DOLGACHEV, I. Introduction To String Theory - Ps

Lecture 1
Quantum Field Theories: An

introduction
The string theory is a special case of a quantum field theory (QFT). Any QFT deals
with smooth maps : ! M of Riemannian manifolds, the dimension of is
the dimension of the theory. We also have an action function S defined on the set
Map(; M ) of smooth maps. A QFT studies integrals
Z
S()
e ~ V ()D[] (1.1)
Map( ;M )
Here D[] stands for some measure on the space of paths, ~ is a parameter (usually
very small, Planck constant) and V : Map(; M ) ! R is an insertion function. The
number e S ()=~ should be interpreted as the probability amplitude of the contribution
of the map : ! M to the integral. The integral
Z
S(
ZE = e ~ d (1.2)
Map( ;M )
is called the partition function of the theory. In a relativistic QFT, the space has a
Lorentzian metric of signature ( ; +; : : : ; +). The first coordinate is reserved for time,
the rest are for space. In this case, the integral (1.1) is replaced with
Z
ZM = eiS()=~ V ()D[]: (1.3)
Map( ;M )
Let us start with a 0-dimensional theory. In this case is a point, so : ! M is

a point x 2 M and S : M ! R is a scalar function. The Minkowski partition function
of the theory is an integral
Z
Z= eiS(x)=~ dx: (1.4)
M
Following the Harvard lectures of C. Vafa in 1999, let us consider the following
example:
1
2 LECTURE 1. QUANTUM FIELD THEORIES: AN INTRODUCTION
Example 1.1. Recall the integral expression for the -function:

Z 1 Z 1
(s) = ts 1 e t dt = 2 t2s 1 e t2 dt: (1.5)
0 0
This integral is convergent for Re (s) > 0 but can be meromorphically extended to the
whole plane with poles at s 2 Z0. We have
(s + 1) = s (s);
p
(1) = 1; ; ( 12 ) = :
By substituting t =
pat0 in (1.5), we obtain the Gauss integral:
Z1 r
(1)
e at2 dt = a1=22 = a : (1.6)
1
Although in the substitution above a is a positive real number, one can show that
formula (1.6) make sense, as a Riemann integral, for any complex a with Re (a) 0.
When Re (a) > 0 this is easy to see using the Hankel representation of (s) as a
contour integral in the complex plane. When a is a pure imaginary, it is more delicate
and we refer to [Kratzer-Franz], 1.6.1.2.
Taking a = , we can use
dG = e t2 dt
to define a probability measure on R. It is called the Gaussian measure. Let us compute
the integral
Z1 Z1
Z M () = eix3 dG = e x2+ix3 dx:
1 1
Here = 1=~ We have
Z1 1
X
Z () = exp( x2 ) (ix3 )n =n!)dx:
1 n=0
Obviously,
Z1
x2m 1 e x2 dx = 0:
1
Also
Z1
x2m e x2 dx = (m + 21 )=m+ 12 = 1 3 (2
(2m 1) =
)m
1
3
(2m)! = P (2m)=(2)m;
22m m!
m
where

P (2m) = (2 m)! = 1 2m 2m 2 2
m
2 m! m! 2 2 2
is equal to the number of ways to arrange 2m objects in pairs. This gives us
1
X
Z () = 1 + ( 1)n 2n P (6n)=(2n)!(2)3n : (1.7)
n=1
Observe that to arrange 6n objects in pairs is the same as to make a labelled 3-valent
graph with 2n vertices by connecting 1-valent vertices of the following disconnected
graph:
b1 b2 b 2n
c2 c 2n
c1
a1 a2 a 2n
Fig. 1
This graph comes with labeling of each vertex and an ordering of the three edges
emanating from the vertex. Let be such a graph, V ( ) be the number of its vertices
and E ( ) be the number of its edges. We have 3V ( ) = 2E ( ), so that V ( ) =
2n; E ( ) = 3n for some n. Let
V( )
W ( ) = ( Vi() )! (2)1E( ) :
Then
X
Z () = 1 + W ( );
where the sum is taken over the set of labeled trivalent graphs. Let N () be the number
of labelled trivalent graphs which define the same unlabelled graph when we forget
about the labelling. We can write W () = N ()W ( ), where N () is the number of
labelling of the same unlabelled 3-valent graph . Thus
X
Z () = 1 + W ();

where the sum is taken with respect to the set of all unlabelled 3-valent graphs. It is
easy to see that
2n
N () = (2#nAut
)!(3!) ;
()
so that
2n n)!(3!)2n ( 6i)V () :
W () = (2( ni)!2) )(2 =
3n #Aut() (2 )E () #Aut()
Given an unlabelled 3-valent graph with 2n vertices, we assign to each vertex a factor
( 6i), to each edge a factor 1=2, then multiply all the factors and divide by the
number of symmetries of the graph. This gives the Feynman rules to compute the
contribution of this graph to the coefficient at 2n . For example, the graph
contributes ( 6i)2 (21 )3 121 = 3

83 and the graph
contributes ( 6i)2 (21 )3 18 = 1693 : The total coefficient at 2 is 15

163 . This
coincides with the coefficient at 2 in Z () given by the formula (1.7).
Recall that the Principle of Stationary Phase says that the main contributions to the
integral
Z
eif (x)(x)dx
R
when goes to infinity comes from integrating over the union of small comapct neigh-
borhoods of critical points of f (x). More precisely we have the following lemma:
Lemma 1.1. Assume (x) has a compact support V and f (x) has no critical points
on V . Then, for any natural number n,
Z1
lim n
!1
eif (x)(x)dx = 0:
1
Proof. We use induction on n. The assertion is obvious for n = 0. Integrating by parts,
we get
Z1 1
i eif (x) (x) 0 dx = i (x) eif (x)1 + Z eif (x)(x)dx =
f (x)0 f (x)0 1
1 1
Z1
eif (x)(x)dx:
1
Multiplying both sides by n+1 , we get
Z1 Z1
eif (x) S((xx))0 0 dx:

lim n+1
!1
eif (x)(x)dx = i lim
!1
n
1 1
Applying the induction to the function f (x)0 (x) 0 we get the assertion.
5
Thus if f (x) has finitely many critical points x1 ; : : : ; xk , we write our function
f (x) as a sum of functions fi (x) with support on a compact neighborhood Ki of xi
and a function F (x) which has no critical points on the support of (x) and obtain, for
any n > 0,
Z1 X Z
eif (x)=~ (x)dx = eifi (x)=~ (x)dx + o(~n ):
1 i Ki
Now let us consider a QFT in dimension 1. Usually we write D = d + 1, where d

is the space-dimension, and 1 is the time-dimension. A QFT in dimension 0 + 1 is the
quantum mechanics. In this case, we take to be equal to R, I = [0; 1] or S 1 = R=Z
parametrized by t. A map : ! M is path in M (infinite, or finite, or a loop). The
action is defined by
Z
S ( (t)) = L( (t); _ (t))dt;

where L : TM ! R is a smooth function defined on the tangent space of M (a La-
grangian). The expression L( (t); _ (t))dt is a density on equal to the composition
of the differential d : T ! TM and L.
For example, take M = Rn so that TM = Rn Rn with coordinates (q; q_ ). For
any L(q; q_ ) and a map x : [a; b] ! Rn , L(x(t); x_ (t)) is obtained by replacing q with
x(t) and q_ with x_ (t).
A critical point of the functional S (x(t)) satisfies the Euler-Lagrange equation
@L d @L
@qi (x(t); x_ (t)) = dt @ q_i (x(t); x_ (t)): (1.8)
For example, let us take the Lagrangian

n
X
q_i2 V (q1 ; : : : ; qn ) (1.9)
i=1
Then we get from (1.8)
2
m d dtx(2t) = rV (x(t)):
Thus a critical path satisfies the Newton Law; it gives the major contribution to the
partition function.
Fix x; x0 2 M and t; t0 2 . Let P (t; x; t0 ; x0 ) be the space of smooth maps
: ! M such that (t) = x; (t0 ) = x0 . The integral
Z
K (t; x; t0 ; x0 ) = eiS( (t))=~D[ (t)] (1.10)
P (t;x;t0;x0 )
can be interpreted as the probability amplitude that a particle in the position x at the
moment of time t moves to the position x0 at the time t0 .
Let us compute it for the action defined by the Lagrangian (1.9) with M = R. We
shall assume that the potential function V is equal to zero.
The space P (t; x; t0 ; x0 ) is of course infinite-dimensional and the integration over
such a space has to be defined. Let us first restrict ourselves to some special finite-
dimensional subspaces of P (t; x; t0 ; x0 ). Fix a positive integer N and subdivide the
time interval [t; t0 ] into N equal parts of length = (t0 t)=N by inserting inter-
mediate points t1 = t; t2 ; : : : ; tN ; tN +1 = t0 . Let us choose some points x1 =
x; x2 ; : : : ; xN ; xN +1 = x0 in Rn and consider the path : [t; t0 ] ! Rn such that
its restriction to each interval [ti ; ti+1 ] is the linear function
i (t) = xi + xti+1 txi (t ti ):

i+1 i
It is clear that the set of such paths is bijective with (Rn )N 1 and so we can integrate
a function F : P (t; x; t0 ; x0 ) ! R over this space to get a number JN . Now we can
define (1.10) as the limit of integrals JN when N goes to infinity. However, this limit
may not exist. One of the reasons could be that JN contains a factor C N for some
constant C with jC j > 1. Then we can get the limit by redefining JN , replacing it
with C N JN . This really means that we redefine the standard measure on (Rn )N 1
replacing the measure dn x on Rn by C 1 dn x. This is exactly what we are going to
do. Also, when we restrict the functional to the finite-dimensional space (Rn )N 1 of
R t0
piecewise linear paths, we shall allow ourselves to replace the integral t L( ; _ )dt by
its Riemann sum. The result of this approximation is by definition the right-hand side
in (1.10). We should immediately warn the reader that the described method of giving
a value to the path integral is not the only possible.
We have
Z1 Z1 N
K (t; x; t0 ; x0 ) = Nlim ::: im X
exp[ 2 (xi xi+1 )2 ]C N dx2 : : : dxN :
!1 i=1
1 1
(1.11)
Here x2 ; : : : ; xN are vectors in Rn and dxi is the standard measure in Rn . The number
C should be chosen to guarantee convergence in (1.11). Using (1.6) we have
Z1 Z1
x1 +x3 2 a (x1 x3 )2
e a(x1 x2 )2 a(x2 x3 )2 dx2 = e 2a x2 2 2 dx2 =
1 1
Z1 r
=e a2 (x1 x3 )2
e 2ax2 dx = 2a e a2 (x1 x3 )2
:
1
Next
Z1
exp[ a2 (x1 x3 )2 a(x3 x4 )2 ]dx3 =
1
7
Z1 r
= exp[ 32a x3 x1 + x4 2 a (x x )2 ]dx = 2 e
3 3 1 4 3 3a
a3 (x1 x4 )2
:
1
Thus
Z1
exp[ a(x1 x2 )2 a(x2 x3 )2 a(x3 x4 )2 ]dx2 =
1
r r r
2 exp[ a (x x )2 ] = 2 exp[ a (x x )2 ]:
2a 3a 3 1 4 3a2 3 1 4
Continuing in this way, we find
Z1 N r
exp[ a
X
(xi xi+1 )2 ]dx N 1 exp[ a (x x )2 ];
2 : : : dxN = NaN 1 N 1 N +1
1 i=1
21
where a = m=2i: If we choose the constant C equal to C = m ; then we will
2i
be able to rewrite (1.11) in the form
m 21 mi(x0 x)2 12 mi(x0
e 2N = 2i(m
x)2
K (t; x; t0 ; x0 ) = 2iN t0 t) e
2(t0 t) : (1.12)
We shall use K (t; x; t0 ; x0 ) to define a certain Hermitian operator in the Hilbert

space L2 (R). Recall that for any manifold M with some Lebesgue measure d the
space L2 (M; d) consists of square integrable complex valued functions modulo func-
tions equal to zero on the complement of a measure zero set. The hermitian inner
product is defined by
Z
hf; gi =
fgd:
M
Example 1.2. An example of an operator in L2 (M; d) is a Hilbert-Schmidt operator:
Z
Tf (x) = K (x; y)f (y)d;
M
where K (x; y ) 2 L2 (M M; ) is the kernel of T . In this formula we integrate
keeping x fixed. By Fubinis theorem, for almost all x, the function y ! K (x; y )
is -integrable. This implies that T (f ) is well-defined. Using the Cauchy-Schwarz
inequality, one can easily checks that
Z Z Z
jjTf jj2 = jTf j2d jjf jj2 jK (x; y)j2 dd;
M M M
i.e., T is bounded, and

Z Z
jjT jj2 = sup jjjjTf jj
2
f jj2 jK (x; y)j2 dd:
f 6=0 M M
We have
Z Z Z Z
(g; Tf ) = f (y)K (x; y)d g(x)d = K (x; y)f (y)g(x)dd:
M M M M
This shows that the Hilbert-Schmidt operator is self-adjoint if and only if
K (x; y) = K (y; x)
outside a subset of measure zero in M M.
In quantum mechanics one often deals with unbounded operators which are defined
only on a dense subspace of a complete separable Hilbert space H. So let us extend
the notion of a linear operator by admitting linear maps D ! H where D is a dense
linear subspace of H (note the analogy with rational maps in algebraic geometry). For
such operators T we can define the adjoint operator as follows. Let D(T ) denote the
domain of definition of T . The adjoint operator T will be defined on the set
D(T ) = fy 2 H : sup jhT (x); yij < 1g:

06=x2D(T ) jjxjj
Take y 2 D(T ). Since D(T ) is dense in H the linear functional x ! hT (x); y i

extends to a unique bounded linear functional on H. Thus there exists a unique vector
z 2 H such that hT (x); yi = hx; z i. We take z for the value of T at y. Note that
D(T ) is not necessary dense in H. We say that T is self-adjoint if D(T ) = D(T )
and T = T . We shall always assume that T cannot be extended to a linear operator on
a larger set than D(T ). Notice that T cannot be bounded on D(T ) since otherwise we
can extend it to the whole H by continuity. On the other hand, a self-adjoint operator
T : H ! H is always bounded. For this reason self-adjoint linear operators T with
D(T ) 6= H are called unbounded linear operators.
Example 1.3. Let us consider the space H = L2 (R; dx) and define the operator
df :
Tf = if 0 = i dx
Obviously it is defined on the space of differentiable functions with square integrable
derivative. This space contains the subspace of smooth functions with compact support
which is known to be dense in L2 (R; dx). Let us show that the operator T : D ! H is
self-adjoint. Let f 2 D(T ). Since f 0 2 L2 (R; dx),
Z t Z t
f 0 (x)f (x)dx = jf (t)j2 jf (0)j2 f (x)f 0 (x)dx
0 0
9
is defined for all t. Letting t go to 1, we see that limt!1 f (t) exists. Since jf (x)j2
is integrable over ( 1; +1), this implies that this limit is equal to zero. Now, for any
f; g 2 D, we have
Z t Z t
0(x)g(x)dx = lim if (t)g(x)+1

(Tf; g) = tlim
!1 0 if t!1 1
if (x)g0 (x)dx =
0
Z t
= tlim
!1 f (x)ig0 (x)dx = (f; Tg):
0
This shows that D D(T ) and T is equal to T on D. The proof that D = D(T )
is more subtle and we omit it.
Let H1 ; H2 be two copies of the space L2 (M; d). Let Tt;t0 be the Hilbert-Schmidt
operator H1 ! H2 defined by a kernel K (t; x; t0 x0 ) which has t; t0 as real parameters:
Z
Tt;t0 ()(x) = K (t; x; t0 ; x0 )(x0 )dM :
M
Suppose our kernel has the following properties:
(M)
Z t00 Z
K (t; x; t00 ; x00 ) = K (t; x; t0 ; x0 )K (t0 ; x0 ; t00 ; x00 )dM dt0 ; t < t00 ;
t M
(N)
Z
jK (t; x; t0 ; x0 )j2 dM = 1;
M
(T)
K (t1 ; x; t01 ; x0 ) = K (t2 ; x; t02 ; x0 ) if t02 t2 = t01 t1 ;

(C) for any ; 2 L2 (M; d), the function
Z
t ! (t) = (x)K (t; x; t0 ; x0 )(x0 )dM dX
M
is continuous for t0 > t and limt0 !t+ (t) = h ; i:
When K is defined by the path integral, property (M) is taken as one of the axioms of
QFT. It expresses the property that any path : [t; t00 ] ! M from x to x00 is equal to
a sum of paths 1 : [t; t0 ] ! M from x to x0 and a path 2 : [t0 ; t00 ] ! M from x0 to
x00 . Property (N) says that the total probability amplitude of a particle to move from x
to somewhere is equal to 1. Notice that property (N) implies that the operator Tt;t0 is
unitary. In fact,
Z
Tt;t0 Tt;t0 dM =
M
Z Z Z
K (t; x; t0 ; x0 )(x)dM K (t; x; t0 ; x0 ) (x)dM dM =
M M M
Z Z Z
(x) K (t; x; t0 ; x0 )K (t; x; t0 ; x0 )dM (x)dM = (x) (x)dM :
M M M
Now we use the following Stone-von Neumanns Theorem:
Theorem 1.1. Let U (t); t 2 R0 be a family of unitary operators in a Hilbert space
H. Assume that
(i) for all u; v 2 H, the function t ! F (t) = (u; U (t)v) is continious for t > 0 and
limt!0+ F (t) = (u; v);
(ii) for all t; t0 2 R0 ; U (t + t0 ) = U (t) U (t0 ):
Then
lim0+ U (t)t I u
D = fu 2 H : t! existsg
is dense in H and the operator defined by
lim0+ U (t)t I u
Hu = i t!
is self-adjoint. It satisfies
U (t) = eitH ; t 0:
Applying this to our situation, we obtain that
Tt0 ;t = e i(t t0 )H ; t t0
for some linear operator H . The operator H is called the Hamiltonian operator asso-
ciated to K (t; x; t0 ; x0 ).
We would like to apply the above to our function
K (t; x; t0 ; x0 ) = 2i(m
21
exp im(x0 x)2 :
t0 t) 2(t0 t)
Unfortunately we cannot take the function K (t; x; t0 ; x0 ) to be the kernel of a Hilbert-
Schmidt operator. Indeed, it does not belong to the space L2 (R2 ; dxdx0 ). In particular
property (N) is not satisfied. One can show that (M) is OK, (T) is obviously true and
(C) is true if one restricts to functions ; from a certain dense subspace of L2 (R).
The way about this is as follows (see [Rauch]).
First let us recall the notion of the Fourier transform in R. It is a linear operator
defined on the Schwartz space S (R) L2 (R) of smooth functions with all derivatives
tend to zero faster than any power of jxj as x ! 1. It is given by the formula
Z 1
F ((x)) := ^( ) = p12 e ix (x)dx:
1
Here are some of the properties of this operator:
11
(i) F : S (R) ! S (R) is an unitary operator;

(ii) F 1 ((x)) = F (( x));
(iii) F ((x)0 ) = i ^( ));
(iv) ^( )0 = F ( ix(x));
p
(v) F ( ) = 2F ()F ( ), where
Z1
(x) = (x y) (y)dy:
1
Let us show that our function K (t; x; t0 ; x0 ) is the propagator for the Schrodinger equa-
tion
@ + m @ 2 u(t; x) := iu + m u = 0; u(0; x) = f (x) 2 L2 (R):
i @t 2 @x2 t 2 xx
We take for simplicity m = 1. Supose f (x) 2 S (R). Let us find the solution in
S (R) using the Fourier transform. Using property (iii), we get u^t = 12 i 2 u^ (we
use the Fourier transform only in the variable x). Integrating this equation with initial
^(0; ) = f^( ), we get u^(t; ) = e it2 =2 f^( ):
condition u
Taking the inverse Fourier transform, we get
Z 1
u(t; x) = F 1 (e it2 =2 f^( )) = p1 e 12 it2 +ix f^( )d: (1.13)
2 1
Clearly, u(0; x) = F 1 F (f ) = f (x): Of course, we have still to show the existence
of a solution. We skip the check that formula (1.13) gives a solution in S (R). This
defines us a linear operator (the propagator)
S (t) : S (R) ! S (R); f (x) ! u(t; x):
We would like to show that it is an integral operator and find its kernel. Let K (t; x) =
F 1 ( p12 e it2 =2 ). Then
Z1 Z1 Z1
p1 e

K (t; x y)f (y)dy = it2 =2 ei(x y) d f (y)dy =
1 1 1
2
Z1 Z1
p1 e

iy f (y )dy eix e it2 =2 d = F 1 (e it2 =2 f^( )) = u(t; x):
1 1
2
2
Unfortunately, this computation is wrong since the function e it =2 does not belong
to S (R). A way about it is to consider this function as a distribution and extend the
Forier transform to distributions.
Recall that a distribution is a continuous linear functional on the space C 1 (R)0 of

smooth functions with compact support (test functions). Any function f which can be
integrated over any finite closed interval (but not necessary over the whole R) can be
considered as a distribution. Its value at a test function is equal to
Z1
f () = f (x)(x)dx;
1
where the bar denotes the complex conjugation. Such a distribution is called a regular
distribution or a tempered distribution. The rest are called singular distributions. We
shall denote the value of a distribution f on a test function by
Z1
f () = (x)f(x)dx:
1
If f is a regular distribution defined by a function f (x) from L2 (R), then
f () = hf; i:
An example of a singular distribution is the delta-function (x a) whose value at a test
function is equal to (a). It is also denoted by a . A linear operator T : D ! L2 (R)
with C 1 (R)0 D \ D(T ) extends to the space of distributions by the formula
Tf () = f (T ):
If f 2 L2 (R), viewed as a regular distribution, we have
Tf () = hT ; f i = h; Tf i
so the two definitions agree.
For example let T = dx d be defined on the space of functions with square integrable
derivative. We have T = T and for any distribution f , f 0 () = f ( ). If f is a

tempered distribution defined by an integrable differential function f such that f 0 also
defines a tempered distribution, then the formula of integration by parts shows that this
definition agrees with the usual definition of derivative.
Since the Fourier transform is an example of an operator defined on S (R) with
F = F 1 , we can define the Fourier transform of a distribution f by
F (f )() = f (F 1 ()):
All the properties (i)-(v) extend to distributions. In property (v) we define the convolu-
tion of a regular distribution f and an element of S (R) by the formula
f () = f ( ):
Lemma 1.2. For any a 2 C with Re (a) 0; a 6= 0,
F (e ax2 ) = p1 e x 2 =4 a :
2a
13
2
Proof. Assume first that Re (a) > 0. Then the function e ax belongs to S (R) and,
using the Gaussian integral, we obtain
Z1 Z1
F (e ax2 ) = p1 2 1
e ax2 e ix dx = p1
2 1

e a(x+ 2ai )2 e 2 =4a dx =
p
p1 pa e 2 =4a = p1 e 2 =4 a :
2 2a
Therefore, for any 2 C 1 (R)0 ,
Z1
F (e ax2 )() = p1 e 2 =4a ( )d:
2a 1
Consider the both sides as functions of a. When Re (a) > 0 each side is a holo-
morphic function, and, for Re (a) 0; a 6= 0, are continuous functions. The unique
continuation principle for holomorphic functions implies that the two sides are equal
for Re (a) 0; a 6= 0. This proves the lemma.
Now we can use the lemma to set
K (t; x) = F 1 ( p1 e it2 =2 ) = F ( p1 e it( )2 =2 ) = p 1 ei x2t :

2
2 2 2it
Property (v) of Fourier transform gives us
Z1
S (t)f = K (t; x) f = K (t; x y)f (y)dy =
1
Z1 Z 1
p 1 ei (x 2ty) f (y)dy = K (t; x; y)f (y)dy:
2
1
2it 1
Thus we see that the integral operator with the kernel K (t; x; y ) = K (t; x y); t > 0;
is well-defined as an operator S (t) on the space S (R). Now observe that
k S (t)f k = k u(t; x) k = k F 1 (e it2 =2 f^( )) k =
ke it2 =2 f^( ) k = k f^( ) k = k f k:

This shows that S (t) is a unitary operator. In particular, S (t) is bounded on S (R) (of
norm 1) and hence continuous. It is known that S (R) is dense in L2 (R). Thus we can
extend S (t) by continuity to an unitary operator on the whole space L2 (R). It satisfies
the property
F (S (t)f ) = e it2 =2 F (f ):
Using property (iii) we get
S (t)f = e itH f; t > 0

2
where H = dxd 2 : This justifies the claim that K (t; x; t0 x0 ) is the kernel of the operator
S (t t) = e i(t0 t)H on L2(R).
0
Finally let us try to justify the following formula from physics books:
K (t; x; t0 x0 ) = hxje i(t0 t)H jx0 i (1.14)
First of all for any ; from a Hilbert space H, physicists employ the bra-ket notation
hj i := h; i:
If T is a linear operator in H, then
hjT j i := h; T i:
Let be a normalized eigenfunction of an operator T with an eigenvalue . Physicists
denote it by ji (although it is defined only up to a factor of absolute value one). To
simplify the notation they set
hjjT jji = hjT ji:

Consider an operator (the position operator)
Q : S (R) ! L2 (R); f ! xf:

It is a self-adjoint operator S (R) ! L2 (R). Its eigenfunctions do not belong to the
space L2 (R) but rather to the space of distributions.
We have
Qa() = a (Q) = a (x) = a(a) = aa ():

Thus a can be considered as an eigendistribution of Q with eigenvalue a. Thus for any
x 2 R0 we have, according to physicists notation, jxi = x. Now we have to compute
e i(t t) x . Recall that we can view it as an integral operator with kernel K (t; x; t0 ; x0 )
defined on the Schwartz space S (R) which obviously contains C 1 (R)0 . We have
Z1
K (t; x; t0 ; x0 )(x0 b)dx0 = K (t; x; t0 ; b);
1
15
ha je i(t0 t) jb i = hb jK (t; x; t0 ; b)i := a (K (t; x; t0 ; a)) = K (t; a; t0 ; b):
Taking a = x; b = x0 we get formula (1.14). We have to understand it as
x (e iH (t0 t) x0 ) = K (t; x; t0 ; x0 ):

For any function V : M !C and a point t 2 one can consider a function on the
set Map(; M ) defined by
V [t]() = V ((t)):
Let V1 ; : : : ; Vn be functions on M and t1 ; : : : ; tn 2 , one can consider the integral
Z
hV1 [t1 ]; : : : ; Vn [tn ]iM := V1 [t1 ]() : : : Vn [tn ]()eiS() D:
Map( ;M )
The right-hand side is called the path integral with insertion functions V1 ; : : : ; Vn . The
left-hand-side is called the correlation n-function. In the example above
hx [t]; x0 [t0 ]i = K (t; x; t0 ; x0 ):
Exercises
1.1 Find the Feynman rules to compute
Z1
Z () = e x2 +ix4 dx:
1
Compute the coefficient at 2 .
1.2 Show that the distribution G(t; x; y ) = K (0; t; x; y) (defined to be zero for t 0)
is a generalized solution of the equation
ut 21 uxx = (t)(x y);

(you have to give the meaning of the right-hand-side).
1.3 Show that, for any 0, the function eix is a generalized eigenfunction of the
d in L2 (R) and any generalized eigenfunction coincides with one of these
operator i dx
functions.
1.4 Find the Fourier transform and the derivative of the Dirac function (x a).
Lecture 2
Partition function as the trace of

an operator
Recall that the trace Tr(T ) of an operator T in a finite dimensional Hilbert space H is
equal to the sum of the diagonal entries of a matrix of T with respect to any basis. If
we choose an orthonormal basis (e1 ; : : : ; en ), then
n
X
Tr(T ) = hTei ; ei i: (2.1)
i=1
If T is a normal operator (e.g. Hermitian or unitary), then one can choose an orthonor-
mal basis of V consisting of eigenvectors of T . In this case
X
Tr(T ) = d(); (2.2)
2Sp(T )
where Sp(T ) is the spectrum of T (the set of eigenvalues) and d() is equal to the
dimension of the eigensubspace corresponding to the eigenvalue . Notice that
Y
det(T ) = d() :
2Sp(T )
This gives
Tr(T ) = ln(det(eT )): (2.3)
There are several approaches to generalize the notion of the trace to operators in infinite-
dimensional Hilbert spaces. We shall briefly discuss them. First assume that T is a
bounded operator. First we try to generalize the definition of a trace by using (2.1).
One chooses a basis (e1 ; : : : ; en ; : : : ) and sets
1
X
Tr(T ) = hTei ; ei i;
i=1
17
18 LECTURE 2. PARTITION FUNCTION AS THE TRACE OF AN OPERATOR
if the series convergent. If the convergence is absolute, then this definition does not
depend on the choice of a basis. In this case T is called a trace-class operator. For
example, one can show that Tr(AB ) = Tr(BA) if both A and B are trace-class. An
example of a trace-class operator is a Hilbert-Schmidt operator in the space L2 (M ; d).
If K (x; y ) is its kernel, then
Z
Tr(T ) = K (x; x)d:
M
When T is a self-adjoint Hilbert-Schmidt operator, the two definitions coincide. This
follows from the Hilbert-Schmidt Theorem.
Example 2.1. Let M be a finite set f1; : : : ; ng equipped with the measure d(A) =
#A. Then L2 (M; d) = Rn with inner product
Z n
X
h; i = d = ai bi ;
M i=1
where = (a1 ; : : : ; an ); = (b1 ; : : : ; bn ). It is clear that K (x; y ) can be identified
with a matrix K = (kij ) and
n
X
T(i) = kij aj ;
j =1
so that T is a linear operator defined by the matrix K . Then its trace is equal to
n
X
Tr(T ) = kjj :
j =1
This agrees with definition (2.1) when we take the standard orthonormal basis of Rn .
As we have already mentioned, in physics one deals with unbounded linear opera-
tors in L2 (M; d) like a differential operator. One tries to generalize definition (2.3).
Notice that
n
X d( ni=1 i s ) :
P
ln det(T ) = ln(i ) = ds s=0
i=1
Now for any T such that H has a basis of eigenvectors of T one can define the
zeta-function of T as follows. Let 0 < 1 < 2 : : : < n < : : : be the sequence of
positive eigenvalues of T . One sets
1
X mn ;
T (s) = s
n=1 n
where mn is the multiplicity of n , i.e. the dimension of the eigensubspace of eigen-
vectors with eigenvalue n . When H = L2 (M; d), where M is a compact manifold
of dimension d and T is a positive elliptic differential operator of order r, one can show
19
that T (s) is an analytic function for Re (s) > d=r and it can be analytically extended
to an open subset containing 0. In this case we define
0
det(T ) = e T (0) :
This obviously agrees with (2.3) when V is finite-dimensional. Also it is easy to see
that for any positive number
det(T ) = T (0) det(T ): (2.4)
This of course agrees with the finite-dimensional case because T (0) = dim V .
2
d 2 which acts on the space L2 (S 1 ), where
Example 2.2. Consider the operator T = dx
1
S = R=2rZ with the usual measure dx descended to the factor. Note that in this
measure the length of S 1 is equal to 2r, i.e. S 1 is the circle of radius r. The measure
dx corresponds to the choice of metric on the circle determined by its radius. The
normalized eigenvectors of T are p21r einx=r ; n 2 Z. The positive part of the spectrum
consists of numbers (n=r)2 ; n 2 Z>0 with mn = 2. Thus
1
X
T (s) = 2 (n=r) 2s = 2r2s (2s);
n=1
where (s) is the Riemann zeta function. It is known to be an analytic function for
Re (s) > 1=2. This agrees with the above since S 1 is one-dimensional and T is an
elliptic operator of second order. We have
(0) = 12 ; 0 (0) = 21 ln 2:

Thus
T0 (0) = 2 ln(2r);

and
det0 (
d2 ) = (2r)2 :
dx2 (2.5)
Example 2.3. Let us consider the path integral when = R=2RZ and M = R=2rZ.
We use the action
Z 2R
1
S ( ) = 2 0 (t)2 dt:
0
A map : ! M extends to a map of the universal coverings ~ : R ! R. It satisfies
~(t) = ~(t + 2R) + 2nr for some integer n (equal to the degree of the map of
oriented manifolds). Let Map(; M )n be the set of maps corresponding to the same n.
It is clear that each 2 Map(; M )n can be uniquely written in the form
(t) = nr
R t + 0 (t) = n (t) + 0 (t);
where 0 (t) satisfies 0 (t + 2R) = 0 (t), hence belongs to L2 (). The value of S
on such is equal to
Z 2R
1
S ( ) = 2 ( 00 (t)2 + 2 00 (t) n0 (t) + n0 (t)2 )dt:
0
We have
Z 2R Z 2R
00 (t) n0 (t)dt = nr
R 00 (t)dt = nr
R ( 0 (2R) 0 (0)) = 0:
0 0
Thus
2 2 Z 2R
S ( ) = nRr + 12 00 (t)2 dt:
0
The Minkowski partition function is
Z X Z
2 2
ZM = eiS( ) D = ei n Rr eiS( )D :
Map(;M ) n2Z Map( ;R)
Observe that we must have
Z Z
eiS( ) D = K (0; x; 2R; x)dx;
Map( ;R) M
where
K (t; x; t0 x0 ) = p 1 ei(x0 x)2 =2(t0 t) ;

0
2i(t t)
to be consistent with the previous computation of the path integral.
This gives
Z
Z = ei n Rr p1 pr
X2 2 X 2 2
dx = ei n Rr :
n2Z 2 iR M iR n2Z
Now we apply the Poisson summation formula
e xn2 = p1x e n2 =x :

X X
n2Z n2Z
Taking x = r2 =iR, we get
Z (r; R) = x
p X X X
e n2 x = e n2 =x = e iRn2 =r2 : (2.6)
n2Z n2Z n2Z
d2
Let us compute the trace of the operator e i2RH = eiR dt2 . Its normalized
eigenfunctions in L2 (M ) are the functions n = p21r einx=r . By (2.1), we have
X 2 X
Tr(e i2RH ) = h n jeiR dtd 2 n i = e iRn2 =r2 :
n2Z n2Z
21
Comparing this with (2.6), we see that
Z = Tr(e i2RH );
where H = 1 d2
2 dt2 .
Remark 2.1. If we repeat the computations for the Euclidean partition function (replac-
ing S with iS ) we get
Z E (R; r) = pr
X X
e n2 r2 =R = e n2 R=r2 :
R n2Z n2Z
This shows that
Z E (R; r) = pr Z E (R; R=r):

R
If we modify the partition function by inserting the factor 1=
pr, we get
Z E (R; r) = Z E (R; R=r):
This is the first glimpse of the T-duality.
Let
X
( ) = ein2 ; 2 U = fx + iy 2 C : y > 0g:
n2Z
be the modular form associated to the quadratic form Q(x) = x2 ( equal to the value at
zero of the Riemann theta function in one variable). It satisfies the functional equation
( 1= ) = ( i ) 12 ( ):
(the proof uses the Poisson summation formula). Observe that
Z E (R; r) = (iR=r2 ):
This is our first encounter with the theory of modular forms.
There is another way to compute the partition function for the action
Z=
Z
exp
1 Z 2R x0 (t)2 dtDx(t);
Map( ;R) 2 0
where is the circle of radius R. Notice that
Z 2R Z 2R
x0 (t)2 dt = x(t)x0 (t)2R

0 x(t)x00 (t)dt = hx(t); x(t)00 iL2 (S1 ) :
0 0
Thus
Z
ZE = e hx(t);Hx(t)i;
Map( ;R)
where H = 1 d2
2 dt2 . The integral
Z
e hx(t);Tx(t)iDx(t);
can be thought as a generalization of the Gaussian integral since hx(t); Tx(t)i is a
quadratic form in V = L2 (S 1 ). If V = Rn and T is a positive-definite self-adjoint
operator, we could use the orthogonal change of variables to diagonalize T and write
Z Z
n
e hx;T xidx1 : : : dxn = e 1 x21 ::: n x2n dx1 : : : dxn =
R Rn
n Z1 n r n=p2 = pn=2 = det( 1 T ) 12 :

Y
e i x2i dxi = Y = Qn
i=1 1 i=1 i i=1 i det(T )
Here we assumed that all eigenvalues i are positive, or equivalently, that the quadratic
form Q = hx; T xi is positive definite. To get rid of let us change the measure on R
replacing dx with p12 dx so that
Z
p
e hx;T xi( ) n dx1 : : : dxn = det(T ) 12 :
Rn
Now, for any normal positive definite operator
P T : H ! H in a Hilbert space H, we
can write any element 2 H as a sum an n , where (n ) is an orthonormal basis
of eigenvectors of T . The coordinate an is an analog of the xi coordinate from above.
This motivates the following definition
Z
C e hv;Qvi Dv = det0 ( 1 Q) 1=2 : (2.7)
V
Here the measure D[v ] is defined up to some multiplicative constant C . In fact we will
be defining the correlation functions by the formula
R
hV1 (t1 ); : : : ; Vn (tn )i = V1 (t1 )R eiSVn( ()tDn )[e ] D[ ] ;
iS ( )
so the choice of the constant will not matter. We would like to appy this to the op-
erator H = dt d22 in L2 (R=2R). However, not all of its eigenvalues are positive.
Constant functions
P form the nullspace of this operator. If we decompose each vector
as a sum n an n of normalized eigenvectors, then coefficients an will be analogs
of the coordinates in Rn . So, we can write our space as the product of the space of
p and functions pwith a0 = 0. The coefficient of the constant function
constant functions
1 at 0 = 1= 2R Ris equal to 2R. Thus the integral over the space of constant
2r p p p
functions is equal to 0 2R(dx= 2) = 2r R: So, using (2.5), we obtain
p p
Z = (2R) 1 (2r R) = r= R:
23
This agrees with the computations in example 2.3 if we switch from the Minkowski
partition function to the Euclidean one.
Here is another application of the Gaussian integral for quadratic functionals. Con-
sider the action functional S ( ) defined by some Lagrangian L : TM ! R. We know
that its stationary points are classical solutions. Write (t) = c (t) + (t), where
c(t) is a classical solution. Then
2
S ( ) = S ( c ) + 21 S2 = ( )2 + terms of higher order in :
c
This gives a semi-classical approximation:
Z 2
exp(iS ( c )=~)[ 21i det( S2 )
X
exp(iS ( )=~)D[ ] ] 21 :
classical solutions
= c
This approximation is exact when the action is quadratic in .

Example 2.4. We take = R=2L and M = R. The Lagrangian is
L(q; q_) = 12 (q_2 !2q2 )
and the Minkowski partition function is
Z Z
Z M = D[ (t)] exp 2i ( _ (t)2 !2 2 (t) dt:

It is called the path integral of the harmonic oscillator. The kernel of the operator
e i(t t0 )H is given by
Z (t1 )=x Z
D[ (t)] exp 2i ( _ (t)2 !2 2(t) dt:

K (t0 ; y; t1; x) =
(t0 )=y
Choose a critical path cl for the action defined by our Lagrangian and decompose
the action in the Taylor expansion at cl .
2
S ( (t)) = S ( cl ) + 21 S2 = ( (t) cl ): (2.8)
cl
The classical path is a solution of the Lagrangian equation:
d @L @L 00 2
dt @ q_ @q = (t) + ! (t) = 0:
Its solution satisfying the initial condition (t0 ) = y; (t1 ) = x is
sin ! (t
1 t) + x sin ! (t t0 ) :

cl (t) = y sin !(t1 t0 ) sin !(t1 t0 )
The value of the action functional on the classical solution is

Z t1 ! (y2 + x2 ) cos !(t1 t0 ) 2xy
S ( cl ) = 12

cl (t)0 )2 2 2
! cl (t) dt = 2 sin !(t1 t0 ) :
t0
The second variation of the action functional is
2 S ( (t)) = Z t1 (t)( d2 + !2 ) (t)dt:
2 t0 dt2
Thus we can rewrite (2.8) in the form
K (t0 ; y; t1 ; x) = exp(iS ( cl ))

Z (t1 )=x i Z t1 ( (t) (t))( d2 + !2 )( (t) (t))dt:
D[ (t)] exp 2 cl
dt2 cl
(t0 )=y t0
Now let us make the variable change replacing (t) cl (t) with (t). The limits in the
path integral change to (t0 ) = (t1 ) = 0. The paths we integrate over are periodic
with the period T = t1 t0 satisfying (t0 ) = (t1 ) = 0. Using the generalization of
the Gaussian integral to functional integrals we have
Z (t1 )=0 i Z t1 (t)( d2 + !2 ) (t)dt =
D[ ] exp 2 dt2
(t0 )=0 t0
Z
D[ ] exp( 21 h ; iD iL2() ) = det( 21i D) 21 ;
Map( ;R)
where
2
D = dtd 2 !2 :
The eigenfunctions of D satisfying the condition (t0 ) = (t1 ) = 0 are the functions
sin(nt=T ), where n 2 Z>0 and T = t1 t0 . The corresponding eigenvalues are
equal to n = (n=T )2 ! 2 . We know that
1 1 1 2 2
det( 21i D) = 21i ((n=T )2 !2 ) = 21i (n=T )2 (1 !n2T2 ):
Y Y Y
n=1 n=1 n=1

Of course here we use a physicistss argument since we dont have the right to write
the product as the product of two infinite products , one of which is divergent (see
the next remark for an attempt to justify the argument). Now we use that the first
product corresponds to the action with ! = 0. So to be consistent with our previous
computation we must have
1 1 (n=T )2 1
= K (t0 ; 0; t1 ; 0) = p 1 :
Y 2
n=1 2i 2iT

25
2
d2
Note that if we compute the product using the zeta function of the operator 21i dx
on the sapce of functions (t) on [t0 ; t1 ] satisfying (t0 ) = (t1 ) = 0 we get
1
Y 1 (n=T )2 = (2iT )C:
n=1 2i
The two computations disagree. The way out of this contradiction is the choice of the
normalizing constant C p
which we used to define the Gaussian integral. It shows that
we have to choose C = 2 . Now we use the Euler infinite product expansion for the
sine function:
1
sin(!T ) = Y 2 2
!T (1 ! T ): n2 2
n=1
From this we deduce that
K (t0; y; t1 ; x) = exp(iS ( cl )) det( 21i D) 1=2 =
! iT!=2
exp(iS ( cl ))( 2i sin( )1=2 = ! 12 p e exp(iS ( cl )):
T!) 1 e 2i!T
Let us rewrite S ( cl ) in the following form
2i!T 4xye i!T =
S ( cl (t)) = i!2 (x2 + y2 ) 11 + ee 2i!T

1 e 2i!T
1
i! (x2 + y2)(1 + 2 X 1
e 2in!T ) 4xye i!T (1 + X e 2in!T =
2 n=1 n=1
1
i! (x2 + y2 ) + X
(2 x2 + 2y 2 4xye i!T )e 2in!T :
2 n=0
Using this we obtain
iT!=2
K (t0 ; y; t1 ; x) = ! 2 p e 2i!T
1
1 e
h 1
! (x2 + y2 ) + X i
exp (2 x 2 + 2y 2 4xye i!T e 2in!T =
2 n=0
1
X p
e i!(n+ 21 ))T An (x; y) = e i!T=2 !=e !(x2 +y2 )=2 + : : : :
n=1
Now recall that the kernel of a Hilbert-Schmidt unitary operator can be written in
the form
X
G(x; y) = n n (x) n (y);
n2Z
where n is the normalized eigenfunction with the eigenvalue n . In our case, the
1
eigenvalues of e iTH must be equal to e iT (n+ 2 )! . Thus the eigenvalues of the
1
Hamiltonian H are ! (n + 2 ). We shall see in the lecture that the eigenvectors of
H are n (x) = (!=)1=4 Hn (p!x)epx2 !=2 ; where
2 2
Hn are the Hermite polynomials.
When n = 0, we get 0 (x) 0 (y ) = !=e !(x +y )=2 = A0 (x; y ): This checks the
first term.
Exercises
1 2 : : : be a non-decreasing sequence of positive real numbers.
2.1 Let : 0 < P
Define (s) = n1 n s provided that this sum converges for Re (s) 0 and has
aQmeromorphic continuation to the whole complex plane with no pole at s = 0. Set
1 = e 0 (0) :
i=1 i
(i) Prove that for any N 1, Q1 Q1
i=1 i = (1 2 N ) i=N +1 i .
p
(ii) Give a meaning to the equality 1! = 2 .
2.2 Let T = Rn = be a n-torus. Here = Z!1 + : : : Z!n and !1 ; : : : ; !n are linear

independent vectors in Rn .
Pn
2 2
(i) Compute the trace of the Laplace operator L = i=1 @xi in L (T; d) where
n
d is induced by the standard volume form on R .
(ii) Compute the Euclidian partition function ZR E (T ) for the maps from = R=2R
to T with the action defined by S ( ) = 21 jj ~ 0 (t)jj2 dt, where ~ is any lift of
to a smooth map R ! Rn .
(iii) Let = fy 2 Rn : y x 2 Z for all x 2 g. Use the Poisson summation
formula
X
f (x) = A 1 X f^(y );
x2 y2
where f 2 S (Rn ), f^ is its Fourier transform and A = det[!1 ; : : : ; !n ], to relate
the partition functions Z E (T ) and Z E (T ).
2.3 Compute the terms A1 (x; y ) and A2 (x; y ) from Example 2.4 to find the eigenfunc-
tions 1 (x) and 2 (x).
Lecture 3
Quantum mechanics
The quantum mechanics is a 0 + 1 dimension QFT.

Let us recall the main postulates of quantum mechanics. A quantum state of a
system is a line in a separable Hilbert space H. It can be represented (not uniquely) by
a vector of norm 1. To each observable quantity (like position, momentum or energy)
one associates a self-adjoint operator A in H (an observable). A measurement of an
observable A depends on the given state and is not given precisely but instead there
is a probability that the value belongs to a subset ( 1; ]. This probability is equal
to pA (; ) = jjPA () jj2 , where PA () is the spectral function of A, an operator-
valued measure on R. In the case when A is a compact operator, H has anP orthonormal
basis (en ) of eigenvectors of A with eigenvalue n . Then PA () = n Pen ,
where Pen is the orthogonal projector operator to the subspace C en . Thus, for any
simple eigenvalue n of A, jh ; en ij2 can be interpreted as the probablility that the
observable A takes value n in the state .
In physics literature one often writes ji for a norm 1 eigenvector of A with
eigenvalue and rewrites h ; i in the form h ji. Also one writes hji instead of
hhjjjii.
The probability amplitude (a complex number of absolute value 1) is defined to
be h ; i. The function ! h ; i is called the wave function of the state with
respect to A. The inner product of two states h; i is interpreted as the probability
amplitude that the state changes to the state . Its absolute value is the probablity
of this event. Note that by Cauchy-Schwarz inequality, this number is always less or
equal to 1 and it is equal to 1 if and only if the two states are equal (as lines in the
Hilbert space).
The expectation value of A in the state is defined to be
Z 1
hAi = hA ; i = xdA; ; (3.1)
1
where dA; is the measure on R defined by dA; (E ) = hPA (E ) ; i.
Example 3.1. Let H = L2 (R; dx) and A = Q is the position operator corresponding
to the measurement of the coordinate x. It is defined by Q(f ) = xf . This is an un-
bounded self-adjoint linear operator. We know from Lecture 1 that ji = (although
27
28 LECTURE 3. QUANTUM MECHANICS
they do not belong to the space H but rather to the space of distributions). We have
Z
h (x); i = (x) dx = ():
The probability amplitude of the value of the observable Q in a state is equal to
(). So, in the realization of H as L2 (R; dx), a state (x) is interpreted as the wave
function of the state with respect to the observable Q. Any 2 H can be written as
Z
= () d:
Of course this has to be understood as the equality of distributions. For any test function
we have
Z1 Z1 Z1 Z
1
() d (x)dx = () (x)dx d =
1 1 1 1
Z1
()()d = ():
1
The expectation value of Q is equal to
Z
hQi = j ()j2 d:
R
Consider the delta-function a as a state (although it does not belong to L2 (R; dx)).
Then the probability of Q to take a value b in the state a is equal to ha ; b i. The inner
product is of course not defined but we can give it the following meaning. We know
2
that a is equal to the limit of tempered distributions p21t e (x a) =2t when t tends to
zero. Thus we can set
ha ; b i := tlim p1
!0 b ( 2t e
(x a)2 =2t ) = lim p 1 e (b a)2 =2t :
t!0 2t
When b = x is a variable, we get
ha ; x i = tlim p1 e (x a)2 =2t =

a = (x a):
!0 2t
Example 3.2. Let H = L2 (R=2R) and A equal to the momentum operator P = i dx d.
1
Its eigenvectors are the functions fn = p2R e inx=R with eigenvalue n. We have
Z 2R
h (x); fn i = p 1 (x)e inx=R dx = an ;
2R 0
29
is the n-th Fourier coefficient of . So, the wave function of is the function n ! a
n
on Z = Sp(P ). The probability that P takes value n at the state is equal to jan j2 .
The expectation value is equal to
X
hP i = njan j2 :
n2Z
The dynamics of a quantum system is defined by a choice of a self-adjoint opera-
tor H , called the Hamiltonian operator. In Schrodingers picture the operators do not
change with time, but the states evolve according to the law
i
~ Ht
t=e :
Here ~ is a fixed constant, the Planck constant. Equivalently, (t) is a solution of the
Schrodinger equation:
i~ dtd t = H (t):
In Heisenbergs picture, the states do not change with time but the observables evolve
according to the law
A(t) = e ~i Ht A e ~i Ht :
We have the Hamiltonian equation:
d
dt A(t) = [A(t); H ]~ ; (3.2)
where
[A; B ]~ = ~i (A B B A):
i
If t is an eigenvector of H , then t = e ~ t and hence the corresponding state
(equal to the line spanned by ) does not change with time, i.e. (x) describes a
stationary state. Usually one measures observables at the stationary states of H .
There are two ways to define a quantum mechanical system. One (due to Feynman)
uses the path integral approach. Here we take H = L2 (M; d) as in Lecture 1 and
define the Hamiltonian by means of the path integral. The choice here is the action
functional. It is defined in such a way that its stationary paths describe the motions
of a classical mechanical system. Another approach is via quantization of a classical
mechanical system. Recall that the latter is defined by a Lagrangian L : TM ! R
which, in its turn, defines an action functional on the space F = Map([a; b]; M ):
Z b
S ( ) = L( (t); _ (t))dt:
a
A critical point of this functional defines a motion of the mechanical system. The
equations for a critical point are called the Euler-Lagrange equations. If one chooses
local coordinates q = (q1 ; : : : ; qn ) in M and the corresponding local coordinates

(q; q_) = (q1 ; : : : ; qn ; q_1 ; : : : ; q_n ) in TM (so that q_i ( @q@j ) = ij ), the equations look as
@L d @L
@qi dt @ q_i = 0; i = 1; : : : ; n: (3.3)
Here the left-hand side is evaluated at a path t ! ( (t); _ (t)) in TM given by qi =

qi (t); q_i = dqdti (t) . For example, if we assume that the restriction of L to each tangent
space TMx is a positive-definite quadratic form, we can use L to define a Riemannian
metric g on M . A critical path becomes a geodesic.
Another way to define the classical mechanics is via a Hamiltonian function which
is a function on the cotangent bundle T M .
Recall that any non-degenerate quadratic form Q on a vector space V defines a
quadratic form Q 1 on the dual space V . If we view a quadratic form as a symmetric
bilinear form, and hence as a linear map V ! V , then Q 1 is the inverse map. Let us
see that, for any 2 V , Q 1 () is equal to the maximum (if Q > 0) or the minimum
(if Q < 0) of the function F : V ! R defined by
F (v) = (v) Q(v):
If we choose the coordinates so that V = V = Rn ; and (v ) = v , then Q(v ) =
1 v A v for some symmetric matrix A. Thus to find an extremum we must have
2
rF = A v = 0, hence v = A 1 and we get
max F (v) = A 1 12 A 1 = 12 A 1 = Q 1 ():
v2V
Using this one can generalize the construction of Q 1 for any function f on V such
that its second differential is non-degenerate. This is called the Legendre transform of
f . By definition, for any 2 V ,
Leg(f )() = (v ) f (v);
where v is the implicit function of defined by = dfv , where dfv : V ! R is
the differential of f at the point v 2 V . In order that this function be defined we
have to satisfy the conditions of the Implicit Function Theorem: det(d2 f (v )) 6= 0. In
general, the implicit function v () is a multivalued function, so the Legendre transform
is defined only locally in a neighborhood of an extremum point of the function (v )
f (v).
We shall apply the Legendre transform to the Lagrangian function L. We denote
the local coordinates in the cotangent bundle T M by
(q; p) = (q1 ; : : : ; qn ; p1 ; : : : ; pn );
where the fibre coordinates (p1 ; : : : ; pn ) are taken to be the dual of the coordinate
functions (q_1 ; : : : ; q_n ) in the tangent bundle TM and can be identified with a basis
( @q@1 ; : : : ; @q@n ) in TMx. The Legendre transform of L is equal to
n
X
H (q; p) = q_i pi L(q; q_);
i=1
31
where q_i are the implicit functions of (p1 ; : : : ; pn ) defined by the equation
pi = @@L
q_ (q; q_); i = 1; : : : ; n
i
The function H : T M ! R is called the Hamiltonian associated to the Lagrangian L.
As we have explained before, in order it is defined the Lagrangian must satisfy some
conditions.
Using the Hamiltonian one can rewrite the Euler-Lagrange equation for a critical
path (t) of the action defined by the Lagrangian in the form:
p_i := dp
dt
i = @H (q; p) ; q_ := dqi = @H (q; p) :
@qi i dt @pi
Here a solution is a path ~ : I ! T M; t ! ~ (t); which satisfies the above equations
after we compose it with the coordinate functions (q; p) : T M ! R. The projection
of the path ~ to the base M (i.e. the composition with the projection map T M ! M )
is the path (t) describing the equation of the motion. The difference between the
Euler-Lagrange equations and Hamiltons equations is the following. The first equation
is a second order ordinary differential equation on TM and the second one is a first
order ODE on T M which has a nice interpretation in terms of vector fields.
Recall that a (smooth) vector field on a smooth manifold X is a (smooth) section
of its tangent bundle TX: Let C 1 (TX ) denotes the set of vector fields. It has an
obvious structure of a vector space. For each smooth function 2 C 1 (X ) one can
differentiate along 2 C 1 (TX ) by the formula
D ()(x) =
X @ ;
@x i
i i
where x 2 X; (x1 ; : : : ; xn ) are local coordinates in a neighborhood of x, and i are
the coordinates of (x) 2 TXx with respect to the basis ( @x @ ; : : : ; @ ) of TXx. We
1 @xn
also have
D () = d( );
where we consider smooth 1-forms as linear functions on vector fields. This defines a
linear map
D : C 1 (TX ) ! End(C 1 (X )):

It is easy to check that D ( ) = D () + D ( ), so that the image of D lies in the
subspace of derivations of the algebra C 1 (X ). Given a smooth map : [a; b] ! X ,
and a vector field we say that satisfies the differential equation defined by (or is
an integral curve of ) if
d := (d ) ( @ ) = ( ()) for all 2 (a; b):
dt @t
The vector field on the right-hand-side of Hamiltons equations has a nice interpre-
tation in terms of the canonical symplectic structure on the manifold X = T M:
Recall that a symplectic form on a smooth manifold X is a smooth closed 2-form

! 2
2 (X ) which is a non-degenerate bilinear form on each T (X )x. If we view !x as
a linear map TXx ! (TXx) = T Xx , then its inverse defines a linear isomorphism
{! (x) : T Xx ! TXx. Varying x we get an isomorphism of vector bundles TM !
T M and by the pull-back of sections an isomorphism of the space of sections
{! :
1 (X ) = C 1 (T X ) ! C 1 (TX ):
Given a smooth function F : X ! R, its differential dF is a 1-form on X , i.e., a
section of the cotangent bundle T (X ). Thus, applying {! we can define a section
{! (dF ) of the tangent bundle, i.e., a vector field. It is called the Hamiltonian vector
field defined by the function F . We apply this to the situation when X = T (M ) with
coordinates (q; p) and F is the Hamiltonian function H (q; p). We use the symplectic
form given in local coordinates by
X
!= dqi ^ dpi :
i
For any v; w 2 T (X )x
X X
!x (v; w) = dqi (v)dpi (w) dqi (w)dpi (v):
i i
In particular,
!x( @q@ ; @q@ ) = !x( @p@ ; @p@ ) = 0 for all i; j;

i j i j
!x ( @q@ ; @p@ ) = !x( @p@ ; @q@ ) = ij :

i j i j
This shows that {! (dpi ) = @q@ i ; {! (dqi ) = @p @ ; hence
i
X @H X @H X @H @ ) + X @H @ :
{! (dH ) = {! dq i + dp i = ( )(
i @qi i @pi i @qi @pi i @pi @qi
So we see that the ODE corresponding to the Hamiltonian vector field (H ) defined
by H is the vector from the right-hand-side of Hamiltons equations. We have
(H )(f ) = ff; H g:
Let (X; ! ) be a symplectic manifold. For any two functions f; g 2 C 1 (X ) one

defines the Poisson bracket
ff; gg = !({! (df ); {! (dg)):

By definition of {! we have ! ({! (df ); ) = df ( ), so that
ff; gg = df ({! (dg)) = {! (dg)(f ) = {! (df )(g):

33
The Poisson bracket defines a structure of Lie algebra on C 1 (X ) satisfying the addi-
tional property :
ff; ghg = ff; ggh + ff; hgg:

One can show (Darbouxs Theorem) that it is always possible to choose local coor-
dinates (q1 ; : : : ; qn ; p1 ; : : : ; pn ) such that
X
!= dqi ^ dpi :
i
In these coordinates
ff; gg =
X @f @g @f @g
i
( @q @p
i i @pi @qi ):
For example,
fpi ; pj g = fqi ; qj g = 0; fqi ; pj g = ij : (3.4)
The corresponding differential equation (= dynamical system, flow) is

df = ff; H g:
dt
A solution of this equation is a path : [a; b] ! M such that
df ( (t)) = ff; H g( (t)):
dt
This is called the Hamiltonian dynamical system on X (with respect to the Hamiltonian
function H ). If we take f to be coordinate functions qi ; pi on X = T (M ), we obtain
Hamiltons equations for the critical path : [a; b] ! M in M .
The flow g t of the vector field f ! ff; H g is a one-parameter group of operators
Ut on O(M ) defined by the formula
Ut (f ) = ft (x) = f ( (t))
where (t) is the integral curve of the Hamiltonian vector field f ! ff; H g with the
initial condition (0) = x, The equation for the Hamiltonian dynamical system defined
by H is
dft = ff ; H g:
dt t (3.5)
Here we use the Poisson bracket defined by the symplectic form of M .

A quantization of a mechanical system is defined by assigning to any observable
f 2 C 1 (X ) a self-adjoint operator Af 2 H. This operator may contain a parameter
~. This assignment must satisfy some natural properties. For example:
Aff;gg = ~lim [A ; A ]
!0 f g ~
(3.6)
Under the quantization the Hamiltonian function of the mechanical system becomes a
self-adjoint operator H , called the Hamiltonian operator of the quantized system. We
have
A(H )f = Aff;H g = ~lim [A ; A ] = lim [A ; H ] :
!0 f H ~ ~!0 f ~
Thus the linear map A ! [A; H ]~ is interpreted as the quantized action of the Hamil-
tonian vector field (H ). The analog of the dynamical system (3.5) is the Hamiltonian
equation in quantum mechanics (3.2).
For example, when a mechanical system is given on the configuration space T (Rn )
with coordinate functions qi ; pi we need to assign some operators to the coordinate
functions:
Qi = Aqi ; Pi = Api ; i = 1; : : : ; n:
By analogy with (3.4), we should have
[Pi ; Pj ]~ = [Qi ; Qj ]~ = 0; [Qi ; Pj ]~ = ij (3.7)
So we have to find an appropriate Hilbert space V and operators Pi ; Qi 2 H(V ) satis-
fying (3.7). We take V = L2 (Rn ) and define
Qi : ! qi ; Pi : ! i~ @q@ :
i
Recall that these are unbounded self-adjoint operators.
The operator Qi (resp. Pi ) is called the position (resp. momentum) operator .
Let us give an example of quantization of a classical mechanical system given by a
harmonic oscillator. It is given by the Lagrangian
L(q; q_) = 12 (mq_2 m!2 q2 ));

where m is the mass and ! is the frequency. The corresponding Hamiltonian function
is
2
H (q; p) = pq_ 12 (mq_2 m!2 q2 ) = 12 ( pm + m!2 q2 );
where we used that p = @L @ q_ = mq_ to express q_ via p. The function H (p; q ) can be
viewed as the total energy of the system.
The corresponding Newton equation is
d2 x = !2x:
dt2
So the motion does not depend on the mass but the total energy does. In the sequel we
shall assume for simplicity that m = 1. The Hamiltonian operator H can be written in
the form
2
H = 12 P 2 + !2 Q2 = aay !2~ = ay a + ~2! ; (3.8)
35
where
a = p1 (!Q + iP ); ay = p1 (!Q iP ) (3.9)

2 2
are the annihilation and the creation operators. We shall see shortly the reason for
these names. They are obviously adjoint to each other. Using the commutator relation
[Q; P ] = i~, we obtain
[a; a] = [ay ; ay ] = 0; [a; ay ] = ~!; (3.10a)
[H; a] = ~!a; [H; ay ] = ~!ay : (3.10b)
This shows that the operators 1; H; a; ay form a Lie algebra H, called the extended
Heisenberg algebra.
So we are interested in the representation of the Lie algebra H in L2 (R).
Suppose we have an eigenvector of H with eigenvalue and norm 1. Since ay is
adjoint to a, we have
jj jj2 = h ; H i = h ; ay a i + h ; ~2! i = jja jj2 + ~2! jj jj2 :

This implies that all eigenvalues are real and satisfy the inequality
~! : 2 (3.11)
The equality holds if and only if a = 0. Clearly any vector annihilated by a is an

eigenvector of H with minimal possible absolute value of its eigenvalue. A vector of
norm one with such a property is called a vacuum vector.
Denote a vacuum vector by j0i. Because of the relation [H; a] = ~!a, we have
Ha = aH ~!a = ( ~! )a :
This shows that a is a new eigenvector with eigenvalue ~! . Since eigenvalues
are bounded from below, we get that an+1 = 0 for some n 0. Thus a(an ) = 0
and an is a vacuum vector. Thus we see that the existence of one eigenvalue of H is
equivalent to the existence of a vacuum vector.
Now if we start applying (ay )n to the vacuum vector j0i, we get, as above, eigen-
vectors with eigenvalue ~2 + n~! . So we are getting a countable set of eigenvectors
n = ayn j0i
with eigenvalues n = 2n2+1 ~!: It is easy to see, using induction on n that that
jj n jj2 n
= n!(~!) : After renormalization we obtain a countable set of orthonormal
eigenvectors
jni = p 1 (ay )n j0i; n = 0; 1; 2; : : : : (3.12)

(~!)n n!
One can show that the closure of the subspace of L2 (R) spanned by the vectors jni is
an irreducible representation of the Lie algebra H.
The existence of a vacuum vector is proved by a direct computation. We solve the
differential equation
p
2a = (!Q iP ) = !q + ~ ddq = 0
and get
j0i = ( !~ ) 41 e !q2 =~ : (3.13)
In fact, we can find all the eigenvectors

p
jni = ~! 1=4 Hn (p!q= ~);

where
Hn (x) = p 1n d n e
x dx x22
2 n!
is a Hermite polynomial of degree n. It is known also that the orthonormal system of
x2
functions Hn (x)e 2 is complete, i.e., forms an orthonormal basis in the Hilbert space
L2 (R). Thus we constructed an irreducible representation of H with unique vacuum
vector j0i: The vectors jni are all orthonormal eigenvectors of H with eigenvalues
(n + 12 )~!.
The function (3.13) gives the probability amplitude that a particle occupies the
position x on the real line in the vacuum state of the system.
According to Example 3.1, the value of the function jni at is equal to the proba-
bility that the observable Q takes value at the state jni.
Finally let us compute the partition function of the Hamiltonian H . The eigenvalues
of H are n = (n + 21 )~! and their multiplicities are equal to 1. So
1 1
eit(n+ 12 )~! = 1 q q ;
X 2
Tr(eitH ) = (3.14)
n=0
where q = eit~! :
Exercises
3.1 Consider the quantum mechanical system defined by the harmonic oscillator. Find
the wave function of the moment operator P at a state jni.
3.2 Consider the Lagrangian L(q; q_) = 21 (mq_2 U (q )) on T (R), where U (q ) = 0
for q 2 (0; a) and U (q ) = 1 otherwise. Quantize this mechanical system, solve the
Schrodinger equation and find the stationary states of the Hamiltonian operator.
37
3.3 Compute the Legendre transform of the function f (x) = ex .

3.4 Let (A) = h(A hAi id)2 i be the expectation value of the operator (A
hAi id)2 (the dispersion of an observable A at the state ). Prove the Heisenbergs
Uncertainty Principle
(A) (B ) 21 jh[A; B ]i j:

Lecture 4
The Dirichlet action
Now we shall move to QFT of dimension larger than 1, i.e. dim > 1, for example,
= T N where T R or S 1 and N is a manifold of positive dimension which we
shall assume for simplicity to be orientable. A map : ! M is given by a function
(t; x); t 2 T; x 2 N . Note that when N is 0-dimensional, say N = f1; : : : ; ng, we
can view (t; x) as a vector function (1 (t); : : : ; n (t)) : T ! M n and get the quan-
tum mechanics on M n . For example, if our QFT is a harmonic oscillator, passing from
R ! R f1; : : : ; ng corresponds to considering n harmonic oscillators. Replacing
f1; : : : ; ng by positive-dimensional N means that we consider the whole manifold of
harmonic oscillators!
Recall that any QFT is defined by an action functional S on the space of paths. In
a one-dimensional theory we defined S by a Lagrangian L : TM ! R. The pull-back
of L under the map (; d) : T ! TM is a function F () on T , so for any density
d on (i.e. a section of top (T )) we can multiply F ( )d to get a density on
which we can integrate. If dim > 1 this is not true anymore since F () is a function
on T and a density is a function on dim (T ). So the definition of the Lagrangian
has to be changed. We are not going into a rigorous mathematical discussion of this
definition referring to Deligne-Freeds lectures at the IAS.
Recall that the jet bundle of order k of a fiber bundle E over a manifold X is a
vector bundle J k (E ) whose local sections are local sections of E together with their
partial derivatives up to order k . Let (e1 ; : : : ; er ) be a local frame of E and x1 ; : : : ; xn
local coordinates on X . A local frame of J k (E ) is a set (e ; ei1 :::is ) where 0 <
i1 + : : : + ik k; 1 i1 : : : ik n. Let (y ; yi1 :::is ) be the corresponding
coordinate functions. Any local section (x) = y e of E can be uniquely extended
to a section ~ of J k (E ) such that
@ i1 +:::+is (x):
yi1 :::is (~(x)) = @x
i1 : : : @x
is
Let F be the space of sections of J k (E ) (fields, and their partial derivatives). Roughly
speaking a Lagrangian of order k is a smooth map from J k (E ) to the space of densities
on . We will be usually dealing with Lagrangians of the first order. Then we can write
39
40 LECTURE 4. THE DIRICHLET ACTION
a Lagrangian as
L = L(xi ; y ; yi )jdn xj:

So, the action will be defined by a formula
Z
S () = L(x; ; @xi ):

One can generalize the Euler-Lagrange equations to the higher-dimensional case:
@L @L
@y @i @ (yi ) = 0; = 1; : : : ; dim M: (4.1)
Here we assume that the equality takes place when we evaluate the left-hand side on .
We also assume here that L is of the first order.
Let us consider an example, which will be very much relevant to the string theory.
First, a little of linear algebra. Let V; W be two vector spaces equipped with non-
degenerate bilinear forms h and g , respectively. We can define a symmetric bilinear
form on the space of linear maps Lin(V; W ) by
hf; i = Tr(f );

where f : W ! V is the adjoint map with respect to the bilinear forms h and g (i.e.
g(f (v); w) = h(v; f (w)) for any v 2 V , w 2 W ). Let us explain this definition.
Choose a basis 1 ; : : : ; n in V and a basis 1 ; : : : ; m in W . Let H be the matrix of
h in the first basis and G be the matrix of g in the second basis. Let A be the matrix of
f with respect to the bases, and B is the same for . Then the matrix of f is equal to
A = H 1 At G so
hf; i = Tr(H 1 At GB ) = hij asi btj gst ; (4.2)
where A = (aji ); B = (bji ); H 1 = (hij ); G = (gij ) and we employ the physics

summation notation. Assume that f is the map defined by f (v ) = i (v )j , where i
is an element of the dual basis ( 1 ; : : : ; n ). Then am
n0 = mi nj . Similarly, take
0
defined by (v ) = i (v )j0 . Then we get hf; i = hii gjj 0 . If we identify Lin(V; W )
0
with V
W , we get f = i
j ; = i
j 0 and
h i
j ; i0
j0 i = hii0 gjj0 :
From this we deduce that the matrix of the bilinear form on V
W with respect to
the basis ( i
j ) is equal to the Kronecker product of the matrices H 1 and G. The
matrix H 1 defines an inner product on V . So, our inner product on V
W could
be taken as the definition of the tensor product of the inner product on V and on W .
Now we are ready to globalize. Let h be a metric on and g be a metric on M .
Define the Lagrangian
L () = jdj2 dX = Tr(d d)dX : (4.3)

41
Here dX is the volume form defined by the metric h and the adjoint d of d is
defined with respect to the metrics h and g .
The corresponding action
Z
S () = jdj2 d (4.4)

is called the Dirichlet action.
If h is given in local coordinates x1 ; : : : ; xd by the matrix (h ) and g is given
in local coordinates y 1 ; : : : ; y D by the matrix (g ) then (4.4) can be rewritten in the
form
Z
S () = @ @ g dx ^ : : : ^ dx :
j det(h)j 21 h @x
@x 1

d (4.5)
This follows from (4.2) and the fact that dX = j det(h)j 2 dx1 ^ : : : ^ dxd . Recall that
1
the volume form on a vector space V is equal to e1 ^ : : : ^ en , where e1 ; : : : ; en is an

orthonormal basis.
Observe the following properties of the Dirichlet action:
(A1) (isometry invariance) For any diffeomorphism : ! 0 preserving the met-

rics,
S ( ) = S0 ();
(A2) (locality) if is glued together from 1 and 2 along their boundaries, then
a
S () = S1 ` 2 (j1 2 )
(A3) (conformal scaling) if h0 = ec(x) h is a new metric on , and S 0 is the new

action, then the action functional does not change if and only if dim = 2:
Proof. Let : V = T ! W be a linear map of inner product spaces and : V 0 !V

be an isometry of inner product spaces. Then, for any v 0 2 V 0 ; w 2 W ,
h((v0 )); wiW = h(v0 ); (w)iV = hv0 ; 1 ( (w))iV 0 :

This shows that ( ) = 1 , hence
Tr(( ) ( )) = Tr( 1 ( ) ) = Tr( ):
Applying this to the case when ; are the maps of the tangent spaces, this implies
that (jjdjj2 d0 ) = jjd( )jj2 d : Property (A1) now follows from the stan-
dard properties of integration of differential forms. Property (A2) is obvious from the
definition of the action. Property (A3) follows easily from formula (4.5).
Example 4.1. Let = R with a local coordinate t and M = Rn with metric defined
by a matrix gij (x) in the canonical basis in TMx = Rn . Define a metric on by
jj1jj2a = h(a). Let f : R ! Rn be defined by the vector function (f 1 (t); : : : ; f n (t)).

Then
0 0 0
jjdfa jj2 = f (a)G(hf(a(a)))f (a) = jjfjj1(ajj)2jj :
2
a
So, if we take h 1; G = 21 ( ), we obtain the action S ( ) 1R 2
= 2 _ (t) dt which
we used in the previous lectures.
If dim > 1, I do not know any geometric meaning of the Dirichlet action. How-
ever, let us see that for a fixed metric g on M one can always choose a metric h on
such that the action acquires a very nice meaning. In fact, the metric h is chosen
to minimize the action. Let us consider the Dirichlet action as a function of h and
compute its variation in the direction h at h = h0

S = (S (h + h) S (h ))=;
h h=h0 0 0
where 2 = 0. Note that, for any invertible matrix A and any square matrix B of the
same size, we have
jA + B j = jAj + Tr(A 1 B )jAj:

Thus
jA + B j 12 = jAj 12 (1 + 2 Tr(A 1 B )):

Also
(A + B ) 1 = A 1 A 1 BA 1 :
Let
@ @ g
= @x @x
and = ( ). The matrix is the matrix of the metric (g ). It can be viewed as the
metric on the image of under the map (called the world-sheet). Then
Z
S ( ) = jhj 12 Tr(h 1 )d (4.6)

and

S

h () =
h=h0
Z Z
[jh0 + h0 j 21 Tr(h0 + h) 1 )d jh0 j 12 Tr(h0 1 )d =

43
Z
= jh0 j 12 ( 21 Tr(h0 1 h)Tr(h0 1 ) Tr(h0 1 hh0 1 ))d :

Set (h 1 ) = h0 1hh0 1 . Then
Z
S j h j 1 1 1 1 1

Tr(h0 1 hh0 1 )
( ) = 2 Tr(h0 (h0 hh0 ))Tr(h0 ) d =
0 2
h h=h0
Z
jh0 j 21 Tr (h 1 )( 21 h0 Tr(h0 1 ) d :

Since this must be zero for all possible h, this implies that a critical metric h0 satisfies
21 h0 Tr(h0 1 ) = 0: (4.7)
This implies
j j1=D = 21 Tr(h0 1 )jhj1=D ;

where D = dim . Plugging this in formula(4.6) we get
Z
S (; h0 ) = 2 j j1=D jhj 22DD d :

We get a wonderful fact: if D = 2, and the metric on is chosen to be critical, the
action has a simple geometric meaning. It is equal to the twice the area of the world
sheet () in the metric induced from the metric of M .
In physics the latter action is called the Nambu-Goto action and the Dirichlet action
is called the Brink-DiVecchia-Howe-Desse-Zumino action, or the Polyakov action for
short.
Remark 4.1. In the case when the metric on is Lorentzian, we have to replace jhj
with j hj. Also physicists use the metric to lower the indices. If (gijP ) is the
P matrix of
basis (e1 ; : : : ; en ) then for any vector ai ei the vector j ( i ai gij )ej
a metric g in aP
is denoted by i ai ei . In this notation formula (4.5) can be rewritten as
Z @
S () = jhj 21 hij @
@x @x d :
i j
Remark 4.2. The tensor
T = (Tij )() = ( ij 21 hij h )dxi dxj :

is called the energy-momentum tensor. By (4.7) this tensor is equal to zero if and only
if h is a critical metric for the action S (; h). Observe that
hij Tij = jhj 12 (hij ij 21 hij hij h )dxi ^ dxj = jhj 21 (1 D2 )hij ij dxi dxj :
In particular, T is trace-less if D = 2.
Assume = RD ; M = Rn and the metrics h; g are the standard Euclidean metrics.
Then
T = ( ij 21 Tr( ))dxi dxj :
Let us write down the Euler-Lagrange equations for the Dirichlet action in the case
when the metrics h and g are flat (i.e. h and g are constant functions). We get the
equations
2
h g @x@ @x = 0; = 1; : : : ; D: (4.8)

In the case when = Rn ; M = R, and h this is just the Laplace equation with
respect to the metric h. When h = , its solutions are harmonic functions. In
general, the Euler-Lagrange equation for the Lagrangian (4.3) can be written in an
invariant form:
Tr(Dd) = 0;
where D is the covariant derivative of a section d of E = TM

TX with respect
to the natural Riemannian connection defined on the bundle E .
Consider the special case when = R2 , with coordinates (x1 ; x2 ) = (t; x), and
h = (h ) = diag(1; 1). Take M = R. Then the Lagrangian density becomes
L = [( @ 2 @ 2
@t ) ( @x ) ]dtdx:
The Euler-Lagrange equation is
@2 @ 2 = 0:
@t2 @x2
Notice the analogy with the Lagrangian for a harmonic oscillator (with m = ! = 1)
2
L = ( dx 2
dt2 x )dt:
We can view (t; x) as the displacement of the particle located at position x at time t.
The Euler-Lagrange equation for the scalar field (t; x) can be thought as the motion
equation for infinitely many harmonic oscillators arranged at each point of the straight
line.
If h is the flat Lorentzian metric in Rn defined by the diagonal matrix diag[ 1; 1; : : : ; 1]
the Euler-Lagrange equation for a scalar field with Dirichlet action is
@ 2 Xn @2
= (@ @ ) = @t2 =2 @x2 = 0:
45
The operator is called the DAlembertian operator or relativistic Laplacian.

A little more general, if we take the Lagrangian
L = 12 (@ @ m2 2 );
the Euler-Lagrange equation is the Klein-Gordon equation
+ m2 = 0:
In many quantum field theories M is a fibre bundle over and : ! M is

a section. When M is a G-bundle with some structure group G a map is called a
classical field, otherwise is called a non-linear -field. For example, when M is
the trivial vector bundle of rank 1, a classical field is called a scalar field. Of course
any map : ! M can be considered a section of a fibre bundle, the trivial bundle
M ! .
Example 4.2. An example of a classical field is a gaugeP
field or a connection on a
principal G-bundle over . It is defined by a 1-form Ai dxi on with values in the
adjoint affine bundle Ad(G). In other words, it is a section of the bundle T
Ad(G).
For example when G = GL(n), a gauge field is a map of vector bundles A : T !
End(E ); where E is a smooth vector bundle of rank r over . It satisfies
A( )(fs) = (f )s + fA( )(s);

where f is a local smooth function and s is a local section of E . It is clear that the
difference of two connections is a morphism of vector bundles and thus a connection
is a section of an affine bundle. Each connection A defines the Lie(G)-valued 2-form,
the curvature form,
FA = dA + 21 [A; A] =
X
Fij dxi ^ dxj :
Here Fij is a local function on X with values in Lie(G). We define the Lagrangian
density on AX by setting
L(A) = FA ^ FA :
Here is the star-operator on the space of differential forms with values in a vector
bundle E equipped with a metric g . It is determined by the property ^ = h; id;
where h; i is a natural bilinear form on the space of such forms (determined by the
Riemannian metric on TM and the metric on E ) and d is the volume form defined by
the metric on M .
The Euler-Lagrange equation for the gauge fields is the Yang-Mills equation:
d
( @F
X ij
i=1 @xi + [Aj ; Fij ]) = 0; j = 1; : : : ; d: (4.9)
There are two approaches to quantization in higher-dimensional QFT. First uses

functional integrals which generalize the path integrals.
Consider a space of fields F = Map(; M ) on a D-dimensional manifold . We
assume that = T 0 , where dim T = 1 ( a time factor). For each t 2 T we denote
by t the restriction of a field 2 F to t = ftg 0 . Let Ft be the space of fields
on t obtained by restrictions of fields from F . Fix two fields fi 2 Fti ; i = 1; 2:
Consider an action S : F ! R and set
Z f2
R(f1; f2 ) = eiS() D[]; (4.10)
f1
where we integrate over the space of fields on such that ti = fi . We use some
measure D[] on F .
Observe the obvious analogy with our previous definition where we take 0 =
fpointg and F is the set of maps ! M .
Now if we consider some Hilbert space Ht of functions on Ft , the integral oper-
ator with kernel (4.10) defines a linear map
Z
Tt1 t2 : Ht1 ! Ht2 ; Tt1 t2 ( )(f ) = R(f; g) (g)D[g]
Map(0 ;M )
It also defines a self-adjoint Hamiltonian operator H : H0 ! H0 such that
Tt1 t2 = e i(t2 t1 )H
which can be used to define a Hermitian map Ht1 Ht2 ! C ,
h 2 ; 1i = h 2 je i(t2 t1 )H 1 iH :
t2
The kernel Rf1 ;f2 has special meaning for f1 = f2 . The integral
Z
Tr(e i(t2 t1 )H ) = Zf;f D[f ]
Map(0 ;M )
is the trace of the operator ei(t2 t1 )H . It is called the partition function of the theory.
More generally, let Oi (Pi ); i = 1; : : : ; n; be a local quantum field at a point Pi on

(a local field is a functional on F which depends only on (P ) and derivatives of
at P ). An example of a local field is the functional ! l((P )), where l : M ! R
is a function on M . We set
Z f2 Y
Zf1 ;f2 (Oi (Pi ); ) = eiS() Oi (Pi )()D[] (4.11)
f1 i
This leads to the correlation function
hO1 (P1 ) : : : On (Pn )i = Zf1 ;fZ2 (Oi (()

Pi ); ) : (4.12)
f1 ;f2
47
We can use (4.11) to define a Hermitian form on the space H of functions on F0 =
Map(0 ; M )
Z Z
R(O(P ); t1 ; t2 )( 2 ; 1 ) = 2 (f2 )
1 (f1 )Zf1 ;f2 (O(P ); )D[f1 ]D[f2 ]:
Ft2 Ft1
(P )
This is still linear in 1 and half-linear in 2 . Thus it defines a linear operator O
such that
R(O(P ); t1 ; t2 )( 2 ; 1 ) = h 2 je i(t2 t1 )H O (P )ei(t2 t1 )H 1 iH :

We can get rid of the parameter t = t2 t1 by letting it go to infinity, i.e. define
itH O (P )eitH 1 i:
!1 R(O(P ); 2 ; 1 ) = tlim
tlim !1h 2 je
(P ) in the Hilbert space H. It is called the vertex
In this way we get a local operator O
operator associated to a functional O(P ).
Another approach to quantization generalizes the one we used for the harmonic
oscillator. Again we assume that = T 0 . For any field : ! M we denote by
@0 the partial derivative in the time variable. By analogy with classical mechanics we
introduce the conjugate momentum field
L ():
(t; x) = @
0
For example, when L() = 21 (( @
@t )
2 ( @ 2
@x ) ) we obtain = @
@t . We can also
introduce the Hamiltonian functional :
Z
1
H () = 2 0 (_ L)dx:

Then the Euler-Lagrange equation is equivalent to the Hamiltonian equations for fields
_ (t; x) = H H
(t; x) ; _ (t; x) = (t; x) ;
where the dot means the derivative with respect to the time variable. Here we consider
and as independent variables in the functional H and use the partial derivatives of
H:
To quantize the fields and we have to reinterpret them as Hermitian operators
in some Hilbert space H which satisfy the commutator relations (remembering that
is an analog of q and is an analog of q_).
[(t; x); (t; y)] = ~i (x y); (4.13)
[(t; x); (t; y)] = [(t; x); (t; y)] = 0: (4.14)

Here we have to consider ; as operator valued distributions, i.e. a continuous linear

functionals on the space of test functions on equipped with some measure with
values in the space of operators in a Hilbert space H. Any function on with values
in the space of operators in H which is integrable with respect to some operator-valued
measure dO defines a distribution
Z
! T (x)(x)dO :

The commutator of two operator valued distributions is a bilinear form on the space of
test functions:
(; ) ! [T1 (); T2 ( )]:

Thus the meaning of (4.13) is
Z
[(); ( )] = (x) (x)d:

2 2
Example 4.3. Assume that 0 = R and L is the Klein-Gordon Lagrangian @@t2 @@x2
m2 2 . A solution of the Klein-Gordon equation can be written as a Fourier integral
Z
(t; x) = p1 a(k)ei(kx !k t) + a(k)ei( kx+!k t) dk;
2 R
where
!k2 = k2 + m2 :
Similarly we have a Fourier integral for (t; x):
Z
(t; x) = p i ! ( eikx a(k)e i!k t + e ikx a(k)ei!k t )dk:
2 R k
To quantize we replace a(k ) with an operator ak and a(k ) with the adjoint operator ayk
and consider the above expansions as operator integrals. This implies that the operators
(t; x) and (t; x) are Hermitian. The commutator relations (4.13) will be satisfied if
we require the commutator relations
[ak ; ayk0 ] = (k k0 );
[ak ; ak0 ] = [ayk ; ayk0 ] = 0:

This is in complete analogy with the case of the harmonic oscillator, where we had
only one pair of operators a; ay satisfying [a; ay ] = 1; [a; a] = [ay ; ay ] = 0 (or n
operators ai ; ayi satisfying [ai ; ayj ] = i;j ; [ai ; aj ] = [ayi ; ayj ] = 0). There is a big
difference however. In our case the Heisenberg Lie algebra generated by 1; ak ; ayk is
infinite-dimensional.
49
Exercises
4.1 Let L = L(; d) be a Lagrangian on with a metric h defined on the space of
maps ! Rn . Define the energy-momentum tensor by
h @ (@L ) () h L()

X
T () =

(i) Show that this definition agrees with the one given for the Dirichlet action;
P @T () = 0 if satisfies the Euler-Lagrange equations.
(ii) Prove that @x
4.2 Consider a system of N harmonic oscillators viewed as a finite set of masses
arranged on a segment [a; b], each connected to the next one via springs of length
. Write the Lagrangian L() describing this system. Show that the limit of L() when
goes to zero is equal to
Z b
c1 ( @(@tt; x) )2 c2 ( @@x
(t; x) )2 ]dx:
a
where c1 ; c2 are some positive constants and (t; x) is the function which measures the
displacement of the particle located at position x at time t.
4.3 Let (X; g ) be a Riemannian manifold of dimension n and E be a vector bundle
over X equipped with a Riemannian metric. Show that there exists a unique linear
isomorphism : k (T X )
E ! n k (T X )
E such that, for any ; 2
(X; k (T X )
E ); one has ^ = g 1(; )vol . Here ^ is defined
locally by extending via linearity the product (a
e) ^ (b
e0 ) = (e; e0 )a ^ b.
Also g 1 is the inverse metric on T (X ) extended to k (T X )
E by the formula
g 1(a
e; b
e0 ) = b(e; e0) ^k (g 1 )(a; b):
4.4 Using the star-operator defined in the previous problem show that the Dirichlet
action can be rewriten in the form
Z
S () = d ^ d;

where d is considered as a section of the bundlle T
(TM ).
Lecture 5
Bosonic strings
From now on we stick with dimension D = 2 of our QFT. This is where strings appear.
Our manifold will be a smooth 2-manifold with a pseudo-Riemannian metric h. It
could be the plane R2 or a cylinder S 1 R, or a torus S 1 S 1 , or a sphere S 2 , or a
compact Riemann surface g of genus g > 1. Of course each time we should specify
a metric on .
We shall begin with the case when is a cylinder S 1 R (closed strings) or
= R [0; ] (an open string). We use the coordinate in the circle direction and the
coordinate t (time) in the R-direction. A map (t; ) : ! M can be considered as a
map
~(t) : R ! L(M ); t ! ( ! (t; ));

where L(M ) is the loop space of M , i.e. the space of smooth maps from a circle to M .
In the case of open strings L(M ) must be replaced with the space P (M ) of paths in M .
We shall consider only closed strings, however occasionally we state the corresponding
results for open strings.
We shall also assume in the beginning that M = Rn with the Lorentzian flat metric
g = (g ) = ( ), where = diag( 1; 1; : : : ; 1). We will write vectors in M
as (x1 ; : : : ; xn ) and denote by (x1 ; : : : ; xn ) the vector ( x1 ; x2 ; : : : ; xn ) equal to
(x g 1 ; : : : ; x gn ). Later on we will of course consider more general target spaces
M . We consider the Dirichlet action
Z Z
S (; h) = T2 jdj2 d = T2
p
jhjh @ @ d : (5.1)

Here T is a certain parameter of a string (the string tension). It is equal to 1=20 for
open strings, where 0 is a certain other constant called the Regge slope. For closed
strings T = 1=0 . We use the subscript h to emphasize the dependence of the action
on h.
S ( ; (h)) = S (; h); (5.2)
51
52 LECTURE 5. BOSONIC STRINGS
where is a diffeomorphism of . This means that the action is invariant with respect
to smooth reparametrizations of the maps. Also, for any f 2 C 1 (), we have
S (; ef h) = S (; h): (5.3)
This means that the action is conformally invariant.

It is known (see, for example, [Modern Geometry] by Dubrovin, Fomenko and
Novikov) that there exists a unique diffeomeorphism such that (h) = ef h0 , where
f is a smooth function and h0 is a flat metric given locally by the diagonal matrix
diag( 1; 1). By (5.2) and (5.3)
S (; h) = S ( ; h0 ); (5.4)
We shall fix the metric on by equipping R with the metric dx2 and taking S 1 =
R=2rZ with the metric induced by the standard metric dt2 on R. Then we have
two constraints on . One comes from the Euler-Lagrange equation for the action
Sh0 and another comes from the vanishing of the energy-momentum tensor. Since the
Lagrangian function for the action Sh0 is equal to
L = T2 (@t @t @x @x ) (5.5)
the Euler-Lagrange equation for the action Sh0 is
(@t2 @x2 ) = 0; = 1; : : : ; D: (5.6)
So our field ( ) satisfies the Klein-Gordon massless equation.

The value of the energy-momentum tensor T at h0 is equal to
T10 = T01 = @t @x = 0; T00 = T11 = 21 (@t @t + @x @x ) = 0:

(5.7)
To solve the wave equation (5.6) we introduce the light-cone coordinates
+ = t + x; = t x:
Let @+ ; @ denote the partial derivatives with respect to these coordinates. We have
@+ = 12 (@x + @t ); @ = 21 (@t @x ):
Thus we can rewrite (5.6) in the form
@+ @ = 0:
This easily implies that a general solution of (??)EQ) can be written as as sum
= L (+ ) + R ( ):

53
Using the boundary conditions, we see that, in the case of a closed string, the functions

@L @
L + R , @ and @ R are periodic with period 2r, so that we can use the Fourier
+
expansion to write
L (t; x) = 21 x + l0 + + ilr 1 e in+ =r ;

X
n
n6=0 n
(5.8a)
R (t; x) = 21 x + l~0 + ilr 1 ~ e in =r ;

X
n6=0 n n (5.8b)
where l = p2rT1 and = ~ . We shall see in a moment a reason for the choice of
0 0
the constant l. Also, since we want to be real,
n = n ; ~n = ~ n :
The field L (t; x) (resp. R (t; x) ) describes the left-moving modes (resp.
right-moving modes) of a closed string.
Note that
X
@+ L (t; x) = l n e in+ =r ; (5.9a)
n2Z
X
@ R (t; x) = l ~n e in =r : (5.9b)
n2Z
It is clear that
Z 2r
x = (0; x)dx; = 1; : : : ; D
0
and can be interpreted as the center-of-mass coordinates.
By analogy with D = 1 QFT the momentum field is defined to be
P = @ (@@ L ) = T@t :

t
The expression
Z 2r
d (0; x) dx = 2rT (2l ) = 2 :
p = T dt 0 l 0 (5.10)
0
is the total momentum coordinate of the string at t = 0. Now we can rewrite the
equations (5.11) in the form
L (t; x) = 21 x + 21 l2 p + + ilr 1 e in+ =r ;

X
n
n6=0 n
(5.11a)
R (t; x) = 12 x + 21 l2 p + ilr 1 ~ e in =r ;

X
n6=0 n n (5.11b)
Remark 5.1. If we choose the Riemannian metric on instead of pseudo-Riemannian,

we will be able to identify the cylinder = R S 1 with the punctured complex plane
C = C n f0g by means of the transformation
(t; x) ! z = e( t+ix)=r :
The Euler-Lagrange equation (3.7) gives
@z @zX (z; z) = 0

The equation of a string becomes
L (z ) = 12 x + 21 il2p ln z + ilr 1 z k ;
X
k
k6=0 k
(5.12a)
R (z ) = 12 x + 21 il2p ln z + ilr 1 ~ zk

X
k
k6=0 k
(5.12b)
The stress-tensor (T ) can be rewritten in the new coordinates too. We have

X X
Tzz = @z @z = Lmz m 2 ; Tzz = @z @z = L~ mzm 2:
m2Z m 2Z
This is a familiar expression from the conformal field theory.
The Hamiltonian of our theory is equal to
Z 2r T Z 2r
H= (@t P L)dx = 2 (@t @t + @x @x )dx: (5.13)
0 0
Observe that it vanishes on a string which satisfies the constraint that the energy-
momentum tensor vanishes. Plugging in the expressions for in terms of n ; ~n ,
we obtain
H = 12
X
( n n + ~ n ~n ) (5.14)
n2Z
Now it is clear the introduction of the constant l. It made our formulas not depend on
T . Observe that we could simplify the sum by getting rid of 12 but we dont do it, since
in a moment the coefficients n ;
~n will become operators.
Using the new coordinates we can also rewrite the constraints (5.7) in the form
@+ @+ = 21 (T00 + T01 ) = 0;
@ @ = 12 (T00 T01 ) = 0:
This immediately gives
@t (R ) @t (R ) = @t (L ) @t (L ) = 0:

55
This can be restated in terms of the Fourier coefficients as follows:

Z
Lm = T2 (@t (R ) @t (R ) e 2imx dx = 12 m n n = 0
X
(5.15a)
0 n2Z
Z
L~ m = T2 (@t (L ) @t (L ) e 2imxdx = 12 ~m n ~n = 0:
X
(5.15b)
0 n2Z
Observe that
H = L0 + L~ 0 (5.16)
Now we quantize as in the previous lecture by taking n as operators in some

Hilbert space. Since we want to be Hermitian we require
n = (n )y ; ~ n = (~n )y :
We need
[P (t; x); X (t; x0 )] = i(x x0 ) ;

Plugging in the mode expansions, we see that this is equivalent to the following com-
mutator relations
[x ; p ] = i (5.17a)

[m ; n ] = [~m ; ~n ] = mm+n;0 ; (5.17b)
all other commutators between, x ; n ; ~n are equal to zero.

We can also quantize the expressions for H; Lm ; L ~ m. In order they make sense
as operators in some Hilbert spaces we will require (by analogy with the harmonic
in our representation n kills any state provided that n is large enough.
oscillator) thatP
Thus the sum n>0 m n n makes sense. Let us set, for any operators Ai ; Aj with
indices in an ordered set I ,
(
if i j ;
: Ai Aj := Ai Aj ; (5.18)
Aj Ai otherwise.
It is called the normal order for the composition of operators. Since [m n ; n ] = 0,

we can rewrite Lm ; m 6= 0; in the form
Lm = 12
X
: m n n :
n2Z
The situation with L0 is more complicated since n n do not commute. Of course
we know that n n = n + n n so we can write
L0 = 21 0 0 + n n + 12 (D 2)
X X
n
n>0 n1
Here we get D 2 because when we sum with respect to , the contributions corre-
sponding to = 0; 1 cancel each other. We shall deal with the last sum later. Now we
define L0 and L~ 0 by dropping out the infinite sum.
Similarly we define the operators L~ m. The operators Lm; L~ m are called the Vira-
soro operators. Notice that Lm and L m (resp. L ~ m and L~ m) are adjoint of each
other.
The expression of the Hamiltonian operator is now straightforward:
1
X
H = L0 + L~ 0 + (D 2) n: (5.19)
n=1
Since the last sum obviously does not make sense, we regularize it by setting
1
X 1:
n = ( 1) = 12
n=1
So, finally we get
H = L0 + L~ 0 D12 2 ; (5.20)
From now on
L0 = 21 0 0 +
X
n n
n1
L~ 0 = 21 ~0 ~0 +
X
~ n ~n :
n1
Let us find the commutator relations between the operators Ln . First we use the
following well-known identity:
[AB; CD] = A[B; C ]D + AC [B; D] + [A; C ]DB + C [A; D]B:

This gives
[m k k ; n l l ] = m k [k ; n l ]l + m k n l [k ; l ]+
[m k ; n l ]l k + n l [m k l ]k = kk;l n m k l + kk; l m k n l +
(m k)m k;l n l k ] + [(m k)m k; l n l k :

Here we skip the upper index . This easily implies
[Lm ; Ln] = 21 (km k n+k; + (m k)m k+n k ):

X
k
57
Changing k to k n in the first sum we obtain for m + n 6= 0

[Lm ; Ln ] = (m n)Lm+n ; if m + n 6= 0:
P
For m+n = 0 we have a problem since k2Z k k is not defined. Since [ k ; k ] =
k, we see that the difference
[Lm ; L m] = 2mL0 + A(m)id
for some scalar A(m). Using the Jacobi identity, we find that, for k + n + m = 0,
(n m)A(k) + (m k)A(n) + (k n)A(m) = 0:

Setting k = 1 and m = n 1 gives
A(n + 1) = (n + 2)A(n)n (21 n + 1)A(1) :

This shows that A(m) = am3 + bm for some constants a; b. We will fix the constants
when we consider the representation of the Lie algebra generated by am ; a
~m in some
Hilbert space.
Exercises
5.1 Let Vect(S 1 ) be the Lie algebra
P of complex vector fields on the circle. Each field
d , where an 2 C . Let Ln = ein d .
is given by a convergent series n2Zan ein d d
(i) Show that [Ln ; Lm ] = i(m n)Ln+m .
(ii) Let Vir = Vect(S 1 ) C . Set
[(; a); (; b)] = ([; ]; B (; ));

where B is a bilinear form on Vect(S 1 ). Show that this defines a structure of a
Lie algebra on Vir if and only if B satisfies
B ([; ]; ) + B ([; ]; ) + B ([; ]; ) = 0:

(iii) Let B (n; m) = B (Ln ; Lm). Show that B (n; m) = 0 unless n + m = 0 and
A(n; n) = an3 + bn for some a; b 2 C .
(iv) Prove that two bilinear forms B and B 0 define isomorphic Lie algebras if and
only if B (n; n) B 0 (n; n) is a linear function in n.
t; t 1 ] be the algebra of Laurent polynomials in one variable. For any P (t) =

5.2 Let C [
P
k k 2 C [t; t ] let Res(P (t)) = a 1 . Define a bilinear form on C [t; t ] by
a tk 1 1
(P (t); Q(t)) = Res(Q dP dt ):
(i) Show that is skew-symmetric and satisfies
(PQ; R) + (QR; P ) + (RP; Q) = 0:

(ii) Let D= d be the Lie algebra of derivations of
Der(C [t; t 1 ]) = C [t; t 1 ] dt
C [t; t 1 ]. Show that D C is a Lie algebra with respect to the Lie bracket
[(P dP dP dP dP
dt ; a); (Q dt ; b)] = ([P dt ; Q dt ]; (P; Q)):
(iii) Let Lm = ( tm+1 dtd ; 0); c = (0; 1). Show that
[Lm; Ln ] = (m n)Lm+n + 12 1 (m3 m)
m; n c:
(iv) Show that any central extension of the Lie algebra Vect(S 1 ) with one-dimensional
center is isomorphic to the Lie algebra defined by the commutator relations as in
(iii).
Lecture 6
Fock space
Let g be a Lie algebra over a field F with the Lie bracket [a; b]. Recall the construction
of the envelopping algebra U(g). It is an associative algebra over F which is universal
with respect to homomorphisms f : g ! A of associative algebras such that f ([a; b]) =
f (a)f (b) f (b)f (a). It is constructed as the quotient of the tensor algebra
U(g) = T (g)=I = (1 n
n=0 T (g))=I;
where I is the ideal generated by elements a
b b
a [a; b], where a; b 2 g.
For example, if g is a commutative Lie algebra (i.e. [a; b] = 0 for all a; b 2 g) U(g) is
isomorphic to the symmetric algebra Sym(g), i.e. a free commutative algebra generated
by the vector space g ( isomorphic to the polynomial algebra in variables indexed by a
basis of g). In general, U(g) has a basis consisting of ordered products ej1 : : : ejk ; j1
: : : jk ; where (ei )i2I is an ordered basis of the vector space g.
Recall that a linear representation of g in a vector space V is a homomorphism of
the Lie algebras : g ! End(V ), where the latter is equipped with a structure of a Lie
algebra by setting [A; B ] = A B B A. We say that V is a g-module. By definition
of the envelopping algebra, this is equivalent to equipping V with a structure of a left
module over U(g). This allows us to extend the terminology of the theory of modules
over associative rings to modules over Lie algebras.
An example of a linear representation is the adjoint representation adg : g !
End(g) defined by adg (a)(x) = [x; a]. The fact that it is a linear representation follows
from the Jacobi identity [x; [y; z ]] + [y; [z; x]] + [z; [x; y ]] = 0.
An ideal l in a Lie algebra g is a linear subspace such that [a; b] 2 l for any a 2
g and any b 2 l, or, equivalently, a submodule in the adjoiont representation. An
example of an ideal in g is the commutator ideal [g; g] generated by the commutators
[a; b]; a; b 2 g. A noncommutative Lie algebra without non-trivial ideals is called a
simple Lie algebra.
We will be mostly dealing with infinite-dimensional Lie algebras. An example
of such an algebra is a Heisenberg algebra. It is characterised by the condition that
its center (the set of elements commuting with all elements in the algebra) is one-
dimensional and coincides with the commutator. Let l be a Heisenberg algebra and let
59
60 LECTURE 6. FOCK SPACE
z be a basis of its center l0 . We define a bilinear alternating form (; ) on l by

[a; b] = (a; b)z:
Its kernel is equal to l0 , and the induced bilinear form on l = l=l0 is nondegenerate. For
example, if l is finite-dimensional, dimF l = 2k and l has a basis (e1 ; : : : ; ek ; e 1 ; : : : ; e k )
such that
(ei ; e j ) = ij ; (ei ; ej ) = (e i ; e j ) = 0; 1 i; j k:

Thus l is completely determined by the commutator relations
[ei ; e j ] = i;j z: (6.1)
So all Heisenberg Lie algebras of the same dimension are isomorphic.

If l is infinite-dimensional, we assume additionally that l is Z-graded, i.e.
l = n2Zln ;
where each linear subspace ln is finite-dimensional, and
l0 = l0 ; [ln ; lm ] ln+m :
Let
l+ = n>0 ln ; l = n<0 ln :
It follows that
[l+ ; l+ ] = [l ; l ] = 0
and the bilinear form on l restricts to a non-degenerate alternating bilinear form on
each ln l n . Thus we can choose a basis (ei )i2Z+ in l+ and a basis (e i )i2Z in
l such that l is determined by the commutator relations as in (6.1). Together with
[z; ei] = [z; e i ] = [z; z ] = 0 these are called the Heisenberg commutator relations.
Notice that
b = l0 l
are maximal abelian Lie subalgebras of l. Consider a linear representation of b+ in the
one-dimensional linear space F defined by
a (z )(1) = a; a (l+ )(1) = 0:

Here a 2 F is a fixed parameter of the representation. Now we can define a linear
representation of the whole Lie algebra l by taking the induced representation:
V (a) = Indlb+ (a ) := U(l)

U(b+) F:
Recall that for any left module M over an associative F -algebra B the extension of
scalars of M to a B -algebra A is a left A-module A
B M defined as the quotient linear
61
space A
M=T , where T is the linear subspace spanned by tensors ab
m a
bm
with b 2 B; a 2 A; m 2 M and multiplication a (a0
m + T ) = aa0
m + T .
We can identify U(b+ ) with the algebra of polynomials F [t0 ; t1 ; t2 ; : : : ] in vari-
ables ti corresponding to the basis (z; e1 ; : : : ; en ; : : : ) of b. Similarly we identify U(l)
(as a linear space) with the linear space of Laurent polynomials F [: : : ; t 1 ; t0 ; t1 ; : : : ]:
However the multiplication is different:
ti t i = t i ti + t0 ; i > 0; (6.2)
and any other pair of variables commutes. The F [t0 ; t1 ; t2 ; : : : ]-module corresponding
to the representation a is the quotient algebra of F [t0 ; t1 ; t2 ; : : : ] modulo the ideal
generated by t0 a; t1 ; t2 ; : : : . Let us first describe the induced module V (a) as a
linear space. A monomial tj1 tjk of degree k in F [: : : ; t 1 ; t0 ; t1 ; : : : ] is called
normally ordered if j1 j2 : : : jk . For any monomial ti1 tik write
: ti1 tik := tj1 tjk ;
where j1 j2 : : : jk and (i1 ; : : : ; ik ) = (j(1) ; : : : ; j(k) ) for some permutation
2 Sk . Using the relations (6.2), we can write
ti1 tik =: ti1 tik : + normally ordered monomials of degree less than k:
We call the above the normal ordering decomposition of the monomial ti1 tik and
write it as n.o.d(ti1 tik ). For example, the normal ordering decomposition of t 2 t1 t 1
is equal to
t 2 t1 t 1 = t 2 t 1 t1 + t 2 t0 =: t 2 t1 t 1 : +t 2t0 :
Observe now that if in a normally ordered monomial ti1 tik the index ik is positive,
the coset of ti1 tik
1 in the induced module V (a) is equal to ti1 tik
tik 1
and hence is zero. If the monomial is equal to ti1 tik t0 then ti1 tik t0
1 =
ati1 tik
1. This shows that V (a) has a basis consisting of elements 1
1 and
ti1 tik
1, where i1 : : : ik < 0. This allows us to identify V (a) with the linear
space F [t 1 ; t 2 ; : : : ]. We have an isomorphism of linear spaces
V (a) = Sym(l ): (6.3)
The vector
j0i := 1
1
is called the vacuum vector. The structure of a U(l)-module on V (a) is given by
tj1 tjk (t i1 t ik
1) = n.o.d(tj1 tjk t i1 t ik )
1 =
n.o.d(tj1 tjk t i1 ti k )j0i:
Note that V (a) carries a natural grading defined by
deg(li1 lik ) = i1 + : : : + ik ;
where li 2 l i and deg j0i = 0.
Remark 6.1. More explicitly the representation V (a) of l can be described as follows.
We identify V (a) with the polynomial algebra C [x1 ; x2 ; : : : ] and assign to ei ; i >
0; the operator @x@ i , to e i ; i > 0; the operator xi : p(x) ! xi p(x), and to z the
scalar operator aid. Then [ @x @ ; xj ] = id and hence we get a representation obvioulsy
i
isomorphic to V (a).
There is an inner product on the space V (a) defined as follows. Let E be any linear
space over a field of characteristic 0 equipped with a symmetric bilinear form g . First
we define the bilinear form in T n (E ) by
g
n(v1
: : :
vn ; w1
: : :
wn ) = g(v1 ; w1 ) g(vn ; wn )
and then extend it to the whole T (E ) = n T n (E ) by requiring that T n (E ) and
T m(E ) are mutually orthogonal. Using the polarization process, we identify S n (E )
with S n (E ) equal to the subspace of symmetric tensors in T n (E ) = E
n and then
restrict g
n to S n (E ) to get a symmetric bilinear form symn (g ). One can show that
this inner product is non-degenerate if g is non-degenerate. Recalling the polarization
isomorphism we see that
symn (g )(ei1 : : : ein ; ej1 : : : ejn ) =

1 X g(e ; e ) : : : g(e ; e );
n! i1 (j1 ) in (jn ) (6.4)
where the sum is taken with respect to all permutations of n letters. Here we identify
Sym(E ) with the space of polynomials in a basis (ei ) of E . This defines a symmetric
bilinear form sym(g ) on Sym(E ). Following the physics agreement we shall drop n1!
in this formula. A similar construction can be given for any hermitian bilinear form. In
fact, if we choose a positive definite hermitian form on E we can complete the tensor
algebra T (E ) with respect to the corresponding norm and obtain a Hilbert space T^(E ).
This space is called the Fock space associated to the unitary space E . The completion
of the subspace Sym(E ) is called the bosonic Fock space. Similarly we can restrict
ourselves with the exterior algebra (E ) identified with the subspace of alternating
tensors in T (E ). Its completion is called the fermionic Fock space. We will deal with
it later.
We apply the construction of the Fock space to the Heisenberg algebra over C by
taking E = l , where l is equipped with a structure of a unitary space.
Let us consider the Lie algebra g with a linear basis 1; n ; n 2 Z; = 1; : : : ; D
with Lie bracket defined by commutator relations (5.17b):
[m ; n ] = mm+k;0 ; (6.5)
Let l be the graded Heisenberg algebra over C with dim ln = D for all n 6= 0. Let
(en ); = 1; : : : ; D; be a basis in ln such that, for any n > 0,
[en ; em ] = m;n z:
Consider the direct sum of Lie algebras lD = l a, where a = RD is viewed as an
abelian Lie algebra. Let (e0 ) be a basis of a. I claim that g is isomorphic to lD . To see
63
this we define the linear map f : lD ! g as follows.

8 i
> p
< n 0n if n 6= 0;
f (en ) = > p1n n if n 6= 0; 6= 0; (6.6)
:
0 if n = 0,
and f (z ) = 1. It is clear that this is an isomorphism of Lie algebras. We call lD the

oscillator algebra of RD . Let V (a) be the linear representation of the subalgebra l
generated by z; ei ; i 6= 0 described above. We can extend the representation to the
whole lD by setting
(e0 )j0i = j0i

for some 2 R. Since e0 belong to the center, it defines the representation. We
denote the obtained representation by V (a; ), where = (1 ; : : : ; D ) 2 RD . It is
more natural to consider as a linear function on the ideal a of lD generated by e0 s so
that = (e0 ). We will be interested only in representations corresponding to a = 1
so that we set V () = V (1; ). Its vacuum state is denoted by ji. Recall that we can
write any element of V () as
j()i := k11;::: ;k 1 n
;::: ;nn k1 : : : kn ji;
where = (1 ;::: ;n (k1 ; : : : ; kn )) is a tensor symmetric in lower and upper indices
defining a linear map S n (l ) ! S n (RD ) with finite-dimensional support. It is called
a Lorentz polarization tensor. Fix k = (k1 ; : : : ; kn ) 2 Z>0 and set
X
j(; k)i = (; k1 ; : : : ; kn ) = k11;::: ;k 1 n
;::: ;nn k1 : : : kn ji:
11 :::n =D
Let us define the inner product in V (). We may assume that ji is of norm 1.
Recall that we want the operators i and i to be adjoint to each other. Then
h i ji; j jii = hj i ji; jii = h( i j + jij )ji; jii = jij :

So we see that p1n n ji form an othonormal basis in Minkowski sense. In particular,
the vectors 0 n ji have squared norm equal to n. Following the discussion above
we can extend the inner product to the whole V (). Two different monomials in i s
are orthogonal and
jj1k1 : : : nkn jijj = k1 : : : kn ;

if all k1 ; : : : ; kn are distinct. We leave to the reader to deal with the general case. Note
that, for any k 2 Z>0, and = (1 ; : : : ; D ) 2 RD ,
jjj(; kijj = kjjjj = : (6.7)

Remark 6.2. One can define the Lie algebra lD and the Fock space V () in a coordinate-
free way. Let E be a vector space over a field K equipped with a non-degenerate sym-
metric bilinear form (x; y ). An element of l(E ) = K [t]
E can be interpreted as a
finite linear combination of tensors vn
tn ; vn 2 E; n 2 Z. Consider the Lie algebra
with generators vn
tn ; z , where z is central, satisfying the commutator relations
[vm
tm ; vn
tn ] = m(vn ; vm )m; n z: (6.8)
If E = RD is the Euclidean vector space, by choosing an orthonormal basis in E , we

see that l(E )
= lD . One defines the Fock space F (E ) = Sym(t 1 K [t 1
E ). Its
elements are finite linear combinations of tensors v n1 : : : v nk
t n ; n1 + : : : + nk =
n > 0; v n1 : : : v nk 2 Symk (E ). The inner product on F (E ) is defined by extending
the bilinear form (vn
tn ; vm
tm ) = (vn ; vm ) on t 1 K [t 1
E to the symmetric
product. The Lie algebra l(E ) has a representation in F (E ) by defining vm
tm to be
the adjoint of v m
t m for m > 0 and letting v m
t m act by multiplication:
v m
t m (v n1 : : : v nk
t n ) = v m v n1 : : : v nk
tm n :
Also we let v0
1 act by
v0
1(v n1 : : : v nk
t n ) = (v0 );
where : E ! K is a fixed linear form. It is easy to see that, in the case E = RD , we
get a representation isomorphic to V ().
There is one more important requirement on the spaces V (). The Lie algebra
of the Poincare group P of the Minkowski space V = R1;D 1 must have a linear
representation in these spaces. Recall that P is the semi-product of the translation
group V and the orhogonal group O(D 1; 1). The Lie algebra of P is the direct
product of the abelian algebra V = RD and the algebra of matrices A = (a ) 2

MD (R) satisfying a + a = 0. It has a set of generators ei ; i = 1; : : : ; D
and eij ; 1 i < j D; satisfying the commutator relations
[ei ; ej ] = 0; i; j = 1; : : : ; D; (6.9a)
[ejk ; ei ] = ji ek ki ej (6.9b)
[eij ; ekl ] = jk eil ik ejl jl eil + il ejk (6.9c)
Here eij corresponds to the matrices Eij eEij , where e = 1 if i 6= 1 and 1 if i = 1.

Define the operators
1
X 1 ( )
J = x p x p i n n n n n (6.10)
n=1
Then one checks that
[p ; J ] = i p + i p

65
[J ; J ] = i J + i J + i J i J :

This shows that the correspondence e ! p ; e ! J is a representation of the
Poincare Lie algebra in the Fock spaces V () (one has only to replace the commutators
[; ] with i[; ]).
Here the operators x p x p correspond to the matrices eij in the natural linear
representation of so(D 1; 1) in the space of vector fields in Rd as the vector fields
= x @x@ x @x@ . We can view a state j(; k)i 2 V () as a polynomial function
on Symn (l ) with values in Symn (Rd ) so that the vector field acts naturally on
such states. The translation part of the Lie algebra of the Poincare group acts via the
operators p .
Remark 6.3. Recall that an irreducible linear representation V of the Poincare group is
described by the following data. First one restricts the representation to the translation
subgroup T . Since the latter is an abelian group the linear space V decomposes into the
direct sum of eigensubspaces V ; 2 T ;, where V = fv 2 V : t v = (t)v; 8t 2
T g: The Lorentz group G = SO(n 1; 1) acts on T . It is easy to see that the
set f 2 T : V 6= f0gg is an orbit O of G. Let H be the isotropy subgroup of
some 0 2 O. Then the restriction of the representation to H defines an irreducible
representation of H in V0 . Now the natural action of G on O = G=H lifts to an action
on the vector bundle E = G H V0 = G V0 =H , where H acts on the product by
h (g; v) = (gh; h 1 v). There is a natural action of G on the space (E ) of sections
of this bundle and the representation V is realized as an irreducible subrepresentation
of (E ).
For example, consider the irreducible representation V which contains a vacuum
vector ji. The translation group T acts via the operators 0 . This shows that ji is
an eigenvector corresponding to the character 2 (RD ) . This shows that the fibres
of the vector bundle E are one-dimensional and the group H = SO(n 1; 1) acts
identically on the fibre over 2 O = SO(n 1; 1) . Thus the data describing the
representation consists of the orbit of determined by jjjj2 (if the norm is positive
then the group SO(n 1; 1) = SO(n 1) is compact) and the trivial representation
of SO(n 1; 1). Physicists say that ji transforms like a scalar.
We can define the similar space V~ () corresponding to right movers. Its vacuum
state is denoted by jiR . Then we consider the tensor product V ()
V~ (). Its vacuum
state is jiL
jiR .
Its vectors look like this
k11;::: ;k ;l ;::: ;l 1 n
;::: ;nn ;11 ;::: ;mm k1 : : : kn jiL
~ 1s1 : : : ~ msm jiR ;
where kn kn 1 : : : k1 and sm sn 1 : : : s1 and the polarization tensor
= (k11;::: ;k ;l ;::: ;l
;::: ;nn ;11 ;::: ;mm )
is symmetric in and (resp. in ki and lj ) separately.
One defines the norm on V~ () similar to the norm on V () and then gets a non-
degenerate inner product on V ()
V~ (). Finally we complete this space to get the
Fock space of the closed bosonic string theory

M
Fclosed = V^ ()
^ V^~ ():
2RD
Finally, let us see the representation of the Virasoro algebra generated by the oper-
ators Ln (resp. L ~ m ) in the space V () (resp. V~ ()). Recall that
L0 = 21 0 0 + N; L~ 0 = 12 0 0 + N;
~
where
X X
N= n n ; N~ = n n :
n1 n1
X X
Lm = m n n ; L~ m = m n n ; m 6= 0
n1 n1
~ are called level operators. It is easy to check that
The operators N and N
N 1k1 : : : nkn jiL = (k1 + : : : + kn )1k1 : : : nkn jiL

~.
and a similar formula holds for N
Recall that [Lm ; Ln ] = (m n)Lm+n + A(m)id. We have to find the constant
A(m) = am3 + bm.
One applies [Lm ; L m ] to some ground states to compute these constants. Notice
that
Lm ji = 21 (
X
1 n n )ji = 0; m > 0 (6.11)
n2Z
L0 ji = 12 0 0 ji = 21 jjjj2 ji (6.12)
Also
L 1 ji = 12 1 n n ji = 21 ( 1 0 + 0 1 )ji = 1 ji:

X
n2Z
Here we used that the operators Lm and L m are adjoint to each other. Thus
hjL1 L 1 jii = hL 1 ji; L 1 ji = 41 jj2 2 0 jijj2 = jj 1 jijj2 = jjjj2 ;

and we obtain
hj[L 1 ; L1 ]ji = hjL 1 L1i = jjjj2 =

67
hj(2L0 + A(1)id)ji = jjjj2 (1 + A(1)2 ):

This gives A(1) = a + b = 0. Also
hjL2 L 2 ji = jjL 2 jijj2 = 14 jj2 2 0 + 1 1 jijj2 =
jj 2 0 jijj2 + 41 jj 1 1 jijj2 = 2jjjj2 + 12 D:

Here we used (6.7) and
jj 1 1 jijj2 = hj1 1 1 1 ji = hj1 ( 1 1 + ) 1 ji =
hj 1 = 2D:
1 + 1 1 1 1 ji = hj2 1 1 ji = 2
Thus
hj[L2 ; L 2]ji = hjL2 L 2 ji = hj4L0 + A(2)idji =
2jjjj2 + A(2) = 2jjjj2 + 21 D:

This gives A(2) = 8a + 2b = 6a = 12 D, hence a = D=12. Finally we obtain that
for all m; n 2 Z, we have the following commutator relation for the Virasoro operators
acting in the space V ().
1 D(m3 m)
[Lm ; Ln] = (m n)Lm+n + 12 m; n : (6.13)
Exercises
6.1 Let l = n ln be a graded Heisenberg Lie algebra. Let Sp(l) be the symplectic
group of linear automorphisms of l which preserve the alternating from (x; y ). Con-
struct a linear projective representaion of Sp(l) in the space P(V (a)) which is compat-
ible with the representation of l in V (a).
6.2 Compute the norm of the state L3 ji.
6.3 Compute the norm of any state 1k1 : : : nkn ji.
6.4 Let p(n) denote the dimension of the space of eigenvectors P of the level operator N
1
with eigenvalue n. Compute the generating function Tr(q N ) = n=0 p(n)q n . Explain
the notation.
Lecture 7
Physical states for bosonic string
The expression for the Hamiltonian provides the mass-squared formula . Recall that
in the special relativity theory the mass is defined as the negative of the norm of the
moment vector in the Minkowski space-time. Let us explain it. We use the metric in
the space-time R4 with coordinates (x0 ; x1 ; x2 ; x3 ) = (x0 ; x) defined by dx20 dx21
dx22 dx23 . Here x0 = ct, where t is time and c is a constant equal to the speed of light.
To describe the motion we use the Lagrangian density
r
1 vc2 dt;
p 2
L = mc c2 jx0 j2 dt = mc2
where x0 = ddtx ; x = (x1 ; x2 ; x3 ) and m is a constant called the mass. The energy and
the moment for this Lagrangian are equal to
r
pi = @ L = mc2 = 1 v2 ; i = 1; 2; 3;
@ x_ i c2
r
1 vc2 :
2
E = pi x_ i L = mc2 =
We have
E 2 c2 jpj2 = m2 c4 ;
so if we set
P = (E=c2 ; p1=c; p2 =c; p3=c);

we obtain that
m2 = jjPjj2 ;
where we use the Minkowski norm defined by the matrix diag[ 1; 1; 1; 1]. The vector
P is called the total momentum vector. In our situation P = (p1 ; : : : ; pD ) so we can
69
70 LECTURE 7. PHYSICAL STATES FOR BOSONIC STRING
define the quantum mass-square operator by
M 2 = pp
1 1
p p = l42 (0 0
1 + 1
~0 ~0 ):
We shall scale the masses to assume that l2 = 8 so that
M 2 = ( L0 + N )
1 + 1
( L~ 0 + N~ ):
Thus the mass-square of the ground state jiL
jiR is equal to
jj2 = 21 22 + : : : 2D :

Recall that in the pre-quantized theory we had Virasoro constraints T = 0. It follows
from (5.15) that the analogs of these constraints in the quantum string theory are the
conditions that Lm = 0 for any element of the Fock space. However, because
[Ln ; L n] = 2nL0 + cid, this would imply that = 0. Thus we have to require
that Lm = 0 only for positive m and L0 = a for some ; and similarly for the
~ m. We set
operators L
Fphys() = f 2 F () : Lm = 0; m > 0; (L0 a) = 0g (7.1a)

F~phys() = f 2 F~ () : L~ m = 0; m > 0; (L~ 0 a) = 0g (7.1b)
Fphys
closed
= 2(RD) Fphys()
F~phys (): (7.1c)
A state satisfying these conditions is called physical. Also for any state and a
physical state , we have
hjL n i = hLn ()ji = 0; n > 0

Thus the intersection of Fphys with the sum Fspur = n>0 L n (F ) belongs to the null-
space of Fphys . The elements of this space are called spurious states. The Hilbert space
which we want will be the quotient
F+closed = Fphys
closed
=Fspur \ Fphys
closed
:
Note that the operators J defined in (6.9) commute with Virasoro operators, so that
the Poincare Lie algebra acts in the spaces of physical states.
Remark 7.1. An abstract Lie algebra is called a Virasoro algebra if it can be defined
by generators z; ln ; n 2 Z with commutator relations
3
[lm ; ln ] = (m n)lm+n + m 12 m m; n z; [z; ln] = 0:
It can be shown that any any Lie algebra obtained as a central extension with one-
dimensional center of the algebra of vector fields on a circle is isomorphic to a Vira-
soro algebra. A representation of the Virasoro algebra in a vector space V is called a
representation with highest weight a and charge c if z acts as a scalar operator c idV
and there exists a vector v0 (called a highest weight vector) such that
lm v0 = 0; m > 0; l0 v0 = av0 :
71
A universal representation with this property is called a Verma module and is denoted
by V (a; c). It can be constructed by using a similar construction as the representations
V (a) we constructed for a Heisenberg algebra. One considers the subalgebra Vir+
generated by the operators Lm ; m 0, then defines a one-dimensional representation
by L0 1 = a; Lm 1 = 0; m > 0 and finally takes the induced representation U(Vir)
Vir
C . Its elements are linear combinations of monomials L n1 : : : L nk v0 with positive
ni s. Any irreducible representation with highest weight a and charge c is isomorphic
to a quotient of V (a; c). So, we see that each nonzero 2 Fphys() generates a
representation space V for the Virasoro algebra with highest weight a and charge
c = D. Its highest weight vector is . As we have seen before any physical state
belongs to V ? .
Since all physical states are eigenvalues of L0 with eigenvalue a, we obtain the
mass-formula for physical states:
M 2 = (2L0 2N ) = 2N 2a: (7.2)
Let us see which ground states in F closed are physical. Since
N ji = 0; Lm ji = 0; m > 0; L0ji = 21 jj2 ;

and the same is true for the right mode operators L~ n , we see that the ground state
ji = jiL
jiR is physical if and only if
jj2 = 2a: (7.3)
For this vacuum state
M 2 ji = 2a
We shall see from the next discussion that a must be equal to 1. Thus the vacuum vec-
tors have negative mass. Such states are called tachyons (they travel faster than light!).
The existence of such states will force us to abandon bosonic strings and consider su-
perstrings.
Let us look for 2 Fphys () of level 1. Each such has the form 1 ji. We
have
L 1 = 1 0 = ; Lm = 0; m > 1; L0 = ( 12 jj2 + 1)
Thus is physical if and only if
= 0; jj2 = 2a 2: (7.4)
If a > 1, we may choose = (0; 1; 0; : : : ; 0), and then = (1; 0; : : : ; 0) satisfies

(7.4) but jj jj = jj2 = 1. This means that we have ghosts, i.e. states of negative
norm. This should be avoided since the the quantum mechanics deals only with Hilbert
spaces with unitary inner product. This forces us to take a 1.
If a < 1, we may take = (1; 0; : : : ; 0) and = (0; 1 ; : : : ; D ) so we have

D 1-dimensional space of physical states of positive norm and no states of non-
positive norm.
If a = 1, we may take = (1; 1; 0; : : : ; 0) and hence 0 = 1 . This shows that
we have a (D 2)-dimensional space of states of positive norm and a one-dimensional
space of states of norm 0. The state L 1 ji = 1 is spurious and is physical if
jj2 = = 0. Thus, if a = 1, Fphys() contains a one-dimensional space of
spurious states of norm 0. Factoring this space out we get a (D 2)-dimensional space
Fphys()=Fspur () \ Fphys(), each element of which can be represented by a state of
PSo far, we find that a 1 and no restriction on D appears.
positive norm.
Let = i i
~i 2 Fphys ()
F~phys (); where i 2 Fphys (); ~i 2 F~phys ():
Applying L0
1 we see that L0 i = a and applying 1
L~0 we see that L ~ 0 ~i = a.
This implies that N i = N ~ ~i . Hence
;::: ;n ;1 ;::: ;m 1 : : : n ~ 1 : : : ~ n ji;
= k11;::: ;kn ;s1 ;::: ;sm k1 kn s1 sn
where k1 + : : : + kn = s1 + : : : + sm . Let us look at the physical states in Fphys
closed
()
of level 1, i.e. N = N~ = . They are of the form
= 1 ~ 1 ji:
We have
L1 = 0 1 ( 1 ~ 1 ji) = ~ 1 ji; Lm = 0; m > 1: (7.5)
Similarly,
L~ 1 = 1 ji; L~ m = 0; m > 1: (7.6)

~ = ( 21 jj2 + 1) ; so that
Also L0 = L
jj2 = 2a 2: (7.7)
In view of (7.5) and (7.6), we get
= = 0:
The norm of the state is equal to
N = ij i j = :

If = 0, we have no restriction on and hence we have physical states of negative
norm. So 6= 0. If a < 1, we may assume that = (1; 0; : : : ; 0) so that 00 = 0
guarantees that is physical. But if we take ij = 0 for i; j 6= 0, we get a state of
negative norm, a ghost. So
a = 1:
If 6= 0, we may assume that = (1; 1; 0; : : : ; 0), and then P the condition is i0 =
i1 ; 0i = 1i ; i = 0; : : : ; D 1 so that the norm is equal to ; 2 2 . We see that
73
all physical states are of nonnegative norm. The states of zero norm satisfy = 0 if
; 2.
~ 1 ji = 1 ~ 1 ji and L~ 1 1 ji = 1 ~ 1 ji
Note that the states L 1
are spurious. Since jj = 2a 2 = 0, these states are also physical because
2
jj2 P = jj2 P = 0. It is easy to see now that any physical state of norm 0
is spurious and we can factor it out. Thus we obtain that for any non-zero light-like
the space Fphys
closed
() is of dimension D2 2D 2D = D(D 4) and all its elements
can be represented by physical states of positive norm.
For any of norm 0, the space of solutions = ( ) of (7.5) and (7.6) is the
direct sum of one-dimensional space of matrices with nonzero trace, the 12 D(D 1)
1-dimensional space of trace-less symmetric matrices and 12 D(D 3)-dimensional
space of antisymmetric matrices. The corresponding physical states are called dilatons,
gravitons and anti-symmetric tensors. These are massless particles (i.e. M 2 = 0).
Let us go to the second level, i.e. consider the physical states in Fphys () of the
form
= 1 1 ji + 2 ji;

where ij = ji . We have
L1 = (0 1 + 1 2 ) = 2( + ) 1 ji = 0;
L2 j; i = (0 2 + 12 1 1 ) = ( + 2 ) 1 ji = 0;
L0 = 12 jj2 + 2 = ( 12 jj2 + 2) :
This implies
+ = 0; + 2 = 0; (7.8)
jj2 = 2a 4 = 2:
The norm of this state is equal to 2N , where
X X X
N = + = 200 2 20 + 2 02 + 2 :
>0 ;>0 >0
Choose a system of coordinates in RD such that = (c; 0; : : : ; 0); c2 = 2. Using
(7.8) we can eliminate i s and 00 so that
D
X1 2 DX1
N = 251 ii + 2ij : (7.9)
i=1 i;j =1
Applying the Cauchy-Schwarz inequality, we obtain

25 DX1 25 DX1
2ij D25 1 2ii = 2ij D 2526 2ii :
X X
N (7.10)
i;j =1 i=1 i;j =1;i6=j i=1
Thus, if D 26, all states are of non-negative norm. If D > 26, the state with ii =
1; i 1 and ij = 0; i; j 6= 0; i 6= j has the norm equal to (D 1)( D + 26)=25 < 0.
Now if we take a physical state ~ from F~phys () of level 2 and of positive norm, then

~ is an element of Fphysclosed
() of negative norm. So we have ghosts. If D = 26, a
state of zero norm must satisfy ij = 0; i; j 1; i 6= j . The states
L 1( 1 jji) + cL 2ji = ( + 2c ) 1 1 jji + c 2 ji

are spurious. It is easy to see that any norm 0 physical state is equal to a physical
spurious state. So we can factor them out and obtain only the space with only postive
norms. Thus we have shown that D 26.
The proof that D = 26 consists of analyzing states of the next level. We skip it.
The result that Fphys
closed
does not contain ghosts if and only if a = 1; D = 26 is called
the No Ghost Theorem. It was proven by R. Brower, P. Goddard and C. Thorn.
Observe that a = 1 and D = 26 agrees with the definitionPof the Hamiltonian
operator H = L0 + L ~ 0 D24 2 using the regularization of the sum 1n=1 n. So physical
states satisfy H = 0.
Remark 7.2. One can show that one can choose a subspace in Fphys closed
invariant with
respect to SO(24) such that its states represent all states of positive norm modulo spu-
rious states. This is achieved by a light-cone gauge which consists of fixing the first
and the last coordinate of the string. The group SO(1; 25) acts in the space in F closed
via its induced representation. Thus F+ closed
defines a linear representation of the group
SO(24). It also leaves the homogeneous parts invariant and hence defines a finite di-
mensional representation in each space of given level. Elements of this space which
belong to an irreducible component are interpreted as elementary particles. For exam-
ple, the anti-symmetric tensors of level 1 define the adjoint representation of SO(24).
The dilatons define the trivial representation and gravitons define the standard repre-
sentation of SO(4) on the space S 2 (R24 ).
Exercises
7.1 Find physical states of level 2 in the Fock space of a closed bosonic string.
7.2 By analyzing physical states of level 3 in F () finish the proof of the No Ghost
Theorem.
Lecture 8
BRST-cohomology
We shall discuss another approach to defining physical states which is called the BRST-
quantization. In this approach one introduces an operator Q in a Fock space F of a
given string theory such that Q2 = 0 and
F phys = Ker(Q)=Im(Q) (8.1)
We shall start with reminding the definition of the cohomology group of a Lie
algebra g with coefficients in its linear representation V . Let F = (g )
V be
the tensor product of the exterior algebra of the space g and the space V . This will
be an analog of our Fock space. If g = Lie(G) for some Lie group G, then F can
be identified with the space
inv L of left-invariant smooth differential forms on G with
values in the trivial vector bundle defined by the space V . Let (ei ) be a basis of g and
(fi ) be the dual basis of g . Let C n (g; V ) = n(g )
V so that F = 1 n
n=0 C (g; V ).
n
Elements of C (g; V ) are linear combinations of the decomposable tensors
fi1 ^ : : : ^ fin
v; i1 < : : : < ik ; v 2 V
Let us assign to fi the operator ai = fi ^ and to ei the contraction operator ai = ei
such that ei (fi1 ^ : : : ^ fin
v ) = ( 1)k+1 fi1 ^ : : : fî : : : ^ fin
v if i = ik for
some k and 0 otherwise. If g = Lie(G) the operator ai is the exterior product with
the differential dxi and ai = @x @ . Here x1 ; : : : ; xdim g are local coordinates on G
i
corresponding to the basis (bi ). We have
fai ; aj g = ij ; fai ; aj g = fai ; aj g = 0:

Here, for any associative algebra A and x; y 2 A,
fx; yg = xy + yx
It is called the anti-commutator or the Poisson bracket.
The structure of g is determined by the constants ckij such that
[ei ; ek ] = ckij ek :
75
76 LECTURE 8. BRST-COHOMOLOGY
Let Ki = (ei ). Define the BRST-operator in F by

Q = ai
Ki 21 ckij ai aj ak
1 2 End(F
V ):
X X
i i;j;k
Lemma 8.1.
Q2 = 0
Proof. Let A1 = ai
Ki , A2 = ckij ai aj ak
1, where we skip the summation sign.
We have
A21 = ai aj
Ki Kj + aj ai
Kj Ki =
X X
ai aj
(Ki Kj Kj Ki ) = ai aj
csij Ks :
i<j i<j
Also
A1 A2 + A2 A1 = 21 (ai aj ak at + aj ak at ai )
ctjk Ki :
Using the anti-commutator relations we see that ai aj ak at + aj ak at ai = 0 unless t = i.
In the latter case ai aj ak ai + aj ak ai ai = aj ak (ai ai + ai ai ) = aj ak so that
A1 A2 + A2 A1 = 21 (cijk aj ak + ckj ak aj )
Ki =
X
aj ak
cijk Ki :
i<j
Here we used that ckij = ckji and aj ak = ak aj . This shows that A21 + A1 A2 +
A2 A1 = 0. It remains to show that A22 = 0. We have
4A22 = fijk fstmai aj ak as at am
1 = fijk fstm(ai aj ak as at am + as at am ai aj ak )
1:
If k 6= s; t; m 6= i; j , the expression in the bracket is equal to zero. So
4A22 = fijk fktm (ai aj ak ak at am + ak at am ai aj ak )

1+
fijk fskm (ai aj ak as ak am + as ak am ai aj ak )

1+
k f m (am aj ak as at am + as at am am aj ak )
1+
+fmj st
k f m (ai am ak as at am + as at am ai am ak )
1:
fim st
Using the Jacoby identity
fijm fmk
l + f m f l + f m f l = 0;
jk mi ki mj
it is easy to see that each of the four sums is equal to zero.
77
Applying the previous lemma we can define the cohomology of the Lie algebra g
with coefficients in V as follows:
H (g; V ) = Ker(Q)=Im(Q):
Example 8.1. Let g be an abelian Lie algebra of dimension n. Its linear representation
is a module over U(g) = C [e1 ; : : : ; en ]. Let V = C 1 (Rn ) with the action of g
defined by (ei ) = Ki : f ! @t@f . Then C n (g; V ) can be identified with the space
n
i
of smooth differential forms of degree n
! = fi1 :::in (x)dxi1 ^ : : : ^ dxin :

The BRST-operator
X
Q= ai
Ki
i
coincides with the exterior derivative d. We know that d2 = 0 and
H n (g; V ) = HDR
n (Rn ) = 0; n > 0:
Example 8.2. Let g = Lie(G). Assume that G is a complex semi-simple group and let
V = C be its trivial representation. Then
H n (g; C )
= H n (G; C );
where G is considered as a smooth manifold.
In our situation we want to take for g the Virasoro algebra Vir and V its represen-
tation in the Fock space F of bosonic string. The space (g ) is called the space of
ghost fields.
We will be dealing with a version of the BRST complex which uses the semi-infinite
cohomology. Instead of differential forms fi0 ^ : : : ^ fik we consider semi-infinite
forms. Let I = (i0 ; i1 ; : : : ) be any strictly decreasing sequence of integers such that
the sets I \ Z0 and Z<0 n (I \ Z<0) are finite. A semi-infinite form is a formal
expression of the form
I = fi0 ^ fi1 ^ : : : :
The number N = #I \ Z>0 #Z0 n (I \ Z0) is called the degree of I .
Let
N = I;
where I = (N; N 1; : : : ): Its degree is equal to N . We extend the operators ak and
ak to semi-definite forms in the obvious manner. Note that any form of degree N can
be obtained from N by applying operators ak an . Also observe that
ak = 0; k > deg ;
ak = 0; k < deg :
Let Vir be the abstract Virasoro algebra with generators ln . We want to construct the
representaion of Vir on the space of semi-infinite forms
1=2 (N ) of degree N . Let ad0
be the coadjoint representation of g on g . It is defined by ad0 (x)(f )(y ) = f ([x; y ]).
Let us identify (fi ) with the dual basis (li ) of Vir. We have
ad0 (ln )(fi )(lm ) = fi ([lm ; ln ]) = fi ((m n)ln+m ) =
(m n)fi (ln+m ) = (m n)i;n+m = (m n)m;i n :

This shows that
ad0 (ln )(fi ) = (i 2n)fi n (8.2)
If n 6= 0, we can set
X
(ln )fi0 ^ fi1 ^ : : : = fi0 ^ : : : ^ (ln )(fik ) ^ : : : =
k0
X
fi0 ^ : : : ^ fik 1 ^ (i 2n)(fik n ) ^ fik+1 ^ : : : :
k0
Observe that the sum is finite because n 6= 0. However it is not defined for n = 0. We
easily check that, for n; m; n + m 6= 0,
[(ln ); (lm )] = (n m)(ln+m ):

So our problem is to define (l0 ) such that all Virasoro commutators work. Next ob-
serve that
X X X
(ln ) = (i 2n)ai n ai = (k n)ak an+k = (n k)an+k ak :
i2Z k2Z k 2Z
We use this formula to set
X
(l0 ) = k : ak ak :
k2Z
Here : 1 : : : k : denotes the normal order of a composition of operators defined by
putting on the right the operator annihilating the vector N and inserting the sign of
the permutation which has been made. If no such operators occurs among the factors
we do nothing. Note that ak ak = ak ak 1 so changing ak ak to ak ak differs from
the usual product. It is easy to see now that each (ln ) is well-defined.
Let us compute [(ln ); (lm )]. We have
[(ln ); (lm )] = (n i)(m j )[an+i ai ; am+j aj ]:

79
Assume n; m 6= 0, i 6= m + j; j 6= n + i. Then, its is easy to see that [: an+i ai :; :

am+j aj :] = 0. Assume n; m 6= 0, i = m + j; j 6= n + i. Then
[: an+i ai :; : am+j aj :] = an+i ai ai aj ai aj an+i ai =
(ai ai + ai ai )an+i aj = an+i aj :
Similarly we get
[: an+i ai :; : am+j aj :] = am+j ai
if n; m 6= 0; j = n + i; i 6= m + j . Note that j = n + i; i = m + j implies m = n.
Thus, if m 6= n, we get
X
[(ln ); (lm )] = (n m j )(m j )an+m+j aj
j
X
(m n i)am+n+i ai = (n m)an+m+j aj = (n m)(lm+n ):
i
Assume n; m 6= 0 and n = m > 0. Then
[(ln ); (l n )] = (2n j )( n j )([: aj a n+j :; : a n+j aj :]:
Now, for any i; j such that i > N; j N we have
[ai aj ; aj aj ] = ai (1 aj aj )ai aj (1 ai ai )aj = ai ai aj aj = 1+ : ai ai : : aj aj :
Similarly, if i N; j > N , we have
[ai aj ; aj aj ] = 1+ : ai ai : : aj aj :
Now
X
[(ln ); (lm )] = (2n j )( n j )(1+ : aj aj : : a n+j a n+j :=
N<j n+N
3
2n(l0 ) 136n (N 2 N + n6 ):
Finally
[(ln ); (lm )] = (n m)(ln+m ) + 13 3 2 n

6 n + (N + N + 6 )n; m : (8.3)
If we fix the vacuum state 1 (i.e. take N = 1) the central charge is equal to 26
12 . Also,
D
(l0 )j 1 = 1 . Recall that the representation of Vir in F has the charge 12 and the
vacuum vector ji is the eigenvector of L0 with eigenvalue a = 1.
We define the BRST operator (Bechi-Rouet-Stora-Tyutin) on
1=2 (N )
F () by
Q = an
Ln 21 (n m) : an am an+m :
1:
Theorem 8.1. If D = 26, then Q2 = 0.

Proof. Let (ln ) = Mn . We have
A22 = 14
X
(an Mn ak Mk + ak Mk an Mn ):
k<n
We use that
Mn ak = (n m)am an+m ak = ak Mn (2n k)ak n :

Using this we get
an Mn ak Mk + ak Mk an Mn = an ak [Mn ; Mk ] (2n k)an ak n Mk :

Changing the index k to k + n in the second sum, and applying (8.3), we get,
an Mnak Mk + ak Mk an Mn = an ak ([Mn ; Mk ] (n k)Mn+k ) =

26 ( n3 + n)an a n :
12
On the other hand,
A21 = an
Ln am
Lm 21 am am
[Ln; Lm ] =
1 n m D 3 n
2 (n m)a a
Ln+m + 24 (n n)a a n :
Finally,
A1 A2 + A2 A1 = k 2 m (an : ak am ak+m :
Ln + : ak am ak+m : an
Ln +
k m (an : am ak a :
L + : am ak a : an
L = k m
L :
2 k+m n k+m n 2 k+m
So (A1 + A2 )2 = 0 if D = 26.
Let 1=2 (n) be the linear space of semi-infinite forms of degree n. It is clear that
Q maps 1=2 (n)
F to 1=2 (n + 1)
F . Let H n(Vir; F ) = Ker(Qn )=Im (Qn 1 ),
where Qn = Qj1=2 (n)
F . We set Hrel n (Vir; F )0 be the subspace of H n (Vir; F )
generated by the cosets of forms which do not contain f0 and which are annihilated by
(l0 )
1 + 1
L0 .
For any graded vector space V = 1 n=0 Vn with dim Vn < 1 we set
1
X
char(V )(q ) = dim Vn qn :
n=0
81
If T :V !V such that Vn = fv 2 V : T (v) = nvg, then dim Vn = Tr(T jVn ) and

char(V )(q ) = Tr(q T ):
We shall apply this to the case when V = F () and T = L0 . Recall that
L0 a1n1 : : : aknk ji = 21 jj2 + (n1 + : : : + nk ):

So it is easy to see that
char(F ()) := Tr(q L0 ) = q 2 jj

1 2
(q) D :
where
Y
(q) = (1 qk ):
k>0
We have already noticed that the representation of Vir in F () is reducible. Let us try
to decompose it into irreducible modules. First we write
F () = F (0 )
F (00 );
where = (1 ; : : : ; 24 ); 00 = 0 . Let us assume that 0 6= 0; 00 6= 0. Let M (h; c)
denote the Verma module for the representation of Vir with central charge c and charac-
ter h (see the previous Lecture). The Verma module M (h; c) has the universal property
with respect to all representations of Vir with central charge c and character h. Any
such representations is a quotient of the Verma module. One can show that M (h; c)
is spanned by the elements L n1 : : : L nk j0i, ni > 0. The grading of M (h; c) is de-
fined by taking M (h; c)n to be the subspace spanned by the monomials as above with
n1 + : : : + nk = n. We have
char(M (h; c)) = (q ) 1 :
It is known that the Verma module M (h; 1) is irreducible if h < 0 and irreducible and
unitary for c > 1; h > 1. Considering F (00 ) as a representation of Vir with character
h = 21 00 2 and central charge c = 1. Comparing the characters, we find that
F (00 ) = M ( 12 j00 j2 ; 1):
The charge a of the representation F (0 ) is equal to 25 and the character is equal to
1 j0 j2 . We have
2
1 02
char(F (0 )) = q 2 j j 25 :
1 02
The character of the irreducible module M ( 21 j0 j2 + k; 25) is equal to q 2 j j +k (q ) 1 .
This shows that
p(24) (k)M ( 12 j0 j2 + k; 25);

X
char(F (0 )) =
k0
where
X
p(d)(n)qn = (q) d :
n
We conclude that
F () = M ( 12 j00 j2 ; 1)
p(24) (k)M ( 21 j0 j2 + k; 25):
X
k0
Let v0 be the vacuum vector of M ( 12 j00 j2 ; 1). Set
T = k0 p(24) (k)fu

v0 : u 2 M ( 12 j0 j2 + k; 25); Ln(u) = 0; n > 0:
We have the following result due to I. Frenkel, Garland, and Zuckerman:
Theorem 8.2. Assume 6= 0. Then Hrel
n (Vir; F ()) = 0 for n 6= 0 and
dim Hrel0 (Vir; F ()) = p(24) (1 21 jj2 )

2 jj is an integer and zero otherwise.
if 1 1 2
Define the map s : Fphys () ! C 0 (Vir; F () by s(v )
= 1
v, where 1 =
f 1 ^ f 2 ^ : : : : We have Ln (v) = 0; n > 0, an ( 1 ) = 0; n < 0, an+m 1 =
0; n + m 0. Thus : an am an+m : 1 = 0 unless n + m 0 and n or m 0.
Assume n 0. Then m 0. If m < 0, : an am an+m : 1 = an+m an am 1 = 0.
0 0
So, m = n = 0 and : a a a0 : 0
1 = a0 ( a0 a + 1) 1 = a0 1 : Therefore, we
obtain
Q( 1
v ) = a0 1
L0 v 2 21 : a0 a0 a0 : 1
v = a0

(Lv v) = 0:
This defines a map from Fphys ! Hrel 0 (Vir; F ()): If s(v ) 2 Im (Q), then v 2 F spur \
Fphys and we get an injective map
Hphys = Hrel0 (Vir; F ()):
On the other hand, dim Fphys \ T = p(24) (k )(1 1 jj2 ) and T \ rad(Fphys ) =
2
0. Thus Fphys \ T is mapped isomorphically to Hrel (Vir; F ()) and hence Hphys =
0
Hrel0 (Vir; F ()).
Exercises
8.1 Show that the equivalent definition of the cohomology of a Lie algebra g with coef-
ficients in a linear representation : g ! V can be given as follows. Let C n (g; V ) be
the space of anti-symmetric n-multilinear maps from g with coefficients in V . Define
the coboundary map : C n (g; V ) ! C n (g; V ) by the formula
X
(f )(x1 ; : : : ; xn+1 ) = ( 1)i+j f ([xi ; xj ]; x1 ; : : : ; xî ; : : : ; x^j ; : : : ; xn+1 )+
i<j
83
nX
+1
( 1)i+1 (xi )(f (x1 ; : : : ; xî ; : : : ; xn+1 )):
i=1
Check that d2 = 0 and set H n (g; V ) = Ker(djC n (g; V ))=Im(djC n 1 (g; V )):
8.2 Consider the trivial representaion of g in a vector space V . Show that H 0 (g; V ) =
V; H 1 (g; V ) = Hom(g=[g; g]; V ).
8.3 A central extension of a Lie algebra g with help of a vector space V is a Lie algebra
g0 containing V as a central abelian subalgebra such that g0 =V = g. Show that such
central extensioncs can be classified by the space H 2 (g; V ), where g acts trivially on
V.
8.4 Prove that H 2 (Vir; R)
= R.
8.5 Let Q be the BRST-operator defined for the Virasoro algebra with coefficients in a
representation F (); 2 RD . Show that the exists a constant a such that (Q + a)2 = 0.

DOLGACHEV, I. Introduction To String Theory - Ps

Cargado por

Información del documento

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

DOLGACHEV, I. Introduction To String Theory - Ps

Cargado por

Copyright:

Formatos disponibles

Lecture 1

Quantum Field Theories: An

Let us start with a 0-dimensional theory. In this case  is a point, so  :  ! M is

Example 1.1. Recall the integral expression for the -function:

contributes ( 6i)2 (21 )3 121 = 3

contributes ( 6i)2 (21 )3 18 = 1693 : The total coefficient at 2 is 15

Now let us consider a QFT in dimension 1. Usually we write D = d + 1, where d

For example, let us take the Lagrangian

i (t) = xi + xti+1 txi (t ti ):

We shall use K (t; x; t0 ; x0 ) to define a certain Hermitian operator in the Hilbert

i.e., T is bounded, and

D(T ) = fy 2 H : sup jhT (x); yij < 1g:

Take y 2 D(T ). Since D(T ) is dense in H the linear functional x ! hT (x); y i

K (t1 ; x; t01 ; x0 ) = K (t2 ; x; t02 ; x0 ) if t02 t2 = t01 t1 ;

is dense in H and the operator defined by

We would like to apply the above to our function

(i) F : S (R) ! S (R) is an unitary operator;

Recall that a distribution is a continuous linear functional on the space C 1 (R)0 of

Now we can use the lemma to set

K (t; x) = F 1 ( p1 e it2 =2 ) = F ( p1 e it( )2 =2 ) = p 1 ei x2t :

k S (t)f k = k u(t; x) k = k F 1 (e it2 =2 f^( )) k =

ke it2 =2 f^( ) k = k f^( ) k = k f k:

Using property (iii) we get

S (t)f = e itH f; t > 0

K (t; x; t0 x0 ) = hxje i(t0 t)H jx0 i (1.14)

hjjT jji = hjT ji:

Q : S (R) ! L2 (R); f ! xf:

Qa() = a (Q) = a (x) = a(a) = aa ():

Taking a = x; b = x0 we get formula (1.14). We have to understand it as

x (e iH (t0 t) x0 ) = K (t; x; t0 ; x0 ):

hx [t]; x0 [t0 ]i = K (t; x; t0 ; x0 ):

ut 21 uxx = (t)(x y);

Partition function as the trace of

Tr(T ) = ln(det(eT )): (2.3)

det(T ) = T (0) det(T ): (2.4)

 (0) = 12 ;  0 (0) = 21 ln 2:

T0 (0) = 2 ln(2r);

K (t; x; t0 x0 ) = p 1 ei(x0 x)2 =2(t0 t) ;

e xn2 = p1x e n2 =x :

Comparing this with (2.6), we see that

Z E (R; r) = pr Z E (R; R=r):

n Z1 n r n=p2 = pn=2 = det( 1 T ) 12 :

This approximation is exact when the action is quadratic in .

The classical path is a solution of the Lagrangian equation:

The value of the action functional on the classical solution is

K (t0 ; y; t1 ; x) = exp(iS ( cl ))

n=1 n=1 n=1

n=1 2i 2iT

K (t0; y; t1 ; x) = exp(iS ( cl )) det( 21i D) 1=2 =

2.2 Let T = Rn = be a n-torus. Here = Z!1 + : : : Z!n and !1 ; : : : ; !n are linear

The quantum mechanics is a 0 + 1 dimension QFT.

ha ; x i = tlim p1 e (x a)2 =2t = 

local coordinates q = (q1 ; : : : ; qn ) in M and the corresponding local coordinates

Here the left-hand side is evaluated at a path t ! ( (t); _ (t)) in TM given by qi =

D : C 1 (TX ) ! End(C 1 (X )):

Recall that a symplectic form on a smooth manifold X is a smooth closed 2-form

!x( @q@ ; @q@ ) = !x( @p@ ; @p@ ) = 0 for all i; j;

!x ( @q@ ; @p@ ) = !x( @p@ ; @q@ ) = ij :

(H )(f ) = ff; H g:

Let (X; ! ) be a symplectic manifold. For any two functions f; g 2 C 1 (X ) one

ff; gg = !({! (df ); {! (dg)):

ff; gg = df ({! (dg)) = {! (dg)(f ) = {! (df )(g):

ff; ghg = ff; ggh + ff; hgg:

Let us start with a 0-dimensional theory. In this case is a point, so : ! M is

contributes ( 6i)2 (21 )3 121 = 3

contributes ( 6i)2 (21 )3 18 = 1693 : The total coefficient at 2 is 15

K (t; x) = F 1 ( p1 e it2 =2 ) = F ( p1 e it( )2 =2 ) = p 1 ei x2t :

k S (t)f k = k u(t; x) k = k F 1 (e it2 =2 f^( )) k =

ke it2 =2 f^( ) k = k f^( ) k = k f k:

hjjT jji = hjT ji:

Qa() = a (Q) = a (x) = a(a) = aa ():

x (e iH (t0 t) x0 ) = K (t; x; t0 ; x0 ):

hx [t]; x0 [t0 ]i = K (t; x; t0 ; x0 ):

ut 21 uxx = (t)(x y);

det(T ) = T (0) det(T ): (2.4)

(0) = 12 ; 0 (0) = 21 ln 2:

T0 (0) = 2 ln(2r);

e xn2 = p1x e n2 =x :

n Z1 n r n=p2 = pn=2 = det( 1 T ) 12 :

n=1 2i 2iT

K (t0; y; t1 ; x) = exp(iS ( cl )) det( 21i D) 1=2 =

ha ; x i = tlim p1 e (x a)2 =2t =

!x ( @q@ ; @p@ ) = !x( @p@ ; @q@ ) = ij :

fpi ; pj g = fqi ; qj g = 0; fqi ; pj g = ij : (3.4)

jj jj2 = h ; H i = h ; ay a i + h ; ~2! i = jja jj2 + ~2! jj jj2 :

(A) (B ) 21 jh[A; B ]i j:

L = L(xi ; y ; yi )jdn xj:

hf; i = Tr(f );

L () = jdj2 dX = Tr(d d)dX : (4.3)

(A1) (isometry invariance) For any diffeomorphism : ! 0 preserving the met-

(A3) (conformal scaling) if h0 = ec(x) h is a new metric on , and S 0 is the new

Proof. Let : V = T ! W be a linear map of inner product spaces and : V 0 !V

h((v0 )); wiW = h(v0 ); (w)iV = hv0 ; 1 ( (w))iV 0 :

Tr(( ) ( )) = Tr( 1 ( ) ) = Tr( ):

jA + B j = jAj + Tr(A 1 B )jAj:

jA + B j 12 = jAj 12 (1 + 2 Tr(A 1 B )):

T = (Tij )() = ( ij 21 hij h )dxi dxj :

where D is the covariant derivative of a section d of E = TM

In many quantum field theories M is a fibre bundle over and : ! M is

A( )(fs) = (f )s + fA( )(s);

hO1 (P1 ) : : : On (Pn )i = Zf1 ;fZ2 (Oi (()

R(O(P ); t1 ; t2 )( 2 ; 1 ) = h 2 je i(t2 t1 )H O (P )ei(t2 t1 )H 1 iH :

[(t; x); (t; y)] = ~i (x y); (4.13)

[(t; x); (t; y)] = [(t; x); (t; y)] = 0: (4.14)

Here we have to consider ; as operator valued distributions, i.e. a continuous linear

(; ) ! [T1 (); T2 ( )]:

[ak ; ayk0 ] = (k k0 );

h @ (@L ) () h L()

~(t) : R ! L(M ); t ! ( ! (t; ));

S ( ; (h)) = S (; h); (5.2)

S (; ef h) = S (; h): (5.3)

(@t2 @x2 ) = 0; = 1; : : : ; D: (5.6)

So our field ( ) satisfies the Klein-Gordon massless equation.

T10 = T01 = @t @x = 0; T00 = T11 = 21 (@t @t + @x @x ) = 0:

= L (+ ) + R ( ):

L (t; x) = 21 x + l0 + + ilr 1 e in+ =r ;

R (t; x) = 21 x + l~0 + ilr 1 ~ e in =r ;

P = @ (@@ L ) = T@t :

L (t; x) = 21 x + 21 l2 p + + ilr 1 e in+ =r ;

R (t; x) = 12 x + 21 l2 p + ilr 1 ~ e in =r ;

Remark 5.1. If we choose the Riemannian metric on instead of pseudo-Riemannian,

@z @zX (z; z) = 0

R (z ) = 12 x + 21 il2p ln z + ilr 1 ~ zk

@t (R ) @t (R ) = @t (L ) @t (L ) = 0:

Now we quantize as in the previous lecture by taking n as operators in some

[P (t; x); X (t; x0 )] = i(x x0 ) ;

[x ; p ] = i (5.17a)

all other commutators between, x ; n ; ~n are equal to zero.

It is called the normal order for the composition of operators. Since [m n ; n ] = 0,