Conjetura Maldacena
Conjetura Maldacena
Conjetura Maldacena
ABSTRACT
These lectures do not at all provide a general review of this rapidly growing field.
Instead a rather detailed account is presented of a number of the most elementary
aspects.
Based on Lectures presented at Quantum aspects of gauge theories, supersymmetry
and unification, School in Leuven, January 18-23, 1999; and on Lectures presented at
Nordic Course on Duality in String and Field Theories, 4-14 August, 1998, Nordita,
Copenhagen. Lecture notes prepared with the help of Troels Harmark.
Introduction
The Maldacena conjecture [1] is a conjecture concerning string theory or M theory on certain
backgrounds of the form AdSd MDd . Here AdSd is an anti de Sitter space of space-time
dimension d, and MDd is a certain compactification space of dimension D d with D = 10 for
string theory and D = 11 for M theory. In addition, the background is specified by a statement
about the flux of a certain field strength differential form. The conjecture asserts that the
quantum string- or M-theory on this background is mathematically equivalent - or dual as the
word goes - to an ordinary but conformally invariant quantum field theory in a space-time of
dimension d 1, which in fact has the interpretation of the boundary of AdSd . This seems
to put the formulation of string/M-theory on a novel and rather unexpected footing. Also
the relation between quantum and classical theory is illuminated in a surprising way by the
conjecture. Several details in Maldacenas original formulation were left unspecified. Most of
those were subsequently given a precise formulation by independent works of Gubser, Klebanov
and Polyakov [2] and by Witten [3]. A priori it might seem very strange that quantum theories
in different space-time dimensions could be equivalent. This possibility is related to the fact
that the theory in the larger dimension is (among other things) a quantum theory of gravity.
For such theories the concept of holography has been introduced as a generic property, and the
Maldacena conjecture is an example of the realization of that (for discussion, see for example
[4]).
In the meantime a large number of checks have been performed which we shall not attempt
to review in these notes (for some recent reviews with many additional references, see for
example [5, 6, 7, 8]). Supposing the conjecture is true, it remains somewhat unclear what the
most significant consequence will be. On the one hand the conjecture allows one to obtain non
perturbative information on ordinary, but mostly conformally invariant quantum field theories,
especially at large N (of a gauge group U(N)), from classical string/M-theory or even classical
supergravity. This is a remarkable unexpected development, and the one that has mostly
been pursued until now. On the other hand it is conceivable that the conjecture will play an
important role in the eventual non-perturbative formulation of M-theory, for which the matrix
model of BFSS [9] was a first proposal. In a somewhat different line of development, Witten
[10] showed how to apparently overcome the original restriction to conformally invariant (and
mostly supersymmetric) quantum field theories, providing in fact an entirely new framework
for studying large N ordinary QCD and similar theories. A rather new idea about how to
achieve the same end in perhaps a more efficient way has recently appeared [11]. That approach,
however will not be covered here at all (see also [5]). In any case, the AdS/CFT development
attracts an enormous interest.
In these lectures we shall attempt a very elementary introduction to a somewhat restricted
number of basic aspects. In sect. 2 we begin by reviewing properties of anti de Sitter spaces,
their isometries, the fact that they may be associated with a boundary and the fact that
the isometry group of anti de Sitter space becomes the conformal group on the boundary. It
follows that if a quantum theory on anti de Sitter space is dual to another quantum theory on
the boundary, then that second theory must necessarily be conformal.
In sect. 3 we expand on the discussion in [1] and provide a short review of classical supergravity solutions in the presence of branes. Both so called extremal (BPS) and non-extremal
1
solutions will be considered for later reference. This subject has already been reviewed on numerous occasions (see for example [12, 13, 14, 15]). We describe how the so called near horizon
approximation in some cases lead to geometries of the form AdSd S Dd . This fact has been
known for several years by the experts, but its full significance was only realized by Maldacena.
In sec. 4 we follow rather closely [3] and describe in detail several instructive albeit rather
trivial examples of how the duality between the bulk theory and the boundary theory works in
the case of free theories. An important object which has a general significance is the generalized
propagator describing propagation of certain modes from a space-time point in the bulk of anti
de Sitter space to a point on the boundary. This propagator was the key object in the
discussions in [2, 3] and will be constructed in a few of the simplest cases. At the same time
the Maldacena conjecture will be made more precise.
In sect. 5 we follow [10] and describe how certain finite temperature scenarios may be used
to provide a mechanism for breaking conformal invariance and supersymmetry, and thereby
obtain a framework for studying large N QCD.
(1)
2D
R =
g
2D
(2)
So these spaces have the property that the Ricci tensor is proportional to the metric tensor:
They are Einstein spaces. We shall be interested in various examples of such spaces, in particular
in ones with maximal symmetry, for which in addition we have
R =
R
(g g g g )
D(D 1)
(3)
Such spaces are (for R 6= 0): spheres, S D , de Sitter spaces, dSD , and anti de Sitter spaces,
AdSD . The difference between de Sitter space and anti de Sitter space is the sign of the
cosmological constant. With the above conventions, AdS spaces have > 0 (see below).
2.1
AdSn+1 by embedding
It is useful to consider an (n + 1)-dim AdSn+1 as a submanifold of a pseudo-Euclidean (n + 2)dimensional embedding space with coordinates (y a ) = (y 0 , y 1, ..., y n , y n+1) and metric
ab = diag(+, , , ..., , +)
with length squared
2
0 2
y (y ) + (y
n+1 2
n
X
(y i)2
i=1
y a y = a b y b ,
a b SO(2, n)
(4)
(5)
For de Sitter spaces we would use a mostly plus metric and the same definition (or equivalently, b2 b2 ) and similarly for the spherical spaces, for which of course the metric is
positive definite. We shall demonstrate below that this implies eq.(3).
If instead of AdSn+1 we consider the n + 1 dimensional Minkowski space, we know that our
theory should be invariant under the Poincare group, which in n + 1 dimensions has dimension
n+1 (for the translations) plus 12 n(n+1) (for the Lorenz transformations) in total 21 (n+1)(n+2).
In fact the Poincare group is exactly the isometry group of flat space: invariant intervals,
squared, are preserved by the Poincare group. With the definition of AdSn+1 just given, it is
a
obvious that the isometry group of that space instead is SO(2, n). In fact, let y0a , y0a + dy(1)
a
and y0a + dy(2)
be 3 points lying in the submanifold given by eq.(5), and let y a0 , y a0 + dy a(1)
and y a0 + dy a(2) be the corresponding images under the SO(2, n) transformation eq.(4). Clearly
these also lie in AdSn+1 , and in particular we have
(6)
2.1.1
Polar/Stereographic coordinates
y0 =
y
(7)
where
x2 (x1 )2 + ... + (xn )2 (xn+1 )2
(8)
where d~y = (dy 1, ..., dy n). From this we may work out the metric in x coordinates. We
get (Notation: = 1, ..., n + 1 and we raise and lower by the flat Minkowski metric
diag(+, +, ..., +, ))
dy
dy
1 + x2
x dx
= d
+ 4
1 x2
(1 x2 )2
2
2x
+
{(1 x2 ) + 2x x }dx
= d
2
1x
(1 x2 )2
42
dx2
2
2
(1 x )
(9)
(10)
We see that in these coordinates the metric factorizes into a trivial radial part and an interesting (angular) AdS part:
4b2
(11)
g = +
(1 x2 )2
(convention: mostly plus).
We now want to verify that this metric indeed satisfies the Einstein equation in vacuum
eq.(2) with a cosmological term:
R g
and determine the constant in terms of the dimension D = n + 1 and the AdS-scale, b. We
shall do even more, and verify that the spaces also satisfy the maximal symmetry condition,
eq.(3).
Let us in fact consider a general conformally flat metric of the form
g (x) = e(x)
In our case
(x) = log 4b2 2 log(1 x2 )
4
1
g ( g + g g )
2
1
( + )
2
+
1
( +
2
) ( )
1
+
2
1
+ + ()2 ( )
4
16x x
(1 x2 )2
4
8x x
+
2
1x
(1 x2 )2
4
(1 x2 )2
1
2 (g + g )
b
(12)
the last equality being the statement of maximal symmetry. Then further
R =
D1
g ,
b2
(D n + 1)
(13)
n(n 1)
(D 1)(D 2)
=
b2
b2
(14)
(Notice that under a shift in convention g g , i.e. mostly plus mostly minus,
R is unchanged, so the cosmological term will appear with the opposite sign.)
Exercise: Show for the conformally flat metric, that in general
D
1
1
D
2
2
R = 1
( ) + 1
()
2
2
2
2
Use this to provide yet another derivation of 13.
We shall also need to consider the case of Euclidean signature or imaginary times. Then
2
dx =
n+1
X
=1
(dx )2
(15)
In such coordinates the Euclidean version of AdSn+1 is topologically the ball Bn+1
n+1
X
(x )2 < 1
=1
The boundary of the ball lies infinitely far away as measured in the AdS metric. We shall
come back to that.
Exercise: In the Euclidean case, AdSn+1 may be viewed as the hyperbola
(y 0)2 r 2 = b2
where
r2
n+1
X
(y )2
=1
Denoting the point (y 0 , r) = (b, 0) as the South Pole, show that the coordinates x eq.(7)
(Euclidean version) are the stereographic projections of AdSn+1 from the South Pole to the
equatorial plane y 0 = 0 (in units of b).
Now define light cone coordinates
u = y 0 + iy n+1, v = y 0 iy n+1
(16)
(17)
( )2
~2
(18)
=1
then
y 2 = uv ~y 2 = uv u2 ~2 = b2
b2
v = 2u +
u
6
(19)
b2 du
u2
dy = ud + du
~2
(ds2 )embedding = dudv dy
b2 du2
= 2 u2 d~2 (y 2 b2 )
u
2
b du2
(ds2 )AdSn+1 = + 2 + u2 d~2 (mostly plus)
u
(20)
The second set is similar to Poincare coordinates on the projective plane. For simplicity
put b 1 and use the set
~ (u1 , )
~
( 0 , )
Then log u = log 0 and
du
u
= d0 . Hence
ds2 =
(d 0)2
d~2
1 0 2
~2
+
=
(d
)
+
d
( 0 )2
( 0 )2
( 0 )2
(21)
2.2
Anti de Sitter space has a kind of projective boundary. The idea is in embedding space to
consider (y 0, y ) very large with y AdSn+1 . Hence define new variables
and take R . Then
y a = R
y a , u = R
u, v = R
v
(22)
2
y 2 = b2 uv ~y = b2 /R2 0
(23)
2
uv ~y = 0
(24)
But since tR is just as good as R for any t R, we have to consider the boundary to be the
projective equivalence classes
uv ~y 2 = 0
(u, v, ~y) t(u, v, ~y)
(25)
so the boundary is n-dimensional as it should be. Using the equivalence scaling, the boundary
may be considered to be represented by (Minkowski signature)
(y 0 )2 + (y n+1)2 = 1 = ~y 2
7
(26)
so that topologically the boundary is S 1 S n1 . In another use of scaling, for points with
v 6= 0 we may scale to v = 1. Then u = ~y 2 and we may use ~y as coordinates on the boundary.
2
Equivalently, if also u 6= 0 we may instead scale u to 1 and use coordinates ~y and have v = ~y .
Clearly the connection between the two sets is
~y = ~y
y2
(27)
When either v = 0 or u = 0, only one of the two sets may be used. For v = 0, ~y = ~0
whereas for u = 0, ~y = ~0. We may think of the (one) point v = 0 as the point at infinity
in the ~y coordinates, and similarly for u = 0. So the boundary is automatically compactified.
The situation is analogous to compactifying the Riemann sphere including the point z =
with z a good coordinate in a neighbourhood of z = 0, and = 1/z a good coordinate in a
neighbourhood of z = .
The above definition of AdSn+1 and its boundary in terms of the embedding space, implies
that the isometry group SO(2, n) (SO(1, n + 1) for Euclidean signature) acts in an obvious way
on points of the boundary. The crucial result on which we would like to elaborate, is that the
isometry group SO(2, n) (SO(1, n + 1)) acts on the boundary as the conformal group acting on
Minkowski (Euclidean) space.
2.3
For definiteness, consider n-dimensional Euclidean space E n . We first want to understand that
the conformal group is SO(1, n + 1). Begin by counting the number of generators = number
of generators in SO(n + 2) = number of linearly independent antisymmetric (n + 2) (n + 2)
matrices
1
dimSO(1, n + 1) = (n + 2)(n + 1)
(28)
2
By comparison, the Poincare group in n dimensions has n translation generators and 12 n(n 1)
rotation generators
1
dim Poincare(E n ) = n(n + 1)
(29)
2
so the difference is n + 1. This just fits with the following extra possible conformal transformations:
Dilations
~x ~x, R
(30)
~x ~x such that
x
x
=
+
(31)
2
2
x
x
involve the n parameters , = 1, ..., n and give rise to the additional n generators. Equivalently we may write
x + x2
(32)
x =
1 + 2~
~x + 2 x2
8
The equivalence between eq.(31) and eq.(32) follows after noting (from eq.(32)) that
2
x =
x2
1 + 2~
~x + 2 x2
(33)
To verify that these are really conformal transformations, consider 3 neighbouring points
~x, ~x + d~x1 , ~x + d~x2
and their images
~x , ~x + d~x1 , ~x + d~x2
The statement that the transformation is conformal, is the statement that the angles are preserved, or
d~x d~x2
d~x1 d~x2
p1
p
(34)
=
(dx1 )2 (dx2 )2
dx21 dx22
x2 dx 2~x d~xx
x 2 dx 2~x d~x x
=
(x2 )2
(x 2 )2
d~xi d~xj
d~xi d~xj
=
, i, j = 1, 2
4
x4
x
(35)
~y
~y
where SO(1, n + 1) i.e. preserves the norm uv ~y 2.
Similarly, a point on the boundary has coordinates (u, v, ~y) subject to
(i)
(ii)
uv ~y 2 = 0
(u, v, ~y) (u, v, ~y)
(37)
and is mapped by to (u , v , ~y ) as before. (Notice that of course (u2 , v2 , ~y2 ) = (u1 , v1 , ~y1 )
(u2 , v2 , ~y2 ) = (u1 , v1 , ~y1 )).
Now consider the infinitesimal transformation = 1n+2 + with infinitesimal. In order
for the relevant norm to be preserved, the (n + 2) (n + 2) dimensional matrix, must be of
the form
a
0
~T
(38)
= 0 a ~ T
1
1~
~ n
2
where
~ , ~ are n-vectors represented as columns and n is an nn antisymmetric matrix. (The
strange looking factors 12 are due to the fact that we have a non-trivial metric on E n+2 in the
9
coordinates (u, v, ~y), and that our matrices have indices like a b rather than two lower indices,
say.)
Indeed
u(1 + a) +
~ ~y
u
u
~
(1n+2 + ) v = v = v(1 a) + ~y
(39)
u~
v
~y
~y
~y + 2 + 2
~ + n ~y
2
~ n .
and one checks that u v y~ = uv ~y 2 to first order in the infinitesimal quantities, a,
~ , ,
2
Now, choose a representative (u, v, ~y) for a boundary point with v = 1, u = ~y . (This can
always be done except for v = 0 corresponding to a point at infinity on the boundary.) We
have seen that with this representation, ~y is a convenient representation of the boundary point.
Now, the image point according to eq.(39) is not in the same convention: v 6= 1 in general.
But the image point is equivalent to (u /v , 1, ~y /v ), which is in the same convention. Thus,
the effect of the mapping is
~y ~y /v = ~y (1 + a ~ ~y ) +
y2 ~ 1
+
~ + n~y
2
2
(40)
Let us verify that this transformation is in fact a combination of infinitesimal (i) translations
(ii) (Lorentz-) rotations (iii) dilations and (iv) special conformal transformations:
(i) Only
~ 6= 0
1
~
(41)
~y ~y +
2
i.e. translations.
(ii) Only n 6= 0
~y ~y + n ~y
(42)
i.e. rotations.
(iii) Only a 6= 0
~y ~y (1 + a)
(43)
1
~y ~y (1 ~ ~y ) + y 2 ~
2
(44)
i.e. dilation.
(iv) Only ~ 6= 0
~ we find
If we compare with eq.(32) and put
~ in that equation equal to /2
~y
~ 2
~y + 12 y
~ ~y ) + + 1 y 2~ + O( 2)
=
~
y
(1
2
1 + ~ ~y + 1 2 y 2
(45)
in agreement with the above, i.e. indeed we find in this case the (infinitesimal) special conformal
transformations.
This completes the main result in this section that SO(1, n + 1) (and SO(2, n) in the
Minkowski case) acts (i) as an isometry on AdSn+1 and (ii) as the conformal group on the
boundary of AdSn+1 .
10
2.4
For completeness let us work out the Lie algebra of the conformal group. We choose the simplest
possible representation, which is in terms of scalar fields (x) with ~x an n-tuple of Cartesian
coordinates. It is trivial to check that the generators are represented as follows:
Translations P = i
(Lorentz-)rotations M = i(x x ) = (x P x P )
Dilations D = ix
Special Conformal Transformations K = i(2x x x2 ) = 2x D + x2 P
One then easily finds:
[M , P ]
[M , M ]
[M , K ]
[D, P ]
[D, K ]
[P , K ]
=
=
=
=
=
=
i(g P g P )
i (g M + g M g M g M )
i(g K g K )
+iP
iK
2i(g D + M )
11
(46)
3
3.1
In this section we first briefly review the construction of solitonic p-branes in low energy effective
supergravity. There are many excellent reviews available, for example. refs. [12, 13, 16, 14] to
which we refer for more details and references to the extensive original literature. We begin by
writing down the effective low energy string action for type II (A or B) strings in the string
frame:
!
Z
X 1
p
1
1
d10 x |g| e2 (R + 4g )
Ss = s
F 2 + ...
(47)
16G10
2 n n! n
(s = 1(+1) for Minkowski (Euclidean) signature, flipping to mostly minus signature introduces an additional sign of ()n in front of Fn2 )). Here the dots represent fermionic terms as
well as the NS-NS 3-form field strength term. is the dilaton, and the n-form field strengths
Fn belong to the RR sector. For the Newton constant in D dimensions we write
16GD = 22D
We shall only be concerned with the terms given. For IIA strings (IIB strings) we only have
even (odd) values of n. For the IIB string the n = 5 field strength tensor is self-dual (In
Minkowski space, see below), and it is not strictly speaking possible to describe the theory by
the simple action above. A more complicated formulation nonetheless exists [17]. However, it
turns out to be suffcient to adopt the above action for deriving the equations of motions, and
imposing self-duality a posteriori (making sure that the normalization of F52 is unchanged). We
shall therefore employ that procedure. It is convenient for various reasons to also represent
the action for fields in the Einstein frame, obtained by a certain Weyl rescaling. In fact the
following identity in D space-time dimensions may be verified [18]
g e2 g
p
p
p
1
|g|e2 R
|g|e((D2)+2) {R + 2(D 1) p ( |g| )
|g|
2
2
(D 1)(D 2)() }
(48)
2
D2
and get rid of a total derivative, specifically in 10 dimensions:
=
g (Einstein) = e 2 g (string)
so we obtain in the Einstein frame
!
Z
X 1
p
1
1
1
SE = s
d10 x |g| R g
ean Fn2 + ...
16G10
2
2 n n!
12
(49)
(50)
with
1
an = (n 5)
2
We shall also be concerned with low-energy M-theory in the form of 11-dimensional super
gravity. The bosonic fields of that theory are just the metric and a 3-form gauge potential C
with a 4-form field strength tensor
K = dC
The bosonic part of the action is
1
Sbosonic (11-dim SUGRA) = s 2
211
Z
Z
p
1
1 2
d x |g|{R K }
CKK
48
6
11
(51)
S = s 2
dD x g{R g
e Fn + ..}
(52)
2D
2
2 n n!
in particular with a = 0 and 0 for 11-dimensional supergravity. The p-brane is a source of
charge for the p + 1 form (RR-) gauge field and the n = p + 2 form field strength. We write
D = (p + 1) + d
(53)
3.2
1
F ... dx1 dxn
n! 1 n
(54)
(55)
From that we define the proper D-form, the volume form with tensor components
1 ...D =
s
1 ...D = 1 ...D
g
g1 ...D ,
(56)
Here s = 1 (s = 1) for Euclidean (Minkowski) metric (both mostly plus). We have dropped
from now on the numerical signs around the determinant of the metric but they are always to
be understood.
13
=
M
dn x gind
(57)
(58)
with (gind )ab the induced metric on the sub manifold M. The Hodge dual satisfies
(F )n+1 ...D =
(F )n+1 ...D =
(F F )01...(D1) =
F
F F
F2
1 2
F
n!
=
=
1 1 ...D
1 s
(59)
which for a given dimension and rank is obviously impossible for both Minkowski signature
and Euclidean signature. In particular, the 5-form field strength in 10 dimensional IIB string
theory is self dual only for Minkowski signature. Even in Euclidean signature it continues to
be true for the D3-brane solution considered below, that it will be of the form
F5 = A5 + A5
(60)
A dF
1
A1 ...n1 = ( gF 1 ...n1 )
g
(61)
Also
3.3
Equations of motion
The equations of motion for the generic problem (Einstein frame) eq.(52) are (often we shall
write a for an ):
n1 2
1 a
1
2 ...n
nF
F2 ...n
+
e
F
R =
2
2n!
D2 n
14
1
a 2
F
2 = ( g g ) =
g
2n! n
( gea F 2 ...n ) = 0
(62)
where for simplicity we have considered the case with Fn 6= 0 only for one value of n, as will
be the case. The Bianchi identity for Fn is
[1 F2 ...n+1 ] = 0
(63)
(64)
ds = g dz dz = sB dt + C
p
X
(dxi )2 + F 2 dr 2 + G2 r 2 d2d1
(65)
i=1
which is a diagonal metric, the components of which are all functions of the transverse distance coordinate,
d
X
2
r =
(y a )2
a=1
d2d1
only. Also
is the metric on the unit sphere S d1 in the transverse space. There is a gauge
freedom which may be disposed of by putting F = G or G = 1 or something else. We shall
leave it for the time being in order to find a convenient form of the solutions in which we shall
be interested. It is furthermore part of the p-brane ansatz to require, that the metric should
tend to a flat value at r , i.e. that all the coefficients B, C, F, G should tend to 1 in that
limit.
A p + 1 form gauge potential couples naturally to the world volume of the Dp brane. The
resulting p + 2 form field strength tensor is termed electric. However, we shall also need the
magnetic possibility, which is a consequence of an electric/magnetic duality in the problem. In
fact, defining
FDn = ea Fn
(66)
and using the Hodge duality relations of the previous subsection, it is possible to verify that
the equations of motions are invariant under the duality transformations:
a a, n D n, Fn FDn
(67)
1
e1 ...D Fn+1 ...D
(D n)!
1 a 1 ...n
g1 ...D
=
e F
n!
1
2
=
ea FDn
(D n)
1
2
ea (D n)F 2 ...Dn F2 ...Dn FDn
=
(D n)!
gea F 1 ...n =
Fn+1 ...D
1 a 2
e Fn
n!
n a 2 ...n
e F
F2 ...n
n!
15
(68)
Since
(69)
g = BC p F (Gr)d1 d1
(70)
i ...i k(r)
B 2 C 2p F 2 1 p
(71)
(72)
Q
(Gr)d1
Q
(Gr)d1
(73)
F1 ...d1 =
d1 1 ...d1 Q
Z
1
d1 Q
p =
Fd1 =
16GD S d1
16GD
(74)
where p is the density of electric charge on the p-brane. The i = 1, ..., d 1 are indices on
the unit sphere in the transverse space, and d1 is the volume of S d1 :
d+1
d
2n1
2n
2 2
=
)
( d+1
2
n
2
=
(n 1)!
2(2)n
=
(2n 1)!!
(75)
(76)
The magnetic ansatz has n = D (p + 2) = d 1 with the non zero components of the field
strength tensor given by
F1 ...d1 = d1 1 ...d1 Q
(77)
and Q = Q(r). But the equation of motion for Fn is trivially satisfied, whereas the Bianchi
identity requires Q to be a constant. The magnetic charge (density) is
Z
1
d1
gp =
Fd1 =
Q
(78)
16GD S d1
16GD
Now
F 1 ...d1 =
Q
1
1 ...d1
d1
(Gr)2(d1)
(79)
from which
1 2
Q2
Fn =
n!
(Gr)2(d1)
Q2
1
2 ...n
F
F2 ...n =
(n 1)!
(Gr)2(d1)
(80)
with non-vanishing only for indices belonging to the sphere S d1 . The similarity to the
electric case will allow us to cover both possibilities at the same time.
To find the form of the equations of motion for our ansatz, we must work out the Riemann
tensor for the metric. We choose to work via the spin connection, expressed in terms of the
vielbein ea as (flat indices are either small latin letters from the beginning of the alphabet, or
bared Greek letters)
1 c
e (cab + bac + bca )
2
ea eb ( ec ec )
abc , ba = ab
Sab + Kab
ab ab
a c cb a c cb
ab =
abc
bac
Rab
Sab
Kab
=
=
=
=
=
(81)
0 2
ds = s(A0 ) (dz ) +
D1
X
(A )2 (dz )2
=1
17
(82)
1
A
A A
1
A
A
1
= 2 log A ( log A )2 + ( log A )( log A )
A
1
+ 2 log A ( log A )2 + ( log A )( log A )
A
X
1
=
2 ( log A )( log A )
A
6=,
=
S
(83)
When the metric only depends on one coordinate, r, these are the only non-vanishing components (up to symmetries). Also notice that S and K are never simultaneously nonvanishing.
We define f (r) by
f (r)r d1 BC p F 1 (Gr)d1
(84)
Using the above formulas, we may then work out
1
(log B) + (log B) (log(f r d1 ))
2
F
1
= 2 (log C) + (log C) (log(f r d1))
F
1
= 2 (log(F f r d1)) (log F ) (log(F f r d1 )) + ((log B) )2
F
+p((log C) )2 + (d 1)((log Gr))2
1
F2
d1
= 2 (log Gr) + (log Gr) (log(f r )) (d 2) 2 2
F
Gr
Rt t =
Rii
Rrr
(85)
The equations of motion for the metric and the dilaton for the electric case then take the forms
(no summation over indices):
(d 2)ean n! 2
K2
Fn (d 2) 2
2(D 2)
F
2
K
(d 2) 2
F
K2
1
(d 2) 2 +
( )2
F
2F 2
K2
(p + 1) 2
F
an (D 2)K 2
Q2
1
ea F 2
2(D 2)
(Gr)2(d1)
Rt t =
Rii =
Rrr =
R =
+ (log(f r d1 )) =
K2
18
(86)
We see already here, that in order for the metric to reduce to a form similar to AdSq S Dq , the
Riemann tensor in each sub space has to be proportional to the metric tensor, and a necessary
condition therefore is that the dilaton decouples and becomes a constant (in particular, zero).
This requires either n = 5 i.e. IIB string theory and AdS5 S 5 , or M-theory (or 11-dimensional
supergravity) where 2-branes and 5-branes are possible corresponding to AdS4 S 7 or AdS7 S 4 .
3.4
B = f 2 H
i.e.
ds2 = H
d2
2 d2
, C = H
sf dt2 +
d2
p
X
, F = f2 H
(dxi )2
i=1
+ H2
p+1
p+1
, G=H
p+1
, e = H a
f 1 dr 2 + r 2 (dd1 )2
d2
r d2
h
0
, f = 1
H = 1+
r
r
1 2
= (p + 1)(d 2) + a (D 2)
2
2
Q
h2(d2) + r0d2 hd2 =
2(d 2)(D 2)
D2
(87)
Notice that indeed the diagonal metric tensor components tend to 1 as r . In the electric
case
Q
(88)
Fti1 ...ip r = i1 ...ip H 2 d1
r
In the magnetic case the solutions are obtained by the above duality relations eq.(67). For the
5-form in IIB string theory we replace F5 F5 + F5 .
The above solutions are not the most general ones, but represent a 2-parameter sub-family
of solutions. For r0 = 0 we have f 1 and we obtain the extremal solution, depending only
on a single parameter, Q related to the common mass and charge density of the BPS D-brane.
For r0 6= 0 a horizon develops at r = r0 .
Thus we are seeking a 2-parameter solution to be represented in a suitable form by some
convenient gauge choice. Let us try the ansatz
F
B
= cB log f, log
= cF log f
(89)
log
C
G
with cB and cF constants to be sought for. Further define
g C p+1 Gd2 = f
CG
= f 1(cB cF )
BF
(90)
19
(91)
or
(log f ) + ((log f ) )2 +
d1
(log f ) = 0
r
(92)
giving
f +
d1
f = 0
r
f = 1
r d2
0
(93)
since we demand f 1 for r , which is our first result for f (r). From the equations of
motion we further obtain
"
2 #
d
2
F
d
1
=0
(94)
(log f ) +
1
(log g) + (log g) (log f r d1 ) +
r
r
G
d2
(1 f 2cF ) = 0
r
(95)
D2
(96)
3.5
The extremal solution is obtained by putting r0 = 0. It corresponds to the brane being in the
ground state in a quantum description. The non-extremal solution (presumably) represents
excitations, corresponding to a definite temperature. In the extremal case we further consider
the case with no dilaton coupling. We have already identified the relevant cases as
(D, n) = (10, 5), (11, 4), (11, 7)
For these values, fortuitously one has the accidental identity
= (p + 1)(d 2) = 2(D 2)
Q
hd2 =
d2
20
(97)
Q
(d 2)r d2
p
d
X
2
2 X
p+1
i 2
2
d2
(dy a )2
(dx ) ) + H
(sdt +
= H
H = 1+
ds2
i=1
d
X
a=1
a=1
(98)
We now want to consider N coincident branes. For a single extremal Dp brane, the flux as
normalized in eq.(74) is given by [20, 16]
p
p = Tp 16G10
(99)
2
(2s )p+1 gs
(2s )8 2
gs
(100)
16G10 =
2
In fact in 11-dimensional supergravity, almost the same formulas apply, with an obvious change
in dimensionality, but with the understanding that gs is absent (say gs 1 in 11 dimensions).
With this somewhat vulgar notation, we may write
Tp =
2
(2)p+1 gs
(2)D2 2
16GD =
gs
(101)
2
The length is the string length 2s = for D = 10 and the 11-dimensional Planck length for
11-dim. sugra. We therefore insist that we should choose Q so that
p
=N
(102)
Tp 16GD
Tp =
or from eq.(74)
Qd1 (2)p+1gs
=N
15GD
2
or
(103)
(2)d2
(2)d2
, hd2
(104)
=
Ng
s
d
d1
(d 2)d1
The Maldacena-conjecture arises by considering the so-called near horizon limit in which we
consider the region very close to r = 0 and subsequently scale this region up in a singular way
to be described. In this limit we simply have
Q = Ngs
hd2
d
r d2
21
(105)
We see in particular that the r 2 in front of d2d1 will get cancelled so that the metric becomes
a direct product with an S d1 . We now consider the various cases in turn.
D = 10, p + 1 = 4, d = 6 AdS5 S 5
This is the case of N D3 branes in IIB string theory. So according to the above prescription
we take (cf. also eq.(75))
4gs N4s
(106)
H =1+
r4
Also define the scaled variable
U = r/2s
(107)
We consider the limit = 2s 0 and again also r 0 in such a way that U becomes the
meaningful variable:
4gs N
U 4 4s
2
p
dU
U2
2
2
2
+ d5
= s
dx + 4gs N
U2
4gs N 4
2
U2 2
2 dU
d
x
+
L
+ L2 d25
=
4
2
2
L
U
H
ds2
with
dx24
sdt +
3
X
(108)
(dxi )2
i=1
and x a suitably scaled version of the coordinate x. Comparing with eq.(20) we see that we
exactly have the metric of AdS5 (L) S 5 (L) where we have indicated the length parameter,
L = b, of AdS5 and and the radius L of S 5 . When s 0 the metric has to be rescaled to get
a finite result - by removing the 2s overall factor. This is the singular blowing up alluded to
above. The radius parameter is given by
L4 = b4 = 4gs N4s
(109)
r3
r3
(110)
U2
22
(111)
Then
dr = 11 2UdU, r 2 = 611 U 4
2
U2
2
2
2
2/3 dU
2/3
2
ds = 11
dx + 4(N)
+ (N) d4
(N)1/3 6
U2
2
U2 2
2 dU
d
x
+
4L
+ L2 d24
=
4L2 6
U2
(112)
(113)
D = 11, p + 1 = 3, d = 8 AdS4 S 7
This is the case of N M2-branes. Here
ds2 = H 2/3 dx23 + H 1/3 (dr 2 + r 2 d27 )
N611 25 2
N611 25 2
H = 1+
r6
r6
(114)
We introduce
1
U2 =
r
3/2
11
25 2 N
H
311 U 3
(
ds2 = 211
=
corresponding to
U2
dx2 +
(25 2 N)2/3 3
1 2
N
2
4U 2 2 L2 dU 2
d
x3 +
+ L2 d27
L2
4 U2
1/3
dU 2
+4
U2
1 2
N
2
1/3
d27
)
(115)
1
AdS4 ( L) S 7 (L)
2
with
2
L =
4211
1 2
N
2
1/3
(116)
Exercise show that the spaces, AdS5 (L) S 5 (L), AdS7 (2L) S 4 (L) and AdS4 ( 21 L) S 7 (L)
are in fact exact solutions to the equations of motion of the appropriate low energy effective
Lagrangians, in particular, that the last term in eq.(51) does not cause any modifications.
Notice, however, that unlike the brane solutions, these solutions do not become asymptotically
flat.
23
3.5.1
Non-extremal branes
For completeness and later reference we give her also the form of the non-extremal p-brane
solutions eq.(87) in the near horizon approximation and with the same scaled U-variables
as in the extremal case. We have used an independent scaling of the boundary coordinates
t, x1 , ..., xn1 in some cases.
D = 10, p + 1 = 4, d = 6 AdS5 S 5
2
U2
2
2
2 dU
(sf
(U)dt
+
d~
x
)
+
L
+ L2 d25
3
L2
f (U)U 2
p
L2 = 2s 4gs N
4
U0
f (U) = 1
U
ds2 =
(117)
D = 11, p + 1 = 6, d = 5 AdS7 S 4
2
U2
2
2
2 dU
(sf
(U)dt
+
d~
x
)
+
4L
+ L2 d24
5
4L2
f (U)U 2
L2 = 211 (N)2/3
6
U0
f (U) = 1
U
ds2 =
(118)
D = 11, p + 1 = 3, d = 8 AdS4 S 7
L2 dU 2
4U 2
2
2
(sf
(U)dt
+
d~
x
)
+
+ L2 d27
2
L2
4 f (U)U 2
1/3
1 2
2
2
N
L = 411
2
3
U0
f (U) = 1
U
ds2 =
3.6
(119)
From the discussion in the previous section we know that IIB string theory on AdS5 compactified
on S 5 or M-theory on AdS4 compactified on S 7 or finally on AdS7 compactified on S 4 , are
all quantum theories with isometry groups (for Minkowski signature) SO(2, 4), SO(2, 3) and
SO(2, 6) respectively. The remarkable Maldacena conjecture [1] is that these various quantum
theories are exactly mathematically equivalent to (dual to) certain quantum theories on the
boundary of the relevant AdS spaces, i.e. on the coincident branes.
What could these brane theories be? They would have to be conformally invariant quantum
field theories according to the discussion in the previous section. And in particular for the case
24
of AdS5 S 5 the argumentation is perhaps not too far fetched. The important point is that
we have seen that what we are looking at, is a small portion of space-time very close to the
branes, and then subsequently blown up by formally letting 0. But in precisely that limit
we think we know what the effective quantum theory on the N coincident D3-branes should
be [19]: It should be N = 4 super Yang Mills with gauge group U(N). (For references on
D-branes, see [20, 21]).
Let us try a heuristic argumentation: Excitations of D-branes may be thought of in terms
of open strings with end-points on the D-branes. We are dealing with IIB string theory which
is a theory of oriented strings, and there are N different branes for the strings to end on,
so these open strings are automatically equipped with Chan-Paton labels relevant to U(N).
They interact with each other, but in the singular limit, 0 we consider, only the zeromass modes need be considered, and in 10-dimensions they form the N = 1 multiplet of pure
Super Yang-Mills (for gauge group U(N)). Also, as has been known since the middle 1970ies
[22], they interact exactly according to that theory, the gauge coupling being related to the
open string coupling gso and . For a given Feynman-diagram, the corresponding open-stringdiagram will have boundaries, which come with a natural orientation, because the string is
oriented (the string connects two boundaries and is oriented, so the two boundaries are different:
have opposite orientations). Further the different boundaries have labels i = 1, ..., N equal to
the label of the D3-brane. Thus the open string-diagram has an appearance identical to the
(super) Yang-Mills Feynman-diagram in the t Hooft double line representation for U(N). In 10
dimensions such a string theory is anomalous, but here we are considering end points restricted
to the D-branes and there is no such problem.
In our case, however, the zero-mass particles are confined strictly to the 4-dimensional world
volume of the coincident D3 branes. Hence the theory is naturally 10-dim SYM dimensionally
reduced to 4 dimensions, and that is exactly the N = 4 SYM theory mentioned above. The
SYM-coupling gY M is essentially the open string coupling constant which is itself the square
root of the closed IIB string coupling constant. More precisely for N p-branes:
gY2 M
= gs (2s )p3
4
(120)
(121)
Here, for the gauge group U(N), F is treated as an N N matrix, but the det refers only
to the (p + 1) (p + 1) index structure of indices . The interpretation of the determinant
in this case is via the symmetrized trace, denoted by T r above. However, in the calculation
needed here, only an expansion to 2nd order is required, and no ambiguity exists. So for a
D3-brane
gY2 M
= gs ,
4
25
D3-brane
(122)
From the perspective of the 4-dim. field theory, N is the N of the gauge group U(N). From
the perspective of IIB string theory on AdS5 S 5 , N is the flux (normalized carefully as in
eq.(102)) through S 5 .
So, the remarkable conjecture is that IIB quantum string theory on AdS5 compactified on
5
S is identically equivalent to the quantum field theory in 4 dimensions. Let us collect a number
of points in favour of this hypothesis. We shall try to compare properties of the two quantum
theories: (i) IIB string theory on AdS5 S 5 with N units of 5-form flux through S 5 - to be
referred simply as IIB; and (ii) N = 4 super-Yang-Mills with gauge-group U(N) in 4 dimensions
- to be referred to simply as SYM.
First compare global symmetries. The IIB-theory has an isometry group SO(2, 4) SO(6)
with the last SO(6) being the isometry group of the 5-sphere. Actually, because spinors are
involved the relevant groups for AdS5 and the sphere, S 5 are the covering groups SU(4) of
SO(6) and SU(2, 2) of SO(2, 4), so we have SU(2, 2) SU(4). But the 32 Majorana spinor
supercharges of the IIB theory (which are all preserved in this background) transform under
this symmetry in such a way that in fact the full invariance is given by the Lie-supergroup
SU(2, 2|4).
Now we should try to understand that this is also the relevant invariance to consider for the
SYM. We have already understood that the SO(2, 4) or SU(2, 2) part is realized as a conformal
invariance. Indeed the SYM is known to have vanishing beta-function and be conformally
invariant. How about SO(6) (or SU(4))? does the SYM theory know about the 5-sphere? Yes,
indeed. That is the R-symmetry of SYM. In fact, consider briefly the field content of SYM:
In 10 dimensions, N = 1 pure Super Yang-Mills contains the gauge field potentials A , =
0, 1, ..., 9 (the gluons) giving 10 2 = 8 bosonic physical degrees of freedom, all in the adjoint
representation of the U(N). Further we have the 8-dimensional Majorana-Weyl gluinos
, = 1, ..., 8, also all in the adjoint representation. The theory has 16 Majorana supercharges
Q , = 1, ..., 16. Under dimensional reduction, the gluon fields turn into 4 2 = 2 gauge
fields and a remaining 6 scalar fields, 1 , ..., 6 . The gluino fields turn into 4 2 Weyl spinors
in 4 dimensions, A
, = 1, 2, A = 1, 2, 3, 4. An N = 1 description in 4 dimensions put one of
these spinors together with the gauge field in a gauge superfield, and the remaining 3 spinors
each combine with a pair of scalars to give 3 scalar chiral superfields. The 16 supercharges turn
A
into 4 sets of complex Majoranas QA
, Q , = 1, 2, A = 1, 2, 3, 4 transforming as a {4} and
of the R-symmetry group SU(4), and the i transform as a {6} (the fundamental rep.
a {4}
of SO(6) or the antisymmetric rank 2 tensor under SU(4)), so we see that SO(6) (or SU(4))
is indeed present.
But the IIB theory had 32 fermionic supercharges, the SYM only 16. Indeed from the
perspective of the N coincident BPS D3-branes, half the IIB supersymmetries are broken. In
any case, where are the remaining 16 fermionic generators? The answer is that they arise as
part of an extension of the conformal group that takes place when supersymmetry is present
as described in the famous paper by Haag, Lopuzanski and Sohnius [23]. For completeness let
us give here their form of the superconformal algebra:
Define first
P P , K K
(123)
and
1 1
1 2
2 2 M M1 2 1 2 + M
1 2
26
(124)
In addition to the generators of the conformal algebra with commutation relations, eq.(46), one
now has the 16 new fermionic generators
(1)A
Q
obtained as
(1)A
,Q
(1)A
[K , QA
] = 2i Q
(125)
[QA
, D] =
(1)B
{QA
, Q
(1)B
} = AB K
{QA
, Q
} = AB ( D + M ) + i T AB
(1)A
(1)A , T BC ] = 2 AB Q
(1)C 1 BC Q
[Q
(126)
Indeed the combination of eq.(46) and eq.(126) together with the standard N = 4 supersymmetry algebra constitute the Lie super-algebra SU(2, 2|4).
It is well known that the IIB theory (almost certainly) contains a (non-perturbative) SL(2, Z)
invariance [24]. It is best viewed as arising from compactification of M-theory on a 2-torus with
modular parameter
= + ie
with the RR-scalar of IIB (the axion). In N = 4 Super Yang-Mills there is a corresponding
SL(2, Z) invariance of the theory with modular parameter in this case
=
4i
+ 2
2 gY M
In this latter case the symmetry is represented by an SL(2, Z) transformation of the lattice of
electric and magnetic charges in that theory: q + ig = gY M (ne + nm ) by treating (ne , nm ) as
a doublet.
This finishes our very brief comparison of symmetries of the two theories.
3.7
We continue for definiteness to focus on the case of AdS5 S 5 . As we have seen in eq.(109)
the common radius- or length-parameter is given be
b4 = 4s 4gs N = 4s gY2 M N = 4s
with
gY2 M N
27
(127)
the t Hooft coupling relevant to large N Yang-Mills theory. Thus, it is tempting in particular
to consider the limit fixed while N . We see that in this limit the string coupling tends
to zero, so that we may perform calculations on the string theory side, simply by restricting
ourselves to string tree-diagrams, the classical limit of string theory! The full quantum nonperturbative description of N = 4 Super Yang-Mills would be obtained from this classical theory
- in the large N limit. This seems like a program which might have some success eventually,
even though the NS-R formulation of IIB string theory on the AdS5 S 5 background with N
units of RR 5-form flux is unknown (for a preliminary attempt, see [25]). In the Green-Schwarzformulation there is a proposal [26], [27], but non-trivial calculations remain to be performed.
In any case, this looks like a concrete proposal for the so called Master-field idea of Witten,
that the large-N U(N) Yang-Mills theory path integral could be described by a single field
configuration, reminiscent indeed of classical theory.
The situation becomes even more astonishing if we furthermore consider the strong coupling
limit, i.e. large limit of SYM. If we keep the AdS/sphere radii fixed, we are therefore dealing
with the 2s = 0 limit of string theory, or in other words, with the limit in which classical
string theory simply becomes classical supergravity! This is the limit mostly considered in
concrete calculations so far. Even in that extreme limit the conjecture has dramatic predictions:
It predicts how the SYM theory at large N behaves in the extreme non-perturbative, strong
coupling regime. String excitations become infinitely heavy and decouple in that limit, but
since we kept the radius of the sphere S 5 fixed, we cannot at all neglect the Kaluza-Klein
states associated with the compactification on that sphere. (Strictly speaking one cannot send
the dimensionful string length to zero. What one means by this is to consider energy scales
for which string excitations may be neglected. If we consider the string length fixed and still
take large, the radius of S 5 tends to infinity. People often phrase the situation that way.
Then all curvatures are small and quantum gravity corrections may be neglected: classical
supergravity is adequate. KK-masses now become very small. Obviously the situation is entirely
equivalent to our formulation, but as usual one properly has to consider dimensionless ratios
for such arguments to make sense.)
28
4
4.1
(128)
This requires short distance singularities of the correlators to be integrable, or else the introduction of some device to render the expression meaningful. In any case we might always go
back to considering individual correlators. Obviously, if the operators Oi on the boundary CFT
have conformal dimension i then the currents i should have conformal dimension n i .
Similarly any other quantum number the operators may have (say as multiplets of SU(2, 2|4)
in the case of AdS5 S 5 ) would have to be supplemented by conjugate quantum numbers of the
currents so that singletsmay be formed. Supposing the generating functional of the currents
{i (x)} (x AdSn+1 E n ) makes sense, the description of this object in AdSn+1 should
29
be by a path integral with fields i (y) in that theory (i.e. with y AdSn+1 ) tending to the
boundary currents {i(x)} in a certain prescribed way, that we shall have to infer. In the large
N limit we would just have to work out the classical action, and in the large N strong coupling
limit, just the classical supergravity action on fields satisfying the equations of motion and
tending to the prescribed boundary currents in some particular way.
4.2
By a scalar field we mean one which transforms as a scalar under the AdS isometry group, hence
it would tend to a boundary value with conformal dimension zero, and couple to operators with
conformal dimension n. In that case it turns out to be meaningful to simply require the path
integral on AdSn+1 with the scalar field in question tending to a definite value on the boundary
(x) 0 (x )
for x AdSn+1 and x E n and (somehow) x x . Let us study the case of free scalar fields.
Here the path integral becomes trivial, and is simply equal to the exponential of (minus or
i times) the classical action, up to a normalization constant. Thus in this case the classical
approximation is the exact result. The action on AdSn+1 is
Z
1
I() =
dn+1 x g
(129)
2 AdSn+1
So we seek a classical field, satisfying the equation of motion
1
D D (x) = g (x) = 0
g
(130)
throughout AdSn+1 , but such that (x) 0 (x ) whenever the point x in AdSn+1 runs away
to in the particular way that defines the boundary point x E n of the boundary.
It is plausible that the classical solution to this problem is unique. Indeed, imagine 1 (x)
and 2 (x) both being solutions of the equation of motion with the same boundary value. Then
(x) 1 (x) 2 (x) has boundary value zero and also satisfies eq.(130). We should show that
vanishes identically. Since it vanishes at infinity we take it to be square integrable. Then
Z
Z
n+1
(131)
0 = d x gD D = dn+1 x g
Since is positive (semi) definite (in the Euclidean case), we find
0
or constant. But as it tends to zero, it must vanish everywhere.
We may solve the problem in terms of a Greens function, the generalized propagator,
K(x, x )
30
(132)
with x AdSn+1 and x E n (or E 1,n1 in the Minkowski case), or since we have seen the
boundary to be compactified, S n , which we will take to denote the boundary. Thus we seek a
solution to the problem
1
(133)
x gx K(x, x ) = 0
g
and somehow K(x, x ) a delta function when x is on the boundary. Then we may construct the
sought for classical solution as
Z
(x) =
dn x K(x, x )0 (x ), x AdSn+1
(134)
Sn
Following Witten [3], we construct this propagator in the coordinates eq.(21), taking for definiteness Euclidean signature:
n
1 X 2
2
(dx )
(135)
ds = 0 2
(x ) =0
Now AdSn+1 is described by the upper half space x0 > 0, and x x , = 1, ..., n are
coordinates in E n , the coordinates of the boundary. However, we know that we must be
dealing with a compactified boundary S n , and there is an extra point at infinity described
at some length around eq.(27). The boundary here is x0 = 0; the single point at infinity is
x0 = , a single point indeed, no matter what the values of the remaining coordinates are,
since the metric tensor vanishes there.
In these coordinates then
g =
1
1
, g = 0 n+1 , g = (x0 )2
0
2
(x )
(x )
(136)
(x0 )n
((x0 )2 + (~x ~x )2 )n
(137)
satisfies Laplaces equation eq.(133) for x0 6= 0 and ~x 6= ~x , and becomes the desired delta
function in the limit x0 0.
We may also infer the above more elegantly from Wittens trick [3], useful in the sequel. First
notice, that just as the scalar propagator in flat space is Poincare invariant so the propagator in
our case is invariant under the AdS isometry group. Then let the boundary point ~x represent
the point, P at infinity:
K(x; P ) = K(x0 , ~x ; P )
(138)
This cannot depend on ~x due to translation invariance, so in that particular case, K is a
function of x0 only, and Laplaces equation becomes
0 g 0 K(x0 ) = 0
Here
0 =
; 0 = g 00 0 = (x0 )2 0
x0
31
d 0 n+1 d
0
(x
)
K(x
)
=0
dx0
dx0
If we try to solve with K(x0 ) = c(x0 )p we find
(139)
p(n + p) = 0
The solution with p = 0 cannot describe something with delta function support at infinity,
x0 = , since such a delta function should vanish on the boundary x0 = 0 for any ~x . Hence
K(x0 , ~x ; P ) = c(x0 )n
Next apply a transformation to map P ~x = ~0:
x z
x
, = 0, ..., n
(x0 )2 + ~x 2
(140)
This transformation is indeed an SO(1, n + 1) isometry of AdSn+1 (in this case with Euclidean
signature). In fact we find with x2 (x0 )2 + ~x 2
n
X
x2 dx x 2x dx
,
(x
dx
x dx )
dz =
(x2 )2
=0
It follows that
dz 2 =
dx2
dz 2
dx2
=
(x2 )2
(z 0 )2
(x0 )2
(141)
(142)
so that the transformation is an isometry and the parametrization space is preserved. Under
this mapping
x0
x0
(x0 )2 + ~x 2
(x0 )n
((x0 )2 + ~x 2 )n
(143)
(x0 )n
((x0 )2 + (~x ~x )2 )n
(144)
is independent of x0 and convergent, as seen by scaling to the new integration variable xi /x0 , so
that a limit is obtained for x0 0, and we may adjust c to get proper normalization if desired.
32
(145)
(146)
(x0 )n+1 = 0
(147)
Now
Z
Z
1
1
n+1
d x g =
dn+1 x(x0 )n+1
I() =
2
2
Z
Z
n
o
1
1
n+1
0 n+1
d x (x )
dn+1 x (x0 )n+1
=
2
2
(148)
The last term vanishes by the equation of motion, and the total derivative term vanishes in all
directions except in the x0 direction where we have a boundary. To avoid the divergence for
x0 = 0 we put first x0 = . Then
Z
1
I() =
dn x(x0 )n+1 (x0 , ~x )0 (x0 , ~x )
(149)
2 x0 =
In the limit x0 0 we may put (x0 , ~x ) = 0 (~x ). We then evaluate
Z
(x0 )n
0
0 (x , ~x ) = c 0 dn x 0 2
0 (~x )
x
((x ) + (~x ~x )2 )n
Z
1
0 n1
0 (~x ) + O((x0 )n+1 ), x0 0
dn x
= cn(x )
(~x ~x )2n
Inserting into eq.(149) we see that the singular x0 behaviour drops out and we get
Z
0 (~x )0 (~x )
cn
dn xdn x
I() =
2
(~x ~x )2n
(150)
(151)
In the classical (super)gravity limit the generating function for operators O(~x ) in the boundary
theory coupling to the source 0 (~x ), is then given by the exponential of (minus) that. It
follows that in this (trivial) example, there are only connected 2-point functions
hO(~x )O(~x )i
33
1
(~x ~x )2n
(152)
This is the expected result: The SO(1, n + 1)-scalar, (x) will couple to conformal operators
of dimension n on the boundary in order for the coupling
Z
dn xO(~x )0 (~x )
to be invariant. And conformal invariance (indeed dilatation invariance) suffices then to fix the
form of the 2-point function to eq.(152). In the case of AdS5 S 5 this operator will turn out
to be the T r(F 2) of the YM field strength (sect. 4.5).
Notice that the 2-point function eq.(152) is characteristic of a quantum theory with a non
trivial short distance singularity. This is so even though it was derived from a classical calculation in the bulk of AdSn+1 .
4.3
We continue to follow ref.[3]. A gauge field A (x) in AdSn+1 with = 0, 1, ..., n gives rise to a
field strength
F = A A
satisfying the free equation of motion (no currents)
1
( gF ) = 0
g
(153)
Now again we seek to construct, first a propagator with one delta function source on the
boundary. With that we then build any gauge field A (~x ) in the bulk with the property that
the components Ai (x0 , ~x ), i 1 tend to prescribed functions on the boundary, corresponding
a certain 1-form on the boundary
A0 (~x ) = ai (~x )dxi
(154)
As before, we use the trick of first working out the propagator when the point on the boundary
is the point P at infinity. Again, in that case we expect the propagator to be independent of
any ~x . Further, the propagator should be a 1-form, but in that case, one with no 0-component.
The remaining components are all treated the same way. We treat the components in the
boundary one by one.
Thus we look for a 1-form in the bulk of AdSn+1 , only depending on x0 , and with a single
component only, say the ith
A(i) (x) = f (x0 )dxi
It should satisfy the equation of motion, eq.(153). We have
(i)
Ai
= f (x0 ), A(i)
0, 6= i
34
(155)
(156)
(the normalization is for later convenience). We hope this 1-form will have a delta-function
singularity at P . We exhibit this as before using the SO(1, n + 1) isometry (inversion)
x
(x0 )2 + ~x 2
n2
x0
xi
n1
d
n 2 (x0 )2 + ~x 2
(x0 )2 + ~x 2
x
A(i)
(157)
This new propagator represents propagation from ~x = ~0 on the boundary to (x0 , ~x ) in the
bulk. When we work out the derivatives we see that this new 1-form propagator will have
components along all the different dx s. We may simplify, using the fact that the propagator
is unique only up to gauge transformation. We shall in fact get a simpler expression if we
subtract the pure gauge
1
(x0 )n2 xi
d
n2
((x0 )2 + ~x 2 )n1
Then
(i)
=
=
=
=
Or,
n2
x0
xi
1
(x0 )n2 xi
n1
d
d
n 2 (x0 )2 + ~x 2
(x0 )2 + ~x 2
n2
((x0 )2 + ~x 2 )n1
1
xi
(x0 )n2
1
0 n2
(n 1)(x )
+
(n
1)
d
dxi
n2
((x0 )2 + ~x 2 )n2
(x0 )2 + ~x 2
((x0 )2 + ~x 2 )n1
(x0 )n2 xi
d
((x0 )2 + ~x 2 )n1
1
(x0 )n2
1
0 n2 i
(x ) x d
+
(n
1)
dxi
n2
((x0 )2 + ~x 2 )n1
((x0 )2 + ~x 2 )n1
(x0 )n2 xi
d
((x0 )2 + ~x 2 )n1
1
(x0 )n2
1
0 n2 i
i
0 2
d (x ) x + (n 1) 0 2
dx
n2
((x ) + ~x 2 )n1
((x ) + ~x 2 )n1
1
0 n3
0 i
0 n2
i
(x
)
dx
x
+
(x
)
dx
(158)
((x0 )2 + ~x 2 )n1
A(i)
(x0 )n3 xi = 0
1
+(x0 )n2
=i
=
0
2
2
n1
((x ) + ~x )
0
otherwise
35
(159)
We may now collect results and obtain for the general classical solution 1-form field
A (x , ~x ) =
d x
dn x
(i)
n
X
i=1
(x0 )n2
ai (~x )dxi
((x0 )2 + (~x ~x )2 )n1
(x x)i ai (~x )
0 n3
0
(x ) dx
((x0 )2 + (~x ~x )2 )n1
(160)
Notice that only the first term acts as a delta-function for x0 0. In fact, a function of the
form
(161)
(2 + ~x 2 )
is a model of n (~x ) only if 0 < = 2 n. Then, namely
Z
Z
1
n x
+n2
n
d
=
d x 2
2
2
( + ~x )
1 + ~x
is a constant, independent of . For x0 = , the first term in 160 is of that form. The last term
in eq.(160) has an extra power of and vanishes for 0. This is perhaps not completely
obvious. In fact it looks like there is a power less of . But first we must remember to scale also
the factor (x x)i , and second we see by expanding ai (~x ) around ~x that the leading term
vanishes since the integrand is odd, and the following terms do have extra powers of . Hence,
clearly eq.(160) will tend to (up to a constant normalization)
A0 (~x ) =
n
X
i=1
(162)
We are now instructed to evaluate the classical action for the classical solution eq.(160).
In form language
Z
1
I(A) =
F F
(163)
2 AdSn+1
where F = dA, and the equation of motion is d F = 0 i.e. d F = 0. Then
Z
Z
Z
1
1
1
I(A) =
dA F =
d(A F ) =
A F
2 AdSn+1
2 AdSn+1
2 boundary()
(164)
where again we take the boundary to be x0 = at first, and only let 0 at the end. On the
boundary, we only need the i = 1, ..., n components of A F . Thus a component of A, Ai has
i = 1, ..., n. Also F has components j1 , ..., jn1 = 1, ..., n, but never 0 or i. Therefore we need
exactly the components F0i . In coordinates on the boundary
Z
(165)
I dn x hA n0 F0
36
1
ij , i, j = 1, ..., n
(x0 )2
The fact that the metric on the boundary is not uniquely obtained (seems singular) is related
to the fact that the boundary theory as a conformal theory knows of no metric - only of a
conformal class. We shall come back to that. n is an outward pointing unit vector normal to
the boundary. We may take
n = (
1
, 0, ..., 0); n = (x0 , 0, ..., 0)
0
x
ai (x )
(x0 )n3 dxi dx0 dn x n1 + terms with no dx0
D
0 n3
F = (n 2)(x )
dx
i
0 n3
0
n ai (x )dx
F = (n 1)(x ) dx d x
D n1
Z
i
0 n1
0
n ai (x )dx
2(n 1)(x ) dx d x
Dn
Z
x ~x ) d~x ai (x )(xi (x )i )
0 n3
0
n (~
2(n 1)(x ) dx d x
+ ...
Dn
Now
I=
k
k
n
ak (x )(x (x ) )
2(n 1) d x (xi xi )
Dn
+O((x0 )n1 )
37
(166)
(167)
(168)
(169)
We use a notation with xi xi . Only the term with (x0 )n3 survives for x0 0, and we find
Z
ij
2(x x )i (x x )j
n
n
(~x ~x )2n2
(~x ~x )2n
This is the final result. To see that this is in accord with the conjecture, notice that a gaugefield 1-form in AdSn+1 is a scalar under SO(1, n + 1). Hence, the components have conformal
dimension +1 on the boundary, so they couple to operators, Ji , in the conformal field theory
on the boundary with conformal dimension n 1, but these currents must be conserved by
virtue of gauge invariance in the bulk. According to the conjecture we have calculated the
generating function for these operators in eq.(170). We see that they have only non vanishing
2-point functions:
2(xi xi )(xj xj )
1
ij
(171)
hJi (~x )Jj (~x )i
(~x ~x )2(n1)
(~x ~x )2
The last term ensures current conservation:
ix hJi (~x )Jj (~x )i
2(xi xi )(xj xj )
ij
= i
(~x ~x )2(n1)
(~x ~x )2n
= 0
(172)
We conclude that the conjecture also works for free massless gauge fields. Furthermore, we
have constructed a propagator also in that case (in a particular gauge).
The case of massless gravitons in the bulk is a little more complicated [2, 3]. They couple
to the energy momentum tensor on the boundary.
4.4
Following Witten again[3] we shall argue that a massive scalar with mass m in AdSn+1 must
couple to operators O with conformal dimension in the boundary theory, given by the
largest root of
( n) = m2
(173)
Of course we have already checked the case of m = 0. In the massive case it turns out that we
have to reinterpret the idea that the field should tend to a definite (current) field on the
boundary.
We take the free massive theory in the bulk to be described by
Z
1
I() =
dn+1 x g( + m2 2 )
(174)
2
n
X
4
(dx )2
(1 x2 )2 0
38
(175)
with
n
X
x2
(x )2 r 2 , 0 r < 1
y
r = tanh , 0 y <
2
1
dy
, 1 r2 =
dr =
2
2 cosh y/2
cosh2 y/2
y
y
1
r2
= sinh2 cosh2 = sinh2 y
2
2
(1 r )
2
2
4
Next write
n
X
(176)
(177)
(dx )2 = dr 2 + r 2 d2n
with dn the metric of the unit S . Thus the metric on AdSn+1 may be expressed as
ds2 = dy 2 + sinh2 yd2n
(178)
In these coordinates
det g = sinh2n y det
with the metric tensor on S n . Now the Laplacian on scalars becomes
1
= g
g
d
1
d
1
sinhn y +
=
n
sinh y dy
dy
d
d
L2
1
n
sinh
y
=
sinhn y dy
dy sinh2 y
where
L2 =
(179)
(180)
is the Laplacian on the sphere, the angular momentum or centrifugal contribution. (The
notation is perhaps slightly confusing here: In 179
= g = g =
2
sinh y
2 = m
sinhn y dy
dy
sinh y
39
d ny d
(e
) = m2
dy
dy
(181)
(182)
near the boundary. But the form ey is rather arbitrary. In fact, the function
ey f (y, ~x )
has a 1st order zero on the boundary (taking into account the metric on AdSn+1 ). It is therefore just of the form needed to build a finite metric from the divergent AdS one. But this
construction has a degree of arbitrariness about it. Thus, if we transform
f (y, ~x ) = ey
into
ew(~x ) ey f(y, ~x )
that new function has an equally good 1st order zero. Thus, demanding an asymptotic behaviour
+
(y, ~x ) f (y, ~x )
0 (~x )
f ew f 0 ew+ 0
(183)
So, the arbitrariness, f ew f , shows that the metric on AdSn+1 in a natural way only defines
a conformal class of metrics on the boundary: if hij is a metric on the boundary defined by
means of f , then e2w hij is a conformally transformed metric obtained from ew f .
40
This is all consistent with the field theory on the boundary being conformally invariant.
Then namely, the field theory is insensitive to the conformal rescalings of the metric. The field
theory can only conceive of a conformal class.
But then the behaviour
hij e2w hij 0 ew+ 0
is the statement that 0 is not a function, but rather a density with conformal weight + .
Indeed, a density of weight d would have the property that for a small length
()d 0
should be invariant under conformal scaling. That works, since the above is transformed into
(ew )d ew+ 0
which is invariant for d = + .
So the only natural procedure is to require that massive fields should tend to densities 0 ,
and therefore should couple to operators O with conformal dimension n + + so that
O 0
is a density with weight n.
We now verify that this state of affairs is in full accordance with the general prescription.
As before we want to solve for a propagator. We use again coordinates with metric (in units
where the radius b of AdSn+1 has been put equal to 1)
n
1 X 2
ds = 0 2
(dx )
(x ) 0
2
Again we begin with a propagator vanishing on the boundary x0 = 0, but developing a deltafunction at P : x0 = , and thus being independent of ~x . Denoting again the propagator as
K(x0 , ~x ; ~x ) and in particular K(x0 , ~x ; P ) = K(x0 ), the equation of motion is
0 n+1
(x )
d
0 n+1 d
2
(x )
+ m K(x0 ) = 0
0
0
dx
dx
(184)
1
(n + n2 + 4m2 )
2
1
(n n2 + 4m2 )
=
2
+ =
41
(185)
Only the solution K(x0 ) = (x0 )n++ will vanish for x0 = 0. As before the propagator K(x0 , ~x ; ~0)
is found by the inversion
x
x 0 2
(x ) + ~x 2
giving
(x0 )n++
((x0 )2 + ~x 2 )n++
(x0 )n++
K(x0 , ~x ; ~x ) =
((x0 )2 + (~x ~x )2 )n++
K(x0 , ~x ; ~0) =
(186)
Notice that according to the rule eq.(161), this does not tend to a delta-function when x0 0,
rather it is
(x0 )n+2+
((x0 )2 + (~x ~x )2 )n++
which tends to n (~x ~x ) for x0 0. Hence, when we build the classical field as
Z
(x0 )n++
0
(x , ~x ) = c dn x 0 2
0 (~x )
((x ) + (~x ~x )2 )n++
Z
(x0 )n+2+
0 +
0 (~x )
= (x )
c dn x 0 2
((x ) + (~x ~x )2 )n++
(187)
1
(~x ~x )2n+2+
(188)
We now verify that this is indeed what is obtained, using the by now well established prescription.
Namely, we evaluate the classical free action on the classical field as follows:
Z
1
I() =
dn+1 x g( + m2 2 )
2
Z
1
1
1
n+1
2
d x g g ( g m
(189)
=
2
g
g
the last term vanishes, and the first term is evaluated as in the massless case
Z
1
dn x(x0 )n+1 (x0 , ~x )0 (x0 , ~x )
I() =
2 x0 =
42
(190)
Now,
0
0 (~x )
((x0 )2 + (~x ~x )2 )n++
+ non leading terms as x0 0
0 n++ 1
0 (x , ~x ) = c(n + + )(x )
dn x
(191)
(192)
in complete agreement with the expectation eq.(188). Since + is the larger root of
( + n) = m2 ,
is the larger root of ( = n + , = n)
( n) = m2
1
n + n2 + 4m2
=
2
(193)
( n + p)( + p) = m2
p
1
2
2
n + n + 4m 4np
=
2
(195)
It is intuitively plausible how to generalize to fields other than scalars. We saw that the massless
1-form gauge-field A = A dx restricting to Ai dxi on the boundary, naturally had component
fields of dimension 1. Likewise a massless p-form field Cp has component fields with dimension
p and couples to operators on the boundary with dimension n p.
A massive p-form field would couple to operators which would have dimensions shifted as
in the scalar case to
= n + + p
(194)
or
4.5
In ref.[3] Witten makes a check on the Maldacena conjecture in the case of AdS5 S 5 . Similar
checks may be performed in more complicated situations. The check is restricted to the case
of the strong coupling (in the boundary theory) and large N approximation, which may be
treated by classical supergravity. Even though we can neglect string excitations in that limit,
the compactification on S 5 gives rise to Kaluza-Klein excitations with massive modes. We
may use the inverse radius of of S 5 as our unit of mass (as in the preceding subsections), i.e.
continue to put that equal to 1. Thus we should at first analyze the spectrum of KK excitations
(for a general discussion, see [28]). This was done some time ago, in [29] by studying small
fluctuations of the supergravity fields around the AdS5 S 5 background, and in [30] by applying
the powerful technique of (super) group representation theory. These analyses lead to several
43
infinite families of massive field modes with definite masses and transformation properties
under SO(1, 5) SO(6) or even better, under SU(1, 3|4) (SU(2, 2|4) for Minkowski signature).
The representation theory of that supergroup has been considered for example in [30, 31, 32].
According to the Maldacena conjecture, these give rise to predictions concerning the spectrum
of conformal operators in the N = 4 U() boundary theory. For a given set of quantum
numbers (conjugate to the ones for the modes in the bulk theory) we may predict conformal
dimensions using the result of the previous subsection. We must ask whether in fact quantum
corrections would upset the result of such a simple analysis. However, it turns out that both in
the bulk theory and in the boundary theory, there exist large classes of small representations
with properties similar to properties of BPS states and for which such quantum corrections
cannot occur. This makes it possible to perform meaningful checks. In the boundary theory
these are the so called chiral primaries (see for example [33]).
Here we shall not attempt an account of this which is anywhere near complete. Instead
we shall restrict ourselves to analyzing certain aspects of one family, the one corresponding to
KK-excitations of the dilaton field. We shall show, that the masses of these excitations obey
the rule
m2 = k(k + 4), k = 0, 1, 2, ...
(196)
In the boundary conformal field theory Witten pointed out that the corresponding operators
are of the form
O(i1 ,...,ik ) (x) = T r((i1 ik ) F F )(x)
(197)
But only when the symmetrized tensor in {i1 , ..., ik } is taken, do these fields belong to a multiplet
of chiral primaries. Precisely then do the fields transform as an irreducible representation of
SO(6) (see below). The trace is over the adjoint of the U(N) gauge group. The scalar (N N
matrix valued) fields i (x), i = 1, ..., 6 have been mentioned before. They transform in the
{6} vector representation of SO(6). The F is the U(N) field strength matrix written as an
N N matrix. It is trivial to count the conformal dimension in the weak coupling limit where
free field dimensions apply. There the scalars have dimension 1, and the field strength tensor
dimension 2, so O11 ,...,ik (x) has dimension k
k = k + 4
(198)
This fits with the formula of the previous subsection eq.(193) (for k = ) in the case of the
mass values eq.(196). In general the check would not be convincing since we used a strong
coupling argument in the bulk and a weak coupling one on the boundary. But because it may
be shown that we are dealing with the above mentioned small representations, the result will
survive quantum corrections. As emphasized we shall not go into these crucial matters, but
here restrict ourselves to an elementary account of KK-modes of the dilaton field.
Exercise: Consider the case of AdS5 S 5 and consider the KK-mode of the dilaton field which
is independent of the coordinates on S 5 , the s-wave. Supposing it couples indeed to T r(F 2 )
as described, work out the 2-point function of that operator in the supergravity picture, using
the result of sect. 4.2. In particular work out the coefficient in the value of the classical action,
left out in the calculation there, taking into account the integration over S 5 and the scale b.
Verify that the coefficient is a numerical constant times N 2 . Argue that the form of the 2-point
44
function is exactly the expected one for the operator T r F 2 in the large N limit (cf. [2]).
As we have seen before, the dilaton decouples in the background AdS5 S 5 and satisfies the
free 10dimensional equation of motion
1
=0
(199)
g g
g
We may use a splitting of this into the 5 components of the AdS5 and the 5 components of the
S 5 . Thus, if we expand the dilaton field on S 5 in eigen modes of the Laplacian on S 5 , we see
that these eigenvalues will play the role of (minus) m2 -values in AdS5 .
Our first task will be to understand the connection between the Laplacian on S 5 (more
generally S n+1 ) and the quadratic Casimir of SO(6) (SO(n + 2)).
We may think of S n+1 as imbedded in Rn+2 in close analogy to the case of AdSn+1 (indeed
many of the results below carry over to results for SO(2, n), but subtle important differences
exist). Thus define S n+1 by the condition
2
y02 + y12 + ... + yn2 + yn+1
=1
(200)
We are interested in scalar fields defined on S n+1 , but we may trivially extend those to scalar
fields on Rn+2 . In fact, define
n+1
X
2
(y )2
=
=0
Then a scalar field, on S n+1 is only defined for = 1, but we may extend the definition
to Rn+2 by demanding independent of along fixed directions in Rn+2 . More concretely,
introduce coordinates (, x ) similar to eq.(7) by
y0 =
2x
1 x2
,
y
=
, = 1, ..., n + 1
1 + x2
1 + x2
y 2 = 2
(201)
(202)
(203)
Further we have the vielbeins and their relation to the metric on S n+1 (the y m ym s are the
flat coordinates, the set (, x ) (x0 , x ) are the curvilinear ones).
45
e0m =
em
=
0
ym em =
em em =
e0n en0 =
ym
x
=
,
e
=
m
y m
y m
m
m
y
y
y m
=
, em
=
x
0, ym em0 = 2 , en en0 = 0
g
yn y n
=1
(204)
(205)
(use: dn+2 y(y 2 1) = dn+1xd g( 1)) and the statement about the connection between
the Laplacian and the quadratic Casimir
1
Cn+2 Lmn Lmn
2
(207)
follows. Hence we work out (the distinction between lower and upper indices has no significance)
mn
m n
n m
Lmn L = 2 ym n y ym n y
x
+
= 2 ym
y n x y n
m x
(m n)
+
y
yn x
yn
= 2 ym (en + e0n )(y)
y m(en + e0n )(y) (m n)
= 2 y 2 (g + e0n en0 ( )2 ) ym e0m e0n y n
(208)
But here we have made the choice that 0, so we simply get
Lmn Lmn = 22 g
(209)
as we wanted1 .
1
It is easy to verify, that putting instead (, x) = N (1, x) for any N , would yield exactly the same result.
46
Thus we are let to perform an analysis of the dilaton field on S n+1 similar to the analysis
of scalars on S 2 according to spherical harmonics Ym (, ). In this latter case the Laplacian is
well known to be identified with (minus) the square of the angular momentum with eigenvalues
( + 1) for integer . Of course the square of the angular momentum is just the Casimir of
SO(3). We want to arrive at a similar understanding in general.
Let us single out as special coordinate
p
u = y0 + iyn+1 Y ei , Y, R, Y = (y 0)2 + (y n+1)2
(210)
On S n+1 we have
y02
2
yn+1
n
X
1
z =
n
X
yi2 Y 2 + z 2 = 1
(211)
~ n+2i
In general a representation of SO(n+2) would be characterized by a highest weight state, |
~ n+2 of SO(n + 2), a vector in weight space, the components of
with a certain highest weight,
which are eigenvalues of a mutually commuting set of Cartan generators of the algebra so(n+2).
Let us take one of these to be
H L0,n+1
(212)
corresponding to rotations in the complex u-plane, and generating an SO(2) subgroup. The
remaining Cartan generators pertain to an SO(n) subgroup commuting with that SO(2). With
this choice our weights are then automatically labelled by (i) the eigenvalue of H and (ii) by a
highest weight of that SO(n), i.e. we classify irreducible representations according to the
SO(2) SO(n)
subgroup. And we write
~ n+2 = (k,
~ n)
(213)
(214)
(215)
X
1
Lmn Lmn =
Lij Lij + L0i L0i + Ln+1,i Ln+1,i + L0,n+1 L0,n+1
2
1i<jn
(216)
(217)
(218)
(219)
and the Casimir has the value above with k = 1 and Cn = 0. Next consider the k-fold tensor
product of the vector representation with k a positive integer. The highest weight is trivially
(k, ~0) but the representation is highly reducible. The unique irreducible representation with
the same highest weight is the symmetrized tensor product. This is the one we encountered in
eq.(197), and we see now that the Casimir of the representation is given by k(k+n), k = 0, 1, 2, ...
(generalizing the result k(k + 1) for n + 2 = 3).
We now consider the generalized spherical harmonics of the scalar (dilaton) field on S n+1 .
In the case of SO(3) these are the usual spherical harmonics Ym (, ) with integer. They
constitute a complete set of scalar functions on S 2 . They carry irreducible representations of
SO(3) with the highest weight member
Y (, ) = N sin ei
(220)
(221)
where we used the definitions in eq.(210) and eq.(211). This is a regular well defined function
on S n+1 only for positive integer values of k. Of course the similarity with the elementary case
is strong (the normalization would depend on n). It is trivial to verify with the form of the
48
generators we have given, that this field is indeed a highest weight state of SO(n + 2) with
highest weight
~ n+1 = (k,
~ n = ~0)
(222)
just as for the symmetrized k-fold tensor representation. Therefore, constructing a representation of SO(n + 2) by rotating this Yk (
y ) in all possible ways, or, equivalently by forming all
possible linear combinations of fields obtained from it by applying lowering operators, we shall
get a finite dimensional irreducible representation of SO(n + 2), in fact precisely the one we
met above by considering symmetrized tensor products.
In the present case of SO(n + 2) there are of course very many more irreducible representations to worry about than for the case of SO(3), and we might wonder if the generalized
scalar spherical harmonics defined by eq.(221) (together with all of its multiplet members),
would really form a complete set of functions on S n+1 for n > 2. This will be so, if we are able
to construct arbitrarily good approximations to delta functions with support at any point on
S n+1 , using (linear combinations of) these functions. It is actually intuitively obvious that this
should be possible. In fact, consider what we may do with the highest weight functions Yk (
y)
themselves, directly. For very high values of k it follows from eq.(221) and the sphere condition
2
y02 + yn+1
= 1 ~y 2 1
that these functions only have appreciable support in the neighbourhood of the unit circle
2
y02 + yn+1
= 1, ~y = ~0
(223)
49
Breaking SUSY and conformal invariance in the boundary theory. A possible new approach to (large N ) QCD
There are of course then (KK-like) excitations with masses n/R. For fermions, we have the
option of considering a non trivial spin structure. Clearly, if we also take the fermions periodic,
they will be quantized with the same integer modes as the bosons, and supersymmetry will be
preserved. But if we take fermions to be anti-periodic around Euclidean time, they have modes
1
(n + )/R
2
In particular, the lowest mode n = 0 is very different for bosons and for fermions. For bosons
we have a massless mode, but for fermions the lowest mode can be considered to decouple for
high enough temperatures. So supersymmetry is broken. If we investigate the theory with very
50
high frequencies, much higher than the temperature, these details are irrelevant and we expect
to regain a supersymmetric situation.
The scalar supersymmetric partners of fermions will get masses due to renormalization. In
the supersymmetric theory the masses are protected from being divergent. Thus in the effective
low energy theory, we shall have divergencies cut off by the cut off of the effective theory,
which is the temperature. Thus we also expect scalar super partners of fermions to become
massive and therefore to decouple at low energies. Finally, the effective theory will have a
non-vanishing beta function, because some of the field modes which make the beta function
vanish in the full theory are absent in the low energy effective theory. Thus, the effective low
energy theory is no longer conformal.
It follows, that apparently we have a scheme for dealing with a realistic (QCD like) theory. In
that theory, the coupling constant should run at high energies to a small value, since the theory is
asymptotically free. The smallest value is the one attained at the cut off, the temperature, and it
would be given by the fixed coupling constant of the unbroken theory. Therefore, in order to use
this scheme in a fully realistic way, we should arrange for the coupling constant of the boundary
theory to be small. We have previously seen, that the very simple supergravity approximation
is obtained when the coupling constant of the boundary theory is large. Therefore, a straight
forward study of large N QCD in 3 or 4 dimensions based on supergravity, cannot be hoped
to be realistic. It is perhaps similar to a strong coupling analysis of lattice gauge theory,
known to be in a wrong phase. What we would rather like to do, would be to study not
supergravity, but the full string theory in the appropriate background. At large N it should
even be enough to study classical string theory - or the tree diagram limit only, in order to
have a realistic framework for large N QCD. This goal has not yet been achieved, although
preliminary proposals have been given [26, 27, 25].
Instead, a large number of studies have been performed in the supergravity approximation.
Such studies can at best be considered to have an exploratory nature, but it is very instructive
to see how several expectations from confinement are brought out in a very simple way (see
refs. [36, 10, 37, 38, 39, 40, 41, 42, 43, 5], to name but a few).
5.1
According to the previous discussion, we are led to consider finite temperature versions of
Anti de Sitter spaces. As shown by Hawking and Page [44] and generalized by Witten [10]
there are two relevant manifolds denoted X1 and X2 which we must consider.
5.1.1
The manifold X1
The first one is the simplest. In the standard case of Poincare isometry, we may arrange for a
finite temperature by taking Euclidean signature, and by compactifying the time on S 1 . That
compactification may be thought of as a periodic identification of Euclidean time. All bosonic
fields would have to be periodic under: t t + 2R. This mapping is a translation in time,
: t 7 (t), and the set of all translations, n , form a group isomorphic to Z. Thus in the case
51
+
of AdSn+1
we may try to do something similar. Consider the embedding condition:
uv
n
X
x2i = b2
(225)
i=1
+
with AdSn+1
being the branch u, v > 0. Now introduce a real, positive parameter (analogous
to R and to be related to the temperature), and define the mapping
u 1 u, v v, xi xi
(226)
This is a mapping, f of AdSn+1 onto itself. The set of all repeated applications of this
mapping (an the inverses) {fn |n Z} constitute a group isomorphic to Z. Generalizing the
case of Poincare isometry we may consider the manifold
+
X1 AdSn+1
/Z
So X1 is the set of equivalence classes where two points P, P in AdSn+1 are equivalent if
they are related by one of these mappings. We see that a fundamental domain for v is
1 v/b
(227)
namely v/b = 1 and v/b = is the same point in X1 . These points parametrize indeed a circle
v/b = /2
2 ln v/b
=
ln
(228)
and use for X1 the coordinates (x1 , ..., xn ) together with the angular coordinate . It follows
that topologically
X1 R n S 1
2
xi /b = 1
(229)
These spheres have two different radii, but only the ratio is relevant for the conformal structure. We shall come back to that.
Let us introduce a convenient metric on X1 . Write
r
n
X
x2i
i=1
n
X
dx2i = dr 2 + r 2 d2n1
(230)
i=1
and define
t = ln v/b
Then work out
Then find
1
1
ln 1 + (r/b)2 = ln
ln 1 + (r/b)2
2
2 2
X
ds2 = dudv
dx2i
2rdr
X
dv
u = (b2 +
x2i )/v du = 2 b2 + r 2 +
v
v
r
dr
dv
b2
dt =
2
v
1 + r/b
r 2 dr 2
b2 + r 2
b2 dr 2
+ r 2 d2n1 , or
= dt2 (b2 + r 2 ) + 2
b + r2
dr 2
= ds2 /b2 = dt2 1 + (r/b)2 +
+ (r/b)2 d2n1
2
1 + (r/b)
(231)
(232)
(233)
(we performed as usual the mostly minus to mostly plus operation). This is our final metric
on X1 .
5.1.2
The manifold X2
This is the case where a temperature is introduced by in fact inserting a black hole into AdSn+1 ,
the Schwarzschild metric generalized to AdSn+1 . Thus we seek a static, spherically symmetric
metric of the general form
ds2 = A(r)dt2 + B(r)dr 2 + r 2 d2n1
(234)
with r = 0 being the position of the black hole. Outside the black hole we have Einsteins
empty space equation (with a cosmological constant)
R =
n
g , D = n + 1
b2
53
(235)
Thus the black hole solution X2 will also be an example of an Einstein space, but one less
symmetric than AdSn+1 , even though the two are asymptotically similar.
We find the following non zero Christoffel symbols (ij denotes the metric on S n1 ):
1 1 r
1
1
A A , tt = B 1 A , rrr = B 1 B
2
2
2
1
= B 1 rij , ijk , irj = ji
r
trt =
rij
(236)
We then find the non vanishing Riemann tensor components contributing to Rtt :
1
1
1
(AB)1 (A )2 B 2 A B (B 1 A )
4
4
2
n 1 1
=
B A
2r
r
Rtrt
=
i
Rtit
(237)
n
(A )2 n 1 A A B 1 A
= 2A
2
4AB
2r B
4B
2 B
b
1
1
1
t
Rrtr
= (A1 A ) A2 (A )2 + (AB)1 A B
2
4
4
n 1 1
i
B B
Rrir =
2r
(238)
(239)
A B n 1 B (A )2 1 A
n
+
= 2B
2
4AB
2r B
4A
2 A
b
(240)
B A1 f ; B = A2 A f + A1 f
Then the two equations become
1
n
1 1
1 n 1
A f
= 2A
A + A + (f ) A
2
r
2
b
1
n
n
1
1 1 n 1
= 2 A1 f
A +
A + f 1 f A1 A +
A
2
r
4
2r
b
(241)
or
n 1
1
2n
A + A + (f 1 ) A = 2
r
2
b
1
n 1
n
1
2n
A + A + (f 1 ) A +
A = 2
f 1
r
2
r
b
f 1
54
(242)
Subtracting, we find
(f 1 ) = 0
so f is a constant, and
A +
n1
f
A = 2n 2
r
b
(243)
(244)
Redefining the scale of t and the meaning of c1 , c2 , we may put f = 1. Also, to get the empty
solution for X1 as a special case, we put the new c2 = 1. Thus we finally have the Schwarzschild
solution in AdSn+1 for X2 :
ds2 =
r2
b2
+1
wn M 2
dr 2
dt
+
r2
r n2
+1
b2
wn
wn M
r n2
+ r 2 d2n1
16GN
(n 1)n1
(245)
(246)
This will turn out to mean that M becomes the mass of the black hole.
Notice, that this of course generalizes the metric of the standard b = , 4-dimensional
black hole (Euclidean, n = 3, rg = 2GN M = w3 M):
rg 2
dr 2
ds2 = 1
dt +
+ r 2 d2n1
r
1 rrg
5.1.3
We expect that a black hole will have a (Beckenstein-Hawking) temperature, and that of course
was our motivation for picking this metric. Let us work out what the temperature is. The metric
eq.(245) is of the form
dr 2
ds2 = V (r)dt2 +
+ r 2 d2n1
(247)
V (r)
where
wn M
r2
+
1
(248)
b2
r n2
and vanishes at various values of r, the largest of which we shall denote, r+ . This is characteristic
of a metric with a horizon: g00 vanishes and grr has a pole at the horizon, r = r+ . Ordinary
physical space is the region r r+ . In our case of a diagonal metric, g rr = 1/grr . It is rather
V (r) =
55
easy to establish the general formula [41] for the temperature, T of the black hole in the case
of such a diagonal metric:
d
g00 |r=r+
2T = g rr
(249)
dr
(other equivalent formulas are easily obtained). To see this, notice that the horizon property
implies that near r = r+
g00 A(r r+ ), g rr B(r r+ )
(250)
and the metric takes the following form in the vicinity of the horizon
ds2
dr 2
+ dt2 A(r r+ ) + r 2 d2n1
B(r r+ )
(251)
This metric has a coordinate singularity at the horizon, which we may remove by a suitable
choice of coordinates, but only if the Euclidean time is periodic with a particular period, which
one then identifies with the inverse temperature. In fact [44] the coordinate singularity may be
made analogous to a standard 2-dimensional polar coordinate singularity
ds2 = d2 + 2 d2
(252)
where is the polar distance and is the polar angle. At = 0 the metric is singular, but
we know very well that the geometry is regular provided the polar angle has period 2. If the
periodicity is anything else than 2 there is a genuine geometrical conical singularity.
We may arrange for the metric eq.(251) to look similar to eq.(252) if we put
d2 =
2
d
1
dr 2
, =
,
=p
r r+
B(r r+ ) dr
B
B(r r+ )
(253)
where we have chosen to vanish at the horizon. Then the first two terms of eq.(251) become
ds2 d2 + 2
AB 2
dt
4
(254)
We see that this metric describes a regular geometry provided the Euclidean time t is periodic
with period
1
4
= =
T
AB
From this eq.(249) is easily obtained. We also see that only r r+ is relevant.
In our case, we have near r = r+
2r+ (n 2)wn M
+
n1
b2
r+
2
(n 2)b2 + nr+
r2
wn M
0 = 1 + +2 n2 V (r+ ) =
b
r+ b2
r+
(255)
4r+ b2
2
nr+
+ (n 2)b2
56
(256)
(257)
r2
b2
wn M 2
dr 2
+ r 2 d2n1
dt
+
r2
wn M
r n2
rn2
b2
(258)
Remarkably, this form of the X2 metric makes it completely equivalent to the non-extremal
brane solutions eq.(117),
P eq.(118), eq.(119). The relation between the two forms is in all cases
U = r and d2n1 1n1 d~x 2 after a suitable scaling (see below). The relations between the
pairs of parameters (L, U0 ) and (b, M) are in the relevant cases:
D = 10, n = 4
D = 11, n = 6
D = 11, n = 3
U04
= w4 M
L2
U06
1
= w6 M
L = b,
2
4L2
4U03
L = 2b,
= w3 M
L2
L = b,
(259)
Although the solutions are expressed in terms of two parameters, one may in fact bee scaled
away (see below). Notice that the p-brane solutions and the black hole solution have totally
different symmetries and asymptotic behaviours. Only in the limits considered here (near
horizon and large mass) do they agree.
In the same large M limit we have
0
4b2
4b2
=
(wn b2 M)1/n
nr+
n
(260)
r 2 2 b2 2
dt + 2 dr + r 2 d2n1
b2
r
At these asymptotically large values of r the metric on S 1 is
ds2
(261)
r2 2
dt
b2
corresponding to a radius of
On the other hand, the radius of the S n1
ratio of the two is therefore
r 0
b 2
(the r 2 d2n1 term in the metric) is simply r. The
0
2b
If we want this ratio to be small, we see that we need 0 small.
We now scale as follows
w M 1/n
w M 1/n
n
n
, t =
r =
n2
n2
b
b
w M 2/n 2
bn2
n
V
bn2
b2
n2
2
bn2 2 2
bn2 1 2 wn M 2/n 2 2
d
+
d +
d
ds2 =
b2
n2
b2
n2
bn2
(262)
Notice that now the radius of S n1 is of order M 2/n , so that indeed we have managed to
replace S n1 effectively by Rn1 as far as coordinates are concerned. The period of is
w M 1/n
w M 1/n
4b2
n
n
=
0
bn2
bn2
n(wn b2 M)1/n
4b
1
(263)
=
n
and we may write
2
2
bn2
bn2 1 2
ds2 = 2 n2 d 2 + 2 n2
d + 2 dx2n1
(264)
b
with the boundary being S 1 (1 ) Rn1 . Notice that in these scaled coordinates, the metric is
characterized by just one parameter, b; any explicit reference to M has disappeared.
5.1.4
In the classical supergravity limit, the CFT partition function should be evaluated as
eI = eI(X1 ) + eI(X2 ) = eI(X1 ) 1 + eI
= eI(X2 ) eI + 1
I I(X2 ) I(X1 )
58
(265)
We shall find
n1
n+1
n+1 b2 r+
r+
2
(266)
4GN nr+ + (n 2)b2
We see then that for small r+ , I > 0 and tends to zero when r+ tends to zero. For large r+
on the other hand I < 0. Thus we have the following situation
I =
small r+ : eI 2eI(X1 )
large r+ : eI eI(X2 )
(267)
n1
n+1
n1 b2 r+
r+
2
4GN nr+
+ (n 2)b2
n1
n1 r+
I(X1 )
I(X1 )
4GN n
(268)
It follows that
eI(X1 ) + eI(X2 ) |r+ 0 eI(X1 ) + eI(X2 ) |r+
(269)
All of that is a consequence of the formula for I eq.(266). We now justify that formula.
The relevant part of the action is
Z
1
n(n 1)
n+1
I=
d x g R+
(270)
16GN
b2
This is clearly badly divergent. In fact the equations of motion give
n(n + 1)
n
g R =
(271)
2
b
b2
so that
Z
1 n
n
n+1
d
Vn+1
(272)
I=
g
=
8GN b2
8GN b2
i.e. a volume divergence. We may hope however, to make sense of the difference, I, between I
evaluated for the two manifolds X1 and X2 . This requires that we arrange for the two manifolds
somehow to be asymptotically identical in their geometrical respect. Now we have for both
ds2 = V dt2 + V 1 dr 2 + r 2 d2 g = r n1
(273)
R =
being the determinant of the metric on the sphere S n1 . V is given by eq.(248) for the
geometry of X2 and by V = 1 + (r/b)2 for the geometry of X1 , eq.(233) To regulate the
integrals, we terminate the integral over r at some large R:
Z (X1 ) Z R
Volume(X1 ) V (X1 ) =
dt
drr n1n1 , 0 r R
0
0
Z 0 Z R
V (X2 ) =
dt
drr n1n1 , r+ r R
(274)
0
r+
59
So we must choose (X1 ) such that the geometry at the R-surface is the same in the two cases,
in particular that the circumference of the S 1 is the same for the two manifolds:
r
R2
for X1 :
1 + 2 (X1 )
b
r
R 2 wn M
1 + 2 n2 0
for X2 :
b
R
1
2
wn M
R2 2 Rn
R
1 + 2 n2 / 1 + 2
V (X1 ) =
0 n1
b
R
b
n
n
wn M R 0
n1 , R
1
2Rn
n
n
R n r+
V (X2 ) = 0
n1
n
n
r+
n1 0 wn M
I = I(X2 ) I(X1 )
(275)
2
8GN
2
b
independent of R. But
2
r+
wn M
+1
=
n2
b2
r+
so
I =
n1
n+1
n+1 b2 r+
r+
2
4GN nr+
+ (n 2)b2
I r+
I
=
0
r+ 0
(276)
60
(278)
which of course is the very famous result generalized from the flat case of b to the present
case of AdSn+1 . Notice incidentally that this kind of calculation does not work in flat space
directly since the classical action vanishes there.
Exercise: Verify the expressions given for the mass and entropy of the black hole.
5.2
We have emphasized that the supergravity approximation is not satisfactory for a serious study
of large N QCD, even supposing the Maldacena conjecture works, and that no unpleasant
phenomena corrupt the set up proposed by Witten. Nevertheless it is very instructive to see how
this approximation very naturally gives a picture in qualitative agreement with expectations
in hadronic physics. Thus the area law for Wilson loops, associated with confinement, comes
out and goes away appropriately at high (physical) temperatures. Several other aspects have
been treated [36, 40, 43]. Here, however, we shall concentrate on just one of these aspects,
generation of a mass gap [3, 10, 39, 37, 45, 38, 42, 46, 47].
When we use the scheme described above in the case of AdS5 S 5 , compactifying Euclidean
time, we expect (as argued) in the boundary theory to obtain an effective (Euclidean) version of
QCD3 at energies much below the temperature cut-off. In the next subsection we shall briefly
indicate the framework for also dealing with QCD4 . In both cases, the hadrons we expect to
find will be stable glueballs, in particular with a minimal positive mass greater than zero: a
mass gap. This then is a non-trivial confinement effect, not visible in classical YM-theory, nor
in perturbation theory.
How would we identify such a mass gap? We would need in the boundary theory to consider
a gauge invariant operator which could create a glueball, the simplest example perhaps being
O(~x ) = trF 2 (~x )
N
X
i,j=1
ij
ji
(~x ) ~x R3
(279)
P 2 a a
with (F )ij = N
a=1 F Tij being the U(N) Yang Mills field strength expressed as an N N
matrix. We should then work out a 2-point function
hO(~x )O(~0)i exp{|~x |m}, |~x |
(280)
and look for the indicated exponential decay at large distances, dominated by the minimal mass,
m > 0, the mass gap. Non leading exponential terms would correspond to gluon excitations.
To perform the calculation in the bulk theory, we would have to identify the field, (y), y
bulk, to which this operator couples. But we have already indicated in sect. 4.5.1 that the
expected candidate is the dilaton field with no KK-excitations on S 5 , the S 5 s-wave, or k = 0
mode [10]. We might then perform a classical supergravity calculation of the action in the AdS5
background with the black hole and with prescribed boundary conditions on the dilaton field.
However, [3, 10] it seems technically simpler to adopt a slightly different point of view. To
see how, begin noticing that (according to the Maldacena conjecture) the two quantum theories,
the bulk theory and the boundary theory, are entirely equivalent. They have the same Hilbert
61
space, the same operators and correlators, only the physical interpretation of operators differ
drastically in the two theories. First consider the boundary theory. At N , glueballs are
free particles. When moving according to plane waves
~
eik~x
their mass is given by
m2 = ~k 2 ,
(281)
(so only imaginary momenta correspond to being on the mass shell, as is usual in Euclidean
space-time). In the Hilbert space there is an operator describing translations in the boundary
theory, and ~k is the eigenvalue of that.
In the bulk theory, the very same operators occur, just with different interpretation. We
need to think of the Hilbert space of the dilaton theory. First recall the situation for a free
scalar in flat space-time. It satisfies the KG equation
( 2 m2 ) = 0
To build the Hilbert space (in this case the Fock space) one finds modes, solutions of the
KG equation, such as eipx , and build field operators as sums over modes, each mode being
multiplied by a creation or annihilation operator. States in the Hilbert space are spanned by
multi-particle states created by the creation operators. Actually, although we often consider
plane waves, we really need to put them in a quantization volume, or better in fact, consider
wave packets, square integrable modes for which
Z
(mode)2 <
space
We wish to imitate this procedure in the present case of a modified AdS5 background with a
black hole in it, the manifold X2 . Thus we must
1. Formulate the equation of motion on X2 ,
2. find the square integrable modes,
3. interpret those in terms of possible m2 (glueball) values.
We follow here the treatments in [10, 37, 45, 39]. In the case of AdS5 S 5 we have
q
p
2
2
2
b = s 4gs N = s gY2 M N
(282)
2
1 2
d2
2 2
ds2
d
+
dx3 + d25
+
=
2
4
b
b4
2
b
2s 4gs N
2 2
62
(283)
+
b4
b2
2
2
2
(284)
(285)
[ g g ] = 0
(286)
We now wish to look for modes that are (i) square integrable on X2 , and (ii) correspond to a
definite momentum in the boundary theory. Thus we consider the ansatz
~
(, ~x ) = f ()eik~x
(287)
b4 1 2 b4 ij
, g = 2 , g = ij 2 , g = 3
(288)
g 00 = 2 2
h
i
5
4
b f () k 2 f () = 0 or
i
h
1 d
4
4
b f () = m2 f ()
d
Put
x = 2 ,
(289)
d
dx d
d
=
= 2
d
d dx
dx
and obtain
d2 f
2
4 df
+
4(3x
b
) k2 f = 0
(290)
dx2
dx
and of course we may scale b away and replace it by 1. This ordinary differential equation is
the equation of motion. It has solutions for any values of ~k 2 , but now we must understand
the additional information coming from boundary conditions and square integrability. These
will imply that for generic ~k 2 there is no acceptable solution, only for a particular spectrum of
strictly positive ~k 2 values do such solutions exist.
4x(x2 b4 )
63
5.2.1
Boundary condition
Putting b = 1 the metric eq.(284) becomes (we write for in the following):
1 1 2
dx2
1 2
1 2
ds2 x
d + ... =
d + ...
d + x
+
x
x
x
4(x2 1)
x
(291)
As before there is a coordinate singularity at the horizon x = 1 which we wish to cast into the
form of a 2-dimensional polar coordinate singularity
dx2
(z is polar radius)
4(x2 1)
1
dz
1
=
, z = cosh1 x, x = cosh 2z
2
dx
2
2 x 1
dz 2 =
(292)
and
1 sinh2 2z
=
4z 2 near z = 0
x
cosh 2z
(293)
ds2 dz 2 + 4z 2 d 2 + ...
near the horizon, showing that we have polar-like coordinates with polar distance z. The
function f () above is then a function of the polar distance only, not of the angle . The
proper boundary condition for such a function to be smooth at the origin is therefore
df
=0
dz
But
df
dx df
df
df
=
= 2 sinh 2z
4z
dz
dz dx
dx
dx
near z 0. So we merely want to ensure that f is regular at x = 1.
5.2.2
Square integrability
But
1
gd = 3 d = 2 d = xdx
2
Hence we should demand, that if f is inverse power bounded, by f O(xa ), then a > 1.
64
5.2.3
(294)
y1 (x)
X
1
1
a
= 2+
n n+2
x
x
n=1
y2 (x) =
X
p2
1
y1 (x) log x +
b
n n
2
x
n=1
(295)
= 1+
X
n=1
a1n (x 1)n
X
n=1
b1n (x 1)n
(296)
Inserting these expansions into the differential equation, one finds recursion relations for the
unknown coefficients with simple unique solutions.
Any solution of the differential equation may then be represented, either as a linear combination of y1 (x) and y2 (x), or a linear combination of y11 (x) and y21 (x). The series expansions
only converge in I() or I(1) respectively, but the solutions may be uniquely analytically continued to the entire complex plane (with the exception of the singularities at x = 0, 1). In the
overlap (containing 1 < x < 2), they may furthermore be directly compared. It is now clear
that y1 (x) but not y2 (x) satisfies the square integrability condition. Also, only y11 (x), but not
y21 (x) satisfies the boundary condition at x = 1.
We may therefore assert, that an acceptable solution is expressed in the yi basis simply as
c y1 (x)
65
and such a solution may be built for any p. However, when analytically continued to I(1), it
will generically be a expressed as a linear combination
1 y11 (x) + 2 y21 (x)
with 2 6= 0, and therefore be unacceptable from the point of view of the behaviour near x = 1.
Similarly, we may assert, that any acceptable solution is expressed in the yi1 basis simply as
c y11(x)
and of course also such a solution may be built for any p, but when analytically continued to
I() it will generically be expressed as a linear combination
1 y1 (x) + 2 y2 (x)
with 2 6= 0, and therefore be unacceptable from the point of view of square integrability.
It follows, that if we restrict ourselves to the overlap 1 < x < 2, we are able to require the
condition
y1 (x) y11 (x)
(297)
and this condition is only satisfied for certain discrete values of p. These represent the sought
for spectrum of mass values. More concretely, we may pick a value such as x = x0 = 3/2
I(1) I(), and require the existence of a common constant, c such that
y1 (x0 ) = cy11(x0 )
(y1 ) (x0 ) = c(y11) (x0 )
(298)
This will ensure that eq.(297) is satisfied throughout, since we are dealing with a 2-dimensional
set of solutions. These conditions are of course just equivalent to the Wronski condition
y1 (x0 )
y11(x0 )
(299)
(y1 ) (x0 ) (y11 ) (x0 ) = 0
Here we should think of yia (x0 ) and (yia ) (x0 ) (a = , 1) as functions of p (for fixed x0 ). Indeed
the coefficients aan are determined uniquely as polynomials in p of order n. Truncating the series
expansions at some high cut-off, eq.(299) becomes the condition for zeros of a certain high order
polynomial in p. As the summation cut-off is taken higher and higher, the numerically smaller
zeros of the polynomial rather quickly converge, whereas the larger zeros take more terms to
stabilize.
Concretely and by way of illustration, it is easy from the differential equation to establish
1
the following recursion relations (using the notation an<0 0, a
0 = 1 = a0 ):
a
n+1
a1n+1
1
=
a (n + 1) + pan
(n + 2)2 1 n1
1
1
2
1
a (n 1) + (3n(n + 1) p)an
=
2(n + 1)2 n1
66
(300)
Thus
p p2 + 12
a
=
, a =
, ...
1
3 2
24
p(p 6)
p
, ...
a11 = , a12 =
2
16
The glueball masses thus determined may be shown to pertain to J P C = 0++ states. Numerically one finds (in units of 1/b) the following strictly positive mass values
11.6, 34.5, 69.0, 114.9, ..
(close, but not equal to the values, 6n(n + 1) [37, 45]). Similar studies have been made for other
J P C quantum numbers. Comparisons with results from lattice calculations indicate reasonable
agreement for the mass ratios.
5.3
On QCD in 4 dimensions
Witten [10] has also explained how to obtain 4-dimensional QCD based on the AdS7 S 4
compactification of 11-dimensional M-theory with an AdS7 scale eq.(113)
b = 211 (N)1/3
In that case the boundary theory is the conformally invariant 6-dimensional so called (2,0)
theory[34]. Excitations of this 5-brane cannot be understood in terms of open strings ending
on it, but in fact in terms of 2-branes ending on it. Upon compactification on a circle of radius
R1 , we obtain type IIA string theory, and for small R1 , IIA string theory in the perturbative
regime. There the 2-branes of M-theory turn into strings and we can use string theory to
describe excitations of that once compactified theory. Now the effective boundary theory is
5-dimensional. We still have to compactify once more, on a circle of radius R2 before we obtain
the desired 4-dimensional theory. In this second step we must take fermions anti periodic
around the second S 1 in order to break supersymmetry and conformal invariance. It is seen
that we assumed it made sense to perform the compactification in two steps, corresponding to
R1 R2 .
In the first step, we take SUSY preserving boundary conditions for the fermions, and we get
a string theory 4-brane with an effective 5-dimensional SYM theory with YM coupling eq.(120)
(10)
g52 = 8 2 s gs = 8 2 R1
(301)
where the last step is the standard identification between M-theory and IIA string theory upon
(10)
compactification on a circle of 10-dimensional radius R1 [48]. In the second compactification
on S 1 (R2 ), we may write
Z
Z
2R2
1
2R2
5 1
d x 2 d4 x 2 2 =
g5
g5
g4
g52
So we get for the 4-dimensional YM coupling
gY2 M
R1
=
4
R2
67
(302)
At low energies this effective D3 brane theory is large N QCD with a running coupling constant,
which at large energies becomes smaller and smaller (asymptotic freedom), until the theory
changes over into a 5-dimensional one at the cut-off R2 (in string units). Thus the YM
coupling eq.(302) is the value of s at that cut-off, and ideally we want that to be small in
accord with the requirement R1 R2 . If the second compactification also preserves SUSY, we
get the N = 4 superconformal theory, but if we take fermions anti periodic around S 1 (R2 ) we
expect to obtain ordinary large N QCD (with no quarks).
g2 M N
denote the t Hooft coupling, so
Let Y4
R1 =
R2
N
(303)
with fixed (preferably small) at N . Let us in fact begin with the second compactification
first. The AdS7 metric with a black hole is given by eq.(264)
ds2 =
2
b2
b4 2
d2
+ 2 dx25
d
+
4
2
4
b2
(304)
2b
4
4b
=
= 11 (N)1/3
n
3
3
Then put
b
= , = b
3
with an angular variable [0, 2]. Then (with a trivial rescaling of the xi )
b2 d2
1 1
b2 2 2
+
dx5
ds2 = 2 4 b2 d2 +
9
9
2 14
(305)
This metric of course is in 11-dimensional units. But we should also compactify on a circle of
radius R1 , say use the coordinate x5 for that, with that radius being /N times the radius of
the first compactification eq.(303), at least at the boundary, i.e. for very large values of . For
large , the circle with coordinate (the R2 circle) has metric
1 2 2 2
b d
9
and therefore radius b/3. For large the x5 coordinate has metric
1 2 2 2
b dx5
9
(before, dx25 meant the flat metric in 5 dimensions, now it refers just to the 5th coordinate) so
to compactify with the correct radius, we want to write
x5 =
68
with an angular variable. The full 11-dimensional metric, including the S 4 part is now
ds2 =
b2 2
1
b2 2 2 2
4 d2 +
d
9
9N 2
4
b2 2 X 2 b2 2
b2 d2
+
dxi + d4
+
9 1
4
2 14
(306)
(10)
= gs2/3 11 , R1
= g s s
(11)
ds2(11)
gs2/3 ds2(10)
(R111 )2 d 2
ds2(10)
R
= 1 ds2(11) (x5 part)
11
gs = e , = dilaton field
(307)
So the final 10-dimensional metric in the string frame becomes independent of N (!)
3
ds2(10)
83
1
1 3
2 d2
3 d2 +
9
3 13
4
1 X 2 1
dxi + d24
+ 3
9
4
1
(308)
This is the form given in [10]. From the radius of the circle parametrized by we find
b
2 1/3 11
=
= gs2/3 11
2/3
3N
3N
and
r
2 3
1
(309)
gs = e =
N
3
So this time we see that we have a non-trivial dilaton field depending on the coordinate . This
fact, actually makes the analysis considerably more complicated. We must set up fluctuation
equations for dilatons around the background provided by the above solution [37]. But a
number of subtleties exist [38, 41, 42] which we do not want to discuss here.
It is instructive to derive an equivalent version of the above, starting from the non-extremal
near horizon limit of the 11-dimensional 5-brane solution eq.(118) in 11 dimensional units
U 6
dU 2
U2
0
2
2
2
2
2
2
+ 4L
dt + d~x
1
(310)
ds =
6 + L d4
2
4L
U
U 2 1 UU0
69
U2
L r=
4L2
2
r2
dr 2
2
3 + d4
1 rr0
(311)
From the discussion above we have seen that when compactifying on a circle to get to the
10-dimensional IIA metric, we must in fact write
2
(11)
(312)
d 2 + gs2/3 ds2(10) = 211 e4/3 d 2 + e2/3 ds2(10)
ds2(11) = R1
Thus we find (in suitable units)
r = e4/3( e = r 3/4
)
4
r 3
X
1
dr 2
0
2
3/2
2
2
1
ds(10) = r
dt + 3
+
dxi + r 2 d24
3
r
r r0
1
(313)
h 3
r
r 3
0
f = 1
r
h 3
r
(
4
r 3
X
0
2
9/8
2
1
ds(10) (Einstein) = r
dt +
dx2i
r
1
dr 2
2
2
15/8
+r
3 + r d4
1 r0
r
e = r 3/4
70
(315)
The dilaton agrees with above, but the metric being in the Einstein frame does not yet. However, we may put it in the appropriate string frame using eq.(49), by multiplying the Einstein
frame metric by
1
e 2 = r 3/8
This precisely reproduces eq.(313).
In the supergravity approximation we should now set up fluctuation equations of motion
for dilatons and other fields in the backgrounds given. However, [38] it becomes important to
diagonalize these fluctuations appropriately in a non trivial way, since the dilaton background
itself is non trivial. We shall not, however, pursue these finer points.
Acknowledgement: I have benefited greatly from discussions with P. Di Vecchia, T. Harmark,
R. Marotta, N. Obers and R. Szabo. I thank the organizers of the two schools: Quantum
aspects of gauge theories, supersymmetry and unification, School in Leuven, January 18-23,
1999; and Nordic Course on Duality in String and Field Theories, 4-14 August, 1998, Nordita,
Copenhagen, for the invitation to present these lectures.
71
References
[1] J. Maldacena, The Large N limit of superconformal field theories and supergravity, Adv.
Theor. Math. Phys. 2:231, 1998, hep-th/9711200
[2] S.S. Gubser, I.R. Klebanov and A.M. Polyakov, Gauge theory correlators from noncritical
string theory, Phys. Lett. B428 (1998) 105, hep-th/9802109
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys.2: 253, 1998,
hep-th/9805028
[4] L. Susskind and E. Witten, The Holographic Bound in Anti-de Sitter Space, hepth/9805114
[5] I.R. Klebanov, From Three Branes to Large N Gauge Theories, hep-th/9901018
[6] J.H. Schwarz, Introduction to M Theory and AdS/CFT Duality, To be published in the
proceedings of 2nd Conference on Quantum Aspects of Gauge Theories, Supersymmetry
and Unification, Corfu, Greece, 21-26 Sep. 1998, hep-th/9812037
[7] A. Sen, Developments in superstring Theory, To be published in the proceedings of 29th
International Conference on High-Energy Physics (ICHEP 98), hep-th/9810356 Vancouver,
Canada, 23-29 Jul 1998.
[8] M.R. Douglas and S. Randjbar-Daemi, Two Lectures on AdS/CFT Correspondence,
hep-th/9902022
[9] T. Banks, W. Fischler, S. Shenker and L. Susskind, M Theory as a matrix model: a
conjecture, Phys. Rev. D55 (1997) 5112, hep-th/9610043
[10] E. Witten, Anti-de Sitter space, Thermal Phase Transition, and Confinement in Gauge
Theories, Adv. Theor. Math. Phys.2: 505, 1998, hep-th/9803131
[11] A.M. Polyakov, The Wall of the Cave, hep-th/9809057;
I.R. Klebanov and A.A. Tseytlin, D-Branes and dual gauge theories in type 0 string,
hep-th/9811156
[12] M. Duff, R.R. Khuri and J.X. Lu, String Solitons, Phys. Rep.259 (1995) 213, hepth/9412184
[13] K.S. Stelle, BPS Branes in Supergravity, hep-th/9803116
[14] R. DAuria and P. Fre, BPS Black Holes in Supergravity, hep-th/9812160
[15] R. Argurio, Brane Physics in M-theory, Doctoral Thesis, hep-th/9807171
[16] S.P. de Alwis, A note on brane tension and M-theory, Phys. Lett. B388 (1996) 291,
hep-th/960711
72
[17] P. Pasti, D. Sorokin, M. Tonin, On Lorentz Invariant Actions for Chiral p-forms, Phys.
Rev. D55, 6292 (1997), hep-th/9611100
[18] C. Hull, String dynamics at Strong coupling, Nucl. Phys. B468 (1996) 113, hepth/9512181
[19] E. Witten.String Theory Dynamics in Various Dimensions, Nucl. Phys. B443 (1995)
85; hep-th/9503124
[20] J. Polchinski, Dirichlet Branes and Ramond-Ramond Charges, Phys. Rev. Letters 75
(1995) 4724, hep-th/9510017;
J. Polchinski, TASI Lectures on D-Branes, hep-th/9611050
73
[32] S. Ferrara and C. Fronsdal, Gauge Fields as Composite Boundary excitations, Phys.
Lett. B433 (1998) 19, hep-th/9802126;
S. Ferrara, C. Fronsdal and A. Zaffaroni, On N = 8 Supergravity on AdS5 and N = 4
Superconformal Yang Mills Theory, Nucl. Phys. B532 (1998) 153, hep-th/9802203;
L. Andrianopoli and S. Ferrara, On Short and Long SU(2, 2|4) Multiplets in the AdS/CFT
Correspondence, hep-th/9812067
[33] P.S. Howe and P.C. West, Is N = 4 Yang-Mills Theory Soluble?, Talk given at the
6th Quantum Gravity Seminar, Moscow, Russia, 6-11 Jun 1996, and at the Workshop on
Gauge Theories, Applied Supersymmetry and Quantum Gravity, London, England, 5-10
Jul 1996. In Moscow 1996, Physics 622-626. hep-th/9611074
[34] E. Witten, Some Comments on String Dynamics, in Strings 95, ed. I. Bars et. al. (World
Scientific, 1997), hep-th/9507121;
A. Strominger, Open p-Branes, Phys. Lett. B383 (1996) 44, hep-th/9512059
E. Witten, New Gauge Theories in 6 dimensions, J. High Energy Phys. 9801: 001,
1998, Adv. Theor. Math. Phys. 2: 61, 1998, hep-th/9710065
[35] G. t Hooft, A Planar Diagram Theory For Strong Interactions. Nucl. Phys. B72 (1974)
461
[36] J. Maldacena, Wilson Loops in Large N Field Theories, Phys. Rev. Lett. 80 (1998)
4859, hep-th/9803002
[37] C. Csaki, H. Ooguri, Y. Oz and J. Terning, Glueball Mass Spectrum From Supergravity,
hep-th/9806021
[38] A. Hashimoto and Y. Oz, Aspects of QCD Dynamics from String Theory, hepth/9809106
[39] H. Ooguri, H. Robins and J. Tannenhauser, Glueballs and Their Kaluza-Klein Cousins,
Phys. Lett. B437 (1998) 77, hep-th/9806171
[40] E. Witten, Theta Dependence in the Large N Limit of 4 Dimensional field Theories,
Phys. Rev. Lett. 81 (1998) 2862, hep-th/9807109;
E. Witten, Baryons and Branes in Anti de Sitter Space, J. High Energy Phys.
9807:006,1998, hep-th/9805112
[41] S.S. Gubser, I.R. Klebanov and A.A. Tseytlin, Coupling Constant Dependence in the
Thermodynamics of N = 4 Supersymmetric Yang-Mills Theory, Nucl. Phys. B534 (1998)
202, hep-th/9805156
[42] J. Pawelczyk and S. Theisen, AdS5 S 5 Black Hole Metric at O( 3 ), J.High Energy
Phys. 9809:010,1998, hep-th/9808126
[43] A Brandhuber, N. Itzhaki, J. Sonnenschein and S. Yankielowicz, Wilson Loops, Confinement and Phase Transitions in Large N Gauge Theories from Supergravity, J. High
Energy Phys. 06 (1998) 001, hep-th/9803263; Wilson Loops in the Large N Limit at
74
75