Liverpool Lectures on String Theory

The Free Bosonic String

Summary Notes
Thomas Mohaupt
Semester 1, 2008/2009

Contents
1 The
1.1
1.2
1.3
1.4
1.5
1.6
1.7

classical relativistic particle

2
Relativistic kinematics and dynamics . . . . . . . . . . . . . . . .
2
Action principle for the massive free particle: non-covariant version 3
Canonical momenta and Hamiltonian . . . . . . . . . . . . . . . .
4
Length, proper time and covariance on the world line . . . . . . .
5
Covariant action for the massive particle . . . . . . . . . . . . . .
6
Canonical momenta and Hamiltonian for the covariant action . .
9
Covariant action for massless and massive particles . . . . . . . .
9

2 The classical relativistic string

2.1 The Nambu-Goto action . . . . . . . . . . . . . . . . . .
2.2 The Polyakov action . . . . . . . . . . . . . . . . . . . .
2.2.1 Action, symmetries, equations of motion . . . . .
2.2.2 Relation to two-dimension field theory . . . . . .
2.2.3 Polyakov action in conformal gauge . . . . . . .
2.2.4 Lightcone coordinates . . . . . . . . . . . . . . .
2.2.5 Momentum and angular momentum of the string
2.2.6 Fourier expansion . . . . . . . . . . . . . . . . . .
3 The
3.1
3.2
3.3

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

11
11
14
14
15
16
17
19
21

quantized relativistic string

23
Covariant quantisation and Fock space . . . . . . . . . . . . . . . 23
Implementation of constraints and normal ordering . . . . . . . . 26
Analysis of the physical states . . . . . . . . . . . . . . . . . . . . 28
3.3.1 The scalar ground state (tachyon) of the open string . . . 29
3.3.2 The massless vector state (photon) of the open string . . 30
3.3.3 Review of the Maxwell Lagrangian . . . . . . . . . . . . . 31
3.3.4 States of the closed string . . . . . . . . . . . . . . . . . . 34
3.3.5 The closed string B-field and axion . . . . . . . . . . . . . 35
3.3.6 The graviton and the dilaton . . . . . . . . . . . . . . . . 37
3.3.7 Review of the Pauli-Fierz Langrangian and of linearized
gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

A Literature

41

1
1.1

The classical relativistic particle

Relativistic kinematics and dynamics

D-dimensional Minkowski space(-time): M = MD = R1,D1 .

Points = Events: x = (x ) = (x0 , xi ) = (t, ~x),
where = 0, . . . , D 1, i = 1, . . . , D 1 and we set c = 1.
Minkowski metric (mostly plus):


1
0

= ( ) =

ID1

Scalar product (relativistic distance between

<0
=0
x x = x x = t2 + ~x2

>0

(1)

events):
for timelike distance,
for lightlike distance,
for spacelike distance.

(2)

World line of a particle:

x(t) = (x (t)) = (t, ~x(t)) .

(3)

where t = time measured in some inertial frame.

Velocity :
d~x
.
dt

(4)

< 1(= c2 ) for massive particles (timelike worldline) ,

= 1(= c2 ) for massless particles (lightlike worldline) .

(5)

~v =
Note:

~v ~v
Proper time:
Z

t2

(t1 , t2 ) =

p
dt 1 ~v 2 .

(6)

t1

To find (t), take upper limit to be variable:

Z

dt0

(t1 , t) =

1 ~v 2 .

(7)

t1

Implies:
p
d
= 1 ~v 2 =
dt

r
1

d~x d~x

.
dt dt

(8)

NB: this makes sense for massive particles, only, because we need ~v 2 < 1. (See
below for massless particles).

Relativistic velocity:
dx
x =
=
d

dt d~x dt
,
d dt d

(1, ~v )
=
.
1 ~v 2

(9)

Note that relativistic velocity is normalised:

x x = 1 .
Relativistic momentum (kinetic momentum):


m~v
m

p = mx = (p , p~) =
,
.
1 ~v 2
1 ~v 2

(10)

(11)

NB: m is the mass (=total energy measured in the rest frame).

For c = 1 note that E = p0 is the total energy, and:
p p = m2 = E 2 + p~2 .

(12)

In the rest frame we have E = m(= mc2 ).

Relativistic version of Newtons second axiom:


d
d
m~v

p~ =
= F~ ,
dt
dt
1 ~v 2

(13)

where F~ is the force acting on the particle.

NB: This is the correct relativistic version of Newtons second axiom, but it is
not (yet) manifestly covariant.

1.2

Action principle for the massive free particle: noncovariant version

Action and Lagrangian:

Z
S[~x] =

dtL(~x(t), ~v (t)) .

Action and Lagrangian for a free massive relativistic particle:

Z p
S = m dt 1 ~v 2 = mass times proper time .

(14)

(15)

NB: we have set ~ = 1, action is dimensionless. The action of a particle is

proportional to its proper time. The factor m is needed to make the action
dimensionless.
Action principle/Variational principle: the equations of motion are found by
imposing that the action is stationary with respect to variations x x + x of
3

the path. Initial and final point of the path are kept fixed: x(ti ) = 0, where
t1 = time at initial position and t2 = time at final position.
Substitute variation of path into action and perform Taylor expansion:
S[~x + ~x, ~v + ~v ] = S[~x, ~v ] + S +

(16)

When performing the variation, we can use the chain rule:

p
p
1 ~v 2 =
1 ~v 2 v i
v i

(17)

(To prove this, you peform a Taylor expansion of the square root).
~v , ~v are not independent quanties:
~v =

d
d~x
= ~x .
dt
dt

(18)

To find S, we need to collect all terms proportional to x. Derivatives acting

on x are removed through integration by parts. Result:
t2

Z t2 

d mvi
mvi

S =
(19)
xi dt +
xi  .
dt 1 ~v 2
1 ~v 2
t1
t1
Now use that the ends of the path are kept fixed (thus boundary terms vanish)
and that the action principle requires S = 0 for all choices of x to obtain the
equation of motion for a free massive particle:
d m~v
d

= p~ = 0 .
dt 1 ~v 2
dt

1.3

(20)

Canonical momenta and Hamiltonian

Canonical momentum:
i :=

For L = m 1 ~v 2 :

L
.
vi

i = pi .

(21)

(22)

Canonical momentum = kinetic momentum (not true in general. Example:

charged particle in a magnetic field).
Hamiltonian:
H(~x, ~ ) = ~ ~v L(~x, ~v (~x, ~ )) .

(23)

Lagrangian Hamiltonian: go from variables (~x, ~v ) (coordinates and velocities)

to new variables (~x, ~ ) (coordinates and canonical momenta) by a Legendre
transform:
L
H(~x, ~ ) =
~v L(~x, ~v (~x, ~ )) .
(24)
~v
4


For L = m 1 ~v 2 :
H = ~ ~v L = p~ ~v L =

m
= p0 = E .
1 ~v 2

(25)

Thus: Hamiltonian = Total energy (not true in general. See below.)

Our treatment of the relativistic particle provokes the following questions:
How can we include massless particles?
How can we make Lorentz covariance manifest?
How can we show that the physics does not depend on how we parametrise
the world line?
We will answer these questions in reverse order.

1.4

Length, proper time and covariance on the world line

Mathematical description of a curve (for concreteness, a curve in Minkowski

space M):
C : R 3 x () M .
(26)
is an arbitrary curve parameter.
Tangent vectors of the curve:

x0 :=

dx
.
d

(27)

Scalar products of tangent vectors, using Minkowski metric:

for spacelike curves,

>0

dx dx
2
= 0 for lightlike like curves (also called null curves),
x0 =
d d
<0
for timelike curves.
(28)
Length of a spacelike curve:
Z 2 r
dx dx
L(1 , 2 ) =

d .
(29)
d d
1
The length is independent of the parametrisation, i.e., it does not change if one
reparametrises the curve:

() , where

6= 0 .
d

(30)

Note that
is an invertible, but otherwise arbitrary function of . Often one
imposes
d

>0,
(31)
d
5

which means that the orientation (direction) of the curve is preserved.

For timelike curves one can define analogously
Z 2 r
dx dx
(1 , 2 ) =
d
d d
1

(32)

and this is nothing but the proper time between the two events located at 1
and 2 .
To find (proper time) as a function of (arbitrary curve parameter), take
the upper limit to be variable:
r
Z
dx dx
0
.
(33)
d 0
() =
d d 0
1
Differentiate:

r
d
dx dx
=
.
d
d d
Compare tangent vector with relativistic velocity:

(34)

dx
dx
dx d
d
=
=q
.
dx
d
d d
dx
d d

(35)

This implies (as required if is really proper time):

dx dx
= 1 .
d d

(36)

Relativistic velocity = normalised tangent vector.

Use proper time instead of as curve parameter:
Z 2 r
Z 2
dx dx
d
d = 2 1 .
=
d d
1
1

(37)

This confirms that is indeed the proper time.

NB: Similarly, the expression for the length of a spacelike curve takes a simple
form when using the length itself as the curve parameter.

1.5

Covariant action for the massive particle

Use new, covariant expression for proper time to rewrite the action:
r
Z
dx dx

0
S[x , x ] = m d
.
d d

(38)

This action does now depend on x (and the corresponding relativistic velocity),
rather than on ~x (and the velocity). It is now covariant in the following sense:
6

 The action is invariant under reparametrisations

() of the world
line. (Reparametrisations are by definition invertible, hence d
d 6= 0. )
The action is manifestly invariant under Poincare transformations,
x x + a ,

(39)

( ) O(1, D 1) and (a ) M

(40)

where
are constant. (This is manifest, because the action is constructed out of
Lorentz vectors).
Compute equations of motion through variation x x + x :
!

d
m x0
S
p
=0
=0.
x
d
x0 x0

(41)

To get the physical interpretation, replace arbitrary curve parameter by proper

time :


d
dx
m
= m
x = 0 .
(42)
d
d
We claim that this is the same as (13) with F~ = ~0.
More generally, the covariant version of (13) is


d
dx
m
= f ,
d
d
where

(~v F~ , F~ )
(f ) =
1 ~v 2

(43)

(44)

is the relativistic force. To see that this is equivalent to (13), note

d~
p
dp0
= ~v
dt
dt

(45)

(The change of energy per unit of time is related to the acting force F~ by
~v F~ . I.p., if the force and velocity are orthogonal, like for charged particle in a
homogenous magnetic field, the energy is conserved.)
Then (13) implies
dp0
= ~v F~ .
dt

(46)

Multiplying by d
dt and using the chain rule we get the manifestly covariant version (43) of (13). The additional equation for p0 is redundant, but is required
for having manifest covariance.

For a free massive particle the equation of motion is

m
x = 0 .

(47)

x ( ) = x (0) + x (0) .

(48)

Solution = straight (world)line:

Actions with interaction

If the force f comes from a potential, f = V (x), then the equation of
motion (43) follows from the action
Z p
Z
2
S = m
x d V (x( ))d .
(49)
For simplicity, we took the curve parameter to be proper time. In the second term, the potential V is evaluated along the worldline of the particle.
If f is the Lorenz force acting on a particle with charge q, f = F x ,
then the action is
Z p
Z
S = m
x 2 d q A dx .
(50)
In the second term, the (relativistic) vector potential A is integrated
along the world line of the particle
Z
Z
dx
A dx = A (x( ))
d .
(51)
d
The resulting equation of motion is


d
dx
m
= qF x ,
d
d

(52)

where F = A A is the field strength tensor. Equation (52) is

the manifestly covariant version of


d~
p
~ + ~v B
~ .
=q E
dt

(53)

The coupling to gravity can be obtained by replacing the Minkowsk metric

by a general (pseudo-)Riemannian metric g (x):
Z
q
S = m d g (x)x x .
(54)
The resulting equation of motion is the geodesic equation
x
+ x x = 0 .
8

(55)

1.6

Canonical momenta and Hamiltonian for the covariant

action

Action and Lagrangian:

Z

x0 2 .

(56)

L
x0
p
= mx .
=
m
x0
x0 2

(57)

Ld = m

S=

Canonical momenta:

Canonical momenta are dependent:

= m2

(58)

Going from a general curve parameter to proper time we see that

= p

(59)

and the constraint is recognised as the mass shell condition p2 = m2 .

Hamiltonian:
H = x L = 0

(60)

Hamiltonian 6= Total energy. Rather Hamiltonian = 0. Typical for covariant

Hamiltonians. Reason: canonical momenta not independent. The Hamiltonia constraint H = 0 defines a subspace of the phase space. Solutions to the
equations of motion live in this subspace. H = 0 does not mean that the dynamics is trivial, it just reflects that the momenta are dependent. To stress this
one sometimes writes H 0 (read: H is weakly zero). One can also write:
H = (p2 + m2 ), where is an arbitrary constant. In this form it is clear that
the Hamiltonian is not identically zero (strongly zero, implying trivial physics),
but that it vanishes whenever the particle satisfies the mass shell condition
p2 = m2 .
We still need to find a way to include massless particles.

1.7

Covariant action for massless and massive particles

Introduce invariant line element ed, where e = e() transforms under reparametrisations such that ed is invariant:

d
e(
) = e() d

ed = ed
.
(61)
d

d
= d d
Action:
1
S[x, e] =
2

Z
ed

1
e2
9

dx
d

2

!
2

(62)

We allow m2 0. (In fact, we could allow m2 to become negative, m2 < 0.)

Symmetries of the action:
S[x, e] is invariant under reparametrisations
.
S[x, e] is invariant under Poincare transformations x x + a .
The action depends on the fields x = (x ) and e. The corresponding equations
of motion are found by variations x x + x and e + e, respectively.
Equations of motion:
d
d

x0
e

x0 + e2 m 2

0,

(63)

0.

(64)

The equation of motion for e is algebraic. e is an auxiliary field, not a dynamical

field. If m2 6= 0, we can solve for e:
2

e2 =

x0
m2

(65)
2

For massive particles we have m2 > 0 and x0 < 0, so that e2 is positive. We

take the positive root:
p
x0 2
e=
.
(66)
m
Substituting the solution for e into (62) we recover the action (38) of a massive
particle. However, the new action allows us to deal with massless particles as
well.
According to (64), e controlls the norm of the tangent vector, which can be
changed by reparametrisation. Instead of eliminating e by its equation of motion, we can bring it to a prescribed gauge (=parametrisation).
For m2 > 0, impose the gauge
e=

1
m

(67)

The equations of motion become:

x

= 1

(68)
(69)

Comment: in this gauge, becomes proper time .

For m2 = 0, impose the gauge
e=1.
10

(70)

The equations of motion become:

x

0,

(71)

x 2

0.

(72)

Comment: for m2 = 0, (64) tells us that the worldline is lightlike, as

expected for a massless particle. In this case there is no proper time, but
one can choose a so-called affine parametrisation. e = 1 (or any other
constant value) corresponds to picking an affine parameter.
Note that in both cases the dynamical equation of motion x
= 0 must be
supplemented by a constraint,
 2

x + 1
=
=0
(73)
x 2
to capture the full information.

The classical relativistic string

2.1

The Nambu-Goto action

Replace particle by one-dimensional string. Worldline becomes a surface, called

the worldsheet .
X : 3 P X(P ) M .
(74)
Coordinates on M are X = (X ), where = 0, 1, . . . , D 1.
Coordinates on are = ( 0 , 1 ) = ( ). The worldsheet has one spacelike
direction (along the string) and one timelike direction (point on the string
moving forward in time). Take 0 to be time-like, 1 to be space-like:
X 2 0 , (X 0 )2 > 0 .

(75)

We use the following notation:

X
X0


X
= (0 X ) =
,
0


X
= (1 X ) =
.
1

(76)

Range of worldsheet coordinates:

Spacelike coordinate:
1 [0, ] .

(77)

0 (, ) ,

(78)

Timelike coordinate

11

for propagation of a non-interacting string for infinite time, or

0
0
].
, (2)
0 [(1)

(79)

for propagation of a non-interacting string starting for a finite intervall of

time).
Nambu-Goto action area of world sheet:
Z
SNG [X] = T A() = T

d2 A

(80)

T has dimension (length)2 or energy/length = string tension (~ = c = 1).

Invariant area element on (induced from M):
q
d2 A = d2 |det(g )| ,
(81)
where
g = X X
is the induced metric (pull back). Note det(g ) < 0.
Action is invariant under reparametrizations of ,
 

0
1

( , ) , where det
6= 0 .

Action is also invariant under Poincare transformations of M.
Action, more explicitly:
Z
Z
q
0 )2 X 2 (X 0 )2 .
SNG = d2 L = T d2 (XX

(82)

(83)

(84)

L is the Lagrangian (density). We compute

L
X

0 )X 0
(X 0 )2 X (XX
= T q
,
0 )2 X 2 (X 0 )2
(XX

L
X 0

0 )X
X 2 X 0 (XX
= Tq
.
0 )2 X 2 (X 0 )2
(XX

= P0 =
P1 =

(85)

is the canonical momentum, P are the momentum densities on .

The equations of motion are found by variations X X + X of the
worldsheet, where the initial and final configuration are kept fixed, X( 0 =
0
0
(1)
, (2)
) = 0. Since the ends of the string are not fixed for other values of 0 ,
variation yields a boundary term, which we need to require to vanish:
Z
d

L
X
X0

1 =

=0.
1 =0

This boundary term must vanish admissible boundary conditions.

12

(86)

1. Periodic boundary conditions:

X( 1 ) = X( 1 + )

(87)

Closed strings, the world sheet does not have (time-like) boundaries.
2. Neumann boundary conditions:

L 
=0
X0 1 =0,

(88)

(also called: free boundary conditions, natural boundary conditions).

Open strings, ends can move freely. (The ends of an open string always
move with the speed of line and thus their worldlines are light-like.)
The momentum at the end of the string is conserved. (P1 is the momentum density along the space-like direction of , i.e., along the string at a
given time.)
3. Dirichlet boundary conditions:
X i ( 1 = 0) = x0 ,

X i ( 1 = ) = x1 .

(89)

Open strings with ends kept fixed along the i-direction (spacelike).
(Dirichlet boundary conditions in the time direction make only sense in
imaginary time, in the Euclidean version of the theory, where they describe instantons).
Consider Neumann boundary conditions along time and p spacelike conditions and Dirichlet boundary conditions along D p directions. Then
the ends of the string are fixed on p-dimensional spacelike surfaces, called
Dirichlet p-branes. Momentum is not conserved at the ends of the string
in the Dirichlet directions (obvious, since translation invariance is broken)
p-branes are dynamical objects. Interpretation: strings in a solitonic
background (6= vacuum).
Equations of motion (with either choice of boundary condition):
0

L
L
+ 1
=0
0

X
X

(90)

or
P = 0

(91)

Canonical momenta are not independent. Two constraints:

X0

0 2

+ T (X )

(92)

Canonical Hamiltonian (density):

L=0.
Hcan = X
13

(93)

2.2
2.2.1

The Polyakov action

Action, symmetries, equations of motion

Intrinsic metric h () on the world-sheet , with signature ()(+).

Polyakov action:
Z

T
d2 hh X X ,
SP [X, h] =
2

(94)

where h = det(h ) = | det(h )|.

Local symmetries with respect to :
1. Reparametrizations
(), which act by
(
X
)

= X () ,

(
h
) =
h () .

(95)

2. Weyl transformations:
h () e2() h () .

(96)

Remarks:
1. A Weyl transformation is not a diffeomorphism, but the multiplication
of the metric by a positive function. Mathematicians usually call this a
conformal transformation, because it changes the metric but preserves the
conformal structure of (, h ).
2. The invariance under Weyl transformation is special for strings, it does
not occure for particles, membranes and higher-dimensional p-branes.
3. Combining Weyl with reparametrization invariance, one has three local
transformations which can be used to gauge-fix the metric h completely.
Thus h does not introduce new local degrees of freedom: it is an auxiliary (dummy) field.
Global symmetries with respect to M: Poincare transformations.

Equations of motion from variations X X + X and h h + h .



1

hh X
= 0,
h
1
X X h h X X = 0 .
2
Boundary conditions: as before.

14

(97)
(98)

The X-equation (97) is the covariant two-dimensional wave equation, alternatively written as
X = 0 X = 0 X = 0

(99)

The h-equation (98) is algebraic and can be used to eliminate h in terms of

the induced metric g = X X . It implies
det(g ) =

1
det(h )(h g )2 .
4

(100)

Substituting this into the Polyakov action one gets back the Nambu-Goto action.
More generally one can show that SP SNG , where equality holds if g and
h are related by a Weyl transformation (i.e., they are conformally equivalent).
2.2.2

Relation to two-dimension field theory

Alternative point of view: is a two-dimensional space-time, populated by D

scalar fields X = (X ), which take values in MD . The Polyakov action has the
form of a standard for a two-dimension scalar field theory. More precisely, it
defines a two-dimensional sigma-model with target space M.
More generally: Mechanics of p-brane (p-dimensional extended object) =
Field theory on a p + 1 dimensional space-time (the worldvolume of the brane).
The energy-momentum tensor of the two-dimensional field theory is:
T :=

1
1
1 1 SP

= X X h h X X

T h h
2
4

(101)

(Defining T by the Noether procedure gives a tensor which might differ from
this by a total derivative.)
The h-equation of motion in terms of T :
T = 0 .

(102)

(Remember that the h-equation is a constraint.)

This equation can also be interpreted as the two-dimensional Einstein equation, since the variation of two-dimensional Einstein-Hilbert action vanishes
identically. The Polyakov action describes a two-dimensional sigma model which
is coupled to two-dimensional gravity.
T is conserved:
T = 0
15

(103)

To show this, one needs to use the equations of motion. The equation holds
only on shell.
T is traceless:
h T = 0

(104)

This follows directly from the definition of T . It holds off shell, independently
of whether the equations of motion are satisfied.
2.2.3

Polyakov action in conformal gauge

Three local symmetries, while the intrinsic metric h has three independent
components can bring h to standard form (locally):


1 0
h =
.
(105)
0 1
!

The condition h = is called the conformal gauge. Strictly speaking,

the gauge condition should be imposed on the equations of motion, not on the
action. For the case at hand the result is correct nevertheless. Therefore, let us
follow the naive procedure and start from the Polyakov action in the conformal
gauge.
The action:
Z
T
d2 X X .
(106)
SP =
2
Equations of motion for X (variation of gauge-fixed action gives correct
gauge fixed equations of motion):
X = (02 12 )X = 0

(107)

This is the two-dimensional wave equation, which is known to have the general
solution:

X () = XL ( 0 + 1 ) + XR
( 0 1 ) .
(108)
Interpretation: decoupled left- and rightmoving waves.
Boundary conditions:
X ( 1 + ) = X ()

X0 1 =0, = 0


X 
=0
1 =0,

periodic
Neumann
Dirichlet

(109)

The equations coming from the h variation must now be added by hand:
T = 0

(110)

Trace(T ) = T = T = T00 + T11 .

(111)

Note that the trace of T is

16

The trace of T vanishes identically (off shell), and therefore T = 0 gives

only rise to two independent non-trivial constraints:
T01 = T10

T00 = T11

1 0
XX = 0
2
1 2
(X + X 02 ) = 0
4

(112)

These equations are equivalent to the constraints derived from the Nambu-Goto
action.
We have two-dimensional energy-momentum conservation (from the X equation):
T = 0
(113)
and tracelessness (without equations of motion):
h T = 0 .

(114)

Note that energy-momentum conservation is non-trivial because T only vanishes on-shell.

2.2.4

Lightcone coordinates

Equation (108) suggests to introduce lightcone coordinates (null coordinates):

:= 0 1 .

(115)

We write a , where a = +, for lightcone coordinates and , where = 0, 1

for standard coordinates.
The Jacobian of the coordinate transformation and its inverse:


D( + , )
1 1
a
=
(J ) =
1 1
D( 0 , 1 )


0
1
D( , )
1
1 1

(Ja ) =
=
(116)
1 1
D( + , )
2
Converting lower indices:
va = Ja v ,

v = J a va .

(117)

Example: the lightcone derivatives

=

1
(0 1 )
2

(118)

Converting upper indices:

w a = w J a ,

w = wa Ja .

(119)

Example: the lightcone differentials

d = d 0 d 2 .
17

(120)

When converting tensors, act on each index. Example: lightcone metric

hab = Ja Ja h .

(121)

Explicitly: metric in standard coordinates



1 0
(h ) =
= (h ) .
0 1
Metric in lightcone coordinates



1
0 1
0
ab
(hab ) =
, (h ) = 2
1 0
1
2

(122)

1
0


.

(123)

The lightcone coordinates of the (symmetric traceless) energy-momentum tensor

are:
T++ =

1
(T00 + T01 ) ,
2

T =

1
(T00 T01 ) ,
2

T+ = 0 = T+ .

(124)

Note that the trace, evaluated in lightcone coordinates, is:

Trace(T ) = ab Tab = 2 + T+ = 4T+ .
Thus traceless T+ = 0.
Action in lightcone coordinates:
Z
T
SP =
d2 X X
2
Z
Z
T
=
d2 = (X 2 X 02 ) = 2T d2 + X X .
2

(125)

(126)

Equations of motion in lightcone coordinates:

X
+ X

= (02 12 )X = 4+ X = 0
=

0.

(127)

Then, it is obvious that the general solution takes the form:

X () = XL ( 0 + 1 ) + XR
( 0 1 ) .

(128)

The constraints in lightcone coordinates

T++ = 0 + X + X = 0 X L2 = 0 ,
2
T = 0 X X = 0 X R
=0.

(129)

We did not list T+ = 0 as a constraint, because it holds off shell. There are
two non-trivial constraints, which are equivalent to the two constraints obtained
for the Nambu-Goto action.
18

Energy-momentum conservation in lightcone coordintes:

T++ = 0

T++ = T++ ( + )

+ T = 0

T = T ( )

(130)

These equations are non-trivial, because T only vanish on shell.

Using this, we see that every function f ( + ) defines a conserved chiral current
(f ( + )T++ ) = 0
(131)
and thus a conserved charge
Z
Lf = T

d 1 f ( + )T++ .

(132)

To show directly that Lf is conserved, dd 0 Lf = 0, you need to use the periodicity

properties of the fields in addition to current conservation. Similarly we obtain
conserved quantities from T using a function f ( ). For closed strings we
get two different sets of conserved quantities. For open strings T++ and T
are not independent, because they are related through the boundary conditons.
Therefore one only gets one set of conserved charges.
The canonical momenta:
=

LP
= T X

The canonical Hamiltonian

Z

 T Z


1

Hcan =
d X LP =
d 1 X 2 + X 02
2 0
0
Z


= T
d 1 (+ X)2 + ( X)2

(133)

(134)

is the integrated version of the constraint T00 = T11 = 0.

NB: The conformal gauge does not provide a complete gauge fixing. The
action is still invariant under residual gauge transformations. Namely, we can
combine conformal reparametrisations with a compensating Weyl transformation (see below).
2.2.5

Momentum and angular momentum of the string

Polyakov action is invariant under global Poincare transformations of M :

X X + a

(135)

Momentum is the quantity which is preserved when there is translation invariance (Noether theorem). Noether trick: promote the symmetry under consideration (here translations in M) to a local transformation: X = a (). The
19

action is no longer invariant, but it becomes invariant when a is constant.

Therefore the variation of the action with respect to the local transformation
must take the form
Z
S = d2 a P .
(136)
Integration by parts gives
Z
S =

d2 a P .

(137)

If we impose that the equations of motion are satisfied, this must vanish for any
a . Hence the current P must be conserved on shell:
P = 0 .

(138)

Interpretation P ( fixed) is the conserved current on associated with translations in the -direction of M, in other words, the momentum density along
the -direction. (From the point of view of the two-dimensional field theory
living on , translations in M are internal symmetries.)
While above we assumed the conformal gauge, the method works without
gauge fixing.
Explicitly:

c.g.
(139)
P = T hh X = T X ,
where we only went to the conformal gauge in the last step.
To find the angular momentum density, do the same for Lorentz transformations in M. Result:

J
= X P X P
(140)
While current conservation holds by construction, it can easily be checked

= 0.
explicitly, using the equations of motion: P = 0, J
Conserved charges are obtained by integration of the timelike component of
the current along the spacelike direction of :
Z
Z
P =
d 1 P 0 = T
d 1 X

Z0
J

0
J

Z0
=T



d 1 X X X X

(141)

Again, these charges are conserved by construction, but one can also check
explicitly 0 P = 0, 0 J = 0.
Summary: P , J are the total momentum and angular momentum of the

string, P , J
the corresponding conserved densities on the world sheet.

20

2.2.6

Fourier expansion

For periodic boundary condition we write the general solution as a Fourier series:
r
r
1 0 i
1 X 1 2in
1 X 1 2in+
i

X () = x +
p +
n e
+

e
T
2 T
n
2 T
n n
n6=0

n6=0

(142)

where x , p are real and n ? = n

,
n ? =
n
.
NB: no linear term in 1 , because of boundary condition. Splitting of zero
mode between left and rightmoving part arbitrary.
p is the total momentum:

d 1 X = p .

P =T

(143)

The center of mass of the string moves along a straight line, like a relativistic
particle:
Z
dx
1 1
d X () = x + p 0 = x (0) +
(0)
(144)
xcms =
0
d
Thus: string = relativistic particle plus left- and rightmoving harmonic oscillations.
Formulae suggest to use string units:
T =

1
.

(145)

(On top of c = ~ = 1).

Fourier components of
1
( X)2
2

T =

(146)

at 0 = 0:

Z
Lm

d 1 e2im T =

:= T
0

m
L

d 1 e2im T++ =

:= T
0

where
0 =
0 =

T X
mn n ,
2 n=

T X

mn
n ,
2 n=

1
T =1 1
p.
p =
2
4T

(147)

(148)

The constraints T = 0 imply

m = 0 .
Lm = L

21

(149)

The Lm are conserved, take f e2im in (132). If the constraint hold at 0 = 0

they hold for all times 0 . The constraints are conserved under time evolution.
Canonical Hamiltonian:
Z
Z


L = T
0
(X 2 + (X 0 )2 )2 = L0 + L
(150)
d 1 X
H=
2 0
0
0 generates translations of 0 , see below.)
(The world sheet Hamiltonian L0 + L
The Hamiltonian constraint H = 0 allows us to express the mass of a state
in terms of its Fourier modes:
H

0 =
= L0 + L
=

T X
(n n +
n
n)
2 n=

p2
) = 0
+ T (N + N
4

(151)

where we defined the total occupation numbers

N=

=
n n , N

n=1

n
n .

n=1

This implies the mass shell condition:

) .
M 2 = p2 = 4T (N + N
0 we have the additional constraint
Since L0 = L
,
N =N
which is called level matching, because it implies that left- and rightmoving
modes contribute equally to the mass.
Open string with Neumann boundary conditions:
X () = x + p 0 +

i X 1 in0
n e
cos(n 1 ) .
T n6=0 n

(152)

This can still be decomposed into left and rightmvoing parts X = XL ( + ) +

XR ( ):

XL/R
( ) =

1
1
i X1
x +
pL/R +
n (L/R) ein ,
2
T
n
2 T n6=0

(153)

but the boundary conditions X 0 ( 1 = 0, ) = 0 imply

pL = pR =

1
p,
2

n(L) = n(R) .

(154)

We see that due to the boundary conditions left and rightmoving waves are
reflected at the boundaries and combine into standing waves. There are only
half as many independent oscillations as for closed strings.
22

Combining the boundary conditions with the constraints we see that (X )2 ( 1 =

0, ) = 0 The ends of the string move with the speed of light.
Since XL and XR are related by 1 1 (world sheet parity), we can
combine them into one single periodic field with doubled period, 1 [, ].
Fourier decomposition gives one set of modes,
Z
2


 T Z
1
1
im 1
im 1
eim X + X 0
Lm = 2T
d e
T++ + e
T =
4
0
X
1
=
T
mn n
(155)
2
n
where we defined 0 = p. The canonical Hamiltonian is H = L0 .
The Lm define the constraints: Lm = 0 and these constraints are conserved
in time. (Note that we only get half as many constraints as for the bosonic
string, due to the boundary conditions.) The Hamiltonian constraint H = L0 =
0 is the mass shell condition:
M 2 = p2 = 2T N .
Dirichlet boundary conditions: linear term in 1 , but no linear time in 0 .
For the osciallators, cos is replaced by sin. Thus X = 0 at the ends (ends dont
move), but X 0 6= 0 at the ends, corresponding to exchange of momentum with
a D-brane.

3
3.1

The quantized relativistic string

Covariant quantisation and Fock space

We set T = 0 in the following (on top of ~ = 1 = c).

Remember quantization of non-relativistic particle:
[xi , pj ] = i ij ,
where i = 1, 2, 3. Here i = pi is the canonical (=kinetic) momentum.
Relativistic particle:
[x , p ] = i .
The canonical (=kinetic) momentum is now constrained by the mass shell condition p2 + m2 = 0.
Free scalar field:
(3)

[(x), (y)]
(~x ~y )
x0 =y 0 = i

Note that the commutator is evaluated at equal times x0 = y 0 . The canonical

momentum of a free scalar field (x) is (y) = (0L

(x)) = 0 (x) = (x).
Free relativistic string:
[X ( 0 , 1 ), ( 00 , 01 )]0 =0 0 = i ( 1 01 ) .
23

Here () = ( + ) is the periodic -function, which has the Fourier series

() =

1 X 2ik/
1 X 2ik
e
e
=
.

k=

k=

We are working in the conformal gauge where X = 0, resulting in the Fourier

expansions (142), (152), depending on our choice of boundary conditions. In
the conformal gauge, the canonical momentum is
=

1
L
= T X = X .
(0 X )

Using the Fourier expansion, we can derive the commutation relations for the
modes

[x , p ] = i , [m
, n ] = mm+n,0 , [
m
,
n ] = mm+n,0 . (156)

Hermiticity of X implies:

+
(x )+ = x , (p )+ = p , (m
) = m
.

We get the commutation relations of a relativistic particle plus those of an

infinite number of harmonic oscillators, (with frequencies m = m > 0, see later)
corresponding to the excitations of the string. To obtain standard creation and
annihilation operators, set

am =
m

(am )+ =
m

for m > 0 ,
for m < 0 ,

(157)

to obtain [am , (an )+ ] = m,n , which generalises [a, a+ ] = 1.

As a first step to constructing the Hilbert space of states we construct a
representation space of the operators x, p, , which we call the Fock space F.
This space is different from the physical Hilbert space H, because we still have
m 0).
to implement the constraints Lm 0 (and, for closed strings L
The ground state |0i of F is defined by two conditions: it is translation
invariant (=does not carry momentum) and it is annihilated by all annihilatin

operators m
, m > 0:
p |0i = 0 ,

m
|0i = 0 for m > 0 .

Momentum eigenstates are defined by the property

p |ki = k |ki ,
where p is the momentum operator and k = (k ) its eigenvalue. Momentum
eigenstates can be constructed out of the ground state using the operator x :
|ki = eikx |0i ,
24

where kx = k x . Exciteted string states are generated by applying creation

operators. I.p.

m
|0i

is a state with excitation number Nm

= 1 in the m-th mode along the direction. A general excited state is a linear combination of states of the form
1
2 |0i ,
m
1 m2

which have definit excitation numbers {Nm

}. A basis for F is obtained by
taking linear combinations of states of the form
1
2 |ki ,
m
1 m2

| = 0, . . . , D
which carry momentum k = (k ) and excitation numbers {Nm
1, m > 0}.
The Fock space F carries a natural scalar product. Let us first define it in
the oscillator sector. Decompose the ground state into an oscillator groundstate
|0iosc and a momentum groundstate |0imom :

|0i = |0iosc |0imom .

The scalar product in the oscillator sector is fixed by the properties of the

creation and annihilation operators m

, once we have normalised the ground
state:
osc h0|0iosc = 1 .
For illustration, we compute:

m
|0iosc , n
|0i = osc h0|m
n |0iosc = osc h0|[m
, n
]|0iosc = m m+n,0 .
This scalar product is manifestly Lorentz covariant, but indefinit, i.e. there are
states of negative norm. This shows that F is not a Hilbert space. This is not
a problem, as long as the physical space of states is positiv definit.
The scalar product between momentum eigenstates is
hk|k 0 i = D (k k 0 ) .
This is not defined if k = k 0 . Momentum eigenstates |ki, including the momentum ground state 0imom , are not normalisable. Normalisable states are obtained
by forming momentum space wave packets
Z
|i = dD k(k)|ki
where (k) (the momentum space wave function) is square integrable:
Z
h|i = dD k(k)(k) < .

25

3.2

Implementation of constraints and normal ordering

Having constructed the Fock space F, we still need to impose the constraints
m . Naively we expect that the resulting space Fphys of physical
Lm 0 L
states is positiv definit. As we will see in due course, the situation is a bit more
complicated. For now we need to decide how to implement the constraints.
m vanish
We decide to require that the matrix elements of the operators Lm , L
between physical states vanish. Naively, this amounts to saying: a state |i is
a physical state, |i Fphys , if
h|Lm |i = 0 ,
and, for closed strings, in addition
m |i = 0 ,
h|L
for all m. But as it stands this is ambiguous, because the Lm are quadratic in
the s, which do not commute any more in the quantum theory. Thus we have
an ordering ambiguity in the quantum theory, which we need to investigate.
One standard ordering prescription is normal ordering, which requires to put
all annihilation operators to the right of the creation operators:

if m < 0, n > 0
m n

if m > 0, n < 0
n m
: m
n :=

any ordering if else.

From the commutation relations (156) we see that normal ordering only has a
non-trivial effect if m = n and = , because creation and annihilation operators corresponding to different excitations levels or different directions commmute.
Compute the normally ordered version of L0 :
LNO
0

1 X
: n n :
2 n=

1 X
1
1X
: n n : + 0 0 +
: n n :
2 n=
2
2 1

1 2
p +N ,
8

where we used that

operator

1
2 0

(158)

0 = 18 p2 for closed strings and defined the number

N=

n n .

n=1

Now compare this to what we might call the classically ordered version of L0 :
LCO
0

1 X
n n
2 n=

26

1 X
1
1X
n n + 0 0 +
n n
2 n=
2
2 1

1 2 1X
p +
(n n + n n )
8
2 1

(159)

Both version of L0 differ by an infinite constant:

LCO
0

LNO
0

DX
=
n.
2 n=1

Going from the classically ordered to the normally ordered L0 amounts to subtracting an infinite constant. We will take the quantum version of L0 to be the
normally ordered one, L0 = LNO
0 . This is prescription usually adopted in quantum field theory. But is this the correct choice? The problem is not that the
constant which we subtract is infinite (we have the freedom to define any undefined object as we like), but that there is a finite ambiguity in this procedure.
O
In other words: why dont we take LN
+ a, where a is some finite constant?
0
(Equivalently, why cant we adopt a slightly different ordering prescription for
finitely many term, which shifts L0 by a finite constant?)
There are two ways to procede from here. The first is to accept that the
ordering of L0 is ambiguous, and to formulate the constraint in the form
0 a
h|L0 a|i = 0 , h|L
|i = 0 ,
0 are normally ordered and a, a
where L0 , L
are finite constants which parametrize
the ambiguity. It then turns out that in order to have a postive definit space
of states, one must choose a = a
= 1, i.e., the ambiguity is completely fixed by
physical requirements.
The other way, which gives the same result, is to find a physcial procedure for
computing a. The basic insight is that the infinite constant which is subtracted
corresponds to the ground state energies of infinitely many harmonic oscillators.
To show this, remember that L0 = 18 p2 + N (closed string). The commutation
relation of N with a creation operator is

[N, m
] = mm
.

Thus

N m
|0i = mm
|0i

and
1
1
N m
2 |0i = (m1 + m2 + )m
2 |0i
1 m2
1 m2

where the eigenvalue m1 + m2 + of N is the total excitation number. Thus

N counts excitations (weighted with their level), hence the name. Now rewrite
N in terms of canonically normalised creation and annihilation operators (157):
N=

n n =

n=1

X
n=1

27

na+
n an

Comparing this to the Hamilton operator of a harmonic oscillator of frequency

n = n > 0 (~ = 1),
1
Hn = n(a+
n an + )
2
P
we see that N differens from n=1 Hn by the (divergent) sum over the ground
state energies n2 . We will discuss later how this sum can be regularized and
computed.1
For the time being we parametrise the normal ordering ambiguity by a, a
. It
is straightforward to show that the Lm , m 6= 0 do not suffer from a normal ordering ambiguity. The resulting definition of physical states is: |i is a physical
state, |i Fphys , if
m |i = 0 for m 6= 0 , and
h|L
m a
h|L
|i = 0 .

h|Lm |i = 0 ,
h|L0 a|i = 0 ,

3.3

(160)

Analysis of the physical states

We will now start to analyse Fphys . The hermiticity properties L+

m = Lm can
be used to write the constraints in the form
Lm |i = 0 ,
(L0 a)|i = 0 ,

m |i = 0 for m > 0 ,
L
m a
(L
)|i = 0 .

(161)

We start with the L0 -constraint.

(L0 a)|i = 0
1
( p2 + N )|i = a|i
8
1 2
k +N =a,
8

(162)

where k is the momentum and N denotes, in the last line, the eigenvalue of the
operator N . In the future, we will use N to denote both the operator and its
eigenvalue. The meaning should be clear from the context.
Since k 2 = M 2 , where M is the mass, we see that this constraint expresses
the mass in terms of the total leftmoving excitation number:
1 2
M =N a.
8
In the rightmoving sector we find
1 2
a
M =N
.
8
1 In QFT non-trivial effects of the ground state energy can usually be ignored, because one
(i) does not couple to gravity and (ii) works in infinite volume. In finite volume, the vacuum
energy depends explicitly on the volume. This is a measurable effect, the Casimir effect. Since
strings have finite size, the Casimir effect cannot be ignored, and its computation leads to
a = 1. See later.

28

Both conditions can be recombined into the mass shell condition

1 2
aa
M =N +N

4
which expresses the mass in terms of left- and rightmoving excitation numbers,
and the level matching condition
a
N a=N
,
which states the left- and rightmoving degrees of freedom contribute symmetrically to the mass.
This formula is given in string units, T = 1. The string tension can easily
be re-installed by dimensional analysis. Traditionally, the mass formula is then
expressed in terms of the Regge slope
0 =

1
2T

rather than in terms of the string tension T :

aa
0 M 2 = 2(N + N
) .

The Regge slope has dimension Length2 .

0 can be regarded as the fundamental string length. String units correspond to setting 0 = 21 .
For open strings we have L0 = 12 p2 +N , and the resulting mass shell condition
is
0 M 2 = N a .
Let us now have a look at the lowest mass states. When listing these we
anticipate that a = a
= 1. Start with the open string:
N
0
1
2

0 M 2
1
0
1

State
|ki

1
|ki

2
|ki

1
1
|ki

Spin
Scalar
Vector
Vector
Symmetric Tensor

We still have to impose the further constraints Lm |i = 0 for m > 0. For

|i = |ki it is easy to see that the other conditions are satisfied automatically.
This state is physical, and having minimal mass-squared it is the physical ground
state of the relativistic string.
3.3.1

The scalar ground state (tachyon) of the open string

Thus the physical ground state of string is a scalar with negative mass-squared,
a tachyon. Note that this state is a tachyon for any positive a. Taken as classical
particles, tachyons propagate faster than the speed of light and are believed to be
unphysical. The interpretation quantum field theory is more subtle. The masssquared of a particle is given by the second derivative of the scalar potential
29

at its minimum. If the second derivative of the potential is negative rather

than positive, this means that one is looking at a local maximum and not at a
minimum of the potential. This means that the true vacuum is elsewhere, at a
finite value of the expectation value of the tachyon (tachyon condensation). In
the standard model, the Higgs field has negative mass squared when expanding
the around the local maximum of the Higgs potential. However, the physical
Higgs particle is found by expanding the potential around the minimum, and
it has a positive mass-squared, given by the curvature of the potential at its
minimum. Something similar happens for the open string. The true vacuum is
believed to be the ground state of the closed string, i.e., the true vacuum does
not have open string excitations.
While this is interesting, for us the relativistic string, open or closed, is only
a toy model anyhow, because, as we will see, it does not contain fermions in its
spectrum. The physically interesting string theories are supersymmetric string
theories, where the ground state is massless rather than tachyonic. Therefore
we will regard the tachyon ground state as an unphysical feature of a toy model.
3.3.2

The massless vector state (photon) of the open string

The first excited state is a massless vector, hence potentially a photon. Note
that the first excited state is massless if and only if a = 1. For generic values of
a there would be no massless states at all. This indicates that the case a = 1 is
special (though it does not prove that this is the correct value).
In this case the L1 -constraint is non-trivial. Let us evaluate the constraint

for a general linear combination |i = 1

|ki of level-one states:2
L1 |i = 0

L1 1
|ki = 0

0 [1, , 1
]|ki = 0

k |0i = 0 .

(163)

Thus physical level-one states satisfy

k2 = 0 ,

k = 0 .

This tells us that the momentum is lightlike, corresponding to a massless particle, and orthogonal (in Minkowski metric) to the vector , which therefore
should be interpreted as the polarisation vector.

Since one can show that Lm 1

|ki = 0 holds automatically for m > 1, all
these states are physical. Can we interprete these states as the components of
a photon? In D dimensions, a photon has D 2 independent components due
to gauge invariance. The condition k = 0 reduces the number of independent
polarisations from D to D 1. We have to show that there are actually only
D 2 independent components, and we would like to see the underlying gauge
invariance.
2 Here

and in the following we write |ki instead of |0iosc |ki for simplicity.

30

To this end, consider states of the form

|ki
|i = k 1

where is a (real) constant.

The scalar product of such a state with itself is zero:

h|i = 2 k k hk|[1 , 1
]|ki = 2 k 2 hk|ki = 0

because k 2 = 0.3 The state |i = k 1

|ki is different from zero, |i =
6 0, yet
orthogonal to itself h|i = 0: it is a so-called null state. The existence of null
states shows that Fphys cannot be postive definit and therefore that it is not
the physical Hilbert state. (More in due course).
One can also show that |i is orthogonal to any state with excitation level
different from 1. Moreover, it is orthogonal to any physical state with excitation
level 1:

hk 0 |1 k 1
|ki = khk 0 |ki = 0 ,

because either k 6= k 0 implying hk|k 0 i = 0 or k = k 0 , implying k = 0.

This shows that |i is a spurious state, because it drops out of any scalar
product between physical states. In other words

( + k )1
|ki

represents the same physical state for any value of . The freedom of changing
the value of is a gauge degree of freedom. The component of parallel to the
momentum does not contribute to any physical quantity, and thus the number
of independent physical polarisations is reduced to D 2, as expected for a
D-dimensional photon.
3.3.3

Review of the Maxwell Lagrangian

For illustration, we review how the conditions k 2 = 0 and k = 0 can be derived

from electrodynamics. We start from the D-dimensional Maxwell action


Z
1
S[A] = dD x F F ,
4
where
F = A A .
By variation of A we obtain one half of the Maxwell equations
F = 0 .
In the presence of charge matter, the action becomes


Z
1
S[A, j] = dD x F F j A ,
4
3 Since hk|ki is not well defined, we should actually consider wave packets. But this is only
a technical complication, which does not change the result.

31

where j is the electromagnetic current. The resulting Maxwell equations are

F = j .
The second half of the Maxwell equations do not follow from the variation of the
action. They are identities (called Bianchi identities), which are the integrability
conditions guaranteeing that the field strength has a vector potential:
[ F] = 0 F = A A (locally) .
The field strength and the action are invariant under gauge transformations
A A + F F ,
where is an arbitrary function. Note that the coupling to a current j is only
consistent if the current is conserved, j = 0.
Let us restrict to free Maxwell theory, j = 0, and compare the properties
of the photon field A to those of the massless vector state of the open string.
Writing out the equation of motion:
F = A A = 0 .
We can partially fix the gauge symmetry by imposing the Lorenz gauge and
obtain
A = 0 , if A = 0 .
There is still a residual gauge symmetry because we can make gauge transformations where the function satisfies  = 0, because
A (A + ) = A +  .
This implies that, in the Lorenz gauge, A and A + with  = 0 represent
the same physical state. The number of independent physical components of
A is D 2.
Consider now a plane wave with momentum k and polarisation vector :
A = eikx .
The general solution can be written a superposition (Fourier integral) of such
plane waves. This allows us to translate the equations of motion, the Lorenz
gauge, and the residual gauge symmetry into momentum space:
A = 0 k k = 0
A = 0 k = 0
A ' A + ' + k

(164)

We can be more explicit about the unphysical, physical and spurious parts of
A , by making an explicit choice for the momentum vector:
k = (k 0 , 0, . . . , 0, k 0 )
32

This corresponds to a massless particle (equivalently, a plane wave) propagation

parallel to the (D 1)-axis (for D = 4, the 3-axis or z-axis). The polarisation
vector must satisfy:
k = 0 k 0 0 + k 0 D 1 = 0 0 = D1 .

(165)

Thus physical polarisation vectors have the form

= ( 0 , 1 , . . . , D2 , D1 ) .
Introduce the following basis for polarisation vectors:
k
k
ei

(k 0 , 0, . . . , 0, k 0 ) ,
1
=
(k 0 , 0, . . . , 0, k 0 ) ,
2(k 0 )2
= (0, . . . , 1, . . . , 0) , i = 1, . . . , D 2

(166)

This is an orthonormal basis which consists of two linearly independent lightlike

vectors spanning the light cone and D 2 spacelike vectors which spann the
directions transverse to the light cone:
kk =0, kk =1, kk =0,
ei ej = ij , k ei = 0 , k ei = 0 .

(167)

In this basis physical polarisation vectors have the form:

= i ei + k ,
i.e. the unphysical direction in polarisation space is the one parallel to k (which
is the only basis vector with a nonvanishing scalar product with k).
However, the part of which is parallel to k is spurious: it has zero norm and
is orthogonal to any physical polarisation vector. The D 2 physical degrees of
freedom reside in the transverse part
transv = i ei = (0, 1 , 2 , . . . , D2 , 0)
We call this the transverse part, because the spatial part
~ = ( 1 , 2 , . . . , D2 , 0)
is orthogonal to the spatial momentum
~k = (0, . . . , 0, k 0 ) .
In D = 4, this is the well-known fact that a photon has two physical polarisations, which are transverse to the momentum.
Group theoretical interpretation. In (165) we have made an explicit choice
for the momentum vector, which corresponds to the choice of a Lorentz frame.
33

This choice is not unique, we can still make Lorentz transformations which are
spatial rotations around the axis spanned by k. The corresponding subgroup
SO(D 2) SO(1, D 1)
is known as the little group. A central result in the representation theory of the
Poincare group states that massless representations (those where p p = m2 =
0) are classified by representations of the little group. The physical components

of 1
|0i and of A = eikx transform in the vector representation [D 2]
of the little group SO(D 2). This is what characterises them mathematically
as a massless vector, or photon.
The case D = 4 is special, because the group SO(2) is abelian, which implies
that its irreducible representations are one-dimensional. In D = 4 the transverse
polarisation vector transv = ( 1 , 2 ) transforms as follows:
 1 

 1 

cos sin

.
2
sin cos
2
To see the decomposition into two one-dimensional representations we take complex linear combinations
= 1 i 2 .
This corresponds to going from a basis of transverse polarisations to a basis of
circular polarisations. Now
 + 
 i
 + 

e
0

0 ei

These states have helicity h = 1. Mathematically, we have used the group

isomorphism SO(2) ' U (1).
In conclusion, the massless vector state of the open string has precisely
the kinetic properties which characterise a photon (=a massless vector boson
with a gauge invariance). Therefore the Maxwell action provides a space-time
description of the massless vector state of the free string: it is the effective action
of this state.
3.3.4

States of the closed string

Physical closed strings satisfy the mass shell and level matching conditions
0 M 2
N

aa
2(N + N
)

= N

(168)

plus
m |i for m > 0 .
Lm |i = 0 = L

34

(169)

Evaluate the mass shell and level matching conditions, anticipating a = a

= 1.
The resulting states are

N =N
0
1
2

0 M 2
4
0
4

State
|ki

n
|ki
1

1
1

1
|ki

1
1

2
|ki

1
|ki
1
2

2
2 |ki

Spin
Scalar
2nd rank tensor
4th rank tensor
3rd rank tensor
3rd rank tensor
2nd rank tensor

The ground state is again a tachyonic scalar, which we disregard as the artifact
of a toy model. The first excited state is massless (for a = a
= 1) and a 2nd
rank tensor. For this state the constraints (169) are non-trivial for m = 1 (but
trivial for m > 1). For a general linear combination of level-two states,

1
|ki

evaluation of the constraints gives

k = 0 = k ,
where k 2 = 0 because of the mass shell condition.
is a 2nd rank tensor under the Lorentz group SO(1, D1). This representation is reducible, it can be decomposed into its symmetric and antisymmetric
part:

s = ()
b = []

= s + b where
1
=
( + )
2
1
=
( )
2

(170)

If we decompose the string state accordingly, the polarisation tensors must satisfy the following constraints:
k s = 0 ,

k b = 0 .

Let us analyse the antisymmetric and symmetric part separately.

3.3.5

The closed string B-field and axion

The antisymmetric part

b 1

1
|ki ,

b = b

corresponds to an irreducible Lorentz representation. Momentum and polarisation satisfy:

k 2 = 0 , k b = 0 .
35

One can show that physical states with

b = k a k a , where k a = 0 ,
are spurious, i.e. they have zero norm and are orthogonal to all physical states.
Thus we have identified a gauge symmetry
b b + k a k a , where k a = 0 .
This state is a massless, 2nd rank antisymmetric tensor and has a gauge symmetry similar to the one of a photon. For a free field, the action is


Z
1
H H ,
S[B] = dD x
2 3!
where
H

=
=

3
( B B + B B + B B )
3!
B + B + B
(171)

is the field strength (completely antisymmetric 3rd rank tensor) and B (antisymmetric 2nd rank tensor) is the gauge field (gauge potential). The equation
of motion is
H = 0 ,
and the Bianchi identity corresponding to the existence of a gauge potential is
[ H] = 0 H = B +
The field strength H and, hence, the action is invariant under gauge transformations
B B + ,
where the gauge parameter is now a vector field (and has its own gauge
invariance + in turn). Writing out the equation of motion gives:
( B + B + B ) = 0
This becomes the wave equation when we impose the analogon of the Lorenz
gauge:
B = 0 , if B = 0 and  = 0 , = 0 .
Solutions can be build up from plane waves. A plane wave with polarisation
b and momenum k has the form
B (x) = b eikx
This allows us to transform the equations of motion, the Lorenz gauge and the
residual gauge symmetry into momentum space:
B = 0 k 2 = 0
B = 0 k b = 0
B B + , b b + k k ,
where = 0

where k = 0 .
36

(172)

This shows that the massless antisymmetric tensor state of the closed string has
the properties of a massless rank 2 antisymmetric gauge field.
Tensor gauge fields and axions
In D = 4 a rank 2 gauge field B is equivalent to an axion, where by axion we
mean a scalar which has a shift symmetry a a + C, where C is a constant.
Therefore the action can only depend on a through its derivative H := a.
This might be viewed as a gauge theory with potential a and field strength H .
Starting from a rank 3 field strength we can use the four-dimensional  tensor
to define
1
H =  H .
3!
Then
H = 0 [ H] = 0 .
The equation of motion for H becomes a Bianchi identity for the vector H ,
implying that it can be obtained from a scalar field: H = a. Moreover
[ H] = 0 H = 0 .
The Bianchi identity for H becomes the equation of motion a for a
massless scalar field a. This shows that in four dimensions a rank 2 gauge field
can equivalently be described as an axion. The dual action is:
Z
S[a] = d4 x a a .
3.3.6

The graviton and the dilaton

We now turn to the symmetric part

s 1

1
|ki ,

s = s

of the massless closed string state. The momentum and polarisation tensor
satisfy
k 2 = 0 , k s = 0 .
One can show that states of the form
s = k + k where k = 0
are spurious. The corresponding gauge symmetry is
s s + k + k where k = 0 .
In contrast to the antisymmetric part, a 2nd rank symmetric tensor is not
irreducible under the Lorentz group. Its trace s is a scalar (because is
an invariant tensor) and therefore a symmetric tensor decomposes into two
irreducible representations: the trace and the traceless part.
37

In order to perform the decomposition explicitly we choose a lightlike vector

k which has unit scalar product with the momentum:
2

k2 = 0 , k = 0 ,

k k = 1 .

Then the traceless part of s is4

1
s k k k k
D2
and the remaining pure trace part of s is


1
s k k k k
=
D2
Note that
s = + , = 0 , = s .
= s

The trace part is physical,5

k = 0
and not spurious (the gauge transformation does not act on the trace).
It describes a scalar, called the dilaton. The purely traceless part still contains spurious states. It is a D-dimensional graviton and its gauge symmetry
+ k + k , where k = 0
is a linearized version of (space-time) diffeomorphism invariance.
As in the case of the photon we can be more explicit about the physical
and spurious components of the graviton/dilaton field, by choosing the explicit
momentum vector (165) and using the basis (167). The transversality k = 0
of the polarisation tensor implies:
k 0 00 + k 0 0,D1 = 0 0,D1 = 00
etc. Combining this with

00
10

..
( ) =
.

D2,0
D1,0

00
01

..
=
.

0,D2
00

symmetry = gives:
01
11

02
12

...
...

0,D2
1,D2

D2,1
D1,1

D2,2
D1,2

01
11

02
12

...
...

0,D2
1,D2

00
01
..
.

1,D2
01

2,D2
02

...
...

D2,D2
0,D2

0,D2
00

. . . D2,D2
. . . D1,D2

0,D1
1,D1
..
.
D2,D1
D1,D1

(173)

1
4 If you ask yourself: why not
= s D s ? (good question!), the short answer is
that this would be the wrong trace. Among the components of s there are several which
correspond to scalars. But only one of them is a physical degree of freedom, the dilaton, while
the others are spourious (gauge) degrees of freedom.
5 In contrast the trace 1 s
is also a scalar, but does not satisfy the physical state
D
condition.

38

Next, we have to work out the spurious part k + k in the basis. Since
k = 0, we have
(k )

(k 0 , 0, . . . , 0, k 0 )

(174)

(k )

(k0 , 0, . . . , 0, k0 )

(175)

( )

(0 , 1 , . . . , D2 , 0 )

(176)

Note that k 0 = k0 . Thus

2k0 0
k0 1

k0 2

(k + k ) =
..

k0 D2
2k0 0

k0 1
0
0

k0 2
0
0

k0 D2
0
0

2k0 0
k0 1
k0 2
..
.

0
k0 1

0
k0 2

0
k0 D2

k0 D2
2k0 0

We can add an arbitrary multiple of a spurious state to a physical state. This

can be used to decompose a physical state in the following way into a transverse
state transv and a spurious state spur :
transv.
spur.
=
+

where

transv.
(
)

and

0
0
..
.

0
0

0
1
..
.

0
11
1,D2
0

transv.
(
)=

D2
0

0
12
2,D2
0

...
...

1,D2 , 0

. . . D2,D2
...
0

..
.
0
0

1
0

2
0

D2
0

0
1
..
.

0
1

0
2

0
D2

D2
0

with arbitrary (subject to D1 = 0 ).

The transversality conditions eliminate D of the D(D+1)
components of .
2
D(D+1)
Among the remaining
D physical components D 1 are spurious. The
2
number of independent physical degrees of freedoms is
D(D + 1)
(D 2)(D 1)
D (D 1) =
2
2
In our parametrisation we see that they form a symmetric second rank tensor of
the little group SO(D 2). This representation is reducible: it decomposes into
39

the trace, which is a scalar, and the traceless part. The traceless, symmetric,
second rank tensor representation of the little group has dimension (D2)(D1)

2
1 and is the representation of the D-dimensional graviton.
In D = 4 the little group SO(2) is abelian and its irreducible representations
are one-dimensional. A traceless, symmetric, second rank tensor transforms as
follows:






11 12
cos sin
11 12
cos sin

12 11
sin cos
12 11
sin cos
To see the decomposition into two irreducible representations, introduce complex components (corresponding to a circular polarisation basis)
= 11 i12
Then
++ e2i ++ ,

e2i .

These sates carry helicity h = 2.

3.3.7

Review of the Pauli-Fierz Langrangian and of linearized gravity

to be written

The Light Cone action

The following section needs to be revised and extended
In conformal gauge we can still make conformal reparametrizations

m . One can fix reparametriza
( ), which are generated by the constraints Lm , L
tion invariance completely and get a formulation without constraints.
Introduce lightcone coordinates in space-time, X + , X , X i , where


1
X = X 0 X D1
2

(177)

are lightlike coordinates and X i , i = 1, . . . , D 2 = nT are spacelike and

transverse to the lightcone spanned by X . In terms of timelike and spacelike
world sheet coordinates 0 , 1 , conformal transformations take the form
0

1 + 0 1

( , ) +
( 0 , 1 ) ,
2


1 + 0 1

( , )
( 0 , 1 ) .
2

(178)

We can take any new

0 as a solution of the two-dimensional wave equations.
1
The new
is then fixed up to a constant (which corresponds to shifting the
origin of the
1 coordinate along the string). Since the X in the conformal
gauge also satisfy the two-dimensiona wave equation, we can use the residual
40

parametrization freedom to set 0 proportional to X + , up to an additive constant. This condition defines the light cone gauge:
X + = x+ + p+ 0

(179)

or n+ = 0 for n 6= 0. In terms of space-time light cone coordinates the constraints are

1
(180)
X X 0 = + (X i X 0i )2 .
2p
This can be solved for the n :
n

nT X
1 X
= +
i
i
2p i=1 m nm m

(181)

Thus in the light cone gauge only the transverse oscillations ni are independent.
In other words the constraints tell us that only transverse oscillations of the
string are physical degress of freedom. We can formulate the theory in terms of
a non-singular, but non-covariant action.
The light cone action is obtained by imposing the gauge and reducing the
Polyakov action to the independent, transverse degrees of freedom:
Z


T
S=
d2 (X i )2 (X 0i )2 .
(182)
2
Equation of motion
X 00 = 0 ,
X

(183)

where X = (X i ) only contains the transverse modes. Canonical momenta

= T X .

(184)

Hamiltonian



 T Z
d 1 X 2 + X 02
d 1 X L =
2 0
0

X
T
LC
(n n +
n
n ) = LLC
0 + L0
2 n=

Z
H

=
=

(185)

Dimensional reduction, T-duality and non-abelian gauge

symmetry
to be written

Literature
Katrin Becker, Melanie Becker and John H. Schwarz, String Theory and
M Theory, A modern introduction (2007).
41

The most recent textbook, attempts to give an introduction from a contemporary perspective and to treats virtually all recent developments.
More advanced than Zwiebach, but according to many colleagues very accessible and not too technical. If you want to have just one book which
covers it all, this is currently the best choice.
Barton Zwiebach, A First Course in String Theory (2004).
The most accessible textbook. Does not cover advanced or technical aspects, but includes recent developments such as brane world model building and black hole entropy.
Joseph Polchinski, String Theory (2 Volumes, 1998).
For many the standard textbook. Includes developments of the midnineties, such as D-branes. Covers many technical aspects, but is (by
opinion of many readers) not detailed enough to learn how its done
without accompanying lectures or further literature.
Michael B. Green, John H. Schwarz and Edward Witten, Superstring Theory (2 volumes, 1987).
The classical textbook. Though it does not cover the modern stuff, it is
a good reference if you need to know the details of the old stuff.
Dieter L
ust and Stefan Theisen, Lectures on String Theory (1989).
Concise, technical exposition of the old stuff. Covers material that is not
in Green, Schwarz, Witten. Out of print (I have a copy), and no easy/first
read.
Thomas Mohaupt, Introduction to String Theory (hep-th/0207249).
My humble attempt to summarize some relevant parts of string theory.
Reasonably up to date, I hope.

42