
Hilbert space HI associated to the interval I by
HI ⊗ M atN ×N
with the hermiticity condition
(ψ ⊗ MIJ )∗ = ψ ∗ ⊗ MJI .
The maps ΦΣ are generalized as follows. Consider for simplicity first a surface Σ
with a single boundary C. Let C contain n ‘incoming’ open string Hilbert spaces
with states ψ1 ⊗ M1 , . . . , ψn ⊗ Mn . These states are now matrix valued. Then the
new morphism is defined as
ΦΣ (ψ1 ⊗ M1 , . . . , ψn ⊗ Mn ) = ΦΣ (ψ1 , . . . , ψn ) Tr(M1 · · · Mn ) .
In case of more than one boundary component, we simply have an additional trace
for every component.
In particular we can consider the disk diagram with three open string inser-
tions. By considering this as a map
ΦΣ : HI ⊗ HI → HI
we see that this open string interaction vertex is now given by
ΦΣ (ψ1 ⊗ M1 , ψ2 ⊗ M2 ) = (ψ1 ∗ ψ2 ) ⊗ (M1 M2 ) .
So we have tensored the associate string product with matrix multiplication.
If we consider the geometric limit where the CFT is thought of as the semi-
classical sigma model on X, the string fields that correspond to the states in the
open string Hilbert space HI will become matrix valued fields on the D-brane Y ,
i.e. they can be considered as sections of End(E) with E a (trivial) vector bundle
over Y .
This matrix structure naturally appears if we consider N different D-branes
Y1 , . . . , YN . In that case we have a matrix of open strings that stretch from brane YI
to YJ . In this case there is no obvious vector bundle description. But if all the
D-branes coincide Y1 = · · · = YN a U (N ) symmetry appears.

6.2. D-branes and K-theory

The relation with vector bundles has proven to be extremely powerful. The next
step is to consider D-branes with non-trivial vector bundles. It turns out that these
configurations can be considered as a composite of branes of various dimensions [6].
There is a precise formula that relates the topology of the vector bundle E to the
16 R. Dijkgraaf

brane charge µ(E) that can be considered as a class in H ∗ (X). (For convenience
we consider first maximal branes Y = X.) It reads [8]

µ(E) = ch(E)A1/2 ∈ H ∗ (X) . (2)

Here ch(E) is the (generalized) Chern character ch(E) = Tr exp(F/2πi) and A is

the genus that appears in the Atiyah-Singer index theorem. Note that the D-bane
charge can be fractional.
Branes of lower dimension can be described by starting with two branes of
top dimension, with vector bundles E1 and E2 , of opposite charge. Physically
two such branes will annihilate leaving behind a lower-dimensional collection of
branes. Mathematically the resulting object should be considered as a virtual
bundle E1  E2 that represents a class in the K-theory group K 0 (X) of X [20]. In
fact the map µ in (2) is a well-known correspondence

µ : K 0 (X) → H even (X)

which is an isomorphism when tensored with the reals. In this sense there is a
one-to-one map between D-branes and K-theory classes [20]. This relation with
K-theory has proven to be very useful.

6.3. Example: the index theorem

A good example of the power of translating between open and closed strings is
the natural emergence of the index theorem. Consider the cylinder Σ = S 1 × I
between two D-branes described by (virtual) vector bundles E1 and E2 . This can
be seen as closed string diagram with in-state |E1  and out-state |E2 

ΦΣ = E2 , E1  .

Translating the D-brane boundary state into closed string ground states (given by
cohomology classes) we have

|E = µ(E) ∈ H ∗ (X)

so that

ΦΣ = ch(E1 )ch(E2∗ )A .
X

On the other hand we can see the cylinder also as a trace over the open string
states, with boundary conditions labeled by E1 and E2 . The ground states in HI
are sections of the Dirac spinor bundle twisted by E1 ⊗ E2∗ . This gives

ΦΣ = TrHI (−1)F = index(DE1 ⊗E2∗ ) .

So the index theorem follows rather elementary.

 The Mathematics of M-Theory 17

7. U-Duality
We indicated that in M-theory we do not want to include only strings but also
D-branes (and even further objects that I will suppress in this discussion such as NS
5-branes and Kaluza-Klein monopoles). So in the limit of small string coupling λ
the full (second quantized) string Hilbert space would look something like
H = S ∗ (Hstring ) ⊗ S ∗ (Hbrane ) .
Of course our discussion up to now has been very skew. In the full theory there
will be symmetries, called U-dualities, that will exchange strings and branes.
There are at present very few formulations of M-theory that present such a
manifest duality-invariant approach. Only for very special compactifications (such
as low-dimensional tori) matrix theory [1] or the famous AdS-CFT correspon-
dence [11] gives a precise non-perturbative definition.
We will give a rather simple example of such a symmetry that appears when
we compactify the (Type IIA) superstring on a four-torus T 4 = R4 /L. In this case
the charge lattice has rank 16 and can be written as
Γ4,4 ⊕ K 0 (T 4 ) .
It forms an irreducible spinor representation under the U-duality group
G = SO(5, 5, Z) .
Notice that the T-duality subgroup SO(4, 4, Z) has three inequivalent 8-dimen-
sional representations (related by triality). The strings with Narain lattice Γ4,4
transform in the vector representation while the even-dimensional branes labeled
by the K-group K 0 (T 4 ) ∼
= ∧even L∗ transform in the spinor representation. (The
odd-dimensional D-branes that are labeled by K 1 (T ) and that appear in the Type
IIB theory transform in the conjugate spinor representation.)
To compute the spectrum of superstrings we have to introduce the corre-
sponding Fock space. It is given by
∞
 
Fq = Sqn (R8 ) ⊗ qn (R8 ) = q N F(N ) .
n=1 N ≥0

The Hilbert space of BPS strings with momenta p ∈ Γ4,4 is then given by
Hstring (p) = F( 12 p2 ) .
For the D-branes we take a completely different approach. Since we only
understand the system for small string coupling we have to use semi-classical
methods. Consider a D-brane that corresponds to a K-theory class E with charge
vector µ = ch(E) ∈ H ∗ (T ). To such a vector bundle we can associate a moduli
space Mµ of self-dual connections. (If we work in the holomorphic context we could
equally well consider the moduli space of holomorphic sheaves of this topological
class.) Now luckily a lot is know about these moduli spaces. They are hyper-Kähler
18 R. Dijkgraaf

and (for primitive µ) smooth. In fact, they are topologically Hilbert schemes [10]
which are deformations of symmetric products
Mµ ∼
2 2
= Hilbµ /2 (T 4 ) ∼ S µ /2 T 4 .
Computing the BPS states through geometric quantization we find that
Hbrane (µ) = H ∗ (Mµ ) .
The cohomology of these moduli spaces have been computed [7] with the result
that

q N H ∗ (HilbN (T 4 )) = Fq .
N ≥0

This gives the final result

Hbrane (µ) = F(µ2 /2) ∼
= Hstring (p)
where µ and p are related by an SO(5, 5, Z) U-duality transformation.

8. Non-Commutative Geometry
From the present point of view, D-branes and the corresponding open strings
seem to indicate that in a more final formulation of M-theory a fundamental role
is played by non-commutative geometry [2, 16]. One of the indications is the occur-
rence of the B-field in string theory. Roughly in the presence of such a (constant)
B-field the space-time coordinates do not longer commute, but instead satisfy
[xµ , xν ] = Bµν .
One of the most striking results is that the D-brane gauge theory, in the presence of
such a B-field indeed is invariant under the T-duality group through the concept
of Morita equivalence. Further explorations of the links between string theory
and noncommutative geometry can well give the key to a final understanding of
M-theory.

References
[1] T. Banks, W. Fischler, S. H. Shenker and L. Susskind, M Theory as a matrix model:
a conjecture, Phys. Rev. D55 (1997) 5112–5128, hep-th/9610043.
[2] A. Connes, M. Douglas and A. Schwarz, Noncommutative geometry and matrix the-
ory: compactifications on tori, JHEP 9802 (1998) 003, hep-th/9711162.
[3] D. A. Cox and S. Katz, Mirror Symmetry and Algebraic Geometry (Mathematical
Surveys and Monographs, No. 68.), AMS, 1999.
[4] Ph. Di Francesco, P. Mathieu and D. Senechal, Conformal Field Theory (Graduate
Texts in Contemporary Physics), Springer, 1996.
[5] R. Dijkgraaf, Les Houches Lectures on Fields, Strings and Duality, in Quantum
Symmetries, les Houches Session LXIV, Eds. A. Connes, K. Gawedzki, and J. Zinn-
Justin, North-Holland, 1998, hep-th/9703136.
 The Mathematics of M-Theory 19

[6] M. Douglas, Branes within branes, hep-th/9512077.

[7] L. Göttsche, The Betti numbers of the Hilbert scheme of points on a smooth projective
surface, Math. Ann. 286 (1990) 193–207; Hilbert Schemes of Zero-dimensional Sub-
schemes of Smooth Varieties, Lecture Notes in Mathematics 1572, Springer-Verlag,
1994.
[8] M. Green, J. Harvey and G. Moore, I-brane inflow and anomalous couplings on
D-branes, Class. Quant. Grav. 14 (1997) 47–52, hep-th/9605033.
[9] N. Hitchin, Lectures on special Lagrangian submanifolds, math/9907034.
[10] D. Huybrechts, Compact hyperkähler manifolds: basic results, alg-geom/9705025.
[11] J. Maldacena, The large N limit of superconformal field theories and supergravity,
Adv. Theor. Math. Phys. 2 (1998) 231–252, hep-th/9711200.
[12] Yu. I. Manin, Frobenius Manifolds, Quantum Cohomology, and Moduli Spaces, AMS,
1999.
[13] J. Polchinski, Dirichlet-branes and Ramond-Ramond charges, Phys. Rev. Lett. 75
(1995) 4724–4727, hep-th/9510017.
[14] J. Polchinski, String Theory (Cambridge Monographs on Mathematical Physics),
Cambridge University Press, 1998.
[15] G. Segal, The definition of conformal field theory, preprint; Two dimensional confor-
mal field theories and modular functors, in IXth International Conference on Mathe-
matical Physics,. B. Simon, A. Truman and I. M. Davies Eds. (Adam Hilger, Bristol,
1989).
[16] N. Seiberg and E. Witten, String theory and noncommutative geometry, JHEP 9909
(1999) 032, hep-th/9908142.
[17] A. Sen and B. Zwiebach, Tachyon condensation in string field theory, JHEP 0003
(2000) 002, hep-th/9912249.
[18] E. Witten, String theory in various dimensions, Nucl. Phys. B 443 (1995) 85, hep-
th/9503124.
[19] E. Witten, Bound states of strings and p-branes, Nucl. Phys. B460 (1996) 335, hep-
th/9510135.
[20] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188.
[21] B. Zwiebach, Closed string field theory: quantum action and the B-V master equation,
Nucl. Phys. B390 (1993) 33–152, hep-th/9206084.

Korteweg-de Vries Institute for Mathematics,

University of Amsterdam,
Plantage Muidergracht 24,
1018 TV Amsterdam, The Netherlands
E-mail address: [email protected]