Arutyunov Notes

Preprint typeset in JHEP style - HYPER VERSION
Lectures on String Theory
Gleb Arutyunova
a Institute for Theoretical Physics and Spinoza Institute, Utrecht University

3508 TD Utrecht, The Netherlands
Abstract: The course covers the basic concepts of modern string theory. This
includes covariant and light-cone quantisation of bosonic and fermionic strings,
geometry and topology of string world-sheets, vertex operators and string scattering
amplitudes, world-sheet and space-time supersymmetries, elements of conformal
field theory, Green-Schwarz superstrings, strings in curved backgrounds, low-energy
effective actions, D-brane physics.
Contents
1. Introduction to the course 2

1.1 Historical remarks 2
2. Relativistic particle 8
3. Classical relativistic bosonic string 13

3.1 Nambu-Goto string 13
3.2 The Polyakov action 16
3.2.1 Symmetries 16
3.2.2 Equations of motion 16
3.2.3 Conformal gauge 18
3.2.4 Polyakov string in the first order formalism. 19
3.3 Integrals of motion. Classical Virasoro algebra. 21
3.3.1 Solutions of the equations of motion 24
3.3.2 Poincaré symmetry. String tension. 25
3.4 Strings in physical gauge 28
3.4.1 First order formalism 28
3.4.2 Poisson structure of the light-cone theory 35
3.4.3 Lorentz symmetry 37
4. Quantization of bosonic string 38

4.1 Remarks on canonical quantization 38
4.1.1 Virasoro algebra 41
4.1.2 Virasoro constraints in quantum theory 43
4.1.3 The spectrum 45
4.1.4 Propagators 47
4.1.5 Vertex operators. Tachyon scattering amplitude 48
4.2 Quantization in the physical gauge 58
4.2.1 Lorentz symmetry and critical dimension 59
4.2.2 The spectrum 63
4.3 BRST quantization 66
5. Geometry and topology of string world-sheet 77

5.1 From Lorentzian to Euclidean world-sheets 77
5.2 Riemann surfaces 79
5.3 Moduli space 84
–1–
6. Classical fermionic superstring 92
6.1 Spinors in General Relativity 92
6.2 Superstring action and its symmetrices 99
6.3 Superconformal gauge and supermoduli 100
6.4 Action in the superconformal gauge 102
6.5 Boundary conditions 106
6.6 Superconformal algebra 107
7. Quantum fermionic string 108

7.1 Light-cone quantization and superstring spectrum 111
A. Dynamical systems of classical mechanics 119
B. OPE and conformal blocks 123
C. Useful formulae 127
D. Riemann normal coordinates 128
E. Exercises 130
1. Introduction to the course

1.1 Historical remarks
String theory arose at the end of sixties in an attempt to describe the theory of
strong interactions. In 1969 Veneziano found a beautiful formula for the scattering
amplitude of four particles. This amplitude comprised many features that physicists
expected to be found in the theory of strong interactions. It was realized very soon by
Nambu and Susskind that the underlying dynamical object from which the Veneziano
formula can be derived is a relativistic string. The fundamental difference of strings
from the theory of point particles is that strings are extended, one-dimensional,
objects; when string moves through space and time it sweeps two-dimensional surface
called the string “world-sheet”. The strings can be of two types: open – with topology
of an interval and closed – with topology of a circle.
Subsequent investigations revealed however severe difficulties to treat the string
as the theory of strong interactions. These difficulties are
1. Existence of a “critical dimension”.
2. Existence of a massless spin two particle which is absent in the hadronic world.
–2–
If one tries to construct the quantum mechanics of relativistic strings one finds
that mathematically consistent theory exists if and only if the dimension of space-
time where string propagates is 26. The number 26 was named the “critical dimen-
sion”. On the one hand, it was pretty remarkable and unexpected to find an example
of physical theory which puts restrictions on space-time where it is defined. On the
other hand, it was certainly not clear why a theory that shared at least some qualita-
tive features with hadronic physics should exist in 26 dimensions only. A subsequent
discovery of QCD (Quantum Chromo Dynamics) as the most appropriate candidate
to describe the theory of strong interactions led to a considerable loss of interest to
string theory.
In 1974, Scherk and Schwarz came up with a proposal to completely alter the
view on string theory. They suggested to consider the massless spin two particle
absent in the hadronic world as the graviton – the quantum of the gravitational
interaction. Indeed, this particle neatly fits the properties of the graviton – string
theory predicts that this particle interacts according to the standard laws of General
Relativity. Gravitational interactions have a natural scale, called the Planck mass,
which is around 1019 GeV. This is a huge number in comparison with characteristic
energies of hadronic physics, 100 − 200 MeV. Thus, according to their view, string
theory could provide the unifying description of all the particles and matter forces,
including gravity and it operates on a new fundamental scale.
Even if one accepts that quantum mechanics of relativistic strings can be defined
in the unusual number 26 of the space-time dimensions, another problem arises. Such
string does not contain fermionic degrees of freedom and it predicts the existence of
a particle with the negative mass squared: m2 < 0. Such a particle, tachyon, is a
source of instability and its existence indicates that either the theory is ill-defined
or it is formulated around a “wrong” ground state, or as physicists say, around a
“wrong” vacuum. Critical dimension, tachyon and absence of fermions were the
puzzling features the string theory had to face.
The status of string theory changed again with the discovery of supersymmetry.
All universe is made of two fundamental types of particles: bosons and fermions.
Fermions constitute all the matter and bosons mediate interactions of the matter
particles. Supersymmetry is a new type of symmetry between bosons and fermions
(Wess and Zumino 1974). Many physicists hope today that supersymmetry could
provide an underlying principle for unification of all interactions.
The first success in incorporating supersymmetry in string theory was achieved
in 1971 by Ramond, who constructed a string analogue of the Dirac equation (the
spinning string). Shortly afterwards, Neveu and Schwarz constructed a new bosonic
string theory. They realized that the two constructions were different facets of a
single theory - an interacting superstring theory containing Neveu and Schwarz’s
–3–
bosons and Ramond’s fermions. The supersymmetry of the two-dimensional string
world sheet was recognized by Gervais and Sakita in 1971. This was advent of the
NSR (Neveu-Ramond-Schwarz) superstring.
In 1972 Schwarz demonstrated the consistency of the superstring theory in 10 di-

mensions. Instead of 26 found for purely bosonic string, the critical dimension for the
NSR string appears to be 10. In 1977 Gliozzi, Scherk and Olive realized that further
conditions should be imposed on the spectrum (the GSO projection mechanism) of
the NSR string which lead to both the so-called space-time supersymmetry (to com-
pare with the world-sheet supersymmetry mentioned above) and to the removal of
tachyon. Thus, superstring theory has at least two advantages in comparison with
bosonic strings: critical dimension 10 < 26 and the absence of tachyon. It also turned
out that the GSO projection can be imposed in two different ways which lead to two
different types of superstrings, called the Type IIA and Type IIB.
String theories have a natural particle limit, when the length of string vanishes.
In this limit superstrings give rise to the low-energy effective theories, known as
supergravities. These theories can be defined in a way completely independent of
string theory: they can be thought of as supersymmetric generalizations of the pure
Einstein gravity. As is known, attempts to quantize gravity in the standard frame-
work of quantum mechanics fail because gravity is a non-renormalizable theory (there
are infinitely many divergent graphs with any number of external legs and with an
arbitrarily high index of divergence, cf. the course on Quantum Field Theory). Su-
persymmetric theories tend to be less divergent than non-supersymmetric ones which
gave initially a hope that supersymmetry could cure the nonrenormalizable infinities
of the quantum gravity. It seems that supergravities themselves are still not capable
to solve the divergency problem1 . Quite remarkably, there is a strong evidence that
the divergency problem of quantum (super)gravities is resolved by string theory.

Resolution of the four-fermi interaction. At high energies the weak force is

mediated by a heavy boson.
1
For instance, it is unknown if the so-called N = 8 supergravity is finite or not.
–4–
To get a better feeling why it happens it is convenient to envoke an analogy
with the theory of weak interactions. Trying to describe the decay of the neutron by
the Fermi-type Lagrangian containing the quartic, point-like, interaction vertex one
finds irremovable ultraviolet divergencies at higher loop orders. Again, the reason
is that the corresponding theory of Fermi-interactions in non-renormalizable. The
solution of this problem lies in a fact that at higher energies (more than 100 GeV),
the pointlike vertex gets resolved and the interaction is mediated by a heavy W-
boson. In the new theory one has qubic vertices and this ultimately makes the
theory renormalizable.
Very similar phenomenon occurs in string theory. Expanding the Einstein-

√
Hilbert action −gR one gets more and more complicated point-like vertices which
render the theory non-renormalizable, analogously to the old Fermi theory. In oppo-
site, in string theory all these vertices get dissolved by the exchange of the massive
string states. String states form an infinite tower of particles of arbitrarily high mass
and spin and all of them participate in the interaction process. Resolving the ultra-
violet divergency problem string theory appears to be a natural candidate for the
theory of quantum gravity.
There is a still an important question how to relate the superstring theory defined
in a ten-dimensional space-time with a our conventional four-dimensional physics.
The basic approach to obtain four-dimensional theories is along the old ideas of
Kaluza and Klein and it consists in compactifying of the ten-dimensional theory down
to four dimensions. In this case, the four string coordinates remain uncompactified,
while the other six are curled up and parametrize a compact space of a very small
size (of the order of the Plank length). It appears that the internal space cannot be
completely arbitrary – it must have vanishing Ricchi-curvature.
One of the major obstacles to build a unified theory is the left-right asymmetry
recorded in the present days experiments. A theory in which there is an asymmetry
between the left and the right must contain chiral fermions. Chiral fermions are
usually a source of anomalies, i.e., of breakdown of classical symmetries by quantum
effects. Anomalies render a theory eventually inconsistent. Only in some special,
“rear” cases anomalies cancel (as it happens for instance for a standard generation
of quarks and leptons, cf. the course on the Standard Model). In higher space-time
dimensions it becomes even more non-trivial to achieve cancellation of anomalies.
–5–
?
Type I
Type IIB
E 8 x E 8 heterotic
Type IIA SO(32) heterotic
String theories
Remarkably, in 1983, Alvarez-Gaumé and Witten demonstrated that anomaly can-

celled out in the chiral Type IIB supergravity theory. This was nice, but the cor-
responding theory was still far from phenomenology. Shortly afterwards Green and
Schwarz (1984) showed that cancellation of anomalies in the open superstring the-
ories singles out two gauge groups, namely SO(32) and E8 × E8 . The gauge groups
SO(32) and E8 × E8 are realized in the so-called “heterotic string” which is a hy-
brid of the old d = 26 dimensional bosonic string and the d = 10 superstring. It
was shown that compactifying the theory down to four dimensions one can obtain
several generations of the chiral fermions. It was a first example of a string theory
whose low-energy limit was not in an immediate conflict with all known physics.
One of the important questions is why there exists non a unique but several
consistent string theories. With a recent advent of the string dualities and D-branes
an interesting new picture start to emerge (but still very far from being complete),
according to which different superstring theories provide different descriptions of
the one and the same theory valid in different regimes of the coupling constant
parameters.
Recommended literature:
1. M. Green, J. Schwarz and E. Witten, “Superstring Theory”, volume I and II,

Cambridge University Press, 1987.
2. D. Lüst and S. Theisen, “Lectures on string theory”, Lecture Notes in Physics,

Springer Verlag, 1989.
3. ’t Hooft, “Introduction to String Theory”, Lecture Notes, 2005.
–6–
4. B. Zwiebach, “A first course in string theory”, Cambridge university press,
2004.
5. J. Polchinski, “String theory”, volume I and II, Cambridge university press,

1998.
–7–
2. Relativistic particle
Consider relativistic particle of mass m moving in d-dimensional Minkowski space:
ηµν = (−1, +1, +1, . . . , +1).
Action Z s
S = −α ds
s0
Rs
Note that s0 ds has maximum along straight lines, this explains the sign “-” in front
of the action.
Embedding xµ ≡ xµ (τ ):
r r
dxµ dxµ dxµ dxν
ds = − dτ ≡ −ηµν dτ
dτ dτ dτ dτ
If xµ = (cτ, ~x) then
√ d~r
ds = c2 − ~v 2 , ~v =
dτ
Thus, the action is r
Z τ1
~v 2
S = −αc 1− dτ
τ0 c2
The Lagrangian in the non-relativistic limit
r µ ¶
~v 2 ~v 2 α~v 2
L = −αc 1 − 2 dτ = −cα 1 − 2 + . . . = −αc + + ...
c 2c 2c
To get the standard kinetic energy one has to identify
α = mc
In what follows we will work in units in which c = 1.

The action is invariant under reparametrizations of τ :
δxµ = ξ(τ )∂τ xµ as long as ξ(τ0 ) = ξ(τ1 ) = 0
Let us show this

p 1 1
δ( −ẋµ ẋµ ) = p µ
(−2ẋν δ ẋν ) = − p ẋν ∂τ (ξ ẋν ) =
2 −ẋµ ẋ −ẋµ ẋµ
1 h i 1 p
= −p ẋ ν
ẋ ν
˙
ξ + ξ ẋ ν
ẍ ν = − p ẋ ν
ẋ ν
˙ + ξ∂τ ( −ẋµ ẋµ )
ξ
−ẋµ ẋµ −ẋµ ẋµ
p p p
= −ẋµ ẋµ ξ˙ + ξ∂τ ( −ẋµ ẋµ ) = ∂τ (ξ −ẋµ ẋµ )
–8–
Therefore, we arrive at
Z h p i
τ1 p
δS = −m ∂τ (ξ −ẋµ ẋµ ) = −m ξ −ẋµ ẋµ |ττ =τ
=τ0 = 0 ,
1
τ0
i.e., the action is indeed invariant w.r.t. the local reparametrization transformations.
The most elegant way to quantize a system is to use the Hamiltonian formal-
2
ism . Let us therefore try to develop the Hamiltonian (canonical) formalism for the
relativistic particle. The canonical momenta
∂L ∂ p ẋµ
pµ = µ
= −m µ
( −ẋµ ẋµ ) = m p
∂ ẋ ∂ ẋ −ẋµ ẋµ
We notice that
p2 ≡ pµ pµ = −m2
Thus, we see that the canonical momenta are not independent, rather they obey the
constraint
φ ≡ p2 + m2 = 0 (2.1)
Constraints which follow just from the definition of the canonical momenta without
using equations of motion are called primary constraints. Mass-shell condition for
the point particle is a primary constraint.
In general the number of primary constraints is equal to the number of zero

eigenvalues of the Hessian matrix:
∂pµ ∂ 2L
Hµν = =
∂ ẋν ∂ ẋµ ∂ ẋν
For the relativistic particle we have only one eigenvector ẋµ with zero eigenvalue
Ã !
∂pµ ν m ẋµ ẋν
ẋ = p δµν + m 3 ẋν = 0
∂ ẋν −ẋµ ẋµ (−ẋµ ẋµ ) 2
The inverse function theorem stats that absence of zero eigenvalues of the Hessian
is a necessary condition to be able to express the velocities ẋµ via the canonical
momenta pµ . Dynamical systems with the Hessian of non-maximal rank are called
singular.
Constraints which have vanishing Poisson bracket:
{φi , φj } = 0
2
About the Hamiltonian approach to dynamical systems of classical mechanics, the reader may
consult appendix A.
–9–
are called the first class constraints. The mass-shell constraint for the relativistic
particle is of the first class.
There is another action for the relativistic particle. It has the following two
characteristic features
• it does not contain square root
• it admits generalization to the massless case
Introduce e(τ ) the auxiliary field on the world-line.

Z
1 τ1 h 1 ³ dxµ dxµ ´ i
S= dτ − em2
2 τ0 e dτ dτ
Equations of motion:
d ³ 1 µ´
for xµ ẋ = 0
dτ e
1 1
for e(τ ) − 2 ẋµ ẋµ − m2 = 0 =⇒ ẋ2 + m2 e2 = 0
2e 2
The last equation can be solved for e:
1 2
e2 = − ẋ
m2
which leads to
dh m µ
i
√ ẋ =0
dτ −ẋν ẋν
which is nothing else as the old eom for xµ . Also if we substitute solution for e into
the new action then this action reduces to the old one:
Z √ Z τ1 √
1 τ1 h m 2 −ẋ2 2 i
S= dτ √ ẋ − m = −m dτ −ẋ2 (2.2)
2 τ0 −ẋ2 m τ0
Also
1 1
pµ = ẋµ =⇒ p2 = 2 ẋ2 = −m2
e e
but this time due to equations of motion for e. Equation of motion for e is purely
algebraic. The Hessian
∂2L 1 ∂ 1
= ẋ ν = ηµν
∂ ẋµ ∂ ẋν e ∂ ẋµ e
is of maximal rank.
The constraint p2 + m2 = 0 does not follow from the definition of the canonical
momentum along, but one has to use equations of motion. Constraints which are
satisfied as consequences of equations of motion are called secondary.
– 10 –
The action has local gauge symmetry which is now
δxµ = ξ ẋµ
δe = ∂τ (ξe)
The first symmetry transformation is generated by the constraint p2 + m2 :

η
δη xµ = η{p2 + m2 , xµ } = 2ηpµ = ẋµ = ξ ẋµ , η ≡ eξ .
e
The second transformation for e is easily derived from e2 = − m12 ẋ2 . Thus, in our
new formulation we have the set of fields (xµ , e) and reparametrization symmetry
which acts on them and leaves the action invariant. This reparametrization freedom
can be used to put e = m1 which results into the following equation
ẍµ = 0
This is not the end however, because there is eom for e which now reads as ẋ2 = −1.
Complete eoms are
ẍµ = 0 , ẋ2 = −1
Thus, the relativistic particle moves freely in Minkowski space over time-like geodesics.
Space-like and light-like straight lines are excluded by the constraint ẋ2 = −1.
In the case of the massless particle we can set e = 1 and get eoms
ẍµ = 0 , ẋ2 = 0
In both, the massive and massless cases, the constraints are integral of motions: they
are preserved in time due to the dynamical equation ẍµ = 0.
Finally, we treat the relativistic particle in the so-called first order (the Hamil-
tonian) formalism. To this end we have to represent the initial Lagrangian in the
form
L = pµ ẋµ + Lrest
where pµ = 1e ẋµ and express in Lrest the derivatives ẋµ via pµ . In doing so we obtain
the phase-space Lagrangian
e
L = pµ ẋµ − (p2 + m2 )
2
We clearly see that the auxiliary field e we introduced in our second formulation
plays here the role of the lagrangian multiplier to the constraint p2 + m2 = 0. By
using the gauge freedom we can fix the gauge e = m1 and the physical Hamiltonian
becomes in this case
1 2
H= (p + m2 )
2m
– 11 –
which is in complete agreement with our previous discussion (We actually have to
identify θ = e). This Hamiltonian3 should be provided with the constraint p2 +m2 = 0
which is the eom for e.
More generally, evolution of a singular system is governed by the Hamiltonian
X
H = Hcan + χ n φn
n
Here {φn } is an irreducible set of primary constraints and Hcan is the canonical
Hamiltonian:
Hcan = pµ ẋµ − L
Only on the constraint surface φn = 0 the Hamiltonian H coincides with Hcan . In
our present case
ẋν ẋν p
Hcan = m p − (−m −ẋµ ẋµ ) = 0
−ẋµ ẋµ
and Hamiltonian dynamics of the system is due to the mass-shell condition only. The
choice of the coefficients χn (τ ) in H is equivalent to the choice of the gauge.
θ 2 θ
H = Hcan + χφ = (p + m2 ) , χ=
2m 2m
We get the time-evolution
dxµ θ 2 θ θẋµ
={ (p + m2 ), xµ } = pµ = p
dτ 2m m −ẋµ ẋµ
Therefore ẋ2 = −θ2 . Choosing θ = 1 means that we identify the time variable with
the proper time. This nicely illustrates the general point: in order to write down the
evolution equations in a system with local gauge invariance one has first to identify
the “time” variable.
Other gauge choices are possible. For instance, the static gauge consists in
imposing the condition t ≡ x1 = τ . Equation for pt ≡ p1 allows us to determine e:
δL dt 1
= − ept = 0 =⇒ e=
δpt dτ pt
The physical Hamiltonian dual to the world-line time τ coincides in this case with
the momentum pt conjugate to t: H = pt . It can be found from the eom for e:
p
p2 = m2 = 0 =⇒ − p2t + p~2 + m2 = 0 =⇒ pt = p~2 + m2 .
Note that here p~ = {pi } with i = 2, . . . , d.

3 1
The gauge e = m is a close analogue of the conformal gauge to be considered for the string
case.
– 12 –
This gauge choice leads to the common Hamiltonian of the relativistic particle
p
H = p~2 + m2 .
This exercise with the relativistic particle also shows how sensitive is the Hamiltonian
to the choice of the gauge. Fixing the gauge e = 1 we get the polynomial Hamiltonian,
while fixing the static gauge the Hamiltonian appears to be a non-linear square root.
Finally, there is another type of gauge known as the light-cone gauge. We intro-
duce the light-cone coordinates
t = x+ − x− xd = x+ + x−
1 1
pt = (p+ + p− ), pd = (p+ − p− )
2 2
and denote the other “transverse coordinates” xi and pi with i = 2, . . . , d − 1 as ~x
and p~. Then the phase space Lagrangian becomes
e
L = ẋ− p+ − ẋ+ p− + ~x˙ p~ − (−p− p+ + p~2 + m2 )
2
The light-cone gauge consists in choosing x+ = τ . From the kinetic term of the
Hamiltonian it is clear that the variable x+ is conjugate to p− and therefore p− is
the physical Hamiltonian. It can be easily found from the equation for e:
1 2
H= (~p + m2 )
p+
The gauge-fixed Lagrangian becomes
1
L = ẋ− p+ + ~x˙ p~ − p− = ~x˙ p~ − H = ~x˙ p~ − + (~p2 + m2 )
p
The variable p+ is canonically conjugate to x− .
Notice that both in the static and in the light-cone gauge the number of physical
degrees of freedom is 2(d − 1). The auxiliary field e was solved in terms of physical
fields. The physical phase space inherits the canonical Poisson bracket.
In the next lecture we extend these different approaches to dynamics of relativis-
tic particles to relativistic strings.
3. Classical relativistic bosonic string

3.1 Nambu-Goto string
Two-dimensional surface traced by string during its time evolution is called world-
sheet. The action for a relativistic string should be a functional of a string trajectory,
i.e. of the world-sheet.
– 13 –
The Nambu-Goto action
Z Z s µ ¶
2 ∂X µ ∂X ν
SN G = −T dA = −T dσ −det ηµν (3.1)
∂σ α ∂σ β
This we can write as

·µ ¶¸
Ẋ µ Ẋµ Ẋ µ Xµ0
−det = (ẊX 0 )2 − Ẋ 2 X 02 . (3.2)
Ẋ µ Ẋµ X 0µ Xµ0
Thus,
Z q Z
2
√
SN G = −T dσ (ẊX 0 )2 − Ẋ 2 X 02 = −T d2 σ −Γ
Here Γ = detΓαβ , where
∂X µ ∂X ν
Γαβ = ηµν (3.3)
∂σ α ∂σ β
is the metric induced on the string world-sheet.
What is a local characteristic of the string world-sheet? Consider a point on the
world-sheet and the space of all vectors tangent to the surface is at this point. These
vectors sweep two-dim vector space. The physical propagation of the string requires
that in these two-dim vector space there is a basis built over two vectors one of them
is time-like and another is space-like.
Recall the standard definitions from the theory of special relativity. We have the
invariant interval between two infinitezimal events:
3
X
2 µ ν 0 0
−ds = ηµν dx dx = −dx dx + dxi dxi . (3.4)
i=1
• If ds2 > 0 the interval is called time-like. In this case the different events which
happen in the same space-point are always time-separated.
• If ds2 < 0 the interval is called space-like. In this case events which happen at
the same time are space-separated.
• If ds2 = 0 the interval is light-like. Vectors v µ obeying the condition v 2 =

ηµν v µ v ν = 0 are called light-like or null.
Identify x0 = t = τ . Then
Ẋ µ Ẋµ = −1 + ~v 2 ≤ 0 ⇐ time − like
X 0µ Xµ0 = (X 0i )2 ≥ 0 ⇐ space − like
– 14 –
This situation should persist in any Lorentz frame. At any point on a string world-
sheet one should always be able to find two vectors: one is time-like and another is
space-like.
µ µ
Consider S µ (λ) = ∂X
∂τ
+ λ ∂X
∂σ
must be space- or time-like as λ varies.
S µ Sµ = Ẋ 2 + 2λẊX 0 + λ2 X 02 ≡ y(λ) .
Discriminant must be positive then there are two roots and therefore the regions of
λ with time- and space-like vectors. Discriminant is
(ẊX 0 )2 − Ẋ 2 X 02 > 0 .
This condition guarantees the causal propagation of string.

Action is invariant under reparametrizations
δX µ = ξ α ∂α X µ , ξ α = 0 on the boundary
Two possibilities:
• Open strings: 0 ≤ σ ≤ π
• Closed strings: 0 ≤ σ ≤ 2π
Equations of motion: Z Z
τ2 π
S= dτ dσL
τ1 0
Variation
Z τ2 Z π µ ¶
δS δL δL
= dτ dσ ∂τ δX µ + ∂ δX µ
0µ σ
(3.5)
δX µ τ1 0 δ Ẋ µ δX
Z τ2 Z π µ ¶
δL δL
=− dτ dσ ∂τ + ∂σ (3.6)
τ1 0 δ Ẋ µ δX 0µ
Z π Z τ2
δL δL
+ dσ δX µ |ττ21 + 0µ
δX µ |σ=π
σ=0 (3.7)
0 δ Ẋ µ
τ1 δX
δL δL
• Open string boundary conditions: δX 0µ
(τ, σ = π) = δX 0µ
(τ, σ = 0) = 0
• X µ (σ + 2π) = X µ (σ).
Canonical formalism. Momentum
δL (ẊX 0 )X 0µ − (X 0 )2 Ẋ µ
Pµ = = −T q
δ Ẋ µ (ẊX 0 )2 − Ẋ 2 X 02
We see that
P µ Xµ0 = 0 (3.8)
µ 2 02
P Pµ + T X = 0 (3.9)
– 15 –
Exercise Show that the Hessian matrix has for each σ two zero eigenvalues cor-
responding to Ẋ µ and X 0µ .
Equations of motion are very complicated:
   
0 0µ 0 2 µ 0 µ 2 0µ
∂  (ẊX )X − (X ) Ẋ  ∂  (ẊX )Ẋ − (Ẋ) X 
q + q = 0.
∂τ 0 2 2 02 ∂σ 0 2 2 02
(ẊX ) − Ẋ X (ẊX ) − Ẋ X
Exercise Think how reparametrization invariance can be used to bring this equa-
tion to the simplest form.
3.2 The Polyakov action

Introduce a word-sheet metric hαβ (σ, τ ). Consider the action
Z
T √
Sp = − d2 σ −hhαβ ∂α X µ ∂β X ν ηµν (3.10)
2
Here h = dethαβ .
3.2.1 Symmetries
The reparametrization invariance
δX µ = ξ α ∂α X µ
δhαβ = ξ γ ∂γ hαβ + hαγ ∂β ξ γ + hβγ ∂α ξ γ
δhαβ = ξ γ ∂γ hαβ − hαγ ∂γ ξ β − hβγ ∂γ ξ α
√ √
δ( −h) = ∂α (ξ α −h)
The Weyl invariance

The Weyl invariance consists in rescaling the metric
hαβ → e−2Λ(σ,τ ) hαβ . (3.11)
3.2.2 Equations of motion

First we discuss the equation of motion for the intrinsic metric hαβ . This discussion
amounts to the introduction of the two-dimensional stress-energy tensor which is a
response of the action to the change of the metric
Z
√
δSp = −T d2 σ −h Tαβ δhαβ
that is
1 δSp
Tαβ = − √
T −h δhαβ
– 16 –
Thus, eom for hαβ is
Tαβ = 0 .
Performing a variation we obtain
1 1
Tαβ = ∂α X µ ∂β Xµ − hαβ hγδ ∂γ X µ ∂δ Xµ . (3.12)
2 4
From here we immediately notice that the stress-energy tensor is traceless
Tαα = Tαβ hαβ = 0
This is a direct consequence of the Weyl symmetry.
Due to the equations of motion for the scalar fields X µ this tensor is covariantly
conserved:
∇α Tαβ = 0 .
This can be easily derived as follows. Consider a variation of the action:
Z ³ δL ´
δL
δSp = d2 σ δh αβ
+ δX µ
δhαβ δX µ
δL
We see that on the equations of motion for X µ , i.e. on δX µ = 0, we have
Z Z
√ √
δSp = −T d2 σ −h Tαβ δhαβ = −2T d2 σ −h Tαβ ∇α ξ β ,
where on the r.h.s. we specified the variation to be a diffeomorphism transformation.

The last expression can be integrated by parts and, die to δSp = 0, we conclude that
∇α Tαβ = 0. This derivation is similar to the derivation of the charge conservation
law in electrodynamics, the latter being a consequence of the gradient invariance.
Let us show directly that ∇α Tαβ = 0. We have
α α µ µ α 1 γδ µ
2∇ Tαβ = ∇ ∂α X ∂β Xµ + ∂α X ∇ ∂β Xµ − ∇β h ∂γ X ∂δ Xµ (3.13)
2
Since ∇α ∂α X µ = 0 is eom for X µ we find
α µ α 1 γδ µ
2∇ Tαβ = ∂α X ∇ ∂β Xµ − ∂β h ∂γ X ∂δ X µ =
2
µ αδ 1 γδ µ γδ µ
= ∂α X h ∇δ ∂β Xµ − ∂β h ∂γ X ∂δ Xµ − h ∂γ X ∂δ ∂β Xµ =
2
µ αδ µ αδ s 1 γδ µ γδ µ
= ∂α X h ∂δ ∂β Xµ − ∂α X ∂s Xµ h Γδβ − ∂β h ∂γ X ∂δ Xµ − h ∂γ X ∂δ ∂β Xµ .
2
Here the first term cancels with the last one and we find
α µ αδ s 1 γδ µ
2∇ Tαβ = −∂α X ∂s Xµ h Γδβ − ∂β h ∂γ X ∂δ Xµ
2
Only symmetric part of hαδ Γs 1 αδ hsp ∂ h

δβ = 2 h δ pβ + ∂β hpδ − ∂p hβδ in the indices α, s matters! This translates into symmetry of
β, δ. Finally,
α 1 αδ sp µ 1 γδ µ
2∇ Tαβ = − h ∂β hpδ h ∂α X ∂s Xµ − ∂β h ∂γ X ∂δ Xµ = 0 . (3.14)
2 2
Equations
Tαα = 0 , ∇β Tαβ = 0
are consequences of the Weyl and diffeomorphism symmetry, respectively. These
two are gauge symmetries, and the relations above can be understood as the second
Noether theorem.
– 17 –
3.2.3 Conformal gauge
Consider the conformal Killing equation
ξ γ ∂γ hαβ + hαγ ∂β ξ γ + hβγ ∂α ξ γ = Λhαβ (3.15)
and solve it assuming the conformal gauge

µ ¶
2φ −1 0
hαβ = e (3.16)
0 1
This equation can be split as follows. First we take α = τ and β = σ and using
hτ σ = 0 we find
hτ τ ∂σ ξ τ + hσσ ∂τ ξ σ = 0 → ∂σ ξ τ − ∂τ ξ σ = 0
Second, we take α, β = τ and then α, β = σ. We get
ξ γ ∂γ hτ τ + 2hτ τ ∂τ ξ τ = Λhτ τ
ξ γ ∂γ hσσ + 2hσσ ∂σ ξ σ = Λhσσ
i.e.
ξ γ hτ τ ∂γ hτ τ + 2∂τ ξ τ = Λ
ξ γ hσσ ∂γ hσσ + 2∂σ ξ σ = Λ .
Subtracting one from the other we get
∂τ ξ τ − ∂σ ξ σ = 0 .
Thus, the conformal Killing equations reduce in the conformal gauge to
∂σ ξ τ − ∂τ ξ σ = 0
∂τ ξ τ − ∂σ ξ σ = 0 (3.17)
or equivalently
(∂τ + ∂σ )(ξ τ − ξ σ ) = 0
(∂τ − ∂σ )(ξ τ + ξ σ ) = 0 (3.18)
By using the world-sheet light-cone coordinates σ ± = τ ± σ this can be reformulated

as4
∂+ ξ − = 0 = ∂− ξ + .
³ ´
4 1
The basic relations are ∂τ = ∂+ + ∂− , ∂σ = ∂+ − ∂− , ∂± = 2 ∂τ ± ∂σ .
– 18 –
Thus,
ξ + = ξ + (σ + ) , ξ − = ξ − (σ − )
solve the conformal Killing equations. This is a freedom (reparametrization + Weyl
rescaling) which does not destroy the conformal gauge condition.
The final remark concerns restrictions on ξ ± following from the periodicity con-
ditions. Indeed, for the Weyl factor Λ we have
ξ τ ∂τ φ + ξ σ ∂σ φ + 2∂τ ξ τ = Λ(σ, τ ) .
The factor Λ must be periodic in σ because the metric hαβ is. Since φ is periodic
the allowed solutions ξ ± of the conformal Killing equation are those which lead to
periodic Λ. One can easily see that this implies that ξ ± must be periodic in σ.
3.2.4 Polyakov string in the first order formalism.

In the first order formalism the density L ≡ L(σ, τ ) of the Polyakov Lagrangian takes
the form
1 ³ ´ γτ σ ³ ´
L = Pµ ∂τ X µ + P µ P µ
+ T 2 0
Xµ X 0µ
+ Pµ X 0µ
(3.19)
2T γ τ τ γτ τ
The conformal gauge consists in imposing the following two conditions
γ τ τ = −1 , γ τ σ = 0.
The gauged-fixed Lagrangian density is

1 ³
µ µ 2 0 0µ
´
L = Pµ ∂τ X − Pµ P + T Xµ X (3.20)
2T
and the Hamiltonian density is
1 ³ ´
H= Pµ P µ + T 2 Xµ0 X 0µ
2T
The phase-space Lagrangian shows that the variables (Pµ , Xµ ) are canonical, i.e. the
corresponding Poisson bracket is
{X µ (σ, τ ), X ν (σ 0 , τ )} = {P µ (σ, τ ), P ν (σ 0 , τ )} = 0 , (3.21)

{X µ (σ, τ ), P ν (σ 0 , τ )} = η µν δ(σ − σ 0 ) . (3.22)
The dynamics of the system in the conformal gauge is governed by the Hamiltonian
Z Z ³ ´
1
H = dσ H = dσ Pµ P µ + T 2 Xµ0 X 0µ
2T
Equations of motion
1 µ
Ẋ µ = {X µ , H} = P , (3.23)
T
Ṗ µ = {P µ , H} = T X 00µ , (3.24)
– 19 –
which result into
Ẍ µ − X 00µ = ¤X µ = 0
We see that we have two constraints
C1 = Pµ P µ + T 2 Xµ0 X 0µ , C2 = Pµ X 0µ (3.25)
We find the following Poisson brackets
{C1 (σ), C1 (σ 0 )} = 4T 2 ∂σ C2 (σ)δ(σ − σ 0 ) + 8T 2 C2 (σ)∂σ δ(σ − σ 0 ) , (3.26)

0 0 0
{C1 (σ), C2 (σ )} = ∂σ C1 (σ)δ(σ − σ ) + 2C1 (σ)∂σ δ(σ − σ ) , (3.27)
{C2 (σ), C1 (σ 0 )} = ∂σ C1 (σ)δ(σ − σ 0 ) + 2C1 (σ)∂σ δ(σ − σ 0 ) , (3.28)
{C2 (σ), C2 (σ 0 )} = ∂σ C2 (σ)δ(σ − σ 0 ) + 2C2 (σ)∂σ δ(σ − σ 0 ) . (3.29)
Instead of the constraints C1 and C2 we can equally consider their linear combi-
nations
1 1
T++ = (C1 + 2T C2 ) = (Pµ + T Xµ0 )2 , (3.30)
8T 2 8T 2
1 1
T−− = (C1 − 2T C2 ) = (Pµ − T Xµ0 )2 . (3.31)
8T 2 8T 2
Their Poisson algebra becomes

1 ³ ´
{T++ (σ), T++ (σ 0 )} = ∂σ T++ (σ)δ(σ − σ 0 ) + 2T++ (σ)∂σ δ(σ − σ 0 ) ,
2T
0 1 ³ ´
{T−− (σ), T−− (σ )} = − ∂σ T−− (σ)δ(σ − σ 0 ) + 2T−− (σ)∂σ δ(σ − σ 0 ) ,
2T
0
{T++ (σ), T−− (σ )} = 0 . (3.32)
Constraints T++ and T−− Poisson commute and form two independent Poisson alge-
bras!
Now one can easily find the evolution equations for T++ and T−− . We have
∂τ T++ = {T++ (σ), H} = ∂σ T++ =⇒ (∂τ − ∂σ )T++ = ∂− T++ = 0 ,

∂τ T−− = {T−− (σ), H} = −∂σ T−− =⇒ (∂τ + ∂σ )T−− = ∂+ T−− = 0 ,
Thus, we see that evolution equations imply that
T++ = T++ (σ + ) , T−− = T−− (σ − ) . (3.33)
The Hamiltonian itself is

Z 2π
H = 2T dσ(T++ + T−− ) ,
0
i.e. it is a sum of zero modes of the left- and right-moving constraints.
– 20 –
3.3 Integrals of motion. Classical Virasoro algebra.
String has an infinite set of integrals of motion (quantities which are conserved in
time due to equations of motion) which are constructed with the help of T±± .
Note that T±± themselves are not the good conserved quantities as they depend
on time! However, if we define the Fourier components
Z 2π
−
Lm = 2T dσ eimσ T−− (σ, τ )
0
5
then
Z 2π ³ ´
dLm
= 2T dσ ∂τ eim(τ −σ) T−− (σ, τ ) + eim(τ −σ) ∂τ T−− (σ, τ ) (3.34)
dτ 0
Z 2π ³ ´
im(τ −σ) im(τ −σ)
= 2T dσ ime T−− (σ, τ ) − e ∂σ T−− (σ, τ )
0
Z 2π ³ ´
= 2T dσ imeim(τ −σ) T−− (σ, τ ) + ∂σ eim(τ −σ) T−− (σ, τ ) = 0
0
Thus, the Fourier components of the stress-energy tensor provide an infinite set of
the conserved constraints. Analogously, we define
Z 2π
+
L̄m = 2T dσ eimσ T++ (σ, τ ) ,
0
which is also an integral of motion. Note that in this derivation we never used the
constraints T++ = 0 = T−− .
The Poisson brackets of the Lm and L̄m generators are
{Lm , Ln } = −i(m − n)Lm+n ,

{L̄m , L̄n } = −i(m − n)L̄m+n , (3.35)
{Lm , L̄n } = 0 .
Z
2 0 −imσ−inσ 0 0
{Lm , Ln } = 4T dσdσ e {T−− (σ), T−− (σ )}
Z
0 −imσ−inσ 0 0 0
= −2T dσdσ e ∂σ T−− (σ)δ(σ − σ ) + 2T−− (σ)∂σ δ(σ − σ ) =
Z
−i(m+n)σ 0 −imσ −inσ
= −2T dσ − T−− (σ)∂σ e + 2e T−− (σ)∂σ e
Z
−i(m+n)σ
= −i(m − n) 2T dσe T−− (σ) = −i(m − n)Lm+n
This is the so-called Wit algebra. This algebra acts on X µ (σ, τ ):

Z 2π
µ 0−
{Lm , X } = 2T dσ 0 eimσ {T−− (σ 0 ), X µ (σ)} =
0
1 − −
= − eimσ (Ẋ µ − X 0µ ) = −eimσ ∂− X µ . (3.36)
2
5
When finding the time dynamics of Lm one has to remember that the function Lm has an
explicit time-dependence and, therefore dL
dτ = ∂τ Lm + {Lm , H}.
m
– 21 –
Analogously,
Z 2π
µ 0+
{L̄m , X } = 2T dσ 0 eimσ {T++ (σ 0 ), X µ (σ)} =
0
1 + +
= − eimσ (Ẋ µ + X 0µ ) = −eimσ ∂+ X µ . (3.37)
2
To summarize
− +
{Lm , X µ } = −eimσ ∂− X µ , {L̄m , X µ } = −eimσ ∂+ X µ . (3.38)
In particular,
{L̄0 − L0 , X µ } = ∂σ X µ (3.39)
i.e. L̄0 −L0 generates rigid σ-translations. The transformations we consider transform
a solution
¤X µ = 0
into another solution of this equation. Indeed, we have for instance
³ −
´ − −
∂+ ∂− eimσ ∂− X µ = imσ − eimσ ∂+ ∂− X µ + eimσ ∂− ∂+ ∂− X µ = 0
as the consequence of ∂+ ∂− X µ = 0.
(P , X)
_
Lm L m =0
L m =0
_
Lm
Fig. 1. The phase space of string. The Virasoro constraints Lm = 0 = L̄m

define a time-independent hypersurface on which the dynamics of string
takes place. This hypersurface remains invariant under the action of Lm ’s
and L̄m ’s.
Even more generally, for any periodic function f we define

Z 2π
Lf = dσf (σ + )T++ .
0
– 22 –
Then we see that
Z 2π ³ ´ Z 2π ³ ´
+
0= dσ∂− f (σ )T++ = ∂τ Lf − dσ∂σ f (σ + )T++ .
0 0
Thus,
Z 2π ³ ´
+
∂τ Lf = dσ∂σ f (σ )T++ = 0 .
0
To summarize: we see that upon fixing conformal gauge we are still left with
gauge freedom. It corresponds to reparametrizations of the special type (solutions
to the conformal Killing equation):
σ + → ξ + (σ + ) , σ − → ξ − (σ − ) ,
where ξ ± are two arbitrary functions periodic in σ.
Finally, for the case of open string we define

Z π ³ ´
+ −
Lm = 2T dσ eimσ T++ + eimσ T−− . (3.40)
0
The point is that there appears additional boundary terms which show that the old
Lm and L̄m are not separately conserved. With our new definition of Lm we obtain
Z π ³ ´
dLm + + − −
= 2T dσ imeimσ T++ + eimσ ∂τ T++ + imeimσ T−− + eimσ ∂τ T−−
dτ
Z0 π ³ ´
+ + − −
= 2T dσ imeimσ T++ + eimσ ∂σ T++ + imeimσ T−− − eimσ ∂σ T−−
Z0 π ³ ´
+ + − −
= 2T dσ imeimσ T++ − ∂σ eimσ T++ + imeimσ T−− + ∂σ eimσ T−−
0
³ ´ ³ ´
im(τ +π)
+ 2T e T++ (π, τ ) − e−2πim T−− (π, τ ) − 2T eimτ T++ (0, τ ) − T−− (0, τ )
The bulk term vanishes as before and we left with the boundary term
dLm ³ ´ ³ ´
= 2T eim(τ +π) T++ (π, τ ) − T−− (π, τ ) − 2T eimτ T++ (0, τ ) − T−− (0, τ )
dτ
Due to the open string boundary conditions X 0µ (π, τ ) = 0 = X 0µ (0, τ ) we obviously

have
T++ (π, τ ) = T−− (π, τ ) , T++ (0, τ ) = T−− (0, τ )
Therefore, the boundary term vanishes and, therefore, Lm are conserved quantities
in the open string case.
– 23 –
3.3.1 Solutions of the equations of motion
Here we are going to discuss the solutions of the string equations off motion
¤X µ = 0
which arises upon fixing the conformal gauge.
• Closed strings. We have
X µ (σ, τ ) = XLµ (τ + σ) + XRµ (τ − σ)
where
1 pµ i X 1 µ −in(τ −σ)
XRµ (τ − σ) = xµ + (τ − σ) + √ αn e (3.41)
2 4πT 4πT n6=0 n
1 pµ i X 1 µ −in(τ +σ)
XLµ (τ + σ) = xµ + (τ + σ) + √ ᾱn e (3.42)
2 4πT 4πT n6=0 n
Since X µ (σ, τ ) are real then (xµ , pµ ) are real as well and
µ µ
α−n = (αnµ )† , ᾱ−n = (ᾱnµ )†
Let us define the zero modes as

1
α0µ = ᾱ0µ = √ pµ .
4πT
Oscillators obey the Poisson relations
µ
{αm , αnν } = {ᾱm
µ
, ᾱnν } = −imδm+n η µν ,
µ
{αm , ᾱnν } = 0 (3.43)
µ ν µν
{x , p } = η .
The Virasoro constraints become

∞ ∞
1 X µ 1 X µ
Lm = α αnµ , L̄m = ᾱ ᾱnµ . (3.44)
2 n=−∞ m−n 2 n=−∞ m−n
• Open strings Solution of the wave equation with the open string boundary
conditions is
pµ i X 1 µ −inτ
X µ (σ, τ ) = xµ + τ+√ αn e cos nσ . (3.45)
πT πT n6=0 n
– 24 –
Oscillators obey the Poisson algebra
µ
{αm , αnν } = −imδm+n η µν ,
{xµ , pν } = η µν . (3.46)
If we define the zero mode as α0µ = √1πT pµ then the generators of the open
string classical Virasoro algebra are realized as
∞
1 X µ
Lm = α αnµ . (3.47)
2 n=−∞ m−n
µ
The hermiticity property of the non-zero modes are α−n = (αnµ )† .
3.3.2 Poincaré symmetry. String tension.

The Noether currents of the Poincaré symmetry
√
Pµα = −T −hhαβ ∂β Xµ (3.48)
√
Jαµν = −T −hhαβ (Xµ ∂β Xν − Xν ∂β Xµ ) (3.49)
Here Pα is a current corresponding to translational invariance X µ → X µ + a, where

a is an arbitrary constant, and J µν is a current corresponding to Lorentz rotations
X µ → Λµν X ν , where Λµν is a (constant) matrix comprising the parameters of the
Lorentz transformation. We see that
Jαµν = Xµ Pνα − Xν Pµα .
Both currents are conserved due to equations of motion of X µ and their τ -components
integrated over σ define the conserved charges (for the open string case integration
rans from 0 to π)
Z 2π Z 2π
µ µ µν
P = dσPτ , J = dσJτµν .
0 0
Imposing the conformal gauge and using the fundamental brackets for (X, P ) one
finds the following Poisson algebra
{P µ , P ν } = 0
{P µ , J ρσ } = η µσ P ρ − η µρ P σ (3.50)
{J µν , J ρσ } = η µρ J νσ + η νσ J µρ − η νρ J µσ − η µσ J νρ .
Substituting solution for X µ one finds

Z 2π
µ
P =T dσ Ẋ µ = pµ ,
0
– 25 –
i.e. the total mass momentum of string coincides with pµ .
The total angular momentum of closed string in the conformal gauge is is defined as
Z 2π
µν
J =T dσ(X µ Ẋ ν − X ν Ẋ µ ) . (3.51)
0
Substituting the oscillator expansion we get
J µν = xµ pν − xν pµ +S µν + S̄ µν ,
| {z }
`µν
where
X∞
µν 1¡ µ ν ν
¢
S = −i α−n αn − α−n αnµ , (3.52)
n=1
n
X∞
µν 1¡ µ ν ν
¢
S̄ = −i ᾱ−n ᾱn − ᾱ−n ᾱnµ . (3.53)
n=1
n
For the case of open string expressions are the same (again integration runs from 0 to
π) except S̄ µν is absent. Here `µν is the angular momentum of string and S µν + S̄ µν
is its internal spin.
Let us show that both P µ and J µν are invariant under the action of the Virasoro
algebra. We have
{Lm , P µ } = {Lm , pµ } = 0 (3.54)
as Lm does not contain xµ . Let us separate the zero mode part6 of Lm

1 X ρ
Lm = α0ρ αmρ + α αnρ .
2 n6=0,m m−n
We first compute
1
{Lm , `µ } = {α0ρ αmρ , xµ pν − xν pµ } = √ αmρ {pρ , xµ pν − xν pµ } =
4πT
ν µ µ ν
= αm α0 − αm α0 . (3.55)
P µ
Second, since S µν = − k6=0 ki α−k αkν we have
X 1 ρ i µ ν
{ αm−n αnρ , − α−k αk } = 0 ,
n,k6=0;n6=m
2 k
X ρ i µ ν µ ν ν µ
{α0 αmρ , − α−k αk } = αm α0 − αm α0 ,
k6=0
k
6
We assume here that m 6= 0.
– 26 –
therefore, {Lm , `µν + S µν } = 0. Thus, Poincaré symmetry descends on the space of
physical states. Physical states will be classified in terms of representations of the
Poincaré group of d-dimensional Minkowski space.
An important invariant of the Poincaré group is the square of the momentum Pµ P µ .

Indeed,
{Pµ P µ , P ν } = {Pµ P µ , J ρλ } = 0 .
It coincides with the mass: M 2 = −Pµ P µ . Of course, P 2 is also invariant w.r.t. to
the action of Lm and L̄m . Consider now one of the constraints, L0 = 0. Separating
the zero mode
n=∞ ∞
1 X 1 2 X 1
L0 = αn α−n = α0 + αn α−n , α0µ = √ pµ
2 n=−∞ 2 n=1 4πT
we have
∞
X
µ
pµ p + 8πT αn α−n = 0 .
n=1
From here we deduce the mass

∞
X
2 2
M = −p = 8πT αn α−n .
n=1
Mass is created due to internal excitations of string! From this expression it is not
obvious that M 2 is non-negative.
Another invariant of the Poincaré group is

µ ¶
2 1 αβ 2 αλ β
J = Jαβ J + 2 Pα J P Jβλ .
2 M
Checking Poincaré invariance of this expression is straightforward, as an intermediate step we note the relations
µν ρσ νσ µρ µσ νρ
{J , Pρ J } = η J Pρ − η J Pρ
µ αβ µρ
{P , Jαβ J } = −4J Pρ .
The terms Jαβ J αβ and Pα J αλ P β Jβλ separately Poisson commute with J µν .
Consider an open string which has P i = pi = 0 for all i = 1, . . . , d − 1. By using reparametrization invariance fix the static
gauge X 0 = τ . The angular moment `µν of this string is zero. Also Pα J αλ P β Jβλ = 0 and, therefore,
2 1 ij
J = Jij J ,
2
where i, j = 1, . . . d − 1. We have
∞
X
2 1 1 i j j i i j j i
J = − α−n αn − α−n αn α−m αm − α−m αm
2 n,m=1 nm
∞
X ∞
X
1 ∗ ∗ ∗ ∗ 1 ∗ ∗
= (αn αm )(αm αn ) − (αn αm )(αn αm ) ≤ (αn αm )(αm αn ) .
n,m=1 nm n,m=1 nm
– 27 –
It follows from the Schwarz inequality that
∗ ∗ ∗ 2 ∗ ∗ ∗ ∗
(αn αm )(αm αn ) = |(αn αm )| ≤ (αn αn )(αm αm ) ≤ nm(αn αn )(αm αm ) .
Therefore,
∞
X ∞
X ∞
X
2 ∗ ∗ 1 1 4
J ≤ (αn αn )(αm αm ) = αn α−n αm α−m = M
n,m=1 4 n=−∞ m=−∞ (2πT )2
because in the open string case

∞
X X
2
M = πT αn α−n = 2πT αn α−n .
n=−∞ n>0
Thus, for an open string motion we found inequality
p 1 2
J ≡ J2 ≤ M .
2πT
Here the parameter

0 1
α =
2πT
is called a slope of the Regge trajectory. The function J = α0 M 2 is a straight line in the (M 2 , J) plane whose slope is α0 .
Consider a closed (pulsating) string solution
x = R cos σ cos τ , y = R sin σ cos τ , t = Rτ .
We see that
P0 E
P 0 = 2πRT =⇒ T = ≡ ,
2πR 2πR
i.e. tension is energy per unit length.
3.4 Strings in physical gauge

As we have seen upon fixing conformal gauge we are still left with the gauge freedom.
It corresponds to reparametrizations of the special type (solutions to the conformal
Killing equation):
σ + → ξ + (σ + ) , σ − → ξ − (σ − ) ,
where ξ ± are two arbitrary functions (periodic in σ). This freedom can be further
fixed leaving only physical excitations. This is achieved by imposing the so-called
light-cone gauge.
3.4.1 First order formalism

Introduce the light-cone coordinates in the d-dimensional Minkowski space
1
X ± = √ (X 0 ± X d−1 ), X i, i = 1, . . . , d − 2 .
2
Consider the Polyakov action and introduce the light-cone momenta conjugate to
the light-cone coordinates
∂L ∂L
P± = , Pi = (3.56)
∂ Ẋ ± ∂ Ẋ i
– 28 –
Express now the velocities via the corresponding momenta
1 1
Ẋ ± = (P∓ − T γ τ σ X 0± ) , Ẋ i = (−P i − T γ τ σ X 0i ) .
T γτ τ T γτ τ
Note that the light-cone indices are raised and lowered according to the rule
P− = −P + , P+ = −P − .
Now we can construct the phase-space Lagrangian in the light-cone coordinates:
L = Pi Ẋ i + P+ Ẋ + + P− Ẋ − + rest .
After explicit computation we find

1 ³ ´
L = Pi Ẋ i + P+ Ẋ + + P− Ẋ − + − 2P P
− + + Pi P i
+ T 2 0 0i
Xi X − 2T 2 0− 0+
X X
2T γ τ τ
γτ σ ³ ´
+ τ τ Pi X 0i + P− X 0− + P+ X 0+ . (3.57)
γ
The phase-space light-cone gauge consists in imposing the following two conditions
1. Closed string
p+ 1 +
X+ = τ, P + = const ≡ p . (3.58)
2πT 2π
2. Open string
p+ 1 +
X+ = τ, P + = const ≡ p . (3.59)
πT π
This gauge choice is done to remove completely the gauge degrees of freedom (recall
that in the conformal gauge we still had a gauge freedom left which was generated
by solutions of the conformal Killing equation).
We further consider the closed string case in detail. We will derive and solve all
the constraints followed from the Lagrangian (3.57) in several steps.
1. Varying the Lagrangian w.r.t. γ τ τ and imposing the light-cone gauge allows
one to solve for P − :
− π³ i 2 0 0i
´
P = + Pi P + T Xi X . (3.60)
p
Thus, equation of motion for γ τ τ allows one to determine P − .
– 29 –
2. Varying w.r.t. γ τ σ leads to determination of X 0− :
1 2π
X 0− = − Pi X 0i = + Pi X 0i . (3.61)
P− p
If we integrate the last equation over σ we obtain
Z
− − 2π 2π
X (2π) − X (0) = + dσPi X 0i . (3.62)
p 0
The closed string periodicity condition requires the fulfillment of the following
constraint
Z 2π
V= dσPi X 0i = 0 . (3.63)
0
This is the only constraint which remains unsolved and it is known as the level
matching condition. We will impose it on physical states of the theory.
3. Now we can determine the world-sheet metric γ αβ . Equation of motion for P+

is
δL P− γ τ σ 0+
0= = Ẋ + − + X ,
δP+ T γτ τ γτ τ
which with our gauge choice gives
γ τ τ = −1 .
4. Equation of motion for Ẋ − gives
d δL δL d p+ δL δL
0= − = − − =⇒ = 0,
dt δ Ẋ − δX − dt 2π δX − δX −
which gives µ ¶
γτ σ
∂σ P− = 0 =⇒ ∂σ γ τ σ = 0 .
γτ τ
For the closed string case this implies that γ τ σ = γ τ σ (τ ) is an arbitrary function
of τ . The presence of this function signals a residual symmetry. Indeed, on the
τσ
solutions of the level-matching constraint V = 0 the ratio γγ τ τ can be shifted by
an arbitrary function f (τ ) of τ without affecting the Lagrangian.
5. Varying w.r.t P− we find an evolution equation for X − :

δL P+ π
0= = Ẋ − − τ τ
= 0 =⇒ Ẋ − = +
(Pi P i + T 2 Xi0 X 0i ) .
δP− Tγ Tp
– 30 –
Thus, the variable X − is not physical and can be solved from the following two
equations we found above
π
Ẋ − = +
(Pi P i + T 2 Xi0 X 0i ) (3.64)
Tp
2π
X 0− = + Pi X 0i (3.65)
p
These two equations can be rewritten as
2πT
∂± X − = +
(∂± X i )2 . (3.66)
p
It is worth noting that two equations (3.64), (3.65) are compatible. Indeed, the
σ-derivative of the first equation must be equal the τ -derivative of the second
one. One can see that this is indeed so due to equations of motion for physical
fields.
Now we are ready to construct the gauge-fixed Lagrangian. Substituting solutions

of all the constraints and the gauge conditions into eq.(3.57) we obtain the density
1 + − p+ − ³ p+ 0− ´
L = Pi Ẋ i − p Ẋ − P − γ τ σ (τ ) Pi X 0i − X . (3.67)
2π 2πT 2π
Thus, the Lagrangian itself is
Z 2π Z 2π
+ −
L= dσL = −p ẋ + dσ Pi Ẋ i −H − γ τ σ (τ )V . (3.68)
0 0
| {z }
defines Poisson structure
Here x− denotes the zero (constant) mode of the variable X − and the Hamiltonian
is Z 2π ³
1 ´
i 2 0 0i
H= dσ Pi P + T Xi X .
2T 0
We also see that γ τ σ (τ ) plays the role of the Lagrangian multiplier to the level-
matching constraint V. Without loss of generality we will choose γ τ σ = 0 which
corresponds the conformal gauge condition discussed above.
From the gauge-fixed Lagrangian we conclude that our physical variables are (Pi , Xi ),
where i = 1, . . . , d−2, and also (x− , p+ ) and they have the following Poisson brackets
{X i (σ, τ ), X j (σ 0 , τ )} = {P i (σ, τ ), P j (σ 0 , τ )} = 0 ,
{X i (σ, τ ), P j (σ 0 , τ )} = δ ij δ(σ − σ 0 ) , (3.69)
+ −
{p , x } = 1 .
– 31 –
The physical Hamiltonian and the Poisson brackets look the same as the ones in the
conformal gauge, however, the important difference is that now they involve 2(d − 2)
physical fields only plus two additional degrees of freedom (x− , p+ ). First, from
eq.(3.65) we find that the zero mode of X − evolves as
1 H
ẋ− = +
H =⇒ x− (τ ) = x− + + τ .
p p
It is this τ -independent mode x− which is conjugate to p+ . Second, equations
2πT
∂± X − = (∂± X i )2 (3.70)
p+
can be solved for the longitudinal oscillators αn− , ᾱn− with n 6= 0 by substituting an
expansion
p− i X 1 ³ − −inσ− +
´
X − (τ, σ) = x− + τ+√ αn e + ᾱn− e−inσ . (3.71)
2πT
|{z} 4πT n6=0 n
H
p+
2πT
We find p− = p+
H and
√∞
πT X i
αn− = + α αi , n 6= 0, (3.72)
p m=−∞ n−m m
√ ∞
− πT X i
ᾱn = + ᾱ ᾱi , n 6= 0 . (3.73)
p m=−∞ n−m m
These formulae give a complete solution for X − .
Thus, the light-cone gauge allows for the explicit solution of the Virasoro constraints.
The variables (Pi , Xi ) are physical excitations while X ± , P + were removed by the
light-cone gauge choice and by solving the constraints. The variable P − plays the
role of the Hamiltonian for physical excitations! Equations of motion for physical
fields are the same as before
Ẍ i − X 00i = 0 , i = 1, . . . , d − 2.
The variables (Pi , Xi ), where i = 1, . . . , d − 2, are called transversal, while X ± , P ±

are longitudinal. The only constraint which we were not able to solve explicitly is
the level-matching constraint V = 0. It is easy to check that
{X i , V} = ∂σ X i , {P i , V} = ∂σ P i , (3.74)
i.e. V generates the rigid σ-rotations. We also have the evolution equations
1 i
{X i , H} = P = ∂τ X i , {P i , H} = T ∂σ2 X i = ∂τ P i . (3.75)
T
– 32 –
One can also see that the τ - and σ-flows generated by H and V respectively commute
with each other because
{H, V} = 0 .
Imposition of the light-cone gauge is possible because in the conformal gauge the
field X + satisfies the wave equation: ¤X + = 0. The corresponding solution for X +
is
p+ i X 1 ³ + −inσ− +
´
X + (τ, σ) = x+ + τ+√ αn e + ᾱn+ e−inσ . (3.76)
2πT 4πT n6=0 n
Our gauge choice (3.58) is taken to be compatible with the form (3.76). Effectively,
it means taken equal to zero all oscillators
αn+ = 0 = ᾱn+ , n 6= 0
and also x+ = 0 or
p+
αn+ = ᾱn+ = √ δn,0 .
4πT
One may wonder why one cannot completely remove X + , i.e. to choose a gauge
X + = 0. To understand this point one has to remember that the infinitesimal
conformal transformations are of the form
X −
X +
X→X+ an einσ ∂− X , X→X+ ān einσ ∂+ X ,
n n
where an , ān are arbitrary constants. In other words these transformations can be
written as
X → X + ξ − (σ − ) , X → X + ξ + (σ + ) ,
where ξ ± are arbitrary functions obeying only one requirement: they must be periodic
in σ. These functions can be used to remove all oscillator modes and the zero mode
x+ but they cannot remove
p+ p+
p+ τ = 2
(τ − σ) + 2
(τ + σ)
because the functions τ − σ and τ + σ are not periodic in σ.

String in the light-cone gauge can be treated in the standard framework of the Hamiltonian reduction. The Hamiltonian
H = L0 + L̄0 is invariant under the symmetry algebra on the surface Lm = 0 = L̄m :
{H, Lm } = imLm , {H, L̄m } = imL̄m .
The symmetry algebra itself is
{Ln , Lm } = −i(m − n)Ln+m , {L̄n , L̄m } = −i(m − n)L̄n+m .
– 33 –
Orbits of the Virasoro algebra
+
Gauge condition for X
Constrained surface
_
L m =L m =0
Phase space (P,X)
Fig. 2. The physical phase space is obtained by solving the Virasoro con-
straints Lm = 0 = L̄m and reducing the action of the Virasoro algebra on
the constrained surface by imposing the light-cone gauge.
One would like to reduce the dynamics of the system over the action of the symmetry algebra. In the framework of the
Hamiltonian reduction conditions
Lm = 0 = L̄m
correspond to fixing the moment map. The reduced phase space P is defined as a quotient space
solutions of Lm = 0 = L̄m
P = ,
isotropy subalgebra
where the isotropy subalgebra is a subalgebra of the Virasoro algebra which leaves the surface Lm = 0 = L̄m invariant. In our present
situation this subalgebra coincides with the algebra itself and, therefore,
solutions of Lm = 0 = L̄m
P = .
action of Virasoro
The action of the Virasoro algebra is factored out by imposing the light-cone gauge, which simultaneously leads to solving the Virasoro
constraints. The transversal coordinates introduced above provide the description of the reduced phase space.
Mass of the string in the light-cone gauge is computed as follows (recall that mass
is a quadratic Casimir of the Poincaré group). Since we have found that p− = 2πT p+
H
we get for the mass
M 2 = −pµ pµ = −(pi )2 + 2p+ p− = −(pi )2 + 4πT H . (3.77)
The physical Hamiltonian is

∞ ∞
1 X ¡ i i i i
¢ (pi )2 X ¡ i i ¢
H= αn α−n + ᾱn ᾱ−n = + αn α−n + ᾱni ᾱ−n
i
. (3.78)
2 n=−∞ 4πT n=1
Thus, we obtain
∞
X ∞
¡ ¢ 2 X¡ i i ¢
M 2 = 4πT αni α−n
i
+ ᾱni ᾱ−n
i
= 0 αn α−n + ᾱni ᾱ−n
i
. (3.79)
n=1
α n=1
This clearly shows positivity of M 2 , a property which was not obvious in the confor-
mal gauge.
– 34 –
Finally, we can write the level-matching condition in terms of transversal oscillators.
We find
1 X³ i i ´
V= ᾱn ᾱ−n − αni α−n
i
= 0. (3.80)
2T n6=0
Thus, the level-matching condition tells that the left- and right-moving oscillators
contribute the same amount of energy.
3.4.2 Poisson structure of the light-cone theory

Using the Poisson brackets of physical fields we can now establish the Poisson rela-
tions between all quantities of interest. We first summarize the basic Poisson relations
for the closed string case
{, } p+ p− pj xj x− j
αm −
αm
p+ 0 0 0 0 1 0 0
i −
p− 0 0 0 − pp+ − pp+ 2πiT
p+
j
mαm 2πiT
p+
−
mαm
pi 0 0 0 − δ ij 0 0 0
pi δ ij δm αim
xi 0 p+
δ ij 0 0 4πT p+
p− α−
x− −1 p+
0 0 0 0 m
+
ij
√ p
2πiT
αni 0 − p+
nαni 0 − δ4πTδn 0 − inδ ij δn+m − i p4πT i
+ nαn+m
√
i
α−
αn− 0 − 2πiT
p+
nαn− 0 − αp+n − n
p+
i 4πT
p+
i
mαn+m {αn− , αm
−
}
Tab. 1. Poisson brackets of the light-cone modes. The variable p− is

essentially the Hamiltonian: p− = 2πT pH+ . The brackets involving ᾱ
variables are the same.
These relations are easy to derive. For instance,
H − 1 H p−
{p− , x− } = {2πT , x } = −2πT H {p +
, x −
} = −2πT = − .
p+ (p+ )2 (p+ )2 p+
Also, one has to remember that the variable αn− contains the zero mode α0i
√
πT pi αi
αn− = + 2α0i αni + . . . = +n + . . .
p p
and, therefore,
1 i j j 1
{xi , αn− } = +
{x , p αn } = + αni .
p p
−
The most complicated bracket is {αm , αn− }.
– 35 –
Let us outline the computation of this bracket.
√ √
n pi αi πT X pj αjn πT X o
− − m i i j j
{αm , αn } = + αm−k αk , + αn−l αl =
p+ p+ k6=m,0
p+ p+ l6=n,0
√ √
n pi αi pj αjn o πT n pi αim X o πT n pi αin X o πT n X X o
m j j j j i i j j
, + , αn−l αl − , αm−l αl + αm−k αk , αn−l αl
p+ p+ p+ p+ l6=n,0
p+ p+ l6=m,0
(p+ )2 k6=m,0 l6=n,0
√ 0 1
pi pi πT i i 2πT X i i
−im δn+m −2i(m − n) p αn+m + @−i(m − n) αn+m−k αk A . (3.81)
(p+ )2 (p+ )2 (p+ )2 k6=m+n,0
| {z }| {z }| {z }
first bracket second and third brackets
forth bracket
Thus, we are getting
0 1
∞
− − pi pi 2πT X i i
{αm , αn } = −im δn+m + @−i(m − n) αn+m−k αk A ,
(p+ )2 (p+ )2 k=−∞
i
where due to αi0 = √ p we combined the second and a third terms into one sum. If m + n 6= 0 then the first term vanishes and we
4πT
can rewrite the last formula as
√
− − 4πT −
{αm , αn } = −i(m − n)αm+n .
p+
If n = −m then in eq.(3.81) contribution from the second and the third term vanishes and we get
0 1 √ 0√ !2 1
− − pi pi 2πT X i i 4πT πT h pi X i i
i
{αm , α−m } = −im + @−2im α−k αk A ≡ −2im @ √ + αk α−k A .
(p+ )2 (p+ )2 k6=0
p+ p+ 4πT k6=0
Therefore,
√ 0√ 1
∞
X
− − 4πT πT i i
{αm , α−m } = −2im @ αk α−k A .
p+ p+ k=−∞
It is therefore natural to define √ ∞

X
− πT i i
α0 ≡ αk α−k .
p+ k=−∞
With this definition we obtained a universal formula (valid for all indices m and n):
√
− − 4πT −
{αm , αn } = −i(m − n)αm+n .
p+
Also we conclude that with this definition √

− − −
p = πT (α0 + ᾱ0 ) .
Thus, one finds the following result

√
− 4πT ¡ ¢
{αm , αn− } = +
−
−i(m − n)αm+n .
p
If we introduce
p+
Ln = √ αn− .
4πT
we therefore find
{Ln , Lm } = −i(n − m)Ln+m
which is the classical Virasoro algebra! Thus, in the light-cone gauge the Virasoro
algebra is carried over by the longitudinal oscillators αn− .
– 36 –
3.4.3 Lorentz symmetry
The light-cone gauge manifestly breaks the d-dimensional Lorentz invariance of the
theory. We have the generators of the Lorentz algebra which in terms of transversal
oscillators become realized non-linearly. For instance, the generator
X∞ X∞
1¡ i − ¢ 1¡ i − ¢
J i− = xi p− − x− pi −i −
α−n αn − α−n αni − i −
ᾱ−n ᾱn − ᾱ−n ᾱni
| {z } n n
n=1 n=1
ì−
is not anymore quadratic in oscillators because p− and α− are non-trivial function

of the transversal oscillators.
In spite of rather non-linear realization one can check that all the Poisson re-
lations involving J i− are still satisfied (this is also consequence of the fact that
{Lm , J µν } = 0 = {L̄m , J µν } , which means that the Poisson bracket of J µν admits a
reduction on the constraint surface Lm = 0 = L̄m ). In particular, from the Poincaré
algebra we must have
{J i− , J j− } = 0 .
Let us check this explicitly. First we have
i− j− i − − i j − − j
{` ,` } = {x p − x p ,x p −x p }=
i − j − − i j − i − − j − i − j
= {x p , x p } − {x p , x p } − {x p , x p } + {x p , x p } .
By using the Poisson brackets from Table 1 we obtain
i− j− pi − j pj − i i j p− − − ij j i p− − − ij
{` ,` } = p x − p x −p x +x p δ +p x −x p δ .
p+ p+ p+ p+
| {z }| {z }| {z }
first bracket second bracket third bracket
and the forth bracket vanishes. Thus,
i− j−
{` ,` } = 0.
Consider the Poisson bracket of the internal spin components
i− j−
X 1 i − j −
{S ,S }=− {α−n αn , α−m αm } =
n,m6=0
nm
X 1 i j − − i − − j − j i − − − i j
=− {α−n , α−m }αn αm + {α−n , αm }αn α−m + {αn , α−m }α−n αm + {αn , αm }α−n α−m =
n,m6=0
nm
− √
ij
X α−
n α−n 4πT X 1 i j − 1 i j − n−m i j −
= iδ +i − α−n+m α−m αn + α−n αn−m αm + α−n α−m αm+n .
n6=0
n p+ m,n6=0
m n nm
Here the first first sum is zero (proved by changing n for −n). In the second and the third summonds one makes the change of
summation indices, n → n + m and m → m + n respectively. After this change these terms cancel exactly against the last one.
However, there are terms which are still left, these are the terms in the second and third summonds containing zero modes, i.e. terms
for which −n + m = 0 and also the term of the last summond for which m + n = 0 . Thus,
√
i− j− i X 1 i j j i − 4πT −
X 1 i j
{S ,S }=− p α−n − p α−n αn + 2i α0 αn α−n . (3.82)
p+ n6=0
n p+ n6=0
n
Analogously, we will get the following contribution from the left-moving modes Then we compute
0 1
B C
i− j−
X 1 i − − i j −
X 1 B B j − αn
i
i 2πiT j − 2πiT j − i j α− C
C
{` , S } = −i {x p − x p , α−n αn } = −i Bα−n p +x − nα−n αn + nα−n αn − p α−n n C .
n | {z } | {z } n B p+ p+ p+ p+ C
n6=0 n6=0 @| {z } | {z }A
A B
from A from B
– 37 –
Thus, we arrive at
i− j− i− j− p− X 1 i j i X 1 i j j i −
{` ,S } + {S ,` } = −2i αn α−n + p α−n − p α−n αn . (3.83)
p+ n6=0 n p+ n6=0 n
Now summing up equations (3.82) and (3.83) we obtain
√
i− j− i− j− i− j− 4 πT − α− −
0 + ᾱ0
X 1 i j
{` ,S } + {S ,` } + {S ,S }=i α0 − αn α−n . (3.84)
p+ 2 n6=0
n
At first glance this expression is non-zero and it can not be compensated by the contribution of left-moving modes
√
i− j− i− j− i− j− 4 πT − α− −
0 + ᾱ0
X 1 i j
{` , S̄ } + {S̄ ,` } + {S̄ , S̄ }=i ᾱ0 − ᾱn ᾱ−n . (3.85)
p+ 2 n6=0
n
However, we have to invoke the level-matching constraint which simply tells that α− −
0 = ᾱ0 and makes both eqs.(3.84) and (3.85)
separately vanish. Thus, we have indeed shown that the most non-trivial relation {J i− , J j− } = 0 is indeed satisfied.
4. Quantization of bosonic string

4.1 Remarks on canonical quantization
According to the standard principles of quantum mechanics canonical quantization
consists in replacing the Poisson brackets of the fundamental phase space variables
by commutators
1
{ , } → [ , ],
i~
where ~ is the Plank constant. Thus, we consider now X(σ, τ ) and P (σ, τ ) as the
quantum mechanical operators which obey the following commutation relations7
[X µ (σ, τ ), X ν (σ 0 , τ )] = [P µ (σ, τ ), P ν (σ 0 , τ )] = 0 ,
[X µ (σ, τ ), P ν (σ 0 , τ )] = iη µν δ(σ − σ 0 ) , (4.1)
These commutation relations induce the commutation relations on the Fourier coef-
ficients
µ
[αm , αnν ] = [ᾱm
µ
, ᾱnν ] = ~mδm+n η µν ,
µ
[αm , ᾱnν ] = 0 , (4.2)
µ ν µν
[x , p ] = i~η .
For the case of open string the modes ᾱn are absent. In what follows we will work
in units in which ~ = 1, so that the commutation relations read as
µ
[αm µ
, αnν ] = [ᾱm , ᾱnν ] = mδm+n η µν ,
µ
[αm , ᾱnν ] = 0 , (4.3)
µ ν µν
[x , p ] = iη .
7
Here the indices µ, ν run from 0 to d − 1.
– 38 –
To restore ~, one has to simply rescale the modes as α → √1 α, ᾱ → √1 ᾱ.
~ ~
Let us take m > 0 and rescale am = √1m αm . Then using the hermiticity property
the commutation relations can be rewritten as
[aµn , a†ν µν
m ] = η δn,m
[āµn , ā†ν µν
m ] = η δn,m
which are the standard commutation relations of two infinite sets of independent
quantum harmonic oscillators.
Introduce the number operator for m’s mode
µ
Nm =: αm α−m,µ :
Then we see that for m > 0
[Nm , αm ] = −mαm ,
[Nm , α−m ] = mα−m .
From here we conclude that
• Modes with m > 0 should be identified with the lowering operators
• Modes with m < 0 should be identified with the raising operators
Construction of the representation of the canonical commutation relations is com-

pleted by introducing the ground state which satisfies the following properties
µ
αm |pν i = 0 ,
p̂µ |pν i = pµ |pν i . (4.4)
The whole infinite-dimensional (Hilbert) space of states is obtained by acting on the

ground state with creation operators.
This construction brings us to the major problem of canonical quantization.
Consider for m positive the following commutator
0 0 0 0 †
[αm , α−m ] = [αm , (αm ) ] = mη 00 = −m .
Thus, we induce from here (let even a ground state carries zero momentum pµ )
0 0 † 0 0 † 0 †
h0|[αm , (αm ) ]|0i = h0|αm (αm ) |0i = ||(αm ) |0i||2 = −m < 0 .
Thus, Minkowskian type of the target-space metric leads to the existence in the
Hilbert space the states with negative norm. States with negative norm are some-
times called “ghosts” and they do not allow for probability interpretation of the
corresponding quantum-mechanical system.
– 39 –
One can correctly anticipate that the problem with negative norm states arose
because we did not take into account the Virasoro constraints. The covariant ap-
proach to quantization consists in defining the subspace of physical states in the
original Hilbert space which obey the Virasoro constraints. One can further show
that in a special dimension of space-time (d = 26) the negative norm states decouple
from the physical Hilbert space.
In the classical theory we have the constraints Lm = 0 = L̄m . However, in
quantum theory expressions for Lm and L̄m are quadratic in oscillators and might
involve operators (quantum oscillators) which do not commute with each other! From
all Lm a constraint which suffers from ordering ambiguity is L0 as
∞
1 X µ
L0 = α α−n,µ (4.5)
2 n=−∞ n
µ
and oscillators αnµ and α−n do not commute with each other. The standard way
to deal with this ambiguity in quantum field theory is to use the normal ordering
prescription
∞
1 X µ
Lm = : αm−n αn,µ : . (4.6)
2 n=−∞
The normal ordering prescription means that
: αm1 . . . αmk := αn1 . . . αnp αs1 . . . αsr

| {z } | {z }
all creation all annihilation
in the operators are ordered in such a fashion that all annihilation operators are
put on the right from all the creation operators. The order of the creation (or
annihilation) operators between themselves does not matter because these operators
commute between themselves and therefore their expression does not have ordering
ambiguity. In particular, for L0 we have
∞
1 2 X µ
L0 = α0 + α−n αn,µ − a , (4.7)
2 n=1
where we include a so far unknown normal ordering constant a. As to the zero modes,
the normal-ordering prescription here is
: pµ xν := xν pµ .
Since the ground state obeys p̂µ |pi = pµ |pi it can be regarded as the usual quantum-
mechanical eigenstate of the momentum operator p̂µ = −i ∂x∂ µ which is in the mo-
mentum representation has the form of the plane-wave
µx
|pi ≡ eip µ
|0i ,
– 40 –
where on the r.h.s. pµ is not an operator and |0i denotes the zero-momentum ground
state. Plane-waves are not square-integrable functions but they form a basis of a
generalized Hilbert space and they are assumed to be normalized as
hp|p0 i = δ(p − p0 ) .
4.1.1 Virasoro algebra

Let us now investigate the algebra of the operators Lm in the quantum case. We
first assume that the normal ordering constant a = 0. We start by computing
∞
X ∞ ³
X ´
µ µ µ µ
[αm , Ln ] = 12 [αm , : αpν αn−p,ν :] = 12 mδm+p αn−p + mαpµ δm+n−p = mαn+m .
p=−∞ p=−∞
Then we have
∞
1 X
[Lm , Ln ] = [: αµ αm−p,µ :, Ln ] .
2 p=−∞ p
We write down the normal ordering explicitly (to simplify the notation we write the
Lorentz summation index on the same level)
0 ∞
1 X µ µ 1X µ
[Lm , Ln ] = [α α , Ln ] + [α α µ , Ln ] =
2 p=−∞ p m−p 2 p=1 m−p p
0
1 X
= µ
pαµ αm−p µ
+(m − p)αpµ αm−p+n
2 p=−∞ | p+n
{z }
p=q−n
∞
X
1 µ µ µ
+ (m − p)αm−p+n αpµ + pαm−p αn+p .
2 p=1
| {z }
p=q−n
In the underbraced terms we make a change of summation index p = q − n and get
1³ X
0 Xn
µ µ
[Lm , Ln ] = (m − p)αpµ αm+n−p + (q − n)αqµ αm+n−q
2 p=−∞ q=−∞
∞
X +∞
X ´
µ µ
+ (m − p)αm+n−p αpµ + (q − n)αm+n−q αqµ .
p=1 q=n+1
Without loss of generality we assume that n > 0. then we have
1³ X
0 Xn
µ µ
[Lm , Ln ] = (m − n)αpµ αm+n−p + (q − n) αqµ αm+n−q
2 p=−∞ q=1
| {z }
not ordered!
∞
X n
X ´
µ µ
+ (m − n)αm+n−p αpµ + (m − q)αm+n−q αqµ .
p=n+1 q=1
– 41 –
µ µ µ
Using αqµ αm+n−q = αm+n−q αqµ + qδm+n δµµ = αm+n−q αqµ + qdδm+n , where d is the
dimension of the Minkowskian space-time where string propagates. Thus, the algebra
relation is
∞ n
1 X µ d X
[Lm , Ln ] = (m − n) : αpµ αm+n−p : + δn+m (q 2 − nq) .
2 p=−∞ 2 q=1
Since
n
X n
X
2 1 1
q = n(n + 1)(2n + 1) , q = n(n + 1) .
q=1
6 q=1
2
one finds the final result

d
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n
12
which is the famous Virasoro algebra. We see that it is different from the classical
Virasoro (Wit) algebra by the presence of the central term.
In a more general setting the algebra is written as

c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n ,
12
where the constant term c is known as the central charge.
If we introduce a normal ordering constant a by shifting the definition of Lm as

Lm → Lm − δm,0 then the linear term in m in the central term changes
³c ¡ c¢ ´
[Lm , Ln ] = (m − n)Lm+n + m3 + 2a − m δm+n .
12 12
We see that the central term has an invariant meaning and cannot be removed for
all Lm by adjusting the normal ordering constant a.
Finally, we comment on the relation to semiclassics. If we restore the Plank the

algebra relations take the form
c
[Lm , Ln ] = ~(m − n)Lm+n + ~2 m(m2 − 1)δm+n .
12
We see that one can define the Poisson bracket
1
{Lm , Ln } = lim [Lm , Ln ] = −i(m − n)Lm+n ,
~→0 i~
which coincides with the Wit algebra. The central term obviously vanishes in the
semi-classical limit.
– 42 –
4.1.2 Virasoro constraints in quantum theory
At first sight it seems that the natural analog of the classical equations Ln = 0 = L̄n
in quantum theory is to require that a physical state should be annihilated by all
Virasoro generators:
Ln |Φi = 0 = L̄n |Φi , n ∈ Z .
Due to the normal-ordering ambiguity in the definition of L0 in quantum theory the

classical conditions L0 = 0 = L̄0 are replaced now by
(L0 − a)|Φi = 0 , (L̄0 − ā)|Φi = 0 , (4.8)
where in fact the normal-orderings constants a and ā must be equal to each other8 and
L0 , L̄0 are understood as the normal-ordered generators. It is easy to see, however,
that eqs.(4.8) cannot be consistently imposed for all m. Indeed, if eqs.(4.8) would
be satisfied for all m we would have
d
[Ln , L−n ]|Φi = 2nL0 |Φi + n(n2 − 1)|Φi ,
12
i.e. µ ¶
d 2
0 = 2na + n(n − 1) |Φi for any n .
12
This is obviously not possible to satisfy unless |Φi = 0. The physical reason for
impossibility to impose in quantum theory the same set of constraints as in the
classical one is an anomaly. Because of the anomaly term the first-class Virasoro
constrains of the classical theory turn upon quantization into the constraints of the
second class!
From the experience with the quantum electrodynamics one can try to impose only
“half” of the constraints, i.e.
(L0 − a)|Φi = 0 ,
Ln |Φi = 0 , n > 0. (4.9)
The conjugate state then obeys hΦ|L−n = 0 for n > 0 and we see that hΦ|Ln |Φi for
all n 6= 0, i.e. expectation values of Ln vanish for all nonnegative n.
Let us recall that the mass operator is obtained from the constraint L0 − a = 0, We
have
X∞
2 2 µ
M = −p = 4πT (−a + N ) , N= α−n αn,µ .
n=1
8
This follows from the constraint (L0 − L̄0 )|Φi = 0.
– 43 –
Here N is the number operator. It turns out that all eigenvalues of the number
operator N are non-negative. Indeed,
∞ ³
X d−1
X ´
0
N= − α−n αn0 + i
α−n αni .
n=1 i=1
We see the “-” sign coming from the time-like oscillators. However, the time-like
oscillators themselves provide only non-negative contribution to N , because for any
m>0 ∞ ∞
X X
[N, a0−m ] = − 0
[α−n αn0 , a0−m ] = − 0
α−n 0
[αn0 , a0−m ] = mα−m
n=1 n=1
since commutator of two time-like oscillators contributes with the negative sign.
Thus, time-like creation operators contribute positively to N .
Virasoro primaries, descendents and physical states

Let us introduce the following useful definitions.
1. States which are annihilated by all positively moded Virasoro operators and
are eigenstates of the operator L0 with an eigenvalue a are called Virasoro
primaries. Number a is called a weight of the Virasoro primary.
2. A Virasoro descendent of a given primary is a state that can be written as a

finite linear combination of products of negatively moded Virasoro operators
acting on the primary state.
3. A state which is both primary and descendent is called a null state.
If |Φi is a primary state then L−1 |Φi is its descendent. If N |Φi = NΦ |Φi then
N L−1 |Φi = (NΦ + 1)L−1 |Φi.
There are two basis descendents with the number NΦ + 2, namely L−2 |Φi and
L−1 L−1 |Φi. The counting of descendents changes at NΦ + 3. Here the candidate
descendents are
L−3 |Φi , L−2 L−1 |Φi , L−1 L−2 |Φi , L3−1 |Φi .
The second and the third states are not identical because the Virasoro operators do
not commute. However, due to the Virasoro algebra there is one relation between
the above states
L−1 L−2 = [L−1 , L−2 ] + L−2 L−1 = L−3 + L−2 L−1 .
Thus, there are only three descendents with number NΦ + 3.
– 44 –
In general, for any fixed level NΦ + n, one can choose an independent basis of
descendents of the form
k
X
L−n1 L−n2 · · · L−nk |Φi , where n1 ≥ n2 ≥ . . . ≥ nk and ni = n .(4.10)
i=1
It is a conventional ordering of the Virasoro operators. Note that the number of

descendents in this basis is equal to a number of partitions of integer n.
For any given primary, there can be a linear combination of descendents (4.10)
which vanishes. The vanishing of such combinations happens not due to the Virasoro
algebra relations but rather due to specific properties of the primary state |Φi. For
instance, one can realize that for the zero-momentum ground (primary) state |0i the
descendent L−1 |0i vanishes identically.
An important property of the descendents is that they are all orthogonal to any
primary. Indeed, a descendent |desi can be written as L−ni |χi for some ni > 0 and
some state |χi. Then for any primary state
hdes|primei = hχ|Lni |primei = 0 ,
since a primary |primei is annihilated by all positively moded oscillators.

Null states, which are both primary and descendents, correspond to pure gauge
degrees of freedom. Any null state has a vanishing norm and it is orthogonal to any
primary and to any descendent. If we alter a primary state by adding a null state,
then the new primary state has the same inner products with all primary states as the
original one. Adding null states to primaries cannot change any physical expectation
value. This motivates the following definition of a physical state:
A physical state is an equivalence class of a primary

state with the weight a = 1 modulo the null states.
The choice a = 1 will be motivated by studying the quantization of strings in the

physical (light-cone) gauge. Note that here we talk about equivalent classes precisely
because of the ambiguity created by the null states. Primary states which differ by a
null state are physically indistinguishable. In the next section studying the spectrum
of open string we will see that the null states are indeed responsible for the gauge
degrees of freedom.
4.1.3 The spectrum

The classical strings cannot provide a reasonable particle physics because the masses
of string states take continuous values. Only the ground state is massless in the
classical open string theory but because it carries no spin we are not able to identify
it with photon. Quantization procedure – this is what alters the nature of the classical
– 45 –
string spectrum and makes it discrete. Simultaneously, states to be identified with
photon emerge in the quantum spectrum because of the downward shift of the M 2
producing thereby the massless states with a proper spin labels.
Consider open strings. The Hamiltonian is
H = L0 − 1 = α0 p2 + N − 1 .
The basis vectors of the Hilbert space are

∞ Y
Y 25
|ψi = (αnµ† )λn,µ |pi ,
n=1 µ=0
are non-negative integers. Generic state |ψi is not physical. A physical state is a
state which obeys the Virasoro constraints (4.9) and is not a descendent.
Let us look for some examples of physical states. The first one is the ground state
|pi. The only non-trivial constraint is
(L0 − a)|pi = (α0 p2 − a)|pi = 0 .
Since p2 = −M 2 we get that the on-shell condition for this state is M 2 = − αa0 . Later
on studying the light-cone quantization we find that the normal-ordering constant
a must be equal to one. Thus, the mass-squared of the ground state is negative:
M 2 = − α10 . The corresponding hypothetic particle moving faster than light is called
tachyon.
µ
The next state to consider is ζµ α−1 |pi. We have
µ µ µ
(L0 − 1)ζµ α−1 |pi = (α0 p2 + N − 1)ζµ α−1 |pi = α0 p2 ζµ α−1 |pi = 0
from which we deduce the on-shell condition p2 = 0, i.e. the corresponding particle
is massless. Further condition gives
µ µ
√
L1 ζµ α−1 |pi = (α0 α1 + α−1 α2 + · · · )ζµ α−1 |pi = 2α0 ζµ pµ |pi = 0 .
Thus, for a physical state the momentum pµ and the polarization vector ζ µ must be
related as ζµ pµ = 0 which is nothing else as the Lorentz gauge condition. All higher
Virasoro modes Ln , n ≥ 2 are automatically annihilate the state. We, however, have
not described the physical state completely. The massless vector particle which is
photon must have d − 2 independent polarizations while the Lorentz gauge lives d − 1
polarizations only. We should now recall that physical states are defined modulo the
null states.
Consider a state (κ is any constant)
κ µ
|di = √ L−1 |pi = κpµ α−1 |pi , p2 = 0 .
2α 0
– 46 –
This state has the same form as before with ζµ = κpµ (longitudinally polarized
photons). It is physical, because ζµ pµ = κp2 = 0. On the other hand, it is null as it
appears to be a descendent of L−1 . Thus, the states
ζµ and ζµ + κpµ
should be identified. The reason for this identification can be understood as follows.
If ζµ pµ = 0 then (ζµ + κpµ )pµ = 0 as well, because p2 = 0. As the result, this
identification reduces the number of independent polarizations to d − 2, as it should
be for a massless photon. Indeed, for any vector ζµ obeying ζµ pµ = 0, one can use
the shift freedom ζµ → ζµ + κpµ to put, e.g., one of the components of ζµ to zero.
4.1.4 Propagators
Here we introduce the concept of propagator or, equivalently, the two-point Green
function.
Consider first the right-moving fields of the closed string. Their propagator is
defined as
¡ ¢
hXR (τ, σ)XR (τ 0 , σ 0 )i = T XR (τ, σ)XR (τ 0 , σ 0 ) − : XR (τ, σ)XR (τ 0 , σ 0 ) : , (4.11)
where T means time-ordering prescription. Thus,
½
¡ 0 0
¢ XR (τ, σ)XR (τ 0 , σ 0 ) , for τ > τ 0 ,
T XR (τ, σ)XR (τ , σ ) =
XR (τ 0 , σ 0 )XR (τ, σ) , for τ < τ 0 .
For τ > τ 0 we have
0 0
hXR (τ, σ)XR (τ , σ )i =
1 µ pµ i X 1 µ −in(τ −σ) 1 ν pν 0 0 i X 1 µ −im(τ 0 −σ 0 )
= x + (τ − σ) + √ αn e x + (τ − σ ) + √ αm e
2 4πT 4πT n6=0 n 2 4πT 4πT m6=0 m
1 µ pµ i X 1 µ −in(τ −σ) 1 ν pν 0 0 i X 1 µ −im(τ 0 −σ 0 )

−: x + (τ − σ) + √ αn e x + (τ − σ ) + √ αm e : .
2 4πT 4πT n6=0 n 2 4πT 4πT m6=0 m
Most of the terms cancelled and we are left with
0 0 1 µ ν 0 0 1 µ ν
hXR (τ, σ)XR (τ , σ )i = x p (τ − σ ) + p x (τ − σ)
8πT 8πT
1
µ ν 0 0 1 µ ν
− : x p : (τ − σ ) − : p x : (τ − σ)
8πT 8πT
1 X X 1 µ ν µ ν −in(τ −σ) −im(τ 0 −σ 0 )
− (αn αm − : αn αm :)e e .
4πT n6=0 m6=0 nm
Thus,
0 0 1 µ ν 1 X X 1 µ ν −in(τ −σ) −im(τ 0 −σ 0 )

hXR (τ, σ)XR (τ , σ )i = [p , x ](τ − σ) − [αn , αm ]e e .
8πT 4πT n>0 m<0 nm
Substituting the commutators we obtain
0 0 i µν η µν X 1 im(τ −σ) −im(τ 0 −σ 0 )

hXR (τ, σ)XR (τ , σ )i = − η (τ − σ) − e e
8πT 4πT m<0 m
0 0 0
1m
i µν η µν X 1 ei(τ −σ )
= − η (τ − σ) + A @
8πT 4πT m>0 m ei(τ −σ)
0 0 0
1
i µν η µν ei(τ −σ )
= − η (τ − σ) − ln @1 − A .
8πT 4πT ei(τ −σ)
– 47 –
Thus, for τ > τ 0 one obtains the propagators9
η µν η µν
hXR (τ, σ)XR (τ 0 , σ 0 )i = ln z − ln(z − z 0 ) ,
8πT 4πT
η µν η µν
hXL (τ, σ)XL (τ 0 , σ 0 )i = ln z̄ − ln(z̄ − z̄ 0 ) ,
8πT 4πT
η µν
hXR (τ, σ)XL (τ 0 , σ 0 )i = − ln z ,
8πT
where we made an identification
z = ei(τ −σ) , z̄ = ei(τ +σ) .
Computation for the case of open string is similar. For τ > τ 0 we have
∞
0 η µν
0 1 X 1 −inτ +inτ 0
hX(τ, σ)X(τ , σ )i = −i τ+ e cos nσ cos nσ 0 .
πT πT n=1 n
Performing the sum one finds

η µν h ³ 0 0
´ ³ 0 0
´
hX(τ, σ)X(τ 0 , σ 0 )i = − log eiτ − e−i(σ−σ ) eiτ + log eiτ − ei(σ−σ ) eiτ
4πT ³ ´ ³ í
0 0 0 0
+ log eiτ − e−i(σ+σ ) eiτ + log eiτ − ei(σ+σ ) eiτ .
4.1.5 Vertex operators. Tachyon scattering amplitude

Here we approach for the first time the question about string interactions. It is
important to realize that the situation here is different to what one usually accounters
in QFT. The interaction of strings cannot be introduced by adding non-linear terms
to the string Lagrangian; in the latter case one would obtain non-linear interacting
theory but still of a single string.
Out
Out
split string
on a mass−shell
Insertion
V of a local operator
emmiting a particle
In In
Fig. 3. Open (closed) strings interact by means of joining and splitting.

Emission of a point particle on mass-shell is represented by insertion of a
local vertex operator.
9
Rigorous justification of these formulae requires an introduction of an IR regularization, because
correlation functions of the massless field in two-dimensions suffer from IR divergencies.
– 48 –
String interactions are introduced by allowing a topology of a wold-sheet to change.
If a split string is on-shell it can be thought as an infinite collection of particles
which form its spectrum. In the limiting case of single emitted particle the process
of scattering can be viewed as application to an initial state |Ini of a local vertex
operator V ≡ V (τ, σ) which depends on the emitted state and transforms |Ini into
the outgoing state |Outi:
|Outi = V |Ini .
Thus, to any physical state |Φi from the spectrum one can put in correspondence a
local vertex operator VΦ .
Conformal operators
Operator A(τ ) is called conformal with the conformal dimension ∆ if10
³ dτ ´∆
0 0
A (τ ) = A(τ ) .
dτ 0
Under infinitezimal variation τ → τ 0 = τ + ²(τ ) one gets
dA d²
δA(τ ) = −² − ∆A
dτ dτ
With ² = −ieimτ
[Lm , A(τ )] = eimτ (−i∂τ + m∆)A(τ )
P
If the operator A is expandable as A(τ ) = An e−inτ , for its Fourier modes the last
relation implies
[Lm , An ] = (m(∆ − 1) − n)An+m
Thus, if A has conformal weight ∆ = 1 then it’s zero mode commutes with all Lm .
Therefore, A0 maps physical states into physical states:
|Φ0 i = A0 |Φi .
An alternative way to understand this is to notice that for ∆ = 1 we have

∂ ³ imτ ´
[Lm , A(τ )] = eimτ (−i∂τ + m∆)|∆=1 A(τ ) = −i e A(τ ) ,
∂τ
R
i.e. the r.h.s. is the total derivative. Thus, A0 = dτ A(τ ) will transform as
Z
∂ ³ imτ ´
[Lm , A0 ] = −i dτ e A(τ ) ,
∂τ
10
This transformation law can be written as
A(τ 0 )(dτ 0 )∆ = A(τ )(dτ )∆ .
For ∆ = 1 this implies that A(τ )dτ is a one-form.
– 49 –
where the expression on the r.h.s. is specified by the periodicity/boundary properties
of A(τ ).
Explicit example of a vertex operator
Consider the following operator acting in the Hilbert space of open string
µ µ
P∞ kµ α−n µ P∞ kµ αn −inτ
√1 einτ cos nσ ikµ (xµ + p τ ) − √1 e cos nσ
V (k, τ, σ) = e πT n=1 n e πT πT e n=1 n .
At σ = 0 this simplifies to
µ µ
P∞ kµ α−n µ P∞ kµ αn −inτ
√1 einτ p
ikµ (xµ + πT τ) − √1 e
V (k, τ ) = e| πT n=1
{z
n
} e| {z } |e
πT n=1
{z
n
}.
V− V0 V+
This operator is “almost” normal-ordered and can be concisely written as
V (k, τ ) = V− V0 V+
Here only zero modes entering V0 are not normal-ordered. Indeed, according to our
conventions : pµ xν := xν pµ , the operator pµ should be always on the right from xµ
which is not the case for V0 .
The normal-ordering of V0 can be achieved with the help of the Baker-Campbell-
Hausdorff formula11
1
eA eB = eA+B+ 2 [A,B]
Thus, we have
µ kµ pµ µ kµ pµ µ
µ +i kµ p τ − τ kµ kν [xµ ,pν ]
: eikµ x ei πT
τ
:= eikµ x ei πT
τ
= eikµ x πT 2πT
Recalling that [xµ , pν ] = iη µν we therefore find that

µ kµ pµ
µ +i kµ p τ 0 2τ µ 0 2τ
eikµ x πT = eiα k : eikµ x ei πT
τ
:= eiα k : V0 :
Thus, we obtain completely normal-ordered vertex operator

0 2
V = eiα k τ V− : V0 : V+
We would like to investigate the transformation properties of this operator under

conformal transformations. To this end we need to compute by using the Leibnitz
rule
0 2
e−iα k τ [Lm , V ] = [Lm , V− ] : V0 : V+ + V− [Lm , : V0 :]V+ + V− : V0 : [Lm , V+ ] .
11
Suppose you forgot an exact coefficient in front of the commutator term. Write the operator
identity in the form eA eB = eA+B+α[A,B] with the forgotten coefficient α. Rescale A and B with a
2
small parameter ² to get e²A e²B = e²A+²B+² α[A,B] . Now expand both sides up to order ²2 keeping
the order of operators. You will find that fulfilment of the operator relation at order ²2 will require
to fix α = 21 .
– 50 –
Assuming for definiteness that m > 0 it is not difficult to find12
1 X ipτ h i
∞
[Lm , V− ] = √ e V− (kαm−p ) + (kαm−p )V−
2 πT p=1
(αm k)
[Lm , : V0 :] =: V0 : √
πT
1
−1
X h i X ∞
e−ipτ
ipτ
[Lm , V+ ] = √ e V+ (kαm−p ) + (kαm−p )V+ = √ : V+ (kαm+p ) :
2 πT p=−∞ p=1
πT
Here the first commutator is of particular importance because it still contains the
terms which are not normal-ordered. Indeed, it can be written in the form
∞
X
1
[Lm , V− ] = √ eipτ : (kαm−p )V− :
πT p=1,p6=m
m−1
1 imτ 1 X ipτ
+ √ e V− (kα0 ) + √ e [kαm−p , V− ] .
πT 2 πT p=1
Further we get
k2
[kαm−p , V− ] = √ ei(m−p)τ V−
πT
This leads to
∞
X
1
[Lm , V− ] = √ eipτ : (kαm−p )V− :
πT p=1,p6=m
m−1
1 imτ k 2 X imτ
+ √ e V− (kα0 ) + e V− .
πT 2πT p=1
| {z }
(m−1)eimτ V−
12
The calculation is as follows:
1 +∞
X µ 1 +∞
X µ µ
[Lm , V− ] = [α αn,µ , V− ] = [α , V− ]αn,µ + αm−n [αn,µ , V− ] .
2 n=−∞ m−n 2 n=−∞ m−n
The two terms on the r.h.s. are computed separately, for instance
+∞ P∞ kν αν
1 X µ 1 m−1
X µ √1 p=1 p
−p ipτ
e 1 X∞ m−1
X µ ν kν ipτ
[αm−n , V− ]αn,µ = [αm−n , e πT ]αn,µ = √ [α , α−p ] e V− αn,µ
2 n=−∞ 2 n=−∞ 2 πT p=1 n=−∞ | m−n
{z } p
n=m−p
X∞
1 ipτ µ
= √ e V− (k αm−p,µ ) .
2 πT p=1
The other terms are computed in a similar way.
– 51 –
Plugging everything together we find
∞
X
0 2 eipτ
e−iα k τ [Lm , V ] = √ : (kαm−p )V− V0 V+ :
p=1,p6=m
πT
| {z }
imτ imτ
e e
+√ : V− V0 (kα0 )V+ : + √ : V− [kα0 , V0 ]V+ : +α0 k 2 (m − 1)eimτ : V− V0 V+ :
| πT {z } πT
X ∞
1 e−ipτ
+√ : V− V0 V+ (kαm ) : + √ : V− V0 V+ (kαm+p ) :
| πT {z } |p=1 πT
{z }
Here the underlined terms are nicely combined with a single sum13 with the range of
summation variable p form −∞ to +∞ and if we further take into account that
k2
[kα0 , V0 ] = √ V0
πT
we will get
X∞
0 2τ e−ipτ
[Lm , V ] = eimτ eiα k √ : V− V0 V+ (kαp ) :
p=−∞ πT
0 2τ
+ α0 k 2 (m + 1)eimτ eiα k : V− V0 V+ :
| {z }
V
It remains to note that

X∞
0 2 iα0 k2 τ e−ipτ
−i∂τ V = α k V + e √ : V− V0 V+ (kαp ) :
p=−∞
πT
With the account of this formula we obtain

³ ´
[Lm , V ] = eimτ − i∂τ + α0 k 2 m V
and, therefore, we conclude that the operator V has the following conformal dimen-
sion ∆:
∆ = α0 k 2 .
In particular, for k 2 = α10 the conformal dimension ∆ = 1 and the vertex operator
we discuss corresponds to emission of the tachyon with the mass m2 = − α10 .
We also see that on the zero-momentum ground state
µ
V (k, 0)|0i = V− eikµ x |0i .
13
It is convenient to shift the summation variable p for p → p − m.
– 52 –
As we will discuss later on, one can perform an analytic continuation τ → −iτ under
which the exponent entering the vertex operator V− will transform as eiτ → eτ . Then
in the limit τ → −∞ we see that V− → 1 and in the Euclidean picture
µ
lim V (k, 0)|0i = eikµ x |0i ,
τ →−∞
i.e. at τ → −∞ this vertex operator creates a particle with the momentum kµ .
Normal-ordering the product of exponents

Consider the normal product : eX :: eY :, where X and Y are two operators with the
propagator hXY i. Clearly one gets
X∞
X Y : Xn : : Y m :
: e :: e := .
n,m=0
n! m!
To apply the Wick theorem we first calculate the number of ways we can pick up k
n!
X’s from X n , which is obviously k!(n−k)! . Analogously, the number of ways to pick
m m!
up k Y ’s from Y is k!(m−k)! . Now we have to pair (i.e. to form propagators) k fields
X with k fields Y
:X
| .{z
. . X} :: Y
| .{z
. . Y} : .
k k
The are k! ways to pair all the terms in the last expression. Thus, application of the
Wick theorem gives
∞ min(n,m)
X X X n−k Y m−k n! m!
X Y
: e :: e := : : k! hXY ik =
n,m=0 k=0
n! m! k!(n − k)! k!(m − k)!
∞ min(n,m)
X X ∞ ∞
X n−k Y m−k hXY ik X hXY ik X X n−k Y m−k
= : : = : :
n,m=0 k=0
(n − k)! (m − k)! k! k=0
k! n,m=k (n − k)! (m − k)!
Thus, we find
: eX :: eY :=: ehXY i+X+Y :
It is now easy to see that the last formula can be generalized for the case of several
vertex operators as follows
Y P P
: eXi := e i<j hXi Xj i : e i Xi : (4.12)
i
Two-point function of tachyon vertex operators
Consider the two-point correlation function
hV (k1 , τ1 )V (k2 , τ2 )i ≡ h0|V (k1 , τ1 )V (k2 , τ2 )|0i

0 2 0 2
= eiα k1 τ1 +iα k2 τ2 h0| : V (k1 , τ1 ) :: V (k2 , τ2 ) : |0i ,
– 53 –
where we assume that τ1 > τ2 . By using the formula (4.12) we find
0 2 0 2 k1 k2
log(eiτ1 −eiτ2 )
hV (k1 , τ1 )V (k2 , τ2 )i = eiα k1 τ1 +iα k2 τ2 e πT h0| : V (k1 , τ1 )V (k2 , τ2 ) : |0i .
The vacuum expectation value of the normal-ordered expression on the right hand
side reduces to the contribution of the zero modes only and we get
0 2 0 2 0 iτ1 −eiτ2 ) µ µ
hV (k1 , τ1 )V (k2 , τ2 )i = eiα k1 τ1 +iα k2 τ2 e2α k1 k2 log(e h0|ei(k1 +k2 )xµ |0i .
| {z }
δ(k1 +k2 )
Thus, the two-point function is non-zero only if k1 = k = −k2 . In this case we get
0 2
eiα k (τ1 +τ2 )
hV (k, τ1 )V (−k, τ2 )i = .
(eiτ1 − eiτ2 )2α0 k2
We thus find that
ei∆(τ1 +τ2 )
hV (k, τ1 )V (−k, τ2 )i = .
(eiτ1 − eiτ2 )2∆
Four-tachyon scattering amplitude

Here we compute the scattering amplitude of four tachyonic particles. This is the
famous Veneziano amplitude which subsequently led to discovery of string theory.
The amplitude is defines as
Z ∞
A= dτ hk4 |V (k3 , τ )V (k2 , 0)|k1 i
0
µ
Here hk4 | is understood in an unusual way hk4 | = h0|eik4 xµ . Using the definition of
the tachyonic vertex operators and the formula (4.12) one finds
0 2
V (k3 , τ )V (k2 , 0) = eiα k3 τ : V (k3 , τ ) :: V (k2 , 0) :
0 2 µ ν
= eiα k3 τ e−k3 k2 hXµ (τ )Xν (0)i : V (k3 , τ )V (k2 , 0) :
Recalling the open string propagator at σ = σ 0 = 0:
η µν ³ ´
iτ
hX(τ )X(0)i = − log e − 1 .
πT
Thus we find
Z ∞
0 2 0 iτ µ µ 0 µ
A= dτ eiα k3 τ e2α (k2 k3 ) log(e −1) hk4 |ei(k2 +k3 )xµ ei2α k3 pµ τ |k1 i
0
Further simplification gives
Z ∞ 4
iα0 k32 τ iτ 2α0 (k2 k3 ) 2iα0 (k3 k1 )τ
¡X ¢
A= dτ e (e − 1) e δ ki
0 i=1
Z 4
∞
iα0 k32 τ 2iα0 (k1 +k2 )k3 τ −iτ 2α0 (k2 k3 )
¡X ¢
= dτ e e (1 − e ) δ ki
0 i=1
Z ∞ X4
0 2 0 0 ¡ ¢
= dτ e−iα k3 τ e−2iα (k3 k4 )τ (1 − e−iτ )2α (k2 k3 ) δ ki .
0 i=1
– 54 –
Recall that for tachyons on shell we have ki2 = α10 . Performing now the Wick rotation
τ → −iτ and changing the integration variable τ for x = e−τ we find
Z 1
0 0
A= dx x2α k3 k4 (1 − x)2α k2 k3 ,
0
where for the sake of simplicity we omitted the δ-function which encodes the con-
servation law of the momenta. Introducing the Mandelstam variables s = (k1 + k2 )2
and t = (k2 + k3 )2 the last formula can be cast in the form
Z 1
0 0 Γ(α0 s − 1)Γ(α0 t − 1)
A= dx xα s−2 (1 − x)α t−2 = .
0 Γ(α0 (s + t) − 2)
In fact this function is known as the Euler beta-function. One of the interesting
properties of the representation of A in terms of the Euler beta-function is that the
latter is explicitly symmetric under the interchange of s and t. Search of amplitudes
with this symmetry property led Veneziano in 1960’s to this amplitude which was
the starting point of modern string theory.
It is interesting to analyze the Veneziano formula in more detail. The Γ-function has
poles at non-positive integers with residues
(−1)n 1
Γ(x) → as x → −n , n ≥ 0.
n! x + n
Thus, when α0 s → 1 − n, n = 0, 1, . . . the amplitude behaves as
(−1)n 1 Γ(α0 t − 1)
A(s, t) →
n! α0 s − 1 + n Γ(α0 t − 1 − n)
Here the dependents of the variable t is polynomial because for n > 0 we have
Γ(α0 t − 1) Γ(w + n)
0 = = (w + n − 1) · · · (w + 1)w
Γ(α t
| {z }− 1 − n ) Γ(w)
w
= (α0 t − 2)(α0 t − 3) · · · (α0 t − n − 1) ≡ Pn (α0 t) ,
i.e. the r.h.s. is a polynomial of degree n. Thus, the scattering amplitude can be
essentially written as
∞
X (−1)n Pn (α0 t)
A(s, t) = , P0 (α t) = 1 .
n=0
n! n − 1 + α0 s
In scattering theory resonances (or simply poles) of the scattering amplitude are
interpreted as an exchange by intermediate particles whose masses are obtained from
the condition of having poles. In our case we see that the poles arise due to exchange
by hypothetical particles whose masses are quantized as
1
Mn2 = −s = (n − 1) .
α0
– 55 –
It also follows from the scattering theory that appearance of a polynomial of degree
n as the residue of the amplitude signals that the exchanged particles of the mass
Mn2 carry spin up to the maximal value Jmax = n. Our later analysis of string in the
physical gauge will reveal that the exchanged particles delivering the resonances of
the Veneziano amplitude are those from the spectrum of open string!
Operator Product Expansion
Consideration of the Veneziano amplitude shows that the object of primary in-
terest in string perturbation theory is a correletation function of local operators
hOi1 (x1 )Oi2 (x2 ) . . . Oin (xn )i ,
where Oi (x) is a local operator. Here x ≡ (τ, σ) is a point on the two-dimensional

world-sheet. It is important to understand the behavior of the correlation function
when two operators are taken to approach each other. The technique to describe this
limit in known as the Operator product Expansion or OPE for short. The Operator
Product Expansion states that a product of two local operators can be approximated
to arbitrary accuracy by a sum of local operators
X
Oi (x)Oj (y) = Cijk (x − y, ∂y )Ok (y) .
k
Let us take as a local operator the stress tensor and try to work out the corre-
sponding OPE. Introduce the short-hand notation T ≡ T++ and X µ ≡ XLµ . Consider
the component T−− of the stress tensor normalized as
1 1
T ≡ T−− = 0
: ∂− X µ ∂− Xµ := 0 : ∂− XLν ∂− XLν : .
α α
0 −σ 0 )
We will also use the concise notation z = ei(τ −σ) and w = ei(τ .
In what follows we consider the product of two stress tensors evaluated at two dif-
ferent points
1
T (τ, σ)T (τ 0 , σ 0 ) = : ∂− X µ (z)∂+ Xµ (z) :: ∂− X ν (w)∂+ Xν (w) :
α02
and try to expand it over a basis of local operators. By using the Wick theorem, we get
1
T (τ, σ)T (τ 0 , σ 0 ) = : ∂− X µ (z)∂− Xµ (z)∂− X ν (w)∂− Xν (w) :
α02
4
+ 02 h∂− X µ (z)∂− X ν (w)i : ∂− Xµ (z)∂− Xν (w) :
α
2
+ 02 h∂− X µ (z)∂− X ν (w)i h∂− Xµ (z)∂− Xν (w)i .
α
– 56 –
We can now rewrite the r.h.s. by using the propagators introduces above
1
α02
4 x y
+ 02 ∂− ∂− hX µ (z)X ν (w)i : ∂− Xµ (z)∂− Xν (w) :
α
2 z w µ
+ 02 ∂− ∂− hX (z)X ν (w)i ∂− z w
∂− hXµ (z)Xν (w)i .
α
A little computation gives
1
α02
2 zw
− 0 : ∂− X µ (z)∂− Xµ (w) :
α (z − w)2
η µν ηµν z 2 w2
+ .
2 (z − w)4
It is further convenient to redefine the stress tensor as follows
T (τ, σ)
T (z) =
z2
so that
1
T (z)T (w) = = : ∂− X µ (z)∂− Xµ (z)∂− X ν (w)∂− Xν (w) :
α02z 2 w2
2 1
− 0 : ∂− X µ (z)∂− Xµ (w) :
α zw (z − w)2
η µν ηµν 1
+ .
2 (z − w)4
Since
1³ ∂ ∂ ´ 1 ³ ∂z ∂z ´ ∂ ∂
∂− = − = − = iz ,
2 ∂τ ∂σ 2 ∂τ ∂σ ∂z ∂z
∂ 1 ∂
we have ∂z
= iz ∂z and, therefore, the last formula can be written as
1
T (z)T (w) = = 02
: ∂z X µ (z)∂z Xµ (z)∂w X ν (w)∂w Xν (w) :
α
2 1 µ η µν ηµν 1
+ 0 : ∂z X (z)∂ w Xµ (w) : + .
α (z − w)2 2 (z − w)4
Expanding the r.h.s. around the point w = z, we will find the following most singular
z → w contribution
d/2 2 1
T (z)T (w) = 4
+ 2
T (w) + ∂w T (w) + . . . (4.13)
(z − w) (z − w) z−w
The first term here reflects the appearance of the conformal anomaly (a purely quan-
tum mechanical effect). In the general setting the coefficient of this term is c/2,
where c is the central charge. The coefficients 2 of the second term coincides with
the conformal dimension of T .
– 57 –
4.2 Quantization in the physical gauge
Quantization of strings in the physical (light-cone) gauge is perhaps the most straight-
forward way to obtain a restriction on the space-time dimension as well as to under-
stand the spectrum of physical excitations.
Using the Poisson brackets and the basic quantization rules we can easily get the
table of the basic commutator relations of the light-cone string theory.
[, ] p+ p− pj xj x− αm j
αm −
p+ 0 0 0 0 i 0 0
i −
p− 0 0 0 − i pp+ − i pp+ 2πT
− p+ mαm j 2πT
− p+ mαm −
pi 0 0 0 − iδ ij 0 0 0
i ij αim
xi 0 i pp+ iδ ij 0 0 i δ4πTδm
i p+
− −
x− −i i pp+ 0 0 0 0 i αp+m
ij
√
2πT
αni 0 p+
nαni 0 − i δ4πTδn 0 nδ ij δn+m 4πT
p+
nαn+mi
i − √
αn− 0 2πT
p+
nαn− 0 − i αp+n − i αp+n − 4πT
p+
i
mαn+m [αn− , αm
−
]
Tab. 2. Canonical structure of the light-cone modes. The variable

p− is essentially the Hamiltonian: p− = 2πT pH+ . The commutators
involving ᾱ variables are the same.
We would like to point out the following commutator

√
i − 4πT i
[αn , αm ] = nαn+m .
p+
−
One of the most important commutators of the light-cone theory is [αm , αn− ]. It can
be computed precisely in the same way as [Lm , Ln ] of the previous section. We find
the same result as before except we have now only d − 2 transversal fields which
contribute to the central charge term with the factor d − 2 instead of d
√
− − 4πT − 4πT d − 2
[αm , αn ] = +
(m − n)αm+n + + 2 m(m2 − 1)δm+n . (4.14)
p (p ) 12
The normal ordering ambiguity
√
4πT
αn− → αn− − aδn,0
p+
leads to the change as
√
− − 4πT − 4πT ³ d − 2 3 d−2 ´
[αm , αn ] = (m − n)αm+n + + 2 m + 2am − m δm+n .
p+ (p ) 12 12
– 58 –
4.2.1 Lorentz symmetry and critical dimension
Studying the classical string in the light-cone gauge we realized that the generators
J i− of the Lorentz symmetry become rather complicated functions of transversal
oscillators
X∞ X∞
i− i − 1¡ i −
− i − i
¢ 1¡ i − −
¢
J =xp −x p −i α−n αn − α−n αn − i ᾱ−n ᾱn − ᾱ−n ᾱni .
n=1
n n=1
n
We would like to ask a question whether we can use this expression in quantum
theory regarding αn and ᾱn as operators to define the quantum Lorentz generators?
Due to the unusual canonical structure of the light-cone theory it is not obvious that
consistent Lorentz generators should exist.
In quantum theory we want to realize a unitary representation of the Poincaré

group and therefore, we require the Lorentz generators to be hermitian, i.e.
(J µν )† = J µν ,
where we treat J µν as an operator acting in the Hilbert space. Also the Lorentz
generators must be normal-ordered to have the well-defined action on the vacuum
state. Consider the following ansatz for the Lorentz generators J i− , which are the
most intricate generators to be defined in quantum theory,
X∞ X∞
1 1¡ i − ¢ 1¡ i − ¢
J i− = (xi p− + p− xi ) − x− pi −i −
α−n αn − α−n αni − i −
ᾱ−n ᾱn − ᾱ−n ᾱni .
|2 {z } n=1
n n=1
n
ì−
One can see that these generators are hermitian and normal-ordered so they can
be considered as candidates to realize the Lorentz algebra symmetry. The latter
requirement is equivalent to
[J i− , J j− ] = 0 .
This is an equation we would like to prove.
First we discuss the orbital part. We have
i− j− i − − i − i j − − j − j
[` ,` ] = [1
2
(x p +p x )−x p , 1
2
(x p +p x )−x p ]
−
1 [xi p− , xj p− ] → i (xj pi −
= 4
i j
x p ) p+
4 p
i − − j −
+ 1 i (pi xj − xi pj ) p
[x p , p x ] → 4
4 p+
i − − j −
− 1 i (x− p− δ ij − xi pj p )
[x p , x p ] → − 2
2 p+
− i j − −
+ 1 i p (pj xi − xj pi )
[p x , x p ] → − 4
4 p+
−
1 [p− xi , x− pj ] → − i p (pj xi − pi xj )
+ 4 4 p+
− i − j −
− 1 i ( p pj xi − p− x− δ ij )
[p x , x p ] → 2
2 + p
− i j − −
− 1 i (xj pi p
[x p , x p ] → − 2
− − ij
−x p δ )
2 p+
− i − j −
− 1 i ( p xj pi − x− p− δ ij ) .
[x p , p x ] → − 2
2 p+
– 59 –
Here by arrow we indicated the explicit expression for the corresponding commutator. Reducing similar terms we get
i− j− − −
[` ,` i [pi , xj ] p
] = 4 i p [pi , xj ] + i [x− , p− ]δ ij = 0 .
+ 4
p+ p+ 2
Thus, the generators of the orbital part of the total momentum have vanishing commutator. Next we compute
∞
i− j− i− j− p− X 1 [i j]
[` ,S ] + [S ,` ] = −2 α αn + (4.15)
p+ n=1 n −n
∞
1 X 1h j − − j i i − − i j
i
+ − (α−n αn − α−n αn )p + (α−n αn − α−n αn )p .
p+ n=1 n
[i j] j
Here and below we use the concise notation α−n αn = αi−n αjn − α−n αin .
Now we are in a position to study the most difficult commutator [S i− , S j− ]. We will do further analysis in several steps.
1. First we consider the following commutator
∞
X
i− − 1 i − − i −
[S , αm ] = −i [α−n αn − α−n αn , αm ] (4.16)
n n
∞ √ √
X 1 4πT i − i − − − − i 4πT − i
= −i − nαm−n αn + α−n [αn , αm ] −[α−n , αm ]αn − nα−n αn+m .
n=1 n p+ p+
| {z }
A
Therefore,
√ ∞
i− − 4πT X 1 i − i − − i − i
[S , αm ] = −i − nαm−n αn + α−n (n − m)αm+n + (n + m)αm−n αn − nα−n αn+m
p+ n=1 n
f (m) i
− i αm .
m
This further gives
√ ∞
i− − 4πT X i − i − − i − i
[S , αm ] = −i α−n αm+n − αm−n αn + αm−n αn −α−n αn+m
p+ n=1 | {z } | {z }
A B
√ ∞
4πT X m i − − i f (m) i
+ i α−n αm+n − αm−n αn −i αm .
p+ n=1 n m
The terms in the first line of the last equation can be partially cancelled upon changing the summation index and we find
that
√ m
i− − 4πT X i − − i
[S , αm ] = −i − αm−n αn + αm−n αn
p+ n=1 | {z } | {z }
A B
√ ∞
4πT X m i − − i f (m) i
+ i α−n αm+n − αm−n αn −i αm .
p+ n=1 n m
Since we have
m
X 0
X m−1
X
− i − i − i
αm−n αn = αk αm−k ≡ αn αm−n
n=1 k=m−1 n=0
we see that
m
X m
X m−1
X
i − − i i − − i
− αm−n αn + αm−n αn = − αm−n αn + αn αm−n
n=1 n=1 n=0
m−1
X
− i i − − i
= α0 αn − α0 αm + [αn , αm−n ]
n=1
Pm−1
Using the fact that n=1 (m − n) = 1
2
m(m − 1) we obtain
√ √ ∞
i− − 4πT i − − i 4πT X m i − − i
[S , αm ] = i (α0 αm − α0 αm ) + i α−n αm+n − αm−n αn
p+ p+ n=1 n
!
4πT m(m − 1) f (m) i
+ i − αm . (4.17)
(p+ )2 2 m
Thus, under the action of the Virasoro operators α−

m the spin components S
i−
transform in a nontrivial manner. Note that
the r.h.s. of eq.(4.17) is normal-ordered.
– 60 –
2. Analogously, we compute (m > 0)
√ √ ∞
i− − 4πT − i i − 4πT X m i − − i
[S , α−m ] = i (α−m α0 − α−m α0 ) − i α−n αn−m − α−m−n αn
p+ p+ n=1 n
!
4πT m(m − 1) f (m) i
+ i − α−m .
(p+ )2 2 m
In fact these formula is also obtained from eq.(4.17) by simply substituting m → −m.
3. The next step consists in finding the commutator
∞
X
i− j 1 i − j − j i − ij
[S , α−m ] = −i α−n [αn , α−m ] − [α−n , α−m ]αn − α−n δ δn−m
n=1 n | {z }
0 as i6=j
√ ∞
4πT X m i j j i
= −i α−n αn−m − α−n−m αn .
p+ n=1 n
4. Finally we compute the commutator
√ ∞
i− j 4πT X m i j j i
[S , αm ] = i α−n αm+n − αm−n αn .
p+ n=1 n
Substituting all our findings into the commutator [S i− , S j− ] we obtain
√ ∞
i− j− 4πT X 1 j − i j i − i − j − i j
[S ,S ] = − α−n α0 αn + α−n α0 αn + α−n α0 αn − α−n α0 αn
p+ n=1 n
∞
!
X 4πT n−1 f (n) [i j]
− − α−n αn
n=1 (p+ )2 2 n2
√ ∞
4πT X 1 j i − − i i j j i −
+ α−m (α−n αm+n − αm−n αn ) − (α−n αn−m − α−m−n αn )αm
p+ m,n=1 n
i − − i j − i j j i
+ (α−n αn−m − α−n−m αn )αm − α−m (α−n αm+n − αm−n αn ) .
We first analyze the last two lines of the equation above, which we write as follows
√ ∞
ij 4πT X 1 j i − i j j i −
I = α−m α−n αm+n −(α−n αn−m −α−m−n αn )αm
p+ m,n=1 n | {z } | {z }
A A
i − j j − i − i j − i j j i
+ α−n αn−m αm − α−m αm−n αn − α−n−m αn αm −α−m (α−n αm+n −αm−n αn ) .
| {z } | {z }
B B
Upon changing the summation variables the A-terms can be partially cancelled, the same is for the B-terms. We therefore obtain
√ ∞ n n
ij 4πT X X 1 i j −
X 1 − j i
I = − α−n αn−m αm + α−m αm−n αn
p+ n=1 m=1 n m=1 n
√ ∞
4πT X j i − i − j j − i − i j
+ α−m−n αn αm + α−n αn−m αm − α−m αm−n αn − α−m α−n αm+n .
p+ n,m=1
One can recognize that in the second line of the equation above the first and the last terms are not normal-ordered. We consider the
first sum, which is not normal-ordered, and try to bring it to the normal-ordered form:
∞
X ∞
X ∞
X
1 j i − 1 j − i 1 j i −
α−m−n αn αm = α−m−n αm αn + α−m−n [αn , αm ]
n,m=1 n n,m=1 n n,m=1 n
∞ ∞ √
X 1X j
j i − i 4πT
= α
α−m−n αm αn +
αn+m
n,m=1 n p+ n,m=1 −m−n
∞ √ ∞
X 1 j − i 4πT X j i
= α−m−n αm αn + (k − 1)α−k αk ,
n p+
n,m=1 k=2
where in the last sum we made a substitution k = m + n and then summed over m, n with the condition m + n = k kept fixed; this
resulted in the factor k − 1. Analogously, we achieve the normal-ordering of the second sum
∞ ∞ √ ∞
X 1 − i j
X 1 i − j 4πT X i j
α−m α−n αm+n = α−n α−m αm+n + (k − 1)α−k αk .
n,m=1 n n,m=1 n p+ k=2
– 61 –
Thus, our commutator takes the form
√ ∞ n n
ij 4πT X X 1 i j −
X 1 − j i
p+ n=1 m=1 n m=1 n
√ ∞
4πT X 1 j − i i − j j − i i − j
+ α−m−n αm αn + α−n αn−m αm − α−m αm−n αn − α−n α−m αm+n
p+ n,m=1 n | {z } | {z } | {z } | {z }
A B A B
∞
X
4πT [i j]
− (n − 1)α−n αn .
(p+ )2 n=1
Again we see that upon change of the summation index the A-terms partially cancel (the same is for the B-terms) and we arrive at
√ n
∞ X
ij 4πT X 1 i j − − j i
p+ n=1 m=1 n
√ ∞ n−1
4πT X X 1 j − i i − j
+ − αm−n α−m αn + α−n αm αn−m
p+ n=1 m=0 n
∞
X
4πT [i j]
− (n − 1)α−n αn .
(p+ )2 n=1
From here we find
√ ∞ X
n
ij 4πT X 1 i j − − j i j − i i − j
I = − α−n α0 αn + α−n α0 αn − α−n α0 αn + α−n α0 αn
p+ n=1 m=1 n
0 1
∞
X n−1
X ∞
X
4πT n−m 4πT
− @ A α[i αj] − (n − 1)α−n αn .
[i j]
−n n
(p+ )2 n=1 m=1 n (p+ )2 n=1
| {z }
1 (n−1)
2
Thus, the commutator of the internal spin components we are interested in acquires the form
√ ∞
i− j− 4πT X 1 j − i j i − i − j − i j
[S ,S ] = − α−n α0 αn + α−n α0 αn + α−n α0 αn − α−n α0 αn
p+ n=1 n
i j − − j i j − i i − j
− α−n α0 αn + α−n α0 αn − α−n α0 αn + α−n α0 αn
∞
!
X 4πT f (n) [i j]
− 2(n − 1) − α−n αn .
n=1 (p+ )2 n2
The final step consists in commuting the factor α−

0 on the left to compare with eq.(4.15).
We thus find
√ ∞
i− 2 πT α0− X 1 [i j]
j−
[S , S ] = 2 α−n αn +
p+ n=1
n
1
∞
X 1 h i
j
+ + (α−n αn− − α−n
−
αnj )pi − (α−n
i
αn− − α−n
−
αni )pj
p n=1 n
X∞ µ ¶
4πT f (n) [i
− + 2
2n − 2 α−n αnj] .
n=1
(p ) n
Finally, adding this expression to eq.(4.15) we arrive at
√ ³´1 X∞
1 [i j]
i− j− − −
[J , J ] = 2 p − 2 πT α0 + α α +
p n=1 n −n n
∞ µ ¶
4πT X h d − 2 i 1h d − 2i [i
+ + 2 − 2 n + 2a − α−n αnj] + (αn → ᾱn ).
(p ) n=1 12 n 12
– 62 –
As for the case of classical string the first term here vanishes due to the level matching
condition α0 = ᾱ0 , while the second sum is an anomaly which appears due to non-
commutativity of the oscillators in quantum theory. Thus, for general values of a
and d the theory is not Lorentz invariant: the quantum effects destroy the Lorentz
invariance which was present at the classical level. However, for special values
d = 26 , a=1
the anomaly term vanishes and the Lorentz invariance is restored!
4.2.2 The spectrum
In the light-cone gauge the spectrum is generated by acting with transversal oscilla-
tors on the vacuum state. We first discuss the spectrum of open strings.
The mass operator is the light-cone gauge for open strings is
1 X³ i i ´
∞
2
M = 0 α α −a ,
α n=1 −n n
where, as was discussed in the previous chapter, the normal-ordering constant a

should be equal to 1 in order to guarantee the Lorentz invariance of the light-cone
theory.
The ground state |pi i carries no oscillators and it has a mass
α0 M 2 |pi i = −|pi i =⇒ α0 M 2 = −1.
This is a tachyon.
i
The first excited state is α−1 |pj i. It is a d − 2 component vector which transforms
irreducibly under the transverse group SO(24). We see that
α0 M 2 (α−1
i
|pi i) = (1 − a)α−1
i
|pi i = 0 ,
i.e. this vector is massless.
– 63 –
level α0 mass2 rep of SO(24) little rep of little
group group
0 −1 |0i SO(1, 24) 1

|{z}
1
1 0 αi |0i SO(24) 24
| −1
{z }
24
j
2 +1 αi |0i i
α−1 α−1 |0i SO(25) 324s
| −2
{z } | {z }
24 299s +1
i i j i j
3 +2 α−3 |0i α−2 α−1 |0i α−1 α−1 αk |0i SO(25) 2900s + 300a
| {z } | {z } | {z −1 }
24 276a +299s +1 2576s +24
Tab. 3. The spectrum of open bosonic string up to level 3.
In general, the Lorentz invariance requires that physical states transform irreducibly
under the little Lorentz group which is
• SO(d − 2) for massless particles
• SO(d − 1) for massive particles (for tachyon SO(1, d − 2))
For tachyon the little Lorentz group is non-compact. Unitary representations of

non-compact groups are either trivial (i.e. one-dimensional) or infinite-dimensional.
Tachyon realizes the one-dimensional representation.
Further analysis reveals that all states corresponding to higher levels are massive
and that being the tensors of SO(24) they combine at any given mass level to repre-
sentations of SO(25), the latter is the little Lorentz group for massive states. This
is highly non-trivial implication of the Lorentz invariance and it occurs only in the
critical dimension and for a = 1!
At level n the mass of the corresponding states is α0 M 2 = n − 1. Among them there
is always a symmetric traceless tensor of rank n. This is a state with maximal spin
Jmax = n and, therefore, we have Jmax = n = α0 M 2 + 1. In general states obey the
inequality
J ≤ α0 M 2 + 1 .
– 64 –
The values of J and M 2 are quantized and the last inequality implies that the states
lie on the Regge trajectories.
level α0 mass2 rep of SO(24) little rep of little

group group
0 −4 |0i SO(1, 24) 1

|{z}
1
1 0 i
α−1 ᾱj |0i SO(24) 299s + 276a + 1
| {z−1 }
299s +276a +1
i
α−2 ᾱj |0i i
α−1 j
α−1 k
ᾱ−1 ᾱl |0i 20150s + 32175
| {z−2 } | {z −1 }
299s +276a +1 299s +1+299s +1
2 +4 SO(25)
αi ᾱj ᾱ1k |0i i

α−1 j
α−1 ᾱk |0i 52026 + 324s + 300a + 1
| −2 −1
{z } | {z −2 }
(24)×(299+1) (299+1)×(24)
Tab. 4. The spectrum of closed bosonic string up to level 2.
Now we discuss the spectrum of closed strings. The mass operator for closed
strings is
2 ³X i i X i i ´
∞ ∞
2
M = 0 α α + ᾱ ᾱ − 2a
α n=1 −n n n=1 −n n
In addition one has to impose the level-matching condition
∞
X ∞
X
i
V= α−n αni − i
ᾱ−n ᾱni = 0 ,
n=1 n=1
which simply means that the excitation (level) number of α-oscillators should be
equal to the excitation number of ᾱ-oscillators.
The ground state is a tachyon which is scalar particle with α0 M 2 = −4. The
i j
first excited state α−1 ᾱ−1 |0i is massless. It can be decomposed into irreducible
– 65 –
representations of the transversal (and simultaneously little) Lorentz group SO(24)
as follows
³ 1 ij k k ´ 1
j [i j] (i j)
α−1 ᾱ−1 |0i = α−1 ᾱ−1 |0i + α−1 ᾱ−1 − δ α−1 ᾱ−1 |0i + δ ij α−1
i k k
ᾱ−1 |0i .
| {z } | 24{z } 24
| {z }
276
299 singlet
The massless excitation of spin two transforming in representation 299 of SO(24)

was proposed to be identified with a graviton, the quantum of the gravitational
interaction. To make this identification one has to relate the string scale α0 with the
Planck scale G = MP−2 , where G is the Newton constant and MP is the Planck mass:
α0 = MP−2 .
Since the masses of the massive string modes are proportional to 1/α0 = MP2 , these
string excitations are extremely heavy due to the large value of MP2 and, by this
reason, they do not show up at the energy scales of the Standard Model.
As in the opens string case the higher massive states of closed string are combined
at a given mass level into representations of the little Lorentz group SO(25). The
relation between maximal spin and the mass is now
α0 2
Jmax = M + 2.
2
4.3 BRST quantization
The path integral approach proved to be a very useful tool for quantizing the theories
with local (gauge) symmetries. The starting point is the Polyakov action and a new
BRST (Becchi-Rouet-Stora-Tyutin) symmetry. We know that the induced metric
Γαβ = ∂α X µ ∂β Xµ and the intrinsic metric hαβ are related classically through the
condition Tαβ = 0. However, quantum-mechanically this is not the case.
The basic idea is to define the path integral using the Polykov action and inte-
grate over
hαβ , X µ
being considered as independent variables:
Z
Z = Dhαβ (σ, τ )DX µ (σ, τ )eiSp [X,h]
Due to the gauge invariance the last integral is ill-defined. This occurs because we
integrate infinitely many times over physically equivalent, i.e. related to each other
by gauge transformations, configurations. This can be understood looking at a much
simpler example of the two-dimensional integral
Z +∞ Z +∞
2
Z= dxdy e−(x−y) .
−∞ −∞
– 66 –
It is divergent since the integrand depends on x − y only. Also one sees that the
group of translations
x → x + a, y →y+a
leaves the measure invariant.
x
Orbits x−y=const
x+y=0
gauge slice
intersect each orbit
at one point
Fig. 4. Divergence caused by a presence of the translational symmetry. One

integrates over the orbits of the gauge group while both the measure and
the integrand are translation invariant.
Let us split the coordinates (x, y) as

³x − y y − x´ ³x + y x + y ´
(x, y) = , + ,
| 2 {z 2 } | 2 {z 2 }
x−y
2
+ y−x
2
=0 shift by (a,a), a= x+y
2
This suggests to introduce new coordinates “along” the gauge orbit and “orthogonal”
to it:
(x, y) → (u, v) u = x − y, v = x + y,
u+v u−v
i.e. x = 2
and y = 2
. Then the integral takes the form
Z +∞ µZ +∞ ¶ √ Z +∞ Z +∞
1 −u2 π √
Z= e du dv = d(x + y) = π da .
2 −∞ −∞ 2 −∞ −∞
| {z
√
} | {z }
π volume
This example illustrates the basic idea to define the path integral – one has to divide
the original Z by the infinite volume of a symmetry group. Thus, our discussion
suggests that a proper definition of the path integral in string theory should be
Z
1
Z= Dhαβ (σ, τ )DX µ (σ, τ )eiSp [X,h] ,
VDiff VWeyl
where we divided over the infinite volumes of the reparametrization and Weyl groups.
– 67 –
Let us illustrate this procedure on our simplified example. The integral can be rewritten by using the δ-function:
Z +∞ Z +∞ Z +∞ Z +∞
−(x−y)2 −z 2
Z = da e dxdy 2δ(x + y) = da e dz .
−∞ −∞ −∞ −∞
Here insertion of the δ-function can be regarded as the gauge-fixing condition; it selects a single representative from each gauge orbit.
What happens if we change the gauge fixing condition? Suppose we take another slice “orthogonal” to the gauge orbits. Let s will be
a free one-dimensional parameter on this slice
s: f (x, y) = 0 =⇒ (x(s), y(s))
Chose now new coordinates

0 0
(x, y) = (x , y ) +(a, a)
| {z }
along s
Then Z +∞ Z +∞ Z +∞ Z +∞
−(x−y)2 −(x0 −y 0 )2
Z = e dxdy = e dsda|J| ,
−∞ −∞ −∞ −∞
where x − y = x0 − y 0 and
∂(x, y)
J =
∂(s, a)
is the Jacobian. Since x = x0 + a and y = y 0 + a we have
! !
∂x ∂x ∂x 1 ∂x ∂y ∂
∂s ∂a ∂s 0 0
J = ∂y ∂y = ∂y = − = (x − y ) .
∂s ∂a ∂s
1 ∂s ∂s ∂s
Along s
∂f ∂x ∂f ∂y ∂f ∂x0 ∂f ∂y 0
0= + = + ,
∂x ∂s ∂y ∂s ∂x ∂s ∂y ∂s
0 0
which allows for a solution ∂y
∂s
= − ∂f
∂x
and ∂x
∂s
= ∂f0 and, therefore,
∂y
∂ 0 0 ∂f ∂f
J = (x − y ) = + .
∂s ∂x ∂y
The integral can be now written as

Z +∞ Z +∞
−(x0 −y 0 )2 ∂f ∂f
Z = da ds e + ,
−∞ −∞ ∂x0 ∂y 0
Here the integral over s is one-dimensional, the functions x0 and y 0 are the functions of s and it is independent of a. One can convert
this one-dimensional integral into a two-dimensional one by substituting the δ-function with the gauge condition
Z +∞ Z +∞
−(x0 −y 0 )2 ∂f ∂f −(x−y)2 ∂f ∂f
da e + = dxdy δ(f (x, y))e + .
−∞ ∂x0 ∂y 0 −∞ ∂x ∂y
The infinite volume arising upon integrating over a can be factored out and taking into account that
∂f ∂f ∂f
|a=0 = + |a=0
∂a ∂x ∂y
we obtain the final and finite expression
Z +∞
∂f −(x−y)2
Zfinite = dxdy δ(f (x, y)) e
−∞ ∂a a=0
∂f
Here ∂a a=0
is known as the Faddeev-Popov determinant.
The first problem in realizing this approach for string theory is to find a mea-
sure for functional integration that preserves all symmetries of the classical theory
(reparametrizations + Weyl symmetry).
Z √
(δh, δh) = d2 σ hhαβ hγδ δhαγ δhβδ
Z √
(δX, δX) = d2 σ hhαβ δX µ δXµ
– 68 –
These scalar products define natural reparametrization-invariant and Poincaré in-
variant measures, however none of them is Weyl invariant.
Let us first assume the simplifying situation when all metric on a world-sheet M
with a given topology are conformally equivalent (i.e. they are related to each other
by diffeomorphism and Weyl rescalings; this is the case when operator P † does not
have zero modes). In this case by using reparametrizations we can bring the metric
to the form
hαβ = e2φ gαβ ,
where gαβ is a fiducial (reference) metric. Under reparametrizations and the Weyl
rescalings the variation of the metric can be decomposed as
1
δhαβ = (P ξ)αβ + 2Λ̃hαβ , Λ̃ = Λ + ∇γ ξ γ ,
2
where P is the following operator
(P ξ)αβ = ∇α ξβ + ∇β ξα − ∇γ ξ γ hαβ ,
which maps vectors into traceless symmetric tensors. Then the integration measure
can be written as follows
¯ ¯
¯ ∂(P ξ, Λ̃) ¯
¯ ¯
Dh = D(P ξ)D(Λ̃) = D(ξ)D(Λ) ¯ ¯,
¯ ∂(ξ, Λ) ¯
| {z }
Jacobian
where in the last formula we changed the variables
(P ξ, Λ̃) → (ξ, Λ)
for the price of getting a non-trivial Jacobian. Here D(ξ) is the measure which gives
upon integration an infinite volume of the diffeomorphism group and
¯ ¯ ¯ ¯ ¯ ¯
¯ ∂(P ξ, Λ̃) ¯ ¯ ∂(P ξ) ∂(P ξ) ¯ ¯ ∂(P ξ) 0 ¯
¯ ¯ ¯ ∂ξ ∂Λ ¯ ¯ ¯
¯ ¯=¯ ¯ = ¯ ∂ξ ¯ = |detP |
¯ ∂(ξ, Λ) ¯ ¯ ∂∂ξΛ̃ ∂∂ΛΛ̃ ¯ ¯ ∂∂ξΛ̃ 1 ¯
In fact, we have
δ(P ξ)αβ (σ) ³ γ γ γ
´
= δ ∇
β α + δα β∇ − h αβ ∇ δ(σ − σ 0 )
δξ γ (σ 0 )
Thus,
Z R √ ¡ ¢
−i T2 2 d2 σd2 σ 0 h bαβ (σ) δβγ ∇α +δα
γ
∇β −hαβ ∇γ δ(σ−σ 0 )cγ (σ 0 )
|detP | = DbDc e σ .
– 69 –
Here cα is called a ghost field, while bαβ is traceless and symmetric, it is called
antighost field. The ghost cα corresponds to infinitezimal reparametrizations while
baβ corresponds to variations perpendicular to the gauge orbits. Both the ghost and
the antighost fields are real. The last formula can be now written as
Z R 2 √
i α β
|detP | = DbDc e− πα0 d σ h bαβ ∇ c .
Thus, the total action is given now by the sum of the Polyakov action and the ghost
action: Z ³ ´
T
S=− d2 σγ αβ ∂α X µ ∂β Xµ + 4ibβγ ∇α cγ
2
There are several subtle issues we have not touched so far
• Conformal anomaly, i.e. possible dependence of the thrown away volume of

the diffeomorphism group on the Weyl (scale) degree of freedom φ.
• Reparametrizations which satisfy P ξ = 0, i.e. conformal Killing vectors. We

see that equations of motion for cα are just conformal Killing equations. There-
fore, in order not to overcount the configurations which are related by a con-
formal transformation one has to exclude integration over the zero modes of cα
ghosts.
• So far we assumed that all symmetric traceless deformations of the metric can
be generated by reparametrizations. This is however not the case if P † has zero
modes. These zero modes correspond to zero modes of the b ghosts.
We can define the stress-energy tensor of the ghost fields

Z
√
δSgh = T d2 σ −h T αβ δhαβ .
Performing the variation one finds

gh
Tαβ = i(bαγ ∇β cγ + bβγ ∇α cγ − cγ ∇γ bαβ − hαβ bγδ ∇γ cδ ) .
Here the last term vanishes on shell. In the deriving this expression we also used
the tracelessness of bαβ . One can verify that this tensor is covariantly conserved
∇α Tαβ = 0.
In the world-sheet light-cone coordinates σ ± the non-vanishing components of
the stress-tensor are
¡ ¢
T++ = i 2b++ ∂+ c+ + (∂+ b++ )c+ ,
¡ ¢
T−− = i 2b−− ∂− c− + (∂− b−− )c− .
– 70 –
Equations of motion are
∂− b++ = ∂+ b−− = 0 ,
∂+ c − = ∂− c + = 0 .
This equations are supplemented by
• by periodicity condition for the closed closed string case
b(σ + 2π) = b(σ) , c(σ + 2π) = c(σ) ;
• by boundary conditions for the open string case

b++ (σ) = b−− (σ) , c+ (σ) = c− (σ) for σ = 0, π ,
which follow by requiring the vanishing of the boundary terms arising upon
deriving equations of motion.
Note that for the closed string case b++ and c+ are left-moving waves, while b−− and
c− are the right-moving ones. The canonical anti-commutation relations are
{b++ (σ, τ ), c+ (σ 0 , τ )} = 2πδ(σ − σ 0 ) ,
{b−− (σ, τ ), c− (σ 0 , τ )} = 2πδ(σ − σ 0 ) .
For the closed string case the Fourier mode expansions look as
+∞
X
+
c (σ, τ ) = c̄n e−in(τ +σ) ,
n=−∞
+∞
X
c− (σ, τ ) = cn e−in(τ −σ) ,
n=−∞
+∞
X
b++ (σ, τ ) = b̄n e−in(τ +σ) ,
n=−∞
+∞
X
b−− (σ, τ ) = bn e−in(τ −σ) .
n=−∞
For the anti-commutation relations this becomes

{bm , cn } = δm+n ,
{bm , bn } = {cm , cn } = 0
and the same for the barred oscillators. The Virasoro generators are
∞
X
Lgh
m = (m − n) : bm+n c−n :
n=−∞
X∞
L̄gh
m = (m − n) : b̄m+n c̄−n :
n=−∞
– 71 –
The ghosts and anti-ghosts are conformal fields. Indeed, it is easy to compute
[Lgh
m , bn ] = (m − n)bm+n , [Lgh
m , cn ] = −(2m + n)cm+n .
Comparing this with the transformation rule of the modes of a conformal operator
of dimension ∆
¡ ¢
[Lm , An ] = m(∆ − 1) − n Am+n
we conclude that b and c are indeed the conformal fields of the conformal dimension
∆ = 2 and ∆ = −1 respectively.
Using the explicit expressions for the ghost generators Lgh it is not difficult to
compute the algebra
gh
[Lgh gh gh
m , Ln ] = (m − n)Lm+n + c (m)δm+n ,
where the central charge appears to be

1
cgh (m) = (2m − 26m3 ) .
12
Now if we introduce the total Virasoro generator as
Lm = LX gh
m + Lm − aδm,0
then it will satisfy the Virasoro algebra
[Lm , Ln ] = (m − n)Lm+n + c(m)δm+n
with
d 1
c= (m3 − m) + (2m − 26m3 ) + 2am .
12 12
We see that the total central charge vanishes for d = 26 and a = 1. We again
found the same values for the critical dimension and the normal-ordering constant
as followed from the light-cone approach! Here these conditions on the theory follow
from the requirement of vanishing of the total central charge.
BRST operator
The concept of the BRST operator is very general. In fact, the BRST operator can
be associated to any Lie algebra and it is a useful tool to compute the Lie algebra
cohomologies.
Consider a Lie algebra with generators Ki satisfying the relations
[Ki , Kj ] = fijk Kk .
Introduce ghost and anti-ghost fields ci and bi satisfying the anti-commutation rela-
tions
{ci , bj } = δji , i = 1, . . . , dimK
– 72 –
Introduce the ghost number operator U :
X
U= ci bi .
i
The eigenvalues of this operator are integers ranging from 0 up to dimK.
The BRST operator is defined as
Q = ci Ki − 12 fijk ci cj bk . (4.18)
First we compute a commutator
[Q, U ] = [ci Ki − 12 fijk ci cj bk , cm bm ]

= −cm {ci , bm }Ki − cm 12 fijk {ci cj , bm }bk − 12 fijk ci cj {bk , cm }bm
= −ci Ki + fijk ci cj bk − 21 fijk ci cj bk = −Q .
Thus, the BRST operator has the following commutator with U :
[U, Q] = Q
and as the result it increases the ghost number by one:
U Q|χi = (QU + Q)|χi = (Ngh + 1)Q|χi .
Second, compute the anticommutator
{Q, Q} = {ci Ki − 12 fijk ci cj bk , cs Ks − 12 fmn

p
cm cn bp }
= ci cs Ki Ks − cs ci Ks Ki − 12 fijk ci cj {bk , cs }Ks − 12 fmn
p
cm cn {ci , bp }Ki
+ 14 fijk fmn
p
{ci cj bk , cm cn bp } .
It is easy to find
p
fijk fmn
p
{ci cj bk , cm cn bp } = 4fijk fkm ci cj cm bp .
Therefore, the expression we are interested in reduces to
p
{Q, Q} = ci cj [Ki , Kj ] − 12 fijk ci cj Kk − 12 fijk ci cj Kk + fijk fkm ci cj cm bp .
Due to the algebra relation [Ki , Kj ] = fijk Kk the first three terms in the last expression
cancel out and we are left with
p
{Q, Q} = fijk fkm ci cj cm bp .
Due to the anti-commuting property of the ghosts the last expression can be rewritten
as
p p p
{Q, Q} = 13 (fijk fkm k
+ fmi fkj k
+ fjm fki )ci cj cm bp .
– 73 –
The expression in the bracket vanishes because this is the Jacobi identity written for
structure constants of the Lie algebra
p p p
fijk fkm k
+ fmi fkj k
+ fjm fki = 0.
Thus, we found that the BRST operator is nilpotent, i.e. it its square is zero
Q2 = 12 {Q, Q} = 0 .
This is the fundamental property of the BRST operator. We also assume that Q is
hermitian, i.e. Q† = Q.
Let Hk will be a Hilbert space of states with fixed ghost number U = k. An element
|χi ∈ Hk is called BRST-invariant if
Qχ = 0 (4.19)
Clearly, any state of the form Q|λi, where |λi is any state with the ghost number14
k − 1, is BRST-invariant because
Q(Q|λi) = Q2 |λi = 0 .
The state Q|λi has zero norm because
hλ|Q† Q|λi = hλ|Q2 |λi = 0 .
The most important BRST-invariant states are those which can not be written in
the form |χi = Q|λi. We will regard two solutions of equation (4.19) equivalent if
|χ0 i − |χi = Q|λi
for some λ.
In fact, we recognize that the BRST-operator mimics all the properties of the de-Rahm operator d which acts on the space of external
(differential) forms on a manifold M. Indeed, it has a property that d2 = 0. A differential form ω is called closed if dω = 0 and it is
called exact if there is another form θ such that ω = dθ. The factor-space of all closed forms over all exact forms of a given degree n
n closed forms
H (M) =
exact forms
is called n-th cohomology group of the manifold M. In our present case the operator Q takes values in the Lie algebra and it defines
cohomologies with values in an given representation of the Lie algebra.
Furthermore, the states with zero ghost charge are of special importance. Such
a state must be annihilated by all bk . For such states the BRST operator reduces to
Q|χi = ci Ki |χi = 0 .
14
As we found the BRST-operator increases the ghost number by one.
– 74 –
Thus, Q|χi = 0 is equivalent to the condition that Ki |χi = 0, i.e. it is invariant under
the action of the Lie algebra. On the other hand, this state cannot be represented as
|χi = Q|λi for some |λi because the ghost number of |λi should be equal −1 which
is impossible.
Let us apply this general construction to the string case. The Lie algebra in this
case is the Virasoro algebra and we supply it with the ghosts cm and ant-ghosts bm .
The BRST operator is now
+∞
X ∞
1X
Q= LX
−m cm − (m − n) : c−m c−n bm+n : −ac0 ,
−∞
2 −∞
where a is the normal-ordering ambiguity constant for L0 . It turns out that this
expression can be written as
+∞ ³
X ´
1 gh
Q= : LX
−m + L
2 −m
− aδm,0 cm :
−∞
The ghost-number operator is

+∞
X
U= : c−m bm :
−∞
We would like to investigate the fulfilment of the relation Q2 = 0 in quantum theory.

We find
1
+∞
X ³ ´
Q2 = {Q, Q} = [Lm , Ln ] − (m − n)Lm+n c−m c−n .
2 n,m=−∞
Here Lm = LX gh 2
m + Lm − aδm,0 is a total Virasoro operator. Thus, Q = 0 for d = 26
and a = 1 as the consequence of vanishing of the total central charge!
Inverse statement is also true: from Q2 = 0 it follows that the central charge of
the Virasoro algebra vanishes. Indeed, we first note that
Lm = {Q, bm }
From here
[Lm , Q] = [{Q, bm }, Q] = (Qbm + bm Q)Q − Q(Qbm + bm Q) = [bm , Q2 ] = 0
as Q2 = 0. Therefore, we see that
[Lm , Ln ] = [Lm , {Q, bn }] = {[Lm , Q], bn }

| {z }
=0
+ {Q, [Lm , bn ]} = (m − n){Q, bm+n } = (m − n)bm+n .
– 75 –
One can check that these objects are charges which correspond to new conserved
currents
gh
J+B = 2c+ (T++
X
+ 12 T++ )
J+ = c+ b++ .
There is new and more fundamental fermionic symmetry present – it is BRST

symmetry. Let λ be a grassman (anticomuting) parameter. The BRST transforma-
tion is defined as
δY = [λQ, Y ]
It is given by
δX µ = λc+ ∂+ X µ + λc− ∂− X µ ,
δc+ = λc+ ∂+ c+ ,
δb++ = 2iλT++ ,
δT++ = 0 .
The ghost number operator

X∞
1
U = (c0 b0 − b0 c0 ) + (c−n bn − b−n cn )
2 n=1
Here cn , bn for n > 0 are annihilation operators.

Zero-modes require special treatment. We have
c20 = b20 = 0 , {c0 , b0 } = 1
There is a two-dimensional representation of this relations:
c0 | ↓i = | ↑i , b0 | ↑i = | ↓i
c0 | ↑i = 0 , b0 | ↓i = 0 .
The ghost numbers are U↓ = −1/2 and U↑ = 1/2.
Physical states should have the ghost number −1/2. They are annihilated by b0 .
Indeed, consider
cn |χi = bn |χi = 0 , n > 0 and b0 |χi = 0 .
The condition of the BRST invariance reduces to
³ X ´
0 = Q|χi = c0 (L0 − 1) + c−n Ln |χi .
n>0
We thus reproduced the conditions for a physical state obtained in the old covariant
quantization approach. Physical states of bosonic string are cohomology classes of
the BRST operator with the ghost number −1/2.
– 76 –
5. Geometry and topology of string world-sheet
5.1 From Lorentzian to Euclidean world-sheets
String world-sheets and the target space both have Lorentzian signature. The con-
nection to the theory of Riemann surfaces can be made by performing the Wick
rotation of both world-sheet and the target space metrics. In particular, for the
world-sheet time τ this means τ → −iτ . Thus, on the string-world sheet this Eu-
clidean continuation corresponds to
σ ± = τ ± σ → −i(τ ± iσ) . (5.1)
A word of caution is needed here: One cannot really prove that the theory of the
Lorentzian world-sheets is equivalent to the theory of the Euclidean ones. However,
treating the world-sheets as Euclidean provides by itself a consistent theory of inter-
acting strings. The Euclidean version can be then used to learn as much as possible
about its Lorentzian counterpart.
We start with considering a single closed string whose world-sheet has topology
of a cylinder. Formula (5.1) suggests to introduce the complex coordinates
w = τ + iσ , w̄ = τ − iσ .
The Euclidean closed string covers only a finite interval of σ on the complex plane
0 ≤ σ ≤ 2π and, therefore, only a strip of the 2dim plane.
Fig. 5. Mapping the cylinder (τ, σ) to a strip (w, w̄) on the complex w-plane.
One can further map the strip to the whole complex plane by using the conformal
map
z = ew = eτ +iσ .
Lines of constant τ are mapped into circles on the z plane and the operation of time
translation τ → τ + a becomes the dilatation
z → ea z .
A procedure of identifying dilatations with the Hamiltonian and circles about the
origin with equal-time surfaces is called sometimes radial quantization. Mapping
– 77 –
from the cylinder to the plane cannot change the physical content of the theory
if the theory is conformally invariant, which is the case of string theory in critical
dimension.
We note that the plane is a non-compact manifold. However, one can compactify
it by adding a point at infinity. The corresponding compact surface arising in this
way is the Riemann sphere. A metric on a plane can be transformed to a metric on
a sphere by a suitable choice of the conformal prefactor. For instance one can pick
up the metric
4dzdz̄
ds2 =
(1 + |z|2 )2
The formula z = cot 2θ eiφ defines a stereographic projection of the sphere onto the
plane and under this projection the metric takes the form
ds2 = dθ2 + sin2 θdφ2 ,
i.e. it is the standard round metric on a sphere.
north

pole(outgoing string)

z−plane south pole (incoming string)
Fig. 6. Stereographic projection of a sphere onto z-plane. Asymptotic

incoming and outgoing strings are mapped to the south and the north poles
of the sphere respectively.
Since cot π2 = 0 and cot(0) = ∞ the incoming and outgoing strings are mapped to
the south and the north poles of the sphere respectively.
The example above can be generalized to general world-sheets corresponding
to interacting strings. The crucial observation is that conformal invariance allows
to consider compact world-sheets instead of surfaces with boundaries corresponding
incoming and outgoing strings. The string boundaries are mapped to punctures on a
compact Riemann surface.
Under the Euclidean continuation the basic equation ¤X = 0 transforms into

∂z ∂z̄ X = 0
with a general solution
X(z, z̄) = X(z) + X̄(z̄) ,
i.e. the left and right-moving excitation correspond now to analytic and anti-analytic
fields on the complex z-plane.
– 78 –
5.2 Riemann surfaces
On a two-dimensional real manifold M the metric can be locally (i.e. in a given
coordinate chart) defined by the line element
ds2 = h11 dx2 + 2h12 dxdy + h22 dy 2
Introducing the complex coordinates
z = x + iy , z̄ = x − iy
the line element can be written as
ds2 = 2eφ |dz + µdz̄|2
with the following identifications

³ q ´
φ 1 2 h11 − h22 + 2ih12
e = 8 h11 + h22 + 2 h11 h22 − h12 , µ= p .
h11 + h22 + 2 h11 h22 − h212
If h11 = h12 and h12 = 0 then µ = 0 and the metric takes in this coordinate chart
the form
ds2 = 2eφ |dz|2 = 2eφ (dx2 + dy 2 ) .
The corresponding coordinate system is called isothermal or conformal and the co-
ordinates (x, y) define a conformal map of a coordinate chart of a manifold to the
Euclidean plane.
A theorem of Gauss
For any real two-dimensional orientable surface with a positive definite metric there
always exists a system of isothermal coordinates (the theorem of Gauss). It is unique
up to conformal transformations. First, assume that we have already found a system
of isothermal coordinates, i.e. the metric is locally in the form
ds2 = 2eφ |dz|2 .
Performing the coordinate transformation with an analytic function of z:
z → f (z)
we get
ds2 → ds2 = 2eφ |f (z)|2 |dz|2 ,
i.e. we get a conformally equivalent metric and, therefore, a new system of the
isothermal coordinates.
– 79 –
coordinate chart

Fig. 6. Covering the Riemann surface with coordinate patches. Every patch
is homeomorphic to an open domain of the Euclidean plane.
Second, in order to prove that isothermal coordinates exist, consider the so-called
Beltrami equation
∂w ∂w
= µ(z, z̄) .
∂ z̄ ∂z
Suppose we solve this equation, then
|∂z w|2 2
|dw|2 = |∂z w + ∂z̄ w|2 = |∂z w|2 |dz + µdz̄|2 = ds .
2eφ
Thus,
2eφ
ds2 = |dw|2 ≡ λ|dw|2 ,
|∂z w|2
i.e. w defines a system of isothermal coordinates. It is a mathematical theorem that
for a sufficiently small coordinate patch and a differentiable metric a solution of the
Beltrami equation with ∂z w 6= 0 always exists.
If two metrics are related by a diffeomorphism and a Weyl rescaling they are said
to define the same conformal structure. If a manifold M is covered by a system of
conformal (isothermal) coordinate patches Uα , then on the overlaps the metrics are
conformally related, i.e. the transition functions on the overlaps Uα ∩ Uβ are analytic
and the complex coordinates are globally defined. A system of analytic coordinate
patches is called a complex structure and it is the same as a conformal structure.
A two-dimensional topological manifold endowed with a complex structure is

called a Riemann surface. Thus, a Riemann surface is a complex manifold.
Another way to understand that Riemann surface is a complex manifold is to

note that in two dimensions the metric provides a globally-defined integrable complex
structure √
I αβ = hhαγ ²γβ
such that I 2 = −1. The conformal structure is conformally-invariant and globally
well-defined. The existence of an integrable complex structure is necessary and suf-
ficient for an even-dimensional manifold to be complex.
– 80 –
Recall the fundamental result from the theory of two-dimensional (real) manifolds.
Any compact orientable connected two-dimensional manifold is homeomorphic to
a sphere with handles. The number of handles, g, is called the genus, a topolog-
ical invariant. Thus, every compact Riemann surface, being a compact orientable
connected one-dimensional manifold, has an associated genus.
Gauss-Bonnet theorem
Let M is a compact orientable two-dimensional manifold of genus g with a metric

hαβ . Then
Z √
1
hR = χ(M) = 2 − 2g (5.2)
4π M
is a topological invariant and it coincides with the Euler characteristic
χ(M) = 2 − 2g .
Here we briefly discuss one way to prove the Gauss-Bonnet theorem. Recall that
in two dimensions the Riemann tensor has only one independent component which
is the scalar curvature:
R
Rαβγδ = (hαγ hβδ − hαδ hβγ ) .
2
Let us choose a system of isothermal coordinates. In this coordinate system the line
element is ds2 = 2eφ dzdz̄ and, therefore, the metric has two components hzz̄ = hz̄z =
eφ . The Riemann tensor simplifies to
1 ¯ .
Rzz̄zz̄ = −hzz̄ Rzz̄ = − (hzz̄ )2 R = eφ ∂ ∂φ
2
The scalar curvature is
√
¯
R = −2e−φ ∂ ∂φ =⇒ ¯ = −4φ ,
hR = −4∂ ∂φ
where 4 is two-dimensional Laplacian (the Euclidean analogue of the ¤ operator).

It is also useful to rewrite the integration measure in the complex coordinates
i
dx ∧ dy = dz ∧ dz̄ ,
2
i.e. we get
µ ¶
√ 2 ∂ 2φ ∂ 2φ ∂ ∂φ
hR d x = −4 dx ∧ dy = −2i dz ∧ dz̄ = 2i dz̄ dz .
∂z∂ z̄ ∂z∂ z̄ ∂ z̄ ∂z
– 81 –
The last formula can be understood in the sense of calculus of exterior differential
forms ∂f = ∂f and ∂f¯ = ∂f . In particular, the de Rahm operator d = dx ∂ + dy ∂
∂z ∂ z̄ ∂x ∂y
can be written as
∂ ∂
d = ∂ + ∂¯ ≡ dz + dz̄ .
∂z ∂ z̄
¯ ¯
Also one has ∂∂ = ∂ ∂ = 0. Therefore, we can rewrite our formula as
√
hR = 2id(∂φ) .
√
This shows that hR is locally a total derivative and therefore, we see again that
eq.(5.2) cannot change under smooth variations of the metric, i.e. it is a topological
invariant.
On a compact Riemann surface M we consider an abelian differential Ω, which is
a meromorphic differential form. It means that in a given coordinate patch (Uα , zα )
it can be written in the form Ω = fα dzα , where fα is a meromorphic function15 .
We can also suppose that the coordinate patches are chosen in such a way that
every Uα contains at most one pole or one zero of Ω. In a patch Uα the metric is
ds2 = 2eφα |dzα |2 . Thus, on the intersections of the patches Uα ∩ Uβ we have
eφα ¯ f ¯2
¯ α¯
φ
=¯ ¯ .
e β fβ
Thus, there exists a globally defined function
eφα
ϕ= on Uα for all α .
|fα |2
This function is smooth except for singularities at zeros and poles of Ω. Since log |fα |2
is harmonic outside zeros and poles of Ω we have
√
hR = 2id(∂ log ϕ) .
Let M² = M − ∪Dk,² , where Dk,² are small disks around the singularities of Ω.
Then, by Stokes’s theorem we have
Z √ Z X Z
hR = 2i lim d(∂ log ϕ) = −2i lim ∂ log ϕ .
M ²→0 M² ²→0 ∂Dk,²
k
To evaluate the integrals over the circles we note that at a zero or pole of Ω the
function ϕ is of the form ϕ = ψ/|z|2m with a smooth function ψ without zeros and
m is the order of zero or pole (m < 0 in the latter case). Therefore,
Z Z Z
−2m dz
lim ∂ log ϕ = lim ∂ log |z| = −m lim = −2πim .
²→0 ∂D ²→0 |z|=² ²→0 |z|=² z
k,²
15
A function f (z) is called meromorphic if it does not have any other singularities except poles.
– 82 –
Summing up we obtain Z √
hR = −4π deg Ω ,
M
where deg Ω is defined as the difference of the number of zeros and the number of
poles of Ω:
deg Ω = # zeros − # poles .
It is now the Poincaré-Hopf theorem that states that deg Ω = 2g − 2, where g is the
genus of the Riemann surface. Thus, the Gauss-Bonnet theorem follows from the
Poincaré-Hopf theorem.
Finally we note that due to the identity
²αβ ²γδ = hαγ hβδ − hαδ hβγ
the Euler characteristic can be rewritten in the form16

Z
1
χ(M) = ²αβ ²γδ Rαβγδ .
8π M
g1 g 2
g1+ g 2 g=0
Fig. 7. If there are two surfaces of genera g1 and g2 then by removing from
each surface a half-sphere we can glue the resulting surfaces into a surface
of genus g1 + g2 and the Riemann sphere of genus zero.
To illustrate the Gauss-Bonnet theorem, we compute the topological invariant

eq.(5.2) for a sphere. We will take a model of a sphere which represent it as the
complex plane (including the point at infinity) with the metric
4dzdz̄
ds2 = = 2eφ dzdz̄ ,
(1 + |z|2 )2
16
This is a non-trivial characteristic class of the tangent bundle to M known as the Euler class.
– 83 –
i.e the conformal factor φ and the action of Laplacian on it are
8
φ = ln 2 − 2 ln(1 + |z|2 ) =⇒ − 4φ = .
(1 + |z|2 )2
Therefore, Z Z
1 √ 1 ∞
8
dxdy hR = 2πrdr = 2,
4π C 4π 0 (1 + r2 )2
which is indeed the Euler characteristic for sphere, a compact orientable manifold
with genus g = 0. It is also known that on a torus, a manifold of genus g = 1, there
exists the globally defined flat metric. Therefore, the Euler characteristic of torus
is χ = 0. In fact, the Euler characteristic χ(g) for a Riemann surface of arbitrary
genus g can be found by using the recurrent formula
χ(g1 + g2 ) = χ(g1 ) + χ(g2 ) − χ(0) = χ(g1 ) + χ(g2 ) − 2
and the fact that χ(1) = 0. This again leads to the formula χ(g) = 2 − 2g.
5.3 Moduli space

The moduli space of all the metrics is the same as the moduli space of Riemann
surfaces and it is defined as the space of all metrics devided by diffeomorphisms and
Weyl rescalings
all metrics
Mg = .
diffeomorphisms × Weyl rescalings
The moduli space is finite-dimensional and it is parametrized by a finite number of

complex parameters τi called moduli. The dimension of the moduli space is another
topological invariant and it depends on the genus g only.
Complex geometry
Since we have a system of well-defined complex coordinates on a Riemann surface

we can consider general tensors
V z...zz̄...z̄ z...zz̄...z̄ (z, z̄) ,
in particular, V z ∂z and V z̄ ∂z̄ are vector fields and Vz dz and Vz̄ dz̄ are one-forms. All
these tensors are one component objects. The metric hzz̄ and hzz̄ can be used to
convert all z̄ indices into z-indices. Tensors with one type of indices (for example,
z indices) are called holomorphic. Holomorphic tensors which depend on z variable
– 84 –
only are called analytic. A holomorphic tensor with p lower and q upper indices has,
by definition, the rank n = p − q:
z| .{z
. . z}
q
V . . z} (z, z̄)
z| .{z ⇐ holomorphic tensor of rank n = p − q;
p
V z...z z...z (z) ⇐ analytic tensor;
Under analytic coordinate transformations z → f (z) a holomorphic tensor of rank n

transforms as follows
µ ¶n
∂f (z)
V (z, z̄) → V (f (z), f¯(z̄)) .
∂z
Denote by V (n) the space of all holomorphic tensors of rank n. This space can be
supplied with the scalar product
Z √ ³ ´∗
(n) (n) (n) (n) (n) (n)
(V1 |V2 ) = d2 z h (hzz̄ )n V1 V2 , V1 , V2 ∈ V (n)
and the associated norm ||V (n) ||2 = (V n |V (n) ). This scalar product is Weyl-invariant
for n = 1 only.
In the analytical coordinate system the Christoffel connection has only two non-
vanishing components
Γzzz = ∂φ , ¯ .
Γz̄z̄z̄ = ∂φ
This connection allows to define two covariant derivatives
∇(n)
z : V (n) → V (n+1) , ∇(n)
z T
(n)
(z, z̄) = (∂ − n∂φ)T (n) (z, z̄)
∇z(n) : V (n) → V (n−1) , ¯ (n) (z, z̄) .
∇z(n) T (n) (z, z̄) = hzz̄ ∇z̄ T (n) (z, z̄) = hzz̄ ∂T
These two differential operators defined on holomorphic tensors of a fixed rank n

commute with the analytic coordinate transformations z → f (z). One can compute
(n)
an adjoint of ∇z and find that
(∇(n) † z
z ) = −∇(n+1) .
The elements of the complex geometry we introduced above allows one to obtain
some information about the moduli space. Consider an arbitrary infinitesimal change
of the metric
X ∂
δhαβ = Λhαβ + ∇α Vβ + ∇β Vα + δτi hαβ .
| {z } | {z } ∂τi
i
Weyl diff | {z }
moduli
– 85 –
The last term here reflects the dependence of the metric on moduli which cannot be
compensated by diffeomorphisms and Weyl rescalings. We can split this variation
into trace and traceless parts
³ 1 γδ X ∂ ´
δhtrace
αβ = Λ + ∇ γ
Vγ + h δτ i hγδ hαβ
2 i
∂τ i
X ³ ∂ 1 ´
γδ ∂
δhtraceless = ∇ V + ∇ V − h ∇ γ
V + δτ h − h h h γδ .
αβ
|α β β α
{z αβ
}γ i
∂τ i
αβ
2
αβ
∂τ i
i
Operator P
We see that we can always shift the Weyl rescaling parameter Λ as

1 X ∂
Λ → Λ − ∇γ Vγ − hγδ δτi hγδ
2 i
∂τ i
so that the trace part of the variation will transform as
δhtrace
αβ = Λhαβ .
In the complex coordinates we have hzz = hz̄z̄ = 0 and therefore rewriting the
variation formulae in these coordinates we find
δhzz̄ = Λhzz̄ ,
X
δhzz = 2∇(1)
z Vz + δτ i µi zz ,
| {z }
Operator P i
where µi zz = ∂τi hzz = hzz̄ µiz̄

z . We see, in particular, that an infinitesimal change of
(1)
hzz̄ can always be written as a Weyl rescaling. Also the covariant derivative ∇z
introduced above should be naturally identified with the operator P . Finally, we
note that decomposition into sum of two terms
X
δhzz = 2∇(1)
z V z + δτ i µi zz
i
is not orthogonal w.r.t. to the scalar product we introduced above. Denote by φizz a
(1)
basis of the orthogonal complement of ∇z :
(φizz |∇(1) z i
z Vz ) = −(∇(2) φzz |Vz ) for any Vz ∈ V (1) .
The last equation is equivalent to
∇z(2) φizz = 0 =⇒ ¯ i = 0.
∂φ zz
Thus, the kernel (or, in other words, the space of zero modes) of the operator P † =
∇z(2) consists of global analytic tensors of the second rank. Such tensors are of special
importance and they are called quadratic differentials. Thus, the dimension of the
– 86 –
moduli space is equal to the number of lineally independent quadratic differentials
on a given Riemann surface of genus g. The theory of quadratic differentials was
developed by Kurt Strebel (I will add more on Strebel theory in due course).
(1)
The kernel of the operator ∇z is spanned by vectors from V (1) :
∇(1) z̄
z Vz = hz z̄ ∂V = 0 =⇒ ∂V z̄ = 0 .
(1)
The globally defined vector fields which span a kernel of ∇z are called conformal
Killing vectors. They generate conformal Killing group (or the group of conformal
isometries, i.e. globally defined diffeomorphisms which can be completely absorbed
by the Weyl rescalings).
Riemann-Roch theorem
An important question of how many moduli for a Riemann surface of genus g exists
(n)
is answered by the Riemann-Roch theorem. Define the index of ∇z as the number of
its zero modes minus the number of zero modes of its adjoint ∇z(n+1) . The Riemann-
Roch theorem states that
1
ind∇(n) (n) z
z = dim ker∇z − dim ker∇(n+1) = −(2n + 1)(g − 1) = (2n + 1)χg .
2
For n = 1 we therefore have
#complex moduli − #conformal Killing vectors = 3g − 3
One can find the number of conformal Killing vectors for a compact Riemann
surface in an independent way. These are globally defined analytic vector fields whose
norm is finite Z √
2
||V || = hhzz̄ V z V z̄ < ∞ .
Mg
4dzdz̄
Here V z = Vn z n . For the case of sphere with metric ds2 = (1+|z|2 )2
one finds that
there exists three independent conformal Killing vectors
∂z , z∂z , z 2 ∂z .
Indeed, for the norm we have

 8π
Z ∞  3 k=0
8
||V ||2 = 2π rdr r2k = 4π3
k=1
0 (1 + r2 )4  8π
3
k=2
where k = 0, 1, 2 for the three vector fields in question. We see that if k ≥ 3 the
integral becomes divergent, therefore, for instance, the field z 3 ∂z has an infinite norm,
– 87 –
i.e. it is not normalizable. The three conformal Killing vectors are well behaved at
∞ as can be seen by making conformal transformation w = 1/z under which
−w2 ∂w , − w∂w − ∂w .
The limit z → ∞ corresponds now to w → 0 and we see that these fields are well-
behaved at this point17 . The three conformal Killing vectors we found correspond to
the Virasoro generators L0 and L±1 and they span (over the complex field) the Lie
algebra sl(2, C). Therefore, sl(2, C) is the Lie algebra of analytic globally defined
maps of the Riemann sphere onto itself.
Thus, the Riemann sphere has three conformal Killing vectors and, according to
the Riemann-Roch theorem, no moduli. That means that all metrics on the sphere
are conformally equivalent, or, in other words, there is a unique Riemann surface of
genus zero.
One can count the number of conformal Killing vectors for higher genus Riemann
surfaces as well. To this end one has to use the Ricci identity
1
∇z(n+1) ∇(n) (n−1) z
z − ∇z ∇(n) = nR .
2
(n)
Let V (n) ∈ ker∇z . Then we have
(n) (n) (n) (n) (n) z (n) (n)
0 = (∇z V |∇z V ) = −(V |∇(n+1) ∇z V )=
1 (n) z (n) (n) 1 (n) z (n) (n)
= − (V |∇(n+1) ∇z V ) − (V |∇(n+1) ∇z V )
2 2
1 (n) (n−1) z 1 (n)1 z (n) (n)
= − (V |∇z ∇(n)
+ nR) − (V |∇(n+1) ∇z V )
2 2 2
1h (n) (n) (n) (n) z (n) z (n) 1 (n) (n)
i
= (∇z V |∇z V ) + (∇(n) V |∇(n) V ) − nR(V |V ) .
2 2
(n)
Therefore, for any vector from the kernel of ∇z the following equality is valid
1
(∇(n)
z V
(n)
|∇z(n) V (n) ) + (∇z(n) V (n) |∇z(n) V (n) ) − nR(V (n) |V (n) ) = 0 .
| {z } | {z } 2
non−negative non−negative
Consider the case of a torus g = 1. On a torus there is a globally defined flat metric
ds2 = dzdz̄ which gives R = 0. Therefore, the equality above leads to two equations
¯ (n) = 0 ,
∂V (n) = ∂V
(n)
i.e. V (n) = const and, therefore, dim ker∇z = 1. Thus, there is a unique generator
of conformal isometries, it corresponds to the rigid U(1)×U(1) rotations of the torus.
The Riemann-Roch theorem gives for g = 1
#complex moduli − #conformal Killing vectors = 3 − 3 = 0 ,
| {z }
=1
17
According to our general discussion of Riemann surfaces the sphere requires at least two coor-
dinate patches to make an atlas. Transformation from one patch to another is analytic and is given
by w = 1/z.
– 88 –
i.e. the torus is characterized by one complex modulus τ .
For g ≥ 2 there is theorem that states that the corresponding manifold always admits
a metric with constant negative curvature. For n 6= 0 this means that − 21 nR > 0 and,
therefore, dim ker∇z(n) = 0. The Riemann surfaces with g ≥ 2 have no conformal
isometries and by the Riemann-Roch theorem that means that the number of complex
moduli n = 1 is 3g − 3.
(0)
For n = 0 we have dim ker∇z = 1, because the corresponding kernel is spanned by
constants. The Riemann-Roch theorem implies then that
dim ker∇z(1) − dim ker∇(0) = (2n + 1)(g − 1) = g − 1 ,

| {z z } | {z }
=1 n=0
i.e. dim ker∇z(1) = g. The kernel of ∇z(1) is spanned by one forms
¯ z=0
∇z(1) ωz = hzz̄ ∂ω =⇒ ¯ z = 0.
∂ω
Thus we arrive at another important consequence of the Riemann-Roch theorem:

on a Riemann surface of the genus g there exists precisely g linearly independent
(globally defined) analytic one-forms. These analytic differential forms are called
abelian differentials of the first kind.
The information we obtained by using the Riemann-Roch theorem is summarized

in the Table below.
(n)
g dim ker∇z dim ker∇z(n+1)
0 2n + 1 0
1 1 1
>1 1 for n = 0 g
0 for n > 0 (2n + 1)(g − 1)
Moduli space of tori
Here we would like to look more closely at the moduli space which describes confor-
mally non-equivalent tori – the Riemann surfaces of the genus g = 1. The torus can
be obtained by performing the following identification on the complex plane
z ≡ z + nλ1 + mλ2 , n, m ∈ Z , λ1 , λ2 ∈ C .
The parameters λ1,2 are subject to conformal transformations z → λz and, therefore,

only their ratio τ = λλ21 is scale-invariant. By using this freedom (the U(1)-rotation
+ real rescaling) one can always bring the parallelogram defining the torus upon
– 89 –
gluing the opposite sides to the canonical form depicted on Fig.9, which corresponds
to Imτ > 0.

Fig. 8. Defining the torus by factorizing the complex z-plane.
The parameter τ takes values in the upper half-plane which is called Teichmüller
space. The parameter τ itself is named the modular or Teichmüller parameter. The
Teichmüller parameter is not yet a parameter describing the moduli space.
+1
0 1
Fig. 9. Canonical representation of the torus by parameter τ taking values
in the Techmüller space which is identified with the upper half-plane.
The reason is that there are global diffeomorphisms which are not smoothly connected
to the identity; they leave the torus invariant but they act non-trivially on the
Teichmüller parameter. They correspond to the so-called Dehn twists
• λ1 → λ1 , λ2 → λ1 + λ2 , which gives τ → τ + 1;
τ
• λ1 → λ1 + λ2 , λ2 → λ2 , which gives τ → τ +1
;
It turns out that these two transformations generate the group SL(2, Z). It is a group
of matrices µ ¶
ab
,
cd
– 90 –
where a, b, c, d ∈ Z and ad − bc = 1. The action on the modular parameter τ is in
the form of the fractional-linear transformation
aτ + b
τ → τ0 = .
cτ + d
One can check that these transformations preserve the area of the parallelogram.
Since two SL(2, Z)-matrices
µ ¶ µ ¶
ab ab
+ −
cd cd
act on τ in the same way, the modular group of the torus is PSL(2, Z) = SL(2, Z)/Z2 .
i

8

Fixed points of S and ST

−1 −1/2 0 1/2 1
Fig. 8. Fundamental domain Mg=1 of the Teichimüller space describing the
moduli space of conformally non-equivalent tori.
The moduli space of conformally non-equivalent tori is then the quotient of the
Teichmüller space of the modular group
Teichmüller space
Mg=1 = .
modular group
One usually uses the following generators of the modular group
1
T : τ → τ + 1, S: τ →− .
τ
Any element of SL(2, Z) is a composition of a certain number of S and T generators:
SST ST T T ST....SST
– 91 –
Any point in the upper-half plane is related by a modular transformation to a point
in the so-called fundamental domain F ≡ Mg=1 of the modular group. It can be
chosen as
n 1 1 o
2 2
Mg=1 = − ≤ Reτ ≤ 0, |τ | ≥ 1 ∪ 0 < Reτ < , |τ | > 1 .
2 2
The modular group does not act freely on the upper-half plane because some of the
modular transformations have fixed
√
points. The point τ = i is the fixed point of S:
1 2πi 1 3
τ → − τ and τ = e 3 = − 2 + i 2 is the fixed point of ST . The existence of the
fixed points implies that Mg=1 is not a smooth manifold, rather it has singularities
of the orbifold type.
Since it does not matter which fundamental region to choose in order to integrate
over in the one-loop path integral the integrand the corresponding string amplitude
should be invariant under modular transformations. The requirement of the modular
invariance is one of the most important principles of string theory. In particular, it
leads to strong restrictions on the possible gauge groups for the heterotic string.
6. Classical fermionic superstring
Introduction of the world-sheet fermionic degrees of freedom requires understanding

of how spinors on curved manifolds (world-sheets) are defined. The discussion in the
next paragraph is general and can be applied to a manifold of an arbitrary dimension
d.
6.1 Spinors in General Relativity
It is not straightforward to introduce spinors in General Relativity. If we have a

i ...i
tensor field Tj11...jqp of rank (p, q) on a manifold M then under general coordinate
transformations of the coordinates xi on M: xi → x0i (xj ), this field transforms as
follows
0k ...k ∂x0k1 ∂x0kp ∂xj1 ∂xjq i1 ...ip
1
Tl1 ...l p
(x0 ) = · · · · · · T (x) .
q
∂xi1 ∂xip ∂x0l1 ∂x0lq j1 ...jq
0i
Here tensor indices are acted with the matrices ∂x ∂xj
which form a group GL(d, R).
This is a group of all invertible real d × d matrices. This group does not have spinor
representations.
– 92 –

e

e

e
d 4

3

x

e

2

e
1

M
Tangent plane at a point x
Fig. 9. Vielbien eaα is a collection of d orthogonal vectors forming a basis

of the tangent space at any point x on the d-dimenional manifold. In string
theory M with d = 2 is the string world-sheet and x ≡ (σ, τ ).
On the other hand, spinors are objects which transform under the spinor repre-
sentations of the Lorentz group. The Lorentz group in d-dimensions is SO(d − 1, 1)
and it is not GL(d, R). At each point of the manifold there is an inertial frame. In
this inertial frame the Lorentz transformations are well defined. One can think that
the Lorentz transformations act in the flat Minkowski space tangent to the manifold
M at any given point x. In the tangent plane we introduce a basis eaα (x), a = 1, . . . , d
of orthonormal vectors:
(ea , eb ) ≡ hαβ eaα ebβ = η ab ,
where η αβ is the flat Minkowski metric. In fact, eaα is an invertible d × d x-dependent

matrix which is called vielbein. The index α is called “curved” and its acted by
the general coordinate transformations (diffeomorphisms) as the usual vector index,
while the index a is called “flat” and its acted by the local (i.e. x-dependent) Lorentz
transformations as we will see in a moment. The inverse matrix is eαa and it obeys
eaα eαb = δba . Because of this relation, we also have that
ηab eaα ebβ = hαβ .
There is no a preferred choice of the basis in the tangent space and one orthonormal
set of tangent vectors can be transformed into the other by means of local Lorentz
transformations
eaα → Λab ebα .
Introduction of the vielbein in favour of the metric introduces additional degrees

of freedom. Indeed, the vielbein being d × d-matrix has d2 components, while the
metric hαβ has only d(d+1)
2
components in d dimensions. On the other hand, if we
require that the theory we consider has the local Lorentz symmetry, then there are
– 93 –
d(d−1)
2
local Lorentz transformations which should allow to remove the additional
(unphysical) components of the vielbein:
d(d − 1) d(d + 1)
d2 −
|{z} =
2 }
| {z 2 }
| {z
vielbein
local Lorentz metric
Thus, the vielbein brings new degrees of freedom, but they can be removed by a
new symmetry which is the local Lorentz transformations. On the other hand, the
vielbein allows one to introduce coupling of the gravitational degrees of freedom with
spinors.
Spinor representations
(Pseudo-)orthogonal groups (SO(d − 1, 1)) SO(d) in addition to the usual tensor

representations have also spinor representations. These are not single-valued but
rather double-valued representations of SO(d).
Consider, for instance, the group SO(3). This group has the so-called universal
covering group which is SU(2). The relation between them is as follows
SO(3) → SU(2)/Z2 .
The group SO(3) is not simply connected, while SU(2) is. Here
½µ ¶ µ ¶¾
10 −1 0
Z2 = ,
01 0 −1
is the center of SU(2), i.e. a set of matrices from SU(2) which commute with all other
SU(2)-matrices. Due to existence of the discrete center Z2 representations of SU(2)
split into two different classes of integer and half-integer spin. Only representations
with integer spin are those of (single-valid) SO(3). Representations of SU(2) with
half-integer spin are spinor (double-valid) representations of SO(3). Indeed, rotation
by the angle φ around the axes given by a fixed 3dim unit vector n = (n1 , n2 , n3 ),
n2i = 1, corresponds the transformation
³ ´ µ ¶
i cos φ2 − in3 sin φ2 − (in1 + n2 ) sin φ2
g(φ, n) = exp − φ ni σi = ,
2 (−in1 + n2 ) sin φ2 cos φ2 + in3 sin φ2
where σi are three Pauli matrices. One can easily verify that g(φ, n) ∈ SU(2).
Rotation by the angle φ = 2π is an identity transformation in SO(3) but it is not
the identity in SU(2). Indeed, one can see that
g(φ + 2π, n) = −g(φ, n) ,

g(φ + 4π, n) = g(φ, n) .
– 94 –
Another example is provided by the Lorentz group of the 4dim Minkowski space-
time, which is O(3, 1). The spinor representation of O(3, 1) is realized on the space
C4 , which is called a space of 4-component spinors.18 This representation looks as
³1 ´
g(ω) = exp γ ab ωab ,
2
where γ ab is the anti-symmetric product of the 4dim γ-matrices and ωab = −ωba are
parameters of the Lorentz transformation.
£ ¤ It is important to note that in general
d
dimension d the spinor has 2 2 complex components.
The spinor ψ̄ = ψ † γ 0 is called the Dirac conjugate of ψ. Its importance is explained
by the fact that the quantity ψ̄ψ = ψ † γ 0 ψ is an invariant of O(3, 1).
In any dimension one can define the charge conjugation matrix C. Indeed, the
Clifford algebra of γ-matrices transforms into itself under operation of transposition
{γ a , γ b }t = {(γ a )t , (γ b )t } = 2η ab ,
therefore by irreducibility of the corresponding representation of the Clifford algebra

there should exists a matrix C which intertwines the original and the transposed
representation of the algebra, namely:
(γ a )t = −Cγ a C −1 .
Matrix C is called the charge conjugation matrix.

Sometimes (depending on the dimension and signature od space-time) it is possible
to define the notion of Majorana spinor. Majorana conjugate spinor is, by definition,
ψ t C. The Majorana spinor is then the spinor for which the Dirac conjugate is equal
to the Majorana conjugate:
ψ†γ 0 = ψtC .
Spinor algebra in two dimensions

18
The group O(3, 1) is not connected and it has four connected components, which however are
not simply connected. The component which contains an identity coincides with SO(3, 1), which
are the transformations preserving orientation of the vierbein. The transformations which preserve
the direction of time are called orthochronous and they form the subgroup SO+ (3, 1). The quotient
group O(3, 1)/SO+ (3, 1) is the Klein four-group Z2 ×Z2 , which is the semidirect product of SO+ (3, 1)
with an element of the discrete group {1, P, T, P T }, where P and T are the space inversion and
time reversal operators
P = diag(1, −1, −1, −1) , T = diag(−1, 1, 1, 1) .
The covering (or spin) group of SO+ (3, 1) coincides with SL(2, C). One can show that SO+ (3, 1) =
SL(2, C)/{I, −I} ≡ PSL(2, C). Thus, SL(2, C) is the double-cover of SO+ (3, 1).
– 95 –
The two-dimensional Dirac matrices ρa , a = 0, 1 obey the algebra
µ ¶
a b ab ab −1 0
{ρ , ρ } = 2η , η = .
0 +1
A particular basis foe the Clifford algebra is given by

µ ¶ µ ¶
0 0 1 1 0 1
ρ = , ρ = .
−1 0 1 0
We also define a matrix ρ̄

µ ¶
0 1 1 0
ρ̄ = ρ ρ =
0 −1
being a 2dim analogue of the 4dim matrix γ 5 . The charge conjugation matrix can be
taken to be C = ρ0 . The Majorana spinor is then ψ † ρ0 = ψ t C = ψ t ρ0 , i.e. ψ = ψ ∗
which simple means that the spinor is real. Thus, in 2dim Majorana spinor is just a
spinor with real components.
Finally, with the help of the vielbein (“zweibein” in 2dim) one can define the
“curved” ρ-matrices:
ρα = eαa ρa .
They satisfy the following algebra
{ρα , ρβ } = 2hαβ .
Spinors in 2dim have various interesting properties. One of them is the so-called
spin-flip identity. If we have two Majorana spinors ψ1 and ψ2 , then the following
identity is valid
ψ̄1 ρα1 · · · ραn ψ2 = (−1)n ψ̄2 ραn · · · ρα1 ψ1 .
It is proved as follows.
ψ̄1 ρα1 · · · ραn ψ2 = (ψ̄1 ρα1 · · · ραn ψ2 )t = −ψ2t (ραn )t · · · (ρα1 )t (ρ0 )t ψ1
= (−1)n ψ2t Cραn C −1 · · · Cρα1 C −1 Cψ1 = (−1)n ψ̄2 ραn · · · ρα1 ψ1 .
Another identity is
ρα ρβ ρα = 0 . (6.1)
Indeed, one has from the Clifford algebra that ρa ρα = 2 and, therefore
ρα ρβ ρα = −ρα ρα ρβ + 2ρβ = −2ρβ + 2ρβ = 0 .
– 96 –
Finally, we discuss the completeness condition for the ρ-matrices. The matrices ρa ,
ρ̄ and the identity matrix I form a basis in the space of all 2 × 2-matrices. Any
2 × 2-matrix M can be expanded as
M = qa ρa + q̄ ρ̄ + qI .
The coefficients of this expansion are found as follows
q a = 12 Tr(M ρa ) , q̄ = 12 Tr(M ρ̄) , q = 21 Tr(M ) .
Plugging this back we obtain
2M = Tr(M ρa )ρa + Tr(M ρ̄)ρ̄ + Tr(M )I
or, more explicitly,

h i
2Mαβ = Mγδ (ρa )δγ ρaαβ + ρ̄δγ ρ̄αβ + δδγ δαβ .
Since M is an arbitrary matrix from here we derive the following completeness rela-
tion
(ρa )δγ ρaαβ + ρ̄δγ ρ̄αβ + δδγ δαβ = 2δαγ δβδ .
Spin connection
We would like to introduce a new local symmetry which is the Lorentz symmetry.
However, we have to guarantee that the the theory we are after should be invariant
w.r.t. these local transformations. As for the case of any local gauge invariance,
the local Lorentz invariance can be achieved by introducing q gauge field ωαa b (x) for
SO(d − 1, d). Here a, b are SO(d − 1, d)-indices and α is the “curved” vector index.
Under the local Lorentz transformations with the matrix Λ this field transforms as
follows
ωα → Λωα Λ−1 − ∂α ΛΛ−1 .
The gauge field of the local Lorentz symmetry is usually called the spin connection.
The spin connection plays the same role for the “flat” indices as the Christoffel
connection plays for the “curved” ones. We have the following substitution of the
basic objects in the theory
³ ´ ³ ´
haβ (x), Γδαβ (x) → eaα (x), ωαa b (x) .
Introduction of the spin connection should not change the gravitational content of
the theory. This means that the spin connection should not be a new independent
field, rather it should be determined in terms of vielbein. The simplest and elegant
way to do it is to notice that we have the covariant derivatives
Dα V β = ∂α V β + Γβαδ V δ
Dα V a = ∂α V a + ωαa b V b
– 97 –
The “flat” and “curved” indices of the vector are related as V a = eaα V α . This way of
transforming “flat” to “curved” indices should be valid for any tensor, in particular,
one has to have
Dα V a = eaβ Dα V β .
This is possible only if
Dα eaβ = ∂α eaβ − Γλαβ eaλ + ωαa b ebβ = 0 .
This equation can be solved for ωα expressing it through the vielbein:

1 1 1
ωαab = eβa (∂α ebβ − ∂β ebα ) − eβb (∂α eaβ − ∂β eaα ) − eλa eγb (∂λ eγc − ∂γ eλc )ecα .
2 2 2
Since the connection is completely expressed via the dynamical vielbein and, by this
means, is not an independent field, it is called sometimes composite.
Spin manifolds
On which manifolds one can introduce spinors? This is rather non-trivial ques-
tion. Vielbein can always be introduces locally in a coordinate patch Uα . It is quite
rare that the vielbein can be also globally defined. In the latter case such manifolds
are called parallelizable. Examples of parallelizable manifolds are Lie groups. On the
other hand, two-sphere is not parallelizable because there is no globally defined vec-
tor field (not talking about the vielbein) which vanishes nowhere. Thus, the vielbein
is defined locally and in the intersection of the coordinate patches Uα ∩ Uβ one has
e(α) (x) = Λ(αβ) (x)e(β) (x) .
Here Λ(αβ) (x) is the local Lorentz transformation (i.e. a matrix from SO(d − 1, 1))
which is called the transition function. In the region of triple intersection Uα ∩ Uβ ∩
Uγ = Uαβγ the transition functions should satisfy the following condition
Λ(αβ) Λ(βγ) Λ(γα) = 1 .
Now if we introduce locally a spinor field ψ(α) then passing from one coordinate patch
to another one the field must transform according to
ψ(α) (x) = Λ̄(αβ) ψ(β) (x) .
Here Λ̄(αβ) is the SO(d − 1, 1) matrix in the spinor representation. For spinorial
transition functions Λ̄ it also makes sense to require that in the triple intersection
region the following relation is satisfied
Λ̄(αβ) Λ̄(βγ) Λ̄(γα) = ±1 . (6.2)
However, since the spinor representation is double-valued, instead of Λ̄ one can

equally use −Λ̄. Thus, to define spinors on a non-parallelizable manifold one has
– 98 –
to pick up the signs for Λ̄(αβ) such that relation (6.2) is satisfied. When it is possible,
the corresponding manifold M is said to admit the spin structure and it is called the
spin manifold. Note that M may admit several inequivalent spin structures.
String theory contains world-sheet fermions and therefore it can be defined on

spin-manifolds. It turns out that in 2 and 3dim any orientable manifold is the spin
manifold. It is not so in 4dim and higher. Finally, we state without proof that on a
Riemann surface of genus g there are 22g inequivalent spin structures.
6.2 Superstring action and its symmetrices

The superstring action is based on two multiplest of 2dim supersymmetry. The first
one is the matter multiplet ³ ´
µ µ µ
X ,ψ ,F .
Here X µ is a bosonic field of the Polyakov string, ψ µ is the Majorana spinor in 2dim
and F µ is the real scalar, which is an auxiliary field to guarantee the equality of
the bososnic and fermionic degrees of freedom off-shell (i.e. without usage of the
equations of motion). Also the target space-time index µ is just a label so that for
µ = 0, . . . , d−1 we have d matter multiplets. the second multiplet is the supergravity
multiplet ³ ´
a
eα , χ α , A .
Here χα is the gravitino, i.e. the Majorana spinor which is also a vector of the 2dim
world-sheet. The field A is an auxiliary scalar field which is needed to guarantee the
equality of the bososnic and fermionic degrees of freedom off-shell. We note that the
kinetic term for the gravitino in any dimension is χ̄α γ αβγ Dβ χγ and it is absent in
two dimensions (because there are only two ρ -matrices while the anti-symmetrized √
product γ αβγ requires at least three to exist). Finally we introduce e ≡ |deteaα | = h.
The superstring action is a generalization of the bosonic Polyakov action to in-

clude the include the world-sheet fermionic degrees of freedom in the supersymmetric
way. It has the following structure
Z µ ³ ´¶
1 2 αβ µ β µ α β α µ i
S=− d σe h ∂α X ∂ Xµ + 2iψ̄ ρ ∂α ψµ − iχ̄α ρ ρ ψ ∂β Xµ − χ̄β ψµ .
8π 4
This action has five local symmetries
1. Local supersymmetry. Let ² is the Majorana spinor. We consider it as the

infinitezimal parameter of local supersymmetry transformations
i
δ² X µ = i²̄ψ µ , δ² eaα = ²̄ρa χα ,
2
1 ³ i ´
δ² ψ µ = ρα ∂α X µ − χ̄α ψ µ ² , δ² χα = 2Dα ² .
2 2
– 99 –
Here Dα ² = ∂α ² − 12 ωα ρ̄² and ωα is the connection with torsion:
i
ωα = ωα (e) + χ̄α ρ̄ρβ χβ ,
4
where ωα (e) = − 1e eαa ²βγ ∂β eaγ is the standard composite spin connection (it has
only one-non-trivial component in 2dim).
2. Weyl invariance. Let Λ be a bosonic local parameter Λ = Λ(σ, τ ). The Weyl

transformations are
δΛ X µ = 0 , δΛ eaα = Λeaα ,
1 1
δΛ ψ µ = − Λψ µ , δΛ χα = Λχα .
2 2
3. Super Weyl invariance. Let η be the Majorana spinor. Under the super Weyl
transformations only the gravitino transforms as
δη χα = ρα η .
The invarince of the action easily follows from the identity ρα ρβ ρα = 0.
4. Local Lorentz symmetry. Let ` be a bosonic local parameter ` = `(σ, τ ). The

local Lorentz transformations are
δ` X µ = 0 , δ` eaα = `²ab ebα ,

1 1
δ` ψ µ = `ρ̄ψ µ , δ` χα = `ρ̄χα .
2 2
5. Reparametrizations. Let ξ α be a bosonic vector parameter ξ α = ξ α (σ, τ ). The

reparametrizations (diffeomorphisms) are
δξ X µ = ξ β ∂β X µ , δξ eaα = ξ β ∂β ebα + eaβ ∂α ξ β ,

δξ ψ µ = ξ β ∂β ψ µ , δξ χα = ξ β ∂β χα + χβ ∂α ξ β .
6.3 Superconformal gauge and supermoduli

The gravitino field is the reducible representation of the Lorentz group. To decom-
pose it into irreducible representations one can use the following trick:
³ 1 ´ 1 1 1
χα = δαβ χβ = δα − ρα ρ χβ + ρα ρβ χβ = ρβ ρα χβ + ρα ρβ χβ
β β
2 2 |2 {z } 2
χ̃α
Here χ̃α part is called ρ–traceless because, due to the identity ρα ρβ ρα = 0 we get
ρα χ̃α = ρa eαa χ̃α = ρa χ̃a = 0. Indeed the gravitino χa transforms under local Lorentz
– 100 –
transformations as the spin-vector:19
1
δ` χa = −`²ab χb + `ρ̄χa .
2
Condition ρa χa = 0 remains invariant under these transformations, i.e. ρa δ` χa = 0.
Decomposition of the gravitino into the ρ-trace and ρ-traceless part is decomposition
into two irreducible representations of the Lorentz group corresponding to helicities
±3/2 and ±1/2 respectively. This decomposition is orthogonal w.r.t. the scalar
R
product (φ|ψ) = d2 σ φ̄α ψα .
The local supersymmetry transformation for the gravitino filed can be also de-
composed into the traceless- and the trace-parts:
δ² χα = 2Dα ² = 2(Π²)α + ρα ρβ Dβ ²
| {z }
trace part
Here we defined the operator

1
(Π²)α = ρβ ρα Dβ ² , =⇒ ρα (Π²)α = 0 .
2
Locally one can show that there always exists a spinor κ such that χ̃α = ρβ ρα Dβ κ.
Comparing this with the supersymmetry transformation for χα we conclude that κ
can always be eliminated (locally!) by a supersymmetry variation. The possibility
to eliminate κ globally depends on the existence of a globally defined spinor ² which
solves the equation
(Π²)α = τα
for arbitrary τα satisfying the condition ρα τα =0. Global solvability of the last ex-
pression relies on the absence of zero modes of the operator Π† : (Π† τ ) = −2Dα τα .
This equation is the supercousin of the bosonic equation
(P ξ)αβ = tαβ
whose global solvability relies on the absence of zero modes of P † . According to our
discussion of the bosonic case it makes sense to call
dim kerP † = moduli

dim kerΠ† = supermoduli
and also
dim kerP = conformal Killing vectors

dim kerΠ = conformal Killing spinors .
µ ¶µ ¶ µ ¶
19 −1 0 0 1 0 −1
The element ²ab cb
= ηac ² is the following matrix ²ab = = .
0 1 −1 0 −1 0
– 101 –
By using reparametrizations and local Lorentz transformations the zweibein can
be brought to the form eaα = eφ δαa which is locally always possible. The gauge
eaα = eφ δαa , χ α = ρα λ
is called superconformal gauge. In classical theory the Weyl symmetry and super
Weyl symmetry can be used to eliminate the remaining gravitational degrees of
freedom φ and λ. In quantum theory it will be possible in critical dimension only.
6.4 Action in the superconformal gauge

In the superconformal gauge the action becomes rather simple
Z ³ ´
1
S=− d2 σ ∂α X µ ∂ α Xµ + 2iψ̄ µ ρα ∂α ψµ . (6.3)
8π
The world-sheet indices are now raised and lowered with the help of the flat world-
sheet metric η aβ and ρα = δaα ρa . This action is invariant w.r.t. local reprametrizations
and supersymmetry transformations which satisfy the requirement
Pξ = 0, Π² = 0 .
We would like to check directly that the action (6.3) is invariant under the super-
symmetry transformations
δ² X µ = i²̄ψ µ ,
1
δ² ψ µ = ρα ∂α X µ ²
2
1
δ² ψ̄ µ = − ²̄ρα ∂α X µ
2
provided the parameter ² satisfies the following equation
ρβ ρα ∂β ² = 0 . (6.4)
To check the invariance we perform the variation

Z ³ ´
1 2 µ α µ µ α β α µ β
δ² S = − d σ 2∂α X ∂ (i²̄ψ ) + iψ̄ ρ ∂α (ρ ∂β Xµ ²) − ²̄ρ ∂α X ρ ∂β ψµ .
8π
Now we integrate by parts the first term and write out the second term more explicitly
Z ³
1
δ² S = − d2 σ − 2i¤X µ ²̄ψµ + iψ̄ µ ρα ρβ ∂α ∂β Xµ ²
8π |{z}
η αβ
´
+ iψ̄ µ ρα ρβ ∂β Xµ ∂α ² − ²̄ρα ∂α X µ ρβ ∂β ψµ .
– 102 –
Now we apply the spin-flip identity in the second and the third term and get
Z ³
1
δ² S = − d2 σ − 2i¤X µ ²̄ψµ + i²̄ψµ ¤X µ
8π
´
+ i∂α ²̄ρβ ρα ψ µ ∂β Xµ − ²̄ρβ ρα ∂β X µ ∂α ψµ .
Finally, we integrate the last term by parts and get

Z
1
δ² S = − d2 σ 2i∂α ²̄ρβ ρα ψ µ ∂β Xµ .
8π
This vanishes as the consequence of eq.(6.4).
We can write equation (6.4) more explicitly

³ ´ ³ ´ ³ ´
β 0 1 0 0 0 1
ρ ρ0 ∂β ² = ρ ρ0 ∂0 + ρ ρ0 ∂1 ² = − ρ ρ ∂0 + ρ ρ ∂1 ² = ∂0 + ρ̄∂1 ² = 0 ,
³ ´ ³ ´ ³ ´
ρβ ρ1 ∂β ² = ρ0 ρ1 ∂0 + ρ1 ρ1 ∂1 ² = ρ0 ρ1 ∂0 + ρ1 ρ1 ∂1 ² = ρ̄∂0 + ∂1 ² = 0 .
Since ρ̄2 = 1 the second equation is obtained from the first by multiplying with ρ̄
and by this reason it is redundant. To analyze the first equation it is convenient to
denote the components of any spinor as follows
µ ¶ µ ¶
ψ+ ²+
ψ= and ²= .
ψ− ²−
We thus see that the first equation reduces to
(∂0 + ∂1 )²+ = ∂+ ²+ = 0 , (∂0 − ∂1 )²− = ∂− ²− = 0
One cab define the spinors with upper indices by using the following convention
ψ − = ψ+ , ψ + = −ψ− .
With this convention we obtain that components of the Majorana spinor which is a
parameter of the supersymmetry transformations satisfy the equations
∂+ ²− = ∂− ²+ = 0 =⇒ ²± ≡ ²± (σ ± ) .
This equations should be contracted with the equations defining the conformal Killing
vectors, i.e. reparametrizations which do not destroy the conformal gauge choice:
∂+ ξ − = ∂− ξ + = 0 =⇒ ξ ± ≡ ξ ± (σ ± ) .
On-shell closer of the supersymmetry algebra
– 103 –
Consider first the commutator of two supersymmetry variations applied to a bosonic
field X µ :
i¡ α ¢
[δ²1 , δ²2 ]X µ = i²̄1 δ²2 ψ µ − i²̄1 δ²2 ψ µ = ²̄1 ρ ²2 − ²̄2 ρα ²1 ∂α X µ = i²̄1 ρα ²2 ∂α X µ ,
2
where the last formula stems from the spin-flip property. We see that the commutator
of two super-symmetry transformations generates a diffeomorphism transformation
[δ²1 , δ²2 ]X µ = ξ α ∂α X µ
with the parameter ξ α = i²̄1 ρα ²2 . In fact, this is not an arbitrary diffeomorohism,

rather it is a conformal transformation, because ξ α is nothing else as a conformal
Killing vector! Thus, bilinear combinations made of conformal spinors
Consider now the commutator of two supersymmetry variations applied to a
fermion ψ µ :
1 1 i i
[δ²1 , δ²2 ]ψ µ = ∂α (δ²1 X µ )ρα ²2 − ∂α (δ²2 X µ )ρα ²1 = ∂α (²̄1 ψ µ )ρα ²2 − ∂α (²̄2 ψ µ )ρα ²1
2 2 2 2
Therefore,
i i
[δ²1 , δ²2 ]ψ µ = (∂α ²̄1 ψ µ )ρα ²2 − (∂α ²̄2 ψ µ )ρα ²1 +
2 2
i i
+ (²̄1 ∂α ψ µ )ρα ²2 − (²̄2 ∂α ψ µ )ρα ²1 (6.5)
2 2
Constraints
In the original (gauge-unfixed) theory we have a world-sheet metric (vielbein) and

the gravitino field. They are removed upon imposition of the superconformal gauge.
However, before we fix the gauge, the metric and the gravitino have their equations
of motion which become the constraints on the other fields of the theory after fixing
the gauge. The stress tensor in now defined as
2π δS
Tαβ = − eαa .
e δeβa
We can also define the supercurrent as response of the action for variation of the
gravitino field
2π δS
Gα = −i .
e δ χ̄α
Analogously to what was in the bosonic case the stress tensor Tαβ will generate
conformal transformations, while the new object Gα appears to be a generator of the
supersymmetries. Equations of motion
Tαβ = 0 = Gα
– 104 –
are constraints on the dynamics of our system. Using the action we find
1 1 i i
Tαβ = ∂α X µ ∂β Xµ − ηαβ ∂γ Xµ ∂ γ X µ + ψ̄ρα ∂β ψµ + ψ̄ρβ ∂α ψµ
2 4 4 4
1 β
Gα = ρ ρα ψ µ ∂β Xµ
4
Note that Gα is ρ-traceless, i.e.
ρα Gα = 0 .
This equation is an analog of Tαα = 0. Finally, by using equations of motion one can
show that the stress tensor and the supercurrent are conserved
∂ α Tαβ = 0 , ∂ α Gα = 0 .
These conservation laws lead to existence of infinite number of conserved charges. To

analyze the algebra of contraints in more detail it is convenient to use the world-sheet
light-cone coordinates σ ± . In the light-cone coordinates the action becomes
Z ³ ´
1
S= d2 σ ∂+ X∂− X + i(ψ+ ∂− ψ+ + ψ− ∂+ ψ− ) .
2π
Equations of motion are
µ µ
∂+ ∂− X µ = 0 , ∂ − ψ+ = ∂+ ψ− = 0.
Solving equations of motion for fermions we get

µ µ µ µ
ψ+ = ψ+ (σ + ) , ψ− = ψ− (σ − ) .
This, it appears that two components of the Majorana fermion are left- and right-
moving fields on the world-sheet.
The components of the stress-tensor T+− = 0 = T−+ , while the other components
are
1 i
T++ = ∂+ X∂+ X + ψ+ ∂+ ψ+ ,
2 2
1 i
T−− = ∂− X∂− X + ψ− ∂− ψ− .
2 2
The components of the supercurrent are
1
G+ = ψ+ ∂+ X ,
2
1
G− = ψ− ∂− X .
2
The conservation laws look in the light-cone coordinates as
∂− G+ = ∂+ G− = 0 , ∂− T++ = ∂+ T−− = 0 .
– 105 –
Now we note that in addition to the conserved charges generated by the stress tensor:
Z
Qξ± = dσ ξ ± (σ ± )T±± (σ, τ )
we will have the conserved charges generated by the supercurrent

Z
G²± = dσ ²± (σ ± )G± (σ, τ ) .
6.5 Boundary conditions

Varying the action to derive the equations of motion we will get the following bound-
ary term Z ³ ´
dσ∂σ ψ+ δψ+ − ψ− δψ− .
For the case of closed string, to make this term vanishing one has to impose the
following condition
³ ´ ³ ´
ψ+ δψ+ − ψ− δψ− (σ) − ψ+ δψ+ − ψ− δψ− (σ + 2π) = 0 .
Since the fermions ψ+ and ψ− are independent this equation implies that
ψ+ (σ) = ±ψ+ (σ + 2π) ,

ψ− (σ) = ±ψ− (σ + 2π) .
The following terminology is standard
• Periodic boundary conditions in σ are called Ramond boundary conditions and

they are denoted by the letter “R”.
• Anti-periodic boundary conditions in σ are called Nevew-Schwarz boundary

conditions and they are denoted by the letter “NS”.
Universally, all fermionic quantities on the world-sheet have the following boundary
conditions
ψ(σ + 2π) = e2πiθ ψ(σ) ,
where θ = 0 in the R-sector and θ = 1/2 in the NS sector.
Boundary conditions for ψ+ and ψ− can be chosen independently, which gives in

total four possibilities
(R, R) , (NS, NS) , (R, NS) , (NS, R)
The boundary conditions for the two components of the supersymmetry parameter
should be chosen in such a way as to make the variation δX µ = i²̄ψ µ periodic.
– 106 –
As we will see the states in the R sector give the space-time fermions, while the
states in the NS sector are the space-time bosons. This further gives that (R,R) and
(NS,NS) sectors are space-time bosons while (R,NS) and (NS,R) are fermions.
For the open string case we have that
ψ+ δψ+ − ψ− δψ−
must vanish at σ = 0 and σ = π. If we assume that ψ+ = αψ− at the end point of

string, then
(α2 − 1)ψ− δψ− = 0
which allows for α = ±1. Thus, at each end of the string we should have ψ+ = ±ψ− .
We can always agree to choose ψ+ (0, τ ) = ψ− (0, τ ) as it is a matter of convention,
then on the other hand of the string we have two possibilities
ψ+ (π, τ ) = ψ− (π, τ ) (Ramond) ,

ψ+ (π, τ ) = −ψ− (π, τ ) (Neveu − Schwarz) .
6.6 Superconformal algebra

In order to compute the Poisson bracket between the components of the stress tensor
and the supercurrent we need the fundamental Poisson bracket for the fermions. The
Dirac action leads to the following bracket
µ ν
{ψ+ (σ), ψ+ (σ 0 )} = −2πiδ(σ − σ 0 )η µν
µ ν
{ψ− (σ), ψ− (σ 0 )} = −2πiδ(σ − σ 0 )η µν .
Using these brackets together with brackets between the bosonic fields one find the
following Poisson algebra of the constraints
³ ´
{T++ (σ), T++ (σ 0 )} = −2π 2T++ (σ 0 )∂ 0 + ∂ 0 T++ (σ 0 ) δ(σ − σ 0 ) ,
³3 ´
{T++ (σ), G+ (σ 0 )} = −2π G+ (σ 0 )∂ 0 + ∂ 0 G+ (σ 0 ) δ(σ − σ 0 ) ,
2
{G+ (σ), G+ (σ )} = −iπT++ (σ)δ(σ − σ 0 ) .
0
This is the so-called N = 1 superconfomal algebra in 2dim. Here N = 1 refers to

the fact that supersymmetry transformations are performed with the help of one
Majorana spinor.
The action of the supercurrent on the bosonic and fermionic fields generate supersym-
metry transformations (bosonic field transforms into fermionic one and vice-versa):
µ
{G+ (σ), X µ (σ 0 )} = −π ψ+ (σ)δ(σ − σ 0 ) ,
{G+ (σ), ψ µ (σ 0 )} = −iπ ∂+ X+µ (σ)δ(σ − σ 0 )
– 107 –
One can also check that the world-sheet fermion transforms under conformal trans-
formations as the conformal field with weigh 1/2.
Consider closed strings. Using the mode expansion

X
ψ+ (σ, τ ) = b̄µr e−ir(τ +σ) ,
r∈Z+θ
X
ψ− (σ, τ ) = bµr e−ir(τ −σ) ,
r∈Z+θ
1
where θ = 0 in the R-sector and θ = 2
in the NS-sector, we obtain the Poisson
algebra of oscillators
{bµr , bνs } = −iη µν δr+s ,
{b̄µr , b̄νs } = −iη µν δr+s ,
{bµr , b̄νs } = 0 .
The reality of the Majorana spinor implies that
(bµr )† = bµ−r , (b̄µr )† = b̄µ−r .
Introducing the modes of the stress tensor and the supercurrent
Z 2π Z
1 −imσ 1 2π
Lm = dσe T−− , Gm = dσe−irσ G− .
2π 0 π 0
Notice that the supercurrent G− satisfies the same boundary condition as the fermion
ψ− . Substituting the mode expansion we get
1X 1 X³ m´
Lm = α−n αm+n + r+ b−r bm+r ,
2 n∈Z 2 r 2
X
Gr = α−n br+n .
n∈Z
These generators generate the classical super-Virasoro algebra

{Lm , Ln } = −i(m − n)Lm+n ,
³1 ´
{Lm , Gr } = −i m − n Gm+r ,
2
{Gr , Gs } = −2iLr+s .
7. Quantum fermionic string

Canonical quantization is again performed by substituting the Poisson bracket for
the (anti)-commutator: { , }pb → 1i [ , ]. Therefore, the anti-commutators for the
quantum fermionic fields are
µ ν
{ψ+ (σ), ψ+ (σ 0 )} = 2πδ(σ − σ 0 )η µν
µ ν
{ψ− (σ), ψ− (σ 0 )} = 2πδ(σ − σ 0 )η µν .
– 108 –
Anti-commutators of the modes are
{bµr , bνs } = η µν δr+s .
We again see that we can split oscillators into creation and annihilation operators
according to the sign of their index, namely
• Oscillators with r > 0 are annihilation operators ,
• Oscillators with r < 0 are creation operators .
However, the modes with r = 0 which occur in the Ramond sector only require
special care. Indeed, in the bosonic modes α0µ and ᾱ0µ correspond to the center of
mass momentum of the string. Analogously, bµ0 and b̄µ0 are distinguished from all the
other modes, in particular, they form the Clifford algebra
{bµ0 , bν0 } = η µν
and analogously for b̄µ0 .
The super-Virasoro generators are again defined as normal ordered expressions

1X 1 X³ m´
Lm = : α−n αm+n : + r+ : b−r bm+r : ,
2 n∈Z 2 r 2
X
Gr = α−n br+n .
n∈Z
Only the generator L0 is ambiguous doe to the undetermined normal ordering con-
stant. Ignoring this constant for the moment we obtain the following answer for the
quantum super-Virasoro algebra
d
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 2ω)δm+n ,
8
³1 ´
[Lm , Gr ] = m − n Gm+r ,
2
d³ 2 ω ´
{Gr , Gs } = 2Lr+s + r − δr+s .
2 2
Here ω = 0 for the R-sector and ω = 12 for the NS-sector. Both the R- and NS-
algebras formally agree except the linear terms in anomalies. The linear term can
be changed by shifting the L0 generator. Indeed, one can see that if one shifts
d 1
LR R
0 → L0 + 16 then both algebras have formally the same structure with ω = 2 . Still
the R- and NS-algebras are very different. For instance, in the NS-sector the five
generators L1 , L0 , L−1 , G1/2 , G−1/2 form a closed superalgebra known as OSp(1|2). In
the R-sector just adding to the generators L1 , L0 , L−1 the generator G0 one generates
the whole infinite-dimensional algebra.
– 109 –
The oscillator ground state is defined in both sectors as
µ
αm |0i = bµr |0i = 0 , m, r > 0 .
Here the dependence on the center of mass momentum is suppressed. In the Ramond
sector one also has the zero mode bµ0 . The level number operator is
N = N (α) + N (b) ,
where
∞
X
(α)
N = α−m αm ,
m=1
X∞
N (b) = rb−r br .
r∈Z+θ>0
Note that the zero mode in the Ramond sector does not contribute to the number
operator! This leads to the fact the mass operator commutes with bµ0 : [bµ0 , M 2 ] = 0,
i.e. the states |0i and bµ0 |0i have the same mass. These states are degenerate. On the
other hand, all other oscillators αnµ , bµr with n, r < 0 increase α0 M 2 by 2n and 2r units
respectively. This means that in the NS-sector the ground state is unique and it has
Lorentz spin zero. In the R-sector the ground state is degenerate and since bµ0 form
the Clifford algebra the ground state is a spinor of the Lorentz group SO(d − 1, 1).
This explains why in the NS-sector all the states are space-time bosons, while in the
R-sector they are all fermions. Indeed, all creation operators have vector Lorentz
index and by this reason they cannot convert a space-time boson into a space-time
fermion or vice versa. If we will write the Ramond ground state as |ai, where a is a
SO(d − 1, d) spinor index, the bµ0 act on it as the usual Γ-matrices
1
bµ0 |ai = √ (Γµ )ab |bi .
2
Here Γµ are the usual Γ-matrices of the d-dimensional Minkowski space and they
satisfy the Clifford algebra {Γµ , Γν } = 2η µν .
We will not go into discussion of the covariant quantization but will just state
that consistency of the quantum theory will impose the following restrictions on the
constant a of the normal ordering ambiguity (for the Ramond and Neveu-Schwarz
sectors) and the dimension d of the target space-time:
1
aNS = , aR = 0 , d = 10 .
2
The same result follows from the condition of non-anomalous Lorentz algebra in the
light-cone gauge. Instead of d = 26 found for bosonic string, quantum fermionic
string chooses to live in a ten-dimensional world.
– 110 –
7.1 Light-cone quantization and superstring spectrum
As we have established imposition of the superconformal gauge does not completely
remove the gauge (unphysical) degrees of freedom. The superconformal transforma-
tions which do not destroy the superconformal gauge choice are left. In order to
remove the remaining unphysical degrees of freedom one can try to fix the light-cone
gauge, similar to as was done for bosonic string. We can also fix
X + = α 0 p+ τ
in our fermionic theory and this choice will completely remove the reparametrization
invariance. However, the local supersymmetry transformations obeying the equations
∂+ ²− = ∂− ²+ = 0
are still left over. These transformations can be used in order to completely elimi-
nate the fermionic field ψ + , where this time ψ ± refer to the target-space light-cone
components
1
ψ ± = √ (ψ 0 ± ψ d−1 ) .
2
This is equivalent to putting to zero the modes b+ r for all r. After this gauge choice
is done we can solve the super-Virasoro constraints and find the longitudinal modes
1
(remember that T = 2πα 0)
1 ³ ´
∂± X − = 0 + (∂± X i )2 + iψ± i
∂± ψ±i
αp
− 2 i
ψ± = 0 + ψ± ∂± X i .
αp
This shows that only the transversal components X i and ψ i are physical degrees of
freedom. In terms of oscillators the previous equations read
− 1 ³ i i X ¡m ¢ i i ´
αm = √ : αn αm−n : + − r : br bm−r : −2aδm
2α0 p+ r
2
2 X i i
b−
r = α b .
α0 p+ q r−q q
For the case of closed strings these expressions must be supplemented by the analo-
gous ones for the left-moving modes. Here we also include a normal ordering constant
a which is a = 12 in the NS sector and a = 0 in the R sector.
The mass operator is

M 2 = MR2 + ML2 ,
where ³X X ´
0
α MR2 =2 i
α−n αni + rbi−r bir −a
n>0 r>0
– 111 –
and similar for ML2 . For the case of open string we have
³X X ´
α0 MR2 = i
α−n αni + rbi−r bir − a .
n>0 r>0
α0 mass2 rep of SO(8) little (−1)F rep of little

group group
NS − sector
− 12 |0i SO(9) −1 1
0 bi−1/2 |0i SO(8) +1 8v

| {z }
8v
+ 12 i
α−1 |0i , bi−1/2 bj−1/2 |0i SO(9) −1 36
| {z } | {z }
8v 28
bi−1/2 bj−1/2 bk−1/2 |0i +1

| {z }
56v
+1 SO(9) 84 + 44
i
α−1 bj−1/2 |0i , bi−3/2 |0i +1
| {z } | {z }
1+28+35v 8v
R − sector
|ai +1 8s
|{z}
8s
0 SO(8)
|ȧi −1 8c
|{z}
8c
αi |ai , bi−1 |ȧi +1 128

| −1
{z } | {z }
8c +56c 8s +56s
+1 SO(9)
i
α−1 |ȧi , bi−1 |ai −1 128
| {z } | {z }
8s +56s 8c +56c
Tab. 5. The lowest levels of the open fermionic string spectrum.
– 112 –
In the closed string case we have in addition the condition of level-matching
which requires that for physical states
ML2 = MR2 .
Finally, we give expressions for the light-cone action

Z ³ ´
1 2 i 2 0i 2 i a i
Sl.c = d σ (Ẋ ) − (X ) − 2iψ̄ ρ ∂α ψ (7.1)
8π
and the Hamiltonian
X X
H = (pi )2 + i
(α−n i
αni + ᾱ−n ᾱni ) + r(bi−r bir + b̄i−r b̄ir ) − 2a .
n>0 r>0
Note that every sector, R and NS, has its own Hamiltonian.
Let us now analyze the closed string spectrum. We first discuss the right-moving
part which (up to a mass rescaling by 2) is equivalent to the spectrum of open
fermionic string.
• NS-sector. The ground state is the oscillator vacuum |0i with α0 M 2 = −a. The
first excited state is bi−1/2 |0i with α0 M 2 = 12 − a. This is a vector of SO(d − 2),
where the critical dimension d = 10. Since the little Lorentz group for massless
states in d-dimensions is SO(d − 2) this state must be massless which gives the
normal ordering constant to be a = 12 . At the next level one has the states
i
α−1 |0i and bi−1/2 bj−1/2 |0i with α0 M 2 = 12 . The number of these bosonic states
is 8 + 28 = 36 = 9×8 2
, they are comprise an antisymmetric representation of
SO(9), the little Lorentz group for massive states.
• R-sector. The Ramond ground state is a spinor of SO(9, 1). The dimension
d
of the Dirac spinor in d = 10 is 2 2 = 25 = 32, i.e. it has 32 complex or 64
real components. In ten dimensions it is possible to impose both Majorana
and Weyl conditions20 which reduce the number of independent components
64
to 2×2 = 16. On shell the number of components is further reduced by two
because the Dirac equation Γµ ∂µ ψ relates half of the components to the other
half (which satisfies the Klein-Gordon equation). The 8 remaining components
can be viewed as the components of the Majorana-Weyl spinor of SO(8), the
latter being the little Lorentz group for massless states in d = 10. Indeed, the
spinor of SO(8) should have
8
(2 2 complex components)/(Majorana × Weyl) = 32/4 = 8
20
In general, for the groups SO(p, q) the Majorana and Weyl conditions can be simultaneously
imposed if and only if p − q = 0 mod 8. For Minkowski space, p = d − 1, q = −1, this gives
d = 2 + 2n and for Euclidean space, p = d, q = 0, this gives d = 2n.
– 113 –
real components. Thus, the Ramond ground state is massless.
It turns out that the group SO(8) has three inequivalent representations of
dimension 8: two of them are spinor representations and another is the vec-
tor one. Spinor representations are commonly denoted by 8s and 8c and the
corresponding representation bases are depicted as
|ai and |ȧi .
The vector representation is 8v and the basis is |ii.

i
The first excited level consist of states α−1 |ai and bi−1 |ai and their chiral part-
ners with α0 M 2 = 1. Once again, for d = 10, all the massive light-cone states
can be uniquely assembled into representations of SO(9), the little Lorentz
group for massive states.
GSO projection
It turns out that fermionic string with all the states we found in the R and NS
sectors is inconsistent. This can seen, for instance, from the fact that the 1-loop
amplitude is not modular invariant. In order to construct a consistent modular
invariant theory one should truncate the string spectrum in a specific way. This
truncation is known as the GSO (Gliozzi-Scherk-Olive) projection. It restores the
modular invariance, removes from the theory the tachyon and, in addition, provides
the space-time supersymmetry of the resulting string spectrum. Below we will use
an inverse argument to motivate the GSO projection – we will show that it allows
to achive a spectrum which exhibits space-time supersymmetry.
Looking at the massless states in the Ramond sector we see that one has two SO(8)
spinors 8s and 8c . On the other hand, the massless states of the NS sector comprise
a vector 8v . If we project one of the two spinors out then there will be match of NS
bosonic (8) and R fermionic (also 8) degrees of freedom. These massless vector and
the massless spinor is indeed a content of the N = 1 super Yang-Mills theory in ten
dimensions.
One has also to get rid of tachyon which is in the NS sector. This all can be achieved
if one first defines an operator
∞
X
G = (−1)F , F = bi−r bir − 1
r= 12
and then requires that all allowed states should have G = 1:
G|Φi = |Φi.
– 114 –
α0 mass2 rep of SO(8) little (−1)F (−1)F̄ rep of little
group group
(NS, NS) − sector
−2 |0iL × |0iR SO(9) −1 −1 1
0 bi−1/2 |0iL × bj−1/2 |0iR SO(8) +1 +1 1 + 28 + 35v

| {z } | {z }
8v 8v
(R, R) − sector
|aiL × |biR +1 +1 1 + 28 + 35s

|{z} |{z}
8v 8v
|ȧiL × |ḃiR −1 −1 1 + 28 + 35c

|{z} |{z}
8c 8c
0 SO(8)
|ȧiL × |biR −1 +1 8v + 56v
|{z} |{z}
8c 8s
|aiL × |ḃiR +1 −1 8v + 56v

|{z} |{z}
8s 8c
(R, NS) − sector
|aiL × bi−1/2 |0iR +1 +1 8c + 56c

|{z} | {z }
8s 8v
0 SO(8)
|ȧiL × bi−1/2 |0iR −1 +1 8s + 56s
|{z} | {z }
8c 8v
(NS, R) − sector
b̄i−1/2 |aiL × |aiR +1 +1 8c + 56c
| {z } |{z}
8v 8s
0 SO(8)
b̄i−1/2 |aiL × |ȧiR +1 −1 8s + 56s
| {z } |{z}
8v 8c
Tab. 6. The lowest levels of the closed fermionic string spectrum.
– 115 –
Since a general state in the NS sector has the form
i1
|Φi = α−n 1
iN
· · · α−n bj1 · · · bj−r
N −r1
M
M
|0i
we get
G|Φi = (−1)M −1 |Φi.
Thus, all states with M even are projected out, in particular, the tachyon. This
removes the tachyon and all the states with half-integer α0 M 2 . Indeed,
rM
X
0 2 1
αM = ri −
i=1
2
|{z}
a
and for M even the sum of half-integers is always an integer α0 M 2 is half-integer

number.
In the Ramond sector the operator G is defined as follows
P∞ i i
G = (−1)F = b10 · · · b80 (−1) n=1 b−n bn .
The transversal zero modes bi0 form the Clifford algebra {bi0 , bj0 } = δ ij . Thus, these
operators can be represented by SO(8) γ-matrices Γi which have size 16 by 16. These
matrices act on the 16-dimensional Majorana (i.e. real) spinor whose components
can be thought to combine two Weyl projections, which are precisely |ai and |ȧi:
µ ¶
|ai8s
|ψi = .
|ȧi8c
In the Majorana representation these matrices Γi can be taken in the block-diagonal

form as
µ ¶
i 0 γi
Γ = .
(γ i )t 0
The fact that Γi obey the standard Glifford algebra {Γi , Γj } = 2δ ij implies that 8 × 8
real matrices γ i satisfy the following algebra
γ i (γ j )t + γ j (γ i )t = 2δ ij .
If we introduce the standard Pauli matrices
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 =
1 0 i 0 0 −1
then the matrices γ i can be defined as
1 2
γ = −iσ2 ⊗ σ2 ⊗ σ2 , γ = iI ⊗ σ1 ⊗ σ2 ,
3 4
γ = iI ⊗ σ3 ⊗ σ2 , γ = iσ1 ⊗ σ2 ⊗ I ,
5 6
γ = iσ3 ⊗ σ2 ⊗ I , γ = iσ2 ⊗ I ⊗ σ1 ,
7 8
γ = iσ2 ⊗ I ⊗ σ3 , γ = I ⊗ I ⊗ I.
– 116 –
The matrix γ i cab be understood as carrying the matrix indices a and ȧ: γa
i
ȧ . The chiral and anti-chiral representations of SO(8) are
constructed then with the help of matrices
ij 1 i j t j i t
γs = γ (γ ) − γ (γ ) ,
2
ij 1 i t j j t i
γc = (γ ) γ − (γ ) γ ,
2
The operator Γ9 = b10 · · · b08 is the chirality operator (the analog of the γ 5 -matrix
in 4dim.) and it projects out one of the two Weyl components of the Ramond
ground state |ψi. We see that G anti-commutes with any mode b−n : {G, bi−n } = 0
and, therefore, the eigenvalues of G in the Ramond sector are ±1, depending on their
chirality, if we define G|ai = |ai and G|ȧi = −|ȧi. Further, a general state in the R
sector is
|Φia = α−n i1
1
iN
· · · α−n bj1 · · · bj−m
N −m1
M
M
|ai
or
i1
|Φiȧ = α−n 1
iN
· · · α−n bj1 · · · bj−m
N −m1
M
M
|ȧi
We therefore find that
P
G|Φia = (−1)M (−1) i δmi ,0 |Φia ,
P
G|Φiȧ = −(−1)M (−1) i δmi ,0 |Φiȧ .
The GSO projection consists in leaving the states which have either G = 1 or G = −1.
To construct the spectrum of the closed superstring we have to tensor left and
right-moving states (such that the level matching constraint ML2 = MR2 is satisfied)
and then impose the GSO projection. Here we have to distinguish four different
sectors
(R, R) , (NS, NS) , (R, NS) , (NS, R)
| {z } | {z }
space−time bosons space−time fermions
The GSO projection is imposed separately for the left- and right-moving modes. In
the NS sector one keeps the states with
G = (−1)F = +1 , Ḡ = (−1)F̄ = +1 .
In the Ramond sector there are essentially two possibilities which lead to super-
symmetric and tachyonic-free spectrum. One of them is to take G = Ḡ = 1. The
massless spectrum is
h i h i
Bosons : (1) + (28) + (35)v + (1) + (28) + (35)s
h i h i
Fermions : (8)c + (56)c + (8)c + (56)c .
In total there are 128 bosonic and 128 fermionic states. The GSO projection imposed
in this way defines the so-called Type IIB superstring and its massless spectrum is
– 117 –
that of Type IIB supergravity in ten dimensions. In particular, 35v are on-shell
degrees of freedom of the graviton, two 28 are two anisymmetric tensor fields and
35s is a rank four antisymmetric self-dual tensor. In addition one has two real scalars.
The fermionic degrees of freedom are two spin-3/2 gravitinos, 56c , and two spin-1/2
fermions. The presence of two gravitinos indicates that the corresponding theory
has N = 2 supersymmetry. Since the gravitions and the fermions are of the same
chirality this theory is chiral.
One can make another choice of the GSO projection by requiring that G = −Ḡ = 1.
This gives the following massless spectrum
h i h i
Bosons : (1) + (28) + (35)v + (8)v + (56)v
h i h i
Fermions : (8)c + (56)c + (8)s + (56)s .
There are on-shell degrees of freedom of the graviton (35v ), antisymmetric rank three
tensor (56v ), an antisymmetric rank two tensor (28), one vector 8v ) and one real
scalar, which is called dilaton. The fermionic degrees of freedom comprise two spin-
3/2 gravitinos and two spin-1/2 fermions. Gravitions and fermions are of opposite
chirality. Thus, this theory has N = 2 supersymmetry also but it is non-chiral. This
string theory is called Type IIA, and the corresponding supergravity is Type IIA
supergravity.
– 118 –
Appendices
A. Dynamical systems of classical mechanics
To motivate the basic notions of the theory of Hamiltonian dynamical systems con-
sider a simple example.
Let a point particle with mass m move in a potential U (q), where q = (q 1 , . . . q n )
is a vector of n-dimensional space. The motion of the particle is described by the
Newton equations
∂U
mq̈ i = − i
∂q
Introduce the momentum p = (p1 , . . . , pn ), where pi = mq̇ i and introduce the energy
which is also know as the Hamiltonian of the system
1 2
H= p + U (q) .
2m
Energy is a conserved quantity, i.e. it does not depend on time,
dH 1 ∂U 1 ∂U
= pi ṗi + q̇ i i = m2 q̇i q̈i + q̇ i i = 0
dt m ∂q m ∂q
due to the Newton equations of motion.
Having the Hamiltonian the Newton equations can be rewritten in the form
∂H ∂H
q̇ j = , ṗj = − .
∂pj ∂q j
These are the fundamental Hamiltonian equations of motion. Their importance lies
in the fact that they are valid for arbitrary dependence of H ≡ H(p, q) on the
dynamical variables p and q.
The last two equations can be rewritten in terms of the single equation. Introduce
two 2n-dimensional vectors
µ ¶ Ã !
∂H
p ∂pj
x= , ∇H = ∂H
q ∂q j
and 2n × 2n matrix J:
µ ¶
0 −I
J=
I 0
Then the Hamiltonian equations can be written in the form
ẋ = J · ∇H , or J · ẋ = −∇H .
– 119 –
In this form the Hamiltonian equations were written for the first time by Lagrange
in 1808.
Vector x = (x1 , . . . , x2n ) defines a state of a system in classical mechanics. The

set of all these vectors form a phase space M = {x} of the system which in the present
P
case is just the 2n-dimensional Euclidean space with the metric (x, y) = 2n i i
i=1 x y .
The matrix J serves to define the so-called Poisson brackets on the space F(M )
of differentiable functions on M :
n ³
X ∂F ∂G ∂F ∂G ´
{F, G}(x) = (∇F, J∇G) = J ij ∂i F ∂j G = j
− j .
j=1
∂p j ∂q ∂q ∂p j
Problem. Check that the Poisson bracket satisfies the following conditions
{F, G} = −{G, F } ,
{F, {G, H}} + {G, {H, F }} + {H, {F, G}} = 0
for arbitrary functions F, G, H.
Thus, the Poisson bracket introduces on F(M ) the structure of an infinite-

dimensional Lie algebra. The bracket also satisfies the Leibnitz rule
{F, GH} = {F, G}H + G{F, H}
and, therefore, it is completely determined by its values on the basis elements xi :
{xj , xk } = J jk
which can be written as follows
{qi , qj } = 0 , {pi , pj } = 0 , {pi , qj } = δji .
The Hamiltonian equations can be now rephrased in the form
ẋj = {H, xj } ⇔ ẋ = {H, x} = XH .
A Hamiltonian system is characterized by a triple (M, {, }, H): a phase space

M , a Poisson structure {, } and by a Hamiltonian function H. The vector field XH
is called the Hamiltonian vector field corresponding to the Hamiltonian H. For any
function F = F (p, q) on phase space, the evolution equations take the form
dF
= {H, F }
dt
– 120 –
Again we conclude from here that the Hamiltonian H is a time-conserved quantity
dH
= {H, H} = 0 .
dt
Thus, the motion of the system takes place on the subvariety of phase space defined
by H = E constant.
In the case under consideration the matrix J is non-degenerate so that there

exist the inverse
J −1 = −J
which defines a skew-symmetric bilinear form ω on phase space
ω(x, y) = (x, J −1 y) .
In the coordinates we consider it can be written in the form

X
ω= dpj ∧ dq j .
j
This form is closed, i.e. dω = 0.

A non-degenerate closed two-form is called symplectic and a manifold endowed
with such a form is called a symplectic manifold. Thus, the phase space we consider
is the symplectic manifold.
Imagine we make a change of variables y j = f j (xk ). Then
∂y j k j km x j km ∂y
p
ẏ j = k
ẋ = Ak J ∇ m H = Ak J m p
∇y H
∂x
|{z} ∂x
Ajk
or in the matrix form

ẏ = AJAt · ∇y H .
The new equations for y are Hamiltonian if and only if
AJAt = J
and the new Hamiltonian is H̃(y) = H(x(y)).

Transformation of the phase space which satisfies the condition
AJAt = J
is called canonical. In case A does not depend on x the set of all such matrices form
a Lie group known as the real symplectic group Sp(2n, R) . The term “symplectic
– 121 –
group” was introduced by Herman Weyl. The geometry of the phase space which
is invariant under the action of the symplectic group is called symplectic geometry.
Symplectic (or canonical) transformations do not change the symplectic form ω:
ω(Ax, Ay) = −(Ax, JAy) = −(x, At JAy) = −(x, Jy) = ω(x, y) .
In the case we considered the phase space was Euclidean: M = R2n . This is not
always so. The generic situation is that the phase space is a manifold. Considera-
tion of systems with general phase spaces is very important for understanding the
structure of the Hamiltonian dynamics.
Dynamical systems with symmetries
Let g(t) will be a one-parametric group of transformations of the phase space:

x → g(t)x. This group does not need to coincide with the one generated by the
Hamiltonian H. The action of this group is called Hamiltonian if there exists a
function C such that
d
g(t)x|t=0 = J · ∇C .
dt
The flow of any function F under the one-parameter group generated by C is then
d d
δF ≡ F (g(t)x)|t=0 = ∇F · g(t)x|t=0 = (∇F, J · ∇C) = {F, C}
dt dt
If we take F = H, we get
δH ≡ {H, C} .
Thus, if Ċ = {H, C} = 0, i.e. if C is an integral of motion, then it generates the
symmetry transformations which leave the Hamiltonian invariant. Infinitezimally,
the symmetry transformations are realized as
δF = {F, C} .
There could be several one-parametric groups which are one-parametric subgroups of

a non-abelian Lie G, the latter being the symmetry of the Hamiltonian. Accordingly,
there are the integrals of motion Ci , i = 1, . . . , dim G. Since Ci are integrals of
motion, from the Jacobi identity
{{Ci , Cj }, H} + {{H, Ci }, Cj } + {{Cj , H}, Ci } = 0
we conclude that {{Ci , Cj }, H} = 0, i.e. {Ci , Cj } is an integral of motion. If one can

chose the functions Ci is such a way that they form the Lie algebra of G under the
Poisson bracket:
{Ci , Cj } = fijk Ck ,
then the corresponding action of G on the phase space is called Poisson. Here fijk
are the structure constants of the Lie algebra of G.
– 122 –
B. OPE and conformal blocks
Every conformal (primary) operator enters in the Operator Product Expansion with
its conformal family (descendants). Contribution of the entire conformal family
associated to a primary operator O into the OPE is called the conformal block.
Conformal blocks are completely fixed by conformal symmetry. As an example, let
us show how to find the conformal block associated to the scalar primary operator O
of conformal dimension ∆O , which arises in the OPE of two scalar operators A and
B of conformal dimensions ∆A and ∆B , respectively.
Assume the Operator Product Expansion
X∞
1 1 k
A(x)B(0) = 1 Λ (x, ∂)O(0) + . . . (B.1)
(x2 ) 2 (∆A +∆B −∆O ) k=0
k!
Here dots indicate the conformal families of other primary operators. We assume
that all primary operators are orthogonal w.r.t. to the two-point functions
CO
hO(0)O(y)i = .
(y 2 )∆O
The three-point functions are fixed by conformal symmetry. In particular,
CABO
hA(x)B(0)O(y)i = 1
(∆A +∆B −∆O ) 1
(∆B +∆O −∆A ) 1
(x2 ) 2 (y 2 ) 2 ((x − y)2 ) 2 (∆A +∆O −∆B ) .
Plugging the OPE into the tree-point function we get
X∞
1 1 r
hA(x)B(0)O(y)i = 1 Λ (x, −∂y )hO(0)O(y)i.
(x2 ) 2 (∆A +∆B −∆O ) r=0
r!
Thus, compatibility of the 3-point function with the OPE results into
X 1 ∞
CABO 1 1
1 1 = Λk (x, −∂y ) 2 ∆ .
C0 (y ) 2 B O A ((y − x) ) 2 A O B
2 (∆ +∆ −∆ ) 2 (∆ +∆ −∆ )
k=0
k! (y ) O
Taking into account that e−x∂y is the shift operator acting as
e−x∂y f (y) = f (y − x) ,
the last relation may be written as

∞
CABO X 1 1 k 1 X 1 k 1
1 (−x · ∂ y ) 1 = Λ (x, −∂ y) 2 ∆ .
C0 k=0 k! (y 2 ) 2 (∆B +∆O −∆A ) (y 2 ) 2 (∆A +∆O −∆B ) k=0
k! (y ) O
It suggests to define
CABO a,b
Λk (x, ∂y ) = Qk (x, ∂y ),
C0
– 123 –
where the operator Qk (x, ∂y ) is defined by the relation
(y 2 )a (x · ∂y )k (y 2 )b = Qk (x, ∂y )(y 2 )a+b (B.2)
and a, b are given by

1 1
a = − (∆B + ∆O − ∆A ), b = − (∆A + ∆O − ∆B ).
2 2
The explicit form of the operator Qk (x, ∂y ) is found by the Fourier transform. Indeed,
we have
d Z
2 a 2a+d d/2 Γ(a + 2 ) 1 e−ip·y
(y ) = 2 π (B.3)
Γ(−a) (2π)d (p2 )a+d/2
and similar for the others. Substituting the Fourier transform of every function of
y 2 one gets
Z
1 Γ(a + d2 )Γ(b + d2 ) Γ(−a − b) e−ipy
dpdq (−ixq)k (B.4)
π d/2 Γ(a + b + d2 ) Γ(−a)Γ(−b) (p2 )a+d/2 (q 2 )b+d/2
Z
e−ipy
= dp 2 a+b+d/2 Qa,b k (x, −ip),
(p )
where Qa,b
k (x, −ip) is defined by
Qa,b
k (x, ∂y )e
ipy
= Qa,b
k (x, −ip)e
ipy
,
and we have used the change of variables p → p − q.

From here one gets that Qa,b
k (x, −ip) is given by the integral
Z
1 Γ(a + d2 )Γ(b + d2 ) Γ(−a − b) 2 a+b+d/2 (−ixq)k
Qa,b
k (x, −ip) = d/2 (p ) dq .
π Γ(a + b + d2 ) Γ(−a)Γ(−b) ((p − q)2 )a+d/2 (q 2 )b+d/2
Thus, the problem is reduced to evaluation of the integral

Z
(−ixq)k
I(α1 , α2 ) = dq . (B.5)
((p − q)2 )α1 (q 2 )α2
One has
Z Z
Γ(α1 + α2 ) 1 α1 −1 α2 −1 k (xq)k
I(α1 , α2 ) = dtt (1 − t) (−i) dq
Γ(α1 )Γ(α2 ) 0 [(q − tp)2 + t(1 − t)p2 ]α1 +α2
Z 1 Z
Γ(α1 + α2 ) (xq + txp)k
= dttα1 −1 (1 − t)α2 −1 (−i)k dq 2
Γ(α1 )Γ(α2 ) 0 [q + t(1 − t)p2 ]α1 +α2
Z [k/2]
X Z
Γ(α1 + α2 ) 1 α1 −1 α2 −1 k 2m k−2m (xq)2m
= dtt (1 − t) (−i) Ck (txp) dq 2 ,
Γ(α1 )Γ(α2 ) 0 m=0
[q + t(1 − t)p2 ]α1 +α2
– 124 –
k!
where Ck2m = (2m)!(k−2m)! . There appear only even powers 2m since for odd powers
the internal integral is zero. Thus, we now need to evaluate the integral
Z
(xq)2m
L= dq .
[q 2 + h]γ
Under the integral the symmetric product qi1 ...qi2m may be substituted for
(q 2 )m
qi1 ...qi2m = (δi i ...δi2m−1 i2m + permutations), (B.6)
d(d + 2)...(d + 2(m − 1)) 1 2
where on the r.h.s. all non-trivial permutations are present. The total number of
this permutations is 2(2m)!
m m! . Therefore, under the integral one may substitute
(x2 )m (q 2 )m (2m)!
(xq)2m = xi1 ...xi2m qi1 ...qi2m = .
d(d + 2)...(d + 2(m − 1)) 2m m!
Now we have
Z Z ∞
(q 2 )m π d/2 (q 2 )m+d/2−1
dq 2 = dq 2 ,
[q + h]γ Γ(d/2) 0 [q 2 + h]γ
h
Performing the change of variables t = q 2 +h
, we arrive at
Z Z
(q 2 )m π d/2 m+d/2−γ 1 γ−m−d/2−1
dq 2 γ
= h dtt (1 − t)m+d/2−1
[q + h] Γ(d/2) 0
d/2
π Γ(γ − m − d/2)Γ(m + d/2) m+d/2−γ
= h .
Γ(d/2) Γ(γ)
Thus, we evaluate the integral L

Z
(xq)2m
L= dq =
[q 2 + h]γ
(x2 )m (2m)! π d/2 Γ(γ − m − d/2)Γ(m + d/2) m+d/2−γ
= h
d(d + 2)...(d + 2(m − 1)) 2m m! Γ(d/2) Γ(γ)
(2m)! Γ(γ − m − d/2) 2 m m+d/2−γ
= π d/2 m (x ) h .
4 m! Γ(γ)
Finally, one gets
[k/2]
d/2 Γ(α1 + α2 ) X 2m
I(α1 , α2 ) = π C (−ixp)k−2m (−1)m
Γ(α1 )Γ(α2 ) m=0 k
Z 1
(2m)! Γ(α1 + α2 − m − d/2) 2 m
× dttα1 +k−2m−1 (1 − t)α2 −1 m
(x ) (t(1 − t)p2 )m+d/2−α1 −α2 ,
0 4 m! Γ(α1 + a2 )
– 125 –
where we have substituted γ = α1 + α2 and h = t(1 − t)p2 . This is simplified to
[k/2]
π d/2 X (2m)!
I(α1 , α2 ) = Ck2m m (−ixp)k−2m (−x2 )m (p2 )m+d/2−α1 −α2
Γ(α1 )Γ(α2 ) m=0 4 m!
Z 1
× dttk−m+d/2−α2 −1 (1 − t)m+d/2−α1 −1 Γ(α1 + α2 − m − d/2).
0
Thus, the final formula reads as
[k/2]
X µ ¶m
π d/2 2m (2m)! k−2m 1 2 2
I(α1 , α2 ) = Ck (−ixp) − xp (p2 )d/2−α1 −α2
Γ(α1 )Γ(α2 ) m=0 m! 4
Γ(k − m + d/2 − α2 )Γ(m + d/2 − α1 )
× Γ(α1 + α2 − m − d/2).
Γ(k + d − α1 − α2 )
For Qa,b
k (x, −ip) one therefore finds
[k/2] µ ¶m
1 Γ(−a − b) X 2m (2m)! 1 2 2
Qa,b
k (x, −ip) = C (−ixp)k−2m
− xp
Γ(a + b + d/2) Γ(−a)Γ(−b) m=0 k m! 4
Γ(k − m − a)Γ(m − b)
× Γ(a + b + d/2 − m).
Γ(k − a − b)
(−1)m
By using the relation (a)−m = (1−a)m
it can be further simplified to give
[k/2]
X µ ¶m
1 k!(−a)m (−b)k−m 1 2
Qa,b
k (x, ∂y ) = (x · ∂y )k−2m
− x ∆y .
(−a − b)k m!(k − 2m)!(−d/2 − a − b + 1)m 4
m=0
Further summation gives the conformal block of the scalar field and it is per-
formed by changing the order of the summation and the shift of the summation
variable:
∞
X 1 a,b
Q (x, ∂y ) =
k! k
k=0
X∞ ∞
X µ ¶m
1 (−a)m (−b)k−m k−2m 1 2
(x · ∂y ) − x ∆y
(−a − b)k m!(k − 2m)!(−d/2 − a − b + 1)m 4
m=0 k=2m
X∞ µ ¶ mX∞
(−a)m 1 (−b)k+m (x∂y )k
= − x2 ∆y .
m!(−d/2 − a − b + 1)m 4 (−a − b)k+2m k!
m=0 k=0
Since
(−b)k+m (−b)m (−b + m)k
=
(−a − b)k+2m (−a − b)2m (−a − b + 2m)k
we finally get
∞
X 1 a,b
Q (x, ∂y ) =
k! k
k=0
∞
X µ ¶m
(−a)m (−b)m 1 1 2
= 1 F1 (−b + m; −a − b + 2m; x∂y ) − x ∆y ,
(−µ − a − b + 1)m m! (−a − b)2m 4
m=0
– 126 –
where 1 F1 is a degenerate hypergeometric function:
∞
X (α)k xk
1 F1 (α, β; x) = . (B.7)
k=0
(β)k k!
C. Useful formulae
In discussion of the central charge of the Virasoro algebra we encounter the sum
n
X
Sn = q2 .
q=1
Here we present a general method of computing this and similar sums. The method
is based on considering the generating function
∞
X
℘(x) = Sn xn ,
n=1
so that
1 ³ ∂ n℘ ´
Sn = |x=0 .
n! ∂xn
For x < 1 we have
∞
X ∞ X
X n ∞
X X∞ ∞
X
n 2 n 2 q 2 xq
n
℘(x) = Sn x = q x = q x = .
n=1 n=1 q=1 q=1 n=q q=1
1−x
We further notice that

∞ ∞ ∞ ∞ ∞ ∞
∂2 X q X q−2 1 X 2 q 1 X q 1 X 2 q 1 X q
x = q(q − 1)x = 2 q x − 2 qx = 2 q x − 2 qx .
∂x2 q=1 q=2
x q=2
x q=2
x q=1
x q=1
Thus,
∞
X ³ ∂2 x 1 ∂ x ´ x2 + x
2 q 2
q x =x + =
q=1
∂x2 1 − x x2 ∂x 1 − x (1 − x)3
and, therefore, we obtain the generating function
x2 + x 1 3 2
℘(x) = 4
= 2
− 3
+ .
(1 − x) (1 − x) (1 − x) (1 − x)4
Finally we compute
µ ¶
1 ³ ∂n℘ ´ 1 2 · 3 · · · (n + 1) 3 · 4 · · · (n + 2) 4 · 5 · · · (n + 3)
= −3 +2 .
n! ∂xn n! (1 − x)n+2 (1 − x)n+3 (1 − x)n+4
The last formula results into
³ 3 1 ´ 1
Sn = (n + 1) 1 − (n + 2) + (n + 2)(n + 3) = n(n + 1)(2n + 1) .
2 3 6
– 127 –
D. Riemann normal coordinates
Consider a Riemannian manifold M of dimension n with coordinates xi , 1 = 1, . . . , m.
The geodesic equation is
ẍi + Γijk (x)ẋj ẋk = 0 .
Let us consider two points p and q with coordinates xi and xi + δxi respectively. We
assume that these points are closed so there is a unique geodesic connecting them.
A parameter t on the geodesic can be chosen proportional to the length of the
arc connecting these two points (the natural parameter). The solution xi (t) can be
chosen so that xi (0) ≡ xi and xi (1) = xi + δxi . The tangent vector to the geodesic
at t = 0 is defined by ξ i = ẋi (0). Then equation for the geodesic can be solved
perturbatively by assuming the following expansion
xi (t) = xi + ci1 t + ci2 t2 + · · · .
Plugging this into the geodesic equation one finds

1 1
xi (t) = xi + ξ i t − Γij1 j2 (x)ξ j1 ξ j2 t2 − Γij1 j2 j3 (x)ξ j1 ξ j2 ξ j3 t3 − · · · ,
2 3!
where
Γij1 j2 j3 = ∂j1 Γij2 j3 − Γlj1 j2 Γilj3 − Γlj1 j2 Γij3 l .
Here all the quantities are evaluated at xi . At t = 1 we have xi (1) = xi + δxi so that
1 1
xi + δxi = xi + ξ i − Γij1 j2 (x)ξ j1 ξ j2 − Γij1 j2 j3 (x)ξ j1 ξ j2 ξ j3 − · · ·
2 3!
Hence, the point xi + δxi is parameterized by the tangent vector ξ i . We can there-
fore take ξ i as the new coordinate system on our manifold. The coordinates ξ i are
called Riemann normal coordinates. They are used to define a map of an open
neighbourhood U of the zero-vector 0 ∈ Tx M to the manifold M:
expx : U ∈ Tx M → M
which is well-defined diffeomorphism of U on its image. Now we are parameterizing

the points of the curved manifold with tangent vectors. The image of the geodesic
through a point x in the Riemann normal coordinates is just a straight line.
How metric will look in the new coordinate system? Upon changing the coordi-
nates form xi to ξ i the metric transforms in the standard way
∂(xk + δxk ) ∂(xl + δxl )
gij0 (ξ) = gkl (x + δx)
∂ξ i ∂ξ j
We have
∂(xk + δxk )
= δik − Γkij ξ j + · · ·
∂ξ i
– 128 –
and also
³ 1 ´
gkl (x + δx) = gkl xi + ξ i − Γij1 j2 ξ j1 ξ j2 + · · · = gkl (x) + ∂k gij ξ k + · · ·
2
Thus, at the linearized level we find that
³ ´³ ´³ ´
gij0 (ξ) = δik − Γkin ξ n δjl − Γljm ξ m gkl (φ) + ∂p gkl ξ p + · · ·
³ ´
= gij (x) + ∂n gij − Γkin gkj − Γkjn gki ξ n + · · ·
| {z }
Dn gij
We see that
gij0 (ξ) = gij (x) + Dk gij (x)ξ k + · · · (D.1)
It is important to realize that the coordinates ξ k depend on x because they define

the tangent vector to the manifold at the point x. Under reparametrizations of
x → x0 the object ξ k (x) transforms as a vector! The expansion (D.1) is covariant
under general coordinate transformations x → x0 because it includes the tensorial
quantities only.
In the case of the minimal connection (connection compatible with the metric)
we have Dk gij = 0. This, the expansion of the metric start from the quadratic order
in ξ i . Extending this calculation to higher orders in ξ one finds
1 1
gij0 (ξ) = gij (x) − Rik1 jk2 ξ k1 ξ k2 − Dk1 Rik2 jk3 ξ k1 ξ k2 ξ k3 + · · ·
3 3!
Also we recall the transformation property of the Christoffel connection
∂xk ³ k ∂xp ∂xq ∂ 2 xk ´

0
0
Γkp0 q0 (ξ) = Γ + ,
∂xk pq ∂xp0 ∂x0q0 ∂xp0 ∂xq0
where xi ≡ φi and x0i ≡ φi + π i . Expanding the r.h.s. of the last equation in ξ i it is
easy to see that expansion does not contain the constant piece because the constant
terms in the bracket cancel against each other, in other words, in the Riemann normal
coordinates we have
0
Γkp0 q0 (ξ) = O(ξ)
In the Riemann normal coordinate system die to the vanishing of Γkpq at ξ = 0 the
geodesic equation takes a form of the free motion λ̈ = 0. This is coordinate system
corresponds to the rest frame of a freely falling observer.
– 129 –
E. Exercises
Exercise 1. Show that the Hessian matrix associated to the Nambu-Goto Lagrangian
has for each σ two zero eigenvalues corresponding to Ẋ µ and X 0µ .
Exercise 2. How reparametrization invariance can be used to bring the equation

   
0 0µ 0 2 µ 0 µ 2 0µ
∂  (ẊX )X − (X ) Ẋ  ∂  (ẊX )Ẋ − (Ẋ) X 
q + q = 0.
∂τ 0 2 2 02 ∂σ 0 2 2 02
(ẊX ) − Ẋ X (ẊX ) − Ẋ X
to the simplest form?
Exercise 3. The Polyakov string. Prove that equations of motion for the fields
µ
X imply conservation of the two-dimensional stress-energy tensor
∇µ Tµν = 0
Exercise 4. Show that Tαβ = 0 implies that the end points of the open string
move with the speed of light.
Exercise 5. Non-relativistic string.
• Consider a string in equilibrium on the x-axis between (0, 0) and (L, 0) and
suppose that the infinitesimal parts of the string can move only in the y-
direction. Derive the Lagrangian with µ the mass density and T the string
tension.
• Derive the equation of motion from this Lagrangian and keep explicit attention
to the boundary conditions.
• Analyze the boundary terms. What must you impose in order to have a sta-
tionary action?
• Construct the momentum function P .
• Calculate the time derivative of the momentum (consider the boundary condi-
tions). What do you conclude?
• Fourier transform the x-coordinate and solve the eom. Do this for both bound-
ary conditions.
– 130 –
Exercise 6. Show that the Polyakov action is invariant under reparametrizations
δX µ = ξ α ∂α X µ
δhαβ = ∇α ξβ + ∇β ξα
√ √
δ( h) = ∂α (ξ α h)
Exercise 7. Show that the Weyl invariance implies the tracelessness of the stress-
energy tensor Tαβ .
Exercise 8. Show that the Gauss-Bonnet term

Z √
1
χ= d2 σ hR
4π
is topological, i.e. it vanishes under smooth variations of the world-sheet metric hαβ .
Take into account that in 2dim the Ricci tensor is proportional to Ricci scalar and
also √ ³ ´
1
δ( hR) ∼ Rαβ − hαβ R δhαβ .
2
Exercise 9. Let S(q, t; q0 , t0 ) be the action of the classical path between (q0 , t0 )
and (q, t). Show that
∂S
= p(t) ,
∂q
where p(t) is the conjugate momentum of q at time t. Show that
∂S ∂S
= −H(q, ),
∂t ∂q
p2
where H is the Hamiltonian. Suppose that H(q, p) = 2m
+ V (q) and define
i
ψ(q, t) = e ~ S(q,t;q0 ,t0 ) .
Show that the schrödinger equation approximately holds for ψ,

∂ψ ³ ∂´
i~ = H q, −i~ ψ + O(~) .
∂t ∂q
This is of course related to Dirac’s idea that the phase of the wave function is pro-
portional to the classical action.
– 131 –
Exercise 10.
• Show that the constraints
C1 = Pµ P µ + T 2 Xµ0 X 0µ , C2 = Pµ X 0µ
have the following Poisson brackets
{C1 (σ), C1 (σ 0 )} = 4T 2 ∂σ C2 (σ)δ(σ − σ 0 ) + 8T 2 C2 (σ)∂σ δ(σ − σ 0 ) ,

{C1 (σ), C2 (σ 0 )} = ∂σ C1 (σ)δ(σ − σ 0 ) + 2C1 (σ)∂σ δ(σ − σ 0 ) ,
{C2 (σ), C1 (σ 0 )} = ∂σ C1 (σ)δ(σ − σ 0 ) + 2C1 (σ)∂σ δ(σ − σ 0 ) ,
{C2 (σ), C2 (σ 0 )} = ∂σ C2 (σ)δ(σ − σ 0 ) + 2C2 (σ)∂σ δ(σ − σ 0 ) .
• Define the linear combinations

1 1
T++ = (C1 + 2T C2 ) = (Pµ + T Xµ0 )2 ,
8T 2 8T 2
1 1
T−− = (C1 − 2T C2 ) = (Pµ − T Xµ0 )2 .
8T 2 8T 2
and show that their Poisson algebra is
1 ³ ´
{T++ (σ), T++ (σ 0 )} = ∂σ T++ (σ)δ(σ − σ 0 ) + 2T++ (σ)∂σ δ(σ − σ 0 ) ,
2T
1 ³ ´
{T−− (σ), T−− (σ 0 )} = − ∂σ T−− (σ)δ(σ − σ 0 ) + 2T−− (σ)∂σ δ(σ − σ 0 ) ,
2T
0
{T++ (σ), T−− (σ )} = 0 .
Exercise 11. For the closed string case define

Z 2π
−
Lm = 2T dσ eimσ T−− (σ, τ )
Z0 2π
+
L̄m = 2T dσ eimσ T++ (σ, τ ) .
0
Show that for any integer m the generators Lm and L̄m are time-independent.
Exercise 12. Compute the Poisson brackets of the constraints Lm , L̄m . What
kind of constraints they are, i.e. the first or the second class?
Exercise 13. It is known in curved space-time that we can transform the metric
locally in the neighborhood of a point xµ = 0 to the following form gµν (x) = ηµν −
– 132 –
1
R
3 µσνρ
xσ xρ , this are Riemann normal coordinates. So that the deviation of g from
the flat metric η is only second order in x (this is a coordinate system of an observer
in free fall). Suppose we have a coordinate system x with metric expanded around
0,
1
gµν (x̃) = gµν (0) + ∂σ gµν (0)x̃σ + ∂σ ∂ρ gµν (0)x̃σ x̃ρ .
2
We want to transform this using a coordinate transformation x̃ → x = x(x̃) to
Riemann normal coordinates. We expand the coordinate transformation to third
order (zeroth order is zero)
∂xµ ν 1 ∂xµ ν σ 1 ∂xµ

xµ = x̃ + x̃ x̃ + x̃ν x̃σ x̃ρ .
∂ x̃ν 2 ∂ x̃ν ∂ x̃σ 6 ∂ x̃ν ∂ x̃σ ∂ x̃ρ
We can use the first term to bring g to η. Show that after we have done this there
µ
are 21 d(d − 1) remaining degrees of freedom left for ∂x
∂ x̃ν
, where d is the dimension.
Where do these remaining degrees of freedom correspond to?
Show that we can bring ∂σ gµν to zero using the second term of the transformation,
by counting degrees of freedom.
For arbitrary dimension there are not enough degrees of freedom to put ∂σ ∂ρ gµν to
zero. Count the number of remaining degrees of freedom and show that it equals
1 2 2
12
d (d −1). This is precisely the number of independent components of the Riemann
tensor.
Exercise 14. Solve equation of motion ¤X µ = 0 for the case of open string (take
into account the open string boundary conditions).
Exercise 15. Consider solution of the closed string equations of motion
X µ (σ, τ ) = XLµ (τ + σ) + XRµ (τ − σ)
where
1 pµ i X 1 µ −in(τ −σ)
XRµ (τ − σ) = xµ + (τ − σ) + √ αn e
2 4πT 4πT n6=0 n
µ 1 µ pµ i X 1 µ −in(τ +σ)
XL (τ + σ) = x + (τ + σ) + √ ᾱn e
2 4πT 4πT n6=0 n
Derive the Poisson algebra of the variables (xµ , pµ , αnµ , ᾱnµ ) by using the fundamental
Poisson brackets of X µ (σ), P µ (σ) variables.
– 133 –
Exercise 16. Prove that that (closed) string has an infinite set of integrals of
motion: for any function f the quantities
Z 2π Z 2π
−
Lf = dσf (σ )T−− , L̄f = dσ f (σ + )T++ ,
0 0
are conserved.
Exercise 17. Obtain an expression for the Virasoro generators Lm and L̄m in
terms of string oscillators.
d
Exercise 18. Show that the operators Dn = −ieinθ dθ obey the Virasoro algebra.
Exercise 19. Consider an open string solution 0 ≤ σ ≤ π:
X 0 = t = Lτ
X 1 = L cos σ cos τ ,
X 2 = L cos σ sin τ ,
X i = 0, i = 3, . . . , d − 1 .
• Show that this solution satisfies the Virasoro constraints and open string bound-
ary conditions.
• Compute the mass of string.
• Compute the angular momentum J ≡ J12 of string.

1
• Show that J = α0 M 2 , where α0 = 2πT
is called a slope of the Regge trajectory.
Exercise 20. By using the Poisson brackets between the generators of the
Poincare group
{P µ , P ν } = 0
{P µ , J ρσ } = η µσ P ρ − η µρ P σ
{J µν , J ρσ } = η µρ J νσ + η νσ J µρ − η νρ J µσ − η µσ J νρ
show that for a certain choice of a function f the following expression

1¡ ¢
J2 = Jαβ J αβ + f (P 2 )Pα J αλ P β Jβλ , P 2 ≡ Pµ P µ .
2
– 134 –
is an invariant of the Poincare group. Find the corresponding f .
Exercise 21*. Show that for any open classical string solution the following
inequality holds √
J ≡ J 2 ≤ α0 M 2 ,
(Hint. Make the computation in the static gauge X 0 = t = τ and use the Schwarz
inequality |(f¯, g)|2 ≤ (f¯, f )(ḡ, g) which holds for any two complex functions f and
g.)
Exercise 22. Show that the generators of the angular momentum commute with
the Virasoso generators
{Lm , J µν } = 0 .
Exercise 23. By using the first-order formalism and imposing the light-cone
gauge for the open string case
• Solve the Virasoro constraints,
• Find the open string light-cone Hamiltonian.
Exercise 24. Show that Virasoro constraints Tαβ = 0 which we have found in
the conformal gauge can be directly solved in the light-cone gauge without using the
first-order formalism. Show how the conformal gauge Hamiltonian turns into the
light-cone Hamiltonian upon substitution the light-cone gauge conditions and the
Virasoro constraints.
Exercise 25. Rewrite the Poisson algebra of the Poincaré generators in terms of
light-cone coordinates P ± , P i and J i± , J ij , J +− .
Exercise 26. For the closed string case in the light-cone gauge compute the
Poisson brackets between the zero mode variables p+ , p− , pi , x− , xi . The full answer
is given in the lecture notes.
Exercise 27. Proof the fulfilment of the Poisson algebra relations between the
generators {J i+ , J j+ } and {J i+ , J j− } .
Exercise 28. The Virasoro algebra relation takes the form

[Lm , Ln ] = (m − n)Lm+n + A(m)δm+n
– 135 –
where A(m) is a function of m. The aim of this exercise is to find the constraints on
A(m) that follow from the condition that the relations above define the Lie algebra.
• That does the antisymmetry requirement on a Lie algebra tells you about
A(m)? What is A(0)?
• Consider the Jacobi identity for the generators Lm , Ln and Lk with m+n+k =
0. Show that
(m − n)A(k) + (n − k)A(m) + (k − m)A(n) = 0.
• Use the last equation to show that A(m) = αm and A(m) = βm3 , for constants
α and β, yield consistent central extensions.
• Consider the last equation with k = 1. Show that A(1) and A(2) determine all
A(n)
Exercise 29.
• Use the Virasoro algebra to show that if a state is annihilated by L1 and L2

then it is annihilated by all Ln with n ≥ 1.
• Consider the Virasoro generators L0 , L1 and L−1 . Write out the relevant
commutators. Do these operators form a subalgebra of the Virasoro algebra?
Is there a central term here?
Exercise 30. Consider open string. The fundamental commutation relation is
[X µ (σ, τ ), P ν (σ 0 , τ )] = iη µν δ(σ − σ 0 ) , σ ∈ [0, π] .
• Show that consistency with the oscillator expansion implies that

∞
01 X
δ(σ − σ ) = cos nσ cos nσ 0
π n=−∞
• Why the fundamental commutation relation compatible with open string bound-
ary conditions?
• Prove this representation for the δ-function by using the fact that any function
f (σ) with σ ∈ [0, π] and vanishing derivative at σ = 0, π can be expanded as
∞
X
f (σ) = An cos nσ .
n=0
– 136 –
Exercise 31. Compute in the light-cone gauge the commutator of the orbital
momenta
[ì− , `j− ] =?
Exercise 32. Show that in the quantum theory the eigenvalues of the covariant
number operator
X∞
µ
N= α−n αn,µ
n=1
are always nonnegative.
Exercise 33. Using the previous exercise show that for any fixed state all but
a finite number of positively moded Virasoro operators automatically annihilate the
state without imposing any conditions. More precisely, show that any state |Φi with
the number eigenvalue N ≥ 0 automatically satisfies
Ln |Φi = 0 for n > N.
Exercise 34. Compute the open string propagator

¡ ¢
hX(τ, σ)X(τ 0 , σ 0 )i = T X(τ, σ)X(τ 0 , σ 0 ) − : X(τ, σ)X(τ 0 , σ 0 ) : .
Exercise 35. Show that the vertex operator of the open string
µ µ
P∞ kµ α−n
pµ
P∞ kµ αn −inτ
√1 einτ µ
ikµ (x + πT τ ) − √1 e
V (k, τ ) = e| πT n=1
{z
n
} |e {z } |e
πT n=1
{z
n
}
V− V0 V+
is the conformal operator with the conformal dimension ∆ = α0 k 2 .
Exercise 36. Compute the two-point correlation function of the tachyon vertex
operators
h0|V (k2 , τ2 )V (k1 , τ1 )|0i
Exercise 37. Compute the three-point correlation function of the tachyon vertex
operators
h0|V (k3 , τ3 )V (k2 , τ2 )V (k1 , τ1 )|0i
– 137 –
Exercise 38. Compute the three-point correlation function of the tachyon vertex
operators
h0|V (k4 , τ4 )V (k3 , τ3 )V (k2 , τ2 )V (k1 , τ1 )|0i
Exercise 40. Verify that ∂−n X µ is a conformal operator. Find the corresponding
conformal dimension. Find the singular terms of the OPE
T−− (τ, σ) ∂−n X µ (τ 0 , σ 0 ) .
Exercise 41. Find the singular terms of the OPE
T−− (τ, σ) V (k, τ 0 , σ 0 ) ,
where V (k, τ 0 , σ 0 ) is the vertex operator of tachyon.
Exercise 42. Find the singular terms of the OPE
T−− (τ, σ) T−− (τ 0 , σ 0 ) .
Exercise 43. Suppose that there are i = 1, . . . , n grassman (anticommuting)

variables ηi and η̄i . Let us define the integration rules as
Z Z
dη = 0 , dηη = 1
for any ηi and η̄i . Show that for any n × n matrix M the following formula is valid
Z
detM = dηdη̄ eη̄M η .
Here η̄M η ≡ η̄i Mij ηj .
Exercise 44. Show that the stress-tensor for the ghost fields implies the following
expression for the ghost Virasoro generators of the closed string
∞
X
Lgh
m = (m − n) : bm+n c−n :
n=−∞
X∞
L̄gh
m = (m − n) : b̄m+n c̄−n :
n=−∞
– 138 –
Exercise 45. Consider conformal transformations z → z 0 = f (z). By definition,
primary fields are the fields which transform as tensors under conformal transforma-
tions ³ ∂z 0 ´h ³ ∂ z̄ 0 ´h̄
φ(z, z̄) → φ0 (z, z̄) = φ(z 0 (z), z̄ 0 (z̄))
∂z ∂ z̄
Find how a primary field transforms under infinitezimal conformal transformations
φ → φ + δξ,ξ̄ φ, where z 0 = z + ξ(z).
Exercise 46. Show that conformal fields of weight h have the following mode
expansion X
φ(z) = z −n−h φn .
n∈integer
Exercise 47. In the radial quantization products of fields is defined by putting

them in the radial order. Using the radial order prescription show that the conformal
transformation
δξ φ(w) = [Tξ , φ(w)]
can be written in the form
I
dz
δξ φ(w) = ξ(z)T (z)φ(w)
Cw 2πi
Here Cw is a small contour in a complex plane around point z.
Exercise 48. Using the previous exercise together with the Cauchy-Riemann
formula I
dz f (z) f (n−1) (w)
n
=
Cw 2πi (z − w) (n − 1)!
show that any conformal field must have the following R-ordered operator product
with T (z):
hφ(w) ∂φ(w)
T (z)φ(w) = 2
+ + regular terms
(z − w) (z − w)
Exercise 49. Show that the following operator product of the stress tensor
c/2 2T (w) ∂T (w)
T (z)T (w) = 4
+ 2
+ + regular terms
(z − w) (z − w) (z − w)
corresponds to the following transformation law
c
δξ T (z) = ∂ 3 ξ(z) + 2∂ξ(z)T (z) + ξ(z)∂T (z)
12
– 139 –
Exercise 50. Under finite transformations z → f (z) the stress tensor transforms
as
c
T (z) → T 0 (z) = (∂f )2 T (f (z)) + D(f )z
12
Here
∂f (z)∂ 3 f (z) − 23 (∂ 2 f (z))2
D(f )z =
(∂f )2
az+b
is the Schwarzian derivative. Show that if f (z) = cz+d
then D(f )z = 0.
Exercise 51. Consider a parallelogram on the complex plane determined by the

following identification
z ≡ z + nλ1 + mλ2 , n, m ∈ Z
Show that a general transformation
λ1 → dλ1 + cλ2 , λ2 → bλ1 + aλ2
with the condition ad − bc = 1 preserves the area of the parallelogram.
Exercise 52. Show that the modular transformations

1
T : τ → τ + 1, S: τ →−
τ
applied to the fundamental region of the modular group
n 1 1 o
Mg=1 = − ≤ Reτ ≤ 0, |τ |2 ≥ 1 ∪ 0 < Reτ < , |τ |2 > 1
2 2
generate the whole upper-half plane.
2πi
Exercise 53. Show that τ = i and τ = e 3 are the fixed points of S and ST
transformations respectively.
Exercise 54. Verify that the action

Z ³ ´
1
S=− d2 σ ∂α X µ ∂ α Xµ + 2iψ̄ µ ρα ∂α ψµ .
8π
is invariant under supersymmetry transformations w
δ² X µ = i²̄ψ µ ,
1
δ² ψ µ = ρα ∂α X µ ²
2
1
δ² ψ̄ µ = − ²̄ρα ∂α X µ
2
– 140 –
provided the parameter ² satisfies the following equation
ρβ ρα ∂β ² = 0 .
Exercise 55. If in the previous exercise the parameter ² is constant then we

deal with global supersymmetry transformations. Derive the Noether current which
corresponds to this global symmetry of the action.
Exercise 56*. Express the spin connection ω via vielbein e by using the condition
of vanishing of the torsion
a
Tαβ = Dα eaβ − Dβ eaα = 0 .
Exercise 57. Derive the on-shell algebra of supersymmetry transformations
[δ1 , δ2 ] =? .
Exercise 58. Using the Noether method derive the currents corresponding to
the Poincaré symmetry for the fermionic string. Express the corresponding Noether
charges via oscillators for both Neveuw-Schwarz and Ramond sectors.
– 141 –

Arutyunov Notes

Uploaded by

Copyright:

Available Formats

Arutyunov Notes

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Arutyunov Notes

Uploaded by

Copyright:

Available Formats

Preprint typeset in JHEP style - HYPER VERSION

Lectures on String Theory

a Institute for Theoretical Physics and Spinoza Institute, Utrecht University

1. Introduction to the course 2

3. Classical relativistic bosonic string 13

4. Quantization of bosonic string 38

5. Geometry and topology of string world-sheet 77

7. Quantum fermionic string 108

A. Dynamical systems of classical mechanics 119

B. OPE and conformal blocks 123

C. Useful formulae 127

D. Riemann normal coordinates 128

1. Introduction to the course

1. Existence of a “critical dimension”.

In 1972 Schwarz demonstrated the consistency of the superstring theory in 10 di-

Resolution of the four-fermi interaction. At high energies the weak force is

Very similar phenomenon occurs in string theory. Expanding the Einstein-

Type IIA SO(32) heterotic

Remarkably, in 1983, Alvarez-Gaumé and Witten demonstrated that anomaly can-

1. M. Green, J. Schwarz and E. Witten, “Superstring Theory”, volume I and II,

2. D. Lüst and S. Theisen, “Lectures on string theory”, Lecture Notes in Physics,

3. ’t Hooft, “Introduction to String Theory”, Lecture Notes, 2005.

5. J. Polchinski, “String theory”, volume I and II, Cambridge university press,

To get the standard kinetic energy one has to identify

In what follows we will work in units in which c = 1.

δxµ = ξ(τ )∂τ xµ as long as ξ(τ0 ) = ξ(τ1 ) = 0

Let us show this

In general the number of primary constraints is equal to the number of zero

Constraints which have vanishing Poisson bracket:

• it does not contain square root

• it admits generalization to the massless case

Introduce e(τ ) the auxiliary field on the world-line.

The first symmetry transformation is generated by the constraint p2 + m2 :

Note that here p~ = {pi } with i = 2, . . . , d.

3. Classical relativistic bosonic string

This we can write as

Here Γ = detΓαβ , where

• If ds2 = 0 the interval is light-like. Vectors v µ obeying the condition v 2 =

Ẋ µ Ẋµ = −1 + ~v 2 ≤ 0 ⇐ time − like

X 0µ Xµ0 = (X 0i )2 ≥ 0 ⇐ space − like

This condition guarantees the causal propagation of string.

3.2 The Polyakov action

The reparametrization invariance

The Weyl invariance

hαβ → e−2Λ(σ,τ ) hαβ . (3.11)

3.2.2 Equations of motion

where on the r.h.s. we specified the variation to be a diffeomorphism transformation.

Since ∇α ∂α X µ = 0 is eom for X µ we find

Only symmetric part of hαδ Γs 1 αδ hsp ∂ h

ξ γ ∂γ hαβ + hαγ ∂β ξ γ + hβγ ∂α ξ γ = Λhαβ (3.15)

and solve it assuming the conformal gauge

Second, we take α, β = τ and then α, β = σ. We get

Subtracting one from the other we get

Thus, the conformal Killing equations reduce in the conformal gauge to

By using the world-sheet light-cone coordinates σ ± = τ ± σ this can be reformulated

3.2.4 Polyakov string in the first order formalism.

The gauged-fixed Lagrangian density is

{X µ (σ, τ ), X ν (σ 0 , τ )} = {P µ (σ, τ ), P ν (σ 0 , τ )} = 0 , (3.21)