String Lectures 2016
String Lectures 2016
A.N. Schellekens
3 String Spectra 24
3.1 Mode Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.1 Closed Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.2 Open String Boundary Conditions . . . . . . . . . . . . . . . . . . . 25
3.1.3 Open String Mode Expansion . . . . . . . . . . . . . . . . . . . . . 26
3.1.4 Open versus Closed . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Negative Norm States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Mode Expansion of the Constraints . . . . . . . . . . . . . . . . . . . . . . 28
3.6 The Virasoro Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7 Operator Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.8 Commutators of Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.9 Computation of the Central Charge . . . . . . . . . . . . . . . . . . . . . . 31
3.10 The Virasoro Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.11 Imposing the Virasoro Constraints . . . . . . . . . . . . . . . . . . . . . . 33
3.12 The Mass Shell Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.12.1 Closed Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.12.2 Open Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.12.3 Tachyons? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.13 Unphysical State Decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.14 Light-Cone Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.14.1 Constraints in Light-Cone Coordinates . . . . . . . . . . . . . . . . 37
3.14.2 Zero-Mode Constraints: Closed Strings . . . . . . . . . . . . . . . . 38
3.14.3 Zero-Mode Constraints: Open Strings . . . . . . . . . . . . . . . . . 39
3.15 The Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.15.1 Open Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2
3.15.2 The Little Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.15.3 Tensor Representations of Orthogonal Groups . . . . . . . . . . . . 41
3.15.4 Open Strings (continued) . . . . . . . . . . . . . . . . . . . . . . . . 42
3.15.5 Closed Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.16 The Critical Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.17 The Lorentz Algebra∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.18 Unoriented Strings∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.19 Chan-Paton labels∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.20 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4 Compactification 49
4.1 Space-time Compactification for Particles . . . . . . . . . . . . . . . . . . . 49
4.2 Space-time Compactification for Strings . . . . . . . . . . . . . . . . . . . . 50
4.3 The Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 T-duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 The Self-Dual Point∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Field Theory Interpretation∗ . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.7 Compactification of More Dimensions . . . . . . . . . . . . . . . . . . . . . 54
4.8 Narain Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.9 Background Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.9.1 The Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.9.2 The Anti-Symmetric Tensor . . . . . . . . . . . . . . . . . . . . . . 57
4.10 Moduli∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.11 Vacuum Selection∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.12 Orbifolds∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3
5.14 The Critical Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.15 The Lorentz Algebra∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.16 The Ramond Ground State . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.17 World Sheet Supersymmetry∗ . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.18 World Sheet Supergravity∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.19 The Constraints∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7 Superstrings 102
7.1 Free Fermion Partition Functions . . . . . . . . . . . . . . . . . . . . . . . 102
7.1.1 Neveu-Schwarz states . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.1.2 Ramond States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3 Modular Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.4 The Modular Invariant Partition Function . . . . . . . . . . . . . . . . . . 107
7.5 Multi-loop Modular Invariance∗ . . . . . . . . . . . . . . . . . . . . . . . . 108
7.6 Superstrings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.7 Supersymmetry and Supergravity∗ . . . . . . . . . . . . . . . . . . . . . . 110
4
7.8 Supersymmetry for the Massive States . . . . . . . . . . . . . . . . . . . . 110
7.9 Ramond-Ramond Particles∗ . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.10 Other Modular Invariant String Theories . . . . . . . . . . . . . . . . . . . 111
7.11 Internal Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.11.1 Modular Invariant Partition Functions for Internal Free Fermions . 112
7.11.2 Factorized Partition Functions . . . . . . . . . . . . . . . . . . . . . 113
7.11.3 Symmetries of Free Fermions . . . . . . . . . . . . . . . . . . . . . . 114
7.11.4 The 18- and 10-Dimensional Theories . . . . . . . . . . . . . . . . . 114
7.11.5 Relation with Even Self-dual Lorentzian Lattices . . . . . . . . . . 115
10 Dualities∗ 131
10.1 Open Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
10.2 Finiteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
10.3 Dual Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
10.4 Historical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
10.5 T-duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
10.6 Perturbative and Non-perturbative Physics . . . . . . . . . . . . . . . . . . 138
10.7 D-branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
10.8 S-duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5
Appendices 143
6
These notes follow rather closely the course given in the fall semester of 1999. They
are based to a large extent on the books by Green, Schwarz and Witten, [1], the book
by Polchinski [2], the book of Lüst and Theisen [3], and the review article [5]. I refer to
these sources for further references. A very useful introductory text that appeared after
the original version of these notes was written is the book by Zwiebach [4].
7
other possibilities exist.
This fact would already be enough to study String Theory seriously. But there is more.
String Theory does not just contain gravity, it comes inevitably with a large number of
other particles and interactions. These particles and interactions have the same features as
the Standard Model. Typically there are quark and lepton-like particles, coupling to gauge
bosons. Furthermore there are usually some scalars. The interactions are often chiral,
which means that left- and right-handed particles couple in different ways. This feature is
observed in the weak interactions. In fact any observed property of the Standard Model
can be identified within String Theory. Among the possibilities are also Grand Unification
or Supersymmetry.
It was once hoped – although for no good reasons – that string theory would uniquely
determine the standard model. Meanwhile we know enough to be able to say that this
is not true. In fact, there are so many possibilities that at the moment nobody is able
to prove whether the Standard Model is among the possibilities. Nevertheless, if String
Theory is correct as a theory of gravity, there is really no other option than that it is also a
theory of all other interactions. According to our present understanding, String Theory is
far more robust than the usual field-theoretic description of particles and interactions. We
cannot take a String Theory description of gravity and simple add the Standard Model to
it. One cannot add particles or interactions to a given String Theory, or remove particles
one does not like.
This seems almost in contradiction with the fact that String Theory is far from unique.
The difference between the String Theory and Field Theory description of nature is illus-
trated by the following picture
A B
B
The first picture shows two possible string theories, the second two possible field
theories. The idea is that in field theory we can extend theory A to theory B without
throwing any part of theory A away. In String Theory we cannot do that. We can modify
theory A to theory B by throwing part of theory A away and adding something new in
its place. Suppose we know experimentally that we need everything in theory A. Then
String theory would actually predict that there are no other particles or interactions. If
an experiment does find something not predicted by theory A, we would be forced to give
up String Theory altogether. On the other hand, if we are in the situation of the second
picture, we could simply add something to our theory.
For this reason one sometimes uses the phrase “Theory of Everything” when talking
about string theory. Not everybody thinks it is a good idea to use this phrase.
8
1.1 Problems of Quantum Gravity
One of the first problems one encounters when one attempts to quantize a field theory is
that certain calculations seem to lead to infinities. For example, loop diagrams may lead
to integrals of the following type
Z
1
d4 p 2 (1.1)
(p + m2 )2
where pµ is a four-momentum, and m some mass. It is easy to see that this integral
diverges for large p.
If one thinks for a moment about what one is doing here, it seems rather silly to inte-
grate over arbitrarily large momenta. Large momenta correspond to short distances, and
by integrating to infinity, we are pretending that we understand physics to arbitrarily short
distances. This is a ridiculous assumption. Perhaps space-time is discrete at extremely
small distances, so that there is a maximum momentum, the inverse of the smallest dis-
tance. Perhaps something else changes at short distances. If anything changes at short
distances, it may happen that our integral is actually finite, proportional to log(Λ), where
Λ is some large, but finite, momentum scale. The trouble is that without detailed knowl-
edge about the short distance physics we cannot compute the integral, all we can do is
introduce a new parameter equal to the value of the integral. However, in quantum field
theory there are integrals of this type for any loop diagram, and if we cannot compute
any of them we would be forced to introduces an infinity of new parameters. To know the
theory completely we would have to measure all these parameters, and that is obviously
impossible.
However, some theories are nicely behaved. In such theories the unknown parameters
appear only in (infinite) linear combinations such as
m + m1 + m2 + m3 . . . (1.2)
where m is a parameter of the original theory (for example a mass), while m1 . . . are the
unknown parameters from loop diagrams. If all physical quantities depend only on this
particular combination there is no need to determine them individually. We simply give
the whole sum a new name, mR , and if we know the value of mR we know enough. The
net result of this procedure is that we end up with an equal number of parameters as we
started with. If a theory has this property, we say that it is renormalizable. The Standard
Model has this property, and for this reason we regard it as a consistent quantum theory.
But gravity does not have this property. In the case of gravity the loop corrections
lead to an infinity of new parameters that cannot be absorbed into the ones we started
with (there is just one, namely Newton’s constant). Strictly speaking that means that
any physical quantity we compute in quantum gravity depends on an infinite number of
parameters, and hence the predictive power of the theory is zero. This is of course an
exaggeration. If true, it would imply that we cannot even believe gravity in its classical
limit, where it is so successful.
9
It turns out that for any reasonable value of the unknown parameters, their effect is
completely negligible at observable energies.
The coupling constant of gravity is Newton’s constant GN , defined through
m1 m1
F = GN (1.3)
r2
Using ~ and c we can construct from it a mass, the Planck mass
r
~c
MP = ≈ 2 × 10−8 kg (1.4)
GN
In high energy physics it is customary to multiply masses with c2 and express them in
units of energy, for which the GeV is used. Then we get
s
~c5
MP = = 1.22 × 1019 GeV (1.5)
GN
Using other combinations of ~ and c we can convert this to an inverse length. This length,
the Planck length is 1.6 × 10−33 cm. Usually we use units where ~ = c = 1, so that the
distinction between energy, mass and inverse length disappears.
The gravitational potential can be obtained from graviton exchange between the two
particles. The first quantum correction is due to two-graviton exchange, and is pro-
portional to G2N . The ratio of the two and the one-graviton term must therefore be
proportional to GN . Since the ratio must be dimensionless, it must be of the form GN E 2 ,
where E is some energy of the process under consideration. Hence we see that the first
correction must be proportional to (E/MP )2 , which is extremely small as long as E is
much smaller than 1019 GeV, and as long as the unknown coefficient is of order 1. This
is far beyond anything we can ever hope to achieve with accelerators.
So the problem of finding a quantum theory of gravity is a purely theoretical one.
There does not appear to be any chance of observing its effects directly.
There are other problems associated with quantum mechanics and gravity. Some very
interesting problems arise in the discussion of black holes. One of the most intriguing
features of a black hole is the existence of a horizon. Outside observers cannot get infor-
mation about what happens inside that horizon. They can allow themselves to fall into the
black hole, inside the horizon, but then they cannot communicate with the outside world
anymore. Objects thrown into the black hole cannot be recovered. This seems to clash
10
with unitarity of the quantum mechanical S-matrix, which states that the total proba-
bility is conserved. Indeed, it has been proposed that the existence of black holes would
force us to give up unitarity of the S-matrix. This is called the information-loss prob-
lem. It is made more severe because of black hole evaporation. Given long enough time,
a black hole radiates away all its energy through Hawking radiation until it evaporates
completely. Then apparently all the information that went into it disappears forever.
Another related problem is that of the black hole entropy. Bekenstein and Hawking
have shown that a black hole has thermodynamic properties: it has black body radiation
with a certain temperature, and it behaves as if it has a certain entropy, proportional
to the area of the horizon. Normally one associates entropy with the logarithm of the
number of states of an object, but such a notion was not available here. Presumably
understanding the counting of states of a black hole in relation to its entropy will also tell
us how it can store information.
String theory has made interesting contributions to both problems. First of all string
theory does indeed change the short distance physics. It has been shown that in certain
string theories the unknown or infinite corrections are in fact finite and calculable. Fur-
thermore it has been possible to obtain a description of the states of special black holes
(in special limits) so that the correct entropy could be obtained.
All of this is still very vague, so it is time to take a closer look.
t t
t
There is still one more distinction to be made: strings may or may not have a definite
orientation. Depending on this we distinguish oriented and unoriented strings. All to-
gether there are at this point then four types of strings: open unoriented, open oriented,
closed unoriented and closed oriented. Oriented strings must keep their orientation when
they interact.
Strings interact by splitting and joining. We can represent this by plotting the strings
as a function of time. For example the following picture represents an open string splitting
11
t
A more interesting process is that of an open string decaying into an open and a closed
string.
This shows that open strings necessarily imply closed strings. The other way around
this is not true. It is impossible to draw a diagram where a closed string emits an open
one.
In general a string diagram can be any two-dimensional surface. Such a surface can
be interpreted in various ways by choosing a time coordinate on it. Then one can give a
string interpretation to the interactions by slicing the surface with planes orthogonal to
the time direction. For example, the previous diagram is obtained by time-slicing in a
suitable way a piece of a plane to which a cylinder is attached.
On a cylinder there are two obvious ways of choosing a time coordinate
12
t
Closed string
propagation
t
=
t
One corresponds to the propagation of a closed string, the other to an open string
making a closed loop.
One of the most important features of string theory is that to each such diagram there
corresponds a certain amplitude. We may try to interpret that amplitude anyway we like,
but there is only one amplitude. For example, to compute the scattering amplitude for
two closed strings to two closed strings one begins by drawing all valid diagrams with two
initial and two final closed strings, propagating from −∞ or to +∞ respectively.
There is a precise procedure that assigns to each such diagram a number, the transition
amplitude. This procedure is based on path integrals. These are infinite dimensional
integrals over all surfaces with certain initial and final states one is interested in. The
surfaces are weighted by an exponential; very roughly the time evolution from initial to
final states is then given by
Z
hout| U |ini = eiS/~ , (1.6)
Surfaces
where S is the (classical) action. In the simplest case S is simply the area of the surface.
Of course such infinite dimensional integrals require a much more detailed discussion.
Which diagrams are allowed depends on the kind of string theory one considers. In
the simplest case, oriented closed strings, the allowed diagrams would be the sphere,
the torus, the double torus, etc, each with four infinite tubes attached. Two diagrams
are considered equivalent if we can deform them into each other in a continuous way,
without breaking anything. It is a basic result in mathematics that all such surfaces
are characterized topologically by the number of handles (the genus), i.e. a torus is a
sphere with one handle, etc. In the case of oriented open and closed strings one must
allow the surface to have holes, corresponding to the ends of open strings. In the case of
unoriented closed strings one cannot have holes, but one must allow closed surfaces that
do not have a definite orientation, such as the Klein bottle (a surface has a well-defined
orientation if one can consistently define a normal vector in a continuous way, on all points
13
on the surface). Finally, in the case of unoriented open strings one must also allow open,
unoriented surfaces. The best known example is the Moebius strip.
External lines are either tubes glued to the surface itself, or strips glued to the bound-
aries. The latter correspond of course to external open strings.
All of this may be compared to the Feynman diagram expansion of quantum field
theory. One obvious difference is that in that case one obtains a strongly increasing
number of diagrams in each order. This is already clear if we look at a theory with
a three-point interaction. Also in quantum field theory there is a precise calculational
procedure for computing an amplitude belonging to such a diagram.
At lowest order, there are three diagrams, whereas in oriented closed string theory there
is just one. At one loop order, there is still only one string diagram, but many more field
theory diagrams.
But there is a more important difference between the two situations. In field theory
diagrams there are special points corresponding to the splitting of a particle into two
(or more) particles. This implies that we have to give additional information about the
strength of the interaction at these points. In a string theory there are no special points.
The surfaces are smooth, and every point is equivalent to any other point. If we time-slice
a surface, there may appear to be special points where the string splits. But this depends
on how we define the equal-time slices. This is not a Lorentz-invariant concept, and hence
two different observers will not agree on a point on the two-dimensional surface which is
special. Hence there are no special points.
One consequence is that by specifying a free string theory, we have already specified
the interactions. As we will see, a string theory is specified by giving an action. The
action is always some two-dimensional Lagrangian density, integrated over the surface. In
the simplest cases the action is nothing but the area of the surface. The action does not
know about the topology of the surface. Computing different string diagrams amounts
then to computing the integrals of the same integrand over different surfaces.
The fact that string theory has no ultra-violet (short distance) divergences is also
related to this fact. The divergences in quantum field theory can be traced back to the
existence of pointlike interactions, and can in fact be removed by “smearing out” the
interactions over a finite region. But in string theory there are no point-like interactions
to begin with.
String theory may nevertheless have divergences, but they are of a different kind, and
can be avoided in certain cases.
14
2 Bosonic String Action
The simplest string theory is the bosonic string. Although simple, it has serious defects:
its spectrum contains no space-time fermions, but it does contain a particle with imaginary
mass, a tachyon. It is therefore not a candidate for a theory of nature, but it is very useful
as a kind of toy-model.
15
2.2 The Nambu-Goto action
As already mentioned above, we want to apply the same idea to String Theory. Now
we generalize from lines to surfaces, which are now called world sheets. To specify the
embedding of a world sheet in space-time we give D functions X µ (σ, τ ). Now there are
two variables, one to describe the direction along the string (σ), and one to parametrize
a time-like direction. The action is proportional to the surface-area of the world sheet,
which can be written as
Z
1 p
S[X] = − dσdτ −det hαβ (2.2)
2πα0
where hαβ is a two-by-two matrix defined as
hαβ = ∂α X µ ∂β Xµ , (2.3)
where ∂α means ∂σ∂ α , and where instead of σ and τ we use variables σ α , α = 0, 1, with
σ 0 = τ and σ 1 = σ. The sign of the determinant is determined by the space-time metric. A
string at rest could be given by X 0 = τ , X i = X i (σ). Then h00 = −1 and h11 = (∂σ X i )2 ,
h01 = h10 = 0, so that the determinant is negative. Then it will stay negative under
continuous deformations (since det h = 0 implies the manifold is singular at that point).
The action (2.2) is called the Nambu-Goto action. The coefficient in front of it is
chosen this way by convention. The parameter α0 has a curious history. It started out as
the derivative at a special point of a function called α, hence the prime. Sometimes it is
also called the “Regge slope parameter”, a name that also suggests a first derivative. But
by now the origin is usually forgotten, and the parameter is called simple α0 . Note that
it has the dimension of [length]2 . This is because X has the dimension of length, and an
action is dimensionless (if one chooses ~ = 1). Furthermore σ and τ are just parameters,
and can be assumed to be dimensionless. One may also give them dimension of a length,
but the dimension of σ and τ cancels out anyway.
Here γαβ is a symmetric, non-singular two-by-two matrix that may depend on σ and τ .
By γ̂αβ we mean the inverse of γαβ . The determinant of γαβ is assumed to be negative
here; this will be justified in a moment.
1
As indicated, γαβ (σ, τ ) may depend on σ and τ , but to simplify the notation we drop the arguments
as long as they are not essential.
16
To see the classical equivalence, note that γαβ is not a dynamical degree of freedom:
it occurs without any derivatives. Hence we can eliminate it by using the equations of
motion. The latter are defined by requiring that the variation of the action with respect
to γαβ vanishes:
δγ S ≡ S[X, γ + ∆[γ]] − S[X, γ] = 0 , (2.5)
where ∆[γ] is an infinitesimal variation of γ; this variation is a two-by-two matrix, and
also a function of σ and τ , but we leave out these details because it is too cumbersome
to write ∆[γ]αβ (σ, τ ) in all formulas. In the rest of the section we just use the symbol δ
to denote δγ , the variation with respect to γ. So in order to be very precise:
for some functional F [γ]. The action depends on γαβ via the determinant and the inverse.
The variation of the inverse is computed as follows:
17
with implicit summation over all repeated indices. This can only vanish for arbitrary
(∆γ)γδ if the terms between square brackets add up to zero; this yields the matrix identity
This allows us to express γ in terms of h. We find that γ̂ = Λh−1 . When we substitute this
into the action S[X, γ] (Eq. (2.4)) we get a factor of 2Λ from Trγ̂h. The function Λ cancels
against a factor from the determinant, and then we do indeed obtain the Nambu-Goto
action S[X].
It is customary to introduce upper and lower indices (analogous to the space-time
Lorentz indices) also for the parameters σ α . Then one defines
γ αβ ≡ γ̂αβ , (2.16)
and we use this matrix to raise indices, and γαβ to lower them. Since these matrices are
each other’s inverse, this makes sense. The action reads now
Z
1 p X
S[X, γ] = − 0
dσdτ −det γ γ αβ ∂α X µ ∂β Xµ . (2.17)
4πα αβ
From the relation between γ and h we see that the natural sign of the determinant is
negative. This action can be recognized as that of D free scalar fields X µ , µ = 1, . . . , D
coupling to a two-dimensional metric γαβ .
Here gµν may depend on the space-time position, but not on the word-sheet coordinates
σ a , like γ and X. Hence from the two-dimensional point of view γ and X are fields, but
gµν is not. If we are considering a string theory embedded in flat, Minkowski space-time,
we must choose gµν = ηµν , η = diag (−1, 1, . . . , 1).
One may also consider a string propagating through curved space-time. This is ob-
tained by simply substituting the metric gµν of that space-time. In general, gµν will
depend on X, so that the resulting two-dimensional action is nonlinear in X. Similarly, a
point particle moving through curved space-time can be described by the action 2.1 with
the implicit flat metric ηµν replaced by gµν (X). The equation of motion can be shown
18
to be equal to the geodesic equation. In both cases, point particle and string theory, one
also calls gµν (X) the “background” through which the particle or string propagate.
In these lectures, however, we restrict ourselves to flat space-times. We remarked
above that µ is just a multiplicity. This is not quite true: because we embed the string
in minkowski space, one of the bosons will have an action of the “wrong” sign. This has
important consequences, as will be discussed later.
2.5 Symmetries
The bosonic string action S[X, γ] has the following symmetries
• Weyl invariance. This is a symmetry not seen in the Nambu-Goto action, and is a
consequence of the free function Λ we found above. Indeed, it is easy to see that
the action is invariant under
0
γαβ (σ, τ ) = Λ(σ, τ )γαβ (σ, τ ) (2.21)
without changing X µ .
19
a new function X 0µ (σ 0 , τ 0 ) which describes the space-time position of every point on the
string in terms of the new parametrization. By definition this new function is therefore
X 0µ (σ 0 , τ 0 ) = X µ (σ, τ ). Now we write the action entirely in terms of the new variables
X 0 , σ 0 and τ 0 . For most variables, this simply implies that we replace them in (2.4)
by the primed variables. In particular, dσdτ is replaced by dσ 0 dτ 0 and ∂α by ∂α0 . Any
action could be rewritten in this manner, but this is only a symmetry if we re-obtain the
original action when we express the primed variables back in terms of the original ones.
Changing back to the original variables introduces two potential problems: the change of
integration measure when expressing σα0 in terms of σα , and the change from ∂α0 back to
∂α . Even without prior knowledge of general relativity, one realizes that in order for this
to be a symmetry we have to transform γ as well. This is allowed, because we view γ as
a two-dimensional field that can be modified under transformations. It is easy to see that
γ has to change exactly as Eq. (2.20) in order for the derivatives to cancel.
The last two symmetries are redundancies of the two-dimensional theory on the world
sheet. This means that the two-dimensional action has fewer variables than it seems to
have. We have three free functions σ 0 (σ, τ ), τ 0 (σ, τ ) and Λ(σ, τ ) at our disposal to modify
0
the dynamical variables γαβ (σ, τ ) and X µ (σ, τ ). In this way we can reduce their total
number by three.
This reduction of degrees of freedom completely analogous to gauge symmetry in
electrodynamics: the invariance under the gauge transformation Aµ → Aµ + ∂µ Λ implies
that one degree of freedom of Aµ can be removed using the function Λ(x). Note that in
addition to this, A0 is non-dynamical, so that the number of physical degrees of freedom
of the photon in D dimensions is reduced from D to D − 2. Hence in four dimensions
the photon has two physical degrees of freedom, namely its two polarizations. As is also
familiar from electrodynamics, one may fix the redundancy by choosing a gauge. This
has to be done in a clever way in order to make quantization of the theory manageable.
Note that the first symmetry only uses constant parameters M µν and aµ , which cannot
remove any degrees of freedom (i.e. reduce the number of functions of σ and τ ). One
would expect the theory to have general coordinate invariance also in target space. It
must have that property, or else it cannot reproduce general relativity. However, general
coordinate transformations lead to changes in the metric gµν , which will be kept fixed
in our treatment of the string action: the components of gµν can be viewed as coupling
constants of the two-dimensional theory. Another way of saying that is that the formalism
is background dependent, unlike the formalism for general relativity. However, it is only
the formalism that has this deficit. It might be nice to have an explicitly general coordinate
invariant formulation of string theory, but it is not a necessity.
20
They cancel only for N = 2. The number of components of the metric is 12 N (N + 1).
Using reparametrization invariance one can put N components of the metric to zero (at
least locally, in a finite neighborhood of a point), reducing the number of components to
1
2
N (N − 1). In two dimensions the extra Weyl invariance then allows us to remove one
more degree of freedom, so that finally none is left! This illustrates why two dimensions is
special, and why string theory is much easier than theories of membranes, whose world-
sheets are N -dimensional, with N > 2.
We can use this freedom to transform the metric to the flat, Minkowski metric ηab =
diag (−1, 1) (with a −1 in the τ direction). This is called conformal gauge. Then the
action looks even simpler
Z
1
S[X] = − dσdτ η ab ∂a X µ ∂b Xµ (2.22)
4πα0
Note that we use indices α, β, . . . for world-sheet indices in arbitrary metrics, and a, b, . . .
in conformal gauge. This is the simplest action one can write down in two-dimensional
field theory. It is the action of a free boson. More precisely, of D free bosons, because
the index µ on the bosons is from the two-dimensional point of view nothing other than a
multiplicity. The action S[X, γ] we had earlier is in fact the action of a free boson coupled
to two-dimensional gravity.
−4πα0 δ
T αβ (σ, τ ) ≡ √ S[X, γ] (2.24)
−γ δγαβ
1
From now on we write “γ” as a short-hand for “detγ”, when no confusion is possible.
21
The factor 4πα0 is for convenience. The derivative appearing in this definition is called
a functional derivative. It is defined as follows. Consider a functional F [φ], where φ is a
function. By definition, a functional maps a function to a real number. The functional
derivative of this functional with respect to φ is defined as
δF [φ] F [φ + δ(x − y)] − F [φ]
≡ lim (2.25)
δφ(y) →0
To compute this, we make use of the fact that we have already calculated the variation
with respect to γ in section 2.3. In that section we found
Z X
δγ S = dσ 0 dτ 0 F (σ 0 , τ 0 )ρη ∆[γ]ρη (σ 0 , τ 0 ) , (2.26)
ρη
this is just Eq. (2.13) with the dummy integration variables σ and τ replaced by σ 0 and
τ 0 , and analogously for the summation indices. The matrix function F αβ can be read off
from Eq. (2.13). To compute the functional derivative with respect to γαβ (σ, τ ) we choose
the following infinitesimal variation
∆[γ]ρη (σ 0 , τ 0 ) = δαρ δβη δ(σ − σ 0 )δ(τ − τ 0 ) (2.27)
Using the definition of the functional derivative, Eq. (2.25), we obtain
1
T αβ (σ, τ ) = ∂ α X µ ∂ β Xµ − γ αβ ∂γ X µ ∂ γ Xµ (2.28)
2
Since the variation of S with respect to γ must vanish, we find that T αβ must vanish.
The other equation of motion is the one for X µ . Although it can be computed for
arbitrary γ, for simplicity we go to conformal gauge. Then the equation is simply
2
∂2
∂
− Xµ = 0 (2.29)
∂τ 2 ∂σ 2
In two dimensions this equation can be solved very easily by writing it in terms of variables
σ + = σ + τ and σ − = σ − τ . The general solution is
X µ (σ, τ ) = XLµ (τ + σ) + XRµ (τ − σ) , (2.30)
where XLµ and XRµ are arbitrary functions. A special point of XL , for example the max-
imum, stays at a fixed value of σ + τ . Hence as τ increases, σ must decrease. Then the
maximum of XL moves to the left, and analogously the maximum of XR moves to the
right. Therefore we call these left- and right-moving components.
22
The same occurs here. It turns out that there are special coordinate transformations
that change the metric by an overall factor:
∂σ 0γ ∂σ 0δ 0 0 0
γ (τ , σ ) = Λ(σ, τ )γαβ (σ, τ ) (2.31)
∂σ α ∂σ β γδ
Then one can remove Λ using a Weyl transformation, and the net effect is that the metric
has not changed at all. Hence if we start with the conformal gauge metric η ab we keep
the same metric, and therefore we have a residual symmetry.
This problem can be formulated in any dimension. One can completely classify all
transformations of the coordinates that satisfy (2.31). It should be clear that this in-
cludes in any case (Lorentz)-rotations, σ 0a = Λab σ b , and translations, σ 0a + a , since they
do not change the metric at all. Another transformation with this property is a scale
transformation, σ 0a = λσ a .
The name of this kind of a transformation derives from the fact that it preserves
angles, but not lengths. Consider two vectors v α and wα , which may be in any dimension.
The angle between these vectors (or its obvious generalization to Minkowski space) is
v α gαβ wβ
p (2.32)
(v α gαβ v β )(wγ gγδ wδ )
A scale factor Λ(σ, τ ) in the metric obviously drops out in the ratio, so angles are indeed
preserved. So in some sense these transformation preserve the “form” of an object, but
not its size. This is why they are called conformal transformations.
Just as symmetries in other systems, these transformations form a group, called the
conformal group. We have already seen that it contains the Lorentz group as a subgroup.
In p space and q time dimensions, the Lorentz group is SO(p, q). The conformal group
turns out to be SO(p + 1, q + 1), if p + q > 2.
If p + q = 2, however, the conformal group is much larger. This can easily be seen as
follows. Let us try the transformation
τ 0 = f (τ + σ) + g(τ − σ)
(2.33)
σ 0 = f (τ + σ) − g(τ − σ)
where f and g are arbitrary continuous functions. The transformation of the Minkowski
metric η ab under such a transformation leads to the matrix multiplication
0
f + g0 f 0 − g0
0
f + g0 f 0 − g0
−1 0 0 0 −1 0
= 4f g (2.34)
f 0 − g0 f 0 + g0 0 1 f 0 − g0 f 0 + g0 0 1
which proves that the metric transforms to a multiple of itself, and hence that this is
a conformal transformation. Thus we see that we may replace σ + τ and σ − τ by
arbitrary functions, f (σ + τ ) and g(σ − τ ). This is a symmetry with an infinite number
of generators, certainly much larger than the conformal group SO(2, 2) we would expect
by extrapolating the higher dimensional result to two dimensions. This group is in fact
contained in these transformations: the SO(2, 2) transformations turn out to correspond
to functions f and g of the form f (x) = a + bx + cx2 (infinitesimally in c).
23
3 String Spectra
Consider an observer looking at a string moving through space time. The string will have
a center-of-mass motion, around which it moves around. The motion around the center-
of-mass can be decomposed in normal modes of vibration. When the string is in one of
these modes, the observer will view the energy of the mode as the mass of a particle.
There is an infinite number of modes of vibration, and when we quantize the theory, each
of these modes will have its own harmonic oscillator wave function.
24
Note that X µ should be real. This immediately implies
µ
(αnµ )∗ = α−n (3.9)
The way this expression is written is based on standard conventions. Note that αnµ and α0
are unrelated, although both contain the symbol α. The former is dimensionless, whereas
α0 has the dimension of [length]2 . Since X and q have the dimension of [length] and p
[length]−1 , the expression is dimensionally correct.
L(X) ≡ ∂a Xµ ∂ a Xν η µν , (3.10)
Hence Z Z π
1
δX S = − dτ dσ∂a (δXµ )∂ a X µ (3.12)
2πα0 0
which after an integration by parts yields
Z Z π Z σ=π
1 a µ 1
µ
δX S = 0
dτ dσ(δXµ )∂a ∂ X − 0
dτ δXµ ∂σ X (3.13)
2πα 0 2πα σ=0
Requiring the first term to vanish for arbitrary δXµ leads to the equations of motion,
∂a ∂ a X ν = 0. The second term must also vanish, and this happens most easily by choosing
∂σ X µ = 0 at the boundaries σ = 0 and σ = π. This is called a Neumann boundary. A
second possibility would be to assume that δXµ vanishes at the boundary, so that X µ is
constant. Then X µ (σ = 0, τ ) = cµ and X µ (σ = π, τ ) = dµ , where c and d are constant
vectors. Clearly this is not Poincaré invariant, and in particular translation invariance is
always broken. This is called a Dirichlet boundary condition. We will come back to it
later, but for now we restrict ourselves to Neumann boundaries.
1
The size of the interval is irrelevant because of scale invariance: any other choice can be rescaled to π.
25
3.1.3 Open String Mode Expansion
The rest of the discussion is the same as in the closed case, except that the σ-modes are
slightly different now. We find
√ X1
X µ (σ, τ ) = q µ + 2α0 pµ τ + i 2α0 (αnµ e−inτ cos nσ) , (3.14)
n6=0
n
assuming Neumann boundary conditions. Note that this is similar to (3.8) if you expand
the cosine in terms of exponentials. However, the coefficients of the left-movers and the
right-movers are the same, because these modes reverse their direction at the boundary.
3.2 Quantization
The system under consideration can be treated using the methods explained in Appendix
A. The Lagrangian is Z ρπ
1
L=− dσ∂a X µ ∂ a Xµ , (3.15)
4πα0 0
R
so that S = dτ L. Here ρ = 1 for open strings and ρ = 2 for closed strings.
Now it is easy to derive the canonical momentum of X µ :
1
Πµ = ∂ τ Xµ (3.16)
2πα0
From the Poisson brackets of the classical theory we infer the correct form of the commu-
tator in the quantum theory:
[X µ (σ, τ ), Πν (σ 0 , τ )] = iη µν δ(σ − σ 0 )
[X µ (σ, τ ), X ν (σ 0 , τ )] = 0 (3.17)
[Πµ (σ, τ ), Πν (σ 0 , τ )] = 0 .
26
These relations can be expressed in terms of modes. The result is1
and
[q µ , pν ] = iη µν (3.19)
We conclude that q µ and pµ can be interpreted as the center-of-mass coordinate and
momentum, as is already clear from the classical expression.
The α-operators look a lot like harmonic oscillator operators. The quantum version
of (3.9) is
µ
(αnµ )† = α−n (3.20)
If we now define
1 1 µ
aµn = √ αnµ , (aµn )† = √ α−n (3.21)
n n
we obtain a harmonic oscillator algebra for each value of µ and n.
However, because of the appearance of η µν one of the oscillators will have the commutation
relation [a, a† ] = −1. This leads to states of negative norm. States with negative norm
often occur in quantum field theory in intermediate stages of a calculation. They are
called ghosts. In sensible theories they should cancel. In particular, there should not exist
scattering processes of physical particles in which ghosts are produced. At this stage, the
string spectrum seems to contains such quantum states. If this were the end of the story,
the theory would be sick.
1
Note that the commutators (3.17) are at some world-sheet time τ = τ0 . Hence also the modes αk are
the coefficients in the expansion of X at τ0 . The classical time evolution of these modes as given (3.8) is
only used to distinguish left- and right-moving modes at τ0 , but of course the time evolution is governed
by quantum mechanics, and not by the classical evolution (3.8).
27
3.4 Constraints
However, it is not the end of the story. So far we have completely ignored the second
equation of motion, Tab = 0. Written in terms of components this tensor becomes
1
T00 = T11 = (Ẋ 2 + X 02 ) (3.24)
2
T10 = T01 = Ẋ · X 0 (3.25)
where the dot and the prime indicate τ and σ derivatives, and Lorentz-indices are implic-
itly contracted.
We may compare this with the Hamiltonian of our system
Z ρπ Z ρπ
1
H= dσ(Ẋ · Π) − L = 0
(Ẋ 2 + X 02 ) (3.26)
0 4πα 0
Hence we see that, up to normalization, H is just the integral over T00 , which has to
vanish. The Hamiltonian is the generator of time translations. The integral over T01 is
proportional to P , the generator of translations in the σ direction. Obviously, it also has
to vanish. But that is still not all, because Tab has to vanish without integration. Put
differently, there is an infinite number of modes of Tab that has to vanish.
28
In the quantum theory the ∗ becomes a †. These mode expansions are done at τ = 0.
With a little effort, we can substitute the mode expansions for X µ and express the L’s
in terms of the modes of X. The result is
1X 1X
Ln = αn−m · αm , L̄n = ᾱn−m · ᾱm (3.32)
2 m 2 m
pµ |0i = 0 (3.36)
This zero-momentum state can be transformed into a ground state with momentum in
the usual way
µ
|k µ i = eiq kµ |0i , pµ |k µ i = k µ |k µ i , (3.37)
where of course pµ is an operator and k µ a momentum vector. Then we obtain all ex-
cited states by action with the negative modes of the operators α and ᾱ. A typical,
unnormalized state the looks like
µ1 µ`
|n1 , . . . , n` , µ1 , . . . , µ` , m1 , . . . , mκ , ν1 , . . . , νκ , k µ i = α−n 1
. . . α−n ᾱν1 . . . ᾱ−m
` −m1
νκ
κ
|k µ i .
For open strings the space of states looks similar, except that one doesn’t have the ᾱ
operators.
29
Now we have to see what the effect is of the energy-momentum constraints. We have
just introduced the modes of the energy-momentum tensor. These modes, the operators
Ln and L̄n , are called the Virasoro operators. Since the energy-momentum tensor T ab
classically has to vanish, one might come to the conclusion that in the quantum theory
we should impose the constraint
Ln |physi = L̄n |physi = 0 (3.38)
This would restrict the number of quantum states, and one might hope that the ghost
states disappear.
It is clear that changing the order of two oscillators changes L0 by a constant. The
fact that we have imposed an ordering does not mean we can ignore that constant. It only
means that there is an unknown constant in the relation between L0 and the Hamiltonian:
H = (L0 − a) + (L̄0 − ā) closed strings
(3.41)
H = (L0 − â) open strings .
In principle, all constants a, ā and â are unrelated, and may be different. This also affects
the condition we must impose on the spectrum:
(L0 − a) |physi = (L̄0 − ā) |physi = 0 (3.42)
1
Here and in the following we will omit the analogous expressions for the ᾱ operators that appear in
the closed string theory.
30
3.8 Commutators of Constraints
But that is not the only problem. For example, consider the state
1X
L−2 |0i = α−2+p · α−p |0i (3.43)
2 p
Here |0i is the vacuum, at zero momentum.1 In the infinite sum, the only term that
contributes to L−2 is the one with p = 1:
1
L−2 |0i = α−1 · α−1 |0i (3.44)
2
Note that the term α−2 · α0 |0i vanishes because α0 is the momentum operator, and any
other terms contain factors αn with n > 1, which annihilate the vacuum. If we try to
impose (3.38) we would like the norm of this state to vanish. But that is clearly nonsense:
1 1 ν 1 1
||L−2 |0i || = h0| α1µ α1,µ α−1 α−1,ν |0i = ηµν η µν = D 6= 0 (3.45)
2 2 2 2
This problem is a consequence of the fact that the Ln ’s have non-trivial commutation
relations. It turns out to be an interesting, but rather cumbersome exercise to compute
[Ln , Lm ]. One may first do this classically, using Poisson brackets instead of commutators
for the modes αn . Although the Poisson brackets have a non-trivial right hand side,2
the modes commute with each other, and there is no ordering problem. It is rather easy
to compute
{Lm , Ln }PB = i(m − n)Lm+n (3.47)
The calculation in the quantum case is essentially identical as long as we neglect ordering.
Hence we get
[Lm , Ln ] = (m − n)Lm+n + re-ordering contributions (3.48)
The reordering contributions can be computed by carefully commuting the oscillators that
are in the wrong place, but there is a more clever way.
31
where A(m) is a c-number, i.e. it commutes with the Ln ’s (such a contribution is called
a “central charge” of an algebra). Obviously A(−m) = −A(m). Furthermore we know
already that A(2) is non-zero:
using h0| L0 |0i = 0. Hence A(2) = 21 D. In a similar way one finds that A(0) = A(1) = 0.
An elementary property of commutators is the Jacobi-identity
[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0 (3.51)
which should hold independent of what A, B and C are. In the case it yields, for A, B, C
= L1 , L−1−n and Ln
This is a recursion relation for the coefficients. Since A(1) = 0 the equation becomes
n+2
A(n + 1) = A(n) (3.53)
n−1
The solution is
1
A(n) = cn(n + 1)(n − 1) , (3.54)
12
and c can be fixed by means of the special case n = 2: this gives c = D. The final result
is a famous commutation relation,
1
[Lm , Ln ] = (m − n)Lm+n + c(m3 − m)δm+n , (3.55)
12
which is called the Virasoro algebra.
32
There is another way to look at c. Remember that in the classical theory c = 0, but
that c 6= 0 is generated by quantum effects. The classical form of the algebra is changed by
quantum effects. This means that the classical symmetry is broken. When this happens,
we speak of an anomaly. The coefficient c is often called the conformal anomaly.
plus the reality condition L†n = L−n . In addition there is the zero-mode condition (3.42).
We are not claiming that (3.57) is a necessary condition for (3.56) to be satisfied,
but there is at least one alternative that can be ruled out: condition (3.56) can also be
satisfied by setting Ln |physi = 0 for n < 0. However, for c > 0 (as is the case here) this
is not an option. This can be seen as follows. Using the reality condition and (3.42) we
find
33
3.12 The Mass Shell Condition
3.12.1 Closed Strings
q
Let us first consider (3.42). For closed strings, substituting α0µ = 1 0 µ
2
αp , this yields
!
X
1 0 2
4
αp + α−m · αm − a |physi = 0 (3.58)
m>0
plus the same condition with bars on all relevant quantities. These two conditions can be
written as (note that we use the metric diag(−1, 1, 1, 1), so that particles satisfy pµ pµ < 0.)
with !
4 X
ML2 = 0 α−m · αm − a (3.60)
α m>0
and !
4 X
MR2 = ᾱ−m · ᾱm − ā (3.61)
α0 m>0
In principle â could be different than a, the shift in the closed string case, but we will see
that it is in fact the same. The factor of 4 in the overall factor in comparison to closed
strings may look like it was the result of some conventions we made in the past but it is
not. For example, we made a different choice for the range of σ in both cases, but because
of scale invariance that should not matter, and indeed it doesn’t. We also normalized the
momenta and oscillators in a certain way, but that was simply done in order to get the
correct commutation relations (3.18).
3.12.3 Tachyons?
It is essential for the particle interpretation of these states that M 2 > 0. If M 2 < 0,
the “particle” is called a tachyon, which is a general name for particles exceeding the
34
speedpof light. To see this note that for normal particles the relativistic energy is given by
1
E = p~2 + M 2 = M γ, where p~ is the 3-momentum and γ = (1 − v 2 )− 2 (with c = 1). We
can solve this relation for v, and we get v 2 = p~2 /(~p2 + M 2 ). This is smaller than 1 for a
massive particle, equal to 1 for a massless particle and larger than 1 for a tachyon, which
therefore moves faster than the speed of light. In field theory the presence of tachyons in
the spectrum indicates an instability of the vacuum. The standard field theory description
of a (free) massive scalar particle involves a scalar field φ(x) with a Hamiltonian
" #
Z 2
dφ
H = d3 x + (∇φ)2 + M 2 φ2 (3.63)
dt
If M 2 > 0 the configuration of minimal energy is φ = 0, but if M 2 < 0 the energy is not
bounded from below, and in any case the standard vacuum (corresponding to φ = 0) is
not a minimum. It is clear that the occurrence of tachyons in the spectrum is not a good
thing.
We have already seen in section 3.11 that a ≥ 0. If a = 0 the lowest state would be a
massless scalar. If a > 0 the lowest state is a tachyonic scalar.
Note that if a 6= 0 the vacuum |0i at zero momentum is not a physical state, because
p = a implies pµ 6= 0. Hence with the benefit of hindsight we see that the heuristic
2
arguments of section 3.8 are valid only for a = 0, since it was assumed that α0µ |0i =
pµ |0i = 0. However, the improved argument in section 3.11 does apply, and proves that
one cannot impose all constraints simultaneously.
35
The last method requires knowledge of path integrals and gauge fixing (the Fadeev-
Popov procedure), as well as BRST-quantization. Nowadays it is considered the most
attractive method, but too much machinery is required to explain it here.
This splitting in left- and right-moving modes is obvious for the oscillator modes, but the
center-of-mass coordinates require a bit more attention. Explicitly we have, for closed
strings
r
α0 X 1 µ −in(τ +σ)
µ 1 µ 1 0 µ
XL (τ + σ) = q + α p (τ + σ) + i (α e )
2 2 2 n6=0 n n
r (3.67)
0 X
1 1 α 1
XRµ (τ − σ) = q µ + α0 pµ (τ − σ) + i (ᾱµ e−in(τ −σ) )
2 2 2 n6=0 n n
36
In order to combine the first two terms into X + we choose bL = bR . Finally we fix the
values of the parameters in such a way that
2
X + = q + + α 0 p+ τ , (3.70)
ρ
dropping the accent on τ for convenience. We choose, as before, ρ = 2 for closed strings
and ρ = 1 for open strings, so that the final outcome is that X + looks exactly like before,
except that all the oscillators have been set to zero.
Note that this almost completely fixes conformal invariance, since bL = bR = ρ/2α0 p+
and aL + aR = −q + ρ/2α0 p+ . Using (2.33) we find for σ 0
ρ
σ 0 = 0 + XL+ − XR+ + aL − aR
2α p (3.71)
= σ + aL − aR + oscillators ,
in other words σ 0 is determined up to a constant shift aL − aR . Such a shift does not
respect the open string range 0 ≤ σ < π, but is allowed for closed strings, where we only
have a periodicity condition, σ ≈ σ + 2π. Note that the oscillator terms do not contribute
to the boundary conditions.
37
r !
2 ρ 1X i
ᾱn− = ᾱ ᾱi (3.77)
α0 2p+ 2 m,i n−m m
where the last equation only applies to closed strings This explicitly solves all modes of
the two constraints with n 6= 0.
The case n = 0√requires a slightly more careful discussion. A similar formula holds,
with α0− = ᾱ0− = ρ1 2α0 p− . However, if n = 0 we have once again an ordering problem in
the quantized theory, which we solve by normal ordering. This introduces an unknown
constant a, the same constant we saw earlier when we only considered L0 . The rest of
the discussion is slightly different for closed and open strings.
and of course the analogous expression for ᾱ. Note that in the last step we have normal-
ordered, and introduced an undetermined constant a to take into account the interchanges
of operators, as before. Bringing all momenta to the left, we get
! !
X X
α0 (2p− p+ − pi pi ) = 2 i
: α−m i
αm : −a + 2 i
: ᾱ−m i
ᾱm : −ā (3.82)
m>0 m>0
38
Note that the momenta nicely combine to a Lorentz invariant expression. One usually
writes the result as follows (it is customary to drop the normal-ordering symbols, and we
will do that from now on).
1
p2 = − (ML2 + MR2 )
2
(3.83)
! !
4 X 4 X
ML2 = 0 i
α−m αmi
− a ; MR2 = 0 i
ᾱ−m i
ᾱm − ā
α m>0 α m>0
The difference between Eqs. (3.78) and (3.79) leads to the condition
p2 = −M 2
!
1 X (3.86)
M2 = 0 αi αi − â
α m>0 −m m
39
two-dimensional theory has a mass M 2 = −â( α10 ). At the next excitation level we find a
state
i
α−1 |0i (3.87)
It has mass M 2 = (1 − â)( α10 ), and there are D − 2 of them. The index i suggests that this
state wants to be a space-time vector. Covariantly a space-time vector has D components.
Not all these components are physical. If a D-dimensional particle is massive, we can go
to its rest frame. We know then that it has D − 1 physical components, corresponding
to its polarizations. If a particle is massless, we can not go to its rest frame, but we
can Lorentz-rotate its momentum to the form (E, k, 0, . . . , 0), with k = E. The physical
components correspond to the rotational degrees of freedom in the last D −2 components.
There are thus D − 2 physical degrees of freedom.
This situation is well-known in electrodynamics. A photon Aµ starts with D degrees
of freedom, but one is eliminated because of the equation of motion for A0 , which contains
no time derivative and hence is just a constraint (∂i E i = 0, also known as Gauss’ law),
whereas another one is eliminated by gauge invariance. The equation of motion for a
vector boson of mass M is
∂µ F µν = M 2 Aν , (3.88)
which has gauge invariance Aµ → Aµ + ∂µ Λ only for M = 0 (whereas the non-dynamical
nature of A0 holds independent of M ).
Turning the argument around, the fact that we have here D − 2 components strongly
suggests that this state must be a massless vector boson. This means that â must be
equal to 1.
40
particles of spin 0, 1, 2, etc. We know that in quantum mechanics also half-integer spins
can occur. This is related to the fact that in quantum mechanics the relevant group is not
SO(3) but its covering group SU (2). But in this chapter we will not encounter particles
with half-integer spin.
Massless particles correspond to representations of SO(2). They are labeled by a
number called the helicity, the projection of the spin along their direction of motion.
In principle, the helicity could be any real number, but since the little group must be
a subgroup of the rotation group SO(3) only integers and half-integers can occur. In
principle, a single helicity is already a valid little group representation, but if we also
want space reflection to be a symmetry of nature, one must have both helicity states.
For half-integer spin particles it is possible to violate space-time reflection symmetry and
have only a single helicity state. A left-handed neutrino might be an example (it is not
known at present if right-handed neutrinos exist). For integer spin particles this is not
possible; they always have two helicity states.
The generalization to D > 4 is straightforward. One looses the notion of “spin” as a
single number labeling representations. This is because irreducible SO(D−1) or SO(D−2)
representations are not labeled simply by integers, as is the case for D = 4. A general
enumeration of such representations requires N/2 non-negative integers for SO(N ) and
SO(N + 1), N even. We will not attempt to do that here, because we are only interested
in the lowest excited states, and hence we only need a few representations.
The three components are called the rank-2 traceless symmetric tensor, the rank-2 anti-
symmetric tensor and the singlet. Since they cannot be split further, they are called
irreducible representations. We will not prove the irreducibility here, but it is easy to
verify that the three components transform into themselves.
41
The number of components of these tensors is respectively 21 N (N + 1) − 1 = 21 (N +
2)(N − 1), 12 N (N − 1) and 1. Note that the first component, the traceless symmetric
tensor, is also invariant if instead of N2 we subtract the trace with another factor. The
reason for the choice of coefficient is that it makes the tensor traceless. This implies a
linear relation among the 21 N (N + 1) components, reducing the number of independent
ones by 1.
By continuing along these lines one can construct irreducible tensor representations
of arbitrary rank (the number of indices i, j, . . .). The general procedure is to impose a
definite symmetry under interchange of indices and remove traces as much as possible.
However, this quickly gets complicated. For rank larger than 2 there are not just sym-
metric and anti-symmetric tensors, but also tensors of mixed symmetry. Furthermore
there are restrictions on the rank of an anti-symmetric tensor, depending on N : clearly
in SO(N ) one cannot have rank N + 1 anti-symmetric tensors. There are more efficient
methods to construct representations, but we will not need them. Later we will also need
another type of representation, the spinor representation, which cannot be obtained from
tensor products.
42
What are these massless fields? The symmetric tensor can be identified with the
graviton. The scalar is called the dilaton, and in addition there is the antisymmetric
tensor Bµν . Experimentally only the first has been observed (or more precisely, its effects
have been observed in the form of gravity). There is no evidence for the other two. In
four dimensions, the field Bµν can be transformed into a pseudo-scalar (a scalar that is
odd under parity), which is often called an axion.
Substituting the quantum mode expansion for X i without interchanging any operators
one gets (for open strings)
∞
D−2 D−2
!
1 XX i 1 XX X
H= 0 α αi = 1 i
: α−m i
αm : + 21 (D − 2) n (3.95)
2α i=1 m −m m α0 2 i=1 m n=1
The sum diverges: it is a sum over the zero-mode energy of an infinite number of oscilla-
tors! However, if we define
X∞
ζ(s) = n−s , (3.96)
n=1
then the sum converges for Re s > 1. This function is the Riemann zeta-function. It can
be uniquely continued to s = −1, and the result is
1
ζ(−1) = − (3.97)
12
Hence we conclude
1
a = (D − 2) , (3.98)
24
and since we have already concluded that a = 1, this can only work if D = 26. This is
called the critical dimension.
43
3.17 The Lorentz Algebra∗
A better way to arrive at the conclusion that the space-time dimension must be 26 is to
consider closure of the Lorentz algebra. In general, one has a set of generators M µν that
must satisfy the commutation relation
[M µν , M ρσ ] = −iη νρ M µσ + iη µρ M νσ + iη νσ M µρ − iη µσ M νρ (3.99)
One can work out the precise form of these quantities in string theory. The result is
∞
µν µ ν ν µ
X 1 µ ν ν µ
M =q p −q p −i (α α − α−n αn ) (3.100)
n=1
n −n n
(plus an ᾱ contribution for closed strings). It is not too difficult to check that this does
indeed satisfy the commutation relation (3.99).
In light-cone gauge, M µν split into components M ij , M i− , M i+ and M +− . The oscil-
lators αn− can be expressed in terms of a quadratic relation in αi (see (3.76)). This makes
M i− tri-linear in αi . The computation of the commutation relations of these operators
M i− with the others is technically of course totally different than before, but the result
should be the same.
The hardest one to check is [M i− , M j− ], which should be zero. However, after some
tedious work one finds
∞
i− j− 1 X 26 − D 1 D − 26 i j j i
[M , M ] = − + 2 m + + 2(1 − a) (α−m αm −α−m αm )
(p ) m=1 12 m 12
(3.101)
This vanishes if and only if a = 1 and D = 26.
Note that this also implies that if we choose these values, then the Lorentz algebra
closes on the sets of states. In particular, all the states must then fall into Lorentz
multiplets. For example, in the open string theory we found at the first excited level a
symmetric tensor plus a vector of SO(D − 2). These nicely combined into a traceless
symmetric tensor of SO(D − 1). Such “miracles” have to occur at any level, or else
Lorentz invariance is violated. The closure of the Lorentz algebra guarantees this.
σ 0 = 2π − σ (closed strings)
(3.102)
σ0 = π − σ (open strings)
These transformations map the world sheet into itself, but with a sign change of σ.
Oriented strings are not invariant under this transformation, whereas unoriented strings
– by definition – are invariant.
44
When one applies this to the mode expansions for open and closed strings one readily
discovers that their effect is equivalent to the following transformation of the modes
On the states we then impose the requirement that they be invariant under the action
of Ω. Now everything is fixed if we just give a rule for the action of Ω on the vacuum.
The correct choice is Ω |0i = |0i (note that choosing a −sign here removes the vacuum
from the two-dimensional spectrum, and hence it removes the tachyon from the space-
time spectrum. Although this looks attractive at first sight, it leads to inconsistencies).
i
Then at the first excited level of open strings the states α−1 |0i is transformed into minus
itself by Ω, and therefore is not in the spectrum. At the first excited level of the closed
i j
string the states α−1 ᾱ−1 |0i is transformed into the states with i and j interchanged, and
hence only the symmetric part of the tensor survives. Hence we only get a graviton and
a dilaton, but no anti-symmetric tensor.
This can all be summarized as follows
One should remember that open strings cannot exist by themselves, but always require the
existence of closed strings ((un)oriented open strings require (un)oriented closed strings).
45
and the boundary at σ = π labeled by “b”. Hence we add the two labels to the vacuum
state: |0; a, bi.
If the open string is oriented the two labels are independent, but consistency requires
them to have the same multiplicity N . Then we find N 2 distinct vacuum states, and at
the first excited level we find N 2 vector bosons. In the free string theory we can only
count them, but when we make the theory interaction we can find out more. It turns out
that these N 2 vector bosons must be gauge bosons of a U (N ) gauge group.
If the string theory is unoriented, we have to remember that the world sheet parity
transformation interchanges the two boundaries, and hence the labels a and b. Then the
invariant vacuum states are the symmetric linear combinations
1
√ (|0; a, bi + |0; b, ai) (3.105)
2
This is a total of 12 N (N +1) states. At the first excited level one has the following invariant
states
1 i i
√ (α−1 |0; a, bi − α−1 |0; b, ai) (3.106)
2
This is anti-symmetric, so there are 12 N (N − 1) vector bosons. They turn out to form a
gauge group O(N ).
Note that the unoriented open string has gauge bosons now. Previously we were
looking at the special case N = 1, and since O(1) is trivial but U (1) is not, we found no
vector bosons for the unoriented open string.
There is one additional possibility if N is even. Then, instead of defining Ω |0; a, bi =
+ |0; b, ai the definition Ω |0; a, bi = − |0; b, ai. is also allowed. The vacuum state is then
1
√ (|0; a, bi − |0; b, ai) , (3.107)
2
which is just a single state for the minimal choice, N = 2 At the first excited level, the
states are
1 i i
√ (α−1 |0; a, bi + α−1 |0; b, ai) , (3.108)
2
which are 21 N (N +1) vector bosons. It turns out that they are gauge bosons of a symplectic
group, Sp(N ).
3.20 Discussion
So far we have looked at free strings. That is to say, we have considered world sheets that
are either an infinite cylinder or an infinite strip. Interactions are taken into account by
allowing world sheets with holes, corresponding to strings splitting and joining, or with
semi-infinite tubes attached to the surface (external closed string states) or semi-infinite
strips attached to the boundaries (external open string states). Hence what we know so
far gives only limited information, but let us look at it anyway.
46
The main good feature that all string theories seen so far have in common is the
existence of a graviton gµν in the spectrum. To be precise, we see that there is a massless
symmetric tensor, which is a candidate for a graviton. To see that it really is a graviton
we need to look at the interactions.
It is known, however, that there is not much choice in making a massless symmetric
tensor interact in a Lorentz-covariant way. The only possibility is Einstein’s theory of
gravity (with possibly higher curvature corrections). Hence either string theory can be
made interacting in a consistent way, and then it must produce general relativity, or it is
inconsistent. The former possibility turns out to be the correct one.
For example, one may compute the three-graviton interaction in string theory and
compare with the analogous computation in general relativity. As expected, the form of
the interaction agrees, and one gets a relation between the coupling constant of gravity –
Newton’s constant – and the only parameter in the string action, α0 .
This is somewhat misleading since we are in 26 dimensions, and not in four. Newton’s
law of gravity in D dimensions is
m1 m2
F = GN (3.109)
rD−2
1
which gives the famous r2
force in 4 dimensions. It follows that the dimension of GN is
dimension dependent:
If we work with the convention c = ~ = 1, so that [t] = [l] = [m]−1 , we get that
[GN ] = [l]D−2 .
The parameter dimension of α0 is determined as follows. X clearly has the dimension
of length, and an action has the same dimension as ~ (so that exp(iS/~) makes sense),
which is 1 in our case. The dimension of the world sheet parameters σ and τ is irrelevant,
since it cancels. Hence α0 has the dimension of [l]2 .
From this analysis we learn that it is inevitable that GN is given by
1
GN = (α0 ) 2 (D−2) × dimensionless constants (3.111)
0
If we manage to construct √ strings in four dimensions, we would expect GN ∝ (α ).
Conversely, 0
√ this implies the α is of order the Planck length, or the fundamental mass
scale 1/ α0 is of order the Planck mass.
We have seen that the masses of all the excited states are multiples of this parameter,
and hence inevitably all excited states have masses that are of order 1019 GeV. These
particles are therefore unobservable.
The only observable string states are the massless ones (assuming we manage to get
rid of the tachyon). Of course very few massless particles have been observed. In fact
the photon is the only one that has really been, the gluons of QCD exist, but cannot be
observed as free particles, and the graviton probably exists as well. Neutrinos were long
conjectured to be massless, but we know that at least two of them (and presumably all
47
three) have a mass. However, the standard model has a symmetry breaking mechanism,
the Higgs mechanism, and we only see the broken phase. In the unbroken theory all
particles (except the scalars that become the Higgs boson) are massless.
The idea is now that this unbroken theory corresponds to the massless particles in
string theory. The origin of the Higgs mechanism and the smallness of the scale at which
it occurs remains a mystery. The smallness of the ratio MW /MPlanck ≈ 10−17 is a mystery
independent of string theory, and at present string theory offers us no solution to this
problem.
If we forget this problem, let us compare the massless string spectrum to what we
observe:
The last column lists the solutions to the mismatch in the first two columns. This
is also a preview of some of the subjects we still have to discuss. Here is very brief
explanation, to be made more detailed later.
• Fermionic strings are string theories which have, in addition to the bosons X µ ,
also two-dimensional fermions. It can be shown that they always contain space-
time fermions in their spectrum. Hence they can give rise to quarks and leptons.
Fermionic strings have a critical dimension of 10, and they may or may not have
tachyons.
• Superstrings are fermionic strings with space-time supersymmetry.
• An axion is a pseudo-scalar that was proposed some time ago to solve certain prob-
lems with QCD. The B-particle is a pseudo-scalar, and hence could perhaps play
the rôle of the axion.
• Supersymmetry is a symmetry between fermions and bosons. So far it has not
been observed in nature. Supersymmetry implies absence of tachyons (but is not
necessary for that).
• Compactification means that some of the space-time dimensions are rolled up into
a small space, so small that it is unobservable to us.
48
• Moduli are changes in the shape of the compactification space. They lead to massless
scalars in the spectrum.
• “Low-E mass” means that by some mechanism these particles are assumed to get a
mass so that they are not (yet) observable. This mass is not directly generated at
the Planck scale, so it must be due to some low energy mechanism. The definition
of “low energy” in this context is anything below the Planck scale.
4 Compactification
The idea of space-time compactification precedes string theory by many decades. already
in the first half of the century Kaluza, Klein and also Einstein considered the idea that
space-time is higher-dimensional.
p2 = pµ pµ + (p4 )2 = −M 2 (4.1)
Now p4 is not on equal footing with the space-time momenta pi . Wave functions must
respect the periodicity:
Ψ(xµ , x4 ) = Ψ(xµ , x4 + 2πR) (4.2)
A free particle wave function,
mx
Ψ = eip m
(4.3)
is only periodic if p4 is quantized:
n
p4 = , n∈Z (4.4)
R
It makes then sense to put (p4 )2 on the other side of the equal sign in (4.1). Then we
get
pµ pµ = −(p4 )2 − M 2 (4.5)
Hence a four-dimensional observer sees a discrete spectrum of particles of mass-squared
M 2 + n2 /R2 .
49
4.2 Space-time Compactification for Strings
We can do the same in string theory. If we do this for closed strings there is an additional
possibility. The string can wrap a couple of times around the compact dimension:
X µ (τ, σ + 2π) = X µ (τ, σ) + 2πmRµ (4.6)
where Rµ = Rδ µ,25 if we compactify the 25th space dimension. Now µ = 0, . . . , 24, so
that we are assuming that we are in 25 space-time dimensions. This will turn out to
be the critical dimension. The rule is that the sum of compactified plus uncompactified
dimensions must be 26.
If we impose this periodicity we must re-derive the string mode expansion. This is
very easy, and results in
r
α0 X 1 µ −in(τ +σ)
µ µ 0 µ µ µ −in(τ −σ)
X (σ, τ ) = q + α p τ + mR σ + i (α e + ᾱn e ) , (4.7)
2 n6=0 n n
and r r
α0 µ α0 X 1 µ −in(τ −σ)
1
XRµ = qµ + p (τ − σ) + i (ᾱ e ) , (4.9)
2 2 R 2 n6=0 n n
with
1 √ 0 µ
1
pµL=√ αp + √ mR µ
2 α0
(4.10)
1 √ 0 µ
µ 1 µ
pR = √ αp − √ mR
2 α0
50
Where, as above, only the compactified components of pL and pR are added to the mass.
When going to light-cone gauge one now uses the components 0 and 24 to define X + and
X − , and one checks Lorentz invariance only in the uncompactified dimensions (since it
is manifestly violated in the compactified ones). The computation is essentially identi-
cal, and yields the result mentioned above, namely that the total dimension (compacti-
fied+uncompactified) must be 26. The constant a comes out as 1, as before.
The quantization of the momenta in the compactified directions is exactly as for par-
ticles. Hence we get
r
α0 n mR
pL = + 0
2 R α
r (4.13)
α0 n
mR
pR = − 0
2 R α
Here we use the notation |pL , pR i. In addition it may happen that (pL )2 and/or (pR )2 is
precisely equal to 2. In that case there are extra massless particles:
i
α−1 |0, pR i −→ vector boson
25
α−1 |0, pR i −→ scalar
i
ᾱ−1 |pL , 0i −→ vector boson (4.15)
25
ᾱ−1 |pL , 0i −→ scalar
|pL , pR i −→ scalar
It turns out that the vector bosons are in fact gauge bosons of some new gauge in-
teraction. This is a second way to obtain gauge symmetry in addition to Chan-Paton
labels.
4.4 T-duality
The spectrum we just found has an interesting symmetry
α0
R ←→
R (4.16)
n ←→ m
51
Under this transformation pL is unchanged and pR changes sign, but this has no influence
on the spectrum.
It turns out that this remains true even for the interacting string theory. This is
remarkable, because it says that strings compactified on a very large circle are equivalent
to strings compactified on a very small circle. Or, in other words, one cannot do any
experiment to determine if the internal compactification manifold is large or small. This
is counter-intuitive, but there is a simple reason why this happens: the momentum and
winding states are interchanged. At small R, the momentum states are very widely
spaced, but it costs very little energy to wind the string around the circle. Therefore
winding states have very little spacing. If we replace R by α0 /R it is just the other
way around. Now momentum states take over the role of winding states and vice-versa.
Experimentally winding and momentum states cannot be distinguished.
momentum
winding
If you take this to the extreme limit R → ∞ it leads to the conclusion that flat space is
equivalent to a point! It is however not very clear what this would mean.
This phenomenon is know as T -duality. Although we have seen it here in a very special
system, it is seen in all compactifications, and is believed to be a fundamental property of
string theory. It seems as if short distances (smaller than the Planck scale) are equivalent
to long distances. In other words, there seems to be a fundamental shortest length scale,
the Planck scale. If one probes smaller distances, nothing new is found, but one simply
sees the same phenomena one has seen at larger distances already.
This is also related to the absence of short-distance singularities (“ultra-violet diver-
gences”) in string loop calculations. What happens is that all integrals are cut off in a
natural way at the Planck scale. Integrating over momenta larger than the Planck mo-
mentum would be double counting: those contributions already are taken into account in
the integral over small momenta.
52
√
this in p2L = 2 leads to the condition R = |n| α0 . Then we√may compute m, and we find
m = n1 . Hence the only possibility is n = m = ±1, R = α0 . Note that the T -dual of
this theory is the same theory. This is therefore called the self dual point on the space
of compactified string theories. In the self-dual point there are precisely two states of
i i
the form α−1 |0, pR i (namely (n, m) = (1, −1) or (−1, 1)) and two of the form ᾱ−1 |pL , 0i
(namely (n, m) = (1, 1) or (−1, −1)).
One may expect these states to correspond to gauge bosons. Then what is the cor-
responding gauge group? If we are not in the self-dual point there are just two vector
bosons, and then there is not much choice; it can only be U (1) × U (1). In the self-dual
point there are six vector bosons, and then the only possibilities are U (1)6 , SU (2) × U (1)3
or SU (2)2 . The latter turns out to be the correct one, but to decide that one would have
to study the interacting string.
53
The presence of a massless spin-2 particle then implies that there must be a corre-
sponding field theory. This is in fact general relativity. Likewise, a massless spin-1 particle
implies a gauge theory.1 These field theories are usually called low-energy effective field
theories. The word “low-energy” refers to the fact that we are ignoring massive string
excitations. It might be possible to include them in the field theory, but since there are
infinitely many that is not practical. Since massive excitations are ignored, any field the-
ory description must fail once we reach energies as big as the string scale (i.e. the Planck
scale. The word “effective” means that it is a good substitute for the real thing, which is
string theory. For example, one can have an effective theory of pion interactions, which
is not as fundamental as QCD but can give good results in certain approximations.
Once we have made this step, we can try to interpret what we have found in terms
of the effective low energy field theory. The relevant field are the graviton, the anti-
symmetric tensor and the dilaton. These massless fields can be decomposed in terms of
25-dimensional fields. This works as follows:
gM N AM
gµν → (4.17)
AM ξ
BM N BM
Bµν → (4.18)
−BM 0
Here µ = 0 . . . 25 and M = 0 . . . 24. We see that the graviton decomposes into the
25-dimensional graviton plus a vector boson and a scalar, whereas Bµν yields BM N plus
another vector. Hence we get two massless vector bosons and a massless scalar ξ, in agree-
ment with what we found. The dilaton in 26 dimensions just yields the 25-dimensional
dilaton. On top of this we get all the momentum and winding excitations of these parti-
cles.
In a Kaluza-Klein compactification one only has the momentum modes of string theory.
They can be seen in the effective field theory by a mode-expansion in the compactified
coordinates, e.g.
X X
ξ(xM , x25 ) = ξn (xM ) cos(2πnx25 /R) + ϕn (xM ) sin(2πnx25 /R) (4.19)
n n
The fields ξn (xM ) and ϕn (xM ) (n 6= 0) are 25-dimensional massive scalar fields, and ξ0 is
a massless field. The winding modes of string theory however can never be found in this
way. Here we see the limitations of effective field theory.
54
A circle was specified earlier by means of a single vector Rµ . The appropriate gener-
alization is to introduce N such vectors (linearly independent, of course) if you want to
compactify N dimensions. The meaning of this set of vectors is that points in space-time
that differ by any such vector are considered identical. If this identification is done in one
direction, space gets rolled up into a cylinder. If we add a second direction, the cylinder
gets rolled up into a torus. Beyond this we cannot even draw the resulting figure anymore,
but it is not hard to describe mathematically.
Obviously, if xµ is the same point as (i.e. is identified with) xµ + 2πR1µ , but also the
same point as xµ + 2πR2µ , and if this is true for any xµ , then it follows that xµ is identified
with xµ + 2πR1µ + 2πR2µ . We see then that the set of vectors Riµ must close under addition.
Such a set of vectors is called a lattice.
A lattice can always be described by choosing a set of basis vectors ~ei , i = 1 . . . N ,
such that any vector ~λ on the lattice Λ can be written as
N
X
~λ = ni~ei (4.20)
i=1
The allowed momenta must have integral inner product with all these lattice vectors in
order for wave functions to have to correct periodicity. A necessary and sufficient condition
for this is
p~ · ~ei ∈ Z for all i (4.21)
This is easily seen to imply that the momenta p must themselves lie on a lattice, the
so-called dual lattice Λ∗ . (This situation also occurs in solid state physics, where such a
lattice is called reciprocal lattice; however string theorists like the word “dual”).
In this general situation the mass formula stays essentially the same, but pL and pR
are higher-dimensional vectors of the form
r
α0 1
p~L = ρ + 0 ~λ)
(~
2 α
r
α0 1 (4.22)
p~R = ρ − 0 ~λ)
(~
2 α
where pL and pR are always the compactified components (for uncompactified ones the left-
right splitting is not very useful anyway), and N is the sum over the “number operators”
55
of the uncompactified and compactified oscillators, i.e.
24 X
X
i i
N = α−m αm , (4.24)
i=1 m>0
Lattices with this property are called Lorentzian integral. The adjective “Lorentzian”
refers to the fact that there is a minus sign: this can be reproduced by introducing a
Lorentzian metric diag (1, . . . , 1, −1, . . . , −1), with N positive and N negative entries.
We then treat (~pL , p~R ) as a 2N dimensional vector in a space with this metric. The
lattice vectors also satisfy
(~pL )2 − (~pR )2 ∈ 2Z (4.26)
Such a lattice is called Lorentzian even (actually this already implies Lorentzian integral,
but not the other way around). Finally we can define a Lorentzian dual to a lattice: the
Lorentzian dual Γ∗ contains all vectors that have integral inner product with all vectors
in Γ, where we use the Lorentzian inner product defined above.
One can easily show that the lattice Γ constructed above has the property that Γ = Γ∗ .
Such a lattice is called a Lorentzian self-dual lattice. We see thus that Γ is a Lorentzian
even self-dual lattice. A compactification on the corresponding torus is called a Narain
compactification. It can be shown (and we will return to this in chapter 5) that any such
compactification yields a consistent string theory.
The point is now that there exist lattices that are Lorentzian even self-dual, but are
not of the form (4.22). The generalization has in fact the following form (here we denote
the components of the vectors with an index I = 1, . . . N ; the original 26-dimensional
index µ is split into (M, I), with the uncompactified space-time index M running from 0
to 26 − N − 1)
r
I α0 I 1
pL = (ρ + 0 (λI − B IJ λJ ))
2 α
r
0
α I 1 (4.27)
pIR = (ρ − 0 (λI + B IJ λJ )) ,
2 α
1
The compactification discussed here is the general solution to the condition of “modular invariance”,
which will be discussed in section 5.
56
where B IJ is an arbitrary anti-symmetric matrix. Upper and lower indices have no special
significance here.
It is interesting to count parameters. The anti-symmetric matrix B IJ has 12 N (N − 1)
real parameters. The lattice Λ we started with is completely specified by the lengths and
relative angles of the basis vectors. This defines a matrix
X
gij = eIi eIj (4.28)
I
57
can write this action back into the old basis, with a diagonal space-time metric. The
advantage of that basis is that it is more convenient for computing the spectrum, and
indeed, that is the basis we used for previous calculations. If we define
X
Bij = eIi eJj BIJ (4.32)
IJ
we get Z
1 ab I J ab I J
S[X] = − dσdτ η ∂ a X ∂b X δIJ + ∂a X ∂ b X BIJ , (4.33)
4πα0
It may seem that this new term forces us to start from the beginning with the derivation
of the equations of motion and the whole quantization procedure. But (for constant BIJ
at least) the effect is minimal. The extra term in the equations of motion is proportional
to ∂a ab ∂b X I . But this vanishes since the two derivatives commute.
The reason this term vanishes is that the extra term in the action is a total derivative:
When the the action is integrated over a surface without boundaries the contribution of
such a term vanishes. Therefore it also does not affect the equations of motion, which
describe variations of the action.
But then one may wonder how it can have any effect at all. The reason is that the
canonical momenta do change. We find
1
ΠI = (δIJ ∂0 + BIJ ∂1 ) X J (4.35)
2πα0
It is thus this quantity that satisfies the canonical quantization rule (3.17).
It is not difficult to see that the extra term does not affect the quantization of the
oscillator modes, but there is a change in the commutators of the zero modes q and p.
The effect of the extra term is, using (4.7), that pI is replaced by π I = pI + α10 B IJ λJ (~λ
is the higher dimensional generalization of mRµ in (4.7).
The expression for the mass spectrum is still (4.23), with pIL expressed in terms pI
and λI exactly as before. But the canonical momentum is π I , not pI , and hence the
quantization condition is that the eigenvalues of the operator π I must lie on the dual
lattice. Then it is more useful to express pIL and pIR in terms of the canonical momenta.
This gives
1 √ 0 I 1 √ 0 I
I 1 I 1 I IJ
pL = √ αp + √ λ = √ α π + √ (λ − B λJ )
2 α0 2 α0
(4.36)
1 √ 0 I 1 √ 0 I
I 1 I 1 I IJ
pR = √ αp − √ λ = √ α π − √ (λ + B λJ )
2 α0 2 α0
If we now put the eigenvalues of π on the dual lattice we do indeed recover (4.27).
58
4.10 Moduli∗
Combining what we saw in section 3.6 and in the previous one, we arrive now at the
following picture. In a compactification to D = 26 − N dimensions, the 26-dimensional
metric decomposes as follows
gM N AIM
gµν → (4.37)
AIM ξ IJ
I
BM N BM
Bµν → I (4.38)
−BM ζ IJ
Here AIM and BM I
are two sets of N vector bosons each, and ξ IJ and ζ IJ are two sets
of 2 N (N + 1) and 21 N (N − 1) scalars. All these quantities are fields in D = 26 − N
1
dimensions, i.e. they are functions of the coordinate xM . They are not functions of the
compactified coordinates, however. The dependence on the compactified coordinate y I
has been mode expanded, as in the case N = 1 discussed above.
We observe now that the parameters of the compactification can be viewed as vacuum
expectation values of the massless scalars in D = 26 − N dimensions.
In general, a classical field theory is quantized by expanding around a certain classical
background: φ = φcl + φqu . The quantum field is expanded in modes, and the vacuum
state is chosen in such as way that h0| φqu |0i = 0, so that h0| φ |0i = φcl . Then one calls
φcl the vacuum expectation value of φ. Also the term background field is often used for
φcl .
If we want a theory to respect certain symmetries, then we must require that at
least the classical values of the fields, φcl , respect it. For example translation invariance
requires that φcl is constant. Lorentz-invariance requires that a vector cannot have a
vacuum expectation value, but a scalar can (whereas for a symmetric tensor the v.e.v.
must be proportional to ηµν ).
A scalar field can get a vacuum expectation value without breaking Poincaré invari-
ance. This is precisely what is occurring here. Consider the coupling of the scalars ξ and
ζ to the string coordinates X. For convenience, we use a choice of variables so that the
string coordinates are periodic on a rectangular lattice, as in Eq. (4.31). Then ξ and ζ
couple as follows
Z
1 n
ab i j ab i j
o
S[X] = − dσdτ η ∂ a X̂ ∂ b X̂ ξij + ∂ a X̂ ∂ b X̂ ζij , (4.39)
4πα0
We may expand the scalar fields as follows
We see now that all parameters of the torus compactification can be interpreted as vacuum
expectation values of scalar fields.
This is not limited to torus compactification. It is often expressed by saying that
string theory does not have free parameters. All freedom is due to changing the vacuum.
59
There is no parameter one fixes “by hand”, like the quark masses or the standard model
couplings.
The difference is mainly philosophical. If the parameters are related to v.e.v’s of fields,
there is a hope that they can be determined dynamically. Maybe one set of parameters
is energetically favorable. If a parameter is in the Lagrangian one starts with, there is no
hope to determine it from the theory itself.
Note that also the string coupling constant is not a parameter, but related to the
v.e.v. of the dilaton. One might object that α0 is a parameter, but that is not true. It is
a quantity with a dimension, and changing it just changes all dimensionfull quantities in
the same way. If we make all the masses in our universe twice as heavy, and also change
Newton’s constant accordingly, this can never be observed. Only dimensionless ratios are
true parameters.
The continuous parameters that relate different compactified strings are often called
moduli. The name comes from mathematics, where it is used for parameters that describe
deformations of a manifold. The space formed by the moduli is called moduli space. For
circle compactifications in field theory it is just a half-line R ≥ 0, corresponding to the
value of the radius R. In string theory it gets a more interesting structure due to T-
duality, which identifies two parts of the half-line. But even if one compactifies just one
dimension in string theory there is already more structure.
60
V is constant (furthermore m = 0, because we consider the massless scalars ξ and ζ).
But then all solutions must be equivalent. [We didn’t really show that the equations of
motion are satisfied for all radii R, but this can be seen in various ways. In the low-energy
field theory limit, the equations of motion for ξ are just Einstein’s equations, and any flat
space satisfies those equations. A torus is a flat space.]
However, there is a problem with this point of view. We have a tachyon T , which
is a scalar with m2 < 0. This means that the point hT i = 0 is an instable point in
the tachyon potential. If we knew the full potential V (T ) we could try to determine the
correct minimum, but we do not know it. The tachyon couples to all other particles, and
in its presence it makes no sense to try to discuss the minima of the moduli potentials.
We present this argument here because it foreshadows an important feature. The
potential picture is inadequate for bosonic strings, but this will be different in superstrings
(string theories with fermions and a boson-fermion symmetry in space-time). There the
moduli are in fact flat directions in the potential (and this can be shown also in the
presence of interactions), and there is no tachyon.
Note in particular that all these compactifications can have different gauge groups.
We have seen at the end of section 4.3 that in special cases there can be additional vector
bosons. This implies that there exist special vacuum expectation values of ξ and ζ, with
a corresponding special lattice, for which the gauge group is enlarged.
Hence we get an enormous vacuum degeneracy: there is a huge set of possible string
compactifications, with different gauge groups, and no principle that selects one over any
other. Perhaps one may dream of deriving one day the Standard Model gauge group
SU (3) × SU (2) × U (1) in this way. At this point too many ingredients are still missing
to start speculating about that, but if it comes from one of these torus compactifications,
the potential will not select it.
This flat potential is unacceptable for other reasons. For example, it implies the
existence of a large number of unobserved massless scalar fields. These scalars are usually
named after the mathematical deformation parameters to which they are related: moduli.
These moduli are a major problem. Somehow from higher order perturbative or non-
perturbative quantum corrections a potential must be generated for them, so that there
are local minima, or at least one of them. Although generating a potential is not that
hard, often one finds that their minima are at infinity in field space. The fields ξ and ζ
then run off to infinity, and the compact space decompactifies. The problem of generating
potentials that do not have such problems is known as the “moduli stabilization problem”.
One may hope that one day we will find that the potential has just a single global minimum
corresponding to the Standard Model, but this is just wishful thinking. Everything we
know about this problem suggests that if there exists one local minimum there will be
many more. This would imply that it is impossible to derive the Standard Model gauge
group in string theory, since there are many alternatives that theoretically are equally
good.
61
4.12 Orbifolds∗
There are many other compactifications one might consider. A very popular one is the
orbifold. We will just sketch the procedure for the one-dimensional case. We start with
the compactification on a circle with radius R, and the declare that all points related by
a reflection are identified. This imposes first of all a condition on the closed string states.
Only string states that are invariant under X → −X can live in such a world. It is easy
to see that this eliminates, for instance, all states created by an odd number of oscillators.
In addition one now has to allow closed strings that begin and end at identified points.
This gives new states in the Hilbert space. This part of the Hilbert space is called the
twisted sector.
Twisted strings
Identification
plane
Since this can be done for any circle radius, one obtains a second infinite half-line
of solutions, with again a T-duality relating small R to large R. For one compactified
dimension (relevant for 25-dimensional strings) the full moduli space is known,
q and there
2
are two surprises. They can be seen in the following picture (here N = R α0
).
62
Orbifold
Isolated
theories
Circle
(self-dual
point)
The first is that the orbifold line and the circle line meet each other. This means that
an orbifold theory (in fact the orbifold of the self-dual circle theory) is in fact equivalent to
a circle theory at a different radius. This is a non-trivial fact that can only be seen in string
theory (indeed, orbifolds are singular as manifolds, and a field theory compactification on
such a manifold is not even well-defined). It is again a kind of duality: two seemingly
different theories are in fact equivalent. The second surprise is the existence of three
isolated theories that have no moduli at all. These were discovered by P. Ginsparg.
63
important rôle in the discussion of the foregoing sections, although the lightcone gauge
derivation of the spectrum does not really show very clearly that these symmetries are
essential. But they are, in the sense that all important results – and in particular the
appearance of gravity – remain true provide we make sure that these symmetries are
present in the theory.
Often one does not write explicitly reparametrization invariant theories but instead one
works directly in conformal gauge, and demands that the theory has conformal invariance.
An rather trivial example is the compactified string. The world sheet action is1
Z
1 ab
µ I
S[X, γ] = − dσdτ η ∂ a X ∂ b X µ + ∂ a X ∂b X I (5.1)
4πα0
Since X I lives on a torus and X M does not, 26-dimensional Poincaré invariance is broken
to D-dimensional Poincaré invariance. But the theory still has conformal invariance (and
it can be written in a reparametrization invariant way be re-introducing a world-sheet
metric γαβ ).
In this example the compactified part of the theory (often called the internal theory)
is still written in terms of free bosons X I , but this is not necessary. Just as in four
dimensions, there are other field theories than those made out of free bosons. In general
any two-dimensional field theory with conformal invariance is acceptable. This kind of
theory is called a conformal field theory, usually abbreviated as CFT. Such theories exist
also in more than two dimensions, but in particular in two dimensions a lot is known, and
many such theories can be constructed. Hence in general one has
Z
1 ab µ
S[X, γ] = − dσdτ η ∂ a X ∂ b X µ + some conformal field theory (5.2)
4πα0
It can be shown that this relation is equivalent to having classical conformal invariance.
One can work out the commutator of two such operators in the quantum theory, and in
general the result is
c
[Lm , Ln ] = (m − n)Lm+n + (m3 − m)δm+n . (5.4)
12
1
From now on we use indices µ and ν for D-dimensional space-time; There will be no need anymore for
an index that covers compactified and uncompactified components simultaneously.
64
The first term is what one expects on the basis of classical/quantum correspondence. We
have seen before that if one allows an extra term in the algebra, the central charge term,
it must have the form shown in (5.4).1 The extra term has no classical correspondence,
and can be interpreted as a violation of the classical symmetry by quantum effects. Such
an effect is generally called an anomaly. Since the symmetry affected by it is the con-
formal symmetry, c is often called the conformal anomaly. It may also be thought of
as counting the number of degrees of freedom of a theory (although it is not necessarily
integer). Indeed, for a free boson theory with N bosons the conformal anomaly is N . The
consistency requirement is now that the total conformal anomaly must be 26, i.e.
D + cint = 26 , (5.5)
65
In D = 2n dimensions, the Dirac matrices are 2n × 2n matrices. Hence in two dimensions
we get 2 × 2 matrices, and it is easy to show that the following set satisfies the anti-
commutation relation
0 0 −i 1 0 i
ρ = ρ = (5.8)
i 0 i 0
Analogous to four dimensions,
ψ̄ = ψ † ρ0 . (5.9)
The representation for the Dirac matrices chosen here is the so-called Majorana rep-
resentation. In this representation the Dirac operator iρa ∂a is real, and hence we may
consistently impose the condition that ψ A is real. This is called a Majorana fermion.
Since it is real, we have in fact
ψ̄ = ψ T ρ0 . (5.10)
The Dirac equation is
iρa ∂a ψ A = 0 (5.11)
Writing this in matrix notation we get
A
0 ∂0 − ∂1 ψ−
=0 (5.12)
−∂0 − ∂1 0 A
ψ+
5.4 Quantization
The quantization of the fermionic action works similar to that of bosons, the main dif-
ference being that instead of commutators we get anti-commutators. The canonical mo-
mentum of a fermion is
i
π(ψkA ) = − ψA (5.14)
2πα0 k
The canonical commutation relation yields
There is a perhaps unexpected factor of two in the canonical momentum. This is related to the fact that
ψ and ψ̄ are actually the same variable. The standard Lagrangian density for complex (Dirac) fermions
is L = −ψ̄6 ∂ψ, and ψ and ψ̄ are treated as independent variables. This leads to a quantization condition
{ψ̄, ψ} = iδ(σ − σ 0 ). The standard Lagrangian for Majorana fermions differs from that of Dirac fermions
by an additional factor 1/2, and ψ and ψ̄ are treated as related variables. Despite these differences,
66
one gets same quantization condition (string actions differ from the “standard” field theory actions by a
factor 1/2πα0 , which appears in the quantization condition in the obvious way). A proper treatment of the
quantization of free fermions requires the Dirac constraint formalism, which produces the correct factor
of two; see e.g. [3] for details. Furthermore there are several sign choices to be made in the quantization
procedure because fermionic variables must be treated as anti-commuting rather than commuting, even
classically. These are just subtleties in the quantization of free fermions in any dimension, and that have
nothing to do with string theory. Therefore this will not be discussed in more detail here.
Now we want to expand the fermionic fields in modes. To do this we need to know
their boundary conditions. For closed strings the possible boundary conditions follow
from the requirement that all physical quantities are periodic under σ → σ + 2π. Physical
quantities for fermions are always bilinears, and hence the fermions themselves can be
either periodic or anti-periodic. This then leads to two distinct kinds of mode expansions
Periodic Fermions
√ X A −in(τ −σ)
A
ψ− (σ, τ ) = α0 d¯n e
n∈Z
√ X A −in(τ +σ) (5.16)
A
ψ+ (σ, τ ) = α0 dn e
n∈Z
Anti-periodic Fermions
A
√ X
−ir(τ −σ)
ψ− (σ, τ ) = α0 b̄A
re
r∈Z+ 12
√ X (5.17)
A −ir(τ +σ)
ψ+ (σ, τ ) = α0 bA
re
r∈Z+ 12
For historical reasons such fermions are called Ramond fermions and Neveu-Schwarz
fermions respectively. The notation b and d for their modes is also traditional. As usual,
the “bar” distinguishes left- and right-moving modes, and is not meant to indicate complex
conjugation.
It is now straightforward to substitute these mode expansions in the canonical com-
mutation relation (5.15) and obtain the commutation relations for the modes:
A B
br , bs = δ AB δr+s
A B (5.18)
dn , dm = δ AB δn+m
and similarly for the “barred” components. Furthermore barred and unbarred components
commute (or anti-commute, if one prefers).
For open strings the discussion is slightly different. As we have seen for the bosonic
string, the boundary condition arise from the surface term that one gets in a careful
67
derivation of the equations of motion:
Z
i A a A A a A
δψA S(ψ) = dσdτ (δ ψ̄ )ρ ∂ a ψ + ( ψ̄ )ρ ∂ a δψ
4πα0
Z (5.19)
i
dσdτ 2(δ ψ̄ A )ρa ∂a ψ A + ∂a (ψ̄ A )ρa δψ A
= 0
4πα
If we require this variation to vanish for general variations δψ A , then the first term requires
the equations of motion to be satisfied, i.e. ρa ∂a ψ A = 0, while the second term is a
derivative, so it only contributes at the boundary. Since there are boundaries only in the
σ direction, we get the requirement
ψ̄ A ρ1 δψ A = (ψ A )T ρ0 ρ1 δψ A = 0 (5.20)
−4πα0 δ
T ab (σ, τ ) ≡ √ S[γ] (5.22)
−γ δγab
where S is an action depending on the two-dimensional metric γ and other fields. This
definition is only usable if we have the action in a form where γ is explicit. In the
case of the bosonic string we started with the action in such a form, but then we went
to conformal gauge, and the explicit dependence on γ disappeared. In the case of free
fermions we started directly in conformal gauge. An action involving γ (i.e. an action
that is reparametrization or general coordinate invariant) can be written down but for
fermions this leads to some extra complications.
68
5.6 Noether Currents
There is a second method for obtaining the energy momentum tensor, the Noether method.
In general this works as follows. Suppose we have an action that is invariant under some
(infinitesimal) symmetry:
S[φ] = S[φ + δS φ] (5.23)
where φ stands for the set of fields, is a constant parameter, and δS φ is a variation of φ
that corresponds to some symmetry. For example translation invariance corresponds to
δS φ = na ∂a φ, where na is the direction of the translation. Now suppose we make depend
on the space-time point x (we work here in arbitrary dimensions, but we will apply the
results to fields on a string world-sheet). Then the action need not be invariant, but since
it must be invariant when is constant, it must be true that
Z
δS = dD xJ µ ∂µ (x) , (5.24)
for some J µ (there may be higher derivatives as well, but by partial integration one can
always get this form). On the other hand, when φ is a classical solution, any variation of
the action around it should vanish, and this should in particular be true for the variation
(x)δφ. Hence Z
dD xJ µ ∂µ (x) = 0 (5.25)
This must be true for arbitrary functions (x), and that implies that the current J µ is
conserved,
∂µ J µ = 0 (5.27)
Note that this is true only under the assumption that the set of fields is a classical solution.
Therefore we expect that to verify ∂µ J µ = 0 one has to use the equations of motion. In
this way one associates with every continuous symmetry a conserved current.
69
Hence
Z
1
δS = − 0
d2 ση ab ∂a [(σ)nc (∂c X)] ∂b X
2πα
Z (5.29)
1
=− d2 ση ab [(∂a (σ))nc (∂c X)∂b X + (σ)nc (∂a ∂c X)∂b X]
2πα0
The last term can be written as
Z Z
2 ab c
d ση (σ)n (∂a ∂c X)∂b X = − d2 ση ab ∂c (σ)nc (∂a X)∂b X
Z
− d2 ση ab (σ)nc (∂a X)∂c ∂b X
Z (5.30)
=− d2 ση ab ∂c (σ)nc (∂a X)∂b X
Z
− d2 ση ab (σ)nc (∂b X)∂c ∂a X
Hence Z Z
2 ab c 1
d ση (σ)n (∂a ∂c X)∂b X = − d2 ση ab ∂c (σ)nc (∂a X)∂b X (5.31)
2
The variation of S is therefore
Z
1 c 2 ab 1 ab
δS = − n d σ η ∂a (σ)(∂c X)∂b X − η ∂c (σ)∂a X∂b X
2πα0 2
Z
1 2 c a 1 db ac
=− nc d σ ∂a (σ)(∂ X)∂ X − η η ∂a (σ)∂d X∂b X (5.32)
2πα0 2
Z
1
=− nc d2 σ∂a (σ)T ac ,
2πα0
where in the last step we substituted (2.28). The last line will now be taken as the
definition of the energy-momentum tensor in the general case.
Applying this to the free fermion term in the action we get (dropping the index A for
convenience)
δψ = nc ∂c ψ (5.33)
so that
Z
0 1
−2πα δS[ψ] = − inc d2 σ (∂c ψ̄)ρa ∂a ψ + ψ̄ρa ∂a (∂c ψ)
(5.34)
2
If we use the equation of motion, ρa ∂a ψ = 0, this reduces to
Z
1 c
− in d2 σ∂a (ψ̄ρa ∂c ψ) (5.35)
2
from which we read off
1
Tac (ψ) = − iψ̄ρa ∂c ψ (5.36)
2
70
Note that this tensor is not symmetric, and hence not equal to the energy momentum
tensor one would get from a variation with respect to γac . This is due to the fact that in
general a current Jµ does not follow uniquely from the Noether procedure, but only up
to terms of the form ∂µ Aµν , where Aµν is an anti-symmetric tensor. One can indeed add
such a term to T ac to symmetrize it (this is called an improvement term; see exercise 5.6
in [6] for more details). In two dimensions the tensor turns out to be both symmetric and
traceless provided one uses the equations of motion. To see the former, note that (with
ρ0 ρ0 = −1)
1
T01 = − iψ̄ρ0 ∂1 ψ
2 (5.37)
1
= i(ψ− ∂1 ψ− + ψ+ ∂1 ψ+ )
2
while
1
T10 = − iψ̄ρ1 ∂0 ψ
2 (5.38)
1
= − i(ψ− ∂0 ψ− − ψ+ ∂0 ψ+ )
2
Now use ∂0 ψ− = −∂1 ψ− and ∂0 ψ+ = ∂1 ψ+ to show that T01 = T10 . For the ++ and −−
components of the tensor we find
1 i i
T++ = (T00 + T01 ) = − ψ̄ρ0 (∂0 + ∂1 )ψ = ψ T (∂0 + ∂1 )ψ (5.39)
2 4 4
and analogously for T−− . Obviously ψ+ appears only in T++ and ψ− only in T−− .
71
5.9 The Neveu-Schwarz Ground State
We can do the same for fermions. For a single Neveu-Schwarz fermion the “naive” classical
contribution to L0 is
∞
1 X
(L0 )cl (ψ) = rb−r br (5.42)
2 r=−∞
Just as in the bosonic case, we define the vacuum by
so that the negative modes create states. Normal ordering is defined as putting all creation
operators on the left of all the annihilation operators. We see then that in the infinite
sum all the terms with r < 0 are in the wrong order, and we use the anti-commutator to
re-order them. This yields
with s = −r. Hence for each positive, half-integer value s we get a contribution −s. They
can be added as follows
∞
1 X 1X
(−s) = − (2n + 1)
2 4 n=0
s∈Z+ 21 ;s>0
" ∞ ∞
#
1 X X
=− m− 2n
4 m=1 n=1
" ∞ (5.45)
∞
#
1 X X
=− m−2 n
4 m=1 n=1
∞
1X 1
= m=−
4 m=1 48
Hence we find that the contribution to the ground state energy of a Neveu-Schwarz fermion
is precisely half that of a boson. We will denote this ground state energy as aN S . As was
the case for bosons, there are better ways to arrive at this result, but this would require
more discussion.
72
In fact there is another problem. One may easily show that
so that acting with dm increase the L0 eigenvalue of a state by −m (this holds also for
the other operators αm and br ). Hence acting with d0 does not change the L0 eigenvalue,
or in other words, the energy of a state. Hence the state dA 0 |0i is degenerate with the
vacuum.
This leads one to consider as the ground state of the Ramond fermions a set of states
|iiR with the property X
dA
0 |iiR = MijA |jiR (5.48)
j
Here MijA is a set of matrices (as many as there are fermions) which must satisfy
A
M , M B = δ AB
(5.49)
γ 1 = σ1 , γ 2 = σ2 , (5.52)
for three fermions one gets γ i = σ i , for four fermions one gets the four-dimensional
Euclidean Dirac matrices (which are 4 × 4 matrices) etc. In general, in 2n + 1 and 2n
dimension one gets 2n × 2n matrices, and the ground states |aiR has 2n components.
The energy of the ground state (which yields the mass in target space) can be evaluated
as we didPfor the Neveu-Schwarz fermions, and one easily finds that the shift in L0 is in
1 ∞ 1
fact − 2 n=1 n = + 24 per fermion, the opposite of the free boson.
− a − aN S (5.53)
73
Here we use units 4/α0 for closed strings and 1/α0 for open strings, and we consider only the
left-moving sector of the closed string. As before, left and right sectors must be combined,
with the condition L0 = L̄0 . Note that the ground states energy gets contributions from
the bosons and the fermions. The first excited state is
bA
−1
|0i , (5.54)
2
1
and it has mass 2
− a − aN S . Then at the second excitation level we find a state
i
α−1 |0i (5.55)
all with mass 1 − a − aN S . Note that |A, Bi must be antisymmetric in A and B. The
value of a + aN S is
D−2 N
a + aN S = + , (5.57)
24 48
where D is the number of space-time dimensions and N the number of fermions (A =
1, . . . , N ).
D−2 N
a + aR = − (5.58)
24 24
The first excited states are the states
i
α−1 |iiR and dA
−1 |iiR (5.59)
with mass 1 − a − aR . Here we have normal ordered the operators d and α with respect
to the states |iiR , using
74
correct approach is to consider the Lorentz algebra. This yields D + 12 N = 26 in both
cases.1 Hence
a + aN S = 1
N
a + aR = 1 −
16
Note that if we could build a Ramond model with sixteen or more fermions, the ground
state would have non-negative mass-squared, i.e. would not be tachyonic. However, we
will see that this is not possible: Purely Ramond or purely Neveu-Schwarz string theories
are not consistent. Inconsistencies arise as soon as one studies interactions.
Now we see immediately that there is a problem: the components d0n and b0r have anti-
commutators of the wrong sign, and will lead to ghost states in the spectrum. In the
case of the bosonic string this was solved by imposing the T ab = 0 constraints. Those
constraints were explicitly solved by going to light-cone gauge, using conformal invariance.
1 1
This is a consequence of the fact that each fermion contributes 2 to the conformal anomaly, as can
easily be verified using the Virasoro generators.
75
This won’t help us with ψ µ . Since we have already used up conformal invariance to reduce
X µ to X i , we cannot achieve the same reduction for ψ µ . The way out turns out to be
an additional symmetry called super-conformal invariance. We will come back to it later,
but let us assume that it does what we expect, and fixes two components of ψ µ , reducing
it to ψ i .
which has only D−2 components, and is the only state at its mass level. By the arguments
used before, it must be massless. This requires a + aN S = 21 . On the other hand,
D − 2 Ntr
a + aN S = + . (5.65)
24 48
Here Ntr is the number of “transverse” fermions, i.e. the fermions that actually appear in
the mass formula, i.e. ψ i , i = 1, . . . , D − 2. Hence Ntr = D − 2, and from a + aN S = 21 one
i
easily deduces that D = 10. If we consider the state α−1 |0i that gave rise to a massless
vector boson in the bosonic string, then we find that now there is no problem making it
massive. It is accompanied by the set of states
76
5.15 The Lorentz Algebra∗
This can be understood as follows. We have argued that the relation D + 21 N = 26 follows
from closure of the Lorentz-algebra in lightcone gauge. Indeed, in the bosonic string that
will always lead to relation (5.5): D + cint = 26. However, the set of Lorentz generators of
the bosonic string, as shown in (3.100), acts only on the bosonic degrees of freedom X µ .
This is also true if we introduce compactified coordinates, or internal fermions ψ A . Since
the Lorentz-rotations do not act on these internal degrees of freedom, the form of (3.100)
stays exactly the same. If one goes to lightcone gauge the generators M i− do depend on
the internal degrees of freedom because we eliminate αn− by solving the condition T ab = 0,
and T ab contains all degrees of freedom. However, the generators M µν still act only on
X µ.
Clearly we get new Lorentz generators if we give the fermions a space-time index
µ. These generators act on ψ µ as well as X µ . The derivation is easy: the generators
are just the Noether currents associated with the global transformation X µ → Λµν X µ ,
ψ µ → Λµν ψ ν , where Λµν is a Lorentz transformation. The calculation gives us a set of
infinitesimal generators M µν . The terms for the bosonic string were given before in Eqn.
(3.100), and in fermionic strings we get the following additional terms:
∞
X
bµ−r bνr − bν−r bµr
NS: −i
r= 12
(5.68)
∞
i µ ν X
dµ−n dνn − dν−n dµn
R: − [d , d ] − i
2 0 0 n=1
Of course this changes the discussion of the Lorentz algebra drastically, not only because
there are additional terms, but also because the components b− −
r and dr are not indepen-
−
dent degrees of freedom, just as αn . They can be expressed as products of bosonic and
fermionic oscillators αi and bi or di (this can be seen explicitly in Eqn. (5.78), which will
be discussed later). Therefore it is no surprise that the result is different. Indeed, in the
NS-sector one finds a + aN S = 21 and D = 10, and in the R-sector a + aR = 0, D = 10.
77
to Φ + iωµν M µν Φ. We see from Eqn. (5.68) that under Lorentz transformations |iiR
transforms as
1 µ ν
|iiR → |iiR + [γ , γ ]ij ωµν |jiR = |iiR + iωµν (Σµν )ij |jiR . (5.70)
4
The matrix Σµν ≡ − 4i [γ µ , γ ν ] is the infinitesimal Lorentz generator in the spinor repre-
sentation. Apparently the Ramond ground states rotate like spinors. The corresponding
particles must thus be space-time spinors and – assuming the spin-statistics relation holds
– they must be fermions. Then all their excited states are fermions as well, since they are
obtained from the ground states by action with space-time bosonic operators dµ and αµ .
δX µ = ¯ψ µ
(5.71)
δψ µ = −iρa ∂a X µ ,
78
when we try to write the action in a manifestly reparametrization invariant way. Then
we would introduce a metric γαβ , which means introducing a new two-dimensional field in
addition to X µ and ψ µ . Supersymmetry acts on this field. It cannot act trivially, because
it transforms bosons into fermions. Hence we are obliged to introduce an additional field,
the superpartner of the word-sheet metric γαβ .
Instead of γαβ a more convenient set of fields is the so-called zwei-bein eαa ,
For more details about this action and how it is obtained see [1]. The first two terms are
just the generally covariant form of the conformal gauge action. The third term couples
the gravitino to the supercurrent. The last term is a four-fermi interaction required by
supersymmetry. Note that there are no kinetic terms for the gravitino. In more than two
dimensions there would be such a term, but it vanishes in two dimensions. This action is
not just supersymmetric, it is in fact invariant under local supersymmetry, which means
that the transformations can be σ-dependent.
Ja = 0 . (5.77)
(“conformal gauge” means a little more than for the bosonic string. It turns out that
one can choose eaα = δαa , and one can use the extra fermionic symmetries to set χα = 0.)
79
The other three equations of motion are the usual ones for X µ and ψ µ , plus the condition
Tab = 0.
An obvious guess is now that the new condition is the constraint we were looking
for earlier. This is indeed correct. In light-cone coordinates ψ µ splits into ψ i , ψ + and
ψ − . Using conformal symmetry and a supersymmetry transformation that survives the
conformal gauge choice, we may bring X + to the same form as before, and we can set ψ +
to zero. Then the conditions Ja = Tab = 0 can be solved, yielding an expression X − and
ψ − in terms of the other components:
1 ⊥
∂+ X − = T
p+ ++
(5.78)
2
ψ − (σ + ) = + J+⊥
p
where T ⊥ and J ⊥ are the the energy momentum tensor and the supercurrent, but with
the summation indices µ = 0, . . . , D − 1 replaced by i = 1, . . . , D − 2. In the case of closed
strings there are analogous relations for the right-moving components ∂− X − and ψ − (σ − ),
whereas in the open string left- and right-moving components of all fields are the same.
These relations express αn− and b− −
r (or dn ) in terms of all the other oscillators. To get
the full spectrum it therefore sufficient to act with all transverse oscillators, as we already
did above, anticipating this result.
80
What we mean by computing the diagram is to compute a number, an amplitude,
that corresponds to the diagram. Since there are no external lines, this amplitude is not
the amplitude for some scattering process, but it turns out that it can be interpreted as
a contribution to the energy of the vacuum in target space. Experimentally, this number
would be measurable because it contributes to the cosmological constant, which affects
the evolution of the universe. Although this is a very interesting subject, it is not our
main interest now. We will not be interested in the result of the calculation, but in its
internal consistency. This will impose strong constraints on the theory, which determine
precisely the relation between R and NS.
Here SE is the Euclidean action, obtained from the action in two-dimensional Minkowski
space by defining σ0 = −iσ2 (see appendix A). Furthermore “fields” stands for all fields
in the theory, such as X µ and ψ µ and any internal degrees of freedom. One integrates
over all values of these fields in all points on the torus, and also over all values of the
two-dimensional metric γ and the gravitino. Hence all X µ (σa ) are treated as integration
variables, labeled by µ, σ0 and σ1 . Obviously this is an infinite-dimensional integral.1
The analytic continuation to imaginary time has the advantage of making the path
integral better behaved, because in Minkowski space the weight factor is exp(iS/~). In
the classical limit, ~ → 0, this integral is dominated by the extrema of the action, the
classical solutions. In the Euclidean formulation it is clear that fluctuations around the
classical solutions, which increase the action, are suppressed by the negative exponential.
1
An additional complication is that fact that some fields are fermionic. The the corresponding path
integral variables must be anti-commuting, which requires a special formalism which will not be discussed
here.
81
However, there may be fluctuations that do not change the action. These are usually
related to symmetries. Let us first consider the integration over the two-dimensional
metric γ. Then the symmetries we have to worry about are reparametrization and Weyl
invariance. If we take the definition we gave above, integrating these symmetries would
simply give rise to an infinite factor. Therefore we change the definition:
A
Afix = (6.2)
Vol(Diff)Vol(Weyl)Vol(Super)
where Vol(Diff) denotes the total integration over the reparametrizations (diffeomor-
phisms), Vol(Weyl) the integration over the Weyl symmetry and Vol(Super) the inte-
gration over the local supersymmetries. Note that we are not yet considering the space-
time Poincaré symmetries of X µ , because we are discussing the integral over the two-
dimensional metric, which is not affected by these symmetries.
Of course Afix is a ratio of two infinite quantities, which requires some care. This is
done by means of the Fadeev-Popov procedure, which will not be described in detail here.
Essentially one writes the integral over γ and χ as an integral over fluctuations of these
fields around the conformal gauge, and the latter integrals then cancel against the “Vol”
factors in the denominator. In doing these manipulations one has to make a change of
integration variables for γ and χ, and this results in some Jacobian factors for the change
of measure. The result is thus
Z Z
Afix = dΩ D[fields]Jac (γ)Jac(χ)e−SE [fields,η,0] , (6.3)
where Jac (γ) and Jac (χ) are the Jacobians. The integral over Ω is over all two-
dimensional surfaces that are really inequivalent: surfaces that cannot be transformed
into each other by coordinate and Weyl-transformations. We will return to this later.
To take care of the Jacobians one introduces extra fields, the Fadeev-Popov ghost
b, c, β and ρ, such that1
Z
Jac (γ)Jac (χ) = DbDcDβDρe−SE (b,c,β,ρ) , (6.4)
for some suitable ghost action. It turns out that this action is gaussian, so that the
integral can be done explicitly. A gaussian path integral is just a generalization of the
well-known result for a finite-dimensional integral
Z
1
dx1 . . . dxn e−xi Aij xj = π n/2 [det A]− 2 . (6.5)
If we consider uncompactified strings, the action of X and ψ is gaussian as well, and can
also be done.
1
In the literature ρ is usually called γ, but unfortunately that character is already in use in this context.
82
In the special case of the torus it turns out that the ghost contribution precisely cancels
against two degrees of freedom of X µ and ψ µ , so that we are finally left with
Z Z
i i
Afix = dΩ D[X]D[ψ]e−SE [X ,ψ ,η,0] , (6.6)
with the action in conformal gauge. This cancellation only works in 26 dimensions for the
bosonic string and 10 dimensions for the superstring.
83
Here the sum is over all surfaces that cannot be deformed continuously into each other,
and the integral is over all parameters of each surface that do not correspond to coordinate
changes and Weyl transformations.
Since string world sheets are two-dimensional we are in the lucky situation that we
can make use of powerful mathematical results regarding such surfaces, which are called
Riemann surfaces. The parameters of such a surface are called the moduli, and for each
topology the moduli are known.
84
6.5 The Rôle of the Dilaton∗
6.5.1 Bosonic String in a Dilaton Background
In the discussion of bosonic string compactification we have seen that we can describe
string propagation in non-trivial gravitational backgrounds gµν and anti-symmetric tensor
backgrounds Bµν . There is one other massless field in the uncompactified closed string,
namely the dilaton. It seems natural to look for a modification of the action that involves
the dilaton φ. There are in fact several natural candidates, for example
√
Z
S(X, γ, φ) ∝ dσdτ γφ (6.11)
or
√
Z
S(X, γ, φ) ∝ dσdτ γX µ Xµ φ (6.12)
to be added to the action we already have. The problem with both proposal is that they
both break Weyl invariance. Weyl-invariance is a transformation of the two-dimensional
fields X and γ. From the two-dimensional point of view the space-time background fields
g, B and φ are not fields, but more like coupling constants. They are not supposed to
transform under reparametrizations and Weyl transformations.
If Weyl invariance is broken, we loose the possibility to go to the conformal gauge,
and later to lightcone gauge. This turns out to have fatal consequences. However, there
is one possibility that turns out to be Weyl invariant:
√
Z
S(X, γ, φ) ∝ dσdτ γR(γ)φ , (6.13)
where R is the two-dimensional curvature scalar. In four dimensions, and without the
factor φ, this is the Einstein action for gravity. In two
R dimensions, however, gravity is
√
trivial: the equations of motion obtained by varying dσdτ γR(γ) with respect to γ
vanish identically.
This means that the gravitational action (or (6.13) with constant φ) does not change
if we change γ in a continuous way. Hence the action is the same for a sphere and any
potato-shaped surface, that can be transformed into each other by continuously varying
the metric. However, for two surfaces that cannot be continuously transformed into each
other, such as a sphere and a torus, the action may be different. This is indeed the case.
√
It turns out that the integral of γR(γ) over a manifold is equal, up to normalization, to
the Euler number of the surface:
√
Z
1
χ= dσdτ γR(γ) (6.14)
4π
85
rôle. Suppose we write the dilaton as φ = c + φ̃. Here c is some constant, and φ̃ represents
the fluctuations of the dilaton field around that constant value. Then the action can be
written as SE = S̃E + cχ (here we have chosen a convenient normalization for the dilaton
action, and we define S̃ as the c-independent part of the action). The result is
X Z Z Z
−cχ
hout| U |ini = e dΩ D[fields] DbDcDβDρe−S̃E (fields,b,c,β,ρ) (6.15)
topologies
We see that c appears as a topology-dependent weight factor for the terms in the sum. It
is customary to define ec = κ, which is called the string coupling constant. The motivation
for this terminology is that each time we add a handle to the surface, we get an extra
factor κ2 in the contribution to the amplitude. It is as if there is one factor κ for each
emission of a closed string.
The string coupling constant is thus related to the classical value (vacuum expectation
value) of a field, the dilaton. In this sense it is not a “parameter” of the theory. We would
speak of a parameter if its value had to be defined a priori. But since κ is related to a
vacuum expectation value of a field, it is possible that some dynamical mechanism (for
example the minimization of a potential) determines its value. However, as long as this
mechanism is unknown there is not much practical difference between the two descriptions.
6.6 Moduli
6.6.1 Moduli of Riemann Surfaces∗
The parameters of the surfaces integrated over by the dΩ integral are called the moduli of
the surface. They should not be confused with the moduli of a compactification surface:
the latter is a space-time surface, whereas here we talk about world-sheets. Mathemati-
cally, however, they are the same.
The number of moduli of a surface of genus g (with b = c = 0) is 2g. This implies
that a sphere has no moduli. Conformally, all spheres are equivalent.
In general, any manifold can be covered with coordinate patches. On each patch one
can go to conformal gauge, so that the metric is ηab . If the manifold has moduli, they are
related to the different ways of sewing the patches together.
On a sphere, one can choose a patch that covers the entire sphere except for one
point. This patch can be mapped to the two-dimensional plane. Infinity in this plane
corresponds to the missing point on the sphere. In this way all spheres are mapped to
the same plane, and hence they are all equivalent.
86
6.6.2 Moduli of the Torus
A torus is made by first making a cylinder, and then gluing the ends of the cylinder
together. A cylinder has just one modulus. One can map it to a rectangle (by cutting it
open), and by scale invariance one can make sure that one side of the rectangle has length
one. The length of the other side, L, can then not be changed anymore, and corresponds
to a modulus parameter.
To see that there really is a free parameter one may a walk on the surface starting in one
corner at a 45 degree angle between the two coordinate axes. Since we are in a metric ηab
this angle is unambiguously defined. If L = 1 we end in another corner, if L 6= 1 we end
on one of the sides. No coordinate or Weyl transformation can change that.
The gluing of the two ends of the cylinder introduces a second parameter. This
parameter can be understood as a rotation of the top of the cylinder before gluing it to
the bottom. Usually one represents a torus as a two-dimensional plane modulo a lattice
X
1
One of the axes is chosen along the x-axis, and the length of a basic x-interval is chosen
equal to 1. Then the lattice is specified by a two-dimensional vector as indicated in the
picture. The projection of this vector along the y-axis corresponds to the length of the
cylinder, the projection along the x-axis to the rotation before gluing.
87
6.7 Complex Coordinates
When we go to Euclidean space the factors e−in(τ ±σ) ≡ e−in(σ0 ±σ1 ) appearing in mode
expansions become e−n(σ2 ±iσ1 ) . This suggests that it may be more natural to introduce
complex coordinates
z = σ2 + iσ1 , z̄ = σ2 − iσ1 (6.16)
This is convenient for other reasons as well. For example, Riemann surfaces are most
easily described as complex surfaces (often called “complex curves”, since they have just
one complex coordinate). The origin of the simplifications is Cauchy’s theorem, which is
a very powerful tool for evaluating contour integrals. Indeed, these techniques are used
abundantly in string theory. Of course Cauchy’s theorem is special to two real, or one
complex dimension. This is another reason why one can get a lot further with strings
than with higher dimensional objects.
In terms of complex coordinates the torus is described by the complex plane modulo a
lattice, and the lattice is described by a point in the complex plane. This point is usually
called τ 1. R
The integral over the moduli, previously denote dΩ, becomes in the case of the torus
d2 τ
Z
, (6.17)
( Im τ )2
with d2 τ = dReτ dImτ . The denominator is a measure factor, that gives the correct
weight to different values of τ . It emerges from a careful analysis of the path integral. We
will not derive it here, but we will see later some justification for its presence.
88
Im
! !+1
0 1 Re
This can be generalized further. One should keep in mind that the torus was defined
by rotating one basis vector along the real axis in the complex plane, and scaling it to 1.
The choice of this basis vector is free; we can also choose the direction “τ ”. This has the
effect of interchanging the two basis vectors. This rotation, combined with a rescaling the
new basis vector along the real axis, has the effect of replacing τ by − τ1 . This is most
easily illustrated by taking τ purely imaginary:
!
rotated
rescaled
- 1
!
The set of such transformations of the torus forms a group, called the modular group.
We have identified two elements of that group, namely
T : τ →τ +1
1 (6.18)
S: τ →−
τ
It turns out that these two transformations generate the entire group. The most general
89
modular transformation has the form
aτ + b
τ→ , a, b, c, d ∈ Z; ad − bc = 1 . (6.19)
cτ + d
This group is isomorphic to SL2 (Z)/Z2 . The group SL2 can be defined by the set of 2 × 2
matrices
a b
(6.20)
c d
with determinant 1. The group SL2 contains the element −1. In the modular transforma-
tion this is indistinguishable from the identity, and for this reason the modular group is
actually isomorphic to SL2 (Z)/Z2 rather than SL2 (Z). One may check that the modular
transformations satisfy
(ST )3 = S 2 = 1 . (6.21)
We see thus that not all parameters τ give different tori.
-1 - 12 1
2
1
The entire upper half plane is covered with an infinite number of such regions of different
shapes and sizes. For example, the lower part of the strip − 12 ≤ Re τ < 12 contains an
infinite number of identified regions. This can be seen as follows. The map τ → −1/τ
takes the outside of the circle to the inside. Suppose we take the dashed region and shift
it by n units, using τ → τ + 1 n times. This gives an infinity of regions. Now apply
τ → −1/τ . This maps any such region to the inside of the circle, and in fact within the
90
range − 12 ≤ Re τ < 21 . Hence the lower part of the strip − 12 ≤ Re τ < 12 contains an
infinite number of copies of the “standard” region, the dashed area in the picture.
When we integrate over each such region we get the same answer. At least we should
get the same answer, since otherwise the result depends on the choice we make among the
equivalent regions, and this would clearly be absurd. The requirement that we always get
the same answer is called modular invariance. Although it is formulated here as a property
of the integral, it turns out that the only way for it to hold in general (for example also
for diagrams with external lines) is that in fact the integrand is invariant.
It turns out that the measure (6.17) is already modular invariant by itself. One may
verify that
aτ + b
Im = |cτ + d|−2 Im τ (6.22)
cτ + d
and that
aτ + b
d2 τ 0 = |cτ + d|−4 d2 τ , τ 0 = , (6.23)
cτ + d
so that the factors precisely cancel. In other words, the integration measure used in (6.17)
is modular invariant.
It follows that the integrand, obtained from the path integral, must also be modular
invariant.
Here q(t) is a coordinate, and one considers a time interval 0 ≤ t ≤ β. The abbreviation
“PBC” means periodic boundary conditions. Concretely it means that q(0) = q(β). On
the left-hand side one is integrating over all paths that return to their starting point after
a Euclidean “time” interval β.
This result can be straightforwardly generalized to theories with many dynamical
variables qk . Here we have infinitely many such variables. The quantities X i (σ 1 ), for each
i and for each value of σ 1 should all be viewed as dynamical variables qk . But this is not
a serious difficulty.
In the lattice description of the torus we regard the real axis as the σ 1 direction, and
the imaginary axis as the Euclidean time direction. If Re τ = 0 we would find for the
91
path integral, as a straightforward generalization of (6.24):1
Z
DXe−SE (X) = Tre−2π Im τ H . (6.25)
The only small subtlety here is the factor 2π. It appears because the torus as depicted
earlier has a periodicity 1 rather than 2π along the σ 1 direction. To make contact with
earlier conventions we had to scale up the entire lattice by a factor of 2π, so that we get
2πτ instead of τ .
What happens if Re τ 6= 0? In that case we have to twist the torus before gluing it
together again, and the periodic boundary conditions in the Euclidean time direction are
defined including such a shift. The operator performing such a shift in the σ 1 direction is
the momentum operator P . A shift by an amount 2π Re τ is achieved by the operator
eiP (2π Re τ )
(6.26)
The correct result is obtained by inserting this shift operator in the trace, so that we get
Z
DXe−SE (X) = Tre−2π Im τ H eiP (2π Re τ ) . (6.27)
The operators H and P are the time and space translation operators on the cylinder,
and they can be derived from the energy-momentum tensor. We have already seen that
H = (L0 − 1) + (L̄0 − 1). The expression for the momentum operator is
Z 2π
1
P = dσ 1 T01 = L0 − L̄0 (6.28)
2πα0 0
Putting everything together we find then
Z
DXe−SE (X) = Tre−2π Im τ [(L0 −1)−(L̄0 −1)]+2πi Re τ (L0 −L̄0 )
(6.29)
2πiτ (L0 −1) −2πiτ̄ (L̄0 −1)
= Tre e
1
To be precise: an infinite factor has been dropped on both sides of the equation. On the left hand side
the integral over constant X gives infinity; on the right hand the trace is proportional to the normalization
of the vacuum h0|0i = δ(0). This is a momentum conservation δ-function which is trivial because there are
no external momenta. This factor is due to the translation invariance of the action; the rest of Poincaré
invariance gives a finite integral over the rotations of X i , which requires no special treatment.
92
where
q ≡ e2πiτ (6.31)
and dn is the number of states with L0 − 1 eigenvalue n. Such a function is usually called
a partition function.
Partition functions first appeared in statistical mechanics, as a sum over all physical
states, weighted by a factor exp (−βE), where E is the energy of the state and β the
inverse temperature. They play a major rôle in thermodynamics. In string theory, these
functions are indeed relevant for understanding thermodynamic properties of strings at
finite temperature, but here we only use them as a tool to investigate modular invariance.
The partition function is the simplest quantity for which those conditions can be explored.
An important difference with partition functions in statistical mechanics is that in
closed string theory the partition function is built out of complex quantities, functions of
τ for the left-movers and of τ̄ for the right-movers. In the simplest cases the full partition
function simply factorizes, so that the full partition takes the form
P (τ, τ̄ ) = F (τ )G(τ̄ ) , (6.32)
but more generally it will be a sum of the form
X
P (τ, τ̄ ) = Fi (τ )Gi (τ̄ ) . (6.33)
i
Since string excitation levels, i.e. the eigenvalues of L0 differ by integers, the typical form
of such a function is (but note that in fermionic strings, discussed in the next section,
half-integer powers of q may also appear in intermediate results)
F (τ ) = q α (d0 + d1 q + d2 q 2 + . . .) with q = e2iπτ (6.34)
Here α is determined by the energy of the vacuum state |0i. It is customary to write
these functions either as functions of q or as functions of τ , whatever is most convenient.
The variable q is most convenient for Taylor expansions to count states, and the variable
τ is most convenient when considering modular transformations. In any case, the two
variables are always related as q = e2iπτ .
Note that as functions of q partition functions have an expansion with real coefficients.
Hence we have
[f (q)]∗ = f (q ∗ ) (6.35)
If we now introduce a function F (τ ) ≡ f (e2πiτ ), then obviously
[F (τ )]∗ = F (−τ ∗ ) ≡ F (−τ̄ ) (6.36)
Note that we denote the complex conjugate of τ as τ̄ . If F and G are described by the
same physical system (e.g. bosonic oscillators), then they will be the same as functions
of q, and hence the partition function than takes the form
P (τ, τ̄ ) = F (τ )F (−τ̄ ) . (6.37)
This will obviously be true for the bosonic string, where both left- and right-movers are
created by oscillators.
93
6.11.2 Assembling the Bosonic String Partition Function
So now it is simply a matter of counting states as each level. This is a simple combinatorial
exercise. Remember that the Hilbert space of physical states is obtained by taking the
i
vacuum |0i and acting with the oscillators α−n , n > 0. Suppose we had just one oscillator
α−n . The set of states is then
State Mass
|0i −1
α−n |0i n−1
2
α−n |0i 2n − 1
3
α−n |0i 3n − 1
... ...
This function – with precisely this factor e(1/24)2πiτ – turns out to have nice properties
under modular transformations. We will see them in a moment. Comparing this with
the string partition function we see that precisely for D = 26 the result can be expressed
completely in terms of η-functions:
Posc. (τ, τ̄ ) = [η(τ )η(−τ̄ )]−24 . (6.42)
note that η(−τ̄ ) = (η(τ ))∗ .
94
6.11.3 The Contribution of Momenta
But we are not quite finished. We have forgotten that our Hilbert space also contains
momentum excitations of the vacuum, |pi i. These excitations work completely indepen-
dently of the oscillators, and again give a multiplicative factor. Since the momenta are
continuous, the trace now becomes an integral
Putting everything together we find for the bosonic string one-loop path integral (up to
normalization)
d2 τ
Z Z
−SE (X) 1
D[all fields...]one-loop e ∝ 2 12
η(τ )−24 η(−τ̄ )−24 (6.45)
( Im τ ) ( Im τ )
Hence the combination that appears in the string partition function transforms as
The factor |τ | in the second√transformation cancel against the transformation factor of the
momentum contribution, [ Im τ ]−1 for each dimension. Since these two basic transfor-
mations generate the entire modular group, it is clear that the bosonic string is modular
invariant.
Note that D = 26 is essential. If D 6= 26 we are left with factors q q̄ to some power
if we try to express the partition function in terms of η-functions. These factors are not
invariant under τ → −1/τ . We discover here a second reason why the critical dimension
of the string must be 26.
95
6.13 Strings versus Particles∗
We have seen that the string spectrum consists of an infinity of modes, which an observer
interprets as particles. One may therefore think that a string loop diagram is in some
way related to the sum of particle loop diagrams. To see if that is true, we will now do
the same computation for a single particle, and at the end sum over all particles.
We follow here the discussion in [1], part II, page 55. The string loop diagrams are
supposed to contribute to the “cosmological constant” Λ. For a massive real free boson
field theory the relevant expression is
Z Z
Λ D 1
µ 2 2
1
e = Dφ exp − d x 2 ∂µ φ∂ φ + m φ = √ (6.49)
det − + m2
The reason we compare the string calculation with the logarithm of the field theory path
integral is that the latter generates not only the one loop diagram, but also all disconnected
diagrams composed by putting loops together. The string diagram we have computed is
however strictly one-loop. Pictorially the field theory path integral is
The factors N1 ! are combinatorial factors, and the whole diagrammatic expansion expo-
nentiates. Hence by taking the logarithm we pick out precisely the one-loop contribution.
The result is equal to
1
Λ = − log det (− + m2 )
2
1
= − Tr log − + m2
2Z
1 dD p (6.50)
=− hp| log(− + m2 ) |pi
2 (2π)D
dD p
Z
1
=− D
log(p2 + m2 )
2 (2π)
There is another way of seeing that this is the right expression. Up to irrelevant
factors, (p2 + m2 )−1 is the propagator of a massive particle. When computing a Feynman
diagram one associates with every internal line such a propagator. If one differentiates a
propagator with respect to m2 it splits into two propagators:
d 1 1
= − (6.51)
dm2 p2 + m2 (p2 + m2 )2
96
The expression
dD p
Z
1
(6.52)
(2π) (p + m2 )N
D 2
is the loop integral for a one loop diagram with N external lines, each with zero mo-
mentum and trivial coupling. Diagrammatically such a diagram is an N -sided polygon.
By differentiating with respect to m2 we go from an N -sided to an N + 1-sided polygon.
Our interest is in a one-loop diagram with zero external lines, which is what we com-
puted in string theory. By inverting the recursion, we should be able to compute that
by integrating the N = 1 diagram with respect to m2 . This of course precisely gives the
logarithm.
To evaluate (6.50) we use the following integral representation
Z ∞
dt −tX
log(X) = − e (6.53)
0 t
This can be “derived” by differentiating w.r.t. X on both sides. The divergence at t = 0 is
independent of X and drops out when one differentiates. So in fact the correct, regulated
expression has a constant term +∞ on the left-hand side:
Z ∞ Z ∞
dt −tX dt −t
log(X) = − lim e − e (6.54)
→0 t t
We drop the m-independent, but infinite constant. The rest of the calculation goes as
follows
dD p
Z
1
Λ=− log(p2 + m2 )
2 (2π)D
1 ∞ dt dD p −t(p2 +m2 )
Z Z
= e
2 0 t (2π)D
(6.55)
1 ∞ dt −tm2 dD p −tp2
Z Z
= e e
2 0 t (2π)D
1 ∞ dt −tm2
Z
1
= e
2 0 t (4πt)D/2
As remarked above, in this calculation an m-independent infinite term was discarded,
proportional to the volume of momentum space. Apart from that, the integral has now
an even stronger divergence at t = 0. The latter divergence can be recognized as the
usual UV divergence in quantum field theory. This can be seen as follows. As seen
above, differentiating (6.50) N times with respect to m2 gives (6.52), which is the basic
integral one encounters when one computes a one-loop diagram with N external lines.
For N = D/2 this diagram diverges logarithmically for large p, and for N < D/2 there is
a power divergence. These are called ultra-violet divergences. On the other hand, if we
differentiate the result of the integration N -times w.r.t. m2 we get
Z
dt N −D/2 −tm2
∼ t e (6.56)
t
97
which is also logarithmically divergent for N = D/2 and power divergent for N < D/2.
We conclude that the divergence at t = 0 must be interpreted as the field theory ultraviolet
divergence. The first, m-independent divergence has a different interpretation. It amounts
to an infinite shift in the vacuum energy, and can be attributed to the contribution of the
ground state energies of the infinitely many oscillators. This is the target space analog of
the infinite sum over “n” we encountered in string theory. Dropping this term amounts
to setting the energy of the vacuum to zero.
If we treat string theory as an infinite set of field theories, we expect to get1
X Z dD p p
1
Λstring = − 2 D
log det (p2 + m2 ) (6.57)
m
(2π)
where the sum is over all physical states in the theory. Using the integral representation
of a single particle we get (for the uncompactified bosonic string)
Z ∞ 13 X
1 dt 1 2
Λstring = e−tm (6.58)
2 0 t 4πt m
The sum is over all vibration modes in the string spectrum, treated as if they were
individual particles. But this is therefore closely related to the sum we have evaluated in
the previous section, the bosonic string partition function
Y Y
η −24 (τ )η −24 (−τ̄ ) = (q q̄)−1 (1 − q n )−24 (1 − q̄ n )−24 (6.59)
n n
In this computation, the parameter q was given by q = e2πiτ . Now we have to match
the parameter τ to the integration variable t introduced above. This can be achieved by
comparing leading term of the two expressions, the tachyon contribution. Consider first
the real part of the partition function. Its leading term is.
e4π Im τ
[1 + . . .] (6.60)
The tachyon mass is m2 = −4/α0 . From the leading term of (6.58) we read off the relation
(4/α0 )t = 4π Im τ (6.61)
If we replace Im τ by t with this relative factor, this is almost the sum we are looking for,
except that we have to deal with the dependence on the real part of τ . A closely related
problem is that the partition function contains more than just physical states. It also
includes states with mL 6= mR , and we have seen before that such states are not physical.
Hence they do not correspond to particles, and do not appear in the sum (6.58). Note
that precisely these states introduce a dependence on Re τ in q n q̄ m for n 6= m. This
1
We may ignore the spins of the particles here. We simple sum over all different spin states, and treat
the particles as scalars.
98
allows us to “project out” these “unmatched” states by integrating over Re τ , so that all
complex phases integrate to zero. The final result is
Z 1
2
−tm2
X
e = d(Reτ )η −24 (τ )η −24 (−τ̄ ) (6.62)
m − 12
Apart from overall factors, which are not of interest here, we see that we have obtained
the string loop diagram, with one important difference: instead of integrating over a
fundamental domain, as shown in the figure in section 6.9, we are integrating over the
entire strip − 12 < Re τ ≤ 21 . But this strip includes the true string contribution an
infinite number of times. Hence if the string contribution is finite, the field theory sum is
necessarily infinite.
Furthermore we see that if we take the standard integration region in string theory, as
shown in the figure, the dangerous Im τ → 0 region is cut out. This is how string theory
avoids the field theory divergence. It includes a natural, built in cut-off that is required
by its internal consistency.
Nevertheless the bosonic string loop integral is not finite. This is entirely due to the
tachyon. Indeed, for a negative mass squared m2 = −A we get an integral of the form
Z ∞ 13
dt 1
eAt , (6.65)
0 t 4πt
which diverges badly for large t. Massless particles (A = 0) do not give a divergence for
large t and massive particles are also completely safe. So if we get rid of the tachyon, we
may expect a finite result.
Note that the tachyon divergence comes from the opposite side of the integration
region as the UV-divergence. This is usually called an infrared divergence. The origin of
such divergences is rather different, and are usually associated with the definition of the
vacuum.
99
that we do not have continuous components pI for √ the compactified momenta. Since
their contribution is absent, we have to drop a factor ( Im τ )−N from the bosonic string
partition function. However, there are also additional contributions to the mass due to
the “momenta” p~L and p~R , see (4.12). Since these momenta can be assigned to any state,
completely independently from the oscillators, this leads to a multiplicative factor in the
partition function. This factor is
2 2
X
PΓ (τ ) = eiπτ p~L e−iπτ̄ p~R , (6.66)
(~
pL ,~
pR )∈Γ
where we are assuming that the momenta (~pL , p~R ) lie on a lattice Γ. The compactified
string partition function is thus given by
√
Pcomp. = Puncomp. ( Im τ )N PΓ (6.67)
Since Puncomp is modular invariant, we have to check modular invariance of the remaining
factors.
To check τ → τ + 1 is, as always, easy. We see that this requires that p~2L − p~2R is an
even integer. We recognize this as the requirement that Γ should be a Lorentzian even
lattice.
To derive the other transformation rule requires a trick, the Poisson resummation
formula. This works as follows. Suppose we have a lattice Λ, and a continuous function
f (w).
~ Consider the quantity X
f (w)
~ (6.68)
w∈Λ
~
The Poisson formula re-writes this sum as a sum over the dual lattice Λ∗ . First define
X
F (~z) = f (w
~ + ~z) (6.69)
w∈Λ
~
Obviously F (~z) is periodic in ~z with the periodicity of Λ. Hence we may Fourier expand
it. The Fourier components of functions with periodicity Λ lie on the dual lattice Λ∗
X
F (~z) = e−2πi~z·~v F ∗ (~v ) (6.70)
~v ∈Λ∗
where vol(V ) denotes the volume of the unit cell, i.e. the result of the integral of 1 over
the unit cell. Now we substitute F (y):
Z
∗ 1 X
F (~v ) = dN ye2πi~y·~v f (w
~ + ~y ) , (6.72)
vol(Λ) unit cell of Λ
w∈Λ
~
100
and we observe that the sum over all cells combines with the integral over the unit cell to
an integral over all of RN , where N is the dimension of Λ. Hence we may write
Z
∗ 1 X
F (~v ) = dN ye2πi(w+~~ y )·~v
f (w
~ + ~y )
vol(Λ) unit cell of Λ
w∈Λ
~
Z (6.73)
1 N x·~v
2πi~ 1 ∗
= d xe f (~x) ≡ f (~v )
vol(Λ) RN vol(Λ)
In the first step we made use of the fact that ~v and w ~ are from mutually dual lattices, so
their inner product is integer. Applying this result to F (0) yields the formula
X X 1 X ∗
f (w)
~ = F ∗ (~v ) = f (~v ) (6.74)
∗
vol(Λ) ∗
w∈Λ
~ ~v ∈Λ ~v ∈Λ
with Z
∗
f (~v ) = dN xe2πi~x·~v f (~x) (6.75)
RN
For our purpose we apply this to the transformed term in the lattice partition function
2 2
f (~pL , p~R ) = e−iπ~pL /τ eiπ~pR /τ̄ . (6.76)
It turns out that we should regard (~pL , p~R ) as a vector on a Lorentzian lattice; only then
do we get a simple answer. The foregoing discussion holds also for such lattices. The
volume of the unit cell is defined as
p
vol(Λ) = |det g| , gij = ~ei · ~ej , (6.77)
where ~ei is a set of basis vectors. This definition generalizes to the Lorentzian case by
simply replacing the inner product by a Lorentzian one. The same is done with all other
inner products in the foregoing paragraph. The Poisson resummation formula allows us
to write the lattice sum of (6.76) as a sum over the Lorentzian dual lattice. The summand
is Z
∗
f (~vL , ~vR ) = d2N xe2πi(~xL ·~vL −~xR ·~vR ) f (~xL , ~xR ) . (6.78)
R2N
As long as Im τ > 0 this is a well-defined gaussian integral, which is easy to do, and
yields
2 2
f ∗ (~vL , ~vR ) = |τ |N eiπτ~vL −iπτ~vR (6.79)
Hence
1 1 X 2 2
PΓ (− ) = |τ |N eiπτ~vL e−iπτ̄~vR , (6.80)
τ vol(Γ)
(~vL ,~vR )∈Γ∗
∗
where Γ is the Lorentzian dual of Γ We observe that the factor |τ |N cancels against the
contribution from ( Im τ )N/2 . The only way to get modular invariance is clearly Γ = Γ∗
and vol(Γ) = 1. The latter condition follows in fact from the self-duality, since vol(Γ) is
always the inverse of vol(Γ∗ ).
The conclusion is thus that the torus compactified string is modular invariant if and
only if the momenta (~pL , p~R ) lie on a Lorentzian even, self-dual lattice.
101
7 Superstrings
Now we will consider the fermionic string theory discussed in chapter 4. Remember that
its critical dimension was 10, and that in lightcone gauge the spectrum was built out of
8 transverse bosons and 8 transverse fermions. The contribution of the bosons to the
partition function does not differ from what we have seen in the previous section. It is
Here we have split the coefficient a in L0 −a into two parts, one associated with the bosons
and another associated with the fermions. For the bosonic contribution we make a choice
per boson as in the case of the bosonic string, and the rest will have to be supplied by the
fermions. Only in this way can we express the bosonic partition in terms of η-functions.
Now we compute the fermionic contributions.
h=0 |0i
h = 12 b− 1 |0i
2
h=1 none
h = 32 b− 3 |0i
2
h=2 b− 3 b− 1 |0i
2 2 (7.3)
5
h= 2
b− 5 |0i
2
h=3 b− 5 b− 1 |0i
2 2
7
h= 2
b− 7 |0i
2
h=4 b− 7 b− 1 |0i , b− 5 b− 3 |0i
2 2 2 2
Note that fermionic oscillators must satisfy the Pauli exclusion principle, so that for
example b−1/2 b−1/2 is zero. For this reason there is no state at level h = 1, and we have
to go to h = 4 to find more than one state.
102
It is straightforward to compute the partition function. Each oscillator b−r can act
once or zero times on the ground state. If there were just one oscillator b−r there would
just be two states, |0i and b−r |0i with h = 0 and h = r. The single oscillator contribution
is thus Trq L0 = 1 + q r , where q = exp2πiτ . All oscillators with different modes act
independently, and it is easy to see that each contributes via additional factors of this
form. Furthermore we have to take into account the shift in the ground state energy.
For a free boson it is natural to assign −1/24 to each boson, so that for 24 transverse
bosons we get precisely the required −1. This gives the factor q 1/24 in the η-function.
For fermionic strings we have seen that the critical dimension is 10, so that there are 8
transverse dimensions. The shift in the ground state energy should be − 12 (see section
5.15), to which the bosons contribute −1/3. Hence the fermions must contribute −1/6,
or −1/48 per transverse fermion.
The result is thus, for each fermion,
∞
1 1
Y
L0 − 48 − 48
PNS (q) = Trq =q (1 + q r ) . (7.4)
r=1/2
by exactly the same arguments as used above. The normalization looks a bit strange,
since one would expect coefficients in partition functions to “count states”. However, for
N fermions we get PR (q)N , and there is then no problem for even N . If we include left-
and right-movers, we get PR (q)N PR (q̄)N , and then all coefficients are integers for any N .
103
numbers. For example, in the case of one coordinate q(t) we integrate over commuting
numbers q, not over operators. In the case of fermions, that would not be correct, because
even “classically” fermionic degrees of freedom must anti-commute. This can be taken into
account by defining integrals over “anti-commuting c-numbers”, also known as Grassmann
variables. Using this formalism one can define the path integral, and derive the relation
with a trace over a Hilbert space, as we did for bosonic variables. The result is almost
the same, except for one small detail:
Z
Dψe−SE (ψ) = Tre−βH (7.6)
ABC
Here “ABC” stands for anti-periodic boundary conditions. These are the partition func-
tions we have just computed.
We may also consider periodic boundary conditions for fermions, but then we get a
different answer. The answer can be obtained by flipping the sign of each fermion before
gluing the top of the cylinder to the bottom, thus changing the boundary condition from
anti-periodic to periodic. This can be achieved by inserting an operator into the trace.
This operator must have eigenvalue −1 for states made with an odd number of fermions,
and +1 for states made with an even number. To construct it we introduce the fermion
number operator F . X X
F = : br b−r ; F = : dr d−r (7.7)
r r
F
Now we insert into the trace the operator (−1) . This operator has the property
To compute such a trace we must know how (−1)F acts on the ground states. In the NS
sector it is natural to require that the vacuum has zero fermion number.
In the Ramond sector it is a bit more subtle. We have seen that d0 transforms the set of
ground states into themselves. On the other hand d0 changes the (−1)F eigenvalue when
it acts. The only way out is then that the ground states |ai must split into two subsets,
one with eigenvalue +1 and the other with eigenvalue −1.
It is quite easy to see how this insertion changes our expressions for the partition
functions. Each fermionic oscillator now contributes a factor 1 − q r instead of 1 + q r .
Hence ∞
1 1
Y
F L0 − 48 − 48
PNS (q)PBC = Tr(−1) q =q (1 − q r ) . (7.11)
r= 12
104
In the Ramond sector we get
PR (q)PBC = 0 (7.12)
since the trace over (−1)F on the ground state manifestly vanishes. It is easy to see that
the same is true for all excited states as well.
Now we can compute four kinds of fermionic partition functions, labeled by the peri-
odicities along each of the two cycles of the torus. For N fermions the result is shown in
the table below.
c N Q∞ N/2
AA TrNS q L0 − 24 q − 48 r N
r=1/2 (1 + q )
θ3
η
c N Q∞ N/2
AP TrNS (−1)F q L0 − 24 q − 48 r N
r=1/2 (1 − q )
θ4
η
c N Q∞ N/2
PA TrR q L0 − 24 2N/2 q 24 n N
n=1 (1 + q )
θ2
η
c
N/2
F L0 − 24 θ1
PP TrR (−1) q 0 η
Here the letters “AP” indicate anti-periodicity along the “space” direction and periodicity
along the “time” direction on the torus, etc. It turns out that these four partition functions
can be expressed in terms of standard mathematical functions, namely the Jacobi θ-
functions and the Dedekind η-function, which we have already seen. The θ-functions are
defined as follows hai 2
X
θ (z|τ ) = eiπ[(n+a) τ +2(n+a)(z+b)] (7.13)
b n∈Z
The last column above indicates the identification of each partition function with ratios
of θ- and η-functions. It is far from obvious that these ratios are equal to the formulas
we derived above. But one may check by expanding in q n that indeed equality holds for
any finite number of levels. Of course the equality can be proved to all orders, but we
will not do that here.
Note that the Jacobi θ-functions have two arguments, but we are only using them at
z = 0 here. The function θ1 (z|τ ) vanishes for z = 0, as does the partition function in the
PP sector, but it can be made plausible that the identification given here is the correct
one. To do this one may examine the dependence on the first argument, which can be
done by considering tori with external legs. Then the contribution from the “PP-sector”
does not vanish, and the identification with θ1 makes sense.
105
7.3 Modular Transformations
Now we discuss modular invariance. Clearly modular transformations change the fermion
boundary conditions. For example, the transformation τ → − τ1 interchanges the two
cycles (“space” and “time”) on the torus, and hence it interchanges AP and PA. The
transformation τ → τ + 1 maps (X,Y) to (X,XY) as shown in the figure1 , where X and Y
stand for A or P, and the multiplication rule is AA=P, AP=A and PP=P. In other words,
it interchanges AA and AP. Since these two transformations generate the modular group,
we generate all permutations of AA, AP and PA. On the other hand PP transforms into
itself. Since S and T generate the modular group we generate all permutations of AA,
AP and PA, whereas PP transforms into itself.
! X
!+1
Y
XY
The explicit form of the modular transformations can be worked out by means of the
Poisson resummation formula, applied to the θ-function form of the partition function.
Note that this form is rather reminiscent of a summation over a lattice, namely the lattice
of integers. Using this Poisson resummation for this lattice it is rather easy to derive
1 √ 1 √
θ1 (− ) = −i −iτ θ1 (τ ); θ2 (− ) = −iτ θ4 (τ );
τ τ (7.15)
1 √ 1 √
θ3 (− ) = −iτ θ3 (τ ); θ4 (− ) = −iτ θ2 (τ )
τ τ
θ1 (τ + 1) = eiπ/4 θ1 (τ ); θ2 (τ + 1) = eiπ/4 θ2 (τ );
(7.16)
θ3 (τ + 1) = θ4 (τ ); θ4 (τ + 1) = θ3 (τ )
We have already seen that
1 √
η(− ) = −iτ η(τ ); η(τ + 1) = eiπ/12 η(τ ) (7.17)
τ
1
To prevent confusion with the multiplication rule we have denoted boundary conditions here with the
notation (A, A), (A, P ), (P, A) and (P, P ), but from now one we will go back to he notation AA, AP, PA
and PP.
106
7.4 The Modular Invariant Partition Function
To construct something modular invariant we clearly need some sort of sum over the
various boundary conditions. Let us consider the correct number of fields for the fermionic
string, namely eight transverse bosons and fermions. The four partition functions are then
of the form
(θi (τ ))4
Pi (τ ) = 12 (7.18)
η (τ )
Under modular transformations
1 1
P1 (− ) = τ −4 P1 (τ ) ; P2 (− ) = τ −4 P4 (τ )
τ τ (7.19)
1 1
P4 (− ) = τ −4 P2 (τ ) ; P3 (− ) = τ −4 P3 (τ )
τ τ
P1 (τ + 1) = P1 (τ ) ; P2 (τ + 1) = P2 (τ )
(7.20)
P3 (τ + 1) = −P4 (τ ) ; P4 (τ + 1) = −P3 (τ )
(2−D)
The full partition function contains a factor [ Im (τ )] 2 just as for the bosonic string.
This cancels factors τ −4 in the τ → −1/τ transformation. Apart from that we see a sign
in the transformation between P3 and P4 . This is canceled by taking as the partition
function
1
PGSO (τ ) = (P3 (τ ) − P4 (τ ) − P2 (τ ) + xP1 (τ )) (7.21)
2
Note that the coefficient of P1 is not determined by modular invariance, because it only
transforms into itself. By considering modular invariance at two loops one does get a
constraint on x, and it turns out that x = ±1.
The full partition function is then
(2−D)
PII (τ, τ̄ ) = [ Im (τ )] 2 PGSO (τ )PGSO (−τ̄ ) (7.22)
Several points require some more discussion. In the foregoing paragraph “GSO” stands
for Gliozzi, Scherk and Olive, who invented this construction in a somewhat different way.
The subscript “II” stands for “type-II”, the name given for historical reasons to this kind
of string theory. Let us see what the linear combination does to the spectrum. In the
Neveu-Schwarz sector the combination
1 1 1
(P3 − P4 ) = TrNS (1 − (−1)F )q L0 − 2 (7.23)
2 2
is a partition function with a projection operator inserted. Indeed, the combination
1
2
(1 − (−1)F ) is a projection operator, that removes all states from the spectrum that are
created by and even number of fermion operators (the “GSO-projection”). This is good
news, because the tachyon corresponds to the state |0i with F = 0. Hence in this string
theory the tachyon is gone! At the first excited level we find the states
107
which stay in the spectrum (note that the normalization factor 12 ensures that there is
precisely one such multiplet). As we have already seen, these states are massless. Together
with the ones from the right-moving sector they produce a graviton, a dilaton and a B µν .
In the Ramond sector a similar projection occurs. Here we have
1 1
(P2 ± P1 ) = TrR (1 ± (−1)F )q L0 −1 . (7.25)
2 2
To interpret this we have to think about the action of (−1)F on the ground state. We
have seen that (−1)F anti-commutes with√all the di0 operators, and we have also seen that
di0 can be represented – up to a factor 2 – by the Dirac γ matrices. Then there is a
natural candidate for (−1)F , namely
Y
γ ∗ = iN/2 γi (7.26)
i
N/2 ∗
(the factor i is inserted to make sure that γ is Hermitean). This matrix is the general-
ization to N Euclidean1 dimensions of the familiar operator γ 5 in four dimension. In four
dimensions the projection operators 12 (1 ± γ 5 ) are also familiar. They project on fermions
with chirality + or −1. The former have their spin aligned with the direction of motion,
and the latter have it anti-aligned. States with definite chirality must be massless, since
otherwise we can flip the velocity relative to the spin by “overtaking” the particle.
In other dimensions the notion of spin changes, but chirality can still be defined,
namely as the eigenvalue of γ ∗ . Hence the operators 12 (1±γ ∗ ) project on the two chiralities.
In case of the 10-dimensional fermionic string the Ramond ground state has multiplicity
2(2−D)/2 = 16. It consists out of 8 states with chirality + and eight with chirality −. We
see thus that the GSO-projection selects half the states. We can choose the chirality by
selecting the + sign or the − sign in front of P1 . Since we can do that for left- and right-
moving modes separately, there are in fact four type-II strings, which we could denote as
II++ , II+− , II−+ and II−− . But by making a reflection in 10 dimensions we can flip all
chiralities simultaneously, so that II++ is equivalent to II−− and II+− to II−+ . Hence there
are really only two string theories, called type-IIA (opposite chiralities) and type-IIB (the
same chiralities).
Note that the Ramond partition function in PGSO is multiplied with −1. This is
good news, because we have seen that Ramond states are space-time fermions. A well-
known feature of fermions is that they contribute to loop diagrams with a −1 sign. This
is closely related to the spin-statistics relation. Since the torus diagram is a one-loop
diagram, bosons and fermions should contribute to it with opposite sign. Remarkably,
this is automatically guaranteed by modular invariance.
108
The essence of the argument goes as follows. Consider a two-loop diagram in the limit
where it almost separates into two one-loop tori. In that limit the two-loop “partition
function” factorizes into two one-loop partition functions, which are described by Jacobi
θ-functions. An n-loop surface has 2n independent loops that cannot be contracted to a
point. This is shown below for the three-torus.
b1 b2 b3
a1 a2 a3
On each loop the fermions can have boundary condition + or −. This gives a total of 22n
boundary conditions on an n-loop surface: Four for the torus, as we have seen, 16 for the
two-torus, etc. On the two-torus there are then 16 distinct terms in the partition function.
In the limit two-torus → torus × torus those 16 partition functions approach the product
Pi Pj . It turns out that modular transformations of the two-torus mix up the boundary
conditions. The 10 combinations (i, j) = (2 . . . 4, 2 . . . 4) and (1,1) are transformed into
each other, and the six combinations (1, 2 . . . 4) and (2 . . . 4, 1) are also mixed with each
other. Invariance under modular transformations at two loops thus requires the coefficient
of P1 P1 , x2 , to be equal to the coefficient of P3 P3 , namely 1 (there is no sign change in
this transformation).
7.6 Superstrings
The complete massless spectrum consists out of states coming from four sectors: NS-NS,
NS-R, R-NS and R-R. The NS-NS states produce the graviton, the dilaton and the B µν ,
as mentioned above. The NS-R and R-NS produce, in lightcone coordinates, a chiral
spinor times a vector. In the type-IIA string NS-R and R-NS sectors have states with
opposite space-time chirality, whereas in the type-IIB theory they have the same chirality.
A theory which is invariant under chirality-flip of all particles is called non-chiral. We see
thus that type-IIA is non-chiral, whereas type-IIB is chiral. Note that the standard model
is chiral: for example, left-handed fermions couple to the W boson whereas right-handed
fermions do not.
Covariantly, the product “spinor times vector” produces a gravitino and an ordinary
spinor with opposite chirality. In four dimensions a gravitino is a particle with spin 3/2.
In any dimension, the presence of a massless gravitino in the spectrum implies that the
spectrum is supersymmetric.
109
7.7 Supersymmetry and Supergravity∗
We have already discussed supersymmetry in two dimensions. The basic idea is the same
in any dimension. One introduces an infinitesimal transformation which takes bosons into
fermions, and vice-versa. In general there can be several supersymmetries.
In gravitational theories the supersymmetry acts in particular on the graviton. Each
supersymmetry produces one gravitino. In the present case we get in fact two gravi-
tini: one from R-NS and one from NS-R. One says therefore that the type-IIA and -IIB
theories have N=2 supersymmetry. Theories with gravity that are supersymmetric are
in fact invariant under local supersymmetry. This means that they are invariant under
infinitesimal transformations with a space-time dependent parameter (x). Such theories
are called supergravity theories.
Given a number of supersymmetries one can work out the possible super-multiplets,
combinations of bosons and fermions that are closed under the action of the supersym-
metry. For example in two dimensions a boson and a fermion form a supermultiplet.
As one may expect, the sizes of the multiplets increase with the number of supersym-
metries. In four dimensions, the maximum number is N=4 if one wants to restrict to
global supersymmetry (i.e. no graviton and gravitino), and N=8 if one does not want to
have particles of spin larger than 2. It turns out to be impossible to make interacting
theories with particles of spin larger than 2, so that the absolute maximum number of
supersymmetries in four dimensions is 8. Similar maxima exist in other dimensions; for
D = 5, 6 the maximum is N=4; for D = 7, 8, 9 and 10 the maximum is N=2, and for
D = 11 the maximum is 1. Beyond D = 11 there are no supergravity theories at all!
The word “abstruse” means “difficult to comprehend”, but to quote Gliozzi, Scherk and
Olive: “This identity is not abstruse to us because it is a consequence of supersymmetry
at the massive string levels”. Indeed, the identity is a consequence of the fact that at
110
every level the number of fermions and bosons is equal, and that bosons and fermions
contribute to the partition function with opposite sign!1
and require invariance under modular transformations. Assuming that the P1 contri-
butions behave as before we find that there are two independent solutions, namely the
type-II strings seen above and furthermore (with a P1 -term added)
1 (2−D)
P0 (τ, τ̄ ) = [ Im (τ )] 2 [P3 (τ )P3 (−τ̄ ) + P4 (τ )P4 (−τ̄ ) + P2 (τ )P2 (−τ̄ ) ± P1 (τ )P1 (−τ̄ )]
2
(7.29)
This is called the type-0 string theory. It also comes in two kinds, namely type-0A and
type-0B, which have opposite signs in the last term. These theories have a tachyon,
because the NS ground states is not projected out, and they have no space-time fermions
since there are no NS-R and R-NS sectors. They are not supersymmetric. In fact it can
be shown that supersymmetry forbids a tachyon. The converse is not true: there do exist
non-supersymmetric strings without tachyons.
1
Technical remark: Only physical boson and fermion degrees of freedom should be counted. These are
also called the on-shell degrees of freedom. For example a massless vector boson has D−2 physical degrees
of freedom. In the light-cone gauge one only sees physical degrees of freedom, so that one automatically
obtains the correct count. 2 perhaps surprisingly, rank-4 tensors in 10 dimension can have a chirality
+ or −, just like fermions, and one finds a rank-4 tensor of chirality + or − (depending in whether one
considers the II++ or the II−− version of the IIB string). Such tensors are called self-dual or anti-self-dual.
111
7.11 Internal Fermions
When we discussed fermions on the world-sheet we started with fermions ψ A that have
an index A rather than a space-time index µ. If one builds theories with such fermions,
the issue of modular invariance also arises. But we lose an important restriction. If µ
is a space-time index, fermions with different values of µ must have the same boundary
conditions, because otherwise Lorentz invariance would be broken. But there is no such
condition for fermions ψ A . They can have boundary conditions that are different for
different values of A. To distinguish these fermions ψ A from the fermions ψ µ the former
are often called “internal” fermions (see also sec. 5.1).
Now that we are considering internal fermions ψ A we are have returned to the bosonic
string. Note that the decoupling of unphysical states from ψ 0 , or equivalently the require-
ment of Lorentz invariance (see sec. 5.15) reduced the dimension of the string theory from
D = 26 to D = 10. But there is no such condition from fermions ψ A . We have seen that
the presence of 2N fermions reduces the dimension from 26 to 26 − N , so that we are
effectively talking about a compactified bosonic string.
θi (τ )
Fi (τ ) = , (7.30)
η(τ )
and similarly for the right-moving modes. A function Fi represents a pair of fermions. A
partition function of a fermionic string contains some combination of N such functions
times a product of 24 − N functions η −1 originating from the transverse bosons X i . Hence
the full partition function is a linear combination of terms of the form
N N
1 Y 1 Y
Fi (τ ) = θi (τ ) (7.31)
η(τ )24−N m=1 m η(τ )24 m=1 m
where the sum is over all possible choices of θ-function labels i = 1, 2, 3, 4 for each fermion
pair. Solving the conditions for modular invariance for this system leads to a huge number
of possibilities.
112
7.11.2 Factorized Partition Functions
For simplicity we will only consider possibilities where left- and right-movers are separately
modular invariant. One can try a linear combination of terms of the form (Fi )N . From
the transformation of θ2 under τ → τ + 1 we see then that N must be a multiple of 8:
otherwise there can be no θ2 term, and then the others must vanish as well. If N = 8
there is indeed a simple solution, which we write in terms of the ratio’s F as
1
(F3 )8 + (F4 )8 + (F2 )8 ± (F1 )8
R8 = (7.33)
2
The normalization is chosen in such a way that the vacuum state occurs only once. If
N = 16 there are two solutions, namely (R8 )2 and R16 , the latter defined in the obvious
way.
The sign of the last term can be + or −, just as the sign of the P1 term in the GSO
partition function. The reason is the same: it is not fixed by one-loop modular invariance,
and fixed in absolute value by two-loop modular invariance.
However, this is not a proof that this is all that is possible. In general, a solution may
contain combinations of factors Fi and Fj , i 6= j, as the (R8 )2 example already shows. In
this way one can write down a solution that is essentially the square of two GSO partition
functions: 2
(F3 )4 − (F4 )4 − (F2 )4
(7.34)
If we were to use this kind of partition function we would encounter the problem that
some mixed terms with factors (F2 )4 and (F3 )4 appear with a minus sign. In the fermionic
string that minus sign gives the correct spin-statistics to space-time spinors. But if the
fermions are internal, they do not transform under the space-time Lorentz group, and
hence the Ramond states are not space-time spinors. In the spectrum, they appear as
scalars (or higher rank tensors) with wrong statistics. This problem cannot be solved
using one loop considerations alone. It turns out that partition functions like Eq. (7.34)
violate the modular invariance conditions for two loop diagrams. Indeed, one can show
that if those conditions are imposed, one always gets theories with a correct spin-statistics
relation.
Note that in Eq. (7.34) we cannot use a single GSO-like factor, because the string
theory with internal fermions is essentially a bosonic string, whereas the theory with space-
time fermions is a fermionic string. This implies that the overall number of η function
factors in the partition function is different. Indeed, terms in the partition functions of
these two theories have the general form
η(τ )−24 θN
η(τ )−12 θ4
for internal and space-time fermions respectively. In the latter case the number of fermion
pairs is restricted to 4 = (D − 2)/2, whereas in the former case it could in principle be any
integer N , including 4. The modular transformations of the η functions under τ → τ + 1
give an extra minus sign in the latter case, which explains why the solutions to one-loop
modular invariance are different.
113
7.11.3 Symmetries of Free Fermions
The free fermion Lagrangian for 2N fermions is symmetric under SO(2N ) rotations
X
ψA = OAB ψB (7.35)
B
where O is an orthogonal matrix. (Note that since ψA is real SU (2N ) is not a symmetry).
This symmetry can be seen in the spectrum provided that the boundary conditions respect
it. In other words, only if all 2N fermions always have the same boundary conditions.
If we consider for example the R16 theory we expect the states in the spectrum to form
SO(32) multiplets. Indeed, bA 1/2 transforms as an SO(32) vector. For the R8 theory we
would only expect an SO(16) symmetry (but see the discussion below).
Note that there is actually an O(16) symmetry in the lagrangian. Here and in the
following we will not worry about the distinction between O(16) and SO(16) since we are
going to be interested in gauge interaction, and they are not sensitive to this distinction.
So we will just denote all these orthogonal groups by SO(N ). A full understanding of
non-perturbative string theory will have to decide if there is an “S” or not.
The foregoing discussion is valid for either left-moving or right-moving fermions ψ A .
If there is a world-sheet fermion ψ A (σ, τ ) which depends on both σ and τ we can split it
into left- and right-moving components: ψ A (σ, τ ) = ψLA (σ + τ ) + ψRA (σ − τ ). Since these
components do not mix in the two-dimensional action, we can actually SO(2N )-rotate
them independently. So then we have an SO(2N ) × SO(2N ) symmetry. However, as we
will see in the next section, there are string theories where only the left-movers (or only
the right-movers) are used.
114
1 8
2
2 = 128 components. Note that one could try to be extremely precise by distinguishing
partition functions R8+ R8+ , R8+ R8− etc. in (7.36), thinking of the analogy with type-IIA and
type-IIB strings. But observe that there is an important difference. In type-II strings the
left-moving and right-moving spinors ψ µ are transformed by the same symmetry group
SO(10) acting on the label µ, which is linked to space-time. But here we have two dis-
tinct groups acting on the labels A of ψ A . This is because ψ A splits into left-moving and
right-moving components, as explained above. Hence the rotations of the labels A are
not linked. Then there does not exist a useful notion of relative chirality of the spinors,
as there was for type-II strings.
If we combine the NS and R sectors we get a total of 120+128=248 space-time scalars.
µ
Furthermore there is a space-time vector created by α−1 . In a short-hand notation we
could denote the result as (V + 248) × (V + 248) When we combine left and right sector
we get the following massless states: 248+248 vector bosons, 248 × 248 scalars, and of
course the graviton, the dilaton and a Bµν .
The presence of massless vector bosons suggests a gauge symmetry. Indeed, to get
the correct counting of degrees of freedom for a massless vector boson we need a gauge
symmetry:
Aaµ → Aaµ + ∂µ Λa + . . . (7.37)
This implies that we must in fact have gauge symmetries for every value of the index a
of Aaµ . At this point, all this tells us that we need a gauge symmetry with 248 + 248
generators, because we have 248 gauge bosons from the left-moving R8 and another 248
from the right-moving R8 . So perhaps the gauge group is U (1)248 × U (1)248 ? To really
determine what it is one has to consider the three-point couplings of the gauge bosons.
This reveals that the Lie algebra of the gauge group is in fact E8 × E8 , where E8 is the
exceptional Lie algebra of rank 8. This implies that the usual non-abelian terms will be
present in addition to the term shown eqn. (7.37). The algebra E8 does indeed have
exactly 248 generators. It has as its sub-algebra D8 , a Lie algebra whose corresponding
Lie group is SO(16). This is of course the symmetry group identified in the previous sub-
section. It turns out that the presence of N world-sheet fermions ψ N gives generically a
SO(N ) gauge symmetry in the string spectrum derived from it, but in some exceptional
cases there can be additional gauge bosons that enhance the gauge symmetry. This is
such an exceptional case.
Using R16 of (R8 )2 we can make theories in ten dimensions. Three different ones in fact,
because we may consider [R16 (τ )][R16 (−τ̄ )], [R16 (τ )][R8 (−τ̄ )]2 and [R8 (τ )]2 [R8 (−τ̄ )]2 . We
will see the spectrum of the distinct two left-moving theories below in more detail.
115
consequence of the following identity
8 8 8 !
1 θ3 θ4 θ2 1 X πiτ~v2
+ + = e (7.38)
2 η η η η8
~v ∈E8
This identity is easy to prove if we write θi as a sum, as in Eq. (7.13). This equivalence is
an example of bosonization. On the left hand side we have a theory constructed out of 16
free fermions, on the right hand side a theory constructed out of 8 free bosons. Indeed,
in general a pair of fermions with appropriate boundary conditions is equivalent to a free
boson on an appropriate torus.
Similarly the two fermionic theories in 26 − 16 = 10 dimensions can be identified with
the E8 × E8 and E16 Euclidean ESDL’s in 16 dimensions. If we combine left and right-
movers we can in fact built three theories: (E8 × E8 )L × (E8 × E8 )R , (E16 )L × (E16 )R and
also the “hybrid” combination (E8 × E8 )L × (E16 )R . The partition functions are of course
for the (E16 )L × (E16 )R theory, and obvious variations on this theme for the other two.
It should be emphasized that there are far more bosonic string compactifications, as
we have already seen in subsection 7.11.1; we have only considered here the ones that
have separate modular invariance in the left and right.
8 Heterotic Strings
The foregoing discussion leads us naturally to the idea of combining the right sector of
the type-II string with the left sector of one of the 10-dimensional compactified bosonic
strings. Since both are separately modular invariant in the left and right sectors, cutting
each theory in half and gluing the parts back together should give a modular invariant
theory. This thought is indeed correct, and the result was so important that a new name
was given to it: the heterotic string.
Since there are two distinct left-moving partition functions in the 10-dimensional
bosonic string, we can build two heterotic strings. The partition functions are given
by
1
4
PGSO (−τ̄ )η(τ )−8 R8 (τ )R8 (τ ) (8.1)
( Im τ )
and
1
PGSO (−τ̄ )η(τ )−8 R16 (τ ) (8.2)
( Im τ )4
These two are called respectively the E8 × E8 heterotic string and the SO(32) heterotic
string. These names refer to the gauge group of these theories, as we will explain in a
moment. In principle we also have the option of choosing the sign in front of P1 in PGSO .
But this only flips the chirality of all the fermions, and has no observable effect, because
there is no other chirality to compare with.
116
8.1 The Massless Spectrum
Note first of all that the spectrum does not contain a tachyon. The left-moving bosonic
sector contains one, but the right-moving superstring does not. Hence there is no physical
tachyon state, because the condition ML = MR cannot be satisfied.
Analyzing the spectrum of the heterotic strings is easy. The right sector gives us a
vector representation and a spinor representation of the transverse SO(8) group in 10
dimensions. The left sector yields a vector of SO(8) plus a number of scalars from the
(R8 )2 or R16 part of the partition function. These scalars come from the NS parts in both
cases, and from the R part in the case of R8 . The total number is
2 1 1 8
(R8 ) : 2× (16 × 15)NS + (2 )R = 496
2 2 (8.3)
1
(R16 ) : (32 × 31)NS = 496
2
Note that in the first case the subscript “R” refers only to the first (or second) group of
16 fermions. The other group of 16 is always in the NS ground state. The total ground
state mass contribution for the 8 bosons, 16 NS fermions and 16 R fermions is thus
1 1 1
(8 × (− 24 ) + 16 × (− 48 ) + 16 × (+ 24 )) = 0. For the second case the analogous computation
1 1
yields (8 × (− 24 ) + 32 × (+ 24 ) = 1, and hence the Ramond ground state is massive.
The complete massless spectrum is thus (V =vector and S =spinor)
(V + 496) × (V + S) (8.4)
which yields the usual graviton, Bµν and dilaton (from V × V ) and 496 massless vector
bosons and 496 additional spinors. Furthermore, the tensor product V × S yields a
gravitino. (To be precise, this tensor product yields a gravitino of one chirality an ordinary
spinor with opposite chirality).
8.3 Supersymmetry∗
The presence of the gravitino suggests that there must be supersymmetry. Indeed, this
theory has N = 1 supersymmetry (only N = 1 because in this case there is only one grav-
itino). As for type-II it is in fact a local supersymmetry, i.e. supergravity. Any symmetry
117
must act on all fields in a theory. Since supersymmetry switches bosons an fermions, it
cannot act trivially. Hence it must act non-trivially on the 496 vector bosons. Obviously,
they are transformed into the 496 fermions and vice-versa. In any supersymmetric theory
vector bosons are accompanied by fermions in the same representation (i.e. the adjoint
representation). These fermions are called gaugino’s (in the supersymmetric extensions
of the standard model one speaks of photino’s, gluino’s, Wino’s and Zino’s; experiments
are looking for such particles, but so far they haven’t been found).
The sum is over all components of the fermion field. The matrices T a form a representation
of the Lie-algebra, which means that they satisfy the commutation relations. The number
of components of ψ I is called the dimension of the representation.
In the present case, the gauge symmetry includes the global SO(2N ) symmetry that
rotates the world-sheet fermions into each other. In the case of the SO(32) heterotic string
this is indeed precisely the full gauge symmetry. However, in the case of the E8 × E8
heterotic string one would expect only an SO(16) × SO(16) symmetry on the basis of
this argument. Here something special happens. From the Ramond sector we get (two
times) 128 extra massless vector bosons. As a result, the gauge symmetry gets extended
as well. The extension is described by a Lie algebra called E8 , which has 248 generators
and contains SO(16) as a sub-algebra. The gaugino’s are in the same representation as
the gauge bosons, the adjoint representation.
The spinor S of SO(8) is chiral because of the GSO-projection in the right sector. It
follows that the gaugino’s are chiral as well (as is in fact the gravitino).
8.5 Compactification∗
When it was discovered in 1984, the heterotic string was almost immediately seen as an
excellent theory for particle physics. The main reasons are that it has a sufficiently large
gauge symmetry to contain the standard model, and that it has massless chiral fermions.
For some people even the presence of supersymmetry in the spectrum is an advantage,
118
although it has not (yet) been seen in nature. The reason is that already at that time
supersymmetry was considered as an attractive mechanism to solve the so called hierarchy
problem. Simply stated, this is the question how to explain the large discrepancy between
the weak interaction scale (≈ 100 GeV) and the natural scale of quantum gravity, the
Planck scale (≈ 1019 GeV) (in fact, supersymmetry does not explain that discrepancy, but
explains why it is stable against quantum fluctuations). Obviously, since supersymmetry
is not observed it can at best be an approximate or “broken” symmetry, but from the
Planck scale point of view it is to first approximation exact, and the breaking is only a
higher order correction, which is usually ignored in string phenomenology.
An observer in four dimensions only sees the first four components, and views the internal
components as a multiplicity. Hence he/she sees a vector and N scalars.
The only difference with the Minkowski case are a few factors of i in some expressions.
Infinitesimal rotations of these spinors are generated by the matrices
1
Σµν = − i [γ µ , γ ν ] , (8.10)
4
where γ µ are the Dirac matrices.
119
We assume that both D and N are even. In M even dimensions, the Dirac matrices
are 2M/2 × 2M/2 matrices. They satisfy the Clifford algebra
{γ µ , γ ν } = 2δ µν (8.11)
We will choose them Hermitean. Since we have three different spaces, we have to introduce
three different set of Dirac matrices:
SO(D) : ΓA , A = 1, . . . , D
SO(D − N ) : γ µ , µ = 1, . . . , D − N (8.12)
SO(N ) : γ I , I = D − N + 1, . . . , D
All these matrices can be written down explicitly in many different ways, but it can be
shown that all choices are equivalent.
In even dimensions there are two kinds of fundamental spinor representations. They
are distinguished by the eigenvalue of the matrix γ ∗ , the product of all Dirac γ matrices
(usually called γ5 in four dimensions).
M
Y
∗ M/2
γ =i γi (8.13)
i=1
120
The two chirality eigenspaces of Γ∗ are projected out by the operators
1
PD± = (1 ± Γ∗ ) (8.18)
2
Writing this in terms of the subgroup we find
−
PD+ = PD−N
+
⊗ PN+ + PD−N ⊗ PN− for N = 0 mod 4 (8.19)
and
PD+ = PD−N
+
⊗ PN− + PD−N
−
⊗ PN+ for N = 2 mod 4 (8.20)
Denoting the spinor representations as S (chirality +) and C (chirality −) we may write
this in the case of interest as
S10 → (S4 , C6 ) + (C4 , S6 ) (8.21)
This is the decomposition of the spinor representation, analogous to the one of the vector,
Eq. (8.7). Such “branching rules” exist for any representation of SO(D) (or any other
group and subgroup), but the vector and spinor decompositions are all we need.
121
coupling for anti-chiral spinors. This is analogous to the Standard Model weak interaction
couplings, where only the left-handed electron and neutrino couple to the W -boson. Even
if the spinor is chiral in ten dimensions, in a torus compactification it splits as in Eq. (8.21).
If furthermore the internal spinors S6 and C6 have the same number of zero eigenvalues
for the Dirac equation (usually called “zero-modes”) an observer in four dimensions sees
the same number of chiral and anti-chiral spinors S4 and C4 coupled to the same gauge
group. Hence the chiral and anti-chiral spinors are in the same representation of any
gauge group, and we end up with a non-chiral theory: we cannot get the standard model.
The foregoing problem is fortunately special to torus compactifications. If one com-
pactifies on other manifolds, a chiral spinor decomposes in a certain number NL of chiral
spinors and another number NR of anti-chiral ones.
On other spaces one has to generalize Eq. (8.22) to take into account curvature. This
implies that the ordinary derivatives ∂ I are replaced by covariant derivatives which contain
the spin connection of the manifold (this was trivial for the torus, since a torus is flat). An
additional complication is that sometimes one allows some of the fields in the space-time
theory to have classical values (an example of that was considered in the discussion of
torus compactification where we allowed a non-vanishing – but constant – background
field BIJ ). Often one chooses a non-vanishing classical gauge field, and in that case the
covariant derivative must also contain a gauge potential term. All of these issues are
beyond the scope of these lectures. However, it is clear that in general a different, more
complicated equation must be solved in the internal dimensions. In general NL 6= NR so
that a chiral theory may be obtained after compactification.
122
8.7 The Standard Model in String Theory∗
We must first review a few basic properties of the standard model. It contains particles of
spin 1/2, the quarks and leptons, particles of spin 1, the carriers of the electromagnetic,
strong and weak forces, and a particle of spin 0, the Higgs boson. The spin-1 bosons
belong to a gauge group SU (3) × SU (2) × U (1), which below about 100 GeV is broken
down to SU (3) × U (1), QCD and electrodynamics. If we ignore the weak interactions the
standard model is non-chiral: QED and QCD couple in the same way to both chiralities.
However, the weak interactions do distinguish chirality, and couple only to the left-handed
and not to the right-handed electron.
A remarkable feature of the standard model is that the particles are grouped in families:
it is natural to associate (e, νe , u, d), (µ, νµ , c, s) and (τ, ντ , t, b) with each other. Although
this grouping is rather arbitrary (it is determined only by the mass hierarchy, which is not
understood, and is furthermore disturbed by mixing), it is certainly true that the basic
SU (3) × SU (2) × U (1) quantum number repeat three times.
When writing one family of the standard model, it is sometimes useful to replace
certain particles by their anti-particles. In general, anti-particles have opposite charge and
complex conjugate non-abelian group representations. Furthermore in four dimensions
(and in all 4m dimensions) particles and anti-particles have opposite chirality (in 4m + 2
dimensions they have the same chirality). For massless particles chirality is equivalent to
helicity: a particle with chirality + is left-handed. Hence, for example, the anti-particle of
e− +
R is eL . This can be used to make all particles in a standard model family left-handed. If
one does that, the SU (3)×SU (2)×U (1) representations are (we are ignoring right-handed
neutrinos here)
Here the “bar” denotes the conjugate of a representation, and representations are denoted
as (dimension).
The standard model gauge group can be embedded into larger groups. The most
familiar embedding chain is
123
The idea is that just as the low energy gauge group SU (3) × U (1) is extended at 100
GeV to the standard model gauge group SU (3) × SU (2) × U (1), perhaps at still higher
energies the standard model gauge group is enlarged further. This can even be done in
such a way that there is just one gauge group and one representation per family. From
this point of view SO(10) looks perhaps most attractive, but the next step, E6 , has also
been studied in the literature.
The point is now that there is a class of compactifications of the E8 ×E8 heterotic string
in which the gauge group is broken to E6 × E8 . Furthermore in these compactifications
the massless fermions are automatically in the representation (27). One can have NL
left-handed fermions and NR right-handed ones, with NL 6= NR . If we represent the
right-handed ones by anti-particles, we may also write that as NL (27)L + NR (27)L , i.e.
with only left-handed particles.
In switching from particles to anti-particles one has to be careful with the notion of
chirality of a theory. The rule is simple: if after flipping all chiralities one obtains a
different theory, the theory is chiral. However, if we can write one theory in the form of
another one by trading some particles for anti-particles, then those two theories are not
different. The following table shows some examples:
Here the first column shows a representation, and the second column shows the parity-
flipped representation, and how it changes under particle - anti-particle interchange. In
the last line the representation (10) of SO(10) appears. This is the vector representation,
124
and it is real: (10) = (10). Hence a four-dimensional fermion in such a representation is
not chiral.
Chiral particles cannot have a mass. The reason is that in a mass term of the form
mψ̄ψ the chiral projection operator is flipped:
†
ψ̄ 12 (1 + γ5 )ψ ≡ ψ † γ4 12 (1 + γ5 )ψ = ψ † 12 (1 − γ5 )γ4 ψ =
1
2
(1 − γ5 )ψ γ4 ψ (8.23)
Hence if the second ψ has chirality +, the first must have chirality −. This implies that
one cannot write down a mass term for (27)L unless there is also a (27)R . On the other
hand, if both (27)L and (27)R are present a mass term is allowed.
An often used principle in quantum field theory is “anything that is allowed is obliga-
tory”. In other words, if there is no reason why a certain parameter should be zero, then
it is reasonable to assume that it is not zero. On the basis of this principle one usually as-
sumes that if for example NL > NR , then NR (27)R ’s combine with NL (27)L ’s to acquire
a large mass (in this context large means “too large to be observed”), so that there are
NL − NR unpaired (27)’s left over. They cannot get a mass, and hence they must appear
as particles in the low-energy spectrum. By the same logic, when E6 break to SO(10) the
leftover (10) + (1) are assumed to acquire a mass of order the symmetry breaking scale
(which must be much larger than 100 GeV). The leftover (1) in the breaking to SU (5) can
also become massive, so that finally only the standard model particles remain massless.
If we accept all this, then we must be looking for a Calabi-Yau manifold with |NL −
NR | = 3. One of the best studied examples has NL = 1 and NR = 101, but thousands of
Calabi-Yau manifolds have been found, including some yielding three families. In addition
there are many variations on the basic idea. These allow breaking E6 to something closer
to the standard model gauge group, while at the same time lowering the number of
families. Although there are many examples that get rather close to the standard model,
there is none that really “fits like a glove”. On the other hand, nothing like a systematic
search has been possible so far due to the overwhelming number of possibilities.
Furthermore it is difficult to do detailed calculations. For example, one may wish
to compute the quark and lepton masses. They get their masses from a three-point
coupling to the Higgs scalar, λx ψ̄x φψx , where “x” denote a particle, and λx is a coupling
constant. This is called a Yukawa coupling. The mass of a particle is equal to the vacuum
expectation value of the Higgs – a universal parameter with the dimension of mass –
times λx . The values of λx can be computed in string theory, but the problem is that the
massless spectrum typically contains a few hundred scalars, so that it is not clear which
one is going to play the rôle of the Higgs scalar φ. In addition the Higgs potential and
the whole symmetry breaking mechanism is poorly understood within string theory. One
has tried to remedy this by adding some field-theoretic assumptions, but pretty soon any
predictive power is completely lost. It seems that in making contact with reality one still
has a long way to go.
However, it is encouraging that the gross features of the standard model are repro-
duced, and that they come out essentially automatically from a theory that in principle
could only hope to be a theory of gravity. These features are
125
• Chiral fermions
• Replication of families
Here the right-moving factors (with subscript R) have an implicit argument τ̄ whereas the
left-moving ones (L) have argument τ . In the left-moving sector the fermions are grouped
into two groups of 16, and e.g. in (F38 − F48 )2 the first factor refers to the first group of 16
and the second factor to the second.
From the partition function one reads off that the gauge group is SO(16) × SO(16),
and is not extended, and that there are no gravitino’s, and hence no supersymmetry. Fur-
thermore there is no tachyon. This shows that it is not necessary to have supersymmetry
to get rid of the tachyon.
There are six other non-supersymmetric heterotic string theories in ten dimensions,
but the other six all have a tachyon.
126
gauge there is an SO(8) symmetry: all partition functions depend on (Fi (−τ̄ ))4 . If instead
we allow combinations like Fi1 Fi2 Fi3 Fi4 with different labels im , then clearly the SO(8)
symmetry is broken, in general to SO(2)4 (one SO(2) for each pair of fermions). If we
associate the first factor Fi1 with the light-cone components of the fermions ψ µ , then the
resulting string theory is four-dimensional.
The difficulty in constructing such a theory is to construct a modular invariant par-
tition function out of the internal degrees of freedom. Concretely, suppose one the right
the 10-dimensional space-time index is split into (µ, M ), M = 1 . . . , 6 and on the left the
26-dimensional space-time index is split into (µ, I), I = 1 . . . , 22, then the internal degrees
of freedom are ψ M , X M and X I . Now the internal degrees of freedom are partly bosons
and partly fermions, and it is difficult to build modular invariant partition functions in
such a hybrid system. The solution is bosonization: in 2 dimension one (compactified)
boson can be replaced by a pair of fermions or vice-versa. Hence we can either replace ψ M
by 3 bosons and construct a modular invariant partition function out of free bosons (this
leads to a description in terms of Lorentzian even self-dual lattices) or replace X M and
X I by 12 and 44 fermions. Then one proceeds as above, by trying to construct modular
invariant combinations of fermion partition functions. A technical complication in both
cases is how to preserve the two-dimensional supersymmetry that relates ψ M and X M .
We will not give further details here. Both methods work, and yield huge numbers of
solutions. So many solutions that a complete listing is impossible. In any case one can
get chiral and non-chiral theories, a large number of possible gauge groups, theories with
and without tachyons and with any allowed number of supersymmetries.
It may seem that there are two different ways to get to four dimensions: compactify a
ten-dimensional string, or construct a string theory directly in 4 dimensions. It turns out
that these two methods are closely related. In many examples a given four-dimensional
string theory yields exactly the same spectrum as a definite Calabi-Yau compactification.
It is believed that then not only the spectrum but also all interactions are the same: the
two theories are identical.
There are other methods for constructing strings in four dimensions, but they are all
(believed to be) connected to each other. The full set of four-dimensional theories is a
complicated parameter space that is still to a large extent unexplored.
127
match anything that has been observed. There are usually extra gauge bosons, and also
additional matter. It is possible to construct examples where all the extra matter is
non-chiral. Then at least it can acquire a mass in principle. One can also construct field-
theoretic mechanisms to break the additional gauge symmetries; basically this requires
some additional Higgs mechanisms. But some problems are even more serious. The two
most important problems are:
• Space-time supersymmetry
• Moduli
Both of these are generic features of string spectra that are not observed.
128
of the gravitational attraction. There are severe limits on such forces. Furthermore, it
is know that light scalars (with a mass smaller than about 10 TeV) lead to cosmological
problems.
All observable parameters of the Standard Model depend on the value of the vacuum
expectations of the moduli. Hence if we want to say something about the gauge couplings,
or the Yukawa couplings, or the parameters of the Higgs potential, we have to understand
how the vacuum expectation values of the moduli are fixed.
This is known as the moduli stabilization problem. The idea is that the flat moduli
potential one observes in the supersymmetric theory is replaced by a non-trivial potential
with one or more local minima. The scalars will move to such a local minimum, and
all parameters of the resulting gauge theory are fixed to some value. However, typically
there are hundreds of moduli. To find local minima of a potential depending on hundreds
of fields is a horrendous problem even if the potential is known exactly. But on top of
that one has to determine all the terms in the potential. This potential can in principle
have an infinite number of terms, because unlike the famous Mexican-hat-shaped Higgs
potential, there is no reason why it should be of at most fourth order in the scalar fields.
The latter requirement is due to renormalizability, but at the Planck scale that criterion
is irrelevant. It goes without saying that there is no hope of computing, for example, the
electron mass from string theory.
129
vacuum energy, is irrelevant for quantum field theory, but not to gravity. Gravity sees ev-
erything. A positive vacuum energy density leads to accelerated expansion of the universe
(mathematically described by a deSitter space), whereas a negative value leads to collapse
of the universe (Anti-deSitter space). Towards the end of last century evidence was found
for accelerated expansion, which one may describe by assuming a vacuum energy. The
trouble is that the natural value for the vacuum energy density one might expect in a
theory of quantum gravity is about a Planck energy per Planck volume, EPlanck /VPlanck .
The observed value for the vacuum energy is much smaller than that: it is smaller by a
factor 10−123 .
It is quite clear that we could never have observed a universe with a vacuum energy of
natural, Planckian order. It would either collapse in a Planck time, or double its size in a
Planck time (about 10−44 seconds). One may ask how small the vacuum energy must be
in order to get a universe that is more or less friendly to its inhabitants. This computation
has been done, and it turns out that if the sign is negative, and with an absolute value
just a few orders of magnitude large than the observed value, the Universe would have
collapsed long before evolution would have produced anything interesting. If the sign is
positive, and just three orders of magnitude larger than the observed value, the universe
expands so rapidly that galaxies are ripped apart before they are formed. In either case,
there would not be any creature living in the universe to observe it.
130
9.2.3 Eternal Inflation
All of this is linked to ideas in inflation, which seem to imply that apart from our own
universe, many other universes may have formed as well. Our own universe would just be
a tiny bubble in a “multiverse”. Some universes in that multiverse would be expanding
indefinitely (unlike our own, where the expansion process, inflation, came to a halt at
a certain moment). This is called “eternal inflation”. During its expansion, such an
eternally inflating universe creates baby universes starting as small, expanding bubbles,
analogous to a boiling liquid. Each universe would end up in one of the local minima
of the moduli potential, so that each universe has different values for the vacuum energy
density and the Standard Model parameters. If our Standard Model is one of the local
minima, inevitably it is selected at some point. Perhaps that is all we can ever say about
the values of the Standard Model parameters.
10 Dualities∗
In ten dimensions there are connections among seemingly different string theories. The
theories involved in these “dualities” are the ten-dimensional supersymmetric string theo-
ries: they include the type-IIA, type-IIB, heterotic SO(32) and heterotic E8 × E8 theories
introduced in the previous chapter. Before discussing dualities, we have to introduce one
more family member, namely
131
10.1 Open Strings
All theories seen so far are oriented closed strings. There exists also a theory of unoriented
open and closed strings in ten dimensions that satisfies all known consistency conditions.
Just as one of the heterotic strings, this theory has N = 1 supersymmetry and an SO(32)
gauge group. It is called the type-I string. The reason why SO(32) appears is rather
different than in the case of heterotic strings. In that case the reason was modular
invariance.
10.2 Finiteness
As we have seen, modular invariance for closed strings allowed us to restrict the one-loop
integration to one “modular domain”, depicted in section (5.9). Furthermore we have
seen that this domain excludes the dangerous Im τ = 0 limit where the field theory
divergences are. Then, if there is no tachyon, the one loop integral is finite. Indeed,
the two supersymmetric heterotic strings have finite one-loop integral, and so does the
non-supersymmetric SO(16) × SO(16) string.
Furthermore the one-loop integral (without external legs) is not just finite, but zero for
superstrings, because the partition function (the integrand) vanishes due to the “abstruse
identity”. This is important for higher loop divergences.
Consider for example the two-loop diagram, the double torus. The modular integration
for the double torus contains a limit where the double torus looks like this
One may show that in the limit where the size of the tube connecting the tori goes to zero
and the length of that tube goes to infinity, only one particle contributes: the dilaton. In
that limit the amplitude factorizes into two so-called tadpole diagrams.
132
It can be shown that if the one-loop diagram without external lines give a contribution
Λ, then the dilaton tadpole is proportional to Λ. The the limit of the double torus diagram,
treated as a field theory Feynman diagram with one propagator ending on both sides on a
one-point vertex, yields something like Λ[1/p2 ]Λ. However, momentum conservation tells
us that p2 = 0. This implies that the double torus contribution becomes infinite, unless
Λ = 0. In that case this simple argument is inconclusive.
Although the discussion here was not very precise, the argument is indeed correct.
The SO(16) × SO(16) heterotic string has a finite, non-zero Λ. Therefore it is finite at
one loop, but divergent at two loops. The supersymmetric heterotic strings have Λ = 0
at one loop. It turns out that the full two-loop integral again yields zero, which does not
follow from the foregoing discussion, but is consistent with it. In fact the corresponding
integral vanishes at any loop order. Diagrams with external lines do not vanish, but give
a finite answer at any loop order (or at least that is what is generally believed, but only
partly proved).
In the case of open and unoriented strings the same arguments apply to the closed
sub-sector, i.e. the torus and higher order generalizations, but the open and non-oriented
diagrams are a separate story. At the lowest non-trivial order, Euler number 1, there two
surfaces, characterized by b = 1 and another by c = 1 (see eq. (6.10)). To both surfaces
one can attach an external dilaton, so that one obtains two dilaton tadpole diagrams (disk
and crosscap)
The point is now that the contributions of these two dilaton tadpoles must cancel. If not,
then they lead to a divergence at Euler number zero due to a dilaton propagator in an
intermediate line. At Euler number zero the relevant diagrams are annulus, Klein bottle
and Möbius strip. It turns out that the integrations corresponding to these diagrams
have limits that look like a dilaton propagation between two disks (annulus), between
two crosscaps (Klein bottle) and between a disk and a crosscap (Möbius strip). Summing
these diagrams we get, schematically
1 1 1 1 1
D[ 2
]D + C[ 2 ]C + C[ 2 ]D + D[ 2 ]C = (D + C)[ 2 ](D + C) . (10.1)
p p p p p
133
that was introduced before. Then, in the same units, the disk contribution D is equal to
N , and the tadpoles cancel if N = 32. In that case one finds precisely the gauge bosons
of SO(32) in the spectrum.
It should be emphasized that this argument is not as nice as the one used for heterotic
strings. There finiteness was a consequence of a sound principle, modular invariance,
whereas here it has to be imposed “by hand”.
We will not discuss the SO(32) open string theory in more detail here, except to
remark that the massless spectrum is identical to that of the heterotic SO(32) string. In
particular both have N = 1 supersymmetry. However, the massive spectra are different.
For example, the heterotic string has massive states in the SO(32) spinor representation,
whereas the open string theory only has symmetric and anti-symmetric tensors in its
spectrum.
where all momenta are assumed to be incoming (in other words, the final state particles
have outgoing momenta −p3 and −p4 ). If the exchanged particle has spin 0, this particular
diagram has an amplitude of the form
1
A = −g 2 (10.3)
t − M2
134
where g is a coupling constant and M the mass of the exchanged particle.
String theory was born by considering the scattering of hadrons by exchanging mesons.
If one considers the exchange of several mesons of different spins J, the amplitude gener-
alizes to
X sJ
At (s, t) = − gJ2 2
(10.4)
J
t − M J
It was observed empirically that in hadronic scattering As (s, t) ≈ At (s, t). This led
Veneziano in to a formula that explicitly had this “s-t duality” built in:
Γ(−α(s))Γ(−α(t))
A(s, t) = (10.6)
Γ(−α(s) − α(t))
with
α(s) = α(0) + α0 s (10.7)
The Γ function satisfies
Γ(u + 1) = uΓ(u) , Γ(1) = 1 (10.8)
so that Γ(u) = (u + 1)! for positive integer arguments. For negative arguments there are
singularities at u = 0, −1, −2, . . ., and near u = −n the behavior is
(−1)n
Γ(u) ≈ (10.9)
n!(u + n)
We see then that the Veneziano amplitude has poles if α(s) = n or α(t) = n, or
1 1
s= (n − α(0)) , t = (n − α(0)) (10.10)
α0 α0
If one interprets these poles as particles exchanged in the t-channel (t = M 2 ) or the
s-channel (s = M 2 ), then we recognize here precisely the open string spectrum (3.62)
135
(apart from multiplicities and spin-dependence, which with a more careful analysis also
come out correctly). The parameter α(0) is called the intercept and plays the role of the
vacuum energy subtraction a. The parameter α0 is called the Regge slope (this explains
finally why the fundamental length scale of string theory was called α0 , and in particular
why there is a prime).
The foregoing remarks suggest that the Veneziano amplitude has a string theory in-
terpretation, and indeed it does. Pictorially the observation in the previous paragraph
is
Although it is beyond the scope of these lectures, the four-point string diagram is not
hard to compute, and yields indeed the Veneziano amplitude. In special limits (namely
near the poles in s or near the poles in t) this single amplitude behaves at the same time
as sum of an infinite number of f -channel exchanges, or as the sum of an infinite number
of s-channel annihilations. It is historically rather strange that first the amplitude was
written down, and only afterwards its origin was understood, but that is indeed what
happened.
Note that by deforming the open string tree diagram one can get the s and t diagrams
but not the u diagram
136
since we cannot interchange external lines. However, if we work with closed instead of
open strings we can interchange all lines, since they are simple four tubes attached to
a sphere. The tubes can be moved around each other on the sphere. Consequently the
four-point amplitude for closed strings has a symmetry in s, t and u, and has the form
Γ(− 14 α(s))Γ(− 41 α(t))Γ(− 14 α(u))
A(s, t, u) = (10.11)
Γ(− 41 α(s) − 14 α(t) − 14 α(u))
The factors 14 reflect the factor of 4 difference in spacing between the open and closed
string spectra.
137
of activity. Calabi-Yau compactifications were studied, and many other compactifications
and four-dimensional string constructions were investigated.
After another period of relatively little activity (although much more than between
1975-1984) in 1994 the “third string revolution” occurred. This was sparked by the
discovery of a set of string dualities, although a rather different kind of dualities than
those discussed above.
10.5 T-duality
In the section on bosonic string torus compactification we saw that the compactified
string has a remarkable duality under R → α0 /R. Similar dualities can be studied for the
compactified type-II and heterotic strings. If we compactify on a circle to 9 dimensions,
it turns out that the two compactified type-II strings are mapped into each other, and
that the two heterotic strings are also mapped to each other. We will not study this in
detail, but only give a rough idea how this works.
The two type-II strings only differ in the relative chirality of the left and right GSO
partition functions. However, chirality is lost if one make a torus compactification, as we
have seen in the previous chapter. Therefore it is not a big surprise that the difference
between IIA and IIB is lost upon compactification.
For the Heterotic compactification we can be more explicit. A general compactification
of one dimension can be described by a Γ1,1 Lorentzian even self-dual lattice. If we combine
this with the 16-dimensional lattice in the left sector (using the bosonic description of
that sector) we get a Lorentzian even self-dual lattice Γ17,1 . A general theorem on such
lattices says that they are all related by Lorentz transformations. It can be shown that
in string theory these Lorentz-transformations are parametrized by the moduli of the
compactification torus, which are massless scalar fields. Giving these scalars a vacuum
expectation value we move from one Γ17,1 lattice to another, and in particular we can
move from E8 × E8 × Γ1,1 to E16 × Γ1,1 . This shows that these two theories are “on the
same moduli space”, and hence they are clearly related. That the relation is actually a
T-duality requires a bit more discussion.
The second term has the property that all its derivatives at g = 0 vanish. Hence its Taylor
expansion around g = 0 is trivial. One can never see such a term in perturbation theory.
138
Therefore, no matter how far we expand, we will never be able to compute the coef-
ficient A from a loop diagram. Effects that behave like this are called non-perturbative
effects. It is known that such contributions occur, for example, in QCD.
In QCD the appearance of such effects is related to the existence of classical solutions
to the equations of motion (without sources). The first terms in (10.12) are coming from
an expansion of the action around the trivial classical solution (with all fields equal to
zero), whereas the second term comes from an expansion around the non-trivial solution
(there are additional perturbative corrections to the last term). In string theory it is a
priori much more difficult to find such classical solutions. This is because in string theory
we only have a perturbative expansion.
To understand this note that in QCD the Feynman diagram are derived from an action,
and that action can also be used to derive classical equations of motion that can be solve.
In string theory we only have the “Feynman diagrams”, but no action to derive them
from (we would need an action in target space, not on the world-sheet).
One way to get around this is to derive from string theory a low-energy effective action.
This is an action for the massless fields only, such that the interactions derived from it
are the same as those derived directly from string theory. This action includes of course
the Einstein action for gravity, as well as the usual actions of the gauge bosons. Then one
can solve the equations of motion of this low-energy effective action, and postulate that
it is the first order approximation to a genuine string theory classical solution.
Classical solutions to field theory equations of motion can be classified according to
their extension in space. A solution can be point-like, meaning that its energy density is
concentrated around a point in space; analogously it can be string-like or membrane-like.
In general we call a solution a p-brane if it has p spatial dimensions. Hence a particle-like
solution is a 0-brane, a string-like solution a 1-brane, etc.
10.7 D-branes
There exists an exception to the statement that we cannot describe classical solutions
exactly in string theory. These are the so-called D-branes (where “D” stands for Dirichlet).
As we have seen, open strings can have two kinds of boundary condition. A Neumann
boundary condition has the property that ∂σ X vanishes at the boundary. This amounts to
conservation of momentum at the boundary. For a Dirichlet boundary condition ∂τ X = 0
at the boundary. This means that X(0) and X(π) are fixed as a function of time.
One may consider situations where – for the bosonic string – D space-like coordinates
have Dirichlet boundary conditions, and the remaining 26 − D have Neumann boundary
conditions. Then we have
where v i and wi are some fixed vectors. The remaining space-like directions are not
fixed. Below we illustrate this for a total of 3 dimensions instead of 26. We plot the two
space-like dimensions x1 and x2
139
D1-branes D0-branes
In the first picture we have Dirichlet boundary conditions in the x1 direction. Hence the
open string has its endpoints fixed to two lines at fixed x1 . The endpoints are allowed to
move in the x2 direction. In the second picture we have Dirichlet boundary conditions in
both directions. Then the endpoints are completely fixed to points in space. Finally we
may take Neumann conditions in all directions. Then the endpoints are free to move in
all space directions. This is what we normally mean by “open strings”.
In the first two cases translation invariance is broken in one or two directions. This
can be understood in terms of objects that are present in space. A particle fixed at
some position obviously breaks translation invariance in all directions. A wall breaks
translation invariance in the direction orthogonal to it. Therefore we interpret the first
picture in terms of two lines, or 1-branes, whose presence breaks translation invariance,
and the second picture is interpreted in terms of two points or 0-branes. The space-time
momentum of modes on the string is not conserved because it can be transferred to these
objects. Because their presence is due to Dirichlet boundary conditions these objects are
called D-branes or Dp -branes: they are p-branes defined by the property that open strings
can end on them.
This only makes sense if we view the D-branes as dynamical objects, that can them-
selves move around and vibrate. This point of view is indeed correct. It turns out that
these D-branes are the exact string analog of certain approximate field theory solutions.
There may be additional solutions, but those are harder to describe.
Note that the D-branes are objects that are not obtained in string perturbation theory.
There are precise rules for which kind of D-branes can occur in which string theories.
There are no D-branes in heterotic strings, but they do exist in type-II strings. This may
look surprising at first, since type-II strings were constructed as closed string theories.
However, that was a perturbative description. It turns out that these theories nevertheless
contain non-perturbative states that are described by open strings.
The rules are that type-IIA theories can have Deven -branes whereas type-IIB theories
can have Dodd -branes. Of special interest are the D0 -branes in type-IIA and the D9
branes in type-IIB. Let us start with the latter. A D9 brane fills all of space, and the
open strings ending on it have endpoints that can move freely through space. Such strings
are traditional open strings with only Neumann conditions, and no Dirichlet conditions.
The type-IIB string either has no D9 branes at all, or it must have exactly 32. In the
latter case one is forced to allow also different world-sheet orientations, and one is led to
140
the type-I string. One can assign a label to each of the 32 branes. This corresponds to
the Chan-Paton labels. From this point of view the type-I string lives in a space built out
of 32 D9 branes stacked on top of each other.
10.8 S-duality
The D0 -branes of type-IIA correspond to extra, non-perturbative particle-like states in
the spectrum. These states turn out to carry a charge with respect to the massless vector
boson Aµ in the perturbative spectrum. They are the only states that are charged, because
it is easy to show that all perturbative states are neutral.
One may also compute the masses of the D0 particles, and then one finds that they
are ∝ |q|
g
, where q is the charge and g the string coupling constant.
Witten made the observation that these particles could be interpreted as Kaluza-Klein
modes of a circle compactification of 11-dimensional supergravity.
The latter theory has two bosonic fields: a graviton, and a rank-3 anti-symmetric
tensor. Upon compactification (as we have seen in section (3.6)) the graviton yields a
10-dimensional graviton, a photon and a scalar. The rank-3 tensor yields a rank-3 tensor
and a rank-2 anti-symmetric tensor. These are precisely the bosonic fields of the ten-
dimensional type-II string at the massless level.
But Kaluza-Klein compactification yields more than that. If we expand any field
in modes, we get also massive excitations. These excitations turn out to carry charge
with respect to the photon, and turn out to match precisely the D0 -particle spectrum.
Furthermore the masses of these excitations are inversely proportional to the radius of
the circle: the smaller the circle, the more energy it costs to excite a mode in the compact
direction.
This suggests to identify the radius of the circle with the strength of the type-IIA
coupling constant. At small coupling constant the masses of the Kaluza-Klein modes or
D0 -particles become large and can be neglected; this correspond to the D = 11 theory
compactified on a very small circle. In the limit of large coupling the D0 particles become
massless, and the radius goes to infinity. In this limit we approach 11 flat dimensions.
The conclusion – which is supported by other arguments – is that the strong coupling
limit of the type-IIA theory is an 11-dimensional theory. This is definitely not a string
theory. It is also definitely not 11-dimensional supergravity: that theory has infinities in
its perturbation theory, which should be absent if it is the strong coupling limit of a finite
string theory. The speculation is now that it is an unknown theory, called “M-theory”
whose low-energy limit is D = 11 supergravity.
One can also consider the strong coupling limit of all other string theories and one
finds
Heterotic SO(32) ↔ Type-I SO(32)
Type-IIB ↔ Type-IIB (10.14)
Heterotic E8 × E8 ↔ 11-dimensional supergravity on S1 /Z2
The first duality requires the existence of SO(32) spinors in the spectrum of the type-I
141
string. They can only because of non-perturbative effects. The second case is, for obvious
reasons, called self-duality. The strong coupling limit of the E8 × E8 heterotic string is
D = 11 supergravity compactified on a circle with mirror image points identified to each
other, as shown here:
The resulting space is a line segment, so that the 11-dimensional space becomes two planes
separated by the length of the segment. The two E8 factors can be associated with the
two planes.
There is a lot more to say about dualities in 10 and 11 dimensions, and even more in
less than 10 dimensions. But we will stop here and conclude with the duality picture that
summarizes the present understanding.
S-d
li ty ua
lity
ua
S-d
Type-IIA Heterotic (E8 × E8)
T-duality
T-duality
Type-I (SO(32))
142
A Functional Methods
In classical field theory we deal with dynamical systems with an infinite number of “coor-
dinates” qi . For example, suppose we want to describe the classical motion of gauge fields
Aµ (~x, t) in electrodynamics. Then the coordinates that enter into the familiar Lagrangian
or Hamiltonian description are Aµ (~x).
An additional complication is that these coordinates are continuously infinite: the
label ‘i’ of qi gets replaced by the set (µ, ~x). This implies that summations become
integrations, and differentiations are replaced by functional derivatives.
Let us first review the kind of formulas we want to generalize to classical field theory.
A system with a finite number of degrees of freedom is usually described by a Lagrangian
L(qi , q̇i ). To every path with initial time t0 and final time t1 one associates an action
Z t1
S= dtL(q(t), q̇(t)) (A.1)
t0
Hamilton’s principle states that the classical trajectory followed by the system is given by
an extremum of the action: δS[qi (t)] = 0. From this one derives the equations of motion,
the Euler-Lagrange equations,
d ∂L ∂L
− =0. (A.2)
dt ∂ q̇i ∂qi
To quantize such a theory one defines canonical momenta
∂L
pi = , (A.3)
∂ q̇i
and a Hamiltonian X
H= pi q̇i − L . (A.4)
i
For two classical quantities a and B one may define a Poisson bracket
X ∂A ∂B ∂A ∂B
{A, B}PB = − (A.5)
`
∂q ` ∂p` ∂p` ∂q`
To go to the quantum theory one replace A and B by by Hilbert space operators, and
furthermore one replaces the Poisson brackets {A, B}PB by the commutator i~1 [A, B]. For
A = qi and B = pj this procedure leads to the well-known Heisenberg relation
143
R
from which one can compute the action as S = dtL. To write down the Euler-Lagrange
equations one needs a functional derivative. A functional is an object that assigns a
number to a function. To only example we will ever encounter is an integral. A functional
derivative can be viewed as the derivative of the functional with respect to the function.
For example, consider a function φ(~x) and the functional
Z
F [φ] = d3 xf (φ(~x)) (A.8)
δφ(~x)
= δ(~x − ~y ) , (A.11)
δφ(~y )
= f 0 (φ(~y )) .
This is not the most convenient form, since in Lorentz covariant field theories time deriva-
tives appear in combination with space derivatives, as ∂µ . The symmetry between space
and time is restored by defining a Lagrangian density, which is simply the integrand of
the Lagrangian Z
L = d3 xL (A.14)
144
In terms of L the Euler-Lagrange equations are
∂L ∂L
∂µ − =0, (A.15)
∂(∂µ φ) ∂φ
These are ordinary derivatives, because we see L as a function (not a functional) of φ and
∂µ φ. This version of the Euler-Lagrangian
R can either be derived from (A.13), or directly
from the variation of S = d4 xL.
Canonical momenta are defined as
δL
π a (~x) = (A.16)
δ φ̇a (~x)
A bit more care is needed when there is a distinction between upper and lower indices. If
gab is a symmetric
P matrix and the action depends on combinations φa gab φb it is customary
b ab
to
P definebc φa =c b gab φ . Indices can be raised by using
R g3 , which is the inverse of gab :
c
b gab g = δa . Suppose we have a functional F = d xVc φ (~x), with Vc a constant
vector. Then
δF
= Vc , (A.17)
δφc (~x)
using
δφa (~x)
= δba δ(~x − ~y ) (A.18)
δφb (~x)
One see this more explicitly by leaving all indices upper, and writing the matrix gab
explicitly. Hence the derivative with respect to a quantity with an upper (lower) index
yields a quantity with a lower (upper) index.
This applies also to canonical momenta and Poisson brackets. The canonical momen-
tum of φa is then (if there is a distinction between upper and lower indices)
δL
πa = (A.19)
δ φ̇a (~x)
The Hamiltonian is Z X
H= d3 x πa (~x)φ̇a (~x) − L (A.20)
a
Often one converts the first index to an upper index, and then the result is of course
a
π (~x), φb (~y ) = i~g ab δ(~x − ~y )
(A.22)
145
A.1 The Euclidean Action
The Euclidean action is obtained by a continuation from Minkowski space, by making
time imaginary. This could mean either t = itE or or t = −itE . The choice between
these two options is determined by the requirement that the analytic continuation should
make the path integral a damped exponential.R Consider the weight factor in the path
integral, exp(iS). In this expression S = dtL, and we consider a simple Lagrangian of
the form 21 q̇ 2 − V (q), so that the Hamiltonian is H = 12 p2 + V (q). Once we know the
correct continuation for this action, we know it for all action.
If the potential V is bounded from below, as it should in a physical system, then
clearly the correct definition for the potential term alone is t = −itE . If the Minkowski
time integral is from 0 to T we get now
Z T 2 ! Z 2 !
TE
1 dq 1 dq
iS = i dt − V (q)) = dtE − − V (q) (A.23)
0 2 dt 0 2 dtE
with T = −iTE . We see that both the kinetic and the potential term give a negative
exponential. It is conventional to choose the Euclidean action in such a way that the
path-integral is dominated by its minima (and not its maxima). Hence one defines
Note that this definition contains an extra minus sign compared to the naively expected
relation, and that in particular the V (q) contribution appears with opposite sign in S and
SE : Z TE
SE = − dtE LM (t = −itE ) , (A.26)
0
where LM is the Lagrangian in Minkowski space.
B Path Integrals
The derivation of (6.24) goes as follows (this discussion is based on appendix A of [2])).
We consider a quantum mechanical system with one coordinate q and a momentum p.
The corresponding quantum operators q̂ and p̂ satisfy [q̂, p̂] = i. Furthermore we define
eigenstates |qi (with q̂ |qi = q |qi) and |pi (with p̂ |pi = p |pi), and we have
hq|pi = eipq
hq|q 0 i = δ(q − q 0 ) (B.1)
hp|p0 i = 2πδ(p − p0 )
146
The integration measures for position and momentum integral are respectively
Z Z
dp
dq |qi hq| = 1; |pi hp| = 1. (B.2)
2π
We divide the time-interval T into N equal steps , T = N , and define tm = m. At
t = 0 a complete set of states is |qi. We define a set of states |q, ti as
This introduces a complete set of states at every time tm . In general we have the com-
pleteness relation Z
hA|Bi = dqhA|q, tihq, t|Bi . (B.4)
The classical Hamiltonian is a function of p and q, H(p, q). The quantum Hamiltonian is
ambiguous because of the ordering of p̂ and q̂. We will define it by requiring that all p̂
are always to the left of q̂. Then
Even if p̂ and q̂ are ordered correctly in H, the exponential exp(−H) still contains
improperly ordered terms. However, they are always of order 2 and will be dropped.
Then we get
Z
dpm −iH(pm ,qm )
hqm+1 , tm+1 |qm , tm i = e hqm+1 |pm ihpm |qm i
2π
Z (B.9)
dpm −i[H(pm ,qm )−pm (qm+1 −qm )]
= e ,
2π
147
We will only be interested in Hamiltonians of the form H = 12 ap2 + V (q), where a is
some constant. In that case the pm integrations in (B.9) can be done explicitly. At this
point it is better to go to Euclidean space, by defining = −iδ. The sign choice is dictated
by the requirement that we get a damped gaussian integral. The relevant integral is
Z Z
dp 2 dp
1
exp(− 2 aδp + ipd) = exp(− 12 [aδ(p − id/aδ)2 + d2 /aδ])
2π 2π
(B.10)
1 1 2
=√ exp(− 2 d /aδ) ,
2πaδ
with d = qm+1 − qm . Now we get
−1 −1
Z NY N N
!
1 X
(qm+1 − qm )2 /aδ − V (qm )δ
1
hqf , T |qi , 0i = dqi √ exp − 2
i=1
2πaδ m=0
−1
Z NY N N −1 ! (B.11)
1 X q m+1 − q m 2
= dqi √ exp − δ 21 ( ) /a − V (qm )
i=1
2πaδ m=0
δ
Here we have absorbed the pre-factor into the definition of Dq. Note that the normal-
ization is anyhow fixed by the limit β → 0, which should yield the normalization of the
state. The result can also be written as
Z
−Hβ
hqf , 0|e |qi , 0i = Dq exp(−SE ) (B.13)
The path integral is over all paths q(t) with q(0) = qi and q(β) = qf . Now we take
qi = qf ≡ q and integrate over all qi . Then we get
Z Z
−βH −Hβ
Tre ≡ dqhq, 0|e |q, 0i = Dq exp(−SE ) (B.14)
P BC
The extra integral has extended the path-integral with an integral over the initial position.
We are now integrating over all paths satisfying q(0) = q(β) = q, with an integration over
q. Hence this is a path integral over all paths satisfying a periodic boundary condition.
C Self-dual Lattices
Here we summarize a few results on even self-dual lattices.
148
Lorentzian even self-dual lattices Γp,q are defined by the requirement that all vectors
have even length, and the Γp,q is self-dual. Both length are duality are defined with an
inner product of the form
p q
X X
2 2
(~v , w)
~ = vi − wj2 , (C.1)
i=1 j=1
where (~v , w)
~ is a vector on the lattice.
Lorentzian even self-dual lattices (ESDL’s) are classified modulo Lorentz transforma-
tions. Up to such transformations there is just one lattice, i.e. all ESDL’s are related to
each other by Lorentz transformations. Physically this information is useless. Particle
spectra depend on ~v 2 and w ~ 2 separately, not just on their difference. Only rotations in
the first p or the last q components make no difference. In other words, the condition for
modular invariance is Lorentz invariant, but the spectrum is not.
Since all Lorentzian ESDL’s are Lorentz rotations of each other, the parameters of the
Lorentz transformation are the moduli of the compactification. The full parameter space
(“moduli-space”) is in fact the coset space
SO(p, q)
(C.2)
SO(p) × SO(q)
This means that we consider the Lorentz-transformations in SO(p, q), but we identify
those Lorentz-transformations that are related to each other by rotations in the first p
or the last q components. The precise moduli space is still a bit smaller: because of T-
duality and its generalizations on higher-dimensional tori some parts of the moduli space
are identified with each other.
The foregoing holds if p 6= 0, q 6= 0. If q = 0, ESDL’s exist only if p is a multiple of 8.
The first case is p = 8, and it turns out that the lattice is then unique and given by
This is an overcomplete basis, but in any case the idea is that one builds a lattice by
adding all these vectors in any integer linear combination. The vectors shown here have
length 2, and in total there are 240 of them: 112 of the first type, and 128 of the second
type (note that by adding the first set one can get all even sign changes of the second
vector). This lattice is called E8 (it is in fact the root lattice of the Lie-algebra E8 ).
By taking this lattice for left- and right-movers we can build a lorentzian ESDL Γ8,8 =
(E8 )L × (E8 )R . The vectors of this lattice are of the form (~vL , ~vR ) with ~vL and ~vR vectors
of the above form. On this lattice we may compactify the bosonic string to 16 dimensions.
i I i
In the resulting spectrum the states of the form ᾱ−1 |vL , 0i and α−1 ᾱ−1 |0, 0i (with vL2 = 2
and I = 1, . . . 8) are massless vector bosons. There are 240 + 8 = 248 such bosons from
the left-moving and the same number from the right-moving modes. It turns out that
massless vector bosons in string theory are always gauge bosons. In this case they are
gauge bosons with a Lie-algebra (E8 )L × (E8 )R . Here E8 is a Lie-algebra specified, just
149
as SU (2), by a set of commutation relations. We will not explain its construction here.
It has of course precisely 248 generators.
If p = 16 there are two solutions: (E8 )2 and a lattice one might call E16 . It is defined as
E8 above, but with 16-component vectors. Because of this the second vector, ( 12 , . . . , 12 )
now has length 4 and would not produce a massless gauge boson when we compactify
on Γ16,16 = (E16 )L × (E16 )R . Indeed, now there are 496 massless gauge bosons from
the left and the same number from the right. They all originate from the first type of
I
lattice vector in (C.3), plus the 16 oscillators α−1 . The number of length-2 vectors is
1
2
× (16 × 15) × 4 = 480. The gauge group is in fact SO(32)L × SO(32)R . The name E16
is not standard. Usually one calls the lattice without the half-integer vectors D16 : it is
the root lattice of SO(32). The half-integer spin vector is an SO(32) spinor weight. So
the correct description of the “E16 ” lattice would be the “D16 root lattice with a spinor
conjugacy class added”.
For profound reasons 496 = 2×248, so that both p = 16 lattices give the same number
of gauge bosons, but different gauge groups.
If p = 24 there are 24 ESDL’s, and for p = 32 the number explodes to an unknown
number larger than 107 (and undoubtedly much larger)!
[T a , T b ] = if abc Tc , (D.1)
with structure constants fabc that are real and completely anti-symmetric.
D.2 Exponentiation
Locally, near the identity, the corresponding Lie group can be obtained by exponentiation
aT a
g(αa ) = eiα . (D.2)
The global properties of the group, involving element not “close” to 1, are not fully
described by the Lie-algebra alone, but will not be discussed here. The space formed by
all the group elements is called the group manifold.
150
D.3 Real Forms
A Lie-algebra is a vector space of dimension dim (A) with an additional P operation, the
a
commutator. An arbitrary element of the vector space has the form a αa T . In ap-
plications to physics αa is either a real or a complex number. If the coefficients αa are
all real and the generators Hermitean, the group manifold is a compact space. For a
given compact group there is a unique complex Lie-algebra, which is obtained simply by
allowing all coefficients αa to be complex. Within the complex algebra there are several
real sub-algebras, called real forms. The generators of such a sub-algebra can be chosen
so that (D.1) is satisfied with all structure constants real, but with generators that are
not necessarily Hermitean. One can always obtain the real forms from the compact real
form (which has hermitean generators) by choosing a basis so that the generators split
into two sets, H and K, so that [H, H] ∈ H and [K, K] ∈ H. Then one may consistently
replace all generators K ∈ K by iK without affecting the reality of the coefficients f abc .
The most common case in physics are the real forms SO(n, m) of the compact real form
SO(n + m). Most of the following results hold for the compact real form of the algebra,
unless an explicit statement about non-compact forms is made.
151
D.7 Roots
If we denote the Cartan sub-algebra generators as Hi , i = 1, . . . , r, then the remaining
generators can be chosen so that
The eigenvalues with respect to the Cartan sub-algebra are vectors in a space of dimension
r. We label the remaining generators by their eigenvalues α
~ . These eigenvalues are called
the roots of the algebra.
where α~ i is a simple root. This matrix is unique for a given algebra, up to permutations
of the simple roots. One of the non-trivial results of Cartan’s classification of the simple
Lie algebras is that all elements of A are integers. The diagonal elements are all equal to
2 by construction; the off-diagonal ones are equal to 0, −1, −2 or −3.
152
Ar
1 2 3 4 r-1 r
Br
1 2 3 4 r-1 r
Cr
1 2 3 4 r-1 r
Dr
1 2 3 4 r-2
r-1
G2
1 2
F4
1 2 3 4
6
E6
1 2 3 4 5
E7
1 2 3 4 5 6
E8
1 2 3 4 5 6 7
153
D.13 Realizations
The compact Lie-algebras corresponding to types A − D are realized by the algebras
SU (n), SO(n) and Sp(n). The correspondence is as follows
Ar :SU (r + 1)
Br :SO(2r + 1)
(D.6)
Cr :Sp(2r)
Dr :SO(2r)
There is no such simple characterization for the algebras of types E, F and G, the excep-
tional algebras.
D.14 Representations
A set of unitary N × N matrices satisfying the algebra (D.1) is said to form a (unitary
matrix) representation of dimension N .
D.15 Equivalence
If a set of hermitean generators T a satisfy the algebra, then so do T̃ a = U † T a U , if U is
unitary. Then T a and T̃ a are called equivalent.
154
D.18 Weights
In any representation the matrices representing the Cartan sub-algebra generators Hi can
be diagonalized simultaneously. The space on which the representation acts decomposes
in this way into eigenspaces with a set of eigenvalues ~λ, i.e.Hi v~λ = λi v~λ . The ~λ’s, which
are vectors in the vector space spanned by the roots, are called weights. The vector space
is usually called weight space.
D.21 Coroots
2~
α
Coroots α̂ are defined as α̂ = α·α
.
155
D.24 The Irreducible Representations of a Simple Lie-algebra
can thus be enumerated by writing a non-negative integer next to each node of the Dynkin
diagram. The states in a representation can all be constructed by acting with the gener-
ators Eα on the highest weight state. This state always has multiplicity 1.
• Vector representations
The N × N matrices that were used above to define the classical Lie groups form
the vector representation of those groups; the expansion of these matrices around
the identity yields the vector representation of the corresponding Lie-algebra.
Acknowledgments
I would like to thank all the students who have contributed to these notes by informing
me about typos and errors. Special thanks to Stan Jakobs and Melissa van Beekveld
for their corrections during the 2014 course and Dennis Obster for corrections during the
2016 course.
156
References
[1] M. Green, J. Schwarz and E. Witten, Superstring Theory, parts 1 and 2, Cambridge
Univ. Press, (1986).
[2] J. Polchinski String Theory, parts 1 and 2, Cambridge Univ. Press, (1998).
[4] B. Zwiebach, A first course in String Theory. Cambridge University Press, Cam-
bridge. 2005.
[5] W. Lerche, A. Schellekens and N. Warner, Lattices and Strings, Physics Reports 177
(1989) 1.
[6] B. de Wit and J. Smith, Field Theory and Particle Physics,(North Holland).
[7] A.N Schellekens, Life at the Interface of Particle Physics and String Theory, Rev.
Mod. Phys. 85 (2013) 1491.
157